Tumour angiogenesis is a complex process based upon a sequence of interactions between tumour cells and endothelial cells. To model tumour/endothelial-cell interactions, we co-cultured U87 human glioma cells with human umbilical vein endothelial cells (HUVECs). U87 cells induced an`activated' phenotype in HUVECs, including an increase in proliferation,migration and net-like formation. Activation was observed in co-cultures where cells were in direct contact and physically separated, suggesting an important role for soluble factor(s) in the phenotypic and genotypic changes observed. Expressional profiling of tumour-activated endothelial cells was evaluated using cDNA arrays and confirmed by quantitative PCR. Matching pairs of receptors/ligands were found to be coordinately expressed, including TGFβRII with TGFβ3, FGFRII and cysteine-rich fibroblast growth factor receptor (CRF-1) with FGF7 and FGF12, CCR1, CCR3, CCR5 with RANTES and calcitronin receptor-like gene (CALCRL) with adrenomedullin. Consistent with cDNA array data, immunohistochemical staining of expressed proteins revealed the upregulation of Tie-2 receptor in vitro and in vivo. Our data suggest that tumour-induced activation of quiescent endothelial cells involves the expression of angiogenesis-related receptors and the induction of autocrine growth loops. We suggest that tumour cells release growth factors that induce endothelial cells to express specific ligands and their cognate receptors coordinately.
Introduction
The development of new blood vessels is essential for local tumour progression and the development of distant metastasis(Folkman, 1995; Paweletz and Knierim, 1989; Folkman and Shing, 1992; Fidler and Ellis, 1994). Tumour angiogenesis involves the degradation of the basement membrane by activated tissue or circulating endothelial precursors, migration of endothelial cells and coalescence into vascular tubes(Ausprunk and Folkman, 1977; Nicosia et al., 1982; Hanahan, 1988; Hanahan and Folkman, 1996). The individual steps in tumour angiogenesis have been identified as targets for therapies employing angiogenesis inhibitors alone or in combination with standard cancer treatments (Denekamp,1993; Denekamp and Hill,1991; Rak et al.,1995; Gastl et al.,1997; Kerbel,1991). Numerous investigations have focused on the tissue origin of tumour endothelial cells (Lin et al.,2000; Asahara et al.,1997) and the individual genes that promote the growth and survival of tumour vessels. Recently, serial analysis of gene expression(SAGE) has provided a preliminary analysis of the genes associated with tumour angiogenesis (St Croix et al.,2000). However, the nature of this analysis precludes quantitative evaluation of gene induction by tumour cells. Analysis of endothelial gene families and the interaction between specific genes and tumour-derived pro-angiogenic factors is not yet available
Tumour cells are proposed to mediate tumour angiogenesis by secreting soluble factors that enhance endothelial cell proliferation, migration and tube formation as well as by direct cellular interactions with endothelial cells. The interactions between tumour cells and endothelial cells are complex, might be analogous to those observed during embryonic development and have been hypothesized to be important in the process of tumour angiogenesis(Darland and D'Amore,1999).
In order to define intercellular interactions in detail, we modelled tumour angiogenesis by developing an in vitro co-culture system consisting of a human glioma cell line (U87MG) and human umbilical vein endothelial cells (HUVECs). Using this system, we investigated the following hypotheses.
Stimulation of endothelial cells by tumour cells establishes an endothelial phenotype consistent with the initial stages of angiogenesis. This phenotypic switch can be modelled by co-cultivation of tumour and endothelial cells.
This phenotypic switch is preceded or accompanied by transcriptional reprogramming of endothelial cells, which can be detected by expressional profiling. Detection of genes involved in reprogramming of endothelial cells by tumour cells is important to elucidate the molecular basis of tumour-induced angiogenesis.
Tumour-induced activation of endothelial cells includes the formation of paracrine loops with soluble ligands (cytokines and growth factors) secreted by tumour cells. It is likely that such activation also involves the formation of autocrine loops in endothelial cells.
