Thematic review series: lipid posttranslational modifications. Structural biology of protein farnesyltransferase and geranylgeranyltransferase type I.

More than 100 proteins necessary for eukaryotic cell growth, differentiation, and morphology require posttranslational modification by the covalent attachment of an isoprenoid lipid (prenylation). Prenylated proteins include members of the Ras, Rab, and Rho families, lamins, CENPE and CENPF, and the gamma subunit of many small heterotrimeric G proteins. This modification is catalyzed by the protein prenyltransferases: protein farnesyltransferase (FTase), protein geranylgeranyltransferase type I (GGTase-I), and GGTase-II (or RabGGTase). In this review, we examine the structural biology of FTase and GGTase-I (the CaaX prenyltransferases) to establish a framework for understanding the molecular basis of substrate specificity and mechanism. These enzymes have been identified in a number of species, including mammals, fungi, plants, and protists. Prenyltransferase structures include complexes that represent the major steps along the reaction path, as well as a number of complexes with clinically relevant inhibitors. Such complexes may assist in the design of inhibitors that could lead to treatments for cancer, viral infection, and a number of deadly parasitic diseases.

More than 100 proteins necessary for eukaryotic cell growth, differentiation, and morphology require posttranslational modification by the covalent attachment of an isoprenoid lipid (prenylation) (1). This modification is catalyzed by three protein prenyltransferases: protein farnesyltransferase (FTase) and protein geranylgeranyltransferase type I (GGTase-I), collectively termed the CaaX prenyltransferases, as well as protein GGTase-II (or RabGGTase) [reviewed in this series in (2)], whose substrates are limited to members of the Rab subfamily of G proteins. FTase and GGTase-I transfer a 15 or 20 carbon isoprenoid [donated by farnesyl diphosphate (FPP) or geranylgeranyl diphosphate (GGPP)], respectively, to the cysteine of a C-terminal CaaX motif, defined by a cysteine (C) residue, followed by two small, generally aliphatic (a) residues, and the X residue, which contributes significantly to specificity ( Fig. 1) (3)(4)(5)(6)(7)(8)(9). Kinetic assays and analysis of lipidated CaaX proteins purified from the cell demonstrate the general preference of FTase for methionine, serine, glutamine, or alanine and the preference of GGTase-I for leucine or phenylalanine in the X position. Here, we review the structural biology of FTase and GGTase-I; the biochemical properties of these enzymes have been reviewed extensively elsewhere (1,(10)(11)(12)(13)(14)(15).
After covalent attachment of the isoprenoid in the cytoplasm, most CaaX proteins undergo two further prenylation-dependent processing steps at the endoplasmic reticulum (1): proteolytic removal of the aaX tripeptide by the CaaX protease Ras and a-factor-converting enzyme (Rce1), and carboxymethylation of the prenylcysteine residue by the enzyme Isoprenylcysteine carboxyl methyltransferase (Icmt). The fully processed proteins exhibit high affinity for cellular membranes and present a unique structure at their C termini that can serve as a specific recognition motif in certain protein-protein interactions (Fig. 2).

STRUCTURAL CHARACTERIZATION OF THE CaaX PRENYLTRANSFERASES
Over the past decade, high-resolution crystal structures of FTase and GGTase-I complexed with various substrates, products, and inhibitors have advanced our understanding of these enzymes. Structures representing each of the major steps along the reaction pathway of CaaX prenyl-transferases have been determined (Fig. 3) (75,76). Kinetic studies suggest an ordered binding mechanism in FTase, in which FPP binds to the apo enzyme first, followed by the CaaX substrate (77)(78)(79)(80). The release of farnesylated product is the rate-limiting step of the reaction and occurs only in the presence of excess of either substrate, particularly isoprenoid diphosphate (3,8,78,81,82). This reaction cycle effectively maintains FTase in a substrate-or product-bound complex at all times. Although the GGTase-I reaction has not been studied in such detail, available evidence indicates that it proceeds through a similar pathway (83,84).

