Antiangiogenic Therapy Decreases Integrin Expression in Normalized Tumor Blood Vessels

Tumor blood vessels normalized by antiangiogenic therapy may provide improved delivery of chemotherapeutic agents during a window of time but it is unknown how protein expression in tumor vascular endothelial cells changes. We evaluated the distribution of RGD-4C phage, which binds α_vβ₃, α_vβ₅, and α₅β₁ integrins on tumor blood vessels before and after antiangiogenic therapy. Unlike the control phage, fd-tet, RGD-4C phage homed to vascular endothelial cells in spontaneous tumors in RIP-Tag2 transgenic mice in a dose-dependent fashion. The distribution of phage was similar to α_vβ₃ and α₅β₁ integrin expression. Blood vessels that survived treatment with AG-013736, a small molecule inhibitor of vascular endothelial growth factor and platelet-derived growth factor receptors, had only 4% as much binding of RGD-4C phage compared with vessels in untreated tumors. Cellular distribution of RGD-4C phage in surviving tumor vessels matched the α₅β₁ integrin expression. The reduction in integrin expression on tumor vessels after antiangiogenic therapy raises the possibility that integrin-targeted delivery of diagnostics or therapeutics may be compromised. Efficacious delivery of drugs may benefit from identification by in vivo phage display of targeting peptides that bind to tumor blood vessels normalized by antiangiogenic agents. (Cancer Res 2006; 66(5): 2639-49)

The morphologic, organizational, and metabolic diversity of endothelial cells exemplifies the heterogeneity of the microvasculature (1). In the brain, continuous endothelial cells of the blood-brain barrier restrict the movement of solutes, whereas fenestrated endothelial cells of the choroid plexus favor solute flux (2). Signals from endothelial cells influence the formation of bone (3) and the development of the pancreas and liver from the primitive endoderm (4, 5). Compared with quiescent established blood vessels, endothelial cells in angiogenic blood vessels express additional proteins, such as the α_vβ₃, α_vβ₅ (6–8) and the α₅β₁ integrins (9). Peptides that bind specifically to distinct vascular beds in normal mice and in a human subject, as identified by in vivo phage display, show the inherent molecular heterogeneity within the microvasculature (1, 10, 11). Intraorgan vascular heterogeneity has also been shown by two peptide phage that are ephrin A-type ligand mimetics that bind solely to the vasculature of normal murine pancreatic islets with increased localization to islet tumor blood vessels (12). Given such inherent molecular diversity of the normal and tumor microvasculature, resident receptor proteins for selective targeting of diagnostic and therapeutic agents to specific vascular beds can be identified and exploited.

An example of a tumor-targeting phage is one displaying the sequence, CDCRGDCFC (termed RGD-4C phage). The double cyclic RGD-4C peptide binds with a 200-fold greater in vitro affinity to α_vβ₃ and α_vβ₅ integrins and a 50-fold greater affinity to the α₅β₁ integrin than the linear GRGDSP peptide (13). Moreover, in vivo studies show RGD-4C phage have a 40- to 80-fold greater selectivity for tumor blood vessels (8). Tumor burden decreased upon i.v. delivery of the RGD-4C peptide fused to either doxorubicin or to a proapoptotic peptide with reduced host toxicity (14, 15). Other studies revealed FITC-labeled RGD-4C peptide binds to both human MDA-MB-435 tumor xenografts and infiltrating murine tumor endothelial cells (16).

In addition to site-directed targeting, inhibition of proangiogenic factors and their signaling pathways has shown promise as tumor therapies (17, 18). Tumor blood vessel regression is enhanced when therapeutic regimens combine vascular endothelial growth factor receptor (VEGFR)-2 inhibitors, such as SU5416, with platelet-derived growth factor receptor (PDGFR)-β inhibitors, such as SU6668 or Gleevec (19), in implanted rat gliomas and pancreatic islet tumors (20, 21). We have recently shown that VEGF Trap, a soluble VEGFR-1/VEGFR-2 chimeric antibody that binds VEGF-A, VEGF-B, and placental growth factor-1 (22), and AG-013736, a potent small molecule inhibitor of VEGF/PDGF receptor tyrosine kinases (23), cause the disappearance of endothelial fenestrations, tumor vessel regression, and the appearance of basement membrane ghosts in murine pancreatic islet tumors (24). Although combination therapies using antiangiogenic compounds with chemotherapy are successful (25, 26), the timing for administration of secondary therapeutic(s) via the normalized tumor vasculature is likely to be a critical factor (27, 28). Although the surviving normalized tumor vessels are functional, molecular changes to the tumor endothelium following antiangiogenic therapy are less well understood.

