Genome-wide association study identifies African-ancestry specific variants for metabolic syndrome

Abstract

The metabolic syndrome (MetS) is a constellation of metabolic disorders that increase the risk of developing several diseases including type 2 diabetes and cardiovascular diseases. Although genome-wide association studies (GWAS) have successfully identified variants associated with individual traits comprising MetS, the genetic basis and pathophysiological mechanisms underlying the clustering of these traits remain unclear. We conducted GWAS of MetS in 1,427 Africans from Ghana and Nigeria followed by replication testing and meta-analysis in another continental African sample from Kenya. Further replication testing was performed in an African American sample from the Atherosclerosis Risk in Communities (ARIC) study. We found two African-ancestry specific variants that were significantly associated with MetS: SNP rs73989312[A] near CA10 that conferred increased risk (P=3.86x10⁻⁸, OR=6.80) and SNP rs77244975[C] in CTNNA3 that conferred protection against MetS (P=1.63x10⁻⁸, OR=0.15). Given the exclusive expression of CA10 in the brain, our CA10 finding strengthens previously reported link between brain function and MetS. We also identified two variants that are not African specific: rs76822696[A] near RALYL associated with increased MetS risk (P=7.37x10⁻⁹, OR=1.59) and rs7964157[T] near KSR2 associated with reduced MetS risk (P=4.52x10⁻⁸, P_meta=7.82x10⁻⁹, OR=0.53). The KSR2 locus displayed pleiotropic associations with triglyceride and measures of blood pressure. Rare KSR2 mutations have been reported to be associated with early onset obesity and insulin resistance. Finally, we replicated the LPL and CETP loci previously found to be associated with MetS in Europeans. These findings provide novel insights into the genetics of MetS in Africans and demonstrate the utility of conducting trans-ethnic disease gene mapping studies for testing the cosmopolitan significance of GWAS signals of cardio-metabolic traits.

1. Introduction

The metabolic syndrome (MetS) manifests as a clustering of risk factors including abdominal obesity, atherogenic dyslipidemia, elevated blood pressure, insulin resistance, and prothrombotic and proinflammatory states [1, 2]. Individuals with MetS have at least a fivefold increased risk of developing type 2 diabetes (T2D), a twofold increased risk of cardiovascular diseases [3], and increased susceptibility to several other disorders including fatty liver disease [4], sleep apnea [5], and some forms of cancer [6].

Recent studies suggest that complex networks of metabolic pathways modulated by interacting genetic and environmental factors underlie MetS [7, 8]. Early evidence for the potential contribution of genetics to MetS was provided in a seminal study of twin pairs with 31.6% of monozygotic twin pairs compared to 6.3% of dizygotic pairs displaying concordance for clustering of three MetS components (hypertension, diabetes, and obesity) [9]. Heritability estimates for MetS vary, with estimates as low as 13% in the Dutch [10], 24% in Caribbean-Hispanic families [10, 11], and as high as 48% in Omani families [12]. Earlier genome-wide linkage studies reported links between MetS and several chromosomal regions including 1p34.1, 10p11.2 and 19q13.4 [13], 1q21–q25 [14], and 1q23–q31 [15]. Candidate gene and whole exome sequencing studies have also identified mutations in patients with rare forms of MetS [16, 17]. Meta-analyses of genome-wide association studies (GWAS) in Europeans and Asians have reported associations between MetS and variants in or near several loci including ZPR1, BUD13, APOA5, LPL, and CETP genes [18–20].

In contrast to the growing success in the identification of variants associated with the individual components of MetS, little progress has been made in the identification of variants underlying the syndromic clustering of the component traits of MetS and variants with pleiotropic effect that may shed light on dysregulated pathways in MetS [21, 22]. Furthermore, the prevalence of MetS shows ethnic disparity in individuals of African descent. For example, analysis of the US National Health and Nutrition Survey (NHANES) serial data from 1999–2000 to 2009–2010 revealed modest decline in prevalence of MetS in Caucasians (25.6% to 21.8%) but a slight increase in African-Americans (22.0% to 22.7%) [23, 24]. Paradoxically, the high prevalence of hypertension and diabetes in African-Americans contrasts with the observed low prevalence of high triglyceride levels [25]. Low prevalence of high triglyceride levels is also observed in west Africans, the ancestral populations of African Americans despite dietary and other differences between the two groups. These observed ethnic disparities in the burden of MetS [25] and other cardiometabolic traits [26] persists even after adjustment for modifiable risk factors, implying the role of background genetic predisposition.

