Structural Basis for the Signaling Specificity of RhoG and Rac1 GTPases*

RhoG is a new GTPase that has high sequence similarity with members of the Rac subfamily (Rac1, Rac2, and Rac3), including the regions involved in effector recognition and binding. To characterize its biological properties, we have compared the activity of RhoG and Rac1 in a number of experimental systems, including the study of their subcellular localization, oncogenic potential, activation of effectors, and effect on F-actin dynamics. Our study indicates that RhoG and Rac1 share overlapping, but not identical, signal transduction pathways. In contrast to previous results, we also provide evidence that RhoG works in parallel to Rac1 rather than as a Rac1 upstream activator. Using an extensive collection of Rho/Rac1 chimeras and point mutants, we demonstrate that the different biological properties of RhoG and Rac1 can be traced to specific amino acid variations in their switch I, β2/β3 hairpin, α5 helix, and C-terminal polybasic regions. Taken collectively, our results highlight the complexity of the signal transduction pathways activated by Rho/Rac GTPases and provide insight into the structural determinants that mediate the differential engagement of biological responses by GTPases of very similar structure.

GTP hydrolases of the Rho/Rac family participate in the generation of coordinated cellular responses to extracellular stimuli (1,2). This GTPase family is composed of numerous members that can be classified according to structural similarity in Rho (RhoA, RhoB, RhoC, RhoD, RhoE, and TTF), Rac (Rac1, Rac2, Rac3, and RhoG), and Cdc42 (Cdc42 and TC10) subfamilies (2). The majority of these proteins are characterized by a common mechanism of activation during signal transduction (2). In non-stimulated cells, Rho/Rac proteins are maintained in an inactive state because of the presence of bound GDP molecules. After cell stimulation, they exchange GDP by GTP, an event inducing structural changes in their switch I and II regions that makes them competent for effector binding. At the end of the stimulation cycle, Rho/Rac GTPases become inactive again because of the hydrolysis of GTP to yield GDP. This cycle is regulated by GEFs 1 and GAPs. GEFs accelerate the activation of the GTPases during cell stimulation by catalyzing the release of the bound GDP. By contrast, GAP catalyze the hydrolysis of GTP into GDP, therefore allowing the rapid inactivation of the GTPases at the end of the stimulation cycle (2). In addition to these regulatory molecules common to all GTP-binding proteins, Rho/Rac proteins utilize regulatory mechanisms that are different from other GTPase families. Thus, the expression of some Rho/Rac family members is inducible during cell stimulation or differentiation (i.e. Rac2 and RhoG) (3,4). Moreover, most Rho/Rac proteins bind to negative regulatory molecules known as Rho GDIs. These proteins regulate the subcellular localization of Rho/Rac GTPases by extracting them from the plasma membrane and keeping them sequestered in the cytosol (2,5,6). In addition, recent data suggest that GDIs may inhibit Rho/Rac signaling by removing effector molecules from the activated GTPase (7).
One key biological function of Rho/Rac family members is the regulation of F-actin polymerization and cytoskeletal dynamics (1,2). For instance, it has been shown that Rac1 promotes the formation of lamellipodia and membrane ruffling, that RhoA induces stress fibers and focal adhesions, and that Cdc42 is in charge of generating microspikes and filopodia in a number of cell types (1). These proteins have also been linked to other cellular processes such as apoptosis/cell survival, immune functions, vesicle trafficking, tissue contractility, cell cycle progression, and transcriptional changes (1,2,8). These activities are the result of the direct activation of different subsets of effector molecules, including adaptor proteins, actin polymerization/ branching agents, and kinases of different substrate specificity (1,2,8). The function of Rho/Rac-dependent pathways have also been associated with a number of human disorders such as cancer, hereditary immunodeficiencies, human immunodeficiency virus replication, X-linked mental retardation, deafness, and other developmental abnormalities (9). These observations have fueled the interest for the functional characterization of all members of this GTPase family.
RhoG is a new member of the Rho/Rac family that is still poorly characterized. This GTPase was initially discovered as a ubiquitously expressed immediate early gene that was induced by a number of extracellular stimuli such as serum, thrombin, and fibroblast growth factor (4). Later on, the activity of this GTPase has been linked to cytoskeletal changes and the abrogation of cell contact inhibition in fibroblasts (10), neurite outgrowth in rat pheochromocytoma cells (11), and axogenesis in primary sympathetic postganglionic neurons (12). More recently, RhoG has been linked to the stimulation of the nuclear factor of activated T-cells (13), a transcriptional factor important for the activation of genes encoding interleukin 2 and other cytokines (14). One interesting property of RhoG is its close structural relationship with Rac1. These two GTPases share a 72% overall sequence identity and, perhaps more importantly, are very similar in the regions involved in effector binding, such as the switch I (92% sequence identity), the ␤2/␤3 hairpin (75% sequence identity), and the switch II domain (89.5% sequence identity). This structural likeness suggests that these two proteins must utilize similar signal transduction pathways. However, one unexpected observation is that RhoG appears to act not in parallel but upstream of Rac1, allowing the activation of this GTPase during specific cellular responses (10,11). A similar relationship has also been reported for RhoG and Cdc42 (10,11). For instance, it has been shown that the constitutively active version of RhoG can induce the formation of membrane ruffles and filopodia that can be blocked in turn by the expression of dominant negative mutants for Rac1 and Cdc42, respectively (10). Moreover, it has been reported that the RhoG-dependent neurite outgrowth can be eliminated by the expression of the same interfering mutants in rat pheochromocytoma cells (11). Although linear pathways of activation between different Rho/Rac GTPases have been previously described (1,2), this is the first example of an activation of one GTPase by one closely related member of the same subfamily.
