Quantitative Analysis of Prenylated RhoA Interaction with Its Chaperone, RhoGDI*

Background: RhoGDI is a key regulator and a chaperon of Rho GTPases. Results: RhoGDI strongly discriminates between GDP- and GTP-bound forms of prenylated RhoA, although both complexes are of high affinity. Conclusion: We provide direct evidence for the existence of two populations of the RhoGDI·RhoA complexes in the cell, characterized by different lifetimes. Significance: The obtained data allows us to formulate the model for membrane delivery and extraction of Rho GTPases. Small GTPases of the Rho family regulate cytoskeleton remodeling, cell polarity, and transcription, as well as the cell cycle, in eukaryotic cells. Membrane delivery and recycling of the Rho GTPases is mediated by Rho GDP dissociation inhibitor (RhoGDI), which forms a stable complex with prenylated Rho GTPases. We analyzed the interaction of RhoGDI with the active and inactive forms of prenylated and unprenylated RhoA. We demonstrate that RhoGDI binds the prenylated form of RhoA·GDP with unexpectedly high affinity (Kd = 5 pm). The very long half-life of the complex is reduced 25-fold on RhoA activation, with a concomitant reduction in affinity (Kd = 3 nm). The 2.8-Å structure of the RhoA·guanosine 5′-[β,γ-imido] triphosphate (GMPPNP)·RhoGDI complex demonstrated that complex formation forces the activated RhoA into a GDP-bound conformation in the absence of nucleotide hydrolysis. We demonstrate that membrane extraction of Rho GTPase by RhoGDI is a thermodynamically favored passive process that operates through a series of progressively tighter intermediates, much like the one that is mediated by RabGDI.

cycle is tightly controlled by guanine nucleotide exchange factors (GEFs) 2 that facilitate loading Rho with guanosine triphosphate (5) and GTPase-activating proteins that dramatically accelerate the intrinsic rate of GTP hydrolysis (6).
Similar to the majority of other Ras superfamily members, the Rho proteins are posttranslationally prenylated on their C termini. Rho GTPases are predominantly singly geranylgeranylated, although farnesylation has also been reported (7). These modifications enable small GTPases to reversibly and dynamically associate with intracellular membranes. In a number of cases, membrane delivery was shown to be functionally coupled to GTPase activation. For instance, GTP loading of the Arf GTPases exposes their otherwise buried myristoylated N-terminal helix and promotes association of Arf proteins with the plasma membrane (8). Similarly, GEF-mediated loading of Rab GTPases with GTP reduces its affinity for its chaperone, Rab GDP dissociation inhibitor (RabGDI), by 3 orders of magnitude and promotes association of GTP-bound RabGTPases with their target membrane (9).
As in the case of Rab proteins, membrane/cytosol cycling of geranylgeranylated Rho GTPases is regulated by functionally similar but structurally unrelated GDP dissociation inhibitors (RhoGDIs). In contrast to the RabGDI, RhoGDIs were reported to interact with both GDP-and GTP-bound Rho proteins with similar efficiency (10 -13). However, the exact affinities of these interactions are still disputed (14 -16).
The lack of clear evidence for the role of nucleotide exchange in the dissociation of Rho⅐RhoGDI complexes has led to a search for alternative mechanisms of the membrane recruitment of Rho. A number of putative GDI displacement factors were proposed to promote dissociation of the RhoGDI⅐Rho * This work was supported in part by Australian Research Council DP Grant complexes in vivo. These include receptors, such as p75 NTR (17) and integrins (18), and associated proteins, such as ERM (19) and merlin (20). However, no thermodynamically sound model has been proposed that would explain how GDI displacement factors could reduce the high affinity of the RhoGDI/Rho interaction. One alternative explanation is that these molecules act as scaffold proteins that recruit RhoGEFs (21)(22)(23) and membrane phosphoinositides (21), which have been proposed to facilitate Rho⅐RhoGDI complex dissociation (22). Phosphorylation of RhoGDI or Rho proteins was also shown to influence the Rho/RhoGDI interaction, strongly affecting their association in the cell (23)(24)(25)(26)(27). Phosphorylation of Rho GTPases is believed to promote their extraction from the membranes by RhoGDI, whereas phosphorylation of RhoGDI diminishes its ability to bind Rho. Despite numerous reports on the possible mechanisms that govern membrane delivery and recycling of Rho GTPases, a kinetic and thermodynamic description of such mechanisms is still outstanding. Therefore, this study reports on the quantitative biophysical analysis of RhoA interaction with RhoGDI with the objective of determining the kinetic and thermodynamic framework of the Rho functional cycle. In contrast to previous findings, we have demonstrated that RhoGDI discriminates strongly between different nucleotide-bound forms of prenylated and unprenylated RhoA. We provide direct evidence for the existence of two populations of the RhoGDI⅐RhoA complex in the cell, characterized by different lifetimes. We also propose a thermodynamic model of RhoGDI-mediated RhoGTPase recycling.

