Genetic associations with carotid intima-media thickness link to atherosclerosis biology with sex-differences in sub-Saharan Africans


 Atherosclerosis precedes the onset of many clinical manifestations of cardiovascular diseases (CVDs). We used carotid intima-media thickness (cIMT) to investigate genetic susceptibility to atherosclerosis in 7894 unrelated adults (3963 women, 3931 men) aged 40 to 60 years resident in four sub-Saharan African countries. cIMT was measured by ultrasound and genotyping was performed on the H3Africa SNP Array. Two new African-specific genome-wide significant loci, SIRPA (p=4.7E-08), and FBXL17 (p=2.5E-08), were identified in the combined dataset. Sex-stratified analysis revealed associations with two male-specific loci, SNX29 (p=6.3E-09) and MAP3K7 (p=5.3E-08), and two female-specific loci, LARP6 (p=2.4E-09) and PROK1 (p=1.0E-08). Regional associations were replicated with known risk loci for atherosclerosis and CVDs with different lead SNPs than in Europeans and significant enrichment for oestrogen response genes for female-specific signals were identified. The genes identified showed biological relevance to atherosclerosis and/or CVDs, as well as sex-differences and transferability of signals from non-African studies.


Introduction
Atherosclerosis is a complex multifactorial trait with an enigmatic genetic aetiology. Despite discoveries from genome-wide association studies (GWAS), little is known about the genetic contributions to atherosclerosis. Meanwhile, the worldwide epidemic of cardiovascular diseases (CVDs), including clinical manifestation of atherosclerosis, is growing and has become the leading cause of deaths worldwide (Fuster, 2014;Roth et al., 2017). Moreover, the health and demographic transition in sub-Saharan Africa (SSA) has shifted the major causes of death from communicable and nutritional diseases to non-communicable diseases (NCDs).
Atherosclerosis results from injury to the arterial endothelium, resulting in an in ammatory response in the vessel wall. The location and morphology of the atherosclerotic lesions predict the nature of the resulting vascular disease. Whereas family and twins studies provided evidence of high heritability of cIMT (20-65%) (Fox et al., 2003;Sacco et al., 2009;Fagnani et al., 2013;Medda et al., 2014), the GWAS studies reported associations that account for only 1.1% of the variance of cIMT (Bis et al., 2012).
The genetic diversity of African populations and their deep evolutionary roots represent opportunities for novel genetic discoveries. Because haplotypes blocks are shorter in Africans compared to other populations (average haplotype block ~ 8.8 kb in Africans, ~ 20.7 kb in Europeans, and ~ 25.2 kb in Han Chinese), identi cation of causal variants is facilitated (Lonjou et al., 2003;Hinds et al., 2005). The role of ancestry in atherosclerosis risk has been established from studies in multi-ethnic settings and admixture studies for atherosclerosis (Shendre, Irvin, et al., 2017;Shendre, Wiener, et al., 2017). African ancestry was reported to be associated with a higher risk of atherosclerosis compared to Europeans, Hispanics and Asians.
Since phenotypic differences between men and women are a pervasive feature of several quantitative traits, studies of sex interactions for complex human traits may shed light on the molecular mechanisms that lead to biological differences between men and women. Sex has been found to play a role in variations between gene expression and genotype across a range of human complex traits (Rawlik et al., 2016). Sex-differences in the transcriptomes of cells involved in the atherosclerotic process have been previously reported (Franconi et al., 2017) and are supported by sex-strati ed GWAS analyses (Dong et al., 2015;Lin et al., 2015). Sex provides two different environmental contexts determined by the hormonal milieu and differential gene expression between the sexes.
Several genetic association studies of cIMT have been performed in the major world populations and provided insights into genes and tissue-speci c regulatory mechanisms linking atherosclerosis both to its functional genomic origins and its clinical consequences in humans. To date, 76 SNPs have been found to be robustly associated with cIMT (GWAS Catalog) (Buniello et al., 2019), but none of the studies focused on sub-Saharan African populations.
The Africa Wits-INDEPTH Partnership for Genomic Studies cohort (AWI-Gen) was developed to examine genetic and environmental contributions to cardiometabolic diseases in Africans. It has over 12 000 participants from four sub-Saharan African countries, Burkina Faso, Ghana, Kenya and South Africa, and the distributions and associated risk factors for cIMT have been described (Ramsay et al., 2016;Ali et al., 2018;Nonterah et al., 2018). This study aimed to investigate genetic susceptibility to atherosclerosis in sub-Saharan Africans in the AWI-Gen cohort. cIMT was used as an endophenotype, with further investigation of sex-differences.

