FOXC2 Expression Links Epithelial–Mesenchymal Transition and Stem Cell Properties in Breast Cancer

Despite the initial effectiveness of conventional therapies, recurrence and the emergence of metastases are major causes of therapeutic failure in cancer patients. These therapies are believed to target the differentiated and proliferative cells that comprise the bulk of the tumor. The relatively high relapse rate of patients with aggressive forms of breast cancer, including the recently identified triple-negative claudin-low/basal B subtype (for brevity referred to as claudin-low throughout), is attributed to a small population of cancer stem cells (CSC) residing within the tumor. In addition to resistance to standard therapies, CSCs are reported to have inherently greater tumor-initiating potential, which is implicated in tumor relapse (1), driving primary tumor growth as well as the seeding and establishment of metastases (2–7). Therefore, therapeutic approaches that specifically target the CSC population, particularly when used in combination with conventional therapies, may provide a therapeutic strategy to substantially improve breast cancer patient outcome. The epithelial-to-mesenchymal transition (EMT) program has recently been linked to the generation of breast CSC-like cells (8) and has well-documented roles in promoting an invasive and metastatic phenotype (9, 10). As such, the development of targeted therapeutics to inhibit the EMT program may provide significant clinical benefits in treating aggressive breast cancers for which current therapies are inadequate.

EMT was initially characterized as an important program during normal embryonic development (11, 12), however, more recent reports suggest that carcinoma cells are capable of reactivating the EMT program during tumor progression (9). Similar to cells that undergo EMT during normal development, carcinoma cells that undergo EMT lose cell–cell contacts, undergo major changes in their cytoskeleton, and acquire a mesenchymal-like morphology endowing them with increased invasive and migratory abilities (11–13). Several signaling molecules present in the tumor microenvironment are capable of initiating EMT and metastasis (14, 15) and importantly, many of these same factors have also been found to be aberrantly activated and/or overexpressed in human cancers (16–20). Because cells undergoing EMT are shown to possess stem cell properties (8), and several recent studies independently showed that CSCs as well as cells that have undergone EMT are relatively resistant to conventional chemotherapies and radiotherapies (1, 21–26), suggests the EMT program may provide a novel therapeutic window for inhibiting CSCs. However, because of the plethora of factors capable of inducing EMT and the hierarchy of the EMT programs, we sought to identify a central functional mediator of EMT independent of the initiating signal that may provide a novel target for anti-EMT-based therapies. Here we report that FOXC2 serves as such a mediator. We show that it is induced by multiple factors and that the expression of PDGFR-β is dependent on the expression of FOXC2, and thus serves as a potential therapeutic target for cells that have undergone EMT.

Immortalized human mammary epithelial (HMLE) cells and V12H-Ras transformed derivatives (HMLER), including cells expressing empty vector (pWZL), Snail, Twist, Goosecoid, or an activated form of TGF-β1 were maintained as previously described (8). Established human breast cancer cell lines were cultured in cell-specific medium as outlined in the Supplementary Materials and Methods. Antibodies used included: anti-β-actin (Abcam), FOXC2 (Dr. Naoyuki Miura), E-cadherin (BD Biosciences), Fibronectin (BD Biosciences), N-cadherin (BD Biosciences), Vimentin (NeoMarkers), and β-catenin (BD Biosciences).

The production of lentiviruses and amphotrophic retroviruses, and the transduction of target cells was followed as previously described (27). See Supplementary Materials and Methods for brief description.

Mammosphere assays were conducted as previously described (8). To test effects of sunitinib on mammosphere formation, 1,000 cells were plated in 96-well ultra-low attachment plates in 100 μL mammosphere media. At 24 and 96 hours, media was replaced with media containing 10 μmol/L sunitinib or vehicle control (dimethyl sulfoxide [DMSO]). Spheres were quantified after 7 days.

The three-dimensional (3D) laminin-rich extracellular matrix (lrECM) on-top cultures were adapted from the procedures previously described (28). See Supplementary Materials and Methods for brief description.

