Page 1
BRAINA JOURNAL OF NEUROLOGY
Evidence for label-retaining tumour-initiatingcells in human glioblastomaLoic P. Deleyrolle,1,2,* Angus Harding,3,* Kathleen Cato,3 Florian A. Siebzehnrubl,1
Maryam Rahman,1 Hassan Azari,1,4 Sarah Olson,5 Brian Gabrielli,3 Geoffrey Osborne,2,6
Angelo Vescovi7 and Brent A. Reynolds1,2
1 McKnight Brain Institute, Department of Neurosurgery, University of Florida, Gainesville, FL 32610, USA
2 Queensland Brain Institute, University of Queensland, Brisbane, QLD 4072, Australia
3 Diamantina Institute for Cancer, Immunology and Metabolic Medicine, Melanoma division, University of Queensland, Brisbane, QLD 4072,
Australia
4 Department of Anatomical Sciences, Shiraz University of Medical Sciences, Shiraz 7134845794, Iran
5 Princess Alexandra Hospital, Department of Neurosurgery, University of Queensland, Brisbane, QLD 4072, Australia
6 Australian Institute for Bioengineering and Nanotechnology, Flow Cytometry Facilities, University of Queensland, QLD 4072, Australia
7 Department of Biotechnology and Biosciences, University of Milano-Bicocca, Milan, I-20126, Italy
*These authors contributed equally to this work.
Correspondence to: Brent A. Reynolds,
McKnight Brain Institute,
Department of Neurosurgery,
University of Florida,
Gainesville, FL 32610,
USA
E-mail: [email protected]
Individual tumour cells display diverse functional behaviours in terms of proliferation rate, cell–cell interactions, metastatic
potential and sensitivity to therapy. Moreover, sequencing studies have demonstrated surprising levels of genetic diversity
between individual patient tumours of the same type. Tumour heterogeneity presents a significant therapeutic challenge as
diverse cell types within a tumour can respond differently to therapies, and inter-patient heterogeneity may prevent the devel-
opment of general treatments for cancer. One strategy that may help overcome tumour heterogeneity is the identification of
tumour sub-populations that drive specific disease pathologies for the development of therapies targeting these clinically
relevant sub-populations. Here, we have identified a dye-retaining brain tumour population that displays all the hallmarks of
a tumour-initiating sub-population. Using a limiting dilution transplantation assay in immunocompromised mice, label-retaining
brain tumour cells display elevated tumour-initiation properties relative to the bulk population. Importantly, tumours generated
from these label-retaining cells exhibit all the pathological features of the primary disease. Together, these findings confirm
dye-retaining brain tumour cells exhibit tumour-initiation ability and are therefore viable targets for the development of thera-
peutics targeting this sub-population.
Keywords: brain tumour; cancer stem cells; glioblastoma; label-retaining cells; tumour-initiating cells
Abbreviations: ABC = adenosine triphosphate-binding cassette; CFSE = carboxyfluorescein diacetate succinimidylester; MCM = minichromosome maintenance; NOD/SCID = non-obese diabetic/severe combined immunodeficient
doi:10.1093/brain/awr081 Brain 2011: 134; 1331–1343 | 1331
Received September 3, 2010. Revised March 1, 2011. Accepted March 4, 2011. Advance Access publication April 22, 2011
� The Author (2011). Published by Oxford University Press on behalf of the Guarantors of Brain. All rights reserved.
For Permissions, please email: [email protected]
Page 2
IntroductionHuman glioblastoma is the most malignant and common primary
adult brain tumour, with a median survival time of 9–15
months, despite aggressive therapy (Vescovi et al., 2006). This
grim prognosis is due to therapy-resistant glioblastoma tumour
cells that initiate tumour regrowth after discontinuing therapy.
The identification and characterization of cell populations with
specific properties that initiate brain tumour recurrence is
essential for the development of effective therapeutics.
Distinct sub-populations of tumour-initiating cells have been
identified in leukaemia (Lapidot et al., 1994) and in many
solid tumours including Ewing’s sarcoma (Suva et al., 2009),
breast (Al-Hajj et al., 2003), prostate (Collins et al., 2005),
lung (Eramo et al., 2008), colon (Ricci-Vitiani et al., 2007),
liver (Suetsugu et al., 2006), pancreas (Li et al., 2007), ovarian
(Bapat et al., 2005) and brain cancer (Ignatova et al., 2002;
Galli et al., 2004; Singh et al., 2004; Bao et al., 2006), opening
up the possibility of characterizing this tumour cell population
for the development of targeted therapies. For human glioblast-
oma, tumour-initiating cells have been identified and isolated
based on the expression of several neural stem cell surface
markers such as CD133 (Singh et al., 2003, 2004), CD15 (Son
et al., 2009) and A2B5 (Ogden et al., 2008; Tchoghandjian
et al., 2010). Tumour-initiating cells have also been identified
based on the functional criteria such as aldehyde dehydrogenase
activity (Bar et al., 2007; Kast and Belda-Iniesta, 2009; Rasper
et al., 2010) and the ability to exclude Hoechst 33 342 dye
(defined as ‘side population’), reflecting elevated expression of ad-
enosine triphosphate-binding cassette (ABC) transporters, such as
breast cancer resistance protein (ABCG2) (Kondo et al., 2004;
Patrawala et al., 2005; Bleau et al., 2009). Recently, several
groups have also utilized the propensity of label retention to iden-
tify tumour-initiating cells from solid tumours such as in breast
(Krishnamurthy et al., 2008; Pece et al., 2010), skin (Roesch
et al., 2010) and pancreatic (Dembinski and Krauss, 2009)
cancer. To determine if a similar population of cells also exists
within human glioblastoma, we exploited the properties of the
prodrug carboxyfluorescein diacetate succinimidylester (CFSE),
which is converted by cellular esterase activity into a fluorescent
compound covalently bound to proteins and retained within the
cells (Lyons, 2000). CFSE dye enables quantification of cell prolif-
eration, as it is equally divided between daughter cells after div-
ision. Here, we describe the isolation and characterization
of an infrequently cycling (i.e. CFSE retaining), tumour-initiating
sub-population in human glioblastoma cells that may represent a
target to improve response to therapy.
Our study provides further evidence supporting the notion
that functional and phenotypic features can be used to identify
cells that initiate and drive tumour growth. These findings
confirm that functional intra-tumour heterogeneity exists
within glioblastoma cell populations, and that identification
of the cells driving tumour initiation may be important for
understanding tumour dynamics and developing effective
treatments.
Materials and methods
Tumour sample, primary culturingand propagationAll brain tumour samples used in this study were collected from pa-
tients undergoing surgical treatment and were obtained following in-
formed consent and Institutional Review Board approval. Biopsies were
classified by neuropathologists as glioblastoma or grade III glioma ac-
cording to WHO guidelines (Louis et al., 2007). After surgical removal,
the tissue was washed and mechanically dissociated before being
placed in an enzymatic cocktail containing trypsin/ethylenediaminete-
traacetic acid (0.05%) for 10min at 37�C, followed by filtration
through a 40-mm filter. Dead cells were quantified using trypan blue
labelling and the cells were then transferred (at a density of 50 000
viable cells per ml) into neurosphere assay growth conditions
(Deleyrolle and Reynolds, 2009). This serum-free culture system con-
taining epidermal growth factor (EGF, 20 ng/ml, R&D) and basic fibro-
blast growth factor (bFGF, 10 ng/ml, R&D) and enables isolation and
expansion in vitro of cells exhibiting stem cell characteristics. Under
these culture conditions, the tumour cells generate gliomaspheres that
can be serially passaged, as reported by Galli et al. (2004). Briefly,
when the gliomaspheres have reached an adequate size (�150mm
diameter), they were dissociated using enzymatic digestion with a so-
lution containing trypsin/ethylenediaminetetraacetic acid (0.05%) for
3–5min. Finally, cells were washed, counted using trypan blue to ex-
clude dead cells and replated in fresh media supplemented with
epidermal growth factor and basic fibroblast growth factor. Using
this technique, which has proven to be a more reliable model than
traditional cancer cell lines to study cancer biology (Lee et al., 2006),
we generated 20 patient-specific human glioblastoma gliomasphere
cultures and one patient-specific grade III gliomasphere culture that
we used in the current study. All lines used were passaged 520 times
and none of the lines expressed any functional or phenotypic changes
over this time span.
