Interaction of paxillin with p21-activated Kinase (PAK). Association of paxillin alpha with the kinase-inactive and the Cdc42-activated forms of PAK3.

p21-activated kinases (PAKs) are implicated in integrin signalings, and have been proposed to associate with paxillin indirectly. We show here that paxillin can bind directly to PAK3. We examined several representative focal adhesion proteins, and found that paxillin is the sole protein that associates with PAK3. PAK3 associated with the alpha and beta isoforms of paxillin, but not with gamma. We also show that paxillin alpha associated with both the kinase-inactive and the Cdc42-activated forms of PAK3 in vivo, without affecting the activation states of the kinase. A number of different functions have been ascribed to PAKs; and PAKs can bind directly to growth factor signaling-adaptor molecule, Nck, and a guanine nucleotide exchanger, betaPIX. Our results revealed that paxillin alpha can compete with Nck and betaPIX in the binding of PAK3. Moreover, paxillin alpha can be phosphorylated by PAK3 at serine. Therefore, paxillin alpha, but not gamma, appears to be capable of linking both the kinase-inactive and activated forms of PAK3 to integrins independent of Nck and betaPIX, as Nck links PAK1 to growth factor receptors. Our results also revealed that paxillin is involved in highly complexed protein-protein interactions in integrin signaling.

The small GTPases Rac and Cdc42 are not the sole activators of the PAK kinase activity. It has been shown that several lipids, particularly sphingolipids, activate PAK1 independent of the action of Rac or Cdc42 (21). The kinase activity of PAKs is also regulated through interaction with the guanine-nucleotide exchange factor PIX/Cool family proteins (reviewed in Ref. 22). It has been shown that PAK1 is activated through interaction with ␤PIX/p85Cool-1 even in the absence of direct binding to Cdc42 (23). ␣PIX/Cool-2 also binds to activate PAK3 kinase (24). On the other hand, binding of p50Cool-1, a splice variant of ␤PIX/p85Cool-1, to PAK3 represses the activation of PAK3 by upstream activators such as Dbl (25). Accordingly, the kinase activity of cellular PAKs has been shown to be regulated by a variety of external stimuli that act through cell surface receptors, including G protein-coupled receptors (4), growth factor receptor tyrosine kinases (26), proinflammatory cytokine receptors (18), Fc receptors (27), and integrins (28,29). However, PAK kinase may not be constantly associated with these cell surface molecules. It has been shown that growth factor stimulations, such as by epidermal growth factor and plateletderived growth factor, recruit PAK1 to the activated receptors, whereby PAK kinase appears to be activated (26,30). Nck, an adaptor molecule recruited to the activated growth factor receptors (31,32), has been shown to bind to PAK1, and implicated to serve as a link connecting the kinase activity of PAK1 to the receptor signaling pathways (30,34,35). A similar scenario has been proposed in photoreceptor guidance receptor signaling in Drosophila (36). However, the molecular mechanisms involved in the recruitment and activation of PAKs in most of the other receptors have not been elucidated.
Paxillin acts as a scaffolding adaptor protein in integrin signaling, by binding to several other integrin-assembly proteins including vinculin, integrin ␤ 1 , Fak, and c-Src (reviewed in Refs. 37 and 38). Paxillin is tyrosine phosphorylated upon integrin activation (39), and thus creates binding sites for several Src homology 2 domain-containing proteins such as Crk-I, Crk-II, Crk-L, and Csk (reviewed in Ref. 38). Tyrosine phosphorylation of paxillin has been shown to be important for cell spreading on the extracellular matrix (40), cell cycle progression (39), as well as adhesion-dependent function of leukocytes (41). Paxillin is composed of multiple isoforms, ␣, ␤, and ␥ (42). These isoforms exhibit different biochemical properties in their binding to other proteins such as vinculin and Fak. For example, both the ␣ and ␤ isoforms bind to Fak whereas the ␥ isoform does not (42).
It has been shown that PAK1 is well co-localized with paxillin at Cdc42-and Rac1-driven small substrate adhesions at the cell periphery, called focal complexes (23); or co-localized with integrin at focal adhesions (43). Turner et al. (44) have shown the association of PAK3 with paxillin. They reported that this association is not mediated by the direct interaction of these two proteins, but mediated through the aid of two other proteins. They showed that paxillin directly binds to p95PKL, which also binds to ␤PIX; and proposed a model in which paxillin associates with PAK3 through three steps: paxillin binds to p95PKL, p95PKL then binds to ␤PIX, and ␤PIX finally binds to PAK3 (44).
To analyze the mechanism involved in the recruitment and activation of the PAK kinases in integrin signaling, we here examined the binding of PAKs to focal adhesion proteins. Consistent with a previous report by Turner et al. (44), we found that paxillin associated with PAK3. On the other hand, other focal adhesion proteins, including Fak, Pyk2, vinculin, talin, tensin, and p130 Cas , did not exhibit a significant association with PAK3. Among paxillin isoforms, the ␣ and ␤ isoforms significantly associated with PAK3, whereas the association with PAK3 was only marginally detectable with the ␥ isoform.
