Requirement for PAK4 in the Anchorage-independent Growth of Human Cancer Cell Lines*

p21-activated protein kinase (PAK) serine/threonine kinases are important effectors of Rho family GTPases and have been implicated in the regulation of cell morphology and motility, as well as in cell transformation. To further investigate the possible involvement of PAK kinases in tumorigenesis, we analyzed the expression of several family members in tumor cell lines. Here we demonstrate that PAK4 is frequently overexpressed in human tumor cell lines of various tissue origins. We also have identified serine (Ser-474) as the likely autophosphorylation site in the kinase domain of PAK4 in vivo . Mutation of this serine to glutamic acid (S474E) results in constitutive activation of the kinase. Phospho-specific antibodies directed against serine 474 detect activated PAK4 on the Golgi membrane when PAK4 is co-expressed with activated Cdc42. Furthermore, expression of the active PAK4 (S474E) mutant has transforming potential, leading to anchorage-indepen-dent growth of NIH3T3 cells. A kinase-inactive PAK4 (K350A,K351A), on the other hand, efficiently blocks transformation by activated Ras and inhibits anchor-age-independent growth of HCT116 colon cancer cells. Taken together, our data strongly implicate PAK4 in oncogenic transformation and suggest that PAK4 activity is required for Ras-driven, anchorage-independent growth. At 28 human STE20 group kinases been identified and comparisons

Mutation of this serine to glutamic acid (S474E) results in constitutive activation of the kinase. Phosphospecific antibodies directed against serine 474 detect activated PAK4 on the Golgi membrane when PAK4 is co-expressed with activated Cdc42. Furthermore, expression of the active PAK4 (S474E) mutant has transforming potential, leading to anchorage-independent growth of NIH3T3 cells. A kinase-inactive PAK4 (K350A,K351A), on the other hand, efficiently blocks transformation by activated Ras and inhibits anchorage-independent growth of HCT116 colon cancer cells. Taken together, our data strongly implicate PAK4 in oncogenic transformation and suggest that PAK4 activity is required for Ras-driven, anchorage-independent growth.
At least 28 human STE20 group kinases have been identified that fall into ten distinct subfamilies based on phylogenic and structural comparisons of their regulatory and catalytic domains (1). To date, six human PAK 1 kinases, which fall into two subfamilies, have been identified. The first subfamily consists of PAK1 (PAK␣), PAK2 (PAK␥, PAK⍜, hPAK65), and PAK3 (PAK␤), which exhibit 80 -90% sequence identity within their catalytic domains. Members of the recently identified second PAK subfamily, PAK4, PAK5, and PAK6, are also highly related to each other but show only about 40 -50% identity to the kinase domains of PAKs 1-3. All PAK family members are characterized by the presence of a p21 binding domain (PBD), which binds activated Rho family GTPases (2)(3)(4)(5)(6)(7). The interaction between activated Rho family GTPases with the PBD motif derepresses the intrinsic kinase activity of PAKs, leading to autophosphorylation in the activation loop and full kinase activity (8 -10). PAKs 1, 2, and 3 interact via their proline-rich domains with the SH3 domain of the PAK-interactive exchange factors PIX/Cool (11,12). In addition to the PIX (PAK-interacting exchange factor) family of DH-PH exchange factors, the PAKs can also bind to adaptors such as Nck and can be recruited directly to growth factor receptors (3,13). Overall, both PAK subfamilies share the same organization and structure in their regulatory and catalytic domains, although PAKs 4, 5, and 6 do not have obvious PIX/Cool binding motifs.
As effectors of Rho family GTPases, PAK kinases play an important role in the regulation of cell morphology and motility by regulating the actin cytoskeleton (14 -19). For example, PAK activity has been implicated in the localized assembly at the leading edge and disassembly at the retracting edge of focal adhesions during cell motility (14,(21)(22). During wound healing of fibroblast monolayers, activated PAK1 rapidly localizes to the leading edge of motile cells (15). Furthermore, overexpression of PAK4 has been shown to induce localized actin polymerization and the formation of filipodia (7). The PAK4dependent changes in the actin cytoskeleton are dependent on PAK4 kinase activity and Cdc42-dependent localization to the Golgi membrane (7). The ability of the PAKs to modulate the actin cytoskeleton is due to kinase-dependent and -independent mechanisms (14, 20 -22).
The role of STE20 family members and Rho family GTPases in regulating mitogen-activated protein kinase pathways is, in general, conserved from yeast to mammals (23)(24)(25)(26)(27)(28)(29). PAK family members have also been implicated in the regulation of the Ras-MEK-ERK MAP-kinase pathway. PAK1 and PAK2 can phosphorylate Raf-1 on a regulatory site that is important for Raf-1 activity (30 -32). PAK1 has also been implicated in the phosphorylation of a specific site in MEK1 that is required for the interaction of MEK1 with Raf1, and PAK1 activity was shown to be required for anchorage-dependent growth factor activation of Raf (33,34).
