Genetics of coronary artery disease: focus on genome-wide association studies

Abstract

With the advent of the genome-wide association (GWA) study, a promising new avenue for identifying genetic markers for complex diseases like coronary artery disease (CAD) has been opened. This avenue, however, is not without challenges and limitations, including the need for carefully designed and executed studies and the risk of false positive associations. Nonetheless, new markers have been identified through such studies that could potentially revolutionize the ways that individuals with CAD are identified and managed.

Background

Coronary artery disease (CAD) and its associated complication, myocardial infarction (MI), is a leading cause of morbidity and mortality worldwide. CAD is a multifactorial disease that can be influenced by many different environmental and heritable risk factors. The progression of atherosclerotic disease involves a complex series of events, each event (e.g. foam cell formation, smooth muscle cell recruitment, etc.) involving multiple biological pathways and genes. Risk factors for CAD include hypertension, smoking status, male gender, age, body mass index, type 2 diabetes mellitus, and heredity. Analyte risk markers associated with CAD include lipid levels, cholesterol particle size and number, C-reactive protein, homocysteine, fibrinogen, and lipoprotein (a). However, many of the current traditional and novel risk markers are unable to fully predict who is at risk for CAD. For example, 35% of CAD has been shown to occur in people with total cholesterol levels <200 mg/dL [1]. Thus, other novel risk markers, including genetic risk markers, may be important for better refining individuals at risk for CAD and CV events.

Both twin and family studies have demonstrated that CAD has a strong heritable component. In one study examining 20,966 twins over a 36-year period, the heritability (h²) for fatal CAD events was 57% for men and 38% for women [2]. Many studies have demonstrated that family history of CAD is an independent predictor of CAD [3–6]. Others have shown that distinct morphological features of CAD display high heritability [7, 8]. In addition, individuals with a family history of CAD generally have earlier onset, are male, have a history of smoking, and are hyperlipidemic [9]. The majority (72%) of early CAD cases (men < 55 years, women < 65 years) and 48% of all CAD cases, regardless of age, have a family history of CAD [10].

While multiple studies have concurred that CAD is highly heritable, the mechanisms underlying the heritable basis of CAD have been elusive. This is likely due to the complex nature of CAD. While a very small proportion of CAD is monogenic, or Mendelian in nature (e.g. familial hypercholesterolemia, familial defective apoB-100), the vast majority of CAD is genetically complex. Multiple genes are thought to contribute to CAD, each gene contributing to only a small percent of the phenotype. In addition, many of the risk factors associated with CAD (e.g. blood pressure, diabetes) are themselves polygenic, further adding to the complexity, and hundreds of genes may be involved in CAD susceptibility [11]. Thus, while some genetic parameters, such as heritability, can be quantified for CAD, determining the number and scope of genes involved in CAD is challenging.

Many different family and population-based studies have identified candidate genes potentially associated with CAD. Unfortunately, the vast majority of results from linkage and association studies have not been reproducible or statistically significant, bringing to question the clinical utility of these markers [12]. However, recent technological and scientific advances have now made possible a new type of study: the genome-wide association (GWA) study. The advent of GWA studies has produced some novel and replicated associations in many disease states, including CAD.

Genome-wide association studies and CAD

GWA studies are essentially unbiased large-scale population-based studies evaluating the association of hundreds of thousands of markers (generally single nucleotide polymorphisms, or SNPs) across the genome with a particular phenotype. Part of the beauty of these types of studies is that they are not hypothesis-driven, allowing for the discovery of novel genetic markers. However, GWA studies are not without challenges and limitations and they must be carefully designed and executed. In the design of a GWA study, it is important to keep phenotypic heterogeneity of cases and misclassification of controls to a minimum [13]. Cases and controls should be well-matched to avoid confounders, such as population stratification. Furthermore, sample sizes should be large, generally in the thousands, and replication sample sizes in an independent population should be even larger. The risk for false positive or spurious associations is high in GWA studies, and strict quality control is essential.

