β1PIX, the PAK-interacting exchange factor, requires localization via a coiled-coil region to promote microvillus-like structures and membrane ruffles

The small GTPases of the Rho family play key roles in transducing extracellular stimuli into distinct responses including cell motility, adhesion, cell division and phagocytosis. The GTPases cycle between GTP-bound and GDP-bound forms and their activation requires the action of guanine nucleotide exchange factors (GEFs) to promote the conversion of the GDP to the GTP state. Individual members of the Rho GTPases are known to cause specific changes to the actin cytoskeleton of the cells. Active RhoA causes actin stress fibre formation, whereas dominant active Cdc42 and Rac1 induce the formation of filopodia and lamellipodia, respectively. The rearrangement of the cytoskeletal structures is pivotal to the outcome of the signal transduction events downstream of the Rho GTPases (Lim et al., 1996; Van Aelst and D’Souza-Schorey, 1997).

Some of the downstream effectors of the Rho GTPases and the pathways they regulate are well studied. In particular, the formation of stress fibres, which is downstream of RhoA-GTP, requires the function of Rho-kinase/ROK (Amano et al., 1997; Leung et al., 1995). ROK can phosphorylate and inactivate the myosin-binding subunit of the myosin light chain (MLC) phosphatase (Kimura et al., 1996). This results in an increase in phosphorylated MLC, which has enhanced actin binding and bundling activity, and hence an increase in stress fibre formation. The effector proteins downstream of Rac1 in lamellipodia formation are not as well characterized although POR1 might be involved in this process (Van Aelst et al., 1996). N-Wasp mediates the link between Cdc42 and the Arp2/3 proteins in actin polymerization, which might participate in the formation of filopodia (Miki et al., 1998; Rohatgi et al., 1999). The ROK-related target MRCK is directly involved in the formation of focal complexes (FCs) and filopodia, as demonstrated by observations that a kinase inactive mutant can block these processes downstream of Cdc42 (Leung et al., 1998).

A less well characterized morphological effect of Rho GTPases is microvillus formation. These apical membrane protrusions, found on polarized epithelial cells, fibroblasts and lymphocytes, are important for the cells to sense extracellular signals. At least one member of the ERM (ezrin, moesin and radixin) family of proteins is required to drive microvillus formation by specifically cross-linking actin filaments to the plasma membrane. The activation of ERM proteins is linked to phosphorylation, phosphoinositide binding and RhoA signalling pathways (Bretscher et al., 1997; Hirao et al., 1996; Matsui et al., 1999; Oshiro et al., 1998; Shaw et al., 1998; Tsukita and Yonemura, 1999). It has also been reported that, by activating Cdc42, RhoG causes the formation of microvilli (Gauthier-Rouviere et al., 1998).

PAK is an effector kinase of Cdc42 and Rac1 (Manser et al., 1994) in promoting the breakdown of Rho-dependent actin stress fibres and focal adhesion complexes (Manser et al., 1997; Sells et al., 1997). We have isolated a PAK-interacting exchange factor (PIX) that exhibits exchange activity towards both Cdc42 and Rac1 in vitro (Manser et al., 1998). The identification of PIX demonstrates a GTPase activator being directly coupled to an effector, thereby providing specificity to the signalling pathway. In the case of T-cell receptor activation the PIX-PAK interaction is indeed required for GTPase-mediated kinase activation (Ku et al., 2001). The complex might also provide the link for cross-talk between Cdc42 and Rac1 pathways because elevated levels of βPAK (which can be recruited by Cdc42) drive a Rac phenotype in PC12 cells (Obermeier et al., 1998; Sells et al., 1999). The PAK-PIX interaction, mediated by the SH3 domain of PIX, plays a key role in these two cell systems.

Other identified domains of PIX include the calponin homology (CH), Dbl homology (DH), pleckstrin homology (PH) and a GIT1 binding domain. Here, we demonstrate that a discrete coiled-coil C-terminal domain appears to regulate PIX function via intermolecular interactions. Although the universal pairing of DH and PH domains in Dbl-family RhoGEFs suggests that the PH domain modulates the activities of the DH domain, the solution structure of β1PIX DH and PH domains does not reveal such a functional coupling (Aghazadeh et al., 1998).

