Anions Modulate the Potency of Geranylgeranyl-Protein Transferase I Inhibitors*

We have identified and characterized potent and specific inhibitors of geranylgeranyl-protein transferase type I (GGPTase I), as well as dual inhibitors of GGPTase I and farnesyl-protein transferase. Many of these inhibitors require the presence of phosphate anions for maximum activity against GGPTase Iin vitro. Inhibitors with a strong anion dependence were competitive with geranylgeranyl pyrophosphate (GGPP), rather than with the peptide substrate, which had served as the original template for inhibitor design. One of the most effective anions was ATP, which at low millimolar concentrations increased the potency of GGPTase I inhibitors up to several hundred-fold. In the case of clinical candidate l-778,123, this increase in potency was shown to result from two major interactions: competitive binding of inhibitor and GGPP, and competitive binding of ATP and GGPP. At 5 mm, ATP caused an increase in the apparent K d for the GGPP-GGPTase I interaction from 20 pm to 4 nm, resulting in correspondingly tighter inhibitor binding. A subset of very potent GGPP-competitive inhibitors displayed slow tight binding to GGPTase I with apparent on and off rates on the order of 106 m − 1s− 1 and 10− 3s− 1, respectively. Slow binding and the anion requirement suggest that these inhibitors may act as transition state analogs. After accounting for anion requirement, slow binding, and mechanism of competition, the structure-activity relationship determined in vitro correlated well with the inhibition of processing of GGPTase I substrate Rap1a in vivo.

Inhibition of prenylation of Ras oncoproteins has long been considered an attractive approach to anti-cancer therapy of Ras-dependent tumors. Four isoforms of Ras are found in mammalian cells, Kirsten-Ras (Ki-Ras) 4A and 4B, Harvey-Ras (Ha-Ras), and N-Ras. Ki4B-Ras is the most frequently mutated isozyme in human cancer, ranging from 20% up to 90% in pancreatic cancers, whereas Ha-Ras mutations are found in a small percentage of bladder cancers. Most oncogenic mutations in Ras result in the elimination of the GTPase activity, locking Ras in its active, GTP-bound conformation. All Ras proteins undergo multiple post-translational modifications on their carboxyl termini that are required for membrane localization and biological activity. The COOH-terminal CAAX motif (where C is cysteine, A stands for aliphatic, and X for any amino acid) is recognized by farnesyl-protein transferase (FPTase), 1 which transfers a farnesyl group from farnesyl pyrophosphate (FPP) to the cysteine thiol. This prenylation step is followed by proteolytic clipping of the three terminal amino acids by a farnesyl-CAAX specific protease and subsequent methylation of the cysteine residue by a carboxymethyl transferase. Ha-Ras and N-Ras are further modified by palmitoylation which provides additional membrane interactions. A polylysine stretch in Ki-Ras near the COOH terminus serves the same purpose. However, the prenylation step has been shown to be the single most critical modification for function of all Ras proteins (reviewed in Ref. 1).
Initial efforts at inhibiting Ras function by many groups focused exclusively on FPTase, the enzyme normally responsible for prenylation of all four Ras isozymes. Specific inhibitors of FPTase showed dramatic efficacy in Ha-Ras-dependent mouse xenograft tumor models (2) and were also effective in Ki-Ras-driven mouse tumor models (3). However, it was soon realized that FPTase inhibitors did not block prenylation of Kiand N-Ras isozymes because of cross-prenylation by the related enzyme geranylgeranyl-protein transferase type I (GGPTase I) (4 -7), resulting in apparently fully functional and, in the case of mutant Ras, transforming protein. It thus became clear that pure FPTase inhibitors were not acting by blocking Ras signaling but by inhibiting farnesylation of one or more as yet unidentified proteins. Blocking Ki-Ras signaling through inhibition of both prenyltransferases remains a viable option, however (8 -10). To further explore the feasibility of preventing Ki4B-Ras signaling in vivo, we developed and characterized potent inhibitors of GGPTase I, FPTase, and of both enzymes (dual inhibitors).
GGPTase I was originally purified from bovine brain (11); the human enzyme was subsequently cloned and purified (12). GGPTase I is an ␣/␤ heterodimer with the same ␣ subunit as FPTase. The enzyme transfers the prenyl moiety of geranylgeranyl pyrophosphate (GGPP) to the cysteine of carboxylterminal CAAX sequences, with a preference for leucine in the X position. It is also capable of prenylating the COOH-terminal CVIM sequence of Ki-Ras, albeit with a K m 500-fold higher than that of FPTase (13). The enzyme mechanism of mammalian GGPTase I has been characterized with enzyme purified from bovine brain as well as from recombinant sources (14,15). The reaction was reported to follow a random Bi Bi pathway with steady-state kinetics, although GGPP binding is the kinetically preferred first step. The order of addition of substrates is reflected in their relative K m values of low nanomolar for GGPP and micromolar for peptide substrate, although the absolute numbers vary dramatically with the particular CAAX sequence and length of the peptide substrate (7,13,15,16). The reaction mechanism thus appears identical to that of FPTase (17). On the other hand, the yeast enzymes yGGPTase I and yFPTase have been reported to proceed through an ordered Bi Bi mechanism (18,19). Interestingly, and in contrast to FPTase, GGPTase I does not require Mg 2ϩ for GGPP binding or catalysis. However, Zn 2ϩ is required for peptide binding and catalysis (15,20).
