CYR61 and αVβ5 Integrin Cooperate to Promote Invasion and Metastasis of Tumors Growing in Preirradiated Stroma

Radiotherapy has proven therapeutic efficacy in human cancer management. Around 50% to 60% of cancer patients are treated by radiotherapy in the course of their disease (1). The antitumor effect of radiotherapy involves tumor cell DNA damage leading to p53-mediated apoptosis, mitotic cell death, and senescence-like irreversible growth arrest (2). In addition, recent findings indicate that radiotherapy rapidly and persistently alters the tissue microenvironment by modulating the cell phenotype, tissue metabolism, and the intercellular cross-talk (3). Increasing evidence indicates that these stromal changes may contribute to the antitumor effects of radiotherapy. However, clinical and experimental observations also indicate that irradiated stroma may have potential deleterious effects on tumor behavior (3). Experimentally, tumors growing within a previously irradiated bed tend to be more invasive and to form more metastases, an effect known as the tumor bed effect (4, 5). Clinically, adjuvant radiotherapy significantly improves local tumor control in patients undergoing conservative surgery (6). However, tumor recurrences within a preirradiated field are associated with higher risk of metastasis and poor prognosis compared with recurrences outside the irradiated area (7–10). The clinical management of this condition is a challenge, and the underlying cellular and molecular mechanisms are still matter of investigation (11).

Although a limiting factor for tumor growth, intratumoral hypoxia has been associated with local tumor invasion, metastasis formation, and shorter disease-free survival in a number of human tumors, including head and neck cancer (12), cervical cancer (13, 14) and soft tissue sarcoma (15). Experimental tumors with a high hypoxic fraction do more frequently metastasize compared with tumors with low hypoxic fractions (16, 17). Experimental tumors growing in preirradiated tissues and human tumors locally recurring after radiotherapy have an increased fraction of hypoxic cells (18–20). Rofstad and colleagues (21) have recently shown that enhanced invasion and metastasis observed in tumors growing within a preirradiated bed involves hypoxia-mediated up-regulation of metastasis-promoting gene products, in particular, the receptor for the urokinase-type plasminogen activator receptor. These results suggest that primary tumors recurring after radiotherapy show aggressive behavior because of increased environmental hypoxia leading to transcriptional activation of metastasis-promoting genes. The transcription factor hypoxia inducible factor (HIF)-1 is likely to play a key role in this process. HIF-1 mediates adaptive responses to hypoxia and activates the transcription of many genes encoding for proteins regulating angiogenesis, glucose metabolism, intracellular pH, cell proliferation/survival, tumor invasion, and metastasis (22, 23). Therefore, activation of a metastatic program by HIF-1 seems as an attractive mechanism by which tumor cells growing into a preirradiated hypoxic microenvironment invade and metastasize. However, hypoxia-induced up-regulation of prometastatic molecules alone may not entirely explain the sustained aggressive behavior of these tumors because once hypoxic tumor cells reach normoxic conditions, they may revert back to their initial phenotype. We therefore hypothesized that, in addition to hypoxia-mediated up-regulation of prometastatic genes, hypoxia forming within tumors growing in preirradiated tissue could generate a selective pressure favoring the outgrowth of highly metastatic tumor cell variants retaining an aggressive phenotype upon escape from hypoxia.

Here, we provide experimental evidence supporting this hypothesis by demonstrating that SCCVII and HCT116 tumor cells derived from tumors growing in a preirradiated bed, or selected in vitro through repeated cycles of sever hypoxia, retain an invasive and metastatic phenotype even when returned to normoxia. We identified the matricellular protein CYR61 and integrin αVβ5 as proteins that cooperate to mediate these effects. Furthermore, we found αV integrin inhibitors to suppress lung metastasis formation of tumors growing within a preirradiated field. These observations extend our knowledge on the mechanisms involved in the tumor bed effect and open novel therapeutic perspective for the management of local tumor recurrences after radiotherapy.

