A Three-protein Charge Zipper Stabilizes a Complex Modulating Bacterial Gene Silencing*

Background: The complex between Hha and H-NS selective represses genes in Enterobacteria acquired by horizontal transfer. Results: A structural model for the regulatory complex is described. Conclusion: A charge zipper formed by interdigitation of residues from three proteins stabilizes the complex. Significance: Charge zippers provide selectivity to electrostatic protein complexes. Understanding selective gene silencing may help fighting antibiotic resistance. The Hha/YmoA nucleoid-associated proteins help selectively silence horizontally acquired genetic material, including pathogenicity and antibiotic resistance genes and their maintenance in the absence of selective pressure. Members of the Hha family contribute to gene silencing by binding to the N-terminal dimerization domain of H-NS and modifying its selectivity. Hha-like proteins and the H-NS N-terminal domain are unusually rich in charged residues, and their interaction is mostly electrostatic-driven but, nonetheless, highly selective. The NMR-based structural model of the complex between Hha/YmoA and the H-NS N-terminal dimerization domain reveals that the origin of the selectivity is the formation of a three-protein charge zipper with interdigitated complementary charged residues from Hha and the two units of the H-NS dimer. The free form of YmoA shows collective microsecond-millisecond dynamics that can by measured by NMR relaxation dispersion experiments and shows a linear dependence with the salt concentration. The number of residues sensing the collective dynamics and the population of the minor form increased in the presence of H-NS. Additionally, a single residue mutation in YmoA (D43N) abolished H-NS binding and the dynamics of the apo-form, suggesting the dynamics and binding are functionally related.

Antibiotic resistance and the appearance of new virulent bacterial strains constitute a major threat to human health (1). The problem is aggravated by the transfer of resistance and virulence genes between bacteria (horizontal gene transfer) (2)(3)(4). In this context, a detailed knowledge of the mechanisms allowing bacteria to tolerate the acquisition of foreign DNA is lacking, and it may open the way to new sustainable strategies to fight infectious diseases. Proteins Hha and YmoA were first identified as environmental regulators of the expression of virulence factors in Escherichia coli (5) and Yersinia sp. (6), respectively. Hha/YmoA bind to the nucleoid-associated protein H-NS (7), enhance its capacity to silence newly acquired genes, and facilitate the smooth integration of foreign genes in the existing genome (8,9). The key role of members of the Hha/ YmoA family in enterobacteria is supported by their conservation in obligate endosymbionts (10) and their presence in conjugative plasmids (11). Hha is also directly involved in the formation and dispersal of biofilms, a microbial life-style that is responsible of many chronic infections and is associated with increased antibiotic tolerance and with resistance dissemination (12)(13)(14). Therefore, Hha/YmoA constitutes promising targets in the development of new antimicrobial drugs.
Hha and YmoA are homologous and functionally replaceable proteins (15,16) with high sequence identity/homology (82.1/ 94.0%) and very similar structures formed by four helices separated by loops (17,18). Interestingly, Ͼ30% of their primary structure corresponds to charged residues. The high degree of conservation ( Fig. 1) and extensive mutational studies (19 -21) suggest the relevant role of charged amino acids in Hha function. The N-terminal region of H-NS interacting with Hha is also unusually rich in charged residues (39%), reinforcing the notion that electrostatic interactions are essential for this interaction (19). The role of electrostatics is also supported by the ionic strength sensitivity of genes regulated through Hha⅐H-NS (22).
The putative role of electrostatic interactions in the Hha/ YmoA complex with H-NS raises the problem of selectivity, as electrostatic interactions are often associated with nonspecific contacts in contrast to van der Waals interactions that display high selectivity by means of an exquisite shape complementarity. The interaction of Hha/YmoA with H-NS is highly specific (19,20).
In this article we have investigated the interplay between electrostatic forces and the constitutional and functional properties of Hha/YmoA. A detailed analysis of the structures of the Hha⅐H-NS complex in crystals (21) and in solution (first reported here) unraveled a charge zipper with conserved residues contributed by three proteins (Hha/YmoA and the two units of an antiparallel H-NS dimer). This feature explains the high selectivity of the Hha/YmoA interaction with H-NS. In addition, electrostatic repulsive interactions within YmoA are also responsible for the presence of collective slow motions (s-ms) in the free form that, we argue, may also contribute to shape the biological function of Hha/YmoA.

