TAK1 Inhibition Promotes Apoptosis
in KRAS-Dependent Colon Cancers
Anurag Singh,1,3Michael F. Sweeney,1Min Yu,1Alexa Burger,1Patricia Greninger,1Cyril Benes,1Daniel A. Haber,1,2,*
and Jeff Settleman1,4,*
1Massachusetts General Hospital Cancer Center and Harvard Medical School, Charlestown, MA 02129, USA
2Howard Hughes Medical Institute, Chevy Chase, MD 20815, USA
3Present address: Department of Pharmacology and Experimental Therapeutics, Division of Medical Oncology and Hematology,
Cancer Research Center, Boston University School of Medicine, Boston, MA 02118, USA
4Present address: Discovery Oncology, Genentech, Inc., South San Francisco, CA 94080, USA
*Correspondence: email@example.com (D.A.H.), firstname.lastname@example.org (J.S.)
Colon cancers frequently harbor KRAS mutations,
yet only a subset of KRAS mutant colon cancer cell
lines are dependent upon KRAS signaling for
survival. In a screen for kinases that promote survival
of KRAS-dependent colon cancer cells, we found
that the TAK1 kinase (MAP3K7) is required for tumor
cell viability. The induction of apoptosis by RNAi-
mediated depletion or pharmacologic inhibition of
TAK1 is linked to its suppression of hyperactivated
Wnt signaling, evident in both endogenous and
genetically reconstituted cells. In APC mutant/
KRAS-dependent cells, KRAS stimulates BMP-7
tion and enhancement of Wnt-dependent transcrip-
tion. An in vitro-derived ‘‘TAK1 dependency signa-
ture’’ is enriched in primary human colon cancers
with mutations in both APC and KRAS, suggesting
potential clinical utility in stratifying patient popula-
tions. Together, these findings identify TAK1 inhibi-
tion as a potential therapeutic strategy for a treat-
ment-refractory subset of colon cancers exhibiting
aberrant KRAS and Wnt pathway activation.
Targeted cancer therapies have exploited specific mutations
that drive survival signals in subsets of tumors, leading to
successful genotype-directed clinical applications of small-
molecule inhibitors (reviewed in Haber et al., 2011). However,
KRAS-activating mutations, which are common in multiple
human cancers, remain a critical therapeutic challenge. KRAS
mutations are generally associated with treatment-refractory
tumors (Downward, 2003). For instance, KRAS mutant lung
cancers are generally refractory to EGFR-targeted small-mole-
cule inhibitors, and in colon cancer, KRAS mutations predict
failure of response to antibodies targeting overexpressed wild-
type EGFR (Normanno et al., 2009). Thus, patients with KRAS
mutant colon cancers are excluded from targeted EGFR thera-
pies and are faced with limited therapeutic options.
To date, pharmacologic targeting of activated KRAS has not
been successful. Mutationally activated KRAS proteins are not
apies thatdisrupt KRAS posttranslational modifications have not
been clinically efficacious (Whyte et al., 1997). More recently,
several genome-wide shRNA screens revealed synthetic lethal
interactions in which knockdown of a candidate gene appeared
et al., 2009; Luo et al., 2009a; Scholl et al., 2009; Singh et al.,
2009). To date, these findings have not led to the development
of effective inhibitors to treat KRAS mutant cancers, possibly
due in part to the contextual complexity of KRAS mutations and
the difficulty in generalizing synthetic lethal interactions across
a broad range of tumor-specific backgrounds.
In analyzing large panels of KRAS mutant tumor-derived cell
lines, we noted that approximately half of these cell lines under-
went apoptosis following shRNA-mediated knockdown of KRAS
(so-called KRAS dependent), whereas the other half maintained
viability (KRAS independent) (Singh et al., 2009). These differ-
ences were independent of the particular KRAS mutation,
although KRAS-dependent cells generally exhibited higher
KRAS protein expression levels. Cellular context appeared to
play a significant role in the dependency of KRAS mutant cells
on continued KRAS signaling for their survival. In both lung and
pancreatic cancer cells harboring KRAS mutations, the pres-
ence of epithelial markers was highly correlated with KRAS
dependency, whereas epithelial to mesenchymal transformation
(EMT) was associated with KRAS independence, despite the
presence of a KRAS mutation. Reversion of mesenchymal cells
to an epithelial phenotype was associated with restored depen-
dence on KRAS, suggesting that mesenchymal-associated
signals may provide alternative survival pathways when KRAS
activity is disrupted (Singh et al., 2009). Interestingly, the associ-
ation of KRAS dependency with expression of epithelial markers
was not evident in KRAS mutant colon cancers, prompting our
interest in identifying lineage-specific determinants of KRAS
dependency in colon cancers.
The cellular context of KRAS mutations in colon cancer is
complex. APC loss-of-function mutations that arise in early
Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc. 639
adenomas lead to inappropriate activation of canonical Wnt
(Behrens etal., 1996). KRAS mutations are most common during
of cases and subsequently followed by frequent deletions of
SMAD4, targeting the transforming growth factor beta (TGF-b)
signaling pathway (Vogelstein et al., 1988). The contribution of
KRAS mutations to colon carcinogenesis is thus uniquely linked
to altered Wnt and TGF-b signaling.
Here, we compared KRAS-dependent and -independent
colon cancer cells using a combination of gene expression and
shRNA knockdown studies, which led to the identification of
MAP3K7, encoding the TGF-b-activated kinase (TAK1), as a
driver of cell survival in KRAS-dependent, APC-deficient cells.
Together with the recently reported enhancement of Wnt
signaling by KRAS (Janssen et al., 2006; Phelps et al., 2009),
these observations point to an unappreciated yet critical
signaling node in a subset of colon cancers. We demonstrate
that, in KRAS-dependent cells, but not KRAS-independent cells,
KRAS activates bone morphogenetic
signaling, leading to TAK1 activation, b-catenin nuclear localiza-
tion, and transcriptional upregulation of Wnt target genes. This is
also accompanied by KRAS- and TAK1-regulated activation of
the NF-kB pathway. Reconstitution studies confirm that activa-
tion of this KRAS-dependent signaling network underlies exqui-
site sensitivity to TAK1 inhibition. Together, these observations
point to a potential therapeutic strategy, based on targeting
a vulnerable node in an identifiable subset of APC/KRAS mutant
Identification of KRAS-Dependent Colon Cancer
We used a lentiviral-based shRNA assay to quantitate KRAS
dependency (Singh et al., 2009) in 21 KRAS mutant colon cancer
cell lines, measuring cell viability at 6 days postinfection (Fig-
ure 1A). KRAS mutant colon cancer cells showed variable
KRAS dependencies (Figures 1A and 1B), allowing us to derive
a quantitative Ras dependency index (RDI) to compare multiple
cell lines with varying viral transduction efficiencies (see Experi-
mental Procedures). An RDI > 2.0 represented a threshold to
classify cells as KRAS dependent. Among the 21 KRAS mutant
cell lines, 10 were classified as KRAS dependent and 11 as
KRAS independent (Figure 1B). KRAS dependency was not
associated with particular KRAS-activating mutations (Table
S2 available online). Examples of two KRAS-dependent cell lines
(SW620 and SK-CO-1) were selected for comparison with two
KRAS-independent lines (LS-174T and SW1463) (Figure 1A).
KRAS depletion in KRAS-dependent colon cancer cells trig-
gered apoptosis, measured by caspase-3 and polyADP ribose
polymerase (PARP) cleavage at 6 days following shRNA knock-
down (Figure 1C). Cells classified as KRAS independent despite
the presence of mutant KRAS showed no such apoptotic
Figure 1. Classification of KRAS Mutant
Colon Cancer Cells into KRAS-Independent
and KRAS-Dependent Groups
(A) Representative 6 day 96-well viability assays in
four KRAS mutant colon cancer cell lines trans-
duced with either control or two independent
KRAS-directed lentiviral shRNAs (A and B) at two
viral MOIs. Cell lines in red text are KRAS inde-
pendent, and those in green text are KRAS
dependent. Quantitation and transformation of
relative cell density values yields the Ras depen-
dency index depicted in Figure 1B.
(B) Ras dependency index plot for a panel of 21
KRAS mutant colon cancer cell lines. Dashed line
represents the ‘‘dependency threshold’’ of 2.0.
Data are presented as the mean of three inde-
pendent experiments ± SEM.
(C) KRAS protein depletion 4 days postinfection
with KRAS-directed shRNAs and effects on
apoptosis, as assessed by caspase-3 and PARP
cleavage, in a representative panel of KRAS-
dependent versus KRAS-independent cell lines.
Lanes 1, 2, and 3 are as in (A). Data are repre-
sentative of two independent experiments.
