The metastasis suppressor NM23-H1 possesses 3'-5' exonuclease activity.

NM23-H1 belongs to a family of eight gene products in humans that have been implicated in cellular differentiation and development, as well as oncogenesis and tumor metastasis. We have defined NM23-H1 biochemically as a 3'-5' exonuclease by virtue of its ability in stoichiometric amounts to excise single nucleotides in a stepwise manner from the 3' terminus of DNA. The activity is dependent upon the presence of Mg(2+), is most pronounced with single-stranded substrates or mismatched bases at the 3' terminus of double-stranded substrates, and is inhibited by both ATP and the incorporation of cordycepin, a 2'-deoxyadenosine analogue, into the 3'-terminal position. The 3'-5' exonuclease activity was assigned to NM23-H1 by virtue of: 1) precise coelution of enzymatic activity with wild-type and mutant forms of NM23-H1 protein during purification by hydroxylapatite and gel filtration column high performance liquid chromatography and 2) significantly diminished activity exhibited by purified recombinant mutant forms of the proteins. Lysine 12 appears to play an important role in the catalytic mechanism, as evidenced by the significant reduction in 3'-5' exonuclease activity resulting from a Lys(12) to glutamine substitution within the protein. 3'-5' Exonucleases are believed to play an important role in DNA repair, a logical candidate function underlying the putative antimetastatic and oncogenic activities of NM23-H1.

NM23-H1 belongs to a family of eight gene products in humans that have been implicated in cellular differentiation and development, as well as oncogenesis and tumor metastasis. We have defined NM23-H1 biochemically as a 3-5 exonuclease by virtue of its ability in stoichiometric amounts to excise single nucleotides in a stepwise manner from the 3 terminus of DNA. The activity is dependent upon the presence of Mg 2؉ , is most pronounced with single-stranded substrates or mismatched bases at the 3 terminus of double-stranded substrates, and is inhibited by both ATP and the incorporation of cordycepin, a 2-deoxyadenosine analogue, into the 3-terminal position. The 3-5 exonuclease activity was assigned to NM23-H1 by virtue of: 1) precise coelution of enzymatic activity with wild-type and mutant forms of NM23-H1 protein during purification by hydroxylapatite and gel filtration column high performance liquid chromatography and 2) significantly diminished activity exhibited by purified recombinant mutant forms of the proteins. Lysine 12 appears to play an important role in the catalytic mechanism, as evidenced by the significant reduction in 3-5 exonuclease activity resulting from a Lys 12 to glutamine substitution within the protein. [3][4][5] Exonucleases are believed to play an important role in DNA repair, a logical candidate function underlying the putative antimetastatic and oncogenic activities of NM23-H1.
nm23-H1 was first classified as a metastasis suppressor gene on the basis of its reduced expression in several metastatic melanoma cell lines relative to nonmetastatic counterparts (1). Subsequently, low expression of the NM23-H1 protein has been linked to increased metastatic potential in human breast carcinoma (2), hepatoma (3), and gastric carcinoma (4). NM23-H1 overexpression has been shown experimentally to inhibit the metastatic phenotype and/or promote differentiation in melanoma (5), breast carcinoma (6), and transformed neural cell lines (7,8). The human nm23-H1 gene is one of eight related NM23 family members identified to date (reviewed in Ref. 9). Each exhibits nucleoside-diphosphate kinase (NDPK) 1 activity, catalyzing the transfer of ␥-phosphate between nucleoside triphosphate and nucleoside diphosphate via a "ping-pong" mechanism (10). NDPK activity does not appear to be relevant to metastasis suppression, however, as catalytically inactive mutants retain metastasis suppressor activity (11,12). In addition to NDPK, a number of other biological activities have been reported for NM23 proteins, some of which have been proposed to underlie metastasis suppression. Two spontaneous mutations in NM23-H1, a serine 120 to glycine (S120G) substitution seen frequently in aggressive neuroblastomas (13), and a proline 96 to serine mutation that corresponds to the killer of prune mutation (Kpn; P96S) in the Drosophila homologue of NM23 (awd), both result in loss of antimetastatic activity (14). Interestingly, both mutations also abrogate a histidine-dependent protein kinase activity of NM23-H1 (15). This activity has been implicated more recently in the serine phosphorylation of the kinase suppressor of ras (KSR), suggesting antimetastatic activity may arise via the suppression of ras-initiated growth signals (16).
Considerable evidence has also been presented to indicate that NM23 proteins interact with DNA. NM23-H2 has been shown to bind and activate the nuclease-hypersensitive element (NHE) of the c-myc promoter (17,18), suggesting a molecular mechanism of oncogenesis and malignant progression. NM23-H2 also cleaves the NHE sequence in vitro when presented in either linear or supercoiled plasmid form (19), suggesting a role in modulating transcription via the remodeling of regulatory elements that exhibit non-B-form, or paranemic, DNA conformations (for a review, see Ref. 20). Each of these interactions with the NHE were independent of NDPK activity, as they were retained with an NDPK-defective mutant form (H118F) of the protein (21). The DNA cleaving activity of NM23-H2 was further shown to occur via a DNA glycosylase/ lyase-like mechanism (22), a hallmark of base excision DNA repair enzymes. Both NM23-H1 and NM23-H2 repress transcription via interactions with paranemic elements in the promoter region of the platelet-derived growth factor-A (PDGF-A) gene (23,24). Repression of this oncogenic and metastasispromoting growth factor (25) is consistent with a potential antimetastatic function of NM23 proteins. Interestingly, NM23-H1 and NM23-H2 also cleaved the PDGF-A regulatory elements in vitro; NM23-H1 appeared to excise nucleotides progressively from the 3Ј terminus of single-stranded oligodeoxynucleotides, whereas NM23-H2 appeared to cleave internally, as observed previously with the c-myc NHE sequence. Interestingly, the DNA cleavage function of NM23 is conserved from the primordial NDPK gene in Escherichia coli (26). A recent study has also shown that NM23-H1 is the DNA-cleaving component of a latent protein complex (SET) that is acti-vated during cytotoxic T lymphocyte-mediated apoptosis (27). While observed in the context of the immune response, this finding suggests that proapoptotic or immunosurveillance functions may underlie the antimetastatic activity of NM23-H1.