In contrast to the model previously proposed (44), however, we found that paxillin ␣ could bind directly to PAK3. We also showed that paxillin ␣ could be phosphorylated by the Cdc42activated PAK3. Our results collectively indicates that paxillin ␣ fulfills several biochemical properties required to serve to link PAK3 to integrin signaling, which have been ascribed to Nck in linking PAK1 to growth factor receptor signaling (30,34,35,45). Moreover, we found that paxillin ␣ associates with both the kinase-inactive and the Cdc42-activated forms of PAK3 in vivo; and that the paxillin ␣ binding to PAK3 could be competitive with the ␤PIX binding to PAK3 in vivo and in vitro. It has been shown that ␤PIX binding to PAK3 activates the kinase activity of PAK3 (23); thus, according to the previous model for the association of PAK3 with paxillin (44), only the ␤PIX-bound and thereby activated form of PAK3 seems to be associated with paxillin. Therefore, our results indicate that the association of paxillin with PAK3 through the aid of p95PKL and ␤PIX (44) may not be the sole way of the interaction of paxillin with PAK3. On the other hand, our model of the direct binding of PAK3 to paxillin ␣ even in the absence of ␤PIX is consistent with the recent finding that the kinase-inactive and/or nonautophosphorylated form of PAK1 is preferentially localized to focal adhesions (45).

EXPERIMENTAL PROCEDURES
Cells and Tissue-3Y1 and COS7 cells were cultured with Dulbecco's modified Eagle's medium (with 4.5 g of glucose/liter) supplemented with 5 and 10% fetal calf serum (Hyclone Laboratories), respectively.
The autoinhibitory domain of PAK3 (78 -146 aa), corresponding to the similar region of PAK1 (83-149 aa) that has been shown to act as inhibitor for PAK1 in vivo (16,48), was synthesized using oligonucleotides 5Ј-ATGGATCCCATACGATTCATGTGGGTTTTGATGCAGTC-3Ј and 5Ј-ATGCGGCCGCCTAACTTTTATCTCCTGACGTAAAGCTC-3Ј, and the resulting PCR BamHI-NotI fragment was ligated into pEBG vector to be fused in-frame to the COOH terminus of GST protein.
Mouse Nck cDNA was a gift from H. Hanafusa (Osaka Bioscience Institute), and amplified using oligonucleotides 5Ј-ATGGATCCATG-GCTGAAGGAGTGGTGGTGGTGGCC-3Ј and 5Ј-ATGGATCCTCAA-GACAAATGCTTGACGAG-3Ј. The resulting BamHI-BamHI fragment was ligated into pGEX2TK vector, to be fused in-frame to the COOH terminus of GST protein.
Nucleotide sequences were confirmed with all these plasmids after the construction. pEFBOS-HA-Cdc42 G12V was a gift from S. Kuroda and K. Kaibuchi (Nara Institute of Science and Technology).
Proteins encoded by pGEX vectors were expressed in Escherichia coli by induction with isopropyl-␤-D-thiogalactopyranoside, and proteins encoded by the pEBG vector were expressed in COS7 cells, as described previously (49). These GST-tagged proteins were then subjected to purification using glutathione-Sepharose beads, as previously described (50). Proteins encoded by the pQE-30 vector were expressed in E. coli by induction with isopropyl-␤-D-thiogalactopyranoside. These His-tagged proteins were then subjected to purification using nickel beads according to the manufacturer's protocol (Qiagen) and further fractionated on Mono Q column using Biologic System (Bio-Rad), if necessary. Nck was purified as followed; GST-Nck was cleaved by thrombin protease at room temperature for 2 h and then purified by Mono Q column chromatography using the Biologic System. Proteins encoded by the pAcG2T and pVL1392 vectors were expressed in Sf9 cells using the baculovirus system according to the manufacturer's instructions (Pharmingen).
Antibodies-Anti-␤PIX antibody was prepared by immunizing rabbit with the GST fusion form of the NH 2 -terminal half of human ␤PIX/ p85Cool-1, which corresponds to the entire region of p50Cool-1. Antipaxillin antibody was previously described (42). Antibodies against the following proteins were purchased from commercial sources: PAK1 and PAK3 from StressGen; talin and vinculin from Sigma; Nck, Fak, and Pyk2 from Transduction Labs; tensin from Chemicon; p130 Cas from Upstate Biotechnology; and HA-tag sequence and GST-tag sequence from BAbCO.
Protein Binding and Immunoblotting Analyses-For in vitro protein binding analysis, cell lysates prepared by 0.5% Triton X-100 buffer, or purified protein preparations which were dialyzed against 0.5% Triton X-100 buffer, were mixed with GST fusion proteins bound to glutathione-Sepharose beads, incubated for 2 h at 4°C and then washed four times with 0.5% Triton X-100 buffer. For in vivo protein binding analysis, 300 g of COS7 cell lysates prepared by 1% Nonidet P-40 buffer were mixed with 5 l of glutathione-Sepharose beads, incubated for 2 h at 4°C. Beads were then washed four times with 1% Nonidet P-40 buffer. For immunoblotting, protein samples retained on glutathione beads were boiled in Laemmli's SDS sample buffer, separated on 8% SDS-PAGE, transferred to membrane filters (Immobilon P, Millipore), and subjected to immunoblotting analysis as previously described (50). Antibodies retained on the filter membranes were visualized by horseradish peroxidase-conjugated secondary antibodies (Jackson Immu-noResearch) coupled with an enzyme-linked chemiluminescence method according to the manufacturer's instructions (Amersham Pharmacia Biotech).