Antibodies, chemicals, proteins, and peptides. Bovine serum albumin (BSA), paraformaldehyde (PFA), fibronectin, vitronectin, and 4′,6-diamidino-2-phenylindole were purchased from Sigma Chemie. Antibodies were purchased as follows: monoclonal antibody (mAb) P1F6 (anti-αVβ5) from Becton Dickinson; mAb LM609 (anti-αVβ3) from Chemicon International; Lia 1/2 (anti-β1) from Beckman Coulter; and rabbit anti-CYR61 from Abcam, goat anti-CYR61 (sc-13100; Santa Cruz Biotechnology), and mAb 356108 (anti-human CYR61) from R&D Systems, biotinylated and purified rat anti-CD31 mAb from PharMingen, and anti–HIF-1α mAb R&D Systems. Antibody 17E6 was produced in house (Merck Farma). Antibodies to mouse αV (RMV-7), β3 (HMβ3), and β1 (9EG7) integrins were kindly provided by Dr. B.A. Imhof (Department of Pathology, CMU, University of Geneva, Geneva, Switzerland). The EMD121974 (the inner salt form of cyclic-(Arg-Gly-Asp-[D-Phe]-[N-Me-Val])) and EMD135981 (the inner salt form of cyclic-(Arg-[β-Ala]-Asp-[D-Phe]-Val)) cyclopeptides were synthesized by Dr A. Jonczyk (Merck KGaA, Darmstadt, Germany; ref. 24).

Cell lines and cell culture. The mouse oral squamous cell carcinoma line SCCVII was kindly provided by Dr. R. Hill (Division of Applied Molecular Oncology, Ontario Cancer Institute, Princess Margaret Hospital, Toronto, Canada). The murine hepatoma lines HEPA1-C1C7 (ARNT-functional) and HEPA1-C4 (ARNT deficient) were obtained from Dr I. Desbaillet-Hakimi (Centre Hospitalier Universitaire Vaudois, Lausanne, Switzerland). The human colorectal carcinoma line HCT116 was purchased from American Type Culture Collection (#CCL-247). For all experiments, cells were grown in DMEM high-glucose supplemented with 10% FCS and 1% Penicillin/Streptomycin and maintained in a humidified incubator at 37°C with 5% CO₂. All cell culture reagents were purchased from Invitrogen. To derive cell lines from tumor grafts, tumor-bearing mice were sacrificed by cervical dislocation, tumors were excised under sterile conditions, mechanically disrupted with a scalpel, and digested for 2 h at 37°C with collagenase A (Roche Applied Bioscience). To obtain a single-cell suspension, digested tissues were filtered through an 80-μm mesh to remove tissue debris. Recovered cells were seeded on tissue culture Petri dishes (TPP) and maintained for several passages under standard culture conditions before use in experiments.

For in vitro hypoxia selection, exponentially growing cells were cultured in a hypoxic chamber (Billups-Rothenberg) at 0.1% O₂ for 5 d. Cells were allowed to recover for 3 d at 21% O₂. A total of three cycles were performed for each cell line. Cell lines nomenclature were as follows: PA, parental lines; NIR and IR, lines derived from tumors grown on a nonirradiated or irradiated mice, respectively; mNIR and mIR, lines derived from lung metastases of tumors grown on a nonirradiated or irradiated mice, respectively. HS, lines derived from parental cells through in vitro hypoxia selection.

In vitro invasion assay. Matrigel (3.5 mg/mL for SCCVII and 0.5 mg/mL for HCT116) was polymerised at 37°C in the upper Transwell chambers (pore size, 8.0 μm; Corning Life Sciences). Serum-starved (24 h) SCCVII (8 × 10⁴) or HCT116 (5 × 10⁴) cells were seeded in the upper chamber and incubated in full medium for 24 h. EMD inhibitors were used at 5 μmol/L, 17E6 at 10 μg/mL. Filters were fixed with 4% PFA, stained with 0.5% CV, and migrated cell counted in 4 random fields per membrane under a microscope (n = 4). Values represent means ± SD.

Microarray hybridization. For each type of cell line (PA, NIR, IR, and HS), three lines (#1, #2, and #3) were derived from three independent tumors or by in vitro hypoxia selections and passaged at least twice in vitro under standard conditions. RNA was obtained using RNeasy kit (QIAGEN). Probe synthesis and Murine Genome 430 2.0 Genechip Array (Affymetrix Ltd) hybridization were performed at DNA Array Facility (DAFL). Microarray data are available at Gene Expression Omnibus⁸

