Advertisement

Diabetologia

, Volume 55, Issue 11, pp 2970–2984 | Cite as

Transferability and fine-mapping of glucose and insulin quantitative trait loci across populations: CARe, the Candidate Gene Association Resource

  • C.  -T.  Liu
  • M. C. Y. Ng
  • D. Rybin
  • A. Adeyemo
  • S. J. Bielinski
  • E. Boerwinkle
  • I. Borecki
  • B. Cade
  • Y. D. I. Chen
  • L. Djousse
  • M. Fornage
  • M. O. Goodarzi
  • S. F. A. Grant
  • X. Guo
  • T. Harris
  • E. Kabagambe
  • J. R. Kizer
  • Y. Liu
  • K. L. Lunetta
  • K. Mukamal
  • J. A. Nettleton
  • J. S. Pankow
  • S. R. Patel
  • E. Ramos
  • L. Rasmussen-Torvik
  • S. S. Rich
  • C. N. Rotimi
  • D. Sarpong
  • D. Shriner
  • M. Sims
  • J. M. Zmuda
  • S. Redline
  • W. H. Kao
  • D. Siscovick
  • J. C. Florez
  • J. I. Rotter
  • J. Dupuis
  • J. G. Wilson
  • D. W. Bowden
  • J. B. MeigsEmail author
Article

Abstract

Aims/hypothesis

Hyperglycaemia disproportionately affects African-Americans (AfAs). We tested the transferability of 18 single-nucleotide polymorphisms (SNPs) associated with glycaemic traits identified in European ancestry (EuA) populations in 5,984 non-diabetic AfAs.

Methods

We meta-analysed SNP associations with fasting glucose (FG) or insulin (FI) in AfAs from five cohorts in the Candidate Gene Association Resource. We: (1) calculated allele frequency differences, variations in linkage disequilibrium (LD), fixation indices (Fsts) and integrated haplotype scores (iHSs); (2) tested EuA SNPs in AfAs; and (3) interrogated within ±250 kb around each EuA SNP in AfAs.

Results

Allele frequency differences ranged from 0.6% to 54%. Fst exceeded 0.15 at 6/16 loci, indicating modest population differentiation. All iHSs were <2, suggesting no recent positive selection. For 18 SNPs, all directions of effect were the same and 95% CIs of association overlapped when comparing EuA with AfA. For 17 of 18 loci, at least one SNP was nominally associated with FG in AfAs. Four loci were significantly associated with FG (GCK, p = 5.8 × 10−8; MTNR1B, p = 8.5 × 10−9; and FADS1, p = 2.2 × 10−4) or FI (GCKR, p = 5.9 × 10−4). At GCK and MTNR1B the EuA and AfA SNPs represented the same signal, while at FADS1, and GCKR, the EuA and best AfA SNPs were weakly correlated (r 2  < 0.2), suggesting allelic heterogeneity for association with FG at these loci.

Conclusions/interpretation

Few glycaemic SNPs showed strict evidence of transferability from EuA to AfAs. Four loci were significantly associated in both AfAs and those with EuA after accounting for varying LD across ancestral groups, with new signals emerging to aid fine-mapping.

Keywords

African ancestry Genetics Genome-wide association LD mapping Minorities Type 2 diabetes 

Abbreviations

AfA

African-American

ARIC

Atherosclerosis Risk in Communities

CARDIA

Coronary Artery Risk Development in Young Adults

CARe

Candidate Gene Association Resource

CEU

HapMap 2 reference panel of Utah residents of northern and western European ancestry

CFS

Cleveland Family Study

CVD

Cardiovascular disease

DIAGRAM

The Diabetes Genetics Replication And Meta-analysis Consortium

EuA

European ancestry

FG

Fasting glucose

FI

Fasting insulin

Fst

Fixation index

HOMA-B

HOMA-beta cell function

HOMA-IR

HOMA- insulin resistance

iHS

Integrated haplotype score

JHS

Jackson Heart Study

LD

Linkage disequilibrium

MAGIC

The Meta-Analyses of Glucose and Insulin-related traits Consortium

MESA

Multi-Ethnic Study of Atherosclerosis

MSA

Metropolitan statistical area

NHLBI

National Heart, Lung, and Blood Institute

SNP

Single-nucleotide polymorphism

varLD

A program for quantifying variation in linkage disequilibrium patterns between populations

YRI

HapMap 2 reference panel of residents of Yoruba in Ibadan, Nigeria

Introduction

In the USA, over 23 million people have type 2 diabetes, but minority groups such as African-Americans (AfAs) are disproportionately affected. In 2005, the prevalence of type 2 diabetes was about 28% higher in AfAs than in individuals of European ancestry (EuA), and risk for diabetes (defined by either elevated fasting glucose or HbA1c) was about twice as high. By 2050, this racial disparity is projected to increase even further [1]. Racial/ethnic variation in health-related behaviours and obesity accounts for much of this disparity [2, 3], but genetic factors could also contribute [4, 5, 6].

Until recently, the majority of genetic studies of type 2 diabetes and related quantitative traits (e.g. levels of fasting glucose [FG] and fasting insulin [FI]) have been conducted in EuA populations [7, 8]. There are strong scientific reasons to expand diabetes-related genetic analyses to AfA, quite apart from the disparity caused by their exclusion from this important research frontier. No single ancestral population is sufficient to fully uncover variants contributing to disease in humans, and of all continental ancestral groups, those of African ancestry contain the richest range of genetic variation. Trans-ethnic genetic analyses can elucidate ancestral differences in common risk-allele frequencies, linkage disequilibrium (LD) patterns and variant effect size differences, as well as guide locus fine-mapping to uncover functional variants identified in EuA genetic studies [4, 5, 6, 9, 10].

Study of AfA populations is likely to be especially important to identify diabetes-related genetic variants [4, 11]. Although one recent study examined whether type 2 diabetes associations found in EuA were also associated in about 2,500 AfA individuals [12], analyses of transferability of type-2-diabetes-associated variants from EuA to AfA populations have not so far been especially instructive, probably because of limited sample sizes and testing of tagging variants rather than functional variants in the setting of reduced LD [13, 14, 15, 16]. On the other hand, analyses of continuously distributed quantitative traits such as FG or FI, elevations of which herald type 2 diabetes [17], offer the advantage of increased power and reduced misclassification by disease status [18]. One recent study of 927 unrelated AfAs confirmed the association of six EuA-identified variants with FG [19]. Another study of 3,024 adults reported that allele frequencies of 16 novel FG-associated single-nucleotide polymorphisms (SNPs) varied across non-Hispanic whites, non-Hispanic blacks and Mexican Americans [9]. In the present study, we investigated whether testing associations of EuA-identified variants in the large Candidate Gene Association Resource (CARe) study would confirm wide risk-allele frequency variation and LD differences in AfA vs EuA and show limited EuA-associated SNP associations in AfA, but when accounting for LD, would show that most FG- or FI-associated loci are associated with diabetes-related quantitative traits in AfAs at different, generally independent, variants.

Methods

Candidate Gene Association Resource study samples

The Candidate Gene Association Resource (CARe) is a National Heart, Lung, and Blood Institute (NHLBI) shared resource for genotype association analyses comprising five parent studies with AfA participants and traits of interest. All individuals provided written informed consent for DNA studies as part of their parent study. CARe study samples are described in greater detail in Lettre et al [15]. Briefly, the studies were as follows.

Atherosclerosis Risk in Communities

The Atherosclerosis Risk in Communities (ARIC) study is an ongoing prospective cohort study originally designed to investigate risk factors of subclinical and clinical atherosclerosis. It includes measurement of cardiovascular and diabetes risk factors. ARIC enrolled 15,792 participants, aged 45–64 years, from four field centres, and has been described in detail elsewhere [20]. This study includes only self-reported AfA participants, and all participants provide written informed consent.

Coronary Artery Risk Development in Young Adults

The Coronary Artery Risk Development in young Adults (CARDIA) study is a prospective multicentre investigation of the development and determinants of clinical and subclinical cardiovascular disease (CVD) and its risk factors. The participants were recruited from four sites and appropriate informed consent was provided [21].

Cleveland Family Study

The Cleveland Family Study (CFS) is a family-based longitudinal study originally designed to investigate genetic and non-genetic risk factors for sleep apnoea [22]. For this analysis, only data from AfAs participating in the most recent examination that included a morning fasting blood draw were analysed. All participants provided written informed consent.

Jackson Heart Study

The Jackson Heart Study (JHS) is a prospective population-based study of CVD among non-institutionalised AfA adults aged 21–95 years and residing in the Jackson, MS metropolitan statistical area (MSA) [23]. The final JHS cohort included 5,301 participants, equivalent to 7% of all AfAs aged 21–95 residing in the Jackson MSA17. Details of the study design and recruitment protocol have been described elsewhere [23, 24, 25, 26]. Appropriate informed consent was provided.

Multi-Ethnic Study of Atherosclerosis

The Multi-Ethnic Study of Atherosclerosis (MESA) is a community-based cohort study designed to investigate characteristics and risk factors for subclinical CVD. Participants were recruited from six centres and appropriate informed consent was provided. From 2000 to 2002, 6,814 participants (38% non-Hispanic white, 28% black, 12% Chinese, 22% Hispanic; aged 45–84 years) without existing clinical CVD were enrolled [27]. Only data from AfA individuals were included in this analysis.

In all studies, FG was measured using hexokinase (in mmol/l) and FI was measured using RIA or ELISA methods in serum or plasma (in μU/ml, converted into pmol/l using a conversion factor of 6).

