Blood Vessel Maturation and Response to Vascular-Disrupting Therapy in Single Vascular Endothelial Growth Factor-A Isoform–Producing Tumors Free

Disodium combretastatin-A4 3-O-phosphate (CA-4-P) is the lead compound of a group of tubulin-binding agents that act as vascular-disrupting agents (VDA) in solid tumors (1, 2). CA-4-P is active in human cancers (3) and is currently in phase II clinical trials in combination with conventional treatments (4).⁴

Despite the progress of VDAs to clinical trials, and some evidence for improved patient survival upon adding VDA treatment to conventional chemotherapy (6), the reasons for the susceptibility of the tumor vasculature to VDAs remain unclear. Tumor blood vessels are generally considered to be immature. Thus, genetic modification of tumor cells to modulate expression of key molecules involved in vascular maturation provides a potential route for investigating the factors that influence the response of the tumor vasculature to VDAs.

Vascular endothelial growth factor A (VEGF-A or simply VEGF) is a key stimulator of tumorigenic angiogenesis (7) and acts as a mitogen, chemoattractant, survival factor, vasodilator, and permeability factor (8). Human and mouse VEGF mRNA is transcribed from eight exons and alternatively spliced to give rise to a number of isoforms of the protein product (9), the most prevalent consisting of 121, 165, and 189 amino acids in the human and 120, 164, and 188 in the mouse (Fig. 1A). VEGF120/121 lacks the heparin-binding site and is readily diffusible, whereas VEGF188/189 is tightly bound to proteoglycans in the extracellular matrix or on the cell surface and VEGF164/165 has intermediate properties. These isoforms are found in most normal tissues and have affinity for the VEGF receptors, FLT1 (VEGFR1) and KDR (VEGFR2; refs. 10, 11). VEGF121 and VEGF165 are the most prevalent forms in human cancers (12, 13).

Using Cre/Lox technology, mice expressing only single isoforms of VEGF, known as Vegfa^120/120, Vegfa^164/164, and Vegfa^188/188 mice, have been developed (14, 15). Their analysis revealed different roles for the individual VEGF isoforms in vascular growth and patterning during development. In particular, VEGF164 was sufficient for developmental angiogenesis and neonatal growth, and VEGF120 supported endothelial cell proliferation and vasculogenesis; however, angiogenic vessel branching was impaired and vessels were hemorrhagic (16). VEGF188 was essential for the formation of a normal cardiovasculature in 50% of mice (14, 15).

In the current study, we used the single VEGF isoform-expressing mice and the counterpart wild-type mice to develop mouse fibrosarcoma cell lines from embryo fibroblasts. We reasoned that the different VEGF isoforms would give rise to tumors with very different vascular characteristics that could be used to investigate susceptibility to CA-4-P. The method of producing the tumor lines was substantially different from previous studies in which overexpression of individual VEGF isoforms was used (13, 17), ensuring that VEGF gene transcription was under the control of the endogenous VEGF promoter and thus susceptible to environmental control. We gained new insights into the differential effects of VEGF isoforms in tumors by using intravital microscopy and tumor uptake of i.v. administered markers to measure functional vascular end points, in addition to measuring vascular morphology and tumor growth, with and without CA-4-P.

Animal experiments were conducted in accordance with the United Kingdom Animals (Scientific Procedures) Act 1986 and with local ethical approval. CA-4-P was kindly provided by Professor G.R. Pettit (Arizona State University, Tempe, AZ).

Tumor cell lines. Primary mouse embryo fibroblasts expressing single isoforms of VEGF (120, 164, or 188) only or all isoforms (Fig. 1A) were isolated from 13.5 dpc embryos produced by heterozygous breeding pairs of single VEGF isoform-expressing mice on a Swiss background. Fibroblast cultures were genotyped, as described (18), to identify wild-type samples and those homozygous for the Vegfa¹²⁰, Vegfa¹⁶⁴, or Vegfa¹⁸⁸ allele. Fibroblasts were immortalized and oncogenically transformed by retroviral transduction with SV40 and HRAS (h-ras; refs. 19, 20). The resulting fibrosarcoma cell lines were maintained in high glucose DMEM (Invitrogen) medium containing l-glutamine, FCS, and the antibiotics G-418 and puromycin.