Nonobese diabetic/severe combined immunodeficient (NOD/SCID) mice were purchased from Jackson Laboratory. All mouse procedures were approved by the Animal Care and Use Committees of M.D. Anderson Cancer Center and conducted in accordance with Institutional policies. For xenograft tumor initiation studies, the indicated number of cells were suspended in 50 μL of Matrigel diluted 1:1 with Dulbecco's Modified Eagle's Medium and injected into the inguinal mammary gland of NOD/SCID mice. Tumor incidence was monitored for 12 weeks following orthotopic injection. To assess the spontaneous metastatic potential of cells, 2 × 10⁶ HMLER–Vector and HMLER–FOXC2 cells labeled with firefly luciferase were injected into the inguinal mammary gland of NOD/SCID mice. Mice were assessed weekly for metastasis via the intraperitoneal injection of D-Luciferin (Caliper LifeSciences) at 150 mg/kg in PBS and in vivo bioluminescence was assessed using the IVIS imaging system 200 series (Xenogen Corporation). Once mammary gland tumors reached 1.5 cm in diameter, mice were euthanized and organs were harvested for confirmation of metastatic tumor burden via bioluminescence.

To ascertain the effects of sunitinib on HMLER–FOXC2 tumor growth and animal survival, 1 × 10⁶ firefly luciferase-labeled HMLER–FOXC2 cells were injected into the mammary fat-pad of NOD/SCID mice as described above. After 24 hours, the mice were treated 5 days a week with 40 mg/kg sunitinib or vehicle (n = 9) by oral gavage. Animal were sacrificed once the tumors reached 1.5 cm³ or 30 days following injection and the dissected organs were immediately analyzed for luciferase activity. Details of additional procedures can be found in Supplementary Materials and Methods.

In our previous study, enhanced expression of FOXC2 was observed following the induction of EMT by several factors in experimentally immortalized HMLE cells (17) strongly suggesting that FOXC2 may be a critical determinant of multiple EMT programs. To assess the functional role of FOXC2 during EMT, we used shRNA-mediated suppression of FOXC2 in HMLE cells that underwent EMT via ectopic expression of Snail, Twist, or TGF-β1. The suppression of FOXC2 had no significant effect on cell growth, but substantially altered the in vitro morphology of all cell lines, including increased clustering of cells into epithelial-like islands with prominent cell–cell contacts and reduced fibroblastic morphology (Fig. 1A). FOXC2 attenuation also led to reduced expression of mesenchymal markers vimentin, fibronectin, and N-cadherin across all cell types tested as well as re-expression (both mRNA and protein levels) of the epithelial marker E-cadherin in HMLE–Snail and HMLE–TGF-β1 cells (Fig. 1B and C). To examine whether FOXC2 could function in parallel to other EMT regulators we examined the expression of Snail, Twist, and Slug in cells ectopically expressing FOXC2 and found only moderate upregulation of Twist but not Snail or Slug expression (Supplementary Fig. S1), suggesting FOXC2 may not regulate expression of these factors.

The passage of cells through an EMT and the acquisition of mesenchymal properties is associated with increased migratory and invasive properties. Suppression of FOXC2 was observed to significantly inhibit the invasion of HMLE–Snail cells through matrigel using Transwell migration assays in response to the soluble chemotactic ligands basic fibroblast growth factor (bFGF) and platelet-derived growth factor-BB (PDGF-BB; Fig. 1D). Mammary cells possessing epithelial traits are known to form organized multicellular acini structures with an intact laminin positive basement membrane in 3D lrECM cultures (29, 30). However, following the induction of EMT, cells became disorganized and gained invasive properties characterized by disrupted laminin staining (31, 32). Using 3D lrECM assays, HMLE–Snail control-shRNA cells grew as highly invasive stellate structures with disrupted basement membrane, displayed by disorganized laminin V staining (Fig. 1E). In contrast, the HMLE–Snail FOXC2-shRNA cells formed noninvasive, multicellular structures with an intact basement membrane as depicted by a continuous laminin V layer (Fig. 1E). Taken together, these results suggest that FOXC2 is necessary for the maintenance of the mesenchymal phenotype of mammary epithelial cells following passage through EMT.

