TL;DR
This paper introduces CMO, a novel gene-level association test that integrates DNA methylation and enhancer-promoter interactions with existing GWAS data. CMO significantly improves the power to identify genes associated with Alzheimer's disease (AD) and discovers six novel AD risk loci missed by previous methods.
What Problem Does This Paper Address?
Alzheimer's disease (AD) is a complex, progressive neurodegenerative disorder with no effective cure. Identifying genetic risk factors is crucial for early detection and intervention. While Transcriptome-Wide Association Studies (TWAS) have been successful in finding genetic loci by analyzing gene expression, they often overlook the crucial roles of epigenetic modifications like DNA methylation and regulatory elements like enhancer-promoter interactions. These factors are known to significantly contribute to AD pathogenesis. Existing methods either focus solely on gene expression or analyze DNA methylation individually, failing to capture the complex interplay of these regulatory mechanisms. This gap limits our ability to fully understand the genetic architecture of AD and identify all relevant risk genes.
Methods
The core innovation of this paper is the Cross-Methylome Omnibus (CMO) test. CMO is a gene-level association testing method that integrates three key data types:
- Genetically regulated DNA methylation (DNAm): Using prediction models derived from existing datasets, CMO estimates DNAm levels based on genetic variants.
- Enhancer-promoter interactions: CMO leverages a comprehensive database (GeneHancer) to link regulatory elements (enhancers) to their target genes.
- Genome-Wide Association Study (GWAS) summary results: This provides the primary trait association data.
The CMO test proceeds in three steps:
- Step 1: Linking CpGs to Genes: CpGs (cytosine bases in DNA) located in enhancers, promoters, and gene bodies are linked to their target genes. This is based on the understanding that DNA methylation in these regions plays a vital role in gene regulation.
- Step 2: Testing for Each Linked CpG: For each linked CpG, CMO tests its association with the trait of interest using a weighted gene-based test. To ensure robust performance across various scenarios, it combines results from multiple established tests (MWAS, SSU, and ACAT) using a Cauchy combination test.
- Step 3: Integrating Over Multiple CpGs: Finally, CMO aggregates the statistical evidence from all CpGs linked to a single gene using another Cauchy combination test to determine the overall gene-trait association.
This multi-layered integration aims to capture a more comprehensive picture of genetic regulation than previous methods.
Datasets
The study utilized several large-scale datasets for both simulation and real-world application:
- UK Biobank: Used for simulations, including genetic data from 10,000 independent individuals.
- APOE gene region data: Imputed genetic data for cis-SNPs within the APOE gene region from the UK Biobank was used for simulations.
- Alzheimer's Disease GWAS Summary Datasets:
- IGAP1: A meta-analyzed AD GWAS dataset with 17,008 cases and 37,154 controls.
- GWAX: A GWAS by proxy (using family history) from UK Biobank with 14,482 proxy cases and 100,082 proxy controls.
- Largest AD GWAS: A dataset with 71,880 proxy cases and 383,378 proxy controls of European ancestry.
- Susceptibility Weighted Imaging (SWI) GWAS: Results from 21 SWI GWAS studies (N = 8,428) were used to evaluate statistical power.
- GWAS Catalog: Used for validating identified genes.
Key Insights and Ideas
- Epigenetics and enhancers are critical for AD risk: The study strongly suggests that DNA methylation and enhancer-promoter interactions are not just bystanders but active players in AD pathogenesis, and integrating them is key to a more complete understanding.
- CMO's integrative approach unlocks new discoveries: By combining multiple layers of regulatory information, CMO can identify genes that are missed by methods focusing solely on gene expression or individual DNA methylation sites.
- Novel AD loci identified: The discovery of six novel AD risk loci highlights the power of CMO and suggests new avenues for AD research and potential therapeutic targets.
- Computational efficiency: CMO's ability to calculate p-values analytically makes it computationally efficient, allowing for rapid analysis of large datasets.
Key Results
- Improved Statistical Power: Through simulations, CMO demonstrated significantly higher statistical power compared to TWAS and other competing methods, especially when genes explained a substantial portion of trait variance.
- More Associations Identified: In analyses of brain imaging-related GWAS, CMO identified an average of 124% more associations than TWAS.
- Novel AD Gene Discovery: When applied to the largest AD GWAS dataset, CMO identified 159 genome-wide significant genes, including six novel loci for AD that were not detected by other methods.
- Replicable Findings: Genes identified by CMO showed good replication rates in independent datasets and were enriched in pathways related to AD.
- Computational Speed: CMO completed p-value calculations for all available genes in the IGAP1 dataset within approximately 1.5 minutes using parallel computing.
Limitations and Future Work
- Association vs. Causality: CMO is an association-based test, meaning identified genes do not imply direct causality. Fine-mapping methods are suggested for prioritizing causal genes.
- Prediction Model Accuracy: The power of CMO is dependent on the accuracy of genetically predicted DNAm values. Current models are based on European ancestry, and improvements are expected with better prediction models and diverse ancestry data.
- Equal CpG Weighting: In its current version, CMO assigns equal weights to all CpGs. Future work could improve power by incorporating prior information or differential weighting for CpGs in different regulatory regions.
- Tissue Specificity: The study focused on blood tissue for illustration. Future research could explore tissue-specific DNAm prediction models and enhancer information (e.g., using brain-specific datasets like PsychENCODE or ROS/MAP) to better understand AD etiology.
Summary updated