In this study, we performed a GWAS of MetS in continental Africans enrolled from Ghana and Nigeria (AF1), and replication testing and meta-analysis with another continental African sample from Kenya (AF2) using ~15 million directly genotyped and imputed single nucleotide polymorphisms (SNPs). Further replication was tested in an African American sample from the Atherosclerosis Risk in Communities (ARIC) study. We also performed a GWAS of MetS in a subset of the samples in the tails of the continuous metabolic syndrome risk scores (cMetS) derived from the sum of standardized residuals of MetS component traits.

2. Materials and methods

2.1. Study samples

Individuals included in this study were participants enrolled in the Africa America Diabetes Mellitus (AADM) study with centers in Ghana and Nigeria (AF1) and Kenya (AF2) [27]. Although the AADM study has been ongoing for over a decade, the majority of participants included in the present analysis was recruited in the year 2008. The study populations, data collection procedures, and ethical processes have been described in detail elsewhere [27, 28].

For developing continuous metabolic syndrome scores and testing their predictive accuracy, we analyzed all 4,820 individuals in the cohorts with non-missing phenotype values (4,023 AF1 and 797 AF2). The discovery genome-wide association analysis was done in 1,427 AF1. Independent replication of genome-wide significant loci was tested in 174 AF2 and 2,475 African Americans enrolled in the ARIC study.

2.2. MetS phenotypes

Based on the definition of the National Cholesterol Education Program (NCEP) improved threshold, an individual was considered to have MetS if they have the following measures for three or more of the five component traits [1]: waist circumference ≥ 102 cm for men or ≥ 88 cm for women; fasting plasma glucose ≥ 100 mg/dL; plasma triglyceride levels ≥ 150 mg/dL; HDL cholesterol < 40 mg/dL for men or < 50 mg/dL for women; systolic BP ≥ 130 mmHg or diastolic BP ≥ 85 mmHg. In our analysis, cases were individuals with MetS and controls were individuals without MetS. We also developed a continuous metabolic syndrome risk scores (cMetS). Previous studies used different approaches including Z-scores, principal components, and percentile rankings to derive cMetS; the scores obtained by using these methods displayed strong correlation with one another [29]. We developed cMetS using the sum of the standardized scores from the components of MetS. Prior to deriving cMetS, normality of each trait was assessed using descriptive statistics and distribution plots (histograms and scatter plots). Plasma glucose and triglyceride values were log₁₀-transformed because their distributions did not conform to normality. Next, standardized residuals were obtained by regressing each trait on age, sex, and broad ethnicity. The sign of the standardized residuals for HDL was reversed so that higher values indicate greater risk of metabolic syndrome. Finally, all standardized residuals of MetS component traits were summed to form cMetS. The higher the cMetS, the higher the tendency for the components to cluster indicating a higher MetS risk. To evaluate the accuracy of cMetS, we developed another continuous metabolic syndrome score based on principal components analysis of MetS component traits. Previous studies have shown that the power of detecting pleiotropic effects of genetic variants on correlated traits can be enhanced by reducing the dimension of the original traits into a single trait using principal components analysis and subsequent analysis of the top principal component in place of the original traits[30, 31]. We used the loadings of the first principal component (PC1) that explain most of the total trait variance as PC-based continuous metabolic syndrome score (pc1MetS).

2.3. Genotyping, quality control, and imputation

The discovery GWAS was done using 1,637 unrelated AF1 in the AADM study genotyped on the Affymetrix Axiom® PANAFR SNP array. The array was chosen because it offers ≥90% genetic coverage of variants with minor allele frequency (MAF) > 2% of the Yoruba (West African) genome and >85% coverage of the Luhya and Maasai (East African) genome. DNA sample preparation and quality control were done in our laboratory at the National Institutes of Health (NIH). We excluded 210 samples (one duplicated, three sex discordant, and 206 with one or more missing MetS component trait values), leaving 1,427 AF1 for the discovery GWAS. Replication of top signals and meta-analysis were done using 185 unrelated AF2 genotyped on the Affymetrix Axiom® PANAFR SNP array. We excluded 11 samples (10 sex discordant, and one with one or more missing MetS component trait values), leaving 174 AF2 for independent replication and meta-analysis. Additional replication was done using 2,612 African Americans in the ARIC study with ~24 million imputed SNP dosage data accessed through the Database of Genotypes and Phenotypes. We excluded 137 samples with one or more missing MetS component trait values, leaving 2,475 ARIC study individuals for replication.