In this report, we demonstrate that U87 cells promote phenotypic changes in HUVECs that include growth stimulation, activation of migration and morphogenetic changes including formation of net-like structures resembling neo-vasculature. We also detected expressional changes that parallel phenotypic changes of endothelial cells and show that these alterations are induced by U87-mediated soluble factors that might activate the formation of autocrine loops in the endothelial compartment. We were able to define several groups of genes that are consistent with observed phenotypic changes.
Materials and Methods
Cell culture and imaging
HUVECs (Clonetics) were maintained in EGM-2 medium (Walkersville, MD) at 37°C and 5% CO2. EGM-2 medium consists of EMB-2 with 1% foetal calf serum and other additives, including growth factors. In experiments where cells are starved, EBM-2 alone is used. The U87 cell line was transfected with green fluorescent protein (GFP) to aid in the discrimination of these cells. Transfection was performed by the standard method using LIPOFECTIN®reagent (Invitrogen, Carlsbad, CA) and the pEGFP-C1 vector system (Clontech,Palo Alto, CA). For growth curves, cells were cultured in six-well plates in duplicate. Cells were counted after trypsinization using a haemocytometer on a fluorescence microscope to discriminate between the colourless HUVECs and the green U87 cells. The microscope system was a Nikon Diaphot TMD inverted microscope adapted for fluorescence with an epifluorescence attachment and High-Q FITC filter cube. In experiments where cells were photographed on Transwell membranes (Corning Costar, Cambridge, MA), cells were imaged using a Nikon Eclipse E800 microscope with a Sony DXC-970 colour video camera and Image Pro Plus software (Media Cybernetics, Springs, MD).
Net-like formation assay
Six-well Transwell cell culture chamber (Costar) inserts with 0.4 μm pores were used with HUVECs plated in the inserts and GFP-U87 in the chambers(in duplicate) at a 1:10 HUVEC:GFP-U87 ratio. After 48 hours, the insert was removed and stained with the PROTOCOL Hema 3® stain set (Biochemical Sciences, Swedesboro, NJ) and photographed at 4× magnification. Completely enclosed spaces were counted to quantify net-like structures. Angiostatin and endostatin (100 ng ml–1) and herbimycin A (1μg ml–1) were purchased from Calbiochem (La Jolla, CA). Anti-VEGF antibody (100 ng ml–1) was purchased from R&D Systems (Minneapolis, MN).
Migration assay
24-well Transwell inserts with a 5 μm pore size were coated with a thin layer of collagen (Type I, rat tail, BD Biosciences, Bedford, MA). HUVECs were cultured in EGM-2 with 10% of the usual growth factors and serum + 0.1% bovine serum albumin for 18 hours before switching to media with no growth factors or serum for 4 hours before the start of the experiment. GFP-U87 cells or HUVECs were washed twice with EBM-2 alone, trypsinized and plated into the lower chambers at a ratio of one cell (in insert) to ten cells (in chamber cell number). HUVECs subject to serum starvation were plated on inserts and cultured for 18 hours before staining the membranes as described above. Cells were photographed at 10× magnification, and four fields each from duplicate samples were counted to quantify migration.
Transfer of activated HUVECs and induction of cytokines expression by tumour conditioned medium
TransWell® chambers with 0.4 μm pores were loaded with GFP-U87 cells or HUVECs in the inserts and HUVECs or EBM-2 only in the wells (see Fig. 6A). After 24 hours in culture, the inserts were removed and HUVECs in the wells were washed four times with EBM-2. Then, inserts that contained HUVECs cultured with only EBM-2 in the wells (acceptor HUVECs) were transferred to the wells with HUVECs cultured with U87 cells (activated HUVECs), HUVECs cultured with HUVECs (naive HUVECs) and HUVECs cultured with medium only (negative control). After 24 hours, cells in the inserts were stained as for the net-like formation assay and structures were counted and graphed as percent of negative control.
