Genetics of Type 1 Diabetes: What's Next?

Although the highly polymorphic HLA class II genes clearly play the most important single role in susceptibility to type 1 diabetes, variation at these loci alone cannot explain all of the evidence of genetic association and linkage of the MHC with type 1 diabetes. To better define genes within the MHC that may affect type 1 diabetes risk and would therefore merit further studies, the T1DGC undertook a comprehensive study of the genetics of the classic 4-Mb MHC region. More than 3,000 SNPs and 66 microsatellite markers were genotyped in 2,300 type 1 diabetes ASP families (∼10,000 individuals) (10). HLA typing using immobilized probes was also performed on these samples for HLA-A, -B, -C, -DRB1, -DQA1, -DQB1, -DPA1, and -DPB1. These data (available for viewing at http://www.t1dbase.org) represent the largest collection of families with type 1 diabetes genotyped at such a detailed level.

Specific combinations of alleles, or haplotypes, of the DRB1, DQA1, and DQB1 genes (in cis and in trans) determined the extent of risk and a distribution of DR-DQ haplotypes and genotypes ranging from highly susceptible to highly protective have been observed (11). Odds ratios (ORs) >40 were observed for some genotypes (e.g., DRB1*0301-DQA1*0501-DQB1*0201/DRB1*0401-DQA1*0301-DQB1*0302). High genetic risks have been reported for islet autoimmunity and type 1 diabetes in DR3/4-DQ8 siblings who shared both HLA haplotypes with their diabetic sibling, although this has yet to be confirmed (12). Further, independent effects of HLA-A, HLA-B, and HLA-DPB1 (13) were also demonstrated. Following adjustment for linkage disequilibrium to haplotypes at the DR-DQ region, both susceptible and protective alleles were found at HLA-B (e.g., B*3906, susceptible, and B*5701, protective), HLA-A (e.g., A*2402, susceptible, and A*1101, protective), and HLA-DPB1 (e.g., DPB1*0301 and *0202, susceptible, and *0402, protective) (13,14).

Other features of the HLA–type 1 diabetes association were also examined; however, only support for an HLA effect by age at diagnosis was found (15,,–18). Presumably, the risk conferred by specific HLA class I and class II alleles and haplotypes reflects the specificity of peptide binding and presentation (19,20). New genomic knowledge will better define the naturally processed peptides from autoantigens in type 1 diabetes. Intriguingly, a decrease in high-risk HLA genetic contribution in new-onset cases over the last decades has been observed in several studies, suggesting a change in environmental impact on penetrance as the incidence of type 1 diabetes increases (21,–23).

The T1DGC MHC fine mapping data and results were published as a supplement to Diabetes, Obesity and Metabolism (10). The T1DGC has made the data available to the scientific community for additional analyses, by request to the National Institute of Diabetes and Digestive Kidney Diseases (NIDDK) Central Repository (https://www.niddkrepository.org/niddk/home.do). The T1DGC is also probing the MHC in greater genomic detail, including a collaboration with the Federation of Clinical Immunology Societies (FOCiS) and with DNA sequence analysis to investigate the association of the “secondary” DRB3, DRB4, and DRB5 variation in the context of the DRB1 haplotypes on which these alleles are found.

The importance of variation at or near the insulin gene (INS) on chromosome 11p15.5 was originally suggested by early association studies (24). The genetic risk conferred by the INS locus is generally ascribed to differing size classes of alleles at a region with a variable number of tandem repeats (VNTR, mini-satellite polymorphisms) flanking the insulin gene. The class I alleles of the INS VNTR, which increase risk of type 1 diabetes, have been associated with lower insulin mRNA and protein expression in the thymus, compared with the dominant protective class III alleles. Decreased central tolerance allows more autoreactive T-cells to escape into the periphery, increasing susceptibility to disease (25,26). Recent studies that highlight insulin and its precursors as the major initiating autoantigen in human type 1 diabetes (27,28) provide support for this hypothesis.

