The neural cell recognition molecule neurofascin interacts with syntenin-1 but not with syntenin-2, both of which reveal self-associating activity.

Neurofascin belongs to the L1 subgroup of the immunoglobulin superfamily of cell adhesion molecules and is implicated in axonal growth and fasciculation. We used yeast two-hybrid screening to identify proteins that interact with neurofascin intracellularly and therefore might link it to trafficking, spatial targeting, or signaling pathways. Here, we demonstrate that rat syntenin-1, previously published as syntenin, mda-9, or TACIP18 in human, is a neurofascin-binding protein that exhibits a wide-spread tissue expression pattern with a relative maximum in brain. Syntenin-1 was found not to interact with other vertebrate members of the L1 subgroup such as L1 itself or NrCAM. We confirmed the specificity of the neurofascin-syntenin-1 interaction by ligand-overlay assay, surface plasmon resonance analysis, and colocalization of both proteins in heterologous cells. The COOH terminus of neurofascin was mapped to interact with the second PDZ domain of syntenin-1. Furthermore, we isolated syntenin-2 that may be expressed in two isoforms. Despite their high sequence similarity to syntenin-1, syntenin-2alpha, which interacts with neurexin I, and syntenin-2beta do not bind to neurofascin or several other transmembrane proteins that are binding partners of syntenin-1. Finally, we report that syntenin-1 and -2 both form homodimers and can interact with each other.

Neurofascin is a member of the L1 subgroup of the immunoglobulin superfamily of cell adhesion molecules, which also includes L1 (NgCAM) 1 itself, CHL1, NrCAM in vertebrates, neuroglian in insects, and tractin in leeches. These transmembrane glycoproteins share a well conserved overall domain organization with six immunoglobulin-like and four to five fibronectin type III (FNIII)-like domains. Their diverse homo-and heterophilic interactions mediate cell-cell contacts and can promote neuronal migration, axonal growth, and fasciculation in the developing nervous system (1)(2)(3). A crucial role of the L1 subgroup in neural development is exemplified by a range of neuroanatomical and neurological disorders caused by knockout of the murine L1 gene (4 -6) and by mutations in the human L1 gene, which affect L1 binding activity and trafficking (7,8).
Unlike L1 and other subgroup members, neurofascin is subjected to extensive alternative splicing that is regulated during embryonic development of the chicken brain (9). This differential splicing has been shown to modulate interactions of neurofascin with axonal NrCAM, F11, axonin-1, and the extracellular matrix protein tenascin-R, and to influence neurite extension in vitro (10,11). Specific isoforms of neurofascin are localized to initial axon segments of Purkinje cells and to the nodes of Ranvier of myelinated nerves, where they interact with the cytoskeleton adapter-protein ankyrin-G (12). In particular, an oligodendrocyte-specific form of neurofascin (termed NF155) was found to localize to the paranodal region, whereas a neuron-specific form (NF186) was confined to the nodal region (13,14). Ankyrin binding appears to be a common feature of all L1-type molecules and is thought to stabilize cell adhesion (15)(16)(17)(18). Interaction with ankyrin requires a highly conserved sequence within the cytoplasmic tails of L1 subgroup members and is inhibited by its tyrosine phosphorylation as demonstrated for neurofascin (19 -21). Furthermore, palmitoylation of neurofascin at a highly conserved cysteine residue in its membrane-spanning segment might affect the targeting of neurofascin to specialized plasma membrane microdomains (22). L1CAM-mediated cellular processes may also be regulated by changes in the expression levels of CAMs on the cell surface. Tyrosine phosphorylation of the endocytic motif YRSL, which represents a binding site of the AP-2 clathrin adaptor complex, regulates not only the internalization of the neuronal L1 form but probably also of NrCAM and neurofascin (23,24). Recently, cross-linking of L1 expressed in heterologous cells has been shown to trigger the activation of ERK2, a component of the MAPK signal cascade. ERK2 activation appears to be coupled with L1 internalization and phosphorylation of two cytoplasmic serines that are conserved in the L1 subgroup (25).
Although the cytoplasmic tails are the most conserved segments of the L1-type molecules, there are also some differences, particularly at their COOH termini. These differences might provide the structural basis for individual intracellular interactions and therefore distinct functional features within the L1 subgroup. To identify proteins that might mediate sig-naling, spatial targeting, or trafficking of L1-type molecules by direct interaction with their cytoplasmic segments, we performed yeast two-hybrid screens of brain cDNA libraries. Here, we demonstrate that syntenin-1 is an intracellular binding partner of neurofascin but not of L1 or NrCAM. Syntenin-1 contains two PDZ domains. PDZ domains are multifunctional protein-binding modules, which were first identified in PSD-95, DlgA, and ZO-1, and are now found in a growing number of other cytoplasmic proteins (26). The PDZ domains of syntenin-1 have been previously shown to interact with the COOH termini of syndecans, class B ephrins, EphA7, pro-TGF-␣, neurexins, and the anion exchanger AE2 (27)(28)(29)(30)(31)(32). In this study, we identified the second PDZ domain of syntenin-1 as a binding site of the COOH terminus of neurofascin and of several other transmembrane proteins mentioned above. Neurofascin was found not to interact with syntenin-2␣ or -2␤, two isoforms of a novel protein closely related to syntenin-1. Furthermore, we observed a homo-and heterodimerization of syntenin-1 and syntenin-2 that appears to involve larger portions of these molecules. This capacity for self-association might be crucial for homo-and heterotypic clustering of neurofascin and other syntenin-binding proteins.

