13.3.1 Genetics of type 2 diabetes mellitus

For years, it has been well known that genetic factors are crucially important for the development of type 2 diabetes. Despite major efforts in seeking to understand the molecular genetic basis, until a few years ago, only a handful of genes responsible for relatively rare monogenic and syndromic subsets of diabetes were detected, and progress in finding genetic predispositions to common type 2 diabetes was lacking. Even though the unravelling of the molecular pathogenesis of type 2 diabetes is still in its infancy, the last few years have, nevertheless, brought some interesting developments. Box 13.3.1.1 provides a glossary of terms used currently in genetics.

Heritability is defined as the proportion of phenotypic variation in a population that is attributable to genetic variation among individuals. The evidence of a genetic component in the pathogenesis of type 2 diabetes (OMIM 125853, http://www.ncbi.nim.nih.gov/omim) comes from studies of large families, twins and sibling pairs, and from adoption studies. It is evident that type 2 diabetes clusters in families and that first-degree offspring have a lifetime risk of developing type 2 diabetes of 35%, if one parent has type 2 diabetes, and 70%, if both parents have type 2 diabetes, compared to circa 10% in the general population. Translated to a recurrence risk for a sibling (λ_s) of an affected person divided by the general population risk, this amounts to a two- to three-fold increased risk in these individuals. Seen in isolation, these family studies say little about the relative importance of heritability and shared family-specific environment. In addition, evidence for a genetic component in type 2 diabetes comes from studies of monozygotic and dizygotic twins in which the relative importance of genetic and non-genetic factors can be estimated rather precisely, under the assumption that twin pairs share the same prenatal and postnatal environment and that twins resemble singletons according to the phenotype in question. Heritability estimates from twin data have shown variable concordance of type 2 diabetes in monozygotic twins of 35–70%, as opposed to 20–30% in dizygotic twins. Similarly, high degrees of heritability of diabetes-related traits have been found. Crude heritability estimates are higher for body mass index (c.70–80%) and serum lipid traits (50–70%) than for insulin secretion (c.50%) and, in particular, insulin resistance (c.40%).

Besides classical genetic variation, which alters the specific nucleotide sequence of the genome, the importance of epigenetic alteration is also becoming increasingly evident. Epigenetic alterations refer to modifications that regulate gene activity. The modifications can affect the DNA itself, or the proteins that package it (histones), but the DNA nucleotide sequence does not change. Examples of such modifications include DNA methylation and modifications of histone tails. Epigentic patterns may change in the individual over their lifetime in an age-dependent manner. However, the extent to which epigenetic modifications contribute to risk of type 2 diabetes and metabolic disturbances is currently unknown. The age-dependent development of type 2 diabetes, showing increased incidence and severity of disease with increasing age, makes epigenetic modifications an attractive disease mechanism. New technological advances are currently taking investigations of epigenetic modifications to a large-scale, genome-wide level, and this will probably lead to new insights into the molecular pathogenesis of type 2 diabetes.

Different approaches have been taken in the search for genetic contributions to complex diseases such as type 2 diabetes. The development of approaches has mainly been guided by technological advances in genotyping and sequencing techniques, statistical handling of data, and also by the collection of larger cohorts suitable for genetic studies. Before the advent of genome-wide association studies, the major approaches for finding susceptibility variation were the candidate gene approach and genetic linkage analysis, both of which are summarized below.

In the candidate gene approach, a specific gene of interest is selected based on biological or bioinformatic knowledge of relation of the gene with type 2 diabetes. This evidence can be based on in vivo studies of genetically engineered animal models or in vitro cell experiments. Variation in the gene of interest is then investigated in genetic association studies looking for evidence that a particular variant allele or genotype is overrepresented in disease cases compared with control individuals.

