EMT Transition Alters Interstitial Fluid Flow–Induced Signaling in ERBB2-Positive Breast Cancer Cells

Breast cancer is the most commonly reported form of cancer in women worldwide (excluding non-melanoma skin cancers) and is responsible for 13.7% of cancer-related deaths in women (1). The mortality rates associated with this and most other solid cancers is a consequence of cancer cell invasion into the surrounding stroma and metastasis to distant organs. Although many factors responsible for invasion and metastasis are still unknown, the influence of the breast tumor microenvironment (TME), which consists of stromal cells, extracellular matrix, and soluble factors, on cancer progression is well established (2).

While their importance in cancer progression has only been realized relatively recently, various biophysical factors are significantly altered in the TME. Changes in matrix density (3), stiffness (4), organization (5), interstitial fluid pressure (6), and flow (7) are all biophysical consequences of tumor growth that affect gene expression, proliferation, differentiation, and invasion. Interstitial fluid flow (IFF), the movement of fluid through the tissue matrix, is elevated in tumors compared with normal tissue (8). This elevated IFF is driven by steep fluid pressure gradients at the tumor margin (9). Measured levels of IFF range from 0.1 to 1 μm/s in normal tissue, and 0.1 to 55 μm/s in tumor tissue (10–13). Increased IFF is directly linked to lymph node metastasis in human cervical carcinoma, and increased cell motility and invasion in vitro in breast cancer, glioma, melanoma, and renal carcinoma cells (11, 14–17). Although increased IFF appears to drive cancer cell invasion, the underlying mechanisms of IFF mechanotransduction are not fully understood.

To date, different studies have proposed mechanisms to explain the strong tumor and stromal cell migration induced by elevated IFF, though most have focused on highly invasive cells. In fibroblasts, IFF induced myofibroblast differentiation and secretion of matrix metalloproteinases (MMP), thus indirectly aiding tumor cell migration through matrix remodeling (16, 18). In metastatic cells, IFF altered extracellular chemokine gradients to accelerate tumor cell invasion via CXCR4- and CCR7-dependent chemotaxis (14, 15), and increased MMP activity via shear stress sensing through the glycocalyx (17). More recently, IFF was shown to induce invasion against the direction of flow due to fluid drag forces and activation of integrins on the upstream side of the cell (19). Although we know some of its effects on invasive cells, the specific pathways triggered by IFF are still unclear, especially those that play a crucial role in the early stages of invasion.

At these early stages of invasion, the process of epithelial-to-mesenchymal transition (EMT) can play an important role. During EMT, epithelial cells lose their distinctive morphology and molecular characteristics and acquire a mesenchymal phenotype that is often associated with invasion. This phenomenon, although indispensable during embryogenesis and tissue remodeling, also occurs during tumor development and likely plays a role in tumor progression (20). EMT is characterized by the loss of epithelial basoapical polarity and cell–cell junctions, spindle-like cellular morphology, and activation of transcription factors that lead to increased invasion (21, 22). The induction of EMT during tumor development is a consequence of external cues in the TME, like changes in levels of cytokines and growth factors (22, 23). To date, the relationship between EMT and how cancer cells respond to IFF has not been investigated.

Thus, the overall goal of this study was to elucidate the molecular pathways activated by IFF in invasive and noninvasive breast cancer cells and determine whether EMT alters IFF signaling and response. Here, we show for the first time that IFF causes an increase in tumor cell invasion via activation of phosphoinositide-3-kinase (PI3K) in ERBB2-positive breast cancer cells. Furthermore, we demonstrate that pre- and post-EMT cancer cells respond to IFF via different PI3K-dependent pathways.

