Prospective Identification of Glioblastoma Cells
Generating Dormant Tumors
Ronit Satchi-Fainaro1., Shiran Ferber1., Ehud Segal1, Lili Ma2, Niharika Dixit2, Ambreen Ijaz2,
Lynn Hlatky2, Amir Abdollahi2,3, Nava Almog2*
1Department of Physiology and Pharmacology, Sackler School of Medicine, Tel-Aviv University, Tel-Aviv, Israel, 2Center of Cancer Systems Biology, Steward Research &
Specialty Projects Corp., St. Elizabeth’s Medical Center, Tufts University School of Medicine, Boston, Massachusetts, United States of America, 3Department of Radiation
Oncology, German Cancer Research Center and University of Heidelberg Medical School, Heidelberg, Germany
Although dormant tumors are highly prevalent within the human population, the underlying mechanisms are still mostly
unknown. We have previously identified the consensus gene expression pattern of dormant tumors. Here, we show that this
gene expression signature could be used for the isolation and identification of clones which generate dormant tumors. We
established single cell-derived clones from the aggressive tumor-generating U-87 MG human glioblastoma cell line. Based
only on the expression pattern of genes which were previously shown to be associated with tumor dormancy, we identified
clones which generate dormant tumors. We show that very high expression levels of thrombospondin and high expression
levels of angiomotin and insulin-like growth factor binding protein 5 (IGFBP5), together with low levels of endothelial
specific marker (ESM) 1 and epithelial growth factor receptor (EGFR) characterize the clone which generates dormant U-
87 MG derived glioblastomas. These tumors remained indolent both in subcutaneous and orthotopic intracranial sites, in
spite of a high prevalence of proliferating cells. We further show that tumor cells which form U-87 MG derived dormant
tumors have an impaired angiogenesis potential both in vitro and in vivo and have a slower invasion capacity. This work
demonstrates that fast-growing tumors contain tumor cells that when isolated will form dormant tumors and serves as a
proof-of-concept for the use of transcriptome profiles in the identification of such cells. Isolating the tumor cells that form
dormant tumors will facilitate understanding of the underlying mechanisms of dormant micro-metastases, late recurrence,
and changes in rate of tumor progression.
Citation: Satchi-Fainaro R, Ferber S, Segal E, Ma L, Dixit N, et al. (2012) Prospective Identification of Glioblastoma Cells Generating Dormant Tumors. PLoS
ONE 7(9): e44395. doi:10.1371/journal.pone.0044395
Editor: Javier S. Castresana, University of Navarra, Spain
Received February 16, 2012; Accepted August 3, 2012; Published September 6, 2012
Copyright: ? 2012 Satchi-Fainaro et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Authors wish to thank Janusz Weremowicz and Clare Lamont for assisting with animal work. The project described was supported (in part) by Award
Number U54CA149233 from the National Cancer Institute (LH). The content is solely the responsibility of the authors and does not necessarily represent the
official views of the National Cancer Institute or the National Institutes of Health. This study was supported (in part) by grant no. 5145–300000 from the Chief
Scientist Office of the Ministry of Health, Israel, by the Israel Science Foundation (grant No. 1309/10), The Swiss Bridge Award and The Israel Cancer Research Fund
given to RSF. The authors disclose any commercial affiliations or financial interests that may be considered conflict of interest regarding this manuscript. The
funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: email@example.com
. These authors contributed equally to this work.
A dormant phase during tumor progression is highly prevalent,
yet it is one of the most neglected areas in cancer research and the
associated biological mechanisms are still mostly unknown [1,2].
Cancer dormancy is a stage in which tumors are kept occult and
asymptomatic for a prolonged period of time [3,4]. It is present as
one of the earliest stages in tumor development, as micro-
metastasis in distant organs, and as minimal residual disease left
after surgical removal or treatment of primary tumors. Dormant
tumors are usually only a few millimeters diameter in size and are,
therefore, undetectable by most imaging technologies currently in
use [5,6]. They can, however, switch to become fast-growing,
clinically-apparent, and potentially lethal.
