Syntenin-Syndecan Binding Requires Syndecan-Synteny and the Co-operation of Both PDZ Domains of Syntenin*

Syntenin is an adaptor-like molecule that binds to the cytoplasmic domains of all four vertebrate syndecans. Syntenin-syndecan binding involves the C-terminal part of syntenin that contains a tandem of PDZ domains. Here we provide evidence that each PDZ domain of syntenin can interact with a syndecan. Isolated or combined mutations of the carboxylate binding lysines in the inter- b A b B loops and of the a B1 residues in either one or both the PDZ domains of syntenin all reduce syntenin-syndecan binding in yeast two-hybrid, blot-overlay, and surface plasmon resonance assays. PDZ2 mutations have more pronounced effects on binding than PDZ1 mutations, but complete abrogation of syn-tenin-syndecan binding requires the combination of both the lysine and the a B1 mutations in both the PDZ domains of syntenin. Isothermal calorimetric titration of syntenin with syndecan peptide reveals the presence of two binding sites in syntenin. Yet, unlike a tandem of two PDZ2 domains and a reconstituted PDZ1 1 PDZ2 tandem, a tandem of two PDZ1 domains and isolated PDZ1 or PDZ2 domains do not interact with syndecan bait. We conclude to a co-operative binding mode whereby nei-ther of these two PDZ domains is sufficient by itself but where PDZ2 functions as a “major” or “high affinity” syndecan binding domain, and PDZ1

Syndecans are type-I membrane proteins that are substituted with heparan sulfate. They function as versatile co-receptors, as there is a growing list of examples where the heparan sulfate chains of these membrane proteins are involved in the docking of heparin binding molecules (e.g. growth factors and adhesion molecules) to cell surfaces and facilitate the interactions of these molecules with specific cognate signaling receptors (1). The strict evolutionary conservation of the structures of the cytoplasmic domains of the syndecans implies that the biological functions of these membrane proteins may also depend on highly specific cytoplasmic interactions and associations. Recently, the cytoplasmic domains of the syndecans were shown to interact with syntenin (2) and CASK (3,4), two proteins that belong to the larger family of proteins that contain one or several PDZ domains (PDZ proteins).
PDZ domains are structural motifs of about 80 amino acids that were initially found in the post synaptic density-95, disclarge, and zonulin-1 proteins but occur in a large variety of proteins (5). PDZ domains interact with the C termini of specific peptide structures. X-ray structures (6,7) and NMR data (8) show that PDZ domains are compact ␣ ϩ ␤ modules containing five to six ␤ strands (labeled ␤A to ␤E) and two ␣ helices (␣A and ␣B). The binding peptide fits into a hydrophobic pocket created by the principal ␣-helix (␣B), the second ␤-strand (␤B), and the "carboxylate binding" loop that connects the ␤A and ␤B strands. The terminal carboxylate of the peptide interacts with a cradle of amide nitrogens from the inter-␤A␤B loop, but an arginine or lysine at the start of this loop also contributes to the stabilization of this carboxylate via a bound water molecule. The C-terminal four residues of the peptide are stabilized by main chain hydrogen bonds with ␤B, whereby peptide binding represents the augmentation of the PDZ ␤-sheet by an antiparallel strand. PDZ domains are selective. This binding specificity is achieved by domain interactions with the residue at the Ϫ2 position of the bound peptide. Type-I PDZ domains bind to peptides with the terminal-(S/T)XV consensus motif, whereby the Ser or Thr hydrogen bonds a relatively well conserved histidine residue near the start of the ␣B helix (7). Type-II PDZ domains, in contrast, bind to peptides with hydrophobic or aromatic amino acids in the Ϫ2 position and feature a non-basic residue as the first residue of the ␣B helix (9). Most often a PDZ domain occurs in association with other functional protein-protein interaction modules, including other PDZ domains (5). The specific associations and most often multiple binding interactions of PDZ proteins implicate many of these proteins in the localization of receptors and cytosolic effectors to specific membrane sites and in linking extracellular signals to the cytoskeleton and intracellular signaling pathways.
Syntenin contains two PDZ domains, which occur as a direct repeat and compose the C-terminal two-thirds of the protein.
Conforming to a PDZ-mediated interaction, the interaction of syntenin with syndecans depends on the integrity of the C-terminal FYA sequence that is common to all syndecans and qualifies as a class II PDZ-interacting peptide (10). Yet, the syndecan cytoplasmic domain in solution does not bind syntenin or not as strongly as syntenin binds to immobilized syndecans, and the only part of syntenin that appears dispensable for the interaction is the N-terminal domain. This suggests that both PDZ domains might be needed in a two-pronged (socketplug) attachment of syntenin to di-or oligomerized syndecans (2). Syntenin is not unique is this respect. In GRIP, which contains two clusters of three adjacent PDZ domains, the minimal region required for binding to the AMPA receptor spans two complete PDZ domains (11), and also in the case of the human Dlg protein, two adjacent PDZ domains are needed for interaction with protein 4.1. (12). On the other hand, there are many more examples of PDZ domains that do work in isolation, and several proteins with single PDZ domains have been identified. CASK, originally identified as a membrane-associated guanylate kinase homolog that binds to the t-YYV sequence of neurexin (13) and recently shown to also bind to syndecans (3,4), is one of these proteins with single PDZ domains that work in isolation. This singles out the PDZ domains of syntenin and several other proteins as peculiar or suggests trivial explanations for this need for paired domains. Conceivably, incorrect folding of PDZ domains that were isolated from their structural contexts may complicate binding assignments. Moreover, some PDZ domains have been shown to form dimers with other PDZ domains (14,15), further complicating the identification of the peptide-binding sites in those instances where binding would be based on co-operative PDZ domain interactions.
