Structural determinants of RhoA binding and nucleotide exchange in leukemia-associated Rho guanine-nucleotide exchange factor.

Rho guanine-nucleotide exchange factors (RhoGEFs) activate Rho GTPases, and thereby regulate cytoskeletal structure, gene transcription, and cell migration. Leukemia-associated RhoGEF (LARG) belongs to a small subfamily of RhoGEFs that are RhoA-selective and directly activated by the Galpha12/13 family of heterotrimeric G proteins. Herein we describe the atomic structures of the catalytic Dbl homology (DH) and pleckstrin homology (PH) domains of LARG alone and in complex with RhoA. These structures demonstrate that the DH/PH domains of LARG can undergo a dramatic conformational change upon binding RhoA, wherein both the DH and PH domains directly engage RhoA. Through mutational analysis we show that full nucleotide exchange activity requires a novel N-terminal extension on the DH domain that is predicted to exist in a broader family of RhoGEFs that includes p115-RhoGEF, Lbc, Lfc, Net1, and Xpln, and identify regions within the LARG PH domain that contribute to its ability to facilitate nucleotide exchange in vitro. In crystals of the DH/PH-RhoA complex, the active site of RhoA adopts two distinct GDP-excluding conformations among the four unique complexes in the asymmetric unit. Similar changes were previously observed in structures of nucleotide-free Ras and Ef-Tu. A potential protein-docking site on the LARG PH domain is also evident and appears to be conserved throughout the Lbc subfamily of RhoGEFs.

vitro nucleotide exchange on Cdc42 and RhoA 11-and 24-fold, respectively (9,20). Conversely, the PH domains of Sos and the C-terminal DH/PH domains of Trio appear to inhibit nucleotide exchange (16,21).
The relative orientation of the PH domain with respect to the DH domain is similar in the structures of intersectin, Dbs, and Trio-N, suggesting a common functional role. However, only in the Dbs-RhoA and -Cdc42 complexes have direct contacts between the PH domain and the GTPase been observed. These contacts are important for PH domain-assisted nucleotide exchange in vitro and Dbs function in vivo (18). Intersectin does not form analogous contacts between its PH domain and Cdc42 and does not exhibit PH domain-assisted nucleotide exchange (22). The PH domain of Trio-N assists in nucleotide exchange (16), however a structure of the Trio-N DH/PH domains in complex with their substrate GTPase is not available.
Leukemia-associated RhoGEF (LARG) and its close homologs, p115-RhoGEF and PDZ-RhoGEF, are RhoA-selective RhoGEFs that are directly regulated by activated Ga 12/13 proteins and thereby play a key role in oncogenic transformation induced by G protein-coupled receptors (23)(24)(25). All three Rho-GEFs contain a regulator of G protein signaling (RGS) homology (RH) domain positioned ϳ200 residues N-terminal to their DH/PH domains and are therefore referred to as the RH-Rho-GEFs (24). The RH domain binds to and serves as a GTPaseactivating protein for activated G␣ 12/13 (26,27). At the same time, the binding of G␣ 12/13 (27,28) and possibly G␣ q (29) stimulates nucleotide exchange on RhoA. The PH domains of LARG (30) and p115-RhoGEF (31) are required for full catalytic activity in vitro and enhance nucleotide exchange ϳ2and 14-to 24-fold, respectively. Whereas the contributions of the LARG and other RhoGEF PH domains toward nucleotide exchange in vitro are relatively small (2-to 24-fold), it is anticipated that they have much more profound effects in vivo, as was shown for Dbs (18).
To better understand the structure and regulation of human LARG, we initiated crystallographic studies of its DH/PH domains (32). Herein we report atomic structures of the DH/PH domains of LARG and their complex with a soluble (unprenylated) form of human RhoA. These structures reveal novel interactions between the LARG DH and PH domains and RhoA. Using site-directed mutagenesis and fluorescence-based nucleotide-exchange assays, we show that these interactions are important for LARG-mediated nucleotide exchange in vitro.

EXPERIMENTAL PROCEDURES
Cloning, Expression, and Protein Purification-The cloning, expression, and purification of the LARG DH/PH fragment and TEV protein were as previously described (32). DNA encoding the LARG DH domain (residues 765-986) was cloned into a modified pMAL expression vector (pMALc2H 10 T) using BamHI and SalI restriction sites. The pMALc2H 10 T expression vector was generated by inserting oligonucleotides encoding a decahistidine (H 10 ) tag followed by a TEV protease recognition site between AvaI and EcoRI of the pMALc2X vector (New England Biolabs). Expression and purification of the LARG DH domain was as described previously for the DH/PH domains except that protein was expressed at 20°C, the MBP-DH fusion protein was dialyzed against buffer containing 100 mM NaCl, and finally digested with 2% (w/w) TEV protease. Fractions containing the DH domain were pooled, concentrated to ϳ5 mg/ml, and stored at Ϫ80°C.
The DNA sequence encoding 1-193 of human RhoA was cloned from the pGEXKG-RhoA vector (T. Kozasa, University of Illinois at Chicago, Medical Center) into pMALc2H 10 T using the EcoRI and SalI restriction sites. Protein expression from Rosetta (DE3) pLysS cells transformed with the pMALc2H 10 T-RhoA vector was induced with 1 mM isopropyl 1-thio-␤-D-galactopyranoside at 30°C and harvested after 4 -6 h. Lysis and purification was as described above for the LARG DH domain, except that cells were lysed in the presence of 50 M GDP. Buffers for nickel-nitrilotriacetic acid columns also contained 10% glycerol, 10 mM MgCl 2 , and 5 M GDP, and the gel filtration buffer contained 1 mM MgCl 2 and 40 M GDP. RhoA was concentrated to ϳ5 mg/ml and stored at Ϫ80°C. The coding region for residues 1-192 of human Rac1 and residues 1-191 of human Cdc42 were amplified from pCDNA-rac1 and pCDNA-cdc42 (gifts from S. Dharmawardhane, University of Texas at Austin), and then cloned into the pMALC 2 H 10 T vector using the EcoRI/ SalI sites. The proteins were then purified as described above for RhoA.
