Single Insulin-Specific CD8⁺ T Cells Show Characteristic Gene Expression Profiles in Human Type 1 Diabetes

Type 1 diabetic patients (n = 49) were HLA-A*0201⁺ and positive for anti-GAD, anti-insulin, or anti-IA2 autoantibodies at diagnosis (Table 1). Recent-onset patients (14 males, 11 females; aged 38.1 ± 11.1 years) were studied within 3 months from diagnosis (median 3 days; range 2–70 days). Long-standing type 1 diabetic patients (13 males, 11 females; aged 41.4 ± 11.6 years) had a median type 1 diabetes duration of 4 years (range 2–40). Control subjects were healthy blood donors (12 males, 9 females; aged 35.6 ± 11.8 years), type 2 diabetic patients under oral treatment (3 males, 4 females; mean age 56.6 ± 9.0 years) or insulin treatment (3 males, 4 females; mean age 55.4 ± 7.7 years) selected as expressing HLA-A*0201. Informed consent was obtained and the study was approved by the local ethics committees. HLA class I alleles were determined by Ambisolv genotyping (Dynal/Invitrogen) on DNA with Gentra Puregene Blood kit A (Qiagen). C-peptide was measured by time-resolved fluoroimmunoassay using Wallac Delfia reagents (Perkin-Elmer SAS, Courtaboeuf, France). Lower detection limit was 3.3 pmol/L. The intra- and interassay coefficients of variation at the level 1.1 ng/mL (364 pmol/L) were 3.9 and 5.2%, respectively, and at the level 6 ng/mL (1986 pmol/L) 3.1 and 3.4%, respectively. Intact insulin did not cross-react with C-peptide in this assay.

PPI peptides were selected from PPI sequence, synthetized, and purified as previously described (20) (Table 2) and were controlled by MALDI-TOF (matrix-assisted laser desorption/ionization−top of flight) on a Bruker Protein TOF mass spectrometer (12). Peptide nomenclature refers to N and C termini along the PPI protein sequence. Peptide binding to HLA-A*0201 was determined by major histocompatibility complex stabilization assays using TAP2-deficient HLA-A*0201–transfected T2 cells (21,22). Percent mean fluorescence intensity (MFI) increase corresponds to (MFI with a given peptide – MFI without peptide)/(MFI without peptide). For TMr assembly, HLA-A*0201 α-chain and human β2-microglobulin were produced in E. coli. HLA-A*0201/peptide monomers were refolded in vitro using 12 mg of peptide, purified by gel filtration (Superdex 200HR 10/30, Pharmacia), and biotinylated using the biotin-protein ligase BirA (Avidity) (23). Purified biotinylated HLA-A*0201/peptide monomers were tretramerized with Extravidin-R-Phycoerythrin conjugate (Sigma) (21,24).

Peripheral blood mononuclear cells (PBMCs) were prepared from fresh blood using Lymphoprep (Axis Shield) separation in Leucosep tubes. TMr staining was performed by incubating 10⁶ PBMCs with 5–10 μg/mL phycoerythrin (PE)-labeled TMrs at 37°C for 30 min. Anti–CD45RA-FITC, anti–CCR7-PE-Cy7, anti–CD8-PerCP (SK1 clone), and anti–CD3-APC-Cy7 antibodies (BD Pharmingen, eBioscience) were then added for 15 min at 4°C. After washing, cells were acquired on a BD FACSAria and analyzed using FlowJo (Tree Star) (25,26). The minimal number of TMr⁺ events analyzed in phenotyping studies was 5.10³.

CD8⁺ PBMCs were negatively selected (Dynal/Invitrogen) by depleting CD4-, CD14-, CD16-, CD19-, CD36-, CD56-, CDw123-, and CD235a-expressing cells. Staining was performed on 10⁶ CD8-enriched cells by adding 10–20 μg/mL of nonpooled PE-labeled TMr at 37°C for 30 min. Anti–CD8-PerCP–labeled antibody was then added for 15 min at 4°C. Small lymphocytes were gated according to forward/side scatter profiles. CD8⁺/TMr⁺ T cells were sorted on a FACSAria at 1 cell/well into 96-well PCR plates containing 5 μl PBS treated with diethyl pyrocarbonate (Sigma) and immediately frozen at −80°C.