MCF10A and MCF10A-ERBB2 (NeuN/NeuT; ref. 24) were maintained in DMEM/F12 supplemented with 5% donor horse serum (Atlanta Biologicals), 20 ng/mL epidermal growth factor (EGF; Peprotech), 10 μg/mL insulin (Sigma-Aldrich), 100 ng/mL cholera toxin (Enzo Life Sciences), 500 ng/mL hydrocortisone (Sigma-Aldrich), and 1% penicillin/streptomycin (Mediatech). SKBR3, a breast cancer cell line that naturally expresses ERBB2, were grown in McCoy's 5A (Mediatech) supplemented with 10% FBS and 1% penicillin/streptomycin. BT474 and MDA-MB-453, other breast cancer cell lines that naturally express ERBB2, were grown in DMEM (Mediatech) supplemented with 10% FBS and 1% penicillin/streptomycin. All cells were maintained in a humidified environment at 37°C and 5% CO₂.

IFF was applied to cells as previously described (14, 16, 25), in serum-free conditions (to minimize the effect of serum factors on the signaling pathways studied). Briefly, cells were embedded in a matrix composed of 1.3 mg/mL rat tail tendon collagen type I (BD Biosciences) and 1 mg/mL Matrigel basement membrane matrix (BD Biosciences) at a final concentration of 5 × 10⁵ cells/mL inside culture inserts with 8-μm diameter pores (Millipore; Supplementary Fig. S1). For static (control) conditions, serum-free media levels inside and outside the insert were kept approximately equal, resulting in a minimal hydrostatic pressure difference across the gel and no measurable interstitial flow. For IFF conditions, serum-free media were added under the insert (100 μL) and above the gel (650 μL). The hydrostatic pressure difference generated was approximately 1 mm Hg. After 24 hours of either physiologic interstitial flow (approximate flow velocity, ∼0.1 μm/s) or static conditions, invasion was measured by counting the number of cells that invaded through the matrix and migrated across the porous membrane. Migrated cells on the underside of the culture insert membranes were fixed in 4% paraformaldehyde in PBS for 30 minutes and permeabilized with 0.5% Triton X-100 in PBS for 10 minutes. Cells were then stained with Alexa Fluor 488–conjugated phalloidin (6 U/mL; Life Technologies) and 4′,6-diamidino-2-phenylindole (DAPI; 2 μg/mL; MP Biomedicals). Labeled cells were visualized on an epifluorescence microscope (Leica Microsystems). DAPI-stained nuclei at five randomly selected locations of each membrane were counted. F-actin staining was used to confirm that positive DAPI stain corresponded to cells. Percentage invasion was calculated from the following equation:

In some instances, percentage invasion was normalized to the static control condition to allow for comparison between independent experiments. To determine the functional role of specific proteins, pharmacologic inhibitors were used to block specific signaling proteins (see Table 1 for a list of inhibitors and corresponding concentrations employed in this study). All experimental concentrations were confirmed to be noncytotoxic and effective on our cell lines (Supplementary Fig. S6). When used, inhibitors were added to the collagen matrix and experimental medium at the appropriate concentrations. Experiments were repeated at least twice with a minimum sample size of n = 3 for each experiment (total minimum sample size n = 6).

EMT was induced in NeuN cells via treatment with recombinant human TGFβ1 (Peprotech). An optimal concentration of TGFβ1 was first determined by treating NeuN cells at varying concentrations of TGFβ1 (3, 5, 10, 20, and 50 ng/mL) for 6 days. A concentration of 20 ng/mL was found to most effectively induce EMT (low E-cadherin expression, high vimentin expression, elongated morphology). NeuN cells were treated with this concentration in full media for 6 days, with a media change after 3 days.

