Side Population is Not Necessary or Sufficient for a Cancer Stem Cell Phenotype in Glioblastoma Multiforme§


  • Disclosure of potential conflicts of interest is found at the end of this article.

  • K.W.R.B.: conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing; M.K.H.: collection and/or assembly of data, provision of study material or patients; K.J.F., K.M.P., C.G., and R.J.M.: collection and/or assembly of data, data analysis and interpretation; I.F.H.: conception and design, data analysis and interpretation, manuscript editing; M.J.M.: conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing, final approval of manuscript.

  • §

    First published online in STEM CELLSEXPRESS January 14, 2011.


There is strong evidence for the existence of cancer stem cells (CSCs) in the aggressive brain tumor glioblastoma multiforme (GBM). These cells have stem-like self-renewal activity and increased tumor initiation capacity and are believed to be responsible for recurrence due to their resistance to therapy. Several techniques have been used to enrich for CSC, including growth in serum-free defined media to induce sphere formation, and isolation of a stem-like cell using exclusion of the fluorescent dye Hoechst 33342, the side population (SP). We show that sphere formation in GBM cell lines and primary GBM cells enriches for a CSC-like phenotype of increased self-renewal gene expression in vitro and increased tumor initiation in vivo. However, the SP was absent from all sphere cultures. Direct isolation of the SP from the GBM lines did not enrich for stem-like activity in vitro, and tumor-initiating activity was lower in sorted SP compared with non-SP and parental cells. Transient exposure to doxorubicin enhanced both CSC and SP frequency. However, doxorubicin treatment altered the cytometric profile and obscured the SP demonstrating the difficulty of identifying SP in cells under stress. Doxorubicin-exposed cells showed a transient increase in SP, but the doxorubicin-SP cells were still not enriched for a stem-like self-renewal phenotype. These data demonstrate that the GBM SP does not necessarily contribute to self-renewal or tumor initiation, key properties of a CSC, and we advise against using SP to enumerate or isolate CSC. STEM CELLS 2011;29:452–461


It is well established that the aggressive, incurable brain tumor glioblastoma multiforme (GBM) contains a subset of cells that are tumorigenic, resistant to therapy and have a self-renewing stem cell-like phenotype [1, 2]. These cancer stem cells (CSCs) have been identified in GBM, and many other tumors, using a variety of methods including growth in serum-free media to allow sphere formation, and expression of surface markers such as CD133, CD15 (SSEA-1/LwX), CD326 (EpCAM), CD49f (ITG6A/VLA-6), and CD44 [3–6]. Cells selected using these techniques have higher expression of the embryonic stem cell self-renewal and pluripotency factors SOX2 and OCT4. They also initiate tumor formation with fewer cells than the parental material, indicating enrichment for a tumor-initiating cell, or CSC. Sphere formation is particularly a good model for GBM CSC, as in addition to the stem-like gene expression and increased tumor initiation, spheres have been shown to accurately recapitulate the diffuse, infiltrative histopathology of human GBM better than parental serum-exposed cells [7, 8].

Another assay often used to isolate or enumerate stem-like cells in cancer is the side population (SP) assay. Originally used to identify hematopoietic stem cells [9], the SP is a subset of cells with differential efflux activity to the main cell population. This is usually measured by efflux of the fluorescent DNA binding dye Hoechst 33342 (H33342), with the main, non-SP cells separating according to DNA content, and hence cell cycle, using the differential emission spectra of H33342 bound to chromatin. The pumps responsible for H33342 efflux, including the MDR1/P-glycoprotein family of transporters, also transport other compounds out of the cell, including chemotoxic drugs. SP has now been identified in many normal and tumor tissues, and the assay has been proposed as a functional marker to identify and isolate CSC [10–12]. This is based firstly on the association of SP with a normal stem cell phenotype and secondly with the drug resistance associated with CSC. Upregulation of drug efflux pumps is a well-described mechanism for resistance to chemotherapeutic agents and has been proposed as a mechanism for the enhanced resistance of CSC. SP cells can be physically separated from the non-SP for phenotypic analysis by flow-assisted cell sorting (FACS) [13], making it an attractive technique to isolate putative CSC.

In addition to many cell lines, SP have been reported in cells isolated from spontaneous murine glioma [14, 15]. These tumors spontaneously arise in transgenic animals with oncogenic mutations, namely, ErbB or PDGF overexpression on a p53−/− or PTEN−/− background. The tumors have histological features of GBM, contain a SP and form spheres when cultured in vitro. Further examination of the relationship between the murine glioma SP, tumor spheres, and the CSC phenotype has suggested that SP is a CSC marker for GBM [15–18].