In this study, we examined the physiologic distribution of the tumor-targeting RGD-4C phage by immunofluorescence microscopy in blood vessels of the RIP-Tag2 pancreatic islet tumor mouse model (29) and its distribution following AG-013736 therapy. The first objective of this study was to compare the distribution of RGD-4C versus fd-tet phage in tumor blood vessels. We sought to quantify the dose dependency of phage distribution and to corroborate the distribution of α_vβ₃ and α₅β₁ integrins to that of RGD-4C phage in tumor blood vessels. In addition, we compared the targeted localization of RGD-4C phage in tumor blood vessels to normal blood vessels in the lung, thyroid gland, cerebral cortex, and liver. Our second objective was to determine whether RGD-4C phage could be used as a biological tool to determine whether endothelial cells that survive antiangiogenic therapy modify their pattern of integrin expression, thereby signaling a phenotypic change in the vasculature.

Animals. RIP-Tag2 transgenic mice from a C57BL/6 background contain the insulin promoter–driven SV40 T-antigen and produce spontaneous multifocal and multistage pancreatic islet tumors (29). RIP-Tag2-positive mice were genotyped from tail-tip DNA by PCR. Male and female RIP-Tag2 mice and nontransgenic littermates between 8 and 12 weeks of age were used in these studies. Mice were housed under barrier conditions at the animal care facility at the University of California, San Francisco. The Institutional Animal Care and Use Committee at University of California, San Francisco, approved all experimental procedures.

Phage preparation. Purified single-stranded fd-tet (30) or CDCRGDCFC (13) phage DNA was electroporated into competent MC1061 Escherichia coli (31), and plated onto Luria-Bertani agar medium containing 100 mg/L streptomycin and 40 mg/L tetracycline (LB/Strep/TET). Single colonies were picked, peptide inserts were amplified by colony PCR, and sequences were verified (12). Phages were amplified overnight at 37°C with agitation from a single transformed MC1061 colony in 100 mL LB/Strep/TET. Phage were precipitated from the bacterial supernatant with 15% NaCl/PEG 8000 (Fisher Scientific, Tustin, CA) for 1 hour on ice, pelleted by centrifugation for 20 minutes at 4°C at 10,400 × g, gently resuspended in 5 mL sterile PBS, and precipitated with 15% NaCl/PEG 8000 on ice for 30 minutes. The final phage pellet was gently resuspended in 100 μL sterile PBS, centrifuged for 2 minutes at maximum g force, and the supernatant was filtered through a 0.22 μm syringe filter. Infectivity titers were determined using established protocols (32).

Phage and antibody injections and tissue preparation. Purified, titered phage preparations were used within 24 hours of preparation. Serial dilutions of phage at 10⁹, 10⁸, and 10⁷ transforming units (TU) were made with DMEM containing Earle salts (University of California, San Francisco Cell Culture Facility) to a final volume of 200 μL. Twenty-five micrograms hamster CD61 (PharMingen, San Diego, CA), rat CD51 (eBioscience, San Diego, CA), monoclonal rat CD49e antibodies (PharMingen), or corresponding amounts of normal hamster (Jackson ImmunoResearch, West Grove, PA) or rat serum antibodies (Jackson ImmunoResearch) were diluted with sterile saline to a final volume of 125 μL each and filtered through 0.22 μm filters. Phage and antibodies were administered i.v. into Avertin (2,2,2-tribromoethanol, 0.015-0.017 mg/g, injected i.p., Sigma-Aldrich Corp., St. Louis, MO)–anesthetized mice (32) and allowed to circulate for 6 minutes. Body temperatures of anesthetized mice were maintained with a heating pad. Mice were systemically perfused with 1% paraformaldehyde in PBS, pH 7.4, and tissues were frozen (12).

