Structural flexibility of RNA as molecular basis for Hfq chaperone function

Abstract

In enteric bacteria, many small regulatory RNAs (sRNAs) associate with the RNA chaperone host factor Q (Hfq) and often require the protein for regulation of target mRNAs. Previous studies suggested that the hexameric Escherichia coli Hfq (Hfq_Ec) binds sRNAs on the proximal site, whereas the distal site has been implicated in Hfq–mRNA interactions. Employing a combination of small angle X-ray scattering, nuclear magnetic resonance and biochemical approaches, we report the structural analysis of a 1:1 complex of Hfq_Ec with a 34-nt-long subsequence of a natural substrate sRNA, DsrA (DsrA₃₄). This sRNA is involved in post-transcriptional regulation of the E. coli rpoS mRNA encoding the stationary phase sigma factor RpoS. The molecular envelopes of Hfq_Ec in complex with DsrA₃₄ revealed an overall asymmetric shape of the complex in solution with the protein maintaining its doughnut-like structure, whereas the extended DsrA₃₄ is flexible and displays an ensemble of different spatial arrangements. These results are discussed in terms of a model, wherein the structural flexibility of RNA ligands bound to Hfq stochastically facilitates base pairing and provides the foundation for the RNA chaperone function inherent to Hfq.

INTRODUCTION

The Escherichia coli host factor Q (Hfq) was originally identified as an accessory factor of the phage Qß replicase >40 years ago, whereas its role in bacterial post-transcriptional regulation became evident only more recently (1). Hexameric Hfq protein (protomer: 102 aa) belongs to the class of Sm-like proteins with multiple functions in eukaryotic RNA metabolism (2,3). In bacteria, small regulatory trans-acting RNAs (sRNAs) can modulate different stress responses through post-transcriptional regulation (4). In general, sRNAs either prevent ribosome loading onto the mRNA by base pairing with or in the vicinity of the ribosome binding site or act by an ‘anti-antisense’ mechanism and abrogate intramolecular inhibitory stem–loop structures that block ribosome binding (5). As many sRNAs display imperfect and non-contiguous target complementary, the requirement for Hfq in riboregulation has mainly been attributed to its RNA chaperone function, which appears to facilitate the interaction between the sRNA and the cognate mRNA (2,6).

Escherichia coli Hfq (Hfq_Ec) homologues have been found in a number of Gram-negative and Gram-positive bacteria. The hexameric Hfq proteins possess an evolutionarily conserved common core consisting of amino acid (aa) residues 7–66, whereas there is considerable variation at the C-terminal end (7). Several high-resolution structures of Hfq proteins of different origin bound to RNA oligonucleotides have been published. The first 3D structure of the 77 aa Staphylococcus aureus Hfq protein in complex with the short 5′-AU₅G-3′ oligoribonucleotide revealed that the poly(U) oligonucleotide was bound in a circular fashion along the inner basic rim of the central pore (8). Afterwards, a mutational analysis performed with Hfq_Ec further indicated that sRNAs bind to the same site, which was termed the proximal face of the hexamer (9). In addition, the crystal structure of Salmonella typhimurium Hfq (protomer: 102 aa) in complex with an (U)₆ RNA oligonucleotide not only corroborated poly(U) binding to the proximal side but also revealed that a free 3′ hydroxyl group of RNA is important for high-affinity binding to the protein, which in turn may impact on the stability of RNA substrates (10).

In contrast to uridine-containing sequences, a recent structural study revealed binding of a poly(A) tract to the distal face of Hfq through six tripartite binding motifs (11). Similarly, a crystallographic analysis demonstrated binding of an A(G)₃A aptamer to the distal side of the Bacillus subtilis Hfq protein, but the overall RNA structure and protein–RNA interaction patterns differed from those of the Hfq_Ec–poly(A) complex (12). Moreover, the crystal structure of a ternary complex between a C-terminally truncated Hfq_Ec variant, ADP and an A(U)₆A RNA oligonucleotide derived from the Hfq_Ec binding region of the sRNA DsrA (see below) has been recently reported (13). In this structure, the ADP was bound on the distal R site(s) of Hfq_Ec, similar to the adenines of poly(A) (11), while the (U)₆A part of the RNA oligonucleotide was bound on the proximal side in a circular fashion, in a manner comparable to A(U)₆G binding to S. aureus Hfq (8). Moreover, the 5′ proximal A nucleotide of A(U)₆A was found inserted into a distal R site of another closely packed Hfq hexamer (13).

A paradigm for positive regulation by an sRNA entails the translational activation of E. coli rpoS mRNA (Supplementary Figure S1), encoding the stationary phase sigma-factor RpoS, by the sRNA DsrA. At low growth temperatures, the DsrA–rpoS mRNA interaction counteracts an inhibitory stem–loop structure that impedes ribosomal access to the ribosome binding site of rpoS (14). In vivo DsrA–rpoS duplex formation at low temperature requires Hfq (15) and the CsdA helicase (16) and creates an RNase III cleavage site within the duplex that would prevent reuse of DsrA (17). DsrA binds to Hfq with a 1:1 stoichiometry (18) on the proximal face (9), whereas rpoS mRNA is believed to be recruited on the distal side of Hfq via A-rich element(s) in the mRNA leader (11,19,20). Mechanistic insights into the sequence of events toward Hfq-mediated DsrA–rpoS duplex formation have been recently obtained by the Woodson laboratory. Soper et al. (6) provided evidence that Hfq increases the stability of the DsrA–rpoS complex by binding to the upstream A-rich regions, which is in line with a spectroscopic study (21), indicating that rapid co-binding of two RNA ligands and their release from Hfq precedes duplex formation. The need for Hfq to cycle off its binding site on DsrA before or during annealing with rpoS (22) could result from the fact that at least part of the Hfq binding site on DsrA (23) base pairs with rpoS mRNA (24). While the dedicated sRNA and mRNA binding surfaces on either site of the Hfq hexamer could readily serve to transiently increase the local concentration of two RNA substrates, the inherent capacity of Hfq to induce conformational changes in RNAs (25–28) could, together with the possibility of several RNA-binding modes within the hexamer, lead to different spatial arrangements of RNA substrates in individual Hfq–RNA complexes, which may in turn increase the likelihood for productive duplex formation.

