Screening for regulators of cortical actin organization

A375 melanoma cells were transfected with siRNA against a set of cytoskeletal regulators (including the entire Rho-GTPase family) and other signalling molecules, and then plated on deformable collagen gels. After 24 h, cells were stained with phalloidin and propidium iodide to reveal the organization of F-actin and overall cell morphology. Approximately 90% of control cells had a rounded morphology with prominent cortical F-actin and membrane blebs. Depletion of RhoA, ROCK1 or ROCK2 disrupted cortical F-actin and produced a more elongated cell morphology (Supplementary Information, Fig. S1). Depletion of paxillin, vinculin and PDK1 altered the organization of cortical F-actin, and cells depleted of Dia1, Dia2 and RhoU had less pronounced phenotypes. In similar experiments, A375 cells were plated on deformable gels for 18 h, then treated with inhibitors for 6 h before fixing and staining (inhibitors and their major targets are listed in Supplementary Information). Consistent with our previous results, we found that TAT-C3 and Y27632 disrupted the rounded cellular morphology⁷ (Supplementary Information, Fig. S1). Of the genes identified in this screen, paxillin, vinculin, Dia1, Dia2 and RhoU have been implicated in cell motility^{18, 19, 20, 21, 22, 23, 24}, whereas less is known about the role of PDK1; we therefore decided to focus on the role of PDK1.

Depletion of PDK1 disrupts cortical actin organization and cell motility

To confirm that these phenotypes were not caused by 'off-target' effects, individual siRNA duplexes were used. Three different siRNA duplexes targeting PDK1 disrupted the rounded morphology and actin organization of A375 cells (Fig. 1a, b and data not shown). The effect of PDK1 siRNA could be reversed by expressing a PDK1 cDNA resistant to the siRNA sequence (discussed below). Higher-resolution analysis showed that cells depleted of PDK1 were elongated, with multiple long protrusions, whereas control cells had strong cortical F-actin and membrane blebs (Fig. 1a). We quantified the morphology of cells depleted of PDK1 by measuring the perimeter and area of more than 100 cells (for a round cell, perimeter²/4area = 1). This value was approximately 1.5 for control cells, indicating that they are generally rounded, whereas PDK1-depleted cells had increased values of 2.5–3 (Fig. 1c). Treatment with the ROCK inhibitor Y27632, which is known to disrupt the rounded morphology of A375 cells⁷, gave a perimeter²/4area value of approximately 3, indicating that the effect of PDK1 depletion was almost as marked as that of total ROCK inhibition. Interestingly, depletion of PDK1 had no effect on the morphology of these cells on 2D substrates (Supplementary Information, Fig. S2a). A more elongated morphology was also observed after depletion of PDK1 in MTLn3 and BE cells. Changes to the organization of the cell periphery were also observed in A431 cells, which invade collectively (Supplementary Information, Fig. S2c). These results confirm that PDK1 regulates cell morphology in a range of cancer cell lines.

Figure 1: Knockdown of PDK1 disrupts rounded cell morphology and reduces cell motility.

(a) Upper panels show phase-contrast images of control and PDK1 siRNA-transfected A375 cells on collagen/Matrigel matrix. Middle panels show 3D reconstructions of the F-actin organization in control and PDK1 siRNA-transfected A375 cells; lower panels show single confocal sections. Scale bar is 20 m. (b) PDK1 western blot showing efficacy of siRNA in A375 cells. (c) Quantification of morphology. Ratio was calculated as perimeter²/4area. (median, quartiles and highest and lowest values are indicated on box and whisker plots; ^** indicates P < 0.01, Mann-Whitney U-test; n > 50). (d) Example of control and PDK1 siRNA transfected A375 or MTLn3 cells expressing GFP–CAAX moving on collagen/Matrigel matrix. Scale bar is 20 m. (e) Distance in microns moved by control and PDK1 siRNA-transfected A375 or MTLn3 cells on collagen/Matrigel matrix over 10 h (median, quartiles and highest and lowest values are indicated on box and whisker plots; ^** indicates P < 0.01, Student's t-test; more than 70 cells were quantified for each condition). (f) PDK1 western blot showing efficacy of siRNA in MTLn3 cells (^* siRNA#1 has a single-base mismatch to the rat mRNA, causing reduced knockdown efficiency). Loading control was tubulin.

