High affinity streptococcal binding to human fibronectin requires specific recognition of sequential F1 modules.

Fibronectin (Fn) binding by the Streptococcus pyogenes protein SfbI has been shown to trigger integrin-dependent internalization of this pathogen by human epithelial and endothelial cells. Here, using nuclear magnetic resonance spectroscopy and isothermal titration calorimetry in a dissection approach, the basis for the specificity and high affinity of the interaction between the N-terminal domain of Fn and SfbI is revealed. Each of the five Fn type 1 modules is directly involved in the interaction and is recognized by short consecutive motifs within the repeat region of SfbI. Crucially, these motifs must be combined in the correct order to form a high affinity ligand for the N-terminal domain of Fn.

Proteins that are natively unstructured (or intrinsically disordered) or those containing significant disordered regions are now recognized to account for a substantial fraction of all proteins (1). The function of most of these non-classical proteins appears to be molecular recognition in regulatory and assembly processes. Unfolded proteins can bind specifically and rapidly to their targets through large interfaces, thereby undergoing a significant disorder-order transition upon binding (2,3). Another striking feature is the over-proportionate occurrence of tandem arrays of sequence repeats in intrinsically disordered proteins (4). Fibronectin (Fn) 1 -binding proteins (FnBPs) that are expressed on the surface of pathogenic Gram-positive bacteria contain functional regions possessing all these properties.
FnBPs belong to a class of adhesins called MSCRAMMs (microbial surface components recognizing adhesive matrix molecules) (5)(6)(7). The Fn binding activity of the FnBP SfbI from Streptococcus pyogenes and its allelic variant, PrtF1, was initially ascribed to the C-terminal third of the molecule where, depending on the bacterial strain, 1-6 Fn-binding repeats (FnBRs) of 37 amino acid residues are found ( Fig. 1A) (8 -10). The FnBRs of SfbI are homologous to tandem repeats in FnBPs of other streptococci and Staphylococcus aureus, and new boundaries for FnBRs of both streptococci and staphylococci have recently been suggested based on structural data (7,11). FnBRs are intrinsically disordered and undergo a conformational change on binding to Fn (12)(13)(14).
Although S. pyogenes was considered previously to be an exclusively extracellular pathogen, it has been shown more recently both to adhere to and invade human epithelial and endothelial cells (15)(16)(17). The FnBP SfbI (18,19) has been shown to efficiently mediate cellular internalization; in the uptake process Fn acts as a bridge between the pathogens and integrin receptors on the cell surface (20). Internalization may contribute to the persistence of S. pyogenes in antibiotic-treated individuals (21) and aid hematogenous dissemination and/or evasion of the host immune system.
Fn is present in a soluble form in human plasma and other body fluids and in an insoluble form in the extracellular matrix. Fn has a mosaic structure and is composed of independently folded Fn type 1, type 2 and type 3 (F1, F2, and F3) modules (22) (Fig. 1C). FnBRs of S. aureus, S. pyogenes, and Streptococcus dysgalactiae bind to the N-terminal domain (NTD) of Fn (8,23,24), which is composed largely of a string of five F1 modules ( 1-5 F1). High resolution structures of subdomains of the NTD have been determined using nuclear magnetic resonance (NMR) spectroscopy (25)(26)(27). The consensus fold of the F1 module consists of a double-stranded anti-parallel ␤-sheet (strands A and B) folded over a triple-stranded anti-parallel ␤-sheet (strands C, D, and E).
The three-dimensional structure of an FnBP-derived peptide (B3 from S. dysgalactiae) in complex with 1 F1 2 F1 from the NTD of Fn was recently determined (11). The structure revealed an anti-parallel orientation of the binding partners with the peptide forming additional ␤-strands at the edge of the triple-stranded ␤-sheets of the two F1 modules. This novel binding mechanism was named a "tandem ␤-zipper" (11). Based on these structural data and sequence analyses, it was proposed that each FnBR forms an even longer tandem ␤-zipper when bound to intact NTD (11).
