Patients ( Table 1 ) were recruited under a human protocol of clinical trial for the access of T1D and healthy donor peripheral blood mononuclear cells (PBMCs) in agreement with the Research Ministry Authorization (MESR number DC-2015-2536/IDRCB number 2015-A01875-44). Recent-onset T1D patients (18 males, 12 females, aged 35.3 ± 17.5 years) were studied within 3 months from diagnosis (9.4 ± 14.5 days, range 1–75 days). HLA genotyping was performed by AmbiSolv genotyping (Dynal/Invitrogen) on DNA with Gentra Puregene Blood kit A (Qiagen). Eleven recent-onset T1D patients expressed HLA-DQ8 (DQA*03:01, DQB*03:02) and 19 co-expressed HLA-DQ8 and HLA-DQ2 (DQA*05:01, DQB*02:01). Long-standing HLA-DQ8-expressing T1D patients (3 males, 3 females, ages 55.25 ± 21.93 years) had a median of diabetes duration of 5.34 ± 2.6 years (range 0.7 to 7 years). Control subjects were healthy donor (n = 9, 3 males, 6 females, aged 34 ± 18.6 years) and T2D controls (n = 4, 1 male, 3 females, aged 67.7 ± 4.57 years) and only express HLA-DQ8 molecule.

YES-RIP-hB7.1 mice were obtained by lentiviral transgenesis of YES mice (H-2 D^b , mouse β2 microglobulin, IAα^b , β^b , IEβ^b quintuple KO mice expressing a chimeric-α3 H-2 D^b domain, human β2 microglobulin-HLA-A*02:01 monochain molecule named HHD- and the HLA-DQB1*03.02 and HLA-DQA1*03.01 genes) (15, 16). The RIP-hB7.1 transgene was inserted in a HIV-derived recombinant lentiviral vector (LV-RIP-hB7.1; insert of 4,640 bp) as previously reported into YES mice (17, 18). Retroviral pseudotypes were injected into fertilized eggs obtained from super-ovulated female YES mice mated with male YES mice (16). Fertile and transduced eggs were reimplanted into pseudo-pregnant C57BL/6xCBA F1 mice. All mice were maintained under specific pathogen-free conditions, and experimental studies were performed in accordance with the Institutional Animal Care and Use Guidelines and accredited by the Ethics Committee n°34 of Paris Descartes under number CEEA34.CB.024.11. Mice were monitored three times a week for glycosuria. When glycosuria was detected, diabetes was diagnosed when two successive glycemic values >250 mg/dl were detected at 24 h interval. Diabetes incidences curves correspond to the percentage of mice diagnosed as diabetic referring to the aforementioned criteria.

YES mice submitted to lentiviral transgenesis were genotyped using the following primers: 5′end (AGGGAACATCACCATCCAAG) and 3′end (TGCCAGTAGATGCGAGTTTG), annealing temperature: 62°C, amplicon: 181 bp. A positive RIP-hB7.1 mouse was selected to perform the molecular characterization of the LV-RIP-hB7.1 transgene. To characterize insertion sites in genomic DNA of the founder, we realized a sequence capture design using the SeqCap EZ system (Roche NimbleGen) targeting 97.6% of the transgene. After nebulization, fragmented genomic DNA was end-repaired and ligated with adapters. A double capture was realized to obtain a GS Junior library ready for emulsion PCR (emPCR). Fragments were then annealed to capture beads and clonally amplified by emPCR (emPCR Amplification Method Manual LibL; GS Junior Titanium Series, Roche). Beads with the cloned amplicons were then enriched, loaded on a 454 picotiter plate, and sequenced on the Roche GS Junior Sequencer according to the protocol of the manufacturer (Sequencing Method Manual, GS Junior Titanium Series, Roche). Image analysis and base calling of the raw sequencing data were performed using the default “shotgun” Roche GS Junior data analysis pipeline. To obtain the position of transgene insertions in genomic DNA, sequence reads were aligned to the 50 bp of each end of the transgene primary sequence using BWASW, an algorithm designed for long reads with more errors. Aligned sequences were then aligned against C57BL/6NJ genomic sequences to obtain scaffold at both ends of the transgene sequence. A 10 kb for each region was extracted and amplified by long range PCR (LR-PCR). A nested PCR targeting the hB7.1 transgene was performed on the long range PCR product to highlight the presence of the transgene and identify the sequence of the insertion site of the LV-RIP-hB7.1 transgene on genomic DNA. A genotyping PCR was designed from the region overlapping genomic DNA and LV-RIP-hB7.1 transgene using the following primers: 5′end (TAAATGCAGGGCTCCAGACT) and 3′end (TAGTGTGTGCCCGTCTGTTG), annealing temperature: 62°C, amplicon: 645 bp.