Purification of the RhoA-DH/PH Complex-The DH/PH domains of LARG were mixed with a 2-fold molar excess of RhoA and diluted 10-fold with complex buffer (20 mM HEPES, pH 8.0, 150 mM NaCl, 10 mM EDTA, 2 mM dithiothreitol). After incubation for 10 min on ice, the complex was loaded onto an S200 16/60 size-exclusion column preequilibrated in complex buffer supplemented with 1 mM EDTA. Fractions containing the 1:1 RhoA-DH/PH complex were pooled, concentrated to about 8 mg/ml, and stored at Ϫ80°C until crystallization.
Crystallization and Data Collection-Crystallization of and data collection from the LARG DH/PH domains were described previously (32) ( Table I). DH/PH-RhoA complex crystals were formed by vapor diffusion using wells containing 50 mM sodium phosphate, pH 7.4, 11% PEG 8K, 0.6 M NaCl, and 5 mM EDTA. The crystals grew as long rods that can approach 1 mm in length and have 72% solvent content. Native data from the DH/PH-RhoA complex were collected from a single crystal at 90 K harvested in cryoprotectant solution (15% PEG 400, 50 mM sodium phosphate, pH 7.4, 20 mM HEPES, pH 8.0, 15% PEG 8K, 0.6 M NaCl, 5 mM EDTA, and 2 mM dithiothreitol) on beam line 8.2.1 at the Advanced Light Source, Lawrence Berkeley National Lab (Table I). Data were reduced by HKL2000 (33).
Structure Determinations-The structure of the LARG DH/PH domains was determined using a combination of MIRAS and molecular replacement. Xenon and NaBr derivatives were generated as described previously (32) and a homology model of the LARG DH domain, based on that of intersectin (10), was manually placed into the resulting MIRAS-phased electron density map. Phases from molecular replacement and MIRAS were then combined, and the DH domain was refined using CNS (34). Subsequently, a solvent-flattened electron density map allowed placement of a homology model of the LARG PH domain. The structure was refined using rounds of maximum-likelihood refinement by either CNS or REFMAC (35) alternating with model building in the program O (36). In the final rounds of refinement, individual isotropic B-factors were used in conjunction with TLS refinement (37). Modeling of the DH/PH domains was ultimately assisted by the structure of the DH/PH-RhoA complex, which facilitated interpretation of poorly ordered regions of the structure, particularly the N-terminal extension of the DH domain, and the ␤1-␤2 and ␤N-␣N loops of the PH domain.
The 3.2-Å crystal structure of the LARG DH/PH-RhoA complex was determined by molecular replacement using as a search model the LARG DH domain modeled in complex with nucleotide-free RhoA (10). The PH domains of the four complexes in the asymmetric unit were later fit by hand. The structure was refined and built as described for the LARG DH/PH domains, except that 4-fold NCS restraints were imposed on structurally equivalent regions of each DH/PH-RhoA complex throughout refinement using REFMAC. As structural differences between subunits became apparent, these restraints were gradually loosened and/or eliminated. For all DH domains and RhoA subunits, main-chain and side-chain densities are well defined. The A and C chain PH domains are better ordered than the E and G PH domains and have even more ordered loops than the PH domain of the 2.1-Å uncomplexed structure. To verify the resulting model, A -weighted phases (38) were generated from the coordinates and refined with twenty cycles of solvent flattening and averaging in the program DM (39). Multidomain 4-fold averaging was used for all parts of the structure except for the RhoA subunits, which were subjected to 3-fold averaging owing to the observed conformational change in the B chain, whose density was omitted from averaging. Correlation between the averaged DM map and the 2͉F o ͉ Ϫ ͉F c ͉ Fourier map generated by REFMAC was 95% for main-chain atoms and 92% for side-chain atoms. The omit map shown in Fig. 5a was generated in a similar fashion, except that the B chain of RhoA was left out of the initial model used to generate phases.
Two residues in each subunit of LARG, Ser-833 and Asp-1054, fall within the disallowed region of the Ramachandran plot. In other atomic structures of DH domains, residues equivalent to Ser 833 have the same strained backbone stereochemistry. Asp-1054 exists in the iϩ1 position of a type I ␤-turn in the PH domain, a position usually occupied by glycine. In the DH/PH structure, the residue Asn-765 of the DH domain and residues 999 -1007 in the ␤N-␣N loop and 1062-1074 in the ␤4 strand of the PH domain could not be modeled. In the "A" and "C" DH/PH chains of the LARG DH/PH-RhoA complex, Asp-765 at the N terminus and residues 1064 -1074 in the ␤4 insertion of the PH domain do not have interpretable electron density and were not modeled. In the "E" and "G" DH/PH chains, the PH domains were substantially more disordered, and various additional loops between secondary structural elements could not be modeled. Refinement statistics for the LARG DH/PH and DH/PH-RhoA structure determinations are shown in Table  I. Atomic coordinates and structure factor files have been deposited for the LARG DH/PH domain and the LARG DH/PH-RhoA complex in the Protein Data Bank with accession codes 1TXD and 1X86, respectively. Visual representations of the models were created using PyMOL (40).
