The human nm23-H4 gene product is a mitochondrial nucleoside diphosphate kinase.

We demonstrate here the catalytic activity and subcellular localization of the Nm23-H4 protein, product of nm23-H4, a new member of the human nm23/nucleoside diphosphate (NDP) kinase gene family (Milon, L., Rousseau-Merck, M., Munier, A., Erent, M., Lascu, I., Capeau, J., and Lacombe, M. L. (1997) Hum. Genet. 99, 550-557). Nm3-H4 was synthesized in escherichia coli as the full-length protein and as a truncated form missing the N-terminal extension characteristic of mitochondrial targeting. The truncated form possesses NDP kinase activity, whereas the full-length protein is inactive, suggesting that the extension prevents enzyme folding and/or activity. X-ray crystallographic analysis was performed on active truncated Nm23-H4. Like other eukaryotic NDP kinases, it is a hexamer. Nm23-H4 naturally possesses a serine residue at position 129, equivalent to the K-pn mutation of the Drosophila NDP kinase. The x-ray structure shows that the presence of Ser(129) has local structural effects that weaken subunit interactions. Site-directed mutagenesis shows that the serine is responsible for the lability of Nm23-H4 to heat and urea treatment, because the S129P mutant is greatly stabilized. Examination of human embryonic kidney 293 cells transfected with green fluorescent protein fusions by confocal microscopy shows a specific mitochondrial localization of Nm23-H4 that was also demonstrated by Western blot analysis of subcellular fractions of these cells. Import into mitochondria is accompanied by cleavage of the N-terminal extension that results in NDP kinase activity. Submitochondrial fractionation indicates that Nm23-H4 is associated with mitochondrial membranes, possibly to the contact sites between the outer and inner membranes.

Nucleoside diphosphate (NDP) 1 kinases (EC 2.7.4.6.) are ubiquitous enzymes responsible for the exchange of ␥-phosphates between tri-and diphosphonucleosides (1). The catalytic reaction follows a ping-pong mechanism in which the enzyme is transiently phosphorylated on a histidine residue conserved in all nucleoside diphosphate kinases. NDP kinases are crucial for the homeostasis of cellular nucleoside di-and triphosphate composition and may control many cellular functions.
More than 60 NDP kinases primary sequences and a number of x-ray structures from prokaryotes to eukaryotes have been reported. Except in a few bacterial species, NDP kinases are hexamers. All have very similar three-dimensional structures, and their subunits retain a characteristic fold with five helices packed on a four-stranded antiparallel ␤-sheet (2). In humans, six nm23 genes coding for NDP kinases or for homologous isoforms have been identified: nm23-H1 (3), nm23-H2 (4), DR-nm23 (5), nm23-H4 (6), nm23-H5 (7), and nm23-H6 (8). The first gene, nm23-H1, was initially identified as a putative metastasis suppressor gene (9,3). It codes for NDP kinase A, which is 88% identical to NDP kinase B, encoded by nm23-H2. NDP kinase B, also known as PuF (10), binds to the promoter of the c-myc proto-oncogene and activates its expression in vitro and in vivo (10,11), independently of the NDP kinase activity (12). Isoforms A and B can form homo-or heterohexamers, resulting in different ratios of the respective subunits (13). The product of the third gene, DR-nm23, inhibits granulocyte differentiation and induces apoptosis of 32Dc13 myeloid cells (5). The nm23-H4 gene was identified in our laboratory (6), and the Nm23-H4 protein, the object of the present study, possesses characteristics for mitochondrial targeting. Recently, we also identified another nm23 homologous gene, nm23-H5, specifically expressed in testis germinal cells (7). nm23-H6 is ubiquitously expressed and does not possess a specific signal sequence for organelle import or vesicular trafficking (8).
Regulatory functions have been attributed to these enzymes, in particular in differentiation, normal and tumoral development (reviewed in Ref. 14) and apoptosis (5). In Drosophila melanogaster, NDP kinase is the product of awd (abnormal wing discs), which is essential for development as mutations lead to larval lethality (15). A natural point mutation in awd is * This work was supported by funds from INSERM, Association pour la Recherche contre le Cancer, Ligue Française contre le Cancer (Nationale and Comité de Paris), and Fédération Nationale des Groupements des Entreprises Françaises dans la Lutte Contre le Cancer. 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.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank TM  called K-pn (Killer of prune). It induces Drosophila lethality in a dominant manner when associated with a null mutation of the prune gene involved in the fly eye pigment synthesis (16). The K-pn mutation substitutes a serine for a proline in a surface loop, named for this reason the K-pn loop, that makes subunit contacts in the hexamer and plays an important part in the stability of the hexamers (17,18). The proline residue is conserved in most NDP kinases, but it is a serine in Nm23-H4 (6) and in the NDP kinase of pigeon liver mitochondria (19).
NDP kinases have been found associated to different cellular compartments, such as the cytosol (1), the nucleus (20,21), the plasma membrane (22), and mitochondria (23,24). Most studies have concerned the cytosolic isoforms, products of the nm23-H1 and -H2 genes, and little is known about the mitochondrial enzyme(s). Studies based on measurement of activity point to a mitochondrial NDP kinase preferentially located in the matrix or in the intermembrane space depending on the tissues and the species (24,25). In the intermembrane space, NDP kinase couples ATP formation through the respiratory chain to the synthesis of the other nucleoside triphosphates (26). In the matrix, NDP kinase plays a role in the availability of nucleotides necessary for the Krebs cycle succinyl thiokinase and the catabolism of short chain fatty acids (27). The presence of NDP kinase would also be required for the synthesis of nucleoside triphosphates as precursors of mitochondrial nucleic acids, of the GTP necessary for the protein synthesis machinery, and of ATP used to fuel Hsp70 and to generate a pulling force for protein import (28).
