Regulatory GTP-binding Proteins (ADP-Ribosylation Factor, Gt, and RAS) Are Not Activated Directly by Nucleoside Diphosphate Kinase*

The expression of nucleoside diphosphate kinase (NDK) genes has been implicated as a negative regu- lator of murine and human tumor metastases and is critical to proper development in Drosophilia mela-nogaster. Molecular mechanisms for the role(s) of NDK in these complex processes have not yet been eluci-dated, but several reports have suggested that these and many other signal transduction pathways may be activated by NDK acting directly on a regulatory GTP- binding protein(s). To test this hypothesis, we examined the ability of NDK to catalyze the phosphorylation of the GDP bound to the following three members of the superfamily of regulatory GTP-binding proteins: G,, Ha-ras p21, and ARF. We have found no evidence to support the hypothesis that NDK can directly activate any GTP-binding protein. Rather, evidence is presented which clearly shows that all of the GTP formed upon incubation of GTP-binding proteins with NDK is the result of NDK utiliz-ing free GDP as substrate. The GDP bound to the regulatory proteins is not a substrate for NDK under conditions in which free nucleotides are rapidly and efficiently phosphorylated. The importance of appropriate controls for dissociation of GDP from the regulatory proteins both during the NDK reaction and dur- ing the analysis of product is demonstrated. We believe there is currently no experimental evi- dence to support the hypothesis that NDK

The expression of nucleoside diphosphate kinase (NDK) genes has been implicated as a negative regulator of murine and human tumor metastases and is critical to proper development in Drosophilia melanogaster. Molecular mechanisms for the role(s) of NDK in these complex processes have not yet been elucidated, but several reports have suggested that these and many other signal transduction pathways may be activated by NDK acting directly on a regulatory GTPbinding protein(s). To test this hypothesis, we examined the ability of NDK to catalyze the phosphorylation of the GDP bound to the following three members of the superfamily of regulatory GTP-binding proteins: G,, Ha-ras p21, and ARF.
We have found no evidence to support the hypothesis that NDK can directly activate any GTP-binding protein. Rather, evidence is presented which clearly shows that all of the GTP formed upon incubation of GTPbinding proteins with NDK is the result of NDK utilizing free GDP as substrate. The GDP bound to the regulatory proteins is not a substrate for NDK under conditions in which free nucleotides are rapidly and efficiently phosphorylated. The importance of appropriate controls for dissociation of GDP from the regulatory proteins both during the NDK reaction and during the analysis of product is demonstrated.
We believe there is currently no experimental evidence to support the hypothesis that NDK can directly activate a regulatory GTP-binding protein.
Nucleoside diphosphate kinase (NDK)' is a ubiquitous enzyme first described in 1953 (1,2). NDK catalyzes the reaction NlTP + NzDP + NIDP + NZTP thus, shuttling the y-phosphate between different nucleotides. The mechanism of catalysis by NDK has been studied extensively and involves formation of a phosphorylated enzyme intermediate in a classic ping-pong mechanism (for reviews * 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. § T o whom correspondence should be addressed Laboratory of Biological Chemistry, National Cancer Institute, Bldg. 37, Rm. 5D- see Refs. 3,4). In mammalian tissues, the enzyme is a hexamer of 17-kDa subunits migrating on sizing columns with an apparent molecular mass of 100,000-120,000 Da. Multiple species with activity can be identified by ion-exchange chromatography. These isozymes differ both in equilibrium kinetics and activation energies. It is likely that these multiple isozymes result from as few as two gene products. In human erythrocytes, two distinct NDK protein chains have been identified and purified to homogeneity (5). One has an acidic PI (A) and the other a basic PI (B). Hexamers formed from different combinations of these monomers (e.g. Ag, A5B1, A&. . ., Be) appear to account for the multiple isozymes and a mixture of these two gene products appears to recreate the complex mixture of isoforms found in crude cell extracts. The A monomer is the product of the nm23-H1 gene, the human gene homologue of the mouse gene, nm23-1 (6,7). B is the product of nm23-H2, a second gene cloned from a human cDNA library using an nm23-1 probe (8).
Initially the function of NDK was thought to be the maintenance of nucleoside triphosphates at the expense of adenosine triphosphate. However, both the abundance and distribution of NDK activity appear somewhat inconsistent with this proposed function. For example, NDK activity is about 10-fold greater than that of the glycolytic enzymes in red blood cells (3,9), and NDK constitutes between 0.1-0.2% of total cellular protein in bovine brain (10). Other reports have found discrepancies between the nucleotide pools found in certain cells and the predicted values based on in vitro derived kinetic constants for NDK and the different nucleotide substrates (3,4). These data have prompted speculation (3,4) on additional, regulatory roles for NDK in cellular physiology. Interest in the possible regulatory roles for this enzyme has recently increased dramatically as NDK genes have been implicated as negative regulators of tumor metastasis in mice (7,11) and as an important regulatory gene in fly development (6,12). It is difficult to imagine how either of these complex processes might be regulated by NDK activity affecting nucleoside triphosphate pools. With the large number of regulatory GTP-binding proteins, and proposed functions for them, it has been proposed (e.g. [13][14][15][16][17][18]) that one mechanism by which NDK may effect either specific or pleiotropic regulation of cellular functions is through direct action on a GTPbinding protein or proteins.