RNA was extracted from sorted cells by direct cellular lysis for 2 min at 65°C. Procedures and primers for coamplification of multiple genes in single cells were designed as described (27,28). Reverse transcription was carried out with murine leukemia virus reverse transcriptase (Applied Biosystems) for 60 min at 37°C. Seminested PCR was then performed with gene-specific primers (Eurogentec) and AmpliTAQ Gold Polymerase (Applied Biosystems) by touch-down PCR. PCR products were resolved on a 2.5% (w/v) agarose gel. Perforin (Prf1), granzyme A (Gzma), granzyme B (Gzmb), Fas ligand (Fasl), IFN-γ (Ifng), TGF-β1 (Tgfb1), TNF-α (Tnfa), IL-2 (Il2), IL-10 (Il10), IL-10Rα (Il10ra), IL-7R (Il7r), MIP-1α (Mip1a), MIP-1β (Mip1b), PD1 (Pd1), RANTES (Ccl5), KLRG1 (Klrg1), CCR7 (Ccr7), and CD3ε (Cd3e) mRNAs were analyzed.

Comparison of distribution scores between patients and control subjects and TMr assessment used the nonparametric Kruskal-Wallis test or Fisher exact test. To study assay reliability, we used the graphical Altman and Bland method, which focuses on means and variability of differences between repeated measurement pairs (29). C-peptide levels were compared using the Mann-Whitney U test.

PPI peptides are recognized by CD8⁺ T cells from type 1 diabetic patients using IFN-γ ELISpot (12–15), which, methodologically, restricts recognition to a functional subset of T cells, precluding characterization of their differentiation stage. To characterize PPI-specific CD8⁺ T cells independently of their function, we prepared HLA-A*0201 PE-labeled TMrs complexed with previously defined PPI peptides (A2/PPI) (Table 2) (12–15). Stable TMrs were obtained with peptides that showed a percent MFI >0.5 in HLA-A2 stabilization assays (Table 2). HLA-A2 TMrs complexed with CMV peptide pp65_495–503 (A2/CMV TMrs) and PDHase_208–216 (A2/PDHase TMrs) were used as positive and negative controls, respectively. A2/PPI TMrs⁺ CD8⁺ T cells were detected (Fig. 1A–E) in 35/49 type 1 diabetic patients (71.43%). Detection thresholds for each individual TMr⁺ were set at the 95th percentile of values obtained in controls. Percentages of TMr⁺ CD8⁺ T cells above 95th percentile of controls were considered positive. (Table 2). No CD8⁺ T cells were stained with A2/PDHase TMrs (Fig. 1F) or in absence of TMrs (Fig. 1G). As expected, more cells were often stained with HLA-A2/CMV TMrs than with A2/PPI TMrs (Fig. 1H).

Significant differences were observed among recent-onset, long-standing type 1 diabetic patients, and control subjects in the frequencies of TMr⁺ CD8⁺ T cells specific for PPI leader sequence peptides PPI_6–14, PPI_15–24, and B-chain peptide PPI_33–42 (B_9–18; Fig. 2A). Percentages of CD8⁺ T cells specific for leader sequence peptide PPI_2–11, B chain peptides PPI_30–39, PPI_34–42, and PPI_42–51, and A chain peptide PPI_101–109 did not differ significantly among groups, but few individual responses were seen against these peptides (Fig. 2A and Supplementary Table 1). There was no significant difference in PPI-specific CD8⁺ T cells between healthy controls and type 2 diabetic patients, including with insulin-treated Type 2 diabetic patients (data not shown), and in CMV-specific CD8⁺ T cells between type 1 diabetic patients and control subjects (Fig. 2A). The number of individuals whose TMr⁺ frequencies were ≥95th percentile of control subjects was different between recent-onset patients and control subjects for PPI_6–14 TMr⁺ cells (8/25, 32.0%; P < 0.03) and PPI_15–24 (8/25, 32.0%; P < 0.03) but not for PPI_33–42 (7/25, 28.0%; P = 0.06) and between long-standing type 1 diabetic patients and control subjects for PPI_33–42 (12/24, 50.0%; P < 0.002) but not for PPI_6–14 (9/24, 37.5%; P = 0.35) and PPI_15–24 (7/24, 29.2%; P = 0.055). In a healthy control subject, a high (0.14%) frequency of PPI_6–14–specific CD8⁺ T cells was observed (Fig. 2A).

