Genetics of type 2 diabetes mellitus and obesity—a review

Abstract

Type 2 diabetes (T2D) and obesity are recognized as conditions of growing biomedical importance to societies worldwide. Despite this, lack of understanding concerning the processes which normally serve to maintain weight and to regulate glucose concentrations, and ignorance about the mechanisms by which these homeostatic processes fail, remains a significant obstacle to the development of improved tools for management and prevention. There has been a long‐standing belief that the identification of the specific genes influencing development of these conditions has the potential to reveal these fundamental processes, thereby providing vital clues to support clinical advances. Furthermore, there has been the hope that this information will translate into the capacity to deliver more ‘personalized’ medical care, whereby management can be tailored in accordance with an appreciation of individual molecular pathogenesis. As this review indicates, these developments are already a reality for selected monogenic forms of diabetes and obesity. Recent advances in the identification of genes underlying multifactorial forms of these conditions will accelerate efforts to effect similar clinical translation across the full spectrum of disease.

• Diabetes mellitus type 2
• genetics
• obesity

Introduction

Type 2 diabetes (T2D) and obesity represent major causes of ill‐health in societies worldwide 1. Specification of the genetic basis of these diseases provides a powerful tool for improved understanding of pathogenesis and opportunities for clinical translation. Monogenic forms of diabetes and obesity provide clear examples of the value of this approach.

Key messages

• Causal genes are known for many monogenic forms of diabetes and obesity, and this information is increasingly used in the clinical setting.

• In contrast to monogenic forms of obesity and diabetes, identification of genes influencing susceptibility to multifactorial forms of these conditions has been proceeding slowly prior to the recent availability of genome‐wide association data.

• New biological insights from global studies of genome‐wide variation and their relationship to disease should translate into clinical advances.

Monogenic forms of diabetes and obesity

Although monogenic forms of diabetes and obesity account for only a small minority of cases worldwide, study of the genetic basis of these conditions has not only generated insight into the specific diseases themselves, but has also provided important clues about the biology of energy balance and metabolic homeostasis that has much broader relevance. The strong correlation between genotype and phenotype which characterizes monogenic and syndromic forms of ‘type‐2‐like’ diabetes has facilitated gene identification for these conditions and resulted in a fairly comprehensive understanding of their molecular aetiology.

The classical phenotype in monogenic diabetes is maturity onset diabetes of the young (MODY). As a result of successes in gene identification, the original clinical definition (based upon a non‐autoimmune loss of beta‐cell function, autosomal dominant inheritance, and early age of diagnosis) has been largely replaced by a molecular classification 2.

This molecular classification has clear prognostic and therapeutic implications that increasingly see molecular diagnostic tools deployed as part of routine clinical care. Individuals with mutations in the glucokinase (GCK) gene 3, 4 account for approximately 14% of cases of MODY, but this form of the disease is characterized by mild, stable fasting hyperglycaemia, with little deterioration of beta‐cell function over time. The majority of cases can be treated with dietary measures alone, and complications are few. A wide spectrum of mutations within GCK gene have been described 5.

Monogenic diabetes resulting from mutations in beta‐cell transcription factors (most commonly in TCF1, encoding HNF‐1α) 6 is, in contrast, more progressive with age, with substantial risk of diabetes complications if treatment is not optimized. Those with TCF1 mutations are particularly sensitive to the hypoglycaemic effects of sulphonylureas 7 and, in contrast to most patients with typical T2D, these, rather than metformin, represent optimal treatment. Obtaining a definitive diagnosis through molecular testing can thus lead to more rational treatment selection.