In this report we show that human NM23-H1 that is expressed in E. coli and purified to apparent homogeneity exhibits biochemical characteristics consistent with 3Ј-5Ј exonuclease activity. 3Ј-5Ј Exonucleases are associated generally with DNA proofreading (reviewed in Ref. 28), with loss of expression and/or function often associated with mutator phenotypes, and increased potential for cancer progression (29). Thus, the 3Ј-5Ј exonuclease activity of NM23-H1 demonstrated in this report would appear to represent a logical candidate function underlying its antimetastatic activity.

EXPERIMENTAL PROCEDURES
DNA and Site-directed Mutagenesis-E. coli plasmids (pET3c, New England Biolabs) for expression of wild-type NM23-H1 and its mutant variants, R34A and H118F, were kindly provided by E. Postel (Princeton). K12Q was generated by the overlap extension modification of the polymerase chain reaction (30) using the sequences 5Ј-cgggatcccatatggccaactgtga-3Ј and 5Ј-ccggatccccgaattctcattcatagatccagt-3Ј as 5Ј-and 3Ј-end primers, respectively, and the mutagenic primers 5Ј-cccatctggctggatcgcaatgaaggtac-3Ј and 5Ј-cattgcgatccagccagatggggtcca-3Ј (codons encoding the mutant glutamine residue are in underlined bold font). The cDNA sequence encoding K12Q was inserted in-frame between the NdeI and BamHI sites of pET3c. The cloned PCR product was sequenced to ensure that only the desired mutation was present.
Overexpression and Purification of Recombinant Human NM23-H1-NM23-H1 and mutant proteins were expressed in E. coli essentially as described (23). Briefly, pET3c vectors containing wild-type and variants of NM23-H1 were used to transform E. coli (BL21 plysS; Promega). For protein expression, freshly transformed bacteria were grown overnight at 37°C in LB-ampicillin medium. Two ml of the overnight culture was used to inoculate 200 ml of LB medium containing ampicillin. When the A 600 reached 0.6, protein expression was induced by the addition of isopropyl-1-thio-␤-D-galactopyranoside (final concentration, 0.4 mM). Three hours following induction, cells were harvested, resuspended in 30 ml of lysis buffer (50 mM Tris, pH 8.0) containing 1 mM EDTA and dithiothreitol, 1 mM leupeptin and pepstatin A, and 0.1 mM phenylmethylsulfonyl fluoride (Sigma) and lysed by sonication. Cell lysates were cleared by centrifugation at 12,000 ϫ g for 30 min and the proteins were precipitated with ammonium sulfate (60 -90% fractions). The 60 -90% fraction was prepared for application to a DEAE-Sephacel column by dialysis into 50 mM Tris buffer (pH 7.5) containing a mixture of protease inhibitors. Under these conditions, NM23-H1 bound to the column and was eluted with a 0 -1 M NaCl gradient with the peak fraction at ϳ350 mM NaCl. Peak fractions containing NM23 proteins were equilibrated in 10 mM phosphate buffer (pH 7.0) by centrifugal filtration (Centricon-10, Millipore) and loaded onto a hydroxyapatite (HTP) column. NM23-H1 was eluted with a 80-ml phosphate gradient of 10 -800 mM. Purity of recombinant proteins was evaluated by SDS-PAGE and Commassie Blue and/or silver staining. Unless otherwise indicated, only the 1-ml peak fraction was employed for cleavage assays.
DNA Cleavage Assays-A number of the oligodeoxyribonucleotide sequences employed for cleavage assays were derived from either the 5Ј-SHS silencer or the NHE elements found in the 5Ј-flanking region of the PDGF-A promoter (Table I; Refs. 23 and 31). Oligodeoxyribonucleotides 12-15 were substrates of the WRN helicase/exonuclease provided by David Orren (University of Kentucky). Radiolabeling of 5Ј termini was performed in a 40-l reaction mixture at 37°C for 30 min using [␥-32 P]ATP and T4 polynucleotide kinase. Double-stranded probes were annealed by incubation of the radiolabeled single strand of interest with its complement at 95°C for 5 min followed by gradual cooling to ambient room temperature over a 3-h interval. For 3Ј-end labeling, the strand to be labeled was first annealed to a complementary strand designed to provide a duplex with a 5Ј-overhang. The recessed 3Ј terminus was filled in with [␣-32 P]dCTP in the presence of the three remaining unlabeled nucleotides and the Klenow fragment of DNA polymerase I. To prepare a doubled-stranded oligodeoxyribonucleotide probe with the 2Ј-deoxyadenosine analogue cordycepin at the 3Ј terminus, the purine-rich strand of NHE was first 5Ј-labeled with ␥-32 P and T4 polynucleotide kinase, then annealed to its complementary pyrimidine-rich strand (9-bp mismatch, oligonucleotide number 3). This du-plex was incubated with 10 mM cordycepin 5Ј-triphosphate and terminal transferase at 37°C for 1 h (specific activity: ϳ1.0 ϫ 10 6 cpm/pmol). Routine DNA cleavage assays were performed in a 15-l reaction mixture containing 20 mM Hepes buffer (pH 7.9), 10 -50 fmol of 5Ј-32 Plabeled oligonucleotide, 2 mM MgCl 2 , and 100 mM KCl as described previously (23). Standard reactions were initiated by the addition of 0.5-1 g of protein and carried out at room temperature. Reactions were terminated by adding an equal volume of sequencing loading dye consisting of 80% deionized formamide (w/v), 10 mM EDTA (pH 8.0), xylene cyanol FF (1 mg/ml), and bromphenol blue (1 mg/ml) followed by heating at 95°C for 5 min. Cleavage products were resolved on sequencing gels ranging in polyacrylamide concentration between 10 and 15% as indicated in the text and visualized by phosphorimaging (Storm 860, Amersham Biosciences).
