Predicting Conversion to Alzheimer Disease using Module‐Based Polygenic Risk Scores Informed by Brain Proteome Profiles

Abstract

AbstractBackgroundNeuropathologically confirmed Alzheimer disease (AD) patients are heterogeneous in their cognitive functions, brain atrophy patterns, and neuropathological traits. Genetic risk profiles informed by brain proteome signatures may predict the conversion of clinically normal individuals to AD.MethodWe discovered co‐expressed proteome networks using 201 AD brains from the prefrontal cortex area in the Religious Orders Study and Memory and Aging Project (ROSMAP) and validated the findings using 328 AD brains in the Emory study. Of the preserved modules, we selected modules with significant association (P<0.01) between neuropathologically and clinically confirmed AD (symptomatic) and neuropathologically confirmed AD but clinically unimpaired brains. We generated polygenic risk scores of the symptom‐associated modules (mbPRSs) using existing genome‐wide summary statistics for AD (Kunkle, 2019) in 309 AD, 483 mild cognitive impairment (MCI), and 301 cognitively normal (CN) subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). We investigated relationships between the derived mbPRSs and previously clustered brain atrophy patterns in the ADNI (Yang, 2021). We defined at‐risk CN and MCI subjects in the ADNI for each module if mbPRSs of the subject > the third quantile. Age‐adjusted Cox proportional hazard models were applied for conversion status to AD with at‐risk subjects of the symptom‐associated modules comparing with all CN and MCI subjects as a reference group.ResultWe detected 12 preserved modules that showed significant associations for symptomatic AD (P‐value<0.05). The mbPRSs of the yellow and the lightcyan modules were significantly associated with brain atrophy cluster 3 (P‐value: yellow‐mbPRS = 5.5e‐03; lightcyan‐mbPRS = 7.1e‐05). At the same time, Of the mbPRSs for the 12 symptom‐associated modules, the at‐risk subjects defined by the lightcyan‐mbPRS were also predicted to be earlier conversion from CN or MCI to AD (Hazard ratio = 1.63, P‐value = 3.8e‐02). Lightcyan‐mbPRS shows a good prediction of conversion to AD and also the association with MRI endophenotypes.ConclusionThis study demonstrates for the first time that PRSs informed by symptom‐associated protein networks may identify at‐risk CN and MCI subjects who will develop clinical symptoms and enable targeted prevention and treatment trials for AD for these at‐risk subjects in preclinical stages of AD.