生物谷报道：vinculin是一种高度保守的细胞内蛋白，它与细胞的粘附和迁移直接相关，但是它是如何作用的不是很清楚。这篇报道从蛋白质结构角度解决了这一现象，本文发表在近期的Nature上。
Nature AOP, published online 13 June 2004; doi:10.1038/nature02610
Structural basis for vinculin activation at sites of cell adhesion
We determined the structure of vinculin at 3.1 Å resolution in three different crystal forms grown from both high and moderate salt buffers. The tertiary and quaternary organization are identical in all cases. Vinculin comprises a 'bundle of bundles', with overall dimensions of 100 100 50 Å (Fig. 1 and Supplementary Fig. S1). The molecular building blocks are eight four-helix bundles, which are further organized into four tandem pairs, each of which resembles the structure of an amino-terminal fragment of -catenin. We therefore call this class of domain a 'vinculin/-catenin repeat' (VCR). Domains 1–3 comprise a VCR, in each of which a central helix runs the length (100 Å) of the domain. In domains 1 and 2, the two four-helix bundles pack end to end. Domain 3 is more divergent, with the two bundles interacting along their sides and with a distinct break in helicity in the middle of the central helix. Domain 4 is a four-helix bundle that packs end to end with domain 5, which is the vinculin tail domain (Vt); together, these two domains topologically and structurally resemble a fourth VCR, but with a large insertion at the centre comprising the proline-rich region and the first helix of Vt. This repetitive tandem organization suggests that the molecule has arisen through gene duplication.
Figure 1 Structure of full-length vinculin in its autoinhibited state. Full legend
High resolution image and legend (63k)
Domains 1, 2 and 3 together form a scaffold that buries a total of 5,000 Å2 of surface area (Table 1) and forms the vinculin head (VH) observed in electron microscopy images6 (note that VH is distinct from the proteolytic fragment 'Vh', which includes D4 and part of the proline-rich region). D1 and D3 form a pair of pincers across which sits D2, locking them in position. The pincers hold Vt in the autoinhibited structure; the conformation of Vt is essentially identical to that seen in the crystal structure of the isolated tail7. The D1–D3 contact at the back of the pincers buries 1,560 Å2 of surface area in a classic protein–protein contact with a hydrophobic core and polar or charged interactions at the periphery. The D2–D3 contact buries the most surface area (2,350 Å2), although it is more polar than the D1–D3 interface.
The N-terminal lobe of D2 (D2a) makes extensive contacts with D3, packing at right angles; the C-terminal lobe, D2b, is more or less free of contacts with other domains (and is not present in Drosophila and Caenorhabditis elegans vinculins). The contacts between D4 and D3, although burying 1,050 Å2 of surface, are largely polar and likely to be weak, suggesting that substantial interdomain movement is possible once Vt is freed from the head. Structural comparisons with the 'M fragment' of -catenin8 further support this notion (see below). In electron microscopy images of the open conformation6, a 'neck' region was appended to the head with variable conformations, and we suggest that this neck is D4.
Vt makes two contacts with the vinculin head and one with the neck (Figs 1 and 2). As expected, D1 packs against Vt to form a major interface (2,290 Å2 buried surface). In the crystal structure of a D1–Vt fragment9, the relative orientation of the two domains is essentially identical to that observed here, suggesting that the interface is rigid. The principal differences are found at the N terminus and the D1–D2 linker, where the linker adopts a non-native conformation in the D1–Vt fragment and three N-terminal residues take a different course. Two further intramolecular contacts contribute to the higher affinity of Vt for Vh than for D1. The first is a small end-to-end contact (530 Å2) between the bottom of the Vt bundle and the top of D4; this contact includes main chain hydrogen bonding and a well-ordered salt bridge between Glu 775 and Arg 978. A second interface occurs on the opposite side to D1, where D3 is brought into close apposition with Vt; this interface buries 770 Å2 and is largely polar.
Figure 2 Accessible surface of Vt in the context of the full-length molecule. Full legend
High resolution image and legend (46k)
Homology between the N-terminal VCR domains and the C-terminal tail domains of vinculin and -catenin was previously established. Our structure further shows that the two domains of the M fragment of -catenin are highly homologous to vinculin D3b and D4 (root-mean-square (r.m.s.) deviation of less than 1 Å; Fig. 3). However, the quaternary disposition of these two domains is highly variable in crystal structures of -catenin8, 10 and very different from that in full-length vinculin (Fig. 3b), and we propose that they provide a model of the open conformation of vinculin.
