While it is self-evident that all extracellular molecules are an integral part of a multicellular organism, it is paradoxical that they are often considered to be dissociated from cells. The reality is that a continuum of dynamic, bi-directional interactions links the intracellular environment through cell-surface receptors to multimolecular extracellular assemblies. These interactions not only control the behaviour of individual cells, but also determine tissue architecture. Adhesion receptor function is partly determined by an ability to tether the contractile cytoskeleton to the plasma membrane, but there is also evidence that integrin receptors modulate signalling events that are essential for cellular differentiation. A major challenge is now to integrate work at the atomic, molecular and cellular levels, and obtain holistic insights into the mechanisms controlling cell adhesion. In the present study, we review current knowledge of the molecular mechanisms employed by cells to integrate with the extracellular matrix. Two main topics are covered: the adaptation of integrin structure for bi-directional signalling and the integration of integrin signalling with other receptors.
Integrin structure, priming and activation
In mammals, 18 α- and 8 β-integrin genes encode polypeptides that combine to form 24 α,β heterodimeric receptors . Both subunits are non-covalently associated, type I transmembrane proteins with large extracellular and mostly short cytoplasmic domains. On the basis of their primary structure, integrins fall into two subfamilies determined by the presence or absence of a von Willebrand factor A-domain in the α subunit. Beginning with the X-ray crystal structure of the isolated αM integrin A-domain in 1995 , recent years have seen major advances in our understanding of integrin tertiary structure. A step-change came with the solution of the first structure of a full-length extracellular dimer (for αVβ3) in 2001 , followed closely by the structure of a ligand mimetic–receptor complex, formed by soaking an RGD peptide into pre-existing αVβ3 crystals . The overall shape of the crystallized conformer of αVβ3 resembles a large ‘head’ on two ‘legs’ (Figure 1). In contrast with models derived from electron microscopic and biophysical analyses , both legs are bent in the crystal structure. The head of the integrin, which is the minimal fragment that contains ligand-binding activity, comprises a β-propeller from the α subunit, an A-domain from the β subunit (the βA-domain) and an immunoglobulin ‘hybrid’ domain below the βA-domain. The α-subunit leg of the integrin comprises three β-sandwich domains, termed ‘thigh’, ‘calf1’ and ‘calf2’. The β-subunit leg contains a plexin–semaphorin–integrin domain, which is disordered in the structure, four EGF-like repeats and a cystatin-like fold. The bend in the integrin is between thigh and calf1 (in αV), and at the conjunction of the hybrid domain, two EGF repeats and the PSI domain (in β3). Since 2001, there have been no further integrin crystal structures reported, and, therefore, the conformational diversity of the family remains unclear.
Cartoon of the straightened form of an integrin which lacks an αA-domain
In the RGD–αVβ3 complex structure, the peptide lies at the interface between the αV β-propeller and the β3 A-domain, and the carboxyl group of the aspartate co-ordinates a bivalent cation at a so-called MIDAS (metal-ion-dependent adhesion site) . A similar receptor–cation–ligand interaction had been observed previously in a crystal structure of a complex between the α2 A-domain and a triple-helical collagen peptide . While complementary conformations of ligand acidic motifs and receptor cation-binding pockets might determine the specificity of receptor-ligand binding, there is also evidence that integrin ligands contain accessory or ‘synergy’ sites for receptor binding. FN (fibronectin) is the best-characterized ligand in this regard, and although it has been suggested that receptor engagement of the synergy site is transient , a combination of gain-of-function analyses with integrin α subunit chimaeras, anti-functional mAb (monoclonal antibody) epitope mapping, and structure determination by small-angle X-ray scattering have led to a detailed topological model for the FN–integrin complex [8,9]. Solution of the structure of a macromolecular ligand–integrin complex remains a priority for understanding how integrins form stable interactions with ECM (extracellular matrix) proteins.
