The Role of the Fibronectin IGD Motif in Stimulating Fibroblast Migration*

The motogenic activity of migration-stimulating factor, a truncated isoform of fibronectin (FN), has been attributed to the IGD motifs present in its FN type 1 modules. The structure-function relationship of various recombinant IGD-containing FN fragments is now investigated. Their structure is assessed by solution state NMR and their motogenic ability tested on fibroblasts. Even conservative mutations in the IGD motif are inactive or have severely reduced potency, while the structure remains essentially the same. A fragment with two IGD motifs is 100 times more active than a fragment with one and up to 106 times more than synthetic tetrapeptides. The wide range of potency in different contexts is discussed in terms of cryptic FN sites and cooperativity. These results give new insight into the stimulation of fibroblast migration by IGD motifs in FN.

Fibronectin (FN) 2 is a multifunctional, multidomain adhesive glycoprotein that plays a prominent role in wound healing, embryogenesis, and hemostasis (1). It is found both in the extracellular matrix and in soluble form in blood plasma. The structure and diverse binding properties of FN have been much studied (2,3). We have previously described a novel truncated form of FN (70 kDa) that was able to act as a potent migration-stimulating factor (MSF) (4,5). MSF was originally identified in fetal and cancer patient fibroblasts (4). In vitro, exposure of human dermal fibroblasts to MSF provokes a change in phenotype, causing cells to migrate into three-dimensional gels of native type I collagen (5). Fulllength FN is, however, devoid of MSF-like activity, suggesting that cryptic sites are exposed in MSF (5) as well as in certain N-terminal FN fragments (6).
The N-terminal fragment of FN is composed of independently folded modules, the majority of which are classified as type I (Fn1), each consisting of ϳ45 amino acids. Fn1 modules have five short ␤-strands that form two sheets and a small hydrophobic core; they are further stabilized by two disulfide bonds in a 1-3, 2-4 configuration (7). These ␤-strands form two anti-parallel ␤-sheets that are linked together by a short loop of three amino acids. Four of the nine Fn1 modules contain a highly conserved Ile-Gly-Asp (IGD) sequence (see Fig. 1) within this loop. The motogenic activity of MSF on fibroblasts has been attributed to this loop since mutation of IGD to DGI in modules 7 and 9 abolishes MSF bioactivity (5). Furthermore, soluble synthetic tri-and tetrapeptides containing the IGD amino acid motif can mimic MSF properties (8). Like MSF, these peptides stimulate the migration of human dermal fibroblasts into three-dimensional type I collagen gels, although with much reduced potency; half maximal activity is observed with femtomolar and micromolar concentrations for MSF and IGD peptides, respectively (8). The addition of Ser or Gln C-terminal to the IGD tripeptide was shown to enhance MSF-like activity by a factor of 10 (8).
The demonstration of bioactivity in a tripeptide is reminiscent of several other activation motifs. The Arg-Gly-Asp (RGD) cell adhesion motif in the tenth Fn3 module was the first tripeptide to be isolated and has been shown to interact with integrin receptors (9,10). Synthetic peptides based on Leu-Asp-Val (LDV) and Arg-Glu-Asp-Val (REDV), found in the IIICS alternatively spliced site in FN, can mimic the activity of intact FN at least partially (11,12). Although the RGD sequence has been shown to bind integrins and promote cell adhesion and fibril formation, the exact binding partner for the IGD motif is unknown, but its ability to promote cell migration, which is abolished by antibody to ␣v␤3 (8), suggests a possible interaction with integrins.
Full-length fibronectin can form fibrils, and although much is known about promotion of fibrillogenesis by cells, little is known about the precise mechanism. Most models implicate integrin binding via the RGD recognition sequence. A recent report surprisingly showed that the absence of a functional RGD motif in FN did not compromise the assembly of FN fibrils in mutant embryos or cells. A possible explanation suggested was that FN contains a novel integrin binding motif in its N-terminal region (13). The importance of the N-terminal 70-kDa fragment was also emphasized recently when it was shown to initiate FN fibril formation in fibronectin-null mouse fibroblasts (14). These reports support the idea that fibrillogenesis might be initiated through integrin cell attachment at the N terminus via a binding site other than the RGD region.
