Novel Conformational States of Peptide Deformylase from Pathogenic Bacterium Leptospira interrogans

Peptide deformylase is an attractive target for developing novel antibiotics. Previous studies at pH 3.0 showed peptide deformylase from Leptospira interrogans (LiPDF) exists as a dimer in which one monomer is in a closed form and the other is in an open form, with different conformations of the CD-loop controlling the entrance to the active pocket. Here we present structures of LiPDF at its active pH range. LiPDF forms a similar dimer at pH values 6.5-8.0 as it does at pH 3.0. Interestingly, both of the monomers are almost in the same closed form as that observed at pH 3.0. However, when the enzyme is complexed with the natural inhibitor actinotin, the conformation of the CD-loop is half-open. Two pairs of Arg109-mediated cation-π interactions, as well as hydrogen bonds, have been identified to stabilize the different CD-loop conformations. These results indicate that LiPDF may be found in different structural states, a feature that has never before been observed in the peptide deformylase family. Based on our results, a novel substrate binding model, featured by an equilibrium between the closed and the open forms, is proposed. Our results present crystallographic evidence supporting population shift theory, which is distinguished from the conventional lock-and-key or induced-fit models. These results not only facilitate the development of peptide deformylase-targeted drugs but also provide structural insights into the mechanism of an unusual type of protein binding event.

Peptide deformylase (PDF), 2 a known essential bacterial metalloenzyme, is responsible for the removal of the N-formyl group from the N-terminal methionine of nascent polypeptides (1)(2)(3)(4). This process is required for bacterial survival, because mature proteins do not retain N-formyl-methionine, and all known N-terminal peptidases cannot utilize formylated peptides as substrate. Eukaryotic cytosolic protein synthesis, however, does not involve deformylase activity. Thus, inhibition of PDF would halt bacterial growth and spare host cell function. In fact, PDF has represented one of the most promising bacterial targets in the search for a novel mode of action antibiotics that lack cross-resistance to existing drugs (5)(6)(7). Peptide deformylase inhibitors appear to be one of the most promising classes of antibacterial agents discovered to date. As a metalloprotease, the high degree of structure-function conservation makes rapid progress possible in the development of peptide deformylase inhibitors. Very recently, a new peptide deformylase inhibitor of clinical importance (LBM415) has been characterized (8). Moreover, a new human peptide deformylase (HsPDF) has been suggested to be involved in the deformylation and processing of mitochondrially encoded proteins and may provide a novel selective target for anticancer therapy based on the observation that inhibition of human peptide deformylase (HsPDF) by actinonin (a naturally occurring competitive inhibitor of PDF) also inhibits the proliferation of 16 human cancer cell lines (9).
As a ubiquitous pathogenic bacterium, Leptospira can cause strong leptospirosis infection of animals (including humans) by entering the host through mucosa and broken skin. The resultant bacteremia, for example, is frequently found as a post-operative complication. Our previous study about peptide deformylase from Leptospira interrogans (LiPDF) revealed some unusual characteristics for the crystal structure obtained at pH 3.0, where no activity is detected (10). Structure comparison suggests LiPDF may represent a new type of PDF, belonging to neither type I nor type II PDFs. As the only dimeric PDF reported so far, the substrate pocket adopts distinct conformations (closed and open) in the two monomers. The primary difference between these two conformations is the structure adopted by a flexible loop (CD-loop, residue 65-76 of LiPDF), which controls access of substrate to the active site of the enzyme. Part of the C-terminal peptide from a symmetry-related molecule is bound to the active site in the open conformation, which was suggested as a mimic of substrate binding (not interacting with an active metal ion). In contrast, access of substrate to the active site is totally blocked in the closed conformation. However, what the conformation is at the active pH (6.5-8.0) and whether the conformational change is triggered by crystal packing remain unknown. Here we present five LiPDF crystal structures (resolutions from 2.3 to 3.1 Å) determined under different conditions. The existence of the closed conformation has been confirmed within the active pH range, with its binding cleft slightly enlarged. Specific interactions between the CD-loop and nearby residues of binding cleft are observed to stabilize the closed conformation. Meanwhile a novel half-open conformation, which has not been observed previously, is defined in the structure of LiPDF bound to an inhibitor molecule actinonin. The CD-loop and Arg 109 have been repositioned in this new conformation to accommodate the bound actinonin. Based on the three conformational states of LiPDF, an inhibitor binding model has been proposed to include a pre-existing equilibrium and a population shift. This may extend the insights into PDF inhibition, facilitating novel drug design. The hypothetical model may provide an example for an unusual type of protein binding event.

