Autoimmunity Without Diabetes in Transgenic Mice Expressing β Cell-Specific CD86, But Not CD80: Parameters that Trigger Progression to Diabetes¹

Ligation of the CD28 receptor, expressed by all T cells, with the B7 molecules provides a crucial costimulatory signal that permits the activation and differentiation of mature T cells into functional effector T cells (reviewed in Refs. 1 and 2). This is exemplified in CD28-deficient mice (CD28^−/−), which present a generalized T cell deficiency, evidenced by reduced in vivo cellular and humoral immunity (3, 4, 5). Two B7 homologues have been identified, CD80 or B7.1 and CD86 or B7.2, that share 25% amino acid identity and similar affinity for the CD28 receptor and were thought to share identical function (6, 7, 8, 9, 10, 11). Functional redundancy was suggested by the demonstration that CD80 or CD86 transfectants have similar efficacy in providing costimulatory signals for mitogen- or anti-CD3-activated T cells (7, 12). Other studies suggested, however, that CD80 and CD86 might express essential functional differences. Expression of these two costimulatory molecules by APCs differ quite substantially (reviewed in 2 . While CD80 is not expressed by unactivated APC, CD86 is expressed at very low levels by some dendritic cells and macrophages (13). Activation by numerous stimuli including cytokines and CD40 induces expression of both CD80 and CD86 (14, 15, 16). The induction of CD86 expression follows faster kinetics and usually reaches higher levels than does CD80 induction (14, 16). Perhaps due to this differential expression pattern, CD86 appears as the predominant costimulatory molecule in in vivo and in vitro T cell responses (6, 8, 10, 17). Indeed, in many experimental systems, blocking CD86 has a more potent effect than blocking CD80, although anti-CD80 and anti-CD86 have additive effects in blocking T cell responses. More essential to the regulation of immune and autoimmune responses was the observation that CD80 and CD86 exhibit different abilities to induce Th cell differentiation, with CD80 favoring Th1 responses and CD86 preferentially inducing Th2 responses. The first evidence for this functional dichotomy was provided by Kruchoo et al. and was further documented by Miller et al., who showed that anti-CD80 Ab treatment reduced the incidence of experimental autoimmune encephalomyelitis, a Th1-mediated disease, while anti-CD86 treatment had no significant effect (18, 19). Protection was indeed associated in this model with a reduced Th1 and increased Th2 response to the encephalogenic Ag myelin basic protein. Similarly, allo-reactive T cells produced higher amounts of IL-4 and similar amounts of IL-2 and IFN-γ following repeated stimulation with CD86-transfected COS cells compared with CD80-transfected COS cells (20). In contrast, anti-CD86 was very efficient at preventing the development of spontaneous autoimmune diabetes in NOD mice, another Th1-mediated autoimmune disease, while anti-CD80 exacerbated disease (21, 22). Interestingly, NOD mice on a CD28^−/− background showed an increased incidence of diabetes associated with an increased representation of IFN-γ-producing cells in the islets (22). This observation raised the possibility that Th2 responses would require CD28 costimulation, while Th1 responses would be less dependent on CD28-mediated costimulation, a possibility that has been substantiated both in vitro and in vivo (23, 24, 25, 26). Other studies, however, do not support such a model (27, 28), confirming the protective effect of CTLA-4Ig treatment on graft rejection (reviewed in 29 .

There are several difficulties that arise in the interpretation of the above data. Most of the studies aimed at resolving the roles of CD80 and CD86 in T cell activation in vivo used Ab administration to block the different B7 molecules. This protocol could generate complications and artifacts resulting from Fc receptor signaling or cross-linking of the B7 molecules that could account for the discrepancies described. And indeed, using CD80- or CD86-deficient APC, such a dichotomy was not substantiated (30, 31). Moreover, expression of CD80 and CD86 is regulated differently depending of the type of APC and these molecules are also expressed on T cells (14, 32). It could thus be possible that the differences observed were linked to a selection or activation of the APC, that itself would have different potentials in stimulating different effector functions.

To circumvent these problems and to directly address the roles of CD80 and CD86 in the activation of self-reactive T cells in vivo and thus in the development of autoimmunity, we generated mice that express one or both molecules on a given tissue. Following this approach we and others have described transgenic mice expressing CD80 on the islets of Langerhans (RIP-CD80) (33, 34, 35). We extended this analysis and generated transgenic mice on a C57BL/6 background that express the CD86 costimulatory molecule on β cells (RIP-CD86). The comparative analysis of these two lines of transgenic mice in different transgenic models of autoimmune diabetes highlights some essential differences in the ability of CD80 or CD86 to induce T cell activation. These differences are however obliterated when the level of TCR signaling is increased, suggesting that CD80 and CD86 mainly differ at the level of the intensity of the signals they deliver.