Apoptosis. Apoptotic cell death in response to CA-4-P was evaluated using the cell death detection ELISA^plus kit (Roche Diagnostics), as previously described (21). Cells were plated at 10⁴ cells/well in 24-well plates, allowed to adhere for 24 h and then exposed to CA-4-P overnight before analysis.

Subcutaneous tumor transplantation. Fibrosarcoma cells expressing only VEGF120, VEGF164, or VEGF188 tumor cells or all three isoforms (control tumor cells) were injected s.c. (1 × 10⁶ cells in 0.05 mL) into the rear dorsum of female severe combined immunodeficiency (SCID) mice (8–12 weeks old, 20–25 g).

VEGF-A mRNA and protein analysis. Excised tumors (6–8 mm diameter) were collected into RNAlater (Ambion). RNA and protein were isolated from tumor samples using the Mirvana PARIS kit (Ambion) and from cell lines in culture using the Cells-to-cDNA11 kit, according to the manufacturer's instructions. VEGF isoforms were amplified using the following PCR primers (forward, 5′CAGGCTGCTGTAACGATGAA3′; and reverse, 5′CTTTCCGGTGAGAGGTCTGG3′). Approximately 20% of each PCR reaction, together with appropriate controls, was then run on 2% agarose gels containing 1 μg/mL of ethidium bromide and products visualized under UV illumination.

For quantification of VEGF protein in culture supernatants, 2 × 10⁶ cells were plated in T-175 flasks, and at 24 h, the medium was replaced with 20 mL of fresh medium. Cells were incubated for a further 48 h and then treated with 0.3 mmol/L of suramin for 3 h to release surface matrix-bound VEGF where medium was collected. VEGF was quantified in the cell medium and tumor extracts using the Quantikine Immunoassay mouse VEGF ELISA kit (R&D systems), according to the manufacturer's instructions.

An antibody recognizing all VEGF-A isoforms (p20, Santa Cruz Biotechnology) was used for Western blotting analysis. Cells were plated as above, but at 24 h, the medium was replaced with medium without serum. Conditioned media was collected 48 h later and concentrated 50-fold (Amicon). Equal amounts of proteins were separated on NuPAGE Novex gels (Invitrogen) and transferred to nitrocellulose membranes. Immunoreactive bands were detected by enhanced chemiluminescence.

Subcutaneous tumor growth. Subcutaneous tumors were measured using calipers and tumor volume (V) was calculated from V = 0.52 × d1 × d2 × d3, where d1, d2, and d3 are the three orthogonal tumor diameters. Untreated tumor growth curves were fitted to “Gompertz” curves [V = exp (a − [b × (exp [−c × (t − d)])])], where a, b, c, and d are fitted variables, and t is time after transplant.

Tumors were treated when they reached 5 to 6 mm in diameter. CA-4-P (50 mg/kg i.p., 10 mL/kg in saline) or saline alone, was given once a day for 10 days, with a 2-day gap between the fifth and sixth doses.

Tumor histology and immunohistochemistry. Subcutaneous tumors (6–8 mm diameter) were formalin- or zinc-fixed, paraffin-embedded, and stained to identify endothelial cells (rat anti-mouse CD31 monoclonal antibody; BD PharMingen, Int.) and pericytes [mouse anti-mouse α-smooth muscle actin (α-sma) monoclonal antibody, Sigma-Aldrich; or rabbit anti-mouse desmin polyclonal antibody, Abcam; or rabbit anti-mouse angiopoietin-1 (ANG1) polyclonal antibody, Chemicon]. Antibody binding was visualized using 3,3′-diaminobenzidine (DakoCytomation) and signal amplification was achieved via the avidin-biotin complex/horseradish peroxidase system (CD31, α-sma, ANG1) or the EnVision system (desmin). Sections were counterstained with hematoxylin.