For induction and measurement of cytokine production, GFP-U87 cells and HUVECs were set up in T150 flasks as monocultures. At 80% confluence, all medium was removed, the cells were washed with EBM-2 and cultured in EBM-2 for 8 hours. Then the conditioned media from HUVECs or U87 was centrifuged (250 g for 5 minutes) to remove floating cells and placed on HUVEC for 18, 24 or 48 hours. The medium was centrifuged again and concentrated on a Speed-Vac from 20 ml to 0.7 ml. 100 μl of each (in duplicate) was assayed using the enzyme-linked immunosorbent serum assay (ELISA; R & D Systems,Minneapolis, MN) for RANTES or FGF-7. Concentrated EBM-2 alone was used as a blank and was also spiked with 25 pg of cytokine standard (22 pg was read by the assay for both cytokines) to ensure that salts in the medium did not interfere with the assay.
cDNA array data analysis
The data were filtered in two steps. First, the data were filtered by intensities using the receiver operating characteristic (ROC) method, which allows the estimation and control of the levels of false-positive and false-negative data based on the levels of signal detection(Pepe, 2000). To the analysis previously described, we applied cut-off levels of signal detection equal to 10% of average intensity Iav. Second, genes were selected based on the ratios of response in co-culture related to monoculture. For the present experiments, we used a twofold change as the cut-off level for ratios. As an alternative approach, when genes were not expressed in monoculture but were expressed in co-culture, we used subtraction of intensities in monoculture from the same intensities in co-culture. Subtracted differences were evaluated by Student's t test and statistically significant differences (P<0.05) were included in the subsequent analysis. For genes selected by P values, zero values were arbitrarily transformed in 100 units of intensity to estimate the ratio of response.
For data clustering, we exported normalized ratio values to JMP software(SAS Institute, Cary, NC) and performed multivariate analysis with hierarchical clustering based on the estimation of the Euclidian distances by Ward's method (Watson et al.,2001). Data were visually presented using TreeView software(http://rana.lbl.gov/EisenSoftware.htm)(Eisen et al., 1998). Out of 351 genes selected, 22 did not match current databases, 39 were not matched with known functions and 290 were assigned to specific functions. For functional classification, we used groups previously described by us(Khodarev et al., 2001) and others (Stanton et al., 2000)with modifications according to the Proteome database(http://www.proteome.com/databases/HumanPD/reports/385.html). Experiments were performed with Research Genetics GeneFilters® GF211 arrays (Research Genetics/Invitrogen, Carlsbad, CA) and two independent control co-cultures and monocultures, and were further confirmed by quantitative real-time PCR and/or independent hybridizations with alternative arrays Atlas® Human and Atlas® Human 1.2 [BD Biosciences/Clontech,Palo Alto, CA; see supplementary Table 1].
Quantitative PCR
cDNA was synthesized using Superscript II® reverse transcriptase(Invitrogen Life Technologies, Carlsbad, CA) following the manufacturer's recommendations, except that the addition of RNase inhibitor was omitted. cDNA was diluted 1:10 in sterile water. Quantitative PCR was performed on an ABI7700 (Applied Biosystems, Foster City, CA) using SYBR Green PCR reagents in a 50 μl reaction mixture containing 5 μl 10× SYBR Green PCR Buffer, 0.5 μl 10 mM primers, 4 μl dNTP mix, 6 μl 25 mM magnesium chloride, 0.5 μl AmpErase, 0.25 μl Amplitaq Gold and 5 μl of the 1:10 diluted cDNA synthesis reaction product. PCR was performed for 40 cycles at 95°C for 15 seconds and 60°C for one minute after initial incubations at 50°C for 2 minutes and 95°C for 10 minutes. PCR product specificity and purity were evaluated by generating a dissociation curve following the manufacturer's recommendations. Sample Ct values were normalized to Ct values for 18S RNA, all of which were calculated from triplicate reactions. Relative gene induction values were calculated following the manufacturer's recommendations.
