Evaluation of pleiotropic effects among common genetic loci identified for cardio-metabolic traits in a Korean population

We performed Exome chip genotyping of 14,616 study participants from three population-based cohorts that are comprised in the Korean Genome Epidemiology Study (KoGES). We selected 7981 participants from the Ansung and Ansan population-based cohort (the Korea Association REsource (KARE) project), 3448 participants from the Health Examinee (HEXA) cohort and 3187 participants from Cardio Vascular Disease Association Study (CAVAS). All participants were aged between 40 and 69 years. More detailed descriptions of these three cohorts are available in published articles [11‐13]. All participants provided written informed consent, and this study was approved by the ethical committee of our institute (Korea Centers for Disease Control and Prevention Institutional Review Board).

Blood levels of glycemic traits [fasting plasma glucose (FPG), glycated hemoglobin (HbA1c)] and lipids [TG, total cholesterol (TC), and HDL] were measured in plasma (FPG), whole blood (HbA1c) and serum (lipids), respectively, using standard enzymatic methods from fasting-individuals. If the TG level was <400 mg/dl, the concentration of low-density lipoprotein (LDL) cholesterol was calculated using Friedewald’s formula [14]. Blood pressure (BP) was measured using a standard mercury sphygmomanometer after participants had been in a sitting position for at least 5 min. The mean of two measures (left and right arm) was taken as the BP. If participants were taking antihypertensive agents, BP values were adjusted for analyses by adding 10 mmHg to systolic BP (SBP) and 5 mmHg to diastolic BP (DBP) [7]. Body mass index (BMI) and waist-hip ratio (WHR) were defined as weight (in kilograms) divided by the square of height (in meters) and the ratio of waist to hip circumferences (in centimeters), respectively.

Individuals taking lipid lowering or anti-diabetes medication and individuals with FPG ≥ 126 mg/dl and HbA1c ≥ 6.5 % were excluded from the analyses to obtain only non-diabetic individuals.

The Illumina HumanExome BeadChip is a genotyping array that includes variants represented by ~250,000 single nucleotide polymorphisms (SNPs) selected from exome and whole-genome sequences in ~12,000 individuals [15]. This chip is focused on protein-altering variants (non-synonymous, stop and splice variants) and represents diverse populations and multiple complex traits [16, 17]. More detailed explanations about the contents of the Exome chip can be found at the Exome chip design web page (http://​genome.​sph.​umich.​edu/​wiki/​Exome_​Chip_​Design).

All 14,616 study participants were genotyped using the Illumina HumanExome BeadChip v1.1 (variants = 242,901 SNPs). For genotype calling, each genotype was automatically clustered by the Illumina GenomeStudio v2011.1 software [18]. To improve the accuracy of variant calling, manual re-clustering and visual inspection were performed for genotypes based on the CHARGE clustering method [15]. After genotype calling, variants with completely missing (n = 1431), monomorphic SNPs (n = 162,803), Hardy–Weinberg equilibrium values of P < 1.0 × 10⁻⁶ (n = 1184) or genotype call rates of < 0.95 (n = 11) were excluded for quality control. Samples with a genotype call rate of < 99 % (n = 43), sex discrepancy (n = 51) or cryptic relatedness (n = 487), and withdrawal of participation (n = 7) were also removed for sample quality control.

After genotype and sample quality control, 77,472 SNPs and 14,028 samples (7524 from KARE, 3436 from HEXA and 3068 from CAVAS) were available for further analyses. The number of SNPs with different minor allele frequencies is summarized in Additional file 1: Table S1.

We focused on 157 loci that have been identified previously in large GWAS on lipids with genome-wide significance (P < 5 × 10⁻⁸) [5]. Different lipids can have different best-associated SNPs in the same locus. After selection of the most strongly associated SNPs for each locus, 157 regions were mapped to 129 genes and 135 SNPs [minor allele frequency (MAF) > 0.001, equivalent to a minor allele count ≥ 5 in each cohort].

