Epithelial to Mesenchymal Transition Promotes Breast Cancer Progression via a Fibronectin-dependent STAT3 Signaling Pathway*

Background: Cells perceive their environment through soluble growth factors and in response to extracellular matrix. Results: STAT3 signaling can be activated by multiple pathways during breast cancer progression. Conclusion: Fibronectin:STAT3 signaling promotes three-dimensional outgrowth of breast cancer cells. Significance: This study demonstrates a novel mechanism by which STAT3 becomes activated by the extracellular matrix independent of the canonical EGF receptor signaling network. We previously established that overexpression of the EGF receptor (EGFR) is sufficient to induce tumor formation by otherwise nontransformed mammary epithelial cells, and that the initiation of epithelial-mesenchymal transition (EMT) is capable of increasing the invasion and metastasis of these cells. Using this breast cancer (BC) model, we find that in addition to EGF, adhesion to fibronectin (FN) activates signal transducer and activator of transcription 3 (STAT3) through EGFR-dependent and -independent mechanisms. Importantly, EMT facilitated a signaling switch from SRC-dependent EGFR:STAT3 signaling in pre-EMT cells to EGFR-independent FN:JAK2:STAT3 signaling in their post-EMT counterparts, thereby sensitizing these cells to JAK2 inhibition. Accordingly, human metastatic BC cells that failed to activate STAT3 downstream of EGFR did display robust STAT3 activity upon adhesion to FN. Furthermore, FN enhanced outgrowth in three-dimensional organotypic cultures via a mechanism that is dependent upon β1 integrin, Janus kinase 2 (JAK2), and STAT3 but not EGFR. Collectively, our data demonstrate that matrix-initiated signaling is sufficient to drive STAT3 activation, a reaction that is facilitated by EMT during BC metastatic progression.


We previously established that overexpression of the EGF receptor (EGFR) is sufficient to induce tumor formation by otherwise nontransformed mammary epithelial cells, and that the initiation of epithelial-mesenchymal transition (EMT) is capable of increasing the invasion and metastasis of these cells. Using this breast cancer (BC) model, we find that in addition to EGF, adhesion to fibronectin (FN) activates signal transducer and activator of transcription 3 (STAT3) through EGFR-dependent
and -independent mechanisms. Importantly, EMT facilitated a signaling switch from SRC-dependent EGFR:STAT3 signaling in pre-EMT cells to EGFR-independent FN:JAK2:STAT3 signaling in their post-EMT counterparts, thereby sensitizing these cells to JAK2 inhibition. Accordingly, human metastatic BC cells that failed to activate STAT3 downstream of EGFR did display robust STAT3 activity upon adhesion to FN. Furthermore, FN enhanced outgrowth in three-dimensional organotypic cultures via a mechanism that is dependent upon ␤1 integrin, Janus kinase 2 (JAK2), and STAT3 but not EGFR. Collectively, our data demonstrate that matrix-initiated signaling is sufficient to drive STAT3 activation, a reaction that is facilitated by EMT during BC metastatic progression.
The extracellular matrix (ECM) 4 plays an integral role in the development and homeostatic maintenance of different organ systems (1). However, the signaling mechanisms that empower cancer cells to aberrantly utilize the ECM remain poorly defined (2). Recent evidence suggests that cooperation between integrins and growth factor receptors alters downstream signaling and may contribute to pathological responses of cancer cells to soluble ligands (3,4). Indeed, we have shown that integrins form physical complexes with EGFR and TGF-␤ receptors giving rise to oncogenic signaling patterns and pathological responses to ligands (5)(6)(7).
High levels of EGFR expression within primary mammary tumors are strongly linked to the poor prognosis of human breast cancer (BC) (8,9). Our previous studies indicate that EGFR is capable of transforming normal mammary epithelial cells (7). These findings and others (10) suggest that paracrine EGF signaling functions in concert with the process of epithelial mesenchymal transition (EMT) to facilitate tumor cell invasion and dissemination (7). Collectively, these studies have led to the clinical evaluation of EGFR inhibitors as potential targeted molecular therapies for BC (9,11). Unfortunately, the results of these trials strongly indicate that targeted inhibition of EGFR does not offer any clinical benefit to BC patients (12). Consistent with this inherent resistance of BC to EGFR inhibition, we recently established that EGFR expression is actually diminished in the later stages of metastasis (13). Interestingly, metastatic breast carcinomas also display enhanced expression and altered distribution of fibronectin (FN) compared with normal breast (14). Along these lines, enhanced expression of FN is a well established marker of the EMT process (15), and survival analysis of BC patients with high levels of FN expression is predictive for an increased risk of mortality (14).