MATERIALS AND METHODS
Protein Expression and Purification-All proteins used in this study were expressed in the Escherichia coli BL21(DE3) Codon plus RIL strain (Stratagene). Expression and purification of FTase, GGTase-I, and RhoA F25N were carried out as described previously (28). RhoGDI was cloned in-frame with maltose-binding protein using the pOPINM vector (29), expressed in E. coli, and purified by nickel-nitrilotriacetic acid chromatography. The tag was removed by proteolysis with Precision protease, and the protein was further purified by gel filtration on a Superdex 200 26/60 column (GE Healthcare). The L193A mutant of RhoA F25N was constructed by PCR mutagenesis of the pGATEV-RhoAF25N plasmid using the forward 5Ј-GATGGGTATCCATGGCTGATG-3Ј, and reverse primer 5Ј-GGATCCCTCGAGTCATGCGACAAGGC-3Ј, and subcloning the resulting PCR product into the NcoI and XhoI sites of the pGATEV vector (30). The C-terminal citrine-RhoGDI was cloned using the in-fusion protocol into the pOPINE vector (29). The CFP N-terminal fusion of RhoA was cloned into the pGATEV vector using the forward 5Ј-CATCACAACACTAG-TATGGTGAGCAAGGGCGAGGAGCTG-3Ј, and reverse 5Ј-ATTCTCATCAGCC-ATGGACTTGTACAGCTCGTCCAT-GCCG-3Ј, primers and cloning of the PCR fragment into SpeI and NcoI sites.
Preparation of GMPPNP-, mantGDP-, and mantGMPPNPbound forms of the RhoA-Nucleotide exchange on the RhoA molecule was performed as described earlier (31).
After incubation for 4 h at room temperature, the reaction mixture was chromatographed on a Superdex 200 10/30 gel filtration column (GE Healthcare) equilibrated with prenylation buffer without ZnCl 2 . The flow rate was 0.35 ml/min, and fractions of 0.3 ml were collected and analyzed by 15% SDS-PAGE. When the fluorescent analog was used as a substrate, the gel was scanned using a Typhoon Trio fluorescent scanner (GE Healthcare) (excitation 488 nm; cutoff filter 520 nm), and the gel was subsequently stained with Coomassie Blue. Isoprenoid conjugation was confirmed by electrospray ionization-mass spectrometry of the selected fractions. The eluted fractions containing the protein complex were combined, concentrated, flash-frozen in liquid nitrogen, and stored at Ϫ80°C.
The complex of geranylgeranylated RhoA and GGTase-I was obtained by size exclusion chromatography on a Superdex 200 column of the in vitro prenylation reaction containing equimolar amounts of GGTase-I and RhoA. The fractions containing the protein complex were collected, concentrated to 10 mg/ml, and stored as described above.
Labeling of the RhoGDI with Maleimide Dyes-The tetramethylrhodamine-maleimide dye was obtained from Molecular Probes, and RhoGDI labeling was performed according to the manufacturer's protocol. The degree of labeling, based on spectroscopic observation, was typically 50%.
Fluorescence Measurements-Fluorescence spectra and longtime base fluorescence measurements were performed at 25°C in 1-ml quartz cuvettes (Hellma, Germany) with continuous stirring on a Spex Fluoromax-3 or Fluoromax-4 spectrofluorometer (Jobin Yvon Inc., USA). Fluorescence titrations were carried out in buffer containing 25 mM Hepes-NaOH, pH 7.2, 40 mM NaCl, 2 mM MgCl 2 , 5 mM dithioerythritol, and 20 M guanosine nucleotide (GDP or GMPPNP). NBD-Geranyl was excited at 479 nm (slit 4 nm), and the emission was monitored at 560 nm (slit 8 nm). The dissociation of the fluorescent CFP-RhoA-GG⅐RhoGDI-citrine complex was monitored with extinction at 436 nm (4 nm slits), and emission spectra at 475-530 nm were read. The FRET efficiency (E FRET ) was defined as intensity of the citrine (acceptor, 530 nm) divided by the sum of the intensity of the CFP (donor, 475 nm) and acceptor.
Primary data analysis was performed with the programs Grafit 5.0 (Erithacus software) and "fluorescence" implementation (Jobin Yvon Inc., USA) of Origin 7.0 (Originlab Corp., USA). Measurements were done on an inverted laser scanning confocal microscope Zeiss LSM 710 FCS, equipped with a Confocor 3 module  (FCCS detector head and APD detectors). Normally, the excitation laser beams were focused on 20 l of the sample, placed on the well of an in-house made silicon plate mounted on an optically clear coverslip. Emitted light from the 458-nm and 561-nm lasers passed through an adjustable pinhole and a 458/ 561 main beam splitter and was separated on a 565-nm diachronic mirror. The 470 -540-nm spectra segment was used to quantify fluorescence from CFP, whereas a long pass 580-nm filter was set for measuring the tetramethylrhodamine dye signal. The laser power was set up to achieve a count rate (CPM) that was not giving detectable cross-talk. Typically, CPM was kept at 8 to 12 kHz. The data acquisition time was at least 100 s and included 10 runs of 10 s each. The collected data were analyzed using the ZEN 2011 Zeiss software package. The theoretical basics of autocorrelation and cross-correlation analysis are described everywhere (32). Experimental autocorrelation functions were fitted in a three-dimensional diffusion model (33) for a 2-component model. The K d for the interaction of geranylgeranylated CFP-RhoA and tetramethylrhodamine-labeled RhoGDI (RhoGDI-TMR) was quantified based on the assumption that the amplitude of the cross-correlation function (CCF) corresponds linearly to the concentration of the CFP-RhoA-GG⅐RhoGDI-TMR complex. The CCF values obtained by titration of the 50 nM CFP-RhoA-GG⅐RhoGDI-TMR solution with farnesylated RhoA were fitted to the 3-component competitive model in Dynafit 4.0 (Biokin Ltd.) (34). are the sum of free and bound species. Under these conditions, the fluorescence is described as,

Fluorescence Cross-correlation Measurements-
where F is the measured fluorescence intensity and F max and F min refer to the maximal and minimal of its value. The parameters [P 0 ], F max , F min , and K d were allowed to vary during the fit. The nonlinear least-square regression was performed in Grafit 5.0 (Erithacus software). Global Fitting Analysis of Competitive Titrations-Competitive titration was performed by titrating the mixture of fluorescent RhoA-GNBD and prenylated RhoA⅐GGTase-I complex with increasing concentrations of RhoGDI. The global fitting procedure of the obtained data were performed in Dynafit 4.0 (Biokin Ltd.) (34).
