Molecular basis of endothelial cell morphogenesis in three-dimensional extracellular matrices

Abstract

Although many studies have focused on blood vessel development and new blood vessel formation associated with disease processes, the question of how endothelial cells (ECs) assemble into tubes in three dimensions (i.e., EC morphogenesis) remains unanswered. EC morphogenesis is particularly dependent on a signaling axis involving the extracellular matrix (ECM), integrins, and the cytoskeleton, which regulates EC shape changes and signals the pathways necessary for tube formation. Recent studies reveal that genes regulating this matrix-integrin-cytoskeletal (MIC) signaling axis are differentially expressed during EC morphogenesis. The Rho GTPases represent an important class of molecules involved in these events. Cdc42 and Rac1 are required for the process of EC intracellular vacuole formation and coalescence that regulates EC lumen formation in three-dimensional (3D) extracellular matrices, while RhoA appears to stabilize capillary tube networks. Once EC tube networks are established, supporting cells, such as pericytes, are recruited to further stabilize these networks, perhaps by regulating EC basement membrane matrix assembly. Furthermore, we consider recent work showing that EC morphogenesis is balanced by a tendency for newly formed tubes to regress. This morphogenesis-regression balance is controlled by differential gene expression of such molecules as VEGF, angiopoietin-2, and PAI-1, as well as a plasmin- and matrix metalloproteinase-dependent mechanism that induces tube regression through degradation of ECM scaffolds that support EC-lined tubes. It is our hope that this review will stimulate increased interest and effort focused on the basic mechanisms regulating capillary tube formation and regression in 3D extracellular matrices. Anat Rec 268:252–275, 2002. © 2002 Wiley-Liss, Inc.

The molecular control of blood vessel formation has been a major topic of investigation over the past several decades. Two major processes—vasculogenesis and angiogenesis—are responsible for blood vessel formation in vivo (Hanahan, 1997; Carmeliet and Jain, 2000; Patan, 2000; Conway et al., 2001). Models of these events using in vitro systems are illustrated in Figure 1. Vasculogenesis refers to the de novo development of capillaries from individual endothelial cell (EC) precursors within tissues, or delivered from the circulation (Risau and Flamme, 1995; Drake et al., 1997; Drake and Little, 1999; Drake and Fleming, 2000; Carmeliet and Luttun, 2001). Angiogenesis refers to new blood vessel formation from preexisting vessels (Folkman, 1995; Hanahan, 1997; Carmeliet and Jain, 2000; Conway et al., 2001). Major cytokines that regulate these events are VEGFs, FGFs, angiopoietins, placental growth factor, various chemokines (e.g., SDF-1α) and other growth factors such as TGF-β and insulin-like growth factors (Pepper, 1997; Yancopoulos et al., 2000; Carmeliet et al., 2001; Hellstrom et al., 2001b). Although many of the factors and their receptors that control these processes are well understood (Yancopoulos et al., 2000), little information exists concerning the molecular mechanisms by which ECs physically assemble into capillary tube structures in three-dimensional (3D) extracellular matrix (ECM) environments. EC morphogenesis is defined here as the process whereby ECs assemble into tubes in 3D extracellular matrices. These events require EC interactions with ECM through integrins, and signaling events involving cytoskeletal elements that control EC shape and cell–cell interactions that dictate the 3D structure of tubes. How these interactions lead to the ability of ECs to assemble into tubes with a fluid-filled lumen, an abluminal surface in contact with basement membrane matrix, and cell–cell junctional contacts remains unclear (Figs. 1 and 2A). However, recent advances in the ability to construct EC-lined tubes using defined in vitro systems, and the ability to manipulate the expression of individual genes within primary ECs now allow for a comprehensive examination of how EC morphogenesis is controlled at a molecular level (Davis and Camarillo, 1996; Bell et al., 2001; Bayless and Davis, 2002; Davis and Bayless, 2003).

Figure 1. In vitro EC morphogenesis assay systems in collagen matrices that mimic blood vessel formation through vasculogenesis or angiogenic mechanisms. A: ECs were seeded within collagen matrices as single cells and were allowed to develop interconnecting networks of tubes for 48 hr (left panels) (Davis and Camarillo, 1996). In other assays, ECs were seeded on the surface of collagen matrices and were allowed to invade and undergo morphogenesis for 72 hr (right panels) (Davis et al., 2000b). Cultures were photographed from a side or bottom view at a focal plane beneath the EC monolayer. Bar = 100 μm. B: ECs were seeded within collagen matrices as single cells and allowed to develop vacuoles (8 hr) and tubes (24–48 hr) over time. Bar = 50 μm. In all cases, the cultures were fixed with glutaraldehyde and stained with toluidine blue.

Figure 2. Electron micrographs of developing capillary tubes in 3D collagen matrices. A: EC-collagen gel cultures were established, and at various times cultures were fixed with glutaraldehyde and embedded in plastic. Thin sections were prepared and examined by electron microscopy. After 72 hr of culture, EC cell–cell junctions were identified; arrows point to two junctional contacts. Higher-power images of these two junctions (JXN-1 and 2) are shown above and below the figure. B: Cultures were fixed at 24 hr. A C-shaped EC is observed surrounding a lumenal space. Arrowheads: EC processes available for junction formation, as illustrated in part A. Arrows: interface between the collagen matrix and the lumenal compartment. Bar = 5 μm.

These latter issues concerning the molecular control of EC tube assembly in 3D extracellular matrices are the subject of this review. We investigate how a matrix-integrin-cytoskeletal (MIC) signaling axis regulates EC morphogenesis, and how critical cytoskeletal signaling molecules (the Rho GTPases) regulate these events. In addition, mechanisms underlying EC lumen formation and sprouting, which represent the major morphologic changes regulating EC morphogenesis, are discussed. Additional topics include how differential gene expression regulates EC morphogenesis, and how a balance of genes, molecules, and cells (such as pericytes) regulate capillary tube formation and maintenance by controlling the processes of EC morphogenesis vs. regression.

MIC Signaling Axis Controls EC Morphogenesis in Three Dimensions

It is apparent from recent work that the MIC signaling axis is a major pathway regulating EC morphogenesis in 3D extracellular matrices (Fig. 3) (Davis and Camarillo, 1996; Salazar et al., 1999; Bayless et al., 2000; Bell et al., 2001; Rupp and Little, 2001; Bayless and Davis, 2002; Davis and Bayless, 2003). As shown in Figure 3, this pathway depends not only on EC exposure to exogenous ECM environments conducive for vascular formation such as collagen type I or fibrin matrices, but also on endogenous synthesis of ECM by ECs (Nicosia and Madri, 1987; Ingber and Folkman, 1988; Maragoudakis et al. 1988; Iruela-Arispe et al., 1991; Sephel et al., 1996; Bonanno et al., 2000; Bell et al., 2001). One of the important conclusions drawn from these studies is that multiple integrins, as well as a number of ECM environments, are permissive for EC morphogenesis. It is clear that both β1 and αv integrins can support vascular morphogenesis, including α2β1, α5β1, α1β1, α6β1, and αvβ3 (Bauer et al., 1992; Brooks et al., 1994; 1995, 1996; Bloch et al., 1997; Senger et al., 1997, 2002; Bayless et al., 2000; Rupp and Little, 2001). The involvement of particular integrins is dependent on the matrix environment to which ECs are exposed, and suggests that signaling pathways common to multiple integrins are capable of directing the MIC axis required for capillary tube assembly in three dimensions. The α2β1 and α1β1 integrins regulate EC morphogenesis in collagen-rich ECM (Davis and Camarillo, 1996; Senger et al., 1997, 2002), α5β1 and αvβ3 in fibronectin-fibrin-rich ECM (e.g., in wound tissue matrix) (Bayless et al., 2000; Kim et al., 2000), and α6β1 in laminin-rich ECM environments (Bauer et al., 1992; Davis and Camarillo, 1995). The ability of multiple integrins to regulate this process creates an important integrin signaling redundancy that may be necessary for blood vessel assembly in different ECM environments. ECM environments vary considerably between tissues (e.g., skin vs. brain) and during development or various types of tissue injury wherein new blood vessel formation occurs (Senger, 1996; Sage, 1997; Davis et al., 2000a).

Figure 3. The MIC signaling axis regulates capillary tube morphogenesis and regression in 3D matrix environments. Molecular regulators of these events are indicated next to where they are thought to act during these events.

Signaling events downstream of ECM–integrin interactions clearly are of critical importance to EC morphogenesis. This signaling affects EC survival, proliferation, migration, shape, and differentiation (Shattil and Ginsberg, 1997; Sastry and Burridge, 2000; Schwartz and Shattil, 2000; Mettouchi et al., 2001). All of these processes are required during EC morphogenesis. Many of the molecules regulating integrin signaling are associated with the actin or microtubule cytoskeletal scaffold (Kiosses et al., 1999; Lee et al., 1999; Schwartz and Shattil, 2000; Paik et al., 2001). In addition, these cytoskeletal molecules control cell shape and are centrally relevant to the question of how capillary tubes form in three dimensions. Our laboratory has identified three actin regulatory proteins that are coordinately upregulated during EC morphogenesis: gelsolin, vasoactive-stimulated phosphoprotein (VASP), and profilin (Salazar et al., 1999). Gelsolin is well known to sever actin filaments, thereby participating in actin assembly/disassembly reactions; profilin binds to VASP (a major phosphoprotein target of protein kinases A and G) (Reinhard et al., 2001), and also sequesters actin monomers. In addition, profilin is involved in Rho GTPase-stimulated actin polymerization (Bishop and Hall, 2000). An interesting question concerns the extent to which actin microfilaments or microtubules are responsible for the 3D shape of capillary tubes in ECM. It is notable that both actin and microtubule structures are directly affected by integrin–ECM interactions (Sastry and Burridge, 2000; Goode et al., 2000). A direct influence of integrins on microtubules has only recently been appreciated (Volkov et al., 2001; Zhou et al., 2001). To assess the relative contribution of actin microfilaments vs. microtubules in the 3D structure of capillary tubes (Bayless et al., unpublished results), established capillary tube networks in collagen matrices were treated with either cytochalasin B (to disrupt actin) or nocodazole and/or colchicine (to disrupt microtubules) (Fig. 4). The microtubule disrupting drugs induced rapid collapse of tubes, while actin depolymerization caused minimal changes in the overall tube structure (Fig. 4) (Bayless et al., unpublished results). These data suggest that microtubules are critical to the overall shape of capillary tubes in 3D ECM, and that actin is not as important in regulating this overall structure.

