Activation of Insulin-Reactive CD8 T-Cells for Development of Autoimmune Diabetes

Insulin-specific TCR transgenic mice were generated by isolating TCR genomic DNA from G9C8 cloned T-cells (23), which have previously been shown to have specific reactivity to amino acids 15–23 of the insulin B-chain (1). TCRα (Vα18S1 Jα18 Cα)- and β (Vβ6S1 Dβ1.1 Jβ2.3 Cβ1)-chain DNA was purified and cloned into pTαcass and pTβcass constructs (28), respectively, kindly provided by D. Mathis. The cloned constructs were injected directly into NOD ova to generate independent TCRα and β founder lines in the Yale Diabetes Endocrinology Research Center (DERC) transgenic core (R.A. Flavell, Yale School of Medicine), which were intercrossed to produce αβ TCR transgenic mice (G9NOD). Three independent founder lines each of TCRα (lines 17, 22, and 24) and β (lines 1, 2, and 3) were generated, and α lines 17, 22, and 24 were crossed with each of β lines 2 and 3. Results with each of these lines was similar in terms of selection and functional phenotype (data not shown), and, thus, TCRα line 22 and TCRβ line 2 were chosen for all subsequent studies.

To generate repertoire-restricted αβ TCR transgenic mice, G9Cα^−/−.NOD mice were generated by crossing the αβ TCR transgenic mice to NOD.Cα^−/− mice. The NOD.Cα^−/− mice were generated by backcrossing TCR.Cα^−/− mice (B6/129 background; kindly provided by A.C. Hayday) (29) >20 generations to NOD/Caj mice in the laboratory of R.S. Sherwin (Yale University). G9RAG^−/−.NOD mice were generated by crossing the αβ TCR transgenic mice to NOD.RAG2^−/− mice, generated by backcrossing RAG2^−/− mice (kindly provided by D. Schatz, Yale University) to NOD/Caj mice >12 generations. To generate G9Cα^−/−.RIP-B7.1 NOD mice, the RIP-B7.1 mice were 10 generations backcrossed to NOD.Cα^−/− mice and intercrossed to generate NOD.Cα^−/−RIP-B7.1 mice and then crossed with the G9Cα^−/−.NOD mice to generate G9Cα^−/−.RIP-B7.1 NOD mice. NOD.scid mice were originally obtained from The Jackson Laboratories but were bred for many years at Yale University. The mice were housed in microisolators or scantainers in barrier rooms. All procedures were performed in accordance with the protocols approved by the U.K. Home Office or institutional animal care and use committee at Yale University.

Single-cell suspensions of 0.1–1 × 10⁶ cells were stained for various cell surface markers (rat anti-murine CD8; CD4, CD62L, and CD44). FcRs were blocked with anti-CD16 (2.4G2; provided by the late C.A. Janeway), and cells were stained for surface markers. Intracellular staining was performed on cells cultured for 15 h with differing concentrations of peptide, for interferon (IFN)-γ and tumor necrosis factor (TNF)-α, after cell surface staining with anti-CD8 following the BD Bioscience protocol. They were analyzed by flow cytometry using a FACSCalibur, and results were analyzed using CellQuest (BD Bioscience) or FlowJo (Tree Star) software. All antibodies were purchased from BD Bioscience, except anti-murine CD8-allophycocyanin (Caltag Laboratories).

Cells were stained with K^d-B15-23-PE tetramer or control K^d-listeriolysin (LLO) 91-99-PE tetramer (obtained from the National Institutes of Health Tetramer Facility) for 1 h at 4°C. Anti–CD8–fluorescein isothiocyanate was added for the last half hour. The cells were then washed and analyzed by flow cytometry.

Enriched CD8 T-cells from splenocytes of G9Cα^−/−.NOD mice were labeled with 5 μmol/l carboxyfluorescein succinimidyl ester (CFSE) for 10 min at 37°C and washed with ice-cold RPMI-1640 with 5% FCS and then with 0.9% normal saline. Ten million cells were resuspended in 0.9% normal saline and transferred intravenously to 4- or 6-week-old NOD mice. The spleen, pancreatic, axillary, and mesenteric lymph nodes were removed 6–10 days posttransfer, and CFSE dilution was analyzed by flow cytometry after staining with anti–CD8-allophycocyanin and anti–CD69-phycoerythrin.