Growing evidence has implicated PAK kinases in oncogenic Ras-driven, anchorage-independent growth and in the regulation of cell survival (35)(36)(37)(38)(39)(40)(41)(42)(43)(44)(45). For example, kinase inactive alleles of PAK1 block Ras transformation of Rat1 and Schwann cells (41,42). Similarly, dominant negative alleles of PAK2 have been shown to interfere with Ras-mediated transformation (44). Several recent studies have implicated PAK1 in the anchorage-independent growth and invasiveness of human epithelial breast cancer cells (46 -48). In addition, PAK1 was shown to play a role in regulating BAD activation (43,49). Recent work has suggested a similar role for PAK4 in protecting cells from stress-induced apoptosis (50) as well as a role in cell adhesion and anchorage-independent growth of rodent fibroblasts (51).
To determine whether any of the recently identified PAKs (PAK4, 5, 6) play a role in human cancer, we analyzed the expression of PAK kinases and other STE20 family members in tumor samples and tumor cell lines. We find that PAK4 is frequently overexpressed in human tumor cell lines and that the PAK4 gene maps to an amplicon associated with human cancers. In addition, we demonstrate that an activated allele of PAK4 has transforming potential and that an inactive PAK4 mutant blocks Ras-dependent, anchorage-independent growth. Our results strongly implicate PAK4 in tumorigenesis and suggest that PAK4 may be particularly important in Rasdriven cell transformation.

EXPERIMENTAL PROCEDURES
Expression Analysis-Northern blots were performed using standard techniques. The cells and tissues were previously described (52). Northern blots were prepared by running 10 g of total RNA isolated from 60 human tumor lines, 22 human adult tissues, and two human fetal tissues, on a 1.2% denaturing formaldehyde agarose gel and transferred to nylon membranes. Filters were hybridized with random primed [ 32 P]dCTP-labeled probes derived from a 602-bp BstXI/SacI cDNA fragment of PAK4 and a 477-bp Eco47-III/BamHI cDNA fragment of PAK6.
Plasmids, Peptides, and Antibodies-2527-bp PAK4 cDNA containing the complete open reading frame was isolated from a Lung ZAP library and is identical to the publicly available GenBank TM sequence (NM 005884). Mammalian expression vectors derived from pcDNA3 (with the MCS modified to contain either an N-terminal Myc or HA epitope) were used to express various alleles of PAK4. To facilitate mutagenesis and subcloning steps we introduced a BamHI restriction site immediately 5Ј of the ATG initiation codon, a silent HindIII site within the open reading frame, and a XhoI site after the stop codon. Mutations were introduced by PCR and confirmed by sequencing of the complete cDNA fragment, which was then subcloned into the appropriate expression vector. The following oligonucleotides were used: 5Ј-CG-GGATCCATGTTT GGGAAGAGGAAGAAGCG-3Ј (BamHI site), 5Ј-AG-CTCGGGCAAGCTTGTGGCCGTCAAGAAG-3Ј and 5Ј-GACGGCCAC-AAGCTTGCCCGAGCTGCG-3Ј(HindIII site), and 5Ј-CCGCTCGAGTC-ATCTGGTGCGGTTCTGGCG-3Ј (XhoI site). The K350A,K351A mutation was introduced with the oligo 5Ј-AGCTCGGGCAAGCTTGT-GGCCGTCGCGGCCATGGACCTGCGCAAGCAGCAG-3Ј. The S474E mutation was introduced with the oligo 5Ј-CAAGGAAGTGCCGCGGA-GGAAGGAGCTGGTCGGCACGCCCTACTGGATG-3Ј, and the oligo 5Ј-CGACTTCCTCCGCGGCACTTCCTTGCTCAC-3Ј was used to introduce a SacII site to facilitate subcloning of the S474E mutant. The oligo 5Ј-GAAGATCTCCGCGCCGTCCAACTTCGAGTTCCGCGTGTTCACG-GGCTTCGAC-3Ј was used to introduce the H19L,H22L mutation. For the generation of GST fusion proteins the kinase domain of PAK4 (aa 291-591) was amplified by PCR, which introduced a 5Ј BamHI and a 3Ј XhoI site and then cloned as a BamHI/XhoI fragment pGEX-4T-1. The 5Ј BamHI site was introduced using the following oligo: 5Ј-CGGGATC-CTCACCACAGCGGGAGCCACAG-3Ј. Mammalian expression vectors: pRK5 Myc-tagged Cdc42 (L61) (gift from A. Hall); pZIPNeo Ha-Ras (L61) (67); pEXV Ha-Ras (V12) (gift from A. Hall); pRK5 human Src (Y530F) (gift from S. Courtneidge); pM chicken Src (Y527F) (gift from S. Taylor); pZIPNeo GST-onco-Dbl (68).
