Catch-bond mechanism of the bacterial adhesin FimH

Ligand–receptor interactions that are reinforced by mechanical stress, so-called catch-bonds, play a major role in cell–cell adhesion. They critically contribute to widespread urinary tract infections by pathogenic Escherichia coli strains. These pathogens attach to host epithelia via the adhesin FimH, a two-domain protein at the tip of type I pili recognizing terminal mannoses on epithelial glycoproteins. Here we establish peptide-complemented FimH as a model system for fimbrial FimH function. We reveal a three-state mechanism of FimH catch-bond formation based on crystal structures of all states, kinetic analysis of ligand interaction and molecular dynamics simulations. In the absence of tensile force, the FimH pilin domain allosterically accelerates spontaneous ligand dissociation from the FimH lectin domain by 100,000-fold, resulting in weak affinity. Separation of the FimH domains under stress abolishes allosteric interplay and increases the affinity of the lectin domain. Cell tracking demonstrates that rapid ligand dissociation from FimH supports motility of piliated E. coli on mannosylated surfaces in the absence of shear force.

C ell-cell adhesion often occurs under dynamically varying conditions and mechanical stress. In many cell-cell adhesion systems, the lifetime of adhesin-receptor complexes is increased under tensile mechanical force via 'catch-bonds', which permit capture or retention of cells under flow conditions while still allowing for release under reduced mechanical force. Catch-bond interactions are prominent in vascular systems and are formed, for example, by selectins for leukocyte recruitment 1,2 , by cadherins controlling tissue integrity 3,4 in the epithelial adhesion of cancer cells 5 and by the interactions between T-cell receptors (TCRs) and peptide-bound major histocompatibility complexes (MHC) on antigenpresenting cells 6,7 . Catch-bonds also play a major role in bacterial adhesion and infection by uropathogenic Escherichia coli strains, which are responsible for the vast majority of urinary tract infections (UTIs) in humans 8 . A first critical step in the establishment of infection is bacterial adhesion to urothelial cells under flow conditions, which is mediated by 0.1 À 2 mm long, proteinaceous filaments on the bacterial surface termed type 1 pili 9,10 . Type 1 pili are composed of up to 3,000 copies of the subunit FimA building the pilus rod, as well as the subunits FimF, FimG and FimH forming the distal tip fibrillum 11 . The adhesin FimH at the fimbrial tip specifically binds in a catch-bond mode 12 to terminal a-D-linked mannoses of N-linked glycans of the receptor uroplakin 1a on urinary epithelial cells 13 . Owing to its important role in establishing infection, FimH is an attractive target for the development of anti-adhesive drugs for UTI treatment 14,15 .
FimH is a two-domain protein, composed of an N-terminal, mannoside-binding lectin domain (FimH L ) and a C-terminal pilin domain (FimH P ). FimH P possesses an incomplete immunoglobulin-like fold that is completed by insertion of an N-terminal donor strand of FimG, the subsequent subunit in pilus assembly 11 . The two-domain architecture of FimH is a prerequisite for catch-bond formation because the interactions between FimH L and FimH P determine the conformational state and ligand-binding properties of FimH L (refs 12,16,17). A 'compressed' FimH L conformation was observed in the crystal structure of FimH in the context of the type 1 pilus tip fibrillum in the absence of ligands, with an open binding site and interactions to FimH P mediated via three loop segments: the swing (amino acids (aa.) [27][28][29][30][31][32][33], linker (aa. 154-160) and insertion loops (aa. 112-118) 17 . In contrast, an 'extended' FimH L conformation was observed in crystal structures of the isolated, ligand-bound FimH L domain [18][19][20][21][22][23] and in the complex between FimH and the pilus assembly chaperone FimC, where FimC prevents the interactions between FimH L and FimH P (ref. 24). This extended form of FimH L is characterized by a closed ligandbinding pocket and rearranged swing, linker and insertion loops.
Notably, isolated FimH L was reported to show a ligand-binding affinity about two orders of magnitude higher than that of fulllength FimH in the tip fibrillum 17,25 . Together with mutagenesis experiments disrupting the interdomain interface 26 , these data indicated that ligand-binding is linked to domain separation in FimH, and that mechanical force shifts the ligand-binding affinity towards that of the isolated FimH L . However, fundamental aspects of the mechanism underlying the force-dependent binding of FimH remained unknown: (i) How is domainassociated, full-length FimH interacting with ligands? (ii) Does ligand-binding directly induce domain separation? (iii) How are interdomain interactions linked to the ligand-binding affinity of FimH and the kinetics of ligand-binding and dissociation?
To address these questions, we designed a stable, soluble variant of full-length FimH that is equivalent in its structural and functional properties to those of FimH in the assembled fimbrial tip. This variant allowed us to obtain high-resolution structural snapshots of all functional states of FimH and to obtain a complete characterization of ligand-binding kinetics in solution. Together with molecular dynamics simulations, these data reveal a three-state mechanism of FimH catch-bond formation. FimH P accelerates ligand release from FimH L via dynamic allostery by 100,000-fold. In addition, using single-cell tracking experiments, we show that the modulation of ligand affinity by FimH P is not only required for adhesion under mechanical stress, but also for efficient bacterial surface motility in the absence of shear force. Our results provide a first complete structural and kinetic description of a catch-bond system and establish a framework for the analysis of the distinct catch-bond mechanisms in other systems, which also commonly couple interdomain interactions to ligand affinity.