Following the three-dimensional (3D) invasion assay, total and phospho-protein levels were determined by Western blot analysis. Cells were isolated from the matrix using 2.5 mg/mL collagenase D (Roche) for 30 minutes. The resulting solution was centrifuged, and the pellet was washed with PBS and resuspended in RIPA lysis buffer (150 mmol/L NaCl, 1% NP40, 0.5% DOC, 50 mmol/L Tris–HCl at pH 8, 0.1% SDS, 10% glycerol, 5 mmol/L EDTA, 20 mmol/L NaF, and 1 mmol/L Na₃VO₄). Lysates were cleared via centrifugation at 16,000 x g for 20 minutes at 4°C. Antibodies used were: rabbit polyclonal anti-phospho PI3K and total PI3K (1:500; Cell Signaling Technology), rabbit polyclonal anti-β-actin (1:3,000; Cell Signaling Technology), rabbit monoclonal anti-vimentin (1:1,000; Abcam), rabbit monoclonal anti-E-cadherin (1:1,000; Abcam), rabbit polyclonal anti-phospho and total CXCR4 (1:500; Abcam), rabbit polyclonal anti-p110α and β (1:1,000; Cell Signaling Technology), and goat anti-rabbit HRP (1:10,000; Abcam).

For visualization of EMT markers, cells were plated onto 8-well chamber slides at 10,000 cells per well. After 16 hours, cells were fixed in 4% paraformaldehyde in PBS for 30 minutes and permeabilized with 0.5% Triton X-100 in PBS for 10 minutes. This was followed by 1 hour in blocking solution (1% BSA, 0.1% goat serum). Cells were stained with rabbit monoclonal anti-vimentin (1:100; Abcam) for 16 hours, then washed in PBS three times. Alexa Fluor 488–conjugated phalloidin (6 U/mL), DAPI (2 μg/mL), and Alexa Fluor 555–conjugated anti-rabbit IgG was incubated on the cells for 16 hours, then washed in PBS three times. Fluorescence was visualized on an epifluorescence microscope (Leica Microsystems).

Changes in gene expression in NeuN and NeuT cells were identified by microarray analysis. Total RNA was extracted from cells embedded in 3D gels (6 biological replicates per sample) using the RNeasy Mini Kit (Qiagen). RNA was submitted to the Wistar Institute genomics core facility (Philadelphia, Pennsylvania) for amplification. Epicentre TargetAmp Nano Labeling Kit for Illumina Expression BeadChip was used at 100-ng RNA and hybridized on an Illumina HT12v4 chip. Data were analyzed with PARTEK Genomics Suite and Database for Annotation, Visualization and Integrated Discovery (DAVID) online tools.

Quantitative RT-PCR (qRT-PCR) was performed on cDNA reversed transcribed using QuantiTech Reverse Transcription kit and QuantiTech SYBR Green PCR kit (Qiagen) on three of the most altered genes within the microarray data to validate the results. Primers for human interleukin 6, desmoplakin, and integrin subunit α6 (Quantitech; Qiagen) were used to validate the microarray analysis and quantify mRNA expression differences.

Data are expressed as mean ± standard error of the mean. Differences among conditions were tested using the Student t tests (for two groups) or two-way analysis of variance (ANOVA; for three or more groups) using GraphPad Prism. When ANOVA identified a significant difference, a Bonferroni post-test was used for multiple comparisons. Differences were accepted as significant at P < 0.05.

To model noninvasive and invasive cells, we used a previously developed model of breast cancer based on MCF10A human mammary epithelial cells engineered to overexpress ERBB2 (24): MCF10A cells retrovirally transduced with either wild-type ERBB2 (NeuN) or a constitutively active mutant of ERBB2 (NeuT). When cultured in 3D conditions, NeuN cells exhibit behavior of preinvasive ERBB2-positive ductal carcinoma in situ (DCIS; ref. 24), while NeuT cells behave like ERBB2-positive invasive ductal carcinoma (IDC; ref. 26). Using a 3D invasion assay where single cells are embedded in stroma-like matrix, we measured invasion by counting the number of transmigrated cells (Supplementary Fig. S1). We observed an increase in invasion of more than 2-fold in response to 0.1 μm/s IFF in both NeuN and NeuT cells (Fig. 1A).