EXPERIMENTAL PROCEDURES
cDNA Constructs Used in the Yeast Two-hybrid System-cDNAs encoding cytoplasmic segments of chick neurofascin and NrCAM, both wild-type and mutants, as well as of rat L1, human syndecan-3, pro-TGF-␣, ephrin-B2, and EphA7, were obtained by PCR using specific primers and inserted in-frame into the pGBT9 vector. pGAD10 vector containing cDNA corresponding to the 435 COOH-terminal amino acid residues of human neurexin I-␣ was obtained by a yeast two-hybrid screen with syntenin-1 as a bait. 2 cDNAs encoding cytoplasmic segments of neuroglian-167 and -180 were subcloned into the pGBT9 vector from pRIT3 plasmids provided generously by M. Hortsch (University of Michigan, Ann Arbor, MI).
The rat syntenin-1 cDNA that was isolated by the yeast two-hybrid screen was subcloned from the library vector pGAD10 into the pGBT9 and pGAD424 (all from CLONTECH). cDNA fragments encoding the 115 NH 2 -terminal amino acids, the 30 COOH-terminal amino acids, or lacking the 101 NH 2 -terminal amino acids of rat syntenin-1 were amplified by PCR using specific primers and inserted in-frame into the pGBT9 and pGAD424 vectors. A syntenin-1 construct encoding the NH 2 -terminal segment, PDZ1 and the first seven amino acids of PDZ2 was generated from pGAD424(ST-1) by restriction digestion with NsiI and PstI and ligation of the overlapping plasmid ends. The cDNA of a mutant lacking 141 NH 2 -terminal amino acids, including 29 residues of the PDZ1 domain, was generated from pGAD10(ST-1) by restriction digestion with XhoI and SpeI followed by filling-in and in-frame ligation of the plasmid ends. The amino acid substitutions G128E and G212D within the PDZ1 and PDZ2 of syntenin-1, respectively, were generated by site-directed mutagenesis using the Transformer site-directed mutagenesis kit (CLONTECH). The corresponding G383A and G635A nucleotide substitutions were introduced into the rat syntenin-1 encoding cDNA by simultaneously annealing the mutagenic primers 5Ј-pG-GATCAAGATGGAAAAATTGAGCTCAGACTGAAG and/or 5Ј-pG-CAGTGGACATGTTGACTTTATCTTTAAAAGTGG, respectively, and the selection primer 5Ј-pCCCTGACTTTCTCGACTTGGT, which carried a mutation of the single XbaI site within the sequence of syntenin-1, to one strand of the denatured pGAD10(ST-1) plasmid. The second strand was synthesized subsequently, and the selection procedure was performed as described in the manufacturer's manual. The resulting full-length cDNA mutants PDZ1*, PDZ2*, and PDZ1*2* were subsequently inserted also into the pGBT9 vector.
All constructs described here and below were verified by automated sequence analysis using the Auto-Read sequencing kit (Amersham Pharmacia Biotech).
Yeast Two-hybrid System-The pGBT9 constructs of wild-type neurofascin, NrCAM and L1, were used to screen an adult rat brain cDNA library cloned into the pGAD10 vector (CLONTECH). The two-hybrid screens were performed with the yeast HF7c reporter strain according to the instructions of the distributor (CLONTECH). Prey plasmids isolated from the positive clones were retransformed together with the baits or control constructs into the yeast reporter strains SFY526 or Y187 for further analysis. The library cDNA clones revealing specific binding in all strains were sequenced as mentioned above.
Cell Cultivation and Transfection-COS7 and L929 cells were cultivated in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, penicillin, and streptomycin at 37°C and 5% CO 2 . COS7 cells were transiently transfected by the DEAE-dextran/Me 2 SO method as described previously (11). To transiently transfect L929 cells, cells were harvested from 10-cm plates with trypsin, washed twice, and resuspended in 0.4 ml of RPMI with 100 M dithiothreitol. Electroporation of the cell suspension was performed in a 4-mm cuvette with 12 g of DNA using an EasyjecT PLUS device (Eurogentec) at 350 V, 600 microfarads, and room temperature. Immediately after electroporation, cells were transferred in a 5-cm PetriPERM dish (Hereaus) precoated with 0.5 mg/ml collagen type I (Sigma) and cultivated for 48 h under standard conditions.
Ligand-overlay Assay-COS7 cells were transfected with pSG5 expression vectors (Stratagene) encoding NgCAM (gift of P. Sonderegger, University of Zü rich), chick neurofascin isoform Nf17 (10,33), or a truncated GPI-anchored construct Nf17GPI. In the latter construct, the transmembrane and the cytoplasmic domain of neurofascin isoform NF17 were replaced by the COOH-terminal region of the cell adhesion molecule F11 (34) containing a part of the third and the complete fourth FNIII domain as well as the GPI anchor signal. After 48-h cultivation, cells were solubilized in radioimmune precipitation buffer and centrifuged as described above. NgCAM and neurofascin were immunoprecipitated from cell lysates using specific monoclonal antibodies (35,36) and Protein G-agarose (Roche Molecular Biochemicals) followed by 2 M. Koroll, and F. G. Rathjen, unpublished data. 7.5% SDS-PAGE and blotting to PVDF membranes (Amersham Pharmacia Biotech). Blots were either stained with corresponding antibodies or incubated overnight at 4°C with 40 g/ml MBP-syntenin-1 or MBP-␤-galactosidase fusion proteins in PBS, pH 7.4, supplemented with 0.2% bovine serum albumin and 0.05% Tween-20. Ligand binding was visualized using anti-MBP antiserum (1:10,000, New England Bio-Labs), horseradish peroxidase-conjugated secondary antibodies (1:10,000, Dianova), and a Metal Enhanced DAB substrate kit (Pierce).
