An Aberrant Transcription Factor
Network Essential for Wnt Signaling
and Stem Cell Maintenance in Glioblastoma
Esther Rheinbay,1,2,5,6,9Mario L. Suva `,1,2,5,9Shawn M. Gillespie,1,2,5Hiroaki Wakimoto,3Anoop P. Patel,3
Mohammad Shahid,8Ozgur Oksuz,2Samuel D. Rabkin,3Robert L. Martuza,3Miguel N. Rivera,2,5David N. Louis,2
Simon Kasif,6,7Andrew S. Chi,1,2,4,5,* and Bradley E. Bernstein1,2,5,*
1Howard Hughes Medical Institute, Chevy Chase, MD 20815, USA
2Department of Pathology and Center for Cancer Research
3Department of Neurosurgery
4Divisions of Neuro-Oncology and Hematology/Oncology and Department of Neurology
Massachusetts General Hospital and Harvard Medical School, Boston, MA 02114, USA
5Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA
7Department of Biomedical Engineering
Boston University, Boston, MA 02215, USA
8Department of Surgery, Massachusetts General Hospital, Boston, MA 02114, USA
9These authors contributed equally to this work
*Correspondence: email@example.com (A.S.C.), firstname.lastname@example.org (B.E.B.)
Glioblastoma (GBM) is thought to be driven by a sub-
population of cancer stem cells (CSCs) that self-
renew and recapitulate tumor heterogeneity yet
remain poorly understood. Here, we present a
comparative analysis of chromatin state in GBM
CSCs that reveals widespread activation of genes
normally held in check by Polycomb repressors.
These activated targets include a large set of de-
coordinated activation is unique to the CSCs. We
demonstrate that a critical factor in the set, ASCL1,
activates Wnt signaling by repressing the negative
regulator DKK1. We show that ASCL1 is essential
for the maintenance and in vivo tumorigenicity of
GBM CSCs. Genome-wide binding profiles for
ASCL1 and the Wnt effector LEF-1 provide mecha-
nistic insight and suggest widespread interactions
between the TF module and the signaling pathway.
Our findings demonstrate regulatory connections
among ASCL1, Wnt signaling, and collaborating
TFs that are essential for the maintenance and
tumorigenicity of GBM CSCs.
The importance of epigenetic regulation in cancer initiation and
progression is now well recognized (Baylin and Jones, 2011; Pu-
jadas and Feinberg, 2012). Aberrant DNA methylation patterns
and recurrent mutations in genes encoding chromatin-modifying
their lineage, differentiation stage, and cellular environment.
These transcriptional programs are driven by transcription
factors (TFs) that interact with regulatory sequences and by
proteins that modulate the chromatin state of specific loci. Tran-
scriptional and epigenetic programs can exhibit striking hetero-
geneity within a tumor and may distinguish cancer stem cells
these malignant programs and their variability remain poorly
Chromatin state maps provide a general and systematic
means for gauging the activity and epigenetic state of pro-
moters, genes, and other regulatory elements within a particular
cell type (Bell et al., 2011; Zhou et al., 2011). These maps are ac-
quired by coupling chromatin immunoprecipitation with deep
sequencing (ChIP-seq) and are typically applied to histone mod-
ifications that mark different types of functional sequence ele-
ments. Chromatin data can be integrated with TF recognition
motifs or binding profiles to gain more specific insight into the
regulators and pathways that activate these sequence elements
in specific cellular contexts (Dunham et al., 2012; Ernst et al.,
2011). Although these approaches have been applied to cancer
models to a limited extent (Baylin and Jones, 2011), their poten-
tial has yet to be explored systematically.
Glioblastoma (GBM) is the most common malignant brain tu-
mor in adults and is associated with poor prognosis despite
aggressive treatment. Transcriptional profiling studies have re-
vealed biologically relevant GBM subtypes associated with sur-
vival and response to therapy as well as specific dysregulated
cellular pathways (Huse and Holland, 2010). Furthermore, epige-
netic regulators, including the Polycomb repressors EZH2 and
BMI1, and specific DNA methylation changes have been linked
to disease pathology, prognosis, and therapeutic responses
(Suva ` et al., 2009; Bruggeman et al., 2007; Lee et al., 2008).
Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors 1567
Fraction of hypermethylated
probes in indicated tumor
NHA promoter state
H3K4me3 H3K4me1 H3K27me3H3K36me3
10 kb10 kb5 kb
Figure 1. Characterization of the GBM CSC Chromatin Landscape
(A) GBM CSCs used for this study grow as gliomaspheres in serum-free neurobasal media.
(B) FACS analysis of MGG8 GBM CSCs shows positivity for the GBM stem cell markers SSEA-1 and CD133.
(C) Mouse brain cross-section after orthotopic xenotransplantation of MGG8 GBM CSCs (left). Higher magnification of tumor tissue depicts cytonuclear pleo-
morphism, mitotic and apoptotic figures (center), and infiltration along white matter tracks (right).
(D) Schematic overview of the study strategy.
(E) Breakdown ofTSSchromatin state inNHA (H3K4me3only, H3K4me3+ H3K27me3,H3K27me3 only, neither mark) and theirconsensuschromatin state inthe
GBM CSCs (see Experimental Procedures). A large fraction (59%) of genes bivalent in NHA become activated (H3K4me3 only, green) in GBM CSCs.
(legend continued on next page)
1568 Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors
Recent work has also shown that epigenetic resetting by
induced pluripotent stem cell reprogramming followed by line-
age differentiation can override the malignant properties of
GBM (Stricker et al., 2013). However, little is known about the
control mechanisms that drive these transcriptional programs
and their contribution to the malignant phenotype in GBM.
Recent studies have documented subpopulations of GBM
cells with tumor-propagating capacity that are believed to
constitute the tumor’s driving force and to play a major role in tu-
mor recurrence and resistance to therapy (Bao et al., 2006; Chen
et al., 2012; Singh et al., 2004). This subpopulation can be pro-
spectively isolated ex vivo with specific cell-surface markers
(Singh et al., 2003; Son et al., 2009) or defined culture conditions
(Lee et al., 2006; Lottaz et al., 2010; Wakimoto et al., 2009) and
expanded in vitro as gliomaspheres. Upon serial xenotransplan-
tation, these cells initiate tumors that closely phenocopy the pa-
tient’s parental tumor both morphologically and genetically (Galli
et al., 2004; Singh et al., 2004; Wakimoto et al., 2009). Because
of their unlimited self-renewal capacity and their ability to seed
malignant tumors in vivo, this subpopulation satisfies major
criteria for cancer stem cells (CSCs) (Valent et al., 2012).
Here, we combined chromatin profiling, computational anal-
ysis, and directed cellular perturbations to characterize tran-
scriptional and epigenetic regulatory programs in GBM CSCs.
Comparative analysis of chromatin maps for GBM CSCs, differ-
entiated GBM cells, and nonmalignant neural cells reveals a
module of developmental TFs that is coordinately activated in
the CSCs. Functional analysis suggests that these TFs play
essential roles in GBM CSCs. In particular, we show that
ASCL1 directly activates Wnt signaling and is essential for
GBM CSC maintenance and in vivo tumorigenicity. Genome-
wide maps for ASCL1 and the Wnt effector lymphoid enhancer
binding factor (LEF-1) suggest specific mechanisms and wide-
spread interplay between the TF module and the signaling
pathway. Our findings thus provide global and mechanistic
insight into the regulatory programs that drive a CSC-like popu-
lation critical for GBM pathogenesis.
Comparative Epigenomic Analysis of GBM CSCs and
We focused our study on four GBM CSC lines derived from
different human tumors that were defined functionally through
kimoto et al., 2009, 2012). These lines grow as gliomaspheres
(Figure 1A; see Experimental Procedures) and express the
CSC cell surface markers CD133 and stage-specific embryonic
diate filament Nestin (Figure S1A). GBM CSCs show differentia-
tion potential toward the neuronal and astrocytic lineages, as
shown by microtubule-associated protein 2 (MAP-2) and glial fi-
brillary acidic protein (GFAP) expression, respectively (Figures
S1B). Orthotopic xenotransplantation of a limited number of
GBM CSCs leads to formation of tumors that recapitulate
GBM morphology with diffuse infiltration of the brain paren-
chyma (Figure 1C).