Signal transducer and activator of transcription 3 (STAT3) is a transcription factor that regulates cell proliferation and survival; it also functions as a major player that drives the growth of stem-like BC cells (16,17). STAT3 can be activated by several signaling systems including those stimulated by interleukin-6 (IL-6) (18), Oncostatin M (19), interferons (20), and EGF (21). In light of our recent study that demonstrated the ability of the matrix to directly activate EGFR (5), we hypothesized that FN:EGFR signaling could play a pivotal role in the activation of STAT3 during BC progression.
We show herein that FN is readily capable of activating STAT3 in the absence of any soluble ligands, a reaction that is markedly increased in metastatic BC cells. Mechanistically, we demonstrate that EMT leads to pathway switching of STAT3 activation, such that nonmetastatic BC cells employ a SRC-dependent EGF:EGFR axis to stimulate STAT3, whereas their metastatic counterparts switch to a FN:␤1 integrin:FAK/PYK2: JAK2 axis to stimulate STAT3. Importantly, interdiction of FN:STAT3 signaling, but not that of EGFR:STAT3, readily prevents the three-dimensional outgrowth of metastatic BC cells in a pulmonary organotypic culture system. In summary, our data suggest a mechanism by which metastatic BC cells exploit EMT to circumvent growth factor signaling, and directly engage the ECM to activate critical downstream signaling pathways.
Cell Harvesting and Immunoblotting-Cells were washed 3 times with PBS supplemented with 5 mM EDTA, 5 mM EGTA, and a phosphatase inhibitor mixture (10 mM NaF, 10 mM Na 4 P 2 O 7 , and 1 mM Na 3 VO 4 ). Cells were lysed with 1% Nonidet P-40 in a solution of 50 mM Tris-HCl, pH 7.4, supplemented with 150 mM NaCl, the phosphatase inhibitor mixture, and a protease inhibitor mixture (100 M phenylmethylsulfonyl fluoride, 10 g/ml of aprotinin, 10 g/ml of leupeptin, 4 g/ml of pepstatin). Cell lysates were clarified by high-speed centrifuga-tion and protein concentrations were determined by Bradford assay (Bio-Rad). Equal aliquots of total cellular protein were resolved by SDS-PAGE and transferred to nitrocellulose filters for immunoblotting using standard methods. Antibody concentrations and suppliers are as follows: Cell Signaling Technologies, EGFR (Tyr-845) Cell Morphometry Measurements-Cells were permeabilized with 0.5% ␤-escin in a solution of 80 mM PIPES, pH 6.8, supplemented with 5 mM EGTA and 1 mM MgCl 2 for 5 min and fixed with 3% paraformaldehyde/PBS for 15 min as described previously (27). Cells were stained with rhodamine-conjugated phalloidin and confocal images were acquired with a Zeiss LSM 510 Meta laser scanning microscope (Carl Zeiss MicroImaging, Jenna, Germany) using a ϫ40 Plan Apo NA 1.4 objective and Zeiss LSM software (Carl Zeiss MicroImaging, Jenna, Germany). Filopodia were counted using MetaMorph software (Molecular Devices, Sunnydale, CA). Briefly, individual cells were filtered by determining threshold values for average pixel intensity. The number of filopodia per cell was determined by counting only filopodia with fluorescent intensity above the average intensity following thresholding that crossed the cell edge and were Ն2 m. Cell shapes were determined by computing shape factor parameters, which measure cell area as a function of cell perimeter (equation 4A/P 2 , where A is the cell area and P is the perimeter) as described previously (5,28). This value varies from 0 to 1 for elongated to more rounded shapes, respectively (29).
