Crystal Structure of the Shank PDZ-Ligand Complex Reveals a Class I PDZ Interaction and a Novel PDZ-PDZ Dimerization*

The Shank/proline-rich synapse-associated protein family of multidomain proteins is known to play an important role in the organization of synaptic multiprotein complexes. For instance, the Shank PDZ domain binds to the C termini of guanylate kinase-associated proteins, which in turn interact with the guanylate kinase domain of postsynaptic density-95 scaffolding proteins. Here we describe the crystal structures of Shank1 PDZ in its peptide free form and in complex with the C-terminal hexapeptide (EAQTRL) of guanylate kinase-associated protein (GKAP1a) determined at 1.8- and 2.25-Å resolutions, respectively. The structure shows the typical class I PDZ interaction of PDZ-peptide complex with the consensus sequence -X-(Thr/Ser)-X-Leu. In addition, Asp-634 within the Shank1 PDZ domain recognizes the positively charged Arg at –1 position and hydrogen bonds, and salt bridges between Arg-607 and the side chains of the ligand at –3 and –5 positions contribute further to the recognition of the peptide ligand. Remarkably, whether free or complexed, Shank1 PDZ domains form dimers with a conserved βB/βC loop and N-terminal βA strands, suggesting a novel model of PDZ-PDZ homodimerization. This implies that antiparallel dimerization through the N-terminal βA strands could be a common configuration among PDZ dimers. Within the dimeric structure, the two-peptide binding sites are arranged so that the N termini of the bound peptide ligands are in close proximity and oriented toward the 2-fold axis of the dimer. This configuration may provide a means of facilitating dimeric organization of PDZ-target assemblies.

Multidomain Shank, proline-rich synapse-associated protein, and somatostatin receptor-interacting protein scaffold proteins bind to various membrane and cytoplasmic proteins within the PSDs 1 in excitatory synapses (1,2). It has been suggested that Shank links N-methyl-D-aspartate receptor-PSD-95 complexes to the actin cytoskeleton, thereby playing a critical role in the organization of cytoskeletal signaling complexes at excitatory synapses (1,2). The three known members of the Shank family (Shank1-3) all contain multiple sites for alternative splicing and show distinct tissue distributions (2). Although shank proteins vary in molecular mass, they share a common domain organization consisting of seven N-terminal ankyrin repeats followed by an SH3 domain, a PDZ domain, a long proline-rich region, and a SAM domain. All of these motifs are potentially involved in protein-protein interactions. For instance, the proline-rich region commonly acts as a binding site for SH3, EVH1, and WW domains and SAM domains can bind to each other in homomeric and heteromeric fashion, enabling oligomerization of Shank and its interacting proteins (3).
PDZs are globular domains containing ϳ80 -100 amino acids (4). The Shank PDZ domain is a class I PDZ recognizing the C-terminal sequence X-(Thr/Ser)-X-Leu (where X represents any amino acid), which enables it to bind a variety of integral membrane proteins; however, it most specifically binds to the C terminus of GKAP, which in turn interacts with the guanylate kinase domain of PSD-95 (5). These interactions may be involved in the synaptic targeting and cytoskeletal attachment of receptors, linking them physically and functionally to the appropriate intracellular signaling pathways (5). In addition, an interaction between the Shank PDZ and the C-terminal PDZ binding motif and leucine zipper domain of ␤PIX was reported recently (6). ␤PIX is a guanidine nucleotide exchange factor that binds p21-activated kinase and promotes the functional coupling of Rac1 and Cdc42 small GTPases with downstream effector kinases (7)(8)(9).
The aim of the present study was to better understand the structural mechanism of the interaction between the Shank1 PDZ and its target protein. To that end, we determined the crystal structures of the Shank1 PDZ domain in its peptide-free form and in complex with the C-terminal hexapeptide of GKAP to resolutions of 1. 8  BL21(DE3) and then purified and crystallized as described previously (10). The C-terminal hexapeptide (EAQTRL) of GKAP used for PDZpeptide cocrystallization was chemically synthesized (Anygen).
Data Collection-A MAD data set was collected from a peptide-free frozen crystal to a resolution of 1.8 Å using an ADSC Quantum 4R CCD detector at beamline 18B at the Photon Factory. Because bromine is a convenient anomalous scatterer for MAD phasing, a bromine MAD data set was collected from a single crystal crystallized in a solution containing 200 mM NaBr (11). The diffraction data from a PDZ-peptide complex to a resolution of 2.25 Å were collected at 100 K at beamline 6B of the Pohang Accelerator Laboratory.