AG-013736 treatment. Nine-week-old male RIP-Tag2 mice (n = 10) were injected with either AG-013736 (25 mg/kg, i.p., BID) or with the vehicle [three parts PEG 400 to seven parts acidified H₂O (pH 2)] for 7 days. AG-013736 is a potent small molecule inhibitor of VEGF/PDGF receptor tyrosine kinases (IC₅₀, 1.2 nmol/L for VEGFR-1, 0.25 nmol/L for VEGFR-2, 0.29 nmol/L for VEGFR-3, 2.5 nmol/L for PDGFR-β, 2.0 nmol/L for c-Kit, and 218 nmol/L for FGFR-1; ref. 23), and was provided by Pfizer Global Research and Development (San Diego, CA). Treated mice were injected i.v. with 10⁹ TU RGD-4C phage and perfused as described above.

Immunohistochemistry. Details for immunostaining fixed frozen sections and sections of phage injected tissues were previously described (12). For immunostaining antibody-injected tissues, sections were incubated with either monoclonal Armenian hamster anti-mouse CD31 (1:1,000, Chemicon, Temecula, CA) or monoclonal rat antimouse CD31 antibodies (1:500, PharMingen) and 5% normal goat serum (Jackson ImmunoResearch) with 1% Triton X-100 in PBS (PBST; pH 7.4). Rinsed sections were incubated in secondary antibody solutions containing goat FITC-conjugated anti-rat antibodies (1:200, Jackson ImmunoResearch) and goat Cy3-conjugated anti-Armenian hamster antibodies (1:400, Jackson ImmunoResearch), or goat FITC-conjugated anti-Armenian hamster antibodies (1:200, Jackson ImmunoResearch), or goat Cy3-conjugated anti-rat antibodies (1:400, Jackson ImmunoResearch) and 5% normal goat serum in PBST. Triple stained sections were incubated with monoclonal rat anti-mouse CD31, monoclonal Cy3-conjugated α-smooth muscle actin (α-SMA, 1:2,000, Sigma-Aldrich), rabbit polyclonal anti-fd-bacteriophage antibody (1:5,000, Sigma-Aldrich), and 5% normal mouse serum in PBST. Rinsed sections were incubated in a secondary antibody solution containing mouse FITC-conjugated anti-rat IgGs, mouse Cy5-conjugated anti-rabbit IgGs (1:400, Jackson ImmunoResearch), and 5% normal mouse serum in PBST. For the AG-013736 studies, frozen 80 μm sections from vehicle or AG-013736-treated tissues were incubated with monoclonal Armenian hamster anti-mouse CD31 and monoclonal rat CD49e antibodies (1:500, PharMingen) and 5% normal goat serum in PBST. Rinsed sections were incubated in a secondary antibody solution containing goat FITC-conjugated anti-Armenian hamster antibodies and goat Cy3-conjugated anti-rat antibodies in PBST.

Imaging. Fluorescence images were acquired using an externally coded, three-chip charge-coupled device camera (CoolCam, SciMeasure Analytical Systems, Atlanta, GA) fitted on a Zeiss Axiophot fluorescence microscope with Fluar objectives or with a Zeiss LSM 510 Laser Scanning confocal microscope with krypton-argon and helium-neon lasers at 488, 543, and 633 nm (Carl Zeiss, Jena, Germany) and analyzed with the Zeiss LSM 510 software v. 3.2.2.