Results
Genetic Association with cIMT Analyses were performed using the imputed dataset of 13.9M SNPs in 7894 participants from the AWI-Gen study and association with mean-max-cIMT. Despite the population sub-structure demonstrated by principal component analysis in the study sample ( Supplementary Figure 1), our results did not show evidence of genomic in ation (l = 0.997). The genome-wide association results for the combined dataset are illustrated in the Manhattan plot and the genomic in ation by the QQ-plot (Figure 1a, 1b). In the combined dataset, we identi ed two new genome-wide signi cant loci in SIRPA on chromosome 20 (rs6045318, p = 4.7E-08) and FBXL17 on chromosome 5 (rs552690895, p = 2.5E-08). These two SNPs are African speci c and have not been observed in European or Asian populations. Other suggestive association signals had lead variants located in an intergenic region on chromosome 8 (rs11781274, p = 1.8E-07), an intronic region in SORCS1 (rs11193156, p = 2.1E-07), an intronic region in ANKK1 (rs11214599, p = 5.4E-07), an exonic region in CTBP2 (rs3781409, p = 6.6E-07) and an intronic region in SMARCA2 (rs1324201, p = 8.6E-07) ( Table 1, Supplementary Table 1).

Replication with GWAS Catalog
In view of the limited number of genome-wide signi cant SNPs for cIMT previously reported, our replication analysis also included screening for phenotypes similar to cIMT (coronary artery calci cation (CAC), abdominal aortic aneurysm (AAA)). Our study replicated (see criteria in Methods section) the locus for association with cIMT in the CBFA2T3 region with a SNP that is 18979 bp from and associated SNP reported in Europeans: rs9934287 (p = 6.6E-06) was suggestively associated in our study in the CBFA2T3 region. The rs844396 (p = 6.00E-09) that was previously reported by Franceschini and colleagues in a European ancestry population (Franceschini et al., 2018) and later replicated by a UK Biobank analysis of cIMT (Strawbridge et al., 2020), did not replicate in our study (p = 0.85). The rs9934287 SNP (MAF=0.047 in our study) is monomorphic in populations of European ancestry from the 1000 Genomes Project.
A previously reported locus for association with carotid plaque in European populations (Pott et al., 2017) at GEM (rs72672639, p = 4.0E-06) was suggestively associated in our female-speci c subset with two SNPs (rs78571209, rs76489670, p = 7.8E-05) located approximatively 2200 bp from the SNP reported for plaque in Europeans. Similarly, the association with the MRPL37 locus (rs11206301, p= 8.00E-06) for plaque in European populations was suggestively associated in our male-speci c analysis for cIMT (rs13374450, p = 3.0E-05). The two SNPs in the MRPL37 locus were not in LD despite their proximity (201 bp). The suggestive variant in our study rs4773141 (p = 4.7E-05, in the combined dataset), located in COL4A1, was previously reported for CAD (p = 4.0E-17) in European populations (Van der Harst and Verweij, 2017).
In our combined analysis, a total of 10 SNPs replicated for CAC (a surrogate marker of atherosclerosis as cIMT) (Inouye et al. 2012;O'Donnell et al. 2011) and for CAC in Type 2 Diabetes African patients (Divers et al., 2017). Fourteen SNPs replicated for coronary heart disease and coronary artery disease (Supplementary Table 2-3).