To resolve this issue for the interaction between syntenin and the syndecan cytoplasmic domain, we have introduced point mutations in the two PDZ domains of syntenin, separately or in combination, using the crystal structures of peptide bound to PDZ3 of human Dlg and PSD-95 as a guide. The mutants were used as prey for syndecan bait in yeast twohybrid reporter assays and expressed as glutathione S-transferase (GST) 1 fusion proteins for use in blotting and surface plasmon resonance experiments. Collectively, the results indicate that both PDZ domains of syntenin bind syndecans, but with unequal strengths, and that only their co-operative binding results in stable syntenin-syndecan interactions. Syntenin does not discriminate between the various syndecans. Syntenin binds also to neurexins and B-class ephrins, with a similar requirement of coupled functional PDZ domains. These data suggest that in vivo syntenin may be recruited to the membrane by the assembly and mixed assemblies of syndecans and other suitable partners, such as neurexins and ephrins.

GCCGATCCA.
Production of GST Fusion Proteins-The syndecan, neurexin, and B-ephrin cDNAs were cloned into the pGEX-5x-2 expression vector (Amersham Pharmacia Biotech) to create fusion proteins between GST and the cytoplasmic domains of these membrane proteins. An expression vector encoding glutathione S-transferase extended by an myc tag was constructed by annealing 5Ј-GATCTCCGAACAAAAACTCATCTC-AGAAGAGGATCTGGG to 5Ј-GATCCCCAGATCCTCTTCTGAGATGA-GTTTTTGTTCGGA and inserting the annealed oligonucleotides into the BamHI site of pGEX-5x-2. Vectors encoding fusion proteins between GST-myc and wild-type syntenin or paired syntenin-PDZ domains were constructed by inserting the corresponding cDNAs in the BamHI site of this modified vector. Syntenin mutants were constructed as non-myctagged fusion proteins in pGEX-5x-2. All constructs were verified by DNA sequencing.
Escherichia coli BL21 or ER2566 cells were used as host cells to express the GST fusion proteins. Expression was induced by adding 0.4 mM isopropyl ␤-D-thiogalactoside to the medium when the culture reached an A 550 of 0.6. After induction, the cultures were allowed to grow for another 3 h at 30°C. Induced cells were collected by centrifugation for 15 min at 8000 ϫ g at 4°C. All the medium was carefully removed, and the pellet was resuspended in 8 ml of TEN buffer (100 mM NaCl, 1 mM EDTA, 50 mM Tris-HCl, pH 8.0). Aprotinin (3 M), benzamidine (5 mM), leupeptin (20 M), pepstatin (3 M), 6-aminohexanoic acid (5 mM), and phenylmethylsulfonyl fluoride (200 M) were added as protease inhibitors. The cell suspension was incubated with 2.5 mg of lysozyme for 1 h on ice. After 1 h, the solution was centrifuged for 20 min at 18,000 ϫ g at 4°C. Except for GST-PDZ2 and GST-PDZ2ϩPDZ2, which proved insoluble, all the GST fusion proteins were mainly found in the water-soluble phase. This phase was applied to a glutathione-Sepharose 4B column (Amersham Pharmacia Biotech). After washing the column with phosphate-buffered saline, the fusion protein was eluted from the column with reduced glutathione.
Surface Plasmon Resonance Measurements-Surface plasmon resonance was measured using a BIAcore 2000 instrument. A total of 900 resonance units of biotinylated synthetic peptide, corresponding to the 32 amino acids that compose the cytoplasmic domain of syndecan-2, was coupled to a streptavidin sensor chip. Analytes (GST fusion proteins) were perfused (10 l/min) over the control (Fc1) and capture (Fc2) surfaces in running buffer (0.1 M NaCl, 0.005% surfactant P20, 0.01M Hepes, pH 7.4). Binding was measured as the difference between the Fc2 and Fc1 binding curves. All bindings were tested at different concentrations of analyte.
