Crystal Structure of Type II Peptide Deformylase fromStaphylococcus aureus

The first crystal structure of Class II peptide deformylase has been determined. The enzyme from Staphylococcus aureus has been overexpressed and purified in Escherichia coli and the structure determined by x-ray crystallography to 1.9 A resolution. The purified iron-enriched form of S. aureus peptide deformylase enzyme retained high activity over many months. In contrast, the iron-enriched form of the E. coli enzyme is very labile. Comparison of the two structures details many differences; however, there is no structural explanation for the dramatic activity differences we observed. The protein structure of the S. aureus enzyme reveals a fold similar, but not identical to, the well characterized E. coli enzyme. The most striking deviation of the S. aureus from the E. coli structure is the unique conformation of the C-terminal amino acids. The distinctive C-terminal helix of the latter is replaced by a strand in S. aureus which wraps around the enzyme, terminating near the active site. Although there are no differences at the amino acid level near the active site metal ion, significant changes are noted in the peptide binding cleft which may play a role in the design of general peptide deformylase inhibitors.

Antibiotics that target cell wall biosynthesis, protein synthesis, or DNA replication in bacteria have been the mainstay for treatment of bacterial infections for decades. However, in recent times antibiotic-resistant bacteria have become a major health threat (1). This unmet medical need has prompted a new interest in the development of novel, potent, and orally available antibiotics. Among the many potential approaches to new antibiotics include those that target novel, bacteria-specific, biosynthetic pathways. Peptide deformylase is such a target. Inhibition of this enzyme would result in blockage of essential protein processing and has been shown to be bacteriostatic (2)(3).
In eubacteria, as well as mitochondria and chloroplasts, the initiation of protein synthesis normally requires an N-formy-lated methionine residue (4 -6). The special initiation tRNA, tRNA f Met , is charged with methionine by the methionyl-tRNA synthetase (EC 6.1.10), which adds a methionine to either of the methionine tRNAs with the consumption of ATP. The formyl group is added to the charged tRNA f Met from 10formyltetrahydrofolate by methionine-tRNA f Met formyltransferase (EC 2.1.2.9). The f-Met-tRNA f Met is transferred to the ribosome where it is the preferred initiator of protein synthesis (7). Mature proteins do not retain N-formylmethionine. In fact, a heterogeneous population of amino acids is normally found at the N terminus of mature proteins: alanine, glycine, serine, threonine, or methionine. Larger amino acids are rarely found (6,8,9). This observation suggests that multiple catabolic processing might occur after or in concert with protein synthesis. This processing may be important for protein turnover (10), conservation of methionine, or it may be required for the activity of some enzymes (11)(12)(13). All known N-terminal peptidases cannot use formylated peptides as substrates (14). Thus, removal of the formyl group is an essential first step in Nterminal processing. The formyl group is removed from the peptide N terminus by a metalloenzyme, peptide deformylase (EC 3.5.1.27) (15)(16)(17). The processed protein is released for possible further processing by methionine aminopeptidase (EC 3.4.11.18). Because the formylation/deformylation cycle is unique to eubacteria and does not occur in cytosolic eukaryotic or Archaebacteria protein synthesis (4), peptide deformylase is an attractive target for the design of new antibiotics.
Previous analysis of peptide deformylase gene sequences revealed that the predicted protein coding regions have a significant sequence similarity and three characteristic stretches of highly conserved amino acids: motif 1 (G⌽G⌽AAXQ), motif 2 (EGC⌽S), and motif 3 (HE⌽DH), where ⌽ is a hydrophobic amino acid (Fig. 1). The Cys of motif 2 and the two His residues of motif 3 are involved in active site metal ion binding (18). The fourth ligand of the tetrahedrally coordinated metal ion is a water molecule that is held in place by the Glu of motif 3. Subsequent analysis of diverse peptide deformylase sequences suggested that the enzyme family could be divided into two classes (19). Peptide deformylase enzyme from Gram-negative organisms represented by Escherichia coli (Class I) have ϳ30% sequence identity with each other; whereas many Gram-positive organisms express a structural variant that includes three major insertions (20,21) relative to the E. coli sequence, some additional single amino acid insertions, and a distinct C terminus (Fig. 1). The overall sequence identity among Class II genes appears to be lower (15%) than Class I, and about half of these identities occur in or near the Class II-specific insertions. The first major insertion occurs after Tyr 39 (E. coli) and consists of 12 amino acids with high sequence similarity. In this region the Mycoplasma genes predict an insertion of only 11 amino acids (22). The second characteristic insertion is of variable length and occurs after Asn 65 . A third insertion of three amino acids follows Ser 81 and extends the sequence preceding the conserved motif 2. The last major difference between the two classes is the C terminus. In the Class II enzymes this sequence is more hydrophobic and nearly always shorter than the E. coli version.