Growth rate assayTo measure cell proliferation and identify a slow-dividing population,
cells derived from 20 human glioblastoma samples and one grade III
glioma sample were loaded with CellTrace CFSE green fluorescent dye
(Molecular Probes) according to the manufacturer’s instructions. After
passage, cells were incubated with 5 mM CFSE. Slow-cycling cells (top
5%) and overall population (bottom 85%) were identified 5–10 days
after CFSE staining based on their CFSE retention level during culture.
Based on the fact that CFSE intensity decreased 2-fold every time a
cell divides, we calculated the time to undergo cell division based on
the decay rate of CFSE intensity, which we normalized to the fluores-
cence decay observed with non-proliferative mouse astrocytes. CFSE
bright cells (top 5%) dilute the dye significantly slower than that of
the overall population. Human glioblastoma cells cultured in the neu-
rosphere assay were observed using a bright field/fluorescent micro-
scope to monitor sphere formation and dye dilution on the day of
loading and 24, 48, 96, 120 and 144h after CFSE labelling. Growth
rate between the different cell populations was also analysed by mea-
suring the cell number obtained at each passage. The cellular fold
expansion was measured by dividing the number of cells quantified
at each passage by the number of plated cells.
1332 | Brain 2011: 134; 1331–1343 L. P. Deleyrolle et al.
Page 3
Sphere forming frequency assayFive to 10 days after passaging the cells, the spheres from seven
different lines were stained using Hoescht 33 342 (1 mg/ml Sigma)
and imaged with a Leica DMI 4000B fluorescent microscope
(Q-Capture-Pro 6.0). Gliomaspheres were quantified using HCA
Vision software. The sphere forming frequency was obtained by divid-
ing the number of observed gliomaspheres by the number of initially
plated cells (10 cells/ml in 384-well plates).
Sphere size measurementSingle cells from four different lines were seeded into 384-well dishes
as above and allowed to proliferate to form spheres. Spheres were
stained using Hoescht 33 342 (1 mg/ml Sigma) and imaged using a
Zeiss Axio Observer. Sphere sizes were measured using ImageJ
software.
Differentiation of stem cell progenyTo assess multipotency, cells were plated at a density of
2.5 � 105cells/cm2 onto poly-L-ornithine-coated glass coverslips in
basal culture media lacking growth factors and containing 10%
foetal calf serum (Singh et al., 2003). Multiple immunofluorescence
assay for neural antigens was performed after 7–10 days (Deleyrolle
and Reynolds, 2009).
Immunostaining and flow cytometryFive to 10 days post-CFSE load, immunostaining was performed using
antibody against CD133 (1:11, Miltenyi Biotec, 10 independent lines),
CD15 (BD Pharmingen, 1:50, 18 independent lines), ABCG2 (BD
Pharmingen, 1:50, nine independent lines) and mini chromosome
maintenance 2 (MCM2) (1:500, Santa Cruz, 12 independent lines).
Staining was quantified by flow cytometry (BD LSRII).
Multipotency assayFive to 10 days post-CFSE loading, the CFSEhigh fraction was isolated
by fluorescence-activated cell sorting (BD FACSAria Flow Cytometer)
and plated in the neurosphere assay for in vitro expansion before
being placed in differentiation conditions for 4–7 days. Multi-lineage
differentiation potential was analysed by fluorescent microscopy using
the antibodies anti-glial fibrillary acidic protein (1:500, Dako), TUJ1
(1:1000, Promega) and O4 (5 mg/ml, R&D Systems) to label astro-
cytes, neurons and oligodendrocytes, respectively.
To isolate and culture the in vivo (intracranial) slow- and fast-cycling
cells, 6–9 weeks post-implantation, the transplanted tissue was mech-
anically and enzymatically dissociated (Galli et al., 2004; Deleyrolle
and Reynolds, 2009). Single cells were stained with propidium iodide
(1 mg/ml) to detect viable cells. We used a specific anti-human CD56
antibody (1:100, BD Biosciences) to identify human cells, which were
isolated by fluorescence-activated cell sorting based on the CFSE level
and subsequently cultured in the neurosphere assay. CD56 staining
was also confirmed using fluorescent microscopy.
ImmunohistochemistryIn situ tumour formation was confirmed using haematoxylin and eosin
staining. Human glioblastoma cells were identified using an anti-
human Nestin antibody (1:500, Millipore) alone or in combination
with CD133 (1:300, Abcam). A human-specific MCM2 antibody
(1:200, Santa Cruz) was used to identify human glioblastoma cells
that were competent to divide. Immunocomplexes were visualized in
3,3’-diaminobenzidine using the ABC-Elite peroxidase method (Vector
Laboratories) or using secondary antibodies conjugated to Alexafluor
488 or 568 (1:500, Invitrogen) together with DAPI (1:1000,
Invitrogen).
Xenotransplantation assayWe used 6- to 10-week-old female non-obese diabetic/severe com-
bined immunodeficient (NOD/SCID) mice for all surgeries, following
institutional and national regulations. Two microlitres of cell suspension
(5000–100 000 live cells/ml) were injected (using a 5 ml Hamilton syr-
inge) into the striatum using a stereotactic apparatus. Injection coord-
inates were 2mm lateral to Bregma and 3mm deep. After tumour cell
implantation, the animals were monitored for any neurological signs
affecting their quality of life. When symptoms were observed (ataxia,
lethargy, seizures or paralysis), the mice were sacrificed and tumour
formation was confirmed by tissue analysis. Tumour-initiation ability of
the slow-cycling fraction and the overall population has been analysed
in three independent human glioblastoma cell lines and one grade III
glioma cell line. Although historic publications have reported injecting
as few as 100 cells and getting tumour formation (though not 100%
of the time), this addresses the issue of the minimal number of cells
sufficient to generate a tumour and does not provide the actual fre-
quency of tumour-initiating cells. It was recently demonstrated that
the frequency of tumour-initiating cells could be calculated in a stat-
istically robust manner by combining a limiting dilution assay with
rigorous statistical analysis (Hu and Smyth, 2009). Therefore, to quan-
tify tumour formation ability, we have used the accepted limiting di-
lution transplantation assay (ranging from 10000 to 200 000 cells
injected) coupled with statistical analysis using the ‘StatMod’ package
(Hu and Smyth, 2009) (http://bioinf.wehi.edu.au/software/limdil/),
part of the R statistical software project (http://www.r-project.org).
Generation of astrocytesMurine astrocytes were generated as described in the Supplementary
materials and methods of the Supplementary material.
5-Ethynyl-20-deoxyuridineincorporation5-Ethynyl-20-deoxyuridine retention was measured by fluorescence-
activated cell sorting in four different lines at different time points
post-labelling (0, 48, 72 and 96 h) after a 45min 5-ethynyl-20-
deoxyuridine pulse (5mM). Labelling was performed according to the
manufacturer’s instructions (Click-iTTM EdU, Invitrogen).
Quantitative polymerase chainreaction analysesTotal RNA was isolated from the sorted sub-populations of two cell
lines on the basis of CFSE labelling, using the RNAqueous�-Micro kit
(Ambion, #AM1931). Complementary DNA production and quantita-
tive polymerase chain reaction reactions were performed as described
in the Supplementary material.
