Ligand-Induced Asymmetry in Histidine Sensor Kinase Complex Regulates Quorum Sensing

LuxQ Kinase Activity Requires Dimerization

Genetic, biochemical, and structural evidence from many two-component sensor kinases suggests that the members of this family function in vivo as dimers (Khorchid and Ikura, 2006; Stock et al., 2000). Although we have not been able to isolate full-length LuxQ protein, the histidine kinase dimerization domain itself is indeed dimeric, as judged by equilibrium analytical ultracentrifugation (R.C.K., B. Kokona, R. Fairman, and F.M.H., unpublished data). To determine whether LuxQ functions as a dimer in vivo, we attempted to “poison” wild-type LuxQ by introducing inactive LuxQ mutants, predicting a dominant negative effect if LuxQ, like other sensor kinases, functions by trans-phosphorylation within homodimers (Heermann et al., 1998; Malpica et al., 2004; Ninfa et al., 1993; Surette et al., 1996; Wang et al., 2001; Yang and Inouye, 1991).

The canonical quorum-sensing readout in V. harveyi— bioluminescence— was used to evaluate LuxQ kinase activity. To achieve precise control over AI-2 concentration, we used a strain lacking the AI-2 synthase, LuxS. AI-2 was generated in situ by adding boric acid and synthetically prepared 4,5-dihydroxy-2,3-pentane dione (DPD), which combine to form V. harveyi AI-2 (Miller et al., 2004; Semmelhack et al., 2005). To avoid the confounding effects of AI-1 signaling, we deleted the gene encoding the AI-1 receptor LuxN (see Figure 1). The control experiment (Figure 2A[a], left bar) shows that the luxN⁻, luxPQ⁺, luxS⁻ strain is dark in the absence of AI-2 because LuxQ functions as a kinase, repressing bioluminescence. In the presence of AI-2, the strain is bright because LuxQ kinase activity is inhibited, derepressing bioluminescence (Figure 2A[a], right bar). To assay dimerization, we introduced a truncated LuxQ (residues 1–704) lacking an intact histidine kinase catalytic domain as well as the receiver domain. LuxQ_1–704 was dominant negative for kinase activity, conferring a bright (kinase-off) phenotype in the presence and absence of AI-2 (Figure 2A[b]). Thus, the kinase activity of wild-type LuxQ is poisoned by coexpression of truncated LuxQ, and we conclude that kinase activity requires dimers (or conceivably higher-order oligomers) of wild-type LuxQ. We observed a slight residual AI-2 responsiveness in Figure 2A[b], which we attribute to a small proportion of wild-type LuxQ dimers. We note that LuxN kinase activity is dominant over LuxQ_1–704. Thus, the effect of LuxQ_1–704 on bioluminescence is due to poisoned LuxQ dimers and not to nonspecific interactions with LuxN or other signaling components.

Regulation of Kinase Activity by the LuxPQ Periplasmic Domain

LuxQ functions in conjunction with the periplasmic AI-2 binding protein LuxP. Previously, we showed that LuxP and LuxQ are constitutively associated (Neiditch et al., 2005). Here, we demonstrate that the entire periplasmic domain of the LuxPQ receptor is dispensible for kinase activity. For these experiments, we engineered a V. harveyi strain lacking the AI-1 receptor (LuxN), the AI-2 receptor (LuxPQ), and the AI-2 synthase (LuxS). This strain (FED119), because it lacks both LuxN and LuxQ kinases, is bright (Figure 2B, vector). Introduction of wild-type LuxPQ on a plasmid restores AI-2 responsiveness (Figure 2B[a]). Introduction of LuxQ only, without LuxP, produces a dark (kinase-on) phenotype (Figure 2B[c]), confirming the earlier finding that LuxQ is intrinsically a kinase (Neiditch et al., 2005). Introduction of LuxP in combination with LuxQ_Δp, a LuxQ deletion lacking the entire periplasmic domain, also generates a dark (kinase-on) phenotype (Figure 2B[d]). Thus, removing either LuxP or the periplasmic domain of LuxQ results in a constitutively active LuxQ kinase, demonstrating that no part of the periplasmic region is needed for kinase activity. Taken together, these results strongly imply that the function of the periplasmic region of LuxPQ is to shut off kinase activity in response to AI-2.