To compare the genetic background of YES-RIP-hB7.1 and YES mice, we performed a GenScan SNP Affymetrix. Briefly, high-quality genomic DNAs (250 ng) were digested with NspI and StyI enzymes. NspI and StyI adaptors were then ligated to restricted fragments followed by PCR using the universal primer PCR002. Each amplicon was purified, pooled, and used for fragmentation and end-labeling with biotin using terminal transferase. Labeled targets were then hybridized overnight to Genechip^® Mouse Diversity Genotyping Array (Affymetrix, ref: 901615) at 49°C. Chips were washed on the fluidic station FS450 following specific Affymetrix protocols, scanned with the GCS3000 7G, and analyzed using GCOS software to obtain raw data (.cel files). Genotypes were identified by the Affymetrix Genotyping Console tool using Dynamic Model (DM) and Bayesian Robust Linear Model with Mahalanobis (BRLMM) mapping algorithms. Genotypes were then extracted for each SNP of the chip. Big data were exploited in R console software. We submitted the data generated by the Affymetrix SNP Array detection to a public repository (http://www.ncbi.nlm.nih.gov/geo) and the GEO Series accession numbers were GSE101551 and GSE151644.

For in vivo T-cell depletion, YES-RIP-hB7.1 mice were injected with four weekly injections (i.p.) of 1 mg of anti-CD4 (GK1.5 clone) (19) and 1 mg of anti-CD8 (YTS 169.4 clone) (20) starting at 2 or 8 weeks of age and monitored for diabetes development. For in vivo poly(I:C) stimulation, YES and YES-RIP-hB7.1 mice (8 weeks old of age) were submitted to seven daily injections (i.p.) of 100 µg of poly(I:C) VacciGrade (vac-pIC, Invivogen) and monitored for diabetes development (16).

T-cell cytotoxicity assay was performed on HHD-transfected P815 cells prepulsed with 10 μg peptide for 2 h at 37°C using the LDH Cytotoxicity Detection Kit^PLUS (Roche) or hPPI/HHD-doubled transfected P815 cells. High control of lysis corresponded to cell killing with a lysis solution (Tween-20), which provide the maximum LDH release. Low control of lysis, which provide the spontaneous cytotoxicity, was evaluated on cells that were not submitted to additional treatment. Specific lysis was calculated as optical density of (targeted condition − spontaneous lysis)/(maximum LDH release − spontaneous lysis) × 100.

IFNγ-ELISpot assays were achieved as previously reported (21). Spots were counted using Bioreader 5000 Pro Sf (BioSys GmbH). Splenocytes from mice were stimulated overnight with peptides and Il-2 (5 U/ml final). The enzyme-linked immunospot (ELISpot) assay was performed using mouse γ-ELISpot antibody pair, from U-CyTech biosciences. Data are the mean of triplicate wells. The background of IFNγ responses was evaluated in the absence of peptide. Specific IFNγ responses were expressed as spot-forming cells (SFC) per 10⁶ cells after normalization of the background. Positive controls were triplicates of cells stimulated by 1 μg/ml ConA (Sigma-Aldrich) and negative controls by pyruvate dehydrogenase (PDHase_208-216) irrelevant peptide.

The cDNA sequence of preproinsulin was mutated to convert the Ala codon in position 3 of the leader DNA sequence to Asp by site-directed mutagenesis (NEB) to abolish the signal-sequence site cleavage of PPI (22). Mutated PPI was cloned in pFastBac vector to generate the PPI-recombinant bacmid, then PPI-recombinant Baculovirus to produce human PPI protein into Bac-to-Bac Baculovirus Expression System (Invitrogen) according to the instruction of the manufacturer. Purification was performed on pelleted infected insect cells, following preparation and extraction procedures for insoluble proteins (inclusion body) from Escherichia coli (23). Purity was confirmed on SDS/PAGE gel and quantification was performed using the BCA Protein assay kit (Pierce).

