Insulin Expression Levels in the Thymus Modulate Insulin-Specific Autoreactive T-Cell Tolerance: The Mechanism by Which the IDDM2 Locus May Predispose to Diabetes

Ins1-KO and Ins2-KO mice (25) were a gift of Dr. J. Jami (Institut Cochin, Paris, France). Chimeric animals were generated with D3 ES cells and crossed with the B6 strain. MHC haplotype, the main determinant of possible differential antigen-specific reactivity, was the same (H-2^b) in all animals used for the studies. Ins1-KO and Ins2-KO mice were crossed, and F2 progeny were used for experiments. Mice were kept in our animal facility under conditions specified by the Canadian Council of Animal Care.

DNA from the tail or ear of mice from F2 progeny was used to genotype for Ins1-wild type (WT), Ins1-KO, Ins2-WT, and Ins2-KO sequences by PCR using specific primers. Tables 1 and 2 show the band size and primers of each specific PCR. The PCR products were resolved by PAGE and quantified by ethidium bromide staining (GelDoc software; Bio-Rad, Hercules, CA). Band identity was confirmed with restriction enzyme digestion.

Pancreatic and thymic RNA from fasted 3-week-old mice were isolated by Trizol (Gibco, Rockville, MD) and treated with DNAse. The concentration of each sample was determined by optical density. An equal amount of each (1–2.5 μg) was reverse-transcribed to cDNA using random hexamer primers and Superscript reverse transcriptase (Gibco). A parallel sample, to which Superscript was not added, was assayed by PCR to confirm the absence of genomic DNA contamination or PCR carryover. An equal volume of each cDNA sample (2 μl from thymic cDNA and 2 μl from 1:300 diluted pancreatic cDNA) was then added to six serial dilutions of competitor (56, 28, 14, 7, 3.5, 1.7, or 0 amol/tube). The insulin competitor is an internally deleted cloned insulin sequence amplifiable by the same primers but 45 bp shorter, constructed as described by Forster (26). For loading control, we used the competitive PCR kit for cyclophilin (Quantum mRNA; Ambion, Austin, TX) as described by the manufacturer. The PCR products were resolved using PAGE, and bands were quantified to obtain a ratio of endogenous to competitor bands. The insulin ratio for each sample was normalized for the amount of starting RNA, using cyclophilin measurement as an indicator of total RNA content, and averaged. This mean represents the amount of insulin mRNA for a set of three total RNA samples. The calculations for comparing thymic and pancreatic levels across samples were as follows:

Thymus and pancreas extracts were obtained by acid/ethanol treatment as described by Vafiadis et al. (4). Total protein concentration of each sample was calculated using DC Protein-Assay (Bio-Rad), and this served for normalization of the insulin amount of each sample. Insulin concentration was measured by the Ultra-mouse insulin kit (Alpco, Windham, NH) as recommended by the manufacturer. The calculation of insulin amount was performed as follows:

Wool-nylon-purified spleen T-cells (2.5 × 10⁵) from unprimed adult mice were incubated with irradiated autologous spleen cells (2.5 × 10⁵) in the presence or absence of antigens: human proinsulin (Sigma Chemical, St. Louis, MO) or ovalbumin (Sigma) for 3 days or phytohemagglutinin (PHA) (Sigma) for 24 h. In another experiment, 3 × 10⁵ spleen cells were incubated with insulin B-chain (B:9–23) (Cambridge Research Biochemicals, Billingham, U.K.) or human insulin C-peptide (Polypeptide Laboratories, Torrance, CA) in the presence or absence of 25 units/ml murine interleukin (IL)-2 (Sigma) for 3 days in RPMI/0.5% heat-inactivated mouse serum and additives, then incubated for 16 h with 1 μCi [³H]thymidine (ICN Biomedicals, Aurora, OH). Supernatant (100 μl) was removed for cytokine analysis, the cells were incubated for an additional 16 h with 1 μCi [³H]thymidine and harvested (Cell Harvester; Inotech, Rockville, MD), and thymidine incorporation was measured by β-counter. IL-4 and interferon-γ production were measured using a two-site ELISA kit (Pharmingen, San Diego, CA).

In rodents, two unlinked, nonallelic genes, Ins1 and Ins2, encode insulin. Homozygous Ins1-and Ins2-KO mice (25) were crossed, and intercrosses between F1 progeny (100% heterozygote) resulted in F2 progeny with one (1-C), two (2-C), three (3-C), or four (4-C) copies of the four possible active insulin gene copies. The presence of each of the two WT and KO sequences allowed us to assign mice according to copy number for each of the two insulin genes.