Analysis of Colocalization of Neurofascin and Syntenin-1-L929 cells were transiently cotransfected with EGFP-syntenin-1 and wild-type neurofascin or the truncated GPI-anchored neurofascin construct described above. To induce and visualize clustering of neurofascin, cells were incubated in culture medium with 30 g/ml Fab fragments of rabbit anti-neurofascin IgGs (35) followed by Cy5-conjugated goat antirabbit secondary antibodies (1:100, Dianova), for 1 h each, at the same conditions. After washing with warm medium and fixation with ice-cold 4% formaldehyde for 10 min, cells were covered with 50% glycerol in PBS and processed for confocal imaging using a Bio-Rad MRC-1000 system.
Surface Plasmon Resonance Analysis-200 RU (resonance units) of synthetic neurofascin peptides corresponding to the last 15 COOHterminal amino acids of either its wild-type or the A(0)S-substituted sequence and containing an additional NH 2 -terminal lysine residue (Biosyntan, Berlin) were immobilized on the CM5 sensor chips activated by 50 mM N-hydroxysuccineimide, 200 mM N-ethyl-NЈ-(dimethylaminopropyl)carbodiimide. Binding of the immobilized surfaces to MBP-syntenin-1 or MBP-␤-galactosidase fusion proteins at the indicated concentrations and 5 g/ml flow rate was analyzed using a BIAcore X instrument (Pharmacia Biosensor) as described elsewhere (37). Data were analyzed by nonlinear curve fitting using the BIAevaluation software (Pharmacia Biosensor).
Gel Filtration Chromatography-Purified MBP-syntenin-1 fusion protein (30 g) was analyzed using the SMART System on a Superdex 200 PC 3.2/30 column (Amersham Pharmacia Biotech) equilibrated with 20 mM Hepes/HCl, pH 7.5, 150 mM NaCl at a flow rate of 50 l/min. Fractions of 80 l were analyzed by SDS-PAGE followed by Coomassie staining. The chromatogram of the molecular mass standards (Bio-Rad) was monitored under the same conditions to generate a calibration curve.
Immunocoprecipitation-To analyze the self-association behavior of syntenin-1 tagged with Myc or FLAG epitopes, syntenin-1 cDNA was inserted in-frame into the pMyc-CMV (CLONTECH) and pFLAG-CMV-2 (Sigma) expression vectors, respectively. 48 h after transfection with one or both of these constructs, COS7 cells cultivated on 15-cm plates were washed with ice-cold PBS followed by solubilization in 1 ml of IP buffer (50 mM Hepes, pH 7.4, 150 mM NaCl, 1 mM EDTA, 10% glycerol, 1% Triton X-100) containing protease inhibitors at 4°C. After 20-min centrifugation at 100,000 ϫ g and 4°C, 0.7 ml of clarified lysates was incubated for 4 h with 3 g of 9E10 anti-Myc monoclonal antibodies (CLONTECH) or 5 g of anti-FLAG M5 (Sigma) and 10 l of Protein G-agarose (Roche Molecular Biochemicals). The immunoprecipitates were collected by centrifugation at 13,000 rpm for 1 min and washed three times with ice-cold IP buffer. For Western blotting, the immunocomplexes were denatured by boiling with 40 l of SDS sample buffer followed by 10% SDS-PAGE and electrotransfer to a PVDF membrane (Amersham Pharmacia Biotech). Myc-and FLAG-syntenin-1 were recognized by using 1 g/ml 9E10 anti-Myc or 2 g/ml M5 anti-FLAG (Sigma) primary antibodies and 1:5000 dilution of AP-conjugated rabbit anti-mouse IgG (Dianova).

Identification of Rat Syntenin-1 as a Neurofascin-binding
Protein-To identify proteins that interact with the cytoplasmic segments of the L1-type cell adhesion molecules, we performed several yeast two-hybrid screens. One such screen of a rat brain cDNA library using the cytoplasmic segment of chicken neurofascin as a bait (90% identity with rat neurofascin at the level of the amino acid sequence) resulted in the identification of eight positive clones out of 6 ϫ 10 6 HF7c yeast transformants. The 1.3 kb cDNA inserts isolated from three positive clones were identical and encoded a 432-amino acid residues segment of NH 2 -terminal repeats of erythrocyte ankyrin. This is consistent with the known interaction of neurofascin with ankyrins (15). The 2-kb-long prey cDNA from the other five positive clones consisted of a 900-bp open reading frame flanked by 15 bp of the 5Ј-and 1.1-kb-long 3Ј-uncoding region. Conceptual translation of the open reading frame revealed a sequence of 300 amino acids that shows 91% identity with human syntenin (27) and can be considered to represent the rat homologue of human syntenin. Subsequently, it will be referred to as syntenin-1 (see below) (Fig. 1A). Syntenin-1 is a cytoplasmic protein consisting of a tandem of two PDZ domains flanked by a NH 2 -terminal segment of 112 amino acid residues and a short COOH-terminal stretch of 26 residues. Both these flanking segments do not show any significant similarities to any known polypeptide modules. Furthermore, on the basis of various data base entries we cloned two isoforms of a protein, termed syntenin-2 in the following, which are highly related at the amino acid level (70% identity over the PDZ domains) and in their overall domain organization to syntenin-1 (Fig. 1,  A and B). The stop codon containing the 5Ј-uncoding region of the cloned cDNA of the shorter isoform, designated syntenin-2␤, which lacks almost the complete NH 2 -terminal segment, is FIG. 1. Amino acid sequences of rat syntenin-1 and human syntenin-2 and their overall domain organization. A, alignment of rat syntenin-1 with mouse and human syntenin-1 and with human syntenin-2. Rat syntenin-1 was identified by its interaction with neurofascin in a yeast two hybrid screen. Two isoforms of human syntenin-2, ␣ and ␤, were cloned on the basis of the GenBank TM data base entries with accession numbers AL136531, AF131809, AF159228 and analyzed by sequencing. Differences between the amino acid sequences are emphasized in black. The PDZ1 domain is underlined, whereas PDZ2 is underlined twice. Conserved glycine residues within the carboxylate-binding loops that were mutated (see Table II) are indicated by asterisks. Amino acid residues positions are given at the right. B, schematic representation of the overall domain organization and homology of syntenins. NH 2 termini are at the left side. Percentages indicate amino acid sequence identity between corresponding domains of rat and human syntenin-1 and -2. Amino acid positions are printed above each scheme. Positions of the start methionine of syntenin-2␤ and of the putative ␤-isoforms of syntenin-1 are indicated in brackets. distinct from that of the longer 2␣-isoform. This finding might be taken as evidence that syntenin-2␤ is generated by alternative splicing instead of using the methionine residue at position 86 within the sequence of syntenin-2␣ as an alternative translational start site (data not shown and Fig. 1, A and B).
