Guanine nucleotide exchange factors regulate specificity of downstream signaling from Rac and Cdc42.

The Rac and Cdc42 GTPases regulate diverse cellular behaviors involving the actin cytoskeleton, gene transcription, and the activity of multiple protein and lipid kinases. All of these pathways can potentially become activated when GTP-Rac or GTP-Cdc42 is formed in response to external cell signals, yet it is evident that each activity must also be able to be controlled individually. The mechanisms by which such specificity of GTPase signaling in response to upstream stimuli is achieved remains unclear. We investigated the action of several well characterized guanine nucleotide exchange factors (GEFRho) to activate Rac- and/or Cdc42-dependent kinase pathways. Coexpression studies in COS-7 cells revealed that the ability of individual guanine nucleotide exchange factors (GEFs) to activate the p21-activated kinase PAK1 could be dissociated from activation of c-Jun amino-terminal kinase, even though activation of both pathways requires the action of the GEFs on Rac and/or Cdc42. In contrast, expression of constitutively active forms of Rac or Cdc42 effectively stimulated both downstream kinases. We conclude that GEFs can be important determinants of downstream signaling specificity for members of the Rho GTPase family.

In mammalian cells the Rho family of small GTPases includes Rac1 and -2, Cdc42hs, RhoA, RhoB, RhoC, RhoE, RhoG, and TC10. They are essential regulators of the dynamics of actin cytoskeletal structures and diverse cellular events requiring the actin cytoskeleton. It is now recognized that Rho family GTPases play an important role in the activation of stress response pathways leading to the generation of pro-inflammatory mediators and/or apoptotic cell death. Rho GTPases have also been implicated in the regulation of transcriptional events that contribute to cell growth regulation. There are numerous known targets for this family of small GTPases that may con-tribute to their ability to modulate cell function (1)(2)(3). To orchestrate both specific (individual) and complex (coordinated) cellular behaviors, a cell needs to be able to activate the diverse pathways downstream of Cdc42, Rac, or Rho in both specific fashion and as a coordinated sequence of events. In light of the tremendous diversity of Rho GTPase effector targets, one of the intriguing questions in understanding the signal transduction pathways regulated by Rho family small GTPases is how signaling specificity in response to distinct stimuli is determined.
Whereas cell type selectivity of downstream effectors is a potential mechanism to explain specific responses (e.g. Rac regulation of the NADPH oxidase in phagocytic leukocytes via p67 phox ), the ubiquitous nature of the majority of Rho effectors suggests that this is not a relevant explanation in most cells. One level of specificity resides in the difference between members in the Rho GTPase family, as different small GTPases do have certain downstream target differences. For example, RhoA specifically associates with PKN and Rhophilin (4,5), whereas p67 phox is a preferred target for Rac2 (6), and WASP interacts preferentially with Cdc42 (7). In many cases however, downstream effectors are shared by more than one family member; for example, Rac and Cdc42 both regulate PAK1-3 (8 -11), and both activate JNK 1 and p38 mitogen-activated protein kinases (12)(13)(14).
Guanine nucleotide exchange factors (GEFs) for the Rho family (GEF Rho ) are thought to be important components in regulating Rho family-mediated cellular behaviors in response to upstream stimuli. Rho GTPases must be converted from inactive GDP-bound forms to active GTP-bound forms; under normal cellular conditions this requires the activity of a GEF. The number of known and putative GEF Rho (currently 35, and the number appears to be increasing rapidly) is much greater than that of the Rho GTPases themselves (about 10 -12 in mammalian cells). GEF Rho are only weakly conserved at the Dbl homology domain responsible for catalytic guanine nucleotide exchange (10 -30% identity), and outside the Dbl homology domain sequence conservation is minimal (15). All GEF Rho are very large proteins (generally over 1000 aa, with Tiam1 having 1591 aa, Trio 2861 aa, and unc-89 being 6632 aa in length) with multiple domains that mediate protein-protein or protein-lipid interactions (15)(16)(17)(18). These domains include Dbl homology, SH2, SH3, zinc finger, and pleckstrin homology (PH) domains (15). It is easy to imagine that they serve as receivers for inputs from different intracellular and extracellular environments. It is less obvious that GEF Rho may also integrate downstream signals involving the network of small GTPases.
