The endocytic adapter E-Syt2 recruits the p21 GTPase activated kinase PAK1 to mediate actin dynamics and FGF signalling

Fibroblast growth factors (FGFs) are part of a large family of growth factors modulating cellular processes (Böttcher and Niehrs, 2005) as diverse as cell survival, differentiation, adhesion, migration and disease. They were the first growth factors shown to possess inductive properties (Kimelman and Kirschner, 1987; Slack et al., 1987) and in Xenopus are essential for mesoderm induction (Kimelman, 2006) and cell migration. Endocytosis is an important aspect of growth factor signalling, and controls both signal strength and specificity (Le Roy and Wrana, 2005). We recently identified the Extended Synaptotagmin-like membrane protein E-Syt2 as an essential endocytic adapter for FGF signalling in Xenopus and human (Jean et al., 2010). E-Syt2 was shown to interact specifically with the activated FGF receptors and with Adaptin-2 (AP-2), and it was suggested to play a role in targeting the receptor to the clathrin-mediated endocytic pathway. Here we show that E-Syt2 also interacts with the p21-GTPase Activated Kinase PAK1 and regulates the actin cytoskeleton, a function implicated in endocytosis.

The PAK family of protein kinases plays key roles in cytoskeletal regulation (Bokoch, 2003), phosphorylating important cytoskeletal regulators like LIM kinase (Edwards et al., 1999), MLCK (Sanders et al., 1999) and MLC (Bisson et al., 2003). PAK has also been shown to play a non-catalytic role in driving actin depolymerization by locally sequestering the active (GTP-bound) GTPases Cdc42 and Rac (Bisson et al., 2007). The role of actin dynamics in endocytosis has been the subject of conflicting reports, but most recently it has been shown that it may play a role at a late stage of clathrin pit internalization (McMahon and Boucrot, 2011). Hence, the recruitment of PAK1 to an endocytic adapter could logically be of functional significance.

To better understand the in vivo functions of Xenopus PAK1 we sought to identify novel PAK1 interacting proteins. The regulatory N-terminal domain and the C-terminal kinase domain of PAK1 were used as bait in two-hybrid screens of a Xenopus laevis cDNA library. 4.7×10⁶ independent clones were screened and eight isolates proved positive after retransformation. One of these, N126, interacted with the N-terminal domain of PAK1 and encoded the last C2 (C2C) domain of Xenopus E-Syt2, a member of a small sub-family of synaptotagmin-like membrane proteins (Min et al., 2007) (Fig. 1a). E-Syt2 is essential for the induction of mesoderm by FGF in early embryos, and E-Syt2 interacts directly and specifically with the activated form of the FGF Receptor (FGFR) (Jean et al., 2010). Further, E-Syt2 loss or overexpression inhibits receptor endocytosis and functional signal transduction via the ERK pathway, suggesting that PAK1 might functionally cooperate with E-Syt2 in mediating FGF signalling. Thus, we first sought to validate the interaction of PAK1 with E-Syt2.

Full length E-Syt2 was found to co-localize with PAK1 at the plasma membrane of Xenopus animal cap (AC) cells (Fig. 1b) and to specifically co-immunoprecipitate with PAK1 (Fig. 1c). Further, we found that the human E-Syt2 orthologue specifically co-immunoprecipitated with human PAK1, demonstrating that this interaction was conserved in evolution (Fig. 1d).

As predicted from the two-hybrid data, interaction with E-Syt2 occurred within the N-terminal control region of PAK1. Full-length PAK1 pulled-down E-Syt2 very effectively while the C-terminal kinase domain (a.a. 220–527) showed only a weak interaction (Fig. 1e) (Fig. 1a for mutant structures). Deletion of the C-terminal C2 domain (C2C) of E-Syt2 eliminated this interaction, while deletion of all three C2 domains had no further effect (Fig. 1e). The N-terminal domain of PAK1 and the C2C domain of E-Syt2 were also sufficient for the interaction (supplementary material Fig. S1a). The site of interaction between the C2C domain of E-Syt2 and the regulatory N-terminus of PAK1 was further delimited to a.a. 47 to 58 of PAK1 using N- and C-terminal deletion mutants (supplementary material Fig. S1b) and deletion of this site, termed the E-Syt2 Box (ESB), essentially eliminated the interaction as determined both in vitro by pull-down and in vivo by immunoprecipitation (Fig. 1f,g). Deletion of the corresponding amino acids (a.a. 49–67) from human PAK1 also eliminated its interaction with human E-Syt2, showing that despite significant sequence divergence in this region, the ESB site was conserved in human (Fig. 1h). Alanine and glutamic acid scanning of the ESB of Xenopus PAK1 did not reveal any single crucial amino acid, but combined mutation of a.a. 48 and 50 (K48/50E) or a.a. 48 to 51 to A (K48-51A) or to E (K48-51E) strongly decreased the interaction (supplementary material Fig. S1c). Thus, the short blocks of lysine residues within the ESB form an essential part of the PAK1 ESB. This said, the solution structure of the C2C domain of E-Syt2 (Protein Data Bank entry, updated in 2009 by Nagashima et al. [http://www.rcsb.org/pdb/explore/explore.do?structureId = 2DMG]) does not reveal any obvious matching acid patch for these adjacent lysines, suggesting that the full ESB probably includes adjacent amino acids.

