Author + information
- Received October 3, 2007
- Accepted November 14, 2007
- Published online July 29, 2008.
- Shu Ye, MD, PhD, MRCPath⁎,⁎ (, )
- Johann Willeit, MD†,
- Florian Kronenberg, MD‡,
- Qingbo Xu, MD, PhD§ and
- Stefan Kiechl, MD†
- ↵⁎Reprint requests and correspondence:
Dr. Shu Ye, Clinical Pharmacology, William Harvey Research Institute, John Vane Science Centre, Charterhouse Square, London EC1M 6BQ, United Kingdom.
Objectives Following the recent novel finding from genomewide association studies that sequence variation on chromosome 9p21 is a genetic factor for coronary artery disease, we investigated whether the genetic variant influenced the development of atherosclerosis and its progression in a population-based, prospective study.
Background Recently, several genomewide association studies revealed a highly significant association between variation on chromosome 9p21 and risk of coronary artery disease.
Methods We studied the rs1333049 polymorphism located on chromosome 9p21 in a cohort of 769 individuals who participated in the Bruneck study with long-term follow-up data on carotid atherosclerosis measured by high-resolution duplex ultrasound and incident cardiovascular disease.
Results The C allele was associated not only with prevalent carotid atherosclerosis (odds ratio [OR]: 1.46 [95% confidence interval (CI): 1.13 to 1.88]; OR: 1.43 [95% CI: 1.11 to 1.84]; and OR: 1.44 [95% CI: 1.11 to 1.87] for each copy of C allele, calculated from data collected in 1990, 1995, and 2000, respectively), but also with progression of atherosclerosis (OR: 1.73 [95% CI: 1.36 to 2.21] during 1990 to 1995, and OR: 1.87 [95% CI: 1.44 to 2.42] during 1995 to 2000). In addition, the C allele was related to incident cardiovascular disease (hazard ratio: 1.37 [95% CI: 1.05 to 1.79]). There was evidence of an interaction between genotype and abdominal obesity on atherosclerosis and cardiovascular risk.
Conclusions The results of this population-based, prospective study indicate that the sequence variation on chromosome 9p21 influences atherosclerosis development and progression.
Recently, 4 genomewide association studies reveal a highly significant association of single nucleotide polymorphisms (SNPs) on chromosome 9p21 with risk of coronary artery disease (CAD) (1–4). Association with the disease was detected for multiple SNPs across a genomic region of approximately 100 kilo base pairs, with substantial linkage disequilibrium between the SNPs (1–4). This genomic region contains no annotated gene but is adjacent to the tumor suppressor genes CDKN2A and CDKN2B, which play an important role in the regulation of the cell cycle. The mechanistic basis for the association is still unknown. It is possible that the genetic variant can increase the development of atherosclerotic plaques, promote atherosclerotic plaque rupture, increase thrombogenesis, or act via other mechanisms such as reperfusion injury.
Taken together, the above-mentioned genomewide association studies indicate that the risk of CAD is increased by ∼15% to 35% in heterozygotes and by ∼30% to 70% in homozygotes of the risk allele, which was estimated by comparing patients with myocardial infarction or a history of coronary revascularization and controls (1–4). True risk assessment in the general population, however, should be addressed in population-based, prospective studies. In addition, it would be warranted to investigate whether this newly identified genetic factor increases vascular risk by promoting atherosclerosis development and/or progression, and whether there are gene-environment and gene-risk factor interactions. In this study, we examined the chromosome 9p21 variants in the population-based Bruneck cohort and sought to address these questions.
Subjects and Methods
The Bruneck Study is a prospective, population-based survey of atherosclerosis (1–4). The study protocol was approved by the appropriate ethics committees, and all study subjects gave written informed consent before entering the study. At the study outset in 1990, a gender- and age-stratified random sample of all inhabitants of Bruneck (Bolzano Province, Italy), ages 40 to 79 years (125 women and 125 men in each of the fifth to eighth decade; total n = 1,000), was drawn and the individuals were invited to take part in the study. A total of 93.6% participated, and data collection was completed for 919 subjects. Between 1990 and the first follow-up in 1995, 63 subjects died or moved away, and the 1995 follow-up was completed for 96.5% (n = 826) of the remaining subjects. A second follow-up took place in 2000, and it was 100% complete for fatal and nonfatal incident cardiovascular disease (CVD) and 95% complete for carotid ultrasound data among the survivors. A third follow-up was undertaken in 2005. Blood samples for deoxyribonucleic acid extraction were taken as part of the follow-up in 1995. Genotyping was performed for 787 participants, of which 769 were successfully genotyped and form the current study population. The characteristics of this sample were identical to those of the entire population.
