Recent advances in understanding the genetic architecture of type 2 diabetes

Abstract

Genome-wide association (GWAS) and sequencing studies are providing new insights into the genetic basis of type 2 diabetes (T2D) and the inter-individual variation in glycemic traits, including levels of glucose, insulin, proinsulin and hemoglobin A1c (HbA1c). At the end of 2011, established loci (P < 5 × 10⁻⁸) totaled 55 for T2D and 32 for glycemic traits. Since then, most new loci have been detected by analyzing common [minor allele frequency (MAF)>0.05] variants in increasingly large sample sizes from populations around the world, and in trans-ancestry studies that successfully combine data from diverse populations. Most recently, advances in sequencing have led to the discovery of four loci for T2D or glycemic traits based on low-frequency (0.005 < MAF ≤ 0.05) variants, and additional low-frequency, potentially functional variants have been identified at GWAS loci. Established published loci now total ∼88 for T2D and 83 for one or more glycemic traits, and many additional loci likely remain to be discovered. Future studies will build on these successes by identifying additional loci and by determining the pathogenic effects of the underlying variants and genes.

Introduction

Type 2 diabetes (T2D) is a common disease with substantial and rapidly increasing global impact. While prevalence varies with age, sex and population, the global age-standardized adult diabetes prevalence is >9.2%, and an estimated >347 million adults have diabetes (1). Diabetes can be diagnosed based on the level of blood glucose after fasting or 2 h after an oral glucose challenge (2hGlu), or based on hemoglobin A1c (HbA1c), which provides a 3-month average of blood glucose (2). In many individuals with T2D, insulin resistance coexists with obesity, adverse lipid profiles, high blood pressure and a proinflammatory state, each likely influenced by genetic and environmental factors (3). Progression to T2D is characterized by abnormalities in pancreatic islet β-cell function in the presence of insulin resistance (4), although these biological processes are only partially defined. Strong evidence for a genetic component exists for T2D risk, insulin secretion and insulin action (5,6).

In the past decade, genome-wide association (GWAS) and sequencing studies have identified genetic loci that help explain the inherited basis of T2D and glycemic traits. These studies are providing insights into the genetic architecture of T2D, including the number, frequency and effect sizes of risk variants in populations around the world. The polygenic nature of T2D is now well established, and multiple risk variants are being identified at some loci, suggesting allelic heterogeneity. Concurrently, increasing numbers of genes and variants have been implicated in monogenic forms of diabetes, including maturity onset diabetes of the young (MODY) and neonatal diabetes (7), and at least five genes have been implicated in both monogenic and polygenic diabetes (8). A recent simulation study evaluated genetic architectures for consistency with results from T2D genetic studies and found that many different disease models were still possible with respect to the number of loci, allele frequencies and level of selective pressure (9). Ongoing studies should more substantially narrow the bounds on feasible architectures (9).

In this review, we describe recent genetic findings from association and sequencing studies for T2D and the related glycemic traits of glucose, HbA1c, insulin and proinsulin. We generally use P < 5 × 10⁻⁸ as a threshold to define significant results, noting in places where less stringent, study-wide thresholds are used. We count distinct loci by considering multiple signals within 500 kb to be part of the same locus and consider traits with and without adjustment for body mass index (BMI) to be the same trait. Finally, we refer to loci by the name of a nearby gene, recognizing that these genes are signposts of a locus and not necessarily the genes affected by the underlying variants.

