Extracellular regulation of VEGF: isoforms, proteolysis, and vascular patterning

2.1 Alternate splicing results in multiple VEGF isoforms that bind differently to receptors, co-receptors, and HSPGs

The human VEGF gene (located on chromosome 6) consists of several exons that can be alternatively spliced (30, 31) to encode several protein isoforms, including: VEGF₁₂₁, VEGF₁₄₅, VEGF₁₆₂, VEGF₁₆₅, VEGF₁₈₃, VEGF₁₈₉, and VEGF₂₀₆. These isoforms are denoted by their length (number of amino acids) and differ in incorporation of exons 6 and 7, which convey HSPG- and neuropilin-1 (NRP1)-binding motifs (Fig. 1B). Compared to human VEGF isoforms, murine isoforms have the same functional components and biological function but are one amino acid shorter, e.g. VEGF₁₂₀, VEGF₁₆₄, VEGF₁₈₈. With ~90% sequence identity between human and murine forms (Fig. 1A), slight structural variations seem to exist in terms of proteolytic processing and antibody recognition (33). Human VEGF isoforms have weakly-angiogenic or anti-angiogenic counterparts, VEGF_xxxb, in which exon 8b is substituted for exon 8a (30, 34). There is some debate over whether the antiangiogenic VEGF_xxxb isoforms are expressed in the mouse (35, 36).

The primary receptor tyrosine kinases (RTKs) for VEGF isoforms are VEGFR1 (Flt-1), VEGFR2 (Flk-1 or KDR), and VEGFR3 (Flt-4). Receptor binding of VEGF depends on the presence of isoform-selective co-receptors such as NRP1, NRP2, and cell-surface glycosaminoglycans (GAGs). VEGFR2 is the primary receptor responsible for endothelial cell mitogenic and migratory responses to VEGF (14), while VEGFR1 is thought to function primarily as a decoy receptor during development, and to modulate signaling of VEGFR2 in the adult (37, 38). Although it is considered a weaker kinase, VEGFR1 does have direct signaling functions (37, 38). NRP1, in contrast, is not thought to directly transduce VEGF signals but selectively enhances VEGF₁₆₅ binding to VEGFR2 (by presenting NRP1-bound VEGF to bind VEGFR2) (39, 40) and alters VEGFR2 intracellular trafficking (41). NRP1-enhanced VEGF signaling has been shown to be important for p38/MAPK activation, and thus is central to vessel branching (27, 42).

VEGF isoforms are typically secreted as cysteine-linked antiparallel dimers (Fig. 1D). Binding to VEGFR1 and VEGFR2 monomers at the cell surface is mediated by the sequence encoded by exons 3 and 4 respectively, epitopes located at either end of the bivalent dimeric ligand (Fig. 1C–D). Binding to HSPG is mediated in a sequence- and charge- specific manner by the basic amino acids from exons 6a and 7 (“heparin-binding domain”), while NRP1 binding is mediated by exon 7 regions. Exon 8a imparts important structural stability and function to exon 7 through disulfide bonding (Fig. 1C–D). Exon 6 has a poorly understood role in VEGF structure, and is discussed further in section 2.3, below.

2.2 Differential cellular effects of VEGF₁₂₁ and VEGF₁₆₅ are mediated by binding to NRP1 and HSPGs

While both VEGF₁₂₁ and VEGF₁₆₅ are capable of signaling through VEGFR2, VEGF₁₆₅ binding to (and activation of) VEGFR2 is potentiated by cell-surface NRP1 (40, 43, 44), cell-surface HSPGs such as glypican-1 (45, 46), and exogenous GAGs such as heparin (47, 48). Unlike VEGFR2, VEGFR1 shows intrinsically greater affinity towards VEGF₁₆₅ than VEGF₁₂₁ especially in the absence of heparin (47); however, the presence of NRP1 may block the VEGF₁₆₅-VEGFR1 interaction, leading VEGFR1 to be a preferential receptor for VEGF₁₂₁ (49, 50).

NRP1 is a co-receptor for several VEGF isoforms, notably VEGF₁₆₅ (40). It facilitates formation of a stable ternary NRP1-VEGF₁₆₅-VEGFR2 complex, increasing the overall avidity of the isoform for VEGFR2 (27, 39). VEGF₁₂₁ may also bind NRP1, despite the absence of the canonical exon 7, via exon 8 sequences (51, 52). However, unlike for VEGF₁₆₅, NRP1 cannot bridge VEGF₁₂₁ and VEGFR2 (44, 52). GAGs similarly potentiate heparin-binding isoforms such as VEGF₁₆₅. Interestingly, heparin and solubilized glypican-1 can stabilize VEGF/VEGFR2 complexes even after pre-treatment with exogenous heparinase (45, 47, 53). NRP2, another member of the neuropilin family, also serves as a co-receptor for VEGFR2, but with slightly different isoform specificity (54, 55).

VEGF_165b (an anti-angiogenic isoform mentioned above) lacks HS and NRP1 binding (30, 43, 44, 56). Unlike VEGF₁₂₁, however, it assumes a reduced signaling state once bound to VEGFR2 that is resistant to rescue by exogenous VEGF₁₆₅ (34, 43, 44). This difference is thought to arise due to the lack of the stabilizing influence of exon 8a (Fig. 1B), rather than the presence of exon 8b itself (43, 56).

2.3 Exon 6a may function as an intrinsic VEGF inhibitor and its cleavage activates VEGF

Exon 6a-containing isoforms such as VEGF₁₄₅ and VEGF₁₈₉ continue to be poorly understood. Exon 6a itself is directly inhibitory for VEGF activity (57). Exon 6a is heavily dominated by basic amino acids, encodes a nuclear localization sequence (58), and interferes with VEGFR2 binding (59). Full-length VEGF₁₈₉ and VEGF₁₄₅ are weaker mitogens and chemotactic agents than even VEGF₁₂₁ (44, 60, 61). In contrast, VEGFR1 activity is unaffected by the presence of exon 6, being similar to that of VEGF₁₆₅ (60, 62), and this allows exon 6-containing isoforms to induce vascular permeability (62, 63) or neutrophil cell migration (64).

Binding of these exon 6a isoforms to VEGFR2 seems to require proteolytic processing by urokinase plasminogen activator (uPA) or plasmin (60, 62), or binding to ECM/heparin (17, 61). For example, uncleaved VEGF₁₈₉ applied directly to bovine adrenal cortex-derived ECs showed no activity (60). However, cleavage by uPA, which only partially removes exon 6a and all of exon 7, activates VEGF₁₈₉ to the level of VEGF₁₄₅, while cleavage by plasmin, which entirely removes exon 6a and 7, activates VEGF₁₈₉ to the level of VEGF₁₂₁ (60) (Fig. 1E). VEGF₁₈₉ activity may depend on cell type as it seems to have similar activity as VEGF₁₆₅ towards HUVECs (65).

VEGF₁₄₅ has low affinity for NRP1 (44, 54) and does not bridge VEGFR2 and NRP1 (44). VEGF₁₈₉ has been shown to have a 10-fold greater NRP1 affinity than does VEGF₁₆₅ (66, 67); however, given that VEGF₁₈₉ is a weak mitogen like VEGF₁₄₅ (60), it is likely unable to form or support a functional ternary complex with NRP1 and VEGFR2.

2.4 Proteolytic processing results in additional VEGF isoforms

Proteases can have several distinct effects on VEGF, including cleavage, activation, liberation from extracellular stores, and degradation. Each of these has consequences for VEGF activity. The most-studied VEGF proteases are plasmin (47, 60, 62, 68) and the matrix metalloproteinases (MMPs), including MMPs 1, 3, 9, 7, 12, 16, and 19 (13, 69). Others include uPA (60), elastase (70, 71), and tissue kallikrein (72, 73).