Immunohistochemistry
U87 tumours were excised 17 days after implantation in the right hind limb of female nude mice (Frederick Cancer Research Institute, Frederick, MD). The care and use of experimental animals was in accordance with institutional guidelines. Excised tumours were placed in 10% neutral-buffered formalin,embedded in paraffin and sectioned at 6 μm thickness. Sections were deparaffinized and rehydrated through xylene and serial dilutions of ethanol to distilled water. Slides were incubated in antigen retrieval buffer (DAKO,Carpinteria, CA) pH 6.0 and heated in a microwave oven at 95°C for 15 minutes. After rinsing, slides were incubated in 3% hydrogen peroxide for 5 minutes and then 10% normal horse serum in 0.025% Triton X-100/PBS for 30 minutes. CD31 (1 μg ml–1, #1506, Santa Cruz Biotechnology,Santa Cruz, CA) and Tie-2 (5 μg ml–1, #AF313, R&D Systems, Minneapolis, MN) were applied on the slides for 1 hour at room temperature in a humidity chamber. Following TBS washing, slides were incubated with biotinylated secondary antibody (Vector Laboratories,Burlingame, CA) followed by ABC reagents (Vector Laboratories). Antigen-antibody complexes were detected using DAB substrate chromogen system with TBS-Tween 20 (DAKO) wash buffer. The entire procedure was performed using an automated staining chamber (DAKO). Slides were briefly immersed in haematoxylin for counterstaining and evaluated using light microscopy.
Quantification of Tie-2 receptor and CD31
For the quantification of the immunohistochemical staining, a colorimetric processing approach was employed (R. Yassari et al., unpublished). Briefly,multiple representative images from each slide were acquired using a Nikon Coolpix 955 camera with fixed optical parameters, light intensity and magnification. Acquired images were stored as TIFF files and evaluated using Image Processing Tool Kit 4.0 (Reindeer Games, Asheville, NC) and Adobe Photoshop 6. Areas of positive staining were selected using a fixed colour range across all images. The hue and saturation of the selected regions were then discarded and luminance used to represent staining intensity ranging from 0 to 255. A similar procedure was employed to perform background subtraction. We used CD31 immunostaining in membranes and tissues as controls. Significance(P<0.05) was determined with unpaired two-tail Student's t test.
Results
Stimulation of endothelial cell growth and morphological alterations following co-cultivation with tumour cells
U87 glioma cells were co-cultured with HUVECs to test the hypothesis that tumour cells stimulate endothelial cells to undergo changes that are characteristic of angiogenesis. When HUVECs were co-cultured with U87 cells in the ratio of 1:5, a significant increase (P=0.046) in HUVEC cell number was detected beginning at day 2 and continuing through day 6(Fig. 1A). The number of HUVECs in monoculture increased, reached a plateau at day 4 and then declined at days 5 and 6. This effect was dose dependent, such that higher U87:HUVEC ratios resulted in an increase in HUVEC cell number. (data not shown). No significant differences in growth kinetics were observed between U87 co-culture and U87 monoculture. These findings demonstrate that interactions between HUVECs and U87 cells in co-culture result in an increase in HUVEC number.
Growth stimulation of HUVECs was also examined in co-cultures in which a permeable membrane was placed between HUVECs and U87 cells (1:5 HUVEC:U87 ratio). At 48 hours, the number of HUVECs in co-culture was 2.0 times greater than the number of HUVECs in monoculture (10.7±2.2×104vs 5.3±0.3×104 cells; Fig. 1B). Data from these experiments suggest that activation of HUVEC growth is associated with soluble factor(s) produced by U87 tumour cells.
