Identification of biomarkers for tumor endothelial cell proliferation through gene expression profiling

Extensive efforts are under way to identify antiangiogenic therapies for the treatment of human cancers. Many proposed therapeutics target vascular endothelial growth factor (VEGF) or the kinase insert domain receptor (KDR/VEGF receptor-2/FLK-1), the mitogenic VEGF receptor tyrosine kinase expressed by endothelial cells. Inhibition of KDR catalytic activity blocks tumor neoangiogenesis, reduces vascular permeability, and, in animal models, inhibits tumor growth and metastasis. Using a gene expression profiling strategy in rat tumor models, we identified a set of six genes that are selectively overexpressed in tumor endothelial cells relative to tumor cells and whose pattern of expression correlates with the rate of tumor endothelial cell proliferation. In addition to being potential targets for antiangiogenesis tumor therapy, the expression patterns of these genes or their protein products may aid the development of pharmacodynamic assays for small molecule inhibitors of the KDR kinase in human tumors.

Angiogenesis is critical to the progression of numerous cancers. Tumors induce endothelial cell proliferation, migration, and differentiation resulting in neovascularization that arises from existing blood vessels and circulating endothelial cell precursors (1, 2). Tumor cells induce angiogenesis primarily through the production and secretion of vascular endothelial growth factor (VEGF), a potent endothelial cell mitogen and ligand for the kinase insert domain receptor (KDR, FLK-1, or VEGF receptor-2; refs. 3–7). KDR is a transmembrane receptor tyrosine kinase expressed in vascular endothelial cells that transduces the primary mitogenic functions attributed to VEGF (3, 8–10). Currently, several small molecule inhibitors of KDR kinase activity are being evaluated as anticancer agents in clinical trials.

In addition to VEGF and KDR, many other proteins contribute to the physiologic process of angiogenesis. To identify novel genes that might prove useful as targets for antiangiogenesis cancer therapy, we employed a genome-wide gene expression profiling strategy in cultured endothelial cells and in animal tumor models. We first identified a general mitogen-induced proliferation signature in cultured primary microvascular endothelial cells. We then identified the subset of genes from the proliferation signature that were endothelial cell specific. Many of these endothelial cell–specific genes were completely uncharacterized and may themselves be developed as novel targets for antiangiogenic therapy. In addition, we were also interested in genes that we could show to be regulated by KDR kinase inhibitors in vivo (and thus more likely involved in angiogenesis), as they could be additional targets in the VEGF/KDR signaling pathway and could potentially serve as biomarkers of KDR inhibition. Using gene expression profiling techniques with animal tumor tissue, we found that several of the endothelial cell–specific genes were regulated in vivo by small molecule KDR kinase inhibitors in a manner consistent with suppression of endothelial cell growth. Endothelial cell–specific expression of these putative biomarkers was confirmed by immunofluorescence microscopy. The genes were further validated by correlating their in vivo gene expression changes to an independent, immunohistochemical measure of endothelial cell proliferation.

We also propose that a gene expression signature specific to proliferating endothelial cells could be used to develop a pharmacodynamic assay to support clinical development of KDR kinase inhibitors. It may be possible to correlate changes in tumor endothelial cell gene expression following exposure to a KDR kinase inhibitor to the rate of endothelial cell proliferation. Such an assay would be useful where immunohistochemistry is inappropriate or impractical, such as with small tissue samples from biopsies (i.e., fine needle aspirates) or from tissue samples with poor morphology. As a real-time quantitative reverse transcription (RT)-PCR assay, it would be sensitive enough to use with small clinical samples and would be compatible with existing clinical laboratory instrumentation. In addition, this type of multivariable assay would likely have the ability to detect inhibition of angiogenesis relatively quickly after initiating therapy, eliminating the longer period of time required to visualize morphologic changes in tumor microvasculature.

Primary human dermal microvascular endothelial cells (HDMVEC) and rat heart microvascular endothelial cells (RHMVEC) were purchased from VEC Technologies (Renneslaer, NY) and grown in culture according to the supplier's directions. Endothelial cell monolayers were maintained at 37°C in a 5% CO₂ humidified atmosphere in tissue culture flasks coated with human fibronectin (Sigma, St. Louis, MO) using complete MCDB-131 medium MCDB-131 supplemented with 10% fetal bovine serum (FBS; Invitrogen, Carlsbad, CA) and the growth factor cocktail ENDOGRO (VEC Technologies).