Mouse monoclonal antibodies (mAb) against the Myc epitope were prepared from ascitic fluid (9E10). For detection of the HA epitope the HA-7 mAb antibody (Sigma) was used. Phospho-specific anti-PAK4 polyclonal rabbit antibodies were raised against the KLH-conjugated phosphorylated peptide 682 synthesized with Ser-474 phosphorylated (CRRKpSLVGTPYWMAPE). The phospho-specific ␣PAK4 polyclonal sera was further purified on a protein A affinity column. The specificity of the protein A purified sera was determined by testing against peptides 681, 682, and 704 (see Fig. 3A). As necessary, any antibodies cross-reacting with the non-phosphorylated peptide were removed by passing the protein A purified sera through peptide 681-coupled Ultralink-Iodoacetyl column (Pierce). For Western analysis the protein A-purified phospho-specific anti-PAK4 (phospho-Ser-474) sera was used at 1/4000 dilution. For immunofluorescence experiments the protein A-purified phospho-specific anti-PAK4 antibody was pre-absorbed with peptides 681 and 704 (see "Immunofluorescence" below).
Immunoprecipitation and Kinase Assays-Cells were harvested in lysis buffer containing 20 mM Tris, pH 8.0, 137.5 mM NaCl, 1% Triton X-100, 10% glycerol, 50 mM NaF, 1 mM EGTA, 1 mM pNPP, 20 mM ␤-glycerol phosphate, 100 m Na 3 VO 4 , and protease inhibitors (4 mM Pefabloc, 1 g/ml aprotinin, 1 g/ml leupeptin, and 0.7 g/ml pepstatin). After vortexing, lysates were clarified by centrifugation, and the supernatant was incubated with anti-Myc mAb and protein G-agarose (Roche Diagnostics). After incubating lysates for 4 h at 4°C, immune complexes were purified by several low speed centrifugations followed by four washes with lysis buffer and by a final wash with kinase buffer (without ATP). In vitro kinase assays were done in kinase buffer containing 20 mM HEPES, pH 7.2, 150 mM KCl, 5 mM MgCl 2 , 1 mM NaF, 1 mM dithiothreitol, 1 mM p-nitro-phenyl phosphate, 20 mM ␤-glycerol phosphate, 100 m Na 3 VO 4 , 20 m ATP, and 0.5 l [␥-32 P]ATP (10 mCi/ml NEN-BLUNEG 502A) with various peptide substrates. Kinase reactions in a final volume of 20 l were incubated with constant shaking at 30°C for 30 min with a thermomixer (Eppendorf) and then stopped by addition of SDS-loading buffer and incubation at 100°C for 10 min. Substrate phosphorylation was then analyzed by SDS-PAGE and autoradiography.
Tissue Culture and Transfections-Low passage mouse NIH3T3 cells that had been previously established in the laboratory of T. Hunter were obtained from M. Broome. 293T cells and low passage rat intestinal epithelial (RIE1) cells were obtained from S. Courtneidge. Cell lines were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (v/v), 1 mM glutamine, penicillin 100 units/ml, and streptomycin 100 units/ml and incubated in a humidified incubator at 37°C and 5% CO 2 . Transfections were carried out using either LipofectAMINE according to manufacturer's instructions (Invitrogen) or calcium phosphate-mediated transfection, as appropriate. Cells were transfected at low density (5-10% confluence) and maintained in low serum until harvested 24 h after transfection for immunoprecipitation kinase assays.
Focus Formation and Soft Agar Assay-Low passage NIH3T3 cells especially derived for their low rate of spontaneous transformation as well as for their growth properties (growth arrest in low serum at low density) were employed in transformation assays. Typically, 14 -18 days after transfection foci were visualized by crystal violet staining. The cells were washed with PBS, then fixed for 10 min with 10% acetic acid (v/v), then stained for 10 min in 0.4% (w/v) crystal violet in 10% (v/v) ethanol, and finally extensively washed with deionized water, inverted, and dried at room temperature (53). For RIE1 cells, foci were visualized 21-24 days after transfection. For transformation assays, media was changed every 3 days. For soft agar assays, cells were plated on top of 0.6% bottom agar in growth medium and overlaid with 0.4% top agar in growth medium. Wells were supplemented with top agar in growth medium every 3-4 days. Typically, colonies growing in soft agar after 2-3 weeks were photographed with a 10ϫ objective using a standard inverted microscope. Activated allele of Ha-Ras (L61) in pZIPNeo was used in transformation and co-transformation assays with various alleles of PAK4 as indicated.
