Molecular cloning of a new member of the p21-Cdc42/Rac-activated kinase (PAK) family.

A number of "target" proteins for the Rho family of small GTP-binding proteins have now been identified, including the protein kinases ACK and p65PAK (Manser, E., Leung, T., Salihuddin, H., Zhao, Z.-S., and Lim, L. (1994) Nature 367, 40-46). The purified serine/threonine kinase p65PAK has been shown to be directly activated by GTP-Rac1 or GTP-Cdc42. Here we report the cDNA sequence encoding a new brain-enriched PAK isoform beta-PAK, which shares 79% amino acid identity with the previously described alpha-isoform. Their mRNAs are differentially expressed in the brain, with alpha-PAK mRNA being particularly abundant in motor-associated regions. In vitro translation products of the alpha- and beta-PAK cDNAs exhibited relative molecular masses of 68,000 and 65,000, respectively, by SDS-polyacrylamide analysis. A specific beta-PAK peptide sequence was obtained from rat brain-purified p65PAK. Recombinant alpha- and beta-PAKs exhibited an increase in kinase activity mediated by GTP-p21 induced autophosphorylation. Cdc42 was a more potent activator in vitro of alpha-PAK kinase, and the fully activated enzyme is 300 times more active than the unphosphorylated form. Interestingly the down-regulation in the binding of p21s to recombinant beta-PAK and brain p65PAK, which is observed upon kinase activation does not occur with recombinant alpha-PAK.

Morphological roles for the most common members of the mammalian Rho family of small GTP-binding proteins, Rac1, RhoA, and Cdc42, have been established in fibroblasts Kozma et al., 1995). Cdc42 in Saccharomyces cerevisiae is required for cell budding and may provide the polarization signal at this site (Ziman et al., 1993); in fibroblasts filopodial formation is dependent on the closely related mammalian homologue of Cdc42 Nobes and Hall, 1995). Although an increasing number of p21 Rho GTPase-activating proteins (GAPs) 1 have been iden-tified (for review, see Lamarche and Hall (1994)), there is as yet no evidence that they are able to exhibit effector function. These proteins can be identified by sequence homology to the GAP domain and by their activity in overlay assays (Manser et al., 1992). The most closely related RhoGAPs comprise the chimaerin family Leung et al., 1993Leung et al., , 1994, acting on Rac and whose activity is regulated through a protein kinase C-like cysteine-rich domain (Ahmed et al., 1993). Although many RhoGAPs are somewhat promiscuous in vitro, they appear to show distinct p21 specificities in vivo (Ridley et al., 1993).
The prototype small GTP-binding protein p21-Ras is an oncogene that has effector targets which include Raf kinases (Vojtek et al., 1993;Warne et al., 1993;Zhang et al., 1993) phosphatidylinositol 3-kinase (Rodriguez-Viciana et al., 1994) and RasGAP itself (Schweighoffer et al., 1992). The use of the p21 GTP/GDP cycle is exemplified by the role of Ras in growth factor signal transduction, where GTP-Ras functions to activate proteins of the "mitogen-activated protein kinase cascade" through the serine/threonine kinase Raf (Warne et al., 1993), and MEK kinase (Lange-Carter and Johnson, 1994). It seems probable that part of p21 Rho family signaling also occurs through associated kinases for which the prototypes are the activated Cdc42-associated tyrosine kinase p120-ACK  and a Cdc42-and Rac1-activated kinase p65-PAK . Both ACK and PAK inhibit intrinsic as well as GAP-stimulated GTPase activity of the p21s. PAK belongs to a family of kinases that includes the S. cerevisiae STE20 gene product (Leberer et al., 1992;Ramer and Davis, 1993) which acts upstream of the pheromone response mitogenactivated protein kinase cascade (Ammerer, 1994). Two other related S. cerevisiae kinases Cla4p 2 and a putative gene product present in the yeast genome we designate as Sc-PAK show homology to PAK in their putative kinase and Cdc42-binding domains.
The use of a [␥-32 P]GTP-p21 overlay technique has allowed us to identify at least eight mammalian candidate target proteins for Rac1, Cdc42, and RhoA . The brain-enriched p65 PAK co-purified with a number of kinases of similar size also identified in [␥-32 P]GTP-Cdc42 overlays. A human PAK designated hPAK65 has been reported to be ubiquitously expressed (Martin et al., 1995) and probably represents the human homologue of the ubiquitous rat p62 Cdc42/ Rac1 binding protein. Thus although PAK kinases are most abundant in the brain, they appear to be a common target for Cdc42 and Rac "molecular switches" in all mammalian cells. Two mammalian Cdc42 isoforms have been identified * This work was supported by the Glaxo-Singapore Research Fund. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank TM /EMBL Data Bank with accession number(s) U33314.