Stopped-flow Measurements-The kinetics of the prenylated RhoA interaction with RhoGDI was analyzed using an SF-61MX stopped flow apparatus (High Tech Scientific).
Measurements were carried out at 25°C in degassed buffer containing 25 mM Hepes-NaOH, pH 7.2, 40 mM NaCl, 2 mM MgCl 2 , and 5 mM dithioerythritol. Fluorescence of the NBD group was excited at 479 nm, and the data were collected with a 515-nm cut-off filter. Transient kinetics of the RhoGDI interaction with RhoA bound to mant-labeled guanine nucleotides was measured under the same conditions, except that fluorescence of the mant group was excited via FRET from tryptophan at 289 nm and detected through a 395-nm cut-off filter. To improve the signal to noise ratio, data from 5 to 10 measurements were averaged and used for further analysis. Data collection and primary analysis of the rate constants were performed with the High Tech Scientific package, and the secondary analysis was performed using Grafit 5.0 (Erithacus software).
Isothermal Titration Calorimetry Measurements-Binding affinities of the unprenylated GDP-and GMPPNP-bound forms of RhoA to RhoGDI were determined using ITC200 (MicroCal). All proteins were kept in buffer containing 25 mM Hepes-NaOH, pH 7.2, 40 mM NaCl, 2 mM MgCl 2 , and 1 mM DTT. The concentration of RhoA protein in the syringe was at least 10-fold higher than the concentration of protein in the cell (typically 10 M). In control experiments, the respective protein was injected from the syringe into the buffer solution, and the recorded background signals were subtracted from the titration experiment data. The titration experiments were carried out at 25°C in triplicate. The data obtained were fitted using the MicroCal-ITC implementation of the Origin 7 software package.
Crystallization of RhoA⅐GMPPNP-GG⅐RhoGDI Complex, Data Collection and Structure Determination-Initial crystallization conditions were determined at room temperature using the PEG Ion kit from Hampton Research in 200-nl sitting drops, set up against a 50-l reservoir with a Mosquito nanoliter dispensing robot (TTP LabTech). Promising conditions were optimized with respect to precipitant composition, pH, temperature, and protein concentration and were transferred to hanging drops prepared by mixing 1 l of protein complex solution with 1 l of precipitant mixture. Crystals for data collection were obtained at 20°C with a reservoir consisting of 20% (w/v) PEG 3350, 0.2 M MgCl 2 , and 0.1 M Hepes-NaOH, pH 7.4. The protein complex was used at 13 mg/ml in buffer containing 25 mM Hepes-NaOH, pH 7.2, 40 mM NaCl, 2 mM MgCl 2 , 10 M GMPPNP, and 1 mM Tris(2-carboxyethyl)phosphine.
Prior to flash-cooling in liquid nitrogen, the crystals were washed briefly in 25% (w/v) PEG 3350, 5% (v/v) glycerol, 0.2 M MgCl 2 , and 0.1 M Hepes, pH 7.4. Diffraction data were collected at 100 K at station MX2 of the Australia Synchrotron (Melbourne, Australia). The data were processed with XDS (55). The crystals belonged to the orthorhombic space group P2 1 2 1 2 1 and contained 1 RhoA-GG⅐RhoGDI complex in the asymmetric unit.
Initial phases were determined by molecular replacement with PHASER (56) of the CCP4 suite, using coordinates of Cdc42⅐GDP-GG⅐RhoGDI (PDB code 1DOA) from which the nucleotide, Mg 2ϩ , and prenylated C terminus of Cdc24 had been deleted. The model was then corrected by alternating rounds of refinement in REFMAC5 (57) and manual adjustment in COOT (58). Restraint libraries were generated with PRODRG (59). Interface analysis was performed with PISA at the EBI website (35). Structure figures were generated using PyMOL.

