Estrogen Receptor 1 Variants and Coronary Artery Disease: Title and subTitle BreakShedding Light Into a Murky Pool

The article by Shearman et al¹ in this issue of THE JOURNAL brings intriguing genetic data to bear on several issues of considerable current controversy—the relationship between coronary artery disease (CAD) and estrogen action, the general utility of CAD associations with variants in candidate genes, and the challenges of identifying causative gene variants for common disease. In this interesting hybrid of a case-control and prospective design, the investigators extracted DNA from blood samples of a subset of unrelated members of the Framingham offspring cohort who were alive at examination 6 (approximately 27 years after the baseline examination). The subset was selected to achieve an approximately equal number of representative men and women from the cohort. The effects of several estrogen receptor α (ESR1) gene variants were tested using primarily a case-control analysis in which offspring who had developed atherosclerotic disease by the sixth examination were considered cases and the remainder controls. A 3.0-fold increase in risk of myocardial infarction (MI) (P<.001) was associated with the CC genotype of the ESR1 c.454-397T>C variant (compared with the TT and CT genotypes combined) after correction for multiple risk factors. Lower odds ratios were associated with broader end points. Virtually identical results were obtained when data were analyzed as a prospective survival study.

The findings by Shearman et al¹ are driven entirely by the results of the men in the cohort, as women contributed few cases. Indeed, the relatively small number of cases, particularly among women, in whom estrogen action would be expected to be more important, is arguably the main limitation of the study (54 men and 5 women with MI). Curiously, MI was the only end point associated with the CC genotype; non-MI end points, if anything, showed an opposite trend. This raises the question of whether the CC genotype marks risk for predisposition to MI rather than promotion of underlying atherosclerosis. Nevertheless, the substantial odds ratio and small P value for MI in this carefully identified, well-studied cohort strongly support a true association for MI, at least in men. Furthermore, the fact that this study confirms previously reported associations between CAD and the same ESR1 variant in a Finnish autopsy study² and among Japanese patients with familial hypercholesterolemia (with stronger risks actually shown for the closely linked −1989T>G, c.454-351A>G, and promoter TA repeat sites in that cohort)³ further underscores the likelihood of a true contribution of the ESR1 c.454-397T>C or a tightly linked locus to CAD.

That the study by Shearman et al¹ confirms prior observations is of paramount importance in light of current thinking about genetic association studies, which are often considered untrustworthy because of the frequent inability to confirm initially reported positive associations.^{4 - 5} In a recent review of 301 separate studies of 25 genetic associations with common disease, 8 of the initial associations could be statistically confirmed by meta-analysis of subsequently reported studies.⁶ Furthermore, if 1 study confirmed the initial report with P<.001 or if 2 studies did so with P<.01, then future replication was strongly predicted. The previously identified tendency of initial genetic association studies to overestimate the true risk⁷ may be explained by an expected statistical phenomenon called "winner's curse"—the likelihood that the first investigator to identify an association has done so because the association is by chance stronger in that particular population or sample.⁶

In addition to large sample sizes, small P values, replication in independent populations, and associations that make biological sense, another distinguishing feature of high-quality association studies is "physiologically meaningful data supporting a functional role of the polymorphism in question."⁴ In this sense, a "functional" variant or mutation can result in a complete or partial loss of function or a gain in function. While more statistically sophisticated means such as linkage or transmission disequilibrium testing may seem more scientific, they usually are not as powerful or efficient as simple association testing in a case-control study (particularly if controls are selected to be "hypernormal").⁸ Both phenotypic and genotypic heterogeneity can make linkage studies of common diseases particularly inefficient. A more intuitive argument can be made for the utility of association testing of functional variants in reasonable candidate genes.

Although linkage studies might reveal the region of a previously unsuspected candidate gene, the likelihood of a misleading or spurious association remains high until the challenging task of locating and characterizing a new functional variant has been accomplished. Testing a known variant with demonstrated functional effects thus sidesteps the laborious gene identification (or positional cloning) phase. Furthermore, association of a genetic variant with a disease eventually must be confirmed using relatively straightforward epidemiologic methods, regardless of how the gene is identified. Association testing of a known functional variant is thus inherently attractive. Unfortunately, the effects of a functional variant can be confounded by the effects of other nearby variants that may be coinherited as a haplotype. Must all such variants and their effects be identified before association is claimed? For variants with very strong effects (eg, virtually any of the >700 mutations of the low-density lipoprotein receptor causing familial hypercholesterolemia⁹ ), the effects of other genes in the haplotype might be expected to be small and of little consequence. Unfortunately, most gene variants associated with common diseases have much milder effects and are more likely to be affected by other nearby and coinherited variants. Variable haplotype frequencies and composition in different populations further complicate the issue with ESR1.¹⁰ The study by Shearman et al¹ does not consider haplotypic information, potentially limiting the interpretation of the data.