To address this question, we used small angle X-ray scattering (SAXS), nuclear magnetic resonance (NMR) and biochemical studies together with available information on high-resolution X-ray structures to assess the biophysical parameters and shape of Hfq_Ec and a truncated version thereof, Hfq_Ec65 (aa 1–65), in complex with a 34-nt segment of DsrA domain II (DsrA₃₄). Taken together, these data revealed that binding of DsrA₃₄ to both the full-length and the truncated protein results in an ensemble of complexes where DsrA₃₄ is bound in a structurally variable manner on the proximal face of a given hexamer. These results are discussed in light of the RNA chaperone function of Hfq_Ec in riboregulation.

MATERIALS AND METHODS

Synthesis and purification of Hfq_Ec and Hfq_Ec65

The Hfq proteins used in this study were purified as previously described (29). For NMR experiments, the proteins were uniformly isotope labelled with ¹⁵N and ¹³C.

Enzymatic probing of DsrA₃₄ with RNase V1

Aliquots containing 0.15 pmol of [³²P]-5′end-labelled DsrA₃₄ were incubated in buffer (50 mM bicine pH 8.8, 100 mM NaCl, 250 mM KCl and 0.5 mM MgCl₂) in the presence or absence of Hfq_Ec (1.5 pmol as hexamer) at 37°C for 10 min. Then, 7.5 × 10⁻³ (Figure 2B, lanes 2 and 4) and 1.5 × 10⁻² (Figure 2B, lanes 3 and 5) units of RNase V1 were added and the incubation at 37°C was continued for 10 min. Then, 2× RNA loading dye was added to stop the reaction. For complete RNA hydrolysis, 0.15 pmol of radioactively labelled DsrA₃₄ was incubated at 85°C for 10 min in 50 mM sodium carbonate (NaHCO₃/Na₂CO₃), pH 9.0, in the presence of 1 µg yeast tRNA. The RNA hydrolysis was terminated by addition of 1 volume 2× RNA loading dye. The samples were analysed on a 10% polyacrylamide gel containing 8 M urea and the labelled RNA fragments were visualized using a PhosphoImager.

Figure 2.

NMR spectra of DrsA₃₄ and enzymatic probing of DsrA₃₄ with RNase V1 in the presence and absence of Hfq_Ec, respectively. (A) NMR spectra of free DrsA₃₄ and in complex with Hfq_Ec65. (B) In vitro RNase V1 cleavage of DsrA₃₄ was performed in the presence (lanes 2 and 3) and in the absence (lanes 4 and 5) of Hfq_Ec. 7.5 × 10⁻³ (lanes 2 and 4) and 1.5 × 10⁻² (lanes 3 and 5) units of RNase V1 were added. Lanes 1 and 6, sequence ladder obtained by alkaline hydrolysis of [³²P]-5′ end-labelled DsrA₃₄. The numbers denote nucleotide positions in DsrA₃₄. The arrows denote RNase V1 cleavage sites on DsrA₃₄ in the absence (C) and presence (D) of Hfq_Ec, respectively, derived from enzymatic probing.

Hfq_Ec–DsrA₃₄ complex preparation and screening of solution conditions

The DsrA₃₄ RNA was chemically synthesized by Thermo Fisher Scientific, Inc., and was further purified on 8% polyacrylamide–8 M urea gels using standard procedures. DsrA₃₄ samples were heated to 75°C for 15 min and immediately cooled down on ice, whereas Hfq_Ec65 was kept at room temperature. Samples were centrifuged for 30 min at 13 000g at room temperature. Before protein–RNA complex assembling, protein and RNA concentrations were determined with a NanoDrop ND-1000 UV/Vis spectrophotometer at 280 and 260 nm, respectively.

To identify suitable conditions for long-term stabilization of Hfq_Ec65–DsrA₃₄ complexes, a series of sample buffer conditions were screened in 96-well plates (Intelli-Plate 96-3, Art Robbins Instruments, Inc.) containing 1 µl of Hfq_Ec65 at a final concentration of 3.5 mg/ml in duplicate. These drops were set up using a Phoenix robot (Art Robbins Instruments, Inc.). Upon addition of stoichiometric amounts of DsrA₃₄ to Hfq_Ec65, the samples were left at room temperature (22°C) and visually examined at time 0 and over the course of 12 h using a Minstrel™ drop imager (Rigaku Americas Corporation). The drops without visible precipitation (14 conditions out of 96) were analysed by dynamic laser light scattering (DLS) experiment. The best four candidate conditions were selected on the basis of the lowest polydispersity index (PdI) derived from the DLS experiment. Then, a time course at different temperatures was performed using DLS to monitor the condition and the time point where the Hfq_Ec65–DsrA₃₄ complex samples acquired a PdI ≤ 15%. A long-term stability and monodispersity of Hfq_Ec–DsrA₃₄ and Hfq_Ec65–DsrA₃₄ complexes were achieved in 50 mM phosphate buffer, pH 7.2, containing 200 mM NaCl and in 50 mM bicine buffer, pH 8.8, containing 100 mM NaCl and 250 mM KCl.

Dynamic laser light scattering

Typically, samples with protein concentration ranging from 1 to 10 mg/ml were centrifuged 30 min at 13 000g to eliminate large aggregates. DLS was measured at a scattering angle of 90° on a DynaProNanoStar™ (Wyatt Technology) using a 10 μl microcuvette and a laser emitting at λ = 662 nm. The data collection was conducted with laser intensity in autoregulation mode, using 10 acquisition frames of 10 s/acquisition at 25 or 35°C. The data were analysed using online Wyatt software to determine the sample PdI, where sample with an average PdI ≤ 15% was considered as monodisperse and suitable for structural characterization. The apparent translational diffusion coefficient, D, calculated from the intensity autocorrelation functions fits, was converted into Stokes radii (R_s) according to the Stokes–Einstein equation (Eq. (1)):

