Polygenic Score to Understand Cancer Etiology and Predict Cancer Risks

Abstract

Genetics have been an important risk factor for cancer. The information we learned from genome-wide association studies (GWAS) provide researchers with tools and new approach to better understand cancer epidemiology. In this dissertation, I present three projects using GWAS discoveries to understand cancer etiology and infer cancer risks. Chapter 1 uses GWAS information as an instrument variable to estimate the causal relationship between adiposity measures at different life stages (at birth, during childhood, at adulthood) and risk of breast, ovarian, prostate, colorectal and lung cancers via Mendelian Randomization analysis. We found that the genetic predicted adult BMI was inversely associated with breast cancer risk but positively associated with ovarian, lung and colorectal cancer risk. Chapter 2 evaluates the performance of a synthetic breast cancer risk prediction model utilizing both classical risk factors of breast cancer and common genetic variants in form of polygenic risk score (PRS). We validated the model using Nurses Health Study and Nurses Health Study II. We found that adding PRS greatly improved the performance of risk prediction models and of all three models validated, the model with both classic risk factor and PRS performed the best. Chapter 3 investigates the joint effect of PRS and pathogenic mutation in nine breast cancer predisposition genes using population based cohort studies in CAnceR RIsk Estimates Related to Susceptibility (CARRIERS) consortium. We also estimated 5-year and lifetime absolute risk using the final model built from penalized regression. We found that PRS is associated with breast cancer in carriers of pathogenic variant as well as in non-carriers but there was no significant difference between these effect (odds ratio associated with one standard deviation change in PRS). More importantly, we found that PRS can be particularly important for managing risk of carriers of pathogenic variants in moderate penetrance cancer predisposition genes such as ATM and CHEK2. Together, the projects presented in this dissertation demonstrated three approaches to utilize genetic information to understand cancer in the post-GWAS era. We hope that these findings could shed light to the underlying genetic architecture of cancer and could contribute to future studies of building breast cancer risk prediction models and generating effective screening guidelines.