The intracellular localization of RhoGEFs is often achieved by specific domains which associate with other proteins or phospholipids at the cell membrane. The PH domain of Dbl mediates the oncogenic activities of the protein by targeting it to specific cytoskeletal components (Zheng et al., 1996). The N-terminal PH domain of Tiam1 localizes it to the plasma membrane allowing Rac-mediated membrane ruffling and JNK activation (Michiels et al., 1997; Stam et al., 1997). The PH domain of the Ras/Rac GEF Sos is involved in membrane targeting and is preferentially localized to the leading edge of the motile cells (Chen et al., 1997). By contrast, other Dbl-family proteins Lfc and GEF-H1 are localized to the microtubule network (Glaven et al., 1999; Ren et al., 1998). The proper presentation of the exchange factors to their respective GTPases is thus critical to their biological activities.

In this paper, we report a βPIX splice variant that is enriched in the brain. All PIX isoforms are substrates of PAK but their exchange activities are not affected by the phosphorylation. β1PIX, but not β2PIX, translocates to the cell periphery, where it drives formation of ruffles and microvillus-like structures. The coiled-coil region in β1PIX appears to be responsible for its localization to the cell periphery and for mediating its cellular activities.

β2PIX was isolated by screening a rat brain cDNA library in λZapII from Stratagene, using [α³²P]dCTP-labelled cDNA fragment encompassing nucleotides 483-671 of αPIX. Various β2 cDNA were assessed by restriction enzyme mapping, and a full length cDNA generated in the pXJ vectors by splicing the 3′ region of the β2 cDNA to the β1PIX sequence at the internal Kpn1 site.

Full length αPIX was cloned into pAS2-1 vector (Clontech) as the bait in a two-hybrid screen for PIX-interacting proteins. The N-terminal (1-360 bp) and C-terminal (2014-2331[stop-codon] bp) of αPIX were amplified by PCR with restriction sites incorporated for the convenience of cloning. Most of the cDNA fragment was cloned by insertion of the original cDNA from the internal BamHI (174 bp) to NcoI (2025 bp) sites. The cDNA library used was a human brain cDNA Matchmaker library in pAct2 from Clontech. The bait recombinant plasmid was first transformed into a reporter yeast strain (Y190), which contained the HIS3 and lacZ genes under the control of a Gal4-responsive element. The matchmaker cDNA library was then transformed into Y190 containing the bait plasmid. The His3 gene allowed a positive growth selection for clones that were then screened using the blue/white β-galactosidase (β-Gal) filter assay to confirm the protein interactions. An estimated 1×10⁶ transformants were screened. After putative positive yeast clones were identified, recombinant plasmids in pAct2 were extracted and retransformed into Y190 containing the bait plasmid. The β-Gal filter assay was repeated and positive clones were sequenced.

Polyclonal anti-PIX SH3 antibodies were raised by injection of glutathione-S-transferase (GST) fusion protein of αPIX (amino acids 155-545) into rabbit. The rabbit antisera obtained were tested for specificity by western blot analysis using protein lysate of rat tissues and cell lysates containing transfected Flag-tagged PIX isoforms. It was found that the antisera was able to recognize specific bands (used at 1:500 dilution). Only a single band was observed with the Flag-PIX transfected cell lysates. The antibodies were affinity purified using MBP-PIX-SH3 column. A similar method was used to purify antibodies against the serine-rich C terminus of β2PIX. The peptide used was amino acids 556-625 of β2PIX.

COS-7 cells were lysed in hypotonic buffer (50 mM HEPES (pH 7.3), 1 mM MgCl₂) without any detergent. NaCl and PMSF were added to 0.3 M and 1 mM, respectively, to the cell lysates. The cell lysates were spun at 100,000 g to separate the S100 soluble fraction and P100 pellets. The P100 pellets were then extracted with buffer containing TritonX-100 (50 mM HEPES (pH 7.3), 1 mM MgCl₂, 0.3 M NaCl, 1% Triton X-100 and 1 mM PMSF) to extract the membrane embedded proteins. After centrifugation at 100,000 g, the pellets were extracted with buffer containing 1% SDS to extract the cytoskeletal proteins. Extracts from different fraction were resolved on SDS-PAGE gels and analysed by western hybridization.

The guanine nucleotide exchange activity was measured using Rac1 assay as previously described (Manser et al., 1998). The GEF activity was determined by the incorporation of [³⁵S]GTPγS (NEN) into Escherichia coli expressed and purified GST-Rac1. The PIX proteins used in these assays were purified from transiently transfected COS-7 cells using anti-Flag M2 beads (Sigma) and quantified by Coomassie staining of duplicates. Bacterially expressed GST-PAK was used to phosphorylate a set of the immunoprecipitated PIX (∼2 μg) prior to the exchange assays. 8 μg of GST-PAK protein was used per reaction, which also contained 500 μM ATP in the kinase buffer (50 mM Hepes pH 7.3, 10 mM MgCl₂, 2 mM MnCl₂). The reaction was incubated at 30°C for 30 minutes. The reactions were stopped by addition of EDTA. The Flag-beads were washed twice with PBS buffer before proceeding with the exchange assay. In each exchange assay about 2 μg of Flag-PIX fusion protein and 5 μg of GST-Rac1 were used. The reaction mixture was incubated at 30°C. Bound [³⁵S]GTPγS was assessed by absorption of protein onto nitrocellulose membranes and liquid scintillation counting. In each experiment, two aliquots were taken at each time. Each experiment was repeated.