Spleen cells (10⁵/well) were incubated with 0.5 μg antigen/well or 0.1 µg peptide/well for 72 h at 37°C in triplicate. Proliferation was evaluated with BrdU Cell Proliferation Assay Kit (Cell Signaling) and expressed as proliferation index (PI). Background and positive controls were evaluated in triplicate wells containing 10⁵ cells/well incubated without antigen or in the presence of 10 μg/ml final concentration of anti-mouse CD3ϵ antibody, respectively.

Immunofluorescence stainings were performed on formalin-fixed paraffin pancreas sections that were deparaffinized in xylene and dehydrated by ethanol. After washing, antigen retrieval was realized by hot incubation in citrate buffer, followed by permeabilization (20 min in PBS 1×/0.4% Triton X-100) and saturation (30 min PBS 1×/1% horse serum) before immunostaining with biotinylated rat anti-human CD3ε(AbD Serotec) and polyclonal rabbit anti-glucagon (DAKO) antibodies overnight. Slides were washed with PBS 1×/1% BSA/0.1% Triton X-100 and stained with an anti-rabbit Ig-FITC antibody (Abcam) and SAV-Cy3 (Abcam) at RT. Sections were mounted in Vectashield mounting medium for fluorescence with DAPI (Vector Laboratories). Observations were made with a spinning disk confocal apparatus at the Cochin Institute Imaging Platform and pictures analyzed with the ImageJ software.

Stable immunohistochemical stainings were performed on paraffin-embedded pancreas sections and stained with polyclonal guinea-pig anti-human insulin (DAKO) or rat anti-mouse CD4 biotin (eBioscience), followed by incubation with peroxidase-labeled antibodies. All reactions were revealed with diaminobenzidine (DAB, Genemed). Sections were counterstained with hematoxylin and mounted. Observations were made with Zeiss AxioObserver Z1 microscope coupled with MRm Axiocam Zeiss and pictures analyzed with the ImageJ software.

Assessment of total β-cell mass was performed on scan-stained microscope slides with Inform software using a guinea pig anti-human insulin antibody (DAKO) as the ratio between β-cell surface (µm²)/pancreas surface (µm²) multiplied by pancreas weight (mg) (24). Assessment of total α-cell mass was performed on scan-stained microscope slides with Inform software using a rabbit anti-glucagon antibody (DAKO) as the ratio between α-cell surface (µm²)/pancreas surface (µm²) multiplied by pancreas weight (mg). Quantification of insulin-positive or glucagon-positive areas was performed from the entire pancreas on serially cut 8-µm-thick sections. Five to 10 pancreatic sections were processed for immunostaining. Cellular mass obtained for each group was compared using the non-parametric Mann–Whitney statistical test.

Tetramers (TMrs) associating the chimeric HLA-A*02:01-H-2K^b and the human β2-microglobulin to hPPI peptides were obtained from the NIH Tetramer Core Facility. Stainings were performed as previously reported (11). Incubation of cells with TMrs was performed for 30 min at RT. The following antibodies were used: anti-mouse CD3ε-AlexaFluo700, anti-mouse CD8α-APC, anti-mouse CD19-PerCP-Cy5, and anti-mouse CD14-PercP-Cy5 antibodies (BD/Pharmingen and eBioscience). Cells were analyzed with a BD LSR-Fortessa flow cytometer and the FlowJo 9.2 software (Tree Star Inc.). For single-cell RT-PCR assays, CD8⁺/TMr⁺ T cells were sorted with a FACSAria II at 1 cell/well into 96-well PCR plates containing 5 μl PBS 1× treated with diethyl pyrocarbonate (DEPC) (Sigma) and immediately frozen at −80°C. Single-cell RT-PCR was performed as previously described (11). Perforin (prf1), granzyme A (gzma), granzyme B (gzmb), Fas ligand (tnfsf6), IFNγ (ifng), TGFβ (tgfb), TGFβ R1 (tgfbr1), TGFβ R2 (tgfbr2), TGFβ R3 (tgfbr3), TNFα (lta), IL2 (il2), IL7R (il7r), IL10Rα (il10ra), IL15 (il15), IL15R (il15r), IL21 (il21), CCR7 (ccr7), and CD3ϵ (cd3e) mRNAs were analyzed ( Table S15 ). Naive and memory CD8⁺/TMr⁺ T cells were differentiated by the Foxo1 expression using 5′ (GGA CAG CCG CGC AAG ACC AG) and 3′ (ACT GTT GTT GTC CAT GGA CGC) first PCR primers and 5′ (ATC ACC AAG GCC ATC GAG AGC) and 3′ (TTG AAT TCT TCC AGC CCG CCG A) second PCR primers.

