Introduction Coronary artery disease (CAD) has a substantial genetic component which is incompletely characterised. Genomewide association (GWA) studies have recently identified several novel susceptibility loci for CAD [1]–[9]. Because GWA studies involve assumption-free surveys of common genetic variation across the genome, they can identify genetic regions responsible for previously unsuspected or unknown disease mechanisms. However, despite the success of the GWA approach, it has potential limitations. Because CAD loci identified through GWA studies have predominantly been found in regions of uncertain biological relevance, further work is required to determine their precise contribution to disease aetiology. Furthermore, in contrast with their high coverage of common genetic variation, GWA studies tend to provide limited coverage of genes with well-characterised biological relevance (“candidate genes”) [2], particularly in relation to lower frequency genetic variants (such as those with minor allele frequencies of 1–5%). Such variants are also often difficult to impute from GWA data. Although candidate gene studies should provide more comprehensive coverage of lower frequency and functional variants than GWA studies, most have been inadequately powered. To complement GWA studies, we undertook a large-scale gene-centric analysis of CAD using a customised gene array enriched with common and low frequency variants in ∼2,100 candidate cardiovascular genes reflecting a wide variety of biological pathways [10]. The array's potential to identify disease-associated lower frequency variants has been demonstrated by previous identification of strong independent associations with 2 variants in the LPA gene - rs3798220 (minor allele frequency 2%), and rs10455872 (7%) - and CAD risk [11]. We have now investigated this gene array in a further 13 studies comprising a total of 15,596 CAD cases and 34,992 controls. To enable interethnic comparisons, participants included 4,394 cases and 4,259 controls of South Asian descent, an ethnic group with high susceptibility to CAD. For further evaluation of putative novel associations, we attempted to replicate them in an additional 17,121 cases and 40,473 controls. Results The experimental strategy used is shown in Figure 1. In the discovery phase we genotyped participants from 12 association studies of CAD/myocardial infarction (MI), including a total of 11,202 cases and 30,733 controls of European descent (10 studies), plus 4,394 South Asian cases and 4,259 South Asian controls (2 studies) (Table 1, Table S1, with further details of the studies given in Text S1). 10.1371/journal.pgen.1002260.g001 Figure 1 Design of the study. 10.1371/journal.pgen.1002260.t001 Table 1 Summary details of discovery and replication stage studies. Stage Study Cases / Controls Male (%) Mean age (SD) of cases at diagnosis Number of cases with MI (%) Version of IBC array** European discovery ARIC 424 / 8447 46.4 -° 368 (82.7) V2 BHF-FHS 2101 / 2426 63.3 49.8 (7.7) 1538 (73.2) V1 BLOODOMICS - Dutch 1462 / 1222 72.6 48.8 (12.0) 1462 (100) V2 BLOODOMICS - German 1910 / 1932 63.3 59.2 (10.9) 1181 (61.8) V2 CARDIA 87 / 1343 46.8 -° 86 (100) V2 CHS 737 / 3155 43.9 -° 381 (50.5) V2 FOS 59 / 6976 45.1 -° 13 (22.0) V2 MONICA-KORA 275 / 1413 57.5 52.9 (9.4) >50%* V1 PennCATH 1027 / 489 66.0 54.2 (8.8) 439 (40.6) V1 PROCARDIS 3120 / 3330 59.2 61.0 (8.7) 2136 (68.5) V2 Total 11,202 / 30,733 South Asian discovery PROMIS 1856 / 1905 82.5 53.3 (10.7) 1856 (100) V1 LOLIPOP 2538 / 2354 83.7 - 1125 (44.4) V2 Total 4394 / 4259 Replication CARDIoGRAM† 15,949 / 38,823 57.0 53.5 (9.8) 10,890 (68.3) N/A EPIC-NL 1172 / 1650 30.6 51.9 (10.6) 341 (30.3) V3 Total 17,121 / 40,473 ARIC = Atherosclerosis Risk In Communities; BHF-FHS = British Heart Foundation Family Heart Study; CARDIA = Coronary Artery Risk Development in Young Adults; CHS = Cardiovascular Health Study; FOS = Framingham Offspring Study; LOLIPOP = London Life Sciences Prospective Population Cohort; PROCARDIS = Precocious Coronary Artery Disease; PROMIS = Pakistan Risk of Myocardial Infarction Study; EPIC-NL = European Prospective Investigation into Cancer & Nutrition (Netherlands) cohort. *All MONICA-KORA cases are either MI or sudden cardiac death. **V2 contains an additional 132 genes (3,857 SNPs) compared to V1. SNPs on V2 were only analysed in studies that used the V2 array. †: Details of studies in the CARDIoGRAM Consortium are presented in Table S6. °: The 4 studies in the CARe Consortium contributed data only on prevalent CAD cases at baseline for whom ages were not available. Associations with known CAD loci 36,799 SNPs passed QC and