Co-cultivation of HUVECs with U87 cells also induced sequential morphological changes in both cell types. U87 cells began to aggregate and form a net-like pattern 5 hours after the initiation of co-culture(Fig. 1C,D). Simultaneously,the morphology of HUVECs began to change. The `teardrop-like' HUVECs observed in monoculture assumed a narrower extended shape as they began to align themselves with the U87 cells (Fig. 1D, inset, white arrow). By 12 hours, the U87 network was complete and spaces began to form between the boundaries of aligned HUVECs and U87 cells (Fig. 1E). At 24 hours,HUVECs formed net-like structures resembling a vascular network(Fig. 1F).
Co-culture of HUVECs with U87 cells activates HUVEC migration and net-like formation, which are suppressed by angiogenesis inhibitors
To investigate the effects of U87 cells on HUVEC motility and differentiation, we employed TransWell® cell culture chamber inserts to restrict interactions between cell types to soluble factors. At a 10:1 U87:HUVEC ratio, we observed an 80% increase in HUVEC migration in co-culture(347.3±6.1) compared with HUVEC monoculture (195.2±3.2, P<0.001; Fig. 2A).
Using TransWell® inserts without collagen coating and a pore size of 0.4 μm, HUVECs were cultured in the inserts with U87 cells in the chamber beneath at a U87:HUVEC ratio of 10:1. The presence of U87 cells increased the formation of net-like structures in the HUVEC compartment(Fig. 2C,D) similar to that described for Matrigel® migration assays (Donovan et al., 2001). The difference between co-culture (21±4) and monoculture (4±1) was statistically significant (P=0.027, Fig. 2B). We noted that the morphology of the endothelial cells involved in the formation of net-like structures also changed (Fig. 2E,F) In these regions, endothelial cells appeared elongated, with protruded cytoplasm and aligned themselves along the perimeter of the enclosed spaces (Fig. 2F, white arrow). These data suggest that tumour cells can induce endothelial cells to differentiate into structures that resemble in vivo neo-vascularization.
As shown in Fig. 2G,angiostatin (O'Reilly, 1997)and endostatin (O'Reilly et al.,1997), two reported inhibitors of in vivo angiogenesis, inhibit the formation of net-like structures. Inhibition of net-like formation in co-culture was also observed following exposure to Herbimycin A and anti-VEGF antibody, whereas no inhibition was detected in HUVEC monocultures (data not shown). These data demonstrate that angiogenesis inhibitors suppress the tumour-induced endothelial phenotype and that tumour cell/endothelial cell interactions are required for specific types of growth inhibition to occur.
Expressional profiling of tumour-activated endothelial cells reveals transcriptional reprogramming coordinated with phenotypic switch
An advantage of our in vitro co-culture system of U87 cells and HUVECs is the ability to evaluate molecular changes that are otherwise too complex to model in vivo. Genes associated with the phenotype of `tumour-activated'endothelial cells might include the expression of the tumour-specific/induced markers of neo-vascularization. To identify the genes associated with the`tumour-activated' phenotype, we compared the gene expression profiles of HUVECs co-cultured with U87 cells with HUVECs grown in monoculture. The complete dataset included 290 genes; 91 genes with consistent changes in at least two consecutive time points are presented in supplementary Table 1. Of these, 16 genes were selected for confirmation by quantitative real-time PCR. 14 of these genes (87.5%) were consistently upregulated by both DNA arrays and quantitative PCR. The functional group corresponding to cell structure/motility/extracellular matrix (ECM)-related genes represented 1.81 times more genes than random samples from the array with significant difference (P<0.0001). Three other functional groups directly correlated with observed phenotypic changes. They are genes for growth response or cell proliferation, for receptors, and for the soluble growth factors, cytokines and chemokines (listed as `ligands' in Fig. 3). In the cell shape/motility/ECM group, we observed significant transcriptional response in several members of collagen, keratin, tubulin and integrin gene families(Fig. 3). Among the receptor and ligand groups, several genes previously described as being involved in in vivo angiogenesis were detected. These genes include the receptors Tie-2,Flt-1 (VEGF RI) and FGFRII (Fig. 3). These findings are consistent with observed morphological changes in endothelial cells induced by tumour cells as reported above.
