Crystal Structure of an EAL Domain in Complex with Reaction Product 5′-pGpG

FimX is a large multidomain protein containing an EAL domain and involved in twitching motility in Pseudomonas aeruginosa. We present here two crystallographic structures of the EAL domain of FimX (residues 438–686): one of the apo form and the other of a complex with 5′-pGpG, the reaction product of the hydrolysis of c-di-GMP. In both crystal forms, the EAL domains form a dimer delimiting a large cavity encompassing the catalytic pockets. The ligand is trapped in this cavity by its sugar phosphate moiety. We confirmed by NMR that the guanine bases are not involved in the interaction in solution. We solved here the first structure of an EAL domain bound to the reaction product 5′-pGpG. Though isolated FimX EAL domain has a very low catalytic activity, which would not be significant compared to other catalytic EAL domains, the structure with the product of the reaction can provides some hints in the mechanism of hydrolysis of the c-di-GMP by EAL domains.


Introduction
Cyclic di-GMP (bis-(39-59) cyclic dimeric guanosine monophosphate, c-di-GMP) was first considered as an allosteric regulator of cellulose production in Gluconoacetobacter xylinus [1,2]. Further studies revealed that c-di-GMP is a second messenger exclusively present in Eubacteria and involved in the regulation of various essential biological processes like twitching motility, swarming, quorum sensing, virulence and surface adhesion [3,4,5]. In several bacteria, the switch between planktonic and biofilm formation is controlled by the concentration of c-di-GMP. The increase in intracellular c-di-GMP concentration leads to biofilm formation while its decrease enhances bacterial mobility and virulence [6]. The targets of c-di-GMP are either proteins with a PilZ domain [7] or RNA riboswitches [8].
Three protein domains, GGDEF, EAL, and HD-GYP, have been reported to be involved in c-di-GMP metabolism. They are named after the conserved amino-acids motifs in these domains [9]: GGDEF have diguanylate cyclase (DGC) activity and manage the synthesis of c-di-GMP. EAL domains exhibit phosphodiesterase A (PDE-A) activity and are responsible for the conversion of cdi-GMP into 59-phosphoguanylyl-(39-59)-guanosine (59-pGpG). HD-GYP domains also carry a phosphodiesterase activity, which can hydrolyze c-di-GMP into 59-pGpG and further into two GMP molecules. Though it has been evidenced in vitro that EAL domains can pursue the hydrolysis of 59-pGpG into two GMP molecules [10], this second reaction is however very slow compared to the reaction carried out by HD-GYP domains, which raises the question as to its occurrence in vivo [11]. Although most recent studies assume that DGC and PDE activities monitor the intra-cellular c-di-GMP concentration, the understanding of the complex molecular mechanisms regulating these activities and the signal transduction by c-di-GMP is very poor [12,13].
Bacteria using c-di-GMP signaling networks produce several proteins containing an EAL domain. EAL domains have been reported to be catalytically active in presence of either Mg 2+ or Mn 2+ . Conversely, the PDE-A activity is inhibited in presence of calcium ions. The comparison of active EAL domains and of degenerated non-catalytic EAL domains has allowed to identify 14 polar residues important for the PDE activity. Based on the structure of the EAL domain containing protein YkuI from Bacillus subtilis and site directed mutagenesis on the EAL domain containing protein RocR from P. aeruginosa, a mechanism explaining the phosphodiesterase activity by the EAL domains has been proposed [14]. It is an Mg 2+ base-assisted catalysis, but the number of Mg 2+ (one, two or three) necessary for catalysis remains elusive [15]. In a more recent study [16], 13 EAL domains from various organisms were purified and their activity was measured. Ten conserved residues important for the PDE-A activity were identified, all these residues being already defined as important in previous studies [14]. This, together with structural comparisons, leads the authors to conclude that the hydrolysis of c-di-GMP involves two Mg 2+ ions.
