INTRODUCTION

In order to react appropriately to environmental changes individual cells in multicellular organisms have to co-ordinate their behavior. Complex signalling mechanisms have evolved to meet this need. Among other mechanisms the cross-talk between cells is accomplished by signalling molecules.

A simple model of communication between two cells is depicted in Figure 1. A producer cell sends a message by secreting a signalling molecule. When the molecule reaches the target cell, it binds to a receptor and the target cell responds by changing its behavior.

The ability of a signalling molecule to deliver a particular message to a distinct subpopulation of cells - its specificity - is not only determined by the molecule itself, but also by its receptor and the intracellular signal transduction. The same signalling molecule can convey different messages depending on the receptor type to which it binds on or in the target cell. Further, the same receptor can transduce different messages depending on the intracellular set-up of transduction pathways, which can differ from one cell type to another.

2.1 Growth factors and their receptors

Cytokines are secreted proteins with a signalling function. After being secreted, they diffuse in the extracellular matrix and act mainly locally in a paracrine manner, unlike hormones, which are distributed through the circulation. The classification of cytokines into "classical" growth factors, colony-stimulating factors, interleukins, lymphokines, monokines and interferons is rather a left-over from the history of their discovery than a classification according to producer cells and target cells (Nicola, 1994). Many growth factors are rich in extended -sheets characteristic for group 2 cytokines. Binding of a growth factor to its corresponding cell surface receptor activates complex multistep signal transduction pathways, involving changes in protein phosphorylation, ion fluxes, metabolism, gene expression, protein synthesis and ultimately a biological response (Nicola, 1994). VEGF and its close relatives VEGF-B and VEGF-C form a subfamily within the PDGF family of growth factors (Joukov et al., 1996), which itself belongs to the cystine knot class of cytokines (McDonald and Hendrickson, 1993).

The members of the PDGF family bind to receptors of the class III tyrosine kinase family, that contain multiple immunoglobulin domains in the extracellular region and a split kinase domain in their cytoplasmic part (Ullrich and Schlessinger, 1990). Receptors of other growth factors include class IV tyrosine kinase receptors (e.g. used by fibroblast growth factors; Basilico and Moscatelli, 1992) and class V tyrosine kinase receptors (e.g. used by nerve growth factors; Barbacid, 1993) and serine/threonine kinase receptors (e.g. used by TGF-; Wang et al., 1991).

Defects in the signalling pathways of growth factors have been connected to pathophysiological processes that involve uncontrolled cell proliferation, e.g. tumorigenesis (Bishop, 1991). The factor itself can be affected (Cross and Dexter, 1991), its receptor (Yarden and Ullrich, 1988) or the intracellular signal transduction pathway (Cantley et al., 1991).

2.2 Interaction between RTKs and their ligands

The receptors for growth factors are transmembrane proteins, whose activation follows the universal mechanism of receptor activation by oligomerization: Upon binding of the ligand, receptors dimerize or oligomerize inducing a conformational change into their active form, that allows the receptor to transmit the signal through the lipid bilayer of the plasma membrane (Lemmon and Schlessinger, 1994). Dimerization of class III tyrosine kinase receptors is thought to result from the dimeric nature of the ligand. So far more then 100 growth factors and 50 RTKs have been cloned and classified (reviewed in Nicola, 1994; van der Geer et al., 1994). A representative overview on RTKs being receptors for growth factors is given in Figure 2 (adapted from Pajusola, 1996). A relationship can be observed between the structure of a growth factor and the type of receptor to which it binds. Nerve growth factors bind to class v tyrosine kinase receptors, members of the PDGF family bind to class III tyrosine kinase receptors, and -trefoil growth factors like FGFs bind to class IV tyrosine kinase receptors and require heparin-bound FGF for dimerization and high-affinity binding.

Ligand binding and receptor dimerization induce the cross-phosphorylation of critical tyrosine residues, which are present in the cytoplasmic domain of the RTK (Honegger et al., 1990). Depending on the receptor type, distinct sets of cytoplasmic tyrosine residues are targets for auto-phosphorylation (reviewed in Carpenter, 1992). These phosphorylation events induce a conformational change within the RTK and create high affinity docking sites for cellular substrates like SH2-domain-containing molecules. SH2-domain-containing molecules bind specifically to certain phosphorylated tyrosines of the RTK recognizing three amino acids immediately C-terminal of the tyrosine residue (reviewed in Koch et al., 1991). Additionally, the kinase activity towards cellular substrates is increased upon phosphorylation, e.g. EGF phosphorylates the SH2-domain-containing phospholipase C-1 (PLC-1; Anderson et al., 1990) and the p85 subunit of phosphatidyl inositol-3-kinase (PI3-K; Hu et al., 1992). Not only the activation by phosphorylation increases the activity of PLC-1 and PI3-K, but also the recruitment from the cytoplasm to the plasma membrane, where the substrates of these enzymes are located (Pawson, 1995).

SH3 domain. SH3-domain-containing molecules have not been shown to interact directly with RTKs, but they are involved inmost signalling pathways of RTKs as adapter molecules (reviewed in Cohen et al., 1995a). An example is the interaction of the SH3-domain of GRB2 with PXXP-strings of the ras-guanine-nucleotide-exchange factor son of sevenless (SOS; Buday and Downward, 1993). SOS replaces the ras-bound GDP by GTP and thus activates the ras/raf/MAP kinase pathway.

Pleckstrin homology. Another homology domain involved in membrane localization and common to signalling molecules is the pleckstrin homology (PH) domain (Haslam et al., 1993), which has been shown to bind to the subunits of heterotrimeric G proteins (G), and to activate together with PIP2 the -adrenergic receptor kinase (-ARK; Pitcher et al., 1995).

2.3 The vascular system

2.3.1 The development of the vascular system

2.3.1.1 The cardiovascular system

The development of the cardiovascular system starts soon after gastrulation, and it is the first organ that starts functioning in a developing embryo (Gilbert, 1991). The inner lining of all blood and lymphatic vessels, as well as the heart endocardium, are composed of endothelial cells, which form a single cell layer. The embryonic vascular development can be divided into three phases[1]:

1. Differentiation of certain mesodermal cells into angioblasts
2. Vasculogenesis[2] (differentiation of endothelial cells in situ from angioblasts)
3. Angiogenesis[3] (vascular sprouting from pre-existing vessels)

Differentiation of mesodermal cells into angioblasts. In the splanchnic mesoderm of the embryo, a distinct set of cells differentiates into angioblasts (also called hemangioblasts), which are the precursors for both endothelial cells and blood cells (His, 1900). Evidence has been presented that the mesodermal cells which become angioblasts are determined already before gastrulation, because the blocking of DNA replication and morphogenetic movements during gastrulation did not prevent the emergence of endothelial cells (von Kirschhofer et al., 1994).

Vasculogenesis. The earliest blood vessels of an embryo form by a mechanism called vasculogenesis, which is defined as differentiation of endothelial cells in situ from angioblasts (Risau et al., 1988). Angioblasts aggregate and form blood islands. Cells at the periphery of the blood island differentiate into endothelial cells; cells in the center of the blood island differentiate into blood cells.

Heart and major blood vessels. Simultaneously with the development of the blood islands, angioblasts start to differentiate from mesodermal mesenchymal cells forming the early heart rudiment and the major blood vessels, such as the paired dorsal aortae (Coffin et al., 1991; Coffin and Poole, 1988; Pardanaud et al., 1987).

The vasculature of certain internal organs of endodermal origin (e.g. lung, spleen and pancreas) forms through a similar vasculogenic mechanism (Auerbach et al., 1985), which was demonstrated e.g. by studies on quail-chicken chimeras (Pardanaud et al., 1989).

Angiogenesis. Angiogenesis, the sprouting of new blood vessels from pre-existent vessels, is responsible for the vascularization of certain other developing organs, e.g. brain, kidney and limbs (Bär, 1980; Ekblom et al., 1982; Wilson, 1983) and neovascularization in adults. New capillaries are formed by sprouting from existing capillaries or postcapillary venules, hence involving the proliferation of the existing endothelium in response to an angiogenic stimulus elicited by the tissue to become vascularized. Since sprouting (that is angiogenesis) is the most frequent, but not the only mechanism, by which new capillaries are derived from pre-existing ones, an alternative terminology was proposed by Sherer (Sherer, 1991). Sherer distinguished between angioblastic development (vasculogenesis) and angiotrophic growth, which employs three different mechanisms:sprouting, intussusceptive growth (a large sinusoidal capillary divides into two smaller capillaries[4]) and intercalation (endothelial cells divide within a blood vessel wall to increase its length and diameter; Folkman, 1987).

The process of sprouting has been divided into five sequential steps (Folkman, 1985) as outlined below:

1. Local degradation of the vascular basement membrane. The importance of proteolytic activity increases with the development of the endothelial basal lamina and is certainly crucial in adult angiogenesis. Capillary endothelial cells are capable of secreting several proteases, which hydrolyze the basal lamina of endothelial cells and certain ECM components (Moscatelli and Rifkin, 1988).

2. Endothelial cells migrate along angiogenic gradients. The extremely high migratory potential of endothelial cells during development has been shown in quail-chick grafting experiments (Wilms et al., 1991). Gradients of soluble angiogenic factors, ECM-bound angiogenic factors and the specific composition of the ECM have been identified as key factors in the directed migration of both embryonic and adult endothelial cells (reviewed in Wilting and Christ, 1996).

3. Lumen formation. Two opposite mechanisms for lumen formation have been described, inter- and intracellular lumen formation. Intercellular lumen formation is the extension of the lumen of the pre-existing capillary (Benninghoff et al., 1930). Of minor importance seems to be the intracellular lumen formation as a result of vacuolation (Sabin, 1920). The specific contributions of these two mechanisms to angiogenesis are not settled (Wilting and Christ, 1996).

4. Mitosis of endothelial cells in the midsection of capillary sprouts. Although vessel formation seems to be possible without mitosis, it was restricted to at least two levels of vascular loops in experiments, that employed endothelial cells, whose capacity to proliferate was destroyed by irradiation/poisoning (Koolwijk et al., 1996; Sholley et al., 1984). Whether under physiological conditions the initial steps of angiogenesis take also place without concurrent mitosis of the involved endothelial cells is still subject of discussion.

5. Fusion of adjacent sprouts and loop formation. It is unknown, whether individual capillary sprouts find each other by a directed mechanism or by trial-and-error migration (Folkman, 1985). A directed mechanism would involve concentration gradients and/or adhesion molecules and, indeed, several candidate molecules have been proposed, e.g. Tek/Tie and their ligands (Dumont et al., 1994).

Hyperpermeability. Microvascular hyperpermeability regularly accompanies angiogenesis, and also seems to play a role in its induction. Fenestration and transcytosis are increased in hyperpermeable vessels (Karnovsky et al., 1967; Kohn et al., 1992; Roberts and Palade, 1995), enabling plasma proteins to leak and to form a new provisional ECM by extravascular clotting, which permits, favors or even induces inward migration of endothelial cells (Dvorak et al., 1995). The mature ECM, by which the provisional ECM is gradually replaced, subsequently would suppress angiogenesis. Notably several markers of vascular maturation inhibit the proliferation of endothelial cells, e.g. laminin (Kubota et al., 1988; Risau, 1991).

ECM in the development of the vasculature. Migration of the early angioblasts and endothelial cells is essential for the proper development of the vasculature. Several ECM components are endothelial cell-derived, and endothelial cells possess multiple receptors for ECM components, such as integrin receptors for types-I and -IV collage, laminin, fibronectin[5] and vitronectin (van Mourik et al., 1990). For example, blood islands and early intraembryonic capillaries secrete fibronectin, whose expression is replaced later by laminin (Risau and Lemmon, 1988). Precardiac mesodermal cells, which later differentiate into endocardial endothelial cells, have been shown to migrate along a fibronectin gradient (Linask and Lash, 1986). Laminin may be involved in the inhibition of proliferation and induction of differentiation of the endothelial cells (Kubota et al., 1988). This illustrates the interdependence between ECM components and the migrational and proliferative behavior of angioblasts and endothelial cells (Grant et al., 1990).

Many angiogenic factors are able to interact with heparin and heparan sulphates in both free and cell-associated forms, and they seem to play important roles in the regulation of angiogenesis. Certain angiogenic molecules like FGFs require interaction with heparin or heparan sulphate for receptor binding (Yayon et al., 1991). Cell-associated and extracellular-matrix heparan sulphate proteoglycans (ECM-HSPGs) appear to provide sustained release reservoirs for angiogenic molecules having affinity for heparin like VEGF (Ferrara et al., 1992) and FGFs (Baird and Ling, 1987), and heparin binding can protect such molecules from degradation. The success of the heparin mimic aluminium sucrose octasulphate (sucralfate) in the treatment of gastric ulcers may be partly based on its ability to protect endogenous bFGF from degradation by gastric acid. The healing of the peptic ulcers seems to depend on the angiogenic activity mediated by the elevated level of endogenous bFGF (Folkman et al., 1991).

