The subcellular localization of type I p21-activated kinases is controlled by the disordered variable region and polybasic sequences

Abstract

p21-activated kinases (PAKs) are serine/threonine kinase effectors of the small GTPases Rac and Cdc42 and major participants in cell adhesion, motility, and survival. Type II PAKs (PAK4, -5, and -6) are recruited to cell–cell boundaries, where they regulate adhesion dynamics and colony escape. In contrast, the type I PAK, PAK1, does not localize to cell–cell contacts. We have now found that the other type I PAKs (PAK2 and PAK3) also fail to target to cell–cell junctions. PAKs contain extensive similarities in sequence and domain organization; therefore, focusing on PAK1 and PAK6, we used chimeras and truncation mutants to investigate their differences in localization. We observed that a weakly conserved sequence region (the variable region), located between the Cdc42-binding CRIB domain and the kinase domain, inhibits PAK1 targeting to cell–cell junctions. Accordingly, substitution of the PAK1 variable region with that from PAK6 or removal of this region of PAK1 resulted in its localization to cell–cell contacts. We further show that Cdc42 binding is required, but not sufficient, to direct PAKs to cell–cell contacts and that an N-terminal polybasic sequence is necessary for PAK1 recruitment to cell–cell contacts, but only if the variable region–mediated inhibition is released. We propose that all PAKs contain cell–cell boundary–targeting motifs but that the variable region prevents type I PAK accumulation at junctions. This highlights the importance of this poorly conserved, largely disordered region in PAK regulation and raises the possibility that variable region inhibition may be released by cellular signals.

Introduction

p21-activated kinases (PAKs)³ are a family of six serine/threonine kinases with key roles in cell adhesion, migration, and survival (1–3). The founding member of the family (PAK1) was first identified in a screen for interactors of GTP-bound Cdc42/Rac (4), and all six PAKs are now known to be Cdc42/Rac effectors (5–10). PAKs have been categorized into two separate groups according to their sequence similarities, with PAK1, -2, and -3 in type I and PAK4, -5, and -6 in type II (11–13). All PAKs have a highly conserved Cdc42/Rac interactive binding (CRIB) domain near their N terminus and a well-conserved C-terminal kinase domain. In type I PAKs, formation of an autoinhibitory dimer is thought to limit kinase activity and be released after Cdc42/Rac binding to the CRIB domain induces a conformational change that releases the autoinhibition and activates the kinase autophosphorylation signaling cascade (14, 15). In contrast, type II PAKs do not form a similar autoinhibitory dimer, and Cdc42/Rac binding is generally not considered to activate type II PAK kinase activity, although this remains somewhat controversial (2, 16, 17). Instead, type II PAK activity is regulated by intramolecular associations between the kinase domain and a pseudosubstrate sequence, located C-terminal to the CRIB domain (16–18). It remains unknown how this interaction is released (2, 11, 16). Despite having little effect on kinase activity, GTPase binding is, however, important for regulating PAK function through control of PAK subcellular localization (19, 20).

PAKs play crucial kinase-dependent and -independent roles in many cellular signaling pathways involved in adhesion, migration, and survival, and dysregulation of PAK function or expression contributes to a range of diseases including cancer (21). Unsurprisingly, given the sequence conservation in their kinase domains, type I and II PAKs share many substrates (e.g. β-catenin, paxillin, and BAD); however, type I– and type II–specific substrates have been reported (22). Both type I and type II PAKs have been implicated in cancer development, and notably, both PAK1 and PAK6 are linked to prostate cancer metastasis (21, 23, 24).

We and others previously showed that PAK6 localizes at cell–cell contacts in the DU145 prostate cancer cell line and that this localization, along with kinase activity, drives escape of DU145 cells from cell colonies (19, 20). In cells expressing GFP-PAK6, we observed a significant increase in the percentage of cells escaping from colonies when compared with control GFP cells (19, 20). In our study, we identified residues 1–48, containing an N-terminal polybasic motif and the CRIB domain, as the minimal sequence sufficient to direct PAK6 to cell–cell contacts. We also demonstrated that knockdown of Cdc42 blocked PAK6 localization at cell–cell contacts (20). Furthermore, we found that both kinase activity and localization to cell–cell contacts were necessary to drive colony escape (20). PAK6 has been reported to form a complex with β-catenin (19), and PAK6 can directly phosphorylate β-catenin in vitro (19). The phosphorylation state of β-catenin regulates adherens junction integrity, and phosphorylated β-catenin is released from E-cadherin, which results in decreased cell–cell adhesions (25). We hypothesize that PAK6 phosphorylation of adherens junction components, such as β-catenin (19), disrupts cell–cell contacts and favors cell escape from colonies (19, 26). Notably, although we find that all type II PAKs target to cell–cell boundaries to varying extents, the type I PAK, PAK1, does not. This differential localization is surprising, considering that PAK1 contains an N-terminal polybasic sequence and a CRIB domain, and PAK1 and PAK6 have shared substrates and both contribute to cancer development. The question of what causes the differential PAK localization is the focus of this study.

Results

Type II but not type I PAKs target to cell–cell contacts

We previously reported that PAK6, a type II PAK, targets to cell–cell contacts in DU145 prostate cancer cells, but PAK1, a type I PAK, does not (20). Furthermore, the other type II PAKs, PAK4 and PAK5, also target to cell–cell contacts, albeit PAK4 targeting is weak (19, 20, 27). Based on these data, we suggested that type I and type II PAKs exhibit differential targeting to cell–cell contacts. To test this hypothesis, we assessed localization of all three type I PAKs using N-terminal GFP-tagged PAK1, PAK2, and PAK3 in DU145 cells. Using immunofluorescent staining of β-catenin as a marker of cell–cell junctions, we showed that type I PAKs fail to target at cell–cell contacts and display a diffuse localization pattern much like the GFP control (Fig. 1A). In contrast, GFP-PAK6 clearly targets to cell–cell contacts (Fig. 1A). The lack of type I PAK targeting was not due to proteolysis, as immunoblotting confirmed that the GFP-tagged PAKs were of the expected molecular weights (Fig. 1B). As described previously (20), we quantified cell–cell targeting using the JACoP plugin (28) on ImageJ to calculate a Mander's colocalization coefficient, the fraction of GFP signal colocalizing with β-catenin. The quantification confirmed that GFP-tagged type I PAKs or GFP alone failed to colocalize with β-catenin, producing a low Mander's coefficient with very small S.D., whereas GFP-PAK6 exhibits a higher Mander's coefficient with a wider distribution probably due to the differences in total GFP-PAK6 expression level (Fig. 1C). Thus, we demonstrated that the type I and type II PAKs exhibit differential targeting to cell–cell contacts, but the molecular basis for this was unknown.

