EGF augments TGFβ-induced epithelial–mesenchymal transition by promoting SHP2 binding to GAB1

Epithelial–mesenchymal transition (EMT) is a cellular program in which epithelial cells lose polarity and strong cell–cell adhesions and partially dedifferentiate to become more motile and invasive (Kalluri and Weinberg, 2009). EMT is accompanied by shifts in transcription factor expression, cytoskeletal arrangement, RNA splicing, and cell signaling (Thiery, 2003; Massagué, 2008; Kalluri and Weinberg, 2009). Although EMT is required for normal development and wound healing, aberrant EMT has also been linked to metastasis (Chaffer and Weinberg, 2011; Rhim et al., 2012) and tumor resistance to therapy (Tan et al., 2014). In lung and pancreatic cancer cells, sensitivity to therapeutic agents can be increased by promoting an epithelial phenotype (Witta et al., 2006; Arumugam et al., 2009; Buonato and Lazzara, 2014). These findings have motivated studies to decipher the signaling and regulatory mechanisms underlying EMT (Davis et al., 2014; Dragoi et al., 2014; Wilson et al., 2014).

EMT is efficiently driven in vitro and in vivo by transforming growth factor β (TGFβ) (Massagué, 2008), which initiates a network of cellular signaling pathways including SMADs, extracellular signal-regulated kinases 1 and 2 (also known as MAPK3 and MAPK1, respectively; ERK1/2), p38 proteins, and c-Jun N-terminal kinase (Zhang, 2009; Gonzalez and Medici, 2014). Activity of ERK1/2 (hereafter referred to as ERK) has been identified as particularly important for TGFβ-driven EMT (Zavadil et al., 2001; Grände et al., 2002; Buonato and Lazzara, 2014). Epidermal growth factor (EGF), which initiates its own signaling network including ERK, is often combined with TGFβ to enhance EMT outcomes in vitro (Saha et al., 1999; Grände et al., 2002; Docherty et al., 2006; Uttamsingh et al., 2008), and the augmented EMT that results has been attributed to ERK activation (Grände et al., 2002; Uttamsingh et al., 2008). In some cell settings, the addition of TGFβ alone to cells indirectly leads to tyrosine phosphorylation of the EGF receptor (EGFR), which augments ERK pathway activation (Murillo et al., 2005; Joo et al., 2008).

Downstream of EGFR and other receptor tyrosine kinases (RTKs) complete ERK activation depends upon the activity of SH2 domain-containing phosphatase-2 (SHP2; also known as PTPN11), a cytosolic protein tyrosine phosphatase. In the basal state, an auto-inhibited SHP2 conformation is stabilized wherein the N-terminal SH2 domain binds and occludes the catalytic domain (Barford and Neel, 1998; Hof et al., 1998). Binding of SHP2 SH2 domains to phosphotyrosines on various RTKs or on adaptor proteins such as GRB2-associated binding protein-1 (GAB1) relieves this intramolecular inhibitory mechanism and allows SHP2 phosphatase domain access to substrates (Lechleider et al., 1993; Pluskey et al., 1995; Eck et al., 1996; Gu and Neel, 2003; Neel et al., 2003). Point mutations of the βB5 arginine residues in the FLVRES motifs of either or both of the SH2 domains reduce the ability of phosphotyrosine peptides to induce SHP2 phosphatase activity, with larger effects for the mutation of the more N-terminal SH2 domain arginine, Arg³² (Sugimoto et al., 1994; Pluskey et al., 1995). Furthermore, SHP2 isoforms with either or both of these SH2 domain point mutants (R32K, R138K) display impaired ability to promote downstream signaling and lead to developmental defects in mice (O’Reilly and Neel, 1998; Kapoor et al., 2004; Stewart et al., 2010), again with larger effects for mutation of Arg³² in the more N-terminal SH2 domain. SHP2-activating mutations occur in cancers including juvenile myelomonocytic leukemia, acute myelogenous leukemia, and lung and colon carcinomas, as well as Noonan syndrome, a developmental disorder leading to increased risk of certain malignancies (Bentires-Alj et al., 2004; Mohi and Neel, 2007; Chan et al., 2008). In mouse tumor xenograft studies of breast and brain cancer cells, SHP2 promotes tumorigenesis (Aceto et al., 2012; Furcht et al., 2014). SHP2 activity also contributes to cellular resistance to EGFR inhibition in non-small-cell lung carcinoma cells (Furcht et al., 2013). SHP2 has been identified as a promoter of TGFβ-mediated EMT in A549 lung carcinoma cells (Li et al., 2014), but whether differential SHP2 activation and consequent regulation of ERK activity play a role in the ability of EGF to enhance TGFβ-mediated EMT is unknown.