Here we describe a novel PAK cDNA (designated ␤-PAK) encoding a protein which is closely related to our previously published sequence (now termed ␣-PAK) and also to the hPAK65 cDNA (Martin et al., 1995). The putative protein products exhibit remarkable conservation of amino acid residues in their kinase and p21-binding domains. Despite this similarity it has been possible to establish the relationship between the ␣and ␤-cDNAs and PAK species found in the brain based on differences in their biochemical properties. In vitro translated ␣and ␤-PAK exhibit relative molecular masses of 68,000 and 65,000 Da, respectively. A peptide sequence derived from purified p65 PAK has been found to be specific for the ␤-PAK isoform. We show that, in a manner similar to purified p65 PAK , binding of Cdc42 to recombinant ␤-PAK kinase, but not to ␣-PAK, is down-regulated upon kinase activation.

MATERIAL & METHODS
Isolation of Rat Brain ␤-PAK cDNAs-A subclone containing bovine PAK cDNA derived from polymerase chain reaction (PCR) with degenerate primers and encoding part of the kinase domain  was used as a probe to screen a rat brain cortex cDNA library (ZAP, Stratagene). Two filters (Hybond N (Amersham Corp.), 20 ϫ 20 cm) containing DNA from 4 ϫ 10 4 bacteriophage plaques were hybridized overnight in 0.5 M sodium phosphate (pH 6.8), 7% SDS, 10% formamide at 55°C and washed in 1 ϫ SSC at 60°C. Positive plaques (ϳ30) were taken through further rounds of purification and excised in vivo as plasmids (in pBluescript SK) according to the manufacturer's protocol. Three similar clones were identified whose restriction maps did not correspond to ␣-PAK. One of these clones contained the complete coding sequence of a predicted 61-kDa protein (␤-PAK). The region from this clone shown in Fig. 1 was sequenced completely on both strands using restriction subclones and specific oligonucleotide primers. Because of the presence of nonconservative residues at ␣-H73, C92, and N258 in our published sequence , we resequenced these regions of the ␣-PAK cDNA with oligonucleotide primers to resolve apparent sequencing ambiguities. The cDNA was confirmed to encode proline at position 73, valine at position 92, and serine at position 258 (Fig. 2); the GenBank entry has been amended appropriately.
Northern Analysis-Total RNA was prepared from fresh rat tissues as described previously . RNA (20 g) was separated on 1% formaldehyde agarose gels and transferred to Hybond N nylon membranes (Amersham). A 876-base pair HindIII restriction fragment from the 3Ј end of the ␤-PAK cDNA (nucleotides 1371-2247) and a 750-base pair fragment of ␣-PAK cDNA (nucleotides 500-1250) were labeled with [␥-32 P]dCTP by random priming (Amersham). Filters were stained with methylene blue to assess equal loading of the lanes, then hybridized overnight at 50°C in 0.5 M sodium phosphate (pH 6.8), 7% SDS, 10% formamide and washed in 1 ϫ SSC at 55°C.
In Situ mRNA Hybridization and PAK Immunocytochemistry-For in situ hybridization oligonucleotides complementary to ␣-PAK nucleotides 1508 -1547 and ␤-PAK nucleotides 1355-1394 were labeled with [ 33 P]-or [ 35 S]dATP (Du Pont NEN) and hybridized to tissue sections . Polyclonal rabbit antiserum raised to ␣-PAK residues 1-251 was applied to sections at 1:500 dilution overnight and developed with horseradish peroxidase-anti rabbit antibodies/3,3Ј-diaminobenzidine according to standard protocols.
Expression and Purification of Recombinant Proteins-Recombinant glutathione S-transferase (GST)/␣-PAK was expressed from pGEX-2T plasmid (BamHI/EcoRI cut) containing PAK cDNA sequence that includes the initiator methionine. At the 5Ј end (N terminus) a PCRengineered BamHI linker sequence (5Ј GGA TCC ACA ATG-3) was inserted, and at the 3Ј end an EcoRI cloning site was used. GST/␣-PAK 1-251 was subsequently created by removing the cDNA coding sequence from the internal BglII site to the 3Ј EcoRI site and ligating the flushed ends. Recombinant ␤-PAK was expressed from pGEX-4T-2 (SmaI/XhoI cut) containing cDNA sequence having at the 5Ј end a PCR-engineered SmaI linker sequence (CCC GGG AAA ATG-3) and at the 3Ј utilizing the SK-plasmid-derived XhoI site. PCR-derived sequences were in all cases checked by sequencing. These constructs were introduced into the BL21 Escherichia coli strain for protein production. Cells grown to A 600 (1 cm) of 0.6 in 50 g/ml ampicillin were induced for 4 -6 h at 25°C in the presence of 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside. Fusion proteins of recombinant kinases and p21s were purified on glutathione-Sepharose (Pharmacia-LKB) as described previously (Manser et al., 1992).