Label-retaining glioma-initiating cells Brain 2011: 134; 1331–1343 | 1333
Page 4
Results
Identification of a label-retainingsub-population in humanglioblastoma
To identify and characterize label-retaining cells in human glio-
blastoma, cells derived from primary tumours cultured in the neu-
rosphere assay (Galli et al., 2004; Singh et al., 2004; Bao et al.,
2006; Lee et al., 2006; Piccirillo et al., 2006) were loaded with the
non-selective cell-permeant fluorescent dye CFSE. This intracellular
fluorescent dye is partitioned evenly between daughter cells upon
cell division, resulting in a 2-fold dilution of the fluorescence in-
tensity, thereby enabling proliferation kinetic quantifications
(Lyons, 2000; Barnes and Melo, 2006). During 8–10 cell divisions,
the original intensity decreases by 28 to 210 reaching a level
equivalent to the autofluorescence of unlabelled cells (Lyons,
2000). We monitored gliomasphere formation by CFSE-labelled
human glioblastoma cells using dual bright field-fluorescent mi-
croscopy (Fig. 1A), and quantified the fluorescence intensity over
time using flow cytometry (Fig. 1B). This process demonstrated
the serial dilution of CFSE with each cell division and the subse-
quent growth of gliomaspheres. Figure 1C shows a typical flow
cytometry histogram of CFSE levels within a gliomasphere culture
together with fluorescent micrographs of varying CFSE intensities.
Using these methods, we were able to identify two populations in
20 individual human glioblastoma cell lines and in one grade III
glioma line (Fig. 1C and Supplementary Fig. 1); a label-retaining
population of cells (top 5% CFSE) and an overall population
(bottom 85% CFSE), separated by a 10% gap to avoid overlap
and contamination between the two fractions. Comparison of
CFSE decay between both populations over time revealed that
the label-retaining fraction diluted CFSE significantly less compared
with the overall population (Fig. 1D). Importantly, CFSE decay in a
non-proliferating control population was significantly reduced
compared with the human glioblastoma cells (Fig. 1D and
Supplementary Fig. 2), supporting the notion that loss of CFSE
intensity observed in the tumour cells was driven by cell division
(Fig. 1D). Based on their CFSE dilution properties, we calculated
the time per cell division for both populations (Fig. 1E). On aver-
age, the label-retaining cells underwent one cell division every
73.22 � 7.94 h, whereas the overall population divided once
every 24.96 � 0.87 h. To further confirm that the decay in
CFSE intensity was due to proliferation-induced dilution of
the label, we next investigated the ability of the CFSE-retaining
and CFSE-diluting populations to retain 5-ethynyl-20-deoxyuridine
labelling. For this purpose, the cells were loaded with
5-ethynyl-20-deoxyuridine for 45min and retention was
measured over a period of 96 h. CFSEhigh-top 5% population dis-
played a significantly greater 5-ethynyl-20-deoxyuridine retention
over time when compared with CFSElow-bottom 85% fraction
(Fig. 1F).
Altogether, these data validate the use of CFSE to identify cel-
lular sub-populations within glioma cells cultured as gliomaspheres
based on their rate of cell division.
Characterization of label-retaininghuman glioblastoma cells
Label retaining and the overall cell population were isolated using
fluorescence-activated cell sorting and cultured in the neurosphere
assay in which both populations generated spheres at a frequency
of 7.70 � 0.97 for the CFSEhigh versus 10.03 � 1.72 for the
CFSElow (Fig. 2A–C). The average size of gliomaspheres generated
by the overall population was significantly higher than that of the
gliomaspheres generated by the slow-cycling cell population, indi-
cating lower cell division frequency occurring within the CFSEhigh
spheres (Fig. 2D). Although long-term cell culture showed
that both populations exhibited cardinal in vitro stem cell
characteristics of extensive self-renewal, generation of a
large number of progeny and multi-lineage differentiation poten-
tial (Fig. 2F and G), the slow-cycling cells expansion rate was
significantly lower when compared with the overall population
(Fig. 2E). The reduction in sphere size and expansion rate of
CFSEhigh cells provides further evidence of the reduced prolifera-
tive rate of CFSEhigh progeny compared with the overall CFSElow
population.
We then analysed CFSE-retaining cells by flow cytometry with a
panel of markers. We demonstrated that human glioblastoma and
grade III glioma CFSEhigh cells expressed cell surface markers used
previously to identify tumour-initiating cells (Singh et al., 2004;
Bleau et al., 2009; Son et al., 2009) (CD133+/CD15+/ABCG2+ )
(Fig. 2H–K). Even though CD133 expression has been described
in tumourigenic and non-tumourigenic cells, this marker is com-
monly used to enrich for tumour-initiating cells in human
brain tumours (Singh et al., 2004; Bao et al., 2006; Shackleton
et al., 2009). Importantly, the slow-dividing population was
enriched in CD133+ cells (Fig. 2H and I). When the entire popu-
lation was evaluated for CD133 immunoreactivity, 14% of the
CD133+ fraction was slow cycling when compared with only
3% of the CD133� cells (Fig. 2M and N). Earlier reports
showed that both CD15 (Son et al., 2009) and ABCG2 (Bleau
et al., 2009) expression enrich for tumourigenic potential. Our
data demonstrating a greater CD15+ and ABCG2+ fraction in
the CFSE-retaining population indicate a potential enrichment
for tumour-initiating cells within the slow-cycling compartment
(Fig. 2H–K).
For analysis of proliferative capacity, the cells were tested for
expression of proteins actively involved in cell proliferation.
MCM proteins (including MCM2-7) are nuclear proteins
involved in the proliferation licensing system by regulating DNA
replication (Blow and Hodgson, 2002). While MCM2 is absent in
differentiated cells, it is highly expressed in human cancer
cells, and plays a vital role in genome duplication in proliferating
cells (Lei, 2005). MCM family proteins (especially MCM2) can be
used to identify cells competent to divide and have, therefore,
been classified as cancer biomarkers (Blow and Hodgson, 2002;
Semple and Duncker, 2004). Using flow cytometry, we
quantified the cells competent to divide using labelling with
MCM2. When MCM2+ cells were quantified, we observed
enrichment in the slow-cycling compartment for proliferative po-
tential compared with the overall population (Fig. 2H and L).
1334 | Brain 2011: 134; 1331–1343 L. P. Deleyrolle et al.
Page 5
Figure 1 Identification of slow-dividing cancer cells in human glioblastoma and grade III glioma. (A) Gliomasphere formation and
fluorescence of CFSE-loaded human glioblastoma cultures were monitored at the indicated times. (B) CFSE intensity was recorded by
fluorescence-activated cell sorting at the indicated times after load. (C) Flow cytometry histogram and fluorescent micrographs of CFSE
intensity revealed in human glioblastoma and grade III glioma cell lines after 5–10 days of growth in the neurosphere assay. A slow-cycling
population of cells and an overall population were identified in 20 independent human glioblastoma cell lines and one grade III glioma cell
line based on their capacity to retain CFSE (CFSEhigh-top 5% and CFSElow-bottom 85%). (D) A time dependent decrease of mean
fluorescence intensity (MFI) was observed over a period of 120 h after CFSE labelling. A non-proliferative mouse astrocyte culture was
used to determine the baseline for CFSE dilution unrelated to cell division. **P50.005, n = 6–11, t-test, four independent cell lines.
