Eur J Clin Invest 2011; 41 (6): 679–692
Background Recent genome-wide association studies enlarged our knowledge about the genetic background of type 2 diabetes.
Aims This review provides an overview of the role of these novel genetic findings for the pathophysiology, prediction and treatment of type 2 diabetes.
Results The genetic susceptibility to type 2 diabetes appears to be determined by many common variants in multiple gene loci with low effect sizes. Although at least 36 diabetes-associated genes were identified, only about 10% of the heritability of type 2 diabetes can be explained. Most of the discovered gene variants have been linked to beta-cell dysfunction rather than insulin resistance, which might challenge established thinking of type 2 diabetes as a predominant disorder of insulin action. Genetic data can lead to statistically significant, but not to clinically relevant contributions to risk prediction for type 2 diabetes. Nevertheless, preliminary evidence suggests interactions between genotypes and response to lifestyle changes or drug treatment.
Conclusions Future studies need to target the issue of hidden heritability and to detect the causal gene variants within the identified gene loci. Improved understanding of the genetic contribution to type 2 diabetes may then help addressing the questions whether genotyping is useful to predict individual diabetes risk, identifies individual responsiveness to preventive and therapeutic interventions or at least allows for breaking down type 2 diabetes into smaller, clinically meaningful subtypes.
The prevalence of diabetes mellitus has been increasing rapidly in recent years . The International Diabetes Federation estimated that 285 million adults aged 20–79 years suffered from diabetes in 2010, and the number of patients with diabetes is predicted to rise to 438 million in 2030, the vast majority of 90–95% most likely due to type 2 diabetes . As these developments are occurring within one or two generations, it is extremely unlikely that genetic variation per se can serve to explain this rapid increase in case numbers. However, it is widely established that a family history of type 2 diabetes markedly increases the risk of the disease particularly in the first-degree relatives . How can these two perspectives be reconciled?
There are numerous nongenetic factors, which contribute to the aforementioned development of diabetes prevalence. Population growth, increasing life expectancy, decreasing mortality among patients with diabetes, earlier diagnosis of type 2 diabetes and higher awareness of the disease represent important demographic and health care–related aspects [1,4]. Moreover, many societies are characterised by intense urbanisation that is usually linked with important environmental and lifestyle changes such as energy-dense diet, physical inactivity, psychosocial stress factors and exposure to environmental pollutants that all lead to metabolic and inflammatory disturbances [5–9]. On the other hand, high concordance rates for type 2 diabetes in monozygotic twins [10–14], increased risk of type 2 diabetes in offspring with at least one affected parent  and differences in diabetes prevalence between ethnic groups  indicate that the heritability of type 2 diabetes exceeds 50%. Taken together, the combined epidemiological evidence strongly suggests that the genetic susceptibility to type 2 diabetes is widespread and that an interaction with modern lifestyle and environmental triggers, which were not present during the evolution of mankind, may be necessary to reveal the full extent of genetic predisposition.
Although the aforementioned family-based studies on diabetes risk point towards the existence of genes predisposing to type 2 diabetes, they do not provide estimates about their number, the effect sizes with which their variants are associated with type 2 diabetes and whether a better knowledge of the genetic architecture of type 2 diabetes may be clinically relevant. To give an overview of the role of novel genetic findings for the pathophysiology, prediction and treatment of type 2 diabetes, this review aims to address the following questions:
- (i) How many genes determine the risk of type 2 diabetes?
- (ii) What do we learn from genetic studies about the pathogenesis of type 2 diabetes?
- (iii) What is the evidence for the use of genetic data for diabetes risk prediction?
- (iv) Can genetic data help to create a personalised view of the ‘type 2 diabetes syndrome’?
How many genes determine the risk of type 2 diabetes?
