Distinct oligomeric forms of the Pseudomonas aeruginosa RetS sensor domain modulate accessibility to the ligand binding site

Authors

  • Florence Vincent,

    Corresponding author
    1. Architecture et Fonction des Macromolécules Biologiques, UMR6098, CNRS et Universités Aix-Marseille I et II, 163 avenue de Luminy, 13288 Marseille, France.
    Search for more papers by this author
  • Adam Round,

    1. Institut de Biologie Structurale J.P. Ebel 41, rue Jules Horowitz 38027 Grenoble Cedex 1, France.
    Search for more papers by this author
  • Aline Reynaud,

    1. Architecture et Fonction des Macromolécules Biologiques, UMR6098, CNRS et Universités Aix-Marseille I et II, 163 avenue de Luminy, 13288 Marseille, France.
    Search for more papers by this author
  • Christophe Bordi,

    1. Laboratoire d'Ingénierie des Systèmes Macromoléculaires, UPR9027, CNRS-IMM, Université de la Méditerranée, 31 Chemin Joseph Aiguier, 13402 Marseille cedex 20, France.
    Search for more papers by this author
  • Alain Filloux,

    Corresponding author
    1. Imperial College London, Division of Cell and Molecular Biology, Centre for Molecular Microbiology and Infection, South Kensington Campus, Flowers Building, London SW7 2AZ, UK.
    Search for more papers by this author
  • Yves Bourne

    1. Architecture et Fonction des Macromolécules Biologiques, UMR6098, CNRS et Universités Aix-Marseille I et II, 163 avenue de Luminy, 13288 Marseille, France.
    Search for more papers by this author

E-mail florence.vincent@afmb.univ-mrs.fr; Tel. (+33) (0)4 91825566; Fax (+33) (0)4 91266720.

E-mail a.filloux@imperial.ac.uk; Tel. (+44) (0)20 7594 9651; Fax (+44) (0)20 7594 3069.

Summary

Bacterial two-component regulatory systems (TCSs) sense environmental stimuli to adapt the lifestyle of microbial populations. For many TCSs the stimulus is a ligand of unknown chemical nature. Pseudomonas aeruginosa utilizes the closely related RetS and LadS sensor kinases to switch between acute and chronic infections. These sensor proteins antagonistically mediate biofilm formation through communication with a central TCS, GacA/GacS. Recently, it was shown that RetS modulates the GacS sensor activity by forming RetS/GacS heterodimers. LadS and RetS are hybrid sensors with a signalling domain consisting of a 7-transmembrane (7TMR) region and a periplasmic sensor domain (diverse intracellular signalling module extracellular 2, DISMED2). The 2.65 Å resolution crystal structure of RetS DISMED2, called RetSp, reveals three distinct oligomeric states capable of domain swapping. The RetSp structure also displays two putative ligand binding sites. One is equivalent to the analogous site in the structurally-related carbohydrate binding module (CBM) but the second site is located at a dimer interface. These observations highlight the modular architecture and assembly of the RetSp fold and give clues on how homodimerization of RetS could be modulated upon ligand binding to control formation of a RetS/GacS heterodimer. Modelling the DISMED2 of LadS reveals conservation of only one ligand binding site, suggesting a distinct mechanism underlying the activity of this sensor kinase.

Introduction

Two-component regulatory systems (TCS) are crucial in the perception of environmental stimuli and are required for bacterial adaptation to changing environmental conditions (Stock et al., 2000). Whereas the TCS are widespread in bacteria, and can be found in plants, they are absent from mammalian genomes. Moreover, many TCSs trigger a series of events contributing to colonization of the host by bacterial pathogens. In this respect they are attractive targets for the development of novel antimicrobials (Matsushita and Janda, 2002).

In the past few decades, wide bacterial genome sequencing led to the identification of hundreds of TCS (D'Souza et al., 2007; Galperin and Nikolskaya, 2007). Among them, the sensor histidine-kinase protein contains a specific domain, called the sensor domain, which is responsible for perception of the environmental stimulus. Whereas the kinase domain is cytoplasmic, many sensor kinases are membrane proteins with a sensor domain exposed to the periplasm in Gram-negative bacteria or to the extracellular medium in Gram-positive bacteria. Detection of a specific stimulus by the sensor domain results in autophosphorylation through the kinase domain, and the phosphoryl group is then transferred onto a response regulator, which in turn modulates the gene expression profile to allow bacterial adaptation to a specific environment (Stock et al., 2000). In some cases, the response regulator has an enzymatic activity rather than being a DNA binding protein (Galperin, 2010). Whereas the series of events in the regulatory cascade is well documented, very little is known about the structure of the sensor domain, the architecture of the ligand binding site and associated conformational changes.

