Structural Determinants of Cadherin-23 Function in Hearing and Deafness

Structures of the cadherin-23 EC1 and EC1+2 repeats

The EC1 repeat and combined EC1+2 repeats of mouse cadherin-23 were refolded from inclusion bodies produced in E. coli (Figure S1A) and used for crystallization and structure determination. The native signal sequence was replaced by a methionine, which is not expected to significantly alter cadherin-23 properties, as the length of the processed N-terminus is not conserved (Figure 1A). Residue numbering throughout the text corresponds to the processed protein (EC1: Q1 to D101; EC1+2: Q1 to D205). Structures of wild-type cadherin-23 repeat EC1 and of repeats EC1+2 at high (1.1 M) Ca²⁺ concentration (“EC1” and “EC1+2 Ca²⁺” in Table 1) exhibit an overall fold closely matching that of classical cadherins: a Greek key motif with seven ß-strands forming a ß-sandwich fold (Figures 1B and 1C). EC1+2 also features a conserved Ca²⁺-binding motif at the linker region with Ca²⁺ ions at sites 1, 2, and 3 bridging acidic residues in canonical sequence motifs (20XEX^BASE22 and 71DRE73 of EC1; 101DXNDN105, 135DXD137, and 185XDX^TOP187 of EC2, Figure 1E; Boggon et al. (2002); Nagar et al. (1996)).

Despite similarities with classical cadherins, several features make cadherin-23 unique. An α-helix between EC1 ß-strands C and D (Figures 1B and 1C) has not been observed in published cadherin EC1 repeat structures. More importantly, cadherin-23's elongated N-terminus includes a 3₁₀ helix within strand A and contributes residues N3 and R4 to an additional Ca²⁺-binding site at the very tip of cadherin-23 EC1, referred to here as site 0 (Figures 1D and S1B). Site 0 is homologous to site 3 in the linker region between EC1+2 repeats (see below) and is further lined by a modified DXD Ca²⁺-binding motif in the loop between ß-strands B and C (36DXDXD40), and an XDX^TOP motif at position 85-87 found in EC1s of classical type-II cadherins, but not type I.

Cadherin-23 EC1 defines the Cr-2 family of cadherin proteins

Placing cadherins into a phylogenetic relationship has been difficult due to the variable number of EC repeats and widely divergent C-termini. Nevertheless, a reasonable phylogeny has been proposed using EC1 sequences to segregate the superfamily into six branches: two branches of true cadherins (C-1 and C-2), a protocadherin branch (Cr-1a), and three other “cadherin-related” branches (Cr-1b, Cr-2, and Cr-3; Hulpiau and van Roy (2009)). In this scheme, cadherin-23 is in branch Cr-2 along with protocadherin-24 and protocadherin-21. Cr-2 and Cr-3 family members sport an N-terminus that is at least six residues longer than classical C-1 cadherins (Figure 2A). All three mammalian Cr-2 family members have all sequence motifs involved in Ca²⁺ binding at site 0. Protocadherin-15, in the Cr-3 group, has an even longer N-terminus, yet it lacks the DXDXD and XDX^TOP motifs of site 0. Instead, protocadherin-15 features two cysteine residues that likely form a disulfide bond at the distal tip in regions aligning closely with cadherin-23 site 0 (Figure 2A). Thus, site 0 is likely a hallmark of Cr-2 family members. Notably, the elongated cadherin-23 N-terminus lacks not only the tryptophan residues at positions 2 and 4, but also the corresponding deep binding pockets that mediate antiparallel binding through strand-exchange in classical type-I and type-II cadherins (Figure 2B). Cr-2 family members must therefore use a different dimerization mechanism.

Cadherin-23's site 0 closely resembles Ca²⁺-binding site 3 of classical cadherins and cadherin-23 itself (Figure 2C), which might suggest that cadherin-23 and Cr-2 family members simply lack a “true” or classical EC1. However, several structural features of cadherin-23 EC1 distinguish it from classical EC2s, namely the subtly different Ca²⁺ coordination (Figures 2D and 2E), the 3₁₀ and α-helices, and the shorter loops between ß-strands B-C and F-G. Thus, cadherin-23 EC1 seems to represent a unique repeat structure, perhaps adapted to perform specific tasks like those of cadherin-23 in the tip link.