Assay reliability was calculated from values obtained from two assays performed in two different samples collected at median intervals of 3 days following diagnosis (range 2–90 days) in the same individuals using PPI_6–14, PPI_15–24, PPI_33–42, and CMV pp65_495–503 TMrs. Differences of up to ±0.038, 0.042, 0.060, and 0.186% were observed for PPI_6–14, PPI_15–24, PPI_33–42, and CMV pp65_495–503 TMrs, respectively. Mean differences represent interassay mean errors and intervals acceptable errors. There is good interassay agreement for PPI_6–14, PPI_15–24, and PPI_33–42 peptides, each value being included in the error interval, which is ≤0.05 (Fig. 2B). Overall, 68% of recent-onset and 83.3% of long-standing type 1 diabetic patients had detectable PPI-specific CD8⁺ T cells above the 95th centile of control subjects. Responses were observed against at least two peptides in 44.9% of patients (22/49), including 32.6% (16/49) who showed a response to either PPI_6–14, PPI_15–24, PPI_33–42.

In long-standing patients who stained positive for A2/PPI TMrs, mean diabetes duration was 6.5 years (range: 3–25) for PPI_6–14 peptide, 14.4 years (3–39) for PPI_15–24 peptide, and 10.7 (3–39) for PPI_33–42 peptide. Basal C-peptide values were available in 11/19 patients in whom increased A2/PPI_33–42 TMr⁺ CD8⁺ T cells (TMr⁺ patients) were detected and 7/30 patients in whom A2/PPI_33–42 TMr⁺ CD8⁺ T cells were in the basal range (TMr⁻ patients). Although differences between both groups were not significant, lower C-peptide values were observed in TMr⁺ patients (median 0.07 ± 0.09 nmol/L; range 0.01–0.27) than in TMr⁻ ones (median 0.22 ± 0.19 nmol/L; range 0.01–0.59; P = 0.07) (Fig. 3A). Moreover, an inverse correlation was observed between percentages of A2/PPI_33–42 TMr⁺ CD8⁺ T cells detected and fasting C-peptide values (r = −0.63; P = 0.05) (Fig. 3B).

Upon antigen encounter, CD8⁺ T cells express differentiation programs that imprint their homing, survival, activation, and functions. Detection of circulating TMr⁺ CD8⁺ T cells allows characterizing membrane markers, especially homing and signaling markers that discriminate effector and memory cells from naïve cells. PPI-specific CD8⁺ T cells were gated to evaluate percent TMr⁺ cells and define (Fig. 4) (30) CD45RA⁺CCR7⁺ naïve T cells (TN), CD45RA⁻ CCR7⁺ central memory T cells (TCM), CCR7⁻CD45RA⁻ effector memory T cells (TEM), and CCR7⁻ CD45RA⁺ effector memory T cells (TEMRA) (Fig. 4A and B). Significant differences were seen in the distribution of CMV-specific and PPI-specific CD8⁺ naïve and memory T cells between type 1 diabetic patients and control subjects using the Kruskal-Wallis test, but not between recent-onset and long-standing type 1 diabetic patients (Fig. 4C). In recent-onset and long-standing type 1 diabetic patients, 31.5 and 32.0% TN PPI-specific CD8⁺ cells were detected, respectively, whereas these cells were almost undetectable among CMV-specific T cells. Similarly, PPI-specific TCMs represented 25.4 and 29.8% of CD8⁺ T cells in the two respective type 1 diabetic populations, whereas they represented only 15.6% of CMV-specific CD8⁺ T cells (P = 0.0071). In contrast, PPI-specific TEM and TEMRA were underrepresented in recent-onset (8.19%) and long-standing (12.1%) patients, as compared with CMV-specific TEM and TEMRA CD8⁺ T cells (50.47 and 29.97%, respectively). There was no difference in distributions of PPI_6–14–, PPI_15–24–, PPI_33–42–specific CD8⁺ T cells. Altogether, these data show that PPI-specific CD8⁺ memory T cells are predominantly TN, TCM, and TEM cells, whereas the majority of CMV-specific CD8⁺ T cells are TEM and TEMRA cells. We observed no correlation between diabetes duration and percentage of PPI-specific CD8⁺ TN or TCM T cells on long-standing type 1 diabetic patients (data not shown).