Common Variants

The development of GWAS spurred considerable progress identifying common variants [minor allele frequency (MAF)>0.05] associated with T2D (Table 1) and glycemic traits (Table 2). After early candidate gene and linkage studies identified common variants associated with T2D in PPARG, KCNJ11-ABCC8 and TCF7L2, the first five GWAS for T2D detected six additional loci, and by early 2008, GWAS and meta-analyses had identified 15 loci for T2D and G6PC2 as a locus for fasting glucose (10). Also in 2008, reports of the first non-European-based GWAS for T2D established KCNQ1 as a T2D locus with variants common in East Asians (MAF = 0.33) but low frequency in Europeans (MAF ∼0.01) (11,12). KCNQ1 risk variants showed similar effect sizes in both populations, demonstrating the role of allele frequency in power to detect loci (13). In 2010, a meta-analysis of European-ancestry individuals identified a second signal of T2D-associated variants near KCNQ1 that are not in marked linkage disequilibrium (LD) with the initial variants (r² < 0.05) and independent from them based on conditional analyses (14). By the end of 2011, further GWAS and meta-analyses in several populations had identified 55 loci for T2D (15,16). Also by 2011, GWAS had identified 32 total loci for one or more glycemic traits, including 17 for fasting glucose (15,17), 2 for fasting insulin (18), 5 for 2hGlu (19), 11 for HbA1c (20–22) and 9 for proinsulin, including 1 identified only in women (23). Incomplete overlap of loci between T2D and glycemic traits showed that not all effects on glucose levels in healthy individuals translate to the risk of T2D and vice versa. Based on the overlap between traits and the biological function of nearby genes, most identified T2D loci appeared to have a primary role in pancreatic islet β-cell function, with far fewer impacting insulin resistance.

Table 1.

Eighty-eight established loci for T2D (P < 5 × 10⁻⁸)

Variant	Chr	Nearby gene	Variant	Chr	Nearby gene	Variant	Chr	Nearby gene
rs2296172	1	MACF1	rs9470794	6	ZFAND3	rs10842994	12	KLHDC5
rs17106184	1	FAF1	rs1535500	6	KCNK16	rs1531343	12	HMGA2
rs10923931	1	NOTCH2	rs2191349	7	DGKB	rs7961581	12	TSPAN8, LGR5
rs340874	1	PROX1	rs864745	7	JAZF1	rs11065756	12	CCDC63
rs780094	2	GCKR	rs4607517	7	GCK	rs2074356	12	C12orf51
rs7578597	2	THADA	rs6467136	7	GCC1, PAX4	rs7957197	12	HNF1A
rs243021	2	BCL11A	rs791595	7	MIR129, LEP	rs1727313	12	MPHOSPH9
rs6723108	2	TMEM163	rs972283	7	KLF14	rs9552911	13	SGCG
rs7560163	2	RBM43, RND3	rs515071	8	ANK1	rs61736969	13	TBC1D4
rs7593730	2	RBMS1	rs896854	8	TP53INP1	rs1359790	13	SPRY2
rs3923113	2	GRB14	rs13266634	8	SLC30A8	rs7403531	15	RASGRP1
rs2943641	2	IRS1	rs7041847	9	GLIS3	rs7172432	15	C2CD4A/B
rs1801282	3	PPARG	rs17584499	9	PTPRD	rs7178572	15	HMG20A
rs7612463	3	UBE2E2	rs10811661	9	CDKN2A/2B	rs11634397	15	ZFAND6
rs831571	3	PSMD6	rs13292136	9	TLE4	rs2028299	15	AP3S2
rs4607103	3	ADAMTS9	rs2796441	9	TLE1	rs8042680	15	PRC1
rs11708067	3	ADCY5	rs11787792	9	GPSM1	rs9936385	16	FTO
rs6769511	3	IGF2BP2	rs12779790	10	CAMK1D	rs7202877	16	BCAR1
rs16861329	3	ST6GAL1	rs1802295	10	VPS26A	rs391300	17	SRR
rs6808574	3	LPP	rs12571751	10	ZMIZ1	rs312457	17	SLC16A11/A13
rs6815464	4	MAEA	rs1111875	10	HHEX, IDE	rs11651052	17	HNF1B
rs1801214	4	WFS1	rs7903146	10	TCF7L2	rs8090011	18	LAMA1
rs6813195	4	TMEM154	rs10886471	10	GRK5	rs12970134	18	MC4R
rs702634	5	ARL15	rs2334499	11	DUSP8	rs12454712	18	BCL2
rs459193	5	ANKRD55	rs2237892	11	KCNQ1	rs10401969	19	CILP2
rs4457053	5	ZBED3	rs5215	11	KCNJ11	rs3786897	19	PEPD
rs35658696	5	PAM, PPIP5K2	rs1552224	11	CENTD2	rs8108269	19	GIPR
rs9502570	6	SSR1, RREB1	rs10830963	11	MTNR1B	rs4812829	20	HNF4A
rs10440833	6	CDKAL1	rs11063069	12	CCND2	rs5945326	X	DUSP9
rs3132524	6	POU5F1, TCF19