Of the 2,217,748 SNPs in the raw genotype data of the AADM cohort, we excluded 46,562 SNPs that had a minor allele frequency (MAF) of <1%, 22,509 that were missing in more than 10% of individuals, 10,282 that had a Hardy-Weinberg p-value <1x10⁻⁶, and 61,087 non-autosomal SNPs. The remaining 2,077,552 SNPs that passed quality filers were used as the basis for imputation. Imputation was performed using the 1000 Genomes Consortium phase 1, version 3 cosmopolitan reference using the programs MaCH [32]/MaCH-ADMIX [33]. The resulting imputed dosage data were filtered for imputed allelic dosage frequency < 0.01 and r² < 0.3, yielding ~15 million SNPs for analysis.

We generated principal components (PCs) from multi-dimensional scaling analysis of a pruned set of uncorrelated genome-wide SNPs (i.e., r² threshold of 0.2, and a sliding window of 50 SNPs by skipping 5 SNPs between consecutive windows) as implemented in the program PLINK. A plot of the first three PCs is shown in Fig. S1. To minimize the effect of population structure in the GWAS, all analyses were adjusted for the first three PCs in AF1 and AF2 and the first two PCs in AA. Quantile-quantile plot of the distribution of the test statistic was consistent with the expected distribution under the null hypothesis (Fig. S2), and the genomic control inflation was 1.014 (AF1) and 1.017 (AF2), indicating a lack of population structure. The power of the study at different scenarios is shown in Fig. S3.

2.4. Statistical analyses

Statistical significance of differences between groups was tested using student’s t-test for continuous variables and the chi-square test for discrete variables. The level of significance was set at 0.05.

The ability of the cMetS variable to be discriminative between individuals with and without MetS was evaluated using a receiver operating characteristic (ROC) curve. The area under the curve (AUC) of the ROC curve was estimated using a nonparametric approach [34] to determine the probability that cMetS was higher in individuals with MetS than those without MetS.

The imputed SNP dosage data were analyzed using a logistic regression model with adjustment for age, sex, and the first three PCs as implemented in mach2dat. For association analyses of MetS component traits, in addition to the above covariates, adjustment was done for BMI and the analysis was performed using linear regression models under the additive genetic model as implemented in mach2qtl. Genome-wide significance was set as P-value < 5x10⁻⁸.

Meta-analysis of the discovery and replication datasets was done using the METAL software program [35]. We combined the evidence of association in two sided p-values using a fixed-effects model. Sample size of each SNP was used as a weight, and the sign of the beta value was used as the direction of association of the effect allele. Evidence for the between-study heterogeneity was tested using Cochran’s Q statistic and I², and the P-value of the chi-square was used to declare significance [36].

3. Results

3.1. Characteristics of the study participants

Characteristics of the AF1 subjects included in the discovery GWAS and the AF2 subjects included in the replication testing is presented in Tables S1 & S2. Of the 1,427 AF1, 602 were cases (i.e., had MetS) and 825 were controls (i.e., did not have MetS). Of the 174 AF2, 118 were cases and 56 were controls. On average, subjects with MetS were ~3 years older than those without MetS (Tables S1 & S2); as a result, age was adjusted for in all analyses models. The prevalence of MetS was 35.6% in the overall sample, and was greater in women than men (43.2% vs. 25.0%; P<0.001) (Tables 1 & S3).

Table 1.

Characteristics of subjects included in developing and testing accuracy of the continuous metabolic syndrome score

	All (n=4,820)	Sample AF1 (Ghana and Nigeria) (n=4,023)	Sample AF2 (Kenya) (n=797)	P-valuea
Mean ± SD
Age, year	47.33±15.64	45.73±16.02	55.43±10.33	<0.001
Waist circumference, cm	88.86±13.07	88.53±12.94	90.55±13.61	<0.001
Systolic BP, mmHg	134.13±23.65	133.56±23.76	137.00±22.87	<0.001
Diastolic BP, mmHg	80.05±13.98	79.39±14.27	83.37±11.88	<0.001
Plasma glucose, mg/dL	127.22±78.62	125.69±75.35	134.97±93.05	0.008
HDL cholesterol, mg/dL	46.50±16.92	46.24±16.89	47.83±17.04	0.016
Triglyceride, mg/dL	107.43±73.13	94.24±51.18	173.97±117.59	<0.001
cMetSb	−0.028±3.20	−0.06±3.14	0.14±3.46	0.095
pcMetSc	−0.016±1.00	−0.04±1.00	0.09±1.02	0.001
n (%)
MetSd	1716 (35.6)	1318 (32.8)	398 (49.9)	<0.001
Women	2811 (58.3)	2389 (59.4)	422 (52.9)	<0.001

^aBased on t-test for differences of means and Pearson’s chi-square test statistic for differences of proportions between AF1 (sample from Ghana and Nigeria) and AF2 (sample from Kenya).