Structure Determination and Refinement-The structure of the peptide-free PDZ domain was determined by MAD phasing using bromine as an anomalous scatterer. The positions of three bromines within the asymmetric unit were located and refined using the program SOLVE (12). The initial phases were improved by solvent flattening using the program RESOLVE (overall figure of merit, 0.52) (13). The resultant map, showing a dimer in an asymmetric unit, was readily interpretable. Model building then proceeded using the program O (14), after which the structure was refined using the program CNS (15). The final crystallographic R value for the peptide-free PDZ model, calculated using data from 27.7-1.8 Å, was 23.4% (R free ϭ 25.1%).
Using the structure of the peptide-free PDZ dimer served as a starting model, the structure of the PDZ-hexapeptide complex was determined by molecular replacement methods with the program MOLREP (16). There were two PDZ domains related by non-crystallographic 2-fold symmetry within the asymmetric unit. After applying a simulated annealing procedure using data to 2.25 Å, the locations of the two bound peptides were determined from a F o Ϫ F c difference electron density map. The electron densities for all of the six residues of the peptide ligands were obvious, indicating that they were well ordered within the structure. The hexapeptide was modeled using program O, after which the ligand-bound PDZ domain was refined to a final crystallographic R value of 25.1% and a free R value of 28.0%. The refined model of the peptide-bound PDZ domain consisted of a PDZ dimer related by a non-crystallographic 2-fold axis, two hexapeptides, and 44 water molecules. The stereochemistry of the model was analyzed with PROCHECK (17). No residues were found in the disallowed regions of the Ramachandran plot. Data collection and refinement statistics are summarized in Table I.

RESULTS AND DISCUSSION
Overall Structure of the Shank1 PDZ Domain-The Shank1 PDZ monomer is a compact, globular domain containing eight segments of secondary structure: six ␤ strands that form an antiparallel ␤ barrel and two ␣ helices (Fig. 1A). The amino acid sequences of all of the Shank protein PDZ domains are nearly identical but differ significantly from other PDZ domains ( Fig. 1B) (18,19). The amino acid sequence of the rat Shank1 PDZ domain (residues 582-690) used in this study is identical to the human domain and shares 88 and 77% sequence identity with the PDZ domains of rat Shank2 and Shank3, respectively. The overall fold of the Shank1 PDZ shows the highest degree of structural variation of any known PDZ (Fig. 1C). In addition, the Shank1 PDZ contains exceptionally long N-terminal ␤A and C-terminal ␤F strands, which span more Ͼ11 residues each and participate in antiparallel ␤ sheet interactions. The ␤A strand is twisted and participates in dimeric interaction by forming an antiparallel ␤ sheet with the ␤A strand of the other monomer (Fig. 1C). The PDZ domain segment that displays the most conformational variation is the loop connecting the ␤B and ␤C strands. The Shank1 PDZ contains the longest known ␤B/␤C loop, spanning 19 residues. Despite the variability in the sequence and length of ␤B/␤C loops, the sequence is strictly conserved among Shank PDZ domains.
As is typical of most PDZ complexes, the peptide ligand is positioned within a groove between the ␤B strand and the ␣B helix, oriented as an additional strand antiparallel to ␤B. Within the crystal structure of the complex, all six amino acid residues of the peptide ligand were well defined as shown by the difference electron density map calculated without inclusion of the peptide. Superposition of the structures of the peptide-free and peptide-bound PDZ domain shows that there is little conformational change upon ligand binding. The C ␣ root mean square (r.m.s.) deviation between two models is only 0.69 Å.
Structural Basis for the Specificity of the Shank1 PDZ-GKAP Interaction-PDZ domains typically bind to the last several residues at the C terminus of interacting proteins. Shank PDZ domains are known to bind to the C-terminal six residues of GKAP1a, which in turn associates with the guanylate kinase domain of PSD-95 (20). Although they recognize the typical consensus sequence X-(Thr/Ser)-X-Leu and are classified as type I PDZ domains, Shank PDZ domains show higher specificity for the GKAP C-terminal EAQTRL sequence. The interaction with GKAP is reportedly specific for the Shank PDZ, because GKAP1a (residues 591-666) does not associate with the other class I PDZ domains (e.g. those of PSD-95, Chapsyn-110/PSD-93, or calmodulin-dependent serine kinase) (3). Moreover, the C-terminal tail of Kv1.4 (ETDV) does not bind the Shank PDZ, even though it interacts with the PDZ domains of PSD-95 (3). The Shank PDZ domain is thus optimized for recognition of GKAP, even though it marginally binds other ligands via the consensus sequence.
