Banting Lecture 2009: An Unfinished Journey: Molecular Pathogenesis to Prevention of Type 1A Diabetes

The first component of the trimolecular complex essential for the development of diabetes in NOD mice is a peptide of insulin, namely amino acids 9 to 23. A growing body of evidence indicates autoimmunity directed at insulin is central to the pathogenesis of type 1A diabetes of the NOD mouse (11,14,,,,,,–21). The major advance in defining a role for T cells targeting insulin was the cloning of T cells directly from islets of NOD mice by Daniel and colleagues (22). They discovered that the majority of CD4 T cell clones derived from such islets reacted with insulin, and of those that reacted with insulin, more than 90% were stimulated by the B:9-23 peptide (23). These T cell clones were able to accelerate the development of diabetes in NOD mice and a subset was able to cause diabetes in immunodeficient mice. T cell clones and TCR transgenics (21) targeting islet autoantigens can produce or prevent diabetes. Though in this review I will emphasize CD4 T cells, CD8 T cells, including T cells targeting the insulin B:15-23 sequence, are of parallel importance. Class II alleles (and thus potentially CD4 T cells), however, predominate in determining diabetes risk of humans and animal models (24,–26).

Mice have two insulin genes. Deleting the insulin 1 gene prevents the development of diabetes for 90% of NOD mice (27). Deleting the insulin 2 gene dramatically accelerates the development of diabetes (17,27). Both insulin genes are expressed in islets, while only insulin 2 is expressed in the thymus. It is likely that the toleragenic influence of small amounts of insulin produced within thymic epithelial cells (i.e., the induction of central tolerance) accounts for the different phenotypes of the insulin gene knockout NOD strains. In contrast, deleting other autoantigens, e.g., GAD65 (28), IGRP (15), IA-2 (29), and IA-2beta (30), does not alter the progression to diabetes of NOD mice.

To test whether targeting of insulin is essential for the development of diabetes we combined insulin 1 and insulin 2 knockouts. To prevent metabolic diabetes in the double knockouts, a preproinsulin transgene was introduced with a mutated sequence of insulin B:9-23 (31). The specific mutation was chosen (Tyrosine to Alanine at position B16 of insulin) because the anti-B:9-23 T cell clones created by Daniel and colleagues were not stimulated by this mutated peptide. NOD mice with only the mutated insulin do not develop diabetes and are protected, albeit not completely, from both expression of insulin autoantibodies and insulitis (31). Replacing the missing native insulin sequence with transplantation of islets with the native insulin sequence or immunization with native B:9-23 peptide, or a transgene coding for native insulin, restores development of insulin autoantibodies (32). These latter studies are complex in that though the transplanted islets have the native insulin sequence, the endogenous islets of the pancreas do not. Thus, to demonstrate induction of diabetes following transplantation of native B:9-23 islets into double-knockout mice, splenocytes from the mice are transferred into an immunodeficient mouse recipient that has the native B:9-23 sequence in its islets. Of note, islets with the native insulin sequence transplanted under the kidney capsule after 4 weeks of age induce insulin autoantibodies, and thus neither the specific anatomic location (i.e., pancreas) nor the neonatal presence of the inducing insulin B:9-23 epitope are critical to disease development (32).

Given the number of islet autoantigens recognized by T cells of the NOD mouse, an important question relates to the hierarchy of antigen recognition. Multiple investigators have demonstrated that techniques designed to induce “recessive” tolerance to insulin prevent diabetes (11,14,,–17). Recent studies by Krishnamurthy and coworkers demonstrate that the immune response to insulin is “upstream” of the response to the major islet autoantigen IGRP (15,33). Of note, though IGRP CD8-reactive T cells are very prominent in the NOD mouse model, mice with I-E promoter-driven proinsulin, introduced for the purpose of inducing tolerance to insulin, do not develop IGRP-reactive T cells. Even TCR transgenic mice targeting IGRP with blocked response to insulin do not spontaneously develop diabetes (33). Thus, the immune response to IGRP is both downstream of the immune response to proinsulin and not necessary for progression to diabetes. Whether responses to all islet autoantigens, except insulin, are dispensable cannot be answered at present in that several prominent autoreactive T cell clones target antigens that are either unknown or have not yet been knocked out in NOD mice (e.g., chromagranin, the target of BDC2.5 T cells) (34a).

Lack of a critical role of specific islet autoantigens, demonstrated by knockout experiments, does not rule out the importance of such autoantigens following intermolecular epitope spreading. Detection of autoantibodies or T cell responses to such autoantigens may also be important for diabetes prediction, and immunization with such autoantigens may still induce forms of dominant tolerance (34). Once NOD mice have developed diabetes, islet transplants with only the B16:A mutated insulin sequence are rapidly destroyed. This indicates that once autoimmunity is advanced, anti-islet autoimmunity directed at noninsulin molecules is sufficient for islet destruction, or that in the transplant setting recognition of the B16:A altered epitopes (or other epitopes of insulin) suffice to target β-cell destruction. Of note, in contrast to transplanted islets, islets within the pancreas with mutated B16:A sequence resist destruction following transfer of splenocytes from diabetic NOD mice. This suggests that though not completely protective in this setting, lack of the critical B:9-23 sequence remains important despite epitope spreading (32).

