L-plastin Ser5 phosphorylation is modulated by the PI3K/SGK pathway and promotes breast cancer cell invasiveness

Background Metastasis is the predominant cause for cancer morbidity and mortality accounting for approximatively 90% of cancer deaths. The actin-bundling protein L-plastin has been proposed as a metastatic marker and phosphorylation on its residue Ser5 is known to increase its actin-bundling activity. We recently showed that activation of the ERK/MAPK signalling pathway leads to L-plastin Ser5 phosphorylation and that the downstream kinases RSK1 and RSK2 are able to directly phosphorylate Ser5. Here we investigate the involvement of the PI3K pathway in L-plastin Ser5 phosphorylation and the functional effect of this phosphorylation event in breast cancer cells. Methods To unravel the signal transduction network upstream of L-plastin Ser5 phosphorylation, we performed computational modelling based on immunoblot analysis data, followed by experimental validation through inhibition/overexpression studies and in vitro kinase assays. To assess the functional impact of L-plastin expression/Ser5 phosphorylation in breast cancer cells, we either silenced L-plastin in cell lines initially expressing endogenous L-plastin or neoexpressed L-plastin wild type and phosphovariants in cell lines devoid of endogenous L-plastin. The established cell lines were used for cell biology experiments and confocal microscopy analysis. Results Our modelling approach revealed that, in addition to the ERK/MAPK pathway and depending on the cellular context, the PI3K pathway contributes to L-plastin Ser5 phosphorylation through its downstream kinase SGK3. The results of the transwell invasion/migration assays showed that shRNA-mediated knockdown of L-plastin in BT-20 or HCC38 cells significantly reduced cell invasion, whereas stable expression of the phosphomimetic L-plastin Ser5Glu variant led to increased migration and invasion of BT-549 and MDA-MB-231 cells. Finally, confocal image analysis combined with zymography experiments and gelatin degradation assays provided evidence that L-plastin Ser5 phosphorylation promotes L-plastin recruitment to invadopodia, MMP-9 activity and concomitant extracellular matrix degradation. Conclusion Altogether, our results demonstrate that L-plastin Ser5 phosphorylation increases breast cancer cell invasiveness. Being a downstream molecule of both ERK/MAPK and PI3K/SGK pathways, L-plastin is proposed here as a potential target for therapeutic approaches that are aimed at blocking dysregulated signalling outcome of both pathways and, thus, at impairing cancer cell invasion and metastasis formation. Video abstract

through its downstream kinase SGK3. Of note, the contribution of the PI3K/SGK3 axis to L-plastin Ser5 phosphorylation strongly depends on the cellular context.
In order to explore the functional outcome of L-plastin expression and, in particular, of L-plastin Ser5 phosphorylation, we have investigated their role in invasion/migration processes, invadopodia formation and extracellular matrix (ECM) degradation. To this end, we have taken two parallel approaches: rst, we have silenced L-plastin expression in two breast cancer cell lines initially expressing endogenous Lplastin and, second, we have neoexpressed L-plastin wild type or the phosphorylation variants L-plastin Ser5Ala (S5A, non-phosphorylatable) or L-plastin Ser5Glu (S5E, phosphomimetic) in breast cancer cells expressing only a low level or no endogenous L-plastin. Our results indicate that L-plastin expression and especially L-plastin Ser5 phosphorylation enhances invasion/migration of breast cancer cells. For immunoprecipitation experiments, HEK 293T cells were transfected with the plasmids pEGFP-N1 L-plastinWT, pEGFP-N1 L-plastinS5E, pEGFP-N1 L-plastinS5A or pEGFP-N1 L-plastinEF-ABD1 using Lipofectamine 2000 (Invitrogen). Cells were harvested 24 h after transfection and used for immunoprecipitation.
HEK 293T cells were used for the production of lentiviral particles. Brie y, HEK 293T cells were transiently transfected with third generation lentiviral vectors using Lipofectamine 2000. The virus-containing supernatant was harvested 24 h and 48 h after medium change, cleared by centrifugation at 2000 rpm and 4 °C for 10 min, and filtered through a 0.45 mm filter. Concentration of lentiviral particles was performed by precipitation with PEG10000 (1:5 volume of 40% PEG10000 solution; Merck KGaA, Darmstadt, Germany) at 4 °C overnight, followed by centrifugation at 2800 rpm and 4 °C for 30 min. The virus pellet was resuspended in serum-free medium, divided in aliquots, and stored at -80 °C. Target cells were transduced in the presence of 8 mg/ml Polybrene (hexadimethrine bromide, Merck) for 16 h. The transduced cells, positive for green fluorescent protein (GFP) expression, were selected with 1 mg/ml puromycin in complete medium for 48 h.