2.3.1.2 The lymphatic vascular system

The lymphatic vascular system starts to develop shortly after the blood circulation has started, but not much is known about its origin. The theory of a venous origin with a centrifugal spread was first proposed by Sabin (Sabin, 1902; Sabin, 1909; Sabin, 1912). According to this theory, lymphatic vessels arise by sprouting from early lymphatic sacs, which themselves develop from large central veins by sprouting at certain locations (around E 12 of mouse development). This theory is supported by the expression pattern of FLT4 (Kaipainen et al., 1995). Around E 15 lymphatic sprouting from the veins ceases, but extensive growth into the periphery continues. Vessels fuse and remodelling continues before the mature vasculature is formed. This includes the regression of earlier capillaries, possibly by apoptosis (Risau, 1995). Except for the left thoracic duct all links to the venous system are disconnected (Zadvinskis et al., 1992).

Lymphatic capillaries differ from blood capillaries. They are blind-ended, possess generally a larger and more irregular lumen and no continuous basement membrane. Lymphatic endothelial cells show no tight junctions (but instead many patent junctions and "complex adherents"; Schmelz et al., 1994), resulting in a generally higher permeability compared to blood capillaries. Via unique anchoring filaments they are connected to the adjacent connective tissue (Leak, 1970).

Although almost no molecular biological research has been done on lymphangiogenesis, the sequential model for the angiogenesis of blood vessels might probably be adapted for the sprouting of lymphatic vessels as well. However, differences between blood and lymphatic vessels should be accommodated to such model; for example, no basal membrane has to be degraded in lymphangiogenesis, and loop formation is not required.

The present work suggests that lymphangiogenesis may involve at least two distinct signals for the establishment of a lymphatic capillary network: one signal that would stimulate migration/sprouting and another, distinct signal, that would stimulate the proliferation of the involved endothelium (see 5.1.6).

It has been proposed that also the angiogenesis of blood vessels is dependent on two distinct signals (Folkman, 1984). Anyhow, angiogenesis of blood vessels inevitably seems to involve VEGF, a molecule that carries signals for both migration/sprouting AND proliferation, yet other molecular inducers of angiogenesis might deliver only one signal (migration/sprouting OR proliferation).

2.3.1.3 Diversity of endothelial cells

Endothelial cells are engaged in a variety of functions: endothelial cells in large vessels participate in the control of vascular tone and blood pressure, whereas capillary endothelial cells are mediating the exchange of gas and metabolic products between blood and tissues, as well as the extravasation of immune cells during infection (Geneser, 1986; Stevens and Lowe, 1992).

The capillary endothelial cells therefore show significant heterogeneity. The degree of permeability of different capillaries depends on the type of intercellular junctions, as well as on the cell morphology itself, and determines the molecular weight cut-off in the separation of blood/lymph and tissue. The liver sinusoids contain a discontinuous endothelium with large holes through the cytoplasm, whereas in the brain the capillary endothelial cells are linked to each other by tight junctions and contain only few transport vesicles thus creating the blood-brain barrier (BBB; Dejana et al., 1995; Geneser, 1986; Stewart and Hayakawa, 1987). Fenestrated capillaries consist of endothelial cells with openings that are closed by a thin diaphragm. These kind of capillaries are found, for example, in the choroid plexus, whose capillaries are responsible for the secretion of the cerebrospinal fluid of the brain ventricles. The high endothelial venules (HEVs) contain endothelial cells with a characteristic morphology. They are found in lymphatic organs and are responsible for the extravasation of mature lymphocytes during their circulation from blood to lymph (Risau, 1995). Leucocytes as well as macrophage extravasation requires various interactions between selectins, integrins and other adhesion molecules, and their respective ligands, expressed on both the leukocytes and the endothelium of the venules and lymphatic capillaries (Tedder et al., 1995). Lymphatic vessels collect and return extra plasma proteins found in the tissue fluid back to the circulation, regulate cell hydration and osmosis. In the immune response, lymphatic vessels provide for macrophages and Langerhans cells an exit pathway from tissues, and later immunoglobulins - produced in the lymph nodes - reach the blood via lymphatic vessels (Geneser, 1986).

2.3.2 Regulation of angiogenesis: Stimulatory and inhibitory factors

Physiological angiogenesis is regulated by a balance involving an interplay of positive and negative angiogenic molecules. Although stimulatory and especially inhibitory angiogenic factors have been almost exclusively discovered in the past decade, their number is already very large. They can be categorized into directly and indirectly acting factors. A factor is classified as indirect if it fails to stimulate endothelial cells in vitro despite its angiogenic activity in vivo. Additionally, indirect angiogenic factors may not only fail to stimulate endothelial cell proliferation in vitro, but actually inhibit it, e.g. TGF- (Baird and Durkin, 1986; Müller et al., 1987).

This apparent paradox was resolved by demonstrating that TGF- is a strong chemoattractant and activator for monocytes/macrophages (Wahl et al., 1987), which in turn are able to secrete a variety of inflammatory and angiogenic molecules including VEGF and TNF- (Berse et al., 1992; Leibovich et al., 1987), and that TGF- induces VEGF and bFGF in fibroblasts and epithelial cells (schematically shown in Figure 4; Pertovaara et al., 1994; Pertovaara et al., 1993).

Figure 4. Example of indirect angiogenic activity: TGF-

Another indirect mechanism is the release of sequestered angiogenic factors by proteolysis. This was shown for cell surface and ECM-bound VEGF (Houck et al., 1992; Park et al., 1993). Inhibitors of angiogenesis often seem to be parts of larger molecules and become activated upon proteolytic cleavage. Interestingly enough, many abundant proteins contain these "cryptic" angiogenic inhibitors. Angiostatin, a potent inhibitor of angiogenesis, is a proteolytic 38-kDa fragment of plasminogen (O'Reilly et al., 1994). Similarly, fibronectin cleavage generates two endothelial cell growth inhibitors (Homandberg et al., 1985). Other recently identified "cryptic" angiogenic inhibitors are collagen-derived and EGF-derived peptides (Nelson et al., 1995; Tolsma et al., 1993), and the moderate anti-angiogenic activity of PF4 can be increased more than 50-fold by proteolytic cleavage (Gupta et al., 1995). The mechanism of rapid generation of large amounts of angiogenic inducers and inhibitors from inactive precursors reflects the requirement for instant upregulation of angiogenesis upon demand as well as for a rigid temporal and spatial limitation, similar to the blood clotting cascade (Furie and Furie, 1988).

Nevertheless, directly acting angiogenic factors may also actin vivo indirectly. It is presently unknown how e.g. FGF elicits angiogenesis in vivo. aFGF and bFGF are strong stimulators of vascular endothelial cell growth in vitro (Lobb et al., 1985; Montesano et al., 1986; Shing et al., 1985; Thomas et al., 1985), but may have only little direct effects in vivo, especially since FGF receptors are not specific to endothelial cells and indirect mechanisms have been described (Goto et al., 1993; Pepper et al., 1992).

The mechanisms underlying the release of angiogenic inducers and inhibitors and of the interactions among them, as well as the integration of their signals inside the endothelial cells, are only beginning to be uncovered. Table 1 gives an overview of well known endogenous angiogenic factors (reviewed in Folkman, 1995b; Klagsbrun and D'Amore, 1991) and a selection of less well characterized ones. Table 2 lists several anti-angiogenic substances including some drugs (reviewed in Auerbach and Auerbach, 1994; Klagsbrun and D'Amore, 1991).

Table 1. Angiogenic factors

polypeptides
acidic fibroblast growth factor (aFGF)	(Lobb et al., 1985; Thomas et al., 1985)
basic fibroblast growth factor (bFGF)	(Montesano et al., 1986; Shing et al., 1985)
angiogenin[6]	(Fett et al., 1985)
vascular endothelial growth factor (VEGF)	(Connolly et al., 1989b; Leung et al., 1989; Plouët et al., 1989)
placental growth factor (PlGF)	(Maglione et al., 1991)
interleukin-8 (IL-8)	(Koch et al., 1992; Strieter et al., 1992)
platelet-activating factor (PAF)	(Montrucchio et al., 1994)
platelet-derived endothelial cell growth factor (PD-ECGF)[7]	(Ishikawa et al., 1989)
granulocyte colony-stimulating factor (G-CSF)	(Bocchietto et al., 1993)
hepatocyte growth factor (HGF)	(Bussolino et al., 1992; Grant et al., 1993a; Rosen et al., 1993)
proliferin	(Jackson et al., 1994)
insulin-like growth factor I (IGF-I)	(Grant et al., 1993c)
epidermal growth factor (EGF)	(Gospodarowicz et al., 1979)
transforming growth factor TGF-	(Schreiber et al., 1986)
transforming growth factor TGF-	(Roberts et al., 1986)
platelet-derived growth factor (PDGF)	(Risau et al., 1992)
tumour necrosis factor TNF-	(Fràter-Schröder et al., 1987; Leibovich et al., 1987)
non-polypeptides
prostaglandins: PGE1, PGE2	(BenEzra, 1978; Form and Auerbach, 1983; Graeber et al., 1990)
nicotinamide	(Kull et al., 1987)
adenosine	(Dusseau et al., 1986; Fraser et al., 1979)
1-butyryl glycerol	(Dobson et al., 1990)
degradation products of hyaluronic acid	(West et al., 1985)
(12R)-hydroxyeicosatrienoic acid	(Masferrer et al., 1991)
okadaic acid	(Oikawa et al., 1992)
Cu2+ complexed to Gly-His-Lys, ceruloplasmin, heparin or SPARC-derived peptides	(Lane et al., 1994; Raju et al., 1984)
GD3, GM1 (gangliosides)	(Ziche et al., 1992)

Table 2. Anti-angiogenic factors

polypeptides
angiostatin (38-kDa fragment of plasminogen)	(O'Reilly et al., 1994)
29-kDa N-terminal and 40-kDa C-terminal fibronectin fragments	(Homandberg et al., 1985)
platelet factor 4 (PF4)/PF4 fragment	(Gupta et al., 1995; Maione et al., 1990)
16-kDa N-terminal prolactin fragment	(Clapp et al., 1993)
protamine	(Taylor and Folkman, 1982)
thrombospondin-1 (TSP-1)	(Iruela-Arispe et al., 1991; Rastinejad et al., 1989)
somatostatin analogue: octreotide	(Grant et al., 1993b)
tissue inhibitors of metalloproteinases: TIMP-1, TIMP-2	(Johnson et al., 1994; Moses, 1993; Murphy et al., 1993)
interferon- (IFN-)	(Sidky and Borden, 1987)
bFGF soluble receptor	(Hanneken et al., 1994)
non-polypeptides
steroids: medroxyprogesterone, 2-methoxyestradiol, dexamethasone	(Blei et al., 1993; Fotsis et al., 1994; Wolff et al., 1993)
retinoids: Re 80, Am 580, Am 80	(Oikawa et al., 1993)
GPA 1734 (quinolizine derivative, BM synthesis inhibitor)	(Maragoudakis et al., 1988)
cyclic peptide 203 (RGDfV, V3 integrin antagonist)	(Brooks et al., 1994)
AGM-1470 (fumagillin derivative)	(Ingber et al., 1990; Oliver et al., 1994)
taxol (microtubule stabilizer)	(Oliver et al., 1994)
pentosan	(Nguyen et al., 1993)
thalidomide	(D'Amato et al., 1994)
placental proliferin-related protein	(Jackson et al., 1994)
GM3 (ganglioside)	(Ziche et al., 1992)

2.3.3 Angiogenesis in the adult organism

Physiological angiogenesis. After the embryonic and postnatal development, the rapid proliferation of the endothelial cells is downregulated almost to quiescence. The turnover time of endothelial cells in the adult is measured in months or years (Denekamp, 1993). However, the capillary endothelial cells maintain their proliferative potential throughout the life of an organism. Almost the only circumstances, under which proliferation is physiologically dramatically upregulate, are the female reproductive cycle and pregnancy (in more detail described under 2.4.1.7.1).

Pathological angiogenesis. Different pathological conditions involve the neovascularization of tissues (reviewed in Folkman, 1995a). Neovasularization can be an accelerator of disease progression (e.g. in many neoplastic diseases) or a mediator of pathogenesis (e.g. in diabetic retinopathy, in more detail described under 2.4.1.7.4-9).

