Staffan Johansson, Gunbjørg Svineng*, Krister Wennerberg*, Annika Armulik, Lars Lohikangas

Department of Medical and Physiological Chemistry, The Biomedical Center, Box 575, S-751 23 Uppsala, Sweden

Received 2/20/97; Accepted 2/27/97; On-line 3/1/97

*GS and KW have made equally important and major contributions to this article.

3. DISCUSSION

3.1. Fibronectin

Fibronectin is a prototype cell adhesion protein, widely distributed in the tissues of all vertebrates and a potential ligand for most cell types (11). It is present as a polymeric fibrillar network in the ECM and as soluble protomers in body fluids. The protomer consists of two subunits linked in an antiparallel manner by a pair of disulfide bridges at the C-termini. Fibronectin is a good example of a mosaic protein since essentially the entire primary sequence, except for the V segment, is comprised of three types of repeating amino acid motifs (Fig. 1).

Figure 1. Schematic model of a plasma fibronectin protomer. The protein is a dimer of two subunits which are identical except for the inclusion of the V-segment in one of the chains. The subunits are covalently connected close to the C-termini by two disulfide bridges. Three types of homologous structural units called I, II, and III make up most of the protein (represented by small ovals, triangles, and squares, respectively). Integrin binding sites are indicated in one of the subunits by arrows. The fibronectin found in blood plasma is a major form of the protein, but other splice variants are present at other locations and at specific developmental stages. The alternatively spliced exons (III_A, III_B, and V) are marked in green and may be present in one or both of the subunits. The known variants of the splicing reactions are illustrated.

Several variants of the protein are formed by alternative splicing of the pre-mRNA at three positions: the type III repeat units A and B can independently be included or excluded, while the V segment is subject for a more complex splicing pattern (Fig. 1).

Several of the repeating units function together as distinct domains with specific biological activities (12). The domains are separated by short hinge regions which allow the molecule to adopt different overall shapes. The native soluble protomer is believed to have a globular configuration (13, 14), while an elongated form would be adopted during polymerization to fibrils (15, 16). However, it should be noted that the procedures for isolation of fibronectin require unphysiological conditions (4 M urea is commonly used), hence the native state of the protein is really not known. Conformational changes in the fibronectin molecule have been demonstrated to be key events for some of its functions (17-20), but it is not known if the regulation of function is related to the gross shape alterations or to more subtle structural rearrangements.

Two regions in each fibronectin subunit possess cell binding activity: III_9-10 and III₁₄-V. The amino acid sequence RGD, a widely occurring cell adhesive motif originally discovered in fibronectin (21), is located in III₁₀ The RGD motif in fibronectin and other cell adhesion proteins is the most important recognition site for about half of all known integrins. The affinity for short peptides containing the RGD sequence varies significantly among these integrins (22, 23), but for all receptors the affinity for larger protein fragments or the intact proteins is orders of magnitude higher than for short RGD-containing peptides (24, 25). This indicates that areas in the vicinity of the RGD site contribute to the contact surface between ligands and integrins. In fibronectin, one such site of major importance for the interaction with some of the fibronectin binding integrins has been located to III9 (26) This site, which acts in synergy with the RGD site, has the minimal amino acid sequence PHSRN (27) (Table 1, Fig. 1 and Fig. 2).

2 Abbreviations: FN, fibronectin; FG, fibrinogen; LN, laminin; MAdCAM, mucosal addressin cell adhesion molecule; OP, osteopontin; TN, tenascin; TSP, thrombospondin; VCAM, vascular cell adhesion molecule; VN, vitronectin; vWF, von Willebrand factor.

Figure 2. Structure of the fibronectin fragment III_7-10 as determined by X-ray crystallography of the recombinant protein fragment. The fragment is presented in four different angles with a stepwise rotation of 60 degrees. The four repeats are shown in different colors: pale purple, blue, green, and yellow for III₇-III₁₀, respectively. The synergy site (PHSRN) in III₉ repeat is marked in blue color, and the GRGDS loop in III₁₀ is marked in red. In fibronectin, as well as in many other proteins, several type III units link to each other as an extended array. The interfaces between the units in fragment III_7-10 are highly variable, suggesting that they may be important for the structure and interactions of the protein. The interdomain surface between III₉ and III₁₀ is smaller than between other units in FN_7-10. Therefore, they are likely to interact less strongly, and this junction may represent a flexible hinge region. Furthermore, the link between III₉ and III₁₀ has an unusually small rotation, resulting in the exposure of the RGD loop in III₁₀ and the synergy site in III₉ on the same side of the molecule. The two sites are separated by 30-40Å, a distance which one integrin molecule could span. The RGD loop is well exposed extending ~10Å away from the body of the molecule. These structure data were obtained from Leahy et al., 1996 (31).

The III₁₄-V region is recognized by two integrins (Table 1), and for both of these receptors three binding sites have been identified within this region (Fig. 1). The so called CS1 site has approximately 20-fold higher affinity for the integrins than the other two sites, namely CS5 and H1 (32). However, CS1 as well as CS5 can be spliced out and are therefore available only in some fibronectin molecules, while H1 is present in all forms of the protein. The related sequence motifs LDV and IDA are the key cell binding structures in CS1 and H1, respectively (33, 34). The active site in CS5 has been identified as REDV in human fibronectin, corresponding to RGDV in the rat and bovine proteins (35). Thus, the binding sequence in CS5 shows similarity to both the RGD and the LDV motifs. It is worth noting that integrin binding sites in fibronectin as well as in other proteins usually contain an aspartic acid as one of the critical amino acids.

