of Epithelial-Mesenchymal Transition Is Essential
for Squamous Cell Carcinoma Metastasis
Jeff H. Tsai,1Joana Liu Donaher,3Danielle A. Murphy,4Sandra Chau,1and Jing Yang1,2,*
1Department of Pharmacology
2Department of Pediatrics
University of California, San Diego, School of Medicine, 9500 Gilman Drive, La Jolla, CA 92093-0636, USA
3Whitehead Institute for Biomedical Research, 9 Cambridge Center, Cambridge, MA 02142, USA
4The Sanford Burnham Medical Research Institute, La Jolla, CA 92037, USA
Epithelial-mesenchymal transition (EMT) is implicated in converting stationary epithelial tumor cells into
motile mesenchymal cells during metastasis. However, the involvement of EMT in metastasis is still contro-
versial, due to the lack of a mesenchymal phenotype in human carcinoma metastases. Using a spontaneous
squamous cell carcinoma mouse model, we show that activation of the EMT-inducing transcription factor
Twist1 is sufficient to promote carcinoma cells to undergo EMT and disseminate into blood circulation.
Importantly, in distant sites, turning off Twist1 to allow reversion of EMT is essential for disseminated tumor
cells to proliferate and form metastases. Our study demonstrates in vivo the requirement of ‘‘reversible EMT’’
in tumor metastasis and may resolve the controversy on the importance of EMT in carcinoma metastasis.
During metastasis, epithelial tumor cells invade surrounding
tion, and then establish secondary tumors in distant sites. A
developmental program termed epithelial-mesenchymal transi-
tion (EMT) has been implicated in giving rise to the dissemination
of single carcinoma cells. During EMT, stationary epithelial cells
lose their epithelial characteristics, including adherent junctions
and apical-basal polarity, and acquire a mesenchymal mor-
phology and the ability to migrate and invade (Hay, 1995).
Biochemically, cells switch off the expression of epithelial
markers, such as adherens junction proteins E-cadherin and
catenins, and turn on mesenchymal markers, including vimentin
and fibronectin. Studies using cell culture and tumor xenograft
models show that activation of EMT promotes carcinoma cells
to dissociate from each other and metastasize to distant organs
(Hay, 1995; Kalluri and Weinberg, 2009; Thiery, 2002, 2009).
However, the involvement of EMT in tumor metastasis in vivo
is still hotly debated (Garber, 2008; Ledford, 2011; Tarin et al.,
2005; Thompson et al., 2005). In human carcinoma, although
primary tumors show many morphological and molecular
features of EMT in subpopulations of invasive cells, distant
metastases present an epithelial morphology (Peinado et al.,
2007).This phenomenoncontradicts the assumptionthatactiva-
tion of EMT in tumor cells should result in metastases with
a mesenchymal phenotype, therefore casting doubts on the
occurrence of EMT during metastasis. This discrepancy could
be due to the interpretation of the EMT program as a permanent
nonreversible course during tumor metastasis. A reversible EMT
model has been proposed to explain this apparent paradox:
carcinoma cells undergo EMT to invade and disseminate from
the primary tumor; once reaching distant sites, tumor cells
need to revert to an epithelial identity to form macrometastases
(Thiery, 2002). However, this hypothesis has not been attested
EMT features are frequently observed in many types of primary human carcinoma, but not their corresponding metastases.
Our findings indicate that reversible EMT likely represents a key driving force in human carcinoma metastasis. Delayed
onset of metastasis following primary tumor removal is thought to be due to resurrection of latent carcinoma cells in distant
organs. Our study raises the possibility that tumor dormancy could be due to the inability of disseminated tumor cells to
revert EMT and proliferate. The dynamic involvement of EMT in metastasis cautions that therapies inhibiting EMT could
be counterproductive in preventing distant metastases when patients already present circulating tumor cells. Instead,
blocking EMT reversion may prevent dormant tumor cells from establishing metastases.
