![part_0_1.png]()
Metabolite To Disease
For metabolite-disease pairs that showed significant associations, we further conducted Mendelian randomization analyses to determine the direction of causation.
In a total of 475 incident diseases showed statistical associations with at least one metabolite, we utilized GWAS statistics for MR analysis.
We accessed FinnGen R10 GWAS statistics for 368 diseases (https://www.finngen.fi/en/access_results) and performed GWAS analysis for the remaining 107 diseases in a subset of 220,000 white British UKB participants who did not have metabolite data available (available at UKB-derived Diseases' GWAS Data Repository).
Metabolites' GWAS summary statistics were derived from a genome-wide meta-analysis of 233 NMR circulating metabolic traits, involving up to 136,016 participants from 33 cohorts1. They are publicly available through the NHGRI-EBI GWAS catalogue (GCST90301941–GCST90302173) and https://www.phpc.cam.ac.uk/ceu/lipids-and-metabolites/. Among these, we included the data of 223 metabolic traits that overlap with the metabolic profile analyzed in our study.
For primary bidirectional MR analyses, instrumental variables were selected using PLINK 2.0 clumping function (clump-kb 500, clump-r2 0.1, clump-p1 5 × 10−8), and the European 1,000 Genomes phase 3 dataset was used as the reference genome.
The Wald ratio and inverse variance weighted (IVW) methods were employed to estimate MR effects for a single or multiple instruments, respectively.
The GWAS analyses were performed using generalized linear mixed models (GLMM) implemented through the Genome-wide Complex Trait Analysis (GCTA, v1.94.0)2 and MR was conducted using TwoSampleMR package (v0.6.1) in R.
As an additional validation, we performed MR analysis using UKB-derived metabolites' GWAS. For each metabolite, outlier values deviating by more than four times the interquartile range from the median were removed, followed by natural log transformation. A total of 189,846 white British participants with both metabolomics and genomic data available were included. GWAS analysis was conducted using an additive linear regression model implemented in PLINK2.0. The covariates included age, ethnicity, sex, fasting time, month of assessment, genotype measurement batch, the top 40 genotype PCs, age indicators by sex interactions, and ethnicity by sex interactions. We then repeated the primary MR and colocalization analyses using this UKB-derived GWAS dataset. Replication rates of metabolite-disease associations assessed with an FDR-corrected q-value threshold of < 0.05. Metabolites' GWAS summary statistics are available to access through UKB-derived Metabolites' GWAS Data Repository.