Nucleoside-diphosphate Kinase Assays-NDPK activity was measured as described (10) in a coupled pyruvate-lactate dehydrogenase assay in which ATP acted as a phosphate donor and dTDP as an acceptor. The procedure was modified for the current study to accommodate the use of 96-well plates and microplate reader. NDPK reactions were initiated by the addition of 10 l of purified NM23-H1 in 20 mM Hepes buffer (pH 7.9) into individual wells of a 96-well plate, each containing 140 l of reaction mixture consisting of 1 mM pyruvate enol phosphate, 5 mM ATP, 2.5 mM dTDP, 6 mM MgCl 2 , 50 mM KCl, 1000 units of pyruvate kinase and lactate dehydrogenase. NAD ϩ formation, which reflected ADP production, was measured every 25 s at 340 nm by a HTS 7000 plus Bioassay Plate Reader (PerkinElmer Life Sciences). Specific activity was expressed as units/mg of protein, with 1 unit defined as the amount of enzyme converting 1 mol of ADP/min at room temperature (⑀ 340 of NADH ϭ 6.22 ϫ 10 3 M Ϫ1 cm Ϫ1 ).
Circular Dichroism Analysis-Far UV-CD spectra were recorded from 260 to 190 nm using a Jasco J-810 spectrometer, with each individual spectrum representing the average of 30 replicate measurements. Prior to analysis, purified NM23-H1 proteins were diluted to a concentration of 0.12 mg/ml in 5 mM phosphate buffer (pH 7.0) (⑀ 280 ϭ 1.35 for a 1 mg/ml solution; Ref. 32). CD spectra were recorded at 5°C in a quartz cuvette with a 0.1-cm path-length. CD data were converted to mean residual ellipticity, assuming a hexameric structure and mean residue molecular mass of 113 Da, the latter calculated from a 152amino acid monomer with a molecular mass of 17,180 Da (33). Secondary structure estimates were derived from the 250 -190-nm region of the recorded CD spectra using the programs CONTILL, SELCON3, and CDSSTR and their reference set of 43 proteins (34) from the CDPro software package. 2 High Performance Liquid Chromatography (HPLC) Gel Filtration-HTP-purified wild-type or mutant forms of NM23-H1 were loaded on a Shodex gel filtration HPLC column (Shodex Protein KW-800, Showa Denko) pre-equilibrated in 50 mM Tris (pH 7.5), 0.1 M KCl. Molecular masses of NM23-H1 were estimated relative to a standard curve generated with a commercially available molecular mass standards kit (Sigma) containing cytochrome c (12.4 kDa), carbonic anhydrase (29 kDa), bovine serum albumin (66 kDa), alcohol dehydrogenase (150 kDa), ␤-amylase (200 kDa), and blue dextran (2,000 kDa).

NM23-H1 and DNA Cleavage Activity Coelute following Gel
Filtration HPLC-DNA cleaving activity co-purifies with NM23-H1 following a three-step protocol consisting of ammonium sulfate precipitation and sequential chromatographic steps of DEAE-Sephacel and hydroxylapatite ( Fig. 1A; Ref. 23). When NM23-H1-containing fractions were incubated with a 5Ј-end labeled oligodeoxyribonucleotide corresponding to the noncoding strand of the 5Ј-SHS silencer (5Ј-SHS antisense strand, or 5Ј-SHSas) from the PDGF-A gene (31), progressively smaller fragments were obtained with increasing amounts of protein. Oligodeoxyribonucleotide substrates employed in this study are summarized in Tables I and II. This DNA cleavage pattern strongly suggested that NM23-H1 was excising nucleotides progressively from the 3Ј terminus of this singlestranded DNA molecule, consistent with a 3Ј-5Ј exonuclease activity. To rule out further the possibility that the 3Ј-5Ј exonuclease-like activity was catalyzed by nuclease contamination from the E. coli expression host, the peak fraction of NM23-H1 a WRN refers to oligodeoxyribonucleotide sequences employed previously as substrates in the study of the Werner syndrome protein, WRN (39). WRN substrate G80b (oligodeoxyribonucleotide 15) contains 21 nucleotides of mismatched sequence (bold letters) in relation to the partially complementary oligodeoxyribonucleotide, C80 (oligodeoxyribonucleotide 14). Annealing of these sequences yield an internal "bubble" structure that serves as a high affinity substrate for the WRN protein. Shown are elution profiles from the final step of purification involving HTP column chromatography. HTP chromatography was conducted using a Waters HPLC system at a flow rate of 1 ml/min, with NM23-H1 proteins eluted using an 80-ml gradient of potassium phosphate (10 -800 mM) and the collection of 1-min fractions. Within each of panels A-D are shown the A 280 profile and accompanying phosphate gradient (top), and analysis of column fractions by SDS-PAGE with Coomassie staining (middle) and nuclease assay (bottom). Fractions in the vicinity of the H1 peak (60 -88 min) were individually concentrated 3-5-fold by Centricon ultrafiltration before assay. Aliquots containing 1 g of protein from the respective peak NM23-H1 fractions (82-83 min for wild-type and K12Q; 63-64 min for R34A; 66 -67 min for H118F) and equivalent volumes from adjacent fractions were used for cleavage assay. The nuclease assay consisted of incubation of column fraction aliquots with 10 -20 fmol of single-stranded 32 P-labeled 5Ј-SHSas as the DNA substrate (oligodeoxyribonucleotide 1; Table I) at 22°C for 2 h in the presence of 2 mM MgCl 2 and 100 mM KCl. Cleavage products were analyzed by electrophoresis through denaturing 10% polyacrylamide gel and visualized by phosphorimaging.
