Virulence Regulation with Venus Flytrap Domains: Structure and Function of the Periplasmic Moiety of the Sensor-Kinase BvgS

Two-component systems (TCS) represent major signal-transduction pathways for adaptation to environmental conditions, and regulate many aspects of bacterial physiology. In the whooping cough agent Bordetella pertussis, the TCS BvgAS controls the virulence regulon, and is therefore critical for pathogenicity. BvgS is a prototypical TCS sensor-kinase with tandem periplasmic Venus flytrap (VFT) domains. VFT are bi-lobed domains that typically close around specific ligands using clamshell motions. We report the X-ray structure of the periplasmic moiety of BvgS, an intricate homodimer with a novel architecture. By combining site-directed mutagenesis, functional analyses and molecular modeling, we show that the conformation of the periplasmic moiety determines the state of BvgS activity. The intertwined structure of the periplasmic portion and the different conformation and dynamics of its mobile, membrane-distal VFT1 domains, and closed, membrane-proximal VFT2 domains, exert a conformational strain onto the transmembrane helices, which sets the cytoplasmic moiety in a kinase-on state by default corresponding to the virulent phase of the bacterium. Signaling the presence of negative signals perceived by the periplasmic domains implies a shift of BvgS to a distinct state of conformation and activity, corresponding to the avirulent phase. The response to negative modulation depends on the integrity of the periplasmic dimer, indicating that the shift to the kinase-off state implies a concerted conformational transition. This work lays the bases to understand virulence regulation in Bordetella. As homologous sensor-kinases control virulence features of diverse bacterial pathogens, the BvgS structure and mechanism may pave the way for new modes of targeted therapeutic interventions.


Introduction
Two-component sensory transduction systems (TCSs) regulate various physiological processes in response to environmental changes [1]. They are abundant throughout the phylogenetic tree except for vertebrates and represent major bacterial signaling pathways [2,3]. TCSs notably regulate the cell cycle, motility, biofilm formation or antibiotic resistance, as well as the virulence of major pathogens [4][5][6][7][8]. TCSs are typically composed of a sensor-kinase activated by environmental stimuli and a response regulator mediating phosphorylation-dependent effects [9,10]. Upon perception of a physical or chemical signal, auto-phosphorylation of a conserved cytoplasmic His residue of the sensor-kinase is followed by transfer of the phosphoryl group to a conserved Asp residue of the response regulator. The phosphorylated response regulator mediates a specific, frequently transcriptional, cellular response [11]. There is considerable diversity among TCSs regarding domain composition and organization [9,10].
Bordetella pertussis, the whooping cough agent, colonizes the upper respiratory tract of humans [12]. Transcription of its virulence regulon is positively regulated by the TCS BvgAS [13]. Over one hundred genes belong to the Bvg regulon, including those coding for the adhesins and toxins and their secretion and assembly machineries [14]. The virulent, Bvg + phase, in which phosphorylated BvgA trans-activates the expression of the virulence regulon, is essential for the development of the infection cycle of B. pertussis and other pathogenic Bordetella species [13,15]. The kinase and phosphotransfer activities of BvgS are maximal (referred to below as the 'kinase-on' state) without specific chemical stimuli and at 37°C, the B. pertussis host body temperature, while low temperatures and specific negative modulators turn these activities off in laboratory conditions (referred to below as the 'kinase-off' state). Thus, millimolar concentrations of nicotinate or sulfate ions result in the dephosphorylation of BvgA, switching the bacteria to the avirulent, Bvgphase [16,17]. Virulence genes are no longer expressed, while a smaller set of virulence-repressed genes (vrgs) are upregulated [18,19]. At low modulator concentrations, an intermediate Bvg i phase occurs in which the reduced concentration of phosphorylated BvgA is sufficient to transactivate 'early' virulence genes as well as specific intermediate genes [13,20,21]. Thus, BvgAS operates like a rheostat, determining several states of gene expression that might correspond to distinct temporal or spatial situations in the course of infection. BvgS is composed of periplasmic Venus flytrap (VFT) domains, a transmembrane segment, a PAS domain, and a kinase and additional domains that make up a phosphorelay (Fig. 1A). The cytoplasmic moiety of BvgS dimerizes, similar to the other TCS sensorkinases [22,23].
BvgS is the prototype of a family of bacterial VFT-domain-containing sensor-kinases [24]. VFT domains have a bi-lobed structure with two mobile jaws delimitating a putative ligandbinding cavity [25,26]. They exist in open and closed conformations that interconvert by clamshell motions. Typically, binding of a ligand in the cavity stabilizes the closed conformation, which triggers downstream cellular events such as transport or signaling. The periplasmic moiety of BvgS is composed of two VFT domains, membrane-distal VFT1 and membraneproximal VFT2. We have previously reported the structure of the isolated VFT2 domain and showed that nicotinate and related negative modulators bind to VFT2 [27]. There are currently more than 2000 predicted BvgS homologs, containing from one to five VFT domains. Some of them are found in major pathogens, including Pseudomonas aeruginosa, Vibrio cholerae, Yersinia enterocolitica and Borrelia burgdorferi, in which they regulate various responses that contribute to pathogenicity [28][29][30][31][32] (Fig. 1B). Unlike those of classical TCSs, the molecular mechanisms of signal perception and transduction by these VFT-containing sensor-kinases are largely unknown.
In this work, we describe the structure of the periplasmic portion of BvgS, revealing a novel homo-dimeric architecture with two highly intricate polypeptide chains wound around each other. A combination of site-directed mutagenesis, functional analyses in vivo and molecular modeling indicated that the integrity of the periplasmic domain is necessary both to maintain BvgS in a kinase-on state by default and to bring about conformational changes that switch the protein to the kinase-off state in response to negative modulation. This study shows that BvgS represents a new paradigm of bacterial two-component sensor-kinases and contributes to our understanding of virulence regulation in Bordetella.