Tie-2 expression is increased in co-culture compared with monoculture
We validated our expression-profiling data by examining the expression of the Tie-2 receptor. Tie-2 is important for the stabilization and maturation of vessels, and has been identified as a marker of tumour angiogenesis(Yancopoulos et al., 2000). We found that, on the transcriptional level, the Tie-2 receptor was not expressed in endothelial cells grown in monoculture, but was expressed by endothelial cells grown in co-culture with increasing levels at 12 and 24 hours(Fig. 3).
To detect protein changes, we stained HUVECs attached to membranes from TransWell® chambers using antibodies to CD31 and Tie-2. Data were quantified as described in Materials and Methods. HUVECs grown in both monoculture and co-culture expressed high levels of CD31 membrane staining and did not show significant differences between mono- and co-cultures. Tie-2 staining was not detected in HUVECs grown in monoculture but was present in HUVECs grown in co-culture with U87 cells(Fig. 4A). Significant Tie-2 staining of endothelial cells was also observed in U87 xenografts(186.05±2.53) compared with endothelial cells in normal mouse brain(176.65±1.67, P<0.03)(Fig. 4B). No significant difference in CD31 staining was observed between U87 xenografts(189.28±0.72) and normal mouse brain (186.11±2.01, P=0.22). These data demonstrate that expressional changes detected in co-culture might correlate with tumours growing in vivo.
Tumour cells induce in endothelial cells the accumulation of mRNAs for receptors and ligands, and the expression of cytokines potentially involving autocrine loops
Our analysis of cDNA array data indicated increased expression of matching pairs of receptors and ligands in HUVECs during co-cultivation with U87 cells(Fig. 3). We evaluated changes in gene expression using quantitative PCR for various receptor-ligand pairs,including members of the FGF family of receptors and ligands (FGF-7/FGFRII and FGF12/CFR1), small inducible cytokine receptors and ligands (RANTES/CCR1,CCR3 and CCR5), CALCRL (CRGP type 1)/adrenomedullin and transforming growth factor(TGF) family members (TGFβRII/TGFβ3). Quantitative PCR data were consistent with cDNA array data (Fig. 5). Co-cultivation of HUVECs with U87 cells resulted in the accumulation of mRNAs for matching receptor and ligand genes. These receptor-ligand gene pairs might be involved in the formation of autocrine loops in tumour-stimulated endothelial cells, which has previously only been shown to occur in glioma cells (Westphal et al., 1997; Tada et al.,1994).
To investigate the functional significance of these observations we first hypothesized that, if tumour-activated endothelial cells express cytokines and growth factors with autocrine potential, such `activated' endothelial cells will (after primary contact with tumour cells) retain an ability to activate naive endothelial cells that have never been exposed to tumour cells. We used HUVECs plated in inserts of Transwell® chambers and cultivated in EBM as`acceptor' cells and negative controls(Fig. 6A). HUVECs loaded in the wells and co-cultivated with U87 were designated `activated' or donor cells and HUVECs co-cultivated with HUVECs only were designated `naive' cells. Transfer of acceptor HUVECs to `activated' HUVECs and subsequent co-culture for 24 hours led to a significant increase in the number of net-like structures compared with transfer to naive HUVECs(Fig. 6B). These data suggest that co-culture leads to the induction of HUVEC self-activating factors by HUVECs, which are released some time after the removal of tumour cells. Our second hypothesis proposes that ligands involved in the formation of tumour-induced autocrine loops will be expressed by HUVECs only after exposure to tumour cells or tumour-conditioned medium, and will not be expressed by HUVECs in monoculture. We investigated this possibility employing two ligands(RANTES and FGF7) based on DNA array data(Fig. 3). Fig. 6C shows that both of these ligands are significantly expressed in HUVEC monocultures only after stimulation by tumour-conditioned medium. These findings are consistent with expression profiling as described above. Taken together, these data suggest the presence of endothelial self-activating autocrine loops, which are induced following stimulation by tumour cells.