To determine whether SP can be used to quantify CSC in human GBM, we took a panel of immortalized GBM cell lines (LN18, U87MG) and tumor-derived primary cells (08/04, 09/06) and generated spheres with a CSC phenotype. We identified a SP at varying frequency in parental GBM primary cells and cell lines but did not find a SP in any spheres. Purified SP cells neither have a self-renewing stem-like phenotype nor did they initiate tumors more efficiently. Exposure to a chemotoxic drug upregulated both self-renewal gene expression and SP, but drug exposure did not skew the SP toward a cancer stem-like phenotype.



Unless otherwise noted, tissue culture plastic ware was from Nunc (ThermoFisher Scientific, Auckland, NZ,, and cell culture reagents were from Invitrogen (Auckland, NZ, NeuroCult stem cell media, heparin, EGF, and bFGF were from Stem Cell Technologies (Victoria, Australia,

Cell Lines

The human GBM lines LN18 and U87-MG were obtained from American Type Culture Collection. The murine glioma line GL261 was obtained from DCTD Tumor Repository, National Cancer Institute at Frederick (Frederick, MD). All cells were maintained at 37°C and 5% CO2. LN18 cells were cultured in RPMI-1640 with 10% serum, 1% penicillin/streptomycin, and 1% glutamax. U87-MG cells were cultured in minimal essential media (MEM) supplemented with 10% serum, 1% penicillin/streptomycin, 1% glutamax. GL261 were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 20% serum, 1% penstrep, and 1% glutamax. Regular testing of cultures by polymerase chain reaction (PCR; e-Myco, Intron BioTechnology, Korea) showed they were mycoplasma free.

Tumor-Derived Primary Cells

Primary cells were isolated and cultured from GBM material obtained from patients undergoing debulking surgery, under informed consent approved by the Central Regional Ethics Committee. Tumors were dissected into 1-mm fragments and mechanically dissociated into a single-cell suspension through consecutive 100-μm, 70-μm, and 40-μm cell strainers. Glial cells were counted and plated out at 1 × 106 glial cells per 10-cm plate, in RPMI-1640 with 10% fetal bovine serum (FBS), 1% penicillin/streptomycin, and 1% glutamax at 37°C 5% CO2. Media were changed 2 days postplating to remove debris. Regular testing of cultures by PCR (e-Myco, Intron Biotechnology, Seoul, Korea, showed they were mycoplasma free.

Sphere Cultures

Dissociated tumor cells were purified over a Percoll gradient to remove red blood cells and debris, then plated out in neural stem cell media at 1 × 105 cells per milliliter. Spheres were cultured in either DMEM/F12 with B27 supplement, heparin 2 μg/ml, recombinant human epidermal growth factor (hEGF) 20 ng/ml, and basic fibroblast growth factor (bFGF) 10 ng/ml (U87-MG and GL261) or in NeuroCult stem cell media with heparin 2 μg/ml, hEGF 20 ng/ml, bFGF 10 ng/ml, following manufacturers instructions (LN18 and primary cultures). When spheres reached 150–200 μM, they were broken up by trituration with a 26-gauge needle 10 times and passed through a 40-μM filter. Spheres were used at passage six and above for LN18, U87-MG, and primary lines. GL261 spheres were used between passage two and four. Regular testing of cultures by PCR (e-Myco, Intron BioTechnology, Korea) showed they were mycoplasma free.