Up to now, three cDNAs encoding mitochondrial NDP kinases have been identified. In the amoeba Dictyostelium discoideum, a gene encoding a mitochondrial NDP kinase has been described (29), clearly distinct from the gene encoding the cytosolic enzyme (30). Characteristics of the N-terminal extension prompted the authors to propose localization to the intermembrane space of mitochondria (29). A mitochondrial NDP kinase was reported in pigeon liver, specifically associated to the matrix (19). We have reported the identification of nm23-H4 cDNA encoding a protein homologous to NDP kinases that presents characteristics suggesting targeting to mitochondria (6). The present study was aimed at demonstrating the NDP kinase activity of the nm23-H4 gene product and its specific mitochondrial localization and investigating the role of the serine residue at position 129 of Nm23-H4 homologous to the Drosophila K-pn mutation by x-ray analysis and by sitedirected mutagenesis.

EXPERIMENTAL PROCEDURES
Materials-Transformed human embryonic kidney (HEK) 293 cells were obtained from the ATCC. Human recombinant A and B NDP kinases were purified as described previously (21,31,32). Hydroxylapatite (Bio-Gel® HT gel) was obtained from Bio-Rad, and DEAE-Sepha-cel® resin was from Amersham Pharmacia Biotech. Lactate dehydrogenase and pyruvate kinase were from Roche Molecular Biochemicals and Fluka, respectively. The antibodies were all raised against human proteins. The monoclonal anti-porin antibody (Calbiochem-Novabiochem) was a kind gift from Dr. C. Prip-Buus. The polyclonal anticytochrome c antibodies were obtained from Santa Cruz Biotechnology, and the monoclonal anti-cytochrome oxidase (subunit II) antibody was from Molecular Probes. The anti-manganese-superoxide dismutase (anti-Mn-SOD) antibodies were a kind gift from Dr. T. D. Oberlay. The affinity purified antibodies against the A and B NDP kinases were prepared using the recombinant NDP kinase A as described previously (21,33). They were not depleted of antibodies recognizing the NDP kinase B and therefore recognized both enzymes. The mitochondrionselective dye MitoTracker TM Red CMXRos was from Molecular Probes.
Plasmid Constructions-To express recombinant Nm23-H4 proteins, the cDNA was generated by polymerase chain reaction using the EcoRI fragment of clone 1.1 inserted into pBluescript SK Ϫ (6) as DNA template. We chose to express the full-length protein (Nm23-H4) and a truncated form (Nm23-H4⌬33) missing the first 33 amino acids that was designed to correspond to the length of the A and B NDP kinases (Ser 34 and Tyr 35 were changed to Met and Gly, respectively). Primers were designed to introduce NdeI and BamHI sites at the 5Ј-and 3Јends, respectively, of the coding sequence allowing for subcloning of inserts at the corresponding sites into pET-21b and pET-28a(ϩ) expression vectors (both from Novagen) for the expression of Nm23-H4 either native or fused to a hexahistidine tag at its N terminus, respectively. The fragment corresponding to the full-length protein (187 amino acids) was obtained using the two following primers: 5Ј-GAGCTAGACATAT-GGGCGGCCTCTTCTGG-3Ј and 5Ј-CATTGGATCCTTCAGGCTGG GTGGATGCT-3Ј. An insert corresponding to the truncated isoform (Nm23-H4⌬33) was obtained using the 5Ј primer 5Ј-GGGCTATACAT-ATGGGGACCCGGGAGCGGACC-3Ј.
The pEGFP-N1 vector (CLONTECH) was used to express the fulllength Nm23-H4 protein fused to the N terminus of the green fluorescent protein (GFP) in the HEK 293 cell line. The nm23-H4 cDNA was amplified by polymerase chain reaction using oligonucleotide primers that contained BamHI (5Ј-TTGGATCCCAGGCTGGGTGGATGCTGCT-GTGCTG-3Ј) and EcoRI (5Ј-ATGAATTC GCGTCATGGGCGGCCTCT-TCTGG-3Ј) sites, and the amplified DNA fragment was cloned into the BamHI-EcoRI sites of the pEGFP-N1 plasmid. The pEGFP-N1 plasmid expressing the native EGFP protein was used as a control.
Mutant constructs with a proline residue instead of a serine at position 129 were generated using the Transformer TM site-directed mutagenesis kit (CLONTECH). The mutagenesis was performed on the pET-28a(ϩ) and on the pEGFP-N1 constructs to express the Nm23-H4⌬33 S129P and the full-length Nm23-H4 S129P /EGFP mutant proteins, respectively. The mutagenic primer designed to produce the desired point mutation was 5Ј-CAGCCTCAGCCGGGTCGGTGTGTC-3Ј (the altered nucleotide is in boldface type). Plasmids were purified using purification kits from Qiagen and were endotoxin-free when used for transfection in mammalian cells.
The construct sequences were verified by automatic sequencing in both directions using an Applied Biosystems sequencing apparatus (Genome Express, Grenoble, France).
Expression and Purification of Nm23-H4 Proteins-The Nm23-H4 proteins were expressed in either native form (Nm23-H4⌬33) or fused to a hexahistidine tag (full-length and Nm23-H4⌬33). The plasmid constructs were transferred into the Escherichia coli strain BL21(DE3) (Novagen). A single colony was inoculated into LB culture medium supplemented with 100 g/ml ampicillin (pET-21b) or 30 g/ml kanamycin (pET-28a(ϩ)). The bacteria were grown to an optical density of 0.5 at 600 nm, and protein expression was induced by addition of 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside for 3 h at 37°C. The bacteria were harvested by centrifugation. The full-length and the truncated Nm23-H4 proteins, either tagged or untagged, were expressed as inclusion bodies and as soluble proteins, respectively. The Nm23-H4⌬33 protein (tagged or untagged) was purified under native conditions, whereas only the tagged form of the full-length Nm23-H4 protein was purified under denaturating conditions (8 M urea).