Figure 4. Disruption of microtubules induces rapid collapse of capillary tubes in 3D collagen matrices. EC-collagen gel cultures were established, and after 48 hr the cultures were left untreated or were treated with cytochalasin B (10 μM), colchicine (10 μM), and nocodazole (10 μM) for 30 min. Cultures were then fixed with glutaraldehyde and photographed. Bar = 50 μm.

Major signaling molecules downstream of integrin–ECM interactions are the Rho GTPases, a family of small GTPases in the Ras superfamily that regulate cytoskeletal structure and function (Kaibuchi et al., 1999; Bishop and Hall, 2000; Hall and Nobes, 2000; Schwartz and Shattil, 2000; Takai et al., 2001; Ridley, 2001a). These molecules regulate one of a series of signaling pathways resulting from integrin–ECM interactions (Giancotti and Ruoslahti, 1999; Sastry and Burridge, 2000; Geiger et al., 2001; Schwartz and Assoian, 2001). Rho GTPases initially were shown to control specific events in actin cytoskeletal dynamics, as reported in the seminal studies of Ridley and Hall (Ridley et al., 1992; Ridley and Hall, 1992). In addition, these GTPases regulate not only the function of the actin cytoskeleton, but also the microtubule and intermediate filament cytoskeletons (Nobes and Hall, 1995; Inada et al., 1999; Meriane et al., 2000; Daub et al., 2001; Palazzo et al., 2001; Ridley, 2001a). They affect many critical steps in cell behavior that are characteristic of EC morphogenesis (i.e., cell migration, cell proliferation and regulation of gene expression, cell shape, permeability, and polarity) (Kaibuchi et al., 1999; Hall and Nobes, 2000; Ridley, 2001a; Settleman, 2001; Takai et al., 2001; Wojciak-Stothard et al., 2001). However, until recently, no direct link between their function and EC morphogenesis in 3D ECM has been made (Bayless and Davis, 2002). The RhoA GTPase has been reported to induce actin stress fiber formation and focal adhesions, and to stimulate acto-myosin contractility and microtubule elongation (Palazzo et al., 2001, Ridley, 2001a). The Rac1 GTPase induces lamellipodia formation, while Cdc42 induces filopodia. Both Rac1 and Cdc42 have been reported to increase microtubule stability by inducing the phosphorylation of stathmin through PAK-1 (Daub et al., 2001). Furthermore, Rho GTPases have been reported to regulate a variety of other cellular processes, such as vesicular trafficking events (including endocytosis, macropinocytosis, and phagocytosis) (Greenberg, 1995; Swanson and Watts, 1995; Garrett et al., 2000; Cardelli, 2001; Ridley, 2001b). In this latter case, RhoA was observed to regulate integrin-mediated phagosome formation while Rac1 and Cdc42 regulate Fc-receptor-mediated phagosome formation (Caron and Hall, 1998). The processes of macropinocytosis and phagocytosis appear to be related to EC intracellular vacuole formation and coalescence, which is an important step in the lumen formation pathway (Davis and Camarillo, 1996; Bayless and Davis, 2002; Davis and Bayless, 2003) (see below). Furthermore, we have recently shown that Cdc42 and Rac1, which regulate both macropinocytosis and phagocytosis, also control the formation of capillary lumens in collagen or fibrin matrices (Bayless and Davis, 2002).

In Vitro Models of Capillary Morphogenesis in 3D Extracellular Matrices

In order to determine how ECs assemble into capillary tubes in 3D space, models of this process are necessary. Thus far, it has been difficult to elucidate how capillary tubes form in vivo, so there is an increasing requirement for utilization of in vitro models that permit a molecular analysis of these events. In vitro systems have the following advantages: 1) defined experimental conditions can be achieved to test hypotheses and define genetic, protein, and morphologic changes in a coordinated fashion; 2) the EC population is relatively uniform; and 3) the function of individual genes during particular events in EC morphogenesis can be directly addressed. In vitro models represent a rapid, defined, and efficient experimental strategy to elucidate the molecular events required for tube formation in 3D extracellular matrices. Although similar experiments can be performed in vivo, it is difficult to define morphogenic steps or assess the role of individual genes/molecules in a complex tissue wherein ECs represent only a small fraction of the total cells. It is increasingly evident that a balanced experimental approach using both in vitro and in vivo morphogenic systems is necessary to elucidate the molecular basis of capillary tube assembly. Some important questions are raised: What is the minimum molecular machinery necessary for EC tube formation in three dimensions? Are EC-specific molecules necessary for these events, and what are these molecules? How are EC lumen formation and EC sprouting/branching events controlled at a molecular level?

In the early to mid-1980s, Montesano and colleagues established capillary morphogenesis assays in vitro that strongly recapitulated the appearance of capillary tube structures in vivo using ECs alone (Montesano and Orci, 1985; Montesano, 1992; Montesano et al., 1992). In these assays, primary ECs were seeded as monolayers onto the surface of collagen or fibrin gels, and over a period of days the ECs were observed to invade these matrices to form capillary tube structures. Histologic and electron microscopic analyses revealed capillary tubes with a central fluid-filled lumenal compartment, with a defined abluminal surface surrounded by ECM. Later, Pepper and Montesano (1990) showed that known angiogenic cytokines such as VEGF and FGF-2 (which together showed strong synergism) stimulated the invasion and formation of capillary tubes from EC monolayers, further demonstrating the relevance of these assays (Pepper et al., 1992). Other models, such as that described by Nicosia and colleagues (Nicosia et al., 1982; Nicosia and Madri, 1987; Nicosia and Ottinetti, 1990; Nicosia and Villaschi, 1999), showed similar capillary tube sprouting events in fibrin or collagen matrices, using the aortic ring model, in response to angiogenic cytokines. Other morphogenic assays developed by Vernon et al. (1992) and Vernon and Sage (1995), and separately by Davis and Camarillo (1995), studied planar EC morphogenesis, whereby ECs were allowed to rearrange to form interconnecting networks of EC cords on the surface of a reconstituted basement membrane matrix. These assays demonstrated how mechanical forces exerted by ECs on ECM (i.e., through the MIC signaling axis) can rapidly induce the assembly of EC network formation by distortion of ECM into “matrical pathways” (Vernon and Sage, 1995) or “matrix guidance pathways” (Davis and Camarillo, 1995) that allow directed EC migration toward neighboring cells. These matrix guidance pathways may play a key role in how ECs locate each other in 3D space and rapidly form multicellular networks in tissues.

More recent assays, such as those developed by our laboratory, have utilized ECs suspended as single cells in 3D extracellular matrices (Davis and Camarillo, 1996; Salazar et al., 1999; Bayless et al., 2000; Bell et al., 2001; Bayless and Davis, 2002) (Figs. 1 and 5). In Figure 5B, a series of time-lapse images reveal two ECs forming intracellular vacuoles, and then the cells interconnecting to form a lumenal structure. Related assays were also developed by Marx et al. (1994) and Ilan et al. (1998). These assays may more closely reflect the process of vasculogenesis as opposed to angiogenesis, since single cells in 3D space assemble into tubes. Examples of the assay systems that mimic vasculogenesis (Davis and Camarillo, 1996) vs. angiogenesis (Davis et al., 2000b) are shown in Figure 1. An experimental advantage of these “vasculogenic” assays is that essentially all of the ECs participate in capillary tube formation, while only a fraction of the ECs in the assays described above (i.e., more similar to angiogenesis, in which subsets of cells invade from a monolayer to undergo morphogenesis) proceed through the morphogenic process. The “vasculogenic” assays have allowed for a comprehensive analysis of differential gene expression during morphogenesis in three dimensions, and the isolation of differentially expressed novel genes (i.e., capillary morphogenesis genes (CMGs)) (Bell et al., 2001) (see below) since the entire population of ECs undergoes morphogenesis during a similar time course. In addition, this analysis of differential gene expression can be correlated with distinct steps in the EC morphogenic process (Fig. 6). Also, EC gene expression can be manipulated using recombinant adenoviruses to address the role of particular genes during these events (Bell et al., 2001; Bayless and Davis, 2002).

Figure 5. Cellular events during distinct stages of EC morphogenesis in vivo and their relationships to in vitro morphogenesis systems. A: ECs convert from a quiescent to an activated state during angiogenesis; then, following morphogenesis and differentiation, they convert back to a quiescent state. B: Time-lapse photographs of EC morphogenesis showing EC lumen development. Two ECs were photographed at the indicated times using phase-contrast microscopy. Arrowheads: EC intracellular vacuoles. Bar = 30 μm.

Figure 6. Steps in EC morphogenesis in 3D extracellular matrices. ECs were suspended in collagen matrices for the times indicated. The cultures were fixed, stained with toluidine blue, and photographed. The role of Rho GTPases and pericytes/VSMCs in regulating different stages of morphogenesis are indicated. Bar = 50 μm.