We also tested whether LuxQ displays phosphatase activity and, if so, whether this activity is regulated by AI-2. To facilitate examination of LuxQ phosphatase activity, we eliminated its kinase function by changing His-492, the presumed target of the kinase reaction, to alanine. As expected, cells bearing LuxQ_H492A were bright (kinase-off; Figure 2B[e]). This construct enabled us to test whether LuxQ_H492A possesses phosphatase activity because we could exploit the unusual circuitry of the AI-1/AI-2 pathways, driving phosphoryl groups in with LuxN and testing whether various LuxQ constructs could pull them out (Figure 2C, left panel). Introduction of LuxQ_H492A into a luxN⁺, luxPQ⁻, luxS⁺ strain caused an approximately 40-fold increase in light production (Figure 2C, compare vector and [e]), demonstrating that LuxQ_H492A is capable of acting as a phosphatase and counteracting the AI-1-pathway-driven accumulation of LuxO phosphate. The LuxQ_H492A phosphatase activity is insensitive to the presence or absence of AI-2 or LuxP (Figure 2[e and f]). Therefore, the phosphatase activity of LuxQ does not appear to be regulated by AI-2. This phosphatase activity resides within the receiver domain of LuxQ since expression of the receiver domain alone (Figure 2C[g] had essentially the same effect as expression of full-length LuxQ_H492A. Overall, these findings reinforce the conclusion that the periplasmic domain of LuxPQ functions to bind AI-2 and shut off kinase activity.

Our results indicate that the functional unit for LuxPQ is the (LuxPQ)₂ dimer. The periplasmic domains cannot be essential for dimerization, however, because they are dispensible for kinase activity, which requires dimerization (Figure 2B). Indeed, no evidence of an interaction between LuxPQ periplasmic domains has been reported. To investigate further, we performed dynamic light-scattering experiments on the purified periplasmic domain of LuxPQ (LuxPQ_p) in the presence and absence of AI-2. To our surprise, these measurements suggested that AI-2 binding promotes dimerization of the periplasmic domain (i.e., formation of [LuxPQ_p]₂). The hydrodynamic radius (R_H), a measure of molecular size and shape, was 3.4 ± 0.1 nm in the absence of AI-2 but increased to 3.8 ± 0.1 nm in the presence of AI-2. While the R_H of unliganded LuxPQ_p is in excellent agreement with calculations based on the crystal structure (Neiditch et al., 2005), the R_H after AI-2 addition is significantly larger. The measured value (3.8 nm) is intermediate between those predicted for LuxPQ_p monomers (3.4 nm) and (LuxPQ_p)₂ dimers (~4.3 nm) and probably reflects an equilibrium mixture of monomers and dimers. After deleting five residues from the N terminus of LuxP, a modification shown to stabilize the AI-2 bound state in vivo (Neiditch et al., 2005), liganded LuxPQ_p displayed a hydrodynamic radius of 4.0 ± 0.1 nm. These experiments, while preliminary, provided the first indication that AI-2 binding promotes interaction between the periplasmic regions of a (LuxPQ)₂ dimer. We note that the interaction between AI-2 bound periplasmic domains could be much stronger in vivo if, as seems likely, the domains were held in proximity and oriented relative to one another by dimeric interactions occurring elsewhere in LuxQ.

Structures of LuxQ_p and LuxPQ_p Bound to AI-2

If indeed the periplasmic domains of LuxPQ dimerize upon AI-2 binding, it is of interest both to determine the structure of these dimers and to evaluate their physiological significance. As a first step, we crystallized AI-2 bound LuxPQ_p, after first deleting five residues from the N terminus of LuxP (see above). Crystals were improved by deleting, from the termini of LuxQ_p, those residues that were poorly ordered in the previously determined AI-2-free LuxPQ_p crystal structure (Neiditch et al., 2005). The 2.3Å resolution structure (Figures 3A–3C) was determined using phases obtained by molecular replacement (see Experimental Procedures, as well as Table S1 in the Supplemental Data available with this article online). The structure indeed contains two similar LuxPQ_p complexes, as discussed below. We also crystallized LuxQ_p alone and determined its structure to 1.7 Å resolution (Table S1).

