A bipartite periplasmic receptor–diguanylate cyclase pair (XAC2383–XAC2382) in the bacterium Xanthomonas citri

The second messenger cyclic diguanylate monophosphate (c-di-GMP) is a central regulator of bacterial lifestyle, controlling several behaviors, including the switch between sessile and motile states. The c-di-GMP levels are controlled by the interplay between diguanylate cyclases (DGCs) and phosphodiesterases, which synthesize and hydrolyze this second messenger, respectively. These enzymes often contain additional domains that regulate activity via binding of small molecules, covalent modification, or protein–protein interactions. A major challenge remains to understand how DGC activity is regulated by these additional domains or interaction partners in specific signaling pathways. Here, we identified a pair of co-transcribed genes (xac2382 and xac2383) in the phytopathogenic, Gram-negative bacterium Xanthomonas citri subsp. citri (Xac), whose mutations resulted in opposing motility phenotypes. We show that the periplasmic cache domain of XAC2382, a membrane-associated DGC, interacts with XAC2383, a periplasmic binding protein, and we provide evidence that this interaction regulates XAC2382 DGC activity. Moreover, we solved the crystal structure of XAC2383 with different ligands, indicating a preference for negatively charged phosphate-containing compounds. We propose that XAC2383 acts as a periplasmic sensor that, upon binding its ligand, inhibits the DGC activity of XAC2382. Of note, we also found that this previously uncharacterized signal transduction system is present in several other bacterial phyla, including Gram-positive bacteria. Phylogenetic analysis of homologs of the XAC2382–XAC2383 pair supports several independent origins that created new combinations of XAC2382 homologs with a conserved periplasmic cache domain with different cytoplasmic output module architectures.

Cyclic diguanylate monophosphate (c-di-GMP) 2 is a nucleotide second messenger responsible for the control of a variety of bacterial features, most of them involved in lifestyle transitions (1). Several reports have demonstrated the role of c-di-GMP in exopolysaccharide production, biofilm formation, chemotaxis, and different modes of motility (swimming, twitching, and swarming), virulence, antibiotic resistance, cell morphology, and cell cycle regulation (2)(3)(4)(5)(6). Generally, high intracellular c-di-GMP levels correlate with a sessile state and biofilm formation, whereas low levels correlate with increased motility.
Intracellular c-di-GMP levels are controlled by diguanylate cyclases (DGCs), which are GGDEF domain-containing proteins, and by phosphodiesterases (PDEs) harboring EAL or the less common HD-GYP domains (7,8). DGCs join two GTP molecules to form one molecule of c-di-GMP and two molecules of pyrophosphate. These enzymes act as dimers because each GGDEF domain can bind only one GTP at its active site (9), and the two domains must come together in the correct orientation to form a competent active site that leads to the generation of the cyclic product. Furthermore, many EAL domains that catalyze the hydrolysis of c-di-GMP to pGpG may depend on dimerization to be active (10 -12). The requirement of a dimerization step and the frequent presence of accessory domains, most often located at the N-terminal of DGCs and PDEs, suggest that these enzymes are under tight regulation to precisely control when, where, and how much c-di-GMP is to be produced or degraded. In fact, over half of these proteins contain partner domains such as PAS (13), GAF (14), and HAMP (15), which could function as sensors or signal transmitters to the catalytic part of the molecule. Furthermore, approximately half of the many thousands of these enzymes annotated in the protein databases are also integral membrane proteins, often with a sensor domain located within the bacterial periplasm and the DGC or PDE domain located in the cytoplasm (16).
Despite the broad existence of inner membrane-spanning DGCs and PDEs in bacterial genomes, only a few studies have so far delved into their mechanisms of activation. In Pseudomonas aeruginosa, the YfiBNR operon, required for cystic fibrosis persistence, codes for a transmembrane DGC, which has its activity finely controlled by the two other proteins coded by the operon, one localized in the periplasm and the other localized in the outer membrane (17). LapD from Pseudomonas fluorescens contains a dual GGDEF-EAL domain and is the bestunderstood member of this class of proteins. Instead of being an active DGC or PDE, LapD senses intracellular c-di-GMP fluctuations through the EAL domain and transmits this information to its periplasmic domain, which in turn recruits a protease (LapG) that cleaves a crucial adhesin that promotes biofilm formation in response to phosphate starvation (18). In this case, information regarding intracellular c-di-GMP concentration is transmitted from the bacterial cytoplasm to the periplasm and is thus labeled an "inside-out" mechanism of signal transduction, opposite to that which occurs in the YifBNR system.
Xanthomonas is a bacterial genus comprising pathogens that infect dozens of plant species, many of which are of significant economic interest, such as orange, rice, sugarcane, and crucifers. To cope with constant environmental changes as they colonize different host tissues, the genomes of these bacteria code for dozens of signaling proteins, among which are transmembrane DGCs and PDEs. Xanthomonas citri subsp. citri (Xac), the causal agent of citrus canker, has 18 GGDEF, 3 EAL, 3 HD-GYP, and 11 GGDEF-EAL domain-containing proteins, evidence of a complex c-di-GMP signaling network in this organism (19). Of these proteins, 15 are predicted to contain membrane-spanning helices and periplasmic domains (19). One of these is XAC2382, which has an uncharacterized N-terminal periplasmic Cache domain followed by a cytoplasmic portion that contains HAMP, GGDEF, and EAL domains. As we show here, the xac2382 gene is co-transcribed with its neighboring gene xac2383. Interestingly, homologs of the XAC2382-XAC2383 pair can be found in a wide variety of bacterial phyla, although in some cases the cytosolic GGDEF-EAL module has been replaced with another signaling output module (i.e. histidine kinase or 54 activator domains). We therefore studied XAC2382 and XAC2383, their interactions, and their in vivo functions. We present genetic and biochemical evidence that supports the hypothesis that XAC2383 acts as a periplasmic sensor, which regulates the DGC activity of XAC2382 via protein-protein interactions. We also determined the crystal structure of XAC2383 in complex with chloride ion, phosphate, pyrophosphate, and adenosine triphosphate. These results suggest that the XAC2383-XAC2382 system responds to the extracellular levels of anionic phosphate-derived ligands.

