The cancer stem cell (CSC) concept has been succinctly summarized by Hans Clevers ([48
], p. 313) in an elegant and critical review:
Central to the stem cell (CSC) concept is the observation that not all cells in tumours are equal. The CSC concept postulates that, similar to the growth of normal proliferative tissues such as bone marrow, skin or intestinal epithelium, the growth of tumours is fuelled by limited numbers of dedicated stem cells that are capable of self-renewal. The bulk of a tumour consists of rapidly proliferating cells as well as postmitotic, differentiated cells. As neither of these latter two classes of cells has the capacity to self-renew, the contribution of these non-CSC tumour cells to the long-term sustenance of the tumour is negligible.
It is not surprising that this CSC concept has fired the imagination of investigators working on drug resistance [49
]. If tumours are driven by CSCs, the stem cells are the cells that need to be killed to eradicate the tumour. Incomplete eradication of cancer must leave some of the CSCs untouched and these are responsible for tumour relapse. Residual disease may therefore consist of stem cells equipped with specialized drug resistance mechanisms. It follows that chemotherapy aiming at cure should therefore target the CSCs rather than the bulk of the tumour cells [50
]. To eliminate the weeds, you have to tear out the roots. If there were drugs that kill rare CSCs without touching the bulk of the tumour cells, they might even have been missed in standard chemotherapy trials.
Since the CSC concept was revived by Dick and colleagues for acute myeloid leukaemia in 1995 and extended to solid tumours in 2003, the concept has become the centre of heated controversies [52
], as summarized by Clevers [48
]. Some investigators think that the CSC concept should guide the search for new cancer therapies. In contrast, others believe that CSCs of solid human tumours are an artefact of the methods used to detect tumour-initiating cells (TICs). This requires dissociation of the tumour into single cells, fluorescence-activated cell sorting (FACS) and seeding in artificial niches in immunocompromised mice. This assay may select more for the ability of cells to survive extreme insults than for stemness. These sceptical investigators stress the flexibility of the tumour cell population, which allows more differentiated cells to dedifferentiate into CSCs. Obviously, if the CSC phenotype is not a stable trait, the development of drugs specifically targeting CSCs becomes less attractive [48
]. If the phenotypic heterogeneity in tumours is reversible, as Morrison and co-workers have shown for melanomas [54
], it becomes irrelevant to distinguish CSCs from the bulk population of cancer cells when considering targeted therapy [48
Although the CSC concept has lost some of its lustre, it is still often invoked to explain residual disease. I shall therefore briefly summarize the evidence that CSCs have specialized defences against chemotherapy that could explain drug-resistant residual disease:
- —Drug transporters . It is often stated that stem cells, including CSCs, are rich in transporters able to extrude drugs from cells. This idea seems to have its origin in the haematopoietic stem cells, which indeed contain high concentrations of the two most versatile drug pumps, P-gp (MDR1, ABCB1) and BCRP (ABCG2). Initially, ABCG2 was even thought to be a general marker of stem cells, but more recent evidence has shown this to be incorrect. For instance, the normal mammary gland stem cell lacks ABCG2 [55,56]. Likewise, gut stem cells lack P-gp . For other transporters present in CSCs, such as the MRPs (ABCCs), a generalized role in drug resistance is improbable. The MRP most generally present in cells, MRP1 (ABCC1), has never been conclusively linked to resistance in either mouse model tumours or human clinical samples . Even if high levels of a MDR-type drug transporter are found in some CSCs, these can only explain resistance to substrates of the transporter, not to the many prominent drugs not touched by MDR transporters, as also pointed out by Dean .
- —Resistance to DNA damage. The most unambiguous results have been obtained with ionizing radiation, which is not complicated by target alterations (e.g. topoisomerase down regulation) or drug uptake problems encountered by DNA-interacting drugs. The CD133-expressing glioma cells with CSC properties are more resistant to ionizing radiation than the CD133-negative tumour cells  and the same holds for the putative CSC fraction of human breast cancer . Why is not known. It could be due to more efficient repair of DNA strand breaks, or to more CSCs being quiescent-like, allowing more time for DNA repair before cells enter S-phase and find their DNA too damaged to survive DNA replication.