For purification of the untagged Nm23-H4⌬33, the bacterial pellet was resuspended in 50 mM Tris-HCl buffer, pH 8.7, containing 1 mM EDTA, 1 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, and the antiprotease mixture tablet from Roche Molecular Biochemicals (at the concentration indicated by the manufacturer) and was lysed by sonication. After centrifugation to remove the bacterial debris, the supernatant was loaded onto a DEAE-Sephacel column equilibrated with the same buffer. The flow-through, devoid of E. coli NDP kinase, was collected and dialyzed against 10 mM sodium phosphate buffer, pH 7, and loaded onto a hydroxylapatite column. The Nm23-H4⌬33 protein was eluted with a linear gradient of 10 -400 mM sodium phosphate. The fractions containing NDP kinase activity were pooled, dialyzed in phosphate-buffered saline, and concentrated using an Amicon cell equipped with a Diaflo YM10 filter. For storage at Ϫ20°C, glycerol was added to a final concentration of 50%.
The His-tagged proteins were purified to near homogeneity by chromatography through a nickel-nitrilotriacetic acid agarose column following the procedure provided by Qiagen under native (Nm23-H4⌬33) or denaturing (with 8 M urea for Nm23-H4) conditions. The eluted fractions containing the recombinant protein were equilibrated in 50 mM HEPES, pH 8, buffer containing 5 mM dithiothreitol and 5 mM EDTA by chromatography on a PD10 column (Amersham Pharmacia Biotech). For storage at Ϫ20°C, glycerol was added to a final concentration of 50%.
The purity of the recombinant proteins were 90 -95% as checked by SDS-PAGE and Coomassie Blue staining performed according to Laemmli (34). The protein concentrations were estimated using the following extinction coefficients, calculated from the amino acid composition (35), at 280 nm for 1 mg/ml solutions: 1.25 for the untagged Nm23-H4⌬33 protein and 1.11 and 1.43 for the tagged Nm23-H4⌬33 and Nm23-H4 proteins, respectively. Unless otherwise specified, the recombinant proteins used in the experiments were the His-tagged proteins.
Assay of NDP Kinase Activity and Autophosphorylation-NDP kinase activity was measured using a coupled pyruvate kinase-lactate dehydrogenase assay (36) with minor modifications. The assays were performed at 25°C in a 1-ml reaction mixture containing 50 mM Tris-HCl, pH 7.5, 75 mM KCl, 5 mM MgCl 2 , 1 mM phosphoenolpyruvate, 0.1 mM NADH, 1 mM ATP, 0.2 mM dTDP, 1 mg/ml bovine serum albumin and 2 units of pyruvate kinase and lactate dehydrogenase. To assay the NDP kinase activity of the full-length Nm23-H4 after renaturation, the protein eluted from the nickel-nitrilotriacetic acid column in urea containing buffer (Qiagen) was incubated overnight at room temperature after a 100-fold dilution in the NDP kinase reaction buffer devoid of pyruvate kinase, lactate dehydrogenase, and dTDP. The reaction was started by the addition of the coupled enzymes and 0.2 mM dTDP. As a control for correct renaturation conditions, the Nm23-H4⌬33 protein previously denatured in 8 M urea was similarly treated.
To detect the autophosphorylated intermediate, wild-type and mutant Nm23-H4⌬33 proteins (100 ng) were incubated in the presence of 1 Ci of [␥-32 P]ATP in 50 mM HEPES buffer, pH 8, containing 1 mM EDTA and 0.5 mM MgCl 2 . After 5 min at 4°C, half of the sample volume was removed and added with the gel sample buffer described by Laemmli (34). GDP (0.2 mM) was added to the other half and incubated further for 5 min at 4°C before addition of gel sample buffer. The samples were subjected to SDS-PAGE without boiling and then transferred to an Immobilon membrane, which was autoradiographed.
Heat and Urea Stability Measurement-The thermal stability of the NDP kinases was analyzed by measuring residual NDP kinase activity after 15 min of incubation at the indicated temperatures and cooling at 4°C. The human NDP kinases were diluted to a final concentration of 4 -13 g/ml in 50 mM Tris-HCl, pH 7.4, 1 mM dithiothreitol, 1 mM EDTA, and 1 mg/ml bovine serum albumin. T m was the temperature corresponding to a 50% loss in activity. To determine the stability upon urea treatment, the enzymes were diluted at the same concentration as above in 50 mM HEPES, pH 8, 10% glycerol and 0.1 M NaCl containing the indicated concentrations of urea and were incubated for 16 h at room temperature. Residual activity after both heat and urea treatments was measured by the coupled assay after a 100-fold dilution in the reaction buffer. Urea did not interfere with the assay.
Crystallization and Data Collection-The protein sample (tagged Nm23-H4⌬33) was dialyzed against 150 mM NaCl, 5 mM EDTA, and 100 mM HEPES buffer, pH 7.5. Hanging drops were prepared by adding 0.7 M Li 2 SO 4 to this protein solution over pits containing 1.4 M Li 2 SO 4 in the same buffer. The protein concentration in the drops was 4 mg/ml. Crystals grew within one week. They have cubic symmetry with cell parameters a ϭ b ϭ c ϭ 112.96 Å, ␣ ϭ ␤ ϭ ␥ ϭ 90°. A single crystal of size 200 m in each direction kept at 4°C was used in the x-ray data collection at the wavelength of 0.965 Å on the W32 station of the LURE-DCI synchrotron radiation center (Orsay, France) using a MA-Research image plate system. Data were processed and reduced with DENZO and SCALEPACK (37). Further processing used CCP4 program suite (38). Statistics are reported in Table I.