Overall enzyme structure
FTase and GGTase-I exist as ab heterodimers and share a common a subunit (48 kDa molecular mass) (16,30) and a homologous b subunit (46 kDa in FTase, 42 kDa in GGTase-I) (3,19,21,31). Both a and b subunits are composed primarily of a helices, which are arranged in a crescent-shaped superhelix (a subunit) wrapped around an a-a barrel (b subunit) (Fig. 4A, B) (76,85).
The structure of the common a subunit is almost identical in both FTase and GGTase-I, with only slight shifts attributable to differences in the interactions with the divergent b subunit. The helices of the a subunit are folded into seven successive pairs, in a series of righthanded antiparallel coiled coils. These "helical hairpins" are arranged in a crescent-shaped superhelix that wraps around the b subunit. The N-terminal domain (z50 amino acids) is disordered in all complexes of FTase and GGTase-I. It has been suggested that this proline-rich domain may interact with other cellular factors, perhaps functioning in enzyme localization, as deletion of this domain does not affect the catalytic activity (86) or the structure of the rest of the protein (87). In the absence of such partners, this domain may be unfolded.
Although the b subunits of FTase and GGTase-I share only 25% sequence identity, they have very similar structures, consisting of 14 a helices in FTase and 13 in GGTase-I. Twelve of the a helices are folded into an a-a barrel (Fig. 4C). Six parallel helices (3b, 5b, 7b, 9b, 11b, and 13b) create the core of the barrel. The other six helices (2b, 4b, 6b, 8b, 10b, and 12b) are parallel with one another and antiparallel with respect to the core helices and form the outside of the barrel. In both enzymes, one end of this barrel is blocked by a loop containing the C-terminal residues of the b subunit, whereas the opposite end is open to the solvent. This arrangement creates a deep, funnel-shaped cavity in the center of the barrel, with an approximate inner diameter of 15 Å and a depth of 14 Å . This cavity is hydrophobic in nature and lined with a number of conserved aromatic residues. The active sites of both FTase and GGTase-I are located within this cavity.
The a and b subunits form an extensive interface, burying .3,000 Å 2 , or z20%, of accessible surface area of each subunit (76,85). The a/b interfaces of both FTase and GGTase-I have lower hydrophobic character than is observed in most subunit interfaces (88). The unusually high polar/charged residue content results in nearly double the typical number of hydrogen bonds. CaaX prenylation and the eukaryotic cell cycle. The prenylation of CaaX proteins (shown here is Ras, in yellow, as an example) by protein prenyltransferases (red and blue ribbon diagram) and further processing by Rce1 (purple) and Icmt (green) are depicted. Also shown is the cholesterol biosynthesis pathway [adapted from (178)], from which the isoprenoid diphosphate substrates farnesyl diphosphate (FPP) and geranylgeranyl diphosphate (GGPP) are derived. A simplified depiction of the Ras signaling pathway, after its insertion into the membrane via its lipid tail, is also illustrated [adapted from (179)]. EGF, epidermal growth factor; GAP, GTPase activating protein; GEF, guanine exchange factor; HMG CoA, 3-hydroxyl-3-methylglutaryl Coenzyme A; Icmt, isoprenylcysteine carboxyl methyltransferase; MEK, mitogen-activated protein (MAP) or extracellular signal-regulated kinase (ERK) kinase; Mnk, MAP kinase-interacting kinase; Rce1, Ras-and a-factor-converting enzyme; RTK, receptor tyrosine kinase.

Zinc binding site
Both FTase and GGTase-I are zinc metalloenzymes, binding one Zn 21 ion per protein dimer (89,90). Crystal structures of FTase and GGTase-I clearly indicate the presence of a single Zn 21 ion bound to the b subunit near the subunit interface (76,85). In both enzyme structures, the Zn 21 ion is coordinated by three strictly conserved residues: D297b, C299b, and H362b in FTase and D269b, C271b, and H321b in GGTase-I. Mutagenesis studies performed on FTase are consistent with Zn 21 coordination by residues D297b, C299b, and H362b (91-93) and additionally implicate D359b in the binding of zinc (92). Crystal structures of FTase (85, 87) reveal a stabilizing hydrogen bond between D359b and the zinc ligand H362b, consistent with a secondary coordination shell effect exerted by the former. A similar interaction is observed in GGTase-I (76).