We used lung and pancreatic cancer cell lines to explore the role of SHP2 in driving EMT in response to TGFβ and/or EGF. Whereas treatment of cells with TGFβ alone produced a partial EMT and treatment with EGF alone produced very modest EMT effects, the combination of TGFβ and EGF produced a more complete EMT as determined by relatively large decreases in E-cadherin expression, large increases in vimentin expression and increased cell scatter rates. In contrast to TGFβ, EGF (alone or in combination with TGFβ) induced SHP2 association with phosphorylated GAB1, which augments SHP2 activity. Knockdown of endogenous SHP2 promoted baseline epithelial cell characteristics and inhibited EMT in response to TGFβ with or without EGF. Reconstitution of SHP2^WT restored EMT in response to TGFβ and the ability of EGF to enhance TGFβ-mediated EMT, but reconstitution with SHP2^R32,138K, a mutant with impaired ability to bind phosphorylated GAB1, did not restore EMT induction compared with empty vector controls. These results suggest that basal SHP2 activity is needed for the incomplete EMT observed in response to TGFβ alone, but that the activation of SHP2 induced by its association with phosphotyrosine-containing proteins such as GAB1 is needed for the more complete EMT observed when ligands such as EGF are combined with TGFβ.

To evaluate the abilities of TGFβ, EGF, or the combination of the ligands to promote EMT, we began with a panel of lung (H322 and H358) and pancreatic (HPAF-II and CAPAN-2) carcinoma cell lines that display baseline epithelial characteristics (Rajasekaran et al., 2004; Thomson et al., 2005; Deer et al., 2010) and represent common genotypes for the respective cancers. H322 is wild-type for EGFR, RAS and RAF; H358, HPAF-II and CAPAN-2 harbor KRAS mutations (Helfrich et al., 2006; Deer et al., 2010). SMAD4 is mutated or deleted in ∼50% of pancreatic cancers, and SMAD4 participates in TGFβ receptor-mediated signaling. However, there are conflicting reports on the effects of SMAD4 mutation on EMT (Levy and Hill, 2005; Bardeesy et al., 2006). HPAF-II and CAPAN-2 cell lines were thus chosen for their wild-type SMAD4 status to avoid the need to consider the uncertain, and potentially variable, effects of SMAD4 mutation.

The cell lines were treated with 50 ng/ml EGF, 10 ng/ml TGFβ, or a combination of the ligands for 5 days. TGFβ treatment reduced expression of the epithelial marker E-cadherin and promoted expression of the mesenchymal marker vimentin (Fig. 1A). EGF treatment produced a modest reduction in E-cadherin expression. When TGFβ was combined with EGF, larger shifts in E-cadherin and vimentin expression were observed than with EGF or TGFβ alone, particularly for H322 and HPAF-II cells. These basic trends were further confirmed in a subset of the cell lines by probing for additional EMT markers (Fig. S1A,B). For H358 and CAPAN-2 cells, differences between the effects of TGFβ alone and the combination of TGFβ and EGF were smaller and not statistically significant. Differences in signaling downstream of TGFβ and/or EGF among cell lines may explain these observations, as suggested by data presented later (Fig. S4). Given the clearer additive effects of ligands in H322 and HPAF-II cells, we selected these for further quantitative studies.