The biological candidate gene approach has generally had limited success in contributing to research on susceptibility genes for common diseases. Several reasons for this exist, and, while some are related to the method itself, others are more a reflection of the era in genetic research in which this approach was widely used. First, a crucial limitation of the candidate gene approach is the fundamental need to have a detailed knowledge of the disease of interest in order to be able to pick a reasonable candidate gene. As the pathophysiology of type 2 diabetes is extremely complex, any single candidate gene will have a low prior probability to affect disease susceptibility. Second, a major obstacle in most reported candidate gene studies is the study design in relation to sample size and phenotypic characterization of the studied sample. Sample sizes have, over the years, generally increased tremendously, in recognition of the very modest effect sizes inflicted by most common variants related to common disease. Statistical power analyses are important in order to assess the limitations of the study, and, for small effect sizes (allelic odds ratios ranging from 1.1 to 1.4), thousands of samples are needed to ensure validity. In order to circumvent problems with lack of statistical power, meta-analysis and large-scale replications have been increasingly applied to detect the effect of genetic variation on risk of type 2 diabetes. Although larger sample sizes, in theory, deliver more confident estimates of association, there are several pitfalls related to meta-analysis. One problem is publication bias, which refers to the fact that negative reports are often not published and a meta-analysis may therefore tend to overestimate association. Also, other biases and heterogeneity can influence the outcome of meta-analyses. Heterogeneity between studies may also be introduced by confounding by ethnic origin, age, sex, or other (measured or unmeasured) variables. Therefore, answers from meta-analysis in genetic epidemiology should be carefully considered and interpreted.

Linkage analysis seeks evidence of cosegregation between genomic markers and disease status within families, and is only reasonably powered when disease status and genotype are strongly correlated. Linkage analysis requires no prior hypothesis of the genomic regions investigated. In principle, linkage studies allow the entire genome to be screened for susceptibility loci using a limited number of highly polymorphic microsatellite markers, and have proven very successful in detecting genes involved in Mendelian monogenic diseases.

Many common low-penetrance variants thought to be involved in the susceptibility to common type 2 diabetes has been identified through genome-wide association studies. Typically, such studies utilize chip-based, high-throughput genotyping of 100 000–1 000 000 single-nucleotide polymorphisms (SNPs) across the entire genome to assess association for each variant with case–control status or a quantitative trait. This approach is agnostic, i.e. it assumes no prior biological hypothesis for each single variant. There is no doubt that genome-wide association studies have advanced the field of genetics and led to progress in understanding the genetic basis of numerous complex diseases. Nevertheless, analysing the vast quantity of data, validating true positive findings, as well as identifying causative genetic variants have proved to be challenging.

Different platforms for genome-wide association studies exist, and up-to-date versions provide high coverage of common variation, i.e. 70–90% of all variation with a frequency above 5% is targeted. Newer platforms also include a range of rare genetic variants and copy number variations. As a result of the vast amount of genetic variants analysed in a genome-wide association study, a high number of statistical tests are performed, increasing risk of false positives due to multiple testing. The crucial need for controlling this problem has resulted in the general use of a more stringent genome-wide significance level before an association is considered to be statistically significant. Current consensus has defined a genome-wide significance level of p <5 × 10⁻⁸ to account for multiple independent hypotheses tested in a dense genome-wide association study.

Until 2006, variation in three genes had shown convincing evidence of association with type 2 diabetes (Fig. 13.3.1.1). Two of these genes were found in a candidate gene approach inspired by known drug targets of antidiabetic medicine. PPARG, which encodes the peroxisome proliferator-activated receptor γ for which the thiazolidinediones are high-affinity ligands, has shown association with type 2 diabetes (1, 2). The PPARG P12A variant influences type 2 diabetes by changing insulin sensitivity (see Fig. 13.3.1.2; Table 13.3.1.1, below, presents an overview of type 2 diabetes susceptibility gene variants).

Similarly, the E23K variant in KCNJ11, which encodes a subunit of the sulphonylurea receptor in the β cell, has been shown to increase risk of type 2 diabetes (3, 4). As anticipated from the function of the gene product, this variant influences insulin secretion from the pancreatic β cell. Interestingly, more severe mutations in KCNJ11 are responsible for cases of permanent neonatal diabetes mellitus (OMIM 606176) and familial hyperinsulinaemic hypoglycaemia (OMIM 256450), underlining the importance of KCNJ11 in glucose regulation.