Insulin-reactive CD8 T-cells from 5- to 8-week-old G9Cα^−/−.NOD mice were cocultured with bone marrow–derived dendritic cells and B15-23 peptide or influenza hemagglutinin peptide IYSTVASSL (control) for peptide responses in RPMI complete medium. Similarly, soluble insulin or keyhole limpet hemocyanin (control) was used as antigen for protein cross-presentation responses. CD8 T-cells were purified from splenocytes of G9Cα^−/−.NOD mice using positive selection beads (Miltenyi Biotec) with >95% purity. After 3 days of incubation, cells were pulsed with [³H]thymidine for 14 h to determine proliferation.

CD8 T-cells purified from 5- to 8-week-old G9Cα^−/−.NOD mice, used directly ex vivo or prestimulated for 24 h with 1 μg/ml insulin B15-23 peptide, were harvested and further cocultured with 10^4 51Cr-sodium chromate (Amersham)-labeled P815 cells together with B15-23 insulin peptide at an effector-to-target (E:T) ratio of 10:1 for 16 h. Specific lysis was calculated as [(cytotoxic release − min)/(max − min)] × 100%, where the minimal release (min) corresponded to the spontaneous lysis, and the maximal lysis corresponded to lysis induced by addition of Triton-X100 (max).

G9Cα^−/−.NOD mice were immunized intraperitoneally with 50 μg B15-23 peptide together with different concentrations of CpG 1826 oligonucleotides (Coley pharmaceuticals). Nine to twenty-one days later, a further injection of the same dose was given. Controls were injected with either 50 μg B15-23 peptide or the different doses of CpG alone.

Diabetes was detected by initially testing for glycosuria and confirmed by blood glucose measurements >13.9 mmol/l/l.

A total of 5–7 × 10⁶ purified CD8 T-cells from G9Cα^−/−.NOD mice, either taken directly ex vivo or preactivated by CpG-matured bone marrow–derived dendritic cells and insulin B15-23 peptide, were washed, resuspended in a volume of 200 μl of normal saline, and transferred intravenously to 6-week-old NOD.scid mice, 3-week-old NOD mice, and 6-week-old NOD mice. CD4 T-cells purified either from young 6-week-old mice (10⁷) or from mice that had become diabetic (4–10 × 10⁶) were transferred intravenously to 3- to 4-week-old G9Cα^−/−.NOD mice.

Pancreata were harvested and fixed, and immunohistochemistry was performed as previously described (30). Insulitis was measured in 6-, 9-, 12-, 15-, 20-, and 25-week-old mice (n = 3–4 each). Sections were stained with hematoxylin and eosin (Vector Laboratories) and scored for insulitis.

The transgenic TCR in G9NOD mice is expressed in both thymus and the periphery, and there is an increased proportion of CD8 T-cells and reduced CD4 T-cells compared with nontransgenic NOD mice (Fig. 1,A–E). The allelic exclusion at the TCRα chain locus is poor (Fig. 1 C and F), and other TCRα chains can be selected as shown by a high frequency of expression of the TCRα chains 2, 3, 8, and 11, which were identified by flow cytometry. Similar results were obtained from different TCRαβ transgenic lines (combination of mice from three different TCRα founder lines with two different TCRβ founder lines; data not shown). Thus, to study the transgenic TCRαβ of the original G9C8 T-cell clone, we selected one TCRαβ transgenic line and crossed the G9NOD mice with TCRCα^−/− and RAG2^−/− mice in order to study the characteristics of the repertoire-restricted TCRαβ transgenic cells.

Despite thymic expression of preproinsulin 2 (31), the G9Cα^−/− and G9RAG^−/− transgenic cells are clearly selected in the thymus (Fig. 2,A and F). There is peripheral expansion of the T-cells in the spleen and lymph nodes (illustrated in Fig. 2,B and C and Fig. 2,G and H, respectively), and, as expected, the cells express TCRVβ6 (shown in Fig. 2,D and I), and the majority of cells expressing CD8 are also positive for K^d-insB15-23 tetramer staining (Fig. 2 E).