Even amongst the various subtypes of transcription factor MODY (which can result from mutations in the genes encoding HNF‐4α and HNF‐1β as well as HNF‐1α), there are subtle phenotypic differences. Although HNF‐4α and HNF‐1α are closely related functionally 8 and both lead to diabetes as a result of progressive beta‐cell dysfunction, only mutations in HNF4A present with fetal and neonatal hyperinsulinaemia resulting in increased birth‐weight and hypoglycaemia 9. Similarly, mutations in TCF2 (encoding HNF‐1β) are associated with a spectrum of abnormalities in the renal and urogenital systems: these, rather than the diabetes, tend to dominate presentation. In families segregating TCF2, pancreatic atrophy (rather than beta‐cell dysfunction) is characteristic and many affected individuals are also insulin‐resistant 10, 11. Once again, the capacity to allocate a specific molecular diagnosis has clear prognostic value for the index case and their families.

Other causes of MODY are extremely rare. Insulin promoter factor 1 (IPF‐1) is a homeodomain transcription factor required for pancreas development and for the transcriptional regulation of genes in the pancreatic beta‐cells: mutations in the PDX1 gene encoding IPF‐1 underlie a rare form of the condition 12. Mutations in NEUROD113 and CEL14 have also been implicated in a small number of families segregating early‐onset non‐autoimmune monogenic diabetes. Around 10% of families segregating MODY have, as yet, no defined genetic cause 15.

The most exciting developments in the field of monogenic diabetes in recent years have been in the area of neonatal diabetes (NDM). Until recently, classification of this condition (which is defined as diabetes with onset in the first 6 months of life) was based upon the clinical course of disease: it was either permanent or transient, the latter characterized by early remission, but frequent relapse in adulthood. It had been known for some time that many of these transient cases are attributable to abnormalities of an imprinted locus on chromosome 6q24 (involving the ZAC and HYMA1 genes) 16, 17.

As a result of recent work, it is now clear that most cases of permanent NDM and many of the remaining cases of transient NDM are the result of activating mutations in the KCNJ11 and ABCC8 genes, each of which encodes one of the two subunits of the adenosine triphosphate (ATP)‐dependent potassium channel (K_ATP) 18–21. By and large, the extent of the clinical phenotype mirrors the degree to which the mutation concerned impacts on channel function (as assessed in vitro). The most serious mutations in KCNJ11 lead to the clinically most severe syndromic form, DEND syndrome (developmental delay, epilepsy, and neonatal diabetes) 20 in which diabetes is accompanied by neurological features thought to reflect the extrapancreatic expression of the gene product.

In most cases, these mutations disrupt the capacity of the ATP signal generated by intracellular glucose metabolism to effect channel closure, depolarization of the plasma membrane, and insulin release 22. Since sulphonylureas lead to membrane depolarization through an ATP‐independent mechanism, it has been possible to transfer many NDM patients off insulin therapy (not surprisingly, most of these individuals have been incorrectly labelled with type 1 diabetes) onto sulphonylureas, a therapeutic change which has generally resulted in improved glycaemic control and amelioration of some of the neurological sequelae 23–25.

Positional cloning efforts have successfully defined the genes underlying a wide range of syndromes which feature diabetes as one component (see 26 for a review) but most of these are vanishingly rare. The one exception is the syndrome of maternally inherited diabetes mellitus and deafness (MIDD) which may account for as much as 3% of diagnosed diabetes. This condition results from a mutation in the mitochondrial genome (hence the exclusively maternal pattern of inheritance) affecting the tRNA^Leu(UUR) gene (at position 3243). The consequence is a condition characterized by quite aggressive loss of beta‐cell function, such that most patients require insulin 27.

Turning to the regulation of fat mass, it has been estimated that monogenic forms account for as many as 7% of children with extremely severe, young‐onset obesity 28: however, as such phenotypes are restricted to <0.01% of the population, these mutations are rare in the general population 29. By far the most common form of monogenic obesity is that due to mutations in the gene encoding the melanocortin 4 receptor (MC4R) 30, 31. Depending on the sample, these account for somewhere between 1% and 6% of early‐onset or severe adult obesity 32. MC4R mutations act in a co‐dominant manner with heterozygotes displaying severe obesity, hyperphagia, increased lean mass, increased linear growth, and severe insulin resistance. The homozygote state is associated with a more severe phenotype. MC4R, expressed in the hypothalamus, acts as the receptor for α‐MSH (melanocyte‐stimulating hormone), a potent anorexigenic agent derived from the processing of the product of the pro‐opiomelanocortin (POMC) gene 33. Of note, the phenotypic effect of MC4R mutations is most marked in childhood and becomes less distinct in adulthood 30, increasingly overlapping with the phenotype of more common multifactorial obesity.

