VEGF-Mediated Angiogenesis Links EMT-Induced Cancer Stemness to Tumor Initiation

A key feature of progressive solid tumors is the acquisition of the potential to invade neighboring tissue and to disseminate throughout the body and form metastatic lesions at distant sites (1–3). Tumor cells achieve this by activating an epithelial–mesenchymal transition (EMT) program to undergo phenotypic changes, such as the loss of cell–cell adhesion and the gain of cell migration capabilities to evade from the primary tumor. The tissue remodeling processes occurring during EMT are shared by embryonic development, wound healing, and metastasis formation. Molecular hallmarks of EMT are the loss of cell polarity, the loss of epithelial markers, such as E-cadherin and ZO-1, the gain of the expression of mesenchymal markers, such as N-cadherin, vimentin, and fibronectin, a dramatic cytoskeletal reorganization accompanied by the change from an epithelial, differentiated morphology to a fibroblast-like, motile, and invasive cell behavior (3–5). Among many growth factors, TGF-β is one of the most potent inducers of EMT.

Interestingly, besides promoting invasiveness, TGF-β–induced EMT has been shown to induce the transition of transformed and immortalized human mammary epithelial cells into mesenchymal cancer cells with stem cell traits, thus linking EMT to tumor cell plasticity. In fact, EMT enriches for cancer stem cells (CSC) and increases tumorigenic potential of cancer cells upon transplantation into immunodeficient mice (6, 7). The ultimate hallmark of CSCs and metastatic cells is their ability to initiate tumors de novo (8). CSCs promote tumor growth through their ability to self-renew and to differentiate. However, the molecular mechanisms promoting the tumorigenicity of cancer cells undergoing an EMT have remained widely elusive. Here, we report that EMT confers increased tumorigenicity to murine breast cancer cells by the upregulated expression of VEGF-A and by increased tumor angiogenesis. Notably, VEGF-A expression is required for the increased tumor initiation capacity of breast cancer cells that have undergone EMT. However, VEGF-A by itself is not sufficient to fully support tumor formation of epithelial breast cancer cells before EMT, indicating that additional factors induced during EMT contribute to efficient tumor initiation. We propose a novel interpretation of the features of CSCs with EMT-induced, VEGF-A–mediated angiogenesis as the connecting mechanism between cancer cell stemness and tumor initiation.

For details, see Supplementary Data.

A subclone of normal murine mammary gland epithelial (NMuMG) cells (NMuMG/E9; hereafter NMuMG) expressing E-cadherin has been described earlier (26). MTflECad and MTΔECad cells have been previously described (9). Py2T cells were established from an MMTV-PyMT mammary gland tumor (10). All these cells have been derived or propagated in house for years and tested for mouse epithelial marker expression by reverse transcriptase (RT)-PCR and immunofluorescence stainings. MDA-MB-231 cells were a kind gift of N. Hynes (Friedrich Miescher Institute for Biomedical Research, Basel, Switzerland) and were tested for human breast cancer markers by RT-PCR and immunofluorescence stainings. All cells were cultured in Dulbecco's Modified Eagle Medium supplemented with glutamine, penicillin, streptomycin, and 10% fetal calf serum (Sigma-Aldrich).

Two-dimensional (2D) cultured cells or second passage mammospheres (M2) were injected as a single-cell suspension in PBS at defined cell numbers into the ninth mammary gland of 7 to 10-week-old females BALB/c Rag2^−/−;common γ receptor^−/− (RG mice; a kind gift from A. Rolink, University of Basel, Basel, Switzerland).

MMTV-Neu mice were injected intravenously with 10⁶ MTflECad and MTΔECad cells resuspended in PBS. Three weeks or 3 days after injections, respectively, mice were sacrificed and lung metastases or GFP-positive cancer cells trapped in the lung were scored.

All experimental procedures involving mice were performed according to the guidelines of the Swiss Federal Veterinary Office (SFVO) and the regulations of the Cantonal Veterinary Office of Basel Stadt (licences 1878, 1907, 1908).

Cryostat and paraffin sections were prepared as described previously (9). Seven micrometer cryostat sections of tumor samples were permeabilized with 0.1% Triton-X100 PBS, blocked with 5% goat serum or bovine serum albumin for 1 hour at room temperature, stained overnight at 4°C with primary antibodies against CD31 (1:50; 440274, BD Pharmingen), NG2 (1:100, AB5320 Millipore) followed by fluorescent secondary antibodies (Alexa Fluor; Invitrogen). Nuclei were counterstained with 4′, 6-diamidino-2-phenylindole (DAPI). To assess tissue hypoxia, mice were injected intraperitoneally with 60 mg/kg hypoxyprobe for 30 minutes before sacrifice and cryostat sections were processed according to the protocol (Hypoxyprobe-1 Kit, HPI). Immunofluorescence pictures were acquired with a Leica DMI 4000. ImageJ was used for processing and analysis of the signal intensity.