Cell surface phenotypes often fail to ascribe functional attributes to memory CD8⁺ T cells (25,31,32). To address how differentiation programs are established at clinical onset and in long-standing type 1 diabetes, we studied 18 genes coding for inflammatory chemokines, cytokines, cytotoxic molecules, and receptors involved in CD8⁺ T-cell responses. Gene expression was evaluated in TMr⁺ CD8⁺ T cells and nonpurified TMr⁺ CD8⁺-naïve and central memory cells. Cells were obtained by purifying CD8⁺ T cells, then sorting TMr⁺ cells from patients in whom a minimum of 0.05% TMr⁺ CD8⁺ T cells were detected. In these patients, the percentage of cells expressing the CD3e gene was 82.06 ± 14.35%, in contrast with patients who did not reach this threshold (16.50 ± 8.96%). Gene expression profiles were heterogeneous between single cells carrying the same peptide specificity, but a significant percentage of single cells coexpressed genes involved in the same functional program (Fig. 5). Gene expression in PPI-specific (Fig. 5A) and CMV-specific (Fig. 5B) CD8⁺ T cells was different. Figure 5A illustrates a representative subject (R18) displaying 0.19% A2/PPI_6–14 TMr⁺ CD8⁺ T cells of which only 4.3% coexpressed Pfr and Gzma or Gzmb, none coexpressed Ccl5 (coding for RANTES) and Mip1b, but 40% expressed Ccr7. Opposite profiles were seen in A2/CMV-specific CD8⁺ T cells. In patient L9 (Fig. 5B), 2.0% A2/CMV TMr⁺ CD8⁺ T cells were detected of which 36% coexpressed Pfr and Gzma or Gzmb but only 4% expressed Ccr7, while 32% coexpressed Ccl5 and Mip1b.

Gene expressions were strikingly comparable in A2/PPI_6–14, PPI_15–24, and PPI_33–42 TMr⁺ CD8⁺ T cells (data not shown). Considering the whole type 1 diabetic population, the mean percentage of PPI-specific CD8⁺ T cells that expressed Pfr and Gzma and Gzmb was 16.1, 11.0, and 6.9%, respectively, as compared with 51.5, 40.7, and 21.4% for CMV-specific CD8⁺ T cells (P = 0.0003, P = 0.0092, P = 0.0125, respectively; Fig. 5D). Differences in chemokine gene expression were also seen, such as Ccl5 and Mip1b, which drive immune cells to inflammatory sites. Indeed, a higher number of CMV-specific CD8⁺ T cells expressed Ccl5 compared with PPI-specific CD8⁺ T cells (81.9 vs. 27.8%; P=0.0026), and the same result was observed for Mip1b (20.6 vs. 6.0%; P = 0.01). By contrast, Ccr7, which controls recirculation through lymph nodes, was more expressed by PPI-specific CD8⁺ T cells (30.9%) than CMV-specific CD8⁺ cells (3.4%; P < 0.001). These data suggest that PPI-specific CD8⁺ T cells display gene expression programs that largely overlap TCM rather than TN phenotypes, based on expression of IFN-γ and Rantes genes (33), whereas CMV-specific CD8⁺ T cells are predominantly TEM cells. Expression of Fasl, Tnfa, Il2, Mip1a, and Il10 genes was remarkably rare. Expression of the Il7ra gene was above 70% in both A2/CMV and A2/PPI peptide TMr⁺ CD8⁺ T cells. In contrast with previous reports (34), the expression of Klrg1 by CMV-specific CD8⁺ T cells was heterogeneous, whereas it was low in most PPI-specific CD8⁺ T cells. A significant difference was further observed in expression of Pd1 between PPI- and CMV-specific CD8⁺ T cells (3.0 vs. 11.1%; P = 0.0117). There was no significant difference in gene expression levels by PPI-specific CD8⁺ T cells between recent-onset and long-standing type 1 diabetic patients (data not shown). In the only control individual who showed detectable PPI_6–14–specific CD8⁺ T cells (Fig. 2A), a unique gene expression program was observed, with 34% CD8⁺ T cells expressing Il10, as compared with 2.2 ± 3.3% of PPI_6–14–specific CD8⁺ T cells in type 1 diabetic patients.

HLA class I TMrs may prove instrumental in characterizing T cells in autoimmunity. Following previous identification of PPI-specific CD8⁺ T cells in type 1 diabetes using IFN-γ ELISpot (12), we here report the presence of circulating PPI-specific CD8⁺ T cells in type 1 diabetes. PPI-specific CD8⁺ T cells were detected against three major peptides, two within the PPI leader sequence (PPI_6–14 and PPI_15–24) and one within the insulin B chain (PPI_33–42; also known as B_9–18). CD8⁺ T cells that were specific for other PPI peptides were only seen in few patients. Overall, CD8⁺ T cells specific for at least one PPI peptide were detected in 68% of recent-onset and 83% of long-standing type 1 diabetic patients. These figures are likely underestimated, since stable TMrs were only obtained with peptides that showed sufficient binding affinity to HLA-A*0201 (Table 2). However, we cannot rule out that PPI-specific T cells not tested here are activated in type 1 diabetes and that different peptides are recognized at different stages of type 1 diabetes because we only tested individuals with clinical diabetes.