Class I peptide deformylase from the Gram-negative organism, E. coli, has been well characterized both in terms of enzymology and structure. Efforts to purify the enzyme were hampered for 3 decades by the rapid loss of activity upon purification. This lability was traced to the Fe 2ϩ metal cofactor (23) and has been overcome by a combination of protein overexpression, elimination of oxygen, and the use of alternative active site metal ions. The Zn 2ϩ form of the enzyme displays about a 100-fold drop in activity relative to Fe 2ϩ peptide deformylase (23)(24)(25)(26), is abundantly produced during overexpression, and is much more stable than the iron version (27). Ni 2ϩ -, Co 2ϩ -, or Cd 2ϩ -substituted peptide deformylase retains activity nearly identical to that of the Fe 2ϩ enzyme, but is much less susceptible to oxidation (26, 28 -30). The x-ray and NMR structures of E. coli peptide deformylase have been reported. The ligand-free structure of the enzyme was first reported by Chan et al. (31). They subsequently determined the structure of the complex of peptide deformylase with a transition state inhibi-tor, (S)-2-O-(H-phosphonoxyl)-L-caproyl-L-leucyl-p-nitroanilide (PCLNA) 1 with Zn 2ϩ and Co 2ϩ peptide deformylase (32). Becker and co-workers (26,33,34) determined the crystal structure of the Ni 2ϩ peptide deformylase in which a polyethylene glycol molecule occupied the active site cleft. They also reported product complex x-ray structures of Zn 2ϩ , Ni 2ϩ , and Co 2ϩ peptide deformylase. Additionally, hydroxamate inhibitor complexes with actinonin and BB-3497 have been reported (2). A model of the thiol inhibitor, thiorphan, bound to E. coli peptide deformylase has been proposed (36). The structures of the Zn 2ϩ and Ni 2ϩ forms of the enzyme in solution have been determined by NMR (37,39,40). 2 Finally, a Class I peptide deformylase from the pathogenic protozoa Plasmodium falciparum has been solved to 2.8 Å resolution (41). All of these structures are essentially identical. The three conserved sequence motifs that characterize peptide deformylase (20,42) contribute side chain residues that coordinate the active site metal ion and form part of a continuous binding pocket depression. Hydrophobic interactions accommodate the methionine side chain, and a number of hydrogen bonds are formed from the peptide substrate or inhibitor to enzyme (2,32,33). The focal point of the overall fold of the protein is ␣-helix 3 that includes the HEXXH metal binding sequence, motif 3. Positioned about this ␣-helix are three small ␤-sheets that cradle the helix and form a shallow depression for peptide binding (18,31). The Class I enzyme has a long C-terminal ␣-helix that extends away from the body of the enzyme. Genetic evidence suggests that this ␣-helix can be deleted without loss of function in E. coli (44).
In this report we present the x-ray structure of Class II peptide deformylase from the Gram-positive pathogen Staphylococcus aureus at 1.9 Å resolution. Comparison of this structure with the prototypic Class I enzyme from E. coli reveals minor structural variation arising from the insertions as well as specific, active site differences that may impact the design of pathogen-specific peptide deformylase inhibitors.

EXPERIMENTAL PROCEDURES
Cloning and Expression-The S. aureus peptide deformylase sequence was identified from Human Genome Science (contig no. 168951) and cloned directly from S. aureus RN4220 genomic DNA to complete the open reading frame. The gene from strain RN4220 has the same peptide deformylase sequence as strain N315 ( Fig. 1) with the exception of N171D. The peptide deformylase gene was cloned by PCR into a vector containing the tac promoter and expressed in E. coli. The plasmid containing the peptide deformylase insert was purified and used to transform a competent strain of E. coli JM109. The cDNA clone used for structural studies contained two mutations (R127K,H186Q). The corrected amino acid sequence was expressed and purified in subsequent experiments, including biochemical experiments discussed herein. There was no effect of the R127K mutation on K m or K cat of the mutant enzyme (data not shown), and the H186Q mutation is outside the open reading frame of the deformylase gene.
The open reading frame of E. coli peptide deformylase was amplified using PCR from an E. coli genomic library. Restriction enzyme BglII sites were engineered onto both ends, and the resulting insert was ligated into the BglII site of the E. coli expression vector pQE60 (Qiagen). Both constructs express peptide deformylase followed by a Cterminal hexahistidine tag.