Other causes of monogenic obesity are far less common, but each discovery has provided an important insight into the mechanisms responsible for normal (and abnormal) weight regulation. Mutations in the POMC gene, for example, cause loss of all POMC‐dependent processes, resulting in adrenal insufficiency and loss of pigmentation, as well as severe obesity and hyperphagia 34. A related cause of early severe obesity is a mutation in the PCSK1 gene: this gene encodes a prohormone convertase responsible for the normal processing of a range of peptide hormones including POMC 35.

The most dramatic discoveries (because of their therapeutic implications) have been seen in rare instances of mutations in the genes encoding leptin (and its receptor) 36, 37. The former is the only genetic form of obesity amenable to replacement therapy, with exogenous leptin, given by injection, leading to sustained weight loss, reduced hunger, and restoration of normal endocrine function.

Finally, a wide range of diverse conditions, including the Prader‐Willi and Bardet‐Biedl syndromes, feature obesity as one characteristic component, and the genes responsible for many of these have been defined. These provide further experiments of nature that can be expected to shed light on the mechanisms of weight regulation.

Multifactorial forms of T2D and obesity

In contrast to the steady progress in the identification of genes involved in monogenic and syndromic forms of T2D and obesity, equivalent success in the domain of multifactorial disease has, until recently at least, been far slower 26.

The reasons behind this disparity in the pace of discovery are obvious. In the vast majority of individuals, in whom diabetes or obesity is not the result of a single gene defect, individual susceptibility depends on the distribution of variants inherited at many sites and on the cumulative effect of the various environments to which the individual has been exposed. Under such circumstances, the magnitude of effect attributable to any single variant will be modest, reducing the power for detection, and making replication difficult. Until recently, most studies in the field have failed to make due allowance for this biological complexity, and the consequences for appropriate study design and the interpretation of results 38. There has been a tendency to overestimate the potential importance of each new candidate gene chosen for consideration, to undertake studies that examine only a small proportion of the genome sequence variation within each gene, to test associations in small sample sizes, and then to make claims of association (and biological significance) without due allowance for multiple testing. The ramifications of such failures of study design are all too clear: true associations will be very hard to find (since studies are underpowered to detect effects of realistic magnitude) whilst the promulgation of spurious associations will be encouraged 39.

Combine all this with an element of publication bias (such that studies reporting apparent associations are more likely to reach the literature), and it is easy to appreciate why most published claims of susceptibility effects are undoubtedly false. For example, the obesity gene map reports many hundreds of linkages and associations for obesity 40, yet few of these have been subjected to the acid test of independent replication, and most will be spurious.

The good news is that many of these obstacles to reliable inference have been overcome through the advent of large‐scale genome‐wide association (GWA) studies. As we report below, the consequence has been a substantial increase in the number of genuine, independently replicated genes implicated in multifactorial diabetes and obesity.

But first, let us consider some of the successes of the pre‐GWA era.

Multifactorial genetics before genome‐wide association studies

Over the past decade, researchers seeking to identify human complex‐trait susceptibility loci have been restricted to two main approaches: genome‐wide linkage analysis and/or candidate‐gene‐based association studies.

Linkage analysis seeks evidence of co‐segregation between anonymous genomic markers and the phenotype of interest. In principle, linkage studies allow the entire genome to be screened for susceptibility loci using a limited number of polymorphic markers (a few hundred microsatellites, or a few thousand single‐nucleotide polymorphisms). The massive disadvantage of the linkage approach, for complex traits at least, is one of power. Even for variants with substantial (but not extreme) effects on susceptibility at the population level, the numbers of families needed to offer reasonable power to detect linkage is prohibitive 41.