We previously reported that HMLE–Snail, HMLE–Twist, and HMLE–TGF-β1 cells acquired properties similar to breast CSCs, including the CD44^high/CD24^low cell surface markers and an increased ability to form mammospheres (8). Here, we found that the attenuation of FOXC2 expression by shRNA in these cells reduced the number of cells with the CD44^high/CD24^low phenotype compared with the control-shRNA expressing cells (Fig. 2A). In the same conditions, there was a marked decrease in the mammosphere-forming ability (Fig. 2B).

Next, the acquisition of stem cell properties and passage through an EMT has been reported to increase the resistance to chemotherapeutic agents (22, 23). In accordance with this, we found that suppression of FOXC2 expression in cells that have undergone EMT (HMLE–Snail, HMLE–Twist, and HMLE–TGF-β1) sensitized them to paclitaxel (Fig. 2C). Together, these results indicate FOXC2 expression is necessary for EMT-derived stem cell-like properties in mammary epithelial cells.

Previously it has been shown that fluorescence-activated cell sorting (FACS) isolation of breast tumor cells with the CD44^high/CD24^low cell surface phenotype (3) as well as isolation of tumor cells from mammospheres (33) can enrich for populations of tumor-initiating CSCs. Using this rationale, we FACS isolated CD44^high/CD24^low and CD44^low/CD24^high cellular fractions from HMLER and SUM159 breast cancer cell lines and observed increased FOXC2 protein expression in the CD44^high/CD24^low stem cell fraction relative to the CD44^low/CD24^high fraction (Fig. 3A). Increased FOXC2 protein expression was also observed in primary mammospheres isolated from HMLER, SUM159, HCC38, and SUM149 cells as compared with the same cells grown in monolayer cultures (Fig. 3B).

Previously, we have reported that ectopic expression of FOXC2 induced partial EMT in Madin–Darby canine kidney epithelial cells and was sufficient to promote the metastasis of EpRas murine mammary carcinoma cells (17). However, whether FOXC2 expression alone can induce EMT and generate CSC-like properties remains unknown. We report here for the first time that ectopic expression of FOXC2 in HMLER cells is sufficient to induce a robust EMT and subsequent CSC-like properties, both in vitro and in vivo. This was displayed as a characteristic switch of morphology to a fibroblastic appearance with a reduction in epithelial-specific E-cadherin, in conjunction with increased expression of mesenchymal markers vimentin, fibronectin, and N-cadherin at the mRNA and protein levels (Supplementary Fig. 2A–2C). Furthermore, HMLER–FOXC2 cells had increased CSC-like properties, including a switch to the CD44^high/CD24^low cell surface phenotype (Fig. 3C), enhanced mammosphere-forming efficiency (Fig. 3D), and increased resistance to paclitaxel (Fig. 3E). Notably, overexpression of FOXC2 led to increased tumor formation of HMLER cells in the mammary fat-pad (Fig. 3F). In fact, as few as 1 × 10³ HMLER–FOXC2 cells could robustly initiate tumors (7/8 sites), whereas at least 2 × 10⁶ HMLER–Vector control cells were required to get a 54% tumor take rate (7/13 sites) within the same 12 week time frame (Fig. 3F).

To assess the effect of FOXC2 expression on the orthotopic growth and spontaneous metastasis of HMLER cells, we injected 2 × 10⁶ luciferase-labeled HMLER–FOXC2 or HMLER–Vector cells into the mammary fat-pad of NOD/SCID mice and analyzed for metastases using bioluminescent imaging once per week until the tumors either reached a volume of 1.5 cm³ (28 days for HMLER–FOXC2) or until the end of the experiment after 12 weeks (HMLER–Vector). FOXC2 expression promoted aggressive growth (Fig. 3G) and metastasis of HMLER cells to the lungs, liver, hind leg bone, and most strikingly, to the brain (Fig. 3H) as soon as 28 days postinjection. In contrast, HMLER–Vector cells did not metastasize to any of the organs examined (Fig. 3H) even at the end of the experiment (data not shown). Collectively, these findings suggest that FOXC2 expression is sufficient to promote EMT and CSC properties, including chemoresistance, tumor initiation, and metastatic competence in transformed human mammary epithelial cells.