xac2382 and xac2383 genes are co-transcribed and code for an integral membrane Cache_HAMP_GGDEF_EAL protein and a periplasmic protein
The xac2382 gene codes for a 76.8-kDa protein containing a periplasmic Cache domain, delimited by two transmembrane helices (residues 10 -32 and 195-217), followed by three cytoplasmic domains: HAMP (named for its occurrence in Histidine kinases, Adenyl cyclases, Methyl-accepting proteins, and Phosphatases, residues 198 -268), GGDEF (diguanylate cyclase, residues 277-438), and EAL (c-di-GMP phosphodies-terase, residues 457-692) (Fig. 1, A and B, and Fig. S1). Cache domains are the largest superfamily of extracellular sensor domains in bacteria, often encountered as an N-terminal module of multidomain transmembrane proteins (20,21). The HAMP domain contains a canonical heptad repeat signature (residues 252-266; Fig. S1) known to mediate the formation of coiled-coil structures (22) and hence may be important for XAC2382 dimerization (23). The GGDEF domain possesses all of the nine conserved residues that are considered important for catalytic activity based on previous mutagenesis studies on DGC enzymes (9,24). The XAC2382 EAL domain is also predicted to be catalytically active because it presents conserved residues at positions predicted to be involved in metal coordination, c-di-GMP binding, and general base catalysis (25,26). Furthermore, the so-called loop 6 motif (DFG(T/A)GYSS) recognized to be important for the activity of this class of PDEs has only two substitutions (Tyr to Phe and Ser to Gly at positions 6 and 8, respectively). Fig. S2 present alignments of the XAC2382 GGDEF and EAL domains with well-characterized homologs and highlights the details of the features described above.
The xac2383 gene (Fig. 1, A and B) was originally annotated as a phosphonate-binding protein (PhnD domain) (27) that belongs to the superfamily of periplasmic binding proteins (PBPs) that bind to different solutes, including amino acids, vitamins, sugars, and ions (28 -31). It has been shown that some PhnD homologs bind phosphates instead of phosphonates, which illustrates how the ligand specificity for these proteins is often not clear from the primary structure (32,33). We note that the stop codon of xac2383 overlaps with the start codon of xac2382, and a fragment extending from the end of the xac2383 gene to the beginning of the xac2382 gene can be amplified from a preparation of Xac cDNA (Fig. 1C). These observations immediately bring up the possibility that these two proteins could work together in a common pathway. Another gene (xac2384) is found upstream of xac2383 (the two genes do not overlap and are separated by 54 nucleotides) and is annotated as a methylase belonging to the SpoU family (Fig. 1A). Transcription sites have been mapped in the genome of the closely related species Xanthomonas campestris pv. campestris strain B100 (34). In that species, the homologous locus has an almost identical genetic structure, and nucleotide sequence and initiation start sites were mapped to position 2,223,411, which is 25 bp upstream of the start site of xcc-b100_1902, the xac2384 homolog, and to position 2,224,252, which is 21 bp upstream of the start site of xcc-b100_1903, the xac2383 homolog ( Fig. S3) (34). No transcription start site was observed in the vicinity of the start codon of xcc-b100_1904, the xac2382 homolog (34). Therefore, the evidence so far is consistent with independent transcriptional control of xac2384 and xac2383/xac2382. Because xac2384 homologs are not present in this cluster in other bacterial species beyond the Xanthomonadaceae family, it was not an object of study in this investigation.

XAC2383 interacts with the XAC2382 periplasmic Cache domain
XAC2383 has a signal peptide for periplasmic localization, and XAC2382 has a predicted Cache periplasmic domain flanked by transmembrane helices (Fig. S1). Because their genes Periplasmic regulation of a diguanylate cyclase are co-transcribed, we asked whether XAC2383 might be interacting with the XAC2382 periplasmic Cache domain (XAC2382 Cache ). To test this hypothesis, we co-expressed in Escherichia coli XAC2383 lacking its signal peptide (XAC2383 31-309 , 31.3 kDa) and residues 37-192 of XAC2382 (XAC2382 Cache_37-192 , 16.9 kDa), the latter carrying an N-terminal His tag that was used to purify the complex by affinity chromatography using a Ni 2ϩ -His-trap column (Fig. S4). The two proteins are co-eluted from a size-exclusion column (GE Superdex 200 10/300 GL) as a single peak with an average mass of 126 Ϯ 8 kDa calculated by multiangle light scattering (Fig.  1D). Additionally, size-exclusion chromatography analyses of isolated XAC2382 Cache_37-192 and XAC2383  show that the proteins elute with apparent masses close to their theoretical monomer masses (Table 1). These results indicate that oligomerization is induced upon interaction between XAC2383 31-309 and XAC2382  . Because XAC2382 is predicted to be an active DGC, we make the assumption that XAC2382 Cache_37-192 is a dimer or multiple of dimers in the complex. For a XAC2382 Cache_37-192 -XAC2383 31-309 complex with 2:2 stoichiometry, we would expect a mass of 96.4 kDa for a (2:2) 1 complex and a mass of 192.8 kDa for a (2:2) 2 complex. For a XAC2382 Cache_37-192 -XAC2383 31-309 complex with 2:1 stoichiometry, we calculate theoretical masses of 65.1 kDa for the (2:1) 1 complex and a mass of 130.3 kDa for a (2:1) 2 complex. The experimentally observed apparent mass lies between the calculated theoretical masses for the (2:2) 1 and (2:2) 2 complexes and very close to the theoretical mass of the (2:1) 2 complex. It is important to point out that the peak in which the complex elutes is asymmetric, which may be indicative of an equilibrium of oligomeric states (Fig. 1D). We therefore cannot at the moment precisely determine the stoichiometry of the XAC2382 Cache_37-192 -XAC2383 31-309 complex. The physiologically relevant stoichiometry may be influenced by the transmembrane and heptad repeat motifs in the XAC2382 HAMP

Periplasmic regulation of a diguanylate cyclase
domain (both absent in the soluble XAC2382 Cache construct used in the interaction studies).