Experimental Procedures
Protein Preparation-Unlabeled and isotopically enriched H-NS 46 C21S (19), H-NS 64 (23), Hha variants (23), and YmoA variants (18) were expressed and purified as previously described. Single cysteine mutants were prepared by site-directed mutagenesis after substitution of the native cysteine by serine (H-NS) or isoleucine (Hha). Site-directed mutagenesis was carried using the QuikChange site-directed mutagenesis kit. All constructs were verified by DNA sequencing. Fluorescence anisotropy titrations were carried out to evaluate the affinity of YmoA toward H-NS 64 (24)  Additional experiments were carried out by observing 15 Nlabeled HhaC18I in the presence of H-NS 46 dimer with an EDTA-Mn 2ϩ tag placed in the C21 residue of each monomer. A third set of PRE experiments was carried out using 15 N-labeled H-NS 46 C21S with an additional C-terminal cysteine tagged with 4-(2-iodoacetamide)-2,2,6,6-tetramethyl-1-piperidinyloxy radical (TEMPO). This sample was used to confirm the antiparallel topology of the H-NS 46 dimer. In all the PRE experiments the buffer was 20 mM HEPES (pH 7.0), 150 mM NaCl, and 0.01% (w/v) NaN 3 . Adding a 2-3 molar excess of ascorbic acid generated diamagnetic reference samples of nitroxide radicals. EDTA-Ca 2ϩ -tagged dimers were used as diamagnetic reference for the manganese-containing samples.
Spin Labeling-Derivatization reagents were 1-oxyl-2,2, Proteins containing a single cysteine residue were incubated with 10 mM DTT at room temperature to ensure the reduction of any intermolecular disulfide bonds. The excess of DTT was removed by passing 2 times the reaction mixture through a PD-10 column (GE Healthcare) (99.9% desalting capacity). After elution, protein solutions in 10 mM Tris (pH 7.4) (ϳ100 M) were mixed with a 10-fold molar excess of the derivatization reagent an incubated for 3 h in the dark at room temperature. To remove any unreacted tag, the resulting protein solution was passed 2ϫ through a PD-10 column, and the buffer was exchanged to 20 mM HEPES (pH 7.0), 150 mM NaCl, and 0.01% (w/v) NaN 3 . Complete incorporation of MTSL was confirmed by mass spectrometry, and the oligomeric state of the conjugated protein was checked by analytical gel filtration.
In the case of the cysteaminyl-EDTA adduct, the reaction was carried out in the presence of 1 mM MnCl 2 or CaCl 2 to generate the paramagnetic and diamagnetic samples. PRE measurements for the backbone amides were carried out at low molar fractions of complex (0. 15 YmoA backbone amide assignments were obtained based on published data (Biological Magnetic Resonance Bank entry 15486) (24) at pH 6.0 and 298 K. 1 H, 15 N HSQC spectra of YmoA were stepwise recorded at pH 7.5, 7.0, 6.5, and 6.0 and then at 285, 292, and 298 K. Small 1 H, 15 N chemical shift differences were observed for most residues and varied linearly with the temperature, such that the assignments at 298 K and pH 6.0 could be propagated to the other temperatures and pH values. 1 H, 15 N HSQC NMR spectra of YmoA variants are similar to those of YmoA wild type and consistent with a well folded, monomeric protein with a similar overall structure.
PRE-guided Modeling-1 H, 15 N HSQC spectra of diamagnetic and paramagnetic samples were acquired at 298 K with a recycling delay between scans of 2.5 s to ensure that magnetization recovery levels are identical for both states and using the same concentration and number of scans for both samples. NMR spectra were acquired on 600 MHz or 800 MHz Bruker spectrometers equipped with a TCI cryo-probe.
