Multiple Factors Confer Specific Cdc42 and Rac Protein Activation by Dedicator of Cytokinesis (DOCK) Nucleotide Exchange Factors*

DOCK (dedicator of cytokinesis) guanine nucleotide exchange factors (GEFs) activate the Rho-family GTPases Rac and Cdc42 to control cell migration, morphogenesis, and phagocytosis. The DOCK A and B subfamilies activate Rac, whereas the DOCK D subfamily activates Cdc42. Nucleotide exchange is catalyzed by a conserved DHR2 domain (DOCKDHR2). Although the molecular basis for DOCKDHR2-mediated GTPase activation has been elucidated through structures of a DOCK9DHR2-Cdc42 complex, the factors determining recognition of specific GTPases are unknown. To understand the molecular basis for DOCK-GTPase specificity, we have determined the crystal structure of DOCK2DHR2 in complex with Rac1. DOCK2DHR2 and DOCK9DHR2 exhibit similar tertiary structures and homodimer interfaces and share a conserved GTPase-activating mechanism. Multiple structural differences between DOCK2DHR2 and DOCK9DHR2 account for their selectivity toward Rac1 and Cdc42. Key determinants of selectivity of Cdc42 and Rac for their cognate DOCKDHR2 are a Phe or Trp residue within β3 (residue 56) and the ability of DOCK proteins to exploit differences in the GEF-induced conformational changes of switch 1 dependent on a divergent residue at position 27. DOCK proteins, therefore, differ from DH-PH GEFs that select their cognate GTPases through recognition of structural differences within the β2/β3 strands.

DOCK (dedicator of cytokinesis) guanine nucleotide exchange factors (GEFs) activate the Rho-family GTPases Rac and Cdc42 to control cell migration, morphogenesis, and phagocytosis. The DOCK A and B subfamilies activate Rac, whereas the DOCK D subfamily activates Cdc42. Nucleotide exchange is catalyzed by a conserved DHR2 domain (DOCK DHR2 ). Although the molecular basis for DOCK DHR2 -mediated GTPase activation has been elucidated through structures of a DOCK9 DHR2 -Cdc42 complex, the factors determining recognition of specific GTPases are unknown. To understand the molecular basis for DOCK-GTPase specificity, we have determined the crystal structure of DOCK2 DHR2 in complex with Rac1. DOCK2 DHR2 and DOCK9 DHR2 exhibit similar tertiary structures and homodimer interfaces and share a conserved GTPase-activating mechanism. Multiple structural differences between DOCK2 DHR2 and DOCK9 DHR2 account for their selectivity toward Rac1 and Cdc42. Key determinants of selectivity of Cdc42 and Rac for their cognate DOCK DHR2 are a Phe or Trp residue within ␤3 (residue 56) and the ability of DOCK proteins to exploit differences in the GEF-induced conformational changes of switch 1 dependent on a divergent residue at position 27. DOCK proteins, therefore, differ from DH-PH GEFs that select their cognate GTPases through recognition of structural differences within the ␤2/␤3 strands.
Rho GTPases are critical regulators of cell motility, polarity, adhesion, cytoskeletal organization, proliferation, gene expression, and apoptosis (1)(2)(3). Conversion of these biomolecular switches to the activated GTP-bound state is controlled by two families of guanine nucleotide exchange factors (GEFs) 2 (4 -6). DH-PH proteins are a large group of Rho GEFs comprising a catalytic Dbl homology (DH) domain with an adjacent PH domain within the context of functionally diverse signaling modules. The evolutionary distinct and smaller family of DOCK proteins activate either Rac or Cdc42 to control cell migration, morphogenesis, and phagocytosis (7,8). DOCK proteins are key switches in the plasticity of tumor cell movement, with DOCK3 regulating mesenchymal modes of motility and DOCK10 controlling amoeboid motility (9,10).