frequency checks and were included in the meta-analysis (reasons for exclusion of variants in each study are given in Table S2). The distribution of association P values in the discovery stage analyses are shown in Figure 2. We found significant associations with CAD for several previous GWA-identified loci contained on the array including 9p21.3 (rs1333042, combined European and South Asian P = 1.1×10−37) and 1p13.3 (rs646776, 3.1×10−17; Table S3). We also confirmed associations of other genes with strong prior evidence including the first association of a variant at the apolipoprotein E locus at genomewide significance (APOE/TOMM40, rs2075650, P = 3.2×10−8), as well as associations at apolipoprotein (a) (LPA, rs10455872, P = 1.2×10−20), and low density lipoprotein receptor (LDLR, rs6511720, P = 1.1×10−8; Table S3). However, we found no persuasive evidence of association of several prominently-studied genes and variants for which the previous epidemiological evidence has been inconclusive, even though the majority of these loci were well-tagged (Table S4) and the current study was well-powered to detect associations of modest effect (Figure S1). Notable variants that did not show significant association included the angiotensin converting enzyme (ACE) insertion/deletion polymorphism, the cholesteryl-ester transfer protein (CETP) Taq1B polymorphism and the paraoxonase 1 (PON1) Q192R polymorphism (Table S4). Perhaps contrary to expectation, apart from the LPA variant rs3798220, we did not observe any other strong association (odds ratio >1.5) among the ∼4,500 low frequency (1–5%) variants and/or variants with suspected or known functional impact on protein structure/function or gene expression specifically selected for the inclusion on the array (Table S3). 10.1371/journal.pgen.1002260.g002 Figure 2 Manhattan plots for discovery stage meta-analyses. Y-axis shows unadjusted −log10(P values) from fixed-effect meta-analysis of discovery stage studies. NB: European and Combined plots are truncated at P = 10−20. Blue horizontal line at P = 10−4 indicates threshold for replication; Red horizontal line at P = 3×10−6 indicates array-wide significance level. Novel CAD loci Based on simulations conducted prior to the analysis (Figure S2), loci were eligible for replication if unadjusted P-values for CAD were 4,500 such variants. Fourth, we have confirmed the relevance of several previously established CAD genes to both Europeans and South Asians, without finding qualitative differences in results by ethnicity. LIPA (lipase A) encodes a lysosomal acid lipase involved in the breakdown of cholesteryl esters and triglycerides. Mutations in LIPA cause Wolman's disease [16], a rare disorder characterized by accumulation of these lipids in multiple organs. However, despite evidence that the risk allele was associated with higher LIPA gene expression (suggesting that both under- and over-activity of LIPA increase CAD risk), it was not significantly associated with altered lipid levels. This finding suggests that the impact on CAD risk is either through an alternative pathway, or that the mechanism is more complex than reflected through conventionally measured plasma lipid levels. Two recent studies have also found associations of variants in the LIPA gene with CAD using a GWA approach, strengthening the evidence for this association [17], [18]. Our identification of the association of variants near interleukin 5 (IL5), an interleukin produced by T helper-2 cells, is interesting given the evidence that both acute and chronic inflammation may play important roles in the development and progression of CAD [19]. Most previous human association studies of inflammatory genes and CAD have focused on other cytokines and acute-phase reactants. Nevertheless, some experimental data predict that IL-5 has an atheroprotective effect and this has been supported by association between higher circulating IL-5 levels and lower carotid intimal-medial thickness [20]–[22]. Our findings now highlight the potential importance of IL-5 in CAD, especially as the IL-5 receptor is already a viable therapeutic target in allergic diseases, although we can not rule out the possibility that another gene at this locus may be mediating the association with CAD risk. The ATP-binding cassette sub-family G proteins ABCG5 and ABCG8 are hemi-transporters that limit intestinal absorption and promote biliary excretion of sterols. Mutations in either gene are associated with sitosterolaemia, accumulation of dietary cholesterol and premature atherosclerosis [23]. Recently, common variants in ABCG8 have also been shown to be associated with circulating LDL-C