Vimentin–ERK Signaling Uncouples Slug Gene Regulatory Function Free

Cancer cell dissemination and metastasis is the underlying cause of mortality in cancer patients and is closely linked to a developmental process called epithelial–mesenchymal transition (EMT). During EMT, epithelial cells lose apical–basal polarity, disassemble cell–cell contacts, and adopt a more mesenchymal and motile phenotype (1, 2). In the course of malignant progression, cancer cells acquire the ability to break through the basal lamina, invade surrounding tissues and capillaries, extravasate and colonize distant organs (3). The acquisition of EMT characteristics is believed to be one important mechanism, whereby cancer cells achieve increased motility and invasiveness to promote metastasis (4, 5). In breast cancer, EMT characteristics are enriched in the aggressive and metastatic triple-negative breast cancer subtype (6, 7), suggesting a role for EMT in breast cancer metastasis. Importantly, vimentin, the mesenchymal intermediate filament (IF) and a hallmark of EMT, is overexpressed in malignant epithelial cancers, including breast cancer, and correlates with poor prognosis (8). Vimentin is also present in mammary and breast cancer stem cells, providing a further link between EMT and malignancy (9, 10). However, the functional role of vimentin in EMT and/or stem cells remains incompletely understood. Strong evidence indicates that vimentin regulates mesenchymal cell shape and mammary epithelial cell migration (11–13), and plays a role in regulating signal transduction, necessary for EMT induction, downstream of mutant H-Ras and TGFβ via a yet unknown mechanism (13).

Activated ERK kinases participate in the regulation of several cellular processes such as cell proliferation, survival, and motility. Recent evidence has highlighted that ERK2, and not ERK1, is involved in EMT induction (14, 15) and cancer cell migration and invasion (16–18). Vimentin has been shown to function as a signaling scaffold in many different cell types and biologic processes (19). In neurons, vimentin fragments interact with ERK and support the spatial translocation of active ERK along axons in response to injury (20). However, it is currently unknown whether a similar mechanism functions outside of the nervous system or whether vimentin upregulation during cancer progression plays a role in regulating ERK signaling.

Slug belongs to the Snail family of EMT-inducing transcription factors (Snail, Slug, and Smuc; refs. 2, 21). Posttranslational modifications of Snail, and as recently described for Slug, have emerged as critical levels of control for their abundance and subcellular localization. In the nucleus, Lats2 interacts with and directly phosphorylates Snail on residue T203, supporting EMT by influencing the stability and nuclear localization of Snail (22). Glycogen synthase kinase 3β (GSK3β) phosphorylates both Slug and Snail and negatively regulates their stability thus maintaining epithelial morphology (23–27). The activation of oncogenic pathways, including PI3K–Akt and Ras–MAPK signaling, suppresses GSK3β activity and GSK3β-dependent reduction in Snail and Slug phosphorylation resulting in increased transcription factor stability and accumulation and EMT induction.

In addition to ERK2 activity-dependent inhibition of GSK3β activity, other direct mechanisms have also been implicated in EMT onset, including Fra1-mediated induction of ZEB1/2 transcription factors (15, 28) and stromal collagen-induced Snail phosphorylation, leading to increased stability and nuclear accumulation of Snail (14). Altogether, a deeper appreciation of the mechanisms regulating the Snail family of transcription factors, including posttranslational modifications, will be vital in our efforts to prevent EMT and EMT-related processes during cancer progression. Here, we identify a pathway linking vimentin expression, ERK activity, and Slug phosphorylation to EMT induction and positive regulation of cancer cell invasion, migration, and the gene regulatory function of Slug.

For antibodies and reagents, see Supplementary Experimental Procedures.

For cell culture and stable cell lines, see Supplementary Experimental Procedures.

The histologic breast cancer material of triple-negative (n = 118) and estrogen receptor (ER) and/or progesterone receptor (PR), and/or HER2-positive tumors (n = 356) consisted of formalin-fixed, paraffin-embedded or frozen samples retrieved from the files of the Department of Pathology, Kuopio University Hospital (Kuopio, Finland). Samples were sectioned, prepared for immunohistochemistry, and stained with appropriate primary and secondary antibodies. For more information on tissue preparation and immunohistochemistry, see Supplementary Experimental Procedures.

For tumor xenografts on chick embryos, see Supplementary Experimental Procedures.