Plasmids
The plasmid pEGFP-N1 L-plastinWT used for transiently transfecting HEK 293T cells was generated from the previously described plasmid pDsRed-Monomer-N1 L-plastinWT (12). Brie y, the L-plastinWT 1.9 kb cDNA fragment obtained by EcoRI/AgeI restriction of pDsRed-Monomer-N1 L-plastinWT was inserted into the EcoRI/AgeI restricted pEGFP-N1 vector. The plasmid pEGFP-N1 L-plastinEF-ABD1 was generated by PCR ampli cation using the plasmid pEGFP-N1 L-plastinWT as a template and using primers that were designed to generate the restriction sites EcoRI and BamHI necessary for cloning the PCR-ampli ed cDNA into the pEGFP-N1 vector. The following primers were used: 5'-TATAGAATTCatggccagaggatc-3' as forward primer and 5'-GCGGATCCGCTTTGTGCAGGGC-3' as reverse complement primer. Lentiviral transduction was performed using third generation lentiviral vectors. The packaging vector psPAX2 and the envelope vector pMD2.G were obtained from Addgene (LGC Standards, Middlesex, United Kingdom). The transfer vector CD527A-1 carried the cDNAs corresponding to GFP, L-plastinWT-GFP, nonphosphorylatable L-plastinS5A-GFP or phosphomimetic L-plastinS5E-GFP. Brie y, the cDNA fragments were obtained by PCR ampli cation using the respective pEGFP-N1 plasmids as templates and using primers that were designed to generate the requested L-plastin mutation as well as the restriction sites necessary for cloning the PCR-ampli ed cDNAs into the CD527A-1 vector. For all cDNAs, XbaI and BamHI restriction sites were generated at the 5'-and 3'-ends, respectively. The following primers were used: 5'-TACTTCTAGAATGGCCAGAGGATCAGTGTC-3' as forward primer for L-plastinWT-GFP, 5'-TACTTCTAGAATGGCCAGAGGAGCAGT-3' as forward primer for L-plastinS5A-GFP, 5'-TACTTCTAGAATGGCCAGAGGAGAAGTGTC-3' as forward primer for L-plastinS5E-GFP, 5'-TACTTCTAGAATGGTGAGCAAGGGCGA-3' as forward primer for GFP and nally 5'-AGTAGGATCCCTTGTACAGCTCGTCCATGC-3' as reverse complement primer for all constructs. All constructs were veri ed by sequencing. The GIPZ short hairpin RNA (shRNA) non-silencing lentiviral vector as well as the target shRNAs for L-plastin (GIPZ Lentiviral shRNA Library, pool of clones V2LHS_133928, V2LHS_133929, V2LHS_238253, V2LHS_311716 and V2LHS_311717) were purchased from GE Dharmacon (Diegem, Belgium). The FLAG-tagged SGK3 plasmids were a kind gift of Professor Dan Liu (Baylor College of Medicine, Houston, TX, US) (characterized in (22)).

Treatment of cells with pharmacologic agents
Cells were cultured in the absence of serum for 16 h and then treatment was performed at 37 °C as follows: 0.1 mM PMA for 1 h, 1 ng/ml EGF for 15 min, 40 ng/ml HGF for 20 min, 100 ng/ml IGF for 20 min, 20 mM AKT inhibitor VIII for 1 h, 5 mM FAK inhibitor II for 1 h, 5 mM RSK inhibitor Bi-D1870 for 30 min, 5 nM Trametinib (Mekinist) for 1 h or 500 nM Apitolisib for 1 h. When activators and inhibitors were combined, the incubation with the inhibitors was performed rst and their presence was maintained during the incubation with the activators.

Immunoblot analysis
In situ cell lysis was performed with a cell scraper in ice-cold lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.1% SDS, 5 mM EDTA, 1% Nonidet P-40, 1% Triton X-100, 1% sodium-deoxycholate, 1 mM Na 3 VO 4 , 10 mM NaF, 100 mM leupeptin, and 100 mM E64D) containing a cocktail of protease inhibitors (Roche Diagnostics GmbH, Mannheim, Germany). Lysate clari cation was done by centrifugation at 13200 rpm for 15 min at 4 °C and total protein concentration was determined by Bradford assay (Bio-Rad, Hercules, CA, USA). Proteins (50 µg per lane) were resolved by SDS-PAGE in a 10% NuPAGE Tris-Base gel (Invitrogen) under reducing conditions, and transferred to a nitrocellulose membrane (GE Healthcare, Chicago, IL, USA) by semidry transfer. Membranes were saturated in Tris-buffered saline containing 1% bovine serum albumin and 0.1% Tween for 1 h at room temperature, then incubated with primary antibodies overnight at 4 °C and with secondary antibodies IRDye 680 RD donkey anti-mouse (#926-68072, RRID:AB_10953628, Thermo Fisher Scientific) and IRDye 800 CW goat anti-rabbit (#926-32211, RRID:AB_621843, Thermo Fisher Scientific) for 1 h at room temperature. Each antibody incubation was followed by at least three wash steps in Tris-buffered saline supplemented with 0.1% Tween. Signal intensities were quanti ed using the Odyssey Infrared Image System (LI-COR Biosciences, Lincoln, NE, USA). The ratio between the intensities obtained for phosphorylated protein versus total protein was calculated to make individual samples comparable and then normalized to the mean of all the ratios calculated for one blot to make blots comparable by accounting for technical day-to-day variability. For representative purposes, data were scaled to the controls present on each blot and are represented as means +/-SEM of three independent experiments. Raw images of the immunoblots are shown in the Supplementary Figure S1.

Modelling
The candidate signalling network upstream of L-plastin was derived from the literature. The experimental data were obtained by immunoblot analysis as described above and the ratios of P-LPL/LPL, P-ERK/ERK, P-AKT/AKT and P-Src/Src were used for model contextualization as follows. Within the FALCON toolbox, Dynamic Bayesian Networks are used to quantitatively simulate the logic of signalling pathways (23). Brie y, networks are initialized in a random state and the activity of 'input nodes' is xed according to the experimental conditions (presence or absence of growth factors and inhibitors). The signals are then propagated according to the laws of probability until convergence, when the activities of the 'output nodes' are compared with the measurements. A gradient descent algorithm is used to optimize the weights of the edges controlling the relative contributions of upstream nodes to downstream nodes in order to minimize the mean squared error (MSE) between the simulations and the measurements.
Regularized optimization was then used to put in evidence the speci c differences in signalling between the cell lines. Two types of regularization were applied to the parameter space during joint optimization of the individual cell line-speci c models. Firstly, we sought to decrease the in uence of experimental noise on the results by including a group partial-norm term penalizing the concurrent activation of a node by more than one activator. The effect of such regularization is to prune the network of edges that are not well supported by experimental evidence. Secondly, uniformity regularization (24) was applied across the four cell lines for each parameter. This density-based regularization term stems from the biological assumption that differences between the cell lines are more likely due to a small number of differences than to large-scale rewiring, and its effect is to remove small differences between cell line-speci c models unless they are well supported by the data. The combined effect of these two regularization terms is to reduce the size of the model and point to the signalling pathways that are differentially activated among the cell lines.
Regularized optimization with the FALCON toolbox was performed on the full dataset, after which the optimal model size was determined using the Bayesian Information Criterion (25) and the topology of the nal multi-cell line model was xed by removing edges with low (<0.01) ux and merging similar (<0.01 standard deviation) edges. This nal model was re-optimized on the full dataset, using unregularized optimization, to retrieve unbiased estimations for the activity of the different signalling proteins and the strength of the interactions between them. To estimate the error on the parameters, we optimized 20 models with synthetic datasets by applying random Gaussian noise on the measurements proportionally to the measurement error.
Files containing the data used for the modelling can be found in the Supplementary Figures S3 -S14.
In vitro kinase assays of full-length recombinant L-plastin The in vitro kinase assay was carried out as described before (Lommel, 2016). Brie y, full-length recombinant L-plastin (10 mg) was incubated with 50 mM ATP and 100 ng recombinant kinase SGK1, SGK2, SGK3 or RSK1 purchased at SignalChem (Richmond, BC, Canada) in a reaction volume of 25 ml, according to the manufacturer's protocol. For the negative control, the respective kinase was omitted. The reaction with RSK1 was performed as a positive control. Following an incubation of 15 min at 30 °C, Laemmli buffer was added, and the samples were boiled at 100 °C for 5 min and then subjected to immunoblot analysis.

