An Unusual Peptide Deformylase Features in the Human Mitochondrial N-terminal Methionine Excision Pathway*

Dedicated machinery for N-terminal methionine excision (NME) was recently identified in plant organelles and shown to be essential in plastids. We report here the existence of mitochondrial NME in mammals, as shown by the identification of cDNAs encoding specific peptide deformylases (PDFs) and new methionine aminopeptidases (MAP1D). We cloned the two full-length human cDNAs and showed that the N-terminal domains of the encoded enzymes were specifically involved in targeting to mitochondria. In contrast to mitochondrial MAP1D, the human PDF sequence differed from that of known PDFs in several key features. We characterized the human PDF fully in vivo and in vitro. Comparison of the processed human enzyme with the plant mitochondrial PDF1A, to which it is phylogenetically related, showed that the human enzyme had an extra N-terminal domain involved in both mitochondrial targeting and enzyme stability. Mammalian PDFs also display non-random substitutions in the conserved motifs important for activity. Human PDF site-directed mutagenesis variants were studied and compared with the corresponding plant PDF1A variants. We found that amino acid substitutions in human PDF specifically altered its catalytic site, resulting in an enzyme intermediate between bacterial PDF1Bs and plant PDF1As. Because (i) human PDF was found to be active both in vitro and in vivo, (ii) the entire machinery is conserved and expressed in most animals, (iii) the mitochondrial genome expresses substrates for these enzymes, and (iv) mRNA synthesis is regulated, we conclude that animal mitochondria have a functional NME machinery that can be regulated.

The N-terminal methionine excision (NME) 1 pathway is an essential pathway that removes the initial Met from two-thirds of the proteins of any proteome (1). Methionine aminopeptidase (MAP; EC 3.4.11.18) has been shown to be involved in this process in all organisms studied, and this activity facilitates subsequent protein modification in Eukaryotes (2,3). In Eubacteria, plastids, and mitochondria, the N-terminal Met residue of nascent polypeptides carries an N-formyl (Fo) group whereas in Eukaryotes and Archaea, nascent proteins synthesized in the cytoplasm start with a free Met (4). The N-terminal Met must be exposed to allow MAP activity in Eubacteria, and this is achieved by systematic removal of the Fo group by peptide deformylase (PDF; EC 3.5.1.88). PDF activity was long considered to be unique to this kingdom (5). Recent studies have demonstrated that PDF orthologs are produced in most eukaryotes, including animals, plants, and many unicellular organisms (6 -8). These orthologs have been shown to display peptide deformylase activity in plants, both in vitro and in vivo. All the eukaryotic orthologs have an extended N-terminal domain. The N-terminal domain targets the proteins to the plastids and mitochondria in plants (7, 9 -11). Similarly, it has been suggested that the PDF orthologs of Apicomplexan parasites are targeted to the apicoplast, an essential plastid (8,12,13). Actinonin, a natural antibiotic that specifically inhibits PDF, targets both bacterial and plant PDFs (10,14). Studies carried out in vivo with this drug have shown that PDF is essential in bacteria and plastids (11,14,15). In plants plastids, NME plays a crucial role in controlling the half-life of a major organellar protein complex, photosystem II (10). The effect on photosystem II stability was similar regardless of whether Met alone or Fo-Met was retained.
MAPs are part of a large family of metallo-aminopeptidases (16). Two types of MAPs have been described to date (17). MAP2s occur in Archaea and in the cytoplasm of Eukaryotes, whereas MAP1s have been found in Eubacteria, in the cytoplasm of Eukaryotes (MAP1A) and in the organelles of plants and Apicomplexa (MAP1D, Ref. 7). The sequences of MAP2 and MAP1 have weak similarity but these two enzymes display similar active site folding patterns. Organellar MAP1s have a cleavable N-terminal extension that is not found in bacterial MAP1s. This extension targets the catalytic domain to the correct cell compartment. Cytoplasmic MAP1A enzymes also have an extension including a conserved zinc finger. Although not involved in catalytic activity, this additional domain is not * This work was supported by an Action Thématique et Incitative sur Programme grant from the C.N.R.S. (to T. M.) and by Grant 4477 from the Association pour la Recherche sur le Cancer (ARC, Villejuif). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. ʈ To whom correspondence should be addressed. Tel.: 33-1-69-82- 36-removed from the mature form and is essential for cell function, possibly allowing interaction with ribosomes (18). PDFs constitute a growing family of hydrolytic enzymes related to the thermolysin-metzincin HEXXH motif-containing family of metalloproteases (6). PDFs contain three distinct short stretches of amino acids (Motifs 1-3; Fig. 1) that constitute the active site (19). Motif 3 contains the HEXXH motif ( Fig. 1). In most PDFs, an Fe 2ϩ cation is bound by three residues of motifs 2 and 3 and plays a crucial role in hydrolytic activity (20,21). This ion is highly unstable, and several agents and procedures have been described that preserve deformylase activity (reviewed in Ref. 6). The iron cation may be replaced by zinc, resulting in a stable but weakly active enzyme. Zinc PDFs have similar substrate specificities to their iron-coordinated counterparts, but lower catalytic constants (22,23). Three types of PDF have been identified on the basis of structural and large scale sequence analysis (6,13,24). The three classes differ in several parts of their three-dimensional structures but their active sites are conserved and entirely superimposable. Type 2 and type 3 PDF enzymes are found only in Grampositive bacteria. In contrast to type 2, type 3 PDF orthologs have no associated deformylase activity, due to amino acid substitutions in motifs 1 to 3 (13,25). Each of the specific amino acids of the motifs appears to be required for enzyme activity and stability, as indicated by near-systematic site-directed mutagenesis analysis followed by in vivo and in vitro studies (19, 26 -29). There are two classes of type 1 PDF. Class 1B (B for bacterial) PDF enzymes are found in Gram-negative bacteria, some Gram-positive bacteria and plants. Eukaryotic class 1B PDFs are targeted to both plastids and mitochondria (7,9). The three-dimensional structure and enzymatic properties of the class 1B PDFs of eukaryotes and bacteria are similar (11,24,30). Class 1A PDFs include plant mitochondrial PDFs and PDF orthologs from animals (9). Biochemical characterization of plant PDF1As has shown that they differ from most previously studied PDFs, including PDF1Bs and PDF2s, particularly in terms of the optimal metal cofactor: zinc rather than iron (11).
