Structural determinants of the Na+/H+ exchanger regulatory factor interaction with the beta 2 adrenergic and platelet-derived growth factor receptors.

The Na(+)/H(+) exchanger regulatory factor (NHERF) binds through its PDZ1 domain to the carboxyl-terminal sequences NDSLL and EDSFL of the beta(2) adrenergic receptor (beta(2)AR) and platelet-derived growth factor receptor, respectively, and plays a critical role in the membrane localization and physiological regulation of these receptors. The crystal structures of the human NHERF PDZ1 domain bound to the sequences NDSLL and EDSFL have been determined at 1.9- and 2.2-A resolution, respectively. The beta(2)AR and platelet-derived growth factor receptor ligands insert into the PDZ1 binding pocket by a beta-sheet augmentation process and are stabilized by largely similar networks of hydrogen bonds and hydrophobic contacts. In the PDZ1-beta(2)AR complex, the side chain of asparagine at position -4 in the beta(2)AR peptide forms two additional hydrogen bonds with Gly(30) of PDZ1, which contribute to the higher affinity of this interaction. Remarkably, both complexes are further stabilized by hydrophobic interactions involving the side chains of the penultimate amino acids of the peptide ligands, whereas the PDZ1 residues Asn(22) and Glu(43) undergo conformational changes to accommodate these side chains. These results provide structural insights into the mechanisms by which different side chains at the position -1 of peptide ligands interact with PDZ domains and contribute to the affinity of the PDZ-ligand interaction.

The Na ϩ /H ϩ exchanger regulatory factor (NHERF) 1 is a cytoplasmic phosphoprotein originally identified as an essential cofactor for the inhibition of the Na ϩ /H ϩ exchanger 3 by the cAMP-dependent protein kinase A in the renal brush-border membrane (1). NHERF, also known as ezrin/radixin/moesinbinding phosphoprotein 50 or EBP50 (2), contains two PDZ (PSD-95/Discs-large/ZO-1) domains that bind to carboxyl-terminal sequences of membrane proteins and an ezrin/radixin/ moesin-binding domain that interacts with the cortical actin cytoskeleton (2,3). Through these interactions, NHERF plays central roles in the membrane targeting, trafficking, and sorting of several ion channels, transmembrane receptors, and signaling proteins (reviewed in Ref. 4).
Previous studies have shown that the amino-terminal PDZ domain of NHERF, referred to as PDZ1, binds with high affinity to the carboxyl termini of the cystic fibrosis transmembrane conductance regulator (CFTR), ␤ 2 adrenergic receptor (␤ 2 AR), and platelet-derived growth factor receptor (PDGFR) (5)(6)(7). The PDZ1-mediated interaction of NHERF with the carboxylterminal motif DTRL of CFTR is essential for the functional expression in the apical plasma membrane and gating of the CFTR channel, and the mutations that destroy this motif result in abnormal apical polarization of CFTR and defective vectorial chloride transport (8,9). Likewise, the binding of the NHERF PDZ1 domain to the ␤ 2 AR carboxyl-terminal sequence NDSLL is essential for the ␤ 2 AR-mediated regulation of Na ϩ /H ϩ exchange (6). This interaction also mediates the sorting of internalized ␤ 2 AR between degradative endocytic pathways and plasma membrane recycling; the truncation of the ␤ 2 AR carboxyl terminus prevents NHERF PDZ1 binding and leads to lysosomal degradation of the receptor (10). Importantly, the phosphorylation of the serine within the ␤ 2 AR sequence ND-SLL (Ser 411 in the full-length receptor) by the G-protein-coupled receptor kinase-5 inhibits the NHERF PDZ1-␤ 2 AR interaction and regulates ␤ 2 AR endocytic sorting (10). In addition, the NHERF PDZ1 interacts with the motif EDSFL, which is present at the carboxyl termini of PDGFR forms A and B, and potentiates PDGFR autophosphorylation and extracellular signal-induced kinase activation (5,7). Notably, a single mutation of the PDGFR carboxyl-terminal leucine to alanine abolishes binding to NHERF PDZ1 and impairs the receptor activity, demonstrating the critical role of this interaction in PDGFR function (7).