The structural similarity of RhoG and Rac1 prompted us to investigate in detail their effector specificities and signaling relationships. To this end, we compared the biological properties of these two GTPases in a number of cellular responses, including subcellular localization, activation of signaling pathways, and oncogenesis. Moreover, we have used two independent techniques to verify the upstream position of RhoG over Rac1 in signal transduction. The results obtained indicate that RhoG and Rac1 act in non-linear, parallel pathways that share overlapping, but not identical, signaling elements. Specifically, we found that RhoG and Rac can promote JNK activation and induce similar changes in the F-actin cytoskeleton. However, they differ in subcellular localization, ability to bind PAK1, and transforming activity. To understand in detail the structural basis that allows the engagement of separate signaling elements by these two GTPases, we used an extensive collection of chimeric proteins in which specific fragments of Rac1 were swapped by the equivalent regions of RhoG. In addition, we utilized point mutants that targeted divergent residues between these two GTPase in the regions presumably involved in assuring substrate selectivity. All these mutant proteins were tested in PAK1 binding, cellular transformation, and subcellular localization. Our results indicate that amino acid residues within the switch I, ␤2/␤3 hairpin, ␣5, and hypervariable region of RhoG and Rac1 are involved in the differential engagement of those cellular responses.

EXPERIMENTAL PROCEDURES
Expression Vectors-All RhoG and Rac1 point mutants described in this work were generated using the QuikChange TM mutagenesis kit (Stratagene) according to the manufacturer's instructions. RhoG Q61L / Rac1 chimeric cDNAs were generated by PCR using the Elongase polymerase (Invitrogen) following a previously described protocol (15). The oligonucleotide sequences and protocols for generating each specific chimera are available upon request. The pGEX vector containing the PAK1 CRIB domain was provided by Dr. R. Cerione (Cornell University, Ithaca, NY). The GST-PAK1 CRIB domain was expressed in Escherichia coli according to standard procedures (16).
Tissue Culture Conditions-All cell types were cultured at 37°C and an atmosphere of 5% CO 2 in a Dulbecco's modified Eagle's medium supplemented with 10% calf serum, 1% L-glutamine, and 1% penicillin-streptomycin. All tissue culture reagents were obtained from Invitrogen.
Immunofluorescence-Cells were grown on uncoated glass coverslips introduced into six-well plates (50,000 cells/plate) and transfected with 1 g of the indicated plasmids for 30 h using liposomes (FuGENE 6, Roche Diagnostics), as described previously (18). Cells were then rinsed in phosphate-buffered saline and fixed with 3.7% formaldehyde (Sigma) in phosphate-buffered saline for 15 min, permeabilized by incubation with phosphate-buffered saline containing 0.5% Triton X-100 for 10 min, and blocked for 10 min in 2% bovine serum albumin, 0.1% sodium azide, 0.1% Triton X-100 in 25 mM Tris-buffered solution. Cells were incubated with anti-AU5 antibodies (Babco) (1:1000 dilution) for 1 h and then incubated with a anti-mouse secondary antibody coupled to Cy2 (Jackson ImmunoResearch) (1:200 dilution) for 45 min. For staining the F-actin cytoskeleton, cells were incubated with rhodaminephalloidin (Molecular Probes) for 20 min (18). After three washes in 25 mM Tris-buffered saline solution containing 0.1% Triton X-100, the coverslips were mounted onto microscope slides using a standard mounting medium (Slowfade, Molecular Probes). Antibody dilutions were made in blocking solution. All steps were conducted at room temperature. Fluorescence images were captured with a Zeiss LSM510 confocal microscope. Fluorochromes were excited using an 18 Ar laser (488 nm excitation-wavelength) for Cy2 and a 2 He/ 10 Ne laser (543 nm excitation wavelength) for both Cy3 and rhodamine.
Focus Formation Assays-NIH3T3 cells (150,000 cells/10-cm plate) were transfected with 20 g of high molecular weight calf thymus DNA (Roche Diagnostics) and the indicated plasmids using the calciumphosphate precipitation method (19). After 24 h, DNA/calcium phosphate precipitates were removed by two washes with Dulbecco's modified Eagle's medium containing 5% calf serum and cultured in Dulbecco's modified Eagle's medium supplemented with 10% calf serum for 15 days. After this period, cells were fixed with formaldehyde and stained with Giemsa to count the foci of transformed cells. All transfections were done in duplicate.
GST Pull-down and Kinase Assays-COS1 cells growing in 10-cm dishes were transfected with 5 g of the indicated constructs using the DEAE-dextran method (20). 48 h after transfection, cells were washed on ice-cold phosphate-buffered saline solution and then disrupted in a lysis buffer containing 20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 5 mM Mg 2 Cl, 0.5% Triton X-100, 5 mM ␤-glycerophosphate, 1 mM dithiothreitol, a protease-inhibitor mixture (Complete, Roche Molecular Biochemicals), and 10 g of the GST-PAK1 CRIB domain fusion protein. Cell lysates were precleared by centrifugation at 14,000 rpm for 10 min at 4°C and then incubated with glutathione-Sepharose beads (Amersham Biosciences) for 2 h at 4°C. Beads were washed three times in lysis buffer without GST-PAK1 CRIB domain and boiled in SDS-PAGE sample buffer to release the bound proteins. Eluates were separated by electrophoresis, transferred to nitrocellulose filters, and subjected to immunoblot analysis using a monoclonal antibody to the AU5 epitope (Babco). Immunocomplexes and pulled down proteins were visualized by chemiluminescence techniques using a commercial kit (ECL, Amersham Biosciences). JNK immunocomplex assays were done as described (20), using a GST-ATF2 fusion protein as phosphate acceptor. Hemagglutinin antibodies used in the immunocomplex assays were from Babco.
