The genetics of Pak

Simple eukaryotes such as yeasts and molds encode multiple Paks that, like their orthologs in other systems, act downstream of Rho-family GTPases to regulate cytoskeletal structure and gene transcription. All Paks contain an N-terminal p21 GTPase-binding-domain (PBD), which confers binding to small GTPases such as Cdc42 or Rac, and a C-terminal protein kinase domain. In addition, some Paks from simple eukaryotes contain an N-terminal pleckstrin homology (PH) domain, a feature not found in Paks from more-complex organisms. In higher eukaryotes, the Pak family is divided into two subfamilies: group A and group B (Bokoch, 2003; Dan, I. et al., 2001; Jaffer and Chernoff, 2002) (Fig. 1A). Group A Paks are characterized by the presence of N-terminal proline-rich motifs that mediate association with various Src-homology 3 (SH3)-domain-containing proteins, a PBD, and a C-terminal kinase domain (Fig. 1B). Group A Paks bind both Cdc42 and Rac, and are strongly activated upon binding these GTPases. Group B Paks contain a PBD at the extreme N-terminus of the protein followed by a C-terminal kinase domain (Fig. 1B). The group B Paks bind Cdc42 and, to a lesser extent, Rac, but, unlike group A Paks, are not appreciably activated upon binding. Thus, the term `p21-activated kinase' is not entirely apt in this case. Instead, the association with Cdc42 is thought to be more important for localization of the group B kinases rather than for their activation per se. Despite these differences, both groups share certain essential functions, such as the regulation of the actin cytoskeleton. However, they are not completely functionally equivalent. For example, mammalian group A Paks can replace the function of the budding yeast Pak Ste20, whereas group B Paks cannot (Cotteret et al., 2003).

Here, we consider genetic analyses of Pak function in yeasts, amoebae, flies and mammals. Although multiple Paks are also present in the nematode Caenorhabditis elegans, which is another genetically tractable model system, very little is known of Pak function in this organism.

For historical and technical reasons, genetic analysis of Pak function has proceeded further in Saccharomyces cerevisiae than in any other organism. Although some of the findings are unique to budding yeast, many of the basic signaling pathways in which yeast Paks have been found to participate are also shared in other organisms. S. cerevisiae encodes three Paks: Ste20, Cla4 and Skm1. Like all members of the Pak family, these three kinases contain an N-terminal PBD and a C-terminal protein kinase domain; however, Cla4 and Skm1 also contain a PH domain N-terminal to the PBD and, together with Pak2 from Schizosaccharomyces pombe, form a distinct subfamily (Fig. 1). As discussed below, these kinases affect cell morphology, polarity, cell-cycle and gene transcription events downstream of the small GTPase Cdc42 and the cell-cycle-dependent kinase Cdc28 (Etienne-Manneville, 2004).

The founding member of the Pak family, Ste20, was uncovered in a genetic screen for suppressors of the mating defects associated with expression of a dominant-negative form of Ste4, a mating-pheromone-receptor-associated Gβ subunit. ste20 cells fail to arrest in G1 phase following pheromone stimulation, fail to form mating projections, and fail to activate transcription of key mating factors, such as Fus1 (Leberer et al., 1992; Ramer and Davis, 1993). In the pheromone mating pathway, Ste20 functions as an activator of mitogen-activated protein kinase (MAPK) cascades, a function that appears to be broadly conserved throughout evolution (Dan, I. et al., 2001).

In budding yeast, the mating MAPK cascade comprises a MAPK (Kss1 or Fus3), a MAPK kinase (MAPKK; Ste7) and a MAPK kinase kinase (MAPKKK; Ste11). Activation of this pathway through the mating pheromone receptor results in changes in transcription and in cytoskeletal structure that are required for mating (Elion, 2000). Epistasis and biochemical experiments have shown that Ste20 operates downstream of Cdc42 and Ste4 near the top of the mating MAPK cascade (Fig. 2A). Ste20 phosphorylates the MAPKKK Ste11 on residues Ser302 and/or Ser306 and Thr307, and alanine substitutions at these positions abolish Ste11 function (Drogen et al., 2000; Wu et al., 1996). Thus, in this system, Ste20 appears to act as a classic MAP kinase kinase kinase kinase (MAPKKKK). As will be discussed, although it is common to extrapolate these findings to other systems, the connections between Paks and elements of the MAPK cascade in other organisms are often more complex than in budding yeast.

Ste20 also functions as an upstream activator of MAPK pathways that regulate invasive growth and the osmotolerance response. As in the pheromone pathway, the main target of Ste20 in both the invasive and the osmotolerance response pathways is thought to be Ste11 (de Nadal et al., 2002; O'Rourke et al., 2002; Posas et al., 1998). Expression of a dominant-negative form of Cdc42, or of a Ste20 mutant that cannot bind to Cdc42, impairs signaling through the invasive growth pathway, indicating that interaction with Cdc42 is required for proper activation of Ste20 in this pathway (Mosch et al., 1996) (Fig. 2A). Cdc42 also plays an important, albeit ill-defined, role in regulating Ste20 in the osmotolerance response pathway (Raitt et al., 2000). Finally, Ste20 and components of the mating, invasive growth and osmotolerance pathways also participate in a vegetative growth pathway that is implicated in the control of cell wall integrity (Cullen et al., 2000; Elion, 2000; Lee and Elion, 1999).

