Predictive Models of Recurrence from Transcriptomic Signatures of the Tumor Microenvironment and Cell Cycle in Stage III Colon Cancer from PETACC-8 and IDEA France Trials.

3523 Background: The objective of this work was to establish predictive models of the risk of recurrence in stage III colon cancer (CC) based on transcriptomic signatures of the tumor microenvironment (TME) and cell cycle from the PETACC-8 (training set) and IDEA- France (validation set) trials. Methods: 3'RNAseq was performed in 1733 patients from the PETACC-8 trial (85%) and 1263 patients from the IDEA-France trial (63%). 4 transcriptomic signatures were analyzed: a signature reflecting T-cells infiltration named "Immunoscore-Like", a signature reflecting M2 macrophage infiltration named "M2-like", the expression of CXCL13 (reflecting B cells infiltration) and a score based on the Oncotype DX® Colon Cancer RS using the same formula from the stromal score and the cell cycle score, named “Oncotype DX Like”. In the 1 st multivariate model, we analyzed the 4 signatures separately with a dichotomization into "high" and "low" with the best cut point value for the prediction of time to recurrence (TTR) in PETACC-8 trial. In the 2 nd multivariate model, we defined a score named “IPS” (Immune Proliferative Stromal), corresponding to the number of deleterious signatures (“high” or “low” depending on the signatures), ranging from 0 to 4. Results: The 1 st multivariate model, built from PETACC-8 trial, showed that these 4 signatures were significantly associated with TTR with a protective effect of Immunoscore Like and CXCL13 "high" (HR: 0.66, p = 0.003 and 0.60, p < 0.001 respectively) and a deleterious effect of M2 Like and Oncotype Like “high” (HR: 1.28, p = 0.05 and HR: 1.37, p = 0.01, respectively), independently of known prognostic factors, with a C-index of 0.73. This model was applied to the IDEA-France cohort by calculating a predictive score for each patient, with TTR significantly different depending on the quartile of this score with a 3-year TTR ranging from 55% for the lowest quartile to 90% for the highest quartile (p < 0.001). In the 2 nd model, in the 2 cohorts, IPS score was independently associated with TTR with a HR increasing with the IPS score, independently of T and N stage and intra-tumoral CMS heterogeneity (Table). Conclusions: Using transcriptomic data of patients with stage III CC from 2 large-scale adjuvant trials, two predictive models based on signatures of the TME and the cell cycle, provide important information in addition to known prognostic factors for patient stratification on risk of recurrence. Beyond T and N stage, for the decision of adjuvant chemotherapy in stage III CC, the combination of these different variables could be exploited in the future for personalized care (de-escalation, intensification). [Table: see text]