Mammospheres were fixed in 4% paraformaldehyde, permeabilized with 0.1% Triton X-100 and stained overnight with antibodies against N-cadherin (33-3900 Zymed) and E-cadherin (13-1900 Zymed). Py2T cells were incubated with 5(6)-carboxyfluorescein N-hydroxysuccinimidyl ester (CFSE; 0.1 μmol/L) before spheroid culturing and unfixed spheroids were stained for living and dead cells with Hoechst and propidium iodide. Confocal sequential images were acquired using a Leica SP5 confocal microscope and three-dimensional (3D) images were generated using IMARIS software.

The statistical significance between the tumor onset of two groups of mice was either tested by the Mann–Whitney U test or Fisher exact test in cases in which mice did not develop any tumors. For all further animal experiments, statistical significance was determined by the Mann–Whitney U test if not stated differently in the figure legend. To compare values acquired in cell culture experiments, statistical significance was assessed by the Student t test. For all calculations, GraphPad Prism software was used.

We employed defined in vitro EMT models of murine mammary epithelial cells and breast cancer cells to investigate the molecular mechanisms by which an EMT promotes the formation of CSCs and increases the tumorigenic potential of cancer cells. MTflECad cells have been established from a mammary tumor of an MMTV-Neu transgenic mouse carrying conditional (floxed) alleles of the E-cadherin gene. These cells undergo EMT upon Cre-mediated genetic ablation of the E-cadherin gene (MTΔECad; ref. 9). NMuMG cells (9) and Py2T breast cancer cells, established from a mammary gland tumor of an MMTV-PyMT transgenic mouse (10), undergo EMT upon treatment with TGF-β. When cultured on antiadhesive plates, epithelial MTflECad, NMuMG, and Py2T cells formed unstructured cell aggregates, whereas the induction of EMT in these cells induced the growth of organized hollow spheres, a hallmark of stem cells (Fig. 1A–C and Supplementary Movies S1 and S2). The mesenchymal MTΔECad hollow spheres expressed N-cadherin at cell–cell contacts, whereas the MTflECad epithelial cell aggregates employed E-cadherin for cell–cell adhesion (Fig. 1B). Culture in methylcellulose-containing media revealed significant clonal colony growth only with mesenchymal cells (Fig. 1D; ref. 11). EMT also provoked an increased sensitivity to the CSC-specific drug salinomycin (Supplementary Fig. S1; ref. 12). Consistent with the controversial debate on the use of cell surface markers for the identification of CSCs (13), the CSC surface markers CD24, CD29, and CD49f (14) failed to identify subpopulations of cells that were increased during EMT in the three murine cellular systems (Supplementary Table S1).

We next assessed the tumorigenic potential of epithelial MTflECad and mesenchymal MTΔECad cells by orthotopic transplantation into the mammary fat pad of immunodeficient BALB/c Rag2^−/−;common γ receptor^−/− (RG) mice. Notably, mesenchymal MTΔECad cells initiated tumors faster and more efficiently than epithelial MTflECad cells. Limiting dilution experiments revealed that MTΔECad cells exerted a significantly higher capacity of tumor initiation than MTflECad cells (Fig. 2A). When cells were cultured as spheroids before implantation (3D/M2), these differences became even more apparent (Fig. 2B). Upon spheroid selection, as few as 10 MTΔECad cells were able to efficiently initiate a tumor, whereas 10 MTflECad cells failed to do so. Epithelial MTflECad cells gave rise to tumors with epithelial morphology, defined borders, and central necrotic areas, whereas tumors formed by MTΔECad cells showed an invasive, fibroblast-like appearance in the absence of necrosis (Fig. 2C).

Consistent with their increased tumorigenicity, mesenchymal MTΔECad cells also formed more and larger lung metastases than epithelial MTflECad cells following orthotopic transplantation into RG mice (Fig. 3A–C). Upon tail vein injection into syngeneic MMTV-Neu transgenic mice, epithelial MTflECad cells seeded many small lung metastases, whereas mesenchymal MTΔECad cells initiated less but larger lung metastases (Fig. 3D and E). No difference was observed in the ability of the cells to home to the lungs upon injection into the tail vein (Supplementary Fig. S2A), suggesting that the difference in metastatic outgrowth was due to tumor growth parameters at the distant site.

