N-glycosylation alters cadherin-mediated intercellular binding kinetics

We present direct evidence that the N-glycosylation state of neural cadherin impacts the intrinsic kinetics of cadherin-mediated intercellular binding. Micropipette manipulation measurements quantified the effect of N-glycosylation mutations on intercellular binding dynamics. The wild-type protein exhibits a two-stage binding process in which a fast, initial binding step is followed by a short lag and second, slower transition to the final binding stage. Mutations that ablate N-glycosylation at three sites on the extracellular domains 2 and 3 of neural cadherin alter this kinetic fingerprint. Glycosylation does not affect the affinities between the adhesive N-terminal domains, but instead modulates additional cadherin interactions, which govern the dynamics of intercellular binding. These results, together with previous findings that these hypo-glycosylation mutations increase the prevalence of cis dimers on cell membranes, suggest a binding mechanism in which initial adhesion is followed by additional cadherin interactions, which enhance binding but are modulated by N-glycosylation. Given that oncogene expression drives specific changes in N-glycosylation, these results provide insight into possible mechanisms altering cadherin function during tumor progression.


Introduction
Classical cadherins are crucial adhesion and architectural glycoproteins that maintain the integrity of intercellular junctions in all cohesive tissues (Gumbiner, 2005;Inoue et al., 2001). These proteins bind cadherins on adjacent cell surfaces to mediate cellcell adhesion and trigger signaling cascades that control differentiation, proliferation and cell migration (Bhadriraju et al., 2007;Desai et al., 2009;Fournier et al., 2008;Li et al., 2010;Liu et al., 2006;Rowlands et al., 2004;Wong and Gumbiner, 2003). The protein architecture comprises an extracellular region, which embeds the adhesive function, a transmembrane domain and a cytoplasmic domain (Yap et al., 1997a). Five, tandemly arranged, b-barrel domains designated EC1-EC5, and numbered from the Nterminal domain, form the extracellular region (Fig. 1A,B). Cadherins form trans adhesive bonds between the EC1 domains of cadherins on apposing cells, and there is some experimental evidence that they also form lateral, cis interactions on the cell membrane (Brieher et al., 1996;Kim et al., 2005;Klingelhöfer et al., 2002;Yap et al., 1998).
Cadherins are essential both for morphogenesis and for maintaining the mechanical integrity and barrier functions of adult tissues. Mutations that compromise adhesive activity or localization at cell membranes correlate with diseases such as cancer. In addition to mutations in the primary structure, aberrant N-glycosylation of cadherin extracellular domains is increasingly linked to the metastatic phenotype of several cancers (Geng et al., 2004;Guo et al., 2009;Jamal et al., 2009;Kitada et al., 2001;Nita-Lazar et al., 2009;Nita-Lazar et al., 2010;Pinho et al., 2011;Vagin et al., 2008). Epithelial cadherin (E-cadherin) has several Nglycosylation sites along the extracellular region. E-cadherin in breast tumors exhibits highly branched N-glycans on extracellular domains EC4 and EC5, and hyper-glycosylation at these sites destabilizes epithelial junctions and increases tumor progression (Pinho et al., 2009a;Pinho et al., 2009b;Pinho et al., 2011;Vagin et al., 2008). Conversely, hypo-glycosylation, achieved by ablating an N-glycosylation site in EC4 enhanced intercellular junction assembly, cytoskeletal remodeling at junctions and barrier integrity (Jamal et al., 2009;Liwosz et al., 2006;Nita-Lazar et al., 2010;Pinho et al., 2011). Neural cadherin (N-cad) contains eight N-glycosylation consensus sequences in the extracellular segment that are post-translationally modified (Guo et al., 2009). Ablating all eight sites increased intercellular junction stability, enhanced ERK signaling, and reduced cell migration in a scratch wound-healing assay (Guo et al., 2009). In contrast to E-cadherin, neural cadherin did not demonstrate significant N-glycosylation on domains EC4 and EC5 (Guo et al., 2009). Instead, eliminating three N-glycosylation sites at N273, N325 and N402 on extracellular domains EC2 and EC3 had the same functional impact as mutating all eight sites. Crosslinking studies further indicated that the loss of N-glycans on EC2-EC3 of N-cadherin increased the prevalence of dimers on cell surfaces, suggesting that N-glycosylation modulates cis dimerization (Guo et al., 2009).
