Benzoquinone ansamycin heat shock protein 90 inhibitors modulate multiple functions required for tumor angiogenesis

Neoangiogenesis is critical to tumor growth and metastasis and hence is an important generic target for therapeutic intervention in cancer (1). Angiogenic cytokines produced in tumors via oncogene activation and/or hypoxia induce a mitogenic and motogenic phenotype in host endothelial cells (2, 3). The vascular endothelial growth factor (VEGF) family (notably VEGF-A and VEGF-C) are key angiogenic stimuli acting via endothelial cell–specific receptors, the VEGF receptor (VEGFR) tyrosine kinases (4). VEGFR-2/KDR is the major receptor on vascular endothelial cells, and VEGFR-3/Flt-4 has been implicated in lymphangiogenesis and lymphatic metastasis (5). VEGFR activation induces signaling cascades, including mitogen-activated protein kinase, phospholipase C-γ, and phosphatidylinositol 3-kinase, which collectively contribute to cell survival, proliferation, migration, and differentiation (6, 7). Activated endothelial cells also produce proteolytic enzymes, including urokinase-type plasminogen activator (uPA; ref. 8) and matrix metalloproteinases (MMP), which degrade the extracellular matrix, release growth factors and cytokines, and allow directional migration toward angiogenic stimuli (9). Endothelial cells then differentiate to form tubular structures, finally recruiting supporting cells to stabilize the newly formed vessels (10). Interruption of the pathways involved in neoangiogenesis may prove a useful adjunct to conventional cancer therapy, sparing normal tissues in which endothelial cells are generally quiescent.

Molecular chaperones, such as heat shock protein 90 (Hsp90), enable the correct folding, maturation, and subcellular localization of their client proteins, particularly under the stress conditions induced by hypoxia and nutrient deprivation frequently found in cancers (11). There are four major members of the Hsp90 gene family: cytosolic Hsp90α and Hsp90β (12), GRP94 in the endoplasmic reticulum (13), and Hsp75/TRAP1 in the mitochondrial matrix (14). They have similar chaperone functions but bind different client proteins depending on their localization and association with various cochaperones. Hsp90 has recently emerged as a key target for cancer therapy (reviewed in refs. 15–18). Research has primarily focused on the direct effects of Hsp90 inhibition on cancer cells. However, many key signaling molecules involved in angiogenesis may be deregulated by Hsp90 inhibition, providing an additional benefit of Hsp90-targeted therapy. We aimed to explore the potential of benzoquinone ansamycin Hsp90 inhibitors to modulate induction of, and response to, angiogenic cytokines and client protein expression in human endothelial cells. We have also determined the functional consequences of these effects in vitro and in vivo.

Inhibitors investigated were geldanamycin, its less toxic derivative 17-allylamino-17-demethoxygeldanamycin (17-AAG), and, in some studies, the soluble derivative 17-(dimethylaminoethylamino)-17-demethoxygeldanamycin (17-DMAG). The latter two agents have now entered clinical trials. The results show the potential of Hsp90 inhibitors to interfere with multiple functional compartments of tumor angiogenesis. Such effects are complementary to the direct effects of Hsp90 inhibitors on cancer cells and would contribute to the ability of these agents to provide combinatorial blockade of the hallmark phenotypic traits of malignancy (19).

Antibodies to the following proteins were used as recommended by the suppliers: AKT, mouse IgG horseradish peroxidase conjugated, and goat IgG horseradish peroxidase conjugated (Cell Signaling Technology, Beverly, MA); glyceraldehyde-3-phosphate dehydrogenase and rat IgG horseradish peroxidase conjugated (Abcam Ltd., Cambridge, United Kingdom); Hsp70, Hsp90α, HSP90β, and GRP94 (Stressgen Biotechnologies Corp., Victoria, British Columbia, Canada); RAF-1, neu, CDK-4, and VEGFR-3 (Santa Cruz Biotechnology, Inc., Santa Cruz, CA); TRAP1 (Neomarkers, Fremont, CA); mouse VEGFR-2/Flk-1 (Zymed Laboratories, Inc., San Francisco, CA); human VEGFR-2/KDR (R&D Systems, Inc., Minneapolis, MN); focal adhesion kinase (FAK; Upstate Biotechnology, Charlottesville, VA); mouse endothelium (MECA-32; Developmental Studies Hybridoma Unit, Iowa City, IA); mouse IgG FITC conjugate (Sigma-Aldrich, Poole, United Kingdom); and biotinylated anti-rat IgG (Vector Laboratories Ltd., Peterborough, United Kingdom).

