CD4⁺CD25⁺ T Regulatory Cells Dependent on ICOS Promote Regulation of Effector Cells in the Prediabetic Lesion

Type 1 diabetes is an autoimmune disease that develops when tolerance mechanisms fail to control immune responses to pancreatic islet β cell proteins (1). Studies of autoimmune diabetes are facilitated by mouse models, such as NOD mice that develop diabetes spontaneously and TCR transgenic mice that express autoimmune TCR specificities present within the natural NOD repertoire (1–8). Among these, BDC2.5 TCR transgenic (BDC2.5) mice express a TCR derived from a diabetogenic CD4⁺ T cell clone, restricted by the MHC class II molecule A^g7 (4, 9). Studies of diabetes pathogenesis in both BDC2.5 and NOD mice demonstrate that diabetes unfolds in stages (10). In BDC2.5 NOD mice, cells invade at 15–18 d of age and set up a massive infiltrate inside the islet (insulitis; references 4, 11). Yet, progression to diabetes occurs in only 10–20% of animals at ∼20 wk of age. In most animals, the lesion settles into a state of “respectful” insulitis, with the inflammatory infiltrate clearly demarcated from functional β cells. CTLA-4 blockade in BDC2.5/NOD or crossing to the B6^H2-g7 background results in a far more aggressive lesion with β cell destruction and rapid diabetes onset (11–13). What cellular mechanisms explain nondestructive insulitis in BDC2.5/NOD? Either T cells entering the islets in young mice are phenotypically incompetent to cause disease, or they are actively prevented from doing so by a CTLA-4–dependent immunoregulatory mechanism.

Immune regulation by CD4⁺CD25⁺ T regulatory cells (Tregs) limits autoimmunity in several mouse models (14–16). The FoxP3 transcription factor appears to play a unique role as a master regulator of their phenotypic and functional properties (17–19). In vivo, CD4⁺CD25⁺ Tregs depend on mediators such as IL-10, TGF-β, and/or CTLA-4, but their exact mechanism of action has not been fully elucidated (20–25). Several studies of Tregs in transgenic models have demonstrated that the cells expand in vitro and in vivo to antigen presented by mature dendritic cells (26, 27), or in draining LNs of sites expressing cognate antigen (28–30). Transfer studies showed that CD4⁺CD25⁺ Tregs home to the lamina propria and proliferate there during regulation of induced colitis, although proliferation in this system could not be separated from the lymphopenic state of the host (30). Still at issue is the localization, proliferative capacity, and function of Tregs in a spontaneous autoimmune disorder. Do the Tregs prevent the initial activation of effector cells in the lymphoid organs, or do they act secondarily on effector cells in the target tissues?

Multiple lines of evidence suggest that CD4⁺CD25⁺ Tregs are important in preventing responses in diabetes settings. Depletion of CD25⁺ cells in a transfer setting or genetic absence of CD28 (leading to low numbers of Tregs in NOD mice) provoked exacerbation of disease (31, 32). Administration of IL-10, vitamin D, or TNF-α to NOD mice promoted the generation of Tregs in conjunction with amelioration of diabetes (33–35). Similarly, CD4⁺ CD25⁺ Tregs may be reduced in the peripheral blood of people with recent onset diabetes (36). Restoration of this cell type may explain exciting recent advances in human diabetes treatment. The generation of TGF-β–producing Tregs by administration of anti-CD3 appears responsible for the efficacy of this drug in early clinical trials (37, 38). Thus, understanding the mechanism and point of action of this important cell type may yield advances in human type 1 diabetes therapy.

Previous studies of regulatory cells in BDC2.5 mice described clonotype^lo suppressor cells that came up in older animals (39) or DX5⁺ populations effective in preinsulitic mice (13). Here, we find that a defined CD4⁺CD25⁺ T cell population with active regulatory properties can account for respectful insulitis in the prediabetes state in the BDC2.5 mouse model of type 1 diabetes. Tregs coexist in balance with aggressive effector cells directly in the autoimmune infiltrate, and express high levels of the immunoregulatory cytokine IL-10 in the pancreas compared with the draining LN. Loss of this balance through blockade of the costimulatory receptor inducible costimulator (ICOS) leads to the rapid onset of diabetes.