(E) Human glioblastoma populations’ division times were calculated from the CFSE intensity decay data and normalized to the decay
obtained with the non-proliferative mouse cells. Based on the fact that at every cell division the fluorescence intensity is divided by two, we
defined the following formula: 2X = B with X as the number of cell division and B as the ratio (initial CFSE MFI)/(final CFSE MFI). Therefore,
X = logB /log2. **P50.01, n = 9, t-test, four independent cell lines. (F) 5-Ethynyl-20-deoxyuridine labelling (EdU) after 4 h of pulse and
different chase period times revealed higher ability to retain 5-ethynyl-20-deoxyuridine staining overtime in the CFSEhigh fraction compared
with the overall population supporting slower proliferation. *P50.001, **P51 � 10�5, ***P5 1 � 10�10, n = 6–12, t-test, four
independent cell lines.
Label-retaining glioma-initiating cells Brain 2011: 134; 1331–1343 | 1335
Page 6
Enrichment in CD133+ and MCM2+ cells in the slow-cycling
population was also confirmed by quantitative polymerase chain
reaction (Fig. 2O). The overall population was used as calibrator
for the quantitative polymerase chain reaction experiments, and
the expression levels of CD133 and MCM2 were greater in the
slow-cycling cells (7.83 � 1.41 and 4.1 � 1.1, respectively)
(Fig. 2O).
Together, these results demonstrate that the label-retaining
CFSEhigh cell population is enriched for replication-competent
CD133, CD15 and ABCG2 immunoreactive cells.
Figure 2 A slow-cycling fraction with stem-like cell features exists in human glioblastoma. The two distinct populations were isolated
using the methodological scheme described in Fig. 1C. The micrographs show gliomaspheres derived from both slow-cycling cells (A)
or overall population (B). (C) The sphere forming frequency between the two populations was compared. P40.25, n = 7, t-test. (D) The
size of the spheres formed by the slow-cycling cells and the overall population was measured. **P5 0.001, t-test. Five hundred and
twelve and 907 spheres were measured from the CFSEhigh and CFSElow populations, respectively. (E) Although the expansion rate,
observed over a period of up to 7–10 passages, for the CFSEhigh cells was lower than the overall population, both fractions exhibit
long-term self-renewal and the ability to generate a large number of progeny. *P5 0.05, n = 7–12, t-test. (F and G) When placed in
differentiation conditions, CFSEhigh (F) and CFSElow (G) cells are multipotent as observed by the expression of glial fibrillary acidic protein
(GFAP, green) and TUJ1 (red) (left) or of GFAP (green) and O4 (red) (right). (H) CD133+ , CD15+ , ABCG2+ and MCM2+ fractions
derived from undifferentiated gliomaspheres were assayed by fluorescence-activated cell sorting. **P50.001, ***P50.0001, t-test,
including 9–18 independent human glioblastoma lines and one grade III glioma. (I–L) Fluorescence-activated cell sorting histograms
comparing CD133 (I), CD15 (J), ABCG2 (K) and MCM2 (L) staining between slow-cycling cells and overall population. (M) Flow
cytometric histogram comparing the CFSE profile between CD133+ cells and CD133� cells. (N) Quantification of the slow-cycling cells
in the different CD133 fractions demonstrated a significant enrichment of CFSE-retaining cells in the CD133+ population compared
with the CD133- cells. **P50.005, n = 7, t-test. (O) Quantitative polymerase chain reaction analysis indicates that in the slow-cycling
cells, CD133 and MCM2 are highly expressed in comparison to the overall population. *P50.05, n = 4, t-test.
1336 | Brain 2011: 134; 1331–1343 L. P. Deleyrolle et al.
Page 7
Label-retaining human glioblastomacells have greater tumour-initiationability
After demonstrating that, slow-cycling human glioblastoma cells
are enriched in cells expressing tumour-initiating markers, we
sought to assess their in vivo tumour-initiation capability.
Transplantation of 200 000 CFSEhigh or CFSElow cells (derived
from three glioblastoma or one grade III glioma cell line) into
the striatum of immunocompromised mice (SCID) resulted in
tumour formation and 100% mortality (Table 1). Our standard
injection of 200 000 cells, based on previously published work
(Galli et al., 2004; Piccirillo et al., 2006; Beier et al., 2007;
Chen et al., 2010), was chosen as it generates a tumour
�100% of the time. The glioblastoma-derived tumours exhibited
typical histopathological hallmarks that define high-grade glioma
(Fig. 3A–G) (Galli et al., 2004; Lee et al., 2006). The slow-dividing
cell-derived tumours displayed migratory and infiltration capability
(Fig. 3A, F and G), nest-like formations (Fig. 3B), vascular prolif-
eration and nuclear pleomorphism with mitotic figures (Fig. 3C) as
well as areas of pseudo-palisading necrosis (Fig. 3D). Anti-human
nestin staining confirmed that slow-cycling cell-derived tumours
were composed of human glioblastoma cells (Fig. 3E). Nestin
labelling demonstrated infiltration of the tumour cells into the par-
enchyma as a result of slow-cycling progeny that would cross the
contralateral hemisphere, migrate along the sub-cortical white
matter tracts towards the ventricular system (Fig. 3F), and infil-
trate the overlying cerebral cortex (Fig. 3G). Each tumour con-
tained cells competent to divide as evidenced by the expression
of MCM2 (Fig. 3H). Tumours derived from the slow-cycling frac-
tion were also immunoreactive for the brain tumour-initiating
marker CD133 (Fig. 3I).
Next, we compared the tumour-initiating efficiency between
CFSE-retaining cells and the overall population by performing lim-
iting dilution transplantation assay into the striatum of immuno-
compromised mice (Fig. 4). This experiment showed that 0.01%
of the CFSEhigh cells could initiate a tumour whereas only 0.003%
of the CFSElow exhibited this ability (Fig. 4 and Supplementary
Fig. 3).
Together, these results reveal that the label-retaining human
glioblastoma sub-population is enriched in tumour-initiating cells,
as the slow-cycling sub-population contains significantly more cells
able to initiate the generation of high-grade brain tumours than
the overall tumour population.
In vivo CFSE-retaining cells displaytumourigenic potential
To address the question of whether label-retaining human glio-
blastoma cells exist in vivo, freshly stained CFSE cells derived from
human glioblastoma gliomasphere cultures were injected into
the striatum of immunodeficient mice (200 000 cells/mouse). All
animals developed invasive tumours and a sub-population of
CFSE-retaining cells was clearly evident 6 weeks after implant-
ation, confirming that label-retaining human glioblastoma cells
exist in vivo after intracranial transplantation (Fig. 5A). One of
the technical challenges in using flow cytometry for the analysis
of human cells in xenograft models in vivo is that the implanted
tumour cells migrate and infiltrate the surrounding host tissue,
thereby creating a chimeric population of human and host
cells. This problem can be overcome by using a human-specific
CD monoclonal antibody (CD56) that recognizes virtually 100% of
the human glioblastoma cells but not mouse cells (Supplementary
Fig. 4). Using this approach, implanted human glioblastoma cells
(expressing CD56 antigen) were separated from the host tissue
(debris and mouse cells). The in vivo slow-cycling (CFSEhigh) and
faster cycling (CFSElow) cells were then isolated and individually
cultured (Fig. 5B and C). Both slow- and fast-cycling in vivo cells
exhibited expansion in culture. Like in vitro CFSE-retaining cells,
in vivo slow-cycling cells generate gliomaspheres (Fig. 5D) and
can drive long-term expansion; however, they have a significantly
reduced mean fold expansion when compared with the in vivo
CFSElow population (Fig. 5E). Finally, to confirm in vivo tumour-
igenicity, we re-implanted the cells (200 000 cells/animal) derived
from in vivo CFSE-retaining or diluting cells cultured in neuro-
sphere assay into immunocompromised mice (Fig. 5F and G).
Following re-transplantation, all animals (Fig. 5F and G) developed
large tumours displaying human glioblastoma-like features
(i.e. vascular proliferation, pseudopalisading necrosis and nuclear
pleomorphism) resulting in the death of the animals (Fig. 5H).