Addressing this question requires to analyse the data on the genetic architecture of type 2 diabetes obtained from recent genome-wide association studies (GWAS) [17–30]. GWAS are association studies using genotype data for hundreds of thousands of genetic variants covering most of the common variation in the human genome. They are typically based on a case–control design in which frequencies of alleles for each genetic variant are compared between two groups of individuals (see references [31,32] for detailed reviews). Alleles are alternative forms of one specific gene variant at a defined chromosomal position. The most thoroughly investigated type of genetic variation in diabetes research are single nucleotide polymorphisms (SNPs), which result from exchanges of single base pairs. To identify SNPs that predispose to type 2 diabetes, GWAS were performed by comparing cases of type 2 diabetes with nondiabetic controls, but also for quantitative traits of glucose metabolism within populations. Altogether, the number of gene loci that contain variants showing significant associations (P < 5 × 10−8) with type 2 diabetes meanwhile increased to at least 36 (Table 1). Shared features of these associations are the facts that risk variants are common, i.e. allele frequency usually > 25% and in several cases even > 50%, and that effect sizes, i.e. the excess risk conferred by a single gene variant, are low. The strongest associations are currently found for gene variants in the loci TCF7L2, initially identified in an Icelandic population [33,34], and KCNQ1, first described in Asian populations [35–37] with subsequent replications in many other cohorts (see Table 1 for extension of all gene names). TCF7L2 encodes a transcription factor that has been implicated in beta-cell development and function, whereas KCNQ1 codes for a pore-forming subunit of a K+ channel that is expressed in several tissues including pancreatic islets. Carriers of one risk allele of the SNPs rs7903146 (TCF7L2) have an approximately 40% higher type 2 diabetes risk than homozygous carriers of the protective allele , whereas risk alleles of most other loci are associated with odds ratios in the range of only 1·05–1·20 [27,29,38–40]. Although the exact genetic models of inheritance are not yet known, it appears that the effects of most risk variants can be described using additive models . Additive models assume that on a log scale, the risk in carriers of two copies of a certain allele is twice than in carriers of one allele. The fact that most risk alleles are fairly common is remarkable, but may be attributed to the study design of current GWAS. Even the combination of multiple cohorts in large-scale association analyses based on c. 100 000 study participants does not have sufficient power to reveal associations of SNPs with type 2 diabetes whose allele frequency is substantially lower than 5% [27,29].
|Gene locus||Full gene name||Associated phenotype*||Putative function(s)||References|
|Candidate gene studies|
|PPARG||Peroxisome proliferator-activated receptor gamma||T2D||IR|||
|KCNJ11||Potassium inwardly rectifying channel, subfamily J, member 11||T2D||B|||
|Large-scale association studies|
|TCF7L2||Transcription factor 7-like 2||T2D, glucose, HbA1c||B||[18,19,28,33]|
|WFS1||Wolfram syndrome 1||T2D||B|||
|HNF1B (TCF2)||HNF1 homeobox B||T2D||B|||
|GWAS for type 2 diabetes|
|FTO||Fat mass and obesity associated||T2D, BMI||IR|||
|SLC30A8||Solute carrier family 30 [zinc transporter], member 30||T2D, HbA1c||B||[17–20]|
|HHEX/IDE/KIF11||Homeobox, hematopoietically expressed/insulin-degrading enzyme/kinesin-interacting factor 11||T2D||B?||[17–20]|
|CDKAL1||CDK5 regulatory subunit associated protein 1-like 1||T2D||B||[17,18,20,22]|
|IGF2BP2||Insulin-like growth factor 2 mRNA binding protein 2||T2D||B||[17,18,20]|
|CDKN2A/CDKN2B||Cyclin-dependent kinase inhibitor 2A/2B||T2D||B||[17,18,20]|
|ADAMTS9||ADAM metallopeptidase with thrombospondin type 1 motif, 9||T2D||B? IR?|||
|NOTCH2||Notch homolog 2||T2D||B|||
|CDC123-CAMK1D||Cell division cycle 123 homolog/calcium/calmodulin-dependent protein kinase ID||T2D||B|||
|THADA||Thyroid adenoma associated||T2D||B?|||
|JAZF1||Juxtaposed with another zink finger gene 1||T2D||B?|||
|KCNQ1||Potassium voltage-gated channel, KQT-like subfamily, member 1||T2D||B||[29,35,36]|
|IRS1||Insulin receptor substrate 1||T2D||IR|||
|DUSP9||Dual specificity protein phosphatase 9||T2D||IR?|||
|ZFAND6||AN1-type zinc finger protein 6||T2D||?|||
|PRC1||Protein regulator of cytokinesis 1||T2D||?|||
|CENTD2||Arf-GAP with Rho-GAP domain, ANK repeat and PH domain-containing protein 1||T2D||B?|||
|TP53INP1||Tumour protein p53-inducible nuclear protein 1||T2D||?|||