Cell Biological Assays-For cell adhesion experiments, cells grown to 80% confluence were serum-starved for 5 h in media supplemented with 0.5% bovine serum albumin (BSA). NMuMG cell populations were serum-starved in DMEM that was also supplemented with insulin (10 g/ml) and if applicable TGF-␤1 (5 ng/ml). Cells were detached from tissue culture plastic with 0.25% trypsin/EDTA, which was inactivated with a 2-fold volume of serum-free media supplemented with soybean trypsin inhibitor (0.5 mg/ml; Invitrogen). Cells were allowed to adhere to polystyrene dishes or glass coverslips coated with ECM proteins (10 g/ml) at a density of 4 ϫ 10 4 cells/mm surface area. Control cells were kept in suspension in polystyrene dishes coated with BSA (10 mg/ml). DNA synthesis was measured by [ 3 H]thymidine incorporation as previously described (26). Cell fractionation was performed using a Nuclear/ Cytosol Fractionation Kit (Biovision, Milpitas, CA) according to the manufacturers' instructions.
Three-dimensional Organotypic Growth Assays-Ninety-sixwell plates were coated with Cultrex (50 l/well; Trevigen Inc., Gaithersburg, MD) and cells were resuspended in DMEM supplemented with 10% FBS and 4% Cultrex (150 l/well). To assess FN-specific growth effects, 96-well plates were similarly coated with Cultrex or a 2:1 mixture of Cultrex:FN using a 1 mg/ml of FN stock. Luciferase expressing MDA-MB-231 or NMuMG-EGFR cells were resuspended in DMEM supplemented with 2% FBS and 2% Cultrex, or with a 2% solution of a 1:3 Cultrex/FN mixture. Cells were seeded at a density of 1 ϫ 10 3 cells/well. Media was replaced every 4 days and organoid outgrowth was detected by adding D-luciferin potassium salt (Caliper Life Sciences, Hopkinton, MA) to induce bioluminescence, which was quantified using a GloMax-Multi detection system (Promega, Madison, WI). Longitudinal cell growth was normalized to an initial reading taken 30 min after seeding as a baseline. Organotypic cultures were also examined by phasecontrast microscopy to assess their morphology.
Tumor Growth-NMuMG cell lines were resuspended in sterile PBS supplemented with 5% Matrigel (2 ϫ 10 6 cells/50 l) and subsequently injected directly into the nipple of 6-week-old female nu/nu mice (Charles River, Wilmington, MA) to allow seeding within the mammary ducts. Tumor growth was monitored by digital caliper measurements at the indicated time points using the following equation: volume ϭ (length 2 ) ϫ (width) ϫ (0.5).
In Silico Analyses-The Cancer Cell Line Encyclopedia contains a repository of log 2 expression data derived from Affymetrix U133A ϩ 2.0 Arrays for 947 unique human cancer cell lines. Human BC cell lines were annotated based on literature search for their basal versus luminal BC status (30,31). Expression data for FN was extracted for each cell line using a robust microarray algorithm and reconverted from a log 2 to a linear scale as described in Ref. 32.
GEO Dataset GSE36953 contains expression data using the Affymetrix U133A ϩ 2.0 for MDA-MB-231 cells under various culture conditions. The dataset contained MAS5.0 normalized expression data, which was used to determine fold-changes between groups. Fold-change in transcript expression was determined by comparing the levels observed in MDA-MB-231 tumors versus those measured in their respective two-dimensional cultured counterparts.
Kaplan-Meier Plots-The Kaplan-Meier plot is an on-line biomarker validation tool that compares the proportional survival of patient groups based on relative biomarker expression using microarray data. This tool was employed to estimate survival probabilities for BC patients split into two groups based on FN gene expression. This analysis was carried out by extracting microarray data for 2878 BC patients and overall survival data for 1027 patients from a database described in Ref. 33 using the sole_at probe (214702_at).
Statistical Analyses-Statistical analyses were carried out using an unpaired Student's t test where p values Ͻ 0.05 were considered statistically significant.