NDK-mediated activation of GTP-binding proteins implies that NDK catalyzes the transfer of phosphate from an NTP to GDP bound to the GTP-binding protein, thereby activating the protein without the need for nucleotide exchange. As the release of GDP has been shown to be the rate-limiting step in the activation of a number of regulatory GTP-binding proteins, the formation of GTP in situ is predicted to dramatically increase the activity of the GTP-binding protein.
Two approaches have been used to investigate this potential regulatory role for NDK. First, the formation of GTP has G Proteins Are Not Activated Directly by N D K 18183 been examined. Thus, NDK has been shown to catalyze the transfer of phosphate from ATP to GDP that had been bound to ras (19), EF-al (19), tubulin (20, 21), G, (15), G, (19), and, most recently (lS), ADP-ribosylation factor (ARF), a 21-kDa GTP-binding protein that regulates protein traffic in eukaryotes. Second, NDK has been shown to activate GTP-binding proteins in vitro. For example, ATPrS-dependent activation of adenylate cyclase in platelet membranes (22), acetylcholine-activated K+ channels in cardiac myocytes (23), and NADPH-oxidase in HL-60 membranes (24) were all reported t o be NDK-mediated. ADP-ribosylation factor activity (18) and the GTPase activity of G, (15) have been reported to increase with added NDK and ATP. However, in very few of these studies has an attempt been made to document that the GDP is actually bound to the GTP-binding protein when phosphorylated by NDK. This is a critical distinction as addition or production of GTP in solution, in the absence of receptor and agonist, is known to have little or no effect on any GTP-binding protein-mediated effector system. A clear demonstration of a GTP-binding protein being activated by NDK absolutely requires proof that the phosphorylation of GDP occurs while bound to the regulatory protein. We believe that this condition has not been met by any published report t o date. Omission or improperly controlled tests of this condition may have led to erroneous conclusions regarding the actions of NDK.
In this report, we have examined GDP bound to ARF, Gt, and Ha-ras p21 as potential substrates for NDK, specifically testing whether phosphate is transferred to the bound GDP. We were unable to demonstrate such a transfer under conditions in which phosphate was efficiently transferred to free nucleoside diphosphate. Several artifacts are described and documented which we believe account for all of the published data that supported the conclusion that a GTP-binding protein is activated by NDK when, in fact, such an event may not occur.

Reagents
Recombinant mARFlp (25), nm23-lp (18), and GPlyZ2 were puril'ied from bacteria or Sf9 cells as previously described. nm23-Hlp and nm23-H2p were expressed in bacteria and purified by the same procedure previously described for nm23-lp (18). Recombinant Harm p21 and (G12V) Ha-ras p21, purified from bacteria as described (27), were the generous gifts from Dr. Richard Michitsch (Oncogene Science Inc., Manhasset, NY) and Drs. Douglas Lowy and Alex Papageorge (NCI, Bethesda, MD). G,, G,a, rod outer segment discs, and urea-washed discs containing rhodopsin were purified from bovine retina as described (28,29). Bovine liver nucleoside diphosphokinase (catalog no. N-2635), ATP, GTP, L-a-dimyristoyl phosphatidylcholine, and Sephadex G-25 were purchased from Sigma. recombinant GPlyB (10 nM) and rhodopsin (10 nM; present in ureawashed rod outer segment discs in a buffer containing 10 mM MOPS, pH 7.5, 1 mM EDTA, 3 mM MgS04, 1 mM dithiothreitol, 3 mg/ml BSA, and 1 p~ [a"P]GTP at 30 "C for 120 min). The membranes and rhodopsin were removed by centrifugation a t 75,000 rpm in a Beckman TL1OO.l rotor at 4 "C for 10 min. Free nucleotides were removed by gel filtration on Sephadex G-25 developed in the binding reaction buffer to which 100 mM NaCl was added. Analysis of bound nucleotides revealed all of the radionucleotide was in the form of GDP, due to intrinsic hydrolysis. The heterotrimer was loaded in the same manner except the purified G, was used and no P-y subunits were added.

Nucleoside Diphosphate Kinase Assay
Coupled Spectrophotometric Assay-NDK activity was determined by the method of Agarwal et al. (4) which uses dTDP as substrate in an enzyme-coupled assay resulting in oxidation of NAD by lactate dehydrogenase.