To study the tissue expression pattern of syntenin-1 and to compare it with neurofascin, we generated antibodies against bacterially expressed GST-syntenin-1 fusion protein in rabbits. To rule out a possible cross-reactivity with syntenin-2, Western blots of syntenin-1, -2␣, -2␤ or EGFP fusion proteins of these expressed in COS7 cells were analyzed by generated antibodies. Syntenin-1 migrated with a molecular mass of ϳ36 kDa (predicted 32.4 kDa), whereas syntenin-2␣ (predicted molecular mass 34.4 kDa) was weakly detected as a single band at ϳ39 kDa, indicating that both proteins can be distinguished on the basis of their apparent molecular masses in SDS-PAGE ( Fig. 2A). Although syntenin-1 and -2 were expressed at equal amounts in COS7 cells as judged from the staining intensities of EGFP-syntenin-1 and -2 fusion proteins using antibodies against GFP (Fig. 2B), comparison of the staining intensities demonstrates that the anti-syntenin-1 antibodies reacted only very weakly with syntenin-2␣ and do not detect syntenin-2␤ (Fig. 2C). To analyze the expression pattern of syntenin-1 and neurofascin, Western blots of detergent extracts of various rat tissues were performed. In contrast to neurofascin, which was found to be expressed exclusively in the brain (Fig. 2E), syntenin-1 with an apparent molecular mass of ϳ36 kDa, as also observed in transected COS7 cells, revealed a wide-spread expression pattern (Fig. 2D). Comparison of the staining intensities indicated strongest expression of syntenin-1 in brain followed by testis, lung, and heart. The lowest level of syntenin-1 staining was detected in skeletal muscles and liver. Syntenin-2 was not detected in this blot, most likely due to the low crossreactivity of the anti-syntenin-1 antibodies as described above. The observed tissue expression pattern of syntenin-1 suggests that functions of syntenin-1 are not restricted to neurofascin, consistent with former studies on syntenin/TACIP18.
Syntenin-1 but Not Syntenin-2 Binds to the COOH Terminus of Neurofascin-Most PDZ domains investigated so far bind directly to the COOH termini of transmembrane proteins. The specificity of these interactions is determined by the structural features of the respective PDZ binding pockets and the COOHterminal amino acids of transmembrane molecules (38). We therefore investigated whether syntenin-1 interacts with the COOH terminus of neurofascin (SLA-COOH in chick and rat) and with the COOH termini of other L1 subgroup members. In particular, NrCAM and neuroglian share a class I PDZ binding motif (S/T)X(V/I) at their COOH termini. First, we tested different deletion and substitution constructs in the yeast twohybrid system. Fourteen of the most COOH-terminal amino acids of neurofascin were sufficient to bind syntenin-1, whereas deletion of the alanine residue at the position 0 abolished the interaction completely. This demonstrates that the COOH terminus of neurofascin is the binding site for syntenin-1 (Table I). Moreover, we showed that the interaction with syntenin-1 is not affected by the alternative splicing of the neurofascin cytoplasmic exon, which encodes the four-amino acid residue stretch RSLE. This sequence motif is also differentially spliced in L1/NgCAM and NrCAM (1). In contrast, syntenin-1 interacted only with the long, nervous system-specific splice isoform (COOH-terminal tripeptide TYV) of the intracellular segment of Drosophila neuroglian but not with its short form (KGL-COOH) that is widely expressed (39). As expected, syntenin-1 failed to interact with the cytoplasmic tail of L1 (ALE-COOH) and, surprisingly, with that of NrCAM (SFV-COOH). Because the fourth vertebrate member of the L1 group CHL1 does not contain any appropriate COOH-terminal PDZ binding motif (LRA-COOH), similarly to L1, its interaction with syntenin-1 was not analyzed. The cloning of syntenin-2␣ and -2␤ allowed us also to test whether they bind to any member of the L1 subfamily or to the other transmembrane proteins targets that are known to interact with syntenin-1. Among these, only neurexin I, which shares its COOH-terminal sequence with neur- against GFP recognized EGFP as a 31-kDa band, whereas EGFP-syntenin-1, EGFP-syntenin-2␣, and EGFP-syntenin-2␤ are detected as 62-kDa, 63-kDa, and 53-kDa polypeptides, respectively. C, anti-syntenin-1 antibodies recognized EGFP-syntenin-1 as a band corresponding to 62 kDa and several proteolytic degradation products. Although EGFP fusion proteins are expressed at a similar level (B), EGFP-syntenin-2␣ is detected very weakly (63 kDa, not visible on this copy) and -2␤ is not recognized by antibodies against syntenin-1. D and E, Western blot analysis of syntenin-1 and neurofascin in various rat tissues: heart (1), brain (2), spleen (3), lung (4), liver (5), skeletal muscle (6), kidney (7), testis (8), stomach (9). D, syntenin-1 is detected at 36 kDa by the rabbit antibodies used in A and C. The very weak syntenin-1 bands in liver and skeletal muscle lanes are not visible on this copy. E, neurofascin, which is generated in several isoforms, was identified by an antibody raised against a cytoplasmic peptide of mouse neurofascin exclusively in brain. The identity of the polypeptide band of ϳ70 kDa that is recognized by the anti-neurofascin antiserum in all tissues with the exception of brain and spleen is unknown. exins II and III, was found to interact with syntenin-2␣ but not with -2␤ in a yeast two hybrid assay (Table II).