For in vitro endothelial cell proliferation experiments, cells were harvested by trypsinization between passages 3 and 6 following initiation of culture from frozen stocks, counted, and seeded in fibronectin-coated tissue culture plates at 75% confluence (1.5 × 10⁶ cells per plate, 100 mm diameter plates). Cell growth was arrested for 24 hours by mitogen withdrawal and then stimulated by the addition of 100 ng/mL VEGF, 100 ng/mL basic fibroblast growth factor (bFGF), or 200 μg/mL ENDOGRO. For growth arrest, the culture medium was changed to prewarmed DMEM supplemented with 10% FBS. For stimulation of cell growth, the growth arrest medium was replaced with MCDB-131 supplemented with 10% FBS and the appropriate growth factor. Matched control plates that received no supplemental stimulatory growth factor were made for each stimulation condition. At the desired time following growth factor stimulation, the culture medium was removed quickly by aspiration, and the cells were lysed in 1.2 mL RLT buffer (guanidine thiocyanate lysis buffer for RNA stabilization and purification, Qiagen, Valencia, CA). Cell lysates were homogenized in QIAshredders, and total RNA was isolated with RNeasy Mini affinity columns (Qiagen). Gene expression profiles from a total of eight independent VEGF-stimulated cultures, seven ENDOGRO-stimulated cultures, and four bFGF-stimulated cultures were determined for HDMVECs. Profiles from four independent VEGF-stimulated cultures, four ENDOGRO-stimulated cultures, and four bFGF-stimulated cultures were determined for RHMVECs.

A rat glial cell line (C6, ATCC CCL-107) and a rat mammary adenocarcinoma (Mat B III, ATCC CRL-1666) were used for our animal tumor models. C6 cells were maintained in culture at 37°C in a 5% CO₂ humidified atmosphere in Ham's F-12 medium supplemented with 2 mmol/L l-glutamine, 1 mg/mL sodium bicarbonate, 15% horse serum, 2.5% FBS, 10 units/mL penicillin, and 10 μg/mL streptomycin (all medium components from Invitrogen). Mat B III cells were grown in McCoy's 5a medium supplemented with 1.5 mmol/L l-glutamine, 10% FBS 10 units/mL penicillin, and 10 μg/mL streptomycin. For RNA isolation from C6 or Mat B III cells, 2 × 10⁶ cells growing in a 100 mm diameter tissue culture plate were lysed directly in 1.2 mL RLT buffer. Following lysate homogenization with a QIAshredder, total RNA was isolated with RNeasy MINI affinity columns.

Experiments were conducted in accordance with the standards established by the U.S. Animal Welfare Acts, set up in Merck & Co., Inc., Institutional Animal Care and Use Committee. C6 glial cells and Mat B III adenocarcinoma cells were chosen for implantation into syngeneic, immunocompetent Fischer 344 rats. Before implantation, cells were collected, washed in PBS, and resuspended in HBSS (Invitrogen) at a density of 2 × 10⁷ (C6) or 2 × 10⁶ (Mat B III) cells/mL.

C6 cells were injected s.c. into the right flank of male Fischer 344 rats (150-175 g, 10⁷ cells per animal). Following cell injection, animals were randomized according to body weight to receive oral doses of vehicle (0.5% methylcellulose) or KDR kinase inhibitor (Table 1). For each oral dose, KDR kinase inhibitor A [3-(5-((4-(methylsulfonyl)-piperazin-1-yl)methyl)-1H-indol-2-yl)quinolin-2(1H)-one] was given at 40 mg compound/kg animal body weight (40 mg/kg/dose). KDR kinase inhibitor B [4-((2-((5-cyano-1,3-thiazol-2-yl)amino)pyridin-4-yl)methyl)-n-methylpiperazine-1-carboxamide] was given at 10 mg/kg/dose. Both inhibitors were formulated in 0.5% methylcellulose (11, 12). Once-daily oral dosing began 7 days after tumor cell implantation and continued for 1, 2, or 3 days at which point the animals were sacrificed. Body weight and tumor size (digital caliper) were monitored on days 0, 4, and 7 and immediately before sacrifice. Blood samples were taken at the time of sacrifice to determine plasma compound concentrations as described previously (13). Excised tumors were bisected, with half preserved for RNA extraction by snap freezing in liquid nitrogen and half fixed for histology or immunofluorescence microscopy. Five vehicle-treated and five compound-treated animals were sacrificed at each time point. RNA was extracted from tumor samples with RNeasy Mini columns according to standard protocols. Briefly, frozen tumor samples were weighed, placed in sample tubes containing RLT buffer (600 μL RLT per 30 mg tissue), and immediately homogenized for 10 to 20 seconds using a rotor/stator homogenizer. Total RNA was isolated from homogenized tissue lysate with RNeasy affinity columns, resuspended in DEPC-treated water, and frozen at −80°C. RNAs from the five tumors in each vehicle-treated cohort were combined to form three reference RNA pools. RNAs isolated from each of the tumor samples from the five compound-treated rats in each cohort were compared with the appropriate time-matched reference pool of RNA during microarray hybridization. In addition, RNA from individual vehicle-treated rats was compared with time-matched vehicle-treated pool to assess interanimal variability.