In the last few years only a very limited number of periplasmic/cytoplasmic ligand binding domains have been characterized. Among those, Escherichia coli CitA (Reinelt et al., 2003) and DcuS (Cheung and Hendrickson, 2008) sensing domains are involved in citrate and fumarate sensing, respectively, Vibrio harveyi LuxQ detects quorum sensing molecule AI-2 (Neiditch et al., 2006), Salmonella typhimurium PhoQ responds to Mg2+ concentration (Chamnongpol et al., 2003) and Pseudomonas putida TodS binds toluene (Busch et al., 2009).

The opportunistic pathogen Pseudomonas aeruginosa is responsible for a wide range of acute and chronic infections (Murray et al., 2007). The transition to chronic infection is accompanied by formation of bacterial biofilm communities (Costerton et al., 1999). Interestingly, many reports have described a balance between the expression of molecular determinants involved in chronic infection (biofilm) and those involved in acute infection (cytotoxicity). In P. aeruginosa the key regulatory components involved in this decision making process are the hybrid sensor kinases LadS and RetS (Goodman et al., 2004; Laskowski et al., 2004; Ventre et al., 2006). These two sensor kinases feed into the central TCS GacS/GacA, in which the response regulator GacA directly controls expression of the two small RNAs, rsmY and rsmZ (Kay et al., 2006; Brencic et al., 2009). These sRNAs titrate out the translational repressor RsmA (Pessi et al., 2001; Brencic and Lory, 2009), which results in a balanced expression between biofilm-associated genes, such as the polysaccharide pel and psl genes involved in biofilm matrix production (Friedman and Kolter, 2004; Vasseur et al., 2005), and virulence genes, such as type III secretion system (T3SS) genes involved in cytotoxicity (Bleves et al., 2005; Engel and Balachandran, 2009). More precisely, upon activation of the RetS pathway, polysaccharide gene expression is repressed whereas T3SS gene expression is induced (Goodman et al., 2004). A retS mutant displays a hyperbiofilm phenotype and is poorly cytotoxic. Conversely, activation of the LadS pathway has the opposite effect and is marked by upregulation in pel gene expression and downregulation of T3SS genes. A ladS mutant is thus defective in biofilm formation but highly cytotoxic (Ventre et al., 2006). Both LadS and RetS sensor kinases have not been attributed to a cognate response regulator but are known to intersect with the GacS/GacA TCS (Goodman et al., 2004; Ventre et al., 2006; Bordi et al., 2010). While the way LadS interferes with GacS/GacA is as yet unknown, it has been shown that RetS inhibits GacS activity by forming a RetS/GacS heterodimer (Goodman et al., 2009).

RetS and LadS are hybrid sensor kinases, which emerge as representative members of a large cluster of proteins identified in the family of the 7TM receptors (Anantharaman and Aravind, 2003). These 7TM receptors are characterized by a sensor domain, which is an N-terminal extracellular/periplasmic domain that belongs to the DISMED2 description in the PFAM classification. This domain is predicted to adopt a β-sandwich fold reminiscent of the carbohydrate binding domain (CBM) superfamily, as confirmed by the resolution of the first crystal structure of RetSp (Jing et al., 2009).

To gain insight into the biological implication of this family of sensor domains, we have solved the 2.65 Å resolution structure of the periplasmic sensor domain of RetS (RetSp). In contrast to the previous structure of RetSp, our study shows two distinct domain folds. A closed monomer and an open domain-swapped dimer forms dictated by flexibility of a hinge loop region. Complementary small angle X-ray scattering (SAXS) and multi-angle light scattering (MALS) studies suggest the presence of a predominant dimeric form and a low abundant tetramer in solution, in agreement with the oligomeric arrangement observed in crystalline RetSp. Most importantly, RetSp structure highlights two potential ligand binding sites with one being buried at the dimer interface, which is not commonly found in the structurally-related CBMs. Yet a model of the LadS sensor domain (LadSp) shows conservation of a single ligand binding site. Based on these new structural insights, a model on how the interaction network between RetS and GacS/GacA may modulate the activity of this signalling pathway is discussed.

Results and discussion

Overall fold of RetSp

The engineered RetSp domain consists of 142 residues ranging from Pro44 to Glu185, preceded by a TEV cleavage sequence, a N-terminal 6×-His tag and a four-residue flag. The 2.65 Å resolution RetSp structure shows well-defined electron density map for most of the protein regions (Experimental procedures, Table 1).

Table 1.  Data collection and refinement statistics.
  • a. 

    Rmerge = (ΣhklΣi|Ihkl − (Ihkl)|/ ΣhklΣi[Ihkl]).

  • b. 

    Rcryst = Σhkl||Fo| − |Fc||/Σhkl|Fo|.

  • c. 

    Rfree is calculated for randomly selected reflections excluded from refinement.