Ste20 is controlled by specific regulators in the invasive growth signaling pathway. Bmh1 and Bmh2, yeast orthologs of mammalian 14-3-3 adaptor proteins, associate with Ste20 in vivo and are required for invasive growth (Roberts et al., 1997). The methyltransferase Hsl7 negatively regulates Ste20 function in this pathway. Hsl7 binds to the N-terminal region of Ste20, possibly competing with Cdc42 (Fujita et al., 1999). Interestingly, Skb1, a fission yeast homolog of Hsl7, has been shown to have the opposite effect: it positively modulates Pak1 function and binds to a site on Pak1 that is different from that which binds Cdc42 (Gilbreth et al., 1996).

Ste20, together with the related protein Cla4, is also required for many of the polarity changes that occur during the vegetative growth cycle, including bud emergence (late G1 phase) and polarized growth of budded cells (G2/M phase), as well as for cell division. The cell-cycle effects are independent of the MAPK pathway, which indicates that Ste20 has one or more additional substrates that are required for cell polarization (Eby et al., 1998; Holly and Blumer, 1999).

During bud emergence, Ste20 is phosphorylated by the CDK-cyclin complex Cdc28-Cln2, and this appears to promote the vegetative morphogenetic functions of Ste20 (Oehlen and Cross, 1998; Wu et al., 1998). Loss of Pak function in late G1 phase (through inactivation of Cla4 in a ste20 strain) leads to a complete loss of polarization and prevents bud emergence (Holly and Blumer, 1999). One of the relevant Pak targets in this process might be Bem1, because this protein is involved in bud emergence, localizes to sites of polarized growth in G1 phase, co-immunoprecipitates with Ste20, and is phosphorylated in vivo (Chenevert et al., 1992; Leeuw et al., 1995). Myosins such as Myo3 and Myo5 might also be involved, as these proteins are required for budding and are phosphorylated in vitro by Ste20 and Cla4 at sites known to be required for function in vivo (Wu et al., 1996; Wu et al., 1997).

The switch from apical to isotropic bud growth (G2/M phase) is regulated by Cdc28-Clb2. Cla4 is hyperphosphorylated during mitosis in a Cdc28-dependent fashion (Tjandra et al., 1998), and again loss of Pak function during this stage (by inactivation of Cla4 in a ste20 strain) leads to a complete loss of polarization (Holly and Blumer, 1999). These effects might be in part mediated through the Nim1-related kinase Gin4, which is dependent on Cla4 for activation and regulates septin dynamics, although no direct activation of Gin4 by Cla4 has been reported (Benton et al., 1997; Tjandra et al., 1998), as described below. The formin-homology protein Bni1 represents another likely target for Ste20 in this process. bni1 is synthetically lethal in a cla4 background (i.e. in cells that are dependent on Ste20 function for viability), and much of the phosphorylation on Bni1 is dependent on Ste20 (Goehring et al., 2003). Bni1 is a component of the `polarisome' and many other components of this pathway are also synthetically lethal when mutations in them are combined with cla4 (Goehring et al., 2003). These results suggest that one function of Ste20 might be to activate the polarisome complex by phosphorylating Bni1.

Paks also participate in a feedback system to signal the end phase of polarized growth. Until the time of bud emergence, Cdc24, which is the activator of Cdc42, resides in the nucleus in a complex with the adaptor protein Far1 (Shimada et al., 2000; Toenjes et al., 1999). Concomitantly with bud emergence, Cdc28 phosphorylates Far1, inducing its ubiquitin-dependent degradation and promoting the release of Cdc24 from the complex. Cdc24 then translocates to the incipient bud site where, together with the adaptor protein Bem1, it activates Cdc42. Once activated, Cdc42 is able to interact with its effectors, including Cla4. Cla4 phosphorylates Cdc24, inducing its dissociation from the adaptor protein Bem1, thus providing a negative-feedback loop to inactivate Cdc42 and to end the phase of polarized growth. The role in the regulation of Cdc24 seems to be specific for Cla4, because Cdc24 phosphorylation occurs in cells lacking STE20 (Bose et al., 2001; Gulli et al., 2000).

Cla4 was initially identified in a screen for mutants that are not viable in the absence of the G1 cyclins Cln1 and Cln2. In this genetic background, cells carrying a mutation in the CLA4 gene show defects in cytokinesis: cells bud and their nuclei divide but cytokinesis does not occur. When Cln1 and Cln2 are present, Cla4 is dispensable for viability, although haploid cells still show aberrant cytokinesis. As noted above, viability is lost in cla4 ste20 double mutants, which suggests that these two kinases share at least one function critical for vegetative growth (Cvrckova et al., 1995).

