Abstract
Growing evidence suggests overlap between Alzheimer's disease (AD) and Parkinson's disease (PD) pathophysiology in a subset of patients. Indeed, 50-80% of autopsy cases with a primary clinicopathological diagnosis of Lewy body disease (LBD)-most commonly manifesting during life as PD-have concomitant amyloid-beta and tau pathology, the defining pathologies of AD. Here we evaluated common genetic variants in genome-wide association with AD as predictors of concomitant AD pathology in the brains of people with a primary clinicopathological diagnosis of PD or Dementia with Lewy Bodies (DLB), diseases both characterized by neuronal Lewy bodies. In the first stage of our study, 127 consecutive autopsy-confirmed cases of PD or DLB from a single center were assessed for AD neuropathological change (ADNC), and these same cases were genotyped at 20 single nucleotide polymorphisms (SNPs) found by genome-wide association study to associate with risk for AD. In these 127 training set individuals, we developed a logistic regression model predicting the presence of ADNC, using backward stepwise regression for model selection and tenfold cross-validation to estimate performance. The best-fit model generated a risk score for ADNC (ADNC-RS) based on age at disease onset and genotype at three SNPs (APOE, BIN1, and SORL1 loci), with an area under the receiver operating curve (AUC) of 0.751 in our training set. In the replication stage of our study, we assessed model performance in a separate test set of the next 81 individuals genotyped in our center. In the test set, the AUC was 0.781, and individuals with ADNC-RS in the top quintile had four-fold increased likelihood of having AD pathology at autopsy compared with those in each of the lowest two quintiles. Finally, in the validation stage of our study, we applied our ADNC-RS model to 70 LBD individuals from 20 Alzheimer's Disease Research Centers (ADRC) whose autopsy and genetic data were available in the National Alzheimer's Coordinating Center (NACC) database. In this validation set, the AUC was 0.754. Thus, in patients with autopsy-confirmed PD or DLB, a simple model incorporating three AD-risk SNPs and age at disease onset substantially enriches for concomitant AD pathology at autopsy, with implications for identifying LBD patients in which targeting amyloid-beta or tau is a therapeutic strategy.