- —Quiescence. A low rate of multiplication is a hallmark of the somatic stem cells of normal tissues, the majority of colon epithelial stem cells being the exception [48,60]. Whether this makes stem cells less vulnerable to chemotherapy is not self-evident. Blanplain and co-workers  have claimed that being in G0/G1 when your DNA gets hit can actually be unhealthy for a stem cell, as duplex breaks in DNA cannot be repaired by the error-free homology-directed system only available during and after DNA replication. Instead, the error-prone system of non-homologous end joining has to be used to fix duplex DNA breaks. Nevertheless, the generally accepted hypothesis is that quiescence of stem cells protects against cytotoxic therapy [48,62–64]. The presence of quiescent cells with CSC properties has been demonstrated in several tumour systems, using retention of DNA label or lipophilic dye. Whether these are the cells in the tumour that result in residual disease and whether their quiescence is responsible for their survival remains to be seen. The most convincing experiments have been published by Andreas Trumpp and co-workers [36,37], who showed that leukaemia stem cells could be targeted by breaking their dormancy.
- —Epithelial to mesenchymal transition (EMT). There is no doubt that EMT provides a formidable version of pan-resistance  and I shall return to this below. The question here is whether residual disease is due to EMT. This question is not easily answered, as EMT can be a transient state that could be easily missed. Moreover, residual disease is usually poorly accessible to detailed analysis, and often the analysis does not include an evaluation of EMT. In the few model systems in which this was verified, no EMT was found  and EMT therefore does not appear to provide a general explanation for residual disease.
Are CSCs responsible for the therapy-resistant fraction resulting in residual disease? This is obviously the key question. There are now several tumour systems in which CSC-like cells are enriched in tumour remnants after therapy. These include gliomas, breast cancer, colon cancer and a sophisticated CML mouse model [48
]. In our laboratory, Pajic et al.
] have studied the issue in a conditional mouse model of human triple-negative breast cancer. In the mouse, the somatic stem cells are well defined in normal mammary glands. Cells with the same surface markers proved to be highly enriched in the tumour-initiating fraction isolated from the tumour. However, the few cells in this tumour repeatedly surviving cisplatin therapy were not enriched in these TICs. This raises the question whether residual disease in other tumour systems is really due to putative CSCs or a consequence of other properties of CSCs, such as quiescence, allowing them to survive drug treatment.
A major effort is under way to find drugs that preferentially target stem cells [67
]. As pointed out by Clevers [48
], the initial ideas driving this effort were too simple. Tumours have no roots that one can specifically tear out, dooming the plant. There is little doubt that some of the more differentiated tumour cells can dedifferentiate to replace the killed CSCs. If CSC-targeted therapy is going to make a contribution, it is only in conjunction with therapy targeting the bulk of the tumour.
Zhou et al.
] and Frank et al.
] have written detailed and optimistic reviews of the new therapeutic opportunities provided by the CSC hypothesis. The drugs under development mainly attempt to target signalling pathways involved in the regulation of self-renewal of normal somatic stem cells, such as the Wnt, the Sonic Hedgehog and the Notch pathways. The drugs should either preferentially block stem cell (and CSC) renewal or drive the stem cells into differentiation, closing down the tumour supply line. As the authors point out, a major problem is specificity, as with all tumour chemotherapy. Indeed, the only small molecule that targets a pathway involving stem cell self-renewal and that has managed to reach a phase II trial at the time the review of Zhou et al.
] was written is a SMO (Sonic Hedgehog) antagonist. This was developed, however, for patients with basal cell carcinoma, most of whom have mutations in Hedgehog pathway components [69
Other approaches attempt to target surface molecules preferentially present on CSCs [68
]. Whether the (limited) effectiveness of these antibodies against metastatic cancer is due to their targeting of CSCs remains to be seen.