VEGF was obtained from Sigma-Aldrich, and human recombinant epidermal growth factor, betacellulin, heregulin β1, and transforming growth factor-α were obtained from R&D Systems. 17-AAG and geldanamycin were kindly provided by the Developmental Therapeutics Division of the National Cancer Institute (Rockville, MD). Compounds were dissolved in DMSO and stored at a stock concentration of 10 mmol/L at −20°C and were diluted in tissue culture medium to provide working concentrations of 20 to 200 nmol/L. 17-DMAG was obtained from InvivoGen (San Diego, CA). Stocks and working solutions were prepared as above. For in vivo studies, 17-AAG was dissolved in DMSO, diluted in the clinical vehicle egg phospholipid (2%) to a final concentration of 10% (v/v) DMSO, and given i.p. at doses and schedules indicated. Unless otherwise stated, all other chemicals were obtained from Sigma-Aldrich.

Human umbilical vein endothelial cells (pooled HUVEC), adult human dermal microvascular endothelial cells, human lymphatic endothelial cells, and their appropriate growth medium and supplements were obtained from TCS CellWorks (Botolph Claydon, United Kingdom). Cells were cultured according to the supplier's instructions and used at passages 3 to 8. MCDB-131 medium supplemented with 0.1% bovine serum albumin, 1 μg/mL hydrocortisone, and 20 μg/mL epidermal growth factor was used to provide serum-free conditions.

Five tumor cells lines were obtained from American Type Culture Collection (Manassas, VA): HT1080 (sarcoma), HCT116 (colon adenocarcinoma), Detroit 562 (pharyngeal squamous cell carcinoma), WM266.4 (melanoma), and PC3 (prostate carcinoma). PC3 was selected for enhanced metastasis by injection into the prostate glands of male NCr athymic mice and collection, subculture, and passage of tumors from the draining lymph nodes for three in vivo cycles (PC3LN3). Cells were cultured in DMEM (Life Technologies, Carlsbad, CA,) with 10% FCS. All cells were grown at 37°C in a humidified atmosphere of 5% CO₂ in air and routinely screened for Mycoplasma contamination using a nested PCR technique.

Tumor or endothelial cells were seeded into 96-well plates, and 24 hours later, fresh medium containing either vehicle or compounds over a range of concentrations was added to triplicate wells. After 96 hours, the cell numbers in each well were determined using an alkaline phosphatase colorimetric assay as described previously (20). Graphs were plotted to determine the GI₅₀, defined as the concentration of compound at which the A_{405 nm} was half that of the (confluent) control vehicle-treated cells. To compare the sensitivity to Hsp90 inhibitors of HUVEC in log-phase growth with those in stationary phase, 1.5 × 10⁴ HUVEC were seeded into T25 flasks (as confluent cells did not survive well in microtiter plates) and left until 90% to 100% confluent (3 days). The medium was replaced with medium containing vehicle, 17-AAG, or geldanamycin over a range of concentrations. The cultures were incubated for a further 96 hours, and the cells were released from the flasks using PBS-EDTA and counted in a hemocytometer.

Cells were lysed in protein lysis buffer [150 mmol/L NaCl, 1 mmol/L EDTA, 50 mmol/L Tris, 1% Triton X-100, 1 nmol/L NaF, 1 mmol/L NaVO₃, 10 μg/mL N-α-tosyl-l-lysyl-chloromethyl-ketone, 1 mmol/L DTT, 5 μmol/L fenvalerate, 5 mmol/L bisperoxo(1,10-phenanthroline)oxovanadate(V), 1 mmol/L phenylmethylsulfonyl fluoride, 1:100 protease inhibitor cocktail]. Protein concentration was determined using the detergent-compatible protein assay kit (Bio-Rad Laboratories, Hercules, CA). Protein (5–10 μg) was separated according to molecular weight on a 4% to 12% Tris-Bis gel (Invitrogen, Ltd., Carlsbad, CA) and transferred onto a nitrocellulose membrane. Membranes were blocked for 1 hour in 5% dried milk (or in the case of the VEGFR-2 antibody in 0.5% bovine serum albumin) in TBST at room temperature and probed overnight at 4°C with rotation in primary antibody diluted in “blocking buffer.” The blot was washed thrice for 10 minutes with TBST and incubated with a horseradish peroxidase–conjugated species-specific antibody diluted in blocking buffer for 1 hour at room temperature with rotation. After three additional washes, the blot was developed by a 1-minute incubation with enhanced chemiluminescent substrate and exposure to Kodak X-OMAT autoradiographic film (Kodak, Hemel Hempstead, United Kingdom).

After 2- or 24-hour treatment with or without Hsp90 inhibitors, cells were serum starved for 1 hour followed by labeling with 1 μmol/L CellTracker Green 5-chloromethylfluorescein diacetate (Invitrogen Ltd.) for 1 hour. Cells (2 × 10⁴ in 350 μL serum-free medium with or without 17-AAG or geldanamycin) were placed into the upper well of a 3 μm pore Fluoroblok membrane insert in 24-well companion plates (BD Biosciences, Franklin Lakes, NJ). Medium (800 μL) containing 5% heat-inactivated FCS was added to the lower chamber as a chemoattractant. The assay plates were incubated at 37°C in 5% CO₂ for 2 or 3 hours. An inverted fluorescent microscope and digital camera (Olympus LX70, Olympus, Middlesex, United Kingdom) were used to obtain images of the migrated cells on the underside of the membrane. Three different fields of view were scored for each well. The number of migrated cells was calculated using Image Pro-Plus 4.0 software (Media Cybernetics, Silver Spring, MD). Migration toward VEGF was done in a similar manner, but inserts were precoated with 100 μg/mL human plasma fibronectin (Invitrogen, Ltd.). Cells were migrated for 16 hours toward 20 ng/mL VEGF and analyzed as before.

