A Novel Family of Cyclic Peptide Antagonists Suggests That N-cadherin Specificity Is Determined by Amino Acids That Flank the HAV Motif*

The classical cadherins (e.g. N-, E-, and P- cadherin) are well established homophilic adhesion molecules; however, the mechanism that governs cadherin specificity remains contentious. The classical cadherins contain an evolutionarily conserved His-Ala-Val (HAV) sequence, and linear peptides harboring this motif are capable of inhibiting a variety of cadherin-dependent processes. We now demonstrate that short cyclic HAV peptides can inhibit N-cadherin function. Interestingly, the nature of the amino acids that flank the HAV motif determine both the activity and specificity of the peptides. For example, when the HAV motif is flanked by a single aspartic acid, which mimics the natural HAVD sequence of N-cadherin, the peptide becomes a much more effective inhibitor of N-cadherin function. In contrast, when the HAV motif is flanked by a single serine, which mimics the natural HAVS sequence of E-cadherin, it loses its ability to inhibit the N-cadherin response. Our results demonstrate that subtle changes in the amino acids that flank the HAV motif can account for cadherin specificity and that small cyclic peptides can inhibit cadherin function. An emerging role for cadherins in a number of pathological processes suggests that the cyclic peptides reported in this study might be developed as therapeutic agents.

The classical cadherins are a family of calcium-binding integral membrane glycoproteins that can account for the ability of cells to segregate from each other during development (1)(2)(3). In this context cadherins have been shown to regulate epithelial (4), endothelial (5), and neural (6 -8) cell adhesion. N-cadherin is predominantly expressed on neural, endothelial, and invasive cancer cells, whereas E-cadherin is predominantly expressed by epithelial cells (9 -11). The classical cadherins generally promote cell adhesion by interacting with one another in a homophilic manner (12,13). In addition, the classical cadherins perform numerous other biological functions (1). For example, E-cadherin has been shown to be capable of acting as a tumor suppressor (14,15), and N-cadherin promotes neurite outgrowth (8).
The classical cadherins are composed of five extracellular domains, a single transmembrane domain, and two cytoplasmic domains (1,16,17). The first extracellular domain (ECD1) 1 of these cadherins contains an evolutionarily conserved His-Ala-Val (HAV) motif (6,18), and several lines of evidence suggest that this sequence is critical for function. In this context, synthetic peptides containing the HAV motif (for example, N-Ac-LRAHAVDING-NH 2 ) have been shown to be capable of inhibiting cadherin-dependent biological processes, such as myoblast fusion (19), neurite outgrowth (8), and embryo compaction (18). Furthermore, antibodies directed against the HAV sequence are also capable of disrupting cadherin-dependent cell adhesion (5,20,21).
Crystal studies have demonstrated that N-cadherin ECD1 monomers will form two types of dimeric structure, one that might reflect a trans adhesion interface, and one that might reflect the formation of a cis dimer (22). Interestingly, the HAV motif is part of a ␤CFG strand that forms one face of the trans adhesion dimer; however, only the histidine and valine make lattice contacts, and these account for less than 5% of the adhesion interface. Furthermore, the alanine at position 80 lines a hydrophobic acceptor pocket for the side chain of tryptophan 2 on a second N-cadherin monomer, and this might underpin the formation of a cis or strand dimer (22). Mutational studies that demonstrate the importance of alanine 80 for stable cell adhesion have been interpreted in support of the hypothesis that formation of the cis dimer is a prerequisite for trans adhesion (23), and this challenges the view that the HAV motif is conserved within cadherins because it is an integral part of the trans adhesion interface (18). However, although there is an emerging view that lateral dimerization of cadherins is required for strong adhesion, structural studies on E-cadherin do not support the strand dimer model (24), and although mutation of alanine 80 inhibits trans binding of E-cadherin, it has no effect on a cis interaction (24). Thus, despite the large number of structural studies and additional mutagenesis data, the precise basis of cadherin homophilic interactions remains contentious (25).