Single cell suspensions were prepared from whole pancreas or indicated LNs using glass slide disruption. Debris was allowed to settle and removed (repeated two to three times for pancreas). Staining with live cell dye was critical, especially with pancreas. We used either 7-AAD (BD Biosciences) or Hoescht 33342 (CN Corp.). Cells were stained using standard methodology with a combination of the following: B220-PE-Texas red (RA3-6B2; Caltag), CD4-PE or PE-Cy7 (GK1.5), ICOS-PE (7E.17G9), CD69-FITC (H1.2F3), CD25-allophycocyanin (7D4 or PC61), or isotype controls rat IgG1 (R3-34) and rat IgG2b (A95-1; BD Biosciences). Cells were analyzed on a MoFlo^® flow cytometer using Summit^® software (DakoCytomation). Sorted cells were collected into media with 5% FBS and assessed for purity (93–98% in transfer studies). For RNA preparation, cells were sorted twice to >98% purity.

BDC2.5/NOD mice were maintained in the Joslin Diabetes Center barrier facility (IACUC protocol 99-19, 99-20). Pancreata or total pooled LNs were isolated from 3–4-wk-old BDC2.5 mice. Live B220⁻CD4⁺ cells were sorted according to CD25/CD69 phenotypes and injected i.p. into 4–12-wk-old NOD.scid recipients (The Jackson Laboratory). Diabetogenic BDC2.5 splenocytes were obtained from 6–10-wk-old mice. Mice were monitored for diabetes as described previously (4).

Total RNA was isolated from sorted cells by standard TRIzol method from 10⁴ cells of each group, and cDNA was prepared using Superscript II^® reverse transcriptase (Invitrogen). cDNA from 1,250 cell equivalents was used for real-time PCR for FoxP3 or IL-10 mRNA, using HPRT as a standard. 2^-ΔΔCT calculations were used to present amounts of expression relative to the whole LN control.

Tregs (CD4⁺CD25⁺CD69⁻) and T effector cells (Teffs; combined CD4⁺CD25⁻CD69⁻ and CD4⁺CD25^loCD69⁺) were sorted from pancreas preparations to >98% purity as aforementioned. For ICOS blockade comparisons, mice were treated with two doses of anti-ICOS mAb or control rat IgG as indicated. Cells were lysed in TRIzol, and total RNA was prepared according to the manufacturer's instructions (Invitrogen). A method developed in-house was used to consistently amplify RNA from small cell populations (Goldrath, A., personal communication). Amplification is primarily linear; additionally, any distortion of data is eliminated when samples amplified the same number of times are directly compared. This procedure permitted analysis of transcripts from as few as 6,000 cells. We performed three to five independent experiments for each cell type. RNA was amplified for two rounds using the MessageAmp^® aRNA kit according to manufacturer's instructions (Ambion). The last biotinylation step was performed using the Enzo Life Sciences BioArray HighYield RNA Transcript Labeling Kit^® (T7). 5–15 μg of twice-amplified biotinylated aRNA was submitted to the Joslin Genomics Core for hybridization to murine genome U74Av2 array GeneChips^® as per the manufacturer's instructions (Affymetrix, Inc.). The raw data were processed with the RMA algorithm for probe-level normalization (40), (S⁺ Array Analyzer; Insightful; modified by R. Park [Joslin Diabetes Center, Boston, MA], D. Mathis, and C. Benoist), and composite expression values were calculated for each of the genes on the chip (averaging after outlier elimination; Distill software). Gene-wise p-values were calculated with Welch's modified Student's t test and converted to LOD scores as −log₁₀.

ICOS-specific 7E.17G9.G1 (41) and CTLA-4–specific UC10.4F10.11 (American Type Culture Collection) mAbs were purified from hybridoma culture supernatant by standard methods, either in-house or by Bio-Express or Harlan, and tested for endotoxin levels (<1.4 EU/mg protein). Each lot was titered, and doses were set based on Ab equivalents by flow cytometric activity (unpublished data). 7E.17G9.G1 mAb was injected i.p. in two doses of 50–100 μg at 9 and 12 d, or mice were injected with an equal amount of rat IgG2b isotype control (A95-1; BD Biosciences), rat IgG (Sigma-Aldrich), or PBS; or 400 μg of UC10.4F10.11 anti–CTLA-4 mAb as indicated. For anti-ICOS treatment of older mice, two doses of 200–400 μg 7E.17G9.G1 mAb were injected i.p. in PBS at 21 and 24 d, or starting at 26 or 35 d as indicated. In Fig. 6 A (right), an equal volume of post-in-house mAb elution supernatant was used as a control for endotoxin contamination (contains small amounts of residual 17G9).