MATERIALS AND METHODS
LiPDF protein was obtained following previously described procedures (10) and was crystallized using the hanging drop method at 277 K. All conditions for experimentation were similar to those reported previously (10,11), except the solution buffer was subsequently changed to 50 mM HEPES, pH 7.5, 100 mM HEPES, pH 7.0, and 100 mM MES, pH 6.5, respectively. The co-crystallization of LiPDF and actinonin at pH 7.5 has been described previously (12).
All phases of the native crystals were obtained by molecular replacement with the AMORE software package (13). The structure of LiPDF at pH 3.0 (Protein Data Bank code 1Y6H) was taken as an initial search model. Models were refined with crystallography NMR software (14) and rebuilt using the software package O (15). After rigid body and energy minimization refinements, manual building of the C terminus and CD-loop followed by simulated annealing were carried out, with 2-fold non-crystallographic symmetrical redundancy exploited. A zinc atom and the coordinating ligands, as well as the small molecules (DDT, HEPES, MES) in the S1Ј pocket, were incorporated into the model by direct examination of 2F o Ϫ F c maps.
The structure of LiPDF complexed with actinonin was determined to a resolution of 3.1 Å using the molecular replacement method. There were eight molecules in each asymmetric unit. Some side chains from the flexible CD-loop were missing. Nevertheless, thanks to the 8-fold non-crystallographic symmetrical redundancy, all of the main chains of CD-loops, as well as the key residue Tyr 72 , were clear enough for comparison with the native structure. Statistics are shown in TABLE ONE.

RESULTS
Crystal Structures of Free LiPDF between pH Values 6.5 and 8.0, Unusual Closed Conformation-Because the previous study of both the "closed" and "open" states in crystals of LiPDF was performed at pH 3.0, where the enzyme is inactive (10), it is possible that the structure that was determined for the closed state is not a representative conformation of the free active enzyme. We have therefore determined four crystal structures (see TABLE ONE) of free LiPDF within the pH range 6.5-8.0, where the enzyme is fully active (16). Surprisingly, all four of these structures adopted the same closed conformation, which was similar to that observed at pH 3.0. Nevertheless, a detailed comparison of the closed cleft at active pH with the one at inactive pH does reveal subtle differences, which could be important for the conformational change required for substrate binding (see "Discussion").
The structure of LiPDF determined at pH 7.5 is representative of those within the active pH range. The crystal of LiPDF at this pH belongs to space group P4 1 2 1 2, with two molecules/asymmetric unit. Each molecule associates with a symmetry-related molecule through hydrophobic interactions, forming a stable dimer (Fig. 1, top right). The structures of the two molecules in the asymmetric unit are highly similar, with a root mean square deviation of 0.125 Å for 171 ␣ carbon atoms. The active site metal is a pentacoordinated zinc ion as we have described previously (10), which is liganded by residues Cys 102 , His 143 , His 148 , and a formate group providing two oxygen atoms. The formate group, which is one of the catalytic products of PDF, comes from the crystallization solution.
Similar to the closed structure observed at pH 3.0 (10), the substrate pocket of all of the structures within the active pH range was fully covered by the extended CD-loop, especially by Arg 71 and Tyr 72 . (Note that the CD-loop is not involved in crystal packing.) These two bulk residues strongly interacted (cation-and hydrogen bonding) with the side chain of Arg 109 , which is on the opposite side of the binding cleft (Fig. 1A). As such, in all of these closed structures, the substrate entrance to the catalytic site was blocked. The specific conformation of the CD-loop was highly stabilized by anchoring at three points (N terminus, C terminus, and middle tip), which appears to account for the predominance of this state in free enzyme. One hydrogen bond between the Asn 69 side chain and the Glu 46 carbonyl oxygen and another between the Glu 70 main chain nitrogen and the Glu 46 side chain formed the N terminus anchor. The middle tip anchor included the cationinteraction between Tyr 72 and Arg 109 and the hydrogen bond between the Arg 109 side chain and the Arg 71 carbonyl oxygen. On the C-terminal where I(hkl) j is the jth measurement of the intensity of reflection hkl and ͗I(hkl)͘ is the mean intensity of reflection hkl.
end of the CD-loop, the hydrophobic packing of residues Pro 76 , Thr 75 , Phe 98 , and Tyr 137 provides the third anchor. Despite the high similarity between closed structures from the active pH and from the inactive pH, the subtle differences observed around the CD-loop anchoring could be significant for the conformational change between the closed and open states. Compared with the structure at pH 3.0, a slightly open substrate binding cleft was observed under active pH (Fig. 2B). In the middle tip anchor for the CD-loop, the hydrogen bonding between the Arg 109 side chain and Arg 71 carbonyl oxygen was weakened, whereas the cation-interaction between Tyr 72 and Arg 109 became stronger (TABLE TWO). This comparison suggests that the middle tip anchoring is flexible and susceptible to environmental factors, such as pH and ionic strength.
Crystal Structure of the LiPDF-Inhibitor Complex, Novel Half-open Conformation-Because what we observed in the active pH range was a closed conformation that blocks the substrate entrance, the structure of the complex of LiPDF with its competitive inhibitor actinonin was then determined (to 3.1 Å resolution at pH 7.5) to examine its substrate binding ability and catalytic activity on the molecular level.
In the crystal of the LiPDF-actinonin complex, each asymmetric unit contained eight PDF molecules organized as four dimers. The structures of all eight molecules were nearly identical with variations observed in the flexible CD-loop region and the side chain of Arg 109 . As expected, one actinonin is observed in the substrate pocket of each PDF molecule. The zinc-coordinating formate group in the free LiPDF structure was then replaced with two oxygen atoms of the bound actinonin.
The overall structures of the LiPDF in the actinonin-bound and free states were similar, except the marked conformational change of the CD-loop. A "half-open" conformation was observed, where the actinonin was intimately bound (Fig. 1B, right). Relative to the closed state, the CD-loop was repositioned with its top end swung outward from the binding cleft by ϳ6 Å (Fig. 1) DECEMBER 23, 2005 • VOLUME 280 • NUMBER 51 of Pro 76 and Thr 75 with Tyr 137 and Phe 98 was preserved due to the synchronous outward movement of these four residues. Another impressive feature of the half-open conformation was the "swing up" of the Arg 109 side chain (rotating the 1 angle ϳ45°) from the previous horizontal orientation (Figs. 1 and 2). There the side chain of Arg 109 formed a salt bridge with Asp 171 from the C-terminal tail helix, as well as a cation-interaction with Trp 99 . The unique topology and positioning of LiPDF C terminus seemed to be required for the interaction between Arg 109 and Asp 171 .