|KLF14||Krueppel-like factor 14||T2D||IR?|||
|ZBED3||Zinc finger BED domain-containing protein 3||T2D||?|||
|BCL11A||B-cell lymphoma/leukaemia 11A||T2D||?|||
|HNF1A||Hepatocyte nuclear factor 1-alpha||T2D||?|||
|CHCHD9||Coiled-coil-helix-coiled-coil-helix domain-containing protein 9, mitochondrial||T2D||?|||
|HMGA2||High mobility group protein HMGI-C||T2D||B?|||
|GWAS for type 2 diabetes–related traits|
|MTNR1B||Melatonin receptor 1B||T2D, glucose, HOMA-B, HbA1c||CR, B||[24,26,27,30]|
|DGKB-TMEM195||Diacylglycerol kinase beta Transmembrane protein 195||T2D, glucose, HOMA-B||B ?|||
|GCKR||Glucokinase (hexokinase 4) regulator||T2D, glucose, insulin, HOMA-IR||GS||[27,28]|
|GCK||Glucokinase||T2D, glucose, HOMA-B, HbA1c||GS||[27,30]|
|PROX1||Prospero homeobox protein 1||T2D, glucose||B|||
|ADCY5||Adenylate cyclase, 5||T2D, glucose, HOMA-B, HbA1c||B?||[27,28]|
|SLC2A2||Solute carrier family 2, facilitated glucose transporter member 2||Glucose||B|||
|G6PC2||Glucose-6-phosphatase 2||Glucose, HOMA-B, HbA1c||B||[27,30]|
|GLIS3||Zinc finger protein GLIS3||Glucose, HOMA-B, HbA1c||B|||
|ADRA2A||Alpha-2A adrenergic receptor||Glucose||B|||
|MADD||MAP kinase-activating death domain protein||Glucose||B?|||
|FADS1||Fatty acid desaturase 1||Glucose, HOMA-B, HbA1c||?|||
|IGF1||Insulin-like growth factor I||Insulin, HOMA-IR||IR|||
|GIPR||Gastric inhibitory polypeptide receptor||Glucose||B|||
|VPS13C||Vacuolar protein sorting 13 homolog C||Glucose||B?|||
|C2CD4B||C2 calcium-dependent domain-containing 4B||Glucose||?|||
|TMPRSS6||Transmembrane protease, serine 6||HbA1c||E|||
|ANK1||Ankyrin 1, erythrocytic||HbA1c||E?|||
|SPTA1||Spectrin, alpha, erythrocytic 1 (elliptocytosis 2)||HbA1c||E?|||
|ATP11A/TUBGCP3||ATPase type 11A||HbA1c||?|||
In addition to GWAS for type 2 diabetes, association studies focusing on values of fasting glucose, 2-hour glucose in an oral glucose tolerance test, glycated haemoglobin A1c (HbA1c), fasting insulin and surrogate markers of insulin resistance (homoeostasis model assessment, HOMA-IR) or beta-cell function (HOMA-B) have shed further light on the genetic susceptibility for type 2 diabetes. The rationale for selecting these quantitative traits was the combination of the assumptions that gene variants which are associated with increased values of these traits should also be associated with type 2 diabetes and that the use of quantitative outcomes may yield additional hits compared to analyses that use the dichotomous variable type 2 diabetes because of improved statistical power.
Initial GWAS for fasting glucose identified MTNR1B and IRS as novel susceptibility genes for type 2 diabetes [24–26], and a recent study from the Meta-Analysis of Glucose and Insulin-related traits Consortium (MAGIC) identified a total of 16 loci (nine of them novel) that were associated with fasting glucose concentrations in a nondiabetic study population . A combined analysis of these loci explained approximately 10% of the genetic variability of this trait. Of these 16 loci, only seven were also associated type 2 diabetes in replication cohorts, suggesting an incomplete overlap between glucose and type 2 diabetes–related loci. Most importantly, not all loci that are associated with fasting glucose within the ‘physiological’ range are also associated with ‘pathological’ glucose levels and risk of type 2 diabetes . A parallel MAGIC study identified variants in the loci for ADCY5, TCF7L2, GCKR, GIPR and VPS13C, which were associated with higher levels of 2-hour glucose . Noteworthy, only the variants in the ADCY5 and TCF7L2 loci were also associated with higher fasting plasma glucose, whereas the variants in the GCKR, GIPR and VPS13C loci that were associated with increased 2-hour glucose showed associations with fasting glucose in the inverse direction. Apart from previously described TCF7L2 SNPs, only SNPs in the GIPR locus were consistently associated with increased 2-hour glucose and type 2 diabetes . Interestingly, the same GWAS found that several type 2 diabetes–associated genes also had effects on HOMA-B, whereas associations with fasting insulin levels and/or HOMA-IR could only be shown for IGF-1 and GCK (Table 1).
Taken together, the MAGIC data clearly demonstrate that certain gene variants may contribute to quantitative phenotypes of glucose metabolism such as glucose levels, but not to type 2 diabetes and vice versa, and that even directions of associations with different traits may be inconsistent. Thus, GWAS for quantitative traits of glucose metabolism do help to identify novel loci that are linked to the genetic predisposition of type 2 diabetes, but not all hits from these GWAS can be considered ‘type 2 diabetes genes’.