Specific genes identified by GeneChip^® array data from pancreatic Tregs and Teffs are presented in Table SI. Datasets are available in the public Gene Expression Omnibus database (accession no. GSE1419).

Initially, we asked whether cells in the prediabetes pancreatic lesion of BDC2.5 mice (Fig. 1 A) are simply incompetent to cause disease, or whether there is an active regulatory process occurring in the islets. Preparation of islets followed by isolation of lymphocytes proved detrimental to identifying all T cells because heavily infiltrated islets fell apart during isolation and were not included in the final yield. Therefore, infiltrating T cells were prepared by mechanical disruption of whole pancreata of young 3–4-wk-old BDC2.5/NOD mice after careful removal of any LNs. The cells obtained (routinely 3–5 × 10⁶ per BDC2.5/NOD pancreas) were a true component of the infiltrate in that lymphocytes could not be isolated from pancreata of strains with no islet infiltration (unpublished data). We divided the CD4⁺ population into three subsets on the basis of the early activation marker CD69 and the activation/regulatory marker CD25: CD4⁺CD25⁺CD69⁻, CD4⁺CD25^loCD69⁺, and CD4⁺ CD25⁻CD69⁻ (Fig. 1 B). Although CD4⁺ cells in the irrelevant inguinal LN (ILN) were mainly naive, those from the pancreatic LN (PLN) or from the pancreatic lesion were highly activated, with 15–40% of CD4⁺ cells displaying a CD25^loCD69⁺ profile consistent with recent activation. 3–11% of pancreatic infiltrating cells had a CD4⁺ CD25⁺CD69⁻ phenotype, consistent with Tregs.

This phenotype suggested that the CD4⁺CD25⁺CD69⁻ cells might contain the regulatory fraction, whereas those CD69⁺ cells that were also CD25⁺ might reflect expression of the CD25 marker as a sign of recent activation. To examine this issue, we sorted each cell type independently from the pancreatic lesion, and assessed the expression of the regulatory cell transcription factor FoxP3. Real-time PCR analyses of the three populations indicated that FoxP3 expression was highly enriched (10–35-fold in different experiments) in the CD4⁺CD25⁺CD69⁻ population relative to the CD4⁺CD25^loCD69⁺ or CD4⁺CD25⁻ CD69⁻ populations (Fig. 1 C), as in CD4⁺CD25⁺ cells isolated from LNs. Even though the CD25^loCD69⁺ cells express some CD25, this did not reflect increased FoxP3 expression. CD69 appears to be a useful marker to separate recently activated cells (which also express CD25) away from CD25⁺ cells expressing the regulatory transcriptional controller FoxP3. A majority of the CD25⁺ CD69⁻ cells (90%) expressed the clonotype receptor, as did the other phenotypes (unpublished data). These data suggest that a balance of putative Tregs and Teffs expressing the BDC2.5 receptor resides together directly in the target organ.

An interesting issue was whether the T cells were actively proliferating in the pancreatic lesion because previous analyses of Treg proliferation have shown expansion in the draining node or in the intestine of lymphopenic hosts (30), but not the target tissue of autoimmunity in mice with a full lymphoid compartment. Flow cytometric analysis of mice fed with BrdU for 5 d showed that all CD4⁺ cell phenotypes in the pancreatic lesion proliferated (Fig. 2). If anything, CD25⁺ CD69⁻ cells were the most active. This was also the case in the draining PLN, but not in the irrelevant ILN. Thus, the insulitic lesion is not inert, and cells with both regulatory and effector phenotypes are turning over rapidly in the presence of a full lymphoid compartment.