Mice irradiation, tumor growth, and metastasis. Six- to 10-wk-old RAGγ2^−/− mice (Charles River Laboratories) were anesthetized (ketamine, 100 mg/kg, i.p.), placed in a lead jig allowing exposure to a 25 × 25-mm field, and irradiated (10–20 Gy single dose) using an X-ray unit (PHILIPS; RT 250) at 220 kV and 20 mA, with a 0.5-mm Cu filter. Two weeks later, tumors were initiated by injecting 10⁴ (SCCVII) or 10⁶ (HCT116 or HEPA1) cells per mouse in 100 μL serum-free medium s.c. in the lower-right back (at the level of the hip) of the mice (6 or more mice per group). Mice were inspected daily and tumors measured with a caliper every 4 d. Tumor volume (V) was calculated as V = π/6 × a × b², where a is the longer and b is the shorter of two orthogonal diameters. For EMD121974, or 17E6 treatment, mice were injected i.p. with 0.75 mg/mouse/d or 1 mg/mouse/wk, respectively, starting when tumors reached a size of ∼10 mm³. For lung metastasis quantification, mice were sacrificed 32 d after tumor cell injection in all conditions (control versus preirradiated mice). Lungs were removed and fixed in 4% PFA, embedded in paraffin, and a total of 4 coronal sections separated by 500 μm per lungs were produced and H&E stained. The total number of nodules per 4 sections (= metastasis per animal) was determined under a binocular microscope (Leica; MZ16). Values stated for each group of animals are means of these numbers ± SD. All experiments with mice were approved by the cantonal veterinary service.

Histopathologic and immunohistochemistry analyses. For tissue morphology evaluation, standard H&E staining procedures were performed on paraffin-embedded tissues. For immunohistochemistry of mouse tumors, detection of microvessels and tumor hypoxia was done on frozen sections using a modified double staining protocol combining a biotinylated rat anti-CD31 for detection of mouse endothelial cells and a monoclonal hydroxyprobe-1 for detection of pimonidazole adducts (Chemicon International). Pimonidazole hydrochloride was injected into tumor-bearing animals 15 to 90 min before sacrifice. Briefly, a total of 4 sections separated by 2 to 5 mm were taken for each tumor, fixed for 10 min in acetone, and blocked for 5 min in 10% normal goat serum. Slides were incubated with primary antibodies (1:50 dilutions in 0.5% BSA) for 2 h at room temperature, and washed and incubated with anti-mouse EnVision horseradish peroxidase and Streptavidin-AP conjugate (Dako) at 1:200 dilution for 30 min. Antibody complexes were revealed by 3,3′-diaminobenzidine (DAB) treatment (Roche) for 9 min followed by Fuchsin+ (Dako) for 20 min. Slides were mounted using aqueous solutions. Quantification of microvascular density was done by calculating the Chalkley score (25). To this purpose, the three most vascular areas were chosen subjectively from each tumor section. A 25-point Chalkley eyepiece graticule was applied to each chosen area, and the number of microvessels hit by the graticule points was recorded. The Chalkley count for an individual tumor was taken as the mean value of the three graticule counts.

Statistical analysis. Results are expressed as mean ± SD. Data were analyzed by Student's t test for both in vitro and in vivo experiments. P values of <0.05 were considered significant. Error bars depict SD. For statistical analysis of gene expression data, see Supplementary Data.

Microarray data analysis, lentiviral constructs, Western blotting, real-time reverse transcription-PCR, and fluorescence-activated cell sorting analyses are described in Supplementary Data.

Tumors growing in previously irradiated stroma are more hypoxic and show increased invasion and metastasis. Tumors locally recurring after radiotherapy (e.g., head and neck, rectal, or breast cancers) often regrow within preirradiated stroma. To experimentally mimic this condition, we locally irradiated the lower back of healthy mice (10 and 20 Gy single dose) and, 2 weeks later, implanted tumor cells s.c. within this area. The biological effect on healthy tissue of such single doses corresponds to cumulative does of up to 60 Gy delivered during fractionated radiotherapy, depending on the chosen α/β values, and are therefore of clinical relevance (26). Stroma preirradiation inhibited local growth of SCCVII (murine squamous cell carcinoma) and HCT116 (human colon carcinoma) tumor cells in a dose-dependent manner (Supplementary Fig. S1A). CD31 staining and pimonidazole staining revealed a dose-dependent decrease of microvascular density and the appearance of larger hypoxic regions in tumors growing in preirradiated skin compared with control tumors (Fig. 1A and B). The appearance of hypoxia was paralleled with the appearance of local invasion (Supplementary Fig. S1B) and with a dose-dependent increase in tumor necrosis (Fig. 1C), and lung metastasis (Fig. 1D).