The main aim of this work is to address two key elements of the tandem ␤-zipper model of Fn/NTD interactions. First, we demonstrate that each of the five F1 modules of the NTD is specifically recognized by short, consecutive (i.e. tandemly arranged) amino acid sequences in FnBRs. Second, functional FnBRs are shown to be linear arrays of specific F1-binding motifs that need to be arranged in the correct order to bind to the NTD with high affinity. Data presented here also support the idea that bacterial peptides bind to all F1 modules of the NTD through a common mechanism, the ␤-zipper. This paper explains how a natively disordered functional domain of a bacterial protein forms a highly efficient ligand for a modular host protein.
Synthetic Peptides-PyOT5, PyOn5, and PyFo5 were synthesized using standard N-(9-fluorenyl)methoxycarbonyl technology and purified by reverse-phase HPLC. All other peptides were purchased from Alta Bioscience (Birmingham, UK) and purified by reverse-phase HPLC, if required, on a C12 silica column (10 m, 90 Å, 25 ϫ 1 cm) with an elution gradient of 10 -20% acetonitrile in 5 mM ammonium carbonate buffer over 15 min at a flow rate of 3 ml/min. The N and C termini of all peptides were capped by acetylation and amidation, respectively.
Isothermal Titration Calorimetry (ITC)-The experiments were carried out with a VP-ITC microcalorimeter (MicroCal Inc., Northampton, MA). In a standard experiment the cell contained 1.4 ml of a solution of an Fn fragment, and the syringe contained 295 l of a solution of an appropriate bacterial peptide at a concentration that was 12-17 times higher than the protein concentration in the cell. The protein concentration in turn was chosen to comply with a c value (29) between 20 and ϳ1000, where c ϭ [cell]/K d , (where K d is the dissociation constant). Titrations using module pairs were carried out at 25°C in 10 mM sodium acetate buffer (pH 5.0) or, for PyTT5, 10 mM sodium/potassium phosphate buffer (pH 7.4). Titrations of Fib1 were performed in phosphate-buffered saline (8.1 mM Na 2 HPO 4 , 1.9 mM KH 2 PO 4 , 137 mM NaCl, 2.7 mM KCl, pH 7.4) at 25°C (except SfbI-5, which was at 37°C). Fib1 (a 30-kDa proteolytic Fn fragment) was purchased from Sigma. Both the cell and syringe solutions were degassed at 15°C for 20 min. The titrations were performed as follows. One preliminary injection was of 2 l, and 30 -44 injections were of 4 -6 l at an injection speed of 0.5 l/min. The stirring speed was 310 rpm; the delay time between the injections was 3-5 min. To take into account heats of dilution, blank titrations were performed by injecting peptide solution into buffer, and the averaged heat of dilution was subtracted from the main experiment. Data were analyzed using MicroCal Origin software, fitting them to a single binding site model. Peptide concentrations, which were defined by weighing freeze-dried peptide, were corrected to result in binding stoichiometries of 1:1. This concentration adjustment never exceeded Ϯ10%. As described previously (11), the association constants for the intact FnBR interactions were expected to be too high to be directly measurable by ITC at the low ionic strength conditions used to study F1 module pair binding. It was observed that the binding of FnBR-derived peptides to F1 module pairs is significantly weakened with increasing ionic strength. 2 Thus, experiments with full-length FnBRs were carried out at physiological ionic strength in order to obtain sigmoidal titration curves that could be deconvoluted (29). In addition, binding of SfbI-5 could only be analyzed at a higher temperature (37°C rather than 25°C), where the affinity was further attenuated, and the signal-tonoise ratio was increased (because the binding is exothermic).
NMR Spectroscopy-All NMR experiments were performed on spectrometers belonging to the Oxford Centre for Molecular Sciences with 1 H operating frequencies of 500, 600, and 750 MHz. The spectrometers are all equipped with Oxford Instruments superconducting magnets, OMEGA software and digital control equipment (Bruker Instruments), home-built triple-resonance pulsed-field-gradient probe-heads (30), and home-built linear amplifiers for 1 H, 15 N, and 13 C nuclei. Spectra were recorded at 25°C. For the assignment of 2 F1 3 F1, a sample containing 0.7 mM U-15 N-labeled 2 F1 3 F1 in 95% H 2 O, 5% D 2 O at pH 5.0 was used. A standard set of two-and three-dimensional spectra was recorded similar to the previously described procedure for U-15 N-labeled 1 F1 2 F1 (27). For the assignment of 2 F1 3 F1 bound to PyTT5, a sample containing 0.5 mM U-15 N-labeled 2 F1 3 F1 and 1.0 mM PyTT5 in 95% H 2 O, 5% D 2 O at pH 5.0 was used. 1 H, 15 N heteronuclear single quantum coher-ence ( 1 H, 15 N HSQC) spectral assignments for U-15 N-labeled 1 F1 2 F1 and U-15 N-labeled 4 F1 5 F1 at appropriate pH values were obtained using previously determined assignments (14,27). All NMR samples were unbuffered.