Transgenic mice expressing the mouse inflammatory cytokine TNF-α (RIP-TNF-α) (36), the class II IE molecule (Ins-IE) (37), or the human CD80 (3B7 or RIP-CD80) (34) on β cells have been described previously. Mice transgenic for a TCR specific for the SV40 large T Ag presented by the H-2K^k class I molecules were provided by T. Geiger (38). RIP-CD86 transgenic mice were generated by microinjection of a cDNA construct encoding the human CD86 molecule driven by the rat insulin promoter (RIP)⁵ into the pronuclei of fertilized (B6×CBA/Ca)F2 eggs as previously described (39). The RIP-CD86 construct was generated as follows. A SmaI-NotI fragment corresponding to the full-length human cDNA isolated from psk-B70 vector (a generous gift from A. Bothwell) was coligated with a NotI-HindIII fragment corresponding to part of the third and fourth exons of Eα (providing a splice site and a polyadenylation site) isolated from RIP-B7 vector (34) into blunt-end EcoRI and HindIII sites of psk-RIP. The psk-RIP vector corresponded to the minimal 0.6 kb of the RIP promoter inserted into Psk (a generous gift from D. Picarella). A SacI-KpnI fragment was isolated for microinjection. All the mice were backcrossed on C57BL/6 and were H-2^b homozygotes except when otherwise specified.

Total RNA from different tissues was extracted by acid guanidinium-thiocyanate (40), treated with DNase I, reverse transcribed with oligo(dT) into cDNA, and then amplified by PCR, as previously described (41). For the PCR reaction, a forward primer within the CD86 cDNA (5′-CGACGTTTCCATCAGCTTGTCTG-3′) and a reverse primer in the fourth exon of the Eα gene (5′-CAAGACTCCAGGGATTTGAGGGA-3′) were used, distinguishing the unspliced DNA contaminant of 1360 bp from the spliced RNA product of 720 bp (Fig. 1). The PCR reactions were run on an 0.8% agarose gel, transferred to nylon membrane, and hybridized with a CD86-specific probe. γ-Actin primers were used for RNA from all tissues as a control for the quality of RNA preparation and the efficiency of cDNA synthesis.

Responding CD8⁺ T cells were isolated from lymph nodes of TG-B transgenic mice that had been crossed on C57BL/6, a nonpresenting H-2^b background. Lymph nodes cells were incubated with a class II-specific (m5/114) (42) and a CD4-specific (GK1.5) (43) Ab, washed, and incubated with a mixture of goat anti-mouse Ig- and goat anti-rat Ig-coated magnetic beads (Collaborative Research, Bedford, MA) followed by two cycles of exposure to a magnetic field. The β cells were obtained by trypsin digestion of hand-picked islets isolated from a collagenase digestion of total pancreas (44). Before trypsinization, isolated islets were incubated overnight in medium containing 100 U/ml IFN-γ (Life Technologies, Gaithersburg, MD) to up-regulate MHC class I expression. T cells (1 × 10⁴) were incubated with 2 × 10⁴ γ-irradiated β cells with or without the 9 mer SV40 peptide (45) and the human CD86-specific Ab IT2.2 at final concentrations of 10 μM and 10 μg/ml, respectively. In parallel TG-B T cells were stimulated with a titration of spleen APC, allowing for a direct comparison, on a per cell basis, of the efficacy of islet cells to activate TG-B T cells. On day 5, 50 μl of supernatant was harvested for IL-2 testing, and the remaining cultures were pulsed overnight with 1 μCi [³H]thymidine. IL-2 production was measured by bioassay using the CTLL-2 cell line as previously described (34).

Mice were bled weekly, and glucose levels were measured using the One Touch (Johnson and Johnson, Milipitas, CA) apparatus. Mice with blood glucose levels of >250 mg/dl were considered diabetic.

For histology, the organs were fixed in 10% formalin, embedded in paraffin, and stained with hematoxylin and eosin. All immunocytochemistry was performed on 7-μm frozen sections (46) in Tris buffer, pH 7.5, with 1% BSA and 0.01% Triton, except for cytokine staining, where 0.5% Triton was used. Abs recognizing CD4, CD8, B220, IL-4, and IFN-γ as well as rat isotype controls were purchased from PharMingen (San Diego, CA); biotinylated anti-rat IgG was obtained from Vector Laboratories (Burlingame, CA); anti-insulin, anti-glucagon, and biotinylated anti-rabbit IgG were purchased from Biogenex (San Ramon, CA). Sections were incubated for 2 h with unconjugated or biotinylated Ab followed, when required, by a 1.5-h incubation with a secondary, biotinylated Ab. Streptavidin coupled to alkaline phosphatase (Zymed, South San Francisco, CA) was then added for 40 min, and development of color was revealed with HistoMark-Red (Kirkegaard and Perry Laboratories, Gaithersburg, MD). All sections were counterstained Gill’s hematoxylin no. 1.

The specificity of the cytokine staining was first determined using cytospin preparations of Th1 or Th2 clones derived from a D10 TCR transgenic mouse (a gift from Bonnie Dittel, Yale University). Specificity was confirmed using OVA/CFA-activated spleen and lung sections from IL-4^−/− and IFN-γ^−/− animals as well as normal animals with OVA-specific Th1 or Th2 cells transferred. In all stainings nonspecific isotype-matched control mAbs were used on parallel sections. All stainings were read blind, and at least four islets per section were analyzed. The number of cytokine-positive cells per islet or infiltrate was calculated by dividing the number of positive cells per slide by the total number of islets analyzed. The different ratios were calculated using these numbers. A dividend of zero was considered as 1.

Mice were injected i.p. with 0.6 mg of BrdUrd (Sigma, St. Louis, MO) 15 to 18 h before death. Seven-micron sections of pancreata or spleen were treated with 70% ethanol for 20 min, followed by a 20-min treatment with 3 N HCl in 0.5% Tween (Sigma) and a final neutralization with 0.01 M tetraborate (Sigma), as previously described (47). Treated slides were stained for 20 min with a 1/10 dilution of FITC-conjugated anti-BrdUrd Ab (Becton Dickinson, Cockeysville, MD).