Comparing AI-2-free and AI-2 bound structures of the LuxPQ_p receptor reveals that the LuxP subunit, as anticipated, undergoes a major conformation change upon ligand binding (Figure 3D). The observed domain closure is similar to that observed for other periplasmic binding proteins binding to their ligands (Bjorkman and Mowbray, 1998; Miller et al., 2004). In contrast to the large movements within LuxP, the LuxQ_p subunit undergoes only minor conformational changes upon AI-2 binding (Figure 3E). Indeed, in three separate crystal structures—LuxQ_p alone, LuxQ_p complexed with unliganded LuxP, and LuxQ_p complexed with ligand bound LuxP—LuxQ_p displayed almost identical tertiary structures (rms d for C_α carbons = 0.7–1.0 Å for all pairwise comparisons). Thus, although LuxP undergoes a major conformational change upon binding AI-2, this change is apparently not transmitted to the periplasmic domain of the histidine sensor kinase subunit LuxQ. Furthermore, within each LuxPQ_p receptor, the subunit interface between LuxP and LuxQ_p is relatively unchanged.

AI-2-Induced Formation of Asymmetric Dimers

The most striking effect of AI-2 binding observed in the crystal structures is the induction of LuxPQ_p-LuxPQ_p dimerization (Figures 3A–3C). Crystals grown in the presence of AI-2 contain two LuxPQ_p monomers per asymmetric unit, denoted LuxPQ_p and LuxP′Q_p′. LuxPQ_p and LuxP′Q_p′ are related to one another by an approximately 140° rotation around an axis between them (Figure 3C). Nearest the rotation axis lie the two LuxQ_p subunits, interacting with one another, while the two AI-2 bound LuxP subunits are situated more peripherally. Importantly, the four LuxQ_p termini, each of which would be attached to a transmembrane anchor in vivo, are located close to one another on the same face of the LuxPQ_p-LuxP′Q_p′ dimer (Figure 3A). This observation suggests that the LuxPQ_p-LuxP′Q_p′ dimer observed crystallographically is plausible topologically and that, in vivo, the face containing all four LuxQ_p termini would be oriented toward the bacterial inner membrane.

Formation of LuxPQ_p-LuxP′Q_p′ dimers coincides with the formation of a large, asymmetric interface between LuxPQ_p and LuxP′Q_p′ (Figures 3B and 3C). When LuxPQ_p and LuxP′Q_p′ interact, 4176 Ų of previously solvent accessible surface area is buried; 60% of this total is attributable to a new interface between LuxP and LuxQ_p′. The new LuxP-LuxQ_p′ interaction appears to be triggered by the closure of the LuxP subunit around AI-2, which creates a new surface to which LuxQ_p′ binds (Figure 3D). As a result, LuxP interacts with both LuxQ_p and LuxQ_p′. Alanine-scanning mutagenesis experiments (below) identified a set of four LuxP residues important for LuxQ_p′ binding: Thr138, Arg139, Phe143, and Trp167 (see Figure 4A, lower left panel). All four residues (red in Figure 3D) lie in the LuxP domain that rotates 29° (relative to the remainder of LuxPQ_p) upon AI-2 binding. This rotation is essential to make room for LuxQ_p′; without it, LuxQ_p′ would clash sterically with LuxQ_p and therefore be unable to bind LuxP.

The AI-2 bound (LuxPQ_p)₂ dimer is strongly asymmetric: LuxPQ_p and LuxP′Q_p′ are related by a 140° rotation (Figure 3C). In a symmetric complex, LuxPQ_p and LuxP′Q_p′ would be related by a 180° rotation, so the magnitude of the observed asymmetry can be thought of as 40°. Because of this asymmetry, LuxP′ and LuxQ_p are separated by approximately 10 Å and make no direct contact with one another (Figure 3C, bottom left). The large asymmetry in the AI-2 bound configuration is likely important in the inhibition of kinase activity.

Activation of Quorum Sensing via LuxPQ_p Dimerization

To test whether the LuxPQ_p-LuxP′Q_p′ interface observed crystallographically in the presence of AI-2 is important for signal transduction in vivo, we systematically substituted interfacial residues with alanine. Nine alanine substitutions were made (one at a time) in LuxP, and 20 alanine substitutions were made in LuxQ (Table 1). The genes encoding each mutant AI-2 receptor were introduced into V. harveyi FED119, and, in all cases, normal levels of the mutant LuxP and LuxQ subunits were confirmed by Western blotting (data not shown). To evaluate the in vivo activity of the mutant receptors, bioluminescence was measured as a function of autoinducer concentration. Although 17 of the alanine substitutions (gray in Figure 4A) did not cause a significant change in AI-2 responsiveness, all but one of the remaining 12 alanine substitutions (red in Figure 4A) desensitized the cells to AI-2 (Figure 4B). The single sensitizing mutation (L59A; green in Figure 4A) is discussed below. The proportion of alanine substitutions causing a significant phenotype (about 40%) is typical for large protein:protein interfaces (see, for example, Bogan and Thorn, 1998).