3.2. Integrins

Most integrins function in cell-matrix contacts, some mediate cell-cell interactions, and a few participate in both types of contacts. (1) All integrins are composed of two noncovalently associated subunits, denoted alpha and ß, which both span the plasma membrane. The combination of alpha and ß subunits determine the specificity for extracellular ligands as well as intracellular signaling events.

A striking feature of many integrins is the ability to bind multiple ligands. For example, integrin alphaVß3 serves as a receptor for vitronectin, fibronectin, fibrinogen, von Willenbrand factor, thrombospondin, and osteopontin (36). Since each of these proteins contain an RGD sequence as the dominating recognition motif for integrin alphaVß3, the ligands can be envisaged to bind to one common binding site in the receptor. In the case of integrin alpha1ß1, laminins and collagens are recognized as ligands, although the two groups of proteins have no known structural similarities. In spite of this and several other examples, there are no indications that there would be separate binding sites for different ligands on an integrin (although the possibility remains open).

It is also commonly found that a particular cell adhesion protein can bind to more than one type of integrin. Fibronectin is an extreme case, for which binding to ten different integrins have been described (Table 1). This should not be interpreted as redundancy of receptors which could compensate for each other if necessary; rather, it is clear that different intracellular signals are generated by the ligand depending on which integrin it interacts with. The point is most convincingly illustrated by the strikingly different mouse phenotypes obtained after disruption of the various integrin genes (see below) (37, 38).

3.3. Molecular mechanisms of ligand-integrin interactions

A central concept for the function of integrins is their ability to shift between active and inactive ligand binding states, by alteration of the conformation of the extracellular domain (39). Transitions between the two states are dynamically regulated by the cell through energy dependent events involving the cytoplasmic tails of the integrin (40). A third conformational state is acquired after ligand-occupation, as demonstrated by exposure of epitopes called ligand-induced binding sites (LIBS) for antibodies. Experimentally, the three conformations can be induced or stabilized by the binding of different monoclonal antibodies to the alpha or ß subunits at epitopes distinct from the ligand binding site (39). In adherent cells the integrins are mainly in the active state, but possibly become inactivated during specific situations, e.g. cell migration and mitosis. In circulating cells the integrins are present on the cell surface in an inactive conformation until the cells are exposed to factors which trigger intracellular reactions leading to activation of the integrins (41, 42).

Intimately related to the conformational rearrangements in integrins is their requirements for extracellular divalent cations for ligand binding ability. There is evidence for at least three functionally important coordination sites for divalent cations in the extracellular integrin domains, each one exhibiting different ion preference (43). Mg²⁺ is most likely the dominating ion which supports ligand binding to integrins under normal physiological conditions. Mn²⁺ stimulates ligand binding to integrins even more strongly, but it is questionable whether this is of physiological relevance. The Mn²⁺ effect is due to the fact that the ion can support ligand binding, possibly by the same mechanism as Mg²⁺. Additionally, Mn²⁺ is capable of inducing/stabilizing a ligand binding conformation of otherwise inactive receptors (43, 44). Thus, Mn²⁺ can activate integrins independently of the intracellular events which normally regulate the extracellular conformation of the receptor (i.e. in the presence of Mg²⁺). In contrast, Ca²⁺ at millimolar concentrations has an inhibitory effect on the ligand binding ability of most integrins, by inducing the inactive conformation. However, the role of Ca²⁺ is complex since low concentrations (µM) appear to stimulate ligand binding. This indicates that Ca²⁺ can bind to at least two different sites in integrins (43-45). The number of binding sites for Mn²⁺ has been directly investigated in one study; by equilibrium gel filtration three Mn²⁺ ions were found to bind to each alphaIIbß3 molecule (46). It is unknown whether Mg²⁺ binds to more than one site in integrins, and it is also unclear if any of the metal ion (Me²⁺) sites can accept alternative divalent ions.

Several good candidates for ion coordination sites have been identified in fibronectin binding integrins; different alpha subunits contain three or four homologous sequences (DxDxDGxxD) which are similar to the Ca²⁺-binding EF-hand motif (47), and a second type of cation binding motif, DxSxS, is located in the ß subunits (Fig. 3) (48, 49).

Direct binding of divalent cations has been demonstrated for the latter site; a 14 amino acid peptide from the ß3 subunit (ß3_118-131; shown in Fig. 3), which is strongly conserved among all ß subunits, was found to bind Mn²⁺, Mg²⁺, or Ca²⁺ with 1:1 stochiometry (46). It is not known if this site has an equally broad ion specificity in the intact receptor. Interestingly, ß3_118-131 could also bind RGD peptides (46). It had previously been suggested that one of the integrin-bound ions may be directly involved in bridging to an aspartic acid residue of the ligand (e.g. RGD or LDV sites) (50). However, the RGD peptides bound also to a mutant form (D119A) of ß3_118-131 which lacks the ion binding ability. Furthermore, the ion bound to the normal ß3_118-131 was displaced upon addition of RGD peptides. Ligand binding to an intact receptor had a similar effect, causing two of the three Mn²⁺ ions bound to alphaIIbß3 to be released (46). Thus, the interactions between integrins and RGD-containing ligands appear to be independent of a direct involvement of divalent ions, but ions and ligands mutually influence their interactions with the receptor through allosteric regulation of the receptor conformation.

3.4. Ligand specificity of fibronectin binding integrins

Integrins alpha5ß1, alphaIIbß3, and alpha4ß1 have been studied in much greater detail than the other fibronectin-binding integrins. These receptors are therefore discussed first, followed by the others in the same order as they appear in Table 1.

Integrin alpha5ß1: This integrin is a prototype fibronectin receptor in the sense that it was the first to be identified (51), and, in contrast to most other fibronectin binding integrins, it is specialized for binding this ligand. alpha5ß1 is expressed by many cell types and is probably the major fibronectin receptor in several of these.