All reported inhibitors of GGPTase I are either competitive with CAAX or GGPP substrate, and many were designed based on the structure of the peptide or prenylpyrophosphate (9,(21)(22)(23)(24)(25)(26). Compounds in both categories show activity in cellbased assays. Some inhibitors of FPTase have been observed to be more potent in the presence of certain anions (27). Increases in potency of up to 50-fold were observed, with phosphate and chloride being the most effective anions. This anion effect was limited to FPP-competitive inhibitors and was not seen with GGPTase I. We report here the identification of GGPTase Ispecific and dual inhibitors with a similar requirement for phosphates for maximal potency. As in the case of the FPTase inhibitors (27), the anion effect is limited to prenylpyrophosphate-competitive compounds; however, with our current dualspecificity compounds, the anion effect is only seen when assayed with GGPTase I. Utilizing an equilibrium binding assay, we show that the anion effect is primarily due to competitive binding between phosphates and GGPP, which results in a dramatic increase in the apparent K d for GGPP. Consistent with this mechanism, a significant anion effect is only observed with GGPP-competitive, but not CAAX-competitive, inhibitors. Although the addition of phosphates increased inhibitor potency, it did not change the mechanism of inhibition in equilibrium binding studies. However, the unexpectedly tight binding of GGPP in the absence of phosphates resulted in an apparently CAAX-competitive mode of inhibition in enzyme activity experiments.
To determine the significance of these observations for the inhibition of GGPTase I in vivo, we evaluated the inhibition of prenylation of the GGPTase I substrate Rap1a in PSN-1 cells. IC 50 values measured in vitro in the absence of phosphates did not predict potency of GGPTase I inhibitors in cell-based assays. However, after accounting for anion requirement, slow binding, and mechanism of competition, the structure-activity relationship determined in vitro correlated well with the inhibition of processing in vivo.

EXPERIMENTAL PROCEDURES
Materials-Biotinylated Ki4B-Ras-derived peptide b-GKKKKKKSK-TKCVIM (single-letter amino acid code) was ordered from Research Organics. [ 3 H]GGPP was purchased from PerkinElmer Life Sciences. Scintillation proximity assay (SPA) beads pre-coated with streptavidin and ECF reagents were obtained from Amersham Pharmacia Biotech. Antibodies specific for Rab6, Rap1a, and for the unprenylated form of Rap1a were from Santa Cruz Biotechnology. YL1/2 antibody was from Harlan Bioproducts for Science. The aspartic acid-phenylalanine (Asp-Phe) dipeptide was from Sigma.
GGPTase I-Human GGPTase I was expressed and purified as described previously (28). A biotinylated version of GGPTase I was produced by inserting the recognition site for biotin ligase, GLNDIFEAQK-IEWH, at the NH 2 terminus of the alpha subunit. Co-expression in BL21(DE3) birA pYCAC, an Escherichia coli strain expressing the biotin ligase gene, in the presence of 50 M biotin resulted in NH 2 -terminally biotinylated GGPTase I. The enzyme was purified by YL1/2 antibody affinity chromatography followed by MonoQ ion exchange chromatography. Liquid chromatography-mass spectrometry analysis confirmed that biotinylation had proceeded to completion (Ͼ95%).
GGPTase I Assay-GGPTase I reactions were run at room temperature with biotinylated Ki4B-Ras derived peptide substrate and tritiated GGPP. Reactions in 100-l volumes were started with the addition of peptide unless otherwise noted and contained the following final concentrations of reagents: 80 pM GGPTase I, 100 nM [ 3 H]GGPP, 1.6 M biotinylated peptide, 50 mM Hepes-NaOH, pH 7.5, 7 mM MgCl 2 , 10 M ZnCl 2 , 0.1% polyethylene glycol (M r 15,000 -20,000), 1 mM dithiothreitol, 4% Me 2 SO. MgCl 2 is not required for GGPTase activity but is essential for FPTase. MgCl 2 was always included in the GGPTase reaction to keep reaction conditions for the two enzymes closely matched. Inhibitors and 5 mM ATP or other anions were included as indicated in individual experiments. For slow-binding inhibitors, GGPTase I and inhibitor were preincubated at room temperature for 30 min in the absence of substrates. After 15 min at room temperature, reactions were stopped with the addition of 100 l of 2 mg/ml SPA beads in 0.2 M NaPO 4 , pH4, 50 mM EDTA, 0.5% bovine serum albumin, 0.05% NaN 3 , and counted after an additional 60 min in a TopCount scintillation counter (Packard Instruments).