Mat B III cells between passages 20 and 30 were injected on the mammary fat pad around the fourth left nipple (10⁶ cells per animal) of female Fischer 344 rats (150-175 g). Before dosing, animals were randomized into groups according to tumor size and body weight. Once-daily oral dosing of KDR kinase inhibitor A (40 mg/kg/dose formulated in 0.5% methylcellulose) or vehicle (0.5% methylcellulose) began on day 7 after tumor cell implantation and continued for 4 additional days. Six vehicle-treated and six compound-treated animals were sacrificed on day 11, 4 hours after final dosing. At the time of necropsy, tumors were weighed immediately on removal. Half of each tumor was fixed for histology or immunofluorescence microscopy and the other half was immediately snap frozen in liquid nitrogen for RNA extraction. Total RNA was isolated in the same manner as with the C6 tumor studies. RNAs from the vehicle-treated cohort were combined to form a control RNA pool. RNAs isolated from each of the tumor samples from the compound-treated rats were compared with the control pool of RNA during microarray hybridization. RNAs from individual vehicle-treated animals were compared with the vehicle-treated pool to assess interanimal variability.

Total RNA isolated from cultured cells or tumor tissue samples was used to make fluorescently labeled cRNA that was hybridized to DNA oligonucleotide microarrays as described previously (14, 15). Briefly, 4 μg of total RNA from an individual tumor sample or endothelial cell culture were used to synthesize dsDNA through RT. cRNA was produced by in vitro transcription and labeled postsynthetically with Cy3 or Cy5. Two populations of labeled cRNA, a reference population and an experimental population, were compared with each other by competitive hybridization to microarrays. Two hybridizations were done with each cRNA sample pair using a fluorescent dye reversal strategy. For animal tumor studies, reference cRNA pools were made by pooling equal amounts of cRNA from each tumor in the appropriate vehicle-dosed group.

Species-specific microarrays were used throughout this study. Human microarrays contained 23,916 oligonucleotide probes corresponding to individual genes or expressed sequence tags. Rat microarrays contained probes to 22,592 genes or expressed sequence tags. Oligonucleotide probe sequences were chosen to maximize gene specificity and minimize the 3′ replication bias inherent in RT of mRNA. In addition, both microarray formats contained ∼1,000 control probes for quality control purposes. All oligonucleotide probes on the microarrays were synthesized in situ with inkjet technology (Agilent Technologies, Palo Alto, CA; ref. 14).

After hybridization, arrays were scanned and fluorescence intensities for each probe were recorded. Ratios of transcript abundance (experimental to control) were obtained following normalization and correction of the array intensity data. Gene expression data analysis was done with the Rosetta Resolver gene expression analysis software (version 3.2, Rosetta Biosoftware, Seattle, WA). For each gene sequence present on the microarrays, statistical significance of differential gene expression was determined by calculating P values according to the following equation:

where Erf is the error function for a Gaussian distribution of zero mean and xdev is the adjusted difference in fluorescence intensities between Cy3 and Cy5 signals calculated by the equation:

where r is Cy5 intensity, g is Cy3 intensity, and s is the error associated with the respective channel.

All gene expression data from this study will be made publicly available through submission to the Gene Expression Omnibus Data Repository at the National Center for Biotechnology Information.³

Quantitative real-time PCR was done with gene-specific PCR primer pairs and amplicon-specific fluorescent probes [Taqman, Applied Biosystems, Inc. (ABI), Foster City, CA] according to published protocols (ABI Assays-on-Demand Gene Expression Protocol, Rev A).⁴

One-step quantitative RT-PCR reactions were done using ABI Taqman One-step RT-PCR Master Mix reagents and 25 ng total RNA template on an ABI PRISM 7900HT Sequence Detection System. Two-step RT-PCR experiments were initiated by cDNA synthesis from 25 ng total RNA as template using ABI High-Capacity cDNA Archive Kit. Two-step quantitative real-time PCR was done with standard reagents (Taqman Universal PCR Master Mix) on the ABI PRISM 7900HT Sequence Detection System. Real-time PCR reactions were done in duplicate in a 25 μL reaction volume in 384-well plates. Primer and probe sequences used for each gene are listed in Table 2. For every RNA sample, transcript abundance of glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was determined. In addition, transcript abundance of genes of interest and GAPDH was determined for calibrator RNA samples, either total human lung RNA or total rat lung RNA. Fold changes in gene expression were calculated using the ΔΔC_T method (ABI User Bulletin 2, Rev B).⁵ Hs refers to Homo sapiens. Rn refers to Rattus norvegicus.

Tumor samples were fixed immediately on removal from sacrificed animals by submersion in a Zn-Tris fixative solution for immunohistochemistry (Zinc Fixative, BD Biosciences-PharMingen, San Diego, CA) for 24 hours at room temperature (22°C) followed by submersion in 70% ethanol at room temperature for an additional 24 hours. All subsequent steps were done at room temperature. Tumor samples were embedded in paraffin (Tissue-Tek VIP Processing/Embedding Medium, Sakura Finetek, Torrance, CA) and cut into 3 μm sections on a Sakura Accu-Cut SRM microtome (Sakura Finetek). Sections were placed onto glass microscope slides and stained with H&E to determine tumor morphology. Adjacent sections were double stained for CD31, a validated endothelial cell marker (16–18), and for Ki-67, a marker of cell proliferation (19, 20).