In contrast to antigen knockout experiments, where to date only knocking out the insulin genes influences the development of diabetes, multiple islet autoantigens and their peptides can be administered in a variety of ways to prevent the development of diabetes of NOD mice (34,–36). Such dominant suppression of disease is typically associated with the enhancement of regulatory T cells. There is evidence for regulatory CD4 (37,,–40), CD8 (24,41), and even B lymphocytes (42). The insulin B:9-23 peptide, when administered by intranasal or subcutaneous injection (especially when given in incomplete Freund's adjuvant) prevents diabetes (43,44). Of note, despite prevention of diabetes, subcutaneous injection of the B:9-23 peptide induces insulin autoantibodies presumably by activating CD4 T cells, and these antibodies bind to insulin but not the immunizing peptide (indicated by absorption studies) (45). Even normal Balb/c mice show such a response to the insulin peptide B:9-23 (46). With the appropriate class II molecules (i.e., I-A^g7 of the NOD or related I-A^d of Balb/c mice that have the same I-A alpha chain sequence), the peptide induces insulin autoantibodies, presumably due to activation of anti-B:9-23 CD4 T cells. Despite induction of insulin autoantibodies, insulitis does not develop unless the innate immune system is activated (e.g., via poly-IC injection) (47).

The TCRs of murine CD4 T cells recognize their target peptides presented in the groove of the antigen-presenting class II, I-E, and I-A molecules, and CD8 T cells target peptides in the groove of class I K, and D molecules (48,49). The homologous molecules of humans are DQ for I-A, DR for I-E (HLA-A, -B, and -C for mouse K and D). These presenting molecules are extremely polymorphic. The amino acid sequence lining the groove binds peptides and determines immune targeting. As in humans (50) and rat models of type 1 diabetes (51), class II molecules are critical for diabetes of the NOD mouse (12,52). The genetic effect is so large that it could be demonstrated with just eight diabetic mice, all being homozygous for the NODs unique major histocompatibility region (having I-A^g7 and lacking I-E) in a cross of NOD mice with a control strain (having I-A^k and I-E^k) (12,52). The class II molecules likely act by altering the TCR repertoire and enhancing the targeting of specific peptides by T cells. In particular, the I-A^g7 molecule, as well as human DR4-associated DQ8 and DR3-associated DQ2, have an unusual binding pocket (pocket 9) that lacks an aspartic acid (52,53). It has been hypothesized that such a lack of aspartic acid would favor presentation of autoantigenic peptides that have a charged residue that could bind in this unusual pocket 9. Recent studies in collaboration with the Kappler laboratory (B. Stadinski, unpublished data) suggest just the opposite, namely that the B:9-23 peptide recognized by pathogenic TCRs recognize the peptide in a low-affinity alternative register. Insulin is produced in the thymus of mouse and humans by specialized thymic epithelial cells and can result in deletion of insulin reactive T cells. It is likely that T cells with TCRs targeting B:9-23 in the thymus rarely encounter the peptide in this low-affinity register and thus escape central (i.e., thymic) deletion. These T cells that escape can then destroy β-cells in the pancreatic islets with their huge local concentration of insulin.

The third element of the trimolecular complex is the TCR made up of an α- and β-chain. TCRs are formed by random selection and recombination of gene segments (e.g., for the TCR α-chain, Vα and Jα TCR segments) that form billions of unique TCRs (48). The final TCR resembles a barrel with six fingers (the complementarity determining regions [CDRs]) extending from the barrel. Three CDR are provided by the TCR α-chain and three by its β-chain. It is these fingers that bind to the MHC-peptide complex and lead to the activation of T cells bearing the receptor. For classic recognition of a target peptide, all of the fingers (i.e., CDR1, CDR2, and CDR3 for both TCR α- and β-chains) are involved in recognition, and the CDR3 region is often predominant. The CDR1 and CDR2 elements are genome encoded sequences of both Vα and Vβ TCR segments. In contrast, the CDR3 α-chain element is formed at the interface of the randomly combining Vα and Jα gene segments, as well as V,D,J for the TCR β-chain. In addition, the joining is imprecise, with addition and subtraction of nucleotides leading to additional CDR3 sequence variation.

Given the specific and unique properties of I-A^g7 and the B:9-23 peptide related to development of autoimmune diabetes, it was perhaps to be expected that the pathogenic NOD anti-B:9-23 TCR would be unusual. Sequencing of the TCR from the clones discovered by Wegmann et al. revealed that the majority utilized a specific Vα segment, termed TRAV5D-4*04. (Note: Each unique Vα segment is given a number.) Unexpectedly, there was no conservation of what is termed the N-region of CDR3 (i.e., the region created by joining V and J plus nucleotide insertions and deletions) and no conservation of the TCR β-chain elements (54). We hypothesized that if we created mice with just this conserved Vα chain, allowing the mice to create multiple β-chains, we would induce islet autoimmunity (14). Using the technology of Vignali and colleagues (55), we have produced transgenic and retrogenic mice with multiple different α-chain sequences containing TRAV-5D-4*04 (i.e., Cα knockout background). The great majority of such mice develop insulin autoantibodies, and a subset develop diabetes (13,56). With the creation of TCR hybridomas from these mice, we estimate that ∼1 per 100 TCRs with a TRAV-5D-4*04 containing anti-B:9-23 α-chain respond to the B:9-23 peptide (56).