Full size image (127 KB)

Next we investigated whether depletion of PDK1 affects cell motility. When plated on a deformable substrate, A375 cells moved with a rounded, blebbing morphology characterized by very rapid translocation of the cell body (up to 5 m min^-1) and frequent changes in direction (Fig. 1d). This highly dynamic behaviour required both myosin function and actin polymerization (data not shown). Transfection of PDK1 siRNA prevented the formation of membrane blebs and the rapid translocation of the cell body. Numerous filopodia were extended and retracted (Fig. 1d, e and see Supplementary Information, Movies S1, S2); however, motility was inefficient and net cell movement reduced. To confirm that PDK1 knockdown reduced motility, we analysed MTLn3 cells, which move more efficiently by extending short pseudopods while retaining a rounded cell body. In vitro, MTLn3 cells can reach speeds of approximately 2 m min^-1, similarly to those observed for amoeboid cancer cells in vivo²⁵. Knockdown of PDK1 reduced the motility of MTLn3 cells, although PDK1-depleted cells were still able to extend protrusions (Fig. 1e and see Supplementary Information, Movie S3). The reduction in cell motility correlated with the efficiency of PDK1 depletion in MTLn3 cells (Fig. 1f). Despite the differences in behaviour and speed, we have shown previously that invasion by both cell types is dependent on ROCK-mediated regulation of the cortical acto-myosin network^{7, 12}. We now demonstrate that PDK1 is important for more persistent amoeboid motility, as well as rapid but inefficient bleb-driven motility. Cell motility on a rigid 2D glass substrate was not affected (Supplementary Information, Fig. S2b).

PDK1 is required for cell motility in vivo

A375 cells stably expressing MLC–green fluorescent protein (GFP) to enable in vivo imaging were infected with a retrovirus containing a PDK1 short-hairpin (sh) RNA cassette or a control empty vector. Cells were then injected subcutaneously into nude mice. Knockdown of PDK1 was stably maintained in vivo (Fig. 2a) and did not significantly affect the rate of tumour growth (data not shown). Two A375 clones stably depleted of PDK1 were analysed and, consistent with the in vitro analysis, both were more elongated than control tumour cells (Fig. 2b, c). Only a minority of control A375 cells were motile in vivo. Figure 2d shows an example of a fast-moving cell and Supplementary Information, Movie S4, shows a range of cell behaviours, including fast movement with a rapidly changing morphology (55%; similar to MTLn3 cells in vitro), slower more chaotic movement with a blebbing morphology (10%; similar to A375 cells in vitro) and extending protrusions without observable movement of the cell rear (35%). Analysis of 14 different areas of four control and four PDK1-depleted tumours also showed that PDK1-depleted cells were less motile than controls (Fig. 2e). Interestingly, in PDK1-depleted tumours, there was a proportionately greater decrease in the number of cells that exhibited rapid, amoeboid movement and chaotic blebbing; the few remaining motile cells extended protrusions within the tumour (Fig. 2f). This correlates well with the in vitro analysis, indicating that PDK1 is required for rapid movement and chaotic blebbing but is not essential for the extension of membrane protrusions. These data confirm the relevance of the altered morphology and motility of PDK1-depleted cells that were observed in vitro.

Figure 2: PDK1 is required for cell motility in vivo.

(a) PDK1 western blot of tumour lysates in control and stably PDK1-depleted A375 tumours (upper panel), total protein levels are shown by Ponceau staining (lower panel). (b) Representative examples of control and PDK1 knockdown in A375 tumours; GFP–MLC in green and collagen in red. (c) Quantification of morphology. Ratio was calculated as perimeter²/4area. (^** indicates P < 0.01, Mann-Whitney U-test; n > 400). (d) Example of an A375 cell (in green, outlined in white) moving into collagen rich matrix (in red) surrounding a tumour. (e) Quantification of the number of moving cells per hour per mm² in control and PDK1-depleted tumours (error bars are s.d.; ^** indicates P < 0.01, Student's t-test; 14 tumour regions imaged for each condition). (f) Quantification of the number of control and PDK1-depleted cells exhibiting different types of motile behaviour (either rapid moving with a smooth amoeboid shape, chaotically moving with bleb-like structures, or extending protrusions without movement of the cell rear) and quantification of cells moving either into tissue surrounding the tumour or within the tumour. The area of the pie-chart segments is proportional to the number of cells. (g) Representative images of mouse lungs injected with GFP-expressing control-transfected cells or mRFP-expressing PDK1 siRNA-transfected cells are shown either 2 or 24 h post injection. Scale bar is 50 m. Bar chart shows the relative proportion of control and PDK1 siRNA transfected cells observed within the lungs at the indicated times (error bars are s.d.; n = 6).