For the same ligand treatment conditions, H322 and HPAF-II cells were imaged using phase contrast and immunofluorescence microscopy (Fig. 1B,C). Whereas EGF had relatively little effect on cellular morphology, TGFβ treatment shifted cell morphology from a cobblestone to a more spindle-shaped appearance and reduced the number of cell–cell contacts. This effect was further enhanced when TGFβ and EGF were used in combination. Quantitative analysis of immunofluorescence images of E-cadherin and vimentin in H322 and HPAF-II cells (Fig. 1C) confirmed the trends shown in Fig. 1A and provided additional information at the single-cell level. Quantifying the ratio of E-cadherin staining at individual cell–cell junctions relative to junction-proximal regions, as in previous studies related to cell adhesion (Loerke et al., 2012; Čáslavský et al., 2013), demonstrated that the overall decrease in E-cadherin expression observed by western blot was accompanied by a delocalization of E-cadherin from cell–cell junctions in response to TGFβ with or without EGF. Analysis of vimentin staining revealed that the higher total levels of vimentin expression measured by western blot reflected an increased percentage of vimentin-positive cells. These effects were most profound when EGF was combined with TGFβ.

It is well known that EGFR activation results in robust activation of the ERK pathway. ERK activity is required for EMT in a number of cell lines (e.g. Buonato and Lazzara, 2014), and was indeed required for complete EMT in H322 and HPAF-II cells (Fig. S1C). Given that TGFβ was unable to promote substantial ERK activity by itself in most cell lines we tested (Fig. 1A), ERK could be involved in the ability of EGF to augment TGFβ-mediated EMT. As EGF alone was unable to drive substantial EMT, however, ERK activation may be necessary but insufficient for complete EMT induction.

Cells undergoing EMT often become more migratory and invasive (Kalluri and Weinberg, 2009). Weakening cell–cell junctions within preformed epithelial cell clusters can lead to disrupted cluster circularity and scatter of cells away from these clusters (Loerke et al., 2012). These changes can be monitored by measuring a cluster circularity index, calculated as 4π(area/perimeter²), which has a value of 1 for a circle or 0 for a line (Čáslavský et al., 2013). To assess the ability of TGFβ and/or EGF to promote cellular migration through such measurements, clusters containing approximately 10–20 H322 or HPAF-II cells were grown from individual cells and monitored for scatter and morphological changes over 24 h in response to EGF, TGFβ or both ligands (Fig. 2A,B). For H322 and HPAF-II cells, neither ligand by itself produced significant deviations in circularity index compared with controls, but the combination of TGFβ and EGF produced statistically significant decreases in circularity index. Similarly, in wound closure assays using H322 cells, pretreatment of cells with EGF or TGFβ alone did not enhance wound closure, but the combination of EGF and TGFβ produced a statistically significant 2.7-fold increase in the wound closure rate (Fig. 2C). This effect was accompanied by individual cells migrating away from the cell sheet (arrowheads in the corresponding phase contrast image). In contrast, HPAF-II cells did not display enhanced wound closure rates for any of the ligand treatments (Fig. 2D). This might have occurred because of a reduced ability to observe EMT induction at the high cell density required for this monolayer assay in HPAF-II, but not H322, cells (Fig. S2).

Given the established role of ERK activity in EMT, we assessed how SHP2, which generally promotes ERK activity, contributes to EMT phenotypes, looking first at baseline cellular characteristics. SHP2 was depleted by stable shRNA expression in the panel of cell lines used in Fig. 1A as well as additional lung (PC9) and pancreatic (AsPC1) cancer cell lines that display clear baseline mesenchymal characteristics (Fig. 3). Cells with SHP2 knockdown tended to group into larger epithelial clusters than control counterparts (Fig. 3A). For some cell lines, SHP2 depletion also led to a clear transition from a spindle-shaped morphology to a cobblestone appearance, an effect especially apparent in PC9 and AsPC1 cells (Fig. 3A, insets). In most cases, SHP2 knockdown promoted significant increases in E-cadherin expression and decreases in vimentin expression as well as reduced ERK phosphorylation (Fig. 3B). These morphological and protein expression shifts with SHP2 knockdown were accompanied by decreased wound closure rates (Fig. 3C). Thus SHP2 knockdown promoted a mesenchymal–epithelial transition (MET) of cells in the absence of EGF or TGFβ.