Tumor growth in preirradiated stroma selects for invasive and metastatic variants independently of HIF. The increased metastasis formation of tumors growing in a preirradiated stroma raised the possibility that tumor aggressiveness was a consequence of sustained tumor hypoxia. Because hypoxia inducible factors, in particular, HIF-1, induce expression of proinvasive and metastatic molecules, we first investigated whether HIF activation was necessary and sufficient for the observed increased rate of metastasis. To address this question, we examined the metastatic capacity of two congenic tumor cell lines with either functional (HEPA1-C1C7) or nonfunctional (HEPA1-C4) HIF (27) by injecting them in control and in locally preirradiated mice. Preirradiation of the prospective microenvironment strongly decreased the growth of both HEPA1-C1C7 and HEPA1-C4 lines in a dose-dependent manner (Supplementary Fig. S1C). The absence of HIF activity in HEPA1-C4 cells inhibited tumor growth on both irradiated and nonirradiated stroma compared with HEPA1-C1C7 grafts (Supplementary Fig. S1C) and strongly reduced the metastatic ability of HEPA1-C4–derived tumors developing on both nonirradiated and preirradiated stroma (Fig. 2A). This result is consistent with the role of HIF as a key mediator of metastasis. However, despite the overall reduced ability of HIF-deficient tumor cells to metastasize, the relative increase (4- to 6-fold) in lung metastasis formation of tumors growing in preirradiated versus nonirradiated stroma, was not affected (Fig. 2A). This result suggested that additional, HIF-independent, mechanisms were involved in the increased aggressiveness of tumors growing in preirradiated microenvironments. One possible mechanism was the selection of highly metastatic tumor cell variants by the hypoxic microenvironment.

To explore this hypothesis, we rederived cell lines from tumors that had developed in preirradiated and nonirradiated stroma (IR and NIR lines, respectively), and compared their invasive and metastatic capacities with their respective PA lines. IR lines were more invasive in vitro under normoxic conditions (Fig. 2B) and more metastatic to the lungs when reinjected s.c. into nonirradiated mice (Fig. 2C). The in vitro growth rates of these lines and the in vivo growth of derived tumors were indistinguishable (data not shown). Importantly, in all derived lines, HIF-1α protein levels were low under normoxia and were effectively induced by hypoxia (Fig. 2D), thereby excluding that invasive/metastatic cell subpopulations were selected on the basis of constitutive HIF-1α activation.

These data show that preirradiated stroma selects for invasive and metastatic tumor cell subpopulations retaining invasive and metastatic capacities even upon return to normoxia and in the absence of constitutive HIF-1 activation.

Gene expression profiling of selected tumor cells. Although the role of HIF in promoting hypoxia-dependent invasion and metastasis is well-established, little is known about alternative molecular pathways conferring a metastatic phenotype enduring beyond this selection process. To identify genes linked to the metastatic capacity of tumor cell escape variants, we performed gene expression studies of SCCVII-PA, SCCVII -NIR, and SCCVII -IR lines under normoxic conditions. Unsupervised gene expression clustering analysis grouped the three conditions separately, confirming distinct and stable gene expression profiles (Fig. 3A and B). Ninety-two probe sets were differentially regulated between the IR and NIR conditions at a cutoff corresponding to a false discovery rate (FDR) of <0.01, and the top 17 probe sets representing 16 genes at a FDR of <0.001 (Supplementary Table S1). We used a quantile-quantile plot to graphically identify those genes deviating more markedly from the bulk of the observations and representing best candidates for significant effects (data not shown). The transcript encoding for cysteine-rich protein 61 (CYR61), a member of the CCN (CYR61/CTGF/NOV) family of matricelluar proteins regulating cell growth, differentiation, survival and migration in development, tissue remodeling, and repair (28), was prominently up-regulated in IR cells. Western blotting analysis validated increased CYR61 protein expression in the SCCVII-IR line (Fig. 3C). Consistent with this observation, we found that levels of CYR61 protein were also stably increased under normoxic conditions in the HCT116-IR line compared with the HCT116-PA and HCT116-NIR lines (Fig. 3C). Elevated levels of CYR61 mRNA and protein were also observed in SCCVII lines derived from lung metastases of tumors growing in preirradiated stroma (mIR) relative to metastases of tumors growing in nonirradiate stroma (mNIR) and parental cells (PA) consistent with a role of CYR61 in selecting metastatic variant from tumors growing in preirradiated stroma (Supplementary Fig. S2).