Since delayed disease recurrence, common in breast cancer,
colon cancer and other tumor types, can be explained by the
concept of tumor dormancy [7,8], eradicating dormant tumors is
currently a major challenge in cancer treatment [9–12]. Tumors
can remain occult and asymptomatic for years, or even decades,
while certain molecular and cellular mechanisms either halt, or are
insufficient to enable, tumor progression and mass expansion.
Clinical data and experimental models have led to the develop-
ment of the concepts of cellular dormancy [13–16] and tumor
dormancy [17–20]. Tumor cell dormancy is observed when
solitary disseminated cancer cells either circulate in the blood
system or settle at secondary sites, and is often associated with
quiescence. Whereas, tumor dormancy is observed when tumors,
as clusters of cells, do not expand in size over a long period of time.
Clearly, dormancy of cancerous lesions depends on crucial signals
from the microenvironment and the tumor stroma [4,16,18,21–
28]. Such signals can induce tumor cell quiescence. Alternatively,
systemic influences – such as the immune system of the host,
hormonal control, or the blockage or insufficiency of tumor
angiogenesis potential – can result in dormant tumors in which cell
proliferation is balanced by cell death.
PLOS ONE | www.plosone.org1September 2012 | Volume 7 | Issue 9 | e44395
A lack of suitable experimental models and limited clinical
access to dormant tumors are two of the major obstacles in the
advancement of research on tumor dormancy . We have
previously established in vivo models of human breast cancer,
glioblastoma, osteosarcoma, and liposarcoma dormancy in severe
combined immunodeficient (SCID) mice [30,31]. These models
were all derived from human tumor cell lines isolated from cancer
patients and no artificial genetic modifications were made to
generate the cell lines that form dormant or fast-growing tumors
when injected into SCID mice. Tumor dormancy in these models
was associated with an impaired angiogenic potential resulting in a
delayed expansion of tumor mass. A high proliferation rate of
tumor cells in dormant tumors is balanced by apoptosis and cell
death. Using these models, we have shown that viable and
metabolically-active, non-angiogenic, microscopic dormant tu-
mors can reside in mice for very long periods of time until they
spontaneously switch to become fast-growing, angiogenic tumors
Next, we sought to identify the molecular determinants of
human tumor dormancy. Using genome-wide expression profiling
assays to compare gene expression profiles in dormant and fast-
growing tumors from our human breast cancer, glioblastoma,
osteosarcoma, and liposarcoma models, we looked for genes with
similar patterns of expression across all tumor types. The
consensus signature of human tumor dormancy was then
determined based on genes that were differentially expressed
between dormant and fast-growing tumors in the same pattern in
all tumor types analyzed . For example: in all dormant tumors,
high expression of thrombospondin and angiomotin with con-
comitant low expression of CD73 and epidermal growth factor
receptor (EGFR) were observed.
Tumor cells are well known to be heterogeneous with respect to
a wide variety of characteristics such as metastatic activity,
angiogenic potential, proliferation rate, and enzymatic activity
. Here, we set out to test whether the tumor dormancy gene
signature that we previously identified can be used for isolation of
tumor cells that will form non-angiogenic dormant tumors. Hence,
this approach can lead to further and deeper understanding of the
molecular mechanisms underlying human tumor dormancy.
Single cell derived clones were generated using a limiting
dilution method from the parental U-87 MG human glioblastoma
cell line. Thirteen clones were chosen according to similar rapid
kinetics of colony formation in tissue culture wells. RNA was
extracted from each clone and the relative expression level of
Thrombospondin (TSP-1), a well-known endogenous inhibitor of
angiogenesis that has been shown to be elevated in all dormant
tumors , was determined using real time PCR (Fig. 1A). When
compared with the expression level of the parental U-87 MG cell
line, most (10 out of 13) of the clones had lower TSP-1 expression,
while only 3 clones (#1, #2 and #6) had elevated TSP levels.
Clone #1 had a significant increase in TSP level (over 25-fold
higher expression than in parental U-87 MG cell line).
Since a high TSP level could suggest slow kinetics of tumor
growth, we chose to focus our analysis on three clones with varying
TSP levels: Clone #1, with the highest TSP level, Clone #6 with
an intermediate TSP level, and Clone #7 with a very low TSP
level (marked by arrows in Fig. 1A). We postulated that Clone #1
might generate dormant tumors, whereas clones #6 and #7
would generate intermediate or fast-growing tumors, respectively.