DH/PH and RhoA Mutagenesis and Purification of Mutants-DH/ PH-W769A, DH/PH-W769D, DH/PH-⌬N, DH/PH-E1023A, DH/PH-E1023R, and DH/PH-S1118D mutants were generated in the pMALc2TH 6 -DH/PH bacterial expression vector by site-directed mutagenesis using the QuikChange mutagenesis kit (Stratagene). RhoA mutants RhoA-V33E, RhoA-K27T, RhoA-R68A, and RhoA-E97A were produced from pMALc2H 10 T-RhoA. The complete coding sequence for each of the mutant proteins was verified by DNA sequencing. The DH/PH and RhoA mutant constructs were expressed and purified as described above for the wild-type proteins.
Nucleotide Exchange Assay-GTPases were loaded with N-methylanthraniloyl-GDP (mant-GDP; Jena Bioscience) by incubating 180 M RhoA with a 10-fold molar excess of mant-GDP in loading buffer (20 mM HEPES, pH 8.0, 100 mM NaCl, 4 mM EDTA, 1 mM dithiothreitol) for 1.5 h on ice. Subsequently, the mant-GDP-loaded GTPase was stabilized by addition of MgCl 2 to a final concentration of 10 mM and incubated for an additional 30 min on ice. mant-GDP-loaded GTPase was then exchanged into reaction buffer (20 mM HEPES, pH 8.0, 150 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol) via a G10 gel filtration column pre-equilibrated with reaction buffer to remove excess nucleotides. Fluorescence assays were performed on a Fluoromax-3 spectrophotometer at 25°C ( ex ϭ 280 nm, em ϭ 430 nm, slits ϭ 2/2 nm), in which 1 M of each mant-GDP-loaded GTPase was incubated with 100 M GTP in reaction buffer in a 200-l cuvette. The exchange reaction was then started by the addition of 100 nM LARG fragment, and k obs was determined by modeling each trace as a one-phase exponential decay using the program Prism version 4.0. All proteins were purified to greater than 95% homogeneity as judged by SDS-PAGE, and their concentrations were determined using the BCA protein assay (Pierce).

RESULTS
The Structure of the LARG DH/PH Domains-As in previously determined DH domain structures, the core of the LARG DH domain is comprised of six major helical segments (9), wherein segments 2, 3, and 5 are broken into several distinct ␣-helices (Fig. 1a). The LARG DH domain, however, has a novel N-terminal extension (residues 766 -781) composed of two short helices, ␣N1 and ␣N2, that bury the side chain of Trp-769 against the ␣1 helix of the DH domain (Figs. 1a and 2a). This extension was included in the DH/PH fragment used for crystallization because of its high sequence homology among the Lbc subfamily of RhoGEFs (4), which includes the RH-RhoGEFs, Lbc, Lfc, Net1, Xpln, and intersectin (Fig. 2b). Otherwise, the LARG DH domain is quite similar to that of its closest homolog of known structure, intersectin (r.m.s.d. of 1.3 Å for 187 equivalent C␣ atoms), whose structure was determined without the corresponding N-terminal extension (10). The most pronounced difference between the LARG and intersectin DH domains occurs within the region spanning the ␣2-␣3 loop and the first helix of the third helical segment, which is on the opposite side of the DH domain from the GTPase binding site.
The PH domain of LARG has notable structural differences from other RhoGEF PH domains of known structure. The ␤3-␤4 loop of LARG is an abrupt ␤-turn, whereas the analogous loops in the Tiam1-Rac1 and Dbs-Cdc42 complexes are extended (8,9) (Fig. 1c). In the case of Dbs, the extended ␤3-␤4 loop forms direct contacts with Cdc42 (9). Although Dbs has a continuous ␤4 strand, intersectin has a three-residue bulge (residues 1515-1517) and LARG has a disordered 17-residue insertion (residues 1060 -1077) within the strand. The function of this insertion in LARG and other RH-RhoGEFs is not known, although it projects along what is anticipated to be the membrane binding surface of the PH domain ( Fig. 1b) and therefore could be involved in membrane association.
Embrace of RhoA by the LARG DH/PH Domains-The LARG DH/PH-RhoA complex crystallized as a tetramer with pseudo-C 4 non-crystallographic symmetry (Fig. 1b, inset). Although oligomers that associate with membranes often have cyclic symmetry, this tetramer is probably not physiologically significant, in part because the observed subunit contacts are not conserved in other RH-RhoGEFs. The DH/PH chains are labeled A, C, E, and G, and their respective RhoA subunits are B, D, F, and H. Thus the four DH/PH-RhoA complexes are referred to as being composed of the A:B, C:D, E:F, or G:H chains.
The most striking change in the LARG DH/PH domains upon binding RhoA is the roughly 30°rotation of the PH domain relative to the DH domain, such that both domains embrace RhoA (Figs. 1b and 3). This conformational change occurs by virtue of a bend at the end of the ␣6 helix of the DH domain that spans residues 976 -983 (thus the "PH domain" of LARG is therefore defined as spanning residues 984 -1138). The result- ing orientation of the LARG PH domain with respect to the DH domain is strikingly similar to those of Dbs in complex with either Cdc42 or RhoA (9, 10), Dbs alone (7), and Trio-N (6). However, the ␣6 segment of LARG is 10 residues longer, and thus the PH domain in the LARG DH/PH-RhoA complex is translated about 13 Å further along the direction of ␣6 (Fig. 1c).
