Self-assembly of the Vascular Endothelial Cadherin Ectodomain in a Ca 2 1 -dependent Hexameric Structure* h S

contrast

The endothelium is a semipermeable barrier that controls the passage of solutes and circulating cells between the bloodstream and the neighboring tissues. The junctions between endothelial cells play a crucial role in controlling this traffic. An adhesion receptor belonging to the cadherin superfamily, vascular endothelial cadherin (VE-cadherin) 1 (1) is specifically localized at these cell-cell junctions.
VE-cadherin is involved in the maintenance and restoration of endothelium integrity (1)(2)(3)(4). In fact, opening of VE-cadherinmediated endothelial cell contacts was suggested to be a crucial step in neutrophil extravasation (5)(6)(7). VE-cadherin is also involved in determining the vascular architecture since recent studies demonstrated that targeted inactivation of VE-cadherin in transgenic mice leads to embryonic mortality due to severe vasculogenic defects (8,9).
Cadherins are Ca 2ϩ -dependent cell-cell adhesion receptors that are able to bind cells together by means of homotypic interactions. In these interactions, only identical cadherin types interact. Consequently, cells expressing various cadherins segregate into like groups when mixed. Due to this selective cell-cell recognition property, cadherins are important regulators of morphogenesis (10) by promoting clustering of cells with identical phenotypes.
The extracellular part of cadherins consists of five homologous protein modules designated EC1-EC5, numbered from the N terminus to the C terminus. Various lines of evidence have implicated the N-terminal module EC1 in determining homotypic binding specificity (11,12). Fixation of Ca 2ϩ on sites generally located between two consecutive extracellular modules (13,14) reduces cadherin sensibility to proteolytic degradation, rigidifies the elongated structure of cadherins (15,16), and is required for homotypic interactions (17). The cytoplasmic domains of cadherins are homologous and interact either with ␤or ␥-catenins in a mutually exclusive fashion. By connecting cadherins to the actin-based cytoskeleton, binding of ␣-catenin to ␤or ␥-catenins (18) strengthens cadherin-mediated cell adhesion.
Comparison of cadherin sequences provides evidence that VE-cadherin is structurally different from class I cadherins, which include N-and E-cadherins. For example, the EC1 module of VE-cadherin lacks the sequence motif HAV (1) involved in the adhesion activity of class I cadherins (11,12). Consequently, VE-cadherin was classified in the class II cadherin group (19), inferring that it might have particular adhesive properties. The elucidation of how the extracellular region of VE-cadherin assembles is fundamental to understanding the mechanism of cell-cell adhesion in the endothelium and to clarifying the process by which VE-cadherin modulates the transmigration of leukocytes. In this study, we focus on the behavior in solution of VE-EC1-4, a fragment consisting of the four N-terminal modules of VE-cadherin.
The molecular determinants at the basis of homotypic interactions and organization of cadherins in the adherent struc-tures at cell-cell junctions have been studied by analyzing the behavior in solution of cadherin fragments and by determining their high resolution structures. Some of the results are conflicting. For example, the E-cadherin N-terminal module (E-EC1) remains monomeric in solution (20,21), whereas the equivalent fragment of N-cadherin (N-EC1) self-associates as dimers (14), suggesting that the homoassociation mechanism is different according to the cadherin type. Comparison of the crystal structures from various cadherin fragments is also ambiguous. Thus, the N-EC1 crystal structure revealed the presence of parallel dimer interfaces (cis-dimers) and antiparallel alignments (trans-dimers) (14). It was suggested that both types of association reflect the interactions occurring between cadherins at the cell surface. cis-Dimers were proposed to mimic the alignment of two molecules emerging from the same cell surface, whereas trans-dimers may correspond to molecules protruding from opposing cell surfaces. The presence of these parallel and antiparallel dimers inspired Shapiro et al. to build a zipper model for the molecular organization of cadherins at the cell surface, in which each extracellular module contributes to lateral dimerization (14,22). Until now, no direct evidence for lateral dimerization of the EC2-EC5 modules has been presented.
This model is in conflict with conclusions drawn from the crystal structures of the two module fragments of E-cadherin (E-EC1-2) (23,24) and N-cadherin (N-EC1-2) (25). In fact, both EC1-2 fragments adopt twisted X structures maintained by a network of calcium-mediated interactions that fixes the orientation between the EC1 and EC2 modules (26). Formation of the antiparallel dimers was not observed in the X-shaped E-EC1-2 and N-EC1-2 structures.
