INTRODUCTION

Top Previous Next References

The establishment of cell polarity is an important component of the overall process of cellular morphogenesis, the complex process by which the three-dimensional organization of subcellular constituents, which ultimately determines an organism's characteristic growth patterns and shape, is generated and maintained. The generation of cell polarity is critical for the control of many cellular and developmental processes such as shape development in early plant and animal embryogenesis, axon migration and neurite outgrowth in early development, the intracellular movement of organelles and proteins in polarized epithelial cells, the stimulated secretion of neurotransmitters, the directed movement of migratory cells, polarized growth within yeast and fungal cells, and the asymmetric partitioning of new cellular constituents during cell division. Establishment of cell polarity involves the generation of cellular asymmetry through the localized temporal and spatial activation of cellular processes and can be divided into several hierarchical and interdependent events. These events include the initial response to endogenous and/or exogenous signals, the determination of an axis of polarization relative to these signals, and the subsequent asymmetric distribution of cellular components along that axis. At the molecular level, cell polarity is best understood in the budding yeast Saccharomyces cerevisiae, but results from studies in fission yeast Schizosaccharomyces pombe, Drosophila, Caenorhabditis elegans, and cultured mammalian cells strongly suggest that the molecular mechanisms controlling cell polarity in S. cerevisiae are highly conserved in other eukaryotes. Due to the abundance of recent reviews on cell polarity and signaling (26, 69, 77, 100, 126, 170, 177, 202, 216, 218, 246, 281, 346, 347, 350, 382, 464, 465, 467, 511, 520, 587) and on the roles of Rho-type GTPases in these processes (46, 61, 120, 141, 171, 186, 192, 225, 241, 288, 290, 325, 326, 337, 344, 419, 441, 473, 477, 478, 544, 550, 552, 572, 578, 638, 644). I will limit this review to a discussion of the Cdc42p GTPase, its identification, its structure and subcellular localization, its function(s) in controlling cell polarity, and its regulators and effectors.

It is becoming increasingly apparent that the Cdc42p GTPase and other Rho-type GTPases play a vital role in regulating the signal transduction pathways that control the generation and maintenance of cell polarity in many, if not all, eukaryotic cell types. The Cdc42p GTPase signaling module consists of regulators of the guanine nucleotide-bound state of Cdc42p, i.e., guanine nucleotide exchange factors (GEFs), guanine nucleotide dissociation inhibitors (GDIs), and GTPase-activating proteins (GAPs), as well as downstream effectors of Cdc42p function (Table 1). The regulators of the guanine nucleotide-bound state of Cdc42p must respond to a variety of exogenous and/or endogenous signals, thereby activating Cdc42p to a GTP-bound state or inactivating it to a GDP-bound state. A myriad of Cdc42p downstream effectors interact with the activated (GTP-bound) form of Cdc42p, thereby inducing a number of downstream events, including rearrangements of the actin cytoskeletal network and protein kinase-dependent induction of transcription, which are increasingly coming into view. Interactions between the 21-kDa Cdc42p GTPase and this host of regulators and effectors must be controlled in a temporal and spatial manner so that Cdc42p can function at different times within the cell cycle and at different places within the cell. Cdc42p function is also regulated by its subcellular localization, which depends on its prenylation state and interactions with its GDI.

CDC42P STRUCTURE AND FUNCTIONAL DOMAINS

Top Previous Next References

Identification of Cdc42

The Cdc42p GTPase was first identified from an S. cerevisiae mutant strain carrying a temperature-sensitive (ts) mutation, cdc42-1^ts, that blocked bud formation at 37°C but allowed the cell mass and volume to increase, resulting in greatly enlarged, unbudded cells (2). Although cell division was arrested at 37°C, DNA replication and nuclear division continued into the next cycle, resulting in multinucleate cells as determined by DNA staining with the fluorescent dye 4',6-diamidino-2-phenylindole (DAPI) and mitotic spindle staining with anti-tubulin antibodies. Fluorescence microscopy with rhodamine-conjugated phalloidin showed that the polarized organization of the actin cytoskeleton (i.e., cortical actin distribution to the regions of new cell growth in the bud and actin cables directed into the enlarging bud) was disrupted, indicating that Cdc42p functioned in the organization of the actin cytoskeleton, which is necessary for polarized cell growth. Chitin and other cell surface materials were deposited uniformly throughout the enlarging cell walls, in contrast to their normal polarized patterns of deposition. Growth of the cdc42-1^ts strain at semipermissive temperatures led to a small percentage of cells with elongated buds. Taken together, these observations suggested that Cdc42p controls polarized cell growth during the cell cycle but that isotropic incorporation of new cell wall material was not impaired through the loss of Cdc42p function. Examination of the cdc42 null phenotype in S. cerevisiae and S. pombe indicated that Cdc42p was essential for viability (242, 390).

DNA and predicted amino acid sequence analysis (242) indicated that Cdc42p belongs to the Rho subfamily of the Ras superfamily of GTPases that act as molecular switches in the control of a variety of eukaryotic processes (191, 239, 241, 242, 642) (see below). At about the same time, a ~25-kDa guanine nucleotide binding protein was purified from bovine brain and human placental membranes (140, 461, 582), and peptide sequences from this protein, termed G_p or G25K, showed a high degree of similarity to S. cerevisiae Cdc42p (242, 461). This protein was shown to be a good in vitro substrate for epidermal growth factor (EGF)-stimulated phosphorylation (209), although the in vivo phosphorylation of Cdc42p has not been reported to date. Subsequent analysis of the predicted amino acid sequence from two independent human cDNA isolates indicated the existence of two highly conserved (95% identical) proteins, the ubiquitously expressed Cdc42Hs (525) and the brain isoform G25K (407). The Cdc42Hs and G25K proteins are identical in both nucleotide and predicted amino acid sequences up to amino acid 163 but diverge from residues 163 to 191, suggesting that these isoforms are differential splicing products of a single gene. Structural and/or functional Cdc42p homologs have subsequently been characterized in the pathogenic yeast Candida albicans (394), S. pombe (390), C. elegans (88, 500), Drosophila (336), chicken (Gallus gallus) cochlea (172), mouse (Mus musculus) liver (172) and brain (367), and dog (Canus familiaris) (GenBank accession no. Z49944), and these homologs are 80 to 95% identical in predicted amino acid sequence (241) (see below) (Fig. 1). S. pombe, Drosophila, and C. elegans Cdc42p, as well as Cdc42Hs and G25K, can complement the cdc42-1^ts mutant (88, 390, 407, 507, 525), suggesting that Cdc42p may have conserved functions in these other eukaryotes.

View larger version (59K): [in this window] [in a new window]   FIG. 1.   Cdc42p mutations and X-ray crystal structure. (A) Comparison of Cdc42p sequences from S. cerevisiae (Sc [242]), C. albicans (Ca [394]), S. pombe (Sp [390]), C. elegans (Ce [88]), D. melanogaster (Dm [336]), chicken (Gd [172]), dog (Cf [GenBank no. Z49944]), mouse (Mm [172]), human (Hs [525]), mouse brain (Mmb [367]), and human brain (G25K [407]). Included are the known Cdc42 mutations mapped onto the primary amino acid sequences and the known functional domains: GTP binding/hydrolysis domains (blue boxes), effector domain (red box), Rho insert domain (purple box), and membrane localization domain (green box). Secondary-structure elements indicated below the sequence are taken from reference 145. (B) Two views of the X-ray crystal structure of Cdc42Hs complexed with GDP (kindly provided by N. Nassar and R. Cerione, Cornell University). Color coding for functional domains is the same as in panel A.

The Cdc42 family of proteins currently has 11 members ranging in size from 190 to 192 amino acids (Fig. 1A). Within this family, there is a very high degree of sequence conservation, ranging from ~75% amino acid identity between C. albicans Cdc42p and the human brain isoform G25K to 100% identity between the dog, mouse, and human Cdc42p and 100% identity between the mouse brain and human brain (G25K) isoforms. Cdc42 proteins display ~40% similarity to other Ras-like GTPases, but this similarity is clustered in the four domains implicated in GTP binding and hydrolysis (Fig. 1, blue boxed domains). The most obvious difference between Cdc42p and Ras protein sequences in these domains is in amino acids 115 to 118. Ras proteins contain the diagnostic sequence Asn-Lys-Xaa-Asp (NKXD, where X is any amino acid), while all Cdc42p proteins contain the signature sequence Thr-Gln-Xaa-Asp (TQXD, with X being predominantly an Ile residue). It has been postulated that these differences may account for the ~10-fold-higher rate of GTP hydrolysis observed with Cdc42p proteins than with Ras proteins (210), but this has not been experimentally tested to date (see "GTPase-activating proteins" below). All Cdc42 proteins contain the C-terminal sequence Cys-Xaa-Xaa-Leu except for the two brain isoforms (Cdc42Mmb and G25K [Fig. 1]), which end in a Phe residue. This conserved domain is necessary for proper membrane anchorage of Cdc42 proteins (see "Prenylation and subcellular localization" below). Much that is known about the functional domains of Cdc42p has been determined by analyzing gain-of-function, loss-of-function, and dominant negative mutations (shown in Fig. 1). A compendium of these mutations, along with their mutant phenotypes, is listed in Table 2.

The numerous cdc42 mutations analyzed to date (see below) have greatly aided in defining functional domains within Cdc42p. However, without the information derived from the crystal structure of purified Cdc42, these mutations do little to clarify the global structure of Cdc42 and hence the multiple interactions between Cdc42 and its regulators and effectors. This problem has recently been resolved with the determination of the solution structure of Cdc42Hs by nuclear magnetic resonance (NMR) spectroscopy techniques (145), along with the determination of the X-ray crystal structure of Cdc42Hs bound to GDP (Fig. 1B) (413a). Several of the more interesting and informative mutations have been mapped onto the crystal structure (Fig. 1), and they highlight the potential functional domains of Cdc42p. The four domains implicated in the binding and hydrolysis of GTP are highlighted in blue in Fig. 1. The structure of these domains is similar to those found in the Ras and Rac crystal structures, highlighting the conservation of structure and function between different guanine nucleotide binding proteins.

Clearly, one of the more interesting and functionally important domains is the effector or switch I domain between residues 26 and 50 (highlighted in red in Fig. 1) (see "Effector domain" below). This domain forms an extended 2-strand/loop structure covering a large proportion of one face of the molecule. Based on its extended structure, it is easy to see how different effectors or regulators of Cdc42p could bind to different subdomains of the effector domain, possibly at the same time, as suggested by the analysis of different effector domain mutations. One effector/regulator could be bound to the N-terminal proximal domain around residue 35, which is in close proximity to the bound nucleotide, while another could be bound to the N-terminal distal region around residue 44. In addition, binding of an effector protein to this domain could interfere with the binding of other effector/regulator proteins to this domain, thereby providing a basis for the regulation of the myriad of Cdc42-dependent cellular processes.

The domain that makes Rho-type GTPases unique within the Ras superfamily is the so-called Rho insert domain (highlighted in purple in Fig. 1). This extra ~13 amino acids is -helical and has been implicated in Cdc42 interactions with one of its downstream effectors, the IQGAPs (379, 605), as well as its GDIs (605). In studies with a chimeric Cdc42Hs in which the insert domain between residues 120 to 139 was replaced with residues 121 to 127 of Ha-Ras, the resulting Cdc42Hs-L8 protein showed a two- to threefold-reduced affinity for the carboxyl-terminal (97-kDa) half of IQGAP1, which contains the GRD Cdc42 binding domain (379, 605) (see "Cdc42p downstream effectors" below). While this Cdc42Hs-L8 protein did not exhibit altered interactions with Rho-GDI from the perspective of the ability of Rho-GDI to extract Cdc42Hs-L8 from membranes, it had a greatly reduced sensitivity to the Rho-GDI-dependent inhibition of GDP dissociation or GTP hydrolysis (605). These results suggest that the insert domain is mediating some of the effects of the Rho-GDI. Rho-GDI binding to Cdc42Hs requires C-terminal prenylation (307) (see "Mammalian GDIs" below), suggesting that Rho-GDI binds to the C-terminal prenylation domain. The fact that the insert domain is on the other side of the Cdc42 molecule from the C-terminal prenylation domain (Fig. 1B) makes it likely that Rho-GDI binding induces a conformational change in the structure of the insert domain, possibly leading to the insert domain shifting its location to block the guanine nucleotide binding pocket, thereby "locking" the Cdc42 protein in either a GDP- or GTP-bound state and inhibiting GDP dissociation or GTP hydrolysis (see "Mammalian GDIs" below). It is interesting that Ras proteins do not have an insert domain and also do not seem to have physiological interactions with GDI proteins. Recently, the insert domain was shown to play a role in the ability of Cdc42 to transform NIH 3T3 fibroblasts (606). The L8 deletion mutation (see above) could intragenically suppress the transforming ability of the Cdc42^F28L mutant protein without affecting its ability to bind GTP, induce Jun N-terminal kinase (JNK) and p21-activated kinase (PAK) activities, or induce filopodium formation (see "Mammals" under "Functional studies" below). These results further highlight the ability of Cdc42p to differentially function in multiple cellular processes through interactions between its different structural domains and downstream effectors (see "Cdc42p downstream effectors" below).

In addition to the effector domain, Cdc24p and other GEFs interact with Cdc42 through other domains, including residues 82 to 100, which encompass the 4-strand-3-helix region, and residues 140 to 150, which encompass the 4-helix. The 3- and 4-helices lie on the same face of Cdc42p highlighted by the S86, W97, and G142 residues (Fig. 1B, right). The dominant negative S86P mutation lies in the loop region between the 4-strand and the 3-helix and it interferes with interactions between S. cerevisiae Cdc42p and Cdc24p (116). Interestingly, this loop region makes close contacts with residues in the Rho insert domain (see above), and this region undergoes chemical shift changes in NMR spectroscopy studies upon binding of the nonhydrolyzable GTP analog GMPPCP (145), suggesting that it may be an additional switch region (i.e., switch III). The nature of the S86P dominant negative phenotype is unknown, but it is not due to sequestration of Cdc24p as is the mechanism of action of the D118A mutant allele (116). The W97R (3-helix) and G142S (4-helix) mutations are ts loss-of-function alleles in S. cerevisiae (391) and although their mechanisms of action are unknown, the W97R mutation leads to a bud site selection defect, implicating Cdc42p in the initial selection of a nonrandom bud emergence site. Taken together, these observations show that the face of Cdc42p defined by the 3- and 4-helices plays a critical role in Cdc42p function. The structure and function of the C-terminal membrane localization domain (highlighted in green in Fig. 1) are discussed below (see "Prenylation and subcellular localization").