VEGF₁₆₅ cleavage by plasmin significantly reduces its overall bioactivity (over 100-fold increase in EC₅₀ (47)) and removes NRP1- and heparin-binding abilities (47, 52). Plasmin-, MMP-, and tissue kallikrein-mediated cleavage of VEGF occurs in exon 5 (Fig. 1C) (13, 47, 56, 72) leading to an N-terminal active fragment of VEGF that has an electrophoretic mobility of 13 kDa (via plasmin) or 16 kDa (via MMP) (13, 70). In contrast, uPA-mediated digestion, as discussed above, likely occurs in the C-terminal portion of exon 6 (60), allowing the main cleavage product to preserve some of its ECM-binding character (60, 61). Elastase cleavage occurs at both N- and C- termini (71). Since VEGF is homodimeric, proteolysis proceeds through a heterodimer intermediate with only one chain cleaved (47, 74). While cleaving fragments on the N- and C- termini, these proteases do not seem to digest the core sequences mediating receptor binding (exons 3, 4) (47), with the exception of elastase which abrogates VEGFR2 binding (71). The effective cleavage rate of VEGF₁₆₅ by plasmin is k_cat/K_m = 328 M⁻¹s⁻¹ with K_m > 1 mM, at 25°C (74), which is similar to that observed by MMP3 (13), but is lower than typical ECM proteolysis reactions, e.g. fibrin/plasmin (k_cat/K_m = 8,100 M⁻¹s⁻¹ at 37°C), or Type 1 collagen/MMP2 (k_cat/K_m = 5,300 M⁻¹s⁻¹ at 25°C) (74). A significant portion (40–80%) of VEGF has been shown to have been proteolyzed in pathological systems (13, 69, 75).

While murine VEGF isoforms can be cleaved by MMPs, this may not be true for human VEGF isoforms (76, 77). While Lee et al. have shown that murine VEGF can be cleaved by MMPs (13), the available evidence that human VEGF is similarly cleaved is a 16 kDa VEGF band in human ovarian cancer ascites fluid which may represent a glycosylation variant of plasmin-cleaved VEGF (13). On the other hand, numerous studies have shown the absence of direct cleavage of human VEGF by MMPs (70, 71, 76–78). This may be due to the fact that murine and human VEGF are highly dissimilar at the proteolytic cleavage sites (amino acids 110–114) (Fig. 1A), and VEGF-cleaving proteases exhibit a high degree of substrate specificity; for example, replacing the plasmin cleavage site in murine VEGF, R¹⁰⁹-T¹¹⁰, with K¹⁰⁹-P¹¹⁰ makes murine VEGF not only plasmin resistant, but also MMP resistant at the MMP cleavage site, amino acids 113–114 (13). It should be noted that MMP-cleaved human VEGF should numerically correspond to VEGF₁₁₄, not VEGF₁₁₃, due to the extra amino acid present in exon 2 not found in murine VEGF. To our knowledge, this has not been demonstrated. A VEGF₁₁₄ isoform proposed by Mintz et al. assumes a 27 instead of 26 amino acid signal sequence (79), and thus should correspond to human VEGF₁₁₅, which lacks the entirety of exon 8.

3.1 ECM sequestration of VEGF affects diffusion and modulates VEGF bioactivity

VEGF binds to the ECM near cells secreting VEGF, as has been shown in retina, brain, various tumors, and other tissues (14, 16, 17). VEGF sequestration is not restricted to matrix binding sites; VEGF binding to cell surface VEGF receptors also reversibly immobilizes VEGF, and before secretion or after internalization may be sequestered intracellularly (81, 82). Sequestration in the ECM and on cell surfaces seems to restrict VEGF gradients (15, 16) and guide endothelial cell migration (14, 28). Sequestered VEGF seems to function haptotactically as migrating ECs extend filopodia to explore sites of deposited VEGF, even towards that found on other cells (14, 28, 69, 83, 84). Matrix-bound VEGF is capable of binding and activating VEGF receptors, serving as a biochemical as well as a mechanical signal; matrix-bound VEGF induces different signaling patterns than soluble VEGF (28, 85). Sequestration by the matrix limits VEGF diffusion and decreases clearance from the tissue, which may facilitate autocrine signaling (64). Sequestration also allows the storage of VEGF that can be released to locally amplify a pro-angiogenic signal (86), and may confer resistance to certain types of proteolytic processing; binding to GAGs or fibronectin may protect against plasmin and MMP9, respectively (13, 87).

Among the many matrix binding sites for VEGF, GAGs are thought to predominate based on heparin elution (17, 62, 68); GAGs can be ECM-associated, e.g. perlecan, or membrane-associated, e.g. syndecan. All known VEGF-A isoforms bind heparin/HS except VEGF₁₂₁, VEGF_xxxb, and the proteolytic isoforms produced after processing by MMPs, plasmin, or elastase (13, 43, 47, 71) (Fig. 1E). The exon 6a-containing isoforms, including VEGF₁₄₅, show greater affinity to the HSPGs than does VEGF₁₆₅; this binding may be partially independent of GAGs (44, 61, 68). Other matrix binding sites include collagen (28), fibrin (88), and fibronectin (28, 84, 89), while Type IV collagen and vitronectin are incapable of binding VEGF (89). VEGF may also indirectly bind to ECM through another ECM-binding soluble mediator, such as sVEGFR1 (21, 90) or ADAMTS1 (91). VEGF cell surface receptors also sequester VEGF until VEGF dissociates or is internalized. This receptor binding induces signaling in endothelial and other cells, but also alters the local gradients of VEGF in the interstitial space (92).

3.2 Soluble carriers of VEGF play a role in vascular quiescence and gradient formation

Despite active secretion of VEGF, some tissues including the cornea (12) and certain precancerous lesions (76, 78), are held in an anti-angiogenic or angiostatic state due to the simultaneous secretion of soluble inhibitors of VEGF, which in binding VEGF prevent VEGF-induced dimerization and activation of VEGFR2 (91). Parallel to ECM sequestration, soluble inhibitors create a diffusible reservoir of VEGF, accessible through unbinding or proteases (76). For example, corneal infection by the herpes simplex virus 1 leads to corneal stromal keratitis through the infiltration of new blood vessels, but requires either overcoming the endogenous sVEGFR1 barrier by upregulation of VEGF, mediated by the virus, or by MMP-mediated degradation of sVEGFR1, mediated by the virus-induced acute neutrophilic response (12, 93). Other soluble VEGF inhibitors include alpha-2-macroglobulin (94) and thrombospondin (TSP) domain-containing molecules, which preferentially bind VEGF₁₆₅ (91): connective tissue growth factor (CTGF) (70, 76) contains one such domain, while thrombospondin-1 (TSP1) (95) and ADAMTS1 (91) each contain three. TSP1 can block VEGF-HSPG association at the cell surface (95) as well as interfere with VEGFR2 signaling (96). Finally, soluble proteins can inhibit VEGF binding to receptors more indirectly, such as platelet factor 4 (PF4) (97). Just as there are soluble VEGF inhibitors, other carriers (soluble molecules that bind VEGF in solution) can permit VEGF to retain its activity. Soluble activators of VEGF include heparin and proteolytically-released heparan sulfates (e.g. syndecan), which can enhance VEGFR2 activation by co-activation of integrins (98), and fibronectin, which upon release by platelets binds VEGF and can crosslink α5β1 and VEGFR2, enhancing endothelial activity (87).

Unlike the stationary ECM, soluble VEGF carriers can mediate the transport and clearance of VEGF. Heparin can release and disperse FGF2 and FGF10, enabling activation of cells at a distance and altering gradients (99, 100). Similarly, VEGF release by proteases and heparanases produces diffusible VEGF that can escape subsequent rebinding to the matrix and is resistant to degradation (13, 101). Carriers may also enhance the spatial complexity of the local VEGF gradients (21, 102). sVEGFR1 secreted by a blood vessel is thought to shape the VEGF concentration field near the vessel, guiding nascent vascular sprouts to migrate perpendicularly from the parent vessel (20).