Mounted sections stained for CD31 and α-sma were quantified according to a random points scoring system (22). One hundred and twenty high-power regions of interest were counted per tumor to give vascular volume as a percentage of tumor volume for CD31 staining. The percentage of necrosis was measured from H&E sections by the same method or by delineation of necrotic regions in MATLAB (The Mathworks, Inc.). Desmin and ANG1 staining were assessed qualitatively.

Double staining immunofluorescence employed the CD31 monoclonal antibody described above and a FITC-conjugated monoclonal anti-mouse α-sma antibody (Sigma-Aldrich). Sections were incubated with a goat anti-rat Alexa-555 Red (Invitrogen) antibody. Sections were mounted in 4′,6-diamidino-2-phenylindole Vectashield (Vector Laboratories).

Window chamber surgery and tumor implantation. SCID mice (12–16 weeks old, 28–32 g) were anesthetized using fentanyl-fluanisone and midazolam i.p., as described previously (23). Briefly, an aluminum window chamber (total weight, ∼2 g), holding two parallel glass windows, was implanted into a dorsal skin flap. A tumor fragment (∼0.5 mm in diameter) from a donor animal was implanted onto the exposed panniculus muscle before closing the chamber, allowing a depth of ∼200 μm for tumor growth. Animals were given a s.c. injection of dextrose saline (1 mL) and an i.p. injection of buprenorphine (0.1 mL, Vetergesic) to aid recovery and then kept in a warm room (28–30°C), until the day of the experiment.

Intravital microscopy. Donor RBC, acquired via cardiac puncture from donor mice were labeled with the fluorescent membrane dye, DiI (Invitrogen), for the measurement of RBC velocity in the tumor vasculature, as described previously (24–26). Tumors, at 3 to 5 mm in diameter, were used for treatment ∼6 to 10 days after surgery.

Treated animals received either CA-4-P (30 mg/kg i.v. at a concentration of 3 mg/mL in 0.9% NaCl) or the same volume of 0.9% NaCl. Transmitted light images and video sequences (25 frames/s) were captured at various magnifications and time points up to 24 h after treatment. For the assessment of tumor vascularization in untreated tumors, imaging was performed once per day for several days after tumor transplantation.

Average vessel length, total vascular length, and fractal dimensions were acquired from transmitted light images using in-house–developed software, as described previously (23, 27). A single vessel was defined as a vascular length with no visible branches. RBC velocities were calculated from ×20 objective video sequences, using epifluorescence illumination, as described previously (26).

Macromolecular vascular leakage. FITC-labeled dextran (FITC-dextran; 40 kDa, 0.013 mol FITC/mol dextran; Sigma-Aldrich) was used as a macromolecular marker for determining tumor blood vessel barrier function. Following extensive washing to remove low-molecular weight contaminants, FITC-dextran (20 mg/mL) was made up in PBS and administered i.v. to awake, restrained mice at 0.05 mL per mouse. Multiphoton fluorescence microscopy, based on a modified Bio-Rad MRC 1024MP workstation, was used for imaging, as described previously (23). Images were captured every 4 min, at a working excitation wavelength of 890 nm. Emission wavelength was 530 to 540 nm. The three-dimensional data consisted of 512 × 512 × 11 voxels each at a volume of 2.6 × 2.6 × 4.5 μm³.