Crystal parameters 
 Space groupI4132
 Cell dimensions 
  a = b = c (Å)174.2
  Nb of Mol/AU2
Data quality 
 Wavelength (Å)0.9792 (ID23EH1)
 Resolution of data (Å)87–2.65
 (Outer shell)2.79–2.65
 Unique reflections13 418
 aRmerge (outer shell)0.099 (0.517)
 Mean (I)/σ(I) (outer shell)24.5 (7.7)
 Completeness (outer shell) %100 (99)
 Multiplicity (outer shell)20.8 (17.2)
Refinement 
 Protein atoms2 225
 Solvent waters66
 bRcryst0.19
 cRfree0.23
 r.m.s. deviation 1–2 bonds (Å)0.014
 r.m.s. deviation 1–3 angles (°)1.64

Crystals of RetSp contain two molecules in the asymmetric unit, each showing a distinct polypeptide fold (Fig. 1A–C). A first molecule displays a single closed domain fold, with dimensions of 40 Å × 25 Å × 24 Å, which consists of two β-sheets flanked by one α-helix inserted between β1 and β2 and a C-terminal α-helix (Fig. 1A). The topology of the β-sheets, each consisting of five (β1, 3, 8, 5 and 6) and four (β2, 9, 4 and 7) anti-parallel β-strands, is reminiscent of the classical β jelly-roll fold (β-sandwich) found in the CBM (Boraston et al., 2004; Guillén et al., 2010), as reported in the structure of RetSp (Jing et al., 2009). In the second molecule, the last three β-strands (β7–9) exchange between two adjacent subunits to form a stable domain-swapped dimeric assembly with dimensions of 72 Å × 40 Å × 26 Å (Fig. 1B). The critical difference in domain fold results from the different conformation of sequence-conserved residues Pro135 to Arg144 located between β6 and β7, which can act as a β-hinge in either a closed β-bend conformation or an open extended β-conformation (Fig. S1). The hinge point distinguishing the β-turn and linear conformational forms of this β-hinge region lies within the stretch of residues Pro140-Leu141-Pro142-Ser143. In fact, the swapped dimer associates with two closed monomers, forms a tightly packed tetramer that might represent a biologically important assembly for RetS regulatory mechanism (Fig. 1D) (see below).

Figure 1.

Overall view and assembly changes of RetSp. Ribbon diagram showing the nine-stranded β-sandwich fold (yellow) labelled β1 to β9 and helix α1 (red).
A. The monomeric form. Two blue arrows delimit the missing region between strands β6 and β7.
B. The swapped dimer (yellow and grey), viewed in a similar orientation.
C. The closed monomer and open swapped dimer present in the asymmetric unit. The brown and orange arrows indicate the position of the hinge region in the swapped dimer and the face-to-face four-stranded β-sheet from a closed and open monomer. The loops and turns connecting the β-strands are in green. The closed monomer and the two extended monomers of the swapped dimer are labelled 1, 2 and 3 respectively.
D. The tetramer oligomeric form, in the same orientation. The swapped dimer is interacting with two monomers through their four-stranded β-sheet. Monomers represented as grey ribbon are symmetric molecules.
E. The same ribbon diagram as (C) with the aromatic residues involved in the four-stranded β-sheet interface highlighted as orange sticks. The van der Waals surfaces of the aromatic residues are shown as orange dots.
The N- and C-termini are labelled. Figures 1–4 were drawn with Pymol (Delano, 2002).

Superposition of a RetSp single domain and RetSp dimer gives a root mean square (r.m.s.) deviation of only 0.5 Å for 86 Cα atoms from residues Gln48 to Ser133, which excludes the last three C-terminal β-strands. In turn, virtually all of the residues responsible for critical structural interactions or forming accessible surface areas are conserved with changes occurring at the two termini and within the β6–β7 loop region. Overall, comparison of the two RetSp conformations suggests that the Pro135 to Arg144 hairpin is the structurally favoured β-hinge conformation and the higher stability of the extended Pro135 to Arg144β-hinge structure is compensated by 12 hydrogen bonds, as well as 33 van der Waals contacts. In contrast to the increased dimer stability of RetSp, the variable conformation of the β-hinge region has intrinsic flexibility in the globular monomeric form, as shown by the lack of interpretable density over this loop region.

RetSp oligomerization state

In addition to the closed monomer and domain-swapped dimer, inspection of the crystal packing environment reveals the presence of a second dimeric interface in the tetramer that buries the face on the four-stranded β-sheet from a closed and open monomer (Fig. 1C–E). These findings are consistent with the reversible dimeric assembly of RetSp with a submicromolar dissociation constant previously reported (Jing et al., 2009). The buried surface area represents 30% of the total surface of the RetSp monomer, consistent with the value found in a biological complex (Jones and Thornton, 1996). To clarify the oligomeric state of RetSp in solution, further analysis was performed using complementary biochemical and biophysical techniques. Size-exclusion chromatography (SEC)-MALS experiments were carried out on RetSp at different concentrations. While RetSp at high concentration (7 mg ml−1) showed the presence of a dimeric assembly with a molecular mass of 32 kDa and a polydispersity of 1.94 ± 0.29 (Fig. S2), a tetrameric form of RetSp along with a monomeric and dimeric forms could be observed at lower concentrations (3.4 to 4.9 mg ml−1), consistent with gel filtration experiments (data not shown). Monomeric, dimeric and tetrameric forms of RetSp were also observed by native gel electrophoresis (data not shown). However, these experiments were not able to discriminate between the swapped dimer and the dimeric interface involving the four-stranded β-sheet packed face to face in the tetramer.