Both primary and metastatic MTΔECad tumors exhibited significantly higher microvessel densities and reduced levels of apoptosis yet unchanged proliferation as compared with MTflECad tumors and metastases (Fig. 4A–D and Supplementary Fig. S2B and S2C). MTΔECad tumors also displayed higher levels of VEGF receptor 2 and 3 expression in their vasculature as compared with MTflECad tumors (Supplementary Fig. S3A), indicating an activation of tumor angiogenesis. Lectin perfusion and pericyte coverage experiments showed that the vessels within the tumors and metastases were all functional and that MTΔECad tumors displays more perfused vessels than the MTflECad ones (Supplementary Fig. S3B–S3D). In contrast to mesenchymal MTΔECad tumors, epithelial MTflECad tumors also exhibited large hypoxic and necrotic areas associated with high levels of apoptosis yet unchanged proliferation (Fig. 4E and Supplementary Fig. S3E and S3F). Consistent with these findings, in a transgenic mouse model of pancreatic β-cell carcinogenesis (Rip1Tag2, RT2; ref. 15) loss of E-cadherin in the tumor cells (RT2;β-ΔECad) not only correlated with increased tumor invasion and metastasis but also with intensified tumor angiogenesis (Fig. 4F and data not shown). Conversely, the genetic depletion of the transcriptional repressor of E-cadherin expression Snail-1 in Rip1Tag2 mice (RT2;β-ΔSnai1) resulted in reduced tumor invasion and tumor microvessel densities (Fig. 4F and data not shown). These results indicate a strong correlation between EMT and tumor angiogenesis. Gene expression profiling of MTflECad, Py2T, and NMuMG cells before and after EMT revealed that, in addition to the activation of genes involved in cellular differentiation, cell motility, and adhesion, EMT also induced the expression of genes involved in the regulation of angiogenesis, including the gene encoding the proangiogenic factor VEGF-A (Supplementary Fig. S4A and S4B). A transient increase of VEGF-A levels by TGF-β has been previously reported in the absence of an EMT (16), yet in our in vitro models, increased VEGF-A expression is associated with the induction of an EMT.

The increased angiogenesis observed in tumors formed by MTΔECad cells correlated with high VEGF-A protein and mRNA levels in cultured MTΔECad cells (Fig. 5A and B). Spheroid culturing caused a further increase in VEGF-A levels (Fig. 5B) Notably, cultured MTflECad cells expressed low amounts of VEGF-A compared with the MTΔECad cells. In contrast, high VEGF-A protein levels were found in large MTflECad tumors (Fig. 5C), most likely due to the tumor hypoxia and necrosis observed in these tumors (Fig. 2C). We conclude that MTΔECad cells exhibit high VEGF-A levels already when implanted into mice, while MTflECad with low levels of VEGF-A expression cells need to undergo an angiogenic switch for efficient tumor outgrowth. Supporting this notion, MTΔECad tumors smaller than 2 mm in diameter were highly vascularized, whereas small MTflECad tumors widely lacked intratumoral microvessels (Fig. 5D).

To assess the functional contribution of VEGF-A to tumor initiation, we generated MTflECad and MTΔECad cells that stably expressed shRNA against VEGF-A (shVA) or a nontargeting control shRNA (shCtr; Fig. 6A and Supplementary Fig. S5A). Upon EMT, silencing of VEGF-A significantly impaired tumor onset: 200 MTΔECad shCtr cells gave rise to tumors after 30 days, whereas no tumors were palpable at this time in mice transplanted with 200 MTΔECad shVA#1 or MTflECad shCtr cells (Fig. 6B). Depletion of VEGF-A in epithelial MTflECad shVA #1 cells only moderately delayed tumor onset as compared with MTflECad shCtr cells. Independent experiments employing additional shRNA sequences against VEGF-A confirmed the reliance of early tumor onset on VEGF-A expression (Fig. 6C and D and Supplementary Fig. S5B). Consistent with their reduced expression of VEGF-A, MTΔECad shVEGF-A tumors exhibited significantly reduced microvessel densities as compared with MTΔECad shCtr tumors; the levels of neovascularization were comparable with epithelial MTflECad tumors (Fig. 6E–G). Orthotopic implantation of 10 spheroid-cultured, VEGF-A–depleted MTΔECad cells (shVA #1, #4, #5, #8) failed to provoke the efficient tumor formation observed with 10 MTΔECad shCtr cells, further underscoring the critical requirement of VEGF-A for the increased tumor initiation of the cells after an EMT (Fig. 6H). Depletion of VEGF-A did neither significantly affect MTflECad and MTΔECad cell proliferation nor mammosphere formation (Supplementary Fig. S5C–S5E), indicating that VEGF-A did not directly affect tumor cell proliferation and survival or cancer stemness. Finally, treatment of mice transplanted with MTflECad or MTΔECad cells with the VEGF receptor inhibitor PTK787/ZK222584 (PTK) significantly impaired vascularization and growth of both tumor types (Fig. 7A and B), indicating that both tumor types depend on tumor angiogenesis. In contrast, PTK treatment had no effect on MTflECad or MTΔECad cells cultured in 2D or as mammospheres (Supplementary Fig. S5F and S5G).