There is, as of yet, no crystal structure of the complete trimolecular complex discussed above (i.e., I-A^g7:peptide B:9-23:TCR with TRAV5D-4*04). We hypothesize that when such a structure is elucidated, it will reveal the anti-B:9-23 TCR bind at an unusual angle such that the primary contacts are the CDR1 and CDR2 portions of the TRAV5D-4*04 Vα-chain interacting with specific side chains of the B:9-23 peptide. The other CDR TCR elements will play a secondary role. Thus, we believe an accident of nature of NOD mice encoded in the genome sequences of the B:9-23 insulin peptide, the presenting I-A^g7 molecule, and the TCR Vα 5D-4 segment results in susceptibility to diabetes if tolerance mechanisms fail (14). The apparently low-affinity register in which the peptide needs to bind for recognition makes deletion of pathogenic anti-B:9-23 T cells in the thymus unlikely, and the extreme concentration of insulin in islet β-cells (i.e., approximately one-third of protein in β-cells is insulin) enhances peripheral immunological recognition. The simple rules for TCR recognition of insulin peptide B:9-23, dominated by a single Vα segment, presumably enhances the numbers of targeting T cells.

Two of the elements of the trimolecular complex targeting insulin are common to many mouse strains (TRAV5D-4 α-chain sequences) or all strains (insulin B:9-23 sequence). That said, I-A^g7 is a unique NOD contribution. Other genetic polymorphisms related to maintenance of tolerance (57) and environmental factors (or lack of protective environmental factors such as certain viral infections) influence activation of autoimmunity (58). Given what may be the very large number of T cells able to target the B:9-23 peptide, we believe the interaction between β-cells and the immune system of the NOD is similar to that of a swarm of bees. Once tolerance is broken, there are numerous individual T cells sharing common features of their TCRs, poised to target insulin.

The aforementioned hypothesis has not yet been subjected to a number of crucial tests. Thus, a crucial prediction is that knocking out the TRAV-5D-4*04 gene segment should prevent diabetes and insulin autoantibodies, similar to the effect of mutating the B:9-23 insulin peptide (11). In addition to genetic approaches to disprove the hypothesis that the shared Vα sequence is critical, development of therapeutics that are able to specifically block the above trimolecular complex are actively being explored.

We are taking two approaches to therapeutic targeting of the above NOD trimolecular complex. In collaboration with David Ostrov, who has defined a series of small molecules using a DOCKING program (59) that screens an NCI library of 140,000 small molecules designed to be “drug” candidates, we have identified small molecules that can modulate anti-B:9-23 TCR signaling (A. Michels, unpublished data). We believe families of small molecules binding to specific pockets of I-A^g7 will be able to both enhance and suppress TCR responses to the B:9-23 peptide, as well as alter specific cytokine responses of the dominant TRAV5D-4*04 B:9-23 T cells of NOD mice.

In addition, we have evidence that antibodies that recognize the B:9-23 peptide bound in the groove of I-A^g7 can block in vitro presentation of the B:9-23 peptide (L. Zhang, unpublished data). With this, the induction of such antibodies for their ability to prevent disease will soon be tested.

In 1986, we proposed a model of chronic autoimmune development of type 1A diabetes with disease pathogenesis divided into a series of stages (Fig. 2) (60,61). Technology to directly assess β-cell mass in humans is still lacking, and thus a useful debate continues as to whether there is a relapsing remitting course contributing to β-cell destruction and whether for some patients, the process of β-cell destruction is acute. In this model, the x-axis of Fig. 2 never specified exact times, with such a lack intentional, given inter-individual heterogeneity (even between monozygotic twins, both progressing to diabetes) (62). Different individuals progress at different rates to overt diabetes, and decades can elapse between the development of diabetes in one monozygotic twin and the development of islet autoantibodies in their twin mate (62). The model highlighted the potential to predict type 1A diabetes, a notion borne of multiple studies (63,64) as well as prevention trials (65,66).

For some of the stages there has been dramatic progress, while for others progress has been more limited; for all stages, however, a research infrastructure (67) that holds the promise of addressing fundamental questions over the next decade has been put in place.

Type 1A diabetes is usually polygenic in etiology, but there are two highly informative rare “monogenic” autoimmune syndromes associated with this disease: APS-1 (autoimmune polyendocrine syndrome type 1) (68) and IPEX syndrome (Immune Dysregulation, Polyendocrinopathy, Enteropathy, X-linked) (69).

IPEX syndrome results from mutations of the FoxP3 gene, a transcription factor that is essential for the development of regulatory T cells (70). In the absence of FoxP3, children develop overwhelming autoimmunity, and it is estimated that 80% develop type 1 diabetes. Diabetes can present as early as the first days of life. This syndrome clearly demonstrates the crucial role played by regulatory T cells and that most humans would develop type 1 diabetes unless pathogenic T cells are held in check. Bone marrow transplantation, by providing dominant regulatory T cells, is a consideration for children diagnosed with this fatal autoimmune syndrome (71). It is possible to generate antigen specific T cell lines expressing FoxP3 with regulatory properties, and this too is being explored as a potential therapy (38).

APS-1 is more common than IPEX syndrome but still rare, though increased in a few populations (e.g., Iranian, Jewish) (68). The syndrome is characterized by mucocutaneous candidiasis, Addison's disease, and hypoparathyroidism and results from mutations of the AIRE (autoimmune regulator) gene, another transcription factor. The disease is usually inherited in an autosomal recessive manner, but a single family with a dominant mutation has been described, and animal models for both have been created (72). The AIRE gene has a major role in enhancing the expression of peripheral antigens such as insulin within the thymic medullary epithelial cells (73,74). Tiny amounts of “peripheral” (i.e., not in the thymus) antigens such as insulin, antigens expressed within the thymus, are associated with negative selection of autoreactive T cells and protection from autoimmunity. Patients with APS-1 develop a series of autoimmune disorders over time and usually patients express more autoantibodies than specific diseases (68). Of note, 100% of these patients are reported to have autoantibodies reacting with interferon-α (75). Type 1 diabetes occurs in ∼18% of these patients, and development of diabetes is influenced by the dominantly protective HLA allele DQB1*0602 and insulin gene polymorphisms (76,77).