These data are suggestive of an active balance of regulatory and pathogenic CD4⁺ T cells in the pancreatic lesion. To evaluate function, we sorted CD25⁺CD69⁻, CD25^loCD69⁺, and CD25⁻CD69⁻ cells from BDC2.5/NOD pancreata were transferred into NOD.scid recipients. Mice that received CD25⁻CD69⁻ cells or CD25^loCD69⁺ cells developed diabetes shortly after transfer (Fig. 3 A). As few as 10–20,000 cells of these phenotypes were enough to provoke diabetes in 50–80% of recipients within 1 mo after transfer. All mice that received CD25⁻CD69⁻ cells or CD25^loCD69⁺ cells, at the doses tested, developed insulitis (unpublished data). In contrast, CD25⁺CD69⁻ cells did not cause diabetes or develop insulitis after transfer. We asked whether the CD25⁺CD69⁻ cells residing in the pancreas did, in fact, have protective capacity by transferring 10–20,000 CD25⁺CD69⁻ cells isolated from the lesion with 5 × 10⁵ diabetogenic BDC2.5/NOD splenocytes into NOD.scid recipients. This small dose of cells, 10–100-fold lower than the dose used in other models of regulation using polyclonal populations (20, 42), provided significant protection from diabetes in this cotransfer context (Fig. 3 B). BDC2.5 LN Tregs (CD4⁺CD25⁺CD69⁻) were also effective at this dose (unpublished data). In vivo regulation was dose dependent because protection from diabetes was lost if only 5,000 Tregs were cotransferred (Fig. 3 B). In protected mice that had received 20,000 CD25⁺CD69⁻ Tregs and 5 × 10⁵ diabetogenic BDC2.5/NOD splenocytes, we observed formidable, but respectful, insulitis in the islet after 30 d (Fig. 3 C), similar to that seen in islets of unmanipulated BDC2.5/NOD mice (compare with Fig. 1 A). Interestingly, Tregs taken directly from the pancreatic lesion were unable to set up a presence there in the absence of other T cells in the NOD.SCID host, suggesting that they may require signals from Teffs to access the pancreas or persist there. These data demonstrate that Teff and Treg subsets coexist in the pancreatic lesion, and that T effectors capable of disease transfer are being actively suppressed by CD25⁺CD69⁻ Tregs during the nondestructive period of insulitis.

To better understand the function and phenotype of effector and regulatory cells residing in the target organ during autoimmunity, we compared the gene expression profiles of CD25⁺CD69⁻ Tregs to the combined Teff populations of CD25^loCD69⁺ and CD25⁻CD69⁻ cells. We sorted cells from the lesion, isolated RNA, prepared double amplified biotinylated aRNA, and hybridized each sample to an Affymetrix murine genome U74Av2 array GeneChip^® for at least three replicates of each population. A false positive rate (FPR) was estimated from a randomized dataset generated by shuffling expression values between samples (Fig. 4 A, randomized control). The CD25⁺CD69⁻ Tregs and Teff cell populations from the prediabetic lesion could be distinguished by the expression of a limited set of genes; with a 10% FPR cutoff, 96 genes were overexpressed in CD25⁺CD69⁻ cells, and 73 were underrepresented, significantly more than in the randomized dataset (Fig. 4 A). Several of the genes overexpressed in CD25⁺CD69⁻ cells such as GITR, CD103, Nrp-1, IL-10, and CTLA-4 are diagnostic of Tregs with suppressive activity (Table SI, available at http://www.jem.org/cgi/content/full/jem.20040179/DC1; references 14–16, 43–46). Overall, this group was dominated by cell surface or cytokine genes. Conversely, the genes underrepresented in Tregs seemed dominated by transcription or signaling factors (OBF-1, Tcf7, Eomesodermin, and Smad1). We compared our data to 77 genes shown previously to be overrepresented in LN CD4⁺CD25⁺ cells from normal mice or from in vitro–activated cultures (45); as illustrated in Fig. 4 B, many of these were also overrepresented in pancreatic CD25⁺CD69⁻ Tregs (21 out of 77 overlap with our list at 10% FPR; Table SI). However, many of the genes on our list have not been identified previously as particularly expressed by Tregs. They might represent a unique subset of genes either induced in a subset of Tregs to make them capable of accessing the tissues, or in Tregs once within the lesion.