In in vitro kinase assays purified recombinant active Akt1, ERK1, or ERK2 (ProQinase GmbH; BSA as the control) were incubated for 20 minutes at 30°C with recombinant Snail or Slug and ATP-γ-P32, in buffer containing 20 mmol/L HEPES pH 7.4, 10 mmol/L MgCl₂, and serine–threonine phosphatase inhibitor cocktail 1. Reactions were stopped with Laemmli sample buffer and heating at 100°C for 10 minutes. Samples were resolved by SDS-PAGE, and the gel was Coomassie-stained and dried before detection by autoradiography. For the alkaline phosphatase protection assays, recombinant active ERK2 (ProQinase) was incubated with either vimentin (Cytoskeleton Inc.) or actin (non-muscle human platelets; Cytoskeleton Inc.) in reaction buffer (125 mmol/L HEPES–NaOH pH 7.5, 150 mmol/L NaCl, 7.5 mmol/L MgCl₂, 10 μmol/L CaCl₂, and 2.5 mmol/L DTT) for 30 minutes at 37°C. Alkaline phosphatase (calf intestinal; Promega) was then added to the preformed protein complexes or to ERK2 alone and incubated for a further 2 or 20 minutes at 37°C. The reaction was stopped by the addition of 8× Laemmli sample buffer followed by SDS-PAGE and Western blotting with the indicated antibodies.

Direct protein binding was assessed by GST and antibody pulldowns using recombinant proteins. For GST pulldowns, recombinant proteins [phosphorylated or nonphosphorylated ERK2 (ProQinase) and GST-vimentin (Spring Bio) or GST as a control] were incubated for 1 hour at room temperature in reaction buffer (20 mmol/L Tris–HCl, 150 mmol/L NaCl, 1 mmol/L MgCl₂, and 10% glycerol) and complexes isolated using preblocked glutathione beads for 1 hour. For antibody pulldowns, recombinant proteins [ERK2 and either vimentin or actin or keratin-8 (Sigma)] were allowed to react for 30 minutes at 37°C and complexes isolated using anti-ERK antibody for 30 minutes followed by the addition of preblocked Protein G Sepharose beads (GE Healthcare) for 1 hour.

For cosedimentation assays, polymerization of vimentin [0.5 mg/mL in 50 μL of assembly buffer: 20 mmol/L Tris, pH 7.4, 3 mmol/L KCl, 0.2% 2-mercaptoethanol, 0.2% phenylmethylsulfonylfluoride (PMSF)] was triggered by the addition of 3 μL of 5 mol/L NaCl at 37°C for 1 hour. Vimentin IFs were then incubated with the indicated recombinant proteins (1–2 μg each) for an additional hour. Proteins were collected by ultracentrifugation at 100,000 × g for 30 minutes and the total soluble (supernatant) and nonsoluble (pellet) fractions were analyzed by SDS-PAGE followed by Western blotting with the indicated antibodies.

In immunoprecipitation assays, cells were treated with EGF (50 ng/mL) or with DMSO as control for 20 minutes, washed with PBS, and detached with HyQTase (Thermo Scientific). Cells were lysed with RIPA lysis buffer (50 mmol/L Tris–HCl, 150 mmol/L NaCl, 1 mmol/L EDTA, 1% NP-40, 0.5% sodium deoxycholate, 1 mmol/L EGTA, and 0.1% SDS) for 30 minutes at 4°C, centrifuged at 13,000 rpm for 10 minutes at 4°C. Lysates were incubated with anti-FLAG antibody (2.5 μg/reaction) for 1 hour, and washed three times. Where indicated immunoprecipitations were performed in cells preincubated with ³²P-ATP–supplemented media.

For luciferase reporter assays and DNA-binding ELISA, see Supplementary Experimental Procedures.

For qRT-PCR methods, see Supplementary Experimental Procedures.

For immunofluorescence, fluorescence recovery after photobleaching (FRAP), and flow cytometry, see Supplementary Experimental Procedures.

Leica TCS SP5 stimulated emission depletion (STED) laser scanning microscope was used to image in super-resolution level (Leica Microsystems GmbH), where approximately 65-nm resolution in x- and y-axis was achieved. Mega-520-labeled vimentin was excited with 532-nm pulse laser (PicoQuant), and respectively Star-635 (for slug or pERK) at 635-nm wavelength (PicoQuant). The channels were scanned sequentially and emission was detected by avalanche photodiode detectors at emission range of 685/40 (Leica Microsystems GmbH). Leica LAS software (Leica Microsystems GmbH) was used do perform background subtraction and then deconvolution of all images. In the deconvolution process, a Lorentzian PSF was generated by using measured PSF value of 62 nm, which was exploited to signal energy based deconvolution algorithm.

For proliferation, migration, and invasion assays, see Supplementary Experimental Procedures.

For proximity ligation assay (PLA), see Supplementary Experimental Procedures.

For mass spectrometry, see Supplementary Experimental Procedures.