There is strong in vivo evidence that specific alterations in the structures of N-glycans expressed on cell surface glycoproteins are common during the oncogenesis of epithelial cells. These changes are driven by the transcriptional regulation of specific glycosyltransferases by oncogene expression (Buckhaults et al., 1997;Fernandes et al., 1991;Murata et al., 2000;Pierce, 2009;Seales et al., 2005;Zhao et al., 2008b). Collectively, several studies suggest that N-glycosylation alters cadherin-associated functions, but evidence that N-glycosylation directly affects intrinsic cadherin functions has been circumstantial. This is because glycosylation could also perturb trafficking to the cell membrane or alter cadherin interactions with other cell surface proteins such as growth factor receptors (Doherty and Walsh, 1996;Gumbiner, 1996;Perrais et al., 2007;Tzima et al., 2005;Williams et al., 2001a). Moreover, the mechanism by which Nglycosylation of the EC2-EC5 region could alter intercellular adhesion is not obvious because these glycosylation sites are distant in the sequence from the adhesive EC1 interface (Fig. 1A,B).
This study uses micropipette manipulation measurements ( Fig. 2) to quantify the impact of N-glycosylation mutations on the intrinsic dynamics of N-cadherin-mediated intercellular binding. Micropipette measurements have been used to determine the two-dimensional affinities and dissociation rates of several membrane-bound adhesion protein receptors (Chesla et al., 2000;Chien et al., 2008;Huang et al., 2004;Huang et al., 2007;Jiang et al., 2011;Leckband, 2000;Long et al., 2001;Williams et al., 2001b;Zhang et al., 2005). Such measurements previously showed that C-cadherin-mediated intercellular binding kinetics exhibit a two-stage process in which an initial, fast EC1-dependent binding step is followed by a short lag period and a further, slower increase in binding probability (Chien et al., 2008). Here we describe the effect on this kinetic signature of ablating N-glycosylation sites in the EC2 and EC3 domains of neural cadherin (N-cad). The findings suggest a binding mechanism in which initial, trans cadherin ligation is followed by additional cadherin interactions, which augment binding but are modulated by N-glycosylation.

Wild-type human N-cadherin displays biphasic kinetics
The micropipette measurements quantify the binding probability P, which is the number of detected cell binding events n b divided by the total number of cell-cell contacts N T . The change in P with cell-cell contact time thus reflects the time-dependent increase in intercellular bonds (Chesla et al., 1998). Figure 3 shows the binding kinetics measured between Chinese hamster ovary (CHO) cells that stably express wild-type human Ncadherin (hN-CHO WT) at 18 cadherins/mm 2 and red blood cells (RBCs) modified with immunoglobulin Fc-tagged chicken Ncadherin (ckN-cad-Fc) at a surface density of 10 cadherins/mm 2 . The binding kinetics exhibits two distinct stages. The first stage is a fast, initial rise to a plateau P 1 at ,0.41 within the first 2 seconds. This is followed by a 2-5 second lag or induction phase, and then a slower, second rise to a higher binding probability at P 2 ,0.55. The second plateau P 2 is stable for contact times up to 30 seconds, and previous studies showed that this plateau persists for at least 45 seconds (Chien et al., 2008).
Previous studies similarly showed that C-cadherin, which is also a Type I classical cadherin, displays biphasic (two-stage) kinetics (Chien et al., 2008). Further studies with cadherin EC domain deletion mutants showed that EC1 is required for the first, fast binding step, and the full extracellular domain is needed for the lag and second binding step (Chien et al., 2008). The first, EC1-dedpendent binding step could therefore be described by a The Asn residues of the N-glycosylation sites are shown in red. Amino acids from the extracellular domain 2 (EC2) portion that contribute to the putative cis-binding interface are in gray. (C) An expanded view of the putative cis binding interface between adjacent cadherins. The N-glycosylation site N273 is red, and is in close proximity to N243 (yellow) from the adjacent cadherin. Fig. 2. Schematic of the micropipette set-up. A CHO cell expressing the membrane bound human N-cadherin is partially aspirated into a micropipette with a diameter of ,7 mm. A red blood cell (RBC), which has been chemically modified with Fc antibody and immobilized, oriented chicken Ncad-Fc is drawn into a micropipette with a 1.5 mm radius. Molecular configurations on the apposing CHO and modified red blood cells are shown below. simple binding mechanism for the formation of a trans adhesive dimer, RzL<LR.
