A Novel Regulator of p21-activated Kinases*

Proteins of the p21-activated kinase (Pak) family have been implicated in the regulation of gene expression, cytoskeletal architecture, and apoptosis. Although the ability of Cdc42 and Rac GTPases to activate Pak is well established, relatively little else is known about Pak regulation or the identity of Pak cellular targets. Here we report the identification of two closely related Pak3-binding proteins, possibly arising from alternative splicing, designated p50 and p85Cool-1 (cloned outof library). Both isoforms of Cool contain a Src homology 3 domain that directly mediates interaction with Pak3 and tandem Dbl homology and pleckstrin homology domains. Despite the presence of the Dbl homology-pleckstrin homology motif, a characteristic of Rho family activators, activation of Cdc42 or Rac by Cool is not detectable. Instead binding of p50Cool-1, but not p85Cool-1, to Pak3 represses its activation by upstream activators such as the Dbl oncoprotein, indicating a novel mechanism of regulation of Pak signaling.

The Rho family GTPases Rac1 and Cdc42 mediate diverse biological events including changes in the cytoskeletal architecture (1)(2)(3), stimulation of DNA synthesis (4), cellular transformation (5)(6)(7)(8), and signaling to the nucleus (9 -14). Many of the signaling pathways leading to the execution of these events involve the p21-activated kinases, Paks1-3, 1 which are direct effectors of Rac1 and Cdc42 (15)(16)(17). Binding of these GTPases to a conserved p21-binding domain (PBD, also known as CRIB for Cdc42/Rac1 interactive binding) stimulates their serine/ threonine kinase activities by a mechanism involving autophosphorylation (15,18). The important roles that Paks play as effectors of Cdc42/Rac1 signaling have been established from genetic and biochemical studies in yeast and mammalian cells. The budding yeast homolog of mammalian Paks, Ste20, acts in concert with Cdc42 in the pheromone response to activate a MAP kinase cascade leading to transcription of genes required for cell cycle arrest. The same protein functions as an effector of Cdc42 to activate a different MAP kinase cascade leading to filamentous growth in response to nitrogen starvation (19). The yeast model in which Pak/Ste20 links Cdc42 to transcriptional and cytoskeletal events is paralleled in mammalian cells; constitutively activated Pak mutants can activate the c-Jun Nterminal kinase MAP kinase cascade leading to transcriptional control (12,13) and can mimic some, although not all, of the effects of Rac1 or Cdc42 on cytoskeletal organization (18,20).
The apparent multiplicity of Pak-mediated signaling pathways suggests that Pak activity must be tightly regulated. This has been made all the more clear from the observations that Pak1 function is required for cellular transformation by Ras (8) and that Pak2 activation is involved in Fas-mediated apoptosis (21,22). Here we describe the identification of two isoforms of a novel Pak-binding protein, probably resulting from alternative splicing of the same message, one of which is able to suppress Pak activation by upstream regulators.

EXPERIMENTAL PROCEDURES
Yeast Two-hybrid Screen-To identify proteins that interact with Pak3, the yeast strain L40 was co-transformed with Pak3 fused to the LexA DNA-binding domain, which regulates expression of both his3 and lacZ, and a HeLa cDNA library was fused with the Gal4 activation domain (23). Clones positive for ␤-galactosidase activity were rescued, and several clones that provided bait-dependent his3 and lacZ gene activation were sequenced. A 1-kilobase (EcoRI-XhoI) fragment of clone Y107 was used to screen an oligo(dT)-primed, size fractionated (4 -9 kilobases) U937 library (a gift from J. Burrows, Massachusetts Institute of Technology, Boston, MA) and a HeLa cDNA library (Stratagene). The U937 library yielded a full-length clone (A6) that was predicted to encode a protein of 436 amino acids, p50 Cool-1 . Several partial clones were recovered from the HeLa library that appeared to represent alternatively spliced forms of Y107. One of these (clone 12a) was identical to a recently cloned cDNA, p85 SPR , but lacked the 3Ј end encoding its C-terminal 31 amino acids. The p85 Cool-1 cDNA was generated by fusing this 3Ј end, derived from the 3Ј-untranslated region of the U937 clone A6, to clone 12a (p85 Cool-1 is therefore identical to p85 SPR ).
Kinase Assays, Affinity Precipitation, and Immunoprecipitations-Kinase reactions were initiated by the addition of 2ϫ kinase buffer (40 mM Hepes, pH 7.4, 20 mM MgCl 2 , and 4 mM MnCl 2 ) and 20 M [␥-32 P]ATP (3000 Ci/mmol) for 3.5 min at room temperature. Reactions were stopped by the addition of 2ϫ SDS sample buffer containing 20 mM EDTA. Affinity precipitation with GST-PBD was as described (24) except that COS cells were lysed in 25 mM Hepes, pH 7.4, 150 mM NaCl, 1% Nonidet P-40, 10% glycerol, 10 mM MgCl 2 , 1 mM EDTA, 10 g/ml leupeptin, and 10 g/ml aprotinin. Lysates containing equal amounts of Pak3 were immunoprecipitated with anti-Pak3 primary antibody (a gift from Dr. S. Pelech, Kinetek Biotechnology Corporation, Vancouver, Canada) Anti-Cool-1 antibody was prepared using His-tagged full-length p50 Cool-1 , treated with thrombin to cleave the His-tag and further resolved on a Mono Q-Sepharose column. Rabbit serum was collected 12 weeks after three injections with p50 Cool-1 .

