Humanized Mouse Model to Study Type 1 Diabetes

The NOD mouse has been instrumental in deciphering mechanisms involved in type 1 diabetes (T1D), but it fails to cover clinical T1D heterogeneity and shows phenotypic differences with human T1D and limitations when developing T-cell assays or therapies to be applied to humans (1). Immunosuppression has shown not only efficacy in preserving β-cells in patients with recent-onset T1D but also side effects precluding its long-term use (2). Strategies to restore immune tolerance to β-cells should thus be prioritized (3) as preserving responses to unrelated antigens. We lack rodent models to evaluate antigen-specific immunotherapy that directly applies to humans as well as rodent models to explore environmental factors involved in triggering diabetes on conventional genetic backgrounds.

We designed a mouse that expresses both a human β-cell autoantigen and a human antigen-presenting module to characterize autoantigen-derived epitopes that translate to humans. On the basis of evidence that insulin is a key autoantigen, we introgressed a human insulin (hINS) transgene in our model. Indeed, insulin is targeted by autoreactive T cells in T1D (4,5). Anti-insulin autoantibodies are the first to be detected in children at risk and carry a high positive predictive value for T1D in patient siblings (6). The variable number tandem repeat located at 5′ of the hINS gene predisposes to T1D. In the NOD mouse, the lack of either the mouse insulin (mINS) 1 or 2 gene markedly alters the diabetes phenotype (4,7). We selected HLA-A*02:01 and HLA-DQ A1*0301/B1*03:02 (i.e., DQ8) as human MHC genes to be expressed in addition to the hINS gene. HLA-DQ8 carries the highest risk for T1D in man (8) and presumably presents epitopes that drive the autoimmunity to β-cells. Class I HLA-A*02:01 also modifies the risk for T1D and is the most common class I gene expressed in Caucasians (9).

We thus generated humanized mice that lack the expression of murine MHC class I and class II genes and insulin genes and express HLA-A*02:01, HLA-DQ8, and hINS transgenes, hereafter called YES mice. YES mice develop T1D upon injection of polyinosinic-polycytidylic acid (pI:C).

YES mice were obtained by crossing mINS^−/− H-2^k mice that express an hINS transgene under the control of the 353-base pair human insulin gene promoter on a mix C57BL/6JxCBA background (10) with double transgenic HLA-A*02:01/HLA-DQ8 mice carrying a dominant C57BL/6J background. The HLA-A*02:01/HLA-DQ8 mouse was obtained by crossing transgenic H-2D^b mouse-β2-microglobulin (mß2m) double-knockout HLA-A*02:01 (11) and IA^bIE^b double-knockout HLA-DQ8 mice (12,13). It expresses a chimeric HLA-A*02:01 HHD monochain containing the HLA-A*02:01 α1 and α2 and H-2D^b α3 domains linked to human β2m by a 15-mer linker under the control of the HLA-A*02:01 promoter (11) and the HLA-DQ8 α and β-chains under the control of their promoter (13) (Supplementary Fig. 1). Mice were maintained under specific pathogen-free conditions, and experiments were performed following Institutional Animal Care and Use Guidelines accreditation CEEA34.CB.024.11 by the ethics committee of Université Paris Descartes.

Insulin PCRs were performed with REDTaq (Sigma-Aldrich) using primers listed in Supplementary Table 1. Screening for mß2m and human HLA transgenes was performed with Advantage GC 2 Polymerase Mix and PCR Kit (Clontech). Real-time quantitative PCRs that were based on three-point dilutions using QuantiTect SYBR Green PCR Kit (QIAGEN) on a LightCycler 480 (Roche) allowed the discrimination of mice carrying a single mINS1 and/or mINS2 allele for further crosses. Cell surface stainings were performed on blood cells for the loss of H-2D^k and IA^kIE^k using anti-mouse T-cell receptor-allophycocyanin (APC)–labeled and anti-H-2D^k-fluorescein isothiocyanate (FITC)–labeled antibodies (BD Pharmingen) or using anti-mouse CD19-APC–labeled and anti-H-2IA^k/IE^k-FITC–labeled antibodies (BD Pharmingen). Data were collected with an LSRFortessa cytometer and analyzed using FlowJo version 9.2 software (Tree Star).