The Cdc42 domains involved in guanine nucleotide binding and GTP hydrolysis (blue boxes in Fig. 1) have been inferred through structural similarities to domains in other GTPases and through the analysis of activated and dominant negative cdc42 mutations. The initial cdc42 mutations were analyzed in S. cerevisiae (642) and were based on the paradigmatic Ras mutations that led to oncogenic transformation. The cdc42^G12V and cdc42^Q61L mutations are analogous to H-ras mutations that cause a decreased level of intrinsic GTPase activity, thereby shifting the mutant proteins to an "activated" GTP-bound state. In S. cerevisiae, these cdc42 mutations were lethal, resulting in large, multibudded cells with aberrant cortical actin structures localized in multiple buds (642). These phenotypes suggested that the mutant proteins were activated and constitutively interacting with downstream components of the pathway, leading to polarization, albeit incorrectly, of the actin cytoskeleton. The H-ras^D119A mutation also leads to an activated phenotype; however, this phenotype was mechanistically due to an increased GDP dissociation rate, which is thought to result in a higher probability of the protein being in a GTP-bound state due to the higher concentration of GTP than of GDP in the cell. The phenotype of the S. cerevisiae cdc42^D118A mutant was unexpected and different from the cdc42^G12V and cdc42^Q61L activated phenotypes. The cdc42^D118A mutant phenotype was temperature-dependent dominant lethal and resulted in large, round, unbudded cells that were phenotypically similar to cdc42^ts mutants grown at restrictive temperatures. This dominant negative phenotype suggested that the Cdc42^D118A mutant protein could bind and sequester the cellular factor(s) necessary for the budding process (see "S. cerevisiae Cdc24p" below).

Expression of equivalent mutant proteins in S. pombe gave different results (390). First, unlike the activated and dominant negative phenotypes seen in S. cerevisiae, the morphological phenotypes of cells overproducing the cdc42^G12V, cdc42^Q61L, and cdc42^D118A mutant gene products were similar to one another. Second, the mutant constructs did not exert a dominant lethal effect in S. pombe cells. Instead, S. pombe cells overproducing these mutant proteins exhibited an abnormal morphology of enlarged, round or misshapen cells with delocalized cortical actin structures, as opposed to the small, round cellular morphology of cdc42 loss-of-function and dominant negative mutants (390, 447). The cdc42^D118A mutant phenotype also was temperature dependent in S. pombe, suggesting that the mutant protein loses either a critical interaction or its three-dimensional structure upon shift to higher temperature, leading to its mutant morphology. Interestingly, septum formation was still evident in these mutant cells, even though the presence of an organized actin ring was not, suggesting that septation can occur in the absence of an actin ring. However, this point must be clarified by experiments with either actin mutants or actin polymerization inhibitors such as latrunculin A.

While activated or dominant negative cdc42 mutations have not been analyzed in C. elegans to date, expression of the cdc42^G12V allele in Drosophila ovaries led to defects in actin distribution whereas expression of dominant negative cdc42 alleles led to defects in actin organization in imaginal discs and wing hairs (see "Drosophila" under "Functional studies" below), reinforcing a role for Cdc42p in regulating actin function. For a discussion of mutations in mammalian Cdc42p, see "Mammals" under "Functional studies" below.

The so-called effector domain between residues 26 and 48 of the Ras GTPase is required for downstream effector function (370, 460). The effector or switch I domain between residues 26 and 50 is highly conserved among Cdc42 proteins (Fig. 1A) but diverges among closely related but not functionally homologous Rac GTPases (239). The current paradigm is that GTPases bind to GEFs when in the nucleotide-free or GDP-bound state and bind to GAPs and downstream effectors when in the GTP-bound state. Since the switch I and II (residues 60 to 76) domains are the predominant regions of GTPases that change conformation upon binding different guanine nucleotides, it is likely that multiple factors interact with these regions. Given that Cdc42p interacts with multiple downstream effectors along with regulatory factors such as GEFs and GAPs (see "Cdc42p regulators" and "Cdc42p downstream effectors" below), it is likely that the specificity of interaction will be through either different residues within the Cdc42p effector domain, competition between effectors and regulators, and/or interactions at different times in the cell cycle.

A predominant binding partner for the effector domain is the CRIB (for "Cdc42/Rac interactive binding") domain (also known as the PBD, GBD, or PAK domain [see Table 3]) found in many Cdc42p downstream effectors, including the PAK family of protein kinases (59). The highest-efficiency binding domain in the CRIB-containing PAK protein was residues 70 to 118, thereby defining the optimal CRIB domain as these 48 amino acid residues (558). In these studies, it was also shown that this domain interacted with Cdc42p at a ~3- to 10-fold-higher affinity than it interacted with Rac and that it interacted with activated (Q61L) alleles at a 5- to 10-fold-higher affinity than it interacted with the wild type, reinforcing the notion that CRIB-containing interacting proteins function as downstream effectors. A recent study in which NMR spectroscopy was used to probe the interactions between Cdc42Hs and 46 amino acids of the PAK CRIB domain showed that the CRIB binding domain surface on Cdc42Hs was the 2 switch I domain and part of the loop between the 1-helix and 2-strand (185). In addition, nuclear Overhauser effect contacts suggested that the formation of an intermolecular -sheet was the basis for the Cdc42Hs-CRIB domain interactions. The CRIB domain of the Wiskott-Aldrich syndrome protein (WASP) downstream effector (see "Cdc42p downstream effectors" below) was dissected by a variety of biophysical techniques including fluorescence spectroscopy, surface plasmon resonance, circular dichroism, and NMR spectroscopy (498). The results indicated that a core 26-amino-acid fragment (residues 221 to 257) was necessary for binding to GST-Cdc42, but higher affinity binding was observed with a larger (120-amino-acid) fragment (residues 201 to 321), suggesting that the CRIB domain was necessary but not sufficient for high-level binding. In addition, these studies suggested that the isolated CRIB domain does not exhibit any apparent secondary structure; it is unknown if the CRIB domain would form a secondary structure, possibly -strands (see above), within the context of the entire protein.

Mutations that disrupt the interaction between Cdc42p and downstream effectors should define the effector domain and should suppress dominant activated cdc42^G12V mutant phenotypes. The T35A allele was thought to be a paradigmatic effector domain mutation in that it could interfere with the ability of S. cerevisiae Cdc42p to interact with the PAK family of protein kinases and could suppress the dominant-activated cdc42^G12V mutant but could not complement the loss-of-function cdc42-1^ts allele. However, the T35A mutation also suppressed the dominant negative S. cerevisiae cdc42^D118A allele and interfered with two-hybrid protein interactions between Cdc42^D118Ap and the Cdc24p GEF (116), suggesting that the effector domain may also interact with the Cdc24p GEF. Corroborating this hypothesis are the results obtained with the Y32K and F37E mutations in the Cdc42Hs effector domain, which caused a loss of Cdc24-stimulated GDP dissociation (323), and the Cdc42Hs F28L mutation, which led to rapid nucleotide exchange and transformation of NIH 3T3 cells similar to that seen with the Cdc24p homolog Dbl (327). Interestingly, the T25K, N26D, and Y40K mutations within the Cdc42Hs effector domain did not show a loss of Cdc24-stimulated GDP dissociation (323), which could be a function of the individual mutational changes or could indicate a level of specificity at the individual amino acid residue for interactions with GEFs.

The S. cerevisiae cdc42^V44A mutation represents a new class of effector domain mutations in that it could complement the cdc42-1^ts allele (475a), suggesting that it did not lead to a nonfunctional protein; it also interfered with interactions with the upstream effector Cdc24p (116). In addition, the cdc42^V44A mutant displayed a morphological phenotype of elongated buds with multiple nuclei, which is suggestive of either a delay at the apical/isotropic switch and morphogenesis checkpoint (see "S. cerevisiae" under "Functional studies" below) and/or a defect in cytokinesis (475a). The V44A mutation interfered with two-hybrid protein interactions between Cdc42p and the S. cerevisiae Cla4p PAK-like kinase but not the Ste20p or Skm1p PAK-like kinases and also between Cdc42p and the Gic1p and Gic2p downstream effectors but not the Bnip or Iqg1p scaffold proteins (see "Cdc42p downstream effectors" below). All of these proteins contain CRIB domains, suggesting that the effector domain may differentially interact with multiple CRIB domain-containing effectors. This hypothesis is substantiated by mutations in the Cdc42Hs effector domain that differentially affected interactions with mammalian downstream effectors (291, 379) (see "Cdc42p downstream effectors" below). The Y40C mutation interfered with interactions between Cdc42p and downstream PAKs and other proteins containing CRIB domains, leading to a loss of p65^PAK kinase activation in transfected COS fibroblasts, but it did not affect Cdc42p-dependent actin reorganization, as evidenced by normal filopodium and integrin complex formation in Swiss 3T3 cells (291). The D38E mutation interfered with in vitro binding to an mPAK-3 CRIB domain peptide (310) but had no effect on the binding of two other downstream effectors, IQGAP1 and IQGAP2. The Cdc42Hs F37A mutation did not affect interactions with CRIB-containing proteins or actin reorganization (291). Taken together, these results suggest that there are different classes of effector domain mutations that can be distinguished by their morphological phenotypes and protein-protein interactions. These different mutations may define interactions with different effectors or regulators, thereby allowing us to dissect the multiple pathways leading from Cdc42p (see "Cdc42p downstream effectors" below). Given the possibility that the Cdc42p effector domain interacts with CRIB-containing proteins through formation of an intermolecular -sheet (see above), it is likely that the nature and orientation of amino acid side chains emanating from the 2 strand (Fig. 1) have an influence on this differential binding. It should be noted, however, that not all Cdc42p-interacting proteins contain recognizable CRIB domains, suggesting that there may be multiple mechanisms by which proteins interact with Cdc42p.

In addition to the effector/switch I domain mutations that affect interactions between Cdc42 and its GEFs, there are mutations in other domains that affect either binding to GEFs or GEF-induced GDP dissociation. Mutations within the switch II domain, including D63L, D65K, R66D, R68A, P69A, and F78L, did not affect Cdc24-stimulated GDP dissociation, but the Q116K mutation in the Cdc42 signature GTP binding domain did (323). Analysis of chimeric Cdc42Hs-RhoA proteins indicated that residues 82 to 120 and 121 to 155 are necessary for Cdc24 responsiveness (323). Within the first domain, the S86P and S89P mutations in S. cerevisiae Cdc42p led to a cold-sensitive, dominant negative phenotype and, in the case of S86P, led to a loss of interaction with Cdc24p (116). Similar mutations in Drosophila Cdc42p also led to a dominant negative phenotype (see "Drosophila" under "Functional studies" below). Other mutations that led to a loss of interaction with Cdc24p in S. cerevisiae included the I117S mutation within the Cdc42 signature GTP binding domain and the T138A and L165S mutations (116). In addition, the C157R mutation within one of the highly conserved GTP binding domains led to a partially cold-sensitive dominant negative phenotype. The C-terminal membrane localization domain (see "Prenylation and subcellular localization" below) does not seem to be necessary for GEF interactions (323). Interestingly, the S86, S89, and T138 residues all lie on the same face of the Cdc42 protein (Fig. 1B), suggesting that this may be a conserved interactional interface between Cdc42p and its GEFs.

In all organisms examined, Cdc42p is prenylated with a C₂₀ geranylgeranyl isoprene group at a C-terminal Cys residue, and this prenylation is necessary for the membrane attachment of Cdc42p. S. cerevisiae and S. pombe Cdc42p fractionated into both soluble and particulate fractions, suggesting that Cdc42p can exist in two cellular pools (390, 643). S. cerevisiae Cdc42p was found predominantly in the particulate fraction, but a significant soluble pool, up to ~20% in some instances, could be observed. Given the existence of GDI proteins in S. cerevisiae and mammalian cells that can interact with Cdc42p and extract Cdc42p from membranes (see "Guanine nucleotide dissociation inhibitors" below), the soluble pool of S. cerevisiae Cdc42p is probably either nonprenylated or complexed with the rho-GDI protein Rdi1p (267, 373). The particulate form of Cdc42p could be solubilized by added detergent but not by added NaCl or urea, suggesting that Cdc42p was tightly associated with either a cellular membrane or a cytoskeletal complex. When synchronous cultures were used, the fractionation pattern of S. cerevisiae Cdc42p did not appear to vary through the cell cycle (643). However, recent studies indicating that Cdc42p functions at different places in the cell at different times of the cell cycle (see below) suggest that fractionation patterns may not be a very sensitive measure of Cdc42p localization.

By using Cdc42p-specific antibodies in immunofluorescence and immunoelectron microscopy, S. cerevisiae Cdc42p was localized to the plasma membrane at sites of polarized growth (642, 643). These sites coincided with the sites of cortical actin localization and included invaginations of the plasma membrane at the site of bud emergence, the tips of growing buds, and the tips of mating projections in pheromone-arrested cells (3, 259). This localization pattern was consistent with Cdc42p functioning in controlling polarized cell growth during the mitotic cell cycle and mating. Recently, functional green fluorescent protein (GFP)-Cdc42p fusion proteins have been localized to the mother-bud neck region in S. cerevisiae and the septum area in S. pombe, suggesting that Cdc42p also plays a role in cytokinesis and/or septation in both yeasts (508a). This localization of Cdc42p at different sites of polarized growth during the S. cerevisiae cell cycle, which is mirrored by the localization of the actin cytoskeletal network, suggests that the subcellular localization of Cdc42p is under temporal and spatial control during the cell cycle. It should be noted that Cdc42p localization to sites of polarized growth was not disrupted by incubation with the actin-depolymerizing drug latrunculin-A (21), suggesting that Cdc42p localization occurs independently of actin localization and of the structural integrity of the actin cytoskeleton.

S. cerevisiae Cdc42p contains the C-terminal ¹⁸³Lys-Lys-Ser-Lys-Lys-Cys-Thr-Ile-Leu sequence which is essential for the membrane localization of Cdc42p. In the Cdc42Hs NMR and crystal structures, this region forms a flexible tether that is separated from the body of Cdc42Hs by two Pro residues at positions 179 and 180 (145) (Fig. 1B), thereby allowing the bulk of Cdc42p to be sterically unhindered by membrane attachment and accessible for binding to other proteins. This region is modified by geranylgeranylation at the Cys¹⁸⁸ residue (indicated by an underbar in the above sequence), which is necessary for its anchoring within the plasma membrane (147, 643). This is thought to be followed by proteolytic cleavage of the last three amino acids and carboxyl methylation of the now C-terminal Cys residue, although this has not been experimentally shown with S. cerevisiae Cdc42p (see below). The geranylgeranylation is deemed necessary because the C188S mutation, which renders the protein incapable of being prenylated, can intragenically suppress the dominant lethality associated with the cdc42^G12V, cdc42^Q61L, and cdc42^D118A mutations (642) and because the S. cerevisiae and S. pombe Cdc42^C188S mutant proteins, either by themselves or as GFP-Cdc42^C188Sp fusion proteins, are nonfunctional, delocalized proteins that fractionate almost exclusively into soluble pools (508a, 642, 643). Whether this prenylation is sufficient for Cdc42p targeting to the sites of polarized growth is not known.