can be estimated for each residue from the line-width of the peak at half-maximum height. Calculated PRE rates were converted into intermolecular distances using the equation, where K is 1.23 10 Ϫ32 cm 6 s Ϫ2 , H is the Larmor frequency of the proton, ⌫ 2 is the experimental determined PRE rate, p b is the fraction of complex, c is the correlation time estimated from Stokes-Einstein equation. The values of ⌫ 2 were corrected by 1/p b to obtain the expected values for fully bound proteins. The correlation time for the electron-nuclear interaction was assumed to be equal to the global correlation time of the complex. PRE effects from spin labels located at positions 37 and 66 were integrated as distance restraints together with mutagenesis data and chemical shift mapping (Fig. 3) in the HADDOCK docking approach (26). Interfacial residues were selected based in the fulfillment of at least three of the following criteria: (i) significant chemical shift perturbation upon the addition of small amounts of the interacting partner, (ii) at least 30% surface accessible area, (iii) as part of a cluster of residues that forms a plausible binding patch, (iv) mutation of the residue to give rise to a properly folded protein but with null or decreased binding ability. The selected residues were defined as "active" and their solvent accessible surface neighbors or residues that were in agreement with criteria (i) and (ii) as "passive." Haddock encodes these interfacial residues in terms of ambiguous intermolecular distance restraints (AIRs) to drive the docking (26). AIRs correspond to ambiguous distance between all atoms of the source residue (active) to all the atoms of all target residues (active and passive). The AIR definitions are provided in Table 1.
The known structures of free Hha (PDB 1JW2) and H-NS 46 dimer (PDB code 1NI8) were used as starting structures in the protein docking protocol. MTSL-modified cysteines were computational designed onto the surface of Hha starting structure. The mobility of the MTSL tags was taken into account by performing ensemble averaging of the PRE-driven restraints with four different MTSL orientations. All docking runs were performed using the HADDOCK standard docking protocol of three consecutive steps: (it0) rigid body energy minimization, (it1) semi-flexible refinement in the torsion angle space, and a final water refinement in explicit solvent (water). Hinges and flexible regions, paramagnetic tags, and/or residues identified as being part of the binding interface of a given model were considered explicitly as flexible in the it1-and water MD-simulated annealing refinement steps. The hinge prediction server HingeProt (27) was used to define the hinge regions of the flexible monomers. The number of structures was increased to 2000, 400, and 400 for it0, it1, and water, respectively. Random removal of AIRs was turned on. Other parameters were left to their default values. Scoring and clustering were performed according to standard HADDOCK procedures (28). PRE effects on 15 N-HhaC18I caused by H-NS 46 with EDTA-Mn 2ϩ placed in each Cys-21 residue were not used in the modeling and, therefore, provided an independent validation of the generated structures. Theoretical intermolecular PRE rates were back-calculated from the structural models using the Solomon-Bloembergen approximation by representing the paramagnetic label by an ensemble of states and calculating the order parameters accounting for the motion of the dipolar proton-electron interaction vector (29). The c of each pose was estimated using HydroNMR (30). PRE rates were scaled by the population of the bound state and compared with the experimental values.
Side chain-directed HCACO Experiments-The standard HCACO experiment was modified to detect correlations between side-chain carbonyls and the adjacent ␤ or ␥ methylene groups. In particular C␤/C␥ excitation was achieved with a 320-s Q5 pulse centered at 35 ppm. The experiments were run at 14.1 T and 298 K in a sample containing 0.7 mM 13 C-YmoA, 20 mM sodium phosphate buffer (pH 7.5) with 100 mM NaCl, 1 mM (tris(2-carboxyethyl)phosphine), 0.2 mM EDTA, and 0.01% (w/v) of NaN 3 in the absence or in the presence of 0.3 mM H-NS 64 to obtain about 40% of bound YmoA. We used a twodimensional version of the experiment with 256 increments and 32 scans. Assignments were based on published data (Biological Magnetic Resonance Bank entry 15486).
CPMG Relaxation Dispersion Experiments-15 N-CPMG single-quantum relaxation dispersion experiments for YmoA were performed at 285K (calibrated with a methanol sample). Data sets were recorded always with two identical aliquots of the same freshly prepared protein sample on Bruker Avance III spectrometers operating at static magnetic fields of 14.1 and 18.8 T, the latter equipped with a cryo-probe.
All 15 N-CPMG relaxation dispersions were collected as a series of two-dimensional planes with 16 interleaved frequency cp values, ranging from 25 to 1000 Hz, including 2 repeats for error analysis and using a constant-time version (31) of the relaxation-compensated TROSY CPMG pulse scheme (32). Each two-dimensional spectrum was recorded with 180 and 200 complex points in t 1 dimension at 14.1 T and 18.8 T, respectively, and a delay between scans of 1.2 s. The total constant-time-CPMG relaxation T relax delay was set to 80 ms. A compensation element was applied at the beginning of each scan to suppress artifacts due to cp -dependent changes in sample temperature.