Immunoprecipitated Flag-PAK or GST-PAK was washed with kinase buffer and then incubated in 30 μl of the same buffer containing 500 μM ATP and 10 μCi [γ-³³P]ATP (Amersham) and 10 μg of myelin basic protein as substrate. The reaction mixtures were incubated at 30°C for 30 minutes and the reactions were stopped by the addition of the SDS sample buffer.

Relevant cDNAs were cloned into the pXJ-Flag vector for mammalian cell expression (Manser et al., 1997). pXJ-Flag-αPIX was cloned similarly as described above. pXJ-Flag-β2PIX was cloned replacing the C terminus of β1PIX by a PCR fragment at the internal KpnI (962 bp) site. The N-terminal deletion mutant, pXJ-Flag-ΔN80β1PIX, was generated by splicing a PCR fragment covering 241 bp to the KpnI site (962 bp) of β1PIX to vector containing DNA fragment C-terminal to the KpnI site. The C-terminal truncation mutant pXJ-Flag-β1PIX^1-555 was generated by splicing a PCR fragment containing the truncated C terminus to the N terminus of β1PIX at the internal KpnI site. pXJ-GST-β1PIX-C-ter was constructed by cloning a PCR fragment containing nucleotides 1612-1941 of β1PIX. The cDNA constructs were transfected into COS-7 or HeLa cells using DOSPER (Boehringer Mannheim) or SuperFect (Qiagen) using the supplier’s protocol. In most cases, 8 μg of plasmid DNA was used per 100 mm dish of 80% confluent cells and 1.5 μg of plasmid DNA was used for 20×20 mm two-well chamber slide (Nalge Nunc International). Cell staining was done as previously described (Manser et al., 1997).

HeLa cells were seeded onto 20×20 mm glass cover slips and injected using an Eppendorf microinjector (no number given on the apparatus) and a Zeiss axiovert microscope. Plasmid (50 ng ml^–1) encoding green fluorescent protein (GFP) was injected with the cDNA of interest (50 ng ml^–1). The injected cells were returned to the incubator for 2-4 hours prior to fixation in 3% paraformaldehyde as described previously (Manser et al., 1997).

Relevant cDNAs were cloned in pXJ-Flag or pXJ-GST mammalian expression vectors (Manser et al., 1997). The cells were harvested in protein lysate buffer (50 mM Hepes (pH7.5), 0.3 M NaCl, 1 mM MgCl₂, 1 mM EGTA, 10 mM β-glycerophosphate, 10 mM NaF, 1 mM sodium vanadate, 5% glycerol, 5 mM DTT, 0.5% Triton X-100). The cell lysates were passed through a 30G syringe (×3) and were cleared by centrifugation at 10,000 g for 5 minutes. The Flag-tag proteins were isolated using anti-Flag Mab M2-beads (Sigma) and GST fusions with glutathione-Sepharose beads (Pharmacia). Protein complexes were dissociated from the beads by heating to 100°C in 1× SDS buffer for 3 minutes.

HeLa cells were first plated on glass coverslips and microinjected with the DNA constructs of interest in pXJ-Flag or pXJ-HA expression vectors (Manser et al., 1997) together with pXJ-GFP plasmid DNA for identifying the injected cells. Injected cells were incubated for 2-4 hours to allow protein expression. Cells that failed to express the GFP marker were removed and remaining cells expressing the protein of interest were fixed with 1% glutaraldehyde. Samples were then gradually dehydrated using increasing ethanol, followed by critical point drying and gold sputtering. The samples were analysed with a Phillips XL30-FEG scanning electron microscope. About 30-50 cells were examined for each experiment.