Results
Construction of a peptide-complemented FimH. Isolated FimH with its non-complemented pilin domain is only marginally stable and shows aggregation tendency under physiological conditions 27 . To establish a stable, isolated FimH molecule with all properties of FimH in the tip fibrillum, we complemented FimH P with the donor-strand peptide of FimG (FimG residues 1 À 14; termed DsG). The FimH Á DsG complex was obtained in good yields and purified after an in vitro reaction, mimicking the first donor-strand exchange (DSE) reaction during pilus assembly in vivo. In this reaction, the FimG donor strand displaces the pilus assembly chaperone FimC from FimH ( Fig. 1a): The experiments described in the following were performed with FimH from the faecal E. coli strain F18 (FimH F18 ), which is structurally identical to the most prevalent variants in uropathogenic infection 25 , and FimH from the wild-type E. coli strain K12 (FimH K12 ), which differ in three amino acids in FimH L (K12-F18: Val27Ala, Asn70Ser, Ser78Asn; Supplementary  Fig. 1a). The isolated lectin domains (residues 1-159) of both FimH variants (FimH L K12 and FimH L F18 ) were produced by direct expression in the E. coli periplasm and were purified as described 27 .
Ligand-free FimH . DsG resembles FimH in the fimbrial tip. The crystal structure of the binary complex FimH F18 Á DsG was determined at atomic resolution by molecular replacement (Table 1). FimH F18 Á DsG comprises the jellyroll fold FimH L and the immunoglobulin-like FimH P domain complemented with the FimG donor strand ( Fig. 1b and Supplementary Fig. 1b). It closely resembles unliganded FimH in the fimbrial tip complex (Fig. 1c) 17 , with a root-mean-square deviation of C a positions (C a r.m.s.d.) of 1.1 Å. The individual FimH P and FimH L domains are even more closely resembling unliganded, fimbrial FimH (r.m.s.d. 0.45 and 0.55 Å, respectively) and undergo only a minimal hingebending rotation of 4° (Fig. 1c).
The DsG peptide in FimH F18 Á DsG is in identical position as compared with the N-terminal FimG extension in the fimbrial tip structure; it interacts with b-strand 2 and 9 of FimH P (Fig. 1d). All contacts in the FimH L -FimH P interdomain region ( Supplementary Fig. 1c,d) 17 as well as the conformation of the empty ligand-binding pocket observed in FimH in the fimbrial tip are preserved in FimH F18 Á DsG. Thus, FimH F18 Á DsG represents the ligand-free state of fimbrial FimH, with Associated FimH L and FimH P (A free state) and is an elegant minimal system to analyse the crosstalk between ligand-binding and interdomain interactions underlying the formation of catch-bonds by FimH.
Persistence of domain association in ligand-bound FimH . DsG. To test whether ligand-binding causes domain separation in FimH, we determined the co-crystal structure of the ternary complex of FimH F18 Á DsG with n-heptyl a-D-mannoside (HM), an established model ligand of FimH 20 , as well as crystal structures of the isolated FimH L F18 and FimH L K12 lectin domains in complex with HM (Table 1). FimH F18 Á DsG Á HM adopts the same closed conformation of the ligand-binding site as previously observed in other FimH Lligand complexes (Fig. 2a,b and Supplementary Fig. 2) 28 . The mannopyranose moiety of HM is coordinated by the side chains of Asp54, Gln133, Asn135 and Asp140, and the main chain of Phe1 and Asp47, and the n-heptyl aglycone of HM is sandwiched between Tyr48 and Tyr137. Compared with the A free form, all loops surrounding the binding pocket close down onto the HM ligand. The most substantial conformational difference to A free is observed for the clamp loop (aa. 8-16), whose tip moves almost 6 Å towards HM ( Supplementary Fig. 1e,f).
Besides the closing of the ligand-binding pocket, the overall conformation of ligand-free FimH L in A free and HM-bound FimH F18 Á DsG is closely similar (C a r.m.s.d. 1.1 Å; Fig. 2a NAMD package ( Supplementary Fig. 3a,b). The domain association remained intact over 100 ns of simulation time without substantial changes in the domain interface; fluctuations were limited to the clamp loop region close to the ligand-binding site.
On in silico removal of the HM ligand after initial equilibration, the A bound state underwent a spontaneous transition to the A free state after B75 ns of simulation time via an opening of the clamp loop ( Supplementary Fig. 3c,d), reproducing the experimentally observed dependence of the binding-site conformation on ligandbinding. Thus, the MD simulations indicate that A bound is a stable conformational state of FimH induced by ligand-binding.