We then continued to analyze the expression levels of several
tumor dormancy-associated genes we had previously identified
. First, we tested the expression levels in the three clones of the
additional genes that were previously shown to be upregulated in
dormant tumors (Fig. 1B). Clearly, angiomotin (Amot) and
IGFBP5 levels were upregulated only in Clone #1. Notably,
IGFBP5 expression was around 1000-fold higher than in the
parental U-87 MG cell line. Expression of TGF-b2 was upregu-
lated in all clones tested.
Genes previously shown to be elevated in fast-growing tumors
were expected to be observed as downregulated in tumor cells that
form dormant or slow-growing tumors. Such downregulation was
indeed observed in Clone #1 for CD73, EGFR, and most
significantly for ESM-1 (Fig. 1B), strengthening our prediction that
this clone could generate dormant tumors.
Tumor growth patterns were then analyzed in SCID mice.
Equal numbers of cells were injected subcutaneously (s.c.) from
each clone and from the parental U-87 MG cell line, and tumor
growth was monitored (Fig. 2A). As expected, the parental U-
87 MG cells generated very small tumors (volume below
100 mm3), which after 3–4 weeks initiated rapid growth. Similar
‘bi-phasic’ growth kinetics were observed for tumors generated
from clones #6 and #7. Although Clone #6, which had an
intermediate level of TSP, initially formed tumors larger than the
parental cell line, its tumors grew slower in the rapid growth phase.
Clone #7, which had a very low level of TSP, formed tumors
smaller than those generated by the parental cell line or by Clone
#6. Importantly, Clone #1 formed dormant tumors which
remained indolent and were barely detectable by gross examina-
tion throughout the experiment (Fig. 2B). This confirmed our
hypothesis that the parental U-87 MG cell line contains cells
which when isolated will form dormant tumors, and that Clone
#1 was generated from such cells.
At the end point of the experiment, tumors generated by clones
#6 and #7 were clearly smaller than those generated by the
parental U-87 MG cells. Although smaller in mass, Clone #6 and
Clone #7 tumors were highly vascularized and tightly capsulated,
similar to tumors generated from parental U-87 MG cells (Fig. 2C).
In contrast, tumors generated from Clone #1 could be detected
only after flipping the skin and seemed avascular. These tumors
were occasionally found attached to the muscle tissue instead of
the skin, like most tumors from U-87 MG, Clone #6, and Clone
#7 (Fig. 2C).
The fate of indolent tumors generated by Clone #1 was
analyzed by following their tumor growth over a prolonged period
of time lasting more than 200 days (Fig. S1). As expected, while U-
87 MG tumors grew rapidly in the first 3–4 weeks after
inoculation, tumors from Clone #1 remained undetectable for
over 70 days. Three of the four tumors from Clone #1 eventually
emerged from dormancy and initiated growth at 81, 122, and 127
days post inoculation (Fig. S1). One mouse injected with Clone #1
cells never developed any detectable tumors during the 270 days of
the experiment (data not shown). Tumors that originated from
Clone #1 cells remained at the site of injection in a constant small
size without expanding in mass for a long period of time (i.e.,
dormant). Importantly, once these tumors emerged from dorman-
cy and started growing, the growth rate could be as rapid as in the
parental U-87 MG cell line derived tumors.
To evaluate tumor properties in their orthotopic microenviron-
ment in a non-invasive manner, both Clone #1 and U-87 MG
parental cells were infected with mCherry as previously described
. Then, in order to assure that the infection did not alter tumor
characteristics, SCID mice were inoculated s.c. with either cell
line, and dormancy periods were monitored and compared.