In combination, these two syndromes illustrate extremes of genetic determination of autoimmune diabetes when one or more pathways that are essential to maintenance of tolerance are disrupted. The genome of humans favors the development type 1 diabetes when mutations in genes controlling tolerance override the normal polygenic prevention of autoimmunity.

APS-2 syndrome is much more common than APS-1 (69). APS-2 syndrome is characterized by the occurrence of multiple autoimmune disorders in the same individual (e.g., type 1A diabetes, Addison's disease, thyroiditis, celiac disease, etc.) (78). It has a complex inheritance similar but not identical to type 1A diabetes (69,79). In particular, the highest risk HLA genotype (DR3/4-DQ2/8) for Addison's disease has DRB1*0404 and not DRB1*0401 with DQB1*0302 (80,81). Patients with type 1A diabetes are at increased risk for the development of the series of autoimmune disorders of the APS-2 syndrome and in particular Addison's disease, celiac disease, thyroid autoimmunity, and pernicious anemia (82). We routinely screen patients with type 1 diabetes for associated autoimmunity targeting the adrenal, intestine, and thyroid with measurement of 21 hydroxylase autoantibodies for Addison's disease (1.5% positive) and transglutaminase autoantibodies for celiac disease (10% positive) and with TSH determination. Within the 1st year after the diagnosis of type 1 diabetes, approximately one-third of patients already express autoantibodies reacting with one or more of these target organs (83).

Type 1A diabetes has become one of the best studied complex genetic disorders (84,85). Approximately 1 in 300 individuals in the U.S. develop type 1A diabetes versus approximately 1 in 20 first-degree relatives. Here too, we note that the risk of siblings and offspring of a father with type 1 diabetes is greater than the risk of offspring of a mother. With very long–term follow up, the majority of monozygotic twins of a patient with type 1 diabetes develop islet autoimmunity (i.e., >70%) and diabetes (i.e., >60%) (62). As many as 30 years can elapse between development of diabetes of the first twin and the second twin. In contrast, the risk of diabetes in dizygotic twins might not differ from that of siblings, with ∼5% developing diabetes (86).

Despite the strong genetic predisposition, as evidenced above, the great majority of individuals developing type 1A diabetes, >90%, do not have a first-degree relative with diabetes. Though 40% of individuals in the general Denver population have the high-risk HLA alleles DR3 or DR4, only 2.4% have the highest risk genotype, namely DR3 and DR4, and such heterozygous individuals make up ∼30% of children developing diabetes (87). Thus, HLA genotype, inherited from both parents, is most important for the development of type 1 diabetes, with susceptible alleles very common in the general population (88).

Similar to the NOD mouse model, the major genetic determinants of type 1A diabetes are polymorphisms of class II MHC genes DQ DR, and DP, in this order (50). The highest-risk genotype consists of a DR3 haplotype DRB1*0301-DQA1*0501-DQB1*0201 on one chromosome and a DR4 haplotype on the other (DRB1*0401-DQA1*0301-DQB1*0302, i.e., in shorthand DR3/4-DQ2/8) (88). Between 30 and 50% of children developing type 1A diabetes have this genotype (89). (Note: The younger the onset of disease, the greater the DR3/4-DQ2/8 genotype.) The absolute risk in the general population with this genotype is ∼5%. The DRB1*0403 allele decreases risk of DR4 haplotypes, and certain DP alleles also alter risk (e.g., decreased with DPB1*0402) (90). In addition, there are several extremely potent protective HLA alleles. In particular, DQB1*0602 occurs in ∼20% of the general population but in only 1% of patients developing type 1A diabetes.

The DAISY study, headed by Marian Rewers, has HLA typed more than 30,000 newborns in Denver, Colorado, and followed prospectively from birth more than 1,000 general population newborns with HLA risk alleles, as well as 1,000 first-degree relatives independent of HLA (91). Siblings of a patient with type 1 diabetes have a diabetes risk that is several-fold that of offspring, despite both siblings and offspring sharing approximately one-half of their genome with their proband. There could be multiple genetic and environmental reasons for the higher risk of siblings, but the simplest explanation would be the presence of additional polymorphisms of genes within the MHC that increase risk for certain haplotypes beyond the sequence of HLA DR and DQ alleles. That this may be the case is suggested by the extreme risk of DR3/4-DQ2/8 siblings who shared both HLA haplotypes identical by descent with their sibling proband compared with DR3/4-DQ2/8 siblings sharing only one or no haplotype identical by descent (Fig. 3) (92). The risk of activating islet autoimmunity in the DAISY study for such siblings was as high as 70%, and such siblings make up a major portion of all of the DAISY children progressing to islet autoimmunity and then diabetes (92). With typing for DR DQ, and DP in the DAISY study, risks as high as 20% can be defined in the general population (90). Nevertheless, a risk of 20% of DR3/DR4 heterozygotes of the general population is much less than the 70% of DR3/4 siblings with MHC inherited identical by descent, suggesting that there remain additional polymorphisms of genes in the MHC to be discovered. We have evidence that one of the loci is at the far telomeric end of the MHC (93), and there is evidence that specific class I HLA alleles (or loci in linkage disequilibrium with these alleles) contribute to risk, in particular HLA-B39, present in ∼2–4% of patients, and HLA-A24 has been associated with earlier onset of type 1A diabetes (94,95). Of note, the presence of HLA B39 increases the risk of DR8 haplotypes to that conferred by DR4 haplotypes (J. Baschal, unpublished data).