Functional annotation
Annotation of the genic positions of the 467, 515 and 581 SNPs respectively from combined, female-speci c and male-speci c analyses with signi cant and suggestive associations (p<1E-05) showed that these were mostly intronic or intergenic. 50 SNPs displayed a CADD score above 12.37 suggestive of being potentially deleterious (19 in the combined; 18 in female-speci c; 13 in male-speci c datasets) (Supplementary Table 4a, 4b, 4c). In the female-speci c sample, the lead SNP in CYMP (rs115473055) had a Regulome DB score of 2a suggesting the variant was likely affecting a transcription binding site (CTCF). Positional mapping, eQTL mapping (matched cis-eQTL SNPs) and chromatin interaction mapping (on the basis of 3D DNA-DNA interactions) is reported (Supplementary Table 5a, 5b, 5c). We found that rs78172571, in high LD with rs150840489 (the top SNP associated in our female-speci c), was involved in HiC type chromatin interactions in multiple tissues including aorta, in which the variant acts as an enhancer of THAP10 (FDR = 2.03E-17).

Gene-based and gene-set analysis
In a gene-based analysis (using MAGMA threshold of p< 2.6E-06) of the combined dataset analysis there was a signi cant association with CALD1 (p = 5.9E-07) (Supplementary Figure 3A) with Mean-Max cIMT, whereas in female-speci c analysis FLT4 (p = 4.3E-07) was signi cantly associated (Supplementary Figure 3B). The results from gene-set analysis in the combined dataset showed signi cant enrichment for "Chemical and Genetic perturbation" gene-set (adjP = 3.9E-05). The female-speci c analysis revealed signi cant enrichment of gene-sets (Supplementary Table 6a, 6b, 6c), with among them "Hallmark gene-sets for Oestrogen response", with "Early Oestrogen response" and "Late Oestrogen response" both being signi cant (2.2E-6).