Interaction Analysis of RhoGDI with Unprenylated RhoA-
To gain insight into the mechanism of the interaction of RhoGDI with RhoA GTPase and the role of the nucleotidebound state in this process, we assessed the thermodynamics of this interaction. We measured the affinity of the interaction between RhoGDI and recombinant unprenylated RhoA, loaded with GDP-or the unhydrolyzable GTP analog GMPPNP, using isothermal titration calorimetry (ITC). We found that RhoGDI forms a complex with GDP-and GMPPNP-bound RhoA with K d values of 0.17 and 2.5 M, respectively (Fig. 1, A and B). To confirm the obtained affinity values using an independent method, we decided to characterize the kinetics of this interaction. We loaded RhoA molecules with the fluorescent nucleotide mant-GDP or mant-GMPPNP, a nonhydrolyzable fluorescent analog of GTP. Rapid mixing of the mant nucleotide-loaded RhoA with RhoGDI in a stopped-flow apparatus resulted in time-dependent fluorescence signal changes. The plots in Fig.  1, C and E, represent a typical fit to the experimental data. The insets show the dependence of the pseudo-first order rate constants on the concentration of RhoGDI. Although the data give a good estimate of the k on values, the k off cannot be extracted from the intercept with y axis reliably, due to its low value in the case of GDP-bound RhoA and low signal amplitudes in the case of the GMPPNP-bound form. Therefore, we determined k off values experimentally by displacing mant nucleotide-labeled RhoA from RhoA⅐RhoGDI complexes with an excess of nonlabeled GDP-bound RhoA (Fig. 1, D and F). Analysis of the kinetic parameters of the RhoGDI interaction with both nucleotidebound forms of RhoA (Table 1) shows that the affinity differences stem from a 15-fold faster off-rate of the RhoGDI complex with the activated RhoA. The affinities that were calculated from the kinetic constants were 0.2 M for GDPbound and 2.3 M for GMPPNP-bound RhoA and are in good agreement with the K d values obtained in the ITC measurements.
To assess the role of the RhoA C terminus in RhoGDI⅐Rho complex assembly, we prepared C-terminal truncated ⌬9 and ⌬13 RhoA mutants that lacked the C-terminal stretch lysines or in addition arginine residues, respectively ( Fig. 2A). Interaction of the truncation mutants with RhoGDI was assessed by ITC as described for the wild-type protein.
The titration experiments showed that deletion of 9 C-terminal residues led to a 15-fold decrease in the affinity of the RhoA/RhoGDI interaction (Fig. 2, B and C). Further deletion of C-terminal residues did not sig-

TABLE 1 Summary of thermodynamic and kinetic parameters of the RhoGDI interaction with GDP-and GMPPNP-associated RhoA
nificantly decrease the affinity (supplemental Fig. S1, A and B). These data suggest that similar to the Rab GTPases, the C terminus of the RhoA significantly contributes to the affinity of the RhoGDI⅐Rho complex (36). The presence of a lysine-and arginine-rich patch in the C terminus of other Rho GTPases suggests that this is a generic feature that may be important for RhoGDI-mediated membrane recycling of Rho GTPases, because it is also involved in the interaction with negatively charged lipids (22,37). Fluorescent Prenylated RhoA as a Sensor of RhoGDI/Rho Interaction-To elucidate the mechanism underlying RhoGDImediated delivery and extraction of Rho GTPases, it is important to characterize RhoGDI interactions with prenylated Rho proteins. Hence, we sought to develop a fluorescent sensor that would report on the interaction of RhoGDI with prenylated RhoGTPases. This is a challenging task, because prenylated proteins display low water solubility, which largely precludes their direct biophysical analysis. All previous experiments relied on detergent-solubilized prenylated proteins, a situation that is known to strongly change the properties of prenylated proteins (41,42). To alleviate this problem, we took advantage of synthetic phosphoisoprenoids modified with the fluorescent NBD group. These analogs of FPP and GGPP can transfer fluorescent lipid moieties onto proteins by protein prenyltransferases. These are less hydrophobic than the native lipids due to the shorter isoprenoid chain (28) (Fig. 3A). We enzymatically prenylated the RhoA protein using recombinant GGTase-I and NBD-geranyl pyrophosphate as a lipid donor. The reaction mixture was resolved on a size exclusion column, and the fluorescently prenylated RhoA was eluted at a position corresponding to a 20-kDa monomer. The protein was homogeneous (Fig.  3B) and remained soluble in the absence of detergent, even at concentrations above 10 mg/ml. ESI-MS analysis of the fractions revealed that the majority of protein was modified with geranyl-NBD (Fig. 3C). Purified RhoA-GNBD was also correctly folded, because it was able to form a stoichiometric complex with RhoGDI, as determined by gel filtration analysis (supplemental Fig. S2A).
Addition of saturating concentrations of RhoGDI to a solution of RhoA-GNBD resulted in an ϳ1.5-fold decrease in NBD fluorescence that could be used for titration experiments (Fig.  3D). Fitting the data to a quadratic equation led to a K d value of 2 nM (Fig. 3E). To confirm this affinity estimate by an independent method and gain insight into the kinetics of the interaction, we rapidly mixed both proteins in a stopped-flow apparatus. The observed trace could be fitted using a single exponential function (Fig. 3F). A linear fit of the observed first-order association rate constants at a range of concentrations allowed us to determine a k on value of 10.7 ϫ 10 6 M Ϫ1 s Ϫ1 for this reaction. To obtain the k off value for this interaction, we mixed 100 nM  Fig. S2B). From the obtained kinetic parameters, we calculated a K d value of 5.6 nM, which is in good agreement with the equilibrium titration experiments. These data demonstrate that RhoA-GNBD binds to RhoGDI with an affinity higher than that corresponding to K d of 7, 30, or 180 nM, respectively, estimated for RhoGDI interactions with the native geranylgeranylated Rho protein Cdc42 (15,16,38). Taken together, these results suggest that NBD-G mimics the native isoprenoid and that the developed sensor can be used to analyze RhoA/RhoGDI interactions.