The T-cells demonstrate antigen-specific cytotoxicity directly ex vivo (shown in Fig. 3,A), but their ability to kill in vitro targets is enhanced by nearly two logs after culturing the cells with antigen, indicating that the cells are relatively naïve when derived directly ex vivo from the transgenic mouse (Fig. 3,A). They produce IFN-γ and TNF-α in response to antigenic peptide (Fig. 3,B). As expected, they proliferate to insulin peptide (Fig. 3,C) and also respond to insulin protein cross-presented by bone marrow–derived dendritic cells (shown in Fig. 3 D), indicating that the peptide is naturally processed and presented. Thus, the transgenic G9 cells maintain the immunological function of their parental G9C8 clone.

In view of the highly pathogenic nature of the original G9C8 T-cell clone, we tested whether the mice developed insulitis. We found that in the G9Cα^−/− TCR transgenic mice, a few islets have a large infiltration by CD8 T-cells but also by B-cells and CD4 T-cells (online appendix Fig. 1 [available at http://diabetes.diabetesjournals.org/cgi/content/full/db08-0800/DC1]). However, infiltration is only seen after 15 weeks of age, and even then most of the islets do not become infiltrated (islets examined from mice at 6, 9, 12, 15, 20, and 25 weeks of age) (insulitis scores shown in online appendix Table 1). Neither the G9Cα^−/− TCR mice (30 mice observed until 35 weeks of age) nor the G9RAG^−/− TCR mice developed spontaneous diabetes, similar to the three lines of G9NOD mice observed that were not on the Cα^−/− background.

As can be seen from Fig. 2, although there are fewer CD4 T-cells in the G9Cα^−/−.NOD mice compared with the G9NOD mice (shown in Fig. 1), the question arises as to whether these CD4 T-cells contain a regulatory T-cell population that prevented the G9Cα^−/−.NOD mice from developing diabetes. To examine whether the CD4 T-cell population inhibited the CD8 T-cells in vitro, proliferation assays were performed that showed that G9Cα^−/−.NOD CD4 T-cells can proliferate to the insulin B15-23 peptide but at a lower level. Furthermore, they do not inhibit the CD8 T-cells from proliferating when they are added to CD8 T-cells in the proliferation assay but rather enhance the proliferation (shown in Fig. 4 A). However, to test further whether these cells were having an inhibitory effect in vivo, we depleted both total CD4 T-cells (n = 6) using anti-CD4 antibody and CD4⁺CD25⁺ T-cells (n = 3) using anti-CD25 antibody injected four times over a period of 6 weeks starting at 4 weeks of age. The experiments were controlled by injection of rat isotype control antibody (n = 3). Efficacy of depletion was shown by testing peripheral blood just before the next injection, and at each time point in mice injected with anti-CD4 or anti-CD25, the CD4 or CD4⁺CD25⁺ T-cells were <1% in the peripheral blood (data not shown). None of the mice in any of the groups, observed until 20 weeks of age, developed diabetes. Thus, the lack of diabetes did not appear to be related to CD4 regulatory T-cells.

To further examine whether CD4 T-cell help was required, G9Cα^−/−.NOD mice were infused with either purified CD4 T-cells from young NOD mice or with CD4 T-cells from diabetic NOD mice. None of the G9Cα^−/−.NOD mice given CD4 cells from young NOD mice developed diabetes, whereas diabetes occurred in G9Cα^−/−.NOD mice transferred with CD4 T-cells from the diabetic NOD spleens (shown in Fig. 4 B). Thus, activated but not nonactivated polyclonal CD4 T-cells are able to promote diabetes in the G9Cα^−/−.NOD mice.

The G9 cells in TCR transgenic mice express a naïve phenotype and function, and the G9Cα^−/−.NOD mice do not develop spontaneous diabetes. Therefore, we sought to test whether activation of the insulin-reactive transgenic T-cells, other than by adding activated CD4 T-cells, could cause diabetes. We injected the mice with insulin B15-23 peptide together with type B CpG oligonucleotides and showed that this approach promoted very rapid diabetes development (Fig. 5,A). This was a specific response requiring peptide, as CpG alone did not induce disease (data not shown). Similarly, the CpG was required to activate the APCs, as injection with peptide alone also did not induce disease (data not shown). The combined CpG and peptide injection activated G9 transgenic T-cells, and their activation profile shows downregulation of CD62L and upregulation of CD44 (illustrated in Fig. 5,B). On development of diabetes, most islets were seen to have extensive infiltration and destruction, whereas little or no insulitis was seen in the islets of mice that were immunized with either CpG or peptide alone (shown in Fig. 5 C).