Immunofluorescence-Coverslips were placed in 24-well tissue culture trays and seeded at a density of 2 ϫ 10 4 cells/well. Cells were transfected the following day. The medium was replaced 6 h post transfection, and the cells were allowed to grow for 48 h. Coverslips were washed extensively with PBS and then fixed with cold 4% paraformaldehyde. The fixed cells were solubilized using a standard solution of 0.5% Triton X-100, washed, and incubated with primary antibodies diluted in PBS containing 1% bovine serum albumin and 5 g/ml goat IgG at 4°C overnight. For detection of phospho-Ser-474 PAK4, protein A-purified rabbit serum directed against peptide 682 (phospho-Ser-474) was blocked with peptides 681 and 704 at a concentration of 100 g/ml and desalted, and specificity for phospho-Ser-474 confirmed by Western analysis prior to use. HA-7 antibody (Sigma) was used to detect exogenous PAK4 and Myc ascites (9E10) was used to detect exogenous Cdc42L61. The coverslips were washed extensively with PBS and incubated in the dark for 2 h at room temperature with a rhodamine-conjugated goat ␣-rabbit secondary antibody (Santa Cruz Biotechnology Inc.) or with a FITC-conjugated goat ␣-mouse-FITC antibody (Santa Cruz Biotechnology Inc., Santa Cruz, CA), together with 1 g/ml Hoechst No. 33342 (bis-benzimide, Sigma). The coverslips were washed and mounted in Fluorosave (Calbiochem Corporation, La Jolla, CA) and allowed to dry overnight. Images were acquired using a Nikon oil emersion 40 ϫ 1.30 objective on a Nikon Eclipse E800 microscope with a SPOT RT camera.

RESULTS
PAK4 is Overexpressed in Human Tumor Cell Lines-To identify genes overexpressed in human cancer, we initiated an expression-based screen for several STE20 family kinases we had previously cloned. Fig. 1 shows the expression profile of PAK4 and PAK6 in normal human tissues and a panel of 60 human tumor cell lines (NCI60 panel) of various tissue origins. All filters were reprobed with actin to confirm equal loading of RNA samples (not shown).
In normal human adult tissues PAK4 was generally expressed at low levels. In contrast, the related kinase PAK6 showed very high, although restricted expression. PAK6 could only be detected in brain, testis, prostate, and placenta. We then examined the expression of PAK4 and PAK6 in the tumor cell line panel. In most of the tumor cell lines, PAK6 was expressed at low levels or was not detectable. Overexpression of PAK6 (relative to the normal tissue control) could only be detected in a low percentage of the cell lines. For example, two of the eight colon cancer cell lines exhibited high PAK6 mRNA levels, while PAK6 expression could not be detected in normal colon tissue. In contrast to PAK6, PAK4 mRNA was detected in virtually all tumor cell lines tested. 47 of the 60 tumor cell lines (78%) exhibited significantly increased PAK4 expression, relative to the normal tissue controls. We extended our analysis by examining PAK4 protein expression in matched human normal and tumor tissue biopsies. PAK4 protein expression in normal tissues varied from undetectable (heart, stomach, brain, and liver) to moderate (ovary and colon) to high (intestine).
Overexpression of a particular gene can be the result of deregulated gene expression or of gene amplification. To deter-mine whether the overexpression of PAK4 message and protein originated from an amplification of the PAK4 gene, we sought to map the chromosomal location of PAK4 using the Genebridge 4 radiation hybrid panel (see "Experimental Procedures" for details). The mapping placed PAK4 9.76 cR from D19S220 with a LOD score Ͼ3.0. Using the mapping tools available at the Genome Data base (gdb.org/gdb/) D19S220 was localized to chromosome 19 within cytogenetic bands q13.2 to q13.3. This mapping was confirmed by Celera's assembly of the human genome: the Pak4 gene resides on the Celera contig GA x2KMHMR58LA, located on chromosome 19, cytogenetic band q13.2. This region of chromosome 19 is amplified in a subset of human pancreatic, colon, and ovarian tumors (Ref. 54, and references therein). Because the gene maps to a tumor amplicon, we used Southern blot analysis to determine whether it was amplified in a subset of primary human tumor cell lines. PAK4 was amplified about 3-fold in Panc-1 (pancreatic carcinoma) and OVCAR-3 (ovarian adenocarcinoma), and about 2-fold in BxPC-3 (primary pancreatic adenocarcinoma) (data not shown).