One representative variant and one or two genes are shown for each locus. Loci are defined as association signals located within 500 kb of each other regardless of LD. The only locus identified first by a low-frequency variant (MAF < 0.05) is shown in bold font.

Chr, chromosome.

Table 2.

Eighty-three established loci for one or more glycemic traits (P < 5 × 10⁻⁸)

Variant	Chr	Nearby gene	Traits	Variant	Chr	Nearby gene	Traits
rs6684514	1	TMEM79	H	rs7034200	9	GLIS3	G
rs2779116	1	SPTA1	H	rs10811661	9	CDKN2A/2B	G
rs340874	1	PROX1	G	rs16913693	9	IKBKAP	G
rs2820436	1	LYPLAL1	I	rs306549	9	DDX31	P
rs1371614	2	DPYSL5	G	rs651007	9	ABO	G
rs780094	2	GCKR	G, I	rs3829109	9	DNLZ	G
rs895636	2	SIX3, SIX2	G	rs16926246	10	HK1	H
rs1530559	2	YSK4	I	rs10885122	10	ADRA2A	G
rs10195252	2	GRB14	I	rs7903146	10	TCF7L2	G, I, P
rs560887	2	G6PC2	G, H	rs7077836	10	TCERG1L	I
rs733331	2	PDK1	G	rs11605924	11	CRY2	G
rs2943645	2	IRS1	I	rs7944584	11	MADD	G, P
rs17036328	3	PPARG	I	rs1483121	11	OR4S1	G
rs11715915	3	AMT	G	rs174550	11	FADS1	G
rs11708067	3	ADCY5	G	rs11603334	11	ARAP1	G, P
rs11920090	3	SLC2A2	G	rs10830963	11	MTNR1B	G, H
rs7651090	3	IGF2BP2	G	rs2657879	12	GLS2	G
rs3822072	4	FAM13A	I	rs150781447	12	TBC1D30	P
rs9884482	4	TET2	I	rs35767	12	IGF1	I
rs6822892	4	PDGFC	I	rs12229654	12	MYL2	G
rs17046216	4	SC4MOL	I	rs2074356	12	C12orf51	G
rs4865796	5	ARL15	I	rs11066453	12	OAS1	G
rs459193	5	ANKRD55	I	rs10747083	12	P2RX2	G
rs7708285	5	ZBED3	G	rs11619319	13	PDX1	G
rs6235	5	PCSK1	G, P	rs576674	13	KL	G
rs1019503	5	ERAP2	G	rs61736969	13	TBC1D4	G, I
rs35658696	5	PAM, PPIP5K2	I	rs7998202	13	ATP11A	H
rs17762454	6	RREB1	G	rs3783347	14	WARS	G
rs9368222	6	CDKAL1	G, H	rs4502156	15	C2CD4A/B	G, P
rs1800562	6	HFE	H	rs1549318	15	LARP6	P
rs6912327	6	UHRF1BP1	I	rs2018860	15	IGF1R	G
rs10305492	6	GLP1R	G	rs1421085	16	FTO	I
rs2745353	6	RSPO3	I	rs9933309	16	CYBA	H
rs9399137	6	HBS1L, MYB	H	rs4790333	17	SGSM2	P
rs2191349	7	TMEM195	G	rs1046896	17	FN3K	H
rs4607517	7	GCK	G, H	rs11667918	19	MYO9B	H
rs6943153	7	GRB10	G	rs731839	19	PEPD	I
rs1167800	7	HIP1	I	rs10423928	19	GIPR	G
rs983309	8	PPP1R3B	G, I	rs6113722	20	FOXA2	G
rs4737009	8	ANK1	H	rs6072275	20	TOP1	G
rs11558471	8	SLC30A8	G, P	rs855791	22	TMPRSS6	H
rs3824420	9	KANK1	G, P