Full size image (136 KB)

Cell motility is also important for extravasation and colonization of new tissues by metastatic cells. To determine whether PDK1 facilitates these processes, the ability of control and PDK1-depleted cells to move from blood vessels to the lungs was assessed. To this end, equal numbers of control and PDK1-depleted cells expressing GFP and mRFP, respectively, were mixed and injected into the tail vein of mice. After 2 h, equal numbers of control and PDK1-depleted cells had lodged in narrow vessels within the lungs (Fig. 2g). After 24 h, the number of PDK1-depleted cells that had entered the lung parenchyma was 60% less than control cells, confirming that PDK1 is required for efficient tumour-cell extravasation.

Organization and phosphorylation of MLC is disrupted in PDK1- depleted cells

The elongated shape of PDK1-knockdown cells suggests that regulation of the cortical acto-myosin network might be aberrant in these cells. The acto-myosin network in cancer cells that move with an amoeboid morphology (such as A375 and MTLn3) is mainly cortical and regulated through phosphorylation¹². In cells that lacked PDK1, pT18S19–MLC and pS19–MLC were no longer localized around the cortex and the total amount of pS19–MLC was reduced (Fig. 3a, b). To determine whether alterations in MLC phosphorylation are a cause or consequence of the altered cell shape following PDK1 depletion, cells were cultured in 2D, where a lack of PDK1 does not affect morphology. Depletion of PDK1 caused a slight reduction in MLC phosphorylation (Fig. 3b), indicating that changes in MLC phosphorylation are not a consequence of changes in cell shape. The organization of phosphorylated Ezrin–Radixin–Moesin (pERM) was also altered in PDK1-depleted cells (Supplementary Information, Fig. S3a); however there was no change in total levels of pERM, suggesting that this change is a consequence of the altered cell morphology (data not shown). The observed changes in cell shape may also be caused by PDK1 modulating the strength of cell matrix adhesions²⁶; however, we found no evidence for this in our system (Supplementary Information, Fig. S3b).

Figure 3: Knockdown of PDK1 disrupts MLC phosphorylation and organization.

(a) Immunofluorescence microscopy of endogenous pS19-MLC and pT18pS19-MLC in control and PDK1 siRNA-transfected A375 cells. Scale bar is 50 m. (b) Western blot of A375 cell lysates showing reduced level of pS19-MLC when PDK1 is knocked down either in cells in 2D or gel culture. Quantification of the reduction in pMLC/MLC, as measured by densitometry. (c) Control, PDK1, or ROCK1 siRNA-transfected A375 cells stably expressing GFP–MLC (in green) plated on collagen/Matrigel matrix (in red). Scale bar is 20 m.

Full size image (104 KB)

Analysis of cells expressing GFP–MLC revealed that the acto-myosin network was localized around the cell cortex, with MLC concentrated around the rear during translocation of the cell body (Fig. 3c, marked with an asterisk, and see Supplementary Information, Movie S5). Imaging of cells that lacked PDK1 showed that MLC–GFP was still mainly localized at the cell cortex but was not correctly coordinated. Cells extended multiple protrusions, but could not organize MLC at the presumptive rear of the cell to drive forward movement (Fig. 3c and see Supplementary Information, Movie S6). This phenotype is similar to that observed in cells that lack ROCK1, although the ROCK1-depleted cells showed a more pronounced change in MLC organization with reduced levels around the cell periphery (Fig. 3c).

PDK1 regulates ROCK1 but not ROCK2 localization

In A375 cells, phosphorylation of MLC is controlled by RhoA and the ROCK kinases²⁷. Thus, we tested whether the observed changes in morphology and motility caused by PDK1 depletion might be due to defective RhoA activation. On the contrary, levels of RhoA-GTP were found to be increased slightly in cells depleted of PDK1 (Supplementary Information, Fig. S4a), indicating that RhoA activation was not defective in these cells and that PDK1 knockdown probably affects signalling downstream of RhoA.

Endogenous ROCK1 was visualized using immunofluorescence microscopy to determine whether its localization depended on PDK1. In control cells, ROCK1 was localized in the plasma membrane, whereas in cells transfected with PDK1 siRNA, it was found mainly in cytoplasmic vesicles (Fig. 4a). We did not observe significant localization of endogenous ROCK2 in the plasma membrane in either control or PDK1-depleted cells (data not shown); however, different subcellular distributions of ROCK1 and ROCK2 have been reported previously for cells grown on 2D substrates²⁸.

Figure 4: PDK1 is required for ROCK1 localization and function.