2.4 The VEGF subfamily of growth factors

2.4.1 Vascular endothelial growth factor (VEGF[8])

VEGF is an endothelial cell-specific mitogen that has the ability to promote angiogenesis in several in vivo models. By alternative splicing of mRNA, VEGF is made in four different isoforms that differ markedly in their secretion pattern, but have similar biological activities. Expression of VEGF mRNA is temporally and spatially related to the proliferation of blood vessels in a variety of physiological and pathological circumstances, such as embryonic development, formation of the ovarian corpus luteum, wound healing, angiogenesis in solid tumours and in rheumatoid arthritis.

2.4.1.1 Discovery and cloning of VEGF

The two main biological activities of VEGF - mitogenic activity and vascular permeability inducing activity - were described, purified and designated independently as VPF (vascular permeability factor) and VEGF (vascular endothelial growth factor):

Vascular permeability factor. The release of a vascular permeability-increasing agent by guinea pig hepatocarcinoma cells was reported more than 15 years ago (Dvorak et al., 1979). Vascular leakage[9] was subsequently used to monitor purification of VEGF from the supernatant of this (Senger et al., 1983; Senger et al., 1986) and of the human histiocytic lymphoma cell line U-937 (Connolly et al., 1989b). Therefore, the factor was later designated as vascular permeability factor (VPF; Senger et al., 1983) or vasculotropin (Plouët et al., 1989).

Vascular endothelial growth factor. Media conditioned by bovine pituitary follicular/folliculostellate cells (FC) contained an agent that specifically promoted growth of vascular endothelial cells. The mitogenic activity of this agent towards vascular endothelial cells was used to monitor its purification, and on the basis of its target cell selectivity the purified agent was designated VEGF (Ferrara and Henzel, 1989; Gospodarowicz et al., 1989).

Soon after the first descriptions by Ferrara, Gospodarowicz and Conolly, several groups reported the purification of VEGF from different sources: human VEGF from U937 cells (Keck et al., 1989), bovine VEGF from pituitary folliculostellate cells (Leung et al., 1989) and AtT-20 pituitary tumour cells (Plouët et al., 1989).

The first human VEGF cDNAs were cloned from a phorbol ester-activated HL60 promyelocytic leukaemia cell library (Leung et al., 1989) and a histiocytic lymphoma cell line U937 library (Connolly et al., 1989b); both cDNAs were screened with oligonucleotides designed on the basis of the amino acid sequence of the previously purified protein.

VEGF-like sequences have also been found in two parapoxvirus genomes. It is thought that they were acquired from a mammalian host and mediate now capillary proliferation in infected goats and sheep (Lyttle et al., 1994).

2.4.1.2 Genomic structure and splice variants of VEGF

Promoter, transcriptional control elements and transcriptional start site. The hVEGF gene is located in chromosome 6p12 (Vincenti et al., 1996; Wei et al., 1996). It has a single major transcriptional start site 1038 bp upstream from the ATG initiation codon. The 5'-UTR and 2.4 kb of the TATA-less promoter have been sequenced, revealing three SP1 binding sites immediately in front of the transcriptional start site and four AP-1 and two AP-2 binding sites located around the transcriptional start site (Tischer et al., 1989). Recently regulatory elements, located within the 5'-UTR, have been described (Cohen et al., 1996).

Alternative splicing and resulting isoforms/alternative polyadenylation of mRNA. The VEGF gene is organized in eight exons (Figure 16), and the its coding region spans approximately 14 kb (Tischer et al., 1989). Four different isoforms are the result of alternative splicing of the mRNA. The shortest isoform, VEGF121, is encoded by exons 1-5 and 8, VEGF165 includes additionally exon 7. VEGF189 and VEGF206 mRNAs contain all 8 exons, and the usage of a variable 5'-splice donor site within exon 6 creates the difference between the VEGF189 and VEGF206 mRNA. The VEGF206 mRNA is an extremely rare species as it was isolated only once from a human fetal liver cDNA library (Houck et al., 1991) and the corresponding form is not present in mouse (Shima et al., 1996). All four mRNA isoforms contain the sequence encoding the hydrophobic signal peptide of 26 amino acids for secretion (Houck et al., 1991; Tischer et al., 1989). Every intron-exon boundary possesses classical consensus splicing signals. The majority of transcripts for VEGF165 contain 1.9 kb of 3'-UTR (2.2 kb for murine mRNA) resulting in mRNA transcripts around 3.9 kb (Conn et al., 1990), although four potential polyadenylation sites are present, and differential termination of transcription has been reported (Levy et al., 1995).

Initially, VEGF165 was thought to be the predominant isoform in all tissues except placenta, where VEGF121 is the most abundant species (Houck et al., 1991). Although VEGF189 has been shown to predominate in rat heart and lung (Bacic et al., 1995), and VEGF121 seems to be expressed by several tissues and cell lines in similar amounts as VEGF165 (Bacic et al., 1995; Minchenko et al., 1994), in this work the term "main form" is used for VEGF165. The longest isoform (VEGF206) was expressed in cell culture (Houck et al., 1991), but evidence of its in vivo-existence is still lacking.

Mouse VEGF gene structure. The mVEGF gene structure is very similar to the human VEGF gene structure (compare Figures 16 and 17; Shima et al., 1996). Three isoforms (VEGF120, VEGF164 and VEGF188) have been shown to be generated by alternative splicing. The existence of a fourth splice variant coding for a putative 145 amino acid peptide (corresponding to the longest human splice variant) is unlikely, a premature stop-codon would be created by the frame shift resulting from the use of the alternative 5'-splice donor site (Shima et al., 1996).

2.4.1.3 Molecular characterization

VEGFs are dimeric proteins. The main form has a molecular weight of approximately 45 kDa and it is composed of the 165 amino acid isoform (Ferrara and Henzel, 1989). The minor human VEGF isoforms contain 121 and 189 amino acids following signal sequence cleavage. Compared with VEGF165, VEGF121 lacks 44 amino acids; VEGF189 has an insertion of 24 largely cationic amino acids and VEGF206 has an additional insertion of 17 amino acids resulting in different affinities for endogenous polyanions such as cell surface heparan sulphates (Figure 16; Park et al., 1993). Rodent and bovine VEGF monomers are shorter by one amino acid. The VEGF monomers have a single glycosylation site at Asp 75 of the mature protein, but glycosylation is not necessary for biological activity (Claffey et al.,1995; Yeo et al., 1991), although it is important for the efficient secretion of VEGF.

Heterodimers have been described in the mouse, which are composed of VEGF121 and VEGF164 (Breier et al., 1992), and the rat glioma cell line GS-9L secretes heterodimers between VEGF164 and PlGF, which were shown to be mitogenically only 3 times less active towards human umbilical vein endothelial cells (HUVEC) than VEGF164 homodimers from the same source (DiSalvo et al., 1995).

The amino-acid sequence of VEGF has a weak homology (~15-20%) to PDGF-A and B and their viral counterpart v-Sis (Conn et al., 1990; Leung et al., 1989), a higher homology to VEGF-C/VRP (~30%; Joukov et al.,1996; Lee et al., 1996) and VEGF-B/VRF (43%; Grimmond et al., 1996; Olofsson et al., 1996a) and the highest homology to PlGF131 (53%, compared in Figure 5; Maglione et al., 1991). In all these molecules the 8 cysteine residues characteristic for members of the PDGF family are conserved. These cysteines are located in the receptor binding domain and are involved in the formation both intra- and interchain disulphide bridges. Thus the folding pattern of these proteins is likely to be similar, resulting in a typical compact cysteine-bonded knot and physical properties like relative stability against heat, acids and mild proteolysis (Thomas, 1996). The crystal structure and mutational analysis of PDGF-BB have shown that the 2nd and the 4th cysteine residue are involved in the dimer formation in an antiparallel fashion (Andersson et al., 1992; Oefner et al., 1992). The same cysteine residues (aa 25 and 67) have been shown to be involved in the interchain disulphide bridge formation in VEGF (Pötgens et al.,1994). Non-covalently linked PDGF-BB, in which the interchain-forming cysteines were mutated, retained biological activity (Kenney et al., 1994), whereas the biological activity of VEGF is abolished upon disruption of its interchain disulphide bonds (Pötgens et al., 1994).

By mutational analysis the acidic amino acids Asp63, Glu64 and Glu67 have been identified as FLT1 receptor binding determinants, whereas the basic residues Arg82, Lys84 and His86 were critical for FLK1/KDR binding (Keyt et al., 1996).

Heparin binding. VEGF121 is a weakly acidic polypeptide that does not bind to heparin due to the lack of the heparin-binding domain encoded by exons 6 and 7. In contrast, VEGF165 is basic and binds to heparin. VEGF189 and VEGF206 are even more basic and bind to heparin with greater affinity (Houck et al., 1992). The differences in affinity for heparin affect the fate of the VEGF isoforms (Houck et al., 1992; Park et al., 1993). VEGF121 is secreted and is freely diffusible in the medium of transfected cells. VEGF165 is also secreted but a significant fraction remains bound to heparin-containing proteoglycans. The longer forms are almost completely bound to the cell surface and extracellular matrix (ECM). They interact probably not only with heparin-like molecules, but also with cell surface proteins distinct from HSPGs via cell surface retention sequences encoded by exon 6 (Boensch et al., 1995). A release of the VEGF165 and VEGF189 isoforms can be achieved by heparin or heparinases. Notably VEGF189 as such seems not to have the same angiogenic activity as the shorter VEGF isoforms (Wilting and Christ, 1996), although this might be solely due to its complete sequestration. It is thought to be a latent form of VEGF, capable of being activated by proteolytic mechanisms. Plasmin is able to cleave its C-terminus after Arg110, releasing a diffusible proteolytic fragment of 34 kDa with 100 fold reduced activity compared to VEGF (Houck et al., 1992; Park et al., 1993). Although the carboxyl-terminal domain (aa 111-165) of VEGF was reported to be critical for mitogenic activity (Keyt et al., 1996), recombinant VEGF121 was fully mitogenic in other studies (Kondo et al., 1995); but comparisons might be difficult due to different experimental settings.

2.4.1.4 VEGF receptors

Two tyrosine kinases, FLT1 and KDR, have been identified as VEGF receptors (Figure 6; de Vries et al., 1992; Millauer et al., 1993; Quinn et al., 1993; Terman et al., 1992a). FLT1 (fms-like-tyrosine kinase 1) was recently renamed as VEGFR-1(vascular endothelial growth factor receptor 1; Shibuya et al., 1990) and KDR (kinase insert domain containing receptor) as VEGFR-2(vascular endothelial growth factor receptor 2; Terman et al., 1991). Non-tyrosine kinase receptors distinct from FLT1 and KDR have been reported recently (Soker et al., 1996).

2.4.1.4.1 Cloning and chromosomal localization of FLT1 and KDR

FLT1. FLT1 was originally cloned from a human genomic library using the v-ros tyrosine kinase as a hybridization probe under low stringency conditions, and the full length clone was subsequently isolated from a placental cDNA library (Shibuya et al., 1990). The mouse and rat homologues of FLT1 have also been cloned (Choi et al., 1994; Finnerty et al., 1993; Yamane et al., 1994). The FLT1 gene localizes to chromosome 13q12 in humans near to the FLT3 gene (Shibuya et al., 1990). A similar linkage is seen on the mouse chromosome 5 (Rosnet et al., 1993).

KDR/FLK1. KDR was cloned from a human endothelial cell cDNA library (Terman et al., 1991), its murine homologue (named FLK1, fetal liver kinase 1) from primitive hematopoietic (Matthews et al., 1991) and embryonic neuroepithelial cells (Oelrichs et al., 1993). The rat homologue (named Tkr-III) was obtained from an aortic cDNA library (Sarzani et al., 1992). The FLK1 gene lies on mouse chromosome 5 in a cluster of class III RTKs closely linked to the genes for c-Kit and PDGF receptor A (Matthews et al., 1991). This linkage is not conserved in humans, as the genes for c-kit and PDGF receptor a map to q11-4q13, whereas KDR maps to chromosome 4q31.2-4q324in a region devoid of any known tyrosine kinase growth factor receptor (Terman et al., 1992b).

2.4.1.4.2 Molecular characteristics of FLT1 and KDR

The FLT1 and KDR proteins belong to the PDGF class of receptor tyrosine kinases and form together with FLT4 (which does not bind VEGF) the FLT-subfamily of endothelial cell receptors (see Mustonen and Alitalo, 1995). The FLT-subfamily differs from other members of the PDGF family having 7 instead of 5 extracellular immunoglobulin-like loops in the extracellular domain. During evolution, the FLT family may have emerged via a duplication of the two immunoglobulin domains adjacent to the membrane, a mechanism already proposed for the emergence of the PDGF-receptor (Claesson-Welsh et al., 1988; Shibuya et al., 1990).