Isoprenoid diphosphate binding in binary complexes
In the cocrystal structures of FTase and GGTase-I with FPP and GGPP prenyl substrate, respectively [Protein Data Bank identifiers (PDB IDs) 1FT2 and 1N4P], the iso-  (75,85,95,99) have produced structures representing the major steps along this path. In each of these complexes, the enzyme acts as a rigid scaffold; therefore, for clarity, only the substrates and products are shown as they bind in the active site. The path begins with the unliganded (apo) enzyme (0) [Protein Data Bank identifier (PDB ID) 1FT1], with the a subunit shown in red and the b subunit in blue. The FPP molecule binds to form a binary substrate complex (1) (PDB ID 1FT2), followed by binding of the CaaX substrate to form a ternary substrate complex (2) (PDB ID 1D8D). The resulting farnesylated product remains bound in the active site (3) (1KZP). Excess substrate, particularly FPP, facilitates product displacement, demonstrated by a complex in which the new FPP molecule and the partially displaced product are bound simultaneously (4) (1KZO). The double dagger symbol indicates a modeled transition state along the reaction coordinate between 2 and 3 (see Fig. 12 for a more detailed view of the transition state). Throughout this figure, the isoprenoid is shown in blue, the CaaX in yellow, and the catalytic zinc ion in magenta. Also shown are the kinetic parameters determined for this reaction (3,8,(77)(78)(79)(80)(81). Reproduced with permission from (75). prenoid binds along one side of the hydrophobic cavity of the b subunit a-a barrel (Fig. 6A, B; complex 1 in Fig. 3) (76,99). The diphosphate moiety of both FPP and GGPP binds in a positively charged cleft at the rim of the a-a barrel near the a/b subunit interface. This group forms several hydrogen bonds in each enzyme (K164a, H248b, R291b, and Y300b in FTase; K164a, H219b, R263b, K266b, and Y272b in GGTase-I). The farnesyl portion of FPP binds in an extended conformation along one side of the hydrophobic cavity of the a-a barrel, interacting with a number of conserved aromatic residues. The first three isoprene units of GGPP bind in a similar conformation (with similar hydrophobic interactions with aromatic residues in the GGTase-I active site), but the fourth isoprene unit is turned z908 relative to the rest of the molecule. This conformation is significantly different from that of the other known structures with bound isopren-oids [i.e., FTase (99), RhoGDP dissociation inhibitor (GDI) (47), and the phosducin-G t bg complex (100)], in which the isoprenoid is completely extended. Superposition of GGTase-I and RabGGTase (101) by alignment of all enzyme C a atoms illustrates the similarity of their respective GGPP binding sites; the 24 residues surrounding the GGPP molecule in GGTase-I are identical or structurally conservative substitutions in RabGGTase. The high degree of homology in the prenyl binding site indicates a similar mode of binding for the GGPP molecule in RabGGTase.
Isoprenoid diphosphate substrate specificity: the ruler hypothesis CaaX prenyltransferase activities are highly selective for their isoprenoid diphosphate substrates: FTase for FPP and GGTase-I for GGPP. Binding studies demonstrate, however, that FTase can bind both FPP and GGPP with nanomolar affinity, yet the enzyme is unable to transfer the geranylgeranyl group to substrate proteins (17,78,79,89,102), indicating that GGPP is an inhibitor of FTase. GGTase-I similarly binds both FPP and GGPP, although it binds FPP with somewhat lower affinity (79,103). Unlike FTase, GGTase-I can (mis)transfer a farnesyl group to substrate proteins, but with reduced efficiency (83).
Superposition of FTase and GGTase-I in binary isoprenoid substrate complexes by aligning all enzyme C a atoms illustrates the similarity of binding of FPP and GGPP in their respective enzymes ( Fig. 7A) (76). There is a slight divergence in the binding of the first isoprene unit, but the second and third units are essentially indistinguish-able. The binding sites for GGPP in GGTase-I and FPP in FTase are very similar, constituting a cavity lined with conserved aromatic residues. The primary differences are at the bottom of this cavity, where the fourth isoprene unit of GGPP binds in GGTase-I. Here, FTase has the bulky tryptophan and tyrosine residues (W102b and Y365b), whereas the corresponding positions in GGTase-I (and RabGGTase) are occupied by the smaller residues threonine and phenylalanine to accommodate the fourth isoprene unit of the GGPP molecule, which cannot fit in FTase. Therefore, FTase and GGTase-I can function as simple length-discriminating molecular rulers. This hypothesis was tested by introducing mutations into FTase that are predicted to convert it to a geranylgeranyltransferase   (76). A mutation at the W102b residue of FTase to a threonine (the corresponding residue in GGTase-I) (W102T mutant) was created. Steady-state activity assays indicated that this single mutation results in an FTase enzyme greatly preferring GGPP over FPP as its isoprenoid substrate without significantly altering CaaX sequence specificity (Fig. 7B).