Cadherin-23 repeats are stiff and Ca²⁺ determines their mechanical strength

Tip links are constantly subjected to small and large forces (1 to > 100 pN) and have been proposed to function as gating springs (Pickles et al., 1984). Mechanical measurements of hair bundles indicate that the gating spring stiffness is about 1 mN/m (Cheung and Corey, 2005; Howard and Hudspeth, 1988). To test cadherin-23 elasticity and its compatibility with the gating spring model, steered molecular dynamics (SMD) simulations of EC1 and EC1+2 repeats were carried out using explicit water and 150 mM KCl (Grubmüller, 2005; Isralewitz et al., 2001; Sotomayor and Schulten, 2007). Simulations are summarized in Table 2.

Cadherin-23 EC1, including Ca²⁺ at site 0, was equilibrated and stretched from both ends in constant velocity SMD simulations (simulations S1a-g in Table 2). Stretching speeds ranged from 0.1 to 10 nm/ns; in all cases the monitored force increased to several hundred pN with little protein extension until unfolding occurred by rupture of non-covalent interactions at site 0 accompanied by a sudden drop in applied force (Figures 3A, 3B, and S2A; Movie I). The slope for the force versus end-to-end distance during initial stretching stages indicated a stiffness of 710 mN/m per repeat for the slowest SMD simulation. Maximum unfolding-force peak values follow the well-known dependency on stretching speed (Evans and Ritchie, 1997; Izrailev et al., 1997), increasing with faster stretching (Figure 3A inset). Simulations in the absence of Ca²⁺ at site 0 show reduced maximum force peaks across all stretching speeds (simulations S2a-f and S3a-g; Figure 3A inset; Figures S2B and S2C). The dependence of force-peak magnitude on the presence of Ca²⁺ is consistent with Ca²⁺-dependent stabilization of classical cadherins (Prasad and Pedigo, 2005; Sotomayor et al., 2005; Sotomayor and Schulten, 2008), indicating that Ca²⁺ at site 0 may further protect cadherin-23 EC1 from large mechanical stimuli.

The overall elastic response of cadherin-23 is likely to be dominated by the elasticity of single cadherin repeats rather than tertiary structure elasticity, since ultrastructural studies of tip links and TEM images of cadherin-23, within their limitations and possible artifacts, show rather straight filaments (Kachar et al., 2000; Kazmierczak et al., 2007). Nevertheless, some elasticity might arise from mechanical failure of linker regions between repeats. Thus, SMD simulations were performed on cadherin-23 EC1+2 structures (simulations S12a-k in Table 2). Stretching simulations of EC1+2 again indicated a stiff protein, with force increasing to several hundred pN with little protein extension (Figure 3C). The overall stiffness was estimated to be 570 mN/m for the slowest SMD simulation (S12f).

After initial stretching, however, a more complex response was observed, with at least two well-defined peaks at several hundred pN at all stretching speeds (Figure 3C). The first peak correlates with rupture of site 0 as residues N3 and R4 detach, followed by unfolding of the 5-residue 3₁₀ helix within strand A and leading to a 2.5 nm extension that corresponds to the distance between both force peaks. The second force peak involves rupture of the EC1+2 linker region, with either residue E21 and ß-strand A detaching from Ca²⁺ at sites 1 and 2 (simulations S12b,c,f; Figures 3C, 3D, 3F, 3G; Movie II) or ß-strands of repeat EC2 detaching from Ca²⁺ _at site 3 (simulations S12d,e). In both scenarios, the synchronized rupture of Ca²⁺-protein interactions, rapid extension of the protein's end-to-end distance, and drop in stretching forces suggest that Ca²⁺ is essential for holding cadherin-23 EC1 and EC2 repeats together. The Ca²⁺ bond provides stiffness, but at the same time confers “break points” when excessive force is applied. This notion is supported by simulations of wild type and mutant EC1+2 stripped of Ca²⁺ presented below.

To determine whether Ca²⁺ alone can keep EC repeats together, we performed additional SMD simulations (S15b-f) in which the peptide backbone was cut between N103 and D104, disrupting the covalent EC1-to-EC2 linker, but which retained all Ca²⁺ ions bridging acidic residues from both repeats. Forces required to unfold this complex were comparable to those observed for the intact protein (Figure 3E). Furthermore, force peaks originated from the same unfolding and detachment of ß-strands observed for the EC1+2 structure with an intact linker, and separation of EC1 from EC2 was never observed in the simulations (Figures S2D and S2E). Ca²⁺-protein interactions alone are therefore sufficient to keep EC repeats together even under tension, and Ca²⁺ binding must be crucial for cadherin strength.