One of the interesting aspects about GWA studies is that usually, loci are identified that have not been previously suspected as candidate genes/loci [14]. Several reasons could be behind this. First, the marker may be a spurious association. Second, the identified marker may be in linkage disequilibrium (LD) with the etiological marker. In this case, the effect of different populations with different LD structures must be taken into account [15]. Third, the identified marker may be an etiological marker, but has not been previously targeted as having a role in the disease. In the latter two cases, novel biological pathways involved in the disease state may be uncovered, leading to a better understanding of disease processes and, potentially, the development of novel therapies.

A search of the National Cancer Institute (NCI)-National Human Genome Research Institute (NHGRI)'s Catalog of Published GWA Studies (http://www.genome.gov/26525834) and the literature uncovered a total of 48 GWA studies examining CAD and/or associated traits or markers with p ≤ 10⁻⁵. Eight GWA studies with risk factor associations of p ≤ 10⁻⁵ have been published looking specifically at CAD or MI (Table 1). The remainder of the 40 identified GWA studies have been published examining traits or markers associated with CAD including coronary artery calcification, blood pressure, LDL cholesterol, HDL cholesterol, triglycerides, Lp(a), C-reactive protein, type 2 diabetes mellitus, and body mass index.

Table 1.

Genome-Wide Association Studies of CAD and MI (p<5 × 10⁻⁵)

Phenotype	Chromo-somal Region	Reported Gene(s)	SNP	Risk Allele Frequency	P-value	Odds ratio	95% CI	Reference
CAD	1p13.3	CELSR2-PSRC1-SORT1	rs5999839	0.23	4 × 10−19	1.29	1.18–1.40	[18]
early-onset MI	1p13	CELSR2-PSRC1-SORT1	rs646776	0.81	1.5 × 10−8	1.17	1.11–1.24	[21]
early-onset MI	1p32	PCSK9	rs11206510	0.81	9.6 × 10−9	1.15	1.10–1.21	[21]
early-onset MI	1q41	MIA3	rs1746048	0.72	5.9 × 10−7	1.13	1.08–1.18	[21]
CAD	1q41	MIA3	rs17465637	0.29	1 × 10−6	1.2	1.12–1.30	[18]
CAD	1q43	NA	rs17672135	0.87	2 × 10−6	1.43	1.23–1.64	[16]
early-onset MI	2q33	WDR12	rs6725887	0.14	1.3 × 10−8	1.17	1.11–1.23	[21]
CAD	2q36.3	pseudogene	rs2943634	0.65	2 × 10−7	1.21	1.13–1.30	[18]
CAD	3q22.3	MRAS	rs9818870	NA	7.4 × 10−13	1.15	1.11–1.19	[92]
CAD	5q21	NA	rs383830	0.22	1 × 10−5	1.60	1.16–2.21	[16]
early-onset MI	6p24	PHACTR1	rs12526453	0.65	1.3 × 10−9	1.12	1.08–1.27	[21]
CAD	6q25	MTHFD1L	rs6922269	0.25	2 × 10−5	1.17	1.04–1.32	[16]
CAD	6q25.1	MTHFD1L	rs6922269	0.25	3 × 10−8	1.23	1.15–1.33	[18]
CAD	6q27	SCL22A3-LPAL2-LPA	rs2048327, rs3127599, rs7767084, rs10755578 (CTTG and CCTC haplotype)	NA	1.2 × 10−9 (CTTG), 4.2 × 10−15 (CCTC)	1.20 (CTTG) 1.82 (CCTC)	1.13–1.28 (CTTG) 1.57–2.12 (CCTC)	[86]*
early-onset MI	9p21	CDKN2A, CDKN2B	rs4977574	0.56	1.1 × 10−30	1.28	1.23–1.33	[21]
MI	9p21.3	CDKN2A, CDKN2B	rs10757278	0.45	1 × 10−20	1.28	1.22–1.35	[19]
CAD	9p21.3	intergenic	rs1333049	0.47	3 × 10−19	1.36	1.27 – 1.46	[18]
CAD	9p21.3	CDKN2A, CDKN2B	rs1333049	0.47	1 × 10−13	1.47	1.27–1.70	[16]
CAD	9p21.3	NA	rs10757274 and rs2383206	NA	NA	NA	NA	[17]
CAD	10q11.21	CXCL12	rs501120	0.13	9 × 10−8	1.33	1.20–1.48	[18]
early-onset MI	10q11	CXCL12	rs1746048	0.84	3.4 × 10−5	1.14	1.08–1.21	[21]
MI	12q24	SH2B3	rs3184504	0.38	8.6 × 10−8	1.13	1.08–1.18	[82]
CAD	12q24.41	HNF1A-C12orf43	rs2259816	NA	5 × 10−7	1.08	1.05–1.11	[92]
CAD	15q22.33	SMAD3	rs17228212	0.30	2 × 10−7	1.21	1.13–1.30	[18]
CAD	16q23	NA	rs8055236	0.20	6 × 10−6	1.91	1.33–2.74	[16]
early-onset MI	19p13	LDLR	rs1122608	0.75	1.9 × 10−9	1.15	1.10–1.20	[21]
CAD	19q12	NA	rs7250581	0.22	3 × 10−5	1.06	0.79–1.43	[16]
early-onset MI	21q22	SLC5A3-MRPS6-KCNE2	rs9982601	0.13	6.4 × 10−11	1.20	1.14–1.27	[21]
CAD	22q12	NA	rs688034	0.31	4 × 10−6	1.11	0.99–1.25	[16]