PIX are GEFs that contain an N-terminal SH3 domain that specifically binds PAK, and are encoded by the αPIX and βPIX (p85^SPR, p85^COOL-1) genes (Bagrodia et al., 1998; Manser et al., 1998; Oh et al., 1997). It appears that the p78 βPIX species is ubiquitous. cDNA coding for a shorter variant of βPIX (p50^COOL-1) and two other cDNA species (β1PIX-b and β1PIX-c) have been reported, although the corresponding proteins have yet to be identified (Bagrodia et al., 1998; Kim et al., 2000). We have now isolated another cDNA splice variant encoding a protein (β2PIX) in which a shorter serine-rich domain replaces the C-terminal coiled-coil sequences (residues 556-646) that are present in the original cDNA product that we now designate β1PIX. β2PIX is 625 amino acids long (Fig. 1A) with a predicted molecular mass of 68 kDa. A sequence comparison of the different C-terminal domains of αPIX, β1PIX and β2PIX is shown in Fig. 1B. The predicted coiled-coil regions of αPIX and β1PIX are well conserved, and are most similar to sequences found in the Myo2 protein. An insert (I) of 31 amino acid residues (417-448) in β1PIX and β2PIX following the PH domain is not present in αPIX.

To characterize β2PIX, antibodies were raised against the serine-rich C-terminal domain. Western analysis revealed that the p70 β2PIX was most abundant in brain (from which the cDNA was isolated) and was also present in testis and thymus (Fig. 1C). Interestingly, the antibody detected a larger (p80) species that was recognized by antibodies raised against the SH3 domain (which is common to all PIX). This might represent a protein species with additional insert sequences flanking the PH domain, as reported by Kim et al. (Kim et al., 2000) or with an N-terminal CH domain that is also present in βPIX (Z.-S.Z., unpublished). Thus, p70 β2PIX shows a more restricted expression pattern than the β1PIX. Western analysis has found β1PIX to be the predominant isoform present in COS-7, HeLa and NIH 3T3 cell lines (Fig. 1C, middle).

DNA constructs of Flag-β1PIX and Flag-β2PIX were transfected into COS-7 cells to determine the distribution of these two isoforms in the cell. From the analysis of fractionated cell extracts, most Flag-β1PIX was present in the Triton-X-100-soluble fraction (Fig. 2A). Some Flag-β1PIX was also found in the water-soluble extract and in the SDS-soluble fraction. However, most of the Flag-β2PIX was found in the water-soluble fraction (Fig. 2A). Very little if any was found in the Triton-X-100-soluble fraction or in the SDS fraction.

Endogenous PIX and PAK proteins were analysed in fractionated COS-7 cell extracts. Western blotting with anti-PIX (SH3) or anti-α/βPAK (PAK1 and PAK3) antibodies revealed that ∼70% of PIX (probably p78 β1PIX) and essentially all PAK were recovered in the cytosolic fraction (Fig. 2A). The remaining 30% of PIX appeared mostly in the detergent-soluble (membrane fraction), with some found in the detergent-insoluble (often defined as the cytoskeletal) fraction.

To determine the intracellular distribution of the various PIX isoforms and the potential role of their various domains, cDNAs encoding different Flag-tagged PIX proteins were transiently transfected into HeLa cells (Fig. 2B). Both αPIX and β1PIX were distributed in the cytoplasm but prominent at the cell periphery, whereas β2PIX was found primarily in cytoplasm and nucleus. When the coiled-coil domain of β1PIX was deleted, the β1PIX^1-555 protein showed cytoplasmic and nuclear localization similar to that of β2PIX. By contrast, removal of N-terminal SH3 domain (ΔΝ80β1PIX) did not affect the peripheral membrane localization of β1PIX. By itself, the coiled-coil C-terminus of β1PIX could localize GST to the cell periphery. These results indicated that the coiled-coil domain is important for the peripheral localization of αPIX and β1PIX, and might represent the key targeting sequence for certain αPIX and β1PIX proteins.

β1PIX but not β2PIX can form homodimers via the coiled-coil domain

In a two-hybrid screen for PIX partners, full-length αPIX was found to interact with a construct containing a C-terminal portion of αPIX (residues 662-776). This C-terminal region of αPIX also interacted with β1PIX, indicating that PIX homo- and heterodimers can be formed (data not shown). Because β1PIX coiled-coil domain could dimerize but did not interact with β1PIX^1-555 (Fig. 3A), we conclude that β1PIX does not interact in a head-to-tail manner. When GST-β1PIX^538-646 (β1PIX-C-ter) was used to pull down various Flag-β1PIX constructs, only those constructs containing the complementary C terminus of β1PIX were precipitated (Fig. 3A). Full-length Flag-β1PIX also brought down GST-β1PIX. By contrast, no co-precipitation was observed between GST-β1PIX and Flag-β2PIX or between GST-β2PIX and Flag-β2PIX (Fig. 3B). Thus, the coiled-coil region is implicated in dimerization. β2PIX appears to be monomeric because it behaves differently from the other two isoforms.