The RGD loop in III₁₀ is the critical recognition site for alpha5ß1 (21), but the synergy site PHSRN in III₉ is required for high affinity binding (27). Screening of peptide libraries for binding to isolated alpha5ß1 by the phage display technique enriched peptides containing the sequence RGDGW/F (52). Notably, the replacement of the R for a K abolished the binding to alpha5ß1, while alphaIIbß3 bound almost equally well to both variants (53).

A curious finding from the phage display screening was the strong and specific binding of the peptide RRETAWA to alpha5ß1 (52). The peptide did not bind to alphaVß3 and only weakly to alphaVß1. There is no sequence homology between this peptide and fibronectin (or any other known ECM protein), but still it competes with RGD for binding to the receptor. However, the interaction may be limited to the human integrin since the peptide was not recognized by mouse or hamster alpha5ß1 (52, 54).

Several different approaches have been taken in order to localize ligand binding sites in alpha5ß1. One such approached was the use of phage display screening of peptide libraries for binding to the FNIII-10 fragment. The peptide sequence WDDG/LWL (55), which is similar to a conserved region in RGD-binding ß subunits (135-140 in ß1; 124-129 in ß3), was found to be enriched in this screening. Furthermore, this sequence is located within the RGD/Me²⁺ binding 14 amino acid peptide of ß3 (see above, Fig. 3) which was identified by an independent method (55) (see further under alphaIIbß3). Cyclic WDDGWL peptide was found to bind directly to RGD peptides; this interaction was independent of cations (55), confirming the result with the D119A mutant of ß3_118-131. Both activating and inhibiting monoclonal antibodies have been mapped to a small region of ß1 (amino acids 207-218) close to the above suggested ligand binding site (56). A sharp and flexible turn is predicted to form within this region by the sequence TNKG (amino acids 206-209). The likely contribution of the alpha subunit to the ligand recognition has been investigated by systematic mutagenesis within an alpha5 segment corresponding to a region in alpha4 in which epitopes for several inhibiting antibodies against alpha4ß1 are located (Fig. 4). Ala-scanning of the third N-terminal repeat of alpha5 showed that mutations Y186A, F187A, and W188A within a predicted ß-turn inhibit ligand binding (57). The strongest effect was obtained by the mutation of F187 (corresponding to Y187 in alpha4, see below). It is yet not known if these mutations indicate the location of an important contact surface with fibronectin, or if they prevent ligand binding indirectly due to conformational alterations of the receptor.

Figure 4. Schematic illustration of the integrin subunit alpha4. The numbers above the bar refer to amino acids starting from the N-terminus of the mature protein. The squares numbered 1-7 represent repeating sequences of ~70 amino acids which are present in the N-terminus of integrin alpha subunits. The last three, or in some subunits four, of these repeats each contain a putative divalent cation binding motif. The horizontal black stripes represent the transmembrane part. The epitopes for antibodies which inhibit ligand binding have been mapped to the region marked with bold blue lines. Within this region, the amino acids shaded in pink have been demonstrated to be important for the ligand binding ability of the receptor. Analogous results were obtained for the corresponding residues in alpha5. The location of the cross-linking regions in alphaV for RGD peptides and in alphaIIb for fibrinogen gamma-chain 12 amino acid peptide, respectively, are indicated by the pink boxes above alpha4. The isolated red region of alphaIIb has been shown to bind directly to fibrinogen via the gamma-chain peptide. Antibodies which induce homotypic aggregation of lymphocytes bind to the N-terminal part (thin green lines), while epitopes for antibodies which can block this activity are located in the region marked with thin blue lines. An LDV sequence which may be the binding site for a homotypic receptor interaction is located within this region.

Integrin alphaIIbß3: Apart from platelets, in which alphaIIbß3 is the major cell surface protein (29, 58), this integrin has only been detected on megakaryocytes (59). Its most important function is to bind fibrinogen during thrombus formation. However, the recognition of fibronectin and several other RGD-containing proteins is probably also physiologically relevant for haemostasis (30).

Both the RGD site and the synergy site DRVPHSRNSIT contribute to the binding of fibronectin to alphaIIbß3 (60, 61), resembling the binding to alpha5ß1. However, the direct binding of the isolated synergy region can only be demonstrated for alphaIIbß3, even though the interaction is ~30-fold weaker compared to RGD peptides (60). Each one of the peptides can efficiently inhibit the binding of fibronectin (or fibrinogen) to isolated receptor or to platelets. Interestingly, the two fibronectin peptides cross-inhibit each other in binding to alphaIIbß3 (60), but still they appear to have separate binding sites (see below). A possible explanation for these findings would be that the integrin acts as a closing trap after triggering at either of the two ligand-binding sites. According to this model the unoccupied site would become closed as a result of the conformational rearrangements induced by binding of one of the peptides, while the corresponding event may result in cooperative binding at both sites when the intact fibronectin molecule is available.

Since short RGD peptides bind with higher affinity to alphaIIbß3 and alphaVß3 than to other integrins, it has been possible to use the peptides in cross-linking experiments to localize binding sites within these receptors. The RGD-containing peptides were found to cross-link mainly to amino acids 109-172 in ß3 of alphaIIbß3 (62), and to 61-203 in ß3 of alphaVß3 (63). Subsequently, the synthetic peptide ß3_118-131 was identified as an RGD-binding site (46). This region also contains the DXSXS motif which is involved in coordination of one Me²⁺(46). Based on molecular modeling and mutagenesis by Ala-scanning, two additional amino acids, D217 and E220, were suggested to contribute to the ion binding site. Mutation of either one of these five oxygenated amino acids results in inactivation of alphaIIbß3 (49). The importance of the region containing D217 and E220 was further demonstrated when overlapping synthetic peptides spanning the N-terminal 288 amino acids were tested for the ability to inhibit binding of fibronectin and fibrinogen to alphaIIbß3 (64).