3.3 Distinct mechanisms of VEGF release promote pro- or anti- angiogenic effects

Among the many roles for proteolytic enzymes in angiogenesis (103), they are important modulators of extracellular VEGF, which is important in numerous pathologies including pancreatic islet carcinogenesis, breast cancer, and oxygen-induced retinopathy (Table 1, Table S1). Proteolytic release of VEGF from the matrix is typically viewed as being pro-angiogenic (76, 86), however in other situations, protease action on VEGF may inhibit angiogenesis (13, 104). Similarly, protease inhibitors have been shown to both decrease angiogenic sprouting (86) and increase sprouting (13). This dichotomy seems to be explained by distinct mechanisms of action that proteases have on VEGF (Fig. 3).

Several extracellular and membrane-bound proteases can directly cleave VEGF. Plasmin and MMP3 cleave only the C-terminal domain (13, 47), while elastase and plasmin are associated with complete degradation involving cleavage at both N-terminal and C-terminal sites (71, 104). In non-healing wounds, interstitial fluid is very proteolytic, resulting in rapid degradation of VEGF, preventing the recruitment of a vascular supply (104, 105). In contrast, cleavage of the C-terminus facilitates VEGF release and increases the level of soluble VEGF; the overall balance still seems to be anti-angiogenic as the released VEGF molecule is less bioactive (13, 73). Breast cancer xenografts engineered to express a C-terminal truncated isoform, VEGF₁₁₃, demonstrate decreased tumor growth rates, lower vascular density, and dilated tortuous vessels compared to VEGF₁₆₄ expressing tumors (13). Despite this, VEGF release by cleavage may be pro-angiogenic in the short term if the release occurs rapidly enough (106), but it is not certain that the pro-angiogenic effects can be sustained. In oxygen-induced retinopathy (OIR), proteolytic VEGF release seems to produce similar irregular endothelial proliferation as in the VEGF₁₁₃-only tumors, apparently mediated by macrophage-secreted MMP12, however the angiogenesis is chaotic and hyperproliferative (69).

Enzymes can also degrade soluble VEGF inhibitors (and presumably soluble activators), altering VEGF transport and/or activity. Cleavage of VEGF inhibitors has been directly associated with increased angiogenesis, and in vivo, seems to be primarily mediated by MMP7. For example, in a Capan-1 xenograft model of pancreatic adenocarcinoma, stromal fibroblasts secrete VEGF inhibited with CTGF; and cancer cells can secrete MMP7 or other MMPs to cleave CTGF (76) (Fig. 3A). Other VEGF carriers can be cleaved: TSP-1 by plasmin and elastase (107); CTGF by plasmin, elastase, MMPs 1, 3, 7, 13, and ADAM28 (70, 108); and sVEGFR1 by MMP7 or weakly by MMP2 and MMP9 (78, 93).

Enzymes can also release VEGF from sequestration by cleaving the ECM, i.e. the GAGs, HSPG core proteins, and other associated ECM molecules, without processing VEGF. This released VEGF may remain complexed with an ECM fragment (98, 99). In vivo studies demonstrating this release mechanism are limited, but it is associated with increased angiogenic potential (109). The primary molecular mediators are heparanase, which cleaves the heparan sulfate chains and possibly MMP9, which can cleave HSPG core proteins (77, 98). Cleavage of the core proteins in a glycoprotein such as perlecan can also occur via MMP1, MMP13, and plasmin (110) and MMP2 or MMP8 (77).

MMP9 is implicated as a major player in VEGF release and the angiogenic switch (86, 111), however its predominant mechanism of action on VEGF is unknown. MMP9 is thought to be primarily secreted by infiltrating macrophages and neutrophils (112–115) stimulated by tumor-secreted cytokines (Fig. 3). MMP9 is thought to raise the concentration of active VEGF in tissues, increasing VEGF-VEGFR2 association (86). While MMP9 has been shown to cleave murine VEGF (13), studies where MMP9 was implicated in release of murine VEGF have not shown evidence of VEGF cleavage (111, 113). MMP9-mediated cleavage of VEGF₁₆₄ may even be inhibited by heparan sulfates (13). Most evidence points to MMP9 releasing VEGF through HSPG cleavage (77, 98). For example, heparanase shows similar pro-angiogenic behavior as MMP9 in the RIP1-Tag2 pancreatic islet model (109) and heparanase action on cell-surface HSPGs can induce MMP9 expression to further cleave HSPGs (98) (Fig. 3D). However, MMP9 is also able to cleave soluble VEGF inhibitors and thus this alternate mechanism of VEGF activation cannot be excluded (70, 93).

3.4 Loss of VEGF activity from the tissue by clearance and degradation

Loss of VEGF via clearance and degradation plays a key role in controlling VEGF activity and in shaping VEGF gradients (19). Relevant mechanisms include lymphatic drainage, transvascular transport into the blood stream, proteolytic degradation, and cellular endocytosis; the relative importance of these mechanisms is not clearly understood.

Endothelial cells can degrade VEGF through internalization of the VEGF-VEGFR complex and through alternate pathways such as via low density lipoprotein receptor-related protein-1 (LRP1) in conjunction with TSP1 (116). In chronic wounds and some tumors, VEGF loss appears to be more dependent on proteolytic degradation (104, 117). This seems to require an initial cleavage by plasmin (13, 104, 117, 118), however the extent of degradation seems inconsistent between studies (47, 104) and it does not seem to occur with MMPs (13). VEGF experiences loss of activity in vitro under cellular conditions shown not be due to cellular uptake, with a rate constant of 2.3–2.8 ·10⁻⁴ s⁻¹ (τ_1/2 ~40 min) (119, 120). Whether this represents inactivation by secreted soluble factors (78) or degradation in solution is not known; factors present in the serum commonly added in many of these studies (120) may be involved. Finally, VEGF also has been demonstrated to degrade in isolation, with a half-life of ~96 min in acellular conditions at 37°C (28, 101). Matrix-sequestered VEGF may be protected from this intrinsic degradation (85, 101).

4.1 VEGF₁₂₀ demonstrates greater dispersion and lower pericellular binding in vivo than VEGF₁₆₄

The clearest examples of VEGF gradients are seen in images from the mouse hindbrain, retina, and cerebellum (14, 15, 18). In the hindbrain, mice engineered to secrete only VEGF₁₂₀ display a broad, diffuse immunostaining pattern, whereas wild-type mice, secreting predominantly VEGF₁₆₄, instead show a higher VEGF level at the midline that falls off more rapidly, i.e. a steeper gradient and shorter propagation distance (Fig. 4B) (15). In the retina, VEGF₁₂₀ secretion from astrocytes results in a dispersed VEGF distribution lacking obvious cellular localization; conversely, VEGF in wildtype mice predominantly associates with the astrocyte cell surface but is lacking in the interstitium (Fig. 4C) (14). VEGF₁₈₈ shows the greatest levels of matrix sequestration and/or pericellular localization (16–18). In support of these observations, expression of VEGF₁₂₀ alone results in greater levels of soluble VEGF (122) and lower levels of matrix deposition (16) in vivo, compared to systems expressing only VEGF₁₆₄ or VEGF₁₈₈.

