The Binding of Syndapin SH3 Domain to Dynamin Proline-rich Domain Involves Short and Long Distance Elements*

Dynamin is a GTPase that mediates vesicle fission during synaptic vesicle endocytosis. Its long C-terminal proline-rich domain contains 13 PXXP motifs, which orchestrate its interactions with multiple proteins. The SH3 domains of syndapin and endophilin bind the PXXP motifs called Site 2 and 3 (Pro-786–Pro-793) at the N-terminal end of the proline-rich domain, whereas the amphiphysin SH3 binds Site 9 (Pro-833–Pro-836) toward the C-terminal end. In some proteins, SH3/peptide interactions also involve short distance elements, which are 5–15 amino acid extensions flanking the central PXXP motif for high affinity binding. Here we found two previously unrecognized elements in the central and the C-terminal end of the dynamin proline-rich domain that account for a significant increase in syndapin binding affinity compared with a previously reported Site 2 and Site 3 PXXP peptide alone. The first new element (Gly-807–Gly-811) is short distance element on the C-terminal side of Site 2 PXXP, which might contact a groove identified under the RT loop of the SH3 domain. The second element (Arg-838–Pro-844) is located about 50 amino acids downstream of Site 2. These two elements provide additional specificity to the syndapin SH3 domain outside of the well described polyproline-binding groove. Thus, the dynamin/syndapin interaction is mediated via a network of multiple contacts outside the core PXXP motif over a previously unrecognized extended region of the proline-rich domain. To our knowledge this is the first example among known SH3 interactions to involve spatially separated and extended long-range elements that combine to provide a higher affinity interaction.

Dynamin is a GTPase that mediates vesicle fission during synaptic vesicle endocytosis. Its long C-terminal proline-rich domain contains 13 PXXP motifs, which orchestrate its interactions with multiple proteins. The SH3 domains of syndapin and endophilin bind the PXXP motifs called Site 2 and 3 (Pro-786 -Pro-793) at the N-terminal end of the proline-rich domain, whereas the amphiphysin SH3 binds Site 9 (Pro-833-Pro-836) toward the C-terminal end. In some proteins, SH3/peptide interactions also involve short distance elements, which are 5-15 amino acid extensions flanking the central PXXP motif for high affinity binding. Here we found two previously unrecognized elements in the central and the C-terminal end of the dynamin proline-rich domain that account for a significant increase in syndapin binding affinity compared with a previously reported Site 2 and Site 3 PXXP peptide alone. The first new element (Gly-807-Gly-811) is short distance element on the C-terminal side of Site 2 PXXP, which might contact a groove identified under the RT loop of the SH3 domain. The second element (Arg-838 -Pro-844) is located about 50 amino acids downstream of Site 2. These two elements provide additional specificity to the syndapin SH3 domain outside of the well described polyproline-binding groove. Thus, the dynamin/syndapin interaction is mediated via a network of multiple contacts outside the core PXXP motif over a previously unrecognized extended region of the proline-rich domain. To our knowledge this is the first example among known SH3 interactions to involve spatially separated and extended long-range elements that combine to provide a higher affinity interaction.
Dynamin is a large GTPase that is a central component of a variety of endocytic processes, including synaptic vesicle endocytosis and activity-dependent bulk endocytosis pathways in nerve terminals (1,2). It is thought to be recruited to sites of endocytosis through interactions between its C-terminal proline-rich domain (PRD) 5 and the SH3 domains of a variety of proteins (3). This set of partner proteins includes several well known endocytic proteins as well as proteins that provide connections between dynamin and the actin cytoskeleton (4).
SH3 domains are small protein-protein interaction units consisting of a ␤-sandwich with five strands; the strands are connected by three loops and a 3 10 helix (5,6). SH3 domains predominantly bind proline-rich sequences containing a short and conserved PXXP motif (where X represents any amino acid), as well as a flanking basic residue(s) such as arginine or lysine within 2-4 amino acids on either side of the PXXP motif. We term this 6 -8-residue sequence the core PXXP motif. PXXP motifs most often adopt a left-handed type II polyproline (PPII) helix conformation and bind to a canonical hydrophobic pocket located between two short loops of the target SH3 domain, the so-called RT and the n-Src loops (5,6). The partner can interact with the SH3 protein in either of the two opposite orientations, depending on the location of the flanking basic amino acids. In the Class I binding mode, basic amino acids located N-terminal to the PXXP motif also contact the SH3, whereas in the Class II binding mode, basic amino acids located C-terminal to the PXXP motif contact the same part of the SH3 (7). These orientation-determining flanking basic residues form salt bridges and are an important specificity determinant for SH3/PXXP interactions.
Flanking residues also determine the binding specificity and sometimes substantially increase the binding affinity of both Class I and Class II proline-rich peptides to the SH3 domain. Ligands comprising a short core PXXP motif alone typically have affinities in the range 5-20 M, with little binding selectivity among SH3 families. However, nearby residues known as short distance elements (SDEs) can additionally bind to less conserved portions of the SH3 surface and contribute to higher (submicromolar) affinity interactions (8). It has also been observed that some peptides, whole proteins, or protein domains have an even higher affinity (low nanomolar) for SH3 domains. For example, the full-length dynamin I PRD (PRD-Ia, Pro-746 -Leu-864) binds amphiphysin and endophilin with a 10 nM dissociation constant (9). However, the sequence or structural features that confer this high affinity binding are not known.