We next investigated whether FOXC2 activity could be observed in metastatic tumors in vivo. As a measure of FOXC2 activity, we generated, using microarray data, a FOXC2 gene expression signature (GES) by comparing the gene expression profile of HMLER cells overexpressing FOXC2 to vector-transduced counterparts. Applying this GES to data from 2 human xenograft models, MDA-MB-231 and CN34, we observed that the FOXC2 GES was significantly higher in metastases compared with the primary tumors in both data sets (Fig. 4A and B). Collectively, these data indicate that the FOXC2-associated signature is enriched in metastases relative to primary tumors.

We previously reported that elevated FOXC2 expression correlated with basal-like breast cancers (17). However, using gene expression profiling, an aggressive claudin-low group has been found within this subtype (34, 35). We found the FOXC2 GES to be significantly enriched in claudin-low human breast tumors (Fig. 4C), as well as in claudin-low cell lines (Fig. 4D) compared with other subtypes. To verify this finding, we conducted a Western blot analysis and found that all claudin-low cell lines analyzed (6/6) expressed FOXC2 at varying degrees, although none of the other cell lines expressed significant levels of FOXC2 (Fig. 4E).

As FOXC2 expression and activity was found to be associated with claudin-low breast tumors, we tested whether FOXC2 expression was required for the mesenchymal and invasive properties of SUM159, MDA-MB231, and HMLER–Snail cells that have a claudin-low gene expression phenotype (35). The suppression of FOXC2 resulted in a less fibroblastic morphology with increased epithelial-like cell clustering (Fig. 5A), decreased expression of mesenchymal markers fibronectin and N-cadherin (Fig. 5B) and re-expression of E-cadherin in HMLER–Snail cells (Fig. 5B). Furthermore, FOXC2 suppression significantly reduced Transwell migration of both SUM159 and HMLER–Snail cells (Fig. 5C and D; P < 0.05). In a similar fashion, SUM159 Control-shRNA cells grew in 3D lrECM cultures with a characteristic stellate morphology (36) and extended multicellular protrusions invading into the surrounding Matrigel (Fig. 5E). The suppression of FOXC2 resulted in a dramatic reduction in the invasive morphology of SUM159 cells (SUM159 FOXC2-shRNA) compared with control cells (Fig. 5E).

We next assessed the effect of FOXC2 suppression on stem cell properties on the same 3 cell lines. Suppression of FOXC2 expression substantially reduced the percentage of cells displaying the CD44^high/CD24^low CSC-like cell surface profile (Fig. 5F) as well as significantly abrogating the mammosphere-forming ability (Fig. 5G) as compared with the Control-shRNA expressing cells. Given that suppression of FOXC2 expression diminished stem cell properties in vitro, we next examined if FOXC2 knockdown affected the tumor-initiating potential of SUM159 as well as HMLER–Snail cells using limiting dilution tumor-initiation assays. FOXC2-shRNA or Control-shRNA expressing cells were introduced into the mammary fat-pad of NOD/SCID mice, and we found that the suppression of FOXC2 expression decreased tumor initiation frequency relative to control cells of both SUM159 and HMLER–Snail xenografts (Fig. 5H). In summary, these findings suggest FOXC2 may be an important functional mediator of the mesenchymal and CSC properties of claudin-low breast cancer cells.

We recently reported that cells induced to undergo EMT by multiple factors upregulated the expression of PDGFR-β (CD140b) (37) similar to FOXC2. Thus, we hypothesized that PDGFR-β might provide a druggable downstream target in FOXC2 expressing cells. We first confirmed that PDGFR-β protein expression was upregulated in a panel of EMT-derived cells (Figure 6A(i) and A(ii)). As observed for FOXC2 (Fig. 3A and B), the expression of PDGFR-β was elevated in the stem cell-enriched CD44^high/CD24^low subpopulation of HMLER cells as well as in SUM159 cells cultured as mammospheres, compared with controls (Figure 6A(iii) and A(iv)). Furthermore, the expression pattern of PDGFR-β and FOXC2 correlated strongly across a panel of established cell lines with increased expression observed in claudin-low cell lines SUM159 and Hs578T (Fig. 6B). Reflecting the increased levels of PDGFR-β in EMT-derived and claudin-low cell lines, the addition of the ligand PDGF-BB significantly elevated Transwell cell migration in all cells tested (Fig. 6C) and led to the further enhancement of the invasive phenotype of HMLER–FOXC2 cells in 3D lrECM cultures (Fig. 6D).