NA = not available

^*Genome-wide haplotype analysis

Chromosome 9, band p21.3

Multiple GWA studies have shown a highly significant association with various SNPs within a large LD block on chromosome 9 at band p21.3 (9p21.3) and CAD [16–21]. While results did not reach statistical significance to be included in the NHGRI catalog of published GWA studies, a GWA study involving the Framingham Heart Study also found an association with a 13 kb region on 9p21.3 with major CHD (p 2.5–3.5 × 10⁻⁴) [22]. Additionally, they observed that 7 SNPs in a 76 kb region around 9p21 had p<10⁻⁵ for major CHD and/or major cardiovascular disease (CVD).

Markers at the 9p21 locus have been shown to lead to a 15–20% risk for CAD in the 50% of Caucasian individuals heterozygous for the allele, and a 30–40% increased risk of CAD in the 25% of individuals homozygous for the allele [17]. Numerous follow-up case-control analyses have been performed and confirm the association of 9p21 and CAD [22–28]. A meta-analysis of case-control studies showed that the odds ratio per copy of the 9p21 risk allele was 1.29 (95% CI 1.22–1.37, p=0.0079) [29]. Abdullah et al demonstrated that four SNPs within the 9p21 region were significantly (p=6.61 × 10⁻⁷ to 1.87 × 10⁻⁸) associated with premature and familial MI and CAD (average age of onset 40.3 +/- 5.1 years) [26]. Another group similarly made the connection between 9p21 and familial CAD in a high-risk population with familial hypercholesterolemia [25]. Although one study found an association between 9p21 and CAD, the association did not hold up with incident events or prevalent MI [27]. Additionally, the Rotterdam Study did not observe an association between 9p21 genotype and coronary heart disease or myocardial infarction in a large cohort of individuals aged 55 years and older [30]. Thus, while the vast majority of studies have identified a strong association with 9p21 and CAD, an association with this locus and MI may be population-dependent and garners further investigation.