X-ray Structure Solution and Refinement-The structure was solved by molecular replacement with AmoRe (39), using an atomic model of human NDP kinase B (40) in which nonconserved residues were replaced by alanine. The solution was compatible with the space group P2 1 3, giving two subunits per asymmetric unit. The resulting model was then fitted to the electron density with the TURBO graphics program (64) and refined with CNS (41). Refinement statistics are shown in Table I.
Cell Culture, Transfection, and Confocal Microscopy-HEK 293 cells were grown at 37°C in a 5% CO 2 atmosphere in Dulbecco's modified Eagle's medium (Biological Industries) supplemented with 10% fetal calf serum (Biological Industries) containing 2 mM glutamine, 100 units/ml penicillin, and 100 g/ml streptomycin. The different GFP constructs (80 ng of plasmid DNA) were transiently transfected into the cells cultured in eight-well Lab-Tek R plates (2.5.10 5 cells/well) using the LipofectAMINE PLUS reagent kit (Life Technologies, Inc.) according to the protocol of the manufacturer. GFP fluorescence could be observed 24 -48 h after transfection. Just before examination, mitochondria were stained for 15 min at 37°C (5% CO 2 ) with 250 nM MitoTracker. Fluorescence in the living cells was examined by confocal microscopy with a TCS SP Leica microscope (Lasertechnik, GmbH) equipped with a ϫ 63 objective (plan apo; numeric aperture ϭ 1.4). A focal series was collected for each specimen with a focus step of 1 m. For each optical section, double fluorescence was acquired simultaneously using a krypton-argon mixed-gas laser adjusted to 488 nm for EGFP and to 568 nm for MitoTracker. Variable windows of spectrophotometers were adjusted to recover green (500 -550 nm) and red (580 -630 nm) fluorescence. The signal was treated with line averaging to integrate the signal collected over four lines in order to reduce noise. Selected paired sections were then processed to produce single overlay images (color merged) using a Power Mac G3 equipped with Photoshop software (version 5.0).
Specific Polyclonal Antibodies and Western Blot-The polyclonal antibodies against the Nm23-H4 NDP kinase were obtained by immunizing rabbits with the purified tagged Nm23-H4⌬33 recombinant protein.
The reactivity and specificity of the antibodies were analyzed by enzyme-linked immunosorbent assay and Western blotting. Proteins from cell extracts and purified recombinant proteins diluted in gel sample buffer (34) were electrophoretically separated on 4 -20% precast ready gels (Bio-Rad) or 12.5% polyacrylamide gels and transferred overnight at 100 mA onto a Immobilon P membrane (0.1 m, Millipore) in 0.025 M Tris-base, 0.192 M glycine, 20% methanol added with 0.02% SDS. The anti-Nm23-H4 (1/2000) antiserum and the polyclonal anti-cytochrome c (1 g/ml), the polyclonal anti-A/B NDP kinase (1 g/ml), the monoclonal anti-cytochrome oxidase (0.5 g/ml), and the monoclonal anti-porin (1/50,000) antibodies, diluted at the concentrations indicated in parentheses in the primary reaction, were reacted, in the secondary reaction, with peroxidase conjugated anti-rabbit (1/20,000) or anti-mouse antibodies (1/10,000) obtained from Biosys. The blots were revealed using the ECL Plus TM Western blotting detection system from Amersham Pharmacia Biotech.
Isolation of Purified Mitochondria and Submitochondrial Fractions-All steps were performed at 0 -4°C. Mitochondria were purified according to Hovius et al. (42) with minor changes. HEK 293 cells grown at confluence (about 10 9 cells) were scraped, washed twice with phosphate-buffered saline, resuspended in Buffer A (250 mM mannitol, 0.5 mM EGTA, 5 mM HEPES, 0.1% (w/v) bovine serum albumin, pH 7.4), and homogenized by nitrogen cavitation (250 p.s.i. for 30 min). The cell homogenate was centrifuged at 750 ϫ g for 10 min, the supernatant was discarded, and the pellet was washed by resuspension in Buffer A and centrifugation. Supernatants were pooled and centrifuged at 8740 ϫ g for 10 min. The pellet was resuspended in 1 ml of Buffer A; layered on one tube containing 8 ml of 30% Percoll (v/v) in 225 mM mannitol, 1 mM EGTA, 0.1% bovine serum albumin, 25 mM HEPES, pH 7.4; and spun for 30 min at 95,000 ϫ g (Beckman 90Ti rotor). The mitochondria were recovered as a diffuse band in the lower part of the gradient and washed twice in Buffer A by centrifugation for 10 min at 6,800 ϫ g. Submitochondrial particles were obtained by a swelling-shrinking procedure (42). The mitochondrial pellet was resuspended in 10 mM phosphate buffer, pH 7.4, at a concentration of 150 g of protein/ml and incubated for 20 min at 4°C with subsequent dilution by addition of the same volume of 32% sucrose, 30% glycerol, and 10 mM MgCl 2 and centrifugation at 10,000 ϫ g for 10 min. The supernatant was discarded, and the pellet containing mitoplasts was subjected to another swelling in 1 mM phosphate buffer for 20 min at 4°C. This mixture (inner compartment) and the previous supernatant (outer compartment) were centri-

Expression and Purification of Nm23-H4 Proteins-The
Nm23-H4 proteins were expressed in E. coli, either untagged (Nm23-H4⌬33) using the pET-21b vector or as hexahistidinetagged proteins (full-length Nm23-H4 and Nm23-H4⌬33) using the pET-28a(ϩ) vector (Fig. 1A). The truncated form is missing the first 33 amino acids and corresponds to the length of the Nm23-H1 and -H2 NDP kinases. The expression yield of the tagged Nm23-H4 was moderate (Fig. 1B). After sonication of the bacteria, the full-length protein, which was expressed as inclusion bodies, sedimented with the bacterial pellet (Fig. 1B,  lane 3), whereas the truncated protein appeared soluble in the supernatant (Fig. 1B, lane 7). The presence of the His tag did not modify the solubility of the proteins. Therefore, the tagged full-length Nm23-H4 and the Nm23-H4⌬33 proteins were pu-rified under denaturing (8 M urea) and native conditions, respectively. The purified recombinant proteins that were obtained were 90 -95% pure as judged by SDS-PAGE after Coomassie Blue staining (Fig. 1C). The proteins migrated slightly less than expected based on molecular masses deduced from the primary sequence that were as follows: 17.2 kDa for the truncated, 19.3 kDa for the tagged truncated, and 22.8 kDa for the full-length tagged proteins. E. coli NDP kinase (15.3 kDa) was added as a control and migrated at a position clearly different to Nm23-H4 proteins, showing that there was no contamination with the bacterial enzyme during Nm23-H4 purification (Fig. 1C, lane 1). In order to study the effects of the mutation of Ser 129 to proline (reverse K-pn mutation) on Nm23-H4, we also expressed and purified the S129P mutant truncated and tagged protein using a similar procedure (data not shown).