Figure 1.

Type II but not type I PAKs target to cell–cell contacts. A, DU145 cells transiently expressing GFP-tagged type I PAKs (PAK1, PAK2, or PAK3) or PAK6 were plated on glass coverslips, fixed after 24 h, and stained with antibody to β-catenin. Scale bar, 10 μm. B, 5 × 10⁶ DU145 cells transiently transfected with GFP-PAK constructs were lysed in buffer X lysis buffer, diluted in sample buffer, and immunoblotted (IB) for GFP. C, colocalization of GFP and GFP-PAKs with β-catenin was assessed by calculating the Mander's coefficient. At least 15 nonoverlapping frames from at least three independent transfections were used in the analysis. Mean ± S.D. (error bars) is indicated. n.s., not significant; ****, p < 0.0001 in an ordinary one-way ANOVA test with Dunnett's correction for multiple comparisons.

We have been unable to identify anti-PAK antibodies that allow immunofluorescent localization of endogenous PAK1 or PAK6. Therefore, to ensure that the differential targeting we observe is not solely a consequence of inserting an N-terminal GFP tag, we also compared PAK1 and PAK6 containing C-terminal mCherry. As shown in Fig. S1A and quantified by Mander's colocalization coefficient (Fig. S1B), PAK6-mCherry localized at cell–cell boundaries, whereas PAK1-mCherry did not. Furthermore, to examine whether the size of the tag influenced localization, we generated C-terminally Myc-tagged PAK1 and PAK6. Again, PAK1-Myc did not localize to cell–cell boundaries, whereas PAK6-Myc colocalized with β-catenin (Fig. S1 (C and D)). We note that PAK6-Myc co-localized with β-catenin with an overall lower Mander's coefficient, and we suspect that this may reflect difficulties with accurate thresholding of images due to the higher background staining obtained with the anti-Myc tag antibody.

Having established that differential localization of PAK1 and PAK6 is independent of the site and size of the tag, we tested whether this trend also occurred in cells other than DU145. Our prior studies used the DU145 prostate cancer cell line, as these cells are known to express PAK1 and PAK6 (20). When GFP-PAK1 or GFP-PAK6 was transiently expressed in HeLa or U2OS cells, GFP-PAK6 robustly co-localized with β-catenin at cell–cell boundaries, whereas GFP-PAK1 did not (Fig. S1 (E and F)). Differential PAK targeting is thus not restricted to DU145 cells.

The variable region contributes to the PAK1 localization difference

The domain architecture of PAK proteins is well-conserved with an N-terminal unstructured region that often contains a polybasic motif, followed by a CRIB domain, a central poorly conserved region of variable length that is believed to be largely unstructured (which we term the variable region), and a C-terminal highly conserved kinase domain (Fig. 2A). We have previously demonstrated that PAK6 targeting to cell–cell contacts depends on the Cdc42-binding CRIB domain and the N-terminal residues containing the polybasic sequence (20). The conservation between PAK isoforms allowed us to use chimeras to investigate the domains responsible for the differential targeting between PAK1 and PAK6. We first generated chimeric constructs swapping the N-terminal and CRIB domains (Fig. 2A). Using fluorescent microscopy with β-catenin as a marker for cell–cell junctions, we showed that a construct containing the N-terminal and CRIB domains from PAK6 (residues 1–51) plus the PAK1 variable region (residues 108–248) and PAK1 kinase domain (residues 249–553), which we term GFP-611, efficiently targets to cell–cell contacts (Fig. 2B). This was consistent with our prior data showing that the minimal sequence needed for PAK6 cell–cell contact targeting is the N-terminal plus CRIB regions (20). However, more surprisingly, the converse chimera, GFP-166, containing the N-terminal and CRIB regions from PAK1 (residues 1–107) and the variable region (residues 52–382) and kinase domain (residues 383–681) from PAK6, also localizes at β-catenin–positive cell–cell boundaries (Fig. 2B). All chimeras were expressed at the correct molecular weight (Fig. 2C), and calculation of Mander's coefficients from many fields in independent transfections showed comparable targeting of GFP-611 and GFP-166 chimeras and of GFP-PAK6, all of which are significantly higher than the PAK1 control (Fig. 2D).

Figure 2.

The variable region influences PAK1 localization. A, E, and I, schematic representations of the GFP-tagged PAK1/PAK6 chimeras used. The amino acid boundaries of the polybasic motif (PB) plus CRIB domain region, the VR, and kinase domain (Cat) are indicated. All GFP tags are positioned at the N terminus of the constructs. PAK1 is shown in yellow, and PAK6 is shown in blue. B, F, and J, DU145 cells were transiently transfected with GFP-tagged chimeric constructs on glass coverslips. 24 h later, cells were fixed and stained with antibody to β-catenin. Scale bar, 10 μm. C, G, and K, immunoblotting (IB) of transfected DU145 cell lysates for GFP. D, H, and L, colocalization of GFP-PAK chimeras with β-catenin was assessed by calculating the Mander's coefficient. At least 20 nonoverlapping frames from at least three independent transfections were used in the analysis. Mean ± S.D. (error bars) is indicated. n.s., not significant; ****, p < 0.0001 in an ordinary one-way ANOVA test with Dunnett's correction for multiple comparisons.