The probabilistic kinetic rate equation describing this mechanism (Chesla et al., 1998) is: where A c is the contact area (mm 2 ), m R and m L are cadherin densities on the two cells (number/mm 2 ), k a is the two-dimensional affinity (mm 2 ), and k off is the dissociation rate (seconds 21 ). Thus, the two-dimensional binding affinity and dissociation rates could be determined from fits of the kinetic data to Equation 1 (Chien et al., 2008). The solid line in Fig. 3 is the nonlinear least squares fit of the data to Equation 1. The best-fit affinity and dissociation rate are, respectively, 4.260.4610 -4 mm 2 and 2.060.5 seconds 21 (Table 1). Importantly, Equation 1 does not describe the entire kinetic profile. Furthermore, the inclusion of a transient encounter complex Sivasankar et al., 2009) predicts an exponential rise to a single plateau, but it does not predict the twostage kinetics shown here.
To identify data that correspond to the first, EC1-dependent step, we used a lack-of-fit test to parse the data into the two, distinct kinetic stages: namely the rise to P 1 and to P 2 . This test compares the least squares residuals of the model (Equation 1) to the intrinsic variability of the data. When the test statistic exceeds the critical value for a given time point, then the model does not describe the data in question. To determine the dissociation rate and two-dimensional affinity for the initial step, Equation 1 was then fit to the maximum number of time points in each data set that did not fail the lack-of-fit test. This approach confirmed that Equation 1 does not describe the second increase in binding probability at longer times.

N-glycosylation alters N-cadherin binding kinetics
To determine the effect of N-glycosylation on N-cadherinmediated intercellular binding kinetics, binding probability data were obtained with human N-cadherin mutants in which all eight of the N-glycosylation sites were mutated to Gln (hN-CHO MuALL) or sites 2, 3 and 4 (N273, N325 and N402) were mutated to Gln (hN-CHO Mu234) (Fig. 1A,B). In subsequent kinetic measurements, N-cadherin mutants were expressed on CHO cells at similar levels, as verified by western blot (Fig. 4). According to quantitative fluorescence-activated cell sorting (FACS) measurements, the expression levels were ,18 cadherins/mm 2 .
The impact of hypo-glycosylation on the kinetics would depend on whether the lesser (limiting) protein ligand is the mutant expressed on CHO cells or the glycosylated ckN-cad-Fc bound to the RBCs. This is because the limiting reagent will dominate the kinetics. We therefore carried out measurements in which we varied the density of ckN-cad-Fc on the RBCs, so that the glycosylated protein was either less than (limiting) or in excess of the human N-cadherin (or mutant) on the CHO cells. The ckN-cad-Fc, which is expressed in mammalian cells, is glycosylated (Fig. 4B). Fig. 5A shows the impact of the hypo-glycosylation mutations on binding kinetics, when the human N-cadherin mutants expressed on the CHO cells are limiting. The time course measured between hN-CHO WT (18 cadherin/mm 2 ) and ckNcad-Fc (37 cadherins/mm 2 ) exhibits the biphasic kinetic signature (Fig. 5A). By contrast, ablating eight N-glycosylation sites (hN-CHO MuALL, ,18 cadherin/mm 2 ) nearly eliminates the induction phase, and the binding probability rises smoothly to the higher probability plateau at P 2 ,0.63. The data do not exhibit a single exponential rise, indicating that at least two processes contribute to the overall kinetics.