Many of the cytokinetic defects observed in cla4 cells are related to abnormal septin ring assembly. Septins are GTPases that, in normal cytokinesis, form a filamentous ring around the mother bud neck. Morphological studies have shown that the septin ring is severely mislocalized in cla4 cells (Cvrckova et al., 1995) and that deletion of CLA4 in strains expressing different combinations of GTP-binding-deficient septins aggravate the aberrant morphology of these cells and, under some growth conditions, lead to lethal effects. Complementation studies have shown that expression of Cla4, but not Ste20 or Skm1, is able to rescue the morphological aberrations and cytokinesis defects of cells expressing GTP-binding deficient septins, suggesting that septins might be a direct and unique target of Cla4. In support of this model, Cla4 phosphorylates at least two septins in vitro (Cdc3 and Cdc10), and phosphorylation of these septins in vivo is largely dependent on Cla4 kinase activity (Versele and Thorner, 2004). Septins, possibly as a result of phosphorylation, are immobilized within the bud neck during S, G2 and M phases, and this immobilization is dependent on both Cla4 and Gin4 (Dobbelaere et al., 2003; Mortensen et al., 2002).

Cla4 also plays at least two roles in the regulation of mitosis linked to its effects on bud neck formation. First, Cla4 is involved in the destruction of the cell-cycle inhibitor Swe1. Swe1 is recruited to the bud neck and is hyperphosphorylated prior to its ubiquitin-mediated degradation (Lew, 2000). Chiroli et al. have shown that cells expressing a dominant-negative form of Cla4 show a delay in the onset of anaphase and this delay requires Swe1 (Chiroli et al., 2003). Sakchaisri et al. have shown that Cla4, along with the Polo kinase Cdc5, phosphorylates and downregulates Swe1, which allows mitosis to proceed (Sakchaisri et al., 2004). Consistent with these data is the observation that Cla4 activity peaks near mitosis and drops as cells complete cytokinesis and enter G1 phase (Benton et al., 1997). Second, Cla4 is required for the phosphorylation of the guanine-nucleotide-exchange factor (GEF) Lte1. Lte1 is an activator of the GTPase Tem1 which, following inactivation of the Cdc28-Clb complex, promotes the release of protein phosphatase Cdc14 from its inhibitor in the nucleus. Active Cdc14 then reverses phosphorylation of CDK substrates, promoting exit from mitosis (Simanis, 2003). Accordingly, cells lacking Cla4 are severely delayed in telophase, which is consistent with the idea that Cla4 plays a key role in this process (Seshan et al., 2002).

Interestingly, both the PBD and the PH domains of Cla4 appear to be essential for its localization to sites of polarized growth and for execution of mitotic exit. Since Cla4 must retain the ability to bind both Cdc42 and phosphoinositides, Wild et al. have termed it a `coincidence detector', responding only when two activation signals are both present (Wild et al., 2004). Whether other PH-domain-containing Paks, such as S. cerevisiae Skm1 or S. pombe Pak2, are also coincidence detectors remains to be determined.

Like Cla4, Skm1 contains an N-terminal PH domain, in addition to the PBD and protein kinase domain (Fig. 1B). Disruption of SKM1 causes no obvious phenotype, even in cells grown under stress stimuli; thus, the protein encoded by this gene is not critical for vegetative growth and morphogenesis. Complementation experiments have shown that the function of SKM1 is not redundant with that of STE20 or CLA4, because disruption of either STE20 or CLA4 in a skm1 background causes no distinguishable difference between single deletion of STE20 and CLA4 and the respective double knockouts. Moreover, overexpression of SKM1 does not compensate for STE20 or CLA4 deletion. When artificially activated by truncation of the N-terminus, Skm1 is able to restore mating to ste20 cells but does not suppress the morphogenetic defects associated with deletion of CLA4. These data suggest that the kinase domain of Skm1 can potentially phosphorylate at least some of the same substrates as Ste20, but that, in the context of the full-length protein, is unable or unavailable to do so (Martin et al., 1997). Interestingly, the lethal effects of expressing activated Cdc42 in budding yeast can be suppressed by deletion of CLA4 or SKM1, but not by deletion of STE20 (Davis et al., 1998). These results suggest that, like the other two yeast Paks, Skm1 is a bona-fide effector of Cdc42, but its precise role remains for the moment obscure. Synthetic lethal analyses using skm1 cells might be useful to clarify its physiological role.

The fission yeast S. pombe encodes two Pak kinases – Pak1/Shk1 and Pak2/Shk2 – both of which, like their counterparts in budding yeast, have been implicated in the regulation of MAPK cascades and cytoskeletal dynamics (Fig. 3). Pak1 most closely resembles budding yeast Ste20, whereas Pak2 contains a PH domain and is most closely related to Cla4 and Skm1 (Fig. 1).