Rabbit polyclonal antibody against phosphorylated Slug was generated following immunization with a synthetic peptide conjugated to OVA (NH2-CYSSSLGRV(Sp)-COOH for Slug(P)87 site. The rabbits were immunized according to the standard procedure of the service provider (Primm srl) with five injections and the antibody was affinity purified from immune serum using CNBr-Sepharose–conjugated phospho-peptides.

All statistical analyses were done using the Student t test. For the clinical samples, cytoplasmic and nuclear Slug and vimentin expression were analyzed on a per patient basis and the statistical significance and associations with the parameters (triple-negative or other samples) was analyzed with the χ² test.

Vimentin, Slug, and ERK kinases, especially ERK2, have been linked to EMT and breast cancer cell invasion. In a cohort of 356 breast cancer tissue samples, we found that vimentin and Slug expression were, as described earlier (29, 30), significantly higher in triple-negative tissues (Fig. 1A, 164 samples). In these samples, 61% of cells were vimentin-positive, 73% showed positive cytoplasmic and 42% positive nuclear Slug staining, 71% were ERK-positive and 49% displayed phosphorylated ERK-1/2. Moreover, among these triple-negative samples, Slug, ERK, and phosphorylated ERK (pERK) expression was more pronounced at the tumor periphery in line with a recent report (Fig. 1B; Slug in 44% and pERK in 39% of the tumors; ref. 31), whereas vimentin was expressed throughout the tumor, including at the periphery. In these triple-negative breast carcinomas, pERK did not correlate with slug (P = 0.12) or vimentin (P = 0.23) expression as determined by the χ² test. As these EMT-linked proteins were coexpressed at the tumor invasive front, we investigated their functional impact on migration and invasion of triple-negative breast cancer cells using the MDA-MB-231 cell line. We reported recently that vimentin is a critical regulator of migration in these cells (13). Here, we further demonstrate that silencing of Slug, ERK1, or ERK2, using either smart pool or individual siRNA oligos, also significantly inhibits MDA-MB-231 cell migration (Fig. 1C and Supplementary Fig. S1A). ERK2 appeared to be the predominant ERK isoform regulating cell migration as recently described (16, 17). Moreover, these effects were motility-specific as suppressed cell motility ((13) and Supplementary Fig. S1A) was not accompanied by defects in cell proliferation (Supplementary Fig. S1B). In addition to roles in cell migration, ERK2, vimentin, and Slug were all required for efficient directional cell invasion through a laminin-rich extracellular matrix (Matrigel; Fig. 1D and Supplementary Fig. S1A). As ERK1 silencing did not appear to have any impact on cell invasion (Fig. 1D), we focused our future efforts on delineating the role of the ERK2 isoform in EMT induction.

Remarkably, in an orthotopic xenograft model, the population of MDA-MB-231 cells that spontaneously metastasized to the contralateral lymph nodes exhibited significantly higher vimentin levels than the cell population remaining in the primary tumor (Fig. 1E), whereas DAPI staining intensity was even between the primary tumor and lymph node metastases (Supplementary Fig. S1C). These data suggest that increased vimentin expression correlates with enhanced tumor metastasis in vivo even in cells that already displays all the hallmarks of EMT. Together, these results demonstrate that vimentin, Slug, and ERK2 are coexpressed in the majority of triple-negative breast cancers and are important regulators of breast cancer cell migration, invasion, and metastasis.

We previously identified a functional role for vimentin in supporting EMT downstream of TGFβ and active Ras signaling (13); however, the molecular mechanisms underlying this observation were unclear. Here, we show that H-Ras–transformed MCF10A cells with elevated vimentin expression exhibit increased ERK phosphorylation relative to the nontransformed low vimentin-expressing cells (Fig. 2A). Moreover, TGFβ-induced vimentin expression in MCF10A breast epithelial cells results in a concomitant increase in ERK phosphorylation (Fig. 2B), indicating a correlation between vimentin expression and ERK activity. Interestingly, silencing of vimentin in TGFβ-induced MCF10A and in vimentin-high MDA-MB-231 cells significantly reduced the phosphorylation and total levels of ERK kinases (Fig. 2B and Supplementary Fig. S2A), suggesting that vimentin expression may be necessary to stabilize activated ERK in cells. To validate these findings, we generated immortalized wild-type (WT) or vimentin-null (Vim KO) mouse embryonic fibroblasts (MEF) and investigated total and active ERK levels in these cells. Congruent with the siRNA data, the levels of pERK and, to a lesser extent, total ERK were significantly decreased in Vim KO MEFs compared with WT MEFs (Fig. 2C). Importantly, ERK levels were rescued by reexpression of vimentin in the null cells (Fig. 2D).