Together, these results show that tumour-initiating label-retaining
cells are present in vivo.
To our knowledge, the results presented here provide the first
evidence for the existence of a label-retaining tumour-initiating
Table 1 Tumour formation ability of the slow-dividing cells
Number of tumours/number of injections Time frame for tumour formation (weeks)
200 000 cells per injection 200 000 cells per injection
L0 L1 L2 L3a L0 L1 L2 L3a
Overall population 36/36 17/17 9/9 5/5 11.1 � 0.5 11.3 � 0.8 27.6 � 2.6 16.4 � 0.6
Slow-cycling cells 13/13 4/4 2/2 3/3 19.1 � 0.9 21.8 � 0.5 19.5 � 1.5 21.3 � 2.0
Flow cytometrically isolated CFSEhigh and CFSElow cells derived from three independent human glioblastoma-derived cells (line 0, 1 and 2; respectively L0, 1 and 2) or one
grade III glioma-derived cells (line 3, L3) were intracranially transplanted. Shown are the numbers of implanted animals and mice bearing tumours at the indicated times.
a Grade III glioma.
Label-retaining glioma-initiating cells Brain 2011: 134; 1331–1343 | 1337
Page 8
cell population within gliomasphere-derived human glioblastoma
cells.
DiscussionThe notion that a self-renewing, infrequently cycling, cancer
stem-like cell population is responsible for tumour initiation is
well-established in leukaemias (Holyoake et al., 2001; Graham
et al., 2002). While an infrequently cycling compartment has
also been described in solid tumours such as breast cancer
(Krishnamurthy et al., 2008; Pece et al., 2010), pancreas adeno-
carcinoma (Dembinski and Krauss, 2009) and melanoma (Roesch
et al., 2010), a similar population has yet to be identified in brain
tumours.
Here, we identified a sub-population of label-retaining cells
within human glioblastoma that exhibited a lower frequency of
cell division, compared with the bulk of the tumour cells, along
Figure 3 Human glioblastoma-derived CFSE-retaining cells form intracranial glioblastoma-like tumours in immunocompromized mice
(representative of three independent cell lines). (A) Haematoxylin and eosin staining demonstrates the presence of a large tumour mass
surrounding the injection site 18 weeks post-implantation of 200 000 slow-cycling cells. (B and C) As shown by haematoxylin and eosin
staining, slow-cycling cell-derived tumours presented peculiar nest-like structures (asterisk, B) demonstrating high mitotic activity also
illustrated by the presence of mitotic figures (yellow arrow, C). Blood vessel neo-formations within the tumour were also identified (black
arrow, C). (D) Low magnification of an area of pseudo-palisading necrosis (arrowhead) that was identified within the tumour generated by
the implantation of slow-cycling cells. (D0) High magnification of neoplastic cells (asterisk) accumulating near an area of necrosis
(arrowhead). (E) Anti-human-specific Nestin labelling confirmed the existence within the host brain of human glioblastoma cells that
compose the bulk of the tumour developed after implantation of slow-cycling cells. (F) The implanted slow-cycling cells (Nestin immu-
noreactive, arrows) displayed infiltration properties as shown by their migration along the sub-cortical white matter (SCWM) towards the
lateral ventricle (LV). (G) Nestin positive slow-cycling tumour cells also spread to the overlying cortex (Cx) (arrows). Brain tumours
generated from slow-cycling cells contain cells competent to divide (MCM2 immunoreactive, H–H0) and cells expressing the cancer stem
cell marker CD133 (arrows, I–I0).
1338 | Brain 2011: 134; 1331–1343 L. P. Deleyrolle et al.
Page 9
with the expression of CD133, CD15 and ABCG2, as well as an
enhanced ability to form tumours in vivo; features that are con-
sistent with a tumour-initiating cell. To track cell divisions and
identify slower cycling cells, we used functional labelling with
the lipophilic, non-fluorescent precursor, carboxyfluorescein diace-
tate succinimidyl ester (CFDASE). The probe is activated by intra-
cellular esterase activity converting it to fluorescent CFSE while
covalently coupling it to amino groups where it becomes cell per-
manent and is diluted in half at each cell division (Lyons, 2000).
Cells exhibiting higher CFSE epifluorescence over time
corresponded to slow-dividing cells, which was confirmed using
nucleoside analogue incorporation [bromodeoxyuridine (BrdU) or
5-ethynyl-20-deoxyuridine (EdU)] followed by a chase period, cor-
relating label-retaining cells with CFSE intensity (Golmohammadi
et al., 2008). This shows that CFSE retention directly correlates
with 5-ethynyl-20-deoxyuridine retention (Fig. 1F), providing direct
confirmation for the hypothesis that a subset of human glioblast-
oma cells cultured as gliomaspheres retain CFSE due to reduced
cell division. In sum, these results, together with the extensive
literature using CFSE to track cell division (Lyons, 2000; Graham
Figure 4 Tumour initiating frequency of human glioblastoma slow-cycling cells. (A) Limiting dilution assay. From 10000 to 200 000 cells,
the slow-cycling cells or the overall population derived from one cell line were injected into the striatum of NOD/SCID mice. The
percentage of animals bearing brain tumours was recorded. (B) The frequency of the cells able to generate a tumour was calculated based
on the numbers presented in the table. Tumour-initiating cell frequency (evaluated using likelihood ratio tests) was greater in the
slow-cycling fraction compared with the overall population (�12 = 4.45, *P = 0.0349). Time frame for tumours to develop is presented
as average � SEM.
Label-retaining glioma-initiating cells Brain 2011: 134; 1331–1343 | 1339
Page 10
Figure 5 In vivo CFSE-retaining cells show stem cell features and tumour formation ability. (A) CFSE+ cells were observed at the
injection site 6 weeks post-transplant of CFSE-loaded cells into the brain of NOD/SCID mice, demonstrating the existence of slow-cycling
cells in vivo. (B) Human-specific anti-CD56 antibody was used to separate the donor cells from the host tissue. (C) In vivo, slow-cycling
cells and overall population were identified from the CD56+ fraction and isolated based on their CFSE level. (D) Biopsy samples from
brains of mice implanted with slow cycling and overall population were re-cultured in the neurosphere assay. The micrograph shows
gliomaspheres derived from the in vivo slow-cycling cells. (E) Re-cultured human glioblastoma cells exhibited long-term self-renewal and
ability to generate large number of progeny. Like in vitro, the in vivo CFSE-retaining cells showed a lower fold expansion rate measured
over a number of seven passages. **P5 0.01, n = 13–24, t-test. (F–G) The progeny of the in vivo CFSEhigh or CFSElow cells cultured in the
neurosphere assay were re-implanted in the striatum of immunocompromised animals. All the animals that had transplants developed
tumours. Survival was also analysed, P5 0.436, Log-rank test, two independent cell lines. (H) Like the in vitro slow-cycling cells, the
in vivo CFSE-retaining cells give rise to progenies able to generate tumours exhibiting human glioblastoma features when transplanted
into the striatum of NOD/SCID mice. In vivo slow-dividing cells were isolated by flow cytometry and cultured in vitro in the neurosphere
assay for several passages and their progenies were subsequently transplanted intracranially. Vascular proliferation (black arrow), necrosis
(arrowhead) surrounded by pseudo-palisade (asterisk), nuclear pleomorphism and mitosis (yellow arrow) were evident 13 weeks
post-transplantation using haematoxylin and Eosin staining. (H0) Higher magnification of the box presented in (H).
1340 | Brain 2011: 134; 1331–1343 L. P. Deleyrolle et al.
Page 11
et al., 2002), demonstrate CFSE labelling as a valid approach to
identify and isolate sub-fractions of cells based on the frequency
of cell division.