The nm23-H4 cDNA Codes for a Nucleoside Diphosphate Kinase-The tagged Nm23-H4⌬33 protein possesses NDP kinase activity with a V max of 450 units/mg, demonstrating that the human nm23-H4 gene codes for a nucleoside diphosphate kinase. The untagged Nm23-H4⌬33 protein exhibits similar specific activity (data not shown). When comparing the kinetic parameters between the wild-type and the S129P mutant proteins, we observed that the K m values were of the same order of magnitude (about 1 mM using dTDP as substrate), whereas the V max increased 6-fold to 2880 units/mg in the mutant (data not shown). The full-length protein, which forms insoluble inclusion bodies, was purified under denaturing conditions (8 M urea). We failed to detect any significant activity of the purified protein after removal of urea by chromatography on a PD10 column (Amersham Pharmacia Biotech). In addition, dilution and incubation of the full-length protein in buffer containing 1 mg/ml bovine serum albumin and 1 mM ATP was ineffective in revealing any NDP kinase activity, whereas these conditions permit renaturation and restoration of the activity of the truncated form previously denatured in 8 M urea (data not shown). This means that the truncated Nm23-H4 was able to recover its hexameric structure, which is required for NDP kinase activity, and that the N-terminal extension could inhibit the folding and/or activity of the full-length protein. Like other NDP kinases, wild-type and S129P Nm23-H4⌬33 rapidly autophosphorylated at 4°C in the presence of EDTA and MgCl 2 when incubated with radiolabeled ATP (Fig. 2) and the label was readily displaced by further incubation with cold GDP.
Stability of the Enzyme and Effect of the S129P Mutation-The thermostability of Nm23-H4⌬33 was compared with that of human NDP kinases A and B by measuring the residual activity of the enzymes incubated at various temperatures (Fig.  3A). As the proline residue of the K-pn loop plays a crucial role in the stability of the enzyme (17, 18), we made the S129P mutation and investigated its effect on the thermostability of

FIG. 2. Autophosphorylation of wild-type and mutant
Nm23-H4 proteins. The wild-type Nm23-H4⌬33 (wt) and mutant Nm23-H4⌬33 S129P proteins (100 ng each) were incubated with radiolabeled ATP. Either reactions were stopped (Ϫ) or GDP (ϩ) was added and the sample was further incubated as described under "Experimental Procedures." The samples were subjected to SDS-PAGE and autoradiography.
Nm23-H4. NDP kinases A and B exhibited T m values of 53 and 59°C, respectively. In comparison, the Nm23-H4⌬33 enzyme was thermolabile with a T m of 40°C. Similar results were obtained with untagged Nm23-H4⌬33 (data not shown). The S129P mutation resulted in a dramatic increase in stability, with an upward shift of 25°C in the T m value. Similarly, Nm23-H4 was observed to be highly susceptible to urea denaturation (Fig. 3B). Whereas NDP kinases A and B retained half-maximal activity in 4 M urea, Nm23-H4⌬33 lost 90% of its activity in 3.5 M urea. The S129P mutant also showed a remarkable increase in stability in this test, retaining more than 80% of its activity in 7 M urea. Interestingly, we observed an increase in the NDP kinase activity at low concentrations of urea, which reached 50% for NDP kinase B. These data indicate that the wild-type Nm23-H4 enzyme is particularly unstable upon heat or urea treatments as compared with the other NDP kinases and that the residue at position 129 plays a crucial role in the stability of the enzyme structure.
X-ray Structure of Nm23-H4 -The x-ray data show that the tagged Nm23-H4⌬33 protein forms a hexamer very similar to that of other eukaryotic NDP kinases (Fig. 4A). In the cubic crystal form, the asymmetric unit contains a dimer, and the hexamer is reconstituted by applying a 3-fold crystal symmetry. The subunits have the characteristic fold shared by all other NDP kinases and based on a four-stranded antiparallel ␤-sheet with five connecting ␣-helices (43). The most significant difference between Nm23-H4 and other NDP kinases is at the C terminus, in which the last 12 residues are disordered and cannot be located in the electron density map. The N-terminal histidine tag is also disordered, and the first visible residue is Met 34 , equivalent to the N-terminal methionine of NDP kinase B.
Compared with NDP kinase B (40) and excluding the disordered C-terminal segment, all 142 common C␣ atoms superimpose with a root mean square distance of 1.0 Å (Fig. 4B). The largest changes are located in helices ␣A and ␣2 (residues 77-97), at the beginning of helix ␣3 (residues 112-119), and in the K-pn loop (residues 126 -132 and 144 -147). In NDP kinase, a groove situated between the K-pn loop and a pair of surface helices (helices ␣A and ␣2) constitutes the nucleotide binding site, and helices ␣A and ␣2 are seen to move by 1-2 Å when a nucleotide is present (40). The NDP kinase B structure contains GDP, which explains most of the difference with Nm23-H4 where the site is empty. However, the crystal packing may also play a part, the loop connecting helices ␣A and ␣2 being in contact with neighboring protein molecules in the cubic crystal form.