Genotyping, imputation and principal components analysis

All samples were genotyped at the Broad Institute using the Affymetrix Genome-Wide Human SNP Array 6.0 (Affy6.0, Santa Clara, CA, USA). Quality-control steps were performed centrally by the CARe analytical group at the Broad Institute [15]. Imputation was conducted centrally at the Broad Institute with MACH 1.0.16 using the HapMap 2 reference panel of Utah residents of northern and western European ancestry (CEU) and the Yoruba in Ibadan, Nigeria (YRI). The allelic concordance rate, defined as 1–1/2*|imputed_dosage − genotyped_dosage| based on the CEU + YRI panel, was 95.6%, comparable with rates for individuals of African descent imputed with HapMap 2 YRI individuals [15]. Filters applied to the imputation data were based on the imputation quality measure RSQ_HAT with a threshold of >0.3 and a minor allele frequency threshold of >0.01. Ancestry principal components (PCs) were generated using EIGENSTRAT (http://genepath.med.harvard.edu/~reich/EIGENSTRAT.htm).

Allele frequency differences and natural selection

We used several approaches to evaluate population differentiation and natural selection at index SNPs. First, we assessed index SNP allele-frequency differences in CARe comparing individuals of EuA vs AfA by calculating the absolute value of (index-risk-allele frequency in EuA − index-risk-allele frequency in AfA), using the trait-raising allele in EuA as the risk allele. Second, we calculated Wright’s fixation index (Fst) to assess the degree of population differentiation as reflected by divergent allele frequencies. We calculated Fst using CARe risk-allele frequencies for AfAs, and the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC) [8] risk-allele frequencies for those of EuA, but because the EuA sample size (46,186) was about seven times the AfA sample size, potentially obscuring population differentiation at any locus, we also calculated the Fst using CEU and YRI HapMap 2 data for which the sample sizes are similar. Third, we calculated the integrated haplotype score (iHS) in HapMap 2 data to assess evidence for recent positive selection at the index SNP. Fst and iHS were calculated using Haplotter [28]. We used varLD (a program for quantifying variation in linkage disequilibrium patterns between populations) to compare EuA vs AfA regional LD patterns. The approach quantifies genomic regions carrying dissimilar patterns of LD, regions at which different haplotypes, structural variants and signs of positive selection may be particularly common [29].

Statistical analysis to generate SNP-trait-association results

We tested SNP associations with quantitative traits in each cohort according to a uniform protocol. The primary study traits were FG and FI; in secondary analyses we also tested associations with HOMA-insulin resistance (HOMA-IR) and HOMA-beta cell function (HOMA-B) [30]. FI and HOMA were analysed on the natural log scale. Ancestry PCs were created in each study; as just the first two PCs were associated with FG (p < 0.005) these were used as covariates in each study to control population stratification. Genomic control correction based on median χ 2 value was calculated across all loci in the genome-wide association study (GWAS) data within each study to adjust for inflation of test statistics. Traits were analysed using linear regression under an additive genetic model, adjusted for age, age2, sex, study site (if applicable) and the first two PCs; relatedness was accounted for when necessary using linear mixed effect (LME) models. PLINK (http://pngu.mgh.harvard.edu/~purcell/PLINK/) and R (v2.9.0) were used for data management and statistical analyses. We performed fixed-effect meta-analyses on study-specific genome-wide association results to generate a combined association effect estimate for each SNP using the inverse-variance weighted approach in METAL (a tool for meta-analysis of genome-wide association scans) [31]. Nominal significance was defined as a SNP association p value < 0.05. We estimated the power of the sample to detect additive SNP-FG (in mmol/l) associations, using the detectable effect size as the 25th, 50th and 75th percentiles of the distribution of β coefficients seen in Dupuis et al [8]. At n ~ 6,000 and α = 4.05 × 10−5, we had low power to detect all but the largest effect sizes (electronic supplementary material [ESM] Fig. 1).

Interrogation of transferability

We defined SNP transferability as an EuA index SNP having the same direction of association with the quantitative trait and a p value < 0.05 in AfA individuals, and locus transferability as a locus containing any SNPs significantly associated with the quantitative trait in AfAs, with significance defined as any SNP with a p value < 0.05 after accounting for the effective number of tests at the locus. We took two approaches to investigate transferability of EuA quantitative trait-associated SNPs and loci in AfA. First, we tested index SNPs reported to be associated with FG or FI at genome-wide significance (5 × 10−8) in EuA individuals in the MAGIC study [8]. We also tested replicated significant type 2 diabetes SNPs [32, 33, 34, 35, 36] that had consistent nominal associations with FG or FI in the Diabetes Genetics Replication and Meta-analysis Consortium (DIAGRAM) or MAGIC studies [7, 8]. Second, we searched for the best SNPs in AfAs accounting for differences in regional LD structure in AfAs vs those with EuA by interrogating the flanking 250 kb regions on each side of the index SNP, with the best SNP defined as the SNP with the smallest FG or FI association p value in the region in AfA individuals. Note that we treated each region independently in this follow-up study of regions that had a higher prior probability of association with the trait of interest than the genomic average, based on previously reported associations in Europeans. As the results for any one region had no inferential bearing on the results for any other region, we made region-wide (not genome-wide) corrections to adjust for multiple comparisons. For a locus, the significance of the identified best SNPs was evaluated using a Bonferroni correction adjusting for the effective number of tests within that locus, estimated using the Li and Ji algorithm [37].

Results

We analysed data from 5,984 (FG) to 5,969 (FI) non-diabetic AfA individuals from five CARe participating cohorts (Table 1). About 40% within each cohort were men and cohorts had a similar mean FG levels (range from 4.5 to 5.4 mmol/l). Mean age ranged more widely (from 24.4 to 61.7 years) across cohorts, as did mean FI (from 40.6 to 97.8 pmol/l).
Table 1

Characteristics of five CARe cohorts

Cohort

ARIC (n = 1,734)

CARDIA (n = 930)

CFS (n = 272)

JHS (n = 1,707)

MESA (n = 1,341)

Men, n (%)

649 (37.4)

362 (38.9)

104 (38.2)

682 (40.0)

596 (44.4)

Age (years)

53 ± 5.8

24 ± 3.9

36 ± 17.7

49 ± 12.1

62 ± 10.3

FG (mmol/l)

5.4 ± 0.5

4.5 ± 0.48

5.1 ± 0.54

5 ± 0.5

5 ± 0.6

FI (pmol/l)a

89.1 ± 65.7

75.1 ± 52.7

84.2 ± 69.5

97.8 ± 55.6

40.6 ± 27.8

HOMA-Bb

160.7 ± 110.1

290.6 ± 248.3

187.4 ± 193.1

229.9 ± 129.7

95.8 ± 65.6

HOMA-IRb

3.6 ± 2.9

2.6 ± 1.9

3.3 ± 2.8

3.7 ± 2.4

1.6 ± 1.2

Data are mean ± SD unless otherwise specified

a1 pmol/l = 6 μU/ml for insulin

bNumbers of missing individuals for HOMA-B and HOMA-IR: ARIC, 9; CARDIA, 17; CFS, 1; JHS, 0; MESA, 4

Allele frequency differences and natural selection

As observed for quantitative glycaemic traits, index SNP allele-frequency differences between those of EuA and AfAs had a wide range (Table 2). Fst exceeded 0.15 at 1/18 loci in CARe but at 6/18 loci in HapMap 2, indicating modest population differentiation based on allele frequency differences at ADRA2A, CRY2, PROX1, FADS1, IGF1 and C2CD4B. The absolute values of all iHSs were <2, suggesting no recent strong positive selection at any locus. For the FADS1 and G6PC2 loci at which the index and best SNPs were weakly correlated, the varLD scores were elevated in AfAs, suggesting allelic heterogeneity of tag SNPs for causal variants or other forms of variation, such as copy number variations (Fig. 1).
Table 2

Allele frequency differences, Fsts and iHSs in AfA and EuA populations

Nearest gene(s)

SNP

Allelesa

EUA EAFb

AfA EAFb

CARe (EAF EuA–AfA)