Western Blots and Immunoprecipitation-Filters were blocked for 1 h in 3% skimmed milk prior to incubation with primary anti-PAK antibodies (1:500) in phosphate-buffered saline, 1% bovine serum albumin, 0.1% Triton X-100 for 2 h at room temperature. Filters were incubated with 1:4000 dilution of horseradish peroxidase-coupled second antibodies (DAKO) for 1 h. Bands were visualized with hyperfilm in the presence of luminol (Amersham).
For immunoprecipitation the rabbit reticulocyte lysate was diluted to 4 mg/ml in tissue extraction buffer. Extracts (100 l) were incubated with 10 l of affinity-purified antibody for 1 h at 4°C collected on 50 l of protein A-Sepharose (Sigma), washed with 200 l of extraction buffer, then 400 l of phosphate-buffered saline ϩ 1% Triton X-100 and eluted in SDS sample buffer.

Isolation of cDNAs Encoding a New p21-activated Kinase-
Overlays of SDS-polyacrylamide electrophoresed tissue extracts with [␥-32 P]GTP-labeled Cdc42 have indicated that there are a number of different proteins of M r 62,000 -68,000 with similar p21 binding characteristics as the purified p65-PAK . Of these, the p62 protein is apparently present at similar concentrations in all tissues, whereas the larger species are more highly expressed in the central nervous system. A rat brain cortex cDNA library was screened with the 400-base pair bovine PCR-generated cDNA corresponding to FIG. 2. Alignment of PAK-related protein kinases. Predicted amino acid sequences of mammalian ␣-PAK , ␤-PAK (Fig. 1), and hPAK65 (Martin et al., 1995), and the S. cerevisiae sequences of Ste20p (GenBank accession number M94719), Cla4p (GenBank accession number X82499) and a putative open reading frame encoding a related kinase present in the yeast genomic DNA (GenBank accession number Z48149) we designate as Sc-PAK, previously assigned as a 36-kDa kinase (GenBank accession number X69322) were aligned using the clustal method (DNAStar).
FIG. 3. Northern analysis of PAK mRNAs. 32 P-Labeled cDNAs derived from the more divergent regions of ␣and ␤-PAK (see "Materials and Methods") were used to probe Northern blots containing 20 g/lane of total RNA from rat tissues. The ␣and ␤-PAK mRNAs were sized relative to RNA markers (Life Technologies, Inc.). The lanes are marked as follows: Th, thymus; Sp, spleen; Lu, lung; T, testis; Br, brain; Ki, kidney; Li, liver; H, heart.
bovine ␣-PAK kinase domain sequence. In addition to the clone reported previously , three clones from a total of 30 "positives" in this screen were found by restriction mapping to consist of overlapping sequences derived from mRNA of a related kinase. One encompassed the entire coding region of this kinase, and a region of 2247 nucleotides was sequenced on both strands (Fig. 1). The entire open reading frame of 544 amino acids, from nucleotide 61 to 1693, shows a FIG. 4. In situ localization of ␣and ␤-PAK mRNAs. Progressive rostral to caudal coronal sections of adult rat brain (A-E) were hybridized with 33 P-labeled ␣or ␤-PAK-specific oligonucleotides (left-and right-hand panels, respectively). A, note high expression of ␣-PAK mRNA in cerebral cortex (Cx) and piriform cortex (Pir); ␤-PAK mRNA in medial preoptic nucleus (MPO) and piriform cortex. B, in the thalamus, ␣-PAK mRNA was highly expressed in certain subdivisions e.g. lateral dorsal (LD) and ventral (V) thalamic nuclei. ␣-PAK mRNA was high in the CA1 pyramidal cell layer of the hippocampal formation (CA1), low in dentate gyrus (dg), where ␤-PAK mRNA levels were high. The medial nucleus of the amygdala (Me) and the ventromedial nucleus of the hypothalamus (VMH) showed enhanced ␤-PAK mRNA expression. C, in midbrain, highest expression of ␣-PAK mRNA was in the subiculum (S) and of ␤-PAK mRNA in the dorsal raphe nucleus (DR). D, the cerebellum (Cb), pontine nucleus (Pn), and reticulotegmental nucleus (RtTg) of the pons showed high levels of ␣-PAK mRNA but not of ␤-PAK mRNA. E, in the medulla ␣-PAK