(p-Erk1/2) and Akt (p-Akt) kinases, following KRAS
depletion in SW837 KRAS-independent versus
SW620 2 KRAS-dependent cells, 4 days post-
infection with three different viral titers (MOIs of
1, 2, and 4) of shKRAS-B. Total protein levels
(t-Erk1 and t-Akt) are shown as gel loading con-
trols. Note: Different exposure times were used for
the individual panels. Data are representative of
two independent experiments.
640 Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc.
response to KRAS depletion. Reduced Erk and Akt phosphory-
lation preceded apoptosis in KRAS-dependent cells, whereas
KRAS-independent cells displayed weak KRAS coupling to
Erk phosphorylation. Moreover, in KRAS-independent cells,
KRAS depletion resulted in paradoxically increased Akt phos-
phorylation, in agreement with recent reports (Ebi et al., 2011)
(Figure 1D). Thus, KRAS-dependent and -independent colon
cancer cells demonstrate distinct patterns of signaling down-
stream of mutant KRAS, with only KRAS-dependent cells
showing suppression of key survival signals following KRAS
TAK1 Is a KRAS Dependency-Associated Kinase
To identify potentially ‘‘druggable’’ prosurvival effectors in
KRAS-dependent colon cancer cells, we first compared gene
expression profiles in four KRAS-dependent and four KRAS-
independent cell lines (Figure 2A and Experimental Procedures).
A core ‘‘KRAS dependency gene set’’ was identified, comprising
687 genes overexpressed in KRAS-independent cells (IND
genes) and 832 genes overexpressed in KRAS-dependent cells
(DEP genes). Hierarchical clustering of this KRAS dependency
gene set across 40 colon cancer cell lines with either wild-type
or mutant KRAS demonstrated three clusters: IND, DEP, and
intermediate (Figure S1A). Gene ontology analysis of the DEP
gene set, using the DAVID algorithm (Dennis et al., 2003), identi-
fied major functional classes, of which kinases were the most
abundant (Figure S1B). We selected these for further analysis,
given the possibility of identifying novel tractable therapeutic
targets. The 47 DEP protein, lipid, and nucleotide kinase genes
showed significant overexpression in KRAS-dependent colon
cancer cells, confirmed for a subset at the protein level (Figures
2B and 2C). The DEP gene set prominently featured genes
relevant to mitotic checkpoint control and DNA replication/repair
pathways (KEGG pathway database) (Figure S1C). Of note, Wnt
signaling components were significantly enriched in KRAS-
dependent cells, compared to KRAS-independent cells, despite
both classes having a comparable frequency of APC mutations
Figure 2. Analysis of Kinases from a ‘‘KRAS
Dependency Signature’’in Colon Cancer Cell Lines
(A) Schematic representation of the methodology used to
derive a colon cancer KRAS dependency gene expression
data set. Gene expression microarray data for four indi-
cated KRAS-independent versus KRAS-dependent cell
lines were analyzed for significantly underexpressed (IND)
or overexpressed (DEP) genes by Student’s t test analysis
(two-tailed, homoscedastic) followed by selection of
lower, yielding 687 IND genes and 832 DEP genes.
(B) Hierarchical clustering of gene expression for 47 DEP
‘‘druggable’’ protein, lipid, or other ATP-dependent kinase
genes or kinase regulatory genes. Heat map shows log2
median-centered intensity values, and similarly expressed
genes are clustered using Euclidean distance as a simi-
larity metric. MAP3K7 (encoding TAK1) is highlighted with
(C) Protein expression levels of indicated kinases in
a panel of KRAS-independent and KRAS-dependent cell
lines. GAPDH serves as a loading control.
(D) Depletion of DEP kinase genes in SW620 versus
SW837 cells. Each colored bar represents an individual
shRNA sequence per gene, with the same color coding as
in Figure S1F. Fold growth inhibition per shRNA per kinase
was computed by dividing the relative cell density of
SW837 by that of SW620 cells and using a weighted
average to account for viral titer. The plot shows cumu-
lative log2-fold growth inhibition for each shRNA per
kinase, i.e., a value of 1 on the plot indicates a 2-fold
greater growth inhibitory effect for a given shRNA in
SW620 compared to SW837 cells. The log2-fold growth
inhibition for each individual shRNA was then cumulated
for each kinase gene. Data are represented as the mean
value corresponding to each shRNA from three indepen-
(E) Knockdown of TAK1 with increasing viral titers of
shTAK1-D-encoding lentiviruses (MOI) and associated
apoptotic effects assessed by PARP cleavage. GAPDH
serves as a loading control. Data are representative of two
See also Figure S1 and Table S1.
Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc. 641
We further selected candidate protein kinase-encoding
genes from the list of 47, based on ranking by DEP scores as
well as literature searches for genes with putative cancer-
associated function. To establish the functional relevance of
these DEP kinases, we compared the consequences of knock-
down in two cell lines with comparable lentiviral infection profiles
(KRAS-independent SW837 cells and KRAS-dependent SW620
cells; Figures S1D–S1F). We targeted each of 17 kinases, using
5 shRNAs at 3 different viral MOIs, measuring relative cell densi-
ties at 6 days postinfection (Figures S1E and S1F). Among
all kinases tested, TAK1 (MAP3K7) depletion had the most
potent and selective effect on viability of SW620 versus
SW837 cells, measured as the cumulative effect of all shRNA
constructs tested (Figure 2D). Two other genes, VRK2-encoding
vaccinia-related kinase isoform 2 and CHUK-encoding I-kB
kinase alpha, also demonstrated selective albeit less potent
effects on SW620 cell viability (Figures S1E, S1F and 2D).
TAK1 depletion in KRAS-dependent SW620 cells was remark-
able in producing a strong, viral titer-dependent apoptotic
response, as assessed by PARP cleavage (Figure 2E). Because
we have not functionally validated other kinase-encoding genes
from the original list of 47, it remains formally possible that addi-
tional untested kinases may play stronger prosurvival roles than
Validation of TAK1 as a Therapeutic Target
in KRAS-Dependent Colon Cancer
To further validate TAK1 as a candidate therapeutic target in this
context, we used a potent and selective TAK1 kinase inhibitor,
5Z-7-oxozeaenol (Rawlins et al., 1999). We tested sensitivity to
5Z-7-oxozeaenol in a panel of 47 colon cancer cell lines with
various genotypes (Figures 3A and S2A). KRAS and BRAF geno-
Somatic Mutations (COSMIC) or determined by targeted rese-
quencing (Table S2). Among KRAS mutant cells, those classified
as KRAS-dependent by virtue of sensitivity to KRAS shRNA
knockdown were also highly sensitive to TAK1 inhibition,
whereas KRAS-independent cells were generally resistant (p <
0.0001). Of note, of ten BRAF mutant cell lines tested, five
were also sensitive to 5Z-7-oxozeaenol (Figure 3A). The majority
of cells with wild-type KRAS and BRAF were 5Z-7-oxozeaenol
resistant, although some with mutations that potentially impact
the Ras pathway (e.g., NF1 mutation in HT55 cells and ALK
mutation in CoCM-1 cells [COSMIC, Sanger Institute]) were
moderately sensitive to 5Z-7-oxozeaenol. Consistent with their
colon cancer derivation, almost all cell lines tested harbored
APC mutations; of note, three cell lines with wild-type APC but
harboring downstream CTNNB1 (b-catenin)-activating muta-
tions (either S33Y or S45 missense mutations) were resistant
to TAK1 inhibition (Table S2).
To determine whether sensitivity to TAK1 inhibition is specific
to colon cancer-derived cell lines, we assessed sensitivity to
5Z-7-oxozeaenol in five KRAS mutant pancreatic ductal adeno-
carcinoma (PDAC) and four nonsmall cell lung cancer (NSCLC)
cells, all of which are APC wild-type (Figure S2B). Whether
previously classified as KRAS dependent or independent (Singh
et al., 2009), PDAC and NSCLC cells were largely refractory
to 5Z-7-oxozeaenol treatment. Finally, two nontransformed
epithelial cell lines were also 5Z-7-oxozeaenol-refractory:
MCF10A (IC50= 5.5mM) and MDCK (IC50= 22mM) (Figure S2A).
Pharmacologic TAK1 inhibition triggered apoptosis in KRAS-
dependent colon cancer cells, as measured by PARP and
caspase-3 cleavage (Figure 3B). In these cells, 5Z-7-oxozeaenol
treatment caused reduced threonine 172 phosphorylation of the
AMP-activated kinase (p-AMPK), an established TAK1 regulated
kinase (Xie et al., 2006). In contrast, KRAS-independent cells
displayed little or no 5Z-7-oxozeaenol-mediated caspase-3 or
PARP cleavage, except at very high doses, and AMPK phos-
phorylation was unaffected. Thus, low concentrations of 5Z-7-
oxozeaenol, in the range of 0.625–1.25 mM, promote apoptosis
selectively in KRAS-dependent colon cancer cells.