Structure of cadherin-23 EC1+2 with a Na⁺ bound to its linker region

Structures and simulations highlight the importance of the three Ca²⁺ ions at cadherin-23's linker for its mechanical response. Ca²⁺ is required for tip-link integrity (Assad et al., 1991), but cochlear endolymph surrounding hair cells has an unusually low Ca²⁺ concentration of 20-40 μM (or 100-300 μM in vestibular endolymph; Bosher and Warren (1978); Salt et al. (1989)). Consequently cadherin-23 must have relatively high affinity for Ca²⁺. The E-cadherin EC1+2 linker region has a K_D for Ca²⁺ of 20 μM (Courjean et al., 2008); if the same is true for cadherin-23 (see below), it suggests that its Ca²⁺-binding sites are only partially occupied in vivo. We crystallized wild-type cadherin-23 EC1+2 in conditions containing >2 M NaCl and ~1 mM CaCl₂, and in this second structure observed replacement of Ca²⁺ with Na⁺ at site 1 (Figures 4A and 4C). Assignment of Na⁺ to the electron density at site 1 was supported by comparison of B-factors of the Na⁺ ion and surrounding residues (see “EC1+2 Na⁺” in Table 1; Supplemental Text; Figures S1C and S1D). Overall, the structure closely resembles the first cadherin-23 EC1+2 structure, obtained in >1 M CaCl₂ (RMSD of 0.45 Å for all atoms, Figure 4A). The substitution of Ca²⁺ by a cation of similar ionic radius (Na⁺) at site 1, although an obvious consequence of the crystallization conditions, indicates a likely hierarchy of Ca²⁺-binding affinities with the outermost site 1 as the weakest. This structure led us to test the elasticity and strength of cadherin-23 when Ca²⁺ is not occupying all sites.

Ca²⁺ ions bound to sites 2 & 3 are sufficient to prevent cadherin unfolding

Low Ca²⁺ concentrations (such as those of the endolymph) can be effectively mimicked in simulations by progressively eliminating Ca²⁺ from different binding sites. Thus, we used SMD simulations to test the elasticity of cadherin-23 EC1+2 Na⁺, a model with Ca²⁺ at both sites 1 and 2 replaced by Na⁺, and a model with all binding sites empty.

Simulations with Na⁺ replacing Ca²⁺ at site 1 (S6a-g) show a two-peak force response similar to that of cadherin-23 with Ca²⁺ at all sites (Figure 5A; c.f. Figure 3C). While the first force peak is again associated with rupture of site 0, the second force peak correlated to detachment of ß-strands in EC2 (Figures S2F and S2G), as opposed to detachment of strand A in EC1 in simulations with all Ca²⁺ sites loaded.

In contrast, simulations of EC1+2 with one more Ca²⁺ ion replaced (Ca²⁺ at sites 0 & 3 and Na⁺ at sites 1 & 2; S11), show a single unfolding-force peak that is significantly smaller than those observed for Ca²⁺-loaded cadherin-23. It coincides with rupture of the EC1+2 linker region and ß-strand detachment (simulations S11a-f, Figure 5B; Figures S2H and S2I). Simulations of cadherin EC1+2 with all binding sites empty (S9) show a further decrease in mechanical strength with unfolding at even lower forces. Force peaks correspond to rupture of hydrogen bonds between ß-strands (S9a-f, Figure 5C). Simulations therefore suggest that Ca²⁺ ions at sites 2 and 3 are sufficient to maintain cadherin-23's mechanical strength, which may gradually decrease as Ca²⁺ concentration is decreased.

All these stretching simulations applied force to C_α atoms at the protein's N- and C-termini. This approach may favor unfolding of the terminal ends or the structures that are in line with the ß-strands to which force is applied. To address this concern, a second set of simulations was performed in which force was applied to groups of ~12 C_α atoms at each end of the protein, effectively distributing the applied force over various ß-strands (simulations S12g-k, S6h-l, S11g-j, S9g-k; Figure S3). Such simulations showed that unfolding of the EC1-EC2 linker was similar to previous simulations, and thus that Ca²⁺ stabilization of cadherin-23 under force is robust and independent of the stretching protocol.