Crystal structure of XAC2383 reveals a periplasmic binding protein topology and a positively charged surface lining the central ligand-binding groove
We solved the crystal structure of XAC2383 31-309 at 1.9 Å resolution using single-wavelength anomalous dispersion (Tables S1-S3 and see details under "Experimental procedures"; structure deposited with PDB code 5UB3). The crystal belongs to the P2 1 2 1 2 1 space group with two protein molecules in the asymmetric unit. The molecules interface one another through an area of 798.8 Å 2 , calculated by PISA (35). This area is not predicted to be sufficient to sustain a dimer in solution, which is consistent with the monomeric structure deduced from analytical gel filtration of XAC2383   (Table 1).
The protein has two topologically similar lobes (lobes 1 and 2) separated by a central groove (Fig. 2, A and B). Each lobe is composed of a core of a mixed six-stranded ␤-sheet and five ␣-helices. Two antiparallel ␤-strand extensions link the two lobes, and an additional helix from the C terminus of the protein lies between the lobes (Fig. 2B). This topology (sometimes called Venus flytrap) corresponds to that of the periplasmic binding protein (PBPs) family that often acts as components of ABC transporters (36) but have also been observed as parts of integral mem-brane-signaling proteins, as a result of gene fusion (37). In all cases in which these proteins have been structurally characterized, the cleft at the interface between the two domains is involved in the binding of a variety of solutes (38). Furthermore, PBPs have been observed to undergo structural changes upon ligand binding, bringing the two lobes into closer proximity (39 -41).
A structural search using the Dali Server (42) identified the phosphonate-binding protein PhnD from E. coli as the database entry with the highest degree of structural similarity despite a low sequence identity (21%). Two PhnD structures are available, one with the 2-aminoethylphosphonate (2AEP) ligand bound (PDB code 3P7I) and one with no ligand (apo; PDB code 3S4U) (41). Although the XAC2383 structure is more similar to the 2AEP-bound form of PhnD (r.m.s.d. for backbone atoms of 3.4 Å) than to apo-PhnD (r.m.s.d. ϭ 5.6 Å), comparison of the structures reveals that XAC2383's central cleft between the two lobes of the protein is more open than that of 2AEP-bound PhnD but less open than that of apo-PhnD (Fig. 2C). When this aperture is measured using the distances between the ␣-carbons of residues at homologous positions in both structures, the XAC2383 aperture is 0 -7 Å more open than in 2AEP-PhnD but is 3-13 Å less open than in apo-PhnD.
Inspection of the electron density in the vicinity of the putative XAC2383 ligand-binding cleft identified a spherically sym- . C, comparison of the aperture between the lobes in-between XAC2383 (PDB code 5UB3), PhnD bound to 2AEP ligand (PDB code 3P7I), and apo-PhnD (PDB code 3S4U). Lobe 1, yellow; lobe 2, orange. The aperture between lobes in XAC2383 is smaller than that observed for apo-PhnD but larger than that observed for the PhnD-2AEP complex. The chloride ion bound to XAC2383 is shown as a green sphere. The 2AEP ligand bound to PhnD is completely buried and therefore not visible.

Periplasmic regulation of a diguanylate cyclase
metric density that could not be satisfactorily modeled as water (Fig. 3A). We modeled a chloride ion to fit this density because (i) the crystallization conditions contained Ͼ300 mM NaCl; (ii) NH groups are frequently observed to coordinate chloride ions (43), and the ligand is coordinated by two main chain NH groups from residues Thr-153 and Ser-154; and (iii) the coordination distances are all greater than 3 Å (3.05, 3.18, 3.26, 3.59; Fig. 3A), consistent with chloride ion coordination (43).
In addition to the main-chain NH groups of Thr-153 and Ser-154, the chloride ion is coordinated by the Ser-154 sidechain hydroxyl group and a water molecule that bridges to the Ser-152 side-chain hydroxyl (Fig. 3A). This Ser-Thr-Ser (STS) motif also participates in the binding of the ϪPO 3 2Ϫ group of the 2AEP ligand bound to E. coli PhnD (Fig. 3B). Furthermore, this STS triad is not absolutely conserved in the annotated phosphonate binding family in the Pfam database (20), which suggests that the phosphonate-binding family contains members that bind other types of ligands as well.
Although XAC2383 and PhnD from both E. coli and P. aeruginosa all have an STS motif, XAC2383 differs from the latter two proteins at positions in which these make contacts with the amino moiety of the 2AEP ligand. In the case of E. coli PhnD (PDB code 3P7I), negatively charged residues Glu-177 and Asp-205 interact with the positively charged amino group ( Fig. 3B) (41). For P. aeruginosa PhnD (PDB code 3N5L), the corresponding residues participate in the coordination of an unidentified ligand whose electron density is also consistent with 2AEP (data not shown). In the case of XAC2383, the corresponding residues are hydrophobic (Val-224 and Ala-241, respectively; Fig. 3C). These observations suggest that the phosphonate-binding protein family has a well-conserved primary STS motif responsible for ϪPO 3 2Ϫ binding plus a more evolutionarily diverse secondary site that may distinguish between different R-groups.

XAC2383 binds to compounds harboring phosphate moieties
The surface representation of XAC2383 shows that the central groove has a high density of positively charged residues that could favor the binding of anionic ligands (Fig. 4A). We therefore performed soaking experiments to test whether XAC2383 crystals could bind different phosphonates or phosphates with anionic R-groups. Crystals soaked overnight with the relatively