The variants associated with type 1 diabetes in the cytotoxic T-lymphocyte–associated protein 4 (CTLA4) gene were identified by association mapping using both NOD mouse and human samples (29,–31). The CTLA4-encoded molecule is a co-stimulatory receptor that inhibits T-cell activation and functions in CD4 T regulatory cells. Several human autoimmune diseases are associated at the same genomic region (2q33) that contains CTLA4. Narrowing down the list of candidate causal variants and their effect on CTLA4 gene splicing has been aided by using samples from patients with Graves' disease (31) and point to variants in the 3′ region of the gene, altering the level of a soluble form of the receptor. CTLA4 genetic variation has a strong effect, presumably via its role in regulation of peripheral tolerance (32), in which the disease-associated CTLA4 haplotype is predisposing to a failure in tolerance to multiple organs or tissues. In the NOD mouse in which convincing statistical gene-gene interactions can be observed, the effect of allelic variation of CTLA4 depends on different combinations of other susceptibility loci, including complete masking of the effect (that is, no association with disease) (33).

A functional variant of the lymphoid-specific protein tyrosine phosphatase (PTPN22) gene on chromosome 1p13 is strongly associated with type 1 diabetes as well as other autoimmune diseases (34,35). LYP, encoded by the PTPN22 gene, is an inhibitor of T-cell activation, acting by dephosphorylating T-cell receptor-proximal signaling molecules such as LCK and ZAP70. A variant of PTPN22 resulting in an amino acid substitution (R620W) has been shown to have functional consequences for PTPN22 function in vitro and in vivo, and may be the causal variant in this region. Provocatively, in the current in vitro immunoassays of β- and T-cell activation and cytokine production, the R620W variant is a gain-of-function allele, suggesting that inhibition of LYP might be a therapeutic target in type 1 diabetes.

The IL-2R α-subunit of the IL-2 receptor complex locus (IL2RA) was found to be associated with type 1 diabetes using a tag SNP approach (36). The gene IL2RA is found on chromosome 10p15.1 and encodes the expression of CD25 on regulatory naive T-cells, memory T-cells, and activated monocytes (37). The regulated expression of the CD25 protein is important for suppressing T-cell proliferation by an immunogenic stimulus. IL2RA has been identified as an associated gene in multiple autoimmune diseases (38,–40). Recent fine mapping and functional studies have identified several variants that make independent contributions to risk for type 1 diabetes, indicating that IL2RA is the causal gene in the region. Different IL2RA variants influence the risk for development of multiple sclerosis, another autoimmune disease (41). In type 1 diabetes, the noncoding variants in IL2RA alter gene transcription, affecting expression of CD25 on the surface of naive and memory T-cells, and IL-2 production by stimulated memory T-cells (42). These human results parallel those observed in mouse studies, in which the CD25 ligand, Il2 (the gene encoding the key cytokine IL-2), has been identified as the major non-MHC risk gene (43).

Previous studies using candidate gene approaches have suggested many additional loci contributing to susceptibility of type 1 diabetes susceptibility (44). However, numerous early studies were underpowered, owing to limitations in genomic information and genotyping technology, as well as small sizes of available cohorts.

The T1DGC, using the same samples as in the MHC and candidate gene investigations, reevaluated 382 SNPs from 21 recently reported candidate genes, assembling nearly 4,000 ASP families and fully characterizing (through tagging SNPs and reported variants) the genetic contributions to type 1 diabetes risk. These results suggest that, aside from the MHC, 11p15 (INS), 2q33 (CTLA and other genes), 10p15.1 (IL2RA), and 1p13 (PTPN22), few of these published candidate genes can be replicated. In addition, a total of 1,715 SNPs were selected from the Wellcome Trust Case Control Consortium (WTCCC) GWA study of type 1 diabetes, and 581 SNPs were selected that exhibited association with autoimmune disease and type 2 diabetes loci (45,46). These studies confirmed established loci (above) (47,48) and suggested additional risk conferred by loci on chromosomes 5q31 (TCF7 [P19T], transcription factor 7, T-cell specific, HMG-box), 18q12 (FHOD3, formin homology two domain containing 3), and Xp22 (TLR8/TLR7 toll-like receptor 8/toll-like receptor 7). Type 1 diabetes has many susceptibility loci and therefore pathways in common with autoimmune diseases. With the recent exception of GLIS3 (49), no genetic overlap was found between type 1 diabetes and type 2 diabetes loci (45,46,50). The dataset established by the T1DGC from its Candidate Gene Workshops is available from the NIDDK Central Repository.