Quantification of phage in blood vessels. Densities of phage immunoreactivity in blood vessels, hereafter called phage vessel area density, for the dose-dependent studies were quantified from 20-μm-thick confocal projections of immunostained RIP-Tag2 islets using ImageJ (http://rsb.info.nih.gov/ij/) with the Zeiss LSM Reader plug-in interface (http://rsb.info.nih.gov/ij/plugins/lsm-reader.html). Total blood vessel densities of tumor islets were determined from CD31-immunoreactive blood vessels and calculated from user-defined islet regions after thresholds were empirically determined. Phage vessel area densities were calculated as the number of pixels of phage immunoreactivity divided by the total vessel area represented by CD31 immunoreactivity. Mean phage vessel area density values and SEs were determined from five islet tumors or acinar regions each from three mice injected with 10⁹, 10⁸, and 10⁷ TU RGD-4C phage, and five islet tumors each from three mice injected with 10⁹, 10⁸, and 10⁷ TU fd-tet phage. For the buffer control, five to six islet tumors each were analyzed from five mice. For the AG-013736 studies, phage vessel area densities were quantified from digital Coolcam fluorescence images acquired with a ×10 Fluar objective. Phage vessel area densities were determined as the percentage of red pixels (phage) to the total number of green pixels (CD31) in islets using the threshold setting of 50 for each image. Mean ± SE values were calculated from five images per mouse (n = 5 mice per group). Statistical analyses were determined by the ANOVA Bonferroni-Dunn test. Cy3 anti-rat CD49e fluorescence intensities of 80 μm pancreatic immunostained sections from vehicle and AG-013736-treated mice were measured from digitized Coolcam images acquired with a ×20 Fluar objective. Cy3 digitized images were converted to grayscale 8-bit images using the look-up table importer plug-in in ImageJ. A customized look-up table that defines the boundaries of fluorescence intensities as a spectrum of color from 0 to 255 was applied to each 8-bit image that was then converted to a surface plot. Increased fluorescence intensities correspond to increased peak heights in the Z-plane.

Distribution of RGD-4C phage in tumor blood vessels. Tumors in RIP-Tag2 transgenic mice were used as a model because they are multifocal and multistage, thereby allowing visualization of phage in tumor blood vessels during various stages of tumorigenesis as well as in normal blood vessels in the acinar pancreas. The immunohistochemical staining pattern of i.v. administered 10⁹ TU of RGD-4C phage in the tumor vasculature was distinct and punctate (Fig. 1). In RIP-Tag2 tumors, RGD-4C phage was abundant in blood vessels 6 minutes after injection such that the majority of the tumor vasculature could be delineated by RGD-4C phage immunoreactivity (Fig. 1A and B). In contrast, an equivalent amount of the insertless, negative control phage, fd-tet, did not localize to tumor blood vessels (Fig. 1C); RGD-4C phage localized much less to blood vessels in normal islets (Fig. 1D,, arrows). RGD-4C phage accumulation was not influenced by tumor size and was distributed throughout the tumor vasculature with focal regions of extravasation (Fig. 1E,, arrowheads). Moreover, the amount of RGD-4C phage was comparatively low to none in the acinar pancreas blood vessels of either RIP-Tag2 or normal mice (Fig. 1B,-E, asterisks). Extravasated RGD-4C phage did not seem to colocalize with α-SMA-immunoreactive pericytes (Fig. 1F , arrow).

Quantification of RGD-4C phage in tumor blood vessels. The affinity of RGD-4C phage binding to blood vessels in RIP-Tag2 tumors was examined further by decreasing the injected dose from 10⁹ to 10⁷ TU (Fig. 2). We reasoned that if in vivo binding of RGD-4C phage was a specific event, then our immunofluorescence studies should show a reduction of phage binding commensurate with decreasing amounts of i.v. administered phage. These experiments should also provide information regarding the threshold for phage detection by immunohistochemistry. The amount of RGD-4C phage in blood vessels in RIP-Tag2 tumors was reduced when the amount of injected phage decreased from 10⁹ to 10⁷ TU (Fig. 2A-C). Background immunostaining in tumor blood vessels from buffer-treated control mice was negligible (data not shown). RGD-4C phage was detectable when injected at 10⁷ TU (Fig. 2C,, right); however, the immunoreactivity was comparable with fd-tet phage injected at 10⁷ TU (data not shown). Mean RGD-4C and fd-tet phage vessel area densities were quantified in CD31-immunoreactive tumor blood vessels at each dose (Fig. 2D). At 10⁹ TU, the mean RGD-4C phage vessel area density was 4-fold greater than that of fd-tet phage in tumor blood vessels and was statistically significant compared with the amount of RGD-4C phage found in normal blood vessels of the acinar pancreas. The amount of RGD-4C phage found in the acinar pancreas blood vessels at this dose was similar to the background immunoreactivity measured in islet tumors from buffer-injected RIP-Tag2 control mice. At 10⁸ TU, the mean RGD-4C phage area density in tumor blood vessels was 3.4-fold greater than that of fd-tet phage. Mean phage vessel area densities for RGD-4C and fd-tet phages were similar at the injected dose of 10⁷ TU. Unlike RGD-4C phage, the mean fd-tet phage vessel area density in tumor blood vessels was similar when injected into RIP-Tag2 mice at either 10⁸ or 10⁷ TU.