Here we set out to explore the structure-function relationships of the IGD sequence in the context of intact Fn1 domains. Of particular interest are the third and fourth IGD motifs in MSF, one in the seventh Fn1 (in this report these modules will be described using a superscript nomenclature: 7 Fn1, etc.), and one in 9 Fn1. Here we examine the role of the IGD motif by studying the properties of a series of fragments, namely the 8 Fn1-9 Fn1 and 7 Fn1-8 Fn1 module pairs, each containing one IGD motif, and the 7 Fn1-8 Fn1-9 Fn1 fragment that contains two IGD motifs. The effects of site-specific mutations on migration and structure are examined using fibroblast migration assays and solution state NMR. Both IGD motifs are shown to have significant migration-stimulating activity while contained within fully folded fibronectin modules.
Transformation by electroporation was performed according to standard Pichia protocols (Invitrogen) and expression of unlabeled and uniformly 15 N-labeled protein was performed in a 1-liter fermenter (Electrolab Ltd., Tewksbury, UK) in an anal-ogous fashion to that described previously for the 4 F1-5 F1 module pair (16). The protein was initially passed through a SP-Sepharose cation-exchange chromatography column at pH 3.0 to partially purify and reduce the volume of the secreted protein. High mannose sugars were trimmed back to single N-linked N-acetylglucosamine (GlcNAc) residues with Endo H f at pH 5.5. High performance liquid chromatography (RP-HPLC) on a C4 column with a gradient of 24 -38% acetonitrile and 1% trifluoroacetic acid was used to achieve homogeneity. The purity of the protein at each stage was assessed with SDS-PAGE, and the identity and final purity were confirmed by electrospray ionization mass spectroscopy (ESI-MS).
Cells-Collagen gel migration experiments were performed with human skin fibroblasts FSF44. Stock cultures were maintained in Eagle's minimal essential medium (MEM) containing 15% donor calf serum, as previously described (17).
Migration Assays-The collagen gel assay was performed in 30-mm plastic culture dishes containing preformed 2 ml of type I collagen gel, as previously described (5). 1 ml of serum-free MEM containing four times the desired final concentration of effector molecule and 1 ml of trypsinized fibroblasts (density 2 ϫ 10 5 cells/ml) suspended in MEM containing 4% donor calf serum were plated in duplicate onto the collagen gel to give a final volume of 4 ml in 1% donor calf serum. After a 4-day incubation period at 37°C, cell migration into the three-dimensional gel in response to the effector was determined as previously described (17). In short, this involved using an inverted phase microscope to count the cells on the surface of the gel and then focusing down through the gel to count the cells within the gel. This operation was repeated with 15 randomly selected areas on the gel surface.

RESULTS
The 8 Fn1*-9 Fn1 (R503K) Protein Is Heterogeneously Glycosylated-Our earlier study concluded that the wild type 8 Fn1-9 Fn1 was proteolytically cleaved by the P. pastoris endoprotease KEX2 between residues Arg-503 and His-504. The R503K mutation was therefore introduced to alleviate this proteolysis problem (15). This 8 Fn1*-9 Fn1 module pair expressed in Pichia was found to be heterogeneously glycosylated at two sites; at least one of the sites was glycosylated and the different fractions could be separated by increasing the retention time on a C4 RP-HPLC column. We previously concluded that glycosylation of Asn-511 was critical for optimal binding to gelatin whereas glycosylation of Asn-497 had no significant effect on binding activity (15). To increase the yield of homogeneous protein and to simplify the purification procedure, Asn-497 was mutated to remove the "non-essential" glycosylation site.