FimX (PA4959) from P. aeruginosa is a 76,000 Da protein identified in mutants lacking twitching motility in response to environmental factors [17]. Twitching motility is a form of surface translocation mediated by extension, tethering and retraction of type IV pili, which are polar structures also required for biofilm formation and adherence. This lack of motility is reminiscent of an increase of c-di-GMP cellular concentration, and thus FimX is expected to down-regulate c-di-GMP hydrolysis. A FimX homolog from Xanthomonas axonopodis, a citrus pathogen, has been reported to interact with PilZ, a protein involved in type IV pilus biogenesis [18]. FimX from P. aeruginosa contains a CheY-like response regulator domain (REC), a PAS-PAC domain commonly involved in environmental sensing, and one GGDEF and one EAL domain. FimX is located at one pole of the bacterial cell probably by its Nterminal REC domain. The full-length protein possesses a PDE-A activity stimulated by Mg 2+ and Mn 2+ but strongly inhibited by Ca 2+ , consistently with what has been described for other proteins with an EAL domain [19]. The FimX GGDEF domain is degenerated and does not display any diguanylate cyclase activity, but may stimulate the PDE-A activity in presence of GTP. Such allosteric regulation of an active EAL domain by a degenerated and inactive GGDEF recalls that described for CC3396, another dual GGDEF-EAL domain protein from Caulobacter crescentus [20]. The isolated FimX EAL domain doesn't show significant catalytic activity [19,21]. A comparative bioinformatics approach showed that FimX-EAL domain lacks several essential residues for catalysis compared to active EAL domains from other proteins [14]. The crystallographic structures of the FimX EAL and GGDEF domains have been solved [21] and show that FimX-EAL can bind c-di-GMP despite the lack of some amino acids predicted to be important for the catalysis. However, neither DGC nor PDE activity was detected for the (full-length) FimX protein used in this study, which is not consistent with what was previously shown [19]. In the complex, c-di-GMP is bound through interactions with 5 of the 10 amino acids identified as essential (Tyr673, Glu654, Asp507, Arg479 and Leu477). Interestingly, the stoechiometry of FimX-EAL in the crystals depends on the presence the ligand. The apo state of FimX-EAL and the dual FimX-GGDEF-EAL crystallize as a dimer but FimX-EAL in complex with c-di-GMP crystallizes as a monomer. FimX-EAL monomers in the apo and complex states differ by few conformational changes. SAXS and analytical centrifugation data showed that FimX would dimerize through its N-terminal REC domain in solution [21].
Here we solved the crystallographic structure of FimX-EAL bound to 59-pGpG and show that this domain doesn't display major structural rearrangements whether bound to 59-pGpG, c-di-GMP or in apo form. We also probe its affinity for c-di-GMP and 59-pGpG in solution by NMR spectroscopy and determined which protons of the ligand are recognized by the protein.

Overall Structures
The FimX-EAL domain (residues 438-686) was crystallized alone and in presence of the substrate c-di-GMP in space groups P2 1 2 1 2 1 and P4 3 2 1 2 respectively. Crystals diffracted to 2.3 Å and 2 Å resolution respectively ( Table 1). The structures were solved using molecular replacement using the EAL domain model tdEAL from Thiobacillus denitrificans (PDB code 2R6O) as search model. Both structures contain two molecules per asymmetric unit related by a non-crystallographic two-fold axis ( Figure 1).
Residual density in the active site of both EAL domains of the asymmetric unit of the P4 3 2 1 2 crystal form leads us to place a molecule in the active site (Figure 2 A). Density maps were improved by simulated annealing omit maps. First attempts were carried out with c-di-GMP substrate but yielded very poor fit. Instead, the accommodation of 59-pGpG, the reaction product, in the cavity was much better and acceptable. Density is clearly defined for 59-pGpG phosphates and sugars but not for the two guanine bases (Figure 2 B). The structures of FimX-EAL crystallized with 59-pGpG (P4 3 2 1 2 space group) will be further referred as the complex form, and that with no density in the active site (P2 1 2 1 2 1 space group) as the apo form.
