Crystal Structure and Oligomeric State of the RetS Signaling Kinase Sensory Domain

Abstract

The opportunistic pathogen Pseudomonas aeruginosa may cause both acute and chronic-persistent infections in predisposed individuals. Acute infections require the presence of a functional type III secretion system (T3SS), while chronic P.aeruginosa infections are characterized by the formation of drug-resistant biofilms. The T3SS and biofilm formation are reciprocally regulated by the signaling kinases LadS, RetS, and GacS. RetS down-regulates biofilm formation and up-regulates expression of the T3SS through a unique mechanism. RetS forms a heterodimeric complex with GacS and thus prevents GacS autophosphorylation and downstream signaling. The signals that regulate RetS are not known but RetS possesses a distinctive periplasmic sensor domain that is believed to serve as receptor for the regulatory ligand. We have determined the crystal structure of the RetS sensory domain at 2.0 Å resolution. The structure closely resembles those of carbohydrate binding modules of other proteins, suggesting that the elusive ligands are likely carbohydrate moieties. In addition to the conserved beta-sandwich structure the sensory domain features two alpha helices which create a unique surface topology. Protein-protein cross-linking and fluorescence energy transfer experiments also revealed that the sensory domain dimerizes with a dissociation constant of K_d=580±50 nM, a result with interesting implications for our understanding of the underlying signaling mechanism.

Introduction

The Gram-negative bacterium Pseudomonas aeruginosa poses a major challenge to the medical community. Antibiotic-resistant strains of this pathogen may cause chronic-persistent as well as acute infections in transplant patients and other immunocompromised individuals ^1-5. Most prominently, chronic P.aeruginosa-associated lung infections are the leading cause of mortality among people with cystic fibrosis ⁶.

Beyond its ability to colonize human tissues P.aeruginosa can persist in a variety of other milieus including plants and soil. This remarkable versatility may be attributed to a diverse array of virulence mechanisms, which enable the bacterium to adapt to vastly different environmental challenges. Expression and activation of these virulence mechanisms are carefully controlled by complex regulatory networks to ensure an optimal adaptive response. The type III secretion system (T3SS), for instance, is a hallmark of acute infections but is not active in chronic infections ^7,8, which are instead characterized by the formation of antibiotic-resistant biofilms ⁹.

The two-component signaling system (TCS) is the primary means of bacteria to translate complex environmental cues into adaptive gene expression patterns. A canonical TCS is composed of a histidine kinase (HK) and a cognate receiver response regulator (RR). Signaling in response to a stimulus involves HK autophosphorylation and subsequent transfer of the phosphate group to a cognate RR. Frequently, the receiver proteins are transcription factors where the RR domains are coupled with DNA binding domains. Phosphorylation and dephosphorylation control gene expression by modulating the affinity of these transcription factors for their DNA binding sites.

P. aeruginosa harbors a particularly broad array of over 60 TCSs to modulate its gene expression¹⁰, including the expression of genes related to the T3SS and biofilm formation. Remarkably, T3SS and biofilm formation are regulated in a coordinated but reciprocal fashion by the signaling kinases RetS, LadS, and GacS ^11-13. RetS and LadS are hybrid sensor kinases, combining both HK and RR domains in a single polypepetide, while GacS is a canonical signaling kinase requiring the response regulator GacA for downstream signaling. RetS is pivotal for the transcription of genes associated with cytotoxicity and acute infections, including the T3SS. On the other hand, a retS mutant displayed a hyperadhesive phenotype and showed elevated levels of transcription for the psl and pel operons which are associated with the synthesis of biofilm oligosaccharides, suggesting that RetS down-regulates biofilm formation in the wild-type strain ^10,11. RetS accomplishes its task by blocking the synthesis of RsmZ, a small regulatory RNA. RsmZ had previously been shown to bind to and sequester the translational repressor RsmA. Free RsmA blocks biofilm formation and favors the expression of the T3SS ^14,15. The kinases LadS and GacS, on the other hand, directly counteract RetS, as they stimulate the expression of the psl and pel operons by up-regulating the expression of the RsmA-antagonist RsmZ ^16,17. Underlying this reciprocal regulation of the T3SSS and biofilm formation is an entirely novel regulatory mechanism that was uncovered in a recent study by Goodman et al. and involves direct contacts between RetS and GacS ¹⁸. Signaling kinases are usually homodimeric and autophosphorylation occurs almost always in trans. According to the proposed model an unknown environmental signal causes the RetS homodimer to dissociate to form a heterodimeric RetS-GacS complex. The asymmetric geometry of the heterodimer is thought to block RsmZ biosynthesis by preventing GacS trans-autophosphorylation. Remarkably, neither the kinase activity nor the RR domains of RetS are required for its unusual interactions with GacS ¹⁸.