Cells were treated and labeled as for the migration assay and allowed to invade through Matrigel-coated inserts (BD Biosciences) for 16 hours toward FCS before the assay was terminated and images obtained as before.

HUVECs at 80% to 90% confluence in 24-well plates were treated with inhibitors for 24 hours before assay. An area of cells was removed with a 20 μL pipette tip drawn across the center of each well. Cells were washed thrice and fresh inhibitors were added with 25 ng/mL mitomycin C to prevent proliferation during the assay. Images of the wound area were captured automatically every 30 minutes for 20 hours to monitor the migration of cells into the cleared area. The time taken for the wound to close was determined using Image Pro-Plus 4.0. Two wells were imaged per condition; the assay was carried out in three independent experiments.

HUVECs were transfected with FAK SMARTpool small interfering RNA (siRNA; Upstate Biotechnology) or a scrambled duplex with the sequence GCGCGCTTTGTAGGATTCG (Dharmacon Research, Inc., Lafayette, CO). siRNA duplexes (200 nmol/L) were transfected into HUVECs using Oligofectamine (Invitrogen Ltd.) on 2 consecutive days and used in experiments 24 hours after the second transfection.

Tumor cells were seeded into microtiter plates at 5,000 per well. After overnight incubation, cells were serum starved for 48 hours before 17-AAG was added at 400 nmol/L followed by activating ligands (100 nmol/L betacellulin, transforming growth factor-α, or heregulin β1) for a further 24 hours. Samples from six replicates were collected and analyzed for levels of VEGF-A by quantitative ELISA (R&D Systems). Cell numbers in the wells were determined by the alkaline phosphatase assay as in proliferation assays.

Endothelial cells were treated with inhibitors for 24 hours, washed with PBS, and incubated in serum-free medium in the presence of fresh inhibitors for 24 hours; after the first hour, cells were stimulated with 20 ng/mL VEGF. Conditioned medium was removed after 24 hours and samples were analyzed by gelatin zymography (Life Technologies). Conditioned medium from phorbol 12-myristate 13-acetate–treated HT1080 cells served as a positive control to identify the positions of MMP-2 and MMP-9.

uPA was measured using a uPA Microtiter ELISA (Oncogene Science, Cambridge, MA). Cell treatment and sample collection were as described for MMP analysis. A protease inhibitor cocktail was added to the samples that were concentrated 10 times using a 10,000 molecular weight cutoff protein concentrator (Vivaspin, Hanover, Germany). Trypan blue exclusion was used to normalize results to viable cell number.

HUVEC differentiation into pseudocapillaries on Matrigel serves as a simple surrogate assay for processes used by endothelial cells during neoangiogenesis (21). Cells were treated with inhibitors for 24 hours prior to assay and then plated at 3 × 10⁴ per well in 24-well plates precoated with 300 μL Matrigel (BD Biosciences). Fresh inhibitor was added and the cells were incubated for 24 hours before images were obtained and quantified for tubular area using Image Pro-Plus 4.0.

HUVECs were treated with Hsp90 inhibitors or vehicle for 24 or 48 hours. Cells were removed from T75 cm² flasks by a 4-minute incubation with 0.05 mol/L EDTA at 37°C and washed with ice-cold PBS, 1 × 10⁶ cells were incubated with 100 μL of a 5 μg/mL concentration of mouse monoclonal antibody against the external domain of human VEGFR-2 (MAB 3572, R&D Systems) or no first antibody (control) for 45 minutes on ice. After two washes with ice-cold PBS, cells were incubated on ice for 45 minutes with 100 μL of 1 mg/mL FITC-conjugated anti-mouse IgG (Sigma-Aldrich). Finally, cells were washed twice with ice-cold PBS, resuspended, and analyzed using a FACScan cytometer (BD Biosciences).