The fact that alphaVß3 does not recognize the synergy site of fibronectin suggests that the a unit in alphaIIbß3 contains an important contact surface for the binding of that region. There is no direct data available to support this assumption, but the information obtained for the interaction of fibrinogen with alphaIIbß3 may be relevant for this issue. Fibrinogen binds to the receptor via RGD site(s) as well as by a 12 amino acid peptide from the C-terminus of the gamma chain (a non-RGD site). The gamma chain peptide was found to crosslink mainly to amino acids 294-314 in repeat 5 of alphaIIb (65). Further, the peptide alphaIIb_296-306 binds directly to fibrinogen and inhibits binding of fibrinogen to alphaIIbß3 (66). It remains to be tested if alphaIIb 296-306 is the binding site also for the fibronectin synergy peptide. The peptide represents the second metal ion binding site in alphaIIb, corresponding to the first in alpha4 (amino acids 278-297). It has been speculated that the second cation which is displaced from alphaIIbß3 upon ligand binding would be released from alphaIIb_296-306 (see above) (46).

Integrin alpha4ß1: The function of alpha4ß1 has been studied mainly in white blood cells, but it is expressed also by several types of adherent cells (67-69). It can mediate cell-cell contacts as well as cell-matrix contacts through the two known ligands, VCAM-1 and fibronectin, respectively (34). Both these interactions are important for the process of leucocyte extravasation, and thereby for a number of immunological and inflammatory events. Activating signals are required to induce strong binding of alpha4ß1 to VCAM-1 or fibronectin, but even in an unactivated state alpha4ß1 is able to mediate a weak binding to VCAM-1, by which lymphocytes can roll along activated endothelial cells (70, 71).

alpha4ß1 interacts primarily with the III₁₄-V region in fibronectin. Among the three known binding sites within this region, CS1 is predominant in the fibronectin splice variants where it is present (32). CS1 and CS5 cross-inhibit each other for binding to alpha4ß1 (35), but the affinity of recombinant fibronectin fragments for the receptor is higher when both of the sites are included (32), suggesting that they are not competing for the same binding site. The results thus resemble those obtained for the interactions of the RGD and synergy sites with alpha5ß1 and alphaIIbß3, although the combined effect of CS1 and CS5 is less dramatic and therefore said to be additive rather than synergistic. H1 was reported not to inhibit the binding of CS1 to alpha4ß1 (72), but higher concentrations of H1 have to be tested, as well as the reverse inhibition experiment, before a firm conclusion can be made. Obviously, determination of the structure of the III₁₄-V region at atomic resolution would greatly increase our understanding of how the multiple cell-binding sites may cooperate.

alpha4ß1 can be induced to recognize also the RGD region of fibronectin (73). The interaction is detectable only in the presence of integrin activating antibodies, hence its physiological relevance remains to be determined. Nevertheless, under these conditions the binding of large RGD-containing fibronectin fragments to the integrin can be inhibited by short peptides with the efficiency decreasing in the order CS1>GRGDSPC>H1 (73).

Monoclonal antibodies against alpha4 which inhibit binding to CS-1, as well as VCAM-1, map to amino acids 108-268 (the so called B epitopes) (74, 75). Ala-scanning mutagenesis within this region revealed that Y187A, W188A, and G190A inhibit binding of CS-1 and VCAM-1 (57). These amino acids are part of a predicted ß-turn between two ß-sheets in the third N-terminal repeat of alpha4 (76). This repeat lacks cation binding motifs. Mutations in the corresponding turn in alpha5 also inhibits binding to fibronectin (see above) (57).

Another activity of alpha4ß1 which is inducible by certain monoclonal antibodies against alpha4 is aggregation of lymphoid cells (77-79). These monoclonal antibodies have all been mapped to the N-terminal 100 amino acids (A epitopes) (74, 75). Antibodies capable of blocking the induced aggregation bind to amino acids 422-606 of alpha4 (C epitopes) (74, 75). Since there are no known ligand for alpha4ß1 on lymphocytes, the possibility of homotypic binding of this receptor to itself has been suggested. In support of this concept, alpha4ß1 on lymphocytes was shown to bind to an LDV-containing region of the isolated alpha4 chain (LDV488-490 located within the C epitope region) (80). Although the data are convincing, they raise several questions: In which physiological situations would homotypic alpha4ß1 interactions occur? Could the interactions reflect lateral interactions between integrins in the same cell (allowing high avidity binding to an unknown ligand) rather than binding between integrin molecules on different cells? Is homotypic binding of alpha4ß1 restricted to certain cells (i.e. lymphocytes), since expression of alpha4ß1 in K562 (erythroleukemia) or CHO cells did not induce aggregation of the cells (81). Is the ability of homotypic interaction unique for alpha4ß1 or shared by other integrin(s)?

Integrin alpha3ß1: The normal distribution of this integrin in vivo is limited mainly to epithelial cells in the skin, endothelial cells of the digestive tract, and kidney mesangial cells (82-85). In contrast, it is expressed by most cells in culture and by many tumors in vivo (86, 87). Laminin 5 (kalinin) appears to be an important ligand for the receptor (88, 89), while a weak binding to fibronectin and several other ligands has been observed only under certain conditions (82, 90-92). Further, alpha3ß1 is unable to mediate initial cell adhesion to any of these proteins except for laminin 5, indicating that it does not function as a fibronectin receptor in intact cells (93, 94). The issue is confused by the fact that alpha3ß1 localizes to focal contacts in cells seeded on fibronectin; however, this localization was shown to occur independently of which protein the cells adhered to (85). alpha3ß1 has therefore been suggested to serve as a secondary receptor with post adhesion functions (e.g. signaling for growth?) (85). It is unknown whether low affinity interactions with fibronectin or other matrix molecules would be of any importance for such functions.