Given the existence of almost total conservation of multiple haplotypes (96) for millions of base pairs across the MHC (the two best known conserved haplotypes: HLA-A1,B8,DR3 haplotype [97,98] and HLA-A30,B18,DR3 “Basque” haplotype [99]) makes the search for such non-HLA disease determinants in the MHC difficult (96). The Basque haplotype is higher risk than the A1,B8,DR3 haplotype even though both haplotypes have the same sequence for their DR and DQ alleles. Both haplotypes, given the presence of DR3, increase diabetes risk (100).

Given the dramatic protection provided by certain specific HLA alleles, the question has been raised as to whether provision of such alleles might be considered therapeutically or assessed in family planning. The alleles protecting and determining risk are presumably “normal” HLA alleles determining the targeting of specific self-molecules; for instance, the DR2-DQB1*0602 haplotype that provides dominant protection from type 1A diabetes is the highest risk haplotype for multiple sclerosis (101).

There has been an explosion in identifying non-MHC loci contributing to genetic risk, with >40 loci confirmed (Fig. 4) (84,102). All of these genetic loci together have a much smaller influence on diabetes risk than the MHC, with many having odds ratios (ORs) of <1.2. (Note: An OR of 1.0 indicates no risk, and the MHC OR is >6). After the MHC, the next most important locus is that of the insulin gene. Protective polymorphisms in a sequence 5′ of the insulin gene increases the amount of insulin expressed in the thymus and is associated with decreased diabetes risk, probably by increasing thymic deletion of T cells targeting insulin (103,104).

The PTPN22 gene is the third most important locus (105,–107). A change of a single amino acid (R640W) increases the risk of diabetes, with an OR of ∼2. The molecule (lyp) encoded by the PTPN22 gene is a lymphocyte-specific tyrosine phosphatase, and the disease-associated polymorphism increases inhibition of TCR signaling, likely through a gain of function (107). How such a gain of function contributes to diabetes risk is unknown, but the predominant hypothesis is that it may decrease negative selection of T cells in the thymus. Multiple other specific genes have been implicated including CTLA4 and CCR5 and the IL2 receptor, as well as multiple loci with one or more potentially relevant genes. A helicase (IFIh1) involved in interferon signaling has several rare variants associated with disease risk as well as common single nucleotide polymorphisms (108). Almost all of the identified loci appear to alter risk of diabetes by effects on the immune system (84). Interestingly, the genetic polymorphisms associated with type 2 diabetes do not generally influence the development of type 1A diabetes (109,110), though there are conflicting reports of an association of polymorphisms of TCF7L2 gene with latent autoimmune diabetes in adults (111,112). On balance, it would appear that the genetics of type 1A diabetes and type 2 diabetes are distinct, with type 1A diabetes an immune-mediated disorder.

The search for rare variants of genes contributing to type 1 diabetes risk has hardly begun, but it is likely that multiple such variants exist. In addition, it is likely that either single or more likely multiple loci within the MHC remain to be discovered, some of which may modulate risk more dramatically than currently discovered common variants outside of the MHC (92,94). Multiple common pathways underlie genetic susceptibility to type 1 diabetes including T cell targeting of peptide antigens (HLA class I and class II) in general, as well as targeting of proinsulin/insulin and, in particular, TCR signaling (e.g., PTPN22, CTLA4), tolerance maintenance (e.g., foxP3 and AIRE), and innate immune responses (e.g., IFIh1). It is now possible to identify individuals with high or extreme risk of type 1A diabetes related to defined HLA genotypes/haplotypes, with smaller contributions from the insulin locus and PTPN22 (113). Common SNPs of loci with ORs <1.2 do not facilitate genetic prediction but will hopefully identify important pathogenic mechanisms (114). Though very high–risk individuals can be identified with HLA analysis (e.g., DR3/4 heterozygotes), they comprise <50% of those developing type 1A diabetes, and there has been a trend over the last 50 years for the percentage of patients with DR3/4 to decrease (115,116). High-risk general population individuals may be appropriate for preventive trials similar to relatives of patients with type 1 diabetes, and their identification has the potential to prevent morbidity and mortality at the time of diabetes onset if genetic screening is combined with islet autoantibody determination (117).

The least amount of knowledge gains over the past two decades has, arguably, been made in defining environmental factors contributing to the development of type 1 diabetes. It is nevertheless clear that environmental determinants are important given the remarkable doubling of type 1 diabetes incidence in most Western countries over the past 20 years (118,–120). This is too rapid a change for a common genetic disorder to be ascribed to genetic alteration (alone) of the population. A leading hypothesis to explain such an increase in incidence is the “hygiene” hypothesis, positing decreased infections increasing multiple immune-mediated disorders including asthma and diabetes (121,122). Both the NOD mouse and BB rat, when raised in a germ-free environment, have been reported to alter diabetes, with recent studies suggesting that intestinal microbiota in animal models modulates development of diabetes (123). Another hypothesis posits that insulin resistance somehow “accelerates” the pathogenesis of type 1 and type 2 diabetes and that type 1 and type 2 diabetes have similar genetic causes, though differing in terms of MHC alleles (124,125). Current genetic analysis as reviewed does not support this hypothesis (126), though insulin resistance associated with obesity influences progression to type 1 diabetes, presumably as insulin secretion fails (127).