DH-PH GEFs often stimulate nucleotide exchange on multiple GTPases in vitro, in contrast to the DOCKs that exhibit very high specificity. In humans, the 11 DOCK proteins are organized into four subfamilies encoding proteins of ϳ2000 amino acids (7,8). The DOCK A (DOCK1/DOCK180, DOCK2, and DOCK5) and DOCK B (DOCK3 and DOCK4) subfamilies activate Rac, whereas the DOCK D subfamily (DOCK9/Zizimin1, DOCK10, and DOCK11) activates Cdc42 (11)(12)(13)(14). Although DOCK6 of the DOCK C subfamily has been implicated as a dual-specificity GEF with activity toward both Rac and Cdc42 (15), other studies detected no activity to any tested Rho GTPase (14). All DOCK proteins contain a catalytic DHR2 domain (also termed the CZH2 or DOCKER domain) of ϳ500 residues situated within their C-terminal regions (11,13). The DHR2 domain is divergent across the family, with the DHR2 domains of DOCK1 (Rac-specific) and DOCK9 (Cdc42-specific) sharing only 22% sequence identity (12,13). A second region of common similarity is the ϳ200 residue DHR1/CZH1 domain located toward the N terminus. The DHR1 domain of DOCK1 adopts a C2-like architecture and interacts with PtdIns(3,4,5)P 3 to mediate signaling and membrane localization (16 -19). DOCK subfamilies are also defined by variations of their constituent domains. The DOCK A and B subfamilies incorporate an N-terminal SH3 domain and an extreme C-terminal polyproline sequence. The SH3 domain and neighboring ␣-helical region mediate their interactions with ELMO subunits (20). In contrast, the DOCK D subfamily incorporates an N-terminal PH domain, whereas the DOCK C subfamily lacks recognizable SH3 and PH domains.
The regulation of DOCK proteins and the signal transduction events responsible for their activation are poorly understood. The Crk and p130 Cas adaptor proteins (9,10,21) are involved in assembling protein complexes that mediate signaling from integrins to DOCK A and B proteins (22)(23)(24). In the A and B subfamilies, GEF autoinhibition is mediated by intramolecular interactions involving the N-terminal SH3 domain and DHR2 (25). Although differing in molecular mechanisms, the N-terminal domains of DOCK9 also markedly suppress DHR2 catalytic activity (12,15,26). Additionally, for DOCK1 and 2, stimulation of Rac-induced cytoskeletal reorganization is dependent on interactions with Engulfment and Cell Motility protein (27)(28)(29)(30). Disruption of DOCK1-ELMO interactions, although not affecting DOCK1 GEF activity in vivo, abrogated the ability of DOCK1 to promote Rac-dependent cytoskeletal changes, indicating that ELMO acts as an adaptor molecule that couples Rac to specific downstream effectors (20).
Recently, the molecular basis for GEF activity of DOCK proteins was elucidated from structures of DOCK9 DHR2 in complex with its cognate GTPase Cdc42 (31). This study showed that GDP release and discharge of the activated GTP-bound Cdc42 are catalyzed by a universally invariant valine residue that functions as a nucleotide sensor. However, the factors conferring Rho GTPase specificity require a comparative study of Cdc42-and Rac-specific DOCKs. Here, we have determined the crystal structure of DOCK2 DHR2 in complex with Rac1. We show that in contrast to specificity mechanisms of DH-PH proteins, multiple differences between DOCK2 DHR2 and DOCK9 DHR2 account for their selectivity toward Rac1 and Cdc42. The selectivity of DOCK A and B for Rac and DOCK D for Cdc42 is conferred mainly by residues at two sites of the GTPase: 1) a Phe or Trp at position 56 of the ␤3 strand, and 2) an Ala or Lys at residue 27 of switch 1 that permits DOCKinduced conformational differences between Cdc42 and Rac1 within switch 1. Our analysis explains why DOCK A and B only activate Rac, and why DOCK D is specific for Cdc42, and suggests that DOCK C may lack the capacity to bind any of the 20 members of the Rho GTPase family.

