Structures of the PelD Cyclic Diguanylate Effector Involved in Pellicle Formation in Pseudomonas aeruginosa PAO1

Background: Bis-(3′–5′)-cyclic dimeric guanosine monophosphate (c-di-GMP) binding to PelD is required for pellicle formation by Pseudomonas aeruginosa. Results: The crystal structures of a cytosolic fragment of PelD show the binding mode of c-di-GMP. Conclusion: PelD has a degenerate active site but binds c-di-GMP through a conserved allosteric site. Significance: PelD represents a novel c-di-GMP effector that has not been structurally characterized before. The second messenger bis-(3′–5′)-cyclic dimeric guanosine monophosphate (c-di-GMP) plays a vital role in the global regulation in bacteria. Here, we describe structural and biochemical characterization of a novel c-di-GMP effector PelD that is critical to the formation of pellicles by Pseudomonas aeruginosa. We present high-resolution structures of a cytosolic fragment of PelD in apo form and its complex with c-di-GMP. The structure contains a bi-domain architecture composed of a GAF domain (commonly found in cyclic nucleotide receptors) and a GGDEF domain (found in c-di-GMP synthesizing enzymes), with the latter binding to one molecule of c-di-GMP. The GGDEF domain has a degenerate active site but a conserved allosteric site (I-site), which we show binds c-di-GMP with a Kd of 0.5 μm. We identified a series of residues that are crucial for c-di-GMP binding, and confirmed the roles of these residues through biochemical characterization of site-specific variants. The structures of PelD represent a novel class of c-di-GMP effector and expand the knowledge of scaffolds that mediate c-di-GMP recognition.


EXPERIMENTAL PROCEDURES
Protein Purification and Crystallization-PelD 158-CT was cloned into pET28 vector and expressed in Escherichia coli Rosetta2. Wild type and mutant proteins were purified using an Ni-NTA affinity column followed by size exclusion chromatography in 20 mM HEPES, pH 7.5, 100 mM KCl. Protein was concentrated to 20 -30 mg/ml and kept at 4°C. Protein was precipitated when stored in this condition but became soluble again at room temperature. The protein concentration was determined by UV absorbance at 280 nm using an calculated extinction coefficient, 11,920 M Ϫ1 cm Ϫ1 . PelD 158-CT crystals were grown by the hanging drop vapor diffusion method at room temperature. Each hanging drop contained 1 l of protein solution and 1 l of mother liquor. The mother liquor conditions are as follows.  4 . The protein concentration was 2-5 mg/ml in all cases. For co-crystallization, 2 mM c-di-GMP was incubated with the protein at room temperature for 30 min prior to crystallization. Crystals were cryoprotected in 15% ethylene glycol before being flash frozen in liquid nitrogen.
Data Collection and Structure Determination-Initial crystallographic studies were carried out using a construct that spanned residues Ile 144 (PelD 144-CT ) through the C terminus. Flash-cooled crystals of PelD 144-CT diffract x-rays beyond a Bragg spacing of 2.5 Å, using an insertion device x-ray beam line (LS-CAT, Sector 21ID, Advanced Photon Source, Argonne, IL). A mercury derivative was prepared by treating crystals with 5 mM ethylmercury bromide for 24 h. Crystallographic phases were determined by single wavelength anomalous diffraction from the mercurial derivative. A 4-fold redundant data set was collected at 100 K to a limiting resolution of 2.6 Å (overall R merge ϭ 9.2%, I/(I) ϭ 1.8 in the highest resolution shell). All diffraction data were integrated and scaled using the HKL2000 package (49). Heavy atom refinement and phase calculation were carried out using PHASER (50) as implemented in the PHENIX software suite (51,52), followed by density modification using DM (53) and cycles of automated building using ARP/wARP (54) and manual rebuilding using XtalView (55). Continuous electron density could only be observed for two of the four molecules that were expected to be in the crystallographic asymmetric unit, and subsequent refinement of all models using REFMAC5 (56) stalled with a free R factor greater than 40%. As subsequent inspection of the model revealed that electron density for the amino terminus could only be observed starting from Asn 158 , all further crystallographic analysis utilized a construct spanning Asn 158 through the C terminus (PelD 158-CT ). Structures of ligand complexes and deletion variants were determined using molecular replacement, as implemented in PHENIX. Ramachandran analysis shows that over 90% of the protein main chain dihedral angles are in the most favored regions, and the rest in generously allowed regions. Data collection, phasing, and refinement statistic are summarized in Table 1.