RESULTS
FN Activates an EGFR:STAT3 Signaling Axis-We previously established that FN and EGFR form a signaling complex coupled to the activation of the RhoA antagonist p190RhoGAP in newly adherent cells (5). However, the extent to which FN:EGFR cross-talk induces other signaling pathways remains unexplored. To address this question, we engineered NR6 cells, which lack endogenous EGFR expression to stably express equivalent levels of either wild-type EGFR (WT-EGFR) or an EGFR mutant with a dialanine substitution for Leu-679 and Leu-680 (EGFR-AA) (5,34,35). Importantly, the EGFR-AA mutant is defective for Tyr-845 phosphorylation by SRC, an event that is necessary for STAT3 activation (36). Accordingly, EGF stimulation of WT-EGFR expressing cells elicited robust phosphorylation of EGFR at Tyr-845 and a STAT3 reaction that failed to occur in EGFR-AA expressing cells (Fig. 1A). Importantly, adhesion to FN induced the phosphorylation of Tyr-845 in cells expressing WT-EGFR, but not in those expressing EGFR-AA (Fig. 1B). Furthermore FN-mediated Tyr-845 phosphorylation of EGFR was dependent on the kinase activities of EGFR and SRC and was not triggered by the alternative ECM protein laminin (Fig. 1C). Along these lines, EGFR expression was specifically required for FN-induced activation of STAT3, which contrasts with that of Erk1/2 that was similarly stimulated in control and EGFR expressing cells (Fig. 1D). Similar to ligand-mediated activation of STAT3, FN-mediated STAT3 signaling was sensitive to EGFR and SRC kinase inhibitors (Fig. 1E). Consistent with its activation and role as a transcription factor, phosphorylated STAT3 underwent nuclear translocation in response to FN exposure, opening the possibility of unique FN-induced STAT3 gene expression profiles (Fig.  1F). In addition to its nuclear function, cytoplasmic STAT3 also plays a role in regulation of cytoskeleton (37). Accordingly, FNinduced formation of filopodial membrane protrusions was potently blocked by treating NR6-EGFR cells with Stattic, an inhibitor of STAT3 dimerization (38) (Fig. 1, G and H). Collectively, these findings show that FN is capable of activating the STAT3 signaling system independent of ligand stimulation.
STAT3 Activation Is Critical for EGFR-mediated Transformation of Mammary Epithelial Cells-We recently established that NMuMG cells are transformed by overexpression of EGFR (hereafter referred as NME cells) (7). Interestingly, we show here that overexpression of EGFR-AA failed to transform NMuMG cells (hereafter referred to as NME-AA cells; Fig. 2, A-D). Consistent with these in vivo findings, parental and NME-AA cells formed small differentiated acinar structures when propagated in three-dimensional organotypic cultures (Fig. 2E). In stark contrast, NME cells produced much larger filled organoids that are characteristic of transformed cells (39) (Fig. 2E). This aberrant growth in three-dimensional culture was normalized by administration of the EGFR kinase inhibitors AG1478 and Erlotinib, as well as by that of two different STAT3 inhibitors, (i) Inhibitor VII, which targets STAT3 pathway members, and (ii) Stattic, which blocks STAT3 dimerization and activation. In contrast, the aberrant growth of the NME cells was not affected by WP1006 (Fig. 2F), which inhibits JAK2-mediated activation of STAT3.
Unlike NR6 fibroblasts (Fig. 1D), the parental NMuMG cells express endogenous EGFR and readily activate Erk1/2 and Akt in response to EGF (Fig. 2, G and H) (13). However, only NME cells displayed robust STAT3 phosphorylation when stimulated with EGF (Fig. 2, G and H). Moreover, EGF-stimulated STAT3 activation was dependent on SRC kinase, whereas Erk1/2 activation was not (Fig. 2H). Collectively, these data establish that STAT3 is aberrantly activated downstream of Src-dependent EGFR signaling, and suggest that this pathway likely plays a critical role in the transformed phenotype of mammary epithelial cells that overexpress this growth factor receptor.