Previous studies have indicated that mouse and human CD86 interact with mouse CD28 and have identical costimulatory activity for mouse T cells (7). Thus, we generated transgenic mice that express the human CD86 costimulatory molecule, which could readily be distinguished from the endogenous mouse homologue, specifically on the β cells of the pancreas using the rat insulin promoter (RIP-CD86). Thirteen transgenic lines were generated, and three of them were further selected on the basis of their tissue specificity and level of expression.

Expression of the transgenically encoded CD86 molecule was analyzed at the RNA level by RT-PCR and at the protein level by immunocytochemistry on tissue sections. For the PCR reaction we used a set of primers that spanned an intron, thus allowing the discrimination between products amplified from an RNA template from those resulting from a direct amplification of contaminating DNA (Fig. 1,A). We found that mRNA corresponding to the human CD86 transgene was detected only in the pancreas of the three different lines studied (Fig. 1,A and not shown). As shown in Figure 1,A, very low levels of mRNA corresponding to the CD86 transgene could be detected in the kidney and thymus of some, but not all, mice from line 12B70. When detected, the level of signal was estimated at 3 to 30 molecules. Similarly, by immunocytochemistry we showed that expression of the human CD86 protein was restricted to the islets of Langerhans (Fig. 1,B), while spleen and thymus remained negative (not shown). Together these results show that only islets express significant levels of the transgenically encoded CD86 molecules. The three lines express different levels of transgene on their β cells. Line 12B70 shows a variably low level of expression, while lines 46B70 and 38B70 express high and very high levels of human CD86, respectively (Fig. 1,B). Furthermore, neither acinar cells nor endothelial cells showed any detectable level of the human CD86 molecule (Fig. 1 B).

We then determined whether human CD86 expressed by β cells had costimulatory activity for T cells in vitro. Normal islets express MHC class I Ags but no MHC class II molecules. To analyze the in vitro response of unprimed T cells specifically, we used T cells from TCR transgenic mice expressing an H-2K^k-restricted TCR specific for the SV40 large T Ag (TG-B). We had previously shown that both proliferation and IL-2 production of TG-B T cells are dependent on costimulatory signals (48). Indeed, as previously reported, normal islets, which do not express B7 molecules, are unable to activate TG-B T cells (Fig. 2). By contrast, islets expressing CD86 efficiently stimulated Ag-specific IL-2 production and proliferation of TG-B T cells (Fig. 2). The response was completely abolished by an Ab specific for the transgenically encoded CD86 molecule. Moreover, the magnitude of the response increased as the level of CD86 expression by islets increased. We compared the proliferative response and IL-2 production induced by varying numbers of spleen APC with that induced by islets from the different transgenic mice. We found that 12B70, 38B70, and 46B70 induced, respectively, 11.9, 39.9, and 100.6% of the proliferative response and 22.9, 32.3, and 150% of the IL-2 production obtained upon stimulation with the same number of spleen APC. Likewise, we could show, by transfer experiments, that CD86-expressing islets were potent activators of allo-reactive T cells in vivo (not shown). Together these results indicate that following CD86 expression islets acquired the ability to activate T cells in vitro or in vivo.

We had previously shown that although CD80-expressing islets are potent stimulators of naive T cells in vitro and in vivo, RIP-CD80 transgenic mice rarely develop diabetes (33). To determine whether CD86 expression by islets, in contrast to CD80, could induce the activation of autoreactive T cells and thus autoimmune diabetes, we analyzed several transgenic mice over a long period of time. We found that neither mice expressing low (12B70) nor those expressing high (38B70 and 46B70) levels of CD86 on β cells showed any sign of autoimmunity until the age of 7 mo (Fig. 3,A). Starting at about 7 to 8 mo of age, however, all mice developed a perivascular/periductal infiltration (referred to hereafter as peri-islet). The onset or severity of the peri-islet infiltration was independent of the level of CD86 expression by islet cells, but was always found in the proximity of an islet, suggesting that it was a direct consequence of CD86 expression by β cells (Fig. 3,A, b–e). We followed the mice over a 14-mo period and found that the infiltrate never progressed to insulitis or diabetes. Indeed, islets from either preinfiltrated or infiltrated pancreata of RIP-CD86 transgenic mice showed normal architecture and a normal staining pattern for insulin-producing β cells or glucagon-producing α cells (Fig. 3, A and B). Furthermore, coexpression of CD80 with CD86 on islets had no additional effect on the time of onset or the severity of the infiltration (Fig. 3 Af).

It appears, then, that while CD86 expression by islet cells was sufficient to induce an inflammatory response, as exemplified by the peri-islet infiltration that developed in all transgenic mice, it could not generate the triggering event(s) required for insulitis and β cell destruction. Interestingly, in the NOD mouse model of autoimmune diabetes prior studies have shown that the absence of CD8⁺ T cells not only prevented disease but also completely eliminated insulitis (49, 50, 51). These observations suggest an essential role for CD8⁺ T cells in the initiation of autoimmune disease. RIP-CD86 mice, however, showed a normal representation of all lymphocyte subpopulations in all peripheral lymphoid organs (spleen and lymph nodes; not shown) as well as within the peri-islet infiltrate. Indeed, we found a large number of B220⁺ B cells (average, 45.4%; n = 9), CD4⁺ T cells (average, 39%; n = 9), some CD8⁺ T cells (average, 15.4%; n = 9), and a few F4/80⁺ macrophage/dendritic cells within the peri-islet infiltrate.