We systematically examined the transcriptional and epige-
netic landscapes of GBM CSCs by profiling histone H3 lysine 4
trimethylation (H3K4me3; a marker of transcriptional initiation),
lysine 36 trimethylation (H3K36me3; transcriptional elongation),
and H3 lysine 4 monomethylation (H3K4me1; candidate en-
hancers) (Figure 1D; Table S1) (Zhou et al., 2011). We also map-
ped H3 lysine 27 trimethylation (H3K27me3), a repressive mark
catalyzed by the Polycomb protein EZH2. For comparison, we
also profiled normal human astrocytes (NHAs) isolated from fetal
brain, embryonic stem cell (ESC)-derived neural stem cells
(NSCs; Figures S1C–S1F) (Conti et al., 2005), and previously
characterized serum-cultured GBM cell lines derived from the
same patient tumors as our GBM CSCs (Wakimoto et al.,
2009). These traditional GBM lines grow as adherent mono-
layers, do not express GBM CSC-surface markers, and fail to
initiate tumors upon orthotopic xenotransplantation in limiting
dilution assays (Figures S1G–S1J).
We initially focused our attention on the four GBM CSC lines
using the NHAs as normal comparators. For each cell type, we
classified over 20,000 gene promoters as ‘‘active’’ (only
H3K4me3 detected) or ‘‘repressed’’ (H3K27me3 detected, with
or without H3K4me3, or neither mark) (Figure S2A; Table S2).
H3K27me3-marked promoters are maintained in an inactive
state by Polycomb complexes, while promoters with neither
mark are frequently repressed through DNA hypermethylation
(Meissner et al., 2008). We compared assignments between
cell types in order to identify genes that are differentially regu-
lated in a majority of the GBM CSC lines relative to NHAs. Pro-
moters that are ‘‘active’’ in NHAs retain this state in GBM
CSCs in a vast majority of cases (n = 11,586; 94% of NHA
H3K4me3 genes) (Figure 1E), consistent with many of these be-
ing housekeeping genes (Mikkelsen et al., 2007). In contrast,
promoters that are ‘‘repressed’’ in NHAs frequently change their
chromatin state in GBM CSCs in the following manner.
Repressed promoters with H3K27me3 and H3K4me3 (‘‘biva-
lent’’) in NHAs predominantly lose H3K27me3 and become acti-
vated in GBM CSCs (n = 1,057; 59% of NHA bivalent promoters)
(Figure 1E). Repressed promoters with only H3K27me3 in NHAs
often lose H3K27me3 and switch to the unmarked state in GBM
CSCs (n = 591; 48% of NHA H3K27me3 genes). Repressed pro-
moters that are unmarked in NHAs tend to retain this state in
GBM CSCs (n = 5,110; 88%) (Figure 1E). Notably, an analogous
comparison of NSCs to GBM CSCs also revealed frequent
switching of loci from bivalent (or H3K27me3 only) in the nonma-
lignant cells to active/H3K4me3 in the CSCs (61%; Figure S2B).
In contrast, only 33% of genes with bivalent chromatin state in
NSCs switched to active/H3K4me3 in the nonmalignant NHAs
(F)FractionofDNAhypermethylated(bR0.75)probesinMGG4,MGG6,andMGG8GBMCSCscontingent onNHAchromatinstateoftheprobe. Probesmarked
with H3K27me3 only are twice as likely to become DNA methylated than those marked with both H3K4me3 and H3K27me3.
(G) Chromatin state of six CSC-TFs in one representative GBM CSC line, matched serum-grown GBM cell line, NSC, and NHA. All TFs are active (H3K4me3 at
promoter, H3K36me3 over the transcript) in GBM CSCs, but not in NHA or serum-grown GBM cells, as indicated by large domains of H3K27me3.
See also Figures S1 and S2.
Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors 1569
(Figure S2C). These statistics suggest that genes marked by
H3K27me3 in nonmalignant contexts frequently lose the Poly-
comb-associated mark and switch their chromatin state in
We next examined how the chromatin state transitions
correspond to changes in gene expression. As expected, genes
that switch from bivalent in NHA to H3K4me3-only in GBM CSC
are expressed at significantly higher levels in the malignant cell
lines (Figure S2D). In contrast, genes that switch from
H3K27me3-only in NHAs to unmarked in GBM CSCs show little
MGG23 CSC MGG4 CSC MGG6 CSCMGG8 CSC
MGG8 serum MGG4 serum MGG6 serum
sequence-specific DNA binding
cell projection organization
transcription factor activity
cell morphogenesis involved in differentiation
MGG23 CSCMGG4 CSCMGG6 CSCMGG8 CSC
Figure 2. Aberrant Activation and Repres-
sion of TF Polycomb Targets in GBM CSCs
(A) Representative top-scoring functional terms
enriched in genes active (H3K4me3 only) in GBM
CSCs but repressed in NHA (see also Table S3).
Scores are calculated based on Benjamini-Hoch-
berg corrected p values (see Experimental Pro-
aberrantly activated TF loci (?2.5kb to +2.5 kb
from TSS) for indicated cell types. Orange in-
dicates overlap of H3K4me3 and H3K27me3
signal (bivalent). Genes were clustered based on
(C) Microarray gene expression data for acti-
indicates low expression normalized by row. The
expression changes are consistent with the
chromatin changes, although the magnitude of
expression change across samples is more
(D) Chromatin state and gene expression data for
NHA-active, GBM-CSC-repressed TF loci. Color
scheme as in (B) and (C).
See also Figure S3.
ing that transcriptional
switch (Figure S2E). We therefore con-
sidered the possibility that these loci
may become DNA methylated in the
CSCs, as would be consistent with prior
reports of Polycomb targets becoming
hypermethylated in cancer (Ohm et al.,
2007; Schlesinger et al., 2007; Widsch-
wendter et al., 2007). To test this,
we profiled DNA methylation in three
GBM CSC lines (see Experimental Proce-
dures). We found that loci marked by
H3K27me3 only in the NHAs become
hypermethylated in the GBM CSCs at a
rate that is 2-fold higher than bivalent
loci (Figure 1F). Taken together, these
findings suggest that repressive Poly-
comb complexes are lost from a subset
of their target loci in GBM CSCs, with
initially bivalent genes undergoing tran-
scriptional activation and H3K27me3-only genes acquiring
Widespread TF Activation in GBM CSCs
To gain insight into the regulatory consequences of the chro-
matin state transitions, we examined the identities of genes
with active chromatin in GBM CSCs but repressed chromatin
in NHAs. Unbiased functional enrichment analysis (Dennis
et al., 2003) revealed a significant overrepresentation of terms
related to development and transcriptional regulation (Figure 2A;
1570 Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors
Table S3). In contrast, we failed to detect significant enrichment
for genes that were repressed in GBM CSCs but active in NHAs
(Table S3). We therefore focused our attention on the set of
developmental TFs (n = 90) that are differentially activated in
GBM CSCs (Table S4). These include factors previously associ-
ated with GBM such as SOX2 (Gangemi et al., 2009), OLIG2
(Ligon et al., 2004), HEY1 (Hulleman et al., 2009), and several
HOXgenes (Muratet al.,2008). However, mostof the TFs wede-
tected have not yet been described in this context. Although the
CSC lines used in this study were derived from different patients
and harbor different genetic aberrations (Wakimoto et al., 2012),
amajority of theimplicated TFsisactive in all four CSC lines (Fig-
ure 2B). Conventional messenger RNA (mRNA) microarray pro-
files confirm that these TFs are expressed at significantly higher
levels in the four CSCs than NHAs (Figure 2C). Consistent with
their role as developmental regulators, most of these TFs are
marked with bivalent chromatin or H3K27me3 in ESCs and
NSCs, as well as in primary nonmalignant brain sections (Fig-
ure S3) (Zhu et al., 2013). In contrast, we identified few TFs
(n = 16) that are selectively repressed in the CSCs (Figure 2D).