and altered serum phytosterol levels with concordant changes in risk of CAD [15], [24]. Our findings confirm that this locus affects CAD risk either directly through its effect on plasma phytosterol levels or through primary/secondary changes in LDL-cholesterol. The association signal on 8q24.13 maps near the TRIB1 gene which encodes the Tribbles homolog 1 protein. Tribbles are a family of phosphoproteins implicated in regulation of cell function, although their precise roles are unclear [25]. However, SNPs in or near TRIB1 - including the lead SNP in our study (rs17321515) - have recently been shown to have highly significant associations with levels of several major lipids [15], providing a possible mechanism for their association with CAD. Our findings confirm the previous suggestion that this variant is also associated with CAD risk [15], [26]. Hepatic over-expression of TRIB1 in mice has been shown to lower circulating triglycerides; conversely, targeted deletion of the TRIB1 gene in mice led to higher circulating triglycerides [27]. The location of the CAD-associated variant downstream of TRIB1 suggests that its effect may be mediated by regulation of TRIB1 expression leading to adverse lipid profiles, although we did not find an eQTL at this locus in monocytes. Our study brings to 33 the number of confirmed loci with common variants affecting risk of CAD (Figure 7). We estimate that in aggregate these variants explain about 9% of the heritability of CAD which is consistent with the recent analysis by CARDIoGRAM [12]. Interestingly, the odds ratios that we observed for the novel loci were generally lower than those of previously identified loci. This suggests that most of the common variants with moderate effects have been identified and that increasingly larger sample sizes will be required to detect further common variants that affect risk of CAD. However, the modest odds ratios associated with such variants do not necessarily imply that they are not of potential clinical or therapeutic relevance. For example, there are only modest effects of common variants in the LDLR gene on CAD risk (Figure 7); yet this pathway has become a major target for the prevention and treatment of CAD with the development of statins. 10.1371/journal.pgen.1002260.g007 Figure 7 Novel loci identified in this study placed in the context of previously confirmed CAD loci. Previously reported variants listed are those from the NHGRI GWA studies catalogue [32] reported as having P 2% tagged at r2>0.8), ‘intermediate priority genes’ moderately covered (all SNPs with MAF>5% tagged at r2>0.5), and ‘low priority genes’ tagged using only non-synonymous SNPs and known functional variants with MAF>1%. A “cosmopolitan tagging” approach was used to select SNPs providing high coverage of selected genes in 4 HapMap populations (CEPH Caucasians, Han Chinese, Japanese, Yorubans). For all genes, non-synonymous SNPs and known functional variants were prioritised on the array. Genotypes were called using standard algorithms (eg, GenCall Software and Illuminus) and standard quality control methods were applied to filter out poorly performing or rare ( 0.8 with all HapMap/Seattle SNPs of MAF≥0.02), 2 (r2>0.5 with all HapMap/Seattle SNPs of MAF≥0.05), 3 (only non-synonymous and known functional variants of MAF>0.01) and GWAS (specific SNPs previously identified in recent GWAS). a rs4343 has r2 = 1 with the insertion/deletion polymorphism in the ACE gene in CEU HapMap 2 population. b rs17443251 has r2 = 0.75 with the more commonly studied R144C variant (rs1799853) in the CYP2C9 gene in CEU HapMap 2 population. c rs9526246 has r2 = 0.97 with the more commonly studied T102C variant (rs6313) in the HTR2A gene in CEU HapMap 2 population. d rs1062535 has r2 = 0.97 with the more commonly studied C807T variant (rs1126643) in the ITGA2 gene in CEU HapMap 2 population. e rs1805096 has r2 = 0.89 with the more commonly studied rs6700896 variant in the LEPR gene in CEU HapMap 2 population. f rs1049897 has r2 = 1 with the more commonly studied A102T variant (rs4236) in the MGP gene in CEU HapMap 2 population. g rs4968624 has r2 = 0.97 with the more commonly studied L125V variant (rs668) in the PECAM1 gene in CEU HapMap 2 population. h rs12944077 has r2 = 1 with the more commonly studied S563N variant (rs12953) in the PECAM1 gene in CEU HapMap 2 population. (PDF) Click here for additional data file. Table S5 27 loci meeting P 5%) in at least one population are displayed. (PDF) Click here for additional data file. Text S1 Supplementary Methods, Supplementary References, Supplementary Acknowledgements. (PDF) Click here for additional data file.