An alternative possibility for the lack of insulitis in RIP-CD86 transgenic mice could be an inability of CD86 to promote T cell activation in vivo. We therefore determined whether activated lymphocytes were present within the peri-islet infiltrate found in RIP-CD86 transgenic mice.

Proper activation of T cells results in entry into the cell cycle as well as induction of T cell-specific effector functions. Cell cycling can be analyzed by staining with BrdUrd, a thymidine analogue that will incorporate in newly synthesized DNA. Thus, RIP-CD86 mice were treated with BrdUrd, and pancreas sections were analyzed 15 to 18 h later. As shown in Figure 4 Aa a large number of lymphocytes found within the peri-islet infiltrate did incorporate BrdUrd. The absence of BrdUrd-positive cells within the spleen of RIP-CD86 mice (data not shown) suggests that the activation is a direct consequence of CD86 expression by islet cells.

Numerous experiments indicate that IFN-γ-producing Th1 cells play an essential role in the development of autoimmune diabetes. By contrast, IL-4-producing Th2 cells have been postulated to be harmless and in at least one transgenic mouse model of spontaneous autoimmune diabetes prevented the development of disease (52). It could be possible, then, that the absence of insulitis and diabetes in the RIP-CD86 transgenic mice resulted from the preferential induction by CD86 expressing islets of a Th2 anti-self response. To test this possibility we stained pancreatic sections of RIP-CD86 mice with IFN-γ- or IL-4-specific Ab and determined the average numbers of IFN-γ⁺ and IL4⁺ cells per infiltrate in 11 different RIP-CD86 transgenic mice. Both IFN-γ- and IL-4-positive cells could be detected in the pancreas within the peri-islet infiltrate (Fig. 4,B, e and f). Interestingly, even though there were significant numbers of IFN-γ⁺ cells in some infiltrates (Fig. 4,B, b and e), the number of IL-4⁺ cells was considerably higher than the number of IFN-γ⁺ cells (Fig. 5). Indeed, we found 3 to 10 times more IL-4⁺ cells than IFN-γ⁺ cells with an average ± SEM ratio of IL-4⁺/IFN-γ⁺ of 7.3 ± 2.1. Moreover, while the representation of IFN-γ⁺ cells did not change much with age, there seemed to be an increased accumulation of IL-4⁺ cells within the pancreatic infiltrate as the mice aged. Indeed, the representation of IL4⁺ cells per infiltrate increased from 8.6 ± 4.0 at 12 mo to 45.8 ± 20.6 at 14 mo (Fig. 5).

Together these data show that CD86 expression by islet cells leads to T cell activation and preferentially induces the differentiation of Th2 cells. This contrasted quite dramatically with transgenic mice expressing CD80 costimulatory molecules on islets that rarely developed peri-islet infiltrates, insulitis, or diabetes (33, 34, 35). Interestingly, despite the presence of activated T cells within the peri-islet infiltrate of RIP-CD86 transgenic mice, lymphocytes seemed unable to persist within the islets. An essential question thus raised by the phenotype observed in RIP-CD86 transgenic mice is whether the absence of islet destruction resulted from an inability of lymphocytes to penetrate the islets and therefore meet their targets or was only due to the preferential development of a nonaggressive Th2-type response.

Inflammatory cytokines such as TNF-α are associated with the development of numerous autoimmune diseases, including diabetes, and are very efficient at inducing lymphocyte migration, potentially through the up-regulation of expression of adhesion molecules by vascular endothelium. Indeed, RIP-TNF-α transgenic mice expressing the inflammatory cytokine TNF-α on islets show a massive lymphocytic infiltration of pancreatic islets (36, 53). Local expression of TNF-α by β cells also results in the up-regulation of the expression of MHC class I Ags by β cells and of ICAM-1 and VCAM-1 by the associated vascular endothelium (36). RIP-TNF-α transgenic mice on a C57BL/6 background, however, never developed diabetes. We have previously shown that all transgenic mice that coexpress the CD80 costimulatory molecule, at low or high levels as observed in the 3B7 line, and the TNF-α cytokine on their β cells (RIP-CD80 × RIP-TNF-α (CD80/TNF)) develop diabetes at 4 to 5 wk of age (34). Diabetes in these double transgenic mice was associated with the activation of islet-specific autoreactive T cells (34). We determined the cytokines produced by islet-infiltrating cells in prediabetic or diabetic CD80/TNF double transgenic mice. Both IL-4- and IFN-γ-positive cells are found within the infiltrates (Fig. 6,A). While the overall number of IFN-γ⁺ cells found in infiltrated islets of prediabetic (4 wk of age) and diabetic animals (6 wk of age) did not differ greatly, the frequency of IL-4⁺ cells increased as the mice became diabetic (Fig. 6 A). Indeed, we found mean numbers (±SEM) of 1.8 ± 1.4 IL-4⁺ cells in prediabetic mice and 4.6 ± 2.4 IL-4⁺ cells in diabetic mice. This change is more clearly exemplified when considering the ratio of IFN-γ⁺/IL-4⁺ cells in CD80/TNF mice, which decreases from 16.1 ± 9.2 in prediabetic mice to 2.1 ± 1.0 in diabetic mice. Thus, in this transgenic model of autoimmune diabetes, IL-4-producing Th2 cells mainly develop at the late stage of the autoimmune response and do not interfere with the primary Th1 response.