The peptide deformylase proteins were expressed using Luria broth with ampicillin (100 mg/liter) in both the seed and production media. Luria broth was prepared using 10 g of Bacto-tryptone, 5 g of Bacto yeast, and 5 g of NaCl added per liter of deionized water. The pH of the media was adjusted to 7.5 with KOH before sterilization. The Luria broth was autoclaved for 20 min in 100-ml volumes in 500-ml wide 1 The abbreviations used are: PCLNA, (S)-2-O-(H-phosphonoxyl)-Lcaproyl-L-leucyl-p-nitroanilide; Ni-NTA, nickel-nitrilotriacetic acid; HPLC, high performance liquid chromatography; MAD, multiple-wavelength anomalous diffraction. 2 Institute for Genomic Research, www.tigr.org. mouth fermentation flasks. Ampicillin was filter sterilized and added just before inoculation. The 100-ml seed stock fermentations were carried out in 500-ml wide mouth flasks and were inoculated from agar cultures and incubated overnight at 37°C with agitation at 200 rpm. The seed fermentations were used to inoculate at 2% the 100-ml production fermentations that were also carried out in 500-ml wide mouth flasks. These fermentations were incubated with agitation at 200 rpm for slightly longer than 2 h and were then induced (A 660 nm reached 0.6). Isopropyl-1-thio-␤-D-galactopyranoside was added to a final concentration of 0.4 mM. The induced fermentations were continued for an additional 3.5 h until the A reached 3.0. Multiple fermentations produced a final harvest of 4 -6 liters for purification.
For expression of selenomethionine peptide deformylase, M9 glucose was utilized in 100-ml volumes containing ampicillin, thiamine, and a trace metal solution. The trace metal salts solution contained per liter of deionized water: 39.44 g of MgCl 2 ⅐6H 2 O, 5.58 g of MnSO 4 ⅐H 2 O, 1.11 g of FeSO 4 ⅐7H 2 O, 0.48 g of Na 2 MoO 4 ⅐2H 2 O, 0.33 g of CaCl 2 , 0.12 g of NaCl, and 1 g of ascorbic acid at 100 mg, 5 mg, and 0.3 ml/liter of deionized water, respectively. Multiple shake flasks were used to attain the desired fermentation volume. Because JM109 is not a methionine auxotroph, incorporation of selenomethionine was accomplished through down-regulation of methionine biosynthesis just prior to induction (45). The culture was grown in 500-ml wide mouth fermentation flasks at 37°C with an agitation rate of 200 rpm until the A 660 nm reached 0.5. At this point, filter-sterilized amino acids were added to achieve down-regulation. DL-Selenomethione, L-lysine, L-threonine, and L-phenylalanine were added to final concentrations of 120 g/ml. L-Leucine, L-isoleucine, and L-valine were added to final concentrations of 60 g/ml. After 15-20 min, protein expression was induced by the addition of filter-sterilized isopropyl-1-thio-␤-D-galactopyranoside to a final concentration of 0.4 mM. Growth of the culture was continued for an additional 3 h when the A 600 nm reached about 2. Cells were then harvested by centrifugation and stored at -80°C.
Purification-Cell paste from a 2-liter E. coli fermentation expressing S. aureus peptide deformylase was lysed in 50 mM Tris-HCl, pH 8.0, with lysozyme dissolved at 1 mg/ml. The suspension sat on ice for 10 min, and large strand DNA was broken by repeatedly shearing with a syringe and 19-gauge needle. Cell extract was collected and centrifuged at 20,500 rpm for 40 -45 min at 5°C. Ni-NTA resin from Qiagen was equilibrated in lysis buffer (without lysozyme) and stirred into the cell extract. The resin suspension was poured into a column, washed extensively with lysis buffer, and peptide deformylase was eluted with 50 mM Tris-HCl, pH 8.0, buffer containing 200 mM imidazole. The eluate from the nickel column was concentrated by ultrafiltration with an Amicon stirred cell under nitrogen at room temperature. Two forms of peptide deformylase were resolved by anionic (Q-Sepharose fast flow, Pharmacia) exchange chromatography (without base-line resolution) as follows. Concentrated sample was injected onto a 1-ml column equilibrated with 25 mM Tris-HCl, pH 8.0. Proteins were resolved with a linear gradient of NaCl (100 -300 mM). The two forms of peptide deformylase were collected separately for further analysis. No differences in SDS-PAGE mobility or purity were observed. The first fraction had optimal activity, whereas the second fraction had much less specific activity. Protein eluted in the early fraction of the gradient was collected and further concentrated in a stirred cell to the desired volume. Peptide deformylase was either prepared for crystallization experiments or could be mixed with 50% glycerol and stored at Ϫ20°C. Enzyme assays have demonstrated that peptide deformylase is stable without catalase for more than 1 year when stored in 50% glycerol at Ϫ20°C.
Selenomethionine peptide deformylase was purified for crystallization efforts as described above with the inclusion of 5 mM ␤-mercaptoethanol.