The key feature of the VEGF gradient is that matrix-binding isoforms displays a higher total VEGF concentration at the source of secretion and at the surface of nearby cells, while VEGF₁₂₀ has a higher concentration than matrix-binding isoforms distant from the source (15, 18). Note that the flatter nature of the VEGF₁₂₀ distribution seems to be due to both the lack of amplification provided by matrix and cell-surface binding as well as VEGF₁₂₀’s ability to travel farther. The VEGF distributions in vivo also appear to be static (or only slowly evolving), indicating that they are not transient dissipative phenomena, but are continually and actively maintained. In the developing mouse hindbrain, the VEGF gradient, as well as the differences between VEGF₁₂₀ and VEGF₁₆₄ gradients, persists for at least three days (E10.5–E13.5), until vascularization of the subventricular zone is complete (15). Similarly, in the retina, the zone of greatest VEGF secretion appears to move with the circumferential expanding vascular front during the first postnatal week (14).

There are limitations to interpreting images of VEGF spatial gradients in vivo. First, the extent to which intracellular VEGF contributes to the immunostaining is not clear (82, 123); a large intracellular fraction of matrix-binding isoforms (17, 58) may make it difficult to determine the quantity of extracellularly diffusing VEGF. On the other hand, VEGF₁₂₀ is thought to elicit biological effects at a longer range than heparin-binding isoforms (13, 16, 124), suggesting that the immunostaining indeed reflects the true extracellular VEGF gradient. There may also be unaccounted factors associated with the biological models themselves. For example, dispersed VEGF gradients in the retina and hindbrain of VEGF^120/120 mice might be due to defective arteriolar development of neural-derived tissues, leading to more extensive hypoxia and hence VEGF secretion (125); global hypoxia can result in a similar retinal vascular phenotype as the VEGF^120/120 mice (126). Constitutive expression models could be informative, but clear examples of VEGF gradients have not been visualized in such systems (16). In contrast to developmental systems, which demonstrate vivid VEGF₁₂₀ staining interstitially (14, 15, 18), VEGF₁₂₀-expressing tumors show little VEGF staining (16, 17). A summary of VEGF isoform patterning is provided in Tables S2–S4.

4.2 HSPG binding cannot solely account for differences in VEGF gradients between isoforms

To study the diffusion and patterning of VEGF isoforms, we developed mathematical models of the microenvironment (19). In the models, VEGF is secreted into the interstitial space and diffuses through the ECM and basement membranes where it can reversibly bind to HSPGs; ultimately the VEGF in solution is either cleared away or degraded (Fig 5). Assuming an initial state of no VEGF, simulations show that upon secretion, diffusing soluble VEGF slowly loads the matrix until HSPGs reach an equilibrium binding capacity at steady state; the tighter an isoform binds the matrix, the more it can load in the matrix, and the longer this process takes (19, 50, 127). Comparing secretion of different isoforms, the models predict the spatial distribution of the unbound soluble fraction at steady state to be the same, regardless of the isoform’s affinity for the matrix. The matrix-bound fraction in each case had the same relative curvature as the soluble VEGF fraction, but with its concentration amplified in proportion to the HSPG-binding affinity (Fig. 5A, Sequestration). Matrix-binding isoforms would thus result in higher VEGF concentrations than VEGF₁₂₀ regardless of the distance from the secretion source, assuming sequestration only. In the model, HSPGs do hinder the rate and time course of diffusion as is commonly thought, but this only has an impact on gradients during the transient period, and not at steady state.

The observation that reversible matrix binding would not impact the VEGF gradient at steady state has a biophysical explanation: only soluble VEGF diffuses, and at steady state, for every soluble VEGF molecule that binds to the matrix, an equivalent molecule dissociates from the matrix, at each point in space (19). Thus, for equal secretion rates of each VEGF isoform, the same amount of soluble (diffusing) VEGF would be present at steady state. Furthermore, since the effective time spent by each isoform in solution, diffusing, would be identical, the underlying soluble VEGF gradient is also identical. VEGF₁₂₀ typically results in high levels of soluble VEGF relative to other isoforms; this is typically attributed to its lack of sequestration by the matrix, but the above analysis shows that this is not the case (19).

4.3 Combined sequestration and degradation can explain VEGF gradient formation

If isoform-dependent sequestration alone cannot explain experimentally-observed VEGF isoform gradients, an alternate mechanism must be responsible. Degradation has an important role in morphogen gradient formation in developmental Drosophila systems (127–129). Unlike the effects of sequestration, degradation is able to make VEGF gradients steeper (Fig. 5A, Degradation, Inset). Biophysically, degradation results in shortened lifespan of VEGF molecules, decreasing the amount of time spent in solution and, thus, the distance the molecules can diffuse. More rapid degradation results in steeper VEGF distributions and reduced soluble VEGF levels, however as a consequence, the total amount of VEGF is also significantly reduced (Fig. 5A, Degradation). Thus, degradation by itself is also insufficient to explain experimental data.

It is only when the effects of both sequestration and degradation are combined, such that VEGF isoforms are degraded at a rate that increases with their sequestration binding affinity, that the model predicts the observed behavior of VEGF isoforms in vivo (Fig. 5A, Degradation & Sequestration; note the intersection of VEGF concentration curves). In this scenario, VEGF₁₈₉ has a steeper distribution than VEGF₁₂₁ for two reasons: first, greater sequestration resulting in a high concentration near the source of secretion; second, increased degradation and therefore decreased time for diffusion in solution and a steeper fall-off in concentration. The in vivo observation that VEGF_120/1 has higher concentration is due to the decreased rate of degradation that it experiences.

The simplest mechanism that accounts for this behavior is sequestration and degradation being coupled. We posit that in vivo, bound VEGF, sequestered in an isoform-dependent manner, is subject to degradation, and that this accounts for patterning differences between isoforms (19). Matrix binding with subsequent intrinsic or proteolytic degradation is a possibility, however, matrix-binding is thought to stabilize VEGF (85, 101) and there do not seem to be indications of VEGF cleavage in the absence of pathology (69). Alternatively, VEGF may bind to the surface of cells in an NRP1- or HSPG-dependent fashion, both of which have progressively greater affinity for the longer isoforms, and be subsequently internalized and degraded (Fig. 5B). In zebrafish, knockout of perlecan has been shown to increase total VEGF levels with associated dispersal of the VEGF spatial gradient, indicating that perlecan may potentiate degradation or loss of VEGF (130). VEGF gradients are observed near the vascular front/avascular border in tissue such as retina and tumors (14, 15), which may be due to high levels of VEGF receptors and NRP1 at the endothelial cell surface (131, 132). A recent study shows that the endothelial cells at the sprouting front exhibit a higher rate of VEGFR2 internalization (133), which not only increases VEGF signaling at the front but could also directly shape the extracellular VEGF gradient. VEGF gradients may also be patterned in the stroma/parenchyma as NRP1 and HSPG co-receptors can also be found on stromal and parenchymal cells (18, 40, 66, 134, 135).

VEGF₁₂₀ gradients are observed in developmental systems but not tumor systems (16, 136), and it may be possible that different tissues operate at different ratios of sequestration to degradation. Gradients observed in developmental systems suggest that sequestration and degradation take place (e.g. Fig. 5A, Degradation & Sequestration), while tumors may have lower degradation of sequestered isoforms (e.g. Fig. 5A, Sequestration). A prediction of the sequestration-dependent degradation model is that total VEGF in the system is relatively fixed with respect to changes in the secreted isoform (or enzymatic release, discussed in Section 4.4) (Fig. 5A). Protein expression studies (western blots) may support this finding (111); while other studies do not (58, 80). A more direct method of validation would be to compare the gradients or half-lives of VEGF₁₆₄ with different VEGF isoforms, e.g. VEGF-E (binds NRP1 but not HSPG), VEGF₁₄₅ (binds HSPGs but not NRP1), or VEGF_{164Δ108–118}, a modified isoform which resists proteolytic degradation (13).