Integrin alpha8ß1: This receptor is expressed mainly in epithelial cells, smooth muscle cells, myofibroblasts, and embryonic neural cells (95, 96). It has been shown to bind to RGD sites in fibronectin and vitronectin (97). In addition, tenascin-C has been reported to bind to alpha8ß1 via both RGD-dependent and RGD-independent interactions (97). alpha8ß1 is able to mediate cell adhesion and neurite outgrowth on these ligands (98, 99). Several other RGD-containing proteins, collagens, and CAMs were not recognized by alpha8ß1 (97).

Integrin alphaVß1: alphaVß1 is present on the surface of some cell lines, and appears to recognize fibronectin as its major ligand (100, 101). Although the subunits alphaV, ß1, and ß3 are expressed simultaneously by many cells, the combination alphaVß1 is seldom detected. This raises the question of how the formation of multiple integrins is regulated inside a cell. Is it just the relative amounts of the different subunits that determine which heterodimeric receptors are formed (simple competition on equal terms), or do the alpha and ß subunits have some preferred partners?

Integrin alphaVß3: This integrin is abundantly expressed by many cultured cells, but its distribution in vivo is restricted mainly to activated endothelial cells, osteoclasts, and tumor cells (36). It has been demonstrated to be important for angiogenesis and is therefore a potential target for inhibition of tumor growth (102-104). alphaVß3 was the first vitronectin receptor to be identified (105), but later it was found to bind also fibronectin and several other cell adhesion proteins (28, 106, 107). In all cases the RGD motif in the ligands is of critical importance for binding of alphaVß3. As discussed above, alphaVß3 does not require the synergy site of fibronectin for stable interaction (60, 108).

Integrin alphaVß6: alphaVß6 is expressed by epithelial cells during development and wound healing, and in many epithelial tumors (109). In this context it is interesting that the C-terminal 11 amino acids of the cytoplasmic part, which is unique to ß6, contains a proliferation promoting activity (110, 111). The cytoplasmic tail of ß6 also contains three regions required for localization of alphaVß6 to focal contacts (the membrane proximal region and two NPXY motifs) (110); the corresponding regions in ß1 and ß3 share the same function.

Fibronectin appears to be the main ligand for alphaVß6 (112), but a weaker interaction with tenascin-C has also been described (113). alphaVß6 binds fibronectin via the RGD site, and like alphaVß3, it does not require the synergy site in fibronectin (114).

Integrin alpha4ß7: This integrin is expressed on subsets of lymphocytes (115). Similar to alpha4ß1, it binds to the III₁₄-V region in fibronectin, VCAM, and the LDV site in the alpha4 subunit (116). In addition, it recognizes MAdCAM-1, an interaction which is important for homing of specific lymphocyte populations to mucosal sites (117).

Integrin alpha?ß8: The ß8 subunit is expressed mainly in brain, spinal cord, kidney, and embryonic muscle (118). The sequence of ß8 is rather different from other integrin subunits; for example, the cytoplasmic domain has no similarity to the other ß subunits (118). The alpha subunit(s) associated with ß8 unit has not yet been identified for the fibronectin receptor. alphaVß8 can form in ß8 transfected 293 cells, but it does not bind fibronectin (118, 119). Antibodies specific for the ß8 subunit co-immunoprecipitated one or more a components from sensory neurons (not alphaV) having unreduced Mr ~110-120kD in SDS-PAGE. The antibody could inhibit neurite outgrowth from these cells seeded on either fibronectin, laminin 1, or collagen IV (120). Since this is an unusual combination of ligands for integrins, it appears likely that ß8 can combine with several different alpha units, including alphaV.

3.5. Fibronectin polymerization

Besides mediating cell adhesion, all integrins can participate in one or more of the following cellular processes: migration, neurite outgrowth, organization of the cytoskeleton and focal contacts, and signaling which affects gene expression, proliferation and apoptosis. In addition, assembly of new extracellular matrix is an activity so far only described for fibronectin binding integrins. In contrast to several other ECM proteins (e.g. collagens, laminins, fibrinogen), fibronectin does not polymerize spontaneously in vitro by entropy driven processes. Instead, the reaction occurs on the cell surface and involves interaction of fibronectin with integrins and possibly with other cell surface components. Since major ECM components such as collagens, fibulins, and proteoglycans (121, 122) may be dependent on interactions with fibronectin for their organized deposition into the matrix, the polymerization reaction of soluble fibronectin protomers to insoluble networks is central for matrix formation. The general inability of malignantly transformed cells to promote fibronectin polymerization (123, 124) further highlights the importance of the process.

alpha5ß1 was the first integrin identified to be involved in fibronectin network formation. It localizes to the sites where fibronectin fibrils are in contact with the cell (125), and antibodies to alpha5 or ß1 can inhibit the polymerization in fibroblast cultures (126). Furthermore, overexpression of alpha5 in CHO cells increases the deposition of fibronectin in the matrix (124). However, mouse embryos deficient in the alpha5 gene still contained extracellular fibronectin networks (127), indicating that also other integrins could promote the polymerization process. At present, alphaVß3 (107) and alphaIIbß3 (128) have been shown to have this potential, although alphaVß3 was markedly less efficient than alpha5ß1, and alphaIIbß3 probably never serves this function in vivo. Expression of alphaVß1 (101) or alpha4ß1 (129) in CHO cells did not promote fibronectin polymerization, illustrating that mere binding of fibronectin to the cell surface by any integrin is not sufficient. A connection of the ß subunit to actin filaments is known to be required for fibronectin fibril formation on the cell surface (128, 130). Other factors which may influence whether an integrin has the fibril promoting activity include the recognition site preference in the fibronectin molecule (RGD, LDV, etc.) and the receptor-ligand binding affinity.