Peptide Deformylase Activity Measurements-The activity of S. aureus peptide deformylase was evaluated using two different assays, based in part on the methods of Wei and Pei (46), for which coupling of the product of the peptide deformylase reaction (Met-Leu-p-nitroanilide) to Vibrio aminopeptidase is required, and Lazennec and Meinnel (47), for which coupling of the peptide deformylase reaction (formic acid generated from formyl-Met-Ala-Ser) to formate dehydrogenase is required. For the former assay method, a Vibrio proteolyticus aminopeptidase strain was obtained from Dehua Pei at Ohio State University. Stocks of the organism were prepared from 16-h cultures grown in Difco marine broth with 20% glycerol added as a cryopreservative and were stored in liquid nitrogen vapor phase. The production of aminopeptidase followed the published procedure (48) but excluded the Q-Sepharose fast flow column step and included an octyl-Sepharose step (49). The resulting preparation was concentrated by Amicon ultrafiltration to greater than 2.5 mg/ml and was stored in 50% glycerol at -20°C.
The final aminopeptidase coupled assay solution contained 50 mM NaHepes, 10 mM NaCl, pH 7.0, 0.1% potassium casein (Sigma), 100 M formyl-Met-Leu-p-nitroanilide, 5 ng of peptide deformylase protein (iron form) or 100 ng of peptide deformylase protein (zinc form), and V. proteolytica aminopeptidase. Assays were conducted in a 96-well plate system (SpectraMax 250 reader) using an absorbance of 405 nm. Time points were taken every 20 s over a 30-min time course at 25°C. The final formate dehydrogenase-coupled assay solution contained 50 mM NaHepes, 10 mM NaCl, pH 7.0, 0.1% gelatin (enzyme immunoassay grade, Bio-Rad), 250 M formyl-Met-Ala-Ser, 20 -80 ng of peptide deformylase protein, 3.125 mM NAD ϩ , and 5-50 g of formate dehydrogenase (Pseudomonas oxalaticus formate dehydrogenase Sigma). Assays were conducted in a 96-well plate system (SpectraMax 250 reader) using an absorbance of 340 nm. Time points were taken every 20 s over a 30-min time course at 25°C. In some cases we also used a third direct C-18 HPLC assay that lacked the coupling enzymes. For the HPLC assay, we followed the generation of the product Met-Leu-p-nitroanilide described in the reaction above, using UV absorbance at 315 nm.
Comparison of the Stability of the Iron-enriched S. aureus Peptide Deformylase with Iron-enriched E. coli Peptide Deformylase-Expression of the iron form of S. aureus or E. coli peptide deformylase in E. coli (both JM109) was completed after the preincubation of the cells in 1.6 mM FeCl 3 in Luria broth. Incubation was allowed to proceed for 1 h prior to the addition of 0.4 mM isopropyl-1-thio-␤-D-galactopyranoside. Induction continued for 3 h at 37°C. Isolated cells were extracted in the presence of 10 g/ml catalase. The extracts were subjected to Ni-NTA chromatography for the purification of the C-terminally hexahistidinetagged versions of both the E. coli and S. aureus enzymes. The purified proteins (concentrated to at least 1 mg/ml by Amicon ultrafiltration) were then stored in 10 g/ml (minimum concentration) catalase. To examine the stability of the iron-enriched forms of the peptide deformylase proteins, dilutions of the enzymes were made in the presence or absence of catalase (at 100 g/ml) followed by enzymatic assays. Assays were run using formyl-Met-Leu-p-nitroanilide as substrate, coupled to the Vibrio aminopeptidase. Similarly, the metal forms of S. aureus peptide deformylase were prepared by supplementation of E. coli K12 with 1.6 mM FeCl 3 , 640 M ZnCl 2, or control media followed by purification by Ni-NTA chromatography as described above (in the absence of catalase) (26,29). The peptide deformylase from each experiment (ironenriched, zinc-enriched, or no supplementation) was then assayed as described above.
Metal Ion Assessment-X-ray fluorescence experiments were carried out at the Advanced Photon Source beamline 17ID near the kappa edge for iron (1.743 Å), nickel (1.488 Å), and zinc (1.284 Å). The characteristic, qualitative signal indicates the presence of each metal ion (data not shown) in the crystallized protein. Atomic absorption spectroscopy confirmed the presence of iron and zinc using protein prepared with unsupplemented Luria broth.