(a) ROCK1 immunofluorescence microscopy of control and PDK1 siRNA-transfected A375 cells plated on gels. Inset panels show ROCK1 in red and farnesylated CFP in green. Scale bar is 2 m. (b) Immunofluorescence microscopy of control and PDK1 siRNA-transfected A375 cells plated on glass, showing membrane blebbing caused by the overexpression of both Myc–ROCK1 and Myc–ROCK2 (ROCK expression was visualized by 9E10 in red and F-actin was visualized using phalloidin in green). (c, d) Quantification of the F-actin phenotype and ROCK localization in control and PDK1 siRNA-transfected A375 cells overexpressing ROCK constructs (results are means s.d. of > 150 cells from at least three independent experiments; ^** indicates P < 0.01, chi-squared test). (e) Cells expressing GFP–ROCK1 were plated on collagen/Matrigel matrix and imaged at 20 frames min^-1 by confocal microscopy. Shown are representative still images from these sequences demonstrating the differences in localization and corresponding kymographs showing the accumulation of GFP–ROCK1 at the rear of the cell as contraction drives the cell forward when PDK1 is present (i, ii) but not when depleted (iii–v). The line of the kymograph is shown on the corresponding still image (rotated 45 ° clockwise in panel ii and 90 ° counter-clockwise in panel v).

Full size image (97 KB)

These data suggest that depletion of PDK1 may affect the ability of ROCK1 to regulate the cortical acto-myosin network. To investigate whether PDK1 was required specifically for ROCK1 to regulate the cortical acto-myosin machinery, ROCK1 or ROCK2 was overexpressed in PDK1-depleted cells. When plated on 2D substrates, these cells were flattened, with some stress fibres (Supplementary Information, Fig. S2a); however, overexpression of ROCK1 or ROCK2 induced contraction, causing the cells to become rounded and form membrane blebs. The ability of ROCK-family kinases to regulate the cortical acto-myosin machinery can be assessed by monitoring the number of ROCK-transfected cells that exhibited membrane blebs²⁹. Transfection with either ROCK1 or ROCK2 caused membrane blebbing in 60% of cells on 2D substrates, compared with 10% of control GFP-transfected cells (Fig. 4b, c and see Supplementary Information, Fig. S5 for examples of the various categories of morphologies and ROCK1 localization patterns). The ability of ROCK1 to induce blebbing was markedly reduced when PDK1 was knocked down, whereas ROCK2-driven membrane blebbing was unaffected (Fig. 4b, c). Depletion of PDK1 also affected the ability of overexpressed ROCK1 to localize to the plasma membrane (Fig. 4d); this was consistent with the behaviour of endogenous ROCK1. Although ROCK1 was aberrantly localized, it retained some activity as it was able to generate large F-actin cables in the cytoplasm when PDK1 was knocked down (Fig. 4b, c). A similar phenotype was observed when membrane recruitment of ROCK1 was disrupted by either knockdown of RhoA or mutation of the ROCK1/Rho-binding domain (Supplementary Information, Fig. S4b). These data suggest that PDK1 may modulate the localization and activity of ROCK1 at the plasma membrane.

We further analysed ROCK1 localization using live imaging of GFP–ROCK1. A375 cells were plated on a collagen/Matrigel matrix and imaged every 3 s. GFP–ROCK was localized at the cortex and accumulated at the rear during translocation of the cell body (Fig. 4e, kymograph in panel ii, and see Supplementary Information, Movie S7). Cells expressing GFP–ROCK but depleted of PDK1 were elongated, as previously observed. GFP–ROCK was no longer localized at the plasma membrane but was either diffuse in the cytoplasm or found in patches at the membrane (Fig. 4e, panels iii, iv). Even when ROCK1 was asymmetrically localized at the membrane it did not induce cell translocation in PDK1-depleted cells (Fig. 4e, kymograph panel v, and see Supplementary Information, Movie S8).