This suggests that a common set of TFs loses Polycomb repres-
sion and is induced in GBM CSCs.
To better understand their specificities, we also studied the
chromatin states of the activated TF genes in representative
cell models, including NHA, ESC-derived NSCs, the CSCs, and
serum-grown GBM cell lines. Clustering the TFs on the basis
of their promoter H3K27me3 signals distinguished four subsets:
(1) ‘‘cancer TFs’’ active in GBM CSCs and traditional GBM cell
lines, (2) ‘‘CSC TFs’’ exclusively active in GBM CSCs, (3)
‘‘stem-cell TFs’’ active in CSCs and NSCs, and (4) NHA-
repressed TFs active in all the other cell types (Figure 2B).
ASCL1 Induces Wnt Signaling in GBM CSCs
The distinctive chromatin patterns of the CSC TFs prompted us
to further explore their functional significance (Figures 1G and
2B). We specifically examined whether any of these TFs affect
signaling pathways that are essential for GBM CSC prolifera-
tion, such as Notch, Hedgehog, and Wnt (Clement et al.,
2007; Shih and Holland, 2006; Zheng et al., 2010). We ectopi-
cally expressed each CSC TF in NHAs and measured the
expression of canonical target genes for each of these path-
ways (Figure 3A).
We found that ASCL1 (also known as MASH1) induces AXIN2,
a canonical Wnt target (Figure 3A). ASCL1 mRNA is highly ex-
pressed in all four CSC cell lines but not in NHAs or NSCs (Fig-
ure 3B). To confirm the functional connection between ASCL1
and AXIN2, we depleted ASCL1 in GBM CSCs by small hairpin
RNA (shRNA)-mediated knockdown (Figure 3C), whereupon
We also used a Wnt-responsive luciferase reporter system to
induction of Wnt signaling and TCF/LEF transcriptional regula-
tion (Firestein et al., 2008; Veeman et al., 2003). Indeed, ASCL1
increased reporter gene expression >10-fold compared to a
control vector with mutated TCF/LEF sites (Figure 3E). ASCL1-
mediated induction of the reporter was dramatically enhanced
by costimulation with Wnt3a, suggesting a synergistic mode of
action with autocrine or paracrine Wnt stimulation. Taken
together, these results establish ASCL1 as a regulator of Wnt
signaling in GBM CSCs.
To assess the functional role of ASCL1 in CSC maintenance,
we performed single-cell sphere-formation assays of GBM
CSCs after shRNA-mediated ASCL1 knockdown (Clement
et al., 2007; Galli et al., 2004). Only 2% of ASCL1-depleted
CSCs retained the capacity to reform gliomaspheres after
14 days compared to nearly 20% of control cells (Figures 3F
and 3G). In addition, spheres generated by ASCL1 knockdown
topically injected with ASCL1-depleted GBM CSCs showed
prolonged survival compared to those xenotransplanted with
control CSCs (p < 0.01; Figure 3I). Thus, ASCL1 is essential for
maintenance and propagation of GBM CSCs.
We next examined ASCL1 expression in primary GBMs using
published gene expression profiles (Verhaak et al., 2010; Sun
et al., 2006). ASCL1 is expressed at markedly higher levels in tu-
mors relative to normal brain. However, its expression varies
significantly between molecular subtypes (Kruskal-Wallis p <
10?16), with significantly higher levels in proneural tumors (Fig-
ure 3K). ASCL1 also correlates with tumor grade, with increasing
proneural subtype (Figure 3L). The functional link between
expression across the GBM tumors (Pearson’s r = 0.52; p <
10?14; Figure 3J). Furthermore, flow cytometric staining in
acutely dissociated human GBM shows that ASCL1 is ex-
pressed in a subpopulation of cells with high levels of SOX2, a
marker for GBM subpopulations with CSCs properties (Fig-
ure 3M; Figure S4). This result is supported by in situ hybridiza-
tion (ISH) in human GBM samples where ASCL1 expression is
restricted to a subset of SOX2-positive cells (Figure 3N).
ASCL1 Promotes Wnt Signaling Directly by Repressing
To identify the direct regulatory targets of ASCL1, weused ChIP-
seq to map ASCL1 in MGG8 CSCs (see Experimental Proce-
dures; Figure S5). The mapped ASCL1 binding sites are highly
enriched for its cognate sequence motif and include the known
enhancers (marked by H3K4me1) in the vicinity of several genes
involved in Wnt regulation, including FZD5, DKK1, TCF7, and
TCF7L1 (Figure 4A). To test whether these candidates represent
direct functional targets, we ectopically expressed ASCL1 in
NHAs and measured changes in their expression. The effect of
ASCL1 on Wnt pathway genes was moderate with the exception
of 10-fold downregulation of DKK1 (Figure 4B). This is consistent
with a previous report of ASCL1 being a negative regulator of
DKK1 in small cell lung cancer cell lines (Osada et al., 2008). In
GBM CSCs, ASCL1 binds to an H3K4me1-marked regulatory
element located 5.7 kb upstream of the DKK1 transcription start
site (TSS) (Figure 4C), which adopts a chromatin configuration
do not express ASCL1, this element is also enriched for
H3K27ac and thus assumes an ‘‘active enhancer’’ state, consis-
tent with the high expression of DKK1 in these cells (Figure 4D).
To test whether repression of DKK1 is the primary mechanism
by which ASCL1 modulates Wnt signaling, we simultaneously
Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors 1571
Relative AXIN2 expression
Notch targetsShh targets
% Sphere formation
Fold expression difference
after expression of indicated TF
AXIN2 LEF1 TCF7 CCND1 HEY1 HEY2 HES1 GLI1 GLI2 GLI3
TCGA AXIN2 expression
TCGA ASCL1 expression
Relative ASCL1 expression
* * * *
TCGA ASCL1 expression
TCGA AXIN2 expression
Log10 ASCL1 expression
III IIIV (GBM)
p<2.5 × 10-7
Sun et al. ASCL1 expression
0 2040 6080100 120 140
Sphere diameter [µm]
p<8 × 10-6
p<4 × 10-17
Figure 3. ASCL1 Is an Upstream Regulator of the Wnt Pathway
(A) Relative expression of Wnt, Notch, and Shh targets after ectopic lentiviral expression of indicated TF in NHA measured by quantitative RT-PCR (qRT-PCR).
(B) mRNA levels of ASCL1 in NSCs, NHAs, and GBM CSCs measured on Affymetrix microarray (GSE46016).
(C) Relative levels of ASCL1 after shRNA-mediated knockdown in MGG4 CSCs by qRT-PCR.
(D) Repression of AXIN2 upon knockdown of ASCL1 in MGG4 CSCs (t test, p < 0.005).
(E) Schematic of Wnt activation experiment and relative luciferase expression for a TCF/LEF-responsive promoter (TOPFLASH-Firefly) relative to a scrambled
response element (FOPFLASH-Renilla) in 293T cells after lentiviral transfection with ASCL1 and addition of Wnt3a. *p < 0.05; **p < 0.01 (one-tailed t test).
(F and G) Individual examples (F) and quantification (G) of MGG4 CSC sphere-forming capacity in control and ASCL1-depleted cells.
(H) Quantification of sphere diameter in control and ASCL1-depleted cells.
(I) Kaplan-Meier survival curve for mice injected with 5,000 control (blue line) or ASCL1-depleted MGG4 CSCs (log-rank p value < 0.01).
(legend continued on next page)
1572 Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors
transfected DKK1 and ASCL1 into NHA and measured Wnt acti-
vation using the Wnt reporter. As shown in Figure 4E, exogenous
DKK1 expression completely abrogates the inducing effect of
ASCL1. These data support a model in which ASCL1 activates
Wnt signaling by repressing a regulatory element upstream of
the negative Wnt regulator DKK1. The association is also sup-
ported in clinical contexts by a significant anticorrelation in
ASCL1 and DKK1 expression across primary GBM tumors
(Pearson’s r = ?0.6; p < 10?20; Figure 4F). Given the depen-
dency of GBM CSCs on Wnt signaling (Zheng et al., 2010; Zhang
et al., 2011), this regulatory function may explain the critical role
of ASCL1 in GBM CSC maintenance.