Whole HUVEC or HCT116 colon carcinoma cell lysates (400 μg total protein in 1 mL) were precleared by the addition of 5 μL mouse serum with 20 μL IgG beads (Santa Cruz Biotechnology) for 30 minutes at 4°C. VEGFR-2/KDR was immunoprecipitated using 1:100 dilution of human VEGFR-2 monoclonal antibody (sc-6251, Santa-Cruz Biotechnology, CA). The complex was recovered by incubation with 20 μL IgG beads for 1 hour at 4°C. Beads were collected by centrifugation at 1,000 × g for 30 seconds and then washed thrice in immunoprecipitation buffer [137 mmol/L NaCl, 50 mmol/L NaF, 20 nmol/L Tris-HCl (pH 7.5), 12 mmol/L β-glycerophosphate, 10% glycerol, 0.2% NP40, 1.5 mmol/L MgCl₂, 1 mmol/L EDTA, 1 mmol/L NaVO₃, 1 mmol/L phenylmethylsulfonyl fluoride, 1:10 protease inhibitor cocktail (Sigma-Aldrich)] and resuspended in 50 μL of 1× electrophoresis sample buffer (Invitrogen Ltd.), boiled for 5 minutes at 95°C, and spun briefly to pellet the beads. Supernatants were then used to probe for Hsp90 by Western blot analysis.

All procedures involving animals were done within guidelines set out by The Institute of Cancer Research's Animal Ethics Committee and the United Kingdom Coordinating Committee for Cancer Research Ad hoc Committee on the Welfare of Animals in Experimental Neoplasia (22). HCT116 human colon carcinoma cells or WM266.4 human melanoma cells (5 × 10⁶) were injected s.c. in the flanks of female NCr athymic mice. Animals were treated with vehicle or 17-AAG when tumors were ∼8 mm in mean diameter. In PC3LN3 prostate carcinoma, 3 × 10⁴ cells were injected into the ventral prostate of male NCr athymic mice under halothane anesthesia as described previously (23), and treatment with 17-AAG or vehicle commenced after 4 weeks.

Control mice received vehicle only (10% DMSO in 2% egg phospholipid) and treated mice received 40 mg/kg 17-AAG twice daily or 80 mg/kg once daily as indicated in the figures. Tissues were harvested 3 or 18 hours after the last dose and snap frozen. The samples included tumor (plus local and distant lymph node metastases in the case of PC3LN3 tumors), mesenteric vessels, and a segment of the vena cava. Samples were lysed with protein lysis buffer under nitrogen, homogenized and analyzed for protein content and expression of selected client proteins.

Bilateral HCT116 xenografts were established as before. Three mice were treated with 17-AAG (80 mg/kg i.p. once daily, 5 days weekly for 2 weeks) and three with vehicle. Tumors were harvested 6 hours after the last dose, and samples were cryopreserved in isopentane cooled to liquid nitrogen temperature, sectioned at 8 μm, and placed on slides coated with 3% 3-aminopropyltriethoxysilane. Sections were dried and stored at −20°C until use. After this, sections were fixed with ice-cold methanol for 30 minutes and three 10-minute washes in PBS preceded incubation in blocking solution (2% bovine serum albumin in PBS) for 1 hour at room temperature. Sections were then incubated in MECA-32 antibody (diluted 1:50) overnight at room temperature, after which three 10-minute washes in PBS preceded a 2-hour incubation in secondary antibody [biotinylated anti-rat IgG (mouse absorbed) diluted 1:200]. This was followed by three further 10-minute washes and incubation in avidin-biotin complex solution diluted 1:1,000 (Vector Laboratories) for 1 hour. Following three 10-minute washes in PBS and a 10-minute wash in stock buffer, the labeled vasculature was visualized using a 3,3′-diaminobenzidine kit (Vector Laboratories) using the manufacturer's protocol.

The stained sections were observed using an Olympus LX70 microscope system at ×20 objective magnification. Three to six fields per section were randomly selected and the microvessel density was determined by image analysis as for HUVEC tubularization. Representative images were captured using a digital camera (CoolSNAP Pro Color, Media Cybernetics) and analyzed using Image-Pro Plus 5.0 software (Media Cybernetics). Data are mean area of vessels in the fields of view.

Data are mean ± SD. Student's t test or Mann-Whitney U test were used for statistical analyses using Prism 2.01 from GraphPad Software, Inc. (San Diego CA).

Figure 1 shows that 17-AAG at 400 nmol/L (∼5× the mean 96-hour GI₅₀ for the cell lines tested) reduced VEGF-A secretion to at least 50% of the levels in vehicle-treated tumor cells stimulated with a variety of ligands (P < 0.001–0.05, Mann-Whitney U test). For example, control HT1080 sarcoma cells stimulated with transforming growth factor-α released 899 ± 293 pg/mL VEGF-A into the culture supernatant, which was reduced to 202 ± 30.5 (78% inhibition) by 24-hour exposure to 17-AAG. We confirmed that, at the time of assay, the cell numbers in treated and control cultures were comparable, and cell viability was >95%.

We and others have shown previously that signaling via receptor tyrosine kinases, including epidermal growth factor receptor, ErbB2/HER-2/neu, and the downstream phosphatidylinositol 3-kinase and mitogen-activated protein kinase signaling pathways, induces expression of VEGF-A and/or VEGF-C (24–26). Hsp90, as a chaperone of many of the key molecular mediators (notably ErbB2, newly formed epidermal growth factor receptor, AKT, and RAF-1), is therefore a potent regulator of oncogene-activated angiogenic cytokine induction. What is more, because VEGF-A is also induced via hypoxia-inducible factor-α, which is stabilized by Hsp90 (27), it is predicted that inhibitors, such as the geldanamycins, will subvert both of these independent stimuli of tumor neoangiogenesis.