Interestingly, recent work identified the basic residues a.a. 48–51, 66–68 and 70 of human PAK1 that overlap the ESB (Fig. 1h), as being required for PAK1 recruitment to phospholipids (Strochlic et al., 2010). These authors suggested that in the cellular context activation of human PAK1 by PDGF depended on these basic amino acids. Given that E-Syt2 binds phospholipids via its C2C domain (Jean et al., 2010; Min et al., 2007), its interaction with PAK1 suggested a different scenario. We, therefore, endeavoured to determine whether E-Syt2 would enhance or would compete for PDGF-dependent activation of PAK1. However, despite strong activation of ERK, in our hands Xenopus and human PAK1 were only weakly if at all activated by PDGF (supplementary material Fig. S1d,e), and we note that others have made similar observations (Deacon et al., 2008; Yuan et al., 2010). The recruitment of PAK1 to the plasma membrane via phospholipids could, however, enhance its subsequent recruitment by E-Syt2. We have found that recruitment of the C2C domain of E-Syt2 to the plasma membrane is independent of the PAK1 interaction site (F.G., unpublished). Thus, E-Syt2 forms a bridge between PAK1 and the plasma membrane.

We found that PAK1 was recruited to the FGFR1/E-Syt2 complex in vivo and, as previously observed for E-Syt2, this recruitment was stimulated by receptor activation (Fig. 2a), suggesting that PAK1 might cooperate in FGF signalling. Overexpression of E-Syt2 was found to function as a dominant negative mutation, inhibiting FGF receptor endocytosis, FGF signalling to ERK and the induction of mesoderm (Jean et al., 2010). We therefore asked if expression of activated forms of PAK1 might rescue the dominant negative effect of E-Syt2. The mildly activated form of PAK1 (L98F) did indeed partially rescued FGFR1-induced mesoderm induction in ACs overexpressing E-Syt2, as judged by expression of the mesodermal marker Xbra (Fig. 2b), and this rescue was somewhat enhanced by the more strongly activated N-terminally deleted PAK1. (Rescue was limited by the fact that higher levels of PAK1 activation lead to a fragmentation of blastomeres caused by its direct activation of myosin II (Bisson et al., 2003)). In contrast, expression of wild type PAK1, or the catalytically inactive forms PAK-K281A or PAK-L98F/K281A, all of which bind E-Syt2, were all unable to rescue Xbra expression in E-Syt2 expressing AC (Fig. 2c). Thus, rescue required the kinase activity of PAK1 and was not simply due to sequestration of E-Syt2. Indeed, these dominant negative forms as well as the dominant negative PAK1 auto inhibitory domain (AID) (Parrini et al., 2002) all mildly suppressed Xbra induction by activated FGFR1 (supplementary material Fig. S2).

A defect in FGF signalling first becomes evident at gastrulation as a delay or inhibition of cell migration and involution of mesoderm, as observed for the dominant negative receptor XFD (Isaacs et al., 1994), (Fig. 2d). Since E-Syt2 and to a lesser extent catalytically inactive PAK-K281A both inhibited FGF signalling when over-expressed, we argued that they should also affect gastrulation movements. Indeed, dominant negative forms of PAK1 were previously shown to inhibit these movements (Nagel et al., 2009). Ectopic expression of both E-Syt2 and PAK-K281A, but not of wild type PAK1, delayed involution in a similar manner to dominant negative XFD, whereas wild type PAK1 had no effect on gastrulation (Fig. 2d). At stage 10.5, involution at the dorsal gastropore lip and the formation of Brachet's cleft were delayed in E-Syt2 and PAK-K281A injected embryos, while by stage 11.5 involution was very limited and no archenteron, clearly evident in the control embryos, had formed. Activated PAK-L98F, shown to partially rescue Xbra expression when co-injected with E-Syt2, did not detectably rescue gastrulation movements. However, this was consistent with the limited rescue of FGF signalling afforded by this mutant.