LEF-1 Mediates Reciprocal Interactions between Wnt
Signaling and CSC TFs
To gain further insight into the Wnt signaling pathway in GBM
CSCs, we mapped LEF-1, a high-mobility group TF that regu-
lates Wnt targets (see Experimental Procedures; Figure S6).
We identified over 3,000 LEF-1 binding sites, which are highly
CCND1 (Clevers, 2006). The vast majority of these sites (82%)
reside outside of promoter regions, consistent with the original
identification of LEF-1 at an enhancer (Travis et al., 1991).
Notably, nearly half of the CSC and stem cell TF genes are prox-
imal to H3K4me1-marked candidate regulatory elements bound
by LEF-1 (Figure 5A; Table S5). To test whether these TFs are
downstream of the Wnt pathway, we measured their expression
in NHAs stimulated with ASCL1 complementary DNA (cDNA)
and Wnt3a protein. We detected significant induction of six
TFs, indicating their responsiveness to Wnt signaling (Figure 5B).
Taken together, these results suggest that crosstalk between a
network of activated TFs and Wnt signaling is critical for main-
taining GBM CSC regulatory programs (Figure 5C).
Cancer genome sequencing and complementary mechanistic
studies have accelerated our understanding of cancer genetics.
However, technical issues and the heterogeneity typical of
rations that contribute to tumor pathology. Here, we combined
recently established epigenomic technologies with an in vitro
the epigenetic and transcriptional programs that drive malignant
brain tumors. By comparing these GBM CSCs to serum-grown
GBM lines and nonmalignant neural cell models, we identify a
large network of TFs activated in the CSCs in combinations un-
likely to occur in normal physiologic contexts. Deregulation of
this network is associated with loss of H3K27me3 at TF pro-
moters and may reflect ineffective Polycomb repression in the
CSCs. We speculate that diminished epigenetic silencing and
by allowing GBM CSCs to respond to the varied requirements of
their malignant state, altered genetic makeup, and environment.
Among the CSC TFs, ASCL1 emerged as a potent upstream
regulator of the Wnt signaling pathway, which was recently
shown to be critical for GBM CSC maintenance (Zheng et al.,
2010; Zhang et al., 2011). ASCL1 overexpression induces Wnt
signaling in normal astrocytes, while its knockdown in GBM
CSCs markedly reduces activity of the pathway. We show that
Wnt activation is mediated through DKK1, a secreted Wnt inhib-
itor that is directly repressed by ASCL1. We also find evidence
that Wnt signaling feeds back upon the CSC TF genes via mul-
tiple LEF-1 target sites in their vicinity. The clinical relevance of
these interactions is supported by increased ASCL1 expression
in primary astrocytomas and GBMs and a correlation between
ASCL1 and the Wnt target gene AXIN2 across tumor samples.
In conclusion, we describe an aberrant epigenetic landscape
in GBM CSCs and the induction of a nonphysiologic TF module
and tumorigenicity. Our findings thus shed light on the regulatory
circuitry of this CSC model and propose specific factors and in-
teracting pathways as candidates for translational study.
Surgical specimens of GBM tumors were collected at Massachusetts General
Hospital with approval by the institutional review board (IRB). Mechanically
minced tissue was triturated and then cells were grown as gliomaspheres in
serum-free neural stem cell medium (Neurobasal medium [Invitrogen] sup-
plemented with 3 mmol/l L-glutamine [Mediatech], 13 B27 supplement
20 ng/ml recombinant human EGF [R&D Systems], 20 ng/ml recombinant hu-
man FGF2 [R&D Systems], and 0.53 penicillin G/streptomycin sulfate), as pre-
viously described (Wakimoto et al., 2009). Genomic copy number alterations
where (Wakimoto et al., 2012). From the same tumors, traditional GBM cells
lines, grown as adherent monolayer in Dulbecco’s modified Eagle’s medium
(DMEM) and 10% fetal calf serum (FCS) were derived as previously described
(Wakimoto et al., 2009). GBM CSC differentiation was induced using 5% FCS
and withdrawal of growth factors for 7 days on poly-L-ornithine- and laminin-
coated plates (see below for details). Staining was performed for nestin (Santa
Cruz, 1: 400), MAP-2 (Chemicon, 1: 150), and GFAP (Sigma, 1: 400).
Human-ESC-derived NSCs generated from H9 ES cells were obtained
from Millipore and grown and passaged in NSC medium consisting of a 1:1
mix of DMEM/F12:Neurobasal (Invitrogen), 0.53 N2 (Invitrogen), 0.53 B27
(J) Scatter plot shows correlation between AXIN2 and ASCL1 expression across 200 primary GBM samples (Verhaak et al., 2010). Each point denotes a single
(K)TCGAASCL1(left)andAXIN2(right)expression (Verhaaketal.,2010)correlateacrossmolecularsubtypes.Distributions forsubtypesaresignificantly different
from each other (Kruskal-Wallis p < 4 3 10?17[ASCL1] and p < 8 3 10?6[AXIN2]).
(L) ASCL1 expression (Sun et al., 2006) is increased in GBM, most strongly in the proneural subtype and lower-grade astrocytomas, relative to nonneoplastic
brain, suggesting that ASCL1 induction may be an early event in gliomagenesis.
(M)Intracellular staining and flowcytometricdetectionof thenuclearstemcellmarker SOX2and ASCL1inprimaryGBM.The population ofASCL1+ cells (23.8%)
is entirely contained within the SOX2+ compartment. Notably, the ASCL1+ subpopulation also displays highest levels of SOX2 expression.
(N) RNA-ISH for SOX2 (blue dots) and ASCL1 (red dots) in primary GBM shows expression of ASCL1 in a restricted subset of SOX2+ cells.
Error bars in qRT-PCR experiments indicate SEM.
See also Figure S4.
Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors 1573
mented with 20 ng/ml of both epidermal growth factor (EGF) and fibroblast
inin-coated plates. Poly-L-ornithine/laminin plates were generated as follows:
A 20 mg/ml solution of poly-L-ornithine (Sigma) in water was added to plates
and plates were incubated at 37?C for 1 hr. The poly-L-ornithine solution was
then removed, plates were washed three times with PBS, and 5 ml/ml solution
of laminin (Sigma) in PBS was added to plates and plates were incubated at
37?C for at least 3 hr. Cells were passaged using manual dissociation.
For differentiation into astrocytes, when cells were 80%–90% confluent,
the media was changed to NSC medium with 3% FCS and without EGF
or FGF-2. After 4 days, cells were fixed for immunofluorescence. For differ-
entiation into neurons, NSCs were grown to 90% confluency. Then, medium
was changed to either NSC medium but without N2 and supplemented
TCGA ASCL1 expression
TCGA DKK1 expression
Expression fold change
Figure 4. ASCL1 Regulates Wnt Signaling through DKK1
(A) ChIP-seq of ASCL1 in MGG8 CSCs (pink track) reveals enrichment at H3K4me1-marked distal elements in several Wnt pathway gene loci (gray shading).
(B) Relative mRNA level change for ASCL1-bound Wnt pathway genes in NHA after ectopic expression of ASCL1.
(C) ChIP-seq maps depict the chromatin environment of the DKK1 gene locus and the ASCL1-bound enhancer (gray shading) in MGG8 CSCs and NHA. In the
absence of ASCL1, the element is activated (as indicated by H3K27 acetylation in NHA) and DKK1 is expressed (increase in active marks H3K4me3, H3K4me1,
H3K27ac, and H3K36me3).
(D) Expression levels of DKK1 in NSCs, NHAs, and GBM CSCs measured by microarray.
