Autoimmunity is generally associated with polygenetic susceptibility, while the initial precipitating event is likely triggered by an environmental stressor (1–4). The major alleles associated with most autoimmune disorders are the human leukocyte antigen (HLA), and several alleles are shared among autoimmune conditions (5–8). This suggests that a common T cell-dependent mechanism is the underlying cause of tissue-specific autoimmunity irrespective of the organ or tissue being targeted. Although several hypotheses have been put forth to explain the HLA-mediated susceptibility, the exact mechanisms are still largely unknown. HLA structure selects for a particular peptide sequence motif and can affect the stability of the peptide:HLA complex (9). It is likely that autoimmune epitopes are not efficiently presented within the susceptible HLA molecules during thymic selection, or alternatively are presented with increased stability or at a higher concentration in the target tissue (10). Clearly, HLA allele structure is not the only parameter that might affect the stability of the tri-molecular complex [T cell receptor (TCR)/HLA/peptide], and not all individuals with T1D possess susceptible HLA alleles. Lower level of tissue antigen expression in the thymus, the relative abundance of self-antigen at the tissue site, an increase in immunogenicity of self-peptides either via post-translational modifications (PTMs) or molecular mimicry could all influence the stimulatory capacity of peptide:HLA complexes in periphery (Figure 1). How these changes in epitope immunogenicity could affect disease development will be discussed in this review.

The spontaneously diabetic non-obese diabetic (NOD) mouse model has been a useful system for identification of the key mechanisms important in the development of autoimmunity due to its significant similarity to human T1D (11, 12). Nearly 6 years after HLA was first associated with T1D in humans (13, 14), the spontaneously generated NOD diabetic strain was obtained by the Jackson Laboratory from CLEA Japan, where it quickly became an invaluable tool in the etiology of T1D (11, 15). The importance of the major histocompatibility (MHC) locus was originally traced by congenic approach, where MHC locus was introgressed onto the NOD background (16, 17). Further analysis of mice that received a non-NOD MHC class II transgene confirmed the important contribution of I-A^g7 to diabetes susceptibility (18). Although MHC II confers most of the susceptibility, there are over 50 genetic loci that make up the NOD diabetic phenotype (19). The polygenetic susceptibility of the NOD mouse strain mirrors human disease, and further underlies the complexity of T1D (20). Importantly, the I-A^g7 MHC II variant has structural similarities with human susceptible DQ8 (DQA1*0301/DQB1*0302) (9, 21, 22). Moreover, many of the antigens targeted in autoimmune diabetes are shared between the two species (19). The similarities of the shallow and positively charged peptide-binding groove characteristic of both human DQ8 and mouse I-A^g7, and significant concordance in antigenic targets have made it possible to uncover sequence characteristics of autoimmune epitopes that are relevant to human disease (23, 24). Nevertheless, the precipitating events that lead to T cell priming and beta cell destruction remain unclear (4, 25). While the NOD mouse model has been a prolific tool for mechanistic insight into the many facets of T1D pathogenesis, recent expansion of HLA-humanized mouse models now allow direct interrogation of human autoimmune tri-molecular complex (TCR/HLA/peptide) and its role in loss of self-tolerance.

A large body of evidence accumulated over several decades has implicated beta cell-specific immune response and, in particular, beta cell-specific T cells as the main drivers of autoimmune tissue damage and development of T1D (12, 26, 27). Progression to disease in humans is associated with islet antigen-specific antibody responses, and T cells specific to islet antigens are found at higher frequencies in T1D patients (28–31). Importantly, both CD4 and CD8 T cells were observed directly in the pancreatic lesions, and islet antigen-specific T cells have been cloned from pancreatic islets of T1D organ donors (32–38). HLA, being the major risk allele, implies that inherent structural differences in HLA and, consequently, TCRs selected on those HLA alleles lead to erroneous T cell reactivity to self (5, 39, 40). While class II HLA alleles confer the majority of the genetic susceptibility, certain class I alleles have been shown to impose a separate risk (41). Multiple antigens are targeted by both CD4 and CD8 T cells in T1D. Beta cell-specific antigens presented by Class II molecules include preproinsulin (PPI), insulinoma-associated antigen (I-A2), glutamic acid decarboxylase (GAD) 65, heat shock protein (HSP)-60, HSP-70, islet-specific glucose-6-phosphatase catalytic subunit-related protein (IGRP), and zinc transporter (ZnT8) (42–44). While MHC class I responses display similar wide range of antigenic targets, including PPI signal peptide, IA2, ZNT8, human islet amyloid polypeptide (IAPP), IGRP, and GAD65 (45). The progression to T1D in humans is associated with accumulation of islet antigen antibody reactivity to IAA, GAD65, IA-2, and ZnT8, which mirrors the intra- and inter-molecular “antigenic spread” of T cell responses (46, 47). Other non-HLA allelic risk variants are associated with pathways involved in T cell development, activation, and function, further highlighting the importance of T cells are the key drivers of autoimmunity (19).