Novel Conformational States of LiPDF
Although the binding cleft in this novel conformation was significantly splayed compared with that of the closed structure, it was still much narrower than that of any other PDF where the substrate pocket remained fully open prior to and following inhibitor/substrate binding. Therefore it was not surprising that such a compact binding pocket did not even allow clear access for actinonin without conformational change (see more under "Discussion"). Comparison of this half-open structure with all of the available inhibitor complexes of PDFs revealed a highly conserved inhibitor binding mode. However, additional hydrophobic interaction between Tyr 72 of the CD-loop and P3Ј Pro of actinonin was observed only in the half-open conformation of LiPDF.

DISCUSSION
Three Conformational States of LiPDF-Our previous study of LiPDF at pH 3.0 (10) revealed a non-free open conformation and a free closed conformation. The open conformation, as a peculiar artifact of crystallization, resulted from the binding of the C terminus peptide of a symmetry-related molecule. Coincidentally, the N terminus of a symmetryrelated molecule was also observed in the substrate pocket of the peptide deformylase from Staphylococcus aureus, which was suggested to mimic substrate binding (17). Thus the open conformation of LiPDF may resemble the state where actual substrate is bound. However, the CD-loop in this state was specifically stabilized by the interaction between Tyr 72 and the bound peptide (10). Removal of the bound peptide would result in a completely solvent-exposed and hence highly unstable CD-loop. All structures of LiPDF indeed adopted a closed (but not open) conformation in the free state. Actinonin could freely dock into the substrate pocket of the open state when bound peptide was removed (Fig. 3C).
The existence of the closed conformational state is now confirmed at the active pH range. Both monomers in the asymmetric unit showed the same closed conformation, even though they faced a different environment in the crystal. Nevertheless, here the closed form of free LiPDF possibly describes a state that was ready to open its binding cleft. Furthermore, the recurrence of the stable closed state for free LiPDF in different conditions suggests that such a conformation would also occur in solution and would predominate in the absence of substrate/inhibitor. On the other hand, the side chain of Tyr 72 would collide with the side chains from P2Ј and P3Ј when the inhibitor actinonin is modeled into the closed substrate pocket. This clearly indicates that the active site in the closed state of free LiPDF is not accessible when no conformational changes occur.
The current structure of the complex substantiates the substrate binding ability as well as the catalytic activity of LiPDF, in keeping with our previous biochemical studies (16). Here a novel conformation was