NMR Competition Experiment-Two identical samples of 600 l of 0.2 mM U-15 N-labeled 2 F1 3 F1 where prepared, and virtually identical 1 H, 15 N HSQC spectra were recorded. Freeze-dried aliquots of 0.12 mol of SfbII-FF3 were added to both samples, again resulting in virtually identical 1 H, 15 N HSQC spectra. One sample was complemented with 0.12 mol of freeze-dried unlabeled 2 F1 3 F1; to the other sample 0.12 mol of freeze-dried unlabeled 4 F1 5 F1 were added, and 1 H, 15 N HSQC spectra were recorded.

RESULTS
Dissection Approach-The interactions of three Fn type 1 module pairs ( 1 F1 2 F1, 2 F1 3 F1, 4 F1 5 F1) with peptides from FnBRs of SfbI were investigated by NMR spectroscopy and ITC ( Fig. 1, Table I). Peptides from S. pyogenes FnBRs were chosen according to the tandem ␤-zipper model (11) to bind to either a specific F1 module or module pair of NTD. Series of 1 H, 15 N HSQC spectra of 15 N-labeled F1 module pairs at increasing concentrations of peptide were acquired (HSQC titrations). Fn residues affected by peptide binding were identified on the basis of observed chemical shift changes of F1 module backbone amide resonances (chemical shift perturbation mapping (33)) ( Fig. 2A and B). K d values were derived directly from the chemical shift changes if the kinetics of binding were fast on the NMR time scale (34) (Fig. 2G). Other K d values were determined using ITC (Fig. 2F). For some complexes, where it was impossible to track the resonances in the HSQC titrations, spectra of fully bound 15 N-labeled module pairs were assigned using three-dimensional heteronuclear NMR spectroscopy. 1 F1 and 2 F1 Are Specifically Recognized by Short Consecutive Motifs of SfbI-The direct involvement of 1 F1 and 2 F1 in interactions with a bacterial peptide from S. dysgalactiae FnBB was demonstrated recently by the determination of the three-dimensional structure of 1 F1 2 F1 in complex with the B3 peptide (11). Here, the involvement of both modules in binding to the homologous peptide from SfbI is confirmed, and the specificity of F1-module recognition is demonstrated.
The peptide PyOT5 bound to 1 F1 2 F1 with a K d of 0.45 M (Fig. 2F) (11) similar to the K d for binding to B3 (35,36). In an NMR binding experiment significant chemical shift changes were induced in residues in both 1 F1 and 2 F1, indicating the direct involvement of both modules in binding to PyOT5 (Fig. 2C).
The tandem ␤-zipper model for NTD recognition is based on the existence of independent, consecutive motifs within FnBRs that interact with specific F1 modules. Therefore, PyOT5 was dissected into predicted single-module binding segments. PyOn5 and PyTw5, representing the C-and N-terminal halves of PyOT5, caused chemical shift changes exclusively in 1 F1 (Fig. 2, A and D) or 2 F1 (Fig. 2, B and E), respectively. The observed chemical shift differences resembled the perturbation patterns for each module on binding of PyOT5. The K d values for binding to 1 F1 2 F1 were 159 and 63 M for PyOn5 and PyTw5 (Fig. 2G), respectively. We conclude that 1 F1 and 2 F1 are recognized specifically by separate motifs of SfbI, resulting in an anti-parallel attachment of the peptide to the module pair 1 F1 2 F1.
