The EGF/CSF-1 Paracrine Invasion Loop Can Be Triggered by Heregulin β1 and CXCL12

An important step in the process of metastasis from the primary tumor is invasive spread into the surrounding stroma. Using an in vivo invasion assay, we have previously shown that imposed gradients of epidermal growth factor (EGF) or colony-stimulating factor-1 (CSF-1) can induce invasion through an EGF/CSF-1 paracrine loop between cancer cells and macrophages. We now report that invasion induced by other ligands also relies on this EGF/CSF-1 paracrine invasive loop. Using an in vivo invasion assay, we show that MTLn3 breast cancer cells overexpressing ErbB3 exhibit enhanced invasion compared with control MTLn3 cells in response to the ErbB3 ligand HRG-β1. The invasive response of both MTLn3-ErbB3 and transgenic MMTV-Neu tumors to HRG-β1 is inhibited by blocking EGF receptor, CSF-1 receptor, or macrophage function, indicating that invasiveness to HRG-β1 is dependent on the EGF/CSF-1 paracrine loop. Furthermore, we show that CXCL12 also triggers in vivo invasion of transgenic MMTV-PyMT tumors in an EGF/CSF-1–dependent manner. Although the invasion induced by HRG-β1 or CXCL12 is dependent on the EGF/CSF-1 paracrine loop, invasion induced by EGF is not dependent on HRG-β1 or CXCL12 signaling, showing an asymmetrical relationship between different ligand/receptor systems in driving invasion. Our results identify a stromal/tumor interaction that acts as an engine underlying invasion induced by multiple ligands. [Cancer Res 2009;69(7):3221–7]

Metastasis is a multistep process involving invasion of the basement membrane and surrounding stroma, intravasation, extravasation, and survival/growth of cancer cells at new sites (1). Our research has focused on elucidating what drives cancer cells to leave the primary tumor and cross the basement membrane, resulting in invasion of the surrounding stroma, with the expectation of finding novel targets of metastasis that may be used to prevent the occurrence of this fatal process. With the development of an in vivo invasion assay in which invasive cells are collected from the primary tumor using needles preloaded with Matrigel and a chemoattractant (2), new insights into the process of invasion have been made. Using this in vivo invasion assay to collect invasive cells from rat MTLn3 breast primary xenograft tumors as well as from transgenic mice in which mammary tumors are induced by the expression of the polyoma middle T oncoprotein from the mouse mammary tumor virus (MMTV) promoter (PyMT; ref. 3), it was observed that macrophages aided cancer cells in invading the surrounding tissue. This was due to a paracrine communication loop involving secretion of colony-stimulating factor-1 (CSF-1) by cancer cells that would activate macrophages to secrete epidermal growth factor (EGF), a chemoattractant for the cancer cells that would drive their invasion (4). Inhibition of either EGF or CSF-1 signaling resulted in decreased invasion to background levels (3), suggesting that EGF/CSF-1 signaling was the key to in vivo invasion in response to either EGF or CSF-1 in this model.

This research implicated the tumor microenvironment, especially stromal cells such as macrophages, in collaborating in the process of metastasis. Tumor-associated macrophages have been correlated with poor prognosis in several cancers including breast, with high density of tumor-associated macrophages associated with metastasis (5, 6). Furthermore, overexpression of CSF-1, a major growth factor involved in the survival, differentiation, and chemotaxis of macrophages, has been shown to correlate with poor prognosis in breast cancer (7, 8). Studies using transgenic PyMT mice carrying a null mutation in the CSF-1 gene showed that recruitment of macrophages to the primary tumor was dramatically decreased in this model, and accordingly, there was slower tumor progression with significantly reduced metastasis that was rescued on expression of CSF-1 in the mammary tumor epithelium (9). Furthermore, imaging of PyMT tumors in which macrophages were labeled with Texas red dextran showed that tumor cell motility was associated with the presence of macrophages, with more frequent motility associated with perivascular macrophages (10), suggesting again a close interaction between carcinoma cells and macrophages within the primary tumor that facilitates invasion.