Thirteen PXXP motifs exist in the dynamin I PRD (Sites 2, 3, and 9 are indicated in Fig. 1A) (10). Each is potentially able to bind to a single SH3 domain-containing protein. Syndapin, endophilin, and amphiphysin I are three of the many binding partners documented to play a direct role in synaptic vesicle endocytosis, and these proteins also contain Bin/amphiphysin/ RVS (BAR) domains that link dynamin to regions of membrane curvature at the neck of budding vesicles (10,11). Using GST pulldown experiments and site-directed mutagenesis, syndapin SH3 was shown to bind the Site 2 motif PXXP motif (Pro-786 -Pro-789), whereas endophilin SH3 recognized the combined Site 2 and Site 3 extended motif (Pro-786 -Pro-793) (10). In contrast, the amphiphysin SH3 binds Site 9 (Pro-833-Pro-836) via the Class II binding mode (11). It is also known that in nerve terminals the binding of syndapin SH3 domain to Site 2 in dynamin is inhibited by phosphorylation of the latter protein at Ser-774 and Ser-778 in the phosphobox sequence that is the N-terminal to Site 2 (Arg-772-Pro-781; Fig. 1A) (10,12). The interaction between syndapin-SH3 and dynamin-PRD is also reported to depend on an N-terminal extension of Site 2, which bridges Site 2 and the phosphobox (residues Gln-782-Ala-785; Fig. 1A) (10). A model was proposed whereby dynamin binds syndapin-SH3 using two adjacent sites within the dynamin PRD: the Site 2 PXXP motif and a pair of arginines at the N-terminal end of the phosphobox (Arg-772/Arg-773), which were originally proposed to act as an SDE.
Relatively few SH3-partner interactions have been demonstrated to utilize SDEs to date. Those that do include Src (PDB ID 1JEG) (6), p67phox (PDB ID 1K4U) (13), Lyn (PDB ID 1WA7) (14), ␤-PIX (PDB ID 1ZSG) (15), and insulin receptor tyrosine kinase substrate (PDB ID 2KXC) (16). In some of these proteins, the additional residues in the SDE make contacts in an acidic groove on a face of the SH3 domain that is not normally involved in PXXP-type interactions. This acidic pocket has been suggested to be termed the "specificity zone" of the SH3 domain, and SH3s that bear such pockets are typically characterized by unusually high affinity substrate interactions (6). Binding specificity and/or affinity can also be enhanced by additional contacts made between the SDE and the variable loops of the SH3 domain.
The SDE alone is generally insufficient to bind to the SH3 binding partner without the associated PXXP motif; however, mutations in the SDE can reduce or abolish binding of the SH3 binding partner, suggesting an important functional role for these elements. For example, the interaction of p46phox with the SH3 domain of p67phox involves a 20-amino acid helixturn-helix SDE in p46phox that extends C-terminal to the core PXXP; this SDE contacts multiple residues on the SH3 domain outside of the PXXP binding groove and greatly increases the affinity of the interaction from 20 M to 24 nM (13). In other situations SDEs have been observed to provide specificity. For example, the 183 PANLG 187 sequence located 11 amino acids away from the PXXP motif in the herpesvirus tyrosine kinase interacting protein confers specificity of binding to the SH3 domain of the Src family kinase Lyn (17). These residues form an additional helical turn adjacent to the PXXP motif, resulting in a 10-fold higher affinity for the Lyn SH3 domain than for other closely related members of the Src family.
In the present study we initially aimed to test the hypothesis that the N-terminal part of the dynamin phosphobox is a SDE that acts as a phosphorylation sensor that is able to inhibit syndapin recruitment. However, our data suggest that no direct contact exists between the phosphobox and the syndapin-SH3. Instead, two non-PXXP elements necessary for their interaction were found on the C-terminal side of the Site 2 PXXP motif of dynamin. To our knowledge this is the first report of two separate and extended long range sequence elements acting to provide higher affinity and specificity to an SH3 domain interacting with its target.
Pulldown Experiments-For synaptosome pulldown experiments, a total rat brain extract was prepared by homogenizing brain tissue in ice-cold lysis buffer (1% Triton X-100, 150 mM NaCl, 25 mM Tris, pH 7.4, 1 mM EDTA, 1 mM EGTA, 20 g/ml leupeptin, 1 mM phenylmethylsulfonyl fluoride, and EDTA-free Complete-Protease inhibitor mixture (Roche Applied Science)). The homogenate was centrifuged twice at 23,000 ϫ g for 30 min at 4°C, and the supernatant was collected. For pulldown experiments, glutathione-Sepharose beads bound with GSTdynamin-PRD or GST-syndapin-SH3 recombinant proteins were incubated with an equal volume of tissue lysate at 4°C for 1 h. Beads with bound proteins were washed extensively with ice-cold 20 mM Tris, pH 7.4, containing 1 mM EGTA, eluted in 2ϫ SDS sample buffer, resolved by SDS-PAGE on 7.5-15% gradient gels, and stained with Coomassie Brilliant Blue.
For dynamin C2 pulldown experiments, GST-tagged C2 was expressed in E. coli (JM109). The GST affinity tag was proteolytically removed with PreScission according to the manufacturer's instructions. C2 was further purified by gel filtration using a Superdex S75 column (GE Healthcare). Beads bound with GST-syndapin-SH3 were then incubated with the same amount of C2 peptide at 4°C for 1 h. Beads with bound proteins were washed extensively with ice-cold 20 mM Tris, pH 7.4, containing 1 mM EGTA, eluted in 2ϫ SDS sample buffer, resolved by SDS-PAGE on 7.5-15% gradient gels, and stained with Coomassie Brilliant Blue.
Antibodies and Western Blots-An anti-syndapin antibody was obtained from Santa Cruz Biotechnology (Santa Cruz, CA). Anti-amphiphysin monoclonal antibody was obtained from Pietro De Camilli (Yale, New Haven, CT). Protein samples were separated by SDS-PAGE on 10% or 12% acrylamide gels and transferred to nitrocellulose membrane as previously described (18). Western blots were analyzed by the enhanced chemiluminescence method using the SuperSignal West Pico Chemiluminescent Substrate (Pierce).