To characterize the predominant oligomeric species of RetSp in solution, SAXS experiments were carried out, which revealed that RetSp behaves as a dominant dimeric assembly in solution with possible tetramers as a low abundant species. The Rg, molecular weight and volume match closely the two dimeric arrangements (dimer 1–2 and swapped dimer 2–3) observed in crystalline RetSp (Fig. 1) while the monomer is approximately half the size observed experimentally (Table 2).

Table 2.  SAXS parameters of RetSp.
 Rg (nm)Molecular weight (kDa)Volume (nm3)χ
Experimental2.1233.953 
Dimer 1–22.0631.6452.966
Dimer 2–32.4332.84812.051
Monomer1.716.02120.655
Tetramer2.5864.39832.450

To confirm the presence of the dimeric and tetrameric assemblies and to determine which conformation was predominantly present under physiological conditions, the calculated theoretical scattering form of each conformation was fitted to the experimental curves. The arrangement observed in dimer 1–2 was most likely the physiologically relevant species giving the best fit, with a χ-value of 2.966 (Table 2). In contrast, the monomer is too small and the swapped dimer 2–3 is too large and appears more extended and open than the size and shape observed by SAXS (Fig. 2). Although the dimer 1–2 closely matches the experimental data, systematic deviations are apparent around 2 nm−1 resulting in a χ-value of only 2.966. As the samples were expected to contain some polydispersity, attempts to improve the fits by testing a mixture of the available oligomeric forms (tetramer, dimers 1–2 and 2–3 and monomer) using Oligomer (Konarev et al., 2003) was undertaken and resulted in a fit showing mostly (93%) dimer 1–2 with a small amount (7%) of the tetrameric form. The use of swapped dimer 2–3 did not provide a satisfactory match with the experimental data. The overall agreement with the SAXS fitting curves in favour of dimer 1–2 suggests that the dominant species reflects a dimer in solution. The arrangement of monomers seen in dimer 1–2 could represent a physiologically relevant entity. However, the small systematic deviations of the fit from the experimental data are likely due to the polydispersity as well as the possibility of conformational mobility of some flexible loop regions resulting in slightly increased values for the experimental parameters (Table 2). As to the presence of a small amount of tetramer in solution, this suggests the possible existence of the swapped dimer 2–3 bound to two closed monomers as observed in the crystal structure.

Figure 2.

The RetSp averaged curve of normalized intensity versus scattering angle (s = 4πSinθ/λ in nm) is represented as blue dots with grey error bars (calculated from Poisson counting statistics). The experimental data are compared with the fit obtained from the theoretical profiles calculated from the non-swapped dimer (green), the swapped dimer (red), the tetramer (cyan) and the closed monomer (pink). The ribbon diagrams of RetSp are shown with corresponding colours.

Altogether these results suggest that RetSp can behave as a monomer, dimer and tetramer most probably involving the four-stranded β-sheet in solution. Thus, RetS provides another example of coexistence of dominant oligomeric forms beyond the existence of a dimeric assembly as previously reported (Jing et al., 2009). While most domain-swapped proteins show a swapped domain from either the N- or C-terminus, only a few cases have been reported in which part or half of the molecule is swapped as observed for RetSp (Liu and Eisenberg, 2002). The structural differences observed between the two RetSp conformations should document how a single domain fold with an unstable bent β-hinge could be projected into a more stable extended β-strand exchange. Moreover, the coexistence of these two domain folds in crystalline RetSp along with the tetrameric arrangement suggests a possible modular mechanism controlling the function of these regulatory sensor domains. In the swapped-dimer form, the N- and C-termini point towards the same face of RetSp, consistent with the topology of the RetS hybrid sensor. Whether the domain-swapped form of RetSp has any physiological relevance to its ligand binding activity remains unclear. While several examples support a functional role of domain swapping (Bennett et al., 1995), this oligomerization mechanism involving closed and extended monomers might be regarded as a consequence of truncation of the whole RetS protein.

Superposition of a RetSp monomer with that independently solved by Jing and collaborators (Jing et al., 2009; PDB accession code 3JYB) reveals a similar overall structure (r.m.s.d. of 0.691 Å for 122 Cα residues). However, conformational rearrangements of the protruding β6–β7 loop and the C-terminal region seem to modulate dimer formation. Indeed, a model of the RetSp dimer involving two tightly packed four-stranded β-sheets based on the structure solved by Jing and collaborators creates steric clashes with residues from the two facing β6–β7 loop regions, preventing formation of a similar dimeric assembly (Jing et al., 2009). Thus, the previously solved structure of RetSp could reflect a monomeric conformation of RetSp while large conformational changes of the β6–β7 loop region, identified as a β-hinge region, are required to promote the dimeric and tetrameric assemblies.