The HAV motif is present in all of the classical cadherins, and it is self-evident that on its own it cannot account for cadherin specificity. One possibility is that the HAV sequence is a primary binding motif for all of the classical cadherins and that selectivity is determined by the flanking amino acids, which differ among the classical cadherins (26,27). In the present study we have addressed this issue by testing a large family of small cyclic peptides for their ability to inhibit N-cadherin function.
We have previously demonstrated that neurons extend longer neurites when cultured on a monolayer of transfected 3T3 cells that express physiological levels of N-cadherin, as compared with untransfected 3T3 cells (8,28,29). This response is driven by the homophilic binding of N-cadherin in the neurons to the transfected N-cadherin present in the 3T3 cell monolayers, and this model is one of the few quantitative assays that measures a biological response to physiological levels of N-cadherin. In the present study we demonstrate that a simple cyclic HAV peptide can act as an N-cadherin antagonist; however, cyclic HAV peptides containing flanking amino acids found in N-cadherin are more potent inhibitors of N-cadherin function. In contrast, cyclic HAV peptides that contain flanking amino acids found in E-cadherin do not inhibit N-cadherin function. Collectively, these data show that the HAV motif is a functional recognition motif within cadherins and demonstrate the pivotal role played by the flanking amino acids in determining cadherin specificity.

EXPERIMENTAL PROCEDURES
Cell Culture and Neurite Outgrowth Assays-Co-cultures of cerebellar neurons on monolayers of control 3T3 cells and monolayers of transfected 3T3 cells that express physiological levels of chick N-cadherin or human L1 were established as described previously (29). In brief, 80,000 3T3 cells (control and transfected) were plated into individual chambers of an 8-chamber tissue culture slide coated with polylysine and fibronectin and cultured in Dulbecco's modified Eagle's medium, 10% FCS. After 24 h, when confluent monolayers had formed, the medium was removed, and 3000 cerebellar neurons isolated from postnatal day 2-3 rats were plated into each well in SATO media (30) supplemented with 2% FCS. All of the test peptides were added immediately before the neurons as a 2ϫ stock prepared in SATO/2% FCS. The co-cultures were maintained for 16 -18 h, at which time they were fixed and immunostained for GAP-43 which is present only in the neurons and delineates the neuritic processes. The mean length of the longest neurite per cell was measured from ϳ150 neurons sampled as described previously (29). The percentage inhibition of neurite outgrowth at various peptide concentrations was calculated as the average of at least three independent experiments. Dose-response curves were plotted, and the IC 50 values (ϮS.E.) were determined.
Peptide Synthesis-All peptides were synthesized using the solidphase method (31,32). The peptides were assembled on methylbenzhydrylamine resin for the C-terminal amide peptides, and the traditional Merrifield resins were used for the C-terminal acid peptides. Acetylation of the N-terminal was performed by reacting the peptide resins with a solution of acetic anhydride in dichloromethane in the presence of diisopropylethylamine after removal of the N-␣-t-butoxycarbonyl by acidolysis using trifluoroacetic acid. All of the cyclic peptides bear the disulfide tether Cys-S-S-Cys. Cyclization was accomplished by reacting the side chain thiol functionalities of the two cysteine residues with a 10% solution of iodine in methanol.
Peptides were generally prepared as a stock solution at a concentration of 5-10 mg/ml in distilled water, and stored in small aliquots at Ϫ70°C. The N-Ac-CHAVDIC-NH 2 peptide (where the underlined residues indicate a cyclic peptide) was made up in tissue culture grade Me 2 SO at a concentration of 20 mg/ml and stored in small aliquots at Ϫ20°C.
Molecular Modeling-Molecular modeling was based on crystals of the ECD1 of N-cadherin, which have been solved as monomeric structures and also found to exist as dimers with protein-protein interfaces that might reflect a cis dimer interface and a trans adhesion interface (Protein Data Base codes 1NCG, 1NCH, 1NCI, and 1NCJ), and on structures of E-cadherin determined by x-ray crystallography (Protein Data Base code 1EDH) and NMR (Protein Data Base code 1SUH). MSI and Swiss PDB software packages were used to isolate the HAV motifs from the intact structures and to make structural and docking orientation predictions for the cyclic peptides. The percentage accessible surface area for individual amino acids at the adhesion interface of the trans dimer of N-cadherin was calculated by submitting the 1NCH structure to the University College London protein-protein interaction server.