To test the combined effects of lipid SNPs on other cardio-metabolic traits, we calculated two types of risk score, an unweighted risk score and a weighted risk score, for each SNP and each individual based on 49 lipid-related SNPs that were replicated in this study. The list of SNPs is provided in Additional file 1: Table S3. The unweighted risk score was determined as the total number of risk alleles per individual, and the weighted risk score was calculated by weighting the risk alleles by their estimated effect size based on our data in this study [19]. Associations between lipid risk scores and cardio-metabolic traits were evaluated based on a linear regression model using the R program (version 2.15.3; http://​www.​r-project.​org/​). In this program, a linear model (function “lm” in the “stats” package) is constructed that includes variables such as lipid risk scores and cardio-metabolic traits. For example, when we choose a model with three variables, the script is as follows; > fit <− lm (y ~ x1 + x2 + x3, data = riskscore), > summary (fit). Age, sex and lipids (TG, TC, LDL and HDL) were included in the analyses as covariates.

Single-variant association analysis was conducted between 135 SNPs and 10 cardio-metabolic traits (TG, TC, LDL, HDL, SBP, DBP, FPG, HbA1c, BMI and WHR) using a linear regression model via EPACTS v3.2.4 (http://​genome.​sph.​umich.​edu/​wiki/​EPACTS) and PLINK 1.07 (http://​pngu.​mgh.​harvard.​edu/​~purcell/​plink/​) programs [20]. The main command of association analysis is “--test q.linear” and “--linear” to each program, respectively. To address the issue of multiple testing, a robust threshold of association significance (⍺ = 0.05/135 for each trait, P < 3.7 × 10⁻⁴) was used based on the Bonferroni correction. Phenotypes used in the analyses were approximately normally distributed, and age and sex were incorporated into the analyses as covariates. Meta-analyses of association results from each cohort were performed with the inverse variance method using the METAL program (http://​genome.​sph.​umich.​edu/​wiki/​METAL) [21].

To evaluate and clarify the type of pleiotropy (independent or mediated) for three pleiotropic SNPs located on region 12q24.12, additional analyses were performed including adjustment of lipids (TC, TG, LDL and HDL) for other cardio-metabolic traits to examine whether or not any association occurred independently of lipids effects.

To search for functional interactions among genes, protein–protein interactions were performed using Wiki-Pi (http://​severus.​dbmi.​pitt.​edu/​wiki-pi/​) [22]. This tool provides information about one or both proteins in the interaction based on automatically updated annotations of individual proteins from databases such as Gene Ontology, Kyoto Encyclopedia of Genes and Genomes, REACTOME, Pubmed2ENSEMBL and DrugBank.

In the meta-analyses, 49 loci with new lead SNPs were significantly associated with one of four lipid traits (10 loci associated with TC, 11 with TG, 14 with HDL and 14 with LDL) after correction for multiple testing (P < 3.70 × 10⁻⁴) (Additional file 1: Table S3). Most signals were associated with two or more lipid traits. The overlap of loci associated with different lipid traits is shown in Fig. 1. Among the 49 replicated SNPs, 22 (45 %) were located in exonic regions and, of these, 18 (82 %) were non-synonymous variants. Several non-synonymous variants, which were located on PCSK9, CETP and APOA1 associated with TC or HDL and were novel protein-coding variants that had not been revealed in the same loci in previous studies.

Eighteen out of 135 SNPs were associated with other cardio-metabolic traits such as FPG, HbA1c, BMI, WHR, SBP and DBP after correction for multiple testing (P < 3.70 × 10⁻⁴) (Table 1; Fig. 2). Our findings replicated those of previous GWASs for five genes (GCKR, FADS1 and FADS2 for FPG, HFE for HbA1c and FTO for BMI) and also identified novel associations for 13 genes (three for FPG, four for WHR, three for SBP and three for DBP). Three associations were observed with one or more of four cardio-metabolic traits (FPG, WHR, SBP and DBP) within the genes BRAP, ACAD10 and ALDH2 in the 12q24.12 region (Table 1; Fig. 2). These associations had not been reported in previous GWASs. The three SNPs were composed of two non-synonymous and one intronic variant and had same direction of effect for each trait. All three SNPs showed monomorphic ethnic differences in allele frequency from European and South Asian descent in comparison with allele frequency from our data (Korean individuals, average MAF = 0.161) and East Asian descent (average MAF = 0.175; calculated using data from the 1000 Genomes Project, http://​www.​1000genomes.​org/​) (Additional file 1: Table S4).

