Genetic Associations Among Appendicular Lean Mass, Alanine Aminotransferase, and Type 2 Diabetes: Multi-Dataset Mendelian Randomization and Colocalization at the SERPINA1 Candidate Locus

Background/Objectives: Appendicular lean mass (ALM), alanine aminotransferase (ALT), and type 2 diabetes (T2D) may be genetically linked. However, the ALT-related pathway and shared signals at the SERPINA1 candidate locus remain uncertain. We used Mendelian randomization (MR) to evaluate associations among ALM, ALT, and T2D and compared candidate-locus signals across datasets. ALM represented muscle mass, not clinical sarcopenia. Methods: Using European-ancestry GWAS summary statistics, we performed six bidirectional two-sample MR analyses, multivariable MR (MVMR) with finite-sample t-based inference, and exploratory path-specific two-step MR. FinnGen R12 provided the T2D outcome data for MR and pathway analyses. Pathway analyses used 5000 shared-sampling bootstrap replicates to assess uncertainty in path estimates. We combined locus-level coloc/ABF with signal-specific SuSiE-coloc in the GRCh37 SERPINA1 region. T2DGGI was used only for colocalization. Molecular-QTL and external-omics analyses were treated as supportive or limiting evidence. Results: Higher genetically predicted ALM was associated with lower ALT and lower T2D risk, whereas higher genetically predicted ALT was associated with greater T2D risk. T2D→ALT was method-dependent. In MVMR, the ALT conditional effect remained positive, whereas the ALM direct effect was model-dependent. Path decomposition was compatible with an ALT-related statistical pathway, but product- and difference-based estimates disagreed, and residual heterogeneity remained. ALM–T2DGGI supported an rs28929474-related shared component, although component-cohort overlap was possible. FinnGen supported other ALT-related signals but did not independently corroborate rs28929474. Tissue eQTL and external omics did not provide consistent mechanistic support. Conclusions: The data support directionally connected MR-based associations and dataset-dependent shared candidate signals. These findings do not establish biological mediation, a unique causal variant, gene-level causal assignment, or a molecular mechanism.