Effects of a Cyclic HAV Peptide on N-cadherin Function-We
reasoned that if the HAV motif plays a direct role in classical cadherin function, then peptides that mimic this short sequence should act as N-cadherin antagonists. A short cyclic HAV peptide (N-Ac-CHAVC-NH 2 , see Fig. 1A) was tested for its ability to inhibit the neurite outgrowth response stimulated by N-cadherin. Neurons were cultured on confluent monolayers of control (untransfected) and N-cadherin-expressing 3T3 cells for 16 -18 h. The cells were then fixed, and the length of the longest neurite on ϳ150 neurons was determined by standard assay (see "Experimental Procedures"). Cells cultured on N-cadherin monolayers extended neurites that are on average twice as long as those measured on control 3T3 cells (Fig. 1B). As previously reported (8) a linear peptide that contained flanking amino acids from the N-cadherin sequence (N-Ac-LRAHAVDING-NH 2 ) inhibited the N-cadherin response with an IC 50 of 0.172 Ϯ 0.007 mM (Fig. 1C). The simple cyclic HAV peptide (N-Ac-CHAVC-NH 2 ) also inhibited the N-cadherin response in a dose-dependent manner with a significant inhibition seen at a peptide concentration of 0.22 mM with a more complete inhibition at 0.44 mM (Fig. 1c). An IC 50 value of 0.323 Ϯ 0.020 mM was obtained from the dose-response curve. Neurite outgrowth over control 3T3 cells requires integrin receptor function (33); this was not inhibited by the N-Ac-CHAVC-NH 2 peptide (see legend to Fig. 1), and this demonstrates that the cyclic peptide has no toxic or nonspecific effects on neurite outgrowth. Interestingly, a similar cyclic peptide that lacked the acetyl group (H-CHAVC-NH 2 ) did not inhibit N-cadherin function, and a peptide where the alanine was substituted by a glycine (N-Ac-CHGVC-NH 2 ) also failed to inhibit N-cadherin function (Fig. 1C). These observations demonstrate that a relatively simple cyclic HAV peptide is an effective antagonist of N-cadherin function and that subtle changes in the structure of this cyclic peptide can ablate activity.
The Effect of the Addition of Flanking Amino Acids from N-cadherin to the Cyclic HAV Peptide-In order to assess the effects of modifying the amino acids flanking the HAV sequence on peptide efficacy, a series of cyclic peptides were made that contained flanking amino acids normally found in human N-cadherin (LRAHAVDING) and compared them with the simple cyclic HAV peptide. The results obtained at a peptide concentration of 250 g/ml are shown in Fig. 2. In some instances IC 50 values were obtained from dose-response curves, and these are given where appropriate. Extending the sequence by three amino acids at the N terminus (to give N-Ac-CLRA-HAVC-NH 2 ) resulted in a peptide with no activity (Fig. 2B). A negligible effect was apparent when a single amino acid was added at the N terminus (N-Ac-CAHAVC-NH 2 ) (Fig. 2C, IC 50 0.36 Ϯ 0.003 mM). In contrast a positive effect was apparent with the addition of a single amino acid to the C terminus (N-Ac-CHAVDC-NH 2 ) (Fig. 2D), and this is more clearly seen in a full dose-response curve (Fig. 3, IC 50 0.114 Ϯ 0.006 mM). Interestingly, addition of 1 or 2 amino acids back to the N terminus (N-Ac-CAHAVDC-NH 2 and N-Ac-CRAHAVDC-NH 2 ) generated inactive peptides (Fig. 2, E and F); however, a peptide with three additional amino acids (N-Ac-CLRAHAVDC-NH 2 ) was active (Fig. 2G). Further amino acid additions to the C terminus could restore activity to an otherwise non-active peptide. For example, addition of an isoleucine to an inactive peptide (N-Ac-CAHAVDC-NH 2 , Fig. 2E), generated a reasonably good N-cadherin antagonist (N-Ac-CAHAVDIC-NH 2 , Fig.  2H) with an IC 50 value of 0.210 mM. As with the simple cyclic HAV peptide, the acetyl group on the N-Ac-CAHAVDIC-NH 2 peptide was essential for activity (Fig. 2I). In summary, five peptides have been identified that can substantially inhibit N-cadherin function when tested at 250 g/ml (see Fig. 2); however, the N-Ac-CHAVDC-NH 2 was the only peptide that retained activity when tested at 125 g/ml.