Acquired images were processed, as described previously (28). Changes in image fluorescence intensity over time were used as an index of tumor vascular leakage of FITC-dextran. In addition, images from the first time point were analyzed for “contiguity” to provide a quantitative measure of vascular features in the different tumor lines. First, three-dimensional images were segmented into intravascular and extravascular classes, based on image intensities, as described previously (28). Then, each voxel within the intravascular class was interrogated to determine the fraction of neighboring voxels that were designated as being in the same class (29). The average of this fraction, derived from analysis of all the intravascular voxels from a particular tumor image, was defined as the contiguity of that tumor's vasculature. A highly contiguous vasculature suggests blood vessels that are well connected, with large diameters and extending into a large fraction of the tumor volume (diffuse).

Statistics. Statistical analysis was carried out using JMP Statistics version 5.1 for the Apple Macintosh (SAS Institute, Inc.). The Student's t test for unpaired data was used to test for significant differences between two groups. ANOVA followed by the Tukey-Kramer honest significance difference (HSD) test was used to test for significant differences between more than two groups. Time courses of intravital microscopy and tumor growth data were fitted to a multivariate model (MANOVA) with repeated measures, and differences in responses of the different lines were tested for significance using an approximate F test. In all cases, differences between groups were described as significant if the probability corresponding to the appropriate statistic was <0.05.

VEGF isoform expression. Fibrosarcoma cell lines expressing only single isoforms of VEGF were generated from primary mouse embryo fibroblasts isolated from isoform-specific mice. Figure 1B shows that the isoform-specific cell lines expressed the expected VEGF isoform transcripts, when grown in vitro or as solid tumors in vivo. Some faint bands corresponding to alternative VEGF isoforms were apparent for each of the isoform-specific tumor lines when grown in vivo. These most likely derive from infiltrating normal mouse cells, as we have observed moderate macrophage infiltration in all four tumor types (data not shown). Control tumor cell lines predominantly expressed the VEGF164 and VEGF120 mRNA (Fig. 1B). Sequencing of the lower band of the doublet that is apparent just below VEGF164 in control tumor cell lines revealed a mixed sequence, which corresponded to VEGF164 and VEGF120 (data not shown), suggesting that these two isoforms could form a heteroduplex under these experimental conditions (30).

Total VEGF-A protein production, measured by ELISA, in conditioned medium from the cells, was not significantly different across the lines, although there was a tendency for the highest production in control cells (Fig. 1C). VEGF levels in cell lysates were very low by comparison (data not shown). For tumors in vivo, total VEGF levels, measured by ELISA, were highest in the VEGF164 line (Fig. 1C), despite equivalent quantitative gene expression levels across the lines (see Supplementary Fig. S1). This confirms a previous report of high VEGF165 protein levels following transfection of MCF-7 tumor cells with VEGF121, VEGF165, or VEGF189 (31). Hypoxia, a common feature of the tumor microenvironment, was reported to affect all isoforms of VEGF-A in a similar manner. However, repeated oxygen fluctuations have been reported to specifically increase VEGF164 in the retina (32) and this may be relevant to our results.

Western analysis confirmed that the cell lines produced significant quantities of the appropriate VEGF isoform, and no others, when cultured in vitro. However, protein expression of the VEGF164 isoform was predominant and VEGF188 expression was low and below detection limits in the control cell line (Fig. 1D). Attempts to carry out Western analysis on solid tumor extracts were not successful, probably because levels were below the sensitivity limits of the assay.

Vascular morphology. Established subcutaneous tumors were examined for vascularity and maturity of their vessel walls using immunohistochemistry. CD31 staining in Fig. 2A shows that all tumor types were well-vascularized. However, α-sma staining shows that control and VEGF188-expressing tumors were better stabilized with pericytes than VEGF120- or VEGF164-expressing tumors. Although this staining was apparent in blood vessel walls, there was also substantial staining in a subset of extravascular cells. Quantification of staining (Fig. 2B) showed that control and VEGF120 tumors were the most vascular (CD31), whereas control and VEGF188 tumors had the most α-sma staining. These differences were statistically significant (see legend for details). Further investigations were carried out in the VEGF120-expressing (most diffusible) and the VEGF188-expressing (nondiffusible) tumors. Figure 2C shows double-staining for CD31 and α-sma, which confirmed the localization of α-sma–positive cells (pericytes) in the vasculature of VEGF188 tumors and very little staining in the VEGF120 tumors. The spiraling morphology of the α-sma staining in the VEGF188 tumor indicates close contact between pericytes and endothelial cells. Figure 2D shows that desmin- and ANG1-positive cells were also preferentially found in VEGF188 tumors rather than in VEGF120 tumors. The staining patterns were also highly suggestive of vascular localization. Taken together, these results indicate the association of VEGF120 with sustained expansion of the tumor vasculature, whereas VEGF188 is uniquely associated with maturation of the blood vessel wall.