Interaction of Farnesylated RhoA with RhoGDI-Several Rho proteins can exist in both farnesylated and geranylgeranylated forms in vivo (39 -41, 45, 46). Thus, it is expected that RhoGDI will govern the interaction of both prenylated forms with membranes, but the biophysical parameters may differ significantly. To analyze the interaction of farnesylated RhoA with RhoGDI, we generated the RhoA L193A mutant, which is a more efficient FTase substrate than wild-type of RhoA. The protein was in vitro farnesylated by recombinant mammalian FTase and purified by gel filtration (supplemental Fig. S3). As expected, farnesylated RhoA L193A (RhoA-F) was soluble at concentrations above 1 mg/ml. To determine the affinity of RhoA-F for RhoGDI, we titrated 100 nM RhoA-GNBD with increasing concentrations of RhoGDI in the presence of 50, 100, 200, or 400 nM GDP bound RhoA-F (Fig. 4A). The titration curves showed an initial "lag" that was most prominent at high concentrations of RhoA-F. This suggests that RhoGDI preferentially forms a complex with the fluorescently silent RhoA-F. Data analysis was performed via least square fitting using a 3-component competitive model (42). The fit to the data shown in Fig. 4A yielded a K d value of 2.5 nM for the RhoGDI interaction with farnesylated RhoA. These data suggest that farnesylated RhoA has the capacity to form a tight complex with RhoGDI in vivo.
Preparation of Solution-stabilized Geranylgeranylated RhoA and Analysis of Its Interaction with RhoGDI-The data described above demonstrate that prenylation of RhoA significantly increases its affinity for RhoGDI. However, so far, it is unclear to what extent an increase in the length of the isoprenoid chain would influence the strengths of this interaction. This is particularly important because Rho proteins are predominantly geranylgeranylated. In contrast to RhoA-GNBD and farnesylated RhoA, geranylgeranylated RhoA is insoluble in water and rapidly aggregates. To overcome this problem, we took advantage of the observation that prenyltransferases form stable complexes with their reaction products in the absence of and excess of lipid donor (43). Hence, we geranylgeranylated RhoA in vitro with recombinant GGTase-I using stoichiometric amounts of both proteins and purified the resulting complex by size exclusion chromatography. The proteins coeluted as a stoichiometric complex at a position corresponding to a molecular mass of 160 kDa (Fig. 4B). MALDI-MS analysis demonstrated that the majority of RhoA in the complex was geranylgeranylated (Fig. 4C). The availability of solubilized RhoA-GG enabled us to measure the affinity of the RhoGDI interaction with RhoA-GG using RhoA-GNBD as a fluorescent reporter. In this experiment, we titrated RhoGDI to a mixture of 50 nM RhoA-GNBD with 100, 200, or 400 nM of the RhoA-GG⅐GGTase-I complex. As depicted in Fig. 4D, the titration resulted in an initial increase of fluorescence, followed by its quenching. This can be interpreted as the initial displacement of GGTase-I from the RhoA-GG complex by RhoGDI and the association of GGTase-I with RhoA-GNBD. At higher concentrations, however, all RhoA-GNBD became sequestered by RhoGDI, leading to a signal decrease. We confirmed this by analyzing the interaction of GGTase-I with RhoA-GNBD (supplemental Fig. S4). The availability of the affinity values for the RhoGDI and GGTase-I interactions with RhoA-GNBD and the corresponding fluorescence yields prompted us to attempt to extract the affinity for the RhoA-GG interaction with RhoGDI and GGTase-I from these data. This is a challenging task because it requires a quantitative description of a 4-component system with 2 sets of competitive interactions, i.e. RhoA-GNBD ϩ RhoGDI N RhoA-GNBD⅐RhoGDI and RhoA-GNBD ϩ GGTase I N RhoA-GNBD⅐GGTase I, and RhoA-GG ϩ RhoGDI N RhoA-GG⅐RhoGDI and RhoA-GG ϩ GGTaseI N RhoA-GG⅐GGTase I, related to each other in a nonlinear fashion. Implementing a recursive least square minimization in Dynafit (Biokin Ltd.) (34) allowed us to calculate the K d values for the interactions of native prenylated RhoA with RhoGDI and GGTase I. The fit- The affinity for the RhoA-GG/RhoGDI interaction was unexpectedly high, particularly in the view that RabGDI interacts with GDP-bound diprenylated Rab proteins with an affinity nearly 500 times lower (9). Thus, we sought to confirm the RhoA-GG/RhoGDI dissociation constant value by an independent and more direct method. We decided to use fluorescence cross-correlation spectroscopy (FCCS), because it allows direct monitoring of interacting molecules at very low concentrations. To this end, we genetically fused the RhoA N terminus with CFP and produced the proteins in pure recombinant form. We then took advantage of the fact that RhoGDI contains only one solution-exposed cysteine residue (Cys-79) that is located on the loop after ␤-strand A and that is not involved in interaction with Rho GTPases. Thus modification of this residue does not affect its association with RhoA (38). We chemically labeled RhoGDI with tetramethylrhodamine-maleimide (RhoGDI-TMR). The complex of RhoGDI-TMR with geranylgeranylated GDP-bound CFP-RhoA was formed by mixing both components and purifying the resulting complex by gel filtration (supplemental Fig. S5). As a measure of CFP-RhoA-GG⅐RhoGDI-TMR complex concentration in the FCCS analysis, we used the cross-correlation function (CCF) (44). We observed a decrease in the CCF value upon addition of farnesylated RhoA, which indicates competitive formation of the RhoGDI⅐RhoA-F complex. Fitting the competitive titration curve of the CFP-RhoA-GG⅐RhoGDI-TMR fluorescent complex with RhoA-F using a 3-component competitive model led to a K d value of 20 pM (Fig.  5A). This provides direct and independent confirmation of the low picomolar affinity of RhoGDI to geranylgeranylated RhoA.