Immunoprecipitation
For immunoprecipitation, 6 x 10 6 HEK 293T cells were transiently transfected with expression vectors encoding GFP, L-plastinWT-GFP, L-plastinS5A-GFP, L-plastinS5E-GFP or L-plastinEF-ABD1-GFP. 24 h after transfection, cells were homogenized in 500 µl lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 1% Triton, 1% glycerin, 1 mM PMSF, 1 mM sodium orthovanadate) containing a cocktail of protease inhibitors (Roche Diagnostics) and incubated on ice for 30 min. After a centrifugation step at 13200 rpm and 4 °C for 10 min, total protein concentration was determined by Bradford assay (Bio-Rad) and sample concentrations were adjusted with dilution buffer (10 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 1 mM PMSF, cocktail of protease inhibitors). 50 µl were added to SDS-containing sample buffer and used for SDS-PAGE (referred to as input). 25 µl of GFP-nanotrap beads (#gta-20, RRID:AB_2631357, Chromotek, Planegg, Germany) were added and incubated for 2 h on an end-over-end rotor at 4 °C. After a centrifugation step of 5 min at 3000 rpm at 4 °C, the supernatant was removed, and 50 µl of the supernatant were used for SDS-PAGE (referred to as non-bound). The bead pellet was washed four times with 300 µl dilution buffer. After the last washing step, the beads were resuspended in 2x SDScontaining sample buffer and boiled for 10 min at 95 °C (referred to as bound). The obtained samples were submitted to immunoblot analysis.

Transwell migration and invasion assays
For the transwell assays, cells were washed in phosphate-buffered saline (PBS) and resuspended in serum-free medium. A cell suspension containing 50000 cells was added to the upper well of transwell migration inserts (pore size: 8 μm, BD Biosciences, San Jose, CA, USA) or 100000 cells to BD BioCoat TM Matrigel TM invasion chambers (pore size: 8 μm, BD Biosciences). In the lower well, complete medium (700 μl) was used as chemoattractant. Cells were incubated for 24 h at 37 °C and 5% CO2, xed in 4% PFA for 10 min and stained with DAPI for 10 min. Cells that did not migrate to the lower compartment were removed with a cotton swab. Inserts were mounted on glass slides and ve random elds at a magni cation of 20x were counted per sample.

Invadopodia quanti cation
To quantify invadopodia formation, MDA-MB-231 cells were plated at low density on top of 0.1% gelatincoated coverslips and cultured for 24 h. All samples from the same replicate were stained simultaneously as described above. Four random elds at a magni cation of 40x were counted per sample using single confocal slices of the ventral surface of the cells. Image analysis was performed using ImageJ software (RRID:SCR_003070, National Institutes of Health, Bethesda, MD, USA). Firstly, the threshold "moments" was applied to the images of cells stained for F-actin and cortactin. To identify invadopodia, the tool "image calculator" was used to show dot-like structures that were present in both images. The GFPpositive invadopodia were determined in the same way using the result image obtained from the calculation of F-actin and cortactin, which was then compared with the GFP signal. Particle frequency was determined using the "analyze particle" command. A cut-off of 0.5-20 µm 2 was set as the size range and a value of 0.2 as the minimal circularity shape.

Gelatin degradation assay
The gelatin degradation assay was adapted from a previously described protocol (26). Firstly, 0.2% gelatin solution (#9000-70-8, Merck) was labeled using the Alexa Fluor 568-gelatin labeling kit (#A10238, Thermo Fisher Scienti c) and aliquots were stored at -20 ºC. To coat glass coverslips, the uorescent gelatin stock was mixed in a proportion 4:1 with non-labeled 0.2% gelatin solution and kept at 50 ºC. A volume of 100 µl of this mixture was given on top of each coverslip and incubated for 5 min. The coverslips were lifted and submerged in PBS in separate wells in a 12-well cell culture plate. When all coverslips were coated, PBS was aspirated and coverslips were incubated for 15 min on ice with prechilled 0.5% glutaraldehyde solution. After washing, the coverslips were incubated for 3 min at room temperature with freshly prepared sodium borohydride solution (5 mg/ml). Finally, the coverslips were extensively washed and stored at 4 ºC in PBS for up to two weeks, protected from light.
To quantify the gelatin degradation ability, 80000 MDA-MB-231 cells were plated on top of Alexa Fluor 568-labeled gelatin-coated coverslips in 12-well cell culture plates and allowed to attach for 6 h. Coverslips were then submitted to immuno uorescence and six random elds at a magni cation of 60x were examined per sample using single confocal slices of the ventral surface of the cells. The cell area was determined by the F-actin staining, using the "ROI manager" tool of ImageJ software. To determine degraded area, a threshold was applied to make visible the dark areas of degraded uorescent gelatin and quanti cation was performed using the "analyze particle" command. Relative degradation area was determined as total degradation area divided by total cell area, normalized to the value obtained for MDA-MB-231 GFP control cells.