An increase in soluble S. cerevisiae Cdc42p was observed in cdc43-2^ts (643) and cdc43-5^ts (434) mutant cell extracts, suggesting that membrane localization of Cdc42p depended on Cdc43p-dependent geranylgeranylation. Cdc43p was originally identified by ts mutations that led to cell cycle arrest of large, unbudded cells, a phenotype similar to cdc42^ts mutants (2). In addition, cdc43^ts cdc42^ts double mutants displayed a synthetic lethal phenotype at 23°C, suggesting that these gene products may interact. Another mutation in CDC43, designated cal1-1, was identified by its calcium-dependent phenotype (437); the cal1-1 mutant required 100 mM CaCl₂ for growth. The predicted amino acid sequence of the CDC43/CAL1 gene (240, 435) showed significant similarity to the S. cerevisiae DPR1/RAM1 gene (173), which encoded the  subunit of the protein farnesyltransferase (FTase), which modified the C termini of Ras GTPases. Type I protein geranylgeranyltransferase (GGTase I) activity was reconstituted from Escherichia coli cells that overproduced both Cdc43p and Ram2p (377, 539, 540), and S. cerevisiae GGTase I activity was decreased in cdc43^ts and ram2 mutants but not ram1 mutants (147, 377, 434), indicating that Cdc43p and Ram2p encoded the  and  subunits, respectively, of the S. cerevisiae GGTase I. The Ram2p  subunit also acts as the  subunit for the S. cerevisiae FTase (214), which may account for the in vitro and in vivo cross-specificity that is observed between FTase and GGTase I activities in S. cerevisiae (66, 565). S. cerevisiae GGTase I is an Mg²⁺-requiring Zn²⁺ metalloenzyme (377, 539), but it can also function with Ca²⁺ as the only divalent cation (377). Added Ca²⁺ could not rescue the reduced in vitro GGTase I activity from cal1-1 mutant cell extracts (434), but only 20 mM CaCl₂ was added, as opposed to the 100 mM CaCl₂ needed to rescue the in vivo cal1-1 growth defect (435, 437). The essential targets for S. cerevisiae GGTase I seem to be Cdc42p and Rho1p, because certain cdc43/cal1 alleles can be suppressed by overexpression of one or both of these GTPases (434, 439).

Another localization determinant consists of the four Lys residues that are next to the C-terminal prenylated Cys residue. This polylysine region is not found in most Ras-like GTPases, and its positive charges may be interacting with negatively charged components, either protein or phospholipid, at the membrane site to play a role in enhancing membrane association or specific targeting of Cdc42p. A similar polylysine domain is found in the K-Ras protein and is important for membrane localization; altering the Lys residues to Gln results in delocalized K-Ras protein (196, 197). In addition, the analogous polylysine domain in Rac1 was recently shown to be important for interactions with PAK effector kinases (266). Mutating the four Lys residues to Gln in S. cerevisiae Cdc42p, creating the cdc42^K183-187Q mutant protein, led to a partial delocalization of the mutant protein (116), suggesting that this domain played a role in targeting or anchoring Cdc42p to the plasma membrane. The K183-187Q mutation could intragenically suppress the dominant lethal cdc42^G12V mutant, and expression of the cdc42^K183-187Q mutant gene on a plasmid could complement the cdc42-1^ts mutant. The ability of the cdc42^K183-187Q mutant gene to complement the cdc42-1^ts mutant (in contrast to the nonfunctional cdc42^C188S mutant gene, which cannot complement the cdc42-1^ts mutant [642]), together with the partial delocalization of the mutant protein, suggested that the K183-187Q mutation had an intermediate effect on Cdc42p function and that the polylysine domain of Cdc42p was necessary but not sufficient for complete plasma membrane localization. Another interesting mutation in this domain, the cdc42^K186R mutant allele, exhibited a ts loss-of-function phenotype in S. cerevisiae (391) and displayed a morphological phenotype of elongated buds and multiple nuclei suggestive of either a delay at the morphogenesis checkpoint (see "S. cerevisiae" under "Functional studies" below) and/or a cytokinesis defect at permissive temperatures (106a). The nature of this mutation (Lys to Arg) suggested that these phenotypes were not due to a change in charge or conformation of the protein but more probably were due to altered interactions with another protein. However, recent results indicate that this mutant protein has an increased intrinsic GTPase activity (see "GTPase-activating proteins" below), suggesting that improper negative regulation of this mutant protein may be playing a role in its phenotypes. Therefore, the mechanism by which this mutant protein exerts its effects remains to be fully elucidated.

All known Cdc42 proteins contain the C-terminal sequence Cys-Xaa-Xaa-Leu, with the exception of the mouse and human brain G25K isoforms, which end in a Phe residue instead of Leu (Fig. 1). Mammalian Cdc42p posttranslational modifications have been analyzed biochemically with protein purified from bovine brain cells (22, 612, 613), cultured murine erythroleukemia cells (354), rat and human pancreatic islet cells (274), or rat kidney cells (45), not with recombinant protein, and so it is unclear whether these studies were performed on the Cdc42Hs or G25K isoform. Regardless, it is clear that the membrane-bound form of mammalian Cdc42p is geranylgeranylated at the Cys residue, the last three amino acids are proteolytically removed, and the now C-terminal prenylated Cys residue is carboxyl methylated, resulting in a protein with a S-(all-trans-geranylgeranyl) cysteine methyl ester at its C terminus (612). These modifications are necessary for membrane localization, and, as with S. cerevisiae Cdc42p, mammalian Cdc42p fractionates to both particulate and soluble pools (45, 354).

Carboxyl methylation of soluble Cdc42p from bovine brain (22), rat kidney cells (45), or pancreatic islet cells (274) seems to be GTP stimulated, but methylation of the membrane-bound form is not (613), presumably because the membrane-bound form is already GTP bound. The methyltransferase activity from brain extracts (613) and insulin-secreting cells (318) was membrane bound. Recently, a human myeloid prenylcysteine carboxyl methyltransferase with in vitro activity against Cdc42Hs was shown to localize to the endoplasmic reticulum membrane (114); the S. cerevisiae Ste14p prenylcysteine carboxyl methyltransferase is also found in the endoplasmic reticulum membrane (490), but it has not been shown to have in vitro or in vivo activity against Cdc42p. Interestingly, addition of glucose to pancreatic islet cells extracts stimulated the carboxyl methylation of Cdc42p (274), and inhibition of Cdc42p function by Clostridium difficile toxins A or B resulted in reductions in glucose-stimulated insulin secretion (273), suggesting that Cdc42p may play a role in glucose-stimulated insulin secretion.

Using affinity-purified anti-Cdc42 antibodies, Cdc42p from rabbit liver was shown to associate with a membrane fraction highly enriched in Golgi membranes (137). In pancreatic islet cells (274) and rat kidney cells (45), Cdc42 was predominantly cytosolic, but it was translocated to the particulate pool upon addition of guanosine 5'-(3-O-thio)triphosphate (GTPS). In NR-6 fibroblasts and rat kidney cells, Cdc42p localized to a perinuclear region that coincided with markers for the Golgi complex including the 110-kDa subunit of the coatomer complex -COP (137). This localization was rapidly altered to general cytosolic localization upon addition of brefeldin A (BFA), a drug which inhibits vesicle formation at the Golgi membrane by inhibiting the guanine nucleotide exchange activity for the Arf GTPase, suggesting that Cdc42p may play a role in or be subject to intracellular membrane trafficking events. BFA-induced delocalization of Cdc42p was inhibited by AlF₄ and by expression of GTPase-defective Arf, while expression of a dominant negative Arf mutant resulted in BFA-independent delocalization of Cdc42p. These results suggest that association of Cdc42p with Golgi membranes is dependent on the guanine nucleotide-bound state of the Arf GTPase. It should be noted that in these experiments, the NR-6 fibroblasts and rat kidney cells did not exhibit polarized growth patterns to a region of their cell periphery. In human HeLa cells transiently transfected with a epitope-tagged Cdc42^G12V protein, the epitope-tagged protein localized to focal complexes and to regions of polarized growth within the Cdc42^G12V-induced peripheral actin microspikes (PAMs) (see "Mammals" under "Functional studies" below) and colocalized with actin and PAKs within these PAMs (127, 356). In addition, HA-tagged Cdc42^G12V protein co-localized with the IQGAP1 downstream effector to cell-cell contact sites of Madin-Darby canine kidney cells (282).

In Drosophila wing disc epithelial cells, Cdc42p localized in a polarized manner to the basal and apical regions (128). In elongating cells, Cdc42p was restricted to the apical and basal membranes, but in nonelongating cells, it was found on lateral membranes as well. This localization pattern was also seen for the actin cytoskeleton, again providing a mechanistic link between Cdc42 and actin rearrangements. C. elegans Cdc42p was shown to fractionate predominantly to a particulate fraction from mixed-stage populations of C. elegans cells and to localize in a polarized manner to both the circumferential and longitudinal boundaries of hypodermal cells during hypodermal cell fusion in embryo elongation, in a pattern similar to that of the C. elegans PAK homolog (87).

In summary, Cdc42 proteins are membrane bound through their posttranslational modifications and are localized to either internal membranes or the plasma membrane at locations where polarized events are occurring. While prenylation is necessary for membrane anchorage, it is not known if it is sufficient for proper targeting. The mechanism by which Cdc42 proteins are targeted to appropriate membranes in regions of polarized cell growth is unknown, but it is likely to be through protein-protein or protein-lipid interactions at the site.

S. cerevisiae alters its morphology in response to both exogenous and endogenous signals, leading to either bud emergence and enlargement during the mitotic cell cycle, mating-projection ("shmoo") formation through the mating/pheromone response pathway in response to exogenous mating-factor pheromones, pseudohyphal formation and filamentous growth in response to starvation conditions, or spore formation during meiosis. Cdc42p has been implicated in regulating the first three processes but not in sporulation to date. The mechanisms by which Cdc42p regulates the generation of, and switching between, these different morphogenetic patterns is still unclear, but Cdc42p interactions with the actin cytoskeleton play a critical role in this regulation. The functional connection between Cdc42p and the cortical actin cytoskeleton has recently been reinforced by the observation that Cdc42p can stimulate actin polymerization in permeabilized S. cerevisiae cells (324).

Mitotic cell cycle. The morphological changes that occur during the S. cerevisiae mitotic cell cycle can be divided into five sequential phases: (i) selection of a nonrandom bud emergence site and the organization of the protein machinery at that bud site, including rearrangement of the cortical actin cytoskeleton; (ii) bud emergence and polarized growth towards and within the emerging bud; (iii) a switch from apical to isotropic bud growth (the "apical-isotropic switch" [see below]); (iv) cytokinesis, septum formation, and cell separation; and (v) isotropic growth of undersized daughter cells after cell separation prior to the initiation of their next cell cycle (reviewed in references 77, 126, and 465) (Fig. 2). A variety of data suggest that Cdc42p can function at multiple stages of the cell cycle. As mentioned above (see "Cdc42p structure and functional domains"), the initial characterization of S. cerevisiae Cdc42p suggested that it plays a role in the actin-dependent generation of cell polarity during the process of bud emergence. Subsequent analysis of ts, dominant activated, and dominant negative cdc42 alleles substantiated this inference and suggested an additional function in the initial selection of the site of bud emergence. Analysis of the cdc42^V44A and cdc42^K186R mutant alleles, along with the subcellular localization of Cdc42p to the mother-bud neck region in large-budded cells, raises the possibility that Cdc42 functions either within the apical-isotropic switch at a morphogenesis checkpoint (Fig. 2) (see below) and/or in controlling actin-dependent events that occur during cytokinesis and septum formation. In addition, GTP-bound Cdc42p functions with the mitotic cyclin Clb2p-Cdc28p kinase complex to lead to the mitosis-specific phosphorylation of several substrates (see below). A potential model that is consistent with the proposed roles for Cdc42p throughout the cell cycle is presented in Fig. 3; a detailed discussion of individual protein components of the model can be found under the individual protein subsections in "Cdc42p regulators" and "Cdc42p downstream effectors" below.

View larger version (34K): [in this window] [in a new window]   FIG. 2.   Polarized cell growth during the S. cerevisiae and S. pombe cell cycles. See the text for details.

View larger version (59K): [in this window] [in a new window]   FIG. 3.   Molecular and cytological model for polarized cell growth in S. cerevisiae. (A) Molecular model for the Cdc42-dependent processes during the S. cerevisiae cell cycle. Shaded boxes attached to GTPases are isoprenyl groups. The stars by Ste20 in the bud emergence complex and Cla4 in the cytokinesis complex indicate that they are the likely PAK functioning at this stage of the cell cycle. See the text for details. (B) Cytological model for bud emergence and cytokinesis. Cdc42 complexes are the same as in panel A. See the text for details.

Mating pathway. The S. cerevisiae mating pathway is a classic signal transduction pathway in which an extracellular signal (peptide pheromone) binds to a G-protein coupled transmembrane receptor, thereby activating a MAP kinase cascade that ultimately leads to a number of cellular events, including the transcriptional induction of genes necessary for the mating process, a G₁ arrest as a prelude to cell-cell fusion and karyogamy, and the generation of unique morphological structures (mating projections or shmoos) that are the sites of contact for cell-cell fusion and mating (for reviews, see references 26, 297, 350, and 368) (Fig. 4A). The notion that Cdc42p plays a role in the S. cerevisiae mating pathway came from the analysis of cdc42 mutants, the subcellular localization of Cdc42p, and its interactions with the Ste20p protein kinase. The mating efficiencies of loss-of-function cdc42 mutants were reduced, and signaling through the pheromone response pathway was diminished (527, 626), while expression of the dominant activated cdc42^G12V mutant allele led to a modest (two- to threefold) increase in signaling as assayed by FUS1-lacZ expression (9), suggesting that Cdc42p plays a role in the activation of the pheromone response MAP kinase cascade (see below). In addition, Cdc42p localized to the tips of mating projections in pheromone-arrested cells (643), suggesting that it plays a direct role in pheromone-induced morphological changes.

View larger version (50K): [in this window] [in a new window]   FIG. 4.   Comparison of Cdc42 interactions and dependent processes in S. cerevisiae (A), S. pombe (B), and mammals (C). Color and shape coding is given in the box at the bottom of figure. Two-headed arrows indicate physical interactions. Single-headed arrows indicate pathways; dotted arrows indicate potential involvement in pathways. For simplicity, not all components of the actin cytoskeleton or JNK kinase cascade are shown. See the text for details.

Pseudohyphal and invasive growth. S. cerevisiae cells can alter their morphogenetic patterns in response to starvation conditions, leading to filamentous growth and the generation of pseudohyphae (for reviews, see references 26, 167, 297, 349, and 350). Diploid cells respond to nitrogen starvation by altering their cell cycles, budding patterns, cell shape, and cell separation patterns, resulting in polarized elongated budded cells that resemble fungal hyphae (168, 280). Haploid cells can also be induced to filamentous growth, which is manifested as invasive growth into agar plates (43, 487). A detailed mutational analysis of actin mutants indicated that the actin cytoskeleton plays a critical role in various aspects of pseudohyphal growth (63). The primary signaling route leading to pseudohyphal growth involves the Ras2 GTPase signaling through Cdc42p to several components of the pheromone response MAP kinase cascade, including Ste20p, Ste11p, Ste7p, and Kss1p, thereby activating the Ste12p transcription factor which, together with the Tec1 transcription factor, induces the expression of genes necessary for filamentous growth (Fig. 4A) (168, 330, 348, 351, 405, 487, 488).