In this work, we provide compelling evidence that the complete machinery, including specific and functional PDF and MAP1D, is expressed and exported to animal mitochondria. Site-directed mutagenesis data, combined with biochemical and structural analyses, showed that HsPDF differs considerably from previously characterized PDFs in terms of its properties. These differences could be used as the basis for chemical modifications for improving PDF inhibitors for use in the clinical treatment of bacterial infections in humans.

Materials and Preparation of Solutions
All chemicals, including Tris(2-carboxyethyl)-phosphine (TCEP), and enzymes for protein analysis were purchased from Sigma. The peptides used have been described elsewhere (28,31). Enzymes for DNA manipulation were purchased from New England Biolabs. Oligonucleotides were synthesized by MWG-AG Biotech. Plasmid DNA was purified with mini-and mid-prep kits (Qiagen). For enzyme purification, we used plastic containers rather than glassware and all containers were thoroughly rinsed with metal-free water (see details in Ref. 23).

Bacterial Strains and Plasmids, Molecular Biology
For the sake of clarity, all PDF sequences were numbered similar to the EcPDF sequence, as previously suggested (24). A substitution of residue X in HsPDF or AtPDF1A indicates that the substitution concerns the residue corresponding to amino acid X in EcPDF, as shown in the alignment in Fig. 1. Residues upstream from Ser 1 of EcPDF are numbered relative to position 1 and take a negative sign.
Cloning in GFP Fusion Vectors-Full-length HsPDF and HsMAP1D sequences and sequences with deletions were inserted between the XhoI and XmaI restriction sites of pEGFP-N1 (Clontech). The plasmid pCB6 (34) encodes the N-terminal mitochondrial domain of mitofusin fused to DsRed1-RFP (mt/RFP), which was used as a control for mitochondrial location. We inserted sequences into pSmRSGFP, expressing GFP under the control of the 35S promoter, between the unique XbaI-BamHI sites, as previously described (7).
Redesign of the HsPDF ORF-The starting material was pQHsdef⌬N62ori, a plasmid encoding codons 63-244 of HsPDF inserted between restriction sites NcoI and HindIII of pQE60 (Qiagen) in-frame with an additional N-terminal Met-Ala dipeptide and a C-terminal His 6 tag. Block I encodes the first 66 codons of the catalytic domain of HsPDF (i.e. codons 62-127 of HsPDF). We introduced a BamHI site into block I, at the 5Ј-end (bases 34 -39), for subsequent block II insertion. Block II encodes the N-terminal targeting sequence of HsPDF (i.e. codons 1-70; Fig. 1). Both synthesized fragments (each ϳ200 bp in length) were built by assembling in vitro six 50 -75 bp oligonucleotides, as previously described (35). The initial remodeling step involved replacement of the NcoI-SacI fragment of HsPDF⌬N62 with block I to yield pQHsdef⌬N62mod. We then inserted block II between the NcoI and BamHI sites of pQHsdef⌬N62mod to yield the full-length redesigned wild-type HsPDF ORF (Hsdefmod). The final optimized nucleotide sequence of Hsdefmod encoding full-length HsPDF is available from Gen-Bank TM under accession number AY368205.
Strain GIF1 Susceptibility Tests-We designed a susceptibility test to determine the susceptibility to drugs of the tolC strain GIF1 derivative producing a given PDF. Insertion of the PDF sequence into the pBAD vector (Invitrogen) render the synthesis of this protein dependent on arabinose concentration (36). The GIF1 strain, which is highly susceptible to several antibiotics including chloramphenicol (37), was first transformed with the pBAD construct and selected on Luria-Bertani (LB) medium supplemented with 50 g/ml ampicillin, 3.4 g/ml chloramphenicol, and 0.5% glucose. Bacteria were cultured overnight in this medium at 37°C and the culture was then diluted 1:100 and used to inoculate 3 ml of medium. When the OD 600 reached 0.9, we diluted the suspension appropriately and layered 2 ϫ 10 4 of bacteria in 100 l on 30 ml of solid LB supplemented with 50 g/ml ampicillin, actinonin (0.1-3 M), and either glucose (0.5%) or arabinose (0.0002-0.2%) in Petri dishes. The minimum inhibitory concentration was defined as the lowest concentration of actinonin causing no growth after 18 h of incubation at 37°C.

