Genetic Prioritization of Plasma Proteins Across Dementia Subtypes: A Proteome-Wide Mendelian Randomization and Bayesian Colocalization Analysis

Dementia is clinically and biologically heterogeneous, and genetically supported plasma protein associations across dementia subtypes remain incompletely characterized. We aimed to genetically prioritize plasma proteins associated with overall dementia and major dementia subtypes using proteome-wide Mendelian randomization (MR) and Bayesian colocalization. We used the UK Biobank Pharma Proteomics Project (UKB-PPP) as a harmonized source of cis-protein quantitative trait loci (cis-pQTLs) for 2,941 Olink-measured plasma analytes. The European-ancestry discovery cis-pQTL analysis included up to 34,557 participants, whereas 54,219 participants were profiled in the overall UKB-PPP. Outcomes included FinnGen R10 genome-wide association studies (GWAS) for dementia, Alzheimer’s dementia, vascular dementia, frontotemporal dementia, and unspecified dementia, plus a European GWAS for dementia with Lewy bodies. At the nominal threshold of P < 0.05, 134, 132, 102, 78, 101, and 112 exploratory protein–outcome signals were observed for dementia, Alzheimer’s dementia, vascular dementia, frontotemporal dementia, dementia with Lewy bodies, and unspecified dementia, respectively; 23 associations remained significant after outcome-specific Benjamini–Hochberg false discovery rate (FDR) correction. Key FDR-significant proteins included APOE, NECTIN2, PVR, SERPINF2, GRN, TREM2, and ATXN2L. The strongest risk and protective associations were NECTIN2 with Alzheimer’s dementia (odds ratio [OR] = 2.650, 95% confidence interval [CI]: 1.875–3.744, FDR-adjusted P = 1.58 × 10 − 5 ) and ATXN2L with dementia (OR = 0.408, 95% CI: 0.255–0.654, FDR-adjusted P = 0.0370), respectively. Strong colocalization support was observed for selected SERPINF2 and TREM2 associations. These findings genetically prioritize several plasma proteins as candidate biomarkers and potential therapeutic targets across dementia subtypes, but do not establish that circulating protein abundance directly mediates dementia risk.