An interesting question is, why does only a subset of ECs participate in sprout formation from an EC monolayer (i.e., “angiogenesis” assay) (Fig. 1A, right panel)? This result is in marked contrast to that observed in the “vasculogenesis” type of assay, wherein essentially all of the ECs participate in the morphogenic response (Fig. 1A, left panel). The phenomenon whereby only a small subset of ECs sprout and form tubes from a monolayer appears similar to a phenomenon observed in neural or hair development, called lateral inhibition (Lindsell et al., 1996; Lanford et al., 1999). Lateral inhibition refers to a situation in which subsets of cells produce factors that inhibit their neighboring cells (subsets of cells that express receptors for the inhibitory factor), which results in selective differentiation of cells in defined spatial areas. For example, these types of interactions regulate the density and spatial arrangement of hair or ciliated cells in the skin and cochlea (Deblandre et al., 1999; Lanford et al., 1999; Renaud and Simpson, 2001). It is interesting to consider that hair or ciliated cell density is analogous to vascular sprout density during angiogenesis. Varying the levels of these inhibitory factors directly influences the density of hair follicles in vivo (Lanford et al., 1999). Thus, these types of molecules could play a role in the control of EC sprout density (as illustrated in Fig. 1A, right panels) through such a lateral inhibition mechanism. Molecules that regulate lateral inhibition include Notch receptors and Notch ligands, such as Jagged and Delta (Lindsell et al., 1996). Zimrin et al. (1996) initially showed that Jagged-1 was upregulated in ECs during morphogenesis, and negatively regulated the degree of EC sprouting and invasion. More recently, Jagged-1 was shown to be highly upregulated during EC morphogenesis in our “vasculogenic” assay (it was the most upregulated gene at 8 hr in more than 7,000 genes screened in a DNA microarray analysis) (Bell et al., 2001). Other studies have shown that the Notch receptors (Notch-4 and Notch-1) are present in ECs during morphogenic events (Zimrin et al., 1996; Uyttendaele et al., 2000; Linder et al., 2001). This pathway regulates the related process of Drosophila tracheal tube development (which is related to EC morphogenesis) (Metzger and Krasnow, 1999) through a lateral inhibition mechanism (Ikeya and Hayashi, 1999; Llimargas, 1999). Further work is necessary to investigate how these molecules regulate EC morphogenesis, and to determine whether they play a role in distinguishing the molecular events that characterize angiogenesis vs. vasculogenesis.

In vitro EC morphogenesis models are also useful for studying how ECs physically assemble into tubes during this process. Currently, little information exists concerning these EC assembly events. Several assay systems appear to be particularly relevant. The first, as described by Vernon and Sage (1999), places beads containing ECs into collagen gels and allows EC sprouting and morphogenesis to proceed radially. Images obtained from this assay should allow for confocal microscopic imaging of sprouts and tubes over time, without optical interference from cells at different focal planes above or below the sprout. Tube formation could be analyzed with time to determine how ECs interconnect with each other and change shape to form tubes. We have recently described a horizontal EC invasion assay that allows for observation of EC sprouting and tube formation without optical interference from the EC monolayer (Davis et al., 2000b). Another assay that appears promising for this purpose is that described by Bayless and Davis (2002). In this assay, fluorescently-labeled individual ECs are suspended in collagen gels, and 1-μl dots of cell-gel mix are placed on coverslips and inverted onto chambers containing culture media. Under these conditions, ECs form interconnecting networks of tubes, as in other 3D assays. These cultures can be visualized as living cultures or following fixation using multiphoton confocal microscopy (Fig. 7) (Bayless and Davis, 2002). To facilitate these experiments, we recently developed GFP-Rho GTPase fusion targeting reagents that label various intracellular components within ECs, including intracellular vacuole membranes (i.e., which regulate lumen formation) (Bayless and Davis, 2002). The development of these new technologies makes it possible to determine how groups of ECs physically assemble into capillary tube networks over time.

Figure 7. EC intracellular vacuole membranes label with GFP-Rac1 during EC morphogenesis in 3D collagen matrices. ECs were infected with a recombinant adenovirus carrying a GFP-Rac1V12 fusion gene. After 48 hr they were cultured in 3D collagen matrices. After 24 hr, cultures were fixed with paraformaldehyde and examined by confocal microscopy. A single optical section of a developing EC lumenal (L) structure is shown. Arrows: GFP-Rac1-labeled vacuole membranes. Arrowhead: the lumenal membrane. Bar = 10 μm.

Critical Steps in the Formation of Interconnecting Networks of EC-Lined Tubes During Capillary Morphogenesis

In vitro models can be utilized to directly test the role of particular molecules in defined steps of capillary tube network formation (Fig. 6). Data from many laboratories indicate a critical role for integrin–ECM interactions, the cytoskeleton, and membrane-type matrix metalloproteinases (MT-MMPs) in these events. Two major morphologic changes that regulate EC tube development include lumen formation and branching/sprouting, which controls how ECs interconnect into networks in three dimensions (Fig. 6). Our data, as well as those of others, suggest that these distinct morphologic steps may be separable. For example, during EC morphogenesis on the surface of basement membrane matrix gels, dramatic branching and network formation occurs with little evidence of lumen formation (Vernon et al., 1992; Davis and Camarillo, 1995). Also, increased expression of the Rac1 GTPase in either its wild type or constitutively active form in ECs using recombinant adenoviruses increases vacuole and lumen formation, but there is little branching and sprouting (Bayless and Davis, 2002). In contrast, increased expression of Cdc42 stimulates vacuole and lumen formation (Bayless and Davis, in press), as well as branching and sprouting, while increased expression of RhoA results in marked increases in EC sprouting with little lumen formation (Bayless and Davis, unpublished results). Using a different experimental approach, another recent study (Kiosses et al., 2002) proposed a role for PAK1 (a Rac1 and Cdc42 downstream effector) in branching phenomena during EC morphogenesis. Much work is needed to elucidate the roles of individual Rho GTPases and their downstream effectors at different stages of morphogenesis (Fig. 6).

It would be interesting to determine the temporal relationship of EC sprouting to lumen formation. During angiogenesis, it is believed that EC sprouts occur first, providing provisional EC networks which then progress to form lumens. In our in vitro model, which more closely mimics a vasculogenic response, the opposite sequence of events occurs. Individual ECs develop vacuoles and lumens first, followed by sprouting events that lead to interconnected tubes (Figs. 5B,6–8). However, it should be pointed out that vacuoles remain visible in sprouting ECs, and we believe they contribute to the process of lumens extending in the direction of EC sprouts during branching morphogenesis (Davis and Camarillo, 1996; Davis et al., 2000b). Recent experiments using our horizontal invasion assay system reveal that intracellular vacuoles are present and participate in lumen formation during initial sprouting events (Bayless and Davis, unpublished results). These data suggest that the initiation of lumen formation may occur concurrently with sprouting.

Figure 8. Time course of EC lumen formation through intracellular vacuole formation and coalescence during morphogenesis in 3D collagen matrices. ECs were placed into collagen matrices and at the indicated times were fixed with glutaraldehyde, embedded in plastic, thin-sectioned, and stained with toluidine blue. Representative fields were photographed. Arrowheads: open lumenal structures (C-shaped ECs) at 24 hr. Arrows: lumenal structures with a continuous EC lining. Bar = 25 μm.

An important question that has been addressed over the years concerns the role of ECM-degrading proteinases in EC morphogenesis (Pepper and Montesano, 1990; Haas and Madri, 1999; Werb et al., 1999; Pepper, 2001; Sternlicht and Werb, 2001). Recent work (Hiraoka et al., 1998; Hotary et al., 2000) has provided very compelling evidence of a requirement for MT-MMPs in capillary morphogenesis. In addition, MMP-2 and MMP-9 knockout experiments have revealed roles for these proteinases during angiogenesis in vivo (Vu et al., 1998; Bergers et al., 2000). Hiraoka et al. (1998) and Hotary et al. (2000) have reported a critical role for MT-MMP-1 and other MT-MMPs as pericellular proteinases which control EC and epithelial cell invasion and morphogenesis in 3D ECM gels. In contrast, many of the secreted MMPs, and even mutated MT-MMPs without a membrane anchor, were unable to support invasion and morphogenesis (Hiraoka et al., 1998; Hotary et al., 2000). Experiments from our laboratory also support these conclusions. When ECs are suspended as single cells in collagen gels, MMP inhibitors that block MT-MMPs (such as TIMP-2 and the chemical inhibitor GM6001) markedly block EC morphogenesis while TIMP-1, PAI-1, aprotinin, and other inhibitors have no effect (Davis et al., unpublished data). In our “angiogenesis”-like assay model (Davis et al., 2000b) (see Fig. 1A, right panels), GM6001 completely inhibited invasion and morphogenesis, while TIMP-2 blocked morphogenesis (i.e., lumen and tube formation) but did not fully block invasion of individual ECs (Bayless et al., unpublished results). In contrast, TIMP-1 did not block this process. The effect of TIMP-2 may relate to the discussion above regarding how EC sprouting/invasion may be regulated separately from lumen formation, and suggests a complex role for MMPs in EC morphogenesis. Another intriguing aspect of the role of MMPs in morphogenesis is that they can expose matricryptic sites in ECM proteins that can regulate cell responses (Sage, 1997; Davis et al., 2000a). Matricryptic sites are biologically active cryptic domains in ECM molecules that are not exposed in mature ECM molecules but are exposed following conformational or enzymatic modifications (Davis et al., 2000a). Recently, Xu et al. (2001) reported that matricryptic sites in collagen type IV regulate injury- and tumor-induced angiogenic responses. Antibodies that recognized these matricryptic epitopes in collagen type IV were able to block in vivo angiogenic responses.

Another question concerns the possible role of secreted proteinases in EC morphogenesis. In addition to their role in degrading ECM, many of these proteinases have been found to have alternate substrates, which could create new roles for these molecules in this process (Sternlicht and Werb, 2001). For example, MMP-9 was recently observed to liberate ECM-associated VEGF, which induces angiogenesis (Bergers et al., 2000). Another role for secreted EC proteinases is to regulate the process of capillary tube regression (Davis et al., 2001), a process that normally follows the wound-healing angiogenic response (Clark, 1996). As discussed below, under appropriate conditions, ECs are capable of degrading the ECM scaffold in which they are suspended, causing regression of capillary tubes (Davis et al., 2001).

