Abstract 1989
Background
The consensus molecular subtypes (CMS) of colon cancer only accounts for inter-tumor heterogeneity so far. To overcome this limitation and describe intratumor heterogeneity, we split colon tumor transcriptomes as weighted linear combinations of the four CMS. This leads to identify pure samples, corresponding to a unique CMS, and mixed samples, corresponding to several CMS. Using the PETACC8 cohort, we evaluate the prognostic impact of this new weighted CMS classification.
Methods
Paired Nanostring and Affymetrix data were available for the CIT training series (n = 198). NanoString expression data (172 genes) were available for 1779 samples from the PETACC8 trial validation cohort. A random forest classifier (rfc) of the CMS was trained on the Nanostring CIT data and then applied to PETACC8 samples. Using linear regressions, we identified for each sample the weighted linear combination of CMS centroids with minimal distance to its transcriptome profile.
Results
Applying rfc to PETACC8 confirmed clinical and molecular data of the CMS classification previously published. Applying the weighted classifier to PETACC8 yielded 42.6% of pure tumors, displaying a unique CMS signature, and 57.4% of mixed tumors, displaying at least 2 CMS signatures (each weighting more than 20%). Mixed tumors were synthetized by their top 2 CMS components (major/minor). Therefore, we defined 16 subgroups with prevalence ranging from 2.1% to 18.3%. MMR deficiency and BRAF mutations were found enriched in pure CMS1, CMS1/CMS3 and CMS3/CMS1 mixed tumors. The pure CMS3 was the only subtype enriched for RAS mutations. For DFS univariate Cox analysis of this 16 subtype classification revealed a significant worse prognostic impact of CMS4 signature either pure or combined to any other CMS signature. The worst prognostic was observed in CMS3/CMS4 subtype (HR = 2.43 CI[1.6-3.7], followed by CMS1/CMS4 (HR = 2.36 CI[1.5-3.6] CMS4/CMS1 (HR = 2.3 CI[1.6-3.5]), CMS1/CMS3 (HR = 2 CI[1.3-3.1]) and CMS4/CMS3 (HR = 1.9 CI[1.2-3]) . These results were confirmed after adjusting for significant clinical variables.
Conclusions
Taking into account the intra tumor heterogeneity brings additional information to colon cancer stage III prognostication.
Clinical trial identification
Legal entity responsible for the study
Fédération Francophone de Cancérologie Digestive.
Funding
Merck KGA.
Editorial Acknowledgement
Disclosure
All authors have declared no conflicts of interest.