Thymi from 6-week-old mice were digested with 0.125% collagenase D with 0.1% DNase I followed by two-step digestions with 0.5 units/mL Liberase TM and 0.1% DNase I (14). Thymic epithelial cell (TEC) enrichments were then depleted using mouse CD45 MicroBeads (Miltenyi Biotec) on an LD column according to the manufacturer’s protocol.

Cell suspensions in RPMI medium/FCS 10% (10⁶ cells/mL) were stained with anti-mouse CD3ε-PE, CD3ε-APC, CD8α-PerCP, CD4-efluor450, CD44-FITC, CD45-APC, CD19-APC, CD11c-APC, CD11b-efluor450 or CD11b-APC, anti-human β2m-FITC (clone TÜ99; BD Biosciences), anti-HLA-DQ-Biot (hybridoma SVPL3), streptavidin-FITC, anti-H-2D^k-FITC, anti-H-2IA^k-FITC, anti-mouse NKp46-PE, CD25-PE, FoxP3-FITC, TCRβ-APC, NK1.1-APC, LY6G-PE-labeled, TER119-FITC (homemade), CD45.2-PE or CD90.2-FITC, biotin-conjugated anti-insulin (DAKO), biotin-conjugated anti-mouse CD62L and SAV-PE-Cy7 from eBioscience if not specified. Data were collected with an LSRFortessa cytometer and analyzed using FlowJo version 9.2 software.

Medullary TECs (mTECs) and cortical TECs (cTECs) were sorted with a FACSAria II using an anti-mouse CD45-Pe-Cy7, an anti-H-2IA^b-FITC or anti-H-2IA^k-FITC or anti-HLA-DQ-FITC, depending on the mouse genetic background; an anti-mouse Epcam-APC; an anti-mouse BP1-PE; an anti-mouse UEA-I-biot (Ulex europaeus agglutinin I); and a Brilliant Violet 650 Streptavidin (eBioscience, except for UEA-I-Biot [VECTOR Laboratories]). CD45⁻ MHC class II⁺ Epcam⁺ cells define TECs. BP1 and UEA-I markers define cTECs as BP1⁺ UEA-I⁻ and BP1⁻ UEA-I⁺ as mTECS (14).

After RNAlater stabilization, RNA extractions were performed on solid organs and inguinal lymph nodes (iLNs) with RNeasy Mini Kit and DNase I treatment (QIAGEN). Sorted cTECs and mTECs were lysed in QIAzol Lysis Reagent, and RNA was extracted with miRNeasy Mini Kit and DNase I treatment (QIAGEN). Gene-specific primers were used for reverse transcription on 10–15 ng RNA (Supplementary Table 2). First-round PCRs were performed on 5 μL amplified cDNA (for ßactin, Aire, and Caseinß, 30 cycles/Tm = 54°C; for insulin genes, 30 cycles/Tm = 58°C) (Supplementary Table 3), and second-round PCRs were performed on 2 μL of the PCR1 product (for ßactin, Aire, and Caseinß, 40 cycles/Tm = 54°C; for all insulin genes, 40 cycles/Tm = 58°C) (15) (Supplementary Table 4). Gene expression data have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus and are accessible through Gene Expression Omnibus series accession number GSE101551 (www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE101551).

Mice were immunized against either 100 μg HLA-A*02:01–restricted influenza matrix protein 2 peptide GILGFVFTL (MatA2_58–66) in Complete Freund’s Adjuvant. For hPPI analysis, YES mice were immunized with 100 μg hPPI peptides (Supplementary Table 5) with 140 μg HLA-DQ8–restricted helper Nef_66–97 peptide or with 100 μg hPPI protein (Supplementary Methods 1) in Incomplete Freund’s Adjuvant (IFA) subcutaneously at the base of the tail followed by two recall injections in IFA every other week. Control immunizations were realized by injecting PBS 1× or KLH in adjuvant.