Characterization of PAK4 Kinase Activity-To better characterize PAK4 kinase activity, we analyzed a series of mutant PAK4 alleles by in vitro kinase assays. For this purpose we examined whether a peptide derived from the PAK4 activation loop could serve as a high affinity substrate for this kinase. We synthesized a set of activation loop peptides either unphosphorylated or phosphorylated on serine 474 or threonine 478 (Fig.   FIG. 1. PAK4  2A). A highly related peptide derived from the Nck-interacting kinase kinase activation loop was used as a control to determine whether PAK4 showed any selectivity toward its own activation loop. The kinase domain of PAK4 was expressed as a GST fusion protein in bacteria, and in vitro kinase assays were performed as described under "Experimental Procedures." Fig. 2A shows that peptide 704 (unphosphorylated) is a substrate for PAK4 kinase, whereas the peptide derived from the activation loop of Nck-interacting kinase shows no detectable phosphorylation at concentrations as high as 20 M. Furthermore, phosphorylation of serine 474 abolishes entirely the phosphorylation of the peptide by PAK4. Interestingly, peptide 681, which contains phosphorylated threonine 478, shows a higher affinity to the PAK4 kinase domain than the non-phosphorylated version, peptide 704 ( Fig. 2A). For peptide 681, we have determined a K m value of 0.5 M (data not shown). Thus, the activation loop sequence of PAK4 is a high affinity in vitro substrate for the kinase, and Ser-474 may be the autophosphorylation site in vivo (Fig. 2). However, we do not yet know whether the PAK4 enhanced affinity for the peptides with a phospho-Thr in the TPY motif is of physiological relevance. The TPY motif is conserved in the activation loop of other STE20 group members and could be a common site for positive or negative regulation of STE20s by other kinase(s).
Using this high affinity substrate, we characterized different mutants of PAK4. An activating allele of PAK4 was made by mutating serine 474 to glutamic acid. To create a kinase-inactive allele, we mutated the conserved lysine residue in the ATP binding region. In addition to the expected lysine at amino acid position 350 in domain II, PAK4 has an additional lysine at position 351; both lysines were mutated to alanine (K350A,K351A mutant). In vitro kinase assays performed on PAK4 and PAK4 mutants, after transient expression in 293T cells, are shown in Fig. 2B (left panel). PAK4 (WT) shows weak kinase activity while the PAK4 (K350A,K351A) mutant had no detectable activity toward either the high affinity peptide substrate or myelin basic protein ( Fig. 3 and data not shown). The isolated catalytic domain exhibits very high kinase activity in vitro, which may be attributed to the removal of an autoinhibitory domain. PAK4 S474E shows increased catalytic activity in vitro, thus confirming the initial finding that Ser-474 is most likely a regulatory autophosphorylation site ( Fig. 2A, left  panel).
In addition to the autophosphorylation site at Ser-474, PAK4 also appears to autophosphorylate on site(s) within the N ter-  minus, though the significance of this is unknown (data not shown). PAK4 291-591 that lacks the N-terminal site(s), fails to show autophosphorylation by in vitro immunoprecipitation kinase assays, while the PAK4 S474E that has the N-terminal site(s) still has detectable autophosphorylation ( Fig. 2A). We believe this is due to the fact that PAK4 (aa 291-591), lacking its auto-inhibitory domain, is stoichemetrically autophosphorylated on Ser-474 when expressed in tissue culture cells (data not shown, and see Fig. 3B). Immunoblot analysis of whole cell lysates revealed comparable expression of all mutants, except PAK4 (K350A,K351A), which was expressed at lower level (Fig.  2B, right panel).
Localization of Active PAK4 to the Golgi Membrane-To further study PAK4 activity and regulation, we developed phospho-specific antibodies directed against serine 474 of PAK4. Rabbits were immunized with peptide 682, and the resulting serum was affinity-purified. To demonstrate the specificity of the purified antiserum, we tested the reactivity against several peptides corresponding to the PAK4 activation loop (Fig. 3A). Our results demonstrate that the antiserum is highly selective for peptide 682, in which serine 474 is phosphorylated. The two control peptides were detected only in dot blots at very high peptide concentrations. To demonstrate that phosphorylation of serine 474 correlates with increased PAK4 kinase activity, we transiently expressed wild type PAK4 and the PAK4 kinase domain (aa 291-591), which is constitutively active (see Fig. 2), and performed immunoblot analysis with the phospho-specific antiserum (Fig. 3B). While both constructs were expressed to comparable levels (Fig. 3, left panel), only PAK4 (aa 292-591) was detected with the phospho-specific antiserum (Fig. 3, right  panel). Together with the data presented in Fig. 2, these results demonstrate that phosphorylation of serine 474 correlates with PAK4 kinase activity.
Previous studies have shown that PAK4 localizes to the Golgi membrane upon co-expression with activated Cdc42 (7). It is not known, however, whether interaction with Cdc42 is able to activate the kinase activity of Golgi membrane-localized PAK4.