EXPERIMENTAL PROCEDURES
Cloning and Mutagenesis-Definition of the N terminus of DOCK2 DHR2 was based on a multiple sequence alignment of DOCK family proteins combined with previous DOCK2-Rac1 binding data and DOCK2 GEF activity (13,14,31). DOCK2 DHR2 was amplified by PCR from a human cDNA library and cloned into the pOPIN-SUMO* vector as described (31). Similarly, Rac1 was amplified from a human cDNA library and cloned with an N-terminal His 6 tag into the pOPIN plasmid. Cdc42 was cloned into pGEX6P. Putative specificity determining mutations, L1941M, Q1944N, and Y2008F of DOCK9 DHR2 and mutants of Cdc42(K27A) and Rac1(W56F) were prepared. Rac1 6M (A3T/A27K/G30S/I33V/N52T/W56F) and Cdc42 6M (T3A/K27AS30G/V33I/T52N/F56W) were generated by gene synthesis from Eurofins MWG Operon. The Uracil-specific excision reagent methodology (32,33) was used for all cloning. The expression constructs made were sequenced and confirmed by GATC Biotech.
Protein Expression and Purification-DOCK2 DHR2 and Rac1 were expressed and purified separately, and a DOCK2 DHR2 -Rac1 complex, formed by incubating a 3-fold excess of nucleotide-free Rac1 to DOCK2 DHR2 , was separated from uncomplexed Rac1 by size exclusion chromatography. All the recombinant proteins were expressed in Escherichia coli B834 (DE3) pRare cells grown overnight at 20°C. To purify DOCK2 DHR2 and mutants of DOCK2 DHR2 and DOCK9 DHR2 , cell pellets were resuspended in five volumes of ice-cold lysis buffer (LB) (50 mM TRIS (pH 8.0), 350 mM NaCl, 2 mM imidazole, 2 mM ␤-mercaptoethanol, 10% (v/v) glycerol). The lysed extract was loaded onto Talon resin (Novagen), with the N-terminal His-SUMO tag cleaved from the eluted protein by Tobacco Etch Virus protease. His-tagged TEV protease and the His 8 -SUMO tag were removed by passage through a second Talon resin column. Protein was further purified by size exclusion (S300) and Source Q anion exchange chromatography (GE Healthcare). His 6 -Rac1 was purified by subjecting the lysate to Talon resin and then to the size exclusion column (S300). GST-Cdc42 (residues 1-188) was purified using glutathione superflow (Generon) followed by removal of the GST tag using PreScission protease. Cleaved Cdc42 was further purified by gel filtration. In the purification of all proteins, to remove nucleotide, 5 mM EDTA was added to the buffer in the size exclusion purification step. To prepare the DOCK2 DHR2 -Rac1 complex, DOCK2 DHR2 and Rac1 were incubated at a 1:2 molar ratio on ice for 30 min. The DOCK2 DHR2 -Rac1 complex was separated from free Rac1 by size exclusion on an S300 gel filtration column. The complex was further purified by ion exchange using Source Q.
Fluorescence Kinetics Analysis-Fluorescence experiments were performed using an Omega-FLUORostar Plate reader. Crystallization and Structure Determination-Crystals of the DOCK2 DHR2 -Rac1 complex were obtained using the hangingdrop vapor diffusion method. Briefly, 1 l of protein solution (4 mg/ml) was mixed with 1 l of mother liquor (0.1 M MES (pH 6.5), 12% (w/v) PEG 3350, 10% (v/v) glycerol, 150 mM NaCl) and allowed to equilibrate at 20°C. Crystal quality was improved by addition of 5% (v/v) 3-(1-Pyridino)-1-propane sulfonate-201 to the protein complex prior to crystallization. Crystals were flash-frozen in liquid N 2 using 25% (v/v) glycerol in the mother liquor as a cryoprotectant. X-ray diffraction data were collected at 100 K at Diamond Light Source beam line I02. The best crystals diffracted to a Bragg spacing of 2.7 Å with space group P2 1 . All data were processed using iMosflm (34) and scaled using SCALA (34). Initial phases were obtained by molecular replacement using the PHASER program (35) program. DOCK9 DHR2 coordinates from the DOCK9 DHR2 -Cdc42 structure (PDB code 2WM9) (31) and Rac1 coordinates from the Rac1-GMPPNP structure (PDB code 1MH1) (36) were used as search models. For molecular replacement, a resolution cutoff of 3.5 Å was used. The Autobuild module of PHENIX (37) was used to build the secondary structural elements of DOCK2 DHR2 . Several rounds of manual model building were carried out with COOT (38), and the structure was refined with PHENIX 6 . Non-crystallographic symmetry restraints and translation/liberation/screw rotation parameters (generated from the TLSMD server) were used throughout the refinement. Water molecules were added toward the end of the refinement. The structure was validated with MolProbity (39). Data collection and refinement statistics are given in supplemental Table 1. A representative 2Fo-Fc electron density map is shown in supplemental Fig. 5.