To determine whether the absence of diabetes in RIP-CD86 mice results from the lack of lymphocytic infiltration into the islets or from an inherent inability of CD86 to activate autoaggressive lymphocytes we intercrossed RIP-CD86 and the same line of RIP-TNF-α transgenic mice discussed above (CD86/TNF). As expected, all nontransgenic (−/−) or single transgenic littermates expressing either the CD86 (CD86/−) or the TNF-α (−/TNF) transgene remained normoglycemic (Fig. 7). However, all double transgenic mice coexpressing CD86 and TNF-α molecules on the islets (CD86/TNF) became diabetic, with 80% of the mice showing elevated blood glucose levels by 6 wk of age (Fig. 7). We further determined the frequency of IL-4- and IFN-γ-positive cells found in the islets of prediabetic and diabetic double transgenic mice. As in CD80/TNF transgenic mice, islets from CD86/TNF double transgenic mice contained both IL-4- and IFN-γ-positive cells (Fig. 6,B). The frequency of IFN-γ⁺ cells, however, was higher in the CD86/TNF transgenic mice (mean ± SEM, 7.4 ± 1.7) than in CD80/TNF (3.0 ± 1.0) and RIP-CD86 single transgenic (3.7 ± 1.8) mice. Interestingly, the frequency of IL-4⁺ Th2 cells found in all double transgenic CD86/TNF was dramatically reduced, especially when compared with their representation in the peri-islet infiltrates of RIP-CD86 single transgenic mice (15.4 ± 6.3 for RIP-CD86 transgenic mice compared with 3.1 ± 1.7 for CD86/TNF double transgenic mice). Together these results indicate that in this double transgenic mouse model, both CD80 and CD86 express similar qualitative costimulatory activities; that is, both induce the activation of Th1 and Th2 cells and promote the development of diabetes. CD86, however, seems more efficient at triggering effector functions in T cells, since we observed an increased representation of cytokine-producing cells within the islets of double transgenic mice expressing TNF-α together with CD86 compared with those coexpressing TNF-α and CD80 (Fig. 6, A and B).

Coexpression of the TNF-α cytokine has at least two consequences that could contribute to the development of diabetes in double transgenic CD86/TNF mice. First, following TNF-α expression by islets, lymphocytes do not accumulate in the perivascular space, but can enter the islets. Moreover, since RIP-TNF-α transgenic mice express the cytokine TNF-α during embryogenesis, insulitis develops very early, contrasting with the delayed peri-islet infiltrate observed in RIP-CD86 transgenic mice (first observed at 7 mo of age). Second, TNF-α expression by islet cells seems to impair the development of Th2 cells occurring in RIP-CD86 transgenic mice. Although TNF-α has not yet been directly implicated in skewing T cell differentiation toward a Th1 pathway, the expression of both TNF receptors I and II by activated T cells leaves open the possibility that TNF-α can affect T cell responses (54).

As previously described (36), both CD4⁺ and CD8⁺ cells are present within the islets of RIP-TNF-α transgenic mice. Interestingly, the representation of CD8⁺ cells is dramatically lower (≤10% (n = 8)) and the representation of CD4⁺ cells increased (≅43%; n = 8) in RIP-TNF-α single transgenic mice compared with that in diabetic double transgenic CD80/TNF and CD86/TNF mice that show 24% (n = 8) and 28% (n = 6) CD8⁺ cells, respectively, and 27.5% (n = 8) and 26.2% (n = 5) CD4⁺ cells, respectively. The activation phenotype of islet-infiltrating lymphocytes in nondiabetic RIP-TNF-α was analyzed to dissect further the consequences of local expression of TNF-α. To our surprise, cytokine-positive cells were also present in the islets of RIP-TNF-α transgenic mice. Both IFN-γ- and IL-4-positive cells are detected, but at a lower frequency compared with prediabetic or diabetic double transgenic CD86/TNF mice (Fig. 6 C). Additional experiments are required to determine whether this accumulation of cytokine-producing cells reflects a direct effect of TNF-α on the activation of islet-specific T cells or simply the preferential recruitment, by TNF-α, of activated cells within the tissue. Likewise, it will be important to determine to what extent the effects of TNF-α expression on lymphocyte activation affect the immunoreactivity toward islet-specific Ags. Nonetheless, these observations raised the possibility that the development of diabetes in CD86/TNF double transgenic mice might result from a direct effect of TNF-α on islet-specific T cells that could over-ride the propensity of CD86 to induce Th2 cells.

We have previously shown that expression of a high level of MHC class II IE Ags on islet cells that by itself was unable to induce autoimmunity would in conjunction with CD80 expression by islets allow the activation of autoreactive T cells and autoimmune destruction of β cells (33). Indeed, we showed that while neither transgenic mice expressing the class II IE molecule (Ins-IE) nor the CD80 costimulatory molecule on islets develop autoimmune diabetes, all double transgenic mice (CD80/IE) develop insulitis at about 2 mo of age that progresses to a specific autoimmune destruction of β cells. As previously reported, Ins-IE single transgenic mice exhibit a nonautoimmune form of diabetes that is also observed in double transgenic mice coexpressing the IE and CD80 molecules on islets (33, 37). However, only double transgenic mice showed insulitis and a specific loss of insulin-producing β cells that was associated with the activation of islet-specific, IE-restricted, autoreactive T cells (33). Thus, providing CD80-expressing islets with a high level of expression of class II IE Ags was sufficient to induce autoimmune diabetes.