Statistics of Arg-involved cation-interactions
This calculation was performed according to the systematic method described in Ref. 29. Etotal indicates the strength of the cation-interaction. According to this method, the candidate interacting pair is considered to be energetically significant only when Etotal Ͻ Ϫ2.0 kcal/mol. Crystal-packing examination also revealed that the CD-loop region is essentially solvent-exposed in all of the eight molecules. In this half-open state, the actinonin molecule was snugly bound in the active crevice. In contrast to studies of PDFs from different species, where P3Ј proline of the bound actinonin is solvent-exposed (18 -19), in LiPDF, the P3Ј proline of the bound actinonin was partially covered by a hydrophobic environment due to the special positioning of the CD-loop. This hydrophobic S3Ј subsite consisted of Tyr 72 and Thr 75 from the CD-loop and the adjacent residues Therefore Arg 109 appears to be a pivotal residue for the conformational changes triggered by substrate binding. Note that Arg 109 structurally corresponds to residue Arg 97 in peptide deformylase from Escherichia coli (20).

PDB
Hypothetical Model for Substrate Binding-Traditionally, conformational changes related to protein binding events have been described by lock-and-key or induced-fit models. However, these two models do not fit the case of LiPDF, because the free enzyme was observed to be in a closed form, and no inducer molecule has been found for LiPDF or for other PDFs.
As an alternative, the pre-existing equilibrium hypothesis (21) is based on protein folding and conformational selection theories of the funnel energy landscape (22)(23)(24)(25), which postulates that the native state of a protein contains an ensemble of conformations at its binding site, rather than a unique fold. In this view, the energy landscape of the protein is dynamic, changing with environmental factors such as pH, ionic strength, and the presence of other molecules. The most populated conformation of a protein may be different prior to and following a binding event, corresponding to a change in the energy landscape. A ligand will bind selectively to an active conformation, thereby biasing the equilibrium toward the binding conformation. An elegant example comes from the study of a monoclonal IgE antibody, SPE7, in which both x-ray crystallography and presteady-state kinetics revealed an equilibrium between at least two different pre-existing isomers (binding structurally distinct antigens) that can both be crystallized in the absence of any ligand (26).
Based on this theory and our observations, we suggest that the closed state of free LiPDF might not be the only conformation adopted in solution. Instead, the open form of LiPDF may co-exist with the closed form, although as the minor population. Because the crystallization process may select one of the populations to form a crystal, but not necessarily the "active" conformers, the three conformational states imply a hypothetical model for substrate binding in LiPDF (Fig. 3). For the free enzyme, a strong, biased equilibrium between closed and open states is suggested, as above. The substrate binding would trigger a redistribution of the closed and open states in the ensemble of conformations. After an initial binding to the open state, the substrate pocket is then induced to shut, leading to the final half-open conformation, as observed in the complex. In this model, we suggest that the pivotal residue Arg 109 , which is on molecular surface, could be sensitive to the environmental disturbances; therefore the locking bridge (mainly contributed by the interactions between Arg 109 and the CD-loop) could be momentarily broken, leading to a conformational ensemble. Among these conformations, the predominant closed state is thermodynamically favored in the absence of ligand, whereas other minor states Conclusion and Perspective-As a novel drug design target from a real pathogen, LiPDF showed unique structural features highlighted by its three conformational states. A pre-existing equilibrium-based model has been proposed for the competitive inhibitor binding. Besides providing structural information for drug development, this model may extend our insights into an unusual type of enzymatic binding event.
Meanwhile, we have noticed that the structures of deformylase from other bacteria do not show significant conformational changes upon inhibitor binding. This suggests that the conformational change observed in LiPDF is not a general feature regarding deformylase activity itself. Why does LiPDF use this machinery? It has been realized that a protein that adopts several different conformations could, in principle, have several different functions (28). Genomics data indicate that many species have two copies of similar but different PDF genes, and PDF genes are also found in eukaryotics. Thus at this stage, we cannot rule out the possibility that peptide deformylase (in at least some species) have additional biological functions besides modifying the newly synthesized peptide.