Abstract
Background: Lean nonalcoholic fatty liver disease (lean NAFLD) occurs in individuals with normal body weight, but the relationship between its molecular features and cardiovascular disease (CVD) remains unclear. We integrated expression quantitative trait locus (eQTL) data from several resources with Mendelian randomization (MR) to investigate druggable genes associated with GWAS-defined lean NAFLD and their potential metabolic links to CVD. Methods: We combined 2532 druggable genes with eQTLGen whole-blood cis-eQTL and lean NAFLD GWAS data. The primary MR analysis used stringent linkage disequilibrium (LD) clumping (r2 < 0.001), with a less stringent threshold (r2 < 0.1) for comparison. Under the less stringent setting, we performed exploratory analyses of the 19 selected genes using GTEx whole-blood eQTL data, conventional and SuSiE-based colocalization, sc-eQTL data from 14 immune cell types, and six CVD outcomes. Circulating metabolic traits were incorporated into two-step MR. Additional analyses included functional annotation using two background gene sets, DSigDB compound-associated gene sets, and molecular docking. Results: Stringent LD clumping yielded 1523 whole-blood MR estimates, of which 69 were nominally significant; none survived Benjamini–Hochberg false discovery rate (BH-FDR) correction. The 19 genes selected under the less stringent setting were retained for exploratory analyses. Under stringent GTEx settings, 15 genes were estimable, and four were nominally significant, but none survived FDR correction. Conventional colocalization was completed for 18 regions, none of which reached PP.H4 ≥ 0.75 under the primary priors. SuSiE identified no credible sets for lean NAFLD, precluding signal-pair colocalization estimates. Of 29 immune-cell MR estimates, only NEK1–CD4NC and SMN2–CD4NC were nominally significant; neither survived FDR correction. Seven whole-blood and one cell-level association with CVD survived correction within their respective testing families. The 19 genes showed 44 nominal associations with 64 preselected metabolic traits. After stringent instrument selection, four of 53 metabolic trait–lean NAFLD estimates survived correction within the preselected family, but no indirect effect survived multiple-testing correction. GO/KEGG analyses identified no FDR-significant terms under either background. DSigDB returned 701 entries, of which 72 survived FDR correction. Vina scores for four candidate target–compound pairs ranged from −11.23 to −8.37 kcal/mol; however, neither reference-ligand redocking met the prespecified pose-reproduction criterion, limiting interpretation of the candidate docking results. Conclusions: Stringent LD clumping and colocalization did not provide sufficient evidence to establish causal therapeutic targets for lean NAFLD. Analyses across eQTL resources, immune cells, CVD outcomes, and metabolic traits identified candidate associations for further investigation, but did not establish shared mediation mechanisms or drug efficacy. These findings apply primarily to the lean NAFLD phenotype defined in the source GWAS and require validation in independent populations, disease-relevant tissues, and functional experiments.