Regulation of Capillary Lumen Formation in Three Dimensions by Intracellular Vacuole Formation and Coalescence: An Integrin and Rho GTPase-Dependent Pinocytic Event

A major direction of our current work is to elucidate how ECs form lumens in 3D ECM environments. We have shown in a series of studies that a major mechanism of lumen formation of ECs in either 3D collagen or fibrin matrices involves intracellular vacuole formation and coalescence (Davis and Camarillo, 1996; Bayless et al., 2000; Davis et al., 2000; Bayless and Davis, 2002; Davis and Bayless, 2003). EC intracellular vacuoles have also been observed both in vivo and in vitro by many investigators (Speidel, 1933; Clark and Clark, 1939; Wolff and Bar, 1972; Guldner and Wolff, 1973; Dyson et al., 1976; Wagner, 1980; Folkman and Haudenschild, 1980; Montesano and Orci, 1988; Konerding et al., 1992; Meyer et al., 1997; Yang et al., 1999; Dvorak and Feng, 2001). As shown in Figures 5B and 6–8, EC vacuole formation and coalescence occurs over time during capillary morphogenesis to regulate the lumen formation process. In Figure 8, cross-sections of fixed cultures at various times illustrate the formation of intracellular vacuoles and lumens in collagen matrices. An electron micrograph of an EC with intracellular vacuoles during morphogenesis is shown in Figure 9. As shown in the figure, collagenous matrix is observed on the preliminary abluminal surface, while the interior of the vacuoles are devoid of ECM. These findings are supported by previous work that suggested a role for an intracellular lumen formation mechanism regulating EC or epithelial cell lumen formation. Wolff and Bar (1972) and Guldner and Wolff (1973) described “seamless” ECs in vivo that contain a lumen and form junctional contacts with adjacent ECs, but contain no junctional site in cross-section. This type of lumen has to form through an intracellular mechanism, such as that observed during intracellular vacuole formation and coalescence (Davis and Camarillo, 1996). Similar “seamless” epithelial cells are observed in the distal tips of the Drosophila tracheal tube system (Samakovlis et al., 1996). Another example of intracellular lumen formation is that observed in the excretory epithelial cell in C. elegans, which is a single cell with a lumen (Buechner et al., 1999).

Figure 9. Electron micrograph of intracellular vacuoles regulating lumen formation during EC morphogenesis in 3D collagen matrices. ECs were placed into collagen matrices, and after 24 hr cultures were fixed and processed for electron microscopy (Davis and Camarillo, 1996). Asterisks: the fluid-filled intracellular vacuole space. Arrowheads: small EC processes observed on the abluminal surface and in the interior of EC vacuoles, indicating pinocytosis of plasma membrane. Arrows: collagen fibrils. Bar = 5 μm.

The in vitro systems developed in our laboratory enable a detailed molecular analysis of the vacuole and lumen formation process. Thus far, our findings indicate that vacuole formation depends on integrin interactions with ECM, and occurs through a pinocytic mechanism that requires both the actin and microtubule cytoskeletons (Fig. 10). The addition of cytochalasin B or nocodazole, which disrupt the actin and microtubule cytoskeletons, respectively, completely blocks vacuole formation and subsequent lumen formation. We have previously shown that integrin–ECM interactions are required for intracellular vacuole and lumen formation. The α2β1 integrin regulates vacuole formation in collagen matrices, and a combination of αvβ3 and α5β1 regulates these events in fibrin matrices (Davis and Camarillo, 1996; Bayless et al., 2000). We originally showed that vacuole formation occurs through a pinocytic mechanism. Plasma membrane markers were detectable in vacuole membranes, and when membrane-impermeant fluorescent dyes (i.e., dextran-fluorescein) were added to the culture media, they strongly labeled the pinocytosed vacuole compartment (Davis and Camarillo, 1996). Plasma membrane markers present in EC intracellular vacuole membranes include PECAM, caveolin-1 (Fig. 11), ICAM-1, and VCAM-1 (Davis and Bayless, 2003). Also, β-catenin, a VE-cadherin and PECAM-associated molecule (Ilan et al., 2000), is associated with vacuole membranes, and von Willebrand factor is detectable within many vacuoles (Fig. 11). This suggests that Weibel-Palade bodies (i.e., which contain von Willebrand factor) fuse with the developing vacuole compartment and contribute membrane and their fluid contents. The presence of caveolin-1 also suggests the contribution of another intracellular vesicular trafficking compartment, caveolae, to these vacuoles (Anderson, 1998). Thus far, we have unable to detect clathrin, early endosomal markers such as EEA-1 or rab5, or lysosomal markers such as LAMP in vacuole membranes.

Figure 10. Pinocytosis of fluorescent dyes into EC intracellular vacuoles with vacuole membrane-associated GFP-Rac1 GTPase. ECs were infected with a recombinant adenovirus carrying a GFP-Rac1V12 fusion gene. After 48 hr they were cultured in 3D collagen matrices in the presence of carboxyrhodamine in the culture medium (Bayless and Davis, 2002). After 24 hr, cultures were fixed and then photographed using fluorescence microscopy. Arrows: GFP-Rac1 labeling of EC vacuole membranes. Arrowheads: carboxyrhodamine labeling of EC vacuoles, indicating that they arise through pinocytosis. Bar = 10 μm.

Figure 11. Immunocytochemical characterization of EC intracellular vacuoles. ECs were cultured in collagen matrices, and after 8 hr they were digested out of the collagen matrices, plated on glass coverslips, fixed with paraformaldehyde, and immunofluorescently stained with antibodies directed to PECAM, caveolin-1, and von Willebrand factor (vWF). Also, some cells were stained with phalloidin-fluorescein to detect F-actin. Arrows: intracellular vacuole membrane staining. Arrowhead: area of increased caveolin-1 staining between EC vacuoles. Bar = 25 μm.

An important consideration is, how do EC intracellular vacuoles enlarge and fuse during the lumen formation process? Do they solely enlarge by fusion with other vesicles, or do they enlarge by accumulation of ions such as Na⁺ or H₂O? DNA microarray or differential display studies have shown that a series of genes associated with sodium and water transport are upregulated during EC morphogenesis (Bell et al., 2001). These genes include NaHCO₃ cotransporters 2 and 3, and melanin-concentrating hormone, which have been reported to regulate ion or water transport (Hawes et al., 2000; Soleimani and Burnham, 2001). It appears that EC intracellular vacuoles represent a novel pinocytic intracellular compartment that is utilized to regulate the lumen formation process. It remains to be determined how EC intracellular vesicles, such as Weibel-Palade bodies and caveolae, contribute to the developing lumen by fusing with pinocytic vesicles that arise through integrin-dependent signaling. It is also important to identify which vesicular fusion pathways, such as those controlled by Rab GTPases (Zerial and McBride, 2001), regulate the fusion of pinocytic vacuoles to each other or to other EC vesicles.

Another event we observe during EC morphogenesis is a presumed exocytic event in which intracellular vacuole fusion occurs with the plasma membrane. This event creates ECs with an open or C-shaped appearance, with processes surrounding an ECM-free space. Such images are frequently observed in electron micrographs and cross-sections of cultures after 24 hr of morphogenesis (Figs. 2B and 8). The EC processes that are generated following exocytosis would be free to interact with each other to form an intra- or intercellular junctional contact with adjacent ECs (Fig. 2). This mechanism allows ECs to form a lumenal compartment through pinocytosis and coalescence of vacuoles, and then, following exocytosis, creates a mechanism for ECs to form a reversible junctional contact site (Davis and Bayless, 2003). Our observation that single ECs can form a lumen and junctional contact are confirmed by the presence of similar cells present during Drosophila tracheal tube morphogenesis (Samakovlis et al., 1996). The junctional site could remain with a single EC or be a site where new ECs are recruited to form multicellular lumenal structures. In the image shown in Figure 2, it appears that some retraction of the processes may have occurred following this exocytic event, as ECM is present (arrows) which is not directly in contact with the EC processes (Fig. 2B, arrowheads). Extension of these processes along the ECM-fluid interface would occur until the processes contact each other or encounter similar processes from adjacent ECs. Thus, this mechanism could create either single ECs with junctions or multicellular capillary tube structures with junctional contacts, as shown in Figure 2A. One of the elegant features of these mechanisms is that the lumenal spaces are already fluid-filled and free of ECM. These mechanisms described in EC morphogenesis may be of general utility for lumen formation for either endothelial or epithelial cells.

Regulation of EC Vacuole and Lumen Formation by Rho GTPases

As discussed above, the EC intracellular vacuole formation process (Davis and Camarillo, 1996) is reminiscent of other pinocytic events, such as macropinocytosis and phagocytosis. Like vacuole formation, both of these latter processes are dependent on actin and microtubules (Greenberg, 1995; Davis and Camarillo, 1996; Cardelli, 2001). In addition, the inner surfaces of the pinocytosed membranes contain F-actin (Greenberg, 1995; Davis and Camarillo, 1996; Cardelli, 2001) (Fig. 11). Recent work has elucidated some of the molecular components involved in the cytoskeletal signaling pathways regulating macropinocytosis and phagocytosis. In both cases, Rho GTPases regulate these events. Rac and Cdc42 GTPases have been found to be required for macropinocytosis (Cardelli, 2001; Ridley, 2001a) and Fc receptor-mediated phagocytosis (Caron and Hall, 1998). In contrast, the Rho GTPase regulates integrin-dependent phagocytosis (Caron and Hall, 1998; Chimini and Chavrier, 2000). Because of the similarities of these events to EC vacuole formation, we tested the hypothesis that Rho GTPases are required for EC vacuole and lumen formation in 3D ECM environments.