Viewed in this light, it is little surprise that efforts to use linkage in T2D and obesity have failed to generate a clear view of genomic patterns of susceptibility. Loci reported in individual studies have usually proved difficult to replicate in others: chance, differences in the types and ethnic origins of families studied and the analytical methods deployed have also played their part. In T2D, regions of chromosomes 1q, 10q, and 20 have shown some evidence of replication, prompting focused efforts at positional cloning therein 42. These linkage findings played a part in the identification of T2D‐associated variants in TCF7L2 (see below) 43. Similarly, evidence of linkage of T2D to chromosome 2q in Mexican Americans led to identification of susceptibility variants within calpain‐10 44, though the effect of these variants on T2D susceptibility in other populations remains controversial 45.

Loci on 2p, 3q, 7q, 10p, 11q, and 20q are the best candidate regions to emerge from linkage studies for obesity 46. As yet, there is no conclusive evidence for an aetiological locus in any of these, though adiponectin may be contributing to the 3q signal 46, 47.

Given the low power of linkage approaches, there are good reasons to favour the application of methods based upon the use of association in the search for disease susceptibility loci 48. However, association mapping is reliant on linkage disequilibrium which, in human populations, typically extends only a few tens of kilobases. Until recently, this has restricted the scope of association‐mapping efforts to the assessment of a limited number of variants selected through a candidacy‐based approach.

For both T2D and obesity, many hundreds such candidates have been selected for study, though very few genes have been tackled comprehensively or in data sets large enough to provide convincing inference. Many apparent associations have been reported, but very few have been replicated.

Nonetheless, the candidate gene approach can claim two major successes in T2D. Common non‐synonymous coding variants in both the PPARG and KCNJ11 genes (P12A and E23K, respectively) have shown consistent evidence for association across many studies, which collectively include many tens of thousands of subjects 49, 50. In each case, the effect size is modest (a per‐allele odds ratio of about 1.2), but the high prevalence of the disease‐associated variant (in European populations, 85% and 40%, respectively) translates into appreciable effects at the population level (on some measures of impact at least).

Several important lessons can be gleaned from these two examples. First, genuine diabetes susceptibility genes can be found, but their effects are modest, and convincing evidence is only forthcoming from studies involving several thousands of individuals. Second, these susceptibility genes are acting through different pathways (adipocyte differentiation and function; glucose‐insulin secretion coupling in the pancreatic beta‐cell, respectively), confirming the heterogeneous nature of diabetes susceptibility. Third, the products of both genes constitute targets for established therapeutic agents (thiazolidinediones and sulphonylureas, respectively) demonstrating the equivalence of genetic and therapeutic perturbations of function with respect to underlying traits. Fourth, rare variants in both genes have been implicated in monogenic syndromes (of insulin resistance, and neonatal diabetes, respectively), illustrating the potential for a spectrum of phenotypes associated with varying degrees of functional disturbance 19, 51.

Whilst these have emerged as the clearest examples of proven susceptibility genes, several others deserve brief mention as ‘contenders’. For each, the evidence in favour of a genuine susceptibility effect is less convincing, usually because compelling evidence of repeated, independent replication is not yet available. Examples include variants in the genes encoding PPARgamma co‐activator 1alpha (PPARGC1A) and leucyl‐tRNA synthetase 2, mitochondrial (LARS2) 52, 53. Also, of particular interest (given the findings at PPARG and KCNJ11) are common variants within genes already implicated in causation of monogenic forms of diabetes. There is good evidence that common variants within the genes encoding hepatocyte nuclear‐factor 4‐alpha (HNF4A), insulin (INS), and lamin A/C (LMNA) have effects on susceptibility to multifactorial T2D 54–56.