In the rod-shaped fission yeast S. pombe, there are three switches in polarized cell growth patterns during the cell cycle (Fig. 2). First, selective and polar growth is initiated at the beginning of the cell cycle at the "old end" of the cell, which is the end distal to the previous division site (reviewed in reference 425). This growth occurs at the end of the cylindrical cell and can be monitored by staining with the dye Calcofluor and by the presence of cortical actin dots (366, 376). Second, after ~0.3 of the cell cycle, a switch in polarized growth, referred to as new-end takeoff, occurs from unidirectional at the old end to bidirectional at both ends (Fig. 2). This growth pattern is visualized by the appearance of both Calcofluor staining and cortical actin dots at the new end and depends on the cell attaining a minimal length and completing the S phase. Third, bipolar growth continues until ~0.75 of the cell cycle, at which time cortical actin reorganizes to the site of septum formation and end growth ceases, resulting in a constant-length stage of the cell cycle. Following cytokinesis and cell separation, polarity must be re-established at the old end as a prelude to unipolar growth in the next cell cycle.

The S. pombe Cdc42p homolog (cdc42⁺) was isolated from an S. pombe cDNA library by functional complementation of the S. cerevisiae cdc42-1^ts mutation (390). The predicted amino acid sequence of S. pombe Cdc42p is 85% identical to those of both S. cerevisiae and human Cdc42p (Fig. 1). Disruption of cdc42⁺ showed that the gene was essential for growth. The S. pombe cdc42 loss-of-function phenotype was originally determined by generating a null allele in a haploid strain that was complemented by the wild-type allele on a plasmid and then inducing plasmid loss to uncover the loss-of-function phenotype (390). The morphological phenotype consisted of small, round, dense, uninucleate cells, which is strikingly different from that associated with cdc42 loss-of-function alleles in S. cerevisiae (2, 642) (see above). The S. pombe cdc42 null phenotype suggested that macromolecular synthesis continued but incorporation of new cellular material into an enlarging cell was inhibited, hence the small, dense, dead cells. Similar morphologies, as well as reduced mating efficiencies, were observed with the cdc42^T17N dominant negative allele (447), suggesting that Cdc42p functions within the mating pathway as well (see below). The uninucleate, 1C phenotype, as assayed by DAPI staining and fluorescence-activated cell sorter analysis, indicated that the mitotic cell cycle was blocked in G₁ phase, which is also different from the S. cerevisiae arrest phenotype of multinucleate cells. It is likely that the cell cycle coordination between DNA synthesis and Cdc42 function is more tightly regulated in S. pombe. Taken together, these data are consistent with Cdc42p functioning in the targeting and incorporation of new growth at the old end in G₁ phase, possibly by affecting protein secretion or secretory-vesicle fusion to the plasma membrane.

Recent data indicating that S. pombe Cdc42p localizes to the septum region (381a) raises the possibility that Cdc42p plays a direct role in septum formation. In S. pombe, cytokinesis begins in early M phase with the assembly of the actin-based medial ring followed by septum formation and cell separation (reviewed in reference 176). Given that GFP-Cdc42p in S. pombe localizes to the medial area in a ring-like structure in some cells that do not have a visible septum, Cdc42p may play a role in the early steps of medial ring formation prior to septum formation. Interestingly, S. pombe Cdc12p, a homolog of the S. cerevisiae Cdc42-interacting protein Bni1p (139, 232) (see "Cdc42p downstream effectors" below), has also been implicated in medial ring formation (74).

To date, two potential downstream effectors of Cdc42p has been characterized in S. pombe, the Pak1p/Shk1p (364, 447) and Pak2p/Shk2p (515, 616) protein kinases (see "PAK-like kinases" below). Pak1p/Shk1p is a CRIB domain-containing ~72-kDa serine/threonine protein kinase that belongs to the PAK family of Cdc42-interacting protein kinases. It can autophosphorylate on Ser and Thr residues and binds preferentially to Cdc42p-GTP. The physiological significance of these interactions was supported by the morphological abnormalities associated with the overexpression or absence of Pak1p and the synthetic-overdose phenotypes observed when overexpressing activated or dominant negative cdc42 alleles together with wild-type or kinase-defective pak1 mutants. Deletion of pak1 is lethal, resulting in small, round cells, a morphology reminiscent of cdc42 null mutants. This result indicates that Pak1p provides an essential function in the cell polarity pathway, which is different from its S. cerevisiae homologs Ste20p, Cla4p, and Skm1p. Pak1p and Cdc42p were also required for mating in S. pombe (see "PAK-like kinases" below). Taken together, the data are consistent with Pak1p being a downstream effector of Cdc42p in the cell polarity and mating pathways (Fig. 4B). Pak2p/Shk2p is a nonessential protein with the greatest similarity in predicted amino acid sequence to S. cerevisiae Cla4p and Skm1p (515). Its role as a downstream effector of Cdc42p is unclear (see "PAK-like kinases" below).

The Candida albicans CDC42 gene was identified by degenerate oligonucleotide PCR and isolated from a C. albicans genomic library by DNA-DNA hybridization to the PCR probe (394). C. albicans Cdc42p is 87.8% identical to S. cerevisiae Cdc42p throughout the entire coding region (Fig. 1A), and DNA-DNA hybridizations suggested that CDC42 is single copy. Analysis of mRNA levels indicated that there is a transient increase in Cdc42p expression in the dimorphic switch to bud emergence, suggesting that C. albicans Cdc42p plays a role in this process. C. albicans homologs of the S. cerevisiae cell polarity proteins Rsr1p/Bud1p (608), Rho1p (272), Cla4p (299), Ste20p and Ste7 (268, 296), Fus3p/Kss1p (110, 595), and Ste12p (329, 352) have also been identified, and many of them have been implicated in hyphal formation and candidiasis. It remains to be seen if Cdc42p also functions in pseudohyphal formation in C. albicans, as it does in S. cerevisiae.

The Caenorhabditis elegans Cdc42 gene seems to be more highly expressed in embryonic cells than in larvae or adults, and when expressed in S. cerevisiae, it could complement the cdc42-1^ts allele (88), indicating that it is a functional homolog. The C. elegans Cdc42 protein, expressed as a glutathione S-transferase (GST) fusion, could bind and hydrolyze GTP at rates comparable to the human Cdc42 protein (88). By using anti-Cdc42 antibodies, Cdc42p was shown to fractionate predominantly to a particulate fraction from mixed-stage populations of C. elegans cells and to localize to both the circumferential and longitudinal boundaries of hypodermal cells during hypodermal cell fusion in embryo elongation, in a pattern similar to that of the C. elegans PAK homolog (87). These results suggest that C. elegans Cdc42 plays a role in the actin-dependent process of embryonic-body elongation. Analysis of cdc42 activated or dominant negative mutant alleles in C. elegans has not been reported to date. The unc-73 gene product exhibits structural homologies to the Dbl family of GEFs (see "Mammalian GEFs" below) and has guanine nucleotide exchange activity against Ce-Rac but not Ce-Cdc42 in vitro (536); whether it is a GEF for Cdc42p in vivo is unknown. Unc-73p localized to the nerve ring and ventral nerve cord and was required for neuronal axon guidance; upon injection into cells, it induced actin polymerization at the plasma membrane. Recently, a C. elegans homolog of the mammalian MKK7 protein kinase, which functions in the Cdc42p-JNK MAP kinase signaling cascade (see "Mammals" below), was identified (148). Characterization of this and other components of the JNK signaling cascade should provide valuable insight into Cdc42p functions in C. elegans.

Studies with Drosophila have indicated a role for Cdc42p in a variety of actin-dependent processes (128, 129, 336, 408). These studies have been based on the analysis of the cellular phenotypes associated with overexpression of activated and dominant negative cdc42 alleles. For example, expression of the activated cdc42^G12V mutant allele in the nervous system resulted in defects in dendrite and axon outgrowth and in embryo death whereas expression in muscle led to abnormally shaped muscle fibers (336). Expression of the cdc42^G12V mutant allele in epithelial cells of the wing imaginal disc led to a high rate of cell death, but expression of a cdc42 dominant negative allele led to changes in the shape of polarized cells, a disruption of apico-basal cell elongation, and a loss of the dense, actin-containing plaques seen on the basal membrane. There were also mild effects on the location of the adherens junctions at the apical face of the cells (128). Expression of the dominant negative cdc42 allele did not affect polarized-protein accumulation in epithelial cells, as evidenced by proper cadherin and yellow-protein localization. In wing epithelial cells that form hairs, expression of the dominant negative cdc42 allele led to a loss or stunting of wing hair formation, which appeared to be a consequence of defects in actin polymerization within the wing hair (128, 129). In ovaries, expression of the activated cdc42^G12V allele led to defects in actin distribution, resulting in altered nurse cell structure (i.e., abnormally fused nurse cells) and delocalized ring canals, structures which function as a cytoplasmic conduit between nurse cells and the oocyte but did not affect border cell migration (408). Therefore, Cdc42p seems to regulate actin distribution and function in a number of different cellular processes. In addition, Cdc42p may be involved with various Drosophila MAP kinase and JNK homologs (195, 482, 530) (see below) that regulate a variety of cellular processes including dorsal closure and establishment of cell polarity (for a review, see reference 233).

Mammalian Cdc42p has been implicated in a wide variety of in vivo functions including receptor-mediated signal transduction pathways leading to induction of transcription and actin rearrangements, cell cycle progression, and apoptosis. Most of these studies have relied on the phenotypic analysis of ectopic expression of dominant activated and dominant negative cdc42 mutants. In addition, the characterization of mammalian Cdc42p-interacting proteins has implicated Cdc42p in multiple pathways (Fig. 4C), whose regulation is still unclear.

Actin rearrangements. One of the earliest studies showed that differentiation of human monocytes into macrophages following the addition of phorbol esters led to an increase in the amount of membrane-bound Cdc42p that correlated with an increase in the actin-dependent cell-spreading activity (6). In subsequent studies, microinjection of wild-type or activated Cdc42^G12V protein into serum-starved or subconfluent Swiss 3T3 fibroblasts led to the induction of peripheral actin PAMs, actin-containing long peripheral filopodia, and vinculin-containing focal complexes and to a reduction in the number of Rho-dependent actin stress fibers (277, 420). Filopodia are thought to be sensory structures involved in the actin-based polarized cell growth observed in fibroblasts (11) and neuronal cells (229, 254, 531). The Cdc42p-dependent induction of vinculin-containing focal complexes in serum-starved Swiss 3T3 fibroblasts was inhibited by microinjection of the dominant negative Rac^T17N protein, and the induction of filopodia was seen as a prelude to the formation of Rac-dependent membrane ruffles and lamellipodia and Rho-dependent stress fibers in time-lapse photomicroscopy (277, 420, 476, 479, 480). These results suggest that Cdc42p may function upstream of Rac and Rho in the generation of actin-dependent subcellular structures in these cells (420).

Cdc42p and the JNK/SAPK and p38 MAP kinase cascades. In addition to regulating actin rearrangements, Cdc42p functions to couple cell surface receptors to MAP kinases, thereby transducing extracellular signals to regulate intracellular events, most notably the transcriptional induction of genes essential for a diverse number of cellular processes. These processes include inflammatory and stress responses, mitogenesis, differentiation, cell growth, cell cycle progression, apoptosis, prostaglandin biosynthesis, myocyte hypertrophy, and immunity gene expression (103, 120, 141, 179, 195, 217, 253, 287, 369, 415, 489, 513, 514, 578). The stress response signaling pathway involves the stress-activated protein kinases (SAPKs) (286), which also can phosphorylate Ser63 and Ser73 in the N terminus of the c-Jun subunit of the AP-1 transcription factor (JNKs) (121). These JNK/SAPK protein kinases can be activated by a variety of cell surface receptors and cellular stresses such as heat shock, UV radiation, and changes in osmolarity, by the protein synthesis inhibitors anisomycin and cycloheximide, and by the cytokines interleukin-1 (IL-1) and tumor necrosis factor alpha (TNF-) (121, 286) and can be inhibited by high cell density (289). The p38/Mpk2 protein kinase, which is another member of this SAPK family and is the mammalian homolog of the S. cerevisiae Hog1p (for "high-osmolarity glycerol") protein kinase, also regulates the ATF2 and Elk-1 transcriptional activators (150, 194, 300, 493).

Ras-mediated transformation, cell cycle progression, and apoptosis. Cdc42p has been implicated in mitogenesis and cell cycle progression, but there are conflicting data concerning its exact role in these processes. Microinjection of activated Cdc42^G12V protein into quiescent Swiss 3T3 fibroblasts led to an increase in the bromodeoxyuridine incorporation into DNA, and expression of dominant negative Cdc42^T17N blocked serum-induced bromodeoxyuridine incorporation (442), suggesting that Cdc42p is necessary for cell cycle progression in these cells. In contrast, microinjection of wild-type Cdc42p into G₁-synchronized NIH 3T3 cells resulted in a dramatic cell cycle arrest at G₁/S, whereas microinjection of activated and dominant negative Cdc42 alleles had a similar but less drastic effect (400). Interestingly, this effect was mediated through the p38 MAP kinase pathway but not the JNK/SAPK pathway. Expression of activated Cdc42^G12V in Rat1 and NIH 3T3 cells did not seem to lead to increases in low-serum growth or growth to high saturation density (466, 494), which is also in contrast to the cell cycle-stimulatory effects seen in Swiss 3T3 cells (see above). It is likely that these discrepancies are due to different signaling mechanisms in these different cell types, but the nature of these important differences has not been elucidated.

Cdc42p and Nef-dependent HIV replication. A fascinating connection has recently been made between Cdc42p and Nef-dependent HIV replication and pathogenesis (for reviews, see references 111 and 564). Nef associated with, and activated, a cellular Ser/Thr PAK-like kinase, the Nef-associated kinase (NAK). NAK shares epitopes with PAK but is not one of the three major PAK isoforms (335, 508) (see "PAK-like kinases" below). Activation of NAK appeared to be mediated through Cdc42p (and Rac), in that expression of dominant negative Cdc42^T17N reduced NAK activity in transfected COS cells and expression of activated Cdc42^G12V enhanced Nef association with and activation of NAK activity (335). This NAK activation also led to serum response element-dependent transcriptional induction that was blocked by Cdc42^T17N expression. In addition, Cdc42^T17N expression led to a reduction in HIV-1 production in transfected COS cells. Determination of the physiological role of Cdc42p in HIV pathogenesis should be active area of investigation in the near future.