Discussion
In this African population GWAS for cIMT as the outcome variable, and as a proxy for atherosclerosis, we identi ed two new loci associated with cIMT in the full dataset, two new loci speci c to the female only analysis and two loci associated in the male only analysis (p < 5E-08). We replicated regional associations with known loci associated with atherosclerosis and CVDs, but with different lead SNPs at almost all loci. In total, 54 loci were replicated with SNPs at p<1E-04 within 25 kb of previously reported genome-wide signi cantly associated variants. Those loci were associated with the following atherosclerosis phenotypes (cIMT, carotid plaque, coronary artery calci cation, and abdominal aortic aneurism) and outcomes (coronary artery disease, coronary heart disease, myocardial infarction, and stroke).
Measurements of cIMT are used clinically to assess vascular pathophysiology and to re ect the atherosclerosis process. Our study identi ed cIMT-associated loci relevant to genes related to macrophage activity and polarisation (SIRPA), to vascular smooth muscle cells (MAP3K7, CALD1), to vascular endothelial growth (PROK1, FLT4), to collagen synthesis and plaque stability (LARP6), and a pathway of blood vessel occlusion (SNX29) (Figure 4).
The associated loci are discussed with regard to their potential functions and biological evidence from previous studies and previous reports from GWAS (Extended discussion in Supplementary Note). FBXL17 (lead SNP:rs552690895; p = 2.5E-08) in the combined data set, is linked to cardiovascular physiology through its involvement in protein degradation where it plays a central role in cardiovascular physiology and disease: from endothelial function, the cell cycle, atherosclerosis, myocardial ischaemia, cardiac hypertrophy, inherited cardiomyopathies, and heart failure. A GWAS in Lithuanian families found that variants in FBXL17 were associated with coronary heart diseases (Domarkiene et al., 2013).
Signal regulatory protein alpha (SIRPA) (lead SNP:rs6045318; p = 4.7E-08 in the combined data set) has a role in the mediation of phagocytosis and polarization of macrophages which is important in the pathophysiology of atherosclerosis (Chen et al., 2019). There is evidence that SIRPA is involved in discrete stages of cardiovascular cell lineage differentiation (Skelton et al., 2014) and that defects in the gene (knock out) reduces atherosclerosis in mice (Szilagyi et al., 2014). SIRPA expression has been found as a signature of in amed atherosclerotic plaque (Puig et al., 2011).
On the chromosome 16, rs147978408 (p = 6.3E-09) was the top cIMT associated variant in SNX29 for the male-speci c analysis. The sorting nexin (SNX) family genes are associated with CVDs, and dysfunction of the SNX pathway is involved in several forms of cardiovascular disease (CVD) (Yang et al., 2019). In a study of genes that regulate smooth muscle cell differentiation and disease risk, SNX29 was involved in pathways for occlusion of blood vessels and atherosclerosis (Iyer et al., 2018). Ito and colleagues identi ed sex-dependent differentially methylated regions close to SNX29 in mouse liver and found that this methylation status was in uenced by testosterone and contributed to sex-dimorphic chromatin decondensation (Ito et al., 2015). This might explain the sex-speci c effect observed in our study. Because of the previous link between SNX29 and hypertension, we ran further GWAS analysis strati ed by hypertensive status and found that the association was driven by the hypertensive group (effect three times higher in hypertensives compared to the non-hypertensives), therefore demonstrating that the association of SNX29 with cIMT might be mediated by the vascular remodeling caused by hypertension.
In the male-speci c analysis, rs284509 (p=5.3E-08) in MAP3K7 region on chromosome 6 to was associated with cIMT. Mitogenactivated protein kinase kinase kinase 7 (MAP3K7) is known to play a role in growth inhibition in vascular smooth muscle cells.
The sex-speci c association observed might be related to the fact that MAP3K7IP3 (located on the X chromosome), which is known to form a ternary complex with MAP3K7 in response to in ammatory stimuli, has shown sex-differential expression in ischemic stroke (Stamova et al., 2012;Rocha et al., 2016). In a study on expression of androgen-modulated micro-RNAs, it was reported that MAP3K7 was a target of mmu-miR-467h and mmu-miR-669i in the angiogenesis and transforming growth factor beta receptor signalling pathways (Bouhaddioui, Provost and Tremblay, 2016). Our study is the rst to report MAP3K7 association with a CVD phenotype. LARP6 (La-related protein 6) is a ribonucleoprotein domain family member 6 with a role in collagen regulation by targeting mRNA encoding Type I collagen (Zhang and Stefanovic, 2016;Stefanovic et al., 2019). Collagen is a hallmark of atherosclerotic plaque stability, thus alteration of the collagen balance may lead to an instability of atherosclerotic lesions, and therefore promote plaque formation and rupture (Puig et al., 2011;Higashi et al., 2016). In the Taiwanese population, the LARP6 locus was found to be associated with coronary artery disease (Assimes et al., 2016). Myocardial gene expression in non-ischemic human heart failure found that LARP6 was differentially expressed between men and women (1.36 fold) (Fermin et al., 2008). The female-speci c effect of this locus in our study may be explained by the enhancer function of rs78172571 in high LD with rs150840489 (the top SNP associated in our female-speci c) on THAP10 gene (FDR = 2.03E-17), known to be regulated by oestrogen.