Further Investigation of the Effects of the Flanking Amino
Acids at the C-terminal Side of the HAV Motif-The results in Fig. 2 demonstrate that addition of an aspartic acid to the C terminus increases the activity of the simple cyclic HAV peptide, whereas addition of an alanine to the N terminus decreases activity. Based on this result, a number of peptides  3. Dose-response curves for the most active N-cadherin antagonists. Two cyclic peptides, N-Ac-CHAVDC-NH 2 (q) and N-Ac-CHAVDIC-NH 2 (E), and a linear version of one of these peptides, N-Ac-HAVDI-NH 2 (ƒ), were tested at the given concentrations for their ability to inhibit the N-cadherin response (determined as in Fig. 1). The results show the percent inhibition of the N-cadherin response, and each value is the mean (ϮS.E.) from three experiments. None of the peptides had a significant effect on neurite outgrowth over the control 3T3 cells (data not shown).
were made that were progressively extended by single amino acids only at the C-terminal side of the HAV motif. Full doseresponse curves are shown for the three most active peptides in Fig. 3. The cyclic HAVD peptide was substantially more active than the cyclic HAV peptide (IC 50 0.114 Ϯ 0.006 mM as compared with 0.323 mM). Activity was further increased by the addition of an isoleucine to give N-Ac-CHAVDIC-NH 2 (IC 50 of 0.065 Ϯ 0.005 mM), but there was no further benefit when an asparagine was added (IC 50 of 0.110 Ϯ 0.006 mM, data not shown). A peptide with an additional glycine (N-Ac-CHAVDINGC-NH 2 ) was inactive (a 1.5 Ϯ 2.2% inhibition of the N-cadherin response at 125 g/ml, mean Ϯ S.E., n ϭ 3). There was an approximate 7-fold reduction in the activity of the most active analogue when it was tested as a linear peptide (N-Ac-HAVDI-NH 2 ) (IC 50 0.44 Ϯ 0.014 mM, Fig. 3), and this aptly demonstrates the benefit of incorporating this sequence into a cyclic peptide. An alanine 3 glycine mutation within the most active peptide (to give N-Ac-CHGVDIC-NH 2 ) led to a dramatic loss of activity, with only a 22.0 Ϯ 4.1% inhibition of the N-cadherin response being found at a peptide concentration of 250 g/ml (ϳ0.32 mM).
The Effect of the Addition of Flanking Amino Acids from E-cadherin to the Cyclic HAV Peptide-A series of E-cadherin peptide mimetics were made by adding flanking amino acids from human E-cadherin to the cyclic HAV peptide. The results in Fig. 4 demonstrate that the E-cadherin peptides either did not inhibit the N-cadherin response (compounds N-Ac-CSHAVC-NH 2 , N-Ac-CHAVSC-NH 2 , N-Ac-CSHAVSC-NH 2 , N-Ac-CSHAVSSC-NH 2 ) or had little effect (compound N-Ac-CHAVSSC-NH 2 ). It is important to note that addition of E-cadherin flanking sequences to the HAV motif always resulted in a loss of activity relative to the simple cyclic HAV peptide.
