Structural Basis for Activation of an Integral Membrane Protease by Lipopolysaccharide

Background: Lipids play an important role in membrane protein function (lipid allostery). Results: The co-crystal structure of lipid-free Pla protease and a substrate peptide shows that the substrate displaces the active site nucleophilic water molecule. Conclusion: Removal of LPS from Pla causes subtle conformational changes that inactivate the enzyme. Significance: The structure provides the first explanation for lipid allostery of an outer membrane protein. Omptins constitute a unique family of outer membrane proteases that are widespread in Enterobacteriaceae. The plasminogen activator (Pla) of Yersinia pestis is an omptin family member that is very important for development of both bubonic and pneumonic plague. The physiological function of Pla is to cleave (activate) human plasminogen to form the plasma protease plasmin. Uniquely, lipopolysaccharide (LPS) is essential for the catalytic activity of all omptins, including Pla. Why omptins require LPS for enzymatic activity is unknown. Here, we report the co-crystal structure of LPS-free Pla in complex with the activation loop peptide of human plasminogen, its natural substrate. The structure shows that in the absence of LPS, the peptide substrate binds deep within the active site groove and displaces the nucleophilic water molecule, providing an explanation for the dependence of omptins on LPS for enzymatic activity.

It is becoming increasingly clear that membrane lipids do not just provide a suitable but inert environment for membrane proteins, but that they can also directly modulate membrane protein function, a phenomenon dubbed "lipid allostery" (1). The biological function of membrane proteins can be affected by membrane lipid composition as well as by specific proteinlipid interactions. Examples of the latter class are the interaction of cholesterol with G protein-coupled receptors and the activation of the inward rectified K ϩ channel Kir2.2 by phosphatidylinositol 4,5-bisphosphate (2,3). To date, however, no explanation for the modulation of bacterial outer membrane (OM) 2 protein function by membrane lipids has been reported.
Yersinia pestis, a highly pathogenic bacterium, is the causative agent of plague in humans and has been classified by the Centers for Disease Control and Prevention as a Category A microbe. An important virulence factor of Y. pestis is the 292amino acid long OM protease Pla (plasminogen activator), the physiological function of which is to activate (cleave) human plasminogen to generate the circulating serine protease plasmin. Plasmin degrades fibrin clots (fibrinolysis), noncollagenous proteins of base membranes and the extracellular matrix, to facilitate cell migration (4). Pla has been shown to be essential for development of bubonic plague by facilitating rapid dissemination of bacteria from the initial site of infection (5). Pla is also important for the progression of pneumonic plague, because it allows Y. pestis to replicate rapidly in lungs (6).
Pla belongs to a unique family of OM proteases known as omptins, which are widely distributed in Gram-negative bacterial pathogens that infect humans and plants (7)(8)(9). Omptins share high sequence identities (40 -74%). The solved structures of two members of the family, Escherichia coli OmpT (10) and Y. pestis Pla (11), show that they have very similar structures with a barrel composed of 10 antiparallel transmembrane ␤-strands connected by five long extracellular loops. Despite their high identity and structural similarity, omptins differ greatly in their target specificity for proteolysis, invasive/adhesive properties, and virulence association, probably due to adaptations to the life style of the host bacterium (7)(8)(9).
Omptins appear to constitute a unique class of proteases because their conserved active site residues (Asp 84 , Asp 86 , Asp 206 , and His 208 in Pla, Fig. 1) differ from the active site consensus sequences found in serine, aspartic or metalloproteases. Crystal structures, mutagenesis studies, and molecular dynamics simulations for OmpT suggest a unique catalytic mechanism for proteases (10,12,13). According to this proposed mechanism, the catalytic Asp 206 -His 208 dyad activates a nucleophilic water molecule that attacks the scissile carbonyl carbon atom. The Asp 84 -Asp 86 dyad likely coordinates the nucleophilic water molecule either directly or indirectly via another water molecule (Fig. 1). In addition to the active site residues, both Ser 99 and His 101 located close to the active site have been shown to be important for omptin activity (14) because they coordinate the Asp 84 -Asp 86 dyad via bridging water molecules (12) (Fig. 1).
Another unique feature of omptins is that catalytic activity requires the presence of bound lipopolysaccharide (LPS), a complex glycolipid that makes up the outer leaflet of the OM.