The presence of a serine at position 129 of Nm23-H4 instead of a proline in NDP kinase B has observable structural consequences in the K-pn loop (Fig. 4C). The substitution introduces two hydrogen bond donor groups, a main chain NH and side chain OH. They donate hydrogen bonds to the main chain carbonyl of residue 144, which requires the 144 -145 peptide bond to flip 180 degrees relative to its orientation in NDP kinase B (Fig. 4D). The same local change takes place in the x-ray structure of the P100S mutant of Dictyostelium NDP kinase, where the equivalent substitution is achieved by sitedirected mutagenesis (18). In the P100S mutant, local changes induced by the peptide bond flipping in one subunit propagate from the mutation site to the C terminus of a neighboring subunit. This cannot be seen in the Nm23-H4 x-ray structure, in which the C terminus is disordered. In addition to the peptide flip, the electron density suggests that a sulfate ion binds to backbone NH groups of residues 144 and 147 in one of the two subunits (subunit A) of the cubic asymmetric unit. No such density is present in subunit B of Nm23-H4, in which this part of the structure is poorly ordered. The anion binding site does not exist in NDP kinase B, in which the NH of the equivalent residues point in the opposite direction.
Although it is a hexamer, the Nm23-H4 protein shows weaker subunit interactions than NDP kinase B, as can be estimated from the area of the surface buried between subunits. The hexamer can be viewed as made of three dimers or two trimers. Whereas the surface buried in a dimer has about the same area as in NDP kinase B, the surface buried in a trimer drops from 1550 to 1020 Å 2 per subunit due to the unstructured C-terminal residues. These residues contribute 660 Å 2 per subunit to the trimer interface in NDP kinase B. Table II lists interactions made at the trimer interface in Nm23-H4. The arginine side chain in position 64 plays a major part in these interactions, giving no less than four hydrogen bonds to main chain carbonyls, which is two more than the equivalent lysine in NDP kinase B.
Nm23-H4 Mitochondrial Localization by Confocal Microscopy-To determine the subcellular localization of Nm23-H4, the protein was transiently expressed into human HEK 293 cells, fused to the N terminus of the green fluorescent protein (EGFP) (Fig. 5). The plasmid containing the cDNA coding for EGFP alone was used as a control. After 24 -48 h of transfection, the fluorescent patterns of the expressed EGFP and Nm23-H4/EGFP fusion proteins were examined by confocal microscopy. Mitochondria were stained by addition of the mitochondrion selective fluorescent probe (MitoTracker) just before examination of the live cells. Cells transfected with the EGFP plasmid showed, as classically observed, a diffuse green fluorescent pattern in both the cytosol and nuclei (Fig. 5A). In contrast, cells expressing the wild-type Nm23-H4 fusion protein showed a green punctiform pattern in the cytoplasm with no labeling of nuclei (Fig. 5B). This pattern was identical to the FIG. 3. Thermal and urea denaturation of NDP kinases. A, thermal inactivation of human NDP kinases: the wildtype Nm23-H4⌬33 (q), the mutant Nm23-H4⌬33 S129P (E), and the A (f) and B (Ⅺ) NDP kinases were incubated for 15 min at the indicated temperatures, and the residual activity was determined as described under "Experimental Procedures." 100% corresponds to the activity of each enzymes kept at 4°C. B, urea denaturation of human NDP kinases: the wild-type Nm23-H4⌬33 (q), the mutant Nm23-H4⌬33 S129P (E), and the A (f) and B (Ⅺ) NDP kinases were incubated for 16 h at room temperature in the presence of the indicated concentrations of urea, and the residual activity was measured as described under "Experimental Procedures." 100% corresponds to the activity of each enzyme under nondenaturing conditions. red pattern of the specific mitochondrial marker (Fig. 5C), as shown by the yellow punctiform areas obtained upon merging of the green and red images of identical confocal sections (Fig.  5D). As it is known that protein stability plays a role in mitochondrial import (44), we also analyzed the subcellular localization of the Nm23-H4 protein stabilized by the S129P mutation. As shown in the confocal section (Fig. 5E), the green fluorescent pattern of the mutant protein was similar to the pattern of the wild-type protein (Fig. 5B) and identical to the red fluorescent pattern obtained with MitoTraker (Fig. 5F), as indicated by the yellow coloration after image merging (Fig.  5G). This demonstrates that the increased stability of the mutant Nm23-H4 did not alter its mitochondrial import.