Interaction of Activated Geranylgeranylated RhoA with RhoGDI-The current model of GTPase action postulates that GTPases are activated upon membrane delivery and recycled to the cytosol following completion of the functional cycle, marked by GTP hydrolysis. However, several reports suggest that RhoGDI can also efficiently interact with GTP-loaded Rho proteins (10,11,15,16) and promote their release from the intracellular membranes (45,46). We therefore measured the affinity of RhoGDI for the activated form of prenylated RhoA. We prepared the GMPPNP-bound form of geranylgeranylated RhoA using the approach described for the production of the GDP-bound form of RhoA-GG. We then titrated RhoGDI to a mixture of 25 nM RhoA-GNBD with 100, 200, 400, or 600 nM of the RhoA⅐GMPPNP-GG⅐RhoGDI complex (Fig. 5B). Global fit of the titration data resulted in a K d value of 3 nM for the RhoA/ GMPPNP-GG/RhoGDI interaction, whereas the affinity of RhoA-GG interaction with GGTase-I was slightly lower for the GMPPNP-bound form of RhoA, with a K d value of 12 nM. Thus, RhoGDI displays a 500-fold difference in affinity between the GDP-and GTP-bound forms of geranylgeranylated RhoA. Our data demonstrate that similar to RabGTPases, the nucleotidebound state of RhoA dramatically affects its affinity for its GDI. Although the discrimination factor is somewhat less than in the Rab/RabGDI interaction (about 2 ϫ 10 3 ), it is still large and likely to play an important biological role. The fact that affinities for both forms are so high provides an explanation for previous confusion in the field, because most of the biochemical methods of protein interaction analysis cannot distinguish between low nanomolar and picomolar affinities.
Analysis of Dissociation Rates of RhoGDI⅐RhoA-GG Complexes-The postulated biological role of RhoGDI is to sequester the Rho proteins in the cytosol and to transport them to a target membrane. Although the exact time scale of this process is unknown, it is expected to be on the order of minutes (47). This would require Rho-GG⅐RhoGDI complexes to have reasonably fast intrinsic or induced off-rates (10 Ϫ2 to 10 Ϫ3 s Ϫ1 ). This notion promoted us to determine the dissociation rates of the RhoA-GG⅐RhoGDI complex experimentally, because the determined low picomolar affinity for RhoA⅐GDP containingand low-nanomolar affinity for RhoA⅐GMPPNP-containing complexes is most likely to reflect their slow off-rate. Unfortunately, the developed 4-component system, based on RhoA-GNBD, is not suitable for kinetic measurements due to a large number of unknown kinetic parameters. Thus, we developed an alternative sensor in which the fluorescence resonance energy transfer occurs from a CFP domain that is fused to the N terminus of RhoA to the citrine domain that is fused C-terminal to RhoGDI (supplemental Fig. S6A). ITC analysis showed that the unprenylated CFP-RhoA⅐RhoGDI-citrine complex has an affinity only 3 times lower than the native RhoA⅐RhoGDI complex (supplemental Fig. S6B). Moreover, a competitive titration of the CFP-RhoA-GG⅐RhoGDI-citrine complex with farnesylated RhoA resulted in a K d value of 16 Ϯ 5 pM (Fig. 5C, inset), indicating that it behaves similarly to the native complex. The dissociation rate was measured by mixing 30 nM of complexes containing either GDP-or GMPPNP-associated RhoA with 5 M farnesylated RhoA and monitoring the change in FRET effi-ciency ( Fig. 5C and supplemental Fig. S6, E and F). As can be seen in Fig. 5C, both the CFP-RhoA⅐GDP-GG⅐RhoGDI-citrine and CFP-RhoA⅐GMPPNP-GG⅐RhoGDI-citrine complexes are very stable, but the addition of excess of RhoA-F led to their dissociation, which is reflected in the decay of FRET efficiency. Remarkably, the extremely slow dissociation rate of the CFP-RhoA⅐GDP-containing complex (2 ϫ 10 Ϫ4 s Ϫ1 ) is increased by a factor of only 25 (Table 1), meaning that the half-life of the CFP-RhoA⅐GMPPNP-GG⅐RhoGDI-citrine complex is 2.5 min. This relatively modest decrease in the half-life of the complex despite almost 3 orders of magnitude change in the affinity on replacing GDP by GMPPNP means that the association rate constants for the complexes also differ by a factor of about 25 ( Table 1). The rate constant for the dissociation of CFP-RhoA⅐GDP-containing complex (2 ϫ 10 Ϫ4 s Ϫ1 ) means that 50% dissociation will take about 1 h, which is probably too slow to allow the complex to operate in cellular systems that require responses on a time scale of minutes.