Figure 3 Comparison of vinculin domains 3b and 4 with the 'M fragment' of -catenin. Full legend
High resolution image and legend (47k)
D4 precedes the -catenin tail and, as in vinculin, it is separated by a flexible linker. This leaves unaccounted only a predicted four-helix bundle, which can be readily modelled by homology to D3a of vinculin. It is further possible that -catenin shares a similar quaternary organization with vinculin. Thus, although crystal structures of -catenin domain 1 lack the first helix11, sequence alignments clearly indicate that it is present in the intact molecule (Supplementary Fig. S1). By assuming a vinculin-like head-to-tail arrangement in -catenin, we find that the D1–tail hydrophobic interface is highly conserved; and by assuming a vinculin-like D1–D3 interface, we also find a similar hydrophobic contact. Direct evidence for an autoinhibited conformation for -catenin is currently limited, however, and dimerization also has a regulatory role10.
We previously showed that Vt undergoes a conformational change on binding phosphatidylinositol-4,5-bisphosphate (PtdIns(4,5)P2), an acidic lipid that is upregulated at cell adhesion sites, and we identified two basic surfaces on Vt as plausible PtdIns(4,5)P2-binding regions: a basic 'collar' surrounding the C-terminal arm, and a basic 'ladder' along an edge of helix 3 (ref. 7). Accordingly, point mutants in the collar (Lys911Ala and Lys924Ala) or basic ladder (Lys952Ala) reduce binding by 50%, whereas mutation of two basic residues in the neighbouring helix has no effect (Supplementary Fig. S2). Lys 911 and Lys 924 are next to His 906, a residue that is essential for PtdIns(4,5)P2-induced conformational changes in Vt12. In the context of the full-length molecule (Fig. 2), both the collar and the ladder point towards the outside of the molecule. The ladder is mostly exposed to solvent, although at its N-terminal end Lys 944 and Arg 945 make salt bridges to acidic residues on the head. The collar is partially obscured by a loop positioned N-terminal to the tail domain (the 'strap'), which lies across the face of the H1–H2 hairpin. Thus, the quaternary organization of the closed conformation partly occludes the PtdIns(4,5)P2-binding surface, as well as inhibiting PtdIns(4,5)P2-dependent conformational changes in the tail.
At the bottom of the strap (which is the end of the proline-rich region) is a PPPP motif that is essential for binding to the Arp2/3 complex3. One of these prolines, Pro 878, packs against the base of Vt and the C-terminal arm, consistent with the low affinity of vinculin in its autoinhibited conformation for Arp2/3. We predict that on binding PtdIns(4,5)P2 the strap is released from its location, which would rationalize the PtdIns(4,5)P2 dependence of Arp2/3 binding3. At the beginning of the proline-rich region, an FPPPP motif is implicated in binding to VASP and vinexin13, 14. VASP binding is stimulated by PtdIns(4,5)P2 (ref. 15), suggesting that in the isolated molecule this region packs against the molecular surface. In our structure, however, this region is involved in a crystal contact and it is unclear what its conformation would be in the isolated molecule. The rest of the proline-rich region is poorly ordered or variable in conformation and does not pack against the molecular surface, consistent with the sensitivity of this region to V8 protease cleavage16.
Talin rod does not bind to full-length vinculin in solution (Fig. 4b). The binding of talin domains to vinculin D1 is strong but characterized by slow kinetics, suggestive of a conformational change on binding17. To test this hypothesis, we expressed a fragment of vinculin corresponding to domains D1–D3 (VH, residues 1–718). In the absence of a binding partner D1 is loosely folded, as judged by its low melting temperature and enthalpy of unfolding (Fig. 4c). On binding Vt, the VH structure becomes highly ordered (Fig. 4d). Binding of the talin rod (relative molecular mass 220,000; Mr 220K) to VH results in the appearance of a similar strong new peak (Fig. 4e), consistent with tight binding to the D1 domain (dissociation constant, Kd 40 nM)17.
Figure 4 Calorimetric analysis of vinculin and ligand complexes. Full legend
High resolution image and legend (91k)
As the binding site for -catenin on VH had not previously been mapped, we expressed a domain-sized fragment (equivalent to D3 of vinculin) of -catenin containing the crucial region for vinculin binding18: this fragment binds to vinculin D1 in a very similar fashion to the binding of talin rod (Fig. 4f). Using isothermal titration calorimetry (ITC), we observed tight binding (Kd = 82 19 nM) that was endothermic (H = +8.2 1.2 kcal mol-1) and entropy driven (+ 60 5 cal K-1 mol-1), consistent with a major conformational change on binding (Supplementary Table S1).