The mechanism of inhibition by selected GGPTase I inhibitors was determined at a fixed concentration of GGPP (100 nM) or peptide (1.6 M) while the concentration of the second substrate was varied. Instead of a graphical analysis via double-reciprocal plots, the entire raw data set was simultaneously fit with the following equation for linear mixed type inhibition using SigmaPlot software.
I is the inhibitor, S is the varied substrate, K i and K ii are the binding constants for the interaction of I with enzyme (E) and with the E-S complex, respectively. For a competitive inhibitor, we expect that K ii Ͼ Ͼ K i , whereas the reverse would be true for an uncompetitive inhibitor.
To obtain qualitative information on competition between inhibitors and GGPP, IC 50 values were determined at three concentrations of GGPP: 20, 100, and 500 nM. The slope of a plot of log(IC 50 ) versus log[GGPP] was used as a measure of the extent of competition, where slopes of 1, 0, and Ͻ0 indicate competitive, non-competitive, and uncompetitive binding, respectively.
Equilibrium Binding Assay-Equilibrium binding experiments were performed under GGPTase I reaction conditions, except for the absence of peptide substrate. In a typical experiment 10 nM biotinylated GGPTase I was preincubated with 0.67 mg/ml streptavidin SPA beads for 1-2 h at room temperature or overnight at 4C. After addition of ligands the mixtures (typically 60-l volumes) were incubated for 30 min, followed by determination of bound ligand on a Packard TopCount scintillation counter. Multiple readings were taken over time to assure that equilibrium had been reached.
For the determination of the K d of the GGPTase I-GGPP interaction, lower enzyme concentrations in correspondingly larger volumes were used. Final concentrations in 15-ml volumes were 15 pM GGPTase I, 5 mM ATP unless otherwise indicated, and variable concentrations of [ 3 H]GGPP. After equilibrium had been reached, free [ 3 H]GGPP was determined by removing 1-ml aliquots, sedimenting the beads by centrifugation, and mixing 0.75 ml of the supernatant with 15 ml of 80% ReadySafe scintillation mixture/20% isopropanol. Free and bead-bound radioactivity were measured in a Beckman scintillation counter. Parallel controls of [ 3 H]GGPP and [ 3 H]biotin of known specific activity were carried along to correct for loss of GGPP due to nonspecific adsorption and to determine counting efficiencies of free and bound ligand. Recovered GGPP at the end of the experiment (free plus bound) was generally in the 30 -70% range.
Data from equilibrium binding experiments were fit with an equation describing the binding of two ligands, one of which is labeled, to two distinct sites on a protein.
E is the protein, S and X are the labeled and unlabeled ligands, respectively, K s and K x are the binding constants for the E-S and the E-X interactions, respectively, and ␣ is the interaction factor between the two binding sites. A value of ␣ of Ͻ1, 1, Ͼ1, and Ͼ Ͼ1 indicates synergistic, independent, partly interfering, and mutually exclusive binding of the two ligands, respectively. ES and ESX are the labeled complexes detected in the SPA-based equilibrium binding format. In all experiments, except for the GGPP K d determination described above, ligand concentrations were substantially greater than the GGPTase I concentrations and [X] t and [S] t were substituted for free [X] and [S].
In the case of a single binding site for two competing ligands, Equation 2 simplifies to Equation 3.
Rap1a Processing-Inhibition of Rap1a geranylgeranylation was measured in the PSN-1 human pancreatic tumor cell line. Cells were treated for 24 h with inhibitor, and lysates were resolved using 10 -20% SDS-polyacrylamide gel electrophoresis gradient gels (Novex Inc) and then analyzed by Western blotting using a polyclonal antibody specific for Rap1a (Santa Cruz Biotechnology Inc., Rap1/Krev-1 antibody 121). Blots were developed using an alkaline-phosphatase conjugated antirabbit IgG (Cappel Inc), developed with a fluorescent detection reagent (ECF, Amersham Pharmacia Biotech) and scanned on a PhosphorImager (Storm ® , Molecular Dynamics). Unprenylated Rap1a was distinguished from prenylated Rap1a by virtue of their different electrophoretic mobilities. The percentage of unprenylated Rap1a relative to total Rap1a was determined by peak integration using Imagequant ® software (Molecular Dynamics), and EC 50 values were derived from dose-response curves with a four-parameter curve-fitting equation using SigmaPlot ® software.  Table I. The dual-specificity inhibitor and clinical candidate, L-778,123 (compound 1), showed a 50-fold preference for inhibition of FPTase over GGPTase I. The most potent GGPTase I inhibitors, exemplified by compounds 2 and 4, resulted from the addition of hydrophobic groups to either end of the basic structure of compound 1. Similarly, the most specific GGPTase I inhibitors, including compound 5, are characterized by the presence of hydrophobic groups at both ends of the molecule. More detailed structure-activity relationships will be presented elsewhere.