HLA-A*0201–restricted leader sequence peptides are commonly presented to CD8⁺ T cells, as previously reported for PPI_6–14 and PPI_15–24 (12,15). PPI_6–14– and PPI_15–24–specific CD8⁺ T cells have previously been shown to be cytotoxic against human islet T cells and HLA-A*0201–transfected P815 target cells, respectively (13,17). Recognition of PPI_15–24 has been reported to increase when human islets were incubated in the presence of high glucose concentrations (17), adding further insight into its role in the natural history of human type 1 diabetes.

The differentiation of memory T cells in relation to antigen exposure has been extensively studied in immune responses to viruses but remains poorly understood in autoimmunity. Clinical diabetes occurs after a preclinical phase that spans over years in many patients. However, the β-cell mass in prediabetes and the level of autoantigen exposure once diabetes is diagnosed remain unknown (35,36). The persistence of CMV infection with time makes it a relevant control, since T cells are kept exposed to low levels of CMV antigens, as is the case for autoimmune T cells exposed to residual β-cells (36). We limited our study of membrane markers of PPI-specific CD8⁺ T cells to CCR7 and CD45RA, which discriminate central memory from effector memory CD8⁺ T cells. However, CD45RA does not fully correlate with effector memory functions. Markers such as CD27 and CD28 further discriminate at least three and four different CD8⁺ subsets within CCR7⁻ CD45RA⁻ and CCR7⁻CD45RA⁺ CMV-specific cells, respectively (25). These markers were, however, less relevant to study PPI-specific CD8⁺ T cells as these cells were mostly within the CD45RA⁻ CCR7⁺ central memory compartment. CD45RA⁺CD45RO⁺ double-positive T cells have been identified as possibly representing a transition population from naïve CD45RA⁺ to memory CD45RO⁺ T cells. This population has been shown to be increased in recent-onset type 1 diabetic patients but not in long-standing patients, possibly reflecting an imbalance between the two CD45 isoforms in autoimmunity (37). Because we did not exclude CD45RA⁺CD45RO⁺ T cells from the analysis of memory cells, we cannot exclude that we lost these transition cells that are likely to be included in naïve cells in our analysis. However, we saw the same increase in memory PPI-specific CD8⁺ T cells in long-standing type 1 diabetic patients in whom the double-positive CD45RA⁺CD45RO⁺ T cell is not increased. A recent work has reported the predominant detection of TN and TEMRA CD8⁺ T cells, but they were specific for different peptides, i.e., PPI_2–10 and PPI_34–42, with higher percentages of detection and in subjects who had possibly been treated with anti-CD3 antibody (38).

Analysis of postactivation CD8⁺ T cells actually illustrates much greater diversity in survival, recall potentials as well as tissue specificity than is defined by current phenotypic markers (32). We turned to single cell PCR to study gene expression programs in CD8⁺ T cells. In contrast to CMV-specific CD8⁺ T cells that showed effector/effector memory phenotypes, PPI-specific CD8⁺ T cells showed profiles that overlap with those of central memory cells. Gene expression profiles of PPI_6–14– and PPI_15–24–specific CD8⁺ T cells in recent-onset type 1 diabetic patients as well as PPI_33–42– and PPI_15–24–specific CD8⁺ T cells in long-standing type 1 diabetic patients were identical. Similarities in gene expression patterns of CD8⁺ T cells specific for PPI leader sequence peptides and B chain PPI_33–42 are remarkable because exposition to B chain epitopes persists in insulin-treated, long-standing, type 1 diabetic patients independently of the persistence of residual β-cells. However, this may not be an exclusive mechanism, since PPI_15–24–specific CD8⁺ T cells were detected in a subset of patients with long-standing diabetes, and thus independently of insulin therapy because insulin preparations are devoid of PPI leader sequence. No correlation was observed between detection of PPI_15–24–specific CD8⁺ T cells and type 1 diabetes duration. Of further note, a significant fraction of PPI-specific T cells displayed a naïve status that was virtually absent in the CMV-specific repertoire, suggesting that there is still an unexpressed potential for further autoimmune activation in most patients. This unexpressed potential may explain the arousal of PPI_33–42–specific CD8⁺ T cells upon insulin treatment in long-standing patients. Indeed, these cells were not detected in type 2 diabetic patients treated with insulin, suggesting that insulin treatment may effectively expand them only in the presence of preexisting autoimmunity. The detection of naïve PPI-specific CD8⁺ T cells may indicate their preexistence and maintenance in the repertoire of type 1 diabetic patients in relation with the slow chronic process that characterizes type 1 diabetes. Adding further value to the extensive functional characterization carried out by single-cell PCR, one single healthy subject harbored PPI_6–14–specific CD8⁺ T cells expressing the regulatory cytokine Il10. Thus, the mere presence of autoreactive CD8⁺ T cells may not be considered pathogenic, as previously suggested for islet-reactive CD4⁺ T cells (39).

