Synergy between Extracellular Modules of Vascular Endothelial Cadherin Promotes Homotypic Hexameric Interactions*

Vascular endothelial (VE) cadherin is an endothelial specific cadherin that plays a major role in remodeling and maturation of vascular vessels. Recently, we presented evidence that the extracellular part of VE cadherin, which consists of five homologous modules, associates as a Ca2+-dependent hexamer in solution (Legrand, P., Bibert, S., Jaquinod, M., Ebel, C., Hewat, E., Vincent, F., Vanbelle, C., Concord, E., Vernet, T., and Gulino, D. (2001)J. Biol. Chem. 276, 3581–3588). In an effort to identify which extracellular modules are involved in the elaboration and stability of this hexameric structure, we expressed various VE cadherin-derived fragments overlapping individual or multiple successive modules as soluble proteins, purified each to homogeneity, and tested their propensity to self-associate. Altogether, the results demonstrate that, as their length increases, VE cadherin recombinant fragments generate increasingly complex self-associating structures; although single module fragments do not oligomerize, some two or three module-containing fragments self-assemble as dimers, and four module-containing fragments associate as hexamers. Our results also suggest that, before elaborating a hexameric structure, molecules of VE cadherin self-assemble as intermediate dimers. A synergy between the extracellular modules of VE cadherin is thus required to build homotypic interactions. Placed in a cellular context, this particular self-association mode may reflect the distinctive biological requirements imposed on VE cadherin at adherens junctions in the vascular endothelium.

Adhesion between cells of identical phenotypes is mediated by receptors belonging to the cadherin superfamily (1,2). This protein family contains 50 different classic cadherins in both vertebrates and invertebrates and the recently described human pro-tocadherins (3,4). Based on their amino acid sequence, classic cadherins were first classified into two subgroups, type I and type II (5,6). More recently, cadherin-5 and -15 that share little sequence similarity with other cadherins or among themselves were considered as two distinct non-type I or II cadherins (7).
Classic cadherin molecules exhibit a similar organization, in particular in their extracellular domain, that consists of five homologous repeats designated EC1 to EC5 and numbered from the N to the C terminus. Type I members show a high degree of protein sequence similarity when compared with the E cadherin sequence and possess, in their N-terminal extracellular module EC1, the HAV cell adhesion recognition sequence (3). By contrast, the HAV sequence is absent on the extracellular domain of either type II members or non-type I or II cadherins (8,9).
The highly conserved cytoplasmic domain of cadherins interacts with ␤-catenin or ␥-catenin (also called plakoglobin) in a mutually exclusive fashion (10,11). Moreover, ␤or ␥-catenins also interact with ␣-catenin which links the cadherin-adhesion complex either directly or indirectly to the actin cytoskeleton (12,13). Cell-cell adhesion is constantly rearranged suggesting that the cadherin-catenin complex is dynamically remodeled.
By mediating homotypic interactions, cadherins are responsible for segregation of different cell types and, consequently, are fundamental for the establishment and maintenance of multicellular structures. Several results demonstrate that the homotypic binding regions reside in the extracellular part of cadherins (14). High resolution structure determination sheds light on the molecular determinants and organization of homotypic cadherin interactions at cell-cell junctions. Based on the first structure derived from the N-terminal domain of neural cadherin (N-EC1 fragment) (15), a model for cadherin-mediated homophilic interactions was proposed (the zipper model) (16). It suggests the formation of parallel dimer interfaces (cis dimers) and anti-parallel alignments (trans dimers). Both types of association may reflect interactions occurring between cadherins at the cell surface. Cis dimers involving the five extracellular modules of cadherins may mimic the alignment of two molecules emerging from the same cell, whereas trans dimers mediated by the N-terminal EC1 module may be elaborated by molecules protruding from adjacent cells. Structural data obtained for the two module fragments of E (E-EC1-2) (17,18) or N (N-EC1-2) (19) cadherin revealed that the length of a single cadherin extracellular module is ϳ45 Å. It can be deduced that the cell to cell distance in the zipper model is about 405 Å, a value incompatible with the dimension of cell-cell adherens junctions that ranges from 200 to 250 Å based on electron microscopy analysis.
By using direct-force measurements, Leckband and co-workers (20,21) proposed a new model for homotypic interactions in which cadherin molecules emerging from adjacent cells elaborate antiparallel, completely interdigitated contacts. These structures, which exhibit multiple adhesive contacts involving successive domains along the extracellular region of the protein, possess a length in the range of 250 Å.
