Canstatin, a Novel Matrix-derived Inhibitor of Angiogenesis and Tumor Growth*

We isolated and identified an endogenous 24-kDa human basement membrane-derived inhibitor of angiogenesis and tumor growth, termed canstatin. Canstatin, a fragment of the α2 chain of type IV collagen, was produced as a recombinant molecule inEscherichia coli and 293 embryonic kidneys cells. Canstatin significantly inhibited human endothelial cell migration and murine endothelial cell tube formation. Additionally, canstatin potently inhibited 10% fetal bovine serum-stimulated endothelial cell proliferation and induced apoptosis, with no inhibition of proliferation or apoptosis observed on non-endothelial cells. Inhibition of endothelial proliferation was not concomitant with a change in extracellular signal-regulated kinase activation. We demonstrate that apoptosis induced by canstatin was associated with a down-regulation of the anti-apoptotic protein, FLIP. Canstatin also suppressed in vivo growth of large and small size tumors in two human xenograft mouse models with histology revealing decreased CD31-positive vasculature. Collectively, these results suggest that canstatin is a powerful therapeutic molecule for suppressing angiogenesis.

Vascular basement membranes are composed of macromolecules such as type IV collagen, laminin, heparan sulfate proteoglycans, fibronectin, and entactin (1). Type IV collagen is composed of six distinct gene products, namely ␣1-␣6 (2). The ␣1 and ␣2 isoforms are ubiquitously present in human basement membranes (3), whereas the other four isoforms exhibit restricted distributions (4). Type IV collagen promotes cell adhesion, migration, differentiation, and growth (3) and via these functions may play a crucial role in angiogenesis, the process of formation of new blood vessels from pre-existing capillaries (5,6). In the adult, new blood vessels arise via angiogenesis, a process critical for normal physiological events such as wound repair and endometrium remodeling (7). It is now well documented that angiogenesis is required for metastasis and growth of solid tumors beyond a few mm 3 in size (6,8). Several studies have shown that inhibitors of collagen metabolism have anti-angiogenic properties, supporting the notion that basement membrane collagen synthesis and deposition is crucial for blood vessel formation and survival (9,10). Additionally, basement membrane organization is dependent on the assembly of a type IV collagen network, which is speculated to occur via the C-terminal globular non-collagenous (NC1) domain of type IV collagen (1,11,27,28). Based on these observations, we postulated that the NC1 domain of collagen type IV could disrupt tumor angiogenesis and thereby inhibit tumor growth. Our laboratory evaluated the anti-angiogenic potential of all six NC1 domains of type IV collagen, and in the present study, we demonstrate the ability of the NC1 domain of the ␣2 chain of type IV collagen, which we have named canstatin, to inhibit endothelial cell tube formation, migration, and proliferation and to induce apoptosis in vitro, as well as suppress tumor growth in vivo.

Recombinant Production of Canstatin in Escherichia coli-
The sequence encoding canstatin was amplified by PCR 1 from the ␣2 NCI (IV)/pDS vector (12) using a forward primer (5Ј-CGGGATCCTGTCAG-CATCGGCTACCTC-3Ј) and a reverse primer (5Ј-CCCAAGCTTCAG-GTTCTTC ATGCACAC-3Ј). The resulting cDNA fragment was digested with BamHI and HindIII and ligated into predigested pET22b(ϩ) (Novagen, Madison, WI). This placed canstatin downstream of and inframe with the pelB leader sequence, allowing for periplasmic localization and expression of soluble protein. Additional vector sequences encode the amino acids MDIGINSD. The 3Ј end of the sequence was ligated in-frame with the polyhistidine tag sequence. Additional vector sequence between the 3Ј end of the cDNA and the His tag encoded the amino acids KLAAALE. Positive clones were sequenced on both strands.
