Inhibition of vascular endothelial growth factor (VEGF) signaling, a key regulator of tumor angiogenesis, through blockade of VEGF receptor (VEGFR)-2 by the monoclonal antibody DC101 inhibits angiogenesis, tumor growth, and invasion. In a surface xenotransplant assay on nude mice using a high-grade malignant squamous cell carcinoma cell line (A-5RT3), we show that DC101 causes vessel regression and normalization as well as stromal maturation resulting in a reversion to a noninvasive tumor phenotype. Vessel regression is followed by down-regulation of expression of both VEGFR-2 and VEGFR-1 on endothelial cells and increased association of α-smooth muscle actin–positive cells with small vessels indicating their normalization, which was further supported by a regular ultrastructure. The phenotypic regression of an invasive carcinoma to a well-demarcated dysplastic squamous epithelium is accentuated by the establishment of a clearly structured epithelial basement membrane and the accumulation of collagen bundles in the stabilized connective tissue. This normalization of the tumor-stroma border coincided with down-regulated expression of the stromal matrix metalloproteinases 9 and 13, which supposedly resulted in attenuated turnover of extracellular matrix components permitting their structural organization. Thus, in this mouse model of a human squamous cell carcinoma cell line, blockade of VEGF signaling resulted in the reversion of the epithelial tumor phenotype through stromal normalization, further substantiating the crucial role of stromal microenvironment in regulating the tumor phenotype.
The importance of the stromal compartment in malignant tumors is increasingly recognized, and many reports indicate that continuous interactions between carcinoma and stromal cells are prerequisite for tumor development and progression (1–8) . On the other hand, earlier and recent studies showed that malignant cells can be reverted to form a normal tissue when exposed to the proper tissue context (for review, see ref. 9). Although the importance of the stromal microenvironment consisting primarily of fibroblasts, endothelial cells, and inflammatory cells as well as extracellular matrix (ECM) components for tumor development and control is increasingly appreciated, the molecular mechanisms underlying the interactions between stromal and tumor compartments are still mostly obscure. In vitro models developed thus far are only partially able to faithfully mimic the complex interactions between tumor cells and stromal elements in vivo (for review, see refs. 9, 10 ). This is particularly true for the process of tumor angiogenesis, a predominant feature of the tumor stroma. The essential role of the formation of a new vasculature for tumor progression has been fully recognized in the last decade and is increasingly used as a target for tumor therapy (for review, see refs. 11–13 ).
The vascular endothelial growth factor (VEGF) has been identified as a key regulator of tumor angiogenesis and VEGF receptor (VEGFR)-2 (also called KDR in humans and flk-1 in the mouse) as the major mediator of the mitogenic and permeability-enhancing effects of VEGF in endothelial cells (for review, see refs. 14, 15 ). In addition, VEGF is a survival factor for endothelial cells, and a marked dependence on VEGF has been shown in newly formed but not established tumor vessels (16). The loss of this VEGF dependence and the concomitant vessel remodeling is marked by coverage of the capillaries by pericytes (17). Although the field of tumor angiogenesis is an area of extensive research, the consequences of enhanced angiogenesis and its reversion on tumor growth and progression, respectively, are only partially elucidated (12, 13, 18) and mostly interpreted in terms of nourishment and oxygenation of tumor cells.
To study the complex tumor-stroma interplay in more detail and to circumvent some of the difficulties encountered in conventional experimental tumor models, we have elaborated an in vivo assay of tumor invasion (7, 10) . This matrix-inserted surface transplantation model (19) combines several crucial advantages over the conventional tumor transplants by injection of cell suspensions (7, 10, 20) . It allows the study of stromal interactions with normal and transformed epithelia of different tumor stages, thereby providing a clear distinction between premalignant and malignant epithelial cells. Benign and premalignant squamous epithelial cells form differentiated stratified epithelia separated by a basement membrane from a wound-like stroma (21, 22) , which is characterized by a transient angiogenesis followed by stromal maturation (10). In contrast, malignant cells form dysplastic epithelia, which invade the host granulation tissue, exhibiting persistent angiogenesis and stromal activation (23, 24) . Coincident with tumor cell invasion, stromal strands with newly formed blood vessels infiltrate the tumor parenchyma, a feature characteristic for malignant transplants (10).
Using this model, we have accumulated evidence that the malignant tumor phenotype, particularly tumor cell invasion, can be modulated by exogeneously induced alterations in the tumor stroma: (a) By disturbing the balance between serine proteases and their inhibitor plasminogen activator inhibitor-1 in the host stroma (using plasminogen activator inhibitor-1 knockout animals), tumor infiltration by blood vessels was blocked; subsequently, tumor invasion was completely abrogated (25). Viral vector–mediated reconstitution of plasminogen activator inhibitor-1 in the knockout animals resulted in reestablished tumor vascularization and invasion (26). (b) By constitutive overexpression of platelet-derived growth factor-B in nontumorigenic immortal HaCaT keratinocytes, conversion to a benign tumorigenic phenotype was induced. This was mediated via stromal activation, because HaCaT cells do not express platelet-derived growth factor receptors (23). 4 (c) Transfection and overexpression of the antiangiogenic ECM component thrombospondin-1 in a skin squamous cell carcinoma (SCC)–derived cell line resulted in blockade of tumor vascularization, invasion, and expansion. When thrombospondin-1 expression was reduced to normal levels by antisense oligonucleotides, tumor vascularization, invasion, and expansion were reestablished (27). (d) More impressively, when interfering with the signaling of VEGFR-2 by using the blocking antibody DC101 (28), reduced microvessel density was accompanied by complete abrogation of tumor invasion (24, 28) . This shifted the malignant and invasive tumor epithelium into a differentiated premalignant phenotype.