Periplasmic regulation of a diguanylate cyclase
small phosphonoacetic acid (negatively charged R-group) or 2-aminoethylphosphonate (positively charged R-group) failed to provide evidence for ligand binding, except for the previously observed chloride ion. However, when we turned our attention to phosphate derivatives, we were able to observe unambiguous electron densities for the following bound ligands: P i (PDB code 5UB4), pyrophosphate (PDB code 5UB6), and the triphosphate nucleotide ATP (PDB code 5UB7) (Fig. 4, B-D, and Tables S1-S3). In all cases, the added phosphates at 2 mM concentration were able to compete with and displace the bound Cl Ϫ ion (present in the mother liquor at a concentration of Ͼ300 mM). The electron density maps show that the terminal phosphate moiety interacts with the 152 STS 154 motif (Fig. 4, B-D). Besides the 152 STS 154 motif, residues Thr-95 and Lys-193 were also frequently found making contacts with the terminal phosphate of all the ligands. Subtle differences were observed in the position of side chains contacting residues in the two molecules of the asymmetric unit. For example, in the phosphate-bound structure, Lys-193 from chain A interacts with an oxygen atom from phosphate molecule 1 at a distance of 3.1 Å, whereas Lys-193 from molecule B is too far to contact oxygen from phosphate molecule 2 (5.0 Å). In the pyrophosphate-bound structure, the same residues interact with the P2 phosphate group, as described above, whereas the P1 phosphate interacts with Lys-193 and with the Arg-240 side chain, which was modeled in a double conformation (Fig. 4C). Furthermore, the water occupancy at the binding site varies in both the number of water molecules and in the diversity of coordination types among the structures.
In the ATP-bound structure, the ␤and ␥-phosphates occupy the site occupied by pyrophosphate in the pyrophosphate-bound structure described above. Arg-240 interacts with both the ␣and ␤-phosphate groups, and an additional basic side chain, Arg-116, interacts with the ␣-phosphate. We observe clear electron density for the whole ATP molecule bound to chain B of the asymmetric unit (Fig. 4D). Here, the ribose 2ЈOH group makes contact with the His-46 side chain and the adenine 6-amino group hydrogen-bonds with the Gln-148 side chain. In chain A of the asymmetric unit, we observe density only for the triphosphate moiety of the bound ligand, indicative of flexibility of the ribose and adenine base. Omit maps for all the ligands bound to the two chains in the asymmetric unit are shown in Fig. S5. We also collected data for XAC2383 crystals soaking with GTP and could observe density for the triphosphate groups but not for the ribose ring and nitrogenous base. Interestingly, we did not observe any ligand electron density in crystals soaked with nucleotide monophosphates, nucleotide diphosphates, hexaphosphate, or phytate.
An atomic sphere representation of XAC2383, whose coloring scheme is based on the degree of sequence conservation derived from an alignment of 44 XAC2383 homologs, was generated ( Fig.  S6) (44). This figure shows that the ligand-binding site in the cleft region between the two lobes is highly conserved. Also, lobe 1 contains a greater portion of conserved residues outside the ligand-binding site than does lobe 2. This suggests that lobe 1 might constitute an important site for interaction with other proteins as, for example, with the XAC2382 Cache domain.
In conclusion, the results of the limited number of soaking assays that we performed seem to indicate that XAC2383 preferentially binds compounds harboring phosphate moieties, and not phosphonates, but further studies must be done to more specifically define ligand-binding preferences in solution. Soaking assays can sometimes conceal the true affinities for potential ligands because crystal packing forces restrict dynamics and may lock the receptor in a suboptimal configuration. In any case, the XAC2383-binding groove is lined by basic residues, which strongly suggest that it binds phosphates or phosphonates with a negatively charged R-group.

XAC2382 and XAC2383 both regulate bacterial motility
Diguanylate cyclases and c-di-GMP phosphodiesterases regulate bacterial lifestyle, often through the control of motility (4,45). We therefore produced a ⌬xac2382 mutant and evaluated its sliding motility on a semisolid agar surface. Sliding motility is impeded by the presence of type IV pili (46,47). After 4 days of growth at 30°C, the knockout strain's motility was clearly greater than that of the WT strain ( Fig. 5A and Fig. S7), a phenotype compatible with a decrease in c-di-GMP levels (48,49). This suggests that under the experimental conditions used, the DGC activity of the GGDEF domain may predominate over the PDE activity of the EAL domain. Complementation of the mutant strain with a pBRA-derived plasmid carrying the full-length xac2382 gene under the control of the araBAD promoter restored the motility to levels observed for the WT strain, even under lowlevel leaky expression under noninducing conditions ( Fig. 5A and  Fig. S7). Furthermore, the ⌬xac2382 knockout strain complemented with the full-length construct presented a tendency to aggregate when cultivated in liquid medium (Fig. S8). This phenotype is also characteristic of high c-di-GMP levels (49,50).
We then sought to evaluate the role of the protein coded by the upstream ORF, xac2383, in this phenotype. Contrary to what was observed for the ⌬xac2382 strain, the ⌬xac2383 strain displayed decreased surface sliding motility (Fig. 5A and Fig.  S7). When ⌬xac2383 was transformed with a plasmid expressing the full-length xac2383 construct, sliding motility was restored to levels even greater than that observed for the WT strain ( Fig. 5A and Fig. S7). The opposite effects on motility observed for the ⌬xac2382 and ⌬xac2383 strains suggest that both proteins may act in the same signaling pathway. The observation that the ⌬xac2383 motility phenotype is similar to that observed in cells overexpressing catalytically active XAC2382 or its fragments raised the hypothesis that XAC2383 interacts with the periplasmic Cache domain of XAC2382 and that this interaction results in an inhibition of XAC2382 DGC activity. Below we present data that are consistent with this hypothesis.

XAC2382 is an active DGC in vivo and requires functional GGDEF and HAMP domains for this activity
Different expression constructs of XAC2382 in the pBRA vector were introduced into the ⌬xac2382 knockout strain to address the importance of each domain on protein function ( Fig. 5A and Fig. S7). We observed that a construct expressing only the HAMP-GGDEF domains (residues 198 -446) was sufficient to revert the knockout strain's sliding motility to WT levels under the experimental conditions tested. Reversion of the phenotype was also observed for the cytosolic fragment containing HAMP-GGDEF-EAL (residues 198 -705) domains. The cellular aggregation phenotype was also observed for the ⌬xac2382 knockout strain complemented with the HAMP-GGDEF and HAMP-GGDEF-EAL constructs (Fig. S8). However, cells expressing constructs for only the GGDEF domain, only the EAL domain, or the GGDEF-EAL dual domain did not abolish the high-motility phenotype (Fig. 5A) nor did they result in cellular aggregation (Fig. S8).
The above results suggest that the high-motility phenotype observed in the knockout strain may be due to the absence of DGC activity of the XAC2382 GGDEF domain. To more clearly test the ability of the GGDEF domain to produce c-di-GMP, we transformed E. coli BL21(DE3) cells with the pBRA-XAC2382 HAMP-GGDEF and HAMP-GGDEF-EAL constructs and grew them under inducing conditions (1% arabinose). Liquid chromatography coupled to MS analysis of the extracts from the cell lysates showed that these strains present 30 and 22 times greater c-di-GMP levels, respectively, than E. coli cells carrying the empty pBRA vector (Fig. S9).
To test whether DGC activity is required for XAC2382-dependent reversion of the high motility phenotype, we abolished  Fig. 1B. B, Xac WT strain containing the empty pBRA vector, the ⌬xac2383 containing the empty pBRA vector, and the ⌬xac2383 strain containing the pBRA vector expressing XAC2383 WT and XAC2383 with the 152 STS 154 motif mutated to AAA. Pictures taken after 4 days growth at 30°C on SB medium 0.5% agar plates.

