Identifying genes associated with Alzheimer’s disease using gene‐based polygenic risk score
Dongbing Lai, Michael Zhang, Rudong Li, Chi Zhang, Pengyue Zhang, Yunlong Liu, Sujuan Gao, Tatiana M. Foroud- Psychiatry and Mental health
- Cellular and Molecular Neuroscience
- Geriatrics and Gerontology
- Neurology (clinical)
- Developmental Neuroscience
- Health Policy
- Epidemiology
Abstract
Background
Recent large‐scale genome‐wide association studies (GWAS) have identified >100 Alzheimer’s disease (AD) associated genes but together they only explain a small portion of AD heritability1‐11. Dramatically increasing the sample size can detect genes with small effects but it is not feasible in the near future. If polygenic risk score (PRS) calculated using SNPs located within genes (gene‐based PRS) has higher predictability, then these genes are likely AD‐associated. Additionally, many drugs target AD‐associated genes and they should be prioritized for future investigations.
Method
We used SNPs having the same directions of effects (concordant SNPs) in different large‐scale European ancestry AD GWAS cohorts (Kunkle et al, 20194,9, the UK Biobank8,9, FinnGen consortium1,12) to calculate gene‐based PRS. Retaining concordant SNPs excludes cohort‐specific findings and reduces false positives, thereby improving PRS predictability. We used megaPRS13 to measure the PRS predictability, which can use GWAS summary statistics as the target dataset13. Leave‐one‐cohort‐out (LOCO) strategy was used and for each LOCO, two GWAS cohorts were used as the discovery dataset and the third cohort was used as the target dataset. Genes identified by all three LOCOs were considered as AD‐associated. Given the large effect of APOE on AD prediction, LOCO was also performed excluding APOE. We then searched the Drug Gene Interaction Database14 for drugs targeting AD‐associated genes. To prioritize identified drugs, we use lmQCM15 and the STRING database16 to identify AD‐associated genes in the same co‐expression module or in the same protein‐protein interaction (PPI) network, respectively, as drugs targeting them may also have effects on other genes via co‐regulation and/or interaction and thus may have large efficacies.
Result
Using this strategy, we identified 389 genes, nearly half of which were not previously reported as AD‐associated. These genes explained large portions of heritability (47%‐97%). Sixty‐four genes were in the same co‐expression modules and/or PPI networks and 688 drugs targeting them were prioritized for further evaluation.
Conclusion
Gene‐based PRS is a cost‐effective way to identify AD‐associated genes using currently available data. Co‐expression modules and PPI networks can potentially identify drugs having large efficacy which can be further investigated for their potentials to treat AD.