The situation becomes even more complex when HbA1c, an indicator of mean glycaemia during the preceding 3 months and recently suggested as a criterion for the diagnosis of type 2 diabetes in the United States , is considered. A recent MAGIC GWAS for HbA1c identified a total of 10 loci with genome-wide significant association with HbA1c . These loci included six new loci near FN3K, HFE, TMPRSS6, ANK1, SPTA1 and ATP11A/TUBGCP3, and the four known HbA1c loci HK1, MTNR1B, GCK and G6PC2/ABCB11. Associations with HbA1c are likely to be caused by glycaemia for only three of the 10 loci (GCK, G6PC2 and MTNR1B), whereas the other hits map to loci with known variants for forms of hereditary anaemia and iron storage disorders and showed no association with type 2 diabetes. Thus, it can be assumed that common variants at these loci influence HbA1C levels via nonglycaemic erythrocyte and iron biological pathways .
In conclusion, the genetic susceptibility to type 2 diabetes appears to be determined by common variants in multiple gene loci with low effect sizes. So far, at least 36 gene loci have been identified that contribute to the genetic risk of type 2 diabetes, but this number can be expected to increase in the future as discussed in the following. This means that the genetic architecture of type 2 diabetes is very similar to the genetic determination of phenotypes such as height  or weight  or to the genetic background of myocardial infarction  as another example of a complex disease with genetic and nongenetic aetiology. Moreover, GWAS for quantitative phenotypes indicate that one has to distinguish between genes for the regulation of glucose metabolism in the physiological range and variants that are important for the transition into the pathophysiological state and thus progression to type 2 diabetes.
What do we learn from genetic studies about the pathogenesis of type 2 diabetes?
This question refers to the relevance of the data from the aforementioned GWAS for understanding the pathogenesis of type 2 diabetes. Although the number of loci that display associations with type 2 diabetes at genome-wide significance level has increased substantially since 2007, for most of the hits it is not clear yet which SNPs or even which genes within the identified chromosomal regions are causal variants that link genetic variation with disease risk. However, it is striking that most regions that harbour hits from GWAS for type 2 diabetes or related traits contain genes coding for proteins with putative roles in the development of beta cells and the pancreas or in the function of the adult pancreas (Table 1). In contrast, genes that have been linked to insulin resistance, obesity, glucose sensing or other aspects of glucose metabolism are much less frequent.
There are several potential explanations for the dominance of ‘beta-cell genes’ in the preliminary list of diabetes-associated loci . The most obvious reason could be the fact that more genes exist that are crucial for physiological beta-cell function compared to insulin sensitivity. However, aspects relating to the study design such as selection of relatively young and/or lean cases or lack of precision in the assessment of insulin resistance, by using HOMA-IR instead of dynamic tests such as euglycaemic–hyperinsulinaemic clamps or frequent-sampling glucose tolerance tests, could also have contributed to the finding. Finally, the interaction of diabetes risk variants and unfavourable environmental and lifestyle factors may be more pronounced for insulin resistance than for beta-cell dysfunction, so that GWAS that do not take these external factors into account are less likely to identify gene variants that predispose to insulin resistance.
One interpretation of the current data could be that a genetic predisposition to beta-cell dysfunction represents one basis of the disease by determining the overall susceptibility of an individual to type 2 diabetes. Exogenous factors that induce insulin resistance are then needed for the manifestation of type 2 diabetes, because genetically challenged beta cells may show less flexibility to respond to increasing insulin demands that result in hyperglycaemia . In this model, beta-cell dysfunction is necessary, but not sufficient for the progression towards type 2 diabetes. In other words, the development of type 2 diabetes depends on the presence of further – genetic or environmental – risk factors in addition to a genetic predisposition to progressive beta-cell failure. In principle, this hypothesis can be tested in epidemiological studies, but this would require a more detailed assessment of lifestyle, environmental and metabolic risk factors on the one hand and of insulin sensitivity and beta-cell function on the other hand than in present GWAS that strive to maximise the number of cases and controls at the expense of detailed phenotyping.
