| dc.description.abstract | Cardiovascular disease (CVD) is the leading cause of death worldwide. The vast possibilities of interaction between genetic and environmental factors that contribute to CVD can be simplified by identifying conditions that favor the emergence of specific risk factor networks. If the relationships among CVD risk factors that give rise to these networks are under genetic control, then such relationships can be considered heritable phenotypes in themselves, amenable to genetic analysis. We characterized correlational networks of cardiovascular risk factors in a large cohort of urban and rural men and women in Ghana, and investigated how they may be perturbed by factors such as sex and urban lifestyle. We also assessed the comparative relevance of individual risk factors to thrombosis within and across networks, using as a proxy their association with an intermediate phenotype of CVD, plasminogen activator inhibitor type-1 (PAI-1). We found that the relationships between risk factors and PAI-1 were far more sensitive to differences in sex and environment than were the relationships among the risk factors themselves. To lay the theoretical groundwork for our subsequent genetic analyses, we modeled multiple types of biological SNP-by-covariate interactions and derived the statistical parameters to which they should give rise. In doing so, we demonstrated that even the strongest gene-by-covariate interactions at the biological level could display weak statistical interactions using general linear models. Moreover, we quantified the expected strength of the interaction relative to the marginal effect, depending on the nature of biological interaction. We then developed the ordinal joint interaction model (OJIM), which can not only identify biological SNP-by-covariate interactions where they exist, but also pick up marginal effects and leverage the change in residual correlation induced by marginal effects. In our analyses of the Ghanaian study population, the OJIM had more power than univariate or bivariate analysis to detect lipid SNPs of known biological significance, indicating that context-dependent genetic effects are probably quite common, and that the OJIM can identify them where they exist. We also used the OJIM to interrogate exome-wide data of our Ghanaian study population, and identified genetic variants that may increase thrombotic risk by influencing the covariance between these risk factors and PAI-1. | |