Cancer Cell 22, 725–736, December 11, 2012 ª2012 Elsevier Inc. 725
The EMT program is orchestrated through a network of tran-
scription factors, including Twist1 (Yang et al., 2004), Snail1/2
(Batlle et al., 2000; Cano et al., 2000; Hajra et al., 2002), Zeb1/2
(Comijn et al., 2001; Eger et al., 2005), and FOXC2 (Mani et al.,
2007). Our previous study found that Twist1 is a potent inducer
of EMT and invadopodia-mediated ECM degradation (Eckert
xenograft models, Twist1 expression can promote tumor metas-
tasis (Yang et al., 2004). Clinical studies have also associated
expression of Twist1 in primary tumors with disease aggressive-
ness and poor survival in many types of human cancers, such as
squamous cell carcinoma, breast cancer, prostate cancer, and
gastric cancer (Eckert et al., 2011; Kallergi et al., 2011; Peinado
et al., 2007; Watson et al., 2007).
Unlike human carcinoma metastases, most established meta-
type (Blick et al., 2008) and cannot be used to address the
dynamic EMT process during tumor metastasis in vivo. Recent
elegant studies using autochthonous mouse tumor models
observed the occurrence of EMT in primary carcinoma, but
how EMT spatiotemporally regulates metastasis has not been
investigated in these models (Hu ¨semann et al., 2008; Rhim
et al., 2012). The chemical carcinogenesis mouse skin model
has been shown to recapitulate the multistep process of human
carcinoma progression, including initiation, growth, invasion,
and metastasis (Kemp, 2005; Perez-Losada and Balmain,
2003). At the molecular and genetic levels, the skin carcinogen-
esis model shares strong similarities with a number of carcinoma
in humans, including activating mutations in Ras family mem-
bers, activation of PI3K- and Stat3-mediated signaling path-
ways, elevated expression of transforming growth factor b1
(TGFb1), and activation of the TGFb/Smad signaling pathways
and, at later stages, Trp53 mutations (DiGiovanni, 1992; Kemp,
2005). Importantly, like human squamous cell carcinoma, this
model develops distant metastases with an epithelial mor-
phology in lymph nodes and lungs (Han et al., 2005), making it
a suitable model to study the involvement of EMT in vivo.
Furthermore, extensive studies have shown that expression of
Twist1 in primary human squamous cell carcinoma, including
esophageal cancer (Sasaki et al., 2009; Xie et al., 2009; Yuen
et al., 2007) and head and neck cancer (Ou et al., 2008; Wushou
et al., 2012), correlates with distant metastasis and poor
prognosis. In this study, we investigate the importance of the
dynamic EMT process in metastasis in vivo using the skin carci-
Induction of Twist1 Promotes Invasive Carcinoma
Previous studies have demonstrated the necessary role of
Twist1 as an inducer of EMT. To understand the contribution of
Twist1 in metastatic carcinoma, we analyzed 99 primary human
carcinomas with patient-matched lymph node metastases for
Twist1 expression. Of the 20 cases with high Twist1 expression
in the primary tumor, we found 16 cases with over 50% drop in
Twist1 levels in the lymph node metastases (Figures S1A and
S1B available online), suggesting Twist1 is activated in the
primary tumor but not distant metastases. To study how
dynamic activation of Twist1 directly impacts carcinoma
progression, we generated skin-specific Twist1 Tet-on inducible
mice by crossing transgenic mice carrying a single copy of a
TetOP-Twist1 transgene with Keratin 5 promoter-driven reverse
tetracycline-controlled transactivator mice (K5-rtTA) (Diamond
et al., 2000). Bitransgenic mice (referred to as K5-Twist1 mice)
showed specific expression of Twist1 protein in the basal
epidermal layer upon doxycycline (dox) treatment (Figure S1C).
Long-term induction of Twist1 alone in K5-Twist1 mice did not
result in visible skin abnormalities (data not shown). To generate
squamous cell carcinoma (SCC), K5-Twist1 mice and control
single transgene littermates were treated with a single dose of
7,12-dimethylbenz[a]anthracene (DMBA) followed by weekly
applications of 12-O-tetradecanoylphorbol-13-acetate (TPA)
for 20 weeks to allow skin tumor development (Abel et al.,
2009; Kemp, 2005; Sun et al., 2007) (Figure 1A). At the end of
TPA treatment, when all mice have developed multiple papil-
lomas, we randomly divided these mice into two groups. One
group of mice received doxycycline in the drinking water to allow
continuous Twist1 expression in K5-positive tumor cells, even if
tumor cells have migrated out of the skin and disseminated
throughout the body. We used this systemic Twist1 induction
group as the model for ‘‘irreversible EMT.’’ The second group
of mice received doxycycline topically on the dorsal skin area
containing papillomas to induce Twist1 only at the primary tumor
site, such that tumor cells would lose Twist1 expression once
they have disseminated from the skin. This local induction of
Twist1 was used as the model for ‘‘reversible EMT’’ (Figures
1A and S1D).