In spite of its name, FLT3 (also called FLK2) does not belong to the FLT-family. FLT3 is mainly expressed in hematopoietic cells and contains 5 extracellular Ig-like loops, thus being more related to c-Fms (the receptor for M-CSF/CSF-1) and the c-Kit/stem cell factor receptors of the PDGFR-family (Rosnet et al., 1991a; Rosnet et al., 1991b).

The overall amino acid sequence identity between FLT1 and KDR is 44% leading to very similar physical characteristics. The homology between the human and murine forms of both receptors is approximately 85% (Matthews et al., 1991). Both FLT1 and KDR have a hydrophobic leader peptide, seven immunoglobulin-like domains in the extracellular domain, a single transmembrane domain, and a consensus tyrosine kinase sequence interrupted by a kinase-insert domain. FLT1 and KDR share about 30% sequence identity with c-Fms (de Vries et al.,1992; Matthews et al., 1991; Millauer et al., 1993; Quinn et al., 1993; Shibuya et al., 1990; Terman et al., 1991; Terman et al., 1992a).

FLT1. FLT1 is produced as a 1338 amino acid residue precursor with a predicted 22 amino acid signal peptide. The extracellular domain of FLT1 is composed of 736 amino acids; its transmembrane spanning domain is 22 amino acids and its cytoplasmic domain is 558 amino acids long. FLT1 contains 13 potential N-glycosylation sites, several of which appear to be utilized since the observed molecular weight of 190-200 kDa in SDS-PAGE is about 25% higher than the predicted 150 kDa (de Vries et al., 1992; Shibuya et al., 1990). Recently, a cDNA encoding a truncated form of FLT1 has been cloned from HUVEC cells, which misses the seventh Ig-like domain, the cytoplasmic domain and the transmembrane sequence (Kendall and Thomas, 1993). This soluble FLT1 protein competes with FLT1 for VEGF and might be a physiological downregulator of VEGF action.

KDR. KDR is expressed as a 1356 amino acid precursor with a predicted 19 amino acid signal peptide. Its extracellular domain is 745 amino acids long after signal peptide cleavage; its transmembrane domain is 25 amino acids and its cytoplasmic domain 567 amino acids long. KDR contains 18 potential N-linked glycosylation sites. Its observed molecular weight is 190 -200 kDa on SDS-PAGE, whereas its predicted molecular weight is only 150 kDa suggesting that, similar to FLT1, some of the glycosylation sites in KDR are actually utilized (Matthews et al., 1991; Terman et al., 1991; Terman et al., 1992a). Four tyrosine autophosphorylation sites were identified in the cytoplasmic domain of KDR, two of which are located in the kinase insert domain (Dougher-Vermazen et al., 1994). KDR does not contain classical consensus sequences for SH2 domain binding. On the other hand, the protein sequence around Y1054 and Y1059 is conserved between FLT1 and FLT4, indicating the existence of an extended recognition motif.

Binding characteristics. Both FLT1 and KDR have been shown to bind VEGF with high affinity[10], although KDR has a somewhat lower affinity for VEGF than FLT1 does. FLT4, a close homologue to FLT1 and KDR, was shown not to bind VEGF, but instead VEGF-C (Joukov et al., 1996; Lee et al., 1996). Cultured endothelial cells express 3000-60000 KDR receptors/cell and 500-3000 FLT1 receptors/cell (Bikfalvi et al., 1991; Myoken et al., 1991; Olander et al., 1991). Recently, the ligand-recognition of FLT1 was localized to the second Ig-like domain in "domain-swapping" experiments (Davis-Smyth et al., in press).

Modulation of VEGF-receptor interaction by heparan sulphate proteoglycans (HSPGs). HSPGs and other heparin-like molecules are integral components of both cell surfaces and ECM. They have been shown to mediate and influence receptor binding of several growth factors. Also VEGF receptor binding is affected by cell-surface heparan sulphates in a complex fashion, depending on the heparin binding properties of the particular VEGF isoform, on the specific VEGF receptor type involved, and on the amount and composition of heparin-like molecules that are present on the cell surface and the ECM (Gitay-Goren et al., 1992; Tessler et al., 1994). Consequently, the effect of heparin on receptor binding of VEGFin vitro shows a concentration optimum. E.g. FLT1 binding of VEGF was claimed to be strongly reduced already at low heparin concentrations, whereas KDR binds VEGF more strongly at high heparin concentrations (Terman et al., 1994). Notably, not only the interaction of the longer VEGF isoforms with their receptors is modulated by heparin, but also VEGF121 binding (Cohen et al., 1995b).

Novel receptors. Binding sites for VEGF of somewhat lower affinity have been described on human endothelial and tumour cells (KD ~200 pM), but nothing is known about the structure and function of these receptors. They interact with VEGF165, but not with VEGF121, and apparently exon 7 encoded sequences are important for this interaction (Omura et al., 1996; Soker et al., 1996). In the light of both old and recent results (Barleon et al., 1996; Clauss et al., 1990) the single class of binding sites identified on mononuclear phagocytes (Shen et al., 1993) represents FLT1, although its affinity was almost two magnitudes lower than expected (KD ~300-500 pM).

2.4.1.5 VEGF expression

VEGF expression is temporally and spatially coupled to the proliferation of blood vessels in the developing embryo (Breier et al., 1992). It starts at the same time as FLK1/KDR expression (E 7.0 in the mouse development) in the juxtaposed endoderm (Dumont et al., 1995). At E 7.5 VEGF expression can be found in the entire endoderm. During further development VEGF mRNA can be found in various epithelial tissues adjacent to FLK1/KDR- and FLT1-expressing endothelium, e.g. primary vasculature, primordial heart, kidney and brain (Breier et al.,1992; Dumont et al., 1995; Jakeman et al., 1993; Yamaguchi et al., 1993). After the embryonic and postnatal development, the endothelium becomes quiescent and, correspondingly, the expression of VEGF declines to a very low level. Despite of no angiogenic activity, VEGF expression persists in adults in distinct tissues, where a specialized endothelium with an increased vessel permeability is required. It has been proposed that VEGF signalling is needed for the differentiation and maintenance of this specialized, fenestrated endothelium and perhaps generally for the "base line"permeability of endothelial cell layers (Berse et al., 1992; Breier et al., 1992; Brown et al., 1992a; Millauer et al., 1993; Risau, 1995).

Kidney. VEGF is expressed at the time of embryonic kidney angiogenesis, but persists in the adult only in the kidney glomerular epithelium, which is adjacent to the fenestrated endothelium expressing both FLT1 and FLK1/KDR (Breier et al., 1992; Simon et al., 1995). These structures, of course, are specialized to ultrafiltration.

Brain. VEGF is transiently highly expressed in the ventricular neuroepithelium and the choroid plexus epithelium of the embryonic brain, and a VEGF concentration gradient probably induces brain vascularization by sprouting of endothelial cells from the perineural plexus that covers the neural tube (Breier et al., 1992; Millauer et al., 1993; Risau and Lemmon, 1988). In the adult brain the choroid plexus epithelium continues to express VEGF at modest levels, whereas elsewhere the mRNA is reduced to very low levels (Breier et al., 1992; Hatva et al., 1995). The VEGF expressing endothelium is - like in the kidney - in immediate vicinity to FLK1/KDR expressing epithelial cells, arguing fora specific role of FLK1/KDR signalling through VEGF in the maintenance of this specialized endothelium.

Placenta. VEGF expression has been observed in the developing ovarian follicles, in the capillaries of the forming corpus luteum, in the endometrium undergoing regeneration and the labyrinth region of the placenta during the implantation of the early embryo. These tissues acquire new capillary networks and VEGF seems to provide the necessary angiogenic stimulus (Jakeman et al., 1992; Phillips et al., 1990; Shweiki et al., 1993). Strong VEGF expression by multinucleated trophoblast cells (which separate fetal and maternal blood vessels in the labyrinth region of the placenta) has been - like in the kidney - connected to the frequent fenestrations of the adjacent endothelial cells (Shweiki et al., 1993).

Liver, lung and heart. VEGF is highly expressed by epithelial cells of the adult lung and by cardiac myocytes (Berse et al., 1992). In these tissues an elevated level of the 189 amino acid isoform of VEGF can be found, suggesting, that it might have a specific function in these organs (Breier et al., 1992).

What are the producer cells? Although a great variety of normal cell types have been shown to express the VEGF mRNAs, including macrophages, neurons, myocytes and many epithelial cells (Berse et al.,1992; Ferrara et al., 1992; Peters et al., 1993), there are no reports of vascular endothelial cells expressing VEGF in vivo (Shifren et al., 1994). This strongly argues that VEGF is mainly a paracrine mediator of angiogenesis.

2.4.1.6 VEGF receptor expression

Whereas VEGF can be expressed by a variety of cell types, VEGF receptors are almost exclusively found on endothelial cells, both quiescent and proliferating. The expression of both VEGF receptors is regulated by cell differentiation, as well as by external stimuli. In the case of FLT1, a 1 kb fragment of the promoter has been shown to direct endothelial cell-specific gene expression (Morishita et al., 1995). A putative regulation of the VEGF receptor genes by hypoxia has been studied, but the results are not nearly as coherent as for VEGF. In vivo, KDR and FLT1 were upregulated upon hypoxia (Thieme et al., 1995), whereas, at least in vitro, KDR was downregulated (Takagi et al., in press-a). The KDR gene is unlikely to be directly responsive to hypoxia. In vitro, KDR upregulation was induced by medium that was conditioned by myoblasts under hypoxic conditions, whereas direct hypoxia showed no effect (Brogi et al., 1996).

2.4.1.6.1 FLT1.

FLT1 expression is largely restricted to endothelial cells and their progenitors (Jakeman et al., 1993; Jakeman et al., 1992; Peters et al., 1993). It starts at E 8.0 during mouse development in the extraembryonic mesoderm, probably in the cells that develop later into the spongiotrophoblast cells of the placenta, which show a very high expression of FLT1 (Dumont et al., 1995). In the embryo proper, FLT1 mRNA can be found first in the embryonic mesoderm, which gives rise to the heart and the major embryonic vessels (Yamaguchi et al., 1993). Later FLT1 is seen in the endothelial cells of the blood vessels and capillaries of the developing organs, a pattern very similar to FLK1 expression (Breier et al., 1992; Simon et al., 1995). However, in the developing placenta FLT1 and FLK1/KDR are not co-localized, since FLT1 is present in the spongiotrophoblast layer, whereas FLK1/KDR is found in the labyrinthine layer nearby the VEGF expressing epithelium. The fact that FLT1 is not always located in the immediate vicinity of VEGF expressing cells was taken to support the idea, that the short VEGF isoform might primarily interact with FLT1 in order to elicit a chemotactic response, whereas FLK1/KDR might be activated mainly by the long, cell-associated VEGF isoforms, thereby stimulating proliferation (Breier et al., 1995; Dumont et al., 1995).

In adult tissues the same local correlation is seen between VEGF and FLT1 expression. In adult rats FLT1 is most highly expressed in the lung.

2.4.1.6.2 Responses to FLT1 stimulation.

Our knowledge of the signal transduction by VEGF receptors is limited. VEGF receptors are endogenously expressed on endothelial cells and the set-up of the signal transduction machinery in endothelial cells is probably very different from other frequently used cell types like fibroblasts. Endothelial cells show among themselves a great diversity, which additionally can account for inconsistencies among the published data. Several endothelial cells express more than one VEGF receptor, which makes it difficult to ascribe a certain response to one of them (Guo et al., 1995; Yamane et al., 1994).

At present, little is know about the intracellular signalling pathways that follow binding of VEGF to FLT1 and the resulting cellular responses. FLT1 shows an exceptional behavior: it binds VEGF with very high affinity, but many attempts failed to show its autophosphorylation upon ligand binding (de Vries et al., 1992; Waltenberger et al., 1994; Yamane et al., 1994). The only report sofar of tyrosine phosphorylation comes from FLT1-over-expressing mouse fibroblast (NIH3T3) cells (Seetharam et al., 1995), but similar to FLT1-over-expressing porcine aortic endothelial (PAE) cells, no mitogenic response was seen upon phosphorylation. The failure to elicit many biological responses via FLT1 comparable to KDR/FLK1 has been independently reported in several publications and is yet unexplained (Waltenberger et al., 1994). PlGF, which also binds FLT1, fails to provoke clear-cut cellular responses (Park et al., 1994). In spite of this, several intracellular transduction pathways are activated by FLT1 stimulation, including Ca²⁺ signalling (de Vries et al., 1992; Myoken et al., 1991; Plouët and Moukadiri, 1990), apparently through PLC phosphorylation (Seetharam et al., 1995) and phosphorylation of the GAP complex. MAP kinase could not be activated by VEGF in FLT1-over-expressing NIH3T3 cells, whereas it was activated in sinusoidal liver endothelial cells (Seetharam et al., 1995), where the response might have been mediated by FLK1/KDR, which is also expressed by these cells.

