New functional roles for non-collagenous domains of basement membrane collagens

Collagens IV, XV and XVIII are major components of various basement membranes. In addition to the collagen-specific triple helix, these collagens are characterized by the presence of several non-collagenous domains. It is clear now that these ubiquitous collagen molecules are involved in more subtle and sophisticated functions than just the molecular architecture of basement membranes, particularly in the context of extracellular matrix degradation. Degradation of the basement membrane collagens occurs during numerous physiological and pathological processes such as embryonic development or tumorigenesis and generates collagen fragments. These fragments are involved in the regulation of functions differing from those of their original intact molecules. The non-collagenous C-terminal fragment NC1 of collagen IV, XV and XVIII have been recently highlighted in the literature because of their potential in reducing angiogenesis and tumorigenesis, but it is clear that their biological functions are not limited to these processes. Proteolytic release of soluble NC1 fragments stimulates migration, proliferation, apoptosis or survival of different cell types and suppresses various morphogenetic events.

Each α-chain is composed of three domains, a cysteine-rich N-terminal 7S domain, a central triple-helical domain and a globular C-terminal non-collagenous NC1 domain (Fig. 1A). The NC1 domain is involved in the assembly of α-chains to form the heterotrimers and the 7S domain is involved in the covalent assembly of four heterotrimers in a spider-shaped structure (Boutaud et al., 2000;Dolz et al., 1988;Tsilibary et al., 1990). Thus, collagen IV forms a complex branching network that serves as a scaffold for the BM (Yurchenco et al., 1987) (Fig. 1B). Although α1 and α2 chains are widely expressed and colocalize in numerous tissues, there is a temporal and spatial regulation of α3, α4, α5 and α6 expression in physiological as well as pathological processes (Dehan et al., 1997;Fleischmajer et al., 1997;Miner and Sanes, 1994;Shen et al., 1990;Tanaka et al., 1997).
Several human genetic diseases have provided insights into the physiological role of collagen IV. Mutations in Col4a5, Col4a3 and Col4a4 are involved in Alport syndrome, which is characterized by a defective glomerular BM and the subsequent development of glomerulonephritis (Mochizuki et al., 1994), whereas some deletions spanning the 5′ regions of the Col4a5/Col4a6 cluster have been associated with Alport syndrome and diffuse leiomyomatosis, a benign smooth muscle tumor (Heidet et al., 1997;Zhou et al., 1993). The Cterminal region of the α3 chain has been identified as an autoantigen involved in Goodpasture syndrome, an immune disease characterized by glomerulonephritis and pulmonary hemorrhage (for a review, see Hudson et al., 1993;Kalluri, 1999). Mouse models for autosomal Alport syndrome have been developed and phenocopy human disease (Cosgrove et al., 1996;Cosgrove et al., 1998;Lu et al., 1999;Miner and Sanes, 1996).
The existence of different α chains for collagen IV and their restricted tissue distribution determine the structural and functional specificity of BM. Moreover, the extremely high conservation of this molecule from lowest metazoans up to vertebrates identifies collagen IV as a key regulator of morphogenesis that is critical for the regulation of adhesion, migration and survival of different cell types.
The genes encoding α1 chains of collagens XV and XVIII have been cloned and mapped to human chromosomes 9 and 21 and mice chromosomes 4 and 10, respectively (Hagg et al., 1997a;Huebner et al., 1992;Oh et al., 1994a).
Collagen XV is highly expressed in heart, skeletal muscles and the placenta and moderately expressed in the adrenal gland, kidney and pancreas (Kivirikko et al., 1995;Muragaki et al., 1994). Its expression is associated with vascular, neuronal, mesenchymal and some epithelial BM, indicating a probable function in adhesion between BM and the underlying connective tissue stroma (Myers et al., 1996). Highly regulated during kidney, heart and lung development in the embryo, collagen XV colocalizes with collagen IV and is a component of continuous or fenestrated capillary BM, with the exception of the blood-brain barrier and liver and spleen sinusoids (Hagg et al., 1997b;Muona et al., 2002). For collagen XVIII several splicing variants exist: two isoforms have been described in humans (Saarela et al., 1998a) and three in mice (Rehn and Pihlajaniemi, 1995). Interestingly, these variants have specific expression patterns. The short isoform is expressed in various organs, whereas the long isoform is more specifically expressed in liver sinusoids and hepatocytes (Musso et al., 2001;Saarela et al., 1998a;Saarela et al., 1998b). Transcription of collagen XVIII also occurs during adipogenesis (Inoue-Murayama et al., 2000) and a strong expression is observed in developing and post-natal eyes in various BM, except Descemet's membrane (Fukai et al., 2002).