While the precise mechanisms that lead to loss of tolerance are multifaceted, HLA-DQ8 susceptibility implies that the stability of the tri-molecular complex is an important aspect that underlies autoimmune T cell responses (Figure 1). The inbred NOD mouse model that possesses a single susceptible MHC class II allele I-A^g7 (I-A^dα/I-A^g7β) has played a vital role in uncovering the mechanisms involved in the development of T1D. The structural similarities characterized by the shallow peptide-binding groove and the positive charge in the p9 peptide-binding pocket present in both I-A^g7 and human HLA-DQ8 point to a similar mechanism of autoimmune susceptibility (22). The potential mechanisms include altered thymic selection due to peptide:MHC instability, and/or preferential binding and presentation of beta cell neo-antigens formed via post-transnational modifications in the periphery (Figure 1). Biochemical analyses revealed a propensity for both DQ8 and I-A^g7 to bind peptides with negatively charged C-terminus (48). In the case of celiac disease, which is also associated with DQ8 susceptibility, gluten peptides targeted in disease have a negative charge at the C-terminus, which results in their stable binding to DQ8 (49). Although this observation suggests that key epitopes targeted in T1D should similarly contain negatively charged residues at p9, most beta cell antigenic epitopes lack this trait. Moreover, the dominant insulin epitope B:9-23 has a positively charged arginine at the C-terminus. Nevertheless, in support of this hypothesis, a mutation of InsB_9–23 at presumptive p9 to a negatively charged glutamic acid increased the immunogenicity of the epitope and augmented the activation of insulin-specific T cells (50). In addition, a recent study has identified IAPP and Chromogranin A (ChgA) epitopes in beta cells that have been modified by peptide fusion to acquire a negative charge at the C-terminus (35). The modified peptides were significantly more immunogenic compared to unmodified wild-type epitopes. This groundbreaking finding offered a potential explanation for lack of efficient thymic selection under conditions of unstable tri-molecular complex formation in the thymus, followed by priming and activation of autoreactive T cells in response to modified and stable peptide:MHC complexes in peripheral tissue.

A body of evidence suggests an important role for altered thymic selection in the development of autoimmunity. Negative selection of autoimmune lymphocytes depends on sufficient amount of self-antigen available for presentation in the thymus, which is regulated by intra-thymic and extra-thymic sources, genetic variation in tissue antigen promoters, and effective antigen presentation on certain HLA alleles (Figures 1 and 2A). Normally, tissue-specific antigens are presented by Autoimmune regulator (Aire) and Fezf2 expressing thymic medullary epithelial cells (mTECs) to aide in the deletion of self-reactive thymocytes (51, 52). MTECs can also transfer antigens, including beta cell antigens, to thymic resident dendritic cells (DCs), which in turn delete self-reactive T cells (53). Both DCs and Aire expressing mTECs are also essential in generating thymically derived Foxp3⁺ regulatory T cells (Tregs), a critical population for the establishment and maintenance of self-tolerance (54, 55). Indeed, there appears to be a correlation between a reduction in DC numbers and residual β cell function in T1D subjects (56), while the NOD mouse exhibits an overall reduction in DCs (57, 58). These observations suggest a relationship between self-tolerance and the absolute number of DCs present in the thymus and periphery. However, not all peripheral antigens are expressed by mTECs and, therefore, negative selection must also rely on peripheral antigen retrieval and delivery to the thymus by DCs. Importantly, studies have shown that the generation of thymic regulatory T cells by antigen-presenting mTECs and DCs early in life (neonatal) is critical in maintaining tolerance to self (55, 59). The idea of peripheral antigen exposure generating tissue-specific Tregs was elegantly demonstrated by Scharschmidt et. al., where skin colonization of S. epidermidis allowed for the development and trafficking of microflora-specific Tregs to the skin. Using sphingosine-1-phosphate receptor antagonist, FTY720, the authors blocked Treg egress and pinpointed the thymus as the main source for Treg development (59).