The role of FLT1 signalling for vascular endothelial cells remains mysterious and to be determined. Recently, several reports showed the importance of heterodimerization for ligands (Pötgens et al., 1994) as well as for receptors (Karunagaran et al., 1996; Wrana et al., 1994). Indeed, FLT1 homodimers might have - similar to e.g. ErbB3 - no direct ligand, and an active role of FLT1 might be confined to combinatorial interactions with other receptors.

Already some time ago it has been proposed that FLT1 might play a role in the physiological establishment and further maintenance of an organized vascular system (Shibuya et al., 1990), a function that might remain unnoticed in vitro, but which is supported by the FLT1 knock-out phenotype (Fong et al., 1995).

It has been also proposed that FLT1 mediates the chemotactic and FLK1/KDR the proliferative response to VEGF. Monocytes express VEGF receptors (Clauss et al., 1990) and their activation and migration seems to be mediated by VEGF signalling via FLT1[11] (Barleon et al., 1996). It is unknown whether the migrational stimulus for endothelial cells is also transduced via FLT1. Some in vivo -data is compatible with this hypothesis, for example the differences between the expression patterns of FLT1 and FLK1/KDR in the developing placenta (Dumont et al., 1995) and the strong overexpression of FLT1 by endothelial cells invading a healing wound (Peters et al., 1993). On the other hand, FLT1-mediated migration could never be demonstrated in response to VEGF in vitro. Recently, receptor-specific mutants of VEGF have been generated (Keyt et al., 1996), which could be helpful clarifying the role of FLT1.

2.4.1.6.3 FLK1/KDR

The expression of FLK1/KDR starts very early during development (mouse E 7.0) in both embryonic and extraembryonic tissues. It occurs in the embryonic mesoderm, that later gives rise to the heart and in the future yolk sac mesoderm, i.e. in regions from which the vasculature emerges. FLK1/KDR may therefore be a marker for the hemangioblast, a hypothetical precursor cell for endothelial and hematopoietic cells. VEGF on the other hand might be an early inducer of these cells, since its expression starts simultaneously with FLK1/KDR in tissues surrounding first the emerging and later the expanding vasculature.

FLK1/KDR becomes restricted to the endothelial lineage during the blood island formation (Dumont et al., 1995; Flamme et al., 1995a; Yamaguchi et al., 1993). Expression of FLK1/KDR continues in differentiated arteries, veins and lymphatic vessels, whereas it decreases in differentiated capillary endothelial cells, which maintain the option of upregulating FLK1/KDR upon stimulation by high VEGF levels (Wilting and Christ, 1996).

The interplay between KDR and its ligand VEGF in the vascularization of developing organs by angiogenic sprouting has been demonstrated for the brain and the kidney. During brain development vascular sprouts of the perineural plexus invade the neuroectoderm by an angiogenic process. During this phase VEGF is expressed in the choroid plexus and in the ventricular layer, whereas FLK1 is expressed in the angiogenic sprouts (Breier et al., 1992; Millauer et al., 1993). A similar distribution of VEGF and its receptors is seen in the developing kidney (Breier et al., 1992; Simon et al., 1995) and in animals that locally over-express VEGF from a retroviral vector with the result of a (most probably KDR-mediated) regional hypervascularization (Flamme et al., 1995b).

The fact that FLK1/KDR expression persists in certain differentiated vessels has lead to the concept of a dual role for VEGF, being on the one hand an angiogenic factor for certain endothelial cells, and on the other hand a survival- or maintenance factor for others (Dumont et al., 1995).

2.4.1.6.4 Responses to FLK1/KDR stimulation

VEGF binding stimulates rapid and strong tyrosine phosphorylation of FLK1/KDR (Millauer et al., 1993; Myoken et al., 1991; Quinn et al.,1993) and induces cytosolic Ca²⁺ fluxes (Bikfalvi et al., 1991; Brock et al., 1991; Quinn et al., 1993). FLK1/KDR has different signal transduction properties from FLT1, in that its activation leads to a strong biological response, that is seen in both transfected cells and in cells endogenously expressing the receptor (D'Angelo et al.,1995; Millauer et al., 1993; Waltenberger et al., 1994). KDR-transfected PAE cells and a clone of Balb/c3T3 A31 cells (a fibroblast cell line of endothelial cell morphology and behavior) undergo edge ruffling and other changes of cell morphology after stimulation with VEGF, in addition to mitotic induction and chemotaxis (Enomoto et al., 1994; Waltenberger et al., 1994), whereas FLT1-transfected PAE cells do not exhibit any of these responses. The exclusive role of FLK1/KDR as the responsible receptor for mitogenesis and proliferation is underlined by the fact that mutated VEGF, which binds FLK1/KDR, but not FLT1, retains its full mitogenic activity (Keyt et al., 1996).

It seems to be dependent on the cellular background, which signalling pathways are initiated by KDR stimulation. Tyrosine phosphorylation of PLC, Ras-GAP and PI3-K can be observed upon VEGF stimulation of bovine aortic endothelial cells, but not upon VEGF stimulation of KDR-transfected PAE cells. This failure might be either due to the different cellular background or, alternatively, the phosphorylation events in bovine aortic cells might have been mediated by FLT1 (Guo et al., 1995). In brain capillary endothelial cells and hepatic sinusoidal endothelial cells, KDR-stimulation by VEGF activates the MAP kinase and elicits a strong proliferative response (D'Angelo et al., 1995; Yamane et al., 1994). Although FLT1 is expressed by the same cells, at least in hepatic sinusoidal endothelial cells MAP kinase activation did not involve phosphorylation of FLT1 (Yamane et al., 1994).

2.4.1.7 VEGF and VEGF receptor expression in neovascular processes in the adult

2.4.1.7.1 Female reproductive cycle and pregnancy

Angiogenesis occurs in a hormonally regulated manner during the female reproductive cycle and pregnancy in the ovary and the uterus (reviewed in Reynolds et al., 1992). In the first third of the ovarian cycle the growing corpus luteum (CL) becomes vascularized by invading blood vessels. During luteolysis and thereafter the newly formed vessels regress. The endothelial cells change the expression pattern of adhesion molecules and detach from the basement membrane, thus leaving areas devoid of the covering cell monolayer. A second mechanism of blood vessel regression is the contraction and occlusion of arterioles and small arteries upon proliferation of smooth muscle cells (Modlich et al., 1996). Hormonally regulated angiogenesis occurs also in the proliferating and regressing endometrium and especially prominently upon implantation of the embryo (Cullinan-Bove and Koos, 1993).

2.4.1.7.2 Exercise induced angiogenesis

Angiogenesis in the heart as a response to physical exercise and its beneficial effects have been noticed long time ago (Przyklenk and Groom, 1985; Tomanek, 1970). Still under discussion is the question, how the neovascular response is elicited and whether exercise-induced neovascularization compensates completely for the increased metabolic demands of myocyte hyperplasia in the heart muscle (Mall et al., 1990; Tomanek and Torry, 1994).

2.4.1.7.3 Wound healing

In wound healing angiogenesis is most prominent during the first seven days after wounding, and this correlates temporarily and spatially with the expression of angiogenic factors. VEGF and its receptor FLT1 are highly over-expressed during wound healing:initially VEGF is expressed by the keratinocytes adjacent to the wound edges and later by invading keratinocytes that migrate to cover the wound surface, and also by macrophages, probably induced by hypoxia (Brown et al., 1992b; Peters et al., 1993; Shweiki et al., 1992). FLT1, on the other hand, is expressed by the endothelial cells of the invading blood vessels (Peters et al., 1993). In a diabetic mouse model delayed wound healing could be correlated with insufficient VEGF expression (Frank et al., 1995).

2.4.1.7.4 Ocular neovascularization

Ocular neovascularization is the most common cause of blindness and dominates many eye diseases, among others age-related macular degeneration (AMD; Bressler et al., 1994) and diabetic retinopathy (Olk and Lee, 1993). Stimulated by VEGF new capillaries in the retina invade the vitreous, bleed, and cause blindness (Adamis et al., 1994; Aiello et al., 1994; Malecaze et al., 1994). In a mouse model for proliferative retinopathy, hypoxia-induced VEGF expression was followed by neovascularization (Pierce et al., 1995). Anti-angiogenic therapy (using either neutralizing antibodies against VEGF or a soluble form of the FLT1 receptor) successfully inhibited ocular neovascularization in a mouse and a monkey model (Adamis et al., 1996; Aiello et al., 1995). It is interesting that already in 1948 a theory was proposed, which assigned a putative diffusible factor, secreted by the retina, as the cause for retinal and iris neovascularization (Michaelson, 1948).

2.4.1.7.5 Ischemic heart and limbs

Collateral vessel formation occurs as a response of the body to arterial occlusions, e.g. after myocardial infarction in coronary heart disease (Shammas et al., 1993). The angiogenic stimulus may be partially due to the oxygen deficiency created by the clogged artery in the affected tissue. In animal models therapeutic benefits could be achieved by supporting the perfusion and neovascular response in ischemic heart or limb through intramuscular or intraarterial VEGF application (Banai et al., 1994; Takeshita et al., 1994). Recently, somatic gene therapy with a VEGF expression vector apparently resulted in angiogenesis in a human patient (Isner et al., 1996). Another potential application of VEGF gene therapy would be the limitation of restenosis after coronary angioplasty by the acceleration of healing of the endothelium with the consequent inhibition of smooth muscle cell proliferation (Finkel and Epstein, 1995).

2.4.1.7.6 Rheumatoid arthritis

Angiogenesis associated with the proliferation of inflammatory synovial tissue is most likely VEGF-mediated. Synovial tissue and macrophages could be identified as sources of the elevated VEGF levels (Koch et al., 1994). Inhibition of angiogenesis was suggested as a possible treatment of the disease, based on an animal model (Oliver et al., 1994).

2.4.1.7.7 Skin diseases

Recent studies suggest that VEGF is a main mediator of angiogenesis in several skin disorders (Brown et al., 1995b; Brown et al., 1995c; Detmar et al., 1994). Microvascular hyperpermeability and angioproliferation - both characterizing psoriatic epidermis -seem to be caused by an autocrine mechanism, in which keratinocytes produce TGF-, which in turn induces their own VEGF expression (Detmar et al., 1994). It will be seen, whether the results on transgenic mice over-expressing VEGF in the epidermis support the concept of the role of VEGF in psoriasis (personal communication by Michael Detmar). Other pathological conditions, in which VEGF upregulation was seen in the skin include bullous pemphigoid, dermatitis herpetiformis, erythema multiforme (Brown et al., 1995b) and the delayed hypersensitivity skin reactions (Brown et al., 1995c).

2.4.1.7.8 Graves' syndrome

Recently Sato added Graves' syndrome of exophthalmic goiter to the increasingly long list of disorders, which are thought to be connected with VEGF overexpression (Sato et al., 1995a).

2.4.1.7.9 Tumour angiogenesis

Diffusion of oxygen sets an upper limit for the size of solid tumours unless the tumour meets its need of oxygen by stimulating nearby capillary endothelial cells to proliferate and to migrate towards the tumour, resulting in its vascularization. Before switching to this angiogenic phenotype, a growing tumour reaches its size limit of several mm³, when net proliferation and necrosis/apoptosis are balanced (Folkman, 1971; Gimbrone et al., 1972). This usually happens before diagnosis, hence such tumours/metastases are called dormant (Holmgren et al., 1995). Many molecules are involved in tumour angiogenesis (Auerbach and Auerbach, 1994; Blood and Zetter, 1990; Folkman and Shing, 1992; Klagsbrun and D'Amore, 1991; Liotta et al., 1991), and the common clinical patterns in cancer metastasis have been associated with different expression patterns of angiogenic stimulators and inhibitors. Surprisingly, the release of angiostatin, a potent inhibitor of angiogenesis, is triggered by certain tumours. It has a much longer half life in the circulation than most angiogenic stimulators which might be secreted by the tumour, and consequently, vascularization and growth of distant metastases are suppressed as long as the primary tumour is not removed (Folkman, 1995a; O'Reilly et al., 1994).