Submitochondrial Localization of Nm23-H4 -We have analyzed the subcellular and submitochondrial localizations of Nm23-H4 using a polyclonal antiserum against the Nm23-H4⌬33 recombinant protein. The specificity of the polyclonal antibodies was analyzed by enzyme-linked immunosorbent assay (data not shown) and Western blotting (Fig. 6A). These antibodies specifically recognized Nm23-H4 (lane 1) because no signal was obtained either with the other human A, B, and DR-Nm23 NDP kinases (lanes 2-4) or with the enzyme from E. coli (lane 5). To confirm the Nm23-H4 mitochondrial localization observed by confocal microscopy, we prepared purified mitochondria by differential and Percoll gradient centrifugations of HEK 293 cell extracts. As shown in Fig. 6B (top panel), Nm23-H4 was highly enriched in mitochondria (lane M) as compared with the homogenate (lane H). A similar enrichment in the specific mitochondrial marker, cytochrome oxidase II subunit was also observed (Fig. 6B, middle panel). There were no A or B NDP kinases associated with mitochondria, whereas they were present in the homogenate and the postmitochondrial supernatant (Fig. 6B, lower panel), as expected from their cytoplasmic localization. A lower band, probably corresponding to proteolyzed enzymes, was also observed. This also demonstrates that our preparation of mitochondria was devoid of cytosolic contamination. The Nm23-H4 protein present in the homogenate and the mitochondrial fraction exhibits a slightly lower molecular weight than the recombinant tagged Nm23-H4⌬33 protein (174 amino acids), indicating in vivo proteolytic cleavage of the protein, probably between residue His 28 and residue Gly 29 , as suggested by computer program analysis of the primary sequence (45). A further submitochondrial fractionation by a swelling-shrinking procedure (42) was used to obtain outer membrane, intermembrane space, inner membrane, and matrix (Fig. 7A). When probing the various subfractions with anti-Nm23-H4 antibodies, Nm23-H4 was shown to be associated with both the outer and the inner mitochondrial membranes and was not found in the soluble fractions, intermembrane space, or matrix. The mitochondrial subfractions were identified by demonstrating the presence of porin and cytochrome oxidase in the outer and inner membrane fractions, cytochrome c in the intermembrane space, and Mn-SOD in the matrix by reaction with specific antibodies. In order to analyze the membrane insertion of Nm23-H4 to the inner membrane, we measured its level of resistance to alkaline treatment (Fig.  7B). This procedure allows distinction between peripherally bound proteins and integral proteins such as cytochrome oxidase (46). When the inner membrane fraction was extracted FIG. 4. Nm23-H4 structure and comparison with NDP kinase B. A, the Nm23-H4 hexamer is viewed along the 3-fold axis. The front trimer is in blue, and the back trimer is in yellow. The boxed region is shown in detail in C. B, superposition of the H4 subunit in yellow and the NDP kinase B subunit with bound GDP in green. Differences between the two structures are localized at the N and C termini and on the sides of the nucleotide binding cleft formed by helices ␣A-␣2 and the K-pn loop. The latter is also involved in the subunit interface forming the trimers. C, details of subunit interactions in the back trimer (boxed region in A). Red balls are the carbonyl oxygens of Ser 129 , Gly 139 , and Ser 142 of the K-pn loop interacting with the side chain of Arg 64 Ј from the neighboring subunit. The Ser 129 side chain H-bonds to the carbonyl of His 144 in the same subunit. In NDPK B (green bonds), the residue corresponding to Ser 129 is Pro 96 . The C terminus, disordered in H4, makes subunit contacts in NDPK B, including a H-bond to Gln 111 equivalent to His 144 (drawn with Molscript (61) and Raster3D (62)). D, close-up of the prolineto-serine substitution at position 129. NDPK H4 is colored according to atom type and NDPK B is in green. In H4, two hydrogen bonds from the serine NH and OH cause a rotation of the carbonyl group of His 144 . The proline in NDPK B cannot make these bonds (drawn with TURBO).

TABLE II Interactions within the Nm23-H4 trimers
The cutoff distance is 4 Å for nonpolar and 3.5 Å for polar interactions. When the interaction is observed in the NDP kinase B structure, the equivalent residue number is indicated in parentheses (63 with sodium carbonate (pH 11.5), the cytochrome oxidase remained in the pellet fraction, whereas Nm23-H4 was recovered in the supernatant fraction. This demonstrates that Nm23-H4 is a peripheral protein. DISCUSSION The present work demonstrates that nm23-H4, a recently discovered member of the nm23-NDP kinase gene family, encodes a catalytically active NDP kinase that is specifically localized in mitochondria. Mitochondrial NDP kinase activity has been reported as early as 1955 (23), but Nm23-H4 is the first mammalian mitochondrial NDP kinase accessible to mo-  lecular analysis. The recombinant Nm23-H4 is catalytically active when the N-terminal extension is removed. Indeed, the truncated isoform missing the first 33 amino acids possessed NDP kinase activity, whereas we were unable to detect activity with the full-length enzyme, suggesting that the N-terminal extension could inhibit enzyme folding and/or activity. In a comparative study, the catalytic properties of the active Nm23-H4 have been shown to be very similar to those of the other human NDP kinases (47). Among the human isoforms, only Nm23-H1,-H2, DR-Nm23, and Nm23-H4 have been characterized as NDP kinases (Refs. 13 and 47 and this paper). Given their broad substrate specificity, the respective function of the various NDP kinase isoforms should reflect their different subcellular localization.
Although their sequences are markedly conserved during evolution, NDP kinases differ in their quaternary structure and can be either hexameric or tetrameric. The stability of the hexamer of Drosophila and Dictyostelium NDP kinases depends on the presence of a proline in the position equivalent to residue 129, occupied by a serine in Nm23-H4, and on the C-terminal residues. The proline and C-terminal residues are implicated in subunit contacts within trimers, and the hexamer becomes unstable when the proline is mutated and/or the Cterminal residues deleted (17,18,48). The x-ray structure shows that Nm23-H4, which has a serine at position 129 and a divergent C-terminal sequence, is a hexamer. However, it is very thermolabile, with a much lower T m than the human A and B enzymes, and highly sensitive to urea denaturation. Ser 129 of Nm23-H4 is equivalent to the K-pn (Killer of prune) mutation of the Awd NDP kinase of Drosophila, which is dominant lethal in the context of the prune mutation. We have mutated Ser 129 to a proline (reverse K-pn mutation) and analyzed its contribution to the catalytic and stability characteristics of the mutated Nm23-H4 enzyme. This single substitution greatly increased the maximal velocity as well as the thermostability of the enzyme with a 25°C gain in T m . The mutated Nm23-H4 was also highly stabilized against denaturation using chaotropic agents because a significant activity could be observed even after 8 M urea denaturation. This suggests stabilization of the monomer, which could easily reassociate upon dilution of urea to form the active hexamer. The large increase in stability is a remarkable consequence of the replacement of a serine by a proline, which actually removes from the structure hydrogen bonds made by the two donor groups of the serine. However, position 129 plays a crucial role at subunit interfaces in all hexameric NDP kinases, and similar but lesser effects have already been described in enzymes from other species. In Drosophila NDP kinase, in which proline is present in the wild-type, the natural K-pn mutation destabilizes the hexamer (17), and in Dictyostelium, site-directed mutagenesis achieves the same result (18). An exception may be the NDP kinase purified from the matrix of pigeon mitochondria, which appears to be highly stable despite the presence of a serine at this position (19).
