The hairpin structure of the ⁶F1¹F2²F2 fragment from human fibronectin enhances gelatin binding

Abstract

The solution structure of the ⁶F1¹F2²F2 fragment from the gelatin-binding region of fibronectin has been determined (Protein Data Bank entry codes 1e88 and 1e8b). The structure reveals an extensive hydrophobic interface between the non-contiguous ⁶F1 and ²F2 modules. The buried surface area between ⁶F1 and ²F2 (∼870 Å²) is the largest intermodule interface seen in fibronectin to date. The dissection of ⁶F1¹F2²F2 into the ⁶F1¹F2 pair and ²F2 results in near-complete loss of gelatin-binding activity. The hairpin topology of ⁶F1¹F2²F2 may facilitate intramolecular contact between the matrix assembly regions flanking the gelatin-binding domain. This is the first high-resolution study to reveal a compact, globular arrangement of modules in fibronectin. This arrangement is not consistent with the view that fibronectin is simply a linear ‘string of beads’.

Introduction

The extracellular matrix glycoprotein fibronectin is a large, multifunctional molecule involved in adhesion and migration events in a range of important physiological processes such as embryogenesis, wound healing, haemostasis and thrombosis (Hynes, 1990). As a soluble dimer in plasma, it is involved in blood coagulation through its affinity for fibrin and platelets. As an insoluble network in the extracellular matrix, it interacts with cell surface receptors and with other matrix components such as collagens and proteoglycans, thus assisting cell migration and the maintenance of tissue integrity (Hynes, 1990).

The interaction between fibronectin and collagen in the extracellular matrix is well documented (Hynes, 1990) but poorly understood at the molecular level. The two proteins are co-distributed in tissues, as shown by immunofluorescence and immunocytochemical studies. Addition of extraneous fibronectin also promotes the attachment of fibroblastic cells to collagen substrates in vitro. Further more, fibronectin is observed in a regularly distributed array along collagen fibres synthesized in culture. Knowledge of the molecular basis of fibronectin’s interaction with collagen would provide a better understanding of the structure and function of the extracellular matrix.

The only viable approach to correlating the structure and function of the large, flexible fibronectin monomer at the atomic level is to dissect it into manageable units (Campbell and Downing, 1998). Fortunately, like many other proteins of the extracellular matrix, fibronectin is a mosaic protein consisting of repeating sequence elements or ‘modules’ that are capable of folding independently (Bork et al., 1996). Its primary sequence is composed almost entirely of three types of module (F1, F2 and F3), which are organized into functional domains (Figure 1A). These domains may be isolated in the form of proteolytic fragments that retain affinity for various ligands. Consequently, many of the ligand-binding sites have been mapped to specific regions of the fibronectin polypeptide. The collagen-binding domain can be isolated as a 42 kDa proteolytic fragment that retains affinity for heat-denatured collagen (gelatin) (Hynes, 1990). This domain has the module composition ⁶F1¹F2²F2⁷F1⁸F1⁹F1, where ⁿFX represents the nth type X module in the native protein. Further dissection of this gelatin-binding domain by proteolysis produces three non-overlapping module pairs (⁶F1¹F2, ²F2⁷F1 and ⁸F1⁹F1) each of which retains some degree of gelatin-binding activity (Ingham et al., 1989; Litvinovich et al., 1991). This suggests that the gelatin-binding site (or sites) spans multiple modules in the domain.

Fig. 1. Binding of fibronectin fragments to immobilized collagen α1(I) chains. (A) The mosaic structure of a fibronectin monomer is shown, with the positions of the alternatively spliced regions EDB, EDA and IIICS depicted below. The major binding sites for cells and for other matrix components are labelled. (B) Assaying the binding of fibronectin fragments to immobilized α1 chains by surface plasmon resonance (SPR). A 20 µl aliquot of 300 µM samples of ⁶F1¹F2²F2 (—), ¹F2²F2 (– – –), ⁶F1¹F2 (........), ¹F2 (– - – -) and ²F2 (– - - –) was injected over the same flow cell of immobilized collagen α1(I) chains at a flow rate of 20 µl/min. The surface was regenerated with 50 mM HCl after each injection and the individual sensorgrams overlaid with the BIAevaluation 3.0 software (Biacore). (C) Response (RU) 25 s after the end of the injection of the fibronectin fragments as a comparison of collagen α1(I) chain binding ability. Significant binding was only detected in this assay for ⁶F1¹F2²F2 and ¹F2²F2. Values represent the mean of triplicate samples, with standard deviation of replicate samples shown as error bars.

Attempts to localize the gelatin-binding site further by recombinant expression in Escherichia coli have yielded conflicting results. Analysis of recombinant fragments produced as β-galactosidase fusion proteins showed that both the ¹F2 module and the ¹F2²F2 module pair could bind immobilized gelatin (Banyai et al., 1990). In an earlier study, however, the ⁶F1¹F2²F2 module construct only bound to gelatin if the 14 N-terminal residues of ⁷F1 were present at the C-terminus of the triplet fragment (Owens and Baralle, 1986). A third study identified the smallest recombinant fragment capable of binding to immobilized gelatin as ⁶F1¹F2²F2⁷F1 (Skorstengaard et al., 1994). The occurrence of F2 modules in other gelatin-binding proteins, such as the matrix metalloproteinases (MMP) 2 and 9, provides evidence for their involvement in gelatin binding by fibronectin (Collier et al., 1988; Wilhelm et al., 1989). Furthermore, recombinant expression of F2 modules from these MMPs has produced fragments with high affinity for gelatin (Banyai and Patthy, 1991; Collier et al., 1992; Banyai et al., 1994), whereas recombinant MMP2 lacking F2 modules was devoid of gelatin-binding activity (Murphy et al., 1994; Allan et al., 1995).