In vitro experiments failed to show any activation of PAK4 by activated Cdc42 (Ref. 7 and data not shown). To address this question, we performed immunofluorescence studies using the phospho-specific anti-PAK4 antiserum to detect activated PAK4. We expressed wild type HA-tagged PAK4 alone or together with Myc-tagged activated Cdc42 in NIH3T3 cells and analyzed PAK4 localization by immunostaining with ␣HA mAb (Fig. 4A). Consistent with previously published data (7), ectopic expression of PAK4 alone results in cytoplasmic staining, while in the presence of activated Cdc42, PAK4 relocalizes to the Golgi membrane (Fig. 4, A and B). Interestingly, when coexpressed with activated Cdc42, the relocalized PAK4 is detectable by the phospho-specific anti-PAK4 antibody. In contrast, PAK4 alone or kinase inactive PAK4 co-expressed with activated Cdc42 were not detected with the phospho-specific antibody. (Fig. 4, A and B).
Effect of PAK4 on Cell Transformation-Given the up-regulation of PAK4 expression in a wide range of human tumor lines, we were interested in determining what role, if any, PAK4 plays in mammalian cell transformation. To examine this question, we co-expressed different alleles of PAK4 in NIH3T3 mouse fibroblasts and rat intestinal epithelial cells (RIE-1) together with activated Ha-Ras (L61). As a control, cells were transfected with the empty vector. Two to 3 weeks after transfection, the cells were stained and analyzed for focus formation (Fig. 5). While overexpression of the WT kinase did not enhance Ras-dependent transformation, kinase-inactive PAK4 strongly inhibited the ability of activated Ras to induce foci. In similar experiments, we tested the effect of wild type and kinase inactive PAK4 on cell transformation by other oncogenes (for example activated Src kinase), but did not observe a significant effect (data not shown). Our findings suggest that kinase inactive PAK4 specifically interferes with NIH3T3 For parts A and B, Cdc42 was visualized with primary mouse ␣-Myc antibody and ␣-mouse FITC-coupled secondary antibody; activated PAK4 was visualized with a rabbit primary ␣-phospho-Ser-474 antibody with ␣-rabbit rhodamine-coupled secondary antibody; and PAK4 was visualized with a primary ␣-HA mouse antibody and ␣-mouse FITC-coupled secondary antibody. transformation by oncogenic Ras. Furthermore, this finding is not limited to fibroblasts, as kinase-dead PAK4 (K350A,K351A) also inhibited Ras transformation of rat intestinal epithelial cells (Fig.  5B). Interestingly, though the kinase inactive PAK4 can interfere with both Ras-and Dbl-driven transformation of NIH3T3 cells, it is not able to interfere with Src-driven transformation (Fig. 5C).
We next tested whether activated alleles of PAK4 could confer anchorage-independent growth to NIH3T3 cells. We established stable 3T3 cells expressing PAK4 WT, S474E, or empty vector control. Different clones were then plated in soft agar and scored for colony formation after 3 weeks. As shown in Fig. 6, cells overexpressing wild type PAK4 and the vector control did not form colonies in soft agar, while the activated mutant PAK4 (S474E) alone was sufficient to confer anchorage-independent growth to NIH3T3 cells (Fig. 6). Interestingly, . Activated PAK4 is able to confer anchorage-independent growth to NIH3T3 cells. Main adaptation to standard protocol was the use of high efficiency transfection, followed by the brief selection for pools of transfected cells that were used in soft agar assay. In parallel, expression of various alleles of PAK4 was confirmed by Western analysis (not shown). Colonies growing in soft agar were photographed with an inverted phase contrast microscope.
foci were visualized by staining with crystal violet. In parallel the expression of the various genes was followed by Western blot analysis (not shown). Assays were optimized by varying the ratio of the PAK4 and Ras expression vectors. Total DNA transfected was kept constant with the empty vector. Shown is a typical result using a ratio of 5:1 of PAK4 expression vector to that of the Ras (L61) expression vector (per 100-mm dish of cells, 0.5 g of Ras (L61) was co-transfected with either 2.5 g of pcDNA derived vector or 2.5 g of the indicated PAK4 expression vectors). Similar results were also obtained with the Ras (V12) expression vector. B, RIE1 cells were grown and transfected under standard conditions. Shown is a typical experiment using a 5:1 ratio of PAK4 expression vector to the Ras (L61) expression vector. 25 days after transfection, transformed foci were visualized by staining with crystal violet. C, the ability of kinase-inactive PAK4 to interfere with the transformation of NIH3T3 cells. Cells were co-transfected with each indicated expression vector: 0.5 g of Ras(L61), 1.0 g of c-Src (Y527F), 2.0 g of h-Src (Y530F), and 2.0 g of GST-onco-Dbl and chart normalized to number of colonies each gave alone in co-transformation assays with vector (arbitrarily set at 100%) and compared with co-transformation with the indicated expression vector for PAK4 (same ratio of expression vectors as in part A). Shown are representative results from a typical experiment, percent of foci relative to vector is relative to each of the tested oncogenes (e.g. each set of co-transformation results with PAK4(WT) and PAK4 (K350A,K351A) are relative to either Ha-Ras (L61), hSrc(Y530F), cSrc(Y527F), or gst-onco-Dbl plus vector arbitrarily set at 100%); vector(s) alone or in combination with either PAK4(WT) or PAK4(K350A,K351A) did not give any foci. Typically 14 to 18 days after transfection, transformed foci were visualized by staining with crystal violet and scored (GST-onco-Dbl and h-Src (Y530F) foci were visualized after 22-24 days). expression of activated PAK4 had only modest effects on the growth rate of NIH3T3 cells (data not shown). Taken together, the ability of activated PAK4 to induce anchorage-independent growth and the ability of kinase inactive PAK4 to interfere with Ras-driven focus formation suggest that PAK4 plays a key role in Ras-driven cellular transformation.