Studies to elucidate mechanisms involved in homotypic interactions have essentially been performed on class I cadherins and more specifically on murine E-and N-cadherins or on Xenopus C-cadherin. In this study, we present results concerning the mechanism of homotypic adhesion for the class II VEcadherin. We show that the recombinant fragment including the four N-terminal modules of VE-cadherin self-associates in a Ca 2ϩ -dependent hexameric structure. Its visualization by electron microscopy allows us to constrain models for homotypic VE-cadherin associations.

Construction of the Recombinant VE-EC1-4 Expression
Vector-Two oligonucleotides (oligonucleotide A, 5Ј-TATACAT ATG GAT TGG ATT TGG AAC CAG ATG CAC-3Ј; and B, 5Ј-CTC GAA TTC TCA CTC CGG GGC ATT GTC ATT CTC ATC-3Ј) were synthesized and used as primers to produce, by polymerase chain reaction technology, a cDNA fragment containing the sequence from nucleotides 166 to 1458 and encoding VE-cadherin (Asp 1 -Glu 431 ). The DNA template was isolated from Escherichia coli strain XL1-Blue transformed with a full-length VEcadherin cDNA inserted into the pBluescript vector (27).
Oligonucleotides A and B were designed to generate a 5Ј-NdeI restriction site containing an ATG start codon (boldface in sequence above) and a 3Ј-EcoRI restriction site (underlined) positioned immediately after an in-frame TGA stop codon (italic). The polymerase chain reaction fragments were ligated with plasmid pCR-Script (pCR-Script TM Amp SK(ϩ) cloning kit, Stratagene, La Jolla, CA) and then inserted, after digestion with NdeI and EcoRI, into the pET-30b ϩ expression vector (Novagen, Madison, WI). The resulting pET-VE-cadherin vector contained the DNA sequences coding for VE-cadherin (amino acids 1-431) fused to an N-terminal methionine. Prior to the expression of the protein, the cDNA constructs were sequenced to verify that mutations had not arisen during the polymerase chain reaction.
Purification of the Recombinant Fragments-To obtain the fragment designated VE-EC1-4 (Asp 1 -Glu 431 ), E. coli host strain BL21(DE3) was transformed with the pET-VE-cadherin plasmid and grown in Lennox broth base containing 30 mg/liter kanamycin. Production of the recombinant fragment was performed according to the supplier's instructions (Novagen, Madison, WI) using a 100 M isopropyl-␤-D-thiogalactopyranoside induction for 4 h. The cells were then harvested and resus-pended in 40 mM Tris (pH 7.4) containing 5 mM EDTA and a protease inhibitor mixture (Complete tablets, Roche Molecular Biochemicals, Meylan, France). After sonication, the insoluble proteins were collected by centrifugation and washed four times with 0.5 M urea and 0.5% Triton X-100 dissolved in Tris-HCl (pH 7.5). The insoluble proteins were then dissolved in Tris-HCl containing 8 M urea and chromatographed on an anion-exchange Mono Q column (Amersham Pharmacia Biotech, Orsay, France) using a NaCl gradient (0 -300 mM). Fractions, eluted at a NaCl concentration of 120 mM, were pooled and concentrated to 3 mg/ml prior to the refolding according to the following method. Purified urea-containing fractions were diluted in renaturing buffer (50 mM Tris-HCl (pH 8) at 4°C) so that the final concentrations of protein and urea were in the ranges of 20 -50 g/ml and 0.1-0.2 M, respectively. To facilitate the refolding of the protein, CaCl 2 was then added at a 5 mM final concentration, and the solution was incubated overnight without stirring at 4°C before addition of NaCl (150 mM final concentration). After concentration to 2 mg/ml, the protein solution was filtered (0.22 m) and eluted on a Superdex S200 column (Amersham Pharmacia Biotech) to eliminate aspecific aggregates.
Determination of the Protein Concentrations-The absorbance coefficients for the VE-EC1-4 fragments, estimated from the amino acid composition using molar absorption coefficients of 5540 M Ϫ1 cm Ϫ1 for tryptophan residues and 1280 M Ϫ1 cm Ϫ1 for tyrosine residues (20,28), were evaluated to 45,800 M Ϫ1 cm Ϫ1 .