Anion Dependence of Inhibition-We
Many compounds showed little or no inhibition of GGPTase I when assayed in Hepes buffer despite the fact that they appeared to inhibit processing of GGPTase I substrates in cell-based assays. For example, compound 1 showed negligible activity against GGPTase I when assayed in 50 mM Hepes buffer but it inhibited Rap1a processing in PSN-1 cells with an EC 50 of 6.8 M. Unexpectedly, the addition of several anions to the in vitro reaction resulted in a decrease of the IC 50 values for compound 1 from greater than 50 M to less than 100 nM (Fig.  2). No inhibition of the reaction by the anions themselves was seen except as noted below. Different anions lowered the IC 50 value of compound 1 to a similar extent, though the required concentrations varied widely. ATP and dithiophosphate were the most effective anions (with respect to the concentration needed), but many phosphate and sulfate-based anions, ranging from nucleotides to glycerophosphate, could substitute. On the other hand, several commonly used buffer anions, such as chloride and acetate, were completely ineffective at concentrations up to 50 mM. The rank order for the tested anions was as follows: dithiophosphate ϳ ATP ϳ dNTP Ͼ thiosulfate ϳ tripolyphosphate ϳ pyrophosphate ϳ sulfate ϳ GTP Ͼ phosphate ϳ glycerophosphate Ͼ Ͼ acetate, carbonate, chloride, or nitrate. Based on intracellular anion concentrations, ATP would be expected to be the major contributor to an anion-effect in cells. We therefore decided to include 5 mM ATP in subsequent GGPTase I assays. When used in the presence of a slight excess of Mg 2ϩ , ATP had no effect on the reaction rate at concentrations up to 20 mM. Mg 2ϩ was not required, however, for either GGPTase I activity or the anion effect per se. In contrast to ATP and inorganic phosphate, pyrophosphate and dithiophosphate inhibited the reaction rate with IC 50 values of 1 mM, likely due to product inhibition in the case of pyrophosphate. ATP at 5 mM did not affect the apparent K m values for GGPP (4 -8 nM, determined at 1.6 M concentration of peptide substrate) and CAAX peptide (0.4 -0.7 M at 100 nM GGPP).
Since our standard assay makes use of a biotinylated peptide substrate, we confirmed the anion effect with full-length proteins as substrates. ATP caused similar shifts in potency of compound 1 (ϳ1000-fold) and several other compounds when assayed with either full-length Ki-Ras or RhoB as substrates. The absolute IC 50 values varied, however, due to the differences in K m values of the protein substrates (data not shown). The extent of the gain in potency in the presence of ATP and other phosphates varied widely with different inhibitors, but was most dramatic for compounds 1 and 5 and closely related structures (Table I and data not shown).
Kinetic Analysis of Inhibition-Substrate competition experiments showed that our GGPTase I inhibitors spanned the entire range of GGPP-competitive to mixed type to CAAXcompetitive compounds, despite the fact that the design of these compounds had evolved from the CAAX peptide sequence. Interestingly, the anion dependence of these inhibitors correlated with their mechanism of inhibition. Compounds that showed a pronounced shift in IC 50 upon addition of ATP, such as compounds 1 and 5, were found to be GGPP-competitive (determined in the presence of ATP). On the other hand, compounds with less than 3-fold shifts, such as compounds 6 and 7, consistently were non-or uncompetitive with respect to GGPP (Table I). This relationship is shown for a large number of compounds in qualitative fashion in Fig. 3. For this analysis the ratio of IC 50 values determined in the absence and presence of 5 mM ATP is plotted against a qualitative measure of GGPP competition in the presence of ATP (see "Experimental Procedures"). In general, the more competitive with GGPP, the more dependent on anions for potency the compounds were found to be. A marked exception was the hydroxyphosphonate 8, a GGPP analog, for which the potency was reduced by phosphates (Table I).
A detailed kinetic analysis was performed for a subset of the above compounds and is shown in Fig. 4 for compound 1. For this analysis the entire data set was fit simultaneously with a general equation for linear mixed type inhibition (see "Experimental Procedures"). In the presence of phosphates (5 mM ATP), compound 1 is competitive with GGPP and noncompetitive with CAAX peptide (Fig. 4, A and B). On the other side of the spectrum, closely related compound 7 showed CAAX-competitive behavior while being un-or noncompetitive versus GGPP. Several other compounds were partly competitive with GGPP, displaying mixed type inhibition versus GGPP with finite ratios of K i to K ii (data not shown).
Unexpectedly, in the absence of phosphates, the mechanism of inhibition of compound 1 changed to competitive with respect to CAAX peptide. Inhibition with respect to GGPP appeared to be mixed type, albeit the data at low GGPP concentrations were not reliable enough to rule out uncompetitive inhibition (Fig. 4, C and D). Similarly, all fully or partially GGPP-competitive compounds became non-competitive inhibitors with respect to GGPP in the absence of ATP. The omission of phosphates thus induces an apparent change in the mechanism of inhibition, from GGPP-to CAAX-competitive. Based on the results from binding experiments, it appears, however, that kinetic uncoupling, rather than a true switch in mechanism, are responsible for this phenomenon (see below).