Structure of the deafness-related mutant D101G of cadherin-23 EC1+2

Multiple missense mutations in cadherin-23's extracellular domain cause the nonsyndromic deafness DFNB12, and most of these are located at its Ca²⁺-binding motifs (Astuto et al., 2002; Baux et al., 2008; Bolz et al., 2001; Bork et al., 2001; de Brouwer et al., 2003; Ouyang et al., 2005; Roux et al., 2006; Schultz et al., 2005; Schwander et al., 2009; Wagatsuma et al., 2007) (Figure S4). Of these, D101G is the only one located in cadherin-23 EC1+2 (E120Q is located at the EC2+3 linker, unresolved in our structures). In compound heterozygous individuals the D101G mutation together with R2442W causes recessive non-syndromic deafness. The nonsyndromic and progressive characteristics of DFNB12 associated with D101G/R2442W suggest that these mutations cause neither protein misfolding nor mislocalization. Thus, the EC1+2 D101G mutant was generated to gain insight into molecular mechanisms underlying deafness.

The mutant protein exhibited a slightly smaller apparent molecular weight than wild type (Figure S1A) and crystallized in a different space group with two EC1+2 molecules in the asymmetric unit. The individual repeats adopt conformations nearly identical to their wild-type counterparts (Figure 4B; Table 1). However, the arrangement of EC2 with respect to EC1 was different for the two molecules of the asymmetric unit, with one similar to wild type, and the other displaying a different inter-repeat arrangement discussed further below. Remarkably, the mutant protein retains a Ca²⁺ ion at site 2 despite the absence of the Ca²⁺-coordinating D101 side-chain (Figure 4D). Density was observed at the previous location of the aspartate's carboxylate group (Figures S1E and S1F) and was assigned to a Cl^- ion located 2.6 Å from Ca²⁺ at site 2. As in the EC1+2 Na⁺ structure, Na⁺ was modeled at site 1. The D101G structure demonstrates that the mutant protein retains the ability to fold and bind Ca²⁺. The mutation may therefore affect the function of cadherin-23 by one or both of two mechanisms: (i) the mutation may directly affect the unfolding strength and flexibility of the linker region or (ii) the mutation may shift the Ca²⁺ affinity.

Mechanical strength of the deafness-related D101G mutant is altered by decreased Ca²⁺-binding affinity

To understand how the D101G mutation affects cadherin stability, SMD simulations of the D101G mutant structure were performed following the same two stretching protocols used for wild type. For all simulations done with the same ions at sites 1-3, observed forces in the mutant EC1+2 were comparable to those in wild-type (simulations S17a-l, S22a-j, S20a-f; Figures 5D-F, S5A-E, S3E, and S3F). Thus, when Ca²⁺ is bound, the D101G mutant is as strong as wild-type.

The D101G mutation may instead change cadherin-23 strength indirectly, by reducing its affinity for Ca²⁺. A competition assay based on a fluorescent Ca²⁺ indicator was thus used to determine Ca²⁺-binding affinities for cadherin-23 EC1+2 and for EC1+2 D101G (Figure 5G). Fluorescence of a Ca²⁺ indicator mixed with EC1+2, at Ca²⁺ concentrations ranging between 20 and 200 μM, was consistently smaller than fluorescence obtained with the D101G EC1+2 mutant, indicating a stronger Ca²⁺-binding affinity for the wild-type protein. Fitting these data using models with three or four Ca²⁺-binding sites, we found that Ca²⁺ dissociation constants (K_D) are larger for the D101G EC1+2 protein. Dissociation constants for the wild-type protein were (in increasing order) 1.9, 5.0, 44.3 and 71.4 μM, whereas for the D101G mutant the best fit was obtained using a three-site model with K_D values of 3.9, 40.6, >100 μM. Interestingly, the same experiment for the EC1 fragment indicates a K_D of 1.0 μM for site 0 (Figures S5F and S5G). Thus, a tentative assignment of K_D values to Ca²⁺-binding sites is possible: The two lowest K_D values in EC1+2 proteins are likely to be site 0 and the structurally similar site 3; site 1 probably has the highest K_D, as Ca²⁺ at this site is replaced by Na⁺ in the crystals prepared in high Na⁺ (Figure 4C and 4D); and site 2 corresponds to the intermediate K_D values (Table S7).