In vitro characterization of invasion typically involves studying cancer cell traversal of a Matrigel barrier. Using this assay, several chemoattractants have been found to stimulate invasion, including CXCL12 (SDF-1) and heregulin β1 (HRGβ1). CXCL12 is a chemokine that binds to CXCR4. CXCR4 can be highly expressed on breast cancer cells as compared with normal breast tissue (11) and has been implicated as a predictor of poor prognosis in patients with breast cancer (12). In MDA-MB-231 breast cancer cells, it has been shown that CXCL12 stimulates their invasion in vitro, as well as plays a key role in the metastatic behavior of these cells in vivo (11). HRGβ1, an EGF-like ligand that binds and activates ErbB3 and ErbB4, members of the ErbB family of receptor tyrosine kinases (13), has been implicated in modulating the invasive behavior of breast cancer cells. HRGβ1 stimulates the in vitro invasive behavior of MDA-MB-231, and down-regulation of HRGβ1 expression results in impaired invasiveness in vitro and metastasis in vivo (14). HRGβ1 has also been shown to induce the migration and invasion of MCF-7 and T47D cells in vitro through activation of the ErbB2 and ErbB3 receptors (15). Overexpression of ErbB3 in MTLn3 cells results in enhanced chemotaxis to HRGβ1 as well as in increased metastatic behavior of MTLn3 cells, including leaving the primary tumor in greater numbers (16).

Because CXCL12 and HRGβ1 have been implicated in breast cancer cell invasion, we tested their ability to stimulate invasion in vivo in the tumor microenvironment using the in vivo invasion assay. Remarkably, we found that although HRGβ1 can stimulate the in vivo invasion of MTLn3 cells overexpressing the ErbB3 receptor, this invasion is inhibited by addition of Iressa, an EGF receptor (EGFR) tyrosine kinase inhibitor, or an EGF neutralizing antibody, as well as by a CSF-1 receptor (CSF-1R) blocking antibody or by macrophage depletion using liposome-encapsulated clodronate. This dependency on EGF/CSF-1 signaling to stimulate invasion in vivo was recapitulated in the MMTV-Neu transgenic animal model of mammary cancer in which ErbB3 is also expressed (17). In addition, we found that in vivo invasion toward a G-protein–coupled receptor ligand, CXCL12, is also dependent on the EGF/CSF-1 paracrine loop. Our results suggest that targeting EGF or CSF-1 signaling can be effective in blocking invasion in response to heterologous ligands through a novel dependence on tumor/stroma interactions revealed in vivo.

Cell lines and animal models. All procedures involving mice were conducted in accordance with the NIH regulations on the use and care of experimental animals. The study of mice was approved by the Albert Einstein College of Medicine animal use committee. The rat mammary carcinoma MTLn3 cells transduced with the empty pLXSN retroviral vector or vector containing human ErbB3 receptor (16) were grown in α-MEM supplemented with 5% fetal bovine serum (FBS) and penicillin/streptomycin solution (Life Technologies). The tumor cells were grown to 70% to 85% confluence before being harvested. Cells were detached using PBS-EDTA and scraped using a rubber policeman. Cells (5 × 10⁵) were injected into the right fourth mammary fat pad from the head of 5- to 7-wk-old female severe combined immunodeficient (SCID) mice (National Cancer Institute). Tumors were allowed to grow for 4 wk before cell collection. FVB mice transgenic for the polyoma virus middle T (PyVT or PyMT) oncogene under the MMTV long terminal repeat 206 (LTR), generated as previously described (18), were used for in vivo invasion assays at 12 to 14 wk of age. FVB mice transgenic for Neu driven by the MMTV LTR containing an activating deletion in the extracellular region (17) were used in in vivo invasion assays and for primary tumor harvest at 31 to 33 wk of age.