In support of the cancer stem cell hypothesis, we demonstrate
that label-retaining cells, isolated from cultured human
glioblastoma-derived spheres using this methodology (CFSEhigh),
possess the characteristics of long-term proliferation, extensive
self-renewal, generation of large number of progeny and multi-
potency (Fig. 2). As previously demonstrated in leukaemia
(Holyoake et al., 1999), breast (Krishnamurthy et al., 2008) and
pancreatic cancers (Dembinski and Krauss, 2009), where
low-frequency cell division was associated with tumourigenicity,
we show here that the human glioblastoma and human grade
III glioma infrequently dividing sub-population is enriched in
CD133, CD15 and ABCG2 (Fig. 2), markers used to identify
brain tumour-initiating cells with enhanced competencies for
self-renewal, tumour formation and treatment resistance (Singh
et al., 2004; Bao et al., 2006; Bleau et al., 2009; Son et al.,
2009). As relative quiescence would be a functional characteristic
providing protection against conventional treatments (such as ra-
diation) targeting actively dividing cells (Anderson et al., 2003;
Zhou et al., 2003), our findings suggest an additional mechanism
underpinning resistance to such treatment of the CD133+ , CD15+
or ABCG2+ fractions compared with their respective negative
populations reported in the literature (Bao et al., 2006; Bleau
et al., 2009).
While there is an overlap with formerly identified tumour-
initiating cell markers, the infrequently cycling CFSE-retaining
cells define a definite and unique population of tumour-initiating
cells. Therefore, this study constitutes a relevant step towards
characterizing the biological activities of sub-populations within
the extensive heterogeneous tumour environment, and provides
further evidence for heterogeneity in solid tumours based on the
functional criteria (i.e. the frequency of cell division).
Importantly, not only in vitro, but also in vivo, the CFSEhigh
fraction demonstrated the ability to establish tumours. Similar ex-
periments have been described in an in vivo breast cancer model
using vibrant CM DiI (Chloromethyl 1,10-dioctadecyl-3,3,3030-
tetramethylindocarbocyanine perchlorate) to isolate slow-cycling
cells (Krishnamurthy et al., 2008). Our study reports on an
in vivo model using CFSE as a tracker in a CNS-derived tumour.
These results further validate the generality of selecting
sub-populations of cancer cells based on their rate of division as
determined by the ability to retain CFSE, which does not reflect a
culture-specific phenomenon. This notion is additionally supported
by the data shown in Supplementary Fig. 5 that demonstrate a
random distribution of the CFSE-retaining cells within glioma-
spheres invalidating the hypothesis of either nutrient or oxygen
deprivation leading to a reduced proliferation rate of this popula-
tion due to its concentration in the core of the spheres.
Importantly, the ability of label-retaining cells to recapitulate
tumours that harbour extensive similarities to the original disease
(i.e. high mitotic activity, pseudopalisading necrosis, vascular
proliferation and invasion) render the slow-cycling population an
essential candidate to brain tumour initiation and progression,
identifying them as a potential novel therapeutic target.
Additionally, these data support the relevance of studying cancer
cell biology using serum-free culture conditions (i.e. neurosphere
assay) as sub-population of cells from gliomaspheres derived from
primary human glioblastoma biopsies demonstrate the capability
to recapitulate the overall in vivo phenotype of the parental
tumour (Lee et al., 2006). Moreover gliomasphere formation abil-
ity has been associated with clinical outcome in malignant glioma
demonstrating the neurosphere assay as a tumour relevant meth-
odology (Laks et al., 2009). However, a limitation of our study is
that our observations included only cells propagated in the neuro-
sphere assay; therefore, one cannot exclude the possibility that a
slow-dividing cell population lacking gliomasphere-generating abil-
ity may exist in vivo.
While the progeny of the slow-cycling cells divided less fre-
quently in culture (Figs 2D and E and 5E), this compartment
demonstrated proliferative and tumour-initiating potential, as evi-
denced by an enrichment in cells that were competent to divide
(i.e. increased MCM2 immunoreactivity) and in expression of
CD133, CD15 and ABCG2, respectively (Fig. 2H–O). On the sur-
face, the decreased effective proliferation of the slow-cycling pro-
geny in culture (Fig. 2D and E) appears to contradict the in vivo
tumour initiation (Fig. 4) and progression (Fig. 5G) properties;
however, these results may be reconciled by appreciating that
tumour initiation and progression are not solely influenced by
cell proliferation. Tumour cell invasion into the healthy brain,
angiogenesis and the tumour cell niche are all likely to contribute
to driving tumour initiation and progression (Brennan et al., 2009;
Witz, 2009; Wong et al., 2009; Qian and Pollard, 2010), and our
study suggests that a high-proliferative rate is not a primary driver
of these mechanisms.
The use of single cell-surface marker expression to identify and
characterize putative tumour-initiating cells in tumours remains
controversial (Bidlingmaier et al., 2008; Cheng et al., 2009).
This is particularly true for CD133, the marker most widely used
to identify brain tumour-initiating cells. CD133 is a cholesterol-
binding membrane protein of unknown biological function and
has been shown by several groups to be preferentially expressed
in tumour-initiating cells (Singh et al., 2003, 2004; Collins et al.,
2005; Suetsugu et al., 2006; Ricci-Vitiani et al., 2007; Eramo
et al., 2008; Suva et al., 2009). However, recent studies have
questioned the utility of using CD133 as a marker for
tumour-initiating cells, as CD133 negative cells have been demon-
strated to be efficient at initiating tumours in a variety of tumour
types, including human glioblastoma tumours (Beier et al., 2007;
Bidlingmaier et al., 2008; Joo et al., 2008; Ogden et al., 2008;
Wang et al., 2008; Cheng et al., 2009; Kelly et al., 2009; Nishide
et al., 2009; Shackleton et al., 2009; Son et al., 2009). In addition,
CD133 expression has been reported to increase in response to
cellular stress, further confusing its utility as a robust marker for
tumour-initiating cells (Griguer et al., 2008). It, therefore, remains
an open question as to whether CD133 expression can be used to
unambiguously identify tumour-initiating sub-fraction in human
glioblastoma. A very recent study may help explain these conflict-
ing results (Chen et al., 2010). These data suggest that the ability
of cells to move between CD133+ and CD133– sub-populations is
a better indicator of tumour-initiation ability that CD133 expres-
sion per se. This interpretation is consistent with our finding that
although the number of CD133+ cells is increased within the
Label-retaining glioma-initiating cells Brain 2011: 134; 1331–1343 | 1341
Page 12
label-retaining population; it is not increased to the extent one
would expect if only CD133 expressing cells can initiate tumours.
In contrast, label retention has consistently enriched for tumour
initiation across multiple tumour types (Krishnamurthy et al.,
2008; Pece et al., 2010), supporting the utility of this approach
for the identification of tumour-initiating cells. However it is im-
portant to note that our analyses have revealed that the
label-retaining population is made up of several tumour cell
sub-populations (L. P. Deleyrolle et al., unpublished data).
Future work analysing lineage relationships and isolating a more
pure tumour-initiating sub-population based on multiple functional
and phenotypic parameters is now underway to formally identify
and quantify the prevalence of tumour-initiating cells within pri-
mary tumours. Nevertheless our study, in combination with recent
publications, reveals label retention as an effective marker for en-
riching tumour-initiating cells from the total human glioblastoma
cell population for further study.