Plasmid constructs encoding canstatin were first transformed into E. coli HMS174 (Novagen, Madison, WI) and then transformed into BL21 for expression (Novagen, Madison, WI). An overnight bacterial culture was used to inoculate a 500-ml culture in LB medium. This culture was grown for approximately 4 h until the cells reached an A 600 of 0.6. Then, protein expression was induced by addition of isopropyl-1-thio-␤-Dgalactopyranoside to a final concentration of 0.5 mM. After a 2-h induction, cells were harvested by centrifugation at 5,000 ϫ g and lysed by resuspension in 6 M guanidine, 0.1 M NaH 2 PO 4 , 0.01 M Tris-HCl, pH 8.0. Resuspended cells were sonicated briefly and centrifuged at 12,000 ϫ g for 30 min. The supernatant fraction was passed over a 5-ml of nickelnitrilotriacetic acid-agarose column (Qiagen) 4 -6 times at a speed of 2 ml/min. Nonspecifically bound protein was removed by washing with 15 ml each of 10, 25, and 50 mM imidazole in 8 M urea, 0.1 M NaH 2 PO 4 , 0.01 M Tris-HCl, pH 8.0. Canstatin protein was eluted from the column with two concentrations of imidazole (125 and 250 mM) in 8 M urea, 0.1 M NaH 2 PO 4 , 0.01 M Tris-HCl, pH 8.0. The eluted protein was dialyzed twice against 2 liters of PBS at 4°C. A portion of the total protein precipitated during dialysis. Dialyzed protein was collected and centrifuged at approximately 3,500 ϫ g and separated into pellet and supernatant fractions. Protein concentration in each fraction was determined by the BCA assay (Pierce) and quantitative (densitometric) SDS-PAGE analysis. The fraction of total protein in the pellet was approximately 40%, with the remaining 60% recovered as a soluble protein. The total protein yield was approximately 15 mg/liter.
Expression of Canstatin in 293 Embryonic Kidney Cells-We used the pDS plasmid containing ␣2(IV)NC1 (12) to PCR-amplify canstatin in a way that it would add a leader signal sequence in-frame into the pcDNA 3.1 (Invitrogen, Carlsbad, CA) eukaryotic expression vector. The leader sequence from the 5Ј end of full-length ␣2(IV) chain was cloned 5Ј to the NC1 domain to enable protein secretion into the culture medium. The canstatin containing recombinant vectors were sequenced using flanking primers. Error-free cDNA clones were further purified and used for in vitro translation studies to confirm protein expression (data not shown). The canstatin containing plasmid and control plasmid was used to transfect 293 cells using the calcium chloride method. Transfected clones were selected by geneticin (Life Technologies Inc.) antibiotic treatment. The cells were passed for 3 weeks in the presence of the antibiotic until no cell death was evident. Clones were expanded into T-225 flasks and grown until confluent. Then, the supernatant was collected and concentrated using an Amicon concentrator (Amicon, Beverly, MA). The concentrated supernatant was analyzed by SDS-PAGE and immunoblotting and enzyme-linked immunosorbent assay for canstatin expression. Strong binding in the supernatant was detected by enzyme-linked immunosorbent assay (data not shown). Canstatin containing supernatant was subjected to affinity chromatography using canstatin-specific antibodies (13). A major peak was identified, containing a monomer of about 24 kDa that was immunoreactive with canstatin antibodies.