These latter findings were observed using an early-stage, differentiated skin SCC cell line (II-4). In the present study, we analyzed whether inhibition of angiogenesis in transplants of a late-stage malignant and metastasizing skin carcinoma cell line (A-5RT3; ref. 29) resulted in a similar reversion of the tumor phenotype. Additionally, we aimed at elucidating the mechanistic details that are at the basis of the stromal alterations induced by inhibition of angiogenesis and may thereby contribute to the observed reversion of the tumor phenotype. Here, we provide evidence that this reversion is probably caused by reduced stromal protease expression resulting in a normalization of the tumor-adjacent stroma. These data clearly support the concept that alterations in the tumor stroma can regulate the tumor phenotype.
Materials and Methods
Cells and Culture Conditions. The highly malignant tumorigenic clone (A-5RT3) was derived from the immortalized human keratinocyte cell line HaCaT (30) after transfection with the c-Ha-ras oncogene and by recultivation of heterotransplants in nude mice as described (20, 29, 31) . All cells were grown in enriched minimal essential medium (4× MEM) supplemented with 5% FCS and 200 μg/mL geneticin as described (29).
Surface Transplantation Assay. Cells were transplanted onto the dorsal muscle fascia of 7- to 9-week-old nude mice (Swiss/c nu/nu back-crosses) as monolayer cultures growing on collagen type I gels using a silicone chamber device (10, 24, 32) . Transplants were dissected en bloc, embedded in Tissue-Tek (Miles Laboratories, Elkhart, IN), and frozen in liquid nitrogen vapor for preparation of cryostat sections. For labeling of proliferating cells, mice received tail vein injections of 5-bromodeoxyuridine (BrdUrd) and 2-deoxycytidine (65 mmol/L each) in 0.9% NaCl (100 μL) 1.5 hours before being killed.
Tumorigenicity Test. A-5RT3 cells (2 × 105) in 100 μL culture medium were injected s.c. in the back of 7- to 9-week-old nude mice (Swiss/c nu/nu back-crosses). Tumor formation was assayed weekly, two diameters were determined by calipers over an observation period of 60 days, and tumor size was calculated. Tumors were removed en bloc and processed like the transplants.
VEGFR-2 Inhibition by DC101. In vivo antiangiogenic activity of the VEGFR-2 neutralizing antibody DC101 was tested in mice carrying s.c. tumors and transplants of the highly malignant keratinocyte clone HaCaT-ras A-5RT3 starting 2 and 14 days after transplantation, respectively. Mice received i.p. injections of the monoclonal antibody DC101 (500-800 μg per mouse as indicated in 150 μL PBS) or PBS at different time intervals (from twice weekly to every second day) for a total of 10 weeks. Transplants were dissected after different time points. The experiment was repeated thrice. All animal experiments were done in compliance with the relevant laws and institutional guidelines with permission of the “Regierungspräsidium Karlsruhe” dated July 4, 1999 and November 3, 2003 (AZ 35-9185.81/G-16/03).
Antibodies and cDNAs. The rat monoclonal antibody DC101 to mouse VEGFR-2 (flk-1) was obtained from ImClone Systems, Inc. (New York, NY) and described in detail previously (28). Rat monoclonal antibody against mouse CD31 was obtained from BD PharMingen (Heidelberg, Germany), guinea pig polyclonal antiserum against cytokeratins (pan) from Progen (Heidelberg, Germany), sheep polyclonal antibody against BrdUrd from NatuTec (Frankfurt, Germany), rabbit polyclonal antibody against tenascin C from Telios Pharmaceuticals (San Diego, CA), biotinylated mouse monoclonal antibody against α-smooth muscle actin (α-SMA) from Progen, rat monoclonal antibody against mouse neutrophil granulocytes from Serotec (Düsseldorf, Germany), sheep polyclonal antibody against mouse matrix metalloproteinase (MMP)-9 was a gift from Prof. Gillian Murphy (University of Cambridge, Cambridge, United Kingdom), and rabbit polyclonal antibody against mouse collagen type IV from Novotec (Lyon, France). Secondary antibodies were obtained from Dianova (Hamburg, Germany) and Hoechst 33258 bisbenzimide for nuclear staining from Sigma-Aldrich (Taufkirchen, Germany).
