Structural basis of filopodia formation induced by the IRSp53/MIM homology domain of human IRSp53

Overall structure and topology

The crystal structure of the IMD of human IRSp53 (residues 1–250) has been determined de novo by multiwavelength anomalous dispersion and refined to a resolution of 2.2 Å (Table I). The experimental density map (Figure 2A) delineates the entire polypeptide with the exception of two N-terminal and two C-terminal residues, the former arising from the expression construct. The asymmetric unit of the monoclinic unit cell comprises two IMD monomers (Figure 2C) forming a zeppelin-shaped dimer with a very extensive dimer interface (Figures 2B and 3A). The dimer is about 25 Å in diameter and measures 180 Å in length, with the dyad relating the two monomers roughly parallel to the crystallographic b-axis. The IMD monomer folds as a coiled coil of three extended α-helices and a shorter C-terminal helix that is separated by an 8-residue linker from the preceding helix 3. Helix 4 packs tightly against the other three helices, and thus represents an integral part of the domain. The fold of the IMD monomer closely resembles that of the BAR (Bin/amphiphysin/Rvs) domain of arfaptin2 and amphiphysin (Tarricone et al, 2001; Peter et al, 2004), but the dimers show markedly different shapes (Figure 2D). The C-terminus of helix 4 is adjacent to the N-terminus of helix 1 of the opposite monomer, with the domain termini falling onto the boundary of the central region, an area approximately defined by the extent of the dimer interface. Helices 2 and 3 protrude beyond the central region, which is made up of a 6-helix bundle, and the poorly ordered 2–3 loops demarcate the extreme ends of the dimer.

The dimer interface

The dimer interface is very extensive burrying about 5800 Ų of solvent-accessible surface, with a contact surface area of ∼1350 Ų per monomer (Figure 3A). By comparison, the crescent-shaped dimer of the BAR domain of amphiphysin buries about 3500 Ų of solvent-accessible surface, with ∼620 Ų contact surface area per monomer (Figure 3B). Hydrophobic and polar surfaces contribute to nearly equal parts to the contact surface area, which involves all four helices (Figure 1B). The interface encloses a substantial cavity of ∼930 ų that contains several water molecules. Cysteines 195 and 230 of opposite chains are juxtaposed such that disulphide bond formation across the dimer interface is geometrically possible, but suppressed due to the presence of 5 mM dithiothreitol (DTT) in the acidic crystallisation buffer (pH 4.6). Disulphide bond formation is indeed observed on a denaturing SDS–acrylamide gel under nonreducing conditions, but not when Cys230 is mutated to alanine (Figure 4A). This provides independent evidence that the dimer interface revealed by the crystal structure is likely physiologically relevant. In order to verify the oligomerisation state of the IMD in solution, we performed sedimentation velocity experiments in the presence of DTT (mirroring the crystallisation conditions), which indicated a single molecular species at a sedimentation coefficient of 3.25s (Figure 4B). Fitting the data to a single species model yields a molecular weight of 56 792 Da (95% confidence interval: 51 517–60 642), representing twice the calculated molecular weight of the IMD monomer, 28 563 Da. No other species can be detected.

In agreement with previous data (Yamagishi et al, 2004), we also observed self-association of full-length IRSp53 in COS7 cell lysates as myc-tagged IRSp53 co-immunoprecipitated with HA-tagged IRSp53 (data not shown). Thus, the IMD forms a dimer in solution and is likely responsible for the observed dimerisation of the full-length protein in vivo.

Structural differences between monomers

The present model of the IMD dimer has been constructed without applying noncrystallographic symmetry restraints throughout phasing, model building and refinement, allowing to detect potential conformational differences between the monomers. The two subunits superimpose very well within the core region (residues 1–130 and 173–248, root mean square deviation (r.m.s.d.) 0.57 Å for 206 Cα positions) that is defined by the extent of the dimer interface. In contrast, when all 248 Cα positions of the present model are included to calculate the superimposition, the r.m.s.d. increases almost three-fold to 1.5 Å, reflecting a significant level of conformational freedom for those parts of helices 2 and 3 that extend beyond the core region. Accordingly, drastic differences between the monomers in terms of side chain or backbone conformation fall outside the core region, with one notable exception: the side chain of Tyr115, which is located adjacent to a kink in helix 2. In chain A, the Tyr115 hydroxyl is observed to form a tight hydrogen bond with Glu189 in helix 3. In chain B, this side chain is rotated by almost 180° away from Glu189 and interacts with a symmetry-related molecule. The side chain rotation opens a sizeable crevice in the surface of the IMD that exposes the guanido group of Arg193 to solvent. We note that both Glu189 and Arg193, albeit not Tyr115, are invariant residues belonging to the IMD signature motif (see below). This fact may suggest that the observed conformational difference might be of functional significance, a question that is not further addressed in this study.