Knowing that the GFP-166 chimera targets to cell–cell contacts when having the variable region and kinase domain from PAK6, we set out to investigate which of these two regions facilitates targeting. We therefore generated additional PAK1 chimeras containing the PAK6 kinase domain (GFP-116) and the PAK6 variable domain (GFP-161) (Fig. 2E). As expected, given the highly conserved PAK kinase domains, swapping only the kinase domain had no impact on targeting (Fig. 2F), and GFP-116 co-localization with β-catenin was not significantly different from PAK1 control (Fig. 2H). Because GFP-166 is a targeting construct, this suggested that the variable region is likely to be responsible. Indeed, when we transiently transfected the GFP-161 construct, containing the PAK6 variable region, it targeted to cell–cell boundaries (Fig. 2F) at levels comparable with the GFP-166 chimera (Fig. 2H). Immunoblot of the lysates showed that all constructs were expressed at the expected molecular weight; however, GFP-161 was consistently less well-expressed (Fig. 2G). Notably, this reduction in expression did not alter our results, as we selected cells with comparable expression levels for imaging and analysis. For completeness, we also generated a complementary set of PAK6 chimeras, containing PAK1 kinase domain (GFP-661), variable region (GFP-616), or both (GFP-611) (Fig. 2I). Knowing that the N-terminal regions and CRIB domain are minimal essential domains for PAK6 targeting to cell–cell contacts (20), we were not surprised to find that all of these chimeras localized to cell–cell contacts (Fig. 2, J and L). Collectively, our results with these chimera sets establish that the variable region of PAK6 enables PAK1 targeting to cell–cell contacts, whereas the presence of the PAK1 variable region and/or the kinase domain cannot stop PAK6 from localizing to the cell–cell boundaries.

PAK1 variable region inhibits PAK1 from targeting to cell–cell junctions

Our data provide the first evidence that the variable region plays a role directing the localization of PAKs to cell–cell contacts. We considered two alternative hypotheses to explain this effect: the PAK6 variable region may be sufficient to drive the chimera to junctions, or removing the PAK1 variable region may release an inhibitory process that keeps PAK1 away from cell–cell boundaries. To distinguish these two possibilities, we generated PAK1 truncation constructs to delete the kinase domain (GFP-PAK1ΔCat) or both the kinase domain and the variable region (GFP-PAK1(1–107)). If there is an inhibitory effect mediated solely by the variable region, we expect to see the construct targeting to cell–cell contacts only when we remove the variable region. Supporting our inhibition hypothesis, we observed that whereas GFP-PAK1ΔCat occasionally displayed increased membrane targeting, it did not colocalize with β-catenin, but upon the removal of the variable region, the truncation mutant GFP-PAK1(1–107) robustly localized at cell–cell boundaries (Fig. 3A). Whereas all truncation constructs were expressed at the correct size (Fig. 3B), Mandar's coefficient analysis showed significantly more GFP-PAK1(1–107) colocalizing with β-catenin compared with GFP-PAK1 and GFP-PAK1ΔCat. These data suggest that the variable region of PAK1 prevents targeting to cell–cell contacts. Notably, we also made truncations deleting the kinase domain and variable regions of PAK2 and PAK3, and, similar to what we observed for PAK1, whereas full-length PAK2 and PAK3 do not target to junctions, GFP-PAK2(1–108) and GFP-PAK3(1–109) target very well (Fig. 3D). We note that some of the truncation mutants appeared to also localize to the nucleus, and we hypothesize that this is due to their smaller size compared with full-length PAKs. GFP is also sometimes observed in the nucleus. Nonetheless, these constructs support the conclusion that the N-terminal and CRIB domain regions of type I PAKs can target to cell–cell boundaries, and the variable region inhibits targeting.

Figure 3.

PAK1 variable region inhibits PAK1 from targeting to cell–cell junctions. A, DU145 cells were transiently transfected with GFP-tagged PAK1, PAK1 truncation mutants, or PAK6 on glass coverslips. 24 h later, cells were fixed and stained with antibody to β-catenin. Scale bar, 10 μm. B, immunoblotting (IB) of transfected DU145 cell lysates for GFP. C, colocalization of GFP-PAK1 truncations with β-catenin was assessed by calculating the Mander's coefficient in at least 20 nonoverlapping frames from at least three independent transfections. Mean ± S.D. (error bars) is indicated. n.s., not significant; ****, p < 0.0001 in an ordinary one-way ANOVA test with Dunnett's correction for multiple comparisons. D, localization of GFP-tagged PAK2 and PAK3 or their truncation mutants was assessed as in A. E, colocalization of GFP-PAK2/3 truncations with β-catenin was assessed by calculating the Mander's coefficient. At least 10 nonoverlapping frames from at least two independent transfections were used in the analysis. Mean ± S.D. is indicated. ****, p < 0.0001 in an ordinary one-way ANOVA test with Tukey's multiple-comparison test.

We next investigated potential mechanisms by which the PAK1 variable region might inhibit accumulation of PAK1 at cell–cell contacts. Whereas the function of the variable region is unclear, and it is largely unstructured in PAK1 crystal structures (29), a binding site for the PIX family of nucleotide exchange factors lies in the PAK1 variable region (30, 31), and PIX binding is proposed to either directly regulate PAK kinase activity or modulate PAK activity by recruiting it to focal adhesions (31, 32). Notably, PIX does not bind to PAK6, and a PAK1 P191G/R192A double mutant has been shown to disrupt PIX binding (30). We therefore investigated whether PIX binding to PAK is responsible for the inhibitory effect of cell–cell boundary targeting. Using fluorescent microscopy, we showed that, like GFP-PAK1, GFP-PAK1 (P191G/R192A) fails to target to cell–cell contacts (Fig. 4, A and B), suggesting that PIX binding does not alter PAK1. We therefore attempted to identify additional binding partners of the PAK1 variable region. We generated GST-tagged constructs of the PAK1 variable region (VR), and the variable region with P191G/R192A mutations (VR-GA), expressed these proteins in Escherichia coli and purified them by GSH affinity chromatography. The purified GST-tagged proteins were loaded on GSH beads and used to pull down binding partners from DU145 cell lysates. Coomassie staining of bound proteins revealed two bands of molecular mass ∼100 kDa, clearly visible in the GST-VR lane only when mixed with DU145 lysates and absent from GST control lanes (Fig. 4C). As the binding of both of bands was abolished in the P191G/R192A double mutant, we anticipated the ∼90 kDa band to be β-PIX, and we confirmed this by immunoblotting with anti-β-PIX antibodies (Fig. 4D). To identify the upper band, we excised the band and analyzed it by MS. This indicated it to be the known PIX binding partner GIT1 or GIT2, GTPase-activating proteins for ADP-ribosylation factor small GTP-binding proteins (33). Immunoblotting with anti-GIT antibodies confirmed our mass spectrometry result (Fig. 4D). However, as the P191G/R192A double mutation did not impact PAK1 targeting to cell–cell contacts (Fig. 4A), we concluded that PIX and GIT were not inhibiting PAK1 targeting. No other variable region binding partners were evident from our pulldown assays, raising the possibility that the variable region may inhibit PAK1 targeting by altering PAK1 conformation rather than through binding proteins.