The primary GWA studies were done in cohorts of individuals of Caucasian, Northern European, and Canadian descent. Follow-up studies in Korean, Japanese, and Italian populations also established the 9p21-CAD association in those populations [31–34]. Additionally, a multi-ethnic Atherosclerotic Disease, Vascular Function, and Genetic Epidemiology (ADVANCE) study confirmed the association in whites and extended it to U.S. Hispanics and U.S. East Asians, but not African Americans [35]. The lack of association in African Americans was consistent with what was observed by McPherson, et al. [17]. However, in both studies examining the association with African Americans and in some of the other non-Caucasian-based studies, the sample sizes were not large, thus potentially limiting the statistical power of the studies. Nonetheless, the studies imply a consistent association of the 9p21 risk allele with CAD in individuals from a wide range of Asian and Caucasian backgrounds.

While the 9p21 association with CAD has been replicated on multiple occasions, associations between surrogate markers of atherosclerotic cardiovascular disease and 9p21 have remained less conclusive. Two studies did not identify an association between the 9p21 locus and carotid intima media thickness (IMT) [36, 37]. An additional study also did not demonstrate an association with 9p21 and abdominal aortic IMT [38]. Samani et al. postulated that CAD risk influenced by 9p21 occurs by a mechanism independent of carotid intimal thickening or endothelial dysfunction, and may instead affect coronary plaque stability [36]. Anderson et al. found that while 9p21 correlated with prevalence of angiographic coronary disease, it was unable to predict extent of disease [39]. The 9p21 locus was significantly correlated with angiographically characterized CAD in another study [24]. Horne et al., however, found that while 9p21 genotype was associated with the CAD phenotype, it was not associated with the severity or extent of CAD assessed angiographically [27]. They also did not find a strong association between history of MI and the 9p21 locus, as described above. They hypothesized that the 9p21 locus acted in the early stages, but not in the progression, of atherosclerosis. On the other hand, Ye et al. observed that the 9p21 genotype was associated with established carotid atherosclerosis and progression of atherosclerotic plaques as determined by carotid duplex scanning [37]. Thus, further investigations are necessary to help define where 9p21 fits into the picture of initiation or progression of atherosclerosis.

Other phenotypic correlations have been investigated with the 9p21 locus and CAD. In the Heart and Soul Study, 9p21 genotype was not associated with any echocardiographic parameter of cardiovascular structure and function (left ventricular hypertrophy, systolic dysfunction, diastolic disfunction, inducible ischemia, exercise capacity, mitral annular calcification, and aortic plaque) in individuals with known CAD [40]. Additionally, Chen et al. observed no significant association between 9p21 and quantitative indices of coronary atherosclerosis, such as minimal lumen diameter and number of coronary lesions or occlusions [41]. However, the 9p21 locus has been associated with coronary artery calcification, a reasonable marker of subclinical atherosclerosis [42].

In addition to CAD, the 9p21.3 locus has been associated with stroke, abdominal aortic aneurysms, and intracranial aneurysms [23, 28, 29, 43–46]. In part because of the relationship between the risk allele and stroke, and as discussed above, it has been suggested that the 9p21 risk allele may be related to plaque stability [36, 45]. The association of 9p21 with aneurysms also suggests an involvement in processes related to vessel wall integrity [23]. Bjorck et al. observed that the 9p21 locus was associated with abdominal aortic compliance and distensibility coefficients (measurements of arterial stiffness), further supporting a link between 9p21 and arterial wall diseases [38]. However, a recent GWA study of a Sardinian cohort did not report an association between 9p21 and arterial stiffness as determined by carotid-femoral pulse wave velocity [47].

Biology of 9p21

While the 9p21 association with CAD has been replicated on multiple occasions, the biological relevance of 9p21 is unclear at this time. The 9p21 region has been commonly implicated in the tumorigenesis of a variety of malignancies [48–51]. However, neither mouse models nor in vitro analyses have specifically implicated atherosclerotic processes with deletion or mutation of the 9p21 locus [52–55].