Structure of the periplasmic domain of BvgS
The periplasmic domain of BvgS (residues Ala 29 -Leu 544 , which includes VFT1 and VFT2) was produced in Escherichia coli and crystallized as a recombinant protein with a 60-residue-long GB1 domain at the N terminus and a 6-His tag at the C terminus. The structure was solved to a resolution of 3.1 Å (Fig. 2, S1 Table). BvgS forms intricate butterfly-shaped dimers in which the A and B polypeptide chains ('protomers') wind around each other, with an extensive dimeric interface of 4000 Å 2 . The two protomers overlap with an RMSD of 1.184 Å. The N-terminal GB1 domain and C-terminal His tag are not visible in the electron density maps.
A two-fold symmetry axis runs parallel to the long axis of the BvgS dimer. The N termini of the two protomers are located on the outer surface of the dimer, and their C termini interrupt α helices at the membrane-proximal end of the structure. VFT1 and VFT2 adopt typical Venus flytrap architectures consisting of two α/β subdomains called lobes 1 and 2 (hereafter L1 and L2) separated by a cleft. They have similar topologies with two crossings between the lobes (S1 Fig). The hinge is formed of anti-parallel β strands in VFT2 and flexible loops in VFT1. The VFT2s are followed by the C-terminal (Ct) domains that encompass the Gly 527 -Pro 532 Ct loops and the H19 Ct helices (Figs. 2 and S1). In the absence of membrane constraints, the H19s adopt divergent orientations in the crystal structure. In full-length BvgS they are predicted to continue across the membrane down to the cytoplasmic PAS domain, with a total length of 60 residues.  (Bvg + ) and avirulent (Bvg -) phases of the bacterium are represented for simplicity. Conditions that turn the bacteria to the avirulent phase include low temperatures and negative modulators such as sulfate or nicotinate (NA) ions. The vags (virulence-activated genes) are trans-activated by phosphorylated BvgA, while the vrgs (virulence-repressed genes) are upregulated in the avirulent phase. An intermediate phase occurs at low modulator concentrations (see text). From N to C terminus, 135 kDa-BvgS is composed of two periplasmic VFT domains, a transmembrane segment, a PAS domain, followed by a histidine-kinase (HK), a receiver (R) and a Histidine phosphotransfer (Hpt) domains that make up a phosphorelay (represented by arrows). BvgA is composed of a receiver domain and a helix-turn-helix DNA-binding domain (HTH). B. Structural organization of selected BvgS homologs, with the same color code as for BvgS. Note that the domain composition varies in the family. The cellular functions regulated by these sensor-kinases are also indicated. The two VFT1s are open, while atypically the VFT2s are closed with no ligand in their inter-lobe cavities (Fig. 3), consistent with the structure of VFT2 alone [27]. The VFT1 cavities are each oriented toward the hinge of the VFT2 domain of the other protomer, and the cavities of the VFT2s are each oriented toward the H19 helix of the opposite protomer (Fig. 3).
The VFT1 L1s interact with each other through several hydrogen bonds between their H8s, while the VFT2s are not directly interconnected. Both lobes of the VFT1s, VFT1 L1 and VFT1 L2 , contact the hinge and lobes of VFT2 of the opposite protomer (Fig. 4), forming the largest dimeric interfaces. Other large interfaces occur between VFT1 L2 and VFT2 of the same protomer, and between VFT2 L2 and the Ct domains. In particular, both the Ct loop and the N terminus of H19 strongly interact with VFT2 L2 of the opposite protomer through hydrogen bonds and through π-stacking interactions that involve a conserved residue in the BvgS family, Trp 535 (Fig. 4).