We observed a differentiated CMV-specific CD8⁺ T cell phenotype, as previously reported (25), with low expression of Ccr7 and high expression of cytolytic effector molecules and significant diversity among individuals. By contrast, PPI-specific CD8⁺ T cells expressed higher levels of Ccr7—presumably allowing them to repeatedly circulate to lymph nodes rather than directly homing to the islets—and lower levels of cytolytic effector molecules. An important factor in driving distinct memory T-cell phenotypes is the strength of T-cell activation in relationship with local factors in secondary lymphoid organs in which the initial T-cell activation takes place and at the inflammatory site (32,40). Expression of homing receptors for lymph nodes have usually been associated with situations in which low strength of activation or a low antigen load is seen (41). Although in most instances antigen persistence is not required for maintenance of memory T cells, recent data in type 1 diabetes suggest that long-standing type 1 diabetic patients may still harbor residual β-cells (42,43).

The detection of B chain–specific CD8⁺ T cells in a significant fraction of long-standing type 1 diabetic patients indicates that memory CD8⁺ T cells persist in the long-term range despite the likelihood that autoimmunity to β-cells is progressively downregulated, since residual β-cells disappear. This persistence of detectable central memory CD8⁺ T cells in the long term raises several nonexclusive hypotheses. They may persist in the absence of residual β-cells or in the presence of a minimal residual β-cell mass, either resisting autoimmune destruction or regenerating as destruction proceeds, as proposed in the NOD model (44). The persistent detection of PPI_6–14– and, to a larger extent, PPI_15–24–specific CD8⁺ T cells in few long-standing patients is compatible with either hypothesis. Alternatively, long-term treatment with exogenous insulin may contribute to activation of T cells against B-chain peptides. This is relevant in the light of the clinical observation that type 1 diabetes is a highly recurrent disease in patients who have been treated with exogenous insulin for years. The negative correlation observed between the percentage of A2/PPI_33–42 TMr⁺ CD8⁺ T cells detected in type 1 diabetic patients and residual basal C-peptide values is compatible with this hypothesis.

In conclusion, a majority of type 1 diabetic patients harbor PPI-specific CD8⁺ central memory T cells. The absence of these cells in most control individuals makes a strong case for the role of PPI-specific CD8⁺ T cells in type 1 diabetes. Although type 1 diabetes is characterized by active autoimmunity many years after disease onset, our observation that exogenous insulin may contribute to expansion of autoreactive CD8⁺ T cells in the long term brings a new light to the natural history of the disease.

This work was supported by Agence Nationale de la Recherche grants COD-2005 MELTD1 and PHRC AOR 05034/P051078 (C.B.), the Juvenile Diabetes Research Foundation (JDRF; grant 1-2008-106), the European Foundation for the Study of Diabetes (EFSD/JDRF/Novo Nordisk European Programme in Type 1 Diabetes Research 2007), the INSERM Avenir Program (R.M.), and the Ile de France CODDIM. S.L. was supported by an Applied Research and Development grant.

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

S.L. performed experiments and contributed to writing the manuscript. J.-P.B., S.M., N.L., and B.R. performed experiments and contributed to discussions. J.C. performed statistical evaluations and reviewed and edited the manuscript. E.L. was responsible for patient recruitment and follow-up. F.L. and R.M. were involved in discussion and manuscript editing. C.B. designed experiments, chaired discussions, and wrote the manuscript.

Sincere thanks are expressed to Unité de Recherche Clinique of Cochin for their support for this study.