In conclusion, our results show that label-retaining cells, defining
a slow-cycling fraction, exist within human glioma (under the ex-
perimental paradigms used) and that this population in human glio-
blastoma cells is enriched in tumour-initiating cells expressing
tumour-initiating cell markers CD133, CD15 and ABCG2 and ex-
hibiting functional characteristics expected of a tumour-initiating
cell population in culture. These findings, together with data from
a growing number of studies (Graham et al., 2002; Krishnamurthy
et al., 2008; Dembinski and Krauss, 2009; Pece et al., 2010; Roesch
et al., 2010), provide a strong rationale for the contribution of
label-retaining cancer cells towards tumour initiation in cancer.
Therefore, identifying agents that effectively target the
label-retaining fraction may lead to improving outcomes in patients
when combined with conventional treatments that target the rap-
idly dividing population (Reya et al., 2001). Glioblastoma tumours
are characterized by the presence of a multitude of cell types, and
this tumour complexity is thought to contribute to the high rate of
therapeutic failure. Defining and understanding tumour heterogen-
eity by the identification of clinically relevant cellular sub-networks
is important to design combinatorial therapeutic interventions
enhancing disease outcome. The results presented here demonstrat-
ing the isolation and characterization of a sub-compartment of in-
frequently dividing cells, consistently identified as tumourigenic,
constitute an important step towards the comprehension of the
pattern of diversity encountered in glioblastoma tumours.
AcknowledgementsWe thank Dr Denis de Assis for his contribution to this article.
We thank Dr Vedam-Mai and Daniel Silver for their constructive
criticisms on this article, Dr Yachnis for his histopathological evalu-
ation of the tumours, Dr Marshall and Dr Chen for their assistance
in performing the irradiation experiments, Amy Poirier and Neal
Benson for cell sorting and Maria Caldeira for her technical support.
FundingNational Institutes of Health (1R21CA141020-01 to B.A.R.); the
Australian National Health and Medical Research Council (511091
to B.A.R, and 569662 to A.H.); the Australian Research Council
(DP1094181 to A.H.); the Cancer Council Queensland (to K.C.);
the Brain Foundation (to A.H.); the Florida Centre for Brain
Tumour Research (to B.A.R.); Preston A. Wells Jr. Center for
Brain Tumour Therapy (to B.A.R.).
Supplementary materialSupplementary material is available at Brain online.
ReferencesAl-Hajj M, Wicha MS, Benito-Hernandez A, Morrison SJ, Clarke MF.
Prospective identification of tumourigenic breast cancer cells. Proc
Natl Acad Sci USA 2003; 100: 3983–8.
Anderson HJ, Andersen RJ, Roberge M. Inhibitors of the G2 DNA
damage checkpoint and their potential for cancer therapy. Prog Cell
Cycle Res 2003; 5: 423–30.
Bao S, Wu Q, McLendon RE, Hao Y, Shi Q, Hjelmeland AB, et al. Glioma
stem cells promote radioresistance by preferential activation of the
DNA damage response. Nature 2006; 444: 756–60.
Bapat SA, Mali AM, Koppikar CB, Kurrey NK. Stem and progenitor-like
cells contribute to the aggressive behavior of human epithelial ovarian
cancer. Cancer Res 2005; 65: 3025–9.
Bar EE, Chaudhry A, Lin A, Fan X, Schreck K, Matsui W, et al.
Cyclopamine-mediated hedgehog pathway inhibition depletes
stem-like cancer cells in glioblastoma. Stem Cells 2007; 25: 2524–33.
Barnes DJ, Melo JV. Primitive, quiescent and difficult to kill: the role of
non-proliferating stem cells in chronic myeloid leukemia. Cell Cycle
2006; 5: 2862–6.
Beier D, Hau P, Proescholdt M, Lohmeier A, Wischhusen J, Oefner PJ,
et al. CD133+ and CD133- glioblastoma-derived cancer stem cells
show differential growth characteristics and molecular profiles.
Cancer Res 2007; 67: 4010–5.
Bidlingmaier S, Zhu X, Liu B. The utility and limitations of glycosylated
human CD133 epitopes in defining cancer stem cells. J Mol Med
2008; 86: 1025–32.
Bleau AM, Hambardzumyan D, Ozawa T, Fomchenko EI, Huse JT,
Brennan CW, et al. PTEN/PI3K/Akt pathway regulates the side popu-
lation phenotype and ABCG2 activity in glioma tumour stem-like cells.
Cell Stem Cell 2009; 4: 226–35.
Blow JJ, Hodgson B. Replication licensing–defining the proliferative state?
Trends Cell Biol 2002; 12: 72–8.
Brennan K, Offiah G, McSherry EA, Hopkins AM. Tight junctions: a
barrier to the initiation and progression of breast cancer? J Biomed
Biotechnol 2009; 2010: 460607.
Chen R, Nishimura MC, Bumbaca SM, Kharbanda S, Forrest WF,
Kasman IM, et al. A hierarchy of self-renewing tumour-initiating cell
types in glioblastoma. Cancer Cell 2010; 17: 362–75.
Cheng JX, Liu BL, Zhang X. How powerful is CD133 as a cancer stem cell
marker in brain tumours? Cancer Treat Rev 2009; 35: 403–8.
Collins AT, Berry PA, Hyde C, Stower MJ, Maitland NJ. Prospective iden-
tification of tumourigenic prostate cancer stem cells. Cancer Res 2005;
65: 10946–51.
Deleyrolle LP, Reynolds BA. Isolation, expansion, and differentiation of
adult mammalian neural stem and progenitor cells using the neuro-
sphere assay. Methods Mol Biol 2009; 549: 91–101.
Dembinski JL, Krauss S. Characterization and functional analysis of a slow
cycling stem cell-like subpopulation in pancreas adenocarcinoma. Clin
Exp Metastasis 2009; 26: 611–23.
Eramo A, Lotti F, Sette G, Pilozzi E, Biffoni M, Di Virgilio A, et al.
Identification and expansion of the tumourigenic lung cancer stem
cell population. Cell Death Differ 2008; 15: 504–14.
1342 | Brain 2011: 134; 1331–1343 L. P. Deleyrolle et al.
Page 13
Galli R, Binda E, Orfanelli U, Cipelletti B, Gritti A, De Vitis S, et al.
Isolation and characterization of tumourigenic, stem-like neural precur-
sors from human glioblastoma. Cancer Res 2004; 64: 7011–21.
Golmohammadi MG, Blackmore DG, Large B, Azari H, Esfandiary E,
Paxinos G, et al. Comparative analysis of the frequency and distribu-
tion of stem and progenitor cells in the adult mouse brain. Stem Cells
2008; 26: 979–87.
Graham SM, Jørgensen HG, Allan E, Pearson C, Alcorn MJ, Richmond L,
et al. Primitive, quiescent, Philadelphia-positive stem cells from patients
with chronic myeloid leukemia are insensitive to STI571 in vitro. Blood
2002; 99: 319–25.
Griguer CE, Oliva CR, Gobin E, Marcorelles P, Benos DJ, Lancaster JR Jr,
et al. CD133 is a marker of bioenergetic stress in human glioma. PLoS
ONE 2008; 3: e3655.
Holyoake T, Jiang X, Eaves C, Eaves A. Isolation of a highly quiescent
subpopulation of primitive leukemic cells in chronic myeloid leukemia.
Blood 1999; 94: 2056–64.
Holyoake TL, Jiang X, Jorgensen HG, Graham S, Alcorn MJ, Laird C,
et al. Primitive quiescent leukemic cells from patients with chronic
myeloid leukemia spontaneously initiate factor-independent growth
in vitro in association with up-regulation of expression of interleukin-3.
Blood 2001; 97: 720–8.
Hu Y, Smyth GK. ELDA: extreme limiting dilution analysis for comparing
depleted and enriched populations in stem cell and other assays.
J Immunol Methods 2009; 347: 70–8.
Ignatova TN, Kukekov VG, Laywell ED, Suslov ON, Vrionis FD,
Steindler DA. Human cortical glial tumours contain neural stem-like
cells expressing astroglial and neuronal markers in vitro. Glia 2002;
39: 193–206.