(1)

with R being the gas constant, T the temperature, N Avogadro’s number and n the solvent viscosity. Control experiments carried out with Hfq_Ec and Hfq_Ec65 alone yielded experimental Stokes radii similar to the ones computed from atomic coordinates of rigid structures by the program HYDROPRO (30) that employs shell-model methodology. In this calculation, the crystal structure of E. coli Hfq (PDB accession code: 1HK9) (7) was used as the structural (PBD) file for Hfq_Ec65, while the SAXS model described by Beich-Frandsen et al. (29) was employed for the Hfq_Ec. The primary solution property derived here was translational diffusion coefficient (D) that was converted into Stokes radii (R_s) according to Eq. (1). For running HYDROPRO, the radius of the atomic elements (AER) was defined to be 2.9 Å. The number of values of the radius of the mini bead (NSIG) was set to be 6, and the radius of the mini beads in the shell was set to 2.0 for SIGMIN and 3.0 for SIGMAX. The solvent viscosity was defined to be 0.0200 poise. Partial specific volume (psv) for protein alone was set to be 0.732 cm³/g, while for Hfq_Ec–DsrA₃₄ and Hfq_Ec65–DsrA₃₄ complexes, the psv values were calculated to be 0.706 and 0.696 cm³/g according to the relative contribution of the pvs of RNA DsrA₃₄ (0.550 cm³/g) in these respective protein–RNA complexes at stoichiometric ratio (1:1). The structural PDB files for the complexes were obtained from the SAXS models. The temperature was defined to be 298 or 308 K.

Nuclear magnetic resonance

For the NMR experiments, the complexes were prepared between unlabelled RNA and doubly ¹⁵N/¹³C isotope-labelled protein. All NMR experiments were performed at 37°C with a Varian Inova 600 MHz spectrometer and with a Varian Inova 800 MHz spectrometer. Before data collection, the Hfq_Ec65–DsrA₃₄ complex samples were incubated overnight to optimize the sample monodispersity level as suggested by the DLS experiment (Supplementary Figure S3). NMR spectra were processed with NMRPipe (31) and analysed with Sparky (32) software. The protein samples for the NMR experiments were prepared by size-exclusion chromatography in 50 mM Na₃PO₄, pH 7.2, 200 mM NaCl and concentrated to ∼1 mM Hfq_Ec65 (with respect to the monomer). Samples were supplemented with 10% (vol/vol) D₂O to provide the deuterium signal for the field-frequency lock, as well as 0.1–0.2% (wt/vol) NaN₃ to inhibit bacterial growth. 1D ¹H-NMR spectra were obtained using the WATERGATE (33) method for solvent suppression. Backbone signal assignment for the C-terminally truncated mutant Hfq_Ec65 was obtained by a suite of standard (sensitivity enhanced) 3D triple resonance experiments: HNCA (34), HN (CO)CA (35), HNCACB (36) and HNCO (34) were recorded for sequential backbone chemical shift assignment of Hfq_Ec65 as described by Beich-Frandsen et al. (29). ¹⁵N relaxation times (T₁, T₂) were measured using gradient sensitivity-enhanced 2D methods with ¹H detection (37,38). Sequential backbone signal assignments in the apo form were available for Hfq_Ec65 for aa residues 6–65 (29) and were used as starting points for signal assignment for the Hfq_Ec65–DsrA₃₄ complex. The fingerprint region of the ¹⁵N-HSQC spectra of Hfq_Ec65 was superimposed onto the ¹⁵N-HSQC spectra of the Hfq_Ec65–DsrA₃₄ complex (Figure 1A). Both spectra were well resolved and largely similar, which allowed for sequential assignment of most of the resonances of the Hfq_Ec65–DsrA₃₄ complex spectra by simple peak comparison. The remainder of backbone signal assignment of the complex between Hfq_Ec65 and DsrA₃₄ was obtained employing HNCA (34)/HN(CO)CA (35) and HNCO (34) experiments. When necessary, ambiguous ¹H–¹⁵N assignments of the Hfq_Ec65–DsrA₃₄ complex were resolved by inspection of their attached ¹³Cα or ¹³C′ shifts.

Figure 1.

NMR analysis of the Hfq_Ec65–DsrA₃₄ complex. (A)¹³C-, ¹⁵N-labelled Hfq_Ec65 was used for solution NMR studies. Assignments are indicated for the complex Hfq_Ec65–DsrA₃₄. Superposition of the ¹H–¹⁵N HSQC spectra of Hfq_Ec65 RNA free form (blue) and in complex with DsrA₃₄ (red). (B) Chemical ¹H–¹⁵N shift differences between Hfq_Ec65 and Hfq_Ec65–DsrA₃₄ complex (calculated as sqrt(Δ¹⁵N²+ 25Δ¹H²) for assigned peaks plotted against residue positions. Multiple sets of signals observed for certain residues are indicated in grey (two signals) and white (three signals), respectively. The Hfq_Ec65 secondary structure is indicated and colour coded according to the position on the proximal (blue) and distal face (yellow). The distal and proximal portion of the β2-strand are termed β2 and β2′, respectively; η1 denotes the small 3₁₀-helix. The red line indicates the chemical shift differences threshold (double the average shift change) above which the chemical shift differences were considered significant. (C) Ribbon diagram of Hfq_Ec65 hexamer. Residues with prominent chemical shift differences are colour coded in red and mapped onto the 3D structure of Hfq_Ec65 derived from the crystal structure pdb1HK9 (7); residues belonging to proximal and distal sites are colour coded as in (B). Residues Gly4 and His57, used in SAXS modelling as contact points with DrsA₃₄, are presented as blue spheres. Semitransparent solvent accessible surface of one Hfq_Ec65 protomer is displayed in grey.

SAXS, ab initio shape determination and molecular modelling

SAXS experiments were performed at the EMBL BioSAXS beamline X33 (39,40) at the DORIS III synchrotron storage ring (DESY, Hamburg, Germany). A1M PILATUS detector (DECTRIS, Switzerland) was installed at a sample-to-detector distance of 2.7 m. At the X-ray wavelength λ = 1.5 Å, this beamline setup records the scattering profiles in the range of momentum transfers between 0.01 and 0.6 Å⁻¹ (s = 4π sinθ/λ, where 2θ is the scattering angle). The data reduction and analysis followed the standard procedures using the ATSAS program package (41). All samples were in 50 mM bicine buffer, pH 8.8, containing 100 mM NaCl and 250 mM KCl and the measurements were performed at 35°C. Hfq_Ec alone was measured at concentrations 4.1 and 15 mg/ml; the complex Hfq_Ec–DsrA₃₄ was measured at 2.3, 4.3 and 9.2 mg/ml, while Hfq_Ec65 alone was measured at 2.5, 3.8 and 9.4 mg/ml and the complex Hfq_Ec65–DsrA₃₄ at 2.6, 4.0 and 5.1 mg/ml. The protein concentrations were determined with a Nanodrop ND-1000 UV/Vis spectrophotometer at 280 nm immediately before X-ray exposure.

