Multiple phenotype modeling in pleiotropic effect studies of quantitative trait loci
MetadataShow full item record
Pleiotropy refers to the shared effects of a gene or genes on multiple phenotypes, a major reason for genetic correlation between phenotypes. For example, for osteoporosis, bone mineral densities at different skeletal sites may share common genetic factors; thus, examining the shared effects of genes may enable more effective fracture treatments. To date, methods are not available for estimating and testing the pleiotropic effects of single nucleotide polymorphisms (SNPs) in genetic association studies. In this dissertation, we explore two types of methods to evaluate the SNP-specific pleiotropic effect based on multivariate techniques. First, we propose two approaches based on variance components (VC) analysis for family-based studies, which quantify and test the pleiotropic effect by examining the contribution of specific genetic marker(s) to polygenic correlation or covariance of traits. Second, we propose a multivariate linear regression approach for population-based studies with samples of families or unrelated subjects. This method partitions the specific effect of the marker(s) from phenotypic covariance. We evaluate the performance of our proposed methods in simulation studies, compare them to existing multivariate analysis methods and illustrate their application using real data to assess candidate SNPs for osteoporosis-related phenotypes in the Framingham Osteoporosis Study. In contrast to existing methods, our newly proposed approaches allow the quantification of pleiotropic effects. The bootstrap resampling percentile method is used to construct confidence intervals for statistical hypothesis testing. Simulation results suggest that the VC-based approaches are affected by the polygenic correlation level. The covariance analysis approach outperforms the VC-based approaches, with unbiased estimates and better power, which remain consistent regardless of the polygenic correlation. In addition, the covariance analysis approach is simple to implement and can be applied to both family data and genetically unrelated data. Using simulation, we also show that existing methods, such as MANOVA, can have high rejection rates when a SNP has a large effect on a single trait, which prevent us from using them for pleiotropic effect analysis. In summary, this dissertation introduces promising new approaches in multiple phenotypic models for SNP-specific pleiotropic effect.