Interferon-γ (IFN-γ)-enzyme-linked immunospot (ELIspot) assays were performed as previously reported (16). Spots were counted using Bioreader 5000 ProSF (BIO-SYS). Data are the mean of triplicate wells and expressed as spot-forming cells per 10⁶ cells, evaluating the background IFN-γ responses in the absence of peptide. Positive controls were cells stimulated by 1 μg/mL ConA (Sigma-Aldrich) and negative controls by irrelevant pyruvate dehydrogenase (PDHase_208–216) peptide. When indicated, responses were inhibited by preincubating splenocytes for 20 min with 50 μg/mL anti-HLA-A*02:01 antibody (BB7.2). For CD4⁺ IFN-γ-ELIspot assays, splenocytes were depleted using mouse anti-CD8 MicroBeads (Miltenyi) on an LD column according to the manufacturer’s protocol. When indicated, responses were inhibited by preincubating CD8⁺ T-cell–depleted splenocytes for 20 min with 50 µg/mL anti-HLA-DQ antibody (hybridoma SVPL3).

In triplicate, 10⁵ spleen cells/well were incubated with 0.5 μg antigen/well or 0.1 μg peptide/well (Supplementary Table 6) for 72 h at 37°C. Proliferation was evaluated with BrdU Cell Proliferation Assay Kit (Cell Signaling) and expressed as the proliferation index. 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. When indicated, the response was inhibited by adding 50 μg/mL anti-HLA-DQ8 antibody (hybridoma SVPL3).

Body weight was monitored weekly for 20 weeks. Intraperitoneal glucose tolerance tests (2 g/kg body weight) were performed after 12–16 h of fasting. Intraperitoneal insulin tolerance tests (0.75 units/kg body weight) were performed by injecting Actrapid (Novo Nordisk) after 4 h of fasting. Blood glucose was measured using a glucometer (BG Star) at 0, 30, 60, and 120 min.

Immunostainings were performed on formalin-fixed paraffin pancreas sections deparaffinized in xylene and dehydrated by ethanol. After washing, antigen retrieval was realized by hot incubation followed by permeabilization (20 min in PBS 1×/0.4% Triton X-100) and saturation (20 min PBS 1×/1% horse serum) before immunostaining with polyclonal guinea pig anti-insulin, polyclonal rabbit antiglucagon (Dako), anti-PDX1, anti-TLR3 (FITC, TLR3.7; Hycult Biotech), or rat anti-human CD3-Biot (AbD Serotec) antibodies overnight. Slides were washed with PBS 1×/1% BSA/0.1% Triton X-100 and stained with an anti-rabbit Ig-FITC antibody, a goat anti-guinea pig-CyA3 (Abcam), or SAV-Cy3 at room temperature. Sections were mounted in VECTASHIELD Antifade Mounting Medium for fluorescence with DAPI (Vector Laboratories). Observations were made with a Zeiss Axio Observer Z1 fluorescence microscope coupled with a Zeiss MRm AxioCam, and pictures were analyzed with ImageJ software. Assessment of β-cell mass was performed on stained microscope slide scans with ImageJ software using a guinea pig anti-hINS antibody (Dako) and calculated as the ratio between β-cell surface (μm²) / pancreas surface (μm²) × pancreas weight (mg) (17).

Six- to 8-week-old mice were daily injected i.p. with 100 μg vac-pI:C (InvivoGen), or 50 μg CpG (ODN 2395; InvivoGen), or 10 μg lipopolysacharide (LPS) (Sigma) for 6 days, tested for glycosuria every 2 days, and diagnosed as diabetic on the basis of blood glucose levels >250 mg/dL. YES mice were injected with six daily doses of 2 mg corticosterone/kg as stress control or with five daily injections of 50 mg streptozotocin (STZ)/kg as diabetes-induced control. To deplete in vivo CD4⁺ and CD8⁺ T cells, we injected 500 μg of GK1.5 and 500 µg of YTS169.4 48 h before the pI:C injection, and we maintain CD4⁺ T-cell depletion for 3 more days (18,19).