Overall parameters, i.e. the forward scattered intensity I(0), radius of gyration R_g and the excluded volume of the particles, were calculated using PRIMUS (42). The program GNOM (43) was employed for calculating the pair distance distribution function and to estimate the maximum dimension (D_max) of the particle. No concentration dependence was found for the Hfq_Ec65–DsrA₃₄ complex, whereas a continuous increase of overall parameters for Hfq_Ec–DsrA₃₄ complex was observed. For further analysis, the data from the lowest concentration of complexes, compatible with 1:1 stoichiometry of the complex, were used. The low-resolution shapes were reconstructed ab initio using the program DAMMIN (44).

To assess the flexibility of DsrA₃₄ in the complexes, the ensemble optimization method (EOM) was used (45). A random pool of 10 000 models for the Hfq_Ec–DsrA₃₄ complex was generated from the atomic coordinates of Hfq_Ec (29) and a chain of dummy residues representing the RNA molecule using RanCh (45). As the scattering density of a nucleotide is about 2.8 times larger than that of an aa residue, and as the volume of a nucleotide is two times larger than the volume of an aa residue, 136 dummy residues with the form-factor multiplier equal to one were used to represent DsrA₃₄. This approach enables EOM to emulate flexible RNA moieties for the Hfq_Ec–DsrA₃₄ complex. A subset of the ensemble was selected using a genetic algorithm such that the calculated scattering of the mixture agreed with the experimental data. The R_g distributions of the selected ensembles were obtained by repeating the selection process multiple times. To better account for the relative difference in scattering density between RNA and protein, we generated a full-atom model of DsrA₃₄ in unfolded conformation. This model on one hand was used to simulate a scattering profile of DsrA₃₄ and on the other allowed us to validate the stoichiometry of the complexes and to estimate the effective difference in scattering density between the Hfq_Ec–DsrA₃₄ complex and Hfq_Ec.

Finally, the rigid body modelling of the protein–RNA complexes was done using the program SASREF (46). To model DsrA₃₄ in the complexes, a tentative 3D model was generated using the programme package RNABuilder (47,48) and subsequently divided into six fragments, in agreement with the persistency length of one unfolded single RNA molecule (49,50), generating in this way a chain of interconnected rigid subunits. Contacts were defined to guarantee the interactions of DsrA₃₄ on the proximal face of Hfq hexamer, with the poly(U) stretch (Supplementary Figure S2) interacting with one of the ‘YKH’ motifs, while the next three nucleotides contact aa residues 2–4 of Hfq, as suggested by the NMR studies (Figure 1A–C). In the case of Hfq_Ec65–DsrA₃₄, the crystal structure of E. coli Hfq (PDB accession code: 1HK9) (7) was used as the rigid subunit, while the model described by Beich-Frandsen et al. (29) was used for the Hfq_Ec–DsrA₃₄ complex.

RESULTS

Hydrodynamic properties of Hfq–DsrA complexes assessed by DLS

The protein–RNA complexes were formed between Hfq_Ec or Hfq_Ec65 and a 34-nt fragment of DsrA (Supplementary Figure S2), spanning DsrA domain II (23). Binding of DsrA₃₄ to both Hfq_Ec and Hfq_Ec65 was confirmed by electrophoretic mobility shift assays (not shown). DLS was used to determine the PdI and the Stokes’ radii (R_s) of both complexes (Supplementary Figure S3). Samples with average PdI values ≤15% were considered monodisperse and suitable for structural characterization. A long-term stability and monodispersity of Hfq_Ec–DsrA₃₄ and Hfq_Ec65–DsrA₃₄ complexes were achieved in 50 mM phosphate buffer, pH 7.2, containing 200 mM NaCl and in 50 mM bicine buffer, pH 8.8, containing 100 mM NaCl and 250 mM KCl. The DLS experiments revealed that the Hfq_Ec–DsrA₃₄ complex is monodisperse up to a final concentration of 3 mg/ml and displayed an R_s of 46 ± 1 Å, whereas the R_s for Hfq_Ec alone was 35 ± 2 Å. The Hfq_Ec65–DsrA₃₄ complex showed monodispersity up to 9 mg/ml with an R_s of 37 ± 2 Å, whereas that of free Hfq_Ec65 was 30 ± 2 Å. Furthermore, global analysis of the DLS autocorrelation functions (Supplementary Figure S3) corroborated the monodispersity of the complex assembled at 1:1 stoichiometry for the samples used for NMR and SAXS experiments.

NMR indicates binding of DsrA₃₄ to the proximal face of the Hfq hexamer with 1:1 stoichiometry

NMR was used to gain more information on the structural and dynamic properties of the Hfq_Ec65–DsrA₃₄ complex in solution. The complex is in a ‘slow exchange situation’ reflecting the high affinity of DsrA₃₄ for Hfq_Ec65. Binding of the RNA to the symmetrical hexameric Hfq_Ec65 changed the appearance of the NMR spectra. Upon the interaction with DsrA₃₄, the original 6-fold symmetry of the Hfq_Ec65 hexamer is broken and additional sets of cross-peaks, corresponding to the different subunits of the Hfq_Ec65 hexamer, started to emerge in the ¹⁵N-HSQC spectra (Figure 1A). In theory, all the individual subunits of Hfq_Ec65 are non-equivalent. However, in practice, only some of the subunit residues in direct contact with RNA experience significant shifts relative to apo-Hfq_Ec65. Typically, only two or in very few cases three sets of signals were detected for a number of residues. This is compatible with a binding mode in which the DsrA₃₄ predominantly interacts with one protomer of the hexameric Hfq_Ec65. Due to the high affinity (nM) of the Hfq_Ec65–DsrA₃₄ interaction, no exchange between the different non-equivalent species could be detected by NMR. When compared with Hfq_Ec65 alone, significant ¹⁵N–¹H^N chemical shift differences between the signal sets of the interacting and non-interacting subunits were observed in the Hfq_Ec65–DsrA₃₄ complex in the N-terminal α-helix, in β-strand β2′, in the proximity of the central pore, within the central pore that comprises the YKH motif (aa 55–57) and in the C-terminal β-strand β5 (Figure 1B and C). The YKH motif of S. aureus Hfq was shown to interact with poly-U (8) and is anticipated to serve as the primary binding site for sRNAs in Hfq_Ec (9). In addition, upon complex formation with DsrA₃₄, the amino acids A2–G4 of the partially disordered N-terminus (29) became apparently conformationally stabilized and thus detectable. As anticipated for close contacts with aromatic nucleobases, high upfield ¹⁵N shifts [∼95 (folded to 125) ppm for residue 2 and ∼90 (folded to 120) ppm for residue 4] were observed (Figure 1A). Hence, the proximal side residues A2 and G4 are apparently in contact with RNA.

