Invasion of tumor cells into the surrounding connective tissue and blood vessels is a key step in the metastatic spread of breast tumors. Although the presence of macrophages in primary tumors is associated with increased metastatic potential, the mechanistic basis for this observation is unknown. Using a chemotaxis-based in vivo invasion assay and multiphoton-based intravital imaging, we show that the interaction between macrophages and tumor cells facilitates the migration of carcinoma cells in the primary tumor. Gradients of either epidermal growth factor (EGF) or colony-stimulating factor 1 (CSF-1) stimulate collection into microneedles of tumor cells and macrophages even though tumor cells express only EGF receptor and macrophages express only CSF-1 receptor. Intravital imaging shows that macrophages and tumor cells migrate toward microneedles containing either EGF or CSF-1. Inhibition of either CSF-1– or EGF-stimulated signaling reduces the migration of both cell types. This work provides the first direct evidence for a synergistic interaction between macrophages and tumor cells during cell migration in vivo and indicates a mechanism for how macrophages may contribute to metastasis.
The tumor microenvironment contains stromal cells that influence the behavior of the tumor (1 , 2) . Of these, there is increasing evidence that macrophages play an important role in modulating the metastatic capacity of the tumor. This includes clinical evidence showing a strong correlation between tumor-associated macrophages (3 , 4) and poor prognosis, and genetic studies in mice in which decreased numbers of macrophages in the tumor bed are associated with a large reduction in the rates of metastasis (5 , 6) . Although macrophages may contribute factors that affect tumor progression by altering the microenvironment with angiogenic and proteolytic factors (6) , these cells also are capable of producing growth factors, including members of the epidermal growth factor (EGF) family, which may directly influence the behavior of tumor cells (3 , 7) . During wound healing or at sites of infection, macrophages synthesize chemotactic factors that recruit other blood cells. Because macrophages migrate to and function within specific tissue sites, it is possible that, within tumors, they also could provide chemotactic cues that promote the egress of carcinoma cells from the tumor (8) .
To image and measure migration and chemotaxis at the cellular level within primary metastatic tumors, we have developed animal models that allow direct examination, by intravital imaging, of the behavior of Green Fluorescent Protein (GFP)-expressing carcinoma cells in primary tumors in vivo (9, 10, 11) . Because tumor growth and metastasis are unaffected by expression of GFP (10) , the behavioral phenotype of cells within GFP-expressing metastatic and nonmetastatic tumors can be correlated with metastatic potential (11) . Intravital imaging of orthotopic rat mammary tumors has shown that increased carcinoma cell orientation and locomotion toward blood vessels correlate with increased numbers of carcinoma cells in blood vessels exiting the primary tumor and with metastasis (9 , 11) . Growth factors potentially chemotactic for tumor cells including EGF are present in blood, macrophages, platelets, and smooth muscle cells near vessels (12, 13, 14, 15) . Overexpression of the EGF receptor has been shown to correlate with metastasis and poor prognosis in a number of tumor types, including small-cell lung cancer, breast cancer, gastric cancer, and prostate cancer (16, 17, 18, 19) . Cell lines that overexpress EGF receptors also are more metastatic in vivo (20) , and experimental expression of the EGF receptor in nonmetastatic cells increases their chemotactic responses to EGF in vitro and metastatic ability in vivo (21 , 22) . Therefore, EGF receptor-mediated chemotaxis within the primary tumor may be important in enhancing invasion, intravasation, and metastasis in addition to the well-characterized effects of EGF receptor signaling on mitogenesis.
We have developed an in vivo invasion assay to test the hypothesis that chemotaxis by carcinoma cells in the primary tumor is an important step in invasion. In this assay, cells are collected by chemotaxis from live primary tumors in rats using microneedles filled with Matrigel and containing growth factors to mimic chemotactic signals that may be present in the primary tumor (23) . To investigate chemotaxis as a determinant of invasion by carcinoma cells in the primary tumor and the mechanism by which macrophages affect invasion, we have combined the in vivo invasion assay with multiphoton-based intravital imaging in mice with mammary tumors produced by the mammary epithelial restricted expression of the Polyomavirus middle T oncogene (PyMT; ref. 24 ). Transgenic mice selectively expressing PyMT in the mammary epithelium, under control of the mouse mammary tumor virus (MMTV) promoter, rapidly develop multifocal mammary adenocarcinomas (25) . Using this approach, we have identified a paracrine interaction involving reciprocal signaling between carcinoma cells and macrophages involving EGF receptor ligands and the macrophage growth factor colony-stimulating factor 1 (CSF-1). This paracrine interaction is involved in the EGF receptor-mediated invasion by carcinoma cells in mammary tumors.