Although the signal that causes RetS-GacS hetero-dimer formation is unknown, the periplasmic RetS sensory domain is believed to serve as the receptor for the elusive ligand. The sensory domain encompasses about 160 amino acids and belongs to the large but poorly characterized 7TM-DISM2 domain family, which has been hypothesized to constitute a new class of carbohydrate binding proteins ¹⁹. However, experimental support for this model is lacking. The here reported crystal structure of the RetS sensory domain not only offers insights into the nature of the molecular signal but it represents the first reported structure for any member of this domain family. Dimerization of the sensory domain and the stability of the dimer were also examined because the currently held model assumes a mechanism where ligand binding shifts the equilibrium from RetS and GacS homo-dimers to a RetS-GacS complex.

Experimental Procedures

Cloning, expression, and purification of the RetS sensory domain

The boundaries of the RetS sensory domain, containing amino acids 27 to 185, are defined by two transmembrane helices. The gene fragment coding for the entire periplasmic domain (retS27-185) was PCR-amplified from P. aeruginosa genomic DNA (ATCC 17933D). During PCR a tobacco etch virus (TEV) protease recognition site and the appropriate recombination sites (attB1 and attB2) were added to retS27-185. Subsequently, the amplicon was recombined into pDONR201 (Invitrogen) to produce the plasmid pDONR201-retS27-185. After verifying the nucleotide sequence via DNA sequencing, retS27-185 was recombined into the destination vector pDEST-HisMBP ²⁰ to create the expression vector pDEST-HMBP-retS27-185. This vector is designed to produce RetS27-185 fused to the carboxy-terminal end of amino-terminally hexahistidine-tagged E. coli maltose binding protein (MBP).

Vector pDONR201-retS41-185, containing a shortened segment of the sensory domain, was generated via PCR with the appropriate primer and using pDONR201-retS27-185 as a template. Following sequence verification retS41-185 was recombined into pDEST-HisMBP to create the expression vector pDEST-HMBP-retS41-185. The RetS41-185-S45C variant used for the dimerization studies was generated via site-directed mutagenesis with pDONR201-retS41-185 serving as template. Subsequent recombination yielded the pDESTHMBP-retS41-185-S45C vector.

The protein expression and purification protocols for RetS27-185, RetS41-185 (RetS_peri), and RetS41-185-S45C (RetS_peri-S45C) were identical. Single colonies of E. coli BL21(DE3) CodonPlus RIL cells (Stratagene, La Jolla, CA) containing the expression plasmid were used to inoculate 100 ml of Luria broth supplemented with glucose at 2 g/L, 100 μg/ml ampicillin, and 30 μg/ml chloramphenicol. The cell culture was grown with shaking (225 rpm) to saturation overnight at 37 °C and then diluted 66-fold into six liters of fresh medium. When the cell density reached mid-log phase (OD₆₀₀=0.5), the temperature was reduced to 30 °C and isopropyl-β-D-thiogalacto-pyranoside (IPTG) was added to a final concentration of 1 mM. After four hours cells were harvested by centrifugation at 5,000 × g for 15 minutes.

All of the following steps were carried out at 4 °C. Cells were re-suspended using a buffer containing 50 mM Tris-HCl, 150 mM NaCl, and 25 mM imidazole, pH 7.4 (buffer A) and 5 μl EDTA-free protease inhibitor cocktail (Sigma P8849) per milliliter of buffer (10 mL of buffer per gram of cell mass). Cells were lyzed through sonication and insoluble debris removed by centrifuging the cell extract at 40,000 × g for 30 minutes. The supernatant was filtered through a 0.45 μm polyethersulfone membrane and loaded onto to a 30 mL Ni-NTA chromatography column (Qiagen, Valencia, CA) pre-equilibrated in buffer A. The column was washed to baseline with buffer A and eluted with a linear imidazole gradient to 250 mM over 10 column volumes. Peak elution fractions were combined and His-TEV(S219V)-Arg (1 mg/100 mg of total protein) was added to effect the cleavage of His-MBP. The TEV protease-digest reaction mixture was dialyzed overnight against buffer A. After dialysis, the protein solution was filtered and applied to a 40-mL Ni-NTA Superflow column (Qiagen) pre-equilibrated with buffer A. Flow-through fractions containing the sensory domain were pooled and dialyzed overnight into buffer of 50 mM MES 50 mM NaCl, pH 6.0 (buffer B). The sample was then applied to a 5 mL Heparin column (GE Healthcare,) pre-equilibrated with buffer B and was eluted with a linear salt gradient to 1 M NaCl. Peak fractions were concentrated and loaded onto a HiPrep 26/60 Sephacryl S-200 HR column (GE Healthcare), pre-equilibrated in a buffer of 25 mM Tris-HCl, 150 mM NaCl, pH 7.4 (buffer C). The protein was judged to be >95% pure by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE).