The lower affinity for Ca²⁺ in the D101G mutant was confirmed using a trypsin sensitivity assay. Significantly higher Ca²⁺ concentrations were required to protect the EC1+2 D101G protein from trypsin digestion than for the wild-type protein (Figures 5H and 5I). Quantification of the trypsin digestion data revealed half-maximal proteolysis rates at 86 μM Ca²⁺ for the wild-type EC1+2 protein, consistent with the K_D values estimated from fluorescence. A Hill coefficient of 3.7 for the fitted curve indicates strong cooperativity in Ca²⁺ binding, which was reduced in the D101G mutant (Hill coefficient 1.3). These results are in agreement with Courjean et al. (2008), where the D103A mutation in E-cadherin EC1+2 (a different aspartate residue in the same DXNDN motif) increased the K_D for Ca²⁺ (from 20 to 240 μM) and eliminated cooperativity. Our data and simulations suggest that the reduced Ca²⁺ affinity of the D101G mutant leads to reduced mechanical strength at the physiological Ca²⁺ concentrations of cochlear endolymph (20-40 μM).

Mechanical strength can be directly altered in other cadherin-23 mutants

Do all deafness mutations in these Ca²⁺-binding sites cause the same structural and functional defects? While structural information on other cadherin-23 EC repeats is not available, sequence alignments can be used to map the mutations either onto our cadherin-23 EC1+2 structure or onto classical cadherin structures, to test their effect on mechanical strength. Among all known cadherin-23 mutations (Figure S4), two classes were selected for in silico studies.

The first modifies the DRE site. The salsa mouse with progressive hearing loss, one of the animal models for human nonsyndromic deafness DFNB12 (Schwander et al., 2009), has the mutation E714V at the EC7+8 linker. A human deafness mutation, E1572K at the EC15+16 linker, modifies the same motif residue. The salsa E-to-V substitution was modeled into the cadherin-23 EC1+2 structure, mapping to residue E73. The resulting model was equilibrated for 10 ns (S25a-b) and subsequently stretched in SMD simulations (S27a-k). This mutant showed a behavior similar to wild-type when Ca²⁺ was loaded at all binding sites (Figures S6A, S6C, S6D, and S6F). Like that of D101G, the phenotype is more likely to arise from changes in Ca²⁺ affinity leading indirectly to altered elasticity and strength.

The second class alters the glutamate of the XEX^BASE motif, e.g., mutations E120Q at the EC2+3 linker and E224K at the EC3+4 linker in humans (Baux et al., 2008; Roux et al., 2006). These correspond to position E21 in cadherin-23 EC1+2, coordinating Ca²⁺ at sites 1 and 2 (Figure 1E). Since either charge neutralization (Q) or reversal (K) causes a phenotype, we chose to evaluate an alanine substitution (E21A), which simply removes the side chain. The E21A mutant shows only one peak in stretching simulations, as opposed to the two peaks observed for wild type, indicating that the mutation directly alters the elastic properties of the protein, reducing its mechanical strength when stretched from C_α atoms (S26a-f, Figures S6B and S6C). The mutation likely affects affinities for Ca²⁺ at sites 1 & 2 as well, which could further impair protein function.

The cadherin-23 EC1+2 background did not affect the behavior of modeled mutations: Equivalent mutations were created in the five-repeat C-cadherin structure and similar behavior was found in simulations (S28 for WT, S29 for E69A, and S30 for E11A; Figures S6G-J).

Thus, deafness mutations targeting Ca²⁺ motifs can affect cadherin-23 in at least two ways: they can reduce Ca²⁺ binding and indirectly affect mechanical strength, or they can modify favorable Ca²⁺-protein interactions to directly reduce the mechanical strength of the protein.

Deafness mutations modulate inter-repeat motion

The two conformations observed in the crystal structure of the D101G mutant (Figure 6A) suggest a third possible effect of deafness mutations: altered inter-repeat orientation and dynamics, that may impair cis and trans interactions. The conformation of classical and tip-link cadherin extracellular domains indeed depends on availability of Ca²⁺, being extended and rigid in its presence and collapsed and flexible in its absence (Cailliez and Lavery, 2005; Kazmierczak et al., 2007; Pokutta et al., 1994; Sotomayor and Schulten, 2008). Equilibrium MD simulations were thus performed to compare inter-repeat motion of wild-type and mutant cadherin-23 EC1+2 repeats in the presence and absence of Ca²⁺.

Simulations of cadherin-23 EC1+2 with Ca²⁺ at all binding sites or with Na⁺ substituting at site 1 show considerable inter-repeat motion (simulations S7a-b, S8a-b, S13a-b, & S14a-b; Figure 6B). Domain motions can be characterized by the tilt angle (θ) between the principal axes of EC1 and EC2 repeats, and the azimuthal angle (φ) of EC2 with respect to EC1 (Figure 6C). Two independent simulations of wild-type EC1+2 with Ca²⁺ at site 1 and two additional simulations with Na⁺ at site 1 display overlapping and similar distributions of tilt and azimuthal angles. Two independent simulations of the D101G mutant show that it explores an overlapping and wider range of values for both θ and φ (S18a-b and S19a-b, Figure 6E), i.e., it both bends and twists more than wild-type.