Conformation and dynamics of the VFT domains
In the crystal structure, the VFT1 domains are open and unliganded, while conversely the VFT2 domains are closed without ligands. We performed normal mode analyses of BvgS motions based on a Gaussian network model to identify the main global motions that are accessible to the protein based on its tridimensional structure. The first, lowest-frequency normal modes are usually most relevant to function. For BvgS, the first two modes of motion consist of large motions of one VFT1 L1 (S2 Fig). In contrast, the VFT2s move together as a rigid body, as shown in mode #3. Mode #4 consists of motions of both VFT1 L2s together with the VFT2s. Thus, the first lobes of the VFT1s in particular can make large motions, while the VFT2 motions are more restrained and mainly coupled to each other and to those of the VFT1s. This was confirmed by performing molecular dynamics simulations to measure the evolution of the As the simulations progress the VFT1 mobility is reduced, which suggests that sustained VFT motions may require the feedback from the transmembrane and cytoplasmic portions of BvgS absent from our model. These in silico analyses thus indicate that the X-ray structure reflects bona fide differences between the VFT1 and VFT2 domains in terms of conformation and dynamics. We then asked whether VFT1 closing-as might happen upon binding of a ligand-would affect BvgS activity. We locked the VFT1 domains in closed conformations by generating a disulfide (S-S) bonds across their cavity [33,34]. Two residues located on the edges of the lobes were replaced by Cys to obtain BvgS E113C+N177C (S3 Fig). The corresponding point mutations were inserted into the chromosomal bvg locus by allelic exchange, and we verified the production of the protein and the formation of the S-S bond by immunoblotting (S4 Fig). The in vivo effect of the substitution on BvgS function was then measured by using a reporter system with the lacZ gene under the control of the Bvg-regulated ptx promoter [35]. In vivo formation of the S-S bond in VFT1 abrogates the kinase activity of BvgS (Fig. 5). This phenotype is reverted by the addition of a reducing agent, TCEP, to the growth medium (S5 Fig), which confirms that the S-S bond forms in vivo and shows that the loss of function is related to its presence and not to the Cys substitutions.
The VFT2s remain closed even when isolated [27]. Nevertheless, to maintain them closed in vivo we also generated an S-S bond between their lobes using a similar method as above, yielding BvgS T355C+D442C . We checked that the S-S bond was formed (S4 Fig; see also below). In contrast to VFT1, closing VFT2 was found to have no effect on the BvgS kinase activity as determined with the ptx-lacZ reporter (Figs. 5 and S5). Altogether thus, closing of the VFT1 domains and/or restraining their mobility abrogate BvgS kinase activity. In contrast, closed VFT2 domains correspond to the kinase-on state of BvgS. The different conformations and dynamics of the two VFT domains thus contribute to BvgS function.