Preparation of selenomethionine-labeled RetS_peri

Selenomethionine-substituted RetS_peri (SeMet-RetS_peri) was produced in the same non-auxotrophic strain of E.coli as used for routine protein production. The incorporation of externally-added selenomethionine into the recombinant protein was accomplished by suppressing methionine biosynthesis ²¹. Initially, a 1L overnight culture of the expression strain was grown in LB medium also containing 100 μg/ml ampicillin and 30 μg/ml chloramphenicol. The cells were washed twice in 100 mL of M9 selenomethionine growth media (Medicilon Inc, Chicago, Illinois). The cells were re-suspended in 100 ml of M9 selenomethionine growth media and used to inoculate four 1L cultures containing M9 selenomethionine growth media and antibiotics. These cultures were grown with agitation at 37°C until the OD_600nm reading reached 0.2. At this point IPTG was added to a final concentration of 1 mM and the temperature lowered to 30°C. Cells were harvested the following morning. The purification of selenomethione-containing protein followed the same protocols as that for the native protein with the exception that 2mM DTT was added to all but final buffer, which contained 1mM tris (2-Carboxyethyl) phosphine (TCEP) instead.

Limited proteolysis of the RetS sensory domain

A 1 mg/ml stock solution of thermolysin (Roche Molecular Biochemicals) in thermolysin buffer (10 mM Tris-HCl, 0.2 M NaCl, and 2 mM CaCl₂, pH 8.0) was used for the limited proteolysis experiments. The RetS27-185 stock solution consisted of the protein at 1 mg/mL in buffer C. The five individual reactions were composed of 25 μL of RetS27-185 stock solution, 25 μL of 2X thermolysin buffer, and 0.5 μL of serial 1:4 dilutions of the thermolysin stock solution. The reactions were allowed to proceed for 1 hour at 37 °C before the protease was inactivated by the addition of 0.5 μL of 0.5 M EDTA. Reaction products were initially visualized by SDS-PAGE. The precise molecular weights of select fragments were determined using LC-electrospray ionization mass-spectrometry. The peptide fragments that corresponded to the observed molecular masses were determined using the FindPep program ²².

Crystallization of RetS_peri and SeMet-RetS_peri

High-throughput crystallization screening was conducted in the sitting-drop format by combining a solution of 17 mg/mL RetS_peri in buffer C with commercially available crystallization matrices at volume ratios of 3:1, 1:1, and 1:3, where the protein solution was maintained at a constant volume of 0.3 μL throughout. Preliminary crystals were obtained from condition D4 of the IndexHT screen (Hampton Research). Hit optimization was carried out using the hanging drop vapor diffusion method at 18 °C. In the optimized conditions crystals for both the native protein and SeMet-RetS_peri were obtained from drops containing a 1:1 mixture of 17 mg/mL proteins in their respective storage buffers and a crystallization solution composed of 0.1M citric acid, 22.5% w/v polyethylene glycol 3350 and 10% v/v glycerol.

Data collection, structure solution, and refinement

Crystals of RetS_peri were loop-mounted and flash-frozen in liquid nitrogen. Data sets were collected at beamline X-29A of the National Synchrotron Light Source using an ADSC Q315 CCD detector. Data were processed at the beamline using the HKL2000 program suite ²³. Details of data collection and processing are provided in Table I. The SeMet-RetS_peri structure was determined using the single-wavelength anomalous dispersion (SAD) method. The location of heavy atoms, initial phase calculations, phase improvement through density modification, initial maps, and automated model building steps were all completed in the PHENIX program suite ²⁴. PHENIX built nearly complete models for both molecules in the asymmetric unit. Iterative cycles of manual model adjustment using COOT ²⁵ followed by refinement in PHENIX rapidly converged to produce the final structures. Model quality was assessed with PROCHECK ²⁶ and the atomic coordinates and structure factors have been deposited in the Protein Data Bank (PDB) ²⁷ with accession code 3JYB.

Table I.