Informatics-Confocal image analysis was carried out with the LSM 5 Image Browser program (version 2.8). Protein tertiary structures were elaborated using the Cn3D 4.0 program.

RhoG and Rac1
Have Overlapping, but Not Identical, Signaling Pathways-To compare the signaling specificities of RhoG and Rac1, we studied their respective activities in four different biological assays. First, we analyzed their subcellular localization and their effects on the F-actin cytoskeleton in COS1 and NIH3T3 cells. As previously described (21), the constitutively active version of Rac1 (Q61L mutant) induces robust membrane ruffling in COS1 cells, showing a preferential subcellular localization in plasma membrane and membrane ruffles (Fig. 1A). A small fraction of Rac1 Q61L was also detected in the cytosol (Fig. 1A). Wild type Rac1 induced no detectable morphological change and displayed a uniform distribution in the cytosol (Fig. 1A), a result consistent with its high GDP content and its constitutive association with Rho GDIs (see Supplementary Information, Fig. S1) (2). Wild type RhoG showed identical distribution as that of Rac1 in COS1 cells (Fig. 1A). However, RhoG Q61L was localized in intracellular vesicles and, unlike Rac1 Q61L , was totally absent from the plasma membrane and ruffles (Fig. 1A). The differential localization of RhoG and RhoG Q61L correlated with their differential binding to Rho GDI (see Supplementary Information, Fig. S1), as previously described for other Rho/Rac GTPases (2). The RhoG-containing vesicles are from endocytic origins, because they co-localize with internalized epidermal growth factor receptor. 2 In contrast to previous reports (22), no association of RhoG Q61L could be found with the endoplasmic reticulum or lysosomes. Despite the differential localization of the activated forms of RhoG and Rac1, both induced strong membrane ruffling when expressed in COS1 cells (Fig. 1A). When tested in murine fibroblasts, Rac1 Q61L and RhoG Q61L induced extensive lamellipodia in the cell periphery (Fig. 1B). In the case of Rac1, this phenotype was associated with the formation of a perinuclear ac-tomyosin ring (Fig. 1B). This morphological phenotype was similar to that induced by the Vav oncoprotein (Fig. 1B), a well known Rac1/RhoG activator (23,24). RhoG Q61L -expressing cells lacked this contractile structure and, instead, displayed a continuous membrane ruffle around the cell periphery. In contrast to previous results (10), no filopodia were found in RhoG Q61L -expressing cells (Fig. 1, A and B). For the sake of comparison, we found that RhoA Q63L induced cell rounding and Cdc42 Q61L promoted extensive filopodia formation when expressed in the same cell context (Fig. 1B). These results indicate that, despite their different subcellular localizations, RhoG and Rac1 induce similar morphological phenotypes.
Next, we analyzed the transforming potential of these GTPases when transfected into mouse fibroblasts (NIH3T3 cells). Rac1 Q61L induces cellular transformation in this system (approximately 1.4 -3.0 ϫ 10 3 foci g Ϫ1 of transfected DNA) (Fig. 1C). This transformation is characterized by the generation of foci of transformed cells of very small size (Fig. 1C). In contrast, RhoG Q61L does not induce cell transformation under identical experimental conditions (Fig. 1C). The lack of transforming activity of RhoG Q61L is not because of lack of GDP/GTP cycling, because a fast cycling mutant of RhoG (F28L) also lacked transforming activity (see Supplementary  showed no transforming activity even at saturating concentrations of transfected DNA. 3 Finally, we compared the activity of RhoG and Rac1 toward two specific Rac1 downstream elements, the serine/threonine kinases PAK1 and JNK (2). Because the activation of PAK1 is mediated by the physical interaction with their upstream GTPases, we measured the PAK1 binding activity of Rac1 and RhoG proteins using pull-down experiments with the GST-PAK1 CRIB domain (see "Experimental Procedures"). As shown in Fig. 1D (left panel), RhoG Q61L could not bind to PAK1 in conditions under which the constitutively active version of Rac1 showed a strong association. Similar negative results were obtained with the fast cycling RhoG F28L mutant (see Supplementary Information, Fig. S3). Instead, Rac1 F28L did show efficient binding to PAK1 under the same experimental conditions (Supplemental Materials, Fig. S3). The different binding properties of Rac1 and RhoG proteins toward PAK1 were not because of different expression levels, since all proteins were detected at similar levels in the transfected cells (Fig. 1D, left panel; Supplemental Materials, Fig. S3, lower panel). Finally, we determined whether these GTPases could activate JNK. To this end, we transfected hemagglutinintagged JNK1 in COS1 in the absence or presence of either Rac1 Q61L or RhoG Q61L . As positive control, we included a cotransfection of JNK with the vav oncogene product, a well known activator of this kinase (20,23). The activity of JNK under these conditions was evaluated 24 h post-transfection using in vitro immunocomplex assays. Rac1 Q61L , RhoG Q61L , and the Vav oncoprotein promoted the activation of JNK in these conditions (Fig. 1D, right panel). These results indicate that RhoG and Rac1 induce overlapping, but not identical, signaling pathways and biological responses despite their high structural similarity.