We have shown that FGFR1 internalization via the clathrin-dependent endocytic pathway is essential for prolonged ERK activation and the induction of mesodermal markers in AC cells (Jean et al., 2010). However, clathrin-dependent receptor internalization requires the dynamic modulation of the actin cytoskeleton, both during clathrin pit formation and subsequently to drive endocytosis (Yarar et al., 2005). PAK1 is both a key downstream effector and upstream regulator of the p21 GTPases Cdc42 and Rac and together with these GTPases is in large part responsible for the regulation of cytoplasmic and cortical actin (Bisson et al., 2007; Bokoch, 2003; Obermeier et al., 1998). In particular, we previously showed that sequestration of activated Cdc42 by PAK regulated cortical actin in a manner independent of PAK catalytic activity (Bisson et al., 2007). Thus, one way that PAK1 might facilitate FGF signalling is by controlling the dynamic changes in the cytoskeleton required for endocytosis.

Since E-Syt2 recruited PAK1 via the ESB, a site immediately adjacent to the GBD/CRIB (a.a. 66–77) (Fig. 1f), it was possible that this recruitment might interfere with the activation of PAK1 either by preventing GTPase binding or the subsequent release of the activated GTPase. Deletion of the ESB (Δ47–58) had no effect on the interaction of PAK1 with Cdc42 (Fig. 3a), while deletion (Δ66–77) or mutation (H74,77L) of the GBD/CRIB GTPase binding site completely eliminated the interaction. Thus, the ESB and GBD/CRIB sites do not overlap to any detectable degree. Further, binding of the C2C domain of E-Syt2 (Fig. 1a) to the ESB of PAK1 was completely unaffected by competition with a large excess of activated Cdc42 or Rac (Fig. 3b). A similar result was obtained for Filamin, another GBD/CRIB interacting protein (Vadlamudi et al., 2002) (supplementary material Fig. S3a). Thus, E-Syt2 and Cdc42 or Rac did not interfere with each other's reaction with their proximal sites on PAK1. To functionally test the significance of E-Syt2 binding, we asked if E-Syt2 affected the activation of PAK1 by constitutively active Cdc42 or Rac in vivo. Increased levels of E-Syt2 led to a significant inhibition of Pak1 activation by Cdc42(L61) or Rac(L61), as judged by autophosphorylation of PAK1 (upshift in figure) and by PAK1 catalytic activity in trans (Fig. 3c,d; supplementary material Fig. S3b,c). This inhibition was not observed for PAK1Δ47–58 in which the E-Syt2 binding (ESB) was deleted (supplementary material Fig. S3c). Thus, E-Syt2 was able to limit the ability of both Rac and Cdc42 to activate PAK1. Since preventing PAK1 activation in turn sequesters the activated GTPases, inhibiting their function and disrupting cortical F-actin (Bisson et al., 2007), we hypothesized that E-Syt2 might in this way locally regulate cortical actin polymerization.

To functionally test the effects of E-Syt2 on actin polymerization we first used the embryo wound-healing assay. In this assay the rapidity of wound-healing depends on the ability to polymerize a ring or purse-string of actin at the leading-edge of the wound, a process dependent on Cdc42 function (Benink and Bement, 2005; Kofron et al., 2002b). Morpholino depletion of E-Syt2 enhanced embryo wound-healing, as well as enhancing actin polymerization in the ACs removed from these embryos (Fig. 4a). In contrast, E-Syt2 overexpression slowed this wound healing, and inhibited actin polymerization in the corresponding ACs.

In agreement with the Xenopus wound healing data, in vitro wound healing (Scratch Test) assays (Coomber and Gotlieb, 1990) (Fig. 4b,c) also showed that in response to FGF stimulation NIH3T3 cells expressing E-Syt2 migrated less rapidly than mock transfected cells. This effect was particularly apparent at 16 h after FGF addition, and was dependent on the presence of the C2C domain of E-Syt2 shown to recruit PAK1.

