O-0025 - Colorectal cancer subtyping consortium (CRCSC) identifies consensus of molecular subtypes

Date 28 June 2014
Event World GI 2014
Session Hot topics in colon cancer
Topics Colon Cancer
Rectal Cancer
Pathology/Molecular Biology
Presenter Rodrigo Dienstmann
Citation Annals of Oncology (2014) 25 (suppl_2): ii105-ii117. 10.1093/annonc/mdu193
Authors R. Dienstmann1, L. Vermeulen2, A. Schlicker3, E. Missiaglia4, C. Soneson4, L. Marisa5, K. Homicsko6, X. Wang7, I. Simon8, L. Wessels9, J.P. Medema10, S. Kopetz11, S. Friend1, J. Guinney1, . Colorectal Cancer Subtyping Consortium Group1, M. Delorenzi4, A. De Reynies5, P. Roepman8, A. Sadanandam12
  • 1Sage Bionetworks, Seattle/US
  • 2Cancer Research UK - Cambridge Institute, Cambridge/UK
  • 3Division of Molecular Carcinogenesis, The Netherlands Cancer Institute, Amsterdam/NL
  • 4SIB Swiss Institute of Bioinformatics, Lausanne/CH
  • 5Ligue Nationale Contre le Cancer, Paris/FR
  • 6Swiss Institute for Experimental Cancer Research, Swiss Federal Institute of Technology Lausanne, Dpt. of Oncology Centre Hospitalier Universitaire Vaudois, Lausanne/CH
  • 7Centre for Biomedical Informatics, Harvard Medical School, Boston/US
  • 8Agendia NV Research and Development Dpt, Amsterdam/NL
  • 9The Netherlands Cancer Institute, Amsterdam/NL
  • 10LEXOR, Academic Medical Center, Amsterdam/NL
  • 11The University of Texas MD Anderson Cancer Center, Houston/US
  • 12The Institute of Cancer Research (ICR), London/UK



Several independent groups have recently reported novel molecular subtypes in colorectal cancer (CRC). A systematic comparison of the reported subtypes is needed to determine common findings and facilitate translation into the clinic. A collaborative effort by the CRCSC was initiated to enable open data sharing and meta-analysis with the goal of establishing a consensus subtyping model.


CRCSC participants, representing more than 15 institutions, analyzed 30+ gene expression sets, including multiple platforms and sample preparation methods. Six previously published classifiers (with 3-6 subtypes each) were applied to a dataset comprising more than 5,000 samples, consisting primarily of stage II-III colon cancer. A central independent team (Sage Bionetworks) provided analysis concordance of subtype calls, and clinical/molecular/pathology annotation with methodology derived through consensus.


Despite variation in the datasets and classification methodology in the separate published subtypes, there were 4 consensus molecular subtypes (CMS1-4) clearly defined by the concordance analysis, as indicated by a high degree of interconnectivity among the subtype calls from the 6 methods. CMS1 (∼15%) was enriched for MSI, right-side tumors, older age, females, hypermutation, BRAF mut and immune pathway activation. Both CMS2 and 3 displayed epithelial markers, canonical WNT and MYC pathway activation. Highly proliferative CMS2 tumors (∼40%) were CIN, MSS, left-sided, TP53 mut with EGFR upregulation and had better survival rates. CMS3 tumors (∼10%) had low CIN status, higher proportion of KRAS/PIK3CA mutations and IGFBP2 overexpression. CMS4 tumors (∼20%) were characterized by mesenchymal/TGF-beta signaling upregulation, younger age at diagnosis, NOTCH3/VEGFR2 overexpression and worse survival outcomes. The remaining samples (∼15%) did not have a consensus assignment, which may be attributed to additional minor subtypes. Further refinement is anticipated.


This consortium represents the first effort to generate consensus cancer subtypes and a major step forward for precision medicine in CRC. By comparing multiple classifiers we were able to identify four biologically distinct subtypes of CRC with unique clinical and molecular markers, a remarkable overlap considering different methodologies and datasets used by individual groups. Our efforts on improving this consensus classification system and correlating subtypes with benefit from adjuvant chemotherapies/response to targeted agents will be presented.