In the pancreas of adult mice, Ins2 is expressed at two to three times higher levels than Ins1 (27,28). Mice homozygous for targeted disruption of either Ins1 or Ins2, expressing only one of the two Ins genes, are healthy and have normal levels of blood glucose, whereas mice homozygous for both Ins-KO genes die 48 h after birth (25). Insulin mRNA expression in the pancreas was measured by quantitative RT-PCR using an internally deleted competitor. The primers amplify the same size band from Ins1-WT and Ins2-WT cDNA. Figure 1 shows insulin mRNA expression in the pancreas of Ins1 (1-C) and Ins1/2 (4-C) mice. Because Ins1 has been shown to be at least twofold less abundantly expressed than Ins2, we chose to make the comparison between Ins1 (1-C) and Ins1/2 (4-C) mice. As shown in Fig. 1A, at the same concentration of the competitor (14 amol), in Ins1 (1-C) and Ins1/2 (4-C) mice, the intensity ratio between endogenous and competitor bands is ∼1, showing that the two mice express similar insulin levels in the pancreas. Cyclophilin, a housekeeping gene, was used to confirm equal RNA loading (Fig. 1B). Figure 1C shows the relative amounts of insulin mRNA in Ins1 (1-C) and Ins1/2 (4-C) mice. The data show, as expected, no difference in pancreatic insulin expression between mice with a single insulin gene copy and mice with four insulin gene copies. Moreover, no difference exists between these mice and mice with two or three insulin gene copies (data not shown).

At the protein level, insulin in the pancreatic extract was measured by the Ultra-mouse insulin ELISA kit (Alpco). The results again show no difference between mice of different genotypes. Thus, insulin production in the pancreas is not dependent on insulin gene copy number, a result we had predicted because the insulin gene is subjected to metabolic feedback control, which can upregulate even a single gene copy to maintain homeostasis.

In the thymus, the same techniques of quantitative RT-PCR were used as for analysis of pancreatic insulin levels. Figure 2A shows that mice with one copy of Ins1 [Ins1 (1-C)] express low levels of insulin, undetectable by our technique. In contrast, mice with four insulin gene copies express easily detectable insulin levels. Both PCR reactions were started with the same amount of RNA, as shown by cyclophilin level (Fig. 2B). Figure 2C shows the relative amount of thymic insulin mRNA in mice with different genotypes. The graph shows that Ins1 (1-C) and Ins2 (1-C) mice express twofold less insulin than Ins1 (2-C) and Ins2 (2-C) mice, respectively. Also, for any given number of copies, Ins2 is expressed at more than twofold higher levels than Ins1, thus approximating the naturally occurring allelic differences in humans (4,5). In contrast to the pancreas, in the thymus the quasilinear correlation between insulin copy numbers and mRNA levels (also confirmed at the peptide level as shown in Fig. 3) is consistent with our assumption that in this organ insulin expression is not subject to metabolic feedback regulation and therefore depends on gene copy number.

Taken together, these data of normal levels of insulin expression in the pancreas and graded deficiency in the thymus validate our mouse model. Thus it is possible to study the effect of thymic insulin levels on insulin-specific autoreactive T-cell selection at quantitative differences very similar to those determined in humans by the IDDM2 locus.

We chose to examine in vitro reactivity of cells from naive, non-diabetes-prone mice as the most direct test of a specific effect of thymic deficiency on central T-cell tolerance, uncontaminated by other events. T-cells vigorously reacting to proinsulin have been found in the pancreatic lymph nodes of such animals, though not in the periphery (8).

Insulin-specific T-cell reactivity was analyzed by a proliferation test. Figure 4 shows spleen T-cell proliferation in vitro after incubation without antigen and with proinsulin (Fig. 4A), ovalbumin (Fig. 4B), or PHA (Fig. 4C). The graphics show that, unlike mice with near-normal thymic insulin levels (Ins1-KO mice), the mice with low thymic insulin expression (Ins2-KO mice) present detectable T-cell reactivity to proinsulin. Both mice respond at similar levels to the control antigen ovalbumin and to nonspecific activation by PHA. Spleen cells were also incubated in the absence or presence of insulin B-chain peptide (B:9–23) or C-peptide. The results show that Ins2-KO mice present T-cell proliferation, whereas the response in Ins1-KO mice, if any, is marginal. The addition of IL-2 to the culture strongly amplifies T-cell proliferation in Ins2-KO mice, whereas it elicits a weak response in Ins1-KO mice (Fig. 5). In these experiments, we were unable to detect IL-4 or interferon-γ in the culture supernatants using a two-site ELISA kit (Pharmingen). This is not surprising, considering the low number of insulin-specific cells in comparison to the frequency of T-cells activated by mitogens.

These results show that thymic insulin expression modulates insulin-specific T-cell activation in peripheral lymphoid organs, even in naive, non-diabetes-prone mice. Together, the above experiments suggest less efficient negative selection of insulin-specific autoreactive T-cells in the immune repertoire of mice with low thymic insulin levels in comparison with mice with near-normal levels.