mRNA was high in the lateral reticular nucleus (LRt). F, in situ hybridization to rat embryo sections; ␤-PAK mRNA (13.5 day embryo) was high in neural structures including brain (Br), spinal cord (SCrd), dorsal root ganglia (DRG), and olfactory epithelium (Olf). Similarly ␣-PAK mRNA (20 day embryo) was high in the brain, spinal cord, and dorsal root ganglia. Magnification: bar equals 2.4 mm (A-E) and 0.45 mm (F). striking similarity to the rat PAK we described previously and hPAK65 (Martin et al., 1995) as shown in Fig. 2. We term the related kinases we have isolated ␣and ␤-PAK. The ␣and ␤-PAKs show 74% identity at the DNA level and 79% identity at the amino acid level. The hPAK65 amino acid sequence is equally related to ␣and ␤-PAKs and is almost identical to a rat isoform we have designated as ␥-PAK. 3 All three mammalian PAKs have conserved p21-binding and kinase domains which they share with the three yeast PAK-like kinases (Fig. 2). The mammalian PAKs also show conservation outside of these regions; the greatest sequence divergence between them is in the "linker" sequence of ϳ100 amino acids that lies between the p21-binding and kinase domains. The N-terminal hPAK65 sequence appears truncated relative to the rat ␣and ␤-isoforms ( Fig. 2; see "Discussion").
Expression Pattern of ␣and ␤-PAK mRNAs-Labeled divergent cDNA sequences of ␣and ␤-PAK were used to probe Northern blots containing mRNA isolated from various rat tissues (Fig. 3). The 4-kilobase ␣-PAK mRNA was expressed predominantly in the brain and at lower levels in the spleen. Surprisingly ␤-PAK (with a 1.7-kilobase open reading frame) appears to be translated from a 9-kilobase mRNA only detected in brain and at low level in testis (the relative weakness of the ␤-PAK signal was consistent with in situ hybridization results).
We performed in situ hybridization with specific oligonucleotides to determine the regional expression pattern of the two kinases (Fig. 4). ␣-PAK mRNA was more abundant than ␤-PAK mRNA, and exposure times were chosen to determine brain regional distribution of ␣and ␤-mRNAs, rather than to compare their absolute levels. Equivalent sections hybridized to ␣and ␤-specific probes are shown side by side (Fig. 4). In both cases sense oligonucleotide probes were used to determine "background" signals (data not shown).
␣-PAK mRNA was expressed in the cortex with highest levels of hybridization over cell layers IV and V, in limbic regions of the cortex, and in piriform cortex (Fig. 4A). ␤-PAK mRNA was also relatively high in piriform cortex while in the cortex enriched in layers II-III and V. Both mRNAs were expressed in the hippocampal formation (Fig. 4B) in both the CA1 and dentate gyrus. ␣-PAK mRNA was highly expressed in subiculum (Fig. 4C) and showed some of its highest expression in the ventral tier nuclei of the thalamus. Here ␤-PAK mRNA exhibited only low expression, but showed greater enrichment in the hypothalamus; in medial preoptic (Fig. 4A), ventromedial and arcuate nuclei (Fig. 4B). Relatively high levels of ␤-PAK mRNA were present in the monoaminergic dorsal raphe nucleus (Fig. 4C) and locus coeruleus. Interestingly, ␣-PAK mRNA was highly expressed in a number of neuronal groups associated with motor function, including the pontine nucleus, reticulotegmental, external cuneate, and lateral reticular nuclei, which send mossy fiber input to the cerebellum (Fig. 4, D and E). ␣-PAK mRNA was also highly expressed in scattered neurons in the pontine and medullary reticular formation, and in patches of cells comprising the linear nucleus of the medulla, with moderate expression in the inferior olivary nucleus which provides climbing fiber input to the cerebellum (see Fig. 5 for detailed comparisons with protein expression). Both ␣and ␤-PAK mRNAs were highly enriched in the embryonic central nervous system (Fig. 4F) with relatively little expression elsewhere, confirming the specificity of the probes.