Blotting Assays-Purified GST fusion proteins or isopropyl ␤-D-thiogalactoside-induced BL21/ER2566 cells expressing these fusion proteins were re-suspended in SDS sample buffer and boiled for 5 min. The samples were cleared by centrifugation and fractionated in 12% SDS-PAGE, each lane containing 0.1 g of fusion protein. Cultured human MCF-7 cells were scraped in phosphate-buffered saline and treated with heparitinase before fractionation by 12% SDS-PAGE. After electro-transfer to Hybond-C-super membranes, the blots were blocked with 5% nonfat milk powder in TBS-Tween (150 mM NaCl, 0.1% Tween 20, 20 mM Tris-HCl, pH 7.6) and incubated with fusion proteins of GST-myc-syntenin, GST-myc deletion mutants, or GST-syntenin point mutants (10 g/ml in TBS-Tween). After a 3-h incubation at room temperature, the blots were washed with TBS-Tween and incubated for 2 h at room temperature with anti-myc monoclonal antibody 9E10 or with anti-syntenin mAb 1C8, both diluted to 1 g/ml in TBS-Tween. After rinsing with TBS-Tween, bound anti-myc and anti-syntenin were visualized with horseradish peroxidase goat anti-mouse (Bio-Rad 1/10.000) and ECL Western blotting detection reagent (Amersham Pharmacia Biotech).
Yeast Two-hybrid Assays-The bait, consisting of the cytoplasmic domain of syndecan-2 fused in-frame to the DNA binding domain of Gal4, was constructed by PCR using the BamHI/PstI restriction sites of the pAS2 vector (CLONTECH Laboratories). The syntenin, mutant syntenins, and PDZ tandem prey were cloned, respectively, in pGAD10, pGAD424, and pACT2 (all from CLONTECH Laboratories). The pAS2 plasmids were transfected in Y187 yeast cells, whereas the pGAD/pACT plasmids were transfected in CG1945 yeast cells. Diploid cells resulted from mating Y187 with CG1945 were selected on double minus plates (Leu Ϫ , Trp Ϫ ), assayed for ␤-galactosidase activity, and assayed for growth on triple minus plates (Leu Ϫ , Trp Ϫ , His Ϫ ) with and without 5 mM 3-aminotriazol. For the ␤A-␤B-loop and ␣B mutants only, the galactosidase staining is shown in the figures; for the shuffle mutants, results of the growth assay are represented in tabular form in Fig. 4.
Tag-free Recombinant Syntenin-Recombinant syntenin, free of any peptide extensions, but with the C-terminal valine substituted by an alanine, was produced in the IMPACT (intein-mediated purification with an affinity chitin binding Tag) T7 system, according to the protocols provided by the supplier (New England Biolabs Inc., Beverly, MA). CDNA-encoding syntenin was produced by PCR using the primers 5Ј-GGAATTCCATATGTCTCTCTATCCATCTCTCGAAG and 5Ј-GGC-CGCTCTTCCGCAAGCCTCAGGAATGGTGTGGTCCA and cloned in the NdeI and Sap I restriction sites of the pTYB1 vector to produce an in-frame fusion between the C terminus of syntenin and the N terminus of an intein-chitin binding domain fusion protein. After propagation of the plasmid in ER2566 E. coli cells and induction of the cells at an A 550 of 0.5 with 0.3 mM isopropyl ␤-D-thiogalactoside at 30°C for 4 h, the tripartite fusion protein was isolated from crude cell extracts by affinity purification on a chitin column. After extensive washings, the column was treated overnight at 4°C with 30 mM 1,4-dithiothreitol to induce the intein-mediated self-cleavage of the fusion protein and the consequent release of its syntenin moiety from the column. SDS-PAGE and Coomassie staining revealed only a single peptide band of 33 kDa in the eluate of the DTT-treated column. Alanine was substituted for the C-terminal valine in syntenin because we noted that this substitution markedly increased the efficiency of the excision process and yield of syntenin. The yield was approximately 1 mg/liter of induced culture.
Isothermal Titration Calorimetry-The syntenin solution was dialyzed overnight at 4°C against 0.1 M NaCl, 0.1 mM EDTA, 20 mM Hepes, pH 8.0 (ITC buffer) before the isothermal titration calorimetry experiment. The titration was performed using an Omega isothermal titration calorimeter from MicroCal Inc. (Northampton, MA). The concentration in the sample cell with a volume of 1.33 ml was 0.0980 mM (3.18 mg/ml). The titrated peptide was dissolved to a concentration of 1.72 mM (6.35 mg/ml) in the ITC buffer and injected in fractions of 18 l. The temperature of the cell was 25°C. The obtained titration thermogram was analyzed with MicroCal™Origin™ 5.0 software.  (7). Recognition of this terminal carboxylate also involves the side chain of a basic residue, Arg or Lys, which invariantly occurs as the first residue of the loop. In syntenin these basic residues correspond to Lys-119 (PDZ1) and Lys-203 (PDZ2). Speculating that either one or both of these residues would be critical for optimal binding, we replaced these lysines by alanines in either PDZ1 or PDZ2 or in both PDZ1 and PDZ2 and tested the influence of these mutations on the synteninsyndecan interaction. The exact positions of these mutations are shown in Fig. 1.