FimX-EAL displays a central barrel composed of 8 b-strands surrounded by 7 a-helices, with a ab(b/a) 6 b secondary structure succession ( Figure 1). This secondary structure topology has already been reported for EAL domains like YkuI [15] and is a variant of the triosephosphate isomerase (TIM) barrel fold. All the structure of EAL domains solved so far share this characteristic TIM-like barrel fold. Superposition of all known EAL domain structures was performed chain by chain. The core barrel of these two EAL domains is invariable. Deviation contributions arise from helices and loops. Interestingly, loop-8 is the only loop highly superimposable. The closest structure to ours is FimX-EAL solved previously (PDB code 3HV8) [21], then tdEAL (PDB code 2R6O) with a r.m.s deviation of 2.37 Å . B. subtilis YkuI (chain A) and our structure of P. aeruginosa FimX-EAL bound to 59-pGpG (chain B) are the most divergent structures, with a r.m.s. deviation rising to 2.59 Å over 220 superimposed Ca positions.

Apo FimX-EAL and FimX-EAL Bound to 59-pGpG both Crystallize as Homodimers
FimX-EAL is monomeric in solution whether it is free or bound to c-di-GMP [21]. However in both our structures, two molecules are present per asymmetric units and form a homodimer (Figure 1 B). The surface of interaction is formed by the N-and C-terminal extremities and external a-helices and loops ( Figure 1 and 3). In both cases, the a-helices and the loops responsible for the inter-monomer interactions are those surrounding the core barrel. In the apo form of the EAL domain, the secondary structures responsible for the interactions are a-helices 1, 2, 6, 7 and loops b2-a2, b7-a7 (previously described as loop 7 in TIMbarrel related literature). In the structure of FimX-EAL bound to 59-pGpG, the interactions are mediated by contacts between ahelices 1, 2, 6, 7 and loops a1-b1, b5-a5. The surface of interaction measured by PISA is around 1410 Å 2 with a score of 0.321 for the complex form and is 1430 Å 2 with a score of 0.401 for the apo form. In both dimers, the computed scores for the interaction are relatively weak but there is however a large interface area due to hydrogen bonds and hydrophobic contacts. Similar homodimer was already described in the asymmetric unit of crystals in space group P4 1 2 1 2 and in the crystal lattice with a symmetry related molecule in space group P6 4 22 [21]. Thus, this homodimer exists in four crystal forms with different crystal packing.
Dimerization of FimX-EAL results in formation of a cavity with a volume of 4876 Å 3 and 3609 Å 3 for the complex and the apo forms respectively (Figure 3). In the complex form, two 59-pGpG molecules are inside the cavity. They are in the same conformation and separated by a distance of 7.39 Å . They do not interact with each other and they are too far away from each other to allow bond formation or any reaction to proceed. It is interesting to note that the cavity is open to the solvent by two channels of 9.76 Å and 13.34 Å diameters (Figure 3 A). Electrostatic surface shows that the edges of these channels are negatively charged (Figure 3 B). The small diameters and the charges of the channels prevent 59-pGpG molecules from escaping the cavity, trapping them inside the dimer like a bird in a cage.
Such a cage was previously described for a dimeric form of apo FimX-EAL [21]. The authors mentioned that if each EAL domain binds a c-di-GMP ligand, clashes would occur between the two cdi-GMP molecules. Our structure of FimX-EAL in complex with 59-pGpG confirms that for the same reasons, FimX-EAL bound to c-di-GMP cannot exist as a dimer.