Molecular specificity of RGD-4C phage localization in tumor blood vessels. Given that the RGD peptide is the recognition sequence for many adhesive proteins, such as the α_vβ₃, α_vβ₅, and α₅β₁ integrins (33), we sought to evaluate the molecular specificity of RGD-4C phage localization by examining the in vivo distribution of the α_v, β₃, and α₅ integrins in RIP-Tag2 tumor blood vessels (Fig. 3). Similar to the phage immunostaining pattern, β₃ integrin (CD61, a binding partner of α_v integrin), immunoreactivity in blood vessels of RIP-Tag2 tumors was not uniformly distributed (Fig. 3A). Some blood vessels were strongly stained whereas others were devoid of β₃ integrin immunoreactivity. α_v Integrin (CD51) immunoreactivity was weaker and nonuniformly distributed in tumor blood vessels as well (Fig. 3B). Consistent with recent reports (34, 35), α₅ integrin (CD49e) immunoreactivity and, hence, α₅β₁ integrin expression, because this integrin is a unique heterodimer (33), was strong and uniformly distributed throughout the tumor blood vessels and was not detectable in adjacent normal acinar blood vessels (Fig. 3C). Neither control hamster (data not shown) nor rat serum antibodies colocalized with tumor blood vessels or normal blood vessels in the acinar pancreas (Fig. 3D,, asterisk). Extravasation of α_v, α₅, and rat serum antibodies from tumor blood vessels was apparent outside tumor blood vessels (Fig. 3B -D, arrows).

Distribution of RGD-4C phage in other tissues. In contrast to its distribution in tumor blood vessels, negligible amounts of RGD-4C phage were found in blood vessels of the lung (Fig. 4A) and in capillaries surrounding the thyroid follicles (Fig. 4B), whereas phages were not detected in the cerebral cortex (Fig. 4C). Interestingly, by increasing the input of RGD-4C phage to 5 × 10⁹ TU, we observed phage in blood vessels of the cerebral cortex (data not shown). Phage immunoreactivity was distinct, punctate, and evenly distributed within the hepatic endothelial sinusoids such that the sinusoids (Fig. 4D,, arrow) surrounding each central vein (Fig. 4D,, arrowhead) were delineated by phage immunoreactivity. Rapid clearance of phage by the reticuloendothelial system and corresponding accumulation of phage in the liver as determined by bacterial infection (8) are consistent with our observations of identical high amounts of RGD-4C phage (Fig. 4D) or fd-tet phage immunoreactivity in the murine liver and spleen (data not shown) after a 6-minute circulation time. The half-life of injected fd-tet phage in the blood was ∼5.5 minutes and recovery of fd-tet phage from the C57BL/6 mouse liver 6 minutes postinjection was typically 4% of total injected phage in a nonsaturated system (data not shown). After a 60-minute circulation, fd-tet phage was not detected in hepatocytes but were present in Kupffer cells with the majority of fd-tet phage associated with sinusoidal endothelial cells (data not shown).

Reduced RGD-4C phage localization after AG-013736 treatment. Preferential localization of RGD-4C phage to the tumor vasculature prompted us to examine whether the distribution of RGD-4C phage in tumor blood vessels following treatment of RIP-Tag2 mice with AG-013736 would be altered (Fig. 5). We reasoned that if tumor blood vessels are normalized (28) following AG-013736 treatment, the amount of RGD-4C phage bound in vivo would reflect a difference in endothelial cell integrin expression. Pancreatic sections from vehicle-treated mice injected with RGD-4C phage 6 minutes before perfusion showed strong phage immunoreactivity in tumor blood vessels (Fig. 5A) that was similar to islet vessels in untreated RIP-Tag2 mice (compare with Fig. 1B and E). Conversely, RGD-4C phage localization in islet blood vessels was greatly reduced in AG-013736-treated mice (Fig. 5B,, arrow). Blood vessels in the AG-013736-treated tumors had notably uniform diameters and decreased tortuosity than vehicle-treated tumors as previously reported (24). RGD-4C phage area density in the AG-013736-treated islet vasculature was ∼4% of that quantified from vehicle-treated vessels (Fig. 5C).