YES mice lack the expression of mINS genes and express the hINS transgene in the pancreas and thymus (Fig. 1A and B). The hINS transgene was expressed in thymic mTEC, coexpressed with the AIRE gene, but not expressed in cortical cTEC (Fig. 1C), a pattern similar to that of hINS in parental mice and of mINS2 in C57BL/6 mice (15,20). We confirmed the loss of mouse MHC class I (H-2D^k and H-2K^k) (Fig. 1D, upper panel) and class II (IA^k) (Fig. 1D, lower panel) and the presence of the transgenic HLA-A*02:01 molecule (Fig. 1E, upper panel) on T cells and the transgenic HLA-DQ8 molecule (Fig. 1E, lower panel) on B cells. We observed the expression of HLA-A*02:01 on CD45⁻ insulin⁺ islet cells from YES mice as of murine H-2D^b in C57BL/6 mice (Fig. 1F).

HLA-A*02:01/HHD expression was confirmed on immune cells in spleen, blood, and iLNs by flow cytometry (Supplementary Fig. 2). In all subsets analyzed, HLA-A*02:01 expression was comparable in YES and in parental HLA-A*02:01/HLA-DQ8 double transgenic mice (data not shown). Expression of both HLA class I and class II transgenic molecules was lower than expression of H-2D^b (Supplementary Fig. 2A) and H-2IA^b (Supplementary Fig. 2B) molecules in C57BL/6 mice.

Because the YES genetic background is a mix of C57BL/6 and CBA strains, we characterized the YES genome relative to C57BL/6 and NOD genomes (Supplementary Methods 2). NOD and C57BL/6 mice showed 80.4% identity. YES mice showed 79.3% identity with NOD but 90.9% identity with C57BL/6 mice (Table 1). We studied T1D-associated regions in the YES genome. Mouse linkage regions associated with T1D were listed and updated on the University of California, Santa Cruz, Genome Browser (https://genome.ucsc.edu/index.html). Single-nucleotide polymorphisms (SNPs) that were identical in NOD and C57BL/6 mice were not further considered. Between YES and C57BL/6 mice, 0.21% of SNPs were different. SNPs were listed according to each chromosome and corresponding Idd loci when relevant. We identified SNPs for which at least one variation was identified. Using the Mouse Genome Informatics Web site (www.informatics.jax.org), we found that 86% of SNPs associated to a CBA genetic background. Fourteen percent of SNPs were not described or covered regions for which PCR probes do not discriminate NOD, C57BL/6, and CBA Idd profiles or could not be defined in the absence of a marker for the relevant region (Supplementary Table 7).

Absolute numbers of spleen, iLN, and thymic cells were comparable in YES and in parental HLA-A*02:01/HLA-DQ8 double transgenic mice (data not shown). Distributions of most immune cell subsets were comparable in YES and C57BL/6 mice, with few differences (Fig. 2). Spleen T-cell frequencies were lower in YES than in C57BL/6 mice (13.0 ± 2.4% vs. 29.0 ± 2.1%), with an inverted CD3⁺CD4⁺/CD3⁺CD8⁺ T-cell ratio (0.7 and 1.35, respectively) (Fig. 3A). A significant fraction of cells (9.0 ± 3.9%) were CD3⁻CD4⁻ CD8⁻CD19⁻CD11b⁻CD11c⁻Nk1.1⁻ and not observed in parental HLA-A*02:01/HLA-DQ8 double transgenic mice (Supplementary Fig. 3) but were present in parental hINS transgenic mice (8.0 ± 2.9%). Corresponding cells were hematopoietic cells expressing CD45. In YES spleens, a majority of cells were analogous to double-negative DN3 (CD44⁻CD25⁺) pre-T cells (59.3%), whereas a minority was analogous to DN1 (CD44⁺CD25⁻, 17.2%), DN2 (CD44⁺CD25⁺, 5.2%) pro-T cells, and DN4 (CD44⁻CD25⁻, 16.3%) pre-T cells compared with 3.4%, 32.2%, 3.1%, and 61.2%, respectively, in hINS transgenic mice (21). In YES spleens, those cells were CD62L⁻, suggesting that they were immature cells incapable of homing in peripheral lymph nodes. In iLN cells (Fig. 2B), an inverted CD3⁺CD4⁺/CD3⁺CD8⁺ T-cell ratio was also observed in YES compared with C57BL/6 mice (0.5 and 1.25, respectively), with a slight difference in T-cell frequency (47.0 ± 8.5% vs. 53.45 ± 5.6%). In peripheral blood, T cells were underrepresented in YES compared with C57BL/6 mice (12.0 ± 9.8% vs. 48.0 ± 7.6%; P = 0.0238 by Mann-Whitney test), with an inverted CD3⁺CD4⁺/CD3⁺CD8⁺ T-cell ratio (0.69 and 1.3, respectively). Dendritic cells (DCs) (12.0 ± 1.83% vs. 0.89 ± 1.1%) and macrophages (30.0 ± 8.1% vs. 8.83 ± 3.4%) were overrepresented (Fig. 2C). Thymus analyses showed no difference in YES compared with C57BL/6 mice. The dominant CD3^+low T-cell subset, which mainly consists of CD3^+lowCD4⁺CD8⁺ double-positive cells, was comparable in YES and C57BL/6 mice (90 ± 3.19% and 95 ± 3.09%, respectively) (Fig. 2D). CD4⁺CD25⁺Foxp3⁺ regulatory T cells were not significantly different in YES and C57BL/6 mice.