In Vitro Kinase Assays-Protein precipitants with glutathione-Sepharose beads were washed four times in 1% Nonidet P-40 buffer and then twice in a kinase buffer (50 mM Tris-HCl (pH 7.4), 10 mM MgCl 2 , 10 mM MnCl 2 , 0.2 mM dithiothreitol). The beads were then resuspended in 20 l of the kinase buffer containing 2 g of myelin basic protein (MBP) (Sigma), and incubated for 20 min at 30°C in the presence of 10 Ci of [␥-32 P]ATP (6,000 Ci/mmol; Amersham Pharmacia Biotech). The reactions were terminated by boiling in Laemmli's SDS sample buffer, and samples were visualized by separating on 12% SDS-PAGE followed by autoradiography.
In Vitro Phosphorylation and Phosphoamino Acid Analysis-To prepare paxillin ␣ protein which was not contaminated with serine/threonine kinase activities, His-paxillin ␣ bound to nickel beads was incubated in a denaturing buffer (6 M guanidine HCl, 0.1 M Na phosphate (pH 8.0)) at room temperature for 1 h, which was then followed by sequential incubation with a renaturing buffer A (3 M guanidine HCl, 0.1 M Na phosphate (pH 8.0), 5 mM dithiothreitol), a renaturing buffer B (1.5 M guanidine HCl, 0.1 M Na phosphate (pH 8.0), 5 mM dithiothreitol), and a renaturing buffer C (100 mM NaCl, 0.1 M Na phosphate (pH 8.0), 5 mM dithiothreitol). Each incubation with renaturing buffers was carried out at 4°C for 1 h. His-paxillin ␣ was then subjected to elution from the nickel beads according to the manufacturer's instruction (Qiagen), and dialyzed against 1% Nonidet P-40 buffer. GST-PAK3 bound to glutathione beads was washed four times in 1% Nonidet P-40 buffer and then twice in the kinase buffer. The beads were then resuspended in 20 l of the kinase buffer containing the "kinase-free" Hispaxillin ␣ and incubated for 20 min at 30°C in the presence of 10 Ci of [␥-32 P]ATP (6,000 Ci/mmol). The reactions were terminated by boiling the sample in Laemmli's SDS sample buffer, and samples were then separated on 8% SDS-PAGE. After excision of the radiolabeled band from the gel, phosphoamino acid analysis was performed as described (51), where 1000 cpm of the radioactivity was separated on cellulose thin-layer plates electrophoresis with pH 3.5 buffer (10:1:189 mixture of glacial acetic acid, pyridine, and water) at 1000 V for 60 min, after being mixed with phosphoamino acid standards (Sigma). 32 P-Labeled phos-phoamino acids were then identified by autoradiography and phosphoamino acid standards were identified by spraying the place with ninhydrin as described (51).

Association of PAKs with Paxillin, but Not Other Focal
Adhesion Proteins-The PAK kinases are localized to focal complexes and focal adhesions, and stimulated upon integrin activation (23,28,29,43). To understand the mechanism for the regulation of the PAK kinases at focal adhesions, we examined several known focal adhesion proteins for their binding to PAKs. We prepared GST fusion forms of PAK1 and PAK3 proteins produced in COS7 cells; and proteins co-precipitated from mammalian cell lysates with these GST fusion proteins were analyzed. As shown in Fig. 1A, paxillin was clearly coprecipitated with PAK1 and PAK3 from extracts of 3Y1 cells and NRB tissues, both expressed PAK1 and PAK3 endog-  7), or GST alone (lanes 4 and 8), each coupled with glutathione-Sepharose beads. After washing, proteins bound to the beads were separated on SDS-PAGE and subjected to sequential immunoblotting analysis using antibodies against focal adhesion proteins, as indicated. Antibodies retained on filter membranes were visualized by the enzyme-linked chemiluminescence method. 20 g of total cell lysates of 3Y1 (lane 1) and NRB (lane 5) were also included (total). Controls were without cell lysates (lanes 9 -11). GST-PAK1 and GST-PAK3 were produced in COS7 cells, and GST was in E. coli; and purified on glutathione beads before use, as described under "Experimental Procedures." B, association of paxillin isoforms with PAK1 and PAK3. 600 g of 3Y1 (lanes [2][3][4][5] or NRB (lanes 7-10) lysates as above were incubated with 5 g of GST-paxillin ␣ (lanes 2 and 7), GST-paxillin ␤ (lanes 3 and 8), GSTpaxillin ␥ (lanes 4 and 9), or GST alone (lanes 5 and 10), each coupled with glutathione-Sepharose beads. After washing, proteins bound to the beads were subjected to separation on SDS-PAGE and sequential immunoblotting analysis as above using antibodies against PAK1 and PAK3. 20 g of total cell lysates of 3Y1 (lane 1) and NRB (lane 6) were also included (total). Controls were without cell lysate (lanes [11][12][13][14]. All GST proteins of paxillin isoforms were produced in the baculovirus system and purified on glutathione beads before use, as described under "Experimental Procedures." enously (see Fig. 1B). On the other hand, other focal adhesion proteins we examined, including tensin, talin, p130 Cas , Fak, Pyk2, and vinculin, were not detected to be co-precipitated with these recombinant PAK proteins from either cell lysate (Fig.  1A).