Other growth factor systems where spatial gradients are attributed to sequestration may also operate based on the principle of sequestration-dependent degradation. The ex vivo epithelial bud model branches in the presence of matrix-binding FGF10 but elongates in the presence of non-matrix-binding FGF7 (100). Interestingly, FGF10 appears to induce strong radially-directed autologous gradients (i.e. gradients generated by the cell), while FGF7 does not. While the original authors attributed these gradients to diffusion, we note that in their study, contrary to the authors’ interpretations, cleavage of cell-surface HS diminished elongation while cleavage of HSPGs in surrounding matrix did not, suggesting that autologous gradients may be due to cell surface HS-mediated internalization. Similarly formed autologous gradients may explain differences in morphogenic behavior observed between different VEGF isoforms in an embryoid model of PAE expressing VEGFR2, which lack NRP1 or sVEGFR1-based countergradients (13).

4.4 Proteolytic release of VEGF increases VEGF spatial range by reducing degradation

In models of carcinogenesis, inflammatory cells are thought to secrete proteases that release matrix-bound VEGF into solution, promoting diffusion and endothelial cell binding, and thereby inducing the angiogenic switch (86). Due to proteolytic release, VEGF is redistributed: matrix-binding and peritumoral VEGF localization is lost and VEGF binding to VEGFR2 on endothelial cells increases, while the total VEGF level is unaffected (69, 111, 137) (Table 1). In pancreatic islets of the RIP1-Tag2 mice, redistribution occurs without upregulation of VEGF or VEGFR2, and without a shift in VEGF isoform expression (86, 138); this permits exclusive study of the extracellular regulation of VEGF. Even in studies where protease expression may be associated with VEGF upregulation (93, 113, 139), redistribution is still evident by an increase in the soluble:total VEGF ratio (113, 137).

Since the angiogenic switch is irreversible, this implies that the proteolytic release of stored VEGF is not a one-time burst but is rather continuously maintained to keep a positive angiogenic balance. Increased vascular binding and VEGF-VEGFR2 association seen upon MMP9-, MMP12-, or heparanase-mediated release appears to be unchanging for weeks (86, 111, 115), a steady state scenario. Computational modeling of proteolytic VEGF release (via cleavage) suggests that, at steady state, VEGF release cannot occur any faster than soluble VEGF binds to the matrix, with the result that proteolytic release, like matrix binding, also does not affect soluble VEGF levels by itself (19). Thus, the observed increase in soluble VEGF due to proteases is not a result of more rapid solubilization of matrix-bound VEGF. While ex vivo experiments where proteases are applied to VEGF-containing matrices can show increased soluble VEGF concentration (77, 86, 140), this may be due to the absence of convective and other mechanisms for VEGF loss that are found in vivo.

Interestingly, the sequestration-dependent degradation model that gives rise to isoform-specific VEGF gradients can also reproduce VEGF redistribution due to proteolytic release. Increased soluble VEGF and increased VEGF-VEGFR2 association in the presence of protease suggests that the protease releases VEGF from sequestration and degradation (19); the released VEGF molecule effectively exhibits both decreased sequestration and degradation. This suggests that the proteases either release a VEGF bound to an ECM fragment that impairs further sequestration (99, 100), or that these enzymes cleave HSPG sequestration sites (77), preventing HSPG-associated VEGF degradation or cellular uptake. Similarly, cleaved VEGF_113/4 satisfies both requirements as it is not able to subsequently bind to either HSPGs or NRP1. When proteases cleave VEGF inhibitors, VEGF may be released and activated, avoiding both its sequestration and its clearance by the inhibitor.

Tip cells are also known to secrete proteases, which have been implicated in local release of VEGF, enhancing spatial gradients and morphogenesis (141). Our model argues against this occurring in vivo. If proteolytic release results in a decrease in VEGF degradation, gradients would be expected to be flatter: absolute gradients can be enhanced far away, but locally they would be reduced. Furthermore, the tip cell by itself has little proteolytic potential in isolation and is unable to alter soluble VEGF levels (74). Instead, we hypothesize that proteases can alter cellular behavior by acting at the cell surface and modulating active, cell-surface associated VEGF. Indeed, VEGF cleavage has been shown to primarily occur at the cell surface in which case it results in decreased morphogenetic behavior (13).

4.5 Tissue degradation impacts VEGF transport throughout the body

In previous sections, we have focused primarily on local VEGF availability within a tissue, such as a tumor or muscle. The VEGF in those tissues comes primarily from the parenchymal and stromal cells of that tissue (134, 142, 143). VEGF in the bloodstream is primarily a result of clearance and extravasation from tissues and plasma clearance; we have predicted the role of luminal VEGF secretion by the endothelium in recent models (144).

Loss of VEGF in tissues explains the counterintuitive result that serum levels of VEGF increase following administration of bevacizumab, an antibody-derived VEGF sequestering molecule (142, 145). Based on experimental (146) and theoretical studies (50, 143), a fraction of VEGF from parenchymal cells is either captured and internalized by endothelial cells or degraded in the interstitium. Pharmacokinetic models show that anti-VEGF agents have a dispersive effect, similar to that of proteolytic release of VEGF. Upon injection, bevacizumab can extravasate into tissues, bind VEGF and inhibit receptor-mediated internalization. Depending on the tumor microenvironment, this may divert a part of the flux of tissue VEGF into the blood stream resulting in an increase in total plasma VEGF (142). Plasma free VEGF may also increase if the VEGF-bevacizumab complex preferentially dissociates in plasma, based on mass action principles, and may explain the counterintuitive observation of VEGF increase following administration of a VEGF sequestering agent (142, 147, 148). Other agents blocking receptor-mediated internalization of VEGF could elicit a similar phenomenon, e.g. sVEGFR1 (143) or isoform-specific agents such as anti-NRP1 antibodies (39, 149). We predict that this phenomenon is not observed with VEGF-Trap (aflibercept) due to its very strong affinity for VEGF (1000 times higher than bevacizumab), resulting in little dissociation of VEGF-Trap–VEGF complex in the blood stream (144).

5.1 Tissue VEGF expression and role of VEGF isoforms in development and adult

VEGF levels must be highly regulated during development, as either a two-fold gain or 50%-loss of VEGF expression results in lethal cardiovascular complications (152, 153). However, across tissues, total VEGF expression has a wide range. In the adult, highest expression rates are found in the omental adipose tissue, lung, retina, ovary; moderate amounts in the kidney, heart, skeletal muscle, and adrenals; and low levels expressed in liver, brain, and breast (25, 79, 154–156). VEGF has numerous roles in the adult organism (157, 158). VEGF seems to be most highly expressed where it is actively being used either for angiogenesis, such as the ovary, or for maintenance of basal permeability such as in the lung, kidney, and heart (154). In contrast, the brain has low VEGF expression and exceptionally low basal permeability (159).

VEGF isoform expression also varies across organs as well as in pathology. VEGF_164/5 is the predominant isoform expressed in normal adult tissues (Fig. 2, Table S5), with VEGF_188/9 also high in tissues such the lungs, heart, and liver which are initially vascularized by vasculogenesis (25). In the lungs, alveolarization and alveolar vascularization are coupled, and loss of heparin binding isoforms results in reduced alveolarization and vascularization. During development, alveolar vasculature begins with sprout formation from the primitive plexus, which then undergoes extensive intussusceptive angiogenesis, the in situ division of a vessel by transcapillary pillars of stroma, to mature and reach vascular complexity (160–162). VEGF₁₂₀ is able to support initial vessel outgrowth, but heparin-binding VEGF isoforms are thought to be needed for continued vascularization and maturation – VEGF₁₈₈ can maintain strong localization and is upregulated by Type II alveolar epithelial cells during primitive alveolar formation (25). Intussusception plays an important role in developmental angiogenesis, however the involvement of VEGF in intussusception is uncertain (162). VEGF seems to be overexpressed in the sprouting phase of network formation and down-regulated during intussusception (160), however other studies show that VEGF can induce splitting forms of angiogenesis (29, 163).