Furthermore, the G9Cα^−/−.NOD CD8 T-cells could become potent diabetogenic T-cells after activation with bone marrow–derived dendritic cells that are matured by CpG. Following a 3-day culture with CpG-matured bone marrow dendritic cells in the presence of insulin B15-23 peptide, the G9Cα^−/−.NOD CD8 transgenic cells were able to transfer diabetes rapidly without the aid of CD4 T-cells into 6-week-old NOD.scid mice (Fig. 6), recapitulating the phenotype of the original CD8 T-cell clones (23). G9Cα^−/−.NOD CD8 T-cells that were not preactivated did not transfer diabetes when a similar number of nonactivated G9Cα^−/−.NOD CD8 T-cells were transferred and observed for 12 weeks after transfer (data not shown). This suggests that activation increases the functional avidity of G9 cells (shown by the data in Fig. 3 A), and this enhances their cytotoxic function and diabetogenicity.

We have previously suggested that expression of the costimulator B7-1 on the islets is able to activate low-avidity T-cells and cause accelerated diabetes (30). To test whether the coexpression of RIP-B7.1 could induce diabetes in G9Cα^−/−.NOD mice, we crossed the G9Cα^−/−.NOD mice to RIP-B7.1.Cα^−/− mice that had been generated for this purpose. In the presence of B7.1 in the pancreas, the G9Cα^−/−.NOD mice developed spontaneous diabetes (shown in Fig. 7). To determine that this was a peripheral effect and not related to changes in selection of the G9TCR T-cells, we had previously demonstrated that the naïve G9TCR CD8 T-cells from G9Cα^−/−.NOD mice could readily transfer diabetes to NOD.scid mice that expressed RIP-B7.1 (30), indicating that the naïve G9TCR CD8 T-cells can be activated in situ in the islets and become highly pathogenic. These results support the data presented earlier that increasing the functional avidity of G9TCR T-cells through activation can promote diabetogenicity of transgenic G9 T-cells.

To investigate the homing and activation of insulin-reactive G9 CD8 T-cells in vivo, we used CFSE-labeled, purified CD8 T-cells from G9Cα^−/−.NOD mice transferred into 4- to 6-week-old NOD mice and analyzed CFSE dilution in different lymphoid tissues. We showed that the G9 CD8 T-cells proliferated, as shown by the decrease in CFSE staining, mainly in the pancreatic lymph nodes (PLNs) but not in the spleen or other lymph nodes. More importantly, the dividing G9 CD8 T-cells express early activation marker CD69 (Fig. 8). This indicates that NOD mice as young as 4 weeks of age have the capability of activating G9-like cells in the pancreatic lymph nodes. However, this single transfer of nonactivated cells into 4- to 6-week-old NOD mice did not result in acceleration of diabetes, when the mice were followed to 15 weeks (the age by which acceleration of disease would be observed).

Our study has shown that G9 TCR transgenic CD8 T-cells that react to an insulin B-chain peptide B15-23, one of the earliest antigenic epitopes to which CD8 T-cells respond (at 3–4 weeks of age), are selected in the thymus. Once released in the periphery, they can be stimulated to become highly pathogenic. The G9 cells are activated by the B15-23 peptide, which has low-affinity binding to the MHC-K^d, and, interestingly, as expected from a naturally processed and presented peptide, transgenic G9 cells can be stimulated by whole insulin protein cross-presented by dendritic cells. Although G9Cα^−/− mice do not spontaneously develop diabetes, they do so when naturally activated polyclonal CD4 T-cells from diabetic NOD mice are transferred, indicating a requirement for CD4 T-cells. Moreover, when transgenic G9 cells are fully activated in vitro, they gain potent diabetogenicity and cause diabetes in immunodeficient NOD.scid mice, without additional CD4 T-cell help, recapitulating the phenotype of the parental clone cells. Importantly, these cells can also be activated in vivo and cause rapid diabetes development by providing exogenous insulin peptide in the presence of immunostimulatory CpG. A similar effect can be achieved if the costimulatory molecule B7.1 is expressed within the islets.