There is no direct evidence for thymic selection deficiencies in individuals with T1D; however, several key observations suggest that there is a role for altered selection in the development of autoimmune responses to insulin. The level of thymic insulin expression in humans is controlled in part by the polymorphic variable number of tandem nucleotide repeats found in the region proximal to the promoter region of the insulin gene (INS–VNTR) (60). It has been shown that VNTR I alleles express 26–63 tandem repeats while the VNTR III carries 141–209 repeats. This difference translates into higher thymic transcript levels for the VNTR III individuals and a threefold to fourfold relative protection from T1D (61, 62). It appears that the number of repeats affects AIRE binding to the insulin promoter region, thus controlling transcriptional regulation of insulin in the thymus (51, 62, 63). In support of alterations in thymic selection, analysis of human peripheral blood from T1D patients and healthy controls revealed that subjects expressing the INS-VNTR I (T1D-predisposing) allele displayed elevated frequencies of high affinity proinsulin-specific T cells compared to INS–VNTR I HLA–DR4 subjects (64). INS-VNTR allelic expression appears to determine insulin reactivity rather than the total number of insulin-reactive T cells, as both VNTR I and VNTR III groups displayed similar total number of insulin-reactive T cells in peripheral blood (64). However, it has only been hypothesized that the differences in thymic insulin expression between VNTR I and VNTR III subjects influence positive and negative selections of insulin-reactive T cells, but this has never been formally demonstrated in vivo due to a lack of VNTR mouse models.

The role of thymic insulin expression in the establishment of central tolerance has been addressed in the NOD mouse model by both deletion and overexpression of insulin in the thymus. Deletion of insulin specifically in thymic Aire expressing mTECs enhanced diabetes development in both male and female mice (65). In addition, transgenic overexpression of proinsulin, but not GAD65 (66) or IGRP (67), significantly delayed (68) or prevented (69) diabetes progression in NOD mice. However, in these studies overexpression of insulin was targeted to all MHC class II expressing APCs and, therefore, the relative role of central compared to peripheral tolerance was not determined (68, 69). A more recent set of experiments determined that a narrow window of ectopic proinsulin expression in APCs (from birth until weaning) could prevent the development of diabetes in NOD mice (70). This timeframe fits with a previous study that showed organ specific autoreactive T cell escape from the thymus is greatest during the first 10 days of life in NOD mice (71). In the former study by Jhala et. al., protection was due in part to the deletion of insulin-specific T cells, but also the inability of the remaining insulin-specific T cells to respond to cognate antigen in periphery (70). In our recent study, we tested two TCRs (4-8 and 12-4.1, Table 1) with defined affinities for InsB_9–23 for their ability to escape negative selection in the presence of ectopic overexpression of insulin. Surprisingly, we did not observe any increase in thymic deletion of the relatively high (4-8) or low (12-4.1) TCRs, although the increase in insulin expression did protect mice from developing autoimmune diabetes. Protection from disease appeared to be due to an increase in Treg development with a significant increase in thymic, splenic, and pancreas-residing insulin-specific Tregs (72). These findings pose an intriguing possibility that the amount or stability of self-peptide:MHC complexes during thymic selection is more important for Treg development rather than deletion of self-reactive T cells (Figure 2A).

Chromogranin A is the only other currently known beta cell antigen necessary for the initiation of autoimmune diabetes in NOD mice; however, expression of ChgA in the thymus has not yet been detected (73). Therefore, tolerance to ChgA may rely in part on transport of antigen by peripheral DCs to the thymus. Whether islet-derived antigens are carried to the thymus to promote islet-specific Treg development has not been explored; nevertheless, the divergent TCR repertoire between islet-infiltrating effector and regulatory T cells suggests a lack of local Treg conversion in favor of thymic lineage being the predominant Treg population in the pancreas (74). The thymic Treg niche was thought to be highly specialized and restricted (75); however, a recent study has demonstrated that the manipulation of either the number of antigen-presenting cells or an increase in antigen exposure within the thymus can expand the Treg niche (54).