VEGF isoforms may also have other roles such as maintenance of permeability (164) or promotion of inflammation (165). For example, the high VEGF expression in the lung, specifically of VEGF₁₆₅ and VEGF₁₈₉, may also have an additional function in recruiting immune cells to provide basal alveolar immunity (165). Pathological states of increased neovascularization, such as tumors, consistently seem to have an increased relative expression of non-heparin-binding isoforms such as VEGF_120/121 (Fig. 2), the significance of which is currently unknown and which seems non-optimal for tumor growth (145), since the most rapid tumor growth is usually found with expression of VEGF_164/5 (16).

5.2 VEGF splice and cleavage isoforms induce a spectrum of vascular phenotypes

The connection between VEGF isoforms and vascular morphology has been investigated using both in vivo animal experiments and ex vivo explant cultures (23). Particularly useful are the VEGF isoform-specific mice, generated by replacing the multiple splice-site vegfa gene with cDNA for a specific isoform, resulting in VEGF secretion rates for that single isoform equal to the total of all isoforms in wildtype mice (14, 15, 24, 166).

Transgenic mice that express only VEGF₁₂₀ display widespread perfusion defects from impaired angiogenesis. Severe ischemic cardiomyopathy in these mice results in death soon after birth or within two weeks (24, 25); the heart almost exclusively expresses heparin-binding isoforms in wildtype mice (Fig. 2) (24, 25). However, vascular defects were found in most organs studied including renal glomeruli (decreased glomerular arteries), pulmonary alveoli, retina, bones, and brain (14, 15, 25, 125, 167, 168). Expression only of VEGF₁₂₀ resulted in fragile, leaky vessels because of poor pericyte coverage, decreased vessel density and poor outgrowth in organs, and impaired arterial and venous development (24, 125). Specifically, sprouting angiogenesis suffered from filopodial disorganization, decreased migration and branching, and formation of closed, blind-ended loops resembling glomeruli (14, 15, 125). In contrast, transgenic mice that express only VEGF₁₈₈ exhibited higher vascular density than wildtype mice, normal pericyte coverage, and normal coronary and glomerular architecture; however these mice had severe deficits in arterial development in the retina (125) and half of these mice died in utero. VEGF₁₆₄-only mice were viable and displayed vascularization that was phenotypically similar to wildtype mice (125).

Specific vascular phenotypes resulting from specific VEGF isoforms have also been observed in tumors. Tumors expressing only VEGF₁₂₀ implanted subcutaneously show poor internal vascularization and poor branching, and peritumoral vessels are enlarged and edematous (16, 80, 169). In contrast, VEGF₁₈₈-expressing tumors exhibit a high density of low caliber vessels, with significant branching, rich pericyte coverage, and low permeability (16, 80, 170).

Thus, there appears to be a spectrum of vascular behavior dictated by VEGF isoforms, with VEGF₁₂₀ and VEGF₁₈₈ at opposite ends. The severity of vascular defects in particular organs, e.g. in the heart and lungs, correlates with endogenous patterns of VEGF isoform expression in wildtype mice (24) (Fig. 2). Furthermore, vascular phenotype seems to be dependent on functional properties such as VEGF gradients, HSPG binding, or NRP1 binding, and as a result, isoforms can be substituted as long as overall function is maintained. For example, dual expression of VEGF₁₂₀ and VEGF₁₈₈ in VEGF_120/188 heterozygote mice or tumors mimics wildtype and VEGF₁₆₄-specific vascular phenotypes despite not expressing VEGF₁₆₄ (15, 16, 23). This functional equivalence of one allele each of VEGF₁₂₀ and VEGF₁₈₈ and two alleles of VEGF₁₆₄ defines a boundary between isoform ratios that yield typical vascular patterns (Fig. 2, gray dashed line). The heart has the closest overall isoform profile to that of VEGF₁₈₈ only, and this organ showed the greatest defects in VEGF₁₂₀-expressing mice. We also note that U87MG (human glioblastoma cell line), despite having a lower fraction of VEGF_164/5 expression relative to normal mouse brain, may have similar net VEGF properties due to the ratio of VEGF₁₂₀ to VEGF₁₈₈.

VEGF₁₂₀, VEGF₁₆₄, and VEGF₁₈₈ have a monotonic relationship between isoform length and properties such as matrix binding and NRP1 binding affinity, and these properties have been suggested as possible key drivers of the above spectrum of vascular phenotypes. VEGF cleavage, producing VEGF_113/4, has a similar phenotypic effect as moving the isoform balance towards VEGF₁₂₀; proteolysis-resistant VEGF₁₆₄ (VEGF_{164Δ108–118}) results in a similar vascular phenotype to VEGF₁₈₈ (13). This was interpreted by Lee et al to indicate that endothelial phenotypes are driven by the levels of matrix-bound VEGF in tissues. In contrast, VEGF₁₄₅ binds HS more strongly than VEGF₁₆₅ but does not bind NRP1 (44), and this isoform shows weak angiogenic activity, leading Kawamura et al to conclude that NRP1-binding is the key driver of angiogenic phenotype (44).

Not all studies are consistent with the above described monotonic spectrum of vascular phenotypes (Table S6). In specific cases, VEGF₁₈₉ has been observed to enlarge vessels relative to VEGF₁₆₅ (66) while VEGF₁₂₁ can result in stronger angiogenesis than VEGF₁₆₅ (58, 124, 171). In addition, implantation of myoblasts transfected with any one of the three main isoforms displayed similar vascular malformations such as glomeruloid-like proliferations in skeletal muscle (172).

5.3 VEGF control of the sprouting branching phenotype through gradients, matrix-sequestration, and NRP1-dependent sensing

The vascular phenotypes resulting from different VEGF isoforms is related to the ability of isoforms to induce the sprouting, migratory phenotype. Sprouting angiogenesis starts with VEGF activation of a nascent vessel, sprout formation, basement membrane degradation, sprout extension, lumen formation, and finally anastomosis to another sprout or vessel to complete a flow circuit (for excellent reviews, see (173, 174)). On completion of angiogenesis, the new capillaries can be stabilized by pericytes or undergo pruning (by cellular apoptosis) based on local metabolic demands.

Angiogenic sprouts consist of a distal, VEGFR2-positive migratory tip cell and one or more trailing stalk cells that are highly proliferative and mediate lumen formation (14); the tip and stalk cells can re-order and be dynamically re-assigned based on local VEGF cues (175). Sprouts extend in the direction of VEGF gradients by stabilization of filopodia, dynamic mechanical and sensory apparatuses primarily found on tip cells (14), possibly assisted by collective shifting of roles (175, 176). Branching density (high for VEGF₁₈₈, low for VEGF₁₂₀) is thought to reflect the rate of new sprout initiation (15); alternately it may reflect decreased vessel pruning due to increased stabilization by pericytes (125). VEGF, particularly VEGF₁₂₀, can induce vessel enlargement (14, 177, 178), possibly due to stalk cell proliferation without accompanying tip cell migration. This can cause tortuosity and coiling of sprouts, and the formation of blind-ended tufts as is seen with VEGF₁₂₀ alone, or in the absence of NRP1 (15, 179).