Crystallization of S. aureus Peptide Deformylase-After purification the protein was stored in a buffer containing 25 mM Tris-HCl, pH 8.0, and ϳ50 mM NaCl at a concentration in excess of 50 mg/ml. The concentration was adjusted to 30 mg/ml (dilution with 25 mM Tris-HCl, pH 8.0). This protein could be frozen immediately in 50-l aliquots for later experiments. Crystallization experiments began with commercially available, random sparse matrix screens (50). Drops of 1 l of protein and 1 l of well solutions were set up in hanging drop vapor diffusion experiments at room temperature. Crystals grew in 1 week from well condition 6 of the Hampton Crystal Screen. Follow-up grid screens were set up to optimize the crystallization condition. After optimization crystals could be routinely grown with seeding from sitting drop vapor diffusion experiments by mixing 4 l of peptide deformylase and 4 l of reservoir solution (17-27% polyethylene glycol 4000, 200 mM MgCl 2 , and 100 mM Tris-HCl, pH 8.5). Crystals were successfully stabilized and slowly transferred into a cryopreservation solution containing 25% polyethylene glycol 4000, 100 mM Tris-HCl, pH 8.5, 100 mM MgCl 2 and 25% glycerol. Crystals were frozen in liquid nitrogen for cryogenic data collection. The crystals belonged to the space group C222 1 with one molecule in the asymmetric unit. The unit cell parameters are a ϭ 94.113, b ϭ 121.873, c ϭ 47.579 Å. Identical crystals were grown with selenomethionine peptide deformylase enzyme.
Structure Determination-The x-ray data for the MAD phasing of peptide deformylase was collected at the Advanced Photon Source on beamline 17ID operated by the Industrial Macromolecular Crystallography Association and consisted of three separate wavelength experiments centered about the selenium kappa edge (low, 1.03321 Å; edge, 0.97939 Å, inflection point; high, 0.97928 Å, peak of the absorption edge). The appropriate x-ray wavelength was selected using the 111 reflection from a cryocooled Si monochromator crystal. This beam was passed through a 0.15 ϫ 0.15-mm pair of slits before striking the sample. The data were collected from the cryocooled selenomethionine peptide deformylase crystal with a 2000 ϫ 2000 Brü ker CCD detector mounted upon a kappa goniometer in an inverse beam mode. Each scan consisted of 300 frames each of which was a 0.5°oscillation about the axis. A second scan beginning at ϭ 180°was accomplished before the wavelength was changed. Each frame was exposed for 0.5 s, and the crystal to detector distance was 15 cm. Data sets at each wavelength were indexed and processed separately with the program SAINT (Siemens Analytical x-ray Systems, Madison, WI) while keeping the anomalous pairs separate. The data collection and statistics are summarized in Table I. The inflection point and peak data sets were scaled to the remote energy data set (low) using the program SCALEIT in the CCP4 Program Suite (51) by treating the remote as the native. Anomalous and dispersive difference Patterson maps revealed six selenium sites whose locations were determined and refined by direct methods using the automated Patterson solution routine in SHELX (52). Heavy atom refinement and phase calculations were carried out using MLPHARE (53) with all the data from 10 to 1.9 Å resolution. The resulting electron density map was readily interpreted and a model built (see Fig. 4a). A density modified map (54) was also calculated, but this map was not very different from the MAD map. Model building was done with the graphics package CHAIN (55) and LORE (56). The structure was refined with CNX 2000.1 (Molecular Simulations Inc.) positional refinement (57) using the remote (low) data set and included three cycles of manual intervention. The active site waters and a zinc atom were incorporated into the model by direct examination of the 2Fo-Fc map, but subsequent water addition was done automatically with CNX using 3 peaks in the Fo-Fc map and a hydrogen bonding requirement. Water without 1 2Fo-Fc electron density and/or B-factors above 65 Å 2 were removed from the model. The final R-factor is 18.0% with a free R-factor of 21.2% for data from 10 to 1.9 Å resolution. The final agreement statistics (Table II) reveal a well refined structure. The final model included 1,452 protein atoms and encompassed residues Met 1 (with a free amino group) through His 184 (from the purification tag). The remaining residues from the purification tag were not ordered in the electron density map. There were no disordered internal main chain atoms. Additional statistics were generated with PROCHECK (58). A view of the initial MAD map and the final protein model in the region of the active metal ion is given in Fig. 4a and was produced using MOSAIC2. 3 The final S. aureus peptide deformylase and the E. coli peptide deformylase complex with PCLNA (32) (1bsj.pdb) were compared using SUPERPDB. 4 This structural alignment was the basis for the sequence alignment ( Fig. 1) for S. aureus and E. coli peptide deformylase. Other sequence alignments were carried out manually so as to conform to the structure-based alignment of each class. Fig. 5 was produced with MOLSCRIPT (59) and Raster 3D (60). Figs. 3, 4, 6, and 7 were prepared with the graphics program MOSAIC2.