Why are these low-avidity T-cells present in the NOD mouse and where are the cells stimulated in vivo? Recently, a multitude of peripheral antigens were shown to be expressed in medullary epithelial cells of the thymus due to the autoimmune regulator gene Aire (32). High-avidity T-cells are removed at the cortico-medullary junction, protecting against autoimmunity, and deficiency of Aire leads to widespread multiorgan autoimmune disease (32). However, some autoreactive T-cells still escape from the thymus, and, using model antigens, it has been shown that low-avidity T-cells are not deleted in the thymus (33,34). PI2 is expressed in the thymus (17,35), and even with the thymic expression, G9-like low-avidity insulin-reactive T-cells can clearly escape from negative selection. Furthermore, in our studies, low-avidity insulin-specific CD8 T-cells are found in the periphery and are not subject to peripheral deletion. As with other islet autoantigen-reactive reactive CD8 (36) and CD4 (37) T-cells in the NOD mouse, G9 cells are activated and proliferate preferentially in the pancreatic lymph nodes compared with the spleen and other lymph nodes, including the mesenteric lymph nodes. However, our earlier study and those of others (1,3,24) demonstrated that the numbers of these cells remain small in lymphoid tissue, but they are measurable in the islets from an early age. It is possible that only in the islets is sufficient antigen presented to these cells for their expansion, as a relatively high concentration of the low-affinity peptide is required to stimulate them. These cells can be sufficiently activated in the pancreatic draining lymph nodes, possibly by dendritic cells that stimulate them to traffic to the islets where a higher concentration of the insulin peptide within the islets induces G9 cell expansion and damage to the islets in the early phases. However, other specificities that may be of higher avidity, such as IGRP-reactive CD8 T-cells, expand more and overtake these cells in number and percentage representation in the islet infiltrate in later phases of the disease development.

Why do these TCR transgenic mice not develop diabetes spontaneously? In the NOD mouse, different islet antigen–specific T-cells are likely to be involved in early phases of the autoimmune process, and it is likely that each may amplify the autoimmune response of the others by producing interleukin-2 locally, which may also increase cells by homeostatic expansion (38). Insulin, as a β-cell–specific autoantigen, is very different from other β-cell–associated antigens. In addition to MHC class I–restricted antigen presentation by the β-cells, insulin is secreted by islet β-cells in response to high glucose from the portal system and circulates in the body at low concentrations, dependent on ambient glucose. The peptide recognized by the insulin-specific T-cells in this study has a low affinity of binding to H2-K^d molecules (39). The widely studied 8.3 TCR transgenic T-cells recognize an epitope of IGRP 206-214, an enzyme within β-cells with high avidity (40). Much less is known of the other autoantigen, dystrophia myotonica kinase, which stimulates the CD8 T-cells from the TCR transgenic mouse AI-4 (3). Both of these mice develop diabetes spontaneously, although 8.3TCR transgenic mice develop more diabetes in the presence of CD4 T-cells (27), whereas this is not a requirement for the AI-4 transgenic mouse (41). In contrast, for CD4 TCR transgenic mice, which also have varying incidence of diabetes, the highly pathogenic BDC2.5 cells develop spontaneous diabetes when expressed on a C57BL/6 genetic background, but the incidence of diabetes is much lower on the NOD genetic background (42), although it is sharply increased when regulatory T-cells are removed by expression on a NOD.scid genetic background. Similarly, the 4.1 TCR transgenic mice recognizing an unknown islet antigen develop diabetes a little earlier on a RAG2^−/− genetic background (27), and the 12-4.1 TCR transgenic mice, recognizing an insulin peptide only, develop diabetes on a RAG2^−/− genetic background (43). Not surprisingly, these mice have varying phenotypes, differing between CD4 and CD8 transgenic T-cells that are likely to reflect important aspects of the biology of the native cells within the normal mice and also interactions with other cells of the immune system, emphasizing the need to study more than one specificity of pathogenic T-cell.