In order to elucidate the mechanism of vimentin-dependent ERK regulation, we first investigated ERK and vimentin localization in cells. Using super-resolution STED microscopy, we detected a subset of pERK coincident with vimentin filaments. Importantly, treatment of cells with ERK inhibitor U0216 noticeably reduced the pERK signal along vimentin filaments (Fig. 2E). Moreover, in situ PLAs demonstrated colocalization between endogenous pERK and vimentin in the cytoplasm of MDA-MB-231 cells (Fig. 2F). Notably, the PLA signal was lost upon inhibition of MEK1/2 with U0126, indicating a specific interaction between ERK and vimentin (Fig. 2F). These data suggest that vimentin and active ERK could form a complex in cells. To clarify whether this vimentin–ERK association was direct and phosphorylation-dependent, we performed in vitro interaction studies with purified recombinant proteins. pERK2 interacted with GST-vimentin in GST pulldowns much more efficiently than nonphosphorylated ERK2 (Fig. 2G). In the reciprocal pulldown, recombinant vimentin was readily detected following immunoprecipitations of pERK2 (Fig. 2H). Importantly, the vimentin–ERK2 interaction appeared to be specific as recombinant keratin 8 and actin failed to associated with ERK2 in immunoprecipitation assays (Supplementary Fig. S2B). Altogether, these data demonstrate a specific and direct interaction between ERK2 and vimentin that appears to be dependent on ERK2 activation status.

ERK phosphorylation is regulated by upstream kinases and by competing phosphatases that rapidly dephosphorylate and inactivate ERK (32). As we revealed a correlation between vimentin expression and active ERK2 levels, we next sought to determine whether vimentin could directly affect ERK2 activity. Intriguingly, we discovered that vimentin supports ERK2 activity, in vitro, in the presence of inactivating alkaline phosphatase, measured by detecting ERK2 autophosphorylation levels with a phospho-specific antibody (Fig. 3A), or by autoradiography in the presence of [y-³²P] ATP (Supplementary Fig. S2C). Notably, maintained ERK2 activity in the presence of alkaline phosphatase was vimentin-specific as actin failed to protect ERK2 from alkaline phosphatase dephosphorylation in vitro (Supplementary Fig. S2D) and induction of actin stress-fibers with active RhoA mutants in MDA-MB-231 cells had no noticeable impact on ERK phosphorylation (Supplementary Fig. S2E). Therefore, the elevated active ERK levels observed in vimentin-expressing cells could be attributed to the ability of vimentin to antagonize phosphatases that inactivate ERK within cells. Elevated ERK1/2 activity has been implicated in many human cancers (33, 34). In line with this notion, we found that vimentin was necessary to support proliferation of tumors in vivo. Moreover, silencing of vimentin reduced ERK activity in MDA-MB-231 cells in vitro (Fig. 3B) and significantly impaired tumor growth of the same cells when implanted as xenografts on chick embryo chorioallantoic membranes (Fig. 3C). Thus, the positive reinforcement of ERK activity by vimentin could represent a crucial mechanism for promoting EMT onset as well as supporting cancer cell proliferation.

Next, we set out to determine whether ERK could likewise support vimentin protein levels and thus further reinforce EMT processes. Intriguingly, ERK2 silencing triggered a significant reduction (40%) in vimentin mRNA levels as measured by qRT-PCR (Fig. 3D). This reduction in vimentin mRNA was further reflected by the diminished fluorescence intensity of the vimentin network in ERK2-silenced cells (Fig. 3E). To validate these findings, we expressed constitutively active ERK2 (MEK–ERK2 fusion; CA-ERK2) in MCF10A epithelial cells with low endogenous vimentin expression. Strong vimentin staining was detected exclusively in CA-ERK2 expressing cells compared with the nontransfected cells (Fig. 3F) indicative of an ERK2-dependent role in inducing vimentin expression. These data together with the observed direct interaction between vimentin and ERK and the vimentin-mediated stabilization of ERK activity strongly support the existence of a reciprocal vimentin–ERK regulatory complex.