The model was generated using a SWISS-MODEL (40) based on alignments shown in supplemental Fig. 1, and energy was minimized with CNS 1.3 (41).
Modeling Rho GTPase and GEF Interactions-Residues at the 27th, 54th, and 56th positions of Rho GTPases in the DOCK2 DHR2 -Rac1 and DOCK9 DHR2 -Cdc42 structures were modeled with those corresponding to 20 Rho GTPases listed in supplemental Fig. 4 using COOT. All rotamers of each residue were used in the analysis. A similar analysis was also carried out for the DOCK6 DHR2 model.

The Overall Structure of DOCK2 DHR2 Resembles DOCK9 DHR2
Despite their low sequence identity (22% compared with 71% for Cdc42 and Rac1), DOCK2 DHR2 and DOCK9 DHR2 adopt similar architectures, comprising three lobes of nearly equal size (lobes A, B, and C) (Figs. 1 and 2 and supplemental Fig. 1 and Table 1) (31). The GTPase-binding site and catalytic center of DOCK DHR2 are generated entirely from lobes B and C (Figs. 1 and 2). Lobe A of DOCK2 DHR2 adopts a Tetratricopeptide repeat-like fold of six antiparallel ␣-helices (one additional ␣-helix relative to DOCK9 DHR2 ) (supplemental Fig. 2). Lobe A of DOCK2 DHR2 also differs from its counterpart in DOCK9 DHR2 because it lacks the small antiparallel ␣-helical hairpin inserted between ␣2 and ␣3, a region of hypervariability within the DOCK family (supplemental Figs. 1 and 2B). Helices ␣4 and ␣5 of lobe A generate the homodimeric interface, previously observed in DOCK9 DHR2 , confirming the prediction that all DOCK DHR2 domains self-associate through a conserved dimerization mechanism (31) and in agreement with findings that the closely related DOCK A subfamily member DOCK1 is an oligomer (19,42).
Lobe B comprises two anti-parallel ␤-sheets related in a loosely packed orthogonal arrangement. Small differences with DOCK9 DHR2 include loss of an edge ␤-strand of sheet1 (␤3 of DOCK9 DHR2 ) and the replacement of the two ␣-helices of DOCK9 DHR2 connecting ␤3 and ␤4 with a single helix in DOCK2 DHR2 (helix ␣7) (supplemental Figs. 1 and 2C). Lobe C of the two proteins adopts an almost identical antiparallel ␣-helical bundle composed of four ␣-helices. Helix ␣10 of lobe C is interrupted by a seven-residue insert incorporating the essential and universally conserved Val residue (Val-1540, DOCK2 numbering, corresponding to Val-1951 in DOCK9) that functions as a nucleotide sensor (31) (supplemental Fig. 2D).

The Dimerization Interface of DOCK Proteins Is Structurally Conserved
Similar to DOCK9 DHR2 , the DOCK2 DHR2 homodimer interface is mediated almost entirely by the anti-parallel interactions of the ␣4 and ␣5 helices of lobe A with their symmetry-related counterparts (Fig. 1). Contacts at the subunit interface involve a mixture of aromatic and aliphatic residues contributed by symmetrically arranged ␣5 helices, augmented by the more loosely packed ␣4 helices that interact via electrostatic interactions, with the ␣4/␣5 turn capped by ␣7Ј (Fig. 1B). Although DOCK2 DHR2 and DOCK9 DHR2 share similar homodimer interface residues (Fig. 1A and supplemental Fig. 1), differences between the two proteins suggest that DOCK2 DHR2 -DOCK9 DHR2 heterodimerization is unlikely. The relative packing of ␣4 and ␣5 with their dimer counterparts differ between the two proteins, altering their overall quaternary structure. Because of this difference in quaternary structure and the longer ␣7 helix of DOCK2 DHR2 , heterodimerization would be prevented by a clash of ␣7 of DOCK2 DHR2 with the ␣4Ј-␣5Ј turn of a symmetry-related DOCK9 DHR2 subunit.

