Proteome-wide association studies using summary pQTL data of brain, CSF, and plasma identify 30 risk genes of Alzheimer's disease dementia

Abstract

Background: A proteome-wide association study (PWAS) that integrates proteomic data with genome-wide association study (GWAS) summary data is a powerful tool for studying Alzheimer's disease (AD) dementia. Existing PWAS analyses of AD often rely on the availability of individual-level proteomic and genetic data of a reference panel. Leveraging summary protein quantitative trait loci (pQTL) reference data of multiple AD-relevant tissues is expected to improve PWAS findings of AD dementia.

Methods: We conducted PWAS by integrating publicly available summary pQTL data of three tissues including brain, cerebrospinal fluid (CSF), and plasma, with the latest GWAS summary data of AD dementia. For each target protein per tissue, we employed our recently published OTTERS tool to obtain omnibus PWAS p-value, testing whether the genetically regulated protein abundance in the corresponding tissue is associated with AD dementia. Protein-protein interactions and enriched pathways of identified significant PWAS risk genes were analyzed by STRING. The potential causal effects of these PWAS risk genes were assessed by probabilistic Mendelian Randomization analyses.

Results: We identified 30 unique significant PWAS risk genes for AD dementia, including 11 for brain, 10 for CSF, and 16 for plasma tissues. Five of these were shared by at least two tissues, and gene MAPK3 was found in all three tissues. We found that 11 of these PWAS risk genes were associated with AD dementia or AD pathology traits in GWAS Catalog; 18 of these were detected by transcriptome-wide association studies (TWAS) in dorsolateral prefrontal cortex brain tissue; and 25 of these, including 8 out of 9 novel genes, were interconnected within a protein-protein interaction network involving the well-known AD risk gene APOE. These PWAS risk genes were enriched in immune response, glial cell proliferation, and high-density lipoprotein particle clearance pathways. Mediated causal effects were validated for 13 PWAS risk genes (43.3%).

Conclusions: Our findings provide novel insights into the genetic mechanisms of AD dementia in brain, CSF, and plasma, and provide targets for developing new therapies. This study also demonstrates the effectiveness of integrating summary pQTL and GWAS data for mapping risk genes of complex human diseases.

Keywords: AD dementia; Alzheimer’s disease; GWAS; OTTERS; PWAS; TWAS; pQTL.

Fig. 1

Manhattan plots of PWAS results (FDR q-values) of AD dementia by OTTERS in Brain (A), CSF (B), and Plasma (C) tissues The -log10(q-values) were plotted on the y-axis, and -log10(0.05) was plotted as the dashed horizontal line. Independent significant genes are labeled. Yellow triangles: genes identified by previous GWAS and TWAS; Blue triangles: genes identified by previous GWAS; Purple square: genes identified by previous TWAS; Red circle: novel findings

Fig. 2

Scatter plots of pQTL weights of example PWAS risk genes of APOE in brain, BSG in CSF, and PROC in plasma that were estimated by individual PRS models The pQTL weights were plotted in the y-axis for all genetic variants in the test gene region and color-coded with respect to -log10 (GWAS p-value). Test SNPs with GWAS p-value <${10}^{-5}$ were colored

Fig. 3

PPI network and enrichment analyses results with PWAS risk genes of AD dementia by STRING Edges represent physical PPI, with different colors representing different sources of connection evidence. PWAS risk genes not connected in the network are not included in the network plot. Node colors represent different enriched GO terms with FDR < 0.05, 6 GO terms with most significant FDR q-value were colored

Fig. 4

Manhattan plot of TWAS-O results of AD dementia in DLPFC tissue The -log10(q-value) were plotted on the y-axis, and -log10(0.05) was plotted as the dashed horizontal line. Plotted TWAS-O results were obtained by the ACAT method, combining TWAS p-values obtained by TIGAR/DPR, PrediXcan/Elastic_Net, and FUSION/BestModel. Independent significant genes are labeled