Periplasmic regulation of a diguanylate cyclase
activity by mutating the conserved GGDEF motif to AAAEF (24). Fig. 5A shows that this mutant protein was not able to revert the knockout strain's high sliding motility phenotype. Furthermore, complementation with this mutant did not result in cellular aggregation, different from that observed for complementation using the WT protein (Fig. S8). These results suggest that XAC2382's effects on Xac motility depends on its diguanylate cyclase activity. For this activity to manifest itself, the adjacent HAMP domain is expected to be required to facilitate dimerization and the proper relative orientation of the two GGDEF domains (51,52). HAMP domains contain heptad repeats (abcdefg), where positions a and d are occupied by hydrophobic residues that are required for coiled-coil formation (Fig. S1). The heptad repeats in the XAC2382 HAMP domain correspond to residues 252-266 (Fig. S1). The influence of the coiled-coil structure important for HAMP domainmediated dimerization was therefore evaluated. A mutation at position a of the second repeat was produced (F259E), replacing a hydrophobic residue with a negatively charged residue. Fig. 5A shows that this mutant is unable to revert the knockout strain's phenotype, and Fig. S8 shows that complementation with the F259E mutant also did not result in cellular aggregation, different from that observed for complementation using the WT protein (Fig. S8). These results indicate that a functional HAMP domain is necessary for diguanylate cyclase activity, probably by facilitating dimerization and proper orientation of the GGDEF domains. Recombinant XAC2382 fragments corresponding to the HAMP-GGDEF (residues 198 -446), HAMP-GGDEF-EAL (residues 198 -705), and GGDEF (residues 269 -446) domains were expressed in E. coli and purified in soluble forms. None of these fragments had detectable in vitro DGC activity (see "Experimental procedures"). We speculate that this may be due to a necessity for the protein to be properly inserted into the bacterial membrane, via the second transmembrane helix at the N terminus of the HAMP domain (Fig. S1), for activity.

STS 154 motif is required for XAC2383-dependent sliding motility
The ⌬xac2383 strain has reduced sliding motility, whereas the ⌬xac2383 strain containing a plasmid expressing WT xac2383 presents sliding motility in excess of that observed for the WT strain (Fig. 5B). To test whether this effect requires ligand binding to XAC2383, we mutated the conserved 152 STS 154 motif to 152 AAA 154 . Complementing the knockout strain with this mutant produced a strain whose motility is very similar to that of the ⌬xac2383 strain and much less than that observed for the mutant complemented with WT XAC2383 (Fig. 5B and Fig. S7B). This result suggests that ligand binding to XAC2383 is required for its proper function.

Homologs of the XAC2382 Cache -XAC2383 pair are found in many bacterial species
The protein-protein interaction between the XAC2382 Cache domain and XAC2383 stimulated us to explore the evolutionary relationships between this pair of periplasmic proteins. Homology searches were performed to evaluate the conservation of the XAC2382-XAC2383 pair in other organisms.
First, four iterations using PSI-BLAST (53) identified 4229 annotated microbial genes coding for proteins with Cache domains similar to that of XAC2382 (residues 32-194; inclusion threshold of 10 Ϫ4 ). Of these genes, 871 (ϳ20%) were found in the same orientation as, and in almost all cases downstream to, a gene coding for a XAC2383 homolog (from a total of 14,522 XAC2383 homologs identified by two PSI-BLAST iterations using an inclusion threshold of 10 Ϫ5 ). In gammaproteobacteria, homologs of operons coding a XAC2382 Cache -XAC2383 pair seem to be well distributed in families of the order Xanthomonadales (175 gene pairs) but are also present in individuals of the Alteromonadales (39 pairs) and Vibrionales (309 pairs) orders, including the medically important species Vibrio cholerae and Vibrio parahaemolyticus. The presence of the pair was also observed in some alphaproteobacteria (25 pairs), betaproteobacteria (47 pairs), and deltaproteobacteria (102 pairs) species as well as in some Gram-positive bacteria from the Firmicutes phylum (78 pairs), for example in Bacillus.
Interestingly, the 871 XAC2382 homologs in the above XAC2382 Cache -XAC2383 homolog pairs exhibit a variety of domain architectures, many of which (67%) are significantly different from the Cache-HAMP-GGDEF-EAL architecture of XAC2382. Fig. 6 shows a sample of the architectures observed and reveals that they can be classified into two main groups, based on their cytoplasmic "output" modules. One group of architecture always presents a cytoplasmic GGDEF or GGDEF-EAL module that is separated from the Cache domain by intervening HAMP, PAS, and/or GAF domains (Fig. 6A). In the other major group of architecture, the XAC2382 Cache homolog has a cytoplasmic portion containing a histidine kinase or histidine kinase and response regulator module, again with intervening HAMP and/or PAS domains (Fig. 6B). A relatively small fraction of XAC2382 Cache homologs have output modules that do not fall into one of the above two groups, including, for example, 54 activation with helix-turn-helix (HTH), HD phosphodiesterase, and guanylate cyclase domains (Fig. 6C). This analysis shows that conserved periplasmic input modules made up of a XAC2382 Cache -XAC2383 pair can be coupled to different cytoplasmic output modules, in some cases controlling second messenger levels, and in other cases regulating protein phosphorylation or even gene expression. Fig. 7 shows the phylogeny of XAC2382 Cache homologs found in association with XAC2383-like genes. Two main groups of XAC2382 cytoplasmic output module architectures are associated with mixed branches of the phylogenetic tree. Interestingly, essentially all members of the Xanthomonadaceae family have a single bona fide ortholog of the fulllength XAC2382, i.e. a single gene that belongs to the same monophyletic group characterized by the fusion of a periplasmic XAC2382 Cache domain to a GGDEF cytoplasmic output domain. In contrast, lineages such as the Vibrionaceae and Betaproteobacteria possess a more numerous and diverse set of XAC2382 Cache homologs in their genomes, including genes that code for fusions of XAC2382 Cache to GGDEF or histidine kinase domains and, most notably, some instances where a single polypeptide chain contains both types of output domains (supporting File S1).