RhoG Is Not an Upstream Activator of Rac1-It has been previously described that RhoG appears to activate Rac1 and Cdc42 (10,11). However, our results showing that RhoG Q61L and RhoG F28L are not transforming argues against such a simplistic interpretation for the RhoG and Rac1 relationship. Indeed, if RhoG were the upstream activator of Rac1, one would expect a transforming activity of this protein identical or higher than that displayed by Rac1 Q61L . Because of this discrepancy, we decided to further evaluate the possible upstream situation of RhoG over Rac1. First, we explored the possibility that RhoG could activate Rac1 or Cdc42 in vivo using focus formation assays. As previously described for Ras GEFs (27), the co-transfection of an upstream activator with the wild type GTPase should result in higher levels of transforming activity. However, the upstream activator should have no effect over the constitutively active form of the GTPase, because this protein is already locked in the GTP-bound, activated state. Using this experimental approach, we observed that RhoG Q61L could not enhance the transforming activity of wild type Rac1 or Rac1 Q61L when co-expressed in rodent fibroblasts ( Fig. 2A, both  panels). In contrast, the latent transforming activity of Rac1 could be increased by the co-expression of a bona fide upstream activator, the vav oncogene product ( Fig. 2A, right panel). RhoG Q61L did enhance the transforming activity of wild type Cdc42, (Fig. 2A, left panel). However, this effect was also observed when wild type Cdc42 was substituted in the experiments by its constitutively active counterpart (Ccd42 Q61L ) ( Fig.  2A, left panel), indicating that this enhanced response is probably a synergistic effect induced by the activation of parallel pathways rather than the direct activation of Cdc42 by RhoG Q61L . In this respect, a similar biological response was found when RhoG Q61L was co-expressed in the same cell system with an unrelated GTPase, the H-Ras oncoprotein (G12V mutant) ( Fig. 2A, left panel). These results indicate that the activated version of RhoG does not seem to act as a bona fide activator of Rac1 and/or Cdc42.
To further confirm these empirical observations, we investigated whether RhoG Q61L could promote the incorporation of GTP into wild type Rac1 in vivo. To this end, we expressed wild type Rac1 in COS1 cells either alone or in the presence of RhoG Q61L or its switch I mutants (F37L and Y40H). As positive control, we included a well known activator of Rac1, the Vav oncoprotein. After the transfection, the levels of GTP-Rac1 were evaluated by GST pull-down experiments using the GST-PAK1 CRIB fusion protein. These experiments revealed that RhoG Q61L was totally inactive as a Rac1 activator under conditions in which Vav promoted effective GTP loading of this GTPase (Fig. 2B). Similar results were obtained when RhoG Q61L was replaced by RhoG F28L in these experiments (see Supplementary Materials, Fig. S4). Collectively, these results indicate that there is no linear signaling relationship between RhoG and Rac1. Instead, these GTPases seem to promote independently similar pathways that share overlapping but not identical signaling elements.
Structural Determinants for the Binding to PAK1: Role of the ␤2 and ␣5 Regions-Because RhoG and Rac1 behave differ-3 R. M. Prieto-Sá nchez and X. R. Bustelo, unpublished data. ently in terms of PAK1 activation, transforming activity, and subcellular localization, we decided to investigate the molecular basis of such signaling specificity. We first focused our attention on the differential binding of the serine/threonine kinase PAK1. To that end, we made two chimeras that combined in the same molecule N-terminal residues of RhoG Q61L (residues 1-85 or 1-120) with the C-terminal domains of Rac1 (Fig. 3A). We hypothesized that if the determinants of the Rac1/PAK1 interaction were present in a discrete area, then its substitution by the same region present in RhoG would lead to the disruption of PAK1 binding. This approach had been useful before to identify the regions that determine the interaction of these GTPases with a number of exchange factors, Rho/Rac GAPs, and effectors (15, 28 -30). Plasmids encoding the two chimeric GTPases were expressed in COS1 cells and the binding activity of their protein products toward PAK1 were determined by pull-down experiments using the CRIB domain of PAK1. As shown in Fig. 3B, upper panel, we found that substitution of the N-terminal regions of Rac1 Q61L by those belong-ing to RhoG Q61L induced the total loss of PAK1 association. As positive control, the PAK1 probe showed optimal binding to Rac1 Q61L and, to a much lower extent, to wild type Rac1 (Fig.  3B, upper panel). As expected, RhoG Q61L did not bind to PAK1 under the same experimental conditions (Fig. 3B, upper panel). All proteins tested in these experiments were expressed efficiently in COS1 cells (Fig. 3B, lower panel). These results indicate that one region that determines the specific binding of PAK1 to Rac1 instead of RhoG appears to be located somewhere between the switch I and II regions of Rac1.
We next sought to identify the amino acids within this region of the GTPases that are used by PAK1 to discriminate Rac1 from RhoG. To aid in the identification of these residues, we used two different criteria: (i) the residue had to be different between RhoG and other PAK1-binding GTPases such as Rac1 and Cdc42; (ii) the candidate amino acid had to be at least partially exposed to the surface of the GTPases, a property indicative of accessibility for the interaction with effectors. Based on these criteria, we found one possible candidate in the Rac1 switch I region (Gly-30), three candidates in the Rac1 ␤2 region (Asn-43, Val-44, and Met-45), and one in the Rac1 ␤3 region (Pro-50) (Fig. 3C, arrows). The residues mentioned above are replaced in RhoG by amino acids Lys, Gln, Ser, Ala, and Thr, respectively (Fig. 3C, arrows). No residue fulfilling these two criteria was found in the switch II region of that GTPase. To check whether these residues were contributing to the binding specificity of PAK1, we generated a Rac1 Q61L mutant in which the aforementioned residues were replaced by those found in the equivalent positions of RhoG. Given the proximity of the residues Val-44 and Met-45, we initially made a Rac1 Q61L double mutant in those positions. Using GST-PAK1 CRIB pull-down experiments, we found that the Rac1 Q61L mutants of the switch I region (G30K mutant) and the ␤3 region (P50T mutant) did not alter the binding of the constitutively active mutant of Rac1 toward PAK1 (Fig. 3D, upper panel on the left). Likewise, one of the Rac1 Q61L ␤2 mutants (N43Q) showed no major effects on such an interaction (Fig. 3D, upper panel on the left). By contrast, the other Rac1 Q61L ␤2 mutant (V44S/M45A) was severely impaired in its binding toward the PAK1 CRIB domain (Fig. 3D, upper panel on the left). We next generated single Rac1 Q61L mutant proteins containing single replacements in either position 44 (V44S mutant) or 45 (M45A mutant) to determine which one of these two residues of Rac1 Q61L was contributing to the specific binding of PAK1. As shown in Fig. 3D (upper panel on the right), only the V44S mutant of the Rac1 Q61L protein disrupted the binding of PAK1. These results indicate that the ␤2 region and, specifically, the valine residue located at position 44, is a major structural determinant for the discrimination of Rac1 over RhoG by the serine/threonine kinase PAK1.