PDZ domains are protein-protein interaction modules that participate in the assembly of membrane receptors, ion channels, and other signaling molecules into specific signal transduction complexes (11,12). These domains bind to short carboxyl-terminal sequence motifs and were initially grouped into two classes based on target sequence specificity. Class I domains bind to peptides with the consensus sequence (S/T)X(V/ I/L) (X denoting any amino acid), whereas class II domains recognize the motif (F/Y)X(F/V/A) (13). However, other PDZ domains have ligand sequence specificities that do not fall into these two categories, suggesting the existence of more PDZ classes (11,12). The PDZ-fold comprises a six-stranded antiparallel ␤-barrel capped by two ␣-helices (14 -22). Peptide ligands * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The specificity of the PDZ-ligand interaction is achieved by the residues at positions Ϫ3, Ϫ2, and 0 of the peptide (position 0 referring to the carboxyl-terminal residue) (11,12). By contrast, initial biochemical and structural studies indicated that the residue at position Ϫ1 of the peptide ligand plays no role in the interaction with PDZ domains (13,14,16). However, more recent studies demonstrated a strong preference of PDZ domains for specific side chains at position Ϫ1, suggesting an important role of these side chains in the PDZ-ligand interaction (17,24,25). In the case of NHERF, affinity selection experiments showed that PDZ1 selected almost exclusively ligands with arginine at position Ϫ1 from random peptides (25), and mutagenesis of the penultimate arginine to other residues decreased the affinity of the PDZ1-ligand interaction (5). A structural explanation for this preference was provided by the crystal structure of the human NHERF PDZ1 domain complexed with the CFTR carboxyl-terminal sequence QDTRL (22). In this structure, the guanidino group of arginine at position Ϫ1 of the CFTR peptide forms two salt bridges and two hydrogen bonds with PDZ1 residues Glu 43 and Asn 22 , respectively, thus contributing to the affinity of the PDZ1-ligand interaction. This finding led to the hypothesis that PDZ residues corresponding to the hNHERF PDZ1 Asn 22 and Glu 43 may play a more general role in the interaction with the penultimate residues of the peptide ligands (22).
Here, we describe the crystal structures of hNHERF PDZ1 bound to the carboxyl-terminal sequences NDSLL and EDSFL of ␤ 2 AR and PDGFR, respectively. The structures provide new insights into the mechanisms by which hydrophobic side chains at position Ϫ1 interact with this domain, contributing to the affinity of the interaction, and reveal the conformational changes that the PDZ1 residues Asn 22 and Glu 43 undergo to accommodate these side chains.

EXPERIMENTAL PROCEDURES
Protein Purification and Crystallization-A DNA fragment encoding the human NHERF PDZ1 domain (residues 11-94) and having the carboxyl-terminal extension N 95 DSLL 99 that corresponds to residues 409 -413 of human ␤ 2 AR was amplified using the polymerase chain reaction and cloned in the vector pGEX-2TJL (22). A similar strategy was used to generate the hNHERF PDZ1 domain (residues 11-94) having the carboxyl-terminal extension E 95 DSFL 99 that corresponds to residues 1085-1089 and 1102-1106 of human PDGFRA and PDGFRB, respectively. The resulting chimeric PDZ1-␤ 2 AR and PDZ1-PDGFR domains were expressed in Escherichia coli BL21 (DE3) cells as glutathione S-transferase fusion proteins purified using glutathione-Sepharose 4B resin and released from the glutathione S-transferase moiety by digestion with thrombin as described previously (21). PDZ1-␤ 2 AR protein (19 mg/ml) was crystallized using the sitting drop vapor diffusion method in 0.1 M citric acid, pH 4.24, and 1 M lithium chloride at 20°C, whereas PDZ1-PDGFR (18 mg/ml) was crystallized in 0.1 M sodium acetate, pH 4.6, 2 M sodium chloride at 20°C. Both crystals were cryoprotected in 20% (v/v) glycerol for 3 min and flash-frozen in a stream of liquid nitrogen. Diffraction data sets were collected at 115 K using CuK ␣ radiation from a Rigaku RU-300 generator and an R-AXIS IV-imaging plate detector. The data were processed and scaled using the programs DENZO and SCALEPACK (26) (Table I). Both crystals belong to space group P3 1 21 with one molecule in the asymmetric unit and unit cell parameters: a ϭ b ϭ 50.37 Å, c ϭ 66.63 Å (for PDZ1-␤ 2 AR) and a ϭ b ϭ 50.07 Å, c ϭ 66.68 Å (for PDZ1-PDGFR).