We demonstrate here by confocal microscopy and Western blot analysis that Nm23-H4 is a specific mitochondrial isoform. Therefore, mitochondrial NDP kinase is a separate entity, and neither the A nor the B cytosolic NDP kinase is associated with mitochondria. The activity that we recovered in the purified mitochondria from HEK 293 cells represents about 4% of the total activity in the homogenate. 2 This is in accordance with data from the literature (1,49) reporting that the NDP kinase activity of the mitochondrial fraction of various tissues represents only a few percent of the total cellular NDP kinase activity, which is mainly cytosolic due to the abundantly expressed A and B isoforms. It is also in accordance with the level of nm23-H4 mRNA, which is at least 10-fold lower than that of nm23-H1 and -H2. 2 It should be mentioned that a protein of the size of the full-length protein could not be detected in the HEK 293 cell homogenate by Western blotting with specific antibodies, indicating that there is no accumulation of the preprotein in the cytoplasm and that close coupling between translation and mitochondrial import should exist for Nm23-H4 as already observed for other mitochondrial proteins (50).
Troll et al. (29), studying the mitochondrial Dictyostelium enzyme, which is also labile, have speculated that the lower stability of the mitochondrial isoform may be important for the import of the protein into mitochondria because it could facilitate unfolding of the precursor during the targeting process, in a way similar to other mitochondrial proteins (51). In fact, the highly stable Nm23-H4 mutant presented here and the pigeon NDP kinase are imported into mitochondria, ruling out the hypothesis that the NDP kinase stability could be a limiting factor for its mitochondrial import.
With regard to the submitochondrial localization of NDP kinases, it has been mainly reported that NDP kinase activity is soluble and can be found within and outside the inner membrane barrier of mitochondria in amounts varying with species and tissues (1,25). More recently, NDP kinase activity was found associated with the contact sites between the outer and the inner membranes (52)(53)(54). In a first attempt to determine the submitochondrial localization of Nm23-H4, we have prepared mitochondrial subfractions of HEK 293 cells. Clean, uncontaminated subfractions were difficult to obtain, perhaps due to the fragility of the inner membrane observed in mitochondria of tumor cell lines (55). Nonetheless, we clearly demonstrate that the Nm23-H4 protein was associated with the mitochondrial membranes and not found in the soluble fractions as assessed by specific markers. Among the mitochondrial NDP kinases purified to date, Nm23-H4 is the first to be shown to be clearly associated with the mitochondrial membranes. The pigeon NDP kinase purified from the matrix fraction was found to be soluble in the supernatant of sonicated mitochondria (19). A pea mitochondrial NDP kinase, which has been partially sequenced, was purified from a soluble fraction proposed to be the intermembrane space based on difference in accessibility of the ATP substrate (56). The mitochondrial D. discoideum enzyme possesses an N-terminal extension similar to the bipartite presequence of protein targeted to the intermembrane space (29) and was found to be soluble. As expected from the sequence lacking any membrane attachment peptide and its oligomeric state, Nm23-H4 is not an intrinsic membrane protein and could be released from the membranes by alkali treatment. The mechanism of the association is unknown. It could be through a protein-to-protein interaction or electrostatic interactions with acidic phospholipids as reported for the human mitochondrial creatine kinase (57). Hydrophobic interactions are less likely because the hydrophobic regions are either hidden or located at the surface of the hexamer but in a region (helix ␣2) conserved between soluble NDP kinases.
It cannot be excluded that another NDP kinase isoform(s) exists in human mitochondria for several reasons. First, we have observed that there was a NDP kinase activity in the mitochondrial soluble fraction (data not shown) in addition to the one bound to the membrane that corresponds to Nm23-H4, as shown by Western blot analyses. Second, given the high degree of identity (90%) between the cytosolic pigeon and the human NDP kinase B, we would expect a similar level of identity between mitochondrial pigeon and human orthologues. In fact, the pigeon and human mitochondrial enzymes are only 72% identical, indicating that they could correspond to different isoforms. Also, although species differences cannot be ruled out, the tissue distribution of the pigeon and human NDP kinases are different. We found that Nm23-H4 is expressed abundantly in skeletal muscle and poorly in kidney (6), whereas the reverse is true for the pigeon mitochondrial enzyme (19).
Submitochondrial fractionation analyses showed that Nm23-H4 behaves like porin, strongly suggesting that it is associated with contact sites. These sites were shown to be involved in the transport of proteins, solute, and energy (54). They contain a polyprotein channel, the permeability transition pore, which includes porin (or voltage anion-dependent channel), the ATP/ADP carrier, and kinases, hexokinase, and creatine kinase (52). The ATP/ADP carrier and porin are the targets of the pro-and anti-apoptotic proteins Bax (58) and Bcl2 (59), respectively, and play a major role in the control of apoptosis (60) through the control of permeability transition pores. Nm23-H4 is a good candidate as a NDP kinase associated with the contact sites at which it could control the ATP/ ADP ratio. The specific association of Nm23-H4 with contact site components, as well as its orientation (toward intermembrane space or matrix sides), will be further analyzed in order to define its specific role in mitochondrial function.