Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2017 Nov 30;7(1):16618.
doi: 10.1038/s41598-017-16747-x.

CMScaller: An R Package for Consensus Molecular Subtyping of Colorectal Cancer Pre-Clinical Models

Free PMC article

CMScaller: An R Package for Consensus Molecular Subtyping of Colorectal Cancer Pre-Clinical Models

Peter W Eide et al. Sci Rep. .
Free PMC article


Colorectal cancers (CRCs) can be divided into four gene expression-based biologically distinct consensus molecular subtypes (CMS). This classification provides a potential framework for stratified treatment, but to identify novel CMS-drug associations, translation of the subtypes to pre-clinical models is essential. The currently available classifier is dependent on gene expression signals from the immune and stromal compartments of tumors and fails to identify the poor-prognostic CMS4-mesenchymal group in immortalized cell lines, patient-derived organoids and xenografts. To address this, we present a novel CMS classifier based on a filtered set of cancer cell-intrinsic, subtype-enriched gene expression markers. This new classifier, referred to as CMScaller, recapitulated the subtypes in both in vitro and in vivo models (551 in total). Importantly, by analyzing public drug response data from patient-derived xenografts and cell lines, we show that the subtypes are predictive of response to standard CRC drugs. CMScaller is available as an R package.

Conflict of interest statement

The authors declare that they have no competing interests.


Figure 1
Figure 1
CMS4-mesenchymal markers in primary cancer are partially lost upon xenografting. (a) Single-sample gene expression enrichment scores for gene sets of TGFβ responses versus MSS-like characteristics identify two PDX models with particularly strong CMS4 characteristics (encircled). Samples are colored according to CMS predictions based on the original CMSclassifier. (b) Differential gene expression between pCRCs and PDX models, plotted against mean overall expression, indicates that genes included as markers in the original CMSclassifier and highly expressed in CMS4 primary tumors (green) show reduced expression in PDXs. The top-5 differentially expressed genes are labeled. Units are log2(signal). (c) Volcano plot of differential expression analysis of CMS4 versus CMS1/2/3 primary CRCs. Highlighted in purple are the genes differentially expressed between pCRCs and PDXs (absolute LFC > 2). The five genes with the largest absolute difference between CMS4 and CMS1/2/3 are labeled. CMS: consensus molecular subtype; LFC: log2fold-change; MSI/MSS: micro-satellite instable/stable; PDX: patient-derived xenograft; pCRC: primary colorectal cancer.
Figure 2
Figure 2
CMScaller feature selection and performance in pCRC. (a) Schematic illustration of gene filtering-approach. Three different datasets (top) were used to identify robust cancer cell-enriched subtype markers (only genes represented in all three datasets were considered). (b) CMScaller performance on the test set of pCRCs from TCGA (n = 143). Heatmap shows the relative expression levels of subtype marker genes (vertical bar) with classifications indicated below (horizontal bar, white indicate prediction confidence p-values). (c) Plot shows results from principal component analysis (expression data batch-adjusted for sequencing-platform). Disagreements between CMScaller and CMSclassifier are indicated with diamonds. (d) Heatmap shows results from mRNA gene set analysis, confirming enrichment of known characteristics in each CMS group (details of the gene sets are given in Supplementary Table 3). Red and blue indicate relative up- and down-regulation, respectively, and color saturation represents increasing statistical significance, as indicated. dn: down-regulation; n/p: number of samples/features; NA: not assigned; PC: principal component; PDX: patient-derived xenograft; pCRC: primary colorectal cancer; RNA-seq: RNA-sequencing; TCGA: The Cancer Genome Atlas; up: up-regulation.
Figure 3
Figure 3
CMS classified PDXs recapitulate relative drug responses observed in patients. (a) Heatmap visualization of mRNA gene set analysis showing selected CMS-informative signatures for comparisons of organoids (n = 48) classified with CMScaller. Red and blue indicate relative up- and down-regulation, respectively, and color saturation reflects statistical significance. (b) Same as a, but for a merged dataset of PDX models (n = 128). (c) Plot showing the number of mutations against the number of copy number aberrations (number of genes affected) per sample. Samples are colored according to CMScaller predictions. Horizontal axis is log-transformed for clarity. (d) Barplot showing distribution of KRAS and BRAF mutations per subtype. (e) Response to cetuximab as measured by change in tumor volume (best average response) among PDXs stratified by CMS subtype. (f) Response to 5-fluorouracil as measured by change in tumor volume (best average response) among PDXs stratified by CMS subtype. CMS: consensus molecular subtype; mut: mutation; PDX: patient-derived xenograft; wt: wild type

Similar articles

See all similar articles

Cited by 18 articles

See all "Cited by" articles


    1. Ferlay J, et al. Cancer incidence and mortality worldwide: Sources, methods and major patterns in GLOBOCAN 2012. Int J Cancer. 2015;136:E359–E386. doi: 10.1002/ijc.29210. - DOI - PubMed
    1. Budinska E, et al. Gene expression patterns unveil a new level of molecular heterogeneity in colorectal cancer. J Pathol. 2013;231:63–76. doi: 10.1002/path.4212. - DOI - PMC - PubMed
    1. Melo FDSE, et al. Poor-prognosis colon cancer is defined by a molecularly distinct subtype and develops from serrated precursor lesions. Nat Med. 2013;19:614–618. doi: 10.1038/nm.3174. - DOI - PubMed
    1. Marisa L, et al. Gene Expression Classification of Colon Cancer into Molecular Subtypes: Characterization, Validation, and Prognostic Value. PLoS Med. 2013;10:e1001453. doi: 10.1371/journal.pmed.1001453. - DOI - PMC - PubMed
    1. Sadanandam A, et al. A colorectal cancer classification system that associates cellular phenotype and responses to therapy. Nat Med. 2013;19:619–625. doi: 10.1038/nm.3175. - DOI - PMC - PubMed

Publication types

MeSH terms