To validate the efficacy of 5Z-7-oxozeaenol in vivo, we gener-
ated subcutaneous xenografted tumors in NOD/SCID mice
using four representative KRAS mutant cell lines: HCT8 and
SW837 (KRAS independent) and SK-CO-1 and SW620 (KRAS
dependent) (Figure 3C). Palpable tumors were evident 2 weeks
postimplantation, at which time mice were treated daily with
either intraperitoneal 15 mg/kg of 5Z-7-oxozeaenol or vehicle
alone (Rawlins et al., 1999). Tumor imaging demonstrated
remarkable regression of both KRAS-dependent tumors after
as few as 6 days of treatment. In contrast, tumors derived from
to TAK1 inhibition. No overt toxicity was evident in 5Z-7-
oxozeaenol-treated mice at the selected dosing regimen.
A Gene Expression Signature Associated
with Sensitivity to TAK1 Inhibition
To identify molecular mechanisms underlying sensitivity to TAK1
inhibition, we isolated subsets of genes within the KRAS DEP
gene set most highly correlated with 5Z-7-oxozeaenol sensi-
tivity. We employed K-means clustering (Gasch and Eisen,
2002) for unsupervised pattern recognition in the KRAS depen-
dency gene set in a test set of 21 colon cancer cell lines whose
sensitivity to TAK1 inhibition had been determined (Figure S3A).
By setting the parameters to k = 3 clusters, we identified
ten nodes (0 through 9) representing synexpression groups of
coregulated genes. We then correlated average expression
scores for the genes in each node with IC50values for 5Z-7-
oxozeaenol by linear regression modeling and computed
the coefficients of determination (r2) and p values for each
node/IC50correlation (Figure S3B). This analysis revealed two
nodes of genes (Figure S3B and S3C) whose expression is
most strongly correlated with sensitivity to TAK1 inhibition. We
combined the genes from these nodes to generate a 32 gene
‘‘TAK1 dependency signature.’’
Clustering of the 32 genes across 21 colon cancer cell lines
demonstrated a high degree of concordance between expres-
sion of the TAK1 dependency gene set, sensitivity to TAK1 inhi-
bition and the degree of KRAS dependency (Figure 4A). Average
expression of the TAK1 dependency signature is very signifi-
cantly correlated with previously derived RDI values for the
KRAS mutant cell lines shown in Figure 4A (p < 0.0001; Fig-
ure S3E). Three general classes of cell lines appear from this
analysis: (1) KRAS-dependent, TAK1 inhibitor-sensitive cell lines
with highest expression of the TAK1 dependency signature; (2)
KRAS-independent, TAK1 inhibitor-refractory cells with weak
642 Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc.
expression of the signature; and (3) a cluster of cell lines with
intermediate levels of expression, demonstrating enrichment
for BRAF mutations (four of six cell lines).
Because we found Wnt pathway enrichment in KRAS-depen-
dent cells (Figure S1C), we overlapped the TAK1 dependency
signature with a data set of binding sites for the Wnt-regulated
transcription factor TCF4, derived from ChIP-on-Chip analyses
(Hatzis et al., 2008). Of the 32 TAK1 dependency genes, 18 con-
tained proximal TCF4-binding sites. A number of these genes,
such as BAMBI, PROX1, and NAV2 (HELAD1), have been previ-
ously linked to colon tumorigenesis in the context of deregulated
Wnt signaling (Lin et al., 2008; Petrova et al., 2008; Sekiya et al.,
2004). Indeed, we found that KRAS-dependent cell lines had
higher basal Wnt signaling activity than KRAS-independent cell
lines, as measured by TOP-FLASH TCF4-responsive luciferase
assays (Figure 4B).
Remarkably, when applied to a primary colon cancer data set
tumors with mutations in both APC and KRAS from those with
only APC mutations (Figure S3C). In particular, the subset of
TAK1 dependency genes identified as being Wnt targets was
expressed at higher levels in APC/KRAS mutant primary colon
cancers compared to APC mutant/KRAS wild-type tumors (Fig-
ure 4C). Though these observations imply increased Wnt
Figure 3. Validation of MAP3K7/TAK1 as a Prosurvival Mediator in KRAS-Dependent Colon Cancers
(A) IC50values (mM) for effects on cellular proliferation and viability with the TAK1 kinase inhibitor 5Z-7-oxozeaenol in a panel of colon cancer cell lines that have
been genotyped as KRAS mutant (KRAS-independent, red circles; KRAS-dependent, green squares), BRAF mutant (blue triangles), or wild-type for both KRAS
and BRAF (OTHER, gray diamonds). Effects on growth were measured 3 days posttreatment. Data are represented as the mean of three independent exper-
iments, and error bars indicate the median ± interquartile range. *p < 0.00001; n.s., not significant.
(B) Effects of TAK1 inhibition on apoptosis and signaling in a representative panel of KRAS-independent and KRAS-dependent cell lines 24 hr after treatment.
PARP and caspase-3 cleavage are shown as indicators of apoptosis, and AMPK threonine 172 (T172) phosphorylation is shown as a downstream indicator of
TAK1 signaling activity. GAPDH serves as a gel loading control.
(C) TAK1 inhibition in mice with xenografted human tumors derived from the HCT8/SW837 (KRAS-independent) and SK-CO-1/SW620 (KRAS-dependent) cell
lines. Cells expressing firefly luciferase were injected subcutaneously into the flanks of nude mice. Tumors are shown as imaged by IVIS detection of lumi-
nescence counts (in photons/sec) following 14 days of tumor growth followed by 6days of treatment with either 15 mg/kg 5z-7-oxozeaenol or vehicle (5% DMSO
in arachis oil) IP delivery q.d. Quantitation of tumor volume (mm3) is also shown. Tumor volume data are represented as the mean of four tumors in two mice for
each group ± SEM.
See also Figure S2 and Table S2.
Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc. 643
signaling in KRAS mutant cancers, we note that some estab-
lished Wnt target genes (e.g., MYC and TCF7 [Figure S3D])
were not enriched in the APC/KRAS mutant tumors. Taken alto-
gether, gene expression analyses suggest that the combination
of APC and KRAS mutations in colon cancers is associated with
Wnt pathway hyperactivation and correlated with susceptibility
to TAK1 inhibition.
KRAS and TAK1 Regulate b-Catenin Transcriptional
Activity and Nuclear Localization
To explore the role of KRAS and TAK1 in modulating Wnt
signaling, we first assessed the effect of KRAS depletion on
b-catenin/TCF transcription in a panel of KRAS mutant cell lines
using the TOP-FLASH reporter (Figures 5A and S4A). The KRAS-
dependent cells SW1116 and SK-CO-1 exhibited decreased
TOP-FLASH reporter activity following KRAS depletion, which
was correlated with the level of KRAS knockdown (Figure 5A).
In contrast, KRAS depletion had no effect in one KRAS-
independent line (SW1463) and increased TOP-FLASH activity
in another (LS174T). In KRAS-dependent cells, 24 hr 5Z-7-
oxozeaenol treatment strongly suppressed TOP-FLASH activity
in a dose-dependent manner (IC500.8–2.5 mM) (Figure 5B). In
contrast, TAK1 inhibition had a much weaker effect on TOP-
FLASH activity in KRAS-independent cells (IC50> 10 mM). Of
note, SW837 cells exhibited a biphasic response to 5Z-7-
oxozeaenol, with increased TOP-FLASH activity at low doses
of TAK1 signaling on a physiological Wnt target gene, we
measured protein expression levels of the endogenous Axin 2
gene (Lustig et al., 2002), following treatment with 5Z-7-
oxozeaenol. TAK1 inhibition resulted in a dose-dependent
reduction in Axin 2 expression in KRAS-dependent cells, but
not in KRAS-independent cells (Figure 5C). Thus, both KRAS
and TAK1 suppression selectively suppress b-catenin-mediated
transcription and Wnt target gene expression in KRAS-depen-
Because activation of Wnt signaling is associated with nuclear
translocation of b-catenin, we analyzed its subcellular localiza-
tion following TAK1 suppression by immunofluorescence
microscopy. Parental- and vehicle-treated KRAS-dependent
SW1116 and SK-CO-1 cells show nuclear b-catenin localization,
in addition to its colocalization with E-cadherin at adherens junc-
tions. TAK1 inhibition in these cells resulted in loss of nuclear
b-catenin within 24 hr. No such effect was seen in KRAS-inde-
pendent LS174T and SW1463 cells (Figures 5D and S4B).
