Crystal structure and catalytic mechanism of the LPS 3-O-deacylase PagL from Pseudomonas aeruginosa

PagL from P. aeruginosa was refolded from inclusion bodies. Folding could be monitored by SDS/PAGE analysis. Like many outer membrane proteins (OMPs), PagL showed heat modifiability (Fig. 1A). Usually, the folded form of an OMP has a higher electrophoretic mobility than its denatured form. In the case of PagL, however, the folded protein has a lower electrophoretic mobility as has been observed also for a few other small OMPs, i.e., OmpA171t, OmpX, and NspA (9, 10). The same mobility shift was observed on Western blot for wild-type PagL folded in vivo (data not shown). Folded PagL was insensitive to denaturation by SDS, as it remained folded even in the presence of 2% SDS when not heated (Fig. 1A).

To verify that PagL was correctly folded in vitro, it was incubated with purified LPS of Neisseria meningitidis as described in the Supporting Text, which is published as supporting information on the PNAS web site. The LPS was converted into a form with a higher electrophoretic mobility (Fig. 1B), in agreement with the expected hydrolysis of the primary acyl chain at the 3 position of lipid A, which is 3-OH C12 in the case of N. meningitidis. Deacylation of the LPS was confirmed by electrospray ionization mass spectrometry (Fig. 6, which is published as supporting information on the PNAS web site). The reaction was independent of the presence of divalent cations, as deacylation of LPS was still observed in the presence of 5 mM EDTA (Fig. 1B). Quantification of the amount of 3-OH C12 released in time revealed a specific activity of 0.40 nmol 3-OH C12 released per min per nmol of in vitro folded PagL. E. coli membranes containing overexpressed in vivo folded wild-type PagL displayed a lipid A-deacylase activity of ≈0.22 nmol/min per nmol of PagL, which is comparable to the specific activity of in vitro folded PagL. We concluded that PagL was correctly folded in vitro into its active conformation and, thus, suitable for structure determination.

PagL was crystallized in the C2 space group, containing two molecules in the asymmetric unit (Fig. 7, which is published as supporting information on the PNAS web site). The structure of PagL was solved with a combination of molecular replacement (MR) using the program phaser (11), together with single-wavelength anomalous dispersion. MR was successfully performed by using a polyalanine model of the β-stranded part of NspA. The calculated 2F_o − F_c maps (even after prime-and-switch in resolve, ref. 12) did not allow for initial model tracing by hand or by automated model building with arp/warp (13). Determination of the position of the single methionine present in PagL, using the anomalous signal of the selenium from a crystal of selenomethionine (Se-Met)-substituted protein, enabled limited model tracing and subsequent automated model building. The structure was refined to 2.0-Å resolution.

The overall structure of PagL consists of an eight-stranded antiparallel β-barrel, which is consistent with our earlier prediction of its topology (8). So far, the N and C termini of the β-barrel part in all OMP structures solved are facing the periplasm (reviewed in ref. 14). Therefore, we propose that the same holds for PagL. In this orientation, the previously identified active-site residues Ser-128 and His-126 (8) are located near the hydrophilic/hydrophobic boundary of the outer leaflet of the outer membrane (Fig. 2A). This position would be consistent with the location of the scissile bond of the substrate LPS. Like other OMPs, PagL has long extracellular loops (L) and short periplasmic turns (T), except for T1, which is exceptionally long (Fig. 2A).

