The Frequency and Immunodominance of Islet-Specific CD8⁺ T-cell Responses Change after Type 1 Diabetes Diagnosis and Treatment

The HLA-A2–restricted β-cell epitopes used (Table 1) have previously been described (17,18,31). A viral peptide pool of Flu MP_58–66 (GILGFVFTL), Epstein-Barr virus BMLF_280–288 (GLCTLVAML), cytomegalovirus pp65_495–503 (NLVPMVATV), and phytohemagglutinin (1 μg/ml; Sigma, Lyon, France) were used as positive controls. HIV gag_77–85 (SLYNTVATL) and DMSO diluent were included as negative controls. All peptides were used at 10 μmol/l and were >80% pure (Schafer-N, Copenhagen, Denmark). Responses against these 9- to 10-amino acid–long epitopes originate from CD8⁺ T-cells, as previously shown (18).

HLA-A2⁺ new-onset adult (aged >16 years) type 1 diabetic patients with acute onset of symptoms requiring permanent insulin treatment from the time of diagnosis (33) were recruited from the Diabetes Registry of the Province of Turin, Italy, and from the GOFEDI Network, France (see appendix for a detailed list of participating centers). All patients had metabolically controlled disease and were free of recent (<2 weeks) infectious or inflammatory conditions at the time of blood draw. Parallel recruitment of Caucasian healthy control subjects took place at the same institutions. Available subjects wishing to continue the study were subsequently drawn and tested a second time under identical conditions. All subjects gave informed consent, and the study was approved by the relevant ethics committees.

All blood samples were shipped overnight at room temperature. Rapid HLA-A2 screening was performed with the BB7.2 monoclonal antibody, followed by subtyping using the Olerup SSP HLA*02 kit (GenoVision/Qiagen, Vienna, Austria) to verify the presence of the HLA-A*0201 allele (>95% of HLA-A2⁺ individuals). PBMCs were isolated by density gradient centrifugation using lymphocyte separation medium (PAA, Les Mureaux, France), and immediately used or stored frozen (10% DMSO in pooled human male AB serum). The second testing was performed under the same condition (i.e., fresh or frozen PBMCs) of the first testing.

Serum GAD, IA-2, and insulin (auto)antibodies were measured by radioligand binding assays, following protocols previously evaluated within the Diabetes Antibody Standardization Program (lab no. 137) (34). Sensitivities in the 2005 Diabetes Antibody Standardization Program were 84% for anti-GAD, 76% for anti–IA-2 and 28% for anti-insulin, where the specificity was set at 95%. Changes >33% between two aAb titer determinations were considered significant.

Ninety-six–well polyvinylidine fluoride plates (Millipore, Saint-Quentin-en-Yvelines, France) were coated overnight with an anti–IFN-γ antibody (U-CyTech, Utrecht, the Netherlands). Plates were subsequently blocked with RPMI plus 10% human serum (PAA) and peptides added (10 μmol/l final concentration) in triplicate wells along with recombinant human interleukin-2 (0.5 units/ml; R&D Systems, Lille, France), as previously described (18). PBMCs were seeded at 3 × 10⁵ cells/well and cultured for 20–24 h. Following PBMC removal, IFN-γ secretion was visualized with a biotin-conjugated anti–IFN-γ antibody (U-CyTech), alkaline phosphatase-conjugated ExtrAvidin, and Sigmafast 5-bromo-4-chloro-3-indolyl phosphate/nitro blue tetrazolium (BCIP/NBT) tablets (both from Sigma). Spots were counted using an AID reader (Strassberg, Germany) and means of triplicate wells calculated. All ISL8Spot readouts are expressed as spot-forming cells (SFCs)/10⁶ PBMCs.

The cutoff for a positive response was set at 3 SDs above the average basal reactivity (i.e., reactivity against HIV gag_77–85 and DMSO diluent alone). This was chosen as the cutoff allowing for the best diagnostic sensitivity (i.e., highest number of positive responses to β-cell epitopes in type 1 diabetic patients) and specificity (i.e., lowest number of positive responses in healthy controls), as determined by receiver-operator characteristics analysis (18).