Ligand Binding Site
The ligand electron density map is well defined for phosphate and sugars but is incomplete for the two guanines ( Figure 2). Several interactions are observed between the protein and the ribose-phosphate backbone by hydrogen bonds and hydrophobic contacts, showing that the ribose-phosphate is tightly trapped while the two bases orientation remains flexible. Residues involved in 59-pGpG and Mg 2+ binding mainly arise from the b-strands forming the barrel. 59-pGpG binding residues are mainly located on strands b4, b5, b6 and loops 7 and 8, and magnesium-coordinating residues are on strands b1 and b2. The EAL domains core barrel is thus essential to ligand binding and consequently probably to the catalytic mechanism. Although our structure of FimX-EAL complexed with 59-pGpG is highly superposable to that of FimX-EAL in complex with cdi-GMP (PDB code 3HV8), the overall position and conformation of the ligand in the active site are different in the two structures ( Figure 4). However noteworthy, the non-hydrolyzed phosphate of 59-pGpG (P2) lies in the same position as that of the c-di-GMP hydrolysable phosphate and share similar interactions with some amino acids, conversely to the hydrolyzed phosphate that is rocked 7.96 Å away from the position it occupies in c-di-GMP ( Figure 4). To accommodate the differences of conformations, the residues involved in ligand binding are also different. In FimX-EAL in complex with c-di-GMP, Phe652 (loop 7) stacks on the guanine base. In FimX-EAL in complex with 59-pGpG, it has rotated to make hydrogen bonds with the 59-hydrolyzed phosphate ( Figure 5). Gln596 (b5) has moved by 1.60 Å in order to allow hydrogen bonding between the amid group and 39 hydroxyl of the ribose G2 of 59-pGpG. Lys616 (b6) binds to the 29 hydroxyl of the G1 ribose, Gln565 (b4) and Ser595 (b5) respectively to the 29 hydroxyl and the 39 hydroxyl of the ribose G2. Hydrophobic  contacts concern principally base binding, with the exception of His533 (b3) that lies above to the second sugar ( Figure 2).

Magnesium Coordination
In our structure with the ligand, only one magnesium ion is present in the active site. The magnesium is coordinated in a tetrahedral state by Gln461 (b1) and Glu475 (b2) from the EAL motif (turned EVL in FimX), a water molecule and the 59phosphate from 59-pGpG ( Figure 2). The magnesium ion is essential for the PDE-A activity [14,19]. A glutamate is involved in magnesium coordination in all known EAL domain structures, with the exception of FimX-EAL in complex with c-di-GMP (PDB code 3HV8) for which no magnesium ion at all is present in the structure. In our structure, the Mg 2+ ion is shifted 5.9 Å apart from the position usually observed, but is still coordinated by Glu475 and one oxygen of the hydrolyzed phosphate.

Phosphodiesterase Activity of FimX
In order to assess the phosphodiesterase activity of FimX and thus to validate the presence of 59-pGpG in our crystallographic structure, we performed enzymatic assays using Bis (p-nitrophenyl) phosphate (BpNPP) as substrate in Mn 2+ buffer. This compound was proved to be an alternative to c-di-GMP to measure PDE-A activities and was used with several EAL domains like HmsP [22], FimX [19], TBD1265 and SFV_3559 [16]. Though BpNPP isn't the natural substrate of EAL domains, previous studies proved that a millimolar K M for BpNPP corresponds to a nanomolar K M for EAL domains for the natural substrate c-di-GMP. However the turnover kcat may display strong variation compared to the turnover for the natural substrate c-di-GMP [16]. The time course of the release of p-nitrophenol was measured for several hours as a function of the BpNPP concentration (1 mM, 5 mM, 10 mM, 15 mM and 20 mM) and initial rates were determined for each concentration. Full-length FimX displays a K M of 1 mM and a kcat of 6.4.10 24 s 21 (Figure 6 A). K M measured for full-length FimX is lower to the one measured for TBD1265 and SFV_3559 (20 mM and 14 mM respectively). In contrary, kcat is weaker than the one measured for TBD1265 and SFV_3559 EAL domains (0.06 s 21 and 8 s 21 respectively). The catalytic efficiency kcat/K M of full-length-FimX is 10-fold lower than the one of TBD1265. These experiments prove that FimX full-length displays a strong affinity for its substrates but a very weak turnover considering its K M and kcat. FimX has a low activity with BpNPP as substrate.
This low phosphodiesterase activity is in agreement with what was already described [19].