Crystal Structure of the RhoA⅐GMPPNP-GG⅐RhoGDI Complex-To gain insight into the molecular basis of RhoA⅐GTP recognition by RhoGDI, we attempted a structural analysis of the activated form of the complex. Using the in vitro prenylation approach described above, we generated preparative amounts of the RhoA⅐GMPPNP-GG⅐RhoGDI complex and subjected it to crystallization trials and diffraction testing, as described under "Materials and Methods." The best diffracting crystal was used to collect the dataset to 2.8 Å using a synchrotron beam source. The structure was solved by molecular c R free is the R-factor calculated using 5% of the data that were excluded from the refinement. d Ramachandran core refers to the most favored regions in the /-Ramachandran plot, as defined by Laskowski et al. (61).
replacement using the Cdc42⅐GDP-GG⅐RhoGDI structure as a search model (48). Full data collection and refinement statistics are given in Table 2. The overall structure of the complex is shown in Fig. 6A. The structure is remarkably similar to the structure of the Cdc42⅐GDP-GG⅐RhoGDI complex (PDB 1DOA), with an overall root mean square deviation on the ␣-chain of 0.87 Å. Total interface area in the RhoA⅐GMPPNP-GG⅐RhoGDI complex excluding geranylgeranyl⅐RhoGDI hydrophobic pocket interface is somewhat smaller than in the structure of Cdc42⅐ GDP-GG⅐RhoGDI complex (1340 versus 1580 Å 2 ) with a significantly lower number of hydrogen bonds and salt bridges forming the RhoA⅐GMPPNP-GG⅐RhoGDI complex. These differences predominantly come from extensive interaction of RhoGDI with residues Arg-186 and Arg-187 of Cdc42. The C terminus of RhoA does not have a diarginine patch, instead there are 3 lysine residues that do not form hydrogen bonds with RhoGDI and are solely involved in the electrostatic interaction.
The position of geranylgeranyl moiety buried into the hydrophobic pocket on the RhoGDI is nearly identical to what is identified on the structure of Cdc42-GG⅐GDP⅐RhoGDI. The minor differences such as involvement of RhoGDI Leu-170 in interaction with the RhoA geranylgeranyl group are unlikely to be functionally significant. Despite the differences in the interface of RhoGDI complexes with RhoA and Cdc42 the total free binding energy calculated by PISA (35) is similar for both complexes: 25.6 kJ/M for RhoA⅐RhoGDI and 27.4 kJ/M for Cdc42⅐RhoGDI.
The Rho-specific helix in the RhoA⅐GDP-GG⅐RhoGDI structure has different conformations than both in the free form of Cdc42 and in the Cdc42 complex with RhoGDI. In contrast to the Cdc42 the Rho insert in the structures of free RhoA and in complex with RhoGDI has an almost identical position. This might reflect differences in the signaling modes of the RhoA and Cdc42 GTPases in the cell (49).
Strikingly, the switch I and II loops of RhoA⅐GMPPNP adopt a nearly identical conformation to those of Cdc42⅐GDP in complex with RhoGDI. As can be seen in Fig. 6C, the conformations of both switch I and switch II regions in the complex differ markedly from the conformations observed in unliganded RhoA⅐GMPPNP. Although switch I adopts a conformation nearly identical to that of GDP-bound RhoA, the switch II region adopts a conformation distinct from both free forms, due to the extensive contacts formed with the N terminus of RhoGDI. The phosphate groups of the bound nucleotide are shifted slightly toward the exterior of the protein, possibly reflecting the fit induced by the complex formation. Overall, the guanine nucleotide in the structure of RhoA⅐GMPPNP-GG⅐RhoGDI has a 15% smaller interaction interface and less number of hydrogen bonds than in the structure of RhoA G14V ⅐GTP␥S alone (50). For instance, there is no interaction between phosphate oxygen atoms and Ala-15 from the P-loop, switch I residues Tyr-34 and Thr-37 are no longer involved in nucleotide ␥-phosphate positioning. Interestingly, important for GTP hydrolysis, the Gln-63 residue in the RhoA⅐GMPPNP-GG⅐RhoGDI structure is located closer than in RhoA G14V ⅐GTP␥S, which results in hydrogen bond formation between its amide group and the oxygen of ␥-phosphate with an overall distance of 3.84 Å.
The observed structure demonstrates that even in the GTPbound form, the loop regions of RhoGTPases are sufficiently flexible to adopt an inactive conformation upon binding to RhoGDI. The resulting conformational strain can be compensated by the very high binding energy of the complex formation and results in a complex that is structurally identical to the inactive conformation but displays a faster dissociation rate.

DISCUSSION
In the present study, we have performed a comprehensive analysis of the interaction of RhoGDI with prenylated and unprenylated forms of RhoA using kinetic and thermodynamic measurements. We have focused on the role of prenylation and the nucleotide-bound state on complex formation and its implications for recycling RhoGTPases.