Here we describe the gelatin-binding properties and solution structure of the ⁶F1¹F2²F2 fragment from fibro nectin (Protein Data Bank entry codes 1e88 and 1e8b). Dissection of this fragment into the ⁶F1¹F2 module pair and the individual ²F2 module results in a drastic reduction in gelatin affinity, suggesting that module– module interactions are essential for optimal binding activity. A comparison of backbone amide chemical shifts of ⁶F1¹F2²F2 with ⁶F1¹F2 and ¹F2²F2 revealed long-range interactions between the non-contiguous ⁶F1 and ²F2 modules. In the solution structure of ⁶F1¹F2²F2, the ⁶F1 and ²F2 modules interact via an extensive hydrophobic interface whose buried surface area is the largest intermodule contact yet seen in fibronectin. The central ¹F2 module shows no non-covalent interactions with either ⁶F1 or ²F2, implying that the ⁶F1–¹F2 interface observed previously (Bocquier et al., 1999) is disrupted in the presence of additional modules. The hairpin topology of ⁶F1¹F2²F2 may facilitate intramolecular contact between the flanking ¹F1²F1³F1⁴F1⁵F1 and ¹F3 fragments, an interaction that is believed to modulate fibronectin fibrillogenesis in the extracellular matrix. Its conformation may also account for the previously noted disruptions in the otherwise uniform strand-like images seen in electron micrographs of fibronectin at high ionic strength. This is the first high-resolution study to reveal a compact, globular arrangement of modules in fibronectin.

Results and discussion

Enhanced gelatin-binding activity in the ⁶F1¹F2²F2 fragment

The isolated ¹F2 and ²F2 modules, the ⁶F1¹F2 and ¹F2²F2 module pairs and the ⁶F1¹F2²F2 module triplet were produced as described in Materials and methods. During their purification, the isolated ¹F2 and ²F2 modules and the ⁶F1¹F2 module pair bound weakly to the gelatin affinity column and were separated from the non-binding contaminants by isocratic elution; the ¹F2²F2 module pair and the ⁶F1¹F2²F2 module triplet bound more tightly and were eluted with a urea gradient.

The binding of ¹F2, ²F2, ⁶F1¹F2, ¹F2²F2 and ⁶F1¹F2²F2 to immobilized collagen α1(I) chains was analysed in greater detail using surface plasmon resonance (SPR) (Figure 1B and C). For the ⁶F1¹F2²F2 binding to the α1(I) polypeptide chains, an acceptable fit (χ² <2) could be obtained with a model that assumed surface heterogeneity, a common observation of immobilization through amine side chains. This treatment resolved the data into two components, with the major component exhibiting association (k_on) and dissociation (k_off) rates of 115 ± 12 M^–1s^–1 and 0.0036 ± 0.001 s^–1, respectively, and an equilibrium dissociation constant of 31 ± 6 µM (K_d = k_off/k_on). No significant changes in the ⁶F1¹F2²F2 binding were observed under acidic conditions (pH 4.5). The ¹F2²F2 bound with much lower affinity, with k_on and k_off values of 94 ± 21 M^–1s^–1 and 0.012 ± 0.001 s^–1, respectively, which is equivalent to an equilibrium dissociation constant of 131 ± 23 µM. The binding of ¹F2, ²F2 and ⁶F1¹F2 to immobilized α1(I) chains was too weak to quantify the kinetics with this assay.

The very low affinity of the ¹F2 and ²F2 modules and of the ⁶F1¹F2 module pair for α1(I) chains demonstrates that the gelatin-binding site of ⁶F1¹F2²F2 does not reside entirely within any single module (Figure 1C). The moderate affinity of the ¹F2²F2 module pair may result from cooperativity between weak, independent binding sites on each module since there is no defined interface between the modules and they tumble independently of each other (Smith et al., 2000). The enhanced binding activity of ⁶F1¹F2²F2 must therefore involve the formation of a composite binding site involving ¹F2 and/or ²F2 and the ⁶F1 module, or cooperativity between independent, weak gelatin-binding sites on separate modules. The structural basis for this affinity was investigated by nuclear magnetic resonance (NMR) spectroscopy.

⁶F1 and ²F2 interact in the ⁶F1¹F2²F2 fragment

Preliminary information regarding the nature and site of any intermodule interaction in a mosaic protein can be derived from a comparison of the NMR chemical shifts of overlapping fragments under identical solution conditions. We compared the backbone amide resonances (N^H and H^N) of the ⁶F1¹F2²F2 fragment with the overlapping ⁶F1¹F2 and ¹F2²F2 module pairs (Figure 2). Removal of the ²F2 module from ⁶F1¹F2²F2 results in shift perturbations at the C-terminus of the ¹F2 module (Figure 2A), as expected from the change in charge. However, many major shift changes are seen in ⁶F1, in the region of Val10–Lys20. Similarly, removal of ⁶F1 from ⁶F1¹F2²F2 results in shift perturbations at the N-terminus of the ¹F2 module, but the largest shift changes are seen in the ²F2 module, in the region of Ser111–His117 (Figure 2B). Therefore, the ⁶F1 and ²F2 modules must interact in ⁶F1¹F2²F2 despite being separated in sequence by the ¹F2 module.