We performed a relative activity assay over nine hours in order to determine whether isolated GGDEF-EAL and EAL domains of FimX also display enzymatic activities. Our results indicate that GGDEF-EAL domains activity is five-fold weaker than FimX fulllength activity. Phosphodiesterase activity of EAL domain is around three-fold weaker than GGDEF-EAL domain activity and fifteenth-fold weaker than FimX full-length activity (Figure 6 B). The observation of a very weak phosphodiesterase activity validates the formation of 59-pGpG during the time course of FimX-EAL crystallization.

c-di-GMP and 59-pGpG both Bind to FimX-EAL in Solution
We wanted to probe c-di-GMP and 59-pGpG binding to the FimX-EAL domain in solution and identify which part of the ligand is involved in the interaction. We have thus carried out a series of reverse NOE-pumping (RNP) NMR experiments [23] in D 2 O (H 2 O contaminated) buffer at 308 K and pH 6.5, for  which only small amounts of unlabeled protein and an excess of ligand are required (Figure 7). The H 2 O contamination allows potential observation of magnetization transfers to sugar hydroxyl protons. No magnesium was added to the NMR buffer to prevent any enzymatic reaction or degradation of the ligands during spectra recording. The RNP experiment contains two blocks, a T2 relaxation filter selecting only the spin systems of the free ligand and a cross-polarization module allowing intermolecular NOE magnetization transfer between the protons of the bound-ligand and those of the protein. Two spectra are recorded sequentially with inverted modules order and subtracted. In the end, the difference spectrum only displays the peaks arising from protons of the ligand that are within a few Å from those of the protein and thus susceptible to interact with the protein (Figures 7 D and G). As controls, in absence of protein, no signal is observed (Figures 7  B and F). The NOE mixing time can be varied to probe short to larger intermolecular distances. When c-di-GMP is mixed FimX-EAL, the RNP experiment discloses two transfers, one to a G H8 (ppm) and the other to a G H1' (ppm) (Figure 7 D). This shows that, the recognition of c-di-GMP involves at least one base and one sugar of the ligand. This is consistent with the crystallographic structure, which discloses that one guanine is stacked on a phenylalanine of the EAL domain [21]. Conversely, in the same experimental conditions (NOE mixing time of 400 ms), no transfers are observed when 59-pGpG is mixed with FimX-EAL. However, longer mixing times (here 700 ms) allow observation of a strong transfer to one OH2' sugar hydroxyl (negative signal at 5.7 ppm) and of a weaker transfer to H59/50 (positive peaks at about 4.5 ppm), the observation of the latter transfer is slightly impeded by traces of the buffer used by Dharmacon to purify 59-pGpG that we were not able to remove without loss of considerable amounts of the product). However, no transfer to G H8 and H19 protons could be recorded, even when temperature was lowered to 298 K (not shown) to increase the affinity. Our RNP experiment shows that in solution and at relatively high temperature, even in the absence of magnesium, FimX-EAL can bind to the 59-pGpG albeit much weaker than to c-di-GMP. The fact that no transfers to GH8 and H19 are observed, are consistently with our crystallographic results which show that none of the guanine aromatic bases are involved in the binding. The specific and reproducible absence of some magnetization transfers also rules out magnetization transfers to non-specifically bounded 59-pGpG molecules.

FimX-EAL is Able to Crystallize as Homodimer
We have crystallized the FimX-EAL in its apo form and in complex with 59-pGpG as a homodimer. It was previously shown that FimX-EAL bound to c-di-GMP crystallized as a monomer because dimerization would cause a large number of ligand-ligand clashes in the binding pocket. Dimerization of apo FimX-EAL was thought to prevent c-di-GMP binding [21]. We observed here that FimX-EAL crystallizes in two new crystal forms as a dimer with a large interface area and the involvement of several polar residues of each monomer. The propensity of FimX-EAL domains to dimerize is emphasized by the observation that they can crystallize as homodimers in different space groups even without ligand. Though the large contact area between the two monomers and the number of residues involved in electrostatic bonding at the interface are compatible with the formation of FimX-EAL dimers in solution, this dimer was never observed in vitro. Gel filtration and SAXS experiments on FimX-EAL with or without 59-pGpG could not reveal the formation of a dimer in solution (data not shown). The dimerization interface involves mainly N-and Cterminal extremities residues, and in less extent helices a2, a6 and a7 helices. The N-and C-terminal extremities may be more flexible in solution than in the crystal context. Structural rearrangements of the flexible N-and C-terminus may be necessary to form the oligomer and crystal packing could stabilize these conformations. Because the residues involved in the interface are mostly polar, burying polar residues during dimerization does not have a large contribution to the entropic effect and will not stabilize the dimer.