While highly self-reactive T cells are removed from the T cell repertoire by negative selection, the quality and the quantity of self-reactive Tregs that develop from the moderately self-reactive thymocyte pool is a critical component of peripheral self-tolerance (Figure 2A). This idea is consistent with the observation that healthy individuals possess significant frequencies of self-reactive T cells, but are free from autoimmunity (76, 77). The escape of self-reactive T cells in itself is not just a byproduct of Treg development, but seems to serve an important immunological purpose, since some level of self-reactivity is associated with enhanced responsiveness to foreign pathogens (78–80). It is likely that the fine balance between beneficial self-reactivity and self-tolerance is uniquely perturbed in individuals with a susceptible genetic background. A slight change in thymic antigen expression or the overall stability of the tri-molecular complex could shift the T cell development spectrum toward Treg insufficiency, rather than escape of higher affinity cells. Therefore, the ratio of beta cell antigen-reactive Tregs vs. effector or memory T cells might be a better predictive biomarker of autoimmunity than the overall frequencies of self-reactive cells.

Autoimmune T cell responses in general, as well as, the T cell population that infiltrates the NOD pancreatic islets, are composed of cells with different T cell lineages, diabetogenic or regulatory capabilities, antigenic specificities, and TCR affinities (10, 81–83). All of these parameters are directly influenced by the TCR (84). Therefore, TCR sequence, specificity, and affinity hold the key to understanding the dynamics of diabetogenic T cell responses during chronic progressive autoimmune disorders, such as T1D. The antigenic specificity of each TCR is dictated by the highly variable CDR3 region found within the α and β chains of the TCR heterodimer. The variability is the result of random genetic recombination events that bring together one of many variable (V) genetic segments with a joining (J) region. The large number of TCR sequences infiltrating an organ, their variability among individuals, and the heterodimeric structure of the TCR has been a significant roadblock in a comprehensive functional analysis of TCRs. In this section, we will summarize the studies that have investigated beta cell-specific TCR parameters for their ability to predict T cell pathogenic potential.

While we have learned a great deal from the NOD mouse, there are certain limitations to the conclusions and parallels we can draw to human T1D. In order to improve the model, several HLA transgenic mouse strains expressing susceptible or protective alleles have been generated, some of these on the NOD background. Surprisingly, NOD mice expressing susceptible DQ8 or DR4 alleles do not develop spontaneous diabetes (141–143). However, HLA-DQ8 humanized mice do develop spontaneous autoimmune cardiomyopathy (144). Still, both DR4 and DQ8 alleles support the development of beta cell-reactive autoimmune T cells but require an additional trigger to initiate beta cell targeted autoimmunity. When DR4 and DQ8 mice were crossed with transgenic mice expressing B7.1 co-stimulatory molecule on beta cells, both HLA-humanized strains developed spontaneous diabetes (141). The main utility for HLA-humanized mice has been realized by performing systematic identification of the key antigenic epitopes presented on human HLAs (24). Future studies should be extended to assess the in vivo functional potential of human autoimmune TCRs specific for key immunogenic epitopes. To date only one beta cell antigen-reactive human TCR transgenic mouse with specificity for GAD65 has been described (101). In vivo functional analysis of TCRs, and human TCRs in particular, has been hindered due to limited access to patient samples, labor, and time involved in generating TCR transgenic mice. We have overcome the limitation of TCR transgenic system by utilizing a TCR retrogenic approach that allows rapid functional analysis of multiple TCRs through retroviral gene delivery (110, 145, 146). Using this approach, we have expressed a GAD65_115–127 reactive TCR isolated from peripheral blood of an individual diagnosed with T1D (99). Although we observed robust development of GAD-reactive T cells in this system, similar to the transgenic expression, we detected a low level of islet infiltration. Future analyses should be expanded to other beta cell protein epitopes targeted in human T1D, including PTM epitopes. The humanized TCR retrogenic approach will allow efficient and relatively high-throughput analysis of autoimmune antigens important in human disease, and can be utilized as a platform for development of antigen-specific immunotherapies. It is likely that many questions pertinent to our understanding of autoimmune T cell development and pathogenicity will be eventually addressed in the context of human susceptible HLA alleles and human TCRs.

The biology of low-affinity autoimmune T cells has been perplexing due to the seeming contradiction between suboptimal in vitro responses and robust in vivo pathogenicity. In many cases, self-reactive autoimmune T cells do not follow the dogma prescribed by studies performed with T cells specific for infectious or model antigens. In addition to unusual TCR:pMHC interactions and downstream signaling, autoimmune antigens themselves can have atypical characteristics. Over the years, it has become clear that antigens targeted in autoimmunity, and particularly in T1D, are often modified versions of self-peptides that are presented during thymic selection. These exceptions to the rule characteristic of autoimmune T cell responses are often centered on the stability of the tri-molecular complex as a master switch from tolerance to autoimmunity.

MB and MLB developed the concept, prepared the figures, and wrote the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.