Peptide and isoprenoid binding in ternary complexes
Crystal structures of the CaaX prenyltransferases with bound peptide substrates and nonreactive isoprenoid analogs revealed the conformation of the bound CaaX substrate (complex 2 in Fig. 3) (76,94,95). Each of these FTase and GGTase-I complexes includes a peptide derived from the K-Ras4B (CVIM Ca 1 a 2 X sequence; PDB IDs 1D8D and 1QBQ) or Rap2b (CVIL Ca 1 a 2 X sequence; PDB ID 1N4Q) protein substrate, respectively, taking advantage of the ability of these enzymes to prenylate peptides as short as the CaaX tetrapeptide (3,6,18). The structure of the enzyme in these complexes is essentially identical to the previously determined structures, with the exception of a few minor side chain rearrangements in the vicinity of the diphosphate binding site compared with the unliganded FTase structure (PDB ID 1FT1) (85).
In each of these structures, the isoprenoid forms a large part of the CaaX sequence binding surface, with its second and third isoprene (and fourth, in the GGPP analog) units in direct van der Waals contact with the CaaX motif, particularly the a 2 and X residues (Fig. 8A, B). The peptide substrate sandwiches the isoprenoid against the wall of the hydrophobic cavity, burying 115 and 140 Å 2 (in FTase and GGTase-I, respectively) of accessible isoprenoid diphosphate analog surface area and completely sequestering the third isoprene unit from the bulk solvent. The location and conformation of the nonreactive isoprenoid analogs used in these studies are similar to those observed in the binary complex with FPP and GGPP. Furthermore, the enzyme residues that interact with the isoprenoid diphosphate substrate molecule were also observed to interact with the analogs. These structures, therefore, are consistent with the ordered binding pattern observed by kinetics in which isoprenoid diphosphate binding precedes peptide binding (77,78).
In the FTase and GGTase-I ternary substrate complexes, the peptide Ca 1 a 2 X motif binds in an extended conformation in the hydrophobic cavity of the b subunit. As predicted from spectroscopic studies (81), the cysteine residue of the peptide Ca 1 a 2 X sequence directly coordinates the zinc ion (Fig. 5). The side chain of the peptide valine residue (position a 1 ) points into solvent without making any hydrophobic interactions with the isoprenoid. The isoleucine side chain of the peptide substrate (a 2 ) is positioned in close proximity to the isoprenoid, making a hydrophobic interaction, whereas the backbone carbonyl oxygen of this residue participates in a hydrogen bond with the enzyme (R202b in FTase, R173b in GGTase-I).
The C-terminal carboxylate of the X residue (methionine in FTase, leucine in GGTase-I) forms hydrogen bonds with Q167a and a buried water molecule coordinated by three enzyme residues (H149b, E198b, and R202b in FTase; H121b, E169b, and R173b in GGTase-I). The residue in position X is also in van der Waals contact with a number of enzyme residues (Y131a, A98b, S99b, W102b, H149b, A151b, and P152b in FTase; T49b, H121b, A123b, and F174b in GGTase-I). These interactions bury a large portion of the accessible surface area of the residue in position X. As noted, the only directional, bonded interactions between the peptide CaaX sequence and the protein are the Zn 21 -to-cysteine thiol(ate) interaction and the hydrogen bonds involving the carbonyl oxygen of the a 2 isoleucine residue and the C-terminal carboxylate. Together, these interactions help anchor the CaaX peptide into its observed substrate binding conformation. Further insight into the role of zinc in peptide substrate binding is derived from a crystal structure of zinc-depleted rat FTase with bound FPP analog and a K-Ras4B-derived peptide (PDB ID 1D8E) (95). The overall structure of the enzyme is essentially identical to the other FTase structures, proving that zinc is not required to maintain the proper three-dimensional fold of FTase. Furthermore, the conformations of the zinc ligands (D297b, C299b, and H362b) remain largely undisturbed. Both the FPP analog and peptide are observed bound in the active site without zinc. The FPP analog conformation is unchanged from the other ternary complexes, supporting the observation that zinc is not required for FPP binding (89). In this complex, the a 2 and X residues of the CaaX peptide bind in the same manner as in the presence of zinc, whereas the cys-teine and valine residues bind much differently, forming a b-turn. The cysteine thiol(ate) has shifted z9 Å from its position when in coordination of the zinc ion. This altered CaaX binding conformation of the peptide in the absence of zinc indicates that it is the zinc-thiol(ate) interaction that anchors the cysteine residue, holding the peptide in an extended conformation. This observation is consistent with the finding that peptide affinity is greatly enhanced in the zinc metalloenzyme (89). This zinc-depleted structure also indicates that, although not strictly required for peptide binding, the zinc ion is required both to orient the cysteine thiol(ate) for catalysis and to stabilize a productive peptide substrate conformation.
A superposition of these ternary complexes by aligning all C a atoms ( Fig. 9) reveals that the CaaX portion of all protein substrates adopts a common binding mode in the FTase and GGTase-I active sites and suggests a series of rules that define peptide substrate selectivity for the CaaX prenyltransferases. The presence of two fixed anchor points within the recognition motif (cysteine and C terminus) that make specific interactions with the enzyme discriminates against peptides that are either too long or too short (i.e., only tetrapeptides) or that lack a cysteine at the correct position. The a 1 residue is solvent-exposed and can accommodate any amino acid, as observed in solution studies (6,76,94,95). Both the a 2 and X positions are buried in the active site; consequently, they are the major determinants of peptide selectivity through steric and electrostatic complementarity between the amino acid side chains and enzyme residues.
In these structural studies, it was observed that the binding sites on FTase and GGTase-I that recognize the a 2 peptide residue have unique steric and aromatic properties, suggesting that the a 2 residue may affect Ca 1 a 2 X recognition. The structures indicate that the a 2 binding site can accommodate a variety of amino acids, including valine, isoleucine, leucine, phenylalanine, tyrosine, proline, threonine, and methionine, consistent with previous studies of FTase peptide selectivity in mammalian and yeast systems (6,104). Comparison of these sites illustrates the increased aromatic character of FTase relative to GGTase-I, which has subtle effects on CaaX specificity but a more marked effect on inhibitor binding (see below) ( Fig. 10).
The X position is the primary sequence determinant that specifies whether a peptide is a substrate for FTase, GGTase-I, both, or neither. The structures reveal that FTase and GGTase-I have different X-residue binding pockets, each with distinct electrostatic properties. The Xresidue binding pocket in FTase is more polar and accepts the residues methionine, serine, and glutamine (as well as alanine, threonine, and cysteine to a lesser extent). This pocket in GGTase-I has more hydrophobic character and accepts leucine and phenylalanine (as well as isoleucine and valine to a lesser extent). Both pockets discriminate against bulky amino acids such as tyrosine, tryptophan, and arginine through steric complementarity. Interestingly, this study also revealed a previously unobserved secondary X residue binding site in FTase, in which the C-terminal phenylalanine residue of the TC21-derived peptide binds. This pocket may also accommodate leucine, asparagine, or histidine.
Together, these structures outline a series of rules that govern substrate peptide selectivity. These rules were used to identify potential protein substrates within the human genome, including a number of possible CaaX prenyltransferase substrates not identified previously.