We next investigated if PDGFR-β expression is dependent on FOXC2 expression. Across a panel of cells, the expression of PDGFR-β was found to be substantially decreased upon suppression of FOXC2 (Fig. 6E and F; Supplementary Fig. S3). Consequently, the suppression of FOXC2 in HMLE–Snail, HMLE–TGF-β1, HMLER–Snail, and SUM159 cells compromised the ability of these cells to migrate toward PDGF-BB (Fig. 6G). To examine the possibility of FOXC2 directly regulating PDGFR-β transcription, we conducted a chromatin immunoprecipation (ChIP) assay using HMLER–FOXC2 cells. FOXC2 preferentially bound to 2 regions at 2.7 kb (–2.7 kb) and 1.3 kb (–1.3 kb) upstream of the PDGFR-β transcription start site (Fig. 6H), thus demonstrating that FOXC2 may be a direct transcriptional regulator of PDGFR-β expression.

As we found that expression of PDGFR-β is regulated by FOXC2, we tested whether sunitinib, a small molecule inhibitor of PDGFR-β, could suppress the stem-like and metastatic properties of FOXC2-expressing cells. Sunitinib treatment was found to specifically inhibit the cell growth of HMLER–FOXC2 but not HMLER–Vector control cells in monolayer cultures (Fig. 7A). The treatment of HMLER–FOXC2 and HMLER–Snail cells with sunitinib was found to significantly decrease mammosphere formation by >8-fold and >20-fold, respectively (Fig. 7B), as compared with vehicle (DMSO)-treated cells. Furthermore, cells expressing endogenous FOXC2 also display increased sensitivity to sunitinib as evidenced by both MTS assay and reduction in mammosphere formation (Supplementary Fig. S4A–S4C).

To investigate whether sunitinib could inhibit FOXC2-expressing tumors in vivo, we orally administered sunitinib to mice following orthotopic injection of luciferase-labeled HMLER–FOXC2 cells into the mammary fat-pad. In accordance with our in vitro observations, sunitinib treatment reduced primary tumor growth (Fig. 7C) and extended event-free survival of mice carrying FOXC2 tumors compared with vehicle-treated control mice (Fig. 7D). We also analyzed the lungs (Fig. 7E) and brain (Fig. 7F) of sunitinib-treated mice for the presence of HMLER–FOXC2 metastases and found that the sunitinib-treated group had significantly reduced metastatic burden compared with the vehicle-treated mice as evidenced by significantly lower photon counts in these organs following dissection (Fig. 7E and F). Taken together, these results indicate FOXC2-expressing tumor cells are sensitive to PDGFR-targeted therapies and suggest sunitinib may be an effective means of targeting FOXC2-expressing cell populations, which may include EMT-derived CSC-like and metastatic phenotypes.

Many recent studies support the emerging dogma that CSCs are responsible for chemotherapy resistance, tumor relapse, and metastatic competence (reviewed in ref. 38). We and others have shown that CSCs and cells that have undergone EMT share many functional and molecular traits, with the corollary that engagement of the EMT program within a tumor may lead to the de novo generation and/or expansion of CSCs (8, 39). Moreover, it suggests that perturbing or targeting EMT signaling pathways may provide an effective therapeutic strategy to deplete the EMT/CSC populations within a tumor. Although this seems like a rational approach, the sheer number and diversity of EMT-inducing stimuli that elicit EMT in different tumor contexts will likely hinder the development of universally applicable therapeutics. On the basis of the increased expression of FOXC2 following experimental induction of EMT by numerous EMT-inducing factors as well as in stem cell-enriched fractions (CD44^high/CD24^low population and mammospheres), we hypothesized that FOXC2 lies at the crossroads of EMT and stem cell properties. Indeed, we found that FOXC2 expression is critical for stem cell properties, including resistance to chemotherapeutics and tumor initiation, using multiple EMT models and claudin-low breast cancer cell lines. These data clearly show that targeting FOXC2 and the associated pathways may provide an additional strategy for diminishing the CSC pool or at least those that arose via EMT.