Sequence conservation

The surface area of the IMD dimer displays a marked polarity in terms of sequence conservation with respect to the five IMD sequences in Figure 1B. When the dyad axis of the dimer is oriented perpendicular to the plane of the paper, the N- and C-termini of the IMD dimer are displayed on the front and back faces of the dimer, respectively (Figure 3C, upper and lower panel). The N-terminal face shows extended patches of invariant solvent-exposed residues, while the C-terminal face is almost completely devoid of invariant or highly conserved residues. The two extended, dyad-related patches of invariant residues seen on the N-terminal side include Met21, Pro26, Glu126, Lys136, Lys143 and Lys147, all of which are part of conserved segments in the sequence of the IMD (Figure 1B). These patches surround two distinct voids in the molecular surface of the IMD dimer, which are bounded by the N-terminus on the core side of the dimer, and by the cluster of conserved basic residues (Lys142–Lys147) on the other side. Interestingly, the signature sequence of the IMD (residues 186–197, ALXEE[RK][RG]RFCX(0,1)F[IL]; Yamagishi et al, 2004) is mostly buried in the dimer. Residues in its core segment ¹⁸⁹EERRR¹⁹³ make contacts with other conserved parts of the opposite monomer (Figure 2A and B), notably with the loop linking helices 1 and 2, and the 8-residue linker separating helices 3 and 4. For instance, Glu190 forms strong hydrogen bonds with the backbone amides of Gln67 and Gly68 in the 1–2 loop (Figure 2A); Arg191 forms salt bridges to residues Glu71 and the carbonyl oxygen of Asp232; Arg193 makes van der Waals contacts with Phe24 and Ile20 of the same monomer, ‘pushing' into a kink of helix 1.

The polarity in terms of sequence conservation is mirrored, albeit to a lesser degree, by a noticeable polarity of the surface charge distribution. The extreme ends of the dimer present a strong positive surface potential on the N-terminal side, but do so to a much lesser degree on the C-terminal side. Elsewhere on the surface, the charge distribution appears more or less even, with positive charges slightly prevailing overall (Figure 3D).

Identification of residues involved in actin binding

At the extreme ends of the dimer are a cluster of basic surface residues that are well conserved between the five known mammalian IMDs, suggesting that these were likely to be significant to IMD function. Electron microscopy of IMD-induced F-actin bundles reveals that the centres of adjacent F-actin fibres are spaced about 11 nm apart (Yamagishi et al, 2004) and this distance correlates with the distance between the conserved basic patches. Moreover, the surface charge of F-actin fibres is highly acidic, suggesting that basic surface areas of the IMD dimer are likely to be involved in F-actin binding. We therefore targeted the extreme ends of the dimer in a mutational analysis. Mutant constructs were first made based on plasmids encoding full-length IRSp53 and assayed in vivo. Mutations that showed compromised filopodia formation were then followed up by making the same mutations in plasmid constructs encoding the IMD alone. We found initially that mutant forms of IRSp53, where only a single site was altered in the Lys142–Lys147 cluster, were indistinguishable from wild-type IRSp53 in the cell-based assay (data not shown). In contrast, altering all four lysines (142, 143, 146 and 147) to glutamic acid dramatically affected the behaviour of both full-length IRSp53 and the IMD.