EXPERIMENTAL PROCEDURES
Purification of Pla for Crystallization-Cloning, expression, and site-directed mutagenesis of Pla (D84N) were carried out as described previously (11). BL21(DE3) T1 phage-resistant cells (New England Biolabs) were transformed with the pB22-Pla D84N construct. The cells were grown to A 600 ϳ 0.6 at 37°C and then induced with 0.1% arabinose at 20°C overnight. Cells were harvested by centrifugation at 4500 rpm for 30 min (J6-MC; Beckman Coulter). Cell pellets were suspended in TSB (20 mM Tris, 300 mM NaCl, 10% glycerol, pH 8.0), and cells were lysed by sonication (3 ϫ 40-s intervals) (Branson Digital Sonifier). Total membranes were obtained by centrifugation at 40,000 rpm for 40 min (Ti-45 rotor, Beckman L8-70M ultracentrifuge). Membranes were homogenized in TSB and solubilized in 1% n-dodecyl-N,N-dimethylamine-N-oxide (LDAO; Anatrace) for 1 h at 4°C followed by centrifugation at 40,000 rpm for 30 min to remove precipitates and unsolubilized membranes. The membrane extract was applied to a 10-ml nickel column. The column was washed with 10 column volumes of TSB containing 0.2% LDAO and 15 mM imidazole. The proteins were eluted with 3 column volumes of TSB containing 0.2% LDAO and 200 mM imidazole. The proteins were further purified by gel filtration chromatography using 10 mM Tris, 50 mM NaCl, and 0.05% LDAO, pH 8.0. For final polishing and detergent exchange, another gel filtration chromatography column was performed in 10 mM Tris, 100 mM LiCl, 0.35% C 8 E 4 , 0.05% decyl-␤-maltoside, pH 8.0. The purified protein was concentrated to 10 -15 mg/ml using 50-kDa molecular mass cutoff filters (Amicon), dialyzed overnight against gel filtration buffer, and flash-frozen in liquid nitrogen. For the preparation of active Pla used for cleavage of the activation loop peptide, protein was extracted in 1.5% ␤-octylglucoside to prevent loss of bound LPS (11). The nickel column and the first gel filtration column were also done in ␤-octylglucoside (0.8%).

Crystallization of Pla-Alp11 and Structure Determination-
An 11-residue activation loop peptide of human plasminogen (Alp11) with the sequence 557 KCPGRVVGGCK 567 was synthesized (Biomatik, Wilmington, DE) with the substitution V567K (numbering for intact human plasminogen) to increase solubility. The peptide was dissolved in protein buffer and added to Pla D84N at a final concentration of ϳ10 mM. Hanging-drop crystallization trials were set up using several commercial screens (MemGold and Structure I/II). Several initial hits were obtained, and one of those (Structure II/#8) was optimized to give thin plates that grew to sufficient size after several weeks of incubation. The crystals were obtained by adding 1 l of protein-peptide solution to 1 l of mother liquor containing 45% (v/v) 2-methyl-2,4-pentanediol, 0.2 M NH 4 KH 2 PO 4 , and 0.1 M NaOAc, pH 5.5. The crystals of the complex belonged to space group I222 and contained one molecule/asymmetric unit. Crystals were flash-frozen in liquid nitrogen directly from the mother liquor.
Diffraction data were collected at 100 K at the National Synchrotron Light Source (Brookhaven National Laboratory) at beamline X6A and processed with HKL2000 (17). The crystals had to be annealed by blocking the cryostream once for ϳ 2 s to improve diffraction and decrease the mosaicity dramatically. The structure of Pla-Alp11 was solved at 2.0 Å resolution by molecular replacement in Phaser (18) using the high resolution apo-Pla structure (Protein Data Bank (PDB) ID 2X55) as the search model. Model (re)building was performed manually within COOT (19), and the structure was refined using Phenix (20). Structure validation was performed within Phenix. 94.3% of the residues are present in favored regions, 5.3% in allowed regions, and 0.4% in disallowed regions of the Ramachandran plot. Final refinement statistics and analysis are presented in Table 1.