One representative variant, one or two genes and one-letter abbreviations for the glycemic traits analyzed in the discovery association studies are shown for each locus. Not all traits are shown. Loci are defined as association signals located within 500 kb of each other regardless of LD. Loci identified first by a low-frequency variant (approximate MAF <0.05) are shown in bold font.

Chr, chromosome; G, glucose (fasting, 1hGlu, 2hGlu or these traits adjusted for BMI); I, insulin (fasting or fasting and adjusted for BMI) or insulinogenic index; H, hemoglobin A1c; P, proinsulin (fasting and adjusted for BMI and insulin).

Building on this success, an additional round of GWAS meta-analysis made use of the Metabochip, a custom genotyping array designed to examine ∼1000–5000 variants nominally associated with each of 23 cardiometabolic traits/diseases and to fine-map 257 established disease- and trait-associated loci (24). This cost-effective array allowed investigators to genotype many additional samples, increasing the power of meta-analyses. The DIAbetes Genetics Replication and Meta-analysis (DIAGRAM) Consortium meta-analyzed Metabochip data from 22 669 T2D cases and 58 119 controls with earlier GWAS data from 12 171 T2D cases and 56 862 controls and identified 10 additional loci (25). These loci showed further evidence of overlap with other metabolic traits including obesity (e.g. MC4R) and lipid levels (e.g. CILP2). CILP2 was also identified in a contemporaneous gene-based study (26). Network analysis of genes located near new and established T2D signals were enriched for transcription factors, especially those that interact with transcriptional co-activator CREBBP, involved in chromatin remodeling (25). Analysis of directional consistency in allelic effects between stages of the meta-analysis supported a genetic architecture consisting of many common causal variants, most of very modest effect.

The Meta-Analyses of Glucose and Insulin-Related Traits Consortium (MAGIC) also incorporated Metabochip data to identify new loci (27). A meta-analysis of as many as 133 010 non-diabetic individuals identified 41 new loci for glycemic traits: 20 for fasting glucose, 17 for fasting insulin and 4 for 2hGlu. Analyses with and without adjustment for BMI allowed additional loci to be detected, especially for fasting insulin. Of 53 total glycemic loci, 34 showed at least nominal association (false discovery rate q < 0.05) with other metabolic traits, including insulin loci for T2D, lipid traits and waist–hip ratio, consistent with insulin resistance and potentially implicating pleiotropic effects. Of 36 fasting glucose loci, 22 showed at least nominal association with T2D (q < 0.05), but many other glucose loci had no apparent effect on T2D.

Genetic studies performed since 2012 have identified many additional T2D loci based on risk alleles common in one population but less common in others. Studies in African Americans identified RND3-RBM43 (28), HLA-B and INS-IGF2 (29). Studies in South Asians identified TMEM163 (30) and SGCG (31). One locus, SLC16A11-SLC16A13, was simultaneously identified in Japanese and Mexican Americans (32,33), and studies in East Asians identified ANK1 (34), GRK5 and RASGRP1 (35), LEP and GPSM1 (32), and CCDC63 and C12orf51 (36). A study of individuals from Greenland identified TBC1D4 (37), and a sequencing-based study of Danes with follow-up in other Europeans identified MACF1 (38). Finally, the largest GWAS to date in American Indians identified DNER at near genome-wide significance (P = 6.6 × 10⁻⁸) (39). Three of these studies imputed GWAS data using the 1000 Genomes Project sequence-based reference panels, providing better genome coverage (29,32,33,40). Taken together, these studies highlight the value of diverse populations, including founder and historically isolated populations, to detect risk loci.