We hypothesize that different GEF Rho not only determine what intra-and extracellular signals feed into a particular set of Rho family switches, but that the GEF Rho can also dictate which downstream pathways can be activated. In this paper we present evidence supporting this hypothesis by establishing that the JNK and PAK1 protein kinases can be differentially activated by GEF Rho family members acting through the same GTPase.

MATERIALS AND METHODS
Plasmids-Construction of the Flag epitope-tagged versions of Rac1 and mutants for Rac1 (pcDNA3-Flag-Rac1Q61L, pcDNA3-Flag-Rac-1T17N) and JNK1 in the pcDNA3 vector have been described (14). Cdc42, Cdc42T17N, and Cdc42Q61L were cloned into pcDNA3 either with or without an NH 2 -terminal myc epitope tag using standard polymerase chain reaction methods, whereas PAK1 and PAK1 mutants were myc-tagged at the NH 2 terminus and subcloned into pCMV6 as described (19).
Three well characterized mammalian GEF Rho were selected for these studies (of 35 members revealed in a data base search). They were FGD1, Dbl and pDbl (proto-Dbl), and Tiam1 and dN-Tiam1 (NH 2terminal-truncated form). GFKG1, a less characterized GEF Rho , was included for comparison (KIAA0006, GenBank TM accession number D25304). 2 The open reading frame of FGD1 is disrupted by a chromosomal translocation in patients with faciogenital dysplasia (Aarskog-Scott Syndrome) (20). FGD1 has been shown to primarily stimulate guanine nucleotide exchange on Cdc42 both in vitro and in vivo (21,22). We evaluated two different FGD1 truncations: RKB1 (aa 391-710), which lacks the zinc finger domain) and RKB3 (aa391-790, which includes the zinc finger domain) (21). For Tiam1 and Dbl, we used both the full-length (FL-Tiam1 and pDbl) and the NH 2 -terminal truncated (dN-Tiam1 and Dbl) forms. Overexpression of pDbl is sufficient to induce transforming activity in NIH3T3 cells, but truncation at the NH 2 terminus further enhances the transforming activity (23)(24)(25). Elevated levels of FL-Tiam1 or truncation of Tiam1 at the NH 2 terminus induces invasiveness of T lymphoma cells and metastases in nude mice (16). The Tiam1 NH 2 -truncated construct (aa 386 -1591) remains sufficient for proper membrane localization and cytoskeletal regulation by this GEF (40). Tiam1 exerts its biological effects primarily through activation of Rac1 (26), whereas Dbl is primarily active toward Cdc42 (27).
Expression plasmids containing these GEF Rho have been described: FGD1 (20), Dbl and pDbl (proto-Dbl) (23,28), Tiam1 and dN-Tiam1, aa 386 -1591 (16). GFKG1 was subcloned (nucleotides 132-2384 of the original cDNA) into pCMV6M (19). All the GEF Rho constructs used in this study are driven by very strong viral promoters (see above references), and the expression of these proteins at similar levels was confirmed to the extent possible, given differences in the antibodies used for detection. None of the empty vectors had effects on PAK1 or JNK1 activity; consequently we show a single vector control for each set of experiments throughout.
Cell Culture and Transfection-COS-7 cells were obtained from ATCC and maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum. 80 -90% subconfluent COS-7 cells on 60-mm culture dishes were transfected with 5 g of total plasmid DNA. If the amount of DNA was less than 5 g, we added pCMV5/LacZ to make up the difference. DNA was introduced into the cell with Lipo-fectAMINE (Life Technologies, Inc.) as recommended by the manufacturer. To ensure that cells expressing the most downstream component were likely to also express upstream factors, more plasmid DNA for upstream components in the signaling pathway was used (in the order GEF Rho ϾRac1/Cdc42ϾPAK1/JNK1 at a cDNA ratio of 2.5:1.5:1). The expression level of endogenous PAK1 was very low, though detectable with PAK1 antibody R626. Between 36 and 48 h after addition of DNA, which we determined to be optimum for protein expression, the cells were collected for experiments.