Inhibition of Endothelial Cell Proliferation-Bovine calf pulmonary aortic endothelial (CPAE) cells were grown to confluence in DMEM with 10% fetal bovine serum (FBS) and kept confluent for 48 h. Cells were harvested by trypsinization (Life Technologies, Inc.) at 37°C for 5 min. A suspension of 12,500 cells in DMEM with 0.5% FCS was added to each well of a 24-well plate coated with 10 g/ml fibronectin. The cells were incubated for 24 h at 37°C with 5% CO 2 and 95% humidity. Medium was removed and replaced with DMEM containing 0.5% FCS (unstimulated) or 10% FCS (stimulated and treated cells). 786-0, PC-3, and HEK 293 cells, used as controls, were also grown to confluency, trypsinized, and plated in the same manner. Cells were treated with concentrations of canstatin ranging from 0.025 to 40 g/ml in triplicate. In thymidine incorporation experiments all wells received 1 Ci of [ 3 H]thymidine at the time of treatment. After 24 h, medium was removed, and the wells were washed 3 times with PBS. Radioactivity was extracted with 1 N NaOH and added to a scintillation vial containing 4 ml of ScintiVerse II (Fisher) solution. Thymidine incorporation was measured using a scintillation counter. In the methylene blue staining method, 3,100 cells were added to each well of a 96-well plate and treated as above, and then cells were counted using the method of Oliver et al. (14). All wells were washed one time with 100 l of PBS, and the cells were fixed by adding 100 l of 10% formalin in neutralbuffered saline (Sigma) for 30 min at room temperature. After formalin removal, cells were stained with a solution of 1% methylene blue (Sigma) in 0.01 M borate buffer, pH 8.5, for 30 min at room temperature. After removal of staining solution, the wells were washed 5 times with 100 l of 0.01 M borate buffer, pH 8.5,. Methylene blue was extracted from the cells with 100 l of 0.1 N HCl, ethanol (1:1 mixture) for 1 h at room temperature. The amount of methylene blue staining was measured on a microplate reader (Bio-Rad) using light absorbance at 655 nm wavelength.
Endothelial Tube Assay-Matrigel (Collaborative Biomolecules, Bedford, MA) was added (320 l) to each well of a 24-well plate and allowed to polymerize (15). A suspension of 25,000 mouse aortic endothelial cells (MAE) in EGM-2 (Clonetics, Walkersville, MD) medium without antibiotic was passed into each well coated with matrigel (15). The cells were treated with either canstatin, BSA, sterile PBS, or ␣5-NC1 domain in increasing concentrations. All assays were performed in triplicate. Cells were incubated for 24 -48 h at 37°C and viewed using a CK2 Olympus microscope (3.3 ocular, 10ϫ objective). The cells were then photographed using 400 DK-coated TMAX film (Eastman Kodak Co.). Cells were stained with diff-quik fixative (Sigma) and photographed again (15). Ten fields were viewed, and tubes were counted and averaged.
Migration Assay-The inhibitory effect of canstatin on VEGF-induced chemotaxis was tested on human umbilical endothelial cells (HUVECs) using the Boyden chamber assay (Neuroprobe, MD). HUVECs were grown in M199 (Life Technologies Inc.) containing 10% FBS and 5 ng/ml DiIC18(3) living fluorescent stain (Molecular Probes, Eugene, OR) overnight. After trypsinizing, washing, and diluting cells in M199 containing 0.5% FBS, 60,000 cells were seeded on the upper chamber wells, together with or without canstatin (0.01 or 1 g/ml). M199 medium containing 1% FBS plus 10 ng/ml VEGF (R & D Systems, Minneapolis, MN) was placed in the lower chamber as a chemotactant. The cell-containing compartments were separated from the chemotactant with polycarbonate filters (Poretics, CA) of 8-m pore size. The chamber was incubated at 37°C with 5% CO 2 and 95% humidity for 4.5 h. After discarding the non-migrated cells and washing the upper wells with PBS, the filters were scraped with a plastic blade, fixed in 4% formaldehyde in PBS, and placed on a glass slide. By using a fluorescent high power field, several independent homogenous images were recorded by a digital SenSys camera operated with Image Processing Software PMIS (Photometrics, CO). Cells were counted by employing the OPTIMIZE 6.0 software program (BioScan, WA) (16).