Thus, inhibition of TAK1 signaling causes reduced b-catenin
nuclear localization in KRAS-dependent cells, but not in KRAS-
Reconstitution of TAK1 Dependency through KRAS
and Wnt Activation
To determine whether TAK1-independent cells could be driven
toward TAK1 dependency by enhanced KRAS/Wnt signaling,
we undertook a series of reconstitution experiments. HT29 and
C2BBe1 colon cancer cells, with mutant APC and wild-type
KRAS, exhibit very little basal TCF/LEF reporter activity and
demonstrate low or undetectable nuclear b-catenin signal
(Figures 5H and S4D). These cell lines are insensitive to 5Z-7-
oxozeaenol. To determine whether activation of KRAS is
sufficient to increase Wnt signaling and hence lead to sensi-
tivity to TAK1 inhibition, we ectopically introduced mutant
KRAS(G12V) in HT29 cells through phosphoglycerate kinase
(PGK) promoter-driven expression. Expression of either the 4A
or 4B splice isoforms of mutant KRAS in these cells resulted in
a 3-fold reduction in the IC50for 5Z-7-oxozeaenol (Figures S4C
and 5E). In contrast, ectopic expression of mutant NRAS at
equivalent expression levels caused slightly increased resis-
tance to 5Z-7-oxozeaenol.
Figure 4. Associations between the KRAS Depen-
dency Gene Set, TAK1 Dependence, and KRAS-
Driven Canonical Wnt Signaling in Colon Cancer
(A) Heat map representation of gene expression most
correlated with TAK1 dependence from the KRAS
dependency gene set across a panel of colon cancer cell
lines of various genotypes. Cell lines are ordered by IC50
values for 5Z-7-oxozeaenol, leftmost being the highest
and rightmost being the lowest. Clustering of genes was
performed with Euclidean distance as a similarity metric.
Values are presented as log2 median-centered intensities.
Genes highlighted in orange text are putative or bona fide
TCF4 target genes.
(B) Basal normalized TCF4 luciferase reporter activity
(TOP-FLASH) in nominal units for a panel of KRAS-inde-
pendent and KRAS-dependent colon cancer cell lines.
Data are represented as the means of three independent
experiments ± SEM.
(C) Average expression of non-TCF4 or TCF4 target genes
depicted in Figure 4A in colon cancer patients genotyped
as either APC mutant/KRAS-wild-type (red circles) or APC
mutant plus KRAS mutant (green squares). p values
represent a comparison of mean expression scores of
genes for each class.
See also Figure S3.
644 Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc.
The increased TAK1 dependency resulting from ectopic
mutant KRAS in HT29 cells was correlated with 5-fold upregu-
lated b-catenin transcriptional activity, which was blocked in a
dose-dependent manner by TAK1 inhibition (Figures 5F). Similar
results were obtained in a second KRAS wild-type, TAK1-inde-
pendent cell line (C2BBe1) (Figures S4D). Ectopic expression
of mutant KRAS resulted in increased TAK1 autophosphoryla-
tion, Erk phosphorylation, elevated Axin 2 levels, and nuclear
b-catenin localization (Figures 5G and 5H). TAK1 inhibition
reversed the KRAS-induced b-catenin nuclear localization.
Finally, pretreatment of cells with 5Z-7-oxozeaenol, prior to their
transduction with mutant KRAS, abrogated the KRAS-mediated
increase in Wnt signaling (Figure S4E). Taken together, ectopic
expression of mutant KRAS is sufficient to activate TAK1 in
APC-deficient cells, leading to increased Wnt signaling and
sensitization to TAK1 inhibition.
Figure 5. KRAS and TAK1 Regulate Canonical Wnt Signaling in KRAS-Dependent Cancer Cells
(A) TOP-FLASH luciferase reporter activity as a function of lentiviral shRNA-mediated KRAS depletion at increasing MOIs in LS174T/SW1463 (KRAS-inde-
pendent) versus SW620/SK-CO-1 (KRAS-dependent) cells. Cell lineswere transducedto stably express luciferaseunder thecontrolof TCF4 responseelements.
Reporter activity is plotted relative to shGFP (vector)-expressing cells. Data are represented as the mean of triplicate experiments ± SEM.
(B)TOP-FLASHactivityinKRAS-independent andKRAS-dependentcelllinesfollowing TAK1inhibitionwithincreasingconcentrationsof5Z-7-oxo(mM).Dataare
represented as means of triplicate experiments ± SEM.
(C) Protein expression levels of theendogenous Wnt target gene Axin 2 following treatment of cells with the indicated concentrationsof 5Z-7-oxo.GAPDH serves
as a loading control.
(D) Laser confocal micrographs of SW1116 KRAS-dependent cells treated with either DMSO vehicle or 5mM 5Z-7-oxo for 24 hr. E-cadherin localization is shown
in the red channel, b-catenin in green, and DAPI-stained nuclei in blue. Scale bar, 20 mM.
5Z-7-oxozeaenol. Expression levels of exogenous and endogenous Ras proteins are shown by immunoblotting with a pan-ras monoclonal antibody. NRAS/
KRAS4B are HA-tagged, and KRAS4A is V5-tagged.
(F) Overexpression of mutant KRAS(12V) followed by TAK1 inhibition in HT29 cells and effects on TOP-FLASH reporter activity. Data are presented as the means
of three independent experiments ± SEM.
(H) Confocal micrographs showing E-cadherin or KRAS (red) and b-catenin (green) localization in vector control or oncogenic HA-tagged KRAS-4B(12V)-
expressing HT29 cells. KRAS expression is visualized using an HA polyclonal antibody. Scale bar, 25mm.
See also Figure S4.
Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc. 645
To further test the role of Wnt signaling in this context, we
investigated two KRAS mutant pancreatic cancer (PDAC) cell
lines, PANC-1 and YAPC, which are APC wild-type. PANC-1
cells are KRAS independent, whereas YAPC cells are KRAS
dependent, a distinction that has been linked to increased
KRAS signaling in YAPC cells (Singh et al., 2009). Activation of
canonical Wnt signaling by inhibition of GSK-3 using the selec-
tive inhibitor BIO caused strong, dose-dependent TOP-FLASH
reporter induction in KRAS-dependent YAPC cells, compared
to weak induction in the KRAS-independent PANC-1 cells (Fig-
ure S4F). Simultaneous treatment of YAPC cells with 5Z-7-
oxozeaenol abrogates the BIO-mediated TOP-FLASH induction.
In contrast, PANC-1 cells undergo stronger induction of TOP-
FLASH activity with combined GSK-3 and TAK1 inhibition
(Figure S4G, lower upper panel). Though the viability of KRAS-
dependent YAPC cells was greatly suppressed by combined
GSK-3 and TAK1 inhibition, no such effect was seen with
activated mutant of b-catenin (CTNNB1-CA) containing S33Y
and S45A missense mutations into KRAS-dependent SW620
cells. This mutant caused partial losses in both KRAS and
TAK1 dependencies (Figures S6A–S6C). Takenaltogether,these
reconstitution studies indicate that KRAS and Wnt pathway
hyperactivation together contribute to TAK1 dependency.
KRAS Activates TAK1 through Enhanced BMP Signaling
TAK1 encodes an effector of the BMP receptor, which is acti-
vated in response to BMP ligand binding. Our TAK1 dependency
signature is notably enriched for TGF-b/BMP pathway compo-
nents, including BMP7, BAMBI, and INHBB (Figure 4A). To test
of TAK1, we measured the expression of the BMP receptor
ligand BMP7 and markers of BMP activation following KRAS
depletion. In KRAS-dependent SW620 and SK-CO-1 cells,
ated phosphorylationof itseffector Smad1,andTAK1autophos-
immunoreactive band in this context is the 40 kD isoform,
although two isoforms (40 and 75 kD)are observed and depleted
by TAK1 shRNA (Figure 2E). Finally, Axin 2 levels were sup-
pressed following KRAS depletion, indicating that KRAS
signaling enhances both BMP signaling and Wnt activation
(Figure 6A). These effects of KRAS depletion were not seen in
KRAS-independent LS-174T and SW837 cells.
Given the observed KRAS-regulated expression of BMP7
in SW620 cells, we tested the functional role of this ligand
using lentiviral shRNA-mediated
BMP-7 depletion using a panel of five different shRNAs caused
pronounced viral titer-dependent apoptosis (Figure 6B). Simi-
larly, knockdown of BMPR1A, encoding the BMP receptor
type 1A (Alk-3) and also a component of the KRAS dependency
gene set, suppressed proliferation and viability of the KRAS-
dependent SW620 cells (Figure 2D). In KRAS-independent
SW837 cells, we did not observe significant proliferation or
viability defects following BMP7 depletion (Figures S5A and
S5B). Thus, autocrine BMP-7 ligand expression and receptor
activation are required to maintain the viability of KRAS-depen-
knockdown (Figure 6B).