In vivo invasion assay. Cell collection into needles placed in the primary tumor of anesthetized mice was carried out as described previously (2). Invasive cells were collected into 33-gauge Hamilton needles (Fisher) filled with Matrigel (Beckton Dickinson) diluted 1:10 with L15-BSA and a chemoattractant for 4 h. At the end of collection, the contents of the needles were extruded with ∼30 μL of 4′,6 diamidino-2-phenylindole (DAPI) using a syringe onto a coverslip. The chemoattractants used in this assay include HRGβ1 at a concentration of 50 nmol/L (R&D Systems), EGF at 25 nmol/L (Life Technologies), and CXCL12α at 62.5 nmol/L (R&D Systems). To inhibit activation of the EGF receptor through EGF binding, an antimouse EGF neutralizing antibody (R&D Systems) was used at a concentration of 20 μg/mL. To inhibit activity of the EGF receptor, Iressa (AstraZeneca), a tyrosine kinase inhibitor specific for the EGF receptor, was used at 1 μmol/L. For inhibiting the CSF-1R, a purified monoclonal antimouse CSF-1R antibody (AFS98; ref. 19) was used at 15 μg/mL. To block activation of CXCR4, AMD3100 was used at 100 nmol/L (Sigma Chemical Co.). To block HRGβ1-mediated signaling, an ErbB3 antibody that blocks binding of HRGβ1 to the ErbB3 receptor was used (HER-3 Ab-5, Fisher). As a control antibody for the experiments involving signaling inhibition using blocking/neutralizing antibodies, a mouse IgG antibody was used at the same concentration used for the experimental antibodies (Jackson ImmunoResearch). To functionally impair macrophages in mice bearing tumors, clodronate liposomes were administered systemically. Empty (control) and clodronate-containing liposomes were administered by tail vein into mice 48 h before the in vivo invasion assay. Liposomes were prepared as detailed in ref. 20 using clodronate at a concentration of 2.5 g/10 mL PBS. Clodronate (or Cl2MDP) was a gift of Roche Diagnostics GmbH. Phosphatidylcholine (Lipoid E PC) was obtained from Lipoid GmbH. Cholesterol was purchased from Sigma. Animals were injected with 100 μL liposome solution/10 g of weight.

Determination of cell types collected in the in vivo invasion assay. After 4-h collection, the invasive cells were extruded from the needles using 10% paraformaldehyde into poly-l-lysine–coated MatTek dishes and fixed for 1 h at room temperature. To block nonspecific binding, the samples were incubated overnight at 4°C in TBS-1% FBS. The blocking solution was removed, and cells were washed with TBS-1% bovine serum albumin (BSA) and incubated with a primary antibody mixture of mouse anti-ErbB3 antibody (Neomarkers) for carcinoma cells and rat anti-F4/80 (21) for macrophages in TBS-1% BSA. After 1-h incubation with the primary antibodies, the cells were washed with TBS-1% BSA and incubated in a mixture of goat anti-mouse Cy3 and sheep anti-rat FITC. The cells were rinsed and left in TBS-1% BSA with DAPI and counted using a fluorescence microscope.

Microchemotaxis chamber assay. A 48-well microchemotaxis chamber (Neuroprobe) was used as described previously (16). Briefly, cells were starved for 3 h in L15 media supplemented with 0.35% BSA (L15-BSA) at 37°C, then detached using PBS-EDTA and loaded into the top wells of the microchemotaxis chamber at 20,000 per well. Cells were allowed to migrate for 4 h at 37°C through a collagen-coated 8-μm pore filter (Neuroprobe) toward bottom wells filled with either EGF or HRGβ1 diluted in L15-BSA with or without EGF neutralizing antibody (R&D Systems) or with L15-BSA alone. The filter was then taken out and cells were fixed in 10% paraformaldehyde for 1 h; nonmigrating cells were removed from the top surface of the filter and the remaining cells on the lower surface were stained overnight in hematoxylin staining solution (Fisher). The membrane was then washed in deionized water and migrated cells were counted using a light microscope.