The dynamic nature of integrin function, by which cells use adhesion to sample their pericellular environment and respond by changing their position and differentiated state, demands a highly responsive receptor structure (Figure 1). Gross conformational changes in integrins have been detected by a variety of biophysical and biochemical techniques (reviewed in ), and much work has been performed on activation-dependent binding of mAbs to integrins. A subset of these mAbs recognize so-called LIBS (ligand-induced-binding sites). It has been hypothesized that LIBS mAbs stimulate integrin-ligand binding by displacing conformational equilibria in favour of ligand-competent and ligand-occupied forms of the integrin. Shape changes reported by LIBS mAbs can also be triggered from the cytoplasmic side of the plasma membrane , but the degree of similarity between the changes induced from either side of the plasma membrane is currently unclear. To separate these two processes conceptually, we have proposed the term ‘priming’ to indicate the acquisition of ligand-binding ability by integrins and ‘activation’ to denote ligand-induced changes .
The availability of the crystal structure of αVβ3 has stimulated structure–function analyses of integrin priming. The relevance of the bend in the integrin dimer has received much attention. Although it has been proposed that integrins remain bent , several lines of evidence indicate that bent integrins are inactive and extended integrins are primed. For example, epitopes that become exposed on integrin priming, and residues that restrain priming, are buried in the bent knee of β2 integrins , and locked-bent integrins containing engineered disulphide bonds have very low affinity for ligand . Furthermore, predominantly bent conformers of β3 integrins are observed by electron microscopy in the presence of Ca2+, whereas extended conformers are found in Mn2+ .
It is now well established that leg separation underpins both priming and activation of integrins. Thus some stimulatory mAb epitopes located in the leg regions are exposed by activating cytoplasmic domain mutations , forced association of the membrane-proximal regions of the α- and β-subunit cytoplasmic domains with an engineered coiled-coil constrains the integrin in an inactive state , unclasping these constrained integrins leads to a spatial separation of the legs , and fluorescence resonance energy transfer studies show that integrin cytoplasmic domains are close to each other in the resting state, but undergo spatial separation during both priming and activation .
Using LIBS mAbs as activation state reporters, we have shown that movement of both the α1 and α7 helices of the β1 A-domain contributes to priming [19,20]. Shape changes in these elements had previously been observed in αA-domain crystal structures. βA-domain conformational changes are coupled with a swing-out of the underlying hybrid domain, as activating mutations in the α7 helix cause increased exposure of LIBS mAb epitopes obscured at a hybrid domain-β-propeller contact . Hybrid domain swing-out has also been observed using electron microscopy  and small-angle X-ray scattering . The extent to which different ligands stabilize different integrin conformers and in turn transduce agonistic effects across the plasma membrane, and the structural mechanisms that couple extension, leg separation and βA-domain conformational change, are not clear, but will underpin our future understanding of how integrins work as dynamic receptors.
Integrin-receptor cross-talk and signalling assemblies in migrating cells
Ligand engagement by integrins initiates signalling responses that include transduction of mechanical force to the cytoskeleton and spatial compartmentalization of signalling complexes. There is now evidence for alterations in the fluxes of almost all known signalling pathways subsequent to integrin engagement, suggesting that adhesion receptor function is integrated with other receptor systems. However, the mechanisms responsible for converting integrin ligation into an efficacious signal are not known. A hierarchy of recruitment of signalling and cytoskeletal molecules to integrins has been demonstrated using immunocytochemical approaches , but as yet evidence for ligand/agonist-specific differences in the composition and/or organization of this hierarchy is lacking.
Early studies employing receptor chimaeras demonstrated that some integrin-specific functions were conferred by the cytoplasmic domain. On the basis of an NMR analysis of a complex between recombinant forms of the αIIb and β3 cytoplasmic domains, it appears that both subunits are largely unstructured, but are held together by a juxtamembrane interaction of two α-helical regions . A variety of approaches including affinity chromatography, equilibrium gel filtration, IP blotting, synthetic peptide binding and two-hybrid analysis have been used to identify molecules that bind integrins (reviewed in ). Interacting proteins include cytoskeletal components such as talin, α-actinin and filamin, adapters such as paxillin, Rack-1, ICAP-1 and β3-endonexin and kinases such as Src.
A central role in integrin signalling has been established for talin. Genetic or translational inhibition of talin disrupts the cytoskeleton and cell adhesion with a similar phenotype to integrin deletions, whereas overexpression of an integrin-binding fragment of talin induces priming triggered by multiple pathways  and causes structural perturbation of the association between the integrin cytoplasmic domains . Crystallization of a talin fragment–integrin β3 cytoplasmic peptide chimaera has revealed a PTB domain-like mode of binding that may be common to several integrin effectors [25,26]. Since talin binds vinculin, actin and FAK, it has the potential to co-ordinate the recruitment of many key integrin-signalling molecules. Furthermore, talin binds and activates PIPKIγ, which leads to increased PIP2 production and in turn activates several cytoskeletal proteins.