Clinical history and examinations
Data on smoking status and regular alcohol consumption (grams/day) were recorded. Hypertension was defined as blood pressure (mean of 3 measurements) ≥140/90 mm Hg or the use of antihypertensive drugs. Diabetes mellitus was defined as fasting glucose levels ≥140 mg/dl (7.8 mmol/l) and/or 2-h glucose values in oral glucose tolerance test ≥200 mg/dl (11.1 mmol/l) (World Health Organization definition). Body mass index was calculated as weight divided by height squared (kg/m2). Waist and hip circumferences (to the nearest 0.5 cm) were measured by a plastic tape meter at the level of the umbilicus and of the greater trochanters, respectively, and waist-to-hip ratios (WHRs) were calculated. Framingham risk score was calculated as described in the literature (5–12).
The ultrasound protocol involves the scanning of the right and left internal (bulbous and distal segments) and common carotid arteries (proximal and distal segments). Scanning was performed with a 10-MHz imaging probe and a 5-MHz Doppler probe (13). Intima-media thickness was measured at plaque-free sections of the far wall of the common carotid arteries (intraobserver coefficient of variation: 7.9%, n = 100) as the distance between the lumen-intima and media-adventitia interfaces. In the analysis, the mean maximum intima-media thickness of the left and right common carotid arteries was used. Atherosclerotic lesions were defined according to 2 ultrasound criteria: 1) wall surface (protrusion or roughness of the arterial boundary); and 2) wall texture (echogenicity). The maximum axial diameter of plaques (in mm) was assessed on the near and far walls at each of 8 vessel segments. The atherosclerosis score was calculated by summing all diameters (intraobserver coefficient of variation: 13.5% and 14.0% [n = 100]) and serves as a measure of atherosclerosis severity (range 0 to 29.7 mm) (5–7). The scans used in the current analyses were performed in 1990, 1995, and 2000 by the same experienced sonographer, who was unaware of the subjects' characteristics. On the basis of the follow-up evaluation, 2 different stages of atherogenesis were differentiated: 1) Early atherosclerosis progression was defined by the occurrence of atherosclerotic lesions in segments previously free of atherosclerosis or enlargement of nonstenotic lesions by a relative increase in the plaque diameter exceeding twice the measurement error of the method. 2) Incident vessel stenosis (advanced atherogenesis) was assumed whenever the progression criterion was met and a narrowing of the lumen >40% (diameter stenosis) was achieved (5–7). The 2 progression categories were highly reproducible (kappa coefficients: >0.8 [n = 100]). Further details of imaging procedures and measurement errors have been published elsewhere (5–7).
Assessment of cardiovascular events
In the main analysis, the cardiovascular end point comprised all incident cases of myocardial infarction (fatal and nonfatal), ischemic stroke (fatal and nonfatal), transient ischemic attack, new-onset symptomatic peripheral artery disease, and any revascularization procedure (n = 103). Myocardial infarction was deemed confirmed when World Health Organization criteria for definite disease status were met. Stroke and transient ischemic attack were classified according to the criteria of the National Survey of Stroke. The diagnosis of symptomatic peripheral arterial disease required a positive response to the Rose questionnaire (typical claudication), with the vascular nature of complaints confirmed by standard diagnostic procedures (ankle-brachial pressure index or angiography), or an acute peripheral artery occlusion requiring revascularization. All other revascularization procedures (angioplasty and surgery) were carefully recorded. Ascertainment of events or procedures did not rely on hospital discharge codes or the patient's self-report but on a careful review of medical records provided by the general practitioners and files of the Bruneck Hospital and the extensive clinical and laboratory examinations performed as part of the study protocols. A major advantage of the Bruneck Study cohort is that virtually all subjects living in the area of Bruneck were referred to the local Bruneck Hospital and that the network existing between the local hospital and the general practitioners allowed the retrieval of virtually all medical information on people living in the area.
