Integration of cerebrospinal fluid pQTL and GWAS prioritizes novel causal proteins for Alzheimer’s disease

Abstract

AbstractBackgroundQuantitative trait loci (QTL) are successful at identifying the functional gene for many loci identified through GWAS. While expression QTL (eQTL) studies robustly examine tissue‐specificity, most protein QTL (pQTL) studies have been limited to plasma. We have previously demonstrated that pQTL associations are highly tissue‐specific. Given the relevance of cerebrospinal fluid (CSF) to Alzheimer’s disease (AD) and other forms of neurodegeneration, we developed a CSF‐specific pQTL atlas and integrated it with AD GWAS to identify novel causal proteins for AD.MethodWe generated CSF proteomics (Somalogic, 7,584 analytes) data for 3,107 individuals. pQTL mapping was performed using a three‐stage approach: discovery, replication, and meta‐analysis. We analyzed the tissue‐ and molecule‐specificity of our associations by comparing to plasma & brain pQTLs and various eQTLs. We performed a proteome‐wide association study (PWAS) and Mendelian randomization (MR) to prioritize proteins affecting AD. We determined shared protein‐disease genetic regulation using colocalization. Using AD‐associated proteins, we built risk models of disease status.ResultWe identified 2,316 significant pQTL (1,247 in cis and 1,069 in trans) for 1,960 proteins, of which 1,720 were novel and not reported in the largest currently available pQTL studies in plasma or brain. Through PWAS, we identified 440 proteins associated with AD risk, which show enrichment in neurons (with APOE region) or microglia/macrophages (without APOE). MR prioritized 37 proteins as causal and colocalization identified 153 proteins that share genetic etiology with AD. Seventeen of these proteins overlap between all three methods and multiple (including PILRA, PRSS8, and SIRPA) represent novel causal proteins for AD. A protein risk score using PWAS‐identified proteins outperformed a polygenic risk score (PRS) at predicting amyloid/tau positivity across ages (AUC: 0.852‐0.906 vs. 0.724‐0.806, P<0.002) and across APOE genotypes (AUC: 0.809‐0.893 vs. 0.70‐0.80, P<0.027).ConclusionWe reported the largest CSF pQTL analysis to date and confirmed that CSF pQTLs are largely tissue‐specific. We identified proteins involved in AD that confirm reported candidate genes and prioritize new ones at GWAS loci. We developed accurate prediction models using prioritized proteins. Our findings offer insights into AD biology undetected with plasma pQTL analyses, supporting proteogenomic databases in neurologically‐relevant tissues.