Fig. 2. Amide chemical shift perturbation upon dissection of ⁶F1¹F2²F2. (A and B) Combined chemical shift perturbation (Δδ) of the H^N and N^H backbone resonances, upon (A) removal of the ²F2 module from ⁶F1¹F2²F2, and (B) removal of the ⁶F1 module from ⁶F1¹F2²F2. In each case, Δδ = {|ΔδH^N| + (|ΔδN^H|/5)}/2, where ΔδH^N and ΔδN^H are the amide proton and amide nitrogen chemical shift differences, respectively.

Solution structure of ⁶F1¹F2²F2

The solution structure of the ⁶F1¹F2²F2 fragment could, in theory, be modelled by combining the NMR restraints derived from the independent studies on the overlapping ⁶F1¹F2 and ¹F2²F2 module pairs (Bocquier et al., 1999; Smith et al., 2000). However, because of the extensive amide chemical shift changes in the ⁶F1 and ²F2 modules, and thus the possibility of significant structural alterations, the solution structure was determined ab initio using only restraints derived from NMR experiments on ⁶F1¹F2²F2 itself. Of the 100 structures calculated, 20 were selected on the basis of their good agreement with the experimental restraints and their minimal deviations from ideal covalent geometry (Table I).

Table I. Experimental restraints and structural statistics.

R.m.s. deviations from experimental dataa
all 2024 NOE restraints	0.009 ± 0.001 Å
515 intraresidue NOEs {i = j}	0.008 ± 0.002 Å
598 sequential NOEs {|i–j| = 1}	0.007 ± 0.002 Å
231 short-range NOEs {1 < |i–j| < 5}	0.014 ± 0.003 Å
539 long-range NOEs {|i–j| > 4}	0.009 ± 0.002 Å
41 6F1–2F2 intermodule NOEs	0.006 ± 0.003 Å
100 ambiguous NOEs	0.011 ± 0.004 Å
106 hydrogen bond restraints	0.010 ± 0.002 Å
69 dihedral φ angles	0.066 ± 0.036°
R.m.s. deviations from ideal covalent geometry
bonds	0.0013 ± 0.0001 Å
angles	0.245 ± 0.004°
impropers	0.123 ± 0.014°
Ramachandran analysisb
residues in favoured regions	55.0%
residues in additional allowed  regions	39.9%
residues in generously allowed  regions	4.4%
residues in disallowed regions	0.7%
Coordinate precision: secondary structure backbone, all heavy atomsc
6F1(1F22F2)	0.46 ± 0.17 Å, 1.21 ± 0.82 Å
(6F1)1F2(2F2)	0.33 ± 0.09 Å, 0.99 ± 0.88 Å
(6F11F2)2F2	0.37 ± 0.06 Å, 0.98 ± 0.69 Å
6F1(1F2)2F2	0.53 ± 0.14 Å, 1.15 ± 0.73 Å
(6F1)1F2(2F2)	11.7 ± 3.48 Å, 10.2 ± 3.59 Å

^aNone of the 20 accepted structures showed distance restraint violations of >0.3 Å or dihedral restraint violations of >2°. No distance or dihedral restraints were consistently violated by >0.1 Å or 1°, respectively.

^bProlines, glycines and terminal residues are excluded.

^cCoordinate r.m.s. deviations were calculated following best-fit superposition over the secondary structure elements of the underlined module(s). Residues from modules in parentheses were excluded from the calculation.

The tertiary structures of the individual modules in ⁶F1¹F2²F2 are very similar to those in previous studies of the ⁶F1¹F2 module pair (Bocquier et al., 1999) and the isolated ²F2 module (Sticht et al., 1998). The back bone heavy atom (N^H, Cα, C) root mean square (r.m.s.) deviations over secondary structure elements are 0.98 Å for ⁶F1, 0.65 Å for ¹F2 and 1.21 Å for ²F2. The ⁶F1 module (Figure 3A) comprises a short N-terminal double-stranded antiparallel β-sheet (AB) that folds over a triple-stranded β-sheet (CDE). The β-sheets are linked by two, conserved disulfide bridges in a 1–3 and 2–4 pattern connecting strands A–D and D–E, respectively. The module core is composed primarily of two highly conserved aromatic residues Tyr12 and Trp18, and a hydrophobic residue, Val36.

Fig. 3. Structural definition of the individual modules in ⁶F1¹F2²F2. (A–C) Individual module overlays for the ensemble of 20 lowest energy structures of ⁶F1¹F2²F2. Each module has been overlaid onto the lowest energy structure by best-fit superposition over the backbone heavy atoms of its secondary structure elements. The secondary structure is coloured green in ⁶F1, gold in ¹F2 and red in ²F2. The disulfide bridges are shown in yellow, and the side chains of non-polar residues that are invariant or highly conserved between modules are shown in cyan for ⁶F1, pink for ¹F2 and purple for ²F2. (D) Average atomic r.m.s. deviation between the 20 accepted structures for the backbone heavy atoms (solid line) and all heavy atoms (dotted line). The vertical dashed lines mark the exon boundaries between the modules. (E) ¹⁵N-{¹H}-NOE for the ⁶F1¹F2²F2 fragment. Lower ¹⁵N-{¹H}-NOE values indicate regions with increased backbone flexibility.