␤A-␤B
In yeast two-hybrid assays, syndecan bait only interacts with syntenin prey that contains the two PDZ domains of syntenin. Moreover, in this particular case bait and prey can not be interchanged (2). In this assay, as assessed from colony outgrowth and ␤-galactosidase staining and using syndecan-2 as bait, none of the Lys 3 Ala mutations markedly influenced the syndecan-syntenin interaction (Fig. 2B). In surface plasmon resonance (BIAcore) experiments, GST fusion proteins that contain the two PDZ domains of syntenin bind avidly to a syndecan-2 cytoplasmic domain peptide that is immobilized on the sensor chip but not inversely. Moreover, peptide in solution does not compete for syntenin binding to immobilized peptide (2). Binding curves obtained from the perfusions of similar amounts and concentrations of the different GST-syntenin fusion proteins indicated a decreased binding to immobilized syndecan-2 peptide, both for the K119A mutant (PDZ1) and for the K203A mutant (PDZ2) (Fig. 2E). Compared with wild-type syntenin, the binding of the K203A mutant (PDZ2) was more markedly reduced than that of the K119A mutant (PDZ1). A similarly significant reduced binding was also observed for the K119A/K203A double mutant. A GST-syntenin fusion protein binds also to GST-syndecan-2 fusion proteins that have been fractionated by SDS-PAGE and blotted on nitrocellulose membranes (2). When used as probes in blotting assays, at similar concentrations of GST-syntenin in solution and for similar amounts of GST-syndecan immobilized on the membranes (as assessed by parallel staining with anti-GST), all three Lys [rarrow] Ala mutants bound still to wild-type syndecan-2 (Fig.  3). The K203A mutant (PDZ2) and the K119A/K203A double mutant produced signals of almost similar intensities, both clearly lower than the signal obtained with the K119A mutant (PDZ1). Like wild-type syntenin, none of these Lys 3 Ala mutants bound to GST-F(C30)S syndecan-2 (substitution of serine for the phenylalanine at the -2 position of the syndecan peptide), consistent with a PDZ-mediated interaction and an unchanged peptide specificity of this interaction (Fig. 3). All together, the negative effects of the lysine to alanine mutations suggested that both PDZ domains were (directly or indirectly) involved in the syndecan binding, possibly with a larger contribution of PDZ2 than of PDZ1.
␣B Helix Mutations-The crystal structures of the PDZ domains of PSD95 and human Dlg (see above) also reveal an important role for the first residue of the second ␣ helix of the domain (the ␣B1 residue). This ␣B1 residue interacts with the residue at the Ϫ2 position of the bound peptide and determines at least in part the specificity of the domain for the peptide. In the PDZ domains of PSD95 and human Dlg the ␣B1 residues are histidines that interact with threonines or serines at the Ϫ2 position of the bound peptides. The Ϫ2 residue in the syndecan peptide is a phenylalanine (t-FYA), and mutating this phenylalanine to alanine indeed abolishes the syndecansyntenin interaction in yeast two-hybrid and overlay experiments (2). Sequence alignments predict that in syntenin the ␣B1 residues of the PDZ1 and PDZ2 domains correspond to serine 171 and aspartate 251, respectively. In attempts to alter the binding specificities of the syntenin PDZ domains and reduce their interactions with wild-type syndecans, we converted their ␣B1 residues into histidines and embedded these in the sequence SHEQ, mimicking the sequence context for the ␣B1 histidines in PSD95 and human Dlg. The exact positions of these mutations (SDK 3 HEQ in PDZ1 and KDS 3 SHE in PDZ2) are represented in Fig. 1. As a complementary test, we also tested whether these mutant PDZ domains would interact with a mutant F(C30)S syndecan-2 cytoplasmic domain, expected to complement for the altered PDZ domain specificity by featuring a serine instead of a phenylalanine at the Ϫ2 position.
In the two-hybrid assay with the wild-type syndecan-2 bait, the mutation of the PDZ1-␣B helix showed no measurable effect, whereas the mutation of the PDZ2-␣B helix (as single mutation or in combination with the PDZ1-␣B mutation) abolished the interaction (Fig. 2C). Substituting F(C30)S syndecan-2 for wild-type syndecan-2 as bait in these assays did not support an interaction with wild-type syntenin or the PDZ1␣B mutant, but it restored the interaction with the PDZ2␣B mutant and the PDZ1␣B/PDZ2␣B double mutant (not shown). In the BIAcore assay (Fig. 2F) using an immobilized wild-type syndecan-2 cytoplasmic domain, the isolated mutation of the PDZ1-␣B helix resulted only in a slight decrease in syntenin binding. The isolated mutation of the PDZ2-␣B helix and the combination of the ␣B helix mutations in both PDZ1 and PDZ2 had clear negative effects. Similar results were obtained in blotting assays, where both the PDZ2␣B mutant and the PDZ1␣B/PDZ2␣B mutant clearly showed reduced binding to wild-type GST-syndecan-2 (Fig. 3). Moreover, unlike wild-type syntenin and the PDZ1␣B mutant, the PDZ2␣B mutant and the PDZ1␣B/PDZ2␣B mutant were both effective at binding to the F(C30)S syndecan-2 mutant. All together the results obtained with the ␣B helix mutants also suggested that both PDZ domains of syntenin were involved in the syntenin-syndecan interaction, but clearly with a larger or more specific contribution of the PDZ2 domain.