Using this transgenic mouse model, we could determine whether CD86 expression on the islets could contribute to the activation of autoreactive T cells and β cell destruction in the absence of a local inflammatory reaction. We thus crossed Ins-IE mice with RIP-CD86 transgenic mice. Double transgenic mice coexpressing CD86 and class II IE molecules on islets (CD86/IE) developed insulitis starting at about 5 mo of age (not shown). As the insulitis progressed, islet architecture became completely disorganized, and islet tissue was replaced by a ductal hyperplasia (Fig. 8,Ab). Similar to CD80/IE double transgenic mice, most CD86/IE double transgenic mice develop clinical signs resembling diabetes, mainly weight loss and dehydration, that led to death, in most cases at about 6 mo of age. Both single transgenic Ins-IE and double transgenic CD86/IE mice showed elevated blood glucose levels. Indeed, as previously reported (37) Ins-IE mice showed a blood glucose level of 300 mg/dl at 8 wk of age that increased to an average of 412 mg/dl at 4 to 6 mo of age. Low levels of insulin production by β cells in Ins-IE mice (Fig. 8,B), although not adequate to fully control glucose levels in blood or urine, appeared sufficient to maintain basal glucose metabolism. Similarly, young CD86/IE double transgenic mice showed blood glucose levels averaging 363 mg/dl at 8 wk of age and rising to 402 mg/dl at 4 to 6 mo of age when clinical signs were marked. Disease progression in double transgenic CD86/IE was associated with a specific loss of insulin-producing cells, while glucagon-producing α cells were spared (Fig. 8 B). As destruction of insulin-producing β cells increased, double transgenic CD86/IE mice developed clinical signs of diabetes and eventually died. In this case, and as was observed in CD80/IE double transgenic mice, disease progression and death could be controlled by daily injection of insulin, further suggesting that CD86/IE double transgenic mice suffer from autoimmune destruction of insulin-producing β cells. Some double transgenic mice, however, survived for longer periods of time while remaining diabetic. Pancreata isolated from these mice also showed a completely disorganized islet architecture with, however, residual insulin-producing cells. Further study indicated that the remaining β cells expressed a reduced level of class II IE Ag and no CD86, possibly allowing escape from immune destruction (not shown). Persistence of a few insulin-producing cells might, in this case, suffice to maintain a basal level of glucose metabolism. CD86/IE double transgenic mice thus resemble CD80/IE double transgenic mice, demonstrating similar autoimmune destruction of β cells.

We further investigated the phenotype of islet-infiltrating lymphocytes. Similar number of B cells and CD4⁺ and CD8⁺ T cells were found in the lymphocytic infiltrate of CD86/IE compared with CD80/IE transgenic mice. Interestingly, the representation of IL-4- and IFN-γ-positive cells detected in either CD80/IE or CD86/IE double transgenic mice greatly resembled the frequencies detected in CD80/TNF or CD86/TNF transgenic mice, respectively. Indeed, while the number of cytokine-positive cells was higher when transgenic mice coexpressed the class II IE molecule together with the CD86 costimulatory molecule on β cells compared with CD80, the ratio of IFN-γ-positive over IL-4-positive cells was similar in both transgene combinations (Fig. 9). Thus again in this double transgenic mouse model, CD80 and CD86 exhibit similar qualitative abilities to activate autoreactive T cells.

To define better the intrinsic properties of CD80 and CD86 in providing costimulatory signals for the activation of tissue-specific autoreactive T cells, we generated transgenic mice that express either costimulatory molecule on the islets of Langerhans. By directly providing tissue cells with costimulatory function we sought to bypass the requirement for professional APC in the initiation of the autoimmune process and thus reduce the resulting complexity. Indeed, some evidence suggests that B cells and macrophages might show a differential ability to induce the differentiation of Th1 or Th2 cells. Moreover, by this approach we could avoid any misinterpretation linked to differential regulation of expression of CD80 and CD86. Indeed, we could compare mice with constitutive and similar levels of CD80 and CD86 expression on the same tissue. To generate our transgenic mice we chose to use a transgene coding for human CD80 or CD86, which could readily be distinguished from the mouse homologue. We could thus unambiguously select transgenic lines that show selective expression of the transgene on the β cells of the pancreas and no aberrant expression of the transgene in any other tissue, including APC.

In this paper we describe three lines of transgenic mice with different levels of expression of CD86 costimulatory molecules specifically on islet cells that all show the same fundamental phenotype and compare them with transgenic mice expressing CD80 on islets (33, 34). Our analysis, for the most part, has been performed using section staining to highlight what was occurring in situ, especially concerning cytokine secretion as well as BrdUrd analysis. This lead us to highlight some intrinsic differences between CD80 and CD86 in their abilities to provide costimulatory signals for the activation of autoreactive T cells in vivo. Moreover, this study suggests additional regulatory steps in lymphocyte trafficking that could contribute to the control of autoimmune responses.