Structural Insights into LPS-dependent Activation of Omptins
Cleavage of Alp11 by Pla D84N-Pla-mediated proteolysis of Alp11 was carried out overnight by incubating Pla D84N purified in ␤-octylglucoside and Alp11 in the crystallization buffer (45% 2-methyl-2,4-pentanediol, 0.2 M NH 4 KH 2 PO 4 , and 0.1 M NaOAc, pH 5.5) at 25°C. The reaction products were analyzed using electrospray ionization MS (Proteomics and Mass Spectrometry Facility, University of Massachusetts Medical School, Shrewsbury, MA) under nonreducing and reducing (10 mM DTT, 30 min, 60°C) conditions. The sequences of the resulting peptides were confirmed by tandem mass spectrometry.

Structure of the Pla-Alp11
Complex-Wild-type Pla crystallized only in its LPS-bound active form, even when the protein was extracted and purified in LDAO (11). Moreover, soaking and co-crystallization attempts of peptides with wild-type Pla proved unsuccessful, presumably due to cleavage of the peptide substrate. We therefore employed the low activity (ϳ15% relative to WT) Pla mutant D84N to obtain a 2.0 Å co-crystal structure of Pla in complex with the Alp11 of human plasminogen ( Table 1). As is the case for the apo-protein, Pla within the complex forms an ϳ70 Å long monomeric vase-like barrel that consists of 10 antiparallel ␤-strands (S1-S10) connected by 5 long extracellular loops (L1-L5). The active site is located in a deep groove formed by the extracellular loops ( Fig. 2A) and consists of two residue couples: Asn 84 /Asp 86 in S3 on one side of the barrel and Asp 206 /His 208 in S7 on the other side. The Alp11 peptide is bound within the active site groove, between both active site couples ( Fig. 2A). In the following discussion, we will use P1 and P1Ј for the peptide residues immediately prior to and following the scissile bond, i.e. Arg 561 is the P1 residue and Val 562 is the P1Ј residue (21). The Alp11 peptide is present in the native, oxidized form, with clear density visible for the Cys 558 -Cys 566 disulfide bond (Fig. 2B) and for most of the peptide. The termini of the peptide are disordered, with Lys 557 and Lys 567 modeled as glycine residues due to the absence of side chain density. The density for the scissile peptide bond is clearly continuous, indicating that the peptide substrate has not been cleaved. Given the fact that the D84N Pla mutant has (low) catalytic activity, this result can be explained by the absence of bound LPS in the protein preparation, rendering the enzyme inactive (10). The lack of LPS is due to the fact that the protein was initially purified in LDAO (see "Experimental Procedures"), which removes LPS from its binding site on the outside of the barrel (11). Indeed, the crystal packing of the complex is such that the residues on the outside of the barrel that are normally involved in LPS binding now mediate crystal contacts, making LPS binding physically impossible within the Pla-Alp11 crystal.
LPS-mediated Conformational Changes-What is the effect of LPS removal on the Pla structure? The previously determined structures of apo-Pla in the presence and absence of LPS are very similar, especially in the membrane-embedded regions, with an overall root mean square deviation of 0.42 Å for the C-␣ atoms (Fig. 3A). The biggest structural differences between LPS-free and LPS-containing Pla are found in loops L4 and L5, both of which are present on the side of the barrel where LPS binds (Fig. 3). Most of loop L5 is not visible in the Pla-LPS structure, most likely due to autocatalytic activity at residue Lys 262 prior to extraction with LDAO (14). Potentially more relevant is the conformation of loop L4 containing the catalytic residues Asp 206 and His 208 , as well as the important residue Arg 211 (13). In the LPS-free structure, the tip of this loop is shifted (ϳ1 Å) outward relative to the structure that contains LPS (Fig. 3A). Although this conformational shift would potentially widen the active site groove, the positions of the catalytic residues at the bottom of the groove are essentially unchanged in both structures (differences Ͻ0.5 Å). Thus, the structures of apo-Pla in the absence or presence of LPS do not give unambiguous clues as to why LPS removal inactivates the enzyme.
Pla contains an active site water network, linking the catalytic residues and other residues important for enzymatic activity (Fig. 1). One of those water molecules is the putative nucleophilic water, located between Asp 84 and His 208 (11). Strikingly, in the Pla-Alp11 structure the peptide has displaced most of the active site waters, including the nucleophilic one (Fig. 3B). In