WFS1 and HNF1B have also been identified as type 2 diabetes susceptibility genes by means of the candidate gene approach. To date, the strongest type 2 diabetes susceptibility gene is TCF7L2, inflicting an increase in risk of circa 40% per copy minor allele. TCF7L2 was discovered by the typing of microsatellite markers under a previously identified linkage peak and one of the genotyped markers associated strongly with type 2 diabetes (5). Unprecedented in 2006, the association of variants in TCF7L2 with type 2 diabetes has since been widely replicated in populations of a range of different ethnic origins. Risk variants in TCF7L2 associate with impaired insulin response, possibly due to an impaired effect of the incretin hormones GLP-1 (glucagon-like peptide 1) and GIP (gastric inhibitory polypeptide).

In 2006, the first genome-wide association study investigating a complex disease was published, and, in 2007, five large genome-wide association studies of type 2 diabetes were reported (6–11) (Fig. 13.3.1.1). These studies contributed a number of novel loci, mostly with no previously known function in type 2 diabetes pathogenesis. A meta-analysis of genome-wide data from three of these studies and large-scale replication has since added six genomic loci to this growing list (12), which, during the summer of 2009, has reached 20 validated type 2 diabetes susceptibility variants (Table 13.3.1.1). For all these variants, the statistical evidence of association is strong, with a P-value lower than the genome-wide significance level, and these variants have thus been firmly established as genuine type 2 diabetes susceptibility variants.

Although 20 gene variants have shown convincing association with type 2 diabetes, it is also evident that the increases in risk conferred individually by these variants are low, ranging from risk increments of 9% to 40% per copy of risk allele. It is evident, too, that all risk alleles are common, with frequencies above 10% in the general population. The characteristics of the currently 20 validated gene variants are heavily biased by the approaches taken to find these risk gene variants, and cannot be generalized to all existing susceptibility genes for type 2 diabetes. Many of the type 2 diabetes genomic signals are located between known genes, and most actual causative variants have not been found. This implies that the disease mechanisms largely remain obscure.

Interestingly, most type 2 diabetes variants have been shown to have an impact on pancreatic β cell function, which seems to be the case for variants in or near KCNJ11, TCF7L2, WFS1, CDKN2A, HHEX, CDKAL1, SLC30A8, KCNQ1, and MTNR1B. Indeed, only the PPARG, IRS1, and ADAMTS9 variants have so far convincingly displayed a diabetogenic potential through affecting insulin sensitivity, and, for FTO variants, by increasing obesity and insulin resistance (Fig. 13.3.1.2). Although these crude mechanisms have been clarified, for most of the associated variants, the more exact and detailed mode of action in the pathways to influence risk of type 2 diabetes have not yet been elucidated.

In quantitative trait genetics of diabetes-related phenotypes, the development has been fast since the emergence of genome-wide association studies in 2007. The most investigated trait has been fasting plasma glucose, and several studies have shown highly statistically significant associations with this trait (although with modest effect sizes). A promoter variant in glucokinase (GCK) has been widely shown to increase fasting plasma glucose (13). Since the emergence of genome-wide association studies, SNPs in glucokinase (hexokinase 4) regulator (GCKR), glucose-6-phosphatase catalytic 2 (G6PC2), and melatonin receptor 1B (MTNR1B) have also shown convincing association with fasting plasma glucose (10, 14–18). MTNR1B has subsequently shown association with type 2 diabetes; however, this is not the case for variants in GCK and G6PC2. For G6PC2, data paradoxically indicate a modest protective effect on type 2 diabetes in carriers of the variant which increase fasting plasma glucose. All four genes seem to exert their pathogenic effect in the pancreatic β cell.