Importance of periplasmic domain integrity for BvgS kinase activity
B. pertussis is in the virulent, Bvg + phase by default at 37°C. To determine the role of the periplasmic domain of BvgS in maintaining this kinase-on state, we loosened the connections between the periplasmic and cytoplasmic moieties of BvgS by replacing Trp 535 with Ala. This residue is located in the C-terminal helix H19 and it contributes to connecting each H19 to the VFT2 L2 of the opposite protomer (Fig. 4C). After allelic exchange, the effect of the substitution on BvgS function was measured by using the ptx-lacZ reporter system. The BvgS W535A variant has no kinase activity (Fig. 5). The presence of BvgS W535A in B. pertussis membranes was verified, showing that the substitution does not affect the structure of the protein in such a way as to prevent its integration in the membrane or to cause its proteolytic degradation in vivo (S4 Fig). Thus, the kinase-on state of BvgS depends on tight connections between the periplasmic domains and the transmembrane H19 helices.
To confirm that the periplasmic portion imposes a specific conformation on the cytoplasmic moiety, we introduced other substitutions in the inter-protomer interfaces between the VFT2s and the Ct domains, by targeting residues whose side chains connect the VFT2 L2s and the Ct loops that precede the H19s (Fig. 4C). Thus, Arg 472 and Tyr 473 located in helix H16 of VFT2 L2 form hydrogen bonds with residues of the Ct loop of the other protomer. Their simultaneous replacement by Ala abolishes BvgS kinase activity, while the single-substitution variants BvgS R472A and BvgS Y473A are partially active (Figs. 5 and S5). This indicates that the interprotomer interface between H16 in VFT2 and the Ct loop is critical and that it is maintained by partly redundant interactions. In contrast, substitutions at the tip of a β hairpin in VFT2 L1 whose residues interact with the other face of the Ct loop do not affect BvgS function, as shown with BvgS R324G/T325G/D326G/E327G . The effect of disrupting of specific interactions between the VFT2 L2s and the Ct loops preceding the H19s is consistent with the effect of the W 535 A substitution, showing that the kinase-on state depends on VFT2-Ct domain interprotomer connections.
To identify additional architectural features of the periplasmic dimer critical to maintain BvgS in its kinase-on state, we disrupted specific interactions in other intra-dimer interfaces of BvgS by site-directed mutagenesis. We targeted residues in the large interfaces between the VFT1s and the VFT2s of the opposite protomers (Fig. 4B). The side chains of Tyr 81 and Glu 86 in a β hairpin of VFT1 L1 and that of Glu 200 in helix H5 of VFT1 L2 form hydrogen bonds with residues at the N-and C-terminal sides of the first hinge strand of VFT2, respectively. Two BvgS variants, BvgS Y81A+E86A and BvgS E200A were generated and analyzed as above (Figs. 5, S4 and S5). Neither of them is functional, demonstrating that connections between the two lobes of VFT1 and the hinge of VFT2 of the opposite protomer are essential to maintain the kinaseon state of BvgS. In contrast, the replacement of Gln 463 by Ala in the same large inter-protomer VFT1-VFT2 interface does not affect activity (Figs. 5 and S5). Gln 463 is part of VFT2 but not located in the hinge, unlike the residues of VFT2 in contact with Tyr 81 , Glu 86 and Glu 200 . The loss of kinase activity of the BvgS Y81A+E86A and BvgS E200A variants might result from the loss of constraints applied by the VFT1 lobes on the VFT2 hinge.
Altogether, thus, we have identified interactions in the inter-protomer interfaces between VFT1 and the VFT2 hinge and between VFT2 L2 and the Ct domain that are necessary to maintain BvgS in its kinase-on state. In particular, the substitutions A 472 A+Y 473 A and W 535 A support the idea that the periplasmic domain exerts a strain on the transmembrane domains, causing the cytoplasmic moiety to adopt a specific conformation corresponding to the kinaseon state. The VFT1s contribute to the strain via the close contacts of their two lobes with the hinges of the tight VFT2 domains. Loosening the periplasmic portion or its connections with the transmembrane helices releases the strain, and therefore the cytoplasmic moiety switches to a distinct, kinase-off state.

Modulation by nicotinate requires multiple intra-dimer interactions
Negative modulators turn BvgS to the kinase-off state at millimolar concentrations in laboratory conditions, and they possibly mimic in vivo ligands that might decrease or turn off virulence genes expression at specific stages of the infection. The sites of interaction of these negative modulators are mostly unknown. We have shown that nicotinate binds to isolated VFT2, even though additional sites cannot be ruled out in the dimer [29], and therefore we used nicotinate to determine how the periplasmic moiety contributes to the response of BvgS to negative modulation. The ability of the BvgS variants described above to respond to nicotinate was thus assessed.
The BvgS T355C+D442C variant with a S-S bond across the VFT2 cavity variant is unresponsive to nicotinate but reverts to the wild type (wt) modulation phenotype when the growth medium is supplemented with TCEP (Figs. 5 and S5). This confirms the in vivo formation of the S-S bond and also shows that it, rather than the Cys substitutions, hampers the response to nicotinate. The S-S bond might prevent nicotinate from binding or hamper a conformational changes involved in the response to the negative modulator.
A number of other substitutions similarly abrogate the effect of nicotinate (Figs. 5 and S5). Interestingly, both inter-protomer and intra-protomer interactions are required for BvgS response to negative modulation. These interactions map to the VFT1 L1 -VFT1 L1 , VFT1 L2 -VFT2 L1 , and VFT1 L2 -VFT2 L2 inter-protomer interfaces and to the VFT1 L2 -VFT2 L1 and VFT2 L2 -Ct domain intra-protomer interfaces (Figs. 5 and S3). Altogether, a large set of both inter-and intra-protomer interactions is required for the response of BvgS to nicotinate. The fact that the response to negative modulation strongly depends on the integrity of the periplasmic moiety indicates that the transition from the kinase-on state to the kinase-off state implies a concerted conformational change.