Autocrine Expression of FN Is Associated with Aberrant STAT3 Phosphorylation-To further examine the potential role of STAT3 in BC progression, we analyzed the HRASdriven human MCF10A BC progression series, consisting of normal (10A), indolent (T24-HRAS-transformed T1K), and malignant (Ca1h) cell lines (40). Fig. 3A shows that malignant Calh cells acquire a mesenchymal phenotype characterized by down-regulation of E-cadherin and up-regulation of FN and ␤1 integrin as compared with their nonmalignant counterparts. Furthermore, autocrine FN expression in Ca1h cells was highly correlated with constitutive STAT3 activation by an EGFR-in-dependent mechanism (Fig. 3, A and B). These findings suggest that EMT-mediated up-regulation of FN and ␤1 integrin may facilitate the activation of STAT3 in more aggressive BC.
EMT Selectively Drives a Pathway Switch Upstream of STAT3 Activation-To further examine the role of EMT in EGFR: STAT3 and FN:STAT3 signaling we utilized TGF-␤1 to induce a mesenchymal state in the NME cells, a measure we previously observed to be capable of increasing the invasion and metastasis of these cells (7). IL-6 stimulation of STAT3 was unaffected by induction of EMT (Fig. 4A). In contrast, short (6 h, 48 h) or long (4 weeks) term exposure of NME cells to TGF-␤1 prior to EGF stimulation led to a dramatic reduction in the ability of these cells to activate STAT3 via this pathway (Fig. 4, B-D). Furthermore, ex vivo NME cells harvested from "post-EMT" tumors selectively displayed reduced EGF-mediated STAT3 signaling as compared with their pre-EMT counterparts despite the fact that ligand-induced Erk1/2 activation was FIGURE 1. EGFR-dependent STAT3 signaling in a heterologous reconstitution system. A, EGFR-null NR6 cells reconstituted with wild-type human EGFR (NR6-EGFR) or EGFR with a 679-LL dialanine substitution (NR6-EGFR-AA) were harvested under basal conditions (Ϫ) or following a 30-min EGF stimulation (ϩ). Equal protein aliquots were immunoblotted with phospho-specific and total EGFR and STAT3 antibodies. B, NR6-EGFR or NR6-EGFR-AA cells were left in suspension (Ϫ) or adhered to FN for 20 min (ϩ). Equal protein aliquots were immunoblotted with phospho-specific and total EGFR antibodies. C, NR6-EGFR cells pretreated with vehicle, EGFR (1 M AG1478), or SRC (10 M PP2) kinase-specific inhibitors and subsequently adhered to FN as in panel A or laminin. Equal protein aliquots were immunoblotted with phospho-specific and total EGFR antibodies. D, parental and EGFR-expressing NR6 cells were adhered to FN for the indicated amounts of time and equal protein aliquots were immunoblotted with phospho-specific (p) and total (t) STAT3 and Erk1/2 antibodies. E, NR6-EGFR cells were pretreated with the indicated kinase inhibitors and subsequently adhered to FN. Equal protein aliquots were analyzed by immunoblotting with phospho-specific and total protein antibodies for STAT3. F, NR6-EGFR cells were kept in suspension or adhered to FN for the indicated amounts. Cells were subjected to cell fractionation prior to immunoblotting with phospho-specific and total STAT3 antibodies. Cytosolic (Cyt) and nuclear (Nu) fractions were confirmed by immunoblot for GAPDH and Lamin B1, respectively. G, representative confocal images of NR6-EGFR cells adhered to FN for 20 min in the presence of vehicle or the STAT3 inhibitor Stattic (5 M) and then stained with phalloidin to visualize the actin cytoskeleton. H, data for NR6-EGFR cells in G quantified as described under "Experimental Procedures." Data are the mean number of filopodia per cell Ն2 mm in length/cell Ϯ S.E. maintained in both cell populations (Fig. 4E). In sharp contrast, TGF-␤1-induced EMT potently enhanced the ability of FN to activate STAT3 (Fig. 5A). Importantly, EGFR and Src inhibitors that abrogated FN:STAT3 signaling in pre-EMT cells had essentially no effect on FN:STAT3 signaling in their post-EMT counterparts (Fig. 5B). Similar to Fig. 4E Erk1/2 by FN was unaffected by the EMT status of the cells (Fig.  5A), and was similarly impaired by EGFR kinase inhibition in both cell populations (Fig. 5B). Concomitant with this enhanced response to FN, STAT3 phosphorylation switched from a pathway regulated by Src and EGFR (Fig. 5B) to one dependent upon JAK2 (Fig. 5C). Importantly, this pathway switching sensitized post-EMT NME cells to pharmacologic inhibition of JAK2 (Fig. 5, D and E). Taken together, these data strongly suggest that EMT programs mediate a pathway switch in STAT3 activation away from EGFR-dependent signaling in pre-EMT cells to an EGFR-independent pathway regulated by FN and JAK2 during EMT and the metastatic progression of BCs.