Combined Mutations of the ␤A-␤B Carboxylate Binding Loops and ␣B Helices-The combination of the Lys 3 Ala mutations in the carboxylate binding loops and of the ␣B helix mutations (K␣B mutations) in either PDZ1 or PDZ2 or in both PDZ1 and PDZ2 yielded results that were consistent with the above interpretation. In the two-hybrid assay with the wildtype bait, the K119A/PDZ1␣B mutant (K␣B mutation of PDZ1) still interacted with syndecan (Fig. 2D). As could have been expected from the effects of the PDZ2␣B mutations (see above), the K␣B mutation of PDZ2 (K203A/PDZ2␣B) or of both PDZ1 and PDZ2 (K119A/PDZ1␣BϩK203A/PDZ2␣B) suppressed this interaction. In the BIAcore assay, the K␣B mutation of either PDZ1 or PDZ2 had a negative effect on the binding, but the mutation of PDZ2 had a much stronger negative effect than the mutation of PDZ1 (Fig. 2G). Yet only the combination of the K␣B mutations of both PDZ1 and PDZ2 completely abolished binding to the syndecan cytoplasmic domain. A similar result was obtained in the blotting assay, where the K␣B mutation of PDZ2 strongly reduced binding and the double K␣B mutant failed to show any binding to wild-type syndecan-2. None of the K␣B mutants bound to F(C30)S mutant syndecan-2 (Fig. 3).
PDZ Domain Shuffling-All the above results suggest that the two PDZ domains of syntenin, together, are both involved in the syndecan interaction, yet with unequal contributions. However, such tentative conclusion assumes that the different mutations affect the two domains to a similar extent and affect similar binding mechanisms. We explored this further by constructing fusion proteins with artificial PDZ tandems (Fig. 4A) and comparing these to the PDZ1-PDZ2 tandem in syntenin.
Unlike the PDZ1ϩPDZ2 and PDZ2ϩPDZ2 tandems, the PDZ1ϩPDZ1 tandem failed to interact with the syndecan-2 bait in yeast two-hybrid assays (summarized in Fig. 4B). In the BIAcore assay (Fig. 4C), the GST fusion protein with the PDZ1ϩPDZ2 tandem bound to the syndecan-2 peptide. The extent of this binding approximated that measured for wildtype syntenin. Yet, the shape of the binding curve was different, suggesting much faster association and dissociation kinetics. The GST fusion protein with the PDZ1ϩPDZ1 tandem, in contrast, bound poorly. Unfortunately, a soluble GST fusion protein with a single PDZ2 or with the PDZ2ϩPDZ2 tandem could not be obtained. A consistent result was obtained in the blotting assay, where the GST fusion protein with the PDZ1ϩPDZ2 tandem, but not that with the PDZ1ϩPDZ1 tandem, bound to the syndecan-2 fusion protein (Fig. 4D). As predicted from prior results (2), GST fusion proteins containing only one copy of PDZ1 did not bind to syndecan in this assay (Fig. 4D).
Isothermal Titration Calorimetry-Since the above results suggested syntenin binds to immobilized syndecans using two PDZ domains but not necessarily by both its PDZ domains, we attempted to obtain independent evidence for the number of syndecan-binding sites in syntenin. For that purpose, syntenin was expressed as a syntenin-intein-chitin binding domain fusion protein and recovered as a tag-free protein after inteinmediated self-cleavage of the chitin-bound fusion protein. Like the GST-syntenin fusion protein, this tag-free syntenin bound to a syndecan-2 peptide that was immobilized on sensor chips (Fig. 4C). The tag-free fusion protein was then used for isothermal titration calorimetry experiments. Repetitive injections of the syndecan-2 cytoplasmic domain peptide in a cell filled with syntenin yielded only very low binding heat values, but a binding isotherm with a breakpoint at molar ratio 2, consistent with a model that supposes two peptide-binding sites in syntenin (Fig. 5).
Alternative Bait for Syntenin-To test whether a two-PDZ domain requirement also applied to syndecan bait other than syndecan-2, similar yeast two-hybrid and blotting experiments were performed with GAL4 and GST fusion proteins that contained the cytoplasmic domains of the syndecans -1, -3, and -4. In yeast two-hybrid assays, all syndecans acted as bait for syntenin or the paired PDZ domains of syntenin (not shown). All syndecans showed a similar requirement for an integer PDZ2-␣B helix in syntenin (as shown for syndecan-1 and -2 in Fig. 2). As a GST fusion protein, syntenin bound equally well to the four different syndecans in blotting experiments (Fig. 6), in contrast to CASK-PDZ, which seems to bind preferentially to syndecan-2 and -4. Thus, unlike syntenin, CASK can interact with syndecans, engaging only a single PDZ domain, and discriminates to some extent between the different syndecans.