3 F1 Is Involved in Binding of SfbI to the NTD-The direct involvement of 3 F1 in FnBR binding has to date not been demonstrated. The K d for the interaction of the predicted 2 F1 3 F1-binding peptide PyTT5 with 2 F1 3 F1 was 3.6 M (11). In NMR binding studies PyTT5 induced chemical shift changes in both 2 F1 and 3 F1 residues (Fig. 3A). To elucidate the role of 3 F1, PyTT5 was dissected into single-module binding halves (PyTw5 and PyTh5). The 2 F1-binding peptide, PyTw5, bound to the module pair with significantly lower affinity (K d ϭ 66 M) than the two-module binding peptide. The pattern of chemical shift perturbations in 2 F1 is very similar to that observed on binding of PyOT5 and PyTw5 to 1 F1 2 F1 (Fig. 2, C and E) and indicates specific recognition of 2 F1 (Fig. 3B). Only one residue in 3 F1, Gly-142, was significantly affected (using a 0.1 ppm cutoff) by PyTw5. Based on the previously determined struc-ture of the highly homologous (48% identity) 4 F1 5 F1 module pair (26), this residue is in a loop in 3 F1 that is expected to be in close proximity to 2 F1 residues. Thus, it is likely that binding of PyTw5 to 2 F1 causes limited chemical shift perturbations in 3 F1 due to indirect (intermodular) interactions. Direct evidence for an intermodule interface awaits the determination of the structure of 2 F1 3 F1.
The putative 3 F1-binding peptide, PyTh5, when added to 2 F1 3 F1 in excess, caused chemical shift changes that were confined to residues in 3 F1 (Fig. 3C). It should be noted that although the chemical shift changes are very small, their significance is clear from their comparison with the even smaller changes observed for 2 F1 residues. Moreover, the perturbation pattern induced by PyTh5 resembles the pattern for 3 F1 residues upon binding of the two-module binding peptide, PyTT5. Thus, it could be shown that 3 F1 is directly involved in interactions with FnBRs and is specifically recognized by the short peptide, PyTh5. The successful dissection of PyTT5 also further demonstrates the specificity of the 2 F1-binding motif in SfbI in that it binds neither 1 F1 nor 3 F1. The PyTh5/ 2 F1 3 F1 interaction was too weak for a K d to be measured. The 3 F1-binding segment, however, contributes significantly to the affinity of the two-module binding peptide, PyTT5. This result is an example where NMR provides unique, residue-specific informa- PyOT5, PyTT5, PyFF5, PyTT4, and PyFF1 are peptides designed to bind two consecutive F1 modules (PyOT5, 1 F1 2 F1; PyTT4, 2 F1 3 F1; PyFF5, 4 F1 5 F1); other peptides are designed to bind to one specific F1 module (PyOn5, 1 F1; PyTw5, 2 F1; PyTh5, 3 F1; PyFo5, 4 F1; PyFi5, 5 F1). In the peptide abbreviations, capitals are used for peptides designed to bind two modules (e.g. PyTT5 is the 2 F1 3 F1(TT)-binding peptide from the FnBR SfbI-5), whereas a combination of a capital and a lowercase letter is used for peptides designed to bind to single F1 modules (for example, PyTh5 is the 3 F1(Th)-binding peptide from SfbI-5). C, modular structure of Fn (one subunit). Module types 1, 2, and 3 are symbolized by pentagons, hexagons, and circles, respectively. The locations of the NTD and GBF, alternatively spliced sites (curved labels), the major integrin binding site (RGD), and the four constructs used for this paper are indicated. tion for a very weak but physiologically relevant interaction. Similar results were obtained for 2 F1 3 F1 binding of peptides from SfbI-4 (data not shown).