The 7 Fn1-8 Fn1**, 8 Fn1**-9 Fn1 and 7 Fn1-8 Fn1**-9 Fn1 Mutants (R503K,N497Q) Are Singly Glycosylated-The mutation N497Q was introduced into all constructs used in this study (with the exception of 8 Fn1*-9 Fn1, which was used as a  DECEMBER 7, 2007 • VOLUME 282 • NUMBER 49 control) as it was judged to be a suitable conservative mutation to remove the Asn-497 glycosylation site. Approximately 80 mg/liter of 8 Fn1**-9 Fn1 was expressed, as judged by SDS-PAGE of crude fermentation supernatant under non-reducing conditions (data not shown). This initial yield was comparable with that of the wild type and single point mutant (R503K) protein but the purification yield was greatly enhanced as the target protein was homogenously glycosylated. SDS-PAGE showed that the molecular mass was reduced from ϳ16 to ϳ11 kDa after treatment with Endo H f, indicating the removal of a single high mannose sugar chain. N-terminal sequencing gave a single sequence DQCIVDDIY, and ESI-MS confirmed the expected molecular mass of 11,075 Da, indicating the presence of the complete module pair with a single GlcNAc sugar residue.

IGD Motifs in Fibronectin and Cell Migration
The 8 Fn1**-9 Fn1 module pair was analyzed by NMR, both to compare the fold of this protein to that of the 8 Fn1*-9 Fn1 single point mutant and to study the sugar attachment site. The module pairs were isotopically labeled by replacing the sole nitrogen source in the fermenter vessel with ( 15 NH 4 ) 2 SO 4 . Both module pairs were expressed, purified, and characterized by ESI-MS, which gave an increased mass of 131 Da, corresponding to label incorporation of 98.5%. The cross-peaks of the 8 Fn1**-9 Fn1-GlcNAc511 and 8 Fn1*-9 Fn1-GlcNAc511 [ 1 H-15 N]-HSQC spectra (a single N-linked GlcNAc sugar residue on Asn-511) were shown to overlay as expected, with the exception of mutated residue Asn-497 (data not shown). Direct comparison of the 8 Fn1**-9 Fn1 spectrum and the 8 Fn1*-9 Fn1 spectrum allowed assignment of the 8 Fn1**-9 Fn1 cross-peaks. After this comparison, a C-terminal His tag was engineered onto the 8 Fn1**-9 Fn1 to aid identification of the module pair by Western blot. This addition did not significantly affect the NMR spectra other than by obvious addition of new peaks from mobile residues.
The 8 Fn1**-9 Fn1 Module Pair Contains Typical Fn1 Folds and the IGD Motif Is Located in a Rigid Loop between Strand B and Strand C-The solution structures of several Fn1-type module pairs have already been calculated by NMR (18,19). The secondary structure of the 8 Fn1**-9 Fn1 was mapped here using inter-strand NOEs. Both modules were found to have a typical type I module fold ( Fig. 2A), and the observed hydrogen bonding pattern agrees well with existing structures. The position of the IGD motif in the 9 Fn1 can be mapped to a tight turn between the B strand and the C strand, and a model based on the structure of the 2 Fn1-3 Fn1 pair confirms that the loop is accessible to the solvent (Fig. 2B). The heteronuclear { 1 H-} 15 N NOE was used to map the internal motion of residues on a subnanosecond time scale, showing the module pair to be relatively rigid, especially in the short linker between modules and in the IGD loop of the 9 Fn1 (Fig. 2C). Those values lower than 0.65 indicate regions with fast internal mobility, the majority falling within the loop regions of the protein.
Expression, Purification, and Characterization of 8 Fn1**-9 Fn1 IGD Mutants-Conservative and non-conservative point mutations were introduced into 8 Fn1**-9 Fn1 to interrupt the FIGURE 2. NMR structural analysis. A, secondary structure of 8 Fn1**-9 Fn1 module pair. Each module adopts a typical Fn1-type fold, with a two-stranded antiparallel ␤-sheet folded over a three-stranded ␤-sheet. Interstrand NOEs between protons are indicated by bold lines within the ␤-sheets. The three-dimensional NOESY strips (inset) show the interstrand N H -H ␣ and N H -N H NOEs between four of the residues in the 8 F1 module. The position of the IGD loop is highlighted, and the mutated residues N497Q and R503K are starred. B, a three-dimensional model of the 8 Fn1**-9 Fn1 module pair based on the secondary structure data in Fig. 1 and the x-ray structure of the 2 Fn1-3 Fn1 module pair (19). A short linker joins the two modules, and the IGD motif falls with a tightly constrained loop in the 9 F1. The side chains of the tightly constrained IGD loop are highlighted (pyMOL). IGD site in the 9 Fn1 module. These constructs were labeled, expressed, and purified; their fold was examined by comparing each HSQC spectrum with that of the unmodified 8 Fn1**-9 Fn1 module pair. All seven mutants were equally soluble and were folded in the same way as shown by the limited 1 H N and 15 N chemical shift differences from the "wild-type" 8 Fn1**-9 Fn1 (supplemental Figs. S1 and S2). As expected, there were minor shifts of peaks from residues in the IGD and of residue Ser-544 which immediately follows. Other observed localized shifts were in the cross-strand residue Cys-530 (in strand A of 9 F1) and Cys-558 (in strand D of 9 F1); these side chains are predicted by structural homology models to point toward the IGD loop (data not shown). Apart from these minor local perturbations around the mutated residue, there was no evidence for a change in the overall fold.