It was surprising to see re-expression of E-cadherin in the FOXC2-depleted HMLE–Snail or TGF-β1 cells but not in Twist cells even though both Snail and Twist are continuously expressed from a retroviral expression vector and known to function similarly. Although these data does not explain the failure of Snail to suppress E-cadherin expression in the absence of FOXC2, we speculate that this could be mostly due to differences in epigenetic alterations.

Claudin-low tumors account for between 25% and 39% of triple-negative breast cancers (ER⁻/PR⁻/HER2⁻), have been shown to resemble most closely with mammary epithelial stem cells; and are also enriched for markers of EMT and CSCs (35, 40). The enrichment of FOXC2 expression and its associated GES in claudin-low tumors (1, 21, 34, 35) provides the first evidence that FOXC2 transcriptional activity may play an important functional role for this molecular subtype.

We speculate that targeting the FOXC2 pathway may be an effective therapeutic strategy for tumors with enriched EMT/CSC properties. Future studies to assess the protein expression of FOXC2 pathway members in clinical specimens will be critical. As transcription factors can be difficult to directly inhibit therapeutically, a potential target for FOXC2-expressing tumors is PDGFR-β, which has U.S. Food and Drug Administration-approved small molecule inhibitors, such as sunitinib. However, as sunitinib is a multitargeted tyrosine kinase inhibitor, future studies using more specific pharmacologic inhibitors of PDGFR-β or RNAi approaches will be required to determine if PDGFR-β is a key functional mediator of FOXC2 in EMT-derived cells and claudin-low tumors. Nevertheless, our results suggest that sunitinib and other PDGFR inhibitors may be effective in patients with claudin-low or therapy resistant tumors displaying elevated FOXC2 expression.

S.A. Mani, S.J. Werden, B.G. Hollier, K.W. Evans, A.A. Tinnirello, and T.R. Sarkar are inventors of a patent application in part based on findings described in this manuscript. No potential conflicts of interest were disclosed by the other authors.

Conception and design: B.G. Hollier, A.A. Tinnirello, S.J. Werden, N. Sphyris, V.L. Battula, R. Guerra, S.A. Mani

Development of methodology: B.G. Hollier, A.A. Tinnirello, S.J. Werden, V.L. Battula, N. Miura, S.A. Mani

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B.G. Hollier, A.A. Tinnirello, K.W. Evans, J.H. Taube, T.R. Sarkar, M. Shariati, S.A. Mani

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.G. Hollier, A.A. Tinnirello, S.J. Werden, K.W. Evans, S.V. Kumar, J.I. Herschkowitz, R. Guerra, J.T. Chang, J.M. Rosen, S.A. Mani

Writing, review, and/or revision of the manuscript: B.G. Hollier, S.J. Werden, K.W. Evans, N. Sphyris, R. Guerra, J.T. Chang, J.M. Rosen, S.A. Mani

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B.G. Hollier, K.W. Evans, S.V. Kumar, N. Miura, S.A. Mani

Study supervision: B.G. Hollier, N. Sphyris, S.A. Mani

The authors acknowledge Michael Lewis, Thomas Westbrook, Jenny Chang, and Charles Perou for insightful discussions and valuable advice.

This work was supported by NIH/NCI CA155243-01 (S.A. Mani) and Susan G. Komen Postdoctoral Fellowship (B.G. Hollier/S.J. Werden). J.T. Chang is supported by NIH R00LM009837 and grant R1006 from the Cancer Prevention & Research Institute of Texas. Flow cytometry, confocal microscopy, and animal imaging were in part funded by the Cancer Center Support Grant from the National Cancer Institute (CA16672).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.