The risk-allele SNPs within the 9p21.3 locus, which spans approximately 50–60kb, have been defined as being in a “desert zone of the genome” (Figure 1) [56]. The 3' end of CDKN2B, encoding the cyclin-dependent kinase inhibitor tumor suppressor p15^INK4B, is in a LD block with a cluster of tightly linked SNPS at the 9p21 locus. Weaker linkage disequilibrium extends through CDKN2B to CDKN2A, which encodes another tumor suppressor p16^INK4B. The cyclin dependent kinases are involved in cell cycle regulation and transforming growth factor-β (TGF-β) cell cycle arrest [57]. In smooth muscle cells, TGF-β has been shown to have impaired signaling and reduced expression in atherosclerotic lesions and overexpression in abdominal aortic aneurysms [58–60]. Other studies have shown increased TGF-β levels in different stages of plaque development [61–63]. In fact, TGF-β1 has been proposed as a marker and potential therapeutic target for cardiovascular disease [60].

Figure 1.

The chromosome 9, band p21 region. The location of the CAD-associated region is shown in red, annotated genes are shown in purple, and the ANRIL anti-sense non-coding RNA is shown in green.

A gene encoding a large antisense non-coding RNA (ANRIL) spans almost the entire 9p21-CAD association region (Figure 1) [55]. ANRIL was discovered through deletion analysis of a family with hereditary melanoma-neural system tumors [64]. It was found that the expression of ANRIL coclusters with the expression of ARF, which is encoded for by an alternative exon 1 and exons 2 and 3 of CDKN2A. Atheromatous human vessels, abdominal aortic aneurysm walls, vascular endothelial cells, monocyte-derived macrophages, and coronary smooth muscle cells have all been shown to express ANRIL [43]. It has been speculated that transcription of the cyclin-dependent kinase genes may be regulated, at least in part, by ANRIL through RNA interference or some other mechanism [43, 44]. In fact, a p15^INK4B antisense construct located in the 5' region of ANRIL, was shown to cause silencing of p15^INK4B (encoded for by CDKN2B) in mouse embryonic stem cells through heterochromatin formation and DNA methylation [65]. Thus, a speculated mechanism for the 9p21 risk allele involves antisense regulation of CDKN2B (and/or CDKN2A), which could then affect signaling of TGF-β and/or additional cytokine(s) involved in cell cycle arrest/proliferation.

Another gene, MTAP, encoding methylthio-adenosine phosphorylase, is part of a different LD block that is closer to the chromosome 9 telomere. MTAP encodes for methylthio-adenosine phosphorylase (MTAP), an enzyme involved in polyamine metabolism that is deleted (along with p16^INK4B) in many cancers. Whether or not this enzyme is involved in CAD requires further investigation.

Clinical utility of genotyping 9p21.3

While the specific phenotypic associations of the 9p21 genotype are somewhat conflicting and not well defined at present, whether or not 9p21 has proven clinical utility in predicting cardiovascular risk is of significant importance. As described above, an approximate 15–40% increased risk for CAD has been observed (depending on number of 9p21 risk alleles carried by the individual). However the Women's Genome Health Study demonstrated that 9p21 genotype did not add significantly to prediction of cardiovascular risk compared to what can be assessed via established risk markers, high-sensitivity C-reactive protein (CRP), and family history of premature myocardial infarction [66]. Thus, while a 20% increased risk (for heterozygous 9p21 risk allele carriers) for CAD may be relevant in some populations, middle-aged women (who are generally at low risk for CAD) who carry the 9p21 allele will not have a high, or even intermediate, risk for cardiovascular disease, despite the additional risk due to the presence of the 9p21 allele. Additionally, Talmud et al. demonstrated that 9p21 genotype did not add significantly to the CHD predictive utility by conventional risk factors in the Framingham risk score algorithm in a cohort of healthy middle-aged Caucasian men [67]. They did note, however, that 9p21 genotype did improve reclassification of CHD risk. Accordingly, based on studies performed in patients with early-onset angiographic CAD, Anderson et al. concluded that the clinical utility of 9p21 genetic assessment might be in refining CHD risk classification [39]. Thus, overall, it is suggested that 9p21 genotype may not be useful in stratifying risk in some low-risk populations but may provide discrimination in intermediate-risk individuals.