Novel glycemic trait loci have also been identified since 2012 in selected populations or by analyzing additional glycemic traits. Studies in African Americans identified SC4MOL and TCERG1L for fasting insulin (41), and the TBC1D4 locus identified in Greenland was also strongly associated with 2hGlu and 2hIns (37). Studies in East Asians identified C12orf51, KANK1, IGF1R and PDK1-RAPGEF4 for fasting glucose; MYL2, C12orf51 and OAS1 for 1-2hGlu and HBS1L-MYB, CYBA and MYO9B for HbA1c (42–44). Additional loci were identified for measures of islet β-cell function derived from glycemic traits: insulinogenic index (HNF1A) and disposition index (ABO) (45). ABO encodes a well-known blood group, was previously implicated in blood cell and lipid traits and was subsequently reported as a locus for fasting insulin (46).

MAGIC also used association evidence from fasting traits and dynamic tests of insulin secretion after glucose challenge to characterize T2D loci (47). Clusters of loci appear to have primary effects on insulin sensitivity (PPARG, KLF14, IRS1, GCKR), reduced insulin secretion and fasting hyperglycemia (MTNR1B, GCK), insulin processing (ARAP1), and insulin processing and secretion without a detectable change in fasting glucose levels (TCF7L2, SLC30A8, HHEX-IDE, CDKAL1, CDKN2A); another 20 loci evaluated did not clearly fit any of these clusters (47). Future studies are needed to better understand pathogenetic effects on physiology.

Meta-analyses across populations provide further opportunities to detect loci with shared risk alleles. Meta-analysis of 17 418 T2D cases and 70 298 controls from European, African-American, Hispanic-Latino, and Asian studies using a gene-based CardioChip array was first to identify the BCL2 locus for T2D (26). A recent genome-wide trans-ancestry meta-analysis of 26 488 T2D cases and 83 964 controls from European, East Asian, South Asian and Mexican ancestry, with follow-up in an additional 21 491 T2D cases and 55 647 controls of European ancestry, identified seven new T2D loci (48). The trans-ancestry part of this latter study was performed using variants imputed based on genotype data from the International HapMap Project (49), and follow-up was limited to variants available in Metabochip-typed datasets, suggesting that future trans-ancestry meta-analyses incorporating data imputed to denser reference panels will identify additional loci.

Another strategy that identified loci for T2D and glycemic traits considered the relationship between these traits and BMI. Analysis of T2D in lean (BMI < 25 kg/m²) compared with obese (BMI ≥ 30 kg/m²) individuals identified LAMA1 as a T2D locus with larger effect in lean (odds ratio 1.13, 95% confidence interval 1.09–1.18) than obese (odds ratio 1.03, 95% confidence interval 1.00–1.06) cases (50). Another study of fasting glucose and insulin applied a joint meta-analysis approach to simultaneously test both a main genetic effect, adjusted for BMI, and potential interaction between each genetic variant and BMI (51). Many of the significant associations in that study also were detected in the concurrent MAGIC Metabochip study; additional loci detected included a fasting glucose locus at DPYSL5 and a locus >1 Mb from MADD, yet still in moderate (r² ∼0.2) LD with a MADD signal. A full understanding of the genetic architecture of T2D and glycemic traits will involve teasing apart genetic contributions that depend on correlated traits.

Low-Frequency and Rare Variants

Technological advances and lower costs of high-throughput sequencing have driven new discoveries of low-frequency (0.005 < MAF ≤ 0.05) and rare (MAF ≤ 0.005) variants associated with T2D or related glycemic traits. Novel signals include both new loci and new low-frequency variants at established loci. Approaches to discover these variants include exome or genome sequencing, genotyping arrays and imputation reference panels that include variants identified by sequencing, and gene-based tests that aggregate variants. The threshold of significance used in current studies varies; while the large number of rare variants suggests the need for thresholds more stringent than P < 5 × 10⁻⁸, some studies use study-wide significance based on the number of variants analyzed or the number of genes analyzed in gene-based tests.