Previous findings indicated that N-glycosylation sites in human N-cadherin domains EC2-EC3 ( Fig. 1B) have the greatest effect on cadherin dimerization, cell migration and ERK signaling (Guo et al., 2009). The binding kinetics of the hN-CHO Mu234 variant is similar to that of hN-CHO MuALL, and the time course also exhibits a rapid rise to P 2 ,0.73 (Fig. 5A). The difference in amplitudes (P 1 and P 2 ) measured with hN-CHO MuALL and hN-CHO Mu234 is due to the different ckN-cad-Fc densities used in the different measurements. Importantly, there is no significant induction phase observed with either of the hypoglycosylated N-cadherin mutants, when their densities are limiting. The altered kinetics relative to hN-CHO WT clearly  shows that N-glycosylation on N-cadherin EC2-EC3 domains directly affects the intrinsic cadherin binding dynamics. By contrast, when hypo-glycosylated N-cadherin is in excess and the glycosylated ckN-cad-Fc is the limiting ligand, the binding kinetics displays the induction phase. Fig. 5B shows the kinetic profiles measured with ckN-cad-Fc on the RBCs at 10 cadherins/mm 2 and 7 cadherins/mm 2 . Measurements with hN-CHO WT and with the two, hypo-glycosylation mutants all exhibit the two-stage kinetic profile, as expected if the glycosylated ckN-cad-Fc densities were rate limiting and the glycans affect the kinetics. Interestingly, the best-fit, twodimensional affinities for the first binding step, as measured with wild-type human N-cadherin and with the two glycosylation mutants (Table 1), are statistically similar, indicating that Nglycan removal does not affect EC1-EC1 binding.

Enzymatic deglycosylation of ckN-cad-Fc eliminates the induction phase
To test whether the lag phase observed in Fig. 3 and Fig. 5B is due to carbohydrates on ckN-cad-Fc, the glycans were trimmed enzymatically with exoglycosidases and EndoH under nondenaturing conditions. The ckN-cad-Fc extracellular domain contains six N-glycosylation sites, including sites at N278, N330 and N407 (Accession P10288), and therefore has a similar Nglycosylation profile to the human protein. Treatment with a mixture of EndoH, neuraminidase, PNGase and 1,4 bgalactosidase reduced the apparent ckN-cad-Fc molecular mass by ,15 kDa relative to the untreated protein (Fig. 4B). Under non-denaturing conditions, the enzymes are less efficient, so it is likely that some core glycans remain.
Consistent with expectations, the kinetics measured with ckNcad-Fc did not exhibit the induction phase (Fig. 6A), at densities where ckN-cad-Fc is limiting (compare with Fig. 5B). In particular, the time course measured with enzymatically treated ckN-cad-Fc versus hN-CHO MuALL exhibits a smooth rise to a plateau at P,0.6. The kinetic features measured with human N-cadherin mutants are similar qualitatively to the kinetics measured when the hypo-glycosylated mutant was limiting (compare with Fig. 5A).
The kinetics measured with hN-CHO WT and RBCs modified with treated (Fig. 6A, black circles) or untreated ckN-cad-Fc (Fig. 6B, black squares) clearly show that ckN-cad-Fc hypo-glycosylation eliminates the lag phase, and accelerates the rise to the higher binding probability P 2 . Fig. 6B contrasts the kinetics measured with glycosylated proteins on both cells with those in which proteins on both cells are hypo-glycosylated. Similarly, the binding kinetics between identical RBCs modified with untreated ckN-cad-Fc (7 cadherins/mm 2 ) exhibited no induction phase, even at low cadherin densities (not shown). A comparison of the kinetics measured with treated ckN-cad-Fc and either N-CHO MuALL or N-CHO WT (Fig. 6A) reveals a slightly higher limiting binding probability with the N-CHO-MuALL cells, despite similar cadherin densities on the cells in both cases. Although this could be due to slight differences in the intrinsic cadherin binding properties, the final binding  probabilities measured with RBCs modified with either treated or untreated ckN-cad-Fc (7 cadherins/mm 2 ) were the same.

Discussion
The extent of N-glycosylation directly alters the intrinsic intercellular binding kinetics of neural cadherin. A large number of studies link aberrant glycosylation to altered cadherin-dependent cell functions, in the context of different cancers. Phenotypic changes include altered cell motility in scratch-wound healing assays, cytoskeleton reorganization, metastasis, ERK signaling and barrier function (Bajpai et al., 2008;Guo et al., 2009;Jamal et al., 2009;Liwosz et al., 2006;Pinho et al., 2009a;Pinho et al., 2011;Vagin et al., 2008;Zhao et al., 2008a). In the case of N-cadherin, the N-glycosylation sites at N273, N325 and N402 are highly conserved. Mutation of these sites ablates glycosylation and alters cell functions, but the loss of glycosylation does not affect cadherin trafficking to the membrane (Guo et al., 2009). Previous findings collectively suggested that glycosylation perturbs cadherin-dependent cell adhesion. The results of this study provide direct evidence that N-glycosylation alters the kinetics of N-cadherin-mediated intercellular bonds.