A plot of (Ϫlog K av )1/2 versus R h for the globular proteins previously mentioned allowed us to determine the hydrodynamic radius of the various VE-EC1-4 forms observed on chromatograms (K av where V e is the elution volume for the protein, V o is the column void volume, and V t is the total bed volume (24 ml)). A knowledge of the R h associated with the sedimentation coefficient allowed an accurate determination of the VE-EC1-4 monomer and multimer molecular masses (see "Analytical Ultracentrifugation").
The time required to reach equilibrium between the monomeric and hexameric forms, after dilution of VE-EC1-4 from 45 to 2 M, was evaluated at 20 and 4°C. The hexameric form rapidly decreased initially and stabilized after 30 h at 20°C and after Ͼ2 weeks at 4°C.
The equilibrium between the monomeric (M) and hexameric (H) forms of VE-EC1-4 can be described as . This unusual unit of measurement for K D(H) does not allow direct comparison with previous results established for the monomer 7 dimer equilibrium of E-cadherin. For comparison, we calculated the C 50% values for each equilibrium. (C 50% corresponds to the concentration at which 50% of the fragments by weight are in the multimeric form.) C 50% can be deduced from K D using the following equation: C 50% ϭ (2/n 1/nϪ1 )(K D ) 1/nϪ1 , where n corresponds to the degree of association.
When n is equal to 2, the C 50% value corresponds to the K D for a monomer 7 dimer equilibrium. In contrast, when n is equal to 6, the C 50% value becomes equal to 1.4 ϫ 5 ͌ K D for a monomer 7 hexamer equilibrium.
Matrix-assisted Laser Desorption Ionization (MALDI) Mass Spectrometry-Mass spectra of the recombinant fragment VE-EC1-4, crosslinked or not, were obtained with a time-of-flight mass spectrometer (VoyagerElite Xl, Perseptive Biosystems, Framingham, MA) equipped with a 337-nm nitrogen laser. Aliquots of 0.5 l of the protein solution and 0.5 l of the matrix solution were mixed on a stainless steel sample plate and dried in the air prior to mass spectrometry analysis. External calibration was performed with bovine serum albumin (m/z 66452; Sigma). All experiments were performed using a saturated solution of 2,5-dihydroxybenzoic acid prepared in a 50% (v/v) solution of acetonitrile and aqueous 0.1% trifluoroacetic acid. The accuracy of MALDI molecular mass determinations is between 0.01 and 0.05%.
Cross-linking of VE-EC1-4 -The recombinant fragment VE-EC1-4 was partially cross-linked using N-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDAC; Sigma) or ethylene glycol bis(succinimidyl succinate) (EGS). The molar ratio between the recombinant fragment and the cross-linker reagent was adjusted by varying the concentrations of the cross-linking reagents from 1 to 3 mM and from 0.1 to 3 mM for EDAC and EGS, respectively. In fact, these experiments were performed at 20°C in MES (pH 7.0) for 2 h and in 2 mM Tris (pH 8.5) for 15 min for EDAC and EGS, respectively; and the concentration of VE-EC1-4 was maintained at 110 M. Electrophoresis analyses were then performed on 4 -15% gradient Phast gels (Amersham Pharmacia Biotech).
To determine the oligomeric state of VE-EC1-4, the fragment was cross-linked with EGS used at a 3 mM final concentration in 50 mM Tris-HCl, 100 mM NaCl, and 5 mM CaCl 2 (pH 7.2) for 30 min at room temperature under gentle stirring. Prior to the reaction, the crosslinker was solubilized in Me 2 SO at a concentration of 50 mM to minimize pH and calcium concentration changes and to avoid addition of a large volume. To maintain the oligomeric state of the fragment, the concentrations of VE-EC1-4 and calcium were adjusted to 20 M and 5 mM, respectively. The cross-linking reaction was terminated by adding 100 mM glycine and cooling the reaction flask in ice.