When discussing the paucity of gene variants that are associated with insulin resistance, a comparison between GWAS results for type 2 diabetes and obesity seems worthwhile, because obesity represents the main determinant of insulin resistance. Recent data from the Finnish Twin Cohort Study demonstrate that the heritability estimates for type 2 diabetes and body mass index (BMI) are rather similar and that 16% and 21% of the covariance of type 2 diabetes and BMI in men and women, respectively, could be attributed to shared genetic influences . This rather modest genetic similarity is in line with the fact that only SNPs in the FTO gene seem to exert their effect on type 2 diabetes risk by increasing BMI , whereas obesity fails to explain the associations between most other GWAS-derived hits and type 2 diabetes [18,20,22,23,49]. Just as putative beta-cell genes are predominant among the current risk variants for type 2 diabetes, putative obesity genes are mainly expressed in the central nervous system . In addition to FTO SNPs, there is some evidence for a BMI-mediated association of SNPs in the obesity loci TMEM18, GNPDA2, NEGR1, SFRS10-ETV5-DGKG, BCDIN3D-FAIM2 and NCR3-AIF1-BAT2 with type 2 diabetes, but these associations did not show genome-wide significance in the most recent meta-analyses [50–52]. If one assumes that obesity explains only a fraction of diabetes risk and that some SNPs may predispose to a ‘healthy obese’ phenotype, then effect sizes in the associations between obesity SNPs and diabetes should be smaller than for the association with obesity, which may explain these largely negative findings.
Whereas a certain overlap between genes that contribute to the susceptibility for obesity and type 2 diabetes could be expected, the comparison of GWAS for multiple diseases and traits based on many different study populations led to some intriguing findings. Table 2 gives examples of loci that were associated with type 2 diabetes and at least one further trait or disease in different GWAS [29,32,53–69]. These traits and diseases include not only such diverse phenotypes like height, lipids, gallstones, cardiovascular diseases, Crohn’s disease and psoriasis, but also a surprising accumulation of neoplastic disorders (see Ref.  for more examples). This pleiotropy is interesting in several ways. First, this knowledge may help to develop new ways of thinking about type 2 diabetes and identify novel pathophysiological pathways that would not have been revealed by candidate-driven approaches. For example, the considerable proportion of genes that has been implicated in the risk of type 2 diabetes and that encode transcription factors involved in cell-cycle regulation and cell proliferation indicates an aetiologic link between inherited beta-cell dysfunction and increased risk of neoplastic disorders that merits further exploration. This putative link between diabetes risk and cancer will have to be integrated in the context of insulin and insulin growth factor (IGF) receptor–associated signalling pathways that are closely linked with mitogen-activated protein kinase and other growth-related pathways that are essential for cell proliferation . Second, pleiotropic associations can pave the way towards a better understanding of general biological phenomena like the coordinated regulation of glucose metabolism and circadian rhythm that was suggested by the association of the gene MTNR1B (encoding the melatonin receptor 1B) with risk of type 2 diabetes [24,26]. Third, overlaps in disease susceptibilities such as type 2 diabetes and cancer may point at an early stage towards potential adverse effects of newly developed antidiabetic drugs for example on cell proliferation and survival. This issue appears most obvious with HNF1B/TCF2, IGF2BP2 or TCF7L2, which harbour gene variants that are protective for type 2 diabetes, but associated with increased risk of prostate cancer [64,71]. It has been hypothesised that decreased cell proliferation and/or regulation may contribute to low beta-cell mass and therefore to increased risk of type 2 diabetes, but may protect against neoplastic transformation and uncontrolled cell proliferation in other tissues such as the prostate, and vice versa . Thus, the GWAS-derived data on pleiotropic associations support epidemiological data on the increased risk of several types of cancer among patients with type 2 diabetes .
|Gene locus||GWAS signal for trait/disease||References|
|Cholesterol, LDL cholesterol||[57,58]|
|Familial atrial fibrillation|||
|CDKN2A/CDKN2B||Coronary artery disease|||
|KLF14||Basal cell carcinoma|||
In conclusion, the wealth of recent GWAS underlined the importance of beta-cell dysfunction in the heritability of type 2 diabetes, although a possible predominance of inherited insulin secretion defects vs. insulin resistance remains uncertain. The pleiotropy in the associations of loci with both type 2 diabetes and seemingly unrelated conditions might provide novel insights into the aetiology of type 2 diabetes and help to define signals of complications of type 2 diabetes and of potential adverse effects of drugs to be developed to target these genes.
What is the evidence for the use of genetic data for diabetes risk prediction?
This question aims to address the relevance of genetic studies for the prediction of type 2 diabetes risk in the general population. As the heritability of type 2 diabetes is substantial, it had been expected that the identification of type 2 diabetes–related genes could improve risk prediction above and beyond traditional risk factors.