^bcMetS: continuous metabolic syndrome score based on sum of standardized components of the metabolic syndrome.

^cpcMetS: continuous metabolic syndrome score based on the first principal component loadings of the metabolic syndrome components in principal components analysis.

^dMetS: the metabolic syndrome (present/absent) based on definition of the National Cholesterol Education Program (NCEP).

Characteristics of the study participants included in the development of cMetS and testing of its predictive accuracy are presented in Tables 1, S3 & S4. Our assessment of cMetS showed that it is a valid predictor of MetS as follows: (i) cMetS for individuals with MetS was significantly higher than for those without MetS (mean difference 3.86, P<0.001) (Table S5); (ii) cMetS was strongly correlated with another continuous metabolic syndrome score metric derived from the first principal component loadings of MetS component traits (Pearson correlation (r) =0.92, Fig. S4); (iii) the AUC of cMetS corresponding to MetS was 0.85 (95% CI 0.84–0.87) (Figs. 1 & S5); and (iv) cMetS was moderately correlated with individual component traits of MetS ranging from a correlation coefficient (r) of 0.40 for fasting plasma glucose to 0.62 for diastolic blood pressure (Table S6).

Fig. 1.

Receiver operating characteristics (ROC) curves of continuous metabolic risk scores corresponding to the metabolic syndrome.

3.2. Genome-wide significant associations

In the discovery sample, two strongly correlated low-frequency variants [rs73989312 (1.64%) and rs73989319 (1.63%)] near CA10 conferred a more than six-fold increased risk of developing MetS (P=3.86x10⁻⁸, OR=6.80; and P=3.97x10⁻⁸; OR=6.85 respectively) (Figs. 2 & S6, Tables 1 & S7). These variants had frequencies of 1.70% and 1.64% respectively in the 1000 Genomes west African ancestry samples (YRI- Yoruba in Ibadan, Nigeria and ASW- African ancestry in Southwest USA), but were absent in all non-African and east African (LWK - Luhya in Webuye, Kenya) samples. In general, we observed that variant frequencies were similar between our samples and those of west African ancestry samples in the 1000 Genomes dataset, validating the genotype accuracy in our datasets (Table S8).

Fig. 2.

Plots of genome-wide association of single nucleotide polymorphisms with the metabolic syndrome. Panels A and B represent plots from analyses of the full data set from the discovery GWAS and meta-analysis, respectively. Panels C and D represent plots from analyses of the two 33% tails of the data set from the discovery GWAS and meta-analysis, respectively. Panels C and D represent plots from analyses of the two 25% tails of the data set from the discovery GWAS and meta-analysis, respectively. Each single nucleotide polymorphism (SNP) is represented by a point, with higher points (higher negative log10 P values), showing more significant association. Points above the blue horizontal line represent SNPs with a P value of less than 10⁻⁵ and points above the red horizontal line indicate SNPs with a P value of less than 5×10⁻⁸.

The meta-analysis of AF1 and AF2 identified six variants with genome-wide significant associations with MetS: rs77244975 in the intron of CTNNA3 (P=1.63x10⁻⁸, OR=0.15, C allele frequency=1.71%) and five strongly correlated (r²>=0.8) common variants (frequencies from 53.61% to 58.54%) near RALYL (lead SNP=rs76822696, P=7.37x10⁻⁹, OR=1.59) (Tables 1 & S9). The CTNNA3 rs77244975[C] variant was present at low frequency (<3%) in the 1000 Genomes African ancestry samples (YRI, ASW, and LWK) but was absent in all non-African samples (Table S8). The rs77244975[C] in CTNNA3 (allele frequency 1.82%) was successfully replicated in African Americans (ARIC study) with directionally consistent significant association with MetS (P=0.011, OR=0.52).

Dichotomous analyses of the extreme tails (⅓ −33.3% and ¼ – 25%) of the empirical distribution of cMetS identified two genome-wide significant associations; the first is an intronic SNP (rs7964157) in KSR2 (P_discovery =4.52x10⁻⁸, P_meta=7.82x10⁻⁹, OR=0.53, frequency=29.73%) and the second is a SNP (rs146816516) near MBNL1 (P_meta=3.51x10⁻⁸, OR=0.22, G allele frequency=3.95%) (Tables 1, S7 & S9). The rs146816516[G] variant was absent in the 1000 Genome’s non-African samples (Table S8). We also identified variants with suggestive genome-wide significant associations (5x10⁻⁸ < P < 9x10⁻⁸) including rs9354671 (P=5.56x10⁻⁸) and rs2061117 (P=6.74x1⁻⁸) near BAI3 and rs149307971 (P=6.66x10⁻⁸) near EDEM1-GRM7 (Table S7).