The polymerization process can be separated into two phases, initiation (nucleation) and extension, both of which are poorly understood. It has been reported in one study that the fibrils grow at only one end, indicating that they are polarized (131). Alternatively, the direction of fibril growth may reflect the migration of the cell. It is not known if new protomers are added to the fibril at the cell surface, or somewhere else along the fibril. The issue is further confused by the fact that the fibrils often appear to be anchored at both ends to cell surfaces, either on one cell or on two different cells (Fig. 5). The inability of monomeric fibronectin chains to become incorporated into the fibrils (132) may be relevant in this context.

Figure 5. Double immunofluorescent staining for fibronectin (green) and integrin subunit alpha5 (red) in a culture of GD25-ß1A cells (107), grown on coverslips for 5 days before fixing and staining. Cells are marked with asterisks (*). Note that most fibrils either seem to be stretched between two cells or are positioned at cell-cell borders, and that the ends of the fibrils colocalize with alpha5. The staining was performed as described (107).

Several regions in fibronectin are involved in one or the other phases of polymerization, through binding to cell surface components or to a neighbouring fibronectin molecule (Fig. 6). Binding of the RGD region to integrins may be the initial event which triggers the subsequent reactions (126), possibly by induction of an altered fibronectin conformation. The inability of recombinant fibronectin lacking the RGD motif to initiate fibril formation supports this idea (133). However, it is unclear whether fibronectin in its soluble folded conformation can bind to alpha5ß1. On hepatocytes alpha5ß1 was found to recognize the protein only after treatments which are known to unfold fibronectin (i.e. interaction with a specific collagen fragment or heparin, limited proteolytic cleavage) (17). Furthermore, fibronectin fragments rather than the intact protein are commonly chosen for affinity isolation of alpha5ß1 (51, 108, 134). It can be speculated that folding at the suggested hinge between III₉ and III₁₀ in soluble fibronectin (31) would prevent alpha5ß1 from reaching both the synergy site and the RGD loop. Thus, modulation of the fibronectin structure by some other interaction may preceed its binding to alpha5ß1.

Figure 6. Hypothetical model of the initial events in fibronectin polymerization. A soluble fibronectin molecule is bound to the surface of a cell via III₁₀ to fibronectin-binding integrins. The interaction induces a conformational change in the fibronectin molecule which becomes elongated. The cell bound fibronectin exposes binding sites (unidentified) for I_1-5 in another soluble fibronectin molecule. This interaction will "open up" the second fibronectin molecule, and so on. Pulling force from the actin filament system may be required for the postulated conformational changes. Domains in fibronectin which may be involved in interactions with other fibronectin domains are marked with black stripes.

The isolated N-terminal I_1-5 domain has been shown to bind cell layers at sites of fibronectin fibril formation, and to inhibit further incorporation of intact fibronectin molecules (135, 136). Deletion or disruption of this functional unit in recombinant fibronectin prevents incorporation of the protein into fibrils (132, 137). The N-terminal domain has been suggested to bind to a non-integrin "matrix assembly receptor" (135), but the nature of the receptor still remains unknown (138). Instead, accumulating evidence indicates that I1-5 binds either to conformationally altered fibronectin (139, 140) or to a site on alpha5ß1 which becomes exposed after binding RGD ligands (141). In either of the latter two models, I_1-5 would be involved in fibril extension rather than initiation. The postulated conformational changes could result directly from the interactions or from a pulling force of the actin filament system. The importance of the cytoskeleton has been demonstrated by use of cytochalasin B, and fibronectin fibrils running between two cells appear to be under tension as indicated by immunofluorescent staining (Fig. 5).

A cryptic site in III₁ can bind to I_1-5 (139, 140) and a similar interaction was described between unfolded III₁₀ and III₁ (142). At present it is not known which of these potential interactions mediate inter- and/or intramolecular bridging between fibronectin domains. The difficulties in studying these events is a major reason why the exact mechanism of the polymerization process still is unsolved.

3.6. in vivo functions of fibronectin and its integrins receptors

During embryogenesis, fibronectin appears before or at the onset of gastrulation in all vertebrates examined, and it is abundant at times and sites of cell migration: during gastrulation, neural crest cell migration and the migration of primordial germ cells (143). Alternative mRNA splicing is used during development as a mechanism to create different forms of fibronectin within the extracellular matrix by inclusion or exclusion of the III_A, III_B, and V segments (144-146). The expression of integrins has also been demonstrated to be developmentally regulated, where some of the fibronectin receptor subunits are continuously expressed, while others are not (147). Injection of antibodies to fibronectin or RGD-containing peptides inhibits gastrulation of several species, indicating that the interactions between fibronectin and integrins are important during that particular stage of development (148-151). To investigate the role of fibronectin and each of its receptors, the genes of both fibronectin and several integrin subunits have been knocked out by homologous recombination (37, 38, 152). Up to now, disruption of the genes for five alpha-subunits (alpha3, alpha4, alpha5, alpha8 and alphaV) (127, 153, 154) and three ß subunits (ß1, ß6 and ß7) (155-159) of fibronectin binding integrins has been reported. Since several of the knock-outs are lethal at embryonic stages, the generation of chimeric mice have been necessary in order to investigate the effects on later stages of development and in adult animals. To generate chimeric animals, embryonic stem (ES) cells in which both alleles of the gene of interest have been disrupted are injected into a normal blastocyst. The resulting mouse will have a mixture of cells which are normal and which are homozygous for the disrupted gene. This enables the investigation of the contribution of the null-cells in different organs and in different processes. Further information about the function of the disrupted gene product can be obtained by in vitro studies of the ES cells. For example, the differentiation process of the mutated ES cells into specific cell types can conveniently be followed and manipulated in cell cultures.

