Inferring the Molecular Mechanisms of Noncoding AD Associated Genetic Variants
Date:
Alexandre Amlie Wolf, Mitchell Tang, Beth A. Dombroski, PhD, Jessica Way, Nikolaos Vrettos, PhD, Yi Fan Chou, MS, Elizabeth E. Mlynarski, PhD, Christopher D. Brown, PhD, Gerard Schellenberg, PhD and Li San Wang, PhD
Background: Genome wide association studies (GWAS) have identified dozens of genetic variants associated with Alzheimer’s disease (AD). Most associations reside in the noncoding genome, where they affect tissue specific enhancers. Furthermore, variants identified by GWAS may be proxies of truly causal variants in linkage disequilibrium (LD). Methods: We applied our INFERNO algorithm to identify causal noncoding variants and their regulatory effects. Significant variants from the top 19 loci from IGAP Stage I were expanded by LD. These variants were tested for overlap with enhancer annotations across 239 tissues and cell types from FANTOM5 and Roadmap and HOMER transcription factor binding sites (TFBSs) for 332 TFs. A Bayesian statistical model was used to identify GWAS signals sharing co localized causal variants (posterior probability >= 0.5) with GTEx expression quantitative trait loci (eQTL) across 44 tissues. GTEx RNA seq data were further used to predict targets of long noncoding RNAs (lncRNAs). Tissue types from individual functional genomic data sources were harmonized into 32 categories for unbiased identification of significantly enriched tissue contexts and variants with concordant tissue specific functional support. Results: INFERNO identified 1,044 LD expanded variants across all 19 loci, 852 of which affected enhancers and TFBSs, with enrichment of enhancer overlaps in the immune related blood (p=0.0126) and the fibroblast containing connective tissue categories (p=0.0038). INFERNO identified 154 co localized GWAS eQTL signals for 67 genes across 15 IGAP regions, including 59 signals overlapping concordant enhancers in matching tissue categories for 35 genes in 10 regions (ABCA7, BIN1, CASS4, CD2AP, CD33, CELF1, CLU, EPHA1, INPP5D, ZCWPW1). Among these were 4 lncRNAs with 332 predicted targets enriched for dozens of immune response related pathways. Other regulatory signals included effects on DNA damage repair and disruption of binding sites for 40 microRNAs. Allele specific enhancer activity was validated by luciferase assays in K562 cells for strongly supported variants at the EPHA1, CD33, and BIN1 loci. Conclusions: INFERNO was used to integrate GWAS signals with functional genomics data to infer the perturbed genes, tissue contexts, and regulatory mechanisms affected by noncoding AD associated variants. This provided novel post GWAS insights into the role of immunity in genetic susceptibility to AD and identified potential therapeutic targets for future investigation.