PDK1-mediated activation of MRCKα regulates directional cell migration and lamellipodia retraction

Directional cell migration is of paramount importance in both physiological and pathological processes, such as wound healing and tumor metastasis (Yamaguchi et al., 2005). Among the different types of directed cell migration, chemotaxis, i.e., migration toward a soluble chemotactic agent, is probably the most studied (Roussos et al., 2011). Because of its ability to bind to phosphatidylinositol (3,4,5)-trisphosphate (PIP3) produced at the leading edge, 3-phosphoinositide–dependent kinase 1 (PDK1) has been recognized as a key regulator of cell migration and chemotaxis. Its role in this process was proved in different cell types and organisms including endothelial cells (Primo et al., 2007), smooth muscle cells (Weber et al., 2004), T lymphocytes (Waugh et al., 2009), neutrophils (Yagi et al., 2009), and Dictyostelium discoideum (Liao et al., 2010). PDK1 is a serine/threonine kinase that phosphorylates residues in the activation segment of AGC (cAMP-dependent protein kinase A, cGMP-dependent protein kinase G, and phospholipid-dependent protein kinase C) family proteins (Alessi et al., 1997; Pearce et al., 2010). PDK1 recognizes phosphoinositides phosphorylated in position 3 by phosphatidylinositol 3 kinase (PI3K), through its C-terminal pleckstrin homology (PH) domain. This event localizes PDK1 to the plasma membrane, where it phosphorylates Akt (Currie et al., 1999). PDK1 substrates lacking the PH domain, such as p70S6K, SGK, RSK, and PKC isoforms (Toker and Newton, 2000), require a different mechanism for their activation. In this case, PDK1 binds the hydrophobic motif (HM) on these substrates through its PDK1-interacting fragment (PIF)-binding pocket, leading to their phosphorylation and full activation (Biondi et al., 2001).

Different mechanisms have been proposed to explain the role of PDK1 in cell migration. The concomitant localization of PDK1 and Akt at the cellular leading edge is essential for endothelial cell chemotaxis and angiogenesis (Primo et al., 2007). Moreover, PDK1 has been shown to regulate cell invasion, in particular of breast cancer and melanoma cells through the activation of PLCγ1 (Raimondi et al., 2012). It has also been reported that PDK1 can control cancer cell motility by regulating cortical acto-myosin contraction in a mechanism involving activation of ROCK1 (Pinner and Sahai, 2008).

Regulation of nonmuscle-myosin activity is essential in directional migration, as well as in multiple cellular processes (Vicente-Manzanares et al., 2009). As regulators of nonmuscle-myosin activity, Rho-activated protein kinases are pivotal regulators of cell migration and tumor cell invasion. This group of kinases belongs to AGC family protein and includes two isoforms of Rho-associated protein kinase (ROCK; Amano et al., 1996)—citron Rho-interacting kinase (CRIK; Di Cunto et al., 1998) and myotonin protein kinase (DMPK; Kaliman and Llagostera, 2008)—and three isoforms of myotonic dystrophy kinase–related CDC42-binding kinase (MRCK; Leung et al., 1998). All these kinases share the ability to increase myosin regulatory light chain 2 (MLC2) phosphorylation either directly, by phosphorylating it on T18 or S19 (Amano et al., 1996), or indirectly, by the phosphorylation of myosin phosphatase target subunit 1 (MyPT1), which results in a further increase of MLC2 phosphorylation (Kimura et al., 1996; Tan et al., 2001a). Phosphorylation of MLC2 results in actomyosin contractility (Ikebe and Hartshorne, 1985).

In contrast to the closely related ROCK kinases that are regulated by the Rho GTPase (Amano et al., 1999), there is relatively little information about MRCKα, MRCKβ, and MRCKγ (Zhao and Manser, 2005). MRCK kinases are downstream effectors of GTPase-CDC42 that play key roles in actin-myosin regulation. The current model of MRCK activation also involves diacylglycerol binding, thereby allowing transautophosphorylation upon appropriate N-terminal interactions. Phosphorylation within the activation loop and the HM provides the means for activation, as demonstrated by the mutation T403A in HM, which completely abolishes MRCK kinase activity (Leung et al., 1998; Tan et al., 2001b).

Here, we show that PDK1 regulates directional migration of breast epithelial cells in a kinase-independent manner by inducing MRCKα activation, myosin phosphorylation, and lamellipodia contraction.

EGF is known to be a potent stimulator of cell migration (Blay and Brown, 1985). The nontransformed mammary epithelial cells MCF10A responded to gradients of EGF in a dose-dependent manner (Fig. S1 A), but in the absence of gradient, EGF was unable to induce a significant increase of migrated cells compared with unstimulated cells (Fig. S1 B). PI3K activity, which plays a crucial role in the EGF signaling pathway (Price et al., 1999), has been demonstrated to be a key regulator of the leading edge in migrating cells (Merlot and Firtel, 2003) and, accordingly, inhibition of PI3K activity resulted in the impairment of MCF10A chemotaxis toward EGF (Fig. S1 C) without affecting their viability at the same concentrations (Fig. S1 D).

Among PI3K downstream effectors potentially involved in directional cell migration, PDK1 has been reported to be required for amoeboid and mesenchymal motility (Primo et al., 2007; Pinner and Sahai, 2008).

To investigate its role in EGF-induced chemotaxis, we stably silenced PDK1 with two different shRNAs transduced by lentiviral vectors (shPDK1 #79 and shPDK1 #81), which were able to efficiently reduce PDK1 protein expression compared with nontargeting control, shScr (Fig. 1 A). PDK1 silencing had a profound effect on chemotaxis in transwell assays at 5 ng/ml EGF, whereas it minimally influenced the still low cell motility in the absence of EGF or when the EGF gradient was abrogated by adding the growth factor in the upper chamber (Fig. 1 A). Moreover, PDK1 knockdown had only marginal effects on cell viability (Fig. S1 E), which is consistent with previous studies (Primo et al., 2007; Liu et al., 2009; Gagliardi et al., 2012). Notably, PDK1 wild-type (PDK1_WT) overexpression (Fig. 1 B) was able to significantly increase chemotaxis toward the gradient of EGF but not cell motility in the presence of a homogenous concentration of EGF (Fig. 1 B). The increased number of migrated cells was not caused by increased cell viability (Fig. S1 F).