To determine whether BMP-7 induction is a direct conse-
quence of KRAS activation, as opposed to an indirect effect of
cell transformation, we introduced an inducible mutant KRAS-
estrogen receptor chimera (ER-KRAS(12V)) into HT29 cells,
which normally express wild-type endogenous KRAS. At 24 hr
following KRAS induction using 4-hydroxytamoxifen (4-HT),
BMP7 mRNA levels were increased, along with cellular and
secreted BMP-7 protein levels (Figure 6C). Endogenous levels
of Axin 2 were also increased following KRAS induction, as
was TOP-FLASH reporter activity. The activation of these down-
stream markers of Wnt signaling by inducible KRAS was effec-
tively suppressed by depletion of BMP7, BMPR1A, and TAK1
(Figure 6D), consistent with their function as key mediators of
the KRAS-potentiated Wnt pathway activation.
To further define the role of BMP signaling in TAK1 depen-
dency, we ectopically expressed a constitutively activated (CA)
variant (Q233D) of BMPR1A (Zou et al., 1997) in HT29 cells.
Expression of BMPR1A-CA conferred increased sensitivity to
5Z-7-oxozeaenol with an IC50value of 1.1 mM compared to
7.7 mM for vector control cells (Figure 6E). TAK1 inhibition in
BMPR1A-CA-expressing cells resulted in apoptosis, as shown
by caspase-3 and PARP cleavage (Figure 6F). This was accom-
panied by dose-dependent decreases in Axin 2 and phos-
phorylated Smad2 levels. Finally, BMPR1A-CA induced nuclear
accumulation of b-catenin, which was suppressed following
TAK1 inhibition (Figure 6G). Thus, in APC mutant cells with low
baseline b-catenin transcriptional activity, artificial BMP activa-
tion enhances Wnt signaling through TAK1 activation, conferring
TAK1 dependency. Surprisingly, introduction of BMPR1A-CA
into KRAS-dependent SW620 cells could not rescue KRAS
depletion-induced cell death (Figure 6D), suggesting that BMP
receptor activation may be necessary, but not sufficient, to pro-
mote cell survival in KRAS-dependent cancer cells. In summary,
mutant KRAS promotes autocrine BMP signaling, causing TAK1
activation and leading to enhanced Wnt signaling in APC-defi-
cient colon cancer cells. In KRAS-dependent cells, all of these
components are necessary for full antiapoptotic signaling.
cancer cells, we uncovered a pathway by which KRAS enhances
WntactivitythroughBMP/TAK1 activation.Approximately halfof
colon cancer cell lines with both KRAS and APC mutations
appear to rely on this pathway for viability, rendering them sensi-
tive to TAK1 kinase inhibition. As such, TAK1 inhibition may
provide a clinical paradigm for context-dependent targeting of
KRAS-dependent colon cancers. Our data suggest that TAK1
functions as a prosurvival mediator in cancer cells displaying
hyperactive KRAS-dependent Wnt signaling. This is seen under
basal conditions in colon cancers with the relevant genotypes or
can be synthetically achieved by activating Wnt signaling via
GSK3 kinase inhibition in KRAS-dependent/APC-wild-type
pancreatic cancer cells and by enforced expression of mutant
KRAS in APC mutant/KRAS wild-type colon cancer cells. The
ability to reconstitute such pathway dependency is unusual
in ‘‘oncogene addiction’’ models and facilitates molecular
dissection of the critical signaling components that drive drug
646 Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc.
Figure 6. Oncogenic KRAS Regulates BMP-7/BMPR1A Signaling
(A) Depletion of KRAS in two KRAS-independent (LS-174T and SW837) and two KRAS-dependent cell lines (SW620 and SK-CO-1) and subsequent effects on
expression ofBMP-7aswellasdownstreameffects onSmad1/TAK1 phosphorylation(p-Smad1/p-TAK1).The20kDsecreted formofBMP7 isshown.Phospho-
TAK1representstheTAK1autophosphorylation siteandisameasureofTAK1activity.TotalSmad1/5/8 andtotalTAK1(t-Smad1/5/8/t-TAK1) proteins areshown
as gel loading controls. Data are representative of two independent experiments.
(B) Effects of BMP7 depletion on proliferation and viability of SW620 KRAS-dependent cells. Plot shows cell density 6 days postinfection with either shGFP
control or five different BMP7-directed lentiviral shRNAs. Data are represented as the mean of three independent experiments ± SEM. Western blots on the right
panel show BMP-7 levels and apoptotic effects as measured by PARP and caspase-3 cleavage following BMP-7 depletion with two independent lentiviral
shRNAs (D and E).
(C) Effects on BMP7 protein and transcript levels following induced activation of ER-KRAS(12V) fusion protein with various doses of 4-HT in HT29 cells. (Left)
KRAS(12V). Total Erk (t-Erk1) serves as a loading control.
(D) TOP-FLASH reporter activity following 4-HT-induced activation of ER-KRAS(12V) and depletion of the indicated genes via lentiviral shRNA delivery at various
viral titers. Reporter activity is shown relative to shGFP control.
(E) Introduction of a V5-tagged constitutively activated (CA) mutant of the BMP receptor BMPR1A (Q233D) or control vector in HT29 cells and effects on 5Z-7-
oxozeaenol sensitivity in terms of IC50values.
(F) Signaling and apoptotic effects of TAK1 inhibition using 5Z-7-oxozeaenol at the indicated concentrations 24 hr posttreatment in BMPR1A-CA-expressing
cells. Caspase-3 and PARP cleavage are indicators of apoptotic cell death. Axin 2 levels are shown as a readout of Wnt signaling. Phosphorylated smad1/5/8
levels serve as a readout of BMP signaling. GAPDH serves as a gel loading control. BMPR1A-CA expression is visualized using a monoclonal V5 antibody.
(G) Effects of BMPR1A-CA expression on b-catenin localization (red) in HT-29 cells following treatment with 5 mM 5Z-7-oxozeaenol or vehicle control for 24 hr, as
assessed by immunofluorescence confocal microscopy. DAPI-stained nuclei are shown in blue. Scale bar, 10 mM.
See also Figure S5.
Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc. 647
susceptibility. An underlying basis for this may be explained by
the emerging concept of ‘‘non-oncogene addiction,’’ describing
the acquired dependence of cells on nonmutated genes that do
not themselves drive malignant progression but whose function
is essential for a cell to tolerate other oncogenic stress-induced
states (Luo et al., 2009b). Though TAK1 dependency may not be
restricted to colon cancer, the elevated Wnt signaling activity in
KRAS-dependent colon cancer cells highlights the importance
of cellularcontext andtherole oflineage-specific pathways inin-
forming an effective therapeutic strategy.
Through a combination of knockdown and reconstitution
experiments, we have attempted to dissect the key signaling
components linking mutant KRAS to TAK1 and Wnt activation
(Figures S6 and 7). A strong relationship between Wnt and
KRAS signaling is underscored by the observation that constitu-
in KRAS-dependent colon cancer cells. Our data indicate that
KRAS regulates TAK1 and Wnt signaling in APC-deficient
cells via upregulation of BMP-7 levels and BMP receptor activa-
tion. However, although BMP receptor activation is necessary
for KRAS-driven survival signaling, it is not sufficient. Thus,
a network of KRAS-regulated signaling components is likely to
contribute to tumor cell survival. For instance, KRAS regulates
NF-kB, in part, via TAK1 activation (Figures S6E and 7). The rela-
tive contribution of the NF-kB pathway to KRAS-driven survival
signaling remains to bedetermined, althoughevidence suggests
that the pathway is critical for KRAS-driven lung tumorigenesis
(Meylan et al., 2009; Starczynowski et al., 2011). Additional
parallel pathways are likely to be components of the KRAS-
TAK1 survival signaling axis, although our data suggest that
Wnt pathway activation is most critical.
Our studies highlight a context-specific role for KRAS in
driving Wnt signaling in the sensitized background of APC
deficiency. This is consistent with recent studies reporting
KRAS-mediated enhancement of Wnt signaling in a zebrafish
developmental model (Phelps et al., 2009). Indeed, we observed
that, in APC-deficient colon cancers with low b-catenin activity,
introduction of mutant KRAS causes a sharp increase in levels
of nuclear b-catenin, accompanied by increased TCF/LEF
transcriptional activity. This effect partly involves KRAS-medi-
ated upregulation of BMP signaling and subsequent TAK1 acti-
vation, leading to enhanced TCF/LEF activity. Interestingly, the
C. elegans TAK1 ortholog Mom-4 promotes nuclear retention
of the b-catenin ortholog Wrm-1 asymmetrically at the 2-cell
stage within the EMS cell, thus defining polarity and axis speci-
fication (Nakamura et al., 2005; Shin et al., 1999). Such
able degree of evolutionary conservation.