To define amino acid residues that determine the binding behavior of neurofascin and NrCAM with respect to syntenin-1, a number of substitution constructs of neurofascin and NrCAM were investigated. The combined results of the yeast two-hybrid assays, which are presented in Table I, are consistent with the conclusion that the binding to syntenin-1 is determined by the residues 0, Ϫ2, and Ϫ3 of neurofascin COOH terminus. Moreover, the failure of NrCAM to bind syntenin-1 is caused by its Ϫ3 asparagine, as indicated by the N(Ϫ3)Y substitution that enabled this mutant to interact with syntenin-1 equally as strong as neurofascin.
To demonstrate the specificity of the interaction between neurofascin and syntenin-1, we employed several other experimental systems besides the yeast two-hybrid one. In Fig. 3A the results of a ligand-overlay assay are shown. The blots of immunoprecipitated wild-type neurofascin, truncated GPI-anchored neurofascin construct, and NgCAM (the chicken homologue of mammalian L1) were incubated either with MBP-␤galactosidase or with MBP-syntenin-1 fusion proteins. After washing, the membranes were stained with anti-MBP antiserum. This assay revealed selective binding of MBP-syntenin-1 to wild-type neurofascin containing the cytoplasmic tail but not to NgCAM or to the truncated construct of neurofascin.
Furthermore, surface plasmon resonance analysis using a BIAcore system also confirmed the specificity of the neurofascin-syntenin-1 interaction (Fig. 3B). We observed a strong doseresponsive binding of MBP-syntenin-1 to an immobilized peptide corresponding to the 15 COOH-terminal amino acid residues of wild-type neurofascin. (Assuming an A ϩ B^AB interaction, the apparent dissociation rate constant was calculated to be ϳ1 ϫ 10 Ϫ4 s Ϫ1 . However, a further analysis of the obtained binding curves revealed that they are not consistent with such a simple model.) In contrast, no significant binding was detected using the A(0)S-substituted neurofascin peptide with MBP-syntenin-1, or the wild-type peptide with MBP-␤-galactosidase.
To demonstrate the association of syntenin-1 with neurofascin in mammalian cells, L929 cells were cotransfected with plasmids encoding syntenin-1 fused to EGFP and either wildtype neurofascin or the truncated GPI-anchored neurofascin construct (Fig. 3C). Colocalization of neurofascin clustered by antibodies and EGFP-syntenin-1 was observed in a subpopulation of cells expressing wild-type neurofascin but not in the control cultures. This further confirmed the specificity of the investigated interaction.
Neurofascin and Several Other Transmembrane Proteins (Neuroglian-180, Pro-TGF-␣, Syndecans, B-ephrins, EphA7, and Neurexins) Bind to the Second PDZ Domain of Syntenin-1-Syntenin-1 contains two PDZ domains and interacts with the COOH terminus of neurofascin and several other transmembrane proteins. This suggests that at least one of these domains is responsible for these interactions. To determine the individual binding specificity of the two syntenin-1 PDZ domains for different known interacting transmembrane proteins, we generated several syntenin-1 mutants and tested these in the two-hybrid assay. We observed that neither of the overlapping deletion constructs of syntenin-1, which were composed of the NH 2 -terminal third only or together with PDZ1, or of the isolated PDZ2 with the COOH-terminal stretch, did interact with the cytoplasmic tail of neurofascin (data not shown). This is consistent with the published data on the interaction of syntenin-1 with syndecan or pro-TGF-␣ (28,29). To address the binding specificity of the syntenin-1 PDZ domains further and to map binding sites within syntenin-1, we substituted the last glycine residue in the carboxylate-binding loop of each PDZ domain (see Fig. 1A) by glutamate (G128E in PDZ1) or aspartate (G212D in PDZ2). This glycine is the most conserved residue throughout all PDZ domains and allows the loop preceding the ␤2 strand to form a turn that is necessary for interaction with the ligand's carboxylate group (40). Three resulting point mutants (PDZ1*, PDZ2*, and PDZ1*2*) were then tested for binding to the cytoplasmic tails of several transmembrane proteins listed in Table II. The mutation in the first domain (PDZ1*) did not cause any significant reduction of the binding as observed in the two-hybrid assay. In contrast, the PDZ2* construct, as well as the double-mutant PDZ1*2*, failed to interact with any of the tested cytoplasmic segments. This indicates that the second PDZ domain is required to bind them. However, the deletion construct of syntenin-1 containing the PDZ tandem but lacking 101 NH 2 -terminal amino acids (N⌬101) also failed to bind neurofascin, neuroglian-180, and pro-TGF-␣, whereas the intensity of its interaction with syndecan-3, ephrin B-2, or EphA7 was reduced (Table II).