Structural similarities with carbohydrate binding domains

The β-sandwich fold of RetSp is related to the large CBM superfamily in which surface exposed aromatic side chains provide hydrophobic platforms to specifically recognize glycan chains (Boraston et al., 2004; Guillén et al., 2010). A search for structural homologues of RetSp in DALI (Holm and Sander, 1997) reveals numerous unknown domains appended to β-galactosidases, as well as representative members from the CBM superfamily. Among the highest scores are the CBM from families 4, 6, 29 and 15. In family 4, the CBMs belong to Cel9B protein from Cellulomonas fimi (CBM4Cf; PDB code: 1GU3) and Lam16A from Thermotoga maritima (CBM4Tm; PDB code: 1GUI) and are bound to β1,4 or β1,3-linked gluco-oligosaccharides (Boraston et al., 2002). A CBM6 appended to an endoglucanase 5A from Cellvibrio mixtus binds xylopentaose and glucan tetrasaccharide on two opposite binding faces (CBM6Cm; PDB code: 1UXX and 1UY0) (Henshaw et al., 2004). In family 29, the CBM belongs to a non-catalytic modular protein from Pyromyces equi and is bound to mannohexose (CBM29PE; PDB code: 1GWM) (Charnock et al., 2002). Finally, a CBM appended to a xylanase10c from Pseudomonas cellulosa and in complex with xylopentaose (CBM15Pc; PDB code: 1GNY) (Pell et al., 2003) exemplifies family 15 (Szabo et al., 2001).

Superposition of the RetSp closed monomer with these top-ranked CBMs shows a high degree of structural similarity in the core β-sandwich, while loop regions connecting β-strands display large conformational adaptations. Yet, the overlay reveals that the loop β1–β2 in RetSp harbouring helix α1 is the primary structural determinant that distinguishes RetSp from members of the CBM superfamily, as recently reported (Jing et al., 2009). Helix α1 is positioned within the carbohydrate binding site in all of the CBM homologues and occludes access to a putative ligand (Fig. 3A). In RetSp, this helical domain is tightly anchored to the protein core through hydrophobic contacts involving residues Val65, Leu66 and Ile62 in helix α1 and Val119, Leu124, Tyr117, Tyr160 and Trp90 from the five-stranded β-sheet. Therefore, a lid movement of this domain to promote ligand access to a putative binding site seems unlikely. A second binding site located on one edge of the β-sandwich has been identified in CBM6 from C. Mixtus (Pires et al., 2004) (Fig. 3B) but none of the key residues involved in the binding of the tetrasaccharide are conserved in RetSp.

Figure 3.

Structural comparison between RetSp and related CBMs.
A. Overlay of the closed monomeric form of RetSp (yellow, green and red) and the Pyromyces equi CBM29 (white) (PDB code: 1GWM) as a representative member of the top ranked six CBMs structures in complex with a ligand. For clarity, only the backbone trace of Pyromyces equi CBM29 (white) (PDB code: 1GWM) is shown.
B. The ligand bound to the carbohydrate binding site is displayed for the other CBMs, with only CBM29 backbone displayed in white. Mannohexose (orange) from Pyromyces equi CBM29 (PDB code: 1GWM); xylopentaose (blue) from Pseudomonas cellulosa CBM15 (PDB code: 1GNY); cellopentaose (yellow) from Cellulomonas fimi CBM6 (PDB code: 1GU3); laminariheptaose (pink) from Thermotoga maritima CBM4 (PDB: 1GUI); xylopentaose (cyan) and a β-1,3-β-1,4 tetrasaccharide (green) from Cellvibrio mixtus CBM6, bound to two opposite binding sites. The additional carbohydrate binding site at one edge of the β-sandwich in CBM6 is not observed in RetSp. The position of the β1–β2 loop harbouring helix α1 in RetSp is indicated by a blue star. Occlusion to the carbohydrate binding site in CBM by RetSp helix α1 is evident. An orange star denotes the position of the second putative ligand binding region in RetSp.

Putative ligand binding sites

In light of the high structural similarity between RetSp and CBMs, several conserved aromatic residues lie at the surface of RetSp and might also forge a putative binding site for a yet unknown ligand. In fact, a combined sequence and structure analysis reveals that two distinct regions at the molecular surface of RetSp could provide such a ligand binding site. The first region lies beneath helix α1 and involves residues Phe72, Trp90, Tyr117 and Tyr160 protruding from the five-stranded β-sheet (strands β1, 3, 8, 5, 6) and partially overlays with the carbohydrate binding site in CBMs (Figs 3 and 4). The second region is located on the opposite face of the β-sandwich and involves residues Trp103, Trp105, Phe107 and Trp172 protruding from the four-stranded β-sheet (strands β2, 9, 4, 7) (Fig. 4). These surface exposed aromatic residues also create a hydrophobic patch that represents a hallmark for carbohydrate recognition. In contrast to the first region, the equivalent region of the second binding site in CBMs has not been shown to be a carbohydrate binding site. Sequence analysis from 20 RetS homologues reveals that most conserved residues are located within β-strands lining the two binding sites, while those in loop regions are less conserved (Fig. 5).