FN Facilitates STAT3 Signaling in Metastatic BC Cells Independent of EGFR-Thus far we have shown that FN is capable of activating STAT3 independent of EGFR following induction of EMT (Fig. 5, A and B). Therefore, we hypothesized that FN adhesion maintains STAT3 signaling independent of growth factor receptor signaling during metastatic outgrowth. This hypothesis was tested in human MDA-MB-231 BC cells, which were originally derived from a pleural effusion of a metastatic BC patient (41). Interestingly, Src-dependent EGFR signaling failed to elicit STAT3 phosphorylation in MDA-MB-231 cells as compared with the MDA-MB-468 cells, in which EGFR is genomically amplified and therefore highly expressed (Fig. 6A) (42,43). In contrast, IL-6 activated STAT3 via a JAK2-dependent manner in the MDA-MB-231 cells (Fig. 6B). Consistent with the lack of EGFR:STAT3 signaling, FN adhesion elicited robust STAT3 phosphorylation in MDA-MB-231 cells (Fig. 6C) by a mechanism that was independent of EGFR and Src kinase activity (Fig. 6D). Treatment of the MDA-MB-231 cells with TGF-␤1 can enhance their pulmonary metastasis (44). Along these lines, treatment of MDA-MB-231 cells with TGF-␤1 led to potent induction of ␤1 integrin and FN expression (Fig. 6E). Interestingly, gene microarray data from in vitro basal-like BC cell lines, including MDA-MB-231 cells, exhibit high levels of FN expression as compared with luminal BC (Fig. 6F). Furthermore, expression of FN is enhanced in MDA-MB-231 primary tumor xenografts versus in vitro cultured cells (Fig. 6G). Thus, enhanced autocrine FN expression by cells at the primary tumor may provide a local oncogenic niche that not only maintains STAT3 signaling independent of other extracellular factors, but also enhances STAT3 signaling in surrounding cells. Accordingly, we found that high FN expression in the primary tumor was linked to poor overall survival in a cohort of 1027 BC patients (Fig. 6H) (33).

FN:STAT3 Signaling Is Regulated by a Focal Adhesion
Kinase-dependent:JAK2 Pathway in MDA-MB-231 Cells-We next sought to elucidate the mechanism whereby FN activates STAT3 independent of EGFR. Integrins are transmembrane receptors that sense the ECM and are linked to intracellular signaling modules via the focal adhesion complex (45). Indeed, we previously established that several integrins and focal adhesion complex proteins are up-regulated during TGF-␤-induced EMT (6,22,26). These findings raise the possibility that an integrin-mediated axis could facilitate FN:STAT3 signaling during EMT and metastasis. Accordingly, adhesion to FN elicited robust activation of two related focal adhesion kinases, FAK and PYK2, as well as that of JAK2 (Fig. 7A). FN-induced STAT3 activation was blocked by treating cells with small molecule inhibitor PF-573228 (PF228) that specifically targets FAK (Fig. 7B). Use of a related compound, PF-562271 (PF271), which targets both FAK and PYK2, was more effective at lower concentrations than PF228, suggesting a preferential role for PYK2 in mediating FN:STAT3 signaling in the MDA-MB-231 metastatic BC cell line (Fig. 7B). Moreover, FN-mediated JAK2 phosphorylation was also dependent on FAK/PYK2, but independent of EGFR kinase activity (Fig. 7C). Similar to post-EMT NME cells (Fig. 5E), FN-induced STAT3 activation was also blocked by the JAK2 inhibitor WP1066 in MDA-MB-231 cells (Fig. 7D).