The residues from RhoA that lose the most accessible surface area upon their interactions with the LARG PH domain are Glu-97 in the ␣3 helix, whose carboxylate interacts with the N terminus of the ␣C helix, and Arg-68 from switch 2, which packs against residues from ␣N and the ␤1 strand of the PH domain (Fig. 3b).
The PH domain of each LARG subunit has a slightly different orientation in the complex, varying as much as 10°when their DH domains are superimposed. Even so, each PH domain maintains similar contacts with RhoA (Fig. 3a). The observed rotation axis among the complexed PH domains is in a direction roughly orthogonal to that of the 30°collapse of the PH domain relative to the unbound DH/PH domains. The accessible surface area buried between the PH domain and RhoA is on  (9) are labeled, as are the ␣N1 and ␣N2 helices of the novel ␣N1/␣N2 extension at the N terminus of the domain. The side chains of Trp-769, which packs in the hydrophobic core of the ␣N1/␣N2 extension, and Glu-1023, which appears to contribute to PH domain-assisted nucleotide exchange, are shown as stick models. Like other RhoGEF PH domains, the LARG PH domain has an N-terminal extension that begins with an ␣-helix (␣N), followed by a ␤-strand (␤N), and then a 3 10 helix. The loop connecting ␤N and the 3 10 helix is highly variable and can contain long inserts (disordered in the uncomplexed LARG DH/PH structure). The ␤3 and ␤4 strands of RhoGEF PH domains are also longer, allowing them to form an additional small ␤-sheet with the ␤N strand. b, the LARG DH/PH-RhoA complex. With respect to panel a, The PH domain has swung ϳ30°downward to engage RhoA (green). The side chains of residues that form a conserved, solvent-exposed hydrophobic patch on the PH domain are drawn as stick models in yellow. This patch forms similar 2-fold dimer interfaces in both the DH/PH and DH/PH-RhoA crystal structures. Based on the position of the C terminus of RhoA (which is geranylgeranylated in vivo), the putative phospholipid binding surface of the PH domain and the flatness and positive charge of the top surface of the complex, the plasma membrane is predicted to run along the top of the panel. The LARG ␤N-3 10 loop becomes ordered upon the binding of RhoA and forms an additional ␣ helix (␣Nb). The inset shows the tetramer observed in the asymmetric unit of the LARG DH/PH-RhoA crystals, with the DH/PH domains rendered as space-filling models and the RhoA chains as green tubes. c, comparison of the DH/PH domains of LARG, intersectin (PDB code 1KI1) and Dbs (1LB1). The structures were aligned by superposition of their GTPase substrates (not shown). The orientation of the LARG PH domain with respect to the DH domain is most similar to that of Dbs, whereas that of intersectin is rotated 18°away from the DH domain and does not contact the GTPase substrate (10). The ␣6/␣N helix of Dbs is shorter than that of LARG, allowing its extended ␤3-␤4 loop to engage its GTPase substrate (not shown) (10). The conformation of the N-terminal DH/PH domains of Trio is essentially the same as that of Dbs (6). LARG, Dbs, and the N-terminal DH/PH domains of Trio exhibit PH domain-assisted nucleotide exchange in vitro, whereas intersectin does not. average 160 Å 2 , varying between 240 Å 2 in the C:D complex to 40 Å 2 in the E:F complex, wherein which the PH domain appears nearly disengaged from RhoA.
The conformation of the individual DH and PH domains in their complex with RhoA is similar to their counterparts in the unbound DH/PH structure (r.m.s.d values of 0.85 and 0.64 Å, respectively). Upon binding RhoA, the largest conformational change in the DH domain occurs within the ␣N1/␣N2 extension and the ␣4 region. The entire ␣N1/␣N2 extension shifts 2-3 Å toward RhoA in the complex. The ␣2-␣3 loop, which interacts extensively with the ␣N1/␣N2 extension, shifts similarly (Fig.  3a). The side chain of Glu-790 from the ␣1 helix assumes a bent conformation that enables it to form two backbone hydrogen bonds with the N terminus of ␣N1 and to pack against the aromatic ring of RhoA-Tyr-34 (Fig. 2a). The largest change in the ␣4 region occurs in the ␣4-␣5 loop, which directly engages RhoA (e.g. 2 Å shift for the C␣ of Arg-923 away from RhoA relative to the DH/PH structure) (Fig. 3a). The internal conformation of the LARG PH domain is essentially unchanged upon binding RhoA, although its loops become better ordered. For example, the 18-residue ␤N-3 10 helix loop becomes ordered in the A and C chains of the complex (Fig. 1b), and contains an extra helix (␣Nb, residues 1005-1013). A similar phenomenon was noted for the Dbs PH domain upon comparing its GTPasefree and bound structures (7).
The LARG DH-RhoA Interface-As observed in other DH-GTPase complexes, a region of switch 2 of RhoA (residues 61-68) reorganizes in the LARG-RhoA complex such that the side chain of Ala-61 projects into the magnesium binding site and Glu-64 occludes the ␥-phosphate binding site of RhoA. Residues 27-40 of RhoA, which includes switch 1, are also restructured such that the nucleotide binding site of RhoA becomes more solvent-exposed. Switch 1 appears to be better ordered in the LARG DH/PH-RhoA complex than in other DH/PH-GTPase complexes, perhaps by virtue of its additional contacts with the ␣N1/␣N2 extension.
