It has been considered for long that genes play a substantial role in susceptibility to coronary heart disease (CHD) [1]. Up to now, a limited number of these genes have been identified through the candidate gene approach and genome wide linkage studies. Recently a number of genome wide association (GWA) studies have identified several genetic variants on chromosome 9p21 associated with the risk of CHD. McPherson et al. found a Single Nucleotide Polymorphism (SNP), rs10757274, on chromosome 9p21 associated with the risk of CHD [2]. Helgadottir et al. found a close-by SNP, rs10757278, in the same 9p21 region associated with the risk of myocardial infarction (MI) [3]. These findings were followed by another GWA study by Samani et al [4], which found rs1333049 to be associated with the risk of coronary artery disease [4]. All three SNPs are located within the same Linkage Disequilibrium (LD) block on chromosome 9 approximately 22 million base pairs from the 9p telomere, adjacent to two tumor suppressor genes, CDKN2A and CDKN2B. These genes are involved in regulation of cell proliferation. Abnormal proliferation is one of the characteristics of atherosclerosis, one of the pathological features of CHD and MI [5].

To date, the findings are replicated in several case-control studies comprising 12285 cases and 23184 controls and two cohort studies comprising 22056 subjects [6–10]. These replications has made this locus one of the best replicated findings for genetic susceptibility to cardiovascular diseases. Though this findings are promising, they will be of more clinical worth if translated to older patients who constitute a larger part of the patients. We chose to study rs10757278 and rs10757274 because they were most strongly and consistently associated with CHD and MI in GWA studies. The leading SNP of the study by Samani et al, rs1333049, is in the same LD block with rs10757278 and contributes to the same haplotype alleles. We attempted to replicate the association in the Rotterdam Study, a population-based cohort study among older subjects, but found no association.

We repeated the analysis with incident cases limited to those occurred before age 70. Age and sex adjusted HR (95% CI) for CHD and MI were 1.00 (0.97, 1.04) and 0.90 (0.74, 1.09) per copy of G allele of rs10757274. The corresponding HRs for MI were 1.00 (0.96, 1.04), and 0.91 (0.75, 1.10) per copy of G allele of rs10757278.

To investigate whether any of the covariates affect the relation of SNPs with CHD and MI, we repeated the analysis in subgroups of age, sex, family history of cardiovascular disease, HDL cholesterol, diabetes, hypertension, smoking, and history of CHD (figure 1). The association was not significant in any of the studied subgroups and no significant interaction was found.

Our main goal was to replicate the results of recent GWA studies on CHD and MI in a population based study. Two SNPs, rs10757274 and rs10757278, which were most consistently and strongly associated with the risk of CHD in GWA studies were studied. We not only found no significant association between these two SNPs and the risks of CHD and MI, but also found, however non-significant, an inverse direction for the risk. We also did not find an association in subgroups of cardiovascular risk factors.

Different approaches have been used in recent decades to discover causal genes for cardiovascular diseases. A novel approach is the GWA study which searches large part of the genome for predisposing variants. Contrary to the formerly common approach, the candidate gene study, the GWA study is a hypothesis free approach, i.e. it holds no prior assumption on the location of predisposing genes. As an advantage, this approach promises a more comprehensive understanding of the causal genes. However, this method is liable to false positive findings. Hence, GWA studies always need to be replicated in independent samples to confirm their findings.

Our study had sufficient power to detect effect sizes as shown in the published studies. In an additive model and for a SNP with a minor allele frequency of 0.45 (lowest minor allele frequency: 0.43, shown in Icelandic population A [3]), our study had more than 80% power to detect a relative risk of 1.15 for CHD (the lowest effect: 1.18, shown in ARIC study [2]) and 1.23 for MI (the lowest effect: 1.25, shown in the Iceland population A) [14].

We did not find any of the SNPs to be associated with the risk of CHD in our study. One legitimate conjecture for the inconsistency of our results with former studies may be heterogeneity of the effect. Compared to the Rotterdam Study, most of the studied populations comprised young CHD or MI patients. If the risk allele on chromosome 9p21 invokes only early onset of CHD, the effect in older subjects may not be large enough to be found in our study. Therefore, our negative finding may point to a heterogeneity of effect by age. In agreement with this conjecture, Helgadottir et al. showed that the association was stronger when only those with early onset MI were considered [3]. However, we failed to find an evidence of age affecting the association in our data. The strength of the association did not change materially when we limited the incident cases to those developed CHD or MI before age 70. Moreover, the strength of the association was not significantly different in age subgroups (figure 1). We emphasis that our study may be underpowered to detect the effect of age on the association. It is noteworthy that the heterogeneity of effect has a particular clinical and public health impact. In Western countries the majority of morbidity and mortality from CHD occurs in elderly people. In the Netherlands, 74% of men and 91% women who experience fatal MI are older than 65 years http://statline.cbs.nl/statweb/. The fact that CHD is less common in younger population implies that these subjects are not a good representative of the general population of CHD patients. Therefore, caution should be taken in generalizing the results of the published studies to an elderly population.

One may also speculate that those carrying the risk allele had developed CHD at early age and were excluded at the baseline in our study. If this is true, the prevalence of risk allele should be higher in those who had a history of CHD at baseline i.e. prevalent cases of CHD. To examine this issue, we studied the association of the SNPs with prevalent CHD cases but found no association (data not shown). Moreover, the frequencies of the alleles in our population were high and comparable to former studies making selection bias unlikely.

Previous studies mainly employed standard case-control association studies. Our study has the advantage of employing a different approach, the prospective study in a large population based sample. One potential limitation of our study is that the participants were not fully followed and healthy subjects who carry the risk allele may later develop the disease.

In conclusion, we showed that the studied SNPs are not major players in development of CHD in the elderly population. Our negative finding offer a new perspective on 9p21 SNPs and shows that the association does not hold for all CHD cases. The lack of association may be due to the difference in genes involved in the development of CHD in young and older people. Individualized preventive measures and therapies constitute a major long-term goal of GWA studies. Heterogeneity of the effect, therefore, has substantial public health impact and needs to be acknowledged.