To determine the signaling pathways involved in the observed IFF-induced invasion, we examined pathways known to be important in invasion of HER2-positive breast cancer cells (HER2, MAPK, and PI3K). We isolated cells from gels after the 3D invasion assay and harvested protein from the resulting cells. Western blot analysis revealed an increase in phosphorylated p85 (the regulatory subunit of class I PI3Ks) in both NeuN and NeuT cells treated with IFF (Fig. 1B). To confirm that our results were not specific to the engineered MCF10A cell lines, exposure of other breast cancer cells overexpressing ERBB2, including SKBR3, BT474, and MDA-MB-453, to IFF also increased invasion and PI3K phosphorylation compared with static conditions (Fig. 2A–C).

To test whether PI3K activity was required for IFF-induced invasion, we treated cells with the PI3K inhibitor LY294002 (a pan-PI3K inhibitor). Treatment with LY294002 decreased IFF-induced invasion in NeuN, NeuT (Fig. 1D), SKBR3, and MDA-MB-453 cells (Fig. 2D). These data suggest that IFF increases invasion in ERBB2-overexpressing breast cancer cells via activation and subsequent downstream signaling through PI3K.

Class I PI3Ks are present in cells as heterodimers composed of a regulatory (typically p85 or p55) and catalytic subunit (p110; ref. 27). The p110 catalytic subunit is responsible for propagating downstream signaling through its kinase activity (28). Using inhibitors that specifically target the β (TGX-221) or α (PIK-75) p110 catalytic subunit of PI3K, we observed that in NeuN cells, only the p110α inhibitor blocked IFF-mediated invasion (Fig. 3A). However, in NeuT cells, inhibiting both p110α and p110β decreased IFF-induced invasion (Fig. 3B). NeuN and NeuT cells expressed similar protein levels of p110α and p110β (Fig. 3C), thus ruling out altered levels of these isoforms as the reason for this difference.

Because previous studies have linked p110α to receptor tyrosine kinase signaling and p110β to G protein–coupled receptors (GPCR; refs. 29–31), we investigated the role of chemokine receptors (a family of GPCRs) in the IFF-induced invasion of NeuT cells. CXCR4 is a known modulator of breast cancer invasion (32, 33) and IFF-induced glioma cell invasion was previously shown to be dependent on CXCR4 (15). Treating NeuT cells with antagonists of the chemokine receptor CXCR4 (AMD3100 or WZ811) significantly reduced IFF-induced invasion (Fig. 4A). In contrast, AMD3100 and pertussis toxin (Gα_i subunit inhibitor of all GPCRs; ref. 34) had no effect on IFF-mediated invasion of NeuN cells (Fig. 4B), suggesting that chemokine receptors, including CXCR4, are not involved in the molecular pathway activated by IFF in NeuN cells. CXCR4 inhibitors also had no effect on IFF-mediated invasion in the noninvasive HER2-positive cell line SKBR3 (data not shown).

To test whether CXCR4 signaling was associated with PI3K activation, we examined the effects of treating NeuT and NeuN cells with AMD3100 on p85 phosphorylation during IFF-mediated invasion. IFF-induced PI3K phosphorylation was decreased in NeuTs when the cells were treated with AMD3100 but phosphorylation was only weakly inhibited in NeuN (Fig. 4C). Interestingly, CXCR4 and CXCL12 protein levels were similar in NeuN and NeuT cells (Supplementary Fig. S2). CXCR4 and its ligand CXCL12 have been previously implicated in IFF-mediated glioma invasion by autologous chemotaxis (15). In autologous chemotaxis, a transcellular gradient is generated by the combination of autocrine chemokine secretion and interstitial flow (35). To determine whether a CXCL12 gradient was necessary for IFF-induced invasion of NeuT cells, exogenous CXCL12 was added to the surrounding media and matrix at a uniform concentration of 80 ng/mL (10 nmol/L). At this concentration, the cells' CXCR4 receptors should be approaching saturation (K_D, ∼14 nmol/L; ref. 36), effectively blocking gradient sensing. Under these conditions, NeuT cells no longer responded to IFF (CXCL12 condition) compared with a 2-fold increase in invasion in the control flow condition (Fig. 4D). When NeuN cells were subjected to IFF in the presence of exogenous CXCL12, the uniform chemokine concentration did reduce invasion, but, the cells still exhibited a significant IFF-induced response (Supplementary Fig. S3), in contrast to NeuT cells. These data suggest that NeuT cells, unlike NeuN cells, require a gradient of CXCL12 for IFF-induced activation of PI3K and invasion, consistent with CXCL12-dependent autologous chemotaxis.