Human cDNA encoding VEGF was provided by H. Weich (Society for Biotechnological Research, Braunschweig, Germany) and D. Marmé (Institute for Molecular Oncology, Freiburg, Germany). Mouse VEGFR-2 (flk-1) and VEGFR-1 (flt-1) cDNAs were provided by G. Breier and W. Risau (Max Plank Institute for Physiological and Clinical Research, Bad Nauheim, Germany). Mouse cDNAs encoding MMP-9 and MMP-13 were a gift from Prof. Gillian Murphy.
Indirect Immunofluorescence. For immunofluorescence staining, frozen sections were fixed for 5 minutes in 80% methanol at 4°C and 2 minutes in acetone at −20°C and rehydrated in PBS. For BrdUrd localization in DNA, sections were additionally denatured in 2 mol/L HCl for 10 minutes at room temperature and washed (3 × 10 minutes). Primary antibodies were incubated in 12% bovine serum albumin/PBS at room temperature for 2 hours or 4°C overnight. After washing (3 × 10 minutes), sections were incubated with appropriate secondary antibodies together with 5 μg/mL Hoechst 33258 bisbenzimide for staining of cell nuclei. Before embedding in Permafluor (Immunotech, Marseilles, France) sections were washed again (3 × 10 minutes) in PBS.
Detection of apoptotic cells was done using the In situ Cell Death Detection kit (Roche, Mannheim, Germany) following the manufacturer's protocol. Stained sections were examined and photographed with a Olympus AX-70 microscope fitted with epifluorescence optics.
In situ Hybridization. In situ hybridization was essentially done as described (33). In brief, 35S-labeled RNA probes for human VEGF, mouse VEGFR-1 (flt-1) and VEGFR-2 (flk-1), and digoxigenin-labeled RNA probes for mouse MMP-9 and MMP-13 were prepared using T3, SP6, or T7 RNA polymerase (for antisense and sense, respectively) according to the manufacturer's instructions (Roche). Cryostat sections were fixed in 4% paraformaldehyde, pretreated, hybridized, and washed at high stringency as described (22). For autoradiography, slides were coated with NTB2 film emulsion and exposed for 3 weeks. After the film was developed, the sections were counterstained with H&E. For nonradioactive in situ hybridization, digoxigenin was labeled by anti–digoxigenin/alkaline phosphatase (Roche), and alkaline phosphatase reaction was detected by nitroblue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate (Life Technologies/Invitrogen, Eggenstein-Leopoldshafen, Germany). After digoxigenin in situ hybridization with MMP-9 or MMP-13 probes, counterstaining was done by indirect immunofluorescence with antisera against pankeratin and collagen type IV. Stained sections were examined and photographed with an Olympus AX-70 microscope fitted with epifluorescence optics. Due to better visualization of digoxigenin in situ signals, colors were reassigned with analySIS software (Soft Imaging Systems, Münster, Germany).
Transmission Electron Microscopy. Fresh samples of A-5RT3 transplants of control and DC101-treated mice were prefixed in ice-cold 4% glutaraldehyde in 0.2 mol/L sodium cacodylate buffer (pH 7.3) for 3 hours and postfixed in 2% osmium tetroxide in 0.1 mol/L sodium cacodylate buffer for 2.5 hours at 4°C. Tissue blocks were then washed with distilled H2O, stained en bloc with 0.5% aqueous uranyl acetate overnight at 4°C, and again washed with distilled H2O. Following dehydration through two graded series of ethanol and infiltration with propylene oxide, specimens were embedded in Epon 812 equivalent (glycidether 100, Serva, Heidelberg, Germany) and finally polymerized at 60°C for 48 hours. Semithin sections of 1 μm were stained with 0.1% toluidine blue for light microscopy. Ultrathin sections (50-90 nm) were cut by a Reichert Young ultramicrotome, counterstained with uranyl acetate and subsequently lead citrate, and examined with a Zeiss EM10B electron microscope.
Morphometric Analysis. Quantification of BrdUrd incorporation, terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling (TUNEL) staining, vessel density, α-SMA-positive cells, and pericyte-associated vessels was done using analySIS software.
In brief, BrdUrd incorporation in tumor cells was detected using two to four fields of view (2.38 mm2) per animal (two to three per time point) and calculated against total tumor cell nuclei. TUNEL staining (two to three per time point) was quantified using whole mount pictures (14.8 mm2) measuring apoptotic (TUNEL positive) and vital (TUNEL negative) against total tumor areas. Mean vessel density was determined using whole mount pictures (3.44 mm2) from CD31 immunostained transplants (two to three animals per time point) and calculating CD31-stained areas, 300 μm below and 500 μm within the tumor, per 100 μm2 area.
The number of vessels associated with α-SMA-positive cells (pericytes) was quantified using two to four vision fields (1.36 mm2) per animal, 300 μm below and 500 μm within the tumor parenchyma, analyzing two to three mice per time point, thereby counting every vessel area with overlapping staining by at least one α-SMA-positive spot and calculating the percentage of pericyte-associated vessels.
Quantification of MMP-9 expression was done in control and DC101-treated transplants (8 + 10 weeks old) in a region from 150 μm below the tumor parenchyma up to 500 μm inside the tumor tissue, measuring three vision fields (1.1 mm2) of three animals per group.