Cell Biology, Transient Expression of GFP in Homologous and Heterologous Systems
Homologous Studies-Mammalian cell lines were cultured in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% fetal calf serum (Invitrogen) with Glutamax-I (Invitrogen) at 37°C, 5% CO 2 . HEK293 cells were plated on poly-L-lysine (Sigma)-coated cover slips 12 h before transfection and cultured in 6-well plates in fresh medium at 37°C, 5% CO 2 . Cells were transfected in the 6-well plates according to the calcium phosphate transfection protocol (CalPhos mammalian transfection kit; Clontech) with 2 g of plasmid DNA/well. Cells were washed with phosphate-buffered saline after 7 h to remove plasmid DNA and fixed in 4% formaldehyde/4% sucrose 48 h posttransfection for localization studies. Fixed cells were permeabilized with 0.05% Triton X-100 for 5 min and stained with 4Ј-6-diamidino-2phenylindole (DAPI, 15 mg/ml; 1:10000 dilution; Molecular Probes) for nuclear DNA. Coverslips were mounted in Vectashield (Vector Laboratories, Inc. CA) for further examination at the confocal microscope. Cells were imaged by a Leica SP confocal microscope through a ϫ100 1.4 NA Planapochromat oil immersion objective. eGFP was excited by a 488 line of an Argon laser and RFP by a 543 nm line of Green HeNe laser. In order to avoid bleed-through, the fluorophores were excited sequentially. The emitted fluorescence was collected separately through a triple dichroic mirror 488/543/633. The emission filter bands for GFP and RFP were restricted to 500 -568 nm and 570 -705 nm, respectively. DAPI staining of nuclear DNA was excited by a 351 nm line of an UV laser and emission fluorescence collected by a 396 -508 bandpass filter. Stacks of confocal sections separated by 0.2 m increments were taken and images analyzed by Metamorph 5.0 software (Universal Imaging Corporation).
Heterologous Studies-We bombarded onion epidermal cells with DNA constructs using the PDS-1000/He instrument (Bio-Rad) as previously described (7). Transient GFP production was examined with an up-right Axioplan 2 imaging fluorescence microscope (Zeiss) with interferential contrast and CCD camera (Sony ICX285). GFP was excited at 460 -480 nm and collected at 505-530 nm (Chroma Technology filters). Images were analyzed by Metamorph 5.0 software.

Protein Analysis
Purification of Peptide Deformylase Variants-BL21-pRares (Rosetta) cells (Novagen) expressing a given plasmid construct of the pET series were cultured to an OD 600 of ϳ0.9 at 22°C for 8 -12 h in 2ϫ TY medium supplemented with 50 g/ml ampicillin and 34 g/ml chloramphenicol. Cells were induced by incubation with 0.4 mM isopropyl-1thio-␤-D-galactopyranoside for another 5-12 h, with shaking. The cells were harvested by centrifugation, and resuspended in 10 -20 ml of buffer A, consisting of 20 mM sodium phosphate buffer pH 7.3 plus 10 mM 2-mercaptoethanol. The samples were subjected to sonication, and cell debris removed by centrifugation. The supernatant (5-15 ml) was applied to a Hi-Trap chelating HP (0.7 by 2.5 cm; Amersham Biosciences) nickel affinity column equilibrated in buffer B (buffer A plus 0.5 M NaCl). The sample was eluted at a flow rate of 0.5 ml/min, in two steps, with buffer C (buffer B plus 0.5 M imidazole) at concentrations of 0.17 and 0.5 M imidazole, respectively. The purity of the protein was monitored by SDS-PAGE (13%) and the final yield was 3 mg for a 400 ml culture. The pooled purified PDF preparation (5 ml) was first dialyzed against buffer A for 12 h and then against buffer A plus 55% glycerol for 24 h before storage at Ϫ20°C. Protein concentration was measured with the Bio-Rad protein assay kit. Bovine serum albumin was used as the protein standard.
Preparation of Cell Extracts-Exponentially growing human cells were collected and frozen at Ϫ80°C. Cells were then resuspended in 200 l of 50 mM Hepes (pH 7.5), 150 mM NaCl, 15 mM MgCl 2 , 1% Triton, 5% glycerol, 10 Ϫ4 M phenylmethylsulfonyl fluoride, and anti-protease mixture (Roche Applied Science) and disrupted in a MM 300 mixer mill (Qiagen). The sample was centrifuged and the supernatant collected for further analysis. Arabidopsis thaliana leaf and root tissues were obtained from 10-day-old seedlings. Samples were frozen in liquid nitrogen and prepared as previously described (10). Protein concentrations were measured with the BC assay protein quantification kit (Uptima). Bovine serum albumin was used as the protein standard.
Enzyme Assays-We used an assay coupling PDF activity and formate dehydrogenase activity to assess PDF activity. We monitored, at 37°C, the absorbance at 340 nm of NADH (⑀ M ϭ 6,300 M Ϫ1 ⅐cm Ϫ1 ), essentially as previously described (38). The reaction was started by adding 5-15 l of purified enzyme. In each case, the kinetic parameters were derived from iterative non-linear least square fits of the Michaelis-Menten equation, using the experimental data (39) as previously described (11).
Determination of Metal Ions by Atomic Absorption Spectroscopy-Protein samples (200 -400 l) were dialyzed overnight against a buffer (1 liter) consisting of 20 mM Hepes (pH 7.5) and 0.1 M KCl. We used a Varian AA220 spectrophotometer equipped with an air-acetylene burner in "peak height" mode and measured atomic absorption at 213.9 nm for 5 s after the injection of 0.1 ml samples. The concentrations of metal ions in serial dilutions of the enzyme samples were calculated by comparison with serial dilutions of standard ZnCl 2 solutions (Merck).

RESULTS
Identification in Animal cDNA Libraries and Cloning of Homologs of Plant Mitochondrial PDF1A and MAP1D-We pre-viously reported the identification in human and mouse of cDNAs encoding homologs of PDF1A (HsPDF and MmPDF) corresponding to a single gene located on chromosome 16 in human (locus NT_019608; Unigene Hs.130849) and 8 in mouse (locus NT_039467; Unigene Mm.246236, previously Mm.34708). We used the sequences corresponding to HsPDF and MmPDF to screen various metazoan cDNA libraries. We identified and annotated cDNAs encoding HsPDF homologs in most major vertebrates species (4 different homologs in fish, 2 in amphibians, 1 in birds, 3 in mammals). We also identified 4 HsPDF homologs in invertebrates: 3 in insects and one in Hydra magnipapillata (Cnidaria). We identified no PDF homologs in nematodes and fungi. The various sequences were aligned and compared with other PDF sequences ( Fig. 1; see "Experimental Procedures" for numbering of PDFs). The various animal PDF sequences were very similar (shown in pink in Fig. 1) and all belonged to the PDF1A class, which includes mitochondrial PDFs from plants and PDFs from Actinobacteria (9).