Cellular Localization of ␣-PAK Expression in the Central Nervous System-Within the nervous system we sought to establish which type of cells might be responsible for the strong expression of ␣-PAK. staining). Within large neurons of the pontine reticular formation both the cell bodies and processes were strongly immunoreactive. In the cerebellum, there were high levels of expression of ␣-PAK mRNA (but not of ␤-PAK mRNA), which at higher magnification could be localized to granule cells, basket cells, and deep cerebellar nuclei (data not shown).
The ␤-PAK cDNA Encodes p65 PAK -PAK antibody immunoprecipitated both ␣and ␤-PAK in vitro translated proteins derived from the two cDNAs (utilizing the endogenous ATG start codons Fig. 6B). The relative masses of the translation products were 68,000 and 65,000 Da, respectively. This is consistent with the presence of three distinct bands in brain extracts, representing the different PAK isoforms, as detected by Cdc42 overlay (Fig. 6A). Heat treatment of the in vitro products in SDS buffer to release the immunoprecipitates caused some streaking but did not alter the relative mobility of the kinases compared with the unheated proteins (Fig. 6B, first two lanes). The mobility of the in vitro translated proteins suggests that ␣-PAK corresponds to the p68 Cdc42-binding protein detected in brain extracts , and ␤-PAK to a brain protein that elutes last from the Cdc42 affinity column during protein purification (the previously described p65 PAK ).
Tryptic peptides obtained previously from within the kinase domain of affinity-purified bovine PAK corresponded to regions of ␣and ␤-PAK which are identical in sequence. However confirmatory evidence for the presence of ␤-PAK in affinitypurified rat brain PAK (Fig. 6C, lane 3) was obtained from a cyanogen bromide digest of the protein. The peptides were subjected to SDS-PAGE, transferred to PVDF membrane, and overlaid with [␥-32 P]GTP-Cdc42. Lanes 1 and 2, with 45 and 5 g of digested protein(s), contains a Cdc42-binding polypeptide with apparent mass of 34 kDa which was subjected to Nterminal amino acid sequencing. The obtained sequence (M)APEEXNXXAxLXXIFPGGG was not informative at every position but corresponds to the predicted ␤-PAK product fol-FIG. 6. The ␤-PAK cDNA encodes p65 PAK . A, [␥-32 P]GTP-Cdc42 overlay of rat brain extract (at concentrations indicated, lanes 1 and 2) predominantly detects three protein bands, the central band appears to migrate at the same position as purified rat p65 PAK (lanes 3 and 4). B, supercoiled cDNA (1 g, pBluescript SKϪ) plasmid was used to generate [ 35 S]methionine-labeled ␣and ␤-PAK proteins by coupled in vitro transcription/translation with T3 polymerase (50 l, Promega; C is control luciferase). Lanes 1-3 show 10 l of total translation mix resolved on a 9% SDS-polyacrylamide gel. Antibodies (ϳ4 g) purified with recombinant ␣-PAK 1-251 were then used to immunoprecipitate 40 l of each translation mix (IP ϩ, lanes 4, 6, 8). These were run after heat treatment in SDS sample buffer to disrupt complexes. As controls 10 l of translation mix (lanes 5 and 7) were heated to 90°C for 3 min in SDS sample buffer showing the same mobility as unheated samples (lanes 1  and 2). C, rat brain PAK protein was purified as described previously . The purified material was precipitated with 2 volumes of acetone and the pellet dissolved in 100 l of 70% formic acid. Lane 3 shows 10 g of this material after lyophilization. Cynaogen bromide was added to 10 mg/ml, and the solution was degassed and left overnight under nitrogen. The digest was then lyophilized and dissolved in SDS sample buffer. Lanes 1 and 2 show 45 and 5 g of material run on 12% polyacrylamide gels and transferred to PVDF membranes. Following Coomassie Blue staining (left-hand panel) the membrane was overlaid with [␥-32 P]GTP-Cdc42 to detect peptides containing the Cdc42 binding domain.

FIG. 7. Tissue distribution of ␣-PAK by Western analysis.
Soluble proteins from tissues (150 g/lane) were run on 9% SDS-polyacrylamide gels and transferred to PVDF membranes. Identical filters were probed with affinity purified ␣-PAK antibodies followed by horseradish peroxidase-coupled second antibody with luminol detection for 30 s (A) and 2 min (B) or analyzed for [␥-32 P]GTP-Cdc42 binding (C). Tissues are as follows: b, brain; h, heart; k, kidney; li, liver; lu, lung; s, spleen; and te, testis. lowing cleavage at M37 (underlined in Fig. 1). Since this region of ␤-PAK is divergent from both ␣-PAK and hPAK65 sequences, it is highly likely that the peptide is derived from the ␤-PAK gene product. Although a mass of Ͼ30 kDa for a CNBrgenerated peptide of ␤-PAK is not consistent with its C terminus being derived from cleavage at ␤-M138, there are many documented cases of methionines which show inefficient CNBr cleavage, and lower bands were also detected by [␥-32 P]GTP-Cdc42 (Fig. 6C). We cannot estimate the stoichiometry of ␣and ␤-PAK by CNBr digestion and overlay because the ␣-PAK p21-binding domain is probably destroyed by cleavage at at an internal methionine (␣-M99), a position occupied by a conservative isoleucine in ␤-PAK(I94).