For CD31/Ki-67 double staining, tissue sections were dewaxed in xylene and rehydrated through graded ethanol washes. Following washes in deionized H₂O (dH₂O) and TBS, a hydrophobic barrier was placed around the tissue section with a hydrophobic pen (Super Pap Pen, EMS 71310). Sections were blocked with Protein Block (Biogenex, San Ramon, CA) for 30 minutes and incubated with anti-CD31 antibodies (mouse anti-rat, Serotec, Raleigh, NC) diluted 1:1,000 in DAKO antibody diluent with blockers (DakoCytomation, Carpinteria, CA) for 2 hours. After several brief washes in TBS + 0.1% Tween 20 (TBST), sections were incubated with biotinylated anti-mouse IgG secondary antibody (DakoCytomation Alkaline Phosphatase Kit Link K-0610) for 10 to 30 minutes, washed several times with TBST, and incubated with streptavidin coupled to alkaline phosphatase (DAKO Alkaline Phosphatase Kit Link K-0610) for 10 to 30 minutes. Sections were then washed again with several changes of TBST, and CD31-bound antibodies were visualized by incubation with Vulcan Fast Red Substrate (Biocare Medical, Walnut Creek, CA) for 10 minutes (color development monitored microscopically). Sections were then washed in dH₂O stored overnight in TBS.

Ki-67 is a nuclear protein expressed only in proliferating cells. To facilitate antibody recognition of Ki-67, we used a high-temperature antigen retrieval strategy. Sections were submerged in Target Retrieval Solution (1× DakoCytomation S1699 diluted with dH₂O) in a decloaking chamber (Biocare Medical, DC2002) and heated to 195°C for 1 minute. Sections were cooled by running room temperature dH₂O into the decloaking chamber and then rinsed in TBS. Residual peroxidase activity was blocked by incubating the sections with 3% H₂O₂ in TBS for 20 minutes. Sections were washed several times in TBS and then incubated with anti-Ki-67 antibodies (rabbit anti-human, Novacastra, Newcastle upon Tyne, United Kingdom) diluted 1:2,000 in antibody diluent for 2 hours. Sections were washed with TBST and then incubated with undiluted biotinylated anti-rabbit IgG (DakoCytomation, Link K-0609) for 10 minutes. Sections were washed in TBST and then incubated with streptavidin coupled to horseradish peroxidase (DakoCytomation, Link K-0609) for 10 minutes. Sections were washed again in TBST, and antibodies bound to Ki-67 were visualized by incubation with 3,3′-Diaminobenzidine Plus substrate (DakoCytomation) for 5 minutes (color development monitored microscopically). Sections were washed in dH₂O, incubated with 3,3′-Diaminobenzidine Enhance for 20 minutes at room temperature and washed again with dH₂O.

CD31/Ki-67 double-stained tumor sections were counterstained with filtered Mayer's hematoxylin (Lillie's formulation, DakoCytomation) for 2 minutes and then washed with tap H₂O until no color remained in the wash water. Sections were then rinsed in dH₂O, dehydrated with 100% ethanol, cleared with xylenes, and mounted with Permount (Fisher Scientific, Hampton, NH).

Sequential brightfield images of CD31/Ki-67 double-labeled tumor sections were obtained with a 3-charge-coupled device color video camera (Optronics) attached to an Olympus BX-51 microscope equipped with a motorized stage (Prior H128, Watertown, MA) and a ×40 objective. The number of images per section varied between 1,000 and 4,000 depending on total tissue area. CD31 staining and Ki-67 staining were quantitated for each image using the ImageProPlus software package (version 4.5, Media Cybernetics, Carlsbad, CA). Proliferating endothelial cells were identified as those cells with cytoplasmic CD31 staining and nuclear Ki-67 staining. Cells staining positive for CD31 but without nuclear staining for Ki-67 were scored as nonproliferating endothelial cells. The percentage of proliferating endothelial cells was calculated by dividing the Ki-67+ nuclear area associated with endothelial cells by the total nuclear area associated with endothelial cells (both Ki-67+ and Ki-67−). Endothelial cell proliferation percentages represent the combined analysis results from at least 100 images with CD31 staining per tumor section.