The occurrence of expressed PDF homologs in most animal genomes suggested that there should be a mitochondrial NME. If this were indeed the case, then we would expect dedicated MAPs to be identified in the same organism. It should be borne in mind that (i) cytoplasmic MAPs (MAP1As) occur in all eukaryotes and have a long N-terminal extension with a zinc finger domain involved in ribosome tagging (18) and (ii) organellar MAPs (MAP1D) have been identified in plants, in Apicomplexa and in Mycetozoa; they resemble MAP1As but have no zinc-finger in their N-terminal domain (7). We used plant organellar MAP1Ds to screen various cDNA libraries and identified MAP1D homologs in vertebrates (humans, pigs, cattle, mice, frogs, fish, and chickens) and in insects. The fulllength cDNA for human MAP1D (HsMAP1D) was cloned from a fetal cDNA library. Its ORF (GenBank TM AY374142; see "Supplementary Materials": Fig. 1) was very similar to that of a full-length cDNA recently isolated from mouse (AK010067) and identical to that deduced from a single gene identified on human chromosome 2 (locus NT_005348; Unigene Hs.406338). Similarly, a single MAP1D gene was identified in the complete genomes of the fish Fugu rubripes and the mouse Mus musculus (chromosome 2; locus NT_039208; Unigene Mm.27075). All known MAP1D genes map to loci different from those encoding MAP1As (chromosome 4 in human and 3 in mouse). One MAP1 pseudogene was also identified in each of the human (chromosome 12) and mouse (chromosome 9) genomes. We used the full-length MAP sequence and our previously identified MAP1As and bacterial MAPs (classified here as MAP1Bs) to construct a phylogenetic tree (Fig. 2). We identified three main branches: (i) bacterial MAP1Bs, found in all bacterial groups (ii) cytoplasmic MAPs (MAP1A) found in all eukaryotes, as previously reported (7) and (iii) MAP1D, found in most eukaryotes but not in fungi (Ascomycota) and nematodes. Actinobacteria has at least two MAP1s: one related to bacterial MAPs (MAP1B) and the other strongly related to MAP1D. This phylogeny is similar to that of PDF1A (see Fig. 1 and Ref. 9). These data suggest that most animal genomes express genes encoding specific machinery for an NME other than the cytoplasmic NME.
Both HsPDF and HsMAP1D Are Located in the Mitochondria-Using the available prediction software at www.expasy. org/tools/, we identified a clear putative mitochondrial targeting signal in the amino acid sequence of HsMAP1D, at position 1-43 (leader peptide LP). We investigated the location of the protein and whether the LP did indeed target the protein to the organelle by fusing sequences encoding the LP and the fulllength HsMAP1D to that encoding GFP. The resulting constructs were used to transfect cells in a heterologous expression system. Both fused proteins were detected in the mitochondria (Fig. 3C), confirming that the putative sequence LP-HSMAP, identified by consensus structural motifs in the predicted amino acid sequence, is genuinely responsible for protein targeting to mitochondria.
In contrast to MAP1D, the prediction software failed to identify a classical mitochondrial signal peptide in HsPDF. However, such software did identify a consensus mitochondrial localization motif composed of the first ϳ20 amino acids in all the other mammalian PDFs. We therefore tried to express full-length (FL) and various C-terminal-deleted forms of HsPDF (Fig. 3A) fused to GFP in a heterologous expression system (i) to demonstrate clear mitochondrial location of the protein and (ii) to define the minimal structural motif controlling HsPDF localization. In transient expression studies, FL-HsPDF/GFP and the C-terminal-deleted forms LP4-HsPDF/ GFP, LP5-HsPDF/GFP and LP1ϩLP2-HsPDF/GFP (Fig. 3A) were located primarily within mitochondria (Fig. 3B). In contrast, the LP1-HsPDF/GFP fusion protein, containing the predicted mitochondrial leader peptide clearly displayed diffuse  (7) were used as input probes for BLAST (59) searches of the data at NCBI. Matching cDNA sequences were compared and aligned if an overlap was detected. The genomic fragments were analyzed with various software packages for the prediction of intron-exon splice sites and amino acid sequence similarity, alignment, and comparison of the cDNAs. The deduced amino acid sequences are shown. Translated sequences were aligned with ClustalX (60) and then by eye. A series of question marks indicates that the corresponding nucleotide sequence is missing. The numbering of each of the three amino acid sequences is indicated below the sequences for each block of 100 residues. Strictly conserved residues within the catalytic core are shown in red or in green for the ligands of the catalytic metal. Residues in pink are conserved in PDF1As. cytoplasmic staining (Fig. 3B). We therefore concluded that the first 44 amino acids were not sufficient to target the protein to the organelle and that the minimal fragment containing all the information required for the efficient targeting of HsPDF to mitochondria corresponded to the first 69 amino acids (LP1ϩLP2-HsPDF). We characterized the leader peptide further by investigating whether the region between amino acids 44 and 69 (LP2-HsPDF) was sufficient for correct localization by creating a construct encoding the LP2-HsPDF fragment fused to GFP. In this case, we obtained a cytoplasmic signal (Fig. 3B). Taken together, these data show that the region encompassing at least LP1ϩLP2-HsPDF is necessary and sufficient for mitochondrial localization.