Remarkably, when assayed against FPTase, none of the above compounds showed a dependence on anions (less than 3-fold shifts in IC 50 values). Furthermore, with the exception of a farnesyl homolog of compound 8, all inhibitors displayed a CAAX-competitive mechanism of inhibition independent of the presence of ATP or inorganic phosphate (data not shown).
In summary, kinetic analyses of inhibition of GGPTase I, but not FPTase, showed a phosphate-dependent partial or complete competition between our inhibitors and GGPP, suggesting an overlap of the respective binding sites.
Equilibrium Binding-In an attempt to better understand the underlying interactions of inhibitors, substrates, and anions with GGPTase I, we designed a binding assay with an immobilized form of GGPTase I. GGPTase I with a biotinylation tag (b-GGPTase I) was expressed and biotinylated in E. coli. Complete biotinylation was confirmed by liquid chromatography-mass spectrometry. b-GGPTase I was then coupled to streptavidin-coated scintillation proximity beads, and binding of tritium-labeled ligands was measured in a standard SPA format. The availability of tritiated forms of GGPP and compounds 1, 3, and 7 allowed pairwise competition experiments among inhibitors, substrates, and anions.
Since initial experiments indicated that the K d for GGPP in the absence of phosphates may be much lower than suggested by the nanomolar K m value, we first performed equilibrium binding experiments with [ 3 H]GGPP at very low enzyme concentration. The concentrations of free and bound [ 3 H]GGPP were measured separately since a substantial fraction of GGPP was bound to enzyme under our assay conditions. Due to the tendency of the hydrophobic prenyl group to adsorb to vessel walls and pipette tips, the measured concentrations of free GGPP were always significantly lower than calculated. In two independent experiments, the equilibrium binding constant for GGPP was found to be remarkably low, 15-30 pM in the absence of phosphates, but nearly 200-fold higher, 2-6 nM, in the presence of 5 mM ATP (Fig. 5A). To determine the mechanism by which ATP raises the apparent K d of GGPP, we measured bound [ 3 H]GGPP as a function of ATP concentration (Fig. 5B).
The nearly complete displacement of GGPP at high ATP indicates that the two ligands bind competitively to GGPTase I. The data could be fitted well with the equation for a two-ligand/ single-site receptor binding model, consistent with a single specific binding site for ATP, rather than multiple nonspecific ionic interactions, and suggestive of competitive binding to the pyrophosphate binding site of GGPP. The curve fit yielded a surprisingly low K d(app) for the ATP-GGPTase I interaction of 50 -100 M. Despite being competitive with one of the substrates, ATP did not inhibit the reaction. This was not unexpected since the association rate of GGPP is limited only by diffusion and the catalytic rate of the GGPTase I reaction is slow (k cat ϭ 0.07 s Ϫ1 at saturating GGPP and 1.6 M peptide, data not shown).
For competition experiments between inhibitors and GGPP, the binding of tritium-labeled inhibitor to GGPTase I was determined in the presence of varying concentrations of cold GGPP. Fig. 5C shows the effect of GGPP on compound 1 binding to GGPTase I in the presence of ATP. Simultaneous curve fitting of all data points with a "two-ligand/two binding sites" model yielded a K d for compound 1 of 30 nM. In the absence of ATP, the K d value rose to 190 nM (data not shown). The interaction factor, ␣, was poorly defined with a value of Ͼ50 under both conditions, but indicative of mutually exclusive binding (␣ values of Ͻ1, 1, Ͼ1, and Ͼ Ͼ1 indicate cooperative, independent, interfering, and competitive binding, respectively). In binding experiments compound 1 is thus competitive with GGPP, independent of the presence of phosphates (see "Discussion"). Analogous experiments with compound 3 yielded K d values of 4 and 24 nM, and ␣ values of 30 and 5 in the presence and absence of ATP, respectively. For this compound simultaneous binding with GGPP is thus possible, although energetically unfavorable. In the case of compound 7, no competition by GGPP was seen under any condition, consistent with its CAAX-competitive behavior in GGPTase I enzyme assays (data not shown).