The oligomeric state of EAL domains and full-length proteins containing an EAL domain is very variable and depends on the ligand bound to the domain. Several EAL domains, like YkuI, BlrP1, YcgF and tdEAL have been reported to also form EAL-EAL mediated homodimers in crystals when they are bound to cdi-GMP and for some of them in the apo form. The existence of homodimeric forms was confirmed in solution by analytical ultracentrifugation, size exclusion chromatography and dynamic light scattering [15,16,24]. RocR from P. aeruginosa is tetrameric in solution [25]. Former X-ray scattering experiments also show that both apo FimX-EAL and apo FimX-EAL-GGDEF are monomeric in solution. Only full-length apo FimX is able to dimerize, most probably through its N-terminal domain [21]. These results of X-ray scattering experiments were the same whether FimX or its constructs were complexed or not with c-di-GMP. As FimX is a long and multi-domain protein, we cannot exclude the possibility of several interfaces promoting dimerization, outside of the Nterminal domains suggested by the SAXS experiments on the apo protein [21].

Product Binding
FimX has a strong affinity for c-di-GMP according to the Kd (about 120 nM) computed from ITC experiments [21,26]. Our RNP NMR experiments confirm that FimX-EAL binds c-di-GMP in solution, and, in much less extent, 59-pGpG, the product of the hydrolysis of c-di-GMP. The 59-pGpG was formed during crystallization by either the very weak phosphodiesterase activity of FimX-EAL or by spontaneous hydrolysis of c-di-GMP. However, the product formed was trapped in the FimX-EAL dimer formed in the crystals. The crystallographic bias allowed us to obtain the structure of FimX-EAL in complex with reaction product.
Interestingly, in the complexes, the 59-pGpG conformation is very different from that of the c-di-GMP preventing clashes between the two ligands inside the dimeric catalytic pocket. The only position to be conserved is that on the non-hydrolyzed phosphate in 59-pGpG which occupies the positions of the target phosphate of c-di-GMP in the hydrolysis reaction. 59-pGpG is thus in position for a potential second round of hydrolysis. Neither the guanines nor the sugars of 59-pGpG are superimposed on their counterpart in c-di-GMP. Conversely to c-di-GMP, and according to both our crystallographic structure and NMR data, none of the two guanines seem involved in the 59-pGpG recognition. We performed a sequence alignment of FimX with other EAL domains in order to define which of the ten conserved residues identified as essentials for the catalytic activity [16] are retained ( Figure 8). We show that FimX-EAL domain has four of them at positions 1, 2, 8 and 10. Glu in position 1 is part of the EAL motif and was defined as an essential residue for magnesium coordination. This is confirmed by our structure where Glu475 coordinates the magnesium ion. Residue number 2 is conserved in FimX-EAL and corresponds to Arg479, but it is neither involved in magnesium coordination nor in 59-pGpG binding. Residue in positions 8 corresponds to Lys616 and is involved in substrate binding. Residue 10 (here Glu671) has no known function and is not involved here in ligand binding.