VEGF initiates endothelial cell activation and induces sprouting (14, 20). VEGF activation through VEGFR2 on an endothelial cell in the parent vessel leads to promotion of the tip cell phenotype (20), with simultaneous inhibition of the tip cell phenotype in adjacent cells via Dll4-Notch-1 signaling. In a vessel with cells undergoing such local mutual competition, the outcome is thought to be dictated by local VEGF and stochasticity in VEGFR2 expression (175). VEGF spatial heterogeneity and gradients are thought to increase the frequency of sprouting (15, 141, 180, 181) and to enhance the migratory phenotype (83, 182, 183). Dynamic filopodial extension causes local variation in VEGFR2 density, and coupled with VEGF heterogeneity, this enhances variability in VEGFR2 activation between cells, leading to more effective distinction of cell fates. The strong, heterogeneous VEGF gradient seen in VEGF₁₈₈ can thus lead to numerous tip cells and sprouts, ultimately resulting in a high branching density (15). The vascular phenotypes resulting from VEGF₁₈₈ expression and from reduced Dll4/Notch signaling (Dll4^+/− mice or Dll4-neutralizing antibodies) are similar, characterized by increased tip cell formation, high branching density, decreased vessel caliber, and overall poor vascular function (16, 80, 184). However, these phenotypes are not exactly the same: with partial loss of Dll4, poor vascular function is due to immature vessels (185), while in VEGF₁₈₈, vessels are pericyte rich (80) but possibly poorly connected to external vessels (16). The high sprouting density may be due to higher variation in tip:stalk cell signal for VEGF₁₈₈, and due to insufficient lateral inhibition in the case of Dll4^+/− (186).

Manipulations that disrupt existing VEGF gradients, e.g. the VEGF^120/120 genotype, hypoxemia, or injection of VEGF into the eye, result in decreased frequency of sprouting, decreased migration with disorganized filopodia, and enlargement of existing vessels (14, 126). Loss of sVEGFR1, which is thought to form a counter-gradient that enhances the local VEGF gradient (20, 21), similarly results in fewer tip cells and increased but disorganized filopodial protrusion (20). In this situation, vessels are more likely to enlarge through proliferation rather than by sprouting, due to the combined effect of fewer tip cells (due to mutual lateral inhibition (180)) and defective migration (14). Interestingly, other experiments suggest the opposite – that high levels of VEGF, expressed from the lens or injected into the eye, result in exuberant tip cell formation though with abnormal sprouting angles (14, 20). How high levels of VEGF can lead to both states of increased or decreased sprouting is not known (180), but sufficiently high levels seem to suppress sprouting behavior according to theoretical models (187).

While branching and migration due to VEGF₁₈₈ may be driven primarily by the intensity and heterogeneity of VEGF gradients, the role of distinct signaling outcomes mediated by NRP1-dependent or matrix-bound VEGF cannot be excluded. The role of NRP1 was not initially obvious, as neither VEGF₁₂₀ nor VEGF₁₈₈ were thought to bind NRP1, which yet when co-secreted displayed a similar phenotype to VEGF₁₆₄ (15). It has since been shown that VEGF₁₈₈ has stronger NRP1 binding than VEGF₁₆₄ (66, 67). Furthermore, injection of VEGF₁₆₄ into the retina, which disrupts endogenous gradients but increases matrix-bound VEGF and NRP1-dependent VEGFR2 signaling, disrupted the sprouting/migratory phenotype which was taken to imply that HSPGs and NRP1 were by themselves insufficient to mediate sprouting (14). That experiment, however, cannot rule out the possibility that VEGF gradients are themselves sensed in a NRP1-dependent or bound VEGF-dependent manner. Subsequent studies comparing VEGF₁₂₁, VEGF₁₆₅, VEGF-E, and VEGF₁₄₅ show that both NRP1 and HSPGs enhance the frequency of sprouting, through increased p38/MAPK signaling (27, 44).

Furthermore, in spheroids of porcine aortic endothelial cells expressing VEGFR2 (PAE-VEGFR2), which do not express NRP1, uncleavable VEGF₁₆₄ (VEGF_{164Δ108–118}) produced strong sprouting as well as proliferation while pre-cleaved VEGF (i.e. VEGF₁₁₃) resulted only in proliferation (13). This suggests that while both matrix-sequestered and soluble VEGF enable proliferation, matrix-sequestered VEGF additionally provides cues necessary for organized sprouting morphogenesis. There may be a role for isoform-specific autologous VEGF gradients in this system which cannot be excluded, as noted in Section 4.3. Subsequent studies showed that whereas soluble VEGF₁₆₅ leads to prolonged activation of Akt involved in proliferation and survival, matrix bound VEGF₁₆₅ resulted in a distinct signaling state characterized by VEGFR2 clustering and prolonged activation of p38/MAPK (28). Even matrix-tethered VEGF₁₂₁ was shown to induce a sprouting migratory phenotype, whereas soluble VEGF₁₂₁ induced proliferation and vessel malformations (106, 188). Simulations further suggest that individual tip cells are unable to appreciably alter local soluble VEGF levels; instead, proteases can act on active, cell-surface associated VEGF to alter VEGF signaling (74). In support of this, loss of HS on endothelial cells result in decreased vessel outgrowth in a spheroid model (189) and decreased vessel branching density in tumor xenografts (46). In retinal angiogenesis, loss of HS and fibronectin binding sites for VEGF on astrocytes results in decreased vascular plexus outgrowth similar to the secretion of VEGF₁₂₀ (84).

5.4 VEGF levels control the spectrum of normal to pathological angiogenesis

While the study of VEGF isoforms helps elucidate mechanisms of sprouting angiogenesis, sprouting does not explain some clinically important vascular behavior. For example, tumor vasculature is characterized by a highly pro-angiogenic balance, leading to an exuberant, irregularly branching, destabilized network (3, 190, 191). Tumor vessels are enlarged, tortuous, leaky, and contain glomeruloid-like swellings, features which lead to stasis/coagulation, edema, resulting in acidic hypoxic microenvironments. Similar dysfunction is also found in retinopathy and neovascular macular degeneration (69, 192) and controlling these excessively angiogenic states is currently a major therapeutic goal.

Tissues can rely on alternate mechanisms of vascularization, e.g. intussusceptive angiogenesis and intraluminal bridging (192–195). Unlike intussusceptive angiogenesis, intraluminal bridging (also known as longitudinal division) divides a parent vessel by intraluminal extension of endothelial processes; it is disputed whether they represent distinct mechanisms (161, 195). Here, we collectively refer to them as splitting angiogenesis. In exercising skeletal muscle, both sprouting and splitting angiogenesis have been shown to occur (29, 196). Splitting is more immediately functional than sprouting angiogenesis as it does not require cellular proliferation, invasion or anastomosis, it retains patency throughout the process, and does not affect the permeability of the original vessels (29, 193). It results in better oxygen transport at high oxygen consumption rates (193, 197). Tumors may switch from sprouting to intussusception following anti-VEGF therapy (198).

What leads to a sprouting response as opposed to these alternative vascularization patterns? Sprouting is primarily a mechanism to vascularize avascular tissues and thus is thought to be dominant when strong VEGF gradients are present, e.g. at the edge of the avascular retina (14). In contrast, in regions with high existing vascular density, such as in muscle (199) or in experimental settings where VEGF is introduced directly into tissues (200), vascularization can proceed more efficiently through splitting of existing vessels. This is consistent with developmental angiogenesis where sprouting creates an initial vessel that can expand into a more complex network through intussusception (160). Which surface of the vessel – luminal or abluminal – receives stimulation may also dictate vascularization (196). Similar to abluminal VEGF gradients, exercise-induced mechanical stretch in muscles and interstitial flow enhance sprouting (141, 181, 196, 201); in contrast, intraluminal shear stress strongly induces intussusceptive angiogenesis (196) and suppresses sprouting (181). Finally, VEGF concentrations play a role. Sprouting is thought to occur best with hypoxia-induced, modest increases in VEGF expression, which preserve existing gradients, whereas splitting angiogenesis occurs more at moderate-to-high VEGF concentrations, which may result in vessel dilation or enlargement (178, 199) and enhanced luminal flow (29).

