Integrated multi-omics of mitophagy-related molecular subtype characterization and biomarker identification in sepsis.

As a life-threatening condition driven by dysregulated host responses to infection, sepsis suffers from high mortality and heterogeneity. Mitophagy is the selective removal of damaged mitochondria, which is implicated in mitigating sepsis-related damage. The systematic identification and validation of key mitophagy-associated genes (MAG) for sepsis diagnosis, stratification, and immune modulation are lacking. Bulk transcriptomic datasets were integrated for differential expression analysis, Weighted Gene Co-expression Network Analysis (WGCNA), and machine learning. We analyzed single-cell RNA-seq data to map MAG expression, performed immune infiltration analyses by ESTIMATE, single-sample Gene Set Enrichment Analysis (ssGSEA) and conducted consensus clustering based on MAG for molecular subtyping. As a screened MAG, the role of NUP93 was functionally validated in mitophagy using lipopolysaccharide (LPS)-stimulated RAW264.7 cells with adenoviral overexpression. Integration of machine learning identified four MAG biomarkers (RPL18, PRPF8, NUP93, CUL1) with high diagnostic power (individual AUCs 0.957-0.975, nomogram AUC = 0.990). Consensus clustering based on these MAG stratified sepsis patients into distinct molecular subtypes with differing MAG expression, immune landscapes, and underlying immune-related pathways. NUP93 overexpression in vitro rescued LPS-induced mitophagy impairment by restoring mitochondrial PINK1 and LC3B levels. This study identifies RPL18, PRPF8, NUP93, and CUL1 as robust diagnostic MAG biomarkers for sepsis, demonstrates their utility in defining molecular subtypes with divergent immune microenvironments, and provides functional evidence that NUP93 promotes mitophagy during sepsis, offering novel tools for precision diagnosis and insights for targeted therapeutic strategies.