Analytical Ultracentrifugation-Sedimentation velocity and sedimentation equilibrium experiments were performed using a Beckman Model XL-A analytical centrifuge. Sedimentation velocity experiments were carried out at 20°C in 1.2-or 0.3-cm path length double-sector cells with quartz windows and run at 42,000 rpm. The recombinant fragment was diluted at 1.2 and 26.2 M in 20 mM Tris-HCl (pH 7.4) containing 150 mM NaCl and 5 mM CaCl 2 5 days before the sedimentation velocity experiments. Sample volumes of 400 l were centrifuged, and radial scans were taken at 12-min intervals. Data analysis was performed using the computer programs SVEDBERG (30) and DCDT (31), provided by J. S. Philo and W. F. Stafford, respectively. The partial specific volume (v) for VE-EC1-4 (0.733 ml/g), the density (, 1.0059 g/ml) and the viscosity (, 1.022 centipoise) of the buffer were calculated according to the program Sedinterp, supplied by D. B. Hayes, T. Laue, and J. S. Philo. By combining the sedimentation coefficient (S) and the hydrodynamic radius (R h ) of VE-EC1-4 determined by ultracentrifugation and gel filtration chromatography experiments, the molar mass (m) of the multimer could be calculated using the following equation: Sedimentation equilibrium experiments were performed in 1.2-cm double-and six-channel centerpieces at 4500, 6000, and 12,000 rpm at VE-EC1-4 concentrations of 3, 12.5, 25, and 40 M. After 24 -48 h of centrifugation, scans were compared at 2-h intervals to ensure that equilibrium had been reached. An experimental base line was determined at 42,000 rpm. Data of the absorbance (A) at 277 nm as a function of radial position (r) were analyzed considering a reversible self-associating system between monomers and hexamers using the Multifit analysis program (Version 4.01, Beckman Instruments). An apparent association constant (K app ; in absorbance units) was obtained from the set of 12 profiles. For each profile, A 1 (r 0 ), the absorbance for monomers at the radial position r 0 (the first point of each data set), and E, the residual base line, were fitted as follows: where m 1 is the molecular mass of the monomer, is the angular velocity in s Ϫ1 , R is 8.31 J⅐K Ϫ1 ⅐mol Ϫ1 (the gas constant), and T is the temperature in Kelvin. The dissociation constant K D(H) was calculated from K app using an extinction coefficient of 48,843 cm Ϫ1 M Ϫ1 .
Electron Microscopy-Negatively stained specimens of 0.05 mg/ml VE-EC1-4 were prepared using uranyl acetate by the mica floatation technique and observed with a Philips CM200 electron microscope operating at 120 kV. Frozen hydrated specimens of 0.4 mg/ml VE-EC1-4 were prepared on a holey carbon film supported on a 400-mesh copper grid. A 4-l sample of the specimen was deposited onto the grid; the excess was blotted with filter paper for 1-2 s; and the grid was then rapidly plunged into liquid ethane at Ϫ175°C. Specimens were observed at a temperature of approximately Ϫ180°C using a Gatan 626 cryo-holder with a Phillips CM200 electron microscope operating at 200 kV. Images were obtained under low dose conditions (Ͻ10 electrons/Å 2 ) at a nominal magnification of ϫ66,000. The images were recorded on Kodak SO163 electron image films and developed in full-strength D19 developer for 12 min at room temperature. The magnification was calibrated in independent experiments using the 23-Å pitch of tobacco mosaic virus.
Micrographs were digitized using an Optronics microdensitometer with a step size of 25 mm, which corresponds to a pixel size of 3.69 Å for the specimen, taking into account the corrected microscope magnification of ϫ67,700. The length and width of the images of 100 particles were measured on a Silicon Graphics computer using the Semper 6 Plus image analysis package.
Individual particles were selected, masked, and normalized by subtracting the mean and any density gradient present and normalizing the S.D. The particles were centered and aligned by cross-correlation procedures prior to averaging. The variation between particle images suggests that they are in different orientations about their long axis, so internal details are lost in the averaged image, but the overall shape is clear.

RESULTS
Expression and Purification-A recombinant fragment of human VE-cadherin encompassing the four extracellular N-terminal modules (EC1-EC4) was expressed in E. coli in sufficient amounts for biophysical and biochemical studies (Fig. 1). The boundary of this fragment, designated VE-EC1-4, was essentially defined according to the domain organization proposed by Tanihara et al. (19). To avoid cross-linking and the associated problems of insolubility, the fifth module of VE-cadherin (EC5), the only module that may contain disulfide bridges, was excluded ( Fig. 1). Moreover, to increase solubility of the fragment, the hydrophobic amino acid Phe 432 in the C-terminal region of the EC4 module was also deleted. Consequently, the fragment starts immediately after the 47-amino acid long signal peptide at Asp 1 and ends at Glu 431 of the mature VE-cadherin molecule. A foreign methionine was added as a start signal at the N terminus of the fragment.