The Hsp90 inhibitors geldanamycin, 17-AAG, and 17-DMAG potently inhibited the proliferation of two types of primary human endothelial cells in log-phase growth measured over 96 hours. The GI₅₀ of 17-AAG was 20.0 ± 2.1 nmol/L for HUVEC and 18.0 ± 3.5 nmol/L for human dermal microvascular endothelial cells; values for geldanamycin were 3.0 ± 1.6 and 6.0 ± 1.2 nmol/L, respectively. The GI₅₀s of 17-AAG and geldanamycin at shorter time intervals (24, 48, and 72 hours) were broadly similar, with only the 24-hour 17-AAG GI₅₀ being higher (by ∼2-fold) than the standard 96-hour value. The response to 17-DMAG was similar to 17-AAG, confirming results recently reported by Kaur et al. (28). Endothelial cells were thus at least as sensitive as a diverse panel of tumor cell lines tested (mean GI₅₀ for four tumor cell lines was 72.0 ± 8.3 nmol/L for 17-AAG and 17.5 ± 4.2 for geldanamycin). These results are in contrast to data reported by Kamal et al. (29), suggesting that normal cells (including human dermal microvascular endothelial cells and HUVECs) were relatively insensitive to 17-AAG.

It was important to know whether contact-inhibited, nonproliferating endothelial cells were equally susceptible to Hsp90 inhibitors. We used such cultures in an attempt to mimic quiescent vasculature. Confluent HUVEC cultures were significantly less susceptible to Hsp90 inhibitors than the same cells in log-phase growth, because after 4-day exposure to 100 nmol/L 17-AAG (5 × 96-hour GI₅₀), cell survival was 82 ± 13% of controls, whereas exposure of actively proliferating cells to this concentration effectively reduced survival to zero, with clear evidence of apoptosis. Similar results were obtained with geldanamycin. This may provide a therapeutic window whereby antiangiogenic effects may be achieved in vivo by concentrations of Hsp90 inhibitors similar to those able to directly inhibit tumor cells while sparing normal vasculature.

Figure 2A shows that HUVEC express the four major isoforms of the Hsp90 family. 17-AAG and geldanamycin target the conserved NH₂-terminal ATP-binding domain of Hsp90 and hence are expected to inhibit all isoforms (11). Depletion of client proteins and up-regulation of Hsp70 provides a clear indicator of response to Hsp90 inhibition, with robust responses being obtained at 5 to 10 × GI₅₀ concentrations (30, 31). Expression levels of RAF-1 and AKT protein were reduced in endothelial cells treated with 17-AAG in a concentration- and time-dependent manner (Fig. 2B and C). Densitometric analysis indicated that AKT and RAF-1 proteins were reduced by 70% and 89%, respectively, in HUVECs exposed to 200 nmol/L 17-AAG, although effects were also evident at lower concentrations. Under the same conditions, Hsp70 expression was induced to a maximum of 796% of controls.

Although the depletion of client proteins is a useful “fingerprint” of Hsp90 inhibition, it is not clear which are rate limiting for specific cellular functions. This is highlighted by the recent identification of a subset of ansamycins that potently inhibit uPA-mediated tumor cell invasion (32). Although initially linked to MET down-regulation, it is now clear that effects occur at compound concentrations well below those required to deplete this and other known client proteins, suggesting that other molecular targets remain to be identified (33).

The development of new generations of Hsp90 inhibitors, perhaps with isoform selectivity, would be greatly accelerated if key pharmacodynamic determinant(s) of response could be identified. In addition, endothelial cell–specific biomarkers may enable dissection of the relative importance of direct (antitumor) and indirect (antiangiogenic) effects in vivo.

Based on these observations on HUVEC sensitivity and biomarker changes, we next explored the effects of Hsp90 inhibition on endothelial cell functions required to complete neoangiogenesis, dissecting the process into separate activities that could be readily assayed in vitro before embarking on in vivo studies.

Two distinct forms of endothelial cell motility were evaluated: chemotaxis, the directional migration in a chemotactic gradient, and haptotaxis, the lateral movement over a solid surface. A short exposure to Hsp90 inhibitors (2 hours) had no effect on either type of endothelial cell migration (data not shown), consistent with the fact that client proteins were expressed at normal levels at this time. However, when the cells were pretreated for 24 hours, there was a significant concentration-dependent inhibition of endothelial cell motility/migration. For example, chemotaxis of HUVEC toward FCS was reduced to 60% of controls at 20 nmol/L 17-AAG (1 × GI₅₀ concentration; P < 0.05) and ∼17% at 200 nmol/L (P < 0.001; Fig. 3A). Migration was more potently inhibited when VEGF rather than FCS was used as a chemoattractant (data not shown; P < 0.001 for 20 nmol/L 17-AAG comparing FCS versus VEGF), suggesting that Hsp90 exerts major effects on HUVEC motility via VEGFRs. Geldanamycin also inhibited HUVEC chemotaxis but less potently and consistently than 17-AAG: concentrations of 8 or 20 nmol/L (∼2.5 and 6.5 × GI₅₀) were required to significantly inhibit HUVEC migration (data not shown). The viability of cells was 96 ± 2% in all cases.