Blood samples were drawn after an overnight fast and 12-h abstinence from smoking. High-density lipoprotein (HDL) cholesterol was determined with the use of an enzymatic method (CHOD-PAP [cholesterol oxidase phenol 4-aminoantipyrine peroxidase] method, Merck, Darmstadt, Germany; coefficient of variation: 2.2% to 2.4%). Low-density lipoprotein (LDL) cholesterol was calculated with the Friedewald formula except in subjects with triglycerides >4.52 mmol/l in whom it was directly measured. Markers of inflammation and all other laboratory parameters were assessed by standard methods as detailed previously (5–7).
Determination of genotypes
The deoxyribonucleic acid was extracted from whole blood samples using a salting out method (Invisorb Blood Universal Kit, Berlin, Germany) within the Genotyping Unit of the Gene Discovery Core Facility of the Innsbruck Medical University. Genotypes for the rs1333049 and rs10757274 SNPs were determined using the 5′ nuclease assay with the Applied Biosystems (Foster City, California) TaqMan probes (assay identification C__1754666_10 and C__26505812_10, respectively). Each assay was performed in a 2-μl final volume with 10 ng of dried deoxyribonucleic acid, 200 μmol/l of each probe, 900 μmol/l of each polymerase chain reaction primer, and 1× Thermo Scientific ABsolute qPCR Rox mix (Thermo Fisher Scientific Inc., Surrey, United Kingdom), and the reactions were performed in 384-well microplates with MJ Research Tetra thermal cyclers (MJ Research, Reno, Nevada). The cycling conditions were 50°C for 2 min and 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. Fluorescence was measured using an ABI Prism 7900HT Sequence Detection System and analyzed using the ABI Prism SDS software version 2.1 (both from Applied Biosystems).
Data were analyzed using the SPSS 12.0 (SPSS Inc., Chicago, Illinois) and BMDP (Statistical Solutions, Saugus, Massachusetts) software packages. Continuous variables are presented as means ± SD or medians (interquartile range), and dichotomous variables are presented as numbers and percentages. Differences in population characteristics and levels of vascular risk factors between subjects with distinct rs1333049 genotypes (GG, GC, and CC) were analyzed using analysis of variance and the chi-square test. Variables with a skewed distribution were loge-transformed to satisfy the assumption of normality and constant variance of the residuals. Cox proportional hazards regression models were used to assess whether rs1333049 genotypes predict the risk for incident CVD. Proportional hazards assumptions were satisfied. The associations of rs1333049 genotypes with presence, severity, and progression of carotid atherosclerosis were tested by means of logistic and linear regression analysis. Base models were adjusted for age and gender. For ease of presentation, all multivariable models were adjusted for the same set of variables (age, gender, smoking status, level of alcohol consumption, presence or absence of hypertension, diabetes mellitus, body mass index, WHR, LDL cholesterol, HDL cholesterol, loge-transformed lipoprotein(a), fibrinogen, antithrombin III, the factor V Leiden mutation, loge-transformed high-sensitivity C-reactive protein, ferritin, and loge-transformed urinary albumin). Alternative models built by a forward stepwise selection procedure or adjusted for sets of covariates that were assessed in previous analyses of the vascular risk profiles of the Bruneck Study (7–12) yielded identical results (data not shown). Differential associations in subgroups (gene × risk factor interactions) were analyzed by inclusion of appropriate interaction terms. All p values were 2-sided.
We genotyped the Bruneck Study subjects for the rs1333049 and rs10757274 SNPs, both located on chromosome 9p21, as rs1333049 was most significantly associated with CAD risk in the genomewide scans by the U.K. Wellcome Trust Case-Control Consortium and the German Cardiogenics Consortium (4) and is in complete linkage disequilibrium (based on data for Caucasians in HapMap ) with the lead SNP (rs10757278) in the genomewide scan by Helgadottir et al. (2), and rs10757274 was the most significant SNP in the genomewide scan by McPherson et al. (1) (data for rs10757274 unavailable for Caucasians in HapMap, and therefore it was not possible to assess whether this SNP was in linkage disequilibrium with rs1333049 based on HapMap data). An analysis of the rs1333049 and rs10757274 genotypes of the Bruneck subjects showed that the 2 SNPs were in strong linkage disequilibrium (r2 = 0.89) (Online Table). Their relationships with atherosclerosis and cardiovascular disease in the Bruneck Study were found to be similar, and therefore only the rs1333049 SNP is presented hereafter.