The structure of LARG is the first of a RhoA-selective Rho-GEF, and, as previously proposed (10), it appears to use the ␣4-␣5 loop region to dictate substrate specificity (Fig. 4). Arg-923 in the ␣4-␣5 loop of LARG forms salt bridges with both Asp-45 and Glu-54 of RhoA, which are substituted by shorter polar side chains in Rac1 and Cdc42 (Fig. 4a). Arg-923 is invariant throughout the Lbc subfamily of RhoGEFs except in intersectin, which is specific for Cdc42 and has a glycine at the equivalent position. Within the same region, Arg-5, Val-43, and Asp-76 of RhoA appear to form additional RhoA-specific con- tacts with LARG (Fig. 4). The side chain of Trp-58 of RhoA is completely buried in the LARG interface and potentially forms a hydrogen bond with the carboxylate of LARG-Asp-928 (Fig.  4b), a residue conserved as aspartate or glutamate among Lbc subfamily RhoGEFs except intersectin, which has a serine at the equivalent position.
Alternative Conformations of Nucleotide-free RhoA-Overall, the conformation of RhoA in the LARG DH/PH-RhoA complex is similar to that of the GTPase in other DH/PH-GTPase complexes. However, there are some interesting structural differences (Fig. 5). In all four complexes of the LARG DH/PH-RhoA structure, the backbone carbonyl of Gly-14 in the P-loop has flipped so that it occludes the binding site for the ␤-phosphate of GDP. Asn-117, positioned next to the P-loop, adopts a rotamer not observed in previously determined RhoA structures. Strong electron density is observed in the ␣-phosphate binding site of all four RhoA subunits and was modeled as inorganic phosphate due to its presence in the crystallization buffer (Fig. 5a).
In the A:B LARG-RhoA complex, crystal contacts have trapped RhoA in a conformation wherein the purine-binding site of the nucleotide is also occluded (Fig. 5a). Residues 160 -164 of RhoA, which contain the SAK motif, shift by up to 3.4 Å into the purine-binding pocket of RhoA. An analogous collapse is observed in the Ras-Sos complex (41). This change is accompanied by a shift of residues 31-35 in switch 1 of RhoA further away from the nucleotide-binding site, such that the C␣ atom of Val-33 shifts by 1.6 Å.
Role of the ␣N1/␣N2 Extension-The structure of the LARG DH domain revealed the presence of a novel N-terminal extension that directly contacts switch 1 of RhoA. We investigated the role of the ␣N1/␣N2 extension in nucleotide exchange by using site-directed mutagenesis and a fluorescence resonance energy transfer-based assay that monitors the release of Nmethylanthraniloyl-GDP (mant-GDP) from RhoA (see "Exper-imental Procedures" and Fig. 6). To test whether the ␣N1/␣N2 extension could influence nucleotide exchange activity, we either deleted the extension (⌬N) or perturbed its hydrophobic core (W769A or W769D). To evaluate the role of Glu-790 from the ␣1 helix of the DH domain (Fig. 2a), which packs against RhoA-Tyr-34 and forms two hydrogen bonds with the backbone of the ␣N1/␣N2 extension, we mutated the residue to glycine (E790G), its equivalent in intersectin. To test whether the extension-switch 1 interface contributes to the substrate specificity of LARG, RhoA-Lys-27 and -Val-33 were mutated to threonine and glutamate, respectively, their counterparts in both Rac1 and Cdc42. All mutant proteins expressed similarly to wild-type and could be purified to the same level of homogeneity (data not shown), suggesting that they are not misfolded or otherwise destabilized.
The LARG-⌬N, W769A and W769D mutants all reduced nucleotide exchange of RhoA to 15-20% of the activity of the wild-type DH/PH domains ( Fig. 6a and Table II). The E790G mutant of LARG was similarly deficient at nucleotide exchange. Furthermore, neither wild-type, W769A, W769D, nor ⌬N mutants of LARG could catalyze nucleotide exchange on Rac1 or Cdc42 in our assays (data not shown). Thus, the ␣N1/ ␣N2 extension and Glu-790 are important for nucleotide exchange on RhoA, but apparently not for substrate specificity.
Surprisingly, neither the K27T nor the V33E mutations of RhoA exhibited a significant loss of nucleotide exchange relative to wild-type RhoA (Table II). Therefore, it appears that the observed contacts between ␣N1/␣N2 extension and switch 1 are not important. However, the ␣N1/␣N2 extension may still have an indirect effect on binding RhoA. Disruption or deletion of the ␣N1/␣N2 extension could allow the side chain of Glu-790 of LARG to adopt a conformation that interferes with GTPase binding. To test this hypothesis, we constructed the LARG-E790G/⌬N double mutant, which could potentially rescue the Because Asp-76 is substituted by a glutamine in both Rac1 and Cdc42, this contact could also contribute to substrate specificity. The mesh cage represents a A -weighted 2͉F o ͉ Ϫ ͉F c ͉ Fourier map contoured at 1.0 , generated using the CCP4 program suite (35). b, the same interface rotated 90°around a vertical axis, highlighting interactions with RhoA-Trp-58, which is buried at the interface. Asp-928 of LARG, a position conserved as an acidic residue in all Lbc subfamily RhoGEFs except for intersectin, is in position to form hydrogen bonds that bridge Trp-58 of RhoA and a backbone nitrogen in the LARG ␣4-␣5 loop. In Cdc42, the equivalent residue to RhoA-Trp-58 is phenylalanine, which would potentially collide with Asp-928 of LARG and leave a cavity in the interface. Val-43 of RhoA (substituted by serine and alanine in Rac and Cdc42, respectively) packs snuggly against the side chain of LARG-Arg-923. Conservative substitution of this valine residue with isoleucine, as in RhoC, could not be easily accommodated. Indeed, it was recently reported that Xpln, another Lbc subfamily RhoGEF selective for RhoA, can catalyze nucleotide exchange on RhoA and RhoB, but not RhoC (50). activity of the ⌬N deletion. However, the E790G/⌬N mutation was just as deficient at catalyzing nucleotide exchange on RhoA as the other ␣N1/␣N2 extension mutations ( Fig. 6a and Table II). Therefore, although Glu-790 may be constrained in a favorable conformation by the ␣N1/␣N2 extension, its contacts with RhoA-Tyr-34 also appear important for nucleotide exchange on RhoA.