In experiments to evaluate PAK1 and JNK activation by different amounts of expressed Rac1Q61L or Cdc42Q61L, we held the level of PAK1 cDNA or JNK cDNA constant and varied the levels of cDNA for each GTPase over a linear range, as indicated; the total amount of vector cDNA was held constant.
Protein Kinase Assays-PAK1 activity was measured essentially as in (10,19). Briefly, cells from a 60-mm plate were washed once with cold 1 ϫ phosphate-buffered saline, then lysed with 300 l of PAK lysis buffer (50 mM Tris, pH 7.5, 1 mM EDTA, 0.1 mM EGTA, 5 mM MgCl 2 , 1% Nonidet P-40, 10% glycerol, 150 mM NaCl, and 10 mM NaF). Cell debris was removed by low speed centrifugation in a microcentrifuge. To the resulting supernatant we added 2-3 g of 9E10 monoclonal antibody against Myc epitope tag and incubated for 2 h at 4°C, followed by precipitation with 20 l of protein G-Sepharose 4-Fast Flow (Amersham Pharmacia Biotech) and washing (19). At this point the beads were split: one-half for Western blotting to normalize the kinase activity and one-half for the kinase reaction itself. The assay was performed in 50 mM HEPES, pH 7.5, 10 mM MgCl 2 , 2 mM MnCl 2 , and 0.2 mM dithiothreitol using 1 g per assay of myelin basic protein (MBP) as exogenous substrate. Analysis was by 12% SDS-polyacrylamide gel electrophoresis.
The JNK assay was performed essentially as in (29). Cells from a 60-mm plate were lysed in 300 l of JNK lysis buffer (25 mM HEPES, pH 7.5, 300 mM NaCl, 1.5 mM MgCl2, 0.2 mM EDTA, 20 mM ␤-glycerophosphate, 0.1% Triton X-100, 0.5 mM dithiothreitol, 1 mM Na 3 VO 4 , 20 g/ml leupeptin, 2 mM phenylmethylsulfonyl fluoride, and 20 g/ml aprotinin). 2-3 g of M2 mouse monoclonal antibody against the Flag epitope was added and incubated at 4°C for 2 h and then the complex was precipitated with 20 l of protein G-Sepharose beads (50% suspension), washed, and suspended in JNK kinase buffer (50 mM HEPES, pH 7.5, 7.5 mM MgCl 2 , 0.5 mM EGTA, 0.5 mM NaF, 12.5 mM ␤-glycerophosphate, and 0.5 mM Na 3 VO 4 . The washed beads were divided: one-half was used for JNK kinase assay using glutathione S-transferase fusion protein consisting of the c-Jun NH 2 terminus (aa 1-79) as substrate (29) and the other half for Western blotting to normalize the kinase activity. For both kinase assays we used the PhosphorImager (Molecular Probes) to quantitate activities and the data was normalized for protein kinase expression.
Western Blots-Western blots were performed as described (30) at 1:1000 dilutions of each primary antibody and protein bands detected using a secondary goat anti-mouse (Myc) or goat anti-rabbit (Flag, Tiam1) antibodies conjugated with alkaline phosphatase at 1:1000 dilution, with color development using the Bio-Rad detection system. The immunoreactive bands were quantitated using NIH Image 1.6. Endogenous PAK was detected using polyclonal antibody R626 (10).