General Comparison of the DOCK2 DHR2 -Rac1 and DOCK9 DHR2 -Cdc42 Complexes
Equivalent C␣ atoms of DOCK2 DHR2 and DOCK9 DHR2 superimpose within a root mean square deviation of 4.5 Å (supplemental Fig. 2A and results), a large deviation that results from differences in the relative orientations of the three lobes in DOCK2 DHR2 relative to DOCK9 DHR2 . Superimposing individual lobes yields smaller deviations (supplemental results). With the exception of the switch 1 region (discussed below), Rac1 and Cdc42 adopt similar conformations in their respective complexes, (root mean square deviation of 1.4 Å, Fig. 2A and supplemental results). In Cdc42 and Rac1, compared with their nucleotide-bound states, pronounced conformational changes occur to switch 1 (residues 25-39) and ␤2/␤3 strands of Cdc42 with little change elsewhere (supplemental Fig. 3) (43)(44)(45)(46). Overall, equivalent residues of both DOCK2 DHR2 and DOCK9 DHR2 interact with their cognate GTPase, with the majority of contacts localized to ␣10 of lobe C (Fig. 2). This is reflected in the close structural correspondence of lobe C of DOCK2 DHR2 and DOCK9 DHR2 when Rac1 and Cdc42 are superimposed (supplemental Table 2), indicating that the lobe C-GTPase interface is structurally conserved throughout the DOCK DHR2 family.

A Conserved Mechanism of Nucleotide Exchange in the DOCK Family
DOCK2 DHR2 catalyzes nucleotide exchange in Rac1 through a similar mechanism to DOCK9 DHR2 -mediated activation of Cdc42. On binding to DOCK DHR2 , conformational changes are confined to switch 1 of the GTPase (supplemental Fig. 3). Nucleotide release is promoted through a concerted mechanism of distorting and opening switch 1 (removing Phe-28 from its interactions with the nucleotide guanine moiety) and projecting the aliphatic side chain of Val-1540 (Val-1951 of DOCK9) into the Mg 2ϩ -binding site of the GTPase, thus displacing Mg 2ϩ from bound nucleotide (Fig. 2). In DOCK9 DHR2 , in contrast to DOCK2 DHR2 , movement of switch 1 is linked to rotation of the P-loop Cys-18 C (the superscripts "C" and "R" refer to residues in Cdc42 and Rac1) thiol group that disrupts a hydrogen bond with the nucleotide ␣-phosphate. In DOCK9 DHR2 -Cdc42, rotation of the Cys-18 C thiol group is linked to rotation of the switch 1 Tyr-32 C side chain. However, the difference in the switch 1 conformation of Rac1 allows Tyr-32 R to maintain its hydrogen bond to Cys-18 R , and Cys-18 R does not alter its conformation relative to the Rac1-nucleotide-bound state.