All RIP-CD86 transgenic mice, independent of the level of CD86 expressed on β cells, develop a peri-islet infiltrate at the age of 7 to 8 mo that almost never progresses to insulitis or diabetes. Quite strikingly, the phenotype observed in RIP-CD86 transgenic mice resembles the phenotype in congenic NOD mice with a MHC class I H-2^b locus, which also develop a peri-islet infiltrate without insulitis (55). By two criteria, proliferation and cytokine production, we could show that activated cells accumulate in the peri-islet infiltrate. An essential question, then, is why are these activated cells not causing diabetes?

It is first important to state that we have no direct proof that this activation is islet specific. Several lines of indirect evidence suggest, however, that the peri-islet infiltrate observed in RIP-CD86 transgenic mice is a consequence of expression of CD86 by β cells and thus involves islet-specific T cells. First, in no other organ did we find expression of the transgenically encoded CD86 costimulatory molecule or observe any inflammatory responses. Second within the pancreas, lymphocytes accumulated specifically in the perivascular or periductal area in the proximity of an islet, and no other pancreatic cell types except β cells expressed the transgenic CD86 molecule. Third the infiltrates associated with the islets contain activated lymphocytes. Under these conditions, the lack of insulitis is more likely due to an inability of lymphocytes to persist within the islets. The only peculiarity of the infiltrate observed in RIP-CD86 transgenic mice is the over-representation of Th2 cells. Reduced expression of inflammatory Th1 cytokines or overexpression of Th2-type cytokines might then preclude the expression of appropriate adhesion molecules required for T cell trafficking. Interestingly, we found no up-regulation of the expression of VCAM-1 or ICAM-1 by the vascular endothelium in the proximity of the infiltrate, although T cell extravasation seemed to have taken place (not shown). More recent studies have suggested an essential role for E- and P-selectins in the migration of Th1, but not Th2, cells toward an inflammatory site (56, 57). Expression of E- and P-selectin that is preferentially induced by inflammatory cytokines is, however, not down-regulated by IL-4 (58, 59). Human Th2 lymphocytes have been shown to express the eotaxin receptor CCR3 (60). Chemokines and their receptors, which are expressed differentially, may also regulate access of different Th subsets into sites of inflammation.

Alternatively, the lack of insulitis and diabetes in RIP-CD86 transgenic mice might result from insufficient CD4⁺ T cell help. Indeed, while CD8⁺ T cells are required for the initiation of autoimmune diabetes (49, 50), activated CD4⁺ T cells are capable of mediating tissue destruction (61, 62). Further, Kurts et al. recently showed that CD4⁺ T cells are essential in preventing deletion of self-reactive CD8⁺ T cells and thus contribute to the development of autoimmune diabetes (63). In RIP-CD86 transgenic mice, CD4⁺ T cell help is probably limiting, since islet cells that do not normally express MHC class II molecules are presumably the only APC. Interestingly, CD86 transgenic mice that also express MHC class II IE on their islets develop autoimmune destruction of their islets. This, however, contrasts with double transgenic mice that coexpress CD86 and TNF-α molecules on their islets and that develop spontaneous diabetes despite a lack of expression of MHC class II molecules by β cells.

Additional experiments are clearly required to elucidate by which mechanisms insulitis is prevented in RIP-CD86 transgenic mice. Interestingly, however, these observations suggest a novel, unsuspected mechanism by which Th2 responses could control self-reactivity. Indeed, in such a model, inhibition of autoaggressivity would not be through the negative regulation of an anti-self Th1 response but, rather, simply by preventing tissue access to potentially harmful cells.

Our comparative study of RIP-CD86 and RIP-CD80 transgenic mice suggests that CD80 and CD86 have different abilities to support self-reactive T cell activation in vivo. Indeed, the phenotype of RIP-CD86 transgenic mice contrasts quite dramatically with that of transgenic mice expressing CD80 on islets that only rarely (<3% of the mice) develop insulitis (34). These differences, however, are obliterated when islets also express high level of MHC class II Ag or the inflammatory cytokine TNF-α. Indeed, then, CD80 and CD86 show similar abilities in sustaining the activation of self-reactive T cells, leading to β cell destruction. We have suggested that tissue cells, even when provided with costimulatory molecules such as CD80, are poorly immunogenic for naive T cells, probably because they express insufficient MHC class I Ag that would be below the threshold required for T cell activation. Increased levels of MHC class I Ag were observed when TNF-α was expressed by islets. Likewise, Ins-IE transgenic mice expressed high levels of transgenic class II IE. It appears, then, that when TCR signaling is low, as is observed in single transgenic mice, RIP-CD80 or RIP-CD86, only CD86 can induce a certain level of activation. In contrast, under conditions where TCR signaling is high, due to higher levels of MHC class I or II expression, as is observed when TNF-α or IE are expressed on islets, CD80 and CD86 are similarly potent at inducing autoimmune destruction of β cells. Increased TCR engagement seems then to over-ride the signaling differences resulting from CD28 interaction with CD80 or CD86.