Radioisotopic Assay-Enzyme activity was measured as GTP formation from [~u-~'P]GDP. Except where noted otherwise in the text, reactions contained 25 mM HEPES, pH 7.4, 2.5 mM MgCl2, 1 mM ditbiothreitol, 1 mM ATP, and 100 pM GTP with the indicated concentrations of nucleotide diphosphate substrate and nucleoside diphosphate kinase in a volume of 100 pl. Reactions were stopped in one of three ways as noted in the text: 1) direct application of a 5-10-pl sample to a PEI-cellulose plate; 2) adding the sample to ethanol to achieve a final ethanol concentration of 50 or 67%; or 3) adding the sample to formic acid to a final formic acid concentration of 2,4, or 6 M. Guanine nucleotides were separated by ascending chromatography on PEI-cellulose plates developed in 2 M LiCI, 2 M formic acid (l:l, v/v). The nucleotides were visualized by autoradiography, identified by comigration with nucleotide standards, and quantified by scintillation spectroscopy. The amount of GTP formed is typically expressed as the percentage of total guanine nucleotide.

Nitrocellulose Filter Trapping Assay
This assay was performed essentially as described in Ref. 31. Samples of 10-20 pl were diluted into 2 ml of ice-cold 20 mM Tris, pH 7.4, 10 mM MgCl,, 100 mM NaCl, 1 mM dithiothreitol (TNMD) with or without 200 pg/ml bovine serum albumin as indicated in the text. The samples were filtered on nitrocellulose filters, and the filters were washed six times with 2 ml of cold TNMD.

RESULTS
We have initiated a formal test of the hypothesis that nucleoside diphosphate kinase can directly activate one or more regulatory GTP-binding proteins. Because activation of a wild type regulatory GTP-binding protein requires the production of a GTP-liganded protein (32) we have addressed whether, in a purified system, NDK can use ATP to produce G T P from GDP while the guanine nucleotide remains bound to a regulatory protein. To test this, we have employed [a-32P]GDP-bound proteins as substrates in NDK reactions after removal of unbound labeled nucleotide. We considered any G T P formed could have been produced by one or more of three possible reactions: 1) the GDP bound to the protein was phosphorylated by NDK while bound 2) the GDP was released from the protein during the incubation with NDK and was phosphorylated by NDK while free in solution; or 3) the G T P was produced during the analysis of products or "postr e a~t i o n . "~ Clearly conditions 2 and 3 must be eliminated or The term post-reaction will be used herein to refer to any reactions or changes in substrates or products after attempts are made to terminate the reaction containing NDK. This would include changes occurring (either chemically or catalyzed by NDK) in or as a result of exposure to ethanol, formate, or PEI-cellulose, as well as those occurring during separation or analysis of products.

G Proteins Are Not Activated Directly by NDK
accurately assessed before it is possible to conclude that possibility 1 has occurred. We chose to examine three regulatory GTP-binding proteins as representatives of this large superfamily. Gt and Gta were used as representatives of the heterotrimeric G protein family as conditions were known which would allow only slow exchange of bound GDP from this protein. Both the monomeric a subunit and the heterotrimer were used to test for any effects of the presence of P-y subunits on the reaction.
Ha-ras p21 was used as a representative of the RAS superfamily of monomeric GTP-binding proteins due to its significant role in oncogenesis. Both the proto-oncogenic form, Ha-ras p21, and an oncogenic form, (G12V) Ha-ras p21 were used as they have very different effects on cells in vivo, and the latter has a lower rate of intrinsic GTP hydrolysis. Finally, human ARFlp was used as a representative of the ARF subfamily of the RAS superfamily as it is quite distinct from the other subfamilies in the RAS superfamily and has unique guanine nucleotide binding properties which make it a likely candidate for an NDK-activated regulatory protein (18).
GDP Dissociation as a Potential Source of Artifact in NDK Assays-Conditions were sought which would minimize the rate of GDP dissociation from the regulatory proteins to allow a clear distinction between the rate of phosphorylation of bound GDP from that of free GDP. To monitor dissociation of GDP, proteins were loaded with [a-32P]GDP in a standard exchange reaction at 30 "C, as described under "Materials and Methods." Unbound nucleotides were then removed by gel filtration at 4 "C. The preloaded proteins were then incubated at 30 "C in the presence of 100 p~ GTP (this typically represents a 1,000-10,OOO-fold isotopic dilution) to exclude dissociation and reassociation as the pathway for formation of bound [a-3'P]GTP. After the indicated periods of time, 32P bound to protein was assessed either by nitrocellulose filter trapping or by fractionation on Sephadex G-25, as described under "Materials and Methods." In addition to those agents previously identified (e.g. [ M e ] , ionic strength, phospholipids (25,32,33)), the rate of GDP dissociatiqn from hARFlp ( Fig. 1) or G, was found to be sensitive to the total protein concentration. Stabilization of nucleotide-bound hARFlp at low concentrations was achieved by addition of a carrier protein, such as BSA. Albumin has no effect on the rate or extent of nucleotide exchange on ARF in a standard exchange reaction. As seen in Fig. 1, incubation of 30 nM [~U-~'P]GDP. ARF at 30 "C resulted in the loss of bound GDP. In the experiment shown, more than 40% of the nucleotide had dissociated after 60 min. This is likely the result of denaturation of the hARFlp in dilute solutions rather than exchange as it is readily prevented by protein addition, and no binding to hARFlp has been observed under these conditions. As little as 1-2 pg/ml protein dramatically slows the dissociation rate, and at total protein concentrations of greater than 10 pg/ml less than 10% of the bound GDP had dissociated (Fig. 1, A  and B ) . Protein stabilization of GDP-bound ARF was nonspecific, with ARF, BSA, and NDK all effectively preventing GDP-dissociation when present a t 25 pg/ml or greater (Fig.   1B). Similar results were obtained when Gt or G,a were tested.