Zymography
To analyze the activity of matrix metalloproteinases (MMPs), cells were cultured in complete medium until 70-80% con uence. Cells were then washed with PBS and cultured in serum-free medium for 24 h. The conditioned medium was collected, cleared by centrifugation and stored at -80 ºC. Zymography acrylamide gels (10%) were prepared according to standard procedures with gelatin added to a nal gelatin concentration of 1 mg/ml. The cell-free supernatant was mixed with 5x non-reducing sample buffer, incubated at room temperature for 10 min, and a volume of 25 µl of the mixture was loaded on the gels. After electrophoresis, the gels were incubated in washing buffer (50 mM Tris-HCl pH 7.5, 5 mM CaCl 2 , 1 µM ZnCl 2 , 2.5% Triton X-100) for 30 min. Finally, the gels were kept at 37 ºC with gentle agitation in incubation buffer (50 mM Tris-HCl pH 7.5, 5 mM CaCl 2 , 1 µM ZnCl 2 , 1% Triton X-100) for at least 24 h. Gelatinase activity was visualized by staining the gels with Coomassie Brilliant Blue G250 (Merck) with subsequent destaining in acetic acid-methanol-H 2 O (1:3:6). To visualize the amount of protein loaded, a parallel 10% polyacrylamide gel was loaded with the same volume of each sample and stained with Rotiblue (Carl Roth, Karlsruhe, Germany) for 1 h. Areas of protease activity and Roti-blue stained gels were analyzed using the Odyssey Infrared Image System (LI-COR Biosciences).

Statistics
All statistical analyses were carried out using Prism 5 (GraphPad Software, RRID:SCR_002798, San Diego, CA, USA). Results are expressed as means +/-SEM of three independent experiments. Statistical signi cance was assessed by performing unpaired Student's t-test or ANOVA for multiple comparison tests.

Results
Analysis of growth factor-stimulated signalling in breast cancer cell lines A candidate network of the regulatory signalling pathways upstream of L-plastin was assembled by manually curating signalling pathways from literature ( Figure 1A). Based on this network and in order to assess the interplay between ERK/MAPK and PI3K/AKT signalling pathways in regulating L-plastin Ser5 phosphorylation, we submitted four breast cancer cell lines to growth factor stimulation, with or without prior inhibition of key components of the two signalling pathways. Using MCF7, SKBR3, HCC38 and BT-20 cells, we rst analyzed the expression level of different growth factor receptors. As veri ed by immunoblotting, the four cell lines express insulin-like growth factor 1 receptor (IGF-IR), although the level of expression is very low for SKBR3 cells. With the exception of MCF7 cells, the receptor for epidermal growth factor EGFR and the receptor for hepatocyte growth factor HGFR (or c-met) could be detected in all the investigated cell lines ( Figure 1B).
The RSK inhibitor II Bi-D1870 (BID) and the AKT inhibitor VIII were chosen to selectively block the ERK/MAPK and PI3K signalling pathways, respectively. As a more central player connecting both pathways, FAK was inhibited using FAK inhibitor II. Following treatment, cells were lysed and the ratio of the phosphorylation level versus the total protein level for different output nodes was determined as an indicator of their activation status. The investigated output nodes were ERK and AKT, which are commonly used as readouts for ERK/MAPK and PI3K/AKT signalling pathway activity, respectively.
Additionally, we assessed the activation of the central player Src as well as L-plastin activation as the nal output node. Growth factor stimulation was carried out based on the expression of the corresponding receptor by the respective cell line, as illustrated in Figure 1B. In summary, we assessed the activation of four different output nodes in 20 different experimental conditions for the cell lines SKBR3, BT-20 and HCC38 and in 12 different experimental conditions for MCF7 cells. An example of the immunoblotting and the respective quanti cation of the ratios between Ser5-P-L-plastin and L-plastin for HCC38 cells is shown in Figure 1C. The quanti cation of all output nodes activation in the four cell lines is shown in the Supplementary Figure S2 and these results are summarized as a heatmap indicating the activity level of the nodes in each condition ( Figure 1D).