SfbI Contains 4 F1-and 5 F1-binding Motifs-Binding of the FnBR region of SfbI to 4 F1 5 F1 has been reported (11, 35), but the specific recognition of 4 F1 and 5 F1 has not yet been dem-  onstrated. Binding of the predicted 4 F1 5 F1-binding peptide (PyFF5) from SfbI-5 to 4 F1 5 F1 (K d ϭ 113 M (11)) resulted in chemical shift changes in both F1 modules (Fig. 3D). On dissecting the peptide into potential 4 F1-and 5 F1-binding peptides, PyFo5, the 4 F1-specific peptide, induced chemical shift changes primarily for residues in 4 F1 (Fig. 3E). Several chemical shift changes in 5 F1 could, however, be observed. These changes are in 5 F1 residues in the A-B double-stranded ␤-sheet and the D-E interstrand loop that were previously shown to form part of the intermodule interface between 4 F1 and 5 F1 (26). The effects observed for these residues are, therefore, likely to be transmitted over this interface rather than to arise from direct interaction with the peptide (14). PyFi5, predicted to interact with 5 F1, did not cause any significant chemical shift changes when added to 4 F1 5 F1 (Fig. 3F). This experiment was, however, limited by the poor solubility of PyFi5. Direct interaction of this 5 F1-binding region of SfbI-5 with 5 F1 is demonstrated by the observation that residues in 5 F1 (for example Trp-247 and Arg-241) undergo significant perturbations on binding of PyFF5 but not PyFo5. Our conclusion that PyFi5 residues recognize 5 F1 is also strengthened by the observation of a very significant enhancement in affinity of the 4 F1-binding peptide for 4 F1 5 F1 when the 5 F1-binding peptide is attached (Table I). Thus, both 4 F1 and 5 F1 are directly involved in the interactions with FnBRs and are specifically recognized by short motifs in FnBRs. Again, these experiments are consistent with an anti-parallel orientation of binding between the NTD and FnBRs.
Evidence for the Role of the EDT Sequence in 4 F1 Binding-A sequence alignment of streptococcal and staphylococcal FnBRs reveals that the glutamic acid residue in an EDT motif, which is conserved in the predicted 4 F1-binding motif in many Fn-BRs, is substituted with a lysine in FnBRs of SfbI (7). In a peptide from S. dysgalactiae FnBA, a Glu/Gln mutation in the EDT motif of the 4 F1-binding region almost completely abrogated the ability of that peptide to inhibit FnBPA (from S. aureus) binding to Fn (37). The relatively weak affinities of PyFF5 and PyFo5 for 4 F1 5 F1 (Table I) also provide strong evidence for the important role of the glutamic acid residue in this motif. To support this finding, SfbII-FF3, a homologous peptide from S. pyogenes SfbII (38) but containing the EDT motif, was used in a binding study with 4 F1 5 F1. SfbII-FF3 bound 4 F1 5 F1 with significantly enhanced affinity (K d ϭ 22 M).
Another putative 4 F1 5 F1-binding peptide, PyFF1, lacks the EDT motif altogether. It was shown to bind to 4 F1 5 F1 but with an even lower affinity than PyFF5 (Table I). PyFF1 binding (Fig. 4A) perturbs chemical shifts of residues in both 4 F1 and 5 F1, including the E-strand of 5 F1 (some distance from the interface with 4 F1; Fig. 4B). This is an important finding because the N terminus of PyFF1 (which contains a predicted 5 F1-binding motif) marks, according to the tandem ␤-zipper model, the N-terminal boundary of the FnBR region in SfbI (Fig. 1). In previous definitions PyFF1 was included in the adjacent functional region of SfbI, the UR segment (see "Discussion") (39,40).
Further Evidence for a Common Structural Theme in F1 Recognition-The chemical shift perturbation maps in Figs. 2-4 provide information regarding bacterial peptide binding sites on each of the F1 modules. For example, a comparison of HSQC spectra of 1 F1 2 F1 in the presence and absence of PyOT5, PyOn5, and PyTw5 (Fig. 2C-E) shows a clustering of the most significant H N and N chemical shift differences for residues in the E strands of both 1 F1 and 2 F1. Very similar chemical shift perturbation maps were observed for the interaction of 1 F1 2 F1 with the B3 peptide (which is homologous to PyOT5) (36).
There is a cluster of significant chemical shift perturbations for residues in the E-strand (by homology with other F1 modules) of 3 F1 when PyTT5 and PyTh5 bind ( Fig. 3A and C), the E-strand of 4 F1 on PyFF5 and PyFo5 binding, and the E-strand of 5 F1 when PyFF5 binds (Fig. 3, D and E). Although no three-dimensional data are available as yet for FnBR peptide binding to 3 F1, 4 F1, and 5 F1, these observations are consistent with a similar mechanism of binding of the FnBR peptides to 3 F1, 4 F1, and 5 F1 to that seen for binding to 1 F1 and 2 F1, i.e. a ␤-zipper. The involvement of A-strand residues observed when PyTw5 and PyFo5 bind to 2 F1 (Fig. 3B) and 4 F1 (Fig. 3E) is consistent with the close proximity of the A and E strands in the F1 module fold. PyFF1, unlike PyFF5 or PyFo5, does not induce significant chemical shift changes for residues at the N-terminus of the E-strand or in the A strand of 4 F1 (Fig. 4A). This finding further supports the predicted anti-parallel binding mode of the tandem ␤-zipper model in which the EDT motif would be found near the beginning of the E strand in proximity to the A strand (Fig. 4B).