Paxillin consists of two isoforms (␣ and ␤) in rodents and three isoforms (␣, ␤, and ␥) in human (42,52). Using GST fusion proteins of each paxillin isoform produced in the baculovirus system, we found that the ␣ and ␤ isoforms exhibited significant associations with PAK1 and PAK3, whereas the association with the PAKs was very weak and only marginally detectable with the ␥ isoform (Fig. 1B). We could not find cell extracts of human origin expressing both PAK1 and PAK3, and thus used rat 3Y1 cell extracts for this analysis.
Direct Binding of Paxillin ␣ to PAK3-It has already been reported that paxillin is associated with PAK3 (44). This association, however, has been proposed to be not mediated by the direct interaction of these two proteins, but mediated with the aid of p95PKL and ␤PIX proteins: paxillin binds to p95PKL, p95PKL binds to ␤PIX, and ␤PIX binds to PAK3 (44). It has been shown that the LD4 motif of paxillin is responsible for the direct binding of paxillin to p95PKL, and therefore also responsible for the association of paxillin with PAK3 and ␤PIX (44). Using GST fusion proteins of the full-length paxillin ␣ and the LD4 motif, we compared amounts of PAK3 and ␤PIX co-precipitated with these fusion proteins. We found that almost equal amounts of PAK3 were precipitated with the full-length paxillin ␣ and with the LD4 motif, under the experimental conditions shown in Fig. 2A. However, reblotting the same membrane filter with an anti-␤PIX antibody revealed that the amount of ␤PIX co-precipitated with the full-length paxillin ␣ was much higher than that with the LD4 motif. The same results were obtained with both NRB and 3Y1 cell lysates ( Fig.  2A) in repeated experiments; indicating a possible independent behavior of PAK3 and ␤PIX in their association with paxillin ␣.
These observations prompted us to investigate whether paxillin ␣ could bind to PAK3, independent of ␤PIX. We prepared His-tagged paxillin ␣, which was expressed in the baculovirus system, purified on nickel beads, and then eluted; and GST-PAK3, which was expressed in COS7 cells and purified on glutathione-Sepharose beads (Fig. 2B). Using these purified preparations of 100 ng of His-paxillin ␣ and 500 ng of GST-PAK3, we tested whether paxillin ␣ and PAK3 bound directly. As shown in Fig. 2B, more than 10% of His-paxillin ␣ was recovered with the GST-PAK3 bound to glutathione-Sepharose beads. Contamination of ␤PIX in these protein preparations as well as in the GST-PAK3 pull-down fraction was minimal (Fig.  2B, lower lanes). Therefore, paxillin ␣ appears to be capable of binding to PAK3 directly in vitro, even in the absence of ␤PIX. Such direct binding was also detected between GST-PAK1 and His-paxillin ␣ (data not shown).
The NH 2 Terminus Domain of PAK3 Is Responsible for the Binding to Paxillin ␣-We next determined the structural basis for the direct binding of paxillin ␣ to PAK3. The coding sequence of PAK3 contains four potential Src homology 3 (SH3) binding Pro-X-X-Pro motifs (where X is any amino acid; P1 to P4), the Cdc42/Rac-binding domain termed the p21-binding domain (65-128 aa), an autoinhibitory domain overlapping with the p21-binding domain, an acidic region, the PIX/Coolbinding region (184 -205 aa), and the kinase domain (268 -523 aa) (7, 23) (see Fig. 3). We prepared various deletion mutants of PAK3 as shown in Fig. 3, each fused to GST, and expressed in COS7 cells (full-length and the M1 mutant) or in E. coli (the M2-M7 mutants), and purified on glutathione-Sepharose beads. Each 5 g of these PAK3 peptides bound to the beads was then incubated with 500 ng of purified His-paxillin ␣ or with 200 ng of purified His-␤PIX, which were produced in the baculovirus system and in E. coli, respectively. As shown in Fig. 3, upper lanes, the M1 and M6 mutants, but not other mutants, bound to His-paxillin ␣, with almost equal affinities seen with the full-length PAK3; suggesting that the NH 2 -terminal domain of PAK3 containing the P1 and P2 motifs is primarily responsible for the binding to paxillin ␣. Contamination of ␤PIX in these protein preparations as well as in the GST-PAK3 pull-down fraction was minimal (Fig. 3, middle  lanes). On the other hand, the M1, M3, M5, and M7 mutants,  8 and 12), each bound to glutathione beads. After washing, proteins retained on the beads were separated on SDS-PAGE and subjected to sequential immunoblotting analysis using antibodies against PAK3 and ␤PIX. Lanes 1 and 9 included 20 g each of the total cell lysates of NRB and 3Y1, respectively (total). Controls were without cell lysate (lanes 13-19). The GST fusion form of the full-length paxillin ␣ was produced in the baculovirus system, and the GST fusion forms of the LD motifs were produced in E. coli; and purified on glutathione beads before use. Note that ␤PIX appeared as multiple bands in NRB lysates as previously shown (23,44), while it showed a single band in 3Y1 lysates, as indicated by arrowheads. B, direct binding of PAK3 to paxillin ␣ in vitro. His-paxillin ␣ (100 ng), which was produced in the baculovirus system, purified on nickel beads and then eluted, was incubated with 500 ng of GST-PAK3 (lane 3), expressed in COS7 cells and purified on glutathione-Sepharose beads, or GST alone (lane 4), each bound to glutathione beads. After washing, proteins bound to the beads were separated on SDS-PAGE and subjected to immunoblotting analysis using anti-paxillin antibody. Lane 2 included 10 ng of the purified His-paxillin ␣. Immunoblotting of the same membrane filter with anti-␤PIX antibody was also performed to show that the binding was conducted essentially free from the contamination of ␤PIX (lower panel); where 10 ng of purified His-␤PIX (see Fig. 3A) was included in but not the M2, M4, and M6 mutants, bound to His-␤PIX, confirming that our preparation of recombinant His-␤PIX binds to the 184 -205 aa region of PAK3 (Fig. 3, lower lanes), as has been previously reported (23). Each 10 ng of ␤PIX and paxillin ␣ proteins was included in Fig. 3, lanes 1 and 2, respectively; and judging from the intensities of the chemiluminescence signals from paxillin ␣ and ␤PIX co-precipitated with PAK3, the binding affinity of ␤PIX toward PAK3 seemed to be at least severalfold higher than that of paxillin ␣ toward PAK3.