Consistent with a potential role for acute STAT3 activation in modulating the cytoskeleton (Fig. 1), administration of JAK2 (WP1066) or STAT3 (Stattic) inhibitors prevented MDA-MB-231 cells from acquiring an elongated mesenchymal morphology when cultured on FN (Fig. 7, E and F). In contrast, targeting EGFR kinase activity under these same conditions had little to no effect on cell morphology (Fig. 7, E and F). Collectively, these findings have delineated a novel FN:FAK/PYK2:JAK2 signaling axis that drives the sustained activation of STAT3 in metastatic BC cells.
The FN:STAT3 Signaling Axis Regulates Three-dimensional Outgrowth of Metastatic BC Cells-We (13,22) and others have previously established the importance of cytoskeletal dynamics (46) and ␤1 integrin (2,47) in regulating the metastatic outgrowth of BC. Therefore, we next sought to identify the impact of inactivating FN:STAT3 signaling in preventing three-dimensional outgrowth. Using a three-dimensional organotypic culture system to recapitulate the compliance of the pulmonary microenvironment, we found that the outgrowth of MDA-MB-231 cells was significantly enhanced by FN supplementation (Fig. 8, A and B). Moreover, supplemental FN enabled MDA- MB-231 cells to transition from branched morphologies to celldense three-dimensional structures (Fig. 8A), a phenotype that we previously linked to enhanced metastatic capacities of BC cells (7,13). Underscoring the importance of ␤1 integrin in mediating these events, we found that administration of neu-tralizing ␤1 integrin antibodies prevented FN-induced threedimensional outgrowth as compared with an isotype control antibody (Fig. 8C). In contrast, MDA-MB-231 cells were resistant to EGFR kinase inhibition in the presence of supplemental FN (Fig. 8, A and D). Pharmacological inhibition of STAT3 using two mechanistically distinct compounds (STAT3 inhibitor VII and Stattic) completely abrogated three-dimensional outgrowth of MDA-MB-231 cells (Fig. 8, A and D). In contrast to NME cells, which are nonmetastatic (Fig. 4H), outgrowth of MDA-MB-231 cells was also blocked by administration of WP1066 to inhibit JAK2 (Fig. 8, A and D). Collectively, these data establish a role for a FN:STAT3 signaling module in driving the three-dimensional outgrowth of metastatic BC. Furthermore, our data demonstrate a role for ␤1 integrin in a complementary STAT3 pathway that employs FN linked to a JAK2: STAT3 signaling network in place of EGFR in late-stage BC.

DISCUSSION
BC can be divided into several genetically distinct subgroups. Clinically, those BCs belonging to the "triple negative" (TNBC)  subtype comprise ϳ10 -20% of all BCs and are unique in their metastatic aggressiveness, increased rate of relapse, and poor overall prognosis (30,48). Unfortunately, TNBC remains a diagnosis of exclusion defined only by their lack of estrogen receptor-␣ and progesterone receptor, and by their absence of human epidermal growth factor receptor 2 (HER2) amplification (48). As such, effective targeted molecular therapies against TNBC do not exist. Interestingly, TNBCs often exhibit elevated levels of EGFR expression, which is associated with decreased overall survival (9,49). These findings suggest that administration of EGFR targeted therapies would alleviate TNBC disease progression, a supposition that has not been born out in clinical trials (50). Unlike other carcinomas where EGFR inhibitors are initially highly effective and resistance is "acquired" through tumor evolution, TNBCs are inherently resistant to EGFR inhibition. The mechanisms that drive this disconnect between the diagnostic and therapeutic efficacy of EGFR in BC remained poorly defined.