VEGFR-2 Blocking Antibody DC101 Inhibits Tumor Growth and Vascularization. On s.c. injection, the late-stage and metastasizing human SCC cell line A-5RT3 (29) formed very fast growing tumors that reached sizes between 1 and 2 cm3 within 4 weeks so that control animals had to be sacrificed ( Fig. 1 ). DC101 treatment by repeated i.p. injections starting 2 days after cell injection caused a significant dose-dependent retardation of tumor growth ( Fig. 1A). On discontinuation of treatment (6 weeks later), tumor growth restarted at the initial rate. When the start of treatment was delayed for 2 weeks allowing tumors to reach an average size of ∼300 mm3, a similar dose-dependent therapeutic effect was achieved ( Fig. 1B).
Importantly, on treatment with DC101 (720 μg/animal, every third day, start 14 days after s.c. injection), the highly vascularized and infiltrating carcinoma phenotype seen in control animals treated with irrelevant antibodies or PBS ( Fig. 2A ) had been modulated into a premalignant phenotype. Histologically, the small residual tumors in DC101-treated animals were largely necrotic, except for a small rim of vital tumor tissue that was essentially avascular and associated with a well-vascularized stromal tissue ( Fig. 2B). The blockade of tumor vascularization by DC101 was transient, and 3 weeks after discontinuation of treatment (see Fig. 1), large and strongly vascularized tumor areas had reformed ( Fig. 2C).
Altered Tumor-Stroma Interactions Abrogate Tumor Vascularization and Invasion. To better understand the altered interactions between tumor and stromal cells following inhibition of angiogenesis, A-5RT3 carcinoma cells were studied in surface transplants. In this assay, tumor cells grow and invade from the upper side and stromal reactions advance from the opposite lower side ( Fig. 3A ), thus rendering tumor-stroma interactions more clearly visible (for review, see ref. 7). On transplantation, tumor cells induced rapid stromal activation with migration and proliferation of fibroblasts and inflammatory cells and growth of new blood vessels in a directed and well-defined pattern. This was followed by tumor cell invasion and reciprocal infiltration of vascularized stromal strands into the tumor parenchyma as visualized by differential immunostaining in 2-week-old transplants ( Fig. 3B). When DC101 treatment (800 μg/animal, every second day) started at this time point (2 weeks), vascularization of the tumor tissue was rapidly reversed and tumor cell invasion was strongly inhibited after 2 weeks of treatment ( Fig. 3C). Whereas control transplants developed into large tumor masses interspersed with well-vascularized stromal regions within 4 to 10 weeks ( Fig. 3D and F), a thin compact but avascular epithelium covered by large necrotic areas remained in transplants treated with DC101 for 8 weeks ( Fig. 3E). Most importantly, the avascular but vital tumor zone exhibited no or only minor invasion into the underlying stromal tissue throughout the entire transplant ( Fig. 4A and B ). Microvessel density was morphometrically quantified in the two areas of most active vessel alterations (i.e., in the stromal area below the tumor parenchyma and in the stromal strands projecting into the epithelial tumor tissue). The drastic reduction in CD31-stained vessel areas was obvious in 4-week transplants (i.e., after 2 weeks of DC101 treatment; Fig. 4C, compare also Fig. 5C and Fig. 5B and D). Initially, this concerns both the areas within and below the tumor parenchyma, whereas at later time points (i.e., in 8-week transplants; 6 weeks of DC101 treatment), vessels were barely found within the tumor tissue, whereas their presence in the stroma below the tumor was maintained, although at a reduced frequency, compared with controls.
The visible reduction in vital tumor tissue coincided with a reduced but not entirely abrogated proliferative activity of tumor cells when compared with untreated controls ( Fig. 4D). Whereas the control transplants exhibited a 1:1 ratio of vital and apoptotic tissue area ( Fig. 4E), up to 80% of the transplants treated for 6 weeks by DC101 were necrotic and stained positive for apoptotic cells ( Fig. 4E).
VEGFR-2 Blockade Causes Vessel Regression and Maturation. These findings suggested that DC101 not only blocked the formation of new blood vessels but also caused regression of preexisting tumor vessels. This was particularly obvious in 4-week-old transplants, 2 weeks after initiation of DC101 treatment (800 μg/animal, every second day), evidenced by the lack of vessels in stromal strands of treated compared with control tumor tissue ( Fig. 5A and B). Intratumoral stromal areas were visualized by staining the ECM with an antibody to tenascin C (green), a predominant component of tumor stroma ECM (34). In control transplants, most of the tenascin C–stained intratumoral stromal strands were vascularized (i.e., costained with an antibody to CD31; Fig. 5A and C). In contrast, in DC101-treated tissues, these fingerlike projections of stroma into the tumor epithelium were essentially free of vessels when counterstained for CD31 (red; Fig. 5B). This effect was even more obvious in 6-week transplants [i.e., after 4 weeks of treatment when the residual fingerlike stromal strands were completely free of vessels ( Fig. 5D) but well vascularized in controls ( Fig. 5C)]. Clearly, preexisting vessels within these stromal strands had regressed and their regrowth was prevented by blockade of VEGFR-2. This regression of small vessels caused by DC101 treatment is first visible after 2 days of treatment as visualized by nuclear magnetic resonance tomography (35) and immunohistochemistry, 5 respectively.