Effective relaxation rates, R 2 , eff , were calculated according to the relation (33), where I 0 is the intensity of a peak in the reference spectrum recorded without CPMG relaxation delay, and I( cp ) is the corresponding peak intensity in the spectrum measured at a frequency of cp .

Structure of the H-NS Hha Charge Zipper Complex
Dispersion profiles for each 15 N were individually and globally analyzed to obtain values of k ex ϭ k ϩ ϩ k Ϫ by fitting the Carver-Richards equation (34) using in-house Matlab scripts either to one field alone or simultaneously using data from two fields.
Electrostatic Potentials and Clustering-We used Poisson-Boltzmann electrostatic calculations and homology modeling to quantitatively identify regions of conserved electrostatic character (35,36) in the Hha/YmoA family of proteins. Homology modeling was used to generate structures for the Hha homologues with unsolved structures based on PDB code 1JW2 using SWISS-MODEL (37). All protein sequences were taken for UniProt (38).
All electrostatic potential calculations were performed using DelPhi (39) with 129 ϫ 129 ϫ 129 grid points, considering solvent dielectric constant of 80 with an ion concentration of 200 mM, whereas the protein dielectric constant was 4. A probe with a radius of 1.4 Å was used to define the dielectric boundary. Each electrostatic potential calculation was centered on each structure to ensure proper alignment of electrostatic potentials before similarity calculations.
Similarity matrices were generated based on the electrostatic similarity index as defined elsewhere (35,36). Cumulative distributions of electrostatic conservation index (ECI) were calculated using (40), Here A represents the electrostatic potential of E. coli Hha to which all other potentials B,i were compared. The electrostatic conservation index is calculated at each grid point and normalized by N, the number of electrostatic potentials comparisons, which are in total 12. The electrostatic conservation only describes the similarity of the electrostatic potential of a set of proteins to one particular protein at a given grid point. The surface projections of electrostatic conservation were generated using UCSF Chimera (41). Small-angle X-ray Scattering (SAXS) Experiments-SAXS data were measured in the beam-line BM29 (ESRF-Grenoble). Data were collected at multiple concentrations from 0.9 to 10 mg/ml at 285 K in 20 mM phosphate buffer (pH 7.5), 100 mM NaCl, 1 mM (tris(2-carboxyethyl)phosphine), 0.2 mM EDTA and 0.01% NaN 3 . A momentum transfer range of 0.032-4.93 Å Ϫ1 was measured. Repetitive measurements indicated that samples did not present radiation damage.
Because the relative populations of species in solution change as a function of the concentration, the final SAXS profiles were directly used without merging them. Initial points of the curves were discarded for the analysis to minimize interparticle interaction effects. Data processing was performed using PRIMUS (42). The theoretical SAXS curves of full-length YmoA, H-N 64 , and YmoA⅐H-NS 64 complexes were computed using the program CRYSOL (43). All theoretical curves were computed with 101 points and a maximum scattering vector of 0.5. Flexibility was accounted for by averaging ensembles of ϳ1000 structures. Theoretical curves for YmoA⅐H-NS 64 mix-tures were computed by linear combination of the theoretical curves of the individual species weighted by their predicted population in the mixture. The binding constant was obtained from fluorescence anisotropy titrations. Experimental and predicted curves were directly compared with no adjustable parameter fitting.

Results
Electrostatics in the Hha/YmoA Family of Proteins-To investigate the role of electrostatic forces in the function of Hha-like proteins, we first examined the conservation of the charge distribution across this family of proteins. For this purpose we chose 13 Hha homologues differing in sequence (37.1-98.6% identity with E. coli Hha) and net charge (Ϫ0.8 to ϩ6) and computed their electrostatic potentials using structural templates generated by homology modeling from the Hha structure (PDB code 1JW2). The isopotential surfaces shown in Fig. 1B clearly show a similar asymmetrical electrostatic potential distribution. Quantitative pairwise electrostatic similarity calculations (35) show that chromosomic Hha paralogues cluster together, whereas Hha-like proteins from plasmids and the obligate endosymbiont Wigglesworthia glossinidia form a separate cluster (Fig. 1C). Fig. 1D shows the local degree of conservation of the electrostatic potential between the 13 homologues, mapped on the Hha structure compared with the conservation in the amino acid sequence. Thus, the abundance of charged residues and their distribution seems to be a conserved feature of members of the Hha/YmoA family, suggesting a functional role.