Fibronectin: Mouse embryos lacking fibronectin (fibronectin-null) die at embryonic day 8.5, and they have defects in the development of mesoderm, neural tube and blood vessels (152, 160). Although fibronectin is expressed in normal preimplantation blastocysts, the fibronectin-null blastocysts hatch and implant into the uterine wall. They initiate gastrulation and form mesoderm in the complete absence of embryo derived fibronectin. Nevertheless, the presence of fibronectin at these early stages is believed to be important, since the oocyte alone is probably contributing with enough maternal fibronectin for these processes to occur. From embryonic day 8.0 and onwards the mutant embryos develop deformities and deteriorate during day 10-11 of gestation. Fibronectin-null embryos show shortened anterior-posterior axes, fail to develop certain mesodermally derived structures like notochord and somites, and develop abnormal heart and blood vessels, all probably a result of a deficit in the mesodermal layer. The lack of notochord and somites has later been shown to be a result of fibronectin being critical for the organization or maintenance of the notochord precursor cells and for the condensation of precursor cells into somites (160). Neural folds are formed in the absence of fibronectin, while the neural tube becomes kinked. Primitive red blood cells do develop in the fibronectin-null embryos, while blood vessels do not, strongly suggesting a role for fibronectin in vasculogenesis but not in hematopoiesis.

Since there are several receptors for fibronectin, the total effect of the fibronectin-null mutation is likely to be made up of separate effects due to lack of binding between fibronectin and its individual receptors. Therefore, studies on knock-outs of each of the fibronectin receptor subunits makes it easier to determine where the different receptors are active, what functions they have, and what their roles are during embryonic development.

ß1: The ß1 subunit can dimerize with at least 10 alpha subunits, of which five have been shown to bind fibronectin (alpha3ß1, alpha4ß1, alpha5ß1, alpha8ß1, and alphaVß1). The homozygous ß1-null embryos develop normally to the blastocyst stage, implant, but die at embryonic day 5.5 (155, 156, 158). The ß1-null throphoblast cells are able to invade the uterine stroma and survive longer than the inner cell mass, suggesting that the inner cell mass requires ß1 integrins for survival while the trophoblasts do not. Analysis of the chimeric embryos demonstrated the presence of ß1 integrin-deficient cells in all germ layers, indicating that ß1-null cells can differentiate and migrate inthe environment provided by most normal tissue. For example, chimeric mice show patches of myocytes that are ß1-null in cardiac muscle, and myotubes show some ß1-null contribution as a result of fusion between wild-type and mutant myoblasts into mixed myotubes. However, lack of the ß1 subunit markedly retards the formation of cardiac and skeletal muscles (161, 162) when the differentiation from ES cells was studied in vitro. In conflict with previous reports (68, 163), studies using in vitro and in vivo differentiated myoblasts have proven that the ß1 integrin is not necessary for sarcomer or myotube formation (162), events which represent late stages of the differentiation process. Migration of neuronal cells as well as neural crest cells can occur in the absence of ß1 integrins. In contrast, the chimeric mice lack ß1-null cells in blood and in haematopoietic organs such as spleen, thymus and bone marrow as a consequence of the inability of ß1-null cells to invade the fetal liver (164). The ß1-null embryos die before the fibronectin-null embryos. The reason for this can be that ß1 containing integrins that bind ligands other than fibronectin are important at earlier stages, before the ß1/fibronectin interactions take place. Alternatively, it can be due to the different life-times of the ß1 and the fibronectin maternal mRNAs provided by the oocyte.

alpha5: The alpha5 subunit is found only in combination with ß1. The knockout of the alpha5 subunit resulted in nearly the same defects as the fibronectin-null mutation did (127), suggesting that the majority of the defects seen in the fibronectin-null embryos probably are due to the lack of alpha5ß1/fibronectin binding. The alpha5 mutant embryos start to show defects by embryonic day 8.5 and die around day 10-11. The alpha5-null embryos have defects in the posterior trunk and yolk sac mesodermal structures, suggesting a role for alpha5ß1 integrin in mesoderm formation, movement or function. The morphogenesis in alpha5-null embryos seems to extend further compared to the fibronectin-null embryos, including formation of notochord, somites and heart, considerable development of the brain, optic and otic anlagen and formation of branchial arches. The anterior part of the alpha5-null embryo develops relatively normally up to about the tenth somite. However, the posterior somites are absent, the paraxial mesoderm are defective, and the neural tube becomes kinked as it does in the fibronectin-null embryos. The alpha5ß1 defective embryo develops a heart but shows defects in vascular development, although again the defects due to the absence of alpha5ß1 are less severe and more limited to the posterior part than those due to the absence of fibronectin. It seems that alpha5ß1 is required for proper formation and maintenance of blood vessels, while other fibronectin receptors are involved in initial steps of vasculogenesis (127).

alpha4: The alpha4 integrins (alpha4ß1 and alpha4ß7) are both capable of binding fibronectin as well as VCAM-1. The alpha4-null embryos die at embryonic day 11 due to the lack of a functional placenta (154). The allantois fail to fuse with the chorion during placentation and the development of the epicardium and the coronary vessels are impaired leading to cardiac haemorrhage. However, the alpha4 subunit seems not to be essential for the formation of either cardiac or skeletal muscle (154, 165). Studies of the VCAM-1 knockout suggest that these alpha4-null phenotypes are a result of abolished alpha4/VCAM-1 interactions rather than alpha4/fibronectin interactions (166, 167). Chimeric mice were also used to study blood cell development (168). During fetal life, T-cell development is alpha4 independent, but after birth further production of T-cells becomes alpha4 dependent. Precursors for both T- and B-cells require alpha4 integrins for normal development within the bone marrow, while monocytes and natural killer cells can develop normally without alpha4 integrins.