The equilibrium binding analysis was subsequently expanded to a large number of GGPTase I inhibitors by measuring displacement of [ 3 H]GGPP, with results consistent with those described above. All inhibitors that showed a GGPPcompetitive mechanism in kinetics experiments were found to compete with [ 3 H]GGPP in the binding assay, with EC 50 values that correlated well with their IC 50 values in the enzyme assay, while CAAX-competitive inhibitors were unable to displace [ 3 H]GGPP. The presence or absence of phosphate did not affect the qualitative outcome of these binding studies, even though the EC 50 values for [ 3 H]GGPP displacement in the absence of ATP were dramatically higher, as expected from the picomolar K d of GGPP under those conditions. The equilibrium binding data suggest a model in which phosphates increase the potency of GGPP-competitive inhibitors by displacing GGPP from the pyrophosphate-binding site, but without impeding binding of the inhibitor itself. In fact, ATP can facilitate inhibitor binding to the enzyme even in the absence of GGPP, at least in the case of compound 1. As shown in Fig. 5D, the apparent K d of compound 1 decreases with increasing ATP concentration. Simultaneous curve fitting of the entire data set with a "two-ligand/two binding sites" model yielded K d values for compound 1 of 220 nM and for ATP of 35 M, consistent with the equilibrium binding constants determined above. The interaction factor ␣ of 0.13 indicates cooperative binding between compound 1 and phosphates. The increased potency of compound 1 in the presence of ATP is thus the combined result of ATP raising the apparent K d of GGPP 200-fold and lowering the apparent K d for compound 1 ϳ7-fold.
The hydroxyphosphonate 8 did not follow the pattern of GGPP-competitive inhibitors, where potency increases with anions (Table I). This initially puzzling result is consistent with the above model, in that the phosphonate head group presumably binds in the pyrophosphate site now occupied by ATP. In contrast to the situation with compound 1, this would result in competitive rather than synergistic binding with phosphates.
Slow Binding Inhibition-In some of the binding experiments described above, the final equilibria established rather slowly. Although the association of GGPP with GGPTase I proceeded with nearly diffusion-limited on rates of ϳ10 9 M Ϫ1 s Ϫ1 (data not shown), several GGPTase I inhibitors displayed slower kinetics. Since these compounds were not available in isotopically labeled form, and association could therefore not be followed directly in our assays, the formation and dissociation of enzyme-inhibitor complexes was monitored in subsequent short activity assays (5-min reactions). Inhibition of GGPTase I activity was determined as a function of preincubation time of enzyme and inhibitor in the absence of GGPP. Fig. 6 shows a typical experiment with compound 2, in which association as well as dissociation of preformed enzyme-inhibitor complexes were measured. Dissociation was initiated by the addition of excess GGPP, followed by a GGPTase I reaction at the indicated times. Curve fitting yielded on and off rates of ϳ2 ϫ 10 6 M Ϫ1 s Ϫ1 and 10 Ϫ3 s Ϫ1 , respectively. Similarly slow association rates were observed with several of the more potent GGPPcompetitive inhibitors, but not with potent CAAX-competitive inhibitors such as compound 6. Preincubation for at least 30 min in the absence of GGPP was required to obtain maximal inhibition of GGPTase I. As expected for GGPP-competitive inhibitors, preincubation in the presence of GGPP resulted in an additional, concentration-dependent lowering of the apparent association rates (data not shown).
Correlation with Cell-based Activity-All of the in vitro phenomena described above were found to have in vivo correlates, i.e. anion effect, GGPP versus CAAX competition, and slow binding had to be taken into account to optimally predict potency of GGPTase I inhibitors in cell-based assays. Fig. 7 demonstrates the improvement in the correlation between in vitro IC 50 and cell-based activity. GGPTase I activity in PSN-1 cells was measured via the inhibition of geranylgeranylation of Rap1a, a substrate for GGPTase I but not FPTase. Inhibition of Rap1a prenylation by nearly 70 compounds showed a complete lack of correlation with the in vitro potency as determined under the initial assay conditions in the absence of phosphates (Fig. 7, left panel). GGPTase I IC 50 values determined in the presence of 5 mM ATP and after preincubation of enzyme and inhibitor result in a dramatically improved correlation (Fig. 7,  middle panel). A separation of inhibitors into GGPP versus CAAX-competitive categories improved the correlation further. Linear regression of the data for the two groups of compounds indicated a 30-fold difference in cell-based versus biochemical potency (Fig. 7, middle panel). After correcting the IC 50 values of CAAX-competitive compounds by a factor of 30, an acceptable correlation with a slope of close to 1 was obtained. The need for the latter correction likely reflects differences in the effective substrate concentrations in cells and in vitro. Differences in cell uptake of compounds are not taken into account in the  data shown in Fig. 7 and may be responsible for the remaining scatter in the in vitro/in vivo correlation.