RESULTS AND DISCUSSION
To identify potential Pak-binding partners we used the yeast two-hybrid screen. One positive clone was used to screen cDNA libraries and data bases (see "Experimental Procedures") to yield two Pak3-binding proteins, p50 and p85 Cool-1 . They share an N-terminal SH3 domain (amino acids 7-65), followed by a Dbl homology (DH) domain (amino acids 100 -279) and an adjacent pleckstrin homology (PH) domain (amino acids 295-400) (Fig. 1A). Because p50 and p85 Cool-1 are identical at the nucleotide level over amino acids 1-418, it is likely that they arise from alternative splicing of the same message. p85 Cool-1 is identical to two recently cloned proteins, p85 SPR (25) and ␤-PIX (26). The DH domain of the Cool proteins shows the highest sequence identity to the DH domains of Dbl (33%), Scd1 (30%), Dbs (29%), Tiam-1 (29%), Still Life2 (29%), Cdc24 (26%), and Vav2 (24%). In addition to the genetic screen, we used recombinant GST-Pak3 (amino acids 148 -239) to purify Pak-binding proteins from Src(Y527F)-transformed NIH 3T3 cells. Sequence obtained from an ϳ85-kDa Pak-binding protein was identical to portions of p50 and p85 Cool-1 (data not shown). We also identified a closely related cDNA from the data bases, and we have found that its product also interacts with Pak3. 2 This protein, Cool-2, is identical to the recently described ␣-PIX (26).
To determine whether the Cool proteins bind to Pak3 in mammalian cells, we transiently co-expressed Myc-tagged p50 Cool-1 and HA-tagged Pak3 in COS cells and assayed for complex formation by immunoprecipitation and Western blot analysis (Fig. 1B). HA-Pak3 was detected in anti-Myc immunoprecipitates (Fig. 1B, lane 8, upper panel) and Myc-p50 Cool-1 was detected in anti-HA immunoprecipitates (Fig. 1B, lane 13 3 and SH3 domains from the Dbl family proteins Dbs and Vav did not bind Pak3 (data not shown), showing specificity of the Cool-1/Pak3 association in vivo. Pak3 contains four conventional (PXXP motif) SH3-binding sites (P1-P4); Pak3 containing Pro to Ala mutations in the P1-P4 sites, alone or in combination, were used to establish that these sites do not mediate Pak3-Cool-1 interactions (data not shown). While this manuscript was in preparation, Manser et al. reported binding of ␣-PIX and ␤-PIX (which correspond to Cool-2 and p85 Cool-1 ) to residues 182-203 of Pak1, confirming the atypical (non-PXXP motif) nature of Cool binding to Paks (26). Using a rabbit antiserum raised against full-length p50 Cool-1 , we detected two predominant proteins (from NIH 3T3 fibroblast lysates) that migrated at ϳ85 and ϳ78 kDa on a SDS-polyacrylamide gel (Fig. 1C, right panel, lane 1). Both of these proteins were detected in anti-Cool-1 immunoprecipitates (Fig. 1C, right  panel, lane 3) as well as in anti-Pak3 immunoprecipitates (Fig.  1C, right panel, lane 2), demonstrating an interaction between endogenous Cool proteins and Pak3 in NIH 3T3 cells. The less reactive band at ϳ50 kDa recognized by the anti-Cool-1 antibody (Fig. 1C, right panel, lane 1) may represent p50 Cool-1 ; however, we were unable to determine whether p50 Cool-1 was present in the immunoprecipitates because of the overlapping signal for IgG.
The presence of a DH-PH tandem motif in Cool-1, a hallmark of Dbl family exchange factors (27), initially led us to consider the possibility that Cool-1 might activate Cdc42 or Rac. However, we were unable to detect stimulation of [ 3 H]GDP dissociation from Cdc42 or Rac by recombinant p50 Cool-1 purified from Escherichia coli or insect cells under conditions where the Dbl oncoprotein strongly stimulated GDP dissociation from Cdc42 or RhoA (data not shown).