Figure 4.

Two putative binding sites in RetSp. Ribbon diagram of the closed monomer (grey) showing:
A. The central five-stranded β-sheet with the conserved patch of aromatic residues (blue) from the first binding region beneath helix α1.
B. The four-stranded β-sheet on the opposite face with the conserved patch of aromatic residues (orange) from the second binding site.
In the two regions, overlapping residues from a LadSp model are shown in white and with italics labels while those in RetSp are underlined.

Figure 5.

Amino acid sequence alignment of annotated hybrid histidine protein kinases with four RetS sequence homologues (group 1) and four LadS homologues (group 2). RetS-PAO1, RetS from Pseudomonas aeruginosa PAO1; Pseudo_men, RetS from Pseudomonas mendocina ; Pseudo_stu, RetS from Pseudomonas stutzeri; Azoto_vin, RetS from Azotobacter vinelandii; Lads-PAO1, LadS from Pseudomonas aeruginosa PAO1; Pseudo_fluo, LadS from Pseudomonas fuorescens; Pseudo_ento, LadS from Pseudomonas entomophila. Strictly conserved and similar residues are boxed with a red background and coloured in red respectively. Conserved aromatic residues involved in the first binding site and conserved among the CBM families are indicated with dark and light blue dots for RetS and LadS respectively. Aromatic residues involved in the second binding site in RetS, not conserved in LadS, are indicated with orange squares. The secondary structure elements of RetS are shown on top of the alignment. The alignment was generated with Multalin (Corpet, 1988) and drawn with ESPript (Gouet et al., 2003).

Modelling of the sensor domain of the LadS sensor kinase

Although RetS and LadS have an antagonistic impact on gene expression, the high (35%) sequence identity between RetSp and LadSp indicates that both domains share a similar fold. Indeed, a model of LadSp generated with the programme Modeller (Eswar et al., 2008) using the structure of RetSp as a template shows a similar overall fold, but unveils large differences in surface residues between the two sensor domains. The aromatic residues involved in the first putative binding region of RetSp are well conserved in LadSp, along with two neighbouring aromatic residues (Phe38 and Tyr117) (Fig. 4), which are highly conserved in the sequences of LadSp homologues (Fig. 5). In contrast, the second binding region in RetSp is not conserved in LadSp, as only a single tryptophan residue, Trp94 (equivalent to RetS Trp103) is present in this putative binding site (Fig. 5).

Such a drastic difference at the molecular surface may suggest that LadS and RetS use different molecular mechanisms for signalling. The lack of the second binding site in LadS may emphasize a different nature of the ligand recognized and a difference in the modulation of the LadS oligomeric forms. Alternatively, the swapped homodimeric assembly could be restricted to RetSp while LadSp may behave only as a globular monomeric fold. Indeed, the low conservation of residues within the hinge β6–β7 loop region in LadSp and RetSp supports a different behaviour in solution as the hinge region contributes to affect the free energy difference between the monomer and the domain-swapped oligomer.

Structural implication of a dimeric assembly for regulating the activity of RetS

The mechanism by which RetS acts on gene expression is uncommon. RetS does not display autokinase activity. Instead, it forms heterodimers with GacS, inhibiting GacS autophosphorylation (Goodman et al., 2009) and indirectly phosphorylation of the GacA response regulator (Fig. 6).

Figure 6.

Model for the RetS signalling mechanism based on the multiple oligomeric forms of RetSp and its interaction with GacS. The membrane bound sensors GacS and RetS are shown with their respective histidine kinase (H, brown) and response regulator receiver (D, green) domains.
A. In absence of external signal, RetS is predominantly found as homodimer, leading to phosphorylated GacS homodimer and GacA (red) phosphorylation. The GacA-dependent regulation increases biofilm formation and represses T3SS activity.
B. In presence of a signal (yellow star) bound to the second ligand binding site within DISMED2, the RetS oligomeric equilibrium is shifted towards GacS/RetS heterodimers, preventing autophosphorylation of GacS and subsequent phosphorylation of GacA. The resulting phenotype, opposite to that in (A), results in an increase of T3SS activity and repression of biofilm formation.