The "angiogenic switch". Both upregulation of positive angiogenic signals and downregulation of negative angiogenic signals is required for the "angiogenic switch"(Rastinejad et al., 1989). According to this balance hypothesis a quiescent endothelium is constantly and actively restrained from proliferating. Some tissues express constitutively angiogenic inducers in the absence of any new blood vessel growth, arguing for the existence of general suppressory mechanisms. Indeed, evidence for angiogenesis suppressor genes have been presented (Parangi et al., 1995). Rather than being predetermined, the angiogenic phenotype is a discrete characteristic (similar to the loss of contact inhibition or immortalization), for which individual cells are selected in the microevolutionary process connected with the establishment of the fully malignant phenotype, (Hanahan and Folkman, 1996). In transgenic mouse models the angiogenic switch was identified in a discrete stage of tumour progression, occurring after the development of carcinoma-in-situ cells, but prior to the appearance of solid tumours (Folkman et al., 1989; Kandel et al., 1991). Biopsies of human breast and cervical cancers of different malignancy grades revealed a remarkable similarity to the mouse models, in that an angiogenic switch was already apparent in the mid- to late dysplastic stage (Guidi et al., 1995). Although the angiogenic switch is one of the indispensable genetic lesions required for rapid tumour growth, it is not the only requirement (Graeber et al., 1996; Kinzler and Vogelstein, 1996).

Proteases. Many proteases are involved in tumour angiogenesis and metastasis, since degradation of the basement membrane underlying the endothelium and the ECM are important steps in cell migration. The plasminogen system includes urokinase plasminogen activator (u-PA) and tissue type plasminogen activator (t-PA) which can cleave plasminogen to release active plasmin. Plasmin has a variety of proteolytic activities including degradation of many ECM and basement membrane components, as well as activation of certain growth factors and inhibitors by proteolytic cleavage (reviewed in Plow et al., 1995). Endothelial cells of both blood and lymphatic vessels are induced to secrete components of the plasminogen system in response to angiogenic factors like VEGF and TNF- (Koolwijk et al., 1996; Mandriotta et al., 1995; Pepper et al., 1994).

Surprisingly, in knock-out mice for u-PA, t-PA or plasminogen activator inhibitor the development of the vasculature was not affected (Carmeliet et al., 1994). On the other hand, application of urokinase receptor antagonists slowed down tumour progression in mice (Min et al., 1996).

The p53 connection. Even vascularized tumours eventually reach an equilibrium with net cell birth equalling net cell death (both necrosis and apoptosis) in the regions farthest away from the tumour vasculature (Kinzler and Vogelstein, 1996). Interestingly, the hypoxia-induced apoptosis/growth arrest is p53-dependent, although the underlying mechanism is unknown (Graeber et al., 1996). Any tumour cell acquiring a p53 lesion may escape from hypoxia-induced p53-mediated apoptosis and can become a predominant clone within the tumour. Additionally the expression of negative angiogenic factors can be dependent on wild type p53, e.g. thrombospondin-1 (TSP-1; Dameron et al., 1994; Van Meir et al., 1994; Volpert et al., 1995). On the other hand, the Ala135-Val mutant of murine p53 is able to upregulate positive angiogenic factors like VEGF (Kieser et al., 1994).

Diagnosis and prognosis. The degree of tumour vascularization - measured directly as the microvessel density in histologic specimens or indirectly by quantitation of angiogenic proteins such as bFGF and VEGF in body fluids - has been shown to be a good prognostic marker, especially for the estimation of the risk of metastasis and disease outcome in various types of tumours (Li et al., 1994; Nguyen et al., 1994; Weidner, 1995; Weidner et al., 1993; Weidner et al., 1991). The correlation between tumour angiogenesis and the risk of metastasis is obvious, since metastases depend on angiogenesis at two stages of their development:

• Firstly, metastatic cells do not usually escape from a tumour until the tumour becomes vascularized (Liotta et al., 1974). The probability of a tumour cell escaping from the tumour into the circulation is thought to increase with the area of the vascular surface within the tumour (Folkman, 1995a).

• Secondly, upon arrival in the target organ, metastatic cells can develop only into detectable metastases if they undergo neovascularization.

Primary tumours usually do not kill patients, since most of them can be excised by a surgeon. Patients usually die because tumour metastases occur in critical organs, and consequently disease outcome is closely linked to angiogenesis via the risk of metastasis.

VEGF overexpression by tumours. VEGF overexpression by tumours is frequently seen (Berse et al., 1992; Shweiki et al., 1992), and many human tumours over-express additionally VEGF receptors. Reports of this include carcinomas of the

• lung (Mattern et al., 1996),
• thyroid (Viglietto et al., 1995),
• breast (Brown et al., 1995a),
• gastrointestinal tract (Brown et al., 1993b; Suzuki et al., 1996; Takahashi et al., 1995),
• kidney and bladder (Brown et al., 1993a),
• ovarian (Olson et al., 1994),
• uterine cervix (Guidi et al., 1995),

as well as:

• angiosarcoma (Hashimoto et al., 1995),
• glioblastoma multiforme (Berkman et al., 1993; Hatva et al., 1995; Plate et al., 1992),
• meningioma (Hatva et al., 1995),
• (oligo)astrocytoma (Hatva et al., 1995),
• capillary haemangioblastoma (Berkman et al., 1993).

VEGF expression is seen usually by tumour cells. The tumour associated vasculature may express FLK1/KDR at elevated levels. These patterns correlate with the degree of vascularization of the tumour (Berkman et al., 1993; Berse et al., 1992; Brown et al., 1993a; Hatva et al.,1995; Plate et al., 1992; Takahashi et al., 1995; Warren et al., 1995). Generally, VEGF is a paracrine stimulator of endothelial cells in tumours. In angiosarcoma, however, VEGF probably acts as an autocrine factor (Hashimoto et al., 1995). Tumour-infiltrating inflammatory cells have been proposed as an additional source of VEGF (Freeman et al., 1995).

The involvement of VEGF and its receptors in tumour progression has been studied in several animal models:

Implantation of VEGF-expressing tumourigenic cancer cells into nude mice resulted in highly vascularized tumours. The size and vascularization of the tumours could be significantly reduced by application of anti-VEGF monoclonal antibodies. This therapeutic effect seemed to be a result of inhibited tumour angiogenesis, since the proliferation rate of tumours cells itself was not downregulated (Kim et al., 1993; Warren et al., 1995). Similarly, VEGF121-transfected human MCF-7 breast carcinoma cells gave rise to more vascularized and faster growing tumours in comparison with implanted, untransfected MCF-7 cells, although the transfected cells had no growth advantage in vitro (Zhang et al., 1995). A direct visualization of in vivo tumour growth blockade by VEGF antibodies was recently presented (Borgström et al., in press).

The dependence of FLK1 signalling in tumour angiogenesis was demonstrated in nude mice bearing implanted glioblastoma cells. A retrovirally introduced dominant-negative soluble FLK1 mutant (lacking the cytoplasmic domain) successfully inhibited tumour angiogenesis and progression (Millauer et al., 1994). Much less is known about the role of FLT1 in tumour vascularization. However, FLT1 does not seem to be frequently upregulated in tumours.

Therapy. The growing knowledge about angiogenesis in tumour progression has guided therapeutic approaches to stop tumour progression by inhibiting angiogenesis, some of which were already outlined above. Several of the anti-angiogenic components mentioned in Table 2 are in clinical trials. The ultimate goal is to archive a high selectivity of action, for which a thorough knowledge of all involved players is needed. Different steps of the angiogenic process are targeted, primarily the balance between positive and negative angiogenic regulators and cell-matrix interactions (reviewed in Folkman, 1995b).

2.4.1.7.10 Tumours derived from vascular endothelium

Hemangiomas. Hemangiomas are non-malignant tumours of blood vessels characterized by an extended network of capillary vessels with large lumens. Although they mostly need no treatment, because they regress after an initial proliferating phase, the tissue damage can be serious if they interfere with vital organs. several angiogenic proteins are over-expressed by hemangiomas, e.g. bFGF. In case corticosteroid therapy fails, anti-angiogenic therapy with interferon-2a (a bFGF antagonist) is perhaps the best alternative (Folkman, 1995b).

Lymphangiomas. Lymphangiomas are congenital lymphatic malformations characterized by enlarged fluid-filled lymphatic vessels. They are benign in character and surgical excision is the treatment of choice (Zadvinskis et al., 1992). Although several theories have been proposed to explain the pathogenesis of lymphangiomas (Lee, 1980; Philips, 1981; Weingast et al., 1988), the molecular basis of the uncontrolled growth of the lymphatic vascular endothelium is unknown. Interestingly, FLT4 and its ligand might play a role in the pathogenesis (this work).

2.4.1.8 Biological roles of VEGF signalling

Different approaches have been undertaken to resolve what is commonly addressed as the "function" of VEGF. The expression pattern of VEGF and its receptors, its biological activities and the phenotype of transgenic animals convincingly establish VEGF as the major regulator in the development and maintenance of the vascular system.

Mitogenicity/angiogenicity. The mitogenic activity of VEGF has first been demonstrated towards capillary endothelial cells. In vitro half-maximal stimulation of bovine capillary endothelial cell growth was obtained at 100-300 pg/ml (2-6 pM) and a maximal stimulation at 1-5 ng/ml (22-110 pM; Ferrara and Henzel, 1989; Plouët et al., 1989). VEGF has no appreciable mitogenic activity on non-vascular endothelial cell types: fibroblasts, smooth muscle cells, epithelial cells or endothelial cells of the lymphatic vascular system do not respond to VEGF (Connolly et al., 1989a; Ferrara and Henzel, 1989; Oh et al., submitted; Plouët et al., 1989; Wilting and Christ, 1996). VEGF exerts its mitogenic activity in a paracrine manner (Qu et al., 1995). This model is supported by the co-localization of producer cells and target cells in both pathological and physiological circumstances and its short half-life of 3 minutes in the circulation (Folkman, 1995a). In vitro, however, an autocrine loop has been identified in some cells (Guerrin et al., 1995; Nomura et al., 1995). The in vivo angiogenic potential of VEGF has been shown indifferent assays, e.g. on the chorioallantoic membrane (CAM) assay (Leung et al., 1989; Plouët et al., 1989; Wilting and Christ,1996; Wilting et al., 1991) and the rabbit cornea assay (Phillips et al., 1994).

Induction of vascular permeability. VEGF has both permeability inducing and mitogenic activity (Shibuya, 1995). On a weight-to-weight basis VEGF is 50,000 times more potent than histamine in increasing vascular permeability (Dvorak et al., 1995), and it does not work via histamine release from mast-cells (Gruber et al., 1995). instead VEGF triggers directly fenestration of endothelial cells, even of those resistant to classical inflammatory mediators such as platelet activating factor (PAF) or fibrin breakdown products (Roberts and Palade, 1995). Related effects of VEGF on the cardiovascular system are vasodilatation in vitro (Doi et al., 1996; Ku et al., 1993), resulting in tachycardia and hypotension in vivo (Yang et al., 1996).

In vivo the mitogenic and permeability-inducing activities of VEGF are closely linked (Dvorak et al., 1995). Apart from VEGF, no other angiogenic factor is able to induce vascular hyperpermeability, which indicates that VEGF could be a general "immediate" angiogenic agent, which is activated by indirect angiogenic factors, a mechanism that was shown already for e.g. TGF- (see Figure 4; Pertovaara et al., 1994).

Migration and differentiation. VEGF is able to promote chemotaxis of endothelial cells (Koch et al., 1994), but also some cells of the myeloid lineage (e.g. monocytes and mast cells) and smooth muscle cells respond to VEGF with migration (Clauss et al.,1990; Grosskreutz et al., 1996; Gruber et al., 1995; Koch et al., 1994). Osteoblasts differentiate in response to VEGF (Midy and Plouët, 1994).

Development. The importance of VEGF-mediated growth and differentiation signals in the development of the vasculature has been shown in knock-out mice. In animals lacking either FLT1, FLK1 or VEGF vascular development was arrested during early embryogenesis (Fong et al., 1995; Shalaby et al., 1995). The developmental block occurred at different stages upon loss of the gene, supporting the view that the signals transmitted by FLT1 and FLK1/KDR are distinct, and that besides VEGF additional factors are capable of stimulating FLK1/KDR.

FLK1 knock-out. Consistent with the early expression of FLK1 in endothelial/hematopoietic cell precursors, the differentiation of both endothelial and hematopoietic cell lineages from hemangioblasts was completely blocked in the embryos lacking FLK1 receptors. No endothelial cells nor hematopoeitic cells were present in homozygous animals and the embryos died between E 8.5 and E 9.5 of development (Shalaby et al., 1995).