Structural Insights into LPS-dependent Activation of Omptins
fact, the position of the nucleophilic water molecule in the apostructure is very close to the carbonyl oxygen atom of the P2 residue Gly 560 within Pla-Alp11 (Fig. 3B). Within the complex, no potential other nucleophile is located close enough to the scissile Arg 561 -Val 562 peptide bond to mediate catalysis. We propose that in the absence of LPS, Pla undergoes subtle conformational changes that widen the active site groove. As a result, peptide substrates can diffuse further down into the active site, replacing essential active site water molecules and rendering the enzyme inactive (Fig. 3C).
In the Pla-Alp11 complex, the Gly 560 -Arg 561 (P2-P1) peptide bond is located between the catalytic residues Asn 84 and His 208 , closest to the position where the nucleophilic water would be in the active enzyme. This contrasts with molecular dynamics simulations of E. coli OmpT with a short substrate peptide, which placed the scissile (P1-P1Ј) Arg-Arg peptide bond between the catalytic residues Asp 83 and His 208 (OmpT numbering) (12). Extrapolating to Pla, this would mean that the Arg 561 -Val 562 peptide bond should be located between Asn 84 and His 208 . Moreover, the negatively charged pocket formed by Glu 29 /Asp 204 /Glu 217 and which was proposed to bind the P1 arginine guanidinium group (10,13), is empty in the structure of the complex (Fig. 3B). Instead, the P1 arginine side chain is located at the position of the P1Ј side chain in the molecular dynamics simulations, close to Asp 97 and Asn 151 . In other words, there is a register shift of one residue between the predictions of the molecular dynamics simulations and our structural data. It is not clear whether the register shift is a consequence of the lack of LPS or whether this is the actual substrate binding mode within the active enzyme. In this context it should be noted that Pla is a very poor enzyme for plasminogen activation, with a k cat of 0.21 min Ϫ1 (K m ϳ120 nM) (22), suggesting that the active site may have an unusual geometry.
Structural studies of active enzyme-substrate complexes as well as biochemical studies based on the current structure will be required to shed light on this issue.
Possible Implications for the Role of Arg 211 -In the Pla-Alp11 structure the guanidinium group of Arg 211 moves by 5-6 Å to interact with the peptide substrate (Fig. 4). The arginine side chain forms strong hydrogen bonds with the carbonyl oxygen of Cys 558 . In addition, a strong hydrogen bond (2.9 Å) is formed with the oxygen atom of the scissile carbonyl group (Fig. 4). Although care must be taken in the interpretation of these structural details because Pla is inactive, the observed interactions might explain the importance of Arg 211 for catalytic activity. Mutations of this residue to alanine or lysine drastically  lower Pla-mediated cleavage of plasminogen (ϳ1% compared with WT) (11,13). The observed conformational change of Arg 211 in the Pla-Alp11 structure may support our previous hypothesis that the positive charge on the arginine side chain could play a role in the stabilization of the oxyanion intermediate during catalysis, in a similar manner to the calcium ion in phospholipase A 2 (23).
Alp11 Proteolysis by Active Pla D84N-To rule out the possibility that the observed position of the Alp11 peptide within the complex is an artifact due to the small size of the peptide relative to full-length plasminogen (ϳ90 kDa), we purified the Pla D84N mutant in ␤-octylglucoside, ensuring that LPS remains bound on the outside of the barrel and preserving enzymatic activity (11). After incubating the protease with the Alp11 peptide in crystallization buffer (see "Experimental Procedures"), the reaction products were analyzed by mass spectrometry (Fig. 5). The results show that (i) the Alp11 peptide is present in the oxidized form, (ii) Pla D84N is active and cleaves Alp11 in the crystallization buffer, and (iii) the Arg 561 -Val 562 peptide bond is cleaved, identical to the cleavage site in fulllength human plasminogen (22). We conclude that the Alp11 peptide is a genuine substrate of Pla that is cleaved in the same way as its physiological macromolecular substrate, provided that LPS is bound to the protein.

DISCUSSION
The structural data presented here provide an explanation for the LPS dependence of omptins for enzymatic activity. An interesting question is why omptins require LPS to be active. By virtue of the fact that the outer leaflet of the OM consists almost exclusively of LPS, omptin activity will always be turned on. Together with the fact that a number of different forms of LPS can activate omptins (15), it is therefore unlikely that LPS serves as a regulator of Pla activity within the OM. An alternative, intriguing possibility is that the requirement for LPS is a consequence of a need to keep the protease inactive within the periplasmic space during OM targeting and integration. This hypothesis would extend the known roles of lipid allostery in membrane protein function and would imply that native-like structure in OM proteins could be formed prior to their insertion into the OM.