Modelling of the signalling network upstream of L-plastin
The averaged, normalized protein measurements were mapped to the corresponding network nodes, and the FALCON toolbox was then used to contextualize this network and retrieve, for each cell line speci cally, the activity of the remaining nodes, the speci c wiring of the signalling network and the ow of information for each experimental condition. Optimized regularization was performed to nd the model that ts best the experimental data taking into account the cell-line speci c parameters ( Figure  2A). The model with the lowest Bayesian Information Criterion (BIC) is considered the most adequate to represent the data ( Figure 2B), which corresponds to a model in which 63 of the 69 network parameters can be parametrized identically for all cell lines. Notably, interactions relating to the PI3K/AKT/mTOR axis showed relatively high heterogeneity compared to the crosstalks between them. The goodness-of-fit was similar for all cell lines ( Figure 2C), with MSE values ranging from 0.008 to 0.017 for individual cell lines, 0.032 for the single model and 0.018 for the nal modelIt should be noted that, in our nal model, RSK, SGK, PKA and PKC appear as the kinases able to phosphorylate L-plastin on its residue Ser5, with RSK and SGK being the most prominent kinases (Table 1). Importantly, SGK as a downstream kinase of the PI3K pathway was pointed out as a novel kinase involved in this process.  The optimized parameter values for the different models are expressed as the strength of the interaction from the parent node to the child node, relative to the total (= 1). Indicated is the value of the parameter for the best of all ts. The indicated error is the standard deviation (SD) of 20 rounds of re-sampling. PI3K is involved in the process of L-plastin Ser5 phosphorylation In an effort to experimentally validate the involvement of the PI3K pathway in the process of L-plastin Ser5 phosphorylation, we veri ed the phosphorylation level of L-plastin upon pharmacological inhibition of the ERK/MAPK pathway and the PI3K pathway, either individually or combined, by treating the cells with the MEK inhibitor Trametinib and/or the dual PI3K/mTOR inhibitor Apitolisib. In BT-20 cells, the combined inhibition of both pathways consistently led to a synergistic reduction of HGF-stimulated Ser5-P-L-plastin levels ( Figure 3A). In SKBR3 cells, Apitolisib treatment reduced L-plastin Ser5 phosphorylation by 50%, whereas Trametinib treatment alone was su cient to decrease this phosphorylation to background levels. In contrast, in HCC38 cells, Apitolisib did not display any effect, whereas Trametinib again reduced HGF-dependent L-plastin Ser5 phosphorylation to background levels. Hence, the relative contribution of the PI3K pathway to L-plastin Ser5 phosphorylation appears to depend on the cellular context.
Next, to further shed light on the possible involvement of SGKs in L-plastin activation, we performed an in vitro kinase assay assessing the ability of the three SGK isoforms SGK1, SGK2 and SGK3 to phosphorylate recombinant full-length L-plastin on residue Ser5. As shown in Figure 3B, SGK3 was able to phosphorylate L-plastin on its residue Ser5, although to a lower extent than RSK1. SGK2 was able to induce a weak L-plastin Ser5 phosphorylation, whereas SGK1 did not exhibit such phosphorylation ability.
We then examined the ability of SGK3 to phosphorylate L-plastin on its residue Ser5 in cells. To this end, we performed a comparative analysis between different SGK3 constructs by co-transfecting FLAG-tagged SGK3 WT or myristoylated SGK3 (Myr SGK3) and L-plastinWT-GFP in HEK 293T cells, which are devoid of endogenous L-plastin. As illustrated in Figure 3C, our immunoblot analysis revealed that the myristoylated form of SGK3 was phosphorylated on residue Thr320 indicating activation of the protein.
Strikingly, we consistently observed a strong L-plastin Ser5 phosphorylation only when co-expressed with the activated myristoylated SGK3 form, and not when co-expressed with the non-activated SGK3 WT, even though SGK3 WT was expressed at a higher level than myristoylated SGK3 in the cells. Altogether, our results provide evidence that L-plastin residue Ser5 can be phosphorylated in cells by SGK3, following activation of the kinase.

L-plastin Ser5 phosphorylation modulates breast cancer cell migration and invasion
Given the important role of L-plastin in cell motility of many different cell types, we next sought to examine the functional impact of L-plastin expression and Ser5 phosphorylation level in breast cancer cells with a speci c focus on cell migration and invasion. To this end, we have selected four cell lines expressing contrasting endogenous levels of this protein.
Initially, we set out to investigate the effect of L-plastin loss-of-function in cells naturally expressing high levels of L-plastin. For that purpose, we have silenced L-plastin expression in BT-20 and HCC38 cell lines.
Characterization of the transduced cells by immunoblotting revealed that the downregulation was highly e cient in both cell lines (Figures 4A, 4B). Following PMA stimulation, L-plastin Ser5 phosphorylation was strongly enhanced as monitored by the stronger band corresponding to Ser5 phosphorylated Lplastin (green). Even in the cells transduced with shRNA targeting L-plastin, PMA treatment could still induce Ser5 phosphorylation of the remaining L-plastin. Next, to determine whether L-plastin silencing has an impact on cell migration and invasion capacity, we performed transwell assays with the transduced cells. We observed that the Matrigel invasion ability was signi cantly reduced in both cell lines whereas the transwell migration was only signi cantly reduced in HCC38 cells (Figures 4C, 4D).
In parallel, we have neoexpressed GFP-fused L-plastin wild type (WT) or the phosphorylation variants Lplastin Ser5Ala (S5A) or L-plastin Ser5Glu (S5E) in BT-549 and MDA-MB-231 cell lines, which display no or a low level of endogenous L-plastin, respectively. Immunoblot analysis showed a band corresponding to L-plastin-GFP (red) for all the transduced cell clones, except for the one transduced with GFP alone ( Figures 4E, 4F). Consistently, L-plastin Ser5 phosphorylation was enhanced following PMA treatment. Of note, the anti-Ser5-P antibody also recognized weakly the phosphorylation-mimetic L-plastinS5E mutant independent of PMA treatment, whereas the phosphorylation-defective L-plastinS5A mutant was not recognized, as expected. Transwell assays showed that neoexpression of the phosphomimetic variant L-plastinS5E signi cantly enhances migration and invasion ability in both cell lines when compared to the GFP transduced cells (Figures 4G, 4H). Importantly, this increase in invasiveness was not observed if the cells expressed the non-phosphorylatable L-plastinS5A variant. A slight increase could also be detected for L-plastinWT-GFP expressing cells, although these differences were statistically signi cant only for BT-549 cell migration.
Overall, these results indicate that, in addition to L-plastin expression, L-plastin Ser5 phosphorylation is required to promote breast cancer cell migration and invasion.

L-plastin Ser5 phosphorylation promotes L-plastin recruitment to invadopodia
To further assess the role of L-plastin Ser5 phosphorylation in regulating cell migration and invasion, we performed confocal microscopy to characterize the subcellular localization of L-plastin phosphorylation variants in MDA-MB-231 cells.
Consistently, we observed that neoexpressed L-plastin was localized in actin-rich migratory structures and colocalized with actin and cortactin in invadopodia, with cortactin being a widely used marker of invadopodia. Invadopodia were identi ed as cortactin-and F-actin-containing punctae ( Figure 5A). To investigate if L-plastin regulates invadopodia formation, the number of invadopodia per cell was determined. We observed a slight increase in invadopodia density in cells expressing the phosphomimetic L-plastinS5E variant when compared to the other conditions; however, this difference was not significant ( Figure 5B). In addition, we assessed the ability of L-plastin variants to be recruited to invadopodia.
Remarkably, the quanti cation of GFP positive invadopodia showed that the non-phosphorylatable L-plastinS5A invadopodia localization was around two-fold lower as compared to L-plastinWT or the phosphomimetic L-plastinS5E, suggesting that L-plastin Ser5 phosphorylation is critical for L-plastin recruitment to invadopodia ( Figure 5C). To strengthen this nding, we investigated the intracellular localization of Ser5 phosphorylated L-plastin using the anti-Ser5-P antibody. This approach revealed that the L-plastin recruited to invadopodia is essentially the phosphorylated form ( Figure 5D). Altogether, our results indicate that L-plastin expression does not affect invadopodia formation, but Ser5 phosphorylation facilitates L-plastin recruitment to these structures.