In the two-stage, kinetic time course, N-glycosylation modulates the duration of the induction (lag) phase and subsequent increase in binding probability, but not the intrinsic two-dimensional affinity of the EC1-EC1 bond. The glycosylation-independence of the affinity agrees with the similar measured affinities of N-cadherin EC1-EC2 fragments expressed in bacteria and in mammalian cells (Katsamba et al., 2009). The basis of N-glycosylation-dependent perturbations to cadherin therefore appears to differ from the sialylation of neural cell adhesion molecule, where the large excluded volume of polysialic acid blocks intercellular binding . Instead, N-glycosylation mutations at N273A, N325A and N402A increased the prevalence of cis N-cadherin dimers on cell membranes (Guo et al., 2009). The latter result, together with the two-stage kinetics and its requirement for the full extracellular region (Chien et al., 2008), suggests a model in which initial adhesion facilitates additional cadherin interactions that enhance intercellular binding (Fig. 7). We speculate that the second, slower rise in binding probability is due to lateral association, which is in turn modulated by N-glycosylation.
Other observations are qualitatively consistent with the second step arising from lateral association. Increasing the cadherin density appears to reduce the induction phase, as expected for a binary reaction (see Fig. 5A,B). In a previous report, hexahistidine-tagged (monomeric) C-cadherin exhibited a longer induction time than Fc-tagged (dimeric) C-cadherin (Chien et al., 2008). Single-bond rupture studies and cell adhesion data suggest that cadherin dimers enhance adhesion relative to monomers ). Cross-linked dimers of C-cadherin extracellular domains also supported greater cell adhesion than monomers (Brieher et al., 1996).
Experimental evidence that cadherins dimerize on free cell membranes is mixed, however. At cell-cell junctions, cadherins form clusters involving many hundreds of proteins (Chtcheglova et al., 2010;Shewan et al., 2005;Yap et al., 1997a;Yap et al., 1997b). However, clustering requires active processes, and is unlikely driven by protein-protein interactions (Shewan et al., 2005). Chemical crosslinking did detect E-cadherin dimers on cells (Troyanovsky et al., 2003), but a similar approach failed to detect dimers of wild type N-cadherin (Guo et al., 2009), and  Table 1. Fig. 7. Proposed model for the effect of N-glycosylation on the dynamics of N-cadherin-mediated intercellular junction assembly. In this model, opposing cadherin monomers initially form trans adhesive bonds. This initial binding in turn facilitates lateral cadherin association, which is slowed by Nglycans on the cadherin ectodomain. antibodies specific to monomeric or to dimeric cadherin detected only monomers on cells (Tsuiji et al., 2007). Results of recent simulations, however, suggest that cell-cell adhesion might be required to overcome the energy barrier to lateral cadherin association . Entropy would contribute to this barrier, but glycosylation might also play a role. Steric interference by N-glycans could alter the free energy difference between monomers and dimers, and shift the dimer-monomer equilibrium, as suggested by crosslinking studies (Guo et al., 2009). Significant pre-dimerization would increase the first plateau P 1 , but the effect of this initial equilibrium on subsequent kinetic steps depends on whether the clusters catalyze further oligomerization. Glycans could also create a kinetic barrier that would slow the dimerization rate, for example, by increasing the induction phase. Hypo-glycosylation through enzymatic treatment, point mutations or downregulation of glycosyltransferases would reduce the barrier and accelerate the rate of lateral association in intercellular gaps.
A similar mechanism is suggested by recent computer simulations of the adhesion-dependent coalescence of cadherin lattices between apposed surfaces Wu et al., 2010;Wu et al., 2011). The simulations constrained lateral cadherin association to mimic contacts in a crystal structure, where a frequently observed interface between EC1 and EC2 (Fig. 1B) is postulated to be the cis bond . It is noteworthy that N273, which is one of the sites investigated in this study, lies at this postulated interface. The crystal structures were of deglycosylated proteins , so it is not possible to visualize how the proposed cis interface might accommodate a large carbohydrate. The latter would likely generate a substantial steric barrier that could impede or disrupt the putative interaction.