Interestingly, in all the transgenic and double transgenic mouse models we analyzed, CD86 was always more potent than CD80 at stimulating cytokine production (Figs. 5 and 9). This suggests that CD86 is more efficient than CD80 at supporting effector T cell activation in vivo. The increased number of activated T cells within the infiltrated islets, however, did not accelerate the disease process in CD86/TNF-α or CD86/IE mice. Two different interpretations might explain this finding. Following on the idea that the level of TCR signaling determines the extent of T cell activation, we would suggest that the differences observed in our transgenic mice simply reflect differences in CD28 signaling. A corollary would be that CD28 engagement lowers the threshold for T cell activation (64). In such a model, the interaction of CD28 with CD86 would be more potent than CD28 interaction with CD80. Thus, the appropriate threshold for T cell activation would be met in RIP-CD86 transgenic mice but not in RIP-CD80 transgenic mice. This is, however, at least in part in contradiction with the finding that CD80 has a slightly higher avidity for CD28 than does CD86 (11). Interestingly, the interaction of CD80 and CD86 with CTLA-4 follows the same hierarchy (11). Recent studies have clearly highlighted the role that CTLA-4 plays in the negative regulation of T cell activation (65, 66, 67). It is thus possible that the reduced level of T cell activation observed in the different transgenic mice expressing CD80 on β cells reflects stronger negative regulation through CTLA-4 or a better balance of the two signals.

It is important to state that such interpretations rely on the reported affinity of human CD28 or CTLA-4 for human CD80 or CD86. It is not yet clear whether the same hierarchy of affinities is observed in mice, or indeed whether this can be applied to mouse receptors and human ligand, as in our transgenic mice models. Mice expressing mouse (35) or human (33, 34) CD80 on the islets of Langerhans, however, showed the same fundamental phenotype, suggesting that the reported lower affinity of human CD80 for mouse CD28 or CTLA-4 does not significantly alter the in vivo response.

We found in all infiltrated islets from RIP-CD86 transgenic mice a high number of IFN-γ-producing cells and a 3- to 10-fold higher number of IL-4-producing cells. The over-representation of Th2 cells in the islets of RIP-CD86 mice is significant, since Th1 responses are, under most priming conditions, dominant in C57BL/6 mice. This bias toward a Th2 response, however, is obliterated when islets coexpress CD86 and TNF-α or a high level of MHC class II Ags. It has been suggested that differentiation of T cells along the Th1 or the Th2 pathway might be determined by the density of ligand expressed by APC (68, 69). Thus, low to intermediate ligand density would preferentially induce a Th2 response, while high ligand density preferentially elicits a Th1 response. Interestingly, in RIP-CD86 single transgenic mice, in which tissue cells express low levels of MHC class I Ags and thus low ligand density, a Th2 response dominates. In contrast, in either CD86/TNF-α or CD86/IE, where islets express high levels of MHC class I or class II Ag, respectively, a Th1 response dominates.

Together these results are compatible with a threshold model for T cell differentiation by which the threshold for the initiation of transcription of the Th2 program of gene expression would be lower than the threshold requirements for the Th1 program of gene expression. Several studies have already suggested that a different level of TCR signaling can induce different T cell responses in both CD4⁺ and CD8⁺ T cells (70, 71). We would thus suggest that at low ligand density, the interaction of CD28 with CD86, but not with CD80, could reach the threshold for T cell activation. This level of signaling would however only reach the threshold required for Th2 transcription. When MHC expression by islets is increased, then both CD80 and CD86 are sufficient, and the threshold allowing Th1 differentiation is now met. As expected in such models, Th1 and Th2 responses could coexist and would simply reflect the heterogeneity of TCR avidity for a given MHC/peptide complex as well as the level of expression of any given autoantigen. Interestingly, as observed in our transgenic mice models, Th1 and Th2 cells can coexist without interfering with tissue damage. This does not, however, preclude the possibility that over-representation of Th2 cells can interfere with the development of diabetes. This is, in fact, exemplified in RIP-CD86 and RIP-TNF-α single transgenic mice that do not develop disease and show ratios of IFN-γ⁺/IL-4⁺ cells of <1 and 2, respectively, while diabetes-prone mice show an average ratio >5.

Numerous studies have, over the past few years clearly shown that an anti-self immune response does not necessarily lead to tissue inflammation and autoimmune disease. This is clearly exemplified in transgenic mouse models where lymphocytic infiltration and autoimmune tissue destruction develop only when the Th2 anti-self immune response subsides while the Th1 response remains unaltered (52, 72). Quite unexpectedly we found, through analysis of both RIP-CD86 and RIP-TNF-α transgenic mice, that in two cases in vivo tolerance can persist despite the accumulation of activated, potentially self-reactive T cells in the pancreas. These results suggest an additional regulatory step in the maintenance of self-tolerance beyond a simple Th1/Th2 imbalance. It is worth noting that 2 of 70 RIP-CD86 mice analyzed developed insulitis and diabetes at 12 mo of age. Interestingly, these two mice demonstrated a severe pancreatitis exemplified by a complete destruction of all pancreatic tissues, including acinar and β cells, that resemble the disease observed in mice suffering from Th2-mediated diabetes (73). These observations strongly indicate that both Th1 and Th2 cells can contribute to autoimmune destruction of tissue by mechanisms that clearly need to be clarified.

We thank A. Bothwell for the CD86 cDNA, and A. Azuma for the IT2.2 hybridoma; Cindy Hughes and Debbie Butkus for generating transgenic mice; Judith Miller for technical help; Irene Visintin for suggestions and support with the cytokine staining; and Fran Manzo for secretarial assistance.