Primary tumor cell culturing and CSF-1 ELISA. After the animal was euthanized, the primary tumor was removed, rinsed briefly in 70% ethanol, and washed with PBS. In a Petri dish with PBS, the tumor was minced until only small clumps remained. The minced tumor was transferred into a 50-mL conical tube into 10-mL digestion medium consisting of growth medium supplemented with 300 units/mL collagenase (Sigma) and 100 units/mL hyaluronidase (Sigma) and then incubated in a 37°C shaker for 30 min. Tumor cells and clumps were centrifuged at 1,000 rpm, washed 3× with PBS, resuspended in growth medium [DMEM/F-12 (Fisher) supplemented with 10% FBS, 50 units/mL penicillin/streptomycin, 10 μg/mL insulin (Sigma), 1 mg/mL BSA, 5 μg/mL linoleic acid complex (Sigma), 20 units/mL fungizone (Invitrogen), 50 μg/mL gentamicin (Sigma), and 1.2 g sodium bicarbonate per 1,000 mL medium], and plated at densities varying from 1 × 10⁶ to 5 × 10⁶ on 10-cm plates (22). The following day, the cultures were washed with growth medium and medium was changed everyday following that. Cells were cultured for 5 to 10 d before they were used for in vitro stimulations. Cells were grown to 85% to 90% confluency, starved for 3 h in L15-BSA, and stimulated for 4 h with 12.5 nmol/L HRGβ1 in L15-BSA. After 4 h, the supernatants were collected, spun down to remove debris, and stored at −20°C. The samples were tested for the presence of secreted CSF-1 using a mouse CSF-1 ELISA kit (R&D Systems).

Statistical analysis. Multiple comparisons were done using ANOVA, and two-condition comparisons were done using t test.

HRGβ1 stimulates the in vivo invasion of MTLn3-ErbB3 cells. ErbB3 overexpression in MTLn3 cells increases their in vitro chemotactic response to HRGβ1 through the formation of active ErbB2-ErbB3 heterodimers (16), as well as their metastatic behavior in vivo. To investigate whether HRGβ1 could also stimulate MTLn3 rat mammary adenocarcinoma cell chemotaxis and invasion in vivo, cells expressing increased levels of ErbB3 (MTLn3-ErbB3) or empty vector control cells (MTLn3-pLXSN) were injected into the mammary fat pads of SCID mice to form orthotopic tumors. Four weeks postinjection, microneedles containing Matrigel and a chemoattractant were inserted into the primary tumors to collect invasive cells. HRGβ1 induced in vivo invasion in MTLn3-ErbB3 tumors (Fig. 1A) in a dose-dependent fashion, with the optimal concentration being 50 nmol/L HRGβ1. Using EGF as a chemoattractant, comparable invasiveness of MTLn3-pLXSN and MTLn3-ErbB3 cells was observed, indicating that overexpression of ErbB3 did not affect the ability of MTLn3 cells to respond to EGF (Fig. 1B). However, MTLn3-ErbB3 tumors showed a greater in vivo invasion in response to HRGβ1 compared with control pLXSN tumors (Fig. 1B).

Previous studies have shown that macrophages comigrate with breast cancer cells in response to EGF stimulation in the needle collection assay (3). To assess whether macrophages also comigrate with cancer cells in HRGβ1-mediated in vivo invasion, the invasive cells collected on HRGβ1 stimulation from MTLn3-ErbB3 tumors were fixed and stained with an ErbB3 antibody to detect MTLn3-ErbB3 cells and an F4/80 antibody to detect macrophages. MTLn3-ErbB3 cells represented 75% (±5%) of the invasive population, and macrophages, 25% (±5%), with no other cell types present in significant amounts, similar to what we have previously reported for the in vivo invasion in response to EGF (3). To test whether macrophages play a role in the invasive behavior of MTLn3-ErbB3 cells in response to HRGβ1, macrophages were functionally impaired using clodronate-containing-liposomes (20). SCID mice bearing MTLn3-ErbB3 tumors were injected i.v. with clodronate-containing liposomes or empty liposomes as a negative control 48 hours before performing the in vivo invasion assay. As described in Supplemental data (Supplementary Fig. S1), although the number of macrophages in the primary tumor is not decreased, clodronate liposome treatment results in reduced macrophage function as measured by macropinocytosis of 70-kDa dextran. There was a significantly reduced invasive response to HRGβ1 in tumors that had been pretreated with clodronate, compared with those pretreated with empty liposomes (ratio of HRGβ1-induced in vivo invasion in the presence of clodronate-containing liposomes to that in the presence of empty liposomes, 0.6 ± 0.1; P < 0.01), whereas basal invasion in the absence of attractant was not affected. This indicates that macrophages contribute to the invasive behavior of MTLn3-ErbB3 cells in response to HRGβ1 stimulation.