To rule out the possibility that the Cl^- ion near site 2 in the D101G mutant was responsible for the observed differences, or that the simulations were biased by the initial conditions used, the structure of wild-type EC1+2 with Ca²⁺ at all sites was mutated (D101G) in silico and equilibrated (S24a-c). Inter-repeat motion was similar to that observed for simulations of the D101G cadherin-23 EC1+2 crystal structure (data not shown). Thus, simulated dynamics and inter-repeat motion of the D101G mutant seem robust and distinct from the wild-type. Similar equilibrium MD simulations were performed on the in silico E21A and E73V mutants of cadherin-23 EC1+2 (S23a-c and S25a-c). E21A shows an overall shift in tilt and azimuthal angles, opposite to the shift observed for D101G, whereas the E73V trajectories show a moderately broadened tilt angle distribution but little azimuthal angle preference (data not shown).

If these mutations also compromise Ca²⁺ binding, as shown for the D101G mutant above, an even more pronounced effect in inter-repeat dynamics is expected in low Ca²⁺ environments: In the absence of ions at all binding sites, and regardless of the presence of the D101G mutation, repeats EC1 and EC2 move dramatically with respect to each other (S10a-b & S21a-b; Figures 6D and 6F). Consistently, the extracellular domain of cadherin-23 loses its filamentous shape in the absence of Ca²⁺ to become a chain of randomly oriented repeats (Kazmierczak et al., 2007). The same simulated behavior has been observed for E-cadherin EC1+2 (Cailliez and Lavery, 2005), C-cadherin EC1-5 (Sotomayor and Schulten, 2008), and the type-II classical cadherin cadherin-8 (data not shown). Overall, simulations suggest that mutations and Ca²⁺ binding modulate cadherin inter-repeat orientation and compromise rigidity, which in turn may affect cis and trans dimerization.

Possible cadherin-23 interfaces differ from the strand-exchanged classical cadherin interface

Although cadherin-23 in the tip link is thought to bind to protocadherin-15 in an antiparallel configuration and bind to the partner-strand cadherin-23 in parallel, it may also engage in antiparallel homophilic binding. Cadherin-23 alone has been reported to enable aggregation of cultured cells (Siemens et al., 2004). In addition, hair-cell bundles are splayed in cadherin-23 mutants, indicating a developmental role perhaps independent of tip links (Söllner et al., 2004). The novel structure of cadherin-23 EC1 indicates that interaction with itself or with other proteins like protocadherin-15 must involve a non-classical binding interface. Although EC1 or EC1+2 dimerization was not observed in size exclusion chromatography (Figure S1A), the cadherin-23 crystal packing hints at possible dimer interfaces.

The crystal lattices include symmetric antiparallel dimeric EC1 interfaces, with two different observed arrangements that we labeled ß and W interfaces. The ß interface observed in the EC1 crystal lattice buries 599 Ų of surface area by forming an intermolecular ß-sheet through residues 88-94 of ß-strands G (Figure 7A). In the EC1+2 structures, including the D101G mutant, the opposite faces pack together to form the W interface, burying 480 Ų with contact between conserved tryptophan residues (W65, Figure 7C). Neither of these interfaces is targeted by known deafness mutations and classical cadherin interfaces bury larger surface areas (850 Ų and 1270 Ų for type I and II, respectively). Consistently, SMD simulations that forced unbinding in either the ß or W interface showed an unbinding strength similar than that observed for classical cadherins. However, a single force peak upon unbinding without unfolding was observed in all cases (Figure 7B, 7D and 7E; Figure S7; Movies III, IV, and V; see also Bayas et al. (2004)). While suggestions from crystal packing interactions must be considered cautiously (Bahadur et al., 2004; Kobe et al., 2008), possible cadherin-23 homophilic interfaces identified here are probably weaker than and likely differ from the mechanically strong heterophilic interaction expected with protocadherin-15.

