Interdomain association in fibronectin: insight into cryptic sites and fibrillogenesis

Abstract

The process by which fibronectin (FN), a soluble multidomain protein found in tissue fluids, forms insoluble fibrillar networks in the extracellular matrix is poorly understood. Cryptic sites found in FN type III domains have been hypothesized to function as nucleation points, thereby initiating fibrillogenesis. Exposure of these sites could occur upon tension-mediated mechanical rearrangement of type III domains. Here, we present the solution structures of the second type III domain of human FN (²FNIII), and that of an interaction complex between the first two type III domains (¹⁻²FNIII). The two domains are connected through a long linker, flexible in solution. A weak but specific interdomain interaction maintains ¹⁻²FNIII in a closed conformation that associates weakly with the FN N-terminal 30 kDa fragment (FN30 kDa). Disruption of the interdomain interaction by amino-acid substitutions dramatically enhances association with FN30 kDa. Truncation analysis of ¹⁻²FNIII reveals that the interdomain linker is necessary for robust ¹⁻²FNIII–FN30 kDa interaction. We speculate on the importance of this interaction for FN function and present a possible mechanism by which tension could initiate fibrillogenesis.

Introduction

Fibronectin (FN) is an extracellular matrix (ECM) protein essential for normal cell adhesion and mobility in vertebrates (Mao and Schwarzbauer, 2005). The disulfide crosslinked FN homodimer is largely composed of three types of repeating domains, FNI, FNII and FNIII. These domain types are ubiquitous in modular proteins and structural comparisons show a high degree of homology even in cases of low sequence identity (Campbell and Spitzfaden, 1994). FN is soluble in blood and tissue fluids, but forms insoluble fibrillar networks in the ECM via a tightly regulated cellular process that depends on the presence of integrins and other cell surface receptors. The mechanism of initiation and regulation of fibrillogenesis is poorly understood, although initiation is believed to depend on cell-generated tension, which uncovers cryptic sites for self-association in the FN molecule (Geiger et al, 2001). Several cryptic sites are known to exist in FN type III domains (Ingham et al, 1997), and it has been hypothesized that they could serve as possible nucleation sites for FN polymerization and fibril assembly. The evidence for tension-induced exposure of interaction sites and the remarkable elasticity observed in FN fibrils (Erickson, 2002; Abu-Lail et al, 2006) led to the hypothesis that FNIII domain unfolding plays an important role in fibrillogenesis (Mao and Schwarzbauer, 2005). A number of recent studies have thus focused on the mechanical unfolding properties of FN type III domains (Oberhauser et al, 2002; Gao et al, 2003; Abu-Lail et al, 2006).

The first two type III domains of FN, ¹FNIII and ²FNIII, make important contributions in the fibrillogenesis process. Denatured ¹FNIII intermediates have been implicated in interactions with the FN N-terminal type I domains (Hocking et al, 1994), and peptide fragments from the same domain (Morla and Ruoslahti, 1992), or anti-¹FNIII antibodies (Chernousov et al, 1991) can inhibit matrix assembly. Anastellin, a C-terminal fragment of ¹FNIII (Briknarova et al, 2003), promotes FN polymerization to an insoluble, matrix-like form (Morla et al, 1994) through direct association with a number of type III domains (Ohashi and Erickson, 2005). Nevertheless, recombinant FN lacking ¹FNIII was reported to assemble in a matrix form highly similar to that of native FN (Sechler et al, 2001). In contrast, ²FNIII is essential for robust matrix accumulation and fibril growth, and it contains at least one FN self-association site (Aguirre et al, 1994; Sechler et al, 2001). These results suggest that the first two type III domains of fibronectin (¹⁻²FNIII) act together as a functional unit to regulate fibril assembly and promote elongation. Mechanical unfolding by atomic force microscopy (AFM) of ¹FNIII or ¹⁻²FNIII in poly-protein constructs showed that ¹FNIII is stabilized in the ¹⁻²FNIII context, and this was interpreted as evidence for a direct ¹FNIII–²FNIII interaction (Oberhauser et al, 2002). Interestingly, ¹FNIII and ²FNIII are connected in fibronectin through a relatively long, 21-amino-acid linker, whereas almost all other type III domains are connected by short, 2- to 4-amino-acid linkers.