The ¹F2 module (Figure 3B) comprises two double-stranded antiparallel β-sheets (AB and CD), oriented approximately perpendicular to each other, with a single α-helical turn located between strands C and D. The cleft between the β-sheets is occupied by side chains of invariant hydrophobic and aromatic residues, which make up the module core. On the opposite face of the second β-sheet, two disulfide bonds link the invariant cysteines, with connectivities 1–3 and 2–4. The topology of ²F2 (Figure 3C) is very similar to that of ¹F2, but includes the additional A′A′′ β-sheet preceding the first cysteine residue.

In general, the individual ⁶F1, ¹F2 and ²F2 modules in ⁶F1¹F2²F2 are well defined (Figure 3A–C) with low backbone heavy atom r.m.s. deviations over their secondary structure elements (Table I). The low ¹⁵N-{¹H}-NOE for the C–D loop in ⁶F1, and the B–C loops in ¹F2 and ²F2, each of which have backbone r.m.s. deviations >1.0 Å, shows that the poorer definition of these regions arises from backbone flexibility rather than a lack of experimental data (Figure 3D and E).

The list of long-range NOEs used in the final round of ⁶F1¹F2²F2 structure calculations included an extensive array of intermodule restraints between nine residues in ⁶F1 and eight in ²F2 (Table I). This resulted in a precise definition of the relative locations and orientations of the ⁶F1 and ²F2 modules in ⁶F1¹F2²F2 (Figure 4; Table I). The ⁶F1–²F2 interface is formed by the docking of the external edges of β-strands C of ⁶F1 and A′ of ²F2, and by the annealing of two extensive hydrophobic surfaces involving Val10, Tyr12, Met16, Leu19 and Leu28 from ⁶F1, and Leu103, Ala114, Leu115 and Thr145 from ²F2 (Figure 5A). These non-polar residues account for 53% of the total buried surface area between the two modules. There are no titratable groups involved in the remaining 47% that could disrupt the hairpin topology within the pH range 4.5–7.4.

Fig. 4. Solution structure of ⁶F1¹F2²F2. (A) A stereoview of the ensemble of the 20 lowest energy structures of ⁶F1¹F2²F2 is shown. The structures were superimposed on the backbone heavy atoms of the secondary structure elements of the ⁶F1 and ²F2 modules of the lowest energy structure. The colour scheme for the protein backbone and side chain residues is the same as in Figure 3A–C. (B) Orthogonal stereoview of the ⁶F1¹F2²F2 ensemble generated by a 90° rotation about the horizontal. For clarity, the disordered ¹F2 module has been omitted.

Fig. 5. Module reorganization upon dissection of ⁶F1¹F2²F2. Ribbon diagrams of the minimized average structures of (A) ⁶F1¹F2²F2 and (B) ⁶F1¹F2. The colour scheme for the secondary structure elements is as in Figure 33A–C. Side chains for which ⁶F1–²F2 intermodule NOEs were observed (V10, Y12, S13, M16, L19 and L28 of ⁶F1, and L103, Q105, S111, N112, A114, L115, T145 and K153 for ²F2) are shown in cyan for ⁶F1 and purple for ²F2. Removal of the ²F2 module allows the side chain of Y68 (pink) in ¹F2 to interact with L19 and L28 in ⁶F1.

No long- or short-range NOEs were observed, either between the ⁶F1 and ¹F2 modules, or between the ¹F2 and ²F2 modules. The ⁶F1–¹F2 and ¹F2–²F2 linker sequences are flexible, as indicated by their low ¹⁵N-{¹H}-NOE values (Figure 3E). Hence, the relative location and orientation of ¹F2 are somewhat ill defined (Figure 4A; Table I) since the only positional constraints on this central module are the covalent tethering of its N- and C-termini, and steric hindrance from ⁶F1 and ²F2.

The gelatin-binding site(s) of ⁶F1¹F2²F2

In each F2 module, five of the core aromatic residues (Tyr62, Trp81, Tyr88, Tyr94 and Phe96 in ¹F2; Tyr122, Trp141, Tyr148, Phe154 and Phe156 in ²F2) form an extensive, solvent-exposed hydrophobic surface (Figure 3B and C). A pocket in each of these surfaces is thought to provide a binding site for non-polar residues in type I collagen (Pickford et al., 1997). This is supported by the binding of the collagen-like peptide (Pro-Pro-Gly)₆ to the ²F2 module from MMP2, which produced backbone H^N and N^H chemical shift perturbations in the equivalent residues (Briknarová et al., 1999). Furthermore, in the crystal structure of pro-MMP2, a phenylalanine side chain in the inhibitory propeptide occupies the equivalent hydrophobic pocket in the ³F2 module, preventing it from binding its gelatin substrate (Morgunova et al., 1999).