MATERIALS AND METHODS
Transgenic mice were maintained on a segregating FVB-C3H/B6 background. The details of the origin and identification of MMTV-PyMT and Csf1op/Csf1op/MMTV-PyMT, Csf1op/Csf1op/MMTV-PyMT/CSF-1 TG, WAP-Cre/CAG-CAT-EGF/MMTV-PyMT, and Lys-GFPKi mice have been described previously (5 , 25, 26, 27) . Lys-GFPKi mice, in which the insertion of GFP into the lysozyme gene locus (lys-GFPKi) created mice with green fluorescent macrophages and granulocytes, were crossed with the MMTV-PyMT mice to produce tumors with GFP-labeled macrophages. The WAP-Cre/CAG-CAT-EGFP transgene resulted in β-actin promoter-driven eGFP expression that is activated by Cre to label ductal epithelial cells in the mammary gland. Tumors, except when noted, were allowed to grow for 16 to 18 weeks before cell collection to ensure late-stage carcinomas and increased metastasis as described previously (5 , 24) .
EGF receptor overexpressing cells were created by transfecting MTLn3 cells using the pLXSN retroviral vector containing the EGF receptor (courtesy of Dr. David Stern, Yale University, New Haven, CT). Cells were selected as a heterogeneous population of G418-resistant clones. Cells were grown in α-MEM with 5% fetal bovine serum (FBS) and harvested using trypsin-EDTA. A total of 1 × 106 cells were injected into the mammary fat pad of severe combined immunodeficiency mice, and tumors were allowed to grow for 4 to 5 weeks before cell collection.
Cell collection into needles placed into anesthetized animals was carried out as described previously (23 , 28) . After 4 hours, the collection needles were removed, and the contents were ejected with ∼30 μL of L15-BSA through a syringe onto a coverslip. The concentration of growth factors in the needle was determined by multiplying the affinity of the growth factor for its receptor by ∼25, which is sufficient to generate a concentration within 100 μm of the bevel of the needle equal to measured concentrations of circulating growth factors in vivo.
To inhibit the EGF receptor, PD153035, a tyrosine kinase inhibitor specific for the EGF receptor (29) , was used. For inhibiting CSF-1 and the CSF-1 receptor, rabbit antihuman urinary CSF-1 (30) and monoclonal antimouse CSF-1 receptor (ref. 31 ; courtesy Dr. S. Nishikawa, Kyoto University Medical School, Kyoto, Japan) antibodies were used, respectively. For PD153035, needles were prepared as described previously, containing 10 nmol/L EGF, 25 nmol/L EGF, 25 nmol/L CSF-1, and 10% FBS in L15-BSA with 1% DMSO or 5 μmol/L or 15 μmol/L PD153035 in 1% DMSO. For the antihuman CSF-1 experiments, needles contained 25 nmol/L EGF or 25 nmol/L CSF-1 in L15-BSA with 10 μg of affinity-purified antibody. For the antimouse CSF-1 receptor experiments, needles contained 25 nmol/L EGF or 25 nmol/L CSF-1 in L15-BSA with either 25% nonimmune ascites or 25% antimouse CSF-1 receptor ascites.
To determine whether PD153035 has any effect on CSF-1 receptor-mediated motility, an in vitro motility assay was performed. Macrophages were allowed to grow to confluence in three dishes containing DEM without CSF-1 or PD153035, DEM with CSF-1 but no PD153053, or DEM with CSF-1 and 5 μmol/L PD153035. A wound was created, and the cells were imaged on an inverted scope for 7 hours.
Calculation of the Shape of the Gradient Emanating from Collection Needles.
The diffusion gradients coming from the needle were estimated by two different methods, assuming diffusion constants of EGF and CSF-1 to be 1.6 × 10−6 cm2/s (32) . In method 1 for the regions immediately inside and outside the needle, linear interpolation, confirmed by relaxation modeling, was used to connect the equations covering the interior and exterior of the needle (33) .3 The results of method 1 were similar to the relaxation model described as method 2 in the Supplemental Data. The results of method 1 are plotted in Fig. 1E⇓ . The net effects of viscosity in both methods is to change the time scale by a factor corresponding to the increase in viscosity relative to water and could be neglected for the long collection times used in this study.
Multiphoton Imaging of Cell Collection.