In addition to ERK and vimentin, we found the EMT transcription factor Slug similarly coexpressed at the periphery of breast carcinoma biopsies (Fig. 1B). A more in-depth examination of Slug localization using STED microscopy and fractionations revealed a proportion of Slug (13% ± 6%; Fig. 4A) in the cytosolic fraction and in regions corresponding to vimentin filaments (Fig. 4B). As a transcription factor, Slug strongly localizes to the nucleus and thus colocalization with vimentin might seem unexpected. However, analysis of Slug dynamics in live-cells using FRAP revealed that the predominantly nuclear Slug actively shuttles between the nucleus and the cytoplasm. Following photobleaching of the whole nucleus to abolish nuclear Slug fluorescence, there was a rapid recovery of nuclear Slug via the import of unbleached cytoplasmic-derived Slug molecules (Supplementary Fig. S3A–S3C). In situ PLAs confirmed endogenous Slug–vimentin interaction in the cytoplasm of cells (Fig. 4C). Moreover, this interaction appeared to be direct as purified recombinant Slug associated with GST-vimentin in in vitro pulldown assays (Fig. 4D). In addition, in cosedimentation experiments, Slug was detected only in the vimentin-positive insoluble pellet fraction, suggesting an association between Slug and vimentin (Fig. 4E). Interestingly and in agreement with a vimentin–ERK interaction, ERK2 was also present in the same vimentin–Slug pellet fractions. Moreover, vimentin expression positively correlated with Slug protein levels as demonstrated by a significant reduction in Slug levels in Vim KO MEFs (Fig. 4F).

Collectively, these results demonstrate that in addition to ERK2, the EMT transcription factor Slug also interacts with vimentin filaments and highlight a direct link between three important EMT regulators.

As we were able to detect vimentin interaction with both active ERK- and Slug, we next investigated whether ERK and Slug could, in turn, associate with each other. In MDA-MB-231 cells transfected with Flag-tagged Slug, active ERK was detected in Flag pulldowns following EGF-mediated ERK activation (Fig. 5A). However, these data are not indicative that the association is specific to active ERK only. In addition, we found that Slug is phosphorylated in cells in a MEK–ERK activity–dependent manner. In vivo, Slug phosphorylation in ³²P-labeled MDA-MB-231 cells was inhibited in the presence of the MEK1/2 inhibitor U0126 (Fig. 5B). To clarify whether ERK phosphorylates Slug directly, we performed in vitro kinase assays using purified recombinant GST-Slug and recombinant ERK1 and ERK2 kinases. Both ERK1 and ERK2 directly phosphorylated Slug (Fig. 5C). Interestingly, phosphorylation of Slug by ERK isoforms appeared to be specific as another kinase, Akt1, was not able to induce Slug phosphorylation (Fig. 5C). The possible ERK phosphorylation sites in Slug were determined by liquid chromatography-tandem mass spectrometry (LC/MS-MS) analysis (Supplementary Fig. S4A). Interestingly, two major serine phosphosites (S87 and S104) were identified that matched the minimal consensus ERK phosphorylation motif (pSP) and ERK2 appeared to be more efficient, than ERK1, in phosphorylating these sites (Supplementary Fig. S4B). Unlike the four conserved central GSK3β phosphorylation sites in Slug and Snail (25, 26) that are implicated in regulation of protein stability, the two ERK consensus motifs identified in Slug are absent in Snail (Fig. 5D), suggesting that ERK may regulate Slug distinctly from Snail (14). To confirm the ERK phosphorylation sites within Slug, we generated a recombinant Slug mutant with alanine substitutions at the specific phosphoserine residues (Slug S87,104AA), identified by LC/MS-MS, and used this as a substrate for in vitro kinase assays. ERK-dependent phosphorylation of Slug S87,104AA was clearly reduced compared with the WT Slug protein (Fig. 5E). The relevant contribution of the two ERK-sites to Slug phosphorylation was further studied in cells. Phos-Tag resolution of cell lysates, expressing exogenous Slug constructs, followed by immunolabeling with anti-Slug antibody revealed a single slow-migrating phosphorylated Slug band in cells expressing Slug WT and Slug S104A only. In contrast, Slug S87A and Slug S87,104AA double mutant migrated at a size corresponding to the control recombinant nonphosphorylated Slug on Phos-Tag gels (Fig. 5F). Together, these data indicate that S87 is the principal ERK phosphorylation site in Slug as disrupting the S104 site does not influence the phosphorylation status of Slug in MDA-MB-231 cells, albeit this residue can be phosphorylated by ERK in vitro.