Tumors generated by cells from Clone #1 remained dormant and
avascular for more than 70 days, while tumors generated from the
Gene Expression Pattern of Dormant Glioblastomas
PLOS ONE | www.plosone.org2September 2012 | Volume 7 | Issue 9 | e44395
Intravital Non-invasive Imaging of mCherry-labeled U-
87 MG and Clone #1 Human Glioblastoma Tumor-
CRI MaestroTMnon-invasive fluorescence imaging system was
used to follow tumor progression of 6-week-old male SCID mice
bearing either mCherry labeled U-87 MG or Clone #1 human
glioblastoma tumors. Tumor progression was validated by caliper
measurement (width26 length 6 0.52). Body weight and tumor
size were also monitored q.o.d. (n=3 mice/group). Mice were
anesthetized using ketamine (100 mg/kg) and xylazine (12 mg/
kg), treated with a depilatory cream (VeetH) and placed inside the
imaging system. Multispectral image-cube were acquired through
550–800 nm spectral range in 10 nm steps using excitation
(595 nm longpass) and emission (645 nm longpass) filter set. Mice
autofluorescence and undesired background signals were elimi-
nated by spectral analysis and linear unmixing algorithm.
Intravital High-resolution Fibered Confocal
Endomicroscopy of Tumor Vasculature
Endomicroscopy imaging was performed using a minimally
invasive high-resolution fibered confocal microscope (CellVizioH).
Mice were anesthetized and a conjugate of Dextran-FITC
(70 kDa) was injected into the tail vein. Blood vessels were imaged
by using a 4.2 mm-diameter optic fiber (ProFlexTMMiniO/30)
and all recordings were obtained in real time.
Microbubbles Contrast-enhanced Ultrasound Analysis of
Ultrasound (US) imaging was performed using a Vevo2100
(VisualSonics Inc, Toronto, Canada) using the 55 MHz 708
probe. Mice were anesthetized and hair was removed over the
tumor. Non-targeted microbubbles (VisualSonics Inc., Toronto,
Canada) were mixed with saline and injected into the tail vein of
For US imaging of size-matched tumors: Contrast-enhanced
US imaging cine loop of the tumors was acquired using the
contrast scans, immediately following administration of the
microbubbles. Tumor regions that are not perfused are easily
discriminated from those receiving blood flow (highlighted in
green) using the Vevo2100 (Visual Sonics) imaging software.
Perfusion curve was calculated using the following formula:
where y = Contrast signal (pixel intensity); A = Peak of curve;
B = Slope of the curve; C = Contrast signal offset; t = Time and
t0= Time offset. Blood flow within U-87 MG and Clone #1
tumors was compared using the maximum and minimum
difference in microbubbles signal intensity.
For US imaging of MatrigelH plugs containing conditioned
media from the cancer cells: 3D contrast-enhanced cine loop of
the tumor was acquired using the 3D acquisition motor
immediately after injection of the microbubbles. Following
microbubbles destruction, a second 3D contrast cine loop was
taken. The difference in video intensity from subtraction of the
pre- and post-destruction image frames was automatically
displayed by the software as a colored (green) overlay on the
gray-scale images. Images were analyzed offline for 3D relative
blood volume (PA) inside the tumor using the Vevo2100 imaging
Immunohistochemistry of tumor nodules was performed using
6 mm thick formalin-fixed, paraffin-embedded tissue sections.
Paraffin sections were deparaffinized, rehydrated, and stained by
hematoxylin and eosin (H&E). For CD34 staining, slides were
deparaffinized and pre-treated with 10 mM citrate, pH 6.0 for
Figure 8. Tumor growth patterns of mixed cancer cell populations. Growth of tumors generated by only U-87 MG or Clone #1 cells was
compared to that generated by mixing U-87 MG and Clone #1 cell lines in ratio of 1:1, 1:10, and 1:100 (U-87 MG: Clone #1). Tumor size (3–5 mice per
group) was measured by caliper.