Structure Determination and Refinement-Both structures were solved by molecular replacement using the program MOLREP (27) and the hNHERF PDZ1 (Protein Data Bank code 1g9o) as the search model. For PDZ1-␤ 2 AR, the model was truncated by removing the last six carboxyl-terminal residues. The rotation function search in the resolution range from 20.0 to 3.0 Å produced a clear solution with a peak height of 6.53 above the mean. The translation function indicated that the correct space group was P3 1 21 with a correlation coefficient of 47.6% and an R-factor of 47.3%. For the structure determination of PDZ1-PDGFR, a polyalanine model of hNHERF PDZ1 was used for molecular replacement in order to remove any model bias. The rotation function in the resolution range from 20.0 to 3.0 Å produced a solution with a peak height of 6.31 above the mean, and the translation function indicated that the space group was also P3 1 21 with a correlation coefficient of 55.6% and an R-factor of 42.7%. Models were built using the program O (28) and were refined by the maximum likelihood method using REF-MAC5 (29). Both structures are well ordered with the exception of the loop region spanning residues 31-35, which is disordered. The crystallized proteins also contained at their amino termini the vector-derived residues GSSRM from which only methionine is ordered and included in both models, whereas arginine is partially ordered and represented as a glycine in the PDZ1-PDGFR model. c R free is calculated as R cryst using 5% of the reflection data chosen randomly and omitted from the refinement calculations. Overview of the PDZ1-␤ 2 AR and PDZ1-PDGFR Structures-To determine the crystal structures of the hNHERF PDZ1 domain bound to the ␤ 2 AR and PDGFR carboxyl-terminal ligands, we adopted a crystallization strategy similar to that used in the hNHERF PDZ1-CFTR structural analysis (22). In this approach, we crystallized the hNHERF PDZ1 domain (residues 11-94) fused to either the ␤ 2 AR carboxyl-terminal sequence N 95 DSLL 99 or the PDGFR carboxyl-terminal motif E 95 DSFL 99 (Fig. 1A). As in the case of hNHERF PDZ1-CFTR, the hNHERF PDZ1-␤ 2 AR and hNHERF PDZ1-PDGFR crystal structures produced infinite head-to-tail polymers of PDZ1 molecules along the z axis with the carboxyl-terminal pentapeptide extension of one chimeric PDZ1 molecule serving as a ligand for a neighboring PDZ1 (Fig. 1B). Both crystal structures were solved by molecular replacement using the hNHERF PDZ1 structure (21) as the search model. The PDZ1-␤ 2 AR structure was refined to an R cryst of 17.1% and an R free of 18.6%, whereas the PDZ1-PDGFR model was refined to an R cryst of 22.2% and an R free of 26.3% (Table I). An evaluation of the stereochemistry of the PDZ1-␤ 2 AR structure using PRO-CHECK (30) showed that 89% of the residues are in the most favored region and 9.6% in the additional allowed regions. One residue (Ala 81 ) is disordered and located in the disallowed region of the Ramachandran plot. A similar analysis of the PDZ1-PDGFR model placed 87.8% of the residues in the most favored and 12.2% in the additional allowed regions.