Consistent with these macroscopic effects on cytoskeletal function, expression of E-Syt2 induced the cell-autonomous depletion of cortical F-actin (Fig. 5a). Analysis of many fields of E-Syt2 expressing and non-expressing cells showed that cortical actin staining was quantitatively reduced at regions positive for E-Syt2. Further, the induction of actin stress fibres in NIH3T3 fibroblasts was strongly affected by E-Syt2 expression, which caused the formation of large F-actin plaques at the expense of fibre formation (Fig. 5b). In contrast, deletion of the C2C domain of E-Syt2 eliminated this effect, allowing normal stress fibre formation. Thus, E-Syt2 suppressed normal F-actin polymerization both functionally in wound-healing assays, and visibly at the plasma membrane and during stress fibre formation.

E-Syt2 is a novel endocytic adapter that is specifically recruited to activated FGFR1 and stimulates a rapid phase of receptor endocytosis required for activation of the ERK signalling pathway (Jean et al., 2010). E-Syt2 also recruits Adaptin-2 (AP-2) to the receptor, suggesting that it plays a role in the assembly of clathrin-coated pits and in this way targets the activated receptor to the clathrin-mediated endocytic pathway. Here we show that E-Syt2 is also a novel interaction partner for the cytoskeletal kinase PAK1. The third C2 domain (C2C) of E-Syt2 binds to a CRIB/GBD adjacent site within the regulatory N-terminal domain of PAK1. This E-Syt2 binding site (ESB) was initially mapped on Xenopus PAK1, but was shown to be functionally conserved in human PAK1. Homologies between the various PAK isoforms suggest that an ESB may be present in the PAK3 and possibly the PAK2 orthologs (supplementary material Fig. S4), but probably not in PAKs 4 to 6.

E-Syt2 can bind PAK1, and both are recruited to the FGF receptor on its activation. E-Syt2 inhibits wound healing possibly by suppressing F-actin polymerization. This may be related to the ability of E-Syt2 to suppress the activation of PAK1 by Cdc42 and Rac small GTPases. We previously found that inactive forms of PAK1 locally sequester these GTPases and prevent their action in catalysing actin polymerization (Bisson et al., 2007). Since, regulation of the actin cytoskeleton has been implicated in endocytosis (Boulant et al., 2011; Liu et al., 2010; Taylor et al., 2011; Yarar et al., 2005). Our data suggest a possible role for E-Syt2 binding of PAK1 in this process. We hypothesize that PAK1 recruitment to E-Syt2 could cause local cortical actin depolymerisation or displacement and open the way for the assembly of a clathrin-coated pit (Fig. 5c). E-Syt2 is not present in early endosomes (Jean et al., 2010) and (F.G., unpublished), and thus probably leaves the coated pit before its endocytosis or even early in its formation. Though a thorough test of this hypothetical series of events awaits the elaboration of an adequate experimental system, it is interesting to note that PAK1 was previously implicated in both clathrin and non-clathrin mediated endocytosis (Pelkmans et al., 2005).

During the course of this study, human PAK1 was shown to bind phospholipids. It was suggested that this property enhanced its recruitment to the plasma membrane as well as its catalytic activation (Strochlic et al., 2010). It is interesting that the phospholipid binding site mapped by these authors overlaps the E-Syt2 binding site mapped here. Unfortunately, we were unable to reproduce the experimental data supporting a role for phospholipid binding in the activation of either human or Xenopus PAK1. On the other hand, using alanine-scanning mutation we have shown that E-Syt2 recruitment to the plasma membrane and its interaction with PAK1 are mediated by distinct binding sites (F.G., unpublished). This suggests that PAK1 could interact with phospholipids via E-Syt2. However, if PAK1 does interact directly with phospholipids in vivo, this might be expected to enhance its subsequent recruitment to E-Syt2 by concentrating it to the plasma membrane.