Our study is the rst to report prokineticin 1 (PROK1) for any trait in a GWAS. It was associated with cIMT in the female-speci c analysis (lead SNP:rs115473055, p=1.00E-08). PROK1 is a speci c placental angiogenic factor that plays a role in the control of normal (e.g. endometrial decidualization) and pathological placental angiogenesis (Hoffmann et al., 2006). The gene is known to be predominantly expressed in the steroidogenic glands, such as ovary, testis, and adrenal cortex, and is often complementary to the expression of vascular endothelial growth factor (VEGF), suggesting that these molecules function in a coordinated manner. The function and particular pattern of this gene's activity might explain why we identi ed the locus only in our femalespeci c analysis.
Our gene-based analysis identi ed caldesmon 1 (CALD1) as signi cantly associated with cIMT in our combined set led by rs7781307 (p = 2.1E-06) on 7q33. This gene plays a major role in the regulation of smooth muscle contraction, cell migration and cell invasion (Mayanagi and Sobue, 2011). CALD1 was linked to advanced coronary atherosclerosis (Tan et al., 2017) and abdominal aortic aneurysm (Wan et al., 2018). Under expression of CALD1 was found to be a key feature of calci cation of vascular smooth muscle cells from atherosclerotic plaque ( Additionally, studies on epigenetic modi cations reported CALD1 to exhibit differential methylation in atherosclerosis (Zaina et al., 2014;Nazarenko et al., 2015;Fernández-Sanlés et al., 2017).
Our study is the rst to report and association of FLT4 ( also known as vascular endothelial growth factor receptor 3 (VEGFR3)) (rs112967731; p = 5.7E-07, female-speci c) with cIMT or any cardiovascular phenotype in GWAS studies. FLT4 is a major signalling protein involved in angiogenesis, vasculogenesis and maintenance of the endothelium. Defect and/or downregulation of VEGR3 was found to lead to cardiovascular failure in embryonic stage and to higher mortality after myocardial infarction in mouse models (Dumont et al., 1998;Vuorio et al., 2018). Biological studies have highlighted the role of FLT4 in atherosclerosis in major pathological processes. The gene has been reported to be involved in plaque instability by two process: the mediation of monocytes/macrophages apoptosis and consequently alteration plaque stability (Schmeisser et al., 2006); and the modulation of vascular remodelling and shear stress resulting in plaques haemorrhages and calci cation in carotid arteries (Baeyens et al., 2015;Tuenter et al., 2016;Camaré et al., 2017).
The SNP rs116517341, which leads the association with the CCDC71L locus in our male-speci c analysis (p=6.30E-05), is located over 100 kb from the lead-SNP found in European, but our lead SNP was closer to CCDC71L than that found by the study of cIMT in Europeans. Therefore, different variants may in uence the association of the CCDC71L gene in cIMT. When analysing the variants in CBFA2T3, our lead-SNP (rs9934287) was located 18,979 bp away from the SNP reported by Franceschini et al. (2018). The regional plot showed more dense signals with numerous SNPs in LD with the lead-SNP in European, whereas our lead-SNP had fewer SNPs in high LD ( Figure 5 C-D). The LD structure using 1000 Genomes Project Europen (CEU) compared to African populations (YRI-LWK-GWD-MSL-ESN) showed that LD blocks were smaller in Africans (Supplementary Figure 4), providing opportunity for extended ne-mapping and reducing the credible set toward identifying causal variants.
Our sex-speci c analyses revealed loci that support the hypothesis that sex differences may be due to sex-speci c epigenetic modi cation, independent of sex hormone levels. When analysing sex-speci c or gene-sex interactions, it is important to keep in mind that they also re ect the in uences of non-genetic factors such as behaviour, as evidence by the previously reported genesmoking interactions (Boua et al., 2020). Hence, environmental exposure, anatomical differences, and sex hormone environment, which create systemic differences between males and females for trait expression, affect disease risk and heritability (Gilks, Abbott and Morrow, 2014).
Our study identi ed signi cant enrichment of oestrogen pathway genes in our female-speci c analysis. Oestrogen-dependent regulation of vascular gene expression and vascular physiology encompasses complex processes involving both nuclear and membrane-associated oestrogen signalling pathways. In recent years we have witnessed major progress in understanding how these regulatory processes contribute to the atheroprotective effects exerted by oestrogens. Animal models of atherosclerosis provided compelling evidence that physiological oestrogen levels potently attenuate both early and advanced stages of atherosclerosis lesion development in females, and suggested similar protective effects in males. The effect of oestrogens on atherosclerosis can target metabolism (lipid, glucose), macrophage function or smooth muscle cells. Nonetheless, hormone replacement therapy during menopause has not been shown to conclusively reduce atherosclerosis risk, suggesting that more studies are needed to fully decipher the biological mechanisms.

