One-sample MR-PheWAS in the UKB population

To explore the effects of nine serum amino acids on a wide range of diseases, we conducted PheWAS analyses. We extracted the phenotypic data of the UKB participants, including patient hospital records, cancer registry data and death registry data defined as ICD-9/10 codes, which were linked to Hospital Episode Statistics, as well as the genotypic data (48–138 amino acid-associated SNPs identified in our GWAS analysis) of the UKB participants (n = 117 944). We selected SNPs with P-value <1 × 10⁻⁵ identified in our UKB GWAS with an F-statistic >10 (calculated by regressing serum levels per each SNP) as genetic instruments. Multiallelic and palindromic SNPs were excluded, and the remaining SNPs were clumped with r² < 0.001. We conducted a one-sample MR analysis approach using the two-stage least-squares method by regressing each serum amino acid on their genetic instruments using a linear regression model and saved the residual values. Subsequently, we used a logistic regression model to regress the saved residual values on each phecode comprising a wide range of clinical diagnoses in the UKB study (n = 617 clinical diagnoses with >200 cases). To avoid overlap with the sample that was used for GWAS, the PheWAS analysis was merely done in the ∼385 000 individuals who were not used for GWAS. In addition, the PheWAS analysis was restricted to participants from European ancestry and a kinship coefficient of >0.088 was used to exclude the related individuals. Our PheWAS analysis was adjusted for age, sex and the first 10 genetic principal components. We used the false discovery rate (FDR) method to account for multiple testing as the Bonferroni procedure is suggested to be too conservative due to the correlation between clinical diagnoses.