Islets of Langerhans were isolated as described previously (25). Islets were dissociated with Cell Dissociation Solution Non-Enzymatic (Sigma) for 30 min at RT. Cells were filtered and stained with anti-HLA-DQ-FITC (BD Biosciences), anti-human β2m-PE (BD Biosciences), guinea pig anti-human insulin (DAKO), anti-guinea pig Ig-Biot and SAV-PECy7, anti-mouse CD4-AlexaFluo700, anti-mouse CD8α-APC, anti-mouse CD19-Percp-Cy5.5, and/or anti-mouse TCRβ-efluo450 antibodies (eBioscience). Acquisition was done with a BD FACSAria flow cytometer and analyzed using FlowJo 10.7.1 software.

Handpicked islets were pooled in 24-well plates for 24 h in RPMI supplemented with 10% FCS, penicillin–streptomycin, and 5.10⁻⁵ M β-mercaptoethanol allowing extrusion of infiltrating lymphocytes (25). Infiltrating lymphocytes were recovered. Cells recovered were used for the analysis of T cells and antigen-presenting cells. T-cell analysis was performed using the Foxp3 intra-staining buffer set according to the recommendation of the manufacturer (eBioscience) with the following combination: anti-mouse CD45-efluo450, anti-mouse CD3ϵ-AlexaFluo700, anti-mouse B220-HorizonViolet500, anti-mouse CD8α-Percp-Cy5.5, anti-mouse CD4-PE, anti-mouse Foxp3-APC, and anti-mouse CD25-BrillantViolet711 mAbs (eBioscience). β Cells and dendritic cells were analyzed with anti-mouse CD45-AlexaFluo700, anti-mouse TCRβ-APC-Cy7, anti-mouse B220-HorizonViolet500, anti-mouse CD11b-efluor450, anti-mouse CD11c-APC, anti-mouse CD8α-BrillantViolet605, and anti-mouse CD4-PE mAbs (eBioscience). Acquisitions were performed with a BD LSR-Fortessa flow cytometer and analyzed using FlowJo 10.7.1 software.

The biostatistic method used to compare diabetes incidences between different groups of mice was the log-rank Mantel–Cox test. Comparison of distribution scores between the different mouse strains used the non-parametric Mann–Whitney test. T-cell reactivity was compared using non-parametric Mann–Whitney test and non-parametric Kruskal–Wallis test. ns, non-significant, *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001. For T-cell proliferation assays, we used the Bland and Altman test to determine the threshold value for a global reliability statistic for significant proliferation response.

Experimental studies were performed in accordance with the Institutional Animal Care and Use Guidelines and accredited by the ethics committee under number CEEA34.CB.024.11. Clinical trial: MESR number DC-2015-2536/IDRCB number 2015-A01875-44.

We submitted the data generated by the Affymetrix SNP Array detection to a public repository (http://www.ncbi.nlm.nih.gov/geo) and the GEO Series accession numbers were GSE101551 and GSE151644.

In order to enforce the development of T1D in YES mice, we introduced the hB7.1 gene under the control of the rat insulin gene promoter (RIP) using the LV-RIP-hB7.1 vector (17). We selected three mice that were positive for hB7.1 for backcrossing onto the YES background. Throughout the crosses, one founder progeny was selected. To further stabilize its lineage, we realized a NimbleGen Sequence Capture of the RIP-hB7.1 transgene of the founder and one of its offspring. Inserted hB7.1 sequences were identified as indicated in Table S2 , matching with regions located on chr. 11, 14, 16, and 19 in the founder mouse. Nevertheless, the only region located in chr.19 remained detectable in the offspring, a chromosome with no Idd known to associate with TID. We did not evidence hot spots of enhancer transcripts close to the LV-RIP-hB7.1 insertions using the mouse ENCODE database from UCSC Genome Bioinformatics (GSE101551 and GSE151644 GEO data accession numbers). We screened 128 mice from the progeny for the selected insertion sites in order to stabilize the lineage at a homozygote status using a classical progeny test.