To examine whether endogenous Slug is phosphorylated on serine 87, we generated a recombinant antibody against phosphorylated S87. The antibody detected Slug from cell lysates, but not the nonphosphorylated recombinant Slug and antibody reactivity, was lost upon Slug silencing (Fig. 5G), demonstrating antibody specificity toward phosphorylated Slug. Importantly, treatment of MCF10A and H-Ras–transformed MCF10A cells with MEK1/2 inhibitor, U0126, clearly reduced phosphorylation of Slug in cells, providing further evidence that Slug is phosphorylated on serine 87 in a MEK/ERK1/2-dependent manner (Fig. 5G and Supplementary Fig. S4C). Antibody specificity was further validated by transfecting cells with Flag-Slug WT/S87A/S87,104AA constructs. S87-P antibody clearly showed reduced immunoreactivity toward overexpressed Slug mutants as compared with WT Slug (Supplementary Fig. S4D). Importantly, TGFβ induced the levels of S87-P in MCF10A cells and this was sensitive to MEK1/2 inhibition (Fig. 5H). Finally, overexpression of a constitutively active MEK–ERK2 fusion plasmid in MDA-MB-231 cells increased the phosphorylation levels of S87. In line with a recent report demonstrating ERK2-mediated phosphorylation and stabilization of Snail (14), cells expressing active ERK2 possessed higher levels of total Snail protein (Fig. 5I).

The apparent vimentin–ERK coregulatory loop (Fig. 3) implied to us that ERK may regulate vimentin levels through a transcription factor, possibly occurring via the ERK-mediated phosphorylation of Slug. To investigate whether phosphorylation of Slug by ERK is functionally relevant for the EMT phenotype, we used MCF7 breast cancer cells that are negative for endogenous Slug and vimentin. We cotransfected these cells with control Flag or Flag-Slug WT/S87A/S87,104AA constructs and a vimentin promoter—GFP reporter (11). Introduction of Slug WT strongly promoted vimentin transcription (GFP signal), whereas Slug S87A or Slug S87,104AA mutants did not (Fig. 6A). Similar results were obtained with another vimentin reporter construct (vimentin promoter fused to luciferase) where vimentin induction was significantly impaired in cells expressing Slug S87A and Slug S87,104AA compared with cells expressing Slug WT, while the DNA-binding-deficient Slug mutant (C161,215A; Slug DM) was fully inactive (Fig. 6B). In agreement with these data, vimentin expression was strongly upregulated in retrovirally transduced MCF10A cells expressing Slug WT and clearly diminished in Slug S87A and S87,104AA expressing cells as analyzed by Western blot analysis (Fig. 6C) and immunofluorescence (Fig. 6D). Furthermore, MCF10A cells expressing GFP_Slug WT exhibited enhanced migration compared with GFP and GFP_Slug S87A-transduced cells (Fig. 6E). Expression of GFP_Slug 87A also enhanced cell migration as compared with GFP construct alone; however, these data failed to reach statistical significance. Thus, it appears that the phosphorylation of Slug is at least partially dispensable for the induction of cell motility.

Remarkably, the requirement for Slug S87 phosphorylation was specific for regulation of vimentin induction and was fully dispensable for Slug repression of E-cadherin. In E-cadherin reporter assays (35), Slug WT and all phosphorylation site mutants of Slug significantly inhibited transcriptional activity of the E-cadherin promoter to the same extent in HEK293 (Supplementary Fig. S5A) and MCF7 cells (Supplementary Fig. S5B). In addition, E-cadherin protein levels were also repressed in MCF10A cells stably transfected with Slug retrovirus (Supplementary Fig. S5C). Slug-mediated repression of E-cadherin involves conserved E-box motifs in the promoter, whereas the vimentin promoter lacks obvious consensus E-box motifs in the 1.5-kb segment used here in the promoter assays (11). Accordingly, we found that recombinant unphosphorylated Slug interacted with the E-cadherin promoter DNA, whereas Slug failed to interact directly with the vimentin promoter DNA regardless of phosphorylation status (Supplementary Fig. S5D).

Taken together, these data suggest an intriguing and novel role for ERK-mediated phosphorylation of Slug that is distinct from the previously reported Slug family protein phosphorylation events. The ERK-mediated phosphorylation of Slug S87 is necessary for the ability of Slug to induce vimentin expression and importantly is separate from the role of Slug as a transcriptional repressor of E-cadherin. As phosphorylated Slug does not directly bind the vimentin promoter, it is likely that vimentin upregulation by Slug involves additional transcription factors that are recruited to Slug in a Slug-phosphorylation–dependent manner. However, this remains to be investigated.

Recently, there has been intense focus on the link between phosphorylation of Slug family transcription factors and their stability and nuclear localization. We found that all GFP-tagged Slug mutants localized predominantly to the nucleus in MCF10A cells with no significant difference detected between the serine phosphosite mutants and Slug WT (Supplementary Fig. S5E). This nuclear staining was not a consequence of the GFP tag as results were reproducible in cells expressing Flag-tagged Slug mutants (not shown). Furthermore, silencing of vimentin had no effect on the relative abundance of endogenous Slug in the nuclear and cytoplasmic fractions in MDA-MB-231 cells (Supplementary Fig. S5F). This suggests that ERK does not regulate Slug nuclear localization and would be in agreement with the fact that Slug lacks the phosphorylation-masked nuclear export motif found in Snail (36). During the course of these studies, Slug appeared to be rather stable in different cell lines regardless of GSK3β activity and correspondingly we did not detect any significant differences in Slug WT or Slug S87,104AA protein stability in cells (Supplementary Fig. S5G).