A flu MP_58–66-specific HLA-A2–restricted T-cell clone was serially diluted into PBMCs from three different HLA-A2⁺ donors showing no response to the MP_58–66 peptide and displaying different levels of basal reactivity in the absence of stimuli. These cell mixtures were challenged in the presence of 10 μmol/l flu MP_58–66 peptide using the same ELISpot format as that described above.

Values are expressed as means ± SD or medians (range) according to their distribution. Comparisons between proportions were made with the χ² test and Fisher's exact test when appropriate. Comparisons of means between two groups were carried out with Student's t test for normal distributed variances or with the Mann-Whitney U test for non-normal variances. P < 0.05 was considered of statistical significance.

The reproducibility of our ELISpot technique has previously been reported (18). Variability was found to be of 14.1% intra-assay, 4.2% at the analytical interassay level (i.e., using thawed PBMC samples frozen on the same occasion), and 9.2% at the preanalytical and analytical level (i.e., using separate blood draws from the same donor). Another problem encountered in a longitudinal setting is that the basal reactivity level (i.e., spontaneous IFN-γ production in the absence of stimuli) can be different not only among different subjects but also within the same individual when analyzed at different time points. To address whether these variations could affect the detection sensitivity of the technique, we took HLA-A2⁺ healthy donors displaying different IFN-γ background levels that demonstrated no response to the Flu MP_58–66 epitope, mixed their PBMCs with serially diluted numbers of a flu MP_58–66-specific clone, and tested these mixtures in ELISpot against the MP_58–66 peptide. As shown in Fig. 1, background levels more than 10-fold different (7.3, 21.7, and 85.5 SFC/10⁶ PBMCs, respectively) did not significantly affect the detection sensitivity of the system (i.e., the basal-subtracted net signal) over a wide range of frequencies. Moreover, all counts were above the basal ±3 SD positive cutoff value calculated from the respective backgrounds.

We considered β-cell epitopes that we previously defined as targets of CD8⁺ T-cells in 19–50% of adult new-onset type 1 diabetic patients but not of age-matched healthy control subjects (17,18). These epitopes were derived from four different β-cell Ags, namely PPI, GAD, IA-2, and IGRP. A complete list is presented in Table 1.

Fifteen new-onset type 1 diabetic patients were available for follow-up within a median of 11 months (range 7–16) (Table 2). Their mean age at diagnosis was 27.6 ± 8.4 years, and their median type 1 diabetes duration at the time of the first blood draw was 13 days (2–180). Fifteen age-matched healthy control subjects (mean age 28.8 ± 5.9) were also tested twice within a median follow-up period of 14 months (12–26).

Results expressed as number of SFC/10⁶ PBMCs are summarized in Fig. 2. All healthy control subjects were negative at follow-up, including two individuals (i.e., H08 and H09) previously found to be weakly positive for 1–2 epitopes. The number of type 1 diabetic patients positive for T-cell responses dropped from 80% (12 of 15) at diagnosis to 53% (8 of 15) during follow-up (P = 0.12). Moreover, while positive patients at the time of diagnosis recognized multiple epitopes (median reactivities/patient 2 [range 0–6]), there were mostly single reactivities that persisted during follow-up (median 1 [range 0–3], P = 0.05). Overall, the number of positive responses out of the total tested decreased from 28.6% (34 of 119) to 13.2% (16 of 121) (P = 0.003).

Those patients with detectable reactivities could be divided in two categories: on one side, patients already positive at diagnosis (i.e., patients P04, P09, P10, P11, P14, and P15) in whom at least one of the original epitopes was still recognized, while new ones sometimes emerged (i.e., P04, P11, and P14). On the other side, patients whose responses were all negative at diagnosis and who developed scattered reactivities at follow-up (1–2 epitopes recognized; patients P06 and P08). CD8⁺ T-cell responses against a pool of viral epitopes were included as internal controls. They decreased or disappeared only in two patients (P08 and P11), remaining unchanged in all other instances (13 of 15 [87%]).

A simplified graphical representation of these data are provided in Fig. 3, where only CD8⁺ T-cell responses testing positive either at diagnosis or follow-up are represented. Such responses are displayed as absent (<3 SDs above the mean basal reactivity), low (3–4 SDs), intermediate (4–5 SDs), or high (>5 SDs), following the quantitative ranking reported in Fig. 2.