L-plastin Ser5 phosphorylation does not enhance Lplastin/cortactin interaction
We have previously shown that the invadopodia marker cortactin e ciently co-precipitated with L-plastinWT extracted from PMA-treated MCF7 cells (12). Given the regulatory role of cortactin in invadopodia formation, function and assembly and knowing that cortactin acts as a scaffold protein (27), we wanted to further explore this interaction and the possible binding preference of cortactin to the Ser5 phosphorylated form of L-plastin, To this end, we performed co-immunoprecipitation experiments using GFP-nanotrap on whole-cell lysates from transfected HEK 293T cells. First, we tested if the presence of the C-terminal ABD2 (actin-binding domain 2) of L-plastin is necessary for its interaction with cortactin. To this end, we used the GFP-tagged L-plastinEF-ABD1 recombinant protein, which lacks ABD2.
We certi ed that PMA stimulation leads to Ser5 phosphorylation of L-plastinEF-ABD1 at a similar level as compared to L-plastinWT ( Figure 6A). Cell lysates of cells transfected with GFP, L-plastinWT-GFP or L-plastinEF-ABD1-GFP treated with PMA were then submitted to GFP-nanotrap with subsequent immunoblotting ( Figure 6B). We con rmed that endogenous cortactin interacts with L-plastinWT, but the expression of L-plastinEF-ABD1 recombinant protein is not su cient to preserve this interaction.
To investigate if Ser5 phosphorylation is essential for L-plastin/cortactin interaction, we then transfected HEK 293T cells with the GFP-tagged L-plastin phosphorylation variants ( Figure 6C). PMA stimulation of HEK 293T cells expressing L-plastinWT or expression of the phospho-mimetic L-plastinS5E did not enhance L-plastin/cortactin interaction. Likewise, the non-phosphorylatable L-plastinS5A mutant was able to co-precipitate cortactin with similar e ciency than L-plastinWT or L-plastinS5E. These observations show that Ser5 phosphorylation is not required for binding and does not enhance Lplastin/cortactin interaction.
L-plastin Ser5 phosphorylation enhances ECM degradation ability ECM degradation activity is typically executed by mature invadopodia (28). To monitor the impact of Ser5 phosphorylation on ECM degradation capacity, transduced MDA-MB-231 cells were plated onto uorescent gelatin-coated coverslips ( Figure 7A). The total number of cells associated with gelatin degradation did not change signi cantly upon expression of L-plastin wild type or phosphorylation variants ( Figure 7B). However, an increase in gelatin degradation area was found for cells expressing L-plastinWT and the cells expressing the phosphomimetic L-plastinS5E variant were found to have the highest gelatin degradation ability as compared to GFP expressing control cells ( Figure 7C).
Greater invasive and migratory capacities are often accompanied by elevated levels of MMPs, such as MMP-2 and MMP-9 (29). To test a possible role for L-plastin in the secretion of MMPs, we performed gelatin zymography on conditioned media of transduced MDA-MB-231 cells. Only the active forms of MMP-2 (65 kDa) and MMP-9 (82 kDa) were detected in the conditioned media. The results in illustrated in Figure 7D indicate that MMP-9, based on its gelatinase activity and apparent molecular weight, is the most prominent secreted protease in MDA-MB-231 breast cancer cells. Although all the L-plastin variants induced an increase in the activity of MMP-9, the quanti cation of the observed differences showed signi cance (~2 fold increase) only for the S5E variant when compared with GFP control samples. Collectively, our results suggest that L-plastin Ser5 phosphorylation enhances MMP-9 activity and concomitant ECM degradation.