An Intact IGD Motif Is Required in 8 Fn1**-9 Fn1 for Stimulation of Fibroblast Migration-The effect of each module pair on the behavior of fibroblast cells was assessed in a collagen gel migration assay. Increasing concentrations of the module pairs were added to cells plated on the surface of three-dimensional gels of native type I collagen fibers. In a similar manner to that of MSF and IGD peptides (8), 8 Fn1**-9 Fn1 with an intact IGD motif stimulated fibroblast migration into this matrix, in a bellshaped dose-response fashion. Maximal bioactivity was expressed at concentrations of 100 pmol (Fig. 3A). As predicted from peptide studies (8), the module pairs with a mutation at Ile-541 (VGD, AGD, RGD) did not affect cell migration (Fig.  3A). The DGI mutant was also unable to stimulate migration (Fig. 3B). In contrast, those module pairs carrying a mutation at Asp-543 (IGE and IGA) showed some bioactivity, but this was significantly reduced with respect to the intact IGD at the same protein concentrations (Fig. 3B). It has also been suggested that the fourth amino acid in the IGD loop could affect migration (8). The amino acid immediately following the IGD was also mutated from Ser to Gln in 8 Fn1**-9 Fn1 but did not change the extent of cell migration (Fig. 3B).
The 8 Fn1**-9 Fn1 pair (R503K,N497Q) was shown to be as active as 8 Fn1*-9 Fn1 (R503K), indicating that the removal of the glycosylation site does not affect the bioactivity of the protein (Fig. 3B). 8 Fn1**-9 Fn1 (His) was also shown to exhibit the same bioactivity as the module pair without the His tag (Fig. 3B).
The Relative Roles of IGD in 7 Fn1-8 Fn1** and 7 Fn1-8 Fn1**-9 Fn1-The 7 Fn1-8 Fn1** and 7 Fn1-8 Fn1**-9 Fn1 fragments were judged by SDS-PAGE to express with yields of ϳ80 and 40 mg/liter, respectively. The lower yield of the module triple is presumably due to some misfolding, and therefore degradation, of this larger recombinant fragment. The identity of the proteins was confirmed by N-terminal sequencing and ESI-MS (expected molecular masses of 12,061 and 18,393 Da). The recombinant material was isotopically labeled with 15 N and examined by NMR. The dispersion of cross-peaks in [ 1 H-15 N]-HSQC spectra was indicative of a fully folded protein.
Following the results from the 8 Fn1**-9 Fn1 mutations the relative roles of the IGD motifs in 7 F1 and 9 F1 were explored by replacing IGD with RGD in 7 Fn1-8 Fn1** and 7 Fn1-8 Fn1**-9 Fn1. The proteins were successfully expressed, purified to homogeneity, and characterized by Western blotting and ESI-MS. The two IGD motifs in 7 Fn1-8 Fn1**-9 Fn1 were mutated in turn by introducing single point mutations I449R and I541R before introducing both mutations to create a double mutant. The structure of the mutants, before and after the Ile to Arg mutation, was judged not to have changed significantly by comparing resolved up-field methyl peaks in one-dimensional NMR spectra (data not shown).