Data collection and refinement statistics for the SeMet-RetS_peri crystal

Data collection statistics
Wavelength (Å)	0.97900
Space Group	P212121
Unit cell parameters (Å)	a=51.164; b=67.193; c = 86.214
Molecules/asymmetric unit	2
Resolution (Å) (last shell 2.11 - 2.04)	30.00 - 2.04
Total reflections	255,128
Unique reflections	19,459
Completeness (%)	99.9 (99.2)a
I/σ	26.3 (3.40)
Rmerge (%)b	9.0 (39.5)
Refinement
Resolution (highest resolution shell) (Å)	29.62- 2.04 (2.1-2.04)
No. reflections	17870 (1265)
Rworkc/Rfreed (%)	19.9 ( 20.1)/ 23.5 (27.5)
Number of total atoms	2403
Number of protein atoms	2298
Number of water atoms	105
Overall mean B factor value ( Å2)	18.747
r.m.s deviations-bond lengths (Å)	0.007
r.m.s deviations-bond angles (°)	1.119

^aThe values in parentheses relate to the highest resolution shell.

^bR_merge = Σ |I-<I>|/Σ I where I is the observed intensity and <I> is the average intensity obtained from multiple observations of symmetry-related reflections after the rejection of significant outliers.

^cR = Σ | |F_o| - |F_c| | / Σ |F_o|

^dR_free defined by Brunger. ³⁷

Protein-protein cross-linking

RetS_peri was dialyzed into a buffer of 20 mM NaH₂PO₄ and 30 mM NaCl, pH7.4. A 2 mM stock solution of Bis(Sulfosuccinimidyl) suberate (BS3) was prepared in water. The 50 μl reaction mixtures contained 10 μM RetS_peri and 500 μM BS3. Cross-linking reactions were quenched by adding 20 μl of each reaction to 10 μl 150 mM Tris-HCl, pH7.4. Cross-linking reactions containing MBP instead of RetS_peri were used as negative controls. Cross-linked samples were analyzed via SDS-PAGE.

Quantitative oligomerization assay

The RetS_peri-S45C variant was fluorescently labeled with Alexa Fluor 488 and Alexa Fluor 555 (Invitrogen) according to the manufacturer's instructions. Following overnight labeling the modified proteins were separated from the unincorporated dye molecules through buffer exchange into a buffer of 20 mM Tris-HCl and 150 mM NaCl, pH 7.4. The degree of labeling of both the Alexa Fluor 555-labeled form of the RetS_peri variant (RetS_peri-555) and the Alexa Fluor 488-labeled form of the RetS_peri variant (RetS_peri-488) were near 100% as assessed by UV-spectroscopy in conjunction with the estimated molar extinction coefficients of RetS_peri and the respective fluorophores.

The fluorescence resonance energy transfer (FRET) measurements involved titrating a solution of 3 nM RetS_peri-488 with RetS_peri-555. Triplicate set-ups of 80 μl reactions containing 3 nM RetS_peri-488 and 0 to 100 μM RetS_peri-555 were transferred to a 96-well half area black polystyrene assay plate (Corning). Background fluorescence produced by RetS_peri-555 was accounted for by measuring and subtracting the signal of solutions that did not contain RetS_peri-488 but identical concentrations of RetS_peri-555.

A TECAN infinite M200 fluorescence intensity scanner (Tecan, inc.) was used for the experiment. The excitation wavelength was set to 430 nm, while the emission spectrum was recorded in between in the wavelength range between 510 and 646 nm using a 4 nm step-size. The integration time was set to 20 μs and all scans were carried out at a room temperature of 24 °C. Dimer formation was monitored by recording the decrease in the peak fluorescence of RetS_peri-488 at λ=522 nm. Data were fit to equation (I) using Matlab (The MathWorks).

$${\mathrm{R}}_{5\mathrm{T}}=\frac{{\mathrm{K}}_{\mathrm{d}}({\mathrm{F}}_{\text{max}}-{\mathrm{F}}_{\text{min}})}{\mathrm{F}-{\mathrm{F}}_{\text{min}}}[\frac{2({\mathrm{F}}_{\text{max}}-{\mathrm{F}}_{\text{min}})}{\mathrm{F}-{\mathrm{F}}_{\text{min}}}-3]+{\mathrm{R}}_{4\mathrm{T}}\frac{{\mathrm{F}}_{\text{max}}-\mathrm{F}}{{\mathrm{F}}_{\text{max}}-{\mathrm{F}}_{\text{min}}}+{\mathrm{K}}_{\mathrm{d}}$$ (I)

R_5T and R_4T are the total concentrations of RetS_peri-555 and RetS_peri-488, respectively, F is the fluorescence measured at λ=522 nm, F_max is the fluorescence measured at λ=522 nm in absence of RetS_peri-555, F_min the residual fluorescence of RetS_peri-488 due to incomplete quenching even when all RetS_peri-488 is bound to RetS_peri-555, and K_d is the dissociation constant of the RetS_peri dimer. Derivation of equation (I) is outlined in supplementary file S1.