SP Assay

SP staining was performed as described [9]. Cells at 70%–80% confluency were trypsinized and washed twice in Hanks Buffered Salt Solution (HBSS)/2% FBS. Cells were resuspended at 106 cells per milliliter in 37°C HBSS/2% FBS, then incubated at 37°C with 5 μg/ml Hoechst 33342 and 15 mM sodium azide. For the inhibitor control, 50 mM 2 deoxy-glucose (2DOG) was added prior to H33342. Human cells were incubated for 120 minutes and murine cells were incubated for 90 minutes, with mixing every 30 minutes. Staining was halted by rinsing in ice-cold HBSS/2% FBS. Propidium iodide (PI) was added at 4 μM for viability assessment. Cells were resuspended at 2–4 × 106 cells per milliliter in HBSS/2% FBS for sorting on the BD FACS Vantage DiVa (BD Biosciences, San Jose, CA, or at 1 × 106 cells per milliliter for analysis on the LSRII (BD Biosciences, San Jose CA). Cells were excited with a UV laser (150 mW on the FACS Vantage DiVa and 20mW in the LSR II). The emission was detected through 250/50 nm (“Hoechst Blue”) and 620 LP (“Hoechst Red”) filters with signal separated by a 610-nm short pass dichroic mirror. All data were collected in linear mode and analyzed using FlowJo (TreeStar, Ashland, OR, Cells were displayed on dot plots gated on live cells, PI negative, and viewed in a Hoechst Blue versus Hoechst Red dot plot to visualize the SP. For sorting experiments, SP cells were sorted at 20 psi sheath pressure, with a 100-μm nozzle and approximately 37,500 Hz drop drive frequency, at a rate of approximately 2,000 cells per second. The SP sort gate was determined by the use of the 2DOG inhibitor-treated cells.

Doxorubicin Exposure

Doxorubicin was titrated to determine the concentration where 80% of the cells die, that is, 09/06 2 μM; 08/04 and LN18 5 μM and GL261 0.1 μM. DMSO was used at the appropriate concentration for a vehicle control. Cells at 60%–70% confluency, exponential growth phase, were exposed to doxorubicin or DMSO (vehicle control) for 16 hours. Cells were then trypsinized and replated into standard tissue culture media and clean flasks, at 1:2 dilution for the doxorubicin cells and 1:5 dilution for the DMSO control. DMSO control cells were passed as required and RNA was collected at passage two to compare with the doxorubicin cells. RNA was collected from doxorubicin-treated cells when they started to divide, and had formed clusters of approximately eight cells.

Real-Time Reverse Transcription Polymerase Chain Reaction

Total RNA was extracted using the Mini RNA Isolation II kit (Zymo Research, Irvine, CA, for cell lines, and RNeasy (Qiagen, Valencia, CA, for primary cells. cDNA was synthesized using iScript (BioRad, Hercules, CA, according to the manufacturers directions. A total of 12.5 ng of cDNA was used for quantitative reverse transcription-polymerase chain reaction (Q-RT-PCR) of SOX2, OCT4, MSI1, NES, CD133, and 18s rRNA (QuantiTect primer assay and SYBR green PCR kit, Qiagen) on the ABI 7500 platform. Cycle threshold (Ct) was determined in the exponential phase of the amplification curve. Ct for SOX2, OCT4, MSI1, NES, and CD133 were normalized to 18s ribosomal RNA (ΔCt). Amplification efficiency of QuantiTect primer assays is equivalent (; so, ΔΔCt method was used to determine fold change from parental to sphere.

Protein Expression

Intracellular staining for SOX2 was performed using the specific anti-SOX2 monoclonal antibody 245610 (MAB2018, R&D Systems, Minneapolis, MN, or isotype control antibody conjugated to Alexa 647 (Invitrogen, Auckland, NZ), at 1:2,000 dilution. Cells were prepared using the CytoFix/Cytoperm kit (BD Biosciences), according to the manufacturers directions. Surface CD133 was detected using the AC133 antibody (Miltenyi Biotech, Germany, or isotype control conjugated to R-Phycoerythrin (PE) or allophycocyanin (APC) at a 1:1,000 dilution. Viable cells were gated using propidium iodide, and the median fluorescence intensity of isotype and specific antibody calculated (FlowJo) to determine relative MdFI for each protein.

Tumor Formation Assay

Breeding pairs of the inbred strain C57BL/6 were originally obtained from Jackson Laboratories, Bar Harbor, ME. Breeding pairs of the inbred strain NOD/scid were obtained from the Hercus-Taieri Research Unit, University of Otago, New Zealand. All mice were maintained at the Biomedical Research Unit of the Malaghan Institute of Medical Research. All experimental protocols were approved by the Victoria University of Wellington Animal Ethics Committee and performed according to institutional guidelines.