The results obtained with the heterologous system guided us to set up the localization experiments in a homologous expression system. Human embryonic kidney (HEK) cells were transfected with constructs encoding GFP fusions with (i) the full-length MAP1D and PDF and (ii) the LP-HsMAP1D and the LP1ϩLP2-HsPDF sequences (Fig. 4). A number of control experiments were carried out in both cases: (i) labeling of the nucleus with DAPI (Fig. 4, A-B) and (ii) cotransfection with a construct encoding a mitochondrial control fused to red fluorescent protein (mt/RFP; Fig. 4C). No bleed-through was observed as confirmed by the absence of colocalization if cells are not cotransfected with mt/ RFP (Fig. 4D). This confirmed the colocalization shown in Fig.  4C. Both HsPDF and HsMAP1D were clearly specifically targeted to the mitochondria by their N-terminal extensions, LP-HsMAP1D and LP1ϩLP2-HsPDF, respectively.
Expression in Escherichia Coli of the Human Peptide Deformylase Homolog, HsPDF-The catalytic centers of animal MAP1Ds resemble those of their counterparts in bacteria and plants ( Supplementary Figs. 1 and 16) and is therefore expected to display MAP activity. In contrast, animal PDFs differ from phylogenetically related plant PDF1As in several features in addition to the conserved N-terminal extension (Fig. 1). We decided therefore to investigate the functional impact of these features. Prior to this, purified HsPDF was needed to raise antibodies against it and analyze the protein form accumulating in vivo.
The full-length ORF of HsPDF was fused to a His 6 tag in FIG. 2. A phylogenetic tree for MAPs. 77 MAP1 sequences were selected as representative of sequence diversity. These sequences were aligned and the tree constructed. The phylogenetic tree was constructed with N-J Tree (60)   various E. coli plasmids. We were unable to detect the protein in either the soluble or the insoluble fraction in all conditions tested. The first 380 base pairs (codons 1-126 of 244, corresponding to amino acids Ϫ62 to 64 in HsPDF) of the HsPDF ORF have a GϩC content of 77% and include 18 codons corresponding to the rarest tRNAs in E. coli (40). Bacteria overproducing a large number of rare tRNAs were therefore transformed with pET derivatives encoding the HsPDF ORF. Unfortunately, this new system did not increase HsPDF production. We therefore redesigned the HsPDF ORF fragment to create a new sequence encoding the same polypeptide, but with codon usage optimized for E. coli (40) and a lower GϩC content and fewer long GϩC stretches (see details under "Experimental Procedures"). Time-course analysis of expression of modified HsPDF ORF in E. coli revealed that the 25 kDa full-length protein was produced, but was subject to proteolysis. Two shorter, more stable fragments of 23 and 20 kDa accumulated in the cell extract throughout induction (Fig. 5, A and C), the shorter predominating in the long term (data not shown). Western blot analysis with anti-His tag antibodies revealed that these fragments corresponded to polypeptides with deletions of ϳ20 -50 N-terminal residues (Fig. 5A). We carried out deletion analysis to characterize these fragments in more detail and to obtain a single, stable form. Constructs encoding HsPDF with N-terminal deletions of 19, 39, 45, 62, and 78 residues (designated ⌬N19, ⌬N39, ⌬N45, ⌬N62, and ⌬N78, respectively; see Fig. 1) were expressed. In all cell extracts except that for ⌬N78, the corresponding protein accumulated normally in the soluble fraction of the crude extract (Fig. 5A). As residue 62 can be aligned with residue 1 of EcPDF, the catalytic domain of HsPDF was predicted to start immediately after this residue. In EcPDF, N-terminal deletions of more than 3 residues result in incorrect folding and inactivation of the enzyme (27). Consistent with this, the ⌬N78 form proved to be unstable and did not accumulate significantly in the soluble fraction (Fig. 5A).
Characterization of the HsPDF Form from Human Mitochondria; Comparison with Plant PDF1A-The various HsPDF deletion variants produced in E. coli were purified and antibodies raised against them. A single cross-reacting band of 22 Ϯ 1 kDa was detected on Western blot analysis of various human cell extracts (Fig. 5B). A band of similar size was also detected in crude extracts from simian kidney cell lines (Fig. 5B), in keeping with the strong conservation of the amino acid sequence of animal PDFs (Fig. 1). The HsPDF that accumulated in human cells corresponded to the ⌬N39 form (Fig. 5C). This indicates that part, but not all of the 60-residue presequence required for import into the mitochondrion is cleaved by mitochondrial leader peptidase (shaded region in Fig. 1). Because the ⌬N39 form displays an additional 6-His C-terminal tag not present in the native HsPDF, our data suggest that ϳ25-30 conserved residues found at the N terminus of all animal PDFs (Fig. 1) are thus retained in the form accumulating in the mitochondria.
A different situation was observed for plant PDF1A. Western blot analysis with various plant extracts and antibodies against AtPDF1A revealed (Fig. 5D) that cleavage occurred just downstream, next to the catalytic domain (shaded region in Fig. 1). We therefore concluded that the PDF accumulating in human mitochondria contained an N-terminal domain of ϳ25-30 residues not present in bacterial PDFs and plant PDF1As.

HsPDF Has Peptide Deformylase Activity in Vivo and in Vitro and Its Extra N-terminal Domain Is Involved in Protein
Stability and Catalytic Activity-We investigated the deformylase activity of HsPDF by trying to complement the def(Ts) phenotype of strain PAL421Tr (32). Complementation was observed with the ⌬N39, ⌬N45 and ⌬N62 forms (Fig. 6A) but not with the full-length form. As a control, we replaced the crucial Glu of motif 3 (HEXXH) by an Ala in ⌬N62, which prevented complementation. This effect is similar to that observed with bacterial PDFs (26). The full-length HsPDF did not accumulate in the bacterium, whereas the shorter deletion mutants did (Fig. 6B), accounting for the lack of complementation observed with the full-length form. The full-length form was found to be toxic to the bacterium in the long term.