We next evaluated how SHP2 knockdown impacted cellular response to TGFβ with or without EGF. Measurements by western blotting demonstrated that SHP2 depletion tended to inhibit reduction of E-cadherin expression and induction of vimentin expression in response to growth factor treatments in H322 and HPAF-II cells (Fig. 4A,B; Fig. S3A,B), although the effects were not always large or statistically significant. Similar trends were observed in H358 and CAPAN-2 cells (Fig. S3C,D). This correlated with impaired ERK phosphorylation in cells with SHP2 knockdown, which was most apparent for EGF treatment conditions (Fig. S3A,C,D). Western blot analysis was supported by immunofluorescence imaging in H322 and HPAF-II cells (Fig. 4C,D), where the effects observed were in some cases larger and more often statistically significant than the effects observed by western blot. SHP2-deficient cells displayed some loss of total E-cadherin in response to TGFβ with or without EGF, but E-cadherin was clearly maintained at cell–cell junctions. The appearance of a vimentin-positive cell population was significantly inhibited by SHP2 knockdown in both cell lines. Importantly, SHP2 depletion inhibited EMT induction in response to TGFβ with or without EGF addition, indicating that the role of SHP2 in this setting extends beyond regulating response to EGFR activation only.

The results of Fig. 4 suggest an important role of SHP2 activity for EMT induction in response to TGFβ alone or the combination of TGFβ and EGF. Given that SHP2 binding to phosphotyrosine-containing proteins through its N-terminal SH2 domains stabilizes an active SHP2 conformation (Sugimoto et al., 1994), we indirectly assessed the extent of SHP2 activation for different ligand treatment conditions by probing for SHP2 associations with phosphotyrosine-containing proteins in response to EGF, TGFβ, or a combination of the two through SHP2 immunoprecipitation experiments in H322 and HPAF-II cells (Fig. 5A). In response to EGF, a phosphotyrosine-containing protein of ∼110 kDa mass co-immunoprecipitated with SHP2 in both cell lines. This protein was confirmed to be GAB1 phosphorylated at Tyr⁶²⁷ by immunoblotting with GAB1-specific antibodies. TGFβ did not promote any SHP2-phosphotyrosine associations by itself, nor did it affect the extent of SHP2-phosphotyrosine associations observed when combined with EGF compared with EGF alone. It should be noted that by the 30 min treatment time chosen for this experiment, the phosphorylation of signaling intermediates downstream of EGFR (ERK1/2) and TGFβ receptor (SMAD2/3) (Fig. 5B) were observed, demonstrating that EGF and TGFβ treatments had both initiated signaling processes by that time.

Using the same SHP2 immunoprecipitation approach, direct comparisons were made among H322, HPAF-II and H358 cell lines (Fig. S4). The results of this experiment demonstrated a relatively low level of EGF-induced GAB1–SHP2 association in H358 cells compared with H322 and HPAF-II cells. This might explain the relatively poor ability of EGF to augment TGFβ-mediated EMT in certain cell lines (Fig. 1A) and the weaker effects of SHP2 depletion on ligand-induced EMT in these cells (Fig. S3C).