Solution Structure of the Hha⅐H-NS 46 Complex; a Three-protein Charge Zipper-To unravel the role of electrostatics in the molecular recognition associated to Hha, we determined the solution structure of Hha bound to H-NS. H-NS contains an oligomerization and a DNA binding domain separated by a linker. Two dimerization regions connected by a long ␣-helix form the oligomerization domain (44). Hha interacts exclusively with the N-terminal dimerization domain (NTD) of H-NS. The NTD is completely included in a construct formed by residues 2-47 (H-NS 46 ) (19,45). The solution structure of the complex formed between Hha and H-NS 46 could not be determined by classical NOE-based methods because of extensive broadening of key NMR signals upon complex formation. Therefore, we turned to PRE NMR experiments as a source of structural restraints (46). This approach allows the observation of residues distant from the paramagnetic tag, even at low molar fractions of complex where intrinsic broadening associated to complex formation is minimized. Paramagnetic tags were introduced by reaction MTSL in four single cysteine variants of Hha: wild type Hha (C18) and three single cysteine mutants at positions 37, 66, and a non-native C-terminal position 74. In these mutants cysteine 18 was mutated to isoleucine. The C18I mutation had been previously shown not to affect H-NS binding (24). Paramagnetic effects were measured on 15 N-labeled H-NS 46 C21S. The native cysteine in H-NS 46 (Cys-21) was mutated to serine to avoid MTSL exchange. The labeled positions ( Fig. 2A) are located outside of the H-NS binding interface inferred by perturbations mapping (Fig. 3). The addition of Hha spin-labeled at positions 37 and 66 to 15 15 N resonances (Fig. 2B), whereas no effect was observed when MTSL was attached at positions 18 or 74 of Hha. PRE data, binding-induced broadening, and mutagenesis information were converted into structural restraints and used in a flexible Haddock docking protocol starting with the known structures of unbound Hha (PDB code 1JW2) (17) and H-NS 46 antiparallel dimer (PDB code 1NI8) (47). The solution structures were validated with an additional set of independent PREs. The structural models correctly predicted the intermolecular PRE obtained by incorporating a paramagnetic tag, EDTA-Mn 2ϩ , in each Cys-21 residue of H-NS 46 dimer (Fig. 2C) and measuring the paramagnetic effects on 15 N-labeled Hha (Fig. 2D).
Structural statistics are given in Table 2. An ensemble of structures derived from PRE-driven modeling has been deposited in the PDB (PDB code 2mw2, RCSB number RSCB104115, Biological Magnetic Resonance Bank accession code 25296). The most relevant feature shared by all the models in the ensemble is the presence of an interdigitated array of residues with alternating charges, forming an electrostatic zipper (Fig.  4). The involved residues belong to three molecules, Hha and the two molecules (A and B) that constitute the H-NS 46 Fig. 3B shows a close-up view of the electrostatic zip-per in the Hha⅐H-NS model structure. Although the electrostatic interactions in the charge zipper probably dominate, additional contacts that also contribute to the complex stability are shown in Fig. 5.
The charge zipper model explains the known effect of mutations in Hha (D48E and E25Q) and H-NS (R12K) (19,20). It also predicted that Lys-32 of H-NS should be essential for the interaction. Indeed, this prediction was experimentally confirmed: mutation to glutamine of Lys-32 in H-NS leads to the complete loss of Hha binding (Fig. 6). The K32Q mutant and the wild type form have nearly identical circular dichroism spectra, confirming that the mutation does not affect the folding of H-NS (data not shown). Similar effects were observed in YmoA and YdgT (20), suggesting that H-NS complexes formed by other members of the Hha/YmoA family involve a similar charge zipper.