Syntenin also binds to B-class ephrins (16,17). In blotting experiments, the bindings of the various GST-syntenin fusion proteins to GST-ephrin closely mimicked their bindings to GST-syndecan-2 (Fig. 7). Since syndecan-2 interacts with the PDZ domain of CASK (Fig. 6) and CASK also binds to neurexins, we also tested whether syntenin would interact with neurexins. When the complete cytoplasmic domain of human neurexin-2 was used as bait in the two-hybrid system an interaction (both growth on His Ϫ and ␤-Gal staining) was observed. When GST-neurexin was immobilized on the blot and probed with GST-myc-syntenin fusion proteins in the overlay assay, we obtained clear binding signals not only with syntenin or the PDZ1ϩPDZ2 tandem but also with the PDZ1ϩPDZ1 tandem (Fig. 7). GST-syntenin fusion proteins containing a combination of the lysine and ␣B1 mutations in either PDZ1 or PDZ2 also bound to neurexin (Fig. 7). GST fusion proteins containing the combined lysine and ␣B1 mutations in both PDZ1 and PDZ2 did not bind to immobilized neurexin (Fig. 7). Thus, as observed for the interaction with syndecan, the interaction of syntenin with neurexins and B-class ephrins appears to depend on the cooperation of PDZ1 and PDZ2 but at least PDZ1 appears to have higher intrinsic affinity for neurexins than for syndecans and B-class ephrins.
To assess the spectrum of the major potential types of syntenin bait in cells, we also performed ligand overlay assays on blots of whole cell extracts obtained from heparitinase-treated human cells (Fig. 8). Only GST-syntenin fusion proteins that contained both PDZ1 and PDZ2 produced signals, binding to a series of prominent bands. Parallel staining of these blots with syndecan-specific antibodies (directed against epitopes in the cytoplasmic domains of these molecules) indicated that these bands corresponded to syndecan core proteins or core protein fragments. Thus, at least in these cells, syndecans figure among the quantitatively most important bait for the paired PDZ domains of syntenin. DISCUSSION The localization and clustering of cell surface receptors to specific subcellular positions can be critical for their proper functioning in the sending and receiving of signals. Syntenin is an adapter-like PDZ protein that binds to the cytoplasmic domain of a number of important signaling and adhesion molecules: syndecans (2), ephrins (16,17), eph-receptors (16), and, as we show here, also neurexins. Deletion of either one of the two PDZ domains from syntenin abolishes or strongly reduces the interaction of syntenin with all the above partner proteins. Together with the present results, this suggests that the interaction of syntenin with all these proteins is PDZ-mediated but that its PDZ domains do not function as independent binding domains. Each domain needs the assistance of the other PDZ domain in a two-pronged binding that engages two contiguous PDZ domains and two contiguous bait peptides. In general, couples of PDZ domains whose interactions are weak taken individually would appear to be designed to bind to specific oligomeric structures rather than to drive the assembly of such structures. Syntenin could therefore be a "detector" of receptor clustering or co-clustering. Two compatible receptors that occur in synteny (the state of being together in location) together bind syntenin. The multiple in vitro interactions or potential partners of syntenin suggest that the PDZ domains of syntenin have a somewhat relaxed specificity. Yet, the promiscuity of syntenin appears limited, and the PDZ domains of syntenin cannot invariably be substituted one for the other in the various binding interactions with syndecans, ephrins, and neurexins. This lack of cross-complementation suggests that the two PDZ domains of syntenin differ somewhat in their affinities for a specific partner and possibly in their partner preferences. Syntenin may therefore bind to and discriminate between a number of specific supramolecular complexes or structures that include assemblies and mixed assemblies of syndecans and other suitable partners. Many ligands for cell surfaces with signal functions contain multiple binding domains, including heparin binding domains that could be instrumental in driving these syndecan assemblies.