Like most other OMPs, PagL contains two girdles of aromatic residues, which are located at the hydrophobic/hydrophilic boundaries of the membrane and are assumed to stabilize the position of the protein in the membrane. Because these girdles are not perpendicular to the β-barrel axis of PagL, as they usually are in OMPs, the orientation of PagL in the membrane may be tilted (Fig. 2A). Therefore, we calculated the net hydrophobicity at the β-barrel surface as a function of the height in the membrane as described (15); this was done for two orientations, i.e., with the β-barrel axis parallel or with a 30° tilt angle relative to the membrane normal (Fig. 2B). In the tilted orientation, the graph resembles the average graph calculated for all OMP β-barrel structures, in which two hydrophobic peaks represent the aromatic girdles (15). Remarkably, the hydrophobic peak at the extracellular side is much lower than that at the periplasmic side, which can easily be explained by the presence of a highly hydrophilic active site at that height in the barrel (see below). Also, in the parallel orientation, two hydrophobic peaks can be discriminated. However, the area of hydrophobicity of the intracellular peak extends considerably into the periplasm, suggesting that a parallel orientation is energetically highly unfavorable. Therefore, we propose that PagL has a tilted orientation in the membrane.

Previously, we identified PagL homologs in the genomes of a wide variety of Gram-negative bacteria (8). These homologs exhibited a low overall mutual sequence identity, with only 10 conserved residues among 14 different homologs. Here, we performed a more extensive search, which revealed additional PagL homologs in Caulobacter crescentus, Collimonas fungivorans, Chlorobium tepidum, Rhodospirillum rubrum, Sinorhizobium meliloti, and Ralstonia eutropha. An alignment of all currently identified PagL homologs is presented in Fig. 8, which is published as supporting information on the PNAS web site. Among all PagL homologs identified so far, only four amino acid residues are fully conserved, i.e., Phe-104, His-126, Ser-128, and Asn-136 of the P. aeruginosa PagL sequence.

Active sites of hydrolases comprise three important regions: the catalytic site, a substrate-binding site, and an oxyanion hole. Classical serine esterases contain a catalytic site consisting of a Ser-His-Asp/Glu triad. The histidine abstracts a proton from the serine residue, after which the serine performs a nucleophilic attack on the carbonyl carbon of the scissile bond. This attack can only be performed efficiently if the Nε2 of the histidine is deprotonated. Deprotonation of the histidine is achieved by stabilization of the proton on the Nδ1 of the histidine via a hydrogen bond with an acidic residue. Usually, this acidic residue is an aspartate or a glutamate. However, in a few exceptions, such as in outer membrane phopholipase A (OMPLA) of E. coli, an asparagine residue is also able to stabilize the deprotonated state of the Nε2 of the histidine (16).

Site-directed mutagenesis studies have shown that His-126 and Ser-128 (8) are important for catalytic activity and constitute part of the active site of PagL. In the crystal structure, two acidic residues, Glu-140 and Asp-106, are located in close proximity to His-126. A hydrogen bond is present between the carboxylate of Glu-140 and the Nδ1 hydrogen of His-126, which implies that Glu-140 is the acidic component of the catalytic triad. This notion is supported by the observation that a Glu140Ala substitution reduced the activity of PagL in vivo in E. coli membranes (Fig. 3). LPS was almost completely converted into its 3-O-deacylated form 75 min after induction of expression of wild-type pagL, whereas ≈50% remained in the acylated form when the mutant PagL was produced. In the quantitative assay, in which the release of 3-OH C12 from exogeneously added Neisserial LPS is measured, the mutated protein appeared 142-fold less active than the wild type.

Another acidic residue, Asp-106, is present on the opposite side of His-126. In the crystal structure, Asp-106 coordinates a Ca²⁺ ion. Because PagL activity is not influenced by EDTA, this Ca²⁺ ion is likely not relevant for activity and may just be a crystallization artifact. Because Asp-106 points toward the hydrophobic region of the membrane, which is not a favorable position for a charged residue, it seems plausible that it has an important function in the catalytic mechanism. To test this possibility, we substituted Asp-106 by Ala. Unfortunately, the mutant protein was poorly expressed (data not shown), indicating that the substitution caused folding and/or stability problems of the protein.