Hsp90 inhibitors also prevented HUVEC haptotaxis in a scratch wound assay (Fig. 3B). Control cells completed wound closure by 10 hours, but cells treated with 100 nmol/L 17-AAG failed to achieve this by 20 hours. A concentration-dependent inhibition was also seen at lower doses with 20 nmol/L 17-AAG (1 × GI₅₀), delaying wound closure by 4 hours. Similar effects were seen with geldanamycin, where at 2 and 5 × GI₅₀ concentrations the monolayer wound remained fully open over the 20-hour observation period (data not shown). Again, no significant cell death occurred during the study and the inclusion of mitomycin C excluded the possibility that cell proliferation was contributing to wound closure.

FAK is a recognized mediator of integrin-mediated cell migration (34), which recruits and activates other signaling molecules leading to reorganization of the actin cytoskeleton (reviewed in ref. 35). As FAK is a known client protein of Hsp90 (36), its loss following Hsp90 inhibition could explain the observed decrease in haptotaxis induced by 17-AAG. To test this hypothesis, we examined FAK protein levels following 17-AAG exposure under conditions corresponding to those used in migration assays. Total FAK protein levels were reduced by 20% following 24 hours treatment with 100 nmol/L 17-AAG. Although not a dramatic reduction, Masson-Gadais et al. (37) showed that Hsp90 inhibition blocked tyrosine phosphorylation of FAK induced by VEGF in HUVECs, a process that is an important component of focal adhesion assembly. We confirmed that phosphorylation of FAK at Tyr^566/567 was reduced by 50% at 40 nmol/L 17-AAG and undetectable at 60 nmol/L (data not shown).

Using siRNA, we investigated further the effect of FAK depletion on HUVEC haptotaxis. A delay in cell monolayer wound closure of 10 hours was obtained compared with scrambled siRNA treatment (Fig. 3C) and densitometric analysis of Western blots (Fig. 3D) showed that FAK siRNA had decreased protein expression by 74%. This implicates FAK in HUVEC haptotactic migration; however, Hsp90 inhibition completely prevented haptotaxis rather than merely delaying it, suggesting that other Hsp90 client proteins may also be involved.

AKT activates the small GTP-binding proteins Cdc42 and Rac1 to affect cell motility (38). This is supported by our current results that showed significant down-regulation of AKT by concentrations of 17-AAG and geldanamycin, which inhibited haptotaxis, and by the fact that a phosphatidylinositol 3-kinase inhibitor (LY294002) and an AKT inhibitor (1L-6-hydroxymethyl-chiro-inositol 2-[(R)-2-O-methyl]-3-O-octadecylcarbonate) both blocked chemotaxis and haptotaxis (39).¹

Localized degradation of the extracellular matrix is required to provide a passage for capillary sprout migration toward the tumor and to assist in the release and activation of survival and motility factors (40). This “invasion” is generally considered to comprise two basic components: cell motility and proteolysis. Invasion can be assayed in vitro by the ability of cells to cross a barrier of Matrigel, a mixture of extracellular and basement membrane proteins (predominantly laminin and collagen IV). HUVEC invasion was inhibited in a concentration-dependent manner by 17-AAG, with a reduction of 74% obtained with 60 nmol/L 17-AAG (Fig. 4A). At this concentration, HUVEC migration was somewhat less effectively inhibited (63%; see Fig. 3A), although this differential did not achieve statistical significance. We hypothesized nevertheless that the apparently more potent effects on invasion might be attributed to a second component (i.e., matrix proteolysis).

A major protease secreted by endothelial cells is MMP-2, which degrades collagen type IV (reviewed in ref. 41). However, neither 24-hour exposure to 17-AAG nor geldanamycin at up to 10 × GI₅₀ concentrations (200 and 30 nmol/L, respectively) decreased latent or active enzyme levels as determined by gelatin zymography (data not shown). A second key protease, uPA, was then investigated. On binding to its receptor, this serine protease cleaves plasminogen to plasmin, causing degradation of the extracellular matrix either directly or indirectly by activating MMPs (8). 17-AAG reduced secretion of uPA in a concentration-dependent manner (Fig. 4B) and similar effects were observed with geldanamycin at equipotent concentrations. Thus, inhibition of two complementary mechanisms (motility and proteolysis) may contribute to the potent effects of Hsp90 antagonists on endothelial cell invasion. Our data are consistent with the results of Webb et al. (32) who showed that that certain geldanamycins are potent inhibitors of uPA-mediated invasion in tumor cells. This effect may be mediated by down-regulation of MET, a known client protein expressed on both tumor cells and endothelial cells (42), although this hypothesis has recently been questioned (33).