Wild-type (IMDwt) and mutant (IMDmut, K142E, K143E, K146E, K147E) forms of the IMD of IRSp53 were expressed as glutathione-S-transferase (GST) fusion proteins in Escherichia coli and purified to homogeneity in their tag-free form. IMDwt and IMDmut were indistinguishable in circular dichroism spectroscopy experiments (Figure 4C). While the sedimentation velocity experiments indicated a slight tendency of the mutant to aggregate (peak at 4.5s, Figure 4B), the predominant species in solution is still the dimer peak and, crucially, its sedimenation coefficient is unchanged from that of wild-type IMD, indicating clearly that the global conformation of the dimer is unchanged by the site-directed mutagenesis. Thus, the mutation did not affect the association state or the fold of the IMD. F-actin binding by IMDwt and IMDmut was analysed using a filament cosedimentation assay varying the concentration of IMD. The pellet fractions were analysed by SDS–polyacrylamide gel electrophoresis and the gels evaluated densitometrically (Figure 4D). The equilibrium dissociation constant K_d derived from these binding curves is 5 and 10 μM for IMDwt and IMDmut, respectively. While the nature of this assay limits the accuracy of the K_d values obtained, it unambiguously demonstrates that the weakened binding affinity of IMDmut for F-actin.

Actin bundling was assayed in vitro using a low-speed cosedimentation assay. In the absence of IMD, 25% of F-actin was typically found in the pellet. Addition of IMDwt resulted in a concentration-dependent increase of the proportion of F-actin in the pellet, demonstrating filament bundling (Figure 4E, left panels), whereas addition of IMDmut had no effect on the partitioning of F-actin between pellet and supernatant (Figure 4E, right panels).

In an independent assay, F-actin coupled to a fluorescent dye (Machesky and Hall, 1997) was incubated with IMDwt and IMDmut at various concentrations and then visualised by fluorescence light microscopy. The presence of IMDwt resulted in the formation of visible actin bundles (Figure 4F). Single bundles as well as branched networks of bundles were observed. The number and size of bundles increased with IMDwt concentration as did the extent of branching. No bundles were observed at any concentration of IMDmut tested. These data demonstrate that although IMDmut is correctly folded and present as a dimer (Figure 4B and C), it cannot bundle actin fibres due to the disruption of the actin-binding sites.

Mutation of basic patches of IMD abrogates filopodium formation by IRSp53

Several groups have observed that overexpression of IRSp53 in a range of cell lines results in the formation of filopodia. Overexpressing myc-tagged IRSp53 in COS7 cells, we observed long filopodia that were often branched (Figure 5A). While filopodia were observed in untransfected COS7 cells, these were usually short (<10 μm) and clearly distinguishable from those induced by IRSp53 overexpression, which were typically 10–60 μm in length. Myc-tagged IRSp53 was localised throughout the filopodia and often concentrated at the tip. As described previously (Yamagishi et al, 2004), we observed that expression of the isolated IMD of IRSp53 also resulted in the formation of filopodia (Figure 5B), while deletion constructs of IRSp53 that lacked the IMD were unable to induce filopodia formation (data not shown). IMD-induced filopodia tended to be less frequently branched than those induced by full-length IRSp53. In marked contrast to the wild-type constructs, myc-tagged forms of IMDmut or of full length IRSp53 carrying the analogous mutation failed to display filopodia formation above the level of untransfected cells (Figure 5). This clearly demonstrates that IRSp53 mutants whose actin-binding and actin-bundling activity is compromised cannot induce filopodia, establishing a molecular basis of IRSp53-induced filopodia formation.

DNA constructs

Human IRSp53 in the mammalian expression vector pRK5myc was obtained from Dr Alan Hall (LMCB, London) (Krugmann et al, 2001). This was used as a template to amplify the region encoding the IMD (amino acids 1–250) of IRSp53 by polymerase chain reaction. Sites for restriction enzymes BamHI and EcoRI were introduced into the forward and reverse primers, respectively, and these sites were used to clone the construct into pRK5myc and pGEX4T2 (Amersham Biosciences) plasmids. Site-directed mutagenesis was performed using the QuikChange procedure (Stratagene) following the manufacturer's recommendations. All primers were purchased from Sigma-Genosys. DNA sequencing of all constructs was performed by the University of Birmingham Functional Genomics laboratory.