We have shown earlier that vimentin regulates EMT induction and breast cancer cell migration by affecting Axl receptor tyrosine kinase expression (13). To test whether ERK-mediated phosphorylation of Slug, which was demonstrated here to be crucial for vimentin expression (Fig. 6), influences Axl protein levels, we stably transfected MCF10A cells with GFP (MCF10A_GFP) or Slug WT (MCF10A_GFP_Slug WT) or Slug phosphosite mutants (MCF10A_GFP_SlugS87A/GFP_S87,104AA). Consistent with the ability to induce vimentin, expression of Slug WT and not Slug S87A or S87,104AA mutants, significantly promoted cell-surface levels of Axl, based on FACS analysis (Fig. 7A) and total Axl expression levels detected by immunofluorescence staining (Fig. 7B). Because vimentin, ERK, and Slug were important for MDA-MB-231 cell invasion and ERK-mediated phosphorylation of Slug is required for Axl induction in MCF10A cells, we tested whether Axl inhibition could also impede cell invasion. Stable silencing of Axl in MDA-MB-231 cells did not influence levels of vimentin, pERK, or Slug but significantly reduced MDA-MB-231 cell invasion compared with control knockdown cells (Fig. 7C). These data suggest that Axl contributes to EMT-linked invasion downstream of the EMT-inducing factors critical for Axl expression. In conclusion, our data show that phosphorylation of Slug at serine 87 is essential for Slug-mediated vimentin induction and Axl expression. Importantly, direct inhibition of Axl expression or of the pathways that impinge on Axl expression hinders MDA-MB-231 cell invasion.

Here, we found that a reciprocal regulatory vimentin–ERK interaction facilitates ERK phosphorylation of Slug at a novel site (S87) that determines the ability of Slug to induce vimentin and Axl expression. Intriguingly, ERK-mediated Slug S87 phosphorylation uncouples the activating and repressing functions of Slug so that ERK-mediated S87 phosphorylation is fully dispensable for E-cadherin repression by Slug. Our working model, posits that the vimentin–ERK axis regulates EMT via phosphorylation of Slug S87 and induces expression of vimentin and Axl (Fig. 7D). In addition, ERK-dependent inhibition of GSK3β activity can enforce this EMT induction indirectly by stabilizing Slug to repress E-cadherin, as implied by data linking GSK3β activity to destabilization of Slug (25). Hence, vimentin functions as an important and central EMT signaling scaffold supporting ERK activity and possibly bringing together Slug and ERK in the cytoplasm (though this remains to be formally shown; Fig. 7D).

As discussed in detail below, several studies have investigated the role of phosphorylation on the subcellular localization and protein stability of Snail family members. In contrast, the link between phosphorylation and the ability of these transcription factors to regulate gene expression remains poorly investigated and the limited data available focuses on transcriptional repression. Phosphorylation of serine 11 and 92 is independently required for Snail repression of E-cadherin expression and for efficient recruitment of the corepressor mSin3A (37). In addition, Slug phosphorylation at serine 4 has been functionally implicated in E-cadherin repression (24); however, the detailed mechanism remains to be investigated. We show here that ERK2-mediated phosphorylation of Slug on S87 specifically influences Slug-dependent transcriptional activation of vimentin, and to the best of our knowledge, this is the first example of how repression versus activation of genes is achieved by an EMT-regulating transcription factor. Our data suggest that in contrast to Slug-mediated E-cadherin repression, the ability of Slug to induce vimentin transcription does not involve direct Slug-recruitment to the vimentin promoter DNA. In the future, it will be important to identify the additional transcription factors necessary for Slug phosphorylation–dependent vimentin induction and to fully dissect the repressor and activator functions of Slug in EMT regulation.

Vimentin is considered a canonical marker of EMT; however, its functional role remains elusive (2, 8). We have recently shown that vimentin is required for EMT induction by H-RasV12, Slug, and TGFβ (13). Accruing evidence highlights the multifaceted role of vimentin in determining cell shape, regulating cell motility, and integrin turnover and as a signaling scaffold that can bind several different proteins (12, 13, 19, 38). Therefore, a functional role for vimentin in cellular processes, including EMT, is evident, albeit incompletely understood at present. In the majority of carcinomas, vimentin is overexpressed and several studies have linked vimentin expression to tumor aggressiveness and metastasis (8). We find here that vimentin expression is critical for maintaining high ERK activity in cells. The ability of vimentin to bind to and to protect ERK from dephosphorylation is most likely a contributing factor to increased ERK activity. However, we also expect other mechanisms to be involved in the regulation of ERK activity as the proportion of ERK sequestered by vimentin in cells in low. Conversely, ERK activity regulates vimentin expression and Slug phosphorylation in breast cancer cells.