Paxillin ␣ Binding to PAK3 Competes with ␤PIX and Nck Binding to PAK3-We next examined the interaction among paxillin ␣, PAK3, and ␤PIX. GST-PAK3 (250 ng) bound to glutathione-Sepharose beads was incubated with His-paxillin ␣ (250 ng) in the presence of increasing amounts of His-␤PIX (0 -800 ng). All of these protein preparations were purified ones, as used above. As shown in Fig. 4A, ␤PIX binding to PAK3 appeared to compete with paxillin ␣ binding to PAK3. On the other hand, the ␥ isoform of paxillin, which did not exhibit significant binding affinity to PAK3, did not interfere with the ␤PIX binding to PAK3 (Fig. 4A). The interaction of these three proteins was also examined in vivo by expressing GST-paxillin ␣, HA-PAK3, and increasing amounts of His-␤PIX in COS7 cells. Consistent with the notion that ␤PIX competes with paxillin ␣ in binding to PAK3, increasing expression of His-␤PIX reduced the amounts of HA-PAK3 co-precipitated with GST-paxillin ␣ (Fig. 4B). We confirmed that ␤PIX does not bind directly to paxillin ␣ (data not shown), as reported previously (44).
Nck has also been shown to bind to the P1 motif of PAK1, via the SH3 domain of Nck (30,34). The amino acid sequence of the P1 motif is highly conserved in all three PAK isoforms, and we thus examined whether paxillin ␣ binding to PAK3 competed with Nck binding to PAK3. GST-PAK3 (250 ng) bound to glutathione-Sepharose beads was incubated with His-paxillin ␣ (250 ng) in the presence of increasing amounts of Nck (0 -1000 ng). Using purified preparations of these proteins, we found that Nck binding to PAK3 was competitive with paxillin ␣ binding to PAK3 (Fig. 4C). Again, 10 ng each of paxillin ␣ and Nck proteins was included in Fig. 4C; and judging from the intensities of the chemiluminescence signals from paxillin ␣ and Nck bands co-precipitated with PAK3, the apparent binding affinities of these two proteins toward PAK3 seemed to be similar.
Paxillin ␣ Associates with Both the Kinase-inactive and the Cdc42-activated Forms of PAK3-The association between PAK1 and Nck has been shown to be constitutive, irrespective of the PAK's activation state (34,53). We thus examined paxillin ␣ binding to PAK3 with regard to the activation state of PAK3. HA-PAK3 was coexpressed with GST-paxillin ␣ in COS7 cells in the presence or absence of the dominant-active (DA) form of HA-tagged Cdc42. As shown in Fig. 5A, GST-paxillin ␣ co-precipitated almost equal amounts of HA-PAK3 irrespective of the presence or the absence of HA-Cdc42 DA. Co-precipitation of HA-Cdc42 DA together with HA-PAK3 and GST-paxillin ␣ was also observed (Fig. 5A). On the other hand, the LD4deletion mutant of GST-paxillin ␣ failed to co-precipitate HA-PAK3 (Fig. 5A). In vitro kinase assay using MBP as a substrate was performed simultaneously. As shown in Fig. 5A, HA-PAK3 bound to GST-paxillin ␣ in the absence of HA-Cdc42 DA showed a very low activity in phosphorylating MBP, whereas it showed a high activity in the presence of HA-Cdc42 DA.