To assess PDK1 involvement in directional migration, we tracked single MCF10A cells in a scratch wound healing assay and in a sparse cells motility assay (Fig. 1, C and F; and Videos 1 and 2). In a wound assay, cells moving collectively showed an increased directionality compared with those moving as sparse cells (Fig. 1, C and F). The directionality was quantified by calculating the persistence (Fig. 1, E and H), the ability of a cell to maintain its direction of migration, and the forward migration index (FMI), a parameter that quantifies how much cells are able to migrate in the same direction (Fig. 1, D and G).

PDK1-silenced cells showed a significantly reduced FMI (Fig. 1 D) and persistence (Fig. 1 E) in a wound assay compared with control cells. In contrast, PDK1-silenced sparse cells displayed the same directionality of control cells. Concordantly, PDK1_WT overexpression increased both FMI (Fig. 1 G) and persistence (Fig. 1 H) of wounded cells, but was not able to modify persistence of sparse cells (Fig. 1 H). To exclude off-target effects of shRNA used to silence PDK1, we transduced shPDK1 #79 cells with a PDK1_WT cDNA resistant to silencing (Fig. S1 G). PDK1_WT reexpression completely rescued the migration ability toward EGF (Fig. S1 H), as well as FMI and cell persistence in wound healing assays (Fig. S1, I and J).

To understand the molecular mechanism activated by PDK1 during cell migration, we investigated which domain of PDK1 is required for this function. MCF10A cells were transduced with lentiviral vectors expressing different PDK1 mutants: PDK1_WT, K111N mutant, which abolishes kinase activity (PDK1_KD); PH domain point mutant, which impedes the binding to PIP3 at the membrane (PDK1_K465E); L155E mutant, which prevents the binding to the HM on PDK1 substrates (PDK1_L155E); PH domain-deleted mutant (PDK1_ΔPH); the deletion mutant covering the 50 N-terminal amino acids that bind to Ral-GEF (PDK1_Δ50); and the double points mutant in both PH domain and PIF-binding pocket (PDK1_L155EK465E; Fig. 2 A). As shown in Fig. 1 B, PDK1_WT overexpression induced an increased migratory ability toward EGF (Fig. 2 B). Surprisingly, PDK1_KD cells had a phenotype comparable to that of PDK1_WT, which suggests that the kinase activity is not necessary to induce cell migration. Similar results were obtained with PDK1_Δ50 mutant. In contrast, PDK1_L155E mutant completely abolished the increase of chemotaxis. The inability to bind membrane-associated PIP3 reduced, but did not completely abrogate, the induction of migration, as shown by PDK1_ΔPH and PDK1_K465E mutants, whereas it was completely prevented by the double mutation of PDK1_L155EK465E (Fig. 2 B).

To exclude the formation of heterodimers between exogenous mutants and endogenous PDK1, which could alter the promoting effect on cell migration, we measured cell migration in the presence of PDK1 mutants expressed in PDK1-silenced cells. Even in these conditions, the expression of either PDK1_WT or PDK1_KD was able to increase cell migration by rescuing the effects of PDK1 silencing (Fig. S2 A). In contrast, both PDK1_ΔPH and PDK1_L155E mutants were unable to rescue the migratory ability of PDK1-silenced cells (Fig. S2 A).

Similar results were obtained in a wound healing assay (Fig. 2 C). Indeed, PDK1_WT, PDK1_KD, and PDK1_Δ50 mutants were able to increase the FMI (Fig. 2 D) and the persistence (Fig. 2 E). In contrast PDK1_ΔPH and PDK1_L155E mutants did not increase FMI or persistence compared with cells transduced with empty vector. To further confirm that the kinase activity of PDK1 is dispensable for the regulation of chemotaxis, we used the specific PDK1 inhibitor GSK2334470, which has been reported to effectively inhibit PDK1 phosphorylation of SGK2, S6K, Akt, and RSK2 at low concentrations (1–3 µM; Najafov et al., 2011; Fig. S2 B). Despite this, GSK2334470 completely failed to block MCF10A chemotaxis toward EGF even at 10 µM (Fig. S2 C), a concentration that caused a significant reduction of cell viability (Fig. S2 D).

Because the PDK1 effect on cell migration requires the integrity of its PIF-binding pocket but not of the kinase domain, we searched for proteins with a putative PIF sequence, even though they did not contain PDK1 phosphorylatable motifs (Tables S1 and S2). The members of the Rho-GTPase–associated kinase family fulfilled these requirements because, though they lack a classical activation loop consensus motif, they exhibited an HM consensus sequence for the interaction with PDK1. The common function of all Rho-GTPase–associated kinases is the activation of nonmuscular myosin (Zhao and Manser, 2005). Therefore, we evaluated whether MCF10A motility was dependent on the regulation of myosin activity using Blebbistatin, an inhibitor of myosin-II ATPase activity (Limouze et al., 2004; Wilkinson et al., 2005). Treatment with Blebbistatin was able to inhibit both cell migration (Fig. 3 A) and FMI (Fig. 3 D) at concentrations reported to be effective in the inhibition of myosin activity and cell migration in other cell lines (Wilkinson et al., 2005). Thus, we examined the role of Rho-GTPase–associated kinase in this process using specific kinase inhibitors. Unexpectedly, we found that the ROCK inhibitor Y-27632 (Uehata et al., 1997) increased MCF10A migration toward EGF (Fig. 3 B) and FMI (Fig. 3 E). In contrast, the MRCK inhibitor chelerythrine chloride (Tan et al., 2011) completely blocked MCF10A directional migration (Fig. 3 C) and FMI (Fig. 3 F). To confirm the role of MRCK in directional cell migration, we silenced α and β isoforms of MRCK, which are both expressed in MCF10A (Fig. 3 G). Knocking down a single MRCK isoform was not sufficient to recapitulate the effects of an MRCK inhibitor. However, the simultaneous down-regulation of MRCKα and MRCKβ resulted in a significant impairment of cell migration (Fig. S2 E). To understand which MRCK isoform was involved in the PDK1-induced migration, we silenced MRCKα or MRCKβ in MCF10A cells overexpressing PDK1_WT or PDK1_KD (Fig. 3 G). Cell migration experiments showed that MRCKα was required for PDK1 regulation of cell migration, whereas MRCKβ appeared to be dispensable (Fig. 3 H). However, because MRCKβ silencing increased the expression of MRCKα, we could not exclude a compensatory effect (Fig. 3 G). Similarly, silencing of MRCKα, but not MRCKβ, cancelled the increase of FMI (Fig. 3 I) obtained by PDK1_WT and PDK1_KD overexpression. To further exclude the role of ROCK in mediating PDK1 effects, as previously described in amoeboid migration (Pinner and Sahai, 2008), we silenced ROCK1 in MCF10A cells (Fig. 3 J). Down-regulation of ROCK1 failed to reduce the effects of PDK1_WT and PDK1_KD overexpression on both cell migration and FMI (Fig. 3, K and L).