3.3. MetS-associated variants displaying pleiotropic effects

The GWAS significant MetS variants and neighboring SNPs within ± 100kb were tested for association with the individual traits used to define the MetS. We found several SNPs near or within two genes displaying Bonferroni corrected significant pleiotropic associations: 1) KSR2 SNPs were associated with both triglycerides and measures of blood pressure; 2) MBNL1 SNPs were associated with BMI and measures of blood pressure.

3.4. Replication of European MetS meta-analyses GWAS loci in African ancestry individuals

We used the local replication strategy to test whether three loci reported to be associated with MetS in Europeans were also associated with MetS in our African ancestry samples. The index European associated SNPs along with neighboring (± 100kb) variants were evaluated for replication [18, 19]. Correlated variants (r²≥0.4 in the 1000 Genomes’ CEU samples) in two of the three loci were replicated by five SNPs: LPL (rs294, P=8.1x10⁻³; rs4523270, P=2.2x10⁻²; rs2165558, P=2.3x10⁻²; rs35237252, P=3.3x10⁻²) and CETP (rs4783961, P=1.3x10⁻²). The BUD13-ZPR1-APOA5 locus did not replicate, mainly because the variants showing significant associations in Europeans are very rare in our African-ancestry population dataset (e.g., rs10790162 and rs2266788 have minor allele frequencies of 0.0045 and 0.0028 in our dataset). Directionally consistent and stronger replication P-values were achieved with meta-analysis of the west Africans (AF1) and African Americans (ARIC) (LPL: rs294, P=3.5x10⁻³; CETP: rs4783961, P=3.0x10⁻⁴). Interestingly, we found neighboring SNPs uncorrelated with the index SNPs that had lower P-values: rs306 (LPL, P=8.16x10⁻⁴), rs734104 (APOC3, P=1.52x10⁻⁴), and rs149959798 (NLRC5, P=1.75x10⁻³) (Table S10). Notably, the rs306[A] allele associated with reduced risk of MetS was present in African ancestry populations at a frequency of about 10% but was absent in non-African 1000 Genomes’ samples. In the ENCODE and Roadmap projects, these variants have been shown to be enhancer histone marks: rs306 in left ventricle and adipose cells, and rs734104 in liver, adipose and pancreatic cells.

4. Discussion

This first GWAS of MetS in continental Africans identified novel common and low-frequency variants that conferred increased risk as well as protection against MetS. Two of these variants (CA10 and CTNNA3) are African ancestry specific. The identified variants are functionally plausible given their regulatory functional predictions and their locations in or near genes involved in glucose homeostasis (KSR2), adipocyte differentiation (RALYL), brain function (CA10, EDEM1-GRM7 locus in 3p26–25 & BAI3), and cardiac and skeletal muscle contractions (CTNNA3, MBNL1). KSR2 and MBNL1 variants displayed pleiotropic properties given their association with two or more cardio-metabolic traits. In addition to these novel findings, we replicated the LPL and CETP loci previously found to be associated with MetS in European ancestry individuals providing evidence for the trans-ethnic properties of these variants.

The rs76822696 A variant (RALYL) associated with increased risk of MetS is predicted to alter binding motifs of C/EBPβ and the glucocorticoid receptor that modulate adipocyte differentiation by regulating the transcriptional induction of PPARG [37, 38]. PPARG is a master regulator of adipogenesis and rare variants in PPARG that reduce adipocyte differentiation have been associated with increased risk of T2D, a major consequence of MetS [39]. Our finding parallels the hypothesis that reduced adipocyte differentiation leads to diabetes [40], and further suggests that reduced adipocyte differentiation attributed to a novel common variant in RALYL increases the risk of the metabolic syndrome.