Figure 4.

Neither PIX binding nor the inhibitory switch region inhibits PAK1 from localizing to cell–cell contacts. A, DU145 cells were transiently transfected with GFP-tagged PAK1 or the PAK1 (P191G/R192A) double mutant (PAK1-GA) on glass coverslips. 24 h post-transfection, cells were fixed and stained with antibody to β-catenin. Scale bar, 10 μm. B, colocalization of GFP-PAK1 and GFP-PAK1-GA with β-catenin was assessed by calculating the Mander's coefficient. At least 15 nonoverlapping frames from at least three independent transfections were used in the analysis. Mean ± S.D. (error bars) is indicated. An unpaired, two-tailed Student's t test was performed. n.s., not significant. C, bacterially purified GST, GST-tagged variable region of PAK1 (VR), or GST-VR containing the P191G R192A double mutation (VR-GA) loaded on GSH beads was incubated with or without DU145 cell lysates, fractionated by SDS-PAGE, and stained with Coomassie Blue. The result shown is representative of seven independent experiments. Bands of interest are marked with an arrowhead. D, samples indicated were taken from the experiment shown in C, analyzed on a separate gel, and immunoblotted (IB) for PIX or GIT. E, localization of GFP-PAK1 or the truncation mutant PAK1(1–149), which includes the inhibitory switch region and the kinase inhibitor segment, was assessed as in A. F, colocalization of GFP-PAK1 and mutant with β-catenin was assessed by calculating the Mander's coefficient. At least 15 nonoverlapping frames from at least three independent transfections were used in the analysis. Mean ± S.D. is indicated. ****, p < 0.0001 in an ordinary one-way ANOVA with Tukey's multiple-comparison test.

Structural studies have shown that unstimulated PAK1 exists as a trans-inhibited dimer where the kinase domain of one PAK1 molecule is inhibited through interactions with the regulatory portion of the other PAK1 molecule and vice versa (14, 29). The regulatory portions include the inhibitory switch (residues 87–136) and the kinase inhibitor segment (residues 137–149), and together they bind the C-lobe of the kinase domain, stabilizing its inactive conformation (29). The inhibitory switch region overlaps with the CRIB domain and the start of the variable region, whereas the kinase inhibitor segment lies in the variable region. Conformational changes in these regions are associated with PAK1 activation and conceivably influence targeting to cell–cell contacts. However, deletion of the kinase domain (GFP-PAK1ΔCat), which would disrupt dimer formation, did not support localization to cell–cell contacts (Fig. 3A). Furthermore, a truncation mutant (GFP-PAK1(1–149)) that includes the entire inhibitory switch and the kinase inhibitor segment colocalized with β-catenin (Fig. 4, E and F) at levels comparable with that seen for GFP-PAK1(1–107), suggesting that residues from 115 to the end of the variable region suppress targeting through a yet to be determined mechanism.

High Cdc42 binding is necessary for targeting

Previously, we showed that Cdc42 binding to the PAK6 CRIB domain is important to recruit PAK6 to cell–cell boundaries, as CRIB domain mutations as well as Cdc42 knockdown blocked PAK6 localization, with the caveat that knockdown cells exhibit disrupted cell–cell junctions (20). We therefore investigated whether the differential targeting of different PAKs are correlated with their ability to bind Cdc42. We bacterially purified GST-tagged Cdc42, a nucleotide-binding defective mutant T17N as negative control, and a constitutively active binding mutant Q61L as positive control. All of the Cdc42 constructs encompassed residues 1–177 but lacked the C-terminal prenylation motif to facilitate expression and purification. We used these proteins in pulldown experiments with lysates from DU145 cells transiently transfected with either GFP-PAK1 or GFP-PAK6. As expected based on our prior studies (20), Cdc42 pulled down GFP-PAK6, and binding was specific as the T17N mutant inhibited binding (Fig. 5A). The activating Q61L mutant also bound GFP-PAK6 at levels comparable with the WT Cdc42. Notably, whereas we also observed specific GFP-PAK1 binding to Cdc42, GFP-PAK1 bound Cdc42 substantially less well than GFP-PAK6 (Fig. 5A). We repeatedly observed on average 3 times more PAK6 than PAK1 binding to Cdc42 (Fig. 5B), consistent with the notion that PAKs that target to junctions bind better to Cdc42 (Fig. 5, A and B). To ensure that the Cdc42 binding differences are not due to any variation in expression levels of GFP-PAK1 or GFP-PAK6, we generated binding curves by serially diluting the amount of GFP-PAK input lysate with untransfected lysates to ensure that total protein levels were the same while decreasing the amount of GFP-PAK present in the lysate. When we plotted the amount of PAK versus the amount of PAK bound to WT Cdc42, we see a robust difference between GFP-PAK1 and GFP-PAK6 at most input concentrations (Fig. 5C), confirming that PAK1 binds less well to Cdc42. We have been unable to purify full-length bacterially expressed PAK6 to confirm this result quantitatively, but experiments assessing binding of GST-Cdc42 to GFP-PAK1 or GFP-PAK6 affinity-purified from cell lysates using anti-GFP nanotrap beads (34) supported the conclusion that Cdc42 binds to GFP-PAK6 much better than GFP-PAK1 or GFP (Fig. S2).

Figure 5.