Trapping of a domain-separated state of full-length FimH. The increase in apparent affinity of FimH to its target glycans under tensile mechanical forces 12,29 has previously been linked to a separation of the FimH L and FimH P (ref. 17). To trap a potential domain-separated state of FimH for structural characterization in the absence of tensile force, we considered FimH variants with weakened interdomain interactions. We had shown previously that FimH P also accepts the donor strand of the non-cognate subunit FimF (DsF). However, FimH P is slightly less stabilized by complementation with DsF than with the natural donor-strand DsG 30 . We hypothesized that such complementation with DsF instead of DsG could also result in a mild destabilization of the interdomain interface in full-length FimH. We then determined the co-crystal structure of FimH K12 Á DsF with HM (FimH K12 Á DsF Á HM) at 3.0 Å resolution with four molecules in the asymmetric unit. Three FimH K12 Á DsF Á HM molecules closely resembled the A bound state (r.m.s.d. of 0.6 Å to FimH F18 Á DsG Á HM) with a preserved interdomain interface. In the fourth molecule, however, the FimH L and FimH P domains were separated and they adopted a drastically different relative orientation with an angle between the domains of B45°instead of B150°in the other three molecules (Fig. 2a). FimH P is virtually identical in all four FimH molecules in the crystal (r.m.s.d. 0.4 Å). In contrast, the FimH L domain differs significantly between the fourth, domain-separated and the three full-length FimH molecules in the crystal. It shows closest similarity to the isolated FimH L Á HM (r.m.s.d. 0.45 Å); in particular, all interdomain loops adopt identical conformations, which are incompatible with domain association (Figs 2c and 3). Remarkably, in the bent fourth molecule, no interactions between FimH L and FimH P other than the direct covalent linkage are detected, equivalent to a breakdown of the total 500 Å 2 interdomain interface of the A bound state (Fig. 3). This molecule thus represents a third state, the domain-Separated, ligandbound state of FimH, S bound . The complete absence of noncovalent interdomain interactions indicates that the S bound state does not possess a defined relative domain orientation in solution, and that the observed, kinked conformation has been selected only by crystal packing.
To analyse the transition trajectory of the A bound to the S bound state, we removed the FimH P domain after equilibration from the A bound state in silico for a 180-ns molecular dynamics simulation ( Supplementary Fig. 3e,f). In contrast to the transition between the A bound and A free states on ligand removal, a sharp transition to the conformation of FimH L in the S bound state was not observed. The conformation only slowly moved towards S bound ; however, the FimH L loops that had formed in the former interdomain interaction kept fluctuating throughout the simulation, indicating lower cooperativity and potentially a higher activation energy for the A bound -S bound compared with the A bound -A free transition.
A comparison of the structural dynamics in the A bound and S bound states clearly reveals differences in the FimH L -FimH P interface region. The root-mean-square fluctuations of atom positions (r.m.s.f.) increase in the swing and insertion loop from a background level of B0.7 Å in A bound to 1.5 and 2 Å in S bound , respectively. Surprisingly, despite the virtually identical conformations of the entire ligand-binding site depicted by X-ray crystallography (Fig. 2b), the clamp loop, which exhibits the most significant conformational changes between the open and closed conformations, exhibits strongly reduced fluctuations in S bound , with r.m.s.f. decreasing by up to 1.5 Å ( Supplementary Fig. 3g,h). This change in clamp loop dynamics provides a mechanistic link between domain association and ligand-binding in full-length FimH.

Domain association alters FimH-ligand-binding kinetics.
To analyse the ligand-binding properties of FimH Á DsG, we exploited the increase in intrinsic tryptophan fluorescence in the FimH Á DsG complexes of B10% on HM binding (Fig. 4a). This difference was used to measure the dissociation constant of HM binding by equilibrium titration (Fig. 4b) and the rates of HM binding and dissociation by stopped-flow fluorescence kinetics (Fig. 4c,    and FimH F18 Á DsG, respectively (Table 2). HM binding to FimH Á DsG is extremely dynamic and was characterized by fast association rates (k on ) of 5.0 Â 10 6 and 4.9 Â 10 6 M À 1 s À 1 , respectively, and rapid dissociation reactions ( Supplementary  Fig. 4). The rates of HM dissociation (k off ) of 22 and 58 s À 1 for FimH K12 Á DsG and FimH F18 Á DsG translate into dissociation half-lives of only 32 and 12 ms, respectively. In contrast to full-length FimH, isolated FimH L K12 showed no change in tryptophan fluorescence on HM binding. We therefore determined the HM affinity of  Table 2). In an inverse competition experiment (Supplementary Fig. 5f-j), in which HM in preformed FimH L Á HM complexes was displaced by GN-FP-4, off-rates of 2.0 Â 10 À 4 and 3.5 Â 10 À 4 s À 1 were determined for FimH L K12 and FimH L F18 , respectively, corresponding to dissociation half-lives of 58 and 33 min (Table 2). On the basis of these measured off-rates and equilibrium dissociation constants, k on rates of 1.8 Â 10 5 and 1.2 Â 10 5 M À 1 s À 1 were calculated for FimH L K12 and FimH L F18 , respectively. The on-rates for the isolated FimH L domains are thus 30-fold lower than those of the corresponding full-length FimH Á DsG complexes.
Together, these results demonstrate that the 3,300-fold higher affinity of the isolated FimH L compared with full-length FimH results from a more than 100,000-fold lower ligand dissociation rate in isolated FimH L , combined with a ligand-binding rate reduced by only 30-fold ( Table 3). The 3,300-fold higher affinity for HM of FimH L relative to FimH Á DsG translates into a free energy of 20 kJ mol À 1 for the interaction between FimH L and FimH P in full-length FimH. This corresponds very well with the mechanical work required for domain separation, as a displacement of FimH L from FimH P by 11 Å for complete domain separation (Fig. 2a) 17 , and a force of 40 pN required to populate the domain-separated state of FimH 31 yields a value of 26.5 kJ mol À 1 .