The LARG PH Domain-RhoA Interface and Its Role in Nucleotide Exchange-Contacts between a RhoGEF PH domain and the GTPase substrate have only previously been observed in complexes of the Dbs DH/PH domains (9,10). We therefore tested the contribution of the LARG PH domain toward nucleotide exchange using site-directed mutagenesis and the fluorescence resonance energy transfer-based nucleotide exchange assay described above. First, we compared the rates of nucleotide exchange on RhoA catalyzed by either the LARG DH/PH domains or the DH domain alone (Fig. 6b). Consistent with previous studies of LARG and the closely related p115-RhoGEF (30,31), the DH domain of LARG catalyzed nucleotide exchange less efficiently than the DH/PH domains (Fig. 6b and Table II).
The two residues of RhoA that bury the most accessible surface area with the LARG PH domain are Glu-97 and Arg-68 (Fig. 3b). RhoA-Glu-97 interacts with the ␣C helix of LARG, where it forms hydrogen bonds with the backbone amides and/or the side chain of Ser-1118. The S1118D mutation, which introduces electrostatic repulsion and/or steric collision with RhoA-Glu-97, reduced the exchange rate to the level of the DH domain alone (Fig. 6b and Table II). However, nucleotide exchange of RhoA-E97A was nearly identical to that of wild-type RhoA. It is possible that the RhoA-E97A mutation was not severe enough to abrogate beneficial packing between the PH domain and RhoA.
RhoA-Arg-68 interacts with the ␣N helix and the ␤1 strand of the PH domain (Fig. 3b). Because our LARG DH domain fragment (residues 765-986) and a previously studied fragment (residues 785-1019) (30) were similarly deficient at nucleotide exchange when compared with the intact DH/PH domains, the residues within ␣N (residues 983-993) appear, at least by themselves, incapable of facilitating nucleotide exchange. We therefore targeted Glu-1023 of the PH domain for site-directed mutagenesis because of its sequence conservation among the RH-RhoGEFs, its contact with RhoA-Arg-68, and its involvement in a hydrogen bond network with the ␣N helix (Fig. 3b). The nucleotide exchange activity of the LARG-E1023A mutant was diminished to that of the DH domain alone (Fig. 6b and Table II). As expected, the E1023R mutation, which introduces electrostatic repulsion and steric collisions with both LARG-Arg-986 and RhoA-Arg-68, was even less active than E1023A.
We then assessed the ability of the wild-type DH/PH and DH domains of LARG to catalyze nucleotide exchange on RhoA-FIG. 5. Two novel and GDP-excluding conformations of nucleotide-free RhoA among the LARG DH/PH-RhoA complexes. a, the "B" chain of RhoA in the LARG DH/PH-RhoA tetramer is trapped, presumably by crystal contacts, in a conformation in which the purine-binding pocket is occupied by the 160 -164 loop (containing the SAK motif) of RhoA. The side chain of RhoA-Asp-120, which in nucleotide-bound structures of RhoA forms hydrogen bonds with the N1 and N2 nitrogens of guanine (b), instead appears to support the new conformation of the 160 -164 loop via two backbone hydrogen bonds. The C␣ atom of Ala-161 shifts 3.4 Å, and residues 31-34 from switch 1 move 1-2 Å further away from the nucleotide-binding site with respect to those of the other RhoA subunits of the tetramer. In addition, the Gly-14 carbonyl is flipped such that it occludes the ␤-phosphate binding site. This is equivalent to the flip of the Val-20 carbonyl in EF-Tu when in complex with EF-Ts (51,52). Asn-117 of RhoA (equivalent to Ras-Asn-116, where its mutation leads to a dominant negative defective in nucleotide binding (53)) adopts a conformation not previously observed for RhoA, wherein its side chain is oriented toward the purine-binding pocket (Fig. 5). The equivalent asparagines in Ras (54) and EF-Tu (51, 52) adopt the same conformation. Although the low resolution of the data used for the structure determination might seem to render the assignment of this side chain ambiguous, the previously observed rotamer of Asn-117 is inconsistent with both omit and ͉F o ͉ Ϫ ͉F c ͉ difference Fourier maps. The white mesh cage represents electron density contoured at 1.5 from a 4-fold averaged, solvent-flattened omit map wherein the "B" RhoA subunit was excluded from initial phase calculation and subsequent map averaging (see "Experimental Procedures"). The carbonyl bulge of Gly-14 is clearly interpretable, as is the conformation of the aforementioned loops and the Asn-117 side chain. Inorganic phosphate (P i , purple mesh) was modeled bound to the ␣-phosphate binding site. In prior DH/PH-GTPase structures, anions have instead been observed bound to the ␤-phosphate binding site (8,10). Oxygen atoms are colored red, nitrogens blue, sulfur yellow, and phosphate purple. b, the three other RhoA subunits in the crystals of the LARG DH/PH-RhoA complex more closely resemble GTPases from previous DH/PH complexes, except for the carbonyl of Gly-14 and the side chain of Asn-117. The P i bound to the ␣-phosphate-binding pocket is omitted from this panel for clarity. Black arrows indicate the direction the 160 -164 and switch 1 loops of RhoA move to achieve the conformation shown in panel a. To help indicate the position of the purine, ␣and ␤-phosphate binding sites of RhoA, GDP (gray carbons) was modeled from the structure of Mg 2ϩ -free RhoA-GDP (44). In nucleotide-bound RhoA, the side chain of Glu-120 (from the NKXD motif) forms hydrogen bonds with the N1 and N2 nitrogens of the guanine ring (dashed lines). The beveled gray ellipse highlights a potential steric clash between the Gly-14 carbonyl and the ␤-phosphate of the nucleotide.