Discussion
The co-transcribed xac2383 and xac2382 genes in the phytopathogen X. citri subsp. citri code for a bipartite periplasmic sensor made up of XAC2383 and the XAC2382 Cache domains that control a cytoplasmic output module made up of GGDEF and EAL domains. The bipartite XAC2383-XAC2382 Cache module is quite widespread among bacteria (from gammaproteobacteria to firmicutes). In contrast, the output modules encountered in the XAC2383-XAC2382 Cache phylogeny do not always possess GGDEF and EAL domains and are therefore not always predicted to participate in the control of c-di-GMP levels. In many cases they instead carry histidine kinases or 54 activators, indicative of participation in pathways involving protein phosphorylation cascades and/or the control of gene transcription. In the XAC2383-XAC2382 system, the communication conduit between the sensor and output modules is the HAMP domain that contains an N-terminal transmembrane helix and a C-terminal heptad repeat that may be required for dimerization and DGC activity. This may explain why in vitro DGC and phosphodiesterase activities could not be detected using purified soluble recombinant XAC2383 fragments in the absence of the bacterial membrane.
Our results, summarized below, support a scheme, depicted in Fig. 8, in which XAC2383 regulates c-di-GMP production by the XAC2382 GGDEF domain. Knocking out the xac2382 gene or overexpressing xac2383 results in increased motility, indicative of reduced c-di-GMP levels. When XAC2383 is not interacting with XAC2382, a state simulated by the ⌬xac2383 knockout strain or by the overexpression of the xac2382 gene due to its presence on a multicopy plasmid, sliding motility is decreased, consistent with higher levels of c-di-GMP. Mutations that abolish XAC2382 DGC activity or destroy the XAC2383 ligand-binding site fail to revert the phenotypes of the ⌬xac2382 or ⌬xac2383 knockout strains. These observations are consistent with the hypothesis that in the absence of XAC2383, XAC2382 is able to adopt a catalytically active configuration. Of course, under physiological conditions, in which xac2383 and xac2382 are co-expressed, the nature of the XAC2383-XAC2382 Cache interaction is probably regulated by ligand binding to XAC2383.
One expected feature of the active XAC2382 configuration is that it exhibits C2 symmetry, at least transiently, during the cyclization reaction (54). We do not have any information regarding the orientation of the two proteins in the XAC2382-

Periplasmic regulation of a diguanylate cyclase
XAC2383 complex except to note that induced asymmetry in PAS-like domains (which include Cache domains) upon binding by a periplasmic protein has been proposed by other groups for the LapD-LapG (55) and LuxP-LuxQ systems (56). In both these examples, the periplasmic domains derived from the membrane-bound component (LapD or LuxQ) interact with its partner in a nonsymmetrical manner. Therefore, if XAC2383 binding to XAC2382, or ligand binding to the XAC2383-  et al. (77)), using default options, from a multiple sequence alignment generated by MUSCLE (75). Only the conserved columns of the alignment, as identified by TrimAL (76) using the heuristic similarity statistics, were used. Species names are colored after the domain effector type, as described in the legend in the bottom left corner. Branch color (when not black) refers to the domain architecture of an occasionally encountered third gene, located downstream to the effector GGDEF or histidine kinase XAC2382 homolog and unrelated to the XAC2383 homologs. Such genes also encode signal transduction proteins and contain effector domains (indicated by the colors in the legend). We note some instances of highly supported sibling branches that correspond to proteins with different effector domains but whose Cache domains are more closely to each other than to any other homologs with a similar architecture, thus pointing to novel and independent origins for these architectures.

Periplasmic regulation of a diguanylate cyclase
XAC2382 complex, can stabilize an asymmetric configuration in the XAC2382 dimer, this could inhibit the latter from adopting a catalytically competent conformation (Fig. 8).
Other sensory pathways that more closely resemble the XAC2382-XAC2383 pair in having one membrane-bound GGDEF-EAL component with a periplasmic domain and a second periplasmic binding protein component have been characterized, but the stoichiometry and organization of the complexes were not determined. The P. aeruginosa YfiBNR tripartite signaling system (17,57) consists of YfiR, a periplasmic protein that interacts with YfiN, a transmembrane DGC that regulates exopolysaccharide production. An additional protein, YfiB, is located at the outer membrane and binds to YfiR, thereby preventing its interaction with the diguanylate cyclase. In the V. parahaemolyticus scrABC system, which regulates bacterial capsule production, the phosphodiesterase activity of ScrC (a GGDEF-EAL protein) is regulated by the periplasmic binding protein, ScrB, which in turn acts as a receptor for the quorum-sensing molecule produced by ScrA (58). Finally, in the Agrobacterium tumefaciens DcpA-PruR system, it was proposed that PruR (a pterin-binding protein) activates the phosphodiesterase activity of DcpA (a membrane-bound GGDEF-EAL protein) by interacting with its EAL domain in the cytoplasm (59). We analyzed the PruR primary structure and found an N-terminal signal peptide for periplasmic localization (Fig. S10). We therefore propose that PruR-DcpA might function in a manner similar to XAC2382-XAC2383, in which PruR is in fact a periplasmic binding protein that interacts with the N-terminal periplasmic domain of the transmembrane GGDEF-EAL protein DcpA.
The physiologically relevant ligand of the XAC2383-XAC2382 signal transduction system is still unknown. The positively charged nature of the ligand-binding site, in lieu of the results obtained by soaking XAC2383 crystals with a limited number of different ligands, are both consistent with a preference for phosphate or phosphonate compounds with an additional negatively charged R-group. The fact that we were able to observe electron density for one complete ATP molecule bound to one XAC2383 subunit, as well as the triphosphate moiety of another ATP bound to the second subunit in the asymmetric unit, is particularly intriguing. Extracellular ATP can be derived from a number of sources, including the lysis of eukaryotic hosts or other bacterial species killed by toxins released by macromolecular secretion systems (60 -62). Mammalian cells can also actively release ATP (63). Furthermore, various bacteria, including P. aeruginosa, E. coli, Acinetobacter junii, Klebsiella pneumoniae, Klebsiella oxytoca, Staphylococcus aureus, Enterococcus gallinarum, Enterococcus mundtii, and Enterococcus faecalis have been shown to secrete ATP (64,65). Finally, extracellular ATP, dATP, and CTP have been shown to inhibit P. aeruginosa twitching motility, although no effect was observed for AMP, ADP, GTP, UTP, c-di-GMP, cAMP, and the nonhydrolyzable ATP homolog AMP-PNP (66). The lack of an effect for AMP-PNP led to the suggestion that the twitching motility response involves hydrolysis of ATP as opposed to direct sensing of the ATP molecule by a signal transduction system (66). An alternative interpretation is that the AMP-PNP molecule is not recognized by the ATP receptor due to loss of specific interactions mediated by the H-bondaccepting oxygen atom linking the ␤and ␥-phosphates when substituted by the H-bond donating NH group. Future studies will have to test these possibilities, determine whether and in what manner X. citri responds to extracellular ATP (and other nucleotide triphosphates), and whether the XAC2383-XAC2382 system is involved in this response.
The phylogeny of XAC2382 Cache homologous domains found in association with XAC2383-like genes (highlighted in Fig. 7) supports several independent origins that created new combinations with different cytoplasmic output module architectures. Such events might be the product of gene fusion and domain loss that create new domain combinations from neighboring genes. This hypothesis is supported by the observation that, in some species, the xac2383-xac2382 homologs have a third downstream gene that codes for another output protein (represented by colored branches in Fig. 7). In many instances, these downstream genes code for proteins with effector/output domains (i.e. GGDEF, histidine kinase, 54 activator) that occur as fusions in nearby sibling lineages. This suggests that XAC2382 Cache homologs are recruited to new domain architectures in combination with a XAC2383like gene, i.e. as a stable pair that remains evolutionarily and functionally linked. The periplasmic binding protein XAC2383 has two topologically similar (nonequivalent) lobes and can exist in two conformations whose equilibrium is determined by the binding of a specific ligand (L, unknown). It is not known whether XAC2383 binding to the XAC2382 Cache domain is affected by ligand binding, nor do we know the precise stoichiometry of the XAC2382-XAC2383 complex (here shown as 2:1 for simplicity). In the absence of the ligand (left), the XAC2383 interaction with the XAC2382 Cache domain permits XAC2382 to adopt a symmetric and catalytically active conformation. In the presence of ligand (right), the XAC2383-XAC2382 interaction changes, breaking the symmetry of the XAC2382 dimer, thereby inhibiting the latter's DGC activity.