These studies were recently described by Bayless and Davis (2002) using human ECs cultured in either collagen or fibrin matrices. Blockade of all three Rho GTPases (Rho, Rac, and Cdc42) with C. difficile toxin B completely blocked EC vacuole and lumen formation in collagen and fibrin gels. In contrast, selective blockade of Rho with the C3 exoenzyme did not block vacuole and lumen formation. These data strongly indicate a role for Rac and Cdc42 in these events. This conclusion is supported by our data using recombinant adenoviruses that induce expression of dominant negative Rho GTPases in ECs during the morphogenic process. Dominant negative Rac1 and Cdc42 markedly blocked EC vacuole and lumen formation in collagen and fibrin matrices, while dominant negative RhoA had no effect. In addition, constitutively active Cdc42 also blocked morphogenesis. Previous studies have shown that either dominant negative or constitutively active forms of these GTPases can in some instances block Rho GTPase-mediated cellular events (Tzuu-Shuh and Nelson, 1998; Tzuu-Shuh et al., 1998). N-WASP, a downstream effector of Cdc42, is known to be activated by Cdc42 and PIP₂. The verprolin cofilin activation (VCA) domain of N-WASP then binds the Arp2/3 complex to stimulate actin polymerization (Rohatgi et al., 1999; Prehoda et al., 2000). Expression of the N-WASP VCA domain alone (i.e., which mimics N-WASP activation and stimulates Arp2/3 binding and actin polymerization) blocks EC vacuole formation (Bayless and Davis, 2002) like constitutively active Cdc42. These data indicate a role for Cdc42 and Rac1 in regulating EC lumen formation through intracellular vacuole formation and coalescence in 3D ECM.

Additional experiments examined the distribution of Rac1, Cdc42, and RhoA within cells during the EC morphogenic process. A series of recombinant adenoviruses were constructed to express either constitutively active or wild-type Rac1, Cdc42, or RhoA fused to GFP (Bayless and Davis, 2002). ECs were induced to express these chimeric proteins, and morphogenesis assays were performed. Cultures at various times of morphogenesis were examined by confocal or conventional fluorescence microscopy (Figs. 7 and 10). In some cases, the fluorescent dye carboxyrhodamine was added to the culture medium to label pinocytic EC vacuoles. In Figure 10, an EC cell is shown with multiple intracellular vacuoles that are labeled with carboxyrhodamine. The vacuole membranes are labeled with GFP-Rac1, showing that Rac1 targets to the pinocytosed membranes (Fig. 10). In addition, we have also shown that GFP-Cdc42 targets to vacuole membranes. Furthermore, we have evidence that increased expression of Cdc42 wild-type protein in ECs induces an increase in intracellular vacuoles within ECs compared to control vectors. In addition, Cdc42 protein is upregulated during EC morphogenesis when the majority of lumenal development and expansion occurs (after 24–48 hrs of morphogenesis) (Bayless and Davis, 2002). Collectively, our data indicate that Cdc42 and Rac1 regulate intracellular vacuole formation and coalescence and localize to vacuole membranes during the process. Our ability to label the intracellular vacuole membranes with either GFP-Rac1 or GFP-Cdc42 will enable future studies using real-time confocal microscopy to analyze vacuole development and fusion during EC lumen formation. As shown in the static confocal microscopic image (Fig. 7), GFP-Rac1 targets to vacuoles which appear to be moving from the EC abluminal surface toward the lumenal membrane, where they fuse to expand the lumenal surface. We have observed images similar to the image shown which indicate that small vacuoles may be arising from areas of EC process formation. The known involvement of Rac1 and Cdc42 in cell process formation and vacuole formation is consistent with this observation. These vacuoles could then fuse with the lumen and extend the lumenal space in the direction of the EC sprout. In previous works we have made similar conclusions based on other images (Davis and Camarillo, 1996; Davis et al., 2000b).

Potential Mechanisms for Creating a Lumenal Space During Endothelial and Epithelial Cell Morphogenesis in 3D Extracellular Matrices

Intracellular vacuole formation and coalescence is one mechanism by which EC lumens can form in 3D matrices. Vacuole structures similar to those described in ECs have been reported in epithelial cells. Original studies by Vega-Salas et al. (1987, 1988) showed that apical domains of polarized epithelial cells can be pinocytosed into vacuoles, which may represent an attempt to form an intracellular lumenal space. These vacuole membranes contain apical-selective markers. A recent study by Akhtar and Hotchin (2001) demonstrated targeting of the Rac1 GTPase to apical pinocytosed membrane-bound vacuoles following disruption of E-cadherin mediated cell–cell contacts in MDCK cells. This result appears similar to our findings of Rac1 GTPase targeting to vacuoles during EC lumen formation (Bayless and Davis, 2002).

A number of other mechanisms have been described that may regulate either EC or epithelial cell lumen formation. Well-described models of epithelial cell morphogenesis include assays using either MDCK cells or breast epithelial cells suspended within or on the surface of various ECM substrates (Montesano et al., 1991, 1997; Pollack et al., 1997; Weaver and Bissell, 1999; Yeaman et al., 1999; Lipschutz et al., 2000; O'Brien et al., 2001). How lumens form in epithelial cell clusters remains unclear. A number of possible mechanisms have been proposed for epithelial and EC lumen formation, including apoptosis of centrally placed cells within clusters (Lin et al., 1999), intracellular vacuole formation (such as that described above), autophagy (where apically placed portions of cells are progressively removed through lysosomal degradation of cellular components (Stromhaug and Klionsky, 2001), intussusception (where smaller caliber vessels are generated from larger precursor vessels (Djonov et al., 2000), and membrane-sorting events regulated by endocytic or exocytic mechanisms. A protein complex called exocyst, which regulates exocytosis, has been found to regulate these latter processes during MDCK lumen formation and morphogenesis in 3D collagen matrices (Lipschutz et al., 2000; Lipschutz and Mostov, 2002). Further support for this mechanism is found in studies of Drosophila tracheal development, wherein exocytic membrane fusion events regulate lumenal diameter of tubes (Beitel and Krasnow, 2000). Of note, lumenal diameter was not regulated through changes in cell number. Our findings regarding the probable role of exocytic mechanisms in EC lumen formation strongly correlate with these findings, revealing the importance of exocytic events in lumen formation.

When cells are suspended in ECM or invade into ECM, they occupy physical space within that ECM. Invagination of a localized area of EC plasma membrane from cell-ECM contact sites creates a fluid-filled pocket in the ECM and a fluid-ECM interface. We have observed ECs forming “suction cup”-like areas with ECM in electron micrographs (Davis and Bayless, 2003). EC processes can migrate along the fluid-ECM interface extending from the boundary of the “suction cup” and enclose the fluid pocket, which creates pinocytic EC vacuoles. As expected, these vacuoles can be readily labeled with membrane-impermeant fluorescent dyes. Similar events can regulate vacuole formation between adjacent cells. If two or more cells are in contact with each other, invagination of plasma membrane in an area of cell–cell contact creates an ECM-free space, which could begin the presumptive lumenal space. We have observed this type of lumen formation in small EC clusters. In these clusters, lumens selectively develop in the contact area between the two cells (Davis and Camarillo, 1996) (see Fig. 5B). Thus, both cell–ECM and cell–cell sites can initiate pinocytic events leading to vacuole and lumen formation. Yang et al. (1999) have shown that blocking antibodies directed to the EC cell–cell adhesion molecules, VE-cadherin and PECAM, interfere with EC vacuole and lumen formation.

Similar types of events appear to regulate lumen formation during EC invasion and sprouting, which can be visualized using our “angiogenesis”-like in vitro assays (Fig. 1). As described above for vacuole formation, fluid-filled spaces in the ECM are generated during invasion events (i.e., invasion tunnels are visualized trailing invading ECs) (Davis and Bayless, 2003) (Bayless et al., unpublished results). These appear to be physical tunnels with a fluid-filled center surrounded by ECM. How such tunnels are generated remains to be determined. They may be generated by proteolysis, but also could be created by physical means or phagocytic removal of ECM by the invading cell. The ECs appear to use these fluid-ECM interfaces to migrate on and facilitate the development of the lumenal space during EC invasion and sprouting. They also serve as matrix guidance pathways (Davis and Camarillo, 1995) (Bayless et al., unpublished results) to facilitate the interconnection of ECs during the tube assembly process. Intracellular vacuoles also appear to play a critical role during this process to regulate lumen formation during invasion (Bayless et al., unpublished results).

Regulation of Capillary Morphogenesis in 3D Matrices by Differential Gene Expression

The original assay system of Davis and Camarillo (1996) has proven to be very useful for analyzing differential gene expression during capillary morphogenesis in 3D extracellular matrices (Salazar et al., 1999; Davis et al., 2001; Bell et al., 2001). The examination of differential gene expression has been successfully applied to many biologic systems with the development of new large-scale genetic screening techniques, such as DNA microarray, differential display, and serial analysis of gene expression (Velculescu et al., 1995; DeRisi et al., 1997; Brown and Botstein, 1999; Martin and Pardee, 1999; Maxwell and Davis, 2000; Peale and Gerritsen, 2001; Bell et al., 2001). These approaches have recently been utilized to examine differential gene expression during capillary morphogenesis in vitro (Glienke et al., 2000; Kahn et al., 2000; Bell et al., 2001) and to compare tumor angiogenic vs. normal endothelium in colorectal cancer (St. Croix et al., 2000). These studies have identified changes in ECM, integrin, and cytoskeletal regulatory protein gene expression, underscoring the importance of the MIC signaling axis in EC morphogenesis. In the study by St. Croix et al. (2000), many of the upregulated ECM genes from tumor-derived ECs were characteristically produced by mesenchymal cells such as collagen type I and collagen type III genes. This suggests the possibility that angiogenic ECs in colorectal carcinomas are undergoing an epithelial-mesenchymal transition, which is observed when epithelial or ECs invade interstitial matrices (Nakajima et al., 1997; Boyer et al., 2000; Savagner, 2001). In contrast, Bell et al. (2001) showed that marked inductions of basement membrane matrix genes characteristic of EC differentiation were observed during EC morphogenesis in collagen matrices. These upregulated genes included collagen type IV αl chain, laminin γ1, laminin α4, heparan sulfate N-deacetylase N-sulfotransferase I (the rate-limiting step enzyme in heparan sulfate biosynthesis), lysyl oxidase (to cross-link EC synthesized ECM components), and the α2 and α1 integrin subunits (both of which can bind to collagen type IV and laminins) (Bell et al., 2001). In addition, another upregulated gene was the novel, capillary morphogenesis gene (CMG)-2, which contains a von Willebrand factor A domain with affinity for collagen type IV and laminin, a signal peptide, and a transmembrane segment, suggesting its possible expression on the EC cell surface (Bell et al., 2001). Preliminary studies have revealed a predominant localization of CMG-2 to the endoplasmic reticulum (ER) (Bell et al., 2001). If CMG-2 can be transported to the cell surface, its cell surface expression may be regulated by an ER membrane retention signal, which is a common way to regulate the expression of cell surface receptors (Bermak et al., 2001; Hardt and Bause, 2002).