Remarkably, PDK1 played a relevant role in cell migration even in the presence of growth factors other than EGF. In a gradient of hepatocyte growth factor (HGF), MCF10A cells, transduced with control vector, efficiently migrated, whereas PDK1-silenced cells completely lost this ability (Fig. S2 F). The overexpression of PDK1_WT or PDK1_KD strongly increased the number of cells that migrated toward HGF (Fig. S2 G), and this effect is completely dependent on MRCKα (Fig. S2 H).

MRCKα, similar to other AGC kinases, has a highly conserved HM, which is the potential binding site for PDK1. This site is phosphorylated and required for MRCKα activity (Tan et al., 2001b). We hypothesized that PDK1 could associate with MRCKα through the interaction of PDK1 PIF-binding pocket with HM of MRCKα. Consistent with our hypothesis, MRCKα was able to coimmunoprecipitate PDK1_WT, but not PDK1_L155E (Fig. 4 A). Moreover, we performed a pull-down assay with the recombinant catalytic domain (CAT) of MRCKα, which contains the HM, to confirm that the PIF-binding pocket of PDK1 is involved in the interaction with MRCKα. As expected, L155E mutation abolished the binding of PDK1 with MRCKα (Fig. 4 B). In addition, we assayed the interaction of MRCKα_CAT with the PDK1_K465E mutant, which is unable to bind PIP3 produced by PI3K. This mutation also prevented the binding between PDK1 and MRCKα catalytic domain (Fig. 4 B). Pull-down experiments with the recombinant catalytic domain of MRCKβ, which holds a conserved HM highly similar to that of MRCKα, showed that PDK1 also associates with MRCKβ (Fig. S3 A).

To further confirm the direct interaction of PDK1 with MRCKα, we performed the reverse pull-down with GST-PDK1. In this case as well, GST-PDK1_WT was able to efficiently pull-down GFP-MRCKα_CAT (Fig. 4 C).

The interaction of PDK1 with MRCKα was also analyzed in intact cells by performing in situ proximity ligation assay (PLA) experiments. Myc-tagged PDK1 and endogenous MRCKα, detected with their respective antibodies, showed fluorescent spots, indicative of their proximity localization, whereas nonimmune Ig did not display any fluorescence signal (Fig. 4 D).

The main biochemical function of MRCKα is the activation of nonmuscular myosin, either directly by phosphorylation of MLC2 or indirectly through inactivation of the MyPT1 by phosphorylation of T696 residue (Tan et al., 2001a). The need for MRCKα in PDK1-induced migration prompted us to investigate whether PDK1 was able to modulate the phosphorylation of MLC2 and MyPT1.

PDK1 knockdown reduced the EGF-induced increased phosphorylation of MCL2 on both S19 (Figs. 5 A and S3B) and T18/S19 (Figs. 5 A and S3 C). This reduction is even more evident on T696 of MyPT1 phosphorylation (Fig. 5, A and B). Moreover, the overexpression of PDK1_WT increased MLC2 phosphorylation at both single and double sites, and a similar effect was produced by PDK1_KD (Fig. 5, C and D). In contrast, the PDK1_L155E mutant was completely unable to increase MLC2 phosphorylation, whereas the PDK1_ΔPH mutant showed an intermediate effect (Fig. 5 C). To confirm that the regulation of MLC2 phosphorylation by PDK1 was dependent on MRCKα, we treated cells with the MRCK inhibitor chelerythrine chloride. No differences were observed between control and PDK1-overexpressing cells in the presence of MRCK inhibitor. Even in cells overexpressing both MRCKα and PDK1, the level of MLC2 phosphorylation was unchanged (Fig. S3 D). To further exclude the contribution of ROCK to PDK1-dependent regulation of MLC2, we treated cells with the ROCK inhibitor Y-27632. In these experimental conditions, MRCKα overexpression was able to induce an important increase of T18/S19 MLC2 phosphorylation, with further increases in cells overexpressing both PDK1_WT and MRCKα (Fig. S3 D).

Because PDK1 kinase activity was not required to increase MLC2 phosphorylation, we investigated whether the direct interaction between PDK1 and MRCKα could regulate the MRCKα activity. We performed in vitro kinase assays with recombinant purified proteins produced in eukaryotic cells. The catalytic region of MRCKα, described as constitutively active (Leung et al., 1998), displayed a high phosphorylation rate that did not change in the presence of GST-PDK1 (Fig. S3 E). The full-length form of MRCKα (GST-MRCKα), which instead is tightly regulated (Tan et al., 2001b), had low kinase activity and, unexpectedly, was not modulated by the addition of GST-PDK1_WT protein (Fig. S3 F). However, we obtained significant alterations of the kinase activity of GST-MRCKα when expressed in cells with different levels of PDK1. Specifically, the overexpression of PDK1_WT resulted in a significantly higher activity of MRCKα (Fig. 5 E), whereas the PDK1 silencing reduced the kinase activity of GST-MRCKα (Fig. 5 F). Collectively these results suggest that other molecular components are involved in the PDK1-dependent regulation of MRCKα.