Domain association in FimH promotes bacterial motility. Uropathogenic E. coli require firm adhesion to the urinary epithelium under flow conditions to escape clearance by urine excretion. On the other hand, bacterial adhesion must be weak enough in the absence of external shear to allow flagellar motility as a prerequisite for the invasion of new tissue areas 32 Table 2). (d) Amplitudes of the reactions monitored in c, plotted against the total HM concentration. Data were fitted (solid line) according to equation (2), yielding a K d value of 12 ± 1 mM.
rapid ligand dissociation under static conditions for flagellar motility remained unclear because of the complex interplay of flagellar swimming and the avidity of multivalent surface interactions by hundreds of E. coli pili. Here we employed single-cell tracking of piliated E. coli cells moving on surfaces coated with mono-mannosylated bovine serum albumin (1M-BSA), an established model system for analysing FimHbased adhesion 12,29 , for a classification of cell motility into two states, attached or mobile (for details see Methods and Supplementary Fig. 6). To study the influence of FimH interdomain interactions, we compared isogenic E. coli strains producing either wild-type FimH F18 or the FimH F18 -variant Ala188Asp, which is characterized by a destabilized interaction between FimH P and FimH L (ref. 26) and serves here to mimic the S bound state in the absence of shear force 35 . The overall fraction of adherent FimH F18 -piliated bacteria on 1M-BSA-coated surfaces was identical to background levels on non-adhesive BSA-coated surfaces at 10-12% of tracked bacteria (Fig. 6a). In contrast, FimH F18 -Ala188Asp-piliated bacteria showed an increased fraction of adherent cells of 24% already at the beginning of cell tracking after a 1-min dead time, which further increased during the 5-min observation period to 48% ( Fig. 6a and Supplementary Movies 1 and 2). Cell tracking permitted quantitative analysis of the transition of individual cells between a mobile and an attached state ( Supplementary Fig. 7). On non-adhesive control surfaces coated only with BSA, less than 2% of all FimH F18 -or FimH F18 -Ala188Asp bacteria showed shifts between the two states of motion ( Fig. 6b and Supplementary Fig. 7c). However, on adhesive 1M-BSA surfaces, 13.9% of all FimH F18 tracks (green in Fig. 6b) exhibited a single transient attachment event with a mean duration of 6.9 s (Supplementary Fig. 7d). For FimH F18 -Ala188Asp-piliated bacteria, only 7.2% of the cells showed attachment/detachment, but with fivefold longer adhesion (35.2 s; Supplementary Fig. 7d). Remarkably, the fraction of cells that permanently stayed attached after adhesion to 1M-BSA until  Protein 3.5 ± 0.8 Â 10 À 4 n.a. n.a.

Discussion
The characterization of full-length FimH had so far been restricted to the analysis of the adhesive properties of piliated E. coli cells and binding studies with the purified type 1 pilus tip fibrillum. With the FimH Á DsG complex, we have now established a model system for quantitative studies of the interaction of FimH with carbohydrate ligands. Soluble FimH Á DsG efficiently mimicks FimH in the context of the assembled tip fibrillum, is readily available in milligram quantities and permits the determination of ligand-binding and release kinetics in solution.    Using FimH Á DsG, we obtained high-resolution snapshots of three functionally relevant states of FimH (Fig. 7). In the absence of ligands, FimH adopts the A free state with associated FimH L and FimH P and an open conformation of the ligand-binding site, which is responsible for the 30-fold faster ligand-binding of full-length FimH as compared with the isolated FimH L domain. Ligand-binding in the absence of shear force induces the A bound state with a closed binding site. In contrast to earlier hypotheses 17 , the transition from A free to A bound is restricted to the ligand-binding site, while all interactions between FimH L and FimH P observed in the A free state remain preserved in A bound . The A free -A bound transition most likely follows an induced fit mechanism, in which the formation of an encounter complex between FimH Á DsG and HM is rate-limiting and followed by a fast, unimolecular rearrangement to the A bound state, in agreement with the observation that binding of the model ligand HM remained rate-limiting for the formation of A bound even at the highest HM concentrations used. Stopped-flowbinding kinetics indicate that the lifetime of the proposed encounter complex before A bound formation is below 1 ms (Fig. 4c). Under tensile mechanical force applied to the FimHligand complex, mimicked here by the destabilized variant FimH Á DsF and crystal packing forces, the domain-separated state of FimH, S bound , is formed. In the S bound state, FimH L and FimH P no longer interact specifically and are only connected via the linker segment comprising FimH residues 154-160. In this S bound state, FimH L adopts a conformation closely resembling isolated FimH L with bound ligand. Notably, ligand dissociation from FimH Á DsG is 100,000-fold faster than that from the isolated FimH L domain. This is striking because the respective crystal structures revealed indistinguishable ligand interactions and binding-site conformations in the FimH Á DsG Á HM and FimH L Á HM complexes (Fig. 2b). MD simulations identified a considerable increase in the conformational dynamics of the FimH L -ligand-enclosing clamp loop in the A bound state as the most likely cause of the dramatic increase in k off in the FimH Á DsG Á HM complex. The altered dynamics in FimH L in the A bound state are the result of the presence of FimH P , which can be described as a negative allosteric regulator [36][37][38] . The allosteric communication from the FimH P -FimH L interface to the ligand-binding site reaches over 40 Å, and is mediated via changes in protein dynamics rather than in static structure, in line with a general model of dynamic allostery 39,40 . Our data demonstrate that the interdomain interactions in FimH (i) maintain the open conformation of the binding pocket and guarantee rapid ligand-binding and (ii) intramolecularly catalyse ligand dissociation by more than 100,000-fold. Rapid ligand-binding and short lifetimes of the FimH ligand complex allow for rapid dissociation of individual pili from their ligands in the absence of shear force. Our biophysical data demonstrate that this mechanism is conserved between the K12 and the F18 E. coli strains.