Endothelial cells form tube-like structures on a layer of Matrigel, in a process involving both cell attachment and migration and described as representing capillary sprout formation (21). Figure 4C shows that 17-AAG induced concentration-dependent decreases in tubular area measured at 24 and 48 hours, and inset images show representative control and treated cultures at 24 hours. Geldanamycin also caused a dramatic reduction in tubule area, with 50% inhibition of tubule area achieved at 6.5 mmol/L and 90% inhibition at 8 nmol/L following 48-hour treatment (data not shown).

The potent effects of Hsp90 inhibitors on endothelial cell functions may be due to depletion of ubiquitous client proteins, such as AKT, FAK, and RAF-1, but endothelial cell–specific clients could also be implicated. Hsp90 chaperone receptor tyrosine kinases, such as ErbB-2 (43) and MET (32); hence, we investigated key endothelial cell–specific receptor tyrosine kinases. For example, VEGFR-2 is a major receptor on endothelial cells with downstream signaling inducing mitogenesis, migration, invasion, and differentiation in neoangiogenesis; VEGFR-3 plays similar roles in lymphatic endothelial cells.

Evidence suggests that VEGFR-2 is a client protein of Hsp90. Masson-Gadais et al. (37) reported previously that VEGFR-2 and Hsp90 form a complex in HUVECs and later showed that deletion mutants of VEGFR-2 that are unable to associate with Hsp90 are able to inhibit endothelial cell migration (44). We show here that VEGFR-2 coassociates with Hsp90 and show by various independent methods that Hsp90 inhibition leads to receptor down-regulation. Firstly, we showed that geldanamycin reduced VEGFR-2 expression in HUVECs in a concentration- and time-dependent manner as measured by Western blotting (Fig. 5A); similar results were obtained with 17-AAG (data not shown). Secondly, independent confirmation was obtained by flow cytometric analysis of HUVECs incubated with a different anti-VEGFR-2 antibody. Following exposure to Hsp90 inhibitors for 24 to 48 hours, there was a significant down-regulation of VEGFR-2 expression (Fig. 5B) with mean fluorescence intensity reduced from 57.2 units in control cells to 28.8 and 26.1 units, respectively, in cells exposed to 17-AAG and geldanamycin for 48 hours. Thirdly, immunoprecipitation of VEGFR-2 from HUVECs probed with an anti-Hsp90 antibody confirmed coassociation of the two proteins as reported previously (ref. 37; Fig. 5C).

We also showed that 17-DMAG decreased expression of VEGFR-3 on cultured human lymphatic endothelial cells over a concentration range that concordantly down-regulated AKT and up-regulated Hsp70 (Fig. 5D), suggesting that other members of the VEGFR family could also represent novel Hsp90 client proteins.

We next investigated whether an effect on client proteins could be detected in murine host endothelial cells in vivo following a short treatment with 17-AAG. Two sources of material were used: vena cava and mesenteric vessels. Because the latter were not dissected from the mesenteric fat, the preparations also contained lymphatic vessels. In the mesenteric samples, we found a clear decrease in all three murine VEGFRs and up-regulation of Hsp70 (Fig. 6A), indicating a response to Hsp90 inhibition similar to that seen in vitro in human endothelial cells and lymphatic endothelial cells. In vena cava samples from the same animals, we again detected clear inhibition of VEGFR-1 and VEGFR-2; VEGFR-3 was not detected (data not shown).

Human tumor xenograft samples were also examined by Western blot with antibodies directed against murine VEGFR-2 to determine if signals could be obtained from host neoangiogenic vessels. In three different tumor models (WM266.4 melanoma, HCT116 colon carcinoma, and orthotopic PC3 prostate carcinoma), we detected decreased expression of VEGFR-2 in response to 17-AAG, with the strongest effect evident in the highly vascular melanoma (Fig. 6B, top). In the PC3 model, decreased VEGFR-2 expression was also evident in local and distant metastases (Fig. 6B, middle). In the same samples, using antibodies recognizing human proteins, clear down-regulation of tumor client proteins was seen (e.g., ErbB2 and CDK-4; data not shown).

These data therefore confirm in multiple independent systems (mouse and human, macrovasculature and microvasculature, in vitro and in vivo) that Hsp90 inhibitors can down-regulate VEGFR-2 expression in endothelial cells, confirming that this receptor is an important client protein and potential direct target of 17-AAG in vivo. Preliminary in vitro and in vivo data suggest that the other VEGFRs (VEGFR-1 and VEGFR-3) are also Hsp90 client proteins, which could be susceptible to targeting by inhibitors, such as 17-AAG.