In addition to vessel regression, the remaining small vessels in the DC101-treated transplants exhibited features of a stabilized vessel phenotype as indicated by the down-regulation of expression of both VEGFRs ( Fig. 5E-H) and endothelial cell proliferation as well as by increased association of vessels with α-SMA-positive cells, presumably pericytes (see below). At the same time, expression of VEGF in A-5RT3 tumor cells remained essentially unaltered independent of the DC101 treatment (data not shown; but see ref. 24). VEGFR-2 was strongly expressed in control transplants throughout the observation period (e.g., at 3 and 7 weeks) in both stromal areas immediately adjacent but also more distant to the tumor epithelium ( Fig. 5E and G). This was reduced in 3-week-old transplants (i.e., after 1 week of DC101 treatment; Fig. 5F) and barely detectable after 5 weeks of treatment (i.e., in 7-week-old transplants; Fig. 5H). The decline in expression was observed both in the areas of vessel regression (i.e., immediately adjacent and within the tumor epithelium but also in the granulation tissue more distant to the tumor stroma border). Comparably, expression of VEGFR-1 declined and disappeared by a similar time course in the stroma of DC101-treated transplants (data not shown). This down-regulation of both VEGFRs and the absence of BrdUrd labeling in the vessel endothelia (data not shown; but see ref. 24) are strong indications that in addition to vessel regression endothelial cells in the remaining small vessels had been rendered quiescent by DC101 treatment.
The quiescent and stable phenotype of vessels in DC101-treated transplants was further substantiated by their association with α-SMA-positive cells, presumably pericytes. In 2-week-old control transplants, the newly formed vessels in the tumor vicinity were mostly devoid of α-SMA-positive perivascular cells, whereas a large number of α-SMA-positive cells, most probably myofibroblasts, were dispersed in the stroma ( Fig. 6A ) resembling wound granulation tissue (36). Only very few of these α-SMA-positive cells overlapped with endothelial lining of small vessels at the tumor border, whereas vessels in the tumor-distant stroma exhibited clear association with α-SMA-positive cells ( Fig. 6B). In addition, at later time points (6 and 10 weeks after transplantation), smaller vessels in the tumor-adjacent stroma of control transplants were virtually devoid of α-SMA-positive cells but still accompanied by isolated myofibroblasts ( Fig. 6C and D). In transplants treated with DC101 for 4 and 8 weeks, however, vessels were increasingly associated with α-SMA-positive cells visible by the yellow color of merged double immunostained vessels, whereas free myofibroblasts were absent ( Fig. 6E and F). Morphometric quantification showed that the percentage of vessels associated with α-SMA-positive cells was low in the tumor stroma of 8 and 10-week-old control transplants (7.8 ± 4.9%), whereas it raised to 43.9 ± 13.7% in the age-matched DC101-treated transplants ( Fig. 6G). Although the overall number of α-SMA-positive cells decreased with time in control transplants, almost no free α-SMA-positive cells (myofibroblasts) were visible in the stroma of treated transplants indicating stromal maturation.
Vessel maturation and stabilization was shown even more convincingly by the normalization of their ultrastructure. In control transplants, capillaries were strongly dilated with a very thin and interrupted endothelial lining and a discontinuous basement membrane, associated with extravasation of erythrocytes into the tumor tissue ( Fig. 7A and B ). In contrast, in DC101-treated transplants, small vessels were lined with well-formed endothelial cells associated with pericytes and encircled by a continuous basement membrane, clearly indicating vessel maturation and normalization ( Fig. 7C and D). Moreover, these mature vessels were embedded in a matrix rich in thick bundles of collagen fibers, separating vessels from the tumor epithelium.
Stromal Normalization Reverses Tumor Phenotype. Normalization of blood vessels in DC101-treated transplants was accompanied by features of stromal maturation, particularly a normalization of the tumor-stroma border zone manifested at the ultrastructural level. In control transplants, large tumor masses with a disorganized contact zone between tumor and stromal cells with multiple protrusions and vesicles of tumor cells projected into the underlying stroma ( Fig. 8A and B ). As a consequence, a clear tumor-stroma border was indistinguishable in later transplants with large tumor masses. In contrast, as seen in the immunostained sections (see Fig. 4A and B), DC101 treatment (800 μg/animal, every second day, start 14 days after transplantation) resulted in a compact and polarized tumor tissue clearly demarcated from a dense stroma ( Fig. 8C). In addition to the closely packed tumor cells, the stromal border zone became sharply delineated defining a rather straight basal tumor cell pole to a stroma rich in bundles of collagen fibers ( Fig. 8D). More importantly, the basal aspect of the tumor epithelium was lined by condensations of basement membrane structures, defined by long stretches of a typical lamina densa connected by anchoring fibrils to collagen type I filaments ( Fig. 8E). Additionally, at the tumor cell membrane, hemidesmosomes had formed with keratin filaments inserted at their inner side ( Fig. 8E and F). These features have never been seen in control transplants and are clearly recapitulating a normalization of the epithelial stromal border typical for transplants of normal and benign tumorigenic keratinocytes (21, 22, 37) . Although immunohistochemical staining of basement membrane components at the epithelial-tumor border has often been seen in transplants of malignant keratinocytes (21), ultrastructural features of basement membrane have not been observed to date. This suggests that basement membrane constituents are continuously produced in malignant tumors but rapidly metabolized due to enhanced protease activity, thus preventing their accumulation and structural organization.