Within the crystal structure, the peptide ligand EAQTRL inserts into the binding pocket antiparallel to the ␤B strand, forming an extensive network of hydrogen bonds and hydro- phobic interactions (Fig. 2, A and B). The GFGF sequence in the carboxylate binding loop of the Shank1 PDZ forms hydrogen bonds with the carboxyl group of the GKAP peptide C terminus in a manner similar to that seen previously in PDZligand complexes. The backbone amides of Phe-602, Gly-603, and Phe-604 of the carboxylate binding loop are involved in hydrogen bonding with the C terminus of the ligand, whereas the side chain of Leu-0 enters the deep hydrophobic cavity formed by Phe-602, Phe-604, Leu-606, and Ile-670. Instead of the side chain of the penultimate residue of the ligand being oriented toward the solvent, the Shank1 PDZ interacts directly with the Ϫ1 position of the ligand (Fig. 2B). The carboxyl group of Asp 634 located at the end of the ␤C strand forms a salt bridge with the guanido group of Arg Ϫ1 . To do so, the long side chain of Arg Ϫ1 twists at the C␤ atom toward Asp 634 . Sequence alignment shows that Asp 634 is strictly conserved among all of the Shank proteins characterized so far (rat Shank1-3 and human Shank1 and Shank2). Moreover, all of the known GKAP family proteins (GKAP/SAP90/PSD-95-associated protein 1-4) terminate with the same four amino acids (ϪQTRL) (21). This finding suggests that a positively charged residue is greatly preferred at ligand position Ϫ1 because it directly contributes to the specificity of the PDZ-ligand interaction. Consistent with that idea, substituting Arg Ϫ1 with Glu significantly reduced the affinity of the interaction (3). The preference for a positively charged residue at the Ϫ1 position was previously reported in the NHERF PDZ1-cystic fibrosis transmembrane conductance interaction where Glu 43 serves a function equivalent to that of Asp 634 in the Shank PDZ (22). In addition, Erbin PDZ domain selectively accommodates ligands with a bulky hydrophobic side chain of tryptophan residue at the Ϫ1 position, revealing the contribution of penultimate residues to the optimal binding of the ligands (23).
In class I PDZ domains, serine or threonine at the Ϫ2 position of the ligand forms a hydrogen bond with the side chain of residues at the start of helix ␣B. Likewise, within the Shank PDZ-ligand complex, the threonine residue at ligand position Ϫ2 contributes to the specificity by hydrogen bonding with His 663 at ␣B1 (Fig. 2B).
The ligand residues at the Ϫ3 and Ϫ5 positions also appear to contribute to the specificity and affinity of the interaction. Involvement of the residues at the Ϫ3 position in the PDZligand interaction was reported in diverse types of PDZ domains (22)(23)(24)(25)(26). In Shank PDZ domain, the guanido group of Arg 607 in the ␤B strand is positioned within the groove formed by the protrusions of the side chains of GlnϪ 3 and Glu Ϫ5 . The O⑀ atom of Gln Ϫ3 forms hydrogen bonds with the Arg 607 guanido group, and Glu Ϫ5 interacts both with the side chain of Arg 607 and the hydroxyl group of Tyr 629 . This means that the presence of Arg 607 may further stabilize the interaction by hydrogen bonding with the hydrophilic residues at positions Ϫ3 and Ϫ5, even though interactions at those positions are not essential for ligand binding.
Dimerization of PDZ Domains-Several well characterized PSD scaffold proteins, including PSD-95 (27,28), the Homer family of proteins (29), and GRIP/AMPA receptor-binding pro- tein (30), have the ability to form homomultimers or heteromultimers. And in some cases, such multimerization is mediated by the PDZ domains (31)(32)(33). For example, this was recently seen in the crystal structure of GRIP PDZ6, which shows a tightly associated dimer (34). In Shank proteins, the SAM domain reportedly multimerizes, which is sufficient for self-association of full-length Shank proteins; however, the present study shows that the PDZ domain also contributes to the multimerization of Shank proteins, suggesting structural conservation of a dimerization mode mediated by PDZ domains.
The crystal structures revealed tightly associated dimers in the asymmetric units of both the peptide free and peptide bound crystals, which belong to the P2 1 and P4 1 2 1 2 space groups, respectively (Fig. 2C). Each asymmetric unit contained two molecules forming a dimer with non-crystallographic 2-fold symmetry. Cross-linking experiments and size-exclusion chromatography confirmed the Shank1 PDZ to be a dimer in solu-tion (Fig. 2, D and E). That the Shank1 PDZ domain has the ability to form a dimer whether it is free or complexed with a ligand suggests the dimer may represent the functional state of the Shank PDZ domain.