Effects of HAV-containing Peptides on the L1 Response-The L1 cell adhesion molecules can stimulate neurite outgrowth (34). This response requires the homophilic binding of L1 in neurons to transfected L1 in the 3T3 cells, and following homophilic binding the response is driven by the same FGF receptor-dependent signaling cascade as the N-cadherin re-sponse (28,34). In order to ascertain the specificity of the most active N-cadherin antagonists, cerebellar neurons were cultured over either control 3T3 cell monolayers or monolayers of 3T3 cells that express physiological levels of human L1 in the presence and absence of the test peptides. As previously reported, L1 stimulates neurite outgrowth from cerebellar neurons with an approximate doubling of neurite length measured in this set of experiments (Fig. 5A). The L1 response was not significantly diminished in sister cultures treated with 125 g/ml of the N-Ac-CHAVDC-NH 2 peptide (Fig. 5B), the N-Ac-CHAVDIC-NH 2 peptide (Fig. 5C), or the N-Ac-CHAVDINC-NH 2 peptide (Fig. 5D). DISCUSSION Despite an extensive series of structural studies on N-and E-cadherin, there remains no consensus on the mechanisms that govern the specificity of cadherin binding (25). Although peptide competition studies support the notion that the HAV motif plays a direct role in binding (8,18), the peptides that were used in the studies contained a number of amino acids other than HAV, and it could be argued that these might have been responsible for the inhibition. Furthermore, although mutation of alanine 80 inhibits stable trans adhesion, this might be an indirect effect as this mutation can prevent the docking of tryptophan 2 from either the same or a second ECD1 monomer into a hydrophobic pocket lined by alanine 80 (23,24).
If the HAV motif is directly involved in trans adhesion, peptide mimetics might be expected to act as cadherin antagonists. One way to increase the efficacy of peptide inhibitors is to make them cyclic and thereby constrain structure. In the present study we have shown that a relatively simple cyclic HAV peptide (N-Ac-CHAVC-NH 2 ) can inhibit the N-cadherin component of neurite outgrowth over 3T3 cells in the absence of any effect on the integrin component of growth over control 3T3 cells. This demonstrates that the cyclic peptide is an effective N-cadherin antagonist and suggests that the cyclic peptide can adopt an active binding conformation. Interestingly, a cyclic HAV peptide with a free amino group at the N-terminal region (H-CHAVC-NH 2 ) was inactive, as was a peptide with an alanine 3 glycine substitution (N-Ac-CHGVC-NH 2 ). These observations highlight the fact that very subtle structural modifications of a cyclic peptide can affect efficacy, presumably by constraining the peptide in a manner that restricts adoption of the active configuration.
The specificity of cadherin binding can be dramatically altered by mutating residues that immediately flank the HAV motif in cadherins. For example, if the amino acids at position Ϫ1 and ϩ2 (relative to HAV) in E-cadherin are replaced by P-cadherin residues, the mutant protein acquires the ability to bind P-cadherin (26). In the present study we made the observation that specificity can also be built into short cyclic peptides. In this context, the addition of a single amino acid to HAV that results in an N-cadherin-specific sequence (N-Ac-CHAVDC-NH 2 ) generated a much more potent N-cadherin antagonist (a decrease in the IC 50 from 0.32 to 0.114 mM), with a clear benefit being seen on further extending the peptide to N-Ac-CHAVDIC-NH 2 (IC 50 , 0.064 mM). In stark contrast, modifications to the amino side of the HAV motif had little effect on peptide activity or were detrimental. For example, removal of the acetyl group from the N-Ac-CHAVC-NH 2 or the N-Ac-CAHAVDIC-NH 2 peptides resulted in a complete loss of antagonistic activity. Likewise, addition of an alanine to the N-Ac-CHAVDC-NH 2 peptide (to give N-Ac-CAHAVDC-NH 2 ) also resulted in a loss of activity. Interestingly, addition of an alanine had little effect when added to the N-Ac-CHAVDIC-NH 2 peptide (to give N-Ac-CAHAVDIC-NH 2 ). Overall, the picture that emerges is one where there appears to be a tussle between the forces on the N and C terminus of the HAV motif in relation to peptide activity, and this initially suggested to us that much of the structure-function relationship between this family of peptides might be accounted for by the precise ordering of the side chains of the histidine and valine within the cyclic peptide (however, see below). The specificity of this class of peptide is reinforced by the observation that the most active N-cadherin antagonists do not inhibit neurite outgrowth stimulated by L1, despite the fact that following the homophilic binding step, both molecules promote neurite outgrowth by activating the same FGF receptor-dependent signal transduction cascade (28,29).