As summarised in Table 3, numerous studies have compared the accuracy of risk prediction using genetic scores or scores based on traditional clinical risk factors or a combination of both approaches [74–85]. Most studies used genotype data for 10–20 diabetes-related SNPs and demonstrated that genetic risk scores alone performed very poorly in predicting type 2 diabetes [area under the receiver-operating characteristic curve (AUC) between 0·55 and 0·63]. Although the accuracy of risk models based on various combinations of traditional clinical risk factors could be improved by addition of genetic data, differences in accuracy were small (changes in AUC ≤ 0·02). Figure 1 illustrates data from the Rotterdam Study, the GoDARTS Study and a Danish sample, all of which found only minor improvements of the accuracy of a very simple prediction model based on age, sex and BMI when genetic data were added [74,75,79]. Thus, genetic data currently provide statistically significant, but not to clinically relevant, contributions to risk prediction for type 2 diabetes.
|Study||Genetic risk score||Clinical risk score||Combination of both scores||References|
|Number of loci||AUC||Variables||AUC||AUC||P*|
|Rotterdam Study||18||0·60||Age, sex, BMI||0·66||0·68||< 10−4|||
|GoDARTS Study||18||0·60||Age, sex, BMI||0·78||0·80||3 × 10−12|||
|Malmö Preventive Project||16||0·63||Age, sex, BMI, fasting glucose, blood pressure, triglycerides, FHD||0·74||0·75||1 × 10−4|||
|Framingham Offspring Study||18||0·581†||Age, sex, BMI, fasting glucose, systolic blood pressure, HDL cholesterol, triglycerides, FHD||0·900||0·901||0·49|||
|CoLaus Study||15||0·59||Age, sex, waist-to-hip ratio, physical activity, triglyceride/HDL cholesterol ratio, FHD||0·86||0·87||0·002|||
|(Danish sample)||19||0·60||Age, sex, BMI||0·92||0·93||n/a|||
|EPIC-Potsdam Study||20||n/a||Diabetes Risk Score (Age, height, waist circumference, hypertension, dietary factors, alcohol, physical activity, smoking), fasting glucose, HbA1c||0·8926||0·8928||0·74|||
|Health Professionals Follow-up Study, Nurses’ Health Study||9‡||n/a||Age, sex, BMI, smoking, alcohol, physical activity, FHD||0·78||0·79||< 0·001|||
|Whitehall II Study||20||0·55||Age, sex, BMI, smoking status, drug treatment, FHD (type 2)||0·72||0·73||0·03|||
|Whitehall II Study||20||0·55||Age, sex, BMI, fasting glucose, HDL cholesterol, triglycerides, parental history of type 2 diabetes||0·78||0·78||0·10|||
|METSIM Study||20||0·552||FINDRISC (age, BMI, waist circumference, physical activity, dietary factors, hypertension, hyperglycaemia, FHD)||0·727||0·730||n/a|||
|METSIM Study||20||0·552||FINDRISC, triglycerides, HDL cholesterol, adiponectin, alanine transferase||0·772||0·772||n/a|||
|Study on Nutrition and Health of Ageing Population in China||17||0·62||Age, sex, region, BMI, FHD, smoking, alcohol use, physical activity, HDL cholesterol, triglycerides||0·77||0·79||< 0·001|||
|(Swedish study population)||73§||0·626||Not done||–||–||–|||
It has been estimated that the gene loci that are associated with type 2 diabetes explain only approximately 10% of its heritability . This raises the question where the remaining heritability is hidden and whether the identification of additional diabetes-related gene loci can be expected to lead to prediction models with clinical relevance. Statistical simulations indicated that the accuracy of prediction models can be improved (AUC 0·8–0·9) if more common gene loci (risk allele frequencies of 5%) are combined. However, the odds ratios for the association with type 2 diabetes need to be in the range of 1·25–1·5, whereas the combination of more than 400 common SNPs (risk allele frequencies of 30%) with odds ratios of 1·1 would not be sufficient to achieve an AUC of 0·8 .
As the genetic architecture of type 2 diabetes (polygenic, common risk variants, low effect sizes) closely resembles that of many other phenotypes and complex diseases, studies on missing heritability for these traits are of interest. Height has a heritability of approximately 80%, but only 5% can currently be explained by about 50 common SNPs that were identified in several GWAS . However, a recent study demonstrated that the combined analysis of SNPs with and without significant association with height (a total number of 294 831 SNPs with minor allele frequencies of ≥ 1% genotyped in 3925 unrelated individuals) extended the explained proportion of variance in height to 45% . These data suggest that most of the ‘missing’ heritability may be hidden in a large number of fairly common SNPs with small effect sizes and that the identification of additional risk variants in even larger GWAS could help to improve prediction models. In addition, exon sequencing and deep resequencing of whole genomes can be expected to increase our knowledge of genetic variation by identifying many rare variants (allele frequencies < 1%) that are associated with intermediate phenotypes and disease outcomes .