The genes that were shown to be differentially expressed in our in vitro morphogenesis model were previously described by Bell et al. (2001). Markedly upregulated genes include jagged-1, stanniocalcin, angiopoietin-2, placental growth factor, sprouty, collagen type IV α1 chain, α2 integrin subunit, myosin IC, alpha-2 macroglobulin, egr-1, CD39, and HMG CoA reductase. Markedly downregulated genes include connective tissue growth factor, RGS-5, RGS-4, frizzled related protein-1, Id-1, Id-3, fibulin-3, β3 integrin subunit, syntenin, PAI-1, Cdc2, cyclin A1, cyclin B2, Cdc20, and Mcm2, thymidylate synthetase, and pentaxin. Many of the downregulated genes are involved in the inhibition of cell cycle progression that occurs during the tube formation process. Further support for this conclusion is that several negative regulators of cell cycle progression (i.e., p16/INK4a and Cdc14) increase in expression during the morphogenic process. The marked downregulation of regulator of G-protein signaling (RGS) genes, which are GAPs for G-proteins (DeVries et al., 2000), suggests that activation of G-protein coupled pathways may be occurring. An interesting finding regarding these RGS genes is that they are known to negatively regulate cell migratory events. This occurs because many of the receptors that mediate chemotactic events require G-proteins (Bowman et al., 1998). Thus, downregulation of RGS genes may result in an increased ability of ECs to sprout and migrate in order to form interconnecting networks of tubes. Other classes of genes identified during these screens include upregulation genes associated with the JAK-STAT pathway, anti-apoptotic signals, EC differentiation genes, cholesterol biosynthesis genes, and genes associated with EC quiescence (Bell et al., 2001) (Fig. 5). Upregulated genes that may function to induce EC quiescence include CD39 (an ecto ATP/ADPase) (Goepfert et al., 2000), CD26 (dipeptidyl peptidase, which inactivates biological active peptides) (Mentlein, 1999), A20 zinc-finger protein (which inhibits TNF-induced NF-κB-dependent gene transcription) (Lee et al., 2000; Klinkenberg et al., 2001), and melanoma-associated antigen (MG50), an ECM-like protein containing a peroxidase domain, multiple Ig repeats, and the precursor sequence of IL-1 receptor antagonist (Mitchell et al., 2000). The melanoma-associated antigen shows homology with the Drosophila basement membrane matrix protein, peroxidasin (Nelson et al., 1994). Other induced genes that may play a role in EC quiescence are alpha 2-macroglobulin and sprouty. Alpha 2-macroglobulin can sequester and inhibit the function of growth factors such as TGF-β, VEGF, and FGF-2 in the extracellular environment (Gonias et al., 2000). Its ability to bind growth factors is induced by conformational changes in α2-macroglobulin that occur through proteolytic cleavage of its bait region. Thus, its potential role in EC quiescence depends on proteinases synthesized and activated by ECs. We have shown that cleaved α2-macroglobulin is detectable during EC morphogenesis (Bell et al., 2001). Sprouty was originally shown by Hacohen et al. (1998) to regulate the development of the Drosophila tracheal system (i.e., networks of epithelial tubes). It was also shown to antagonize receptor tyrosine kinases (which are directly relevant to angiogenic signaling) by inducing turnover of internalized receptors (Reich et al., 1999; Wong et al., 2000). Increased expression of sprouty by viral gene transfer blocked angiogenic responses, further supporting these conclusions (Lee et al., 2001). Thus, all of the above genes have in common the potential ability to suppress EC activation through biologically active molecules such as ATP, peptides, hydrogen peroxide, IL-1, TNF, and angiogenic cytokines.

An intriguing possibility is that the basement membrane matrix itself (through proteins such as melanoma-associated antigen, laminins, and possibly by the presentation of proteinase inhibitors such as TIMP-3 and TFPI-2 via heparan sulfate proteoglycans (Liu et al., 1999; Yu et al., 2000; Herman et al., 2001)) also directly contributes to the development of EC quiescence. A recent study by Mettouchi et al. (2001) showed that α2β1-dependent interactions with laminin cause EC cell cycle arrest in G1. In contrast, they also showed that α5β1-dependent EC interactions with fibronectin lead to cell cycle progression and EC proliferation. In our system, fibronectin mRNA expression by ECs was downregulated during EC morphogenesis in collagen matrices (Bell et al., 2001). Also, downregulation of the β3 integrin subunit may represent another sign of EC quiescence. This protein has been described to be associated with angiogenic ECs and to either be expressed at very low levels, or not at all, in quiescent ECs (Brooks et al., 1994).

We have also utilized our microassay systems to isolate mRNA from different time points of EC morphogenesis to create cDNA libraries representative of these different stages (Figs. 1B and 6). Using these libraries, we isolated novel genes that were differentially expressed during EC morphogenesis and termed them capillary morphogenesis genes (CMGs) (Bell et al., 2001). To date, four full-length CMGs have been isolated, and we are in the process of characterizing their function. The sequences of CMG-1 and CMG-2 were recently reported by Bell et al. (2001), and the sequences of CMG-3 and CMG-4 will be reported in forthcoming publications. CMG-2 is a 45 kDa protein with a signal peptide, a transmembrane segment and an N-terminal region containing a von Willebrand factor A domain. These domains have been reported in other proteins to be ECM-binding domains (Colombatti and Bonaldo, 1991; Perkins et al., 1999). Furthermore, CMG-2 shows considerable sequence homology with a transmembrane protein recently identified in colonic tumor angiogenic ECs (TEM8) (St. Croix et al., 2000). Like CMG-2, TEM8 contains a von Willebrand factor domain, and this latter receptor was also recently found to be identical to the anthrax toxin receptor (Bradley et al., 2001). Bacterial expression of the CMG-2 von Willebrand factor A domain yielded a 20 kDa recombinant protein fragment with affinity for the basement membrane matrix proteins, collagen type IV, and laminin (Bell et al., 2001). Since CMG-2 is induced during EC morphogenesis alongside other basement membrane matrix proteins (see above), it is our hypothesis that CMG-2 participates in basement membrane matrix synthesis/assembly or EC–basement membrane matrix interactions during EC morphogenesis. It should be noted that blockade of basement membrane matrix assembly during angiogenesis or EC morphogenesis in vitro can markedly inhibit these events (Ingber and Folkman, 1988; Maragoudakis et al., 1988; Iruela-Arispe et al., 1991; Bonanno et al., 2000; Bell et al., 2001).

We also have considerable information concerning another novel gene, CMG-4, a 45 kDa protein, which is markedly downregulated during EC morphogenesis (Mavila et al., unpublished results) and is selectively expressed in ECs. This gene appears to regulate the EC cell cycle and shows affinity for the cdc2-cyclin B complex, which regulates the G2/M transition. Furthermore, the gene efficiently targets to the EC nucleus and is degraded by the proteasomal machinery, a characteristic of many cell cycle regulatory proteins (Koepp et al., 1999). Preliminary work with CMG-3, which is induced during EC morphogenesis, suggests that it associates with the EC cytoskeleton. CMG-3 contains a pleckstrin homology domain as well as a coiled-coil domain. A GFP-CMG-3 fusion protein targets to submembranous areas in ECs resembling lamellipodia (Mavila et al., unpublished results).

These data from our laboratory, as well as from others, reveal the central importance of the MIC signaling axis in EC tube formation, as well as new insights into how EC differentiation occurs during this process. This work also strongly supports the utility of in vitro models for investigating capillary morphogenesis where specific hypotheses can be formulated and then directly tested under defined conditions. These developing technologies allow for a rapid assessment of the contribution of particular genes to distinct stages of capillary morphogenesis in 3D ECM environments.

Capillary Tube Formation and Maintenance: A Balance Between Morphogenesis and Regression

The formation of blood vessels during development and following tissue injury responses is balanced by signals regulating both capillary morphogenesis and regression (Fig. 12). In normal wound-healing responses, the formation of vessels in granulation tissue is followed after several days by a capillary regression response (Clark, 1996). The final stages of this capillary regression response characteristically overlap with wound contraction (Clark, 1996). We recently described a model of capillary tube regression in vitro wherein these two processes occur concurrently (Davis et al., 2001). Thus, regression of these tube structures occurs alongside contraction of the collagen matrices in which they are suspended. This result describes an understudied ability of ECs to be highly contractile and capable of exerting strong mechanical forces on the ECM. As discussed above, this ability helps ECs locate each other in 3D space by creating “matrix guidance pathways” that represent physical channels or aligned ECM fibrils that facilitate directed cell migration (Vernon and Sage, 1995; Davis and Camarillo, 1995).

Figure 12. Capillary tube formation and maintenance in three dimensions is regulated by a balance of EC morphogenesis (M) and regression (R). The indicated factors regulate the described steps in a stimulatory (+) or inhibitory (–) manner.

The molecular mechanisms regulating capillary tube regression are for the most part unknown. Elucidation of this pathway is critical to our ability to induce regression of tumor-associated angiogenic blood vessels to inhibit tumor growth and progression (Folkman, 1997). A variety of studies have identified molecules that appear to induce tumor vessel regression in experimental animals, including ECM protein fragments, integrin targeting reagents, and fragments of proteins that regulate hemostasis (Brooks et al., 1994; O'Reilly et al., 1994, 1997; Sage, 1997; Browder et al., 2000; Maeshima et al., 2002).