Here we investigate the mechanism of homotypic interactions mediated by the non-type I or II cadherin 5 (or VE 1 cadherin). The distinct phylogenetic attribution of this cadherin is likely to be at the basis of its specialized function. Indeed, VE cadherin is expressed at the surface of a peculiar tissue, the vascular endothelium (9). This tissue is formed of a continuous monolayer of endothelial cells that constitute a physical barrier between blood and underlying tissues. The highly dynamic modulation of endothelial adherens junctions allows the passage of leukocytes from blood toward inflamed tissues (22). In fact, recently, we and others (23, 24) established a critical role for VE cadherin in the modulation of endothelial monolayer permeability and consequently in the control of leukocyte trafficking. Moreover, targeted inactivation of the VE cadherin gene in mice affects remodeling and maturation of endothelial vessels leading to the death of the embryos from severe vascular defects at mid-gestation (25,26). This demonstrates that VE cadherin is also required for vascular morphology. Furthermore, in contrast to other cadherin family mem-bers, VE cadherin is both connected to the actin cytoskeleton and to intermediate filaments (27,28).
Recently, we have presented evidence (29) for the Ca 2ϩ -dependent hexameric association of the extracellular region of VE cadherin. This type of oligomerization is difficult to reconcile with structural data obtained for shorter fragments of other cadherins. We have now extended our previous study by analyzing a series of VE cadherin-derived arrays of modules. We show that the degree of oligomerization is related to the number of modules within the array indicating that each module acts in synergy with the neighboring ones during the homotypic assembly of VE cadherin. Altogether, our results suggest that VE cadherin molecules first self-assemble as intermediate dimers involving extracellular modules EC3 and EC4 before elaborating mature hexameric structures.

MATERIALS AND METHODS
Recombinant Fragments-The single module fragments were named VE-ECi, with i corresponding to the position of the respective extracellular module within the extracellular domain of VE cadherin. The multidomain fragments were designated VE-ECi-j, where i and j correspond to the positions of the most N-and the most C-terminal modules, respectively. For instance, VE-EC1-3 starts at the N terminus of module EC1, ends at the C terminus of module EC3, and consequently overlaps the three modules EC1, EC2, and EC3 (Fig. 1A).
The oligonucleotide pairs used to produce the cDNA fragments encoding the VE cadherin-derived proteins are summarized in Table I. PCR-amplified products were cloned into different expression vectors. The single module fragments, cloned into the vector pGEX-4T1 (Amersham Biosciences), were expressed as glutathione S-transferase fusion proteins, whereas multimodule fragments, cloned into the vector pET30bϩ (Novagen), were expressed as native proteins. The resulting pET-VE cadherin plasmids allowed the expression of VE cadherin multiple module fragments fused to an N-terminal methionine. The coding sequence of all VE cadherin-derived constructs was verified by sequence analysis. The mutated fragment VE-EC1-4m was elaborated using the Quick-Change site-directed mutagenesis kit (Stratagene Europe, Amsterdam, The Netherlands). The mutations N98A and D99A were introduced using the synthetic oligonucleotide primers shown in Table I.
Purification of Recombinant Fragments-The single module fragments VE-EC1 and VE-EC4 were purified as GST fusion proteins directly from bacterial lysates using the affinity matrix glutathione-Sepharose 4B (Amersham Biosciences). Removal of the GST tail was achieved by cleavage with thrombin (2ϫ 1 unit/mg fusion protein, 2ϫ for 2 h at room temperature). Following complete digestion, GST was retained by an affinity chromatography step on glutathione-Sepharose 4B (Amersham Biosciences), whereas the VE cadherin fragments eluted in the flow-through. After concentration to 1 mg/ml, VE cadherinderived proteins were run on a gel filtration Superdex S200 column (Amersham Biosciences) to eliminate degradation products.
Determination of Protein Concentrations-Molar extinction coefficients for VE cadherin fragments were calculated based on their respective amino acid composition using the computer program Protparam Tools of the Expasy server (www.expasy.ch/tools/protparam.html).
Analytical Gel Filtration Chromatography-Multimers of VE cadherin fragments were fractionated at 4°C by analytical gel filtration on a Superdex S200 column (fractionation range 10,000 -500,000 Da, Amersham Biosciences). The hydrodynamic radii (Rh) corresponding to the oligomeric forms of the different VE cadherin fragments were deduced from chromatograms as described previously (29).
Two calibration curves connecting Rh to the molecular weights of the corresponding proteins were established using either standard globular proteins (30) or VE cadherin fragments. Molecular weight values, used to establish the VE cadherin curve, were calculated from sedimentation coefficients (see the experimental section under "Analytical Ultracentrifugation") or deduced from the degree of oligomerization determined using cross-linking experiments.