Insulitis and spontaneous diabetes developed in founder YES-RIP-hB7.1 offspring that were submitted to 20 brother–sister mattings. In the founder progeny, 46 out of 128 YES-RIP-hB7.1 mice developed diabetes, while diabetes was not observed in non-transgenic YES mice. Diabetes incidence of stabilized YES-RIP-hB7.1 mice is shown in Figure 1A . Age at onset varied from one mouse to another, spanning from 9 to 51 weeks of age. The overall prevalence of diabetes was similar in female (53.8%) and male (50.0%) YES-RIP-hB7.1 mice. The difference in diabetic incidence was not statistically different between males and females. The average glycemia in YES-RIP-hB7.1 mice when diagnosed as diabetic was 520 ± 115 and 532 ± 99 mg/dl in female and male, respectively. The average glycemia in non-diabetic YES-RIP-hB7.1 mice at the end of the experiment (40 weeks) was 143 ± 35 and 125 ± 18 mg/dl in female and male, respectively. The average glycemia in YES mice was 107 ± 13 and 102 ± 12 mg/dl in female and male, respectively. In order to address whether diabetes was immune-related, we analyzed hematoxylin–eosin-stained pancreas paraffin sections from diabetic and non-diabetic YES-RIP-hB7.1 mice. As shown in Figure 1B , insulitis was detected by immunofluorescence staining using an anti-CD3ϵ antibody and a rabbit anti-glucagon antibody to locate remnant islets, showing the infiltration of islets by CD3⁺ T lymphocytes ( Figure 1C ). Glucagon-positive cells were dispersed in remnant islets that showed a dislocated architecture in YES-RIP-hB7.1 diabetic mice ( Figure 1D ) compared with the YES mice control ( Figure 1E ). Stable immunohistochemistry staining with an anti-CD4 antibody confirmed the detection of CD4⁺ T cells within the infiltrate ( Figure 1F ). As shown in Figure 2 , the number of islets expressing insulin was decreased by 70.9% in diabetic YES-RIP-hB7.1 mice as compared with YES mice or non-diabetic YES-RIP-hB7.1 mice. Islet size ( Figure 2B ) was decreased by over 60% in diabetic YES-RIP-hB7.1 mice as compared with YES controls. A dramatic decrease of the β-cell mass was observed in diabetics YES-RIP-hB7.1 mice as compared with age-matched YES mice (0.60 ± 0.49 versus 3.93 ± 0.65, respectively, p ≤ 0.02) and non-diabetic YES-RIP-hB7.1 mice (7.01 ± 2.06) ( Figure 2C ). The α-cell mass ( Figure 2D ) also showed a significant decrease in diabetic YES-RIP-hB7.1 mice as compared with non-diabetic YES-RIP-hB7.1 mice and YES controls (0.60 ± 0.49 versus 3.93 ± 0.65, respectively, p ≤ 0.02). In non-diabetic YES-RIP-hB7.1 mice, the α-cell mass was heterogeneous (7.01 ± 2.06). Infiltrates from three diabetic mice were recovered and pooled to be analyzed ( Figure S1 ). They were composed of 52% T cells, among which 81% were CD8⁺ T cells, 8% were CD4⁺ T cells, 21% were β cells, 0.8% were dendritic cells, and 0.6% were macrophages. Among CD4⁺ T cells, 7% were CD4⁺CD25⁺FoxP3⁺ T cells. Infiltrates recovered from three non-diabetic YES-RIP-hB7.1 mice ( Figure S1 ) were composed of 56% T cells, among which 56% were CD8⁺ T cells, 34% were CD4⁺ T cells, 34% were β cells, and 2.3% were dendritic cells or macrophages. Among CD4⁺ T cells, 1% were CD25⁺FoxP3⁺. These data demonstrate the presence of an immune response along the development of diabetes in YES-RIP-hB7.1 mice, a dramatic decrease in β cells and a reduced α-cell mass. As the extent of infiltration was milder than the infiltrate seen in the NOD mouse model, we addressed whether autoimmune development could be delayed by a transient treatment with anti-CD4 and anti-CD8 monoclonal antibodies. As shown in Figure 3A , in vivo depletion of T cells by anti-CD4 and anti-CD8 antibodies from either 2 to 5 or 8 to 11 weeks of age significantly delayed T1D development in YES-RIP-hB7.1 mice. There was no difference in the protection observed in early-treated and late-treated mice. As previously described in the YES mouse model (16), we addressed whether diabetes development could be triggered in YES-RIP-hB7.1 mice by poly(I:C). As shown in Figure 3B , diabetes was induced within 6 to 17 days following the first poly(I:C) injection.