The observations that EMT depended on SHP2, which associated with GAB1 in response to EGF but not TGFβ, and that EGF augmented TGFβ-driven EMT outcomes suggested that EGF-induced SHP2 activation might be responsible for the ability of EGF to augment EMT. To test this model for the contribution of SHP2 binding to phosphotyrosine-containing proteins in EMT, we reconstituted cells depleted of endogenous SHP2 with either wild-type SHP2 or the double SH2 domain mutant SHP2^R32,138K. For these experiments, knockdown of endogenous SHP2 was accomplished using two independent shRNA constructs targeting the 3′ untranslated region (UTR) of SHP2 (SHP2 shRNA-2&3) in HPAF-II cells to allow for RNAi-resistant reconstitution of SHP2 using expression constructs lacking the UTRs (Fig. 6; Fig. S3B). The R-K mutations antagonize the ability of SHP2 SH2 domains to bind phosphotyrosines (O’Reilly and Neel, 1998; Kapoor et al., 2004). In HPAF-II cells, SHP2^WT reconstitution rescued the ability of TGFβ treatment, with or without EGF, to reduce E-cadherin expression and promote vimentin expression (Fig. 6A). Cells with SHP2^R32,138K reconstitution, however, responded more closely to cells with SHP2 knockdown, displaying E-cadherin maintenance and impaired vimentin induction in response to TGFβ and EGF. Although cells transduced with shRNA-2&3 and the empty expression vector did not display obviously impaired ERK phosphorylation compared with cells transduced with control shRNA and empty expression vector, ERK phosphorylation was significantly impaired in cells with SHP2^R32,138K reconstitution compared with cells with SHP2^WT reconstitution, which was associated with impaired EMT induction (Fig. 6A). The lack of apparent effect on ERK phosphorylation with shRNA-2&3, at least for the time point tested, could result from higher residual endogenous SHP2 expression than that remaining after transduction of the same cell line with shRNA-1 (Fig. 4B; Fig. S3A), where ERK phosphorylation was indeed reduced by SHP2 depletion. Immunofluorescence imaging confirmed that SHP2^R32,138K reconstitution impaired shifts in E-cadherin and vimentin expression in response to TGFβ with or without EGF compared with cells with SHP2^WT reconstitution (Fig. 6B). Immunoprecipitation experiments confirmed that SHP2^R32,138K did not associate with tyrosine phosphorylated proteins in response to EGF treatment (Fig. 6C). Thus, the ability of SHP2 to promote EMT in response to exogenous ligands depends upon SHP2 engagement of phosphotyrosine residues on cognate adaptor proteins such as GAB1, and the complete EMT observed when EGF and TGFβ were combined was due to SHP2 activation driven by EGF specifically.

Our results identify a mechanism wherein SHP2 SH2 domain engagement of a phosphotyrosine-containing protein can contribute to EMT and thus uncover a new aspect of the documented ability of EGF to augment EMT in response to TGFβ, as summarized in Fig. 7. In our proposed model, TGFβ binding to its receptors activates some pathways that promote a partial EMT. Although TGFβ does not elevate SHP2 activity above the basal level, basal activity of SHP2 plays a role in EMT driven by TGFβ alone. In response to EGF binding to EGFR, GAB1 becomes tyrosine phosphorylated and SHP2 SH2 domain engagement of phosphorylated GAB1 increases SHP2 activity, but this effect alone is insufficient for substantial EMT. When SHP2 activity augmentation in response to EGF occurs with simultaneous signaling through the TGFβ receptor, however, strong EMT occurs. Of course, the ability to increase SHP2 activity sufficiently to augment TGFβ-mediated EMT may not be unique to EGF. Growth factors including hepatocyte growth factor (HGF) and oncostatin-M (Argast et al., 2011), platelet-derived growth factor (PDGF) (Gotzmann et al., 2006) and IL-6 (Yao et al., 2010) have been reported to enhance TGFβ-mediated EMT, and signaling downstream of many of these ligands involves SHP2. It is unclear, however, whether SHP2 participates in the augmented EMT that accompanies the combination of these ligands with TGFβ.

Although the model proposed in Fig. 7 captures the effects observed in the two cell lines explored here in most detail (H322 and HPAF-II), it should be emphasized that not all aspects of the model held in other cell lines. For example, although SHP2 knockdown clearly increased baseline epithelial characteristics in H358 and CAPAN-2 cells (and indeed most of the cell lines analyzed in Fig. 3), EGF did not substantially augment EMT in response to TGFβ in those cell lines. At least within the panel of cell lines we tested, this might suggest a broader relevance of basal SHP2 activity than growth factor-induced SHP2 activity in determining epithelial or mesenchymal characteristics. The results in Fig. S4 offer one possible explanation for the inability of EGF to augment TGFβ-mediated EMT in some cell lines.