The EGF/CSF-1 paracrine loop is required for the in vivo invasion of MTLn3-ErbB3 cells in response to HRGβ1. The previous section suggested that macrophages could play an active role in HRGβ1-induced in vivo invasion of MTLn3-ErbB3 cells. We therefore tested whether the EGF/CSF-1 paracrine loop was necessary for HRGβ1-induced in vivo invasion. We first evaluated the role of CSF-1 signaling by blocking the CSF-1R on macrophages using a CSF-1R blocking antibody. As a positive control, we confirmed that the invasion of MTLn3-ErbB3 tumor cells toward EGF was inhibited on blockage of the CSF-1R (P < 0.05; Fig. 2A,, left). Similarly, addition of the CSF-1R antibody also inhibited HRGβ1-stimulated in vivo invasion by MTLn3-ErbB3 tumors (P < 0.005; Fig. 2A , right), indicating that activation of the CSF-1R in macrophages is required for MTLn3-ErbB3 cells to invade in response to HRGβ1.

To determine whether EGF signaling also plays a role in the invasive response of MTLn3-ErbB3 cells to HRGβ1 stimulation, an EGFR-specific tyrosine kinase inhibitor, Iressa (23), was added to microneedles loaded with HRGβ1 (Fig. 2B). Addition of Iressa inhibited the in vivo invasion of MTLn3-ErbB3 cells to HRGβ1, bringing the response down to basal levels (P < 0.05). This suggests that EGFR signaling is required for MTLn3-ErbB3 cells to invade when stimulated with HRGβ1. To test whether EGF specifically was required, we determined the effect of an EGF antibody that blocks the binding of EGF to its receptor, EGFR (Fig. 2C). The EGF antibody significantly inhibited the in vivo invasion of MTLn3-ErbB3 cells to EGF as anticipated (P < 0.05, Fig. 2C,, left) but also resulted in inhibition of the in vivo invasion of MTLn3-ErbB3 cells to HRGβ1 (P < 0.005; Fig. 2C,, right). To check that the neutralizing EGF antibody was specific in blocking EGF-mediated signaling and does not affect HRGβ1 binding and chemotaxis, a Boyden chamber was used to analyze the in vitro chemotaxis behavior of MTLn3-ErbB3 cells in the presence of the EGF antibody. Whereas the EGF antibody itself had no effect in the basal motility of MTLn3-ErbB3 cells (P < 0.6, data not shown), the EGF antibody blocked the chemotaxis of MTLn3-ErbB3 cells to EGF (P < 0.005) but not to HRGβ1 (P < 0.15; Fig. 2D), showing that the antibody was not directly affecting HRGβ1 chemotaxis. This result indicates that HRGβ1-induced in vivo invasion is dependent on the secretion of EGF within the primary tumor. In summary, these results indicate that HRGβ1-mediated in vivo invasion of MTLn3-ErbB3 tumors is dependent on EGF/CSF-1 signaling within the primary tumor.

An ErbB2 transgenic model recapitulates the MTLn3-ErbB3 xenograft model of in vivo invasion to HRGβ1. To determine whether HRGβ1 mediated in vivo invasion is dependent on EGF and CSF-1 signaling in other breast cancer models, we compared the MMTV-PyMT and MMTV-Neu transgenic models (17, 18). We have previously shown that HRGβ1 did not induce invasion of MMTV-PyMT tumors (3), but MMTV-Neu tumors showed high levels of expression of ErbB3 (17). Consistent with their ErbB3 expression, MMTV-Neu tumors showed enhanced in vivo invasion in response to HRGβ1 (P < 0.005) whereas MMTV-PyMT tumors did not (Fig. 3A). We next tested the effect of blocking EGF and CSF-1R signaling on HRGβ1-induced in vivo invasion (Fig. 3B). Addition of the EGF neutralizing antibody (P < 0.005) or CSF-1R blocking antibody (P < 0.005) resulted in inhibition of MMTV-Neu tumor invasion in response to HRGβ1. We conclude that HRGβ1 stimulates in vivo invasion in MMTV-Neu tumors also by activating the EGF/CSF-1 paracrine loop. To test how HRGβ1 could trigger the EGF/CSF-1 paracrine loop, we stimulated MMTV-Neu primary tumor cells in vitro with HRGβ1 and tested the supernatants for CSF-1 secretion. Cultured Neu primary tumor cells were stimulated for 4 hours with HRGβ1 (the same time used for the in vivo invasion assay) and the supernatants were collected and assayed for the presence of CSF-1 protein using a commercially available ELISA. There was a modest but significant increase in CSF-1 secretion on HRGβ1 stimulation of Neu tumor cells (ratio of HRGβ1 stimulated to buffer stimulated, 1.60 ± 0.66, mean and SD; n = 6; P < 0.05). Thus, HRGβ1 induction of CSF-1 production could induce the EGF/CSF-1 paracrine loop.