It is noteworthy that certain medications clearly induce development of type 1A diabetes. In particular, interferon therapy is associated with diabetes in animal models (128), and in humans this treatment can induce islet autoantibodies and accelerate diabetes progression leading to ketoacidosis (129,130). Methimizole (131,132), penicillamine (133), and lipoic acid (134) (i.e., sulfhydryl-containing drugs) can induce autoantibodies, with titers of the antibodies, at times, high enough to influence metabolism (e.g., insulin autoimmune syndrome: Hirata disease associated with DRB1*0406) (131).

Other contenders for environmental factors are dietary, with reports that cows milk (135), early introduction of cereals (136,137), decreased levels of n-3 fatty acids (138), and vitamin D contribute to diabetes risk (139). Large current trials and prospective observational studies will hopefully rigorously test these hypotheses (67). Of note, both metabolomics (140) and analysis of mRNA arrays (141) are being applied to identify abnormalities potentially preceding the occurrence of anti-islet autoantibodies. Both a power and potential weakness of these studies is that the analysis of multiple parameters, and thus testing of thousands of hypotheses at the same time, will yield many false-positive results. Thus, replication studies will be crucial.

Many studies of viral infections inducing diabetes preceded the discovery that type 1A diabetes is a chronic autoimmune disease. Thus investigators searched for viral infection at the time of diabetes onset, including studies of pancreas from patients that died at diabetes onset. Congenital rubella infection and enteroviral infections have been studied in most detail, with at present a lack of clear consensus as to their importance (142,143). A major difficulty may relate to the induction of islet autoimmunity years prior to the development of diabetes and the likelihood that viruses triggering islet autoimmunity may be many, ubiquitous, and act over a very short time period. All of the preceding would be lessons learned from virus-induced autoimmune diabetes in rat models.

Perhaps the best animal model of the triggering of type 1A diabetes comes from the discovery that infection with the Kilham rat virus (KRV) induces diabetes in diabetes-resistant BB rats (144). BB-DR rats were bred to be a normal rat control strain and lack the lymphopenia of the BB original strain that spontaneously developed diabetes (51). The KRV infection was discovered to induce diabetes when a subset of BB-DR rats spontaneously developed diabetes and it was subsequently discovered that the strain had become infected with the virus. It is now known that multiple rat strains, all with the same class II MHC alleles (i.e., RT1-U) and specific TCR loci, develop diabetes (at varying frequency) when infected with several different viruses or when stimulated with immunologic inducers including toll-like receptor (TLR) agonists (145,146). In a similar manner, mice engineered to express the B7 molecule on the surface of islet β-cells develop diabetes when given poly-IC, a viral RNA mimic and activator of TLR3 (147). In this model, poly-IC induces interferon-α. Antibodies to interferon-α block diabetogenesis, and interferon itself can induce diabetes. The KRV does not infect islet β-cells and only needs to be transiently present to induce diabetes. If a virus induces diabetes in humans, we believe its discovery will depend upon monitoring individuals for acute infections and initial appearance of islet autoantibodies on a relatively short time scale, and the inducing virus(es) will be ubiquitous.

The immunocytochemical (ICA) assay measures a combination of autoantibodies reacting with multiple defined autoantigens (except insulin), and its discovery was central to the search for defined molecular targets (148). The ICA assay is difficult to standardize and interpret given this heterogeneity and the need to utilize frozen human pancreatic sections as substrate.

The clearest indication of the presence of islet autoimmunity is the expression of islet autoantibodies (149). At present, four major islet autoantigens (i.e., insulin, GAD65, IA-2, and Znt8) have been identified, and assays for these autoantibodies have been validated in international workshops. ZnT8, the most recently defined autoantigen, is the islet-specific zinc transporter involved in transport of zinc into insulin secretary granules (150). It is of interest that a common polymorphism of ZnT8 that changes a single amino acid is a major target of anti-ZnT8 autoantibodies. Patients homozygous for specific ZnT8 variants more often make autoantibodies to their own genome encoded variant, demonstrating the autoimmune nature of the targeting (151).

The simplest rule for predicting type 1A diabetes is that expression of two or more of the four “biochemical” autoantibodies predict diabetes (63,152). It is currently unknown why such combinatorial prediction is so effective, but the null hypothesis is that it simply relates to statistical probabilities. With a combination of the binomial theorem and Bayes theorem, one can calculate positive and negative predictive values of combinatorial autoantibody testing. Specifically, one can derive the probability for instance of detecting one or more, or two or more, islet autoantibodies with the four assays currently utilized, with assays having a specificity set at the 99th percentile. Approximately 4% of normal individuals will be positive for one or more autoantibodies, while only 0.06% will be positive for two or more autoantibodies (i.e., 6 in 10,000 normal individuals). For assays with specificity at the 95th percentile, the corresponding false-positive frequencies are 19 and 1.4%. Both for research and clinical care, assays with false-positive rates of 5%, when multiple assays are combined, are very problematic and potentially underlie some confusion concerning expression of islet autoantibodies in populations at low risk for type 1A diabetes, such as patients clinically identified as having type 2 diabetes. In such populations, expression of two or more autoantibodies, set at the 95 percentile, remains reasonably specific (i.e., 1.4% “false positive”) but not the presence of one or more autoantibodies.

A “biologic” false-positive islet autoantibody result is not synonymous with lack of autoantibodies binding to the assayed autoantigen or confirmation, even over time, of the presence of the autoantibody. Thus, true autoantibodies may be present, but the epitope that they recognize, the level of the autoantibody, or the lack of expression of multiple autoantibodies is such that its presence may not indicate an increased diabetes risk. This is well illustrated by the isolated presence of low-affinity insulin autoantibodies, where most children with such autoantibodies do not progress to diabetes (64,153). Recently Hampe and colleagues have reported the presence of anti-idiotypic antibodies in normal individuals whose removal results in normal sera becoming positive for GAD65 autoantibodies (154). The experimental system is complex, with the potential for monoclonal human anti-GAD65 autoantibodies to be eluted off of beads. Further studies are needed to test this novel finding (154).