The physiological roles of collagens XV and XVIII are not well understood. Mice lacking collagen XV show a higher sensitivity to exercise-induced muscle injury and progressive degeneration of skeletal muscles with collapsed capillaries and endothelial cell degeneration. Thus collagen XV might be involved in the survival and stabilization of muscle fibers and endothelial cells on the subjacent BM (Eklund et al., 2001). For collagen XVIII, a mutation affecting the short isoform has been recently associated with Knobloch syndrome, an autosomal recessive disorder characterized by high myopia, vitreoretinal degeneration with retinal detachment, macular abnormalities and occipital defects (Sertie et al., 2000). Interestingly, mice lacking collagen XVIII develop the same ocular abnormalities (Fukai et al., 2002). Thus, collagens XV and XVIII regulate critical functions within specialized BMs.

Structure, localization and receptors for C-terminal NC1 fragments of collagen IV, XV and XVIII
The NC1 domains of collagens IV, XV and XVIII have been isolated from invertebrate and vertebrate BM and identified as circulating fragments in serum. Collagen XV and XVIII share similar structures within their NC1 domains, which are organized into three different subdomains (Fig. 2B) Sazaki et al., 2000). Endostatin, the C-terminal subdomain of NC1 (XVIII), was the first endogenous collagen fragment characterized with anti-angiogenic properties (O'Reilly et al., 1997). On the basis of its high degree of sequence similarity with collagen XVIII, the C-terminal subdomain of NC1 (XV), named restin (Ramchandran et al., Roles for non-collagenous domains of collagens 1999) or endostatin-like § , was also identified as an endogenous angiogenesis inhibitor.
For collagen IV, the NC1 domains from α1, α2 and α3 chains have also been identified as inhibitors of angiogenesis and named arresten, canstatin and tumstatin, respectively (Colorado et al., 2000;Kamphaus et al., 2000;Maeshima et al., 2000b). These NC1 domains have been implicated in the self-association of the heterotrimers (Boutaud et al., 2000;Timpl and Brown, 1996;Tsilibary et al., 1990). The recent crystal structure of the collagen IV NC1 domain showed that NC1 monomers fold into a novel tertiary structure comprising β-strands and two homologous subdomains, N and C. The trimers are assembled through unique three-dimensional domain swapping (Sundaramoorthy et al., 2002), and two trimers can be stabilized head to head by an uncharacterized covalent crosslink (Than et al., 2002).
The crystal structure of endostatin revealed a compact fold with a zinc-binding site and an extensive basic patch of 11 arginine residues, which explains the high affinity of endostatin for heparin. The overall structure of the endostatin domain is related to the C-type lectin carbohydrate-recognition domain, and the domains are present as dimers in the crystals (Ding et al., 1998;Hohenester et al., 1998;Hohenester et al., 2000). The structure of endostatin-like is very similar to that of endostatin (60% sequence identity) but lacks the zinc and heparin-binding sites .
Endostatin and endostatin-like fragments colocalize with collagen XV and XVIII in the BM of numerous organs, with the exception of the liver sinusoids, where endostatin-like but not collagen type XV staining is present (Miosge et al., 1999;Sasaki et al., 2000;Tomono et al., 2002).
What are the receptors for these fragments? Numerous integrins have been identified as major cellular receptors for NC1 fragments (Table 2). For endostatin two cell surface binding sites with Kd values of 18 pM and 200 pM have been described. The low-affinity receptor corresponds to glypicans, whereas the high-affinity receptor has not yet been identified. These receptors are not specific to endothelial cells; they are also present on epithelial cells .
Recently it has been shown that endostatin binds to VEGF-R2, a receptor involved in proliferation of endothelial cells. Besides integrins, two other types of receptors have been described for collagens, glycoprotein VI in platelets and discoidin domain receptors (DDR) in various cell types (for a review, see Vogel, 1999). But whether or not these interactions occur through NC1 domains of collagen IV, XV and XVIII is not known. Furthermore, several ECM proteins interact with these fragments (Table 2). Taken together, these data suggest that NC1 fragments might be involved not only in the regulation of angiogenesis but also in other morphogenetic processes.