The fact that the G9TCR transgenic mice do not develop spontaneous diabetes is highly reminiscent of the LCMVgp/p14 double transgenic mouse, in which a transgenic TCR that recognizes a peptide of lymphocytic choriomeningitis virus (LCMV) glycoprotein was coexpressed with the target LCMVgp as an autoantigen in the pancreas using the rat insulin promoter (RIP) (44). In that model, the TCR transgenic cells were “ignorant” and diabetes did not occur spontaneously but could be induced to occur on infection with LCMV or administration of the LCMVgp peptide with lipopolysaccharide (44). Here, we show that this mechanism can operate for a naturally occurring autoantigen (insulin) and low-avidity (G9) T-cells. In the NOD mouse, which develops diabetes spontaneously, infection is clearly not the strong stimulus necessary for activation of low-avidity cells, and, indeed, many infections prevent diabetes in these mice. However, there may be endogenous stimuli of the innate immune system that could activate low-avidity cells (45,46), leading to further immune activation in both diabetes and other autoimmune diseases. Indeed, we have shown that activation, using a ligand for TLR9, can stimulate the insulin-reactive cells to become highly potent cytotoxic T-cells. The challenge will be to identify how this process occurs in vivo.

What relevance does this low-avidity T-cell have in diabetes? A recent study (15) showed that diabetes was inhibited if NOD mice were deficient in both PI1 and PI2 but expressed a transgenic insulin that was mutated at position 16 of the B-chain. The alteration of B16, in fact, will undoubtedly affect both pathogenic CD4 cells recognizing insulin B9-23 (47) as well as pathogenic CD8 T-cells recognizing insulin B15-23 (23). This B16 mutation overlaps the CD8 epitope B15-23 and is the major MHC-K^d binding residue for the B15-23 epitope (39). Its effects in protecting against diabetes were likely to be linked to removing not only the CD4 epitope affecting diabetogenic CD4 cells but also the CD8 epitope, abolishing the ability of G9-like CD8 T-cells to recognize and bind natural B15-23, a weak binder to the MHC-K^d (39). Thus, although the study by Nakayama et al. (15) did not distinguish effects on CD4 insulin– versus CD8 insulin–reactive T-cells, CD8 T-cells are highly likely to be affected (48,49). G9-like cells are present in the islets at a very early stage of islet infiltration (1), and others have shown that reactivity to B15-23 is the earliest CD8 antigenic specificity so far identified when compared with the more dominant IGRP-reactive T-cells (24). Cytotoxic T-cells specific for IGRP were not detected in PItg NOD mice that were tolerant to PI, suggesting that immune responses against PI are required for IGRP-specific T-cells to develop (50). The current study reports insulin-reactive CD8 transgenic T-cells can cause rapid diabetes after activation, and this study does not address whether insulin is a “prime autoantigen” for CD8 T-cells in NOD mice. Further work is ongoing to establish whether these cells require “help” from CD4 insulin–reactive cells or whether other specificities can provide this.

In conclusion, we have shown that a CD8 T-cell responsive to a natural peptide of insulin, a major autoantigen in autoimmune diabetes, is selected in the thymus and can become highly pathogenic once activated in the periphery, causing diabetes. These T-cells are “ignorant” in a naïve state, but this can be readily overcome by stimuli that include direct activation or activation of antigen-presenting cells, either in pancreatic lymph nodes or in the islets. CD4 T-cells are likely to play a significant role, and this is currently under investigation. Further experiments aim to identify the endogenous stimulators that facilitate the “switch” of naïve and “ignorant” G9 cells into potent β-cell–destructive effector cells in the NOD mouse. Our study provides an important model not only for the study of immunopathogenic role of insulin-specific CD8 T-cells in autoimmune diabetes but also for better design of insulin-based immunotherapy.

These studies were funded by the Wellcome Trust, the Juvenile Diabetes Research Foundation, and the National Institute of Diabetes and Digestive and Kidney Diseases (DERC, transgenic core, and NOD mouse genetic core). Tetramers were obtained from the National Institutes of Health Core Facility.

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

In memory of Charles A. Janeway, Jr., without whose support in the early stages, this work would not have been possible. We thank Dr. Andrew Herman and the University of Bristol Cellular and Molecular Medicine Flow Cytometry Facility for assistance in these studies and Dr. Alexander Chervonsky for helpful comments on the manuscript. We also thank Ewan Basterfield, Jenny Radcliffe, Louise Phelon, and Craig Husher for expert care of the animals.