We found a common variant (rs7964157) in KSR2 associated with a 47% reduced risk of developing MetS. KSR2 (Kinase suppressor of Ras 2) is a molecular scaffold of the RAS-MEK-ERK pathway that regulates glucose uptake and glycolysis [41, 42], and promotes activation of the AMPK enzyme that regulates cellular energy homeostasis [43]. Rare KSR2 loss-of-function mutations have been linked with early onset obesity and insulin resistance in humans [44]. These KSR2 mutations impair the RAS-MEK-ERK pathway similar to gain-of-function mutations in DYRK1B that was recently associated with a form of metabolic syndrome characterized by early onset coronary artery disease, central obesity, hypertension, and diabetes [17]. Individuals carrying the KSR2 loss-of-function mutations exhibited childhood hyperphagia, reduced basal metabolic rate, and severe insulin resistance [44]. Recapitulating these metabolic phenotypes observed in humans, mice that are deficient in Ksr2 develop profound obesity [43, 45], glucose intolerance, and insulin resistance in liver, muscle and adipose tissues [46, 47], possibly due to reduced AMPK activation leading to impairment of fatty acid oxidation and enhancement of triglyceride storage [43]. Identification of a novel variant associated with MetS in our study showed that, in addition to the previously identified rare KSR2 mutations [44], common variants in KSR2 contribute to metabolic disorders. Our KSR2 GWAS locus rs7964157 is predicted to alter binding motifs of Maf, a family of transcription factors involved in regulating pancreatic β cell function, and β-cell-specific insulin gene transcription (ENCODE/RoadMap project). MafA knockout mice displayed pancreatic islet abnormalities and glucose intolerance, and developed T2D [48].

An important pleiotropic role of the KSR2 locus is shown by our findings of statistically significant associations of KSR2 variants with MetS component traits (systolic, diastolic blood pressure, and triglycerides) (Table 3) as well as published associations with LDL cholesterol and waist circumference (Tables S13 & S14). Previously, we reported linkage signals at chromosome 12q24, the locus harboring KSR2, in a genome-wide linkage study of T2D in a cohort of 343 affected sibling pairs from west Africa [27]. Other family-based studies have also linked the 12q23–24 locus to obesity [49–51], T2D [52–54], percent body fat [55], and total cholesterol [56]. Genome-wide studies have found suggestive associations (that fell short of genome-wide significance threshold) of KSR2 SNPs with visceral adipose tissue [57], LDL cholesterol [58], and total cholesterol [58] (Table S11). A recent meta-analysis reported a pleiotropic effect of a nearby locus in 12q24.31 (rs12310367 in ZNF664) for lipid and inflammatory traits (triglycerides, HDL cholesterol, and adiponectin) [22]. Taken together, our finding and those of others provide strong evidence in support of the pleiotropic properties of the KSR2-dependent signaling.

Table 3.

Metabolic syndrome-associated loci showing Bonferroni-corrected significant association with individual components of the MetS and related cardiometabolic traits^a

Gene	MetSb GWAS SNP (chr: position)	Replication trait	Replicated SNP	Position	Alleles	A1c freq	P-value	Data set
MBNL1	rs146816516 (chr3:152262078)	Systolic BP	rs76204242	152291018	A,C	0.98	3.74E-05	A
		Body Mass Index	rs116815255	152192393	T,A	0.98	1.29E-05	B
		Body Mass Index	rs145202083	152193084	C,A	0.98	1.32E-05	B
		Body Mass Index	rs150947994	152195940	T,A	0.98	1.25E-05	B
		Body Mass Index	rs114991118	152199365	T,A	0.98	1.16E-05	B
		Body Mass Index	rs184693776	152201330	C,T	0.98	1.07E-05	B
		Body Mass Index	rs114284899	152203823	G,T	0.98	1.32E-05	B
		Body Mass Index	rs192244039	152209252	C,T	0.98	1.17E-05	B
		Body Mass Index	rs183566325	152221527	G,T	0.98	2.66E-05	B
		Body Mass Index	rs6804557	152334938	C,A	0.98	1.44E-05	B
		Body Mass Index	rs6773118	152339917	C,T	0.98	1.05E-05	B
KSR2	rs7964157 (chr12:117982220)	Systolic BP	rs78683574	117980191	T,A	0.89	1.39E-05	A
		Systolic BP	rs7964157	117982220	C,T	0.70	3.56E-05	A
		Diastolic BP	rs7964157	117982220	C,T	0.70	3.56E-05	A
		Systolic BP	rs1501635	117973298	T,C	0.60	2.70E-05	B
		Systolic BP	rs78683574	117980191	T,A	0.89	1.35E-05	B
		Triglycerides	rs11610896	117973915	A,G	0.64	1.70E-05	B
		Triglycerides	rs7964157	117982220	C,T	0.70	9.92E-06	B
		Triglycerides	rs11610896	117973915	A,G	0.64	1.14E-05	C
		Triglycerides	rs7964157	117982220	C,T	0.70	2.06E-05	C
EDEM1-GRM7	rs149307971 (chr3:5854876)	HDL cholesterol	rs1377212	5767686	G,C	0.27	7.20E-06	A
		HDL cholesterol	rs1377212	5767686	G,C	0.27	1.54E-05	B
		HDL cholesterol	rs1377212	5767686	G,C	0.27	1.71E-05	C
		HDL cholesterol	rs116357511	5769812	G,T	0.87	2.16E-05	B
		HDL cholesterol	rs116357511	5769812	G,T	0.87	2.11E-05	C
CTNNA3	rs77244975 (chr10:69207975)	Waist circumference	rs188455459	69273893	C,T	0.98	1.40E-05	B

^aSNPs within 100kb distance from the variants showing genome-wide significant association with MetS were tested for association with components of MetS in the discovery datasets. Here we reported those showing Bonferroni-corrected (0.05/number of SNPs tested in the respective locus) significant association.