Disruption of the PAK1 gene has shown that this kinase is essential for viability (Marcus et al., 1995; Ottilie et al., 1995). Promoter-shutoff experiments suggest that this lethality is due to polarity defects, because PAK1-null spores germinate but arrest after several rounds of cell division as small spherical cells. This phenotype is similar to that observed in CDC42-null cells (Miller and Johnson, 1994) and suggests that Cdc42 and Pak1 function in the same signaling pathway. Deletion of PAK1 or expression of kinase-dead Pak1 in fission yeast induces a spherical morphology and delocalization of actin patches, which is consistent with an inability to polarize growth towards the cell tips, and severely reduces mating efficiency. Similar defects have been noted in orb2-32 strains of S. pombe, which bear a temperature-sensitive mutation in the PAK1 gene. Cells carrying this allele can grow only in a monopolar fashion, because of an inability to recognize their ends as sites for growth (Sawin et al., 1999).

Trans-species complementation experiments showed that Ste20 is able to restore viability, mating and morphology to PAK1-null S. pombe; likewise, Pak1 is able to induce the mating pheromone response pathway in S. cerevisiae lacking STE20 (Marcus et al., 1995; Ottilie et al., 1995b). These observations indicate that, despite the wide evolutionary distance between fission and budding yeasts, these two kinases are functionally related.

Pak1 interacts with the SH3-domain-containing adaptor proteins Scd2/Ral3 and Skb5 (Chang et al., 1999; Yang et al., 1999), the protein methyltransferase Skb1 (a homolog of budding yeast Hs17 (Gilbreth et al., 1996) and the WD-repeat protein Skb15 (Kim et al., 2001) (Fig. 3). Scd2 acts as a scaffold and stimulates Pak1 activity by modulating the interaction between Pak1 and Cdc42 (Chang et al., 1999). Skb1 binds to and activates Pak1, and forms a complex with Pak1 and Cdc42 in vivo (Gilbreth et al., 1996). Together with Skb1, Pak1 directly associates with the cyclin-dependent kinase Cdc2 and negatively regulates mitosis (Gilbreth et al., 1998). By contrast, Skb15 acts as a negative regulator of Pak1. Inhibition of Pak1 activity by Skb15 is essential for the proper regulation of actin remodeling and cytokinetic function (Kim, H. et al., 2003). In addition to these connections, there are genetic interactions between PAK1 and the genes encoding the mating-pathway kinase Byr2 (a MEKK ortholog) (Tu et al., 1997) and Orb6 (an ortholog of mammalian Rho-activated kinase, p160 ROCK) (Verde et al., 1998). Pak1 appears to act upstream of both Byr2 and Orb6, although neither of these latter proteins has been reported to be a direct substrate of Pak1. Interestingly, like Pak1, Orb6 inhibits mitosis in a Cdc2-dependent manner (Verde et al., 1998). Whether Pak1 and Orb6 constitute a Pak1-Orb6-Cdc2 pathway or act on Cdc2 in parallel remains to be established. In the case of Byr2, it is the physical interaction with Pak1, rather than phosphorylation, that relieves autoinhibition and promotes activation of the downstream MAPK cascade (Tu et al., 1997). Thus, the link between Pak1 and the MAPK mating cascade in fission yeast appears to operate on a different principle to that seen for Ste20 in budding yeast, involving functions of Pak that are independent of protein kinase activity.

Qyung and co-workers have established that Pak1 function is required for proper regulation of interphase and mitotic microtubule (MT) dynamics, and that Pak1 activity is reduced by the destabilization of MTs but returns to normal levels as MTs polymerize (Qyang et al., 2002). In this regard, it is interesting to note that mammalian Pak1 phosphorylates and inactivates the MT-associated protein Op18/Stathmin, which promotes destabilization of MTs (Wittmann et al., 2004). Thus, a role for Pak in the regulation of MT stability seems to be common in many organisms.

Pak1 is localized to the cell ends during interphase and to the septum-forming region during mitosis, and associates with mitotic spindle and interphase MTs. The Pak1 inhibitor Skb15 is also associated with mitotic spindles, which suggests this protein has a role in the regulation of Pak1 functions required for spindle/MT dynamics in fission yeast (Sawin et al., 1999). Pak1 is also connected to the cell polarity apparatus through the cell polarity factor Tea1, which is phosphorylated by Pak1 in vivo (Qyang et al., 2002). Phenotypes associated with Pak1 activation caused by the loss of Skb15 (e.g. defects in the actin cytoskeleton, chromosome segregation and cytokinesis) are suppressed by the loss of TEA1, which suggests that Tea1 is a mediator of Pak1 polarity functions (Kim, H. et al., 2003).