The hNHERF PDZ1 core domain consists of six ␤-strands (␤1-␤6) and two ␣-helices (␣1 and ␣2) (Fig. 1, A, C, and D). The ␤ 2 AR and PDGFR carboxyl-terminal sequences NDSLL and EDSFL, respectively, insert into the PDZ1 binding pocket as additional ␤-strands (␤7) antiparallel to ␤2 and extend the ␤-sheet of PDZ1. In this arrangement, the invading pentapeptide ligands are highly ordered as indicated by the high quality electron density maps (Fig. 1, C and D) and average B factors of 13.5 Å 2 for ␤ 2 AR and 30.7 Å 2 for PDGFR.
Specificity Determinants of the hNHERF PDZ1 Interaction with ␤ 2 AR and PDGFR-The stabilization of the PDZ1-␤ 2 AR and PDZ1-PDGFR interactions is achieved through networks of hydrogen bonds and hydrophobic interactions, which to a large extent are similar in the two protein-peptide complexes. The side chains of residues Leu 0, Ser Ϫ2, and Asp Ϫ3, which are common in both ligands, are engaged in numerous interactions with PDZ1 residues, consistent with biochemical evidence on the central role of these residues in the specificity and affinity of the NHERF PDZ1 interaction with ␤ 2 AR and PDGFR (5)(6)(7)(8)(9)(10). Specifically, in both PDZ1-␤ 2 AR and PDZ1-PDGFR structures, the side chain and carboxylate group of Leu 0 enter into a deep cavity formed by Tyr 24 , Gly 25 , Phe 26 , Leu 28 , Val 76 , and Ile 79 residues. In this position, the carboxyl oxygen of Leu 0 hydrogen bonds with the amide nitrogens of Gly 25 and Phe 26 , whereas the carbonyl oxygen of Leu 0 hydrogen bonds directly with the amide nitrogen of Tyr 24 (Fig. 2, A-D). The isobutyl group of Leu 0 is stabilized by hydrophobic interactions with PDZ residues that are largely similar in both structures (Fig. 2, C and D). Moreover, in both complexes the amide nitrogen and carbonyl oxygen atoms of Ser Ϫ2 hydrogen bond with the carbonyl oxygen and amide nitrogen of Leu 28 , respectively, whereas the O ␥1 atom of Ser Ϫ2 hydrogen bonds with the N ⑀2 atom of the conserved His 72 . These interactions corroborate previous studies that demonstrated the critical role of a serine or threonine at position Ϫ2 of the ligand and a histidine at the beginning of the ␣2 helix for the specificity of class I PDZ-peptide interactions (5,6,10,12,13). In the case of the ␤ 2 AR ligand, the phosphorylation of Ser Ϫ2 by G-protein-coupled receptor kinase-5 has been shown to abrogate the interaction with NHERF PDZ1 and play an important role in the regulation of ␤ 2 AR endocytic sorting (10). Based on the current PDZ1-␤ 2 AR structure, it could be predicted that the added phosphoryl group on the Ser Ϫ2 would sterically clash with the imidazole group of His 72 , dislodging the main chain of the peptide away from the binding pocket and disrupting several hydrogen bonds that stabilize the PDZ-peptide ␤-sheet.
As observed previously in the hNHERF PDZ1-CFTR structure (22), in both PDZ1-␤ 2 AR and PDZ1-PDGFR complexes the O ␦1 atom of Asp Ϫ3 hydrogen bonds with N ␦1 of His 27 , the O ␦2 atom of Asp Ϫ3 forms a salt bridge with N 1 of Arg 40 , and the carbonyl oxygen of the residue at position Ϫ4 hydrogen bonds with the amide nitrogen of Gly 30 (Fig. 2, A-D). However, in the PDZ1-␤ 2 AR structure the O ␦1 and N ␦2 atoms of Asn Ϫ4 make two additional hydrogen bonds with the amide nitrogen and carbonyl oxygen of Gly 30 , respectively (Fig. 2, A and C), that contribute to the affinity of this interaction.