Multiangle Laser Light Scattering (MALLS)-Size-exclusion chromatography combined with multiangle laser light scattering was carried out using a Superdex 75 10/300 GL column (GE Healthcare) equilibrated in a buffer containing 20 mM HEPES, 100 mM NaCl, 1 mM DTT, 1 mM PMSF, pH 7.2. The column was run at a flow rate of 0.5 ml min Ϫ1 on an AKTAbasic liquid chromatography system (GE Healthcare) coupled to a miniDawn MALLS detector and an Optilab refractive index detector (Wyatt Technology). The MALLS technique provides a molecular mass estimate that is independent of shape (21). The molecular masses of bovine serum albumin (BSA, used as a control) and dynamin-PRD/syndapin-SH3 were calculated using a differential refractive index (dn/dc) value of 0.185.
Isotopic Labeling of Protein for NMR Spectroscopy-Production of 15 N-labeled or 13 C, 15 N-labeled dynamin PRD Ia and Ib and syndapin-SH3 was achieved by expression in shaker flasks with 15 NH 4 Cl and [ 13 C]glucose as sole nitrogen and carbon sources, respectively (22). The isotopically labeled proteins were purified in the same manner as the non-labeled proteins.
NMR Spectroscopy-For all NMR experiments, recombinant GST fusion proteins (of C2, PRD-Ia, PRD-Ib, and syndapin-SH3) were proteolytically cleaved to remove the GST tag and were purified by gel filtration. All NMR spectra were acquired at 298 K on a Bruker Advance III 600 spectrometer equipped with a triple-resonance cryogenically cooled probe and z axis pulsed-field gradients. 15 N HSQC and triple-resonance experiments were recorded using standard pulse sequences from the Bruker library. All spectra were processed with the TopSpin software (Bruker) and analyzed using SPARKY (T. D. Goddard and D. G. Kneller, SPARKY 3, University of California, San Francisco, CA). Chemical shift perturbations (CSPs, ⌬␦) in 15 N HSQC experiments were calculated as, Affinities for the syndapin⅐PRD and syndapin⅐C2 interactions were calculated from titrations of the unlabeled syndapin-SH3 domain into 15 N-labeled C2 or PRD. In the former case the changes in amide proton chemical shift of Arg-784, Ala-785, and Val-788 were separately plotted as a function of the concentration of syndapin-SH3, and the data were fitted to simple 1:1 Langmuir binding isotherms. The dissociation constant was taken as the average of the three derived values, and the error was estimated as 20%. In the latter case, however, the exchange between free and bound 15 N-labeled PRD is slower (the socalled intermediate exchange regime), and the peak positions are, therefore, not a simple population-weighted average of the chemical shifts of the free and bound forms. In this situation, the line shape at a given point in the titration is given by the imaginary component of the complex quantity G() (23), which is given by, where ϭ P F /k off , k off is the off rate for dissociation of the complex, and C is a scaling factor. P F and P B are the proportions of free and bound dynamin-PRD, which is a function of the dissociation constant K D and the concentrations of each protein. The quantities ␣ F and ␣ F are given by, and where T 2F and T 2B are the transverse relaxation times for the free and bound forms of the resonances in question. We, therefore, chose an intermediate point in the titration (0.4 or 0.8 mol eq of added syndapin-SH3) and extracted one-dimensional cross-sections of the two signals that underwent the largest chemical shift change (the amide nitrogen of Gly-801 and an unassigned Gln/Asn side-chain amide proton). The equations above were coded into Mathematica (Wolfram Research), and simulated line shapes were generated and compared with the experimental data by calculating the sum of the squares of the differences between the simulated and actual data ( 2 ). The values of K D , k off , C, T 2F and T 2B were varied in a grid search to find the best fit of the data to the model. The fits obtained were of good quality. Molecular Docking-The crystal structure of the SH3 domain of mouse syndapin I (24) (PDB ID 2X3W) was docked to dynamin I (residues 783-802) using HADDOCK (25)(26)(27). Initial docking attempts using full-length dynamin I PRD including the novel LDE element (Arg-838 -Gly-842) (Fig. 6) were unsuccessful, and we, therefore, limited our calculations to the shorter construct incorporating Sites 2 and 3 and the SDE (Fig.  1A). Residues 363-367 of the polyproline-region of p47phox (corresponding to dynamin residues 786 -790 of Sites 2 and 3) in the crystal structure with the SH3 domain of the p40phox protein (28) (PDB ID 1W70) were used as a structural template for the calculations based on the sequence identity and the high structural conservation of this site. During docking, all dynamin residues were defined as semi-flexible (side chains allowed to move freely), whereas full flexibility was restricted to the C-terminal part of dynamin (residues Ala-791-Pro-802) to maintain the overall conformation of Sites 2 and 3. For the SH3 domain, residues 390 -404, 415-422, and 431-439 were defined as semi-flexible based on the chemical shift perturbations and mutational data (Figs. 4 and 6). A total of 32 ambiguous interaction restraints were introduced between all atoms of dynamin (13 residues) and all atoms of 19 residues of syndapin (391, 393, 395, 397, 398, 400, 401, 402, 403, 416, 419, 420, 421, 432, 433, 435, 436, 437, 438) based on the NMR and mutational analysis. 1000 structures were calculated in the semi-flexible simulated annealing step, and the best 10 structures based on the HADDOCK score were subject to another annealing procedure (1000 structures) using the same restraints. Finally, the solutions were clustered using a cutoff of 0.5 Å root mean square deviation based on the pair-wise backbone root mean square deviation matrix. The best 20 structures (root mean square deviation of 0.7 Å) of this cluster were analyzed using standard HADDOCK protocols and were used to represent a model of the complex. HADDOCK models are available from the authors upon request.