Similarly, lack of all islet autoantibodies cannot be equated with absence of type 1A diabetes (155). As children are followed to the development of diabetes, a wide fluctuation in autoantibodies over years is often observed, with one autoantibody falling and another autoantibody increasing. A small subset of autoantibody-positive children loses expression of all autoantibodies prior to diabetes onset. If a child presents with diabetes and all four standard biochemical autoantibodies are absent, consideration of monogenic forms of diabetes is in order, as rare forms of diabetes can greatly influence choice of therapy, e.g., sulfonylureas, or prognosis, e.g., DIDMOAD (diabetes insipidus, diabetes mellitus, optic atrophy, and deafness)/Wolfram syndrome (156). It is estimated that 10% of children with diabetes lacking all islet autoantibodies have definable monogenic forms of diabetes (A. Hattersley, personal communication).

Insulin autoantibodies are unique in that their levels and frequency of positivity are inversely related to the age at which diabetes develops (157). Within several weeks of the introduction of insulin therapy, almost all individuals express insulin antibodies that cannot at present be distinguished from insulin autoantibodies. Insulin autoantibodies are often, but not always, the first autoantibody to appear in children followed from birth, followed by the appearance of GAD65, and then IA-2 and ZnT8 (153,158,159). Standard ELISAs should, in general, be avoided for the measurement of these autoantibodies (160), though a novel GAD assay using plate-bound anti-GAD autoantibody rather than plate-bound antigen (as for standard ELISAs) performs well in international workshops (161).

There will likely be need for point-of-care screening assays for islet autoantibodies, and this need will be acute if preventive therapies are introduced into clinical practice. Most of the cost of screening for islet autoantibodies now relates to shipping samples to laboratories and the paperwork involved. If point-of-care screening assays were available, it would allow direct testing in a physicians office and referral for confirmation and staging of the relatively uncommon individual expressing one or more autoantibodies. At present, the insulin autoantibody is the most difficult for laboratories to master, and most fluid-phase radioassays for GAD65, IA-2, and ZnT8 perform well in international workshops (162).

There are multiple ways to assess metabolic progression of patients developing diabetes. One of the most specific is the intravenous glucose tolerance test. Most patients within 1 year of diabetes have a 1 + 3 min insulin secretion following a bolus of glucose less than the first percentile of normal control subjects (163). This is also one of the least convenient methodologies utilizing an intravenous injection. Oral glucose tolerance testing with measurement of glucose is also highly predictive of deterioration (164). In studies such as DAISY, we now utilize fingerstick measurement of A1C, with the great majority (but not all) of children developing diabetes demonstrating a gradual rise of A1C in the normal range within the 1 to 2 years prior to overt diabetes (165). With rising A1C, an oral glucose tolerance is utilized to confirm diabetes.

The development of overt type 1A diabetes is often acute (166). This is perhaps not unexpected for an autoimmune organ-specific disease, and we see similar sudden dramatic increases in ACTH in patients expressing 21 hydroxylase autoantibodies, who develop overt Addison's disease. It is likely that the majority of patients presenting with type 1A diabetes have had the disease for months to a year prior to diagnosis; they present with high A1C and at times glucose >1,000 mg/dl (117). The children who die at onset frequently have a history that the first health care providers to evaluate the child missed the diagnosis of diabetes, and a several day delay in treatment can be fatal. Such acute presentations are prevented in children pre-identified for diabetes risk in studies such as DAISY (117). Once diabetes has developed, loss of C-peptide secretion is the primary parameter to follow further disease progression (167).

A controversy has arisen relative to the quantitative preservation of C-peptide and β-cells in patients with long-term type 1 diabetes, with important evidence of preservation in some insulin-expressing cells within the pancreas of some long-term patients with childhood-onset diabetes, as well as apoptosis of β-cells (168,169). Fortunately a large set of pancreases of cadaveric donors is rapidly becoming available for histologic analysis through the nPOD program headed by Mark Atkinson. An advantage of these pancreases is the availability of the whole pancreas for analysis of multiple regions of the organ. Pancreases from cadaveric donors with type 1 diabetes, as well as nondiabetic cadaveric donors, are being analyzed, and the slides are posted on a Web site for international scientific study (www.jdrfnpod.org). The slides can be viewed via desktop computer as if through a microscope. It is already apparent that a subset of childhood-onset patients with a clinical diagnosis of type 1 diabetes has large numbers of β-cells, with β-cells in 100% of their islets, but it is likely that these patients do not have type 1A diabetes (170), and their disease may be related to ketosis-prone forms of diabetes occurring in minority populations of the U.S (171). Most of the pancreases from patients with long-term childhood-onset diabetes in nPOD have no β-cells within islets. Approximately 10% have limited areas with insulin containing β-cells. This heterogeneity with lobular areas with β-cells remaining (reported decades ago by Foulis and coworkers studying recent-onset patients [172,173]) likely explains the slow progression to type 1A diabetes. It is reminiscent of the destruction of melanocytes in patches of skin of patients with vitiligo (Fig. 5).