Analytical Size Exclusion Chromatography-Oligomerization of PelD 158-CT was examined using analytical size exclusion chromatography (Superdex 200 HR 10/30, GE Healthcare) in 20 mM HEPES, pH 7.5, 100 mM KCl. 600 l of sample with a protein concentration of 0.5 mg/ml was applied to the column. For protein complexed with c-di-GMP or cGMP, 70 M c-di-GMP or 10 mM cGMP was included in the sample, and 10 M c-di-GMP or 1 mM cGMP was included in the mobile phase. The molecular weight standards were blue dextran (ϳ2,000,000 Da), ␤-amylase (ϳ200,000 Da), albumin (ϳ66,200 Da), carbonic anhydrase (ϳ29,000 Da), and cytochrome c (ϳ12,400 Da) and purchased from Sigma.
[ 32 P]c-di-GMP Binding Assay-[ 32 P]c-di-GMP binding assay was adopted from the method in Ref. 42 with some major modifications. [ 32 P]c-di-GMP was synthesized by YdeH (57) using [␣-32 P]GTP. It was then incubated with purified wild type PelD 158 or each of the mutants that have N-terminal His tags under the following conditions: 7 nM [ 32 P]c-di-GMP, 30 M protein, 50% Ni-NTA-agarose beads (GE Healthcare), 10 mM Tris, pH 7.5, and 50 mM NaCl. The mixture was incubated at room temperature for 30 min and then transferred to a Spin-X 0.22-m centrifuge tube filter (Costar). The remaining beads in the original tube were washed with 50 l of wash buffer (10 mM Tris pH 7.5, and 50 mM NaCl) and also transferred to the filter. Free [ 32 P]c-di-GMP was removed from the mixture by centrifugation. The flow-through was collected in the 2-ml centrifuge tube that held the filter insert. The filter was washed twice with 300 l of wash buffer and the flow-through was collected in the same centrifuge tube. The filter insert and the flow-through were both counted in the scintillation counter and the fraction of bound [ 32 P]c-di-GMP was calculated by dividing the filter counts with the sum of both counts.
Isothermal Titration Calorimetry-Measurements were carried out on a Nano ITC (TA Instruments, Waters LLC) with protein protomer typically at 30 M in the cell and c-di-GMP at 0.5 mM in the syringe. For D370A mutant, the protein concentration was 100 M and the c-di-GMP concentration was 3 mM. An initial 1-l injection was followed by 24 injections of 2 l each at 240-s intervals. c-di-GMP was synthesized by YdeH using GTP and purified according to the method Zähringer et al. (57), and quantitated based on UV absorbance using a extinction coefficient of 26 mM Ϫ1 cm Ϫ1 at 260 nm (58). Heat of dilution for c-di-GMP was estimated from the last 7-10 injections and subtracted from raw data before fitting the binding isotherm in NanoAnalyze (TA Instruments). Curve fitting was conducted using single independent site binding model. When using the nominal ligand concentration we consistently obtained a stoichiometry of more than 2 c-di-GMP molecules per protomer of PelD. This physically unreasonable value of stoichiometry lead us suspect that the ligand concentration was overestimated due to the presence of UV-absorbing contaminants, a similar situation as reported previously (26). Therefore, the ligand concentration was empirically adjusted by a 2-fold reduction to yield a stoichiometry of approximately 1 when using the single independent site binding model. cGMP or cAMP binding experiments were carried out similarly except that the protein concentration was 100 M and cNMP (Sigma) concentration was 10 mM. Curve fitting was conducted using a single independent site binding model without any concentration adjustment.