Proteolytic pathways generating NC1 fragments
For collagen IV the proteolytic pathways involved in generation of NC1 fragments are not clear, even though different fragments are detected in the serum. For collagens XV and XVIII, several different NC1 fragments have been extracted from tissue homogenates, and circulating forms have been isolated from human blood filtrates, suggesting that these fragments exist as physiological cleavage products (John et al., 1999;Sasaki et al., 1998;Standker et al., 1997). Endostatin originally was purified from conditioned medium of a murine hemangioendothelioma cell line (EOMA) as a 20 kDa  (Palmieri et al., 2000) propeptide Various fragments Chemotactic for dermal fibroblasts and peripheral monocytes (Albini and Adelmann-Grill, 1985;Malone et al., 1991) Col II Various fragments Inhibit collagen synthesis and MMP secretion by chondrocytes (Jennings et al., 2001) C-propeptide or chondrocalcin Cartilage calcification (Alini et al., 1992;Van der Rest et al., 1986) Col IV NC1 and 7S domains Inhibit morphogenesis in Hydra vulgaris (Sarras et al., 1993) NC1 domain Promote axonal growth (Lein et al., 1991) α1 NC1, arresten Anti-angiogenic (Colorado et al., 2000) α2 NC1, canstatin Anti-angiogenic (Kamphaus et al., 2000;Petitclerc et al., 2000) α3 NC1, tumstatin Anti-angiogenic (Maeshima et al., 2000;Petitclerc et al., 2000) α6 NC1 Anti-angiogenic (Petitclerc et al., 2000) Col VIII NC1 domain vastatin Angiogenic (Xu et al., 2001) Col XIV N-terminal fragment Chemotactic for neutrophils (Nakagawa et al., 1999) Col XV NC1 Anti-angiogenic  Restin Anti-angiogenic (Ramchandran et al., 1999;Sasaki et al., 2000) Col XVIII NC1 Induce migration of neural and non-neural cells in C. elegans ) Endostatin Anti-angiogenic (O'Reilly et al., 1997) Inhibit branching morphogenesis  fragment that binds to heparin (O'Reilly et al., 1997). In vitro studies with the same cell line have shown that the NC1 hinge domain contains cleavage sites for matrix metalloproteinases (MMPs) and cathepsin L. MMPs generate fragments of 30 kDa containing the endostatin domain, whereas cathepsin L directly and specifically releases the 20 kDa endostatin domain . Other in vitro studies confirmed these results and shown that generation of endostatin is also mediated by elastase after a first processing of NC1 by MMPs (Ferreras et al., 2000;Lin et al., 2001;Wen et al., 1999). In corneal epithelial cells, MMP-7 generates a 28 kDa NC1 (collagen XVIII) fragment (Lin et al., 2001). Thus, several distinct proteolytic pathways may be involved in the generation of endostatin in various tissues. Although we have some clues for in vitro processing, the physiological pathways still need to be more extensively investigated (Fig. 2B).

Biological functions of NC1 fragments and morphogenesis
Collagens IV, XV and XVIII, which are ubiquitously present in vascular and epithelial BM, are well conserved among different phyla. Interestingly, the biological functions of the NC1 domain also seem to be conserved throughout evolution.  and a globular C-terminal non-collagenous NC1 domain. The NC1 and central triplehelical domains are of an equivalent size, whereas the 7S domains are shorter in the case of α3, α4, α5 and α6 compared with α1 and α2. On the basis of sequence homology these different chains can be divided in two groups, the α1-like (α1, α3, α5) and the α2-like (α2, α4, α6). Assembly of collagen IV α chains. (B) The assembly of trimers is dependent first on the association of the NC1 domains, then the triple-helical structure forms and 7S domains are covalently associated. Four trimers interact through their 7S domains in a spider-shaped structure, and two trimers interact head to head through their NC1 domains, forming a sheet structure. Several trimers can also lace together along their triple-helical domain, thickening the structure (Timpl et al., 1981).