Though multiple interventions have failed to prevent either progression to type 1 diabetes or loss of β-cell function postdiagnosis, we are entering an era where several immunotherapies can almost certainly ameliorate β-cell loss (174,175), with several phase III trials underway or planned. Key information nevertheless remains lacking in terms of the long-term efficacy and safety and each therapeutic pathway, thus necessitating the need for further study. One can broadly divide therapies into those that are generally immunosuppressive and immunomodulatory versus antigen specific. The immunomodulatory/immunosuppressive trials have a higher probability of efficacy, based on current data, but likely involve greater risk. I will highlight just four therapies, each with some evidence of efficacy, given both the void of “positive” trials and paucity of large well-controlled studies.

The National Institutes of Health (NIH) Diabetes Prevention Trial evaluated both parenteral insulin and oral insulin (to induce “oral tolerance”) in cytoplasmic islet autoantibody–positive relatives of patients with type 1 diabetes. Relatives were staged in terms of their expression of insulin autoantibodies, metabolic abnormalities, and presence of the protective HLA allele DQB1*0602. In addition, GAD65 and ICA512 autoantibodies were measured though not included in the prediction of disease risk. These large studies, as well as the ENDIT prevention study utilizing nicotinamide, demonstrated the ability to predict type 1 diabetes on a relatively large scale and to assign various levels of risk for disease (65,66,159,176,177). Overall, none of these studies influenced progression to diabetes, though in one subgroup of the oral insulin trial, those with high levels of insulin autoantibodies, oral insulin significantly slowed progression to diabetes (Fig. 6). Given the subgroup analysis, the NIH TrialNet group is now seeking to confirm whether oral insulin can ameliorate progression to diabetes of islet autoantibody–positive relatives expressing multiple islet autoantibodies as well as those having insulin autoantibodies.

The primary outcome variable in trials at the onset of type 1 diabetes is preservation of C-peptide. GAD65 is a prominent target of the autoimmunity of humans, with GAD65 autoantibodies one of the best predictors of progression to diabetes. GAD65 in the adjuvant alum in a randomized placebo controlled small trial delayed loss of C-peptide, though without improving A1C or decreasing insulin requirements in the treated patients (178). Confirmatory studies of GAD65 vaccination are underway, including a new-onset TrialNet study, with plans for a prevention study.

Two studies, using single courses of two different anti-CD3 antibodies, significantly delayed the loss of C-peptide secretion (173,174). The anti-CD3 antibodies were modified to decrease cytokine release and limit acute toxicity. The delay in loss of C-peptide lasted for ∼6–12 months, followed by what appears to be resumption of C-peptide decline parallel to (though delayed) control subjects. Decreased insulin utilization was associated with decreased loss of C-peptide. There is evidence that in addition to acute and transient depletion of T cells, the therapy induces regulatory T cells and, in particular, evidence for regulatory CD8 T cells (41,179). Longer follow up and evaluation of a larger number of individuals in phase III trials will be essential to judge efficacy and potential toxicity.

A recent TrialNet study indicates that the anti–B-cell antibody, anti-CD20, as a single course, significantly delays loss of C-peptide, improves A1C, and decreases insulin requirement (180). Anti-CD20 essentially eliminates B-lymphocytes from the circulation for months to a year but does not target plasma cells. Thus, many antibodies are unchanged with this therapy, while selected autoantibodies are markedly inhibited (181). Anti-CD20 has been utilized as therapy for B-cell lymphoma and shows efficacy in multiple sclerosis and rheumatoid arthritis. Thus, there is considerably more experience with this therapy than with anti-CD3. Nevertheless, further study is essential (including more than a single drug course) to evaluate both efficacy and safety as a single agent in patients developing diabetes and those with new-onset diabetes. Newer anti-CD20 antibodies and a series of therapeutics targeting B-cells are in development. From animal studies there is evidence of regulatory B-cell induction, and B-cells themselves are likely important for the presentation of islet autoantigens (42).

To date none of the aforementioned therapies have demonstrated long-term permanent arrest of disease progression. It is remarkable that despite being such a slow destructive process, the autoimmunity underlying type 1A diabetes is so resistant to immunotherapy. It is likely that the therapies showing promise to date in the above trials do not eliminate the underlying T cell memory driving β-cell destruction, and that multiple courses of therapy or combinatorial therapies will need to be developed to achieve long-term immunologic remission (182). That said, long-term continuous immunosuppression is almost certainly not an option as a therapy for type 1 diabetes.

Approximately 1 in 300 randomly selected cadaveric donors from the general population express multiple islet autoantibodies (183,184), and thus it is likely that ∼1 million individuals in the U.S. are in the process of developing type 1 diabetes. As immunomodulatory/immunosuppressive therapies improve or, hopefully, antigen specific therapies are developed, we believe it is likely that chronic active insulitis will become a treatable entity similar to chronic active hepatitis. North American patients with new-onset diabetes and relatives of patients with type 1 diabetes can be evaluated for islet autoantibodies and participation in trials by calling 1-800-HALT-DM1 or accessing the TrialNet Web site (www.diabetestrialnet.org).

This work is supported by grants from the National Institutes of Health (DK32083, DK55969, DK62718, AI50864, DK32493, DK064605), the Immune Tolerance Network, the Diabetes Endocrine Research Center (P30 DK57516), the American Diabetes Association, the Juvenile Diabetes Foundation, the Helmsley Foundation, the Children's Diabetes Foundation, and the Brehm Coalition.

G.S.E. has been a paid consultant for and has received stock options from Bayhill Therapeutics. No other potential conflicts of interest relevant to this article were reported.

I thank my many fellows, students, collaborators, and a world of investigators who generated the bulk of what I reviewed and the families who have been partners for decades in deciphering the natural history of type 1 diabetes. I thank Dr. Mark Atkinson for his excellent editing.