The equilibrium between the monomeric (M) and dimeric (D) forms of both VE-EC3-4 and VE-EC2-4 fragments can be described as Trypsin Digestions-Prior to digestion, the fragments were equilibrated in 5 mM Ca 2ϩ , and their concentrations were adjusted so that they remained oligomeric. Limited proteolysis of the VE cadherin-derived fragments was performed with trypsin at room temperature for 20 min and then blocked using 2 mM phenylmethylsulfonyl fluoride (ICN, Biomedical Inc, Aurora, OH). To establish a comparison between trypsin sensitivity, the various VE cadherin-derived fragments were simultaneously digested using identical standard conditions.
The mixtures of peptides were separated by SDS-PAGE, and the gels were stained by Coomassie Blue. The different protein fragments generated by trypsin digestion were individually excised from polyacrylamide gels (lanes 5, Fig. 2A) and subjected to in-gel total proteolysis with trypsin. Practically, excised bands were washed with 50 mM ammonium bicarbonate, destained with acetonitrile (diluted 50/50 in 50 mM ammonium bicarbonate), and dried in a Speedvac evaporator. The gel pieces were re-swollen in 20 l of 50 mM acetate buffer, pH 7.2, containing 1 g of trypsin. Following a 2-h digestion at 37°C, the peptides were extracted with 20 l of 60% acetonitrile containing 1% trifluoroacetic acid. The combined extracts were desalted on Poros R2 resin. One microliter of the eluate was applied on the dried matrix spot and analyzed as described (31). The molecular masses of the resulting peptides were determined using MALDI mass spectrometry that allowed assignment of the cleavage sites.

Matrix-assisted Laser Desorption-Mass Spectrometry (MALDI)-
Mass spectra of the recombinant fragments, cross-linked or not, were determined as described previously (29).
Cross-linking of Recombinant Fragments-The recombinant fragments were cross-linked using N-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC from Pierce). EDC was chosen because the coupling reaction is restricted to carboxyl and amino groups separated by only 6 Å, thus preventing unwanted cross-linking between two oligomers.
To maintain the oligomeric states, concentrations of the fragments and calcium were adjusted to 50 M and 5 mM, respectively. Various cross-linking assays were systematically performed with molar ratios between the recombinant fragments and the cross-linker reagent of 37.5, 75, 150, 300, and 600. These experiments were carried out for 2 h at 20°C in MES buffer, pH 7.0. The cross-linking reactions were terminated by adding 1 M Tris, pH 8.0. Analysis of cross-linked products was then performed on 4 -15% gradient Phast gels (Amersham Biosciences).
Analytical Ultracentrifugation-Sedimentation velocity experiments were performed using a Beckman model XL-A analytical centrifuge as described previously (29). By combining the sedimentation coefficients (s) and the Rh of VE-EC1-3, determined by ultracentrifugation and gel filtration chromatography experiments, the molar mass (M) of this fragment could be calculated using the following equation: The partial specific volumes (v ) for VE-EC1-3 (0.718 ml/g), the density () (1.0059 g/ml), and the viscosity () (1.022 centipoise) were calculated according to the program Sedinterp supplied by D. B. Hayes, T. Laue, and J. Philo (software available at www.bbri.harward.edu/rasmb/rasmb.html). Sedimentation velocity experiments were performed at 42,000 and 60,000 rpm.

Recombinant VE Cadherin
Fragments-To evaluate the relative contribution of each extracellular module of VE cadherin to homotypic binding, several recombinant fragments, encompassing variable parts of the extracellular region of human VE cadherin, were expressed in Escherichia coli. The boundary of these fragments was defined according to the cadherin domain organization proposed by Tanihara et al. (6). To improve the stability of some of them, the boundaries of the fragments were also delineated by means of limited proteolysis experiments. Individual modules or arrays of modules are depicted in Fig.  1A.
The single module fragments VE-EC1 and VE-EC4 were expressed as N-terminal glutathione S-transferase (GST) fusion proteins and purified by affinity chromatography using glutathione-Sepharose prior to proteolytic cleavage of the GST tag. The multimodule fragments were purified from inclusion bodies and refolded by dilution as described previously (29).
The purity of the VE cadherin fragments was verified by SDS-PAGE analysis (Fig. 1B). N-terminal sequencing revealed that cleavage by thrombin left the exogenous sequence GS at the N termini of fragment VE-EC1 (Table II). Concerning the multimodule fragments, N-terminal sequences corresponded to the expected ones, except for those of the inner fragments VE-EC3-4 and VE-EC2-4 for which the N-terminal methionine was cleaved off (Table II). MALDI mass spectrometry analysis confirmed the identity of the VE cadherin fragments (Table II).
