The current, clinically applicable understanding of the pathogenesis of CHD is generally limited to RF exposures and basic genetics in the form of the family history. Yet this base of knowledge alone does not begin to explain the complex etiology of CHD. Recent GWAS have identified more than 30 genetic loci associated with CHD [10,11] and other loci (both overlapping and novel) associated with CAC quantity . However, these identified variants have small effect sizes, which have raised the question of what accounts for the missing heritability of CHD. There are multiple theories to account for the missing heritability of CHD. These include, but are not limited to, unscreened control data , biased sample collection , non-additive genetic effects, and that unidentified variants, rare and common, account for a majority of the risk for CHD and other common diseases . These theories, though, do not account for the possibility of gene-by-environment interactions and how they could contribute to missing heritability for traits such as CAC quantity. Accordingly, we have demonstrated a novel, replicated finding for a SNP in the 9p21.3 region: that a specific common variant on chromosome 9 provides a context for gene-by-BP interactions, thus defining the range of influence that blood pressure can have in atherogenesis and plaque deposition.
Specifically, we have extended accumulated empirical evidence regarding the 9p21.3 region and CHD and present significant and replicated evidence that rs2069416 interacts with DBP to exacerbate the extent of CAC in a genotype dependent manner. While in the CAC GWAS, that ignored any interactions, GENOA and FHS had discordant results for the 9p21.3 region, once the interaction with DBP was considered, inferences for a specific SNP became concordant. Moreover, no SNPs considered were significant for primary association with CAC in both cohorts; it was only in the context of a SNP-by-DBP interaction that the association with CAC was identified and replicated.
The lack of evidence for significant SNP-by-SBP interactions in the GENOA discovery cohort may reflect the high (67.6%) use of anti-hypertensive medications that specifically lower SBP, while largely leaving DBP levels unaffected. Moreover, our findings of CAC quantity being higher with lower DBP likely reflects pulse pressure due to the inclusion of both SBP and DBP (and the control for one measure of BP when the other is being tested for interaction). Pulse pressure has been demonstrated to increase with age due to increases in SBP and decreases in DBP . While tests for interactions between SNPs in 9p21.3 and pulse pressure were not significant in GENOA (data not shown), there remains the possibility that pulse pressure is contributing to the observations in this study. There is a complex relationship between pulse pressure and quantity of CAC: at younger ages (<50 years of age), SBP and DBP have been shown to be positively associated with quantity of CAC and likely act as surrogates of arterial resistance; however, quantity of CAC is more closely associated with pulse pressure in subjects older than 50 years, which likely reflects large-artery stiffness . Similar age-related effects of pulse pressure have also been reported in the FHS, with an increase in pulse pressure after 50–60 years of age  and pulse pressure becoming the strongest predictor of CHD risk after 59 years of age in FHS . In GENOA, as pulse pressure increased, CAC quantity also increased, regardless of genotype at rs2069416 (Additional file 1: Figure S2). However, both cohorts included participants younger than age 50 and older than age 50. Moreover, there was not sufficient power to consider a SNP-by-pulse pressure interaction in age specific strata or to consider three-way interactions that included age, SNP and pulse pressure.
A recent large GWAS for blood pressure identified 28 loci that, to the best of our knowledge, do not overlap with GWAS loci identified for CAC quantity or CHD . Our novel finding of a replicated interaction between a 9p21.3 SNP and DBP suggests a new genetic pathway through which BP variation contributes to variation in CAC quantity and thereby the potential pathogenesis of CHD. Specifically, it suggests that the CDKN2B-AS1 region, which is critical for numerous atherosclerotic phenotypes [15,34,35], but not associated with blood pressure phenotypes , may confer genetic risk through differential atherosclerotic plaque development for a given level of blood pressure. This mechanism may also be shared with other genomic regions associated with CHD and other atherosclerotic phenotypes, but not with blood pressure.
Despite findings of a replicated SNP-by-DBP interaction in the 9p21.3 region, knowledge is still lacking about the specific mechanisms by which genetic variants in the 9p21.3 region contribute to CAC extent and subsequent CHD pathogenesis. The 9p21.3 region has previously been consistently associated with CHD and related phenotypes, such as familial, premature CAD , abdominal aortic aneurysms , vascular wall stiffness , increased platelet reactivity , subclinical carotid artery disease , and ankle-brachial index . The implicated SNP within the 9p21.3 region is near protein-coding genes and also overlaps with an antisense non-coding RNA (also known as ANRIL or DQ485453) . Our top SNP-by-DBP interaction occurred with a non-coding variant, rs2069416, which is not in LD with any coding or obvious regulatory variants. Additionally, Visel et al. found that deletion of the mouse-analogue of the 9p21.3 non-codingregion resulted in a severe decrease in cardiac CDKN2A/B expression , suggesting that the presence of risk alleles in the 9p21.3 region may affect development of CHD through alteration of vascular cell proliferation.
Strengths of the present study include data from large community-based studies, similarity in CAC measurements from the different CT scanners used in the two cohorts, and similarity in imputation strategies and statistical methods. In addition, we utilized strict imputation quality control (only using SNPs with an imputation RSQ ≥ 0.8), which likely decreased the potential for false positive results from poor imputation quality of genotypes. As well, the minor allele frequency of our significant and replicated SNP, rs2069416 (0.37 and 0.33 for GENOA and FHS, respectively), closely matches the reported frequency of 0.36 for a European population in the 1000 Genomes Project data (http://browser.1000genomes.org). Finally, multiple RFs were included in the association and interaction analyses in contrast to most other studies that have included just age and sex .
Limitations include the differences in cohort inclusion criteria. The cohorts were entirely of European descent, thus limiting generalizations that can be drawn from these findings. More work is needed to replicate that the 9p21.3 region is involved in gene-BP interactions in other racial and ethnic groups. In addition, this investigation examined a restricted number of SNPs that had first been identified in the CHARGE CAC GWAS . This limited the inferences, as there are likely other loci that contribute to gene-BP interactions in the extent of CAC. Finally, multiple testing is an issue in all genetic association studies and must be accounted for in the results. We have applied Bonferroni correction for the number of SNPs tested to the final meta-analysis results, even though this approach is likely too conservative since the seven SNPs considered here are not completely independent of one another. As well, there is limited statistical power to detect gene-by-environment interactions. Due to these considerations, we used P < 0.05 in both cohorts as a cut-off for declaring significance. One SNP, rs2069416, had significant interactions with DBP under these criteria.