Reduced α₅β₁ integrin expression in AG-013736-treated tumor blood vessels. To understand the molecular basis for the marked decrease of RGD-4C phage distribution in the surviving islet blood vessels of AG-013736-treated mice, the expression of the α₅β₁ integrin was evaluated in vehicle versus AG-013736-treated pancreatic sections by immunohistochemistry (Fig. 6). Expression of α₅β₁ was chosen because its expression was robust and uniform in untreated RIP-Tag2 tumor blood vessels (Fig. 3C). In vehicle-treated mice, α₅β₁ integrin expression in tumor blood vessels was similar to that found in untreated RIP-Tag2 mice (Fig. 6A and B). Alternatively, α₅β₁ integrin immunoreactivity in the remaining islet blood vessels from mice treated with AG-013736 for 7 days was much less and heterogeneous (Fig. 6C and D). In some islet blood vessels, α₅β₁ integrin expression was strong (Fig. 6C, and D, arrow), whereas expression was weak (Fig. 6C, and D, yellow arrowhead) or absent in other blood vessels (Fig. 6C, and D, white arrowhead). Acinar blood vessels were also α₅β₁ immunoreactive because the antibody was not administered i.v.. α₅β₁ Integrin fluorescence intensity in vehicle versus AG-013736-treated islet blood vessels of large islet tumors was decreased in the latter (Fig. 6E and F). Qualitative differences of α₅β₁ integrin fluorescence intensities in the corresponding surface plot illustrate significantly decreased peak heights in surviving islet blood vessels from AG-013736-treated mice (Fig. 6G).

We show the in vivo cellular distribution of RGD-4C phage by immunofluorescence microscopy is highly sensitive and reflects the expression of vascular receptors. We established RGD-4C phage colocalize with CD31-immunoreactive tumor blood vessels and that extravasation of phage from tumor blood vessels occurs in focal regions within a few minutes. Quantification of phage area density in tumor blood vessels showed a dose dependency that is consistent with specific binding and to α_v, α₅, and β₃ integrin expression. Treatment of RIP-Tag2 mice with AG-013736 markedly decreased the amount of RGD-4C phage bound to surviving blood vessels, which was consistent with decreased α₅β₁ integrin expression. RGD-4C phage binding to the normal endothelium was barely detectable, whereas systemic clearance of phage occurred mainly via the hepatic sinusoidal endothelial cells.

Differential integrin expression in vehicle versus AG-013736-treated blood vessels. Consistent with earlier reports (6–9), our studies show the α_vβ₃ and α₅β₁ integrins that bind the RGD-4C phage in normal blood vessels were up-regulated in tumor blood vessels. Following AG-013736 treatment, our studies showed endothelial expression of the α₅β₁ integrin was substantially down-regulated. Studies using primary endothelial cells isolated from human umbilical cord veins show α_vβ₃, α₅β₁, and α₂β₁ integrins are physically associated with VEGF₁₆₅-bound VEGFR-2 (36, 37). Moreover, endothelial cell migration and proliferation are enhanced when cells were plated on vitronectin, fibronectin, and type I collagen, the primary cognate ligands for α_vβ₃, α₅β₁, and α₂β₁ integrins, respectively. Adhesion proteins, such as fibronectin, vitronectin, laminin, as well as collagen, contain the RGD binding motif as recognized by the α_vβ₃, α₅β₁, and α₂β₁ integrins, respectively (33). The presence of basement membrane ghosts lacking endothelial cells in AG-013736-treated RIP-Tag2 islet tumors (24) is consistent with our current finding of integrin down-regulation because integrins mediate cell adhesion to the basement membrane. Others have shown decreased endothelial cell expression of integrins such as α₂β₁ may account for the loss of capillary lumen maintenance and tube formation (38). We have shown decreased expression of VEGFR-2 and VEGFR-3 in AG-013736-treated RIP-Tag2 tumors (24). Thus, down-regulation of VEGF receptors by AG-013736 and decreased RGD-4C binding to AG-013736-treated tumors may have effectively disrupted the crosstalk between VEGF and adhesion receptors (39, 40). Whether integrin expression is down-regulated as a result of inhibiting VEGFR-2 signaling cascades or by directly inhibiting integrin transcription is unknown. The net effect of AG-013736 treatment in RIP-Tag2 tumors indicates that the surviving tumor blood vessels are functional and have decreased levels of integrin expression despite the high VEGF concentration in these tumors (41). Our results suggest that AG-013736 therapy may act synergistically with integrin antagonists (42) in combination therapy to effect a faster rate of tumor blood vessel regression.