We previously characterized preproinsulin-specific CD8⁺ T cells in human T1D diabetes (11). In order to validate the YES-RIP-hB7.1 mice as a model to study T cells in T1D, we characterized CD8⁺ T-cell responses in 12- to 20-week-old diabetic mice and control mice that remained diabetes-free up to 30 weeks of age as previously reported (11). Using an IFNγ-ELISpot assay, we evaluated CD8⁺ T-cell responses to a library of hPPI peptides that were selected for presentation by HLA-A*02:01. Most hPPI peptides that were previously defined as recognized by human CD8⁺ T cells (11, 26) were recognized by T cells obtained from diabetic YES-RIP-hB7.1 mice ( Figure 4A ). Responses were detected against hPPI_2-11 (p ≤ 0.03), hPPI_6-14 (p ≤ 0.001), hPPI_15-24 (p ≤ 0.02), and hPPI_33-42 (p ≤ 0.013). No responses were observed against any of these hPPI peptides in control YES mice and in non-diabetic YES-RIP-hB7.1 mice. When considering responses that were over the mean ± 3SD of responses seen in control YES mice, significant responses were observed against hPPI_6-14 (9/26, p ≤ 0.016), hPPI_15-24 (11/26, p ≤ 0.004), and hPPI_33-42 (9/26, p ≤ 0.016). Considering individual mice, responses were observed against hPPI_2-11 (19.2%), hPPI_30-39 (23.1%), hPPI_34-42 (19.2%), hPPI_42-51 (15.4%), and hPPI_101-109 (15.4%), although they did not reach statistical significance in the whole population of diabetic mice analyzed ( Table S3 ). As a whole, 93.75% YES-RIP-hB7.1 diabetic mice showed a response to at least one of the HLA-A*02:01-restricted hPPI peptides studied (defined as > mean ± 3SD of the response of YES mice against each given peptide, Table S3 ). An IFNγ-ELISpot response of pancreatic infiltrating cells was also documented against pooled hPPI peptides ( Figure 4A , in red symbol) as the limited number of infiltrating cells prevented testing individual peptides. Using chimeric HLA-A*02:01 tetramers, we detected significant expansions of CD8⁺ T cells against hPPI_6-14 (47.1%), hPPI_15-24 (29.4%), and hPPI_{33-42 (}47.1%) in YES-RIP-hB7.1 mice as compared with YES mice ( Figure 4B ). Considering individual mice, an expansion of CD8⁺ T cells was seen in 70.6% YES-RIP-hB7.1 mice against either hPPI_6-14, hPPI_15-24, or hPPI_33-42. Single hPPI-specific CD8⁺ T cells showed gene expression profiles that were characteristic of memory T cells ( Figure 4C ). hPPI-specific CD8⁺ T cells from diabetic YES-RIP-hB7.1 mice expressed gzmA (59.38%), among which 100% co-expressed ccr7, while only 4.33% hPPI-specific CD8⁺ T cells from both YES-RIP-hB7.1 non-diabetic mice and YES control mice expressed gzmA, of which 76.7% co-expressed ccr7. To confirm a naive T-cell profile of cells from non-diabetic mice, we performed gene expression analysis for foxo1, which is expressed in basal naive T cells (27). An increased percentage of CD8⁺ T cells expressed foxo1 in non-diabetic as compared with diabetic mice ( Figure S2A ). A significant level of foxo1 expression was detected in 10-cell and single-cell batches ( Figure S2B ) in non-diabetics YES-RIP-hB7.1 mice when normalized to cd3ϵ expression. Finally, we determined ex vivo cytotoxic effects of hPPI-specific CD8⁺ T cells ( Figure S3 ) in five diabetic YES-RIP-hB7.1 mice as compared with three YES controls in a LDH-release assay using peptide-pulsed HLA-A*02.01-transfected P815 target cells. Cytotoxic responses were detected against all hPPI peptides in diabetic YES-RIP-hB7.1 mice. Mice considered individually responded to different hPPI peptides. All mice that responded against at least one hPPI peptide also responded to full-length hPPI. Three mice showed cytotoxic responses against several hPPI peptides and two mice showed no cytotoxicity against any of the eight hPPI peptides. A heterogeneity of the cytotoxic responses in diabetic YES-RIP-hB7.1 was observed against hPPI.