At concentrations of 100 nM or less with no other protein present, all the preloaded GDP had dissociated within 2 h at 30 "C. Addition of 3 mg/ml BSA completely prevented nucleotide dissociation from Gt or Gta at 4 "C and substantially reduced dissociation at 30 "C. The denaturation of ARF in dilute solution motivated the re-evaluation of the nitrocellulose filter trapping assay as a quantitative method for the determination of guanine nucleotide binding. This is a standard means of monitoring nucleotide binding to proteins (31) in which protein-bound nucleotide and free nucleotide are rapidly and efficiently separated. Typically, the binding reaction is stopped by a combination of dilution and lowering the temperature. Inclusion of high concentrations of Mp"', which often slows nucleotide dissociation, is also useful to maintaining binding after dilution. For the assay to accurately reflect binding, exchange and denaturation must be minimal while the protein is in the stop solution and during filtration. If these conditions are met, recovery of binding sites is independent of time in stop solution. These properties have been documented previously for G proteins, ras p21, and ARF under specific conditions. Typically, ARF in a solution containing 3 mM DMPC and 0.1% sodium cholate (conditions required for nucleotide exchange on ARF (34)) is diluted 100-200-fold into exchange stop solution (TNMD) at 4 "C. As shown in Fig. 2, ARF. GDP recovery is independent of time in TNMD in this case. If, however, ARF is incubated in an NDK reaction mixture without other protein or DMPC and then diluted into TNMD, recovery of the ARF.GDP complex decreases exponentially with time. BSA (100 pg/ml) added to either the NDK reaction mixture or in the TNMD stabilizes ARF, and recovery is independent of time in TNMD at 4 "C. Thus, addition of protein or DMPC is sufficient to stabilize ARF when in dilute solutions. The earlier characterization of the stability and guanine nucleotide binding properties of ARF proteins, performed in a combination of DMPC and cholate, clearly cannot be extended to those conditions which lack phospholipids.
,. These data indicate that close attention must be paid to the conditions used to test for dissociation of GDP, including total protein concentration, phospholipids, salts, and metals. Without such controls the amount of free GDP produced as a possible substrate for NDK can be grossly underestimated (18).
Post-reaction Product Formation as a Potential Source of Artifacts in NDK Reactions-The method used to terminate the NDK reaction prior to analysis of product formation was also examined. Two methods of stopping NDK reactions, addition of the reaction mixture to ethanol (e.g. 20) or directly to the surface of a PEI-cellulose plate (e.g. 18,35), were tested to determine if complete arrest of NDK activity had occurred. Formic acid was also examined for the ability to instantly arrest any NDK activity as it would not interfere with subsequent nucleotide separations. Each of the GTP-binding proteins tested, G,, ARF, or Ha-ras p21, were preloaded with [ ~u-"~P]GDP, as described under "Materials and Methods" and used as substrates to test for any NDK activity occurring after the reactions were presumably stopped. Two solutions, one containing (G12V)Ha-ras p21. [G-~'P]GDP (0.3 pmol) in a buffer containing ATP and GTP, the other containing either 10 or 100 ng bovine liver NDK, were added simultaneously to either 1) the surface of a PEI cellulose plate, 2) 50% ethanol (final concentration), or 3) 6 M formic acid (final concentration). As seen in Fig. 3A, only the formate solution actually arrested the NDK activity. Stopping in either ethanol or on PEI-cellulose plates resulted in the conversion of more than 30% of the GDP to GTP. Thus, production of large amounts of [a-:'ZP]GTP occurred independently of any incubation at 30 "C and under conditions that have been used previously to terminate the reaction, specifically on PEI plates or in ethanol solutions. Similar results were obtained when the labeled GDP was bound to ARF or G,a. For G,a (Fig. 3C), 10 pg/ml nm23-H2p was added on ice to a reaction mixture containing 100 nM [a-"'P]GDP.G,a, 1 mM ATP, and 100 p M GTP, in the NDK reaction mixture. This mixture was then either spotted directly onto PEI-cellulose plates or added to ethanol (final ethanol concentration 67%) or made 2, 4, or 6 M formate before spotting onto PEI plates and resolving products. Again, only the formate was an effective inhibitor of post-reaction artifacts. The protocol described above and shown in Fig. 3C was repeated with one change, the [a-"P]GDP.Gta was heated at 90 "C for 5 min prior to addition, thus releasing the nucleotide. As seen in Fig. 3B, the free nucleotide was almost completely converted to GTP when the mixture was stopped * . e @ @ @ @ One p1 of 10 pg/ ml nm23-H2p was added to each sample, and 10 p1 was analyzed by chromatography on PEI-cellulose plates, as described under "Materials and Methods." Panel C, the conditions were identical to those in panel B, but the substrate was native [a-:"P]GDP-Gta.
immediately, with either spotting onto PEI-cellulose plates or addition to ethanol. Lanes [5][6][7][8][9][10] show that formate at each concentration tested completely arrested the conversion of free GDP to GTP. Although not demonstrated directly, it is likely that denaturation of the GTP-binding proteins is occurring (facilitated by ethanol) prior to the complete loss of NDK activity, and the consequent free GDP is then converted by residual NDK. At higher concentrations of NDK, there is more activity remaining in the "stopped" reaction to produce the GTP (Fig. 3A, solid bars). For all three GTP-binding proteins examined, the contribution of post-reaction GTP formation could be eliminated by stopping the reactions with formic acid.