Taken together, we conclude that neurofascin and the other tested transmembrane proteins bind to the second syntenin-1 PDZ domain that is inactive if separated from other parts of the molecule. In addition, the specificity of this domain appears to be unusual, in that it is able to bind class I (neurofascin, neuroglian, and pro-TGF-␣) and class II (syndecans, class B ephrins, EphA7, and neurexins) COOH termini. Abolishment or reduction of binding caused by the deletion of the PDZ1 and/or the NH 2 -terminal segment of syntenin-1 might be the result of inappropriate folding of the obtained polypeptides. Alternatively, these domains might be indirectly involved in the interaction between syntenin-1 and the transmembrane proteins. One such conceivable mechanism might include the oligomerization of syntenin-1.
Homo-and Heterodimerization of Syntenin-1 and Syntenin-2-To further our understanding of the molecular functions of syntenin-1 and syntenin-2, we examined the ability of these proteins to self-and heteroassociate using the two-hybrid assay. Full-length syntenin-1 as well as syntenin-2␣ showed a The COOH-terminus of neurofascin is required to bind to syntenin-1 Binding of syntenin-1 to the two wild-type isoforms (ϮRSLE exon) as well as to the various COOH-terminal point mutants of the cytoplasmic segment of neurofascin (Nf) were analyzed by a yeast two-hybrid assay. The five most COOH-terminal amino acid residues are printed. Mutated residues are printed in boldface. The Nf-c14 construct contains only the last 14 COOH-terminal amino acid residues of neurofascin. In addition, neuroglian (Ngl-167 and Ngl-180 isoforms), L1, NrCAM (Nr), and two mutants of the latter were investigated. In the NrCAM mutant Nr-Nf the five COOH-terminal residues are replaced by those of neurofascin. The results of the ␤-galactosidase filter assays (the time it takes colonies to start turning blue) were scored as follows:  (Tables III and  IV). To gather additional evidence for the homotypic oligomerization of syntenin-1, we transiently cotransfected COS7 cells with plasmids encoding Myc-and FLAG-tagged syntenin-1. Specific coprecipitation of Myc-or FLAG-syntenin-1 from detergent lysates of double-transfected cells with either anti-FLAG or anti-Myc monoclonal antibodies, respectively, was readily observed (Fig. 4A). Full-length syntenin-1 without any epitope extensions also coprecipitated with FLAG-as well as with Myc-syntenin-1, suggesting that oligomerization might not be caused by the Myc or FLAG epitopes (data not shown). In addition, size exclusion chromatography of purified bacterially expressed MBP-syntenin-1 fusion protein revealed three peaks. One peak, estimated at a molecular mass of ϳ90 kDa, fits relatively well with the monomeric form (predicted molecular mass 77 kDa), whereas the other at ϳ163 kDa represents most likely the dimeric form of recombinant MBP-syntenin-1 (Fig. 4B). A third MBP-syntenin-1 peak was observed in the exclusion volume of the column and probably contains supermolecular aggregates. Taken together, these investigations indicate that syntenin-1 self-associates and forms homodimers.
To address the question of the structural basis of the selfand heteroassociation of syntenin-1 and -2, we again used the yeast two-hybrid system to test different combinations of wildtype, deletion, and point mutants of syntenin-1 as well as the two isoforms of syntenin-2. No interaction of full-length syntenin-1 with overlapping deletion constructs consisting of either the NH 2 -terminal domain alone or combined with PDZ1, of PDZ2 with the COOH-terminal stretch, or of the COOH-terminal stretch of 30 amino acid residues alone could be detected by the two-hybrid assay (data not shown). In contrast, full-length syntenin-1 bound to the syntenin-1 construct lacking its NH 2terminal third with only slightly reduced intensity, whereas this N⌬101 mutant was not able to self-associate (Table III). These observations indicate that the NH 2 -terminal third might be required but is not sufficient for the homotypic interaction of syntenin-1 and that it might bind to some part of the molecule other than itself. Similarly, syntenin-2␤, which is comparable to the N⌬101 construct of syntenin-1 (see Fig. 1), was found to self-associate significantly weaker than the long ␣-isoform (Table IV). We tested whether the point mutations within carboxylate-binding loops of PDZ domains can affect the self-association of syntenin-1. For this reason, PDZ1*, PDZ2*, and PDZ1*2* mutants were assayed for their ability to interact with each other, wild-type, or deletion constructs of syntenin-1.
Only results were considered that could be confirmed by a vice versa exchange of the two-hybrid vectors (BD and AD) in which a particular pair of constructs was expressed. The data illustrate an apparent implication of the PDZ domains in the selfassociation mechanism of syntenin-1, but details remain to be investigated in the future.
Taken together, our observations indicate that the individual domains alone are not sufficient and the overall integrity of syntenin-1 is required for the self-association of syntenin-1. The inhibition of the homodimerization of syntenin-1 might concomitantly affect its PDZ-mediated binding to neurofascin and to other tested transmembrane proteins.

DISCUSSION
In this study, we identified the PDZ domains containing molecule syntenin-1 as an intracellular neurofascin binding partner. The interaction of syntenin-1 with the cytoplasmic domain of neurofascin was observed in the yeast two-hybrid system and confirmed by overlay assay, by surface plasmon resonance measurements, and by the colocalization of both proteins in transfected cells. In addition to rat syntenin-1, we isolated a novel human syntenin-1-related molecule, syntenin-2, which can be expressed as a long isoform (syntenin-2␣) that has the same domain organization as syntenin-1 and as a short isoform (syntenin-2␤) that contains the PDZ-tandem but lacks mostly the NH 2 -terminal third. Syntenin-2␣ but not syntenin-2␤ was shown to interact with neurexins, but neither with neurofascin nor with several other transmembrane proteins. Syntenin-1 and syntenin-2 both were able to self-associate and to interact with each other in the two-hybrid system. The homodimerization of syntenin-1 was confirmed by coimmunoprecipiation experiments and gel filtration chromatography. Although the binding sites sufficient for the homodimerization of syntenin-1 are still unknown, we consider a homotypic binding mode, in which the PDZ tandem and at least a part of the NH 2 -terminal domain are essential. The binding of the syntenin-1 deletion mutant N⌬101 to wild-type syntenin-1 and its failure to interact with itself exclude the possibility that dimers are formed by association of the NH 2 termini of two syntenin-1 molecules. More likely, binding of the NH 2 terminus of syntenin-1 to an unknown site within the COOH-terminal two thirds of the molecule enables an antiparallel or "head to tail" association.