Gene Expression Pattern of Dormant Glioblastomas
PLOS ONE | www.plosone.org 13September 2012 | Volume 7 | Issue 9 | e44395
50 min in a steam pressure cooker (Decloaking Chamber, BioCare
Medical, Walnut Creek, CA, USA). All further steps were
performed at RT in a hydrated chamber. Slides were covered
with Peroxidase Block (Merck, Germany) for 5 min to quench
endogenous peroxidase activity, followed by blockage of nonspe-
cific binding sites. Blocking was performed by incubation with
10% of rabbit serum in 50 mM Tris-HCl, pH 7.4, for 30 min or
10% horse serum, 1% BSA, 0.1% triton-X and 0.05% tween-20 in
PBS, pH 7.2, for 30 min. Primary rat anti-murine CD34 antibody
(MEC 14.7 1:50 dilution; Abcam, Cambridge, MA, USA) and
mouse anti TSP-1 antibody (1:15 dilution; Abcam, Cambridge,
MA, USA) were applied in 1% rabbit and goat serum respectively
in Tris-HCl, pH 7.4 at RT for 1 hour. Slides were washed in
50 mM Tris-HCl, pH 7.4 and rabbit anti-rat antibody (1:750
dilution; Vector Laboratories, CA, USA) was applied for 30 min,
followed by anti-rabbit horseradish peroxidase-conjugate antibody
(ABC detection kit, Vector Laboratories, CA, USA). Following
further washing, immunoperoxidase staining was developed using
ImmPACTTMDAB diluent kit (Vector Laboratories, CA, USA)
per the manufacturer’s instructions and counterstained with
methyl green. Microvessel density (MVD) was calculated as
previously described . Biotin anti-human PCNA antibody
(Biolegend, San Diego, CA, USA); HRP streptavidin (Biocare
medical, Concord, CA, USA). For fluorescent CD34 staining, Cy3
goat anti-rat antibody (1:100 dilution; Invitrogen) was applied for
2 hours, followed by anti-fade mounting solution containing DAPI
Data were expressed as mean 6 s.d. for in vitro assays or 6
s.e.m. for in vivo. Statistical significance was determined using an
unpaired t-test. All statistical tests were two-sided. All in vitro
experiments were performed in triplicates and repeated at least
87 MG parental cell line and by Clone #1. Each line
represents one tumor. Red lines indicate tumors generated from
U-87 MG cell line. Blue lines indicate tumors generated from
CD34 staining of size-matched (, ,2 mm3) U-
87 MG and Clone #1 tumor-sections, divided into
Growth kinetics of tumors generated by U-
Authors wish to thank Janusz Weremowicz and Clare Lamont for assisting
with animal work and to Melissa Klumpar for editing the manuscript.
Conceived and designed the experiments: SF RSF ES AA NA. Performed
the experiments: SF RSF ES LM ND AI LH. Analyzed the data: SF RSF
ES LM ND AI LH NA. Contributed reagents/materials/analysis tools:
RSF AA NA. Wrote the paper: SF RSF NA.
1. Folkman J, Kalluri R (2004) Cancer without disease. Nature 427: 787.
2. Almog N (2010) Molecular mechanisms underlying tumor dormancy. Cancer
3. Hart IR (1999) Perspective: tumour spread–the problems of latency. J Pathol
4. Bissell MJ, Hines WC (2011) Why don’t we get more cancer? A proposed role of
the microenvironment in restraining cancer progression. Nat Med 17: 320–329.
5. Black WC, Welch HG (1993) Advances in diagnostic imaging and overestima-
tions of disease prevalence and the benefits of therapy. N Engl J Med 328: 1237–
6. Harach HR, Franssila KO, Wasenius VM (1985) Occult papillary carcinoma of
the thyroid. A ‘‘normal’’ finding in Finland. A systematic autopsy study. Cancer
7. Goss PE, Chambers AF (2010) Does tumour dormancy offer a therapeutic
target? Nat Rev Cancer 10: 871–877.
8. Brackstone M, Townson JL, Chambers AF (2007) Tumour dormancy in breast
cancer: an update. Breast Cancer Res 9: 208.
9. Ribatti D, Nico B, Crivellato E, Roccaro AM, Vacca A (2007) The history of the
angiogenic switch concept. Leukemia 21: 44–52.
10. Aguirre-Ghiso JA (2007) Models, mechanisms and clinical evidence for cancer
dormancy. Nat Rev Cancer 7: 834–846.
11. Pantel K, Alix-Panabieres C, Riethdorf S (2009) Cancer micrometastases. Nat
Rev Clin Oncol 6: 339–351.
12. Uhr JW, Scheuermann RH, Street NE, Vitetta ES (1997) Cancer dormancy:
opportunities for new therapeutic approaches. Nat Med 3: 505–509.
13. Naumov GN, MacDonald IC, Chambers AF, Groom AC (2001) Solitary cancer
cells as a possible source of tumour dormancy? Semin Cancer Biol 11: 271–276.
14. Townson JL, Chambers AF (2006) Dormancy of solitary metastatic cells. Cell
Cycle 5: 1744–1750.