It has been shown that paxillin is associated with some cellular serine/threonine kinases whose identities were unknown (54). The autoinhibitory domain of PAK1 has been shown to act as a specific inhibitor for the PAK1 kinase activity (16,48). To confirm that the kinase activities co-precipitated with GST-paxillin ␣ in the presence of HA-Cdc42 DA was primarily due to the kinase activity of PAK3, the PAK3 autoinhibitory domain was coexpressed. As shown in Fig. 5B, coexpression of the autoinhibitory domain of PAK3 with HA-PAK3 and HA-Cdc42 DA in COS7 cells largely reduced the kinase activity co-precipitated with GST-paxillin ␣, although the amounts of HA-PAK3 co-precipitated with GST-paxillin ␣ were almost unchanged. Thus, PAK3 accounts for the major kinase activity that is associated with GST-paxillin ␣ in the presence of HA-Cdc42 DA. The results shown in Fig. 5, A and B, collectively indicate that the binding of paxillin ␣ to PAK3 per se does not significantly activate its kinase activity, or interfere with the Cdc42-mediated activation of the PAK3 kinase activity.
Paxillin ␣ Is Directly Phosphorylated by PAK3 at Serine-Finally, we examined whether paxillin ␣ could be phosphorylated by PAK3, again with the analogy to Nck phosphorylation by PAK1 (34,45). Paxillin is well phosphorylated on serine in vivo (54 -56), and is known to associate with some cellular serine/threonine kinase activities (54). Likewise, we found that  4 -10), or GST alone (lane 11), each bound to glutathione beads. After washing, His-paxillin ␣ proteins bound to the beads were subjected to immunoblotting analysis using anti-paxillin (upper lanes), and His-␤PIX proteins bound to the beads were subjected to immunoblotting analysis using anti-␤PIX antibody (lower lanes). Possible contamination of ␤PIX in these protein preparations as well as in the GST-PAK3 pull-down fraction was assessed (middle lanes). Lanes 1 and 2 included 10 ng each of His-␤PIX and His-paxillin ␣, respectively (authentic control). Controls were also GST fusion proteins only (lanes [12][13][14][15][16][17][18][19][20]. Each of the deletion mutants of PAK3 is shown at the bottom panel, where each number corresponds to amino acid residues of mouse PAK3 (7) (see "Experimental Procedures" and text for details). GST fusion forms of the full-length PAK3 and the M1 mutant were expressed in COS7 cells, and all the other mutants (M2-M7) were produced in E. coli; and all purified on glutathione beads before use. His-paxillin ␣ was the same purified preparation as in the His-paxillin ␣ protein preparation, which was produced in the baculovirus system and collected on nickel beads, was still highly contaminated with serine/threonine kinase activities, although the purity of the protein seemed to be more than 95% as assessed by Coomassie staining (data not shown). To remove the contaminating kinase activities, His-paxillin ␣ protein bound to nickel beads was treated with guanidine HCl, and then subjected to a renaturing process. GST fusion form of PAK3 has been known to be constitutively active in the kinase activity (4). The kinase-free His-paxillin ␣ (see Fig. 6, lane 1) was then incubated with a purified preparation of GST-PAK3 in the presence of ATP and divalent cations; and found to be well phosphorylated by GST-PAK3 in vitro (Fig. 6A). Phosphoamino acid analysis revealed that the phosphorylation had exclusively taken place at the serine residue(s) (Fig. 6B). Note that phosphorylation of paxillin ␣ in vitro in GST-PAK3 precipitants was already observed in Fig. 5, A and B, where coexpression of the PAK3 autoinhibitory domain substantially, but not completely, suppressed the phosphorylation. DISCUSSION We showed in this paper that among representative focal adhesion proteins, paxillin is the sole protein that can associate with PAK1 and PAK3. Moreover, our results revealed that the ␣ isoform of paxillin can bind directly to PAK3. In vivo association of paxillin ␣ with PAK3 took place both with the kinaseinactive and the Cdc42-activated forms of PAK3; and this association did not seem to change or interfere with the activation states of the kinase. We furthermore, demonstrated in vitro that paxillin ␣ can be phosphorylated by PAK3 at serine.
These biochemical properties of paxillin ␣ are similar to those ascribed to Nck in its binding to PAK1 (30,34,35,45). Nck binding to PAK1 had no stimulatory effect on PAK1 kinase activity, and it did not alter the ability of PAK1 to be stimulated by Rac-or Cdc42-GTP␥S; and Nck can be phosphorylated by PAK1. Nck has been shown to associate with several growth factor receptors, and is thus thought to link PAK1 to growth factor receptor signalings. Nck has been shown to bind to PAK1 at its NH 2 -terminal proline-rich sequence. We showed that paxillin binds to the NH 2 -terminal region of PAK3, and the binding competes with the Nck binding to PAK3. Our results also indicated that the binding affinities of paxillin ␣ and Nck in their direct interaction with PAK3 seem to be similar. It has been well documented that PAKs are activated by integrinmediated cell adhesion to the extracellular matrixes (28,29). Therefore, we collectively propose that paxillin ␣ is capable of linking both the kinase-active and -inactive forms of PAK3 to At 36 h after transfection, GST-paxillin ␣ was precipitated with glutathione beads from 300 g of each cell lysate; and proteins retained on the beads were separated on SDS-PAGE and subjected to sequential immunoblotting analysis using antibodies against HA-tag and GST, and shown in the lower panel (pull down with GST-pax ␣). In the upper panel, 20 g of each total cell lysate was analyzed by immunoblotting to show the expression levels of each exogenous protein (total). Each arrow indicates positions of each corresponding protein. C, competitive binding of paxillin ␣ and Nck toward PAK3 in vitro. Purified GST-PAK3 (250 ng) bound to glutathione beads was incubated with purified Hispaxillin ␣ (250 ng) in the presence of increasing amounts of purified Nck (0, 200, 400, and 1000 ng in lanes 3-6, respectively), or only with Nck (1000 ng in lane 7). After washing, proteins bound to the beads were processed and analyzed as above using antibodies against paxillin and Nck. Lanes 1 and 2 included 10 ng of His-paxillin ␣ and Nck, respectively. GST-PAK3 and His-paxillin ␣ were the same purified preparations as in Fig. 2. Coomassie Brilliant Blue (CBB) staining of the Nck preparation, which was produced as a GST fusion form in E. coli, then cleaved from the tag and purified as described under "Experimental Procedures," is shown on the right (lane 8). A protein band corresponding to Nck is shown by an asterisk. Molecular sizes are shown on the left. integrins, as Nck serves to link PAK1 to growth factor receptors. However, since our results revealed that paxillin ␣ and Nck compete with each other in their binding to PAK3, a single molecule of PAK3 does not appear to bridge these two proteins.