Oligomeric States of the VE Cadherin Recombinant Fragments-The capacity of the fragments to self-associate was analyzed by gel filtration chromatography, chemical cross-linking, and analytical ultracentrifugation experiments.
As illustrated in Fig. 2, the chromatographic profiles differ according to the nature of the fragments. Thus, fragments VE-EC1, VE-EC4, VE-EC1-2 (not shown), and VE-EC1-3 gave single elution peaks for fragment concentrations tested up to 200 M. In contrast, double distributions (peaks I and II) were observed for the fragments VE-EC3-4, VE-EC2-4, and VE-EC1-4, even at the relatively low concentration of 10 M, indicating that they possess two different oligomeric states.
From the elution volumes of the peaks observed on the chromatograms, the hydrodynamic radii of the different species of each fragment were deduced (Table III and "Materials and Methods").
To determine the oligomeric states of the fragments, they were first cross-linked using the heterobifunctional reagent EDC that covalently couples primary amino to carboxyl groups located in close proximity. Fig. 3 shows the electrophoretic separation of the cross-linked products.
Cross-linked fragments VE-EC3-4 and VE-EC2-4 exhibited a two band pattern. The upper bands have molecular masses of ϳ50 and 72 kDa for the cross-linked fragments VE-EC3-4 and VE-EC2-4, respectively, indicating that both fragments form dimers. By using similar cross-linking experimental conditions, a pattern of six bands was obtained with VE-EC1-4 (29).
Altogether, these data confirm our gel filtration chromatography results from which it can be deduced that, in Fig. 2, peaks I and II observed with both VE-EC3-4 and VE-EC2-4 correspond to the dimeric and monomeric forms of these fragments, respectively, and that peak I of the fragment VE-EC1-4 represents its hexameric association.
In contrast, under similar cross-linking conditions, fragments VE-EC1-2 and VE-EC1-3 gave single bands of 24 and 36 kDa, respectively, confirming the absence of oligomerization seen by gel filtration experiments for these fragments. Similar results were obtained for fragments VE-EC1 and VE-EC4 (results not shown). These observations were confirmed when high concentrations of fragment VE-EC1-3 were analyzed by analytical ultracentrifugation. Indeed, sedimentation velocity experiments performed using VE-EC1-3 at concentrations ranging from 1 to 108 M showed a single species (Fig. 4). The sedimentation coefficient remained constant as the concentration of the fragment increased and was measured to 2.2 Svedberg. Combined with the hydrodynamic radii determined by gel filtration chromatography (Table III), these values allowed the determination of the molar mass of the fragment VE-EC1-3 (36 kDa) and confirmed that this fragment does not self-associate even at high protein concentrations.
We have established a standard curve connecting the Rh to the corresponding molecular masses (M r ) of the VE cadherin fragments. The hydrodynamic radii were determined from gel filtration chromatography as mentioned above (Table III), whereas the molecular weights were calculated from the degree of self-association deduced from either cross-linking or analytical centrifugation experiments. A linear relationship connecting log M r to Rh was observed for the different oligomeric states of the various VE cadherin fragments (Fig. 5). All the points for VE cadherin fragments are down-shifted when compared with the curve established using globular proteins reflecting a difference in the molecular shape between globular proteins and VE cadherin fragments. This can be explained by the elongated form adopted by VE cadherin fragments as observed for VE-EC1-4 in cryoelectron microscopy (29). These data confirm, for each fragment, the multimerization results obtained either by gel filtration chromatography or cross-linking or ultracentrifugation experiments.
Equilibria between the Monomeric and the Multimeric Forms-We have demonstrated recently (29) the existence of an equilibrium between the monomeric and the hexameric forms of fragment VE-EC1-4. Here this study was extended to fragments VE-EC3-4 and VE-EC2-4 that self-associate as dimers. Indeed, gel filtration chromatography experiments showed that the intensities of peak I increased with concomitant decrease of those of peak II for increasing concentrations of the fragments (Fig. 6A). This result demonstrates the existence of an equilibrium between the monomeric and the dimeric species of both fragments VE-EC3-4 and VE-EC2-4. For comparison, results concerning the equilibrium between the monomeric and hexameric forms of fragment VE-EC1-4 were also included in Fig. 6A.
To study the parameters governing the monomer-oligomer equilibria, time courses of the multimer dissociation, established by diluting the fragments from 150 to 20 M, were followed by comparing the areas of peaks I and II obtained by gel filtration chromatography. The time required to reach the equilibrium at 4°C was estimated to 24 h for the fragments VE-EC3-4 and VE-EC2-4 and to be longer than 2 weeks for VE-EC1-4 (results not shown). Thus, during the chromatographic runs, which lasted 30 min, the dissociation of multimers was negligible (results not shown). Consequently, the dissociation constants K D calculated from the ratios [dimer]/ [monomer] 2 can be directly evaluated from the areas of peaks I and II. The K D values are thus estimated to 80 and 25 M for the fragments VE-EC3-4 and VE-EC2-4, respectively.