From a clinical perspective, the role of secreted BMP-7 is of
particular interest because autocrine or paracrine activation
of this pathway could be detectable and targetable in tumors.
Importantly, expression of BMP pathway components should
help to stratify colon cancer patients into TAK1 inhibitor
response groups. To that end, we suggest that the top ten genes
from the in vitro-derived TAK1 dependency signature (GGH,
BMP7, BAMBI, MBOAT2, HSPA12A, FYN, NAV2, RGL1, SYK,
and RUNX1) will provide a clinically annotated signature once
TAK1 inhibitors are sufficiently developed for clinical trials. This
could eventually be applied as a clinical diagnostic test to
measure the relative mRNA levels corresponding to the ten-
gene TAK1 dependency signature in patient tumors. We note
that as many as half of all KRAS mutant colon cancer cell lines
are KRAS dependent and sensitive to TAK1 inhibition, which
may account for as many as a quarter of all colon cancers. As
such, when guided by accurate molecular profiles, TAK1 inhibi-
tors may provide significant clinical benefit for the most recalci-
trant form of colon cancer. Beyond tool compounds such as
5Z-7-oxozeaenol, synthetic TAK1 inhibitors have been tested
in preclinical models (Melisi et al., 2011). However, given poten-
tial toxicity, administration regimens will need to be modeled
using highly TAK1-dependent cancers. Finally, our study illus-
trates that the presence of a KRAS mutation does not identify
histological type. Instead, degrees of KRAS dependency in
different cancers are modulated by associated signaling path-
ways such as the Wnt pathway in colon cancers. This adds
complexity to their analysis but is ultimately expected to inform
unique therapeutic opportunities.
Derivation of the Ras Dependency Index
Weighted averages for relative cell densitiesfor MOIs of 5 and 1 with the KRAS
A and B shRNAs were calculated. The inverse of these averages was then
calculated. This number was multiplied by the transduction efficiency for
each respective cell line (the proportion of cells expressing the control shRNA
Figure 7. A Model for Context-Specific KRAS Dependency in Colon
In KRAS-independent colon cancers, APC loss of function results in hyper-
activation of canonical Wnt signaling through stabilization of b-catenin in
canonical Wnt signaling in these cells. In KRAS-dependent cells, oncogenic
KRAS upregulates BMP-7 expression/secretion, activating the BMP receptor
and resulting in TAK1 activation. KRAS and TAK1 in these cells are activators
of Wnt signaling by promoting b-catenin nuclear localization, which is stabi-
lized by virtue of APC loss-of-function mutations. KRAS-mediated anti-
apoptotic signaling could also be facilitated by NF-kB activation. Dashed lines
represent unknown molecular interactions.
See also Figure S6.
648 Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc.
following puromycin selection compared with those not treated with puro-
mycin), yielding the RDI value. An RDI of 2 is calibrated as a 50% reduction
in cellular proliferation following KRAS depletion.
Gene Expression Microarray Analyses
U133A Microarrays. The data set for the colon cancer cell lines used in this
study ispubliclyavailable via the
Project. Expression data were normalized using GCRMA (Bolstad et al.,
2003). To derive the KRAS dependency gene set, p values were computed
comparing average normalized probe intensity for each probe set between
the cell lines shown in Figure 2A. Fold differences for the average probe inten-
sities were calculated. The p value was log transformed, and fold difference
was log2 transformed. The product of these log-transformed values was
designated as the ‘‘DEP SCORE’’ (Table S1), and genes were ranked based
ence > 2) were identified by this method. To generate heatmaps, complete
linkage hierarchical clustering by Euclidean or city block distance was per-
formed using Cluster (Eisen et al., 1998) and Java Treeview (http://
sourceforge.net/projects/jtreeview/). Statistical analyses of correlations with
gene expression data were performed with Graphpad Prism.
In Vivo Pharmacology with Xenografted Mouse Tumors
Human colorectal cancer tumor cells were trypsinized and resuspended as
single cell suspensions and injected into left and right flanks of NOD/SCID
General Hospital and were handled in strict accordance with Good Animal
Practice as defined by the Office of Laboratory Animal Welfare, and all animal
experiments were done with approval from Massachusetts General Hospital
Subcommittee on Research Animal Care. Tumor size was monitored daily,
and once tumor volume had reached ?200 mm3, treatment with 5Z-7-
oxozeaenol was initiated (7–14 days postimplantation). Mice were injected
daily with 15 mg/kg 5Z-7-oxozeaenol via intraperitoneal administration.
0.5 mg FOP-FLASH or TOP-FLASH plasmids or pGL3-kB-LUC plus 50 ng of
pRL-TK (expressing Renilla luciferase). Normalized luciferase activity was
obtained by using the Dual-Luciferase Reporter Assay System (Promega
Inc). Alternatively, stable cell lines expressing the TOP-FLASH reporter were
generated by transducing cells with 7TFP (Fuerer and Nusse, 2010) recombi-
nant lentiviruses followed by drug selection.
Patient data are available through the NCBI GEO database (accession
Supplemental Information includes Extended Experimental Procedures,
six figures, and two tables and can be found with this article online at
This work was supported by NIH K99 CA149169 to A.S. as well as NIH R01
CA109447 and a Lustgarten Foundation grant to J.S. and NIH R01
CA129933 to D.A.H. We are grateful to Randall Peterson, David Ting, and
Andre Bernards for critical evaluation and comments on the manuscript. We
thank Laura Libby and Brian Brannigan for technical assistance with mouse
xenograft experiments and KRAS/BRAF sequencing analyses, respectively.
We thank Vijay Yajnik for providing TOP-FLASH reporter plasmids. Finally,
we also thank Kevin Haigis, Moon Yee Hang, Miguel Rivera, Jianmin Zhang,
Mingzhu Liu, and Matt Zubrowski for intellectual and technical advice.
Received: June 17, 2011
Revised: October 28, 2011
Accepted: December 29, 2011
Published: February 16, 2012
Barbie, D.A., Tamayo, P., Boehm, J.S., Kim, S.Y., Moody, S.E., Dunn, I.F.,
Schinzel, A.C., Sandy, P., Meylan, E., Scholl, C., et al. (2009). Systematic
RNA interference reveals that oncogenic KRAS-driven cancers require
TBK1. Nature 462, 108–112.
Behrens, J., von Kries, J.P., Ku ¨hl, M., Bruhn, L., Wedlich, D., Grosschedl, R.,
and Birchmeier, W. (1996). Functional interaction of beta-catenin with the tran-
scription factor LEF-1. Nature 382, 638–642.
Bolstad, B.M., Irizarry, R.A., Astrand, M., and Speed, T.P. (2003). A compar-
ison of normalization methods for high density oligonucleotide array data
based on variance and bias. Bioinformatics 19, 185–193.
Campeau, E., Ruhl, V.E., Rodier, F., Smith, C.L., Rahmberg, B.L., Fuss, J.O.,
Campisi, J., Yaswen, P., Cooper, P.K., and Kaufman, P.D. (2009). A versatile
viral system for expression and depletion of proteins in mammalian cells.
PLoS ONE 4, e6529.
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.
Downward, J. (2003). Targeting RAS signalling pathways in cancer therapy.
Nat. Rev. Cancer 3, 11–22.
Ebi, H., Corcoran, R.B., Singh, A., Chen, Z., Song, Y., Lifshits, E., Ryan, D.P.,
Meyerhardt, J.A., Benes, C., Settleman, J., et al. (2011). Receptor tyrosine
kinases exert dominant control over PI3K signaling in human KRAS mutant
colorectal cancers. J. Clin. Invest. 121, 4311–4321.
Eisen, M.B., Spellman, P.T., Brown, P.O., and Botstein, D. (1998). Cluster
analysis and display of genome-wide expression patterns. Proc. Natl. Acad.
Sci. USA 95, 14863–14868.
Wnt signaling pathway. PLoS ONE 5, e9370.
Gasch, A.P., and Eisen, M.B. (2002). Exploring the conditional coregulation of
yeast gene expression through fuzzy k-means clustering. Genome Biol. 3,
Haber, D.A., Gray, N.S., and Baselga, J. (2011). The evolving war on cancer.
Cell 145, 19–24.
Hatzis, P.,van der Flier, L.G., vanDriel, M.A., Guryev, V.,Nielsen, F., Denissov,
S., Nijman, I.J., Koster, J., Santo, E.E., Welboren, W., et al. (2008). Genome-
wide pattern of TCF7L2/TCF4 chromatin occupancy in colorectal cancer cells.
Mol. Cell. Biol. 28, 2732–2744.