We confirmed the participation of aspartic and glutamic acid side chains of YmoA in its complex with H-NS 64 , a truncated H-NS construct comprising residues 1-64, using a HCACO experiment modified to detect the correlation between sidechain carbonyl groups and their adjacent methylene groups (Fig. 6C). The experiments were done with 13     Our solution structural model of the Hha⅐H-NS complex can be compared with the recently published crystal structure of an equivalent complex (21). Despite the fact that the crystal structure has missing electron density for atoms in 50 side chains, including most of the interface region, the x-ray and NMR models have very similar backbone structures. The closest structures have a root-mean-square difference (r.m.s.d.) of ϳ1.6 Å, confirming the overall geometry of the complex. Modeling the missing side chains in the x-ray structure also leads to a charge zipper (results not shown). Fig. 7 shows a comparison of the NMR and x-ray models. The x-ray structure was obtained with an excess of Hha to saturate the two binding sites in the H-NS dimer. Based on the measured binding constant, the population of the species with two Hha molecules bound to the H-NS dimer should be very low under our experimental NMR conditions. The structure of the YmoA⅐H-NS 64 complex was predicted by homology based on the Hha⅐H-NS structure (Fig. 6D). SAXS of YmoA in the presence of H-NS 64 was in good agreement with predictions based on the structures of YmoA, H-NS 64 , and models of the YmoA⅐H-NS 64 complexes (one or two YmoA molecules per H-NS dimers) weighted according to the populations predicted by the experimental binding constant, assumed to be identical for the two sites in the H-NS dimer (Fig.  8). These results confirm the structure of the Hha⅐H-NS complex, the independent binding to the two sites, and the similarity between the complexes formed by Hha and YmoA.
The two bound Hha molecules in the crystal structure are not equivalent by symmetry. Although all the PRE-based models present the charge zipper, they also fall into two structural clusters. One of the x-ray determined sites is similar to one of the clusters, but it has a r.m.s.d. higher than 2.7 Å to the members of the second cluster. The second x-ray model shows intermediate r.m.s.d. values to the PRE-based models. The major difference between the various models and the structure of free Hha involves changes in the position of helix 4 that is displaced to allow H-NS binding and conformational changes in residues spatially close to Asp-48 (Tyr-44 and Trp-68) to allow key residues to interact with H-NS. The observed structural variability, the missing electronic density within the crystallographic Hha⅐H-NS interface, and previous NMR data (23) suggest the existence of dynamics in the Hha protein family.
YmoA Shows Concerted Pervasive Microsecond-Millisecond Dynamics-Dynamics of free YmoA was studied using CPMG relaxation dispersion measurements for the backbone amide group and at two magnetic fields (14.1 and 18.8 T). Hha samples tend to aggregate at the concentrations required for these experiments, whereas the YmoA construct used by McFeeters et al. (18) is highly soluble and stable over long periods at 1.1 mM in 20 mM sodium phosphate, 200 mM NaCl, pH 7.5 and 285 K. Fig. 9 shows representative dispersion curves. CPMG relaxation dispersion curves were analyzed as an equilibrium between two sites characterized by resonance frequencies A and B with populations p A and p B , respectively, exchanging at a rate k ex ϭ k ϩ ϩ k Ϫ (34). A preliminary analysis of the relaxation dispersion curves of individual residues showed dynamics in 12 residues with similar apparent k ex values distributed along the entire sequence: Leu-7 in helix1, Glu-24 and Lys-27 in helix 2, Asp-33, Tyr-39, Ala-42, and Asp-43 in helix 3, Tyr-55 and Ile-58 in a loop, and Val-62, Gln-64, and His-65 in the C-terminal helical region. The spatial distribution of the dynamic residues and the similar k ex values found for the exchanging residues suggest concerted dynamics affecting large regions or the entire protein. Therefore, the dispersion curves were analyzed collectively assuming that groups of residues move in a concerted way (i.e. with the same k ex and p B ). A protocol for residue incorporation, based on F-test analysis, was created. Two groups were identified, a first group including residues 55, 58, 64, and 65 that can be fitted collectively with a k ex ϭ 267 Ϯ 85 s Ϫ1 and p B ϭ 6 Ϯ 2%. These residues are located in helix 4 and the loop that connects it to helix 3, a region showing major structural differences among the different existing structures of the Hha/YmoA family.  The solid and dashed-line boxes denote residues, whose mutation completely or nearly abolishes the binding, respectively. C, surface representations of the electrostatics potential of Hha and H-NS 46 show that the complex is stabilized by charge complementary.