ERK Phosphorylation Assay-HUVEC cells were cultured overnight in McCoy's medium supplemented with 20% FBS, 1% penicillin/streptomycin, 100 g/ml heparin, and 50 g/ml endothelial mitogen (Biomedical Technologies). The following day cells were washed and grown for 4 h in low serum medium (McCoy's medium supplemented with 1% penicillin/streptomycin, 100 g/ml heparin, and 5% FBS). After 4 h, the medium was replaced with fresh low serum medium with or without 20 g/ml canstatin. One hour later the serum concentration was adjusted to 20%, and endothelial mitogen was added to 50 g/ml. At 0, 5, 10, 25, and 40 min the cells were washed with PBS and lysed with passive lysis mix (Promega, Madison WI) plus leupeptin, phenylmethylsulfonyl fluoride, NaF, Na 3 VO 4 , ␤-glycerophosphate, and sodium pyrophosphate. Lysates were quantified for protein concentration and separated on 12% SDS-PAGE gels. Western blots of serum or serum ϩ canstatintreated HUVEC lysates were performed to detect levels of phospho-ERK using anti-phospho-ERK antibodies (New England Biolabs, Beverly, MA). Total ERK protein was detected on the same nitrocellulose membrane with anti-ERK antibody (Santa Cruz Biotechnology, Santa Cruz, CA) illustrating equal amounts of total ERK protein.
FLIP Protein Level Quantification-HUVEC cells were cultured overnight in McCoy's medium supplemented with 20% FBS, 1% penicillin/streptomycin, 100 g/ml heparin, and 50 g/ml endothelial mitogen (Biomedical Technologies). The following day, growth factor-containing medium was removed and replaced with medium containing 20% serum alone (with or without 20 g/ml canstatin) or medium with 20% serum and growth factors (with or without 20 g/ml canstatin). All wells received polymyxin B to eliminate potential effects of endotoxin. Cells were lysed at 30, 60, and 180 min with passive lysis mix (Promega, Madison, WI) plus leupeptin, phenylmethylsulfonyl fluoride, NaF, Na 3 VO 4 , ␤-glycerophosphate, and sodium pyrophosphate, and the lysates were analyzed for FLIP levels by Western blot with anti-FLIP antibody (17).
Annexin V-FITC Binding-0.5 ϫ 10 6 CPAE cells were added to each well of a 6-well tissue culture plate in 10% FBS-supplemented DMEM (BioWhittaker, Walkersville, MD) overnight. The next day the fresh medium was added to all wells together with 15 g/ml canstatin or 40 ng/ml tumor necrosis factor-␣. Control cells received an equal volume of PBS. After 24 h of treatment medium containing detached cells was collected, and attached cells were trypsinized and added together with detached cells and centrifuged at 3,000 ϫ g. Cells were then washed and phosphatidylserine externalization (an early apoptotic indicator) was measured by labeling with FITC-labeled annexin V (CLONTECH, Palo Alto, CA) according to manufacturer's instructions. Annexin-FITC-labeled cells were counted using a Becton-Dickenson FACScan Plus flow cytometer. For each treatment 15,000 cells were counted and stored in listmode. The data were then analyzed using standard Cell Quest software.
In Vivo Tumor Studies-Human prostate adenocarcinoma cells (PC-3) were harvested from culture, and 2 ϫ 10 6 cells in sterile PBS were injected subcutaneously into 7-9-week-old male SCID mice. The tumors grew for approximately 4 weeks after which animals were divided into groups of 4 mice. Experimental groups were injected daily intraperitoneally with canstatin at a dosage of 10 mg/kg in a total volume of 0.1 ml of PBS. The control group received equal volumes of PBS each day. In a second PC-3 experiment, PC-3 cells were harvested from culture, and 3 ϫ 10 6 cells were injected into 6 -7-week-old athymic nude mice, and tumors grew for approximately 2 weeks after which animals were divided into groups of 4 mice. Experimental groups were injected daily intraperitoneally with canstatin at a dosage of 3 mg/kg in a total volume of 0.2 ml of PBS or recombinant mouse endostatin at a dosage of 8 mg/kg in the same volume of PBS. The control group received equal volumes of PBS each day. Tumor length and width were measured using a Vernier caliper, and the tumor volume was calculated using the standard formula length ϫ width squared ϫ 0.52 (18). Volume Ϯ S.E. is plotted over the treatment period. For the renal cell carcinoma cell model, 2 ϫ 10 6 786-0 cells were injected subcutaneously into 7-9-week-old male athymic nude mice. The tumors were allowed to grow to ϳ700 or ϳ100 mm 3 . Each group contained 6 mice. Canstatin in sterile PBS was injected intraperitoneally daily (10 mg/kg) for 10 days. The control group received the same volume of PBS.