The most strongly desensitizing mutations invariably involved aromatic residues in the LuxPQ_p-LuxP′Q_p′ interface (LuxP W167A; and LuxQ F132A, Y133A, and F196A). In several cases, the concentration of AI-2 required to activate the expression of bioluminescence was increased more than 100-fold (Table 1 and Figure 4B). The results provide strong confirmation that the crystallographic LuxPQ_p-LuxP′Q_p′ interface is physiologically relevant. We infer that destabilizing this interface destabilizes the kinase-off (i.e., high cell density) state of the LuxPQ receptor. Because the kinase-off state is rendered less stable relative to the kinase-on state, more AI-2 is required to convert the mutant LuxPQ receptors to their kinase-off states. To confirm that the substitutions destabilize the LuxPQ_p-LuxP′Q_p′ interface, we used light scattering to test whether two of the mutant proteins (LuxP W167A and LuxQ F132A), when paired with wild-type partners, were capable of forming (LuxPQ_p)₂ dimers. As expected, neither mutant LuxPQ_p displayed significant dimerization in the presence of AI-2 (R_H = 3.5 ± 0.1 nm for LuxP W167A; R_H = 3.4 ± 0.1 nm for LuxQ F132A).

To extend this analysis, random mutagenesis of the periplasmic domain of LuxQ was carried out, followed by screens for mutations conferring altered AI-2 signaling phenotypes (see Experimental Procedures). These screens identified six mutations; all six mutant strains produced wild-type levels of LuxQ as judged by Western blotting (data not shown). Similar to the alanine substitutions discussed above, LuxQ S81R, F132L, and A221P (red in Figures 5A and 5B) desensitized V. harveyi to AI-2 (Figure 5C and Table 1). These three residues lie in or near the LuxPQ_p-LuxP′Q_p′ interface, presumably functioning— like the engineered alanine substitutions—to destabilize the interaction. The finding that three random mutations—isolated in a genetic screen because they desensitize cells to AI-2—map to the LuxPQ_p-LuxP′Q_p′ dimer interface provides additional support for the in vivo significance of this interface. By contrast, three other mutations (L59F, M199V, and K203R; green in Figures 5A and 5B) were identified that conferred a constitutively bright (kinase-off) phenotype to V. harveyi; these mutations allow V. harveyi to produce light in the absence of AI-2 (Figure 5C and Table 1). The alanine substitution L59A, like L59F, also conferred a constitutively bright phenotype (see Figure 4). The bright phenotypes were, however, largely LuxP independent (data not shown). Therefore, these substitutions appear to inactivate LuxQ kinase activity by an allosteric mechanism unrelated to AI-2 signaling.

LuxPQ Oligomeric State In Vivo

Bacterial chemotaxis receptors, which have served as a paradigm for bacterial sensory transduction, form large arrays of membrane-spanning dimers (Maddock and Shapiro, 1993). Clustering is proposed to be important for enabling small changes in attractant and repellent concentrations to generate dramatic changes in swimming behavior (Bray et al., 1998; Keymer et al., 2006; Sourjik and Berg, 2004). To test whether LuxPQ forms large clusters in vivo, we tagged the C terminus of full-length LuxQ with yellow fluorescent protein (YFP) and examined its cellular localization (Figure 5D, left panel). The LuxQ-YFP fusion retained wild-type signaling function (data not shown). LuxQ-YFP localized homogenously over the entire cell membrane, and addition of AI-2 did not noticeably alter this pattern (Figure 5D, left panel, and data not shown). By contrast, LuxQ_cyto-YFP, containing only the cytoplasmic portion of LuxQ, displayed diffuse fluorescence throughout the cytoplasm (Figure 5D, middle panel). As a control, we fused the V. harveyi chemotaxis protein CheW to GFP. CheW-GFP localized to punctate membrane arrays (Figure 5D, right panel). We conclude that LuxPQ, unlike the chemotaxis receptors, does not form detectable clusters either in the presence or absence of ligand.