PDK1 regulates ROCK1 independently of its kinase activity

Examination of the ROCK1 amino-acid sequence showed a potential PDK1-binding site centred around Thr 398 in the hydrophobic extension of the ROCK1 kinase domain (http://scansite.mit.edu/motifscan_seq.phtml) and a phospho-acceptor site in the activation loop of ROCK1, Thr 233, which corresponds to known PDK1 phosphorylation sites in other AGC kinases. We hypothesized that PDK1 may regulate ROCK1 kinase activity by phosphorylating its activation loop. However, the activity of endogenous ROCK1 was unaffected by PDK1 knockdown (Fig. 5a). Furthermore, addition of purified PDK1 did not alter phosphorylation of MLC or RhoE by ROCK1 in vitro (Fig. 5b, lanes 1, 3, c, lanes 5, 7), nor was phosphorylation of ROCK1 amino acids 17–535 by PDK1 observed (Fig. 5c, lane 7), suggesting that regulation of ROCK1 by PDK1 is not dependent on PDK1-kinase activity. Consistently, transfection of PDK1-depleted cells with PDK1 cDNA resistant to siRNA restored the ability of exogenous ROCK1 to induce membrane blebbing (Fig. 5d) and the localization of ROCK1 to the plasma membrane (data not shown) independently of PDK1 kinase activity (Fig. 5d). However, a pleckstrin-homology (PH) domain-deleted PDK1 mutant, which does not localize in the plasma membrane, was unable to restore the ability of ROCK1 to induce blebbing or its localization to the plasma membrane (Fig. 5d and data not shown). Similarly, transfection of either wild-type or kinase-dead siRNA-resistant PDK1 was equally effective at restoring a rounded morphology to cells depleted of PDK1 (Fig. 5e, f). Immunostaining revealed that a proportion of ROCK1 and wild-type or kinase-dead PDK1, but not PH-domain-deleted PDK1 colocalize at the plasma membrane (Fig. 5g). However, expression of PDK1-PH did not interfere with the ability of endogenous PDK1 to recruit ROCK1 to the plasma membrane. These results confirm that PDK1 does not require kinase activity to regulate ROCK1 but does require plasma membrane localization.

Figure 5: PDK1 does not regulate kinase activity of ROCK1.

(a) Endogenous ROCK1 was immunoprecipitated from control and PDK1 siRNA-transfected A375 cell lysates and used to phosphorylate recombinant MLC in vitro. (b) Recombinant ROCK1 (amino acids 17–535) was used to phosphorylate purified MLC in vitro in the presence of -³²P-ATP. Where indicated, recombinant PDK1 or 10 M Y27632 was added. (c) In vitro kinase assay using recombinant PDK1, ROCK1, Akt and RhoE: upper panel shows incorporation of ³²P-phosphate and lower panel shows total protein. PDK1 efficiently phosphorylates Akt and itself, but not ROCK1. (d) Graph quantifying the actin phenotype of PDK1 siRNA-transfected A375 cells co-expressing ROCK1 and siRNA-resistant PDK1 mutants (mean s.d. of > 100 cells from at least two independent experiments; ^** indicates P < 0.01, chi-squared test). (e) The proportion of A375 cells with different morphologies following PDK1 depletion and transfection of wild-type (WT) or kinase-dead (KD) PDK1 resistant to siRNA is shown (mean of two independent experiments, 100 cells counted for each point, ^* indicates P = 0.34, ^** indicates P < 0.01, chi-squared test). (f) F-actin staining of PDK1-depleted cells transfected with either GFP, wild-type or kinase-dead PDK1 resistant to PDK1 siRNA is shown in left-hand panels. Expression of GFP or PDK1 is shown in right panels. Scale bar is 50 m. (g) Immunofluorescence microscopy showing colocalization of FLAG–PDK1 (upper panels) and GFP–ROCK1 (middle panels) and regions of colocalization (pixels with an intensity value > 128/255 in both channels, lower panels) in A375 cells. Scale bar is 20 m.

Full size image (111 KB)

PDK1 binds directly to ROCK1 to regulate ROCK1 localization

We tested whether PDK1 could bind to the hydrophobic extension of ROCK1, as predicted by SCANSITE analysis. Figure 6a shows that purified PDK1 binds to a fragment of ROCK1, including the hydrophobic extension (amino acids 375–415). Although comparison with other AGC-family kinases suggested that binding of PDK1 to this region might require phosphorylation on Thr 398, this was not the case because PDK1 bound equally well to a T398A mutant (data not shown). PDK1 did not bind to an equivalent region of ROCK2, confirming the functional differences between the two kinases. These data suggest that PDK1 may positively influence the ability of ROCK1 to regulate cortical MLC phosphorylation by binding directly to ROCK1 and promoting its localization to the plasma membrane.

Figure 6: ROCK1 associates with RhoE when PDK1 is absent and results in reduced cell motility and blebbing.