A second group of dynamic residues (24,27,39,42, and 43) is located in helices 2 and 3, constituting the structural core region of the Hha/YmoA family, well conserved in Hha, YmoA, and YdgT. These residues could be collectively fitted using k ex ϭ 1200 Ϯ 130 s Ϫ1 and p B ϭ 2 Ϯ 1%. Under the assumption of the two-state model, the derived kinetic constants for the process are: k ϩ ϭ 1177 Ϯ 65 s Ϫ1 and k Ϫ ϭ 23 Ϯ 8 s Ϫ1 .
Residues that could not be incorporated in any of the two groups were further analyzed using a three-state model based on the simultaneous occurrence of the two previously defined processes. A good fit was obtained, but an F-test analysis showed that the improvement with respect to an independent two-state model was not significant.
The existence of concerted dynamics raises the question of the functional coupling between YmoA dynamics and H-NS binding. To address this question, we measured CPMG relaxation dispersion of YmoA in the presence of increasing amounts (40 and 46 ) or green (Hha) sticks. In addition to electrostatics complementary (Fig. 3), the complex is also stabilized by hydrophobic contacts, flanking the salt bridge formed by Asp-48 (Hha) and Lys-32 (H-NS 46 ). ond dynamics increased from 12 to 21 (38) residues of YmoA. Although H-NS 64 binding in the presence of intrinsic dynamics in the apo form would in principle call for at least a three-state model, an F-test analysis showed that the improvement of the fit was not statistically significant with respect to the two-state model. The small concentration of complex present thus appears as a perturbation of the apoprotein dynamics, and data measured in the presence of H-NS were analyzed with a twostate model with concerted dynamics, as in the free YmoA but incorporating additional residues to the group in the core region that participate in the concerted dynamics. The similarity of the structures of the core regions of free and bound forms in the case of Hha support the idea that H-NS binding affects mostly the dynamics of YmoA.
Residues in helix 4 showed a complex response to the addition of H-NS. The structural variability in the position of helix 4 in free YmoA is probably aggravated by the formation of the complex with H-NS. The effect of H-NS 64 addition on the side chain of a key residue in the loop that connects helices 3 and 4 supports the idea of more complex structural changes in this region, although a three-state model did not improve the fit with respect to the two-state one, and no further analysis was attempted.
The population of the minor species sensed by the core residues increased from 2% in the free form to 5-6% in the presence of 6% H-NS-bound species (Table 3) Table 3). The simplest model explaining the experimental observations is that binding of H-NS is coupled with the intrinsic dynamics of YmoA, which may also have an electrostatic origin.
Ionic Strength Effects and Mutagenesis Show That YmoA Dynamics and H-NS Binding Are Related-Because of the electrostatic character of the YmoA⅐H-NS interface, we characterized the stability of the complex as a function of ionic strength (200 mM, 100 mM, and 50 mM NaCl). The dissociation constants (K D ) for the YmoA⅐H-NS 46 complex increases linearly with the salt concentration (Fig. 9), suggesting that electrostatic effects are important for complex formation but through their effect on the off-rate, as an effect of the ionic strength in the on-rate should result in strong deviations from linearity (48).
The increase in stability of the complex at low ionic strength is consistent with the charge zipper motif identified in the NMR structure of the Hha⅐H-NS complex. The high charge density in free YmoA, when not compensated by the interaction with H-NS, is expected to create intramolecular electrostatic interactions, which may be the origin for the dynamics observed in YmoA. To test this hypothesis we compared CPMG relaxation dispersion experiments of free YmoA at 200 mM, 100 mM, and 50 mM NaCl ( Fig. 10 and Table 3). When the salt concentration  was reduced from 200 mM to 50 mM NaCl, the number of core residues showing collective motion increased from 5 to 16 residues. 11 of these residues belong to helices 2 and 3, and the short loop that connects them and 4 residues belong to helix 1. Thus, by reducing the ionic strength the collective motion was extended to embrace helix 1. The increased number of residues involved in the collective dynamics is expected to result in a reduction in the timescale of the motional process. Indeed, the exchange rate was progressively reduced nearly 2-fold and 8-fold when NaCl concentration was reduced to 100 and 50 mM, respectively.
Thus, electrostatic interactions are responsible for both the intrinsic dynamics of YmoA and the association to H-NS, stabilized through a charge zipper. This connection is strengthened by the observation that mutation of residue Asp-43 to asparagine completely abolished dynamics in the microsecondmillisecond time scale even in the presence of 80 M H-NS 64. Residue Asp-43 plays a crucial role in the charge zipper, and mutation of the equivalent residue in Hha (Asp-48) completely abolished binding to H-NS (data not shown).