Here, we present structural and biochemical evidence for the existence of a cryptic site in ¹⁻²FNIII that interacts with the N-terminal FN 30 kDa fragment (FN30 kDa). In contrast to canonical type III domains, the first β-strand of ²FNIII is disordered in solution, although the remaining ²FNIII structure is highly similar to other type III protein domains. The linker connecting ¹FNIII with ²FNIII is flexible and includes the disordered ²FNIII region. The presence of a direct ¹FNIII–²FNIII interaction was confirmed using isolated domains and intact ¹⁻²FNIII, and a restrained model of the ¹⁻²FNIII structure was calculated using the ¹FNIII structure (Gao et al, 2003) and the new ²FNIII structure. Wild-type ¹⁻²FNIII exhibits only a weak association interaction with FN30 kDa. In contrast, a ¹⁻²FNIII variant that disrupts the ¹FNIII–²FNIII interaction shows significantly stronger association with the same fragment. A series of protein constructs of both ¹⁻²FNIII and ²FNIII, with variable lengths of the interdomain linker, show that the strong ¹⁻²FNIII–FN30 kDa interaction depends on the presence of that linker. This leads directly to a model, where disruption of the ¹FNIII–²FNIII complex by cell-generated tension creates an open ¹FNIII–²FNIII conformation that initiates fibrillogenesis by interaction with FN30 kDa.

Results

²FNIII structure

Over 21 000 FNIII domain entries exist currently in the SMART database (http://smart.embl.de), whereas over 80 distinct entries in the RCSB protein structure database contain one or more FNIII domains. The canonical structure of this domain type, a three-stranded and a four-stranded β-sheet in β-sandwich arrangement, is well understood. Unexpectedly, backbone dynamics studies (Supplementary Figure 1B and C) of the canonical ²FNIII domain indicated that approximately 20 amino acids of the ²FNIII N-terminus are largely flexible and unstructured. This N-terminal amino-acid region corresponds to the first β-strand (β-strand A) of the canonical FNIII structure (Supplementary Figure 1A). ²FNIII is, to the best of our knowledge, unique among the structurally characterized members of this domain class in lacking a secondary structure element in its native form.

The high-resolution solution structure of ²FNIII was determined using standard triple-resonance NMR methods (Figure 1A and B). The final ensemble of structures was calculated from 2827 distance, geometry, and global orientation restraints, with an average of approximately 36 restraints per residue of the structured core domain (Supplementary Table I). The resulting ensemble precision is 0.20±0.04 Å for all backbone and 0.43±0.09 Å for all heavy atoms of non-mobile residues ({¹H–}¹⁵N NOE⩾0.6), with over 94% of those residues in the most favored region of the Ramachandran plot. The overall structure of ²FNIII, other than β-strand A, closely resembles that of canonical FNIII domains. Structure alignments show that the backbone of essentially all structured ²FNIII residues can be superimposed with equivalent residues of other FNIII domains with average r.m.s. deviations smaller than 2 Å (Figure 1C and Supplementary Table II). The largest deviations are observed in the middle of the strongly twisted β-strand G (Figure 1D), which, in canonical FNIII domains, forms close contacts with the missing β-strand A.

Figure 1.

Solution structure of the ²FNIII domain and comparison with other type III domains. Shown here are (A) the 25-structure ensemble of ²FNIII structures. Purple-colored areas correspond to secondary structure elements. (B) Schematic stereo-representation of the minimized average structure of ²FNIII. β-strands are numbered according to canonical type III structures. (C) Superpositions of ²FNIII (purple) with human FN ⁸FNIII (turquoise) and avian tenascin ⁶FNIII (yellow). (D) The β-strands F and G of ²FNIII (purple) and human FN ⁸FNIII (turquoise) are shown here, as well as β-strand A of the latter. The side chains of the ²FNIII amino-acid residues likely responsible for the ²FNIII structural divergence (I802 and L803) are shown in purple. Also shown (turquoise) are the structurally equivalent ⁸FNIII residues (P1318-R1320) and L1242, which anchors the A and G β-strands.

Sequence alignment of the closest ²FNIII structural homologs and all identifiable ²FNIII domains from fibronectin orthologues (Figure 2) revealed a likely cause of the strand A disorder. In all ²FNIII domains, one of three hydrophobic residues that anchor β-strand A to β-strand G is substituted by a glutamate residue, E726. The remaining two hydrophobic residues are present (V728 and I731) but apparently insufficient to maintain an ordered structure. The volume occupied by the missing hydrophobic residue in canonical FNIII domains (L1242 in Figure 1D) is filled by two bulky hydrophobic residues of β-strand G in ²FNIII (I802 and L803 in the same figure); these do not follow the alternating side chain orientation rule of typical β-strands, and are thus responsible for the strong β-strand twist observed. As shown in Figures 1D and 2, other FNIII domains have one or no hydrophobic residues at the same position. We expect that, collectively, these substitutions cause disorder in the N-terminal β-strand of ²FNIII, but stabilize the remaining fold.

Figure 2.

CLUSTAL-W sequence alignment of (A) ²FNIII and the five closest structural homologs of ²FNIII as identified by Dali (Holm and Sander, 1998), and (B) ¹⁻²FNIII sequences from different FN proteins. The residue spans of ¹FNIII and ²FNIII are denoted by brackets, and the positions of the two amino-acid substitutions engineered in this study, K669A and D767A, are indicated by arrows. Highlighted by yellow bars are the three ²FNIII residues (E726, I802 and L803) identified in our analysis as possibly responsible for the disordered ²FNIII N-terminus, and their sequence counterparts in other domains. Highlighted in a purple box are the residues of the disordered ¹FNIII–²FNIII linker in all FN molecules. Shown in a cyan box are residues of the interdomain linker removal of which adversely affects the ¹⁻²FNIII–FN30 kDa interaction.