Despite the fact that, at present, 20 genomic loci have been shown to cause susceptibility to type 2 diabetes, the genetic background for this disease remains mostly obscure. First, this is due to the lack of biological and functional knowledge of mechanisms behind these new loci and genes. Massive efforts are needed to find causal variants and elucidate biological pathways and pathogenic impact for this growing list of associated variants. Second, for type 2 diabetes and intermediary diabetes-related phenotypes, the explained proportion of the genetic contribution is rather low, indicating the existence of a number of other, as yet undetermined genetic risk elements. For instance, for fasting plasma glucose, a number of variants have been found to influence levels of glycaemia in the general population, yet the proportion of variance in fasting plasma glucose explained by these variants remains low. Together, the 20 type 2 diabetes susceptibility variants combined with the three additional loci described, i.e. with an impact on fasting plasma glucose, explain circa 2–3% of the variance in this trait in the general population. Also, for type 2 diabetes, a large proportion of the genetic risk remains unexplained. It has been estimated that the sibling relative risk, λ_s, attributable to the initial nine identified gene variants was merely circa 1.07, compared to an estimated λ_s of 2 to 3 for type 2 diabetes, indicating that the currently identified gene variants explain less than 10% of the genetic component in type 2 diabetes.

An ultimate goal of the search for the genetic determinants of type 2 diabetes would be to add this information to conventional risk markers, and, thereby, optimize algorithms for diabetes risk assessment in high-risk proportions of the population. One way of assessing this ability, based on the current understanding of the genetics of type 2 diabetes, is to estimate the prediction potential of the common validated type 2 diabetes variants by receiver–operating characteristic (ROC) curves. This procedure evaluates the potential to predict type 2 diabetes cases from a glucose-tolerant population, and the area under the ROC curve (AUC) can range from 0.5 (as by random) to 1 (perfect discrimination). A number of such studies have all reported an AUC of 0.60–0.65 in combined analysis of a high number of the presently 20 validated risk variants (Fig. 13.3.1.3) (19–21). However, these studies have also shown that genetic information applied on top of conventional risk factor modestly, but statistically significantly, increases the discriminative power. Together, these findings indicate that the genetic risk variants known at present have a weak potential for prediction of type 2 diabetes and harbour no clinical relevance at this time. However, although the effects of common type 2 diabetes loci are modest and cannot serve as a tool to predict future type 2 diabetes, these associations may still provide important insights into biological mechanisms and highlight targets for future drug developments.

The genome-wide association studies have been a valuable tool for finding common gene variants with modest impact on risk of type 2 diabetes. The statistical power estimates from these studies make it likely that many more common variants with low impact on type 2 diabetes will appear when sample sizes of genome-wide association studies increase further. However, power estimates also make it unlikely that additional common variants with impact as high as TCF7L2 (odds ratio, c.1.4) will be found since such variants would probably already have been detected by current approaches.

Most of the studies of genetics of type 2 diabetes have been based on the HapMap resource of sequence variation forming the basis for selection of variants for candidate gene studies and for genome-wide arrays (see Box 13.3.1.1). While HapMap offers good coverage for most common SNPs with a frequency above 5%, the coverage rapidly declines for alleles with lower frequency. Such low-frequency variants may be particularly important, as deleterious variants are maintained at low frequency in the population by natural selection and such variants probably also contribute to the genetic risk of the common form of type 2 diabetes. Novel variation with low frequency in the general population may be found by sequencing approaches preferentially in high numbers of both cases with type 2 diabetes and glucose-tolerant individuals. If variants with low frequency (0.5–2% in the population) and higher effect sizes (odds ratio, c.2) are detected, they will significantly increase the clinical value of genetic testing.

Finally, it must be underlined that knowledge about main genetic defects in type 2 diabetes is the entrance to the future essential studies of combined and interactive effects of genes and environment to be elucidated in a prospectively followed population-based cohort, with incident cases probably giving the most detailed information necessary to address these issues. Also, the efforts within type 2 diabetes genetics may lead to detection of differential drug responses in carriers and non-carriers of specific combinations of genes, potentially offering genotype-driven therapeutic initiatives with greater efficacy and less side-effect.