RESULTS
PAKs have been demonstrated in previous studies to be capable of stimulating the JNK pathway under appropriate circumstances (14,31,32). In the strain of COS-7 cells used for this study, we observed that expression of constitutively kinase-active forms of PAK1 had only a weak effect on JNK activity (Fig. 1). Expression of either the PAK1(T423E) mutant or PAK1(Y107F), both of which exhibit a high level of constitutive serine/threonine kinase activity in the absence of Rac-or Cdc42-GTP (33,34), caused very little increase in JNK activity, with only the Y107F mutant causing a consistent modest (2-3-fold) increase in activity (34). Furthermore, whereas JNK activity was effectively stimulated (5-10-fold) by expression of Rac1Q61L or Cdc42Q61L constitutively active mutants (Figs. 1 and 2B), this stimulation was not substantially inhibited by coexpression of a dominant negative PAK1 (H83L, H86L, K299R) mutant (33). Thus the ability of Rac1 and Cdc42 to stimulate the JNK cascade was largely independent of PAK1 under the experimental conditions.
In vitro and in vivo experiments indicate that PAK1 becomes active only when Rac1 and Cdc42 are present in GTP-bound forms (8 -11). 3 We used cotransfection experiments to assay which of the previously characterized Rac/Cdc42 GEF (termed GEF Rho ) induced PAK1 activation by stimulating guanine nucleotide exchange on either cotransfected Rac1 or Cdc42. We observed only weak activation of PAK1 or JNK when wild-type Rac or Cdc42 were introduced into the cells (Fig. 2), although Cdc42 had some stimulatory effect toward PAK1 which likely reflects basal activation of this GTPase in COS-7 cells, as suggested in a previous report (35). Similarly, the various exchange factors had little effect on PAK1 or JNK activity when expressed by themselves, apparently limited by the levels of endogenous GTPase. An exception was RKB3, which by itself induced an ϳ5-fold increase in JNK activity.
The comparative abilities of the various GEF Rho to stimulate PAK1 and JNK through Rac1 and Cdc42 were carefully quantitated by normalizing PAK1 activity (MBP phosphorylation) and JNK activity (glutathione S-transferase-Jun NH 2 terminus phosphorylation) to the amount of expressed protein in the immunoprecipitate. Intriguingly, the activation profile of JNK by GEF Rho was clearly different from that for PAK1 (Fig. 2). The strongest stimuli for PAK1 activation (Tiam1 and Dbl) induced only modest activation of JNK. FL-Tiam1 and dN-Tiam1, which activated PAK1 by more than 15-fold, stimulated JNK 2.1-and 3.6-fold, respectively, in the presence of Rac1 and in the presence of Cdc42 by 2.5-and 3.8-fold, respectively. This was not significantly different from the effect of expression of these GEF in the absence of either GTPase. Dbl and pDbl (6 -12-fold increase in PAK1 activity with each GTPase) each stimulated JNK weakly and only in the presence of Cdc42 by 2.3-and 2.6-fold, respectively. GFKG1 showed modest, though statistically significant, activation of JNK with either Rac1 or Cdc42 (2.7-and 3.2-fold, respectively), which was in contrast to its inability to activate PAK1. FGD1 had the most dramatic effect, with both the RKB1 and RKB3 forms activating JNK through Cdc42 about 20-fold. This was in contrast to its total inability to activate PAK1. The levels of PAK1 and JNK activity achieved with active GEF Rho were comparable with or greater than those obtained by expression of RacQ61L or Cdc42Q61L (Fig. 2). Furthermore, these effects were not a result of differential activation of PAK1 and JNK by active Rac and Cdc42, as we observed similar activation of each effector at each level of RacQ61L or Cdc42Q61L expression (Fig. 1D).