Conformational Changes Caused by AI-2 Binding to LuxPQ_p

To address such questions, we used X-ray crystallography to define ligand-induced structural changes in the two-component sensor kinase complex LuxPQ. LuxPQ is required for detection of a quorum-sensing signal, AI-2, produced and detected by diverse bacteria and proposed to foster interspecies bacterial communication. Comparison of the ligand-free and ligand bound structures of the periplasmic region of LuxPQ reveals that large conformational changes accompany ligand binding. Specifically, the two domains of the LuxP subunit move relative to one another as they clamp down on AI-2 (Figure 3D). Because both LuxP domains contact the periplasmic region of the kinase subunit LuxQ, we initially supposed that the LuxP conformational change might drive a corresponding conformational change in LuxQ_p. This does not, however, appear to be the case: crystal structures of LuxQ_p by itself, or in complex with either unliganded or liganded LuxP, reveal only slight differences in LuxQ_p conformation (Figure 3E). Rather, upon ligand binding, the periplasmic regions of two LuxPQ receptors interact to form an asymmetric (LuxPQ_p)₂ dimer (Figures 3A–3C). In the crystal structure, LuxP′Q_p′ is rotated relative to LuxPQ_p by 140° around an axis normal to the bacterial inner membrane. Extensive mutagenesis results support the argument that the asymmetric interface between LuxPQ_p and LuxP′Q_p′ is biologically relevant and that its role is to stabilize a conformational state in which kinase activity is shut off (Figures 4 and 5).

Deleting the periplasmic domain of LuxQ yields a constitutively active kinase, demonstrating that the periplasmic regions— and, by extension, any interaction between them— are dispensible for kinase activity (Figure 2B). Likewise, in the absence of AI-2 (corresponding to the kinaseon state), we have been unable to detect any interaction between two LuxPQ periplasmic domains using crystallographic or biochemical approaches. We conclude that the kinase-on state— unlike the kinase-off state— is not stabilized by interactions occurring within the periplasm. Rather, the kinase-on state is apparently defined primarily by the transmembrane and cytoplasmic regions of LuxQ. A recent X-ray structure shows that the intact cytoplasmic region of the histidine sensor kinase Thermotoga maritima HK853 is dimeric and symmetric (Marina et al., 2005). If this is also the case for LuxQ, then the kinase-on to kinase-off transition would involve a symmetry-breaking 40° rotation of one LuxPQ receptor relative to the other. Rotation of LuxPQ_p relative to LuxP′Q_p′ presumably repositions the LuxQ transmembrane helices (Figure 6). We hypothesize that this repositioning destabilizes or rearranges the association between the two LuxQ cytoplasmic domains, disabling trans-phosphorylation. Testing this model will require future structural studies that include cytoplasmic and transmembrane regions of LuxQ.

A Clasp Preventing Dimerization

In the previously determined crystal structure of AI-2-free LuxPQ_p monomers (Neiditch et al., 2005), the N terminus of LuxP was observed to bind in a deep groove between the two domains of LuxQ (Figure S1A). Unexpectedly, progressive truncation of the N terminus was found to sensitize V. harveyi to AI-2 (Neiditch et al., 2005). Additional alanine-scanning mutagenesis within the LuxP-LuxQ_p interface confirmed this finding: disrupting the interaction between the LuxP N terminus and the LuxQ groove (specifically, by changing LuxQ residues Asn152 or Trp153 to Ala) lowered the EC₅₀ for AI-2 (Table 1, Figure S1). This observation can be rationalized in light of our new crystallographic results. Specifically, we hypothesize that the intimate interaction between the LuxP N terminus and the LuxQ groove serves as a clasp, locking LuxPQ_p monomers into a conformation unable to form stable (LuxPQ_p)₂ dimers. AI-2 binding causes LuxP to transition from an open conformation to a closed one, simultaneously releasing the clasp. According to our model, these changes would allow the periplasmic domains of two LuxPQ monomers to dimerize, shutting off kinase activity. Importantly, mutations that weaken the clasp would permit dimerization at reduced AI-2 concentrations. Therefore, these mutants would display increased sensitivity to AI-2, as observed (Table 1, Figure S1; Neiditch et al., 2005).

Biochemical evidence provides additional support for the unclasping model: truncating the N terminus of LuxP increased the propensity of AI-2 bound LuxPQ_p to dimerize as judged by dynamic light scattering, and the same truncation was crucial for generating AI-2 bound LuxPQ_p crystals. Taken together, our results lead us to propose that AI-2-induced receptor inactivation proceeds by a two-step mechanism: a first step in which AI-2 binds and the LuxP-LuxQ_p clasp becomes unfastened, and a second step in which LuxPQ_p and LuxP′Q_p′ associate asymmetrically. Although we have not characterized the unclasped intermediate directly, it seems plausible that releasing the clasp might permit the modest reorientations (3–7°) about the LuxP-LuxQ_p and LuxP′-LuxQ_p′ interfaces observed in the liganded structure.