Because differences were seen in immune cell distributions, we evaluated immune responses to conventional antigens. After immunization against MatA2_58–66, YES splenocytes showed a strong IFN-γ-ELIspot response (Supplementary Fig. 4A) that was inhibited in the presence of the anti-HLA-A*02:01BB7.2 antibody. CD8⁺ T-cell–depleted HLA-DQ8–restricted CD4⁺ spleen T cells from YES mice immunized against KLH showed significant IFN-γ responses (Supplementary Fig. 4B) that were inhibited by an anti-HLA-DQ antibody (P ≤ 0.03). Altogether, these experiments indicate that YES mice respond to conventional antigens as expected.

Because hINS binding to the mouse insulin receptor is in the normal range (22), YES mice are expected to remain normoglycemic. They maintained a normal weight in the long-term range (Fig. 3A and B). Their overall islet morphology was comparable in YES and HLA-A*02:01/HLA-DQ8 parental mice, as were insulin, glucagon, and PDX1 staining (Fig. 3C). The β-cell mass was comparable in YES mice and parental hINS and HLA-A*02:01/HLA-DQ8 mice at 15 weeks of age (Fig. 3D). No significant difference was observed in YES responses to the 2 mg/kg intraperitoneal glucose tolerance test at 6 months of age compared with HLA-A*02:01/HLA-DQ8 parental mice (Fig. 3E). Glycemic responses to 0.75 units/kg i.p. insulin were similar in female YES mice and parental HLA-A*02:01/HLA-DQ8 females, whereas a difference was observed in male mice (Fig. 3F). In the long-term range, YES mice developed neither diabetes nor insulitis.

We previously identified CD8⁺ T-cell responses to hPPI peptides in human T1D (23). After immunization of YES mice against individual HLA-A*02:01–restricted hPPI peptides, a significant response was detected against hPPI_2–11 (P ≤ 0.03) and hPPI_6–14 (P ≤ 0.03) using an IFN-γ-ELIspot assay. A significant, but weak response was detected against hPPI_33–42 (P ≤ 0.04) and against hPPI_101–109 in an individual mouse (Fig. 4A).

YES mice were immunized against hPPI and tested for responses against a panel of hPPI-overlapping peptides. They remained insulitis free upon immunization (data not shown). Significant IFN-γ responses (Fig. 4B) were detected against peptides overlapping the whole hPPI sequence. When considering individual mouse responses, 93.75% of mice showed IFN-γ responses against at least one hPPI peptide (Supplementary Table 8). Observed predominant individual IFN-γ responses were to hPPI_20–35 (62.5% of mice; P ≤ 0.008), hPPI_25–40 (56.25% of mice; P ≤ 0.02), hPPI_46–61 (56.25% of mice; P ≤ 0.02), hPPI_55–70 (68.75% of mice; P ≤ 0.005), hPPI_61–76 (62.5% of mice; P ≤ 0.008), hPPI_92–110 (68.75% of mice; P ≤ 0.005), and hPPI (81.25% of mice; P ≤ 0.0005).