F1 Modules and Module Pairs Are Specifically Recognized by FnBP Peptides-
The specificity of binding of peptides to 1 F1 versus 2 F1, 2 F1 versus 3 F1, and 4 F1 versus 5 F1 has been demonstrated above. However, a comparison of 2 F1 with 4 F1 and of 3 F1 with 5 F1 shows that these modules share 51 and 45% sequence identity, respectively. This seems to be reflected in a certain degree of homology between F1-binding motifs in FnBRs. In particular the important (E/K)DT motif (see above) is found in both the 2 F1-and 4 F1-binding regions (Fig. 5). Thus, in an even more stringent test of F1-binding specificity, binding (for example) of a 2 F1-binding peptide to 2 F1 (match binding) and to 4 F1 (mismatch binding) was compared.
As predicted from sequence analyses, a degree of "mismatch" binding to module pairs was observed for both 2 F1-and 4 F1binding peptides from SfbI. That is, PyTw5 induced chemical shift changes in 4 F1 residues (and 5 F1 residues close to the intermodule interface) when added to 4 F1 5 F1. PyFo5 induced chemical changes in 2 F1 residues when added to 1 F1 2 F1 or 2 F1 3 F1. PyTw5 bound with a significantly higher affinity to its match module 2 F1 (in 1 F1 2 F1 or 2 F1 3 F1) than its mismatch target 4 F1 (in 4 F1 5 F1; Table I). A quantitative analysis for match and mismatch binding by PyFo5 was hampered by difficulties in fitting binding data for the very weak binding to all module pairs (Table I). To allow further quantitative analysis, a similar approach to probing specificity was used for the higher affinity two-module binding peptides. PyTT5, PyFF5, and SfbII-FF3 all bound to their mismatch target but with considerably lower affinity (Table I). It was also shown that 4 F1 5 F1 efficiently competed with U-15 N-labeled 2 F1 3 F1 for binding to the 4 F1 5 F1-binding peptide SfbII-FF3 (Fig. 6).
Functional FnBRs; the Correct Order of F1-binding Motifs Is the Key to High Affinity-To test the validity of the dissection approach and the functional relevance of the measured specificities of F1 module recognition, we investigated the interaction of recombinantly expressed full-length FnBRs with Fib1, a proteolytic fragment consisting primarily of the NTD of human Fn (Table II). As reported previously, the 2-5 F1-binding FnBR SfbI-4 and the 1-5 F1-binding FnBR SfbI-5 both bound to Fib1 with relatively high affinities (K d values of 62 and 2 nM, respectively) (11). SfbI-1, the first FnBR from SfbI, that is predicted to bind 2-5 F1, was found to have a somewhat lower affinity for Fib1 (K d of 168 nM) than SfbI-4. This reflects the presence of the weak 4 F1 5 F1-binding segment PyFF1 (see above) in this repeat. All three binding processes were characterized by high exothermic enthalpies and unfavorable entropies (Table II), consistent with the considerable disorder-order conformational transition of FnBRs on Fn binding.