Discussion
Post-translational modi cations (PTMs) are critical for protein function and play pivotal roles in cellular homeostasis. Aberrant PTM may contribute to pathogenesis and has been associated with numerous diseases, including cancer (30). This implies that the exclusive analysis of genetic variations and protein expression levels may often be insu cient or even misleading (31). In particular, protein phosphorylation displays the largest number of disease associations among PTMs and is especially relevant for breast cancer (30). Since protein phosphorylation is frequently altered as a consequence of cancer driver gene mutations and concomitant aberrant signal transduction, phosphoproteome analysis is indispensable for the understanding of disease mechanisms and may have diagnostic and therapeutic relevance (32).
Here, we aimed at exploring the interplay between ERK/MAPK and PI3K/AKT pathways in regulating Lplastin Ser5 phosphorylation. To this end, we performed immunoblot analysis to assess the activation status of four different output nodes in four breast cancer cell lines in 20 different experimental conditions. In addition to phosphorylated L-plastin (pSer5) as an output node, we analyzed phosphorylated ERK (pTyr204) and phosphorylated AKT (pSer473) as read-outs for activated ERK/MAPK and PI3K/AKT pathways, respectively. As a fourth output node, we monitored phosphorylated c-Src (pTyr416), since activated c-Src is known to be important for tumorigenesis and cancer progression and has been previously implicated in the regulation of several signalling processes, including ERK/MAPK and PI3K/AKT pathways (33). We have stimulated the cells with the growth factors for which they express the corresponding receptor. As inhibitors, we selected a MEK and an AKT inhibitor, to block the ERK/MAPK and PI3K/AKT pathways, respectively. Additionally, a FAK inhibitor was used, as FAK is a key molecule in invasion and metastasis and activates both ERK/MAPK and PI3K/AKT pathways (34).
Performing computational modelling based on our experimental data, we con rmed that the ERK/MAPK pathway plays a major role in L-plastin Ser5 phosphorylation. In addition, we found that the PI3K pathway likely contributes to this process via downstream SGK kinases, rather than via AKT. Our in vitro kinase assay comparing the three isoforms SGK1, SGK2 and SGK3 showed that Ser5 phosphorylation of full-lengh L-plastin is primarily mediated by the SGK3 isoform. Importantly, L-plastin Ser5 phosphorylation was highly increased when co-expressed with activated myristoylated SGK3, corroborating the capacity of SGK3 to phosphorylate residue Ser5 of L-plastin in cells.
To further assess the contribution of the PI3K pathway in L-plastin Ser5 phosphorylation, we treated BT-20, HCC38 and SKBR3 cells with two drugs Apitolisib as a PI3K/mTOR inhibitor and Trametinib as a MEK inhibitor to inhibit the PI3K pathway and the ERK/MAPK pathway, respectively. The dual PI3K/mTOR inhibitor Apitolisib was chosen in order to block both PDK1, which is downstream of PI3K, and mTORC2, and thus, to inhibit the activation of SGK3, which requires phosphorylation by the two kinases (35). In HCC38 cells, PI3K/mTOR inhibition had no major effect on HGF-triggered L-plastin Ser5 phosphorylation as compared to MEK inhibition, which strongly impaired this phosphorylation event. In contrast, in SKBR3 cells, inhibition of the PI3K/mTOR pathway reduced L-plastin Ser5 phosphorylation by 50%, and in BT-20 cells, PI3K/mTOR inhibition and MEK inhibition acted in a synergistic way to reduce this phosphorylation to background levels, suggesting that PI3K and mTOR are important for HGF-dependent L-plastin Ser5 phosphorylation in these two cell lines. Taken together, our results corroborate that the ERK/MAPK pathway is predominant for triggering L-plastin Ser5 phosphorylation. Importantly, the PI3K pathway contributes to this phosphorylation event depending on the investigated cell line. For BT-20 cells, this involvement may be explained by the presence of an activating mutation in the PI3KCA gene, which confers a gain of function to the p110α catalytic subunit of class IA PI3K and promotes PI3K-dependent tumorigenesis (36). Oncogenic mutations in PIK3CA are found in approximately 25% of breast cancers (3). Importantly, PI3KCA-mutant cells that lack AKT activation display a functional dependency on SGK3, which shares a consensus phosphorylation motif with AKT (Vasudevan et al. 2009). The mechanism linking oncogenic PI3KCA to SGK3 activation and concomitant AKT suppression involves the phosphoinositide phosphatase INPP4B, and both SGK3 and INPP4B have been suggested to have oncogenic functions (37).
Only few studies have reported so far a link between the PI3K pathway and L-plastin regulation and most of these studies have focused on the PI3K/AKT axis. In chronic lymphocytic leukemia, inhibition studies provided evidence for a role of PI3K in B-cell receptor-induced L-plastin activation through promoting Ser5 phosphorylation (38). Moreover, in a prostate cancer study, the PI3K/AKT pathway was found to upregulate L-plastin expression levels through upregulation of the transcription activator AP4 (39). This study showed that L-plastin is a key player in AP4-mediated prostate cancer cell migration, invasion and proliferation. Importantly, L-plastin and AP4 protein levels were also found to be upregulated in prostate cancer tissues as compared to adjacent normal tissues and correlated with lymph node metastasis. On the other hand, L-plastin has also been shown to play an upstream role in the regulation of the PI3K/AKT pathway, by promoting SDF-1 -dependent AKT Thr308 phosphorylation in human T-lymphocytes (40) or AKT Ser473 phosphorylation via regulation of the mTORC2 complex activity in a very recent hypereosinophilia study (41). Most interestingly, immunohistochemical staining of bladder cancer tissues revealed a signi cant positive correlation between pAKT and L-plastin expression as well as a signi cant correlation between L-plastin expression and tumor histological grade, stage and growth pattern (42). In our study, our experimental data corroborate our computational modelling predictions and indicate that, depending on the cellular context, the PI3K/SGK3 axis represents an alternative oncogenic signalling pathway involved in L-plastin activation. Knowing that this PI3K-dependent, AKT-independent signalling axis signi cantly contributes to cancer progression (43), it deserves close attention and SGK inhibitors might be promising therapeutic agents.
In a next step, we examined the functional impact of the L-plastin Ser5 phosphorylation event in breast cancer cells. Whereas calcium binding and oxidation of L-plastin inhibit the actin-bundling activity of Lplastin (44)(45)(46)(47), Ser5 phosphorylation is known to promote its targeting to actin-rich structures and to increase its F-actin bundling activity in vitro and in cells (11,12). L-plastin phosphorylation on residues Ser5 and Ser7 has been linked to bone resorption activity via nascent sealing zone and sealing ring formation in cultured osteoclasts and in mice (48,49). In T-cells, L-plastin Ser5 phosphorylation is important for immunological synapse maturation and stability and, thus, for proper T-cell activation (50,51) and, in podocytes, this phosphorylation event promotes lopodia formation (52). Most importantly, previous results obtained in cancer cells point to the importance of this phosphorylation event for in vitro invasion and in vivo metastasis formation of melanoma cells (53,54). Consistent with these ndings, we demonstrate here that L-plastin Ser5 phosphorylation strongly promotes cell migration and invasion capacities in a breast cancer cell model.