Integrins undergo cis interactions with a number of different receptors, and thereby spatially regulate diverse signalling responses (reviewed in ). Direct extracellular associations with members of the TM4 family, the urokinase receptor uPAR and CD47 have been established, as has indirect co-clustering with several growth factor and cytokine receptors. Cross-talk at the level of signalling occurs with semaphorins, ephrins and syndecans. It is remarkable that most ECM molecules possess both integrin- and syndecan-binding sites, and a clear synergistic relationship exists between these two families. Thus α5β1-dependent FA (focal adhesion) formation on FN requires engagement of, and signalling through, a syndecan co-receptor [28,29].
In spread fibroblasts adherent to FN in vitro, adhesion signalling complexes are distributed focally rather than diffusely, and are manifested as asymmetric patches, flecks and stripes. These contact points are found all over the ventral surface, and are usually associated with the contractile polymers of the cytoskeleton. Detailed morphological and functional analyses have defined three major forms of adhesion contact: FC (focal complexes), FA and FB (fibrillar adhesions; Figure 2; reviewed in ). These contacts reflect different stages of interaction of cells with the ECM, and each is formed and disrupted in a cyclical manner as cells translocate. Initially, FC form at the posterior edge of ruffling membrane, where they anchor the short filopodial struts and lamellipodial meshes of actomyosin that mediate membrane protrusion. When protrusion ceases or the lamellipodium retracts, FC transform into larger FA, which provide a more robust anchorage through transcellular actomyosin-containing stress fibres. In turn, FA evolve into FB, which are the major sites of FN matrix deposition [31,32]. Adhesion contact-like structures have been observed in vivo in smooth-muscle cell plaques and myotendinous junctions and, recently, in embryonic three-dimensional ECM , thereby validating the use of cell cultures for analysis.
Indirect immunofluorescence analysis of human fibroblasts showing FAs and FBs
The members of the Rho family of small GTPases are recognized to play a central role in modulating the actin cytoskeleton (reviewed in ). Cdc42 and Rac1 promote FC formation and membrane protrusions at the leading edge through filopodia and lamellipodia, whereas RhoA mediates FA formation and causes retraction of the trailing edge. Co-ordination between these pathways involves activation of p190RhoGAP, which may permit membrane protrusion by suppressing RhoA activity. Subsequently, the FC to FA transition is dependent on RhoA and ROCK-mediated myosin II contractility through phosphorylation of myosin light chain. Integrin occupancy controls the activity of Rho family GTPases and regulates their translocation to plasma membrane microdomains [35,36], but the molecular links between integrins and Rho GTPases are not understood.
At least 50 different proteins have been localized to FA , but analyses of the differences between FC, FA and FB are still in their infancy. FC and FA share many components, but FC tend to lack zyxin and tensin , whereas FB contain α5β1 and tensin, but lack most of the other FA components, including αVβ3 . Thus while compositional changes do take place as adhesion contacts mature, the signalling events that regulate these transitions are not well understood. In particular, how integrin activation with ligands in the ECM modulates the assembly, maturation and turnover of different adhesion contacts will be an important area of future study. Preliminary evidence from our laboratory (Figure 2) suggests that integrin conformation can vary with location in the cell, implying that bi-directional communication is important for co-ordinating cell–matrix integration.
Signalling Outwards and Inwards: A Focus Topic at BioScience2004, held at SECC Glasgow, U.K., 18–22 July 2004. Edited by J. Challiss (Leicester, U.K.), A. Harwood (University College London, U.K.), M. Humphries (Manchester, U.K.), C. Isacke (Institute of Cancer Research, London, U.K.), R. Liddington (Burnham Institute, La Jolla, CA, U.S.A.), T. Palmer (Glasgow, U.K.), K. Siddle (Cambridge, U.K.), C. Sutherland (Dundee, U.K.), H. Wallace (Aberdeen, U.K.) and M. Welham (Bath, U.K.).
This work was supported by grants from the Wellcome Trust.