To verify whether this position was sufficient for PAK1 binding, we decided to generate the reciprocal mutation (S44V/ A45M) in RhoG Q61L . We assumed that if this position was the only one important for PAK1 binding, then this RhoG mutant would be able to associate with PAK1. Contrary to this hypothesis, we found that the S44V/A45M mutant of RhoG Q61L could not bind to the PAK1 CRIB domain (Fig. 3D). As positive control, Rac1 Q61L and Rac1 M45A/Q61L showed optimal association to PAK1 under identical experimental conditions (Fig. 3D, upper panel on the right). These results indicate that the presence of the valine residue at position 44 of the ␤2 region is necessary, but not sufficient, to allow the binding of PAK1 to their upstream GTPases. As a control, we corroborated that all the GTPases used in these experiments were expressed at similar levels, using immunoblot analysis with anti-AU5 antibodies (Fig. 3D, lower panels). To rule out the possibility that the Rac1 V44S/Q61L and RhoG S44V/Q61L mutant proteins were incapable of associating with PAK1 because of spurious structural effects induced by the mutations, we also tested all our Rac1 and RhoG mutants in a different biological read-out, the ability to induce F-actin polymerization and membrane ruffling. To this end, COS1 cells expressing transiently these proteins were fixed, stained with both phalloidin-rhodamine (to detect F-actin) and anti-AU5 antibodies (to detect the GTPases), and analyzed by immunofluorescence. All Rac1 and RhoG mutants were similarly active in these two cellular responses (Fig. 4). Moreover, all mutants showed the expected subcellular localization for Rac1 Q61L and RhoG Q61L in membrane areas and intracellular vesicles, respectively (Fig. 4). These results confirmed that the mutations of residues 44 and 45 of Rac1 and RhoG were not deleterious for the overall function of these proteins.
To identify the other region(s) outside the ␤2 region responsible for specific binding of PAK1 toward Rac1, we generated a new series of RhoG/Rac1 chimeric proteins (Fig. 5A). In these constructions, our aim was to replace progressively the Cterminal region of Rac1 Q61L with the same regions present in RhoG (Fig. 5A). In addition, a subset of these chimeric proteins included the S44V/A45M mutation in the RhoG ␤2 region, because these residues were demonstrated to be essential for the binding of PAK1 to Rac1 Q61L (see above, Fig. 3). Pull-down experiments indicated that all RhoG Q61L /Rac1 chimeric proteins containing the S44V/A45V mutation were capable of binding to PAK1, but only when the ␣5 region of Rac1 was present in the chimeric protein (Fig. 5, B and C, upper panels; compare the negative binding of chimera G with the positive binding of chimeras C and F). In good agreement with our previous results (Fig. 3), this binding was strictly dependent on the presence of the RhoG to Rac1 (S44V/M45M) mutation in the ␤2 region of RhoG, because chimeras lacking these amino acid changes could not bind PAK1 even when the Rac1 ␣5 helix was present (Fig. 5, B and C, upper panels, see chimeras B and E). All these RhoG/Rac1 chimeras were expressed at similar levels in vivo, as determined by Western blot analysis using anti-AU5 antibodies (Fig. 5, B and C, lower panels). These chimeras were also active in other biological assays, such as morphological change (see below, Fig. 8A). Taken together, these results indicate that PAK1 utilizes residues present in both the ␤2 and ␣5 regions of the GTPases to distinguish Rac1 from RhoG.
The examination of the ␣5 region of Rho/Rac GTPases under the same criteria used in Fig. 3 suggested the presence of two possible residues (Thr-167 and Asp-170) that could contribute to the specificity of the Rac1/PAK1 physical interaction (Fig.  6A). These two residues are replaced in RhoG by Glu and Ala, respectively (Fig. 6A, arrows). To investigate whether any of these residues was the second structural signal required for the specific interaction of PAK1 with Rac1, we tested whether the Rac1 to RhoG mutants (T167E, D170A, and T167E/ D170A) could bind PAK1 in vitro. We found that while the Rac1 Q61L/T167E mutant retained PAK1 binding activity, the Rac1 Q61L/D170A mutant displayed a 20-fold lower binding affinity toward PAK1 than both Rac1 Q61L and Rac1 Q61L/T167E mutants (Fig. 6B, upper panel on the left). Interestingly, a Rac1 Q61L mutant containing a double mutation in the ␣5 helix (T167E/D170A) partially recovered the binding to PAK1, although it was still significantly less active than the normal protein (Fig. 6B, upper panels). All these mutant proteins were expressed normally in cells, as determined by anti-AU5 immunoblotting experiments (Fig. 6B, lower left panel). In addition, we also found that the Rac1 Q61L ␣5 mutants were still active in F-actin polymerization and membrane ruffling, indicating that these mutations did not impair the overall signaling activity of this protein (Fig. 6C). 3 Interestingly, residue Asp-170 of Rac1 is located from a spatial point of view close to the Val-44 residue, the other position important for PAK1 binding (Fig. 6D). This result indicates that the groove existing between the ␤2/␤3 and ␣5 regions of Rac1 and RhoG is important for their discrimination by PAK1.