In fission yeast, Pak1 also regulates a potential regulator of small GTPases, the Rho-GAP Rga8. Pak1 phosphorylates Rga8, and this phosphorylation is required for the proper localization of Rga8, which, like Pak1, is found at cell ends and in the septum-forming region. Interestingly, despite compelling evidence that Rga8 acts as a GAP for Rho1 in cells, Rga8 does not appear to act as a negative regulator of Rho1 function. Instead, these two proteins have a positive functional interaction in S. pombe cells: gain of Rga8 function exacerbates phenotypes caused by gain of Rho1 function and vice versa, and both Rho1 and Rga8 antagonize Pak1 function (Yang et al., 2003). This linkage between a member of the Pak family and a regulator of Rho is not unique to fission yeast, because two groups have recently reported that both group A and group B mammalian Paks phosphorylate and thereby alter the localization and/or activity of additional Rho-GEFs (Barac et al., 2004; Zenke et al., 2004).

Pak2, like Cla4 and Skm1 from budding yeast, contains an N-terminal PH domain (Sells et al., 1998). Unlike PAK1, the PAK2 gene is not required for viability or fertility in fission yeast (Sells et al., 1998; Yang et al., 1998). Interestingly, high-level expression of PAK2 partly suppresses the morphological defects of PAK1-null cells but does not restore mating competence. This partial complementation requires the PH domain, the PBD domain and a functional kinase domain. Overexpression of Pak2 also restores rod-shaped morphology to RAS1-deleted ovoid cells, indicating that Pak2 participates in the Ras1-dependent morphological pathway (Yang et al., 1998). Such a pathway might be analogous to the Ras/Ste20 pathway for haploinvasive growth in budding yeast discussed previously. In cross-species complementation assays, Sells et al. found that PAK2 cannot rescue cla4 morphological defects, ste20 mating defects or cla4/ste20 lethality in budding yeast. These observations suggest that, despite its structural resemblance to Cla4, Pak2 is not functionally equivalent to this kinase.

The signaling pathways downstream of Pak2 are not well defined. Overexpression of activated Pak2 leads to morphological defects (Sells et al., 1998) and these defects can be suppressed by loss of the stress-pathway kinases Mkh1 and Spm1 (Merla and Johnson, 2001). Because expression of activated Pak2 does not rescue the growth defects conferred by overexpression of Mkh1, these epistatic data place Pak2 upstream of Mkh1 in this pathway. Pak2, but not Pak1, interacts with Mkh1, which is consistent with a specific function for Pak2 in the Mkh1-Pek1-Spm1 pathway, which regulates cytokinesis and cell division in fission yeast (Merla and Johnson, 2001).

In Dictyostelium, three Pak genes have been identified, encoding myosin I heavy chain kinase (MIHCK/PakB), PakA and PakC. The proteins encoded by the first two have been studied in some detail, whereas PakC appears only as a database entry. These three kinases are more closely related to one another than to Paks in other organisms (Fig. 1).

MIHCK is activated by Rac-GTP and also by acidic lipids such as phosphatidylserine, phosphatidylinositol and phosphatidylinositol 4,5-bisphosphate. As in the case of Ste20 in budding yeast, Dictyostelium MIHCK phosphorylates and regulates the activity of myosin. MIHCK might thus link small GTPase signaling pathways to motile processes requiring myosin I molecules (Brzeska et al., 1997; Lee et al., 1998). However, a direct test of this model (e.g. by mihck gene disruption) has not been reported.

Chung and Firtel reported that PakA colocalizes with myosin II to the cleavage furrow of dividing cells and to the posterior of polarized, chemotaxing cells (Chung and Firtel, 1999). Upon disruption of cell polarity and rounding up of the cells, PakA becomes uniformly distributed along the membrane cortex of the entire cell. Chung and Firtel found that paka-null cells fail to complete cytokinesis in suspension, appear more elongated, and have a less-polarized actin cytoskeleton than wild-type cells. They also observed that PakA is required for maintaining directional cell movement: in their studies, paka-null cells or wild-type cells expressing a kinase-dead PakA mutant produced many random, lateral pseudopodia and made wrong turns at a much higher frequency than wild-type cells (Chung and Firtel, 1999). In contrast to these findings, Müller-Taubenberger et al. have found no obvious abnormal phenotype in paka-null cells nor could they confirm the localization of PakA to the cleavage furrow of dividing cells (Müller-Taubenberger et al., 2002). It is possible that these discordant results are related to the different genetic strains of Dictyostelium used. It is also possible that overexpression of full-length PakA in the Müller-Taubenberger study affected cell morphology and cleavage furrow formation such that PakA became delocalized.