High Cdc42 binding is necessary for targeting. A, pulldown of GFP-tagged PAK1 or PAK6 from transiently transfected HEK293T cell lysates by bacterially purified GST-Cdc42, GST-Cdc42 T17N, or GST-Cdc42 Q61L. Samples were immunoblotted (IB) for GFP, and 3% input is shown. Ponceau staining was used to assess equal bead loading. B, PAK binding was quantified from four independent experiments (mean ± S.D. (error bars)). *, p < 0.05 in a paired, two-tailed Student's t test. C, representative binding curve of GFP-tagged PAK1 or PAK6 pulled down by GST-Cdc42. Input amounts were serially diluted with untransfected 293T cell lysates. D and E, Cdc42 binding of GFP-tagged PAK1, PAK1(1–107), or constructs containing inactivating H83L/H86L double mutations in the PAK1 CRIB domain (HHLL) was assessed as in A and quantified as in B from seven independent experiments. ***, p < 0.001. F, DU145 cells on glass coverslips were transiently transfected with GFP-tagged PAK1 mutants, and 24 h later, cells were fixed and stained for β-catenin. Scale bar, 10 μm. G, GFP-PAK colocalization with β-catenin was assessed by calculating the Mander's coefficient. At least 20 nonoverlapping frames from at least three independent transfections were used in the analysis. Mean ± S.D. is indicated. ****, p < 0.0001 in an unpaired, two-tailed Student's t test.

We next tested our gain-of-function GFP-PAK1(1–107) truncation construct, and this also showed on average a 2.5-fold increased Cdc42 binding over full-length PAK1 (Fig. 5, D and E). We further confirmed the specificity of this interaction by showing that a double H83L/H86L) mutation in the Cdc42-binding site in the PAK1 CRIB domain (PAK1 HHLL) abolished binding to Cdc42 (Fig. 5D). Importantly, this also inhibited the ability of GFP-PAK1(1–107) to target to cell–cell junctions (Fig. 5, F and G).

A polybasic motif is necessary for PAK1 targeting once the inhibition from the variable region is released

The preceding data show that Cdc42 binding is required for PAK1 targeting to cell–cell contacts and that targeting constructs bind high levels of Cdc42; however, they do not establish whether strong Cdc42 binding is sufficient to target PAKs to cell–cell contacts. As the CRIB domain is the minimal Cdc42-binding domain, we generated GFP-tagged PAK1 CRIB domain (amino acids 74–107) and performed pulldown experiments with GST-Cdc42 proteins. As expected, GFP-PAK1 CRIB bound strongly and specifically to Cdc42 at levels comparable with GFP-PAK6 and higher than full-length GFP-PAK1 (Fig. 6A). However, GFP-PAK1 CRIB failed to target to cell–cell contacts when transiently transfected into DU145 cells (Fig. 6B). Comparison with GFP-PAK1(1–107), which includes the N-terminal region plus the CRIB and targets effectively to cell–cell contacts (Fig. 6C), suggests that the N-terminal region plays a role in GFP-PAK1(1–107) targeting. The N-terminal region of PAK6 (amino acids 1–10) is also important for PAK6 targeting (20), but this region is much longer in PAK1. Whereas PAK6 has a polybasic lysine-rich motif adjacent to CRIB domain, the PAK1 N-terminal region contains a polybasic stretch of four lysines more N-terminal from the CRIB domain and a KEKE sequence adjacent to the CRIB domain corresponding to the position of the PAK6 polybasic motif. To test the importance of the PAK1 N-terminal polybasic region, we first generated a construct containing PAK1 CRIB and 10 amino acids upstream, GFP-PAK1(64–107), a size equivalent to the PAK6 N-terminal region. PAK1(64–107) failed to target to cell–cell boundaries, indicating that the sequence prior to CRIB is important and that a KEKE sequence is not sufficient to help the CRIB construct localize to junctions (Fig. 6, D, H, and I). To ensure that the N-terminal GFP tag, positioned close to the polybasic motif, was not interfering in targeting, we confirmed our results using C-terminal mCherry-tagged constructs (Fig. S3). We subsequently mutated the four lysines (residue numbers 48–51) in the polybasic stretch of PAK1(1–107) to alanines to investigate the importance of the polybasic motif. Whereas GFP-PAK(1–107) targets robustly to cell–cell contacts, when the lysines are mutated to alanines, targeting is abolished (Fig. 6, E, H, and I), supporting the importance of the polybasic region for targeting. To confirm the relationship between an intact polybasic region and targeting to cell–cell boundaries, we fused the PAK6 N terminus and the PAK1 CRIB domain and showed that this efficiently targeted to cell–cell contacts (Fig. 6, F, H, and I). However, this substitution in the context of full-length PAK1 was insufficient to trigger localization to cell–cell contacts (Fig. 6, F–I), consistent with the inhibitory action of the variable region of PAK1 identified above. We therefore conclude that PAK1 targeting to cell–cell contacts requires both the polybasic motif and the Cdc42-binding CRIB domain but that it is inhibited by the variable region between the CRIB and the kinase domains. Release from the variable region inhibition is therefore likely to be a key step in directing PAK1 to cell–cell contacts (Fig. 6, F–I).

Figure 6.

Polybasic is necessary for targeting once the inhibition from the variable region is released. A, pulldown of GFP-tagged PAK constructs from transiently transfected 293T cell lysates by bacterially purified GST-Cdc42, GST-Cdc42 T17N, or GST-Cdc42 Q61L. Samples were immunoblotted (IB) for GFP, and 3% input is shown. Ponceau staining was used to assess equal bead loading. B, DU145 cells were transiently transfected with GFP-tagged PAK1(1–107) or the PAK1 CRIB domain, and 24 h later, cells were fixed and stained with antibody to β-catenin. Scale bar, 10 μm. C, colocalization of GFP-PAK chimeras with β-catenin was assessed by calculating the Mander's coefficient. At least 15 nonoverlapping frames from at least three independent transfections were used in the analysis. Mean ± S.D. (error bars) is indicated. ****, p < 0.0001 in an ordinary one-way ANOVA test with Dunnett's correction for multiple comparisons. D–F, targeting of GFP-PAK constructs was assessed as in B. A schematic representation of the constructs with PAK1 shown in yellow and PAK6 shown in blue is provided. H, colocalization was assessed by Mander's coefficient as in C. I, immunoblotting of transfected DU145 cell lysates for GFP.

