AI-based phenotyping of hepatic fiber morphology to inform molecular alterations in metabolic dysfunction–associated steatotic liver disease

BACKGROUND AND AIMS: Hepatic fiber morphology may significantly enhance our understanding of molecular alterations in metabolic dysfunction-associated steatotic liver disease (MASLD). We aimed to comprehensively characterize hepatic fiber morphological phenotypes in MASLD and their associated molecular alterations using multilayer omics analyses. APPROACH AND RESULTS: To quantify the morphological phenotypes of hepatic fibers, the artificial intelligence-based FibroNest algorithm (PharmaNest) was applied to 94 MASLD-affected liver biopsies, among which 12 (13%) had concurrent HCC. FibroNest identified 327 fiber phenotypes that were summarized into 8 major principal components, named FibroPC1-8. Next, molecular alterations captured by morphological fiber phenotypes were evaluated by comparison with genome-wide transcriptomics of paired liver samples. Pathway analyses revealed that FibroPCs more sensitively captured MASLD-related molecular alterations, such as upregulation of interleukin-6 and susceptibility to resmetirom, compared with the histological fibrosis stage. Among them, FibroPC4, which reflects reticular fibers, was associated with a gene signature predictive of incident HCC from MASLD. Furthermore, we used a spatial single-cell transcriptome, CosMx, to reveal the cell-cell interactions driving MASLD pathogenesis, as captured by FibroPC4. CosMx revealed that the FibroPC4-rich microenvironment contains HCC-promoting HSCs located adjacent to periportal endothelial cells. Neighboring cell analyses suggested that the HCC-promoting phenotype of HSCs was acquired by insulin growth factor-binding protein 7 secreted from senescent periportal endothelial cells. Consistently, in vitro experiments showed that insulin growth factor-binding protein 7 transformed HSCs into an HCC-promoting phenotype. CONCLUSIONS: Hepatic morphological fiber phenotyping can reveal the disease progression and underlying mechanisms of MASLD.