Within 7 days of doxycycline treatment through either oral or
topical routes, papillomas on the K5-Twist1 mice began to
invaginate into the skin and converted to SCCs at similar rates
in both groups (Figures 1B and 1C). By three weeks, both groups
of K5-Twist1 mice presented over 3-fold higher conversion
frequencies than their control littermates (Figure 1C). Impor-
tantly, induction of Twist1 by oral or topical doxycycline resulted
in similar conversion rates and frequencies of papillomas to
SCCs (52% for oral treatment versus 40% for topical treatment;
Figures 1C and 1D), demonstrating similar efficacy of Twist1
induction at the primary site using both doxycycline delivery
methods. Histological analysis confirmed that papillomas have
converted to poorly differentiated SCCs, with many regions pre-
senting a spindle-cell phenotype in both groups of K5-Twist1
mice, while the naturally converted SCCs in the control group
showed a well- to moderately differentiated epithelial mor-
phology (Figure 1E). In Twist1-induced SCCs, tumor cells
invaded through the underlying basement membrane, demon-
strating a role of Twist1 in matrix degradation (Figure 1F).
Together, these data indicate that Twist1 is sufficient to promote
invasive carcinoma progression in vivo.
Reversible Induction of Twist1 Promotes Carcinoma
To understand how irreversible versus reversible induction of
Twist1 impacts metastasis, we examined individual mice for
distant metastases by macroscopic and histological analysis.
Starting at 5 weeks after doxycycline induction, mice with heavy
metastasis burden were sacrificed together with mice in the
comparison groups, and all mice were terminated by 8 weeks.
Reversibility of EMT Is Essential for Metastasis
726 Cancer Cell 22, 725–736, December 11, 2012 ª2012 Elsevier Inc.
Consistent with published data, 27%–33% of control SCC-
bearing mice developed distant metastases in the lymph node
and/or the lung (Abel et al., 2009; Kemp, 2005). Strikingly, 12
out of 14 K5-Twist1 mice (86%) receiving topical doxycycline
developed distant metastases. In contrast, only three out of 13
K5-Twist1 mice (23%) receiving oral doxycycline developed
distant metastases (Figure 2A). K5-Twist1 mice receiving topical
doxycycline also developed significantly more metastatic
papillomaoral doxtopical dox
oral dox SCC topical dox SCC
SCC conversion frequency
day 0day 6 day 13
SCC conversion rate
0 1015 20
oral dox (n=5)
topical dox (n=6)
Figure 1. Induction of Twist1 Promotes Invasive Carcinoma Conversion
(A) A schematic of the DMBA/TPA skin tumor model and two doxycycline (dox) induction approaches in K5-Twist1 mice.
(B) Representative images of tumor lesions in control and doxycycline-treated K5-Twist1 mice over time. Control mice are single transgene littermates that
received oral or topical doxycycline.
(C) Graph of conversion rates from papillomas to SCCs over time for a representative cohort ± SEM at each time point.
(D) Scatter plot of SCC conversion frequency in control and doxycycline-treated K5-Twist1 mice. Each dot represents one mouse, and the bar represents the
mean of each group. *p < 0.0001 compared to control group, Student’s t test.
(E) Histologic sections of tumors stained with hematoxylin and eosin. Papillomas have well-defined cellular organization, whereas control SCCs are well- to
moderately differentiated. In contrast, doxycycline-treated tumors are disorganized and poorly differentiated. Bar = 50 mm.
(F) Frozen tumor sections were costained for tumor cells (K5, green), basement membrane (laminin 5, red), and nuclei stain (blue) to examine the breachment of
basement membrane by tumor cells. Bar = 50 mm.
See also Figure S1.
Reversibility of EMT Is Essential for Metastasis
Cancer Cell 22, 725–736, December 11, 2012 ª2012 Elsevier Inc. 727
lesions per mouse than mice receiving oral doxycycline, high-
lighting the drastic difference in metastasis incidences between
these two groups (Figure 2B). It is also important to note that this
difference is not due to nonspecific effects of doxycycline, since
control mice receiving oral or topical doxycycline presented
similar metastasis frequencies (Figures 2A and 2B).