Results

The RetS sensory domain assumes a beta-sandwich fold reminiscent of carbohydrate binding proteins

In order to ensure that the complete sensory domain was contained in the cloned construct, the entire periplasmic region of RetS encompassing residues 27 to 185 was initially cloned and over-expressed. Subsequently, the domain boundaries were more precisely mapped through limited proteolysis of RetS27-185 in order to facilitate crystallization. Mass-spectrometric analysis of the obtained fragments identified a relatively stable fragment comprising RetS residues 41 to 185 (RetS_peri). RetS_peri, over-expressed from a newly engineered plasmid, readily crystallized. SeMet-RetS_peri crystals, while very small (~ 0.05*0.02*0.01mm³), produced x-ray diffraction to a resolution of 2.0 Å.

Forming an asymmetric dimer there are two independent RetS_peri molecules in asymmetric unit of the crystal. The structural differences between the backbone atoms of the two molecules are small as reflected in a root-mean-square-deviation of 0.4 Å. The electron density throughout the structure was excellent. However, the amino-terminal regions ranging from Ala-41 to Asn-47 are disordered and produced no discernable electron density in either molecule. All non-glycine residues of the final model reside either in the most favorable or in the allowed regions of the Ramachandran plot; and the overall geometry was comparable to other structures solved at the same resolution.

A cartoon drawing of the final RetS_peri structure model is depicted in figure Fig. 1a, while the correlation between protein sequence and tertiary structure is visualized in figure Fig. 1b. The sensory domain of RetS adopts β-sandwich or jelly-roll fold formed by two opposing antiparallel β-sheets. The two sheets have a β1-β3-β8-β5-β6 and a β2-β9-β4-β7 topology, respectively. The β-sandwich structure is augmented by two alpha-helices, α-1 follows strand β-1 and the short α-2 helix is formed by the carboxy-terminal residues of RetS_peri. The observed fold is characteristic of carbohydrate binding modules (CBM). Consistent with this notion, a DALI-guided ²⁸ search of the PDB identified CBMs and enzymes involved in carbohydrate degradation as the closest structural homologs of RetS_peri. While the structural conservation extends to both beta sheets, helix α-1 appears to be unique to RetS_peri. The position of this helix directly above the β1-β3-β8-β5-β6 sheet is interesting because in most of the related CBMs the equivalent β-sheets form the carbohydrate binding sites. We examined the temperature factors as well as intermolecular and intramolecular contacts of helix α-1 to evaluate whether or not this helix might be conformationally flexible and its conspicuous position perhaps the result of packing contacts. However, we found no intermolecular contacts involving α-1 within a 5 Å radius. On the other hand, α-1 forms a total of 18 intramolecular contacts with neighboring residues if a 4 Å distance cutoff is applied, burying a surface area of 525 Ų. Furthermore, the positions of the two α-1 helices are identical for both molecules in the asymmetric unit and the average temperature factor of ~20 Ų for the main-chain atoms also suggest a stable conformation. Therefore the unique topology of this region suggests that the 7TM-DISM2 domains constitute a novel class of carbohydrate binding proteins.

Fig. 1.

a. Two orthogonal views of the PYMOL ⁴¹-generated cartoon representation of the RetS_peri structure. All secondary structure elements are labeled. b. Alignment of the secondary structure elements identified from the crystal structure with the primary structure RetS_peri. This figure was generated with ESPript ⁴². c. The asymmetric unit of the crystal consists of molecules A and B. The contacts between A and B closely resemble the packing contacts between A and the symmetry-related B’ molecule, suggesting that the dimer in the asymmetric unit arose from crystal packing contacts.