Role of the Penultimate Residue in the PDZ-Peptide Interaction-As mentioned above, in the hNHERF PDZ1-CFTR structure, the guanidino group of Arg Ϫ1 forms two salt bridges with O ⑀2 of Glu 43 and two hydrogen bonds with the carbonyl oxygen of Asn 22 (22). This unexpected finding suggests that the residue Ϫ1 of the peptide plays an important role in the affinity of the PDZ-ligand interaction. It also raises the question of whether nonpolar side chains at position Ϫ1 of peptide ligands, such as the large hydrophobic side chains of the penultimate residues of ␤ 2 AR and PDGFR, interact with the hNHERF PDZ1 or are disordered facing toward the solution. The present structures clearly demonstrate that both the isobutyl group of Leu Ϫ1 and the phenolic ring of Phe Ϫ1 of the ␤ 2 AR and PDGFR ligands, respectively, are well ordered (Fig. 1, C and D) and engage in hydrophobic interactions with PDZ1 residues. Specifically, the C ␦1 atom of the ␤ 2 AR Leu Ϫ1 contacts the C ␦2 atom of His 27 , whereas the C ␦1 and C ␤ atoms of the PDGFR Phe Ϫ1 contact the C ␦2 and C ⑀1 atoms of His 27 , respectively (Fig. 2, A and B). These results provide compelling structural evidence for the importance of the penultimate ligand residue in the affinity of the PDZ-peptide interaction. In agreement with this conclusion, the recent crystal structure of the InaD PDZ1 complexed with the carboxyl-terminal sequence of NorpA revealed the presence of an intermolecular disulfide bond formed between Cys Ϫ1 of the peptide ligand and a cysteine residue of the PDZ domain (31).
Another notable feature in the present structures is that the carbonyl oxygen atoms of the penultimate residues of both ␤ 2 AR and PDGFR ligands make direct hydrogen bonds with the N 2 atom of Arg 80 (Fig. 2, A and B). In this respect, the PDZ1-␤ 2 AR and PDZ1-PDGFR structures differ from that of PDZ1-CFTR where the carbonyl oxygen of Arg Ϫ1 does not hydrogen bond directly with Arg 80 (22). Moreover, the carbonyl oxygen of Leu 0 interacts indirectly with the N ⑀ atom of Arg 80 through two ordered water molecules in the PDZ1-␤ 2 AR and PDZ1-CFTR structures but not in the PDZ1-PDGFR complex. These findings demonstrate that the networks of hydrogen bonds and ordered water molecules participating in hydrogen interactions observed between the hNHERF PDZ1 residues (orange) and ␤ 2 AR (C) or PDGFR (D) peptide ligands (purple). Dashed lines denote hydrogen bonds, and numbers indicate hydrogen bond lengths in Å. Hydrophobic interactions are shown as arcs with radial spokes. C and D were generated using LIGPLOT (38). E, superposition of the hNHERF PDZ1 bound to CFTR (yellow), ␤ 2 AR (blue), and PDGFR (pink) peptide ligands. For clarity, the ␤2 strand and ␣2 helix of the PDZ1 domain are shown as ribbon diagrams, and only the C ␣ traces are shown for the remaining PDZ1 main chain. Side chains of the peptide ligands and the PDZ1 residues Asn 22 and Glu 43 are shown as ball-and-stick models.
bonding between the PDZ domain and the peptide differ among various ligands bound to the same PDZ domain.