CXCL12-mediated in vivo invasion in MMTV-PyMT tumors is also dependent on activation of EGF/CSF-1 signaling. We had previously found that CXCL12 also induced in vivo invasion in MMTV-PyMT tumors and that this invasive response could be blocked on addition of Iressa (Supplementary Fig. S2).⁷

Given our studies showing that HRGβ1-induced invasion was dependent on the EGF/CSF-1 paracrine loop, we decided to test whether a different type of ligand, such as CXCL12, which activates the G-protein–coupled receptor CXCR4, could also activate the EGF/CSF-1 paracrine loop. Similar to our results with HRGβ1-induced invasion, both the neutralizing EGF antibody (P < 0.005) and the CSF-1R blocking antibody (P < 0.005) resulted in decreased in vivo invasion of MMTV-PyMT tumors in response to CXCL12 (Fig. 4). To evaluate the contribution of macrophages to CXCL12 in vivo invasion, MMTV-PyMT mice were injected with either empty or clodronate-containing liposomes via tail vein 48 hours before measurement. Clodronate treatment resulted in significantly reduced in vivo invasion to CXCL12 in PyMT tumors (the ratio of in vivo invasion to CXCL12 in tumors pretreated with clodronate-containing liposomes to that in tumors pretreated with empty liposomes, 0.4 ± 0.2; P < 0.04).

CXCL12 and HRGβ1 signaling are not required for invasion in response to EGF. Given that CXCL12- and HRGβ1-induced invasion are dependent on EGF/CSF-1 signaling, we then wished to determine whether EGF/CSF-1–induced invasion was in turn dependent on CXCL12 or HRGβ1. We first tested whether CXCL12 signaling was necessary for EGF/CSF-1 in vivo invasion. We added the CXCR4 inhibitor AMD3100 (24) to needles containing CXCL12 as a positive control to confirm that AMD3100 could block CXCL12-mediated in vivo invasion from PyMT tumors. As expected, addition of AMD3100 resulted in inhibition of CXCL12-mediated in vivo invasion from PyMT tumors (P < 0.005; Fig. 5A); however, addition of AMD3100 to EGF-containing needles did not result in impaired in vivo invasion to EGF in PyMT tumors (P < 0.4; Fig. 5B). These results show that although CXCL12-induced in vivo invasion is dependent on EGF/CSF-1 signaling, EGF-induced in vivo invasion is not dependent on CXCL12 signaling.

Similarly, to test the role of HRGβ1 in EGF-induced in vivo invasion, we used an anti-ErbB3 blocking antibody. ErbB3 is the major receptor for HRGβ1 in MTLn3-ErbB3 cells (16), and this antibody blocks the binding site for HRGβ1 on ErbB3. Addition of the ErbB3 blocking antibody inhibited invasion of MTLn3-ErbB3 tumors to HRGβ1 (P < 0.005; Fig. 5C) but not to EGF (P < 0.4; Fig. 5D), indicating that although HRGβ1-induced in vivo invasion is dependent on EGF signaling, EGF-induced in vivo invasion is not dependent on HRGβ1.