In the primitive invertebrate Hydra vulgaris, addition of NC1 (IV) alters morphogenesis, blocking cell aggregate development (Sarras et al., 1993;Zhang et al., 1994). In vitro, NC1 (IV) promotes axonal but not dendritic growth in sympathetic neurons from rat embryos (Lein et al., 1991), and hexameric NC1 supports attachment and migration of chicken neural crest cells. By contrast, intact dimers of collagen IV do not (Perris et al., 1993). These results imply that biological functions supported by NC1 domains are conformation dependent; proteolytic digestion of collagen IV may induce appearance of cryptic sites involved in new signals between cells and BMs. A collagen homologous to type XV-XVIII collagens has been identified in C. elegans. Deletion of the NC1-encoding region of this gene (cle-1) causes defects in axon guidance and migration of neural and non-neural cells. This phenotype can be rescued by ectopic expression of NC1, but not endostatin . These functional differences could be explained by NC1 being trimeric, whereas endostatin is monomeric. In analogous data, NC1 (XVIII) inhibits endothelial tube formation in Matrigel and stimulates cell motility of endothelial and non-endothelial cells. Monomeric endostatin has no effect by itself, but blocks the migration induced by NC1. The artificial oligomerization of endostatin stimulates cell motility , indicating that proteolytic processes may regulate endogenous NC1 functions. The α1 chains of collagen XV and XVIII are structurally homologous; they define a new collagen subfamily, the multiplexin family, on the basis of their central triple-helical domain with multiple long interruptions (green boxes). They are also characterized by a long non-collagenous N-terminaldomain-containing Thrombospondin sequence motif with two splicing variants in human collagen XVIII and a long non-collagenous globular Cterminal domain or NC1 domain. (B) Functional sub-domains of human NC1(XVIII) and protease cleavage sites. The NC1 domain contains three functionally different subdomains: these domains consist of a N-terminal non-covalent trimerization domain necessary for the association of trimers, a hinge domain containing multiple sites sensitive to different proteases and an endostatin globular domain covering a fragment of 20 kDa with anti-angiogenic and antibranching morphogenesis activities. Numerous enzymes can generate fragments containing endostatin. Cathepsin L and elastase are the most efficient, but in contrast to MMP cleavage leading to accumulation of endostatin, cathepsin L and B degrade the molecule [cleavage sites are indicated according to the data published by Ferreras (Ferreras et al., 2000)].
These effects were not observed with the NC1 (XV) and endostatin-like domain. This lack of activity may not be surprising, since these molecules may have different spectra of activity depending on the identity of the motility signals.
More recently, murine endostatin has been shown to inhibit HGF-induced migration and branching morphogenesis of renal epithelial cells and the ureteric bud. These processes are dependent on the presence of glypican3. Ureteric bud expresses endostatin, and addition of neutralizing anti-endostatin antibodies enhances ureteric bud outgrowth and branching .
Thus, high levels of endostatin or endostatin-like molecule may interfere with different pathways, downregulating morphogenetic processes. They may act as dominant-negative ligands that interact with the same receptors as their native molecule and, thus, inhibit proliferation and migration, or they may interact with different receptors and induce apoptosis. They may also have a mechanical effect in interacting with the original collagen trimers and disrupt them, leading to the loss of anchors between BMs and cells. Differential degradation of NC1 thus constitutes a negative feedback loop.

Biological functions of NC1 fragments in angiogenesis and tumorigenesis
Tumstatin and other collagen IV NC1 fragments Biological functions of collagen IV NC1 fragments have recently been extensively studied in mammals. A synthetic peptide encompassing residues 183-205 of α3 chain NC1 domain and containing an SNS triplet unique to α3 specifically inhibits activation of polymorphonuclear leukocytes (Monboisse et al., 1994). This peptide binds to a CD47/αvβ3 integrin complex, promotes adhesion and chemotaxis and inhibits proliferation of various human cancer cell lines (Han et al., 1997;Shahan et al., 1999a;Shahan et al., 1999b;Shahan et al., 2000). Indeed, direct interaction between α3NC1 and β3 integrin, independently of CD47, stimulates focal adhesion kinase (FAK) and PI 3-kinase phosphorylation (Pasco et al., 2000a). Furthermore, the inhibition of migration observed with melanoma and fibrosarcoma cells on native collagen IV or on the 185-205 α3NC1 peptide correlates with a decrease in expression of MT1-MMP and the β3 integrin subunit and a decrease in the level of activated membrane-bound MMP-2 (Pasco et al., 2000b). MMP-2 is involved in tumor progression and metastasis and its activation depends on MT1-MMP/TIMP-2 complexes Kinoshita et al., 1998). In vitro, α3NC1 fragment decreases the expression of MT1-MMP in a bronchial tumor cell line and the 185-205 α3NC1 peptide inhibits their invasion through [α1(IV)]2 α2(IV) collagen (Martinella-Catusse et al., 2001). Altogether, these data indicate that the ability of collagen IV to inhibit proliferation and regulate cellular adhesion and motility resides in the NC1 domain. The α3(IV) chain has a specific interaction with invasive cancer cells, and, in contrast to α1 and α2 chains, which favor migration, α3 limits the invasive phenotype. Thus, in the context of tumor progression and metastasis, the presence of either the α3(IV) collagen chain or the α3NC1 fragment may negatively regulate the invasion process. Interestingly, in the lung, where α1, α2, α3, α4 and α5 (IV) collagen chains are expressed in normal alveolar BM, development of bronchioalveolar carcinoma correlates with a loss of α3, α4 and α5 chain expression and an increase in α1 and α2 chain expression (Nakano et al., 2001). The heterotrimer [α1(IV)]2 α2 is permissive for the invasion of different cancer cell lines and mediates pro-MMP-2 activation (Maquoi et al., 2000). By contrast, α1NC1 or α2NC1 monomers have some structural similarities to TIMP-1 ¶ and inhibit MMP-2 and MMP-3 activity towards small synthetic peptide substrates in vitro (Netzer et al., 1998). Thus, the presence of different collagen chains or NC1 fragments during tumorigenesis might represent a key regulatory mechanism for the acquisition of an invasive phenotype.