p value for EAF differencec

Fst CARe

EAF HapMap 2 CEU

EAF HapMap 2 YRI

Fst HapMap 2 CEU vs YRI

iHS HapMap 2 CEU

iHS HapMap 2 YRI

FG-associated loci

  PROX1

rs340874

C/T

0.520

0.169

0.351

<1 × 10−384

0.051

0.508

0.092

0.207

−0.06

−0.09

  G6PC2

rs560887

C/T

0.700

0.943

0.243

<1 × 10−384

0.031

0.700

1.000

−0.26

  GCKR

rs780094

C/T

0.620

0.817

0.197

8.78 × 10−196

0.017

0.617

0.883

0.095

−0.39

0.14

  ADCY5

rs11708067

A/G

0.780

0.843

0.063

3.68 × 10−29

0.002

0.783

0.917

0.035

−0.36

−0.79

  SLC2A2

rs11920090

T/A

0.870

0.645

0.225

<1 × 10−384

0.040

0.853

0.667

0.051

1.17

0.60

  GCK

rs4607517

A/G

0.160

0.101

0.059

6.49 × 10−33

0.003

0.200

0.059

0.045

−0.26

−0.20

  DGKB-TMEM195

rs2191349

T/G

0.520

0.593

0.073

1.60 × 10−26

0.002

0.467

0.593

0.016

−0.13

−0.07

  SLC30A8

rs13266634

C/T

0.680

0.908

0.228

2.96 × 10−290

0.026

0.750

0.942

0.070

−1.87

−0.55

  GLIS3

rs7034200

A/C

0.490

0.630

0.140

3.79 × 10−92

0.008

0.525

0.585

0.003

0.47

−0.24

  ADRA2A

rs10885122

G/T

0.870

0.334

0.537

<1 × 10−384

0.191

0.900

0.217

0.484

−0.23

−0.25

  TCF7L2

rs7903146

T/C

0.290

0.284

0.006

3.20 × 10−1

0.000

0.250

0.292

0.002

0.09

−0.27

  MTNR1B

rs10830963

G/C

0.300

0.073

0.227

<1 × 10−384

0.027

0.300

0.050

0.108

−0.05

−0.55

  MADD

rs7944584

A/T

0.750

0.950

0.200

2.27 × 10−254

0.024

0.692

1.000

−0.31

  FADS1

rs174550

T/C

0.640

0.916

0.276

<1 × 10−384

0.036

0.633

0.983

0.198

−0.62

  CRY2

rs11605924

A/C

0.490

0.867

0.377

<1 × 10−384

0.059

0.542

0.941

0.209

−1.50

−1.02

  C2CD4B

rs11071657

A/G

0.630

0.868

0.238

4.61 × 10−292

0.026

0.592

0.942

0.171

1.07

−1.65

FI-associated loci

  GCKR

rs780094

C/T

0.620

0.817

0.197

2.70 × 10−192

0.020

0.617

0.883

0.095

−0.39

0.14

  IGF1

rs35767

G/A

0.850

0.570

0.280

<1 × 10−384

0.061

0.900

0.525

0.172

−1.290

0.374

The loci in this table were obtained from Dupuis et al (2010) [8]

aEffect (trait-raising)/other allele

bEAF, effect allele frequency

c P value for testing whether the effect allele frequency EAF in AfAs is the same as the effect allele frequency in EuA populations

Fig. 1

500 kb regional association plots centred at the index SNPs identified from EuA samples at GCK, MTNR1B, FADS1 and G6PC2. The x-axis denotes genomic position and the y-axis denotes the − log(p value) for each SNP. The two larger-sized data points represent the best SNPs within the region in the AfA CARe sample (black arrows) and the index SNPs (white arrows, locating the centre of the region) previously reported from EuA population samples. The colour of each data point indicates its LD value (r 2) with the index SNP based on HapMap 2 YRI: white, r 2 = 0.0–0.2; yellow, r 2 = 0.2–0.5; orange, r 2 = 0.5–0.8; red, r 2 = 0.8–1.0. The blue line represents the recombination rate. The green line shows the varLD score at each SNP and is highlighted with dark brown if the varLD score is ≥95th percentile of the genome-wide varLD score, comparing LD information between YRI and CEU HapMap 2 samples [29]. (a) GCK region: the index SNP rs4607517 has a p value of 1.42 × 10−6 in AfA CARe individuals, while the best SNP rs1799884 has a p value of 5.79 × 10−8. Their r 2 values are 0.469 in HapMap 2 YRI and 1.0 in HapMap 2 CEU. (b) MTNR1B region: rs10830963 is the index SNP and the best SNP within the region in AfA CARe individuals, with a p value of 8.51 × 10−9. (c) G6PC2 region: the index SNP rs560887 has a p value of 4.08 × 10−3 in AfA CARe individuals and the best SNP rs853789 has a p value of 1.62 × 10−3. The r 2 value is not known in HapMap 2 YRI and is 0.692 in HapMap 2 CEU. A known functional SNP in the region, rs13431652 [38], has r 2  = 0.923 with rs560887 and r 2  = 0.595 with rs853789 in HapMap 2 CEU, but in HapMap 2 YRI the r 2 of rs13431652 with rs853789 is 0.004. (d) FADS1 region: the index SNP rs174550 has a p value of 0.399 in AfA CARe individuals and the best SNP rs7937198 has a p value of 2.22 × 10−4. Their r 2 values are 0.00 in HapMap 2 YRI and 0.24 in HapMap 2 CEU

Interrogation of EuA FG- and FI-associated loci in AfAs

Trait-raising allele frequencies of EuA index SNPs in AfAs varied widely, from a minimum of a 0.6% difference (TCF7L2) up to a 54% difference (ADRA2) between ancestry groups (Table 3). Of EuA index SNPs tested directly in AfAs, all 18 were directionally consistent; three of the 18 were nominally (p < 0.05) associated with FG and one with FI. In addition, all of the published 95% CIs around point estimates for association of index SNPs in those of EuA overlapped with those tested here in AfAs. We also performed formal heterogeneity tests comparing effects in EuAs with effects in AfAs. Heterogeneity tests yielded only one nominally significant result for rs10830963 (p value = 4.26 × 10−2 at MTNR1B), indicating that the effect sizes in those of EuA vs AfAs for at least 17 index SNPs were not statistically different.
Table 3

Transferability to AfAs of FG- and FI-related loci previously reported in EuA populations

Nearest gene(s)

Index SNP

Chr.

Position

Allelesb

EuA association

AfA association for index SNP

Heterogeneitya

EAF

Effect ([mmol/l] per effect allele)

95% CI

EAF

Effect ([mmol/l] per effect allele)

95% CI

p value

p value

FG-associated loci

  PROX1

rs340874

1

212225879

C/T

0.520

0.013

0.007, 0.019

0.169

0.006

−0.020, 0.031

6.63 × 10−1

6.00 × 10−1

  G6PC2

rs560887

2

169471394

C/T

0.700

0.075

0.069, 0.081

0.943

0.060

0.019, 0.100

4.08 × 10−3

4.80 × 10−1

  GCKR

rs780094

2

27594741

C/T

0.620

0.029

0.023, 0.035

0.817

0.022

−0.002, 0.046

7.05 × 10−2

5.72 × 10−1

  ADCY5

rs11708067

3

124548468

A/G

0.780

0.027

0.021, 0.033

0.843

0.015

−0.010, 0.041

2.31 × 10−1

3.68 × 10−1

  SLC2A2

rs11920090

3

172200215

T/A

0.870

0.020

0.012, 0.028

0.645

0.005

−0.014, 0.025

5.81 × 10−1

1.64 × 10−1

  GCK

rs4607517

7

44202193

A/G

0.160

0.062

0.054, 0.070

0.101

0.084

0.050, 0.118

1.42 × 10−6

2.09 × 10−1

  DGKB-TMEM195

rs2191349

7

15030834

T/G

0.520

0.030

0.024, 0.036

0.593

0.016

−0.003, 0.036

1.02 × 10−1

1.80 × 10−1

  SLC30A8

rs13266634

8

118253964

C/T

0.680

0.027

0.019, 0.035

0.908

0.004

−0.030, 0.038

8.16 × 10−1

1.88 × 10−1

  GLIS3

rs7034200

9

4279050

A/C

0.490

0.018

0.012, 0.024

0.630

0.018

−0.001, 0.037

6.15 × 10−2

1.00 × 10−0

  ADRA2A

rs10885122

10

113032083

G/T

0.870

0.022

0.014, 0.030

0.334

0.006

−0.014, 0.026

5.68 × 10−1

1.37 × 10−1

  TCF7L2

rs7903146

10

114748339

T/C

0.290

0.023

0.015, 0.031

0.284

0.013

−0.010, 0.035

2.63 × 10−1

3.93 × 10−1

  MTNR1B

rs10830963

11

92348358

G/C

0.300

0.067

0.061, 0.073

0.073

0.104

0.069, 0.139

8.51 × 10−9

4.26 × 10−2

  MADD

rs7944584

11

47292896

A/T

0.750

0.021

0.015, 0.027

0.950

0.017

−0.024, 0.059

4.18 × 10−1

8.50 × 10−1

  FADS1

rs174550

11

61328054

T/C

0.640

0.017

0.011, 0.023

0.916

0.014

−0.019, 0.048

3.99 × 10−1

8.62 × 10−1

  CRY2

rs11605924

11

45829667

A/C

0.490

0.015

0.009, 0.021

0.867

0.016

−0.012, 0.043

2.71 × 10−1

9.44 × 10−1

  C2CD4B

rs11071657

15

60221254

A/G

0.630

0.008

0.002, 0.014

0.868

0.006

−0.027, 0.038

7.37 × 10−1

9.08 × 10−1

FI-associated loci

  GCKR

rs780094

2

27594741

C/T

0.620

0.032

0.024, 0.040

0.817

0.025

−0.00, 0.053

7.67 × 10−2

6.31 × 10−1

  IGF1

rs35767

12

101399699

G/A

0.850

0.010

−0.002, 0.022

0.570

0.001

−0.022, 0.023

9.49 × 10−1

5.02 × 10−1

The loci in this table were all obtained from Dupuis et al (2010) [8]

aHeterogeneity test of the β coefficients for EuA populations and AfAs

bEffect (trait-raising)/other allele

Chr chromosome; EAF effect allele frequency

Further, in the ±250 kb flanking regions around index SNPs (Table 4), nearly all (17 of 18) had a best SNP nominally significant in AfAs, and four (at GCK, MTNR1B and FADS1 for FG and GCKR for FI) had SNPs that remained significantly associated after adjusting for the effective number of SNPs tested within the regions. At 15 of 18 loci in AfAs, the index SNP and the best SNP were not in substantial LD (all r 2 < 0.2 in AfAs). At GCK, MTNR1B and CRY2, the index and best SNPs were in strong LD or were the same SNP. Of the best AfA SNPs, five of 18 (at G6PC2, GCKR, GCK, MTNR1B and CRY2) were also at least nominally associated (p < 0.05) with FG or FI in those of EuA in the MAGIC data (ESM Table 1).
Table 4