To establish a comparison with the fragment VE-EC1-4, which self-associates as a hexamer, the concentrations C 50% for which 50% of the fragments are multimeric are estimated using the following equation: C 50% ϭ (2/n 1/nϪ1 ) (K D ) 1/nϪ1 , where n represents the oligomeric state of the fragment. For the fragments that self-associate as dimers, C 50% ϭ K D and for the fragment VE-EC1-4, C 50% ϭ ⅐(2/6 1/5 )(K D ) 1/5 , with K D being expressed in this case in M Ϫ 5. As indicated in Table III, the C 50% values are calculated to be 80, 25, and 0.5 M for the fragments VE-EC3-4, VE-EC2-4, and VE-EC1-4, respectively. Clearly, as the length of the fragments increases, the affinity for self-association increases. This reflects a synergy existing between the extracellular modules to elaborate VE cadherin homotypic interactions.
As illustrated in Fig. 6B, oligomerization of the fragments was demonstrated, by gel filtration chromatography experiments, to be Ca 2ϩ -dependent. Indeed, upon elution of fragments, previously equilibrated in the presence of Ca 2ϩ , using an EDTA-containing buffer, the multimers of fragments VE-  Folding of VE Cadherin Fragments-The varying capacity of the recombinant VE cadherin fragments to self-associate may reflect differences in their efficiency to adopt a correctly folded conformation. Consequently, their stability was studied by limited trypsin digestion. Proteolytic experiments were performed at concentrations at which recombinant VE cadherin fragments were mainly oligomeric. Fig. 7A shows the electrophoretic separation of the digestion products. The single module fragment having its proximal inter-module region VE-EC1 (Asp 1 -Arg 107 ) appeared resistant to trypsin (Fig. 7A). In contrast, the short single domain fragments VE-EC1 (Asp 1 -Phe 104 ) and VE-EC4 (Gln 326 -Phe 432 ), which did not possess inter-module extensions, were almost completely digested with very low amounts of trypsin (data not shown). Moreover, digestion patterns for the various multimodule fragments exhibited a limited number of bands. It could be deduced that most of the potential cleavage sites were not accessible to the protease indicating that the fragments of VE cadherin are folded (Fig. 7B).
Cleavage sites were accurately localized using an approach that combines limited enzymatic proteolysis, in-gel proteolytic digestion, and MALDI mass spectrometry (33,34). Results are given in Tables II and IV. Following limited trypsinolysis, VE-EC1 (Asp 1 -Arg 107 ), extended by the EC1-EC2 inter-module region exhibited two distinct proteolytic bands (band b, Ile 24 -Arg 107 ; band c, Tyr 35 -Arg 107 ) (Fig. 7A). This means that, among the 13 putative   The s values were calculated from sedimentation profiles obtained at 20°C after 120 min using the computer program Svedberg (50). The path length was 1.2 cm for the 1 M sample and 0.3 cm for the other samples. The slight decrease of the sedimentation coefficient is due to a nonideality term observed in such experiments as described (14).

FIG. 5. Relationship between the molecular masses (Mw) and
the Rh of different VE cadherin fragments. K av values, estimated from the elution volumes, allowed the determination of the Rh corresponding to the different oligomeric states of the various fragments ("Materials and Methods"). The Rh values were correlated to the molecular weights of the oligomeric species determined either by analytical centrifugation or by MALDI mass spectrometry after fragment cross-linking. For comparison, a standard curve established with globular proteins (ribo, ribonuclease; chymo, chymotrypsin; oval, ovalbumin; alb, albumin; aldo, aldolase; cata, catalase; ferri, ferritin) is drawn using the same protocol (----). The oligomeric state of each fragment is mentioned as follows: x1, monomer; x2, dimer; x6, hexamer.
trypsin-susceptible sites, only two (Lys 23 and Lys 34 ) remain effectively accessible to the enzyme.