Mice were challenged with 5 × 104 tumor cells (GL261) or 1 × 106 tumor cells (U87-MG) injected subcutaneously into the flank, in PBS (GL261) or Geltrex (U87-MG). Tumor appearance was monitored and size obtained by measuring the short and long diameters then expressed as mean product of diameter ± SEM. Onset of tumor growth was the point at which a tumor was first palpable. Growth rates were determined by measuring tumor size over time. Five mice were used in each group, and experiments repeated between two and four times. For intracranial xenograft experiments, 5 × 104 tumor cells were implanted into the right striatum in a 2 μl volume of PBS (2.1-mm lateral to the bregma, at a depth of 3 mm) under general anesthesia using a Stoelting small animal stereotactic frame. Time to tumor progression was defined as the time to 10% weight loss or the appearance of neurological symptoms, including head tilt, reduced activity, and increased respiration. Onset of symptoms occurred at 90–100 days, and tumors had developed in all cases. After sacrifice by cervical dislocation, mouse brain tissue was formalin-fixed and embedded in paraffin. Four micron sections were cut, mounted onto glass slides, and stained with Harris's H&E stain. Sections were imaged at ×20 or ×40 magnification on the Olympus BX 40 microscope.


GBM Spheres Have Increased Self-Renewal and Tumor Initiation Activity

Primary cells derived from GBM tumors were established directly in neural stem cell media (08/04) or transferred to neural stem cell media at early passage (09/06). Cells from LN18 [19], U87MG [20] (human), and GL261 [21] (murine) were also grown in stem cell media and spheres established from each line (Fig. 1A). All spheres could be maintained for multiple (human >5, murine >4) passages, indicating that they all derived from, and contained cells with a stem-like self-renewal activity [22]. The human GBM cell lines T98G [23] and DBTRG-05MG [24] were examined but did not form spheres capable of multiple passage (data not shown). Self-renewal potential was measured by gene expression analysis. Expression of the stem-like factors SOX2, OCT4, and MSI1 and the neural stem/progenitor filament protein NES were quantified by real-time RT-PCR and compared between matched parental serum-derived and sphere cultures. Increased transcription of NES (twofold to fivefold), SOX2 (10–20 fold), and MSI1 (2-fold to 10-fold) was observed in all spheres (Fig. 1B), indicating increased self-renewal potential and the neural stem cell nature of spheres. In addition, SOX2 protein was measured in parental and sphere cells by flow cytometry and expressed as the median fluorescence relative to an isotype control. Spheres had increased SOX2 protein, confirming the stem-like phenotype of these cells (Fig. 1C).

Figure 1.

Characterization of glioblastoma multiforme (GBM) neurosphere model. (A): Primary human GBM cultures (08/04 and 09/06), human GBM lines (LN18 and U87-MG), and a murine cell line (GL261; parental, left hand panel) were used to establish long-term neurosphere cultures (spheres, right hand panel). (B): Expression of self-renewal genes in parental (white bars) and neurosphere culture (black bars) was measured by real-time reverse transcription polymerase chain reaction. Fold change in gene expression in spheres was expressed relative to the corresponding parental gene expression, which was normalized to 1. Data represents the average ± SD (error bar) of three independent experiments post sphere passage six (P6). (C): CD133 (left) and SOX2 (right) were measured by flow cytometry in parental (upper) and sphere (lower) cells. Gray, isotype control; black line, specific antibody. Median fluorescence intensity (MdFI) was calculated for both isotype and specific antibody, and specific MdFI expressed relative to the isotype MdFI. Relative MdFI shown for each plot. Data representative of multiple experiments, all experiments showed the same trend.

OCT4, which controls multilineage differentiation, did not increase to the same extent as the self-renewal genes in the human lines, no protein could be detected, and was not expressed at all in the murine GL261 cells, parental, or sphere. However, spheres were pluripotent and could be differentiated, demonstrated by induction of the glial cell marker GFAP and the oligodendrocyte marker MBP in 08/04 and LN18 spheres after growth in differentiation media (data not shown).

In parallel, we measured expression of the putative, and controversial, GBM CSC marker CD133 [25] in spheres. Increased CD133 transcript was observed only in primary spheres, not in LN18, U87MG, or GL261 spheres, suggesting loss of CD133 expression in long-term culture (Fig. 1B). Similarly, CD133 protein was significantly upregulated in primary spheres, while induced but barely detectable in LN18 and U87MG cells (Fig. 1C).

Interestingly, LN18 spheres closely resembled those derived from primary tumors, in both sphere behavior and self-renewal gene expression. In contrast, U87MG, like the GL261 murine cell line, had lower levels of stem-like gene induction (Fig. 1B) and shorter passage of long-term sphere cultures (data not shown).