(E) Expression changes of a TCF/LEF-responsive reporter (TOPFLASH-Firefly) relative to scrambled response elements (FOPFLASH-Renilla) in 293T cells after
stimulation with the indicated combinations of Wnt3a protein and lentivirally transfected ASCL1 and DKK1. DKK1 overexpression abrogates ASCL1-mediated
(F) DKK1 and ASCL1 expression patterns are inversely correlated in 200 GBM samples (Verhaak et al., 2010). Each point denotes a tumor sample.
Error bars in qRT-PCR experiments indicate SEM. See also Figure S5.
1574 Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors
with 13 B27 or ENStem-A neuronal differentiation medium (Chemicon)
supplemented with L-glutamine (2 mM). After 2 weeks, cells were fixed for
For immunofluorescence, cells were washed once with PBS and then fixed
with 4% paraformaldehyde for 30 min. Cells were washed three times with
PBS and blocked for 2 hr (5% normal goat serum, 0.3% Triton X-100, PBS).
Then primary antibodies in blocking solution were added and cells were incu-
bated overnight at 4?C. The next day, cells were washed twice with PBS and
added and cells were incubated for 2 hr at room temperature. Cells were then
washed three to five times with PBS and then counterstained with DAPI/
1XPBS solution. Primary antibodies include anti-nestin 1:500 (Chemicon Cat
SCR060), anti-Sox2 1:200 (Chemicon Cat SCR019), anti-BLBP 1:500 (Chem-
icon Cat AB9558), anti-GFAP (Chemicon Cat SCR019), anti-betaIII tubulin
1:500 (Chemicon Cat SCR060), and anti-MAP-2 1:200 (Chemicon SCR019).
Secondary antibodies were Alexa Fluor 488 (goat anti-rabbit immunoglobulin
G[IgG], Invitrogen CatA-11008)at1:200 andAlexaFluor 555(goatanti-mouse
IgG, Invitrogen Cat A-21422) at 1:200.
NHAswereobtained fromLonzaandpropagated accordingtothemanufac-
CD133 (Miltenyi Biotec CD133/1 PE cat # 130-080-801) and SSEA-1 (BD Bio-
sciences cat # 560127) antibodies were used according to the manufacturer’s
instructions. For TF staining in primary tumor, primary human glioblastomas
were obtained from patients operated at Massachusetts General Hospital in
accordance with an approved IRB protocol (2005-P-001609/16). Briefly,
tumors were dissociated to single-cell suspension using a brain tumor disso-
ciation kit (Miltenyi Biotec) depleted for CD45-positive immune cells using
microbeads and a MACS separator (Miltenyi Biotec). Cells were stained with
SOX2 (R&D Systems) and ASCL1 (BD Pharmingen) antibodies conjugated to
Alexa Fluor 647 or Alexa Fluor 488 using Alexa Fluor protein labeling kits (Invi-
(BD Biosciences) and analysis was performed withFlowJosoftware (Treestar).
ChIP assays were carried out on cultures of approximately 1 3 106cells per
sample and epitope, following the general procedures outlined previously
(Ku et al., 2008; Mikkelsen et al., 2007). Immunoprecipitation was performed
using antibodies against H3K4me3 (Millipore 07-473), H3K27me3 (Millipore
07449), H3K36me3 (Abcam 9050), H3K4me1 (Abcam 8895), ASCL1
Expression fold change
H3K4me3 H3K4me1 LEF1
GBM serum cells and NHA
Cancer stem cells
Figure 5. Crosstalk of the TF Module and Wnt Signaling in GBM CSCs
(A) ChIP-seq of LEF-1 (purple track) in MGG8 CSCs reveals enrichment at H3K4me1-marked distal elements near several TF loci.
(B) Relative mRNA level changes for Wnt-responsive TFs with distal elements bound by LEF-1 after transfection of NHA with ASCL1 and stimulation with Wnt3a
measured by qRT-PCR. Error bars indicate SEM.
(C) A model for crosstalk between aberrantly activated TFs and Wnt signaling in nonstem GBM cells and NHA (top) versus GBM CSCs (bottom). In nonstem GBM
is transcribed, and expressed DKK1 inhibits Wnt signaling. In GBM CSCs, Polycomb repression is lost at many TF loci, including ASCL1. ASCL1 binds to the
DKK1 regulatory element, thereby repressing DKK1 expression and activating Wnt signaling. Active Wnt signaling feeds back upon loci encoding several other
TFs that are aberrantly active in GBM CSCs.
See also Figure S6.
Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors 1575
(Epitomics T091), or LEF-1 (Abcam 53293). ChIP DNA samples were used to
prepare sequencing libraries, which were then sequenced on the Illumina
Genome Analyzer II or HiSeq by standard procedures. We sequenced an
input control for each cell type for use in normalization. Read alignment to
the hg19 human reference genome was performed with Bowtie (Langmead
et al., 2009) and density maps were generated with read extension to
200 bp with IGVtools (Robinson et al., 2011; Thorvaldsdo ´ttir et al., 2013).
When several reads with same start position and direction were present,
only one was kept for further analysis. Two replicates that were available for
MGG8 GBM CSCs were merged into a single track. Visualization was per-
formed with the Integrative Genomics Viewer. ChIP-seq data set statistics
are summarized in Table S1 and data are available for viewing at http://
DNA Methylation Assay and Analysis
For each sample, about 1 3 106cells were harvested and genomic DNA was
isolated using the QiaAMP DNA mini kit following the manufacturer’s instruc-
tions. DNA was eluted in 100 ml water, treated with RNase (37?C for 30 min),
and cleaned up again with the QiaAMP DNA mini kit.
Data were processed using Illumina BeadStudio software. Probes with p
value calls above 0.05 were discarded and b values for two replicates for
each sample were averaged. Probes with b R 0.75 were classified as ‘‘hyper-
methylated.’’ Data are available through the Gene Expression Omnibus
Detection of Regions Enriched for Histone Modifications
Genomic regions enriched for a histone mark were identified using a sliding
window approach as previously described (Mikkelsen et al., 2007) with several
modifications. We adapted the previous approach for the highly copy-number
variant genomes of the GBM CSCs with the help of an unenriched input
sequencing track generated from whole-cell extract (WCE). In short, a fixed
size window of 1 kb was used to scan the genome in 100 bp steps for local
enrichment of ChIP signal. Significance of signal in each window was as-
sessed based on the assumption that random read alignment would follow a
Poissondistribution withparameterlChIP.lChIPwasadjustedforlocal variation
in genome copy number by multiplication with the observed-to-expected ratio
(O/E) for unenriched input reads in this region (Mikkelsen et al., 2010). To in-
crease numeric stability in regions of heterozygous deletions, we calculated
this O/E ratio based on input reads in the scoring window as well as in a 10
kb and a 100 kb region centered at the current window and used the maximal
valueof these three. When all three input O/E ratios were zero, ladjustedwas set
to equal lChIP. Poisson p values were then calculated for each window with
ladjusted. All p values were corrected for multiple-hypothesis testing using
the Benjamini-Hochberg procedure (Benjamini and Hochberg, 1995), and
only windows with significance p < 10?5were kept. Finally, adjacent (distance
ifications in NSCs and NHAs, we applied the same algorithm and parameters
albeit without background correction. Genomic regions enriched for ASCL1 or
LEF-1 were identified with MACS (Zhang et al., 2008). No background correc-
tion as described above was applied; instead, peaks identified in the WCE
track served to remove spurious TF peaks.
TSS for genes from the hg19 human genome assembly were defined as re-
gion from 500 bp upstream to 2 kb downstream as previously described (Mik-
kelsen et al., 2007). Chromatin state calls for TSS were then made based on a
minimal overlap of 500 bp of enriched intervals with the 2.5 kb TSS region. A
consensus set of TSS chromatin states in the CSCs was generated with the
chromatin state of the majority (at least three of four CSC lines) assigned to
summarized ‘‘CSC’’ cell type. 87% (n = 20,422) of TSS satisfied the majority
criterion and were thus included in the consensus set.