The pleiotropic effect of region 12q24.12 was identified for three SNPs (rs3782886 on BRAP, rs11066015 on ACAD10 and rs671 on ALDH2) on lipids and other cardio-metabolic traits (Table 2). These SNPs were associated with three lipid traits (TG, HDL and LDL), SBP, DBP, FPG, BMI and WHR (P < 0.05) and had the same direction of allelic effect for all cardio-metabolic traits (the linkage disequilibrium of these SNPs in our data, r² = 0.988 between rs671 and rs11066015, r² = 0.926 between rs671 and rs3782886) (Additional file 1: Figure S1). To identify the type of pleiotropy, adjusted association analyses were repeatedly performed across the eight cardio-metabolic traits. Mediated analyses showed that the association between rs671 and the three lipid traits (TG, HDL and LDL) remained significant after adjustments for all other cardio-metabolic traits (Table 3; Fig. 3; Additional file 1: Table S5). In addition, the association between rs671 and FPG, SBP, DBP, BMI and WHR remained significant after adjustments for the three lipid traits, suggesting that rs671 on ALDH2 is independently associated with each cardio-metabolic trait (Table 3; Fig. 3; Additional file 1: Table S5). The adjusted associations of rs3782886 and rs11066015 with each cardio-metabolic trait showed the same effects as rs671 (data not shown).

To assess the physical interactions among the three pleiotropic genes in the 12q24.12 region, protein–protein interactions were tested using the Wiki-Pi database (http://​severus.​dbmi.​pitt.​edu/​wiki-pi/​) [22]. There were numerous direct protein–protein interactions for the three genes: 12 interactions for ALDH2, one interaction for ACAD10 and 15 interactions for BRAP. According to this analysis, the three pleiotropic genes were functionally connected by one gene, APP, which encodes amyloid beta (A4) precursor protein (Additional file 1: Figure S2). This protein participated in several biological pathways and showed 1995 interactions with other proteins within Wiki-Pi data (Additional file 1: Table S6). In particular, APP has a role in the cholesterol metabolic process and in blood coagulation (Additional file 1: Table S6).

To examine the cumulative genetic effect of the 49 replicated lipid SNPs on other cardio-metabolic traits such as FPG, SBP, DBP, BMI and WHR, a lipid risk score was calculated for TG, TC, HDL and LDL from different sets of SNPs, and two types of lipid risk score were calculated for each lipid. The weighted risk score for LDL was associated with all five cardio-metabolic traits (P < 0.05) (Table 4). The unweighted risk score of the four lipids was associated with FPG (P < 0.05) (Table 4).

In this study, we have presented lipid-related genetic variants with pleiotropic effects on other cardio-metabolic traits. Recent studies suggested that pleiotropic effects on complex traits may be widespread. Genetic variants associated with two or more traits, such as rs7138803 on FAIM2 that related to obesity, T2D and myocardial infarction, rs8192673 on PGC-1α and rs1801282 on PPAR-γ associated with waist circumference and T2D, were revealed by previous studies [32‐34]. The increased use of genetic information in clinical practice, has underscored the importance of understanding pleiotropy and its implications for genetic testing and personal genomics. A gene or multiple genes in a biological pathway that affect more than one trait present new chances and challenges for drug development. However, most studies have focused only on the detection of pleiotropic effects. Functionally characterizing identified variants through experiments and understanding the mechanisms of shared pathophysiology are necessary to establish causality. In addition, gene-environmental interactions (G × E) could provide insights the role of environmental factors in disease risk and hence to investigate their role as genetic factor modifiers [35]. For example, rs671 on ALDH2 was found to be associated with TG when the interaction between SNP and alcohol consumption was considered [36]. By integrating biological knowledge with genetic pleiotropy and environmental factors, G × E interactions may provide the potential to shed light on biological processes leading to diseases and could explain ‘missing heritability’ for phenotypic variance in GWASs. And also, it may be necessary to search for additional shared genetic patterns on multiple phenotypes in larger sample sizes across diverse ethnic cohorts as well as Korean population.