After purification of VE-EC1-4, a single 49-kDa band was observed by SDS electrophoresis. N-terminal sequencing (Fig.  1) gave a unique signal, confirming the purity and integrity of the extremity of the molecule. Moreover, the molecular mass of the purified fragment (48,948 Da), obtained by MALDI mass spectrometry, matches the calculated one (48,941 Da), indicating the absence of any internal cleavage within the molecule. Circular dichroic measurements attested that VE-EC1-4 possesses a large fraction of ␤-sheets and indicated that it is correctly folded (data not shown).
Self-association of VE-EC1-4 -The capacity of the recombinant fragment to self-associate was evaluated by gel filtration chromatography (Fig. 2). Purified VE-EC1-4 was first injected at a concentration of 1 M onto an analytical gel filtration column ( Fig. 2A). Two peaks with elution volumes equal to 12.4 Ϯ 0.1 ml (peak I; R h ϭ 68 Ϯ 1 Å) and 15.74 Ϯ 0.05 ml (peak II; R h ϭ 38.6 Ϯ 0.8 Å) were detected on the chromatograms. SDS electrophoresis analysis of the chromatographic fractions corresponding to elution peaks I and II showed the presence of a unique 49-kDa band (Fig. 2B). This attested that both elution peaks correspond to the VE-EC1-4 fragment, thus excluding the possibility that peak II was due to a contaminating protein.
It can be predicted that peaks I and II contain the oligomeric and monomeric VE-EC1-4 forms, respectively.
Oligomeric State of VE-EC1-4 -To determine the oligomeric state of VE-EC1-4 in solution, it was cross-linked using the homobifunctional cross-linking reagents EGS (which reacts covalently with amino groups) and EDAC (which reacts covalently with amino and carboxylic groups). Before adding EGS or EDAC, the concentrations of both VE-EC1-4 and Ca 2ϩ were adjusted so that the fragment was mostly in its multimeric form. Electrophoresis analysis of the cross-linked product revealed the formation of a multiple band pattern that varied with the concentration of the cross-linking reagent. In the presence of 0.1 mM EGS, six bands clearly appeared on the gel (Fig. 3A, lane 1). As the concentration of EGS was increased, the intensity of the five lower molecular mass bands decreased, whereas that of the upper band increased (Fig. 3A, lane 2). At 3 mM EGS, only the higher molecular mass band was detected (Fig. 3A, lane 3). Cross-linking with EDAC was comparatively less efficient since the six previously mentioned bands were observed at an EDAC concentration of 3 mM instead of 0.1 mM for EGS (Fig. 3A, lane 5), whereas three bands were observed with 1 mM EDAC (lane 4). The five lower molecular mass bands are partially cross-linked VE-EC1-4 products, and the highest one reflects the oligomeric association found in solution. By attributing a degree of oligomerization to each band (Fig. 3A), it was possible to relate the theoretical molecular mass to the electrophoretic mobility. Indeed, a linear curve was obtained by plotting the log theoretical molecular mass against the corresponding electrophoretic mobility for each band detected on the gel (data not shown). This suggests that the six discrete bands correspond to multimers containing one, two, three, four, five, or six covalently cross-linked molecules from the bottom to the top of the gel.
Using MALDI mass spectrometry, the molecular mass of the fragment cross-linked with 3 mM EGS was measured to be 315 kDa. There are 26 lysines on VE-EC1-4, all susceptible to coupling with EGS. Assuming saturating cross-linking, the The apparent sedimentation coefficient distribution functions (g(s*), i.e. normalized for loading concentrations, but without correction for diffusion effect) were evaluated using DCDT (32), considering six profiles of absorbance obtained for ͐ 2 t between 1.1 ϫ 10 11 and 1.4 ϫ 10 11 s Ϫ1 . molecular mass of the VE-EC1-4 multimer can be calculated as (49 ϩ 26 ϫ 0.224) ϫ n, where 49 corresponds to the mass of a VE-EC1-4 monomer, n is the degree of oligomerization, and 0.224 corresponds to the mass of an EGS molecule coupled to the fragment. Accordingly, the molecular masses were theoretically evaluated as 274 and 329 kDa for pentameric and hexameric associations, respectively. Due to the lack of accessibility to EGS of some lysines buried inside the core of the protein, these calculated values are likely overestimated. This suggested that the 315-kDa experimental value corresponds to a hexameric association.