Regional interrogation in AfAs for FG- and FI-related loci previously reported in EuA populations

Nearest gene(s)

Index SNP in EuA

Interrogation in AfA within 500 kb region of index SNP

Best SNP in AfA

Position

LDa in YRI r 2 (D’)

LDb in CEU r 2 (D’)

Allelesc

EAF

Effect ([mmol/l] per effect allele)

SE

95% CI

p value

Eff. Nd

Adjusted p value

FG-associated loci

  PROX1

rs340874

rs7519597

212261507

0.004 (1.000)

N/A

T/C

0.038

−0.0879

0.0276

−0.142, −0.034

1.42 × 10−3

90

1.28 × 10−1

  G6PC2

rs560887

rs853789

169509734

N/A

0.692 (0.916)

A/G

0.105

−0.0494

0.0157

−0.080, −0.019

1.62 × 10−3

105

1.70 × 10−1

  GCKR

rs780094

rs7586601

27438170

0.169 (0.491)

0.549 (0.956)

A/G

0.239

−0.0303

0.0111

−0.052, −0.009

6.64 × 10−3

22

1.46 × 10−1

  ADCY5

rs11708067

rs10934645

124552510

0.009 (1.000)

0.022 (1.000)

A/G

0.091

0.0573

0.0184

0.021, 0.093

1.86 × 10−3

66

1.23 × 10−1

  SLC2A2

rs11920090

rs9827202

172296993

0.005 (0.105)

0.000 (0.118)

T/C

0.443

0.0274

0.0100

0.008, 0.047

5.86 × 10−3

65

3.81 × 10−1

  GCK

rs4607517

rs1799884

44195593

0.469 (1.000)

0.003 (0.196)

T/C

0.177

0.0659

0.0121

0.042, 0.090

5.79 × 10−8

34

1.97 × 10−6

  DGKB-TMEM195

rs2191349

rs17327498

15111235

0.001 (0.184)

0.003 (0.196)

A/G

0.972

0.1132

0.0343

0.046, 0.180

9.66 × 10−4

90

8.69 × 10−2

  SLC30A8

rs13266634

rs2649098

118304453

0.008 (1.000)

0.004 (0.088)

T/C

0.106

0.0477

0.0153

0.018, 0.078

1.83 × 10−3

82

1.50 × 10−1

  GLIS3

rs7034200

rs3895473

4319209

0.003 (0.098)

0.006 (0.158)

A/G

0.800

−0.0417

0.0117

−0.065, −0.019

3.70 × 10−4

169

6.25 × 10−2

  ADRA2A

rs10885122

rs7908674

112913361

0.070 (0.426)

0.028 (0.500)

A/G

0.159

0.0345

0.0141

0.007, 0.062

1.42 × 10−2

83

1.00 × 10−0

  TCF7L2

rs7903146

rs7909517

114852926

0.006 (0.255)

0.003 (1.000)

C/G

0.103

0.0606

0.0182

0.025, 0.096

8.42 × 10−4

72

6.06 × 10−2

  MTNR1B

rs10830963

rs10830963

92348358

Same SNP

Same SNP

C/G

0.927

−0.1038

0.018

−0.139, −0.069

8.51 × 10−9

86

7.32 × 10−7

  MADD

rs7944584

rs1685404

47200241

N/A

0.258 (1.000)

C/G

0.241

−0.032

0.0109

−0.053, −0.011

3.35 × 10−3

27

9.03 × 10−2

  FADS1

rs174550

rs7937198

61525549

0.000 (0.016)

0.003 (0.055)

A/G

0.491

−0.0341

0.0092

−0.052, −0.016

2.22 × 10−4

65

1.44 × 10−2

  CRY2

rs11605924

rs11605924

45829667

Same SNP

Same SNP

A/C

0.867

0.0155

0.0141

−0.012, 0.043

2.71 × 10−1

58

1.00 × 10−0

  C2CD4B

rs11071657

rs893157

60165933

0.019 (0.488)

0.174 (0.527)

T/C

0.584

−0.0264

0.0097

−0.045, −0.007

6.30 × 10−3

60

3.78 × 10−1

FI-associated loci

  GCKR

rs780094

rs7586601

27438170

0.169 (0.491)

0.549 (0.956)

A/G

0.239

−0.0452

0.0131

−0.071, −0.020

5.87 × 10−4

22

1.29 × 10−2

  IGF1

rs35767

rs7970320

101508605

0.011 (0.126)

0.001 (0.067)

T/C

0.557

−0.0300

0.011

−0.052, −0.008

6.60 × 10−3

59

3.89 × 10−1

The loci in this table were all obtained from Dupuis et al (2010) [8]

aLD in YRI between the best SNP in AfAs and top SNP in EuA populations

bLD in CEU between the best SNP in AfAs and top SNP in EuA populations

cEffect (trait-raising)/other allele

dEffective number of independent (typed) SNPs interrogated in CARe samples of AfAs

N/A, not available

Associations with HOMA-B (for FG loci) and with HOMA-IR (for FI loci) for both the index SNPs and the best FG or FI SNPs are shown in ESM Tables 2 and 3. Seven SNPs out of 16 EuA FG index SNPs showed nominally significant association with HOMA-B in AfAs, while three best AfA SNPs reached nominal significance. Neither of two EuA FI index SNPs were nominally significant, while the best AfA FI SNPs at both loci were nominally significant (at GCKR and IGF1).

Interrogation of FG or FI-associated type 2 diabetes loci in AfAs

Fourteen of the ~40 known type 2 diabetes loci have also been nominally associated with FG or FI in individuals of EuA [7, 8]. Of these 14, only TP53INP1 was directionally consistent and nominally associated with FG in AfAs (Table 5). Overall, 8/14 index EuA SNPs had directionally consistent FG associations in AfAs, ten out of 13 best AfA SNPs (one SNP was monomorphic in AfA) showed 95% CIs that overlapped with EuA index SNPs in AfA and eight out of 13 best AfA SNPs did not reach nominal significance level in formal heterogeneity tests comparing those of EuA with AfAs. Note that the loci MTNR1B (p value = 1.38 × 10−6 for index SNP rs1381753) is again among the loci (THADA, ADAMTS9, HHEX and KCNQ1) with heterogeneous effect (Table 5) and its allelic effect of 0.060 is remarkably larger in the sample of EuA compared with 0.008 in the sample of AfAs. After accounting for ancestral LD by querying SNPs flanking the index SNP, all 14 best SNPs around the 14 index SNPs showed nominally significant associations in AfAs, and four of these best SNPs (at KCNQ1, MTNR1B, ADAMTS9 and CDC123/CAMK1D) remained significantly associated with FG after adjusting for multiple comparisons (Table 6 and ESM Fig. 2). At all 14 loci the best SNP was in low LD with the index SNP (all r 2  < 0.2). The best SNPs at ZBED3, TP53INP1, KCNQ1 and MTNR1B in AfAs were nominally associated with FG in EuA individuals in the MAGIC data (ESM Table 1).
Table 5

Association with FG in AfAs of FG-associated type 2 diabetes risk SNPs previously reported in EuA populations

Nearest gene(s)

Index SNPd

Chr.

Position

Allelesb

EuA FG result for index SNP

AfA FG result for index SNP

Heterogeneitya

EAF

Effect ([mmol/l] per effect allele)

95% CI

p value

EAF

Effect ([mmol/l] per effect allele)

95% CI

p value

p value

THADA

rs7578597

2

43586327

C/T

0.9

0.026

0.014, 0.037

2.40 × 10−5

0.72

−0.003

−0.023, 0.017

7.81 × 10−1

1.29 × 10−2

ADAMTS9

rs4607103

3

64686944

T/C

0.76

−0.010

−0.018, −0.001

2.64 × 10−2

0.71

0.028

0.008, 0.048

6.85 × 10−3

4.18 × 10−4

IGF2BP2

rs4402960

3

186994381

T/G

0.29

0.011

0.003, 0.018

8.34 × 10−3

0.51

−0.003

−0.021, 0.015

7.63 × 10−1

1.55 × 10−1

ZBED3

rs4457053

5

76460705

G/A

0.26

0.018

0.009, 0.026

9.15 × 10−5

N/A

N/A

N/A

N/A

N/A

CDKAL1

rs7754840

6

20769229

C/G

0.31

0.010

0.002, 0.017

1.41 × 10−2

0.57

0.008

−0.010, 0.027

3.74 × 10−1

8.39 × 10−1

TP53INP1

rs896854

8

96029687

T/C

0.48

0.012

0.005, 0.019

8.79 × 10−4

0.69

0.022

0.002, 0.042

3.40 × 10−2

3.53 × 10−1

SLC30A8

rs13266634

8

118253964

C/T

0.75

0.027

0.018, 0.036

5.46 × 10−10

0.91

0.004

−0.030, 0.038

8.16 × 10−1

1.88 × 10−1

CDKN2A/B

rs10811661

9

22124094

T/C

0.79

0.019

0.009, 0.028

9.99 × 10−5

0.93

0.026

−0.011, 0.063

1.68 × 10−1

7.22 × 10−1

CDC123/CAMK1D

rs12779790

10

12368016

A/G

0.18

−0.016

−0.025, −0.006

1.20 × 10−3

0.87

−0.014

−0.044, 0.015

3.40 × 10−1

8.99 × 10−1

HHEX

rs1111875

10

94452862

C/T

0.56

0.009

0.002, 0.016

1.41 × 10−2

0.77

−0.017

−0.039, 0.004

1.16 × 10−1

2.63 × 10−2

KCNQ1

rs231362

11

2648047

G/A

0.52

0.019

0.011, 0.027

5.29 × 10−6

0.78

−0.019

−0.044, 0.006

1.46 × 10−1

5.21 × 10−3

CENTD2 c

rs1552224

11

72110746

A/C

0.88

0.019

0.010, 0.029

1.20 × 10−4

0.97

−0.005

−0.065, 0.055

8.73 × 10−1

4.45 × 10−1

MTNR1B

rs1387153

11

92313476

T/C

0.28

0.060

0.052, 0.069

6.59 × 10−45

0.38

0.008

−0.012, 0.028

4.38 × 10−1

1.38 × 10−6

PRC1

rs8042680

15

89322341

A/C

0.22

0.010

0.002, 0.017

1.30 × 10−2

0.84

0.019

−0.007, 0.045

1.50 × 10−1

5.08 × 10−1

FG results in EuA are from MAGIC, Dupuis et al (2010) [8], as reported in Voight et al (2010) [7]