VEGFR-2 Blockade Abrogates MMP-9 and MMP-13 Expression. Matrix-degrading metalloproteases, particularly interstitial collagenases and gelatinases, have been found to be differentially regulated in benign and malignant tumor cells and, importantly, in the adjacent stroma (38–41) . Whereas stromal MMP-9 and MMP-13 were only transiently up-regulated in benign transplants, these MMPs were strongly and persistently expressed in the stroma of malignant transplants (7). 6 In malignant control transplants, stromal MMP-9, as visualized by a mouse specific antibody, was localized in the tumor-associated stroma, particularly in the infiltrating stromal strands mostly adjacent to blood vessels ( Fig. 9A ) and colocalized with neutrophil granulocytes. 6 On reduction of the infiltrating stromal areas in the tumor epithelium of DC101-treated transplants, MMP-9 labeling was reduced within the tumor tissue but still present in the underlying tumor-adjacent stroma ( Fig. 9B). Morphometric quantification of MMP-9 labeled areas in transplants from 150 μm below the tumor parenchyma up to 500 μm inside the tumor tissue revealed only minor and nonsignificant reduction. In brief, in control transplants, the MMP-9-positive area revealed 7.3 ± 2.35%, whereas DC101-treated transplants showed 6.2 ± 2.2% positive MMP-9 staining.
At the RNA level, however, analyzed by in situ hybridization with a mouse-specific probe, striking differences in MMP-9 expression between control and DC101 transplants were observed. This effect became even more obvious at the RNA level analyzed by in situ hybridization. Stromal MMP-9 expression in control transplants was restricted to distinct areas in the vicinity to blood vessels at the immediate tumor border ( Fig. 9C). Following DC101 treatment (800 μg/animal, every second day, start 14 days after transplantation), this expression was drastically reduced to a few spots at the tumor-stroma border ( Fig. 9D). For murine MMP-13, a mouse interstitial collagenase, a similar localization of RNA expression closely associated with intratumoral vessels was seen in control transplants ( Fig. 9E). Again, DC101 treatment abolished this expression, with the exception of a few spots at the tumor-stroma border zone ( Fig. 9F). These data indicate a drastic and consistent down-regulation of a major collagenase (MMP-13) and gelatinase (MMP-9), respectively, in the tumor-stroma border zone following VEGFR-2 blockade by DC101.
The role of the decline of these proteases for the reversion of the tumor phenotype has still to be substantiated. Nevertheless, it is tempting to speculate that their reduced expression was causal for the observed normalization of the stromal microenvironment, which had important consequences on the tumor phenotype and can revert a highly malignant into a premalignant tissue.
VEGFR-2 Blockade Causes Normalization of Tumor Stroma and Reverts Tumor Phenotype. The results of this study show that in a mouse model of tumor invasion using the human skin SCC line A-5RT3 inhibition of VEGF signaling by blocking its interaction with VEGFR-2 has wide-ranged consequences for tumor-stroma interactions leading to reversion of the tumor phenotype. Blockade of VEGFR-2 resulted in inhibition of tumor growth and in abrogation of tumor cell invasion. This not only applies to low-grade malignant tumor cells as shown earlier (24) but also is even more impressive in transplants of the high-grade and metastatic SCC cell line A-5RT3 used in this study. VEGFR-2 blockade induced a phenotypic shift from a highly malignant to a premalignant (i.e., noninvasive tumor phenotype). This was associated with regression of immature and maturation of preexisting vessels as well as with the remodeling of the activated tumor stroma into a stabilized connective tissue. Most importantly, blockade of VEGFR-2 resulted in normalization of the tumor-stroma border with restoration of mature basement membrane structures, including hemidesmosomes at the basal pole of tumor cells. This maturation of the tumor-stroma border zone with accumulation of collagen bundles and reformation of basement membranes is most likely due to a down-regulation of matrix-degrading proteases in the stroma as shown for mRNA expression of murine MMP-9 and MMP-13. This decrease in essential ECM-degrading enzymes could then result in a reduced turnover of crucial basement membrane constituents, enabling their accumulation and structural organization. As a consequence, structured basement membranes and a stable connective tissue were formed indicated by the accumulation of collagen bundles.