The number or coordinated magnesium ions in the catalytic site of the EAL domains varies amongst the structures: in BlrP1 and tdEAL, two magnesium ions are present [16,24], in YkuI, only one magnesium ion is present in the catalytic site [15]. The coordination state of magnesium in EAL also presents some variability: the two metal ions in BlrP1 are hexa-coordinated [24], but in a magnesium of tdEAL and our structure of FimX-EAL bound to 59-pGpG, magnesium ions are tetra-coordinated. The coordination of magnesium in our structure corresponds to the tetra-coordinated calcium found in the structure of B. subtilis YkuI bound to c-di-GMP [15]. We confirm the essential role of one glutamate from the EAL motif, in this coordination as suggested by previous bioinformatics and sequence alignment studies. Our structure supports a previous model of c-di-GMP hydrolysis in which Glu475 would act as a general base catalyst, activating Mg 2+ . After the hydrolysis of the c-di-GMP, the Mg 2+ is not located at the conserved position observed in other EAL domains. It remains coordinated by a water molecule, by one of the two oxygens of the hydrolysed phosphate, by a carboxylic oxygen from Glu475 of the EAL motif and by the backbone carbonyl of Gln461. This network of interactions supports a catalytic mechanism in which a water molecule activated by the cation would perform a nucleophilic attack on one of the cyclic phosphates and hydrolyse the c-di-GMP as proposed for YkuI and BlrP1. According to this model, during the reaction, the hydrolysed phosphate would move away from its original position, dragging the Mg 2+ , and induce conformation changes of the whole hydrolyzed ligand backbone, resulting also in a change of the positions of the bases. The fact that only Glu475, a water molecule and one magnesium ion are essential to carry out the hydrolysis of the c-di-GMP, could explain why the overall structure of the FimX-EAL domain doesn't undergo major structural rearrangements during the hydrolysis. Interestingly, the position of the hydrolysed phosphate is highly superimposable to the equivalent position Leishmania mexicana TIM in the complex with 2-(N-formyl-N-hydroxy)-aminoethyl phosphonate (PDB code 1IF2), suggesting the existence of a conserved phosphate-binding motif involving the C-terminal loops of TIM and EAL domains. Our results also suggest that the exact number of cations involved in the catalysis may not be fixed amongst EAL catalytic pockets.
We can't exclude the possibility that other proteins could interact with FimX. Our results coupled to previous ones, allow us to assess that there is no major structural rearrangement on EAL domain of FimX whether it is in the apo-form or bound to 59-pGpG or c-di-GMP. Recent results obtained by mass spectrometry on full-length FimX indicate that the binding of c-di-GMP induces long-range conformation changes and that this allosteric regulation could modulate the binding of a partner protein [26]. Interestingly, NMR and two-hybrid assays in Xanthomonas proved that FimX homolog is able to bind PilZ, a major protein in type IV pili biogenesis, via its EAL domain [18]. It would be interesting to determine what is the effect of 59-pGpG binding on the overall structure of full-length FimX in solution and what would be the consequences on oligomerization of the protein.
As a conclusion, we bring here the first high-resolution structure of an EAL domain complexed with 59-pGpG. This is the first time this reaction product of c-di-GMP hydrolysis is trapped within its ligand binding pocket, like a bird in a cage. The analysis of the new network of interactions between the protein, the ligand, a water molecule and the metallic cation support the model of hydrolysis of c-di-GMP by EAL domains proposed previously.
Escherichia coli BL21 strains were transformed with the recombinant plasmids for protein expression. All the N-terminal his-tagged proteins were purified using the same protocol. Bacteria were grown at 37uC up to OD 0.6. Recombinant protein production was induced by addition of 0.5 mM IPTG. Bacteria were then grown for a further 3 hours at 37uC. Cells were harvested by centrifugation at 4000 g for 20 minutes. The bacteria pellet was re-suspended in lysis buffer (NaH 2 P0 4 50 mM pH 8.5, NaCl 300 mM, Imidazole 20 mM, lysozyme 0.5 mg/ml, PMSF 0.5 mM, bensonase 100 U) and sonicated for 3 minutes. The sonicate was clarified by centrifugation at 10000 g for 30 minutes. The supernatant was loaded onto a Ni 2+ -NTA column (Qiagen). The column was extensively washed with lysis buffer and the protein was eluted with NaH 2 PO 4 50 mM, NaCl 300 mM, Imidazole 500 mM. The protein was then purified by size exclusion chromatography on a HiLoad Superdex 200 16/60 (GE Healthcare) equilibrated in Tris-HCl 50 mM, NaCl 300 mM at a flow rate of 1 ml/min. Fractions of interest were analysed on SDS-PAGE and pooled. Proteins were further concentrated on centricon (Amicon).