Synthesis of collagen IV by vascular BM is a prerequisite for angiogenesis (Maragoudakis et al., 1993;Haralabopoulos et al., 1994). Moreover, α1 and α2 NC1 induce adhesion and spreading of endothelial cells (Koliakos et al., 1989;Tsilibary et al., 1990). On the basis of these observations and the data previously described, several groups have focused their attention on potential anti-angiogenic properties of NC1 fragments. Kalluri's group identified anti-angiogenic activities for α1, α2 and α3 NC1, named arresten, canstatin and tumstatin, respectively (Colorado et al., 2000;Kamphaus et al., 2000;Maeshima et al., 2000b). In vitro, these molecules inhibit endothelial cell proliferation and migration. Tumstatin seems to be the most efficient. None of the whole NC1 fragments inhibited the proliferation of cancer cell lines, as observed with the 185-205 α3NC1 peptide, which indicated that this effect is dependent on partial degradation of the NC1 domain. Arresten, canstatin and tumstatin molecules inhibit angiogenesis: in vitro, they block the formation of tubular structures by mouse aortic endothelial cells embedded in Matrigel; in vivo, they block the recruitment of capillaries in Matrigel plugs and inhibit the growth of large and small tumors in mouse xenograft models. Brooks' group generated similar NC1 fragments and described the anti-angiogenic effects of α2, α3 and α6 NC1 in chorioallantoic membrane (CAM) assays (Petitclerc et al., 2000). However, none of these NC1 fragments Journal of Cell Science 115 (22)  (α1 and α2 chains) Axonal growth of Hydra regeneration (Sarras et al., 1993) sympathetic neurons (Lein et al., 1991) Activation of leukocytes (a3)  VEGF-induced (O'Reilly et al., 1997;Yamaguchi et al., 1999;You et al., 1999;Taddei et al., 1999;Dhanabal et al., 1999a) Tumor regression (O'Reilly et al., 1997;Boehm Tumor angiogenesis (O'Reilly et al., 1997;Boehm et al., et al., 1997) 1997; Bergers et al., 1999;Blezinger et al., 1999;Boehle et al., 2001;Dhanabel et al., 1999b;Kisker et al., 2001;Sorensen et al., 2002;Yoon et al., 1999) Down-regulation of gene transcription (Shirichi Choroidal neovascularization Hirata, 2001;Hanai et al., 2002) G1 arrest of endothelial cells (Hanai et al., 2002) Arthritis (Matsuno et al., 2002;Yin et al., 2002) In vitro stabilization of microvessels (Ergun et Microvessel formation in rat aortic or human vein ring al., 2001) angiogenesis assay Ergun et al., 2001) Chorioallantoic membrane angiogenesis FGF2-induced  Migration of neural and non-neural cells in C. elegans  Migration and branching morphogenesis of renal epithelial cells  FAK phosphorylation, focal adhesion and actin stress fibers formation growth factor induced (Dixelius et al., 2002;Wickstrom et al., 2001) Endothelial and cancer cell invasion through Matrigel (Kim et al., 2000) Immobilized monomeric Endothelial cell spreading, FAK endostatin phosphorylation  NC1 trimer Cell motility of endothelial and non-endothelial cells  Endothelial tube formation in Matrigel  inhibited proliferation of cancer cell lines or endothelial cells in vitro. In contrast to the results shown for arresten, no inhibitory effects were observed for the α1NC1 fragment. Although identical mammalian expression systems were used for the production of the fragments in both cases, these discrepancies may be attributed to the use of different endothelial cell types and different tumorigenesis models (xenografts in mice versus development of tumor on chorioallantoic membranes).