The oligomeric state of VE-EC1-4 was determined independently by analytical centrifugation. Sedimentation velocity experiments were performed using VE-EC1-4 at 1.2 and 26 M. At 1.2 M, the concentration profiles exhibited bimodal boundaries (Fig. 3B). Consistent with this observation, g(s*) versus s*, generated from this set of concentration profiles with the software DCDT, showed a double distribution (Fig. 3C). This reflected the heterogeneity existing within the molecular masses of the VE-EC1-4 protein at 1.  (Fig. 3C).
The molar masses for each distribution were then calculated from the sedimentation coefficients combined with the hydrodynamic radius determined by gel filtration chromatography. Under these conditions, the molar masses were estimated to 49 and 301 kDa for the 2.9 S and 10.2 S distributions, respectively. These results indicate that the 2.9 S peak distribution corresponds to the monomeric form of VE-EC1-4, whereas the 10.2 S peak distribution corresponds to the hexameric form of VE-EC1-4.
To determine the molecular mass by a method that does not depend on the shape of the molecule, sedimentation equilibrium experiments were performed. VE-EC1-4 at various initial concentrations was equilibrated at 4500, 6000, and 12,000 rpm (Fig. 4, A and B).The mean molecular mass increased with increasing protein concentration and reached a plateau value of 285 kDa, compatible with a hexameric form (Fig. 4C).
Altogether, the results from gel filtration chromatography, analytical ultracentrifugation, and MALDI spectrometry demonstrate that the multimer of VE-EC1-4 is a hexamer. Moreover, variations in the relative amounts of VE-EC1-4 hexamer over monomer as a function of protein concentration show that both species are at equilibrium in solution.
Dissociation Rate Constants of the Hexameric Form-To quantify the equilibrium between the monomer and hexamer, the stability of the hexamer was first analyzed by gel filtration chromatography. It was noticed that, following dilution of the VE-EC1-4 stock solution, the level of hexamer relative to monomer slowly decreased with time. Equilibrium between the monomeric and hexameric forms required at least 30 h at 20°C (Fig. 5A). At 4°C, the dissociation rate was considerably slowed down since equilibrium was reached 2 weeks after the dilution (data not shown). From this kinetic study, the hexamer dissociation rate constants (k d ) at 4 0 and t, respectively). The dissociation rate constants of the hexamer (k d ) were calculated to be 6 ϫ 10 Ϫ6 and 1 ϫ 10 Ϫ6 s Ϫ1 at 20 and 4°C, respectively.

Apparent Dissociation Constant of the Hexamer-
The parameters of the equilibrium between the monomeric and hexameric forms were further established by loading VE-EC1-4 at a range of initial concentrations onto the gel filtration column. The chromatographic runs were carried out at 4°C, taking advantage of the low hexamer dissociation rate constant at this temperature. Indeed, we verified that, at 4°C, the hexamer was not dissociated during the chromatographic runs despite the occurring dilution.
The chromatographic profiles showed that, as loading concentrations of VE-EC1-4 were increased, a simultaneous increase in the intensity of peak I and a decrease in that of peak II were observed ( Fig. 2A). Based on the previous peak attribution, it was deduced that increases in VE-EC1-4 concentration favor formation of the hexamer, thus confirming results generated from sedimentation velocity experiments.
The monomer/multimer composition could be estimated from the areas of peaks I and II seen on the chromatographic profiles. The relationship between the percentage of multimer in solution and the initial concentration of VE-EC1-4 was then Sedimentation equilibrium experiments were also used to determine the apparent dissociation constants (K D(H) ) independently of the determination of R h . As shown in Fig. 4, our data were successfully described by the simplest model considering an equilibrium between the monomeric and hexameric forms of VE-EC1-4. A K D(H) value of 0.5 M 5 , similar to that determined from gel filtration chromatography experiments, was deduced. From this K D(H) value, C 50% was estimated as 1 M (see "Materials and Methods"). This value is comparable to that experimentally determined by gel filtration chromatography (0.5 M).