Product formation
In both FTase and GGTase-I, the structures of the product complex (complex 3 in Fig. 3) (PDB IDs 1KZP and 1N4R) and ternary substrate complexes are virtually identical (75). There are no observed changes in the polypeptide backbone; differences are confined to changes in side chain rotamers of those active site residues that interact with the diphosphate moiety of the isoprenoid diphosphate substrate and are a consequence of the pyrophosphate product release. Equally, the peptide exhibits no appreciable differences between its reactant and product structures (Fig. 11A). The isoprenoid, however, moves toward the peptide in the product structure, with the C 1 atom nearly 7 Å closer to the Zn 21 ion compared with its position as a substrate. This motion is localized, as the third (and fourth, in the geranylgeranyl moiety) isoprene unit retains its substrate binding position; only the first and Fig. 9. Comparison of Ca 1 a 2 X substrate binding in FTase and GGTase-I. Superpositions of four FTase substrate complexes (left panel) and three GGTase-I substrate complexes (right panel) demonstrate that cognate and cross-reactive peptides adopt a common binding mode (PDB IDs 1D81, 1TN6, 1TN7, 1TN8, 1TNO, 1TNU, and 1TNB). In FTase, however, the C-terminal phenylalanine residue of TC21 binds in a different pocket than the Cterminal methionine, glutamine, and serine residues. Reproduced with permission from (97). Fig. 10. Comparison of the a 2 binding site in FTase and GGTase-I. Superposition of FTase and GGTase-I shows residues that interact with the a 2 residue of the Ca 1 a 2 X motif (PDB IDs 1D81 and 1N4Q). Regions of the FTase ternary complex forming the a 2 binding site (enzyme residues W102b, W106b, and Y361b and isoprene 3) are colored red. Corresponding regions of the GGTase-I ternary complex (enzyme residues T49b, F53b, and L321b, isoprenes 3 and 4, and substrate peptide X residue) are colored blue. Portions of the FPP and GGPP analogs (gray) and of the CVIM (pink) and CVIL (light blue) peptides not forming the a 2 site are also shown. Reproduced with permission from (97). second isoprene units shift, and they do so by rotating through the second isoprene unit. These results suggest that only a portion of the isoprenoid moves substantially during catalysis.

Product release
We note that the product complex described above (complex 3 in Fig. 3) does not represent the final step in this reaction. As described above, the rate-limiting step of the prenylation reaction is product release (78,105), requiring binding of the next substrate(s) (82). This final step is represented by structures of FTase and GGTase-I complexed with both isoprenoid substrate and prenylated peptide product (complex 4 in Fig. 3) (PDB IDs 1KZO and 1N4S), which reveal that binding of the fresh isoprenoid diphosphate substrate molecule moves the prenyl group of the prenylated peptide to a new binding site, the "exit groove" (Fig. 11B) (75,76). This movement is accompanied by a conformational change in the CaaX moiety. Therefore, there are two CaaX peptide conformations in the product: extended (complex 3 in Fig. 3) and type I b-turn (complex 4 in Fig. 3). NMR studies of peptides bound to FTase revealed a similar b-turn, which was postulated to correspond to a reactant rather than a product complex (106,107). In the displaced product, the X and a 2 residues remain in their original positions, indicating that the peptide substrate is competitive with respect to the displaced product. The three C-terminal amino acids of the prenylated peptide product make extensive van der Waals contacts with the new isoprenoid substrate, suggesting that the amino acid sequence of the CaaX motif may modulate product release. Multiple binding sites for product molecules have been observed in processive enzyme systems, in which substrate binding is involved in product release or translocation [e.g., the P and E sites of the ribosome (108)]. We note that this complex represents a direct observation of both substrate and product bound simultaneously to an enzyme that catalyzes an apparently nonprocessive reaction.

Implications of substrate-mediated product release
Unlike FTase and GGTase-I, RabGGTase is a processive enzyme, adding two prenyl groups to its Rab substrates (52) via a process that appears to proceed without dissociation of the monoprenylated intermediate (109) [reviewed in this series in (2)]. Biochemical and structural studies indicate that, like FTase and GGTase-I, RabGGTase has only one prenyl diphosphate binding site (101,110). The structure of RabGGTase (PDB ID 1DCE) revealed a striking 7.5 Å diameter hydrophobic tunnel, originating from a shallow groove analogous to the exit groove in FTase and GGTase-I and extending through the b subunit (101,110). The first geranylgeranyl isoprenoid attached to Rab may be sequestered in this tunnel after translocation of the monoprenylated product from the site of catalysis, which may be associated with the binding of fresh GGPP substrate, as observed for FTase and GGTase-I (109,111). This translocation would allow the unmodified cysteine residue to interact with the zinc ion in preparation for the second catalytic reaction. This mechanism would explain why RabGGTase preferentially modifies Rab substrates that have one geranylgeranyl group already attached (52,109). In addition, the potential to insert different lengths of the geranylgeranyl moiety through the tunnel may explain how RabGGTase is able to modify cysteine residues at a variety of positions within a few residues of the C termini of Rab substrates.
Additionally, this unusual substrate-mediated product release may have a significant biological role. We propose that it provides a mechanism for the regulated handover of the prenylated protein product to the next step in this processing pathway (76). In the CaaX prenyltransferases, the prenylated product remains tightly bound; therefore, it is shielded from the cytoplasm, preventing aggregation or association with an incorrect membrane compartment. Only upon binding of an additional isoprenoid diphosphate molecule is the product isoprenoid displaced into the exit groove, presenting the product for delivery to the next processing step. The mechanism for this handover is analogous to one of the roles of Rab escort protein (REP), which escorts RabGGTase products to their final destination (112), and is similar to that of the GDI proteins, which extract prenylated G proteins such as Rho from membranes by binding the isoprenoid moiety in a hydrophobic pocket and transporting them throughout the cell (47).