To see whether hypoxia alone was sufficient to drive the selection of aggressive tumor cells expressing elevated levels of CYR61, we selected SCCVII hypoxia-resistant lines in vitro through repeated cycles of cell culture under harsh hypoxic (0.1% pO2) conditions (called HS line). Global gene expression analysis of three hypoxia-selected versus three PA SCCVII lines confirmed that CYR61 was among the top up-regulated gene in HS cells (FDR < 0.001; data not shown). Western blotting analysis validated increased CYR61 protein expression in HS cells (Fig. 4A). As for IR cells, HIF-1α protein levels in HS cells were low under normoxia and high under acute hypoxia, such as in PA cells (Fig. 4B). When tested in vitro for their invasive capacity, or in vivo for their ability to form lung metastasis, HS cells behaved similarly as IR lines (Fig. 4C).

CYR61 promotes invasion and metastasis. Experimental data indicate that CYR61 promotes tumor progression by stimulating tumor cell growth, angiogenesis, anchorage independence, and invasion (29, 30). Thus far, however, evidence for a prometastatic role of CYR61 was only correlative (28, 30–33). To directly test whether the up-regulation of CYR61 in the IR and HS lines played a causal role in their aggressive phenotype, we performed gain and loss of function experiments on CYR61. CYR61 overexpression in PA lines increased Matrigel invasion in vitro and promoted lung metastasis formation in vivo (Fig. 5A). Down-regulation of CYR61 protein expression by shRNA in IR lines attenuated Matrigel invasion in vitro and reduced lung metastasis in vivo (Fig. 5B). Identical results were obtained with SCCVII cells (data not shown).

These data showed that CYR61 is necessary and sufficient to promote invasion and tumor metastasis formation in the lung.

Invasion and metastasis induced by CYR61 require αVβ5 integrin. CYR61 is reported to exert many of its biological activities through integrin adhesion receptors (33, 34), including those integrins (e.g., αVβ3 and αVβ5) promoting invasion and metastasis (35–37). We found SCCVII and HCT116 cells to express αVβ5 but not αVβ3 integrin by cell surface detection of the receptor (Fig. 6A and data not shown). Modulation of CYR61 expression level (i.e., overexpression and knockdown) had no effect on αVβ5 cell surface level (data not shown).

To test for the functional role of αVβ5 in mediating the proinvasive and prometastatic effects of CYR61, we performed invasion and lung metastasis experiments in the absence or presence of αV integrin inhibitors. Because of the availability of the 17E6, an anti-human–specific αV blocking antibody (38), functional experiments were mostly done using the human HCT116 lines. The αVβ3/αVβ5-specific antagonist EMD121974 (24), currently in clinical development in oncology as cilengitide (39), was also used in vitro and in vivo to obtain clinically relevant proof-of-concept experiments. Treatment with EMD121974, or with 17E6, inhibited in vitro Matrigel invasion by IR, HS, and CYR61-overexpressing HCT116 lines (Fig. 6B). EMD121974 and 17E6 treatment reduced invasion of PA, IR, and CYR61-overexpressing cells below basal (i.e., no treatment) levels, consistent with a role of αVβ5 in both basal and CYR61-enhanced invasion. These results also show that the αVβ5-dependent invasion-promoting effect of tumor cell–derived CYR61 was cell autonomous because no other cells were present in the assay.

To test whether such a cell-autonomous effect also occurred in vivo, we injected HCT116 CYR61 or mock transfectants into nonirradiated mice and started 17E6 treatment when tumors were palpable (∼10 mm³). MAb 17E6 specificity for human αV allowed us to rule out any putative effect on the host microenvironment. 17E6 treatment inhibited enhanced lung metastasis capacity of HCT116 cells overexpressing CYR61 (Fig. 6C). To obtain a clinically relevant proof-of-concept whether αVβ5 integrin inhibition could be a therapeutic approach to prevent metastatic spreading of tumors growing within a preirradiated field, we injected HCT116 cells into preirradiated or control mice and treated them with EMD121974. EMD121974 treatment effectively prevented metastasis formation (Fig. 6D). Importantly, primary tumor growth was not affected (data not shown).

Taken together, these results show that tumor cell–derived CYR61 cooperates with tumor cell αVβ5 integrin to mediate invasion and metastasis.

Tumors recurring within a previously irradiated field have increased probability of developing metastasis (4, 8–10). This is a clinically relevant situation because 10% to 20% of the patients treated by radiotherapy will relapse within the irradiated field and progress to form metastases. To date, there are no effective therapeutic options to prevent or treat these relapses. The presented data give new insights in the mechanisms involved and provide a rational for a therapeutic approach to prevent postradiation tumor escape.