^bMetS: the metabolic syndrome based on definition of the National Cholesterol Education Program (NCEP).

^cA1 freq: frequency of the first of the two alleles listed under the ‘Alleles’ column in the West African discovery dataset.

Position: base-pair position in NCBI Build 37 coordinates

A: Full dataset (602 cases from Ghana and Nigeria (AF1), 825 AF1 controls)

B: 33.3% tails of cMetS (402 AF1 cases, 550 AF1 controls)

C: 25% tails of cMetS (310 AF1 cases, 405 AF1 controls)

cMetS: continuous metabolic syndrome score based on the sum of standardized residuals of the metabolic syndrome component traits.

Two Africa-specific GWAS variants we found in the CTNNA3 and MBNL1 genes that modulate cardiac muscle contraction are likely relevant to the understanding of the patho-biology of heart diseases. The alpha-T-catenin protein encoded by CTNNA3 is involved in muscle contraction and is predominantly expressed in the heart [59]. Published GWAS reported suggestive associations between CTNNA3 variants and serum levels of the cardiac function markers alanine aminotransferase/aspartate aminotransferase [60] (Tables S11 & S12). Furthermore, the 10q21 locus harboring CTNNA3 is a common fragile site linked to dilated cardiomyopathy [61]. Similarly, MBNL1 is highly expressed in cardiac muscle [62]. MBNL1 inhibits pre-mRNA expression of cardiac troponin T (cTNT) that regulates contractile function of cardiac muscle [63, 64]. Overexpression of MBNL1 in muscle cells reduces inhibition of insulin receptor (IR) protein [63, 65] leading to enhanced glucose uptake in skeletal muscle [66]. According to the ENCODE project, our GWAS associated variant rs146816516 [G] in MBNL1 is predicted to alter binding motifs of Foxa and STAT transcription factors that regulate glucose homeostasis [67] and adipocyte metabolism [68]. SNP rs116264158 which is in strong LD (r²=0.87) with rs146816516 has been shown to be a promoter/enhancer histone mark in adipose and skeletal muscle cells, in a DNAse hypersensitivity site in pancreatic islets, and is predicted to alter binding motifs of the glucocorticoid receptor protein. SNP rs150505636, which is strongly correlated with rs146816516 (r²=0.87), is a weak enhancer in pancreatic islets (ENCODE and RoadMap).

Our finding that variants near CA10 (Carbonic anhydrase X) substantially increased the risk (OR=6.80) of developing MetS provides evidence in support of the reported pathophysiological link between brain function and metabolic syndrome. CA10 is exclusively expressed in the brain and is considered to have an essential, yet unknown, functional role because it is highly conserved across animal species [69]. The encoded CA10 protein is catalytically inactive due to absence of zinc-binding histidine residues. When the missing histidines are restored in the sequence, the mutated CA10 exhibits high catalytic activity [70]. Most carbonic anhydrases (CAs) are involved in important biosynthetic processes including lipogenesis and gluconeogenesis. Inhibition of mitochondrial CA isozymes using sulfonamide analogues such as the antiepileptic drugs topiramate (TPM) and zonisamide (ZNS) induces weight loss [71–75]. FDA approved chronic weight management drug Qsymia (phentermine and TPM extended-release) leads to weight loss, and improvements in weight-related comorbidities such as improved glycemia, decreased blood pressure, and improved cholesterol levels [76]. The rs73989312 [A] variant that we found to be associated with a substantially increased risk (OR=6.80) of MetS is predicted to alter binding motifs of transcription factor activator protein-2 (AP-2) (ENCODE). The AP-2 transcription factors are involved in the pathogenesis of metabolic diseases through dysregulation of adipocyte function [77] and regulates nervous system and other developmental processes [78]. The obesity-linked Drosophila homolog TfAP-2 is expressed in octopaminergic neurons that regulate satiation and the mouse homolog AP-2β is located in parts of the brain involved in modulating feeding behavior [79].