Cell collection needles were prepared as described previously and placed in WAP-Cre/CAG-CAT-EGFP/MMTV-PyMT–, Tie2-GFP/MMTV-PyMT–, or MMTV-PyMT/lys-GFPKi–generated primary tumors in an isoflurane-anesthetized mouse placed on an inverted microscope and imaged at 960 nm for GFP fluorescence. WAP-Cre/CAG-CAT-EGFP/MMTV-PyMT–generated primary tumors were imaged by multiphoton microscopy performed as described previously (11 , 34 , 35) .
Determination of Cell Types Collected.
Cells collected into needles were extruded into a poly-l-lysine–coated MatTek dish (MatTek Corp., Ashland, MA) containing 20 μL of 10% paraformaldehyde and fixed for 30 minutes. Nonspecific binding was blocked with 100 μL Tris-buffered saline (TBS) with 1% FBS overnight at 4°C. The blocking solution was removed, and a primary antibody mixture of rabbit anti-pankeratin for carcinoma cells and rat anti-F4/80 (36) for macrophages was added in TBS with 1%BSA (TBS-BSA) for 1 hour at room temperature. The cells were rinsed three times with TBS-BSA and incubated in a mixture of goat antirabbit Cy3 and sheep antirat FITC secondary antibodies in TBS-BSA for 1 hour at room temperature. The cells were rinsed as described previously and left in TBS-BSA; 4′,6-diamidino-2-phenylindole (DAPI) was added; and the cells were counted.
Quantitative real time-PCR analysis of mRNA was performed with sequence-specific primer pairs for different cell type markers. Keratin 18 is a type I intermediate filament protein that is expressed in nearly all of the epithelial malignancies. Mac-1 is a macrophage antigen. For the paracrine loop components, mRNA was extracted from fluorescence-activated cell-sorted (FACS) carcinoma cells and FACS macrophages from primary tumors originating in WAP-Cre/CAG-CAT-EGFP/MMTV-PyMT mice and used with sequence-specific primer pairs for the EGF receptor, EGF, CSF-1, and the CSF-1 receptor. Carcinoma cells were sorted by their GFP fluorescence using FACS, whereas macrophages were sorted using the F4/80 primary antibody and R-phycoerythrin–labeled antirat secondary antibody (PharMingen, San Diego, CA). Experiments were performed by following standard procedures described previously (11) .
Growth Factor Specificity of Cell Collection.
We used MMTV-PyMT–induced mammary tumors in mice (25) and the in vivo invasion assay described previously (23) to study chemotaxis by carcinoma and host cells within live primary tumors. Initially EGF was used to visualize the migration of chemotactic cells. Using multiphoton microscopy, the collection of carcinoma cells and macrophages into microneedles in the living tumor was observed directly by time-lapse imaging (Fig. 1A and B)⇓ . Mice with PyMT tumors resulting from crosses of either MMTV-PyMT/WAP-Cre/CAG-CAT-EGFP (GFP-expressing carcinoma cells; ref. 26 ) or MMTV-PyMT/lys-GFP mice (GFP-expressing macrophages; ref. 27 ) were used. In these animals, macrophages and carcinoma cells were observed to actively move toward collecting needles filled with EGF at velocities of several micrometers per minute for both cell types.
Chemotaxis was analyzed in response to a variety of growth factors reported to be involved in progression to malignancy. The growth factors used were chosen for the following reasons: EGF receptor expression is correlated with poor prognosis in breast cancer, and EGF and transforming growth factor α (TGF-α) are chemotactic for breast carcinoma cells (37, 38, 39) . CSF-1 and CSF-1 receptor expression is correlated with invasive mammary tumors in human populations and animal models (40 , 41) , and CSF-1 is chemotactic for macrophages (38 , 42) Heregulin has been shown to enhance motility and migration of cancer cells (43) and, as the ligand for the ErbB2/ErbB3 heterodimer, has been shown to enhance cell proliferation in breast cancers (44) . Platelet-derived growth factor (PDGF) B/B is produced by macrophages and stimulates cell motility in connective tissue cells, monocytes, and neutrophils (45) and is correlated with invasion in a number of human cancers (46) . Furthermore, PDGF receptor β, which responds to PDGF B/B, is found on monocytes and macrophages (47 , 48) . Vascular endothelial growth factor α (VEGF-α) is correlated with angiogenic response (49) and has been shown to stimulate invasion in breast cancer cells (50) . FGF-1 also has been shown to induce malignant behavior in breast cancer (51) .