To confirm the relationship between increased self-renewal in vitro and enhanced tumorigenicity, we assessed frequency of tumor initiation with U87MG and GL261, which are both competent for subcutaneous growth. One million U87MG spheres initiated subcutaneous tumors in NOD/scid mice in 11–17 days, faster and more uniformly than 106 parental cells (11–22 days; Fig. 2A). Similarly, GL261 spheres consistently initiated subcutaneous tumors in C57BL/6 mice at day 12, whereas the parental cells initiated tumors between day 21 and 27 (Fig. 2A). These data clearly demonstrate increased frequency of tumor-initiating cells in spheres.

Figure 2.

Glioblastoma multiforme neurospheres have a cancer stem cell phenotype. (A): Time in days for subcutaneous tumor formation from parental (circle) and neurosphere (square) cultures for U87MG (upper panel) and GL261 (lower panel). *, p value, calculated with log-rank (Mantel-Cox) test (Prism 5.0, GraphPad Software). (B): H&E staining of orthotopic xenograft of 08/04 parental (left)- and sphere (center)-derived tumors and original patient tumor (right), magnification ×20. (C): H&E staining of orthotopic GL261 parental (left)- and sphere (right)-derived tumor edge, magnification ×20.

The stem-like nature of the primary spheres was confirmed by orthotopic xenograft of the human primary cells 08/04. Fifty thousand parental and sphere cells were implanted intracranially into NOD/scid mice, and the resulting tumors examined by H&E staining (Fig. 2B). The parental cells formed diffuse infiltrative tumors with relatively round to oval nuclei, with occasional perinuclear halos. There was some evidence of organization, including satellitosis around neurons and a delicate arborizing vascular pattern. The overall pattern of growth resembled oligodendroglioma. In contrast, the sphere-derived tumors were diffuse, highly cellular, and demonstrated greater cellular pleomorphism than the parental tumors. Sphere-derived tumor cells contained angular nuclei and a haphazard organization that recapitulated the original patient tumor.

Orthotopic parental and sphere GL261 tumors (Fig. 2C) were highly pleomorphic with tumor giant cells and distinct vasculature. Although clearly more circumscribed than the human xenografts, three of five sphere-derived tumors contained focal areas where tumor cells had invaded the surrounding cortex, in contrast to zero of five parental tumors. This is in accordance with the invasive, aggressive nature of the glioma stem cell, and confirmed the CSC phenotype in the GL261 spheres.

SP Is Not Necessary for Self-Renewal In Vitro or Tumor Initiation In Vivo

The presence of a SP in a tumor or tumor cell line has been proposed as a direct measure of CSC numbers, and FACS of SP has been used to isolate CSCs from many tumor types, including murine glioma. We examined our cells for a SP using DNA staining with the fluorescent dye H33342. SP cells were detected in parental GBM cells, at frequencies ranging from 0.01%–4% (Fig. 3A). The SP was confirmed by use of 2DOG, an inhibitor of the ABC transporters involved in dye efflux. In striking contrast, a SP could not be found in any sphere culture (Fig. 3B), even with analysis of a large number of events, demonstrating loss of the high-efflux population in cells enriched for self-renewal and tumor initiation. Loss of SP was observed early during sphere formation, and was stable throughout on-going passage (data not shown).

Figure 3.

Glioblastoma multiforme (GBM) neurospheres do not have a side population (SP). (A): Hoechst 33342 staining of GBM primary cultures and cell lines (right hand panel). 2DOG was used to inhibit SP formation (left hand panel). Location of SP indicated by gate. Dot plots are representative of more than 3 separate experiments. Average percentage of SP over more than 3 experiments given. The number of events collected for each sample is shown in the lower right corner of each plot. (B): Hoechst 33342 staining of corresponding neurosphere cultures, with 2DOG used as in (A). *, less cells counted than in the corresponding “no 2DOG” sample. Abbreviation: 2DOG, 2-deoxy glucose.

Notably, 08/04 had the lowest SP frequency of all the parental cells (0.01%) but reproducibly initiated orthotopic tumors, further demonstrating that SP is not required for tumorigenicity. We also examined two primary human metastatic melanoma cell lines for a SP. Neither of these lines contained a SP, despite the ability of each line to be maintained (Supporting Information Fig. 1), and serially transplanted (data not shown) as a subcutaneous xenograft in NOD/scid mice. This confirmed that SP is not necessary for either tumor initiation or a CSC phenotype.