The surprising normalizing effects on tumor cells with disappearance of membrane protrusions and vesicles and reformation of hemidesmosomes was not caused by blockade of a direct VEGF effect on tumor cells. The human A-5RT3 tumor cells express VEGFR-2 (the human KDR) in vitro 7; however, the mouse-specific (anti-flk-1) DC101 antibody does not cross-react with the human VEGFR-2 (28). Thus, the normalizing effect on the tumor cell phenotype must be caused indirectly by stromal alterations resulting in the formation of features reminiscent of a fibrotic tissue and highlighting again the essential role of the tumor microenvironment (3–5) . Because these stromal alterations were initiated by down-regulation of endothelial cell activity, endothelial cell activation and maturation, respectively, may have essential functional consequences for the pathophysiology of the tumor stroma. Although these features of stromal maturation and tumor phenotype reversion are also seen in s.c. tumors, they are particularly well shown in surface transplants with their distinct geometry and the well-defined stromal development (for review, see ref. 7). In this in vivo model, early changes at the tumor-stroma border are manifest and amendable for detailed analysis, although, as usual in in vivo models, quantification of results is rather difficult and remains restricted to morphometric analysis of vascularization by determining mean vessel density (12, 42, 43) or by counting nuclei with incorporated BrdUrd as mean of cell proliferation.
Abrogation of VEGF Signaling Results in Vessel Regression and Maturation. The consequences of blocking VEGF signal transduction observed in this study are clearly exceeding the classic role of VEGF as mitogenic, motogenic, and survival factor for endothelial cells (11–15) . For more than a decade, the central role of VEGF in the regulation of angiogenesis, particularly in tumor angiogenesis, has been substantiated by many studies (for review, see refs. 14, 15, 18, 43, 44 ). However, questions concerning its functional significance in the interactions between tumor cells and the surrounding stroma have remained largely untouched.
Due to the elimination of VEGF signaling as a critical survival factor for endothelial cells of immature, pericyte-free small vessels (16), DC101 treatment causes vessel regression with retraction of preformed blood vessels, particularly visible in the stromal projections within the tumor parenchyma. Surprisingly, the fingerlike stromal projections still persist after vessel retraction as indicated by staining for tenascin C, a major and supposedly stable component of the tumor-stroma ECM (34, 45) . The retraction of vessels was first visible 2 days after the beginning of DC101 treatment and further increased up to 1 week as shown by both magnetic resonance imaging of A-5RT3 s.c. heterotransplants (35) and immunohistochemistry in surface transplants. 5 A similar selective ablation of nascent blood vessels in human and experimental tumors following VEGF withdrawal (16) as well as regression of coopted vessels by blockade of VEGF signaling (42) has been reported.
This vessel regression and the associated blockade of formation of new vessels was enforced by the down-regulation of expression of both VEGFR-1 and VEGFR-2 starting 1 week after DC101 treatment (i.e., in 3-week-old transplants) and resulting in a complete loss of signal between weeks 2 and 3 of treatment. Because ligand interaction is considered a major regulator of VEGFR-2 expression (46), blocking this interaction may have been causal for VEGFR-2 down-regulation as shown recently (47). This would, however, not explain why VEGFR-1 is also down-regulated, because DC101 specifically blocks ligand binding to VEGFR-2 (28) and VEGF expression in tumor cells is ongoing at a high level (48). Down-regulation of both VEGFRs coincident with the enhanced association of vessels with α-SMA-positive perivascular cells, which we considered to represent pericytes, can be understood as hallmarks of vessel maturation (49). Although staining with a α-SMA-specific antibody is not a consistent and exclusive criterion for pericytes, other antibodies, such as NG2 and platelet-derived growth factor receptor-β, did not provide an exclusively labeling either, as these proteins are also expressed by activated fibroblasts in our hands (data not shown). Despite these difficulties, we could clearly show a massive increase in the number of vessels associated with α-SMA-positive cells (presumably pericytes) in DC101-treated transplants. The interpretation of vessel normalization is further supported by the substantial change of vessel ultrastructure from typical tumor vessels with dilated, thin, and interrupted endothelial lining to mature pericyte-associated and basement membrane–surrounded capillaries embedded in a collagen fiber–rich ECM. Whether this normalization of vessels was induced by withdrawal of VEGF signaling exclusively or additionally mediated indirectly in the context of stromal maturation (i.e., by down-regulated protease activity) awaits further analysis. Further data on normalization of tumor vessels as a result of blockade of VEGF signaling have been reported quite recently during the review process of this article, demonstrating reduced vascular permeability (50) and reversion of vascular fenestration (51).
Stromal Maturation by Reduced MMP Expression and Extracellular Matrix Turnover. The most remarkable consequences of DC101 treatment were the changes in the tumor-stroma border zone. Here, a massive accumulation of collagen bundles on the stromal side and a regular structure of the basal pole of tumor cells were prominent features resulting in a smooth and straight tumor surface, which strongly contrasted the irregular border zone in control transplants. This was second only to the reappearance of long stretches of a well-structured basement membrane typically anchored to the basal cell pole on one side through hemidesmosomes and by anchoring fibrils to the collagen fibers on the other. This surprising feature of a normalized epithelial stromal border zone formed by a poorly differentiating and metastasizing carcinoma cell line (29) was similarly seen in DC101-treated s.c. growing A-5RT3 tumors (data not shown). Although immunohistologic staining of malignant tumor-stroma borders with antibodies specific to basement membrane components, such as collagen type IV or laminin, is observed, although often as interrupted labeling in well-differentiated carcinomas (21, 22, 36) , a structured basement membrane with hemidesmosomes has not been reported to date. Formation of a structured basement membrane is a diagnostic feature of premalignant tumor lesions and classically seen in benign tumor stages. In addition, considering the inverse correlation observed between the amount of retained peritumoral basement membrane and the degree of tumor aggressiveness (52), reformation of basement membrane structures is a clear indication for a phenotypic shift of the invasive malignant transplants to a premalignant, well-delineated tumor phenotype following VEGFR-2 blockade.