Pericyte changes associated with AG-013736 treatment. Comparison of pericyte morphology in wild-type mice to the RIP-Tag2 tumor mouse model (43) and a mouse model of prostate cancer (44) show pericytes are loosely associated with tumor blood vessels, have processes that extend away from the vessel wall, overlap with other pericytes, and accompany and extend beyond endothelial sprouts. Due to the heterogeneity of islet tumors in the RIP-Tag2 model, the amount of α-SMA-immunoreactive pericytes increases with tumor size, whereas vessels in hyperplastic islets have mostly desmin-immunoreactive pericytes (43). Subsequent to this work, Inai et al. (24) showed pericyte processes are more tightly associated with normalized blood vessels in tumors from RIP-Tag2 mice treated with AG-013736 for 7 days by scanning electron microscopy. An interesting finding of this work is that although the area density of CD31-immunoreactive blood vessels decreases by 79% following AG-013736 treatment, the reduction of α-SMA-immunoreactive pericyte area density (33%) is not coincident with vessel regression but is similar to the reduction of basement membrane area density. Indeed, immunohistochemistry with all three marker proteins, α-SMA, CD31, and type IV collagen, shows the α-SMA-immunoreactive pericytes associate with either normalized blood vessels or with basement membrane sleeves that do not contain endothelial cells. Recent studies showed that pericytes of RIP-Tag2 mice treated with AG-013736 for 7 days and then withdrawn from treatment for 4 days reverted to that morphology described by Morikawa et al (43).⁴

Immunohistochemical studies using antibodies that recognize PDGFR-β showed the area density of PDGFR-β-immunoreactive pericytes was unchanged in the absence or presence of AG-013736 and after AG-013736 treatment was withdrawn. Consistent with earlier studies (24), however, the ratio of α-SMA and PDGFR-β area densities decreased during AG-013736 treatment and returned to baseline levels 7 days after withdrawal of AG-013736. These results suggest sustained AG-013736 treatment reverted pericyte protein expression to a normal phenotype rather than decreased the total number of pericytes. The biochemical transformation of pericytes following AG-013736 treatment corresponds with their gross morphologic changes to a more normal phenotype (24). Whether vessel normalization and pericyte transformation occurs concurrently or in a sequential fashion is unknown. Decreased endothelial expression of α₅β₁ integrin and the findings of pericyte transformations upon treatment with a broad-spectrum compound, such as AG-013736, lends further support for effective treatment of tumors with combination therapies.

Physiologic evaluation of tumor blood vessels by RGD-4C phage. Recent reports show that combination therapies of antiangiogenic inhibitors are more efficacious when used with low-dose chemotherapy (25–27). In addition, because tumors activate multiple angiogenic pathways, others propose that clinical therapies should include cocktails of angiogenic inhibitors that target multiple angiogenic pathways (18, 28). The paradox of abrogating tumor blood vessels in antiangiogenic therapy when delivery of therapeutics is dependent on a functional vasculature can be resolved because angiogenic inhibitors gradually “prune” the dysfunctional tumor vasculature to a more “normalized” state (28). The period in which the normalized functional vasculature can be effectively used to deliver appropriate therapeutics, however, is a critical temporal window (26). Accurate assessment of the tumor vasculature during antiangiogenic therapy by peptide phage such as RGD-4C may provide one variable by which the rate and extent of tumor blood vessel normalization can be evaluated for appropriate application of secondary treatment regimens.

Targeting specificity of RGD-4C phage. The distribution of 10⁹ TU RGD-4C in islet tumor blood vessels reflected the expression patterns of α_vβ₃, α₂β₁, and α₅β₁ immunoreactivity. Our results do not contradict the fact that many different integrins recognize the RGD binding motif (33). Nevertheless, it would be difficult to unequivocally assign RGD-4C binding to be exclusively limited to either α_v integrins or α₅β₁ integrins based on the current findings described here until reliable mouse α_vβ₃ and α_vβ₅ antibodies become available. Although α_v integrins are also expressed on pericytes and tumor cells, the limited amount of RGD-4C phage extravasation we observed here may be insufficient to detect on these perivascular cell types and/or may be confined within the vascular basement membrane to allow binding to surrounding tumor cells.