aHeterogeneity test of the β coefficients for EuA populations and AfAs

bEffect/other allele

cAlso known as ARAP1

dVoight et al (2010) [7]; Zeggini et al (2008) [32]; Rung et al (2009) [53]; Altshuler (2000) [54]; Zeggini et al (2007) [33]; Scott et al (2007) [34]; Diabetes Genetics Initiative of Broad Institute et al (2007) [55]; Sladek (2007) [36]; Gloyn (2003) [56]; Winckler et al (2007) [57]; Gudmundsson (2007) [58]

Chr., chromosome; N/A, not available

Table 6

Regional interrogation in AfAs for FG-associated type 2 diabetes loci previously reported in EuA populations

Nearest gene(s)

Index SNP

Interrogation for FG in AfA within 500 kb region of type 2 diabetes index SNP

 

Best SNP

Position

LDa in YRI r 2 (D’)

LDb in CEU r 2 (D’)

Allelesc

EAF

Effect ([mmol/l] per effect allele)

95% CI

p value

Eff. Nd

Adjusted p value

THADA

rs7578597

rs6724325

43529402

0.066 (0.322)

N/A

T/C

0.189

0.036

0.012, 0.059

2.80 × 10−3

62

1.74 × 10−1

ADAMTS9

rs4607103

rs4688504

64659981

0.020 (0.394)

0.003 (0.080)

T/C

0.319

0.040

0.018, 0.062

3.82 × 10−4

125

4.78 × 10−2

IGF2BP2

rs4402960

rs7627308

186919528

0.031 (1.000)

N/A

A/G

0.013

0.166

0.052, 0.279

4.19 × 10−3

55

2.30 × 10−1

ZBED3

rs4457053

rs12517269

76454133

0.001 (0.029)

0.003 (0.080)

A/G

0.883

−0.049

−0.077, −0.021

6.06 × 10−4

85

5.15 × 10−2

CDKAL1

rs7754840

rs9368255

21008207

0.002 (0.250)

0.019 (1.000)

T/C

0.949

0.057

0.015, 0.099

7.88 × 10−3

72

5.67 × 10−1

TP53INP1

rs896854

rs509594

96036332

0.069 (1.000)

0.737 (1.000)

C/G

0.795

−0.035

−0.058, −0.012

2.54 × 10−3

61

1.55 × 10−1

SLC30A8

rs13266634

rs2649098

118304453

0.008 (1.000)

0.004 (0.088)

T/C

0.106

0.048

0.018, 0.078

1.83 × 10−3

82

1.50 × 10−1

CDKN2A/B

rs10811661

rs10965319

22293465

0.092 (1.000)

0.008 (0.216)

A/T

0.840

−0.041

−0.068, −0.014

2.86 × 10−3

72

2.06 × 10−1

CDC123/CAMK1D

rs12779790

rs7074893

12446905

0.000 (1.000)

0.001 (0.068)

T/G

0.977

−0.209

−0.312, −0.105

8.29 × 10−5

89

7.38 × 10−3

HHEX

rs1111875

rs11187186

94576008

0.000 (0.017)

0.000 (0.017)

T/C

0.457

−0.029

−0.048, −0.010

2.24 × 10−3

42

9.41 × 10−2

KCNQ1

rs231362

rs2011766

2490838

0.000 (0.033)

0.001 (0.085)

T/C

0.332

−0.044

−0.068, −0.021

2.02 × 10−4

91

1.84 × 10−2

CENTD2 e

rs1552224

rs6592481

72067275

N/A

0.001 (0.273)

T/C

0.220

0.038

0.012, 0.064

4.79 × 10−3

51

2.44 × 10−1

MTNR1B

rs1387153

rs10830963

92348358

0.036 (0.710)

0.705 (0.874)

C/G

0.927

−0.104

−0.139, −0.069

8.51 × 10−9

87

7.40 × 10−7

PRC1

rs8042680

rs10083635

89195519

0.006 (1.000)

N/A

A/G

0.145

0.060

0.025, 0.094

6.53 × 10−4

98

6.40 × 10−2

Index SNP data are from Voight et al (2010) [7]; Zeggini et al (2008) [32]; Rung et al (2009) [53]; Altshuler (2000) [54]; Zeggini et al (2007) [33]; Scott et al (2007) [34]; Diabetes Genetics Initiative of Broad Institute et al (2007) [55]; Sladek (2007) [36]; Gloyn (2003) [56]; Winckler et al (2007) [57]; Gudmundsson (2007) [58]

aLD in YRI between the best SNP in AfAs and top SNP in EuA populations

bLD in CEU between the best SNP in AfAs and top SNP in EuA populations

cEffect/other allele

dEffective number of independent (typed) SNPs interrogated in CARe samples of AfA

eAlso known as ARAP1

N/A, not available

Discussion

We found that among 18 index SNPs robustly associated with glycaemic quantitative traits in the MAGIC EuA populations, just three SNPs (at GCK, MTNR1B and G6PC2 for FG) were directly transferrable to the ~6,000 AfA individuals from five separate cohorts in CARe, but all 18 published 95% CIs around association-point estimates in MAGIC overlapped with the CIs based on associations in AfAs. However, after accounting for ancestral differences in LD by searching more broadly in regions flanking the EuA index SNP, 17 of 18 loci were nominally associated with FG or FI in AfAs, and SNPs at four of these loci (GCK, MTNR1B and FADS1 for FG and GCKR for FI) remained significantly associated with FG or FI in AfAs after accounting for multiple comparisons. For FG-associated type 2 diabetes SNPs discovered in EuA populations, only one (TP53INP1) was directly transferable to AfAs, but ten of 13 published 95% CIs overlapped with the CIs based on associations in AfA. These overlapping intervals indicate that the trait-raising allelic effects for glycaemic loci are similar in individuals of EuA and AfAs. Four FG-associated type 2 diabetes loci (ADAMTS9, CDC123/CAMK1D, KCNQ1 and MTNR1B) had a best AfA SNP significantly associated with FG in AfAs after adjusting for multiple comparisons. As expected, allele frequencies and regional LD varied widely: there was evidence of modest population differentiation for six index SNPs (including FADS1), but none showed evidence of recent positive selection. These results provide evidence that some genetic loci associated with glycaemic regulation are reproducibly shared across human populations, with a suggestion of potential transferability for most loci, at least for individuals of European or African ancestry. The data also provide evidence for allelic heterogeneity at many loci and evidence for shared haplotypes with proximity of an index SNP with a functional variant at several loci.

These results illustrate the value of examining AfA cohorts for fine-mapping. Where there is strong LD between the index SNP and the best SNP, association with FG in an ancestral group with reduced LD increases confidence that the index signal is on the same haplotype or is otherwise close to a functional mutation (GCK, MTNR1B and CRY2). For instance, at MTNR1B, the FG–type 2 diabetes signal is the same for both those of EuA and AfAs (rs10830963). Where there is weak LD between the index SNP and the best SNP (all other loci in AfAs), there are two alternative explanations: (1) these SNPs represent independent signals where the index and best SNPs were significant in only one population (population-specific allelic heterogeneity); or (2) they share LD with a tagged untyped functional SNP in the region where the differential LD pattern between those of EuA and AfAs can focus fine-mapping of the association signal. For instance, at G6PC2 in EuA the FG index SNP (rs560887) and best SNP (rs853789) are in strong LD with a known functional SNP (rs13431652; r 2 with rs560887 in CEU = 0.923, r 2 with rs853789 in CEU = 0.595) [38], but AfA rs853789 is not correlated with rs13431652 (r 2 in YRI = 0.004). KCNQ1 also demonstrates multiple independent signals identified in different ancestry groups [7, 39, 40]. Variation in LD structure across these ancestral groups has focused regions of interest and identified new signals that may aid future fine-mapping to identify functional variants involved in FG and FI regulation.