In addition to SNPs, structural variations, like copy number variations (CNVs), which are defined as chromosomal segments with varying numbers across individuals or copy neutral variations such as inversions and translocations, could contribute to the risk of type 2 diabetes. It is known that causal CNVs exist for many diseases [89,90]. Large CNVs are more frequently found in obese than in lean individuals and may represent another level of genetic variation that is causally involved in disease risk in addition to SNPs that were identified in GWAS . First analyses within the DIAbetes Genetics Replication And Meta-analysis (DIAGRAM) study yielded no evidence that common CNVs contribute substantially to the risk of type 2 diabetes , but additional studies will be needed to address this issue more in depth. Furthermore, it remains to be tested how helpful existing biobanks will be for the analysis of CNVs and disease risk, because CNV analyses based on blood samples only have limitations. A recent study reported differences in the distribution of CNVs between tissues throughout the body so that CNVs characterised in blood samples may not reflect the genetic variation in more relevant tissues such as beta cells, myocytes or adipocytes .
Epigenetic variations add another level of complexity to the question of where the remaining heritability is hidden. DNA methylation and other epigenetic modifications may affect multiple phenotypes by heritable alterations of gene transcription and thereby contribute to the risk of type 2 diabetes. Although genome-wide analyses of the combination of SNPs, methylation, transcription and metabolic phenotypes have not yet been performed, a study from Iceland provided preliminary evidence that the effects of susceptibility variants for type 2 diabetes and other diseases may depend on whether they are paternally or maternally inherited and that imprinting mechanisms are involved in this regulation . However, epigenetic variations may also be tissue-specific and change over time so that the accuracy of their measurement based on a single blood sample may be lower than for SNPs .
In conclusion, the proportion of heritability of type 2 diabetes that we can explain with currently available data from GWAS based on common SNPs is small, but might increase once further types of genetic variations like CNVs or rare variants have been analysed. For the time being, it needs to be stated that the incremental value of genetic data over and above anthropometric and clinical risk factors of type 2 diabetes (Table 3, Fig. 1) is so negligible that genetic tests to assess the individual risk of this disease are of no relevance to clinicians and patients likewise.
Can genetic data help to create a personalised view of the ‘type 2 diabetes syndrome’?
This question hints at a yet to be explored future where genetic information could help to provide the basis for a concept of ‘personalised diabetology’. Such concept would provide a tool for individualised prevention and treatment options and thereby replace the current view of treating all patients with type 2 diabetes as if they suffered from one homogeneous disease. Although currently type 2 diabetes–related gene variants do not improve the prediction of the disease, one has to keep in mind that the data were obtained from heterogeneous study populations and that type 2 diabetes is already recognised as a heterogeneous disorder with regard to phenotyping. It has not been tested yet in great detail whether there are subgroups in the population for whom the association between the presence of certain gene variants and risk of type 2 diabetes is more pronounced than for the rest of the population. In addition, it is not clear yet whether carriers of specific risk variants will respond better or worse to lifestyle interventions or drug treatments than carriers of protective variants. The following section summarises the current knowledge on the interactions of the TCF7L2 gene, which has been investigated in most detail so far, with other diabetes risk factors, nonpharmacological and pharmacological prevention and treatment of type 2 diabetes.
Although it seems that cross-sectional associations between genetic risk variants with type 2 diabetes are more pronounced in nonobese that in obese individuals, this has been shown convincingly only for the TCF7L2 SNP rs7903146 in a recent large DIAGRAM study . Data from the Nurses’ Health Study indicated that the risk association of the T allele of the TCF7L2 SNP rs12255372 (in tight linkage disequilibrium with rs7903146) may also depend on dietary factors because it was stronger in individuals with high glycaemic load or glycaemic index . There is some evidence for an interaction of TCF7L2 gene variants with response to lifestyle changes, because both the Diabetes Prevention Program and the Finnish Diabetes Prevention Study demonstrated that the association of TCF7L2 risk genotypes with progression from impaired glucose tolerance to type 2 diabetes was present in the control groups, but not in the intensive lifestyle intervention groups [96,97]. These data suggest that lifestyle changes can effectively lower the TCF7L2-related genetic risk of type 2 diabetes. However, several smaller studies failed to provide evidence that carriers of certain TCF7L2 risk variants responded more favourably to lifestyle changes with regard to changes in fasting glucose, insulin resistance or weight or found even opposite effects [98–100]. With respect to pharmacological therapy, interactions with treatment are conceivable because TCF7L2 variants affect beta-cell function and insulin secretion rather than mechanisms of insulin resistance . In line with this expectation, the GoDARTS Study reported an increased odds in homozygous carriers of the TCF7L2 risk allele rs12255372T for failure to respond to sulfonylureas, whereas the response to metformin did not depend on genotype . However, these data need to be replicated and extended to other diabetes-related loci.