A variety of findings have been made that appear to address how vascular regression may be regulated during normal physiologic events such as development, the female menstrual cycle, and wound-healing. Work by Benjamin and Keshet (1997) and Benjamin et al. (1999) has shown that withdrawal of VEGF induces marked capillary regression in in vivo models where tumor cells express VEGF under the control of a tetracycline-regulated promoter. Hyperoxia, which can cause physiological suppression of VEGF expression, is often a problem in the treatment of premature infants, and can cause severe visual impairment due to decreased VEGF-induced signaling in the retina (Hellstrom et al., 2001a). In contrast, hypoxia induces VEGF expression and other hypoxia-regulated genes, such as PAI-1, a plasminogen activator inhibitor (Pinsky et al., 1998). PAI-1 appears to play a role along with VEGF in promoting vascular stability (see below). Furthermore, findings have shown that angiopoietin-2 can participate in vascular regression by competing for the binding of angiopoietin-1 to the tie-2 receptor (Suri et al., 1996; Holash et al., 1999; Yancopoulos et al., 2000). Other studies have revealed elevations of angiopoietin-2 relative to angiopoietin-1 during physiological capillary tube regression in the ovary and adipose tissue (following leptin treatment) (Goede et al., 1998; Cohen et al., 2001).

The tendency of EC tubes to regress after formation appears to be affected by a number of factors. One such factor appears to be the association of specialized vascular smooth muscle cells (i.e., pericytes) with EC tubes (Allt and Lawrenson, 2001; Hellstrom et al., 2001b). Disruption of this interaction is observed in a variety of genetically altered mice (e.g., angiopoietin-1, edg-1, endoglin, and TGF-β knockout mice) (Dickson et al., 1995; Suri et al., 1996; Pepper, 1997; Lee et al., 1999; Li et al., 1999; Liu et al., 2000; Urness et al., 2000), as well as in human diabetic retinopathy (Sima et al., 1985; Ruggiero et al., 1997). This disruption causes marked instability of newly formed blood vessels or preexisting vessels. In the models described above using tumors expressing VEGF in a regulated fashion, the susceptibility of those vessels to VEGF withdrawal correlated with the presence of pericytes (Benjamin et al., 1999). Those vessels with pericytes were much more resistant to VEGF withdrawal than those without. However, it is not known how pericytes contribute to this stability. Histologic analysis of angiopoietin-1 knockout mice revealed abnormally dilated and tortuous vessels, which led to embryonic lethality (Suri et al., 1996). Some ECs in these vessels did not appear to be adhering well to their underlying ECM. In support of such conclusions is a recent report describing the ability of angiopoietin-1 to directly promote EC cell attachment through integrins (Carlson et al., 2001). Another possibility is that the pericytes are contributing factors in the specialization of the ECM environment underlying ECs. One important example of basement membrane matrix specialization is the presence of unique laminin isoforms in EC basement membranes. Our recent data suggest the presence of at least laminin-8 and laminin-10 isoforms based on the differential regulation of laminin subunit mRNAs during EC morphogenesis (Bell et al., 2001). This conclusion is also supported by previous results (Miner et al., 1998; Lefebvre et al., 1999). These mRNAs were upregulated during later stages of morphogenesis, suggesting that these laminins may play an important role in the maturation and stabilization of developing tubes. There are many examples of epithelial–mesenchymal interactions which are necessary for basement membrane matrix formation and specialization (Kadoya et al., 1997; Smola et al., 1998; Hedin et al., 1999; Erickson and Couchman, 2000). These interactions contribute different combinations of basement membrane components (laminin and collagen type IV isoforms, etc.) which are critical to basement membrane assembly. The extent to which pericytes contribute to EC basement membrane matrix formation and maintenance is unclear.

Since basement membrane matrix contributes to the formation, stability, and maintenance of blood vessels, its proper formation may underlie many of the described EC–pericyte interaction defects reported in the various genetic knockout models. Such defects include failure to properly recruit pericytes to the EC tubes, and defects in the EC–pericyte interactions necessary to stabilize the tube. One likely participant in this event is TGF-β, including its receptors and the signaling pathway molecules downstream of these receptors (Dickson et al., 1995; Pepper, 1997; Li et al., 1999; Urness et al., 2000). It has been well described that ECs or pericytes alone are unable to properly activate secreted TGF-β (Sato et al., 1990; Munger et al., 1997; Hirschi et al., 1998). In contrast, when both cells are combined, latent TGF-β can be activated by a mechanism that is typically mediated through plasmin-mediated proteolysis. TGF-β is well known to markedly stimulate ECM synthesis and the production of factors such as PAI-1, which can facilitate ECM stability by inhibiting matrix proteolysis (Pepper, 1997; Denton and Abraham, 2001). Another very interesting finding in this regard is that the EC-derived cytokine, connective tissue growth factor (CTGF), markedly enhances the ability of TGF-β to stimulate matrix production and deposition (Grotendorst, 1997; Denton and Abraham, 2001). In our DNA microarray work, the mRNAs of CTGF and a related cytokine (Cyr61) were markedly downregulated early during morphogenesis but were then returned to near-baseline levels as EC morphogenesis progressed (Bell et al., 2001). Other regulated cytokines, such as PDGF-B, which was induced during EC morphogenesis, may also play a role in this process by regulating events such as pericyte recruitment to EC tubes by stimulating migration and proliferation. CTGF can similarly induce smooth muscle cell proliferation and migration (Fan et al., 2000). Thus, EC–pericyte interactions by means of TGF-β, CTGF, and PDGF may be responsible for ensuring the formation, stability, and maintenance of EC-lined capillary tubes through effects on a number of events, such as EC basement membrane matrix assembly and EC quiescence.

Recent efforts to examine gene expression changes during EC morphogenesis offered some insights into how a balance between EC morphogenesis vs. regression may be regulated (Fig. 12). It remains to be determined whether EC tube regression genes are differentially regulated during EC morphogenesis, and whether genes exist whose primary purpose is to regulate the process of EC tube regression rather than EC tube formation. At the moment it is not clear that such genes exist, but often in studies of biological systems, molecules that regulate positive events are identified first, followed by molecules that regulate negative events. An excellent example of this point are studies of nervous system development wherein factors that stimulate neurite outgrowth were discovered first, while factors that interfere with this process (e.g., collapsins and semaphorins) were discovered later (Yu and Bargmann, 2001). It has recently become clear that there is a striking overlap between factors that regulate neural growth and vascular growth. For example, the recent discovery by Soker et al. (1998) and Miao and Klagsbrun (2000) that the semaphorin III receptor, neuropilin, is a VEGF receptor indicates that there are similarities between these systems. These investigators have also shown that semaphorin III blocks both capillary tube sprouting and neurite outgrowth (Miao et al., 1999). In addition, we have detected semaphorin III mRNA in ECs and shown that its expression is markedly downregulated during the period of sprouting and branching that occurs during EC morphogenesis (unpublished observations). Sprouting EC processes often contain “growth cone”-like projections which regulate the directional navigation of ECs in three dimensions (Speidel, 1933; Davis et al., 2000b). The extent to which other regulators of neural growth cone guidance may play a role in EC morphogenesis or regression is not yet known (Yu and Bargmann, 2001).

Differentially regulated genes that could play a direct or indirect role in regulating the process of EC tube regression are MMP-1, PAI-1, tissue factor pathway inhibitor (TFPI)-2, gelsolin, angiopoietin-2, thrombospondin (TSP)-1, and TSP-2. The latter two ECM proteins have been shown to negatively regulate blood vessel formation, with TSP-2 appearing to have the most potent activity (Kyriakides et al., 1998; Streit et al., 1999; Adams, 2001). In a study by Bell et al. (2001), TSP-1 mRNA was downregulated during EC morphogenesis while TSP-2 was upregulated. In another work (Davis et al., 2001), PAI-1 (which negatively regulates plasmin- and MMP-dependent capillary tube regression) was markedly downregulated as well. In recent studies (Rao et al., 1996; Herman et al., 2001) using adenoviral gene transfer, upregulated expression of the serine and MMP proteinase inhibitor, TFPI-2, in ECs blocks either plasminogen/plasmin or plasma kallikrein-induced capillary tube regression (Davis et al., unpublished results). Our model of regression involves not only ECM degradation but also EC tube collapse and apoptosis (Davis et al., 2001) (Davis et al., unpublished results). We previously reported that gelsolin is markedly upregulated during EC morphogenesis (Salazar et al., 1999). Gelsolin has been identified as a major caspase-3 target and an important regulator of apoptosis (Kothakota et al., 1997; Kamada et al., 1998). In our model of EC capillary tube regression, caspase-dependent cleavage of gelsolin was detected (Davis et al., 2001). Thus, gelsolin may be induced in part to regulate the process of EC tube regression if this process is initiated. The most induced gene (of more than 7,000 genes profiled) detected at 48 hr of culture during EC morphogenesis was angiopoietin-2. Angiopoietin-2 has been reported to antagonize the function of angiopoietin-1 by competing with its binding to the Tie-2 receptor (Suri et al., 1996; Yancopoulos et al., 2000) and it also has been reported by other groups to be produced by ECs (Mandriota and Pepper, 1998). The important question here is, why do ECs make angiopoietin-2? The induction of angiopoietin-2 by ECs during morphogenesis could be envisioned as an attempt by ECs to regress unless some signal is provided to overcome this inhibitory signal.