Tumor samples were fixed, embedded, sectioned, dewaxed, and rehydrated as described for immunohistochemistry above. All subsequent steps were done at room temperature. After a brief rinse in TBS, tissue sections were blocked by incubation with Sniper Blocking Reagent (Biocare Medical) for 5 to 10 minutes, rinsed in TBS, and incubated with primary antibodies diluted 1:1,000 in DAKO antibody diluent for 2 hours [antibodies against angiopoietin-2 (ANGPT2), clusterin, and PLAU or urokinase-type plasminogen activator (uPA) were from Santa Cruz Biotechnology (Santa Cruz, CA) and raised in goat or rabbit; antibodies against EDNRB were from Calbiochem (San Diego, CA) and raised in sheep; antibodies against CD31 were from Serotec and raised in mouse]. Sections were then washed with TBS containing 0.2% Tween 20 (Sigma) and incubated with appropriate secondary antibodies diluted 1:200 (10 μg/mL) in DAKO antibody diluent with blocking serum for 45 minutes (Alexa Fluor 488 donkey anti-goat IgG, Alexa Fluor 488 goat anti-rabbit IgG, and Alexa Fluor 488 donkey anti-sheep IgG, Molecular Probes, Eugene, OR; normal donkey and normal goal blocking serum, Sigma). Following additional washes with TBST, sections were counterstained with 4′,6-diamidino-2-phenylindole (Molecular Probes, 1:2,000 dilution of 1 mg/mL stock in dH₂O) for 30 minutes. Sections were then washed in TBST, dehydrated in 100% ethanol, cleared in xylene, and mounted under coverslips with Permount. Images were captured with a Zeiss Axiocam HRm charge-coupled device camera connected to a Zeiss Axiovert 135 inverted fluorescence microscope equipped with a ×40 objective. For each fluorophore, all images were captured using equal camera integration times.

To identify genes that are regulated in proliferating endothelial cells relative to quiescent cells, we employed an in vitro angiogenesis model in which primary cultured microvascular endothelial cells were driven to proliferate from a quiescent state by incubation with growth factors. Primary HDMVECs or rat RHMVECs between passages 4 and 7 were grown in monolayers in tissue culture dishes and mitogen starved for 24 hours. The cells were then induced to proliferate by exposure to VEGF, bFGF, or ENDOGRO (a bFGF-rich bovine brain extract) or left untreated and quiescent (Fig. 1). We found that whereas growth medium supplemented with 10% FBS was not sufficient to drive endothelial cell proliferation 10% FBS plus additional growth factor induced rapid endothelial cell proliferation. VEGF-stimulated proliferation was selectively inhibited by simultaneous treatment with either of two KDR kinase inhibitors (11) that are 20- to 100-fold less active toward fibroblast growth factor receptors (FGFR) 1 and 2 (Table 1). These compounds had no effect on bFGF- or ENDOGRO-induced proliferation (Fig. 1).

We then compared the gene expression pattern from mitogen-starved, quiescent endothelial cells with the expression pattern in actively dividing cells. Cells were harvested for isolation of total RNA 24 hours following addition of VEGF, bFGF, or ENDOGRO. Matched control cultures and RNA samples were produced for each stimulation condition by treatment of an identical plate with growth medium not supplemented with growth factors. We found significant gene expression changes (P < 0.05) in proliferating HDMVEC and RHMVEC cultures (displayed graphically in Fig. 2). Tables with additional information about the genes shown in Fig. 2 are provided in the supplementary material. (Supplementary Tables 1 and 2).⁶

To determine if the growth factor–induced proliferation signatures that we observed were sensitive to KDR kinase inhibitors, we stimulated HDMVECs or RHMVECs with VEGF or bFGF for 24 hours in the presence of 100 nmol/L KDR kinase inhibitor B (Table 1). Consistent with our cell proliferation data, we found that the VEGF-induced gene expression signature was effectively suppressed by KDR kinase inhibitor B, whereas the bFGF-induced signature was unaffected (Fig. 3). VEGF binds to and activates the Fms-like tyrosine kinase (FLT1) and KDR (21, 22). KDR kinase inhibitors A and B inhibit both FLT1 and KDR (Table 1). bFGF binds to FGFR1 and FGFR2 but not to FLT1 or KDR. Both FGFR1 and FGFR2 are relatively insensitive to KDR kinase inhibitor B. Similar results were observed in parallel experiments done with RHMVECs (data not shown).

The previous experiments generated gene expression profiles from proliferating endothelial cells. However, the majority of genes regulated during endothelial cell proliferation will also be expressed in other types of proliferating cells (genes that regulate cell cycle and metabolic processes, for example). Human tumors contain a complex mixture of cell types, where ∼1 in 2,000 cells (0.05%) are proliferating endothelial cells.⁷