Receptor Clustering and Biological Function

In the X-ray crystal structure, both LuxP and LuxP′ are bound to AI-2 ligands. Nonetheless, the LuxPQ_p-LuxP′Q_p′ complex is asymmetric. This asymmetry arises because only one of the two LuxP subunits is able to bind both LuxQ_p and LuxQ_p′ simultaneously (Figure 3C). The other LuxP (P′ in Figure 3C) is separated from LuxQ_p by approximately 10 Å, so it does not make contact with LuxQ_p. Nonetheless, the close apposition of the two binding surfaces would block the formation of higher-order oligomers.

Indeed, we do not observe higher-order complexes of (LuxPQ)₂ receptors (Figure 5D). This finding contrasts with elegant analyses demonstrating that chemotaxis receptors form large arrays (Maddock and Shapiro, 1993; Parkinson et al., 2005). We suspect that the fundamentally different biological decisions that are elicited by the chemotaxis and the quorum-sensing circuits underpin the striking difference in their respective receptor organizations. In chemotaxis, the receptors detect extremely small fluctuations in chemoattractant concentrations and amplify this sensory information into large alterations in motility. Furthermore, signaling is rapidly reversible, allowing the bacteria to alter swimming direction frequently. For this reason, organization of the chemotaxis receptors into cooperative arrays is proposed to be critical for rapid, high-gain signaling. Importantly, no changes in gene expression occur in chemotaxis signaling. By contrast, the receptors that mediate the quorum-sensing response measure the relatively gradual accumulation of autoinducer signals over time. This sensory information is used to invert the patterns of expression of hundreds of genes. Thus, the quorum-sensing transition is a slow all-or-none biological decision that is not readily reversed. We presume that receptor clustering, which potentially amplifies small signal changes, might have undesirable consequences in quorum sensing by rendering bacteria vulnerable to minor fluctuations in signal concentrations. Thus, at least for AI-2 signaling, the architecture of the LuxPQ receptor is such that clustering— and therefore high-order cooperativity— is precluded, and this ensures that the cells avoid the deleterious consequences that would be incurred from mounting a community-wide response to minor changes in signal concentrations.

Protein Production

V. harveyi LuxQ (residues 53–271; LuxQ_p) was overproduced as a glutathione S-transferase (GST) fusion protein using the expression vector pGEX-4T1 (GE Healthcare). This LuxQ fragment corresponds to the predicted periplasmic region, except for 14 N-terminal and 6 C-terminal residues, which were disordered in the X-ray crystal structure of the AI-2-free periplasmic portion of LuxPQ (Neiditch et al., 2005). V. harveyi LuxP (residues 27–365) was overproduced using the expression vector pBB75. This LuxP fragment corresponds to the entire mature protein (residues 22–365 after signal-sequence cleavage), lacking only the N-terminal five residues previously shown to stabilize the AI-2-free state of the LuxPQ complex (Neiditch et al., 2005). GST-LuxQ_p was expressed in E. coli strain BL21; GST-LuxQ_p and LuxP were coexpressed in E. coli strain BL21 (DE3). Proteins were purified from cell lysates by glutathione agarose affinity column chromatography followed by thrombin cleavage of the GST tags. LuxQ_p contains two heterologous N-terminal residues (Gly-Ser) derived from the thrombin cleavage signal. Proteins were further purified by anion exchange (Source 15Q; GE Healthcare) and size exclusion (S200; GE Healthcare) chromatography.

Dynamic light scattering experiments were performed using a Wyatt Technologies MiniDAWN instrument at 23°C. All samples contained 30 µM LuxPQ_p in 14 mM Tris-HCl (pH 7.5), 105 mM NaCl, 4 mM ammonium acetate, and 2 mM boric acid; AI-2 bound protein samples were supplemented with 40 µM DPD, prepared as described (Schauder et al., 2001). LuxPQ_p containing full-length LuxP (residues 22–365 after signal-sequence cleavage) was prepared as described (Neiditch et al., 2005). Theoretical hydrodynamic radii were calculated from crystal structures using HYDROPRO (Garcia De La Torre et al., 2000).