Our EDEM1-GRM7 and BAI3 findings provide additional evidence for the link between neural connectivity in the brain and MetS. GRM7 is the major excitatory neurotransmitter in the central nervous system, and BAI3 regulates excitatory synapses and is involved in brain angiogenesis. EDEM1 is directly involved in endoplasmic reticulum-associated degradation (ERAD) of misfolded proteins. Dysfunctional ERAD mediates the link between chronic ER stress and impairment of cellular functions that lead to metabolic dysregulations and neurodegenerative diseases [80–82]. The 3p26–25 locus that harbors the EDEM1-GRM7 region has been shown to have pleiotropic effects on obesity, lipids, and blood pressure traits [83]. These SNPs are located 3kb-17kb from the less frequent Africa-specific SNP (rs149307971) that showed association with MetS in our study. In all, our findings open novel research areas for understanding the genetic basis of the roles of the brain in regulating energy metabolism, body fat content, and glucose metabolism and the link between defects in brain function and metabolic disorders [84].

We acknowledge that our replication sample from Africa is small because of the paucity of well phenotyped genome-wide data from continental Africa. Despite this limitation, our study identified novel low frequency and common variants in genes with plausible pathophysiological role in metabolic and cardiovascular diseases. The functional diversity of the proteins encoded by the identified genes including glucose homeostasis, adipocyte differentiation, cardiac and muscle contractions, neuro-endocrine signaling, brain and pancreatic β cell function indicates that the underlying pathology in MetS is a complex network of dysregulated pathways. The identification of African ancestry specific variants influencing MetS is noteworthy for several reasons including providing strong supportive evidence for including multi-ethnic global populations in disease mapping strategies to facilitate discovery of novel variants influencing complex traits. Finally, we showed that a continuous measure of metabolic syndrome (cMetS) can be used to identify individuals at the extremes of metabolic disease risk profile in GWAS, and that cMetS is a valid predictor of MetS in continental Africans as previously observed in other ancestral populations [85–87].

Supplementary data to this article can be found online at Molecular Genetics and Metabolism.

Table 2.

SNPs showing genome-wide significant association with the metabolic syndrome

Variant	Chr	Position	Frequency (%)	Nearest gene	P-value	ORa	SE
Discovery
Full datasetb
rs73989312[A]c	17	50464800	1.64	CA10	3.86E-08	6.80	0.398
rs73989319[C]c	17	50477498	1.63	CA10	3.97E-08	6.85	0.401
Extreme 33.3% tails of cMetSd
rs7964157[T]	12	117982220	29.73	KSR2	4.52E-08	0.53	0.121
Meta-analysis
Full dataset
rs77244975[C]	10	69207975	1.71	CTNNA3	1.63E-08	0.15	0.430
rs76822696[A]	8	84929136	53.61	RALYL	7.37E-09	1.59	0.080
rs16912410[T]	8	84926391	55.42	RALYL	8.00E-09	1.56	0.077
rs62526240[G]	8	84927369	56.22	RALYL	8.94E-09	1.58	0.079
rs55752635[A]	8	84926274	56.46	RALYL	1.16E-08	1.55	0.077
rs188622356[T]	8	84931781	58.54	RALYL	3.28E-08	1.56	0.080
Extreme 33.3% tails of cMetS
rs7964157[T]	12	117982220	29.73	KSR2	7.82E-09	0.52	0.116
Extreme 25% tails of cMetSe
rs146816516[G]	3	152262078	3.95	MBNL1	3.51E-08	0.22	0.363

Chr: chromosome; Position: base-pair position in NCBI Build 37 coordinates; OR: odds ratio; SE: standard error

^aIn the meta-analysis, odds ratio was calculated using the SCHEME STDERR command that weights effect size estimates using the inverse of the corresponding standard errors as implemented in METAL

^bFull dataset (602 cases from Ghana and Nigeria (AF1 sample), 825 AF1 controls)

^cThis has not replicated because the variant allele is absent in east African populations.

^dExtreme 33.3% tails of cMetS (402 AF1 cases, 550 AF1 controls)

^eExtreme 25% tails of cMetS (310 AF1 and 54 cases from Kenya (AF2), 405 AF1 and 32 AF2 controls)

Highlights.

• African-specific CA10 and CTNNA3 variants is associated with metabolic syndrome.

• A common KSR2 variant is associated with a 47% reduced risk for metabolic syndrome.

• KSR2 and MBNL1 loci display pleiotropic effects across two or more traits.

• A continuous metabolic score identifies loci implicated in metabolic profiles.

Abbreviations

cMetS

continuous metabolic syndrome score

GWAS

genome-wide association study

MetS

the metabolic syndrome as defined by the National Cholesterol Education Program

NCEP

the National Cholesterol Education Program

T2D

type 2 diabetes