RESULTS AND DISCUSSION
Comparison of the Stability of Iron-enriched S. aureus and E. coli Peptide Deformylase-We have found that the expression of S. aureus peptide deformylase containing a C-terminal hexahistidine tag fusion provides a facile method for purification of this protein. During the course of this work, we also found that the enzymatic activity of the purified form of the iron-enriched protein was stable, in sharp contrast to the instability reported for the Fe 2ϩ form of E. coli peptide deformylase (25). To validate this finding we expressed the iron form of S. aureus or E. coli peptide deformylase in E. coli after the preincubation of the cells in 1.6 mM FeCl 3 in Luria broth as described under "Experimental Procedures." After induction with isopropyl-1-thio-␤-D-galactopyranoside, isolated cells were extracted in the presence of 10 g/ml catalase, to act as an oxygen scavenger and to prevent the inactivation of the iron form of each peptide deformylase. The C-terminally hexahistidine-tagged versions of the E. coli and S. aureus peptide deformylase enzymes were purified by Ni-NTA chromatography and then were stored in 10 g/ml catalase. To examine the stability of the iron-enriched forms of the peptide deformylase proteins, dilutions of the enzymes were made in the presence or absence of 100 g/ml catalase, followed by enzymatic assays under standard conditions as described above. Assays were run using formyl-Met-Leu-p-nitroanilide as substrate, coupled to the Vibrio aminopeptidase. The E. coli iron-enriched peptide deformylase enzyme rapidly loses activity upon dilution in the absence of catalase (24). On the other hand, the S. aureus iron-enriched peptide deformylase remains stable regardless of whether catalase is added. Thus, the iron-enriched form of S. aureus peptide deformylase remains highly active under conditions that result in the rapid inactivation of identically prepared E. coli peptide deformylase (Fig. 2).
Similarly, the metal forms of S. aureus peptide deformylase were prepared by supplementation of E. coli K12 with 1.6 mM FeCl 3 , 640 M ZnCl 2, or control media, followed by purification by Ni-NTA chromatography as described above (in the absence of catalase) (26,29). The peptide deformylase from each experiment (iron-enriched, zinc-enriched, or no supplementation) was then assayed. The addition of zinc lowers the specific activity of the peptide deformylase, whereas the addition of iron increases the specific activity by 3-6-fold (average) compared with control media (Fig. 3). These data support the notion that the S. aureus, but not E. coli, peptide deformylase enzyme is capable of forming a stable metal complex with iron, whereas the zinc metal complex with S. aureus peptide de-  formylase behaves similarly to that described for the Zn 2ϩ form of the E. coli enzyme. Structure of S. aureus Peptide Deformylase-The Class II peptide deformylase structure is composed mostly of ␤-sheet with helical regions near the N terminus (7 to 4 and 27 to 53) and from residues 147 to 161. The N-terminal helical segments form a knot-like cluster (see Fig. 5, right) whereas the ␤-sheet regions are confined to the C-terminal two-thirds of the protein.
The other helical region forms the core of the structure and is also involved in catalysis. The conserved motif 3 is found on this central helix and is involved in the coordination of the active site metal ion (Fig. 4, a and b) (33,61,62). The ␤-sheet regions surround the centrally located helix and help create the shallow active site cavity. The ␤-sheet-rich section is composed of three ␤-sheet elements: an N-terminal antiparallel threestranded ␤-sheet, a central antiparallel three-stranded ␤-sheet, and a C-terminal mixed ␤-sheet. The C terminus of the protein   forms a last short strand of mixed ␤-sheet and is poised at the mouth of the active site (Fig. 5).
Comparison of Class I and Class II Peptide Deformylase-With the availability of numerous E. coli peptide deformylase x-ray and NMR structures, we have carried out a detailed comparison between the Class I and Class II enzymes. The root mean standard deviation between S. aureus and E. coli peptide deformylase x-ray structures for 134 common ␣-carbon atoms is 1.101 Å (1.457 Å for all 861 common atoms; 1.189 Å for 536 common main chain atoms). This superposition is the basis of the structure-based alignment of the protein sequences for selected members of both classes (Fig. 1). The strictly conserved residues (in boldface) across both classes are limited to the conserved motifs with three exceptions: Arg 124 (S. aureus), Asn 92 , and Val 122 . This valine is not conserved across a previous 34-organism alignment and will not be considered further (20). Arg 124 makes a salt bridge with Glu 109 from conserved motif 2 in both the E. coli and S. aureus crystal structures (Fig.   4b). The hydrophobic portions of the glutamic acid side chain contribute to the surface of the formyl-methionine substrate binding pocket. The side chain of Asn 92 makes a set of hydrogen bonds to both the main chain amide and carbonyl of Leu 22 . This interaction presumably stabilizes the tertiary structure and is also present in both crystal structures. The conserved residues of motif 1 form part of the wall of the active site crevice and provide loci for hydrogen bonding of peptide substrates (32). Motifs 2 and 3 contribute the protein side chains of His 154 , His 158 , and Cys 111 which coordinate the active site metal ion (Fig. 4b). Motif 3 (63) is a signature motif that is found in many metalloproteases including thermolysin (64 -67). The glutamic acid residue, Glu 155 , of this motif probably plays a dual role in metal coordination and in the protonation/deprotonation of reaction intermediates during the catalytic cycle in a manner similar to the role of the conserved glutamate in thermolysin (31,68). Examination of the E. coli and S. aureus structures in the immediate vicinity of the active site metal ion reveals no differences within 5.5 Å of the metal ion. The sulfur of Met 163 (S. aureus) is 6.0 Å from the metal ion. This residue is Leu 141 in E. coli.