Vascularization of single VEGF isoform–expressing tumors was further analyzed by intravital microscopy. Figure 3A and Supplementary Fig. S2 illustrate the vascularization of VEGF120- and VEGF188-expressing tumors at early and late phases of growth in the window chamber. Initial vascularization of the VEGF120-expressing tumors was typified by dilated preexisting normal blood vessels, which were prone to hemorrhage and indistinct tumor margins (Fig. 3A). In contrast, preexisting normal vessels were narrower in the VEGF188 implants, there was very little or no hemorrhage and the tumor margins were distinct. As tumors grew, hemorrhage within the VEGF120-expressing tumors tended to resolve but vascularization was often limited to the periphery of the tumor mass. In contrast, VEGF188-expressing tumors were vascularized throughout the tumor mass (Fig. 3A). Control and VEGF164-expressing tumors had intermediate properties.

Quantitative morphologic analysis of tumor blood vessels was carried out on two peri-peripheral regions of interest within each tumor, using a ×10 objective, once the tumors became optimally vascularized. The vascular patterning in the VEGF188 tumors stood out from the others in that the blood vessels were arranged more regularly and individual vessels were uniformly narrow. Blood vessels in the other tumor types were irregular in width but generally wider and very chaotic in their arrangement. Typical regions of interest in a VEGF120- and VEGF188–expressing tumor are shown in Fig. 3B. Quantitative analysis revealed no significant difference between the tumor lines in terms of mean vascular lengths or fractal dimensions (data not shown). However, the total vascular length was higher in the VEGF188 tumors than in any of the other tumor types (Fig. 3C).

Vascular function. Leakage of i.v. injected 40 kDa fluorescent dextran was monitored in VEGF120 and VEGF188 tumors growing in window chambers, using multiphoton fluorescence microscopy (Fig. 4). Figure 4A shows typical images with diffuse vascularization and rapid leakage in a VEGF120 tumor compared with a VEGF188 tumor. Contiguity data for the whole group (Fig. 4B), is consistent with the diffuse nature of the vasculature in VEGF120 tumors, wider diameter vessels, and leakiness. Figure 4C shows the kinetics of fluorescence intensity in tumors with time after injection of FITC-dextran. There was a significant difference between the VEGF120 and VEGF188 tumors, with average intensity decreasing with time in the VEGF188 tumors and maintained or increasing in VEGF120 tumors. This difference was consistent with more rapid leakage of FITC-dextran from the intravascular to extravascular space in the VEGF120 tumors, which maintained average image intensity in the face of vascular clearance of the marker.

Response to CA-4-P. Vascular response to CA-4-P was determined in tumors growing in window chambers. Figure 5A shows typical images from VEGF120- and VEGF188-expressing tumors at 0, 1, and 24 h following treatment with a moderate single dose of CA-4-P (30 mg/kg i.v.). Response in both tumor types was typified by a substantial loss of the visible vasculature, primarily at the tumor center, at 1 h after treatment. Subsequently, VEGF120-expressing tumors progressed to almost complete avascularity and opacity of the central tumor region at 24 h. The vascular morphology was substantially less affected in the VEGF188 tumors, with very little avascularity by 24 h (Fig. 5A).