Functional Gene Enrichment Analysis
Functional enrichment in gene sets was determined using the DAVID func-
tional annotation tool with ‘‘FAT’’ Gene Ontology terms (Dennis et al., 2003;
version 6.7). Benjamini-Hochberg p values correcting for multiple hypothesis
testing were used for further interpretation.
Generation of Aberrantly Active TF Set
TFs were identified as those aberrantly active genes that were contained in the
GO:0003700 transcription factor activity set or in the set of human TFs defined
by Vaquerizas et al. (2009). We manually removed TFs whose chromatin state
was incorrectly of ambiguously called by our algorithm to generate the final list
of TFs. MYCN was also removed because of focal amplification in the MGG8
cell line. For the TF chromatin state heatmap, we extracted H3K4me3 and
H3K27me3 signal, respectively, in 40 bins covering a 5 kb region around the
annotated TSS from density maps. For both H3K4me3 and H3K27me3,
several control genes with similar chromatin state in all samples were chosen
(Table S4)and servedtoscale signalfor eachsample.TFs wereordered based
on their H3K27me3 signal using hierarchical clustering (R Development Core
Team, 2008), and H3K4me3 and H3K27me3 maps were overlayed to generate
the final map. Cells exceeding 15% of maximum signal for H3K4me3 and
H3K27me3 (bivalent) were additionally enhanced with orange.
RNA Extraction and Gene Expression Analyses
Total RNA was isolated from cells using TRIzol Reagent (Invitrogen) and puri-
fied using the RNeasy Kit (QIAGEN). Gene expression data were acquired with
Affymetrix Human Genome U133 2.0 arrays. CEL files were normalized with
robust multichip average and multiple probe sets per gene were collapsed
by taking the maximum expression value using the GenePattern package
(Reich et al., 2006). Gene expression data for NHA were included from Balani
etal. (2009) (GSE12305). Normal brain and astrocytoma transcriptomeprofiles
from Sun et al. (2006) were used (GSE4290) and processed as described
above. TCGA combined expression data and subtype information was ob-
tained from Verhaak et al. (2010).
We used the HOMER software package (Heinz et al., 2010) to search for de
novo enriched motifs in TF peak regions.
Overexpression and Knockdown Experiments
Human cDNA for ASCL1, OLIG1, OLIG2, HEY2, LHX2, and EN2 was cloned
into a lentiviral plasmid and sequence verified. Primers used are listed in Table
S6. For knockdown experiments, an ASCL1 lentiviral shRNA set from Open
Biosystems was used (RHS4533-NM_004316), of which TRCN0000013551
(CCCGAACTGATGCGCTGCAAA) yielded sufficient knockdown. The same
sequence was also used in vector pGIPZ (RHS4430-101103529) to allow for
GFP sorting. Lentiviruses were produced as previously described (Barde
et al., 2010). Briefly, cDNA coding or shRNA plasmids were cotransfected
with GAG/POL and VSV plasmids into 293T packaging cells to produce the vi-
rus used to infect the target cells (NHA or GBM CSC). Viral supernatant was
concentrated by ultracentrifugation using an SW41Ti rotor (Beckman Coulter)
at 28,000 rpm for 120 min. Using GFP control, efficiency of infection was esti-
mated as greater than 90% (data not shown). For maximal homogeneity, NHA
were selected using 0.75 mg/ml puromycin for 5 days and GBM CSC were
either selected using 2 mg/ml puromycin or sorted for GFP depending on vec-
tor used. After selection, RNA was extracted (QIAGEN RNeasy kit) following
the manufacturer’s instructions.
Real-Time Quantitative RT-PCR
cDNA was obtained using Moloney murine leukemia virus reverse transcrip-
tase and RNase H minus (Promega). Typically, 250 ng of template total RNA
fication was performed using Power SYBR mix and specific PCR primers in a
7500 Fast PCR instrument (Applied Biosystems). Relative quantification of
each target, normalized to an endogenous control (GAPDH), was performed
using the comparative Ct method (Applied Biosystems). Error bars indicate
SEM. Primer sequences are listed in Table S6.
TOPFLASH-Firefly and FOPFLASH-Renilla plasmids were cotransfected with
ASCL1 lentivirus or control vector in 293T cells using Fugene6, as previously
described (Firestein et al., 2008; Veeman et al., 2003). When indicated,
Wnt3a was added at 100 ng/ml (R&D 5036-WN-010). Luciferase activity was
1576 Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors
measured after 48 hr using a dual-luciferase reporter assay system (Promega
E1910) according to the manufacturer’s instructions.
GFP-sorted GBM CSC spheres, infected either with lentiviral control vector
or with ASCL1 shRNA vector, were mechanically dissociated into single cells
number was assessed 2 weeks later under a fluorescence microscope.
For sphere diameter quantification, five pictures were taken per condition at
1003 magnification. At least 60 spheres per conditions were measured with
Intracranial injections of 5,000 cells from acutely dissociated gliomaspheres
were performed with a stereotactic apparatus (Kopf Instruments) at coordi-
nates 2.2 mm lateral relative to Bregma point and 2.5 mm deep from dura
mater. Four severe combined immunodeficient mice were used per condition.
Kaplan-Meier curves and statistical significance (log-rank test) were calcu-
lated with the R survival package (R Development Core Team, 2008). Animal
experiments were approved by protocol number 2009N000061.
mRNA was detected in formalin-fixed, paraffin-embedded tissue sections
using Quantigene ViewRNA (Affymetrix). Probes for ASCL1 (type 1, red,
VA1-11908; Affymetrix) and Sox2 (type 6, blue, VA-11765) were used following
the manufacturer’s instructions for two-color chromogenic ISH. Tissue
sections were prepared for hybridization by fixation in 10% formaldehyde, de-
paraffinization, boiling for 10 min, and digestion with protease for 20 min. Hy-
bridization was performed for 2 hr at 40?C. Signal amplification and detection
were performed using standard conditions for fast red and fast blue sub-
strates. Tissues were counterstained with hematoxylin and visualized with a
standard bright-field microscope.
The Gene Expression Omnibus accession number for the microarray gene
expression data, DNA methylation profiles, and ChIP-seq density maps re-
ported in this paper is GSE46016.
Supplemental Information includes six figures and six tables and can be found
with this article online at http://dx.doi.org/10.1016/j.celrep.2013.04.021.
This is an open-access article distributed under the terms of the Creative
Commons Attribution-NonCommercial-No Derivative Works License, which
permits non-commercial use, distribution, and reproduction in any medium,
provided the original author and source are credited.
We thank Tarjei Mikkelsen, Timothy Durham, and Noam Shoresh for computa-
tional assistance and Chuck Epstein, Robbyn Issner, and the Broad Institute
Genome Sequencing Platform for help with data production. We thank David
ney for assistance with graphics. A.S.C. is supported by a Joan Ambriz Amer-
ican Brain Tumor Association basic research fellowship and early career
research award from the Ben and Catherine Ivy Foundation. M.L.S. is sup-
ported by Oncosuissegrant BIL-KFS-02590-02-2010 and aMedic Foundation
grant. M.N.R. is supported by awards from the Burroughs Wellcome Fund and
the Howard Hughes Medical Institute. S.D.R. supported in part by NIH (RO1
CA160762). This research was supported by funds from the Starr Cancer Con-
sortium (B.E.B.), the Harvard Stem Cell Institute, the Burroughs Wellcome
Fund (B.E.B.), and the Howard Hughes Medical Institute (B.E.B.). M.N.R.
and M.S. receive research funding from Affymetrix.
Received: October 15, 2012
Revised: February 6, 2013
Accepted: April 20, 2013
Published: May 23, 2013
Balani, P., Boulaire, J., Zhao, Y., Zeng, J., Lin, J., and Wang, S. (2009). High
mobility group box2 promoter-controlled suicide gene expression enables tar-
geted glioblastoma treatment. Mol. Ther. 17, 1003–1011.