The differences observed were also not because of differential expression of Rac and Cdc42, individual GEF Rho , nor kinase components. Fig. 3 shows a representative immunoblot of protein expression under the experimental conditions. Within the limits imposed by the use of different antisera, there were roughly equal amounts of Rac and Cdc42 expressed. Tiam1 and FGD1 were expressed to similar levels in the presence of PAK1 or JNK, and similar results were obtained with the myc-tagged GFKG1 (not shown). We were unable to detect Dbl expression with available Dbl antisera, but functional effects observed with the Dbl constructs, plus the lack of variation in the expression levels of Tiam1, GFKG1, or FGD1 when cotransfected with either PAK1 or JNK cDNAs, strongly suggest that Dbl expression levels were similarly unaffected. Finally, there were

FIG. 2. Comparison of the ability of GEF Rho to stimulate PAK1
versus JNK1 activation through Rac1 and Cdc42. Cell lysates were prepared from COS-7 cells transfected with JNK1, PAK1, and various other plasmids as indicated, then assayed for PAK and JNK activity as described. A represents the quantitation of the effects of various GEF Rho expressed as -fold activation over PAK1 transfection alone (ϭ1.0). The values shown, which represent quantitation of the phosphorylation of MBP, were normalized to the level of PAK1 protein in each immunoprecipitate. Error bars indicate standard deviation of at least three independent experiments. In B, phosphorylation of glutathione S-transferase-c-Jun (aa 1-79) was normalized against the amount of immunoprecipitated JNK protein, as determined by Western blotting. JNK activity was consistently stimulated to a greater extent by Cdc42Q61L than by RacQ61L in all experiments. Results shown are the averages and S.D. of at least three independent experiments except where error bars are absent, which represents the average of two determinations. R ϭ Rac1 wild-type; C ϭ Cdc42 wild-type; dN-Tiam1 ϭ NH 2 -terminal truncated Tiam1 (similar results were obtained with full-length Tiam1; not shown); Dbl ϭ oncogenic Dbl (similar results were obtained with proto-Dbl; not shown); RKB3 ϭ truncated form of FGD1 (similar results were obtained with the RKB1 construct; not shown), R QL and C QL ϭ Rac1Q61L and Cdc42Q61L constitutively active mutants, respectively. no differences in levels of kinase expression when coexpressed with different GTPases and/or GEF Rho .

DISCUSSION
The existence of diverse effector molecules for Rho family GTPases has been established. Whereas some of these effectors appear to be specifically targeted by a single member of the Rho family (e.g. PKN is a Rho target only), other effectors may interact with more than one Rho family GTPase (e.g. PAK is both a Rac and Cdc42 target). Additional overlap of signaling pathways may result from crosstalk between individual Rho GTPases (e.g. Cdc42 can activate Rac; Rac can activate Rho). Models to explain the activation of Rho family GTPases by upstream stimuli, however, must account for the selective activation of specific subsets of effectors for the subsequent stimulation of divergent cellular responses. Whereas phosphatidylinositol 3-kinase has been implicated in control of Racmediated signaling pathways, possibly through its ability to control a RacGEF (36), a recent paper has demonstrated the selective activation by phosphatidylinositol 3-kinase of only a subset of potential Rac (and Rho) responses (37). Moreover, selectivity in the activation of effector pathways is necessary to explain how distinct effector responses are elicited by the diverse stimuli that make use of Rho GTPases for intracellular signaling.
The data presented here indicate that at least a part of the specificity in Rho GTPase signaling is a result of regulation through the action of GEF Rho . Thus, we observed that it was possible to induce the activation of specific downstream kinase pathways with certain GEF (Fig. 2). The best example of this was the ability of FGD1 to dramatically stimulate JNK activity in a Cdc42-dependent manner while having essentially no effect on PAK1 activity. Less dramatically, GFKG1 showed significant activation of JNK through both Rac1 and Cdc42 (2.7and 3.2-fold, respectively), which was in contrast to its inability to activate PAK1. Conversely, whereas Tiam-1 and Dbl were effective Rac-and Cdc42-dependent stimulators of PAK1 activity, they caused only small increases in JNK activity. These differences in effects of individual GEF Rho are unlikely to be due to experiment-to-experiment variations in the production of Rac-GTP or Cdc42-GTP, because we observed very little variation in the levels of these GTPases expressed (see Fig. 3) and obtained identical results when the same population of COS-7 cells was used in simultaneous transfections. We were also careful to control for expression of individual GEF Rho and kinase components by immunoblotting (Fig. 3). It should be noted that, whereas we cannot state definitively that the endogenous GEF Rho of the types we have used for this study have the same specificity for PAK1 versus JNK1 activation as we observe after transient expression, we believe that our data establish the principle that these GEF are capable of directing downstream signaling to specific effector pathways.