CD31 Immunohistochemistry-At the end of treatment, mice were sacrificed, and the tumors were excised. The removed tumors were dissected with a scalpel into several pieces approximately 3-4 mm thick and then fixed in 4% paraformaldehyde for 24 h. Tissues were then switched to PBS for 24 h before dehydration and paraffin embedding. After embedding in paraffin, 3-m tissue sections were cut and mounted. Sections were deparaffinized, rehydrated, and pretreated with 300 g/ml protease XXIV (Sigma) at 37°C for 5 min. Digestion was stopped in 100% ethanol. Sections were air-dried, rehydrated, and blocked with 10% rabbit serum. Next, slides were incubated at 4°C overnight with a 1:50 dilution of rat anti-mouse CD31 monoclonal antibody (PharMingen), followed by two successive incubations at 37°C for 30 min with 1:50 dilutions of rabbit anti-rat immunoglobulin (Dako, Carpinteria, CA) and rat APAAP (Dako). The color reaction was performed with new fuchsin. Sections were counterstained with hematoxylin.

RESULTS AND DISCUSSION
Human canstatin was produced in E. coli as a fusion protein with a C-terminal six histidine tag using pET-22b(ϩ), a bacterial expression plasmid, and also produced as a secreted soluble protein in 293 embryonic kidney cells using the pcDNA 3.1 eukaryotic vector. The E. coli-expressed protein was isolated predominantly as a soluble protein, and SDS-PAGE analysis revealed a monomeric band at 27 and 3 kDa that arises from polylinker and histidine tag sequences. Uncleaved signal peptide may account for additional molecular weight seen in some preparations (Fig. 1, A and B). Canstatin protein was analyzed by SDS-polyacrylamide gel electrophoresis (PAGE) and detected by Western blotting with an anti-polyhistidine tag antibody (Fig. 1, C and D). Canstatin antibodies also detected bacterially expressed recombinant canstatin protein (data not shown). The 293 cell-produced recombinant protein (without any purification or detection tags) was isolated using affinity chromatography, and a pure monomeric form was detected in the major peak by SDS-PAGE and immunoblot analyses (Fig.  1, E and F). In addition, human canstatin was isolated from human placenta by gel filtration, high pressure liquid chromatography, and affinity chromatography techniques; a 24-kDa molecule was detected by SDS-PAGE and immunoblot analyses (data not shown).
We tested the ability of canstatin to disrupt endothelial cell tube formation in matrigel, a solid gel of mouse basement membrane proteins derived from sarcomas. When mouse aortic endothelial cells are cultured on matrigel, they rapidly align and form hollow tube-like structures (15). Canstatin, produced in 293 cells, selectively inhibited endothelial tube formation in a dose-dependent manner, with a near-complete inhibition of tube formation seen with the addition of 1 g of canstatin protein (Fig. 2, A and C). Neither a control protein, bovine serum albumin (BSA), nor the NC1 domain of type IV collagen ␣5 chain had an effect on endothelial tube formation (Fig. 2, A  and D).
During the process of angiogenesis, endothelial cells proliferate and migrate to form new blood vessels. Therefore, we assessed the effect of canstatin on endothelial cell migration. HUVECs were cultured in a Boyden chamber and stimulated to migrate toward a 10 ng/ml VEGF gradient (16). Canstatin, produced in 293 cells, inhibited the migration of HUVECs with a significant effect observed at 10 ng/ml (Fig. 2, E and F). The ability of canstatin to inhibit both proliferation and migration of endothelial cells may indicate that it works at more than one step in the process of angiogenesis or that the effect is common to both migratory and proliferative processes.