A first key finding of this study is the demonstration that tumors growing within a preirradiated stroma are selected for escape variants that retain an invasive and metastatic phenotype even upon return to normoxia. Hypoxia seems as the main driving force in this process because in vitro selection through repeated cycles of hypoxia generated highly metastatic tumor cells that behaved indistinguishably from the in vivo obtained escape variants (i.e., increased levels of CYR61 expression, enhanced αVβ5-depedent invasion, and metastasis). The role of hypoxia in promoting tumor progression and metastasis is well-documented by experimental and clinical observations, including in the context of the tumor bed effect (13–16, 21, 40–42). It has been largely considered that hypoxia promotes invasion and metastasis by eliciting an adaptive response through HIF-1–dependent transcriptional activation of metabolic and invasive programs (22, 23). Our results, however, suggest that HIF-1–dependent events, although important, are not essential for metastatic spreading of hypoxic tumors: HIF-deficient cells can form enhanced metastasis when growing in irradiated skin; HIF-1 levels are not constitutively increased in escape variants; and no HIF-1 gene expression signature was seen in IR and HS lines. The fact that escape variants retain enhanced adhesive, invasive, and metastatic capacities upon return to normoxia, suggest a mechanism involving hypoxia-mediated selection rather than adaptation, which by definition is transient. A selection mechanism is consistent with an emerging paradigm of cancer metastasis, whereby cells with a basal predisposition to metastasize are already present in the primary tumor, and the full development of their metastatic capacity requires the expression of only a few specific complementary functions in addition to the bulk of features promoting growth and survival at the primary site (43, 44). Thus, the risk of metastasis is proportional to the fraction of tumor cells expressing these complementary functions. Indeed, we have found a limited number of differentially expressed genes between parental and in vivo escape variants or in vitro hypoxia-selected variants. These differentially expressed genes (signature) include those (i.e., CYR61 and ITGB5) causally involved in metastasis.

The second key finding of this work is the identification of CYR61 and αVβ5, as molecules cooperating in promoting metastasis. CYR61 may belong to the class of prometastatic protein expressed in the primary tumor, such as ID1, CXCL1, cyclooxygenase 2, EREG, and matrix metalloproteinase 1, whose expression increases with metastatic ability (45). Importantly, CYR61 was not induced by hypoxia, consistent with a previous study demonstrating that, in advanced melanoma, CYR61 expression occurs independently of hypoxia (46). Intriguingly, HRG, fibroblast growth factor 2, platelet-derived growth factor, and transforming growth factor β, four cytokines promoting cancer progression were reported to induce CYR61 expression (47), raising the possibility that CYR61 expression may be driven by those genetic or epigenetic events promoting malignant progression. Integrin contribution to cancer progression is well-recognized (48). The role of αVβ5 in promoting metastasis, however, has remained unclear because divergent results were reported using different experimental models (35, 49). αVβ5 was shown to mediate CYR61-stimulated fibroblast migration (50) and to promote resistance to apoptosis in MCF-7 cells (51). Here, we provide original evidence that αVβ5 is essential for CYR61-induced metastasis. It remains to be seen whether CYR61 induces these effects by acting as a canonical αVβ5 ligand, or by modulating αVβ5 function indirectly through an alternative receptor and intracellular signaling events.

In conclusion, this study represents a conceptual advance to the understanding of the tumor bed effect by demonstrating that preirradiated tissue selects for stable, aggressive tumor cell variants. Our data identifies CYR61 and αVβ5, as molecules that cooperate to promote metastasis, and suggest a therapeutic approach for preventing escape, which can be promptly tested in the clinic. The proposed model is summarized in Supplementary Fig. S3.

No potential conflicts of interest were disclosed.

Grant support: Swiss National Science Foundation (SNF, 3100A0 118079), Oncosuisse (OCS 0181212-12-2005 and 02020.02.2007), Gebert Rüf Stiftung, the National Center for Competence in Research Molecular Oncology, a research instrument of the SNF, and the Medic Foundation.

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.

We thank F. Lejeune for continuous support, G. Christofori, F. Bosman, Ph. Monnier, B.A. Imhof, and I. Stamenkovic for discussion; R. Hill, B.A. Imhof, I. Debaillet Akimi, L. Lau, L. Naldini, R. Iggo, A. Follenzi, A. Jonczyk, and M. Frech for kindly providing reagents; O. Hagenbuchle, K. Harshman (DAFL), L. Ponsonnet, and the Microscopy, Imaging Morphology Facility at Swiss Institute for Experimental Cancer Research for help with experiments, R. Driscoll for reading of the article and the University of Lausanne DAFL and Swiss Institute for Bioinformatics Vital-IT project for infrastructure support.