The full-length, ⌬N19 ⌬N39, ⌬N45, and ⌬N62 forms were characterized. The full-length, ⌬N19, and ⌬N39 variants were very sensitive to proteolysis and the most active variant was the ⌬N39 form, corresponding to the form that accumulates in vivo (Table I). Proteolysis of the N terminus led to cleavage in the Ϫ10/Ϫ15 region. Flanking sequences containing Lys or Arg, preceding regions containing Pro-Glu-Ser-Thr, are commonly found in proteins with high turnover (41). The extra N-terminal domain of HsPDF also displays such a structure (Fig. 1). We therefore constructed two ⌬N39 variants with substitutions of the two conserved twin-Arg (Table I). These substitutions  6. Complementation of strain PAL421Tr by HsPDF. pBAD plasmids encoding the corresponding His-tagged version of the indicated PDF were used to transform strain PAL421Tr at 30°C. Strains were streaked out in parallel at 42°C on LB Petri dishes and on LB medium containing the optimal concentration of the PDF expression inducer arabinose (0.02%), with the exception of the strain producing EcPDF, which was streaked in the presence of glucose (0.5%) to reduce its production to a level similar to that for the other PDF forms. Control corresponds to the empty cloning vector pBAD. Panel A, image of the Petri dish after 24 h of growth. Panel B, Western blot analysis, with anti-His tag antibodies, of the cell extract. resulted in forms that displayed increased resistance to Nterminal proteolysis in vivo (Fig. 5E).
The purified enzyme was found to have a modest catalytic activity (Table I). We noticed that the final specific activity was improved by the addition of preservative agents such as nickel anion salts (23), cobalt (42) or a mixture of the strong reducing agent TCEP and catalase (21) during purification (Table I).
Despite testing the effects of many different conditions, including incubation with various metal cations, we were unable to preserve full activity, as already reported for some bacterial and plant PDF1Bs (TtPDF and LePDF; Refs. 11 and 23). The purified ⌬N39 form contained zinc bound in a stoichiometric manner, suggesting that zinc had replaced the physiological metal cation, which was probably iron, as in PDF1Bs (11,20,43).
Thus, functional HsPDF contains an unstable metal cation and an additional N-terminal domain that is retained in the mature protein that accumulates in mitochondria. This domain is important for activity and import into mitochondria.

Site-directed Mutagenesis of HsPDF and Plant PDF1As Reveals the Importance and Functional Linkage of Changes at
Residues 43 and 91-We next addressed the distinction in metal binding between animal and plant PDF1As, using HsPDF and AtPDF1A as prototype examples. The most striking difference concerns the strictly conserved motifs. First, motif 1 (Gly 43 -⌽-Gly-⌽-Ala 47 -Ala-X-Gln 50 , where ⌽ is a hydrophobic amino acid) is not conserved: the Gly is replaced by a Cys and the Ala by a Ser at positions 43 and 47, respectively. Second, the Leu 91 of motif 2 (Glu 88 -Gly-Cys-Leu 91 -Ser) is systematically replaced by another residue: Met, Ala, Ser, or, most frequently, Glu.
Site-directed mutagenesis targeting the unusual residues of motifs 1 and 2 of HsPDF was carried out to generate residues in these positions identical to those in the highly stable plant AtPDF1A. Substitution of two residues, in motifs 1 and 2 (C43G and E91L) was sufficient to render the enzyme as active as the plant form and insensitive to nickel cations or the addition of TCEP (Table II). A single substitution was insufficient, indicating a functional link between residues 43 and 91. Substitution at residue 47 had little effect. Reciprocally, a single substitution of residue 43 or 91 in the A. thaliana PDF1A (AtPDF1A), converting the corresponding residue to that found in HsPDF, dramatically decreased activity and dependence on preservative agents (Table II). Various residues other than Glu are found in position 91 in PDFs from animals (Fig. 1). However, the replacement of Glu 91 by Met (as in insects) or Ala (as in fish) had little effect on HsPDF (Table II).
We therefore conclude that the physiologically active form of HsPDF coordinates an iron cation, rather than the zinc ion coordinated by plant PDF proteins. The evolutionary substitutions at positions 43 and 91 in animal PDFs are responsible for this effect.
Substrate Specificity of HsPDF and Sensitivity to Common PDF Inhibitors Including Actinonin; Comparison with Other PDFs and Role of Residue 125-The substrate specificity of HsPDF was investigated and compared with those of AtPDF1A, and PDF1Bs (i.e. AtPDF1B and EcPDF) (Table III). Both HsPDF and AtPDF1A hydrolyzed peptides with bulky P1Ј chains much less efficiently than EcPDF. This lower efficiency probably results from the smaller size of the S1Ј pocket including the conserved residue specific to all PDF1As, Trp 125 . This conclusion is consistent with the behavior of the mutant forms at position 125 (Table II and Table III). However, like PDF1Bs and unlike plant PDF1As, animal PDF was insensitive to the nature of hydrophobic bulky P3Ј chains (Table III). This effect probably results from the differential impact of residue 87 (one of the two residues, together with Arg 97 , involved in the S3Ј subsite; Ref. 24) in both PDF1A subclasses. Indeed, residue 87 is a conserved bulky hydrophobic Phe in plant PDF1As and a small Pro in animal PDFs; Pro imposes no specific structural restraint on the subsite whereas Phe, which is much larger, does (Fig. 1).