(a) Recombinant PDK1 was incubated with the indicated GST-fusion proteins (RhoE, ROCK1 (R1) or ROCK2 (R2)) before washing and analysis by SDS–PAGE and western blotting. (b) The binding of ROCK1 amino acids 17–535 to GST–RhoE was determined in the presence of increasing concentrations of PDK1. Western blot analysis is shown with the antibodies indicated, except Ponceau staining was used for RhoE. (c) Control and PDK1 siRNA-transfected A375 cell lysates were incubated GST–RhoE and, where indicated, exogenous PDK1. The amount of ROCK1 bound to RhoE after washing was determined by western blotting. (d) ROCK1 was immunoprecipitated from control and PDK1 shRNA cell lines and the amount of RhoE bound was determined by western blotting. (e) Quantification of the localization and F-actin phenotype of myc-ROCK1 expression in control, PDK1 siRNA, RhoE siRNA, or combined PDK1 and RhoE siRNA-transfected A375 cells (results are mean s.d. of > 100 cells from three independent experiments; ^** indicates P < 0.01, chi-square test). (f) The actin phenotype of Myc–ROCK1-expressing cells in control, PDK1 siRNA, RhoE siRNA, or combined PDK1 and RhoE siRNA-transfected A375 cells. (results are mean s.d. of > 100 cells from three independent experiments; ^** indicates P < 0.01, chi-squared test). (g) F-actin and pT18pS19-MLC staining of control, PDK1 siRNA, RhoE siRNA, or combined PDK1 and RhoE siRNA-transfected A375 cells plated on collagen/Matrigel matrix. (h) Quantification of morphology. Ratio was calculated as perimeter²/4area (median, quartiles and highest and lowest values are indicated on box and whisker plots; ^** indicates P < 0.01, Mann Whitney test U-test; n > 60). (i) Speed of control. PDK1 siRNA, RhoE siRNA, or combined PDK1 and RhoE siRNA-transfected MTLn3 cells on collagen/Matrigel matrix during 10 h (^** indicates P < 0.01 Student's t-test; over 50 cells were quantified for each condition).

Full size image (109 KB)

ROCK1 associates with RhoE in the absence of PDK1

The fragment of ROCK1 that binds to PDK1 (Fig. 6a) also includes a region required for RhoE binding¹⁵. This suggests that RhoE and PDK1 may bind to a similar region of ROCK1 and therefore, compete with each other. Indeed, ROCK1 bound to RhoE, but the complex formed was disrupted by adding increasing amounts of PDK1 (Fig. 6b). On the basis of these results, endogenous ROCK1 might be expected to associate with RhoE when PDK1 is depleted. In agreement with previous studies¹⁶, we found that GST–RhoE interacted poorly with full-length ROCK1 derived from A375 cell lysates; however, this interaction was significantly increased when PDK1 was depleted (Fig. 6c). The increased binding of ROCK1 to RhoE could be reversed by either the addition of recombinant PDK1 to the cell lysate, or by the co-expression of PDK1 resistant to siRNA knockdown in the PDK1-depleted cells. Furthermore, endogenous ROCK1 co-immunoprecipitated with endogenous RhoE only in cells lacking PDK1 (Fig. 6d). This demonstrates that endogenous PDK1 and RhoE compete for binding to ROCK1.

Combined depletion of PDK1 and RhoE restores acto-myosin contractility

If PDK1 promotes ROCK1-driven organization of cortical F-actin by opposing the inhibitory effects of RhoE, then loss of RhoE should reverse the effect of PDK1 depletion, which disrupts the localization of ROCK1. When both PDK1 and RhoE were depleted, ROCK1 localization to the plasma membrane was restored (Fig. 6e); furthermore, it was able to promote acto-myosin contraction and blebbing (Fig. 6f). Combined depletion of PDK1 and RhoE also restored phosphorylation of endogenous MLC around the cell cortex and produced a more rounded morphology in both A375 and MTLn3 cells (Fig. 6g, h and see Supplementary Information, Fig. S6b, c). This result was verified by quantifying the morphology of more than 60 cells and using an additional siRNA that targeted RhoE (efficacy of siRNA knockdown was confirmed by western blot; see Supplementary Information, Fig. S6a). Crucially, combined depletion of PDK1 and RhoE also corrected the defect in cell motility observed in cells depleted of only PDK1 (Fig. 6i and see Supplementary Information, Movie S9). Together, these data demonstrate that defective regulation of cortical acto-myosin and reduced motility in cells lacking PDK1 is due to inhibition of ROCK1 by RhoE.

Cell lines.

A375, A431 and BE cells were grown in Dulbecco's Modified Eagle's Medium (DMEM) containing 10% fetal calf serum (FCS). MTLn3 cells were grown in MEM containing 5% FCS. Polyclonal cell lines stably expressing GFP–MLC were generated by selection with G418 (1 mg ml^-1) followed by fluorescence-activated cell sorting (FACS) for GFP. Monoclonal cell lines with stable knockdown of PDK1 were generated by transfection with pSuperRetro constructs containing the sequences detailed below. Clones were chosen following selection with puromycin.

Time-lapse microscopy.