Testing GDP Bound to Proteins as Possible Substrate for NDK-With the appropriate controls for GDP dissociation and post-reaction product formation, it was possible to test the hypothesis that GDP could be converted to GTP while remaining bound to a regulatory protein. The regulatory protein was loaded with [a-"PIGDP and incubated at 30 "C with the indicated concentrations of NDK, ATP, and GTP. Formation of GTP was determined after separation of products on PEI-cellulose, and the amount of GDP that had dissociated during the reaction was determined by gel filtra-tion. When ARF was examined as a substrate for NDK, no phosphorylation of the bound GDP was detected. Any G T P formed from GDP bound to ARF observed in previous studies was likely the result of intra-incubation and post-incubation dissociation of GDP. In Fig. 4A, the effect of eliminating the post-reaction artifact on GTP formation is shown. Filled symbols represent data obtained when the experiment was performed exactly as previously reported (18) and without controlling for dissociation or post-reaction artifacts. These reactions were run in the absence of BSA or DMPC and stopped by directly spotting on PEI-cellulose plates. When the reactions were stopped in formic acid (open symbols) less G T P was formed, and increasing the NDK concentration led t o a decrease in the net GTP produced. The latter is likely the result of the higher protein concentration resulting in less GDP dissociation, as demonstrated above in Fig. 1B. The effect of dissociation of GDP on GTP production during the incubation is shown in Fig. 4B. All reactions were stopped in formic acid. In the absence of BSA, 30-40% of the GDP was converted to GTP, whereas in the presence of 100 pg/ml BSA, which inhibits dissociation, no G T P was formed. This amount of BSA has no inhibitory effect on NDK activity using either dTDP or free GDP as substrates (not shown). After 60 min, samples were taken from each of the three reactions, shown in Fig. 4B, and analyzed by gel filtration to determine the proportions of bound and free nucleotide. In each case the amount of GTP formed during the reaction exactly correlated with the amount of dissociated nucleotide (not shown). These results indicate that any GTP formed in the reaction was formed in solution and not while bound to the ARF.
Using the same technique, the GDP bound to Ha-ras p21 and (G12V) Ha-ras p21, were tested as substrates for NDK. Again, when the appropriate controls for nucleotide dissociation were used, there was no GTP formed on the protein ir any case. In Fig. 5A, the phosphorylation of free GDP (100 JLLM, squares) was compared to that of GDP bound to either 100 nM (G12V)Ha-ras p21 (triangles) or 100 nM Ha-ras p21 (circles), in the presence of 1 mM ATP, 100 ~L M GTP, NDK, and 100 p~ BSA. Within 10 min, 25 ng/ml NDK converted more than 80% of the free GDP to GTP. However, less than 25% of GDP from either ras protein was phosphorylated by 10 pg/ml NDK after 60 min at 30 "C. This was similar to the a,nount of GDP dissociating from (G12V)Ha-ras p21 in this experiment. As seen in Fig. 5A, the rate of G T P formation from 100 nM GDP that had been bound to (G12V)Ha-ras p21 was the same whether 1 pg/ml (triangles) or 10 pg/ml (inverted triangles) NDK were present with 22% of the GDP converted to GTP after 60 min at 30 "C. Hence, the rate of GTP formation was not limited by the availability of catalytic sites on the enzyme but, rather, was limited by the GDP dissociation rate. When samples from this experiment were analyzed by gel filtration (Fig. 5 B ) , at the start of the reaction (0 min, triangles) all [m3'P]GDP eluted in the void volume (bound to protein). After 60 min (circles), 22% of the radio- Comparison of GDP and GDPORas as substrates for NDK. Panel A , Ha-ras p21 (100 nM, circles) or (G12V)Ha-ras p21 (100 nM, triangles, inverted triangles) was preloaded with [a-"'P]GDP and incubated, as described in the legend to Fig. 4 with the addition of 100 pg/ml BSA and 1 pg/ml (triangles) or 10 pg/ml (circles, inverted triangles) bovine liver NDK. Alternatively, free [a-"*P]GDP (squares) was incubated in the same buffer with 25 ng/ml liver NDK. Samples were taken at the indicated times and the NDK reaction was terminated in formate prior to analysis of products by thin layer chromatography, as described under "Materials and Methods." The percentage of t h e [ Q -~~P ] G D P converted to [a-3'P]GTP is shown. Panel B , samples (80 pl) from the above incubation containing (G12V)Ha-ras p21 and 10 pg/ml NDK were taken at 0 min (triangles) and 60 min (circles) but prior to formate addition, and bound and free nucleotides were separated by gel filtration at 4 "C, as described under "Materials and Methods." activity was recovered in the salt volume (fractions 10-15), having dissociated from (G12V)Ha-ras p21. In fact, whenever

examined, any [a-"P]GTP formed from protein-bound [CY-