Crystallization and Diffraction Data Collection

Preliminary AI-2 bound LuxPQ_p crystallization conditions were identified by high-throughput screening using a variant of LuxQ_p containing residues 39–278. High-throughput screening was performed by the Hauptman-Woodward Medical Research Institute using the microbatch-under-oil technique at room temperature (Luft et al., 2003). Following protein engineering to remove the LuxQ_p N and C termini and optimization of the crystallization conditions, X-ray diffraction quality LuxPQ_p crystals were produced by vapor diffusion at 23°C using a 1:1 mixture of protein (20 mg/ml in 20 mM Tris-HCl [pH 7.5], 150 mM NaCl) and well solution (27.5% [w/v] PEG 400, 100 mM potassium nitrate, 100 mM 2-morpholinoethanesulfonic acid monohydrate (pH 6.0), 100 µM DPD, and 1 mM boric acid). DPD was prepared as described (Schauder et al., 2001). LuxQ_p was crystallized in a similar manner using a 1:3 mixture of protein (10 mg/ml in 10 mM Tris-HCl, 150 mM NaCl, pH 7.5) and well solution (8% [w/v] PEG 2000 monomethyl ether [MME], 10 mM Tris-HCl [pH 7.0], and 10 mM NiCl₂). LuxQ_p crystals were cryoprotected by a brief soak in mother liquor solutions containing 25% (w/v) PEG 2000 MME and 25% glycerol; LuxPQ_p crystals did not require cryoprotection prior to freezing. Data were collected on nitrogen-cooled crystals [Table S1] at NSLS beamline X25. Diffraction data were processed using the HKL software package (Otwinowski and Minor, 1997).

Structure Determination and Refinement

The AI-2 bound LuxPQ_p structure was determined by molecular replacement with PHASER (Storoni et al., 2004) using AI-2 bound LuxP (1JX6) and LuxQ_p from AI-2-free LuxPQ_p (1ZHH) as search models. AI-2 was not included in the LuxP search model to ensure that any density observed in the LuxP ligand binding pocket was uninfluenced by phase bias. AI-2 was built into this density during subsequent rounds of refinement and was found to be identical to the structure of AI-2 in LuxP crystallized in the absence of LuxQ_p. The LuxQ_p structure was determined by molecular replacement with MOLREP (Vagin and Teplyakov, 1997) using LuxQ_p′ from the LuxPQ_p structure as a search model. Iterative cycles of building in O (Jones et al., 1991) and refinement in CNS (Brunger et al., 1998) or REFMAC5 (Murshudov, 1997) were then performed.

The final LuxPQ_p model lacks LuxQ residues 53–57 (N terminus), 238–247, and 271 (C terminus); LuxQ′ residues 239–247 and 271 (C terminus); and LuxP′ residues 109–118. The model includes one heterologous Ser residue, derived from thrombin cleavage, at the N terminus of LuxQ′. The LuxPQ_p and LuxQ_p structures have excellent geometry with only one residue (Asn216) falling outside of the additional allowed regions of the Ramachandran plot as determined by PROCHECK (Laskowski et al., 1993) both in uncomplexed LuxQ_p and in LuxQ_p′ of the LuxPQ_p structure (Table S1). The building of four Ni²⁺ ions into the electron density of uncomplexed LuxQ_p was justified by their hexameric coordination state, their binding to histidine and aspartic acid residues, and the presence of 10 mM NiCl₂ in the crystallization conditions.

Structure-based sequence alignment was guided by DALI (Holm and Sander, 1993) and PyMOL (DeLano, 2002). Analysis of protein domain motion was carried out using DynDom (Hayward and Lee, 2002; Lee et al., 2003), LSQMAN (Kleywegt, 1996), and PEK2 (P.D.J., unpublished data). Buried surface and contacts were analyzed in CNS. Molecular graphics were produced using PyMOL.

V. harveyi Strain Construction

To make the luxN, luxPQ, luxS mutant V. harveyi strain FED119, the luxS gene of V. harveyi BB170 (luxN::Tn5) (Bassler et al., 1993) was deleted using plasmid pKAS46-ΔluxS as described (Bassler et al., 1993; Henke and Bassler, 2004b), generating strain JMH610. Cosmid pBB713 (Bassler et al., 1994) carrying luxPQ was modified to delete the DNA specifying amino acids 34–366 and 1–843 of LuxP and LuxQ, respectively. This plasmid, pDLS100, was introduced into JMH610, and the deletion transferred to the chromosome to generate V. harveyi strain FED119.