DISCUSSION
Prenylation and membrane localization of Ki-Ras are essential for its oncogenic potential. In vivo Ki-Ras is normally predominantly farnesylated on its COOH-terminal cysteine, but in the presence of FPTase inhibitors the 21-kDa protein is prenylated by GGPTase I. Geranylgeranylated Ki4B-Ras is biologically active, implying that inhibition of both enzymes would be required to down-regulate Ki-Ras signaling. Interestingly, however, inhibition of either FPTase or GGPTase I alone can result in tumor growth inhibition in nude mouse tumor models, although it is not clear which prenylated protein(s) is responsible for the observed efficacy. In the case of FPTase, inhibition of RhoB farnesylation, followed by geranylgeranylation, has been suggested as the critical step (29, 30). Inhibition of farnesylation subsequently leads to induction of p21 and a cell cycle block at G 1 (31) or G 2 -M (Ref. 32 and unpublished data). The G 2 -M block may be the result of a block of prenylation of the mitotic kinesin CENP-E (33). G 1 cell cycle arrests, including hypophosphorylation of retinoblastoma and effects on cyclin-dependent kinase inhibitors, were observed after treatment of cells with specific GGPTase I inhibitors (32, 34 -36). Nevertheless, inhibition of either prenyltransferase alone does not block signaling via Ki-Ras. Since we wished to examine the effects of inhibition of Ki-Ras in vivo and to conduct tumor efficacy studies with combinations of FPTase and GGPTase I inhibitors, we synthesized and evaluated specific GGPTase I inhibitors and dual inhibitors.
While analyzing FPTase inhibitors for their ability to inhibit prenylation of Ki-Ras in cell-based assays, we found that some compounds with little activity against GGPTase I were capable of completely blocking farnesylation and geranylgeranylation in cells. This apparent paradox led to the discovery that we were underestimating the potency of GGPTase I inhibitors in vitro because of a specific anion requirement for maximum inhibition. Many different phosphate-based anions could satisfy this anion requirement, including inorganic phosphates, nucleotides, and phospholipids. The fact that some anions were effective at significantly lower concentrations than others immediately suggested that this effect was not due to nonspecific ionic interactions but rather to a direct effect on inhibitor or substrate binding.
Despite being closely related and derived from the same basic structural motif, our GGPTase I inhibitors separated into CAAX-competitive and GGPP-competitive classes. Only the latter compounds showed a significant gain in potency in the presence of anions, suggesting that the anion effect was directly related to the GGPP-competitive mode of inhibition. A similar phenomena has been reported for certain inhibitors of FPTase, which also became much more potent in the presence anions (27). Increases in potency of up to 50-fold were observed in the presence of phosphate. As is the case for the GGPTase I inhibitors described here, this anion effect was limited to pre- nylpyrophosphate-competitive inhibitors. In contrast, however, the anion effect was only seen with FPTase and not with GGPTase I. Furthermore, the rank order of anions differed substantially from that described in the present report for GGPTase I inhibitors. For example, chloride, one of the best anions for FPTase, was completely ineffective for GGPTase I inhibitors.
What is the basis of the anion effect? Scholten and co-workers proposed a model in which anions would compete with the pyrophosphate moiety of FPP for binding to the enzyme. To test this hypothesis for the case of GGPTase I, we developed an equilibrium binding assay to directly measure binding of GGPP and other ligands. Using carefully controlled conditions with picomolar enzyme concentrations, we found that GGPP binds to GGPTase I with a much lower K d than expected, 15-30 pM in the absence of phosphates. Previous reports had found K d values in the low nanomolar range (3-16 nM), a result that may be explained at least in part by the presence of phosphate in some of the buffer systems used (15,20,37). We show that phosphates behave as competitive ligands with respect to GGPP. Although our data do not prove that the two ligands bind to physically overlapping sites, anion binding to the pyrophosphate site on GGPTase I would be the simplest explanation. ATP, our anion of choice, raises the apparent K d of the GGPPenzyme interaction ϳ200-fold into the low nanomolar range at physiological concentration.
Equilibrium binding experiments were also used to analyze the extent to which GGPTase I inhibitors compete with GGPP binding. In the presence of phosphates (i.e. ATP), the results were entirely consistent with the data from enzyme assays. GGPP-competitive inhibitors such as compound 1 were able to displace GGPP completely without significant formation of the ternary enzyme-GGPP-inhibitor (E-GGPP-I) complex (interaction factor ␣ Ͼ 50). In contrast, CAAX-competitive inhibitors, including compounds 6 and 7, did not affect GGPP binding at all. Several compounds showed partially competitive behavior where the ternary E-GGPP-I complex was observable but energetically unfavorable (interaction factors of greater than 1). These data are consistent with a model in which the inhibitor primarily occupies the CAAX binding site but, depending on the particular compound, partly overlaps with the GGPP binding site. Inhibitor binding in the CAAX site is expected since the design of all compounds described here (with the exception of the phosphonate 8) evolved from the CAAX peptide structure. In the case of FPTase, compound 1 was shown by x-ray crystallography to bind in the CAAX pocket alongside FPP. 2 Whether or not a given inhibitor overlaps with the GGPP binding site seems to depend on minor structural alterations, as demonstrated by the orthogonal behavior of closely related compounds 1 and 7. In contrast to GGPP, ATP does not compete with inhibitor binding; instead, anion and inhibitor binding are cooperative, at least in case of compounds 1 and 3 (interaction factor Ͻ1). The overlap between GGPP and these inhibitors thus appears to be limited to the prenyl binding pocket, whereas the triphosphate moiety of ATP occupies the pyrophosphate binding site. Consistent with this model, a reverse anion effect is observed with the GGPP analog 8. In this case anions directly compete with inhibitor binding in the pyrophosphate pocket.