The boundaries for these three constructs were chosen based on the tandem ␤-zipper model for FnBR/NTD interactions to contain specific F1-binding motifs arranged in the correct order to bind 2-5 F1 (SfbI-1 and SfbI-4, Fig. 7A) or 1-5 F1 (SfbI-5, Fig.  7C). Using different boundaries, however, it is possible to design a native peptide SfbI-4.5 that contains the C terminus of SfbI-4 and the N terminus of SfbI-5 (Fig. 1B). Of course, like SfbI-4, this construct contains four F1 module binding segments, but they are now arranged in the wrong order to bind 2-5 F1 in a single NTD molecule (Fig. 7B). Choosing conditions similar to the SfbI-4 experiment, injections of SfbI-4.5 into a Fib1 solution only gave rise to small exothermic responses. The titration curve was linear and, therefore, could not be deconvoluted (Fig. 7B). Although it is not possible to infer from the data if four or only two F1 modules of Fib1 take part in binding to SfbI-4.5, the titration curve clearly demonstrates that SfbI-4.5 binds Fib1 with much lower affinity than SfbI-4 (Fig. 7, A  and B). DISCUSSION The striking similarities between FnBRs of different streptococcal and staphylococcal FnBPs had been noticed and discussed extensively (6). However, the lack of structural information for FnBRs hampered the interpretation of sequence alignments in terms of the binding sites in Fn. In addition, sequence comparisons could not always predict or explain the binding properties of apparently homologous repeats. The cross-reactivity observed in inhibition studies with FnBRs from different species was also difficult to interpret (5,35,41).
The first structural information for FnBP-Fn interactions paved the way to a better understanding of the molecular basis of this important host-pathogen interaction, and a novel mech- anism of protein:protein interaction for the binding of unstructured FnBRs to the NTD of Fn was proposed (11). This tandem ␤-zipper model is based on the following key elements. (a) Each of the five F1 modules of the NTD is specifically recognized by short, consecutive amino acid sequences in FnBRs. (b) A functional FnBR is a linear array of these F1-binding motifs, arranged in the correct order to bind 1:1 to the NTD in an anti-parallel fashion. (c) The common structural theme of F1 recognition is the ␤-zipper. This paper confirms the first two elements of the model and presents data consistent with the third.
PyTT5 bound specifically to 2 F1 3 F1, and PyFF5 bound specifically to 4 F1 5 F1 (Table I), as predicted. Within these peptides it was demonstrated that 2 F1-and 4 F1-binding motifs lie in the C-terminal half (Fig. 3). It was also demonstrated that the boundaries of the FnBRs we identified previously (11) are functionally relevant. That is, the choice of "incorrect" boundaries resulted in low affinity binding (Fig. 7). This is a key point because FnBPs have multiple FnBRs and, therefore, multiple 2 F1 3 F1-and 4 F1 5 F1-binding peptides. Taken together, these results confirm that larger regions of the bacterial and human proteins interact in the anti-parallel orientation that had been observed in our previous study of a 1 F1 2 F1/peptide interaction (11).
Using three different module pairs in NMR binding studies, the direct involvement of all five F1 modules of the NTD in FnBR (SfbI-5) binding was demonstrated. Single-module binding peptides induced chemical shift changes in 1 F1, 2 F1, 3 F1, or 4 F1 residues. Evidence for 5 F1 involvement comes from chemical shift changes in 5 F1 residues on binding of a 4 F1 5 F1binding peptide and its higher affinity of binding when compared with a 4 F1-binding peptide. 3 F1-and 5 F1-binding motifs did not show any significant binding activities when on their own but greatly enhance binding to module pairs when linked with 2 F1 or 4 F1 binding sequences, respectively. Reflecting the sequence homology between F1 modules, peptides containing 2 F1-or 4 F1-binding motifs all displayed 4 F1 and 2 F1 binding (mismatch) activity. However, in the context of an intact FnBR, this mismatch binding activity is very unlikely to be functionally relevant.
It should be noted that the chemical shift perturbation experiments (Figs. 2-4) report changes in backbone amide chemical shifts. Thus, although they are useful in identifying a binding surface and can detect backbone atoms involved in interactions (e.g. in ␤-zipper formation), specific side chains that play an important role in the binding may not always be revealed. Although a full study of bacterial residues important for the binding to F1 modules using mutagenesis is beyond the scope of this study, the importance of the EDT sequence in 4 F1 binding is suggested by the comparison of sequences and binding affinities of three 4 F1 5 F1-binding peptides (two from SfbI and one from SfbII). Fig. 4 shows that the lack of this EDT sequence (as in PyFF1) results in fewer significant chemical shift perturbations in the N terminus of the E-strand and in the A-strand, which also suggests the location of the binding site of the EDT sequence.