In the experiments of Chung and Firtel, the distribution of PakA and the phenotype of paka-null cells were similar to those described for myosin II. Normally, Dictyostelium cells increase the level of myosin II in the cytoskeleton in response to cAMP; this response is not observed in paka-null cells, which suggests that the assembly of myosin II into the cytoskeleton requires PakA function. Because PakA is activated by cAMP, PakA might thus control chemotaxis by regulating myosin II function. paka-null cells also cannot properly retract the rear end of the cell when chemotaxing and exhibit defective myosin II assembly: the myosin II cap in the posterior of chemotaxing cells and myosin II assembly into the cytoskeleton upon cAMP stimulation are both absent in these cells. By contrast, constitutively active PakA leads to an upregulation of myosin II assembly. Interestingly, these effects on myosin II assembly do not appear to result from direct phosphorylation of myosin II by PakA; rather, PakA might indirectly regulate myosin II assembly by negatively regulating myosin II heavy chain kinase (Chung and Firtel, 1999).

PakA is phosphorylated by protein kinase B (PKB/Akt) in vitro at Thr579 between the N-terminal proline-rich motif and the PBD (Chung et al., 2001). Mutation of this site to alanine blocks PakA activation and redistribution in response to chemoattractant stimulation. Conversely, mutation of the same site to the phosphomimic aspartic acid increases the basal level of association with the cytoskeleton (Chung et al., 2001). In cells lacking PKB or phosphoinositide 3-kinase (PI 3-K), PakA does not significantly incorporate into the cytoskeletal fraction upon cAMP stimulation. This suggests that PKB/Akt and PI 3-K regulate cell polarity and chemotaxis at least in part through control of PakA activation and localization.

Drosophila encodes three Paks: one group A (DPak1), one group B (Mbt) and a third that does not fit easily into either classification (DPak3, accession number NP_650545). The functions of DPak1 and Mbt, but not DPak3, have been analyzed by genetic techniques. As will be discussed below, these studies have revealed key functions for Paks in sensory organ development, as well as in organismal morphogenesis.

DPak1 binds to the SH3/SH2 adaptor protein Nck (Dock), which is thought to facilitate its translocation to insulin receptors (InRs) at the cell membrane. Mutations in InR, dock or dpak1 result in similar errors in photoreceptor axon guidance and targeting (Garrity et al., 1996; Hing et al., 1999; Song et al., 2003). Dock and DPak1 are also necessary for the guidance of olfactory axons (Ang et al., 2003). Interestingly, Kim and co-workers showed that DPak1 is probably not involved in axon outgrowth per se but rather in axon guidance, concluding that it regulates filopodial activity in the growth cone (Kim, M. D. et al., 2003).

In addition to its role in attractive axon guidance, the Dock-DPak1 complex might also be involved in axon repulsion. Dock binds to the cytoplasmic domain of the Roundabout (Robo) receptor, and loss of Dock or DPak1 function compromises Robo-mediated repulsion (Fan et al., 2003).

The mushroom bodies tiny (mbt) gene of Drosophila encodes a group B Pak. Mbt was uncovered during a genetic screen for genes involved in the formation of the mushroom body, a structure in the adult fly corresponding to the human hippocampus that is involved in learning and memory. The mbt-null mutants have fewer neurons in the brain; this leads to a dramatic reduction in mushroom body volume, which correlates with a reduced number of Kenyon cells in this structure. It is therefore thought that Mbt has a role in cell proliferation, differentiation or survival of these neuronal cells (Melzig et al., 1998).

In addition to mushroom body defects, mbt mutant flies display loss of a variable number of photoreceptor cells (R-cells) in many ommatidia of the eye, whereas the innervation pattern in the medulla appears to be normal (Schneeberger and Raabe, 2003). Thus, unlike DPak1, Mbt is believed to be involved in photoreceptor cell morphogenesis rather than photoreceptor axon guidance. Schneeberger and Raabe showed that Mbt specifically localizes to adherens junctions of photoreceptor cells. In mbt mutants, the rhabdomers of the differentiated photoreceptor cells show severe morphological defects, with disorganized adherens junctions. As in the case of other group B Paks, binding of activated Cdc42 has no influence on Mbt kinase activity but instead is required for recruitment of the kinase to adherens junctions. Interestingly, although the PBD appears to be essential for the in vivo function of Mbt, kinase activity is not absolutely required, since a kinase-dead Mbt mutant can partially rescue the mbt mutant eye phenotype (Schneeberger and Raabe, 2003).

Mammals encode six Paks: three group A and three group B (Fig. 1A). As in simpler eukaryotes, Paks in mammalian cells regulate MAPK pathways and cytoskeletal organization (Fig. 4). The connection to the ERK MAPK pathway is particularly interesting, because Paks appear to phosphorylate both Raf1 (at Ser338) and Mek1 (at Ser298); that is, they behave both as MAPKKKKs and as MAPKKKs. These phosphorylations are in themselves not sufficient to activate Raf1 or Mek1, but are required for the activation of these kinases by Ras and Raf, respectively (Frost et al., 1997; King et al., 1998). Paks also activate the stress-activated JNK and p38 MAPK cascades, but these effects are modest in most cell types (Bagrodia et al., 1995; Brown et al., 1996).