Discussion
Stimulation of endothelial cells by tumour cells establishes an endothelial phenotype consistent with the initial stages of angiogenesis
Angiogenesis is defined as the process of new vessel sprouting from pre-existing vessels (Ruoslahti and Rajotte, 2000). The induction of angiogenesis is required for tumour growth and the development of metastases(Yancopoulos et al., 2000; Carmeliet, 2000). Important components of angiogenesis include endothelial cell proliferation, migration,interaction with the ECM, morphological differentiation, cell adherence and tube formation (Paweletz and Knierim,1989; Denekamp,1993; Folkman,1995). The `angiogenic switch' hypothesis proposes that angiogenesis is induced and controlled by the relative balance of pro- and antiangiogenic factors present in the tumour microenvironment(Hanahan et al., 1996). These factors include different soluble ligands produced by tumours, ligands present in the ECM and molecules expressed on cell membranes that participate in direct cell-cell and/or cell-ECM interactions(Hanahan and Folkman, 1996). Although some of these ligands and molecules are known, it is likely that other tissue- and tumour-specific factors remain unidentified.
We described in this report an in vitro co-culture system that simulates direct and indirect interactions between tumour cells and endothelial cells. Our data show that our co-culture system recapitulates components of the angiogenic phenotype, including endothelial cell proliferation, migration and differentiation into the net-like structures. We also demonstrate that anti-angiogenic compounds inhibit the formation of net-like structures by endothelial cells grown in co-culture with tumour cells. Further investigations might detect new potential targets for therapeutic intervention and provide an approach for screening of compounds with anti-angiogenic potential.
Tumour-induced phenotypic changes coordinate with transcriptional reprogramming of endothelial cells, which can be detected by expression profiling
One of the major advantages of our co-culture system is the ability to investigate the molecular changes that occur as a result of tumour cell/endothelial cell interactions. Although the `cross-talk' between cell types might be bidirectional, in this study, we limited our investigations to tumour-cell-induced changes in endothelial cell gene expression. Our data demonstrate significant differences in the expression of endothelial genes,which coordinate with phenotypic changes observed in HUVECs co-cultured with U87 cells, with the cell structure/motility/ECM, receptors, ligands and cell proliferation genes as major response groups(Fig. 3; supplemental Table 1). These observations on the molecular level are consistent with the phenotypic features we observed.
Several independent laboratories have reported differences in endothelial cell gene expression during Matrigel-induced differentiation into the net-like structures in vitro. Differential gene expression has been assessed using suppression subtraction hybridization (SSH)(Glienke et al., 2000) or gene-calling techniques (Kahn et al.,2000). Also, SAGE was applied for investigation of gene expression in endothelium (St Croix, 2000; Lal, et al., 1999). These methods and models differ from our approach. However, in spite of the differences in techniques, model systems, and methods, some commonalities exist between our data and the data reported by others. For example, we detected changes in the transcriptional response of bone morphogenetic proteins (BMP) 6 and 8, as well as BMP receptor II. BMP6 was found to be a marker of tube formation(Glienke et al., 2000),whereas BMP1 was identified as a possible tumour endothelial marker (TEM) in vivo (St Croix et al., 2000). BMP receptor II was reported to be upregulated during endothelial cell differentiation in Matrigel (Glienke et al., 2000). SAGE databases identified overexpression of angiogenesis-related genes including VEGF, fibronectin and adrenomedulin in glioblastoma, compared with normal brain tissue(Lal et al., 1999). SAGE analysis was also used to evaluate transcriptional responses to MCP1 in endothelial cells treated with macrophage-conditioned media(de Waard et al., 1999). SSH was used to compare differences in expressional profiles between freshly isolated high endothelial