In all structures of the ⁶F1¹F2²F2 ensemble, solvent access to the putative binding sites on the F2 modules is unhindered by the ⁶F1–²F2 interface. In the average ⁶F1¹F2²F2 structure (Figure 5A), the binding sites are oriented in opposite directions. This configuration of F2 modules observed here is reminiscent of the pro-MMP2 crystal structure where the three F2 modules are also oriented with their binding sites facing outwards (Morgunova et al., 1999). Recombinant ¹F2²F2³F2 from MMP2 has been shown to be capable of binding multiple collagen triple helices (Steffensen et al., 1995), suggesting that F2 modules form separate binding surfaces that intercalate between molecules in a collagen fibril. The flexibility of the ¹F2 module with respect to ⁶F1 and ²F2 may permit a variety of collagen-binding conformations. It has also been proposed that the observed pliability of the ⁹F3¹⁰F3 module pair may allow it to accommodate some variation in the integrin structure to which it binds (Copié et al., 1998).

The structural basis for the 4-fold increase in affinity for α(I) chains from ¹F2²F2 to ⁶F1¹F2²F2 is not known, but may arise from an extension of the ²F2 binding site onto ⁶F1. The ⁶F1 module docks onto ²F2 alongside its gelatin-binding site in an orientation that may allow solvent-exposed residues in the AB sheet of ⁶F1 to contribute to binding (Figure 4B). Alternatively, the binding enhancement may arise from stabilization of the ²F2 module due to the ⁶F1–²F2 interface. The backbone H^N protons of residues within the AB sheet of ⁶F1 and the A′A′′ sheet of ²F2 undergo much slower solvent exchange in ⁶F1¹F2²F2 than observed previously for ⁶F1¹F2 (Bocquier et al., 1999) and ²F2 (Sticht et al., 1998). Stabilization of the ²F2 module was observed previously in differential scanning calorimetric studies on the ⁶F1¹F2²F2⁷F1 proteolytic fragment, although this was originally attributed to an interaction between ¹F2 and ²F2 (Litvinovich et al., 1991).

The contributions of ⁷F1, ⁸F1 and ⁹F1 to collagen binding have yet to be determined. The only structural information available for these domains is the isolated ⁷F1 (Baron et al., 1990), but there is no information on its interactions with adjacent modules. Given the hairpin structure observed in ⁶F1¹F2²F2, we believe that the extension of the ⁶F1¹F2²F2 nucleus with these modules will provide a better examination of the collagen-binding domain function.

Module reorganization upon dissection

The two techniques capable of providing high-resolution structural information on binding surfaces, namely X-ray diffraction and NMR, are both limited in the size of mosaic proteins that can be studied (Campbell and Downing, 1998). In the case of X-ray diffraction, the limitation arises from the difficulty in crystallizing a protein that is inherently flexible, whilst for NMR the molecular weight is the limiting factor. Hence, for both techniques, large mosaic proteins must frequently be dissected into fragments that are amenable to analysis. However, a comparison of the solution structures of ⁶F1¹F2 and ⁶F1¹F2²F2 illustrates a potential complication associated with this dissection strategy (Figure 5).

The F1 and F2 modules in ⁶F1¹F2 were found to interact via a small hydrophobic interface of ∼340 Å², involving the side chains of Leu19 and Leu28 from ⁶F1, and Tyr68 from ¹F2 (Bocquier et al., 1999). This interaction resulted in a significant upfield shift in the N^H resonance of Ser69 relative to the isolated ¹F2 module (Hashimoto et al., 2000). However, this ⁶F1–¹F2 interface is inconsistent with the solution structure of ⁶F1¹F2²F2 presented here. The upfield shift of the Ser69 N^H resonance is reversed in ⁶F1¹F2²F2, consistent with a break up of the weak ⁶F1–¹F2 interface (Figure 2A). None of the 18 weak intermodule NOEs previously observed between ⁶F1 and ¹F2 is apparent in the spectra of ⁶F1¹F2²F2; manual incorporation of these restraints into the structure calculations of ⁶F1¹F2²F2 also resulted in structures with 20–25% higher potential energy. Therefore, this contact between the modules must have arisen from a module reorganization; removal of the ²F2 module relieves the covalent and steric constraints on the ¹F2 module, allowing a hydrophobic collapse of non-polar residues that are distant and/or buried in the intact protein (Figure 5). Such rearrangements are a well-known phenomenon in intracellular proteins, for example, the rearrangement of SH2 and SH3 domains in Src family tyrosine kinases (Sicheri et al., 1997).

A globular domain in a fibrillar protein

The ⁶F1–²F2 interface gives the ⁶F1¹F2²F2 fragment a compact, hairpin topology (with average dimensions 15 × 19 × 32 Å). This is in sharp contrast to the extended, near-linear arrangement of F3 modules in the crystal structures of the cell-binding ⁷F3⁸F3⁹F3¹⁰F3 (Leahy et al., 1996) and heparin-binding ¹²F3¹³F3¹⁴F3 (Sharma et al., 1999) fragments of fibronectin (Figure 6). The global topology of ⁶F1¹F2²F2 is in agreement with previous calorimetric studies, which suggested that the gelatin-binding domain has a relatively compact structure (Litvinovich et al., 1991).

Fig. 6. Global topologies of multimodule fibronectin fragments. Solvent-accessible surfaces have been superimposed over ribbon diagrams for the minimized average structure of ⁶F1¹F2²F2, and the crystal structures of ⁷F3⁸F3⁹F3¹⁰F3 (Leahy et al., 1996) and ¹²F3¹³F3¹⁴F3 (Sharma et al., 1999). The fragment structures are mapped onto the mosaic illustration of fibronectin, which has been folded to account for the hairpin structure of ⁶F1¹F2²F2.