We then considered the possibility that Cool-1 exchange activity may require cellular co-factors or post-translational mod-  ifications as proposed for Vav, Sos, and Tiam-1 (28 -31). To measure Rac1 or Cdc42 activation in vivo, we used a modification of a recently described assay for activated, GTP-bound Ras (24). The PBD of Pak3 was expressed as a GST fusion protein and immobilized by binding to glutathione-Sepharose beads. The immobilized GST-PBD was used to precipitate activated Cdc42 or Rac1 from transfected COS cell lysates (Fig. 2). In untreated control cells, relatively low levels (Ͻ5%) of HA-Rac1 (upper panel) or HA-Cdc42 (lower panel) were precipitated with GST-PBD (compare lane 1 (cell lysates) and lane 9 (affinityprecipitated GTPase) in both panels). When Dbl was co-expressed with HA-tagged GTPases, there was a marked increase in the amounts of activated HA-Rac1 and HA-Cdc42 that were precipitated with GST-PBD (lane 16). The cellular activation of both Cdc42 and Rac by Dbl is consistent with results from micro-injection studies (32). In contrast, p50 Cool-1 did not detectably activate either HA-Rac1 or HA-Cdc42 (compare lanes 9 and 10), even after exposure to a number of different biological stimuli (lanes 11-15). Activation of Rac1 or Cdc42 by p85 Cool-1 was also not detected, even when co-expressed with Pak3, under conditions where Tiam-1 resulted in the activation of Rac1 but not Cdc42 (data not shown). We have not detected Cool-1-mediated Rac1 activation in cells co-expressing potential activators of Rac including Src(Y527F), Ras(G12V), and Cdc42(Q61L) 2 . We next tested whether p50/p85 Cool-1 could directly modulate Pak activity. Although p85 Cool-1 was without significant effect (Fig. 3A, compare lanes 2 and 4), p50 Cool-1 inhibited Dbl-stimulated Pak3 activity (lanes 2 and 3). p50 Cool-1 (W43K) did not inhibit Dbl-activated Pak3 (lanes 2 and 5), indicating that binding of p50 Cool-1 to Pak3 was required for its inhibitory effect. Immunoblotting confirmed that wild type and mutant Cool-1 proteins were equally expressed and that Dbl expression levels were unaffected by Cool-1 expression (data not shown); immunoprecipitated Pak3 levels were equal (Fig.  3A, lower panel). We also found that purified, recombinant p50 Cool-1 completely blocked Cdc42(Q61L)-stimulated autophosphorylation of Pak3 and strongly inhibited the phospho-rylation of myelin basic protein in Pak3 immunoprecipitates (data not shown).
Unlike p50 Cool-1 , expression of p85 Cool-1 did not inhibit Cdc42(Q61L)-stimulated Pak3 autophosphorylation (Fig. 3C,  compare lanes 8 and 9), which is consistent with the results shown in Fig. 3A. As expected from the presence of identical SH3 domains, p50 and p85 Cool-1 competed for binding to Pak3 in vitro (lanes 9 -12). In these experiments, p50 Cool-1 co-precipitating with Pak3 was not detected with anti-Myc (upper panel, lanes 8 and 10 -12) but was readily detectable with an anti-Cool-1 antibody (lower panel, lanes 8 and 10 -12). The fact that an ϳ3-fold excess of either p50 Cool-1 or p85 Cool-1 was able to significantly inhibit the binding of the other to Pak3 suggests that they have similar affinities for Pak3.
Our results suggest that p50 Cool-1 , by competing with endogenous p85 Cool-1 or other Cool proteins for binding to Pak3, might sequester Pak3 away from its site(s) of activation. On the other hand p85 Cool-1 , which has a permissive effect on the stimulation of Pak activation by Dbl or other Rho family exchange factors, could play an important role in recruiting Pak to its sites of activation, possibly via its C-terminal region, which is not present in p50 Cool-1 . In support of this model, our preliminary data indicate that p50 Cool-1 has a diffuse cytoplasmic localization, 4 whereas p85 Cool-1 is concentrated at focal adhesions (25). Moreover, HeLa cell p85 Cool-1 (␤-PIX) has recently been shown to localize to focal complexes and appears to mediate Pak1 recruitment to these sites by activated Cdc42 (26). However, this would not account for the ability of p50 Cool-1 to inhibit Cdc42-stimulated Pak3 activity in vitro. Because both p50 Cool-1 and p85 Cool-1 appear to bind to a common site on Pak3, the specific inhibition of Pak3 activity by p50 Cool-1 may be due to the differences in the C-terminal regions of p50 and p85 Cool-1 .
Protein kinases are often subject to multiple levels of regu-lation. The involvement of Paks in cellular signaling pathways leading to changes in gene expression or cytoskeletal architecture and their participation in both Ras-mediated transformation (8) and Fas-mediated apoptosis (21,22) mandates that their activities be tightly controlled. The ability of p50 Cool-1 to suppress and p85 Cool-1 to permit Pak activity indicates that signaling through Pak-dependent pathways may be regulated by cell type, cell cycle, or developmental-specific expression patterns. It will be important to establish the role of DH and PH domains in Cool function and how differential expression of Cool impacts upon Pak signaling to different effector pathways.