Significantly, preliminary studies indicate that Hsp90 inhibitors can indeed target tumor neoangiogenesis. We showed that treatment of mice with 17-AAG (two 5-day schedules of 80 mg/kg/d i.p.) was able to reduce the microvessel density in HCT116 xenografts as determined by immunohistochemistry and semiquantitative image analysis (Fig. 6C). The mean area of vasculature was reduced from 42,468 μm² in the controls to 13,257 μm² in the 17-AAG treated tumors, representing a 69% decrease. This did not quite achieve statistical significance probably due to the small sample size (n = 3; P = 0.0523, Mann-Whitney U test). Studies with 17-DMAG have also recently shown inhibitory effects in a nontumor angiogenesis model of s.c. implanted Matrigel plugs (28).

Although down-regulation of VEGFR-2 in normal, mature mouse vasculature (vena cava and mesenteric vessels) was detected following treatment with 17-AAG, this would not necessarily be a cause for concern as quiescent endothelial cells are VEGF independent for survival. We found that the survival in vitro of confluent, contact-inhibited HUVEC was not significantly affected by 17-AAG or geldanamycin at concentrations up to 5 × IC₅₀. We therefore believe that compromised signaling through VEGFR-2 in normal tissues would potentially only be an issue should these cells be required to respond to injury. Indeed, inhibition of tumor neoangiogenesis in xenografts by 17-AAG shown here was not associated with any obvious adverse effects or evidence of vascular damage (e.g., hemorrhage) in normal tissues.

In summary, the ability of tumors to induce angiogenesis is vital for their growth and dissemination and is an area of intense interest for the development of novel therapies as an adjunct to conventional cytotoxic agents (1). Hsp90 chaperones several proteins implicated in the angiogenic cascade (e.g., FAK, hypoxia-inducible factor-1α, and AKT). The present studies explored the potential role of Hsp90 inhibitors to target tumor angiogenesis, from the production of angiogenic cytokines by tumor cells to the endothelial responses that they induce. We showed that all four major Hsp90 chaperone proteins are expressed in HUVECs and that benzoquinone ansamycins potently inhibit not only VEGF-A secretion by diverse types of tumor cells but also all major endothelial cell responses to angiogenic stimuli: proliferation, migration (haptotaxis and chemotaxis), invasion, proteolytic enzyme secretion (uPA), and tubular differentiation. The effects were exemplified with three Hsp90 inhibitors, geldanamycin, 17-AAG, and 17-DMAG, and dose-response relationships were shown in several instances. We obtained strong data from a variety of independent studies to support the recognition of VEGFR-2 as an important Hsp90 client protein and have shown that geldanamycins can significantly down-regulate its expression both in vitro and in vivo. This is associated with compromised neoangiogenesis in human tumor xenografts. Preliminary in vitro and in vivo evidence indicates that Hsp90 inhibitors also down-regulate VEGFR-1 and the key lymphangiogenic mediator VEGFR-3, with potential implications for the control of lymphatic metastasis.

Hsp90 inhibitors reportedly show selectivity for cancer cells compared with many normal cells. This has been attributed to the fact that, in the former, Hsp90 is primarily found in a superchaperone complex that has higher affinity for, and is more readily inhibited by, 17-AAG than the uncomplexed form, which predominates in normal cells (29). Selectivity may also occur because of the greater dependency of cancer cells on oncogenic pathways affected by Hsp90 inhibition (45) and because of the stressed state of many tumor cells (11). A clear therapeutic window is seen in xenograft models (30, 46, 47) and early clinical trials are encouraging as responses are seen at well-tolerated doses (48).

Interestingly, our in vitro studies show that activated endothelial cells are as sensitive to Hsp90 inhibition as tumor cells, whereas quiescent HUVECs are much less sensitive. Consistent with this, tumors in mice treated with 17-AAG showed reduced microvascular density, but there was no sign of vascular damage in normal tissues. It will be important (but challenging) in future preclinical and clinical studies to quantify the contribution of antiangiogenic effects to tumor growth inhibition with agents that can target both cellular compartments, because understanding of this may affect dose scheduling and appropriate pharmacodynamic end points. It has been shown that tumors grown in mice with deficient angiogenesis (Id−/− mice) are more sensitive to 17-AAG (96% growth inhibition) compared with those in wild-type mice (48% growth inhibition; ref. 49). Therefore, “classic” angiogenesis inhibitors in combination with inhibitors of Hsp90 function may be useful in anticancer therapy. Hsp90 inhibitors reduce expression not only of endothelial cell–specific receptors and oncogenic tyrosine kinase receptors but also shared targets on downstream signaling pathways (e.g., RAF and AKT). Hsp90 inhibitors may therefore target multiple pivotal points in both tumor and endothelial cell survival, proliferation, and invasion, providing a combinatorial approach to attack multiple features of the malignant phenotype (19, 50).

We thank Jenny Titley for the analysis of fluorescence-activated cell sorting samples and Dr. Edward Sausville (National Cancer Institute) for providing 17-AAG and geldanamycin.