Cdc42p/JNK pathway and ion homeostasis. Activation of mammalian NHE1, an integral membrane Na⁺-H⁺ exchanger subtype involved in regulating intracellular ion homeostasis and pH (421), by the G₁₃ GTPase is mediated through the Cdc42-MEKK1-JNK pathway (40, 222, 580, 581). Although it is unclear whether these JNK pathway effects are due to transcriptional activation of the NHE1 gene or induction of NHE1 exchanger activity, expression of dominant negative MEKK1 inhibited the rapid activation of a G₁₃/G_z chimera by the D₂-dopamine receptor (222), suggesting that transcriptional regulation is not involved. Recently, the p115 Rho-GEF has been shown to act as a GAP for G₁₃ and G₁₂ (275), and G₁₃ has been shown to stimulate the guanine nucleotide exchange activity of p115 Rho-GEF (207), which can lead to Tec/Bmx nonreceptor tyrosine kinase-dependent induction of serum response factor-dependent transcription (361-363). These results suggest a regulatory connection between these G-protein  subunits and members of the Rho/Rac family of GTPases. Analysis of G₁₃ knockout mice indicated that G₁₃ functions in thrombin-dependent cell migration, probably in a Cdc42p/actin-dependent manner, and in vascular system development (433). In addition, expression of constitutively active G₁₃ and G₁₂ in differentiated PC12 cells led to Rho-dependent neurite retraction and cell rounding (255). Another interesting connection between Cdc42p and ion channels was the recent identification of the human homolog (IBP72) of the S. pombe Skb1 protein kinase, which interacts with the S. pombe PAK homolog Shk1/Pak1 (166) (see "S. pombe PAK-like kinases" below) in a two-hybrid protein screen for proteins that interacted with the pICln protein, a putative component of the chloride channel (279). Given the unclear nature of the role of pICln in chloride channel activation, the significance of this interaction remains to be determined.

Cdc42p and host cell responses to bacterial invasion. Cdc42p also plays a role in host signaling pathways that are activated in response to invasive bacteria (for reviews, see references 125 and 234). These pathways include actin rearrangements seen in Salmonella and Shigella invasion (see below), as well as JNK pathway-dependent induction of proinflammatory cytokines in epithelial cells in response to Neisseria gonorrhoeae invasion (414). Salmonella typhimurium can induce actin rearrangements and macropinocytosis in host cells (149), and expression of the dominant negative Cdc42^T17N mutant protein in COS1 fibroblasts inhibited this induction and prevented internalization of the bacteria into the host cells (85). Expression of the activated Cdc42^G12V mutant protein did not alter these processes with wild-type Salmonella, but it did allow an invasion-defective Salmonella mutant to invade COS1 and Rat-1 cells. Salmonella invasion also induced JNK kinase activity, and expression of the Cdc42^T17N mutant protein prevented this bacterially induced activation. These Salmonella-induced effects were mediated by the Salmonella SopE protein (599), as evidenced by the ability of SopE both to act as a GEF for Cdc42p and Rac1p and to induce cytoskeletal rearrangements and JNK kinase activation (200). Therefore, aspects of Salmonella invasion are mediated through bacterial proteins impinging on host cell signal transduction pathways.

CDC42P REGULATORS

Top Previous Next References

Guanine Nucleotide Exchange Factors

The activation of G proteins from an inactive GDP-bound state to an active GTP-bound state requires the action of a GEF. While the structures of GEFs for different G-protein families (i.e., Ras, Rho/Rac/Cdc42, Rab, Arf, and heterotrimerics) are quite dissimilar, the mechanism of guanine nucleotide exchange seems to be conserved (for a review, see reference 533). GEFs function by stabilizing the nucleotide-free state of the G protein, through the disruption of both Mg²⁺ and nucleotide binding sites and subsequent GDP dissociation. However, the means by which GEFs promote these effects can be quite different. GEFs for the Rho/Rac/Cdc42 family of GTPases all contain a Dbl homology (DH) domain (Fig. 5), which is a highly -helical (331) catalytically active domain, and a PH domain, which functions in membrane localization and has recently been shown to enhance DH-domain-dependent nucleotide exchange (331).

View larger version (24K): [in this window] [in a new window]   FIG. 5.   Structure of the representative Dbl family member GEFs Cdc24, Scd1, Dbl, and Fgd1. The DH and PH domains are indicated, along with the potential Ca2+ binding domains and Ste4p, Rsr1p/Bud1p, and Bem1p binding domains in Cdc24. See the text for details.

S. cerevisiae Cdc24p. Cdc24p is believed to be the sole GEF for Cdc42p in S. cerevisiae. CDC24 mutants were among the original cdc mutants isolated and characterized by Hartwell et al. (211-213). The initial analysis of cdc24^ts mutants indicated that Cdc24p plays a role in bud emergence, with cdc24^ts mutants exhibiting a first cycle arrest as large, round unbudded cells with multiple nuclei. The presence of multiple nuclei in arrested cells was the first evidence that the budding cycle was independent of the DNA synthesis and nuclear division cycles in S. cerevisiae. Interestingly, overexpression of Cdc24p led to similar morphological phenotypes (504, 641), indicating that either a lack or an excess of Cdc24p leads to a loss of cell polarity, presumably through the disruption of multiprotein complexes (see below). Other phenotypes associated with cdc24^ts mutants were delocalized chitin deposition throughout the cell instead of the typical chitin ring formation at the mother-bud neck region (528, 529) and defects in bud site selection and localized deposition of mannan (528), localized secretion of acid phosphatase (146), and mating (416, 472) (see below). Delocalized chitin deposition in cdc24 mutants is probably due to the altered assembly of the 10-nm filament septin ring, which is necessary for proper chitin ring formation (119). Disruption of CDC24 led to death (106), indicating that Cdc24p had an essential function in cell growth. The cdc24-3 and cdc24-4 mutant alleles displayed abnormal elongated buds when grown at semipermissive temperatures (528), suggesting that Cdc24p plays a role in the apical-isotropic switch (see "S. cerevisiae" under "Functional studies" above). They also displayed defects in bud site selection to a discrete, nonrandom site in both haploids and diploids (528), suggesting that Cdc24p is involved in the bud site selection process (see below). When cdc24-4^ts cells were arrested in S phase with hydroxyurea and then released into media at the restrictive temperature, the predominant arrest phenotype observed consisted of large mother cells with small buds (528), suggesting that Cdc24p also functions after bud emergence to direct growth preferentially into the enlarging bud instead of in the nonenlarging mother cell.

(i) Cdc24p-Rsr1p/Bud1p interactions. Besides interacting with Cdc42p (see "GEF interaction domains" above), Cdc24p interacts with Rsr1p/Bud1p, Bem1p, and the Ste4p G subunit. Rsr1p was originally identified in a genetic screen for multicopy suppressors of a cdc24-4^ts mutant (35). DNA and predicted amino acid sequence analyses indicated that Rsr1p belonged to the Ras subfamily of the Ras superfamily of GTPases. Rsr1p showed 57% identity to the first 120 amino acids of c-Ha-Ras protein and yeast Ras1p and Ras2p; its closest homolog is the Krev-1/rap1a protein, with 56% identity over the entire protein. Deletion of Rsr1p did not result in death, indicating that Rsr1p is not essential for growth, but it did result in a random bud site selection pattern, as occurs with certain cdc24 alleles (see above), suggesting that the role of Rsr1p may be in the selection of the nonrandom site for bud emergence. This possibility was confirmed when rsr1 mutants were identified in a screen for mutants defective in establishing a normal axial budding pattern (80). This screen identified five genes, designated BUD1 through BUD5, of which BUD1 was shown to be allelic to RSR1. Subsequent genetic and biochemical analyses have indicated that Bud2p is a GAP and Bud5p is a GEF for the Rsr1p/Bud1p GTPase (34, 64, 79, 81, 448, 450, 462).

(ii) Cdc24p-Bem1p interactions. Bem1p was identified in three different genetic screens: as a synthetic lethal mutant with msb1 (36), a mutant that was identified as a multicopy suppressor of the cdc24-4^ts and cdc42-1^ts alleles (35); as a synthetic lethal mutant with a bud5 allele (79); and as a mutant with mating defects that were due to an inability to form mating projections (90, 91). The bem1^ts alleles analyzed displayed large, unbudded, multinucleate cells with delocalized chitin and actin reminiscent of cdc24 and cdc42 alleles. Bem1p is not essential but is important for growth, in that bem1 cells grew slowly at 23 and 30°C, but were ts and cs for growth (90); however, this phenotype was strain specific in that the bem1 mutation was lethal in other strain backgrounds (294). Bem1p fractionated to both particulate and soluble pools and appears to be a phosphoprotein, as evidenced by a protein mobility shift on sodium dodecyl sulfate-polyacrylamide gel electrophoresis upon phosphatase addition (303). Bem1p localized to the site of incipient bud emergence and to the tips of small buds (21, 465), as do Cdc24p and Cdc42p. This localization was not disrupted by incubation with the actin-depolymerizing drug latrunculin-A (21), suggesting that Bem1p localization occurs independently of actin localization.

rho3 rho4

boi1 boi2

(iii) Cdc24p-Ste4p interactions. Several lines of evidence suggest that Cdc24p functions within the S. cerevisiae mating pathway through interactions with the Ste4p G subunit and Cdc42p and that it functions prior to cell-cell fusion, possibly by affecting the orientation of mating-projection formation in response to localized high concentrations of pheromone (91, 416, 472, 527, 626). The original cdc24^ts mutants had reduced mating efficiencies (472), and the cdc24-4^ts strain had modest defects in pheromone-induced, Ste4p-dependent transcriptional activation. These defects could be suppressed by overexpression of wild-type or activated CDC42 alleles (527, 626). New cdc24 mutant alleles have been identified in genetic screens for mutants that have either reduced mating efficiencies with mating-enfeebled partners (91) or reduced mating efficiencies with wild-type partners but no effects on the vegetative role of Cdc24p (i.e., wild-type growth, morphology, bud site selection, and actin distribution [416]). The latter cdc24-m1, cdc24-m2, and cdc24-m3 mutants had wild-type phenotypes with respect to pheromone-induced cell cycle arrest, transcriptional activation, mating-projection formation, and actin polarization but had defects in cell-cell fusion, an inability to properly orient to a mating-pheromone gradient, and decreased mating efficiencies with mating-enfeebled partners, suggesting that Cdc24p may be playing a role in mating-projection orientation in response to mating pheromone.

S. pombe GEF. A potential Cdc24 homolog, named Scd1p, was identified in S. pombe in a genetic screen for mutants with mating defects and round cells (73). Scd1p showed 32% identity to Cdc24p, and this identity was found throughout the coding region, including amino acids 194 to 254 (containing the DH domain) (Fig. 5). Scd1p is the same protein as Ral1p, also identified as a mutant with mating and morphological defects (159). Surprisingly, a scd1 deletion did not lead to cell inviability, as a cdc24 deletion does in S. cerevisiae, but it did result in mating defects and round cells. This result brings into question whether Scd1p is the sole physiological GEF for the essential GTPase Cdc42p; this has not been tested biochemically to date. Overexpression of S. cerevisiae Cdc24p could partially suppress the mating and morphological defects of a scd1 mutant, and although overexpression of S. pombe Cdc42p could not suppress the same defects, it did enhance the suppression by S. cerevisiae Cdc24p. This result is again different from results with S. cerevisiae, in which overexpression of Cdc42p can suppress cdc24^ts mutants. In two-hybrid protein assays, Scd1p could interact with Scd2p, a potential Bem1p homolog that was identified in the same genetic screen. In addition, Scd1p could interact with S. pombe Cdc42p, but only when Scd2p or S. pombe Ras1p was overexpressed in the same cells, suggesting that Scd1p may have to be bound to Scd2p or activated by Ras1p in order to interact with Cdc42p. Scd1p also interacted with Ras1p in the presence of overexpressed Scd2p or after the N-terminal 671 amino acids of Scd1p were deleted (Scd1Np); it could also interact with activated mutations of human H-Ras. Interactions between Scd1Np and Scd2p were corroborated by GST affinity chromatography experiments. While interactions between S. pombe Scd1p, Scd2p, and Ras1p are reminiscent of interactions between S. cerevisiae Cdc24p, Bem1p, and Rsr1p/Bud1p (see above), there remain questions about the physiological role of Scd1p in S. pombe as a potential GEF for Cdc42p.

Drosophila and C. elegans GEFs. Three potential Cdc42 GEFs have been identified in Drosophila to date: Drt-GEF (589), still life (sif) (532), and DRho-GEF2 (29, 190). However, none has been shown to have either in vitro or in vivo GEF activity against Cdc42p. The 658-amino-acid Drt-GEF contains an N-terminal SH3 domain followed by a DH domain and a PH domain. Drt-GEF mRNA is expressed during oogenesis and embryogenesis and seems to be concentrated in the ventral furrow, cephalic furrow, posterior midgut, and anterior midgut involutions, areas which undergo actin-dependent morphological changes during gastrulation. The still life (sif) mutant was identified by reduced locomotor behavior, and the sif gene encodes two differentially spliced products. Both contain two PH domains, a PDZ domain, and a DH domain as well as potential PEST sequences. The sif mRNA was found predominantly in the brain and ventral nerve cord, and anti-Sif antibodies were localized to the neuropils, the location at which neurites form synapses, in both embryonic and adult brains. Overexpression of full-length Sif was not associated with a phenotype, but overexpression of a N-terminal truncation mutant missing the sequences before the first PH domain resulted in defects in axonal extension in Drosophila and induced membrane ruffling in human KB cells. This mutant protein colocalized with actin structures at the ruffles. DRho-GEF2 was identified as a dominant suppressor, Su(Rho1)2B, of DRho1 overexpression (29) and in a screen for maternal effects of zygotic lethal mutations (190). The Su(Rho1)2B mutation could not suppress the DRac1 and Cdc42Dm overexpression phenotypes, suggesting that DRho-GEF2 has Rho1-specific functions. DRho-GEF2 is ~284 kDa and contains an N-terminal PDZ domain followed by a potential phorbol ester/diacylglyceride binding domain, a DH domain, and a PH domain. Mutants defective in DRho-GEF2 function have abnormalities in ventral-furrow formation and anterior and posterior midgut invaginations that are qualitatively similar to those associated with ectopic expression of a dominant negative Rho1^N19 mutant protein and not seen with ectopic expression of dominant negative Cdc42 or Rac1 mutant proteins. This result reinforces the possibility that DRho-GEF2 is an in vivo GEF for Rho1.