from the hydroxylapatite column (82-83 min) was further analyzed by gel filtration HPLC. Two peaks of NM23-H1 protein were observed (Fig. 2, top), a primary peak at approximately 18.8 min with an estimated molecular mass of 82 kDa (Table  III) and a minor peak of an (ϳ200 kDa) aggregate. No peaks corresponding to monomeric or dimeric NM23-H1 were observed. The derived molecular mass estimate of 82 kDa is somewhat less than the predicted molecular mass of 105-120 kDa reported for hexameric NM23-H1 and NM23-H2, based on a monomeric molecular mass of 17-19 kDa (35)(36)(37). This difference could be the result of anomalous interactions with the gel filtration matrix, as has been observed by others. 3 It should also be noted that tetrameric structures have been reported for NM23-H2 based on gel filtration analysis, as well as for Myxococcus xanthus NDPK by direct assessment of the crystal structure (36). Nevertheless, DNA cleavage activity (Fig. 2, bottom) coeluted precisely with the primary peak of oligomerized NM23-H1 protein (middle), providing additional evidence that the activity is not because of copurifying or nonspecifically associated nuclease contaminants.
Cleavage activity was also inhibited by KCl, with the optimal KCl concentration less than or equal to 20 mM (panel B). ATP and its non-hydrolyzable analogue, ATP␥S, inhibited DNA cleavage activity to the same extent (50 -60%; panel C). This equivalence indicates that phosphate transfer does not underlie the inhibitory action of ATP. It has been suggested that ATP inhibits the cleavage activity of NM23-H2 indirectly through the sequestration of Mg 2ϩ (19), although interactions with active or allosteric sites of the enzyme have not been ruled out.
NM23-H1 Is a 3Ј-5Ј Exonuclease-To demonstrate directly that single nucleotide moieties were liberated from the 3Ј terminus by NM23-H1 via a bona fide 3Ј-5Ј exonuclease mechanism, cleavage reactions were conducted with two distinct DNA substrates that were 32 P-radiolabeled at their respective 3Јterminal nucleotides. Incubation of NM23-H1 with 3Ј-end labeled 5Ј-SHSas resulted in the time-dependent appearance of a diffuse band corresponding to the mobility of a single nucleotide unit, consistent with authentic 3Ј-5Ј exonuclease activity (Fig. 4A, left). In a series of reactions conducted simultaneously with the 5Ј-end labeled 5Ј-SHSas reactions, 3Ј-5Ј exonuclease activity proceeded with significantly faster kinetics (right; note rate of probe disappearance). A second DNA substrate derived from the NHE sequence of the PDGF-A promoter, consisting of a 20-bp double-stranded portion and a single-stranded 8-nt extension at the 3Ј terminus of its G-rich coding strand, was analyzed in similar fashion. As was seen with the 5Ј-SHSas sequence, 32 P-labeled nucleotide units were released from the 3Ј terminus by NM23-H1 (Fig. 4B, left), and the process was significantly slower than that seen with 5Ј-end labeled DNA (right). Interestingly, analysis of the 5Ј-end labeled DNA revealed a strong pause in cleavage upon reaching the final unpaired nucleotide, suggesting the enzyme activity was much higher with single-stranded DNA and required at least 1-2 3 I. Lascu, personal communication. overhanging nucleotides at the 3Ј terminus for optimal activity. Taken together, these results demonstrate that NM23-H1 is a 3Ј-5Ј exonuclease that cleaves single-stranded DNA in either Cor G-rich forms.
Characterization of the DNA Substrate Specificity of NM23-H1-The current studies strongly suggested that the 3Ј-5Ј exonuclease activity of NM23-H1 was much more pronounced with single-stranded than double-stranded DNA. To examine this apparent preference for single-stranded DNA, four substrates were prepared with each containing a double-stranded region and mismatched 3Ј-overhangs of progressively increasing length (1, 2, 3, and 9 nt; Fig. 5A). In each case, the enzyme proceeded rapidly through the mismatched 3Ј-terminal nucleotides but slowed markedly upon reaching the final unpaired 3Ј-terminal nucleotide and duplex DNA. In an enzyme titration experiment using the 9-nt mismatched substrate, DNA cleavage occurred at lower than stoichiometric amounts of NM23-H1 (Fig. 5B). Blunt-ended NHE sequences containing short internal mismatches of 1 and 3 nt were poor substrates for the enzyme (data not shown).