Site-directed mutagenesis
In-frame deletions of targeted genes were performed using a two-step allelic exchange procedure as described previously (46). Briefly, four primers were designed, two flanking the ends of the gene and two internal complementary primers containing the desired mutation (Table S5). First, two fragments were generated, one from the beginning of the gene to the desired mutation and a second from the desired mutation to the end of the gene. The PCR products were then combined, and another PCR was performed using the primers that flank the ends of the gene. The resulting PCR product was cloned into pET-28a and pBRA for expression in E. coli or Xac cells, respectively. The mutations were confirmed by sequencing.

Recombinant XAC2383 purification and crystallization
The xac2383 gene coding for the protein lacking the signal peptide was cloned (residues 31-309) into the pET3a vector and used to transform BL21 DE3 cells, and colonies were selected with ampicillin. Cells were grown in 2ϫ TY medium to an absorbance of 0.8 at 600 nm. Cells were induced with 400 M IPTG for 4 h at 37°C and harvested by centrifugation. Cells were resuspended in 50 mM Tris, pH 8.0, 25% sucrose, 1% glycerol, 0.03% Triton X-100, 0.03% Tween 20 and purified using a 55-ml SP-Sepharose (GE Healthcare) cation-exchange column. The column was previously equilibrated with 50 mM Tris, pH 8.0, and bound proteins were eluted using a 0 -1 M NaCl gradient in the same buffer. The protein spontaneously crystallized 3 h after elution in the chromatography buffer containing ϳ0.3 M NaCl. The crystals were collected with a pipette, and no further purification step was required. For selenomethionine (Se-Met)-derivatized protein production, cells were grown in 5 ml of 2ϫ TY medium overnight at 37°C, washed, and transferred to 500 ml of minimal medium. After the cells reached an absorbance of 0.8, the following amino acids were added: 60 mg of selenomethionine, 50 mg each of valine, isoleucine, lysine, threonine, and phenylalanine. Recombinant gene expression was then induced by the addition of 400 M IPTG and allowed to grow for 6 h at 37°C.

Diguanylate cyclase activity
XAC2382 constructs, HAMP-GGDEF (residues 198 -446), HAMP-GGDEF-EAL (residues 198 -705) and GGDEF (resi-dues 269 -446), at a concentration of 20 M, were incubated at 30°C in a solution containing 20 mM Tris, pH 8.0, 100 mM NaCl, 2 mM MgCl 2 , and 1 mM GTP in a final volume of 100 l. After 16 h, the reaction was applied in a Resource Q column (GE Healthcare), and the nucleotides were eluted using a gradient 0 -125 mM NaCl in 10 mM solution of HCl. The production of c-di-GMP was evaluated by comparison with c-di-GMP standard purchased from Biolog (Bremen, Germany).

X-ray data collection and structure determination
XAC2383 Se-Met crystals were used to collect a single wavelength anomalous dispersion dataset at the MX2 beamline at the Laboratório Nacional de Luz Synchrotron (LNLS) in Campinas, Brazil, using a wavelength of 0.979 Å at Ϫ180°C under a nitrogen gas flux (Table S1). Data were processed using the HKL2000 suite, and phases were solved using the Autosol software from the Phenix package (67)(68)(69). The initial model was further used for molecular replacement using Phaser (70) to obtain a higher resolution model (1.9 Å) from a native crystal, also collected at the MX2 beamline. The model was improved by cycles of automatic refinement using REFMAC5 and Phenix Refine and manual real-space refinement using Coot (71,72). For soaking experiments, crystals were transferred from the purification tube (where they were initially grown) to Eppendorf tubes containing the same solution plus 2 mM ligands and incubated at 4°C overnight prior to data collection. Data were collected at the Institute of Chemistry of the University of São Paulo using a rotatory copper anode MicroMax 007 (Rigaku) X-ray source at a wavelength of 1.542 Å with crystals maintained at Ϫ180°C under a nitrogen gas flux. The crystals were rotated by 180°and images were collected at 1°intervals. Phases were solved by molecular replacement using Phaser (70). All models were validated using MolProbity (73).