Tissue Distribution of PAK Using Anti-␣-PAK Antibodies-Affinity-purified antibodies to ␣-PAK residues 1-251 showed strong immunoreactivity to proteins of 65-68 kDa on Western blots of brain extracts (Fig. 7A). Similar sized proteins were recognized by the antiserum in spleen also expressing the ␣-PAK mRNA. In the testis an immunoreactive band was detected at longer exposure times whose mobility was greater than the broad band detected in the spleen (Fig. 7B), consistent with the presence of testicular ␤-PAK mRNA. It is probable that the presence of multiple bands on Western blots is also the result of different phosphorylation states of the PAKs in vivo. All of these immunoreactive bands corresponded to proteins larger than the ubiquitous p62 band detected by [␥-32 P]GTP-Cdc42 overlay (Fig. 7C). As discussed later, the p62 is also a PAK isoform we designate as ␥-PAK 3 , but which is not recognized by the antiserum.
Cdc42 and Rac1 Activate Recombinant PAK-We have expressed full-length ␣and ␤-PAK as ϳ95-kDa GST fusion proteins in E. coli to study kinase activation using homogeneous preparations. Because the ␤-PAK sequence contained additional polylinker-derived sequence, it has a lower mobility than recombinant ␣-PAK. Thrombin cleavage of this GST/PAK fusion protein caused some internal cleavage of the kinase and thus gave truncated products. We therefore used the GST fusion protein, since it showed similar p21 activation to the "native" purified protein . Fig. 8A shows the activities of glutathione-Sepharose-purified GST/PAK in the absence and presence of GST/Cdc42. Autophosphorylation of recombinant ␣and ␤-PAK and concomitant phosphorylation of myelin basic protein (MBP) substrate were significantly activated in vitro by adding purified recombinant GTP␥S-Cdc42. This was observed previously with affinity-purified native brain PAK, which from data described in this paper corresponds to ␤-PAK gene product. The recombinant ␤-PAK exhibited higher basal activity, which might be related to its phosphorylation state in E. coli. Since the behavior of the p68 shows 32 P-phosphorylated proteins. B, effect of various p21s on GST/␣-PAK activity; Rho-p21s (1 g) were preloaded with the appropriate nucleotide (0.5 mM) and incubated with 0.5 g of GST/␣-PAK and 5 g of MBP at 30°C. Activity incorporated into MBP was quantified using a PhosphorImager (Molecular Dynamics). C, Cdc42 and Rac1 activation of of GST/␣-PAK with 0.25 mM unlabeled ATP. Left-hand panel shows Coomassie Blue-stained kinase; in the right-hand panel the proteins were resolved on a 9% polyacrylamide gel co-polymerized with 0.1 mg/ml MBP. Following the denaturation/renaturing steps (see "Materials and Methods"), [␥-32 P]ATP was added to the gel in kinase buffer to detect MBP kinase activity (2-h exposure). D, activity of inactive and Cdc42-activated ␣-PAK (å); 2 g of recombinant kinase was dialyzed for 2 h against kinase buffer and incubated at 32°C with 20 g of MBP (total volume, 100 l) with 50 M [␥-32 P]ATP. At the times shown 25-l aliquots were removed, quenched by the addition of SDS buffer and fractionated on a 12% gel. Radioactivity associated with MBP is in arbitrary units (an averaged value of three determinations is shown); errors for the unactivated PAK (f) were within the values covered by the symbol.
␣-PAK has not been described previously, we examined its activation in more detail. Activation of recombinant ␣-PAK occurred only with GTP␥S-Rac and GTP␥S-Cdc42, but not GTP␥S-RhoA (Fig. 8B). In their GDP forms all p21s were inactive (exchange of GDP into GTPase-deficient GTP-Cdc42 V12 protein probably being incomplete during the standard 4-min period). Quantification of the MBP phosphorylation (indicated as relative activity) showed that GTP-Cdc42 was more effective than GTP-Rac1 under these assay conditions.