Having identified separate IFF-induced pathways in NeuN and NeuT cells, our next goal was to determine some of the fundamental differences between these cells that are responsible for the observed differences in IFF-induced invasion. NeuN cells adopted more epithelial-like shapes on 2D cell culture plates and invaded in clumps of cells similar to their parental MCF10A counterparts (Supplementary Fig. S4). However, NeuT invaded as single cells and were characterized by a spindle-like morphology. Thus, we hypothesized that NeuT cells may have undergone EMT and this may explain the difference in IFF response. We probed for protein expression of epithelial and mesenchymal markers, and observed loss of E-cadherin and increased levels of vimentin in NeuT cells when compared with NeuN cells (Fig. 5A and B). Both of these proteins are known markers of EMT (37). To examine gene differences between NeuN and NeuT cells further, we profiled gene expression using microarray analysis. Table 2 lists a number of known EMT-associated genes that were significantly different when comparing NeuT and NeuN mRNA expression (log₂-transformed fold change is represented). These data have been deposited in NCBI's Gene Expression Omnibus (GEO; ref. 38) and are accessible through GEO Series accession number GSE64670 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE64670). The expressions of all these genes agree with published EMT-associated data (39, 40). We validated the microarray result with qRT-PCR of three of the identified genes (Supplementary Fig. S5). These results supported the hypothesis that NeuT cells display a more mesenchymal-like phenotype when compared with NeuN cells.

To determine whether EMT contributes to the altered IFF response, EMT was induced in NeuN cells using the known EMT-inducing growth factor TGFβ1 (22). The cells were treated for 6 days with 20 ng/mL TGFβ1 and EMT induction was confirmed using a series of protein, morphological, and functional markers. NeuN^EMT cells exhibited a more marked spindle-like morphology consistent with EMT (Supplementary Fig. S7A). NeuN^EMT cells show lower levels of E-cadherin and higher levels of vimentin when compared with their control NeuN counterparts (Fig. 5C and D), as well as increased levels of fibronectin (Supplementary Fig. S7B). These are all well-established EMT markers (22). As expected, the NeuN cells induced to undergo EMT had higher levels of basal invasion (Fig. 6B, Supplementary Fig. S7C). In line with our previous findings in NeuT cells, we observe that IFF-induced invasion in NeuN^EMT cells was inhibited by both pertussis toxin and CXCR4 (Fig. 6A and B). In addition, IFF-induced PI3K phosphorylation in the presence of AMD3100 was reduced (Fig. 6C) in NeuN^EMT cells, similar to NeuT cells (Fig. 4C). Consistent with this result, treatment of NeuN^EMT cells with p110β inhibitor (TGX-221) reduced IFF-mediated invasion, in contrast to regular NeuN (Fig. 6D), but similar to NeuT cells (Fig. 3B). Thus, these data support the hypothesis that ERBB2-expressing breast cancer cells that have undergone EMT invade in response to IFF through a CXCR4- and p110β-dependent mechanism.

In this study, we examined the role of key intracellular signaling pathways that mediate IFF-induced invasion in ERBB2-expressing breast cancer cells. These experiments were performed under serum-free conditions to decrease confounding factors associated with serum components and to reduce basal cellular activities. In all breast cancer cells tested, we found that IFF induced activation of PI3K as measured by p85 phosphorylation. In addition, these breast cancer cells required PI3K activity for IFF-mediated invasion as inhibition of PI3K and the p110α catalytic subunit blocked IFF-mediated invasion. However, cells that have undergone EMT (NeuT and NeuN^EMT cells) required additional signals, including activation of chemokine receptors. Indeed, we showed that inhibitors of CXCR4 and p110β, a p110 isoform that is GPCR-regulated (29–31), specifically blocked IFF-mediated invasion in breast cancer cells that have undergone EMT. Our data suggest a model where, as cells undergo EMT, the signaling pathways activated by IFF to induce invasion change significantly (Fig. 7).