DOCK2 DHR2 and DOCK9 DHR2 Induce Different Conformational Changes in Their Cognate GTPase Switch 1 Regions
A notable conformational difference between Cdc42 and Rac1 induced by their respective DOCK DHR2 proteins, confined to the switch 1 region (residues 27 to 35), contributes to the selectivity of DOCK2 DHR2 and DOCK9 DHR2 for Rac1 and Cdc42, respectively. Thus, DOCKs differ from DH-PH proteins that induce virtually identical conformations of switch 1 and switch 2 in their cognate GTPases. The different orientation of the B and C lobes of DOCK2 DHR2 results in a small displacement (ϳ3 Å) of lobe B away from Rac1 compared with lobe B of DOCK9 DHR2 relative to Cdc42 (Figs. 2 and 3). In DOCK2 DHR2 , the ␤3/␤4 loop is withdrawn by ϳ4 Å relative to DOCK9 DHR2 . This accommodates an extended conformation of the N-terminal region of the Rac1 switch 1 incorporating Phe-28. Additionally, the short ␤-hairpin of the ␤5/␤6 loop of DOCK9 DHR2 is absent from DOCK2 DHR2 , hence ␤5/␤6 is also more extended in DOCK2 DHR2 and, unlike DOCK9 DHR2 , does not contact Tyr-32 R of switch 1 (Figs. 2 and 3). The buried solvent-accessible surface area at the DOCK2 DHR2 -Rac1 interface at 1744 Å 2 is less than that at the DOCK9 DHR2 -Cdc42 interface (2146 Å 2 ) but ϳ30% larger than at DH-PH protein-GTPase interfaces. JULY 15, 2011 • VOLUME 286 • NUMBER 28

Multiple Factors Account for DOCK DHR2 Specificity for Rac and Cdc42
Unlike DH-PH domain GEFs that can often stimulate nucleotide exchange on multiple GTPases in vitro, DOCK GEFs exhibit very high specificity, and there are no confirmed reports of multispecific DOCKs. Comparing the DOCK2 DHR2 -Rac1 and DOCK9 DHR2 -Cdc42 complexes provides a rationale for the specificity of DOCK2 and DOCK9 for their cognate GTPases. The regions of Cdc42 and Rac1 that bind to DOCK DHR2 are mainly confined to the N-terminal 74 residues, with a small contribution provided by the C-terminal ␤6/␣5 loop ( Fig. 2 and  supplemental Fig. 4). The DOCK DHR2 contact residues that differ between Rac1 and Cdc42 are confined to the N-terminal 56 residues and comprise only six residues located in ␤1, switch 1, and ␤3 (supplemental Table 2). These six divergent residues and associated differences in switch 1 conformation are responsible for conferring DOCK-GTPase specificity, detailed in the supplemental results and Fig. 4. Compared with Cdc42, Rac1 binding to DOCK9 DHR2 would be unfavorable because of loss of three hydrogen bonds, and the generation of steric clashes from the replacement of Val-33 C and Phe-56 C with the more bulky Ile-33 R and Trp-56 R residues, respectively. Conversely, Cdc42 cannot be activated by DOCK2 because of steric clashes involving Thr-3 C (instead of Ala-3 R ), whereas Phe-56 C instead of Trp-56 R would form less optimal van der Waals interactions with lobe C of DOCK2 DHR2 , including an unfavorably close contact with the C␥-atom of Met-1529 (Fig. 4E). Finally, because of the different switch 1 conformations, Lys-27 C , as in Cdc42 (Ala in Rac), could not be accommodated in DOCK2 DHR2 -bound Rac1 because of intramolecular steric clashes (Fig. 4F).

Switching Cdc42 and Rac1 Specificities for DOCK2 DHR2 and DOCK9 DHR2
Cdc42 and Rac1 Specificities-To test our model for the structural basis of DOCK DHR2 specificity, we generated mutant Cdc42 and Rac1 proteins (Cdc42 6M and Rac1 6M ), where the six residues predicted to determine the specific response to either DOCK2 DHR2 or DOCK9 DHR2 were exchanged (supplemental Table 2). Thus, Cdc42 6M and Rac1 6M would be expected to have the capacity to be specifically activated by DOCK2 DHR2 and DOCK9 DHR2 , respectively. Fig. 5 shows that although Cdc42 is not a substrate of DOCK2 DHR2 , Cdc42 6M is strongly activated by DOCK2 DHR2 (50% of Rac1) but is virtually unresponsive to DOCK9 DHR2 . Conversely, Rac1 6M is a substrate of DOCK9 DHR2 (40% of Cdc42) but is not activated by DOCK2 DHR2 . Thus, we have reciprocally exchanged the selectivity of both GTPases for DOCK2 DHR2 and DOCK9 DHR2 .