Discussion

PAKs play key roles in cell adhesion, cytoskeletal organization, and cell motility (1). However, despite conserved domain organization and extensive sequence and structural similarity in their Rac and Cdc42-binding CRIB domains and their kinase domains, type II PAKs (PAK4, PAK5, and PAK6) co-localize with β-catenin at cell–cell contacts, whereas PAK1 does not (20, 27, 35, 36). Here, we extend these observations to show that none of the type I PAKs localize at cell–cell contacts in DU145 cells, and we use truncation mutations and PAK1/PAK6 chimeras to identify domains involved in this differential targeting. Surprisingly, rather than revealing a unique targeting domain in type II PAKs, we discovered that the nonconserved central variable region inhibits cell–cell contact localization of type I PAKs. Notably, removal of this region allows CRIB domain–dependent accumulation of PAK1 at cell–cell contacts. Furthermore, we show that this localization depends on an N-terminal polybasic motif as well as the small GTPase binding activity of the CRIB domain. These data explain why type I PAKs are not routinely found at cell–cell contacts but also highlight the importance of future studies to understand how the variable region modulates PAK localization and raise the interesting possibility that cellular modulation of variable region inhibition may control PAK1 subcellular localization.

In all six PAKs, the highly conserved kinase domain is located at the very C terminus, and the regulatory Cdc42/Rac-binding CRIB domain is found near the N terminus, often accompanied by a stretch of basic residues, the polybasic motif (11). The PAKs were identified as Cdc42/Rac binding partners, and in all six PAKs the highly conserved CRIB domain binds Cdc42/Rac, albeit with various affinities (5–10). Functionally, all PAKs have key roles at either cell–matrix or cell–cell junctions, where they regulate adhesion turnover and cell motility (11). Consistent with this, overexpression of PAKs are commonly seen in more aggressive and metastatic cancers (23). Despite these many similarities, the type I and II PAKs harbor a number of unique characteristics. For example, the mode of activation differs between groups. Type I PAKs are maintained as autoinhibited homodimers with the CRIB domain masked by association in trans with a part of CRIB domain called the dimerization segment, and the kinase inhibitor segment passing through a kinase cleft, stabilizes the inactive catalytic domain. Cdc42/Rac binding to the CRIB domain disrupts this inhibitory interaction, permitting kinase activation (11, 14, 29, 37). In contrast, type II PAKs exist as monomers in solution, and their activation is generally considered not to require the binding of Cdc42/Rac, with type II PAK kinase activity instead controlled by binding of an inhibitory pseudosubstrate region (located in the variable region) to the kinase domain in cis (2, 16). The key regulator that releases this inhibitory interaction remains to be identified. Rather than controlling kinase activation, GTPase binding is instead considered to act mainly by controlling type II PAK localization (11). In this regard, it is noteworthy that whereas both type I and type II PAKs have been reported at cell–matrix junctions, where they apparently influence focal adhesion stability and integrity, only type II PAKs have been found at cell–cell junctions, where they drive adherens junction turnover and favor colony escape in epithelial cancer cell lines (6, 20, 36). Given the highly similar domain organization, we sought to understand what caused the differences in type I and II PAK localization.

We have previously shown that for PAK6, the N-terminal region along with Cdc42 binding to the CRIB domain is necessary and sufficient for targeting to cell–cell junctions (20). Here we expanded this observation to confirm the differences in type I and II PAK localization to cell–cell junctions. Then, using a series of PAK1/6 chimeras, we found switching the variable region from PAK6 into PAK1 directed the chimeric PAK1 construct to cell–cell junctions. Deletion experiments indicated that this gain of targeting was not due to the addition of the PAK6 variable region but instead due to the loss of the PAK1 variable region, leading us to suggest an inhibitory function mediated by the variable region (Fig. 7). Notably, a PAK6 construct containing the PAK1 variable region retained cell–cell contact targeting ability, demonstrating that the PAK1 variable region only inhibits PAK1 localization at cell–cell contacts, suggesting that it might exert its effect via association with another region of PAK1. Attempts to identify such an interaction using a bacterially purified GST-tagged PAK1 variable region to pull down PAK domains expressed in 293T cells have thus far failed to reveal any interactions (data not shown). The possibility remains that the interaction, which might normally occur in cis or in the context of a preformed dimer, might be too weak to be detectable in our pulldown assay. However, we note that structural studies indicate the variable region to be largely disordered and only revealed interactions of residues Asp-126 and Glu-129 of the inhibitory switch region in addition to the kinase inhibitor segment (residues 136–149) with the kinase domain. Deletion of the kinase domain does not trigger targeting to cell–cell contacts, and inclusion of residues Asp-126, Glu-129, and the kinase inhibitory switch in truncation constructs PAK1(1–149) did not alter targeting to cell–cell contacts compared with minimal PAK1(1–107) constructs. Thus, currently we cannot explain the inhibitory effect of the variable region through inhibitory PAK1–PAK1 interactions.

Figure 7.

Schematic showing potential mechanisms for regulation of PAK recruitment to cell–cell contacts. Inactive PAK1 is thought to form an autoinhibited dimer (14), whereas PAK6 is inhibited in cis by binding of a pseudosubstrate sequence (18). PAKs are recruited to cell–cell contacts through direct interactions between their CRIB domain and Cdc42 and potentially through interactions of their polybasic (PB) motifs with the membrane. In PAK1, recruitment is blocked by the variable region, possibly through effects on PAK1 conformation or through interactions with other proteins.

Our studies did, however, confirm the interaction of the PAK1 variable region with the Rho GEF β-PIX. The β-PIX–binding site is conserved on all type I PAKs, but not in type II PAKs, and the complex between PAK, PIX, and the PIX binding partner GIT localizes to focal adhesions and promotes junction disassembly by interacting with paxillin (31, 32, 36). Surprisingly, studies in C. elegans identified a novel PAK1-specific regulation of cell shape and migration, where the GIT/PIX/PAK signaling pathway acts independently of Cdc42/Rac, and complex recruitment and formation is negatively regulated by PAK1 autophosphorylation (38). However, mutation of the PIX-binding site in PAK1 or PAK2 (PAK2 data not shown) did not alter PAK localization to cell–cell contacts, suggesting that PIX binding cannot explain the differential adherens junction targeting of type I and type II PAKs. In our study, PIX and GIT were the only visible bands from a Coomassie-stained GST-PAK1-variable region pulldown experiment of DU145 cell lysate; however, we cannot rule out the possibility of other important binding partners interacting at low levels. Therefore, although our data clearly point to an inhibitory role for the variable region of PAK1, the molecular basis for this effect remains to be fully elucidated.