DISCUSSION
In this study we have been able to extend the earlier work on the bioactivity of soluble synthetic IGD peptides (8) to that of IGD motifs in the context of folded Fn1-type modules. Various Fn1 module fragments, including several IGD mutants, were produced in P. pastoris, and their ability to stimulate fibroblast migration was assessed in a collagen gel migration assay. All of the proteins produced had a similar fold, as judged by NMR. A 10,000-fold increase in potency was observed in moving from peptides to recombinant module pairs.
The conservative I541V mutation was designed to mimic the VGD of the first Fn1-type module; it was unable to promote cell migration in the context of 8 Fn1**-9 Fn1. The I541R mutation was designed to introduce an RGD sequence, a motif that promotes cell adhesion when in other regions of fibronectin, e.g. 10 Fn3 (9). This mutation did not, however, promote cell migration here; neither did I541A. These results suggest that the isoleucine is critical for the observed bioactivity.
The bioactivity of 7 Fn1-8 Fn1** is very similar to that of 8 Fn1**-9 Fn1 (Figs. 3A and 4). Substitution of I449R in the 7 Fn1-8 Fn1** module pair also resulted in a protein devoid of migration-stimulating activity. About half the bioactivity remained in the protein with the conservative D543E mutation. Some activity also remained for the D543A mutation but it was significantly lower than the wild type 8 Fn1**-9 Fn1. These results suggest that the aspartate residue in the IGD is required for optimal bioactivity but is less critical than the isoleucine. The "reverse" mutant DGI was also devoid of motogenic activity.
It has been suggested that the IGDS of 9 Fn1 is more active than the IGDQ in 7 Fn1 (8), but the mutation S544Q did not change the bioactivity of the protein. This finding suggests that the fourth amino acid is less important in the context of the 8 Fn1**-9 Fn1 module pair than in the unconstrained peptide form.
The increased potency of the protein in comparison to the peptide is remarkable. Constraining the ends of the IGD motif in the 9 Fn1 module may favor the orientation of ligand binding that accounts for migration stimulation at much lower concentrations. A cyclic IGD peptide containing 20 amino acids from around the IGD site is also more potent than a similar length unconstrained IGD peptide (data not shown). Cyclic RGD peptides can inhibit cell attachment at a 20-fold lower concentration than linear peptides (20). The insertion of the RGD motif into mutant lysozyme significantly increased the cell adhesion activity (21), again implying that the introduction of conformational constraints can increase affinity to the integrin receptor, although a constrained RGD sequence in the IGD loop of Fn1 modules does not enhance cell migration. In the current study, the IGD motif as presented by the 9 Fn module structure has markedly increased the bioactivity compared with free peptides.
The heteronuclear { 1 H-} 15 N NOE data on 8 Fn1*-9 Fn1 here, as well as previous dynamics studies of 2 Fn1-3 Fn1 and 4 Fn1-5 Fn1 (18), show that the IGD motif is in a relatively immobile part of the module. This inflexibility in the IGD loop is quite different from the considerable flexibility of the RGD loop observed in 10 F3 (22).
We also explored possible cooperativity by studying the effect of two IGD motifs in one protein using a triple module construct ( 7 Fn1-8 Fn1**-9 Fn1) containing an IGD motif in both 7 Fn1 and 9 Fn1. The maximal activity of two IGD motifs occurred at significantly lower concentration than a single IGD (the concentration dependence of the MSF activity of different peptides and proteins is summarized in Table 1). It is interesting to note that the gelatin-binding domain, which contains the same two IGD motifs, is some 1000-fold more potent than the triple module (6). Migration-stimulating activity was only lost from the triple module protein after mutation of both of the IGD sites, confirming that they are both able to stimulate migration independently as well as by acting together.
The profile of cell migration at different concentrations takes a bell shape, suggesting that a threshold concentration is required to initiate cell migration and that a negative feedback mechanism occurs at higher concentrations. This is somewhat similar to a previous observation (23) that a 21-kDa fragment of FN stimulated the migration of fibroblasts in a bell-shape doseresponse pattern. We have previously shown (8) that fibroblast migration stimulated by IGD peptides is dependent on both ␣v␤3 functionality and tyrosine phosphorylation of focal adhesion kinase. The reduced migration at high concentration may thus arise from effects on integrin clustering and/or subsequent downstream signaling.