SP Is Not Sufficient for Self-Renewal In Vitro or Tumor Initiation In Vivo

We, and others, have observed that SP frequency changes in any condition that changes the rate of cell division, including alterations in confluency, culture media, and serum concentration (data not shown and [26]). Spheres are grown in a serum-free media that could potentially inhibit the SP phenotype and confound the analysis of SP in sphere cells. Therefore, we asked whether SP isolation would directly enrich for a self-renewing, tumor-initiating cell population similar to those in spheres. We attempted to sort the SP from the primary line 09/06, which was present at high (0.7%) frequency. However, optical limitations on sorting primary GBM cells, which are large, irregularly shaped cells with unusual light-scattering properties, made it impossible to sort a SP with sufficient purity for our analysis (>95%, data not shown). The LN18 cells had a similar SP frequency to 09/06, and LN18 self-renewal activity resembled the primary GBM lines, so SP were sorted from LN18 and from GL261 to examine tumor growth. Purity of the sorted SP was always >95% (average 98%), whereas the non-SP cell purity was >98% (average 99.5%).

Serial sphere formation was compared between SP and non-SP cells from each line, to determine if SP had greater potential for self-renewal in vitro. SP and non-SP were equally able to form and reform spheres at the same rate over more than six passages (Fig. 4A), demonstrating no differential acquisition of a stem-like phenotype. Expression of the stem-like self-renewal gene panel SOX2, OCT4, and MSI1 was compared in the sorted SP and non-SP cells. Neither cell line showed significant enrichment of self-renewal gene expression in the SP population compared with the non-SP (Fig. 4B). Curiously, the murine GL261 SP showed consistently higher CD133 mRNA levels than non-SP. This was never observed in the human SP cells and was not seen in the GL261 sphere-forming cells.

Figure 4.

SP cells do not have a cancer stem cell phenotype. (A): Neurosphere cultures of SP and non-SP cells at passage 6. Pictures representative of three independent experiments. (B): Expression of self-renewal genes in non-SP (white bars) and SP (black bars) cells was measured by real-time reverse transcription polymerase chain reaction. Fold change in gene expression in SP was expressed relative to the corresponding non-SP gene expression, which was normalized to 1. Data represents the average ± SD (error bar) of three independent experiments. (C): Time taken for subcutaneous tumor formation from parental (square), SP (triangle), and non-SP (circle) cells. Abbreviation: SP, side population.

Tumor initiation could be independent from increased self-renewal activity in vitro. To determine whether SP enriched for tumor-initiating cells compared with parental or non-SP, GL261 SP and non-SP cells were sorted, serially diluted, and assayed for subcutaneous tumor growth. Unexpectedly, there was a significant increase in time to tumor appearance (Fig. 4C) and decreased rate of tumor growth (data not shown) in SP-derived tumors seeded with 50,000 cells, compared with those from non-SP and parental cells. At 10,000 cells, SP completely failed to initiate tumors by 7 weeks, compared with 10,000 non-SP or parental cells, which both initiated in less than 7 weeks (data not shown). This was in marked contrast to the increased rate of tumor onset of sphere-forming cells (Fig. 2A). We conclude that unlike sphere formation, SP did not enrich for a tumor-initiating or stem-like population in our GBM cell lines and is not sufficient for a CSC phenotype.

SP Is Regulated Independently from the Self-Renewal Phenotype

SP is often increased by drug exposure, and chemo-resistance mediated by drug efflux is proposed to be a CSC property. We tried to skew the GBM SP toward a self-renewing CSC phenotype by exposure to doxorubicin. The primary lines, 09/06, 08/04, and the LN18 and GL261 cells were exposed to a transient (16 hours) doxorubicin exposure that resulted in ∼80% cell death. Surviving cells started to divide approximately 5 days postdrug exposure. At this point, self-renewal gene expression was measured and was significantly increased in doxorubicin-treated cells, similar to spheres and consistent with enrichment for a CSC phenotype by drug selection (Fig. 5A). As previously reported [27], CD133 expression also increased with drug exposure. Next, we attempted to purify the SP from doxorubicin-treated cells. However, the SP in treated cells was obscured, both by a “waterfall” of cells predominantly from the G2 population and by the increased apoptotic population adjacent to the SP (Fig. 5B). By 13 days postdrug exposure (dependent on the rate of cell division, data shown are for the fast dividing LN18 cells), a clean SP could be identified, showing threefold increase over the basal SP (Fig. 5B). By 21 days postdrug exposure, SP had reduced to basal level, demonstrating the fluctuation of SP in response to stress.