R68A, a mutant that had no effect on nucleotide exchange rates catalyzed by Dbs, although mutation of the contacting Dbs PH domain residue (Tyr-889) diminished nucleotide exchange (9). Interestingly, the LARG DH/PH domains could catalyze nucleotide exchange about 4-fold more efficiently on RhoA-R68A than wild-type RhoA (Table II). The molecular basis for this effect could be due to the removal of electrostatic repulsion between the RhoA-Arg-68 and LARG-Arg-986 side chains (Fig. 3b). RhoA-R68A was an equally good substrate for our DH domain fragment of LARG (residues 766 -986, Table II) as it was for the DH/PH domains. These results suggest that at least one role of the LARG PH domain could be to help compensate for unfavorable, yet apparently necessary, contacts between RhoA-Arg-68 and the ␣6/␣N helix of LARG, perhaps by burying additional accessible surface area (e.g. between ␣C and RhoA-Glu-97).
A Potential Protein Docking Site on the LARG PH Domain-In the DH/PH crystals, the PH domain forms a 2-fold crystal-lographic dimer interface that buries 800 Å 2 of surface area. The interface consists of a solvent-exposed hydrophobic patch on the ␤5-␤7 sheet of the PH domain, which includes the side chains of Leu-1086, Phe-1098, Ile-1100, Ala-1107, and Ile-1109 (Fig. 1b). Strikingly, the hydrophobic patch of all four PH domains in the asymmetric unit of the LARG-RhoA crystals form similar non-crystallographic dimer contacts. However, there is no evidence from size exclusion chromatography that the LARG DH/PH domains are dimeric in solution (data not shown). The residues that compose this solvent-exposed patch are highly conserved among the Lbc subfamily RhoGEFs, suggesting a functional role. DISCUSSION The structures of both the GTPase-bound and free states of a DH/PH domain have only previously been reported for Dbs (7,9). Although there were no major conformational differences  (Table II). b, mutation of Glu-1023 eliminates PH domain-assisted nucleotide exchange. The PH domain of LARG assists nucleotide exchange on RhoA, as seen by comparing the curves for DH/PH (OE) versus that of the DH domain (residues 765-986, ) alone. The E1023A mutation (*), which is predicted to disrupt a network of hydrogen bonds with the ␣N helix near the RhoA interface (Fig. 3b), reduces the rate of nucleotide exchange to the level of the DH domain. The E1023R mutation introduces a side chain that creates steric overlap and charge repulsion, and, as expected, reduces the catalytic rate even further. The S1118D mutation (not shown), which was designed to disrupt the contact between RhoA-Glu-97 and the N terminus of the ␣C helix of the PH domain (Fig. 3b), yields a curve that overlaps with the E1023A mutant (see Table II). between the two states of the Dbs DH/PH domain, there is a dramatic conformational change between the DH and PH domains of LARG upon binding RhoA (Fig. 1, a and b). Given the longer ␣6 helix of the LARG DH domain, which extends the PH domain further from the surface of the DH domain than in Dbs (Fig. 1c), it is possible that the LARG PH domain simply has greater conformational freedom when not in complex with RhoA. However, it is striking that the relative orientation of the DH and PH domains of LARG while in complex with RhoA are essentially the same as those of Dbs and Trio-N, even though they are distantly related within in the RhoGEF family ( Fig. 1c) (4). The distinct conformations exhibited by structures of the Sos (5) and Tiam1 (42) DH/PH domains may therefore represent exceptions, and not the rule. A characteristic shared by both LARG and Dbs that may help account for the structural similarity of their DH/PH domains is the fact that residues from their PH domains interact directly with the GTPase substrate. In Dbs RhoGEF, a triad of interacting residues (Dbs-His-814, Dbs-Tyr-889, and RhoA-Asp-65) appears most important for PH domain-assisted nucleotide exchange in vitro (9). Upon superposition, there is no residue in LARG equivalent to Dbs-Tyr-889, although the side chain of LARG-Lys-979 occupies the same approximate position as Dbs-His-814. LARG-Lys-979 does not, however, appear to make an analogous contact with RhoA-Asp-65. These differences may account for the smaller catalytic rate enhancement provided by the LARG PH domain compared with that of Dbs. In LARG, we have instead identified Ser-1118 (of ␣C) and Glu-1023 (of ␤1) as residues in the PH domain whose mutation lead to reduced nucleotide exchange. Because Glu-1023 of LARG only forms a small part of the interface with RhoA (Fig. 3b), its more important functional role could be to stabilize the ␣6/␣N helix in a conformation more competent to bind RhoA (Fig. 3b). An analogous role was proposed for Tyr-889 in the Dbs PH domain (9).