Numerous biophysical and biochemical studies have shown that intact fibronectin undergoes a substantial change in structure from a compact conformation at low ionic strength to a more extended conformation in high salt (Engel et al., 1981; Erickson et al., 1981; Erickson and Carrell, 1983; Rocco et al., 1983; Lai et al., 1993). However, even at high ionic strength, numerous kinks are seen in electron micrographs of the molecule (Engel et al., 1981; Erickson et al., 1981; Erickson and Carrell, 1983; Rocco et al., 1983; Lai et al., 1993; Johnson et al., 1999). The resistance of these bends in the protein to high salt concentrations suggests that the interactions involved are predominantly non-polar in nature. Thus, the solution structure of ⁶F1¹F2²F2 presented here, with its hairpin conformation and extensive hydrophobic interface between the ⁶F1 and ²F2 modules, is one possible explanation for the kinks observed towards the N-terminal ends of the fibronectin dimer (Figure 6).

Many of the models for the compact structure of fibronectin at low ionic strength involve a folding over of the ¹F1²F1³F1⁴F1⁵F1 fragment allowing it to form interdomain interactions with the F3 modules in the protein (Williams et al., 1982; Homandberg and Erickson, 1986; Rocco et al., 1987; Ingham et al., 1988; Khan et al., 1990). Of particular interest is the potential contact between the ¹F1²F1³F1⁴F1⁵F1 and ¹F3 fragments (Figure 1A), an intramolecular interaction that is believed to suppress fibronectin fibrillogenesis (Aguirre et al., 1994; Schwarzbauer and Sechler, 1999). While this contact would be unlikely to take place if the gelatin-binding domain had a more extended organization, akin to those of the cell-binding and heparin-binding fragments (Figure 6), the hairpin conformation of ⁶F1¹F2²F2 in the gelatin-binding domain may facilitate intramolecular contact between the flanking ¹F1²F1³F1⁴F1⁵F1 and ¹F3 fragments. Interestingly, the addition of collagen or the cyanogen bromide fragment CB7 of the α1(I) chain results in a reorganization and increased accumulation of fibronectin fibrils at the surface of collagen-deficient MOV-13 cells (Dzamba et al., 1993). Thus, collagen binding may induce a conformational change that disengages the inhibitory ¹F1²F1³F1⁴F1⁵F1–¹F3 intramolecular interaction, thus facilitating matrix assembly; this is also supported by the observed partial unfolding of plasma fibronectin upon binding of the α1(I)-CB7 fragment (Williams et al., 1982).

It has been proposed that stretching of the fibronectin molecule by the cell might regulate matrix function by exposing new binding sites and affecting cell adhesion (Schwarzbauer and Sechler, 1999). Applied tension could extend the molecule through a breakdown of the above interdomain interactions, the local disruption of intermodule interfaces or even complete unfolding of F3 modules (Erickson, 1994; Oberhauser et al., 1998; Ohashi et al., 1999). For example, the disruption of the ⁹F3–¹⁰F3 interface due to an extension of the intermodule linker resulted in a reduction of integrin-mediated cell adhesion and intracellular signalling (Grant et al., 1997). Thus, it could be argued that stretching might also provide a means for actively modulating the affinity of fibronectin for collagen by disrupting the ⁶F1–²F2 intermodule interface that enhances binding. However, the buried surface area between ⁶F1 and ²F2 in the ⁶F1¹F2²F2 structure ensemble is on average 868 Å² (±48 Å²), much greater than the 340 Å² buried between ⁹F3 and ¹⁰F3 (Leahy et al., 1996). Therefore, the smaller and more flexible ⁹F3–¹⁰F3 interface (Copié et al., 1998) is likely to deform more easily in response to stress, with the result that the cell retracts from the fibronectin matrix before it can apply the necessary tension to disrupt the fibronectin–collagen interaction.

Materials and methods

Preparation of recombinant fibronectin modules

Fragments of the collagen-binding domain of human fibronectin were prepared by recombinant expression from the methylotrophic yeast Pichia pastoris. Expression and purification of the isolated ¹F2 and ²F2 modules and the ⁶F1¹F2 and ¹F2²F2 module pairs have been described previously (Pickford et al., 1997; Sticht et al., 1998; Bocquier et al., 1999; Smith et al., 2000). The P.pastoris clone expressing the ⁶F1¹F2²F2 triplet (corresponding to residues 274–433 of mature human fibronectin) was produced in analogous fashion to that described previously for the ¹F2 module (Pickford et al., 1997). Expression of unlabelled and uniformly ¹⁵N-labelled ([u-¹⁵N]) proteins was carried out in a 1 l fermentor (Electrolab Ltd, Tewkesbury, UK) following the detailed protocol for the ⁴F1⁵F1 module pair (Bright et al., 1999). Each fragment was purified by a combination of cation exchange chromatography on SP-Sepharose Fast Flow (Amersham Pharmacia Biotech), affinity chromatography on gelatin–Sepharose 4B (Amersham Pharmacia Biotech) and reverse phase high-performance liquid chromatography (HPLC) on a C8 column (Rainin). Prior to the last purification step, those fragments containing the ²F2 module were treated with Endo H_f (New England Biolabs) to trim the high mannose sugar attached to residue Asn25 back to a single N-acetylglucosamine (GlcNAc) (Sticht et al., 1998). The identity and purity of each fragment were confirmed by electrospray mass spectrometry and N-terminal sequence analysis.

