Statistical methods in genome-wide association studies

Abstract

The aim of genome-wide association studies (GWAS) is to identify sequence variants that influence human traits or diseases. Previous GWAS have mostly focused on finding variants that affect the mean of a trait or disease risk under an additive model. However, variants can contribute to traits in different ways, such as under a recessive mode of inheritance and by affecting the variance of quantitative traits. In this thesis we use different statistical models to detect variants associating with sensory traits and explore relationships between correlated phenotypes. Furthermore, we implement a variance model to detect sequence variants that affect the variance of quantitative traits and we explore the effect of variants on the variance of glucose levels. In Paper I we estimate the effect of 36 glucose variants on the between subject and within subject variance of glucose levels. We found that some variants that affect the mean also affect the variance. The trend was that variants that increased mean and between subject variance of fasting glucose increased type 2 diabetes (T2D) risk, while variants that increase the mean but reduce the variance do not. We found that the effect of variants on the between subject variance of glucose levels are as important for genetic risk prediction of T2D as the effect of variants on the mean. Furthermore, the variants that increased between subject variance created correlation between close relatives and will thus increase heritability estimates. In Paper II we conduct a GWAS on structural measures of the corneal endothelium that are used in clinic to evaluate the health of the cornea. We detected associations at 7 novel loci, one of which is an intergenic variant near ANAPC1 that strongly associates with decreased endothelial cell density and accounts for a quarter of the population variance of cell density. The variant near ANAPC1 does not affect risk of corneal diseases or glaucoma in our data, which shows that even though low endothelial cell density is associated with ocular diseases, low cell density does not in and of itself lead to the development of disease. In Paper III we conduct GWAS meta-analysis of age-related hearing impairment (ARHI) using both the additive and recessive models. Previous GWAS on ARHI have reported common variants with small to moderate effects, while in this study, 13 of the 21 novel variants have rare genotypes with large effects. Six of the novel variants associate with ARHI under the recessive model, some of which would not have been detected under the additive model. We constructed an ARHI genetic risk score (GRS) using common variants and show that individuals in the top GRS decile develop ARHI 10 years earlier than those in the bottom decile, and their risk of ARHI is comparable to carriers of rare highly penetrant ARHI variants while the rare ARHI variants predispose to more severe ARHI than the common variants. Our findings shed a new light on the genetics of glycemic traits, the corneal endothelium and ARHI and highlight the importance of applying different statistical models when analyzing the effects of variants on phenotypes.