To explore this possibility further, we sought to determine whether PAK4 could modulate the growth properties of a Rasdependent tumor line. For this purpose, we established stable lines of HCT116 human colon carcinoma cells expressing different alleles of PAK4. HCT116 cells harbor an activating mutation of K-Ras (58). HCT116 cells were transfected with vector control, wild type PAK4, or the kinase-inactive PAK4 mutant (K350A,K351A). After selection, the cells were analyzed for anchorage-independent growth in soft agar. As shown in Fig. 7, expression of kinase-inactive PAK4 (K350A,K351A) almost completely abolished colony formation of HCT116 cells in soft agar.

DISCUSSION
Deregulation of cellular signaling pathways is a hallmark of neoplastic growth and is frequently caused by activation of proto-oncogenes either by mutation and/or gene amplification. Activation of various protein kinases, particularly members of the protein tyrosine kinase family such as EGFR, MET, PDGFR, Kit, Abl and Src (see reference Ref. 63 for review), are frequently associated with cancer. Until recently, however, only a few members of the serine/threonine kinases have been shown to have oncogenic potential and to be overexpressed in cancer. These include such serine/threonine kinases as Raf, Akt, and Aurora2 (52,64,65).
Here we demonstrate that the serine/threonine kinase PAK4 is overexpressed in a broad range of human tumor cell lines. This increased PAK4 expression in transformed cells could be the result of deregulated gene expression and/or gene amplification. We tested the latter possibility by determining the chromosomal location of the PAK4 gene, mapping it to a region on chromosome 19 that is amplified in a subset of human tumors (cytogenetic bands q13.2 to q13.3). 2 In fact, we detected amplification of PAK4 in two tumor cell lines. The region 19q13.1-q13.4 contains a cluster of genes whose expression is elevated in colon and pancreatic cancers: CEA (carcinoembryonic antigen) at 19q13.3; SAP-1, a stomach cancer-associated PTP at 19q13.4; and AKT2 at 19q13.3 (54 -57). Several studies have suggested that, in addition to AKT2, other unknown amplified gene(s) in the region 19q13.1-q13.2 may play a role in pancreatic and ovarian carcinogenesis (54,56,57). The results of our expression analysis together with the mapping data suggest that PAK4 may be one these genes.
Rho family GTPases such as Rho, Rac, and Cdc42 play an important role in mediating many of the effects of Ras on cellular proliferation, morphology, and transformation (35)(36)(37)(38)(39)45). Various members of the PAK family of protein kinases have been implicated in the signaling pathways of Rho family GTPases (15, 34, 47-48, 51, 59 -61). It was recently demonstrated that PAK4 is a specific downstream effector for Cdc42 and that it mediates the Cdc42 effect on the actin cytoskeleton (7). Although PAK4 has a PBD domain, its N-terminal regulatory region is fairly divergent from PAKs 1-3. For example, PAK4 does not contain a recognizable inhibitory switch domain, which is involved in negative regulation of PAKs 1-3 (7,9). Deletion of the PAK4 N-terminal regulatory domain (aa 291-591), mutation of its PBD (H19L,H22L), and mutation of its activation loop (S474E) all lead to derepression of the PAK4 catalytic domain to a similar extent as has been shown for the other PAKs (Fig. 2B) (for discussion, see Ref. 10). Thus, PAK4 may be regulated by a mechanism similar to PAKs 1, 2, and 3.
Consistent with previously published data (7), we show that PAK4 is localized to the Golgi in cells co-expressing activated Cdc42 (Fig. 4A). However, while PAK4 specifically interacts with GTP-bound Cdc42, no effect of activated Cdc42 on PAK4 activity was detected in in vitro kinase assays (7). Using immunofluorescence with a phospho-specific anti-PAK4 antibody, we clearly show that Golgi-localized PAK4 is autophosphorylated in the activation loop in the presence of activated Cdc42 (Fig. 4B). In the absence of Cdc42, PAK4 shows a diffuse cytosolic staining pattern and is not detected by the phosphospecific anti-PAK4 antibody. Together with our finding that phosphorylation of Ser-474 correlates with PAK4 catalytic activity (see Fig. 3B), these data suggest that Cdc42 activates PAK4 in vivo and that activated PAK4 localizes to the Golgi. We do not yet know, however, if interaction with other Golgi components is required for PAK4 activation We also demonstrate that PAK4 plays an important role in Ras-driven, anchorage independent growth. Expression of kinase-dead PAK4 is able to interfere with Ras-driven transformation of rodent fibroblast and epithelial cells, and an activated mutant of PAK4 (PAK4 S474E) is sufficient to confer anchorage-independent growth to NIH3T3 cells (Figs. 5 and 6). Kinase-dead PAK4 also interferes with anchorage-independent growth of a human tumor line, HCT116, which is driven by an activating mutation in K-Ras (Fig. 7). These results confirm and extend recently published data (50,51) and further implicate PAK4 in Ras transformation and anchorage-independent growth cells.