We showed that PDK1 mutants that are unable to bind to PIP3 (PDK1_ΔPH and PDK1_K465E mutants) were defective in the regulation of directional migration (Fig. 2, B, D, and E). This suggested that PDK1 localization at the plasma membrane could be involved in this process. We analyzed by total internal reflection fluorescence (TIRF) microscopy the translocation at the plasma membrane of GFP-PDK1 and mCherry-MRCKα fusion proteins. EGF stimulation caused a rapid plasma membrane localization of GFP-PDK1_WT, reaching a maximum at 100 s after stimulation and slowly decreasing (Fig. 6 A and Video 3). Similarly, PDK1 mutants, GFP-PDK1_L155E, and GFP-PDK1_KD increased their plasma membrane localization upon EGF stimulation (Fig. 6 B and Video 4). As expected, GFP-PDK1_K465E was totally excluded from the plasma membrane (Fig. 6 B and Video 4). Stimulation with EGF also induced an increase of mCherry-MRCKα membrane localization (Fig. 6 C and Video 5). GFP-PDK1_WT and GFP-PDK1_KD cotranslocated with MRCKα to the plasma membrane with maximal colocalization at 200–300 s after EGF stimulation. Remarkably, the colocalized pixels were not randomly distributed but clustered in protrusive regions of adherent cell surface (Fig. 6, D and E; and Videos 6 and 7). Although we were not surprised that GFP-PDK1_K465E mutant, which is not able to bind PIP3, did not colocalize with MRCKα, more interesting was the lack of colocalization between GFP-PDK1_L155E and MRCKα (Fig. 6 E and Video 7). This indicates that both HM/PIF-binding pocket and plasma membrane anchorage are required to increase PDK1 and MRCKα colocalization in response to EGF.

TIRF microscopy experiments suggested that PDK1 and MRCKα specifically colocalize at protrusive regions of the plasma membrane. We investigated whether these regions could be considered lamellipodia by transfecting cells with a plasmid coding for LifeAct-mTurquoise, which expressed a peptide that binds polymerized actin (F-actin), and by performing supplemental TIRF experiments. The colocalization of MRCKα/PDK1 upon EGF stimulation was much more evident in highly dynamic cellular protrusions with homogenous distribution of F-actin and rear-delimited by a more stable region characterized by condensed linear actin bundles (Fig. 7 A and Video 8). This description is coherent with the definition of lamellipodium and lamella, respectively (Ponti et al., 2004; Cai et al., 2008; Dang et al., 2013). The spatial proximity between PDK1 and MRCKα in lamellipodia was confirmed by in situ PLA and TIRF microscopy observation. The distribution of fluorescent spots showed a sharp accumulation in lamellipodia at 100 and 200 s upon EGF stimulation (Fig. 7 B). The overall number of fluorescent spots per cell significantly increased upon EGF stimulation, reaching a maximum at around 200 s (Fig. S4 A).

Starting from these observations, we reasoned that PDK1 could regulate MRCKα localization to lamellipodia. To investigate this hypothesis, we evaluated by immunofluorescence the localization of MRCKα in cells either knocked down or overexpressing PDK1_WT and PDK1 mutants. The accumulation of MRCKα in lamellipodia by EGF stimulation (Fig. 8 A) was significantly reduced in PDK1-silenced cells (Fig. 8 B). In contrast, PDK1_WT overexpression determined an increase of cells showing high MRCKα localization to lamellipodia (Figs. 8 C and S4 B). A similar phenotype was observed in cells expressing PDK1_KD and PDK1_Δ50, whereas both PDK1_ΔPH and PDK1_L155E failed to allow the accumulation of MRCKα in lamellipodia (Fig. 8 C).

The results showing that both PDK1 and MRCKα were enriched in lamellipodia, together with their role in directional migration, suggested that PDK1 could regulate lamellipodia dynamics through MRCKα. In response to EGF stimulation, MCF10A cells rapidly produced lamellipodia, which cause the extension of the cell surface to a maximum around 200–300 s from stimulation (Fig. 9 A). The extension of lamellipodia was then followed by lamellipodial retraction (Fig. 9 A). We precisely quantified these lamellipodia dynamics with a method based on the measurement of impedance as an indicator of cell surface extension. This method allows to clearly visualize both protrusion and retraction phases of lamellipodia on the whole cell population (see Materials and methods; Fig. 9 B).

PDK1 silencing significantly reduced both the protrusion and retraction phases (Fig. 9, C, D, and E). Furthermore, PDK1 overexpression increased both protrusion and retraction induced by EGF (Fig. 9, F, G, and H). Remarkably, MRCKα silencing specifically blocked the promoting effects of PDK1 overexpression in a retraction slope but not in a protrusion slope (Fig. 9, F and H). Collectively, these results indicate that PDK1 controls lamellipodia dynamics, but that only the retraction phase is completely dependent on PDK1-mediated regulation of MRCKα.

Cell migration plays a critical role in tumor cell dissemination and tissue invasion. Although most studies on this subject focus on two-dimensional setups, it is now well established that cell migration in three-dimensions displays distinctive features that cannot be found in 2D. To test whether PDK1 also drives cell migration and invasion in a 3D model of collective migration, we used MCF10DCIS.com cells, derived from MCF10A, which form DCIS-like lesions in in vivo mouse models, similar to primary human DCIS lesions, and spontaneously progress to invasive cancer.

Single MCF10DCIS.com cells seeded and embedded in 3D extracellular matrix gel grew as multicellular spheroids, which occasionally produced invasive protrusions (Fig. 10 A). Overexpression of PDK1_WT and PDK1_KD in MCF10DCIS.com cells resulted in a dramatic increase of the invasiveness when compared with control cells (Fig. 10, B and C). The overexpression of PDK1 affected the spheroidal structure and the actin basal localization observable in control spheroids. Moreover, PDK1 expression induced multicellular protrusions, typically composed by several cells, that maintained cell-to-cell contacts with actin cortical distribution (Fig. 10 D). We observed that the tip cells of these protrusions showed a particular actin enrichment at their leading edge (Fig. 10 D).

The quantification of the invasiveness of spheroids confirms the pro-invasive effect of both PDK1_WT and PDK1_KD overexpression (Fig. 10, E, F, and G). The PDK1 kinase–independent effect suggests a mechanism similar to that described in 2D migration. Thus, we verified whether MRCKα mediated the pro-invasive action of PDK1 by stably silencing MRCKα in MCF10DCIS.com overexpressing PDK1_WT or PDK1_KD (Fig. 10 E). The MRCKα knockdown reduced the invasive ability of spheroids overexpressing PDK1 to the level of control cells (Fig.10, F and G). In contrast, ROCK1 silencing (Fig. 10 H) did not significantly affect the number of invasive spheroids (Fig. 10 I) and invasive cells/spheroid (Fig. 10 J).

Altogether, these data suggest a specific role of PDK1 in 3D collective migration and invasion.