Bao, S., Wu, Q., McLendon, R.E., Hao, Y., Shi, Q., Hjelmeland, A.B., Dewhirst,
M.W., Bigner, D.D., and Rich, J.N. (2006). Glioma stem cells promote radiore-
sistance by preferential activation of the DNA damage response. Nature 444,
Barde, I., Salmon, P., and Trono, D. (2010). Production and titration of lentiviral
vectors. Curr. Protoc. Neurosci. Chapter 4, Unit 4, 21.
Baylin, S.B., and Jones, P.A. (2011). A decade of exploring the cancer epige-
Bell, O.,Tiwari,V.K.,Thoma ¨,N.H., andSchu ¨beler, D.(2011). Determinants and
dynamics of genome accessibility. Nat. Rev. Genet. 12, 554–564.
Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery rate: a
practical and powerful approach to multiple testing. J. R. Stat. Soc. Series B
Stat. Methodol. 57, 289–300.
Bruggeman,S.W., Hulsman, D., Tanger, E.,Buckle,T.,Blom, M.,Zevenhoven,
J., van Tellingen, O., and van Lohuizen, M. (2007). Bmi1 controls tumor devel-
opment in an Ink4a/Arf-independent manner in a mouse model for glioma.
Cancer Cell 12, 328–341.
Chen, J., Li, Y., Yu, T.S., McKay, R.M., Burns, D.K., Kernie, S.G., and Parada,
L.F. (2012). A restricted cell population propagates glioblastoma growth after
chemotherapy. Nature 488, 522–526.
Clement, V., Sanchez, P., de Tribolet, N., Radovanovic, I., and Ruiz i Altaba, A.
(2007). HEDGEHOG-GLI1 signaling regulates human glioma growth, cancer
stem cell self-renewal, and tumorigenicity. Curr. Biol. 17, 165–172.
Clevers, H. (2006). Wnt/beta-catenin signaling in development and disease.
Cell 127, 469–480.
Conti, L., Pollard, S.M., Gorba, T., Reitano, E., Toselli, M., Biella, G., Sun, Y.,
symmetrical self-renewal of a mammalian tissue stem cell. PLoS Biol. 3, e283.
Dennis, G., Jr., Sherman, B.T., Hosack, D.A., Yang, J., Gao, W., Lane, H.C.,
and Lempicki, R.A. (2003). DAVID: Database for Annotation, Visualization,
and Integrated Discovery. Genome Biol. 4, 3.
Dunham, I., Kundaje, A., Aldred, S.F., Collins, P.J., Davis, C.A., Doyle, F.,
Epstein, C.B., Frietze, S., Harrow, J., Kaul, R., et al.; ENCODE Project Con-
sortium. (2012). An integrated encyclopedia of DNA elements in the human
genome. Nature 489, 57–74.
Ernst, J., Kheradpour, P., Mikkelsen, T.S., Shoresh, N., Ward, L.D., Epstein,
C.B., Zhang, X., Wang, L., Issner, R., Coyne, M., et al. (2011). Mapping and
analysis of chromatin state dynamics in nine human cell types. Nature 473,
Firestein, R., Bass, A.J., Kim, S.Y., Dunn, I.F., Silver, S.J., Guney, I., Freed, E.,
Ligon, A.H., Vena, N., Ogino, S., et al. (2008). CDK8 is a colorectal cancer
oncogene that regulates beta-catenin activity. Nature 455, 547–551.
Galli, R., Binda, E., Orfanelli, U., Cipelletti, B., Gritti, A., De Vitis, S., Fiocco, R.,
Foroni, C., Dimeco, F.,and Vescovi,A.(2004).Isolation and characterizationof
tumorigenic, stem-like neural precursors from human glioblastoma. Cancer
Res. 64, 7011–7021.
Gangemi, R.M., Griffero, F., Marubbi, D., Perera, M., Capra, M.C., Malatesta,
P., Ravetti, G.L., Zona, G.L., Daga, A., and Corte, G. (2009). SOX2 silencing in
glioblastoma tumor-initiating cells causes stop of proliferation and loss of
tumorigenicity. Stem Cells 27, 40–48.
Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors 1577
Heinz, S., Benner, C., Spann, N., Bertolino, E., Lin, Y.C., Laslo, P., Cheng, J.X.,
Murre, C., Singh, H., and Glass, C.K. (2010). Simple combinations of lineage-
determining transcription factors prime cis-regulatory elements required for
macrophage and B cell identities. Mol. Cell 38, 576–589.
D., Gaetani, P., Tunici, P., Finocchiaro, G., et al. (2009). A role for the transcrip-
tion factor HEY1 in glioblastoma. J. Cell. Mol. Med. 13, 136–146.
Huse, J.T., and Holland, E.C. (2010). Targeting brain cancer: advances in the
molecular pathology of malignant glioma and medulloblastoma. Nat. Rev.
Cancer 10, 319–331.
Ku, M., Koche, R.P., Rheinbay, E., Mendenhall, E.M., Endoh, M., Mikkelsen,
T.S., Presser, A., Nusbaum, C., Xie, X., Chi, A.S., et al. (2008). Genomewide
analysis of PRC1 and PRC2 occupancy identifies two classes of bivalent do-
mains. PLoS Genet. 4, e1000242.
Langmead, B., Trapnell, C., Pop, M., and Salzberg, S.L. (2009). Ultrafast and
memory-efficient alignment of short DNA sequences to the human genome.
Genome Biol. 10, R25.
Lee, J., Kotliarova, S., Kotliarov, Y., Li, A., Su, Q., Donin, N.M., Pastorino, S.,
Purow, B.W., Christopher, N., Zhang, W., et al. (2006). Tumor stem cells
derived from glioblastomas cultured in bFGF and EGF more closely mirror
the phenotype and genotype of primary tumors than do serum-cultured cell
lines. Cancer Cell 9, 391–403.
Lee, J., Son, M.J.,Woolard, K., Donin, N.M., Li, A., Cheng, C.H., Kotliarova,S.,
Kotliarov, Y., Walling, J., Ahn, S., et al. (2008). Epigenetic-mediated dysfunc-
tion of the bone morphogenetic protein pathway inhibits differentiation of glio-
blastoma-initiating cells. Cancer Cell 13, 69–80.
Ligon, K.L., Alberta, J.A., Kho, A.T., Weiss, J., Kwaan, M.R., Nutt, C.L., Louis,
D.N., Stiles, C.D., and Rowitch, D.H. (2004). The oligodendroglial lineage
marker OLIG2 is universally expressed in diffuse gliomas. J. Neuropathol.
Exp. Neurol. 63, 499–509.
Lottaz, C., Beier, D., Meyer, K., Kumar, P., Hermann, A., Schwarz, J., Junker,
M., Oefner, P.J., Bogdahn, U., Wischhusen, J., et al. (2010). Transcriptional
profiles of CD133+ and CD133- glioblastoma-derived cancer stem cell lines
suggest different cells of origin. Cancer Res. 70, 2030–2040.
Meissner, A., Mikkelsen, T.S., Gu, H., Wernig, M., Hanna, J., Sivachenko, A.,
Zhang, X., Bernstein, B.E., Nusbaum, C., Jaffe, D.B., et al. (2008). Genome-
scale DNA methylation maps of pluripotent and differentiated cells. Nature
Mikkelsen, T.S., Ku, M., Jaffe, D.B., Issac, B., Lieberman, E., Giannoukos, G.,
Alvarez, P., Brockman, W., Kim, T.K., Koche, R.P., et al. (2007). Genome-wide
maps of chromatin state in pluripotent and lineage-committed cells. Nature
Mikkelsen, T.S., Xu, Z., Zhang, X., Wang, L., Gimble, J.M., Lander, E.S., and
Rosen, E.D. (2010). Comparative epigenomic analysis of murine and human
adipogenesis. Cell 143, 156–169.
Murat, A., Migliavacca, E., Gorlia, T., Lambiv, W.L., Shay, T., Hamou, M.F., de
Tribolet, N., Regli, L., Wick, W., Kouwenhoven, M.C., et al. (2008). Stem cell-
related ‘‘self-renewal’’ signature and high epidermal growth factor receptor
expression associated with resistance to concomitant chemoradiotherapy in
glioblastoma. J. Clin. Oncol. 26, 3015–3024.