Qualitative and quantitative evidence for RetS_peri dimerization

The reversible oligomerization of RetS is believed to play a pivotal role in the regulation of RetS function. Consequently, we sought to establish whether or not RetS_peri dimerization contributes to the stability of the RetS homodimer. Intuitively, we would have expected to observe a symmetric dimer in the crystal, but the two molecules that form the asymmetric unit display no two-fold symmetry. While unusual this asymmetric dimer could nevertheless represent a biologically relevant dimer. However, when we compared the intermolecular contacts within the asymmetric unit to packing contacts between symmetry-related asymmetric units in the crystal, we found them to be very similar. These packing arrangements are visualized in figure Fig. 1c, where molecules A and B represent the original asymmetric unit and B’ a molecule from a symmetry-related asymmetric unit. The buried surface area of the A•B and A•B’ interfaces are very similar at 610 and 629 Ų, respectively. There are twenty-one interacting residue pairs at A•B interface and twenty-three such pairs at the A•B’ interface when a 4 Å distance cutoff is applied (this cutoff was chosen because it represents the default value in programs such as Ligplot²⁹ that are designed to analyze intermolecular contacts). Twenty of these interactions are found at both interfaces. Distances for the four contacts that did not match ranged between 3.7 and 4 Å, suggesting that they make only small contributions to the stability of the respective intermolecular interactions. These similarities suggest that the dimer observed in the asymmetric unit arose due to crystal packing contact and does not represent the biological unit of the protein. Otherwise, if the interactions in the asymmetric unit were indeed representative of the solution state then RetS_peri should be forming higher order oligomers or even polymerize in solution. This was not observed in the cross-linking experiments described below.

Since the non-physiological pH of 3.0 of the crystallization conditions could have caused a biological RetS_peri oligomer to dissociate, protein-protein cross-linking using chemical BS3 was employed to probe for dimer formation at neutral pH. Monomeric E.coli MBP served as negative control. After cross-linking MBP still migrated as a monomer on an SDS-polyacrylamide gel, while a substantial portion of RetS_peri had shifted to a higher band corresponding to the molecular weight of a RetS_peri dimer (Figure Fig. 2a). Higher order oligomers were not observed.

Fig. 2.

a. SDS-PAGE analysis of the BS3-mediated cross-linking experiment with RetS_peri. Shown in the lanes are from left to right: MBP, the molecular weight makers, and RetS_peri. b. Sample plots of fluorescence spectra obtained during the FRET-based dimerization assay at three different concentrations of RetS_peri-555. The spectra are generated by subtracting the background fluorescence produced by the same concentrations of RetS_peri-555 in absence of RetS_peri-488 from the raw data. c. Shown is the resulting curve when equation (I) is fit to the FRET data where $\Delta \mathrm{F}=\frac{{\mathrm{F}}_{\text{max}}-\mathrm{F}}{{\mathrm{F}}_{\text{max}}-{\mathrm{F}}_{\text{min}}}$. The error bars signify the standard deviation of the ΔF calculated from the readings obtained from triplicate set-ups. For comparative purposes the data were also fit to the conventional binding isotherm (supplementary file S1. part c.), which does not consider the increasing amounts RetS_peri-555 dimer. This fit is shown as a dashed line.

Although the cross-linking result demonstrates RetS_peri dimerization, a substantial fraction of RetS_peri failed to cross-link even when large excess of cross-linking agent and extended reaction times were used. This finding is consistent with the presence of monomer-dimer equilibrium rather than a purely dimeric state. A FRET-based assay was developed to validate this hypothesis and quantify the strength of the RetS_peri-RetS_peri interactions. A low concentration of Alexa-Fluor-488-labeled RetS_peri-S45C (RetS_peri-488) was titrated with an Alexa-Fluor-555-labeled form of the same RetS_peri variant (RetS_peri-555). Dimerization was monitored by observing the quenching of the Alexa Fluor 488 fluorescence at λ=522 nm. The concentration of the RetS_peri-488 was maintained more than a hundred-fold below the dissociation constant (K_d) for RetS_peri dimer formation because here virtually all of the protein should be monomeric. This optimal concentration of RetS_peri-488 was determined in an iterative process where titration experiments were carried out at concentrations of 50, 10, and finally 3 nM of RetS_peri-488 and fit to equation (I). The two higher concentrations of the protein resulted in a poor fit and a K_d value that suggested that a substantial proportion of RetS_peri-488 was dimeric at the outset of the experiment. A second important consideration relates to the fact that, as the concentration of RetS_peri-555 is increased during the titration experiment, a large fraction of this protein is already dimeric and therefore not available for binding to RetS_peri-488. In order to properly model RetS_peri dimerization it was therefore necessary to express the concentration of the RetS_peri-555 monomer as a function of the total concentration of RetS_peri-555 and the K_d. Usually, equations for binding isotherms are arranged to express the amount of complex formed in terms of the variable concentration of one of the binding partners. However, when this was done for the modified isotherm the resulting cubic equation was solvable but the obtained solution was awkwardly long (supplemental file S.1 part b.). Therefore, the data were fit to the rearranged and thus simpler equation (I). This equation is valid under the assumption that the fluorescent labels do not influence the monomer-dimer equilibrium, which is based on the observation that the S45C mutation used to mediate thiol-based labeling is located in a structurally disordered region that does not appear to constitute a pivotal part of the sensory domain. Background-corrected example graphs are presented in figure Fig. 2b. The resulting FRET data are plotted in figure Fig. 2c; and fitting equation (I) to these data yielded a dissociation constant of K_d=580± 50 nM for RetS_peri dimerization. If RetS_peri-555 dimerization is not taken into consideration and the data are simply fit to a conventional isotherm, the fit is poor and the obtained K_d incorrect (dashed line in figure Fig. 2c).