We studied HLA-DQ8-restricted T-cell responses to an overlapping hPPI peptide library. Significant T-cell proliferative responses were detected against hPPI_1-15 (p ≤ 0.003), hPPI_8-23 (p ≤ 0.008), hPPI_16-30 (p ≤ 0.015), hPPI_25-40 (p ≤ 0.014), hPPI_33-47 (p ≤ 0.05), hPPI_55-70 (p ≤ 0.029), hPPI_61-76 (p ≤ 0.032), hPPI_80-97 (p ≤ 0.022), and hPPI_92-110 (p ≤ 0.028) peptides in YES-RIP-hB7.1 diabetic mice as compared with YES controls ( Figure 5A ). A non-significant trend for proliferative responses was detected against the same peptides in a few non-diabetic YES-RIP-hB7.1 mice. As in the case of CD8⁺ T cells, individual responses were observed against peptides covering the whole hPPI sequence. After determination of the threshold value for each peptide in controls using pairs of measurements and the Bland and Altman test ( Figure S4 , Table S4 ), proliferative responses were observed in a significant number of mice against all peptides except for hPPI_16-30, hPPI_40-55, and hPPI_46-61. A significant response to full-length hPPI was observed in 51.5% of mice (p ≤ 0.009). Significant responses were observed in more than 30% diabetic YES-RIP-hB7.1 mice against hPPI_1-15 (30.3% mice, p ≤ 0.008), hPPI_8-23 (36.4% mice, p ≤ 0.002), hPPI_18-30 (30.3% mice, p ≤ 0.04), hPPI_25-40 (36.4% mice, p ≤ 0.002), hPPI_33-47 (30.3% mice, p ≤ 0.009), hPPI_55-70 (39.4% mice, p ≤ 0.009), hPPI_61-76 (30.3% mice, p ≤ 0.04), hPPI_70-86 (30.3% mice, p ≤ 0.04), and hPPI_92-110 (45.4% mice, p ≤ 0.002) ( Table S5 ). We observed a significant correlation between proliferative responses against the hPPI protein and several hPPI peptides in diabetic YES-RIP-hB7.1 mice, including hPPI_16-30, hPPI_25-40, hPPI_33-47, hPPI_55-70, hPPI_61-76, and hPPI_92-110 ( Figure S5 ). Proliferative responses were seen against at least one hPPI peptide in 87.9% diabetic mice.