Transition state model
The chemical mechanism and nature of the transition state is still an area of active investigation. We have developed a transition state model for catalysis by CaaX prenyltransferases consistent with the observed structural and kinetic information (Fig. 12) (75, 76). The proposed model retains elements of both the postulated electrophilic and nucleophilic components of the reaction mechanism (80,113,114). In this model, the CaaX motif binds as observed in the ternary substrate complex, coordinating the Zn 21 ion, whereas the isoprenoid diphosphate rotates through its second isoprene unit to bring the C 1 atom into alignment for thioether bond formation, maintaining the binding conformation of the b-phosphate in its substrate conformation. In this reaction, the isoprenoid C 1 atom undergoes inversion of configuration during the reaction (115)(116)(117). The developing negative charge on the diphosphate moiety (localized mainly on the a-phosphate) is stabilized by several interactions with the enzyme, particularly residues K164a and a b subunit tyrosine (Y300b in FTase, Y272b in GGTase-I). Kinetic studies demonstrated that FTase, but not GGTase-I, requires millimolar levels of Mg 21 for full catalytic efficiency (19,89). It has been hypothesized that the Mg 21 ion required by FTase stabilizes the developing negative charge on the diphosphate as the bond breaks between the a-phosphate and the C 1 atom of the farnesyl group. Structural studies suggest that Mg 21 is coordinated by residue D352b of the enzyme and the FPP diphosphate moiety (75). Sequence alignments reveal that in GGTase-I, a lysine occupies the position corresponding to D352b in FTase. Superposition of the structures of FTase (75,85,95,96,99) and GGTase-I (76) shows that this lysine residue adopts a conformation that positions the positively charged side chain N e at the site of Mg 21 in FTase, allowing it to stabilize the diphosphate group. Mutagenesis studies (118,119) indicate that Mg 21 sensitivity is dependent on these two residues; a lysine at 352b of FTase abolishes Mg 21 dependence, whereas an aspartate at 311b in GGTase-I introduces Mg 21 dependence.

Isoprenoid diphosphate-derived inhibitors
A number of FPP and GGPP analogs have been developed that are competitive inhibitors of FTase and GGTase-I, respectively (135)(136)(137)(138)(139)(140)(141). Systematic structureactivity analysis of a series of FPP-based inhibitor molecules indicated that the most potent inhibitors retain a hydrophobic farnesyl group and a negatively charged moiety mimicking the diphosphate (138), consistent with the interactions observed previously between the enzyme and the terminal phosphate in the FTase and GGTase-I complexes (75, 76, 94-97, 99, 142, 143). That study also revealed the dramatic effect of hydrophobic chain length on FTase activity, because elongation of the farnesyl group by a single carbon resulted in a decrease in IC 50 of .200-fold, consistent with the structure-derived ruler hypothesis.
Several isoprenoid analog inhibitors have been characterized in crystal structures of various complexes (76, 94-97, 144, 145). Some of these inhibitors are described in this review (see above), including the FPP analogs farnesyl protein transferase inhibitor II (95-97) and a-hydroxyfarnesylphosphonic acid (94) and the GGPP analog 39-aza GGPP (76,97). In each of these structures, the isoprenoid analogs bind in a manner similar to their substrate counterparts, interacting extensively with both the enzyme and the CaaX substrate. These inhibitors continue to be invaluable in the structural characterization of FTase and GGTase-I; they bind much like the isoprenoid diphosphate substrates but will not react with the peptide, allowing the crystallization of ternary substrate complexes.