Enzyme-linked Immunosorbent Assay (ELISA) Microtiter Plate Assay-ELISA microtiter plate assays were performed using 384-well plates (Pierce). All wells were coated with 320 ng of His-syndapin SH3 or full-length His-syndapin in 50 l of screening buffer (10 mM HEPES, pH 7.4, 50 mM NaCl, 1 mM DTT, 1 mM PMSF). After a 1-h incubation at room temperature, all wells were blocked with 50 l of blotting buffer (10 mM HEPES, pH 7.4, 50 mM NaCl, 1 mM DTT, 1 mM PMSF, 2% BSA, 2.5% milk powder), sealed with aluminum foil, and stored overnight at 4°C. Plates were then washed 3 times with chilled wash buffer (10 mM HEPES, pH 7.4, 50 mM NaCl, 0.05% Tween 20) to remove unbound protein. Between 1 ng and 640 ng of dynamin was applied to quadruplicate wells for each sample and incu-bated at 22°C for 1 h. For each sample an equivalent amount of screening buffer was added to adjacent wells in quadruplicate as a control for nonspecific binding. The plate was incubated for 1 h at 22°C and then washed 3 times with wash buffer. Anti-GST HRP antibody solution (50 l, Sigma) diluted 1:5000 with screening buffer was added to each well, and the mixture was incubated at 22°C for 15 min, after which time the excess antibody was removed by washing the wells 3 times with wash buffer. 50 l of 1-step Turbo TMB-ELISA (Pierce Biotechnology) was added, and the mixture was incubated for ϳ10 min then treated with 50 l of 1 N H 2 SO 4 to stop the reaction. ELISA plates were read at 450 nm on a Victor3 MultiLabel Plate reader (PerkinElmer Life Sciences). Data were analyzed using GraphPad Prism.
Structural Alignment of SH3 Domains-To generate a structure-based sequence alignment of SH3 domains and look for conservation in the RT-under-groove of SH3, the coordinates of the ϳ295 hits were downloaded from the Protein Data Bank and aligned in PyMOL using the 1EFN.pdb structure as the template SH3 domain model. For structures with multiple chains or multiple models, only the first SH3 domain was considered. Structures were examined, and obvious outliers were removed. The model of the syndapin SH3 domain with dynamin peptide was also included in the structure alignment. The syndapin SH3 domain sequence was uploaded into the software along with the sequences from all of the above PDB files. A structure-based alignment was conducted using 1NEB.pdb as the template model. In conjunction with this, a structure-based sequence alignment was conducted using STRAP program for statistical analysis.

Results
The Site 2 PXXP Motif in Dynamin I Is the Core Binding Site for Syndapin I-Previous work showed that the binding of syndapin to dynamin I required two components (the Site 2 PXXP and the N-terminal SDE that encompasses the phosphobox) in the region 772 RRSPTSSPTPQRRAPAVPPARPGSR 796 of dynamin (10) (note that throughout the paper, residue numbers between Arg-386 and Ile-441 indicate residues in syndapin I SH3 domain, whereas residue numbers between Asn-746 and Leu-864 refer to residues in the dynamin I PRD). Mutations of each underlined residue to alanine greatly or completely reduced the ability of this sequence to bind syndapin in GST pulldown experiments (10). To test the hypothesis that the proposed N-terminal SDE directly binds to the SH3 domain independently of the Site 2 core, a range of dynamin I PRD truncations and point mutants was generated, each containing different regions of the PRD (Fig. 1A). Dynamin has two major alternatively spliced tail variants, dynamin Ia and dynamin Ib (in this study a or b refers to the two main alternative C-terminal tails, also known as long tail a and short tail b, whereas additional splicing occurs in the middle domain, which is not discussed further herein). The dynamin Ia (dynIa) variant was primarily used in this study unless otherwise specified. The PRD was truncated from either the N-terminal (C1 and C2) or C-terminal ends (C6 and C10; C7 was truncated using dynamin Ib (dynIb) as template), and a double point mutation (S774E/ S778E; termed EE) was also introduced into the dynIa-PRD and C2 constructs as phospho-mimetics.
These constructs were expressed in E. coli as GST fusion proteins, immobilized on glutathione-Sepharose beads, and used as bait proteins to pull down wild type (WT) syndapin from rat brain synaptosome lysates. Western blot analysis revealed that GST-dynamin bound syndapin and that the expected decrease in binding to the phospho-mimetic form GST-dynIa-PRD-EE was observed (Fig. 1B). Among the truncated forms, only GST-C2 pulled down WT syndapin. The phospho-mimetic GST-C2EE showed markedly reduced binding to syndapin (Fig.  1B). Failure of the C1 construct, which contains the phosphobox but not the core PXXP motif, to bind syndapin shows that the proposed N-terminal SDE is not sufficient for autonomous syndapin binding. As expected, GST-C6, GST-C7, and GST-C10, which lack Site 2 and Site 3, were also incapable of binding syndapin. The pulldown samples were also probed for endogenous amphiphysin I, which binds to the Site 9 PXXP (11). Consistent with this specificity, GST-C6, GST-C7, and GST-C10, but not the other truncation constructs, pulled amphiphysin I out of the lysate (Fig. 1B, lower panel).