RESULTS
Overall Structures of PelD 158-CT -A series of soluble constructs encompassing the cytoplasmic region of PelD were purified and crystallized, and these constructs were started from Met 105 , Leu 123 , Asp 133 , or Ile 144 (PelD 144-CT ) through the C terminus, respectively. Crystals of PelD 144-CT diffracted beyond 2.5-Å resolution and crystallographic phases were determined to 2.6 Å using a mercurial derivative. Clear electron density could only be observed for two of the four molecules in the crystallographic asymmetric unit and while the quality of the experimental map was sufficient to allow building of an initial, near complete model, the structure could not be satisfactorily refined. Subsequent inspection of the model revealed that electron density for the amino terminus could only be observed starting from Asn 158 . A new construct encompassing Asn 158 through the C terminus (PelD 158-CT ) was generated and structural analysis was carried out on crystals of both PelD 158-CT (2.0-Å resolution) and its complex with c-di-GMP in two different crystal forms (form I, 2.0-Å resolution; form II, 2.05-Å resolution) (see Table 1 for data collection and refinement statistics). The structures all occupy the same space group of P2 1 , but with different unit cell dimensions and, consequently, different packing. There is one molecule in the asymmetric unit of the apo structure and in the form I co-crystal structure, whereas the form II co-crystal structure contacts two molecules in the asymmetric unit that appear to be the result of crystal packing and are biologically irrelevant, as illustrated by both the small contact area between protomers (818 Å 2 ) (59) and the solution behavior of PelD 158-CT as a monomer, both in the presence or absence of c-di-GMP ligand ( Fig. 1).
Although the apo structure is well ordered throughout its entirety, the region between Glu 251 and Val 263 is poorly defined in both c-di-GMP complex structures. Hence, description of the overall-fold will be based on the apo structure unless otherwise stated. The structure of PelD 158-CT can be clearly divided into two domains of similar size, an N-terminal domain (composed of Gln 158 through Ser 309 ) and a C-terminal domain (encompassing Asp 318 through Ala 454 ), which are connected by a loop composed of residues Asp 310 -Ala 317 ( Fig. 2A). Binding of the ligand does not induce any local or global changes in the structure (Fig. 2B) and the two structures can be aligned with an average r.m.s. deviation of 1.0 Å over 285 C ␣ atoms.
A DALI search (60) against the Protein Data Bank using the complete structure of PelD 158-CT failed to identify any structure with significant similarities over the entire polypeptide. However, a search using either the N-or C-terminal domains identified several candidates that show significant structural homology to each of the individual domains (supplemental Tables S1 and S2). The PelD N terminus consists of a GAF domain found in various cyclic nucleotide receptors including the cyclic GMP-regulated phosphodiesterases, adenylyl cyclases. The closest structural homolog is the GAF-A of human cyclic GMP (cGMP)-specific 3Ј,5Ј-cyclic phosphodiesterase PDE5A1 (61) with a r.m.s. deviation of 2.7 Å over 128 C ␣ atoms (sequence identity: 15%; PDB code 3MF0). The C terminus of PelD shows an architecture similar to the GGDEF domain found in diguanylate cyclases and the closest homolog is the GGDEF domain of P. aeruginosa c-di-GMP receptor FimX (32), with an r.m.s. deviation of 3.9 Å over 124 aligned C ␣ atoms (sequence identity: 15%; PDB code 3HVA). Consequently, we refer to the N-and C-terminal domains of PelD 158-CT as the GAF and GGDEF domains, respectively.
The GGDEF Domain of PelD 158-CT Shows Significant Differences to Canonical GGDEF Domains-The GGDEF domain of PelD 158-CT is composed of a central four-stranded ␤-sheet, sandwiched between two pairs of ␣-helices (Fig. 3A) and is topologically similar to other GGDEF domains such as those noted above (supplemental Table S1). However, in PelD the GGDEF domain is ϳ20 -25 residues shorter than the canonical equivalents found in these other polypeptides and lacks some of the highly conserved secondary structure features (Fig. 3, A and  B). For example, a comparison of the GGDEF domain of PelD 158-CT with that of PleD reveals that the canonical central ␤-sheet in the PleD GGDEF domain possesses one extra strand (␤i3). Additionally, canonical GGDEF domains contain an additional helix (␣i1), as well as two additional anti-parallel ␤-strands (␤i1 and ␤i2) that are peripheral to the core structure (Fig. 3, B and C). PelD also lacks the catalytically requisite GGDEF sequence characteristic of active DGCs, and instead contains RNDEG (Arg 376 -Gly 380 ) at the equivalent position (Fig. 3C). In active DGCs such as PleD, this motif is located on a loop between two ␤-strands and is involved in binding to GTP and metal ion. However, in PelD, both ␤-strands are extended and thus occlude the GTP binding pocket. Even though the two catalytically requisite metal ion-coordinating residues (Asp 378 and Glu 379 ) are conserved in PelD, they point away from the position of the GTP binding pocket and thus cannot contribute to either substrate binding or catalysis. The absence of both the requisite active site residues, as well as the additional secondary structural elements found in all active DGCs contribute to the lack of a competent active site in the GGDEF domain of PelD.