Mammalian GEFs. There are multiple potential Cdc42p-GEFs in mammalian cells, including the Dbl, Dbs, Ost, Bcr, and Abr oncoproteins, the Tiam-1 invasion-inducing protein, the PAK-interacting exchange factor PIX, the FGD1 faciogenital dysplasia protein, and the Brx estrogen receptor binding auxiliary protein (for reviews, see references 44, 71, 441, and 594). The prototypical mammalian GEF is the Dbl oncoprotein, originally identified by malignant transformation of NIH 3T3 cells with transfected DNA from a human B-cell lymphoma (138, 534). Comparison of the predicted amino acid sequence of proto-Dbl with the Cdc24p GEF and the Bcr oncoprotein identified a ~200-amino-acid domain (residues 498 to 674 in Dbl) with significant sequence similarity (491). Deletion of or mutations in this DH domain (Fig. 5), including the replacement of the highly conserved LLLKELL sequence at amino acids 640 to 646 with the conservative IIIRDII sequence, resulted in loss of Dbl-transforming activity and GDP dissociation activity (206, 491), suggesting that the DH domain is necessary for both activities. Expression of the Dbl DH domain by itself in NIH 3T3 cells did not lead to cellular transformation (634), but fusion of the DH domain to GST resulted in a protein with fully functional exchange activity against Cdc42Hs (206), suggesting that the DH domain is not sufficient for transformation but is sufficient for GEF activity. It should be noted that DH domains have been found in all known or potential Cdc42-GEFs identified to date but that the presence of a DH domain does not determine that a protein will have in vivo Cdc42-GEF activity. Dbl has in vitro GEF activity (i.e., ability to stimulate GDP dissociation from and promote GTP binding to a GTPase) against platelet and recombinant Cdc42Hs (205, 206) as well as RhoA and membrane-bound Rac1 (206, 443, 611), and it is able to bind to the Rho family members Cdc42Hs, RhoA, and Rac1 (206) and murine Cdc42, Rho, RhoC, Rac1, and RhoG (389) in GST affinity chromatography experiments, suggesting that it has a broad in vitro specificity range. Dbl binding to Cdc42 was observed with nucleotide-free Cdc42p and to a lesser extent with GDP-bound Cdc42p but not with GTP-bound Cdc42p (206, 389), suggesting that Dbl may stabilize a nucleotide-free or GDP-bound state of Cdc42p.

The transition of G proteins from an active GTP-bound state to an inactive GDP-bound state occurs through the intrinsic hydrolysis of GTP to GDP + inorganic phosphate (P_i), a process that can be significantly stimulated by the action of GAPs. Cdc42p displays a ~10-fold higher intrinsic (GAP-independent) rate of GTP hydrolysis compared to Ras proteins (210), and this rate can be further stimulated by the addition of GTPase-activating proteins. It has been postulated that differences in the GTP binding domain (residues 115 to 118 in Cdc42p) may account for the higher rate of GTP hydrolysis, but this has not been experimentally tested to date. Recently, it was shown that Cdc42p can undergo homodimer formation in vitro and that this homodimer formation can lead to a Cdc42-GTP-stimulated increase in intrinsic GTPase activity (623). In addition, the C-terminal polybasic region of Cdc42p, and specifically the R186 residue, was shown to be necessary for homodimer formation and for this GAP activity (622, 623). Introduction of the K186R mutation into S. cerevisiae Cdc42p leads to an increase in intrinsic GAP activity in vitro, a ts loss-of-function phenotype in vivo, and abnormal cell morphologies at the permissive temperature (622), reinforcing a potential physiological role of the polybasic region in intrinsic GAP activity.

S. cerevisiae Cdc42p GAPs. Bem3p, Rga1p/Dbm1p, and Rga2p are three potential Cdc42p GAPs identified in S. cerevisiae, but only Bem3p has been shown to have GAP activity against Cdc42p in vitro (630). Bem3p was originally isolated as a multicopy suppressor of bem2^ts mutants (36, 630). The 125-kDa Bem3p contains a C-terminal domain (residues 977 to 1140) with significant similarity to Bem2p and other Rho-GAPs (630), and this GAP domain can be subdivided into three subdomains with various levels of sequence similarity (633). Bem3p also contains a PH domain (residues 633 to 739). An E. coli-produced GST-Bem3p GAP domain (residues 751 to 1128) fusion protein had in vitro GAP activity against a GST-Cdc42p fusion protein that was not competed by the GST-Bem2p GAP domain fusion protein (630). The GST-Bem3p GAP domain fusion protein also had in vitro GAP activity against human Cdc42Hs but not against the GTPase-defective Cdc42^G12V mutant protein, and it did not affect the binding of GTP to Cdc42Hs (633). Analysis of Bem3p GAP domain deletions indicated that all three GAP homology subdomains were necessary for GAP activity, but analysis of chimeras between Bem2p and Bem3p GAP subdomains indicated that the two N-terminal subdomains were sufficient for Cdc42p binding and GAP activity, albeit at ~30% of the GST-Bem3p levels (633). Bem3p interacted with Cdc42p, but not with other Rho-type GTPases, in a two-hybrid protein assay; this interaction was enhanced with the GTPase-defective Cdc42^Q61L mutant protein (537). A bem3 strain was viable with no morphological abnormalities (references 537 and 630 and data not shown).

rga1 bem3

Drosophila and C. elegans GAPs. To date, the only potential Cdc42-GAP identified in Drosophila is the RnRac-GAP, the product of the rotund locus (7, 180, 181). Although RnRac-GAP has not been shown to have GAP activity against Cdc42p, overexpression of RnRac-GAP led to defects in actin organization similar to those seen with Drosophila cdc42 mutants (see "Drosophila" under "Functional studies" above).

Mammalian GAPs. At least 12 mammalian proteins have in vitro GAP activity against Cdc42p, including CDC42GAP/p50rhoGAP (28, 163, 164, 210, 293, 403), Bcr (124), Abr (95, 215, 548), p190GAP (518), n-chimaerin (8, 276, 357), 3BP-1 (98, 99), Graf (219, 553), RalBP1/RLIP76/RIP1 (65, 244, 451), MgcRacGAP (562), PARG1 (506), myr5 (406, 474), and Cd-GAP (292). However, most of these proteins also have in vitro activity against Rac and Rho proteins (for review, see references 290 and 572), and so the assignment of a subset of these GAPs as specific in vivo Cdc42-GAPs has proven difficult.

S. cerevisiae Rdi1p. Rho-GDI was purified from S. cerevisiae cytosolic fractions by assaying for an activity that inhibited the dissociation of [³H]GDP from bovine rhoA (373). Based on peptide sequences, the Rho-GDI gene, RDI1, was isolated and it encoded a ~23-kDa polypeptide with 36% identity to human and bovine Rho-GDIs. A GST-Rdi1 fusion protein was active on prenylated yeast Rho1p and mammalian RhoA and Rac1 but not on nonprenylated Rho1p, suggesting that Rdi1p interacts with Rho-like GTPases through the C-terminal membrane localization domain. Disruption of Rdi1p did not lead to death or defects in mating, sporulation, heat shock sensitivity, or budding pattern, suggesting that the protein is not essential for growth or morphogenesis. However, overexpression of Rdi1p, as well as bovine Rho-GDI, resulted in cell death. The morphological phenotypes associated with this cell death have not been reported, but if Rdi1p can extract Cdc42p from cellular membranes as mammalian Rho-GDI can, one would predict that the phenotypes would resemble cdc42 loss-of-function phenotypes. Myc-tagged Rdi1p coimmunoprecipitated with HA-tagged Rho1p and Cdc42p (267), suggesting that Rdi1p could interact with both GTPases within the cell. Rdi1p fractionated exclusively into soluble fractions, and immunofluorescence microscopy with HA-tagged Rdi1p indicated that Rdi1p was present in the cytosol. There was no change in the fractionation pattern of overexpressed HA-Cdc42p in a rdi1 mutant compared to the wild type, but overexpression of Rdi1p led to an increase in soluble Cdc42p (267). However, it should be noted that in these experiments, HA-tagged Cdc42p was found predominantly in soluble fractions, which is opposite from the predominant particulate fractionation pattern of endogenous Cdc42p (643). It remains to be seen if Rdi1p displays GDI activity against Cdc42p as exhibited by mammalian Rho-GDI (see below).

Mammalian GDIs. Three Cdc42-GDI proteins have been identified to date in mammalian cells; they are currently designated RhoGDI (previously rhoGDI [for a review, see reference 549]), RhoGDI (previously LD4, LyGDI, D4, or D4/LyGDI [4, 306, 409, 459, 509]), and RhoGDI (also known as RhoGDI-3). The first Cdc42-GDI was purified from bovine brain cytosol by its ability to inhibit the dissociation of labeled GDP from Cdc42Hs (307); it effectively inhibited Dbl-catalyzed GDP dissociation as well. Limited peptide sequence from this 28-kDa purified protein suggested that it was identical to the previously isolated 28-kDa Rho-GDI (160, 569), which was shown to act as a GDI against Cdc42Hs (307) and to cosediment with Cdc42Hs (471). Rho-GDI mRNA and protein were expressed in the brain, lungs, thymus, spleen, small intestines, and kidney (160, 524), suggesting that Rho-GDI was a ubiquitous and promiscuous regulator of Rho-type GTPases. This GDI activity required the isoprenylation of Cdc42Hs, in that unprenylated Cdc42Hs isolated from E. coli was not responsive to the GDI (208, 307) and addition of the prenylation inhibitor lovastatin altered the Cdc42Hs sedimentation profile toward a noncomplexed protein (471). Addition of purified bovine brain Rho-GDI to either human placental membranes or membranes from human epidermoid carcinoma (A431) cells containing Cdc42Hs resulted in a significant dissociation of Cdc42Hs from the particulate pool into the soluble pool, and this dissociation seemed to be insensitive to the Cdc42Hs-bound guanine nucleotide (307), suggesting that the GDI can interact with both GDP-bound and GTP-bound Cdc42Hs. A GST-Rho-GDI fusion protein could also efficiently extract RhoA and Cdc42Hs, but not Rac1, from rat liver membranes (353). The Rho-GDI, either native or as a GST fusion protein, also inhibited the intrinsic and GAP-stimulated GTPase activity of Cdc42Hs, and this GTPase inhibitory protein activity also required isoprenylation of Cdc42Hs (208).

S. cerevisiae PAK-like kinases. Ste20p, Cla4p, and Skm1p are the three members of the PAK family of serine/threonine protein kinases found in S. cerevisiae (for a review of PAKs, see reference 516). All three have a highly conserved protein kinase domain in their C termini as well as a CRIB domain in their N termini. Cla4p and Skm1p differ from Ste20p in that they contain a PH domain located N-terminal to the CRIB domain. There are clear instances of overlapping functions for these PAK-like kinases in regulating actin-dependent growth during the cell cycle, the pheromone response pathway, and filamentous growth. However, it remains to be determined whether these overlaps are physiologically relevant or are due to imposed artificial stresses uncovering biochemical redundancies.

(i) Ste20p. Ste20p was identified in two different genetic screens by its ability, when overexpressed on a multicopy plasmid, to either suppress the sterility caused by overexpression of the dominant negative ste4^D62N mutant allele in a ste4 background (295) (Ste4 encodes the G subunit of the pheromone response pathway heterotrimeric G protein [596]) or induce the transcriptional activation of a FUS1-lacZ fusion protein independent of added pheromone (469). Disruption of the Ste20p kinase domain resulted in cells with defects in mating, pheromone sensitivity, transcriptional induction of mating-specific genes, and induction of mating projections (295, 469), while overexpression of N-terminal truncation mutants in which the CRIB domain and other sequences, but not the kinase domain, were deleted, caused death (298, 469). Epistasis experiments with mutant alleles of other components of the pheromone response pathway indicated that Ste20p functioned at or just below the level of the G protein but above the Ste11p-Ste7p-Fus3p/Kss1p MAP kinase signaling module. Recent data indicates that Ste20p interacts directly with the Ste4p G subunit and that this interaction is enhanced upon pheromone addition (304). A small (14-amino-acid [ANSSLAPLVKLARL]) domain C-terminal to the kinase domain was necessary and sufficient for this interaction, and mutating the underlined Ser and Pro residues led to loss of this interaction. Interestingly, the Cla4p kinase (see below) does not have this highly conserved domain and interacts weakly with Ste4p, suggesting that these interactions play a role in the in vivo specificity of the PAKs.

ste20 cla4

cla4 ste20

(ii) Cla4p. Cla4p was originally identified in two genetic screens designed to identify mutants unable to survive in the absence of the two G₁ cyclins Cln1p and Cln2p (39, 113). The cla4/erc10 cln1 cln2 triple mutants displayed abnormal morphologies, including elongated buds, wide mother-bud necks, and multinucleate cells, indicative of a delay in either the apical-isotropic bud growth switch and/or nuclear division and/or a cytokinesis defect. Another mutant allele identified in the screen was cla10, which was allelic to CDC12, a member of the septin family of proteins that comprise the 10-nm filaments laid down at the mother-bud neck region in late G₁ post-START and are necessary for proper cytokinesis (see below). Cla4p showed significant sequence similarity to Ste20p within its kinase and CRIB domains, with the exception that Cla4p contained a PH domain in its N terminus (112). Deletion of the CRIB domain or the PH domain resulted in a nonfunctional or partially functional protein (38). Deletion of the Cla4p catalytic kinase domain did not result in death but did result in morphogenetic defects similar to the original cla4 mutant (112).

cla4 ste20

cla4-75^ts ste20

(iii) Skm1p. Skm1p was identified on chromosome XV through the S. cerevisiae genome-sequencing project (371). It exhibited higher sequence similarity to Cla4p than to Ste20p, including the presence of a PH domain in its N terminus. Skm1p showed weak interactions with Cdc42p in two-hybrid protein assays, but it interacted more strongly with Cdc42^G12Vp (475a), indicating that it was a bona fide downstream effector of Cdc42p. As with Ste20p, this interaction was not affected by the Cdc42^V44A effector domain mutation, suggesting that Skm1p interacts with another subdomain of the Cdc42p effector domain. Disruption of Skm1p did not cause death (371), indicating that Skm1p does not play an essential role in cell growth. Loss of Skm1p also did not show adverse effects on cellular morphologies, bud site selection, growth on high-osmolarity media, or mating, and did not show a synthetic lethal phenotype with a cla4 or ste20 mutant or mutations in the CDC10 septin, the Rho-GAP BEM2 or CDC42. Expression of Skm1p on a high-copy-number plasmid could not suppress the cla4 or ste20 mutant phenotype; overexpression under an inducible promoter was not tested. However, overexpression of a GST-Skm1p fusion protein led to an abnormal cellular morphology of large, round, multinucleate cells with small, sometimes multiple, buds, suggesting that Skm1p may be functioning in cellular morphogenesis. Deletion analysis indicated that overexpression of the catalytic kinase domain was responsible for this phenotype as well as for an ability to suppress ste20 mating defects. Overexpression of the Skm1p catalytic domain also led to severe growth defects similar to those seen with N-terminal truncation mutants of Ste20p and Cla4p, reinforcing the notion that the N-terminal domain of these PAK-like kinases containing the CRIB domain plays a negative regulatory function on kinase activity that is relieved by binding of Cdc42p. A detailed genetic analysis of SKM1 is needed to pinpoint its cellular function.