Mapping of the C2⅐Syndapin-SH3 Binding Interface-Because the phosphobox did not bind independently to syndapin, we used NMR CSP experiments to probe the dynamin/ syndapin interface. First, chemical shift assignments of backbone H, N, C␣, and C␤ atoms were made for 13 C, 15 N-labeled C2 (which comprises Ser-751-Pro-798 of dynamin) using standard approaches ( Fig. 2A). Assignments could be made for all expected signals, and analysis of the chemical shifts clearly demonstrated that C2 is primarily disordered in solution: this was confirmed by the program TALOSϩ (29). Uniformly 13 C, 15   , which share identical amino acid sequence up to Pro-844, are indicated. Thirteen PXXP motifs exist in the PRD and in vivo phosphosites Ser-774 and Ser-778 are located before the Site 2 PXXP motif. The phosphobox is defined from Arg-772 to Pro-781, the SDE is from Gly-797 to Gly-801, and the LDE is from Arg-836 to Gly-842 The lower half of this figure shows the dynamin PRD truncation mutants and the phospho-mimetic mutants used in this work. PH, PH domain; AD, assembly domain. B, the effect of various dyn-Ia-PRD fragments on syndapin binding in rat brain synaptosome pulldown experiments. Samples from pulldown experiments with the indicated GST-tagged PRD mutants were run on gels, blotted, and probed with antibodies to syndapin or amphiphysin. Ponceau staining was used as loading controls. Amphiphysin I was used as a positive control that is known to bind to the Site 9 PXXP motif. C1, C2, and C2EE lack Site 9 and were unable to bind amphiphysin I. Results are shown from one of four independent experiments. nant syndapin-SH3 (residues Arg-386 -Ile-441; Fig. 2B), and triple resonance experiments were used to reassign backbone atoms in C2 in the bound state. Significant CSPs were observed for a distinct subset of signals, confirming the existence of a direct interaction between syndapin-SH3 and C2. Observation of fast exchange on the chemical shift timescale is consistent with a relatively weak interaction. CSPs were plotted against residue number, and all significant CSPs were found to occur for residues in the region Arg-784 -Ala-791 (Fig. 2C). This shows that it is the Site 2 PXXP motif that drives binding to syndapin and that the phosphobox makes no direct contacts with syndapin-SH3 that are detectable by this method. This is not consistent with the hypothesis that the N-terminal phosphobox region is an SDE. Arg-784 and Ala-785 in the flanking region also displayed large chemical shift changes and, therefore, are likely to make a contribution to SH3 binding. This is consistent with their role as part of the core PXXP motif.
In the reciprocal NMR experiment backbone assignments were made for 13 C, 15 N-labeled syndapin-SH3 (all expected signals could be assigned, Fig. 3, A and B), and 15 N-labeled syndapin-SH3 was then titrated with unlabeled C2 (Fig. 3C). A clear subset of residues in the SH3 domain displayed significant CSPs (Fig. 3D). Those residues with changes that were larger than the average CSP plus 1 S.D. are shown in red on the x-ray structure of syndapin-SH3 (24) (PDB ID 2X3V; Fig. 3F) and are concentrated around a hydrophobic groove on one surface of the SH3 domain (referred to as top in this study) where the RT and the n-Src loops are located. This area corresponds to the canonical PXXP binding surface of SH3 domains. The complex of myosin SH3 domain bound to abcan125 (PDB ID 2DRK) is shown for comparison (Fig. 3E), as this SH3 domain has high sequence homology to syndapin SH3. Two CSPs, namely for Gly-395 and Gln-398 in the RT loop, were unexpected and are atypical judging from other structures. Gln-398, which is at the tip of the RT loop, underwent a significant CSP despite being relatively far from the expected PXXP binding groove (Fig. 3F). Together, these data do not support the hypothesis that the phosphobox is a SDE, which directly contacts the SH3.
To corroborate the NMR data, 21 residues in syndapin SH3 (including 9 non-alanine residues displaying a significant CSP) were mutated to alanine, and the ability of each mutant to interact with C2 was assessed using GST pulldown experiments. GST-tagged syndapin mutants were immobilized onto glutathione-Sepharose beads, and their ability to bind purified C2 was assessed by SDS-PAGE. In total, 14 of 21 SH3 mutations abolished detectable C2 peptide binding. The data show that five of nine non-alanine SH3 residues displaying significant CSPs (Tyr-393, Gly-395, Glu-397, Asp-416, and Gly-419) are also important for C2 peptide binding (Fig. 4A). It is notable that two classes of mutants dramatically reduced GST-SH3 binding to the C2 peptide: (i) point mutations in the hydrophobic pocket at positions conserved in virtually all SH3 domains (Tyr-393, Gly-395, Gly-419, Trp-420, Leu-432, Tyr-433, and Tyr-437) and (ii) point mutations of acidic residues thought to be important in coordinating the orientation of PXXP binding (Asp-394, Glu-397, Asp-399, Glu-400, Asp-416, and Glu-417). Those alanine mutants that abolished C2 binding (marked in magenta on the syndapin-SH3 crystal structure) clearly cluster around the core polyproline binding site pocket (Fig. 4B).
Full-length PRD Binds the SH3 with Higher Affinity Than the C2 Peptide-Because the experiments with C2 utilized less than half-of the dynamin PRD sequence, we next explored syndapin-SH3 binding by the full-length (FL) native dynamin protein.
The syndapin-SH3 mutants described above were used as bait in GST pulldown experiments from rat brain synaptosomes. In contrast to the C2 peptide, FL dynamin bound to almost all mutants, although to varying extents (Fig. 4C). Only 3 of 21 SH3 mutations abolished FL dynamin binding in this experiment (Y393A, W420A, and Y437A), whereas Y433A also produced a drastically reduced level of binding. In addition, the ability of the four mutants to fold correctly was confirmed by one-dimensional 1 H NMR spectroscopy. These residues lie in a small cluster around the RT loop and adjacent to the n-Src loop (Fig.  4B), which make up the core PXXP pocket, suggesting they constitute the focal point for dynamin binding on syndapin-

SH3. A similar situation is observed in most other known SH3
domains with PXXP interactions. Importantly, these data also suggest that the SH3 domain interacts with the full-length dynamin PRD with a higher affinity than with C2. These differences might arise from additional interaction sites in the PRD or FL dynamin or from additional components of the synaptosome lysate.