Different mechanistic models, such as the two-pathway 41 , the deformation 42 and the sliding re-binding model 43 , have been developed to describe catch-bond interactions, often based on powerful single-molecule atomic force measurements. These models included the principle of allosteric control of ligandbinding affinity 26,31 , which was clearly fully confirmed in our present study. However, these conceptual models did not reveal the underlying atomic-scale mechanisms in different catch-bond systems. For most catch-bond systems, including the cadherin-catenin binding to actin filaments 3,44 , integrin epithelial cell adhesion 45,46 and TCR-MHC interactions 6,7,47 , structural information is, if at all, available only for one state or from computer simulations. One exception is the selectins, which employ catch-bond binding for leukocyte recruitment. Selectins are multidomain cell surface receptors, which consist of a lectin domain for complex carbohydrate binding, linked via an epidermal growth factor (EGF)-like domain to a variable number of short consensus repeat domains and a transmembrane-anchoring helix. Selectins exist in two conformations, a bent and an extended one, which differ in the angle between their lectin and EGF-like domain. Ligand-binding and conformational changes in the ligand-binding site are directly linked via a complex allosteric coupling mechanism to the adoption of the extended conformation 48,49 . Tensile mechanical force under flow conditions acts along the axis of the ligandbinding site and the Lec-EGF interface resulting in a stabilization of the extended conformation and thus increased ligand complex lifetimes 2,49 . Moreover, in FimH, catch-bond behaviour is mediated by the interplay of a lectin and an anchoring domain that does not interact with the ligand. Ligand-binding by FimH in the absence of shear force results in a closing of the ligand-binding site, but, in contrast to selectins, is not directly linked to altered interdomain interactions. Here mechanical force promotes domain separation and completely releases FimH L from FimH P , which acts as an activator of ligand release via dynamic allostery. Remarkably, selectins and the fimbrial adhesin FimH thus employ entirely different mechanisms for establishing catch-bond behaviour by crosstalk between a lectin and an anchoring domain that provides tethering to a shaft. In both systems, the selectins and fimbrial adhesion, the shaft structures linking the terminal lectin/coupling domains to the cell surface, may contribute to the overall catch-bond behaviour, either via directly influencing coupling domain behaviour or via their general elastic properties 50,51 .
The cell-tracking experiments indicate the importance of rapid ligand release from the high-mannose-type glycoprotein receptor uroplakin 1a in the lower urinary tract 52 for flagellar motility of piliated bacteria, and hence their ability to colonize new tissue areas under certain conditions during infection 12,29,53 . This provides a plausible explanation for the fact that low-affinity FimH variants were preserved in numerous uropathogenic E. coli strains. Binding of terminal mannoses with low affinity in the absence of shear force may also play a role in preventing the clearance of uropathogenic E. coli from the urinary tract by competitive binding to the Tamm-Horsfall protein in the urine 54 . In turn, populating the S bound state with an extremely low dissociation rate ensures tight bacterial adhesion under the mechanical forces of urine excretion. FimH is a promising target for anti-adhesive therapy of UTI because FimH antagonists, in contrast to antibiotics, are not exerting selection pressure towards resistance formation 18,55,56 . Previous ligand-binding studies on the isolated FimH L domain mimic the domain-separated S bound state of FimH. This state is characterized by extremely low off-rates and is promoted in vivo only after ligand-binding and the onset of flow conditions. Our kinetic data on ligand dissociation from full-length FimH demonstrate that rapid, competitive displacement of FimH from its carbohydrate ligands by FimH antagonists is well possible in the absence of shear force. Thus, full-length FimH (for example, in the form of the FimH Á DsG complex established in this study) instead of the isolated FimH L domain is the relevant target for the development of anti-adhesive drugs. Importantly, the concept of the FimH Á DsG model system can now be expanded to other related adhesive pilus adhesins. In combination with the novel fluorescent GN-FP-4 ligand, this model system paves the way for efficient screening for anti-adhesive drug candidates.

Methods
Materials. The synthetic DsG (sequence: ADVTITVNGKVVAKR) and DsF peptide (sequence: ADSTITIRGYVRDNG; 495% purity) were purchased from JPT (Germany). Guanidinium chloride ('AA-Grade' for spectroscopy) was NATURE COMMUNICATIONS | DOI: 10.1038/ncomms10738 ARTICLE obtained from NIGU Chemie (Germany). Standard chemical of highest purity available was obtained Sigma, Merck or AppliChem. If not mentioned otherwise, chromatography media for protein purification were purchased from GE Healthcare (UK). Oligonucleotides were from Microsynth (Switzerland).