Besides the N-terminus, a number of clusters of significant ¹⁵N–¹H^N chemical shift changes were identified. One large ¹⁵N–¹H^N chemical shift change maps to residue Ala14, residing in the central region of the N-terminal α-helix on its solvent-exposed side (Figure 1B and C). Ala14 could sense the conformational changes at the N-cap of the α-helix, caused by RNA interacting with the N-terminal aa residues 2–4. As expected, the YKH motif with a shell of surrounding residues experiences a difference in chemical environment upon binding of DsrA₃₄ (Figure 1B and C). The third group of notable chemical shifts locates to the outer rim of the Hfq hexamer, i.e. to aa residues 34–37, in a conformationally labile region at the end of the β2-strand and turn connecting to the following β2′-strand (Figure 1B and C). In summary, the chemical shift analyses suggested that the RNA is bound to the central pore and that it emerges from the proximal side, tethered by interactions with the N-terminal aa residues 2–4, which precede the α1-helix on the proximal side and point toward the central pore (Figure 1B and C). In support, the Hfq_EcK31A mutant protein, which was shown to be strongly impaired in poly(A) binding to the distal site (51), was not impaired in binding to DsrA₃₄, corroborating the chemical shift analyses (not shown).

The effect of adenine binding in the distal R binding site was independently investigated by titration with ATP (not shown). This ruled out simultaneous binding of DsrA₃₄ to both the proximal and the distal side of Hfq_Ec and a 2:1 stoichiometry as observed in complex with the RNA molecules used in the study by Wang et al. (13). This different behaviour can be best rationalized by the lack of a 5′ terminal A in our DsrA₃₄ construct, which is obviously required for binding to the distal side (11).

Next, the ¹⁵N relaxation rates were measured to obtain information on the hydrodynamic radius, thus the effective molecular mass and consequently on the stoichiometry of the Hfq_Ec65–DsrA₃₄ complex. The NMR ¹⁵N relaxation measurements revealed for Hfq_Ec65 (43.2 kDa) a T₂ = 40.0 ± 2.1 ms (R₂ = 25.0 ± 1.3 s⁻¹) at 14.1 T (i.e. 600 MHz ¹H frequency). For the Hfq_Ec65–DsrA₃₄ complex (50.5 kDa), T₂ was determined with 29.0 ± 6.5 ms (R₂ = 34.5 ± 7.7 s⁻¹) at the same field strength for the signals of the core region. This agreed with the molecular weight increase expected for a 1:1 stoichiometry between Hfq_Ec65 and DsrA₃₄. Thus, the NMR studies strongly suggested that under these conditions, complex formation predominantly occurs in a 1:1 ratio. Moreover, no imino ¹H resonances indicative of stable base pairs in double-stranded nucleic acids, which are observable in the 1D ¹H NMR spectra of free DsrA₃₄, were detected in the 1D ¹H NMR spectra of the Hfq_Ec65–DsrA₃₄ complex (Figure 2A). The lack of stable base pairing in the Hfq_Ec65–DsrA₃₄ complex suggested an extended and single-stranded conformation of DsrA₃₄, where the imino ¹Hs are prone to solvent exchange in the absence of base pairing. Similarly, the H₃ imino Hs of the poly U tract at the 5′-end of DsrA₃₄, which are expected to be bound to the central pore of Hfq_Ec65 (similar to A(U)6G binding to S. aureus Hfq (8)), are outward-facing towards the solvent and thus not expected to yield observable signals. To obtain additional experimental support for an extended conformation of DsrA bound to Hfq, enzymatic probing with the double-stand specific endoribonuclease V1 was performed. This analysis revealed that the stem–loop structure in DsrA₃₄ is at least partially formed in the absence of Hfq, whereas no indications for base pairing were obtained when a 1:1 complex of Hfq_Ec65 hexamer and DsrA₃₄ was treated with the enzyme (Figure 2B–D). These chemical probing experiments are in agreement with a FRET study by Večerek et al. (28), showing that both Hfq_Ec and Hfq_Ec65 induce structural changes in full-length DsrA.

Low-resolution shape of Hfq_Ec65–DsrA₃₄ and Hfq_Ec–DsrA₃₄complexes

SAXS measurements were performed in parallel with the complexes Hfq_Ec–DsrA₃₄ and Hfq_Ec65–DsrA₃₄ at three different sample concentrations (Hfq_Ec–DsrA₃₄: 2.3, 4.3 and 9.2 mg/ml and Hfq_Ec65–DsrA₃₄: 2.6, 4.0 and 5.1 mg/ml). To avoid aggregation or concentration-induced artefacts, we used only the lowest concentrations of both complexes, corresponding to a 1:1 stoichiometry, which is in agreement with a study by Updegrove et al. (18) who reported that DsrA domain II and Hfq_Ec form 1:1 complexes.

The processed experimental data are presented in Figure 3A and the overall molecular parameters are summarized in Table 1. The data recorded for Hfq_Ec and Hfq_Ec65 are fully compatible with the results obtained by Beich-Frandsen et al. (29). The calculated scattering profile of the crystal structure of Hfq_Ec65 (PDB accession code: 1HK9) (7) agreed with the experimental SAXS data with a discrepancy χ = 1.06 (Figure 3A), indicating that Hfq_Ec65 forms a stable hexamer in solution and that the doughnut shape of the crystal structure is retained. For the full-length protein, the model described by Beich-Frandsen et al. (29), where the C-terminal regions stretch outward from the central core, also agreed with the present experimental data (χ = 1.65; Figure 3A). The models generated by Beich-Frandsen et al. (29) were therefore used in the subsequent analysis of the Hfq_Ec–DsrA₃₄ and Hfq_Ec65–DsrA₃₄ complexes.