The most efficient cell collection occurred in response to EGF and TGF-α, and CSF-1 (Fig. 1C)⇓ . FBS, which contains several of the growth factors used in Fig. 1C⇓ , also was effective in collecting cells. This collection was inhibited with PD153035, an inhibitor of the EGF receptor, and was therefore EGF receptor dependent. Heregulin, VEGF-α, FGF-1, and PDGF B/B were not effective at collecting cells above levels obtained with buffer alone.
Cell collection into microneedles filled with various concentrations of EGF, TGF-α, and CSF-1 followed reproducible dose-response curves in MMTV-PyMT–derived tumors (Fig. 1D)⇓ . Approximately 1000 cells were collected from primary tumors of 18-week-old wild-type MMTV-PyMT animals in each needle in 4 hours at the optimum concentration of each EGF receptor ligand and CSF-1, whereas no cells above background were collected with the same range of concentrations of the other growth factors (Fig. 1C and D)⇓ . Unlike CSF-1, TGF-α did not show up on array analysis as being up-regulated in invasive carcinoma cells collected using the in vivo invasion assay (data not shown); therefore, it was not investigated further.
The concentrations given in Figs. 1⇓ and 2⇓ are those loaded within the collection needles. The concentration of the growth factor delivered within 100 μm of the opening of the needle containing 25 nmol/L ligand, a place where cell migration in response to the needle was observed by multiphoton imaging (Fig. 1B)⇓ , was calculated based on diffusion in a low Reynolds number environment, such as whole tissue, to be ∼0.2 to 1.3 nmol/L (Fig. 1E⇓ and Materials and Methods). The circulating concentrations for these growth factors in vivo are reported to be 0.18 to 1.5 nmol/L, suggesting that cells near the collection needle were responding to physiologic concentrations of these growth factors (52 , 53) .
To determine whether cell collection efficiency was related to EGF receptor expression, the collection of cells was scored from mammary tumors made by injecting carcinoma (MTLn3) cells, overexpressing the EGF receptor, into the mammary fat pads of severe combined immunodeficiency mice. Overexpression of EGF receptor caused a significant increase in the number of carcinoma cells collected from mammary tumors at both concentrations of EGF tested (Fig. 2A)⇓ . These results are consistent with previous studies in which cell lines expressing low levels of EGF receptor generated tumors from which only background levels of cells were collected, indicating a contribution of the EGF receptor to cell collection (23) .
Macrophages were collected from the mammary tumors in response to either EGF or CSF-1 (Fig. 2B)⇓ and were observed by multiphoton imaging directly during collection in lys-GFPKi mice with tumors (Fig. 1)⇓ . In response to PDGF, however, the number of cells collected was at the level of the buffer background (∼150 cells), and few were macrophages based on imaging of GFP-macrophages (Fig. 2B)⇓ . When needles were placed into normal mammary fat pads of similarly aged lys-GFPKi mice, only ∼20 cells were collected in response to CSF-1, and all were GFP labeled, suggesting a greatly decreased response exclusively by macrophages (Fig. 2C)⇓ . During the same collection time interval, a mixture of >1000 carcinoma cells and macrophages was collected from mammary tumors (compare scales in Fig. 2B and C⇓ ).
Only Macrophages and Carcinoma Cells Migrate into Needles.
To determine the cell types collected from PyMT mouse mammary tumors into EGF- and CSF-1–containing needles with more precision, DAPI stain was used for all of the cells, anti-pankeratin was used for carcinoma cells, and anti-F4/80 was used for macrophages. F4/80 is a macrophage lineage restricted antigen and is not found on neutrophils (36) , although it is found on eosinophils, which can be easily distinguishable by morphology (54) . As shown in Fig. 3A⇓ , in response to EGF, carcinoma cells comprised ∼73% of the cell population collected, whereas macrophages comprised 26%, collectively accounting for >99% of the cells collected. Similar results were obtained with CSF-1–containing needles (not shown). Furthermore, the same results were obtained with another animal model with mammary tumors in rats prepared by injecting cultured carcinoma cells (MTLn3) into the mammary fat pads (Fig. 3B)⇓ . Cell collection from the rat tumors showed that carcinoma cells comprised 76% and macrophages accounted for 23% of collected cells (Fig. 3B)⇓ . These results indicate that the comigration of macrophages and carcinoma cells is a common property of mammary tumors in different animal models regardless of how the primary tumor was formed and may reveal a common underlying mechanism for migration and possibly invasion.