To assess whether there are elements of the PRD located outside the C2 region that might contribute to the binding of dynamin to syndapin-SH3, we used an ELISA to measure the affinity of syndapin-SH3 for recombinant C2 and dyn-Ia-PRD (Asn-746 -Leu-864). Use of the purified proteins eliminates the potential confounding influence of other proteins in tissue lysates. Either His-tagged mouse syndapin SH3 domain (Fig.   5A) or full-length syndapin (Fig. 5B) was coated onto a microtiter plate and incubated with various concentrations of GST-tagged dynamin-C1 or -C2 or dyn-Ia-PRD. Strikingly, the binding affinity of dyn-Ia-PRD was in the low nanomolar range for both syndapin SH3 domain (2.7 Ϯ 0.8 nM) and FL syndapin (9.2 Ϯ 1.9 nM). Affinities for the C2 peptide were 23-or 17-fold lower for syndapin-SH3 or FL syndapin, respectively. The GST-C1 construct that failed to bind syndapin-SH3 in previous pulldown experiments (Fig. 1B) again displayed no interaction with either syndapin construct. Therefore, although the Site 2 region of PRD contains the sequence elements necessary to bind the SH3, additional elements must exist within the full PRD that account for a significant affinity increase.   We also used NMR spectroscopy to measure K D values for the interaction between syndapin-SH3 and the C2 and PRD polypeptides. In the former case the chemical shift of three selected signals in a 15 N HSQC of 15 N-labeled C2 were plotted as a function of added syndapin-SH3 concentration and fitted to a standard 1:1 binding isotherm. In the latter case two signals were chosen in a 15 N HSQC of 15 N-labeled PRD and line-shape analysis used to extract K D values from a titration with syndapin-SH3 (see "Experimental Procedures"). For each titration the values obtained were consistent from signal to signal, and we obtained K D values of 20 Ϯ 4 M for the C2/SH3 interaction and 5 Ϯ 1 M for the PRD/SH3 interaction. These affinities are significantly weaker than those measured by ELISA (here) or surface plasmon resonance (9), two techniques where binding occurs to a solid-phase-attached protein rather than wholly in solution. It is also possible that the affinities observed by ELISA reflect the use of GST-syndapin polypeptides, which possess the capacity to form GST-mediated dimers, although this is not always the case for all GST-tagged proteins (30). It is worth noting that dynamin forms functional multimers at synapses, and it is, therefore, possible that the "true" affinity in vivo is substantially higher than that measured by our NMR titration method. Importantly, both sets of data are consistent with our central conclusion that elements outside the PXXP motif are important for the dynamin/syndapin interaction.
Two Extended Elements in the PRD Bind Syndapin-Given that dyn-Ia-PRD binds syndapin-SH3 with higher affinity than the truncated C2 peptide, we asked whether there might be additional interactions between dynamin and syndapin not contained within the bounds of the C2-SH3 complex. One possible scenario is that the full-length PRD is able to bind multiple SH3 domains by virtue of its 13 PXXP motifs. To assess this possibility, we expressed and purified dyn-Ia-PRD and assessed the stoichiometry of its interaction with syndapin-SH3 using size-exclusion chromatography coupled MALLS. Both dyn-Ia-PRD (M r theor ϭ 11.9) and syndapin-SH3 (M r theor ϭ 6.5) ran predominantly as single species, and the molecular weights determined from MALLS data were consistent with both proteins existing as monomers in solution (M r obs ϭ 13.0 Ϯ 0.2 and  . Site-directed mutagenesis verifies residues in the syndapin-SH3 domain involved in dynamin binding. A, GST-syndapin-SH3 alanine mutants were tested for their ability to bind to the GST tag cleaved C2 peptide. Excess GST-SH3 was used, but note that the C2 peptide did not stain well with Coomassie Blue. The experiment was repeated three times, and a typical result is shown. B, the alanine mutants that eliminate C2 binding are marked in pink, and the three aromatic residues identified in Fig. 3C are shown in red. C, GST-tagged syndapin-SH3 alanine mutants were used to pull down fulllength rat dynamin from rat synaptosome lysates and subjected to SDS-PAGE and Coomassie staining. The experiment was repeated three times, and each experiment included two replicates. A typical result is shown. 6. 7 Ϯ 0., respectively) (Fig. 5, C and D). Injection of a mixture of the two proteins at a molar ratio of 1:3 (dyn-Ia-PRD:syndapin-SH3) yielded an early eluting peak with a MALLS-derived molecular weight of 18.8 Ϯ 0.6, consistent with a 1:1 complex and a late-eluting peak corresponding to unbound syndapin-SH3 (Fig. 5D). The enhanced binding of dyn-Ia-PRD to syndapin is, therefore, unlikely to be a consequence of multiple syndapin SH3 domains binding to the PRD simultaneously. Consistent with this conclusion, a titration of 15 N-labeled dynamin-C6 (which contains the Site 5 through to Site 13 PXXP motifs) was carried out. No significant CSPs were observed (data not shown). We next used triple resonance spectra to make backbone assignments of 13 C, 15 N-labeled dyn-Ia-PRD (Asn-746 -Leu-864; 93% of expected resonances assigned; Fig. 6A). Similar to the situation for C2, inspection of the chemical shift distribution for dyn-Ia-PRD alone indicated that this polypeptide does not adopt a regular secondary structure in solution. However, in contrast to C2, a large number of the signals in the PRD-a 15 N HSQC spectrum displayed substantial line broadening, suggesting that this protein interconverts between two or more conformations on a s-ms timescale.