Given the structural identity near the active site metal, the differences we observe in activity between the E. coli and S. aureus enzymes are perplexing. Our data show that the S. aureus enzyme retains high activity even in iron-enriched preparations. E. coli peptide deformylase is very labile and loses iron-enhanced activity rapidly when catalase is absent. No immediate structural explanation for the stability of the S. aureus enzyme or the lability of the E. coli enzyme is present. The origin may be found in protein dynamics and/or sequence variation beyond the conserved zone which affect the rate of oxidation of the metal or the amino acid side chains that are metal ion ligands.
The low overall sequence identity between Class I and Class II peptide deformylase is not reflected in overall structural similarity. Both enzymes possess similar features in tertiary structure. The common fold is embellished in S. aureus peptide deformylase by seven insertions with respect to the E. coli sequence which correspond roughly to the gaps identified by previous analysis of the protein sequences (Fig. 1). The first insertion Thr 3 -Met 4 adds some additional hydrophobic surface area that forms a small helical interface with third insertion of 12 amino acids (the Class II extended N-terminal helix) Asn 43 -Gly 54 . The second insertion after Pro 25 adds one additional residue to the turn, which leads into the common N-terminal helix of peptide deformylase (Fig. 6). Together these three FIG. 4. a, stereo view of the electron density from the initial MAD map. The final refined model including water (red spheres) and the metal ion (yellow spheres) for S. aureus peptide deformylase enzyme is included. The metal is coordinated by two histidine residues, Cys 111 , and a water is held in place by Glu 155 . b, stereo pair of the superposition of the conserved metal binding site of the S. aureus peptide deformylase structure (gray carbons) and the E. coli peptide deformylase structure (1bsj.pdb) (green atoms). Conserved motif 1 (yellow carbon atoms) motif 2 (purple carbon atoms), and motif 3 (blue carbon atoms) from the E. coli structure are indicated. Arg 124 (S. aureus) makes a conserved salt bridge with Glu 109 . The yellow spheres are the active site zinc atoms. The fourth metal ligand is a water molecule (red spheres). helices form a Class II-specific helical cluster that precedes conserved motif 1. In the E. coli structure the N-terminal helix lacks the extending residues and is followed by a ␤-turn and motif 1. The conserved motif 1 sequence forms the third (and edge) strand of the N-terminal ␤-sheet. The insertion of resi-dues Gly 81 -Gly 83 in the S. aureus structure extends the turn between strands II and III of the N-terminal ␤-sheet. These turn residues project near the mouth of the active site depression in Class II peptide deformylase and can vary in length. The fifth insertion, Val 100 , is in the turn between strand I of the central antiparallel ␤-sheet and the central strand of the Cterminal mixed sheet. The sixth insertion occurs at the end of the central strand of the mix sheet and includes Pro 106 and Thr 107 . These residues are positioned at the opening of the active site crevice and may be important determinants of S. aureus specificity. The subsequent conserved residues (motif 2) form the other wall of the active site crevice. Residue Cys 111 at the center of this sequence is one of the active site metal ligands. The conserved glutamic acid projects downward to form a part of the crevice wall and makes a conserved salt bridge with Arg 124 , which is found in the center of the first strand of the mixed ␤-sheet. The insertion of Ala 119 results in a slight bulge of the connecting strand which precedes the first strand of the C-terminal mixed ␤-sheet. This seventh insertion,  Cys 129 the sixth insertion (Pro 106 /Thr 107 ), and the C-terminal extension are all in close proximity and constitute a surface specific to S. aureus peptide deformylase and Class II in general. The insertion after Asp 134 is specific to Mycoplasma and occurs in a surface loop (Fig. 1). From the simplest comparison of these two x-ray structures one is immediately struck by the obvious difference at the C terminus (Fig. 5). The E. coli enzyme has a long protruding ␣-helix that abuts the protein surface behind the active site cavity. The C terminus of the S. aureus enzyme does not contain an equivalent ␣-helix but wraps around to complete a short stretch of ␤-sheet, terminating near the opening of the active site cavity. This is the only major topological difference between the two structures (Fig. 6). It follows from the low overall sequence identity between E. coli and S. aureus peptide deformylase that the lining of the active site cavity would not be identical. This expectation is in fact borne out by the present structure (Fig. 7). Analysis of the active site cavity suggests that nine residue changes are found in the crevice and the annulus about the active site depression (Table III). Some particularly interesting differences include the structural replacement of Arg 97 (E. coli) for Arg 56 (S. aureus) where an arginine side chain is retained in a similar three-dimensional position, but the side of the cavity from which it projects is altered. Leu 125 (E. coli) forms part of the hydrophobic surface that binds the peptide substrate methionine side chain. This residue is replaced with Tyr 147 in S. aureus, and the side chain hydroxyl is positioned near the mouth of the active site cavity. In addition, a number of subtle hydrophobic-hydrophobic changes are observed as are a number of polar-polar changes.