PIX and PAK form multimeric complexes

Because PIX can form dimers and also binds tightly to PAK via its SH3 domain, we investigated whether PIX and PAK can exist as multimeric complexes in the cell. GST-PAK, HA-β1PIX and Flag-ΔΝ80β1PIX were transfected together into COS-7 cells. Flag-ΔN80β1PIX could be detected in the GST-PAK complex. The results indicated that PAK binds to PIX in dimeric form (Fig. 3C, lane 2). GST-PAK itself does not precipitate with ΔN80β1PIX because the SH3 domain that binds PAK is missing in this mutant (Fig. 3C, lane 1). When DNA constructs of GST-PAK, HA-PAK and Flag-β1PIX^1-459 (lacking the dimerization domain) were transfected together into COS-7 cells, Flag-β1PIX^1-459 was found in the complex but HA-PAK could not be detected in the GST-PAK complex (Fig. 3C, lane 3). Neither could HA-PAK be recovered from the GST-PAK complex (ΔN80β1PIX was included to drive non-productive dimers with endogenous β1PIX) (Fig. 3C, lane 4). However, GST-PAK could complex to HA-PAK when β1PIX (wild type) was present. The results suggest that a GST-PAK-(Flag-PIX)₂-HA-PAK tetramer could be formed (Fig. 3C, lane 5). Hence, PAK and PIX proteins associate as multimeric complexes.

PIX was first identified as a protein that both binds to and is phosphorylated by PAK (Manser et al., 1998). We immunoprecipitated various Flag-tagged isoforms and mutants of PIX from transfected COS-7 cells and subjected them to in vitro phosphorylation by GST-PAK. We found that a major PAK phosphorylation site(s) was located between residues 459 and 555 of β1PIX (Fig. 4A, lanes 3,4). A deletion mutant termed β1PIXΔpro (deletion of 460-495) was still phosphorylated by PAK (Fig. 4A, lane 6), suggesting that a prominent phosphorylation site(s) resides in β1PIX 496-555. We have mapped the major phosphorylation sites to S525 and T526 of β1PIX (data not shown). This region is conserved in αPIX and therefore phosphorylation might regulate a common activity among PIX proteins. One testable function is regulation of Rac1 or Cdc42 GEF activity.

Given the differences in the in vivo behaviour of the two βPIX splice variants (see next section), we assayed the guanine nucleotide exchange activities of the three PIX isoforms towards Rac1. Immunoprecipitated Flag-αPIX, β1PIX, and β2PIX were quantified by Coomassie staining (not shown) and also analysed for their GEF activities after in vitro phosphorylation by PAK. The Rac1-GTP exchange assay (Fig. 4B) indicated that the exchange activities of all three isoforms are similar in vitro and that PAK phosphorylation of αPIX, β1PIX or β2PIX neither enhances nor inhibits their GEF activities.

Although a truncated form of βPIX suppresses PAK activation (Bagrodia et al., 1998) a truncated αPIX was reported to enhance PAK activity (Daniels et al., 1999). We therefore compared the effects of βPIX isoforms on PAK activity in vivo. PIX constructs were expressed with GST-αPAK (Pak1) and Cdc42 in COS-7 cells. Wild-type Cdc42 stimulates PAK activity to much more limited degree than Cdc42^V12 (Fig. 4C) and allows an assessment of potential activation and suppression. GST-PAK was isolated on glutathione-Sepharose beads and assayed using myelin basic protein (MBP) as substrate (Fig. 4C). Consistent with previous observations, αPAK was inactive in the absence of Cdc42 (∼2% of that in the presence of Cdc42^V12) (Manser et al., 1997) and β1PIX had no activating effect on PAK in these cells (data not shown). However, with Cdc42, β1PIX inhibited αPAK activity. Similar inhibition did not occur with the β1PIX^1-555 mutant lacking the coiled-coil domain, and β2PIX showed an intermediate effect (Fig. 4C). Because inhibition was also observed in a β1PIX mutant (ΔN80) lacking the PAK-binding SH3 domain, it seems these effects are not mediated by the direct binding of PIX to PAK.