Peptide-competitive inhibitors
Peptidomimetics. Initial characterization of FTase demonstrated that this enzyme is inhibited by short Ca 1 a 2 X peptides (3,4,6). The most effective inhibitory peptides contain a nonpolar aliphatic or aromatic amino acid in the a 1 or a 2 position, particularly the latter, although some of these peptides can also serve as FTase substrates. Inclusion of an aromatic residue such as phenylalanine in the a 2 position eliminates prenylation, however, resulting in a purely competitive inhibitor with respect to the peptide. Interestingly, the removal or substitution of the amino group on the cysteine of a tetrapeptide inhibitor restores Fig. 13. Comparison of CaaX nonsubstrate and substrate conformations. A: CaaX tetrapeptide nonsubstrate inhibitor binding conformation. L-739,750 peptidomimetic (PDB ID 1JCQ) and CVFM tetrapeptide (PDB ID 1JCR) are shown with the FPP molecules observed in these two complexes colored yellow (with peptidomimetic) and blue (with CVFM). B: Peptide substrate binding conformation. TKCVFM (PDB ID 1JCS) and the K-Ras4B-derived peptide TKCVIM (PDB ID 1D8D) are both substrates and adopt the same backbone conformation. The N-terminal residue and the lysine side chain of each of these peptides are omitted here for clarity. Also shown are the FPP analogs, which are colored according to their observed conformations with these peptides. C: Comparison of the hexapeptide substrate TKCVFM and the nonsubstrate tetrapeptide inhibitor CVFM. A boxed region highlights the differences between the substrate and nonsubstrate conformations. D: Energetically unfavorable steric contact (red spikes) between the C g2 methyl group of the isoleucine residue in the a 2 position and the carbonyl oxygen of the a 1 residue when the phenylalanine of the CVFM tetrapeptide is replaced with isoleucine. Consequently, a tetrapeptide with a b-branched residue (isoleucine or valine) in the a 2 position cannot adopt the nonsubstrate binding mode observed for CVFM. Reproduced with permission from (96). catalytic activity (4,7). Because these molecules are peptides, they have three significant disadvantages for potential in vivo activity: poor cellular uptake, rapid degradation, and FTase farnesylation (in some cases), which results in depletion of the inhibitor. To overcome these disadvantages, a number of peptidomimetic inhibitors have been developed and characterized (96,121,(146)(147)(148)(149)(150).
The mechanism for tetrapeptide inhibition was studied using the crystal structures of FTase complexed with the nonsubstrate tetrapeptide CVFM (PDB ID 1JCR), the substrate hexapeptide TKCVFM (PDB ID 1JCS), and the peptidomimetic inhibitor L-739,750 (PDB ID 1JCQ), developed by Merck Research Laboratories (96). This inhibitor compound, mimicking the tetrapeptide CIFM, is the metabolic product of the prodrug L-744,832; administration of this prodrug in rats resulted in tumor regression without any observed systemic toxicity (121). Each of these peptides/peptidomimetics bind in the peptide binding site, anchored by cysteine coordination of the catalytic zinc ion and hydrogen bonds (both direct and water-mediated) between the C terminus and the enzyme. The two inhibitors (the nonsubstrate CVFM peptide and L-739,750) adopt a similar conformation, in which there is an z3 Å distance shift of the inhibitor cysteine C a atom relative to the substrate, although the cysteine still coordinates the zinc ion (Fig. 13A, C). This alternative conformation is stabilized by a hydrogen bond between the N terminus of the inhibitor and an oxygen on the a-phosphate of the FPP molecule, as well as by hydrophobic interactions between the phenylalanine-derived side chain of the inhibitor and several aromatic residues of the enzyme. In both complexes, the inhibitor binds in the space between the isoprenoid and the zinc ion and interferes with the movement of the isoprenoid during the reaction, preventing bond formation. The substrate TKCVFM hexapeptide, however, adopts the same conformation as the K-Ras4Bderived TKCVIM peptide in an analogous complex (Fig. 13B). In each of these complexes, the side chain in the a 2 position adopts essentially the same conformation; this conformation places steric constraints on the f backbone angle between the a 1 and a 2 residues. b-Branched amino acids (isoleucine and valine) in the a 2 position form an unfavorable steric interaction between their C g2 methyl group and the carbonyl oxygen of the a 1 residue in the nonsubstrate backbone conformation adopted by the CVFM peptide (Fig. 13D). Therefore, it was predicted that, of the CaaX cognate tetrapeptides, only peptides with branched residues (i.e., isoleucine or valine) in the a 2 position are substrates, whereas peptides with other nonbranched aliphatic amino acids are poor substrates or nonsubstrates (provided that the N terminus is not blocked).
Nonpeptide inhibitors. Another class of peptide-competitive inhibitors includes molecules that are not derived from peptides but still bind in the CaaX binding site of the enzyme; this class is by far the most prevalent and the most characterized (122-125, 144, 151-165). This class of inhibitors includes compounds produced by the drug discovery and design efforts of Merck Research Laboratories, Janssen Research Foundation, Bristol-Myers Squibb Co., Schering-Plough, and Abbott Laboratories, some of which have been evaluated in phase I, II, and III clinical trials for the treatment of a variety of human cancers (122,123). Crystal structures of a number of these inhibitors complexed with the CaaX prenyltransferases have been determined. In each of these structures, the compounds bind in the peptide binding site, typically without inducing large structural changes in the enzyme. Each of these compounds participates in extensive hydrophobic interactions with both the enzyme and the FPP/FPP analog molecule, as well as in a combination of direct and watermediated hydrogen bonds with the enzyme.
The structures of FTase complexed with FPP (or an FPP analog) and each of the inhibitors R115777 (Janssen; PDB ID 1SA4) (142), BMS-214662 (Bristol-Myers Squibb; PDB ID 1SA5) (142), and ABT-839 (Abbott Laboratories; PDB ID 1N94; one of several Abbott compounds with determined structures) (161) have been determined. All three inhibitors bind in the FTase peptide binding site, overlapping with the binding of the Ca 1 a 2 portion of the substrate Ca 1 a 2 X peptide (Fig. 14). R11577 and BMS-214662 coordinate the catalytic Zn 21 ion via an imidazole group; however, ABT-839 does not. ABT-839 is also unique in that it possesses a methionine mimic at one end of the molecule that binds as in the substrate complex (95), making it the only nonpeptidomimetic inhibitor described here that extends into the X residue binding site.
Design efforts from Schering-Plough have resulted in a number of FTase inhibitors, of which a series of tricyclic compounds have been structurally characterized (144). These compounds are composed of three connected rings, with a tail containing two or more rings extending from the central seven carbon ring, representing a series of mono-, di-, and trihalogenated species. SCH66336, an inhibitor in this series, is observed in an unusual conformation in the FTase active site (PDB ID 1O5M) (Fig. 15). This compound does not coordinate the Zn 21 ion and only overlaps the a 1 and a 2 portion of the substrate Ca 1 a 2 X peptide. Additionally, a large portion of the inhibitor is observed binding in the exit groove described above, where the displaced prenyl group of the farnesylated peptide product binds in the translocated product complex. The latter observation may explain the binding affinity toward FTase that this inhibitor exhibits, despite its lack of Zn 21 coordination.
Although most of these inhibitors are highly selective toward FTase in preference to GGTase-I, L-778,123 (Merck Research Laboratories) is a nonpeptide inhibitor of both FTase and GGTase-I (154,155). No steric blocks prevent compounds such as R11577 and BMS-214662 from binding in GGTase-I, although binding is not favored because the GGTase-I a 2 binding site has a reduced aromatic character that does not permit the stabilizing aromatic stacking interactions (166,167) observed in FTase (see above). These differences in the a 2 binding site may cause the alternative binding mode adopted by L-778,123 in GGTase-I (Fig. 16) (143). In FTase-I (PDB ID 1S63), the peptide-competitive binding mode is stabilized by aromatic stacking interactions with the aromatic residues that constitute the a 2 binding site. In GGTase-I (PDB ID 1S64), these stacking interactions are not possible, and L-778,123 adopts a GGPP-competitive binding mode that permits the formation of a hydrogen bond in lieu of stacking interactions. In this structure, the inhibitor also interacts with a bound sulfate anion; this synergistic interaction may additionally induce a small change in the GGPP binding site that enhances L-778,123 binding (168,169).