The interface between dimeric PDZ domains involves a ␤A strand and a ␤B/␤C loop from each monomer (Fig. 3, A and C). The ␤A strands form antiparallel ␤-sheets around the center of a 2-fold axis so that the N and C termini of each PDZ domain point in opposite directions. The amount of surface area buried upon dimer formation is 611.6 Å 2 or 9.5% of the total surface area of each monomer, which implies the existence of specific contacts between the two monomers that may have biological significance. The dimeric interface is composed of 64.9% of non-polar atoms and 35.0% of polar atoms. Within the dimer, the monomers are held tightly by six hydrogen bonds, four water-bridged hydrogen bonds, and numerous van der Waals interactions. All of the six hydrogen bonds originate from backbone atoms in the ␤A strands forming the antiparallel ␤-sheet. In contrast, mostly side chain atoms in the ␤B/␤C loops, which comprises 41% of the interface area, participate in the hydrophobic interactions at the interface. That residues 613-625 in the otherwise variable ␤B-␤C loop are strictly conserved in all of the Shank proteins identified so far implies a potential role for the ␤B/␤C loop in both dimerization and protein-protein interactions. Recently, the crystal structure of Erbin PDZ⅐ErbB2 complex revealed the important function of the loop ␤B-␤C in ligand interactions, implying the possibility that the variable loop ␤B-␤C could be exploited for the specific function of the PDZ domain (24).
Similar to Shank PDZ, GRIP PDZ6 dimerizes by forming antiparallel ␤-sheets with N-terminal ␤A strands (Fig. 3, A and  B), although the configurations of the ␤A strands differ in the two dimers (34). Shank PDZ has a longer more twisted ␤A strand, and half of the strand is involved in the ␤-sheet interaction causing the peptide binding pockets to be oriented differently. In both dimers, the peptide binding pockets are located at the distal sides of the complex and oriented in antiparallel fashion. In the Shank PDZ dimer, however, the N termini of the ligands are oriented toward the center so that they are in close proximity around the 2-fold axis of the dimer (Fig. 3A).
Biological Implications of Shank PDZ Dimer-Shank proteins have a SAM domain at their C terminus, which can mediate oligomerization of the proteins. Sharpin, which interacts with Shank anykrin repeats, has a coiled-coil domain that contributes to the multimeric association of the Shank-Sharpin complex (35). Notably, the crystal structure of the Shank1 PDZ domain suggests that the dimeric configuration of the PDZ domain may facilitate multimeric organization of Shank proteins. Although multimeric association of GKAP and Shank PDZ has yet to be reported, the interaction of ␤PIX and Shank PDZ provides some insight to the biological significance of the Shank PDZ dimer. The Shank1 PDZ domain reportedly interacts with the C-terminal domain of ␤PIX, and the leucine zipper domain mediates the homodimerization of ␤PIX (6, 36). This means that both the C-terminal PDZ binding motif and the preceding leucine zipper domain of ␤PIX are involved in the B, ribbon diagram of the GRIP PDZ6 dimer (Protein Data Bank code 1N7F). C, Shank PDZ dimer interface. Side chains of hydrophobic residues at the interface are shown as ball and stick models. D, amino acid sequence and predicted secondary structure of the C-terminal domain of ␤PIX. The secondary structure was predicted using the PredictProtein Web server (cubic.bioc.columbia.edu/predictprotein/). E, proposed model of the Shank1 PDZ and ␤PIX C terminus complex. The figures were made using PyMOL (www.pymol.org).
interaction and that an additional interaction besides that taking place at the peptide binding pocket of the PDZ domain is involved in the binding of ␤PIX (6). The leucine zipper motif commonly forms a coiled-coil homodimer of two ␣-helices terminating at their C termini with a 2-fold symmetry. Prediction of ␤PIX secondary structure indicates that six amino acid residues connect the C-terminal PDZ binding motif and the preceding leucine zipper helix (Fig. 3D). As mentioned above, the ligands binding to the PDZ are in close proximity around the 2-fold axis of the dimer. And given the strict conservation of the ␤B/␤C loop among Shank proteins, there are few tertiary configurations available to the PDZ-␤PIX complex. With that in mind, we propose a model of the Shank PDZ-␤PIX C-terminal complex in which the C-terminal root of the leucine zipper domain interacts with the ␤B-␤C loops at the center of PDZ dimer and the C-terminal tails are inserted into the peptide binding pockets (Fig. 3E). This model suggests that the novel surface interacting with the leucine zipper domain is the conserved ␤B/␤C loop. Such a configuration is advantageous, because it would enable dimeric PDZ domains to efficiently colocalize with dimeric target proteins.