Turner et al. (44) were the first to show the association of paxillin with PAK3. However, they could not detect direct binding of paxillin with PAK3 in vitro, and instead proposed that PAK3 associates indirectly with paxillin through the aid of other proteins, p95 PKL and ␤PIX, as already mentioned earlier.
Contrary to this, we could detect direct binding of PAK3 to paxillin ␣ in vitro, in the absence of ␤PIX. For the in vitro binding, we used PAK3 molecules produced in COS7 cells and paxillin ␣ molecules produced in the baculovirus system. On the other hand, Turner et al. (44) used the 54 -313 amino acid region of paxillin produced in E. coli and PAK3 molecules synthesized by in vitro transcription/translation system. We also found that the in vitro translated product of PAK3 was almost incapable of binding to GST-paxillin ␣. 2 To examine the bindability of the in vitro translated PAK3 molecules, Turner et al. (44) have confirmed that it could bind to ␤PIX, which we also detected. Interaction between PAKs and ␤PIX is, however, mediated through the proline-rich sequence of PAKs and the SH3 domain of ␤PIX; thus the binding of these two proteins can be detected even with a protein overlay assay using ␤PIX as a probe, where PAK1 protein was run on SDS-PAGE and thereby denatured (23). Binding of the in vitro translated PAK3 to ␤PIX may therefore have been able to be detected, even if the PAK3 may not have preserved its native structure. We found that the in vitro translated PAK3 protein ran as doublet bands on SDS-PAGE, as also seen previously (Fig. 7 in Ref. 44). Some artificial events may have occurred with PAK3 molecules in the in vitro transcription/translation system, and might have hampered its capacity to bind to paxillin.
Paxillin consists of three isoforms (␣, ␤, and ␥), which exhibit different properties of protein binding (42). Association with PAK3 has been reported to be primarily mediated through the LD4 motif (44), which we also confirmed. The ␤ and ␥ isoforms are generated by the insertion of each specific exon just after the LD4 motif of the ␣ isoform. We have thus tested and found that the ␣ and ␤ isoforms of paxillin exhibit significant affinities to PAK1 and PAK3, while the ␥ isoform exhibits a very weak and only marginally detectable level of association with these PAKs. It is interesting to note that this property of PAKs in the association with paxillin isoforms is similar to that of FIG. 5. Paxillin ␣ associates with PAK3 in vivo without changing its activation states. A, paxillin ␣ associates with the kinaseinactive and the Cdc42-activated forms of PAK3 in vivo. COS7 cells were transfected with 2.5 g of pEFBOS-HA-PAK3 in the presence or absence of 2.5 g each of pEFBOS-HA-Cdc42 G12V, pEBG-paxillin ␣, or the pEBG-paxillin ␣ (⌬LD4), as indicated. At 36 h after transfection, GST-paxillin ␣ or the LD4-deletion mutant was precipitated with glutathione beads from 300 g of each cell lysate. After washing, proteins retained on the beads were separated on SDS-PAGE and subjected to sequential immunoblotting analysis using antibodies against HA-tag and GST, and shown in the middle panel (pull-down with GST-pax ␣ or GST-pax ␣ (⌬LD4)). Each half-amount of the precipitants was simultaneously subjected to in vitro kinase assay using [␥-32 P]ATP and MBP as a substrate as described under "Experimental Procedures," followed by SDS-PAGE and autoradiography; and shown in the bottom panel (in vitro kinase assay); where molecular sizes are shown on the left. In the top panel, 20 g of each total cell lysate was analyzed by immunoblotting to show the expression levels of exogenous proteins (total). Each arrow indicates positions of each corresponding protein. Protein bands corresponding to HA-PAK3 and HA-Cdc42, both detected by an antibody against the HA-tag, were distinguished by their molecular sizes. B, assessment of kinase activities bound to the GST-paxillin ␣ by use of the PAK3 kinase inhibitor. COS7 cells were transfected with 2.5 g each of pEFBOS-HA-PAK3, pEFBOS-HA-Cdc42 G12V, and pEBG-paxillin ␣, in the absence (lane 1) or presence (lane 2) of 2.5 g of pEBG-PAK3 KI, which encoded the autoinhibitory domain for the kinase activity (see "Experimental Procedure"). At 36 h after transfection, GST-paxillin ␣ was precipitated using glutathione beads from 300 g of each cell lysate and subjected to in vitro kinase assay as above, and the resulting autoradiograph is shown (in vitro kinase assay). Each halfamount of the precipitants was separated on SDS-PAGE and analyzed using antibodies against HA-tag and GST, to examine the amounts of HA-PAK3 co-precipitated with GST-paxillin ␣ and that of GST-PAK3 KI, respectively; and shown below (pull-down with GST-pax ␣). Each arrow indicates positions of each corresponding protein. phosphorylated His-paxillin ␣ was then excised from the gel, and subjected to phosphoamino acid analysis. Phosphoamino acid standards were visualized by ninhydrin to identify the position of each phosphoamino acid. 1000 cpm of the radioactivity was loaded for thin layer chromatography. In A and B, resulting autoradiographies are shown; and each arrow indicates positions of each corresponding protein, phosphoamino acids, free phosphate, and the origin of the sample spot. Note that the His-tag is composed only of six consecutive histidine residues; thus the serine phosphorylation occurred at His-paxillin ␣ is within the paxillin moiety.