FLT1 knock-out. FLT1 does not affect the differentiation or proliferation of endothelial or hematopoietic cells, but instead results in the formation of a disorganized vasculature, which leads to death from embryonic day E 8.5 onwards. In the yolk sac and in the embryo proper large vascular channels developed instead of organized blood islands. This migrational misbehavior was assigned to altered cell-cell or cell-matrix interactions (Fong et al., 1995).

VEGF. The phenotype of transgenic animals over-expressing VEGF in specific tissues substantiates the idea of VEGF being a major regulator of vascularization in embryogenesis and of maintenance of vasculature in certain organs (Flamme et al.,1995b). Vice versa, appropriately timed vascularization might be crucial for the proper morphogenesis of certain organs like the lung (Zeng et al., submitted). Quail embryos treated with recombinant VEGF during vasculogenesis appeared similar to FLT1 knock-out mice. The organization of the vasculature of VEGF-treated embryos was severely affected showing numerous extra vessels and enlarged lumens of fused vessels (Drake and Little, 1995). The effect of exogenous VEGF was seen already 5 h after injection[12], supporting the conclusion drawn from the FLT1 knock-out studies, that in addition to endothelial cell proliferation VEGF influences migrational patterns of endothelial cells and restricts vascularization by its paracrine manner of action.

VEGF knock-out. Loss of a single VEGF allele is lethal in the mouse embryo between E 10.5 and E 12, thus being the only known lethal heterozygous autosomal knock-out. It seems that VEGF levels have to be tightly regulated during vasculogenesis. Most steps of the early vascular and hematopoietic development were impaired, although not abolished and numerous abnormalities occurred in organogenesis connected with the failure of blood-vessel ingrowth, lumen formation, large vessel formation and the establishment of interconnections. In contrast to FLK1 knock-outs, homozygous VEGF-deficient embryos survive until E 10.5, implying the possibility of a partial activation of FLK1 by another ligand (Carmeliet et al.,1996; Ferrara et al., 1996), perhaps VEGF-C (Joukov et al., 1996).

Tie/Tek. The development of an intact and properly organized vasculature was also inhibited in mice deficient of Tie or Tek/Tie-2, two other RTKs, that are restricted to endothelial cells (Dumont et al., 1994; Puri et al., 1995; Sato et al., 1995b).

Induction of other genes by VEGF. VEGF increases or reduces the synthesis of several other proteins/peptides in vascular endothelial cells, most of which have been implied in playing a role in angiogenesis:serine proteases, urokinase-type and tissue-type plasminogen activators (uPA and tPA), PA inhibitor 1 (PA-1) and urokinase-type plasminogen activator receptor (uPAR; Mandriotta et al., 1995; Pepper et al., 1991; Pepper et al., 1994). VEGF induces synthesis of interstitial collagenase in human umbilical vein endothelial cells but not in dermal fibroblasts (Unemori et al., 1992). The co-expression of PAs, their cognate receptors and collagenase by VEGF is expected to break the basement membrane and to alter the ECM, thus facilitating migration of endothelial cells, a fundamental step in the angiogenesis cascade (see 2.3.1.1; Folkman and Klagsbrun, 1987).

The relaxing effect of VEGF on smooth muscle cells is indirectly mediated via nitric oxide/endothelium-derived relaxing factor (EDRF; Ku et al., 1993) and a downregulation of the synthesis of the C-type natriuretic peptide (CNP) in ECs (Doi et al., 1996).

2.4.1.9 Regulation of VEGF gene expression

The VEGF gene can be induced by phorbol esters like PMA and TPA (Finkenzeller et al., 1995; Garrido et al., 1993; Tischer et al., 1989), PDGF and cAMP analogues, consistent with the presence of binding sites for the transcription factors AP-1 and AP-2, suggesting that VEGF expression is controlled by protein kinase A- and C-mediated signal transduction pathways (Finkenzeller et al., 1995; Garrido et al., 1993).

Hypoxia-inducibility. The most prominent feature of VEGF regulation is the upregulation of VEGF by hypoxia involving a putative oxygen sensitive heme-protein, whose reduced state can be mimicked by cobalt or manganese. Hypoxia occurs in wounds, ischemic organs or around necrotic areas of tumours. VEGF is the only angiogenic factor for which hypoxia-inducibility has been shown[13]. Upregulation of VEGF by hypoxia has been shown both in vitro and in vivo (Ladoux and Frelin, 1993; Shweiki et al., 1992) and works probably by a mechanism similar to the upregulation of erythropoietin (Epo) by hypoxia[14] (Goldberg and Schneider, 1994). Glucose deprivation is able to upregulate VEGF expression to a similar extent as hypoxia does, although when both inducers are applied simultaneously, no change of VEGF expression occurs (Shweiki et al., 1995)[15]. The induction of VEGF in response to both hypoxic and hypoglycemic stress requires several hours and new protein synthesis; these parameters distinguish VEGF from protein products of immediate early genes like c-Myc and c-Fos (Stein et al., 1995).

Teleologically speaking, the ultimate aim of VEGF upregulation by hypoxia or hypoglycemia is to eliminate the hypoxic or hypoglycemic condition. In the retina transient hypoxia during development seems to induce physiological vascularization; normal oxygen levels reached upon sufficient perfusion of the newly vascularized tissue correspondingly seem to downregulate VEGF expression (Alon et al., 1995; Stone et al., 1995).

mRNA stabilization. Hypoxia and hypoglycemia mainly affect VEGF mRNA stability and to a smaller extent increase the transcription rate (Finkenzeller et al., 1995; Goldberg and Schneider, 1994; Ikeda et al., 1995; Levy et al., 1995; Stein et al., 1995). The mechanisms underlying the prolonged VEGF mRNA half life in hypoxia are not known, but the 3'-UTR of the rat VEGF gene has been found to contain sequence motifs that probably participate in the regulation (Levy et al., 1995).

Transcriptional activation. The transcriptional activation is thought to occur through binding of a hypoxia-inducible factor (HIF-1) to a 28-bp element (hypoxia-induced enhancer), which is present in the VEGF promoter region as well as in the 3'-flanking region in the Epo gene (Levy et al., 1995). Increased promoter activity under hypoxic conditions and binding of hypoxia-regulated factors in an oxygen-regulated manner have been demonstrated (Levy et al., 1995; Liu et al., 1995), but the overall significance of transcriptional activation in hypoxia-induced upregulation of VEGF is still being questioned (Shima et al., 1995b).

The angiogenic activity of nicotinamide and its analogues (Kull et al., 1987) was recently clarified by the finding that adenosine accumulation and the resulting elevation of intracellular cAMP levels are involved in the hypoxic upregulation of VEGF mRNA levels (Fischer et al., 1995; Hashimoto et al., 1994; Takagi et al., in press-a; Takagi et al., in press-b). Protein kinase A and C (Claffey et al., 1992; Finkenzeller et al., 1995; Takagi et al., in press-a; Takagi et al., in press-b), c-Src tyrosine kinase (Mukhopadhyay et al., 1995) and, at least in transformed cells, Ras (Rak et al., 1995) have been shown to participate in the signalling that induces VEGF.

Upregulation by inflammatory mediators and growth factors. Numerous inflammatory mediators and growth factors are able to upregulate VEGF. Therefore many of them have been shown to be indirect angiogenic factors (compare with Table 1). Synergism in the induction of angiogenesis has been demonstrated between bFGF, TGF- and VEGF (Asahara et al., 1995; Goto et al., 1993; Pepper et al., 1992; Pepper et al., 1993). Because VEGF and FGF seem to accompany many types of angiogenic tumours, it was assumed that VEGF and FGF complement each other in a yet unknown way to induce tumour angiogenesis. Aside from its direct effect on endothelial cells, bFGF upregulates VEGF and the VEGF receptor FLK1, which might explain its high angiogenic potential (Flamme et al., 1995b). A list of VEGF inducers includes:

• EGF, TGF-, TGF-, KGF and TNF- in human keratinocytes (Detmar et al., 1994; Frank et al., 1995),
• PDGF in smooth muscle cells and fibroblasts (Brogi et al., 1994; Finkenzeller et al., 1992),
• EGF in malignant glioma cells (Goldman et al., 1993),
• Thyroid-stimulating hormone (TSH) in thyroid follicle cells (Sato et al., 1995a),
• Thromboplastin/tissue factor in transfected meth-A sarcoma cells; (Zhang et al., 1994),
• bFGF and IL-1 in smooth muscle cells (Li et al., 1995; Stavri et al., 1995),
• EGF, PDGF-BB, and bFGF in human glioma cell lines (Tsai et al., 1995),
• IL-1, and TGF- in fibroblasts (Ben-Av et al., 1995; Brogi et al., 1994; Pertovaara et al., 1994),
• PGE2 in osteoblasts and fibroblasts (Ben-Av et al., 1995; Harada et al., 1995),
• estradiol in endometrial carcinoma cells (Charnock-Jones et al., 1993),
• Luteotrophic hormone (LH) in ovarian granulosa cells (Garrido et al., 1993),
• IL-6 in several cell lines (Cohen et al., 1996).

2.4.2 Placenta growth factor (PlGF)

The placenta growth factor (PlGF) was cloned from a placental DNA library. Sequence analysis revealed a 53% identity with VEGF in the PDGF homology domain (Maglione et al., 1991). The molecular and genomic structure of PlGF resembles that of VEGF: PlGF is a dimeric glycoprotein, alternative splicing creates two isoforms (PlGF131/PlGF-1 and PlGF152/PlGF-2[16]), which differ by the basic domain encoded by alternative exon 6. Mature rat PlGF (the equivalent to human PlGF152) shows a heterogenous NH2-end, due to differential signal peptide cleavage or additional proteolytic processing, resulting in mature peptides of 132 and 135 amino acids (PlGF132 and PLGF135; DiSalvo et al., 1995). Because of the absence of the heparin-binding domain, PlGF131 is a soluble factor similar to VEGF121, whereas PlGF152 binds to cell surface and ECM HSPGs. PlGF is highly expressed only in placenta (Hauser and Weich, 1993; Kaipainen et al., 1993; Maglione et al.,1993), but its biological function is unknown. PlGF may have some mitogenic activity towards certain endothelial cells (Hauser and Weich, 1993; Maglione et al., 1991).

PlGF binds to the FLT1 receptor with high affinity, but fails to elicit a mitogenic response in endothelial cells known to express FLT1. However, heterodimers between PlGF and VEGF do occur naturally and endogenous PlGF/VEGF heterodimers from a rat glioma cell lines showed mitogenic activity towards endothelial cells. This activity was only 3 to 7 times weaker than that of VEGF homodimers (DiSalvo et al., 1995; Park et al., 1994). Thus heterodimer formation might be a negative regulatory mechanism for VEGF expression (Cao et al., 1996). In contrast to VEGF, PlGF seems to be moderately downregulated by hypoxia (Gleadle et al., 1995).

2.4.3 VEGF-B[17]

2.4.3.1 Cloning of VEGF-B

A partial murine cDNA for VEGF-B was discovered when searching for new retinoid acid binding proteins from a murine library by the yeast two-hybrid method. Subsequently, the murine full length cDNA was obtained from an adult heart cDNA library. Homologous human sequences were cloned from the fibrosarcoma cDNA library HT-1080 and the whole coding region from an erythroleukemia cell cDNA library. Genomic clones were also obtained for both species.

Independently from the above described discovery, Grimmond et al. used a chromosome 11q13 specific cosmid probe to isolate the human and mouse VEGF related factor (VRF) gene, which then appeared to be identical with VEGF-B (Grimmond et al., 1996).

2.4.3.2 Genomic structure and splice variants of VEGF-B

Human VEGF-B maps to chromosome 11q13 (Paavonen et al., 1996). It contains 7 exons, and at least two different isoforms with unrelated C-terminal domains (VEGF-B167 and VEGF-B186) are generated by alternative splicing of exon 6 due to the use of distinct splice acceptor sites. Unlike in the alternative splicing of the VEGF mRNA, the alternative splice-acceptor site of VEGF-B exon 6 results in a frameshift and thereby in two overlapping reading frames (rarely seen among higher eukaryotes) coding for the distinct non-homologous C-termini of VEGF-B186 and VEGF-B167, the C-terminus of VEGF-B186 being much more hydrophobic than the C-terminus of VEGF-B167.