We assessed the sensitivity of HsPDF to various inhibitors in vitro (Table IV). HsPDF was not inhibited by compound FRN but was efficiently inhibited by actinonin. This is consistent with animal PDF displaying restricted selectivity for bulky P1Ј side chains. We investigated whether actinonin inhibition could be observed in vivo. The HsPDF ORF was inserted into an arabinose inducible vector, which was then used to transfect tolC actinonin-permeable cells. Resistance to actinonin was then assessed. The overproduction of HsPDF increased the resistance of the bacterium to actinonin (data not shown). Thus, HsPDF, like AtPDFs (11), is the primary target of actinonin in vivo.

DISCUSSION
The Two Components of the Mammalian NME Have Unusual but Similar Phylogenies-In this study, we found that most animal cells produce the two enzymes required for mitochondrial N-terminal methionine excision and that the corresponding mature forms are localized to the correct cell compartment. These two enzymes are a PDF and a dedicated methionine aminopeptidase (MAP1D). At the time we submitted this article, similar data concerning the ⌬N62 form of a The substrate was Fo-Met-Ala-Ser. The value of k cat /K m is given as a percentage of the value obtained with variant ⌬N39 purified in the presence of TCEP-catalase. b A mixture of various forms including the full-length (FL) and N-terminally truncated forms (see Fig. 5, panels A and C) occur in this protein preparation. c nm, not measurable because K m value is too high (Ͼ10 mM). d ND, not determined.
HsPDF were released (44). Seven of the thirteen ORFs of the human mitochondrial genome are predicted to be substrates of HsMAP1D, according to the known rules of MAP substrate selectivity (reviewed in Ref. 1). The products of these ORFs are ATP synthase subunit 8, NADH dehydrogenase subunits 1, 4L, and 5, cytochrome c oxidase subunits 2 and 3 and cytochrome b.
The putative cleavage sites for MAP1D are conserved in all mitochondrially encoded ORFs in organisms in which a MAP1D has been identified. Both animal PDFs and MAP1Ds strongly resemble enzyme forms found in Actinobacteria, although mitochondrial genes are thought to have originated from Rickettsiae bacteria (45). Thus, both mitochondrial NME genes probably originate from an ancient horizontal gene transfer from an actinobacterium to an ancestor of eukaryotic cells. This event may be similar to that generating the second type of glutamine synthetases, as the phylogenies determined were similar (46), but is probably unrelated to mitochondrial gene transfer to the nucleus. Our present identification of a mitochondrial MAP (MAP1D) confirms the predictions of Keeling and Doolittle (47). According to the branching pattern of the phylogenetic tree (Fig. 2), cytoplasmic MAP1As were probably derived from MAP1Ds by gene duplication and fusion to a ribosome-targeting domain.
A Functional and Regulatable NME System in Human Mitochondria-Mitochondrial NME has been shown to function in vivo in plants by (i) the discovery of dedicated enzymatic machinery cleaving the initial Met (7) and (ii) the compilation  (11). AtPDF1B and LePDF1B were in variant ⌬N80 (11). PfPDF corresponds to a ⌬N63 variant (12). The substrate was Fo-Met-Ala-Ser. The catalytic efficiency measured with each wild-type PDF was set at 100. b -, corresponds to the amino acid sequence without any substitution. c All enzymes were prepared without adding preserving agent, unless otherwise stated. (TC), a mixture of TCEP and catalase was added. (Ni), NiCl 2 was added. The respective concentrations of each compound were described previously (11,21   of a number of N-terminal protein sequencing data (reviewed in Ref. 9). Although phylogenetically related to plant PDF1As ( Fig. 1 and Refs. 7 and 9), the possible existence of an active peptide deformylase in animals has been a controversial issue (6,7,13,44). One objection raised was that the sequences of mitochondrially encoded proteins from mammals indicate that their Fo groups are retained, whereas this is not the case for the corresponding proteins of plant mitochondrial genomes (see data compiled in Ref. 13). This suggested that there was no PDF and MAP activity in animal mitochondria. However, the same tissues (adult bovine heart) were always used to prepare mitochondria. Moreover, the method used to determine whether an Fo group was present was indirect. It was suggested that the protein was N-formylated because mild acid treatment of the N terminus exposed the terminal residue for Edman sequencing. Direct mass spectrometric analysis of the N termini of mitochondrial proteins from several tissues would be required to draw definitive conclusions. The identification in this study of MAP and PDF enzymes in animals strongly suggests that NME is functional in mitochondria. Interestingly, the crystal structure data of cytochrome bc 1 (48) lack the electron density of the N-terminal Met of the mitochondrially encoded cytochrome b, suggesting Fo-Met removal. This result is consistent with NME activity as the cytochrome b sequence starts with Fo-Met-Thr, which strongly favors Met removal (1).
In this study, we provide compelling evidence that HsPDF is active in vivo, as it allows complementation of a PDF gene (def) defect, and in vitro, although its activity is highly unstable (Table I).
A recent study showed that human cells are resistant to the PDF-specific inhibitor actinonin and its derivatives, if used at concentrations with antibacterial effects (49). However, if used at higher concentrations, these inhibitors have a toxic, antiproliferation effect in vivo and in vitro, the intracellular target of which is unknown (50,51). Possible effects of inhibition of the cell surface zinc aminopeptidase CD13 and other metzincins has been excluded. The most likely target is therefore HsPDF, suggesting that PDF is functional in humans. The PDF gene is expressed in many mammalian cell types (Fig. 5B) and a similar gene is expressed in most animal genomes (Fig.  1). In all these genomes, genes encoding a dedicated mitochondrial MAP (MAP1D) can be identified (Fig. 2). As the function of PDF is to facilitate MAP activity (52), these data indicate the genetic and functional linkage of the two components of mitochondrial NME in animals. Finally, it was recently shown that the PDF gene is part of a cluster that is tightly regulated in skeletal muscle regeneration in mice (53), with down-regulation after injury and up-regulation during cell differentiation and re-entry into the cell cycle (see H3113G11 in cluster 7). This is consistent with NME activity being tightly controlled in animal mitochondria.