To compare more directly the differences between Tregs found in tissue and their LN counterparts, we examined the expression of several regulatory molecules by flow cytometry or real-time PCR. IL-10 is an important effector molecule for Tregs in several in vivo systems (for review see references 14–16) and has been shown to be critical for regulation in the BDC2.5 model (47). We observed that IL-10 was highly up-regulated in Treg in the pancreatic lesion compared with effector cells (Table SI). We compared mRNA levels of IL-10 in Tregs isolated from the pancreatic infiltrate or LN. Unlike FoxP3 levels, which were similar between LNs and pancreatic Tregs (Fig. 1), IL-10 levels were 15–36-fold higher in pancreatic Tregs compared with those isolated from the PLN or irrelevant mesenteric LN (MLN; Fig. 5 A). The GITR molecule was also consistently present at somewhat higher levels on pancreatic versus LN Tregs (Fig. 5 B), as well as increasing on activated cells in the pancreas. ICOS, a CD28 family member, is known to be an important costimulatory receptor for the production of IL-10 (48–51). Although ICOS is usually thought to act as a positive costimulator, genetic deficiency or Ab blockade of ICOS leads to exacerbation of disease in mouse models of multiple sclerosis (48, 51). ICOS was expressed on a small subpopulation (5–10%) of CD25⁺CD69⁻ Tregs in the LN, although interestingly ICOS was always lower or absent on PLN Tregs (Fig. 5 C). Strikingly, 30–80% of CD25⁺CD69⁻ Tregs in the pancreas expressed ICOS. Recently activated cells (CD25^lo CD69⁺) showed no comparable ICOS expression. In some experiments, ICOS was present on CD25^loCD69⁺ cells in the pancreas (2–16%), but was of lower intensity. CD25⁻ CD69⁻ cells did not express ICOS (unpublished data). These data suggest that Tregs in the autoimmune lesion acquire functional and phenotypic properties distinct from their LN counterparts, in particular marked by high expression of ICOS. Higher levels of the regulatory cytokine IL-10, GITR, and ICOS in Tregs specifically from the pancreatic lesion suggest that Tregs function directly in the target organ to mediate regulation.

Abrogation of costimulatory signals in both the NOD and BDC2.5 models has been shown to lead to rapid onset of diabetes (32); in particular, blockade of CTLA-4, known to be important for Treg function, led to immediate onset of diabetes in young BDC2.5/NOD mice if administered when they first exhibited insulitis (11, 21, 22). To test if ICOS on Tregs in the pancreas was important to maintain the balance of regulation, we treated young BDC2.5/NOD mice with anti-ICOS mAb. These animals developed diabetes rapidly at the onset of insulitis, all becoming diabetic by 25 d of age (Fig. 6 A, left). In contrast with the CTLA-4 blockade, which is only effective when given before islet infiltration, ICOS blockade could promote diabetes for a period after insulitis onset as well (Fig. 6 A, right), although the effect also waned in older animals.

We analyzed the frequency of CD4⁺ CD25/CD69 T cell subsets within the pancreas after ICOS blockade. ICOS blockade led to an increased frequency of CD25^loCD69⁺ Teff cells and a decreased frequency of CD25⁺CD69⁻ Tregs, resulting in a significant change in the ratio of the two subpopulations (Fig. 6 B). This change only occurred in the pancreatic infiltrate, but not in the PLN or irrelevant LN, reflective of ICOS expression specifically in the pancreatic infiltrate. Anti-ICOS treatment did not lead to deletion of Tregs because the absolute numbers of Tregs in the lesion did not change; rather, an increase in Teff was responsible for the skewing of the Teff/Treg ratio (unpublished data). The blockade of ICOS perturbed the Teff/Treg ratio, thereby disrupting the balance of regulation in the lesion.

Microarrays were used to analyze gene expression changes induced by ICOS blockade in the two populations. Teff or Treg CD4⁺ subsets were sorted as aforementioned from 4-wk-old BDC2.5/NOD mice 5 d after treatment with anti-ICOS or control rat IgG (Fig. 7 A), and RNA probes were prepared for microarray analysis as aforementioned. Changes induced by ICOS blockade were more limited in the effector population than in Tregs; the comparative plot was much tighter for Teff than for Tregs (Fig. 7, B and C). In Teff, only 33 genes were induced by a factor of ≥1.5-fold, and only 1 was repressed. Interestingly, a number of these induced genes have also been found activated in a recent analysis of pancreatic destruction induced by cyclophosphamide treatment of BDC2.5/NOD mice, including IFN-γ and several IFN-γ–regulated genes (Matos, M., personal communication).