In the case of ESR1, many levels of genetic complexity are evident. A host of ESR1 variants have been identified, many in strong or modest linkage disequilibrium.³ Limited haplotyping has been applied with considerable success to the association of ESR1 with bone mineral density.^{10 - 12} In these studies, the c.454-397C allele (which appears most often with the c.454-351G variant), the same allele studied by Sherman et al,¹ was more often associated with higher bone density and lower fracture risk, suggesting increased estrogen activity. This association appears to be consistent with apparent increased estrogen action in CC homozygotes in other settings, including a greater increase in high-density lipoprotein cholesterol¹³ and a greater decrease in E-selectin¹⁴ in response to oral estrogen therapy. Also consistent with increased estrogen action was an increased likelihood of premenopausal hysterectomy due to several uterine disorders, including fibroids and menometrorrhagia associated with the CC genotype.¹⁵ In in vitro studies using an artificial promoter construct, the C allele showed enhanced reporter gene transcription due to formation of a B-myb binding site not present with the T allele.¹⁴ B-myb may itself be up-regulated by estrogen, suggesting a potential direct or indirect increase in estrogen action by the C allele.¹⁴ Paradoxically, this same c.454-397C allele was associated with increased MI risk in the report by Shearman et al.¹ Also consistent with a possibly reduced response to estrogen, other studies reported that the C allele was associated with decreased breast cancer risk¹⁶ and decreased risk for endometrial cancer.¹⁷ Thus, a straightforward, causal molecular mechanism to explain the various genetic associations reported to date does not seem possible.

Perhaps these complexities of ESR1 genotype associations with estrogen-related phenotype can be explained by the complexities of its function as an estrogen receptor. For example, there are multiple alternative transcription or splice variants of the gene, including variants that exclude the estrogen binding site. In addition, recent studies indicate rapid, nongenomic effects of estrogen due to its binding to some form(s) of ESR1 at the cell surface.^{18 - 23} These novel cell surface receptor-mediated effects include stimulation of endothelial and inducible forms of nitric oxide synthase with increases in nitric oxide, vasodilation, endothelial rounding, and decreased immunocyte adherence.^{24 - 25} The many traditional nuclear receptor-mediated effects of estrogen include increases in high-density lipoprotein cholesterol and triglyceride levels, decreases in low-density lipoprotein cholesterol and lipoprotein(a) levels, smooth muscle proliferation, and a decrease in antithrombin III.²⁶ Thus, there appears to be great physiologic complexity in ESR1 effects beyond which are many potential contributions from estrogen receptor β (ESR2).²⁰

Given these complex and contradictory data regarding the effects of one of the estrogen receptors on atherosclerosis-related factors, perhaps it is understandable and even expected that the effects of exogenous estrogen could be equally complex and contradictory. For example, estrogen therapy in rodents that have excess lipoprotein remnant particles (apolipoprotein E knockout mice) appears to reduce fatty streak formation but not progression or rupture of advanced plaques.²⁷ Similar results were seen in monkeys.²⁸ In postmenopausal women, oral estrogen increases C-reactive protein (CRP) levels²⁹ and matrix metalloproteinases.³⁰ However, the estrogen-induced increase in CRP may be of little or no consequence, since CRP failed to predict cardiovascular risk in postmenopausal women taking oral estrogen.³¹ Furthermore, while estrogen has been thought to have a proinflammatory effect with regard to CRP, estrogen may also be considered anti-inflammatory through its demonstrated decrease of a number of other markers of inflammation, including plasma-soluble E-selectin, intercellular adhesion molecule 1, vascular cell adhesion molecule 1, and monocyte chemoattractant protein 1.^{32 - 35} Raloxifene, an estrogen variant with both antiestrogenic and proestrogenic actions, has prothrombotic effects similar to estrogen yet may protect against CAD events in higher-risk women.³⁶

Of greater clinical importance is the striking paradox that observational studies in humans (and interventional studies in nonhuman primates) have suggested antiatherogenic effects when oral estrogen therapy is started in the perimenopausal or early menopausal period but proatherogenic effects when started later in the period of estrogen deficiency.³⁷ Even as ongoing and future clinical trials grapple with these issues, population genetic studies of estrogen receptor variants are beginning to shed valuable, if not always definitive, light on the multifaceted relationship between estrogen and atherosclerosis.