Ohm, J.E., McGarvey, K.M., Yu, X., Cheng, L., Schuebel, K.E., Cope, L.,
Mohammad, H.P., Chen, W., Daniel, V.C., Yu, W., et al. (2007). A stem cell-
like chromatinpattern may predispose tumorsuppressorgenes toDNA hyper-
methylation and heritable silencing. Nat. Genet. 39, 237–242.
Osada, H., Tomida, S., Yatabe, Y., Tatematsu, Y., Takeuchi, T., Murakami, H.,
logue 1 in DKK1 and E-cadherin repression and neuroendocrine differentiation
in lung cancer. Cancer Res. 68, 1647–1655.
Pujadas, E., and Feinberg, A.P. (2012). Regulated noise in the epigenetic land-
scape of development and disease. Cell 148, 1123–1131.
R Development Core Team (2008). R: a language and environment for statis-
tical computing. R Foundation for Statistical Computing, Vienna. ISBN 3-
Reich, M., Liefeld, T., Gould, J., Lerner, J., Tamayo, P., and Mesirov, J.P.
(2006). GenePattern 2.0. Nat. Genet. 38, 500–501.
Robinson, J.T., Thorvaldsdo ´ttir, H., Winckler, W., Guttman, M., Lander, E.S.,
Getz, G., and Mesirov, J.P. (2011). Integrative genomics viewer. Nat. Bio-
technol. 29, 24–26.
Schlesinger, Y., Straussman, R., Keshet, I., Farkash, S., Hecht, M., Zimmer-
man, J., Eden, E., Yakhini, Z., Ben-Shushan, E., Reubinoff, B.E., et al.
(2007). Polycomb-mediated methylation on Lys27 of histone H3 pre-marks
genes for de novo methylation in cancer. Nat. Genet. 39, 232–236.
Shih, A.H., and Holland, E.C. (2006). Notch signaling enhances nestin expres-
sion in gliomas. Neoplasia 8, 1072–1082.
Singh, S.K., Clarke, I.D., Terasaki, M., Bonn, V.E., Hawkins, C., Squire, J., and
Dirks, P.B. (2003). Identification of a cancer stem cell in human brain tumors.
Cancer Res. 63, 5821–5828.
Singh, S.K., Hawkins, C., Clarke,I.D., Squire, J.A., Bayani, J.,Hide,T.,Henkel-
man, R.M., Cusimano, M.D., and Dirks, P.B. (2004). Identification of human
brain tumour initiating cells. Nature 432, 396–401.
Son, M.J., Woolard, K., Nam, D.H., Lee, J., and Fine, H.A. (2009). SSEA-1 is an
enrichment marker for tumor-initiating cells in human glioblastoma. Cell Stem
Cell 4, 440–452.
Stricker, S.H., Feber, A., Engstro ¨m, P.G., Care ´n, H., Kurian, K.M., Takashima,
Y., Watts, C., Way, M., Dirks, P., Bertone, P., et al. (2013). Widespread
resetting of DNA methylation in glioblastoma-initiating cells suppresses
malignant cellular behavior in a lineage-dependent manner. Genes Dev. 27,
Sun, L., Hui, A.M., Su, Q., Vortmeyer, A., Kotliarov, Y., Pastorino, S., Passaniti,
A.,Menon, J., Walling, J., Bailey, R.,et al. (2006). Neuronal and glioma-derived
stemcellfactorinduces angiogenesiswithin thebrain. CancerCell9,287–300.
Suva `, M.L., Riggi, N., Janiszewska, M., Radovanovic, I., Provero, P., Stehle,
J.C., Baumer, K., Le Bitoux, M.A., Marino, D., Cironi, L., et al. (2009). EZH2
is essential for glioblastoma cancer stem cell maintenance. Cancer Res. 69,
Thorvaldsdo ´ttir,H.,Robinson,J.T., andMesirov, J.P.(2013). IntegrativeGeno-
tion. Brief. Bioinform. 14, 178–192.
Travis, A., Amsterdam, A., Belanger, C., and Grosschedl, R. (1991). LEF-1, a
gene encoding a lymphoid-specific protein with an HMG domain, regulates
T-cell receptor alpha enhancer function [corrected]. Genes Dev. 5, 880–894.
Valent, P., Bonnet, D., De Maria, R., Lapidot, T., Copland, M., Melo, J.V., Cho-
mienne, C., Ishikawa, F., Schuringa, J.J., Stassi, G., et al. (2012). Cancer stem
cell definitions and terminology: the devil is in the details. Nat. Rev. Cancer 12,
Vaquerizas, J.M., Kummerfeld, S.K., Teichmann, S.A., and Luscombe, N.M.
(2009). A census of human transcription factors: function, expression and evo-
lution. Nat. Rev. Genet. 10, 252–263.
Veeman, M.T., Slusarski, D.C., Kaykas, A., Louie, S.H., and Moon, R.T. (2003).
Zebrafish prickle, a modulator of noncanonical Wnt/Fz signaling, regulates
gastrulation movements. Curr. Biol. 13, 680–685.
Verhaak, R.G., Hoadley, K.A., Purdom, E., Wang, V., Qi, Y., Wilkerson, M.D.,
Miller, C.R., Ding, L., Golub, T., Mesirov, J.P., et al.; Cancer Genome Atlas
Research Network. (2010). Integrated genomic analysis identifies clinically
relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA,
IDH1, EGFR, and NF1. Cancer Cell 17, 98–110.
Wakimoto, H., Kesari, S., Farrell, C.J., Curry, W.T., Jr., Zaupa, C., Aghi, M.,
glioblastoma-derived cancer stem cells: establishment of invasive glioma
models and treatment with oncolytic herpes simplex virus vectors. Cancer
Res. 69, 3472–3481.
Wakimoto, H., Mohapatra, G., Kanai, R., Curry, W.T., Jr., Yip, S., Nitta, M.,
Patel, A.P., Barnard, Z.R., Stemmer-Rachamimov, A.O., Louis, D.N., et al.
(2012). Maintenance of primary tumor phenotype and genotype in glioblas-
toma stem cells. Neuro. Oncol. 14, 132–144.
1578 Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors
Widschwendter, M., Fiegl, H., Egle, D., Mueller-Holzner, E., Spizzo, G., Marth, Download full-text
C., Weisenberger, D.J., Campan, M., Young, J., Jacobs, I., and Laird, P.W.
(2007). Epigenetic stem cell signature in cancer. Nat. Genet. 39, 157–158.
Zhang, Y., Liu, T., Meyer, C.A., Eeckhoute, J., Johnson, D.S., Bernstein, B.E.,
Nusbaum, C., Myers, R.M., Brown, M., Li, W., and Liu, X.S. (2008). Model-
based analysis of ChIP-Seq (MACS). Genome Biol. 9, R137.
Zhang, N., Wei, P., Gong, A., Chiu, W.T., Lee, H.T., Colman, H., Huang, H.,
Xue, J., Liu, M., Wang, Y., et al. (2011). FoxM1 promotes b-catenin nuclear
localization and controls Wnt target-gene expression and glioma tumorigen-
esis. Cancer Cell 20, 427–442.
J.H., Zhang, H., Xiao, Y., Perry, S.R., et al. (2010). PLAGL2 regulates Wnt
signaling to impede differentiation in neural stem cells and gliomas. Cancer
Cell 17, 497–509.
Zhou, V.W., Goren, A., and Bernstein, B.E. (2011). Charting histone modifica-
Zhu, J., Adli, M., Zou, J.Y., Verstappen, G., Coyne, M., Zhang, X., Durham, T.,
Miri, M., Deshpande, V., De Jager, P.L., et al. (2013). Genome-wide chromatin
state transitions associated with developmental and environmental cues. Cell
Cell Reports 3, 1567–1579, May 30, 2013 ª2013 The Authors 1579