Multi-cohort and single-nucleus transcriptomics identify a five-gene lactylation-related panel and a disease-dependent hepatocyte glycolysis-lactylation-related expression program in metabolic dysfunction-associated steatohepatitis

Lysine lactylation couples glycolytic flux to protein and chromatin regulation, but the expression landscape of lactylation-related genes (LRGs) in metabolic dysfunction-associated steatohepatitis (MASH) has not been mapped. We analysed 320 LRGs across eight bulk, single-nucleus and spatial transcriptomic datasets. Sixty-six LRGs were differentially expressed in a merged training cohort (51 controls, 49 MASH). A severity-anchored funnel and four machine-learning algorithms then converged on a five-gene panel (SIRT1, CEBPZ, TKT, RPL14, CD2BP2). A neural network on the panel reached a nested cross-validated AUC of 0.89 and detected MASH against histologically normal liver in a classifier-external cohort (AUC 0.90-0.97). The panel did not separate MASH from simple steatosis (NAFL), a boundary shared by all four benchmarked signatures (AUC 0.50-0.70). In 265,271 nuclei from 48 livers, the LRG expression program localised to hepatocytes and rose most significantly in Hep3, a sterol- and lipid-biosynthetic state that expanded 7.2-fold in MASH. Hepatocyte transcriptomic LRG scores tracked glycolysis across MASH patients (rho = 0.93, n = 19) but not across controls (rho = 0.01, n = 23; P = 1.3 × 10⁻⁶ for the difference). This disease dependence exceeded that of 1,000 expression-matched random gene sets. Hep3 and glycolysis signals were oriented centrilobularly in Visium sections from four donors. The panel offers a transcriptomic MASH detection axis, and the Hep3-centred program a hypothesis for hepatic lactylation that lactylomic profiling can now test.