Our results have shown that when we turn the cyclic HAV peptide into an N-cadherin peptide by incorporating flanking amino acids that are specific to N-cadherin, we can substantially increase the antagonistic properties of the peptide. Remarkably, we find the opposite result when we turn the peptide into an E-cadherin peptide. Whereas addition of N-cadherin amino acids at position Ϫ1 or ϩ1 (relative to HAV) has little effect or increases peptide activity, addition of the corresponding amino acids from E-cadherin (to give N-Ac-CSHAVC-NH 2 and N-Ac-CHAVSC-NH 2 ) generated peptides that no longer inhibit N-cadherin function. Likewise, three other cyclic HAV FIG. 6. Crystal and model structures of the HAVDI motif. The HAVDI segment of N-cadherin, isolated from the rest of the structure of the trans dimer crystal (Protein Data Bank code 1NCH) is shown in A. A model of the cyclic N-Ac-CHAVDIC-NH 2 peptide, based on a possible backbone/side chain configuration, demonstrates that the histidine and valine side chains are able to adopt a similar structure (B). The docking of the isolated HAVDI segment of one N-cadherin monomer to the surface of the second N-cadherin monomer, as observed in the trans dimer crystal, is shown in C. A possible docking conformation of the cyclic N-Ac-CHAVDIC-NH 2 peptide, biased toward the histidine-and valine-binding sites on the surface of the second monomer, is shown in D. E, the HAV motifs from N-cadherin and E-cadherin were isolated by superimposing the intact domains of each crystal (Protein Data Bank codes 1NCH and 1EDH) onto each other by minimizing root mean square differences in the side chain configurations using Swiss PDB software. The HAV motifs are closely aligned, and this demonstrates an essentially identical structure in both crystals. In a similar vein, the HAV motif was isolated from the monomeric, trans, and cis dimer crystals of N-cadherin (Protein Data Bank codes 1NCG, 1NCH, and 1NCI); the fact that these are closely aligned demonstrates that the structure of the HAV motif does not change following homophilic binding.
peptides that contained flanking amino acids from E-cadherin had little or no effect on the N-cadherin response. The same panel of cyclic peptides have been tested in assays that measure E-cadherin function, in contrast to the above results the E-cadherin set of peptides inhibit E-cadherin function. and the N-cadherin set of peptides have little or no effect. 2 Collectively, these results suggest that the HAV sequence is a general classical cadherin-binding sequence and that specificity is determined by the amino acids that immediately flank this motif in both N-cadherin and E-cadherin.
The question of how the HAV-flanking amino acids induce specificity within the cyclic peptides and cadherins themselves is fundamental to understanding cadherin function. In this context, there would seem to be no problem in the side chains of the HAV motif within the most active cyclic peptide adopting a similar structural configuration to that found in the intact ECD1 (Fig. 6, A and B), and the cyclic peptide might be expected to antagonize homophilic binding by competing for the histidine-and valine-binding sites on the adhesion face of ECD1 (Fig. 6, C and D). The possibility that the flanking amino acids confer specificity by differentially orienting the side chains of the histidine and valine within E-and N-cadherin is attractive; however, there appears to be no obvious difference in the structure of the HAV motif in E-and N-cadherin ( Fig.  6E) nor any evidence for a change in conformation of the HAV motif within N-cadherin following homophilic binding (Fig. 6F). An alternative possibility is that the flanking amino acids contribute directly to the binding energy. In this context, the aspartic acid and isoleucine of the HAVDI motif both contribute to the adhesion face, with the isoleucine in particular accounting for a significant percentage of the interface surface (7.45%). Although this might explain why the cyclic HAVDI peptide is a very good N-cadherin antagonist (IC 50 ϳ0.064 mM) relative to the cyclic HAV peptide (IC 50 ϳ0.32 mM), it does not explain why the cyclic HAVS peptide has no activity when tested at up to 250 g/ml. One possibility is that flanking amino acids can additionally hinder the docking of the histidine and valine side chains to an inappropriate cadherin.
In summary, a novel family of cyclic peptides containing the HAV motif has been developed as N-cadherin antagonists. Our studies indicate that specific N-cadherin antagonists can be developed based on the incorporation of 1 or 2 flanking amino acids from native N-cadherin onto the HAV motif. The results obtained from this study demonstrate the importance of the flanking amino acids in determining the specificity of cadherin interactions and provide novel insights into the mechanisms that govern the specificity of cadherin binding.