Janssen, K.P., Alberici, P., Fsihi, H., Gaspar, C., Breukel, C., Franken, P.,
Rosty, C., Abal, M., El Marjou, F., Smits, R., et al. (2006). APC and oncogenic
KRAS are synergistic in enhancing Wnt signaling in intestinal tumor formation
and progression. Gastroenterology 131, 1096–1109.
Lin, Z., Gao, C., Ning, Y., He, X., Wu, W., and Chen, Y.G. (2008). The pseudor-
eceptor BMP and activin membrane-bound inhibitor positively modulates
Wnt/beta-catenin signaling. J. Biol. Chem. 283, 33053–33058.
Luo, J., Emanuele, M.J., Li, D., Creighton, C.J., Schlabach, M.R., Westbrook,
T.F., Wong, K.K., and Elledge, S.J. (2009a). A genome-wide RNAi screen iden-
tifies multiple synthetic lethal interactions with the Ras oncogene. Cell 137,
Luo, J., Solimini, N.L., and Elledge, S.J. (2009b). Principles of cancer therapy:
oncogene and non-oncogene addiction. Cell 136, 823–837.
Lustig, B., Jerchow, B., Sachs, M., Weiler, S., Pietsch, T., Karsten, U., van de
Wetering, M., Clevers, H., Schlag, P.M., Birchmeier, W., and Behrens, J.
(2002). Negative feedback loop of Wnt signaling through upregulation of con-
ductin/axin2 in colorectal and liver tumors. Mol. Cell. Biol. 22, 1184–1193.
Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc. 649
Melisi, D., Xia, Q., Paradiso, G., Ling, J., Moccia, T., Carbone, C., Budillon, A.,
Abbruzzese, J.L., and Chiao, P.J. (2011). Modulation of pancreatic cancer
chemoresistance by inhibition of TAK1. J. Natl. Cancer Inst. 103, 1190–1204.
Meylan, E., Dooley, A.L., Feldser, D.M., Shen, L., Turk, E., Ouyang, C., and
Jacks, T. (2009). Requirement for NF-kappaB signalling in a mouse model of
lung adenocarcinoma. Nature 462, 104–107.
Moffat, J., Grueneberg, D.A., Yang, X., Kim, S.Y., Kloepfer, A.M., Hinkle, G.,
Piqani, B., Eisenhaure, T.M., Luo, B., Grenier, J.K., et al. (2006). A lentiviral
RNAi library for human and mouse genes applied to an arrayed viral high-
content screen. Cell 124, 1283–1298.
Nakamura, K., Kim, S.,Ishidate, T.,Bei, Y.,Pang, K., Shirayama,M.,Trzepacz,
C., Brownell, D.R., and Mello, C.C. (2005). Wnt signaling drives WRM-1/beta-
catenin asymmetries in early C. elegans embryos. Genes Dev. 19, 1749–1754.
Naldini, L., Blo ¨mer, U., Gallay, P., Ory, D., Mulligan, R., Gage, F.H., Verma,
I.M., and Trono, D. (1996). In vivo gene delivery and stable transduction of
nondividing cells by a lentiviral vector. Science 272, 263–267.
Normanno, N., Tejpar, S., Morgillo, F., De Luca, A., Van Cutsem, E., and
Ciardiello, F. (2009). Implications for KRAS status and EGFR-targeted thera-
pies in metastatic CRC. Nat. Rev. Clin. Oncol. 6, 519–527.
Petrova, T.V., Nyka ¨nen, A., Norrme ´n, C., Ivanov, K.I., Andersson, L.C.,
Haglund, C., Puolakkainen, P., Wempe, F., von Melchner, H., Gradwohl, G.,
et al. (2008). Transcription factor PROX1 induces colon cancer progression
by promoting the transition from benign to highly dysplastic phenotype.
Cancer Cell 13, 407–419.
Phelps, R.A., Chidester, S., Dehghanizadeh, S., Phelps, J., Sandoval, I.T., Rai,
K., Broadbent, T., Sarkar, S., Burt, R.W., and Jones, D.A. (2009). A two-step
model for colon adenoma initiation and progression caused by APC loss.
Cell 137, 623–634.
Rawlins, P., Mander, T., Sadeghi, R., Hill, S., Gammon, G., Foxwell, B.,
Wrigley, S., and Moore, M. (1999). Inhibition of endotoxin-induced TNF-alpha
production in macrophages by 5Z-7-oxo-zeaenol and other fungal resorcylic
acid lactones. Int. J. Immunopharmacol. 21, 799–814.
Reid, J.F., Gariboldi, M., Sokolova, V., Capobianco, P., Lampis, A., Perrone,
tive approach for prioritizing cancer genes in sporadic colon cancer. Genes
Chromosomes Cancer 48, 953–962.
Scholl, C., Fro ¨hling, S., Dunn, I.F., Schinzel, A.C., Barbie, D.A., Kim, S.Y.,
Silver, S.J., Tamayo, P., Wadlow, R.C., Ramaswamy, S., et al. (2009).
Synthetic lethal interaction between oncogenic KRAS dependency and
STK33 suppression in human cancer cells. Cell 137, 821–834.
Sekiya, T., Adachi, S., Kohu, K., Yamada, T., Higuchi, O., Furukawa, Y., Naka-
mura, Y., Nakamura, T., Tashiro, K., Kuhara, S., et al. (2004). Identification of
ing growth factor-beta signaling, as a target of the beta-catenin pathway in
colorectal tumor cells. J. Biol. Chem. 279, 6840–6846.
Shin, T.H., Yasuda, J., Rocheleau, C.E., Lin, R., Soto, M., Bei, Y., Davis, R.J.,
and Mello, C.C. (1999). MOM-4, a MAP kinase kinase kinase-related protein,
activates WRM-1/LIT-1 kinase to transduce anterior/posterior polarity signals
in C. elegans. Mol. Cell 4, 275–280.
Singh, A., Greninger, P., Rhodes, D., Koopman, L., Violette, S., Bardeesy, N.,
and Settleman, J. (2009). A gene expression signature associated with ‘‘K-Ras
addiction’’ reveals regulators of EMT and tumor cell survival. Cancer Cell 15,
Starczynowski, D.T., Lockwood,W.W.,Dele ´houze ´e, S.,Chari, R., Wegrzyn, J.,
Fuller, M.,Tsao,M.S., Lam, S., Gazdar, A.F., Lam, W.L., and Karsan, A. (2011).
TRAF6 is an amplified oncogene bridging the RAS and NF-kB pathways in
human lung cancer. J. Clin. Invest. 121, 4095–4105.
Vogelstein, B., Fearon, E.R., Hamilton, S.R., Kern, S.E., Preisinger, A.C.,
Leppert, M., Nakamura, Y., White, R., Smits, A.M., and Bos, J.L. (1988).
Genetic alterations during colorectal-tumor development. N. Engl. J. Med.
Whyte, D.B., Kirschmeier, P., Hockenberry, T.N., Nunez-Oliva, I., James, L.,
Catino, J.J., Bishop, W.R., and Pai, J.K. (1997). K- and N-Ras are geranylger-
anylated in cells treated with farnesyl protein transferase inhibitors. J. Biol.
Chem. 272, 14459–14464.
Xie, M., Zhang, D., Dyck, J.R., Li, Y., Zhang, H., Morishima, M., Mann, D.L.,
Taffet, G.E., Baldini, A., Khoury, D.S., and Schneider, M.D. (2006). A pivotal
role for endogenous TGF-beta-activated kinase-1 in the LKB1/AMP-activated
protein kinase energy-sensor pathway. Proc. Natl. Acad. Sci. USA 103,
Zou, H., Wieser, R., Massague ´, J., and Niswander, L. (1997). Distinct roles of
type I bone morphogenetic protein receptors in the formation and differentia-
tion of cartilage. Genes Dev. 11, 2191–2203.
650 Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc.
EXTENDED EXPERIMENTAL PROCEDURES
Derivation of the Ras Dependency Index
Weightedaveragesforrelativecelldensities forMOIsof5and1withtheKRASAandBshRNAswerecalculated. Theinverseofthese
averages was then calculated. This number was multiplied by the transduction efficiency for each respective cell line (the proportion
of cells expressing the control shRNA following puromycin selection compared those not treated with puromycin), yielding the RDI
value. An RDI of 2 is calibrated as a 50% reduction in cellular proliferation following KRAS depletion.
Lentiviral shRNA Experiments
293T cells were seeded (3ml at density of 2 3 105cells per ml) in duplicate wells of a 6 well plate per shRNA construct. Constructs
were from the Broad RNAi Consortium. Lentiviral particles were generated using a three-plasmid system, as described previously
(Moffat et al., 2006; Naldini et al., 1996). To standardize lentiviral transduction assays, viral titers were measured in a benchmark
cell line, A549. For growth assays, titers corresponding to multiplicities of infection (MOIs) of 5 and 1 in A549 cells were employed.