We previously demonstrated that αPIX causes morphological changes in HeLa cells consistent with Rac1 activation. However, these are somewhat different from the morphology of Rac1^V12-producing cells (Manser et al., 1998). Microinjection of αPIX and β1PIX plasmid DNA caused similar phase-dark ruffles at the cell periphery (Fig. 5A). This morphological change was not observed in cells microinjected with plasmids encoding β2PIX, β1PIX^1-555, ΔN80β1PIX, β1PIX-DHm (an exchange-activity-deficient mutant) or the β1PIX C-terminal domain. Thus, the ability of β1PIX to generate ruffles was dependent on the integrity of domains involved with PAK binding as well as dimerization. These observations are consistent with the dependence of lamellipodia formation on PAK-PIX interactions in PC12 cells (Obermeier et al., 1998). The formation of these ruffles was blocked by co-injection with dominant negative Rac1^N17 but not with dominant negative Cdc42^N17 (Fig. 5B). Rac1^N17 by itself did not result in any obvious change in the cell morphology (data not shown). It has been shown that PIX has exchange activity towards Cdc42 in vitro (Manser et al., 1998) and in vivo (Yoshii et al., 1999). However, β1PIX neither induces filopodium-like peripheral structures nor drives the cell rounding that is characteristic of other Cdc42 GEFs, such as hPem2 (Reid et al., 1999) (Fig. 5C). Co-injection of β1PIX with wild-type Cdc42 gave a phenotype not seen with Cdc42 alone, and more similar to cells overexpressing Cdc42^V12 (Fig. 5C). Co-injection of β1PIX with wild-type Rac1 elicited ruffle-like structures and enhanced cell spreading, a phenotype associated with overexpression of Rac1^V12 but not wild-type Rac1 in these cells (not shown). Similarly, co-injection of β2PIX with Cdc42 (Fig. 5C) produced an activated Cdc42 phenotype. A Rac1^V12 phenotype was observed when β2PIX was co-injected with wild-type Rac1 (data not shown). Thus, overexpression of β1PIX or β2PIX elicits phenotypes associated with activated Rac1 and Cdc42 only when levels of the wild-type GTPase are increased, unlike with Tiam1 or hPEM2. This activity was apparent even though a proportion of β2PIX was targeted to the nucleus.

Because we were unable to observe morphological changes other than the phase-dark ruffles by light microscopy, scanning electron microscopy (SEM) was used to investigate changes at the cell surface. HeLa cells were microinjected with β1PIX cDNA and fixed after 2 hours for imaging by SEM. The β1PIX expressing cells exhibited numerous membrane protrusions on their surface compared with uninjected cells, although the peripheral ruffles seen by light microscopy were not particularly evident by this technique (Fig. 6A). These surface protrusions resembled microvilli and were not induced by β2PIX. By contrast, Cdc42^V12 produced apical structures that were significantly longer than the PIX-induced microvillus-like structures, which explains why these structures cannot be seen by light microscopy. The β1PIX induced structures were blocked by co-injection with Cdc42^N17 but not with Rac1^N17 (Fig. 6A, bottom) implying that the formation of microvillus-like structures was indeed downstream of Cdc42. Cells injected with vector DNA alone exhibited a smooth plasma membrane surface (Fig. 6A).

Surface structures were either absent or sparse in cell expressing β1PIX mutants (β1PIX-DHm, β1PIX^1-555 or ΔN80β1PIX) (Fig. 6B). Constitutively active Rac1^V12 caused cell spreading but no filopodium- or microvillus-like structures were observed (not shown). At higher magnification, we observed that the β1PIX-induced microvillus-like structures arose from ruffles (Fig. 6C, arrowheads), indicating that these might be hybrid structures. These observations suggesting that PIX can activate both Cdc42 and Rac1 in vivo are in agreement with previous reports (Daniels et al., 1999; Manser et al., 1998; Yoshii et al., 1999) and we now show these activities to be dependent upon the coiled-coil domain at the C-terminus.

In this paper, we report an alternate spliced isoform of βPIX with distinct properties and tissue distribution. The cellular distributions of β1PIX and β2PIX differ in that most β1PIX is membrane bound but β2PIX is mainly cytosolic (Fig. 2A). A small proportion of PIX might bind either to lipid rafts or the cortical cytoskeleton because it cannot be extracted by nonionic detergent. The predicted coiled-coil C-terminal domain of αPIX and β1PIX drives both dimerization and targeting to the membrane. One possibility is that dimerization potentiates a relatively weak membrane-binding activity of the PIX monomer (e.g. through the PH domain). Constructs that lack the coiled-coil C-termini are distributed primarily in the cytoplasm and nucleus. However, because the coiled-coil sequence alone localizes GST protein to the cell periphery (Fig. 2B), it appears that this property is intrinsic to this domain.