The enzyme failed to exhibit significant cleavage with a variety of other substrates derived from the GC-rich NHE, including a 22-bp blunt-ended duplex (Fig. 6A) and a counterpart containing a recessed 3Ј terminus 6 nt in length (panel B). Additional substrate specificity of the NM23-H1 exonuclease was demonstrated by the lack of cleavage of an 80-bp duplex in the absence (panel C, lanes [1][2][3][4] or presence (lanes 5-8) of a large internal 21-bp bubble, the latter shown previously to promote the 3Ј-5Ј exonuclease activity of the Werner syndrome protein, WRN (39). As expected, another WRN substrate consisting of the same sequence but harboring a 3Ј-overhang of 56 The primary peak of wild-type NM23-H1 protein as indicated by A 280 was observed at 18.76 min. Fifty-l fractions were collected and a 6-l aliquot from each fraction was analyzed by either SDS-polyacrylamide gel electrophoresis and Coomassie Blue staining (middle, ϳ2 g in peak fraction) or DNA cleavage assay (bottom, ϳ1 g in peak fraction). Indicated above the middle and bottom panels are the elution times of the fractions analyzed, with each number representing the beginning of the collection interval. Cleavage assays were conducted with 10 fmol of 32 P-labeled 5Ј-SHSas probe in the presence of 2 mM MgCl 2 and 100 mM KCl at room temperature for 1 h. Cleavage products were separated on a denaturing 10% polyacrylamide gel and visualized by phosphorimaging. Mobility of the 32 P-labeled 5Ј-SHSas probe (arrow) and cleaved products (bracket) are identified at the right.  15-20). In a similar vein, DNA cleavage was impaired by the addition of cordycepin, a dAMP analogue (3Ј-dAMP), to the 3Ј terminus of the 9-nt mismatched NHE-based substrate (Fig.  7). In summary, the 3Ј-5Ј exonuclease activity of NM23-H1 cleaves single-stranded DNA substrates efficiently without apparent nucleotide sequence specificity, whereas failing to cleave blunt-ended DNA or regions of internal mismatching,

FIG. 3. DNA cleavage activity is dependent on magnesium and is greater in low salt conditions.
A, 1 g of wild-type NM23-H1 protein was incubated for 1 h at room temperature with 10 fmol of 32 P-labeled 5Ј-SHSas and the concentrations of divalent cations as indicated at the top of the figure. Reaction products were analyzed on a denaturing acrylamide gel and visualized by phosphorimaging. B, cleavage assays were conducted for 2 h at room temperature with 0.5 g of H1 protein in the presence of KCl (20, 50, and 100 mM) and MgCl 2 concentrations (0, 2, 5, and 10 mM) denoted in the figure. Reaction products were analyzed on nondenaturing 6% acrylamide gels. Represented below panels A and B is oligonucleotide 1 (*1; see Table I), the radiolabeled DNA substrate used in these analyses. C, 1 g of NM23-H1 was incubated with 10 fmol of 5Ј-32 P-labeled 5Ј-SHSas in the absence or presence of 1 mM ATP or ATP␥S at room temperature for 20 min. DNA products were resolved by denaturing gel electrophoresis (12% polyacrylamide gel) and visualized by phosphorimaging (left). Bands corresponding to cleavage products were quantified to derive the amount of nucleotides cleaved by NM23-H1 (total nucleotides cleaved), as shown at right. P-labeled at either their 5Ј or 3Ј termini and incubated with 1 g of wild-type NM23-H1 protein under standard cleavage assay conditions ("Experimental Procedures"). A, NM23-H1 was incubated with 5Ј-SHSas substrate radiolabeled at the 3Ј (left) or 5Ј terminus (right) for the indicated lengths of time. The electrophoretic mobility corresponding to that of a single nucleotide unit (1 nt) is indicated with an arrow. B, shown are results obtained with a partial duplex DNA substrate containing an 8-nt overhanging 3Ј terminus, radiolabeled at either the 3Ј (left) or 5Ј terminus (right). A strong stop in cleavage activity corresponding to the onset of duplex DNA 20 nt (20 nt) from the 5Ј-terminal radiolabel (right panel) is indicated with an arrow. Reaction products were analyzed on denaturing 14% polyacrylamide gels. and it requires the presence of a free 3Ј-hydroxyl group for maximal catalysis.
Lys 12 , but Not His 118 , Is Required for DNA Cleaving Activity of NM23-H1-To characterize the active site of the 3Ј-5Ј exonuclease-like activity of NM23-H1, three different point mutations were introduced into key amino acid residues of the molecule. The first of these, a lysine 12 to glutamine (K12Q) substitution, was employed previously to demonstrate the necessity of Lys 12 for both the cleaving and NDPK activities of NM23-H2 (22). Similarly, an Arg 34 to alanine mutation (R34A) was introduced into NM23-H1 based on its utility in revealing the importance of this residue for the DNA binding properties of NM23-H2 (40). Finally, an amino acid substitution at His 118 (H118F) that prevents the formation of a phosphorylated in-termediate required for NDPK activity (21,41) was employed to examine further the relevance of NDPK to the DNA cleaving mechanism.
For functional analysis, each of the three NM23-H1 mutants was purified to apparent homogeneity by the same procedure used for wild-type NM23-H1 (Fig. 1, panels B-D). Detectable amounts of 3Ј-5Ј exonuclease-like activity coeluted with each mutant NM23-H1 variant during HTP chromatography. Whereas the K12Q eluted at precisely the same position of the phosphate gradient as the wild-type protein (panel B, 82-83 min), the R34A and H118F mutant proteins (panels C and D) eluted much earlier (64 and 67 min, respectively). The coelution of DNA cleavage activity with the R34A and H118F proteins despite their very different elution behavior provides FIG. 5. NM23-H1 digests overhanging mismatched 3 termini from double-stranded DNA templates. A, DNA substrates derived from the NHE of the PDGF-A promoter region and containing varying lengths of overhanging nt (1, 2, 3, and 9 nt) at their 3Ј termini as shown below the figure were 32 P-labeled on their respective C-rich strands and incubated with 1 g of NM23-H1 at room temperature for the indicated times. Arrows shown at the right of each time course indicate the size of DNA segments remaining after trimming of unpaired nucleotides. B, the indicated amounts of NM23-H1 (1 fmol to 10 pmol) were incubated with ϳ15 fmol of the indicated 3Ј-overhanging substrate at 22°C for 20 min. Bands corresponding to cleavage products were quantified by phosphorimaging to derive the amount of nucleotides (NT) cleaved, as shown in the inset.
further evidence the observed exonuclease activity is associated with NM23-H1 and not a contaminant. Moreover, 3Ј-5Ј exonuclease-like activity was only observed in NM23-H1-containing fractions; no activity was detected between 60 and 81 min for wild-type and K12Q, nor between 71 and 89 min for R34A and H118F.
