The concentration of growth factors required for proliferation of cancer stem-like cells depends on the cell context and the kind of original cell line/ tumor types. There is a plasticity, however, in which some of the cancer stem cells undergo pseudo-differentiation into non-CSCs, so that you have to check the CSC marker expression or perform FACS to enrich CSCs during the course of in vitro culture.   

In the following article you will probably find an answer to your problem.

Yang B, Lu Y, Zhang A. Zhou A, Zhang L et al. Doxicycine induces apoptosis and inhibits proliferation and invasion of human cervical carcinoma stem cells. Plos One 2015;10(7):e0134201 http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0129138

A part of your problem is that the so-called biomarkers (fluorescent antibodies against specific cell surface proteins that have nothing to do with being stem cells) identify a fraction of the tissue or tumor population that is much too large to be the stem cells. Al Hadj et al used CD24- and CD44+ to identify a fraction of breast cancer cells that could initiate tumors in immuno-suppressed xenograft hosts. Supposedly the cancer stem cells (CSCs). Although the non-selected fraction did not initiate tumors, the selected fraction represented 11-35% of the tumor cells. Clearly too high a fraction to be the breast cancer stem cells. Kuperwasser's lab added ESA+ (CD326+) and reduced this fraction to 2% of the tumor. However, in limiting dilution assays they found that injections of 100 cell directly into the previously "humanized" breast fat pad in NOD/SCID mice yielded tumors in only 60-62.5% of the mice. Based on the Poisson distribution and Hewitt & Wilson's 1958 paper (first in vivo cell survival curve for ionizing radiations)  If a single cell (injected directly into the optimal site for its growth) can cause the tumor and an average of one cell is injected, 37% will get no cell and hence no cancer, while 63% will get one or more cells and get the cancer. So 60-62.5% getting the cancer indicates that only 1 out of the 100 injected cells was the CSC and 1% of 2% is 2x10^-4. This agrees with our hybrid spheroid assay (which provides a CSC niche) results of 4.5x10^-4 as the CSC fraction in breast cancer samples directly from patients. Read also Kern & Shibatta (Cancer Research 2007) and RP Hill (ibid. 2006) to see why these biomarker generated numbers (as opposed to functional assays) are internally inconsistent and hence do not serve to establish the cancer stem cell hypothesis. 

Also, as your CSCs grow, some progeny differentiate, so the CSC fraction eventually returns to a steady state with a much lower CSC fraction than at the start for those that had a CSC and therefore grew.

One can expand CSCs in non-adherent condition with the required stem cell specific media and growth factors. However, the markers those were used to isolate them as cancer stem cells should be tested at different time intervals to check for loss due to differentiation during the expansion. It is very similar to performing a serial sphere assay. Addition of serum is not a good idea as serum might enhance their differentiation process.

In agreement with 

You should check if your media contains antibiotics. As I stated above, tetracyclines added to culture media will impede stem cell growth.