Therefore, in attempting to identify genes involved in endothelial cell proliferation, we were most interested in the endothelial cell–specific portion of the HDMVEC and RHMVEC proliferation signatures. Candidate endothelial cell–specific genes will be regulated during a proliferative response to mitogens in our in vitro experimental system but expressed at relatively low levels in nonendothelial cells. We used microarray intensity data, which correspond to the number of labeled cRNAs bound to each array feature and proportional to mRNA copy number, from previous expression profiling studies and compared it with the microarray intensity data from our HDMVEC proliferation experiments. Existing intensity data from a panel of actively growing tumor-derived cell lines (MOLT-4, HL-60, Raji, SW480, Daudi, G361, A549, K562, and MCF7) were used to remove from consideration those genes with endothelia cell/tumor microarray intensity ratios <3:1. We selected 702 HDMVEC gene sequences as endothelial cell specific in this manner. (Supplementary Table 3).⁶ We identified many known endothelial cell–specific genes by this method (i.e., ESM-1, KDR, and FLT1) as well as numerous novel sequences. In parallel, we obtained a measure of endothelial cell specificity for genes regulated in proliferating RHMVECs by comparing microarray intensity data from the RHMVEC experiments with data from gene expression profiling experiments with rat C6 glioma cells actively growing in culture. We identified 493 genes with RHMVEC/C6 intensity ratios >3:1 (Supplementary Table 4). ⁶

To validate the endothelial cell–specific proliferation signature in vivo in the context of a complex tumor, we employed two syngeneic rat tumor models and assessed the effects of small molecule KDR kinase inhibitors on tumor gene expression. The tumor models used C6 glioma and Mat B III mammary carcinoma cell lines both derived from Fischer 344 rats. These cell lines each secrete VEGF, do not express KDR, and form highly vascularized tumors that are sensitive to KDR kinase inhibitors.⁸

In the first animal model, C6 cells were injected s.c. into the right flank of rats and allowed to form tumors for 7 days. At that time, once-daily oral dosing with KDR kinase inhibitor A, KDR kinase inhibitor B, or vehicle commenced and continued for a total of 1, 2, or 3 days (Fig. 4A). Under the dosing schedule used, the achieved plasma concentrations led to complete suppression of KDR as measured by determination of phospho-KDR levels in tumor tissue (13).⁸ Genome-wide gene expression in tumors isolated from compound-treated animals was compared with gene expression from tumors isolated from vehicle-treated animals. We observed that both compounds A and B induced robust gene expression changes in tumor gene expression over multiple days, particularly after ≥48 hours of compound exposure (Fig. 5A and B; P < 0.05 for individual sequences).

In the second tumor model, Mat B III mammary adenocarcinoma cells were injected into a mammary fat pad of female rats. After allowing tumors to establish for 7 days, once-daily oral dosing of KDR kinase inhibitor A began and continued for a total of 5 days (Fig. 4B). When the pattern of tumor gene expression from compound-treated rats was compared with vehicle-treated controls, we again found significant differences (Fig. 5C; P < 0.05 for individual sequences). Although there was overlap in the gene expression changes induced by KDR kinase inhibition between the three studies, the majority of gene expression changes were study specific (Fig. 6A). This indicated that differences between the tumor cell types and differences in the kinase inhibitory profiles of compounds A and B accounted for the majority of gene expression changes observed in our tumor models.

In each of the three animal studies we did, we found that we could detect KDR kinase inhibitor-induced changes in expression for a fraction of those genes we had identified as specific to proliferating RHMVECs in culture (Fig. 6B; Supplementary Table 5).⁶ Most interestingly, we found in each study that some of those genes were regulated in a manner consistent with suppression of endothelial cell proliferation. In effect, these genes were oppositely regulated in our in vitro proliferation experiments compared with our in vivo tumor studies. In both cases, the genes were highly expressed when endothelial cells were proliferating and expressed at low levels under nonproliferating conditions. Thus, we identified endothelial cell–specific genes that were “oppositely regulated” in each of the three animal tumor studies and identified genes that were regulated as such in multiple studies (Fig. 6C; Supplementary Table 6).⁶

In addition to being new potential targets for antiangiogenesis therapy, these genes may also be used as biomarkers of KDR kinase inhibition. Imposing a requirement that genes to be considered as biomarkers should have compound-induced in vivo expression changes of at least 1.6-fold, we identified six genes that were “oppositely regulated” in both animal tumor studies with KDR kinase inhibitor A and two genes that were “oppositely regulated” in all three studies. Because an assay using these genes would be designed to measure the pharmacodynamic effects of KDR kinase inhibitor A in the clinic, we selected the six genes [ANGPT2, EDNRB, PLAU, CLU, fucosyltransferase-4 (FUT4), and IFN-induced protein with tetratricopeptide repeats 3 (IFIT3)] identified by both KDR kinase inhibitor A animal studies as potential biomarkers for tumor endothelial cell proliferation (Table 3). Each of these genes has been reported to be involved or implicated in endothelial cell function.

To confirm that the six genes we identified as potential biomarkers of tumor endothelial cell proliferation were specifically expressed in tumor vasculature and whose expression levels reflected endothelial cell proliferation rates, we did several validation experiments. First, we independently assessed gene expression levels in the animal tumor RNA samples by quantitative real-time PCR to confirm the microarray hybridization results. We found that the results obtained by real-time PCR closely matched those from the DNA microarrays (Fig. 7). We also measured the expression levels of the biomarker genes in an additional set of rat Mat B III tumors from a fourth animal study for which no gene expression profiling was done. In this independent study, animals with established Mat B III breast tumors (7 days after cell implantation) were dosed orally once daily with KDR kinase inhibitor A or vehicle for a total of 8 days. The expression levels of all genes changed as expected in response to exposure to the KDR kinase inhibitor.