Our comparison of Class I and Class II peptide deformylase identifies the parameters that will influence the discovery of general inhibitors of this enzyme, as well as those that might be specific for Gram-positive or Gram-negative bacteria. The sequence and structural identity between enzymes in the immediate vicinity of the active site metal ion suggest that this is the determining locus for general inhibitors (Fig. 4b). Common modes of interaction are expected between various metal binding inhibitors across all bacteria (2, 3, 32, 69 -73) and may extend to parasite peptide deformylase (19). In addition to direct metal chelation, general inhibitors may also be expected to exploit interactions with the conserved catalytic Glu 155 ; Gln 65 (the side chain that projects above the metal from the back of the active site, Fig. 4a); and main chain interactions with Leu 112 and Gly 110 . Hydrogen bonds with different subsets of these conserved amino acids are observed in the published complexes (2,32,33). Although all peptide deformylase enzymes are expected to bind a methionine side chain optimally in the long and fairly narrow hydrophobic S1Ј subsite (74), the lack of strict conservation of side chain residues that compose this subsite affords an opportunity for specificity. The left side of the pocket is composed of strictly conserved residues, Glu 109 , Gly 110 , and Cys 111 . The remaining subsite surface is defined by variable residues: Val 59 (Ile, E. coli), Tyr 147 (Leu), Ile 150 (Ile), Val 151 (Cys), and Leu 105 (Ile). Accommodation of bulkier hydrophobic groups in S1Ј may be achieved more readily in certain peptide deformylase enzymes. It may also be possible to exploit electronic differences in this subsite, i.e. the free cysteine of the E. coli enzyme is replaced by Val 151 in the present structure (Fig. 7).
Because of more prevalent differences in the S2Ј, S3Ј, and S4Ј subsites between Class I and Class II, designing generality into inhibitors becomes increasingly difficult. The S. aureus enzyme S2Ј subsite differs from the E. coli enzyme in several important ways. The path of the main chain is altered to narrow the subsite at Ser 57 and to project the backbone carbonyl of this residue more directly into the subsite pocket. Arg 56 forms a lid across the back of the S2Ј subsite further restricting its size. In the product complex bound to E. coli peptide deformylase (33), the product makes several key interactions. These include a hydrogen bond between the peptide backbone carbonyl of the product and the side chain of Arg 97 . Equivalent interactions have been seen in other complexes including thermolysin (68). The possibility for this hydrogen bond is lost in S. aureus peptide deformylase because Arg 97 is replaced by Asn 117 . The shorter asparagine residue cannot reach to influence substrate binding. In addition, Arg 56 of S. aureus cannot replace the loss of Arg 97 . In S. aureus, Arg 56 projects from the other side of the binding pocket (Fig. 7) and places the guanidine group a few angstroms distal in S2Ј and inappropriately positioned for substrate hydrogen bonding. Several differences can also be noted in the region around S3Ј when the two enzymes are compared. There is an insertion at Pro 106 which changes the shape of the subsite. The Thr 107 (Glu, E. coli) introduces a different polar group that may interact with substrates or inhibitors on the left side of the pocket. In S. aureus the Tyr 147 hydroxyl on the right side of the pocket introduces a new hydrogen bonding group into the binding cleft. Superposition of the product complex (33) with the present structure places the new hydrogen bond donor/acceptor within 3 Å of the product peptide backbone nitrogen. A similar tyrosine hydroxyl hydrogen bond has been observed in human aspartic proteases (75,76). The S4Ј subsite is completely exposed to solvent and does not form a distinct pocket. However, protein surface residues in this region derive from the Class II-specific insertion of variable length and may modify the path by which peptide substrates (or inhibitors) approach the active site.
The comparison of E. coli and S. aureus peptide deformylase reveals the structural differences between Class I and II peptide deformylase enzymes but fails to explain the stability differences we observe between the prototypic enzymes. Examination of the active site variation suggests that general inhibitors of peptide deformylase will possess specificity determinants found near the metal and in S1Ј. However, compounds with realistic drug-like molecular masses on the order of 400 -500 Da would most certainly reach into the more divergent S2Ј and S3Ј regions. Thus, careful consideration of structural variation across classes will be required for the design of general inhibitors. Conversely, specificity to S. aureus over other bacteria may be introduced by incorporation of specific hydrogenbonding groups into inhibitors that exploit the S2Ј and S3Ј subsites. This structure determination is the essential first step toward a program of structure-based antibiotic discovery centered about a pathologically important target.