Integration of single-cell and bulk RNA sequencing reveals programmed cell death-associated transcriptional programs in sepsis-induced acute lung injury.

BACKGROUND: Sepsis-induced acute lung injury (ALI) is a frequent and life-threatening complication of sepsis, yet clinically actionable transcriptomic biomarkers remain limited. Regulated/programmed cell death (PCD) pathways shape inflammatory injury and barrier dysfunction, but their cell-type-specific transcriptional signatures in sepsis-induced ALI are incompletely defined. METHODS: Bulk transcriptomes from E-MTAB-5273 (training; sepsis-induced ALI vs sepsis) and E-MTAB-5274 (external validation) and scRNA-seq data from GSE207651 (CLP vs sham) were analyzed. Thirteen curated PCD gene sets were scored by ssGSEA, and differential PCD pathways were identified using the Wilcoxon rank-sum test with multiple-testing correction. WGCNA and differential expression analysis (limma) were integrated to obtain differentially expressed PCD-related genes (DE-PCDRGs). We benchmarked 113 machine-learning model combinations (12 algorithms) under cross-validation to select an optimal classifier, and interpreted predictions using SHAP and LIME. Associations between the PCD score and immune/metabolic signatures were assessed by ssGSEA. Cell types enriched for model-gene expression were localized in scRNA-seq, and key genes were validated in a CLP-induced ALI rat model using Western blot and immunohistochemistry. RESULTS: Five PCD processes differed between sepsis-induced ALI and sepsis, including increased apoptosis and pyroptosis and decreased lysosome-dependent cell death, NETosis, and alkaliptosis. Twelve DE-PCDRGs were identified, and an 8-gene signature (PADI4, IFI6, POLB, IFI27, GZMB, CD3E, CRIP1, CASP5) yielded the best performance (glmBoost feature selection + random forest classifier; AUC 0.988 in training and 0.817 in validation). Enrichment analyses linked model genes to ribosome-related pathways, cell adhesion molecules, and the intestinal immune network for IgA production. High vs low PCD score groups differed in 11 immune-cell signatures and 23 metabolic pathways. Single-cell analyses highlighted endothelial cells as a major compartment expressing multiple model genes. In vivo experiments confirmed differential protein abundance of PADI4, POLB, and IFI27 in CLP-induced ALI lungs, supporting their potential as biomarkers. CONCLUSION: Integrating bulk and single-cell transcriptomes delineated PCD-associated molecular features in sepsis-induced ALI and identified an externally validated 8-gene classifier signature. These results nominate endothelial-cell-linked PCD programs and the PADI4/POLB/IFI27 axis for further mechanistic studies and biomarker development.