''"PIGDP exactly correlated with the amount of GDP dissociated. Similar protocols were used to test for the ability of NDK to activate G, or Gta, as seen in Fig. 6. Three different recombinant NDK proteins, nm23-lp (lanes 1-4), nm23-Hlp (lanes [5][6][7][8], and nm23-H2p (lanes 9-12), were incubated with [cY-:'~P]GDP. G,CY (panel B ) or with free [a-"PIGDP (panel A ) , produced by heat denaturation of the G1a, in the standard NDK reaction buffer for 0 min (lanes 1, 2,5, 6, 9, and 10) or 5 min (lanes 3,4, 7,8,11, and 12) at 30 "C before stopping in formate. Reaction times were kept short because dissociation of nucleotide from G, occurs more readily than from ARF or ras proteins. As seen in panel A, the free nucleotide was completely converted to product within 5 min at 30 "C. Only trace amounts (less than 1% of substrate) of GTP were produced by the same NDK preparations when the substrate used was [a-"P]GDP. Gta (panel B ) . This amount of GTP could not be detected in the salt volume of a Sephadex G-25 column, but we believe the small amount of product formed in this experiment was produced in solution and not while the GDP was bound to the G,a. When longer incubation times were employed and significant amounts of GTP were formed, we observed a quantitative correlation between the amount of GTP formed and the appearance of free nucleotide.
Attempts to Demonstrate Other Interactions between ARF und NDK-In an attempt to recover ARF. GTP from an NDK reaction, ARF. [a-"P]GDP, NDK, ATP, and GTP were incubated at 30 "C for 60 min before fractionation of bound and free nucleotides was performed by gel filtration chromatography. All of the radioactivity eluting in the void volume, bound to protein, was found to be in the form of GDP. A similar protocol was performed several times and the chro-  (nm23-lp, lanes 1-4; nm23-Hlp, lanes 5-8; or nm23-H2p, lanes 9-12) was added to the remaining 16-pl reaction solution, and the samples were incubated for 5 min at 30 "C prior to termination by addition of 5-10 pl of 6 M formate. Products of the reaction were then analyzed as described under "Materials and Methods. Panel A , free GDP was produced by heat denaturation (90 "C, 5 min) of the same preparation of [a-:"P]GDP-Gta used in B. Panel B, conditions were identical to those in panel A , except that the substrate was native [a-'"PIGDP-Gta. matography developed in a variety of buffers, including addition of ATP and GTP to prevent back reactions, but in every case only [cY-"~P]GDP was found in the protein fractions. All of the [cY-"P]GTP eluted in the salt fractions.
Further evidence against the hypothesis was obtained from inhibition studies. ARF, at concentrations up to 10 pM, had no effect on the rate or extent of NDK-catalyzed formation of dTTP or GTP with either dTDP or GDP as substrates (not shown). If both ARF.GDP and GDP are each substrates for NDK then the presence of the ARF. GDP should inhibit the rate of conversion of GDP to GTP in proportion to their K, values. Ha-ras p21 (5 PM) or (G12V) Ha-ras p21 (5 pM) were also found to be inactive as inhibitors of the NDKcatalyzed conversion of free GDP to GTP. Hence, on the basis of competition studies also, GDP bound to these GTPbinding proteins does not appear to be a substrate for NDK.
Studies examining physical interaction between ARF and NDK also suggested that ARF is not directly activated by NDK. Two approaches were used in an effort to detect ARF. NDK interaction. First, 3 pg of [a-"PIGDP-ARF was incubated in 25 mM HEPES, pH 7.4, 2.5 mM MgC12, and 1 mM dithiothreitol in the absence or presence of 3 pg of NDK and in the absence or presence of 1 mM ATP for 30 min at 30 "C in a total volume of 100 pl before applying to 5-ml sucrose density gradients (5-20%) containing 85 pg of NDK and spun at 48,000 rpm for 16 h at 4 "C. Approximately 20-24 fractions were collected, and the migration of protein-bound nucleotide was then monitored by scintillation spectroscopy. NDK had no effect on the migration of ARF. In a second approach, [CY-32P]GDP.ARF (0.1 PM) was again incubated in the absence or presence of NDK (1 p~) in a final volume of 100 p1 in 25 mM HEPES, pH 7.4, 2.5 mM MgCL, 1 mM ATP, and 1 mM dithiothreitol. This reaction was then applied to an Ultrogel AcA 44 column (5 x 80 mm), previously equilibrated with 25 mM HEPES, pH 7.4, 2.5 mM MgC12, 1 mM dithiothreitol in the absence or presence of 1 p~ NDK. As NDK migrates as the hexamer with apparent molecular mass of over 100,000 Da, an ARF. NDK complex was predicted to appear at an apparent mass of >120,000 Da. However, NDK had no effect on the migration of [a-"P]GDP.ARF. Thus, in addition to the failure to find any evidence for conversion of bound GDP to GTP, we were also unable to find any evidence for physical interaction between ARF and NDK either by competition studies or by looking for a change in the migration of ARF in the presence of NDK and resolution by either gel filtration or sucrose density gradients.