We have shown a detectable T-cell response to proinsulin in naive, non-diabetes-prone mice with deficient thymic proinsulin expression, a response absent in animals expressing normal or near-normal levels in the thymus. The result was robust, reproduced over several experiments with either proinsulin or specific epitopes (C-peptide and B:9–23 of the B chain). The genetic background of the animals was mixed B6 and 129 strains, but all animals had the H-2^b haplotype at the MHC locus, the main determinant of antigen-specific differential T-cell response. It is very unlikely that the differential reactivity seen between the two knockouts is due to random genetic variation at other loci, since it was specific for insulin and the two strains were no different in T-cell proliferative response to nonspecific stimuli.

Applied to the human situation, our findings directly imply that the low thymic insulin expression associated with the predisposing VNTR alleles (4,5) could be responsible for the generation of high numbers of insulin-specific autoreactive T-cells. The thymus is the organ where T-cell precursors undergo negative selection by apoptosis if they encounter self-antigen (10–13). Studies using T-cell receptor (TCR) transgenic mice have established that susceptibility to negative selection depends on TCR avidity and that thymic self-antigen plays a dose-dependent role (12,13,29,30). Many tissue-specific proteins, including hormones, have been shown to be expressed in the thymus (14,15,31). Insulin has been reported to be produced by rare thymic dendritic cells dispersed in the medulla (16), but in a more rigorous recent study, these cells showed all the markers and properties of medullary epithelial cells (15).

Evidence that thymic expression is important in self-tolerance (14,15) has until now come mostly from transgenic overexpression studies. Targeted disruption of endogenous proteins—myelin basic protein, for example (32,33)—results in strong T-cell reactivity in the homozygous state, as expected of a xenoantigen. However, those findings do not differentiate between central thymic and peripheral tolerance.

To answer this question in the case of insulin, we have established a mouse model with normal pancreatic insulin levels but graded thymic insulin deficiency and used it to show the modulatory effect of insulin expression levels in the thymus on insulin-specific autoreactive T-cell selection. What makes this model relevant to human diabetes is the possibility of exploiting the fact that Ins1 is two to three times less abundantly expressed in the thymus than Ins2. This difference is similar to that seen between the predisposing and protective IDDM2 alleles in the human (4,5), making our mouse model relevant for the study of immunological effects on T-cell reactivity to insulin in the periphery by the genetically determined thymic level discrepancy in the human.

The only other demonstration of a modulation of T-cell tolerance to nontransgenic peripheral self-antigen by physiologic thymic expression levels comes from a very recent study of thymus cross-transplantation between animals null for SAP, a liver protein, and syngeneic controls (15). In that study, thymic expression alone was sufficient to induce tolerance to SAP in the absence of peripheral expression but not necessary, as animals expressing SAP in liver but not thymus were equally tolerant. In our study, by contrast, we find a substantial effect of reduced thymic insulin in the presence of full peripheral expression, suggesting that the importance of central versus peripheral tolerance may depend on the specific autoantigen. Size of the organ where expression is confined may also be a factor: contact of any given T-cell with the tolerizing peripheral tissue is much more likely to occur in the liver than in the endocrine pancreas, an organ three orders of magnitude smaller.

We did not observe insulitis or diabetes in mice with low thymic insulin expression levels, despite the presence of autoreactive T-cells. This is consistent with the adoptive transfer literature showing that the mere presence of autoreactive T-cells in the periphery is not sufficient to cause either insulitis or autoimmune diabetes. Neither is insulitis required for the generation of insulin-autoreactive T-cells. These cells can be found before histologically detectable insulitis in the NOD mouse, and even in the pancreatic lymph nodes of normal, non-diabetes-prone mice (8). It has been shown that T-cell reactivity to self-antigen in vitro can occur even in the presence of self-tolerance in vivo (34,35). As a disease of multifactorial etiology, type 1 diabetes results from the association of several factors, of which self-reactivity to insulin is only one. In the NOD mouse, many non-antigen-specific factors are also required, such as aberrant cytokine expression (36), MHC II and costimulatory molecule expression in the peripheral organ, and pancreatic inflammation (37–39). The state of peripheral cell tolerance is not yet fully understood, but it has been shown that the expression levels of molecules such as B7 or MHC class II in the target organ (such as the pancreas) in the presence of sufficient numbers of autoreactive T-cells can result in tolerance breakdown and development of diabetes (40–42).

We conclude that thymic insulin expression levels play a critical role in insulin-specific autoreactive T-cell selection and that the IDDM2 locus predisposes to diabetes by the low thymic levels of insulin expression via failure of negative selection of insulin-specific autoreactive T-cells.

This work was supported by an operating grant from the Juvenile Diabetes Foundation. A.A.C. is supported by a Canadian Diabetes Association postdoctoral fellowship.

The authors thank Rosemarie Grabs and Marylène Rousseau for technical assistance. We thank Dr. Jacques Jami for his generous gift of the knockout mice.