15. Chambers AF, Groom AC, MacDonald IC (2002) Dissemination and growth of
cancer cells in metastatic sites. Nat Rev Cancer 2: 563–572.
16. Barkan D, El Touny LH, Michalowski AM, Smith JA, Chu I, et al. (2010)
Metastatic growth from dormant cells induced by a col-I-enriched fibrotic
environment. Cancer Res 70: 5706–5716.
17. Wikman H, Vessella R, Pantel K (2008) Cancer micrometastasis and tumour
dormancy. APMIS 116: 754–770.
18. Holmgren L, O’Reilly MS, Folkman J (1995) Dormancy of micrometastases:
balanced proliferation and apoptosis in the presence of angiogenesis suppression.
Nat Med 1: 149–153.
19. Achilles EG, Fernandez A, Allred EN, Kisker O, Udagawa T, et al. (2001)
Heterogeneity of angiogenic activity in a human liposarcoma: a proposed
mechanism for ‘‘no take’’ of human tumors in mice. J Natl Cancer Inst 93:
20. Udagawa T, Fernandez A, Achilles EG, Folkman J, D’Amato RJ (2002)
Persistence of microscopic human cancers in mice: alterations in the angiogenic
balance accompanies loss of tumor dormancy. FASEB J 16: 1361–1370.
21. Gimbrone MA, Jr., Leapman SB, Cotran RS, Folkman J (1972) Tumor
dormancy in vivo by prevention of neovascularization. J Exp Med 136: 261–276.
22. Folkman J, Watson K, Ingber D, Hanahan D (1989) Induction of angiogenesis
during the transition from hyperplasia to neoplasia. Nature 339: 58–61.
23. Almog N, Ma L, Raychowdhury R, Schwager C, Erber R, et al. (2009)
Transcriptional switch of dormant tumors to fast-growing angiogenic phenotype.
Cancer Res 69: 836–844.
24. Quesnel B (2008) Dormant tumor cells as a therapeutic target? Cancer Lett 267:
25. Moserle L, Amadori A, Indraccolo S (2009) The angiogenic switch: implications
in the regulation of tumor dormancy. Curr Mol Med 9: 935–941.
26. Koebel CM, Vermi W, Swann JB, Zerafa N, Rodig SJ, et al. (2007) Adaptive
immunity maintains occult cancer in an equilibrium state. Nature 450: 903–907.
27. Kenny PA, Bissell MJ (2003) Tumor reversion: correction of malignant behavior
by microenvironmental cues. Int J Cancer 107: 688–695.
28. Barkan D, Green JE, Chambers AF (2010) Extracellular matrix: a gatekeeper in
the transition from dormancy to metastatic growth. Eur J Cancer 46: 1181–
29. Uhr JW, Pantel K Controversies in clinical cancer dormancy. Proc Natl Acad
Sci U S A.
30. Almog N, Henke V, Flores L, Hlatky L, Kung AL, et al. (2006) Prolonged
dormancy of human liposarcoma is associated with impaired tumor angiogen-
esis. FASEB J 20: 947–949.
31. Naumov GN, Bender E, Zurakowski D, Kang SY, Sampson D, et al. (2006) A
model of human tumor dormancy: an angiogenic switch from the nonangiogenic
phenotype. J Natl Cancer Inst 98: 316–325.
32. Segal E, Pan H, Ofek P, Udagawa T, Kopeckova P, et al. (2009) Targeting
angiogenesis-dependent calcified neoplasms using combined polymer therapeu-
tics. PLoS One 4: e5233.
33. Kaur B, Cork SM, Sandberg EM, Devi NS, Zhang Z, et al. (2009) Vasculostatin
inhibits intracranial glioma growth and negatively regulates in vivo angiogenesis
through a CD36-dependent mechanism. Cancer Res 69: 1212–1220.
34. Weidner N, Semple JP, Welch WR, Folkman J (1991) Tumor angiogenesis and
metastasis–correlation in invasive breast carcinoma. N Engl J Med 324: 1–8.
Gene Expression Pattern of Dormant Glioblastomas
PLOS ONE | www.plosone.org14September 2012 | Volume 7 | Issue 9 | e44395