PD-003Identification of a nanostring signature that differentiates early pancreatic cancers according to stromal composition and predicts clinical outcome

Introduction: Surgery followed by adjuvant chemotherapy represents the only form of curative treatment for pancreatic adenocarcinoma (PDAC). However, in patients amenable to surgical resection the 5 years survival rate is still less than 50%, suggesting that peri-operative strategies need to be improved to maximise the chances of cure. PDACs are known to be rich in stroma. Whole genome transcriptomic data identified the prognostic value of genes that reflect the enrichment of activated stromal cells in the tumour tissue. However, large scale transcriptomic analyses, such as microarray and RNA-seq profiling are difficult to implement in routine clinical practice. On the contrary, the nCounter platform provides a fast and easily accessible solution for multiplex analysis of formalin fixed paraffin embedded tissues. Methods: A custom Nanostring-panel including probes for 198 genes associated with stromal activation was run in the tumour tissue of a cohort of clinically-annotated human early stage PDACs. Kaplan Meier analysis was used to investigate the association between the signature and survival outcomes [Relapse Free Survival, RFS and Overall Survival, OS). The prognostic value of the signature was derived in an exploratory cohort [Royal Marsden Hospital (N = 44)], and was confirmed in a validation cohort [The Cancer Genome Atlas (N = 141)]. The C-MAP dataset was used to retrieve data on perturbations that can reverse the gene changes occurring in the bad-prognosis group. Results: A signature based on the unsupervised clustering of the exploratory cohort (coefficient variation 1.1) and the expression of the 100 top-ranked genes could distinguish the population in two groups with different prognosis (good prognosis: median OS 24 months, bad-prognosis: median 0S 14 months; p:0.02). Gene ontology analysis of the genes identifying the two subgroups showed enrichment of transcripts linked to the matrisome and the extracellular space (bad-prognosis group) and enrichment of genes associated to the immune response (good-prognosis group). The group identified by the immune-signature had a reduced risk of both relapse (hazard ratio (HR) 0.46, p 0.03) and death (HR 0.46, p 0.02), which maintained statistical significance at the multivariate analysis (p 0.003). The signature confirmed good performance in the validation cohort (p:0.02 for OS). The bad prognosis signature was used as an input to the C-MAP dataset, which identified a number of compounds able to reverse the transcriptomic changes associated with the increased risk of recurrence, including modulators of the tankyrase, CD40, Interferon, and Tumour Necrosis Factor pathways. Conclusion: We identified a 100 gene-expression signature that can be run on the nCounter platform and can 1) predict the prognosis of resected PDAC patients, and 2) inform drug discovery projects for the development of therapeutics that reverse the stromal transcriptomic changes associated to the increased risk of death from PDAC.