Based on our previous work and evidence from other studies demonstrating the importance of ERK activity in EMT (Shin et al., 2010; Buonato and Lazzara, 2014), SHP2-mediated augmentation of ERK activity is likely to be a critical factor in SHP2-mediated effects on EMT. Although ERK pathway regulation by SHP2 is important downstream of EGFR, SHP2 regulates other signaling processes, some of which may regulate EMT. For example, SHP2 has been shown to dephosphorylate focal adhesion kinase (FAK) at the leading edge of a cell to promote focal adhesion turnover (Hartman et al., 2013). By regulating FAK and other components of focal adhesions, SHP2 promotes cellular migration, polarity and mechanical sensing (Hartman et al., 2013; Lee et al., 2013), phenotypes with obvious connections to EMT. Inhibiting or stably depleting FAK inhibits TGFβ-driven signaling, EMT and tumorigenesis in mice (Wendt and Schiemann, 2009). In addition, we recently described a mechanism by which EGFR activation leads to formation of SHP2-GAB1 complexes that can remain associated distal from the receptor (Furcht et al., 2015). In the mechanism we identified, Src family kinases (SFKs) served as effectors activated by EGFR that maintained GAB1 phosphorylation, and therefore GAB1–SHP2 complexes, distal from EGFR through multiple rounds of GAB1-SHP2 dissociation and GAB1 tyrosine dephosphorylation. Other studies have identified SFKs as necessary intermediates for EGFR-dependent signaling initiated by TGFβ (Murillo et al., 2005; Joo et al., 2008), which could be related to the ability of SFKs to promote SHP2 activation via GAB1–SHP2 complex formation or could involve other SFK signaling functions.

Another effect observed in this study was the ability of high cell density to inhibit EMT in HPAF-II, but not H322, cells. Such cell context-dependent effects could arise from perturbations to signaling processes regulating EMT that occur preferentially in specific cell lines. Indeed, previous work showed that high cell density can significantly impair TGFβ-induced SMAD nuclear localization in a cell line-dependent manner (Zieba et al., 2012; Nallet-Staub et al., 2015). An effect of cell density on EMT could be particularly important in vivo where epithelial cells are packed tightly in three-dimensional structures. Of course, other tumor microenvironmental cues could potentially override any EMT inhibitory effects on SMAD signaling resulting from high cell density.

Within the SMAD pathway, it would also be informative to evaluate the effects of SMAD4 deletion and mutation, which are present in ∼50% of pancreatic carcinomas, 10–35% of colorectal carcinomas, and a small percentage of lung and other carcinomas (Schutte et al., 1996; Miyaki and Kuroki, 2003), on the TGFβ- and EGF-mediated effects reported here. HPAF-II and CAPAN-2 cells are both SMAD4 wild-type, whereas AsPC1 cells have a SMAD4 mutation. Though all three cell lines demonstrated SHP2-dependent mesenchymal traits, we did not specifically investigate the differential ability of cell lines with or without SMAD4 mutation to undergo EMT in response to TGFβ and/or EGF. Previous work on this specific point has produced conflicting results (Levy and Hill, 2005; Bardeesy et al., 2006). More concrete determination of the role of SMAD4 in EMT in response to growth factor signaling will require further studies in isogenic cell backgrounds.

Despite the aforementioned nuances of the specific and context-dependent roles SHP2 plays, it is clear that SHP2 can participate in EMT. Of course, a role for SHP2 in EMT is consistent with its identification as a proto-oncogene and the concept of SHP2 as a driver of malignancy and therapeutic resistance in carcinomas of the colon, lung and breast (as well as non-epithelial cancers including leukemia and glioblastoma). These and other findings motivate the ongoing search for specific inhibitors of SHP2. New strategies to develop SHP2 inhibitors that target allosteric binding sites unique to SHP2 have shown promise in achieving SHP2 specificity (Yu et al., 2013; Chio et al., 2015). Therapies developed using such strategies would have potential application in many diseases. In particular, our findings suggest that SHP2 may be a potential target for reversing the mesenchymal dedifferentiation process in malignant tumor cells through chronic inhibition prior to administration of another targeted cancer therapeutic, analogous to a strategy we described previously wherein chronic ERK inhibition led to enhanced EGFR inhibitor response in non-small cell lung carcinoma (NSCLC) cells (Buonato and Lazzara, 2014).