Quantitation of the vascular damage in window chamber tumors following CA-4-P treatment is shown in Fig. 5B and C for all the tumor types. There is a clear distinction in response between the VEGF120 and VEGF164 tumors on the one hand, and the wild-type and VEGF188 tumors on the other, confirming the observations shown in Fig. 5A and Supplementary Fig. S3, which gives additional images from all tumor types. The distinction in response of the different tumor types was clearly apparent from both a morphologic end point (total vessel length; Fig. 5B) and a functional end point (red cell velocity; Fig. 5C).

In order to determine whether these differences in early vascular response to CA-4-P translated into differences in therapeutic response, a repeat dosing schedule of CA-4-P was administered to mice bearing subcutaneous tumors (Fig. 6A). VEGF120 and VEGF164 tumors were, again, the most sensitive. Interestingly, although the control and VEGF188 tumors seemed to be completely resistant to the effects of CA-4-P in the first week of treatment, there was some growth retardation during the second week. In a separate experiment, VEGF120-expressing tumors were significantly more necrotic than VEGF188-expressing tumors at 24 h after a single 100 mg/kg dose of CA-4-P (79.5 ± 9.3% versus 42.8 ± 9.3% of tumor volume, respectively, P < 0.05 for Student's t test). However, after five doses of 50 mg/kg CA-4-P, as in Fig. 6A, the necrosis levels in the two tumor types were very similar (34.3 ± 7.2% and 41.1 ± 9.2%, respectively), suggesting that the responsiveness of the 188 tumors in the second week was due to increased sensitivity of the vasculature. Vascular volume in viable regions of VEGF120-expressing tumors, as measured by CD31 staining, significantly increased from 7.2 ± 0.5% to 11.0 ± 0.6% of tumor volume after a week's treatment (P < 0.05 for Student's t test). A similar trend in the VEGF188-expressing tumors was only borderline significant. Figure 6A also shows that untreated VEGF120- and VEGF164-expressing tumors reached a measurable size ∼4 days before the VEGF188 and control tumors, but then, all lines grew at the same rate, consistent with comparable blood flow rates (Supplementary Fig. S4). CA-4-P (100 μmol/L) increased apoptosis in all the cell lines growing in vitro, with no significant difference in induction rates between the different cell lines (Fig. 6B).

We have shown that mouse fibrosarcoma cells, which express only single isoforms of VEGF, under the control of the endogenous VEGF promoter, produce tumors with very different vascularization patterns, vessel wall structure, and barrier function. The method used for production of the tumor lines ensured normal overall VEGF levels. Expression of VEGF188, even at very low levels, was uniquely associated with extensive recruitment of vascular support cells (pericytes) and tumor resistance to the VDA CA-4-P. Overall, our results support the view that VEGF188 expression results in a different type of vascular bed, which renders the tumor vasculature less susceptible to CA-4-P.

Our data support previous findings that VEGF120/121 is associated with hemorrhage and vasodilation around tumors (33, 34). They are in agreement with Grunstein et al. (17) who found that VEGF120 and VEGF164 were associated with rapid initiation of fibrosarcoma growth. In contrast with our findings, Grunstein et al. found poor tumor growth associated with a fibrosarcoma cell line expressing only VEGF188. This may be due to high levels of VEGF188 in their overexpressing tumor line, leading to overvascularization of tumors. Indeed, Grunstein et al. found particularly poor vascular filling of contrast agent in their VEGF188-expressing tumors. In both studies, endogenous VEGF production from normal cells such as macrophages might be expected to contribute to vascular development. We found moderate macrophage invasion in all tumor types. However, our reverse transcription-PCR results showed that host-derived expression of VEGF isoforms was very low in all tumors and this was consistent with substantial differences in the vascular characteristics we observed. Quantitative differences in expression levels may explain why some authors have found that tumor cells transfected to overexpress VEGF188/189 are nontumorigenic (13), whereas others are tumorigenic (34–36). Interestingly, selective down-regulation of VEGF189 expression in a non–small cell lung cancer (35) and pancreatic cancer cell line (36) significantly reduced their xenotransplantability, suggesting that VEGF188/189 does play an important role in tumorigenesis.