In contrast, the changes affecting Tregs after ICOS blockade are far more profound, 215 and 21 genes were affected at the 1.5-fold cutoff. The plot of Fig. 7 D is highly informative. First, it shows that essentially all genes that are under- or overexpressed in control-treated Tregs compared with control-treated Teff remain so after anti-ICOS administration. This provides an independent validation of the “Treg signature” described in Fig. 4. Most importantly, it also becomes clear that the differential expression of the entire profile between Treg and Teff is lessened after ICOS blockade, as evidenced by the off-diagonal shift. This “muting” of the Treg signature affects genes that are both over- or underrepresented in Tregs (IL-10, GITR, and SOCS-2 or Tcf7, Eomesodermin, and Myb, respectively). It is as if ICOS blockade induced a broad shift of gene expression away from the Treg profile, toward one that is more similar to other CD4⁺ cells. We considered the caveat that the CD25⁺CD69⁻ cell population might be more contaminated with effector cells after ICOS blockade, but ruled out this possibility by simulating computationally what the Treg gene expression values would look like given 5, 10, or 50% contamination by the Teff population. Greater than 35% contamination would be required for the shift seen in Fig. 7 D, an unlikely scenario given the decreased frequency yet static absolute numbers of CD25⁺CD69⁻ cells after ICOS blockade (Fig. 6 B and not depicted), and given the similar FACS^® profiles with anti-ICOS or control treatment (Fig. 7 A). Another possible explanation is that a population of CD25⁺ICOS⁺FoxP3⁺ cells is lost via antibody-induced cell death, and replaced by a CD25⁺CD69⁻ effector cell population, thus keeping the numbers of cells with this phenotype constant. However, ICOS⁺ cells could still be detected at similar frequencies after anti-ICOS treatment using labeled anti–rat IgG (unpublished data), suggesting that there was no large composition change within the CD25⁺ population.

The two standout genes that were exceptions to the down-modulation of the Treg profile were IFN-γ and Ccl-3 (MIP1α). Paradoxically, these proinflammatory genes were overrepresented in Tregs without treatment and further induced by anti-ICOS in both Treg and Teff populations. Cytokines represented in the lesion switch from an immunoregulatory to a proinflammatory profile (Fig. 7 E). Thus, ICOS blockade induces changes in the relative numbers and in the broad gene expression profiles of the Treg population, suggesting that ICOS engagement is necessary to maintain the activity of the Treg populations in the lesion, and thus prevent β cell destruction.

Several mouse models of autoimmune diabetes share the feature of exhibiting a long quiescent period between the initiation of insulitis and the onset of diabetes (10). Similarly, the presence of islet-directed antibodies and T cells in the blood of prediabetic individuals years before diagnosis of diabetes suggests that the human disease might unfold in these same stages (52, 53). In animal models, where it can be assessed, the prediabetic state coincides with a local autoimmune infiltrate that refrains from terminal β cell destruction. This work aimed to unravel the localization as well as cellular and molecular mechanisms that control nondestructive insulitis, critical because this is likely to be the stage at which many therapeutic interventions will be attempted. We demonstrated the presence of both pathogenic Teffs and active CD4⁺CD25⁺ Tregs within the lesion of young prediabetic mice. Both were highly effective, very few cells being required to effect full islet destruction or, conversely, to exert full protection. The ability to recreate a nondestructive lesion much like the one in unmanipulated young BDC2.5 or NOD mice by cotransferring diabetogenic splenocytes and Tregs provides evidence that locally acting Tregs are an important element in the establishment of nondestructive insulitis.

How do the tissue Tregs described here relate to previously described regulatory populations in the BDC2.5 system? Kanagawa et al. (39) found that BDC2.5 clonotype^lo cells had regulatory properties, but these cells were not CD25⁺ and developed as the mice aged, thus likely represent a different regulatory population. Gonzalez et al. (13) demonstrated that cells capable of protecting 10-d-old BDC2.5/RAG^o mice from fulminant diabetes were not enriched in the CD25⁺ fraction, but resided, at least in part, in a population with markers, such as DX5, typical of NK T cells. This population also promoted a state of “respectful” insulitis, and had an effect directly in the islet lesion. The explanation for the divergence of these findings from the current ones may reflect differences in timing between the two experimental systems as follows: the protective effect of the DX5⁺ fraction required administration before the onset of insulitis. One might speculate that these cells are needed upstream of CD4⁺CD25⁺ Tregs, perhaps to promote their establishment.

We observed that both Teffs and Tregs were actively cycling in pancreatic infiltrate. Similarly, Tregs could be observed proliferating in LNs draining their specific antigen site (28, 29), or in the lamina propria during regulation of colitis in a lymphopenic environment (30). Together, these observations clearly indicate that, in vivo, Tregs are not anergic. Previous studies have suggested that antigen-specific Tregs are not required for immunoregulation in TCR transgenic models of multiple sclerosis, but are more effective than nonspecific cells (54, 55). Here, we found that CD25⁺CD69⁻ Tregs essentially all expressed the clonotypic TCR, but although CD3 levels were equivalent to clonotype levels, we could not rule out the presence of dual TCR-expressing cells. The numbers of BDC2.5 Tregs required for effective protection were 10–100-fold lower than those needed in polyclonal cotransfer models of inflammatory bowel disease or autoimmune gastritis (20, 42). These data imply that TCR specificity of Tregs for antigens in the target organ may be important for their effectiveness, probably because it allows triggering of the cells at the appropriate location.