Quantitative real time-PCR was used to determine the relative enrichment of macrophages and carcinoma cells in the microneedles relative to the primary tumor using macrophage (MAC-1) and carcinoma cell (keratin)-specific primers. mRNA was isolated from the cells collected in the microneedles with 25 nmol/L EGF (Fig. 3C)⇓ . The results show a sixfold enrichment of macrophages (MAC-1 gene) over that found in the primary tumor, whereas the carcinoma cells were not significantly enriched. Quantitative real time-PCR of pieces of tissue obtained from needle biopsies of the same tumor showed no enrichment for macrophages (Fig. 3C)⇓ , indicating that the macrophages in EGF and CSF-1 microneedles were actively collected by chemotaxis and not passively collected by punching the needle into the tumor as in the biopsy. This is consistent with observations made during multiphoton imaging, in which active motility of cells toward the collection needles was observed (Fig. 1B)⇓ .
To investigate why macrophages and carcinoma cells are collected by needles containing either EGF or CSF-1, the expression pattern of these growth factors and their receptors in tumor-associated macrophages and carcinoma cells was determined by real time-PCR. As shown in Fig. 4⇓ , carcinoma cells isolated from WAP-Cre/CAG-CAT-EGFP/MMTV-PyMT–generated primary tumors by FACS express EGF receptor and CSF-1 but neither EGF nor CSF-1 receptor. The GFP-labeled carcinoma cells also were shown to express the PyMT antigen by real time-PCR (Fig. 4A)⇓ , and the carcinoma cells also stained positively for the PyMT antigen in histologic sections (Fig. 4B)⇓ . The reciprocal pattern of expression in FACS macrophages from the same tumor was observed. The macrophages expressed CSF-1 receptor and EGF but not CSF-1, EGF receptor, or the PyMT antigen (Fig. 4A)⇓ . CSF-1 radioimmunoassays (30) of extracts of tumors at different stages showed that small tumors with more stromal cells than carcinoma cells had undetectable levels of CSF-1, whereas large tumors that consisted of mainly carcinoma cells had modest levels of CSF-1 (not shown; ref. 5 ). This, along with the real time-PCR data, indicates that the carcinoma cells produce CSF-1.
The PCR results are consistent with in situ hybridization results showing that macrophages are the only cells adjacent to carcinoma cells in PyMT-derived tumors that express the CSF-1 receptor (5) . Immunostaining using anti–F-4/80 for macrophages and anti–middle-T antigen for carcinoma cells shows that macrophages are found near carcinoma cells in the invasive margin of the mammary tumor in histologic sections (Fig. 4B)⇓ . Our results raise the interesting possibility that carcinoma cells and macrophages are engaged in a paracrine interaction causing them to move as coupled cells toward a source of either EGF or CSF-1. Other cell types that express the EGF receptor, such as fibroblasts (55 , 56) and vascular endothelial cells (57) , were not collected in the needles, possibly because they do not enter either an autocrine- or paracrine-mediated amplification of the chemotactic signal by secreting chemotactic cytokines in response to exogenous EGF stimulation.
Macrophages Are Required for Carcinoma Cell Migration in Response to EGF and CSF-1.
In transgenic mice susceptible to mammary cancer (MMTV-PyMT mice) that also are homozygous for the Csf1op allele (CSF-1 deficient), neither the incidence nor the growth of primary tumors is affected by the absence of CSF-1. However, these mice have a low density of tissue macrophages because of the chronic absence of CSF-1 (58) . This is correlated with delayed onset of metastasis (5) . To investigate the importance of macrophages in carcinoma cell migration, cell collection in microneedles from PyMT-generated tumors in a Csf1op/Csf1op background was compared with cell collection from tumors in wild-type mice (Fig. 5A)⇓ . Here we show a large reduction in the collection of cells from mammary tumors in the Csf1op/Csf1op background by needles filled with either EGF or CSF-1 as compared with cells collected from wild-type PyMT tumors (Fig. 5A)⇓ . Macrophages comprised 5 to 7% of the population of cells collected from the PyMT-generated tumors in the Csf1op/Csf1op background mice using a needle containing 25 nmol/L EGF, consistent with low macrophage densities in the Csf1op/Csf1op tumors. These results indicate that the chemotactic and migratory responses of carcinoma cells to EGF depend on the presence of macrophages.