Figure 5.

Side population (SP) is enhanced by drug exposure, but is not associated with increased self-renewal. (A): Five days postdoxorubicin exposure, expression of self-renewal genes in DMSO (white bars)- and doxorubicin (black bars)-treated cells was measured by real-time reverse transcription polymerase chain reaction (PCR) in the cell lines indicated. Fold change in gene expression in dox-treated cells was normalized to DMSO-treated cells. Data represent the average ± SD of at least two independent triplicate experiments. (B): DMSO- or doxorubicin-treated LN18 cells were stained with H33342 at the indicated time points. SP is gated, with SP frequency given. Number of events collected indicated in lower right-hand corner. (C): DMSO- or doxorubicin-treated LN18 and GL261 cells were stained with H33342 at day 13 (LN18) or day 14 (GL261) postdrug exposure. SP is gated, with SP frequency given. Number of events collected indicated in lower right-hand corner. Data are representative of 2–3 independent experiments. (D): Expression of self-renewal genes in nonside population (white bars) and side population (black bars) was measured by real-time reverse transcription PCR. Fold change in gene expression in SP was expressed relative to the corresponding non-SP gene expression, which was normalized to 1. Data represents the average ± SD (error bar) of three independent experiments. Abbreviations: apop, apoptotic cells; DMSO, dimethyl sulfoxide; W, “waterfall” effect.

At 13–14 days postdrug treatment the SP was sorted from doxorubicin-treated LN18 and GL261 cells, and self-renewal gene expression compared between the drug-induced SP and non-SP cells. Despite the twofold to threefold increase in SP (Fig. 5C), the doxorubicin-treated SP cells showed no significant increase in self-renewal gene expression (Fig. 5D). This was essentially the same pattern of gene expression as the basal SP and non-SP cells (Fig. 4B). Again, as in the basal SP, CD133 expression was increased in the murine GL261 drug-induced SP versus the drug-induced non-SP cells.

Exposure to doxorubicin can both increase the SP and select a CSC-like population defined by the increased self-renewal gene signature in surviving cells. However, these are independent phenomenon, as the increased self-renewal gene expression was not seen in the isolated SP cells.


Human GBM cell lines and tumor-derived primary cells were shown to form spheres that fulfilled all the criteria for CSC, that is, increased self-renewal gene expression, increased tumor initiation, and the ability to differentiate into the lineages involved in GBM. However, the GBM CSC spheres did not contain a SP, regardless of whether they were derived from immortalized cell lines or from tumor-derived primary cells. Isolation of SP cells directly from GBM cell lines did not enrich for a self-renewing phenotype demonstrated by gene expression analysis in vitro and did not enhance tumor initiation in vivo. Drug treatment enriched for a CSC phenotype, and transiently increased the frequency of the SP, but did not alter the self-renewal profile of the SP. These data lead us to conclude that SP is not a useful marker of a CSC phenotype in human GBM. The absence of a SP in a set of tumorigenic melanoma cell lines extends this to other tumor types, a conclusion supported increasingly by the literature [26, 28–30].

SP did not occur in spheres, in any of our GBM cells. This lack of overlap between SP and other CSC enrichment techniques has been observed previously [31, 32]. This could mean that there is little similarity in cancer stem-like cells selected by different techniques, but we suggest this reflects the variability of cells identified in the SP.

Analysis of SP by flow cytometry separates cells according to DNA content using differential emission spectra of H33342 bound to DNA, and is highly reliant on dividing, viable cells for optimal identification of the highly effluxing SP [13]. Anything that alters cell division potentially alters H33342 efflux or emission of fluorescence, changing the composition of cells that make up the SP. Further complicating SP analysis is the lack of standard gating strategies. The SP is often identified merely as a continuous tail emerging from a homogenous main population, rather than a separate, distinct population to the side [33]. Changes in SP can be transient and induced by many cellular stresses, including drug [34, 35] and phytochemical [36] exposure, hypoxia [37], DNA damage, and genomic instability [38]. These stresses often transiently obscure the SP, as demonstrated by disruption and recovery of the doxorubicin-treated SP in our study. This makes accurate quantification or sorting of SP very difficult. In addition, SP sorting is technically demanding, requiring careful setup of laser strength and alignment, optics, and cell collection. All of these factors need to be optimized and carefully controlled to consistently identify or isolate the same subpopulation of cells.