Structural Basis for the PDZ Interaction with Different Penultimate Residues-The ability of the hNHERF PDZ1 domain to interact with different side chains at position Ϫ1 of the peptide ligand raises the question of whether PDZ residues around the binding pocket undergo conformational changes to accommodate different ligands. The superposition of the hNHERF PDZ1 domains bound to the ␤ 2 AR, PDGFR, and CFTR peptides reveals the structural differences among the three complexes and provides insights into the mechanisms underlying PDZ-ligand recognition. The three liganded hNHERF PDZ1 structures are very similar with overall rootmean-square deviations ranging from 0.53 to 0.88 Å (Fig. 2E). Larger differences in the backbone positions are observed in the ␤2-␤3 and ␣2-␤6 loops with root-mean-square deviation ranges of 1.13-1.98 and 1.08 -1.59 Å, respectively. The main chains of the three peptide ligands and the side chains of residues at positions 0, Ϫ2, and Ϫ3 are superimposed extremely well, whereas large deviations are observed in the orientations of the side chains at position Ϫ4 (Fig. 2E). To a large extent, the side chains of Phe Ϫ1 and Leu Ϫ1 follow a path similar to that of the aliphatic portion of the Arg Ϫ1 side chain, facing toward the PDZ residues Asn 22 and Glu 43 . Notably, the side chains of these two PDZ residues exhibit the largest conformational changes seen in all three complexes. In the PDZ1-␤ 2 AR structure, the side chain of Asn 22 is oriented toward the isobutyl group of Leu Ϫ1, whereas in the PDZ1-PDGFR and PDZ1-CFTR complexes, the amide group is rotated away from the longer side chains of Phe Ϫ1 and Arg Ϫ1 because of steric effects (Fig. 2E). Remarkably, the Glu 43 side chain occupies three different positions for optimal ligand interaction. These findings indicate that the conformational changes of Asn 22 and Glu 43 underlie the hNHERF PDZ1 flexibility to accommodate ligands with penultimate side chains of different hydrophobicity and polarity and suggest similar roles for the corresponding residues of other PDZ domains.
Perspective-This work provides the structural basis for the interaction of ␤ 2 AR and PDGFR with hNHERF PDZ1 and expands our insights into the molecular mechanisms underlying peptide recognition by class I PDZ domains. The observed differences in the hNHERF PDZ1 interaction with the CFTR, ␤ 2 AR, and PDGFR ligands also indicate that although PDZ domains are structurally simple protein interaction modules sharing the same fold, more biochemical and structural studies are needed to derive the structural principles governing the selection of their target peptides.
From a clinical standpoint, because of the central roles of ␤ 2 AR and PDGFR in human physiology and pathology, the structural determinants of the hNHERF PDZ1 interaction with these receptors may have important applications in molecular medicine. For example, this information may be used for the design of a new class of specific inhibitors of these receptors, which would act by disrupting the hNHERF PDZ1-receptor interaction and block the functional localization in plasma membrane and activity of these receptors. It was previously shown that ␤ 2 AR agonists induce conformational changes in the receptor that promote a direct physical association of its carboxyl terminus to the hNHERF PDZ1 (6). Therefore, small molecules that would block specifically the hNHERF PDZ1-␤ 2 AR interaction could act as ␤ 2 AR antagonists. Likewise, the inhibitors of the hNHERF PDZ1-PDGFR interaction could be useful in reducing the mitogenic activity of this receptor in tumor cells. In this context, it is of particular interest to note that the hNHERF gene expression is rapidly activated by estrogen in estrogen receptor-positive breast cancer cells (32), and estrogen receptor levels correlate closely with hNHERF expression in breast carcinomas (33), raising the intriguing possibility that hNHERF may play an important role in breast cancer development through potentiation of the PDGFR mitogenic activity (7). Moreover, it was recently shown that PDGFRA is significantly up-regulated in metastatic medulloblastoma and enhances the migration properties of medulloblastoma cells, suggesting that strategies aiming at inhibiting PDGFRA function could constitute novel therapeutic approaches against medulloblastoma (34). Therefore, it is conceivable that the hNHERF PDZ1-PDGFR atomic structure may provide the framework for the design of small molecules that would block this interaction and decrease the PDGFRproliferative activity in cancer cells.