A number of genome-wide scans for linkage to type 1 diabetes have been reported (4,51,,,–55). All these studies consistently demonstrated linkage of type 1 diabetes to the MHC and specifically to the HLA genes on human chromosome 6p21.3. Additional regions with evidence of linkage have been identified, but many of these regions have not been reproduced in independent studies.

The T1DCG has completed genome-wide linkage studies, including a meta-analysis of data from previous linkage studies with a subset of T1DGC families (4), as well as the largest ASP linkage study in type 1 diabetes (56). Five non-HLA regions (Table 1), and a distinct locus located in the broad HLA linkage peak, showed some evidence of linkage to type 1 diabetes. In general, the peaks delineated broad regions with multiple identified associated loci (www.t1dbase.org). Both INS and CTLA4 are included among the identified regions from linkage. By applying family-based association testing to the linkage data from T1DGC families, one novel region associated with type 1 diabetes was identified, the UBASH3A region on chromosome 21, which has been confirmed in additional datasets (57). UBASH3A is expressed exclusively in T-cells, and animal studies implicate it in T-cell signaling.

Data from T1DGC genome-wide linkage experiments are available to the scientific community by request. Linkage studies in complex human disease are now recognized to have limited sensitivity due to the typical small locus-specific effect sizes. A major focus of current research is on the identification of putative risk genes with rarer or structural variants that could contribute to disease, and it is possible that the regions showing some evidence of linkage harbor variants that are not common SNPs well covered by the currently available genotyping platforms (58).

During the past few years, GWA studies have represented a paradigm shift in strategies for identifying risk genes for complex (multifactorial) human diseases, including type 1 diabetes (Fig. 1). This research has been made possible by the developments of high-density SNP genotyping arrays, analytical methods that build on the synthesis of population genetics, statistical genetics and genetic epidemiology, and the use of large clinically well-characterized case and control populations, as well as family collections, as that provided by the T1DGC. Careful attention to study design has been essential to eliminate or minimize bias (59,,–62). Type 1 diabetes genetic research has benefited from the collaboration among investigators, since T1DGC members have provided access to their own large collections to complement the T1DGC collection of families, case subjects, and control subjects. The family collections have proved invaluable not only for replicating case-control results but also for providing additional validation of the selection of control cohorts and their geographical and ethnic group matches to case subjects and investigating parent-of-origin effects.

The recently completed T1DGC GWA study, meta-analysis, and replication study included data from >30,000 individuals (47; http://www.t1dbase.org). Excluding HLA, there were 41 regions in the human genome that provided evidence of association with type 1 diabetes (P < 10⁻⁶) (Table 2 and Fig. 2). Fifteen of these 41 regions were previously reported. Of the 26 novel regions, 18 were replicated in independent case-control and family collections (overall P < 5 × 10⁻⁸). Four additional SNPs were associated (P < 0.05) in the replication study but failed to reach genome-wide significance (overall P < 5 × 10⁻⁸) (Table 2). Over 100 other regions had SNPs that achieved associations with type 1 diabetes at borderline levels of significance (10⁻⁶ < P < 10⁻⁵). Overall, the T1DGC GWA study and meta-analysis (48,63) provided convincing evidence for >40 non-HLA type 1 diabetes risk loci, with effect sizes of alleles ranging from OR = 2.38 (11p15.5, INS) to OR = 1.05 (17q21.2; SMARCE1). Many of these loci contain genes that affect the immune response (Table 2 and Fig. 2), although alternative, and as yet unknown, pathways may be implicated, including, for example, several genes such as IFIH1, GLIS3, and PTPN2 strongly expressed in β-cells.

In type 1 diabetes, initial analyses suggest that the risk conferred by non-HLA loci appears to be lower in ASP families (already enriched for high-risk HLA genes) than in sporadic cases (47). For example, it has been observed that some non-HLA SNPs (e.g., TCF7 P19T) show evidence of association with type 1 diabetes only in families that are not HLA-DR3/DR4, the highest HLA risk (64). This result suggests the presence of interaction, or a departure from the multiplicative model (statistical independence of the distributions of genotypes at two nonlinked loci). Further analysis is needed to fully clarify this observation (65). It seems likely that these interactions are small and, as a result, the biological interpretation and impact of such interactions will be difficult.