Discussion

The putative ligand binding site of RetS_peri

The crystal structure suggests that RetS is regulated by a carbohydrate-based moiety. However, since the natural ligand of RetS is currently not known, the common strategy for mapping the ligand binding site involving a mutational analysis coupled with in vitro binding studies could not be employed. Nevertheless, insights could be obtained by comparing RetS_peri to the altogether eight structurally homologous proteins, where the positions of the ligand binding sites are known. Protein-carbohydrate complexes have been reported for the related family 4 CBM of Lam16A from Thermotoga maritima (TmCBM4-2, PDB code 1gui) ³⁰, the family 6 CBM SdAga16B-CBM6-2 of the β-Agarase Aga16B from Saccharophagus degradans (PDB code 2CDO) ³¹, and the family 15 CBM of the Cellvibrio japonicas Xylanase 10C, (CjXyn10C-m, PDB code 1US2) ³². In addition, the binding sites of the following proteins have also been mapped and were therefore included in the analysis: the carbohydrate-recognition domain of the human glycoprotein sorting receptor p58/ERGIC-53 (PDB code 1R1Z) ³³, the carbohydrate recognition domain of the cargo receptor Emp46p from Saccharomyces cerevisiae (PDB code 2A6W) ³⁴, the carbohydrate binding module of xylanase 10A from the thermophilic bacterium Rhodothermus marinus (RmXyn10A-CBM4-2, PDB code 1K42) ³⁵, the catalytic domain of the endoglucanase CelB from Streptomyces lividans, (SlCelB-2, PDB code 1NLR) ³⁶, and family 32 carbohydrate-binding protein YeCBM32 from Yersinia enterolitica (PDB code 2JDA) ³⁷. Although all eight proteins share the same fold, their ligand binding sites cluster to not one but two regions when superimposed. In YeCBM32 and SdAga16B-CBM6-2 the ligand binding sites are located between the two beta sheets, involving residues equivalent to the carboxy-terminal end of β-2, the amino-terminal end of β-9, and residues from the two connected loop regions in RetS_peri. This pocket is not present in RetS_peri suggesting that its binding cleft is located elsewhere.

In the other six homologous structures the critical residues map to structurally conserved section of the proteins that corresponds to the β1-β3-β8-β5-β6 sheet in RetS_peri. In the TmCBM4-2 and CjXyn10C-m complexes, the entire beta-sheet forms an extended cleft, where their respective oligosaccharide ligands are bound, while the ligand binding site in the other four proteins are smaller, covering only part of the beta-sheet. In RetS_peri helix α-1 alters the topology of this the corresponding surface area. In fact, α-1 effectively occupies the same space as the bound ligands in TmCBM4-2 and CjXyn10C-m (Figure Fig. 3a.). However, together with residues from the beta sheet and several surrounding loops α-1 does create three large grooves that appear well-suited for ligand binding. The pocket that most closely coincides with the prevalent binding pockets observed in the related proteins is formed by the loop connecting β-1 and α-1, the amino terminal regions of strands β3 and β5, the carboxy-terminal ends of strands β1, β8, and β6 and the loop regions between β-2 and β-3 as well as those between β-8 and β-9 (Figures Fig. 3b and 3c).

Fig. 3.

a. Superposition of RetS_peri, TmCBM4-2, and CjXyn10C-m. TmCBM4-2 and CjXyn10C-m have both been crystallized in complex with xylopentaose. The positions of the ligands in the latter complexes coincide with helix α-1 in RetS_peri. b. The backbone structure of RetS_peri has been colored according to the level of sequence conservation among proteins with related 7TM-DISM2 domains. Also, shown are side chains and labels of the seven broadly conserved amino acids that are concentrated in this section of the protein and of those amino acids that form the putative binding site. c. The solvent-accessible surface area of RetS_peri, viewed from the same perspective and also colored–coded according to the level of sequence conservation, reveals a large cavity that is proposed to constitute the ligand binding site.