The number of epitopes that led to proliferative responses in vitro after hPPI immunization was restricted compared with IFN-γ responses (Fig. 4C). As for IFN-γ responses, proliferative responses were diversified. After determination of the threshold value for each peptide in YES mice (Supplementary Table 9) using pairs of measurements and the Bland-Altman test (Supplementary Table 10), responses were seen against both hPPI_55–70 peptide (58.3% of mice; P ≤ 0.006) and hPPI (66.6% of mice; P ≤ 0.02) in a significant number of mice. Proliferative responses were seen against at least one hPPI peptide in 91.67% mice upon immunization against hPPI. These experiments show responses that preferentially cluster around a region covering the leader, B-chain, and C-peptide sequences.

We addressed whether YES mice were amenable to T1D induction. After six daily 100-μg pI:C injections (24), diabetes developed within 1 week after the last injection (P ≤ 0.0003) but not after LPS or CpG injection (Fig. 5A). As expected, diabetes occurred after multi-STZ low-dose stimulation (P ≤ 0.0001) but was not induced under cortisol stress conditions that were tested as a control. Administration of anti-CD8 and -CD4 monoclonal antibodies prevented diabetes induced by pI:C stimulation (P ≤ 0.04) (Fig. 5B). An islet infiltration was observed in diabetic mice (Fig. 5C), with islets showing a loss of cells expressing insulin (Fig. 5D) and the presence of CD3⁺ T cells (Fig. 5E). No significant TLR3 expression was observed in control mice that remained untreated (Fig. 5F). TLR3 expression in diabetic pI:C-treated mice showed a speckled distribution within the islet cell cytoplasm, which mostly colocalized with insulin (Fig. 5G). A few glucagon-stained cells showed TLR3 colocalization. Spleen CD8⁺ T cells from diabetic pI:C-treated mice showed IFN-γ responses to hPPI_6–14 (P ≤ 0.016) and hPPI_15–24 (P ≤ 0.02), although not to hPPI_2–11 or hPPI_33–42 as observed in hPPI-immunized YES mice (Fig. 6A). Individual responses were observed against at least one hPPI peptide in 80% of mice. No CD8⁺ T-cell responses were observed in nondiabetic pI:C-treated mice. We next evaluated CD4⁺ T-cell responses in pI:C-treated and control YES mice (Fig. 6B). Diabetic pI:C-treated mice showed significant proliferative responses compared with control YES mice as follows: hPPI_16–30 (P ≤ 0.048), hPPI_18–30 (P ≤ 0.008), hPPI_33–47 (P ≤ 0.05), hPPI_40–55 (P ≤ 0.016), hPPI_46–61 (P ≤ 0.022), hPPI_61–76 (P ≤ 0.031), hPPI_80–97 (P ≤ 0.041), and hPPI_92–110 (P ≤ 0.005) and the hPPI protein (P ≤ 0.011). As for CD4⁺ T-cell responses in hPPI-immunized YES mice, individual responses were observed against at least one hPPI peptide in 86.7% of mice (Supplementary Tables 10 and 11). Responses were against hPPI_1–15 (53.3%; P ≤ 0.02), hPPI_33–47 (46.7%; P ≤ 0.05), hPPI_46–61 (46.7%; P ≤ 0.05), hPPI_70–86 (53.3%; P ≤ 0.02), hPPI_92–110 (73.3%; P ≤ 0.02), and hPPI (53.3%; P ≤ 0.05) (Fig. 6C). Responses clustered in a region covering the leader sequence and C-peptide sequence. A strong correlation was observed between the responses against hPPI protein and most of all hPPI peptides in diabetic pI:C-treated YES mice (Supplementary Fig. 5). In contrast with CD8⁺ T cells, we detected proliferative responses against hPPI peptides in splenocytes from nondiabetic pI:C-treated YES mice (Fig. 6B). All mice also were tested for ZnT8 and GAD T-cell responses (Fig. 6D). Significant proliferative responses were observed against ZnT8_166–179 and GAD_536–550 peptides that are common to humans and mice. Stronger CD8⁺ T-cell (Fig. 7A) and CD4⁺ T-cell (Fig. 7B) responses were observed in pancreatic infiltrates in diabetic versus nondiabetic pI:C-treated YES mice. Infiltrating cells from nondiabetic pI:C-treated mice showed an increased frequency of regulatory T cells but a decreased frequency in CD11b⁺ and DCs compared with diabetic pI:C-treated mice (Fig. 7C). In pancreatic lymph nodes, populations maintained similar frequencies with the exception of DCs expressing the activator marker CD103 specifically in diabetic pI:C-treated YES mice (Fig. 7D).