PelD 158-CT Binds to c-di-GMP Molecule through the I-site-In the PelD 158-CT -c-di-GMP co-crystal structure, one molecule of c-di-GMP molecule is bound to the GGDEF domain through the I-site, in an open, shallow pocket, with only one guanine ring (Gua-1) in the pocket and the other guanine ring (Gua-2) completely exposed to the solvent (Fig. 4A). The two adenine rings are parallel to each other and both vertical to the 12-membered macrocycle formed by two phosphodiester bonds between the two GMP molecules, engaging in a 2-fold symmetrical clip-shaped manner. This binding configuration of c-di-GMP is similar to the one observed in co-crystal structures of PleD (16,19) and WspR (18,20), and the PilZ domain VCA0042 from V. cholera (26), but distinct from that in FimX (32) and LapD (34), where the c-di-GMP ligand adopts an extended conformation and is inserted in a deep binding pocket.
The interactions of PelD 158-CT with c-di-GMP occur mainly through two segments: a loop composed of residues Arg 367 through Asp 370 , and a ␤-loop-␣ segment encompassing Leu 387 to Arg 402 . Within these regions, Arg 367 , Asp 370 , and Arg 402 are responsible for the majority of the interactions with c-di-GMP (Fig. 4B). One of the carboxylate oxygens of Asp 370 forms a hydrogen bond with N-2 of Gua-1, and the second forms a hydrogen bond with N-1 of Gua-1. The guanidinium nitrogen of Arg 402 forms hydrogen bonds with O-6 and N-7 of Gua-1. The guanidinium group of Arg 367 inserts under the Gua-2 ring and forms hydrogen bonds with the diphosphate backbone, and is in electrostatic proximity with both purine rings (Fig. 4B). Importantly, Arg 367 and Asp 370 belong to the conserved RXXD motif and are equivalent to the I-sites in the GGDEF domain of many DGCs like PleD (19) and WspR (18), where the product c-di-GMP binds and allosterically inhibits DGC activity. In both of these DGCs, two c-di-GMP molecules are bound as intercalated dimers (Fig. 4, C and D). The interactions between the RXXD motif and one c-di-GMP are similar to those observed in PelD, but intercalated Gua rings stabilize each other in a manner similar to the interactions observed between Arg 367 and c-di-GMP in the cocrystal structure of PelD (Fig. 4, C and D).
Binding Affinity of PelD Variants for c-di-GMP-To characterize the importance of the residues implicated in c-di-GMP binding in our co-crystal structure, we carried out mutagenesis studies and tested the binding affinity of wild type and variant PelD proteins to c-di-GMP. Residues Arg 367 , Asp 370 , and Arg 402 , which form extensive hydrogen bonds with the ligand, and Tyr 399 , which provides a hydrophobic floor of the binding pocket, were individually mutated to Ala. Gly 395 , which also lines the floor of the binding pocket, was mutated to Pro to introduce steric hindrance within the pocket. All these mutants, along with the wild type protein, were purified with poly-His tag. The protein variants were immobilized on Ni-NTA-agarose beads, and their ability to retain [ 32 P]c-di-GMP was tested (Fig. 5A). R367A, Y399A, and R402A failed to retain c-di-GMP, whereas the binding ability of D370A was reduced by more than 60%. G395P bound c-di-GMP at a similar level to the wild type.