We found, however, that decreasing the dose of RGD-4C phage by 10-fold did not alter its specificity for tumor blood vessels. These findings support the targeting specificity of RGD-4C phage to angiogenic tumor blood vessels and corroborates previous reports of decreased toxicity to normal tissues of RGD-4C-based delivery of targeted therapeutics (14, 15). Although targeting peptide sequences outside the context of phage confers an improved specificity, phage display is an appealing tool by which small peptides can be rapidly propagated by bacterial amplification and screened in vivo due to its low toxicity to mammalian cells. Given our experimental variables, we show by immunofluorescence microscopy that 10⁹ TU is an optimal dose for visualization of targeting phage and may be the optimal dose for identifying peptides from a random phage library in in vivo phage display.

Localization of RGD-4C phage in other tissues. Although RGD-4C phage bind to tumor blood vessels, we were interested in exploring this further by examining normal blood vessels of other tissues from RIP-Tag2 mice that represent a variety of endothelial cell types. Given the extensive vascularity of the lung, the small amount of phage detected indicates that RGD-4C phage binding to tumor blood vessels was not determined by the concentration of endothelial cells. To address whether binding of RGD-4C phage was specific to fenestrated endothelial cells, which are abundant in RIP-Tag2 tumors and in the thyroid gland, we found a minimal amount of RGD-4C phage in the follicular capillaries of the thyroid. These results indicate that endothelial cell type was also not a key determinant in RGD-4C phage binding in the RIP-Tag2 tumor blood vessels. The brain and liver are typically treated as negative and positive control organs, respectively, in in vivo phage display experiments, and our results confirmed this. Identical amounts of fd-tet or RGD-4C phage in the liver suggests phage localization in the hepatic endothelial sinusoids seems to be a property of the phage protein coat and not determined by the peptide targeting sequence thereby confirming previous findings (8).

Although in vitro phage display was originally developed to map antigenic sites on antibodies in vitro (45), this method was successfully adapted to identify in vivo vascular addresses in mice and in a human subject (1, 12, 46–49). Our studies here show that visualization of targeting phage in blood vessels at the cellular level by immunofluorescence microscopy allows direct identification and quantification of phage homing to the endothelium of tumor blood vessels. Decreased vascular localization of RGD-4C phage following AG-013736 treatment is consistent with down-regulation of integrin expression during normalization of surviving tumor blood vessels. Decreased integrin expression on tumor vascular endothelial cells raises the possibility of impaired targeting of chemotoxins that are dependent on integrin overexpression. These findings also illustrate the use of phage-displayed peptides as tools to probe the vascular addresses of abnormal blood vessels in disease and identify changes in the microvasculature in response to antiangiogenic treatment. Phage-displayed peptides that exploit the molecular changes in the normalized tumor endothelium may be judiciously used for targeted delivery of secondary chemotherapeutics.

Grant support: NIH grants HL-24136 and HL-59157 from the National Heart, Lung, and Blood Institute (D.M. McDonald); the Vascular Mapping Project (D.M. McDonald); National Cancer Institute grant P50-CA90270 (D.M. McDonald and W. Arap); NIH grants CA88106 (R. Pasqualini) and CA90810 (W. Arap); The Gillson-Longenbaugh Foundation and the V Foundation (R. Pasqualini and WA); and the AngelWorks Foundation (D.M. McDonald, R. Pasqualini, and W. Arap).

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 Erin Ator, Michael Mancuso, and Barbara Sennino (University of California, San Francisco) for overseeing the care of the RIP-Tag2 colony; Gyulnar Baimekanova and Jie Wei (University of California, San Francisco) for genotyping the mice; Dana Hu-Lowe and David Shalinsky for providing AG-013736 (Pfizer Global Research and Development, San Diego, CA); and Jonas Fuxe and Beverly Falcón (University of California, San Francisco) for critical reading of the manuscript.