Mutagenesis of luxPQ

The luxPQ operon was amplified from wild-type V. harveyi by PCR and inserted into pEVS143 (Dunn et al., 2006) at the SphI and BamHI sites, making pFED367. A chloramphenicol resistance cassette from pHP45Ω-Cm (Fellay et al., 1987) was inserted into pFED367 at a SmaI site located in the kanamycin resistance gene, generating pFED368. Site-directed mutagenesis was conducted on pFED368 using complementary PCR primers containing targeted mutations in luxP and luxQ. The entire plasmid was amplified using PfuUltra high-fidelity polymerase (Stratagene).

To generate random amino acid substitutions in the periplasmic domain of LuxQ, luxQ_p was amplified with Mutazyme DNA polymerase (Stratagene). A library of mutant fragments was subcloned into pFED132 (Neiditch et al., 2005) to regenerate luxPQ, and these constructs were introduced into V. harveyi strains. To identify luxPQs conferring a bright phenotype to V. harveyi in the absence of AI-2, the library was introduced into V. harveyi lacking luxN and luxS (JMH610). To identify luxPQ mutations conferring a dark phenotype in the presence of AI-2, the mutant library was introduced into a V. harveyi strain lacking luxN (BB170). In both cases, luxPQ genes from plasmids that altered the V. harveyi Lux phenotype were collected and sequenced.

Additional site-specific mutations were engineered in luxP and luxQ to examine kinase and phosphatase activities. The LuxP_ss secretion mutant, containing substitutions L9R and I10K in the signal sequence (Neiditch et al., 2005), and the LuxQ_H492A mutant were constructed by site-directed mutagenesis as above. The LuxQ_Δp allele was constructed by deleting amino acids 45–265. The in-frame deletion of luxP has the entire open reading frame of luxP replaced with an NheI site and maintains the native ribosome binding site and first codon of luxQ. To express the receiver domain of LuxQ, DNA encoding LuxQ amino acids 708–859 (LuxQ_receiver) was amplified by PCR and cloned into pEVS143 at the NheI/BamHI restriction sites (Dunn et al., 2006), generating plasmid pFED553. Expression from pFED553 was driven by the Ptac promoter and induction with 10 µM IPTG.

Luciferase Bioassays

AI-2 responses were measured from V. harveyi FED119 carrying pFED368 or plasmids harboring mutant luxPQ alleles. Strains were grown overnight in AB medium (Greenberg et al., 1979) containing 10 µg/ml chloramphenicol. Cultures were diluted 1/10,000 into identical medium supplemented with 0.5 mM boric acid and dispensed into microtiter plates containing from 0.1–10,000 nM synthetically prepared DPD (Semmelhack et al., 2005). The plates were incubated at 30°C for 16–20 hr. EC₅₀ values are defined as the concentration of DPD required to produce 50% maximal bioluminescence and were obtained by curve fitting using Prism (GraphPad Software). Bioluminescence assays were also performed on strains of V. harveyi that endogenously produce AI-2 (luxS⁺). For these assays, overnight cultures were diluted 1/5,000, and the relative light production (counts per minute per colony-forming unit × 1,000) was monitored throughout growth. Bioluminescence values reported in bar graphs show the point in growth at which time wild-type V. harveyi produces the lowest level of light, typically at 1 × 10⁶ cfu/ml.

Construction and Visualization of YFP Fusions

The complete luxQ ORF and a region encoding only the cytoplasmic domain of luxQ (nucleotides 1114–2577; luxQ_cyto) were each fused to the A206K allele of eyfp (Zacharias et al., 2002) via a flexible Gly-Gly-Gly-Ala linker. luxQ-eyfp was cloned in place of luxQ on pFED368, and luxQ_cyto-eyfp was cloned under the Ptac promoter of pEVS143 (Dunn et al., 2006), to generate plasmids pFED376 and pFED537, respectively. egfp was fused to the C terminus of the V. harveyi cheW gene by inserting the cheW open reading frame into pEVS143 at an NheI site. All fusions were visualized using a Nikon Eclipse TE200 microscope with a 100× objective equipped with a Photometrics Cool- SNAPHQ CCD camera. Images were deconvolved using SoftWoRx v. 2.5 (Creative Softworx).

Immunoblotting

V. harveyi whole-cell extracts were sonicated, electrophoresed on SDS-PAGE gels, and blotted to nitrocellulose for immunostaining with anti-LuxP or anti-LuxQ rabbit antisera followed by anti-rabbit IgG HRP conjugated antibody. Protein was detected using Chemi Glow substrate by a FluorChem imaging system (Alpha Innotech).