As suggested by the results in Table 2 and Fig. 2, many of the identified non-HLA regions contain candidate genes that are plausible by functional considerations. The median size of the identified regions is 255 kb (range 68 kb to 1.9 Mb), and they contain between 0 and 27 known genes. This suggests that there are >300 candidate genes, if we assume that the causal gene(s) is in the linkage disequilibrium, LD, region. However, since a causal variant in an associated region could affect transcriptional regulation of a gene several thousand base pairs away, owing to the existence of long-range regulatory elements or enhancers, including the number of candidate genes within 0.5 Mb on either side of an associated region brings the number of candidates in the order of 1,000 protein coding genes and ∼500 non-protein coding pseudogenes and RNA-encoding sequences (http://www.t1dbase.org). It is evident that a combination of further more detailed genetic mapping, and genotype-phenotype correlation studies, are necessary for identification of the causal genes within these regions. Some of these studies are underway—a recent initiative on Fine Mapping and Gene Function in Type 1 Diabetes, supported by the National Institutes of Health, supports several different approaches.

Studies to evaluate the molecular differences in gene regulation or function that are due to the supposed causative genetic risk variants (e.g., protein expression level and differences in cellular function between case and control subjects) are needed to explore the mechanisms through which the causal variants generate disease risk. Even when a gene has an obvious potential to explain pathogenesis and to be a component in the disease mechanism, inferences concerning function may be limited. Furthermore, several of the identified loci do not suggest genes with known functions: in fact, some of the associated regions do not contain annotated genes, pointing to potential contribution of long-range gene expression regulatory elements and/or nonprotein coding RNA genes. The vast majority of the currently reported associations also do not point unambiguously to a particular gene, but to several within, and outside, a block of linkage disequilibrium. Thus, genes that may clearly be implicated are often not annotated with respect to function.

In silico analyses and experimental data indicate that up to 50% of conserved cis-acting elements in the human genome may be 1 Mb from target genes, sometimes in the introns of neighboring genes, although most regulatory sequences are within 50 kb of the gene (66). Thus, “true” type 1 diabetes genes may be some distance from the association signal, although the reported association provides an anchor point on which to base functional studies.

Recent data (67) and ongoing investigations indicate that other types of common genetic variation (e.g., copy number or structural variants, such as deletions and duplications) may contribute little to the observed familial clustering of type 1 diabetes risk. However, rare loss-of-function structural gene variants could still make an important contribution to type 1 diabetes risk, through identification of which particular gene in a region of association could harbor a causal variant. With further advances in array and sequencing technologies, it is anticipated that such loss-of-function variants will be identified that influence susceptibility to type 1 diabetes (68).

Each newly identified association of a candidate locus with type 1 diabetes presents new challenges. Finding the causal genes and the causal variants, understanding how they affect disease pathophysiology, and dissecting their contribution to type 1 diabetes risk remain the major undertakings. For some genes, the effect sizes of risk alleles are such that larger collections of patients will be needed to identify the causal genes and limit the number of potential causal variants. Genotype-phenotype fine-mapping studies, however, can be performed with much smaller sample sizes while still achieving convincing statistical evidence (e.g., 42). Each confirmed gene, based on both statistical and functional evidence, provides a key piece of the etiology of type 1 diabetes, regardless of the magnitude of the odds ratio as a measure of the population association.

Combinations of many alleles, possibly hundreds, combine with effects of environmental factors (probably numerous and ubiquitous) to establish the risk profile for type 1 diabetes. Each common variant in isolation has a subtle effect on disease risk, but each may alter a key function in the immune system and its interaction with pancreatic β-cells. Recent discussion of “missing heritability” for complex human traits has considered the source of this variation and appropriate research strategies to detect these genetic effects (61). Studies in populations that are distinct from Europeans or European ancestry, such as populations of recent African ancestry or from Asian countries, are likely to narrow the large chromosomal regions of association identified in current studies and to increase the yield of rare variants (69). Future studies examining rare variants, structural variation, and polymorphisms not well imputed should be helpful in uncovering the remaining missing heritability in type 1 diabetes.