We found striking differences in the maturation status of the vascular wall in the different VEGF isoform–producing tumors, which related to response to CA-4-P. In particular, the presence of VEGF188 resulted in tumor vascular recruitment of mural cells. This is consistent with the observations that mice selectively expressing VEGF188 recruit mural cells normally to the developing retinal venules and capillaries, whereas mice selectively expressing VEGF120 show major defects in mural cell recruitment (15). In addition, we showed that ANG1 was expressed in a subset of vessels in VEGF188 tumors and that barrier function in the VEGF188 tumors was more effective than in VEGF120 tumors, consistent with the observed structural differences. The processes involved in mural cell recruitment are complex and poorly understood but undoubtedly involve growth factors in addition to VEGF, such as platelet-derived growth factor B (PDGFB; ref. 37). Nevertheless, the current result clearly shows a key role for VEGF188 in mural cell recruitment in tumors that is associated with a functional effect. In addition, positive staining for α-sma was found in a subset of extravascular cells, presumably tumor cells or fibroblasts, of the VEGF188-expressing and control tumors but not in the VEGF120- or VEGF164-expressing tumors. This suggests a novel role for VEGF188/189 in pericyte differentiation.

Vascular network development was highly influenced by differential expression of VEGF isoforms, with VEGF120 being associated with a high vascular volume in established subcutaneous tumors and VEGF188 being associated with regular, narrow vessels and a high vascular length per tumor volume. In tumors growing in window chambers, which are subjected to high tissue pressure induced by the restraining glass, VEGF120 and VEGF164 were associated with a failure to effectively vascularize the center of tumors. This is likely to be due to the fragility of blood vessels, suggested by their lack of mural cells. Expression of VEGF164 was associated with a relatively low vascular volume (similar to VEGF188) but rapid initiation of tumor growth (similar to VEGF120). This suggests a highly proliferating phenotype for VEGF120- and VEGF164-expressing tumor cells, at the expense of vascular maturation. This is consistent with our recent in vitro data, showing higher proliferation rates for VEGF120- or VEGF164-expressing tumor cells than VEGF188-expressing or control cells (ref. 20; Supplementary Table S1). Any differences between the vasculature in VEGF120- and VEGF164-expressing tumors were subtle. However, the difference in vascular volume and the qualitative observation of less hemorrhage in the VEGF164 tumors does suggest some differences in their control of vascular maturation, an observation that warrants further investigation.

Several explanations for the different effects of individual VEGF isoforms on tumor vascular morphology and function can be envisaged. In the developing brain of Vegfa^188/188 mice, there is an increased extension of endothelial filopodia and vascular branch formation, compared with brains of Vegfa^120/120 mice (16), and this might also apply to tumor angiogenesis. There is also evidence for differential activation of VEGF receptors by the different isoforms. First, VEGF164/165 (and probably VEGF188/189) but not VEGF120/121, can bind to the accessory receptor, neuropilin-1 (NRP1; ref. 38). However, impaired signaling through the NRP-1 receptor could not account for all vascular branching defects in Vegfa^120/120 mouse embryos (16) and would not explain the significant differences between VEGF164- and VEGF188-expressing tumors. Heparan sulfate proteoglycans on the cell surface have also been implicated in modifying VEGF signaling, partly by controlling the distribution and bioavailability of secreted VEGF (39, 40). Other possibilities, which remain poorly understood, include the translocation of intracellular VEGF188/189 to the nucleus (31) and the cross-talk of VEGF signaling pathways with integrins (41).