To more accurately quantitatively assess the c-di-GMP binding affinities of D370A and G395P variants, we conducted isothermal calorimetry titration (ITC) analysis (Fig. 5B) of these two mutants as well as on the wild type. Wild type PelD 158-CT binds c-di-GMP at a K d of ϳ0.5 M (see "Experimental Procedures"). These results are comparable with previously reported values (42). The K d for D370A and G395P PelD 158-CT are 28.6 and 3.4 M, a 60-and 7-fold decrease in the affinity, respectively (Fig. 5, C and D). These results confirmed the importance of both Asp 370 and Gly 395 in c-di-GMP binding, which could not be determined directly from the [ 32 P]c-di-GMP binding assay.
As noted, the only significant difference between the structure of unliganded PelD 158-CT and the two c-di-GMP co-crystal structures is the lack of ordered electron density in the region spanning Glu 251 and Val 263 in the latter. In both crystal forms of the ligand bound structure, residues in this loop from the unliganded would clash with a bound c-di-GMP from a symmetry related molecule. To demonstrate that the crystal contacts are not physiologically relevant, we generated a deletion variant in which 10 residues from this loop (Leu 249 -His 258 ) were replaced with a Gly-Gly linker (⌬-loop). The binding affinity of this mutant was tested using both [ 32 P]c-di-GMP binding assay and ITC (Fig. 5, A and E). The ⌬-loop variant showed a binding affinity similar to the wild type in the [ 32 P]c-di-GMP binding assay, and ITC yielded a K d of 0.4 M, which is similar to that of the wild type. These results demonstrate that this loop does not interfere with c-di-GMP binding and the symmetry-related interactions are a consequence of crystal packing. To further corroborate these results, we solved the co-crystal structure of the ⌬-loop-di-c-GMP complex to 1.7-Å resolution and showed that the binding mode of the ligand is identical to that observed in the wild type structure.
GAF Domain of PelD 158-CT -The GAF domain is part of various multidomain proteins that participate in numerous signal transduction processes (62,63). The PelD 158-CT GAF domain displays structural similarity to those of cAMP-or cGMP-specific PDE, including PDE2A (64, 65), 5A(61), 6C (66), and 10A (67), as well as the adenylyl cyclase CyaB2 (68) (supplemental Table S2). The GAF domains in PDEs and adenylyl cyclases have the capacity to bind cyclic nucleotide (cNMP), including cAMP and cGMP, which allosterically regulate the catalytic activity of these enzymes. Several PDEs are shown to bind cGMP or cAMP with nanomolar affinities (10 -200 nM) (69 -  Fig. 1) with that of (B) the active diguanylate cyclase C. vibriodes PleD (shown in gray), with the secondary structural elements that are found in most GGDEF domains but are lacking in PelD 158-CT , shown in magenta. The A-site is colored in cyan and the I-site is colored in green. In the PleD co-crystal structure, two molecules of the c-di-GMP product stack and occupy the autoinhibitory I-site. C, comparison of the secondary structural elements between PelD and PleD near the ligand binding sites. The I-site is highlighted in green, the A-site is colored in cyan, and the additional secondary structure elements in PleD (␣i1, ␤i1, and ␤i2) are colored in magenta. 74) and the cyanobacterial adenylyl cyclases are shown to be activated exclusively by cAMP at submicromolar concentrations (68).
To determine whether the GAF domain of PelD 158-CT is functional in nucleotide binding, we carried out ITC analysis using either cAMP or cGMP. Calorimetric analysis demonstrated that PelD 158-CT does not bind cAMP with any appreciable affinity (data not shown), and only binds cGMP weakly, with a K d of 221.7 M that is several orders of magnitude higher than those reported previously for other GAF domains (Fig. 6A).
Given the low intracellular concentration of cGMP in bacteria (below 100 nM) (75) and the experimentally determined high K d value of PelD, the GAF-like domain of PelD 158-CT likely does not bind cyclic nucleotides.