Our identification of Slug, active ERK, and vimentin coexpression at the invasive edges of triple-negative breast carcinomas and the role of these proteins in regulating breast cancer cell motility and invasion in vitro suggest that this pathway would be an important clinical target. Furthermore, the fact that vimentin-positive MDA-MB-231 cell metastases, from orthotopic xenografts, express higher level of vimentin than in the primary tumors (Fig. 1E) and that vimentin contributes to xenograft growth in vivo (Fig. 3C), underlines the apparent role for vimentin in cancer progression and metastatic dissemination. Therefore, understanding the role and regulation of vimentin in EMT may unravel new strategies to target this important pathway in cancer.

GSK3β has been suggested to regulate the stability of Slug and Snail in epithelial cells via posttranslational modification. Phosphorylation-deficient alanine substitutions of four key phosphorylation sites, which appear to be conserved between Slug and Snail, results in a more stable protein with increased efficiency in EMT induction (14, 23, 25, 26). One of these phosphorylation sites (S104 in Slug), identified here as an ERK phosphorylation site in vitro, is also phosphorylated by GSK3β. According to our data, this site may not be prominently phosphorylated at least in MDA-MB-231 cells where mutagenesis of another serine (S87) clearly reduces overall Slug phosphorylation. This further emphasizes the seemingly dominant role of ERK in the regulation of Slug function because in addition to controlling GSK3β activity ERK-mediated phosphorylation of Slug at S87 controls the overall phosphorylation of Slug in cells.

It is possible that these findings are also relevant for our understanding of the regulatory networks that determine the mammary stem cell state and regulate human breast cancer stem cells (9). Vimentin is abundantly expressed in stem cells and is induced alongside Slug during conversion of luminal mammary epithelial cells into MaSC (9). The vimentin–ERK axis is critically important for Axl tyrosine kinase expression in breast cancer cells (13), and Axl expression is also associated with the expression of stem cell genes in addition to metastasis-linked genes (39). Therefore, it is likely that vimentin–ERK coregulation and contribution to Slug phosphorylation is also functional in the regulation of stemness. However, more studies are required to further examine the activity of this pathway and its potential implications in tumorigenesis, metastasis, and the maintenance of human breast cancer stem cells.

J.B. Lorens is a CSO at BerGenBio and has ownership interest (including patents) in the same. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R. Virtakoivu, E. Mattila, J.B. Lorens, J. Ivaska

Development of methodology: E. Mattila, G. Corthals

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. Virtakoivu, A. Mai, E. Mattila, N. De Franceschi, S.Y. Imanishi, G. Corthals, M. Saari, F. Cheng, E. Torvaldson, A. Mannermaa, G. Muharram, Y. Soini, J. Ivaska

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R. Virtakoivu, E. Mattila, S.Y. Imanishi, G. Corthals, F. Cheng, Y. Soini

Writing, review, and/or revision of the manuscript: R. Virtakoivu, A. Mai, N. De Franceschi, G. Corthals, R. Kaukonen, V.-M. Kosma, A. Mannermaa, C. Gilles, J. Eriksson, Y. Soini, J.B. Lorens, J. Ivaska

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. Virtakoivu, A. Mai, R. Kaukonen, M. Saari, A. Mannermaa, C. Gilles, Y. Soini

Study supervision: J. Ivaska

Other (expertise and model systems related to vimentin): J. Eriksson

The authors thank L. Lahtinen, J. Siivonen, S. Kaustara, K. Vuoriluoto, and H. Marttila for excellent technical assistance and H. Hamidi for scientific writing and editing of the article. The authors are grateful to Dr. J. Westermarck for sharing his reagents and critically reading the article.

This study has been supported by the Academy of Finland, ERC Starting Grant (#202809), ERC Consolidator Grant (#615258), the Sigrid Juselius Foundation, and the Finnish Cancer Organization. R. Virtakoivu has been supported by the K. Albin Johansson Foundation, Lounais-Suomen syöpäyhdistys, Instrumentariumin tiedesäätiö, Orion Research Foundation and by the Turku Doctoral Program of Biomedical Sciences. N. De Franceschi has been supported by the Drug Research Doctoral Program. E. Mattila has been supported by the Academy of Finland postdoc grant.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.