Leveraging Novel Multivariate Methods of Subphenotypes in Genetic Association Studies of Sjogren’s Syndrome

Harvard School Of Public Health, Boston MA

Investigators

Linked publications & trials

Abstract

Abstract Genetic epidemiology has entered the big data era with many cohort studies having access to not only genome-wide genotyping data but also large number of disease-related traits and a variety of biomarkers. These extensive datasets hold great promise for increasing our understanding of human diseases and improving public health. However, statistical tools to leverage these data are severely lacking and the development of innovative methodological approaches remains a key component for future successes. Indeed, most genetic association studies still utilize standard univariate approach, testing each measured phenotype independently for association with each single genetic variant. Our recent work has shown that phenotypes sharing genetic and environmental underpinnings can be leveraged in multi-phenotype analyses to increase statistical power to detect associated genetic loci. Diseases showing heterogeneity and/or evidence for subtypes that can be partially characterized by endophenotypes and biomarkers, including several autoimmune diseases such as the Sjögren's syndrome (SS), are particularly good candidates for multi- phenotype methods. In this proposal we aim to apply two new multi-phenotype methods for the analysis of over 50 SS related phenotypes from the Sjögren's International Collaborative Clinical Alliance (SICCA) study. SICCA has generated a unique collection of SS case/control and SS related phenotypes along genome-wide genotypes data among more than 3,500 individuals. The first proposed approach is an extension of the multivariate method based on a principal component analysis framework that we recently developed. Unlike standard univariate approaches, our method is capable of detecting associations even when there exist multiple genetically heterogeneous subphenotypes of the disease. It is based on composite null hypothesis (all phenotypes are tested jointly), so that single phenotype-genotype association cannot be established. This limitation is the cost for dramatic increase in statistical power to identify genetic variant with positive and negative pleiotropic effect (concordant and discordant genetic effect respectively). The second approach relies on a new and innovative strategy that will be developed as part of this proposal. As oppose to multivariate methods, this approach keeps the univariate properties of determining association between a single outcome and a single genetic variant, but as in multivariate approaches, it leverages correlation with other available phenotypes. Using the proposed approaches we expect to dramatically increase our ability to identify genetic variants associated with SS phenotypes. We fully expect that these two methods will reveal important insights into the genetic basis of SS and will go on to serve the broader the genetics community.