Effects of Different Preparations of NDK-Most of the experiments reported above were performed with a commercially available preparation of NDK purified from bovine liver. We have also tested the protein products of the murine nm23-1 and human nm23-HI and nm23-H2 genes, purified from bacterial strains which over-produced each protein, for NDK activity and in protocols described above (e.g. see Fig.  6). In the standard NDK assay (4), using dTDP as substrate the specific activities of these preparations of NDK were: bovine liver (Sigma N-2635), 1.60 units/pg;4.48 units/pg;3.01 units/pg,4.63 units/ pg. We have found no significant differences between any of these preparations with regard to their activities relating to any GTP-binding protein. Thus, we believe that the results reported above may be extended for all NDK.

DISCUSSION
We have described a rigorous test of the hypothesis that regulatory GTP-binding proteins can be directly activated by NDK and have found no evidence to support it. GDP bound to representatives of the trimeric G protein (both monomeric and heterotrimeric G,) or RAS superfamilies (including ras p21 and ARF) was found to be unaffected by NDK, using conditions in which the NDK was fully functional and utilized free guanine nucleotide as substrate. During the course of these studies, we have documented a number of artifacts which we believe have contributed to the erroneous conclusions that NDK acts on GTP-binding proteins. We cannot exclude the possibility that another GTP-binding protein can be directly activated by an NDK, although we currently consider such a possibility unlikely. Re-examination of our own previous work (18) as well as that of others (e.g. 13,15,19,36) has left us unable to cite a single faultless piece of evidence which supports the conclusion that a GTP-binding protein can be directly activated by any NDK through a mechanism which may have physiological significance.
In addition, results from functional and structural studies of GTP-binding proteins provide strong arguments against a role for NDK as a regulator of these proteins. A paradigm describing the regulation of heterotrimeric G protein activation by hormone receptors has emerged from a number of well-controlled kinetic studies of reconstituted G proteincoupled systems (32, 37-39). Results from these and other studies provide strong evidence for the release of bound GDP as the rate-limiting step in activation of G proteins and the step which is most dramatically effected by receptors. Much less is known about the mechanism of activation of members of the RAS superfamily, but it is likely that the rate of GDP dissociation also is limiting. These conclusions make the hormone receptors (in the case of G proteins) or "exchange factors" (in the case of the members of the RAS superfamily) the catalysts for the signal generation inherent in both classes of regulatory proteins. In contrast, the role proposed for NDK would short circuit this regulation by generating the active species without release of bound GDP. The omnipresence and abundance of NDK activities in cells and tissues make it difficult to imagine that both proposed mechanisms of regulation of GTP-binding proteins could be operative at the same time. Further, the structure of ras p21 has been determined by x-ray crystallography (40-43) and shows the guanine nucleotide buried in the protein. It is difficult to imagine how such a buried nucleotide could access the same binding site on NDK as a free nucleotide. Nevertheless, reports which appear to support the hypothesis of NDK-activated regulatory GTP-binding proteins persist. By focusing on the NDK reaction itself and examining the bound GDP as potential substrates for NDK, rather than downstream consequences, we have provided a direct and definitive test of this hypothesis.
Inaccurate assumptions concerning the behavior of the GTP-binding protein have led to erroneous conclusions in previous publications. As shown above, failure to take total protein concentration into account as a variable led to a gross underestimation of the dissociation rate of GDP from ARF or G,. In combination with the post-reaction formation of product, it is clear that all of the data indicating an apparent activation of ARF in our previous study is the result of NDK acting on free nucleotide after release from ARF. The reported (18) apparent K , of 0.16 p~ for NDK activation of ARF was found in reality to be the concentration of ARF which prevented loss of prebound GDP by 50% (0.16 p~ ARF = 4 pg/ ml, see Fig. 1B). Dissociation was not detected after gel filtration in the previous report as it was only performed on samples containing high concentrations of NDK which stabilized the ARF.GDP.
When nucleotide dissociation was carefully monitored, the nucleotides bound to ARF, Ha-ras, G,, or Gta were not found t o be substrates for NDK. It is difficult to assess the extent t o which the other artifact reported above, post-reaction prod-uct formation, have affected results in other studies. Certainly in those instances where ethanol or spotting in thin layer plates was the means of terminating the reaction, the results must be viewed with skepticism.
In many previous studies examining formation of GTP from GDP that had been bound to G,, ras, and E F q (19), nucleotide exchange was not measured and therefore cannot be excluded as the likely cause of apparent phosphorylation of bound GDP. Hence, these studies are incomplete as they cannot exclude the likelihood of NDK acting solely on GDP after its dissociation from each binding protein.