In addition to TCF7L2, there is preliminary evidence that variants in other loci may also interact with preventive and therapeutic measures. These loci include both gene loci for which an association with type 2 diabetes with genome-wide significance has been shown (e.g. PPARG and KCNQ11 [103,104]) and genes that represent interesting candidates based on the pathophysiology of type 2 diabetes. Examples for the second class of genes are SLC2A2, ABCC8, SIRT1 and NDUFB6. SLC2A2 and ABCC8 encode glucose transporter-2 (GLUT2) and sulfonylurea-1 receptor (SUR1), respectively, which are both expressed in beta cells and play a role in insulin secretion. Variants in these genes were found to interact with changes in moderate-to-vigourous physical activity in the risk of progression to type 2 diabetes in the Finnish Diabetes Prevention Study . SIRT1 codes for sirtuin 1, while NDUFB6 codes for a component of complex 1 of the mitochondrial respiratory chain. These proteins are involved in cellular energy metabolism, and variants in both genes were associated with resistance to lifestyle intervention in high-risk individuals with a family history of type 2 diabetes [106,107].
Taken together, it seems premature to conclude that genotyping data can be used for individual treatment recommendations. However, large-scale cohort studies or small-scale studies in specifically treated subgroups should provide more knowledge about additional risk variants and help to answer the question whether genotyping can be used to better predict individual responsiveness to preventive or therapeutic interventions.
Conclusions and outlook
Recent GWAS and the derived meta-analyses demonstrate that the genetic architecture of type 2 diabetes is determined by common risk variants in multiple gene loci with low effect sizes. However, the loci that were identified so far explain only 10% of the estimated heritability of type 2 diabetes. Ongoing and future studies will have to address the issue of hidden heritability by (i) even larger sample sizes to identify additional common variants, (ii) using exon sequencing and deep resequencing to find rare susceptibility variants, (iii) characterising CNVs and epigenetic variation as further determinants of gene expression and (iv) investigating intermediate traits including more accurate estimates of insulin resistance.
Fine mapping and molecular as well as cellular studies will be required to reveal the causal gene variants and to test whether some of these candidates may be targets in the prevention and treatment of type 2 diabetes. Additional information on the pathophysiological role of gene variants will most likely be gained from the combination of different ‘-omics’ technologies, i.e. the combined analysis of data from genomics, transcriptomics, proteomics and metabolomics in relation with disease phenotypes .
The current knowledge does not suffice to support the use of genetic data for the prediction of diabetes risk or for decisions regarding specific prevention or treatment measures. However, a more complete understanding of the genetic contribution to type 2 diabetes on the one hand and lifestyle as well as environmental factors on the other hand will lead to a reassessment of the clinical relevance of genotype data. Although it is not known whether additional common variants with low effect sizes will enable a ‘personalised diabetology’, it remains to be seen whether rare variants with higher effect sizes can be identified and used to predict individual diabetes risk and individual responses to prevention and treatment or at least to break down type 2 diabetes into smaller, clinically meaningful subtypes [109,110].
CH and MR are supported by the Federal Ministry of Health (Berlin, Germany) and the Ministry of Innovation, Science, Research and Technology of the state North-Rhine Westphalia (Düsseldorf, Germany). This study was supported in part by a grant from the German Federal Ministry of Education and Research (BMBF) to the German Center for Diabetes Research (DZD e.V.). The authors thank Dr Harald Grallert (Helmholtz Zentrum München) for contributions to Table 1.
Conflict of interest
The authors have no conflicts of interest to disclose.
Study concept and design: CH, MR; acquisition of data: CH; analysis and interpretation of data: CH, MR; drafting of the manuscript: CH; critical revision of the manuscript for important intellectual content: CH, MR.
Institute for Clinical Diabetology, German Diabetes Center, Leibniz Center for Diabetes Research at Heinrich Heine University Düsseldorf, Düsseldorf, Germany (C. Herder, M. Roden); Department of Metabolic Diseases, Heinrich Heine University Düsseldorf, Düsseldorf, Germany (M. Roden).