Of the total positive T-cell responses to islet epitopes observed at diagnosis, the majority (26 of 42 [62%]) dropped to nondetectable levels, while few of them (8 of 42 [19%]) remained detectable at follow-up. Only in five patients (i.e., P04, P06, P08, P11, and P14) did new epitopes become targeted (8 of 42 [19%]). Except for patient P04 who recognized proinsulin_B18–27, in all other instances (7 of 8 [87.5%]) the newly targeted epitopes were IA-2_206–214 and IGRP_265–273. These two epitopes were recognized together in three of four patients.

While the number of patients responding to epitopes derived from PPI and GAD significantly decreased (60 to 20% for PPI and 67 to 20% for GAD; P < 0.02) (Fig. 4A), IA-2_206–214 became the major immunodominant epitope detected at follow-up. This was true both in terms of IA-2_206–214–responding patients (5 of 14 [36%]) (Fig. 4B) and of frequency of IA-2_206–214 among all targeted epitopes (5 of 16 [31%]) (Fig. 5). Second in order came IGRP_265–273, which was recognized by 27% (3 of 11) of patients (Fig. 4B), thus constituting 19% (3 of 16) all reactivities detected (Fig. 5). Another 20% of patients and 19% of reactivities were covered at follow-up by GAD_114–123 and proinsulin_B18–27, which were the two immunodominant epitopes at diagnosis (53 and 36% prevalence and 23 and 14% relative frequency, respectively). Thus, four epitopes (i.e., the newly immunodominant IA-2_206–214 and IGRP_265–273 along with GAD_114–123 and proinsulin_B18–27) accounted for 88% of all reactivities detected at follow-up. Three epitopes (proinsulin_B10–18, GAD_536–545, and IGRP_228–236) were instead no longer recognized. This focusing of T-cell responses toward more selected epitopes contrasted with the situation at diagnosis (Fig. 4B and Fig. 5) where, besides GAD_114–123 and proinsulin_B18–27 emerging as the immunodominant specificities, the other seven epitopes were evenly ranked (13–29% prevalence; 6–12% of total reactivities).

A sharp dichotomy was observed when comparing the progression of aAb titers and T-cell responses (Table 2). While the fraction of positive ISL8Spot responses decreased from 28.6 to 13.2% (P = 0.003), that of positive GAD or IA-2 aAb responses did not change (63.3 vs. 60.7%; P = 0.83). Even the titers of GAD and IA-2 aAbs remained nearly identical in the majority of cases (21 of 28 [75.0%]). A decrease in titers was registered only in 17.9% (5 of 28) of case subjects, while both GAD and IA-2 further rose in only one patient.

The picture was different when we compared (auto)antibody and T-cell responses against PPI. In line with their adult onset type 1 diabetes, all but two (13 of 15 [86.7%]) patients were negative for insulin autoantibodies at diagnosis (Table 2). Following initiation of insulin therapy, insulin autoantibodies titers became positive in most patients, with the exception of P03 and P10. P10 was also one of three patients for whom proinsulin-specific T-cell responses were still detectable at follow-up.

Here, we describe the progression of β-cell–directed CD8⁺ T-cell responses after type 1 diabetes clinical onset and initiation of insulin therapy. As the aim of this study was the longitudinal evaluation of type 1 diabetic patients rather than their comparison with healthy control subjects, the fact that these two groups were matched only with respect to age does not bias our conclusions. In this longitudinal setting, it is instead essential to distinguish between changes in T-cell responses due to the variability of the ISL8Spot technique from those reflecting true biological fluctuations. In this respect, the 9.2% interassay variability registered when analyzing separate blood draws from the same individual (18) is reassuring. Although some fluctuations in the ELISpot background noise are unavoidable, most of these fluctuations were shown to have a minimal impact on the final basal-subtracted signal obtained (18). Moreover, detection of a serially diluted clone over different basal backgrounds was also similar. Three main conclusions can thus be drawn: 1) most CD8⁺ T-cell responses are short-lived after clinical onset, 2) some new reactivities appear, which preferentially target the previously subdominant IA-2_206–214 and IGRP_265–273 epitopes, and 3) GAD and IA-2 aAb titers display less fluctuations.