¹FNIII–²FNIII interaction

The flexible N-terminal region of ²FNIII increases the length of the ¹FNIII–²FNIII linker to approximately 34 residues between structured domains. If flexible, this linker effectively uncouples the molecular motions of the two domains and allows the adoption of completely independent orientations in the absence of any interdomain interactions. We tested ¹FNIII and ²FNIII for such interactions by monitoring their ¹H–¹⁵N HSQC spectra for chemical shift perturbations upon addition of unenriched ²FNIII or ¹FNIII, respectively.

Upon titration, several ¹FNIII and ²FNIII resonances were perturbed significantly in a dosage-dependent manner (Supplementary Figure 2A–D), consistent with a weak interdomain interaction with dissociation rates that are fast on the NMR timescale. Fitting the chemical shift changes for different domain ratios and concentrations enabled us to estimate the K_d as approximately 100 μM for this interaction under the experimental conditions used (Supplementary Figure 3). The observed perturbations were subsequently mapped on the domain amino-acid sequences (Supplementary Figure 4A and B) and tertiary structures. The ¹FNIII perturbations primarily localize along the β-strand connecting loops and turns at the C-terminal end of the molecule, while those of ²FNIII form a largely continuous surface on the exposed side of the four-stranded β-sheet (β-strands C, D, F and G).

A number of the larger perturbations involve charged ¹FNIII and ²FNIII residues, suggesting that electrostatic forces contribute substantially to this interaction. ¹FNIII has an overall positive charge under physiological conditions (pI 9.45), whereas ²FNIII is negatively charged (pI 3.79). To determine whether the interaction observed is specific between these two domains or simply a result of the opposite charges, we performed HSQC titration experiments using ¹FNIII and the negatively charged ³FNIII under conditions identical to those used for the ¹FNIII–²FNIII titrations. No observable changes could be detected in the HSQC spectra upon mixing ¹FNIII/³FNIII (data not shown), thus indicating that the observed ¹FNIII–²FNIII interaction is specific for this domain pair. In addition, significant chemical shift perturbations could be detected in the ¹FNIII/²FNIII HSQC spectra at physiological ionic strength (data not shown).

Previous studies in FNIII domain pairs have shown a significant reduction of molecular motions in the second domain compared with isolated FNIII domains (Carr et al, 1997; Spitzfaden et al, 1997; Altroff et al, 2004). Heteronuclear {¹H–}¹⁵N NOE and ¹⁵N T₂ data collected for ¹⁻²FNIII (Supplementary Figure 5) show that this is not the case for ²FNIII, since the N-terminal region of this protein, as well as the interdomain linker, remain flexible. Comparison of the backbone amide chemical shifts of ¹FNIII and ²FNIII, isolated and in the ¹⁻²FNIII context, shows that the interdomain interaction is present in this construct (Supplementary Figure 4C and D). The chemical shift perturbation patterns observed in ¹⁻²FNIII best match those of ²FNIII in the presence of ¹FNIII. ¹FNIII perturbation patterns are relatively complicated, presumably due to the unstructured interdomain linker. The unstructured linker does not affect ²FNIII chemical shifts as its N-terminus is already flexible.

We performed simple calculations of the effective local ¹FNIII and ²FNIII concentrations when connected by the linker; single ¹FNIII and ²FNIII molecules were assumed to be present inside a sphere with a diameter that is the sum of the diameters of gyration of ¹FNIII (26.6 Å) and ²FNIII (24.6 Å), plus the average end-to-end distance of a flexible 34-amino-acid chain. The latter was calculated as approximately 37.6 Å (C_α–C_α distance) from a large library of simulated linker chains. Local ¹FNIII and ²FNIII concentrations were thus calculated to be approximately 4.5 mM, which, given an estimated K_d of 0.1 mM, implies that over 85% of ¹FNIII molecules would be complexed with ²FNIII in the ¹⁻²FNIII context. In practice, this estimate of the local concentration is expected to be low, since the linker will probably not be able to access all possible conformations. Note that we have only considered intrachain interactions in solution, because intermolecular interactions are considered to be unlikely under the conditions used. Analytical gel filtration of ¹⁻²FNIII (Supplementary Figure 6) shows the presence of a single species; ¹⁵N T₂ relaxation rates of ¹⁻²FNIII (Supplementary Figure 5B) and analytical ultracentrifugation experiments (Supplementary Figure 7A) also indicate that this species is monomeric in solution.