Four new loci for T2D or glycemic traits were recently identified using low-frequency or rare variants (45,46,52,53) (Tables 1 and 2). The first exome array-based study to identify novel loci reported low-frequency missense variants (MAF = 0.020–0.029) associated with proinsulin levels at TBC1D30 (Arg279Cys, P = 5.5 × 10⁻¹¹) and KANK1 (Arg667His, P = 1.3 × 10⁻⁸) (Fig. 1), and borderline low-frequency missense variants (MAF = 0.053) in the PAM (Asp563Gly, P = 1.9 × 10⁻⁸) and PPIP5K2 (Ser1228Gly, P = 2.3 × 10⁻⁸) genes, located ∼100 kb apart, associated with insulinogenic index (45). A whole-genome sequencing and imputation study showed that the missense variants in PAM and PPIP5K2 were also associated with T2D (P = 3.9 × 10⁻¹⁰) (52). That study reported an additional rare (MAF = 0.006) missense variant in PAM with independent evidence of T2D association (Ser539Trp, P = 1.7 × 10⁻⁵), implicating PAM as the more likely T2D gene at this locus. Most recently, an exome array-based study identified missense variant GLP1R Ala316Thr (MAF = 0.014) associated with fasting glucose levels (P = 3.4 × 10⁻¹²) (46). A concurrent exome array study using partly overlapping samples supports this association (P = 4.6 × 10⁻⁷) (53). The latter study also identified a novel fasting insulin locus at study-wide significance (URB2 Glu594Val, MAF = 0.001, P = 3.1 × 10⁻⁷) (53). These association signals represent the first of potentially many novel loci that will be identified based on low-frequency variants.

Figure 1.

A low-frequency Finnish missense variant and a common non-coding East Asian variant may tag the same signal at KANK1. The KANK1 locus was originally identified in an exome array study in Finns by a low-frequency missense variant associated with fasting proinsulin (rs3824420, red arrow) (45). Many additional non-coding variants analyzed subsequently, shown here, are also associated with proinsulin in Finns (LD coloring, 1000 Genomes EUR). Also at KANK1, an East Asian GWAS meta-analysis detected a common variant (rs10815355) associated with fasting glucose (43). The East Asian signal (r² > 0.8, purple dotted rectangle) does not include the missense variant (43).

Recent studies have also identified low-frequency variants associated with T2D at established GWAS loci but distinct from the common GWAS variants based on LD and conditional analysis. At CCND2, Steinthorsdottir described an intronic variant (rs76895963, MAF = 0.015) that reduces T2D risk (P = 5.0 × 10⁻²¹) and is correlated with increased CCND2 expression (54). Two studies identified low-frequency variants in MODY genes associated with T2D in the general population, frameshift variant PDX1 Gly218Alafs*12 associated with the increased risk of T2D at study-wide significance (MAF = 0.002, P = 7.3 × 10⁻⁷) (52) and HNF1A Glu508Lys associated with reduced T2D risk (MAF = 0.021, P = 2.4 × 10⁻⁹) (55).

Low-frequency variants distinct from GWAS loci have also been described for glycemic traits. An exome array study in Finns identified variants in SGSM2 (Val996Ile, MAF = 0.014) and MADD (Arg776X, MAF = 0.037) associated with proinsulin and distinct from the GWAS variants (45). At the MADD locus, another study identified a rare (MAF = 0.00068) variant associated with fasting glucose within NR1H3 intron 2 that shows evidence of a functional regulatory effect (56). Also, an exome array study of Europeans identified missense variant G6PC2 His177Tyr (MAF = 0.008) associated with fasting glucose (P = 3.1 × 10⁻⁸) (53). Evidence of allelic heterogeneity may help identify the underlying genes and facilitate biological studies.