Conversely, the forced expression of VEGF-A in MTflECad cells significantly increased tumor microvessel density and reduced tumor necrosis but only moderately accelerated tumor onset (Supplementary Fig. S6A–S6D). The increased vasculature was not associated to diminished hypoxia and only an intermediate increase in the number of perfused vessels could be observed (Supplementary Fig. S6E–S6H). In these tumors, apoptosis was significantly decreased by VEGF-A expression, whereas tumor cell proliferation was unaffected (Supplementary Fig. S6I and S6J). The modulation of VEGF levels had no impact on the capacity or on the quality of spheroid formation (Supplementary Fig. S5D, S5E, and S5G), indicating that VEGF-A upregulation plays a role only in the context of an EMT-induced mechanism. These results indicate that, while critical for tumor angiogenesis, VEGF-A by itself is not sufficient to efficiently promote tumor initiation. Besides VEGF-A, the other family members VEGF-B, C, and D were also found upregulated in their expression during EMT and in spheroid culture of MTflECad and MTΔECad cells and tumors, yet to varying degrees and without affecting tumor lymphangiogenesis (Supplementary Fig. S7A–S7D). Moreover, shRNA-mediated ablation of VEGF-C expression failed to affect MTΔECad tumor formation (Supplementary Fig. S7E and S7F). Hence, other factors were likely necessary to support VEGF-A in the tumorigenic abilities of mesenchymal MTΔECad cells. High levels of matrix metalloproteinases (MMP), inflammatory cytokines, and activated endothelial cell markers were found in lysates of MTΔECad tumors as compared with MTflECad tumors (Supplementary Table S2), suggesting that additional proangiogenic signaling pathways contributed to the efficient tumor initiation of MTΔECad cells. Consistent with this notion, conditioned medium of MTΔECad cells supported the growth of human umbilical vein endothelial cells (HUVEC), while MTflECad-conditioned media did not (Supplementary Fig. S8). Accordingly, forced expression of VEGF-A in MTflECad cells increased, whereas VEGF-A knockdown in MTΔECad cells decreased conditioned medium-induced HUVEC proliferation.

To investigate the possibility that VEGF-A upregulation could have implications on the CSC's intrinsic properties, we analyzed the VEGFR1 and Nrps levels in MTflECad and MTΔECad in 2D or spheroids culturing, as well the effect of downregulation of Nrp1 and Nrp2 on cell proliferation and cell cycle. In contrast to what has been previously shown in other models (17), we did not observe any effect of VEGFR and Nrp expression on cell proliferation or mammosphere formation in our cellular models (data not shown).

To determine VEGF-A expression by tumorigenic CSCs of human breast tumors, we isolated bona fide CSCs from patients' primary breast tumors that have been previously serially transplanted in immunocompromised mice (18) as well as from the human breast cancer cell line MDA-MB-231. Single-cell suspensions were stained with antibodies against CD44 and CD24. CD44⁺/CD24^−/low putative CSC and CD44⁺/CD24⁺ non-CSCs (19) were isolated by flow cytometry, and the expression of VEGF-A mRNA was determined by qRT-PCR. Indeed, the putative CSC population exhibited significantly higher levels of VEGF-A mRNA expression than their non-CSC counterparts (Fig. 7C and D), indicating that also in human breast cancer cells, increased cancer cell stemness correlated with high VEGF-A expression.