Cloning, expression, and purification of cadherin-23 repeats

Mouse cadherin-23 repeats EC1 and EC1+2 comprising residues Q1 to D101 (Q24 to D124 in NP_075859.2) and Q1 to D205 respectively were sub-cloned into the NdeI and XhoI sites of the vector pET21a. The D101G mutation in EC1+2 was generated using the QuikChange Lightning mutagenesis kit (Stratagene). Cadherin-23 repeats were expressed in BL21CodonPlus(DE3)-RIPL (Stratagene) cultured in LB and induced at OD₆₀₀=0.6 with 100 μM IPTG at room temperature for ~16 hrs. Cells were lysed by sonication in denaturing buffer (20 mM HEPES at pH 7.5, 6 M guanidine hydrochloride, 2 mM CaCl₂, 20 mM imidazole at pH 7.0). The cleared lysates were loaded onto Ni-sepharose (GE Healthcare), eluted with denaturing buffer supplemented with 500 mM imidazole, and refolded by overnight dialysis against 20 mM HEPES at pH 7.5, 150 mM NaCl, 2 mM CaCl₂, and 50 mM KCl using MWCO 2000 membranes. Refolded proteins were further purified by size-exclusion chromatography on a Superdex75 column (GE Healthcare) in 20 mM TrisHCl pH 7.5, 2 mM CaCl₂, and 200 mM NaCl (EC1 and EC1+2 Na⁺) or 150 mM KCl & 50 mM NaCl (EC1+2 Ca²⁺ and EC1+2 D101G) and concentrated by ultrafiltration to 10 mg/ml for crystallization (Vivaspin 10 kDa).

Crystallization, data collection, and structure determination

Crystals were grown by vapor diffusion at 4°C by mixing equal volumes of protein and reservoir solution of (0.1 M sodium cacodylate pH 6.0, 40 % MPD) for cadherin-23 EC1, (0.1 M MES pH 6.5, 1.1 M CaCl₂) for EC1+2 Ca²⁺, (0.1 M Tris HCl pH 8.0, 2.7 M NaCl, 10 % glycerol) for EC1+2 Na⁺, and (0.1 M sodium cacodylate pH 7.1, 1 M NaCl) for cadherin-23 EC1+2 D101G. Cadherin-23 EC1+2 crystals were cryoprotected in reservoir solution plus 25% glycerol. All crystals were cryo-cooled in N₂. X-ray diffraction data were collected as indicated in Table 1 and processed with HKL2000 (Otwinowski and Minor, 1997). The cadherin-23 EC1 and EC1+2 Na⁺ structures were determined by molecular replacement using a cadherin-23 EC1+2 homology model based on E-cadherin EC1+2 (pdb code 1FF5, Pertz et al. (1999)) as a search model with Phaser (McCoy et al., 2007), while cadherin-23 EC1+2 Na⁺ was used for the EC1+2 Ca²⁺ and D101G EC1+2 structures. Model building was done using COOT (Emsley and Cowtan, 2004) and restrained TLS refinement using REFMAC5 (Murshudov et al., 1997). The final models include residues M0 to D101 (EC1), M0 to E207 (EC1+2 Ca²⁺ and EC1+2 Na⁺), and Q1 to H208 (EC1+2 D101G). Data collection and refinement statistics are provided in Table 1. Coordinates have been deposited in the Protein Data Bank with entry codes 2wbx (EC1), 2wcp (EC1+2 Na⁺), 2whv (EC1+2 Ca²⁺), and 2wd0 (EC1+2 D101G).

Ca²⁺-binding affinity measurements and trypsin digestion

F-buffer (100 mM KCl, 10 mM MOPS, pH 7.5) was incubated with 2% Chelex-100 (Bio-Rad) resin in a dialysis bag and stirred for 3 days before use. Cadherin-23 EC1, EC1+2, EC1+2 D101G, and calmodulin (CaM) at 5-10 mg/mL were stripped of Ca²⁺ by three consecutive batch incubations of 100 μl Chelex-100 resin per ml of protein solution for 1 h. This procedure yielded buffer and protein solutions that contained <0.2 μM Ca²⁺ as assessed by Ca²⁺-dependent Fura-2 fluorescence. Protein concentrations were determined by averaging absorbance reads at 280 nm and data from amino-acid analyses. Ca²⁺-competition assays used for the determination of Ca²⁺ dissociation constants (K_D) were modified from André and Linse (2002). Assays at 25°C were performed in cuvettes containing 2 mL of 2.5 μM mag-fluo-4 (Invitrogen) and 3-4.2 μM protein in F-buffer. CaCl₂ solutions were titrated in 5 μl aliquots; after a 10 min incubation the change in fluorescence was recorded on a fluorescence spectrometer (Fluorolog-3, Instruments S. A., Inc.) set at excitation and emission wavelengths of 430 nm and 530 nm, respectively. Experiments were conducted in duplicate and repeated twice. Titration of the chelator in absence of protein was used to determine the apparent K_D of the chelator as 33.88 μM as described in (http://probes.invitrogen.com/media/pis/mp03008.pdf). Fluorescence signals were scaled according to sample volume and fitted with models assuming one, three, or four Ca²⁺-binding sites using the CaLigator software (André and Linse, 2002). For trypsin proteolysis protection assays, 15 ng of trypsin was mixed with 15 μg of decalcified EC1+2 or EC1+2 D101G in 50 μL F-buffer supplemented with Ca²⁺ concentrations ranging from 0 to 1.6 mM and incubated for 1 h at 37°C. Digestions were stopped by addition of PMSF to a final concentration of 1 mM. Samples were analyzed on Coomassie-stained 20% SDS-PAGE gels quantified with ImageJ.