PDZ domains are widespread among signaling and cytoskeletal proteins and are thought to localize the activities of these proteins to appropriate sub-membranous protein complexes FIG. 4. The combination of PDZ1 and PDZ2 is the minimal requirement for the syntenin-syndecan interaction. A, schematic representation of the tandem constructs. B, summary of the two-hybrid results obtained with these PDZ tandems. C, BIAcore curves for the interaction of non-tagged syntenin (syntenin Val 3 Ala, produced by self-splicing from a syntenin-intein fusion protein) and of GST-myc-syntenin, PDZ 1ϩ2, PDZ1ϩ1, or PDZ1 fusion proteins, with the wildtype (wt) syndecan-2 cytoplasmic domain peptide. D, ligand overlay results obtained with these fusion proteins, excerpted from the data shown in Fig. 7. RU, relative units. (18). Most known PDZ domain interactions occur through recognition of short C-terminal peptide motifs, which include the terminal carboxylate of the ligand. Binding specificity is achieved by domain interactions with the residue at the Ϫ2 position of the bound peptide, serine/threonine (coordinating with histidines near the start of the ␣B-helix) in the case of type-I domains and hydrophobic or aromatic residues (coordinating with non-basic residues in the ␣B-helix) in the case of type-II domains. Syntenin binds to C-terminal peptide structures such as t-FYA (syndecans), t-YKV (ephrin-Bs), t-IQV (Eph A7), and t-YYV (neurexin) and features Ser as the ␣B1 residue in PDZ1 and Asp as the ␣B1 residue in PDZ2. Moreover, syntenin does not bind to t-SYA, as in syndecan-2 Overlays of blotted GST fusion proteins representing the cytoplasmic domains of syndecan-2, neurexin-2, and ephrin-B1 as well as the C30 deletion mutant (missing the last two residues) of the syndecan-2 cytoplasmic domain, the C-terminal peptide EFYA (C2 region) that is common to all syndecans, and the C-terminal sequence of Kv1.4 that corresponds to a class I PDZ-interacting peptide. An anti-GST antibody shows that, except for the ephrin construct that appeared to be unstable and partially degraded, equal amounts of GST fusion proteins were loaded on the gels and transferred. These blots were incubated with syntenin-FL (full-length), K␣B mutant syntenins, and PDZ tandem GST fusion proteins. Bound proteins were detected with the anti-syntenin mAb 1C8 or the anti-myc antibody 9E10. The K␣B mutation in PDZ2 severely reduces the syntenin interaction with syndecan and ephrin but less significantly than with neurexin. The PDZ1ϩ1 tandem binds only significantly to neurexin.
FIG. 8. Syndecans are major bait for syntenin. Extracts from heparitinase-treated MCF7 cell suspensions were fractionated by SDS-PAGE and blotted on nitrocellulose membranes. Heparitinase treatment, which removes the GAG chains, and suspension of the cells, which induces syndecan shedding, ensures that the syndecan core proteins and cytoplasmic domains will migrate as discrete bands. The blot was incubated with GST-myc-syntenin, and bound fusion protein was detected with the anti-myc antibody 9E10. In parallel staining, the blots were incubated with mAb 2E9 (directed against the cytoplasmic domains of the syndecans 1 and 3) and mAb 6G12 (directed against the cytoplasmic domain of syndecan-2, but also showing some cross-reactivity with the cytoplasmic domains of the other syndecans). The major bands highlighted by GST-myc-syntenin correspond to the bands stained by the anti-syndecan antibodies. F(C30)S, indicating that at least one of its PDZ domains is a type-II domain. Mutations of either the carboxylate binding lysines or the ␣B1 residues in either PDZ1 or PDZ2 all affect the syndecan-syntenin binding, and combinations of these mutations have additive negative effects on this binding. All this together is compatible with a bivalent binding model, engaging a syndecan dimer (acting as a plug) and the two type-II PDZ domains of syntenin (functioning as a socket).
Nonetheless, there is an increasing number of PDZ-mediated interactions that do not conform to the canonical mode of PDZ recognition and where the interaction is based on the recognition of internal peptide motifs. Crystal structures indicate that, in this case, a ␤-hairpin finger docks in the peptide binding groove, and a sharp ␤-turn replaces the normally required C terminus. Such alternative internal interaction modes support the heterotypic dimerization of the PDZ domain of neuronal nitric-oxide synthase with PDZ domains from PSD95 and syntrophin (15,19). The demonstration of interactions between PDZ domains in heterodimeric complexes raises the further possibility that intramolecular PDZ-PDZ interactions may serve to regulate the conformations and functions of proteins with two or more PDZ domains (20). All this indicates that tandem arrays of PDZ domains may mediate multiple interactions and complicates the interpretation of the effects of deletions in terms of binding assignments. The effects of the point mutations that we introduced in the PDZ domains of syntenin confirm that directly or indirectly both domains are involved in the binding of syntenin to di-or multimeric syndecans. A single PDZ domain is clearly insufficient for syndecan binding, but the effects of the mutations could also result from disruptions or weakening of homotypic PDZ-PDZ interactions that support the formation of syntenin dimers or multimers. In that case, it remains unclear what PDZ domain directly mediates binding to the immobilized syndecan peptide. PDZ1-supported syntenin dimers may bind to syndecans via PDZ2 (dimers), or alternatively, PDZ2-supported syntenin dimers may bind to syndecans via PDZ1 (dimers). The first of the two alternatives would be consistent with the binding modes observed for the various PDZ tandem constructs, since the PDZ1ϩPDZ1 tandem fails to bind syndecans, whereas the PDZ2ϩPDZ2 tandem does interact (although the latter could only be deduced from yeast twohybrid experiments). However, full-length syntenin does not interact with full-length syntenin in yeast two-hybrid and surface plasmon resonance experiments (not shown). A full-length syntenin that is produced in a self-splicing intein expression system (which allows the production of recombinant protein that is free of any extensions such as GAL4 or GST) does not show any evidence for self-association in gel filtration and cross-linking experiments (not shown). Moreover, highly sensitive isothermic micro-calorimetry experiments, using syntenin produced from intein fusion proteins and free peptide, can detect a syntenin-syndecan interaction in solution and suggest a 1:2 syntenin-syndecan binding stoichiometry. All together, this strongly favors a model wherein stable binding results from the engagement of two intramolecular PDZ domains and two oligomerized syndecans. Although both PDZ domains of syntenin appear to directly participate in the binding to syndecan, the consistently more pronounced effects of the PDZ2 mutations suggest that PDZ2 contributes to a larger extent than PDZ1. Relative affinity estimates based on the current data can only be very tentative, but all together the results are compatible with a model where the contribution of PDZ2 is twice as strong as that of PDZ1, and binding to syndecan requires a combination of strengths that exceeds that of a single PDZ2 domain.
