Tumour and stroma RNA signatures predict more accurately distant recurrence than clinicopathological factors in resected pancreatic adenocarcinoma.

ABSTRACT

BACKGROUND: Few patients with pancreatic adenocarcinoma (PAC) are eligible for surgery. Patients with early relapse have a poor prognosis and might be better candidates for a medical approach. Clinical and pathological parameters only partially predict recurrence and are only obtained after surgery. PAC subtypes based on gene expression were proposed, and we assessed if they could predict the risk and type of recurrence independently of clinicopathological parameters. METHODS: Patients with curative-intent surgery for PAC without pretreatment were selected and divided into two independent cohorts defined as discovery (n = 381) and validation (n = 149) cohorts. Transcriptomic analyses were performed on formalin-fixed paraffin-embedded surgical samples to characterise tumour and stroma compartments using previously defined signatures. We associated molecular and clinicopathological characteristics with general, distant, and local recurrences using Cox regression analyses. RESULTS: We found that tumour biology predicted distant recurrence contrary to local recurrence, which was directly related to resection margin status. Pure basal-like and stroma-activated subtypes were strongly associated with distant recurrence, independently of clinicopathological factors (hazard ratios [HRs] = 5.85, p < 0.001 and HR = 1.75, p = 0.007, respectively). By dissecting tumoural and stromal compartments, we demonstrated that the basal-like tumour component positively correlated with distant recurrence in both cohorts (HR = 1.45, p < 0.001 and HR = 1.90, p < 0.001), whereas the inactive structural stroma component was protective against distant recurrence (HR = 0.68, p < 0.001 and HR = 0.72, p < 0.001). CONCLUSIONS: In addition to suggesting a different mechanism for local and distant relapse (incomplete resection and high metastatic potential, respectively), our results show the potency of molecular phenotype to predict patient outcome regarding distant recurrences.