Although integrin ␣v␤3 cannot be confirmed as the prime candidate receptor for the IGD motif, it is clear that the presence of ␣v␤3 and its downstream signaling components (phosphatidylinositol 3 kinase and focal adhesion kinase) are required for IGD motogenic activity (8). ␣v␤3 integrin also appears to cooperate with activated growth factor receptors, such as epidermal growth factor receptor, in mediating their respective motogenic signaling in response to ligand binding (24). The RGD loop has been established as a key recognition sequence for ␣v␤3 (10), but changing IGD to RGD in Fn1 domains results in loss, not gain, of motogenic activity. This apparent discrepancy may be due to the inherent inflexibility of the RGD in this context.
A recent report suggested that Asn can spontaneously deamidate to isoD to create a DGR sequence (25) that can interact with ␣v␤3. There is a potential isoDGR in 7 Fn1 that could constitute a binding ligand for the ␣v␤3. However, because the 8 Fn1**-9 Fn1 lacks the NGR sequence that converts to isoDGA, and 7 Fn1-8 Fn1** and 8 Fn1**-9 Fn1 both exhibit the same potency in our assay, it is unlikely that this receptor activation mechanism is part of the IGD response. Of course, integrins could play a critical mechanistic role in mediating motogenic signaling by IGD ligation to some unknown cell surface receptor. The four IGD motifs at the N terminus of FN are invariant in human, rat, and mouse (see Fig. 1). Three are also invariant in frogs, with a conservative mutation in the fourth. From the current work and knowledge of Fn1 module structure it seems likely that all these IGD motifs will display bioactivity as long as the Fn1 modules are exposed. Only two of the four IGD motifs of MSF ( 7 Fn1 and 9 Fn1) appear to be active regarding stimulation of fibroblast migration (5). Consistent with this conclusion, a proteolytically derived gelatin-binding domain of FN (Sigma) that contains both 7 Fn1 and 9 Fn1 displays motogenic activity, whereas the proteolytically derived 1 Fn1-5 Fn1 (Sigma) was inactive (5). This suggests that 3 Fn1 and 5 Fn1 may not be exposed in MSF or 1 Fn1-5 Fn1. Even if all four IGD motifs were exposed and bioactive in MSF, a simple association between exposed IGD motifs and motogenic activity would not explain the huge differences in potency between MSF and IGD-containing peptides ( Table 1). MSF (with four IGDs in four Fn1 modules) has been shown to stimulate fibroblast migration at 100-fold lower concentration than the gelatin-binding domain (with two IGDs) (5,6). This suggests that IGD motifs in larger fragments can act cooperatively to give a much increased response. The enormous enhancements of IGD potency in different contexts is, however, not easy to explain. One possibility is that the increased bioactivity arises from a precise geometric presentation of the different binding motifs. Another, more likely, possibility is that several steps are involved; for example, an initial binding to matrix, followed by receptor encounter and intracellular signaling. It is additionally possible that the IGD-containing Fn1 modules might partially denature the collagen triple helix, exposing RGD triplets to which the ␣v␤3 integrin could then bind. This would explain the high potency of those IGDcontaining Fn1 modules that have high collagen/gelatin binding activity and distinguish them from the N-terminal domain, which has no motogenic activity. Of further interest is the observation that MSF displays proteolytic activity (26).
Another intriguing observation is that full-length FN does not show bioactivity (5). The usual explanation is that IGD motifs are hidden in assembled, intact FN but are bioactive when exposed; for example, as shown here and previously (6), bioactivity appears in smaller fragments. Fibronectin fragments created by proteolysis exhibit many properties not seen in full-length FN (27,28), and cryptic sites can be exposed simply by selective sequence deletion (29). Mechanical tension has also been shown to unravel FN (30), and potential new sites can be revealed through conformational change and molecular rearrangement. The number of cryptic sites revealed at any one time could define the size and nature of the response.
Previous studies have suggested that biological activities of fibronectin, such as fibrillogenesis, can be controlled by concealing active motifs within the tertiary structure (31). The 70-kDa FN fragment has been shown to be critical in fibril assembly (32), and much work has gone into establishing the matrix assembly site. A possible explanation is that the IGD motifs studied here could also contribute to fibrillogenesis.