Phase contrast: cells were plated on top of a deformable gel (1.8 mg ml^-1 Matrigel, 3.2 mg ml^-1 collagen type 1 (BD Biosciences)) on glass-bottomed cell culture plates (MatTek) 24 h before imaging. Cells were incubated at 37 °C with 10% CO₂ in DMEM containing 10% FCS during imaging. Confocal imaging: A375 cells stably expressing MLC–GFP or transiently transfected with ROCK–GFP or GFP–CAAX were plated on top of deformable collagen gels as detailed above and imaged using a Zeiss LSM510 confocal microscope.

Analysis of cell morphology.

Using Volocity software, the area and perimeter of cells were determined by manually drawing around the cell shape using phase contrast images. Ratios were calculated as follows: perimeter²/4area. Volocity was also used to determine the cell centroid and its translocation.

Expression vectors and mutagenesis.

Full-length Myc–ROCK1 and Myc–ROCK2 have been described previously^{41, 42}. PDK1 was cloned with an amino-terminal FLAG tag into pRK5.1 using EcoR1 sites; kinase and PH domain mutations were K111A and truncation at Ser 408, respectively. Site-directed mutagenesis was then performed to introduce resistance to siRNA oligo1 (by changing TTTTCGACAAGATCCCTA to TTCTCCACGAGGTCTCTT). Stable knockdown of PDK1 was achieved by cloning siRNA sequences (GACCAGAGGCCAAGAAUUUUU, Oligo2) into pSuperRetro vector. GST–ROCK1 (amino acids 338–415, 375–415 and 338–378) and ROCK2 (amino acids 353–433) fragments were cloned into pGEX 6P3 by PCR. pGEX–MLC was made by cloning MLC (image clone AU18-h7 Gene Service) into pGEX–KG using Xba I and HindIII, MLC–GFP is described elsewhere²⁹ and pGEX 2T RhoE was a gift from M. Way (Cancer Research UK).

Transfection of siRNA oligos and expression vectors.

100 nM siRNA oligos were transfected using Oligofectamine diluted 1:200 (for A375, MTLn3 and BE cells) or Dharmafect 2 diluted 1:400 (for A431 cells) in Optimem (Gibco #31985) and harvested approximately 56 h later. For subsequent analysis, the following siRNA targeting sequences were used: PDK1#1 CAAGAGACCUCGUGGAGAAUU, PDK1#2 GACCAGAGGCCAAGAAUUUUU, ROCK1 GCCAAUGACUUACUUAGGAUU, RhoE#2 GAAAUUAUCCAGCAAAUCUUU, RhoE#3 UAGUAGAGCUCUCCAAUCAUU, RhoE#4 AGAAUUACACGGCCAGUUUUU. Control transfections were performed using non-targeting siRNA#1 from Dharmacon (D-001210-01 — UAGCGACUAAACACAUCAA). Expression vectors were transiently transfected using Effectene (Qiagen) 36 h prior to assay end-point. For experiments combining siRNA with plasmid transfection, the siRNA transfection was performed the day before the plasmid transfection and Fugene-6 (6 l ml^-1) was used instead of Effectene for plasmid transfection.

Western blotting.

Standard methods were used for western blotting with either nitrocellulose or PVDF membranes. The following antibodies were used: 9E10, pS19-MLC (Cell Signaling Technology #3671), MLC (Cell Signaling Technology #3672), ROCK1 (Chemicon), RhoE (Sigma) and PDK1 (Cell Signaling Technology #3062).

In vitro kinase assays.

Cells were lysed in lysis buffer (50 mM Tris pH 7.5, 1% Triton X-100, 10 mM MgCl2, 150 mM NaCl, 1 mM dithiothreitol (DTT), 40 mM Na pyrophosphate, 1 mM NaVO4, 2 mM PMSF, 0.025% deoxycholate, 10 g ml^-1 aproteinin, 10 g ml^-1 leupeptin). Endogenous ROCK1 was immunoprecipitated from cells using an anti-ROCK1 antibody (Santa Cruz, sc-6056). Recombinant ROCK1 amino acids 17–535 was obtained from Upstate (14-601). Beads were washed extensively in lysis buffer before the final wash in kinase assay buffer (50 mM Tris pH 7.5, 1 mM EDTA pH 8.0, 10 mM MgCl2, 50 mM NaCl, 0.03% Brij35, 1 mM DTT, 40 mM Na pyrophosphate, 1 mM NaVO4, 2 mM PMSF, 0.025% deoxycholate, 10 g ml^-1 aproteinin, 10 g ml^-1 leupeptin and 10 M ATP. Kinase assay buffer (50 l) and bacterially expressed MLC/reaction (50 ng) were added to the beads and incubated at 30 °C for 30 min. The reaction was stopped by adding 10 l 5 SDS sample buffer and boiling. Phosphorylation of MLC was detected by western blotting using an anti-phospho-S19MLC antibody (Cell Signaling Technology #3671). Equal loading of MLC was determined by Ponceau staining. Equal input of kinase was determined by western blotting: Myc–ROCK1 and endogenous ROCK1 were detected using 9E10 (CR-UK antibody service) and anti-ROCK1 (AB3885 Chemicon International), respectively.