S. pombe PAK-like kinases. There are two known PAK homologs in S. pombe, Pak1p/Shk1p (364, 447) and Pak2p/Shk2p (515, 616). The pak1⁺ (447) and shk1⁺ (364) gene product was isolated in two independent screens by degenerate oligonucleotide PCR based on S. cerevisiae Ste20 sequences; pak2⁺ was isolated in a cDNA library screen with the pak1 PCR product (447). The pak1⁺/shk1⁺ gene encoded a 72-kDa protein with significant amino acid identity to the PAK family, especially within the kinase domain and the N-terminal CRIB domain, and was shown to have in vitro autophosphorylation activity predominantly on Ser residues (447). Analysis of pak1/shk1 mutants indicated that Pak1p/Shk1p was essential for cell growth, with mutant cells exhibiting a small, round cellular phenotype reminiscent of cdc42 null mutants (390). Pak1p/Shk1p preferentially interacted with GTP-Cdc42p in GST affinity chromatography experiments and with the Cdc42^G12V activated allele in two-hybrid protein assays (447). This interaction was abolished with the T35A effector domain mutation (447) and was not seen with the Cdc42^T17N dominant negative allele (364, 447), suggesting that Pak1p/Shk1p is a bona fide downstream effector of Cdc42p. This point was corroborated by the observations that co-overexpression of mutant alleles of cdc42 and pak1 led to lethal growth and morphology defects (447) and that overexpression of wild-type Pak1p or a kinase-defective K415,416R mutant protein (447) or a C-terminal truncation mutant protein that still contained the CRIB domain (364) resulted in cells with abnormal morphologies and delocalized cortical actin structures. Both Pak1p/Shk1p and Cdc42p also functioned in the mating pathway, as evidenced by the reduced mating in the cdc42^T17N dominant negative mutant and the pak1^K415,416R kinase-defective mutant (364, 447), by the ability of Pak1p/Shk1p to partially suppress the cdc42^T17N mating defect (364) and the S. cerevisiae ste20 mating defect (447), and by the ability of a Pak1p/Shk1p N-terminal deletion mutant protein to activate the S. cerevisiae pheromone response pathway (364) and to interfere with two-hybrid protein interactions between the S. pombe Byr1 and Byr2 protein kinases involved in Ras-mediated pheromone response (568).

Drosophila and C. elegans PAK-like kinases. The Drosophila PAK homolog (DPAK) was identified by low-stringency DNA-DNA hybridization from an embryonic cDNA library (199). The 76-kDa DPAK contained a highly conserved CRIB domain as well as a Ser/Thr kinase domain. A GST fusion to the DPAK N-terminal CRIB domain bound to Drosophila RacA (DRacA) and Cdc42p (Dcdc42) and human Rac1 and Cdc42p (data not shown) in an overlay assay (199). DPAK mRNA and protein were localized ubiquitously throughout embryonic development, with elevated localization in epidermal cells associated with the dorsal vessel and muscle attachment sites as well as the central nervous system. DPAK colocalized with antiphosphotyrosine antibodies to focal adhesions and focal complexes, and with F-actin caps in the syncytial blastoderm and the leading edge of epidermal cells during dorsal closure over the amnioserosa, a process that is inhibited by expression of a dominant negative DRacA transgene (198). Therefore, it is unclear whether DPAK is a physiological Cdc42 effector.

Mammalian PAK-like kinases. Mammalian PAKs act in response to a variety of intracellular and extracellular signals to mediate a number of different cellular events including growth factor- and stress-induced actin rearrangements and activation of the JNK/SAPK and p38 MAP kinase pathways (see below) (see "Mammals" under "Functional studies" above), Nef- and Nef-associated kinase-dependent HIV-1 replication and pathogenesis (111, 335, 508, 564), thrombin cleavage in platelets (554), cleavage arrest in frog embryos (492), Schwann cell transformation (551), CD28-dependent antigen-specific activation of T cells (248, 249), and T-cell receptor-mediated activation of the nuclear factor of activated T cells transcription factor (609). The original mammalian PAK, designated p65^PAK, was identified as a rat brain protein that interacted with [-³²P]GTP-GST-Cdc42 and [-³²P]GTP-GST-Rac1, but not with [-³²P]GTP-GST-RhoA, in an overlay assay (359). This interaction was specific for the GTP-bound form of Cdc42p, suggesting that p65^PAK could function as a downstream effector of Cdc42p function. Purified p65^PAK displayed Ser/Thr autophosphorylation activity that was stimulated by GTP-bound Cdc42p and Rac1p and had kinase activity against the exogenous myelin basic protein substrate. Currently, there are three major 62- to 68-kDa PAK isoforms in mammalian tissues, designated (in the nomenclature of Sells and Chernoff [561]) PAK1 (previously p65^PAK, PAK, and hPAK-1 [24, 55, 104, 265, 359]), which is found predominantly in brain, muscle and spleen tissue; PAK2 (previously -PAK, PAKI, and hPAK65 [235, 265, 492, 554]), which is ubiquitous; and PAK3 (previously PAK and mPAK-3 [24, 355]), which is found in brain tissue (for reviews, see references 264 and 516 and references therein). These three isoforms all contain several N-terminal proline-rich domains followed by a CRIB domain that interacts with GTP-bound Cdc42p through its effector domain and a C-terminal Ser/Thr protein kinase domain, whose activity is stimulated by binding of Cdc42p and Rac to the CRIB domain. Although it was believed that the CRIB domain played a role in autoinhibition of PAK activity, recent data suggest that a highly conserved domain C-terminal to the CRIB domain functions in this capacity (151). The PAK N-terminal proline-rich domains can interact with SH3 domains within the Nck adapter protein (47, 161, 263, 334, 468, 517, 542), thereby forming a linkage between growth factor receptors and activation of the PAKs. Interestingly, another Cdc42p effector, the Wiskott-Aldrich syndrome protein (WASP), also interacts with Nck through its SH3 domains (486) (see "Wiskott-Aldrich syndrome proteins mediate actin rearrangements" below).

ACK tyrosine kinases. Two tyrosine protein kinases, enriched in mammalian brain and skeletal muscle tissue, have been identified as specific effectors of Cdc42p. These tyrosine kinases, designated ACK-1 (358) and ACK-2 (617), specifically interact with GTP-bound Cdc42p in vitro and in vivo and contain a CRIB domain along with a tyrosine kinase catalytic domain, an SH3 domain, and a proline-rich domain. The function of ACK-1 is unknown, but incubation of ACK-2-transfected, detached (not adherent) COS7 cells with EGF or bradykinin resulted in an increase in ACK-2 phosphorylation, suggesting that ACKs may link serpentine/G-protein-coupled receptors to Cdc42p signaling pathways.

Bni1p (139, 232, 236, 269, 619) is a ~220-kDa protein that can interact with Rho-type GTPases in S. cerevisiae. It contains four functional domains including a N-terminal Cdc42/Rho interaction domain contained within amino acids 90 to 343; a proline-rich formin homology 1 (FH1) domain (amino acids 1230 to 1330) found in a number of formin family members including the S. pombe genes fus1 (457) and cdc12 (74), Drosophila genes diaphanous (68) and cappuccino (134), the Aspergillus nidulans gene figA/speA (203, 365), and vertebrate formins (342, 566, 585, 600); a formin homology 2 (FH2) domain (amino acids 1516 to 1616); and a C-terminal Bud6p/Aip3p binding domain (within amino acids 1647 to 1953). It should be noted that the only other formin-like protein that has been shown to interact with Rho-type GTPases is murine p140mDia (585), a homolog of Drosophila diaphanous that specifically interacted with GTP-bound RhoA (in vitro interactions with Cdc42p have recently been reported [10]), localized to spreading lamellae and cleavage furrows in Swiss 3T3 cells, and colocalized with RhoA and profilin (see below) in membrane ruffles in HT1080 human fibrosarcoma cells. However, several other formin-like proteins have been implicated in cell polarity processes in their respective organisms.

Bni1p interacted with Cdc42p in two-hybrid protein assays, and this interaction was specific for GTP-bound Cdc42^G12Vp and not GDP-bound Cdc42^D118Ap (139). This interaction was substantiated by the in vitro binding of an HA-tagged fragment of Bni1p (amino acids 1 to 1214) purified from S. cerevisiae on Sepharose beads to GTPS-bound, but not GDP-bound or nucleotide-free, Cdc42p. Bni1p was also identified by a two-hybrid protein interaction with the activated Rho1^Q68L mutant protein (269). This interaction was between Rho1^Q68Lp and amino acids 90 to 489 of Bni1p, and deletion analysis indicated that amino acids 90 to 343 were capable of interacting with Rho1^Q68Lp. The two-hybrid interaction was abolished by the Rho1^T42A effector domain mutation, suggesting that Bni1p was a downstream effector of Rho1, but a maltose binding protein fusion to amino acid 1 to 524 of Bni1p could bind nonspecifically in vitro to both GDP-bound and GTPS-bound Rho1p. Bni1p may also interact with Rho3p and Rho4p (unpublished results cited in reference 139), suggesting that Bni1p may be a general effector of Rho-like GTPases in S. cerevisiae.

In two-hybrid protein assays, GST affinity chromatography experiments, and a maltose binding protein tag overlay assay, Bni1p also interacted with the S. cerevisiae Pfy1p profilin, and this interaction occurred through the Bni1p FH1 domain (139, 232). This binding was also substantiated by a loss of interaction with a profilin mutant protein, Pfy1p-3, that had defects in polyproline binding but not actin binding (139). Several C-terminal fragments of the actin-binding protein Bud6p/Aip3p (15), as well as the Act1p actin protein, displayed two-hybrid protein interactions with Bni1p (139). Bni1p-Bud6p two-hybrid protein interactions occurred through the C-terminal ~300 amino acids of Bni1p, and interactions with actin occurred through the FH1 domain; these interactions may be mediated through Bni1p interactions with profilin (see above). Bni1p can also interact with elongation factor 1 (EF1), a protein that has actin binding activity, through a domain between the FH1 and FH2 domains, and the binding of Bni1p to EF1 led to a loss of EF1-actin binding (570). Bni1p also interacts with the SH3 domain of the Myo3p myosin as assayed by two-hybrid protein assays and GST affinity chromatography (48a). Interestingly, the SH3 domains of the Myo3p and Myo5p myosins also bind to the proline-rich protein verprolin (Vrp1p) (16), which has previously been implicated in cell polarity (571). Bni1p displays genetic and physical interactions with Spa2p, a protein of unknown function that localizes to regions of polarized growth, and localization of Bni1p to the tips of enlarging buds requires the presence of Spa2p and the N-terminal Cdc42/Rho interaction domain (157), suggesting that Bni1p is tethered to the plasma membrane through binding to either a Rho-type GTPase or Spa2p or both. Recent results indicate that Spa2p can interact with Bud6p as well as components of several MAP kinase cascades and the Pea2 cell polarity protein (523), suggesting that actin, profilin, verprolin, Myo3 and Myo5 myosins, Bni1p, Spa2p, Bud6p, EF1, Pea2p, and possibly other actin binding proteins may form a multiprotein complex with Cdc42p during the bud emergence process (Fig. 3).

BNI1 was identified in a screen for mutants that exhibited a randomization of bud site selection in cells exhibiting a bipolar budding pattern (619). A bni1 disruption mutant, in which amino acids 1228 to 1414 containing the FH1 domain were replaced, grew poorly at high temperatures (269); whether this was a true null mutant is unclear, given that the Cdc42/Rho interaction domain would be predicted to still be expressed in this truncated Bni1p. In addition, a transposon insertion mutation in BNI1 led to a defect in filamentous growth (404), suggesting that Bni1p may either mediate actin rearrangements during pseudohyphal growth or be required for the bipolar budding pattern necessary for pseudohyphal formation. Genetic evidence for a role of Bni1p in Rho1p function came from synthetic lethal phenotypes observed between this bni1 disruption mutant and a mutant expressing mammalian RhoA in place of S. cerevisiae Rho1p and between a pkc1 mutant defective in protein kinase C (269), a known downstream effector of Rho1p (251, 423). BNI1 was also shown to be allelic to SHE5, mutants of which are defective in transcriptional expression from the HO endonuclease promoter in mother cells (236); another mutant identified in this screen, she1, was found to be in the MYO4 gene, a class V type minimyosin (188). An HA-epitope-tagged Bni1p localized to the tips of mating projections in pheromone-arrested cells (139), which correlated with the isolation of bni1 mutants in a screen for mutants defective in mating (139). This defect was due to an inability to form pheromone-induced mating projections in bni1 mutants, which was a consequence of a depolarized cortical actin cytoskeleton. This phenotype could not be suppressed by a Bni1 mutant protein lacking its FH1 domain (unpublished results cited in reference 139), reinforcing the role of the FH1 domain in actin interactions.

Overexpression of full-length Bni1p had no phenotypic effect, but overexpression of the Bni1N N-terminal truncation protein, which was missing amino acids 1 to 451 containing the Cdc42/Rho interaction domain, resulted in cell death and a dominant negative phenotype of large, round, unbudded cells with a delocalized cortical actin cytoskeleton and an increased number of cortical actin patches and actin cables (139). This result suggested either that the essential polarization of cortical actin to the site of bud emergence was dependent on Cdc42p binding to Bni1p or that overexpression of the C-terminal portion of Bni1p could lead to the nonproductive sequestration of actin or actin-binding proteins. The striking appearance of cortical actin structures around the periphery of these cells suggested that plasma membrane localization of cortical actin, albeit nonpolarized, may be possible in the absence of Cdc42p binding to Bni1p; it would be very interesting to determine if the Bni1N truncation protein localizes to the plasma membrane and is capable of cross-linking actin in the absence of Cdc42p. The Bni1Np dominant negative phenotype could be suppressed by overexpression of Pfy1p profilin and the two tropomyosins Tpm1p and Tpm2p, suggesting that loss of Cdc42p-Bni1p binding resulted in a Bni1p that can sequester Pfy1p profilin or other actin binding proteins in a nonfunctional manner. Interestingly, overexpression of Pfy1p profilin inhibited the growth of a bni1 disruption mutant (232). Although the cellular morphologies and actin localization patterns associated with this profilin-based inhibition of growth were not reported, this result suggests that the interactions between these two proteins is important for their functions.

A sequence homology search of the S. cerevisiae genome database revealed the presence of a protein, designated Bnr1p, with significant amino acid sequence homology to Bni1p (232). The smaller (1,374-amino-acid) Bnr1p exhibited 19% identity in the Cdc42/Rho interaction domain, 44% identity in the FH1 domain, and 35% identity in the FH2 domain of Bni1p. Bnr1p also interacted with the Pfy1p profilin protein in two-hybrid protein assays, and this interaction also occurred through the Bnr1p FH1 domain. While a bnr1 disruption mutant was viable at all temperatures tested, a bni1 bnr1 double mutant displayed a ts growth defect at 33°C and an arrested phenotype of large, round, unbudded, multinucleate cells with delocalized actin and chitin, highly reminiscent of cdc42 loss-of-function alleles. Interestingly, bud site selection is randomized in haploid bnr1 mutants but not diploids, which is the opposite of the diploid bud site selection defects in a bni1 mutant (619). Also, a bni1 bnr1 double mutant was sensitive to growth on 1 M sorbitol (232), as were pfy1 (187) and certain act1 (424) mutants, suggesting that these proteins function in osmoregulation. In two-hybrid protein assays, Bnr1p interacted only with wild-type and activated (Q70L mutant) Rho4p but not with Rho1p, Rho2p, Rho3p, or Cdc42p, and maltose binding protein-tagged Bnr1p containing the Rho interaction domain of amino acids 63 to 421 bound in vitro to GTPS-Rho4p but not GDP-Rho4p (232). In addition, Bnr1p and Bni1p interacted with the SH3 domain-containing protein Hof1p, which displayed sequence similarity to the S. pombe Cdc15 protein involved in cytokinesis, and HA-tagged Bnr1p and Hof1p localized to the mother-bud neck region (252). Taken together, these data suggest that Bni1p and possibly Bnr1p can act as scaffold proteins juxtaposing Cdc42p and other Rho-like GTPases with actin and actin binding proteins during bud emergence and possibly cytokinesis, thus serving as a critical link between the Cdc42p-dependent signal transduction machinery and its ultimate target, the cortical actin cytoskeleton (Fig. 3).