The clinical management of T1D faces the lack of fully accurate biomarkers of autoimmunity and of efficient and specific immunotherapies. Major improvements in glucose monitoring and insulin therapies have narrowed the mortality gap within the general population and have shifted the safety line that balances risks and benefits toward the safest immunotherapy approaches. Meanwhile, antigen-specific immunotherapy has a long way to go. The autoantigen dose, delivery route, stage of the disease process at which immunotherapy is applied, and use of peptides rather than full-length autoantigens remain open issues in humans (25). Models that would allow direct translation of experimentally defined peptides in man are lacking. In addition, environmental factors that trigger diabetes on a susceptible genetic background remain elusive. We almost exclusively rely on the NOD and MHC class I humanized NOD mice to address such issues (26,27). The NOD mouse shows a unique phenotype that is unlikely to summarize the heterogeneity of T1D mechanisms. It has been selected for spontaneous diabetes development, precluding its use in strategies to explore the triggering of diabetes by environmental factors on conventional genetic backgrounds.

The YES mouse is a new model that lacks mouse class I, class II, and insulin genes and expresses HLA-A*02:01, HLA-DQ8, and hINS. Parental mice carrying the hINS transgene have normal insulin levels and glucose homeostasis (28). Most differences seen between YES and control mice were minor and likely represent the expected interstrain variability. YES mice remain normoglycemic up to 1 year of age, respond normally to an intraperitoneal glucose challenge, and maintain a normal β-cell mass. The distribution of most immune cell subsets in YES mice is comparable to that of a conventional mouse strain. However, T cells were underrepresented in YES mice. Given an increased percentage of CD3⁺CD4⁻CD8⁺ single-positive T cells in the thymus, shifts in peripheral T-cell representation in YES mice is likely to relate with shifted efficiency of human MHC molecules to select T cells. The inverted CD4⁺/CD8⁺ T-cell ratio seen in YES and parental HLA-A*02:01/HLA-DQ8 mice suggests a lower efficiency of class II HLA-DQ8 to select CD4⁺ T cells than of class I HLA-A*02:01 to select CD8⁺ T cells. Because the expression of the insulin gene controls the selection of insulin-specific T cells in the thymus (4,29), we evaluated the expression of the hINS transgene in thymic cells. YES mice exclusively express the hINS transgene in mTECs, a pattern identical to expression patterns observed in C57BL/6 or NOD mice and in humans. Immunization with hPPI induced neither insulitis nor diabetes in YES mice, indicating that they are tolerant to the hPPI transgene.

The main bias toward T1D development in YES mice is the expression of DQ8. The YES genetic background was closer to the C57BL/6 and CBA than to the NOD strain. YES mice were tolerant to the islets and remained insulitis free in the long-term range, as do parental hINS transgenic mice (10). Upon immunization against hPPI or hPPI peptides, YES mice developed CD8⁺ T-cell IFN-γ responses to a diverse repertoire of HLA-A*02:01–restricted peptides but remained insulitis free. On the basis of evidence that enterovirus infections are involved in T1D development (30), we addressed whether it was possible to induce diabetes in YES mice. Upon pI:C injections, YES mice developed acute T1D. Diabetes-free mice were refractory to ultimate T1D development upon a second set of pI:C injections or hPPI immunization, suggesting an on/off mechanism in T1D triggering (31,32). Individual responses were observed against a set of hPPI peptides that partially overlaps with peptides that induced CD8⁺ T-cell responses upon hPPI immunization. Significant responses were seen against leader sequence hPPI_6–14 and hPPI_15–24 peptides, which we previously reported as driving significant expansions of CD8⁺ T cells in patients with recent-onset T1D, pointing to the relevance of the YES model to human T1D (16,23,33).