An analogy that occurs to us in this regard refers again to relationships between neural and vascular systems. During peripheral nervous system development, neurons extend axonal processes to targets such as skeletal muscle to form synaptic contacts. These neuron–target interactions provide a trophic role to maintain survival of the neurons and to stabilize both cell types (Ernfors, 2001; Sanes and Lichtman, 2001). Do developing vascular networks need an analogous target to provide a similar type of signal to facilitate EC tube survival and stabilization? We hypothesize that pericytes/vascular smooth muscle cells (VSMCs) represent such a target in most tissue environments (Fig. 13). The EC–pericyte interaction would serve to stabilize and support the survival of the developing tube, and through these effects would facilitate the development of EC specializations. Such specializations may include properties characteristic of arterial, capillary, or venous ECs. The ephrins, which regulate neural development, have been reported to regulate arterial vs. venous identity for ECs in these locations (Wang et al., 1998; Wilkinson, 2001). In addition, EC–target interactions may regulate the development of unique EC properties characteristic of different tissue locations. Also, in other specialized tissues such as the brain, kidney, and lung, ECs are known to closely interact with astrocytes, podocytes, and alveolar eptithelial cells, respectively. These other cell types may perform functions similar to those of pericytes to regulate EC stability and create further EC specializations. This EC–target hypothesis may explain many of the observations regarding the instability of newly formed tubes in the absence of pericytes or other supporting cells. Furthermore, it may explain data explaining why factors such as angiopoietin-2, CTGF, PDGF, and TGF-β are produced by ECs during the morphogenic process. These factors are probably synthesized by ECs to facilitate the development of this EC–target interaction. If the EC–target interaction does not occur, the EC tubes will regress, which is analogous to what happens when neurons fail to reach an appropriate target and undergo apoptosis (Ernfors, 2001). With respect to the EC–pericyte/VSMC interaction, it is notable that ECs can synthesize neurotransmitters such as acetylcholine and nitric oxide to act on VSMC targets, similar to what is observed in neuron–target interactions.

Figure 13. The EC–target hypothesis. An analogy between neuron–target interactions and EC–target interactions is shown. This hypothesis suggests that EC networks interact with target cells to stabilize and develop specialized characteristics. These interactions also prevent capillary tube regression events. Potential EC target cells include pericytes/VSMCs, astrocytes, podocytes, and alveolar epithelial cells.

A major new insight into the molecular regulation of capillary tube regression stems from recent studies showing that ECM proteolysis, which disrupts the matrix scaffold regulating EC morphogenesis, induces capillary tube regression and EC apoptosis (Zhu et al., 2000; Davis et al., 2001). This relatively simple model of capillary tube regression has considerable relevance for the development of strategies and reagents aimed at inducing vascular regression, and is particularly relevant in the context of cancer. This work is highly related to the pioneering work of Werb and colleagues, who studied the molecular control of mammary gland involution following the cessation of lactation (Talhouk et al., 1992; Sympson et al., 1994; Werb et al., 1996). In this model, it is clear that ECM proteolysis plays a major role in epithelial tube regression as well as the development of adipose tissue to replace this regressing tissue (Selvarajan et al., 2001). Other studies (Marbaix et al., 1996; Kokorine et al., 1996) have revealed an important role of MMP-1 in, for example, the menstrual cycle and endometrial shedding, where it was shown to be hormonally regulated and to be involved in this cyclical tissue regression event.

Several laboratories, including ours, have found evidence that vascular regression is controlled by ECM proteolysis. In one set of studies, it was observed that PAI-1 knockout mice demonstrated poor angiogenic responses, and that, consequently, tumors grew poorly in these mice (Bajou et al., 1998, 2001). In the absence of PAI-1, an increase in matrix dissolution occurs via plasmin-dependent degradation of ECM (see below) which destabilizes ECM environments supporting EC morphogenesis. In addition, PAI-1 has been observed to be a hypoxia-induced gene (Pinsky et al., 1998). Thus, hypoxia induces ECM stability to promote angiogenic responses. In our model of capillary tube regression in vitro, PAI-1 was shown to negatively regulate plasmin and MMP-dependent regression events (Davis et al., 2001). Blocking antibodies to PAI-1 markedly accelerates the regression response, and the addition of PAI-1 to this model blocks capillary tube regression. Another study (Bacharach et al., 1998) showed that PAI-1 expression was increased by EC–fibroblast interactions to inhibit proteolysis and stabilize the ECM environment during angiogenesis. Thus, in some cases, factors that promote ECM stabilization may be pro-angiogenic while factors that promote ECM dissociation/proteolysis may be anti-angiogenic.

Key regulators of this proteinase-dependent capillary tube regression response are plasmin and MMPs (such as MMP-1) (Fig. 3). Plasmin is capable of degrading fibrin, a major ECM scaffold for EC morphogenesis, and of activating MMPs such as MMP-1, MMP-3, and MMP-9, which can degrade other critical ECM scaffolds, such as collagen types I and IV, laminin, and fibronectin (He et al., 1989; Jeffrey, 1998; Lund et al., 1999; Ramos-DeSimone et al., 1999). In our serum-free system, the addition of plasminogen leads to the generation of plasmin through EC-derived plasminogen activators. This in turn activates MMP-1 to initiate collagen type I proteolysis and denaturation, which leads to capillary tube regression (Davis et al., 2001). Without plasminogen addition, ECs fail to activate MMP-1 (based on assays of conditioned media) and do not regress. Activated MMP-1, which degrades native collagen, acts in concert with activated MMP-9 (activated by plasmin) and MMP-2 (activated by MT-MMPs), which degrade denatured collagen, to cause capillary tube regression and collagen gel contraction. The addition of proteinase inhibitors such as TIMP-1, PAI-1, and aprotinin, which block MMPs, plasminogen activators, and plasmin, respectively, block the regression response. We have also recently found that the addition of activated plasma kallikrein can independently activate MMP-1 to initiate the capillary tube regression response (Davis et al., unpublished results). This experiment was performed in response to similar observations recently reported by Selvarajan et al. (2001) that plasma kallikrein can act in concert with plasmin to facilitate mammary gland regression and adipose tissue development. In our experiments, plasma kallikrein and plasmin synergistically acted to induce the capillary tube regression response by initiating the activation of MMP-1 (Davis et al., unpublished results).

To address the hypothesis that VSMCs may serve as EC target cells that facilitate the stabilization of capillary tubes, we performed coculture experiments with ECs and VSMCs. We sought to determine whether plasminogen/plasmin-induced capillary tube regression would occur in the presence of VSMCs. EC-dermal fibroblast cocultures were used as a control. As shown in Figure 14, the addition of VSMCs to EC cultures completely blocked the capillary tube regression response, while dermal fibroblasts did not. Capillary tube regression is accompanied by collagen gel contraction, which indicates that regression has occurred (Fig. 14A). Conditioned medium from VSMCs suspended in collagen gels also blocked capillary tube regression and collagen gel contraction, suggesting that a soluble factor was involved (Fig. 14C). Examination of MMP-1 and MMP-9 activation revealed that the presence of VSMCs markedly blocks MMP activation, which explains the inhibitory effect (Fig. 14B). The nature of the VSMC-derived capillary regression inhibitor is not known at present. These data show that VSMCs can stabilize developing tubes that would otherwise regress, suggesting that inhibition of ECM proteolysis may represent one mechanism by which pericytes/VSMCs stabilize EC tubes.

Figure 14. VSMCs prevent plasminogen-induced capillary tube regression in EC-VSMC cocultures in 3D collagen matrices. ECs were cultured in the presence or absence of either human coronary artery smooth muscle cells (CASMCs) or dermal fibroblasts (HDFs). The CASMCs and HDFs were added at 5 × 10⁵ cells/ml of gel, and the ECs were added at 10⁶ cells/ml of gel. Culture media contained plasminogen (Plg) at 1 μg/ml, or contained none, as previously described (Davis et al., 2001). A: After 72 hr, cultures were fixed, stained, and photographed. The circular dark structures represent contracted collagen gels. Arrowheads indicate CASMCs. B: Conditioned media samples collected from the indicated cultures were run on SDS-PAGE for immunoblotting (left panel) with anti-MMP-1 antibodies, or for gelatin zymogram analysis (right panel). Left panel, arrowhead: pro-MMP-1; arrows: activated MMP-1 forms. Right panel, arrowheads: pro-MMP-9 and pro-MMP-2; arrows: activated forms of these enzymes which directly underlie the proenzyme bands. C: Indicated conditioned medium (CM) samples were added at a 1:4 dilution to EC-only cultures to determine whether CASMC CM could block the capillary tube regression and collagen gel contraction response. Cultures were fixed after 48 hr and photographed. Bar = 25 μm.

Other experiments support the general conclusion that ECM proteolysis is a key regulator of capillary tube regression responses. In a recent study, transgenic animals overexpressing TIMP-1 showed increased vascularity (Yamada et al., 2001), which could be indicative of decreased vascular regression (Fig. 12). TIMP-1 and PAI-1 may be particularly important capillary regression inhibitors because of their ability to block EC tube regression (Davis et al., 2001) and inability to block EC morphogenesis (Davis et al., unpublished observations). Marked overexpression of TIMP-1 or PAI-1 using recombinant adenoviral vectors in ECs does not block morphogenesis, but completely inhibits plasminogen/plasmin or plasma kallikrein-induced and MMP-dependent capillary tube regression. Furthermore, recent studies have indicated surprisingly high levels of TIMP-1 in the plasma of patients with a variety of cancers, including colon, gastric, breast, and lung cancer (Pellegrini et al., 2000; Yoshikawa et al., 2001; Yukawa et al., 2001). These data indicate that TIMP-1 and PAI-1, and perhaps other proteinase inhibitors synthesized by tumors, may block the vascular regression response to facilitate the maintenance of the tumor angiogenic vasculature. Thus, proteinase inhibitors may be critical positive regulators of tumor angiogenesis by blocking vascular regression. This new information suggests that ECM-degrading proteinases, which clearly play a role in EC morphogenesis, are also critical regulators of capillary tube regression in 3D ECM environments.

CONCLUSIONS

In this review, we have addressed many issues regarding how EC morphogenesis and regression are regulated at a molecular level in 3D extracellular matrices. It is apparent that this field is still developing, and that considerable information about these processes remains to be discovered. The molecular tools and cell biological approaches necessary to answer the basic question of how ECs assemble or disassemble in three dimensions are currently available. It is our hope that this review will provide some new insights into how the study of basic mechanisms regulating EC morphogenesis or regression will rapidly lead to the identification of novel molecular targets to modulate these events in the context of human disease.

Acknowledgements

The authors thank Dr. Scott Bell and Dr. René Salazar, who made many contributions to our current understanding of EC morphogenesis in 3D ECM environments. This work was supported in part by grants from the NIH (NHLBI; HL59373 to G.E.D., and F32 HL69603 to K.J.B.) and the Texas Higher Education Coordinating Board (89-57-2001 to G.E.D.).