The genotype distribution of rs1333049 in the Bruneck population was 196 (GG), 373 (GC), and 200 (CC) and was consistent with Hardy-Weinberg equilibrium (p = 0.41). The C allele frequency was 0.50, similar to reported frequencies in the United Kingdom (0.55 in CAD cases and 0.47 in controls) and Germany (0.54 in CAD cases and 0.48 in controls) (4).
The primary end point measurement in the Bruneck Study was carotid atherosclerosis, which was determined by high-resolution duplex ultrasound measurements carried out at the outset of the study in 1990 and subsequently in the 1995 and 2000 follow-ups. There was an association between the C allele of the rs1333049 SNP and prevalent carotid atherosclerosis determined in 1900, 1995, and 2000, respectively (odds ratio [OR]: 1.46 [95% confidence interval (CI): 1.13 to 1.88], OR: 1.43 [95% CI: 1.11 to 1.84], and OR: 1.44 [95% CI: 1.11 to 1.87], respectively, for each copy of C allele), with adjustment for age and gender (Table 1). The relationships remained significant in multivariable analyses with adjustment for age, gender, smoking, alcohol consumption, hypertension, diabetes mellitus, body mass index, WHR, LDL cholesterol, HDL cholesterol, lipoprotein(a), C-reactive protein, fibrinogen, antithrombin III, factor V Leiden mutation, ferritin, and urinary albumin concentration (Table 1).
There was an association between the C allele and progression of carotid atherosclerosis. During the period of 1990 to 1995, odds ratio of atherosclerosis progression for each copy of the C allele was 1.73 (95% CI: 1.36 to 2.21, p < 0.001), adjusting for age and gender, and 1.69 (95% CI: 1.31 to 2.17, p < 0.0001) in a multivariable analysis with adjustment for the additional variables described in the previous paragraph (Table 1). Similarly, during the period of 1995 to 2000, odds ratio of atherosclerosis progression for each copy of the C allele was 1.87 (95% CI: 1.44 to 2.42, p < 0.0001), adjusting for age and gender, and 1.98 (95% CI: 1.51 to 2.61, p < 0.001) in the multivariable analysis (Table 1).
A relationship with the C allele was also observed for advanced atherogenesis defined by the development of vessel stenosis >40%, and this was seen for both the 1990 to 1995 and 1995 to 2000 periods (Table 1). There was also a tendency of association between the number of copies of the C allele and severity of carotid atherosclerosis estimated by the atherosclerosis score (Table 1). The highest C allele frequency was seen in subjects with the most advanced atherosclerosis and this was especially true for younger individuals. When categorizing subjects in 3 groups based on 1995 measurements—no carotid atherosclerosis, mild-to-moderate atherosclerosis, severe atherosclerosis (defined as an atherosclerosis score exceeding the 90th percentile corresponding to 10 mm)—C allele frequency was 47.7%, 55.0%, and 80.8%, respectively, in subjects age <65 years of age (p < 0.001) and 43.4%, 49.5%, and 56.2%, respectively in subjects age ≥65 years (p = 0.046). No association was detected for carotid intima-media thickness, a measurement of vascular pathology prior to formation of atherosclerotic plaques.
In addition, the C allele was associated with incident CVD defined by coronary, cerebral, and/or peripheral artery events. Hazard ratio of overall CVD for each copy of the C allele was 1.37 (95% CI: 1.05 to 1.79), adjusting for age and gender, and 1.35 (95% CI: 1.02 to 1.78) in a multivariable analysis (Table 1). Figure 1 shows cumulative hazard curves of incident CVD from 1995 to 2005 for groups of subjects of the GG, GC, and CC genotypes, with multivariable adjustment. Relationships with the C allele appeared to be similar for coronary events and cerebral and peripheral artery events (Table 1).
There was no association between the rs1333049 SNP and classic cardiovascular risk factors including gender, age, smoking, hypertension, body mass index, and Framingham risk score nor was there an association between the SNP and plasma levels of LDL cholesterol, HDL cholesterol, lipoprotein(a), C-reactive protein, and fibrinogen (Table 2). The percentage of diabetics was higher among subjects with the CC genotype (Table 2), but there was no significant difference in number of new diabetic cases (1995 to 2005) among the different genotype groups (data not shown).