For VE-EC1-2, one cleavage site, localized in the EC1-EC2 linker region at position Arg 107 , yielded two bands of 15 and 12 kDa corresponding to fragments Leu 108 -Arg 244 and Asp 1 -Arg 107 , respectively (bands d and e, Fig. 7, A and B, and Table  IV). In addition, two N-terminal products with molecular masses of 25 and 23 kDa resulting from cleavage sites at residues Lys 23 and Lys 34 (bands b and c, Fig. 7, A and B, and Table IV) were also detected. Similarly, N-terminal truncations of fragment Asp 1 -Arg 107 at positions Lys 23 and Lys 34 generated the 10-and 9-kDa bands (bands f and g, Fig. 7, A and B, and Table IV Table III. B, cleavage sites generated by trypsin digestion. Trypsin cleavage positions were deduced from the molecular weights of in-gel-digested peptides. Small black tic marks represent theoretical trypsin cleavage sites, and highly accessible cleavage sites are indicated by black arrows, and moderately accessible cleavage sites are represented by open arrowheads. cleavage site located at position Arg 183 (band g). Nevertheless, compared with fragment Asp 1 -Asp 203 (results not shown), fragment Asp 1 -Asp 244 exhibits a higher trypsin resistance indicating that addition of the EC2-EC3 inter-domain improves the stability of the fragment VE-EC1-2.
Limited proteolysis of the dimeric VE-EC3-4 fragment revealed cleavage sites at Arg 300 and Arg 252 , yielding four distinct bands migrating at 21, 15, 10, and 6 kDa (bands b-e, Fig.  7A, and Table IV). Even at the highest trypsin concentrations used, the intensity of band c remained unchanged indicating that the amino acid stretch Tyr 301 -Glu 431 is trypsin-resistant. Consequently, within VE-EC3-4, module EC4 is probably correctly folded.
Moreover, when compared with VE-EC3-4, VE-EC2-4 exhibited two additional highly accessible cleavage sites at positions Arg 107 and Arg 183 within module EC2 (Fig. 7, A and B, and Table IV).
Limited proteolysis of the monomeric VE-EC1-3 fragment generated a major 27-kDa band corresponding to peptide Asp 1 - Arg 244 (band b, Fig. 7A, and Table IV). In addition, multiple minor fragments with molecular masses between 33 and 30 kDa (bands c-f, Fig. 7A) appear for a trypsin/VE cadherin fragment ratio of 1.3 units/mol, indicating secondary cleavage events at residues Lys 23 , Lys 34 , and Arg 300 (lane 6, Fig. 7B, and Table IV).
Concerning VE-EC1-4, limited proteolysis, performed using enzyme to fragment ratios ranging from 0.162 to 0.650 units/ mol, yielded two bands of 35 (band b) and 14 kDa (band c), respectively (Fig. 7A, lanes 3-5). With a larger amount of trypsin (enzyme/fragment ratio of 1.3 units/mol), an additional band of 22 kDa (band d) appeared (Fig. 7A, lane 6). The 35-and 14-kDa bands were identified by mass spectrometry to be the N-terminal Asp 1 -Arg 309 part and the C-terminal Arg 309 -Glu 431 part, respectively ( Fig. 7B and Table IV). Cleavage of Asp 1 -Arg 309 at position Arg 107 generated the secondary 22-kDa product corresponding to the sequence Leu 108 -Arg 309 (band d, Fig. 7B, and Table IV). Fig. 7B summarizes the theoretical and experimental trypsin cleavage sites for the different fragments. It can be concluded that, in general, neighboring inter-module regions seem to mutually limit access to potential trypsin cleavage sites for each domain, probably by interactively favoring correct folding or by masking sites due to homotypic association.
Ca 2ϩ -binding Site Mutations at the EC1-EC2 Linker Region Impair Hexamer Formation-The tridimensional structure of N and E cadherin-derived EC1-EC2 fragments revealed that Ca 2ϩ promotes dimerization of cadherins (17,19). It is therefore possible that formation of a dimer represents a step in the VE cadherin assembly pathway. We have investigated the role of Ca 2ϩ -binding sites of VE cadherin by introducing mutations within the EC1-EC2 inter-region domain.
As indicated by amino acid sequence alignment, most of the amino acids involved in Ca 2ϩ binding (17) are conserved among E and VE cadherins and particularly the VE cadherin amino acids Asn 98 and Asp 99 (Fig. 8A). Consequently, two mutations, N98A and D99A, were simultaneously introduced within the Ca 2ϩ -binding region of the VE cadherin EC1-EC2 inter-domain to generate a new fragment designated as VE-EC1-4m (Fig.  8A).