Whether these characteristics of stromal maturation are due to altered expression levels of ECM components or their reduced turnover as consequence of lowered protease activity requires further detailed studies (53). However, our data indicate that altered protease expression may play an important role. One of the matrix-degrading proteases most frequently associated with malignant tumors is MMP-9 (gelatinase B), which is known to cleave components of the ECM, particularly of the basement membrane (54, 55) , thereby facilitating endothelial cell migration as well as tumor cell invasion (56, 57) . Blockade of VEGFR-2 by a tyrosine kinase inhibitor reduced MMP-2 and MMP-3 secretion in endothelial cells and inhibited their migration (58). In MMP-9-deficient mice, reduced angiogenesis, tumor progression, and metastasis has been reported highlighting the role of this protease for angiogenesis and cancer (40, 41, 59) . However, the molecular mechanisms by which MMPs promote tumor invasion and angiogenesis are still poorly understood (60). More recent studies suggest that ECM proteolysis contributes to angiogenesis by exposing cryptic regulatory elements within ECM components (61, 62) and by release of ECM-sequestered angiogenesis factors (41).
In our study, localization of MMP-9 protein in the stroma of malignant transplants revealed a rather diffuse perivascular staining in the stromal strands and predominantly localized to neutrophil granulocytes as seen by counterstaining. 6 Neutrophil granulocytes are known to store considerable amounts of MMP-9 in their granules (63) and sequestration of proMMP-9 as well as other MMPs to ECM components has been shown (64). Following DC101 treatment, there was some decrease in protein staining in the stromal strands in the tumor tissue, but staining was not significantly reduced in the underlying stroma where granulocytes were still abundant (data not shown). 6 However, when RNA expression was analyzed, MMP-9 expression was restricted to distinct spots in intratumoral stromal strands, in close association to vessels and tumor cells. Whether the expressing cells were endothelial cells, pericytes (65), or other stromal cells awaits further investigations. In any case, this expression was drastically reduced in DC101-treated transplants with down-regulation of endothelial activation by blocking VEGF signaling (66). We hypothesize that this reduced expression of MMP-9 was responsible for reduced degradation of basement membrane components, thus allowing their accumulation and structural reorganization. In particular, laminin is a critical determinant of morphogenesis and differentiation and directs tissue-specific gene expression in tissue-type in vitro models (67). Its reduced turnover due to MMP-9 down-regulation may contribute not only to basement membrane reconstitution but also to normalization of epithelial polarity and thus to the reverted tumor phenotype.
Comparably, the reduced expression of murine MMP-13 in the tumor-adjacent stroma indicated decreased turnover of its substrate collagen type I, which supposedly resulted in the accumulation of bundles of collagen fibers in the tumor-neighboring stroma. Although the MMP-13 synthesizing cell type has not yet been identified, the localization of RNA signals indicates that perivascular cells, most probably fibroblasts, are promising candidates. Their localization close to endothelial cells and tumor cells suggests that MMP-13 expression is induced by paracrine signals from both neighboring cells, a regulatory mechanism that may well extend to MMP-9. Blockade of endothelial cell activation obviously resulted in down-regulation of both MMPs, suggesting a major paracrine role of endothelial cell–derived factors to control these stromal MMPs.
This rather complex process of stromal maturation into a dense fibrotic tissue together with the reformation of a structural basement membrane obviously resulted in phenotypic reversion of the invasive malignant tumor into a premalignant, noninvasive tissue. This is in line with a rapidly increasing body of evidence from several groups demonstrating that the microenvironment has crucial regulatory functions on tumor phenotype (2–5) . We and others have thus shown with different tumor cell types using in vitro and in vivo model systems that signals from the microenvironment can enhance or revert the malignant phenotype without affecting the abnormal genotype of tumor cells. These results further document the regulatory potential of the stroma on the tumor phenotype and suggest promising new tools for tumor therapy.
Grant support: Deutsche Forschungsgemeinschaft (Special Research Project “Angiogenesis,” FU 91-5/2), European Union QLK-CT-2002-02136 (N.E. Fusenig), and ImClone Systems.
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 G. Murphy for the kind donation of antibodies and murine cDNA probes for MMP-9 and MMP-13.
↵4 W. Lederle et al., unpublished data.
↵5 D. Miller et al., unpublished data.
↵6 S. Vosseler et al., unpublished data.
↵7 M. Koci et al., unpublished data.
- Received December 23, 2003.
- Revision received December 8, 2004.
- Accepted December 15, 2004.
- ©2005 American Association for Cancer Research.