Microinjection of activated Pak1 protein into quiescent Swiss 3T3 cells induces the rapid formation of lamellipodia, filopodia and membrane ruffles (Sells et al., 1997), which is similar to the effect produced by microinjection of Cdc42 (Nobes and Hall, 1995). Expression of various constitutively active forms of Pak1 induces disassembly of stress fibers and focal adhesion complexes (Manser et al., 1997; Sells et al., 1997). The basis for these activities is not completely understood, but involves the phosphorylation of multiple substrates that affect cytoskeletal structure, including LIM kinase, myosin light chain kinase, merlin, filamin, p41^Arc, Rho-GEFs and stathmin. Paks also phosphorylate the apoptotic protein BAD and the estrogen and progesterone hormone receptors (reviewed by Bokoch, 2003) (Fig. 4). It is important to note that, as in Drosophila, some of the signaling functions of mammalian Paks, particularly certain cytoskeletal effects, are independent of its kinase function (Daniels et al., 1998; Frost et al., 1998; Sells et al., 1999; Sells et al., 1997). For example, Pak1 plays a key role in the polarization of chemotaxing cells; however, it appears that the main role of Pak1 in this context is not as a kinase per se but as a scaffold protein, bringing the GEF PIX alpha to the plasma membrane, where it encounters and activates Cdc42 (Li et al., 2003). It is also becoming apparent that mammalian Paks have GTPase-independent functions. For example, Pak1 can be directly activated by Akt (Tang et al., 1999; Tang et al., 2000), and Pak2 can be activated by cleavage by caspase 3 (Lee et al., 1997; Rudel and Bokoch, 1997).

The genetic analysis of Pak function in mammals is ongoing. In mice, the Pak1, Pak2, Pak4 and Pak5 genes have been disrupted (Table 1). Pak1- and Pak5-null mice are viable and healthy, whereas loss of Pak2 or Pak4 results in embryonic lethality. In humans, mutations in the Pak3 gene are associated with an X-linked mental retardation syndrome (see below).

Pak3 is the only member of the Pak family known to be associated with a human genetic disease. Mutations in the Pak3 gene are associated with X-linked, nonsyndromic mental retardation (MRX) syndromes, which are forms of mental retardation accompanied by grossly normal brain development and few other signs or symptoms.

The cause of MRX30 is a mutation in pak3 that generates a truncated, kinase-dead mutant (R419Stop) (Allen et al., 1998). Two more independent MRX kindreds have been found to have pak3 mutations. One of these (R67C mutation) is expected to affect GTPase binding, whereas the other (A365E mutation) affects a highly conserved region within subdomain VIA of the protein kinase domain (Bienvenu et al., 2000; Gedeon et al., 2003). Affected males bearing this latter mutation show mild-to-borderline mental retardation with no additional clinical manifestations other than mental impairment and relatively long ears, and occasionally neuropsychiatric problems.

The absence of severe brain defects in these MRX patients suggests that Pak3 function is not absolutely required for neuronal proliferation, migration or cortical gyration. However, the observation that pak3 mutations result in mental retardation might reflect a later requirement for Pak3 function in the adult cortex. Perhaps, by analogy with the function of DPak1 in Drosophila, Pak3 is necessary for the normal development of axonal connections and hence there are aberrant or absent axonal connections in these MRX patients. Alternatively, because Rac and Pak signaling also appears to be important for dendritic spine morphogenesis, Pak3 might be necessary for dendritic development or for the rapid cytoskeletal reorganizations in dendritic spines associated with synaptic plasticity (Park et al., 2003; Penzes et al., 2003). Pak3 might have additional functions in the adult brain that involve its interaction with the amyloid precursor protein. Indeed, it might be one of the proteins that mediate the increased DNA synthesis and apoptosis found in neuronal cells of patients with familial Alzheimer's disease (McPhie et al., 2003). Through interference with these and/or similar pathways, mutation of pak3 might cause mental retardation.

To date, a pak3-knockout mouse model has not been reported. Considering the phenotype of the MRX patients, the expression pattern of mouse pak3 and pathways in which the Pak3 protein is known to be involved, one might expect a pak3-knockout mouse to develop normally, but to show some brain abnormalities and/or signs of memory and learning defects. Such a model would be invaluable as a tool to investigate the molecular basis for cognitive function. Indeed, loss of all group A Pak function in the brain, induced by transgenic expression of a Pak-inhibitory peptide in mice, induces aberrant synaptic morphology in cortical neurons and is associated with defects in hippocampus-dependent, long-term memory consolidation (Hayashi et al., 2004).

Pak4 was the first member of the group B Paks to be identified (Abo et al., 1998). It binds preferentially to activated Cdc42 and promotes filopodium formation in response to activated Cdc42 in fibroblasts and other cell types (Abo et al., 1998; Qu et al., 2001). Expression of activated Pak4 in fibroblasts decreases adhesion to the extracellular matrix and promotes proliferation, leading to anchorage-independent growth and increased cell migration (Callow et al., 2002; Qu et al., 2001). Pak4 substrates include the cytoskeletal regulatory kinase LIMK1, the pro-apoptotic protein BAD and probably PDZRho-GEF (Barac et al., 2004; Dan, C. et al., 2001; Gnesutta et al., 2001).