In other studies, dissociation of GDP from GTP-binding proteins has been improperly measured. For instance, exchange was defined (15) as loss of [3H]GDP from G, determined as loss of radiolabel trapped to nitrocellulose filters. However, no free unlabeled nucleotide was included to prevent the rebinding of [3H]GTP after NDK acted on the free nucleotide. Another apparent error was introduced in at least one study (20) which failed to take into account the contribution made by the bound GDP in the determination of the specific activity of guanine nucleotides. It has been reported that many GTPbinding proteins are purified as a 1:l molar complex with GDP (34, 44, 45). Therefore, to obtain an accurate determination of guanine nucleotide-binding sites, the nucleotide which starts bound to the protein must be included in the calculation of specific activity, unless the bound nucleotide can be shown to remain bound throughout the incubation (e.g. the non-exchangeable site of tubulin). Others have attempted to avoid these difficulties by indirectly assessing the NDK-catalyzed phosphorylation of GDP on heterotrimeric G proteins (15). In these experiments, the transfer of 32P from [y3'P]ATP to GDP and subsequent release of 32Pi upon hydrolysis of GTP has been used to measure the ability of NDK to use G proteins as substrates. However, such studies cannot unambiguously be interpreted to demonstrate NDK phosphorylation in situ unless the GTPase rate measured exceeds the GDP dissociation rate. This criterion was not met for the affect of NDK on G, (46). As we have shown with ARF, the rate of formation of GTP was not affected by increasing NDK; rather, the reaction was limited by the GDP dissociation rate.
In a number of cases the ability of NDK to affect the activity of a GTP-binding protein was assessed. However, these studies also suffered from a t least one conceptual flaw in experimental design which precludes the conclusion that NDK was acting as a direct activator of a GTP-binding protein. We note that in our previous report (18, Fig. 2) of the functional consequences of NDK addition to the ARF assay, the reaction required the addition of GTP, DMPC, and cholate, a condition which facilitates GDP-GTP exchange. ARF activity was unaffected by added NDK unless conditions were used which promoted nucleotide exchange. We cannot exclude, but rather favor, the possibility that in these cases the NDK is simply acting as a nucleotide triphosphate regenerating system, removing the inactivating GDP to a point where the GTP can compete more effectively and become a more potent activator. This explanation is particularly attractive in the case of ARF as it has a much greater affinity for GDP than GTP (25), and even under the best conditions only a small fraction of the protein will bind GTP without removal of the released GDP. Such a mechanism is unlikely to be of physiological significance as cells have other nucleoside triphosphate regenerating systems. Thus, we conclude that there are currently no data to support the hypothesis that NDK plays any role in the activation of ARF proteins.
Results from other attempts to document the effects of NDK on a G protein-mediated process in membranes are ambiguous. A source of confusion in these studies is in the use of GTPyS, or ATPyS which can produce GTPyS via NDK (47) in crude in vitro systems. The ability of ATPyS to activate a GTP-binding protein-mediated activity, e.g. activation of adenylate cyclase (22), NADPH oxidase (24), or Gkmediated potassium conductance (23), was interpreted as evidence of NDK catalyzed thio-phosphate transfer to bound GDP or to GDP in a pool that can be "channeled" to the binding site (26, 48). The finding that removal of free GDP by an NTP regenerating system eliminated the effect of ATPyS was thought to further support the latter conclusion (16). However, these arguments fail to consider that while added GDP or GTP reaches a steady-state with any binding sites present in the reaction, GTPyS does not. In the case of the trimeric G proteins, GTPyS binding under most conditions is essentially irrever~ible.~ Furthermore, GTPyS, unlike GTP, is not hydrolyzed by GTP-binding proteins. Consequently, even in the presence of excess GTP there is still a slow but irreversible increase in the activation of a G protein when low levels of GTPyS are formed. Hence, these effects of NDK i n vitro seem best explained by the formation of GTPyS available for exchange. The finding that NDK could be used to produce GTPyS from ATPyS and GDP i n vitro has arguable physiological significance and does not address the argument that NDK may activate GTP-binding proteins in cells (22,24,26,48). Likewise, the suggestion (26, 48) that cells contain different pools of compartmentalized nucleotides, at least one of which is affected by NDK, has not been critically tested.
The importance of the essentially negative data reported above may lie more in allowing researchers to pursue more productive paths in the future. It is hoped that the focus of research into mechanisms of G protein activation can remain on the recept0r.G protein interactions and the contribution played by the bilayer or other, modulating proteins. Finally, the involvement of nm23 in tumor metastasis and awd in fly development still lacks a molecular explanation. These results make it unlikely that expression of an NDK gene is affecting tumor progression or metastasis or development through a GTP-binding protein-mediated pathway. The presence of a leucine zipper motif (8) and the nuclear localization of the nm23 gene product (6) have prompted workers to suggest that NDKs may have distinct functions in the cytosol and in the nucleus. It is hoped that researchers may now focus more of their attention on more productive lines of research that may lead to a better understanding of human tumor metastasis, development, and the physiological functions of NDK.