H322 cells (Dr Pasi Jänne, Dana-Farber Cancer Institute), H358 cells (Dr Russ Carstens, University of Pennsylvania), PC9 cells [Dr Douglas Lauffenburger, Massachusetts Institute of Technology (MIT)], and AsPC1 and HPAF-II cells (Dr Carl June, University of Pennsylvania) were maintained in RPMI medium supplemented with 10% fetal bovine serum (FBS), 1 mM L-glutamine, 100 units/ml penicillin and 100 μg/ml streptomycin. CAPAN-2 cells (Dr Carl June) were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented as RPMI above. Cell culture reagents were from Life Technologies (Carlsbad, CA). Recombinant human EGF and TGFβ were purchased from Peprotech (Rocky Hill, NJ).

DNA oligonucleotides encoding short hairpins targeting human SHP2 (Integrated DNA Technologies, San Jose, CA) were cloned into the pSicoR vector (Tyler Jacks, MIT; Ventura et al., 2004). shRNA-1 targeted nucleotides 1780–1798 of SHP2 mRNA (5′-GGACGTTCATTGTGATTGA-3′); shRNA-2 targeted nucleotides 5890–5908 (5′-GTATTGTACCAGAGTATTA-3′); and shRNA-3 targeted nucleotides 4931–4949 (5′-GCTGGTGGGTATTAAATAT-3′). Control vectors were created using shRNA sequences that do not target a known human mRNA. For stable shRNA expression, lentiviral particles were produced by calcium phosphate-mediated transfection of 293FT cells (Life Technologies) with pSicoR plasmids along with the packaging plasmids pCMV-VSV-G, pMDL-gp-RRE and pRSV-Rev (Dr Marilyn Farquhar, University of California San Diego). Virus-containing media was collected 48 and 72 h post-transfection and filtered using 0.45 µm syringe filters before infecting target cells, which were subsequently selected and maintained in 2 μg/ml puromycin.

SHP2 cDNA encoding SHP2^WT or SHP2^R32,138K (Ben Neel, New York University) was inserted between the BamHI and EcoRI sites of the pBabe.Hygro vector. Retroviral particles were produced by calcium phosphate-mediated transfection of amphotropic Phoenix cells (Dr Gary Nolan, Stanford University) with vector. Virus-containing media was harvested 24 and 48 h post-transfection and used to infect target cells, which were selected and maintained in 200 μg/ml hygromycin. All expression and shRNA constructs were validated by sequencing.

pERK T202/Y204 (4377), ERK (4695), Slug (9585), ZEB1 (3396), ZO-1 (8193), pSMAD2/3 S465/467/S423/425 (8828) and pGAB1 Y627 (3233) antibodies were from Cell Signaling Technology (Danvers, MA). E-cadherin (sc-8426), vimentin (sc-373717), SHP2 (sc-7384), GAB1 (sc-9049) and GAPDH (sc-32233) antibodies were from Santa Cruz Biotechnology (Dallas, TX). Phosphotyrosine antibody (4G10) was from Millipore (Billerica, MA). Infrared dye- and Alexa Fluor^®-conjugated secondary antibodies were from Rockland Immunochemicals (Gilbertsville, PA) and Life Technologies, respectively.

Whole cell lysates were prepared in a standard cell extraction buffer (Life Technologies) supplemented with protease inhibitors and phosphatase inhibitors (Sigma-Aldrich, St Louis, MO). Lysates were cleared by centrifugation at 16,100 g for 10 min, and total protein concentrations were determined by micro-bicinchoninic acid (BCA) assay (Thermo Fisher Scientific, Waltham, MA). Proteins were resolved on 4–12% gradient polyacrylamide gels (Life Technologies) under denaturing and reducing conditions and transferred to 0.2 μm nitrocellulose membranes (Bio-Rad Laboratories, Hercules, CA). Membranes were imaged using a LI-COR Odyssey Imaging System. As needed, membranes were stripped with 0.2 M NaOH.

Cells were lysed in an immunoprecipitation lysis buffer (Cell Signaling Technology) supplemented with protease and phosphatase inhibitors. Lysates were prepared and assayed for total protein as described above, and 500 μg of protein was incubated with protein G agarose beads conjugated to SHP2 or normal mouse IgG control antibodies at 4°C overnight. Beads were washed three times with cold lysis buffer, resuspended in LDS sample buffer (Life Technologies) and boiled before western blotting.