Two-hybrid baits were constructed by amplifying sequences encoding a.a. 1–186, 240–528 and 1–528 of kinase dead Xenopus laevis PAK1 using cloned PFU polymerase. The fragments were subcloned in pGBKT7-AD (Clontech) and used to screen a stage 30 Xenopus laevis cDNA library derived from head and tail mRNA, using the Matchmaker Library Contruction and Screening Kit (Clontech) according to the manufacturer instructions. An average of 4.7×10⁶ independent clones were screened using each bait and positive clones verified by retransformation. Mutant and GST constructs were created by PCR amplification where appropriate using the QuikChange protocol (Stratagene) and fully sequenced. The PAK1 and E-Syt2 constructs and antisense reagents have been described previously (Bisson et al., 2003; Jean et al., 2010). Human PAK1 was obtained from J. Chernoff. The coding region for a.a. 1795 to 2647 of Filamin was subcloned by PCR from I.M.A.G.E. clone 4800733 and verified by sequencing. Embryo expression constructs for Xenopus FGFR1 and its activated form (K562E) were obtained from R. E. Friesel and Flag-tagged FGFR1 was obtained from C. Niehrs (Böttcher et al., 2004). Both were cloned in the pCS2+ vector. Wherever the source of E-Syt2, PAK1 and FGFR1 is not specified the Xenopus forms were used. Human Cdc42 and Rac1 GTPase T/C expression constructs were provided by N. Lamarche-Vane and the corresponding bacterial expression vectors were provided by A. Hall. The position of epitope tags is indicated as N- or C-terminal by prefixing or suffixing the epitope indicator, e.g. HA, Myc, etc.

Embryos were staged using the Nieuwkoop and Faber tables (Nieuwkoop and Faber, 1967) and injection performed as described elsewhere (Jean et al., 2010). To assay expression of mesodermal marker Xbra Animal Caps AC), explants were removed from injected embryos at stage 8–9 using forceps on agarose coated dishes in 1× Barth's medium and cultured in 0.5× Barth's medium in parallel with control embryos and stored at −80°C before gene expression analysis. For actin staining ACs were cultured in 1× MMR for 10 minutes before staining with Alexa488-coupled phalloidin (Invitrogen) (Tao et al., 2007), while the vegetal region of the embryo was culture for 1 h in 1× MMR and then fixed in MEMFA. The diameter of the wound was measured using a graticule mounted on binocular microscope and normalized to the embryo diameter. The wound size is indicated as the relative surface area of a circle corresponding to the normalized diameter. For blastula and gastrula sections, embryos at the indicated stage were fixed in MEMFA overnight at 4°C and cut sagittally with a scalpel blade.

HEK293T and NIH3T3 cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (Wisent).

Transfection, immunoprecipitation and pull-down assays were carried out as in (Jean et al., 2010; Vadlamudi et al., 2002). In the case of competition pull-down assays with Cdc42, Rac and Filamin, interacting components were added simultaneously.

HEK293T cells were transfected with the indicated constructs and lysed in MGE, (20 mM MOPS pH 7.0, 10% glycerol, 0.5 mM EDTA, 5 mM EGTA, 1 mM NaVO₄, 5 mM NaPPi, 50 mM NaF, 80 mM β-glycerophosphate, 1% Triton X-100, 1 mM benzamidine, 1 mM DTT, 1 mM PMSF). Otherwise the procedure was as previously described (Bisson et al., 2003).

Immunofluorecence and GFP/RFP imaging were carried out essentially as previously described (Jean et al., 2010). Stage 8 animal caps (AC) were processed for actin staining with Alexa 488-conjugated phalloidin as described by (Kofron et al., 2002a; Lloyd et al., 2005). HEK293T and NIH3T3 cells expressing HA tagged E-Syt2 or E-Syt2 mutant were fixed in 4% PFA in PBS, 20 min R/T and permeabilized with 0.5% Triton X-100 in PBS, 5 min. R/T. E-Syt2 revealed with anti-HA mouse (12CA5) and Alexa568 conjugated anti-mouse second antibody (Invitrogen) and Alexa 488-conjugated phalloidin used to reveal F-actin. Stress fibres were induced in NIH3T3 cells by 5 min. treatment at 37°C with 250 µM H₂O₂ in PBS before fixing (Huot et al., 1998).

The effects of E-Syt2 and mutants on NIH3T3 cell migration were determined in a wound healing assay (scratch test) (Coomber and Gotlieb, 1990). Cells were tranfected with E-Syt2 constructs or mock transfected, 12 h later serum was withdrawn and cells incubated for a further 12 h before the assay. After scratching with a 2 µl pipette tip cells were incubated for the indicated times in the presence or absence of 20 ng.ml⁻¹ bFGF (Sigma-Aldrich) and 5 µg.ml⁻¹ heparin (Sigma-Aldrich).

The authors are grateful to those who provided material: R. Friesel, N. Lamarche, J. Chernoff and L. Poitras. They also wish to thank Nicolas Bisson and Alexander Mikryukov for helpful discussion during the course of this work. The work was supported by an operating grant from the National Cancer Institute of Canada with funds from the Canadian Cancer Society, FRSQ Santé and CIHR scholarships to S.J.