Titration of 15 N-dyn-Ia-PRD with unlabeled syndapin-SH3 induced numerous CSPs and an overall increase in signal intensity and a narrowing of line-widths (Fig. 6, B and C). In line with the size-exclusion chromatography coupled with MALLS data, no further changes were observed in the spectrum after the addition of 1.2 mol eq of syndapin-SH3. As expected, large CSPs were observed for residues in the Site 2 PXXP motif, and the pattern of changes closely matched those observed above for C2 (Fig. 6D). However, a number of additional residues exhibited CSPs of a similar magnitude, and these were clustered in two additional regions of the sequence: Gly-797-Gly-801 and Arg-838 -Gly-842. Gly-797-Gly-801 is nine amino acids C-terminal to the Site 2 PXXP motif, and we, therefore, label it a SDE, whereas Arg-838 -Gly-842 is Ͼ50 residues away, and we designate it a long distance element (LDE, Fig. 6D). Essentially identical CSPs were observed using a dyn-Ib-PRD construct corresponding to the short splice isoform, which terminates at Pro-851 (Figs. 1A and 6, E and F). These observations suggest that in addition to the core PXXP motif at least two additional non-PXXP elements are involved in the SH3 interaction.
To corroborate these observations and to determine if the additional elements are required for SH3 domain binding, we prepared a panel of point mutants of dyn-Ia-PRD as GST fusion proteins and used them to pull down syndapin and amphiphysin from rat synaptosome lysates. Mutations across the SDE at G797A and P798A abolished the dyn-Ia-PRD/syndapin interaction (Fig. 7). However, mutations of the next four C-terminal residues did not impair binding to either syndapin or amphiphysin (Fig. 7), suggesting that the SDE is encompassed by the dynamin PRD sequence Gly-797-Gly-801 (Fig. 1A). Similarly, mutation of Pro-844, but not R846 or P852, substantially reduced the interaction with syndapin. Mutation of Arg-838 caused a marginal reduction in syndapin binding; the mutation R838A also disrupted the dynamin/amphiphysin interaction, as expected (Arg-838 lies within Site 9, the PXXP motif that binds amphiphysin). Interestingly, the P798A mutation also signifi-cantly reduced the dynamin/amphiphysin interaction, suggesting that amphiphysin can also make contacts with this element, which is 35 amino acids N-terminal to the Site 9 PXXP binding site for the amphiphysin SH3 domain. Overall, these data are consistent with the idea that both an SDE and an LDE in dynamin PRD are involved in binding the syndapin SH3 domain. In addition, it appears that the SDE for syndapin binding might act in a reciprocal way as an LDE for dynamin in its interaction with amphiphysin.
To assess how these additional regions of dynamin might contact syndapin-SH3, 5 N-labeled syndapin-SH3 was titrated with dyn-Ia-PRD (Fig. 8, A and B). Most of the CSPs were the same as observed earlier with the C2 (Fig. 3, B and C), with four notable differences. Comparison of the CSPs with those observed for the syndapin-SH3/C2 interaction indicates that Glu-397, Leu-401, Ser-402, and Arg-424 exhibit substantially larger chemical shift changes after titration with dynamin PRD-a. Of these residues, Leu-401, Ser-402, and Arg-424 are located on a different surface to the canonical PXXP binding groove (Fig. 8D). All four residues define a groove "under" the RT loop and distinct from the classical the PXXP binding area, which we call the RT-under-groove.
To understand how these additional residues in the SH3 might contact dynamin we calculated a structural model using HADDOCK (25)(26)(27) (Fig. 9, A and B). Restraints were based on our NMR and mutational data, and the structure of the SH3 domain of p67phox bound to p47phox (28) (PDB 1K4U) was used as a template to pre-orient the Site 2 PXXP motif (for details see "Experimental Procedures"). The residues facing into the PXXP binding groove are highly conserved between syndapin and p67phox SH3 domains. Initial attempts using full-length dyn-Ia-PRD did not result in convergence. We, therefore, used a shorter version that incorporates both Sites 2 and 3 as well as the SDE element (residues Arg-783-Pro-802) in our calculations. Fig. 9A shows the 20 lowest energy models of the complex. As expected, the N terminus of PRD sits in the conserved hydrophobic groove made up by Tyr-391, Trp-420, Pro-434, and Tyr-437 and makes contacts via the two prolines (Pro-786 and Pro-789). However, the newly identified SDE element almost completely encircles the RT loop of the SH3 domain and contacts the SH3 in the same RT-under-groove (encompassing residues Gln-398, Glu-400, Leu-401, and Ser-402) as identified above by NMR CSP experiments and sitedirected mutagenesis.

Discussion
The Dynamin PRD Utilizes Both Short and Long Range Sequence Elements to Recognize Syndapin-The molecular basis for recognition of PXXP motifs by SH3 domains is well known (31)(32)(33), but our study extends this understanding in the case of the interaction between dynamin and syndapin, two proteins that are essential for normal synaptic vesicle endocytosis in mammals.
A combination of NMR titration and mutagenesis data confirms that the primary syndapin binding site in the dynamin PRD is the Site 2 PXXP motif, but that, unexpectedly, two additional sequence elements contribute to syndapin recognition. The first of these (the SDE) is immediately C-terminal to the  Site 2 ϩ 3 sequence, and the second (the LDE) is ϳ50 residues farther toward the C terminus. Neither the SDE nor LDE site is sufficient to independently bind syndapin, but they complement the PXXP motif, increasing the binding affinity. It is notable that high affinities have been observed for interactions between dynamin and both endophilin and amphiphysin (9), suggesting perhaps a requirement for strong interactions during synaptic vesicle formation. Furthermore, both dyn-Ia-PRD and C2 bind syndapin SH3 ϳ3-fold tighter than they bind full-length syndapin. This observation is consistent with the hypothesis that the SH3 domain of syndapin might be at least partly autoinhibited within the context of the full-length protein. In fact, in the absence of PRD, the SH3 domain has previously been proposed to bind to its own BAR (Bin/amphiphysin/ RVS) domain (24).