In this report, we have utilized standardized assays for CSCs to delineate the mechanisms underlying EMT-induced cancer stemness. While spheroid growth, tumor initiation, and metastatic spread revealed a correlation between EMT and hallmarks of CSCs (shown here), the analysis of cell surface markers (Supplementary Table S1), aldehyde dehydrogenase-positive populations, label-retaining cells, and general drug resistance (data not shown) did not correlate with EMT. While the identification of appropriate stem cell markers for the isolation of CSCs remains challenging (13), growing evidence links tumor-initiating cells with proangiogenic signals (17, 20, 21). Notably, we have identified VEGF-A–mediated angiogenesis as one critical determinant of the increased tumorigenicity of cells undergoing EMT: (i) VEGF-A and several other angiogenic factors and cytokines are upregulated in murine breast cancer cells during EMT, (ii) CSCs of human primary breast cancers or from a human breast cancer cell line also display increased levels of VEGF-A, (iii) VEGF-A is required for the increased tumorigenicity of cells undergoing EMT, yet (iv) VEGF by itself is not sufficient to promote tumorigenicity of epithelial cancer cells, indicating that additional angiogenic factors regulated in their expression during EMT are critical for an effective tumor initiation. Our results also show that VEGF-A is required for EMT-induced angiogenesis in the early events of tumor initiation, in line with similar results in skin and brain (17) and with an established central role for VEGF-A in tumor progression (22–24).

As few as 10 mesenchymal cells efficiently initiated tumors upon orthotopic transplantation, while their epithelial counterparts were much less efficient. The calculated CSC frequency of MTΔECad [1:14.3 and 1:23.3 with cells cultured as mammospheres (3D/M2) and on plastic (2D), respectively] and of MTflECad cells (1: 398 and 1:73.7 with 3D/M2 and 2D cells, respectively could not be statistically verified, since extreme limiting dilution analysis (25) rejected a single hit model for MTΔECad cells, thus arguing in favor of a multi-hit event (LR test P = 0.073 for 3D/M2 and P = 0.0065 for 2D in MTΔECad cells, and P = 0.178 for 3D/M2 and P = 0.193 for 2D in MTflECad). These results are consistent with the observation that VEGF-A by itself is not sufficient to increase the tumorigenicity of epithelial cancer cells and that other factors are required as well. Together, these findings propose a novel interpretation of cancer stemness by functionally linking EMT with VEGF-A–mediated tumor angiogenesis and the capacity to initiate tumors de novo. These results provide important insights into the mechanisms underlying the increased tumorigenicity of CSCs and cells undergoing EMT and open avenues for the design of therapeutic interventions. Our work also challenges the hypothesis that the ability of CSCs to de novo initiate tumors is exclusively due to features specifically attributed to CSCs, such as asymmetric cell division and self-renewal. Rather, a high angiogenic potential, for example, induced by an EMT and the upregulated expression of VEGF-A, contributes to the tumorigenic phenotype of CSCs. Our data indicate that high angiogenic capabilities are an intrinsic feature of CSCs and metastatic cells. The data also raise the caveat that, while tumorigenicity is a hallmark of metastatic cells and CSCs (tumor-initiating cells), tumorigenicity assays may not be the ultimate criteria to definitely identify CSCs.

No potential conflicts of interest were disclosed.

Conception and design: A. Fantozzi, D.C. Gruber, G. Christofori

Development of methodology: A. Fantozzi, D.C. Gruber, L. Pisarsky, C. Heck, M. Yilmaz, U. Hopfer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Fantozzi, D.C. Gruber, L. Pisarsky, C. Heck, A. Kunita, K. Cornille, U. Hopfer, M. Bentires-Alj, G. Christofori

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Fantozzi, D.C. Gruber, L. Pisarsky, C. Heck, A. Kunita, N. Meyer-Schaller, U. Hopfer, G. Christofori

Writing, review, and/or revision of the manuscript: A. Fantozzi, D.C. Gruber, C. Heck, M. Yilmaz, N. Meyer-Schaller, U. Hopfer, M. Bentires-Alj, G. Christofori

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Fantozzi, D.C. Gruber, A. Kunita, G. Christofori

Study supervision: A. Fantozzi, D.C. Gruber, G. Christofori

The authors thank M.J. Wheelock (University of Nebraska,), A. Banfi (University Hospital Basel, Basel, Switzerland), A. Rolink (University of Basel), J. Jonkers (NKI, Amsterdam, the Netherlands), T. Gridley (The Jackson Labs), and A. Welm (University of Utah) for providing cell lines, mice, primary breast tumors and reagents. They also thank T. Barthlott, C. Berkemeier, and C. Mayer for flow cytometry cell sorting and H. Brinkhaus (FMI, Basel, Switzerland), H. Antoniadis, P. Schmidt, I. Galm, U. Schmieder, and R. Jost for technical support.

This work was supported by EU-FP7 TuMIC HEALTH-F2-2008-201662 (G. Christofori), the SystemsX.ch RTD project Cellplasticity (G. Christofori), the Novartis Research Foundation (M. Bentires-Alj), and a Marie-Heim Vögtli grant by the Swiss National Science Foundation (N. Meyer-Schaller).

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.