In thymidine incorporation assays of endothelial cell proliferation, using E. coli produced soluble protein, a dose-dependent inhibition of 10% serum-stimulated CPAE cells was detected with an ED 50 value of 0.5 g/ml (Fig. 3A). Furthermore, we found that HUVECs were also inhibited by canstatin with a similar ED 50 value (Fig. 3B). These results support earlier observations that ␣1 and ␣2 type IV collagen chains isolated from the Engelbreth-Holm-Swarm sarcoma tumor may be inhibitory to capillary endothelial cell proliferation (19). No significant effect was observed on the proliferation of renal carcinoma cells (786-0), prostate cancer cells (PC-3), or human embryonic kidney cells (HEK 293) at canstatin doses up to 40 g/ml (Fig. 3C). Methylene blue colorimetric proliferation assay confirmed the results of the thymidine assay for the CPAE, PC-3, and HEK 293 cells (data not shown).
FIG. 2. Endothelial tube and migration assays. Mouse aortic endothelial cells (MAE) (25,000 per well) in EGM-2 medium without antibiotic were passed into Matrigel-coated wells (15). The cells were treated with either canstatin, BSA, sterile PBS, or ␣5-NC1 domain in increasing concentrations for 48 h at 37°C and viewed using a CK2 Olympus microscope and then photographed. Cells were stained with diff-quik fixative, and 10 fields were viewed, tubes counted, and averaged. A, average number of tubes counted per field represented as a percentage of the PBS-treated wells. Bars represent the S.E. B, representative PBS control-treated MAE cells forming tubes (ϫ 40 magnification). C, AE cells treated with ␣5-NC1 domain (ϫ 40 magnification). D, canstatin-treated (0.4 g/ml) MAE cells (ϫ 40 magnification). E and F, migration assay. The inhibitory effect of canstatin on VEGF-induced chemotaxis was tested on HUVECs using a Boyden chamber assay as described under "Materials and Methods." HUVECs (60,000 cells) stained with DilC18(3) living fluorescent stain overnight were passaged into the upper chambers in M199 containing 0.5% FBS, with or without canstatin (0.01 or 1 g/ml). M199 medium containing 1% FBS plus 10 ng/ml VEGF was placed in the lower chamber as a chemotactant. Cells were allowed to migrate for 4.5 h through an 8-m pore size polycarbonate filter and fixed as described under "Materials and Methods." Using a fluorescent high power field, several independent homogenous images were recorded by a digital SenSys camera operated with Image Processing Software PMIS, and cells were counted by employing the OP-TIMIZE 6.0 software program (E). Representative pictures of a single high power (ϫ 40) field are shown in F. Chemotactant and inhibitor added are as follows: 1, without VEGF or serum; 2, 1% FCS and 10 ng/ml VEGF; 3, 1% FCS and 10 ng/ml VEGF treated with 0.01 g/ml canstatin; 4, 1% FCS and 10 ng/ml VEGF treated with 1 g/ml canstatin.