To gain further insight into the catalytic mechanism, we modeled lipid X, the smallest known substrate of PagL in vitro (6) into the active site of PagL by using the program haddock (17). Several restraints were used. The distances between the hydroxyl oxygen of Ser-128 and the carbonyl carbon of the scissile bond and between the Nε2 of His-126 and the main-chain oxygen of the ester bond were both restrained to 3 Å. Furthermore, the acyl chains of lipid X were restrained to point toward the periplasm. In the resulting model (Fig. 4), the residual acyl chain (i.e., the chain to be cleaved off from the substrate) of lipid X is bound in a well defined hydrophobic groove, whereas the acyl chain of the leaving group is loosely bound into a second hydrophobic groove. The conserved residue Phe-104, which makes a hydrophobic interaction with the residual acyl chain, is likely to be a key residue for positioning the substrate into the correct orientation. Furthermore, the hydroxyl group of the residual acyl chain makes a hydrogen bond to Asp-106, suggesting a possible role for this residue in substrate specificity.

The structure of one other outer membrane esterase has been solved, i.e., that of OMPLA, which forms a 12-stranded β-barrel with a catalytic site composed of a Ser-His-Asn catalytic triad (16). To compare the catalytic sites of OMPLA and PagL, we superposed Ser-128 and His-126 of PagL onto Ser-144 and His-142 of OMPLA (Fig. 5). In both PagL and OMPLA, the active sites are located at comparable heights in the barrel relative to the membrane. Furthermore, Glu-140 of PagL occupies the structurally equivalent position of Asn-156, the acidic residue in the catalytic triad of OMPLA, consistent with Glu-140 being the acidic residue of the PagL catalytic triad. In many serine hydrolases, the oxyanion hole consist of two backbone amide groups that form a slightly positively charged center that stabilizes the transient negative charge on the carbonyl oxygen of the substrate during the reaction. The oxyanion hole of OMPLA is formed by two backbone nitrogens from two glycine residues and two water molecules, coordinated by a calcium ion (16). For PagL, the oxyanion hole could well be formed by the two backbone nitrogens of the highly conserved Ala-130 and Gly-131 residues, which are located at the equivalent positions of the oxyanion-hole glycines in OMPLA. Comparison of the OMPLA and PagL active sites further showed that the side-chain nitrogen of Asn-136, the fourth completely conserved residue among the PagL homologs, occupies the position of one of the oxyanion-hole waters in OMPLA (Fig. 5). Therefore, we speculate that the side-chain amide group of Asn-136 is part of the oxyanion hole of PagL. This possibility is supported by the observation that an Asn136Ala mutant derivative of PagL was less active in vivo than wild-type PagL (Fig. 3). In the quantitative assay, the activity of this mutant protein was reduced 304-fold. In addition to the three highly or even completely conserved residues Ala-130, Gly-131, and Asn-136, which we postulate to form the oxyanion hole of PagL, one other residue, Asn-129, shows a high conservation among the PagL homologs (see Fig. 8). Furthermore, the same residue is present at the equivalent position in OMPLA (Fig. 5). We propose that Asn-129 of PagL might play a crucial role in stabilizing the conformation of loop 4 and thereby that of the oxyanion hole and the active site. The importance of Asn-129 for the enzymatic activity is supported by the observation that an Asn129Ala mutant derivative of PagL showed no activity in vivo (Fig. 3), nor in the quantitative assay even after 27 h incubation with the substrate.

Mutations were introduced in pagL by using the QuikChange Site-Directed Mutagenesis kit (Stratagene) and the primers listed in Table 2, which is published as supporting information on the PNAS web site. Plasmid pPagL_(Pa) encoding wild-type PagL, including the signal sequence (8), was used as the template in which the mutations were created. The presence of the correct mutations was confirmed by nucleotide sequencing in both directions.