Fak (42), whose interaction with paxillin is also primarily mediated by the LD4 motif of paxillin (44). Among the three isoforms of paxillin, no significant difference in the subcellular localization is detected (52). Thus, the ␥ isoform of paxillin may act as a dominant-negative form with regard to the PAK and Fak signalings mediated by paxillin ␣.
Three isoforms (PAK1, PAK2, and PAK3) of PAKs have been identified in mammals. PAK1, the first isolated PAK (2), has been most throughly studied. Because size and sequence are highly conserved among PAK isoforms, most of the biochemical properties as well as the possible physiological functions ascribed to PAK1 pertain to the other isoforms (reviewed in Refs. 1, 10, and 22), although there may be also some important differences. For example, PAK3 is released from Rac1 following activation, whereas PAK1 and PAK2 remain bound to this GTPase (2, 3, 6). Our analysis was performed mostly with PAK3, but we also showed that paxillin ␣ can associate with PAK1. Our preliminary experiment also suggested that PAK2 seemed to be co-precipitated with GST-paxillin ␣ from NRB lysates, 2 but this has not been able to be confirmed due to the insufficient quality of PAK2 antibody.
The results described in this paper, however, do not preclude a complex formation of the four proteins, paxillin ␣/p95PKL/ ␤PIX/PAK3, in a linear configuration as Turner et al. (44) have proposed. We have isolated a cDNA for Git2 (33), which encodes a paxillin-binding protein. 3 Git2 is highly homologous to p95PKL; and we indeed detected ternary complexes of paxillin ␣/Git2/␤PIX, and Git2/␤PIX/PAK3. 2 However, since paxillin ␣ itself can bind to PAK3, we could not properly assess whether or not these four proteins make the tetra-protein complex in a linear configuration. On the other hand, we showed that both the kinase-inactive and the Cdc42-activated forms of PAK3 can be co-precipitated with paxillin ␣ from COS7 cells. ␤PIX binding to PAK3 activates the kinase activity of PAK3 (23); thus, our results strongly suggest that the tetra-protein complex formation proposed by Turner et al. (44), in which the ␤PIX binding to PAK3 is required, is not the sole way of the association of paxillin ␣ with PAK3. Moreover, we showed that paxillin ␣ association with PAK3 can compete with ␤PIX association with PAK3 in vitro and in vivo. PAK1 has been shown to be colocalized with paxillin at focal complexes at the cell periphery (23); and it has been shown recently that the kinase-inactive and/or non-autophosphorylated form of PAK1 is preferentially localized to focal adhesions (45). Consistent with this, we demonstrated that paxillin ␣ association with PAK3 per se did not activate the kinase activity of PAK3. A fraction of PAK3 was detected at focal adhesions, largely, but not completely colocalized with paxillin ␣. 2 Collectively, it seems reasonable to suppose that paxillin ␣ can associate with PAK3 at focal adhesions, and at focal complexes may as well, even in the absence of ␤PIX.
Additional signaling seems to be required for the activation of the paxillin-bound PAK3. We showed that the catalytically active form of PAK3 can stay associated with paxillin ␣. However, our preliminary results indicate that when paxillin ␣ was highly phosphorylated by PAK3 in vitro, these two phosphorylated proteins could no longer stay bound to each other. 2 Thus, it is possible that when PAK3 is activated and phosphorylates paxillin ␣ at high levels, this activated PAK3 molecule is released from the paxillin ␣. Similarly, PAK1 has been proposed to be dissociated from Nck, when PAK1 is highly activated and phosphorylated (45). However, it should be also noted that the PAKs have two types of effects, one related to its kinase activity and one that is kinase independent (12,13,15,23). Elucidation of the mechanism and timing of the recruitment of PAKs to paxillin ␣, the regulation of the kinase activity as well as the autophosphorylation of the paxillin-bound PAKs, and the phosphorylation of paxillin ␣ by PAKs, will be thus required for a precise understanding of the PAK-mediated integrin signalings.