Once the variable region inhibition of PAK1 targeting to cell–cell contacts is released by deletion or substitution, PAK1 targets to junctions in a process that requires a functional CRIB domain. In our study, all of the PAK constructs that target to cell–cell junctions exhibited high Cdc42 binding in pulldown assays, and disrupting binding to Cdc42 with point mutations in the CRIB domain abolished targeting to cell–cell contacts. This is consistent with our prior studies on PAK6 and the impairment of PAK6 targeting in Cdc42 knockdown cells (20). Thus, high Cdc42 binding is apparently required for PAK targeting to cell–cell contacts but is not sufficient, as not all high Cdc42 binders are targeting constructs, meaning the high binding is not the sole mechanism directing constructs to cell–cell junctions.

We previously implicated the polybasic sequence of PAK6 in targeting to cell–cell junctions (20), and once we observed that the PAK1 CRIB domain alone was unable to target despite a high level of binding to Cdc42, we recognized that the largely unstructured N-terminal region of PAK1 must provide an additional regulator for targeting. Previous studies have suggested that the polybasic sequence N-terminal to the CRIB domain aids Cdc42 binding to Cdc42 (39) in PAK6, but our semiquantitative pulldown experiments using GST-Cdc42 and lysates from DU145 cells overexpressing GFP-tagged PAK1 constructs did not reveal differences between the N-terminal region of PAK1(1–107) and the PAK1-CRIB alone. However, we abolished targeting by either truncating the N-terminal region or by mutating the polybasic sequence in PAK1(1–107). Notably, the length and position of the polybasic sequence are apparently not important, because when we created a construct expressing the short PAK6 polybasic motif fused directly to the PAK1 CRIB domain, this localized in cell–cell contacts. We hypothesize that the polybasic motif aids membrane targeting and so facilitates recruitment or retention of PAKs at cell–cell contact sites (Fig. 7). Thus, strong Cdc42 binding and an intact polybasic motif are needed to target PAK1 to cell–cell contacts, but this only occurs if the PAK1 variable region is removed, illustrating that the inhibitory effect of the variable region is dominant. Although we do not yet know how this region inhibits targeting, our results raise the intriguing possibility that cellular signals may release this inhibition such that under certain circumstances, PAK1 may localize to cell–cell contacts, potentially leading to adhesion turnover.

Experimental procedures

Antibodies

All primary antibodies were purchased and diluted at 1:1000 for immunoblotting and 1:100 for immunofluorescence staining; these include antibodies against total β-catenin (Cell Signaling, D10A8 rabbit mAb 8480), GFP (Rockland, 600-101-25, goat polyclonal), β-PIX (Santa Cruz Biotechnology, Inc., H-3, mouse monoclonal, sc-393184), GIT1/2 (Santa Cruz Biotechnology, 13, mouse monoclonal, sc-135925), glutathione S-transferase (Sigma-Aldrich, AB3282, rabbit polyclonal), and Myc tag (Cell Signaling, 9B11, mouse monoclonal, 2276). Secondary antibodies used in immunoblotting were diluted 1:10,000; these antibodies include LI-COR IRDye 680 donkey anti-goat (926-68074, lot C70712-05) and LI-COR IRDye 680 goat anti-mouse (926-68070, lot C60405). Alexa Fluor 568 donkey anti-rabbit (Invitrogen, A10042, lot 1134929), diluted 1:500, was used as the secondary antibody in immunofluorescence staining.

DNA constructs

pEGFP-C2-human PAK1 (Uniprot Q13153) and PAK6 (Uniprot Q9NQU5) were described previously (20). Human PAK2 (Uniprot Q13177) in pEGFP-C2 vector was generated by PCR amplification from a PAK2 cDNA kindly provided by Clair Wells (Kings College London). Human PAK3 (Uniprot O75914) in pEGFP-C1 vector was generated using PCR amplification from pDONR223-PAK3, a gift from William Hahn and David Root (Addgene plasmid 23439) (40). All chimera and mutant constructs were generated with PCR amplification or QuikChange mutagenesis. pGEX 6p1 was kindly provided by Titus Boggon (Yale University). pGEX6p1-VR was generated from PCR amplification of pEGFP-C2-PAK1. pET-28a-Cdc-42 and mutants were described previously (41).

Cell culture and transfection

DU145 prostate cancer cells were kindly provided by Raymond Baumann at Yale University. HEK293T cells were purchased from ATCC. Cells were cultured in Dulbecco's modified Eagle's medium, containing 4.5 g/liter d-glucose and l-glutamine (Gibco, 11965-092), supplemented with 9% fetal bovine serum (Life Technologies). Cells were incubated at 37 °C with 5% CO₂. Cells were transfected using PEI (linear polyethyleneimine, Polysciences, Inc.) at 8 μg/ml overnight in serum-free media.

Immunoblotting

Transfected DU145 or 293T cells were harvested using 0.25% trypsin with EDTA (Gibco, 25200-056). Cells were lysed with buffer X lysis buffer (1 mm Na₂Vo₄, 50 mm NaF, 40 mm NaPP_i, 50 mm NaCl, 150 mm sucrose, 10 mm PIPES, 0.5% Triton X-100, 0.1% deoxycholic acid, complete mini protease inhibitor mixture, EDTA-free (Roche Applied Science, 11836170001)). Cell lysates were diluted in sample buffer, fractionated on 10% SDS-polyacrylamide gels, and transferred to 0.45-μm nitrocellulose membrane at 0.25 mA for 1.5 h. Membranes were blocked with blocking buffer (5% milk in TBS with Tween 20 (TBS-T)) at room temperature for 1 h before incubation with primary antibody in blocking buffer overnight at 4 °C. Membranes were washed with TBS-T before incubating with secondary antibody diluted in blocking buffer for 1 h at room temperature. Membranes were washed with TBS-T before imaging with an Odyssey IR imaging system (LI-COR). All quantification of the immunoblots was done using Image Studio (LI-COR).