In the yeast two-hybrid assay, several bait fusion proteins are bound to the GAL4-responsive promoter and are, thus, automatically presented as a homotypic assembly. This display system would appear neutral with respect to the detection of PDZ domains that function in isolation and well suited for the detection of coupled PDZ domains with similar or shared specificities. Yet, it may be strongly negatively biased with respect to co-operative PDZ domains with distinct target specificities. Syndecan bait is very efficient at recruiting syntenin prey, which invariably contain the combination of PDZ1ϩPDZ2 and typically represent 80% of the clones isolated from a hippocampal library screen. Unless over-represented in this library because of the relative abundance of the corresponding transcript, this result suggests that homotypic assemblies of syndecans selectively or preferentially recruit syntenin. The overlay experiments suggest that homotypic assemblies of Bephrins or neurexins may do this with similar or even greater efficiencies. This does not ensure that this bait represents the optimal fit (and therefore potentially natural partners) for the syntenin-PDZ domains. At minimum, our failures to detect binding partners in yeast two-hybrid screens using single syntenin PDZ domains as bait 2 and in ligand overlays of blotted cell extracts that use GST with a single syntenin PDZ1 domain as a probe support the contention that the PDZ domains of syntenin do not work in isolation. Yet, whereas strong binding to any of the arrayed bait requires the combination of PDZ1 and PDZ2, the relative contributions of PDZ1 and PDZ2 appear to not always mimic the hierarchy observed in the syndecansyntenin binding. The separate mutations of PDZ1 and PDZ2 affect the binding to neurexin less differentially than they affect the binding to syndecan or ephrin-B1, and the combination of PDZ1ϩPDZ1 binds to neurexin but not to syndecan or ephrin-B1. By inference this implies that at least PDZ1 adapts better to neurexins than to syndecans or B-class ephrins. Both the PDZ domains of syntenin may have a somewhat relaxed specificity but differential preferences for different subsets of type-II C termini. Differential peptide preferences of PDZ1 and PDZ2 imply that in a physiological setting syntenin may bind to specific heterotypic assemblies rather than to homotypic assemblies that would represent a less than optimal fit. As PDZ1 prefers neurexin over syndecan-2, the combination of a neurexin and a syndecan could potentially figure as an example of a supramolecular complex that represents a better fit for syntenin than homotypic syndecan assemblies. Neurexin is present at the presynaptic side of the neuronal synapse. Syndecan-2 accumulates in central neuronal synapses, where it appears to be specifically associated with both the postsynaptic density and the presynaptic terminal (21), indicating that from the known syndecan and neurexin expressions, this suggestion could make sense. Resolving this issue will require the identification of the bait that is physiologically associated with syntenin in cells. It is clear that there are also different PDZ proteins competing for the same bait. Both CASK and syntenin bind to neurexins and syndecans. Syntenin is widely expressed. CASK is appropriately expressed and localized to interact with both syndecan-2 and -3 in different compartments of the neuron throughout postnatal development (21). As we show here, the single PDZ domain of CASK binds to all four syndecans but clearly more easily to the syndecans -2 and -4 than to the syndecans -1 and -3, which respects the structural similarities among the syndecans. Indeed, the sequence immediately upstream of the EFYA terminus of syndecan-1 is highly similar to the corresponding region in syndecan-3, and the corresponding region in syndecan-2 is highly similar to that in syndecan-4 but more distantly related to these sequences in the syndecans -1 and -3. Syntenin, in contrast, does not show a similar preference and binds equally well to all four syndecans. Ligand overlays suggest that syntenin binds better to assemblies of syndecan or neurexin bait than CASK binds to this same bait, but we do not have data that allow a precise quantitative comparison of the relative strengths of the neurexin/syndecan-CASK and neurexin/syndecan-syntenin bindings. Yet, should these be of the right order, it is conceivable that regulated co-expressions and associations of syndecans and neurexins could result in a switch in transmembrane connections and signaling from CASK-mediated to syntenin-mediated modes. Such a recruitment model could fit in a broader "dual" receptor concept, where switches in cytoplasmic connections are supported by ligands with cooperative receptor binding domains that realize the required syntenies of the receptor cytoplasmic domains. Clearly, further investigations are needed to substantiate these proposals.