The hypothesis that this site indeed constitutes the ligand binding pocket may be evaluated from two additional angles. The solvent-accessible surface area of RetS_peri was color-coded according to the degree of sequence conservation using the ConSurf ³⁸. Those regions with the three highest conservation scores were colored brightly, while the remainder of the surface was kept white (Figures Fig. 3b and 3c.). Remarkably, seven of the ten amino acids that received the highest conservation scores map to the β1-β3-β8-β5-β6 beta sheet. Four of these seven residues, Trp-90, Tyr-117, Arg-162 and Ser-164 are part of the pocket that was identified from the structural comparison with other binding sites. Trp-90 is one of only two amino acids that are strictly conserved in all thirty-two of the homologous protein sequences that were used for this analysis.

Lastly, the predicted binding site may also be assessed by analyzing the types of amino acids that form this pocket. Tryptophan, arginine, asparagine, glutamic acid, tyrosine, and histidine display the highest propensities for participating in carbohydrate binding ³⁹. The highly conserved Trp-90, Tyr-117, and Arg-162 fall into this category. Other residues in the binding pocket also feature prominently in carbohydrate binding sites: Gln-59, Arg-61, Gln-112, Arg-134, and Asn-166, and the aromatic residues Tyr-113 and His-129. Pro-86 and Thr-164 constitute the only unusual residues among the amino acids that are form the putative ligand binding site (Figures Fig. 3b. and 3c.). In summary, three independent indicators -structural conservation, sequence conservation, and composition of the putative binding pocket- all point toward the same location of the ligand binding site and suggest that the elusive ligand is a carbohydrate moiety. Ultimate confirmation of these findings, however, will require the identification of the natural ligand of RetS.

Implications of RetS_peri dimerization for the RetS-GacS signaling mechanism

RetS is an unusual sensor kinase because at least one of its signaling modes does not require its kinase activity but relies on direct interactions with GacS kinase to prevent phosphate transfer in the GacS/GacA TCS ¹⁸. The kinase domains of both enzymes appear to be sufficient for this interaction. The model assumes ligand binding to the periplasmic sensory domain disrupts the RetS dimer, but how is this signal communicated across the inner cell membrane? The hypothesis that the RetS sensory domain mediates the up-regulation of T3SS-related gene expression also appears to be at odds with a previous study, where deletion of the sensory domain did not cause a decrease but a slight increase in virulence associated with the T3SS ¹³. Our finding that RetS_peri alone is capable of dimer formation offers a possible explanation for this apparent contradiction as it suggests that the sensory domain is required for repression as well as activation of the T3SS. Ligand-free RetS_peri likely dimerizes and prevents RetS-GacS interactions by stabilizing the RetS homodimer. The unknown ligand could act by disrupting the interface between the two sensory domains, and thus favor RetS-GacS hetero-dimer formation. Consistent with the previously published results this model would explain why a deletion of the sensory domain did not result in a repression of the T3SS, since such a deletion, just as ligand binding, would favor a RetS-GacS complex.

Conclusion

The crystal structure of the RetS sensory domain, the first representative structure for the 7TM-DISM2 domain family, revealed a fold closely related to carbohydrate binding modules, suggesting that RetS function is regulated by a carbohydrate-based ligand. While the crystal structure does not unambiguously reveal the ligand binding site, structural homology, sequence conservation, and amino acid composition suggest that the ligand likely binds to a pocket formed by one of the two beta sheets and helix α-1. RetS_peri dimerizes with a submicromolar dissociation constant, suggesting that dimerization of the RetS sensory domain plays a role in both stabilizing the RetS homo-dimer and, upon ligand binding, in shifting the equilibrium towards RetS-GacS hetero-dimer formation.

Abbreviations used

BS3

Bis(Sulfosuccinimidyl) suberate

CBM

carbohydrate binding module

DTT

dithiothreitol

EDTA

ethylenediaminetetraacetic acid

FRET

Fluorescence resonance energy transfer

IPTG

isopropyl-β-D-thiogalacto-pyranoside

MBP

Maltose Binding Protein

PDB

Protein Data Bank

rms

root mean square

SAD

single-wavelength anomalous dispersion

SDS-PAGE

sodium dodecyl sulfate polyacrylamide gel electrophoresis

TCS

two-component signaling system

TCEP

tris(2-Carboxyethyl) phosphine

TEV

tobacco etch virus

T3SS

type III secretion system