The PI3K pathway plays a central role in numerous cellular processes crucial for cancer progression. This pathway has been implicated in invasion, proliferation, transformation, and cell survival (41, 42). Mutations or alterations of this pathway are the most frequent in all human cancers, making it an important target of cancer therapeutics (41). This work supports the importance of targeting the PI3K pathway to curtail breast cancer invasion, especially in those with prominent IFF. PI3K is present in cells as a heterodimer complex made up of a regulatory and catalytic subunit. Cancer-specific mutations have been observed in the four catalytic p110 isoforms (α, β, δ, and γ) of the class I kinase (42). IFF led to phosphorylation of p85 in all cells tested, suggesting that class I PI3K is central to signals activated by IFF. However, as cancer cells underwent EMT, we found changes in p110 subunit dependency. Only breast cancer cells expressing a mesenchymal phenotype (determined by E-cadherin and vimentin expression) required p110β activity. As mentioned above, p110α has been linked to downstream activity of receptor tyrosine kinases, while p110β has been linked to GPCRs (29–31). This is consistent with our data that EMT cells required the GPCR and chemokine receptor CXCR4 for IFF-induced invasion. Our findings suggest that breast cancer cells undergoing EMT require both PI3K catalytic subunits p110α and p110β for IFF-mediated invasion and that this pathway may be targeted to block invasion of aggressive breast cancers. Because we could decrease IFF-induced invasion in EMT-like cells by specifically blocking either p110α or p110β, drugs targeting these p110 isoforms may be used to block invasion of aggressive breast cancers and this may have less toxic effects compared with using pan-PI3K inhibitors.

We demonstrated that IFF-induced PI3K activation occurs through separate upstream receptors, leading to downstream signaling in a p110 isoform-specific fashion (Fig. 7). IFF-specific activation of PI3K in cells that have undergone EMT (NeuT and NeuN^EMT) occurred via CXCR4. CXCR4 is a known prognosis marker for breast cancer metastasis (32) and has been implicated in breast cancer invasion (43). The role CXCR4 plays in IFF-induced glioma invasion was previously described as a result of autologous chemotaxis through its ligand CXCL12 (15). Here, as the cells secrete CXCL12, IFF creates a CXCL12 gradient around the cells in the direction of flow, leading to a higher chemokine concentration on the downstream side of the cells. We did not observe an increase in total or phosphorylated CXCR4 due to IFF or a difference in CXCL12 levels in these cells (Supplementary Fig. S2). This suggests that IFF does not alter the levels of these proteins but may alter how the cells migrate by creating a small but biologically relevant transcellular gradient (35). The fact that this phenomenon is only observed in cells that have undergone EMT suggests that in addition to inducing invasion in preinvasive cells, IFF may play a role in sustaining invasion as cells move through the stroma individually. Invasive cells, like those which have undergone EMT, have been shown to be more sensitive to chemokine gradients (44, 45). This may allow them to home to specific organs producing a chemoattractant (46, 47), but it may also facilitate IFF-induced invasion, as we have outlined here.