Figure 3.

Analysis of the SAXS data. (A) Comparison of the experimental SAXS curves (empty circles) with the CRYSOL calculated scattering curves for Hfq_Ec65 (curve 1) and Hfq_Ec (curve 2) using the crystal structure pdb1HK9 (7) and the model reported by Beich-Frandsen et al. (29). Typical fits of the ab initio models (curves 3 and 5) and rigid body models (curves 4 and 6) against the experimental data for Hfq_Ec65–DsrA₃₄ (fits with red lines) and Hfq_Ec–DsrA₃₄ (fits with green lines) complexes, respectively. (B) Comparison of the pair distance distribution functions for Hfq_Ec and Hfq_Ec65 alone and the corresponding complexes with DsrA₃₄. (C) Frequency distributions of R_g obtained in EOM to assess the flexibility of DsrA₃₄ in complex with Hfq_Ec65 (red) and Hfq_Ec (green). Dashed curves (i) and (iii) correspond to the distributions for the random pool of 10 000 conformers and curves (ii) and (iv) to the distributions for the optimized ensembles.

Table 1.

SAXS parameters calculated from experimental data compared with Stokes radii from DLS experiment and with Stokes radii calculated from the ab initio models using the program HYDROPRO (30)

Sample	Rg [SAXS] (Å)	Dmax [SAXS] (Å)	Rs [DLS] (Å)	Rs [HYDROPRO] ab initio model (Å)
HfqEc	34 ± 1	105 ± 5	35 ± 2	36
HfqEc–DsrA34	44 ± 2	145 ± 10	46 ± 1	48
HfqEc65	23 ± 1	63 ± 3	30 ± 2	30
HfqEc65–DsrA34	33 ± 2	120 ± 8	37 ± 2	38

As shown in Figure 3B, the distance distribution function P(r) of the Hfq_Ec–DsrA₃₄ complex revealed a skewed appearance typical for an elongated particle. The significant increase in R_g and D_max after binding of DsrA₃₄ again suggested that the 34-nt RNA molecule is fully extended after binding to the protein. The observed R_g and D_max are incompatible with the presence of a stable stem–loop and can thus be explained by flexible and disordered single-stranded RNA (possibly in a fast exchange equilibrium with other loosely base-paired species). For Hfq_Ec65 and the Hfq_Ec65–DsrA₃₄ complex, the changes in R_g and D_max follow the same pattern as observed for Hfq_Ec and the Hfq_Ec–DsrA₃₄ complex. The fact that truncation of the C-terminus did not change the behaviour of DsrA₃₄ binding and that Hfq_Ec65 was previously shown to be proficient in DsrA binding (28) suggested that C-terminal residues are not involved in Hfq–DsrA₃₄ interactions.

To obtain direct information on the structural changes upon RNA binding, low-resolution ab initio models (∼25 Å) were reconstructed from the SAXS data. The model of Hfq_Ec65–DsrA₃₄ shows an elongated particle consisting of a bulk region with an extended arm. The crystal structure of Hfq_Ec (aa 1–71; PDB accession code: 1HK9) (7) can be docked into the bulk part of the model and the arm, apparently consisting of DsrA₃₄, is stretching outward (Figure 4A–C). The elongated ab initio model for the complex is further supported by the agreement of a Stokes radius of 38 Å determined from this model by the program HYDROPRO (30) with the measured value (37 ± 2 Å) using DLS (Table 1). Similarly, the model for the Hfq_Ec–DsrA₃₄ complex displays an arm protruding outwards the protein part represented by the crystal structure of Hfq_Ec with the C-terminal residues added as in (29) (Figure 4D–F).

Figure 4.

Ab initio models of Hfq_Ec65–DsrA₃₄ and Hfq_Ec–DsrA₃₄ derived from the SAXS data. (A–C) Models of Hfq_Ec65 (blue) and (D–F) Hfq_Ec (green) represented with their respective solvent accessible surfaces were docked manually into the bulk region of the ab initio shapes. The Hfq_Ec65 atomic coordinates were derived from the crystal structure pdb1HK9 (7). The Hfq_Ec structure was derived from the same atomic coordinates with the C-terminal segment modelled using SAXS data (29).

Flexible arrangements of DsrA₃₄ in both complexes

As the NMR, SAXS and enzymatic probing studies suggested that DsrA₃₄ is unfolded, extended and lacks defined secondary structure when bound to Hfq_Ec and Hfq_Ec65, the EOM (45) approach was used to assess the flexibility and the accessible conformational space of the complexes. EOM takes into account flexibility by allowing for the coexistence of different conformations of the complex in the population of molecules in solution, contributing to the experimental scattering pattern. In EOM, a large pool of random configurations was generated and ensembles were selected from this pool using a genetic algorithm, such that the average computed scattering over the ensemble fitted the experimental scattering data. As shown in Figure 3C, the R_g distributions of the random pools for both samples are rather broad, whereas the selected ensembles both for Hfq_Ec–DsrA₃₄ and Hfq_Ec65–DsrA₃₄ display relatively narrow distributions, where the most compact and most extended configurations are not present. The EOM ensembles fitted the scattering data rather well, with a discrepancy χ of 1.3 and 1.0 for Hfq_Ec–DsrA₃₄ and Hfq_Ec65–DsrA₃₄, respectively (fits not shown). This indicated that a confined range of conformationally variable Hfq_Ec65–DsrA₃₄ and Hfq_Ec–DsrA₃₄ complexes exists in solution. The major insight from EOM was that DsrA₃₄ in the complex is unfolded, as the experimental data could not be fitted by compact RNA structures.

The confined flexibility of the complexes permitted rigid body modelling against the SAXS data to visualize the spatial arrangement of DsrA₃₄ relative to the hexameric core of the protein. As shown in Figure 3A, the rigid body models generated by SASREF (46) agreed with the experimental data with a typical discrepancy χ = 1.2 for Hfq_Ec–DsrA₃₄ and χ = 1.1 for Hfq_Ec65–DsrA₃₄. Ten superimposed independently constructed rigid body models for Hfq_Ec–DsrA₃₄ and Hfq_Ec65–DsrA₃₄ revealed that in both complexes the RNA extends like an arm away from the hexamer core, thus leading to an elongated shape. The orientation and the contour length of the RNA arms may vary such that DsrA₃₄ explores a manifold of configurations around the core (Figure 5). The obtained conformational space of the RNA molecule is confined to the proximal side of the hexamer, in agreement with the EOM results.