To verify whether the amino acids located at positions 44 and 170 were the only ones important for such binding, we generated a final RhoG Q61L mutant in which the serine and Although this mutant recovered a small amount of PAK1 binding with respect to RhoG S44V/A45M/Q61L , its affinity toward PAK1 was still very low compared with Rac1 Q61L or even the Rac1 Q61L/D170A mutant (Fig. 6B, upper panel on the right). All these versions of Rac1 and RhoG were similarly expressed at the protein level (Fig. 6B, lower panel on the  right). This result suggests that despite the importance of Val-44 and Asp-170 for PAK1 binding, there have to be additional residues located within (or downstream) the ␣5 helix that ensure optimal PAK1 binding.
Structural Determinants for the Differential Transforming Activity of Rac1 and RhoG-To achieve this end, we utilized a similar strategy than that used above for the identification of the structural cues that determine the differential binding of PAK1. First, we used focus formation assays in NIH3T3 cells to evaluate the transforming activity of the seven RhoG/Rac1 chimeric proteins used before in our PAK1 binding assays (see Figs. 3A and 5A). As shown in Fig. 7A, no chimeric protein showed transforming activity in this assay under conditions in which Rac1 Q61L induced elevated levels of cell transformation. This indicated that the N-terminal region of Rac1 also had important structural determinants for engaging the signaling pathways that contribute to cell proliferation. To identify them, we then tested the transforming activity of Rac1 Q61L mutants with six Rac1 to RhoG mutations distributed in the switch I, ␤2, and ␤3 regions (G30K, N43Q, V44S/M45A, V44S, M45A, and P50T). In contrast to the results with PAK1 (Fig. 3), we found that the V44S/M45A and the V44S Rac1 mutants did not abolish Rac1 Q61L transforming activity (Fig. 7B). Rac1 N43Q/Q61L and Rac1 Q61L/P50T mutants also showed similar transforming activities as that of Rac1 Q61L (Fig. 7B). In contrast, the transforming activity of the Rac1 G30K/Q61L mutant was 15-fold lower than Rac1 Q61L (Fig. 7B). This result indicates that position Gly-30 of the switch I region of Rac1 represents a key residue for the transforming activity of this protein. Moreover, these observations suggest that the structural elements used to discriminate binding of PAK1 and cellular transforma- tion are different, because they are located in the ␤2 and switch I regions, respectively.
When we did the reciprocal mutation in RhoG Q61L (K30G), we recovered one-third of the levels of transforming activity displayed by Rac1 Q61L , confirming that position 30 of Rho/Rac GTPases is important for this biological response (Fig. 7C). However, this mutant could not recover the full biological activity of Rac1 (Fig. 7C), indicating that other regions were also contributing to the robustness of this signaling response. Because of this, we focused our attention in the ␣5 region, testing the transforming activity of the previously described mutants of Rac1 Q61L (T167E and D170A). As shown in Fig. 7C, both mutants severely affected the transforming activity of Rac1 Q61L . Collectively, these results indicate that the switch I and ␣5 regions are important for triggering cell transformation.
Structural Determinants for the Differential Subcellular Localization of Rac1 and RhoG-As shown in Figs. 1A and 4, the constitutively active versions of Rac1 and RhoG have different subcellular localizations. To identify the regions responsible for this property, we studied the subcellular localization of 4 RhoG Q61L /Rac1 chimeras (B, E, F, and G; Figs. 3A and 5A). Interestingly, we found that all these chimeras showed a Rac1like distribution, with a preferential localization at the plasma membrane (Fig. 8A). In agreement with our previous results (Fig. 4), this localization was independent on the presence of the RhoG to Rac1 V44S substitution in the ␤2 region of RhoG (Fig. 8A). The expression of all these chimeras also induced membrane ruffling in the transfected cells, an indication of the biological activity of all these chimeric proteins (Fig. 8A). These results indicate that the signal for the differential subcellular localization of RhoG and Rac1 is located between residues 180 and 192, the region encompassing the polybasic and the CAAX box. In this case, and unlike the results found for PAK1 binding and cell transformation, the switch I, the ␤2 strand, and the ␣5 region are totally dispensable for the specificity of this biological property unshared by Rac1 and RhoG. DISCUSSION One feature of Rho/Rac proteins is the versatility with which they exploit a similar molecular scaffold to induce specific and, in some instances, quite diverse biological functions. Thus, despite a similar three-dimensional structure, members of this family show distinct subcellular localizations, differential binding to upstream and downstream effectors, distinct affinities toward GAPs, and even different dynamics of their GDP/GTP exchange cycles (1, 2). As a consequence, the biological responses induced by these proteins are almost as diverse as the number of members of the Rho/Rac family. This structural flexibility is well exemplified by the results presented here, because we have observed that two highly related GTPases such as RhoG and Rac1 are still capable of inducing different cellular effects despite their high sequence similarity. Thus, we have shown that these GTPases share some common properties (cytoskeletal change and JNK activation), but differ in other important biological features such as the binding to PAK1, the ability to induce cell transformation, or their specific subcellular localization. This functional disparity is remarkable if we take into consideration that RhoG and Rac1 share almost identical switch I and II regions, the domains involved on the recognition and binding of effector molecules.