What are the cell surface receptors involved in these functions? It is clear that integrins are key targets of NC1 (Table 2). Tumstatin binds αvβ3 integrin in an RGDindependent manner and interacts through two different sites, one site, composed of tumstatin residues 54-132, is involved in the anti-angiogenic effect, whereas the other, composed of residues 185-203, is involved in the anti-proliferative activity on cancer cell lines (Maeshima et al., 2000a;Maeshima et al., 2001b;Shahan et al., 1999b). Adhesion of endothelial cells (HUVEC or C-PAE) to tumstatin also seems to occur through α 6β1 integrin binding (Maeshima et al., 2000b). Although it is well known that the central triple-helical domain, as well as the NC1 domain of collagen type IV, interacts with cells via α1β1 and α2β1 integrins (Eble et al., 1993;Setty et al., 1998), these new results indicate that NC1 domains support novel integrinmediated cellular interactions involved in the regulation of angiogenesis. Interestingly, collagen IV also contains cryptic integrin-binding sites. During angiogenesis these sites are exposed and induce a switch in integrin recognition, with a loss of α1β1 binding and a gain of αvβ3 binding (Xu et al., 2001), which might be due to denaturation and concomitant degradation of collagen IV by MMPs such as MMP-2 (Eble et al., 1996).
In the case of tumstatin, a peptide composed of residues 45-132 of α3NC1 fragment is sufficient to inhibit in vitro and in vivo angiogenesis by increasing apoptosis of endothelial cells and is tenfold more active than endostatin. Because of the involvement of this fragment in Goodpasture syndrome, deletion of the Goodpasture epitope (residues 45-54) has been done, and the anti-angiogenic properties are preserved (Maeshima et al., 2000a). The effects of tumstatin are independent of disulfide bonds and are located in a 25-residue peptide (residues 74-98) (Maeshima et al., 2001a;Maeshima et al., 2001b). Apoptosis induced specifically in endothelial cells by this peptide is associated with inhibition of capdependent translation through negative regulation of mTOR signaling and depends on the presence of β 3 integrin (Maeshima et al., 2002).
Thus, NC1 domains of collagen type IV exhibit specific regulatory subdomains controlling adhesion, proliferation or apoptosis of various cells. The specificity of these subdomains for endothelial or cancer cells is very interesting, particularly in the case of α3NC1. Indeed, the recently published crystal structure of the collagen IV NC1 domain reveals a 3D structure with two homologous subdomains, N and C, with the major difference between these subdomains for each chain occurring in the region composed of residues 86-95 in the N subdomain and 196-209 in the C subdomain. Curiously these regions overlap two sequences identified previously as having antiangiogenic activity and cancer cell anti-proliferative effects, respectively. It will be interesting to identify more precisely the integrin-binding sites on these NC1 domains. Indeed, their interactions with integrins might be involved in the disruption of the contacts between endothelial cells or tumor cells and the basement membrane, leading to apoptosis of these cells. They might be also involved in the disruption of the C-terminal association that occurs during the assembly of a collagen IV network and induce disorganization of this matrix, thus disturbing migration, proliferation or survival of the cells. Moreover, the unique properties of the tumstatin NC1 domain in regulating neovascularization are very interesting and suggest a promising new family of integrin-dependent angiogenesis inhibitors.

NC1(XV), NC1(XVIII), endostatin and endostatin-like fragments
Since its discovery by O'Reilly and co-workers in 1997, endostatin has been the object of extensive research, trials and controversy in the angiogenesis field. In vitro, endostatin inhibits proliferation of bovine capillary endothelial cells but not cancer cells. In vivo, it inhibits angiogenesis on CAM assays and growth of various primary tumors. Moreover, no signs of toxicity, drug resistance or regrowth of tumors are observed as long as mice are treated (O'Reilly et al., 1997). Another striking effect is that endostatin when administrated on repeated cycles allowing the tumor to re-grow between each of them is still efficient and induces a prolonged dormancy of the tumor without resistance after two to six cycles, depending on the tumor model (Boehm et al., 1997). Thus far, multiple studies have demonstrated anti-angiogenic properties of endostatin in pathological models of tumorigenesis (Bergers et al., 1999;Blezinger et al., 1999;Boehle et al., 2001;Dhanabal et al., 1999a;Kisker et al., 2001;Sorensen et al., 2002;Yoon et al., 1999), choroidal neovascularization  and arthritis (Matsuno et al., 2002;Yin et al., 2002). However, in a controlled angiogenic process such as wound healing, endostatin does not affect the overall neovascularization. Ultrastructural analysis demonstrated some abnormalities in vessel maturation, but the blood vessel density is not affected (Bloch et al., 2000;Berger et al., 2000).