Usually, activity of the sensor kinase is stimulated by ligand binding on the sensor domain. Finding a putative binding site at a RetSp dimer interface may indicate that ligand binding at this site, disfavours or destabilizes RetS homodimers. Consequently, level of RetS monomers increase to promote association with GacS and the resulting RetS/GacS heterodimers prevent biofilm formation and induce T3SS (Fig. 6). In a previous study, it was shown that deletion of DISMED2 from RetS does not prevent RetS function, as seen by normal level of T3SS activity (Laskowski and Kazmierczak, 2006). The lack of the sensor domain could disfavour RetS homodimerization, favour RetS/GacS heterodimerization and thus promote T3SS activity. In vivo studies showed that a retS mutant is less virulent in a mouse infection model likely because of low levels of T3SS (Laskowski and Kazmierczak, 2006). Interestingly, this study showed that complementation with a gene encoding RetS lacking the DISMED2 domain resulted in increased virulence. Again the lack of DISMED2 prevents homodimerization through the RetSp domain, increases formation of RetS/GacS heterodimers and, in turn, T3SS activity and virulence. We propose that having no DISMED2 or a ligand bound at the second RetSp binding site would result in a similar phenotype. Once a ligand is bound at the second binding site, RetS homodimers are destabilized to promote formation of RetS/GacS heterodimers (Fig. 6).

Conclusions

The crystal structure of RetSp reveals two dimeric interfaces and is marked by two putative binding sites. RetS and LadS share the first binding site, highly conserved in CBMs, but its accessibility is blocked by a helical region covering the binding area in RetSp. In contrast, the second binding site in RetSp that lies at the interface between two monomers is not conserved in LadS. Although both dimeric interfaces have been observed in the RetSp crystal, the dimeric and tetrameric forms corresponding to tightly packed four-stranded β-sheets have been also evidenced in solution. The presence of a second putative ligand binding site, which is buried at the interface of the dimeric assembly, favours the hypothesis of RetS monomerization upon binding of an environmental signal. In turn, destabilization of the RetS dimer in presence of a ligand would trigger RetS heterodimerization with monomers of GacS and therefore would inhibit GacA phosphorylation. The non-conservation of this second binding site in LadS suggests that the mode of action of this sensor protein differs from the RetS homologue, which might explain their antagonistic activity in controlling regulation of the T3SS and biofilm formation. With the identification of two putative ligand binding sites and distinct oligomeric species in RetSp, the next challenge is to identify the nature of the environmental stimulus and how it can control the oligomerization process. Site-directed mutagenesis and chimeras of RetS and LadS are under way and may be anticipated to document the antagonist activity of these two hybrid sensor kinases.

Experimental procedures

Cloning and protein production

The DNA fragment encoding the RetSp domain was amplified by PCR from an entry clone containing the PAO1 retS gene (Gateway®, Invitrogen) (Labaer et al., 2004). Oligonucleotides were designed to incorporate the attB1 and attB2 recombination sites for cloning into the T7-promoter pDest17 expression vector (Gateway®, Invitrogen). A construct was designed (pDest17-RetSp) to encode the periplasmic domain of RetS (RetSp) ranging from residue 44 to 185 and harbouring a N-terminal 6×-His-tagged followed by a TEV cleavage site.

Escherichia coli Rosetta(DE3)pLysS cells (Novagen) with pDest17-RetSp were cultured in Superior Broth medium containing 50 µg ml−1 ampicillin and chloramphenicol at 37°C to mid-exponential phase (OD600 = 0.6). IPTG was added to a final concentration of 100 µM, and the cultures were further incubated overnight at 25°C. Cells were harvested by centrifugation and disrupted by sonication. The lysate was clarified by centrifugation and applied to a 5 ml Ni-chelating column using an ÄktaXpress (GE-Healthcare). Proteins bound to the resin were eluted using a linear gradient 10–500 mM of imidazole. RetSp-containing fractions were concentrated and further purified by gel filtration on Superdex75 26/60 (GE-Healthcare) equilibrated with 50 mM Bicine pH 8.5 150 mM NaCl buffer. Pure fractions of RetSp were pooled and concentrated to 8 mg ml−1 using a 9 kDa cut-off ultracentrifugation membrane (Thermoscientific). A selenomethionine-substituted RetSp was expressed according to standard conditions for the methionine-biosynthesis inhibition pathway (Doublié, 1997) and purified as for native RetSp. Native RetSp and selenomethione-substituted RetSp were mixed with the Nvoy polymer (Novexin) in a fivefold molar excess in order to improve RetSp solubility and stability.

Size-exclusion chromatography-multi-angle laser light scattering characterization

Size-exclusion chromatography experiments were carried out on an Alliance 2695 HPLC system (Waters) using a Silica Gel KW803 column (Shodex) eluted with 50 mM Hepes pH 7.5, 150 mM NaCl (flow rate of 0.5 ml min−1). Detection was performed by a combination of UV spectrophotometry, multi-angle static light scattering (MALS) and refractometry. UV, MALS and refractometry measurements were achieved with a photo Diode Array 2996 (Waters), a MiniDawn Treos (Wyatt Technology) and an Optilab rEX (Wyatt Technology) respectively. Molecular weight and hydrodynamic radius determination were determined with the ASTRA V software (Wyatt Technology) using a dn/dc value of 0.185 ml g−1 for the protein and 0.131 ml g−1 for Nvoy. RetSp and RetSp + Nvoy were loaded at 3.5, 5 and 7.8 mg ml−1, respectively, in 50 mM Hepes pH 7.5 150 mM NaCl. A two-component analysis has been applied to accurately quantify the RetSp–Nvoy ratio and recover the molecular weight of RetSp alone (Veesler et al., 2009).