We constructed a sensor of the RhoGDI/RhoGTPase interaction by in vitro prenylation of RhoA with the NBD-deriva-tized fluorescent isoprenoid. The developed enzymatic route is significantly simpler than the previously described semisynthesis-based approach for construction of fluorescent lipidated proteins (51) and can be easily extended to other GTPases. Using the developed sensor, we were able to determine the affinities of geranylgeranylated and farnesylated RhoA for RhoGDI. The affinities for RhoA-GG/RhoGDI interactions were much higher than previously reported (16), with RhoGDI binding to GDP-bound geranylgeranylated RhoA with a K d value of 5 pM. We demonstrated that the length of the isoprenoid chain has a dramatic impact on the RhoA interaction with RhoGDI. The affinity difference of nearly 3 orders of magnitude between farnesylated and geranylgeranylated complexes indicates that farnesylated and geranylgeranylated Rho proteins will be regulated quite differently by RhoGDI in vivo. Clearly, in the absence of a large pool of free RhoGDI, farnesylated Rhos will be outcompeted by the geranylgeranylated form and therefore are likely to distribute through alternative mechanisms. We also demonstrate that activation of RhoA results in a 500fold decrease in the affinity of the complex. The high affinity of RhoGDI complexes with both GDP-and GMPPNP-bound prenylated RhoA molecules implies their slow dissociation rates, which we found to be 10 Ϫ4 and 10 Ϫ2 s Ϫ1 , respectively. These results have revealed that the RhoA⅐GDP-GG⅐RhoGDI complex is very tight and has a half-life of ϳ60 min. In contrast, the half-life of the GMPPNP-loaded RhoA-GG⅐RhoGDI complex is only 2.5 min, which fits better into the physiological context (47) and suggests the presence of two types of RhoGDI⅐RhoA complexes in vivo with dramatically different half-lives. These findings further support the idea that RhoGDI is actively involved in the recycling and distribution of activated RhoGTPases in the cell (52,53). To gain insight into the mechanism underlying the ability of RhoGDI to stably associate with the activated form of RhoA, we solved the structure of the activated form of the complex and demonstrated that complex formation induces adoption of the inactive conformation without nucleotide hydrolysis. Formation of an extensive set of contacts between RhoGDI and switch II partially through hydrophobic interaction with Leu-67 and Leu-70 of RhoA stabilizes the loop and, importantly, secures the essential GTP-hydrolysis Gln-63 residue, which results in inhibition of GTPase hydrolytic activity (11). The nucleotide independence of the conformation of switch II, which forms the second largest interface in the RhoA⅐GMPPNP-GG⅐RhoGDI complex, most likely promotes association of the proteins, which involves energetically unfavorable changes in the switch I region. Notably, the role of switch II in the recognition of both nucleotide-bound forms of Rho GTPases by RhoGDI was suggested earlier (48). The structural transition of the switch I loop occurs at the cost of interaction the free energy, which is reflected by a difference of almost 3 orders of magnitude in affinities for RhoA-GG⅐ RhoGDI complexes (Table 1). The association of RhoGDI with activated RhoA in solution is 25 times slower than with inactive RhoA, because it requires a structural transformation of the latter molecule. Additionally, the dissociation rate of such a complex is faster by almost the same factor, due to the strain in switches I and II. The ITC data, furthermore, demonstrate that the binding enthalpy of RhoGDI to unprenylated RhoA⅐GMPPNP is 2-fold lower than the enthalpy of the RhoGDI interaction with the GDP-bound RhoA protein. Comparison of the structures of the Cdc42⅐GDP-GG⅐RhoGDI and RhoA⅐GMPPNP-GG⅐RhoGDI complexes indicates the lack of major changes in the interaction surface area. Thus, change in the binding enthalpy most likely reflects the free energy cost of the conformational change in the switch I loop that is required for RhoA⅐GMPPNP⅐RhoGDI complex formation.
The results obtained in this work, in combination with data from previous biochemical and crystallographic studies, allow us to formulate a detailed mechanistic model of the RhoGDImediated recycling of prenylated RhoGTPases in vivo. Our analysis of the interaction of unprenylated RhoA with RhoGDI provides insight into the initial stages of RhoA extraction from the membranes, because the isoprenoid moiety is embedded into the membrane and unlikely to contribute to the interaction at this stage. It is likely that RhoGDI initially forms a micromolar-affinity complex with RhoA by recognizing the switch II and I regions but not the C terminus. Next, the positively charged C terminus of RhoA binds to the negatively charged residues in the C-terminal RhoGDI domain. This increases the RhoA⅐RhoGDI complex residence time on the membrane and allows progression to the next step in extraction, in which the isoprenoid group spontaneously flips out of the membrane and is captured by the lipid-binding pocket of RhoGDI. This eventually triggers release of the complex from the membrane (Scheme 1). In general, RhoGDI and RabGDI appear to operate in a similar way in extracting the respective GTPases from membranes, using sequential formation of progressively tighter complexes that function as a thermodynamic trap. However, the quantitative data suggest that Rho and RabGDI operate differently in the sense that the former distributes both nucleotide-bound forms between membranes and cytosol, whereas the latter acts only on inactive conformations, albeit with quantitative differences. Given the exceptionally high affinities of RhoGDI complexes with prenylated RhoA of both nucleotidebound forms, the local concentration of RhoGDI in the cell will govern the extraction process of GDP-and GTP-bound Rho proteins. Low and limiting RhoGDI concentrations will mostly lead to the extraction of GDP-associated Rho proteins that will outcompete short-lived and weak RhoA⅐GTP⅐RhoGDI complexes. In contrast, a local abundance of RhoGDI will result in extraction of both inactive and activated RhoGTPases. The transient RhoGDI association with GTP-bound Rho molecules will favor its local distribution. The presence of slow and fast cycling pools of the RhoGTPase cycle in the cell was proposed earlier (27,53,54). The mechanism of dissociation of these complexes is unclear at present and, according to our data, cannot be based solely on nucleotide exchange. Therefore, unlike the case of RabGTPases, where GEFs have been suggested to be sufficient to drive targeted membrane insertion of Rabs, it seems likely that dissociation of RhoGDI⅐RhoGTPase complexes must be further accelerated by other factors. Recently demonstrated Arl-GTP-mediated displacement of farnesylated Rheb from its GDI-like chaperon PDE␦ provides an example of a possible mechanism (60). Alternative factors may include some types of lipids and phosphorylation of both Rho and RhoGDI. Therefore, further research is needed in this area to fully understand the role of RhoGDI in the membrane targeting and recycling of RhoGTPases in the cell.