NEW YORK (GenomeWeb News) – Colorectal cancer can be divvied up into six subtypes based on its transcriptome, and those subtypes are further related to patient prognosis, investigators led by Valérie Boige from Université Paris Descartes reported in PLOS Medicine yesterday.
Using whole-genome and whole-transcriptome arrays, the researchers screened tumor samples from 750 patients. The six molecular subtypes they uncovered were also linked to certain tumor characteristics and deactivated signaling pathways.
"The biological relevance of these subtypes is illustrated by significant differences in prognosis," Boige and her colleagues wrote. "This analysis provides possibilities for improving prognostic models and therapeutic strategies."
To develop their molecular characterization signature, the researchers turned to 750 tumor samples from people with stage I through stage IV colorectal cancer from the Cartes d'Identité des Tumeurs national program in France. Of those, 566 samples met the researchers' RNA quality standards.
The discovery set of 443 samples was analyzed for gene expression levels using the Affymetrix U133 Plus 2.0 chip. Those samples were also studied for common gene mutations, such as KRAS and BRAF mutations, as well as for their mismatch repair and CpG island methylator phenotype statuses. For 19 samples, matched normal tissue was also examined.
Through hierarchical clustering of the gene expression profiles of the discovery set, Boige and her colleagues homed in on six clusters, termed C1 through C6. Four of those clusters were distinct while two, C1 and C5, exhibited some overlap in classification. The subtypes, they added, did not appear to be associated with pathological stage.
The researchers developed a classifier based on 57 genes that they then validated using the other 123 CIT samples as well as data from seven public databases comprising some 900 other colorectal cancer samples.
The subtypes, the researchers noted, appeared to correlate with biological characteristics of the tumors. Colorectal cancers belonging to the subtypes C1, C5, and C6, for instance, more commonly showed signs of chromosomal instability and harbored mutant TP53, but did not have a CpG island methylator phenotype. Meanwhile, the C2, C3, and C4 subtypes were more likely to be have the CpG island methylator phenotype.
Certain subtypes also appeared to have different upregulated signaling pathways. For example, C2 subtype tumors, which tended to have deficient mismatch repair, had upregulated immune system and cell growth pathways, while the C4 and C6 subtypes had downregulated cell growth pathways, but upregulated epithelial-mesenchymal transition/motility pathways.
Restricting their analysis to stage III and stage IV tumors, the researchers noted that patients with C4 and C6 subtypes tended to have worse outcomes.
Then by combining their classification into a binary high-risk or low-risk category — with high risk including the C4 and C6 subtypes and low risk including the others — the researchers found an even stronger association with relapse-free survival.
"Our study, like others, supports the idea that the unsupervised analysis of transcripts in primary tumors yields information of prognostic value," the researchers wrote.
The C4 subtype, investigators pointed out, includes a number of deregulated genes, especially ones involved in stem cell regulation that have been included in a previous signature of poor prognosis. "These genes may therefore be markers of the aggressiveness of [colon cancer] cells and may constitute potential therapeutic targets," they added.
Boige and her colleagues also examined how their classifier worked in comparison to Genomic Health's Oncotype Dx recurrence score and the 17-gene signature included in Agendia's ColoPrint. Oncotype Dx, the researchers found, also had prognostic value in their dataset as it classified 97 percent of their C4 subtype samples as high risk, but it was not prognostic for all subtypes. The ColoPrint signature, though, did not identify risk in their dataset.
"Our classification added prognostic information that remained significant in the multivariate analysis adjusted for [tumor node metastasis] and Oncotype DX score," Boige and her team wrote. "This suggests that the 'one size fits all' prognostic signature approach can be difficult to apply because of the heterogeneity of [colon cancer]. This may explain, in part, the poor concordance of [gene expression profile] prognostic signatures in [colon cancer]."