Although static invasion levels between NeuN and NeuT are similar, the mechanism by which these cells invade appears to be different. NeuN cells have a tendency to invade collectively and NeuTs invade as single cells (Supplementary Fig. S4). To identify possible phenotypic differences responsible for their differential responses to IFF, we compared NeuN and NeuT gene expression through microarray. More than 2,500 genes were significantly differentially expressed between NeuN and NeuT. Among them, genes known to suppress the PI3K pathway, such as PTEN and PIK3IP1, were downregulated in NeuT cells. Genes associated with the ERBB2 pathway, such as GRB2 and BCL2, were upregulated in NeuT cells. We further profiled these cells' mRNA and protein levels and observed that NeuT lacked key epithelial cell–associated proteins/genes (E-cadherin and integrin α6) and expressed higher levels of mesenchymal markers (N-cadherin and vimentin; Fig. 5A and B), suggesting that EMT alters the mode of invasion of cells, and this leads to changes in the response of cells to IFF. To test this idea specifically, we showed that induction of EMT in NeuN converted the IFF invasion response from chemokine-independent to chemokine receptor–dependent. Previous studies have implicated ERBB2 expression and activation in EMT (31, 48, 49), therefore our findings could suggest that the observed EMT-specific response to IFF was related to expression of HER2. Although the HER2 pathway is constitutively active in NeuT, both cell lines overexpress similar levels of HER2 so this alone could not explain the differential effects of IFF on these cells.

One limitation of our study is the difference in pressure gradients between our model and that of advanced tumors. The hydrostatic pressure difference generated by our 3D system was approximately 1 mm Hg, comparable with physiologic fluid pressure levels in normal tissue and early lesions (11) but the hydrostatic pressure differences present in advanced tumors are much larger (10, 11). Nevertheless, the in vitro collagen gels used in our experiments are much more permeable than normal and cancerous tissues, resulting in IFF velocities that are within a tumor-relevant range (12, 13). We cannot, however, discount the possibility that the larger absolute pressure magnitudes or pressure gradients associated in more advanced tumors may have an effect on cell behavior. Future studies should therefore consider how activation of the signaling pathways described here is affected by fluid pressures and pressure gradients closer to those levels observed in vivo. In addition, although our data suggest potential relevance of our findings to treatment of HER2-positive DCIS, the study focuses on invasion of single cells embedded in a matrix. DCIS cells in vivo are organized in 3D complex structures that are highly dependent on cell-to-cell interactions and basement membrane restrictions (50). Their invasion through the stroma is also dependent on stroma-associated cells, which are well known to have both positive and negative influences on cancer progression (2). In the future, investigating the role of IFF on 3D DCIS structures in a more clinically relevant matrix will lead to better understanding of the role of IFF on breast cancer invasion. We must also consider the contribution of other proposed mechanisms of IFF mechanotransduction, including mechanosensors such as integrins and the tumor cell glycocalyx, which have both been implicated in IFF-induced cell invasion (17, 19). Examining these other mechanisms may help identify the upstream mediator of PI3K phosphorylation in NeuN cells. Further studies should also focus on identifying the implications of EMT in altering IFF response in other cancers (e.g., melanoma, glioma and renal cell carcinoma).

Taken together, our results demonstrate for the first time that IFF increases invasion of cells via separate mechanisms depending on the stage of the cancer cells. This is the first study to examine the effect of IFF on separate stages of breast cancer, especially in the context of how it may influence the early steps of tumor progression. We showed that IFF increases the invasion of different types of human mammary breast cancer cells through activation of PI3K. We further demonstrated that EMT alters how cells respond to IFF, engaging a CXCR4/CXCL12-dependent autologous chemotaxis mechanism that signals through p110β. Understanding these cellular responses to IFF will increase our understanding of how biophysical forces interact with molecular factors to drive breast cancer progression and may help identify novel therapeutic regimens.

No potential conflicts of interest were disclosed.

Conception and design: A.M. Tchafa, M.J. Reginato, A.C. Shieh

Development of methodology: A.M. Tchafa

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.M. Tchafa, M. Ta

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.M. Tchafa, M. Ta, M.J. Reginato, A.C. Shieh

Writing, review, and/or revision of the manuscript: A.M. Tchafa, M.J. Reginato, A.C. Shieh

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.M. Tchafa

Study supervision: A.C. Shieh

The authors thank Arpit Shah and Rawan Shraim for their technical support.

This work was funded, in part, by a CURE grant from Drexel University College of Medicine and the Pennsylvania Department of Health (to M.J. Reginato).

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.