Premium
Defining Alzheimer’s disease subtypes using polygenic risk scores integrated with genomic and brain transcriptomic profiles
Author(s) -
Hu Junming,
Chung Jaeyoon,
Panitch Rebecca,
Zhu Congcong,
Beecham Gary W.,
Mez Jesse,
Farrer Lindsay A.,
Stein Thor D.,
Crane Paul K.,
Jun Gyungah R.
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.046449
Subject(s) - disease , transcriptome , neuroimaging , medicine , genome wide association study , alzheimer's disease , gene , oncology , biology , gene expression , genetics , single nucleotide polymorphism , psychiatry , genotype
Background In light of the complex etiology of Alzheimer’s disease (AD), classifying AD patients into clinical subtypes is important for precision medicine. We tested whether polygenic risk scores (PRSs) combined with brain expression profiles in AD patients are correlated with clinically defined AD subtypes. Methods We generated 143 co‐expressed gene networks (modules) using RNA sequencing (RNA‐Seq) data derived from 64 autopsied AD brains from the Framingham Heart Study and Boston University Alzheimer’s Disease Center. These modules were validated using brain RNA‐seq data from participants with AD in the Religious Orders Study and Memory and Aging Project (n = 363) and the Mayo Clinic Study of Aging (n = 82). Fourteen AD‐related modules were selected by enrichment analysis using the significant genes (P < 0.001) identified in a genome‐wide association study for neuropathological traits ( Beecham, 2015 ) and then characterized by brain cell‐type specific expression profiles. We computed polygenic risk scores of the AD‐associated genes within each module (module‐based PRSs) using the estimates from Beecham et al . in the 449 AD patients of the Alzheimer’s Disease Neuroimaging Initiative. Next, we examined correlations of the 14 module‐based PRSs with (1) scores for 9 different cognitive tests and six cognitively‐defined AD subtypes including memory, language, visuospatial, executive, multi, or no domains ( Mukherjee, 2018 ). Results Eleven of the 14 AD‐related modules were significantly enriched (P < 0.05) for specific brain cell types including four in neurons (best P = 6.8 × 10 −104 ), four in astrocytes (best P = 5.2 × 10 −76 ), two in endothelial cells (best P = 9.9 × 10 −87 ), and one in microglia (P = 1.7 × 10 −129 ). The most significant correlation between the PRSs from the 14 modules and clinical phenotypes was found with an astrocyte‐specific module and language‐related cognitive functions (correlation r 2 >0.2; P = 6.7 × 10 −3 ), and this module contained 8 AD genes (best gene: DOCK1 , P = 3.4 × 10 −5 ). The individual PRSs of the astrocyte‐module were significantly higher in the language‐specific AD patients compared with in other five subtypes (best P = 0.01 from comparison with AD patients in the no domain group). Conclusion These findings illustrate for the first time that genetic risk scores integrated with system biology may define clinical subtypes of AD. Our findings will facilitate genome‐guided precision medicine efforts in prevention and treatment of AD.