To explore this further, we designed a mutant that would stabilize the conformation of D1 observed in our vinculin crystal structure. The mutant Ala50Ile fills a cavity in the hydrophobic core between helices H1 and H2 (Fig. 4a). Melting profiles show that the Ala50Ile mutation in the context of VH does indeed stabilize the D1 fold to a significant extent (Fig. 4c); moreover, Vt binds more strongly to the mutant than to the wild-type VH (Fig. 4d). However, binding of the mutant VH to -catenin is reduced by a factor of about 10 (Kd = 710 40 nM) and talin binding is also greatly reduced (Fig. 4e, f). Thus, binding of both talin and -catenin to VH requires conformational changes in the D1 bundle that are inhibited by Vt binding. A crystal structure of D1 in complex with a short talin-derived peptide supports this hypothesis9.
Head-to-tail binding has been estimated with proteolytic and recombinant fragments (Kd 20–50 nM), but this value presumably underestimates the intramolecular affinity when the head and tail are covalently linked. A peptide derived from the talin rod (residues 1,944–1,969) activates full-length vinculin when present in a 500–1,000-fold molar excess by binding to vinculin D1 (ref. 17). We measured the binding of this peptide to VH by ITC (Kd = 400 150 nM). Assuming that binding to full-length vinculin is negligible and that the system has achieved equilibrium, this gives an estimate for the intramolecular VH–Vt Kd of less than 1 nM. A similar result is obtained for a phage-display peptide that binds to the head with an affinity of 100 nM and activates vinculin when present in a 300-fold excess, implying an intramolecular Kd of about 0.3 nM (refs 19, 20). These values imply an affinity that is roughly 100 times higher than that for the isolated VH and Vt domains, explaining why talin rod (Kd 30 nM for binding to the isolated head) and -catenin (Kd = 80 nM) do not bind significantly to full-length vinculin when present at an equimolar concentration.
Our data show that the binding sites on the head, tail and proline-rich regions are structurally distinct, but conformationally and thus thermodynamically linked. Therefore, activation of vinculin could be achieved by a combinatorial input of ligands to these distinct regions in which the activating potentials (that is, the difference in energy between binding to the closed and open states) are essentially additive. Once these potentials exceed the intramolecular binding energy, the molecule should open and binding should ensue, provided that a kinetic pathway to equilibrium is available. Estimates of phospholipid (10% PtdIns(4,5)P2 in phosphatidylcholine vesicles) binding to Vt are in the micromolar range, whereas binding to the full-length molecule is negligible at physiological salt21. This binding energy is insufficient to activate vinculin by itself, but in combination with PtdIns(4,5)P2 micelles (in a 100-fold molar excess) talin forms a ternary complex with vinculin and PtdIns(4,5)P2 (ref. 22). Because the PtdIns(4,5)P2-binding site is only partly occluded in the full-length molecule, we can propose a kinetic pathway to activation in which PtdIns(4,5)P2 binding and the ensuing conformational changes transiently release the tail from the head, allowing talin or -catenin to bind to the head and ligands such as VASP and vinexin to bind the proline-rich region. The full repertoire of input signals remains to be determined but may also include the binding of tensin to the neck (D4; D. Lin, personal communication).
Similar considerations apply to the binding of F-actin, where the binding affinity between F-actin and Vt in physiological salt has been estimated at Kd 1 µM, whereas binding to the full-length protein is very weak. In the structure of Vt in complex with F-actin23, the actin-binding surface comprises two regions of Vt. One of the binding surfaces is exposed in the full-length molecule and is available for binding. The second site is only partly exposed and is blocked by the vinculin head. The principal steric clash occurs with the end of helix 1 in D1, which has been implicated in structural rearrangements on binding talin9. This arrangement provides a kinetic pathway for F-actin to bind to vinculin and to transiently loosen intramolecular interactions, leading to activation if a head-binding ligand is available. Because F-actin binding and PtdIns(4,5)P2 binding are exclusive20, these ligands provide two alternative pathways to activation.
The organization of tandem domains into an autoinhibited conformation that provides a combinatorial output to numerous input signals is an emerging theme among signal transduction proteins24, 25. Vinculin extends this theme to proteins that lack a catalytic domain, that seem to have arisen from gene duplication of a single protein interaction domain, and that respond primarily to the spatial colocalization of their binding partners rather than to posttranslational modification.