Just one paper has investigated the transferability of EuA diabetes-related quantitative trait loci in AfAs. Ramos et al examined the index and correlated SNPs (r 2 ≥ 0.3 in CEU) for 16 fasting glucose loci reported by MAGIC in 927 AfAs [19], demonstrating associations of index and/or correlated SNPs at TMEM195 (also known as AGMO), SLC30A8, TCF7L2, G6PC2, GCKR and MTNR1B. This is consistent with our index SNP association results at G6PC2 and MTNR1B and regional association at GCKR. In addition, the direction and magnitude of effect of the T allele in rs2191349 at TMEM195 were comparable (β = 0.014 [Ramos] vs 0.016 [this study]), despite not reaching significance in this study. Taken together with the significant association of AfA best SNPs at GCKR, GCK and MTNR1B with FG or FI in EuA populations in MAGIC, the loci GCKR, GCK and MTNR1B appear likely to share the same causal variant(s) in populations with EuA and AfAs. Differences in LD such as those found at GCK and GCKR between those with EuA and AfAs also facilitate the fine-mapping of causal variants in these loci. The associations of independent SNPs at G6PC2 and FADS1 in EuA populations and AfAs probably point to distinct signals requiring further study.

Our data add to the list of glycaemic loci confirmed and informatively fine-mapped by cross-ancestry analyses [41, 42, 43]. For instance, at KCNQ1, studies in South and East Asian populations identified the type-2-diabetes-associated SNPs rs2237892, rs2237895 and rs2237897 [39, 40], weakly correlated with the signal seen in EuA populations (rs231362) [7]. Here, we add an FG-associated signal in AfAs (rs2011766), bringing to five the total number of apparently independent signals at KCNQ1 associated with glycaemia in humans. SNP rs2011766 is intronic and is in very weak LD with the other SNPs in the 1,000 Genomes Pilot 1 reference panel: rs231362 vs rs2011766, r 2  = 0.06; rs2011766 vs rs2237895 r 2  = 0.03; and rs2011766 vs rs2237892 r 2  = 0.02. However, rs2011766 does not appear to be a type-2-diabetes-risk locus, with an odds ratio for type 2 diabetes in the DIAGRAM discovery meta-analysis of 1.04 (p = 0.19) [7]. Multi-ethnic allelic heterogeneity in association with FG or type 2 diabetes has also been reported for the C2CD4A/B locus [44]. Other loci at which cross-ancestry fine-mapping has confirmed the index SNP to be the same in EuA populations and AfAs are TCF7L2 [10, 13, 45, 46], PPARG, IGF2BP2, JAZF1 [12] and FTO [47].

Type 2 diabetes disproportionately affects AfAs. This differential risk may be due partly to genetic factors [5]. Most AfAs have differential mixed ancestry from Western Africa and Europe [48]. AfA genetic architectures are substantially different and generally have lower LD compared with other US populations [5], which is theoretically able to assist fine-mapping of causal variants. Our observation of large differences in risk-allele frequency and high Fst between AfAs and EuA populations at some SNPs is consistent with studies of other complex-disease-associated SNPs [4, 49]. Differences in risk-allele frequencies or in ability to increase FG (ancestral vs derived allele) did not seem to have strong impact on the association as supported by many of the loci showing transferability with similar effect sizes in AfAs [16]; of 18 glycaemic loci, only SNPs at FADS1 had evidence of population differentiation and significant association with FG at weakly correlated EuA and AfA signals. A limitation of our study is that lower risk-allele frequencies at some loci may be due to insufficient study power. In addition, lack of transferability may arise from the confounding effects of social and environmental risk exposures that vary across populations and that have not been considered here [50, 51, 52].

Conclusions

Cross-ancestry studies can show new population-specific risk variants and confirm shared risk variants. Allele frequencies and LD variation influence power and the ability to resolve the signal to a smaller region containing possible functional variants. In the AfA CARe cohort of ~6,000 person, there was substantial allele frequency variation but modest EuA index SNP transferability. However, most glycaemia-associated loci showed at least nominal evidence of similar association in AfA and EuA groups after accounting for LD differences. Four FG loci (G6PC2, GCK, MTNR1B and FADS1) and three type 2 diabetes-FG loci (KCNQ1, ADAMTS9 and CDC123/CAMK1D) were convincingly transferable from EuA populations to AfAs. Cross-ancestry analyses that leverage varying LD across ancestral groups can identify signals that may aid fine-mapping. In the near future one can expect that new or confirmatory variants associated with type 2 diabetes and diabetes-related quantitative traits are likely to be revealed by more extensive sequencing and genome-wide analysis ongoing in other ancestral population groups, further illuminating the genetic architecture of type 2 diabetes risk and glycaemic regulation in all humans.

Notes

Acknowledgements

The authors wish to acknowledge the support of the NHLBI and the contributions of the research institutions, study investigators and field staff involved, and the study participants of ARIC, CARDIA, CFS, JHS and MESA in creating the CARe for biomedical research (CARe; http://public.nhlbi.nih.gov/GeneticsGenomics/home/care.aspx). The following five parent studies have contributed parent study data, ancillary study data and DNA samples through the Massachusetts Institute of Technology - Broad Institute (N01-HC-65226) to create this genotype/phenotype database for wide dissemination to the biomedical research community: ARIC; CFS; JHS; CARDIA; and MESA.

Funding

This work was supported by NIDDK R01 2 DK078616 and NIDDK K24 DK080140 to J. B. Meigs, a Doris Duke Charitable Foundation Clinical Scientist Development Award to J. C. Florez, RO1 DK066358 to D. W. Bowden, 5RC1HL099911-02 5UC2HL103010-02 and the NHLBI MESA contract N01HC95165-21-0-1 to S. S. Rich.

ARIC is carried out as a collaborative study supported by NHBLI contracts N01-HC-55015, N01-HC-55016, N01-HC-55018, N01-HC-55019, N01-HC-55020, N01-HC-55021, and N01-HC-55022. CFS: Case Western Reserve University (NIH HL 46380, M01RR00080).

JHS is carried out as a collaborative study supported by Jackson State University (N01-HC-95170), University of Mississippi (N01-HC-95171), Tougaloo College (N01-HC-95172);

MESA is conducted and supported by the NHLBI in collaboration with MESA investigators. Support is provided by grants and contracts N01-HD-95159 through N01-HC-95169, RR-024156, and R01-HL-071205.

CARDIA is carried out as a collaborative study supported by University of Alabama at Birmingham (N01-HC-48047), University of Minnesota (N01-HC-48048), Northwestern University (N01-HC-48049), Kaiser Foundation Research Institute (N01-HC-48050), University of Alabama at Birmingham (N01-HC-95095), Tufts-New England Medical Center (N01-HC-45204), Wake Forest University (N01-HC-45205), Harbor-UCLA Research and Education Institute (N01-HC-05187), University of California Irvine (N01-HC-45134, N01-HC-95100).

Contribution statement

The experiments were conceived and designed by SR, WHK, DS, JCF, JIR, JGW, DWB and JBM. C-TL, DR, JD, MCYN, SRP, WHK and KLL analysed the data. The paper was written by: C-TL, MCYN, JCF, JD and JBM. SR, WHK, DS, JCF, JIR, JGW, DWB and JBM recruited individuals. Analysis and interpretation of data and critical review of manuscript were performed by C-TL, MCYN, DR, AA, SJB, EB, IB, BC, YDIC, LD, MF, MOG, SFAG, XG, TH, EK, JRK, YL, KLL, KM, JAN, JSP, SRP, ER, LR-T, SSR, CNR, DS, DS, MS, JMZ, SR, WHK, DS, JCF, JIR, JD, JGW, DWB and JBM.

JBM and C-TL are the guarantors of this work, had full access to all the data and take full responsibility for the integrity of data and the accuracy of data analysis. All authors have approved the final version of manuscript.

Duality of interest

The authors declare that there is no duality of interest associated with this manuscript.

Supplementary material

125_2012_2656_MOESM1_ESM.pdf (12 kb)
Supplementary Figure 1 Power analysis. The sample size required vs. minor allele frequency is plotted to attain 80% power with a type I error rate of 4.08 × 10−5 for various effect sizes. Three effect sizes, which represent the 25th (blue), 50th (red), and 75th (green) percentiles for previously reported effect sizes in European ancestry samples [8] were evaluated, with 4.08 × 10−5 chosen to maintain the overall type I error rate at 0.05 considering the total number of independent SNPs tested at each FG or FI-associated locus (PDF 12 kb)
125_2012_2656_MOESM2_ESM.pdf (795 kb)
Supplementary Figure 2 The 500 kb regional association plots centered at the index SNPs identified from European ancestry samples at ADAMTS9, TP53INP1, KCNQ1, MTNR1B, CDC123/CAMKI, CDKN2A/2B, IGF2BP2, CDKAL1, and DGKB. The x-axis denotes genomic position and the y-axis denotes the −log(p value) for each SNP. The two larger sized data points represent the best SNPs within the region in the African American CARe sample and the index SNPs (locating the center of the region) previously reported from European ancestry samples. The color of each data point indicates its LD value (r 2) with the index SNP based on HapMap 2 YRI. The blue line represents the recombination rate. The green line shows the varLD score at each SNP and is highlighted with dark brown if the varLD score is greater or equal to 95th percentile of the genome-wide varLD score, comparing LD information between YRI and CEU HapMap 2 samples [35]. (a) ADAMTS9 region: rs4607103 is the index SNP with p value of 6.85 × 10−3 and the best SNP rs4688504 with p value 3.82 × 10−4. b DGKB/TMEM195 region: rs2191349 is the index SNP with p value of 1.02 × 10−1 and the best SNP rs1732749 with p value 9.66 × 10−4 (c) TP53INP1 region: rs896854 is the index SNP with p value of 3.40 × 10−2 and the best SNP rs509594 with p value 2.54 × 10−3 (d) CDKN2A/CDKN2B region: rs10811661 is the index SNP with p value of 1.68 × 10−1 and the best SNP rs10965319 with p value 2.86 × 10−3 (e) CDC123/CAMK1D region: rs12779790 is the index SNP with p value of 3.4 × 10−1 and the best SNP rs7074893with p value 8.29 × 10−5 (f) MTNR1B region: rs1387153 is the index SNP with p value of 4.38 × 10−1 and the best SNP rs10830963 with p value 8.51 × 10−9 (g) KCNQ1 region: rs231362 is the index SNP with p value of 1.43 × 10−1 and the best SNP rs2011766 with p value (PDF 794 kb)
125_2012_2656_MOESM3_ESM.pdf (20 kb)
ESM Table 1 PDF 20 kb
125_2012_2656_MOESM4_ESM.pdf (8 kb)
ESM Table 2 PDF 8 kb
125_2012_2656_MOESM5_ESM.pdf (11 kb)
ESM Table 3 PDF 10 kb