A recent sequencing study provides an example of detection of rare variants in type 1 diabetes. Targeted sequencing in a series of candidate coding regions resulted in IFIH1 being identified as the causal gene in a region associated with type 1 diabetes by GWA studies (58). IFIH1 encodes a cytoplasmic helicase that mediates induction of the interferon response to viral RNA. The discovery of IFIH1 as a contributor to susceptibility to type 1 diabetes has strengthened the hypothesis (70) about a mechanism of disease pathogenesis involving virus-genetic interplay and raised type 1 interferon levels as a cofactor in β-cell destruction. Nonetheless, it should be recognized that a component of the missing heritability (familial aggregation) in type 1 diabetes could well be due to unrecognized intra-familial environmental factors.

Contemporary models of pathogenesis of type 1 diabetes support the involvement of two primary dramatis personae: the immune system and the β-cell. The known and newly identified genetic risk factors for type 1 diabetes present exciting opportunities to build on to the current cast of disease mechanisms and networks. Most of the listed genes of interest (Table 2) and those in extended regions are assumed to regulate immune function. Some of these genes, however, may also have roles in the β-cell (insulin being the most obvious example). Another gene, PTPN2, encoding a protein tyrosine phosphatase, was identified as affecting the risk for type 1 diabetes as well as for Crohn disease (47,71). PTPN2 is expressed in immune cells, and its expression is highly regulated by cytokines. However, PTPN2 is expressed also in β-cells, where it modulates interferon (IFN)-γ signal transduction and has been shown to regulate cytokine-induced apoptosis (72). Other candidate genes, such as NOS2A, IL1B, reactive oxygen species scavengers, and candidate genes, identified in large GWA studies of type 2 diabetes, have not been found to be significant contributors to the susceptibility of type 1 diabetes (73).

Recently, new relationships between type 1 diabetes and other autoimmune and inflammatory diseases have been uncovered (63,71,74) (Table 3). Certain HLA haplotypes have long been known to strongly influence genetic predisposition to autoimmunity (75). The contribution of the specific HLA component differs considerably among different autoimmune diseases, but most relate to the function of the adaptive immune response and the binding and presentation of specific peptides. The results of GWA studies have reinforced the belief that type 1 diabetes is an autoimmune disease and that HLA is the major genetic determinant of risk for type 1 diabetes. Importantly, there is a substantial overlap in non-HLA susceptibility loci between type 1 diabetes and other autoimmune diseases (76). This overlap in genetic susceptibility locus (although not necessarily the same causal variant [41]) supports the concept that genetic risk in autoimmunity is determined in part by variation in genes that act on control mechanisms of the immune system. It will be important to identify loci that are distinct to type 1 diabetes (such as the INS locus), since these loci may illuminate type 1 diabetes–specific pathways.

The T1DGC is participating in a follow-up study of multiple autoimmune disease consortia. This project identifies significant loci from GWA studies to develop the ImmunoChip, a 200,000-SNP custom array that will provide dense SNP mapping of regions that have been associated (at genome-wide significance) with autoimmune diseases. Both individual and shared regions of the genome will be assayed across autoimmune diseases. These results will be made available through T1DBase, and the data will be made available from the NIDDK Central Repository.

Recently, Clayton (65) evaluated the genetic architecture of type 1 diabetes from the GWA meta-analysis study conducted by the T1DGC. It was concluded that the principal value of the newly discovered SNPs would be to increase our understanding of disease pathogenesis, rather than increase our ability to predict disease development on an individual level. Even if we could explain all the familial clustering of the disease (genetic and environmental factors), of which the largest contributor is the HLA, receiver operator curve analyses showed that the positive predictive value is limited, where a trial designed to capture 80% of all future cases has to treat 20% of the general population, of whom only <0.5% will develop type 1 diabetes.

The ultimate objective of genetic research is the translation of genetics findings into advances in clinical care. An obvious question is, “What can a risk gene with an OR in the range of 1.05–1.2 add to clinical treatments for type 1 diabetes?” However, a low OR does not disqualify the encoded protein as a potential drug target. Both PPARG and KCNJ11 are genes that have a low OR for type 2 diabetes risk, yet they encode for major drug targets. A major contribution of genetics to type 1 diabetes will be the identification of important disease pathways that can be examined for new therapeutic targets or biomarkers, including the stratification of subjects at risk for interventions or patients for effective treatment (and prevention of complications).