Our studies demonstrate the ability of FN to activate STAT3, thus contributing to aberrant cell growth within a physiologically relevant tumor microenvironment. These finding are supported by other recent studies demonstrating the ability of the ECM and cytoskeletal dynamics to directly initiate critical signaling pathways (51) and metastatic outgrowth (46). Furthermore, FN is a secreted matrix protein whose production is potently increased upon the induction of EMT. Therefore, our data establish a mechanism by which EMT may act in trans to activate STAT3 and influence the growth and/or metastasis of surrounding tumor cells. Moreover, disseminated tumor cells that have undergone EMT have the ability to establish a meta- static niche through EMT-induced autocrine FN production that supports STAT3 activation and metastatic outgrowth. Indeed, using a BC model in which transformation is initiated by EGFR overexpression, we show that EGF-mediated STAT3 signaling is actually diminished following the induction of EMT. Importantly, loss of growth factor responsiveness is associated with a concomitant increase in the ability of FN to stimulate this critical oncogenic pathway. Mechanistically, our data suggest that BCs switch away from SRC:EGFR-dependent STAT3 signaling to a compensatory pathway whereby ␤1 integrin receptors signal to PYK2/FAK during metastatic progression.
Of course the new FN:JAK2:STAT3 signaling pathway that we have elucidated in this study is only one facet of the unique molecular phenotype of metastatic BC cells that is acquired through both genomic and nongenomic means. Other potentially critical factors that may be relevant in TNBC include the interplay between EGFR and insulin-like growth factor recep-tor signaling systems and production of cell surface effectors such as the subclass of heparin sulfate proteoglycans known as syndecans (52)(53)(54)(55)(56). Understanding how these different factors are integrated to drive metastatic progression will provide the basis for improved prognostic screening, guide individualized therapeutic choice, and further illuminate clinical resistance to EGFR-targeted therapies in TNBC patients.
Somewhat surprisingly, our data also appear to exclude a role for autocrine IL-6 because EMT programs fail to diminish the ability of NME cells to activate STAT3 in response to exogenous IL-6. Other investigators have shown that PYK2 forms a specific molecular complex with JAK2 responsible for activating selective JAK2 signaling responses (57,58). Thus, it is reasonable to assume that FN-activated PYK2/FAK may fulfill a similar role in a JAK2:STAT3 signaling network, which is governed in an EMT-dependent manner. Underscoring the importance of this switch in upstream STAT3 signaling, we show that inhibition of JAK2 had no effect on BC organoid growth prior to induction of EMT. However, once these same cells have under-gone EMT pharmacologic inhibition, JAK2 prevents BC outgrowth. These findings are quite remarkable as several recent studies suggest that the process of EMT is strongly linked to the acquisition of a stem-like and drug-resistant state (59). Indeed, our data clearly shows that BC cells acquire the ability to engage the matrix to facilitate cell signaling in the face of EGFR inhibition following EMT. Therefore, our JAK2 findings are quite unique in that they represent the first known example whereby EMT actually sensitizes BC cells to a targeted chemotherapy. These findings support the notion that JAK2 inhibition may be a viable option for treatment of late stage BC. Accordingly, inhibition of JAK2 is effective in abrogating three-dimensional growth of the human metastatic MDA-MB-231 cells that are resistant to inhibition of EGFR. Interestingly there are already phase II clinical trials addressing the effectiveness of JAK2 inhi- bition in BC (60). Human MDA-MB-231 cells harbor activating Ras and BRAF mutations that are associated with constitutive MEK-Erk1/2 activation, which could account for EGFR-independent metastatic outgrowth (61). However, the effectiveness of ␤1 integrin, JAK2, and STAT3 inhibition in decreasing the outgrowth potential of these cells strongly implicates STAT3 as a critical signaling node mediating tumor outgrowth in FN-rich microenvironments even in cells harboring an alternative oncogenic pathway.
Collectively, our findings establish a novel form of EMTmediated "pathway switching" that engages the tumor microenvironment to facilitate activation of critical signaling pathways. Our data illustrate a switch away from SRC-dependent EGFR:STAT3 signaling downstream of ligand stimulation or FN engagement as tumor cells undergo EMT. This is accompanied by a concomitant activation of an alternative pathway whereby FN activates ␤1 integrin receptors to elicit a FAK/ PYK2:JAK2:STAT3 signaling pathway (Fig. 9). In addition to supporting the use of JAK2 inhibitors to specifically target post-EMT and metastatic BCs, our findings also indicate that development of novel strategies to prevent the switch in upstream STAT3 signaling may represent an important first step in resensitizing late-stage BCs to EGFR inhibitor therapies.