References

  1. 1.
    Cowie CC, Rust KF, Ford ES et al (2009) Full accounting of diabetes and pre-diabetes in the U.S. population in 1988–1994 and 2005–2006. Diabetes Care 32:287–294PubMedCrossRefGoogle Scholar
  2. 2.
    Mokdad AH, Bowman BA, Ford ES et al (2001) The continuing epidemics of obesity and diabetes in the United States. JAMA 286:1195–1200PubMedCrossRefGoogle Scholar
  3. 3.
    Mokdad AH, Ford ES, Bowman BA et al (2003) Prevalence of obesity, diabetes, and obesity-related health risk factors, 2001. JAMA 289:76–79PubMedCrossRefGoogle Scholar
  4. 4.
    Myles S, Davison D, Barrett J et al (2008) Worldwide population differentiation at disease-associated SNPs. BMC Med Genome 1:22CrossRefGoogle Scholar
  5. 5.
    Guthery SL, Salisbury BA, Pungliya MS et al (2007) The structure of common genetic variation in United States populations. Am J Hum Genet 81:1221–1231PubMedCrossRefGoogle Scholar
  6. 6.
    Adeyemo A, Rotimi C (2010) Genetic variants associated with complex human diseases show wide variation across multiple populations. Pub Health Genome 13:72–79CrossRefGoogle Scholar
  7. 7.
    Voight BF, Scott LJ, Steinthorsdottir V et al (2010) Twelve type 2 diabetes susceptibility loci identified through large-scale association analysis. Nat Genet 42:579–589PubMedCrossRefGoogle Scholar
  8. 8.
    Dupuis J, Langenberg C, Prokopenko I et al (2010) New genetic loci implicated in fasting glucose homeostasis and their impact on type 2 diabetes risk. Nat Genet 42:105–116PubMedCrossRefGoogle Scholar
  9. 9.
    Yang Q, Liu T, Shrader P, Yesupriya A et al (2010) Racial/ethnic differences in association of fasting glucose-associated genomic loci with fasting glucose, HOMA-B, and impaired fasting glucose in the U.S. adult population. Diabetes Care 33:2370–2377PubMedCrossRefGoogle Scholar
  10. 10.
    Palmer ND, Hester JM, An SS et al (2011) Resequencing and analysis of variation in the TCF7L2 gene in African Americans suggests that SNP rs7903146 is the causal diabetes susceptibility variant. Diabetes 60:662–668PubMedCrossRefGoogle Scholar
  11. 11.
    Casto AM, Feldman MW, Casto AM et al (2011) Does selection affect unlinked SNPs with shared trait associations? PLoS Genet 7:e1001266PubMedCrossRefGoogle Scholar
  12. 12.
    Waters KM, Stram DO, Hassanein MT et al (2010) Consistent association of type 2 diabetes risk variants found in Europeans in diverse racial and ethnic groups. PLoS Genet 6:e1001078PubMedCrossRefGoogle Scholar
  13. 13.
    Lewis JP, Palmer ND, Hicks PJ et al (2008) Association analysis in African Americans of European-derived type 2 diabetes single nucleotide polymorphisms from whole-genome association studies. Diabetes 57:2220–2225PubMedCrossRefGoogle Scholar
  14. 14.
    Sale MM, Smith SG, Mychaleckyj JC et al (2007) Variants of the transcription factor 7-like 2 (TCF7L2) gene are associated with type 2 diabetes in an African-American population enriched for nephropathy. Diabetes 56:2638–2642PubMedCrossRefGoogle Scholar
  15. 15.
    Lettre G, Palmer CD, Young T et al (2011) Genome-wide association study of coronary heart disease and its risk factors in 8,090 African Americans: the NHLBI CARe Project. PLoS Genet 7:e1001300PubMedCrossRefGoogle Scholar
  16. 16.
    Ioannidis JP, Ntzani EE, Trikalinos TA et al (2004) ‘Racial’ differences in genetic effects for complex diseases. Nat Genet 36:1312–1318PubMedCrossRefGoogle Scholar
  17. 17.
    Wilson PW, Meigs JB, Sullivan L et al (2007) Prediction of incident diabetes mellitus in middle-aged adults: the Framingham Offspring Study. Arch Intern Med 167:1068–1074PubMedCrossRefGoogle Scholar
  18. 18.
    Rasmussen-Torvik LJ, Alonso A, Li M et al (2010) Impact of repeated measures and sample selection on genome-wide association studies of fasting glucose. Genet Epidemiol 34:665–673PubMedCrossRefGoogle Scholar
  19. 19.
    Ramos E, Chen G, Shriner D et al (2011) Replication of genome-wide association studies (GWAS) loci for fasting plasma glucose in African-Americans. Diabetologia 54:783–788PubMedCrossRefGoogle Scholar
  20. 20.
    The ARIC Investigators (1989) The Atherosclerosis Risk in Communities (ARIC) Study: design and objectives. The ARIC investigators. Am J Epidemiol 129:687–702Google Scholar
  21. 21.
    Friedman GD, Cutter GR, Donahue RP et al (1988) CARDIA: study design, recruitment, and some characteristics of the examined subjects. J Clin Epidemiol 41:1105–1116PubMedCrossRefGoogle Scholar
  22. 22.
    Buxbaum SG, Elston RC, Tishler PV et al (2002) Genetics of the apnea hypopnea index in Caucasians and African Americans: I. Segregation analysis. Genet Epidemiol 22:243–253PubMedCrossRefGoogle Scholar
  23. 23.
    Taylor HA Jr, Wilson JG, Jones DW et al (2005) Toward resolution of cardiovascular health disparities in African Americans: design and methods of the Jackson Heart Study. Ethn Dis 15(S6):4–17Google Scholar
  24. 24.
    Fuqua SR, Wyatt SB, Andrew ME et al (2005) Recruiting African-American research participation in the Jackson Heart Study: methods, response rates, and sample description. Ethn Dis 15(S6):18–29Google Scholar
  25. 25.
    Wilson JG, Rotimi CN, Ekunwe L et al (2005) Study design for genetic analysis in the Jackson Heart Study. Ethn Dis 15(S6):30–37Google Scholar
  26. 26.
    Taylor HA Jr (2005) The Jackson heart study: an overview. Ethn Dis 15(S6):1–3Google Scholar
  27. 27.
    Bild DE, Bluemke DA, Burke GL et al (2002) Multi-ethnic study of atherosclerosis: objectives and design. Am J Epidemiol 156:871–881PubMedCrossRefGoogle Scholar
  28. 28.
    Voight BF, Kudaravalli S, Wen X et al (2006) A map of recent positive selection in the human genome. PLoS Biol 4:e72PubMedCrossRefGoogle Scholar
  29. 29.
    Ong RT, Teo YY, Ong RT-H et al (2010) varLD: a program for quantifying variation in linkage disequilibrium patterns between populations. Bioinformatics 26:1269–1270PubMedCrossRefGoogle Scholar
  30. 30.
    Matthews DR, Hosker JP, Rudenski AS et al (1985) Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 28:412–419PubMedCrossRefGoogle Scholar
  31. 31.
    Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta-analysis of genomewide association scans. Bioinformatics 26:2190–1Google Scholar
  32. 32.
    Zeggini E, Scott LJ, Saxena R et al (2008) Meta-analysis of genome-wide association data and large-scale replication identifies additional susceptibility loci for type 2 diabetes. Nat Genet 40:638–645PubMedCrossRefGoogle Scholar
  33. 33.
    Zeggini E, Weedon MN, Lindgren CM et al (2007) Replication of genome-wide association signals in UK samples reveals risk loci for type 2 diabetes. Science 316:1336–1341PubMedCrossRefGoogle Scholar
  34. 34.
    Scott LJ, Mohlke KL, Bonnycastle LL et al (2007) A genome-wide association study of type 2 diabetes in Finns detects multiple susceptibility variants. Science 316:1341–1345PubMedCrossRefGoogle Scholar
  35. 35.
    Saxena R, Voight BF, Lyssenko V et al (2007) Genome-wide association analysis identifies loci for type 2 diabetes and triglyceride levels. Science 316:1331–1336PubMedCrossRefGoogle Scholar
  36. 36.
    Sladek R, Rocheleau G, Rung J et al (2007) A genome-wide association study identifies novel risk loci for type 2 diabetes. Nature 445:881–885PubMedCrossRefGoogle Scholar
  37. 37.
    Li J, Ji L (2005) Adjusting multiple testing in multilocus analyses using the eigenvalues of a correlation matrix. Heredity 95:221–227PubMedCrossRefGoogle Scholar
  38. 38.
    Bouatia-Naji N, Bonnefond A, Baerenwald DA et al (2010) Genetic and functional assessment of the role of the rs13431652-A and rs573225-A alleles in the G6PC2 promoter that are strongly associated with elevated fasting glucose levels. Diabetes 59:2662–2671PubMedCrossRefGoogle Scholar
  39. 39.
    Yasuda K, Miyake K, Horikawa Y et al (2008) Variants in KCNQ1 are associated with susceptibility to type 2 diabetes mellitus. Nat Genet 40:1092–1097PubMedCrossRefGoogle Scholar
  40. 40.
    Unoki H, Takahashi A, Kawaguchi T et al (2008) SNPs in KCNQ1 are associated with susceptibility to type 2 diabetes in East Asian and European populations. Nat Genet 40:1098–1102PubMedCrossRefGoogle Scholar
  41. 41.
    Sim X, Ong RT, Suo C et al (2011) Transferability of type 2 diabetes implicated loci in multi-ethnic cohorts from Southeast Asia. PLoS Genet 7:e1001363PubMedCrossRefGoogle Scholar
  42. 42.
    Hu C, Zhang R, Wang C et al (2010) Variants from GIPR, TCF7L2, DGKB, MADD, CRY2, GLIS3, PROX1, SLC30A8 and IGF1 are associated with glucose metabolism in the Chinese. PLoS One 5:e15542PubMedCrossRefGoogle Scholar
  43. 43.
    Ng MC, Tam CH, So WY et al (2010) Implication of genetic variants near NEGR1, SEC16B, TMEM18, ETV5/DGKG, GNPDA2, LIN7C/BDNF, MTCH2, BCDIN3D/FAIM2, SH2B1, FTO, MC4R, and KCTD15 with obesity and type 2 diabetes in 7705 Chinese. J Clin Endocrinol Metab 95:2418–2425PubMedCrossRefGoogle Scholar
  44. 44.
    Yamauchi T, Hara K, Maeda S et al (2010) A genome-wide association study in the Japanese population identifies susceptibility loci for type 2 diabetes at UBE2E2 and C2CD4A-C2CD4B. Nat Genet 42:864–868PubMedCrossRefGoogle Scholar
  45. 45.
    Yan Y, North KE, Ballantyne CM et al (2009) Transcription factor 7-like 2 (TCF7L2) polymorphism and context-specific risk of type 2 diabetes in African American and Caucasian adults: the Atherosclerosis Risk in Communities study. Diabetes 58:285–289PubMedCrossRefGoogle Scholar
  46. 46.
    Dabelea D, Dolan LM, D’Agostino R Jr et al (2011) Association testing of TCF7L2 polymorphisms with type 2 diabetes in multi-ethnic youth. Diabetologia 54:535–539PubMedCrossRefGoogle Scholar
  47. 47.
    Grant SF, Li M, Bradfield JP et al (2008) Association analysis of the FTO gene with obesity in children of Caucasian and African ancestry reveals a common tagging SNP. PLoS One 3:e1746PubMedCrossRefGoogle Scholar
  48. 48.
    Tishkoff SA, Reed FA, Friedlaender FR et al (2009) The genetic structure and history of Africans and African Americans. Science 324:1035–1044PubMedCrossRefGoogle Scholar
  49. 49.
    Myles S, Tang K, Somel M et al (2008) Identification and analysis of genomic regions with large between-population differentiation in humans. Ann Hum Genet 72:99–110PubMedGoogle Scholar
  50. 50.
    Martinez-Marignac VL, Valladares A, Cameron E et al (2007) Admixture in Mexico City: implications for admixture mapping of type 2 diabetes genetic risk factors. Hum Genet 120:807–819PubMedCrossRefGoogle Scholar
  51. 51.
    Florez JC, Price AL, Campbell D et al (2009) Strong association of socioeconomic status with genetic ancestry in Latinos: implications for admixture studies of type 2 diabetes. Diabetologia 52:1528–1536PubMedCrossRefGoogle Scholar
  52. 52.
    McCarthy MI (2008) Casting a wider net for diabetes susceptibility genes. Nat Genet 40:1039–1040PubMedCrossRefGoogle Scholar
  53. 53.
    Rung J, Cauchi S, Albrechtsen A et al (2009) Genetic variant near IRS1 is associated with type 2 diabetes, insulin resistance and hyperinsulinemia. Nat Genet 41:1110–1115Google Scholar
  54. 54.
    Altshuler D, Hirschhorn JN, Klannemark M et al (2000) The common PPARgamma Pro12Ala polymorphism is associated with decreased risk of type 2 diabetes. Nat Genet 26:76–80Google Scholar
  55. 55.
    Diabetes Genetics Initiative of Broad Institute of Harvard and MIT, Lund University, and Novartis Institutes of BioMedical Research (2007) Genome-wide association analysis identifies loci for type 2 diabetes and triglyceride levels. Science 316:1331–1336Google Scholar
  56. 56.
    Gloyn AL, Weedon MN, Owen KR et al (2003) Large-scale association studies of variants in genes encoding the pancreatic beta-cell KATP channel subunits Kir6.2 (KCNJ11) and SUR1 (ABCC8) confirm that the KCNJ11 E23K variant is associated with type 2 diabetes. Diabetes 52:568–572Google Scholar
  57. 57.
    Winckler W, Weedon MN, Graham RR et al (2007) Evaluation of common variants in the six known maturity-onset diabetes of the young (MODY) genes for association with type 2 diabetes. Diabetes 56:685–693Google Scholar
  58. 58.
    Gudmundsson J, Sulem P, Steinthorsdottir V et al (2007) Two variants on chromosome 17 confer prostate cancer risk, and the one in TCF2 protects against type 2 diabetes. Nat Genet 39:977–983Google Scholar