In contrast with class I–restricted epitopes, HLA-DQ8–restricted CD4⁺ T-cell responses remain ill-defined in man. IFN-γ CD4⁺ T-cell responses observed after immunization of YES mice against hPPI covered a wide spectrum of peptides scattered along the hPPI sequence, indicating the absence of a prevalent response. Proliferative CD4⁺ T-cell responses were observed against a more restricted set of peptides overlapping the B-chain and C-peptide sequences. CD4⁺ T-cells clones obtained from pancreatic infiltrates of a patient with T1D have been characterized as recognizing two HLA-DQ8–restricted C-peptide epitopes that were overlapping with peptide hPPI_61–76 in our model (34). Upon induction of T1D by pI:C, significant CD4⁺ T-cell proliferative responses were observed against hPPI_61–76 and against hPPI peptides that were only partially overlapping with responses induced by immunization of YES mice against hPPI. The partial dissociation of IFN-γ and proliferative responses on one side and of responses induced by immunization against hPPI and upon induction of diabetes by pI:C on the other side was not unexpected. Adjuvant-mediated immunization and pI:C indeed involve different pathways in antigen-presenting cells and CD4⁺ T-cell activation (35–37).

Mechanisms inducing T1D through the loss of immune tolerance to β-cells remain elusive. Our data point to the importance of islet-environment interactions through signals carried by pattern recognition receptors (38,39) and TLRs (toll-like receptors) in the induction of T1D through innate immune upregulation (40,41). This is reminiscent of data involving T1D induction by Coxsackie B4 virus (32,42). The IFN induced with the helicase C domain I (IFIH1/MDA5) gene associates with T1D while mediating the early IFN response to viral RNA (43). Using a TLR3 agonist (44), we induced T1D in YES mice along with a T-cell infiltrate. The β-cells from pI:C-treated diabetic mice showed increased TLR3 expression. On the basis of data indicating TLR3 and MDA5 expression in the islets in vitro (24,45), we analyzed TLR3 expression after pI:C treatment. pI:C itself had a direct effect on β-cells in boosting speckled TLR3 expression predominantly in β-cells. Indeed, a key role of plasmacytoid DCs has been reported in the NOD mouse (32,46).

In conclusion, the YES mouse is a new model to characterize HLA-A*02:01– and HLA-DQ8–restricted hPPI epitopes involved in T1D and study mechanisms of T1D induction and heterogeneity in humans. This model will provide a new avenue to evaluate immune tolerance strategies that may directly apply to immunotherapy of T1D in humans.

Acknowledgments. This work was performed within the Département Hospitalo Universitaire Autoimmune and Hormonal Diseases project. The authors acknowledge Sébastien Jacques and the genotyping platform at Cochin Institute, Paris, France. The authors also acknowledge Raphaël Scharfmann for reading the manuscript and Latif Rachdi and Virginie Aiello-Lorenzo for technical assistance in evaluating β-cell mass and PDX1 staining (all from Cochin Institute, Paris, France).

Funding. This work was supported by Agence Nationale de la Recherche grants R11189KK-RPV11189KKA and ANR2010-Biot-00801 and European Foundation for the Study of Diabetes grant 1-2008-106.

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

Author Contributions. S.L. performed experiments, was involved in discussions, and contributed to writing the manuscript. S.G. performed metabolic experiments. A.G. performed DAB staining. F.Let. was involved in discussions and manuscript editing. B.B. addressed the question of cortisol stress. P.N. performed big data analysis. E.P., M.V., and F.Lem. were involved in Affymetrix genotyping array discussion and manuscript editing. C.B. designed experiments, chaired discussions, and wrote the manuscript. C.B. is the guarantor of this work and, as such, had full access to all data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.