To determine if expression of the biomarker genes was specific to the endothelial cells within tumors, we visualized their protein products in rat tumor tissue sections by immunofluorescence microscopy. Half of each tumor from our animal tumor studies was fixed and preserved for sectioning and immunohistochemistry as described in Materials and Methods. Antibodies are available commercially for four of the six biomarker protein products, and we were able to determine optimal conditions for immunofluorescence staining. Tumor sections were incubated with antibodies against a putative biomarker protein and with antibodies against the endothelial cell surface protein CD31 (to label the tumor vasculature). We found that the four proteins we examined (ANGPT2, clusterin, EDNRB, and PLAU) were each localized specifically to the tumor vasculature in Mat B III tumors (Fig. 8).

Finally, we correlated the confirmed expression changes of the biomarker genes in KDR kinase inhibitor A–treated tumors with an independent measure of tumor endothelial cell proliferation. Using a modification of the method described by Mundhenke et al. (23), we measured endothelial cell proliferation rates by double immunohistochemical staining of tumor sections for the endothelial cell marker CD31 and the nuclear proliferation marker Ki-67. We analyzed sections of C6 flank tumors from five vehicle-treated animals and five KDR kinase inhibitor A–treated animals (three doses of vehicle or compound over 72 hours). These tumors were the same as those used for our gene expression studies, so that we could determine if there was a correlation between changes in biomarker gene expression and endothelial cell proliferation. Using this modified immunohistochemical method, we determined the endothelial cell proliferation rate in tumors from vehicle-treated animals to be 34 ± 5%. In contrast, we determined that the endothelial cell proliferation rate was 19 ± 5% in tumors from animals treated with KDR kinase inhibitor A.

We have described the identification of a set of genes that are regulated both in vitro during mitogen-induced proliferation of primary microvascular endothelial cells and in vivo in response to systemic exposure to KDR kinase inhibitors. Changes in expression levels of these genes in response to inhibition of KDR are indicators of change in tumor endothelial cell proliferation rate. Beginning with a large set of genes shown to be regulated in vitro by mitogen-induced proliferation of primary endothelial cells, we identified a subset that was relatively specific to endothelial cells. We then employed two distinct syngeneic tumor models to show that exposure to KDR kinase inhibitors generates robust gene expression changes in established tumors. Gene expression changes consistent with inhibition of VEGF signaling and inhibition of endothelial cell proliferation were detected in tumors from each animal model.

Genes regulated by systemic exposure to KDR kinase inhibitors in at least two of the three tumor models were selected for further study. Changes in expression of these genes (as determined by microarray hybridization) were confirmed by quantitative real-time PCR both in tumors that were profiled and in tumors from an additional independent animal tumor study. We further validated the selected genes by correlating the compound-induced gene expression changes to compound-induced differences in proliferating tumor endothelial cell number as determined by immunohistochemical staining (again in the same rat tumors that were profiled). We also verified the endothelial cell specificity (in the context of our rat tumor models) of their expression by showing that their protein products were restricted to CD31-expressing cells. Some of these genes may prove to be attractive targets for future antiangiogenesis therapies. Experiments are currently under way with three genes that we found to be down-regulated in tumors by KDR kinase inhibition (FUT4, ANGPT2, and EDNRB) to investigate the effects of gene knockdown by RNA interference on endothelial cell proliferation and differentiation.

Although the expectation by random chance of identifying a gene that met each of our selection criteria was low, we identified six. We biased our gene selection toward endothelial cell–specific genes, but there was no guarantee that genes meeting our multiple criteria would have any known function in endothelial cells. Nevertheless, nearly all the genes identified have been implicated or shown to be directly involved in the regulation of endothelial cell function. The ANGPT2 protein (ANG2) is a well-characterized ligand for the Tie-2 receptor tyrosine kinase that functions in concert with VEGF and angiopoietin-1 to regulate vascular remodeling (24). ANGPT2 gene expression has been reported previously to be directly up-regulated by VEGF both in vivo and in vitro, consistent with our results (25).

The endothelin receptor B [EDNRB/ET(B)] is a seven-transmembrane, G protein-coupled receptor that is mutated in Waardenburg-Hirschsprung disease, a congenital malformation of neuronal ganglia in the hindgut (26). Most published studies of EDNRB describe its role in the neuronal system during neural crest development. However, it does control vasoconstriction and vascular cell proliferation induced by the endothelins, and EDNRB is overexpressed in primary melanomas (27). EDNRB antagonists have been reported to inhibit vascular cell proliferation and human melanoma cell growth in vitro and in vivo (28, 29).

FUT4 is an α1,3-fucosyltransferase involved in the synthesis of myeloglycan, the major physiologic binder of E-selectin (30). It is also involved in the synthesis of many other glycosylated proteins but is reported to be highly expressed in some tumors with inverse correlation to prognosis (31).