In order to understand further the molecular mechanisms involved in the anti-proliferative and anti-migratory activities of canstatin, we assessed the effect of canstatin on ERK activation induced by 20% fetal bovine serum and endothelial mitogens. Fig. 3D shows that ERK phosphorylation in HU-VECs was evident within 5 min after growth factor stimulation, and treatment with 20 g/ml canstatin did not alter early activation of ERK. A decrease in ERK phosphorylation was observed at later time points, a profile that is consistent with responses observed with several mitogens (20,21). These ob-  HUVEC (B), or 786-0, PC-3, and HEK 293 cells (C) were grown, passaged, and plated onto 24-well plates as described under "Materials and Methods." Cells were treated with concentrations of canstatin ranging from 0.025 to 40 g/ml for 24 h in thymidine incorporation experiments (or 48 h for HUVEC). All groups represent triplicate samples measured by a scintillation counter. Graph bars represent mean cpm Ϯ S.E. D, ERK phosphorylation. HUVECs were cultured overnight as described under "Materials and Methods," and the following day cells were washed and grown for 4 h in low serum medium. After 4 h, the medium was replaced with fresh low serum medium with or without 20 g/ml canstatin. One hour later the serum concentration was adjusted to 20%, and endothelial mitogen was added to 50 g/ml. At 0, 5, 10, 25, and 40 min (phospho-ERK) the cells were lysed and quantified for protein concentration and then separated on 12% SDS-PAGE gels. Western blot of phospho-ERK is shown in the top panel, and the bottom panel shows total ERK protein illustrating equal amounts of total ERK protein. E, annexin V-FITC binding. CPAE cells were treated for 24 h with 15 g/ml canstatin (red peak) or 40 ng/ml tumor necrosis factor-␣ (blue peak) or an equal volume of PBS (green peak). Detached cells and attached cells were collected, and phosphatidylserine externalization (an early apoptotic indicator) was measured by labeling with FITC-labeled annexin V according to manufacturer's instructions. For each treatment 15,000 cells were counted and analyzed using standard Cell Quest software. The rightward shift on the x axis of the canstatin and tumor necrosis factor-␣ peaks indicates increased annexin V-FITC binding of apoptotic cells. F, FLIP protein levels. HUVECs were grown in growth factor medium and switched (at time 0) to serum only medium or growth factor medium, with or without canstatin and harvested at 0, 30, 60, and 180 min. FLIP protein in HUVEC lysates was detected by Western blot using anti-FLIP antibody (17). Vinculin protein was also detected on the same blot as a protein loading control. G, FLIP protein levels were quantified by densitometric scanning software and normalized for protein loading using levels of vinculin and plotted as a percentage of the 0-h time points. servations indicate that canstatin does not primarily work by inhibiting proximal events activated by VEGF or basic fibroblast growth factor receptors.
In order to establish apoptosis of endothelial cells as a potential mode of action for canstatin, we used annexin V-FITC labeling of externalized phosphatidylserine to assess for apo-ptotic cells. Our results show that canstatin specifically induces apoptosis of endothelial cells (Fig. 3E) with no significant effect observed on PC-3, 786-0, or HEK 293 cell lines (data not shown). It is probable that canstatin acts by predisposing cells to apoptosis, as evidenced by a steady decrease in FLIP protein levels during treatment with canstatin in the presence of both FIG. 4. In vivo tumor studies. A and B, 786-0 cells (2 million cells) were injected subcutaneously into 7-9-week-old male athymic nude mice. The tumors were allowed to grow to ϳ100 (A) or ϳ700 mm 3 (B). Each group contained 6 mice. Canstatin in sterile PBS was injected intraperitoneally daily (10 mg/kg) for 10 days. The control group received the PBS vehicle. C, human prostate adenocarcinoma cells (PC-3) xenografts grown subcutaneously in male SCID mice were treated with daily intraperitoneal injections of canstatin at a dosage of 10 mg/kg in a total volume of 0.1 ml of PBS. The control group received equal volumes of PBS each day. Tumors were measured, and the volume was calculated as described under "Materials and Methods." Initial (day 0) tumor volumes ranged from 88 to 135 mm 3 for control and from 108 to 149 mm 3 for canstatin-treated. Each group contained 5 mice. The calculated volume on a given day was divided by the volume of treatment day 0 to give a fractional tumor volume (V/V 0 ). Fractional tumor volume Ϯ S.E. is plotted over the treatment period. D, PC-3 