We used our NMR CSP and mutagenesis data to build a partial model of the dynamin-syndapin complex. The residues lining the PXXP binding groove in syndapin are highly conserved in the NADPH oxidase component p67phox, and in addition, the PXXP motif of dynamin Site 2 (PAVP) is identical to that of p47phox, the partner of p67phox. We, therefore, constrained this portion of our model using the p67phox-p47phox crystal structure. In this arrangement, the Pro-786 side chain (the first Pro in the PAVP) is sandwiched between the aromatic side chains of Tyr-391 and Tyr-437, whereas Val-788 directly contacts Tyr-393, and Pro-789 contacts Trp-420. The side chain of Arg-792 lies adjacent to a cluster of acidic side chains on the RT loop, Glu-397, Asp-399, and Glu-400, making a small electrostatic network. Mutations of essentially all of these aromatic and acidic residues reduce the apparent affinity of the interaction with C2 (Fig. 4A), consistent with our model.
Our data suggest that the dynamin PRD SDE binds a previously unrecognized groove on the surface of syndapin SH3, which we have called the RT-under-groove; this groove is centered on Leu-401, a residue that is also a part of the hydrophobic core. Leu-401 is conserved (as Leu, Ile, or Val) in 93% of the nearly 200 SH3 domains with known structures (supplemental Table S1 and S2). Furthermore, of the other residues whose side chains contact the PRD in our model (Gly-395, Glu-397-Ser-402, G431 and Leu-432), almost all show significant conservation despite not being important for recognition of the core PXXP motif (supplemental Table S2). Thus, Gly-395 is Gly or Ala in ϳ50% of SH3 structures, Gln-398 is Gln or acidic in ϳ60% of structures, Asp-399 is acidic in 40% of structures, Glu-400 is acidic in 85% of structures, Leu-401 is Leu/Val in 85% of structures, Ser-402 is Ser/Thr in 60% of structures, Gly-431 is Gly in 85% of structures, and Leu-432 is hydrophobic in 90% of structures. This level of conservation is consistent with the idea that the RT-under-groove could be relevant for interactions of other SH3 domains with SDEs in their target proteins.
Although our HADDOCK modeling shows the SDE binding this groove, the CSP changes occurred in the context of the dyn-Ia-PRD polypeptide, and it is possible that the LDE contacts either this groove or an additional, unidentified region of the SH3 domain. This hypothesis awaits further investigation. Efforts to measure intermolecular NOEs to resolve this issue were unsuccessful, most likely due to the intermediate exchange regime and consequent line broadening observed for complex formation. Overall, our data are consistent with the idea that one of the two remote elements in dyn-Ia-PRD contacts this novel RT-under-groove on the syndapin SH3 surface.
The binding mode predicted for dynamin from our data is significantly different to SH3-binding proteins previously reported to utilize SDEs (13)(14)(15)(16)(17)34). Fig. 9, C and D, shows the SH3 domain of p67phox (PDB ID 1K4U) bound to p47phox. Here, the additional residues of the SDE element make contacts on a distinct face of the SH3 domain, the specificity zone (Fig.  9D). Our data do not support that the dynamin PRD makes any contacts in this region of syndapin but passes over it. In future studies it will be important to determine whether SDEs and LDEs in other PRD-containing proteins can be discovered by using more extensive PRDs or domains rather than using shorter peptide motifs.
How Does the PRD Act as a Phospho-sensor for Syndapin SH3?-In nerve terminals, the dynamin PRD is the major site for both phosphorylation and protein binding at the synapse (35)(36)(37). The dynamin phosphorylation cycle plays a key functional role in bulk endocytosis (37); for example, dephosphorylation of dynamin promotes recruitment of syndapin, which in turn triggers activity-dependent bulk endocytosis of synaptic vesicles (1). We have shown previously that binding of syndapin SH3 domain to Site 2 in dynamin is blocked by phosphorylation at Ser-774 and Ser-778 in the adjacent N-terminally positioned phosphobox sequence (10,12). Consistent with these findings, our pulldown assays show that the phospho-mimic double mutant S774E/S778E displays significantly reduced binding to syndapin in the context of both C2 and dyn-Ia-PRD polypeptides. Our original hypothesis was that dynamin binds syndapin via two contacting sites: the Site 2 motif and the phosphobox (12). However, the NMR data in our current study failed to provide evidence that the phosphosites directly interact with syndapin SH3, ruling out the existence of an SDE N-terminal oriented to the core PXXP. A surprising aspect of our study is how a broad range of mutations across large stretches of the PRD can reduce or abolish syndapin binding, particularly the mutations in the phosphobox region. It is difficult to reconcile these observations with previous reports where a phosphobox synthetic peptide bearing phospho-deficient mutations (Dyn 769 -784 AA with Ala replacing Ser; this peptide completely lacks the Site 2 PXXP motif) reduced the dynamin/syndapin binding in vitro and also reduced bulk endocytosis in nerve terminals (12,37). A potential explanation for the discrepancy is that the synthetic peptide was not from native dynamin but had two point mutations and was additionally tagged with a highly basic penetratin heptapeptide sequence to facilitate intracellular delivery (a total of 10 amino acid changes). These changes may have resulted in an artifactual, yet serendipitously useful construct that does not accurately reflect the behavior of the native protein.
Binding amphiphysin I to the dynamin PRD (which is focused around Site 9 in the PRD) may also involve an LDE. Consistent with previous reports (11,12), mutations at sites close to and within Site 9 (P836A and P840A) abolish or greatly reduce amphiphysin binding, as expected. Surprisingly, the P798A mutation also reduced amphiphysin binding to the PRD despite its location ϳ40 residues from Site 9 (Fig. 7). Similarly, it has been proposed that basic amino acids near Sites 2 and 3 in the PRD could also regulate amphiphysin binding despite being located Ͼ50 amino acids away from its binding site (10).
Are SDEs and LDEs Involved in Other SH3 Interactions?-Finally, how common might be an involvement of these novel elements in interactions between PRDs with other SH3 domain proteins? Most published biochemical and structural data have focused primarily on very short proline-rich sequences rather than the whole protein or protein domain that contacts an SH3