Figure 5.

Rigid body models of Hfq_Ec65–DsrA₃₄ and Hfq_Ec–DsrA₃₄ complexes. (A–C) Ten typical models for the Hfq_Ec65–DsrA₃₄ complex are superimposed and shown at three different plane rotations. Hfq_Ec65 atomic coordinates were derived from the crystal structure pdb1HK9 (7) (solvent accessible surface representation in blue). The models were generated using SASREF (46) and SAXS data combined with constraints obtained from NMR. (D–F) Ten typical models of the Hfq_Ec–DsrA₃₄ complex are superimposed and shown at three different plane rotations. The Hfq_Ec structure (solvent accessible surface representation in green) was derived from the high-resolution crystal structure pdb1HK9 (7) with the C-terminal segment modelled using SAXS data (29). In both complexes, one DsrA₃₄ molecule is represented as full-atom model with atoms colour-coded to highlight the overall structure of the RNA in the complex, while the other nine models are presented with their solvent accessible surface (in different tonalities ranging from light yellow to deep orange). In both cases, the best DsrA₃₄ models that fit the SAXS data display an extended conformation. The average root mean square deviation between the DsrA₃₄ portions in the models, (60 ± 20 Å), is similar to that between the selected models in the EOM ensembles.

The theoretical Stokes radii (R_s) calculated for the Hfq_Ec–DsrA₃₄ and Hfq_Ec65–DsrA₃₄ complexes are equivalent to the R_s derived experimentally from the DLS data (Table 1) and agree with DsrA₃₄ binding to Hfq_Ec and to Hfq_Ec65 in extended conformation with a 1:1 stoichiometry. Moreover, the RNA does not interact with the C-terminus of Hfq_Ec (Figure 5D–F), suggesting that the C-terminal region of Hfq_Ec may not affect the overall conformation of DsrA₃₄ and vice versa. These results corroborated previous synchrotron radiation circular dichroismn spectroscopy studies, which revealed that DsrA₃₄ did not alter the structure of Hfq_Ec (29). In contrast, the presence of a longer RNA fragment comprising the 5′ upstream region and the immediate coding region of the hfq gene, which was shown to require the C-terminal extension of Hfq_Ec for binding (28), did lead to an ordering of the C-terminus of Hfq_Ec (29).

DISCUSSION

We used complementary biophysical and structural biology methods to study the arrangement of domain II of the sRNA DsrA on the surface of the Hfq_Ec hexamer at a stoichiometry ratio as reported by Updegrove et al. (18). Previous electrophoretic mobility shift experiments (22) showed that Hfq can bind to full-length DsrA with a 2:1 ratio, although binding of a second Hfq hexamer required high concentrations of the protein. In addition, a recent biophysical study indicated likewise a 2:1 ratio between Hfq and DsrA (13). Moreover, crystallographic studies by Wang et al. (13) revealed that a short oligonucleotide A(U)₆A was bound with the (U)₆A segment to the proximal side of one Hfq hexamer, whereas the 5′A nucleotide was found inserted into a distal R site of a second Hfq hexamer. In our study, the increase of SAXS overall parameters for Hfq_Ec–DsrA₃₄ at higher concentrations could be explained by the formation of 2:1 (or 2:2) Hfq_Ec–DsrA₃₄ complexes, which would be in agreement with (13). As no concentration effects were observed for the Hfq_Ec65–DsrA₃₄ complex (Supplementary Table S1), this interaction is probably not RNA mediated and therefore could occur through interactions of C-terminal residues of Hfq_Ec. However, given that Hfq appears to be rather limiting for riboregulation during normal growth conditions (52), the biological significance of the observed higher Hfq–RNA complexes remains uncertain. In addition, the studies by Wang et al. (13) are at variance with the observations by Updegrove et al. (18). Using different experimental approaches, the latter authors showed by mimicking the cellular environment in terms of concentration and ionic strength conditions that Hfq_Ec and Hfq_Ec65 form a 1:1 complex with DsrA or DsrA domain II in solution.

Similarly as observed by Updegrove et al. (18), our studies strongly suggest that Hfq_Ec and Hfq_Ec65 form a 1:1 complex with DsrA₃₄. Furthermore, the NMR study identified three clusters of residues affected by binding of DsrA₃₄ to Hfq_Ec65. (i) The YKH motif located in the central pore of the Hfq hexamer together with the surrounding residues residing in the adjacent β-strands. In the YKH motif, K57 and H58 participate in RNA binding in S. aureus Hfq (8), whereas Y56 is involved in aromatic stacking interactions stabilizing H58 in an appropriate orientation for interaction with the RNA base. (ii) The N-terminal segment of Hfq, where structuring of aa residues 2–4 was observed. These residues precede the α-helix located on the proximal face of the Hfq hexamer (7) (Figure 1C), and their interaction with RNA might lead to concomitant perturbation of the hydrogen bonding network in the α-helix itself and to some subtle repacking of the α-helix as a whole. (iii) As a consequence, the residues spatially adjacent to the C-cap of the α-helix, i.e. the aa residues 34–37 located at the end of β2-strand and in the turn connecting to the following β2′-strand, are also affected. Although the reason(s) for these chemical shift variations is unclear, it seems possible that this loop has some conformational variability and can assume different conformations in different environments. This hypothesis is supported by our NMR studies as well as by the crystal structure (pdb1HK9 (7)) in that the 34–37 region displays some flexibility, which might result in different conformations and hence in the observed chemical shifts variations in the Hfq_Ec65–DsrA₃₄ complex. These NMR studies corroborate mutational analyses (9), which likewise indicated that DsrA binds to the proximal site of Hfq.