SAXS data collection and analysis

Data were collected at the ID14EH3-BioSAXS station (ESRF, Grenoble, France). Samples were exposed using 30 µl of RetSp solution loaded into a 2 mm quartz capillary mounted in vacuum using an automated system that enables the sample to pass through the beam during exposure to minimize the effect of radiation damage. A range of RetSp concentration ranging from 2.68 to 8.19 mg ml−1 in 10 mM Bicine pH 8.5, 150 mM Nacl and 2% glycerol was measured to assess and account for inter particle effects. 2-D scattering images were collected using a Pilatus 1M detector (Dectris) with a sample-detector distance of 1.83 m. Standard data collection consisting of 10 frames with a 10 s exposure was used for all experiments. Individual time frames were processed automatically and independently using the BsxCUBE programme developed at the ESRF, yielding individual radially averaged curves of normalized intensity versus scattering angle (s = 4πSinθ/λ in nm). Time frames are combined to give the average scattering curve for each measurement excluding data points affected by aggregation induced by radiation damage. Scattering from the buffer alone was measured before and after each sample measurement and the average value was used for background subtraction using the programme PRIMUS (Konarev et al., 2003) from the ATSAS package (EMBL, Hamburg). Preliminary analysis to estimate the radius of gyration, volume and molecular weight were also done with PRIMUS Crysol (Svergun et al., 1995). Version 27 was used to fit the theoretical structures to the experimental data and to calculate the theoretical values for Rg and volume. The plots were produced with a beta version of SASPLOT from the ATSAS package.

Crystallization, data collection and processing

The Nvoy polymer has been used during the concentration step of the RetSp-SeMet protein in order to increase its solubility by masking hydrophobic surface patches and has been conserved during the crystallization process. Crystals of RetSp-SeMet were obtained at 20°C by screening three crystallization kits (Pact1er, Stura, JCSG, Molecular dimension) using a nanolitre sitting drops setup with automated crystallization TECAN Genesis and Cartesian robots. A total of 300 µl of RetSp-SeMet-Nvoy complex (6.6 mg ml−1) was mixed to 100 µl of reservoir solution composed of 0.2 M Ammonium chloride pH 6.3 and 20% polyethylene glycol (PEG) 3350. A single crystal of RetSp-SeMet was mounted on a rayon loop and flash-cooled to 120 K. A SAD data set was collected at 2.65 Å at the selenium K-edge on a single RetSp-SeMet crystal at the ID23EH1 beamline (ESRF, France). Data were integrated, scaled and reduced with MOSFLM and SCALA (Collaborative Computational Project, Number 4, 1994). Unless otherwise cited, all further crystallographic computations were carried out using the CCP4 suite of programmes (Collaborative Computational Project, Number 4, 1994).

Crystals of RetSp devoid of Nvoy were obtained through crystallization screening with 300 µl of protein concentrated at 5 mg ml−1, and 100 µl of reservoir solution containing 0.1 M Hepes pH 7.5 25% PEG 3000. A data set (up to 4 Å resolution) was collected on the Proxima I beamline (SOLEIL, France).

Phasing, model building and refinement

HySS from the PHENIX suite (Adams et al., 2002) identified six selenium sites corresponding to the six methionines present on each molecule in the asymmetric unit. Selenium positions were refined, and SAD phases were calculated using Autosol (Adams et al., 2002), giving an overall figure of merit of 0.39. The resulting phases were used as a starting set for phase improvement by solvent flattening using RESOLVE (Terwilliger, 2000), giving an overall figure of merit of 0.7. The improved phases were of sufficient quality to allow subsequent tracing of 80% of the molecule using AUTOBUILD (Adams et al., 2002). The model was refined using Phenix with manual correction using Coot (Emsley and Cowtan, 2004). Missing electron densities correspond to the β6–β7 loop (residues Ser133–Ala140) in the closed monomer. Despite the low resolution (4 Å) of the structure of RetSp, arising from a RetSp solution devoid of the Nvoy polymer, the structure shows the same closed and open monomers in the asymmetric unit, indicating that the polymer does not affect the equilibrium between the two oligomeric forms. The final RetSp model consists of 2225 non-hydrogen protein atoms with 66 water molecules (Table 1). Stereochemistry of the model was assessed with PROCHECK (Laskowski et al., 1993). Data collection and refinement statistics are reported in Table 1.

Acknowledgements

We thank ESRF (Grenoble) for their excellent beamlines facilities. The help of Gerlind Sulzenbacher for SAXS data collection is gratefully acknowledged. We thank Heather Combe for careful reading of the manuscript. This work was supported by the French association for cystic fibrosis ‘Vaincre la mucoviscidose’, the Fondation pour la Recherche Médicale and the CNRS. A.F. is supported by the Royal Society.

Ancillary