The story in obesity is similar, though with fewer successes than in T2D. The single ‘proven’ effect involves mutations within the MC4R gene (encoding the target of α‐MSH) which account for up to 5% of severe obesity 57. As described earlier, in children these mutations generally present as severe monogenic obesity 30. As carriers move into adulthood, the phenotype becomes less dramatic and the clinical overlap with multifactorial obesity greater.

As with diabetes, the candidate gene approach in obesity has been relatively unproductive with many genes studied but few, if any, well replicated findings. Recent studies have raised interesting questions about the role of variants in the GAD2 and ENPP1 genes 58–60, both positional candidates lying within linked regions. However, initial reports that GAD2 variants predispose to extreme obesity 58, 59 have not been sustained in subsequent publications 61, 62.

Multifactorial genetics in the genome‐wide association era

The limited gains from linkage and candidate‐based association studies had led to growing concerns that the benefits of genetics research had been overstated. However, the recent explosion in the identification of susceptibility genes through GWA approaches makes it clear that the major problem lay with the limited scale of much previous research.

Three developments have contributed to the capacity to undertake large‐scale GWA studies. First, the efforts of the International Hapmap Consortium 63 and related endeavours enumerated the patterns of common variation in human populations, and showed that it was possible to capture most common variation genome‐wide with as few as 300,000 carefully selected single nucleotide polymorphisms (SNPs) 64. Second, advances in technology made possible high‐performance genotyping using array‐ or bead‐based platforms capable of generating up to a million genotypes per assay 65. Third, researchers recognizing that most susceptibility variants would have modest effects had started to amass ‘fit‐for‐purpose’ study samples of adequate size, and to establish collaborations that allow rapid replication and meta‐analysis.

Even before the current frenzy of GWA activity, this kind of biology‐agnostic approach claimed what may turn out to be its largest single success—the identification of T2D susceptibility variants within the TCF7L2 gene 43. Although the study of anonymous markers in this gene was partly prompted by efforts to locate association signals within the chromosome 10 region of replicated T2D linkage, the variants within TCF7L2 so far identified do not explain that linkage signal. With several GWA scans now complete (see below) it is clear that common variants in TCF7L2 have an unusually large impact on T2D susceptibility 66. In European populations, the 10% of individuals who are homozygous for the high‐risk allele have approximately twice the risk of diabetes as those with no copies. This finding has been widely replicated, including studies of non‐European populations: these have helped to identify a single intronic SNP, rs7903146, as the strongest statistical candidate (though the precise functional mechanism is not yet clear) 67. TCF7L2 is known to operate in the Wnt (wingless)‐signalling pathway, and the bulk of evidence points to an effect on T2D risk mediated through an adverse effect on islet function 68.

As described in a flurry of papers in early 2007, GWA data are now available on over 6700 T2D cases and 11600 controls (all of Northern European origin) examined as part of five different studies 69–73. This effort has turned out to be spectacularly successful, with six novel T2D susceptibility genes (all of them independently replicated) identified to date.

In the first reported GWA scan on T2D, Sladek and colleagues (studying a total of 1363 cases and controls from France), reported strong associations with five loci, one of them TCF7L2, the other four novel 73. The two with the least strong evidence for association (EXT2 and LOC387761) have not so far been replicated, but the remaining two (involving SNPs within the HHEX/IDE region and the SLC30A8 gene) have been seen in other scans and are clearly genuine. Whilst the causal gene in the HHEX/IDE region has yet to be confirmed (both HHEX and IDE are excellent biological candidates, and both contain strongly associated SNPs), the findings with respect to SLC30A8 (which encodes a zinc transporter) point to the importance of zinc in maintaining normal beta‐cell function.