Analysis of the capacity of DOCK2 DHR2 and DOCK9 DHR2 to activate single site mutants of Rac1 and Cdc42 identifies key specificity-determining residues at position 27 of switch 1 and position 56 of ␤3. Replacing Trp-56 for Phe, as in Cdc42, allows Rac1(W56F) to be weakly activated by DOCK9 DHR2 (Fig. 5B), indicating that a bulky Trp residue at the Phe-56-binding pocket of DOCK9 DHR2 precludes activation of Rac1. Interestingly, compared with Rac1, Rac1(W56F) is a poor substrate of DOCK2 DHR2 (Fig. 5A), probably because of unfavorable contacts between Phe and the Trp-56-binding pocket of DOCK2 DHR2 (discussed below), a finding in agreement with recent data of Wu et al. (47), and is reminiscent of the ablation of activation of Rac1(W56F) by the Rac-specific DH-PH protein Tiam (47)(48)(49). The ability of either a Phe or Trp at position 56 to modulate the sensitivity of DOCKs for Rac and Cdc42 is also reminiscent of the striking activation of Rac1(W56F) by the Cdc42-specific DH-PH protein Intersectin (49), indicating a detrimental consequence of Trp-56 for Cdc42-specific GEFs. However, in contrast to findings that replacing Phe-56 with Trp renders Cdc42 responsive to the Rac-specific GEFs Tiam, GEF-H1, and TrioN (47)(48)(49), a previous study showed that Cdc42(F56W) was not activated by DOCK1 DHR2-C (47). Thus, factors additional to a Phe at position 56 are responsible for the inability of DOCK2 DHR2 to activate Cdc42.
A second key factor determining specificity for Rac-specific DOCKs is the identity of residue 27. Structural comparison of DOCK2 DHR2 -Rac1 and DOCK9 DHR2 -Cdc42 complexes indicates that Lys-27 sterically hinders the conformational change of switch 1 necessary for DOCK2 DHR2 -Rac interactions. Consistent with this notion, other studies showed that Rac2(A27K) is unable to bind DOCK2 (50) and that Rac(A27K) is only weakly activated by DOCK1 DHR2-C (47). However, replacing Lys-27 of Cdc42 with Ala is not sufficient to allow either DOCK2 DHR2 (Fig. 5A) or DOCK1 DHR2-C (47) to activate Cdc42, indicating that additional features, specifically Phe-56, as indicated by activation of the double mutant Cdc42(K27A, F56W) by DOCK1 DHR2-C (47), prevent Rac-specific DOCKs from activating Cdc42.
DOCK DHR2 Specificities-Because of the low sequence conservation between DOCK2 DHR2 and DOCK9 DHR2 , multiple tertiary structural differences account for the distinctive specificities of DOCK2 DHR2 and DOCK9 DHR2 toward Rac1 and Cdc42, respectively. Thus, rational efforts to switch their specificities by exchanging a relatively small number of amino acid residues are probably not feasible. Consistent with this prediction, attempts to confer Rac1 exchange factor activity on DOCK9 DHR2 by modifying the Phe-56-binding pocket were unsuccessful. Substituting Met (as in DOCK2 DHR2 ) for Leu-1941 in DOCK9 DHR2 to interact with the Trp residue of Rac1 (Fig. 4E) did not confer Rac1 GEF activity on DOCK9, even at 10-fold and 5-fold higher concentrations of DOCK9 DHR2 (L1941M) and Rac1, respectively (data not shown). Because DOCK9 DHR2 (L1941M) was 4-fold less active toward Cdc42, a Met at residue 1941 selects against Phe-56, and our structures show that the C␥-atom of Met (instead of Leu) would form unfavorable van der Waals contacts with the aromatic ring of Phe-56 C (Fig. 4E). This observation also accounts for the low activation of Rac1(W56F) by DOCK2 DHR2 (Fig. 5) and DOCK1 DHR2-C (47) and explains why replacing Met with Leu in DOCK1 DHR2-C decreased the GEF activity of DOCK1 DHR2-C toward Rac1 by 80% (47). Interestingly, the C␥-atom of Met (Met-1529 of DOCK2) forms favorable contacts with a Trp residue, not possible with a Leu side chain (Fig.  4E). Significantly, this indicates that a either a Leu or Met at residue 1941 (DOCK9 numbering) is necessary but not sufficient to confer optimal activity toward Cdc42 and Rac, respectively.
A triple mutation of the Phe-56-binding pocket of DOCK9 DHR2 to mimic the Trp-56-binding site of DOCK2 DHR2 (L1941M, A1944N, Y2008F) did not confer activity toward Rac1 but completely abolished activity toward Cdc42 (data not shown). This result highlights the importance of favorable van der Waals contacts between DOCK and GTPase, and is reminiscent of mutations of the Phe-56-binding pocket of Intersectin that reduce activity for Cdc42 (51). However, in contrast to the DOCKs, the Intersectin mutation confers activity toward Rac1. Thus, the DOCK family differs from DH family GEFs, where a relatively small number of residues were identified as being responsible for conferring specificity toward Cdc42, Rac, and Rho (51).