The role of the unique 1 F1-binding motif at the C terminus of SfbI-5 is not clear. This highly conserved segment occurs once in the most C-terminal FnBRs of FnBPs, with the exception of FnBPA and FnBPB from S. aureus and FnZ from Streptococcus equisimilis (7). Potentially, several copies of Fib1 can be bound to the FnBR region of FnBPs (11). It is possible that NTD binding to the high affinity (Table II) C-terminal FnBR recognizing 1-5 F1 may facilitate binding of more Fib1 to the lower affinity FnBRs recognizing 2-5 F1.
Chemical shift perturbation maps presented here support the idea that all modules in the NTD are recognized by segments of FnBRs in a similar fashion. The observed chemical shift changes upon peptide binding, particularly in the E-strand of F1 modules, are consistent with the bacterial motifs binding the individual modules primarily by forming an additional ␤-strand along the E-strand of each F1 module, as was observed previously for 1 F1 2 F1 binding to an FnBR peptide (11). Confirmation of this mechanism of binding awaits the determination of other F1 module-FnBR peptide complex structures. This work is currently under way.
A strong test for the tandem ␤-zipper model and the relevance of the results obtained with the dissection approach is the measurement of binding using intact FnBRs. Assembly of specific F1-binding segments in the correct order to match the array of F1 modules in the NTD results in a high affinity interaction, whereas their assembly in the incorrect order does not (Fig. 7). This could convincingly be demonstrated by comparing the binding of the FnBRs SfbI-4 and SfbI-4.5 to Fib1. The failure of the mismatch FnBR SfbI-4.5 to bind Fib1 with high affinity also demonstrates that the mismatch binding activity identified using module pairs is unlikely to be functionally relevant.
The existence of subsites that recognize specific F1 modules was proposed previously for PrtF1 (42). In that study Fn frag- ments including variants of the 70-kDa fragment (comprising NTD and the gelatin-binding fragment (GBF)) with deleted F1 modules in the NTD were used. The data were interpreted in terms of parallel binding of streptococcal peptides to NTD. Parallel binding is not consistent with either the data presented here or previously (11). In the light of the tandem ␤-zipper model, where an array of binding sites in the correct sequence is crucial for activity, the use of gap constructs of the NTD (42) might explain the apparent discrepancy between the results.
Full Fn binding activity of SfbI (or PrtF1) requires both FnBR and a second, upstream Fn-binding region (UR, Fig. 1) as well as both the NTD and GBF from Fn (39). Upstream regions with differing boundaries have also been referred to as spacer (43), UFBD (44), or FUD (42). It was suggested that UR and FnBRs independently mediate adherence of bacteria to Fn, with UR acting as the high affinity Fn-binding site (39). However, the UR construct, as defined by Ozeri et al. (39), contains, according to our definition of FnBRs, an almost complete FnBR (SfbI-1) that would bind to the NTD. Because the N-terminal region of this UR construct contains a GBF-binding site (40), this might explain the high affinity binding of this construct to Fn. Talay and co-workers (40) suggest an anti-parallel binding mode of SfbI to Fn where the FnBR region interacts with the NTD and activates binding of UR to GBF. It will be interesting to examine how SfbI binding to the NTD extends into the GBF and whether any of the four F1 modules in GBF ( 6 F1 and 7-9 F1; Fig. 1C) is involved in binding to UR.
We have demonstrated with this paper that the FnBR/Fn interaction involves the binding of four or five short consecutive motifs in FnBRs to four or five sequential F1 modules in the NTD. As a result, bound FnBRs form a large intermolecular surface with their modular target. The very significant conformational change that natively unfolded FnBRs undergo upon binding is entropically highly unfavorable. The disordered nature of unbound FnBRs may be the perfect starting point for the adoption of an extended conformation. This is a striking example for the efficiency of intrinsically disordered proteins, which can present much larger interaction surfaces than a globular protein of the same size (45). Many mosaic proteins contain strings of modules with external ␤-strands, so it is tempting to suggest that the Fn/FnBR interaction is just the first example of this unusual mechanism of protein-protein interaction.  (11). b n/f, non-fittable.
FIG. 7. Isothermal titration calorimetry profile for the interaction of NTD with SfbI FnBRs. Binding of Fib1 (containing the NTD of Fn) to SfbI-4 (match binding) (A), SfbI-4.5 (mismatch binding) (B), and SfbI-5 (match binding) (C) is shown. Ways in which the two components in each titration are likely to interact are indicated in each panel.