In interpreting our data, we cannot discount the possibility that an adaptive response to the depletion of VEGF120 or VEGF164, resulting in increased expression of alternative vascular-related growth factors in the VEGF188-expressing cell line, could explain our results. Notwithstanding potential affinity issues of the VEGF antibody used in the ELISA kit for VEGF188, VEGF protein expression in the VEGF188 line in vivo seems to be relatively low (Fig. 1C), highlighting this possibility. However, real-time PCR data (Supplementary Fig. S1) did not reveal any alternative candidate genes for explaining our results. Transformation significantly lowered VEGFR2, NRP1, TIE2, Ang2, and Hey2 gene expression across all the lines, but only Ang2 and PDGFB differed in their expression levels between the VEGF188-expressing tumor cells and the other cell types. Increased Ang2 and lowered PDGFB levels, as found in the VEGF188-expressing line, are unlikely explanations for the observed increase in pericyte recruitment of this line in vivo because Ang2 is normally associated with vascular immaturity and PDGFB is a well-known chemoattractant for pericytes. In addition, administration of VEGF receptor kinase inhibitors in vitro and in vivo and a pan-isoform neutralizing antibody for VEGF in vitro, equalized both tumor cell proliferation in vitro and vascular morphology in vivo, across the different cell lines, suggesting that baseline differences in these variables were VEGF-induced (Supplementary Fig. S5; Supplementary Table S1).

Differential expression of VEGF isoforms was clearly associated with outcome following treatment with the VDA, CA-4-P, both in terms of initial vascular response and tumor growth response to a repeated dosing schedule. A modest growth response, as observed here even for the VEGF120-expressing tumors, is not unusual for VDAs and could mask significant cell-killing effects. This is thought to be due primarily to poor clearance of dead cells because of blood flow reduction (42). The presence of VEGF188, even at low levels in the control tumors, conferred resistance to treatment. Although speculative, it is possible that very tight association of VEGF188 with the cell surface minimizes the amount of protein necessary for a given outcome, as well as negatively feeding back on VEGF188 production. Necrosis data showed that VEGF188 tumors suffered vascular damage in the first week of treatment despite the lack of growth response. This may have been sufficient to sensitize the tumors to continued treatment. CD31 staining indicated an increase in vascularization of re-growing viable tumor regions after the first week of treatment, which could relate to homing of bone marrow–derived vascular progenitor cells following CA-4-P treatment, as previously reported (43). This increased vascularization may also relate to increased sensitivity to CA-4-P in the second week of treatment, although potential mechanisms require further investigation. Finally, the larger size of tumors at the start of treatment in the second week may have contributed to increased sensitivity (44).

Previously, we hypothesized that high vascular permeability and the immaturity of the vascular wall were major factors associated with the sensitivity of tumor vasculature to CA-4-P and similar agents (45, 46). The current results substantiate this hypothesis, and furthermore, suggest that VEGF188 expression is uniquely predictive of treatment outcome. Interestingly, several studies have suggested that VEGF189 expression in human cancers is associated with tumor progression and poor outcome from conventional treatments (47–50).

In conclusion, the main VEGF isoforms, under the control of the endogenous VEGF promoter, have very different effects on vascularization of fibrosarcomas in an animal model. In particular, VEGF188 and/or associated gene expression plays a crucial role in tumor vascular maturation and conferring resistance to the VDA, CA-4-P. Therefore, VEGF isoform expression may be useful for predicting tumor susceptibility to vascular-disrupting cancer therapy.

Grant support: Cancer Research UK Programme grant C1276/A3307.

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 gratefully acknowledge Dr. Gabi Dachs, Dr. Andrew Steele, and Claudia Coralli for their roles in developing the tumor lines used in this study; Finuala Hylands, Ian Wilson, Frances Daley, and Matthew Fisher for their technical support; Professor Boris Vojnovic and Dr. Simon Ameer-Beg for their expertise and support with multiphoton-fluorescence microscopy; the Gray Cancer Institute, University of Sheffield and Cancer Research UK London Institute for care of the animals; and Professor Bob Pettit for supplying the CA-4-P.