Clusterin is a secreted glycoprotein that seems to be overexpressed in apoptotic cells (32–34) but whose function is still largely unknown (32). Clusterin expression is antiproliferative (35) and down-regulated in advanced prostate cancer (36–38). Reduction in serum clusterin levels also correlates with esophageal squamous cell carcinoma tumorigenesis (39).

PLAU is a proteolytic enzyme that plays a critical role in angiogenesis, tumor invasion, and metastasis by contributing to remodeling of the extracellular matrix (40, 41). The effect of PLAU activity is the conversion of plasminogen to plasmin. It is unclear why we observe an increase in PLAU gene expression in tumors exposed to KDR kinase inhibitors rather than the decrease we would have expected to accompany a decrease in neovascularization. We can surmise that increased PLAU expression is a compensatory mechanism elicited by inhibition of the VEGF signaling pathway, but clearly, more investigation is required to determine the mechanism underlying our observations.

IFIT3 [also known as glucocorticoid-attenuated response gene-49 (GARG-49) and IFN-responsive gene 2 (IRG2)] is a gene that yet has no known function. Cloned from the mouse as part of studies to identify GARGs induced by lipopolysaccharide or IFN, the highly conserved tetratricopeptide repeat domains of IFIT3 are believed to mediate protein-protein interactions (42–45). No human orthologue of IFIT3 has been identified in human cells, but a homologous gene, designated IFT4, is 60% identical and 78% similar by protein sequence [BLASTP (46)].

Directly assessing the pharmacodynamics of antiangiogenesis therapeutics targeted to the VEGF signaling pathway is challenging. Inhibition of the KDR tyrosine protein kinase suppresses endothelial cell proliferation, but it is difficult to assess the rate of proliferation of these cells in vivo (47). KDR is not expressed at high levels in readily accessible biological materials, such as peripheral blood or bone marrow aspirates. Current pharmacodynamic assays for KDR kinase inhibition typically rely on surrogate protein kinase markers whose activity is also sensitive to the compound being evaluated (i.e., FLT3 tyrosine phosphorylation in many KDR kinase inhibitors) or on in vivo imaging techniques, such as dynamic contrast-enhanced magnetic resonance imaging, that can assess changes in vascular permeability. These methods have the disadvantage of being indirect measures of KDR function and endothelial cell proliferation.

One described method to assess in vivo endothelial cell proliferation involves dual immunohistochemical staining of tumor sections for the endothelial cell marker CD31 and the nuclear marker of cellular proliferation Ki-67 (23). Although this method is able to determine the fraction of proliferating endothelia cells, the experimental protocol is technically complex and the analysis required for each stained tumor section is lengthy.

In contrast, a gene expression assay for a few endothelial cell–specific genes could be done with relatively little effort using quantitative PCR. However, it remains to be shown that such an assay can distinguish human tumors with active angiogenesis from tumors containing mostly quiescent endothelial cells. The 39% endothelial cell proliferation rate observed in our untreated rat C6 gliosarcomas was significantly higher than the ∼5% proliferation rate observed in human breast tumor specimens or the 1% to 5% rate generally reported for other human tumor types (23).⁹

The elevated endothelial cell proliferation rate in rat tumors is consistent with the observation that C6 and Mat B III tumors grow much more rapidly than the average human tumor (48). However, higher endothelial cell replication rates have been reported in aggressive human cancers, such as noninflammatory breast tumors (11% proliferating endothelia cells) and hepatocellular carcinomas (35% proliferating endothelia cells; refs. 49, 50).

In the future, it may be possible to identify gene expression signatures that accurately reflect the proliferation rate of tumor cells or tumor responsiveness to anticancer therapies. Although additional studies would be required, our current results suggest that this is feasible. For example, in the data we collected for this study, interleukin-6, osteoprotegerin, and bone morphogenetic protein-2 were all consistently up-regulated in tumor tissue in response to KDR kinase inhibition. Conversely, pineal-specific PG25 protein and the potassium voltage-gated channel KCNE3 were consistently down-regulated. None of these genes were endothelial cell specific by our analyses.

In the present study, we were interested primarily in assessing the in vivo effects of a KDR kinase inhibitor on gene expression in tumor endothelial cells. However, the genes we have described may prove useful as a general pharmacodynamic readout for cancer therapies that inhibit proliferation of endothelial cells in tumor vasculature. Furthermore, the gene expression data we generated identified numerous uncharacterized genes that seem to be specifically expressed in endothelial cells.

Expression profiling-based monitoring of pharmacodynamic effects of cancer therapy has many benefits. Carefully designed, these assays have the potential to make dosing of antineoplastic agents more efficient, to identify patient populations most likely to benefit from specific therapies, and to reduce clinical development time of novel therapeutics. Each of these aspects will lead to increased tumor response rates and improved human health.