The capacity of the fragment VE-EC1-4m to self-associate was analyzed by gel filtration chromatography (Fig. 8B). Chromatographic profiles exhibited one major (peak III) and two minor peaks (peaks I and II) indicating that the fragment possesses three different oligomeric states. By comparing the chromatographic profiles of the wild-type to those of the mutated VE-EC1-4m fragment, it could be deduced that the minor peaks I and II correspond to the hexameric and monomeric states, respectively. The major peak corresponds to a species intermediate between a monomer and a hexamer. The hydrodynamic radius of this intermediate species, evaluated from its elution volume to 50.5 Å, allowed the estimation of its molecular mass to be 95 kDa using the standard curve from Fig. 5. This result demonstrated that the intermediate species (peak III) is a dimer (theoretical molecular mass of 98 kDa). We can conclude that the ability of the mutant protein to form a hexamer is greatly impaired by introducing mutations within the EC1-EC2 interdomain. Indeed, at 0.5 M, 50% of the wild-type fragment VE-EC1-4 is hexameric, whereas for the mutated fragment at 22 M, the dimer is the preponderant species. Altogether, the results are consistent with the existence of a dimeric intermediate appearing when the fragment VE-EC1-4 is deprived of interactions involving Ca 2ϩ -binding sites between modules EC1 and EC2. DISCUSSION Suzuki and co-workers (6) classified VE cadherin as a class II cadherin because the tripeptide HAV, involved in cellular adhesion of classical group I cadherins, is missing within its N-terminal module EC1. Based on low sequence similarity scores, Shimoyama et al. (7) recently attributed VE cadherin to a separate phylogenetic class. To analyze whether this classification can be justified by a specific self-association mechanism, possibly different from that of the classical group I and II cadherins, various recombinant fragments encompassing one or several extracellular modules of VE cadherin were recombinantly produced, and their propensity to self-associate in solution was analyzed.
Limited proteolysis experiments show that single module fragments devoid of their adjacent inter-module regions are unstable most likely because they lack Ca 2ϩ -binding sites that are known to rigidify cadherin molecules (17). Extension by adding inter-module regions at the C or N terminus stabilizes the single module fragments. This correlates with the fact that all multidomain fragments are stable. Although they possess well conserved overall structures, some subtle structural differences are detected around positions Lys 23 , Lys 34 , Arg 244 , Arg 252 , Arg 300 , and Arg 309 that are digested within some but not all fragments. This may reflect differences in the capacity of the fragments to self-assemble or in the relative orientation of modules as their number increases.
Fragment VE-EC1-4 appears to be the most stable among the VE cadherin fragment series. Cleavage sites at positions Arg 244 /Arg 252 and Lys 23 /Lys 34 observed on shorter fragments were not accessible to trypsin on VE-EC1-4, probably because they were masked by its particular multimerization capacity (see below) (29). VE cadherin single module fragment EC1 is not able to self-associate. This is in good agreement with results obtained for the N-terminal module EC1 of E cadherin, which also remains monomeric in solution (14,35). By contrast, the single N-terminal fragment of N cadherin self-associates as dimers as shown by x-ray crystallography (15). However, the intermolecular contacts seen in this single domain structure might be artificially induced by crystal packing (36).
Whereas the single module fragment VE-EC1 is not able to self-associate, some of the fragments consisting of two (VE-EC3-4) or three modules (VE-EC2-4) form dimers. Moreover, the fragment overlapping the four N-terminal extracellular modules (VE-EC1-4) was recently demonstrated to associate as a hexamer in solution (29). Thus, a synergy between the extracellular modules of VE cadherin is necessary to build homotypic interactions. This cooperative effect is confirmed by the C 50% values for which 50% of the fragments are present under their respective oligomeric form. These C 50% values are 0.5, 25, and 80 M for the hexameric and dimeric self-associations of the fragments VE-EC1-4, VE-EC2-4, and VE-EC3-4, respectively. It can be deduced that, as the length of the fragment increases, the affinity of self-association increases concomitantly. VE cadherin homotypic interactions therefore require multiple inter-module interactions to elaborate tight cell-cell adhesion. These results are in agreement with those presented in the paper of Chappuis-Flament et al. (37) on C cadherin who demonstrated that homophilic binding and adhesion are mediated by multiple cadherin extracellular repeats.
VE-EC1-2 and VE-EC1-3 do not self-associate. Specific regions necessary for homotypic association of VE cadherin are possibly missing on these fragments. In fact, only fragments containing the EC3-EC4 tandem module such as VE-EC3-4, VE-EC2-4, and VE-EC1-4 are able to self-assemble. We therefore propose that module EC4 and/or the hinge region between modules EC3 and EC4 are required for elaborating VE cadherin self-associating interactions. In fact, the hinge region between modules EC3 and EC4 may constitute a new selfassociating interface indispensable for VE cadherin multimerization. This is in agreement with the results of Corada et al. (38) who mapped the epitope of a monoclonal anti-VE cadherin antibody able to increase endothelial permeability between amino acids 296 and 301, therefore possibly interfering with the self-associating process described here.
Whereas VE-EC2-4 is able to form dimers, VE-EC1-4 selfassociates as a hexamer. This underlines the particular role of module EC1 in homotypic VE cadherin association, thus confirming results obtained on other cadherin members in particular for E (17) and N cadherins (15,19).