In a quiescent venule or capillary, appropriate concentrations of VEGF can induce sprouting or splitting angiogenesis without resulting in abnormal vascular structures (29, 202). VEGF increases vessel diameter (178), but at high concentrations, VEGF induces a strong morphogenetic and proliferative response, with pericytes detaching and the vessel enlarging to form an abnormal, highly permeable sinus called a mother vessel (199, 200). Mother vessels can subsequently undergo splitting to form capillaries, can re-associate with pericytes to form stable enlarged vessels called vascular malformations, or can form glomeruloid microvascular proliferations (191). Glomeruloid proliferations arise through continued proliferation of the vessel wall intraluminally and abluminally, subsuming the original mother vessel lumen and forming small channels (203). Other mechanisms of formation of glomeruloid bodies have also been proposed (204). Glomeruloid bodies are loosely associated with perivascular cells, are leaky, and can become hemangiomas in the presence of continued high VEGF (202).

Pathological angiogenesis results from vascular destabilization. VEGF induces EC proliferation, and loss of pericyte coverage occurs through disruption of PDGFR-β activity and an increase in Ang-2 signaling, while NRP1 may promote pericyte coverage (205). Among the VEGF isoforms, VEGF₁₂₀ has the highest propensity for unstable angiogenesis due to its shallow gradients and low NRP1 association; VEGF₁₈₈ induces significant pericyte coverage and is protective (206, 207). Despite this, high levels of any of the VEGF isoforms can give rise to glomeruloid bodies (172). Isoforms also vary in terms of their matrix binding affinities. Soluble VEGF₁₂₁-loaded fibrin gels placed on the chorioallantoic membrane released VEGF in a diffusive burst, and induced splitting angiogenesis and hemangioma formation; gels formulated with fibrin-tethered VEGF₁₂₁, which allowed for low levels of release over time in a cell-demanded manner, resulted in stable, branched vessels with pericyte coverage (106, 188). The release of VEGF from matrix sequestration into solution is associated with a pro-angiogenic and hence destabilized state (86), but matrix-bound VEGF appears to have similar proliferative activity as soluble VEGF (13, 28, 208). It is unknown whether matrix-bound VEGF achieves its increased vessel stability relative to soluble VEGF due to differences in VEGF localization or to differences in stability-promoting signaling pathways. Numerous other biological factors are also involved in vascular stability, including hypoxia, which induces endothelial HIF2α, pericytes which induce Ang-1 or Ang-2 signaling, and perivascular NO gradients, reviewed in (3).

The goal of anti-angiogenic therapy has typically been to prune unwanted vasculature, however this can be difficult because pathological angiogenesis, especially in tumors, can result in leaky, low-flow vessels, e.g. mother vessels and glomeruloid proliferations, make the surrounding tumor or tissue a poor target for drug delivery. Furthermore, the underlying acidosis and hypoxia, while reducing tumor growth rate, favor long term tumor expansion by promoting epithelial-to-mesenchymal transition and malignancy potential, and hindering the action of immune cells, radiation, and chemotherapy (209). Glomeruloid proliferations, indicative of a highly angiogenic phenotype and of vascular disorganization, are associated with poor patient prognosis (210, 211). Anti-angiogenic therapy has shown poor clinical success, possibly because a large subset of tumor vessels may be VEGF-independent (191). An alternate therapeutic strategy is to use anti-angiogenic therapy for a short period to enable normalization of pathological structures, e.g. mother vessels and glomeruloid proliferations (212), thereby enabling normal blood flow patterns and delivery of cytotoxic agents (3, 191, 209, 213). An alternate hypothesis proposes that vascular normalization occurs by intussusception (198). Normalization may be more successful in tumors showing signs of vascular disorganization, such as the presence of glomeruloid proliferations, or those that secrete higher levels of VEGF₁₂₁; for example, VEGFR2 blockade specifically increased blood flow in VEGF₁₂₀-expressing but not VEGF₁₈₈-expressing tumors (206, 214).

1. Regulation of angiogenesis by VEGF isoforms

Vascularization is controlled by overall VEGF concentrations and by the balance of the various VEGF splice and proteolytic isoforms. These isoforms regulate the balance of branching/migratory and proliferative behavior. The endothelial cell decisions are based on isoform-specific differences in: spatial patterning and gradients; binding to cell surface receptors, the VEGFRs and NRPs; binding to the ECM. Recently, the role of matrix-bound VEGF in directly binding to VEGF receptors has been recognized. Clinically, control of vessel morphogenesis through the VEGF family would be of use in a wide range of diseases.

2. Role of VEGF isoforms and vascular morphology in perfusion, function, and in tumor growth

Stimulation with different VEGF isoforms results in structurally different vasculature, and this can lead to different vascular architecture in different tissues, as well as in pathological situations. It is not clear in all cases what functional differences – for example, in perfusion and oxygen delivery – result from these differences. For example, tumors seem to be characterized by relatively high expression of VEGF₁₂₁; however, single-isoform xenograft studies suggest that VEGF_164/5-tumors grow most efficiently. It is unclear whether VEGF₁₂₁, which causes deranged vasculature, poor tumor growth, and hypoxia, may result in greater malignant potential in the long run, for example through hypoxia-related hypermutation.

3. VEGF gradients arise from a balance of sequestration and degradation in the microenvironment

VEGF isoforms, like other growth factors, show increasing localization with increasing affinity for ECM proteoglycans. We hypothesize that in order to give rise to the observed VEGF isoform gradients, isoforms with increased matrix-binding affinity are also lost from the microenvironment at higher rates (19). This suggests that VEGF sequestered in an isoform-dependent manner (either in the ECM or via cell surface co-receptors such as NRP1 or HSPGs) is subsequently lost via proteolytic degradation or internalization. It is difficult to verify which (or both) of these mechanisms holds in vivo without more detailed observations of VEGF gradients. This also explains observations following VEGF release by proteases, which we hypothesize disperse VEGF by interfering with sequestration-dependent degradation. The balance of sequestration and degradation may be tissue specific. Our analysis assumes a dynamic equilibrium or pseudo-steady state for VEGF, which seems to hold true in vivo.

4. Different enzymes involved in VEGF release have different effects on angiogenic potential

VEGF-cleaving proteases seem to suppress angiogenesis and vascularity, while VEGF release by heparanase is pro-angiogenic. The mechanism by which MMP9 releases VEGF is still unknown, but its pro-angiogenic role tends to support a HSPG-cleaving activity. An interesting question is whether rapid release of cleaved VEGF is able to overcome diffusible VEGF’s intrinsically lower activity compared to matrix-bound VEGF and induce angiogenesis in vivo. Furthermore, if cleaved VEGF levels subsequently return to baseline, would the pro-angiogenic effect be sustained? It has previously been shown that even a temporary burst of VEGF can induce stable angiogenesis, which has been termed the Spike Hypothesis (215).

5. Soluble VEGF inhibitors play an important role in VEGF patterning

The role of soluble VEGF inhibitors in altering VEGF gradients is only recently being recognized. Similar to HSPGs, soluble VEGF inhibitors seem to have multiple effects, from maintenance of quiescence to control of VEGF gradients and sprouting morphogenesis. Understanding how proteases cleave VEGF inhibitors and thereby induce VEGF redistribution and exert control on angiogenesis is an emerging area of study.

6. Computational models in coordination with experimental studies can provide deeper insight into complex biological mechanisms

There have been significant attempts to simulate angiogenesis in computational models based on VEGF transport and endothelial cell responses (183, 216–221); however, it is clear from experimental studies that our knowledge of VEGF transport in tissues is incomplete. The strength of computational models lies in their ability to simulate complex interactions and determine non-intuitive relationships (183). Models can be used to study biological mechanisms, and to generate new hypotheses that can then be tested experimentally.