TL;DR
This paper introduces T-GEN, a novel method that leverages epigenetic information to improve the identification of genes associated with complex diseases. By prioritizing functionally relevant genetic variants, T-GEN successfully identifies more disease-associated genes and novel disease loci, particularly for late-onset Alzheimer's disease, than existing methods.
What Problem Does This Paper Address?
Identifying the specific genes underlying complex diseases (like diabetes, heart disease, or Alzheimer's) is a major challenge in genetics. Genome-Wide Association Studies (GWAS) can pinpoint genetic regions associated with these diseases, but these regions often contain many genes, making it difficult to determine which gene is truly responsible. Transcriptome-Wide Association Studies (TWAS) aim to bridge this gap by using gene expression data as an intermediary. TWAS methods predict gene expression levels from genetic variations and then link these predicted expression levels to disease phenotypes. However, a key limitation is the accurate identification of "expression quantitative trait loci" (eQTLs) – the genetic variants that actually influence gene expression. Many identified eQTLs might be in linkage disequilibrium (LD) with the true functional variant, leading to misattribution of disease associations. This paper addresses the challenge of accurately identifying functional eQTLs and, consequently, more reliable disease-associated genes.
Methods
The core innovation of T-GEN lies in its integration of epigenetic information to prioritize SNPs (Single Nucleotide Polymorphisms) for gene expression imputation.
- Epigenetic Prioritization: T-GEN uses epigenetic marks (like H3K4me1, H3K4me3, H3K9ac) from the Roadmap Epigenomics Project to identify SNPs that are likely to be functionally active and regulate gene expression. This is based on the observation that active epigenetic marks are enriched in regulatory DNA regions.
- Bayesian Variable Selection: A bi-level variable selection model, employing a spike-and-slab prior, is used to model the genetic effects of SNPs on gene expression. This model allows for the selection of SNPs that are likely to have a regulatory role, guided by their epigenetic annotations.
- Tissue-Specific Imputation: Gene expression imputation models are built for each tissue separately, leveraging both genotype and epigenetic data.
- Association Testing: Imputed gene expression levels are then used in a univariate regression model to test for associations with disease phenotypes, similar to existing TWAS methods.
T-GEN is compared against several established methods, including elastic net (elnt), variational Bayes (vb), and their annotated variants (elnt.annot, vb.annot).
Datasets
- GTEx (Genotype-Tissue Expression) Project: Used for gene expression data across 26 human tissues and genotype data.
- Roadmap Epigenomics Project: Provided epigenetic annotation data.
- LD Hub: A database of summary statistics for 207 complex traits, used for trait-gene association analysis.
- Independent GWAS for AD: Used for replication of findings in late-onset Alzheimer's disease.
Key Insights and Ideas
- Epigenetics as a Functional Filter: The paper strongly supports the idea that epigenetic marks can serve as a powerful filter to identify SNPs that are not just statistically correlated with gene expression but are also biologically functional.
- Improved eQTL Identification: By prioritizing functional SNPs, T-GEN identifies eQTLs that are more likely to be truly regulatory, leading to more accurate gene expression imputation.
- Novel Disease Loci Discovery: The method's ability to pinpoint functional variants allows it to uncover novel genetic loci associated with diseases, which might be missed by methods that don't incorporate functional information.
- Tissue-Specific Relevance: T-GEN's tissue-specific approach, combined with epigenetic data, helps identify disease genes in tissues most relevant to the disease's biology.
Key Results
- Functional eQTLs: T-GEN identified a significantly higher percentage of eQTLs (18.7% to 47.2% increase) that are inferred to be functional by ChromHMM and have higher deleteriousness scores (CADD).
- More Associated Genes: Across 207 complex traits, T-GEN identified more trait-associated genes (7.7% to 102% increase) and loci compared to existing methods.
- High pLI Score Genes: T-GEN was better at identifying genes with high probability of loss-of-function intolerance (pLI > 0.99), suggesting a focus on evolutionarily conserved and functionally important genes.
- Late-Onset Alzheimer's Disease (AD): T-GEN identified 96 genes at 15 loci for AD, including two novel loci not previously implicated in GWAS. Fifty of these genes were successfully replicated in an independent AD GWAS.
- Replication Rates: Genes identified by T-GEN showed higher replication rates in external datasets compared to those identified by other methods.
Limitations and Future Work
- Imputation Accuracy: While T-GEN improves eQTL identification, the accuracy of gene expression imputation can still be a limiting factor. The authors suggest that integrating non-parametric Bayesian models and considering epigenetic annotations could further improve this.
- Trans-eQTLs: The current method primarily focuses on cis-eQTLs (variants near the gene). Incorporating trans-eQTLs (variants far from the gene) could potentially identify more trait-associated genes and co-regulatory peripheral genes.
- Experimental Validation: The identified eQTLs and gene-trait associations require further experimental validation.
- Epigenetic Data Limitations: The use of a reference panel for epigenetic data (Roadmap) limits the potential for leveraging individual-specific epigenetic variations, which can change over time and across individuals.
- Complexity of Regulation: The authors acknowledge that the relationship between epigenetic signals and gene regulation is complex, and their current model makes simplifying assumptions.
- Future Directions: The paper suggests that using more experimental epigenetic data, integrating trans-eQTLs, and developing benchmarking datasets for TWAS methods could advance the field. Mendelian randomization and gene-level fine-mapping are also proposed for validating causal relationships.
Summary updated