Function of BvgS heterodimers
The importance of the structural integrity of the periplasmic domain for the kinase-on state and for the transition to the kinase-off state was further probed by generating in vivo BvgS heterodimers that harbor one wt periplasmic domain and another one with a substitution. A merodiploid containing two inactive but complementary bvgS copies, one with a substitution of the phosphorylable Asp of the receiver domain (D 1023 N) and the other with a substitution of the phosphorylable His of the Hpt domain (H 1172 Q), will form inactive homodimers and active heterodimers (Fig. 6) [36,37]. Indeed, only heterodimers will be able to restore the phsophorylation cascade of BvgS. We set up this merodiploid expression system in B. pertussis. As shown in Fig. 6, the homodimers formed by BvgS D1023N or by BvgS H1172Q are inactive using the ptx reporter, but the heterodimer BvgS D1023N/H1172Q is functional, displaying kinase activity in the default state and responding to nicotinate like wt BvgS.
We disrupted critical contacts in one side of the dimer by combining the W 535 A variant with the wt periplasmic protomer. The kinase activity of BvgS was measured using the ptx-lacZ system as above. The resulting BvgS homodimer is not functional, similar to the homo-dimeric BvgS W535A variant (Fig. 6). Another variant that harbors the Y 81 A+E 86 A substitutions in the inter-protomer VFT1-VFT2 interface was similarly combined with the wt periplasmic moiety. The heterodimer is also not functional, a phenotype similar to that of the BvgS Y81A+E86A homodimer (Fig. 6). Both results support the model that the periplasmic architecture and more specifically the crucial inter-protomer interfaces identified above impose a kinase-on conformation onto the cytoplasmic moiety via the H19 helices. Releasing the strain in one half of the dimer is sufficient to lose the kinase-on conformation.
We also combined the wt periplasmic moiety with that harboring a S-S bond across the VFT1 cavity. The resulting BvgS heterodimer has no kinase activity (Fig. 6). Thus, both protomers must have the proper conformation and dynamics for BvgS function.
We finally used the heterodimer strategy to test the effect of a substitution that makes BvgS unresponsive to nicotinate. We thus combined a protomer harboring a wt periplasmic domain with that harboring the N 231 A substitution. Asn 231 from VFT1 L2 makes interactions with the Ct loop of the other protomer (S3 Fig), and the BvgS N231A homodimer does not respond to nicotinate (Figs. 5 and S5). The recombinant strain expressing the heterodimer has β-galactosidase activity and interestingly, its sensitivity to nicotinate is partially restored. Thus, the heterodimer responds to 20 mM nicotinate, although it is not fully modulated (Fig. 6). This intermediary phenotype indicates that the transition to the kinase-off state requires higher modulator concentrations when the integrity of the periplasmic domain is slightly compromised.

Discussion
Although the BvgAS system was identified more than 25 years ago [38], the mode of regulation of Bordetella virulence has remained a puzzle. With its kinase-on state by default and its extracytoplasmic domain different from those of classical 'PDC' (for PhoB/ DcuS/CitA) TCS sensor-kinases, BvgS was initially considered an oddity. However, the realization that many bacterial sensor-kinases harbor similar sensor domains and the first clues about its structure and mode of action have made BvgS a model for the family [23,24,27]. Importantly, some of the BvgS homologs are found in major pathogens, including other Bordetella species as well as P. aeruginosa, E. coli, V. cholerae, Y. enterocolitica and B. burgdorferi, in which they control programs such as biofilm formation, efflux pump expression, type III secretion, or nutritional adaptation [28][29][30][31][32]. The BvgS structure establishes the foundations to decipher the molecular mode of action of this poorly characterized family of VFT-containing sensor-kinases, and it may pave the way to develop new, highly specific, anti-infective therapeutic strategies [39].