αPIX contains an additional CH domain at its N terminus that is not present in βPIX. Although the CH domain is found in many actin-binding proteins and signalling molecules, a single CH domain is not sufficient to bind actin (Gimona and Mital, 1998). It has been proposed that actin binding requires two CH domains (Stradal et al., 1998), and so dimerization might confer actin binding on αPIX.

PIX isoforms containing coiled-coil C-termini might form heterodimers. None of the cell lines we have tested contain multiple PIX isoforms (Fig. 1C), although both αPIX and β1PIX are present in Jurkat cells (Ku et al., 2001). Membrane localization could certainly facilitate dimer formation, as in the case of Ras (Inouye et al., 2000), where dimerization of the GTPase is essential for the activation of Raf-1. Yet another example is the N terminus of amphiphysin II, which contains sequences responsible for both plasma membrane targeting and dimerization (Ramjaun et al., 1999). It has been reported recently that the DH domain of the Dbl oncoprotein forms oligomers and that oligomerization is essential for Dbl-induced transformation (Zhu et al., 2001). It was suggested that oligomerization of Dbl could result in a signalling complex that further augments and co-ordinates the GEF activities of Dbl. Clearly, dimerization of PIX provides the possibility of forming multimeric complexes with PAK, GIT1/p95PKL (Bagrodia et al., 1999; Turner et al., 1999; Zhao et al., 2000a) and associated proteins (Fig. 7). Interestingly, PAK has also recently been reported to form dimers via the Cdc42/Rac1-binding domain (Lei et al., 2000).

The importance of PIX for PAK function is demonstrated in T-cell receptor signalling. Efficient αPAK (PAK1) activation requires its binding sites for Rho GTPases and for PIX. Overexpression of β1PIX that either cannot bind PAK or lacks GEF function prevents PAK1 activation (Ku et al., 2001); it is also suggested that the kinase needs to be localized by GIT1. We have previously shown that PAK requires PIX to localize to FCs (Manser et al., 1998). Significantly, PIX binding to GIT1 links it to the central FC components paxillin and FAK. This protein complex promotes focal adhesion turnover and Rac1-dependent motility (Turner et al., 1999; Zhao et al., 2000a). Although PAK activation and autophosphorylation lead to dissociation from its partner, PIX (Zhao et al., 2000b), the ability of PIX to inhibit kinase activity (Fig. 4C) potentially increases the lifetime of the complex.

Although PAK can bind PIX tightly, their subcellular location confirms our previous observations of a dynamic association between the pair. Thus, essentially all of the αPAK and βPAK are found in the cytosol, whereas ∼30% of PIX is membrane associated and also present in a detergent-insoluble fraction (Fig. 2A). Because active (autophosphorylated) PAK dissociates from PIX (Zhao et al., 2000b), PAK will only transiently complex to PIX at the membrane owing to the presence of kinase activators at this site (Lu et al., 1997). However, introducing a PAK inhibitor into cells allows PAK to be stabilized within FCs (Zhao et al., 2000a) (where it is usually not visible). Thus, we propose a model in which PAK can be recruited to the membrane and FCs through binding to membrane-associated PIX (Fig. 7). Indeed addition of a CAAX box to PAK has a similar effect of driving αPAK to FCs (Manser et al., 1997). This localization is further enhanced by PIX binding to GIT1, which unmasks the cryptic paxillin binding site in the GIT1 C-terminal region (Zhao et al., 2000a). Upon activation, PAK is immediately released back to the cytosol while PIX-GIT1 remains at the membrane (and FCs). This might explain why most of PAK fractionates into the cytosolic fraction, whereas PIX is distributed between the cytosol and the membrane-cytoskeleton fraction.

The interactions that lead to PIX-mediated cell shape changes are complex. Although the in vitro and in vivo GEF activities of PIX are very low compared with other Dbl family members (Manser et al., 1998), PIX can apparently still cause morphological changes but in a more restricted manner. This restriction reflects PIX’s localization to the membrane, where bound phosphatidylinositol-3-kinase reportedly co-operates with αPIX in activating Cdc42/Rac1 (Yoshii et al., 1999). Here, we observe that only PIX isoforms that dimerize cause ruffle and microvillus-like structure formation. Nonetheless, the binding of PAK to PIX is important because the SH3 deletion mutant (ΔN80) is ineffective. This is consistent with previous data showing that PAK-induced lamellipodium formation is dependent upon PIX binding (Obermeier et al., 1998). We suggest that the β2PIX isoform would become active by recruitment through as yet unidentified membrane-associated partners.