To measure the extent to which the mutations affected secondary structure of the NM23-H1 molecule, the purified pro-teins were further analyzed by gel filtration HPLC and circular dichroism (CD) spectrometry. Molecular weight estimates obtained by gel filtration HPLC for R34A and H118F were not significantly different from that of wild-type NM23-H1 (Table  III). K12Q exhibited a very small, but statistically significant decrease (6 kDa) in apparent size. This small increment was not consistent with an effect on oligomeric structure, and was possibly because of anomalous interactions of this charge-al-

3Ј-5Ј Exonuclease Activity of NM23-H1
tered variant with the gel filtration matrix. The CD spectra obtained for the wild-type form of NM23-H1 (Fig. 8) were similar to those reported in an earlier report using this procedure (32) and were consistent with a highly ordered structure, with secondary structure estimates of 22% total ␣-helix, 32% ␤-sheet, 17% turns, and 29% random (Table IV). The CD spectra associated with the H118F and K12Q variants were similar to each other, but yielded estimates of subtly higher helical and lower sheet content than the wild type. Interestingly, the CD profile for the R34A mutant was highly atypical, displaying an almost total loss in ␣-helical structure and a shift toward higher random and possibly higher ␤-sheet content. Crystal structure analyses of NM23-H2 indicates that Arg 34 is located within a ␤-strand (␤2) interposed between two ␣-helices (␣1 and ␣A; Ref. 36), suggesting that the alanine substitution at this cationic residue may have destabilized these adjacent structures. Whereas the destabilizing effect of the R34A mutation has been seen in a majority of preparations, some do exhibit unaltered structure (data not shown). Taken together, the gel filtration and CD spectral analyses indicate that each of the NM23-H1 proteins employed in this study exhibited normal oligomeric structure, with the R34A mutant displaying conformational instability.
NDPK activity was measured using a coupled pyruvate kinase/lactate dehydrogenase assay as described previously (10,21). The average specific NDPK activity of wild-type NM23-H1 was 570 units/mg (Table III). The K12Q and H118F mutations resulted in dramatic reductions in NDPK activity, whereas the R34A mutation was only modestly inhibitory. The specific activity of wild-type NM23-H1 and the effects of the disruption of His 118 were consistent with previous reports in the context of NM23-H1 (21, 41) and NM23-H2 (42). The identification of Lys 12 as a critical participant in the NDPK activity of NM23-H1, while not unexpected (22), is novel.
To assess the effects of these amino acid substitutions on DNA cleaving activity, equal amounts of wild-type and mutant proteins were incubated with 5Ј-end labeled 5Ј-SHSas over a 2-h time course. Wild-type NM23-H1 elicited the rapid appearance of products representing removal of 1-2 nucleotides from the 3Ј terminus within 5 min (Fig. 9A, lanes 2-5). An accumulation of DNA fragments ranging between 5 and 9 nt was observed with longer incubation times, probably representing the minimal DNA substrate length for the enzyme. These fragments are unlikely to be the result of endonuclease activity because 3Ј-end labeled substrates fail to yield fragments consistent with internal cleavage of 5-9 nt from the 5Ј terminus (Fig. 4). This is reminiscent of similar observations with the 3Ј-5Ј exonuclease activity from bacteriophage Ø29 (43) and a WRN-like 3Ј-5Ј exonuclease recently described in Arabidopsis thaliana (44). Longer incubations also yielded an array of fragments ranging between 15 and 33 nt, suggestive of a time-dependent decay of enzyme activity. The NDPK-null mutant, H118F, generated significantly less of the fully digested fragments (5-9 nt) than the wild-type protein, while still yielding a considerable quantity of the 15-33-nt series of fragments (Fig.  9A, lanes 6 -9). Production of fully and partially digested DNA fragments was even more markedly reduced with the K12Q and R34A mutant proteins, evidenced clearly by the much slower rate of depletion of the full-length DNA substrate (lanes 10 -13 and 14 -17, respectively). This hierarchy of cleavage activity (wild-type Ͼ H118F Ͼ Ͼ K12Q Ͼ R34A) was consistently seen with three or more replicate preparations of each protein (data not shown). The impairment in 3Ј-5Ј exonucleaselike activity for the K12Q mutant indicates a potential role for Lys 12 in the catalytic mechanism, as shown previously for NM23-H2. In contrast, the reduced 3Ј-5Ј exonuclease-like activity for the R34A mutant was possibly secondary to effects on the overall secondary structure of the molecule. When taken in the context of the previously reported inhibitory effect of the R34A substitution in the NM23-H2 molecule on binding of DNA substrates (19), our data indicates this mutation may result in an overall unraveling of the DNA binding pocket. Importantly, the diminished 3Ј-5Ј exonuclease-like activity of the K12Q and R34A proteins argues strongly against the possibility of artifactual contamination of our purified preparations by bacterial nuclease(s).