There are several hypotheses which, without being mutually exclusive, could explain the rapid disappearance of most β-cell–specific IFN-γ–producing CD8⁺ T-cells after type 1 diabetes diagnosis. The simplest explanation is that these T-cells are no longer detected just because they are no longer present at sufficient frequencies. They may have been deleted or suppressed as part of the natural blunting of (auto)immune responses. Although some β-cells are thought to survive for a long time after type 1 diabetes onset, the increasing paucity of islet Ags may doom most of these T-cells to die for lack of Ag stimulation. These cells may also become undetectable because of a change in their recirculation or homing behavior. Alternatively, β-cell–specific T-cells may have stopped secreting IFN-γ, changing their phenotype to a different one. There are indeed examples in the setting of both chronic virus infections (35) and of neoplastic diseases (36), where activated virus- or tumor-specific CD8⁺ T-cells persist but lose their effector function (i.e., cytotoxicity and cytokine production).

A third explanation is that the epitope specificity of the CD8⁺ T-cell population may have shifted toward different targets. Although it is possible that other unknown specificities become prevalent, an epitope focusing phenomenon was observed. Indeed, only six of the nine original epitopes were still targeted at follow-up. Furthermore, a shift in the epitopes preferentially targeted by CD8⁺ T-cells was observed, as two previously subdominant IA-2 and IGRP epitopes became targeted by 27–36% of patients, while the previously immunodominant responses against proinsulin and GAD epitopes became less prominent. IA-2_206–214 and IGRP_265–273 (but not IGRP_228–236) were also the two epitopes newly targeted in the vast majority (87.5%) of case subjects. These results are relevant in light of the recent debate as to whether there is a hierarchy of epitope spreading in the type 1 diabetes pathogenic process. Data in the NOD mouse suggest that proinsulin may stand at the source of this spreading (10,11), while IGRP would lie downstream in the cascade (9). Our human data are consistent with this model. Indeed, proinsulin- and GAD-specific responses decreased, while IA-2_206–214–specific and, to a lesser extent, IGRP_265–273-specific T-cells became immunodominant. While the decreasing proinsulin-specific responses and the increasing of IGRP_265–273 reactivity are reminiscent of NOD mouse data (6), the finding of a similar phenomenon for GAD and IA-2 is unparalleled. This is not surprising, since GAD and IA-2 are thought to be important target Ags in human type 1 diabetes (15,17,26,37–39) but not in the NOD mouse (40–43). Consideration of additional epitopes poorly recognized at type 1 diabetes onset (17,18) may reveal further specificities becoming immunodominant at later time points. It is also interesting that, similarly to the overlapping proinsulin_B9–23 and proinsulin_B15–23 mouse epitopes, the immunodominant human CD8⁺ T-cell epitopes at diagnosis (i.e., GAD_114–123) and at follow-up (i.e., IA-2_206–214) overlap with the DR4-restricted epitopes GAD_115–127 (44,45) and IA-2_206–221 (46).

The observed shift in epitope specificity could be due either to the emergence of novel clonotypes or to an avidity maturation of preexisting ones. Indeed, if IA-2_206–214–specific CD8⁺ T-cells undergo avidity maturation, this would lower the ISL8Spot detection threshold and thus increase their measured frequencies. Alternatively, the measured frequencies could increase as a result of higher precursor numbers following de novo generation and/or expansion. The study of epitope-specific CD8⁺ clones obtained at different time points will clarify this issue.

Another important question is whether the observed changes in CD8⁺ T-cell responses are part of the type 1 diabetes natural history or, rather, reflect an immunological effect induced by insulin therapy. Indeed, daily insulin treatment in these patients may have modulated T-cell responses. Consistent with this hypothesis, proinsulin-specific responses were frequently undetected at follow-up (12 of 15 patients [80%]), a trend which contrasted with the opposite behavior of insulin antibody responses, which rose in most (10 of 12 [83%]) instances. The increase in insulin Ab titers could also contribute to this shift in T-cell responses, as Abs could change the efficiency of presentation of proinsulin-derived T-cell epitopes (47,48). Besides the possibility of a direct effect of proinsulin as an Ag on its cognate T-cells, an indirect tolerogenic effect could also be mediated by the hormonal activity of proinsulin. Insulin treatment can indeed reduce the functional pressure on the β-cell, which may translate to reduced β-cell apoptosis and/or reduced β-cell Ag release.