The reproducibility of our SP isolation is indicated by the low variation in the gene expression data, which are the average of several experiments. Tumor experiments were also consistent, indicating that the SP from different sorts was composed of similar cells. Sorting of SP from parental material did clearly separate a population of cells with a different phenotype to the non-SP cells, as the SP was much less efficient at forming tumors than the non-SP cells, and at low cell number (1 × 105) SP completely failed to form tumors. This presumably reflects a lower proliferation rate or loss of cells with tumor-initiating potential in the SP selection procedure and certainly does not fit the paradigm for enrichment or selection of a CSC phenotype.

Only the GBM cell lines were used for self-renewal gene expression and tumor formation in the SP-sorting experiments. Ideally, the SP would have been sorted from primary GBM cells, as SP could clearly be visualized using the fiber-optic laser system of the LSRII flow cytometer. However, many technical limitations prevented these cells from being physically sorted. The large and irregular cell size combined with the stream-in-air delivery system and the optical setup of the BD FACS Vantage DiVa sorter made it impossible to identify and separate a clean SP from primary GBM cells with sufficient purity. Based on the similarity of the self-renewal phenotype to primary cells, LN18 cells were used instead. We did not sort the SP from U87MG even though it was present, albeit at lower frequency than LN18. U87MG is used more commonly by other researchers than LN18, as it readily forms subcutaneous tumors. However, our data suggest U87MG is not a good choice to model GBM CSC, as it does not form spheres that can be maintained over extended (>6) passage, neither does it upregulate self-renewal genes to high levels, unlike LN18 and primary cells. The cell line data support the conclusion that CSC can be selected or induced from immortalized cell lines such as LN18 and GL26.

The absence of a SP in our spheres, and the lack of a CSC phenotype in SP cells contrasts with reports from murine models, indicating that SP cells purified from spontaneous gliomas contain a SP with a CSC phenotype [14, 15]. There are several potential reasons for this. The simplest is that clean discrimination and sorting of a SP requires predominantly viable cells, which may be difficult to achieve using primary tumor material. Sorting on a continuous tail or indistinctly separated population would result in different cellular composition and phenotype of the sorted population. Alternatively, although we have demonstrated that cells grown in vitro can exhibit a CSC phenotype, the SP from such in vitro-adapted cells could have a different composition to SP cells isolated directly ex vivo from the spontaneous tumor tissue. In this context, it should also be noted that the spontaneous tumors in the murine models referred to above were accelerated by loss of p53 or PTEN and were treated with inhibitors of various signaling pathways. The relationship between cell viability, division rate, and SP may have been altered by these changes in signaling, again altering composition of the SP. Finally, a cancer stem-like cell phenotype can also be induced by mycoplasma infection [39, 40], something we have excluded by regular testing for mycoplasma.

Identification and composition of the SP is influenced by many factors, and the relationship between SP and a stem cell phenotype is not absolute, even in normal, noncancer cells. Although the bone marrow SP contains hematopoietic stem cells [9], the proportion of SP in bone marrow increases with age while the properties of those cells change [41], indicating variation in cells with dye efflux activity. Further, SP is not found in all pluripotent stem cells [42] and not all SP cells have a stem-like phenotype [30]. We have demonstrated in GBM that the SP cells are neither necessary nor sufficient for a CSC phenotype. Our data are consistent with recent reports from other tumors that SP is not a universal CSC marker [26, 28, 29].


In the absence of definitive cell surface markers for CSC, the SP is a functional assay that can be associated with stem-like characteristics. There are tumor types where a clean SP has been correctly identified, sorted, and associated with enhanced tumorigenicity over the non-SP cells. This includes breast [43], lung [32], and mesenchymal [44] tumors, demonstrating that SP can potentially mark a CSC population in some circumstances. However, SP as a CSC marker always requires confirmation by tumor experiments, and we strongly caution against using the presence, absence, or change in SP as an indicator of CSC activity.


We thank Brigitta Mester for flow cytometry assistance, Dr. Catherine Koleda for pathology assistance, Catherine Wood for coordination of GBM tissue collection and the Biomedical Research Unit of the Malaghan Institute of Medical Research for animal husbandry. This study was supported by the Wellington Division of the Cancer Society of New Zealand and a Cancer Society of New Zealand Project Grant (08/10). M.J.M. was supported by the McKenzie Medical Research Fellowship. M.I.F.H. was supported by the Neurological Foundation and the Royal Australasian College of Surgeons.


The authors indicate no potential conflicts of interest.