In the absence of phosphate, equilibrium binding and enzyme assays gave apparently contradictory results, in that compounds that competed with GGPP in binding studies behaved as CAAX-competitive inhibitors in enzyme assays. This discrepancy has probably two underlying reasons: (i) kinetic uncoupling in the enzymatic assays at high concentrations of GGPP (in contrast to equilibrium binding experiments where GGPP concentrations could be lowered into the low picomolar range to show competition by less potent inhibitors, steadystate reactions needed to be run in the mid-nanomolar range), and (ii) the ability of most, if not all, GGPP-competitive compounds to form the energetically unfavorable ternary E-GGPP-I complex at high enough concentrations (i.e. interaction factor ␣ Ͼ Ͼ 1, but not infinite). The resulting inhibition would be weak but CAAX-competitive, as exemplified by compound 1. It follows from this explanation that the less favorable the formation of the E-GGPP-I complex, the greater the expected anion effect. This prediction is in fact borne out in that the extent of GGPP-competition correlates with the magnitude of the anion effect (Fig. 3). In the presence of phosphates, which compete with GGPP and thus facilitate the formation of E-I complexes even at higher GGPP concentrations, equilibrium binding and enzyme assays gave consistent results.
Several potent GGPP-competitive inhibitors displayed slow binding to the enzyme even in the absence of GGPP. On-rates 2 orders of magnitude below diffusion limited on rates were observed, and enzyme-inhibitor complexes showed half-lives of up to 1 h. Synergistic inhibition of enzymes in the presence of anions has been reported in several cases and has been dis-2 S. B. Long and L. S. Beese, manuscript in preparation. cussed in the context of FPTase inhibitors (27). The observation of slow tight binding in combination with anion dependence furthermore suggests the formation of inhibition complexes that may be mimicking the transition state of GGPTase I. For FPTase, as well as for the related reactions catalyzed by squalene synthase, isopentenyl pyrophosphate isomerase, and prenyltransferase, the reaction mechanism has been shown or proposed to proceed through an allylic carbonium-pyrophosphate intermediate (38 -43). Nitrogen-containing phosphates and phosphonates have been reported to be transition state analogs of isopentenyl-pyrophosphate isomerase (44,45). Most of our slow-binding GGPTase I inhibitors have at least one electrophilic nitrogen that may play the role of the carbonium cation, whereas the anion (ATP) would take the place of the pyrophosphate leaving group.
We have identified closely related compounds with either GGPP or CAAX-competitive mode of inhibition in vitro. This difference in mechanism was reflected in cell-based assays, where GGPP and CAAX-competitive compounds fall into different categories, resulting in the need for a correction factor when predicting cell potency from in vitro IC 50 values. This is not entirely surprising since the effective concentrations ([S]/ K m ) of the two GGPTase I substrates differed ϳ10-fold for experimental reasons; GGPP was used at 100 nM (20-fold over K m ), and the CAAX peptide was used at 1.6 M (2-fold over K m ). After correction of the IC 50 values of CAAX-competitive compounds by a factor of 30, enzyme assays predicted cell potency fairly reliably, implying that the effective concentrations of GGPP and Rap1a in cells are in fact very similar.
Interestingly, none of the dual inhibitors in this study showed a significant anion-dependence when assayed with FPTase. Although the two prenyltransferases are analogous in terms of function and general reaction mechanism, their differences obviously extend beyond the different substrate preferences. Crystal structures of the enzyme-inhibitor-anion complexes will eventually shed light on some of these idiosyncrasies and should help further refine inhibitor specificity and potency. We show here that relative subtle structural changes affect the mechanism of inhibition of GGPTase I, which may positively impact drug development. The rank order of inhibition of prenylation of various GGPTase I protein substrates depends on their relative K m values. Since this rank order is affected differently by CAAX and GGPP-competitive inhibitors, it should allow for the design of inhibitors with different substrate profiles within the same structural class.
Compound 1 (L-778,123) is a dual-specificity inhibitor with a 50-fold preference for FPTase over GGPTase I in enzyme assays. In cell culture this compound shows a 35-fold preference for FPTase, as measured by the inhibition of prenylation of Rap1a and FPTase substrate HDJ2. Inhibition of geranylgeranylation of Rap1a in PSN-1 cells is detectable at low micromolar concentrations. Preliminary pharmacodynamic analysis of Phase I clinical data shows that L-778,123 partially inhibits Rap1a prenylation in patients at the highest dose tested. 3 The potential consequences and therapeutic benefits of inhibition of both FPTase and GGPTase I in the clinic are still uncertain; however, we have created a solid basis for selecting compounds for further development.