We demonstrated that other vertebrate members of the L1 subgroup of cell adhesion molecules bind neither syntenin-1 nor syntenin-2. Interestingly, syntenin-1 was also able to interact in yeast with the nervous system-specific isoform of the cytoplasmic tail of Drosophila L1-type protein neuroglian. This TABLE II Two-hybrid analysis of interactions between different membrane proteins and syntenin-1 or -2 A mutant of syntenin-1 lacking 101 NH 2 -terminal amino acids (N⌬101) and point mutants of the first and/or second PDZ domain (PDZ1*, G128E; PDZ2*, G212D) as well as syntenin-2␣ and -2␤ were tested for binding to the cytoplasmic segments of neurofascin and various other transmembrane proteins (their four most COOH-terminal amino acid residues are given). The results were scored as indicated in the legend of Table I finding raises the question whether there is a syntenin-1-like molecule in flies. Because neuroglian has been considered as the sole L1-type molecule in Drosophila (3), it might be able to substitute for several functions that are conferred by diverse L1-type molecules (neurofascin, NrCAM, L1, CHL1) in vertebrates, including binding to a putative Drosophila syntenin-1related protein. Although we failed to identify a syntenin-1-like protein in the complete Drosophila genomic data base, another PDZ protein might interact with the long neuroglian isoform in the insect cells.
Although the majority of known PDZ domains bind to specific COOH-terminal peptides of transmembrane molecules, several of them interact with internal sequences or with other PDZ domains. Here, using site-directed mutagenesis, we mapped the binding sites to the COOH terminus of neurofascin and to the PDZ2 domain of syntenin-1. Moreover, we found that PDZ2 is also responsible for the interaction with neuroglian-180 and several other transmembrane proteins that were previously reported to bind syntenin-1. One of them is pro-TGF-␣, which, together with neurofascin and neuroglian-180, contains a threonine or a serine at the COOH-terminal position Ϫ2 and belongs therefore to the class I PDZ-binding proteins (38). Another group of syntenin-1-interacting proteins tested here consists of syndecans, neurexins I-III, class B ephrins, and EphA7, all of which belong to the class II PDZ-binding proteins that contain an aromatic or hydrophobic residue at position Ϫ2.
On the basis of their sequences, the PDZ domains of syntenin-1 should interact with the class II-specific sequence motifs (28). Indeed, affinity of syntenin-1 to class II COOH termini appears to be higher than to class I COOH termini. Nevertheless, our data together with the observations of Fernandez-Larrea et al. (29) on pro-TGF-␣ strongly support that the syntenin-1 PDZ2 domain also binds to specific class I COOH termini. Screens of oriented peptide libraries and yeast two-hybrid studies have demonstrated that PDZ binding may require side-chain interactions in addition to those at the COOH-terminal amino acid positions 0 and Ϫ2 (38,41). Here, we showed that binding to syntenin-1 PDZ2 domain is also determined by the were immunoprecipitated from detergent extracts of transfected COS7 cells, followed by SDS-PAGE, and transfer to PVDF membranes. To compare the relative amounts of the immunoprecipitated proteins, one blot was stained using antibodies against NgCAM and neurofascin. Two other identical blots were incubated overnight either with MBP-syntenin-1 or with MBP-␤galactosidase. Binding of MBP fusion proteins was analyzed by anti-MBP antibodies. Only MBP-syntenin-1 was found to bind to wild-type neurofascin but not to NgCAM or to GPI-linked neurofascin. Bands corresponding to the heavy chains of antibodies used for immunoprecipitation are indicated by the arrow. In the GPI-anchored form of neurofascin, the transmembrane and cytoplasmic segments were replaced by one and a half FNIII-like domains followed by the GPI attachment signal of the F11 molecule. B, surface plasmon resonance measurements. MBP-␤-galactosidase or MBPsyntenin-1 were perfused over the BIAcore sensor chips, which contained immobilized synthetic peptides corresponding to the last 15 COOH-terminal amino acid residues of either wild-type or of the A(0)S-substituted neurofascin. Binding curves monitored at different protein concentrations are shown. Curves representing interaction of MBP-syntenin-1 with the mutant and of MBP-␤-galactosidase with the wild-type peptide are indicated by single and double asterisks, respectively. In two independent sets of experiments, only MBP-syntenin-1 was found to bind selectively and in a dose-responsive manner to the wild-type but not to the mutated COOH terminus of neurofascin. (RU, resonance units). C, colocalization of neurofascin and syntenin-1 in heterologous cells. EGFP-syntenin-1 was coexpressed in L929 cells either with wild-type neurofascin (upper panel) or with GPI-anchored neurofascin construct. Clustering of neurofascin was achieved by subsequent application of polyclonal anti-neurofascin and Cy5-conjugated secondary antibodies at 37°C in culture medium. Following fixation, the subcellular localization of neurofascin and EGFP-syntenin-1 was analyzed by confocal microscopy. Arrows indicate colocalization. (The scale bar represents 10 m.)