There were no differential effects of the rs1333049 SNP on atherosclerosis and CVD risk in men and women nor in subgroups defined by age (<65 years, ≥65 years), diabetes, hypertension, smoking, body mass index (<median, ≥median), LDL cholesterol (<median, ≥median), and high-sensitivity C-reactive protein (<median, ≥median). Remarkably, however, we found a significant gene-to-risk factor interaction with the WHR defining abdominal obesity. Odds ratios for atherosclerosis progression were 3.41 (95% CI: 1.96 to 5.92) and 2.06 (95% CI: 1.41 to 3.02) during 1990 to 1995 and 1995 to 2000, respectively, in subjects with WHR equal to or above the median (=0.933) and 1.40 (95% CI: 1.06 to 1.84) and 1.70 (95% CI: 1.18 to 2.43) during 1990 to 1995 and 1995 to 2000, respectively, in those with WHR below the median (p = 0.001 and p = 0.044 for interaction between genotype and WHR on atherosclerosis progression measured in the 1990 to 1995 and 1995 to 2000 periods, respectively, with adjustment for age and gender, and p = 0.003 and p = 0.028 in the multivariable analysis with adjustment for the additional variables mentioned previously). Hazard ratios of CVD were 1.74 (95% CI: 1.22 to 2.48) in subjects with a WHR equal to or above the median and 1.01 (95% CI: 0.65 to 1.57) in those below the median (p = 0.006 for interaction between genotype and WHR on CVD adjusting for age and gender, and p = 0.019 in the multivariable analysis). The interactions remained significant after exclusion of subjects with diabetes mellitus and also remained significant after adjustment for homeostasis model assessment of insulin resistance index and plasma adiponectin levels. In all, the effect of rs1333049 SNP on vascular disease appears to depend on the degree of abdominal obesity and vice versa the effects of abdominal obesity on vascular disease risk appear to be determined by the rs1333049 SNP. Figure 2 shows hazard ratios of incident CVD calculated for each 0.1 unit increment in WHR in groups of subjects of the GG, GC, and CC genotypes.
In this population-based, prospective cohort study of atherosclerosis and incident CVD, we assessed the effects of the rs1333049 SNP that was most significantly associated with risk of CAD in the genomewide scans by the U.K. Wellcome Trust Case-Control Consortium and the German Cardiogenics Consortium (4) and is in nearly complete linkage disequilibrium with the lead SNPs in the genomewide scans by McPherson et al. (1) (lead SNP rs10757274) and Helgadottir et al. (2) (lead SNP rs10757278). The results of our prospective study confirm the finding from the genomewide association studies that the genetic variant (rs1333049 per se or another SNP in linkage disequilibrium with rs1333049) is a novel genetic factor for CAD (1–4). The population-based, prospective nature of our study allows true assessment of cardiovascular risk related to the genetic variants in the general population. In brief, the hazard ratio for CVD related to each copy of the C allele of rs1333049 was 1.37 with adjustment for age and gender and 1.35 in the multivariate analysis. This finding corroborates the results from the case-control analyses by the U.K. Wellcome Trust Case-Control Consortium and the German Cardiogenics Consortium, in which the odds ratio for CAD for each copy of the C allele of rs1333049 was 1.37 and 1.33, respectively (4). The risk estimate for rs1333049 in the aforementioned 3 studies is similar but not identical to the risk estimate for rs10757274 in the study by McPherson et al. (1) and rs10757278 in the study by Helgadottir et al. (2). In the study by McPherson et al. (1), rs10757274 was associated with an ∼15% to 20% increase in CAD risk in heterozygotes and ∼30% to 40% increase in homozygotes of the risk allele. In the study by Helgadottir et al. (2), the odds ratio for myocardial infarction associated with rs10757278 was 1.26 in heterozygotes and 1.64 in homozygotes of the risk allele.
Another key finding of our study is that the influence of the chromosome 9p21 variant is not restricted to CAD. In particular, the study shows that the genetic variant is associated with atherosclerosis in carotid arteries, and that it influences not only the development but also progression of atherosclerosis. This novel finding is useful for understanding the mechanistic basis of this genetic factor. It indicates that the association between the genetic variant and CVD is in part mediated by promoting the development and progression of atherosclerosis, the pathology underlying the majority of CAD and stroke.