Glycans on N-cadherin in melanoma lines have been characterized, but the glycan structure and molecular mass at each site is unknown and would vary with cell type and culture conditions. The 15 kDa shift in N-cadherin molecular mass following enzymatic treatment (Fig. 4B) gives an upper bound for expressed proteins in these studies. Additionally, prior characterization of the N-glycosylation mutants of human Ncadherin showed that each of the N273, N325 or N402 mutants alone did not generate the Mu234 or MuALL phenotype. Instead, the elimination of the induction phase required ablating all three N-glycosylation sites on both EC2 and EC3 domains, and suggests that the putative cis interface involves more extensive contacts between ectodomains -a possibility supported by the requirement of EC3 for the induction phase and second kinetic step (Chien et al., 2008).
Despite apparent parallels between our findings and the simulations , there are important differences. First, the induction and second step in the kinetic profiles requires EC3 (Chien et al., 2008), which was not implicated in the structure-based model. Second, the effect of N325 glycosylation on the putative cis interface and junction organization is unknown. Third, it is unclear whether the sparse cadherin densities used in our studies would nucleate clusters, as depicted in the simulations. Although there might be different interpretations of the structural basis of these experimental observations, the results presented here are consistent with a process in which adhesion triggers subsequent changes in cadherin organization at cell-cell junctions that further enhances binding and is modulated by N-glycans. This study focused on N-cadherin glycosylation, but E-cadherin hyper-glycosylation has been implicated in cadherin dysfunction in cancer (de Freitas Junior et al., 2011;Geng et al., 2004;Pinho et al., 2009a;Pinho et al., 2011;Zhao et al., 2008a). Although phenotypic changes of E-cadherin hypo-glycosylation (Jamal et al., 2009;Nita-Lazar et al., 2010) are qualitatively similar to those reported for N-cadherin (Guo et al., 2009), the glycosylation sites with the greatest impact on E-cadherin-based cell functions appear to be on membrane proximal domains EC4-EC5 (Liwosz et al., 2006). Despite the significance of those sites for E-cadherin, ablation of the same glycosylation sites on human N-cadherin did not alter ERK signaling or wound healing assays perceptibly (Guo et al., 2009). Studies have, however, documented altered N-linked glycan expression on specific receptors on tumor cells, driven by oncogene expression (Buckhaults et al., 1997;Fernandes et al., 1991;Murata et al., 2000;Pierce, 2009;Seales et al., 2005;Zhao et al., 2008b). It remains to be determined how the location and size of N-glycans on the extracellular domain, particularly those N-glycans whose structures change during oncogenesis, alter intrinsic cadherin function.
Finally, our use of human and chicken N-cadherin does not impact the conclusions of this study, primarily because we compared the effect of glycosylation on identical pair-wise interactions. The EC1 binding affinity between ckN-cad-Fc and chicken N-cadherin on CHO cells is slightly lower than that between ckN-cad-Fc and human N-cadherin (Table 1). Despite .90% sequence identity, small differences between the Ncadherins could affect binding in two ways. EC1 binding pocket differences could alter the affinities, or sequence variations could affect glycosylation. Chicken N-cadherin contains the consensus sequence for N-linked glycosylation at N278, N330, N407, N578, N628 and N657, and the first three sites correspond to the sites that were mutated in the human N-cadherin Mu234 variant. Similar N-glycosylation and our finding that glycosylation does not affect EC1-dependent affinities (Table 1) indicate that the affinity differences are due to binding site variations. However, neither the qualitative two-stage kinetic signature nor the induction phase depends on the EC1-EC1 affinity.
In summary, these results demonstrate that the N-glycosylation of neural cadherin directly impacts the intrinsic dynamics of cadherin-dependent intercellular binding. Glycosylation does not affect EC1-dependent affinities, but the two-stage kinetic fingerprint and prior crosslinking results support a binding mechanism in which initial, EC1-dependent trans binding is followed by additional cadherin interactions that enhance intercellular binding. N-glycans localized at three sites in the EC2-EC3 domains of N-cadherin alter the second step in the two-stage kinetic profile to modulate the intercellular binding dynamics.