There are several possible mechanisms through which GEF Rho could act to specify downstream signaling. The first possibility involves specific localization of GEF within the cell such that only particular Rho GTPase pools become activated. For example, FGD1 would localize to a site where it activates Cdc42 able to initiate a JNK activation cascade, but where it would not have access to PAK1. There is evidence in support of such a mechanism. PH domains appear to be invariant features of GEF Rho (15). Deletion of these domains impairs the normal biological activities of GEF Rho , whereas replacement of the PH domain with a CAAX isoprenylation and membrane-targeting motif restores normal activity (38,39). Expression of an isolated PH domain from Dbl inhibited the ability of Dbl to transform cells, presumably because it competed with intracellular targeting sites (38). One of our laboratories (J. Collard) has established the critical importance of the NH 2 -terminal, but not the COOH-terminal, PH domain of Tiam1 for membrane localization and downstream signaling through Rac (40). PH domains may serve to localize exchange factors to the plasma membrane (or other locations) through interactions with G protein subunits and/or phosphatidylinositol lipids, where they would have access to their GTPase substrates and/or other signaling components (41)(42)(43)(44).
A second possibility is that GEF Rho can interact with their GTPase targets in such a way that the GEF still associates with the GTPase in the GTP form after the GDP/GTP exchange reaction has taken place. This interaction may then influence the interaction of the GTPase with its downstream targets such that activation of one is favored. A corollary of this model is the possibility that different GEF may also form complexes with downstream targets to ensure signaling specificity. There is biochemical evidence that small GTPases in the GTP form can interact with GEF Rho (45)(46)(47); for example, complexes of Rac-GTP with the GEF Rho Ect2 and Dbl have been described (47). It is now clear that different protein targets of Rac and Cdc42 interact with distinct sites on the GTPase (48 -50). Rho GTPases have at least three regions that are part of the overall effector domain. For example, whereas two of the regions on Rac1 (aa 22-45 and aa 143-175) are essential for activating both PAK and p67 phox , a third region (aa 74 -90) in conjunction with the NH 2 -terminal effector region (aa  forms the Bcr GAP interaction site (48). Even individual residues within the effector domain present at the Switch I region (aa  can determine target protein interactions (51,52). It is possible that the binding of a GEF could mask, for example, a specific Rac or Cdc42 effector interaction by sterically blocking interactions involving aa 37 versus aa 40, each of which is required to mediate interactions with individual effector targets (51,52). The interactions of Rho GTPases with their GEF regulators FIG. 3. Expression levels of proteins used in these studies. The immunoblots shown establish the levels of protein expression of the various cDNAs used in the coexpression experiments. The gels were immunoblotted using 1:1000 dilutions of the 9E10 myc epitope antibody, the anti-Flag antibody (Santa Cruz), and Tiam1 antibody (16) simultaneously. Abbreviations are as in Fig. 2, and J ϭ JNK1; P ϭ PAK1. The arrows at the right indicate the presence of a nonspecific band that appears across the gel; the asterisks indicate the position of Tiam1 breakdown products, which cross-reacted with the Tiam1 polyclonal antibody.
has not been mapped in detail, although both the Switch I and Switch II regions appear to be important (53), as is the COOH terminus (54).
We conclude that guanine nucleotide exchange factors can contribute some of the specificity that determines which effector proteins become activated when a Rho GTPase is activated. The large size of most GEF Rho , the presence of subcellular targeting motifs and other protein-protein interaction motifs in these molecules, and their existence as a large, growing, and diverse protein family make these GTPase regulatory molecules ideal candidates to mediate Rho GTPase signaling specificity.