¹⁻²FNIII structure

The extent of chemical shift perturbations caused by the ¹FNIII–²FNIII interaction is small, suggesting that the structures of the two domains remain highly similar to those calculated in the absence of this interaction. A possible approach toward determining the complex structure is then to use a rigid-body docking protocol to compute restraint models of intermolecular complexes (Clore and Schwieters, 2003; Dominguez et al, 2003). These protocols utilize chemical shift perturbations to delineate the interaction surfaces, and residual dipolar coupling (RDC) restraints measured for both complexed components to provide global orientation. Due to the ambiguous nature of such distance restraints, no specific information is entered into the calculation concerning interdomain interface packing. Despite the absence of specific packing information, the final model is often highly similar (backbone r.m.s. deviations ∼1 Å) to models derived from traditional methods (Clore and Schwieters, 2003).

This approach was used to calculate the structure of ¹⁻²FNIII. Ambiguous NOE restraints were applied between residues of ¹FNIII and ²FNIII that exhibit significant chemical shift perturbations and satisfy the solvent accessibility and surface continuity criteria described previously (Clore and Schwieters, 2003). Shift perturbations from both the ¹FNIII/²FNIII titrations and ¹⁻²FNIII (Supplementary Figure 4) were considered in order to reach a consensus. The final residues selected for distance restraints were S621, S625, K669, G674, V675, T698 and S701 for ¹FNIII, and R751, E753, E755, D767, N789, Q799 and S800 for ²FNIII. In addition, ¹D_NH RDC restraints were collected in a strained polyacrylamide medium in the context of ¹⁻²FNIII and used in the structure calculation. One hundred structures were calculated and the 50 lowest-energy models were retained for further analysis. The final family of structures consisted of two distinct population ensembles that fitted the experimental data equally well (Figure 3A and B). All ambiguous NOE restraints were satisfied in the vast majority of the calculated models, and the RDC restraint R-factors (Clore and Garrett, 1999) increased only slightly in the complex compared with those computed for isolated domains (¹FNIII isolated R-factor ∼14.5%, complex R-factor ∼16.8%; ²FNIII isolated R-factor ∼9%, complex R-factor ∼9.8%).

Figure 3.

(A, B) Representative structures of the two different ¹⁻²FNIII forms present in the structure calculation. Occupancy of form A in the final ensemble was 78%, whereas that of form B was 22%. Side chains for two residues, K669 and D767, involved in an electrostatic interaction in form A but not form B, are shown. D767 is also involved in electrostatic interactions with K672 in population A, and participates in an extended hydrogen bond and electrostatics network on the ¹FNIII binding interface of ²FNIII. The interdomain linker is schematically represented as a dashed line. (C, D) ¹⁻²FNIII structure. Shown here are: the final 39-structure ensemble of ¹⁻²FNIII structures displaying (C) or omitting (D) the flexible ¹FNIII–²FNIII linker. Secondary structure elements are colored cyan and purple for ¹FNIII and ²FNIII, respectively.

The two populations present in the final ensemble are related through an approximately 180° rotation and a small translation of the two domains with respect to each other. Due to the two-fold degeneracy present in the RDC restraints, the two models could not be discriminated by these initial data. Model A in Figure 3 represents 78% of the final family of structures (population A), whereas model B corresponds to the remaining 22% (population B). Two different approaches were used in order to determine which of the two population ensembles corresponds to the correct ¹⁻²FNIII structure. First, we attempted to locate and disrupt interdomain interactions unique to one particular model, by designing specific amino-acid residue substitutions. Second, we collected additional RDC restraints for ¹⁻²FNIII in a polyethylene glycol (PEG)/hexanol medium to break the two-fold symmetry inherent in the RDC data.

An electrostatic (salt-bridge) interaction between residues K669 of ¹FNIII and D767 of ²FNIII is present only in population A (Figure 3A). The same residues in population B are exposed to solvent and are unlikely to contribute any stabilizing interactions. K669 experiences significant chemical shift perturbations in the ¹FNIII–²FNIII interaction (Supplementary Figure 4), consistent with the structural interactions seen in population A. Domain variants with alanine substituting these two amino acids were created and tested for interdomain interactions in our HSQC chemical shift perturbation assay (Supplementary Figure 2E–H). Both of these variants appear to reduce the chemical shift change upon domain interaction for given protein concentrations and ¹FNIII/²FNIII ratios. The dissociation constant for ¹FNIII K669A and wild-type ²FNIII was estimated to be approximately 10-fold higher than that of the wild-type domains (Supplementary Figure 3), whereas shift perturbations for the interaction of ²FNIII D767A with wild-type ¹FNIII were too small to allow an accurate K_d estimation. When ¹FNIII K669A was added to ²FNIII D767A, no observable chemical shift perturbations were observed (data not shown), and analytical gel filtration experiments with a ¹⁻²FNIII variant that contains both amino-acid substitutions show that the great majority of this ¹⁻²FNIII construct is in a conformation with significantly larger hydrodynamic radius than wild-type ¹⁻²FNIII (Supplementary Figure 6). NMR spectra of ¹⁻²FNIII K669A/D767A (referred to henceforth as ¹⁻²FNIII KADA) showed that this variant remains well folded and in a non-supramolecular aggregate state (data not shown); this was confirmed by analytical ultracentrifugation experiments (Supplementary Figure 7B). We expect that ¹FNIII and ²FNIII completely dissociate in this variant, thus corresponding to a ¹⁻²FNIII open conformation with an average hydrodynamic radius of approximately 110 Å, a two-fold increase over the 58 Å length of population A. In contrast, the K669A/D767A substitutions should not significantly affect the interactions in population B, thereby arguing against the validity of population B as a structural model of ¹⁻²FNIII.