CONCLUSIONS
Our structural analysis, supported by biochemical data, reveals that multiple factors determine the specificity of DOCK proteins for their cognate GTPase. Two sites on the GTPase confer dominant roles in defining specificity for Rac and Cdc42. Residue 56 is a Phe, Trp, or Tyr in 19 of 20 Rho GTPases (supplemental Fig. 4 Fig. 4). DOCK9 DHR2 can accommodate Phe but not Trp (Figs. 4 and 5), and modeling shows that Tyr would experience steric clashes at the Phe-56-binding pocket of DOCK9 DHR2 . Trp and Tyr at residue 56 precludes activation by Cdc42-specific DOCKs, which therefore can only activate Cdc42 (supplemental Table 3). Rac-specific DOCKs, although preferring Trp, can also accommodate Phe (Fig. 5). Thus, Rac(W56F) is activated by both Cdc42-and Rac-specific DOCKs. A critical difference between DOCK and DH-PH GEFs is that switching between a Phe and Trp at residue 56 is sufficient to toggle the Cdc42/Rac specificity of Rac and Cdc42specific DH-PH GEFs (47)(48)(49). The conformation of switch 1 of Rac bound to DOCK2 DHR2 is compatible with Ala at residue 27, whereas large side chains would sterically hinder this conformation. Thus, non-Ala amino acids tend to select against DOCK2, whereas DOCK9 is permissive for all residues at residue 27.
RhoA is not activated by either DOCK2 DHR2 or DOCK9 DHR2 (11)(12)(13)(14). The more bulky Ala at residue 54 of RhoA compared with Gly of Cdc42 and Rac would possibly sterically hinder the interaction of Rho with DOCK2 DHR2 , analogous to the inability of Tiam1 and Intersectin to activate Rho (49). This prediction is consistent with a recent study showing that DOCK1 DHR2-C was unable to activate a G54A Rac mutant (47). Additional factors would prevent RhoA activation by DOCK GEFs. In the instance of DOCK9, Trp-58 of RhoA (equivalent to Phe-56 of Cdc42 and Trp-56 of Rac) would undergo a steric clash with Val-1941 and Gln-1944 of the Cdc42 Phe-56-binding pocket, as predicted for Rac. Furthermore, our modeling study shows that Gln-29 of Rho (equivalent to Lys-27 of Cdc42 and Ala-27 of Rac1) would introduce internal steric clashes with other switch 1 residues when bound to either DOCK2 DHR2 or DOCK9 DHR2 (supplemental Table 3). Because the Trp-56-binding pocket of DOCK C is similar to that of DOCK9 DHR2 , DOCK C-family GEFs would also be incapable of binding RhoA.
Our comparative analysis of the DOCK2 DHR2 -Rac1 and DOCK9 DHR2 -Cdc42 complexes allows us to define a set of rules and apply this to generate a matrix to predict which of the 20 Rho GTPases have the capacity to bind DOCK DHR2 domains (supplemental Table 3 3). This analysis shows that Rac-specific DOCKs (DOCK A and B subfamilies), in addition to their Rac GEF activity, may also have the capacity to bind and activate RhoH. The DOCK D subfamily, on the other hand, is specific for Cdc42. Furthermore, the Phe-/Trp-56-binding pocket of DOCK C (Met replacing Leu-1941 of DOCK9) is incompatible with Phe, Tyr, and Trp, suggesting that the DOCK C subfamily lack the capacity to bind and hence activate any Rho GTPase. We found that DOCK6 DHR2 expressed in E. coli and the insect cell system failed to activate either Cdc42 or Rac1. However, the conservation of the essential catalytic Val residue of the nucleotide sensor in DOCK C proteins (supplemental Fig. 1), a hallmark of a catalytically active DOCK GEF (31), suggests that DOCK C proteins may stimulate another member of the small GTPase family.