Protein expression and purification

Plasmids pGEX4T2 containing the coding sequence of the IMD were transformed into the methionine-auxotroph strain E. coli B834(DE3), obtained from Professor JB Jackson (University of Birmingham), for the purpose of crystallographic structure determination, or into E. coli BL21(DE3) for the biochemical assays. The transformants were used to inoculate 4 l of minimal M9 medium (Sambrook et al, 1998) supplemented with 50 μg/ml seleno-methionine (Fisher Scientific) and 100 μg/ml ampicillin (Sigma), while unlabelled protein was produced in Luria-Bertani (LB) medium supplemented with 100 μg/ml ampicillin. Cultures were grown at 37°C to OD₆₀₀ 0.45, and protein expression was induced with 1 mM IPTG (Melford labs) for 14 h at 25°C. Cells were lysed by sonication in PBS/1% Triton X-100 containing a protease inhibitor cocktail. The cleared sonicate was then passed three times over a glutathione–agarose column. The column was washed with the sonication buffer followed by thrombin buffer (150 mM NaCl, 50 mM Tris–Cl (pH 8.0), 5 mM MgCl₂, 2.5 mM CaCl₂, 1 mM DTT). Thrombin (40 U) (Calbiochem) was added in one column volume of thrombin buffer and the column incubated for 15 h (4°C). Thrombin-cleaved IMD was eluted with thrombin buffer and dialysed into 140 mM NaCl, 20 mM Hepes (pH 7.0), 5 mM MgCl₂, 2.5 mM CaCl₂, 1 mM DTT, bound to a carboxymethyl–sepharose column and eluted with a linear gradient of 140–500 mM NaCl. Fractions were analysed by SDS–PAGE, and those containing pure IMD were pooled and concentrated to 50 mg/ml protein using Vivaspin 15 (Vivascience) and centricon 10 (Amicon) ultrafiltration columns.

Crystallisation and structure determination

Crystals of the seleno-methionine-labelled form of the IMD of IRSp53 were grown by hanging drop vapour diffusion over a reservoir containing 15–18% (w/v) polyethylene glycol (PEG) 6000, 0.1 M sodium acetate (pH 4.6), 15–70 mM ammonium acetate, 5 mM DTT and small concentrations of either sucrose (2–6% (w/v)) or ethylene glycol (10–12% (v/v)). Crystals appeared and grew to final size over night, but showed signs of decay only a few days later. Therefore, crystals were flash frozen in liquid nitrogen promptly as follows. High-quality diffraction patterns were obtained by soaking crystals in a cryoprotection buffer based on the reservoir solution, but with PEG 6000 increased to 30% (w/v) and 20% ethylene glycol or sucrose prior to data collection from flash frozen crystals. Crystals were in space group P2₁ (Table I) with two molecules in the crystallographic asymmetric unit. A three-wavelength MAD data set was recorded on station 14.2 at SRS Daresbury (Warrington, UK) integrated and scaled using MOSFLM/SCALA (Leslie, 1992; CCP4, 1994). An initial set of 10 out of 16 possible selenium positions was determined from the anomalous differences of the ‘absorption peak' data using SnB v2.1(Weeks and Miller, 1999). Five additional positions, including that of the initial Met residue of chain A, were located by SOLVE (Terwilliger and Berendzen, 1999) and included in the phase calculation (30–2.2 Å), followed by solvent flattening using RESOLVE (Terwilliger, 2002). The resulting map was of a very good quality and allowed to unequivocally build the entire backbone (chain tracing using ARP/WARP (Morris et al, 2002) combined with manual building in O (Jones et al, 1991)), except for two residues at the C-terminus of both chains A and B, and two N-terminal residues that arose from the expression construct. Side chains were added manually and the model was refined (CNS 1.1 (Brünger et al, 1998), REFMAC5 (Murshudov et al, 1997)) against the peak wavelength data restrained by experimental phases. Several rounds of manual rebuilding interspersed with refinement lowered the free R-factor (5% of reflections) to 29.5%, at which point water molecules were added. Met residues were modelled as seleno-methionine, whereby the free R values were only marginally different when using sulphur–methionine instead. This probably reflected the fact that at present anomalous scattering factors cannot be modelled in the CCP4 distribution of REFMAC5 (Murshudov et al, 1997). Density for loops 2–3 is weak and a number of side chains in this region could not be built with confidence, in particular in chain A. The final model comprises 496 protein residues (248 each in chains A and B) and 221 water molecules. Temperature factors were refined applying a TLS correction (Winn et al, 2001) (two groups) and the minimisation converged to a free R value of 26.7% with excellent stereochemistry (Table I).