The presence of hPPI-specific CD4⁺ T cells in the peripheral blood of T1D diabetic patients was analyzed by ex vivo proliferative assay and compared with HLA-DQ8 healthy donor and type 2 diabetes (T2D) patients as controls in order to address whether hPPI regions recognized in human matched with peptides recognized in YES-RIP-hB7.1 mice. Clinical features of T1D patients and control subjects (who were negative for detection of anti-GAD, anti-IA2, and anti-ZnT8 autoantibodies) are summarized in Table 1 . Considering the possible trans-complementation between HLA-DQ8 and HLA-DQ2 class II MHC molecules, T1D patients were included according to their HLA-DQ8 or HLA-DQ2/DQ8 genotype. A proliferative response was observed against full-length recombinant hPPI in HLA-DQ8 T1D patients (p ≤ 0.04) and in HLA-DQ2/DQ8 T1D patients (p ≤ 0.003) as compared with controls ( Figure 5B ). When dissecting epitopes recognized, significant proliferative responses were observed against leader sequence hPPI_1-15 (p ≤ 0.007), hPPI_8-23 (p ≤ 0.003) and hPPI_18-30 (p ≤ 0.02) peptides, overlapping leader sequence and B-chain hPPI_20-35 peptide (p ≤ 0.007), B-chain hPPI_25-40 (p ≤ 0.03) and hPPI_46-61 (p ≤ 0.03) peptides, C-peptide hPPI_70-86 (p ≤ 0.05) peptide, overlapping C-peptide and A-chain hPPI_80-97 peptide (p ≤ 0.02), and the A-chain hPPI_92-110 peptide (p ≤ 0.03) in HLA-DQ8 patients ( Figure 4B ). Significant proliferative responses were observed against leader sequence hPPI_8-23 peptide (p ≤ 0.009), B-chain hPPI_20-35 (p ≤ 0.05) and hPPI_25-40 (p ≤ 0.035) peptides, C-peptide hPPI_55-70 (p ≤ 0.003) peptide, and A-chain hPPI_92-110 (p ≤ 0.04) peptide in HLA-DQ2/DQ8 T1D patients. After determination of the threshold value for each peptide in controls using pairs of measurements and the Bland and Altman test ( Figure S6 , Table S6 ), proliferative responses against hPPI_1-15 (72.75%, p ≤ 0.01), hPPI_18-30 (45.4%, p ≤ 0.041), hPPI_20-35 (72.7% of T1D, p ≤ 0.04), hPPI_40-55 (54.4%, p ≤ 0.003), hPPI_46-61 (54.4%, p ≤ 0.02), and hPPI_80-97 (63.6%, p ≤ 0.03) were seen in a significant number of HLA-DQ8 recent-onset T1D patients ( Table S7 ). Significant proliferative responses to hPPI_33-47 (42.1%, p ≤ 0.05), hPPI_40-55 (36.8%, p ≤ 0.03), hPPI_55-70 (57.9%, p ≤ 0.0006), and hPPI_92-110 (47.4%, p ≤ 0.03) were observed in HLA-DQ2/DQ8 recent-onset T1D patients ( Table S8 ). Distribution of proliferative responses to hPPI peptides was different in HLA-DQ8 and HLA-DQ2/DQ8 patients. As a total, proliferative responses were seen against at least one hPPI peptide in 94.44% T1D patients. Overall, the proliferative responses in HLA-DQ8 T1D patients largely matched the responses in YES-RIP-hB7.1 mice.

Based on the observation of insulin mRNA alternative splicing events when reanalyzing databases generated in the context of human islet inflammation (28, 29), we designed an HLA-A*02:01-restricted spliced hPPI peptide, sp-hPPI_60-68, and three HLA-DQ8-restricted spliced peptides, named sp-hPPI_55-74, sp-hPPI_52-74, and sp-hPPI_51-77 using HLA-restrictor and NetMHCII 2.3 server, respectively ( Figure 6A ). Using an IFNγ-ELISpot assay, we detected HLA-A*02:01-restricted IFNγ responses against sp-hPPI_60-68 peptide ( Figure 6B ) and HLA-DQ8-restricted responses to the sp-hPPI_51-77 peptide (p ≤ 0.05) in splenocytes from diabetic YES-RIP-hB7.1 mice as compared with non-diabetic YES-RIP-hB7.1 or YES mouse controls ( Figure 6C ). Proliferative responses were further observed against the sp-hPPI_52-74 and sp-hPPI_51-77 peptides (p ≤ 0.009 and p ≤ 0.049, respectively, Figure 6C ) in diabetic YES-RIP-hB7.1 mice as compared with non-diabetic YES-RIP-hB7.1 or YES controls. Seemingly, we addressed whether proliferative responses were detected in T1D patient PBMCs. Significant proliferative responses were seen against the sp-hPPI_55-74 and sp-hPPI_52-74 peptides in recent-onset HLA-DQ8 patients (p ≤ 0.0001 and p ≤ 0.001, respectively, black circle, Figure 6D ) and in recent-onset HLA-DQ2/DQ8 diabetic patients (p ≤ 0.008 and p ≤ 0.03, respectively, black square, Figure 6D ) as compared with control donors. Threshold values were determined in YES mice ( Table S9 ) and in human controls ( Table S10 ) using pairs of measurements and the Bland and Altman test ( Figure S7 ). The frequency of recognition of sp-hPPI_55-74, sp-hPPI_52-74, and sp-hPPI_51-77 peptides was not significantly different in diabetic YES-RIP-hB7.1 mice ( Table S11 ). By contrast, the frequency of recognition of sp-hPPI_55-74 and sp-hPPI_52-74 peptides was significant in recent-onset HLA-DQ8 (57.9%, p ≤ 0.009 and 52.6%, p ≤ 0.02, respectively, Table 2 ) and HLA-DQ2/DQ8 (63.6%, p ≤ 0.009 and 72.5%, p ≤ 0.003, respectively, Table 2 ) T1D patients.