The launch of a clinical genotyping assay for 9p21, in October, 2007, was met with mixed enthusiasm (http://www.theheart.org/article/817629.do). Concerns regarding genetic testing for 9p21 include inherent problems with testing for the 9p21 locus in isolation, lack of biological knowledge of this locus and how it relates to CAD, uncertainty of the value and applicability of information obtained from genotyping, and potential for false negative reassurance in individuals who do not carry the risk allele. Thus, while 9p21 has been repeatedly associated with CAD, demonstrates a risk that is independent of traditional risk factors, and may be a useful marker in some populations (e.g. intermediate risk individuals), 9p21 genotyping as standard of care is debatable.

Chromosome 1p, band 13.3

A risk allele at 1p13 was found to be associated with CAD and early-onset MI, and this correlation was confirmed in a large scale association analysis [18, 20, 21]. This risk allele is in a 97-kb region of LD containing the CELSR2, PSRC1, and SORT1 genes, and has been found to be strongly associated with low-density lipoprotein (LDL) and total cholesterol concentrations [68–74]. Based on expression studies correlated with LDL cholesterol levels, it was concluded that SORT1 and CELSR2 were the most likely candidate susceptibility genes at the 1p13.3 locus [75]. Sortilin is a pro-neurotrophin receptor encoded for by the SORT1 gene, and is involved in adipocyte and muscle glucose metabolism. Sortilin expression is downregulated in obesity and has been implicated in insulin resistance [76]. Sortilin has also been shown to be involved in uptake and degradation of lipoprotein lipase, an important enzyme for lipid hydrolysis [77]. In relation to this, the gene encoding lipoprotein lipase (LPL) at 8p21 has also been implicated in multiple GWA studies as associated with high-density lipoprotein (HDL) cholesterol and triglyceride levels and other metabolic traits [69, 70, 73, 74, 78–80]. CELSR2 encodes for cadherin EGF LAG seven-pass G-type receptor 2, but very little is known about its function. The third gene in the 1p13.3 gene complex is PSRC1, which encodes for proline/serine-rich coiled-coil 1, a microtubule-associated protein. The functional connection between PSRC1 and CAD and/or lipid metabolism is not known. Thus, while there is a potential connection between the biology of SORT1 and cholesterol levels and CAD, the picture involving CELSR2 and PSRC1 is less clear and further studies are needed to elucidate those mechanisms.

Other loci

Other loci that have been identified through multiple, independent GWA studies and confirmed in follow-up CAD association analyses include 1q41 and 10q11 (Table 1) [16, 18, 20, 21]. Interestingly, 10q11 was shown to have a significant relationship to CAD in women but not in men [20]. Another locus, 6q25, was also identified to be associated with CAD in multiple, independent GWA studies, but the correlation was not statistically significant in a large-scale association analysis [20]. Loci within the 12q23-24 region were found to be associated with CAD, MI, and coronary artery calcification in three separate GWA studies [42, 81, 82]. The reported genes in these studies are DRIM (12q23), SH2B3 (12q24), and HNF1A-C12orf43 (12q24.41). The HNF1A gene has also been implicated for its association with LDL cholesterol and C-reactive protein in other GWA studies [79, 83, 84].

Two genes, LDLR and PCSK9, in which mutations can lead to autosomal dominant hypercholesterolemia, were reported in a GWA study of early-onset myocardial infarction [21]. Mutations in LDLR (on 19p13), which encodes for the LDL receptor, leads to familial hypercholesterolemia. Gain of function mutations in PCSK9 (on 1p32), which encodes for proprotein convertase, subtilisin/kesin-type 9 and is involved in recycling of the LDL receptor, leads to autosomal dominant hypercholesterolemia 3. SNPs in these genes have also been identified in other GWA studies as associated with LDL cholesterol [69, 70, 74, 79].