The other four scans (conducted in UK, Icelandic, Finnish, and Swedish subjects) have added four further loci to the tally 69–72. One of these, FTO, exerts its effect on T2D risk through an effect on adiposity (see below), whereas the other three directly impact on T2D susceptibility through as yet unknown mechanisms. Two of the signals, mapping within or near the CDKAL1 and CDKN2A genes, may well be acting through effects on beta‐cell development and regeneration, since there is evidence that the products of both genes inhibit cyclin‐dependent kinases that influence beta‐cell function and regenerative capacity 74, 75. If such mechanistic speculations are confirmed, these genetic discoveries will have contributed to resolution of the on‐going controversy over the role of reduced beta‐cell mass in the pathogenesis of T2D. The sixth and last locus to emerge from these studies maps to the IGF2BP2 gene, the product of which is involved in translation of insulin‐like growth factor (IGF) 2.

Several important inferences can be made from these studies. First, the GWA approach has proved to be extremely rewarding in terms of identifying T2D susceptibility loci and delivering novel biological insights into disease pathogenesis. Second, the effect size of each of the variants uncovered is modest: save for TCF7L2, the per‐allele odds ratios lie between 1.1 and 1.2. Third, detection and replication of such effects requires access to very substantial numbers (tens of thousands) of well characterized DNA samples. Fourth, initial functional studies confirm that individual risk of DNA can be influenced by perturbation in any one of several pathways (including fat mass, adipocyte function, beta‐cell function, beta‐cell mass). Fifth, there are likely to be many more T2D susceptibility genes to be found in the months and years ahead, both through more systematic analysis of existing GWA data, and through the advent of technologies designed to capture sources of genome sequence variation (rare SNPs, structural variants) not accessible to current genotyping reagents.

Compared with the situation in T2D, GWA efforts directed towards variation in BMI and obesity have been slower to emerge. However, several large scans currently in progress will report in the months ahead. Existing GWA efforts have, however, identified two loci of interest. In a low‐density scan initiated in the Framingham cohort, Herbert and colleagues published evidence that variation at the INSIG2 gene may be responsible for a modest increase in the risk of obesity 76. Replication of this finding has been patchy and large‐scale meta‐analysis is going to be required to establish the validity of this association 77.

Evidence for the second gene, FTO, is more robust 29. Variants in this gene were initially revealed through the UK GWA scan for T2D 69, whence it emerged as the second strongest effect after TCF7L2. It became clear that this effect on T2D susceptibility was entirely the result of a marked effect of FTO variants on weight and BMI (a between‐homozygote difference in adult weight of ∼3 kg), all of which reflected fat mass. FTO has also been implicated in the development of early‐onset and severe obesity in adults and children in French, German, and Swiss cohorts 78. Approximately one‐sixth of European populations are homozygous for the high‐risk allele: these individuals have a 70% increase in their odds ratio for obesity compared to the one‐third with no copies. The function of the FTO gene is unknown but has become an issue of considerable research interest, given the valuable clues it may offer to the pathogenesis of this condition.

Table I. Replicated type 2 diabetes susceptibility genes identified to date with genome‐wide data.

Signal	Chromosome	Risk allele frequency in controls^a	Per‐allele odds ratio^a
TCF7L2	10	0.30	1.37
FTO	16	0.40	1.23
CDKN2A/CDKN2B	9	0.86	1.20
PPARG	3	0.84	1.20
IGF2BP2	3	0.32	1.17
KCNJ11	11	0.47	1.15
HHEX/IDE	10	0.54	1.14
CDKAL1	6	0.32	1.12
SLC30A8	8	0.65	1.12

Genes in descending order of per‐allele odds ratio (OR).

^aApproximate values of OR and risk allele frequency are based on currently available data. OR for FTO effect on type 2 diabetes based on case‐control samples not matched for body mass index.

Conclusion

These are exciting times in the field of multifactorial genetics. Now that it is possible to undertake global studies of genome‐wide variation and its relationship to disease, in sample sizes adequate to the unequivocal identification of variants of modest effect, we can expect significant advances in our understanding of these conditions. We can also start to explore how these findings can be translated into improvements in the clinical management of these important conditions.