Surface plasmon resonance

The collagen α1(I) chain was purified from human placental type I collagen (Sigma) by size-exclusion chromatography through Sephacryl-S400HR (Amersham Pharmacia Biotech) followed by cation exchange chromatography at 42°C on a Mono-S HR5/5 column (Amersham Pharmacia Biotech). SPR experiments were performed on a BIAcore 2000 instrument (Biacore AB, Uppsala, Sweden). Purified collagen α1(I) chains were immobilized on the dextran matrix of a CM5 sensorchip (Biacore AB) in 10 mM acetate buffer pH 4.0, and covalently bound using the amine coupling method as described in the BIAapplications handbook (Biacore AB). A control flow cell was also created by derivatizing the surface for amine coupling in the absence of protein. For comparative binding of the fibronectin fragments, a collagen α1(I) immobilization level of ∼7000 resonance units (RU) was used. In order to avoid mass transport limitation, this immobilization level was reduced to ∼2000 RU for the kinetic analysis of fibronectin fragment binding. The same sensorchip surface was used in each set of experiments. Binding experiments were carried out at 25°C in HBS-EP running buffer [10 mM HEPES pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.005% (v/v) Surfactant P20] at a flow rate of 20 µl/min. A regeneration step of 30 s exposure to 50 mM HCl was applied after each injection to return to the baseline. To compare the binding of the fibronectin fragments, 20 µl of 300 µM samples of ⁶F1¹F2²F2, ¹F2²F2, ⁶F1¹F2, ¹F2 and ²F2 were injected over immobilized collagen α1(I) chains, and the amount of protein bound to the sensorchip was monitored by the change in RU. For the kinetic analyses, triplicates of 20 µl of five dilutions (25–200 µM) of the fibronectin fragments ¹F2²F2 and ⁶F1¹F2²F2 were injected over the flow cells. The sensorgram data were analysed using global fitting procedures in the BIAevaluation 3.0 program (Karlsson and Falt, 1997). The association (k_on) and dissociation (k_off) rates were evaluated by analysing appropriate components of the sensorgram curve.

NMR spectroscopy

All NMR experiments were acquired at 25°C on a spectrometer built in-house at the Oxford Centre for Molecular Sciences incorporating an Oxford Instruments magnet (750.1 MHz for ¹H) and a GE/Omega computer. Experiments were recorded in a phase-sensitive manner using the States/TPPI method for quadrature detection in the indirectly detected dimensions. In all heteronuclear experiments, ¹H–¹⁵N decoupling was achieved using a GARP pulse-train with a 1.7 kHz decoupling bandwidth. Samples for amide chemical shift comparison were prepared by dissolving either [u-¹⁵N]⁶F1¹F2, [u-¹⁵N]¹F2²F2 or [u-¹⁵N]⁶F1¹F2²F2 to a final concentration of 1.0 mM in 90% H₂O/10% D₂O containing 1 mM 1,4-dioxane, and adjusting the pH to 4.5 (meter uncorrected for deuterium). For each sample, a one-dimensional ¹H spectrum and a two-dimensional gradient-enhanced [¹H–¹⁵N]-HSQC spectrum (Kay et al., 1992) were acquired. Samples for ⁶F1¹F2²F2 assignment and structure determination were prepared by dissolving [u-¹⁵N]⁶F1¹F2²F2 to a final concentration of 2.0 mM in either 90% H₂O/10% D₂O or 99.9% D₂O, adding 1,4-dioxane to 1 mM, and adjusting the pH to 4.5. The following spectra were recorded in H₂O: a two-dimensional [¹H–¹⁵N]-HMQC-J (Kay and Bax, 1990), a three-dimensional gradient-enhanced [¹H–¹⁵N]-TOCSY-HSQC with 46 ms mixing time (Marion et al., 1989) and a three-dimensional gradient-enhanced [¹H–¹⁵N]-NOESY-HSQC with 60 ms mixing time (Marion et al., 1989). The following spectra were recorded in D₂O: a two-dimensional [¹H–¹H]-DQF-COSY (Rance et al., 1983), a two-dimensional [¹H–¹H]-NOESY with 60 ms mixing time (Kumar et al., 1980) and a two-dimensional [¹H–¹H]-TOCSY with 46 ms mixing time (Davis and Bax, 1985). For measuring the ¹⁵N-{¹H}-NOE, two experiments were recorded, either with (NOE) or without (NONOE) ¹H saturation, during the recycle delay (Kay et al., 1989). Slowly exchanging amide protons were identified by lyophilizing [u-¹⁵N]⁶F1¹F2²F2 from H₂O, dissolving in D₂O and recording multiple gradient-enhanced [¹H–¹⁵N]-HSQC spectra at two-hourly intervals.