Following complete activation (1 h in the presence of GTP␥S-Rac or GTP␥S-Cdc42), ␣-PAK exhibited slower mobility under SDS-polyacrylamide electrophoresis (Fig. 8C), as seen with the purified protein  and many other protein kinases. The hyperphosphorylated kinase exhibited some size heterogeneity in its activated state. After separation from the GTP␥S-p21s by SDS-PAGE, ␣-PAK was still active as determined by an in-gel kinase assay against MBP (Fig. 8C). No labeling was seen in the absence of MBP in the gel (data not shown). This supports the model we presented previously in which the autophosphorylated kinase, after dissociation of p21, would remain active in the absence of dephosphorylation. The ability of this active kinase to phosphorylate MBP was tested (under similar conditions to the co-activation assay in Fig. 8B). Based on the initial rates of phosphorylation (Fig. 8D), the activity of the phosphorylated form was found to be more than 300 times higher than the unphosphorylated ␣-PAK.
Activation of ␤-PAK Leads to a Decrease in p21 Binding-We had observed previously that purified p65 PAK upon activation exhibited lower affinity for both GTP-Rac1 and GTP-Cdc42 in overlay assays . Similar experiments were conducted with recombinant ␣and ␤-PAK proteins. Fig. 9A shows the binding of [␥-32 P]GTP-Cdc42 to normal and activated kinase in a typical overlay experiment. Averaged data from three independent experiments are shown below. The data clearly demonstrated that only ␤-PAK exhibits downregulation of p21 binding upon activation. Since this effect is probably related to the phosphorylation state of the kinase, we sought to mimic the effect using the N-terminal of ␤-PAK (and of ␣-PAK both expressed as a GST fusion proteins cf Fig. 9C) exogenously phosphorylated by full-length ␤-PAK. Although phosphorylation to completion could be observed by monitoring the upward mobility shift of the N-terminal fusion proteins, similar to that seen with the whole kinase (Fig. 9B), this did not result in any significant change in affinity for the [␥-32 P]GTP-Cdc42 by ␤-PAK. It therefore seems that the down-regulation of Cdc42 binding either requires the whole ␤-kinase (conformational effects) or a specific intramolecular phosphorylation site is involved. Interestingly when the recombinant ␣-PAK kinase domain alone was expressed in E. coli, the fusion protein exhibited no kinase activity, nor could it be phosphorylated by full-length PAKs (data not shown).
The Rac1/Cdc42-binding domain of PAK shows sequence homology to the Cdc42-specific binding domain of the tyrosine kinase ACK . This region is highly conserved in ␣and ␤-PAK (Fig. 2). Although PAK binding to Rac1 appeared significantly weaker than to Cdc42, their ability to activate the autophosphorylation and MBP kinase activity of purified p65-PAK depended on the assay conditions . As illustrated in Fig. 9C, by normalizing the labeling of the p21s using excess (cold) GTP in the "exchange" reaction, the amount of [␥-32 P]GTP-Cdc42 or [␥-32 P]GTP-Rac1 bound to the N-terminal region of ␣and ␤-PAK in overlays was found to be the same (Ϯ10%). Note there is no potential autophosphorylation of the construct. The labeling of Rac1 with [␥-32 P]GTP is normally poor, probably because of its high intrinsic GTPase activity during the exchange reaction in low magnesium buffer (Menard et al., 1992). DISCUSSION The heterogeneity in Cdc42-binding proteins with apparent molecular mass between 60 and 70 kDa in different tissues FIG. 9. Decrease in Cdc42 binding to activated ␤-PAK. A, fulllength GST/PAK was activated in the presence of 2-fold excess GTP␥S-Cdc42 and 0.5 mM ATP for 1 h at 30°C (ϩActivation) or in the presence of 1 unit of alkaline phosphatase (unphosphorylated control). Proteins separated by SDS-PAGE and stained with Coomassie Blue are shown in the left-hand panel. Indicated amounts of the proteins were also transferred to PVDF and processed for [␥-32 P]GTP-Cdc42 overlay. The filters were placed against a phosphor screen at Ϫ20°C (output from one experiment shown) and analyzed using the Imagequant software (Molecular Dynamics) to determine radioactivity associated with each band. Average data from three independent experiments are presented. B, residues corresponding to ␣-PAK 1-251 (BamHI/BglII cDNA fragment) and ␤-PAK 1-231 (SmaI/Sca1 cDNA fragment) fused to GST were phosphorylated as in A with full-length GST/␤-PAK. The unphosphorylated and phosphorylated proteins were resolved on a 9% gel, transferred to PVDF, and overlaid with [␥-32 P]GTP-Cdc42. C, 2 g of fusion protein, as in B, were overlaid and bound counts at the position of the 60-kDa band were quantified and found to be Ϯ 10% between Rac1 and Cdc42 in two independent experiments. suggested these kinases to be expressed from a number of related genes. The identity and relationship of two of these kinases (␣and ␤-PAK) have now been established through isolation of their cDNAs. Amino acid sequence comparison between the two mammalian PAKs described here and hPAK65 (Martin et al., 1995) reveals functionally important regions of the protein. In particular the p21-binding domain showed almost no sequence divergence. PAK kinases also contain polyacidic and proline-rich sequences between the p21-binding and kinase domains. The use of purified recombinant proteins has enabled us to confirm that addition of GTP␥S-p21 was sufficient to activate the kinase in the absence of any factor that might have co-purified with brain-derived native PAK. This activation by Rac1 or Cdc42 was achieved through the p21mediated autophosphorylation of the kinase. Recombinant (p68) ␣-PAK showed robust activity toward MBP in the presence of GTP␥S-p21, as we have described for purified p65 PAK (␤-PAK). Here we show with recombinant ␣-PAK protein that there is a 300-fold increase in activity following p21-mediated autophosphorylation.
