PHIT score integrates tumor biology to predict PDAC prognosis and therapy response

Background: Pancreatic ductal adenocarcinoma (PDAC) exhibits wide variation in tumor plasticity (P), heterogeneity (H), immune suppression (I), and treatment resistance (T). Existing clinical and molecular classifiers capture only part of this biology. We developed PHIT, a 12-gene transcriptomic score integrating these four axes, and evaluated its prognostic and therapeutic relevance across resected and metastatic PDAC. Methods: RNA-seq from TCGA, PACA-CA/ICGC, and CPTAC (n=419) was batch-corrected and analyzed with weighted Cox models to generate gene-level survival coefficients. These were integrated with chemotherapy-response metrics from the metastatic COMPASS cohort (n=188; continuous tumor-volume change and PR/SD/PD categories across all patients and within FOLFIRINOX and gemcitabine/nab-paclitaxel subgroups) to build a joint survival-chemo composite. Bootstrap selection yielded a 12-gene PHIT signature. PHIT scores were computed using training-cohort means/SDs and Cox logHR weights. Prognostic performance was tested using Kaplan-Meier, Cox models, and c-index. Validation included PACA-AU, Chen, Nones, and Moffitt cohorts (n=339). Hallmark analyses compared PHIT-high vs PHIT-low tumors within molecular subtypes. Results: PHIT significantly stratified overall survival (OS) in the 419-patient training cohort (median 32.7 vs 10.4 months for PHIT Q1 vs Q4; HR 3.49, p=7×10⁻¹⁴), improved a tumor grade/stage model (c-index 0.590→0.685), and remained independently prognostic after adjustment for Moffitt, Collisson, and Bailey molecular subtypes (HR 2.36, p<0.001). In COMPASS, PHIT correlated with chemotherapy resistance (tumor-volume change r=0.40 overall; 0.45 FOLFIRINOX; 0.34 GA) and stratified metastatic OS (12.6 vs 5.6 months; HR 2.92, p=1.4×10⁻⁵). In validation cohorts, PHIT consistently separated risk: PACA-AU (50.4 vs 10.9 months; HR 4.90, p=0.008), Chen (31.5 vs 15.6; HR 3.09, p=0.01), Nones (24.5 vs 15.1; HR 3.47, p=0.01), and combined arrays (30.1 vs 15.1; HR 2.45, p=6.45×10⁻⁵). Mechanistically, PHIT-high Classical tumors showed selective enrichment of EMT, ECM remodeling, and inflammatory NF-κB pathways, whereas PHIT-high Basal-like tumors showed glycolytic and metabolic plasticity with hypoxia, mTORC1, and proliferative G2M/E2F programs. These subtype-specific patterns were reproducible across Collisson and Bailey frameworks. Conclusions: PHIT is a concise transcriptomic score that integrates four dimensions of aggressive PDAC biology. It predicts survival across seven cohorts (n=946), correlates with chemotherapy response, and reveals distinct EMT- and glycolysis-associated high-PHIT phenotypes, supporting its utility for biological risk stratification and treatment selection.