We performed singular value decomposition fits of the PEG/hexanol RDC restraints to the isolated ¹FNIII and ²FNIII domains, as well as to structural models from populations A and B of our ¹⁻²FNIII structure ensemble. The latter models were chosen to have the smallest backbone r.m.s.ds from the average of their respective populations. The structures of the individual domains are in good agreement with the new RDC restraints, having r.m.s.ds from those restraints of approximately 2.9 and 1.75 Hz for ¹FNIII and ²FNIII, respectively (data not shown). The model of population A was also in good agreement, with an r.m.s.d. of approximately 3.1 Hz (3.9 Hz for the ¹FNIII part and 2.3 Hz for the ²FNIII part) and a correlation R² of experimental and calculated RDCs of approximately 97% (Supplementary Figure 8A). In contrast, the model of population B could not be fit well using the same procedure, having an r.m.s.d. to the data of over 10 Hz and a correlation R² of only 67% (Supplementary Figure 8B). Therefore, both mutagenesis and NMR data indicate that population A is a better representation of the ¹⁻²FNIII structure.

The 39-structure ensemble of population A is shown in Figure 3C and D, and a structural analysis is presented in Supplementary Table III. ¹FNIII residues on the loops connecting β-strands A and B, and E and F interact with ²FNIII residues located on the strongly twisted four-stranded β-sheet, creating an overall appearance of ¹FNIII being ‘wedged' along an N-terminal-to-C-terminal axis on the curved ²FNIII β-sheet. The interaction surface of the two domains is relatively small, approximately 540 Ų. Complex formation introduces only a relatively small change in the overall protein chain direction, as the N- and C-terminal direction vectors of ¹⁻²FNIII differ by approximately 32°. This is comparable to the 10–40° angle formed by canonical type III domain pairs (Leahy et al, 1996; Spitzfaden et al, 1997; Altroff et al, 2004). The length of the interdomain linker is almost invariable in diverse species (Figure 2) and is considerably larger than is necessary to bridge the ¹FNIII C-terminus and the ²FNIII N-terminus in this conformation. Indeed, calculations based on our model suggest that a linker length of about 15 residues would be sufficient to allow the observed ¹FNIII–²FNIII interaction.

The almost 10-fold decrease in affinity of our K669A substitution (Supplementary Figure 3) corresponds, in energetic terms, to a loss of approximately 1.3 kcal/mol in stabilization energy. This value is similar to that previously reported in proteins (Horovitz et al, 1990) and model peptides (Mayne et al, 1998) for disruption of single salt-bridge interactions. Nevertheless, the ¹FNIII–²FNIII interaction is still present in this variant both for isolated domains (Supplementary Figure 2E and F) and ¹⁻²FNIII K669A (data not shown). In contrast, the D767A substitution has an effect larger than that expected for a single salt-bridge disruption, effectively abolishing the interdomain interaction (Supplementary Figure 2G and H). Indeed, in addition to K669, D767 can also interact favorably with K672 and possibly K655 in our model. Further, D767 participates in an extended hydrogen bond and electrostatics network on the surface of the twisted β-sheet of ²FNIII; R751 and Q799, two other ²FNIII residues that show significant perturbations upon ¹FNIII binding are also involved. Therefore, it is likely that D767 is a key residue for maintaining the ²FNIII surface in a conformation favorable for ¹FNIII interaction. D767 is conserved or conservatively substituted in all FN proteins (Figure 2), whereas K669 is absent in some of them.

¹⁻²FNIII interaction with FN30 kDa

Binding of ¹FNIII and ²FNIII to the N-terminal FN fragment has been shown previously (Aguirre et al, 1994; Hocking et al, 1994; Sechler et al, 2001), although only the ²FNIII interaction has been demonstrated under native conditions. We performed surface plasmon resonance (SPR) experiments with immobilized FN30 kDa under physiological conditions to determine whether the observed ¹FNIII–²FNIII interaction affects binding to FN30 kDa. Wild-type ¹⁻²FNIII interacts with FN30 kDa weakly, with a K_d of approximately 85 μM, as estimated from equilibrium analysis (Figure 4A and Supplementary Figure 9A). In contrast, ¹⁻²FNIII KADA, an open conformation of ¹⁻²FNIII, binds to FN30 kDa strongly (Figure 4A and Supplementary Figure 9C). Kinetic analysis of binding by this variant fitted a simple 1:1 Langmuir model with a dissociation constant of approximately 1.4 nM (k_on 4.04 × 10⁴ M⁻¹s⁻¹, k_off 5.56 × 10⁻⁵ s⁻¹).