Lamellipodia, together with filopodia, are the most frequently observed protrusive structures in cells migrating in a 2D environment (Ridley, 2011). Nevertheless it has been shown that lamellipodia are also present in 3D migration, together with protrusive structures that cannot be found in 2D, such as lobopodia or blebs (Petrie and Yamada, 2012; Petrie et al., 2012;). Here we describe a new regulative pathway involved in cell migration that links EGFR signaling to lamellipodia retraction through PDK1 and MRCKα.

We report that breast epithelial cells migrate toward an EGF gradient via a PI3K-dependent mechanism, in which PDK1 covers an essential role. Furthermore, PDK1 overexpression increases directional cell migration in nontransformed MCF10A cells and the invasive properties of preinvasive MCF10DCIS.com cells. This is consistent with data showing that PDK1 is amplified in some breast (Maurer et al., 2009) and prostate cancers (Choucair et al., 2012). Additionally, its overexpression correlates with more aggressive and invasive phenotypes (Maurer et al., 2009; Gagliardi et al., 2012; Scortegagna et al., 2013). Unexpectedly, we found that PDK1 exerts its pro-migratory activity in a kinase-independent manner. Indeed, PDK1 usually achieves its biological functions by phosphorylating substrates, such as Akt and RSK. However, it was reported that PDK1 activates, through kinase-independent mechanisms, Ral-GEF (Tian et al., 2002), PLCγ1 (Raimondi et al., 2012), and ROCK1 (Pinner and Sahai, 2008). Moreover, we found that PDK1-dependent regulation of cell migration requires the integrity of the PDK1 PIF-binding pocket. For this reason, we focused our attention on Rho-GTPase–associated kinases, which have a highly conserved HM able to bind the PIF-binding pocket of PDK1, but lack a PDK1 phosphorylation site. All members of this family are myosin activators that directly phosphorylate MLC2 or the myosin phosphatase target subunit-1 MyPT1 (Pearce et al., 2010). The role of myosin contraction in migrating cells has traditionally been reported to be confined to tail retraction, but recent new data suggest a broader role, particularly in 3D cell migration (Petrie et al., 2012). Our data show that MCF10A migration toward the EGF gradient requires myosin activity, which is controlled by MRCK but not ROCK activity. Although ROCK is required for amoeboid-like cell migration, it has been suggested that MRCK is involved in mesenchymal cell migration (Wilkinson et al., 2005). Actually, MCF10A cells, which are nontransformed epithelial cells, retain the ability to migrate both as single cells, in a mesenchymal-like manner, and collectively, as an epithelial sheet. Therefore, it is conceivable that MRCK, instead of ROCK, regulates the MCF10A migration. In fact, MRCKα knockdown or inhibition hampers PDK1-induced directional cell migration, which suggests that MRCKα is locally activated in a PDK1-controlled manner. We also demonstrate that MRCKα associates with PDK1 through the interaction of the PDK1 PIF-binding pocket. Less clear is the role of MRCKβ, which associates with PDK1 but is not required for PDK1-induced migration. However, the silencing of MRCKβ up-regulates MRCKα expression, which may compensate for the loss of MRCKβ.

Experiments with spheroids of MCF10DCIS.com support the involvement of PDK1/MRCK signaling even during collective migration in a three-dimensional environment, which suggests a significant role for MRCK in the invasive process. Interestingly, it has been previously reported that collective invasion of the squamous cell carcinoma was dependent on MRCK signaling (Gaggioli et al., 2007).

Despite the direct interaction with MRCKα, PDK1 is not able to activate a purified MRCKα in vitro while it regulates MRCKα activity in intact cells or on crude cell lysates. This suggests an indirect mechanism of activation in which, for example, PDK1 could compete with a negative regulator of MRCKα. A similar model has been already proposed for ROCK1 (Pinner and Sahai, 2008). In any case, further efforts must be undertaken in order to understand such an unconventional type of regulation.

Conversely, results showing that PDK1, upon binding with PIP3, colocalizes with MRCKα, indicate that the interaction between PDK1 and MRCKα is increased by membrane translocation, similar to as previously described for PDK1 and Akt (Alessi et al., 1997). The regulation of MRCKα activity and the concomitant membrane localization constitute a precise spatio-temporal regulation orchestrated by PDK1 downstream of PI3K activation. This type of regulation is typically involved in processes that require a fine tuning of spatially and temporally localized intracellular signals, such as directional cell migration. In MCF10A cells, this regulation is particularly evident in lamellipodia retraction. We show that MRCKα does not regulate the formation phase of lamellipodia, but in contrast is involved in the lamellipodia retraction. However, PDK1 is able to regulate both lamellipodia protrusion and retraction, the latter being exerted through MRCKα. The localization of MRCKα in membrane protrusions has been previously reported, but its role in lamellipodia retraction and cell migration has not been fully explored (Tan et al., 2008).

Molecular mechanisms leading to lamellipodia formation or regulating cyclic protrusion of the leading edge have been described previously (Giannone et al., 2004; Bisi et al., 2013). PKA inhibition, for example, reduces the number of protrusions formed at the leading edge and increases their duration through a mechanism mediated by RhoA and RhoGDI (Tkachenko et al., 2011).

In contrast, the signaling pathways involved in lamellipodia retraction have received less attention, and the role of myosin in this process is particularly unclear. In the traditional model of cell motility, the generation of new actin filaments ensures lamellipodial growth, and protrusion continues until myosin II pulls the rear of the lamellipodial actin network, causing edge retraction and initiation of new adhesion sites (Ponti et al., 2004). In this model, it has been reported that MLCK, a kinase that phosphorylates myosin, is bound to the lamellipodial edge but that MLCK activation of myosin II is concentrated at the back of the lamellipodia, which suggests that periodic contractions release a fraction of MLCK that will be transported as periodic waves with actin filaments (Giannone et al., 2007). In contrast, a recently proposed model suggested that at the peak of protrusion myosin II filaments form in the lamellipodium, and a local network occurs that drives actin-arc formation and edge retraction (Burnette et al., 2011). In this model, MRCK localization to lamellipodium could locally activate myosin II and then could promote cell motility (Burnette et al., 2011). Whether PDK1/MRCK complex controls lamellipodia retraction through a mechanism like that of MLCK or if it activates myosin directly in the lamellipodium has yet to be investigated.

In summary, our work describes a new pathway regulating directional cell migration, which, through PI3K activity and PDK1 translocation to the membrane, activates MRCKα in cell protrusions resulting in lamellipodia retraction (Fig. 9 K). In addition, these findings call for a novel role for lamellipodia retraction linked to directional cell migration.