The slower changes observed for aAb titers compared with T-cell responses have important implications. Autoreactive memory B-cells are long-lived and can keep replenishing the plasma cell compartment without further need for Ag-specific stimulation (49). It is not completely clear whether the same rules apply to CD8⁺ T-cell memory (50). The efficient B-cell homeostasis may make it possible for the aAb titers to be maintained longer than T-cell responses. The faster dynamics of CD8⁺ T-cell responses could make their measurement more suitable to promptly reflect the autoimmune modifications of the disease. In this scenario, aAb and T-cell measurements could find complementary clinical applications. We hypothesize that the slow changes of aAb titers may make them more suitable for predicting the “ifs” of type 1 diabetes (i.e., 5-year probability to develop disease in at-risk subjects). Conversely, the faster kinetics of T-cell responses may be less useful to predict long-term odds but better suited to reflect more proximal events, i.e., the “whens” of type 1 diabetes once disease onset, remission, or relapse are approaching.

Another important application for T-cell assays will be for the etiologic classification of borderline cases. One such example is patient P03, whose diagnosis of type 1 diabetes could be questioned, having ketoacidosis and insulin dependency at diagnosis but being seronegative and genetically protected (DR15⁺ with no HLA class II susceptibility allele). Nonetheless, his robust CD8⁺ T-cell responses at diagnosis hint at an autoimmune component for his diabetes.

The disappearance of most CD8⁺ T-cell responses following type 1 diabetes onset and initiation of insulin therapy also has important implications for the application of T-cell assays in the clinics. There is quite some parallelism between human and NOD mouse data, as in both cases T-cell responses significantly wane after clinical onset (8). If this human/mouse parallel also holds true for the prediabetic period, we may find the zenith of T-cell responses before diabetes onset. Longitudinal follow-up of at-risk subjects will shed further light on the time course of these events.

The following colleagues and Institutions contributed to the study with patient recruitment:

Coordinators, G. Novelli and G. Bruno: S. Cianciosi, Avigliana; A. Perrino, Carmagnola; C. Giorda and E. Imperiale, Chieri; A. Chiambretti and R. Fornengo, Chivasso; V. Trinelli and D. Gallo, Ciriè-Lanzo; A. Caccavale, Collegno; F. Ottenga, Cuneo; R. Autino and P. Modina, Cuognè; L. Gurioli and L. Costa-Laia, Ivrea; C. Marengo and M. Comoglio, Moncalieri; T. Mahagna, Nichelino; M. Trovati and F. Cavalot, San Luigi Hospital, Orbassano; A. Ozzello and P. Gennari, Pinerolo-Pomaretto-Torre Pellice; S. Bologna and D. D'Avanzo, Rivoli; S. Davì and M. Dore, Susa; S. Martelli and E. Megale, Giovanni Bosco Hospital, Turin; S. Gamba and A. Blatto, Maria Vittoria Hospital, Turin; P. Griseri and C. Matteoda, Martini Hospital, Turin; A. Grassi and A. Mormile, Mauriziano Hospital, Turin; E. Pisu, G. Grassi, V. Martina, V. Inglese, and R. Quadri, Molinette Hospital, Turin; G. Petraroli and L. Corgiat-Mansin, Ophthalmologic Hospital, Turin; F. Cerutti and C. Sacchetti, Regina Margherita Pediatric Hospital, Turin; A. Clerico and L. Richiardi, Valdese Hospital, Turin; and G. Bendinelli and A. Bogazzi, Venaria.

Coordinator, L. Chaillous: P.H. Ducluzeau and V. Rohmer, Angers; M. Dolz, V. Kerlan, and E. Sonnet, Brest; B. Charbonnel, Nantes; R. Marechaud, Poitiers; and P. Lecomte, Tours.

This study was supported by the Juvenile Diabetes Research Foundation International grants 1-2005-39 and 20-2006-1095. R.M. was recipient of an Accueil Chercheur Etranger Fellowship of the Fondation Recherche Médicale.

This study was presented at the 9th International Congress of the Immunology of Diabetes Society and American Diabetes Association Research Symposium, Miami, Florida, 14–18 November 2007.