In contrast to VE cadherin, E (14,17) and N cadherins (19) dimerize via their EC1-EC2 inter-module region. X-ray structure determination demonstrated that within these dimers, cadherin molecules are not only connected by their EC1-EC2 linker region but also bridged by an arrangement of three calcium ions that are associated with modules EC1 and EC2. Cadherin sequence alignments predicted that calcium is similarly bound to the successive extracellular EC3 and EC4 modules of VE cadherin and thus might play a similar stabilizing role in VE cadherin multimers. In fact, formation of VE cadherin multimers is shown to be Ca 2ϩ -dependent because, in the presence of EGTA, the VE-EC1-4 hexamer and both the VE- Altogether, the results presented here concerning the selfassociating behavior of VE cadherin are different from those published previously on type I E (14) and N cadherins (19). For instance, fragments EC1-2 of E and N cadherins form dimers, whereas the VE cadherin equivalent fragment does not selfassociate. In contrast, the VE-EC3-4 fragment forms dimers with a C 50% of 80 M comparable with that of E-EC1-2 suggesting that modules EC3 and EC4 are the basic interacting modules of VE cadherin. This result is confirmed by the fact that the mutated fragment VE-EC1-4m, whose EC1-EC2 hinge region was destabilized by introducing mutations within its Ca 2ϩ -binding domain, forms mainly dimers. This reveals the existence of a transient dimeric intermediate involving interactions between modules EC3 and EC4. Once formed, these dimers can elaborate hexameric structures by establishing new inter-module interactions involving the EC1-EC2 interdomain Ca 2ϩ -binding sites of VE cadherin.
Cryoelectron microscopy images obtained with VE cadherin EC1-4 allowed us to determine the length of the hexameric structure to be 233 Ϯ 10 Å (Fig. 9C) (29), a value compatible with that determined by others (32) for VE cadherin. This value also correlates with the 200 -250-Å intercellular gap distance measured on electron micrographs of endothelial cell adherens junctions (39). The hexameric structure possesses a length equivalent to the length of five cadherin modules and therefore probably results from inter-digitation of six molecules of VE cadherin as proposed in Fig. 9, D and E.
Our results cannot distinguish between cis or trans interactions within the hexamer. Nevertheless, as illustrated in Fig. 9, due to symmetry only two orientations of VE cadherin molecules within the hexamer are possible, either the six molecules adopt a parallel orientation (A and B) or they are alternatively arranged in an anti-parallel manner (D and E). Constitution of parallel hexamers would need, as a prerequisite, the formation of parallel intermediate dimers. Consequently, parallel hexamers would have to assemble into anti-parallel multimers to constitute a cell-cell adhesive interface. The lack of any detectable multimers of hexamers for the fragment VE-EC1-5 seems to exclude this possibility. Therefore, we favor models Fig. 9, D and E, where VE cadherin molecules within the hexamer form trans interactions that lead directly to cell-cell binding.
The model for trans-binding proposed in Fig. 9E depicts six molecules of VE cadherin on apposing cells aligned anti-parallel so that extracellular modules EC1 are paired with EC5, EC2 with EC4, and EC3 with EC3. This face-on view of the hexameric VE cadherin structure is simplified for reasons of clarity and does not reflect the overall complexity of the interdigitation of the extracellular modules. Understanding the detailed interface will be possible when the tridimensional structure of the hexamer is solved. Moreover, VE cadherin homotypic binding is influenced by polysialic acids (40) suggesting that the overall mechanism is more complex and may involve additional interactions. Our interaction model is compatible with a model proposed by Sivasankar and co-workers (20, 21) who demonstrated by direct-force measurements that the ectodomain of C cadherin exhibits multiple adhesive contacts involving successive domains along the extracellular region of the protein. The proposed anti-parallel alignment of the six molecules within the hexamer favors trans interactions, a prerequisite for cadherin-mediated cell-cell adhesion (17,18,(41)(42)(43)(44).
Formation of VE cadherin hexameric structures described here may correspond to an early clustering event occurring during the formation of endothelial cell-cell contacts. In fact, these hexamers that probably correspond to the cylindershaped particles observed on freeze-etch images of adherens junctions may first act as discrete units (45)(46)(47). Binding of catenins to the cytoplasmic domain of VE cadherin molecules may result in indirect connection of these hexameric structures to each other by anchoring them to the actin and/or intermediate filament cytoskeletons. By increasing the local concentration of VE cadherin in the plane of the cellular membrane, these clusters first increase cell adhesion as demonstrated previously for C cadherin (48) and, second, may promote the generation of different intracellular signals as observed for N cadherin (49).