Colon cancer is one of the best-understood neoplasms from a genetic perspective, yet it remains the second most common cause of cancer-related death, indicating that some of its cancer cells are not eradicated by current therapies. What has yet to be established is whether every colon cancer cell possesses the potential to initiate and sustain tumour growth, or whether the tumour is hierarchically organized so that only a subset of cells--cancer stem cells--possess such potential. Here we use renal capsule transplantation in immunodeficient NOD/SCID mice to identify a human colon cancer-initiating cell (CC-IC). Purification experiments established that all CC-ICs were CD133+; the CD133- cells that comprised the majority of the tumour were unable to initiate tumour growth. We calculated by limiting dilution analysis that there was one CC-IC in 5.7 x 10(4) unfractionated tumour cells, whereas there was one CC-IC in 262 CD133+ cells, representing >200-fold enrichment. CC-ICs within the CD133+ population were able to maintain themselves as well as differentiate and re-establish tumour heterogeneity upon serial transplantation. The identification of colon cancer stem cells that are distinct from the bulk tumour cells provides strong support for the hierarchical organization of human colon cancer, and their existence suggests that for therapeutic strategies to be effective, they must target the cancer stem cells.

Summary DNA demethylating agents have shown clinical anti-tumor efficacy via an unknown mechanism of action. Using a combination of experimental and bioinformatics analysis in colorectal cancer cells, we demonstrate that low-dose 5-AZA-CdR targets colorectal cancer initiating cells (CICs) by inducing viral mimicry. This is associated with induction of dsRNAs derived at least in part from endogenous retroviral elements, activation of the MDA5/MAVS RNA recognition pathway and downstream activation of IRF7. Indeed, disruption of virus recognition pathways, by individually knocking down MDA5, MAVS or IRF7, inhibits the ability of 5-AZA-CdR to target colorectal CICs and significantly decreases 5-AZA-CdR long-term growth effects. Moreover, transfection of dsRNA into CICs can mimic the effects of 5-AZA-CdR. Together; our results represent a major shift in understanding the antitumor mechanisms of DNA-demethylating agents and highlight the MDA5/MAVS/IRF7 pathway as a potentially druggable target against CICs.