You would likely benefit from reading my 2 reviews on this area.
One way to target Cancer Stem Cells (CSC) is to counteract their own biology. For example, all CSC have increased expression of BMI1 an oncogene that mediates ubiquitylation of histones which silences genes. In the CSC this process has become aberrant in almost every type of cancer, i.e., the over expression of BMI1 leads to silencing of may genes that regulate cell proliferation, apoptosis, cell cycle arrest etc..
However, BMI1 is also involved in DNA repair, being one of the very first factors in the DDR (DNA damage response) to modify the area around DNA double strand breaks.
We find that the higher the expression of BMI1 the better CSC can repair their DNA, including the DNA damage induced by therapies whose aim is to eradicate the tumor. Therapy can kill most of the tumor but if not all CSC are eradicated the cancer will eventually relapse.
Unfortunately CSC are very good at repairing this damage making current treatments that don't target CSC biology obsolete.
Unfortunately, the field of CSC identification has been plagued with circular reasoning. The recommended cell surface markers and/or their combinations merely enrich the selected populations for CSCs, but most of the selected cells are not themselves CSCs. Even the claim that CSCs express certain genes is fraught with the problem of how one defines the CSCs in the first place. For normal hematological stem cells, we have only the functional stem cell definition of Till & McCulloch: (1) Replication (make more stem cells), (2) Proliferation (have unlimited cell proliferation ability, i.e., immortality), (3) Differentiation (be able to differentiate into the functional cells of the tissue, i.e., form red cells, the 3 types of white cells, and megakaryocytes). For CSCs a 4th requirement is to be able to form tumors in vivo. Unfortunately, the recent literature does not distinguish between populations of cells that have amongst them cells that meet the definition's requirements and that ALL the selected cells meet these requirements. The best example of this are the data of Kuperwasser's lab, in which they selected CD26-CD44+CD326+(CD326 = ESA = Epcam) cells from patient tumors (or tumor cell lines) and found these selected cells to be 2% of the initial population. Using a limiting dilution assay into the "humanized" (human fibroblasts injected into the mouse breast fat pad 10 days earlier) NOD-SCID mouse breast fat pads, they found that 10,000 cells and 1,000 cells per injection yielded 10/10 (100%) mice getting a breast cancer, but injecting 100 cells yielded 6/10 (60%) and 5/8 (62.5%) of the mice getting the breast cancer. They concluded that these must therefore be the breast CSCs. However, injecting cells from a single cell suspension is subject to the Poisson distribution, in which an average of a single CSC per injection yields: 37% get no cell and 63% get one or more cells and therefore the cancer. So their data very nicely demonstrate that only 1 of the 100 injected cells was in fact a CSC, so 1% of 2% = 2/10,000 = 0.02% of the original tumors cells were CSCs. This is in excellent agrement with the results my lab obtained , 0.04%, using our patented Hybrid Spheroid Assay, which is, like the Till & McCulloch spleen colony assay, a functional assay. The 1957 (or 1958) paper of Hewitt and Wilson, showing the first in vivo radiation survival curve for an isogenic mouse leukemia, also demonstrates the Poisson distribution succesfully applied to such data. Please see my recent papers and abstracts listed in my ResearchGate profile.