PagL was produced in E. coli as described in Supporting Text and folded in vitro by 2-fold dilution from a stock solution in 8 M urea into 10% (wt/vol) lauryldimethylamine-oxide, followed by 10 min of sonication using a Branson 1210 Sonifier. Refolded PagL was diluted 3- to 4-fold with buffer A [20 mM Tris·HCl, pH 8.5/0.08% (wt/vol) pentaethyleneglycol monodecyl ether (C₁₀E₅)] and purified by FPLC using a 1-ml monoQ (Amersham Pharmacia) ion-exchange column that was preequilibrated with buffer A. PagL was eluted with a linear gradient of 0–1 M NaCl in buffer A. Fractions containing PagL were pooled and concentrated to ≈10 mg/ml by using Centricon concentrators with a molecular mass cutoff of 3 kDa (Amicon). The protein was then dialyzed three times overnight against 10 ml of 2 mM Tris·HCl, pH 8.5/0.06% (vol/vol) C₁₀E₅ using a membrane with a molecular mass cutoff of 3.5 kDa.

PagL was crystallized by using the hanging-drop vapor diffusion method. Crystals grew in the C2 space group, using a condition containing 3% PEG3000/25% glycerol/10 mM calcium acetate/0.1 M Tris·HCl, pH 8.5 at 20°C. Se-Met PagL crystals were grown in the same condition, with 1 mM DTT added to the protein solution. Crystals were harvested from the drops with a cryo-loop and directly cooled into liquid nitrogen. Data sets were obtained at 100 K on a charge-coupled device detector at beamline ID14-EH4 at the European Synchrotron Radiation Facility. Native data were collected to 2.0-Å resolution. Data of the Se-Met PagL crystal was collected at λ_peak = 0.9795 to 2.8-Å resolution. All data were indexed and processed by using denzo and scalepack (24). Table 1 summarizes data collection information. The structure of PagL was solved by using a combination of Se-Met single-wavelength anomalous dispersion (SAD) and MR. From the Se-Met SAD data set, the two selenium atoms in the asymmetric unit (one per PagL molecule) were located by using standard programs (12, 25, 26). Unfortunately, the anomalous signal was too low to obtain an interpretable density-modified electron-density map. Therefore, we attempted MR. Whereas all commonly used MR programs failed, the program phaser (11) provided an MR solution using a polyalanine β-barrel model of the NspA structure, lacking the loops. The poor phases obtained from the incomplete model were sharpened by using the prime-and-switch algorithm in the program solve (12). The positions of the seleniums indicated the locations of the methionines, which enabled us to correctly build a few β-strands by using the program o (27). Subsequently, the complete model could be built automatically by using arp/warp (28), which was not the case directly after MR. The structure was refined by using refmac 5.0 (29), using TLS groups for the separate molecules in the asymmetric unit. Refinement statistics are summarized in Table 1, and the electron density map is shown in Fig. 9, which is published as supporting information on the PNAS web site.

Models of lipid X docked onto PagL were created by using the program haddock (17). The coordinates of PagL were taken from chain A of the structure and the lipid X coordinates were based on those of the LPS molecule that was cocrystallized with FhuA (30). Ambiguous interaction restraints (AIRs) were defined, based on the information on the reaction mechanism: 3-Å-distance restraints were defined between the Ser-128 oxygen of PagL and the carboxyl carbon of lipid X and between the His-126 nitrogen (Nε2) and the ester oxygen of lipid X. Additional AIRs were imposed merely to ensure that the lipid X acyl chains were positioned at least as close to the PagL molecule as they could be in the outer leaflet of the outer membrane. First, an ensemble of 20 conformations of lipid X was generated by simulated annealing and molecular dynamics refinement in DMSO, which was subsequently used to generate 1,000 rigid body docking solutions. The best 200 solutions based on their intermolecular energy (sum of van der Waals, electrostatic, and AIR energies) were then submitted to a semiflexible refinement, during which lipid X was treated as fully flexible, whereas PagL could only move at defined flexible regions (side chains first, subsequently both side chains and backbone). All 200 models were finally refined in explicit solvent (DMSO was used to mimic the hydrophobic environment imposed by the membrane) and clustered based on root mean square deviation criteria. The lowest energy structure from the lowest energy cluster was taken as best solution.