Immunofluorescence microscopy.

Cells were routinely fixed with 4% paraformaldehyde, and permeablized using 0.3% Triton X-100. Samples were blocked for 1 h with 3% BSA/0.1% Triton, in PBS before incubation with the primary antibody. The following antibodies were used: 9E10, JAK6, pS19-MLC (Cell Signaling Technology #3671), pT18S19-MLC (Cell Signaling Technology #3674), phospho-ERM (Cell Signaling Technology #3141), ROCK1 (sc-6056 Santa Cruz) and phospho-Y118 paxillin (Biosource 44-722G). For endogenous ROCK1 and phospho-ERM staining, cells were fixed with 10% TCA in PBS for 15 min, washed three times with 30 mM glycine in PBS and blocked with 3% BSA/0.1% Triton in PBS. Either Molecular Probes or Jackson Stratech secondary antibodies were used at 1:200. F-actin was visualized using Alexa-633, FITC or TRITC coupled to phalloidin. Images were taken with a Zeiss LSM510 confocal microscope. Cell phenotypes were quantified manually by placing cells into categories depending on the organization of the actin cytoskeleton and the localization of the overexpressed ROCK construct.

GST-pulldown and binding assays.

ROCK1–PDK1 interaction: GST–ROCK fragments were incubated with 100 ng purified PDK1 (521270 Calbiochem), diluted in kinase assay buffer (without ATP) for 2 h at 4 °C. Beads were then washed four times with kinase assay buffer. Proteins were eluted with 2 SDS sample buffer and analysed by western blot using an anti-PDK1 antibody (Cell Signaling Technology #3062).

RhoE–PDK1 competition assay: Approximately 0.5 g of GST–RhoE was bound to glutathione–agarose. ROCK1 (amino acids 17–535, 0.2 g) was added to 200 l of kinase assay buffer containing no ATP, and increasing amounts of PDK1 (as indicated) were added up to a maximum of 1.2 g. This mixture was left for 30 min at 4 °C before mixing with GST–RhoE for 90 min and then washing four times in kinase assay buffer. Proteins were eluted with 2 SDS sample buffer and analysed by western blot using anti-PDK1 (Cell Signaling Technology #3062) and anti-ROCK1 (Chemicon) antibodies.

RhoE–ROCK1 pulldown from cell lysates: GST–RhoE bound to glutathione agarose was incubated with A375 cell lysates expressing Myc–ROCK1 for 3 h at 4 °C (lysates made in the same way as for kinase assays). Beads were washed four times with kinase buffer containing no ATP. Bound protein was eluted using 2 SDS sample buffer, then binding was analysed by western blotting using 9E10.

Immunoprecipitation.

Cells were lysed in 20 mM HEPES pH 7.5, 2mM EDTA, 150 mM KCl, and 1 mM DTT containing protease inhibitors and passed through a 19G needle repeatedly before centrifugation at 1000g for 5 min at 4 °C. NP-40 was added to 0.05% before addition of 2 g of an anti-ROCK1 antibody (Santa Cruz, sc-6056) and protein G–sepharose. This mixture was then tumbled for 3 h at 4 °C before three washes with the lysis buffer containing 0.05%NP-40.

Tumour imaging

Nude mice were injected sub-cutaneously with either 106 control or PDK1 shRNA stably transfected A375 cells also expressing MLC-GFP. When tumours were 3-7mm diameter mice were anaesthetised and imaged as described in 12. Four control and four PDK1 depleted tumours were imaged, four different regions were imaged for 20 minutes for each tumour. Moving cells were defined as those that moved 10m or more during a 20 minute movie.

Lung extravasation assay

MTLn3 cells stably expressing either membrane-tethered GFP or mRFP were transfected with either control or PDK1 siRNA#2. 48 hours after transfection cells were trypsinized and mixed in equal numbers prior to injection into the tail vein of nude mice. Mice were sacrificed after either 2 or 24 hours and the lungs examined for the presence of GFP or mRFP expressing cells.

Note: Supplementary Information is available on the Nature Cell Biology website.