Redefining and stratifying human disease, especially with regard to pharmacological response, in the post-GWA era is essential. A new approach to classifying human disease that both appreciates the uses and limits of reductionism and incorporates the tenets of the nonreductionist approach of complex systems analysis is necessary. Disease phenotypes reflect consequences of variation in complex genetic networks operating within a dynamic environmental framework. Further genetic and functional evaluations, conducted at the highest levels of experimental rigor and repeatability and reproducibility, are necessary to establish and confirm involvement of such networks in type 1 diabetes, to fully elucidate the biological mechanisms of the networks and to identify the strongest risk phenotypes (77,78). Some phenotypes will be regulated by several of the type 1 diabetes genes and may well be precursors of disease, appearing at the earliest stages of the development of type 1 diabetes and perhaps even preceding aggressive autoimmunity (79).

Recently, it was suggested that since the vast majority of disease genes show no tendency to encode highly connected protein hubs but are localized to the functional periphery of networks (80), they are not essential for explaining disease pathogenesis. The counterargument is to consider that cellular networks are modular, consisting of groups of highly interconnected proteins responsible for specific cellular functions. Disease pathogenesis represents the perturbation of probably many specific functional modules caused by a variation in one or more of the components producing recognizable developmental and/or physiological dynamic instability (81). Such a model offers a hypothesis for the emergence of complex or polygenic disorders—a phenotype often correlates with the inability of a particular functional module to carry out its basic function. For extended modules, many different combinations of gene variants might incapacitate the module and lead to the same clinical phenotype. The correlation between disease pathogenesis and functional modules can improve our understanding of cellular networks by helping us to identify which genes are involved in the same cellular function or network module. Pathogenic processes may progress to clinical disease such as type 1 diabetes; alternatively, these processes may be interrupted at subclinical levels. The identification of such phenotypes or disease precursors is therefore a key aim. Comprehensive gene expression studies in cells and tissues relevant to type 1 diabetes will help lead to identification of relevant networks. Importantly, the association of disease with functional networks may also influence our choice of new therapeutic targets.

The greatest genetic risk (both increased risk, susceptible, and decreased risk, protective) for type 1 diabetes is conferred by specific alleles, genotypes, and haplotypes of the HLA class II (and class I) genes. There are currently about 50 non-HLA region loci that also affect the type 1 diabetes risk. Many of the assumed functions of the non-HLA genes of interest suggest that variants at these loci act in concert on the adaptive and innate immune systems to initiate, magnify, and perpetuate β-cell destruction. The clues that genetic studies provide will eventually help lead us to identify how β-cell destruction is influenced by environmental factors. While there is extensive overlap between type 1 diabetes and other immune-mediated diseases, it appears that type 1 and type 2 diabetes are genetically distinct entities. These observations may suggest ways to help identify causal gene(s) and, ultimately, a set of disease-associated variants defined on specific haplotypes. Unlike other complex human diseases, relatively little familial clustering remains to be explained for type 1 diabetes. The remaining missing heritability for type 1 diabetes is likely to be explained by as yet unmapped common variants, rare variants, structural polymorphisms, and gene-gene and/or gene-environmental interactions, in which we can expect epigenetic effects to play a role. The examination of the type 1 diabetes genes and their pathways may reveal the earliest pathogenic mechanisms that result in the engagement of the innate and adaptive immune systems to produce massive β-cell destruction and clinical disease. The resources established by the international T1DGC are available to the research community and provide a basis for future discovery of genes that regulate the earliest events in type 1 diabetes etiology—potential targets for intervention or biomarkers for monitoring the effects and outcomes of potential therapeutic agents.

This research uses resources provided by the Type 1 Diabetes Genetics Consortium, a collaborative clinical study sponsored by the National Institute of Allergy and Infectious Diseases (NIAID), National Human Genome Research Institute (NHGRI), National Institute of Child Health and Human Development (NICHD), the Wellcome Trust and the National Institute for Health Research Cambridge Biomedical Centre, the Juvenile Diabetes Research Foundation International (JDRF), and Grant U01 DK062418 from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK).

No potential conflicts of interest relevant to this article were reported.