Strengths and limitations
Our study is the rst population-based study to investigate the genetic architecture of cIMT in sub-Saharan African populations. In addition to our sex-speci c analyses, we tested for sex-difference between the two strata using a minimal model (adjustment for age and PCs) to avoid the "collider bias". We used an analysis framework allowing us to identify genetic effects that point in opposite directions in men and women and to detect genetic effects that are only (or more pronounced) in one stratum, a method that has been shown to have better power to identify qualitative gene-sex interactions (Winkler et al., 2017). The use of a new SNP genotyping array with better representation of common African variants and imputation reference panels from African participants has improved the SNP coverage in ethnically diverse African populations.
The lack of an ethnically matched replication cohort is a limitation in our study, and it will be important to replicate these ndings in additional suitable cohorts. We identi ed African-speci c variants in new loci and replicated previously reported loci, revealing opportunities for trans-ancestry ne-mapping.
We found evidence of gene set enrichment for biological processes. Our study is the rst GWAS to report signi cant enrichment of genes in the oestrogen pathway for cIMT in our female-speci c analysis. The ndings from our study support the notion that genomics studies in Africa are likely to contribute to the understanding of complex traits, such as atherosclerosis.

Materials And Methods
Study population and phenotype assessments This is a cross-sectional study that investigated populations from six sub-Saharan African sites in West Africa (Burkina Faso (Nanoro) and Ghana (Navrongo)), East Africa (Kenya (Nairobi)) and South Africa (Agincourt, Dikgale and Soweto) as part of the AWI-Gen study (Richter et al., 2007;Derra et al., 2012;Kahn et al., 2012;Oduro et al., 2012;Alberts et al., 2015;Beguy et al., 2015;Ramsay et al., 2016;Ali et al., 2018). The participants for this study include 10,703 black African men and women from two urban settings (Nairobi and Soweto) and four rural settings (Agincourt, Dikgale, Nanoro and Navrongo), aged 40 to 60 years.
Participants completed a questionnaire requesting information on demography, health history and behaviour. Anthropometric measurements were taken and blood collected for genotyping (H3Africa SNP array) and phenotyping (biomarkers) (Ali et al., 2018). Ultrasound scans were performed to assess cIMT of the right and left carotid arteries. No cIMT data was collected for female participants from Soweto because they were drawn from the Study of Women Entering and Endocrine Transition (SWEET) study, and they were therefore not included in the subsequent GWAS. This study received approval from the Human Research Ethics Committee (Medical), University of the Witwatersrand, South Africa (M121029, M1706110). All the participants provided written informed consent prior to enrolment and participation in the study.
cIMT Measurement cIMT was measured using Dual B-mode ultrasound images of the carotid tree showing a typical double line for the arterial wall.
Details of the method for measurement are provided in Ali et al. 2018(Ali et al., 2018. The cIMT values were QCed according to the Mannheim Consensus de ning the use of cIMT in population-based studies. The Mean Max cIMT was generated as the average of the maximum cIMT from the left and right, and this value was used for the GWAS analyses.

Genotyping and Imputation
The H3Africa genotyping array (https://chipinfo.h3abionet.org), designed as an African-common-variant-enriched GWAS array (Illumina) with ~2.3 million SNPs, was used to genotype genomic DNA using the Illumina FastTrack Sequencing Service (https://www.illumina.com/services/sequencing-services.html). The following pre-imputation QC steps were applied to the entire AWI-Gen genotype data set. Individuals with a missing SNP calling rate greater than 0.05 were removed. SNPs with a genotype missingness greater than 0.05, MAF less than 0.01 and Hardy-Weinberg equilibrium (HWE) P-value less than 0.0001 were removed. Non-autosomal and mitochondrial SNPs, and ambiguous SNPs that did not match the GRCh37 reference alleles or strands were also removed. Imputation was performed on the cleaned dataset (with 1,729,661 SNPs and 10,903 individuals) using the Sanger Imputation Server and the African Genome Resources as reference panel. We selected EAGLE2 (Loh et al., 2016) for pre-phasing and the default PBWT algorithm was used for imputation. After imputation, poorly imputed SNPs with info scores less than 0.6, MAF less 0.01, and HWE P-value less than 0.00001 were excluded. The nal QC-ed imputed data had 13.98 M SNPs, and only participants with both good quality cIMT and genotyping data (n = 7894) were used for the GWAS analyses.