Cells were plated onto glass coverslips and maintained in 6-well plates. Following various durations of treatment with growth factors, cells were washed and fixed for 20 min in 4% paraformaldehyde and permeabilized with 0.25% Triton X-100 in phosphate-buffered saline (PBS) for 5 min. Washed coverslips were incubated with an E-cadherin or vimentin antibody in a humidified chamber for 3 h at 37°C. Washed coverslips were incubated with Alexa Fluor-conjugated anti-mouse secondary antibodies and Hoechst-33342 DNA stain (Sigma-Aldrich) for 1 h at 37°C. Coverslips were mounted on microscope slides using Prolong Gold Antifade mounting media (Life Technologies) and dried overnight. Fixed slides were imaged on a Zeiss Axiovert 40 CFL microscope using an A-Plan 20× objective (vimentin) or 100× oil objective (E-cadherin) and a SPOT Insight CCD camera. Identical acquisition settings were used across all images from a single experiment.

E-cadherin junctional localization was quantified using ImageJ and a method similar to that described in Loerke et al. (2012). Each cell–cell junction was traced with a 15-pixel (1.1 μm)-wide freehand line (corresponding to the average thickness of E-cadherin stained junctions in the untreated condition), for which the area (A_j), integrated density (ID_j), and area-averaged mean intensity (M_j) were measured. M_j was taken to be the average junctional signal. Three measurements, A_total, ID_total and M_total, were then taken for an increased thickness of 100 pixels (7.5 μm) encompassing junctional and peri-junctional E-cadherin staining. In order to find the mean cytoplasmic density (M_c), A_j and ID_j were first subtracted from A_total and ID_total, respectively, to obtain the cytoplasm-only quantities A_c and ID_c. M_c was then calculated by dividing ID_c by A_c. In order to quantify the extent of junctional localization of E-cadherin, the ratio M_j/M_c was calculated for each junction. All junctions contained within an image were analyzed.

Quantification of the fraction of cells positive for vimentin was also performed using ImageJ. Nuclei were counted in each image from Hoechst staining to obtain the total number of cells. In the corresponding vimentin image, cell outlines were drawn using a freehand tool from which mean intensities of vimentin staining were recorded. Cells with mean intensities higher than the intensity of background non-specific staining were considered to be vimentin-positive, and the total number of vimentin-positive cells was divided by the number of nuclei to obtain the fraction of vimentin-positive cells in each image.

Cells were plated in complete medium at low density (500 cells per well in 12-well plates). When individual cells had expanded into well-defined colonies 4–5 days later, the scatter experiment was initiated by the addition of complete medium containing EGF (50 ng/ml), TGFβ (10 ng/ml), or a combination of the two ligands. Individual cell clusters were monitored, with phase contrast images captured with a Zeiss Axiovert 40 CFL microscope (10× objective) over 24 h at time points indicated in the corresponding figures. Image segmentation was performed using the MATLAB image processing toolbox (MathWorks, Natick, MA) to identify all objects (cell clusters) within each image and record their area and perimeter. As described previously (Čáslavský et al., 2013), a circularity index was calculated as 4π(area/perimeter²) (=1 for a circle) for each cluster.

Scratches were made on confluent cell monolayers in 6-well plates using a pipette tip, and media was changed immediately thereafter. Phase contrast images were taken with a Zeiss Axiovert 40 CFL microscope (10× objective) every 1–3 h for up to 11 h. Open scratch areas were quantified using ImageJ, and closure rates were calculated from linear fits of scratch areas versus time, and reported as a percentage of total image area closed per hour. For experiments with different pretreatment conditions, cells were plated at appropriately modified densities such that all wells within an experiment reached confluence on the same day.

Statistical significance was determined using one- or two-way ANOVA in Kaleidagraph 4.0 (Synergy Software, Reading, PA). One-way ANOVA P-values were determined using a Student–Newman–Keuls test for multiple corrections in Kaleidagraph 4.0. Two-way ANOVA P-values were determined using the Bonferroni correction for multiple comparisons using the GraphPad online QuickCalcs calculator for ANOVA P-values (GraphPad Software, La Jolla, CA).

The authors thank Drs Carl June, Pasi Jänne, Russ Carstens and Ben Neel for providing reagents.