Tumorigenesis is not affected in mice lacking collagen XVIII, indicating that even if endostatin is detected as a circulating molecule, the physiological levels may not be sufficient to decrease tumor progression (Fukai et al., 2002). By contrast, collagen XVIII is required for normal regression of hyaloid vessels and for anchoring vitreal collagen fibrils to the retina inner limiting membrane. Thus, collagen XVIII may act as a gatekeeper, inducing regression of vessels in nonpermissive territory for angiogenesis. Although endostatin is efficient in reducing choroidal neovascularization, a role for endogenous endostatin has still to be fully demonstrated.
Ex-vivo, endostatin decreases and stabilizes microvessel formation in rat aortic or human vein ring angiogenesis assays Ergun et al., 2001). In vitro, endostatin inhibits basal and FGF2 or VEGF-induced proliferation and migration of different endothelial cell types (Dhanabal et al., 1999b;O'Reilly et al., 1997;Taddei et al., 1999;Yamaguchi et al., 1999;You et al., 1999). The circulating form purified from human plasma lacks 12 N-terminal residues and does not inhibit proliferation (Standker et al., 1997); so the antiproliferative activity of endostatin requires the full-length fragment but is independent of its zinc-or heparin-binding Roles for non-collagenous domains of collagens capacity (Yamaguchi et al., 1999). Soluble endostatin induces endothelial cell apoptosis by altering Bcl-2 expression (Dhanabal et al., 1999c) and inducing tyrosine phosphorylation of the protein adaptor Shb (Dixelius et al., 2000).
The molecular targets of endostatin are not yet clear. In endothelial cells growing exponentially, endostatin mimics serum deprivation in downregulating the transcription of genes involved in proliferation, apoptosis and cell migration. In the presence of serum, endostatin affects only migration of endothelial cells (Hanai et al., 2002;Shirichi and Hirata, 2001).
Inhibition of migration and survival are the most constant functions reported for endostatin. It is clear that migration of endothelial cells involves assembly and disassembly of focal adhesions in concert with integrin signaling, and endostatin seems to interfere effectively with both systems. Immobilized endostatin promotes and soluble endostatin inhibits α 5β1 and αvβ3 integrin-dependent endothelial cell migration and survival . Depending on the cell type and the growth factor environment, inhibition of migration with soluble endostatin correlates with an increase or a decrease in focal adhesion and actin stress fiber formation (Dixelius et al., 2002;Wickstrom et al., 2001). Inhibition of VEGF-induced migration induces eNOS dephosphorylation (Urbich et al., 2002). Furthermore, endostatin induces a downregulation of the urokinase plasminogen activator system (Wickstrom et al., 2001), indicating that the signals induced by endostatin not only modify the cytoskeletal architecture and survival signals, but also affect pericellular proteolytic activity.
Interestingly, as observed in the case of collagen IV NC1 fragments, some functions of endostatin may interfere with MMP signaling (for a review, see Egeblad and Werb, 2002). Indeed, endostatin has been reported to inhibit endothelial and cancer cell invasion through Matrigel. This effect seems to be mediated by its association with pro-MMP-2, which inhibits MMP-2 activation (Kim et al., 2000). Recent data have shown a direct interaction of endostatin with the catalytic domain of MMP-2 . MMP-2 is very inefficient in generating endostatin fragments from collagen XVIII (Ferreras et al., 2000), but it may be one of the endostatin key targets for downregulating expression or activation of other MMPs and proteases.
A direct interaction between endostatin and VEGFR2 has been described previously . This interaction may be due to the basic character of endostatin, similar to the interaction demonstrated between VEGFR2 and the transactivator protein Tat of HIV-1 (Albini et al., 1996). Thus endostatin may act essentially by interfering with the binding of VEGF to its receptors, VEGF-R2 as well as VEGF-R1**.