While ␣-PAK was expressed predominantly in the brain, both mRNA and protein were detected in the spleen. Within the immune system it appears that PAK-related kinases are relatively abundant; the report that neutrophil p62 and p68 Cdc42/Rac-binding proteins are not reactive to antibodies to the conserved PAK kinase domain (Martin et al., 1995) is in conflict with our demonstration that antibodies generated against ␣-PAK recognize the neutrophil p68 protein (Prigmore et al., 1995). The p65 ␤-PAK appears to be more restricted in its expression outside of the nervous system. We have been unable to correlate the PAK expression pattern with any known signaling molecule or receptor, but the strikingly restricted expression pattern of both isoforms lends credence to the idea that ␣and ␤-PAKs may be coupled to specific Rho-p21 pathways in the brain. Another Rho-p21 interacting protein with highly restricted expression is the RacGAP ␤2-chimaerin, which is only present in cerebellar granule cells  where high expression of ␣-PAK mRNA is also observed.
The hPAK65 mRNA has been reported to be expressed in a wide variety of tissues and in cultured cells. We have obtained peptide sequence data allowing the cloning of the ubiquitous rat p62 Cdc42/Rac binding protein which we designate as ␥-PAK. 3 Because the published hPAK65 cDNA contains, in the 5Ј-noncoding sequence and 38 base pairs of putative open reading frame, sequence identical to human placental lactogen (somatomammotropin, Seeburg et al., 1977), this region of the hPAK cDNA may have arisen as a library artifact. The only biochemical property reported for this recombinant enzyme that differs significantly from those of brain-purified p65 PAK is that there is no decrease in Cdc42 binding of activated hPAK65 (Martin et al., 1995). Our study clearly shows that a difference in this respect exists between the ␣and ␤-isoforms in that ␤-PAK, but not ␣-PAK, binds GTP-Cdc42 more weakly after activation of the kinase. This is consistent with ␤-PAK cDNA encoding the purified p65 PAK protein which exhibits similar behavior. Autophosphorylation of ␣-PAK occurs on at least 6 residues in the N-terminal regulatory region. 4 There is, therefore, potential for similar phosphorylation events in the ␤-PAK, perhaps altering the conformation of the p21-binding domain such that its affinity is affected. The loss of affinity of ␤-PAK for GTP-Cdc42 after activation of the kinase suggests that the p21 in this case could amplify its signal by activating several kinase molecules, or alternatively it allows the activated kinase to act at a different site. In contrast, ␣-PAK activity may co-localize with GTP-p21 and remain as part of the signaling complex.
Although functions for PAK kinases in relation to the roles of Cdc42 and Rac1 in mammalian cells is yet to be established, we have been able to demonstrate that Ste20p, a kinase required for the mating response pathway in S. cerevisiae (Leberer et al., 1992;Ramer and Davis, 1993), requires GTP-Cdc42p for its function. 5 As in higher organisms, there appear to be a family of PAK kinases in yeast (Blumer and Johnson, 1994); the 36-kDa yeast kinase previously thought to only contain a kinase domain is now confirmed as having an N-terminal p21binding domain (Fig. 2). The role of Ste20p as a kinase acting above the pheromone-responsive mitogen-activated protein kinase cascade may point to the involvement of mammalian Cdc42 and Rac proteins in cell proliferation and differentiation.