Immunofluorescence and scoring of cell–cell targeting

DU145 cells were seeded on coverslips in complete culture medium. 24 h post-transfection, coverslips were fixed in 4% paraformaldehyde in PBS for 15 min at room temperature. Cells on coverslips were permeabilized with 0.05% Triton in PBS for 5 min at room temperature. Coverslips were washed with PBS and blocked with blocking solution (5% BSA in PBS) for 1 h at 37 °C in a humidity chamber. Anti-β-catenin antibody was diluted 1:100 in blocking buffer and incubated with the coverslips for 1 h at 37 °C in a humidity chamber. Coverslips were washed three times with PBS before incubation with secondary antibody diluted 1:500 in blocking buffer for 1 h at 37 °C in a humidity chamber. Coverslips were washed three times and rinsed with milliQ water before mounting with ProLong Diamond Antifade mountant (Thermo Fisher Scientific, P36962). Images were captured using either a Nikon Ti1 or Ti2 microscope with ×100 immersion oil objective.

Mander's colocalization coefficient, which measures the fraction of one protein that colocalizes with a second, in our case the fraction of GFP co-localizing with β-catenin staining, was calculated using JACoP (just another colocalization plugin) in ImageJ (28). This requires effective thresholding to remove background, allowing identification of pixels that are negative for β-catenin or GFP signal (42). There are a variety of mechanisms for removing background, but all require checking visually against the parental image (42). We found that manually setting fixed thresholds in each experiment produced the best overall visual concordance between the thresholded and parental images to ensure the incorporation of all positive structures. As all images in an experiment were recorded at identical exposure times and illumination settings, a fixed background (∼1500 in the GFP channel and 5000 in the red channel) was applied to all images. Mander's coefficient was calculated, and the score showing the fraction of GFP signal overlapping with β-catenin signal was recorded. At least 20 nonoverlapping frames from at least three independent transfections were analyzed for quantification.

Protein production and purification

pET-His-GST-cdc421-177, pET-His-GST-cdc421-177-T17N, pET-His-GST-cdc421-177-Q61N, pGEX 6p1, pGEX 6p1-VR, and pGEX 6p1-VR-GA were transformed into BL21 E. coli. Upon reaching A₆₀₀ of 0.6, cultures were induced with 1 mm isopropyl 1-thio-β-d-galactopyranoside (for pET vector) or 0.2 mm (for pGEX 6p1 vectors) for 3 h at 37 °C. Cells were harvested, and the pellets were lysed in lysis buffer (PBS with 1% Triton X-100, 100 μm phenylmethylsulfonyl fluoride, aprotinin (1 μg/ml), lysozyme (1.2 μg/ml), DNAse I (2 μg/ml), MgCl₂ (2 μg/ml), protease inhibitor (EDTA-free), and 0.1 mm tris(2-carboxyethyl)phosphine) on ice. Lysates were sonicated on ice with four rounds of 15 s on and 15 s off. Lysates were then clarified and incubated with 80% slurry GSH-Sepharose 4B beads (GE Healthcare) overnight rolling at 4 °C. Beads were washed three times with cold PBS and eluted with elution buffer (50 mm Tris, pH 8.2, 150 mm NaCl, 0.1 mm tris(2-carboxyethyl)phosphine, 20 mm GSH, protease inhibitor complete). Buffer exchange was done using PD-10 columns and eluted with PBS and subsequently quantified using Nanodrop.

Cdc42-binding assays

DU145 cells were transiently transfected with pEGFP-PAK constructs overnight with PEI. Cells were harvested and lysed in buffer X lysis buffer. Lysates were clarified and incubated with GSH-Sepharose beads loaded with GST-Cdc42 proteins. Samples were diluted in buffer X-T (5 mm PIPES, pH 6.8, 0.5 mm Na₃VO₄, 25 mm NaF, 20 mm NaPP_i, 25 mm NaCl, 75 mm sucrose, 0.05% Triton X-100) and incubated overnight on a roller at 4 °C. Beads were washed three times with buffer X-T before elution in SDS-PAGE sample buffer at 95 °C for 5 min. Samples were run on 10% SDS-polyacrylamide gels, transferred to nitrocellulose, and immunoblotted for GFP.

HEK293T cells were transiently transfected with pEGFP-PAK constructs overnight with PEI. Cells were harvested and lysed in PBS, pH 7.4, 1% Triton X-100 plus EDTA-free complete mini protease inhibitor mixture (Roche Applied Science, 11836170001). Cells were sheared with a 23-gauge needle and sonicated briefly, and insoluble material was removed by centrifugation. GFP proteins from the clarified lysate were immobilized on GFP-nanotrap beads (prepared as described previously (34, 43)) by mixing for 2 h at 4 °C with rotation. Beads were washed three times with PBS, 0.1% Triton X-100, and the binding of purified GST-Cdc42 was assessed by incubating for 2 h at 4 °C with rotation. Beads were washed three times with PBS, 0.1% Triton X-100 before elution in SDS-PAGE sample buffer at 95 °C for 5 min. Samples were fractionated on 10% SDS-polyacrylamide gels and immunoblotted for GST and GFP.

GST-VR pulldown

DU145 cells were harvested and lysed in buffer X lysis buffer. Lysates were sheared with a 30-gauge needle and clarified. GSH-Sepharose beads loaded with GST, GST-VR, or GST-VR-GA were incubated with or without lysates overnight on a roller at 4 °C. Beads were washed three times with buffer X-T and diluted in sample buffer. Samples were then heated to 70 °C for 15 min and spun down, and supernatants were transferred to new tubes and heated to 95 °C for an additional 10 min. Samples were run on 10% polyacrylamide gels and stained with Coomassie Blue.

Author contributions

X. S. and D. A. C. conceptualization; X. S. formal analysis; X. S. and V. L. S. investigation; X. S. writing-original draft; X. S., V. L. S., and D. A. C. writing-review and editing; V. L. S. validation; D. A. C. supervision; D. A. C. funding acquisition; D. A. C. project administration.