S. cerevisiae Iqg1p/Cyk1p functions during cytokinesis. Mammalian IQGAP proteins are potential scaffold proteins that interact with Cdc42p, actin, and calmodulin (see below). In S. cerevisiae, a potential IQGAP termed Iqg1p (135, 446) or Cyk1p (328) was identified in three independent studies. Iqg1p/Cyk1p is a ~165-kDa protein that contains several, but not all, of the structural motifs found in mammalian IQGAPs, including an N-terminal calponin homology (CH) domain predicted to interact with actin, four or eight IQ domains predicted to interact with calmodulin, a coiled-coil domain that may function in the dimerization of IQGAPs, and a C-terminal GAP homology domain (GRD) predicted to interact with Cdc42p. In two-hybrid protein assays, Iqg1p preferentially interacted with activated (GTP-bound) Cdc42^G12Vp, suggesting that it is a downstream effector (446). Iqg1p/Cyk1p coimmunoprecipitated with actin from S. cerevisiae cell lysates (446) and cosedimented with polymerized rabbit skeletal muscle actin; this cosedimentation was dependent on the Iqg1p/Cyk1p CH domain (135). While Iqg1p/Cyk1p has not been shown to interact with calmodulin as do mammalian IQGAPs (see below), calmodulin is delocalized in a iqg1 strain (446).

Mammalian IQGAPs mediate Cdc42p-calmodulin-actin interactions. There are two identified mammalian IQGAPs, designated IQGAP1 and IQGAP2. Human IQGAP1 was originally identified serendipitously in a PCR-based search for matrix metalloproteinase family members (588). IQGAP1 was also identified from COS cell lysates by its ability to interact with GTPS-GST-Cdc42 on agarose beads (204). This ~195-kDa protein contains a GRD (residues 997 to 1270) with significant similarity to the catalytic domain of Ras-GAP proteins, as well as a CH domain (residues 48 to 161), which is believed to interact with actin, and four tandemly repeated IQ domain (residues 745 to 865), which are present in a number of proteins that interact with calmodulin. IQGAP1 also contains a single WW domain implicated in protein-protein interactions and six copies of a unique 50- to 60-amino-acid domain with no known matches in the database. The human IQGAP1 gene mapped to chromosome 15p-15q1.1, and RNA blot analysis indicated that IQGAP1 was highly expressed in the placenta, lungs, kidneys, and skeletal muscle but was absent from the brain. A GST-IQGAP1-GRD fusion protein or an in vitro-translated C-terminal IQGAP1 domain polypeptide did not exhibit GAP activity against Ras, Rho, or Cdc42Hs (204, 588), and affinity-purified IQGAP1 did not display GAP activity against Ras, RalA, Cdc42Hs, Rac1, or RhoA (reference 283 and data not shown), suggesting that IQGAPs do not have catalytic GAP activity. Interestingly, expression of the IQGAP1 C-terminal domain peptide in S. cerevisiae resulted in a dominant negative loss-of-polarity phenotype that could be suppressed by overexpression of wild-type S. cerevisiae Cdc42p (204), suggesting that this C-terminal domain could interact with Cdc42p in vivo (see below).

View larger version (24K): [in this window] [in a new window]   FIG. 6.   Molecular model for the generation of an IQGAP-Cdc42p-actin ternary complex. See the text for details.

S. cerevisiae Bee1p/Las17p. Another family of potential scaffold proteins linking Cdc42p and the actin cytoskeleton are the WASPs, with the prototypical WASP being encoded by the gene that is defective in Wiskott-Aldrich syndrome patients (see below). Bee1p (321) was identified from the S. cerevisiae genome database by its similarity to the mammalian WASP. Bee1p contains a WASP homology domain (WH1) that is similar to PH domains, as well as a proline-rich domain that binds SH3 domains; it does not seem to contain a CRIB domain for binding Cdc42p as does the mammalian WASP. Deletion of Bee1p resulted in a slow-growth phenotype at room temperature and no growth at high temperatures (321). Morphological characterization of bee1 cells indicated defects in bud growth, cytokinesis, and actin organization. In addition, Bee1p localized to cortical actin patches and bound to actin and the actin binding protein Sla1p in immunoprecipitation experiments (321) and to the actin binding protein verprolin in a two-hybrid experiment (413). The bee1 cells were also defective in the ability to assemble cortical actin to the buds in an in vitro permeabilized cell system. While Bee1p clearly plays a role in actin cytoskeleton organization, it is unclear if Bee1p interacts with Cdc42p as does its human counterpart (Fig. 3 and 4A).

Mammalian WASPs. Patients with Wiskott-Aldrich syndrome have multiple immunological defects, including thrombocytopenia with small platelets, eczema, T- and B-lymphocyte defects, and an increased risk of malignancies and autoimmune diseases (for reviews, see references 144, 261, and 427). The severity of these defects has been directly correlated with mutations within the X-linked recessive gene WASP and with defects in cellular actin cytoarchitecture (122, 257, 270, 285, 399, 428, 475, 577, 636, 637). The ~62-kDa WASP (19, 271, 545) contains several functional domains including an N-terminal WH1 domain (residues 8 to 105), a CRIB domain (residues 238 to 257), a proline-rich domain (residues 312 to 404), two potential actin binding sites with similarity to verprolin and cofilin sequences (residues 430 to 446 and 469 to 489), and an acidic C-terminal region. A human WASP-GST fusion protein containing residues 48 to 321 bound to GTP-Cdc42Hs, but not GTP-bound RhoA or Rac1, in GST affinity chromatography experiments, and this binding depended on the presence of the WASP CRIB domain (545) and the Cdc42 effector domain (291), suggesting that WASP is a Cdc42p-specific effector. Ectopic expression of FLAG-tagged WASP in rat kidney epithelial cells indicated that WASP was present predominantly in cytosolic clusters and colocalized with actin structures excluding stress fibers. Formation of these clusters was inhibited by addition of cytochalasin D and by coexpression of the dominant negative Cdc42^T17N mutant protein but not the dominant negative Rho or Rac mutant proteins (545), reinforcing the specific interactions seen between WASPs and Cdc42p. It was reported that the Cdc42 Y40C effector domain mutation abolished interactions with WASP and other CRIB domain-containing proteins but did not affect the generation of actin-dependent morphological structures, suggesting that other downstream effectors mediated Cdc42p-actin interactions (291). However, examination of the data indicates that the Y40C mutation, while interfering with p65^PAK kinase activity in immunoprecipitates, reduced but did not abolish the binding of GST-WASP (containing amino acids 201 to 321 of WASP) to [-³²P]GTP-loaded Cdc42^{Q61L, Y40C} mutant protein in a nitrocellulose overlay assay (291, 388), leaving open the question of the physiological role of WASP-Cdc42 interactions.

Gic1p and Gic2p. The S. cerevisiae Gic1p and Gic2p downstream effectors were identified in two independent studies (54, 83). Gic1p was identified by its ability to suppress the bem2-101^ts growth and bud site selection phenotypes (83), and Gic1p and Gic2p were identified through a search of the S. cerevisiae genome database for proteins that contained CRIB domains (54). Gic1p and Gic2p were 39% identical and 54% similar in predicted amino acid sequence and were not homologous to other proteins in the database. Deletion of GIC1 or GIC2 did not lead to abnormal phenotypes, but a gic1 gic2 double mutant did not grow at high temperatures and displayed numerous morphological abnormalities at semipermissive temperatures, including the presence of a large percentage of large, unbudded, multinucleate cells, delocalized chitin deposition, aberrant actin organization, and abnormal mitotic spindles, suggesting a role for Gic1p/Gic2p in cellular morphogenesis, as well as defects in mating-projection formation and reduced mating efficiencies, suggesting a role in the mating pathway.

gic1 gic2 ts

gic1 gic2 ts

gic1 gic2

gic1 gic2 cla4

gic1 gic2

gic1 gic2

Zds1p and Zds2p. S. cerevisiae Zds1p and Zds2p were identified in multiple genetic screens (see below) including a screen for negative regulators of Cdc42p function (42). The 915-amino-acid Zds1p and the 942-amino-acid Zds2p exhibited ~42% similarity and contained potential coiled-coil domains but did not contain recognizable CRIB domains. Overexpression of Zds1p reduced the restrictive temperatures of the cdc24-10 and cdc42-1 mutants, led to a large proportion of enlarged unbudded cells when expressed in the cdc24-12 mutant, and altered chitin and actin localization in haploid cells (42), reinforcing its potential role as a negative regulator of the cell polarity pathway. Deletion of Zds1p or Zds2p singly had no phenotypic effect (however, see below), but the zds1 zds2 double mutant showed a reduced growth rate and a high percentage of elongated budded cells with a single nucleus and 2C DNA content, indicating a defect in the apical-isotropic switch and/or a G₂/M cell cycle delay, suggesting that Zds1p and Zds2p together are needed for cell cycle progression (42, 618). Subcellular localization of a GST-Zds1p fusion protein by using anti-GST antibodies suggested that Zds1p localized to the sites of incipient bud emergence as well as to tips of enlarging buds and occasionally to the mother-bud neck region, suggesting a colocalization with Cdc42p. However, it should be noted that a direct interaction between Zds1p or Zds2p and Cdc42p has not been reported.

Bem4p/Rom7p. S. cerevisiae Bem4p was identified in three separate screens as a multicopy suppressor of the cdc42-1 mutant simultaneously overexpressing the SRO4 gene, as a mutant that required multiple copies of CDC42 to grow, and as a synthetic-lethal mutant with the cdc24-4 mutation (343). The allelic Rom7p was identified as a multicopy suppressor of a dominant negative rho1 mutant (221). In two-hybrid protein studies, Bem4p was shown to interact with Cdc42p, Rho1p, Rho2p, and Rho4p, and it interacted equally well with the Cdc42^G12V, Cdc42^Q61L, and Cdc42^D118A constructs, suggesting that it can interact with both GTP-bound and GDP-bound Cdc42p. However, in vivo interactions between these proteins have not been reported. Deletion of BEM4 led to cell inviability at 37°C with a cellular morphology of large, round, unbudded, multinucleate cells containing delocalized actin, and this ts phenotype could be suppressed by simultaneous overexpression of both Cdc42p and Rho1p. It should be noted, however, that these phenotypes varied in different strain backgrounds. Due to its ability to interact with multiple Rho-type GTPases in S. cerevisiae, the physiological role for Bem4p is elusive, but an interesting observation was that deletion of the Rho-GDI, Rdi1p, led to cell inviability in the presence of a bem4 mutation (unpublished results cited in reference 343), suggesting that Bem4p and Rdi1p, while showing no sequence similarity to each other, may have overlapping functions in vivo.

70-kDa S6 kinase. The mammalian 70-kDa S6 kinase (pp70^S6k) is involved in growth control, translation initiation, and progression through the G₁/S phase of the cell cycle through its phosphorylation of ribosomal protein S6 and the subsequent translation of 5' terminal oligopyrimidine tract-containing mRNA (for reviews, see references 92, 238, and 557). The pp70^S6k kinase activity is activated by a number of different regulatory signals including growth factors, phorbol esters, and cytokines, and this activation is inhibited by the immunosuppressant drug rapamycin, which functions through its binding to the FK506 binding protein rapamycin-associated protein FRAP (also known as RAFT and TOR), a PI 3-kinase-like protein kinase (52, 53, 502, 503). The pp70^S6k kinase activity is also activated by PI 3-kinase (82, 97, 401), and this activation is blocked by the PI 3-kinase inhibitors wortmannin and LY294002. Immunoprecipitated HA-tagged pp70^S6k from NIH 3T3 and COS cells had in vitro kinase activity against ribosomal protein S6 as a substrate, and this activity was enhanced by cotransfection with activated GST-Cdc42^G12V and GST-Rac1^G12V, but not GST-RhoA^G12V, and by the Cdc42-GEF Dbl and was inhibited by dominant negative Cdc42 and Rac mutants (93), suggesting that activation of Cdc42p leads to activation of pp70^S6k kinase activity. Cotransfection with activated GST-Cdc42^G12V and GST-Rac1^G12V constructs also increased the phosphorylation of pp70^S6k necessary for its kinase activity, and this activation was lost in the Cdc42^T35A effector domain mutant and in the Cdc42^C189S prenylation mutant, suggesting that proper subcellular localization is necessary for the interaction between Cdc42p and pp70^S6k. The growth factor-induced activation of pp70^S6k activity seemed to be independent of the activation of the JNK and p38 kinase activities, but the Cdc42p-induced pp70^S6k activation was blocked by the addition of rapamycin and wortmannin, suggesting that FRAP and PI 3-kinase function in the activation pathway.

So what are the future research directions for deciphering Cdc42p function? The answer to this question will be determined partly by the organism in which experiments are performed. For instance, genetic and biochemical studies in S. cerevisiae and in cultured mammalian cells, and to a lesser extent in S. pombe, have identified a myriad of Cdc42p regulators and effectors, but only recently have experiments designed to test specific protein-protein interactions and multiprotein complex formation been performed. In addition, little is known about the in vivo specificity of assorted GEFs, GAPs, GDIs, or downstream effectors or about the targeting mechanisms for Cdc42p to the plasma membrane at sites of polarized growth in response to different signals or at different times in the cell cycle. Therefore, future experiments with these organisms will probably focus on these issues. Few Cdc42p effectors and regulators have been identified or characterized in Drosophila and C. elegans, and so these proteins must be isolated before detailed mechanistic questions can be addressed. However, the mechanistic studies in yeast and mammalian cells should develop useful paradigms that will allow for more defined questions to be addressed in Drosophila and C. elegans. Interestingly, no bona fide Cdc42p homologs have been identified in fungal systems outside of the unicellular yeast or in plant systems, although multiple Rac homologs have been identified (118, 319, 563, 598). Given the high degree of cellular polarization seen in fungal and plant cell growth patterns, it would be surprising if Cdc42p homologs did not exist and were not involved in these processes. Finally, given the recent determination of the NMR and X-ray crystal structures of Cdc42p and Cdc42p complexed with one of its GAPs, future molecular modeling studies could provide valuable insight into the effects of various loss-of-function, gain-of-function, and effector domain mutations on Cdc42p structure and function and its interactions with downstream effectors and regulators. The much anticipated NMR and/or X-ray crystal structure determinations of Cdc42p complexed with a downstream effector or GEF or GDI should greatly enhance our knowledge of the mechanisms of action of these proteins. All in all, the explosion of research centered on Cdc42p over the past 5 to 10 years has only served to whet our appetite for more details, which will certainly be forthcoming in the very near future.