Figure 4.

¹⁻²FNIII–FN30 kDa interaction. Shown here are sample SPR sensogram traces of (A) the interaction of wild-type ¹⁻²FNIII and ¹⁻²FNIII KADA (5 μM protein concentration) and (B) the interaction of different ²FNIII linker length variants (20 μM protein concentration) with immobilized FN30 kDa. Data points close to injection start and end (denoted by red vertical lines) were removed for clarity. (C, D) Response levels at a time point 50 s after the end of the injection for the interaction of (C) different ¹⁻²FNIII variants (5 μM protein concentration) and (D) different ²FNIII variants (20 μM protein concentration) with immobilized FN30 kDa. (E) The amino-acid sequences of the interdomain linker for the different ²FNIII and ¹⁻²FNIII variants are shown here. The ¹FNIII end and the start of the structured part of ²FNIII are denoted. Highlighted in cyan are linker residues, removal of which significantly affects FN30 kDa binding.

Further testing under the same conditions showed no appreciable binding of ¹FNIII and the interdomain linker (data not shown) to immobilized FN30 kDa, whereas ²FNIII bound with intermediate kinetics and affinity (Supplementary Figure 9B). However, a series of ¹⁻²FNIII KADA and ²FNIII constructs with variable lengths of the interdomain linker (Figure 4E) exhibited binding that decreased as the linker sequence was truncated from the N-terminus (Figure 4B–D). This trend is observed in both sets of constructs, suggesting that the effect is not due to the KADA substitutions but to the presence of the linker sequence. Removal of ¹FNIII also affects binding adversely, as full-length ¹⁻²FNIII KADA shows stronger binding affinity compared with a ²FNIII variant with the entire interdomain linker (Figure 4). Thus, it is likely that the high affinity of ¹⁻²FNIII KADA for FN30 kDa is the result of cooperativity along multiple, independently weak binding sites. Interestingly, our variable linker constructs delineate a specific series of residues (Figure 4E), removal of which significantly decreases binding affinity to FN30 kDa. These residues form a generally conserved patch (Figure 2) in the middle of the linker, indicating that the interaction described here is of biological significance.

Discussion

An understanding of fibrillogenesis requires knowledge of the structural states and interactions of FN and FN components, both in solution and in fibrils. Although type III domains have been extensively studied both in FN (Leahy et al, 1996; Sharma et al, 1999; Gao et al, 2003) and in other systems (Leahy et al, 1992; Bisig et al, 1999), ²FNIII constitutes a unique case. It represents a well folded and stable (Litvinovich and Ingham, 1995) type III domain but with a flexible β-strand A in the native form. In spite of this flexibility, we have demonstrated that a ¹FNIII–²FNIII interdomain interaction orients the two domains in FN in a way that maintains the chain direction. Recognition of residues responsible for abnormal chain deviations could be important for understanding the role of type III domains that form structural blocks in large proteins (Campbell and Spitzfaden, 1994).

The interaction between ¹FNIII and ²FNIII presented here has been previously alluded to in mechanical unfolding (Oberhauser et al, 2002) and in vitro functional experiments (Chernousov et al, 1991; Morla and Ruoslahti, 1992; Sechler et al, 2001). Oberhauser et al (2002) showed, using AFM, that ¹FNIII in the ¹⁻²FNIII context exhibits greater mechanical stability than ¹FNIII attached to the titin immunoglobulin domain I27. If the time constant for ¹FNIII–²FNIII dissociation is smaller than the time constant in the mechanical unfolding experiment (∼40 ms), then ²FNIII could protect the C-terminal end of ¹FNIII from unfolding. Under the conditions used in our study, the lifetime of the interaction between isolated ¹FNIII and ²FNIII domains is relatively short (sub-millisecond); however, it is possible that, in the context of ¹⁻²FNIII, and under the conditions used for the AFM experiments, the interaction is slow enough to allow such stabilization.

Functional studies suggest that ¹FNIII and ²FNIII interact with FN30 kDa (Aguirre et al, 1994; Hocking et al, 1994; Sechler et al, 2001), an FN fragment essential for fibrillogenesis (Schwarzbauer, 1991). ²FNIII is also essential for robust fibril accumulation, whereas recombinant FN lacking ¹FNIII was reported to form normal fibrils (Sechler et al, 2001). Our results indicate that ¹⁻²FNIII and isolated ²FNIII bind FN30 kDa relatively weakly in their native forms. In contrast, the open ¹⁻²FNIII conformation represented by our KADA variant binds FN30 kDa strongly and almost irreversibly. A truncation analysis of this open conformation showed that all three components of this construct (¹FNIII, interdomain linker and ²FNIII) are necessary to achieve tight binding, although it is possible that in vivo some of these requirements, notably that of ¹FNIII, could be relaxed due to the presence of additional interacting domains. We have established a strong correlation in vitro between the presence of the interdomain linker and FN30 kDa binding. Of particular interest is a series of generally conserved residues in the middle of this linker (Figures 2 and 4E) that cannot be explained in structural terms as the entire linker is flexible. Removal of these residues decreases FN30 kDa binding in vitro.