Sliding motility assay
X. citri cells were grown overnight with agitation in liquid 2ϫ TY medium at 30°C. Cells were washed with water, and the absorbance was adjusted to 0.3 at 600 nm. Plastic Petri plates were prepared with 30 ml of SB medium containing 0.5% agar, and 3 l of the cell suspensions were pipetted onto the center of the plates. After drying the drops, the plates were sealed and incubated at 30°C for 4 days on a flat surface (46). Spectinomycin was used at 100 g/ml, and no arabinose was added.

RT-PCR
Xac cells were collected from plates after 2 days of growth at 30°C. Cells were washed with water, and RNA was isolated using TRIzol reagent (Invitrogen). DNase I (ThermoFisher Scientific) was added to avoid genomic DNA contamination. The reverse transcription reaction was performed using 1 g of RNA and RevertAid H Minus First-Strand cDNA synthesis kit (Fermentas) with random hexamer primers, according to the manufacturer's instructions. The cDNA was then used as a template to amplify three regions of the XAC2382-XAC2383 gene locus: (i) the first 186 bp of the xac2382 gene (oligos RTxac2382F RTxac2382R); (ii) the final 193 bp of the xac2383 gene (oligos RTxac2383F RTxac2383R); and (iii) 396 bp overlapping the XAC2382-XAC2383 junction (oligos RTxac2383F and RTxac2382R). The oligos sequences are shown in Table S5. A control reaction in which reverse transcriptase was omitted was also performed to confirm the absence of genomic DNA in the sample.

Production of Xac gene knockout strains
In-frame deletions of targeted genes were performed using a two-step allelic exchange procedure as described previously (46). Briefly, forward (F) and reverse (R) primers were designed to amplify 1000-bp flanking regions upstream (oligos F1 and R1) and downstream (oligos F2 and R2) of the target genes (Table S5). These upstream and downstream 1000-bp fragments were ligated to produce a deleted version of the gene. This ϳ2000-bp fragment was cloned into the pNPTS138 suicide vector, and Xac cells were transformed by electroporation. Transformants were selected by sucrose sensitivity and kanamycin resistance. Cells were then grown without selection for 3 days and then plated and selected for sucrose resistance and kanamycin sensitivity. Deleted alleles and WT alleles were identified by PCR.

XAC2382-XAC2383 interaction
The pET-28a-derived construct for the expression of the XAC2382 Cache domain (residues  with an N-terminal His tag and the pET3a-derived construct for the expression of XAC2383 (residues 31-309) were used to simultaneously transform BL21(DE3) cells. Transformants were selected for kanamycin and ampicillin resistance, and recombinant protein production was induced with 400 M IPTG at an A 600 nm of 0.8 for 4 h at 37°C with agitation. The cells were harvested by centrifugation and resuspended in 50 mM Tris, pH 8.0, 25% sucrose, 1% glycerol, 0.03% Triton X-100, 0.03% Tween 20. The cells were lysed by five passages through a French press and clarified by centrifugation at 31,000 ϫ g. The soluble fraction was applied to a Hi-trap column (GE Healthcare) loaded with Ni 2ϩ and previously equilibrated with 50 mM Tris, pH 8.0, 150 mM NaCl, 10 mM imidazole at 1 ml/min. Bound proteins were eluted with a 10 -500 mM imidazole gradient (10 column volumes). Eluted fractions were analyzed by SDS-PAGE, and those containing both proteins were pooled and applied to a Superdex 200 gel-filtration column (GE 10/300) with a flow rate of 0.5 ml/min coupled to a multiangle light-scattering system (miniDAWN TREOS) and to a refractive index detector (Optilab rEX). A refractive index increment dn/dc ϭ 0.185 ml/g was assumed for the mass calculation.

c-di-GMP quantification in E. coli cells
c-di-GMP extraction was performed as described (74). Briefly, E. coli BL21(DE3) cells carrying pBRA vector constructs expressing XAC2382_HG (residues 198 -446) and XAC2382_ HGE (residues 198 -705) were grown in 2ϫ TY media overnight. An aliquot of each overnight culture was inoculated in 10 ml of fresh 2ϫ TY media and incubated at 37°C. After the bacterial culture reached an absorbance between 0.6 and 0.8, arabinose was added to a final concentration of 1% (w/v), and the culture was incubated for 2 h at 37°C at 200 rpm. One ml of the culture was transferred to a 1.5-ml Eppendorf tube and centrifuged at 9300 ϫ g for 2 min at 4°C. The cellular pellet was washed twice with ice-cold PBS, pH 7.2, and resuspended in 100 l of ice-cold PBS and incubated at 100°C for 5 min. Ice-cold ethanol was added to a final concentration of 65% and vortexed for 15 s. The sample was centrifuged at 9300 ϫ g for 2 min, and the supernatant was transferred to a new microcentrifuge tube. This extraction procedure was repeated two more times, and all the supernatants were pooled together. The insoluble fraction was retained for subsequent determination of protein content by BCA assay. The pooled ethanol-soluble fraction was vacuum-dried after which the pellet was resuspended in 100 l of distilled water. 40 l of these samples were subjected to HPLC-ESI-MS analysis using a C-18 column (Jupiter 5u, Phenomenex, 250 ϫ 4.6 mm) coupled to a C-18 precolumn (Phenomenex, 4 ϫ 3 mm) equilibrated with 2% buffer B (100% methanol) and 98% buffer A (25 mM ammonium formate, pH 7) at 0.6 ml/min. The mixture was eluted using 2% buffer B for 7 min followed by a 2-80% B gradient up to 17 min, followed by 80% B up to 25 min. Under these conditions, c-di-GMP elutes at ϳ15 min. The separated components from this c-di-GMP peak were analyzed with an Amazon Speed ETD (Bruker Daltonics) mass spectrometer equipped with an ESI interface. The operating parameters were set as follows: drying gas flow rate, 12 liters/min; temperature, 300°C; nebulizer, 70 p.s.i.; capillary voltage ϭ 4500 V. The samples were analyzed in positive mode, and mass spectra data were recorded across the m/z range of 50 -1000. Peak areas of the MS spectrum corresponding to the reference mass for c-di-GMP (691 Ϯ 0.2) were recorded and corrected with reference to the total protein content of the particular sample. Three biological replicas were performed, each in triplicate.