The molecular mechanisms behind immunoregulation in the lesion are partially revealed by the gene expression profiles of the cells residing there. Although there were similarities to expression profiles reported for LN Tregs (45), some of the genes overexpressed in CD25⁺ Tregs are highly evocative of local suppressive action. The chemokine receptors CCR-2, CCR-5, and CXCR3 may control the entry of cells into the site, especially given reported expression of the cognate chemokines in the islets (56). CCR-2 has been found recently to correlate with regulatory activity in a model of autoimmune arthritis (57). Expression of mRNA for the cytokine IL-10 was highly up-regulated in the pancreas compared with the LNs, which was consistent with its antiinflammatory activity and its implication in Treg function in several studies (for review see references 14–16). In BDC2.5 mice, IL-10 was observed to be protective and important for dominant regulation of diabetes in older animals (47). The striking enhancement of IL-10 transcripts in the pancreas suggests that regulation specifically occurs within the target organ. Localization and increased production of immunoregulatory molecules in the target tissue is likely critical for preventing spontaneous autoimmunity from evolving into overt autoimmune disease.

This important balance of regulation is disrupted by the loss of ICOS signals in the pancreatic lesion. Tregs are diluted out of the lesion and turn down the complete Treg signature. Dependence on ICOS for expression of regulatory molecules such as IL-10 by other Treg types has been shown in a model of airway hypersensitivity (58). Our data demonstrate that ICOS is expressed on CD4⁺CD25⁺ Treg particularly in the target organ during autoimmunity. There, it is essential for maintaining the Treg gene program and, hence, preventing a switch to a proinflammatory environment during the early phase of insulitis. Although we cannot completely exclude that rather than dampening the Treg gene profile, ICOS blockade somehow leads to an increase in a CD25⁺CD69⁻ effector population that contaminates the detection of CD25⁺CD69⁻ Treg genes, cytometric analysis shows no evidence that the CD25⁺CD69⁻ populations are significantly altered by ICOS blockade, nor that ICOS blockade causes the loss of CD25⁺CD69⁻ ICOS⁺ cells. It had been postulated that a Th1/Th2 switch of effector cells was responsible for the exacerbation of autoimmunity after ICOS blockade in EAE models (48, 51). The present data demonstrate that a loss of immunoregulatory capacity by CD4⁺CD25⁺ Tregs is actually the mechanism by which ICOS blockade functions early to exacerbate autoimmunity.

Understanding the molecular and cellular basis of immunoregulation in the lesion could lead to the development of therapies that favor Tregs and promote long-term tolerance, even after an immune response has been well-established in the islet lesion. Costimulatory receptor blockade of ICOS upsets the balance of regulation directly in the lesion when it is first being established, and not in the draining node. It is interesting to consider what other signals might provoke diabetes after establishment of nondestructive insulitis in mice or humans. Perhaps viral infections, other environmental triggers, genetic factors, or age-dependent waning of Treg efficacy observed in NOD mice (31) disrupt the fragile balance of immunoregulation in the lesion and provoke diabetes.

The authors would like to thank J. Johnson and J. Yee of the Joslin Genomics Core and R. Park for microarray analysis; M. Roy for statistical analysis; E. Hyatt for animal assistance; G. Losyev, H. Shen, and M. Handley for flow cytometry; and K. Elder for administrative assistance.

This work was supported by the National Institutes of Health Program Project grant AI 39671 (to D. Mathis, C. Benoist, and G.J. Freeman). Additional support was provided by the Joslin Diabetes and Endocrinology Research Center cores (Z P30 DK36 836-17) and funds from the W.T. Young Chair for Diabetes Research (to D. Mathis and C. Benoist). A.E. Herman was supported by the Juvenile Diabetes Research Foundation International (grant 3-200-239), the National Institutes of Health support for the Joslin Diabetes Center (grant 2 T32 DK07260-25), and the Dana-Farber Cancer Institute (grant T32 CA 70083-07). G.J. Freeman was also supported by the National Institutes of Health (grant CA 84500).