DISCUSSION
The 3Ј-5Ј exonuclease activity of NM23-H1 represents a novel catalytic property of the NM23/NDK family of proteins. Previous studies of NM23-H2 have revealed a DNA cleavage activity that recognizes internal sites within the NHE of the c-myc promoter (19). This cleavage occurs in the context of double-stranded DNA and is directed to repeated sequence motifs within the NHE, yielding double-stranded breaks with 5-nt overhangs at the 3Ј termini of the cleavage fragments. The cleavage mechanism was further characterized as DNA glycosylase/lyase-like, a feature often associated with base excision repair nucleases, and was shown to occur via a nucleophilic attack by the amine group of Lys 12 on the glycosylic bond of deoxyribose (22). In addition, the E. coli homolog NDK has recently been shown to possess a uracil-DNA glycosylase function that excises uracil bases from both single-and doublestranded DNA templates (45). NDK also excises apurinic (AP), uracil-deleted sites via an AP endonuclease/lyase activity. Finally, the enzyme also exhibits a phosphodiesterase activity   (48), and 5Ј AP endonuclease (49) activities, the latter proposed as its main physiological role (50). APE1 also exhibits many of these same multiple functions (for review, see Ref. 51). The extent to which each of the DNA modifying activities that have been associated with NM23/NDK proteins is shared across different isoforms and species of origin remains to be determined. Site-directed mutagenesis of NM23-H1 highlighted both similarities and differences with the NM23-H2 isoform in terms of structure and function. Amino acid substitutions within the NM23-H1 molecule at residues Lys 12 and Arg 34 , residues implicated previously in the DNA cleaving (22) and DNA binding (40) properties of NM23-H2, resulted in significantly reduced 3Ј-5Ј exonuclease activity. Although the inhibitory effect of the K12Q mutation in NM23-H1 appears to reflect an important role for Lys 12 in the initial attack of the glycosidic bond, as described for NM23-H2 (22), some residual 3Ј-5Ј exonuclease activity was apparent. This contrasts somewhat with the nearly total ablation of DNA cleavage activity achieved with the analogous mutations in NM23-H2 and suggests either that a residue other than Lys 12 is the primary player in 3Ј-5Ј exonucleolytic attack, or that an alternative residue might partially substitute for Lys 12 in the K12Q mutant protein. The diminished activity of the R34A mutant was apparently secondary to a destabilizing effect on protein structure, an effect not seen on either the K12Q or H118F mutant forms. The H118F mutant displayed a robust 3Ј-5Ј exonuclease activity concomitant with a complete loss in NDPK activity, appearing to exclude an absolute interdependence between these functional domains. However, the H118F mutant consistently yielded smaller amounts of fully digested DNA fragments relative to wild-type NM23-H1 (Fig. 9A). It should be noted that the only DNA template producing these very small (6 -9 nt) fragments was the single-stranded 5Ј-SHSas sequence. The significance of this observation is unclear at present, but could reflect unusual conformations assumed by this single-stranded oligode-oxyribonucleotide, novel interactions of the protein with those structures, and a possible role for His 118 in recognizing them.
The current study adds NM23-H1 to an expanding list of mammalian proteins that exhibit 3Ј-5Ј exonuclease activity. 3Ј-5Ј Exonucleases are believed to be primarily involved with maintaining the fidelity of DNA synthesis, often functioning as proofreading enzymes during DNA replication, repair, and recombination (for review, see Ref. 28). Many of these activities are associated with DNA polymerases themselves (e.g. DNA polymerases ␥, ␦, and ⑀; Refs. [52][53][54]. However, the lack of DNA polymerase activity associated with NM23-H1 places it in the alternative category of autonomous 3Ј-5Ј exonucleases, as described by Shevelev and Hü bscher (28). This grouping includes the TREX proteins Trex 1 and TREX2 (55), Werner syndrome protein WRN (56,57), hRAD1 (58), hRAD9 (59), hMRE11 (60), the apurinic endonuclease APE1 (61,62), the tumor suppressor p53 (63), and the V(D)J recombination enzyme VDJP (64). The overall knowledge of the biochemical properties and function of autonomous 3Ј-5Ј exonucleases is quite limited at present, although evidence is accumulating to suggest they can augment the proofreading function of DNA polymerases (65), as well as associate with and improve the fidelity of higher order protein complexes (replisomes) associated with DNA replication (66).
NM23-H1 concentrations employed in the current study for characterization of 3Ј-5Ј exonuclease activity were in the range of 200 -600 nM, although DNA cleavage was detected at concentrations as low as 3.3 nM (Fig. 5B). These concentrations, which are similar to those reported for the study of the DNA cleaving activity associated with the NM23-H2 isoform (19,22), are relatively high as compared with those used in assays of some 3Ј-5Ј exonucleases. For example, only picogram quantities of enzyme were required for determination of K cat values for 3Ј-5Ј exonuclease activity for DNA polymerases of bacterial and phage origin, which were reported in the range of 170 -550 s Ϫ1 (43,(67)(68)(69). Trex1 and TREX2, which have been reported to constitute the bulk of autonomous 3Ј-5Ј exonuclease activity in mammalian cells, also exhibit relatively high specific activity (K cat ϳ 10 -20 s Ϫ1 ; Ref. 70). The 3Ј-5Ј exonuclease activity of NM23-H1 is consistent, however, with the majority of other autonomous 3Ј-5Ј exonucleases, which possess considerably less activity and require higher enzyme concentrations for functional characterization. In this regard, WRN and a WRN-  (H118F, K12Q, R34A) of NM23-H1 protein was incubated with 10 fmol of 32 P-labeled 5Ј-SHSas oligodeoxynucleotide under standard cleavage assay conditions ("Experimental Procedures") for the indicated lengths of time. The cleavage products were resolved on a 12% denaturing polyacrylamide gel. Indicated to the right of the WT lanes are the DNA fragment lengths generated, assigned relative to sequencing ladders (not shown). B, bands corresponding to cleavage products were quantified by phosphorimaging to derive the amount of nucleotides cleaved by wild-type and mutant NM23-H1 (total nucleotides cleaved), as shown at right. like protein from Arabidopsis (AtWRNexo-p) both exhibit significant DNA cleaving activity only when analyzed at concentrations ranging between 1 and 20 nM, with incubation at 30°C for 5-30 min (39,57,71,72). Even higher concentrations were employed to observe 3Ј-5Ј exonuclease activity in the analysis of hRAD1 ( 64). For these proteins exhibiting lower 3Ј-5Ј exonuclease activity, precise kinetic assessments have yet to be reported. In the case of NM23-H1, we have obtained an estimate of K m on the order of 1 M (data not shown), a low affinity that essentially precludes measurement of maximal enzymatic rates in the presence of DNA substrate excess required. Application of the heparin trap method has revealed that the 3Ј-5Ј exonuclease activity of NM23-H1 is essentially nonprocessive (data not shown), providing a potential explanation for its lower activity relative to the highly processive 3Ј-5Ј exonuclease activities of DNA polymerases (69,73).
