Genetically regulated expression in late-onset Alzheimer's disease implicates risk genes within known and novel loci

Abstract

Late-onset Alzheimer disease (LOAD) is highly polygenic, with a heritability estimated between 40 and 80%, yet risk variants identified in genome-wide studies explain only ~8% of phenotypic variance. Due to its increased power and interpretability, genetically regulated expression (GReX) analysis is an emerging approach to investigate the genetic mechanisms of complex diseases. Here, we conducted GReX analysis within and across 51 tissues on 39 LOAD GWAS data sets comprising 58,713 cases and controls from the Alzheimer's Disease Genetics Consortium (ADGC) and the International Genomics of Alzheimer's Project (IGAP). Meta-analysis across studies identified 216 unique significant genes, including 72 with no previously reported LOAD GWAS associations. Cross-brain-tissue and cross-GTEx models revealed eight additional genes significantly associated with LOAD. Conditional analysis of previously reported loci using established LOAD-risk variants identified eight genes reaching genome-wide significance independent of known signals. Moreover, the proportion of SNP-based heritability is highly enriched in genes identified by GReX analysis. In summary, GReX-based meta-analysis in LOAD identifies 216 genes (including 72 novel genes), illuminating the role of gene regulatory models in LOAD.

Fig. 1. Miami plot of tissue-specific GReX association tests with LOAD from meta-analysis of AGDC and IGAP (top) or ADGC-only (bottom).

The green region labels the ±10 Mb of known AD loci, the blue dot indicates the novel gene, which locates outside the known AD region (green) and is never reported in previous GWAS, and the brown dot in ADGC-only meta-analysis suggests the gene holds significance in conditional analysis. P-values less than 1 × 10⁻¹⁵ were truncated (pink dot) to improve the clarity of peaks in other, less significant regions.

Fig. 2. Miami plot of cross-all GTEx tissue (blue dots, top) and cross-brain tissues model (gray dots, bottom).

p-values less than 1 × 10⁻¹⁵ (pink dots) were truncated to improve the clarity of peaks in other, less significant regions.

Fig. 3. Scatter plot of log-transformed p-values from SNP-based (PrediXcan) and summary statistic (S-PrediXcan) GReX approach.

Each dot shows the gene-level p-value from a SNP-based approach (PrediXcan, y-axis) and summary statistic approach (S-PrediXcan, x-axis). Color indicates that the signal (tissue-specific GReX) was only identified in PrediXcan (red, FDR < 0.05) or only in S-PrediXcan (green, FDR < 0.05).

Fig. 4. Scatter plot of sample size used for model building against the true positive rate.

X-axis represents the number of subjects used in PrediXcan model building and Y-axis represents the true positive rate for each model in our analysis. Color represents tissue source (red for brain tissues and gray for other tissues).