Simulated systems

The psfgen, solvate, autoionize, and mutator VMD (Humphrey et al., 1996) plugins were used to build all systems (Table 2 and Figure S8). Hydrogens were automatically added to protein structures and crystallographic water molecules. Structures with non-native N- and C-terminal tails were modified back to native sequences. Disulfide bonds in C-cadherin were explicitly modeled. Residues D, E, K, and R were assumed to be charged. Histidine residues were assumed neutral and protonation states chosen to favor the formation of evident hydrogen bonds. Additional water molecules and randomly placed ions were used to solvate the systems at the desired KCl concentration (150 mM for cadherin-23 systems). For SMD simulations, molecules were aligned such that the vector joining the terminal residue C_α atoms was oriented along the x-axis.

Molecular dynamics

MD simulations (Karplus and Petsko, 1990) were performed using NAMD 2.6/2.7 (Phillips et al., 2005), the CHARMM27 force field for proteins with CMAP correction (MacKerell et al., 1998, 2004), and the TIP3P model for water (Jorgensen et al., 1983). A 12-Å cutoff (switching function starting at 10 Å) for van der Waals interactions was used along with periodic boundary conditions. The Particle Mesh Ewald method was used to compute long-range electrostatic forces without cut-off and with a grid point density of >1 Å^-3. A uniform 2-fs integration time step was used along with SHAKE. Langevin dynamics was utilized to enforce constant temperature (T = 300 K) when indicated, with a damping coefficient of 0.1 ps^-1 unless otherwise stated. Constant pressure simulations (NpT) at 1 atm were conducted using the hybrid Nosé-Hoover Langevin piston method with a 200 fs decay period and a 50 fs damping time constant.

Simulations and analysis tools

Each system was energy-minimized, then equilibrated in the constant number, pressure, and temperature ensemble (NpT), and the resulting state used to perform subsequent equilibrium and SMD simulations (Tables S1 to S6). Coordinates of all atoms were saved for analysis every picosecond of simulation. Constant velocity stretching simulations used the SMD method and NAMD Tcl Forces interface (Isralewitz et al., 2001; Sotomayor and Schulten, 2008). The stretching direction was set along the x-axis matching the vector connecting terminal regions of the protein. SMD simulations were performed by attaching C_α atoms of N- and C-terminal residues to virtual (independent) springs of stiffness k_s = 1 (kcal/mol)/Ų, or where indicated, by attaching the center of mass of groups of C_α atoms to the same type of virtual springs. In unbinding simulations, virtual springs were attached to C_α atoms of C-terminal residues of independent molecules. The stretching direction was set along the x-axis matching the vector connecting terminal regions of the protein, the free ends of springs were moved away from the protein in opposite directions at a constant velocity, and applied forces were computed using the extension of the virtual springs. Plotted forces correspond to those applied to N-terminal atoms unless otherwise stated. Stiffness was computed through linear regression fits of force-distance plots. Maximum force peaks were computed from 50 ps running averages used to eliminate local fluctuations. End-to-end and CM-to-CM distances were computed as the distances between individual SMD atoms or the center-of-mass of SMD atoms at opposite protein ends, respectively. Regression fits to data points of maximum force peaks versus stretching speeds were performed using a logarithmic expression of the form y = a + b log x. Principal axes of EC repeats were computed using the VMD Orient plugin. Sequence alignments were performed using ClustalX. Structural alignments were performed using VMD/STAMP (Roberts et al., 2006; Russell and Barton, 1992). Plots and curve fits were prepared using xmgrace. Molecular images in this paper were created with the molecular graphics program VMD (Humphrey et al., 1996).