A structure-based comparison of the PelD 158-CT GAF domain with those of nucleotide-activated GAF domains reveal several important features that may result in the low affinity of PelD 158-CT -GAF for cGMP (Fig. 7). First, in PelD 158-CT -GAF a loop that connects strand ␤2 and helix ␣3 travels through the potential cyclic nucleotide-binding pocket and leaves very little A, stereo view of electron density maps calculated using Fourier coefficients F obs Ϫ F calc with phases derived from the final refined model of the 1.7-Å resolution co-crystal structure of ⌬-loop PelD 158-CT with c-di-GMP. The map was calculated by omitting the coordinates of the cyclic di-GMP prior to one round of crystallographic refinement and is contoured at 2.3 (blue mesh) and 7 (red mesh). A ribbon diagram of the co-crystal structure is superimposed, and the ligand is shown as yellow ball-and-stick and I-site residues Arg 367 and Asp 370 (of the RXXD motif) are shown in green. B, stereo diagram showing the interactions between the cyclic di-GMP (in yellow) and PelD residues (in green) that are critical for ligand binding (as confirmed by biochemical analysis of site-specific variants; see text for further details). Note that Arg 367 wedges between the two guanines, which helps to engage the ligand in a clip-like fashion. C and D, comparison of the interactions between cyclic di-GMP and the I-site in the co-crystal structures of C. vibriodes PelD (C) and P. aeruginosa WspR (D). Note that in both of these co-crystal structures, two molecules of c-di-GMP stack in order and engage their respective protein effectors in a manner analogous to the interactions provided by Arg 367 in PelD. room to accommodate any ligands (Fig. 7A). Second, strand ␤i1 in the GAF domains of PDEs undergoes a significant movement toward the ligand pocket upon cGMP binding (61,71). However, this strand is absent in the PelD 158-CT -GAF domain (Fig.  7, A and B). Last, several residues that are shown to be required for cyclic nucleotide binding are not conserved in the PelD 158-CT -GAF domain (Fig. 7C). For example, in PDE10A, residues Cys 287 , Phe 304 , Asp 305 , Phe 352 , Thr 364 , and Gln 383 are involved in cAMP binding (67), but none of these residues are conserved in the PelD 158-CT -GAF domain. In addition, a conserved NKFDE motif (64,65,76) is largely degenerate in the PelD 158-CT -GAF domain.
Of particular note, the region in the PelD 158-CT -GAF domain that is disordered in the c-di-GMP co-crystal structure (Glu 251 -Val 263 ) corresponds to a portion of the cyclic nucleotide binding pocket in the PDEs. The structure of cAMP-bound PDE10A GAF-B showed that the cAMP molecule is deeply buried in a pocket that uses the antiparallel ␤-sheet as the floor and a short helix (␣i1, Asn 353 -Gly 361 in PDE10A) as the roof (67) (Fig. 7B). To test whether the region between Glu 251 -Val 263 plays a role in the inability of PelD 158-CT to bind cyclic nucleotides, we carried out ITC analysis on the ⌬-loop mutant with cGMP (Fig. 6B). The result showed that deleting this region led to only a modest increase of the K d (415.8 M). Hence, this loop region does not play any role in the inability of PelD 158-CT GAF to bind ligands.
We also examined the possibility that PelD 158-CT -GAF may mediate homodimerization, a typical feature of GAF domains (74). Analytical size exclusion chromatographic analysis failed to identify any changes in the elution profile of PelD 158-CT in the presence of a high concentration of cGMP (Fig. 1). Although the dimerization interface and the domain orientation differ for many GAF domains, they all involve the two or three helices (␣1, 2, and 4) located on the opposite side of the ligand binding pocket (64,65,67,77). However, the corresponding helices in the PelD 158-CT -GAF domain, consisting of residues Gln 158 -Glu 174 (␣1), Leu 231 -Gly 240 (␣2), and Glu 290 -Leu 307 (␣4), are oriented toward the GGDEF domain and partly buried in the domain interface (Fig. 2). Thus, the conformation of PelD observed in our structures is not competent to mediate dimerization, consistent with the results from our analytical size exclusion data (Fig. 1).