MCF10A (CRL-10317), 293T (CRL-11268), and HeLa (CCL-2) cell lines were obtained from the ATCC resource center. MCF10DCIS.com (Miller et al., 2000) were kindly provided by the G. Scita laboratory (IFOM, Milan, Italy). All experiments were performed on cell lines that had been passaged for <3 mo after thawing. 293T and HeLa cells were cultured in DMEM (catalogue no. D5546; Sigma-Aldrich). The culture medium was supplemented with 10% FBS (catalogue no. 10270; Gibco), 200 U/ml of penicillin, and 200 µg/ml streptomycin (catalogue no. 67513; Sigma-Aldrich). MCF10A and MCF10DCIS.com cells were cultured as described previously (Debnath et al., 2003). SiRNAs were transfected in MCF10A cells using Oligofectamin (Life Technologies). Pools of siRNA targeting MRCKα (L-003814-00), MRCKβ (L-004075-00), ROCK1 (M-003536-02), and control nontargeting siRNA #1 (D-001810-01-05) were purchased from GE Healthcare.

Rabbit anti-Akt1 (2H10), rabbit anti-pT308Akt (244F9), rabbit anti-pS241PDK1, rabbit anti-PDK1, rabbit anti-pS19MLC2, rabbit anti-pT18/S19 MLC2, and rabbit anti-MLC2 were purchased from Cell Signaling Technology. Rabbit anti-MRCKα (H-90), goat anti-actin (C-11), mouse anti-GST (B-14), goat anti-ROCK1 (K-18), mouse anti–c-Myc (SC-40), and mouse anti-PDK1 (E-3) were from Santa Cruz Biotechnology, Inc. Mouse anti-FLAG and mouse anti-CDC42BPB clone 2F4 (anti-MRCKβ) were from Sigma-Aldrich. Rabbit anti-pT696MyPT1 was from EMD Millipore. Rabbit anti-MyPT1 was from EMD Millipore. Rabbit anti-GFP was from Life Technologies. Human EGF was purchased from R&D Systems. Human HGF was from PeproTech. LY294002 was from Cell Signaling Technology. Y-27632 was from EMD Millipore. Chelerythrine chloride was from Sigma-Aldrich. GSK2334470 was provided by the MRC Protein Phosphorylation Unit. Alexa Fluor 647, 555, and 488 Phalloidin were obtained from Invitrogen.

Myc-tagged PDK1, PDK1-KD (K111N), PDK1-ΔPH, PDK1-L155E, and PDK1-Δ50, previously cloned into PINCO retroviral vector (Primo et al., 2007), were subcloned into a third-generation lentiviral vector pCCL.sin.cPPT.polyA.CTE.eGFP.minhCMV.hPGK.Wpre (Follenzi et al., 2000; Amendola et al., 2005) as described previously (Gagliardi et al., 2012). K465E point mutation was introduced using the following primers: FW (5′-GGCCCAGTGGATGAGCGGAAGGGTTTATTT), and RE (AAATAAACCCTTCCGCTCATCCACTGGGCC-3′).

PDK1_WT, PDK1-KD (K111N), PDK1-L155E, and PDK1_K465E mutants were cloned in pDONR/zeo using Gateway BP Clonase (Invitrogen). To perform the BP recombination, we performed a PCR reaction using the following primers: attB1-PDK1 (5′-ACAAGTTTGTACAAAAAAGCAGGCTTAACCATGGCCAGGACCACCAGCCAGCTGTATGAC-3′), attB1-PDK1Δ50 (5′-ACAAGTTTGTACAAAAAAGCAGGCTTAACCATGGACGGCACTGCAGCCGAGCCTC-3′), attB2-PDK1 (5′-ACCACTTTGTACAAGAAAGCTGGGTTGCCCTGCACAGCGGCGTCCGGGTGG-3′), and attB2-PDK1ΔPH (5′-ACCACTTTGTACAAGAAAGCTGGGTTGCCCTGGTGCCAAGGGTTTCCGCCAGCCTG-3′).

MRCKα WT and MRCKαCAT_WT were provided by the T. Leung laboratory (Institute of Molecular and Cell Biology, A-STAR, Singapore; Leung et al., 1998; Tan et al., 2001b) and cloned into pDONR/zeo using Gateway BP Clonase (Invitrogen). We used the following primers: attB1-MRCKα (5′-ACAAGTTTGTACAAAAAAGCAGGCTTAACCATGTCTGGAGAAGTGCGTTTGAGGCAGTTGGAGCAG-3′), attB2-MRCKα (5′-ACCACTTTGTACAAGAAAGCTGGGTTGCCCGGGTCCCAGCTCCCGCGGTC-3′), and attB2-MRCKαCAT (5′-ACCACTTTGTACAAGAAAGCTGGGTTGCCCTGCAGAGCTTGGACAGTCTGTGTTGACTCTTG-3′).

The sequence of catalytic domain of MRCKβ, Addgene plasmid 50759 (Ando et al., 2013), was cloned into pDONR/zeo using the following primers: attB1-MRCKβ (5′-ACAAGTTTGTACAAAAAAGCAGGCTTAACCATGTCGGCCAAGGTGCGGCTCAAGAAGCTGGAGC-3′) and attB2-MRCKβCAT (5′-ACCACTTTGTACAAGAAAGCTGGGTTGCCGTGGAGGGACTGCACGGTCTGGGTGGAC-3′).

The constructs previously cloned into pDONR/zeo were transferred through Gateway LR Clonase in the following destination vectors: pcDNA-DEST47 (C-Term GFP), in pcDNA-DEST53 (N-Term GFP), in pcDNA-C-mCherry (C-term mCherry), and in pDEST27 (GST-N-Term Tag; all from Invitrogen).