Plasmids and cell lines
The extracellular region of chicken N-cadherin fused to a mouse Fc tag (ckN-cad-Fc) was previously stably expressed in CHO K1 cells (Prakasam et al., 2006). Cells were cultured in Dulbecco's modified Eagle's medium (DMEM) containing 10% FBS (Invitrogen) and 0.4 mg/ml G418 (Sigma, St Louis, MO) as a selection marker. The ckN-cad-Fc construct was purified from cell supernatant using a Protein A affinity column (Bio-Rad) followed by gel filtration chromatography, as described previously (Chappuis-Flament et al., 2001;Prakasam et al., 2006). CHO-K1 cells expressing the full-length human N-cadherin and glycosylation mutants of N-cadherin (Guo et al., 2009) were cultured in DMEM containing 10% FBS and 0.4 mg/ml G418. Quantitative flow cytometry and immunoblotting determined the cadherin expression levels.

Quantification of cadherin surface expression levels in CHO cells
Quantitative flow cytometry determined the density of cadherins expressed on the cell surface (Chesla et al., 1998;Chien et al., 2008). CHO-K1 cells expressing human N-cadherin were labeled with mouse, anti-N-cadherin antibody and then with fluorescein-isothiocyanate (FITC)-conjugated anti-mouse IgG antibody (Sigma). The antibody labeling was in phosphate buffered saline (PBS) [137 mM NaCl, 2.7 mM KCl, 10 mM sodium phosphate dibasic, 2 mM potassium phosphate monobasic, 0.5 mM EDTA and 1 w/v bovine serum albumin (BSA) at pH 7.4]. An LSR II flow cytometer (BD Biosciences) measured the fluorescence intensities of labeled cells and of fluorescent bead standards (Zhang et al., 2005). The fluorescence calibration curve was obtained with calibrated FITC-labeled standard beads (Bangs Laboratories, Fisher, IN).

Immobilization of oriented cadherin extracellular domains on erythrocytes
Human whole blood was collected from healthy donors. The whole blood was stored in Vacutainers. The erythrocytes were isolated with Histopaque 1119 solution (Sigma) following the manufacturer's protocol. A mixture of 7 ml whole blood and 7 ml of 0.9% w/v NaCl was added to a 12 ml aliquot of Histopaque 1119 in a 50 ml centrifuge tube. The RBCs were separated by centrifuging the tube at 800 g for 20 minutes at room temperature in an Eppendorf 5810R benchtop centrifuge. The supernatant was discarded as biological waste. The concentrated RBCs at the bottom of the tube were resuspended in 7 ml of 0.9% w/v NaCl, followed by the addition of 1.5 ml of 6% w/v Dextran. The cells were incubated at room temperature for 45 minutes, during which they settled to the bottom of the tube. After discarding the supernatant, the RBCs were washed twice at room temperature with 0.9% w/v NaCl and resuspended in 12 ml EAS45 solution (2 mM adenine, 110 mM dextrose, 55 mM mannitol, 50 mM NaCl, 10 mM glutamine and 20 mM Na 2 HPO 4 , at pH 8.0) (Dumaswala et al., 1996). The purified RBC suspension in EAS45 was maintained at 4˚C for up to 3 weeks, after which the RBCs were treated with bleach and discarded.
Goat polyclonal anti-mouse IgG Fc-specific antibodies (Sigma) were covalently coupled to the RBCs using CrCl 3 activation (Gold and Fudenberg, 1967;Kofler and Wick, 1977). Approximately 10 6 RBCs were washed five times with 0.85% w/ v NaCl, and then resuspended in 250 ml of 0.85% NaCl with 1 mg of the antibody. A CrCl 3 solution was diluted to concentrations below 0.01% w/v in 0.02 mM sodium acetate containing 0.85% w/v NaCl. Next, 250 ml of diluted CrCl 3 solution was mixed with an equal volume of the red blood cell and antibody mixture. After 5 minutes, 500 ml of 'stop solution' (PBS with 5 mM EDTA and 1% BSA) was added, and cells were washed twice with the 'stop solution'. The labeled RBCs were stored in EAS45. The resulting density of antibodies immobilized to the RBC surface was variable, therefore the CrCl 3 concentration was titrated to achieve the desired coverage. After labeling, a sample of 10 5 RBCs was taken, and incubated with 8 ml of 1 mg/ml ckN-cad-Fc. The density of cadherin-Fc bound to the antibody-modified RBCs was determined by quantitative flow cytometry using the same antibody-labeling protocol that was used for the CHO cells expressing Ncadherin (Chesla et al., 1998;Chien et al., 2008).

