Genome-wide landscape establishes novel association signals for metabolic traits in the Arab population

Abstract

While the Arabian population has a high prevalence of metabolic disorders, it has not been included in global studies that identify genetic risk loci for metabolic traits. Determining the transferability of such largely Euro-centric established risk loci is essential to transfer the research tools/resources, and drug targets generated by global studies to a broad range of ethnic populations. Further, consideration of populations such as Arabs, that are characterized by consanguinity and a high level of inbreeding, can lead to identification of novel risk loci. We imputed published GWAS data from two Kuwaiti Arab cohorts (n = 1434 and 1298) to the 1000 Genomes Project haplotypes and performed meta-analysis for associations with 13 metabolic traits. We compared the observed association signals with those established for metabolic traits. Our study highlighted 70 variants from 9 different genes, some of which have established links to metabolic disorders. By relaxing the genome-wide significance threshold, we identified 'novel' risk variants from 11 genes for metabolic traits. Many novel risk variant association signals were observed at or borderline to genome-wide significance. Furthermore, 349 previously established variants from 187 genes were validated in our study. Pleiotropic effect of risk variants on multiple metabolic traits were observed. Fine-mapping illuminated rs7838666/CSMD1 rs1864163/CETP and rs112861901/[INTS10,LPL] as candidate causal variants influencing fasting plasma glucose and high-density lipoprotein levels. Computational functional analysis identified a variety of gene regulatory signals around several variants. This study enlarges the population ancestry diversity of available GWAS and elucidates new variants in an ethnic group burdened with metabolic disorders.

Fig. 1

Flow of data from the two genotyping platforms to the steps of imputation, meta-analysis and to the characterization of the resultant association signals. Characterization of the signals was carried out by way of comparing to association signals published in GWAS Catalog for the 313 search terms relating to metabolic traits

Fig. 2

Bubble plots illustrating the distributions of observed established variants (a) and gene loci (b) associated with the study-specific metabolic traits onto types (direct, indirect or broad) of relationships with observed traits in GWAS Catalog (as accessed in February 2019). (c) Presents the count of global studies (publications) reporting the observed associations in GWAS Catalog for each of the ‘direct’, ‘indirect’, and ‘broad’ categories. 83 variants (from 42 genes) formed the ‘direct’ category; such variants were associated with 11 traits (namely Weight, height, WC, BMI, TC, TG, LDL, HDL, FPG, HbA1C and DBP). The count of unique variants per each of these 11 traits were: weight:2 (both at suggestive p values); height:13 (1 variant at borderline p value and 12 at suggestive p value); WC:4 (all at suggestive p values); BMI:5 (all at suggestive p values); TC:12 (1 variant at borderline p value and 11 at suggestive p value); TG:28 (9 at borderline p value and 19 at suggestive p value); LDL:8 (8 at suggestive p value); HDL:13 (9 variants at genome-wide significant p values, 2 at borderline p value and 2 at suggestive p value); DBP:1 (at suggestive p values); FPG:1 (at suggestive p value); and HbA1C:1 (at suggestive p value)

Fig. 3

Ethnic transferability of SNP association signals for metabolic traits among populations. The figure presents the counts of transferable association signals (at the level of variants) from each global population to Arab population through ‘Direct’  (a) or ‘Indirect’  (b) relationships

Fig. 4

Trends in the mean of effect sizes of the established variants replicating  (a) in our study and those that are not replicating  (b). For every 5% incremental MAF of replicating variants (at borderline and suggestive p values) and non-replicating variants, we formed bins of effect size (beta) and then calculated summary statistics (mean ± standard deviation [SD]) for each bin for each of the 13 traits. Note the differences in the scaling of Y axis

Fig. 5

Venn diagrams showing genes associated with members of each of the four classes of tested 13 metabolic traits among the Kuwaiti population