Copyright information

© Springer-Verlag 2012

Authors and Affiliations

  • C.  -T.  Liu
    • 1
  • M. C. Y. Ng
    • 2
  • D. Rybin
    • 3
  • A. Adeyemo
    • 4
  • S. J. Bielinski
    • 5
  • E. Boerwinkle
    • 6
  • I. Borecki
    • 7
  • B. Cade
    • 8
  • Y. D. I. Chen
    • 9
  • L. Djousse
    • 8
    • 10
    • 11
  • M. Fornage
    • 6
  • M. O. Goodarzi
    • 9
  • S. F. A. Grant
    • 12
  • X. Guo
    • 9
  • T. Harris
    • 13
  • E. Kabagambe
    • 14
  • J. R. Kizer
    • 15
  • Y. Liu
    • 2
    • 16
  • K. L. Lunetta
    • 1
    • 17
  • K. Mukamal
    • 10
  • J. A. Nettleton
    • 6
  • J. S. Pankow
    • 18
  • S. R. Patel
    • 8
  • E. Ramos
    • 4
  • L. Rasmussen-Torvik
    • 19
  • S. S. Rich
    • 20
  • C. N. Rotimi
    • 4
  • D. Sarpong
    • 21
  • D. Shriner
    • 4
  • M. Sims
    • 22
  • J. M. Zmuda
    • 23
  • S. Redline
    • 8
  • W. H. Kao
    • 24
  • D. Siscovick
    • 25
  • J. C. Florez
    • 10
    • 26
    • 27
  • J. I. Rotter
    • 9
  • J. Dupuis
    • 1
    • 17
  • J. G. Wilson
    • 22
  • D. W. Bowden
    • 2
    • 28
  • J. B. Meigs
    • 10
    • 29
    Email author
  1. 1.Department of BiostatisticsBoston University School of Public HealthBostonUSA
  2. 2.Center for Genomics and Personalized Medicine Research, Center for Diabetes ResearchWake Forest University School of MedicineWinston-SalemUSA
  3. 3.Boston University Data Coordinating CenterBostonUSA
  4. 4.National Human Genome Research InstituteBethesdaUSA
  5. 5.Mayo ClinicRochesterUSA
  6. 6.University of Texas Health Science Center at HoustonHoustonUSA
  7. 7.Washington UniversitySt LouisUSA
  8. 8.Brigham and Women’s HospitalBostonUSA
  9. 9.Cedars Sinai Medical CenterLos AngelesUSA
  10. 10.Department of MedicineHarvard Medical SchoolBostonUSA
  11. 11.Boston VA Healthcare SystemBostonUSA
  12. 12.Children’s Hospital of PhiladelphiaPhiladelphiaUSA
  13. 13.National Institute on AgingBethesdaUSA
  14. 14.University of AlabamaBirminghamUSA
  15. 15.Cornell UniversityNew YorkUSA
  16. 16.Department of Epidemiology and PreventionWake Forest UniversityWinston-SalemUSA
  17. 17.National Heart, Lung, and Blood Institute’s Framingham Heart StudyFraminghamUSA
  18. 18.University of MinnesotaMinneapolisUSA
  19. 19.Northwestern UniversityChicagoUSA
  20. 20.University of VirginiaCharlottesvilleUSA
  21. 21.Jackson State UniversityJacksonUSA
  22. 22.University of Mississippi Medical CenterJacksonUSA
  23. 23.University of Pittsburgh, Graduate School of Public HealthPittsburghUSA
  24. 24.Johns Hopkins UniversityBaltimoreUSA
  25. 25.University of WashingtonSeattleUSA
  26. 26.Diabetes Unit and Center for Human Genetic ResearchMassachusetts General HospitalBostonUSA
  27. 27.Program in Medical and Population Genetics, Broad InstituteCambridgeUSA
  28. 28.Departments of Biochemistry and Internal MedicineWake Forest University School of MedicineWinston-SalemUSA
  29. 29.General Medicine DivisionMassachusetts General HospitalBostonUSA

Personalised recommendations