Biophysical studies (18) have recently shown that a ternary complex between Hfq, DsrA domain II and polyA₁₈, the latter of which was shown to bind to the distal site (11), is rather unstable. However, Soper et al. (6) provided evidence that Hfq forms persistent ternary complexes when the two ligands are complemetary. These experiments also indicated that at least transient co-binding between Hfq, rpoS mRNA and DsrA contributes to riboregulation, which can be reconciled with the observed rapid binding and release of Hfq from ternary complexes during annealing (21). Hence, the two distinct binding sites on Hfq could at least transiently increase the local concentration of both the sRNA and the mRNA, which could in turn facilitate the interaction of ligands with medium affinities and circumvent the need for a high concentration of either substrate. Taken the above-mentioned studies together with the observation that stable binding of longer mRNAs by Hfq appears to involve contacts with the intrinsically flexible C-termini of Hfq (28,29) and with the structural data presented herein, we suggest the following model for the function of Hfq in riboregulation. The model entails five steps: (i) fast binding of both RNA ligands (21) to Hfq leading to their increased local concentration (Figure 6), followed by (ii) restructuring of the substrates by Hfq (25,26,28). Hereby, the Hfq-induced conformational fluctuations in both the sRNA and mRNA may occur separately. While the intrinsically unstructured region of the C-terminus of Hfq (29) may contribute to mRNA binding and may induce conformational changes in these ligands, it seems to be dispensable for doing so in sRNAs, as the C-terminally truncated Hfq_Ec65 was proficient to alter the structure of DsrA (28). In addition, at least in the Hfq_Ec–DsrA₃₄ complex, the DsrA₃₄ does not interact with the C-terminus of Hfq_Ec (Figure 5D–F); (iii) initiation of base-pairing between the ligands (21). In this step, the inherent capacity of Hfq to present RNA in extended conformation and different spatial orientation(s) (Figures 5D–F and 6) would allow to cover a large space over Hfq, which would firstly favour the encounter and secondly the initial annealing of two cognate RNAs. In addition, it has been recently shown that Hfq can bind to the U-rich sequence following the rho-independent terminator of sRNAs (53). Thus, different binding sites of Hfq on one sRNA could be likewise important in terms of increasing the geometric variability of these ligands on the proximal face of Hfq. Moreover, not only the sRNA may be presented in different orientations on the proximal face but—given the presence of six tripartite-binding motifs (11)—also the mRNA bound on the distal side may adopt different orientations. Hence, the geometric variability of the ligands in individual complexes would facilitate annealing in a stochastic manner; (iv) ligand release from Hfq (21,22) followed by (v) stable duplex formation between the RNA substrates. At this juncture, RNA displacement from Hfq may occur by invasion of other RNAs (54). In this model, fast binding, restructuring and the presence of conformationally variable complexes would not only ensure a fast turnover of cognate RNAs but could also provide a means of proofreading for non-cognate ligands, i.e. where initial base pairing cannot take place. Similarly, Doetsch et al. (55) suggested that a human immunodeficiency virus-1 derived Tat peptide accelerates annealing of two RNA ligands by changing the population distribution of RNA structures to favour an annealing-competent RNA conformation.

Figure 6.

Model of Hfq RNA chaperone function. The sRNA displayed by the full-atom model with atoms colour-coded is shown bound to the proximal face of Hfq_Ec (green, solvent accessible surface). Through conformational fluctuations, the sRNA can cover a larger conformational space (nine representative DsrA₃₄ models are displayed with their solvent accessible surface coloured as in Figure 5). The mRNA is bound on the distal side (polyA₉ orange flat cartoon representation) to one of the six tripartite binding motifs as shown in the crystal structure pdb3GIB (11). The model of a hypothetical mRNA chain is displayed in orange oval cartoon. Hfq_Ec acts by restructuring the mRNA (6), which may be accomplished by the conformationally flexible C-termini (29). The structural variability of both RNAs in a transient ternary 1:1:1 complex (18) would allow to sample large spaces. In this way, Hfq_Ec would not only act as a platform for binding and by increasing the local concentration of both ligands but would also serve to promote their flexibility and consequently successful annealing in a stochastic manner.

As mentioned above, the increased local concentration of two RNA ligands together with the structural plasticity of RNA bound to the RNA chaperone Hfq would allow the RNA to sample large spaces and therefore enhance the probability of successful annealing between RNAs. This capacity of Hfq appears to be of particular importance when the free energy of sRNA–mRNA pairing interactions is low, i.e. when the complementarity of Hfq-dependent sRNAs and their target mRNAs is non-contiguous, as it is the case for many studied sRNA–mRNA pairs in GC-rich Enterobacteriaceae (56). In contrast, Hfq is dispensable for duplex formation between the sRNA IstR-1 and tisAB mRNA in E. coli (57) as well as for base pairing between RNAIII and the target mRNA sa1000 in S. aureus (58). In both cases, the free energy of base-pairing is ∼2–3-fold higher than observed for Hfq-dependent sRNA–mRNA duplexes in E. coli. Experimental support for the link between the Hfq requirement and the free energy of sRNA–mRNA pairing has also been obtained by Woodson et al. (21). Overexpression of DsrA resulted in increased rpoS translation even in the absence of Hfq, whereas ArcZ and RprA, which are also known to stimulate RpoS synthesis, were unable to do so. These authors further showed that DsrA binds the rpoS leader more tightly in the absence of Hfq than RprA and ArcZ and concluded that the stability of the RNA duplex between rpoS mRNA and the sRNA rather than the kinetics of formation is important.

In conclusion, the proposed molecular mechanism for Hfq-mediated RNA annealing is reminiscent of that suggested for intrinsically unstructured proteins and their interactions with protein partners. In these proteins, flexible, intrinsically disordered regions are believed to provide conformational fluctuations, which can facilitate intermolecular interactions, forming complexes with high specificity and relatively low affinity. This is critical for processes in which not only specific association but also subsequent dissociation of binding partners is required (59).

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online: Supplementary Table 1 and Supplementary Figures 1–3.

FUNDING

Special Research Program (SFB17) on ‘Modulators of RNA fate and function’ by the Austrian Science Fund [F1722 to K.D.-C. and F1720 to U.B.]; Bundesministerium für Bildung und Forschung grant SYNC-LIFE [contract number 05K10YEA to D.I.S.]; European Union FP7 e-Infrastructure grant WeNMR [contract number 261572 to D.I.S.]. Funding for open access charge: Austrian Science Fund.

Conflict of interest statement. None declared.