Lentivirus vectors were produced as described previously (Gagliardi et al., 2012). In brief, for PDK1 stable silencing, two pLKO.1 lentiviral vectors carrying PDK1 targeting shRNA called, respectively, shPDK1 #79 (TRCN0000039779; Sigma-Aldrich) and shPDK1 #81 (TRCN0000039781) were used. For MRCKα and ROCK1 stable silencing, we used the following constructs: TRCN0000001332, TRCN0000001333, TRCN0000195202, and TRCN0000121312. A vector leading the expression of a scrambled not targeting shRNA, called shScr, Addgene plasmid 1864 (Sarbassov et al., 2005), was used as a negative control. pCCL.sin.cPPT.polyA.CTE.eGFP.minhCMV.hPGK.Wpre lentiviral vector (Follenzi et al., 2000; Amendola et al., 2005) was used for PDK1 constructs expression. This vector led the expression, through a bidirectional hPGK promoter, of both PDK1 constructs and GFP. A plasmid expressing only GFP was used as a negative control (empty vector). All viruses were produced as described in The RNA Consortium’s shRNA guidelines. Cell infection was performed with an MOI equal to 1 for pLKO.1 and an MOI equal to 5 for pCCL.sin.cPPT.polyA.CTE.eGFP.minhCMV.hPGK.Wpre lentiviral vectors, in the presence of 8 µg/ml Polybrene (H-9268; Sigma-Aldrich).

1 d before the assay, MCF10A cells were split in order to reach a confluence <30% at the time of the assay. MCF10A were detached from culture plates using a Trypsin-EDTA solution (Sigma-Aldrich), then resuspended in DMEM at 2.5 × 10⁵ density. 400 µl of cell suspension (containing 10⁵ cells) were seeded into 8.0-µm pore size inserts (BD). Cells were left to migrate toward 750 µl of DMEM containing various concentrations of EGF or HGF and compared with DMEM alone, as a negative control. Then cells were maintained in a humidified incubator at 37°C plus 5% CO₂ for 6 h (EGF) or 24 h (HGF). Nonmigrated cells were removed with a cotton tip. The inserts were fixed in PBS and 2.5% glutaraldehyde for 20 min and stained with Crystal Violet dissolved in H₂O 20% Methanol. Two fields for each insert were imaged with a 5× objective lens. Cells were counted by image thresholding with ImageJ.

Wound healing scratch assay was performed as described previously (Primo et al., 2010). In brief, 2 × 10⁶ GFP-tagged MCF10A were seeded on 24-well plates the day before the assay. Cells were wounded by dragging a plastic pipette tip across the cell layer. Alternatively, GFP-tagged MCF10A were seeded at low density (10⁴ cells) on 24-well plates to ensure a single cell distribution. Cell movement was followed by inverted-phase contrast and fluorescent time-lapse microscopy (Microsystems LAS AF 6500/7000 [Leica] equipped with an HC PL FLUOTAR 10×/0.30 PH objective lens) every 20 min. Time-lapse image sequences of wound healing or sparse cell motility were analyzed and cell tracked with MetaMorph 7.0 (Molecular Devices). The persistence was calculated as the distance between the start and the end point divided by the path length of a cell trajectory, whereas the FMI was calculated as the orthogonal projection in the direction of migration of the distance between the start and the end point divided by the path length of a cell trajectory.

Viability was evaluated with an MTT assay. Cells were cultured for 24 h in 6-well plates, then 0.2 mg/ml of MTT in culture medium without Phenol red was added. After 3 h of incubation with MTT, the supernatant was removed, and 200 µl of DMSO was added to dissolve the formazan crystal. The optical density value of each sample was measured at a wavelength of 595 nm.

HeLa cells stable transduced with empty vector, PDK1_WT, and PDK1_L155E lentivirus were transfected with pXJ40-FLAG-MRCKα (Leung et al., 1998; Tan et al., 2001b). Cell proteins were extracted with lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM PMSF, 100 µM ZnCl₂, 1% Triton X-100, 1 mM Na₃VO₄, 10 mM NaF, and 1:1,000 protease inhibitor cocktail; Sigma-Aldrich). Cell lysates were incubated with anti-Flag antibody (Sigma-Aldrich) or negative control mouse IgG for 2 h; immune complexes were recovered on anti–mouse IgG-Agarose (Sigma-Aldrich). Proteins were separated by SDS-PAGE electrophoresis, transferred to polyvinylidene difluoride (PVDF) membrane (Bio-Rad Laboratories), incubated with primary antibody, and visualized by ECL.

pDEST27-MRCKα_CAT or pDEST27-MRCKβ_CAT plasmids were transfected in 293T cells with calcium phosphate (Promega). After 36 h, cell lysates were extracted using lysis buffer. GST-tagged protein was isolated through Glutathione Sepharose 4B beads (GE Healthcare). Meanwhile, lysates from MCF10A cells stably infected with the empty vector, PDK1_WT, PDK1_L155E, or PDK1_K465E were extracted using the same lysis buffer. The GST-tagged proteins isolated from 293T bound to glutathione beads were used to pull down proteins from MCF10A extracts. The pulled-down proteins were dissociated using reducing Laemmli buffer (62.5 mM Tris-HCl, pH 6.8, 2% SDS, and 10% glycerol) and analyzed by immunoblotting.

MRCKα kinase assay was performed by transfecting pDEST27-GST-MRCKα in 293T or HeLa cells overexpressing or silenced for PDK1. After 36 h, cell lysates were extracted using the kinase buffer (25 mM Hepes, 300 mM NaCl, 1 mM PMSF, 1.5 mM MgCl₂, 0.5% Triton X-100, 20 mM Na-β glycerophosphate, 1 mM Na₃VO₄, 0.2 mM EDTA, and 1:1,000 protease inhibitor cocktail; Sigma-Aldrich). Purified GST-MRCKα was assayed for its ability to phosphorylate T696 of recombinant MyPT1 (CSA001; Millipore).

PLA was performed according to the Olink Bioscience’s protocol using Duolink II reagents with minor changes. After blocking and primary antibody staining, PLA anti–mouse MINUS and anti–rabbit PLUS probes were diluted 1:5 in PBS 1% donkey serum and incubated for 1 h at 37°C in a preheated humidity chamber. Subsequent ligation and amplification steps were performed using Duolink II Detection kit according to the manufacturer’s instructions.