DISCUSSION
The GGDEF domain of PelD 158-CT represents a novel class of c-di-GMP receptor that binds c-di-GMP through a conserved I-site. GGDEF domains constitute the active sites of DGCs, such as PleD from C. vibriodes (19), WspR from P. aeruginosa (18), and XCC4471 from X. campestris (78), but are also found in enzymatically inactive c-di-GMP receptors, such as FimX (32). The GGDEF domains of PleD and WspR both have an active GGEEF motif (A-site) that binds to substrate and catalyzes the cyclization of two molecules of GTP into one molecule of cyclic di-GMP. Both PleD and WspR contain a conserved RXXD motif (I-site) located amino-terminal to the A-site, and this I-site binds to c-di-GMP and accounts for allosteric product inhibition (18,19). In contrast, XCC4471 has a conserved A-site but a degenerate I-site, and (competitive) product inhibition is achieved by c-di-GMP binding directly to the A-site (78). Last, the GGDEF domain of FimX is degenerate at both the A-site and I-site, and lacks both DGC activity and c-di-GMPbinding capability (32). A similar feature is also observed in another c-di-GMP receptor LapD from P. fluorescens (34). Remarkably, distinct from all these domains, the GGDEF domain of PelD 158-CT has a degenerate A-site ( 375 RNDEG) but a conserved I-site ( 367 RGLD) (Fig. 8). This combination is also conserved in some other potential c-di-GMP receptors, for example, CdgG from V. cholera (43) and PopA from C. vibriodes (6) (Fig. 8). Thus the PelD 158-CT -GGDEF domain is representative of the class of c-di-GMP receptors that contain a degenerate GGDEF active site but a conserved I-site that can engage c-di-GMP.
Our structures showed that PelD binds to one molecule of c-di-GMP through the conserved I-site in the GGDEF domain. Mutations of the conserved residues in this site, Arg 367 and Asp 370 , abolished the binding of PelD to c-di-GMP. Importantly, it was shown in previous in vivo studies (42) that such mutants failed to restore the ability of P. aeruginosa PA14 to form pellicles and bind Congo red. Combining these functional studies and our structural data, a strong correlation may be concluded between the binding of PelD to c-di-GMP and the formation of pellicles in vivo.
Although the GAF domain of PelD 158-CT is topologically similar to canonical GAF domains that can bind to cNMP molecules, our calorimetric studies show that PelD 158-CT does not bind either cAMP or cGMP with affinities that are physiologically meaningful. A number of GAF-containing proteins have tandem GAF domains and the two GAF domains are proposed to have distinct functions (74). For example, only one of the two GAF domains in PDE2A (GAF-B) binds a cyclic nucleotide, whereas the second domain (GAF-A) lacks ligand binding ability but is proposed to function as a dimerization locus (64,65). cGMP binding to the GAF-B domain has been shown to allosterically increase the PDE activity at the catalytic domain (79). A second example is that cGMP binding to the GAF-A domain of PDE5A stimulates phosphorylation through a cGMP-dependent protein kinase, which in turn increases the catalytic activity of PDE5A and cGMP binding affinity of its GAF-A domain (80 -82).
Our results showed that the GAF domain of PelD has only very weak affinity to cyclic nucleotides, thus it is unlikely regulated by these molecules. On the other hand, the location of the PelD 158-CT GAF domain between the C-terminal c-di-GMPbinding GGDEF domain and an N-terminal transmembrane region implies that it might serve as a c-di-GMP signal relay between these two domains. Sequence-based genome analysis demonstrates that the association of a GAF domain with a GGDEF domain occurs in many diguanylate cyclases (3). However, only a few of these proteins have been biochemically characterized. The diguanylate cyclases DgcA from Rhodobacter sphaeroides (83), and MSDGC-1 from Mycobacterium smegmatis (84) have been shown to largely lose their DGC activity after partial or complete removal of the GAF domain. However, the rationale behind this remains elusive, as a ligand-dependent function of the GAF domains in these proteins has not yet been   Fig. 1) with the (B) human PDE10A-cyclic GMP co-crystal structure (shown in brown with ligand colored in green ball-and-stick), with the secondary structural elements that are found in most GAF domains but are lacking in PelD 158-CT , shown in magenta. The binding pocket for cyclic GMP is partially occluded in PelD 158-CT by the loop that joins ␤2 and ␣3. C, comparison of secondary structural elements between PelD and PDE10A near the cyclic GMP binding sites. The I-site is highlighted in green, the A-site is colored in cyan, and the additional secondary structure elements in PleD (␣i1 and ␤i1) are colored in magenta. Residues in PDE10A that involved in direct contact with the cyclic nucleotide are highlighted in yellow and the NKFDE motif is highlighted in green.
established. In the case of DgcA, neither cAMP nor cGMP stimulated DGC activity (83).