Intersectin, another Lbc subfamily RhoGEF, contains a PH domain that does not facilitate nucleotide exchange on its GTPase substrate in vitro (22). Accordingly, its PH domain does not contact either the DH domain or the GTPase substrate in the intersectin-Cdc42 crystal structure (10). The residues of the LARG PH domain that make specific contacts with RhoA ( Fig. 3b) are conserved in intersectin with the exception of LARG-Glu-1023, which is substituted by serine. This change may be enough to abrogate PH domain-assisted nucleotide exchange in intersectin, just as the LARG E1023A and E1023R mutations abolish it in LARG ( Fig. 6b and Table II).
The LARG DH domain has a novel ␣N1/␣N2 extension that directly interacts with RhoA. Other Lbc subfamily RhoGEFs, such as Lbc and Lfc, are predicted by sequence analysis to have a similar extension (Fig. 2b). We have shown through deletion or point mutation of this extension in LARG that catalytic activity in vitro is diminished when the extension is removed or its structure perturbed ( Fig. 6a and Table II). Based on our data, one possible role for the ␣N1/␣N2 extension may be to fix the conformation of Glu-790 in a manner that allows it to pack optimally against the side chain of RhoA-Tyr-34, a switch 1 residue. Given its influence on the rate of nucleotide exchange (Table II), and its conformational flexibility (Fig. 3a), the ␣N1/ ␣N2 extension could serve as a "switch" that can be perturbed and/or reorganized by contacts with other domains of LARG (e.g. the RH domain) or with other regulatory proteins (e.g. G␣ 13 ). Perhaps not coincidentally, Tyr-174 of the RhoGEF Vav, which is phosphorylated by Src family kinases to relieve autoinhibition (43), is equivalent by sequence alignment to Trp-769 of LARG (Fig. 2b). If the ␣N1/␣N2 extension is indeed such a "switch," then nature will have delegated regulation of RhoGEF activity to the same small region N-terminal to the DH domain in even as divergently related enzymes as LARG and Vav.
As predicted by extensive analysis of the Dbs-Cdc42, Dbs-RhoA, Tiam1-Rac1, and intersectin-Cdc42 complexes (10), structural elements that dictate the specificity of LARG for RhoA are primarily found in the ␣4-␣5 loop region of the DH domain (Fig. 4). Salt bridges between LARG-Arg-923 and the N-terminal region of the GTPase (Fig. 4a) help select for RhoA, but they are not sufficient to select against Cdc42, because Dbs has Lys-758 in the equivalent position and can utilize both Cdc42 and RhoA as substrates. Accommodation of RhoA-Trp-58 (Phe-56 in Cdc42) will help LARG discriminate against Cdc42, but not Rac1, which also has tryptophan at the equivalent position (42). However, the ␣4-␣5 loops of LARG, Tiam1, and Dbs are not structurally equivalent, and multiple interactions between these loops and the GTPase are likely responsible for additional selectivity. For example, RhoA-Val-43, which is buried by the ␣4-␣5 loop of LARG (Fig. 4b), is substituted by serine in Rac1 and alanine in Cdc42.
In all reported DH/PH-GTPase structures, the GTPase assumes a conformation that is incompatible with GTP binding due to a reorganization of switch 2 that disrupts its Mg 2ϩ and ␥-phosphate binding sites. However, this conformation is not necessarily GDP-exclusive. The Mg 2ϩ -free structure of RhoA in complex with GDP (44) demonstrated that RhoA can bind GDP even when switch 2 adopts the same conformation as it does when bound to a DH domain. Furthermore, LARG-RhoA crystals can be grown in the presence of GDP, and strong difference density corresponding to GDP can be observed bound in the RhoA active site. 2 Therefore, the novel changes we observe in the P loop and the purine-binding pocket of the LARG-RhoA complex (Fig. 5) probably reflect the natural plasticity of the GTPase active site when nucleotides are absent, rather than an active exchange mechanism exerted by the LARG DH domain. This idea is further supported by the fact that no part of the LARG DH domain comes into obvious direct or indirect contact with either the P loop or the NKXD and SAK motifs of the GTPase. It remains to be shown whether the DH domainmediated exchange mechanism requires RhoA to assume distinct nucleotide-free, GDP-bound, and GTP-bound states while bound to the DH domain, as has been proposed for Cdc25 on Ras (45) or Sec-7 on Arf (46). Our preliminary data indicate that at least the DH/PH-RhoA and DH/PH-RhoA-GDP states occur for LARG, and that their structures will be quite similar.
Finally, the PH domain of LARG, and perhaps all Lbc subfamily RhoGEFs, has an exposed hydrophobic patch (Fig. 1b) that could interface with other domains of LARG or other regulatory proteins. If the LARG PH domain docks with the plasma membrane in the same general orientation as other well characterized PH domains (e.g. PLC-␦1 (47)), this patch would be able to interact laterally with other peripheral membrane proteins/domains. Because the PH domain of Lfc binds specifically to tubulin (14), and that of Lbc to actin (15), one intriguing possibility is that this conserved hydrophobic patch could be involved in targeting these RhoGEFs to the cytoskeleton. Other actin and microtubule binding domains likewise feature a solvent-exposed hydrophobic patch that is thought to be important for filament binding (48,49). A better understanding of the roles of this putative protein-docking site and the ␣N1/␣N2 extension of LARG will be facilitated by determining atomic structures of larger fragments of LARG that include its RH domain and by evaluating the effects of point mutants that perturb these structural elements on the function of full-length LARG in vivo.