We next examined the expression of Keratin 5 and Twist1 in
the primary tumors and metastatic nodules. We found that all
skin tumor cells express Keratin 5 both with and without Twist1
induction, suggesting that Keratin 5 can be used to specifically
mark skin tumor cells in this model. Importantly, we detected
robust nuclear Twist1 expression in the primary tumors following
both oral and topical doxycycline treatment; in contrast, all
distant metastatic lesions showed no Twist1 expression (Fig-
ure 2C). Together, these results indicate that only reversible,
but not irreversible, induction of Twist1 significantly promotes
Twist1 Regulates EMT in a Reversible Fashion during
Metastasis In Vivo
To understand whether Twist1 indeed regulates EMT in a revers-
ible manner during metastasis, we examined both primary
tumors and metastatic nodules for the expression of Twist1
and key EMT markers. In control mice, the naturally converted
SCCs showed strong expression of epithelial markers, including
E-cadherin, b-catenin, and g-catenin (Figures 3A–3C and S2A)
and no expression of mesenchymal marker vimentin in the tumor
cells (Figure 3D). Primary tumors from K5-Twist1 mice receiving
either oral or topical doxycycline presented diminished epithelial
markers and strong vimentin expression, indicating that Twist1
can effectively induce EMT in primary tumors (Figures 3, S2A,
and S2B). In contrast, all corresponding distant metastases in
the topical induction group present an epithelial morphology
with no vimentin expression and strong E-cadherin staining
(Figures 3C, 3D, and S2C). The fact that topical induction of
control SCCoral doxtopical dox
number of metastatic events
oral doxoral doxtopical dox
Figure 2. Reversible Induction of Twist1 Promotes Carcinoma Metastasis
(A) A histogram showing metastasis frequencies in control and K5-Twist1 mice group receiving oral or topical doxycycline. The fraction of mice developing
metastases in individual groups is represented above each bar. Fisher’s exact test analysis was performed to determine statistical significance.
(B) An event was defined as a tumor nodule in an individual lymph node and/or the presence of at least a single nodule in the lung tissue. Each dot represents
a single mouse. Student’s t test statistical analysis was performed to compare average events per group.
(C) Representative images of tumor sections costained for Twist1 (brown) and K5 (green). Paraffin-embedded tumor sections were stained for Twist1 using
immunohistochemistry (brown) followed by immunofluorescent staining for K5 (green) on the same section to identify tumor cells. Bar = 50 mm.
Reversibility of EMT Is Essential for Metastasis
728 Cancer Cell 22, 725–736, December 11, 2012 ª2012 Elsevier Inc.
Twist1 drastically increased metastasis incidence and that
distant metastases presented an epithelial phenotype indicates
that ‘‘reversible EMT’’ can effectively promote tumor metastasis.
To our surprise, while oral doxycycline induction of Twist1
reduced E-cadherin expression and induced EMT in primary
tumors in all 13 mice (Figure 3), the rare metastatic nodules
developed in three micealso presented anepithelial morphology
(Figure S2D). Immunostaining analyses of these metastases
showed strong E-cadherin along with weak Twist1 expression
in the tumor cells (Figure S2E), suggesting that these rare
xo d l ac i p o txo d l a r ol o r t noc
Relative E-cadherin levels
s i s a t sa t emxo d l ac i p o txod l a r ol o r t noc
control oral doxtopical dox metastasis
Figure 3. Twist1 Regulates EMT in a Reversible Fashion during Metastasis
(A) Primary and metastatic tumor samples were costained for E-cadherin (red) and K5 (green) to identify tumor cells undergoing EMT. Bar = 50 mm.
(B) The relative E-cadherin levels in K5+tumor cells were quantified in individual tumor samples from (A). Values were normalized to control samples and plotted
on a histogram ± SEM. Student’s t test statistical analysis was performed.
(C) Representative images of tumor sections from control and K5-Twist1 mice costained for Twist1 (brown) and E-cadherin (green) expression. Boxed regions
highlight areas of Twist1-positive tumor cells with disrupted or absent E-cadherin expression. Bar = 50 mm.
(D) Representative images of tumor sections from control and K5-Twist1 mice costained for Twist1 (brown) and vimentin (green) expression. Arrows indicate
Twist1-positive tumor cells that express vimentin. Bar = 25 mm.
See also Figure S2.
Reversibility of EMT Is Essential for Metastasis
Cancer Cell 22, 725–736, December 11, 2012 ª2012 Elsevier Inc. 729