One consequence of the functional specialization of Rho/Rac proteins is that, to assemble a coordinated and coherent cellular response, the stimulated cell has to make sure that several GTPases are stimulated simultaneously. For instance, cell motility can only be achieved efficiently if Cdc42, Rac1, and RhoA are activated in a specific time frame to promote the advancement of the leading edge and the retraction of the posterior end of the migrating cell (1). This can be achieved by different means. On one hand, the cell can activate a GEF that can stimulate several members of the Rho/Rac family, thus allowing the simultaneous engagement of their signaling pathways (31). On the other hand, cells have developed signal transduction cascades in which the activation of one particular GTPase translates into the stimulation of other GTPases further downstream. Thus, it has been shown that fibroblasts have a linear pathway in which the activation of Cdc42 leads to the stimulation of Rac1 that, in turn, promotes the activation of RhoA (2). Interestingly, several publications have indicated that a pathway of this type is assembled by RhoG, because the activated version of this protein appears to activate the GTPases Rac1 and Cdc42 (10,11). This model is primarily based on the indirect observation that dominant negative mutants of Rac1 and Cdc42 can block the effects induced by RhoG Q61L in cell morphology and neurite extension (10,11). Although this model is congruent with previous observations with other GT-Pases, there are several observations that speak against it. At the empirical level, our results showing the lack of transforming activity of RhoG argue against the upstream position of this GTPase over Rac1, because the latter displays high oncogenic potential. If a linear pathway existed, one would expect that RhoG were at least as transforming as its downstream element. This is not the case in our system. Moreover, we have demonstrated that the expression of RhoG Q61L does not have any effect in the activation of Rac1 either in vivo (using focus formation assays) or in vitro (by detecting the GTP levels of Rac1). Our results showing that RhoG is preferentially localized in cytoplasmic vesicles far away from the membranelocalized Rac1 indicate also that a close connection to allow the activation of Rac1 by this GTPase is difficult to achieve in vivo. Taken together, all this evidence is consistent with the idea that these two GTPase are functionally autonomous, regulat- ing parallel pathways that share some, but not all, signaling elements. This is probably not an exclusive property for the RhoG and Rac1 relationship, because preliminary evidence indicates that RhoG and Cdc42 work also through parallel pathways. 3 The lack of a linear relationship between RhoG and Rac1 is understandable from a mechanistic point of view, because RhoG and Rac1 share GEFs such as Trio, Vav1, Vav2, and Vav3 (24,32). So, the stimulation of one of these GEFs should suffice to ensure the simultaneous activation of both RhoG and Rac1 without the need of signaling intermediates. Although we have not attempted a careful experimental analysis to explain the basis of our discrepancies with previous publications, it is worth noting that similar conclusions have been reached by Wennerberg et al. (33) during the final elaboration of our work. In this case, they have explained the disagreement with previous reports by demonstrating that the inhibitory effect of the dominant negative mutant of Rac1 (Rac1 N17 ) on RhoG Q61L -mediated morphological change is probably because of either nonspecific or cytotoxic effects derived from the overexpression of Rac1 N17 for extended periods of time (33). In agreement with this hypothesis, no inhibitory effects of Rac1 N17 on RhoG signals are observed when the former protein is expressed during short periods of time (2-3 h instead of the usual 24 -48 h) (33). These observations highlight the importance of taking into consideration the multiple caveats associated with the use of this type of dominant negative mutants (34).
The different biological activities of RhoG and Rac1 suggest that, despite their high sequence similarity, they must diverge in specific residues that determine the pattern of effector binding. We have used an extensive mutagenesis approach to identify some of those molecular cues. Our data indicate that the proper connection of RhoG and Rac1 GTPases with specific cellular responses involves at least four different structures (switch I, the ␤2/␤3 hairpin, the ␣5 helix, and the C-terminal polybasic region), each of them accomplishing specific signaling tasks (Fig. 8B). In the case of PAK1 binding, we have shown that the binding specificity is meditated by residues located in the ␤2 (Val-44) and ␣5 regions (Asp-170) (Fig. 8B). Mutation of each of these two residues of Rac1 into the equivalent amino acids present in RhoG leads to either the total (Val-44) or partial (Asp-170) loss of PAK1 association by Rac1. This is not because of deleterious effects of the mutations in the structure or activity of Rac1, because the mutations of those areas do not affect the subcellular localization or the morphological changes induced by Rac1 Q61L . Conversely, the S44V substitution in RhoG Q61L along with the inclusion of the ␣5 region of Rac1 (RhoG Q61L /Rac1 chimera F) allows the binding of PAK1, a result that demonstrates the importance of these two regions for the proper docking of PAK1 to the upstream GTPases. However, it should be noted that the RhoG S44V/Q61L/D170 mutant does not recover PAK1 binding as efficiently as the RhoG Q61L /Rac1 chimera F, indicating that, in addition to residue Asp-170 of the ␣5 helix, the overall conformation of this area or the presence of additional residues in the polybasic region must also contribute to the overall binding affinity of PAK1 toward the GTPases. Interestingly, other divergent residues between Rac1 and RhoG located in the switch I (position 30, Gly and Lys in Rac1 and RhoG, respectively), ␤2 (position 45, Met and Ala in Rac1 and RhoG, respectively), ␤3 (position 50, Pro and Thr in Rac1 and RhoG, respectively), and ␣5 regions (position 167, Thr and Glu in Rac1 and RhoG, respectively) do not contribute to the differential binding of PAK1 to these GTPases. Our results are consistent with the type of interaction that PAK1 establishes with Cdc42 and Rac1. As described earlier by Morreale and collaborators (35), PAK1 wraps around half of the Cdc42 molecule, making contacts in the switch I (Pro-34 and Val-36 residues), switch II (Leu-70), the ␤2 regions (residues 40 -46), and ␣5 helix (Phe-169, Asp-170, and Ile-173). In agreement with such structure, the amino acid residues identified in our analysis match some of those areas. However, it was proposed that the ␣5 of Rac1 and Cdc42 worked as a passive contact site for PAK1 because, unlike the case of other effectors such as activated Cdc42-associated kinase and WASP, its replacement by the ␣5 region of RhoA (a non-CRIB domain binding protein) did not affect the overall affinity of the PAK1 association (29,35). Our results suggest that the role of ␣5 is not entirely passive in this process, because the Rac1 Q61L/D170A binds PAK1 poorly. One possible explanation for this discrepancy is that RhoA, Rac1, and Cdc42 share the same aspartic residue in position 170 (position 172 in RhoA) that is not conserved in RhoG. Accordingly, our prediction is that the ␣5 region of RhoA with the D172A mutation would not be compatible with PAK1 binding when transplanted into the Rac1 backbone.