Structure and Function of Sensor-Kinase BvgS
Our functional analyses based on the BvgS structure support the following model. Specific inter-protomer interactions are necessary to maintain the kinase-on state. The tight architecture of the periplasmic moiety together with the differential dynamics of the VFTs imposes a strain onto the transmembrane H19 helices. In response, the cytoplasmic moiety, beginning with the PAS domain, adopts specific conformation and dynamics that support the kinase and phosphotransfer activities of BvgS. The bacteria are thus in the virulent, Bvg + phase, and they can establish an infection. Switching BvgS to the kinase-off state involves a conformational change of the periplasmic moiety, which modifies the conformation, and possibly the dynamics, of the downstream cytoplasmic PAS and kinase domains. The roles of the avirulent or intermediate phases of B. pertussis are unclear, and in vivo stimuli that may trigger the shift to phosphatase or lower kinase states of activity remain to be identified. However, this work shows that the shift to the kinase-off state can easily be hampered by point mutations at various periplasmic sites. That the ability to reversibly perform the shift that regulates BvgS activity has been conserved through evolution supports the importance of the avirulent or intermediate phases in the lifestyle of B. pertussis. It also strongly argues that the VFT domains perceive negative in vivo signals, which explains the good conservation of their cavities in Bordetella [35].
As shown in this work, one can artificially turn BvgS to the kinase-off state by disrupting specific inter-protomer interactions between the VFT2 domains and the H19 helices. The release of constraints on these helices causes the cytoplasmic portion to adopt an alternative conformation in which BvgS functions as a phosphatase. We have shown that other events putatively relevant to BvgS function, i.e. the closing of VFT1 domains, which might mimic the binding of a ligand, or the binding of nicotinate to VFT2 [27], also turn BvgS to the kinase-off state. Both most likely cause conformational-and/or dynamic-changes to the periplasmic domain, with repercussions below the membrane. A number of BvgS variants with looser connections between the VFT domains are blocked in the kinase-on state and cannot respond to nicotinate, which shows that the shift to the kinase-off state implies a concerted conformational change. Modulation therefore facilitates the transition by shifting the equilibrium from the kinase-on to the kinase-off conformations. It is likely that these two stable states will also differ in their dynamics, and we have indeed obtained preliminary indications that VFT1 dynamics is modified in the modulated state. Similarly, VFT1 dynamics probably contributes to the transition, in line with the emerging paradigm that the dynamics of signaling proteins relates to their function [40].
In the default situation-i.e., at 37°C and without modulators-, the equilibrium is strongly shifted towards the kinase-on state of BvgS, which is therefore fully populated, while conversely the equilibrium is strongly shifted towards the kinase-off state in the presence of high modulator concentrations. This two-state model is compatible with intermediate levels of activity of the BvgAS system, such as those obtained at intermediate modulator concentrations [15], in which kinase-on and-off BvgS proteins may co-exist in equilibrium. It is also most likely the case for the BvgS wt/N231A heterodimer, in which the lack of a critical interaction on one side of the dimer hampers the transition, and therefore only a proportion of the BvgS molecules shift to the kinase-off state at high modulator concentrations. The transition between the two conformations will likely imply relative rotation, translation or shearing movements of the helices that join the periplasmic and cytoplasmic domains, similar to what has been proposed in other signaling proteins [41][42][43][44].
With its clamshell motions, VFT1 behaves like a typical VFT domain. As stated above, the conservation of the VFT1 cavity residues in Bordetella [35] suggests that it binds specific ligand (s) in vivo, and if so our results show that ligand binding to VFT1 will likely cause BvgS to shift to the kinase-off state. In contrast, the VFT2s remain closed in the kinase-on state with no bona fide ligand in their cavity. Whether nicotinate binding to VFT2 opens the cavity or causes another type of deformation remains unknown, but the thermal stabilization of VFT2 upon nicotinate binding argues against the former possibility [27]. The crystal structure of the single VFT domain of a BvgS homolog, the HK29 histidine-kinase of Geobacter sulfurreducens interestingly shows that this VFT is also closed unliganded [45]. Its hinge is composed of two β strands, like that of VFT2 in BvgS, leading those authors to propose that it might not be able to open. Sequence analyses of BvgS homologs indicate that the regions forming the hinge of the membrane-proximal VFT domain contain fewer Gly and more Pro residues than those of classical VFT domains. Therefore, we speculate that in the BvgS family the membrane-proximal VFT domain should be closed and tight for the regulation of sensor-kinase activity. BvgS also responds to various organic and inorganic ions [46]. The binding of these modulating molecules might not necessarily involve the cavity but possibly also interfaces, as in some other VFT-containing receptors [47,48].
The periplasmic moiety of BvgS adopts a highly compact dimeric structure. The helical and strongly intertwined architecture of BvgS may explain how some of its homologs could be functional with three, four or even five predicted VFT domains in tandem [24]. The multiple VFT domains of these sensor-kinases potentially enable the perception of several chemical signals that must be integrated to determine the appropriate response. A compact structure like that of BvgS appears to be better suited for inter-domain communication than more linear arrangements such as those found in the VFT-based iGlu receptors of higher eukaryotes, which might dissipate information coming from the most distal VFT domains [49,50] (S6 Fig). This study of BvgS will undoubtedly serve as a basis to elucidate the function of the other family members. Not all BvgS homologs are in a kinase-on state by default [51], but our mechanistic model can perfectly accommodate sensor-kinases that are regulated in the opposite manner.

Crystallization of BvgS, data collection and processing
The bvgS sequence was amplified by PCR and introduced into pGEV2 [52]. The resulting plasmid encodes the periplasmic portion of BvgS (A 29 -L 544 ) with N-terminal GB1 and C-terminal His tags. The recombinant protein was purified on a Ni 2+ -Sepharose affinity column (GE Life Sciences) and eluted in 10 mM Tris-HCl (pH 8.8), 500 mM NaCl, 200 mM at 4°C. BvgS was concentrated by ultrafiltration to 20 mg/mL. The initial crystallization screening was carried out using the sitting-drop, vapor-diffusion technique in 96-wells microplates with a Cybi-Workstation (Cybio) and commercial crystallization kits (Nextal-Qiagen and JBSscreen). Extremely fragile crystals were obtained at 19°C by manual refinement in 100 mM sodium acetate (pH 4.6), 1.6 M NaCl, in 5 to 7 days. All manual crystallization attempts were carried out using the hanging-drop, vapor-diffusion technique in 24-well plates. The crystals were soaked in a stepwise fashion to a final concentration of 20% glycerol in the crystallization buffer.
A preliminary diffraction screening was performed on 80 crystals. On the best crystal, diffracting at 3.10 Å, a single diffraction dataset (160 images with an oscillating range of 1°) was collected at an X-ray wavelength of 1.5418 Å and a temperature of 100 K using an in-house Mardtb goniostat and a Mar345 image plate detector. Diffraction images were indexed and scaled using the XDS program package [53]. The crystal belongs to the space group P2 1 2 1 2 1 , with cell parameters a = 72 Å, b = 286 Å and c = 128 Å. According to the calculated Matthews coefficient of 2.52 Å 3 Da -1 , a solvent content of 51.3% was estimated.