Construction of expression plasmids. Expression plasmids for the periplasmic production of the E. coli F18 FimH lectin domain (FimH L F18 ) and for the periplasmic co-expression of full-length FimH F18 with FimC were based on the expression plasmids pfimH L and pfimH-fimC-ATG, respectively, for the analogous proteins from E. coli K12 (ref. 27). Six silent mutations replacing rare codons were introduced into the E. coli F18 fimH gene (fimH F18 ) contained in the plasmid pGB2-24 (ref. 57) with the QuikChange mutagenesis kit (Agilent Technologies, Switzerland) to improve periplasmic expression. The coding sequence of the modified fimH F18 gene was amplified by PCR using the primers 5 0 -GATCCTCTA GAGGAGGGATGATTGTAATGAAACGAG-3 0 and 5 0 -TTTCAAGCTTATTGA TAAACAAAAGTCACG-3 0 and cloned into pfimH-fimC-ATG 27 via the XbaI and HindIII sites (thereby replacing the fimH K12 gene) and yielded pfimH F18 -fimC-ATG. The gene encoding FimH L F18 was amplified with the primers 5 0 -GATCCTCT AGAGGAGGGATGATTGTAATGAAACGAG-3 0 and 5 0 -CAGCCAAGCTTAG CCAGTAGGCACCACCAC-3 0 and ligated via the XbaI and HindIII sites into ptrc99a-f1-stopp 27 . Protein production in the resulting plasmid pfimH L F18 is under control of the trc-promotor/lac operator.
Protein production and purification. For purification of the complexes FimC Á FimH K12 and FimC Á FimH F18 , E. coli HM125 harbouring the corresponding co-expression plasmid was grown at 30°C in 2YT medium containing ampicillin (100 mg ml À 1 ). At an OD 600 of 1.5, isopropyl-b-D-thiogalactoside (IPTG) was added to a final concentration of 1 mM. The cells were further grown for 12 À 18 h, harvested by centrifugation, suspended in cold 50 mM Tris-HCl pH 7.5, 150 mM NaCl, 5 mM EDTA, 1 mg ml À 1 polymyxin B sulfate (18 ml l À 1 of culture) and stirred at 4°C for 1.5 h. After centrifugation, the supernatant (periplasmic extract) was dialysed against 20 mM Tris-HCl pH 8.0 and applied to a QA52 (Whatman, Maidstone, UK) column equilibrated with the same buffer. The flow-through containing the respective FimC Á FimH complex was dialysed against 20 mM MOPS-NaOH pH 7.0, loaded onto a Resource S column equilibrated with the same buffer and the complexes were eluted with a linear NaCl gradient (0 À 400 mM). Fractions containing FimC Á FimH were pooled and loaded onto a Superdex 75 (HiLoad 26/60) column equilibrated with 20 mM NaH 2 PO 4 -NaOH pH 7.4, 50 mM NaCl. Fractions containing the pure complex were pooled and stored at 4°C until further use. Typically, 3 À 5 mg of the purified complex were obtained per litre of bacterial culture.
For expression of the isolated E. coli FimH L K12 and FimH L F18 , E. coli HM125 transformed with the respective expression plasmid was grown at 30°C in M9 medium containing ampicillin (100 mg ml À 1 ) to an OD 600 of 1.0, and expression was induced with 1 mM IPTG. After further growth for 12 h, cells were subjected to periplasmic extraction (see above). The extracts were mixed with 0.11 volumes of 1 M acetic acid-NaOH pH 4.5, dialysed against 10 mM acetic acid-NaOH pH 4.5 and then loaded onto a SP-Sepharose column equilibrated with the same buffer. The flow-through was collected and its pH was adjusted to 8.0 by addition of 1 M Tris-HCl pH 8.2. This solution was then applied to a Q-Sepharose column equilibrated with 20 mM Tris-HCl pH 8.0. The flow-through containing FimH L was loaded onto a Resource S column dialysed against 20 mM formic acid-NaOH pH 4.0. The protein was eluted with a linear NaCl gradient (0 À 1 M). Fractions containing pure FimH L were pooled, dialysed against water and stored at À 20°C. The identity of the purified proteins was confirmed by electrospray ionization Production of FimH . DsG and FimH K12 . DsF complexes. The respective FimC Á FimH complex (40 mM) was incubated with a threefold molar excess of the DsG peptide and incubated in 20 mM NaH 2 PO 4 -NaOH, pH 7.0, 50 mM NaCl for 48 h at 37°C. The reaction mixture containing isolated FimC, the FimH Á DsG complex and excess DsG was dialysed against 20 mM acetic acid-NaOH pH 4.5 and loaded onto a Resource S (6 ml) column equilibrated with the same buffer. The FimH Á DsG complex was eluted with a linear NaCl gradient (0 À 400 mM). Fractions containing the pure complex were pooled, dialysed against water and stored at 4°C. The FimH Á DsG partially dissociated during ESI-mass spectrometry analysis, so that masses of the intact complexes and free FimH were obtained:  29,036.0 Da. The overall yields of the purified FimH Á DsG complexes relative to the initial amount of FimC Á FimH were in the range of 50-55%. The FimH K12 Á DsF complex was generated and purified as described for the FimH Á DsG complexes (FimH K12 Á DsF: calculated mass: 30,702.2 Da; measured mass: 30,702.5 Da). The FimH Á DsF complex was prepared from the FimC Á FimH complex after mixing with DsF exactly according to the protocol described above for FimH Á DsG and obtained in similar yields. Despite the non-natural interaction between FimH P and DsF, the FimH K12 Á DsF complex was formed four times faster than the FimH K12 Á DsG complex at pH 7.0 and 37°C, with a rate constant of 2.2±0.5 M À 1 s À 1 . The FimH Á DsF complex was stable against dissociation and unspecific aggregation.