A possible model of fibril formation and elongation emerges from our results and previously published data (Figure 5). ¹⁻²FNIII exists in the closed conformation in the solution state of FN and the weak interaction between ¹⁻²FNIII and FN30 kDa might help maintain the FN globular structure (Figure 5A). During fibril formation, FN binds to cell surface receptors and is stretched by cell-generated tension (Figure 5B). This disrupts the globular structure and causes ¹FNIII and ²FNIII to dissociate; subsequently, the ¹⁻²FNIII open conformation interacts strongly with the FN30 kDa region of other FN molecules, thereby creating a FN homodimer along the FN N-terminus (Figure 5C). As FN is disulfide crosslinked at the C-terminus, this FN self-association interaction could potentially create a FN fibril that, upon elongation, is rendered insoluble. Thus, our proposed mechanism places a regulatory role on the ¹FNIII–²FNIII interaction at the initiation step of fibrillogenesis. The presence of this interaction would ensure that fibril formation does not occur spontaneously, but that, through the requirement for tension-induced dissociation, it is under firm cellular control. We believe that this model of weakly interacting domains, dissociating to uncover cryptic sites, may well also be applicable to other stress-sensing molecules.

Figure 5.

Possible fibrillogenesis mechanism: FN molecules exist in solution, with the ¹⁻²FNIII domain pair in closed conformation, likely interacting with the FN N-terminus as part of a larger globular structure (A). Under tension, the FN globular structure and the ¹FNIII–²FNIII interaction are disrupted (B, C). This allows the ¹⁻²FNIII open conformation to strongly associate with the N-terminus of other FN molecules (C) and, along with the disulfide crosslinks at the FN C-terminus, create FN fibrils. The different domain types of FN are shown in gold (FNI), purple (FNII) or cyan (FNIII). The different FN fragments shown correspond to: FN30 kDa, ¹⁻⁵FNI; gelatin binding domain, ⁶FNI-¹⁻²FNII-⁷⁻⁹FNI.

Materials and methods

Protein preparation and analysis

Gene fragments encoding human FN residues 608–701 (¹FNIII), 721–809 (²FNIII), 700–809 (²FNIII₃₄), 712–809 (²FNIII₂₂), 729–809 (²FNIII₅), 734–809 (²FNIII₀), 810–900 (³FNIII) or 608–809 (¹⁻²FNIII) were cloned as GST fusions using a pGEX-6P-2 expression vector (GE Biosciences). Modified genes with amino-acid substitutions or internal truncations were created using a PCR-based method and cloned in a modified pET-16 expression vector (Novagen) incorporating a 3C protease cleavage site. These genes include the ¹⁻²FNIII K669A/D767A double mutant (¹⁻²FNIII KADA), as well as variants of this double mutant with linker truncations, ¹⁻²FNIII KADA₂₉ (FN residues 608–701 and 707–809), ¹⁻²FNIII KADA₁₉ (residues 608–701 and 717–809), ¹⁻²FNIII KADA₁₂ (residues 608–701 and 724–809) and ¹⁻²FNIII KADA₇ (residues 608–701 and 729–733). Details of the protein expression and purification protocols can be found in the supporting information. As prepared, ²FNIII includes six N-terminal non-FN residues (cloning artifacts, amino-acid sequence GPLGSH), whereas all other constructs include five N-terminal residues (GPLGS). The proteolytic N-terminal FN 30 kDa fragment (FN30 kDa) was purchased from Sigma.

Gel filtration chromatography was performed using a calibrated Superdex-75 analytical gel filtration column (GE Biosciences) equilibrated with PBS buffer. Analytical ultracentrifugation equilibrium experiments were performed on samples in PBS buffer using a Beckman Optima XL-A analytical ultracentrifuge. UV absorbance was monitored at 280 nm. The duration of the run was 48 h at 20 000 r.p.m. and it was conducted at 25°C. The data were fit to an ideal monodisperse model using the program Origin (OriginLab).

NMR experiments

All experiments were performed at 30°C using 11.7 T, 14.1 T, 17.6 T and 22.3 T spectrometers equipped with triple-resonance, triple-axis gradient probeheads (Soffe et al, 1995). NMR sample buffers corresponded to 20 mM NaCl, 20 mm NaPi pH 7.0 for ¹FNIII, ²FNIII, ³FNIII wild-type and variants, and 150 mM NaCl, 20 mM NaPi pH 7.0 for ¹⁻²FNIII wild type and variants. Details of the NMR experiment performed are provided in the supporting information. All assignments have been deposited in the BioMagResBank (http://www.bmrb.wisc.edu) under accession numbers 7127 and 7128 for wild-type ²FNIII and ¹⁻²FNIII, respectively.