Immunoblot analysis of cell surface expression levels
Subconfluent cells were washed and detached using 2 mM EDTA. Cells were washed with ice-cold PBS and incubated with 1 mg/ml NHS-LC-biotin in PBS for 20 minutes at 4˚C on a rocking platform. After washing, cells were lysed. Cell surface proteins were precipitated with streptavidin-agarose at 4˚C overnight and detected by western blot as described (Guo et al., 2002).

Micropipette measurements of cell-binding kinetics
The micropipette manipulation technique (Chesla et al., 1998;Chien et al., 2008;Evans et al., 2004;Huang et al., 2010;Zhang et al., 2005) experimentally determines the binding probability as a function of intercellular contact time. The binding probability P(t) is the ratio of the number of binding events n b to the total N T cell-cell touches, n b /N T . The binding kinetics of full-length human N-cadherin expressed on CHO cells was probed with the Fc-tagged extracellular region EC1-EC5 of ckN-cad-Fc immobilized and oriented on the adjacent red blood cell (Fig. 2). The CHO cell expressing membrane-bound wild-type human N-cadherin has been previously described (Gavard et al., 2004). The wild-type human Ncadherin was expressed on CHO cells (hN-CHO WT) at 18 cadherins/mm 2 . The sequences of the N-terminal extracellular domain 1 (EC1) of chicken N-cadherin and of human N-cadherin are 95% identical (BLAST, P10288|165-729 and P19022|160-723).
Throughout the experiment, the cells were maintained in L15 medium (Invitrogen, Carlsbad, CA) supplemented with 1% w/v BSA. Cell-cell contact was observed at 1006 magnification on a Zeiss Axiovert 200 microscope, interfaced with a DAGE-MTI CCD100 CCD camera (DAGE-MTI, Michigan City, IN) to record images. The cells were positioned with a programmable piezo-electric controller, which was used to cyclically bring two cells into contact for a defined period. The contact area was controlled at ,7 mm 2 (,1.5 mm radius). The contact area was photographed several times during an experiment using a Zeiss Axiocam MRm and measured using the distance calibrations included with the Zeiss Axiovision software. Adhesion events are identified from the RBC deformation during cell-cell separation and the recoil at bond rupture. Each cell pair was tested for 50 cell-cell touches (N T ,50), and each contact time represents measurements with at least three cell pairs (N.150). The probability P values are the mean 6 s.d.
In these measurements, it is not necessary to always use identical cadherin densities on either the red cell or on the CHO cell because Equation 1 predicts quantitatively how the kinetics and amplitude of P 1 depend on the receptor and ligand surface densities. This was confirmed in our previous study, which demonstrated that the amplitude of the plateau P 1 varied with the product m R 6m L exactly as predicted by Equation 1 over a wide range of protein surface densities (Chien et al., 2008).

Enzymatic de-glycosylation of cadherin-Fc constructs
The N-cad-Fc constructs were enzymatically treated to remove glycans, by incubating the protein with exoglycosidases (PNGase F, b-1,4-Galactosidase, Neuraminidase) and endoglycosidase (EndoH). The exoglycosidase reactions used 16 G6 buffer (50 mM sodium acetate, 5 mM CaCl 2 , pH 5.5), whereas the EndoH reaction was carried out in 16 G5 buffer (50 mM sodium citrate, pH 5.5). Reactions containing both EndoH and the exoglycosidases were done in 16 G6 buffer. To determine optimum conditions, 5 mg protein was used for each reaction, which was carried out for 24 hours at 37˚C in a water bath. Then 4 ml of 66 SDS Coomassie Blue solution was added to stop the reaction. The contents were heatdenatured and separated on a 7.5% SDS polyacrylamide gel. The combination of all enzymes gave the greatest decrease in molecular mass, so this condition was used to treat 250 mg of ckN-cad-Fc, which was then purified from the reaction mixture on a Protein-A affinity column.

Funding
This research was supported by the National Science Foundation CBET [grant number 0853705].