Genome-wide association analysis
Linear regression of Mean Max cIMT was performed with covariates in R (https://www.R-project.org/). Residuals were extracted from the linear regression analyses and used for the GWAS analysis. We used as covariates age, sex and 8 principal components (PCs) computed on genetics data. In our sex strati ed analysis (3963 women, 3931 men), the covariates were age and 5 PCs. The number of PCs to include in each model were determined using a stepwise regression. We performed all association testing with the residuals in BOLT-LMM, which implement testing using a Linear Mixed Model (LMM). To run e ciently, BOLT-LMM required three components: the (imputed) genotypic data for association testing; a reference panel of LD scores per SNP, calculated using 1000 Genomes Project African samples; and genotype data used to approximate a genetic relationship matrix (GRM) (Using a subset of the SNP Array genotypes with LD ltering). This method is expected to account for all forms of relatedness, ancestral heterogeneity in the samples and other (potentially hidden) structure in the data. The analyses were run on the automated work ow of H3abionet/H3agwas (http://github.com/h3abionet/h3agwas/) (Baichoo et al., 2018).
We screened the output for a genome-wide signi cance threshold (p-values < 5.E-08). To assess genomic in ation, we compared our observed distribution of −log10(P) values to that expected in the absence of association (Lambda) and illustrated the results in QQ plots. The same process was applied for sex-strati ed analyses.
We used EasyStrata (Winkler et al., 2015) to test for the joint effect calculated from sex strata results (Aschard et al., 2010) and to test for the difference between two strata results as a means to test for sex effects (Randall et al., 2013). The joint and strati ed frameworks were found to be the most e cient way to test for gene-environment interactions (Sung et al., 2016). Power calculations were performed with Quanto (Version 1.2.4) (http://biostats.usc.edu/Quanto.html ).

Replication from the GWAS catalog
The GWAS Catalog database was downloaded (https://www.ebi.ac.uk/gwas/, accessed on 12 Jan 2019) and a subset of the data generated using the following key words relevant to our study: coronary artery disease, carotid atherosclerosis, cIMT, coronary artery calci cation and abdominal artery aneurism. The marker co-ordinates from the GWAS Catalog are given in build 38. Since our dataset was in build 37, we performed lift-over of GWAS Catalog to build 37 in order to allow accurate comparison.
In order to look whether our study was replicating previous ndings, we searched for the same marker or any markers within 25 kb (considering the highest mean size of LD blocks). We then searched for SNPs in a 25 kb region of all SNPs with suggestive associations (p-value < 1E-04) found in our study. We further de ned loci by grouping SNPs with p-value < 1E-04 within 250 kb of each other. These loci were used for regional replication and transferability analyses.

Functional analysis
The FUMA online platform (http://fuma.ctglab.nl/) (Watanabe et al., 2017) was used to annotate, prioritize, visualize and interpret GWAS results. GWAS summary statistics (p<1E-05) from out study was used as the input. FUMA provided extensive functional annotation for all SNPs in genomic areas identi ed by lead SNPs. From the list of gene IDs (as identi ed by SNP2GENE option in FUMA) FUMA annotated genes in a biological context (Watanabe et al., 2017). We selected all candidate SNPs in the associated genomic region having r 2 ≥ 0.6 (with 1000 Genome Project African references) with one of the independently signi cant SNPs, with a suggestive P-value (p < 1E-05) and MAF > 0.01 for annotation. Predicted functional consequences for these SNPs were obtained by matching the SNP's chromosome base-pair position, and reference and alternate alleles, to databases containing known functional annotations, including ANNOVAR (Wang, Li, and Hakonarson 2010)

MAGMA Gene-based and gene-sets analysis
Multi-marker analysis of genomic annotation (MAGMA, v1.6) gene analysis were performed using summary statistics of our association results as input. Gene-based analysis enabled summarization of SNPs associations at the gene level and association of the set of genes to biological pathways. MAGMA employs multiple linear regression to obtain gene-based pvalues (de Leeuw et al., 2015;Watanabe et al., 2017). The window for gene annotation was set for 25kb and genome-wide signi cance was set at 0.05/number of tested genes. MAGMA gene-set analysis used a competitive testing framework, with gene-sets from MsigDB (v6.2, 10678 gene sets (curated gene sets: 4761, GO terms: 5917)) (Liberzon et al., 2015). MAGMA analysis was implemented within FUMA.    Figure 1 QQ and Manhattan plots for cIMT association results in AWI-Gen study (7894 participants).

Figure 2
Miami plot showing female and male-speci c associated p-values for mean max cIMT.

Figure 4
Biological relevance of identi ed loci