To date, all PDEs that have been structurally characterized form dimers, although the functional significance of PDE dimerization remains unclear. The regulatory N-terminal region of these proteins, including the tandem GAF domains, are suggested to provide dimerization contacts as the isolated catalytic domains from PDE2A (85), PDE5A (86,87), and PDE10A are monomeric (88). Unlike the GAF domains found in PDEs, the PelD 158-CT -GAF domain does not form a dimer, neither in the crystal nor in solution. The helices that provide the dimerization interface in other GAF domains are partially buried between the GAF and GGDEF domains of PelD 158-CT , and dimerization through the GAF domain would require significant conformational movements.
In most of the characterized receptors, c-di-GMP binding usually induces large conformational changes, and these changes have been proposed to propagate signal transduction. For example, binding of c-di-GMP to the PilZ domain of VCA0042 from V. cholerae induces a 123 o rotation, resulting in a more compact overall structure and drastically different accessible protein surface, which is proposed to interact directly with downstream effectors (26). Another example is a membrane-bound effector LapD from P. fluorescens in which autoinhibitory interactions between the degenerate EAL and HAMP domains are relieved upon c-di-GMP binding to the EAL domain, promoting the interaction of the HAMP domain with other effectors (33,34). Last, c-di-GMP binding to the I-site of two DGCs, PleD (16,19) and WspR (18,20), allosterically inhibits the enzyme activity by forcing the protein dimer into a nonproductive conformation.
Unlike these other c-di-GMP binding targets, c-di-GMP binding to PelD 158-CT does not result in any structural changes, either globally or local to the ligand-binding site. The relative orientation of the GAF and GGDEF domains are retained upon c-di-GMP binding, and PelD 158-CT remains as a monomer regardless of the absence and presence of the ligand. Similarly, the c-di-GMP binding EAL domain of FimX from P. aeruginosa is monomeric, and does not undergo any structural changes upon ligand binding (32). It was proposed that c-di-GMP binding might facilitate complex formation between the EAL domain and an unidentified binding partner (32). A similarly plausible model can also be considered for PelD, and this is further strengthened by the fact that c-di-GMP in the complex structure is significantly surface exposed. Binding of the ligand might either interrupt the interaction between the PelD GGDEF domain and an unidentified binding partner, or bridge the GGDEF domain to its binding partner. An example of protein-protein interactions triggered by c-di-GMP was reported recently on a PilZ domain-containing protein (36).
There are a large number of GGDEF domain-containing proteins in bacteria, which have diverse functions and thus account in part for the complexity of c-di-GMP signaling. Although studies have begun to investigate the hierarchy and specificity of GGDEF domain-mediated signaling pathways (89,90), the knowledge of protein binding partners of GGDEF domains is still limited. The binding or dissociation of a protein partner to the GGDEF domain might cause a conformational change in the N-terminal transmembrane domain of PelD, using the GAF domain as a signal relay. A BLAST search (91) reveals that the N-terminal transmembrane domain of PelD (N terminus, Ala 104 ) is categorized as a domain of unknown function under the DUF4118 superfamily (Pfam 13493). Domains in this superfamily exist in a wide variety of bacterial signaling proteins, and these may play a role in signal transduction. Along with two other proteins in the Pel pathway, PelE and PelG (44), which are predicted to contain transmembrane helices, PelD could facilitate the export of carbohydrate-containing substances. Such export processes might require interactions between these transmembrane proteins and PelC, an outer-membrane lipoprotein that has been shown to facilitate exopolysaccharide transport (47). We are continuing additional biochemical, microbiological, and structural studies aimed at identifying the effectors downstream of PelD that mediate pellicle formation.