Copy number variant and haplotype GWA analyses

Genome-wide association studies involving the univariate analysis of single nucleotide polymorphisms (SNPs) have been the standard to date. In contrast to polymorphisms at single base positions (SNPs), copy number variants (CNVs) are polymorphic deletions or duplications of large, (1 kilobase to several megabase) regions of the genome, and could likewise be associated with specific disease states. In that regard, The Myocardial Infarction Genetics Consortium explored the possibility of an association between copy number variants (CNVs) and early-onset myocardial infarction [21]. However, they did not detect a significant difference in CNVs in cases vs. controls, in genes vs. the genome, or at any individual locus.

In addition to univariate analysis of SNPs, there is the potential for SNPs in haplotype to be associated with certain phenotypes. Tregouet et al. recently performed a genome-wide haplotype association study in CAD, using a sliding-windows approach [85]. Using this type of analysis, a haplotype of four SNPs in the SLC22A3-LPAL2-LPA gene cluster was found to be associated with CAD. Of particular interest in this gene cluster is LPA, which encodes apolipoprotein (a), the protein component of lipoprotein (a) [Lp(a)]. Elevated levels of Lp(a) have been associated with an increased risk for CAD and MI [86–88].

Conclusion

GWA analyses are a powerful tool for evaluating genetic associations in complex disease. However, as described above, careful design and execution of GWA and follow-up association studies are imperative to providing robust, meaningful data. A major challenge with GWA analyses is separating the true markers from the spurious associations.

Having stringent p-values is important, as is replicating results in independent cohorts. Additionally, comprehensive phenotyping of the cohort is also beneficial for understanding biological pathways. Extensively characterized cohorts will provide the foundation for ascertaining genetic and environmental risk factors in an integrated, interdependent manner.

For the vast majority of published GWA studies, the ethnic background of the primary study population has been European Caucasian. However, there is a need to extend association analyses to populations with other ethnic backgrounds. Linkage disequilibrium blocks and allele frequencies can vary widely between European Caucasians, African Americans, Asians, Hispanics, and other ethnic groups. Thus, associations found in one ethnic group may not translate to the same association in other ethnic groups. A case in point is 9p21.3, in which an association between this locus and CAD was observed in many different populations, but the association was not present in African Americans [17, 35].

In addition to confounding issues in association studies due to ethnicity, age also should be taken into account. In fact, a stronger impact of genetic factors on CAD is observed in younger individuals, possibly due to heterogeneity of effect associated with advancing age [2]. The 9p21 risk allele had a more pronounced effect in individuals with earlier onset MI, as observed by Helgadottir et al. [19]. Additionally, the Rotterdam Study was unable to make a correlation with 9p21 and CHD or MI in a prospective cohort study of nearly 8000 participants aged 55 years and older [30]. The authors cautioned, however, that the results of the study may be too underpowered to generalize them to an elderly population. Proper characterization of cohorts, as well as matching for age, gender, and ethnicity, is important to prevent misrepresentation of the findings.

While GWA studies have identified novel potential markers for CAD and other complex diseases, there are several limitations to the results of these studies [89]. These limitations include the risk of false positive results, lack of detection of rare variants due to the insensitive nature of the assay, biases from incomplete phenotyping of cases and controls, and confounders due to population stratification. Furthermore, very few functional variants have been identified through GWA studies. Thus, further studies are necessary to ascertain the relationship of the identified variants to functional variants and biology of the disease.

Because of the nature of complex diseases, it is likely that multiple genetic risk alleles are needed to accurately assess risk. The risk alleles may be identified through association and linkage studies, but many of them may be identified by downstream studies investigating interactions between multiple risk loci. One study, using computer simulation, estimated that over 200 alleles were required to provide a reasonable assessment of CAD risk [90]. Thus, while GWA studies may be useful in identifying potential novel markers for complex disease, essential follow-up investigations involve determining the biology of the markers, interactions between loci, effect of risk markers in subclinical phenotypes, and association in other populations.