The induction of phase dark ruffles by full-length β1PIX was not observed with β2PIX, the exchange deficient mutant β1PIX-DHm or β1PIX lacking either the SH3 domain or the coiled-coil region. These results suggest that the C-terminal sequences target PIX to the cell periphery, where its exchange activity is closely linked to PAK association. These ruffles were unlike those induced by microinjection of a more active Rac1-specific GEF, Tiam1 (Michiels et al., 1995) (data not shown), which gives a similar phenotype to cells injected with Rac1^V12, as reported previously (Manser et al., 1997). This suggests that β1PIX drives a more localized production of Rac1-GTP, although co-injection of β1PIX with Rac1 did result in cell spreading, resembling that generated by expression of Rac1^V12 (data not shown). The phase-dark ruffles are not blocked by Cdc42^N17, indicating that Rac1 activation is direct.

We show here for the first time that PIX also promotes localized changes on the cell membrane resembling microvilli. Microvilli and filopodia do share many similarities and components. Although villin has been recognized as a tissue-specific component, these structures can form in its absence, which explains why microvilli are seen in epithelial cells and fibroblasts that do not express villin. The structures induced by PIX are microvillus-like, based on their position, number and ultrastructure by SEM (i.e. they are not visible by light microscopy). Such short microvillus-like structures have been reported with RhoA and RhoG, although, with the latter, they are thought to occur indirectly via activation of Cdc42 (Gauthier-Rouviere et al., 1998; Shaw et al., 1998). Consistent with this, β1PIX-induced microvillus-like structures are blocked by Cdc42^N17 but not Rac1^N17. That β2PIX and β1PIX derivatives (β1PIX-DHm, β1PIX^1-555, ΔN80) do not induce microvillus-like structures confirms that the localization of (functional) PIX is important in its induction of microvillus-like structures, which is potentiated by an association with PAK.

Recent genetic evidence implicates the large GEF Trio in a pathway that includes DPak and the Drosophila Nck homologue Dock. (Newsome et al., 2000). The N-terminal GEF domain of Trio, TrioGEF1, can activate both RhoG and Rac1 (Bellanger et al., 1998; Blangy et al., 2000; Debant et al., 1996). Whether Trio and RhoG lie upstream of PIX remains to be addressed. We have not ruled out the possibility that PIX activates other Rho GTPases such as TC10 (Neudauer et al., 1998). Apart from the induction of long filopodia, TC10 also caused the formation of microvilli (Vignal et al., 2000). Although TCL is very similar to TC10, it elicits different effects on cell morphology, including long, thin extension at the cell periphery and large dorsal protrusions (Vignal et al., 2000). Hence, it is unlikely that PIX preferentially activates TCL.

Ezrin/radixin/moesin (ERM) proteins are essential for microvillus formation and their connection to PIX signalling is of interest. Antisense phosphorothioate oligonucleotide mixtures against ERM mRNAs induce disappearance of microvilli in epithelial cells (Takeuchi et al., 1994). The breakdown of microvilli is commonly observed in the early stage of apoptosis, when the ERM proteins are found to translocate from the microvilli to the cytoplasm (Kondo et al., 1997). Moesin has been reported to be phosphorylated by the Rho kinase and the related myotonic-dystrophy-kinase-related Cdc42-binding kinase (MRCK), which is an effector of Cdc42 (Leung et al., 1998; Oshiro et al., 1998). Thus, PIX at least can play a role in the activation of the ERM proteins by recruitment of MRCK.

In conclusion, we show that a C-terminal domain of αPIX and β1PIX plays a key role in the dimerization and localization of PIX, which are required to drive changes in cell morphology. A proposed model of how the PAK, PIX and GIT1 function (Fig. 7) suggests that the FC provides an important docking site, which is consistent with PAK playing a key role in these structures (Manser et al., 1997). When PAK is activated by Cdc42-GTP or Rac1-GTP, subsequent phosphorylation of PIX and GIT1 must modulate some as-yet-unidentified function. PIX isoforms are substrates of PAK but their in vitro exchange activity was unaffected by phosphorylation. By contrast, the Rac1 GEF Vav1 is potently activated upon tyrosine phosphorylation by Src-family members (Crespo et al., 1997). Although β1PIX associates with the membrane via the coiled-coil domain, we do detect a significant amount of soluble β1PIX, suggesting this process is regulated. As membrane association is critical for PIX to drive the local formation of microvillus-like structures and membrane ruffles, it will be important to determine what other molecular interactions play a role in this process.

This work is supported by the Glaxo Singapore Research Fund. The GenBank accession number for β2PIX is AY034823. While we were revising the manuscript, Kim et al. published similar observations that βPIX could form homodimers