MetScore and classical-basal subtype intersect to define chemotherapy response and tumor microenvironment architecture in pancreatic ductal adenocarcinoma

Background Transcriptomic heterogeneity in pancreatic ductal adenocarcinoma (PDAC) is conventionally modeled along a single classical-basal axis, but this framework incompletely explains clinical behavior. We hypothesized that MetScore, a single-sample measure of metastatic colonization potential, defines a complementary axis. Objective To determine how MetScore and classical-basal identity jointly shape PDAC biology and clinical outcomes. Design We integrated RNA sequencing, DNA sequencing, clinical, and digital pathology data from 512 tumor biopsies from 510 patients in the Pancreatic Cancer Action Network Know Your Tumor real-world cohort. We assessed associations with biopsy site, survival, chemotherapy response, and tumor microenvironment composition; validated metastatic discrimination in an independent cohort; and examined cancer cell-intrinsic drug sensitivity in vitro. Results MetScore distinguished metastatic from primary biopsies, a finding validated externally. High MetScore and basal identity were independently associated with inferior survival. High MetScore and classical identity were each associated with preferential response to 5-fluorouracil (5-FU)-based rather than gemcitabine-based therapy, with the strongest enrichment for 5-FU responses among classical/MetScore-high patients. In vitro drug-sensitivity experiments did not recapitulate the clinical treatment-response pattern, motivating investigation of cancer cell-extrinsic factors. Transcriptomic deconvolution and histopathologic machine learning-based cell annotation demonstrated that high-MetScore and basal states converged on macrophage/monocyte enrichment but diverged in their associations with cancer-associated fibroblasts. Conclusion Integrating MetScore with classical-basal identity establishes a two-axis framework that better explains metastatic behavior, prognosis, treatment response, and microenvironmental composition than the conventional one-axis model and provides a clinically testable strategy for transcriptome-guided patient stratification.