Ca 2ϩ Stabilization of the Hexamer-To study the Ca 2ϩ dependence of the monomer 7 hexamer equilibrium by gel filtration chromatography, a modified elution buffer containing 10 mM EGTA was used to elute VE-EC1-4 loaded in the presence of Ca 2ϩ . The resulting progressive removal of free Ca 2ϩ prevented the formation of aggregates that clog the column. As illustrated in Fig. 5C, progressive Ca 2ϩ depletion induced a large increase in peak II at the expense of peak I. This shows that the multimer dissociates into monomers in the absence of Ca 2ϩ .
Electron Microscopy of VE-EC1-4 -In the presence of 5 mM Ca 2ϩ at a protein concentration of 2 M, conditions for which the hexameric form is the preponderant species, the electron micrographs of negatively stained VE-EC1-4 show a quite homogeneous population of particles with an elongated ellipsoidal shape (Fig. 6A). In the presence of EGTA, smaller objects with various ill defined shapes replaced these structures, indicating that formation of the hexamer is dependent on the presence of Ca 2ϩ (data not shown).
A more precise picture of the shape of the hexamer was obtained by VE-EC1-4 in the frozen hydrated state (Fig. 6, B and C). It has a hollow cigar shape when viewed from the side (Fig. 6C) and appears as an annulus when viewed from the end (Fig. 6B). There is apparently a specific interaction between VE-EC1-4 and the air/water interface, so there are only two views of the molecule. In each view, the protein density of the hexamer is subdivided into smaller domains as expected for VE-EC1-4, but it is too variable from particle to particle to allow determination of the multimeric state from these images. The average dimensions of 100 hexameric VE-EC1-4 particles, visualized by cryoelectron microscopy, are 233 Ϯ 10 ϫ 77 Ϯ 7 Å. DISCUSSION A recombinant fragment encompassing the four extracellular modules of VE-cadherin was produced in E. coli. Recently, this fragment was demonstrated to be functional since it is able to inhibit the aggregation of Chinese hamster ovary cells transfected with VE-cadherin in a concentration-dependent manner (27). Here, we demonstrated that this fragment spontaneously forms a unique well organized multimer. This study provides the first in vitro evidence that the extracellular part of cadherins can homoassociate in solution in a structure possessing an oligomeric order higher than 2. Furthermore, this calciumdependent oligomer self-assembles at lower concentrations than those previously published for cadherin dimeric associations (20). FIG. 5. Gel filtration analysis of the monomer 7 hexamer equilibrium. All the chromatographic runs were performed at 4°C to avoid dissociation of the hexamer. A, kinetic study of the hexamer dissociation. VE-EC1-4 was diluted from 45 to 2 M and stored at room temperature. The percentage of VE-EC1-4 present as hexamers was estimated from chromatographic profiles performed at different time intervals after dilution. B, relationship between hexamer formation and total concentration of VE-EC1-4. Multimer/monomer molar ratios were estimated from molecular exclusion profiles by integrating the area of peaks I and II (see "Materials and Methods") for various loading concentrations of VE-EC1-4. C, Ca 2ϩ -dependent association of the hexameric VE-EC1-4 form. VE-EC1-4 at 18 M (equilibrated in 5 mM Ca 2ϩ ) was injected onto a Superdex S200 gel filtration column. Elutions were performed in Tris/saline (pH 7.4) running buffer containing either 5 mM Ca 2ϩ (OO) or 10 mM EGTA (---). In the presence of 10 mM EGTA, the hexamer (peak I) dissociated into the monomer (peak II). Analysis of the oligomeric state of the VE-cadherin VE-EC1-4 fragment both by analytical centrifugation and, after cross-linkage, by mass spectrometry revealed a hexameric association. This is the first description of such an organization in solution for a cadherin. Until now, the extracellular recombinant fragments of other cadherins, such as E-cadherin, were demonstrated either to stay monomeric (15,33) or to form dimers, as for Xenopus C-cadherin (34). Oligomeric self-associations were observed for the extracellular region of E-cadherin; but in this case, the molecule was fused to the assembly domain of the cartilage oligomeric matrix protein to force oligomerization (24,33). Removal of the fifth module (EC5) to the extracellular part of VE-cadherin does not change the properties of the molecule since we recently verified that the VE-EC1-5 fragment (Asp 1 -Asp 542 ), which corresponds to the whole extracellular domain of the receptor, was also able to form a hexameric structure.
