Replication of Yu et al. (2024) estimating the causal relationship between GLP-1 receptor agonists (GLP-1RA) and ischemic stroke using Mendelian Randomization.
library(data.table)
library(dplyr)
library(ggplot2)
library(TwoSampleMR)
library(ieugwasr)
library(MRPRESSO)
data/GLP1RA-eqtls - Sheet1.csv — 22 significant cis-eQTL SNPs for the drug target gene GLP1R, sourced from Yuming et al. (2024). All instruments pass F-statistic threshold (F > 47), indicating strong instrument relevance.
European ancestry-specific T2D GWAS summary statistics from the DIAGRAM Consortium, used as a positive control given the glucose-lowering mechanism of GLP-1RAs.
GWAS summary statistics extracted from Open GWAS — Malik et al. (2018), Open GWAS ID: ebi-a-GCST005843.
Two sample Mendelian Randomization was performed using the TwoSampleMR package in R. Primary causal estimates were obtained using the inverse variance weighted (IVW) method, with MR-Egger, weighted median, and weighted mode as sensitivity analyses.
Positive control: Proxies were first tested against T2D outcomes to validate instrument relevance, given the established glucose-lowering mechanism of GLP-1RAs.
Sensitivity analyses:
- Horizontal pleiotropy: MR-Egger intercept test and MR-PRESSO global test
- Heterogeneity: Cochran's Q statistic (MR-Egger and IVW)
- Outlier influence: leave-one-out analysis
GLP-1RA proxies were significantly associated with reduced odds of T2D (IVW: OR = 0.82, p = 2.3×10⁻¹⁷), validating instrument relevance. No heterogeneity (Q-statistic p > 0.05) or pleiotropy (MR-Egger intercept = 0.046, p = 0.625) was detected.
Consistent with the original paper, genetically proxied higher GLP-1R expression was associated with increased odds of ischemic stroke (~11.6% higher odds, IVW). Tests for pleiotropy (MR-Egger intercept = 0.048, p = 0.704) and heterogeneity (Cochran's Q p = 1.0) were non-significant. MR-PRESSO global test was also non-significant (p = 1.0, RSS = 0.73).
The paradoxical direction of effect i.e. increased stroke risk contrary to cardioprotective effects of GLP1-RAs in CVOTs, may reflect limitation of whole-blood eQTLs as proxies for drug action. Randomised trials studying comparative effectiveness of drugs against health outcomes like MACE, often include non-healthier group of people eg. One who have Type 2 diabetes, to ensure that outcome occurs. It is hard to achieve similar condition in mendelian randomisation study. Also, drug-target action are often acute, and germline genetic proxies are fixed at conception.