Data processing and analysis

Data processing was performed using the FELIX 2.3 software package (Biosym Technologies Inc.). Homonuclear DQF-COSY, TOCSY and NOESY experiments were processed as described previously (Pickford et al., 1997). The three-dimensional gradient-enhanced [¹H–¹⁵N]-TOCSY-HSQC and [¹H–¹⁵N]-NOESY-HSQC data sets were processed using a Lorentzian–Gaussian multiplication in t₃, linear prediction and apodization in t₂ using a 70° phase-shifted squared sine-bell window function, and a Kaiser function for apodization in t₁. Proton chemical shifts were referenced relative to the internal standard 1,4-dioxane at 3.743 p.p.m., with indirect referencing in the ¹⁵N dimension using a ¹⁵N/¹H frequency ratio of 0.101329118 (Wishart et al., 1995). The program NMRView v3.0.b1 (Merck and Co., Inc.) was used for spectral assignment and the derivation of structural restraints. Complete assignment of the backbone N^H, H^N and Hα resonances, and of most side chain proton and nitrogen resonances of ⁶F1¹F2²F2 at 25°C and pH 4.5 was achieved using standard homonuclear and heteronuclear techniques. The assignment process was assisted by previous studies on the ⁶F1¹F2 module pair (Bocquier et al., 1999) and the isolated ²F2 module (Sticht et al., 1998) under the same conditions. Backbone N^H and H^N resonances of the ¹F2²F2 module pair, which had previously been assigned at pH 6.0 (Smith et al., 2000), were reassigned at pH 4.5. An iterative procedure was used in the assignment of NOEs: those that could not be assigned unambiguously were included as ambiguous restraints during initial structure calculations (Nilges, 1995) followed, where possible, by resolution of the ambiguity by inspection of preliminary structures. NOEs were calibrated using interproton distances in regions of regular secondary structure, and converted into three distance restraint categories (‘strong’, ‘medium’ and ‘weak’) with upper distance limits of 2.8, 3.5 and 5.0 Å, respectively. Hydrogen bond restraints were introduced in the final round of the calculation if three criteria were met: slow solvent exchange of the H^N proton, an H^N–O distance <2.3 Å and an O–H^N–N^H angle >120° in at least 70% of the unrestrained structures. For each hydrogen bond, two distance restraints were introduced into the calculation (d_H^N_–O = 1.7–2.3 Å and d_N–O = 2.4–3.3 Å). Backbone φ torsion angle restraints were derived by measuring ³J_H^N_–Hα spin–spin coupling constants from the [¹H–¹⁵N]-HMQC-J spectra using spectral simulations (Redfield et al., 1991). For those residues with ³J_H^N_–Hα <6 Hz or ³J_H^N_–Hα >8 Hz, estimates of φ angles were obtained using a modified Karplus equation (Pardi et al., 1984) and included as restraints in the structure calculations with an error of ±30°.

Structure calculations and analysis

Structure calculations were performed using an ab initio simulated annealing protocol within the program CNS v0.9 (Brünger et al., 1998). The ‘parallhdg.pro’ forcefield (version 5.1) was used to describe the covalent and non-bonded interactions for the polypeptide (Linge and Nilges, 1999). Parameters for the N-linked GlcNAc on Asn25 of the ²F2 module were derived as previously described (Sticht et al., 1998). The non-bonded energy was calculated using a purely repulsive function with a final value of the van der Waals radii scaled by a factor of 0.75 (Linge and Nilges, 1999). A total of 100 structures were calculated using a simulated annealing profile similar to that described previously (Sticht et al., 1998). It comprised four stages: a high temperature conformational search phase in cartesian space (50 ps at 2000 K with a 2 fs time step), two cooling phases (2000 to 1000 K in 25 ps, and 1000 to 100 K in 25 ps, each with a 1 fs time step) and a final minimization phase. The final values for the force constants were K_bond = 1000 kcal/mol/Ų, K_angl = 500 kcal/mol/rad², K_impr = 500 kcal/mol/rad², K_vdw = 4 kcal/mol, K_noe = 50 kcal/mol/Ų and K_cdih = 200 kcal/mol/rad². The structures were refined using an additional cycle of simulated annealing similar to the second cooling phase above, followed by extensive restrained energy minimization. The floating assignment of prochiral groups was achieved using a novel procedure named SOPHIE (for ‘spinning of prochiral hydrogens’). Throughout the conformational search and cooling phases, the diastereospecifically unassigned groups were allowed to rotate freely about the bond connecting their pseudoatom and prochiral centre. Then, during the final minimization phase, each group was eased into either the pro-R or pro-S position by enforcing the correct bond angles at the prochiral centre. The stereochemical quality of the structures was assessed using the program PROCHECK_NMR (Laskowski et al., 1993). Buried surface areas were calculated in CNS using a probe radius of 1.4 Å. Atomic r.m.s. deviations were calculated following best-fit superposition of each accepted structure onto the secondary structure backbone heavy atoms of the lowest energy structure. The average structure was calculated by superimposing over the backbone heavy atoms (N, Cα, C) of the secondary structure elements of each module. Geometric strain was removed from this average structure by extensive restrained energy minimization in CNS (Brünger et al., 1998). Molecular models were generated with the programs MOLMOL (Koradi et al., 1996) and POV-Ray (http://www.povray.org).

Accession codes

The list of ¹H and ¹⁵N resonance assignments of ⁶F1¹F2²F2 at pH 4.5 and 25°C has been deposited at the BioMagResBank with the accession number 4830. The coordinates of the ⁶F1¹F2²F2 NMR structure ensemble and the minimized average structure have been deposited in the Brookhaven Protein Data Bank with the ID codes 1e88 and 1e8b, respectively.