Structure determination and refinement
The crystal structure containing four monomers in the asymmetric unit was determined by molecular replacement using MOLREP [54] and the crystallographic structure of the isolated VFT2 domain (PDB code: 3MPK) as a search model. Eight copies of the model were located, four occupying the actual positions of VFT2 domains and the other four those of VFT1 domains, whose sequence identity to VFT2 is 24%. The former four copies were positioned using the conventional Patterson search. The latter four copies were found using an iterative procedure alternating refinement of a partial structure with REFMAC [55] and molecularreplacement search in the electron density maps [56]. Subsequent model rebuilding and refinement of the 3.10 Å structure were conducted iteratively using Coot [57] and phenix.refine [58], with the use of local non-crystallographic symmetry restraints. Torsion angles of the structure were optimized by using the Godzilla web server (http://godzilla.uchicago.edu/) [59]. The structure was refined to final R work of 18.1% and R free of 24.4%. The two BvgS homodimers (AB and CD) found in the asymmetric unit can be superimposed with a Cα rmsd of 1.234 Å. A Ramachandran analysis performed with the program Phenix indicated that 94.4% of residues are in preferred conformations and 1.4% in disallowed conformations. The GB1 domains are not seen in the electron density. Analysis of crystal packing revealed an empty space close to the N-terminal segment of each polypeptide chain, indicating that they might be unseen because of crystallographic disorder.

Structure analyses
The 1026-residue AB dimer was used for all analyses. The opening angles for VFT1 and VFT2 were measured using three residues structurally equivalent between the two VFTs, one on the lip of each lobe and one in the hinge. They are Tyr 70 , Gly 244 and Ser 199 for VFT1 and Leu 314 , Glu 490 and Pro 444 for VFT2. The inter-domain interfaces were defined using the PISA server (http://www.ebi.ac.uk/pdbe/prot_int/pistart.html) [60], and http://capture.caltech.edu/ was used to identify cation-π interactions. A model for the closed VFT1 domain was made with Modeller [61] based on its closest homologous structure (PDB code: 1WDN), and residues to be replaced by cysteines were chosen by using http://cptweb.cpt.wayne.edu/DbD/ [62].

Molecular modeling
The methods used for the normal mode analysis and the molecular dynamics simulations with their associated references are described in S1 Protocol. For the analyses of the MD simulations, the opening angles of the VFT domains were calculated based on the geometric centers of the C-α atoms of each lobe and the hinge using a slightly extended definition of the hinge region, encompassing residues 146-151 and 241-246 for that of VFT1, and 390-395 and 486-491 for that of VFT2, to make them less susceptible to noise.

Measurement of BvgS activity
Point mutations were introduced into the chromosome of B. pertussis BPSM by allelic exchange [35]. The BvgS sequence corresponds to that of TohamaI except for a Glu residue at position 705, as in most B. pertussis strains [35]. A ptx-lacZ transcriptional fusion was generated in each recombinant strain [63]. The strains were grown in modified Stainer-Scholte medium [64] non-supplemented or containing 1 to 8 mM of nicotinate. Tris(2-carboxyethyl)phosphine (TCEP, SIGMA) was added to 3-10 mM where indicated. TCEP did not affect the activity or the response to nicotinate of wild type BvgS. The bacteria were grown to mid-exponential phase, harvested by centrifugation, resuspended to an OD 600 of 5 and broken by using a Hybaid Ribolyser apparatus for 30 s at speed 6 in tubes containing 0.1 mm silica spheres. β-galactosidase activities were measured and calculated as described [63]. Each experiment was performed with 3 different clones at different times. The bars represent the standard errors of the mean.

Detection of inactive BvgS variants
The inactive proteins were detected by immunoblotting of B. pertussis membrane extracts using anti-BvgS polyclonal antibodies [23] to verify that the substitution(s) generated no major structural defect that might cause BvgS to misfold and to be degraded intracellularly. BPSM ΔbvgA and BPMS ΔbvgS were described previously [23,35].