Joo KM, Kim SY, Jin X, Song SY, Kong DS, Lee JI, et al. Clinical and
biological implications of CD133-positive and CD133-negative cells in
glioblastomas. Lab Invest 2008; 88: 808–15.
Kast RE, Belda-Iniesta C. Suppressing glioblastoma stem cell function by
aldehyde dehydrogenase inhibition with chloramphenicol or disulfiram
as a new treatment adjunct: an hypothesis. Curr Stem Cell Res Ther
2009; 4: 314–7.
Kelly JJ, Stechishin O, Chojnacki A, Lun X, Sun B, Senger DL, et al.
Proliferation of human glioblastoma stem cells occurs independently
of exogenous mitogens. Stem Cells 2009; 27: 1722–33.
Kondo T, Setoguchi T, Taga T. Persistence of a small subpopulation of
cancer stem-like cells in the C6 glioma cell line. Proc Natl Acad Sci
USA 2004; 101: 781–6.
Krishnamurthy K, Wang G, Rokhfeld D, Bieberich E. Deoxycholate pro-
motes survival of breast cancer cells by reducing the level of
pro-apoptotic ceramide. Breast Cancer Res 2008; 10: R106.
Laks DR, Masterman-Smith M, Visnyei K, Angenieux B, Orozco NM,
Foran I, et al. Neurosphere formation is an independent predictor of
clinical outcome in malignant glioma. Stem Cells 2009; 27: 980–7.
Lapidot T, Sirard C, Vormoor J, Murdoch B, Hoang T, Caceres-Cortes J,
et al. A cell initiating human acute myeloid leukaemia after transplant-
ation into SCID mice. Nature 1994; 367: 645–8.
Lee J, Kotliarova S, Kotliarov Y, Li A, Su Q, Donin NM, et al. Tumour
stem cells derived from glioblastomas cultured in bFGF and EGF more
closely mirror the phenotype and genotype of primary tumours than
do serum-cultured cell lines. Cancer Cell 2006; 9: 391–403.
Lei M. The MCM complex: its role in DNA replication and implications
for cancer therapy. Curr Cancer Drug Targets 2005; 5: 365–80.
Li C, Heidt DG, Dalerba P, Burant CF, Zhang L, Adsay V, et al.
Identification of pancreatic cancer stem cells. Cancer Res 2007; 67:
1030–7.
Louis DN, Ohgaki H, Wiestler OD, Cavenee WK, Burger PC, Jouvet A,
et al. The 2007 WHO classification of tumours of the central nervous
system. Acta Neuropathol 2007; 114: 97–109.
Lyons AB. Analysing cell division in vivo and in vitro using flow cyto-
metric measurement of CFSE dye dilution. J Immunol Methods 2000;
243: 147–54.
Nishide K, Nakatani Y, Kiyonari H, Kondo T. Glioblastoma formation
from cell population depleted of Prominin1-expressing cells. PLoS
ONE 2009; 4: e6869.
Ogden AT, Waziri AE, Lochhead RA, Fusco D, Lopez K, Ellis JA, et al.
Identification of A2B5+CD133- tumour-initiating cells in adult human
gliomas. Neurosurgery 2008; 62: 505–14; discussion 514–5.
Patrawala L, Calhoun T, Schneider-Broussard R, Zhou J, Claypool K,
Tang DG. Side population is enriched in tumourigenic, stem-like
cancer cells, whereas ABCG2+ and ABCG2- cancer cells are similarly
tumourigenic. Cancer Res 2005; 65: 6207–19.
Pece S, Tosoni D, Confalonieri S, Mazzarol G, Vecchi M, Ronzoni S, et al.
Biological and molecular heterogeneity of breast cancers correlates
with their cancer stem cell content. Cell 2010; 140: 62–73.
Piccirillo SG, Reynolds BA, Zanetti N, Lamorte G, Binda E, Broggi G, et al.
Bone morphogenetic proteins inhibit the tumourigenic potential of
human brain tumour-initiating cells. Nature 2006; 444: 761–5.
Qian BZ, Pollard JW. Macrophage diversity enhances tumour progression
and metastasis. Cell 2010; 141: 39–51.
Rasper M, Schafer A, Piontek G, Teufel J, Brockhoff G, Ringel F, et al.
Aldehyde dehydrogenase 1 positive glioblastoma cells show brain
tumour stem cell capacity. Neuro Oncol 2010; 12: 1024–33.
Reya T, Morrison SJ, Clarke MF, Weissman IL. Stem cells, cancer, and
cancer stem cells. Nature 2001; 414: 105–11.
Ricci-Vitiani L, Lombardi DG, Pilozzi E, Biffoni M, Todaro M, Peschle C,
et al. Identification and expansion of human colon-cancer-initiating
cells. Nature 2007; 445: 111–5.
Roesch A, Fukunaga-Kalabis M, Schmidt EC, Zabierowski SE,
Brafford PA, Vultur A, et al. A temporarily distinct subpopulation of
slow-cycling melanoma cells is required for continuous tumour growth.
Cell 2010; 141: 583–94.
Semple JW, Duncker BP. ORC-associated replication factors as biomark-
ers for cancer. Biotechnol Adv 2004; 22: 621–31.
Shackleton M, Quintana E, Fearon ER, Morrison SJ. Heterogeneity in
cancer: cancer stem cells versus clonal evolution. Cell 2009; 138:
822–9.
Singh SK, Clarke ID, Terasaki M, Bonn VE, Hawkins C, Squire J, et al.
Identification of a cancer stem cell in human brain tumours. Cancer
Res 2003; 63: 5821–8.
Singh SK, Hawkins C, Clarke ID, Squire JA, Bayani J, Hide T, et al.
Identification of human brain tumour initiating cells. Nature 2004;
432: 396–401.
Son MJ, Woolard K, Nam DH, Lee J, Fine HA. SSEA-1 is an enrichment
marker for tumour-initiating cells in human glioblastoma. Cell Stem
Cell 2009; 4: 440–52.
Suetsugu A, Nagaki M, Aoki H, Motohashi T, Kunisada T, Moriwaki H.
Characterization of CD133+ hepatocellular carcinoma cells as cancer
stem/progenitor cells. Biochem Biophys Res Commun 2006; 351:
820–4.
Suva ML, Riggi N, Stehle JC, Baumer K, Tercier S, Joseph JM, et al.
Identification of cancer stem cells in Ewing’s sarcoma. Cancer Res
2009; 69: 1776–81.
Tchoghandjian A, Baeza N, Colin C, Cayre M, Metellus P, Beclin C, et al.
A2B5 cells from human glioblastoma have cancer stem cell properties.
Brain Pathol 2010; 20: 211–21.
Vescovi AL, Galli R, Reynolds BA. Brain tumour stem cells. Nat Rev
Cancer 2006; 6: 425–36.
Wang J, Sakariassen PO, Tsinkalovsky O, Immervoll H, Boe SO,
Svendsen A, et al. CD133 negative glioma cells form tumours in
nude rats and give rise to CD133 positive cells. Int J Cancer 2008;
122: 761–8.
Witz IP. The tumour microenvironment: the making of a paradigm.
Cancer Microenviron 2009; 2 (Suppl 1): 9–17.
Wong ML, Prawira A, Kaye AH, Hovens CM. Tumour angiogenesis: its
mechanism and therapeutic implications in malignant gliomas. J Clin
Neurosci 2009; 16: 1119–30.
Zhou BB, Anderson HJ, Roberge M. Targeting DNA checkpoint kinases in
cancer therapy. Cancer Biol Ther 2003; 2: S16–22.
Label-retaining glioma-initiating cells Brain 2011: 134; 1331–1343 | 1343