The process of tumor cell invasion and metastasis is associated with the assembly of invadopodia, which are F-actin-rich protrusive structures capable of degrading the ECM. These structures are characterized by the presence of the core proteins cortactin and Tks5. Moreover, they concentrate proteolytic activity and constitute the point of convergence of a plethora of signalling pathways (55). Although, a dendritic actin network mediated by the Arp2/3 complex is known to be essential for invadopodia formation, there is increasing evidence that linear, bundled F-actin is also inherent to invadopodia (56,57). In addition to other actin-bundling proteins including fascin and -actinin, a recent report has added L-plastin to the list of actin-bundling proteins found in invadopodia (19). This study provided a model where L-plastin contributes to invadopodia extension, while protrusive force is guaranteed by fascin, which in turn, confers structural rigidity and stability (19). In line, our results show that L-plastin ectopic expression does not affect invadopodia density in MDA-MB-231 cells, suggesting that L-plastin is not involved in the initial stage of invadopodia formation. Importantly, we found that Ser5 phosphorylation promotes L-plastin recruitment to invadopodia and L-plastin localized in invadopodia is essentially the phosphorylated form.
Since cortactin is considered as a scaffold protein in invadopodia (27), we wanted to better characterize the interaction between cortactin and L-plastin that we have shown before (12). Our results indicate that the presence of the L-plastin ABD2 domain is necessary for this interaction, but that Ser5 phosphorylation is not required for this interaction. Given that we have shown that Ser5 phosphorylation enhances Lplastin localization in invadopodia, we can speculate that the invadopodial recruitment of L-plastin does not solely rely on its interaction with cortactin.
Degradation of the ECM by podosomes and invadopodia occurs by localized secretion of specialized proteases (58). However, the association between L-plastin and the regulation of matrix-degrading enzymes in these structures is largely unknown. Here we demonstrated that L-plastin Ser5 phosphorylation facilitates ECM degradation although it increases only weakly MMP-9 activity. An interesting approach to study the role of L-plastin in macrophage podosomes and cancer cell invadopodia was taken by the group of Jan Gettemans, who produced nanobodies (Nb) either inhibiting L-plastin bundling activity (Nb5) or locking L-plastin in an inactivated state (Nb9) (59). In macrophages, the expression as well as secretion and localization of the most prominent proteases MMP-2, MMP-9 and MMP-14, were found to be unaffected by both nanobodies (60). Instead, the nanobodies induced podosome instability by affecting the cyclic turnover of actin in the podosomes and, thereby, decreased podosome lifespan. The authors concluded that defective ECM degradation observed in the nanobodyexpressing cells was most likely associated with structural malformation of the podosomes (60). Similarly, in PC-3 prostate cancer cells, the use of the L-plastin-speci c nanobodies led to reduced degradation capacity of the cells (19). Again, the bundling-inhibitor Nb5 had no effect on MMP-9 secretion and activity, con rming that Nb5-mediated reduction of degradation is not dependent on MMP-9 secretion and activity. More recently, Balta and collaborators highlighted the existence of a link between L-plastin expression, total MMP activity and MMP-2 release in MV3 melanoma cells (47). L-plastin and MMP-2 were found to localize in invadopodial extensions and to co-immunoprecipitate, indicating that Lplastin may help MMP-2 translocate to invadopodial structures. Similarly, cortactin was shown to regulate membrane tra cking to promote protease secretion for invadopodia-associated ECM degradation (61). A more recent study revealed a role for both cortactin and fascin in the release of extracellular vesicles containing MMPs (62). Notably, the involvement of fascin was not linked to its actin-bundling activity, but rather to its function in microtubule-regulation and endosomal tra cking.
According to our results, L-plastin expression in MDA-MB-231 cells slightly increases MMP-9 activity and has no in uence on MMP-2. Despite the weakness of its in uence on MMPs, we could demonstrate that L-plastin Ser5 phosphorylation strongly increases gelatin degradation capacity. Given that L-plastin Ser5 phosphorylation also increases its actin-bundling activity (11), our results suggest that enhanced degradation capacity of Ser5 phosphorylated L-plastin as compared to non-phosphorylated L-plastin is mainly due to enhanced bundling activity and only minimally due to enhanced MMP-9 activity.
Activation of invasion is a hallmark of metastatic cancers and understanding the cellular machinery that underlies invasion is critical for the establishment of novel predictive biomarkers and therapeutic targets. In this study, we have found that L-plastin Ser5 phosphorylation promotes breast cancer cell invasion, Lplastin recruitment to invasive structures and degradation of the ECM. Furthermore, we have established an involvement of the ERK/MAPK and PI3K pathways in promoting L-plastin Ser5 phosphorylation in breast cancer cell lines. In a next step, we wanted to con rm these results in tissues by analyzing the existence of potential correlations between L-plastin and ERK/MAPK or PI3K pathways in breast cancer tissue microarrays (TMAs). Surprisingly, we could not detect L-plastin in the carcinoma cells of the analyzed TMAs when performing immunohistochemistry (IHC). As L-plastin expression and Ser5 phosphorylation were detectable in in ltrated leukocytes, epitopes seem to have been preserved during sampling. This result is surprising, because L-plastin expression in breast cancer tissues has been shown before by IHC (63,64). Other studies have also detected an expression of L-plastin in breast cancer samples, but their methods did not take into account the heterogeneity of cancer tissues (65,66) (67). Whether L-plastin expression in the carcinoma cells of our breast cancer TMAs was below threshold for detection in IHC remains to be established.

Conclusions
Given the extensive crosstalk between the ERK/MAPK and PI3K pathways, blocking of one of the two pathways in anti-tumor therapy might be counteracted through activation of the other pathway.
Therefore, combined therapy approaches concurrently blocking both pathways are expected to have more e cient anti-tumor activities. However, complete blocking of both signalling pathways might result in signi cant toxicity for non-transformed cells (5). Taking all this into consideration, one can speculate that blocking a subset of downstream functions of both pathways might be an e cient approach. As illustrated in Figure 8, our study has provided evidence that L-plastin is a downstream molecule of both ERK/MAPK and PI3K/SGK signalling pathways and, hence, represents a potential target for such an approach. Since L-plastin Ser5 phosphorylation promotes the recruitment of L-plastin to invadopodia, ECM degradation and invasion/migration of breast cancer cells, blocking this phosphorylation event might be an interesting alternative to reduce breast cancer cell invasiveness.