We have previously shown that breast cancer cells invade surrounding breast tissue with the help of macrophages (3). This process involves a paracrine communication loop between cancer cells and macrophages in which cancer cells secrete CSF-1, a chemoattractant for macrophages. Macrophages are in turn stimulated by CSF-1 and secrete EGF, a ligand that binds to the EGF receptor in breast cancer cells and directs their chemotaxis (4). Within the primary tumor, externally imposed gradients of EGF or CSF-1 will induce cancer cell and macrophage invasion up the gradient. We report here for the first time that other ligand/receptor systems can induce in vivo invasion responses that are dependent on this EGF/CSF-1 paracrine loop. Both a xenograft model (MTLn3-ErbB3 ref. 16) overexpressing the receptor for heregulin, ErbB3, and a transgenic model (MMTV-Neu ref. 17) that expresses ErbB3 show heregulin-induced invasion. This heregulin-induced invasion is dependent on macrophage function as indicated by clodronate liposome inhibition, as well as by inhibition using a blocking CSF-1R antibody. Both an EGFR-specific inhibitor, Iressa, as well as an anti-EGF antibody show that EGFR function and signaling via extracellular EGF are required for the heregulin-induced invasion, but not for heregulin chemotaxis in vitro. A ligand for CXCR4, CXCL12 (SDF-1), effectively induced invasion in the MMTV-PyMT model (which showed no invasion in response to heregulin). CXCL12-induced in vivo invasion was also dependent on macrophage function, CSF-1R function, and extracellular EGF signaling.

Whereas inhibition of the EGF/CSF-1 paracrine loop blocked in vivo invasion in response to HRGβ1 and CXCL12, inhibition of either CXCR4 or ErbB3 did not block in vivo invasion induced by EGF. These results show that the EGF/CSF-1 in vivo paracrine invasion loop is independent of HRGβ1 and CXCL12 but can be triggered by these chemoattractants. Based on these results, we propose a novel model for in vivo invasion (Fig. 6) in which the EGF/CSF-1 paracrine loop is an important driver of invasion within the primary tumor. In its simplest form, invasion is induced by ligands of EGFR or CSF-1R present in the tumor microenvironment, which directly feed into this “invasion engine.” In addition, if the tumor cells (or potentially the macrophages) express other receptors such as ErbB3 or CXCR4, the corresponding ligands such as heregulin or CXCL12 can activate the engine, which can then dramatically enhance the invasion response to the ligand. Both CXCL12 and heregulin are expressed in breast cancers (25–29). CXCL12 can be secreted by cancer-activated fibroblasts (30, 31) whereas heregulin has been reported to be expressed by endothelial cells (32–34). In consequence, the invasion induced in response to the ligands secreted by these cells in the local tumor microenvironment can contribute to enhanced intravasation and metastasis by directing invasion out into the stroma and toward blood vessels. We have previously reported that the MTLn3-ErbB3 cells show enhanced intravasation and metastasis (16), and CXCR4 has also been reported to enhance breast cancer metastasis (11, 35), consistent with this hypothesis. However, not all ligands are able to trigger the invasion response in the models we have tested; the macrophage chemoattractant vascular endothelial growth factor did not induce invasion (3).

A clinical implication of these results is that inhibition of EGFR or CSF-1R signaling could provide a broader inhibition of tumor cell invasiveness than might be anticipated based simply on EGFR expression. EGFR expression levels in MTLn3 cells are slightly elevated (∼50,000 per cell; ref. 36) but are not overexpressed to the level associated with typical EGFR overexpressors such as the MDA-MB-231 cells (700,000 receptors per cell; ref. 37) or MDA-MB-468 cells (1.9 × 10⁶ receptors per cell; ref. 38). Similarly, CSF-1R expression in MTLn3 cells is low (4), and MTLn3 cells do not show CSF-1–induced lamellipod extension responses (data not shown). Thus, our data suggest that invasion by breast tumors that do not show high EGFR or CSF-1R expression can still be sensitive to inhibition of these receptors due to in vivo paracrine interactions that can occur in the tumor microenvironment.

No potential conflicts of interest were disclosed.

Grant support: Grants CA107050 (D. Cox), CA110269 (L. Hernandez), CA77522 (T. Smirnova and J.E. Segall), and CA100324 (E.R. Stanley, J. Wyckoff, J.W. Pollard, and J.E. Segall); The Canadian Breast Cancer Research Alliance and Terry Fox Foundation (W.J. Muller); and grant CA131270 (J.W. Pollard). J.E. Segall is the Betty and Sheldon Feinberg Senior Faculty Scholar in Cancer Research.

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.