Omnibus proteome-wide association study identifies 43 risk genes for Alzheimer disease dementia

Abstract

Transcriptome-wide association study (TWAS) tools have been applied to conduct proteome-wide association studies (PWASs) by integrating proteomics data with genome-wide association study (GWAS) summary data. The genetic effects of PWAS-identified significant genes are potentially mediated through genetically regulated protein abundance, thus informing the underlying disease mechanisms better than GWAS loci. However, existing TWAS/PWAS tools are limited by considering only one statistical model. We propose an omnibus PWAS pipeline to account for multiple statistical models and demonstrate improved performance by simulation and application studies of Alzheimer disease (AD) dementia. We employ the Aggregated Cauchy Association Test to derive omnibus PWAS (PWAS-O) p values from PWAS p values obtained by three existing tools assuming complementary statistical models-TIGAR, PrediXcan, and FUSION. Our simulation studies demonstrated improved power, with well-calibrated type I error, for PWAS-O over all three individual tools. We applied PWAS-O to studying AD dementia with reference proteomic data profiled from dorsolateral prefrontal cortex of postmortem brains from individuals of European ancestry. We identified 43 risk genes, including 5 not identified by previous studies, which are interconnected through a protein-protein interaction network that includes the well-known AD risk genes TOMM40, APOC1, and APOC2. We also validated causal genetic effects mediated through the proteome for 27 (63%) PWAS-O risk genes, providing insights into the underlying biological mechanisms of AD dementia and highlighting promising targets for therapeutic development. PWAS-O can be easily applied to studying other complex diseases.

Keywords: Alzheimer disease; Alzheimer disease dementia; GWAS summary data; aggregated cauchy association test; pQTL; proteome-wide association study.

Figure 1

Two-stage PWAS framework Stage I: Protein abundance imputation models are trained using cis-SNPs as predictors and proteomic abundances as the outcomes, which estimate pQTL effect sizes (weights) per protein-coding gene using the reference proteomics data. Stage II: Individual PWAS predicts the genetically regulated protein abundances using these trained imputation models and individual-level genotype data of test samples, and then tests the association between the predicted genetically regulated protein abundances and the trait of interest. Summary-based PWAS takes the pQTL effect sizes as variant weights to conduct an equivalent gene-based association test with GWAS Z scores and reference LD.

Figure 2

PWAS-O framework Protein abundance imputation models are first trained using the reference training data by TIGAR/DPR, PrediXcan/EN, and FUSION tools, and two-stage PWASs (Figure 1) are conducted based on each set of models with GWAS data. Our pipeline uses the TIGAR tool to train protein abundance imputation models by both DPR and EN (the same method used by PrediXcan with equal proportions of LASSO and Ridge penalties) methods. Next, we obtain a PWAS-O p value per protein-coding gene by aggregating all PWAS p values by ACAT.

Figure 3

Test R² and PWAS power comparison in simulation studies Different scenarios with proportions of true causal cis-pQTL p_causal = (0.001,0.01) and protein abundance heritability $h_p^2$ = (0.01, 0.05) were considered in the simulation studies. Boxplots of test R²s of 50,000 simulations per scenario for 3 PWAS tools were presented (A). Simulations that fail to train valid protein imputation models would have R² test scores of zero. The median is depicted as a black bar, while the lower and upper hinges represent the 25th and 75th percentiles, respectively. Additionally, the whiskers extend from the hinges to a maximum of 1.5 times the interquartile range. Any data points beyond the whiskers were plotted individually. The power of 50,000 simulations were plotted with respect to various GWAS sample sizes, comparing PWAS-O to these three individual PWAS tools (B).

Figure 4

Manhattan plot of PWAS-O results for AD dementia A total of 43 significant risk genes were identified with FDR q value < 0.05. A total of 19 independently significant genes as listed in Table 2 were labeled and plotted in red. The remaining significant risk genes are plotted in orange. The −log10(q values) were plotted in y axis, and −log10(0.05) was plotted as the dashed horizontal line.

Figure 5

pQTL weights estimated by TIGAR/DPR and FUSION (BestModel:BLUP) for the PWAS-O significant genes PPM1N and DCAKD The pQTL weights were plotted in the y axis for all test SNPs in the test gene region, color coded with respect to their −log10 (GWAS p value). SNPs with GWAS p value < 10⁻⁵ were plotted in yellow. PPM1N (A) was found significant by both TIGAR and FUSION, as multiple test SNPs with non-zero pQTL weights (by TIGAR/DPR and FUSION/BestModel:BLUP) for significant GWAS signals (yellow dots). By contrast, the significance of DCAKD (B) was mainly driven by FUSION/BestModel:BLUP, for having fewer significant GWAS signals with a relatively large magnitude of pQTL weights. These two example genes were not tested by PrediXcan/EN for having CV R² < 0.5% by EN method. Thus, their pQTL weights by EN method were not plotted.

Figure 6

PPI networks and enriched pathways among 43 PWAS-O risk genes of AD dementia by STRING Edges represent physical PPI, with different colors representing different sources of connection evidence. Node colors represent different enriched GO terms or KEGG pathways with FDR < 0.05.