Bisubstrate mimics
The final class of prenyltransferase inhibitors includes a variety of molecules that mimic a bisubstrate complex and are competitive for both the isoprenoid diphosphate and CaaX binding sites. Several of these inhibitors have been synthesized and characterized (170)(171)(172), although none has been studied by structural methods. Fig. 16. L-778,123 binding in FTase and GGTase-I suggests a molecular mechanism for inhibitor selectivity. Superposition of FTase (blue) and GGTase-I (red) complexes illustrates the binding modes of L-778,123 in each enzyme (PDB IDs 1S63 and 1S64, respectively). Only the residues that contact the a 2 residue of the Ca 1 a 2 X peptide, the "a 2 binding site," and a conserved arginine (Arg) residue are shown. Aromatic stacking interactions are not possible between L-778,123 and the GGTase-I a 2 binding site, encouraging the inhibitor to adopt a lipid-competitive binding mode that permits the formation of a hydrogen bond. Reproduced with permission from (143).

STRUCTURAL ANALYSIS OF CaaX PRENYLTRANSFERASE AND HUMAN THERAPEUTICS
Since their discovery, FTase and GGTase-I have been targets for cancer therapeutics (73,74). To this end, a number of FTIs and GTIs have been studied kinetically, structurally, and clinically (reviewed in this series in 120). More recent studies have begun to illustrate the potential of FTIs in the treatment of infection by pathogenic microorganisms, including Plasmodium falciparum (the causative agent of malaria) (29,128), Trypanosoma brucei (African sleeping sickness) (126,173), Trypanosoma cruzi (Chagas disease) (35), and Leishmania mexicana and Leishmania major (leishmaniasis) (35,127) (reviewed in this series in 174). Another potential application of FTIs and GTIs is as a new class of antifungal and antiviral agents. Candida albicans and Cryptococcus neoformans (opportunistic fungal pathogens causing life-threatening infections in many immunocompromised patients with AIDS) (24,25,175,176) and viruses (through the use of host prenyltransferase enzymes), including hepatitis C virus and hepatitis delta virus, appear to require prenylation for viability (25,129,132,133).
Although structural characterization of mammalian FTase and GGTase-I has advanced our understanding of these enzymes, there is no structural information available for any of the protein prenyltransferases from human pathogens. FTIs are attractive drug candidates to develop for the treatment of parasitic, fungal, and viral infections because the chemistry, pharmacokinetics, and toxicity of a number of these compounds have been established in humans. Using this "piggy-back" approach to drug development bypasses many of the hurdles involved in designing drugs de novo (177). Structural studies of the protein prenyltransferases from human pathogens would facilitate the design of potent, highly selective inhibitors.