The genomic structure of the VEGF-B gene is highly conserved between humans and mice (compare Figures 18 and 19) and further resemble closely the structure of the VEGF gene. The main difference is that VEGF-B lacks a homologue of exon 6 of VEGF. Thus, exon 6B of VEGF-B corresponds to exon 7 of VEGF (Figure 7). Both code for the heparin-binding domain, of which VEGF-B186 is devoid due to the frameshift by alternative splicing.

2.4.3.3 Molecular characterization of VEGF-B

The two secreted murine and human isoforms of VEGF-B are 186 and 167 amino acids long. They show electrophoretic mobilities of 21 and 32 kDa under reducing conditions. Amino acid sequence homology is about 88% between mouse and human VEGF-B with highest divergence in the terminal regions of the protein. The pairwise comparison with other members of the PDGF family revealed that its closest relative is VEGF164/165 with approximately 43% amino acid sequence homology. The homology of mVEGF-B167 to PlGF131 is approximately 30% and to PDGF-A and B approximately 20% (amino acid sequence alignment in Figure 5).

The cysteine motif defining PDGF family members (PXCVXXXRCXGCC) is present in VEGF-B. VEGF-B forms disulphide-linked homodimers and heterodimers with VEGF. VEGF-B167 remains cell surface or ECM-associated after secretion by 293 EBNA cells, but can be released by heparin treatment. VEGF-B186 is secreted in a soluble form by Cos-1 cells, probably because it lacks the basic domain encoded by exon 6B.

The glycosylation pattern of VEGF-B is unique in the PDGF family. Unlike other growth factors of this family no N-glycosylation occurs in VEGF-B, whereas VEGF-B186 is the only PDGF family member that is O-glycosylated. This is rare among growth factors generally (Hudgins et al., 1992; Lu et al., 1995).

2.4.3.4 VEGF-B receptor(s)

The receptor for VEGF-B remains to be identified. Exon 6B of VEGF-B reveals a high homology with exon 7 of VEGF, which recently has been shown to code for a domain mediating the interaction with a novel class of VEGF receptors (Soker et al., 1996). However, VEGF-B binding to such receptors has not been reported to date. Two of the three acidic determinant amino acids for FLT1 binding of VEGF (Asp63 and Glu64) are conserved in VEGF-B, suggesting that VEGF-B could bind to FLT1. VEGF-B was shown to induce proliferation of FLT1- and KDR-expressing HUVEC and KDR-expressing bovine capillary endothelial (BCE) cells.

2.4.3.5 VEGF-B expression

VEGF-B is abundantly expressed in cardiac and skeletal muscle. Its expression pattern is partly overlapping with VEGF, thus showing that heterodimerization between VEGF and VEGF-B might be of biological significance in regulating the release, bioavailability and signalling properties of VEGF. The two isoforms seem to be present in equimolar amounts in various tissues.

VEGF-B gene expression starts early during embryonic development. At day 14 p.c. VEGF-B is expressed in most cells of murine embryos. Spinal cord, cerebral cortex and especially the heart display much higher expression levels than other tissues and around day 17 p.c. significant expression is seen mostly in the heart, brown fat tissue and the spinal cord (Lagercrantz et al., 1996).

2.4.3.6 Biological roles of VEGF-B signalling

VEGF-B is an endothelial cell mitogen as shown with experiments on HUVEC and BCE cells. Its tissue distribution implies a paracrine fashion of action similar to that of many other growth factors. The tight association of VEGF-B167 with the cell surface indicates either a requirement for direct cell-cell interactions for stimulation or is a means of regulating the bioavailability of VEGF-B (similar to VEGF). VEGF-B-mediated stimulation through direct cell-cell interactions may guide migrating endothelial cells and organize the correct spatial distribution of the vascular network in the development and maintenance of organs.

2.4.4 VEGF-C

2.4.4.1 Cloning of VEGF-C

VEGF-C was first purified from the supernatant of PC-3 human prostatic adenocarcinoma cells based on its activity to bind and to stimulate the autophosphorylation of FLT4 expressed in transfected NIH3T3 cells. Its cDNA was cloned subsequently from a PC-3 library, both human and murine genomic sequences were obtained from the EMBL3 library. Independently, a matching cDNA sequence was later reported and named VRP (vascular endothelial growth factor related protein). This cDNA was obtained from a human glioma library by screening with a probe derived from an expressed sequence tag (EST), that showed 36% homology to VEGF (Lee et al., 1996)[18].

2.4.4.2 Genomic structure and splice variants of VEGF-C

The genes for human and mouse VEGF-C are very similar to each other. All exon lengths are conserved except for exon 2, where the human gene has a 12 nucleotide insertion (compare Figures 20 and 21). Compared to other genes of the VEGF family the VEGF-C gene is much larger and encompasses at least 35 kb (Figure 20). The VEGF-C gene has one exon less than the VEGF gene. It lacks sequences that correspond to the VEGF exon 6 and confer heparin binding to the protein.

Three mRNA splice variants have been reported for the human VEGF-C gene (Figure 20; Lee et al., 1996) and two for the murine gene (Figure 21). These shorter mRNAs are rare species and it is not known whether they are translated.

2.4.4.3 Molecular characterization of VEGF-C

Biologically active VEGF-C migrates as a 23 kDa band on SDS-PAGE under reducing conditions. The amino acid sequence of VEGF-C shows in its N-terminal half homology to VEGF and related proteins. In the VEGF-homology domain (encoded by exon 3 and 4 in both VEGF and VEGF-C) VEGF-C is 30% identical with VEGF165, 27% with VEGF-B167,25% with PlGF131 and approximately 22-24% to PDGF-A and PDGF-B. The C-terminal half of VEGF-C is characterized by four complete and three incomplete cysteine repeats, which are encoded by exon 6. Cysteine repeats of this type were first identified from the larval saliva of the midge in the Balbiani ring 3 protein (BR3P), which plays a constitutive role in the formation of the silk-like fibers of the salival secretory proteins. Interestingly, also in the very N-terminal part of VEGF the BR3P motif can be identified (Figures 5 and 7).

Processing. VEGF-C is synthesized as a prepropeptide and cleaved after removal of the signal peptide at least in two positions. The resulting fragments give rise to multiple forms of monomers and dimers apart from the biologically active 23 kDa form, among which the 32 kDa form is the most prominent one. Depending on the expression system variable relative amounts of the different forms are produced.

2.4.4.4 VEGF-C receptors

VEGF-C, having been isolated as the FLT4 ligand, binds and stimulate sFLT4 (fms-like tyrosine kinase 4, also called VEGFR-3). Reports concerning the second putative VEGF-C receptor (KDR/FLK1) are contradictory. Although VEGF-C is probably capable of activating KDR/FLK1, little is known about the significance of this interaction. A partial activation of FLK1 by VEGF-C might be important during early vascular development, and it might be responsible for the differences between the VEGF and FLK1 knockout phenotypes. KDR/FLK1 has been described as a receptor for VEGF under 2.4.1.4.

2.4.4.4.1 Cloning and chromosomal localization of FLT4

FLT4 was discovered from a human erythroleukemia (HEL) cell cDNA library (Aprelikova et al., 1992; Pajusola et al., 1992), as well as from a placental cDNA library (Galland et al., 1992; Galland et al., 1993). The mouse and quail homologues of FLT4 (mFLT4 and Quek2 = quail endothelial kinase 2) have been cloned from embryonic cDNA libraries (Eichmann et al., 1993; Finnerty et al., 1993). The gene for human FLT4 is located in chromosome 5q35 telomeric to the genes for PDGFR and Fms and centromeric to the FGFR4 gene, a region affected by several translocations. In these translocations the FLT4 gene stays intact, whereas the ALK RTK becomes activated (Armstrong et al., 1993; Morris et al., 1994; Warrington et al., 1992).

2.4.4.4.2 Molecular characteristics of FLT4

Having seven extracellular Ig like domains, FLT4 is a member of the FLT subfamily within the PDGF class of receptor tyrosine kinases and thus it shares many features with FLT1 and KDR. FLT4 is about 35% homologous with FLT1 and KDR in its extracellular domain, to about 80% in its kinase domain, but differs greatly in its kinase insert and C-terminus (Finnerty et al., 1993; Pajusola et al., 1992; Pajusola et al., 1994).

In humans (but not in mice; Galland et al., 1993) two isoforms of FLT4 are generated by alternative mRNA polyadenylation and splicing. These isoforms differ mainly in the length of their C-terminal tails (Pajusola et al., 1993). Only the longer isoform (FLT4l) seems to be the biologically active one, probably due to a tyrosine residue (Y1337), which is involved in the receptor autophosphorylation, but which is absent in the shorter form (FLT4s; Fournier et al., 1995).

FLT4l is produced as a 195 kDa glycosylated precursor, which is proteolytically cleaved into fragments of 125 and 70 kDa. the 125 kDa C-terminal fragment spans the cell membrane and contains the tyrosine kinase domain, whereas the 70 kDa N-terminal fragment forms a major part of the extracellular domain. These fragments are held together by disulphide bonding. In reducing SDS-PAGE bands of 195, 175, 125 and 70 kDa can be detected, which correspond to the fully glycosylated unprocessed form[19], the unglycosylated unprocessed forms and the two cleavage products of the mature receptor (Pajusola et al., 1994).

2.4.4.5 VEGF-C and VEGF-C receptor expression

2.4.4.5.1 VEGF-C

VEGF-C is expressed both in embryonic and in adult mice. It is strongly expressed on day 7 p.c., before the onset of the expression of its main receptor FLT4. This suggests that VEGF-C interacts also in vivo with KDR/FLK1, as proposed (Joukov et al., 1996).

VEGF-C is seen in developing perinephric, mesenterial and jugular regions as well as in the non-neural parts of the cephalic region and becomes - like its main receptor FLT4 - restricted to endothelial structures, from which the lymphatic vessels originate according to the theory of Sabin. In the mesenterium, which is rich in developing lymphatic vessels, VEGF-C is strongly expressed by mesenchymal cells adjacent to endothelia that express its receptor. A paracrine action in the regulation of lymphatic angiogenesis is therefore presumed (Kukk et al., in press).

2.4.4.5.2 FLT4 and FLK1/KDR

The expression of FLT4 starts on day E 8.5 of mouse development, when it can be seen in the angioblasts of the head mesenchyme, in the cardinal vein and extraembryonically in the allantois (Kaipainen et al., 1995). Similarly, the onset of expression of its quail homologue Quek2 occurs later and recedes earlier than that of Quek1, which is homologous to both FLT1 and KDR/FLK1 (Eichmann et al., 1993). Quek2 is expressed in almost all endothelial cells of 2 day-old quail embryos. During further development, the expression pattern of FLT4/Quek2 becomes more restricted. FLT4 is seen in veins one 8.5 and on E 12.5, FLT4 expression concentrates in the lymphatic premordia, lymph sacs and sprouting lymphatic vessels (Kaipainen et al., 1995). Later, on days E 14.5 and E 16.5, the endothelium of the main lymphatic vessel, the thoracic duct, as well as many other smaller vessels devoid of red blood cells were found to express FLT4. From adult human tissues only the lymphatic endothelia and some high endothelial venules maintain FLT4 expression (Kaipainen et al., 1995). Similarly, in differentiated CAM FLT4 expression is restricted to the endothelial cells of lymphatic vessels (Wilting and Christ, 1996). FLT4 has not been detected in malignant primary tumours in vivo, but increased expression was seen in metastatic lymph nodes and in lymphangiomas (Hatva et al., 1995; Kaipainen et al., 1995).

2.4.4.6 Biological roles of VEGF-C signalling

Little is known about the biological function of VEGF-C. In vitro VEGF-C stimulates migration of bovine capillary endothelial cells (Joukov et al., 1996) and mitogenesis of human lung endothelial cells (Lee et al., 1996). Based upon their expression patterns, VEGF-C and its receptor are thought to play a general role in early vascular development and a more specialized role in the development of the venous and in particular the lymphatic vascular system. My own work adds proof to this point (see 5.1.6) in demonstrating that the biological function of VEGF-C is targeted to the lymphatic system.

Before discovery of its ligand, the intracellular signal transduction pathways following FLT4 activation were studied with receptor chimeras, which could be activated by heterologous ligands (Borg et al., 1995; Fournier et al., 1995; Pajusola et al., 1994). In these studies SHC and GRB2 were shown to interact strongly with FLT4, although SHC interacted probably not via its SH2 domain (Fournier et al., 1995). No interaction was seen with Ras-GAP, the p85 PI3-K subunit or c-Src. In spite of efficient activation by the heterologous ligand CSF-1, the chimeric receptor elicited a mitogenic response only in NIH3T3 cells, whereas no response was seen in transfected endothelial cell lines (Pajusola et al., 1994).