We found no association between the genetic variant and classic cardiovascular risk factors, which is consistent with the findings from the study of McPherson et al. (1). In addition, we detected no association between the genetic variant and carotid artery intima-media thickness, a measurement of vascular pathology prior to formation of atherosclerotic plaques. However, we observed evidence of an interaction between the genetic variant and abdominal obesity measured by WHR on carotid atherosclerosis and CVD, such that the association between the genetic variant and atherosclerosis and CVD was more pronounced in individuals with high WHR and vice versa effects of abdominal obesity on vascular risk were determined by the genetic variant (Fig. 2). This finding, although observed both for clinical end points and the main ultrasound outcome measure (progression of atherosclerosis), awaits confirmation in future investigations and should be viewed as preliminary. If this finding proves to be true, it could provide an interesting clue for a better understanding of the association between rs1333049 and vascular risk.
Multiple SNPs in a genomic region of approximately 100 kilo bases on chromosome 9p21 were associated with CAD in the genomewide association studies (1). These SNPs were in strong linkage disequilibrium (1–4), indicating that they are likely to represent the same genetic signal. This genomic region contains no annotated gene but only several expressed sequence tags, none of which contains an open reading frame. The genes that are located nearest to this region are CDKN2A and CDKN2B, which are important regulators of cell cycle and could potentially have an influence on the development and progression of atherosclerosis through affecting vascular cell proliferation or apoptosis (1–4). The functional effects of the genetic variants in the 9p21 region, however, remain unknown. It is possible that certain genetic variant(s) in this region affect the expression of CDKN2A, CDKN2B, and/or other genes located nearby (15,16).
In summary, the results of this study corroborate the findings from recent genomewide association studies that genetic variability on chromosome 9p21 is an important cardiovascular risk factor. The population-based, prospective nature of our study provides a good estimate of the population risk associated with this genetic variant. In addition, our study shows that the association of this genetic factor is not restricted to CAD, but is also related to carotid atherosclerosis, suggesting a general effect on atherosclerosis-related CVDs. The study shows that the genetic variant has an influence on both susceptibility to and progression of atherosclerosis, providing an understanding of the mechanistic basis of this genetic factor. Finally, the study provides preliminary novel evidence that effects of the genetic variant are linked to the degree of abdominal obesity, a key feature of the metabolic syndrome.
For a supplementary table outlining the genotype data (number) of rs1333049 versus rs10757274, please see the online version of this article.
Association of Genetic Variation on Chromosome 9p21 With Susceptibility and Progression of Atherosclerosis: A Population-Based, Prospective Study
Supported by the “Genomics of Lipid-associated Disorders—GOLD” grant of the Austrian Genome Research Programme GEN-AU (to Dr. Kronenberg); funding from Pustertaler Verein zur Prävention von Herz- und Hirngefaesserkrankungen, Sanitaetseinheit Ost, and Assessorat fuer Gesundheit Province of Bolzano, Italy; and grants from the British Heart Foundation (Drs. Ye and Xu).
- Abbreviations and Acronyms
- coronary artery disease
- confidence interval
- cardiovascular disease
- high-density lipoprotein
- low-density lipoprotein
- single nucleotide polymorphism
- waist-to-hip ratio
- Received October 3, 2007.
- Accepted November 14, 2007.
- American College of Cardiology Foundation
- McPherson R.,
- Pertsemlidis A.,
- Kavaslar N.,
- et al.
- Helgadottir A.,
- Thorleifsson G.,
- Manolescu A.,
- et al.
- Kiechl S.,
- Willeit J.
- Kiechl S.,
- Willeit J.
- Willeit J.,
- Kiechl S.,
- Oberhollenzer F.,
- et al.
- Tsimikas S.,
- Kiechl S.,
- Willeit J.,
- et al.
- Mayr M.,
- Kiechl S.,
- Tsimikas S.,
- et al.
- Kiechl S.,
- Schett G.,
- Schwaiger J.,
- et al.
- Kiechl S.,
- Willeit J.,
- Mayr M.,
- et al.
- Wilson P.W.,
- D'Agostino R.B.,
- Levy D.,
- Belanger A.M.,
- Silbershatz H.,
- Kannel W.B.
- ↵International HapMap Project. http://www.hapmap.org/. Accessed July 2007.
- Matthews C.,
- Gorenne I.,
- Scott S.,
- et al.