We have obtained a different cartography of the residues responsible for the different behavior of RhoG and Rac1 in cell transformation. In this case, the important regions for this biological response are the switch I (position 30, amino acids Gly and Lys in Rac1 and RhoG, respectively) and the helix ␣5 (positions 167 and 170, Thr-Glu and Glu-Ala in Rac1 and RhoG, respectively) (Fig. 8B). Substitution of any of these three residues in Rac1 by the equivalent amino acids present in RhoG severely affects Rac1 Q61L transforming activity. Conversely, the substitution of Lys-30 in RhoG by the glycine residue present in Rac1 induces a low, but significant (approximately 1 ϫ 10 3 foci/g of transfected DNA), transforming activity in the mutant RhoG Q61L protein. In contrast to the results with PAK1, the ␤2/␤3 region was found not relevant for determining the specificity of this biological property. The common implication of residues of the ␣5 helix in the assembly of PAK1 binding and transforming activities is not entirely unexpected given the available evidence derived from other signaling and structural studies. For example, it has been shown that a C-terminal residue of the ␣5 region of Cdc42 (Leu-174) is the key specificity determinant for the binding to WASP and activated Cdc42-associated kinase (35). Likewise, the C-terminal region of the Rab3A ␣5 helix determines the specific binding to rabphilin 3A (36). N-terminal residues of the RhoA ␣5 helix also contribute to the binding of the serine/threonine kinase protein kinase N (37). Finally, residue Asp-170 of yeast Cdc42 appears to be very important for the assembly of specific signaling responses through the interaction with a hitherto unknown effector (38). These observations indicate that the contribution of GTPase ␣5 regions to signaling specificity is not restricted to RhoG but, rather, it is a usual event in the signal transduction of different GTPase subfamilies.
Our studies on the subcellular localization of RhoG and Rac1 indicate that these proteins also diverge in this biological property. As expected, the wild type versions of RhoG and Rac1 show a cytoplasmic distribution, without detectable localization in membrane-enriched areas. This is in agreement with the arrest in the cytoplasm of the GDP-bound forms of most Rho/Rac family GTPases through the binding of Rho GDIs (2). However, when the constitutively active versions of RhoG and Rac1 were analyzed, RhoG Q61L was detected mainly in intracellular vesicles that, in contrast to previous observations (22), did not match with lysosomal or endoplasmic reticulum markers. Instead, they seem to belong to the endocytic compartment, because they co-localize with typical markers for this pathway such as the epidermal growth factor receptor and its ligand. 2 A percentage of these vesicles also colocalize with caveolin, a marker for caveolae. 2 No RhoG Q61L was found associated with the plasma membrane or membrane ruffles, indicating that the effect of this GTPase on F-actin polymerization is probably mediated by a second messenger intermediate. In contrast to these observations, Rac1 Q61L was enriched at the plasma membrane and areas of membrane ruffling, in good agreement with previous observations (21). Using Rho Q61L /Rac1 chimeras, we have observed that this differential behavior of RhoG and Rac1 can be mapped to a totally distinct area than the previously described for PAK1 binding and oncogenesis (Fig. 8B). Indeed, RhoG protein can be relocated to the plasma membrane just by replacing its hypervariable, C-terminal region by the equivalent area of Rac1.
In this report, we have given a new example of exquisite effector selectivity of GTPases. The case of RhoG and Rac1 is even more remarkable than other cases of signaling divergence in this family, because these two GTPases are highly homologous. Through the use of numerous mutations, we have been able to associate the functional differences between these GT-Pases to specific residues located in different areas of the molecule. One question that remains to be addressed is why these proteins need to be so selective. For instance, it not obvious at first sight the biological reason for the binding of PAK1 to Rac1 and not to RhoG when this serine/threonine kinase binds with very high affinity to Cdc42. This question cannot be answered until we have genetic models for Rac1 and RhoG with which to dissect the actual contribution of each GTPase to cell signaling in vivo. Until that time, the reagents and information generated here provide a broad molecular foundation for continuing the analysis of the functional specificity of these two highly related GTPases. In addition, the elucidation of the mechanisms underlying these intermolecular interactions may serve as clues for the design of specific pharmacological agents that may inhibit in a specific fashion the signal transduction pathways connecting extracellular stimuli with cytoskeletal and mitogenic events.