Animal PDFs Are Unusual PDF1s with Non-random Substitutions in the Conserved Motifs-Unlike HsMAP1D, animal PDFs differ markedly from the many other PDFs characterized to date. HsPDF has an additional 20-amino acid N-terminal extension (Figs. 1 and 5). This extension is strongly conserved, predicted to fold as an ␣-helix (www.embl-heidelberg.de/predictprotein/predictprotein.html, Ref. 54), has a very high pI due to the presence of five conserved arginines (Fig. 1). It is separated from the rest of the molecule by a region with high local concentrations of Pro, Glu, Ser, and Thr. This suggests a high turnover of the protein in vivo. Interestingly, all the arginine residues occur on one side of the helix and the hydrophobic residues occur on the other. The N-terminal domain may fold back toward the E1 extended structure of the enzyme stabilizing the active site. This would provide a role for this domain in deformylase activity. However, we cannot exclude the possibility that this domain is also involved in binding to mitoribosomes, like the additional N-terminal domain of MAP1As to cytosolic ribosomes. All animal PDFs display systematic substitutions altering crucial amino acids in motifs 1 and 2: Gly 43 , Ala 47 , and Leu 91 (Fig. 1). Our results have shown that the Cys 43 and Glu 91 substitutions are responsible for the unusual behavior of HsPDF, as the replacement of these two residues by Gly and Leu, respectively, makes HsPDF as active as plant PDF1As (Table II). The side chains of residues 43 and 91 are located in close proximity (within 4.5 Å) in the crystal structure of known PDF1s (24). They are involved in recognition of the Fo group and are located close to the P2Ј binding site. There is unlikely to be a true S2Ј binding pocket because PDF has no marked specificity for a given P2Ј side chain (28,55). The other residue close (3.7 Å) to residues 43 and 91 is residue 41, a Glu in E. coli and an Arg in human PDF. We further investigated the three-dimensional structure of this region, by constructing a model of the active site of HsPDF (Fig. 7). This model suggests that a salt bridge could be created in HsPDF between the side chains of residues 41, 43, and 91. The alignment in Fig. 1 shows that if residue 91 is a Glu, then residue 41 is positively charged. Conversely, if residue 91 is another non-charged residue, as in amphibians, then residue 41 is uncharged. We conclude that the substitutions of residues 41, 43, and 91 in animal PDFs are not random and that these substitutions are interdependent. This may make it possible to fine tune PDF activity in mitochondria, possibly facilitating adaptation to metal availability or to the highly oxidative environment in this organelle. Finally, we cannot rule out the possibility that these three residues interact with the N-terminal extension of HsPDF.
Implications of the Existence of Human PDF for PDF Inhibitor-based Antibacterial Treatments-Bacterial PDF inhibitors (PDFI) constitute one of the most promising families of new antibiotics (6, 56, 57). However, our characterization of a functional PDF in humans raises a major concern as to the use of PDFI in human medicine. Clearly, the unambiguous mitochon- FIG. 7. A putative electrostatic interaction in the active site of HsPDF. The amino acid sequence of HsPDF was aligned with that of EcPDF, using InsightII software (Accelrys). The three-dimensional structure of PDF bound to actinonin (24) was used to construct several three-dimensional models of HsPDF, with the homology modeler module. The lowest energy structure of HsPDF was further minimized with the CHARMM forcefield and superimposed on the structure of actinonin bound to EcPDF. A close-up of the deep end of the active site in the vicinity the metal cation is shown. A schematic diagram of the proposed electrostatic interaction between the cysteine Cys 43 side chains, and the Glu 91 and Arg 41 side chains is shown. The metal cation (Me 2ϩ ) is shown as a gray sphere, and the substrate analog actinonin is shown in purple. The distance between two heavy atoms is indicated as a white line, above which the corresponding value is indicated. Hydrogen atoms are not shown. drial location of HsPDF lessens this concern to some extent and accounts for the high concentration of PDFI required to obtain a toxic effect. Millimolar amounts of PDFI are also required to inhibit plant growth (11). Mitochondrial membranes are less permeable to small molecules than most other membranes and act as an efficient supplementary filter for many drugs. For this reason, a number of antibiotics (tetracycline, macrolides, etc.) that inhibit mitochondrial targets are nevertheless currently used in human medicine. However, as PDFI toxicity was observed (see above), the use of PDFI might cause side effects in the long term due to inhibition of HsPDF. We therefore think that great care should be taken when designing PDFI to reduce the potency of these molecules against HsPDF. Our data are interesting in this respect because they show that animal PDFs are intermediate between PDF1As and PDF1Bs in terms of biochemical properties. Like PDF1Bs, animal PDFs are highly sensitive to preservative agents, indicating that they are highly unstable iron-coordinated molecules whereas plant PDF1As are zinc-coordinated (11). This suggests that the nature of the optimal PDFI chemical head, a formylhydroxylamine as in BB-3497 (58) or a hydroxamate as in actinonin, that binds the metal cation cannot be modified. Animal PDFs, like PDF1Bs but unlike plant PDF1A, are stimulated rather than inhibited by the presence of a hydrophobic P3Ј side chain (Table III). Conversely, animal PDFs resemble plant PDF1A in not tolerating a bulky hydrophobic group at position P1Ј (Tables III-IV) This intolerance results from the presence of a Trp rather than a Leu at position 125, causing a decrease in the size of the S1Ј pocket of the enzyme (Tables II-IV). Thus, modification of the P3Ј group in PDFI will not result in discrimination between PDF1B and HsPDF. Instead, replacing the P1Ј side chain with a bulkier group, such as a phenyl group, dramatically decreased the affinity of PDFI for HsPDF, without affecting their potency with respect to bacterial PDFs.