TABLE III
Self-association of syntenin-1 Syntenin-1 and different mutant polypeptides of syntenin-1 were analysed for self-association by a yeast two-hybrid assay. All possible combinations were tested. Only results are shown that were confirmed by a vice versa exchange of the GAL4 expression vectors (AD or BD) in which one particular pair of constructs was inserted. The results were scored as indicated in the legend of Table I.
tenin-1 PDZ2 domain. In the future, the issue of the binding specificity of the syntenin-1 PDZ1 domain might be addressed by screening for proteins interacting with it.
In accordance with other publications on syntenin-1, the two-hybrid construct composed only of the PDZ2 domain was found not to interact with neurofascin nor with other transmembrane proteins tested here. How might this inability of the isolated PDZ2 domain to bind be explained? The overall structural integrity of the syntenin-1 molecule might be required either directly, for its interactions with transmembrane proteins by means of multiple ligand binding sites, or indirectly, for the stabilization of the active conformation of the PDZ2 domain. In this context, Grootjans et al. (28) proposed recently a cooperative binding mode of specific class II peptides with both PDZ domains of syntenin-1. In our experiments, mutation in the carboxylate-binding loop of PDZ1, in opposite to that of PDZ2, hardly impaired the interactions of syntenin-1 with any of the tested transmembrane proteins. In contrast, interactions of class I COOH termini with syntenin-1 lacking its NH 2terminal domain (N⌬101) were abolished, whereas the binding of class II COOH termini to this mutant was retained. As it was shown that this deletion construct does not self-associate, dimerization of syntenin-1 might be considered as a prerequisite for interactions with class I but not with class II COOHterminal peptides. Although there is currently no direct evidence proving this binding model, future studies might reveal whether the homodimerization allosterically enhances the affinity of the PDZ2 to specific ligands of class II and even enables this domain for interactions with class I proteins such as neurofascin and pro-TGF-␣.
Because we are primarily concentrating here on the molecular aspects of syntenin interactions, the biological functions of these interactions, in particular between syntenin-1 and neurofascin in the developing nervous system, remain to be established. PDZ domains have been investigated in a number of so-called scaffolding proteins that are restricted to polarized subcellular sites where they cluster transmembrane and cytosolic components to multimolecular complexes (42). For example, in the nervous system, PDZ-containing proteins have been implicated in the targeting and clustering of pre-and postsynaptic proteins (43,44). In particular, the self-association of syntenin-1 and its heterodimerization with syntenin-2 might suggest a function to cluster neurofascin and other transmembrane proteins to subdomains of the plasma membrane of neural cells. However, syntenin-1 was not found to co-cluster with neurofascin at nodes of Ranvier in the optic nerve of adult mouse, 3 making a scaffolding function of syntenin-1 at least at this site less likely. On the other hand, homo-and heterodimerization of syntenin-1 may allow a large spectrum of transmembrane receptors belonging to different protein families to be assembled together with neurofascin.
Neurofascin has been shown to be implicated in axonal growth and fasciculation (10,11,35). These dynamic cellular processes require a mechanism to regulate the cell surface expression of neurofascin. One possible way to modulate the number of neurofascin molecules on the neural surface might be to internalize or to target it to specific cell surface domains. Insertion and removal of plasma membrane components are part of the growth cone machinery, and evidences have been accumulated in the past that vesicular transport and the subcellular targeting of proteins in neurons are involved in neurite extension (45). Because syntenin-1 appears to be necessary for correct targeting of pro-TGF-␣ to the surface of Chinese hamster ovary cells (29) and is colocalized with internalized trans-3 M. Koroll and F. G. Rathjen, unpublished observations. FIG. 4. Syntenin-1 forms homodimers. A, coimmunoprecipitation of FLAG-syntenin-1 with Myc-syntenin-1 and vice versa. FLAG-and Myc-tagged syntenin-1 were expressed either separately or together in COS7 cells by transient transfection. FLAG-syntenin-1 was immunoprecipitated with monoclonal antibody M5 against the FLAG epitope followed by Western blotting and detection of Myc-syntenin-1 by monoclonal antibody 9E10 against the Myc epitope, whereas Myc-tagged syntenin-1 was immunoprecipitated by monoclonal antibody 9E10 followed by Western blotting and detection of FLAG-syntenin-1 by monoclonal antibody M5. (IP, immunoprecipitation; WB, Western blot; P, precipitate; S, supernatant; H, heavy chain; L, light chain of antibodies; ST, Myc-or FLAG-syntenin-1.) B, size exclusion chromatography. Purified MBP-syntenin-1 fusion protein (30 g) was analyzed using a Superdex 200 gel filtration column that had been calibrated by the molecular mass standards indicated in the diagram by small squares (thyroglobulin (670 kDa), bovine gamma globulin (158 kDa), chicken ovalbumin (44 kDa), equine myoglobin (17 kDa)). Two observed peaks at 90 kDa and 163 kDa (larger squares) contained MBP-syntenin-1 and presumably represent its monomeric and dimeric forms. A third peak (not shown) was observed in the exclusion volume of the column.

TABLE IV
Self-association of syntenin-2 and its heterodimerisation with syntenin-1 Syntenin-2␣, -2␤, and -1 were tested for self-or heteroassociation by a yeast two-hybrid assay. The results were scored as indicated in the legend of Table I.   BD AD ␤-Galactosidase ST-2␤ ϩϩ ferrin in early apical recycling endosomes in Madin-Darby canine kidney cells (46), it might function by linking bound neurofascin or other transmembrane proteins to trafficking or recycling pathways also in neural cells. The identification of syntenin-1 as an intracellular binding partner of neurofascin may allow us to study the removal and insertion of neurofascin in the context of neurite extension during early neural development or its targeting to axonal initial segments and to the nodes of Ranvier in the differentiated nervous system. Further insights into the functions of syntenin-1 and -2 might also be obtained by identifying cytoplasmic proteins linking them to trafficking or signaling pathways.