xenografts grown subcutaneously in male athymic nude mice were treated with daily intraperitoneal injections with 3 mg/kg canstatin in a total volume of 0.2 ml of PBS or endostatin at a dosage of 8 mg/kg in the same volume of PBS. The control group received equal volumes of PBS each day. All groups contained 4 mice. Tumors were measured and fractional tumor volumes calculated as in the previous study. Initial tumor volumes ranged from 26 to 73 mm 3 . E and F, CD31 immunohistochemistry. At the end of xenograft tumor studies, mice were sacrificed and tumors were excised, fixed in 4% paraformaldehyde, and paraffin-embedded. Pictures show 3-m tissue incubated at 4°C overnight with a 1:50 dilution of rat anti-mouse CD31 monoclonal antibody, followed by two successive incubations at 37°C for 30 min with 1:50 dilutions of rabbit anti-rat immunoglobulin and rat alkaline phosphatase anti-alkaline phosphatase. The color reaction was performed with new fuchsin, and sections were counterstained with hematoxylin. E, control (PBS-treated) PC-3 tumor; F, 10 mg/kg canstatin-treated PC-3 tumor. serum and growth factor (basic fibroblast growth factor and VEGF) stimulation (Fig. 3F). Canstatin did not effect FLIP levels in the absence of growth factors. A transient increase in FLIP levels was seen with the addition of serum only medium, probably due to the removal of growth factors. In contrast, the addition of growth factor-containing medium induced a transient decrease in FLIP levels that returned to base line or increased in the absence of canstatin. Since endothelial cells express both Fas and FasL constitutively (17), it is likely that this decrease in FLIP sensitizes endothelial cells to the apoptotic signal. Furthermore, the lack of effect of canstatin on endothelial cells in the absence of growth factors indicates it may target angiogenic and not pre-formed endothelium.
For in vivo experiments, established xenograft tumor models in mice were used to test the effectiveness of canstatin as an inhibitor of angiogenesis-dependent tumor growth. Canstatin, produced as soluble protein in E. coli, inhibited the growth of small (Fig. 4A) and large (Fig. 4B) renal cell carcinoma (786-0) tumors by 4-and 3-fold with respect to placebo-treated mice. Established human prostate (PC-3) tumors in severe combined immunodeficient (SCID) mice or athymic (nu/nu) mice exhibited fractional tumor volumes of 1.8 -2.4-fold less than placebotreated mice when treated with 10 ( Fig. 4C) or 3 mg/kg ( Fig.  4D) canstatin, respectively. This decrease in tumor size was consistent with a decrease in CD31-positive vasculature (Fig. 4, E and F). A similar suppressive effect was observed with 8 mg/kg of endostatin (Fig. 4D). Further studies will be required to determine the minimal effective doses of these proteins, but results in our laboratory indicate that a 5 mg/kg dose of endostatin is not able to suppress tumor growth (data not shown), suggesting that human canstatin may be more potent than endostatin. In all of the in vivo studies, mice appeared healthy with no signs of wasting, and none of the mice died during treatment.
The ability of canstatin to inhibit specifically endothelial cell proliferation, migration, and tube formation strongly suggests that it is an anti-angiogenic agent and that it may function via a cell surface protein/receptor. Integrins are potential candidate molecules for binding to canstatin based on their extracellular matrix binding capacity and ability to modulate cell behavior such as migration and proliferation (22). In particular, ␣ v ␤ 3 integrin is a possible canstatin receptor, due to its induction during angiogenesis and its promiscuous binding capacity (23). Also, integrins ␣ 1 ␤ 1 and ␣ 2 ␤ 1 have been implicated as cell adhesion receptors that bind to type IV collagen (24), but whether canstatin is the collagen IV domain that is mediating this interaction or any other integrin binding has yet to be determined. The recent identification of other collagen-derived inhibitors of angiogenesis, such as endostatin and restin, C-terminal fragments of collagen XVIII and XV, respectively (18,25,26), lends credence to the theory that collagen matrix is critical to angiogenesis, although no significant sequence (Ͻ14% at the amino acid level) or structural homology is evident between canstatin and these other inhibitors. Furthermore, canstatin, endostatin, and restin are non-collagenous fragments of collagen molecules underscoring the importance of this portion of the protein in mediating cell behavior (2,18,25). Future comparative studies with these inhibitors may also give insight into their unique mechanism of action.