Construction of heterodimers
To construct a bvgAS locus deletion strain from BPSM, a Tohama I streptomycin-resistant derivative, sequences on either side of the locus (i.e., the 5' end of the fhaB gene and the 3' end of the bvgR gene) were amplified by PCR using the pairs of oligonucleotides iEco-up and Xma-lo, and Xho-up and HindIII-lo (S2 Table). All the amplicons were first introduced into pCRII-TOPO (Invitrogen) and sequenced. The amplicons were introduced as EcoRI-XhoI and XhoI-HindIII fragments into pUC19 by performing a triple ligation, yielding pUC19 newΔbvgAS . The EcoRI-HindIII insert was then introduced as in [35] into pSORTP1, a mobilizable plasmid for allelic replacement, resulting in BPSM newΔ .
The bvgAS locus was then constructed as a mosaic gene for allelic replacement in BPSM newΔ . We replaced the EcoRI-SpeI part of pUC19 mos [35] using a triple ligation with the EcoRI-XmaI fragment obtained as above and a XmaI-SpeI fragment generated using the primers XmaI-up and SpeI-lo. In the latter amplicon, a natural EcoRI site was eliminated by site-directed mutagenesis with a synonymous mutation. The XbaI-HindIII part of pUC19 mos was replaced by 3 fragments: a XbaI-NcoI PCR fragment generated using the primers XbaI-up and NcoI-lo, a NcoI-XhoI PCR fragment generated using the primers NcoI-up and XhoI-lo, and the XhoI-HindIII fragment described above. The latter fragment contains a natural NcoI site, which was eliminated as above. The final plasmid was called pUC19 mint . The 5.5-kb EcoRI-HindIII insert of pUC19 mint was transferred into pSORTP1 for allelic exchange.
A plasmidic construction of the bvgAS locus was also created starting from pUC19 mint and replacing the EcoRI-SpeI fragment by that generated using the primers pEcoRI-up and SpeI-lo. The natural EcoRI site of this latter fragment was eliminated as above. Finally, the NcoI-HindIII fragment of pUC19 mint was replaced by another fragment generated using the primers NcoI-up and pHindIII-lo, yielding pUC19 mpla . The 4.7-kb EcoRI-HindIII insert was transferred into pBBR1-MCS4 [65], a low-copy, mobilizable and replicative plasmid.
The residues Asp 1023 and His 1172 were replaced by Asn and Gln, respectively, using sitedirected mutagenesis (QuikchangeXL, Agilent). The first mutation was inserted in pUC19 mint and then in pSORTP1 for allelic replacement in BPSM newΔ . The second mutation was inserted in pUC19 mpla and then in pBBR1-MCS4, yielding pBBR mpla to be introduced in Bordetella as an episome.
Successive conjugations were then performed to generate the merodiploids. The first one introduced pSORTP1 containing the bvgAS locus with the D 1023 N substitution into BPSM newΔ , yielding an avirulent strain. Then, pFUS-S1 was integrated to generate the ptx-lacZ transcriptional fusion [63], and the resulting strain was finally transformed with pBBR mpla containing bvgAS with the H 1172 Q substitution. The mutations of the periplasmic domain were introduced via restriction fragment exchange in pUC19 mpla and then in pBBR mpla .

Accession numbers
Atomic coordinates and structure factors for the BvgS periplasmic moiety have been deposited in the Protein Data Bank under the accession number 4Q0C.
Supporting Information S1  The open structure of VFT1 is shown, although the selection of the residues for S-S bond formation was performed using a closed model based on the closest homolog (see Methods). B. Position of the substitutions that make BvgS unresponsive to modulation. One protomer is shown in surface representation, while the other is outlined and colored gray. The pink balls represent the modified residues. A zoom delimited by a dashed orange box shows specific residues whose replacement affects the responsiveness of BvgS to nicotinate but not its kinase activity. Residues Ser 271 to Ser 275 are in the α helix H8 that forms the VFT1 L1 -VFT1 L1 interface. Residues Arg 160 , Phe 230 , Arg 234 , Ser 287 are in the intra-protomer VFT1-VFT2 interface, and Arg 526 is in the intra-protomer VFT2-Ct interface. Residues Gln 463 and Asn 231 are part of the inter-protomer VFT1-VFT2 and VFT1-Ct interfaces, respectively. ΔS and ΔA represent strains with deletions of bvgS and bvgA, respectively. In the right panel in A, the BvgS E113C+E177C band was most likely too faint and fuzzy for detection under non-reducing conditions, but the left panel confirms that the protein is produced and membrane-localized as expected. The amounts of BvgS are generally lower in avirulent strains because the bvgAS operon is positively auto-regulated. The asterisk in the right panel denotes that the oxidized BvgS T355C+D442C variant migrates slightly faster than the wild type control. Note that in vivo S-S bond formation was confirmed by the observation that the recombinant strain producing the BvgS T355C+D442C variant does not respond to nicotinate modulation, unless the S-S bond is reduced (see S5 Fig)