Molecular data are more limited for the related endostatinlike molecule. Inhibition of FGF2-induced migration, but not proliferation, of endothelial cells has been reported. In a model of renal cell carcinoma xenograft, a reduction of the tumor growth was observed but, in contrast with endostatin, endostatin-like did not induce any regression (Ramchandran et al., 1999). However, in CAM assays, only endostatin-like and NC1(XV) inhibited angiogenesis induced by VEGF, whereas angiogenesis induced by FGF2 was inhibited only by endostatin and NC1(XV) . Thus these fragments might have different inhibition properties depending on their angiogenic environment and motility signaling, as discussed in the previous paragraph. These data correlate with previous observations showing that VEGF and FGF2 mediate their effects through different integrins, αvβ5 and αvβ3, respectively (Friedlander et al., 1995).

Conclusions and perspectives
The detailed biological functions of these different NC1 collagen fragments are not completely identified, but the similarities within the pathways they affect are troubling ( Table  3). As immobilized substrates, NC1 fragments from collagen IV, XV or XVIII can induce migration, proliferation or survival depending on the cell type; however, soluble forms act in the opposite way, inhibiting proliferation, migration and inducing apoptosis (Fig. 3). This balance between negative and positive pathways may be part of a highly controlled regulatory mechanism. In the context of an invasive process such as angiogenesis or branching morphogenesis, the cells need to migrate and proliferate, and these processes are dependent on protease activity. In the first steps of these processes, partial degradation of the matrix may unmask some cryptic sites exposing NC1 immobilized domains that bind to cell surface integrins and activate different pathways depending on the integrin pattern. Sustained proteolysis of the extracellular matrix may release high local concentration of soluble fragments that then act as dominant-negative molecules. They can inhibit proliferation and migration, allowing stabilization of new vessels or tubular structures or induce complete regression of these new tubules through the activation of apoptotic pathways. Interestingly, the activities of these fragments are dependent on the activation of integrins or proteoglycans, such as glypicans in the case of endostatin. The integrin α vβ3 is a common ligand for collagen IV, XV and XVIII NC1 fragments and endostatin. In mice lacking β3 integrin, tumor angiogenesis as well as VEGF or hypoxia-induced angiogenesis are enhanced, suggesting a role for this integrin in limiting angiogenesis in vivo (Reynolds et al., 2002). The antiangiogenic activities of different NC1 fragments dependent on binding to β 3 integrin might also support this hypothesis. In that case, the NC1-fragment-β3-integrin interaction might be a key regulator of this negative feedback. Another feature in the biological activities of these collagen fragments is the downregulation of gene expression: the collagen IV α3NC1 domain as well as endostatin decreases the expression of various genes involved in cell cycle regulation, migration and survival. However these activities seem to be correlated with environmental factors, and the anti-angiogenic activities of various fragments may depend on the specificity and the concentrations of growth factors locally released.
Besides β3 integrin, another common target emerging for these fragments is MMP-2. A direct interaction with the catalytic domain has been shown in the case of endostatin. For endostatin-like and the collagen IV α3NC1 domain, such an interaction has not been demonstrated, but the three molecules induce a decrease in MMP-2 membrane-bound activity, which might be part of the negative feedback loop mentioned previously. A decrease in the basal level of pro-MMP-2 activity is also observed in collagen-XV knockout mice (Eklund et al., 2001). Moreover, several MMPs are involved in the generation of the fragments themselves. Thus, in a process such as angiogenesis, proteolytic activity of MMPs might be part of a biphasic regulation: proangiogenic in early steps, critical for the rupture of basement membrane and migration of the endothelial cells, and anti-angiogenic in late steps, generating endogenous inhibitor fragments. Endogenous inhibitors or activators derived from larger precursor proteins now appear to be a common theme in the context of remodeling processes. If endostatin or tumstatin have received increased attention recently because of their strong anti-angiogenic potential, it is clear that their activity is not specific for microvascular endothelial cells, but this work has opened new interest in the potential cryptic biological functions of these ubiquitous collagen molecules. Thus, these studies might be interesting not only in the context of cancer but also, considering the large number of diseases linked to collagens, in numerous other human disorders. This work was supported by grants from the National Institutes of Health (NS39278, CA72006, AR46238), funds from the Sandler Family Sustaining Foundation and a fellowship from the Osteogenesis Imperfecta Foundation. We would like to apologize for our failure to cite many of the important and relevant studies in this field because of space limitations. We are grateful to Bryony Wiseman for critical reading of the manuscript.