Multi-omics mapping of PSC genetic risk to a high TWAS-active JAML+ lipid-associated macrophage program: bridging single-cell heterogeneity, spatial fibrotic niches, and machine learning selection

Background Primary sclerosing cholangitis (PSC) is a chronic cholestatic liver disease with progressive peribiliary inflammation and fibrosis. Disease-modifying therapies are lacking, and the cell-type-specific mechanisms linking genetic susceptibility to pathogenic immune states remain incompletely understood. Methods We performed a transcriptome-wide association study (TWAS) using PSC GWAS summary statistics and GTEx v8 liver eQTL weights. TWAS-prioritized genes were mapped onto a PSC liver single-cell RNA-seq atlas (GSE247128) using integrated gene set scoring (irGSEA). Monocyte/macrophage subsets were re-clustered and lipid-associated macrophages (LAMs) were stratified by TWAS activity. Core genes distinguishing high-versus low-activity LAMs were identified using seven machine-learning feature selection algorithms. JAML was validated in bulk transcriptomic cohorts (GSE119600, GSE177044), evaluated by immune infiltration analysis, examined by cell–cell communication inference (CellChat), assessed by virtual knockout (scTenifoldKnk), and spatially localized using Visium FFPE spatial transcriptomics (PSC: GSE245620; control: GSE240429), and validated at the protein level using Western blot and ELISA in an in vitro macrophage model. Results TWAS activity was predominantly enriched in the monocyte/macrophage lineage. Within this lineage, LAMs (TREM2+ GPNMB+ APOC1+) showed the highest TWAS activity and occupied late pseudotime states. High-activity LAMs were enriched in fibrosis-related pathways (e.g., TGF-β, NOTCH, WNT/β-catenin) and innate immune pathways (e.g., TLR2/4-MAPK, NLRP3 inflammasome). Multi-algorithm feature selection identified JAML (AMICA1) as a core discriminator of high-versus low-activity LAMs. JAML was upregulated in PSC in bulk cohorts, associated with higher macrophage and lower activated CD8 + T-cell infiltration, showed enhanced cell–cell communication signatures, and spatially co-localized with LAM and fibrosis scores in PSC tissue sections. In vitro , JAML upregulation accompanied NF-κB p65 phosphorylation and PD-L1 expression in LPS/IL-6-stimulated macrophages, and JAML knockdown attenuated these responses. Conclusion Integrating TWAS with single-cell and spatial transcriptomics highlights a genetically linked, highly active LAM state in PSC and nominates JAML + LAMs as a testable effector population potentially connecting genetic susceptibility to peribiliary fibrosis, with in vitro evidence supporting a JAML/NF-κB/PD-L1 inflammatory axis.