Sedimentation velocity

Sedimentation velocity experiments were carried out in a Beckman XL-A analytical ultracentrifuge (Beckman Coulter, Palo Alto, CA, USA) equipped with absorbance optics. Protein samples were dialysed into storage buffer as above including 1 mM DTT and loaded into cells with two channel Epon centre pieces and quartz windows. Data were recorded at 40 000 r.p.m., 20°C, using an An50Ti rotor and an absorbance wavelength of 280 nm, with scans taken every 6 min. Solvent density was measured using an Anton Paar DMA 5000 high-precision density metre. Partial specific volumes were calculated using the program SEDNTERP (Laue et al, 1992). Data were analysed using the program SEDFIT (Schuck, 2000). Sedimentation coefficient distributions were calculated using the Lamm equation modelling implementing maximum entropy regularisation. A total of 95 scans for each sample were analysed, representing the full extent of sedimentation of the sample. For the wild-type sample, data were fitted to a single species model using finite element solutions to the Lamm equation (Schuck, 1998).

F-actin-binding and F-actin-bundling assays

G-actin was purified from rabbit muscle as described previously (Spudich and Watt, 1971). G-actin was maintained in G-buffer: 2 mM Tris–Cl (pH 8.0), 0.2 mM ATP, 0.5 mM DTT and 0.2 mM CaCl₂. Polymerisation of the actin was achieved by addition of 50 mM KCl, 2 mM MgCl₂ and 0.1 mM EGTA, the resulting buffer being F-buffer. Concentrations measures given for F-actin in this paper refer to the initial concentration of G-actin prior to polymerisation. Plasmid constructs of wild-type and mutant forms of IMD were cloned into the pGEX4T2 expression vector and transformed into E. coli BL21(DE3) cells and purified as described in the above section. Proteins were dialysed against F-buffer and centrifuged at 435 000 g for 60 min at 4°C, immediately prior to use in actin-binding/bundling experiments. Protein concentrations were measured using the Biorad protein assay and confirmed by SDS–PAGE analysis.

Actin-binding experiments. F-actin (2.5 μM) was incubated with a range of concentrations (0.5–10 μM) of either wild-type or mutant IMD in F-buffer for 30 min on ice, and then centrifuged at 350 000 g (30 min, 4°C). Supernatants were removed, pellets rinsed with F-buffer then resuspended in a volume of F-buffer equal to that of the supernatant. Samples were analysed by SDS–PAGE and Coomassie staining, and quantified by densitometry using GENETOOLS (Syngene), and normalised with respect to the F-actin band intensity.

Actin-bundling experiments. F-actin (5 μM) was incubated with varying concentrations of either wild-type or mutant IMD in F-buffer at room temperature for 1 h, and then centrifuged at 16 000 g for 30 min at room temperature. Supernatants were removed and pellets resuspended in a volume of F-buffer equal to that of the supernatant.

Fluorescence microscopy of actin bundling. Rabbit muscle G-actin was labelled using FluoroLink Cy3-reactive dye (Amersham Biosciences) as described previously (Machesky and Hall, 1997). This was mixed with unlabelled G-actin, such that 15% of monomers carried the Cy3 label. The actin was polymerised and varying concentrations of the labelled F-actin was mixed with varying concentrations of wild-type and mutant IMD in F-buffer and incubated at room temperature for 1 h. The samples were then mounted between a slide and coverslip and imaged by fluoresence microscopy using a Zeiss Axioskop2 and Hamamatsu C47442-95 camera. Images were processed using Openlab (Improvision) and Photoshop 7.0 (Adobe).

Cell culture

COS7 cells were maintained in Dulbecco's modified Eagle's medium containing 10% foetal calf serum and antibiotics. Cells were seeded onto glass coverslips and transfected 24 h later using Gene Juice (Novagen) according to the manufacturer's recommendations. At 24 h after transfection, cells were fixed with 4% paraformalehyde and permeabilised in PBS containing 0.1% Triton X-100. Cells were stained with TRITC–phalloidin (Sigma) and anti-myc monoclonal antibody 9E10 (obtained from Cancer Research, UK), followed by an anti-mouse IgG secondary antibody conjugated to Alexa 488 (Molecular Probes).

Protein Data Bank (PDB) ID code

The coordinates of the IMD of IRSp53 have been deposited at the PDB (http://www.rcsb.org/pdb) under accession number 1Y2O.