Discussion
Hha-like proteins modify the capacity of H-NS to silence specific DNA regions (8). H-NS binds and silences DNA   (44). A functional result of this interaction is that the Hha⅐H-NS complexes preferentially repress horizontally acquired genes. The mechanism by which Hha binding induces changes in selectivity with respect to the genes that are repressed is not completely clear. Hha-like proteins and the NTD of H-NS contain an unusually large proportion of charged residues and a marked electrostatic asymmetry. The positive region of H-NS NTD has been suggested to interact with DNA according to the model of the DNA-H-NS complexes presented by Arold et al. (44) on the basis of the x-ray structure of the H-NS oligomers. According to this model, Hha-like molecules bind and mask the positively charged region of H-NS NTD. However, because of the electrostatic asymmetry of Hha itself, a new positively charged patch (from Hha) is displayed (19,20), although at a different position with respect to the H-NS oligomer axis. The presence of Hha-like proteins is believed to change the "matching condition" between DNA binding and H-NS oligomerization and, therefore, the DNA silencing specificity (20,21,50). Mutagenesis in the positively charged region of Hha clearly demonstrates that this region is critical for transcriptional repression but not for H-NS binding (21,24).
Binding of Hha to a strongly charged region of H-NS imposes the need for an electrostatic complex. However, electrostatic complexes are usually associated with a poor selectivity. In contrast, the tight network of multiple complementary interactions from three different proteins in the Hha complex with a H-NS dimer, conforming the observed charge zipper, ensures an exquisite selectivity (19,20). A charge zipper combining intraand intermolecular interactions stabilizing the translocation pore of TatA protein has been recently described (51).
Although electrostatic interactions play a clear role in intermolecular interactions (with H-NS or DNA), intramolecular electrostatic interactions may induce forces distorting the low energy packing of the helical core. The pervasive dynamics observed in the free form of YmoA may originate from the balance between electrostatically driven distortions and the hydrophobic interactions restoring forces. The fact that dynamics has been preserved by evolution suggests it may result in functional advantages.
Our results do not provide direct information on the possible conservation of the Hha/YmoA dynamics in the complexes, although the missing residues and structural variations between the two sites in the x-ray structure suggest that this may be the case. A direct functional correlation between dynamics and complex formation is supported by the changes observed in the number of residues involved in the microsecond-millisecond dynamics and the population of the minor conformation. This correlation is also supported by the effect of the D43N mutation of YmoA that completely abolished H-NS binding as well as s-s dynamics of the free form.
We can only speculate about how Hha/YmoA dynamics may influence function. We suggest at least three possible effects. (a) Due to its positively charged surface, a direct interaction between Hha and DNA was suggested (21). The existence of pervasive motions in the free molecule would increase the entropic penalty upon binding and may decrease spurious binding of Hha to DNA. (b) Hha binding to the NTD domain and the intrinsic dynamics of Hha-like proteins may change the flexibility of H-NS oligomers (52), thus modifying the selectivity. The fact that horizontally acquired DNA regions silenced by H-NS tend to be longer than those in the core genome (53) could make the flexibility of H-NS oligomers an important control element (50). (c) YmoA/Hha dynamics may facilitate binding to and/or dissociation of charge zipper complexes. Considering that electrostatic interactions are long range, the formation of the charge zipper may be compromised by local minima and dynamics may facilitate the optimization of the complex. Similarly, complete dissociation of an electrostatic complex may require large displacements of the charged elements before the interaction is effectively lost, and dynamics, if retained in the complex, may facilitate its dissociation. The similarity in the exchange rates observed in the presence and in the absence of H-NS would be compatible with a dynamic coupling between the intrinsic motions of Hha-like molecules and the formation or dissociation of the complex with H-NS.
Author Contributions-T. N. C. performed experiments leading to the determination of the structure of the complex and did the electrostatic analysis of the Hha family. J. G. generated and measured mutants to test the models and prepared complexes. P. B. contributed to SAXS measurement and analysis and to the general discussion. O. M. measured and analyzed CPMG relaxation dispersion experiments and contributed to the design and interpretation of experiments. M. P. conceived and coordinated the study and wrote the paper with input from all the other authors.