A productive approach to analyzing exonic variants is to aggregate them into gene-based tests. Aggregating SLC30A8 variants identified by sequencing or genotyping ∼150 000 individuals, Flannick showed that carriers of 12 rare protein-truncating variants had a 3-fold reduced risk of T2D (P = 1.7 × 10⁻⁶) (54). Another study used gene-based tests of up to 15 variants as evidence that three coding variants in G6PC2 (Val219Leu, His177Tyr, Tyr207Ser) are associated with fasting glucose levels independent of each other and the non-coding GWAS signal. Wessel also reported a significant gene-based association at G6PC2 (P = 6.8 × 10⁻⁶) (46). Gene-based tests are most powerful when they combine only variants that have a functional effect, without neutral variants, as elegantly demonstrated by two studies that tested the function of multiple missense variants in MTNR1B or PPARG using cell-based assays (57,58). Compared with aggregating all known missense variants in those genes, aggregating only the variants that showed functional effects strengthened evidence of T2D association (57,58).

Other Aspects of Genetic Architecture

The genetic architecture of T2D and glycemic traits is also influenced by parent-of-origin effects and sex differences. An Icelandic study was first to report that variants located near KCNQ1, MOB2 and KLF14 showed stronger evidence of association when maternally inherited than when paternally inherited (59). The parent-of-origin effect at these three variants and others was subsequently confirmed in Pima Indians (60). A GWAS meta-analysis for proinsulin reported a locus at DDX31 more significant in women (P = 2.0 × 10⁻⁸) than men (P = 0.17)(23), and the DIAGRAM consortium reported the CCND2 signal was more significant in men (P_men = 1.1 × 10⁻⁹, P_women P = 0.036; P_het = 0.013) and the GIPR signal was more significant in women (P_women = 2.2 × 10⁻⁷, P_men = 0.0037; P_het = 0.057); the sex difference at CCND2 was not replicated in a study of individuals from Iceland, Denmark and Iran (52). Among low-frequency variants, loci with stronger evidence observed in one sex include a missense variant in PAM (rs35658696) associated with T2D (P_men = 5.2 × 10⁻¹⁰, P_women = 0.0044; P_het = 0.033) (52); the association of this variant with insulinogenic index was observed in an all-male study (45). More extensive study of these and other influences may improve understanding of the mechanisms by which variants in or near these genes influence traits or confer susceptibility to disease.

Conclusion and Future Directions

Together, published loci identified at P < 5 × 10⁻⁸ total ∼88 for T2D and 83 for one or more of glucose, HbA1c, insulin and proinsulin. Counts are slightly more or less depending on whether variants hundreds of kilobases apart represent the same or different loci. Combining T2D and these glycemic traits, ∼137 loci have been identified. Of these, only four were identified based on low-frequency variants, and none based on rare variants alone. Loci were discovered in many different populations and typically share direction of effect across populations when they are observed.

New loci will continue to be observed as sample sizes increase and as more variants are analyzed. Imputation panels built using larger numbers of individuals, such as the >39 million variants in ∼65 000 haplotypes from the first release of the Haplotype Reference Consortium (www.haplotype-reference-consortium.org/ last accessed on 6 July 2015), will improve the genome coverage of meta-analyses and help identify signals with previously untested variants. New loci will be identified with lower-frequency variants based on large numbers of sequenced exomes from the T2D-GENES, PROMIS, CHARGE, UK10K and other consortia. Additional loci will be detected based on gene-based tests, especially as annotation for both coding and non-coding regions improves. Results of these studies will more narrowly define the feasible models of genetic architecture for T2D and glycemic traits. Ultimately, the greatest value from these studies will come with the discovery of the underlying genes and biological mechanisms responsible for disease.

Conflict of Interest statement: None declared.

Funding

This work was supported by the National Institutes of Health (DK072193, DK093757 and DK105561 to K.L.M., and DK062370, DK085584 and DK098032 to M.B.).