GFP-PDK1_WT, GFP-PDK1_KD, GFP-PDK1_L155E, GFP-PDK1_K465E, mCherry-MRCKα, LifeAct-EGFP (Riedl et al., 2008), and LifeAct-mTurquoise (Addgene Plasmid 36201; Goedhart et al., 2012) were transfected in MCF10A cells using X-tremeGENE Transfection Reagents (Roche). Then cells were seeded at low density on glass bottom plates (MatTek Corporation) coated with 1 µg/ml fibronectin (Sigma-Aldrich). After starvation overnight in medium without growth factors (DMEM supplemented 200 U/ml of penicillin and 200 µg/ml streptomycin), cells were placed on an inverted microscope equipped with a 37°C humidified chamber with 5% CO₂, and visualized using a fluorescence microscope (True MultiColor Laser TIRF Leica AM TIRF MC; Leica) equipped with a 63× oil immersion objective lens (HCX Plan-Apochromat 63×/1.47 oil CORR TIRF) and an EM-CCD camera (C9100-02; Hamamatsu Photonics). Images were acquired with LAS AF6000 modular system software (Leica). We quantified fluorescence intensity in regions of interest in multiple channels with ImageJ software. Colocalization analysis was performed with the Imaris 6.3 (Bitplane) software colocalization module by determining thresholds in a systematic way. In the image histogram of the prestimulus fluorescence channels, we identified two distinct peaks: one for the background and one for the fluorescence intensity within the cell. The mode of the second peak for each channel was used as the threshold for the whole sequence of images. Once the images were thresholded, we considered the pixels that were in the intersection of the two thresholded images (in each time frame) to be colocalized pixels. For each time frame, we then calculated the sum of the intensity of the MRCKα channel in the intersected pixels divided by the total MRCKα intensity within the thresholded image and multiplied it by 100.

MCF10A cells cultured on chamber slides were fixed in freezing cold methanol or in 3.7% paraformaldehyde, permeabilized for 10 min in PBS 0.1% Triton X-100 (T8294; Sigma-Aldrich), washed three times with PBS, and saturated in 10% donkey serum (Sigma-Aldrich). The primary antibodies were left on the slices overnight in PBS plus 10% donkey serum at a 1:100 dilution at 4°C. The secondary staining was performed at 25°C for 1 h with fluorescent dye–conjugated antibodies (Alexa Fluor 488, Alexa Fluor 555, and Alexa Fluor 405; Invitrogen). Images were acquired at room temperature with a confocal laser-scanning microscope (SPEII DM5500 CSQ; Leica) equipped with a 63×/1.30 HCX Plan-Apochromat oil-immersion objective lens (ACS APO 63×/1.30 oil CS 0.17/E, 0.16) using Leica LAS AF software. Images were analyzed with ImageJ software.

Lamellipodia formation was measured by the Xcelligence system (Acea Biosciences, Inc.). 3 × 10³ MCF10A cells were seeded into each well of E-plate 16. After 24 h of adhesion, cells were starved overnight in 100 µl of serum-free DMEM. The impedance was measured with an RTCA DP Analyzer (Acea Biosciences, Inc.) every 7 s for 5 min after the beginning of the measurement. 100 µl of 5 ng/ml EGF was added to each well, and cells were monitored for about an hour. The baseline Δ cell index was calculated using as Δ time the time of EGF addition and as a baseline the unstimulated cells curve. The protrusion slope was calculated on the first 100 s from EGF stimulus, whereas the retraction slope was calculated from the peak of the curve to 600 s after the peak itself. To visualize time-dependent lamellipodia extension and retraction at TIRF microscopy, we infected MCF10A with a virus expressing LifeAct-EGFP.

3D culture of MCF10DCIS.com spheroids was performed by seeding isolated MCF10DCIS.com cells (10³) embedded in 50 µl of Matrigel. Medium was refreshed every 2–3 d. After 8 d, spheroids were fixed in 3.7% paraformaldehyde, permeabilized for 10 min in PBS with 0.5% Triton X-100 (T8294; Sigma-Aldrich) and stained with Phalloidin-555 and DAPI. Spheroid images were acquired at room temperature with a confocal laser-scanning microscope (SPEII DM5500 CSQ; Leica) equipped with a 20× oil-immersion objective lens (ACS APO 20×/0.60 Imm CS) using Leica LAS AF software. Images were analyzed by ImageJ software. The percentage of invasive spheroids and the extent of invasion were obtained by manually counting the presence and the number of cells invading the extracellular matrix from the spheroid, identifying invading cells by both DAPI and Phalloidin staining.

For box plot representations, the central line depicts median values; the upper and the lower hinges represent the 75th and 25th percentiles, respectively; and the upper and the lower whiskers represent the 90th and 10th percentiles, respectively. All the remaining data are represented as mean value of both technical and biological replicates, whereas error bars give the standard error of the mean. Statistical analysis was performed using GraphPad Prism software (GraphPad Software). Statistical significance was determined by a Student’s t test or one-way analysis of variance, with P < 0.05 considered significant.

Fig. S1 shows additional experiments and controls confirming that PDK1 specifically regulates directional cell migration toward EGF. Fig. S2 shows additional experiments supporting the kinase-independent role of PDK1 in directional cell migration toward EGF and HGF. Fig. S3 shows PDK1 pull-down with GST-MRCKβ_CAT and additional experiments on PDK1 regulation of MRCKα activity. Fig. S4 shows the quantification of the PLA signal with TIRF microscopy and an immunofluorescence of MRCKα in PDK1-overexpressing cells compared with control cells. Tables S1 and S2 show the Blast search for proteins having highly conserved HM consensus sequences. Videos 1 and 2 show time-lapse movies of wound healing assays made with cells silenced for or overexpressing PDK1, respectively. Videos 3, 4, 5, 6, and 7 are the time-lapse movies relative to the images shown in Fig. 6. Video 8 is the time-lapse movie relative to the images shown in Fig. 7.

We thank Dr. Giorgio Scita for suggestions and critical reading of the manuscript.

This work was supported by Associazione Italiana per la Ricerca sul Cancro (AIRC) investigator grants IG (10133 to F. Bussolino, and 9158 to L. Primo); AIRC 5x1000 (12182 to F. Bussolino), Fondazione Piemontese per la Ricerca sul Cancro-ONLUS (Intramural Grant 2010 to L. Primo), Fondo Investimenti per la Ricerca di Base RBAP11BYNP (Newton to F. Bussolino and L. Primo), University of Torino Compagnia di San Paolo (RETHE to F. Bussolino, GeneRNet to L. Primo), FP7-ICT-2011-8 Biloba (contract 318035), and Consiglio Nazionale delle Ricerche (CNR; grant “Personalized Medicine” to F. Bussolino). P.A. Gagliardi is supported by a triennial FIRC fellowship (15026).

The authors declare no competing financial interests.