NM23-H1 and other autonomous 3Ј-5Ј exonucleases are likely to function as components of higher order protein complexes, a potential explanation for their relatively low specific activity in purified form. In this regard, hRAD1 and hRAD9 have been shown to associate with the protein Hus1 in the formation of the so-called 9-1-1 complex, a replisome that serves a critical checkpoint function in response to DNA damage (74 -76). Other illustrations of such interactions are provided by: the enhancement of 3Ј-5Ј exonuclease activity of hMRE11 by Rad50, both of which are present in the MRE11 complex associated with double-strand DNA break repair (60); the stimulation of 3Ј-5Ј exonuclease proofreading activity of T4 DNA polymerase by other T4 replication proteins (77); the stimulation of RAG1/RAG2-mediated V(D)J cleavage by the high mobility group proteins HMG1 and HMG2 (78); and, the induction of a switch from exonuclease to endonuclease activity for the Artemis protein by DNAPK cs (79). Interestingly, NM23-H1 has been recently shown (27) to be an important component of a latent 270 -450-kDa complex in the endoplasmic reticulum that also contains HMG2 and APE1, the nucleosome assembly protein SET, and the tumor suppressor pp32 (also known as PHAPI, I 1 PP2A ). The latent SET complex is liberated following attack of virus-infected or tumor cells by cytotoxic T lymphocytes, with NM23-H1 playing a critical role as the initiator of DNA nicking in a caspase-independent pathway of apoptosis induction. The coordinated release of NM23-H1 with APE1 and HMG2, proteins implicated previously in DNA repair processes, suggests the possibility of functional collaboration not only in the induction of apoptosis but in the physiological response to DNA damage, including the various DNA modifying activities of NM23-H1. Indeed, we have observed that NM23-H1 accumulates within nuclear microstructures in response to treatment of human tumor cell lines (HeLa and HepG2) with such DNA damaging agents as etoposide and cisplatin, suggesting recruitment to sites of DNA repair. 4 Studies are currently being initiated to isolate the NM23-H1-containing nuclear complex and to assess 3Ј-5Ј exonuclease activity further in that context.
The relatively low specific activity of the 3Ј-5Ј exonuclease associated with recombinant human NM23-H1 necessitated a rigorous demonstration that the activity was not conferred by contaminants from the E. coli expression host. First, a thorough analysis of individual column fractions from the final HTP-HPLC purification step demonstrated a precise and reproducible coelution of 3Ј-5Ј exonuclease activity with each preparation of wild-type and mutant NM23-H1. This coelution was observed even with two mutants (H118F and R34A) that eluted much earlier in the phosphate gradient than the wildtype and K12Q variant. Second, coelution of cleavage activity with the NM23-H1 protein during gel filtration HPLC strongly suggests identity between the two and effectively excludes the presence of a bacterial exonuclease in physical association with the 82-kDa oligomer of NM23-H1. Third, preparations of K12Q and R34A exhibited significantly lower exonuclease activity than wild-type NM23-H1, a reproducible result that was inconsistent with expectations for an enzyme contaminant.
The 3Ј-5Ј exonuclease activity we have identified in NM23-H1 represents a highly plausible candidate function underlying metastasis suppression in human cancer. Deficiencies in proofreading 3Ј-5Ј exonuclease and mismatch repair activities have been shown to elicit the mutator phenotype in yeast (80) and in mice (81). Acquisition of the mutator phenotype is widely recognized to be a key event in the accelerated chromosomal and microgenetic changes underlying the progression of cancer to its ultimate lethal forms (29). In this context, the observation that disruption of the NDPK gene (ndk) in E. coli elicits a mutator phenotype (38) is noteworthy. Although the mechanism was attributed to imbalances in nucleotide pools, our studies suggest that loss of ndk-mediated DNA repair and an associated 3Ј-5Ј exonuclease function may also play a role. Whereas these observations support a model in which suppression of metastasis and the mutator phenotype by NM23-H1 is mediated by its 3Ј-5Ј exonuclease activity, other viable candidate functions must also be considered. The recent identification of NM23-H1 as the DNA nicking component of the SET complex in cytotoxic T lymphocyte-induced apoptosis suggests a role for the 3Ј-5Ј exonuclease in mediating both immunologic and cell cycle checkpoint elimination of tumor cells. Moreover, both NM23-H1 and NM23-H2 exhibit transcription regulatory activities, which could be attributable to nuclease-mediated remodeling of regulatory elements with unusual non-B-form conformations (for reviews, see Refs. 20 and 24). In this regard, NM23 proteins have been shown to regulate transcription of such cancer-relevant genes as c-myc, which is enhanced by NM23-H2 expression (17,18), and PDGF-A, which is repressed by both NM23-H1 and NM23-H2 (23). Further research into the linkage between the 3Ј-5Ј exonuclease activity of NM23-H1 and these potential mediators of malignant progression appears to be warranted.