Enzymatic Assays
Phosphodiesterase activities of FimX-EAL, FimX GGDEF-EAL and full-length FimX was measured with Bis(p-nitrophenyl) phosphate (BpNPP) (Sigma-Aldrich) as substrate. BpNPP was diluted in 200 mM NaCl, 50 mM Tris-HCl pH 8.5 and 1 mM MnCl 2 to the final concentrations of 1 mM, 5 mM, 10 mM, 15 mM and 20 mM [22]. Standard assays were performed at 37uC for 5 hours and 30 minutes and the release of the reaction product (p-nitrophenol) was monitored and quantified at 410 nm at different times. Absorption curves were plotted against time and kinetics parameters extracted from Hanes-Woolf linearization [27].
For comparison of catalytic activities of FimX full-length, GGDEF-EAL and EAL domains, enzymatic assays were performed in reaction buffer 200 mM NaCl, 50 mM Tris-HCl pH 8.5, 1 mM MnCl 2 and BpNPP at a final concentration of 5 mM. Reaction mixture was incubated at 37uC for 9 hours and absorption was measured at 410 nm.

Crystallography
FimX-EAL (0.5 mM) alone or in equimolar complex with MgCl 2 and c-di-GMP (Biolog Life Science Institute, Gmbh) was used for crystallization. Crystallisation conditions were screened with the JCSG kit (Qiagen) on a CyBIO crystallisation robot. FimX-EAL alone crystallized in Na-Acetate 0.1 M pH 4.5, K 2 HPO 4 0.4 M, NaH 2 PO 4 1.2 M. Crystals of FimX-EAL complex were obtained by vapour diffusion in Hepes 0.1 M pH 7.5, NaH 2 PO 4 0.8 M and K 2 HPO 4 0.8 M. Crystals were cryoprotected with a brief soaking in a solution containing crystallisation buffer and 30% (v/v) glycerol. Diffraction data were collected on Proxima 1 beamline at SOLEIL synchrotron, processed with XDS [28]. Molecular replacement was done using Balbes [29] with the tdEAL structure (PDB code 2R6O) as template. The molecular replacement solution structure was automatically rebuilt using Phenix [30]. Manual reconstruction and ligand fitting were achieved using Coot [31]. Refinement was performed with Phenix. Structural analysis of the surface and interface was done with Protein interfaces, surfaces and assemblies service PISA at European Bioinformatics Institute [32]. Figures for molecular structure description were done with PyMOL [33]. NMR NMR spectra were recorded on a 600 MHz Bruker Avance spectrometer equipped with a TBI probe and X,Y,Z-gradients and samples loaded in a 5 mM Shigemi tubes. The ligands (c-di-GMP, BioLog; 59pGpG, Dharmacon) were dissolved in water to a concentration of 0.5 mM and the pH adjusted to 6.5 by the addition under stirring of small amounts of NaOH 0.1 N. The solutions were then lyophilized and re-suspended in the same volume of D 2 O. P. aeruginosa FimX-EAL was dissolved from a concentrated aqueous stock solution by the 0.5 mM ligand NMR solution to a final concentration of 30 uM. 1D proton spectra were recorded using gradient sculpting for water suppression. Reverse NOE-pumping experiments [23] were acquired with a 50 ms CMPG sequence and 400 and 700 ms NOE mixing times. Data were processed using Topspin 2.0 (Bruker).

Accession Numbers
Coordinates and structure factors have been deposited in the Protein Data Bank with accession number 4AG0 (FimX-EAL apo) and 4AFY. (FimX-EAL in complex with 59-pGpG).