venule endothelial cells and high endothelial venule endothelial cells cultivated in vitro(Girard et al., 1999). Differential gene expression was detected for thrombospondin-1, proteases ADAMTS1 and ADAMTS4, fibronectin, integrins α2, α5 and αV,and receptors Tie-2, Flt-1 and BMPRII. In cDNA experiments, temporal co-clustering of serum-responsive genes revealed co-clustering of VEGF with FGF2, ICAM-1, MCP1 and FGF7 (Iyer et al.,1999). Collagen IVα2 was identified as a pan-endothelial marker in vivo using SAGE analysis (St Croix et al., 2000) while collagen IV was detected in differentiating endothelial cells in collagen gels using GeneCalling(Kahn et al., 2000). Another isoform, collagen V, was detected in Matrigel-induced endothelial cells using the SSH technique (Glienke et al.,2000). Keratin K7 and integrin α2 were reported to be upregulated in differentiating endothelial cells(Kahn et al., 2000) as was integrin α5 (Glienke et al.,2000). Therefore, comparison of our dataset with those published by other investigators indicates the presence of several gene groups that are common to all models. These include components of ECM and ECM-related signalling molecules (integrins, collagens, keratins, proteases, cellular adhesion molecules), several cytokines and growth factors (VEGF, FGF, BMPs,MCP1), receptors and signalling molecules. Expansion of endothelial cell expression databases and meta-analysis will elucidate the gene groups, gene families and pathways associated with general angiogenesis and tumour-specific angiogenesis.
Tumour-induced activation involves the induction of autocrine loops in endothelial cells
Our data suggest that tumour-induced transcriptional reprogramming of endothelial cells might involve the establishment of autocrine loops in endothelial cells. Examples include coordinated transcriptional expression of FGF receptors and FGF proteins, CCR receptors 1, 3 and 5 and their putative ligand RANTES, TGFβRII receptor and TGFβ3, and CALCRL (CGRP1)receptor and adrenomedullin (Fig. 5). Consistent with these DNA array data, we found that tumour cells can induce the release of self-activating factors by endothelial cells. Examples of such factors include RANTES and FGF7, which are not expressed by HUVECs in monoculture but are significantly expressed following exposure to tumour-conditioned medium. Additional experiments are required to evaluate different receptors and ligands potentially involved in tumour-induced endothelial autocrine loops. However, our data indicate the existence of such loops and provide a system for further investigations. It has been recently suggested that FGF2 transfected murine aortic endothelial (MAE) cells, which constitutively express FGFRII, might be stimulated through formation of autocrine loops (Dell'Era et al.,2001). It has also been proposed that the Streptococcus pneumoniae cell wall (PCW) can induce autocrine loops involving tumour necrosis factor α in cerebral endothelial cells(Freyer et al., 1999). TGFβ3 has been shown to regulate the differentiation of embryonic cardiac endothelial cells (Potts et al.,1991). Interestingly, transcriptional upregulation of TGFβ3 was found in Kaposi's sarcoma associated herpes virus (KSHV)-infected dermal microvascular endothelial cells during the transformation from cobblestone-shaped cells to spindle-shaped cells, which are characteristic for Kaposi's sarcoma lesions (Poole et al.,2002). Adrenomedullin has been demonstrated to be an autocrine regulator of proliferation in human endometrial endothelial cells during endometrial angiogenesis (Nikitenko et al., 2000). Adrenomedullin signalling is induced via its interaction with CALCRL (CGRP1) receptor and downstream activation of adenylate-cyclase/protein-kinase-A-dependent cascades(Belloni et al., 2001). Therefore, some pathways of autocrine regulation of endothelial cells are described in the literature, but induction of endothelial autocrine loops by tumour cells has yet to be described and characterized.
In summary, the experimental system reported here, may provide useful tools for further detailed examination of the interactions between tumour and endothelial cells.
Acknowledgements
This work was supported in part by NIH/NCI grant CA71933-05 (R.R.W.). We are grateful to Samuel Hellman and Akira Imamoto for helpful discussions of this manuscript.