NMR structure calculations

Structure calculations were performed using the XPLOR-NIH software package (Schwieters et al, 2003). ²FNIII calculations were based on established simulated annealing protocols. In contrast, ¹⁻²FNIII calculations employed a rigid-body docking protocol described previously (Clore and Schwieters, 2003), appropriately modified for two structured domains and a flexible linker in a single polypeptide chain. XPLOR-NIH was also used for short simulations of possible linker conformations under conditions that generated random coil conformations with favorable Ramachandran statistics. Details of these different types of calculations can be found in the supporting information. The structures and structure calculation restraints for ²FNIII have been deposited in the RCSB Protein Databank under accession numbers 2H41 and 2H45 for the average minimized structure and the 25-structure ensemble, respectively; the population A structure ensemble of ¹⁻²FNIII is deposited under accession number 2HA1, along with the calculation restraints.

Surface plasmon resonance

SPR experiments were performed on a BIAcore 2000 instrument (Biacore AB, Uppsala, Sweden). The FN30 kDa was immobilized on the dextran matrix of the sensor chip by amine coupling (Biacore, 1994) in 10 mM sodium acetate buffer at pH 5.0. All experiments were carried out at 25°C in HBS running buffer (20 mM HEPES at pH 7.4, 150 mM NaCl and 0.005% (v/v) Tween 20), with flow rates of 50 and 1 μl/min for the kinetic and equilibrium experiments, respectively. Regeneration of the sensor surface was achieved with 30-s exposure to 50 mM HCl. The equilibrium binding constant was measured for the binding of wild-type ¹⁻²FNIII by injecting 100 μl of this construct (at concentrations of 17–270 μM) over immobilized FN30 kDa at 1 μl/min until equilibrium was reached. Responses at equilibrium, corrected for background from control flow cells, were recorded as R_eq for individual injections. Scatchard analysis was used to evaluate the binding constants from triplicate data sets using linear regression, which gives –1/K_d as slope and R_max as intercept on the x-axis. The on- and off-rates (k_on and k_off) of ²FNIII and the ¹⁻²FNIII KADA variant binding to 30 kDa FN30 kDa were measured by injecting a series of sample dilutions (²FNIII: 20–80 μM and ¹⁻²FNIII variant 31–250 nM) over the immobilized fibronectin fragment. To reduce mass transport effects, the immobilization level was kept low (1000 RU) and the flow rate was kept high (50 μl/min) (Kortt et al, 1999). Due to the slow k_off for ¹⁻²FNIII KADA (5.6 × 10⁻⁵ s⁻¹), the dissociation rate was measured over 15 min to obtain a reliable value. The k_on and k_off were obtained by fitting the 1:1 Langmuir binding and the heterogeneous ligand models to components of the sensorgrams. Experiments of similar type at a specific protein concentration level were conducted for the different ²FNIII (20 μM protein concentration) and ¹⁻²FNIII KADA (5 μM protein concentration) variants with variable linker lengths, over FN30 kDa immobilized at 5000 RU to increase the sensitivity of the sensor. The sensogram response levels at a fixed point 50 s after the end of the injection, corrected for background from the control flow cell, were used as a measure of the relative binding affinities of the variants. The data were analyzed using the BIAevaluation software (Biacore, 1997) provided with the instrument.

Figures and notes

Amino-acid numbering for the different proteins corresponds to that in their respective entries; the accession numbers for the different FN proteins are as follows: Homo sapiens UniProt P02751, Brachydanio rerio FN3 Q6JAN2, B. rerio FN1a O93405, Xenopus tropicalis Q501R6, X. laevis Q91740, Bos taurus P07589, Rattus norvegicus P04937 and Mus musculus P11276. The FN protein of Gallus gallus used here corresponds to translation of the ensemble ENSGALT00000005654. The ²FNIII structural homologs identified correspond to the RCSB entries 1FNF (H. sapiens FN ⁸FNIII), 1QR4-B (G. gallus Tenascin ⁶FNIII), 1FNH (H. sapiens FN ¹³FNIII), 1TDQ-A (R. norvegicus Tenascin ³FNIII) and 1TEN (H. sapiens Tenascin ³FNIII). The subscript in the variable linker length ¹⁻²FNIII KADA and ²FNIII constructs denotes the length or the linker in amino acids. Using the same nomenclature, our original constructs would correspond to ¹⁻²FNIII KADA₃₄ and ²FNIII₁₃. The exact amino-acid sequences of all variants are described in the methods and are also shown in Figure 4E.

Supplementary Material

Supplementary Information