Gel filtration chromatography and ultracentrifugation experiments showed evidence of a concentration-dependent monomer 7 hexamer equilibrium with a dissociation constant K D(H) ϭ 0.3-0.5 M ϩ5 . Although the detailed mechanism resulting in the elaboration of the hexamer is not known, it can be assumed that this occurs via intermediate species such as dimers or trimers. These transient associations probably appear for VE-EC1-4 concentrations lower than those required for hexameric associations, suggesting that VE-EC1-4 dimers may be formed at concentrations lower than 0.5 M. The unusual unit used for this K D(H) value does not allow a direct comparison between our value and those previously published for other cadherins. To palliate this inconvenience, C 50% was evaluated for comparison with other cadherins. The VE-EC1-4 fragment associates as a hexamer with a C 50% of ϳ0.5 M, whereas the E-EC1-2 fragment, containing the two N-terminal modules of E-cadherin, forms dimers with a C 50% ranging from 80 M (20) to 170 M (35). Differences in propensity to selfassociate observed for E-and VE-cadherin fragments suggest that class I and II cadherins elaborate various types of homotypic interactions.
Our results also indicate that the hexamer is able to dissociate after dilution, with a very slow k d dissociation rate constant that increases from 10 Ϫ6 to 6.10 Ϫ6 s Ϫ1 when the temperature is raised from 4 to 20°C. This reflects the stability of the hexameric structure.
Formation of the hexameric structure observed for the VEcadherin fragment is strictly dependent on the presence of Ca 2ϩ . This result reflects the behavior of cadherins expressed at the cell surface and is in agreement with studies done on various cadherin-derived recombinant fragments. Indeed, except for the earliest works of Shapiro et al. (14) and Brieher et al. (34), multimers of cadherins were formed only in the presence of Ca 2ϩ (20,(23)(24)(25)35).
Cryoelectron microscopy allows us to visualize the hexameric structures as elongated cigar-shaped particles. From these images, the average dimensions of the hexamers were evaluated as 233 Ϯ 10 ϫ 77 Ϯ 7 Å (Fig. 6, B and C). The length of the hexameric structure is comparable to the distance separating two adjacent endothelial cells, estimated by electron microscopy to be ϳ200 Å (36). It is not clear from the cryoelectron microscope images whether the six cadherin molecules adopt a parallel or an antiparallel orientation within the hexamer. Determination of the relative orientation of the six molecules of VE-cadherin within the hexamer is a key step in understanding how adherent junctions are formed between endothelial cells. A parallel orientation would indicate formation of hexamers with molecules emanating from a single cell surface, whereas an antiparallel orientation would indicate formation of adhesive hexamers consisting of molecules protruding from neighboring cells. The crystal structure of the hexameric VE-EC1-4 complex should resolve this question. Nevertheless, the fact that VE-cadherin was demonstrated, by atomic force microscopy, to be able to constitute multiple trans-interactions is compatible with an antiparallel orientation of the hexamer (37). This trans-orientation is also in agreement with results recently published demonstrating that formation of antiparallel dimers requires Ca 2ϩ , in contrast to parallel dimers (38,39). Moreover, as attested by the determination of the N-EC1 structure (14) and as deduced from the electron microscopy measurement for the E-EC1-5 cadherin construct (i.e. 220 Å) (33), the average dimension of a cadherin module is 45 Ϯ 1 Å. Consequently, the length of the VE-EC1-4 hexamer (233 Ϯ 10 Å) corresponds approximately to the length of five cadherin modules. Thus, this is compatible with a model in which the six monomers are disposed in an antiparallel fashion to form a barrel with three overlapping modules along its length.
Due to its relatively high affinity constant, the hexamer appears at low concentrations. Placed in a cellular context, this result indicates that this hexameric structure may be physiologically relevant at the cell-cell junctions where VE-cadherin is concentrated. In fact, this hexameric assembly may constitute a basic association motif whose assembly may promote formation of the continuous sealing region between endothelial cells.