Kinetics of HM binding to FimH . DsG. The rate constants of binding (k on ) and dissociation (k off ) for the complex between FimH Á DsG and HM were measured at 25°C in 20 mM MOPS-NaOH pH 7.4 in a SX20 stopped-flow instrument (Applied Photophysics, UK). A constant FimH Á DsG concentration of 1 or 2 mM was used. FimH Á DsG was mixed with different concentrations of HM (2 À 100 mM), and binding was monitored by the increase in fluorescence above 320 nm (excitation at 280 nm). The fluorescence traces were globally fitted with Dynafit 59 according to a second-order binding and first-order dissociation reaction. As an additional control, the fluorescence amplitudes of the individual reactions were plotted against the total HM concentration and fitted according to equation (2). The deduced dissociation constants reproduced the K d values obtained with equilibrium titration within experimental error.
Equilibrium titration of FimH . DsG with HM. The binding equilibrium between FimH Á DsG and HM was followed at 25°C in 20 mM MOPS-NaOH pH 7.4 on a QM 7/2003 spectrofluorometer (PTI) by the increase in fluorescence at 320 nm on HM binding (excitation at 280 nm). Measurements were performed with a stirred 1 Â 0.4-cm quartz cuvette. The concentration of FimH Á DsG was kept constant at 2 mM and the concentration of HM was varied between 0 and 200 mM. The samples were equilibrated overnight, and their fluorescence intensities were recorded for 30 s and averaged. The fluorescence intensities were plotted against total HM concentration and fitted according to equation (2) where F is the monitored fluorescence signal, F 0 is the fluorescence signal in absence of ligand, F N is the fluorescence signal at full saturation with ligand, K d is the dissociation constant, [P] 0 is the total concentration of FimH Á DsG and [L] 0 is the total concentration of HM.  Crystallographic data collection. All crystals, except for FimH F18 Á DsG Á HM, were cryo-preserved by the addition of ethane-1,2-diol to a final concentration of 20% (v/v). The precipitant solution used for the crystallization of FimH F18 Á DsG Á HM already contained 30% (v/v) methyl-2,4-pentanediol, which acts as cryoprotectant. Crystals were flash-cooled in liquid nitrogen. All measurements were carried out at the SLS beamline X06DA and X06SA (Swiss Light Source, Paul Scherrer Institute, Switzerland) at 100 K. All data were integrated, indexed and scaled using the XDS software package 60 (5% of the reflections were set aside as test set). Data collection statistics are summarized in Table 1.

Equilibrium titration of FimH
Crystallographic structure determination. All structures were solved by molecular replacement using structures of isolated FimH L (AA1-158, PDB ID: 3MCY 18 , and the pilin domain of FimC Á FimH (AA160-297, PDB ID: 1QUN 24 , as search models with the programme Phaser 61 ). Model building and structure refinement were performed with Coot (ref. 62) and PHENIX (ref. 63). Twelve out of thirteen residues could be built for the FimG donor strands in the crystal structures, and only the C-terminal lysine residue had weak electron density. Refinement statistics are summarized in Table 1.
Molecular dynamics simulations. Four molecular systems were prepared for FimH F18 . The first system was constructed using the A bound state of FimH F18 Á DsG Á HM (Supplementary Fig. 3a,b) and the second system is equivalent but HM was removed ( Supplementary Fig. 3c,d). The third system contains only the FimH L and HM from the A bound X-ray structure ( Supplementary Fig. 3e,f). The fourth system was prepared for FimH L and HM based on the S bound state in the FimH K12 Á DsF Á HM crystal structure ( Supplementary Fig. 3g,h).
The CHARMM-GUI web server 64 was used to prepare the molecular systems, which were solvated with TIP3 water molecules and ionized with 50 mM NaCl. Each system contains between 50,000 and 60,000 atoms. All simulations were performed with the NAMD simulation package (version 2.9) (ref. 65). The CHARMM36 force field was used for the protein, and parameters for HM were generated using the CHARMM General Force Field programme (version 0.9.7 beta). Electrostatic interactions were calculated using the particle-mesh Ewald method 66 with a grid spacing of 1 Å. The cutoff for the van der Waals interactions was taken at 12 Å with a switching function used after 10 Å. Time step for the integration of dynamics was 2 fs. Simulations were performed in an isothermalisobaric ensemble, with a pressure of 1 atm and a temperature of 300 K.