We also sought to address the broader questions of whether PAK4 selectively inhibited Ras-dependent transformation and whether this was mediated via signaling through Rho family GTPases. Cdc42 and other Rho family GTPases are activated by various members of a broad family guanine nucleotide exchange factors whose hallmark are their tandem Dbl and pleckstrin homology domains (66). Activated alleles of a large num-2 D. Whyte, unpublished observation. FIG. 7. Kinase-dead PAK4 is able to interfere with anchorageindependent growth of a Ras-driven human colon cancer cell line, HCT116. Stable cell lines expressing wild type PAK4 or kinaseinactive Pak4 were assayed for growth in soft agar. A HCT116-stable line that had been transfected with the empty vector was used as a control. Similar to the vector control, HCT116 cell lines expressing wild type PAK4 readily formed colonies in soft agar. In contrast, two independent HCT116 clones expressing kinase-inactive PAK4 were unable to form colonies in soft agar (bottom panels).
ber of these guanine nucleotide exchange factors have been shown to be transforming (66,69). Interestingly, though the kinase-inactive PAK4 is able to interfere with Ras-and Dbldriven transformation of NIH3T3 cells, kinase-inactive PAK4 was not able to interfere with Src-driven transformation. Recently, Qu et al. (51) reported that a kinase-dead allele of PAK4 could interfere with Dbl-mediated cell transformation although they did not observe an inhibitory effect on Ras. The inactive mutant of PAK4 used in their study was generated by changing lysine 350 to methionine (PAK4 K350M). In contrast, we mutated both lysines in subdomain II of the PAK4 kinase to alanine in order to generate an inactive PAK4 allele (PAK4 K350A,K351A). It is possible that mutation of lysine 350 alone is not sufficient to completely abolish PAK4 kinase activity in vivo (see Ref. 62 for discussion of residual kinase activity of kinases with catalytic lysine mutations). Alternatively, the differing results with regard to inhibition of Ras transformation may merely be due to different transfection conditions, expression vectors, and/or cell lines.
Similar to our findings with PAK4, dominant negative alleles of PAK1 and PAK2 were reported to interfere with Ras transformation, and Pak1 was also implicated in anchorage-independent growth and invasiveness of breast cancer cells (41, 42, 44, 46 -48). Thus, while distinct in their N-terminal and kinase domains, members of both PAK subfamilies have similar effects on Ras transformation. We do not yet know, however, if dominant negative versions of PAK1/PAK2 and PAK4 interfere with the same Ras-dependent signaling pathways or if members of the two PAK subfamilies regulate distinct aspects of Ras function. Importantly, our finding that dominant negative PAK4 does not block transformation by other oncogenes, such as Src, indicates that the effect is most likely due to inhibition of one or more Ras-dependent signaling pathways.
To date, the physiological function of PAK4 and its downstream effectors has not been completely elucidated. Recent reports have implicated PAK4 in the regulation of cell survival. For example, cells overexpressing PAK4 are resistant to stressinduced apoptosis, and PAK4 can phosphorylate the apoptotic regulator BAD in vitro (50). In tumor cells, including Rastransformed cells, the Akt survival pathway is often deregulated. Therefore, we conducted preliminary experiments to explore the possible effect of PAK4 on Akt. Similar to the results reported by Gnesutta et al. (50), we found that BAD is a substrate for PAK4 in vitro. However, we also found that overexpression of PAK4 in tissue culture cells can result in an increase in phosphorylation of Akt on Ser-473 (data not shown). Thus, we are not certain yet whether PAK4 ability to regulate BAD is only direct and could in part be also due to activation of the upstream kinase Akt. We are currently investigating this possibility. It will be of great interest to determine whether the observed effects of PAK4 on cell transformation are, at least in part, dependent on modulation of the Akt survival pathway.
In summary, our data strongly implicate PAK4 in tumorigenesis and suggest that PAK4 might be particularly important for Ras transformation. Elucidation of the signaling pathways and the mechanisms by which PAK4 and other PAK family members influence Ras signaling will further contribute to our understanding of cell transformation and the development of human cancer.