ß7: The ß7 subunit can dimerize with alpha4 and the integrin alpha4ß7 is expressed on a subset of lymphocytes together with alpha4ß1. It has been shown that lack of ß7 leads to an impaired formation of the gut-associated lymphoid tissues (159). This is probably a result of inability of the lymphocytes to adhere to the endothelium and subsequently to populate the gut-associated lymphoid tissues. Whether this is a result of disrupted interactions of alpha4ß7 with VCAM-1, MAdCAM, or fibronectin is not known.

alpha3: The alpha3-subunit can dimerize with the ß1 subunit, and alpha3ß1 is a receptor for laminin 5 and possibly for fibronectin. The homozygous alpha3 mutant mice die during the first day after birth (153). The alpha3 mutants have several kidney and lung abnormalities, consistent with the epithelial expression in these organs of normal mice. The tubules in the medullary region of the kidney are dilated, and the bronchial airways of the mutant lung are wider than normal. The mutant kidney cells have a greatly decreased number of foot processes along the glomerular basement membrane, and the basement membrane is wider and more disorganized when compared with the normal kidney. Since a recent report indicates that alpha3ß1 does not really bind fibronectin but that it rather has a secondary function (85) that may be important for fibronectin assembly (169), further investigation is needed to determine if the defects in the alpha3-knockout is a consequence of lack of binding to its major ligand laminin 5, rather than a fibronectin related effect.

alpha8: The alpha8ß1 integrin binds fibronectin and is expressed in adult smooth muscle and epithelia, and in mesenchymal and neural cells during development. The alpha8 gene has been knocked out, but the original results have not been published yet. Apparently, lack of alpha8 results in defect development of the kidney, indicating a role for alpha8ß1 in kidney morphogenesis (37, 38).

alphaV: The alphaV subunit can dimerize with five ß subunits, three of which are able to bind fibronectin (alphaVß1, alphaVß3, and alphaVß6). The alphaV subunit has been knocked out, but the original data has not yet been published. Apparently, the mouse die on the first day after birth, probably as a cause of vascular haemorrhage (37, 38). The alphaV containing integrins have a wide range of ligands in addition to fibronectin, and this complicates the interpretation of the results with regard to fibronectin and its functions.

ß3: The ß3 subunit has, to our knowledge, not been knocked out. However, the lack of, or mutations in, ß3 lead to bleeding disorders in man (Glanzmann's disease) due to the impaired ability of the platelets to aggregate at injury sites. Surprisingly few other abnormalities are observed in these patients.

alphaIIb: The alphaIIb subunit has not been reported knocked out (however, see ß3).

ß6: The integrin alphaVß6 is the only integrin expressed exclusively in epithelial cells and has been found to bind fibronectin and tenascin-C (112, 113). The expression of alphaVß6 is rapidly induced following epithelial injury while only barely detectable on normal adult epithelia. During embryogenesis, alphaVß6 is expressed in lung, skin, and kidney. The ß6-null mice have juvenile baldness associated with degenerated hair follicles and infiltration of macrophages into the skin, and accumulation of activated and cytokine producing B- and T-cells around conducting airways in the lungs (157). The lungs show enhanced sensitivity to acetylcholine, a hallmark feature of asthma, suggesting that alphaVß6 participates in the modulation of epithelial inflammation. On the other hand, the ß6-null mice were capable of healing cutaneous wounds, which had been suggested to be alphaVß6 dependent. Further studies are needed to determine whether it is fibronectin or tenascin-C that is involved in these conditions, but the lack of the corresponding abnormalities after disruption of the tenascin-C gene strongly indicate that the interaction between alphaVß6 and tenascin-C is of minor importance (170).

Taken together, these knock-out experiments of both fibronectin and its integrin receptors clearly demonstrate that these proteins are of vital importance for the organism. They also show that the different integrins have distinct functions and distribution, and that they in most cases do not compensate for each other. Nevertheless, alphaV integrins have been found to be able to compensate for alpha5ß1 integrin function with regard to adhesion to fibronectin and fibronectin matrix assembly in vitro (107, 165). The other important aspect of these knock-out investigations is that they complement the earlier antibody or peptide inhibition studies, by providing new and often contradictory conclusions of the in vivo functions of fibronectin and integrins. They also show that a protein can be present while not necessarily have a critical function at that particular location at that particular time. Further studies using tissue specific knock-outs, knock-ins, and disruption of one splice variant at the time, will hopefully give even more detailed and interpretable information about the in vivo roles of fibronectin and its integrin receptors.

3.7. Perspectives

Integrin alpha5ß1 was isolated in 1985 as the first fibronectin receptor (51) and its cDNA sequence was reported two years later (171). Since then an impressive amount of information on the structure, interactions, and functions of fibronectin and its receptors has accumulated. As convincingly demonstrated by gene knock-out technology, these proteins are of fundamental importance in various physiological situations. Therefore, they will continue to be subject for intense research, and the progress rate will probably be faster than we can foresee. Some of the more immediate goals to reach include information on the three dimensional structures of whole integrins and fibronectin, and understanding of how conformational changes in these proteins are regulated. One area where such information would be of practical use is in development of reagents which could modulate receptor function during treatment of various adhesion-related disorders. These reagents could potentially be designed to act directly on the ligand binding sites, to affect conformation, or to modulate interactions with integrin associated proteins inside or outside of the cells.