Abstract / Summary
Abstract Pancreatic ductal adenocarcinoma (PDAC) is shaped by extensive interactions among malignant, stromal and immune compartments, but connecting phenotype-associated cell states to reproducible tissue-level molecular patterns remains challenging. We developed a systems-level multi-omics framework integrating bulk transcriptomics, single-cell RNA sequencing, spatial transcriptomics, somatic alterations and targeted cell-line experiments. Scissor linked myeloid single-cell states to survival phenotypes in bulk cohorts, identifying enrichment of survival-associated cells within cDC_LAMP3, pDC_LILRA4 and MARCO_NLRP3 states and inflammatory, interferon and epithelial–mesenchymal-transition-related programs. Genes associated with these states were integrated with PDAC-upregulated genes and benchmarked across 113 machine-learning combinations, yielding an eight-gene Scissor-positive myeloid-derived signature (SMDS: FN1, SPP1, XAF1, IL1RAP, PLAU, CD109, IFI27 and MYOF). SMDS reproducibly discriminated PDAC from non-tumor pancreas across retrospective cohorts (AUC 0.918–0.990) and localized spatially to tumor-rich, macrophage- and fibroblast-associated niches with reduced lymphocyte signals. Genomic analyses linked high FN1 expression to copy-number gain and TP53-mutant contexts, while pharmacogenomic modeling predicted lower trametinib IC50 in SMDS-high tumors. FN1 silencing reduced PDAC-cell viability, although TP53 perturbation produced discordant FN1 mRNA and protein responses, arguing against a simple regulatory mechanism. Together, these results establish an integrative systems biology framework that connects phenotype-associated myeloid states with tissue architecture, genomic context and therapeutic hypotheses in PDAC, while supporting SMDS as a candidate research signature requiring prospective clinical and mechanistic validation.