Mouse pak4-null embryos die around embryonic day 10.5 (Qu et al., 2003). The most likely cause of death is a cardiac defect, although its basis is not yet known. Pak4-null embryos also show severely abnormal development and migration of neurons. Neuronal progenitors form normally, but differentiation of these cells is mostly inhibited, axonal outgrowth is impaired, and neurons do not migrate to their correct target areas. Neuroepithelia of pak4-mutant embryos are abnormally thin in the hindbrain and forebrain, which results in the appearance of a nearly translucent head and neural tube. Neuronal differentiation and migration is also defective in motor neurons of the developing spinal cord and neural tube. In addition to these abnormalities, the neural tube in these mice is improperly folded (Qu et al., 2003). The development of the neural tube depends on proper cell migration and therefore dynamic cytoskeletal reorganization in these cells. Interference with the underlying signaling pathways may therefore lead to defects in neural tube development. Previous studies have shown that Pak4 is indeed involved in regulation of focal adhesions and filopodia, both of which are structures that are important for cell migration (Callow et al., 2002; Qu et al., 2001). Neural tube defects have previously been found in mice lacking other proteins that are involved in the regulation of the cytoskeleton (Brouns et al., 2000; Snapper et al., 2001).

Pak4-deficient cultured cells have increased numbers of focal adhesions under serum-starved conditions, whereas cells overexpressing activated Pak4 have fewer focal adhesions compared with wild-type cells (Qu et al., 2001; Qu et al., 2003). Therefore, increased adhesion of neuronal cells might contribute to the observed defects in the knockout mice.

Like Pak3, Pak5 is expressed preferentially in the brain (Pandey et al., 2002). Pak5 has a high basal kinase activity that is not significantly increased by binding Cdc42, and overexpression of Pak5 activates the JNK MAPK pathway but not the p38 or ERK MAPK pathways (Dan et al., 2002; Pandey et al., 2002). In N1E-115 cells, overexpression studies suggest that Pak5 has a role in filopodium formation and neurite outgrowth, and probably acts downstream of Cdc42 and Rac (Dan et al., 2002). In common with other family members phosphorylating BAD, Pak5 is thought to have a role in the regulation of apoptosis (Cotteret et al., 2003). In CHO and neuronal HMN1 cells, overexpressed Pak5 localizes at mitochondria, independently of Cdc42 binding or its kinase activity (Cotteret et al., 2003).

Li and Minden have reported that pak5-null mice develop normally and are fertile. Although Pak5 is expressed at high levels in neurons in the brain and the eye, all parts of the nervous system studied develop normally in the mice, and histological analysis showed no obvious differences between them and wild-type mice (Li and Minden, 2003). One possible explanation is functional redundancy, because all members of the Pak family are expressed in the central nervous system. Although other members of the Pak family were not found to be upregulated in whole brain extracts of 8-week-old pak5-knockout mice (Li and Minden, 2003), it is nevertheless possible that these can compensate for the deficiency in Pak5. Interestingly, the expression patterns of Pak5 and Pak6 are very similar (Lee et al., 2002; Li and Minden, 2003). In addition, Pak6 mRNA levels in the adult mouse brain are significantly higher than those of Pak4 (Li and Minden, 2003). Although Pak5 does not appear to be required for brain development, it may yet prove to be important for cognitive function of the adult brain, as is thought to be the case for Pak3. Detailed behavioral and learning/memory studies of the pak5 knockouts could reveal such defects, if they indeed exist.

Paks are named for their ability to bind and become activated by small GTPases, and their functions are usually considered in relation to their roles as GTPase effectors. It is clear from the wealth of biochemical, cell biological and genetic data that the regulation of MAPK cascades and regulation of cytoskeletal organization are ancient and conserved functions of Paks. However, the mechanisms by which they carry out these functions differ considerably, and the existence of kinase-independent effects further complicates the analysis of Pak signaling. It is also becoming evident that Paks shoulder an increasingly heavy signaling burden with increased organismal complexity and that, in higher eukaryotes, some of these signaling tasks are independent of small GTPase function. In other words, Paks have evolved into something more than simple GTPase effectors. A vigorous and systematic search for such GTPase-independent activities is likely to yield new insights, and the ongoing genetic analysis of Paks in mammalian systems should help clarify the cellular and developmental signaling pathways regulated by this fascinating family of protein kinases.

Work in the Chernoff lab is funded by the National Institutes of Health and the American Cancer Society, as well as by a center-wide CORE grant and an appropriation from the Commonwealth of Pennsylvania. We are grateful to our many colleagues who provided information and insights into Pak function, as well as E. Golemis and J. Peterson for comments on the manuscript. We regret the inevitable omissions that were necessary to keep to the designated word limits.