Methods
Protein expression and purification Recombinant His-tagged chicken vinculin was cloned into a pET15b vector, expressed in Escherichia coli BL21(DE3) cells, and purified with a HiTrap affinity column (Amersham Pharmacia Biotech). After dialysis against 20 mM sodium phosphate (pH 7.5), 150 mM NaCl and 0.1 mM EDTA, the protein was further purified by ion exchange chromatography (DE52) and concentrated in an ultrafiltration cell (Amicon) to 6 mg ml-1. Wild-type and Ala50Ile mutant vinculin D1–D3 (residues 1–718) and -catenin D3 (residues 273–510) were cloned, expressed and purified by a similar protocol. We expressed His-tagged talin rod (residues 397–2,541) in a pET30a vector and purified it as described26. Vt was expressed and purified as described7. Wild-type vinculin was purified from chicken gizzard smooth muscle as described27. Recombinant selenomethionine (SeMet) vinculin was expressed in minimal media supplemented with SeMet as described for Vt7.
Structure determination Crystals of recombinant vinculin were grown at 30 °C by a hanging-drop vapour diffusion method, with protein at 4–6 mg ml-1 in 20 mM NaCl, 20 mM Tris pH 8.0 and 5 mM dithiothreitol, against a reservoir of 100 mM cacodylate (pH 6.5) and 1–1.2 M (NH4)2SO4. Three crystal forms were obtained (C2: a = 57 Å, b = 126 Å, c = 170 Å, = 94.9°; P21: a = 57 Å, b = 353 Å, c = 69 Å, = 114.1°; and C2221: a = 56 Å, b = 127 Å, c = 352 Å). Crystals of chicken gizzard vinculin grew in 30% PEG 5000 MME, 0.2 M (NH4)2SO4, 100 mM MES pH 6.5 and 10 mM MgSO4, and were very similar to the orthorhombic recombinant crystals, although they diffracted less well (dmin 3.9 Å). Crystals were transferred directly to a cryobuffer consisting of 1.5–1.8 M (NH4)2SO4, 100 mM cacodylic acid (pH 6.5) and 20% glycerol, flash-frozen in an N2 stream, and then annealed by re-equilibration in cryobuffer for a few minutes before being returned to the N2 stream.
Multiwavelength anomalous diffraction data were collected from orthorhombic SeMet crystals at the Stanford Synchrotron Radiation Laboratory (SSRL) beamline 9.1 (Supplementary Table S2). Crystal lifetime limited data collection to two wavelengths: the high-energy remote and inflection point (0.979 and 0.94 Å, respectively). Because an increase in mosaic spread and cell volume was observed, the data were reprocessed using local scaling. The anomalous component of the structure factor was calculated by XPREP, and Patterson correlation analysis was done by the program SHELXD28 at a resolution of 4.5 Å. Because there was no clear occupancy drop for the sites, phase refinement was carried out with SHARP29, which refined coordinates and occupancy but not temperature factors. By this approach, 25–27 selenium sites were confirmed out of an expected total of 39, leading to a map with visible helical density.
Phased molecular replacement using BRUTEPTF (http://russel.bioc.aecom.yu.edu/server/NYSGRC.html) located the intact Vt (Protein Data Bank (PDB) accession code 1QKR) and -catenin dimerization (PDB 1DOV) domains, as well as one half of the -catenin M-fragment (PDB 1H6G). Chain tracing for the remainder of the molecule was guided by the selenium positions. Refinement was done with the CNS package30 against the maximum-likelihood target using amplitudes and phase probability distribution to 3.1 Å resolution. No cut-off was used in refinement. Convergence was reached at an Rwork of 31.6% and an Rfree of 35.7%. The current model includes residues 1–855 and 875–1,065. Of these, 87.4%, 10.9% and 1.5% are in the most favoured, additionally and generously allowed regions, respectively, of the Ramachandran plot. The structure was used as a molecular replacement search model for the other crystal forms. Refinement proceeded for each to Rfree values of 0.35–0.40. Minor differences were observed in interhelical loops and the proline-rich region at sites of crystal contacts. Coordinates for the model of -catenin are available from the authors on request.
Calorimetry Differential scanning calorimetry (DSC) experiments were done at a scanning rate of 1 K min-1 under 3.0 atm of pressure on an N-DSC II differential scanning calorimeter (Calorimetry Sciences). Before measurements, all protein samples were dialysed against PBS buffer, which was used as the reference solution. All thermal transitions were irreversible under the conditions used. See Supplementary Methods for a further description of the DSC method. ITC was carried out on a VP-ITC calorimeter (Microcal). We injected 8-µl aliquots of solution containing either 940 µM -catenin D3 or 1.0 mM talin peptide into a cell containing 70–100 µM vinculin D1–D3 construct (wild type or the Ala50Ile mutant). In each experiment 35 injections were made. The experiments were done at 23°C. Before ITC titrations, all protein samples were dialysed against PBS buffer. Experimental data were analysed with Microcal Origin software provided by the ITC manufacturer (Microcal).
Supplementary information accompanies this paper.
Received 1 April 2004;accepted 29 April 2004