For KRAS knockdown, cells were plated on day zero at 3x104cells/ml in 96 well plates (100 ml per well) or 6 well plates (3 ml per well).
Cells were spin infected, as described previously (Moffat et al., 2006). 24 hr post-infection, cells were treated with 1 mg/ml puromycin
for 3 days to eliminate uninfected cells. Media was replaced and cells were grown for 2 more days, then fixed with 4% formaldehyde
and stained with 1 mM Syto60 dye (Invitrogen Inc) for 1 hr. Syto60 fluorescence was quantified with a Li-Cor fluorescence scanner in
Vanadate and 25mM NaF). Lysates were normalized for total protein using Pierce BCA reagent and resolved by SDS-PAGE followed
by transfer to PVDF.
KRAS and BRAF Genotyping
To determine the mutation states of KRAS and BRAF in colorectal cancer cell lines used in this study, total RNA was extracted from
cells with the RNeasy Kit (QIAGEN). RNA was reverse transcribed with an Applied Biosystems Reverse Transcriptase Kit. KRAS
exon4 was sequenced from cDNA with the following primers: forward: CCA TTT CGG ACT GGG AGC GAG C and reverse: CCT
ACT AGG ACC ATA GGT ACA TCT TC. BRAF exon15 was sequenced with TCA TAA TGC TTG CTC TGA TAG GA (F) and GGC
CAA AAA TTT AAT CAG TGG (R).
In Vivo Pharmacology with Xenografted Mouse Tumors
Human colorectal cancer tumor cells were trypsinized and resuspended as single cell suspensions at 3x107cells per ml in PBS.
100mL (3x106cells total) of this suspension were injected into opposite left and right flanks of NOD/SCID mice. All mice were housed
in a pathogen-free environment at the Massachusetts General Hospital and were handled in strict accordance with Good Animal
Practice as defined by the Office of Laboratory Animal Welfare, and all animal experiments were done with approval from Massachu-
setts General Hospital Subcommittee on Research Animal Care. Tumor size was monitored daily and once tumor volume had
reached approximately 200mm3, treatment with 5Z-7-oxozeaenol was initiated (7 to 14 days post-implantation). Mice were injected
daily with 15mg/kg 5Z-7-oxozeaenol. The drug was resuspended as a 25mg/ml stock in DMSO. This was further diluted 10-fold in
Arachis Oil (Sigma Inc.) to yield a 2.5mg/ml stock in 10% DMSO. Approximately 120ml of this stock was delivered to 20 g mice intra-
peritoneally. Alternatively, 10% DMSO in Arachis Oil was delivered as a vehicle control.
Western Blotting and Antibodies
The following antibodies were used for Western blotting: KRAS OP-24, Pan-Ras OP-40 (Calbiochem); PARP (BD Pharmigen,
4C10-5); BMP-7 (Abcam); phospho-ERK, Axin2, phospho Smad1 and total Smad1/5/8, phospho- and total AMPK, phospho- and
total AKT, cleaved Caspase-3 (Cell Signaling); GAPDH (Chemicon); E-Cadherin, b-catenin (BD Pharmigen) Syk, TAK1, total ERK1
(Santa Cruz). For secreted BMP-7 levels, 1x106HT29 cells stably expressing ER-KRAS(12V) were plated in 10cm dishes. 24h
post-plating, 10ml serum-free DME/F12 medium (GIBCO) was added. Conditioned media was collected 24h post-induction of
ER-KRAS(12V) with 4-HT and concentrated to 500mL using Amicon? Ultra-4 Centrifugal Filter Units with 3kDa membranes. To
assess BMP-7 levels, 60mL of this concentrated conditioned medium was used for Western blotting.
TOP-FLASH Reporter Assays
For data shown in Figure 4B, cells were plated in 12-well tissue culture plates at a density of 5x104cells/ml and 1ml per well. Cells
were then co-transfected with either 0.5mg FOP-FLASH or TOP-FLASH plasmids (kind gift from Vijay Yajnik, Massachusetts General
Hospital) plus 50ng of pRL-TK (expressing Renilla luciferase). Normalized luciferase activity was obtained by using the Dual-Lucif-
erase Reporter Assay System (Promega Inc). For data shown elsewhere, stable cell lines expressing the TOP-FLASH reporter
were generated by transducing cells with 7TFP (see below) recombinant lentiviruses and selecting with 2mg/ml puromycin for 5 days.
Lentiviral vectors were used throughout this study. 7TFP, encoding a 7xTcf-FFLUC was obtained from Addgene (Fuerer and Nusse,
2010). Human epitope-tagged HA-NRAS(G12V), V5-KRAS4A(G12V) and HA-KRAS4B(G12V) were cloned into pLenti-PGK-Hygro
Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc. S1
(Campeau et al., 2009) by Gateway cloning (Invitrogen Inc.) – LR reaction from pDONR-223 Entry clones. Human BMPR1A was
cloned into pDONR-223 by Gateway Cloning with a mutated stop codon. The Q233D missense mutation was introduced using
the Stratagene QuikChange site-directed mutagenesis kit. BMPR1A(Q233D) was cloned into pLenti6-CMV-V5 by Gateway cloning
to allow for expression of a C-terminal in-frame V5 fusion construct. Constitutively-active CTNNB1 was generated via Gateway
cloning into pDONR-223 and subsequent site-directed mutagenesis to generate a S33Y/S45A double mutant. This mutant was
subcloned into the pWPI lentiviral expression.
Cells were fixed in EM grade 4% formaldehyde and permeabilized with 0.1% Triton X-100. Staining with primary antibodies was
carried out overnight at 4?C. For mouse monoclonal antibodies, an Alexa594-conjugated goat anti-mouse secondary antibody
was used (Molecular Probes). For rabbit polyclonal antibodies, Alexa-488 conjugated goat anti-rabbit secondary antibody was
used (Molecular Probes). Nuclei were visualized using DAPI. Micrographs were either captured on an IX81 Spinning Disk Deconvo-
lution Microscope equipped with 100X Plan-Apo Oil objective (Figures 5H, 6G) or a Zeiss Laser Confocal Microscope equipped with
a 63X Plan-Apo Oil objective (Figures 5D and S4B). Digital images were processed with Slidebook, Zeiss LSM Browser and Adobe
HT29, SW620 or SKCO1 cells were infected with recombinant lentiviruses encoding either BMPR1A-CA and CTNNB1-CA or vector
control (containing the ccDB gene). For BMPR1A-CA stable expression, cells were selected in 5mg/ml Blasticidin for 7 days and
pooled clones were established. Stable expression was verified using the V5 epitope tag on the BMPR1A transgene product. For
CTNNB1-CA, the pWPI recombinant lentiviruses encode GFP driven by IRES. Thus, stable cell clones were obtaining by FACS
live cell sorting to obtain the top 10% of GFP expressing cells. The SW620-CTNNB1-CA stable cell clones were passaged 1:5 every
2 days and assayed for KRAS dependency after the fifth passage.
S2 Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc.
Figure S1. Computational Analyses of KRAS Dependency in CRC Cell Lines, Related Figure 2
(A) Heatmap representation of hierarchicalclustering analysis of median-centered log2 transformedprobe intensities for the KRAS Dependency Gene Set across
a panel of 40 CRC cell lines of various genotypes.
(B) Panther Molecular Function classifications for DEP genes, using the DAVID gene ontology algorithm.
(C) KEGG pathway enrichment in the DEP genes from the KRAS Dependency Gene Set using the DAVID algorithm.
(D)Viral titration curveforSW837 and SW620 cells. Cells wereinfected withlentiviruses encoding control shGFP and treated withpuromycin toselect for infected
cells, 24h post-infection. Relative cell density was quantitated 6 days post-infection. Data are presented as the mean of three experiments +/? SEM.
(E) Representative examples of kinase knockdown assays. Scans of cells fixed and stained with syto-60 dye in 96-well plates following knockdown of indicated
kinases with 5 different shRNAs per gene (shA through shE). shGFP is shown as a control.
(F) Quantitation of well intensities for the scans shown in Figure S2A. Representative examples of effects on growth and viability following shRNA knockdown of
kinase expression for MAP3K7, VRK2 and CHUK in SW837 and SW620 cells. Five individual shRNAs were used (shA through shE), represented by different
colors. Relative cell densities are shown, normalized to shGFP control expressing cells for 3 different viral titers (MOIs of 4, 2 and 1). Data are represented as the
mean of triplicates +/? SEM.
Cell 148, 639–650, February 17, 2012 ª2012 Elsevier Inc. S3