Cell tracking on mannose-BSA-coated surfaces. The E. coli KB18 strain 67 was kindly provided by Professor Evgeni Sokurenko and served as host for the generation of recombinant strains. KB18 contains the pPKL114 plasmid 57 , which encodes the whole fim operon with a translational stop linker upstream of the fimH gene. KB18 was co-transformed with the pGB2-24 plasmid, which was isolated from the ELT115 strain and encodes fimH J96 (kindly provided by Professor Evgeni Sokurenko). Single-nucleotide point mutations were introduced in fimH J96 using overlap extension PCR following standard molecular techniques to obtain fimH F18 and fimH F18 -Ala188Asp. The PCR products were cloned into pGB2-24 by the ApaLI and SphI sites, and KB18 was transformed with the resulting plasmid.
E. coli strains were grown from frozen stocks in LB medium supplemented with antibiotics (100 mg ml À 1 ampicillin and 25 mg ml À 1 chloramphenicol) until late log phase (OD 600 of 1.0-1.2) and diluted to an OD 600 of 0.01 before movie acquisition.
Cell culture dishes (35 mm, Corning Inc., Corning, NY) were incubated with 50 ml of 50 mg ml À 1 1M-BSA in 0.02 M bicarbonate buffer for 75 min at 37°C. The dishes were then washed three times and quenched with 0.1% PBS-BSA to remove unbound 1M-BSA and block remaining sites on the plastic surface to prevent nonspecific binding of bacteria. Controls were prepared by treating cell culture dishes with 0.1% PBS-BSA only. The bacterial suspension was added to the cell culture dishes for microscopy studies.
Cell tracking was carried out at room temperature under static conditions. A bacterial suspension of 50 ml in the late logarithmic growth phase was placed onto the cell culture dishes (diluted to OD 600 of 0.01), and a cover slide was placed on top. The delay between sample placement and start of the movie acquisition was about 1 min. Time-lapse movies were recorded with a Â 20 phase contrast objective using a CMOS digital camera (The Imaging Source Europe, Bremen, Germany) mounted on a Nikon Ti Eclipse inverted microscope and using the NIS Elements Basic Research software (Nikon, Zurich, Switzerland). Phase contrast images in an B5-mm-thick surface layer were taken at four to five frames per second over 5 min. The dead time of movie acquisition was B1 min. The resulting images were segmented by creating a projection of the average intensities over all frames to remove the background and by subsequent thresholding using the Maximum Entropy method in Fiji 68 to obtain binary images (examples shown in Supplementary Movies 1 and 2). The segmented movies were imported into Imaris (Bitplane, Zurich, Switzerland) and tracked through the autoregressive algorithm. A time filter was applied to exclude all tracks with a length below 15 s. Tracks longer than 15 s were reviewed individually and edited manually, if necessary. Five to seven independent movies were recorded for each experimental set-up: FimH F18 or FimH F18 -Ala188Asp on 1M-BSA-coated dishes and FimH F18 and FimH F18 -Ala188Asp on BSA-coated cell culture dishes. E. coli piliated with FimH F18 or FimH F18 -Ala188Asp binding to 1M-BSA in the absence (1,815 and 1,283 individual tracks, respectively) and in the presence of 200 mM HM (1,175 and 1,071 individual tracks, respectively) were analysed respectively. For E. coli piliated with FimH F18 or FimH F18 -Ala188Asp binding to BSA 1,314 and 1,065 individual tracks, respectively, were analysed. Bacteria with a speed of o0.5 mm s À 1 were classified as attached, all other bacteria were classified as mobile. Owing to limitation in the spatial and temporal resolution of movie acquisition, we did not further subdivide bacterial swimming into motility behaviours as 'rolling' 29,53 , 'roaming', 'orbiting' and so on. The individual cell tracks were classified into four classes: no motility change during observation (pre-attached or mobile), transient attachment, permanent attachment and permanent detachment. For FimH F18 on 1M-BSA surfaces, 13.9% (251 out of 1,815 tracks) of all tracks showed a single transient attachment event (Supplementary Fig. 7d). In total, 67 out of the 251 bacteria that underwent a first transient adhesion attached and detached from the surface a second time. For these cells, the average time between detachment and re-attachment was only 13.5 s (Supplementary Fig. 7e), suggesting that re-binding may be favoured by proximity to the surface as compared with the initial attachment. The mean velocity on 1M-BSA, as compared with BSA-coated surfaces, was reduced for both FimH F18 -piliated (4.2 and 7.4 mm s À 1 , respectively) and FimH F18 -Ala188Asp-piliated bacteria (3.5 and 8.1 mm s À 1 , respectively; Supplementary Fig. 7a,b). This reduction of the mean velocity originates from two different phenomena: in FimH F18 -piliated cells it is caused by a change from fast swimming to a slower mode of motion (Supplementary Fig. 7b; Supplementary Movie 1), which is consistent with bacterial surface rolling due to weak, short-lived mannose-based interactions 29,53 . In contrast, for FimH F18 -Ala188Asp-piliated bacteria, the reduction of the mean velocity results from an increase in the fraction of adherent cells on 1M-BSA compared with BSA (see main text). In the presence of 200 mM HM, the mean velocity on 1M-BSA is increased for both FimH F18piliated (6.5 mm s À 1 ) and FimH F18 -Ala188Asp-piliated bacteria (5.9 mm s À 1 ; Supplementary Fig. 7a) and transient and permanent attachment is reduced by 75% and 85%, respectively ( Fig. 6 and Supplementary Fig. 7c).