To directly test the requirement for CSF-1 in invasion, 18-week-old tumors in Csf1op/Csf1op/PyMT CSF-1 TG mice, which contain the MMTV-driven CSF-1 transgene, were used. The CSF-1 transgene only expresses CSF-1 in the mammary and salivary gland because of the MMTV promoter and causes the acceleration of tumor progression and metastasis (5 , 53) . In these animals, 1.7-fold more carcinoma cells were collected in needles containing 25 nmol/L EGF than from similar aged Csf1op/Csf1op/PyMT tumors (Fig. 5B)⇓ . These results indicate that expression of CSF-1 in the mammary gland potentiates the migration of carcinoma cells in response to EGF.
To further investigate the relationship between macrophages and the collection of carcinoma cells by chemotaxis, CSF-1 was infused locally into the MMTV-PyMT tumor 4 hours before needle collection. CSF-1 has been shown to recruit macrophages to the point of administration in the pleural and peritoneal cavities (59) . Previous introduction of CSF-1 into the tumor significantly increased the collection of cells into microneedles containing EGF (Fig. 5C)⇓ . The cells collected into an EGF-containing needle after injection of CSF-1 into the tumor showed a similar ratio of carcinoma cells to macrophages as noninjected tumors (72% to 27%), indicating the enhanced collection of carcinoma cells and macrophages in response to priming the tumor with just CSF-1. These results suggest that the presence of macrophages is essential for the full chemotactic potential of carcinoma cells to be realized in PyMT tumors during cell collection by microneedles.
Requirement of EGF- and CSF-1–Mediated Signals for Cell Migration.
Our results suggest the presence of a paracrine loop in which the carcinoma cells are a source of CSF-1, which attracts macrophages, whereas macrophages respond by releasing EGF, which stimulates carcinoma cells. To further investigate the existence of a paracrine loop, PD153035, a tyrosine kinase inhibitor specific for the EGF receptor (29) , and antibodies that block CSF-1 receptor activity were added to the collection needles to investigate the relative contributions of these growth factor receptors to cell migration in PyMT tumors of 18-week-old wild-type mice. PD153035 is reported to have no effect on the activity of CSF-1 receptor (29) . PD153035 also had no effect on the motility of BAC1.2F5 macrophages in response to CSF-1 in vitro (data not shown). The addition of PD153035 to needles containing EGF or CSF-1 inhibited cell collection (Fig. 6A)⇓ . In CSF-1–containing needles, the number of macrophages collected from tumors in the presence of PD153035 was similar to that collected from normal tissue in response to CSF-1 (Fig. 2C)⇓ , consistent with in vitro results showing that PD153035 does not inhibit macrophage motility. Similar inhibitory results with PD153035 were obtained with an independent animal model (described in Fig. 3B⇓ ) using rats with mammary tumors derived from injection of MTLn3 cells (Fig. 6D)⇓ .
Addition of antibodies to needles that block the activity of murine CSF-1 receptor also resulted in the inhibition of cell collection in EGF- and CSF-1–containing needles to background levels (Fig. 6B)⇓ . The percentage of cells that were macrophages in EGF and CSF-1 needles decreased to <3%, indicating the macrophage motility requires CSF-1 receptor activity.
The addition of antibodies to needles that block human but not mouse CSF-1 resulted in only a slight inhibition of EGF-mediated collection of cells, indicating an endogenous source of mouse CSF-1 during collection. Inhibition to background levels of cell collection was observed in needles containing human CSF-1 and its antibody as expected (Fig. 6C)⇓ . These results are consistent with the ability of the antihuman CSF-1 antibody to inhibit only human CSF-1 placed in the needle and not the mouse CSF-1 generated in situ in response to EGF.
In aggregate, these experiments indicate that the activities of the receptors for EGF and CSF-1 contribute to the collection of carcinoma cells and macrophages from mammary tumors in these two independent animal models in response to either EGF or CSF-1, consistent with the presence of a paracrine loop.
We have shown that in mammary tumors derived from the expression of the PyMT oncogene in mice and in rats and severe combined immunodeficiency mice injected orthotopically with carcinoma cell lines, a paracrine loop operates involving the interaction of carcinoma cells and macrophages. In mouse mammary tumors, in response to needles containing either EGF or CSF-1, only carcinoma cells and macrophages are collected. This is a true paracrine loop requiring the activity of the EGF and CSF-1 receptors on separate cell types because CSF-1 receptors are expressed only on macrophages and EGF receptors are expressed only on carcinoma cells, and inhibition of collection of both cell types results from inhibition of either receptor type. These data, along with the studies described previously (5) , support the existence of a paracrine loop involving the mutual signaling and chemotaxis between macrophages and carcinoma cells that is essential for motility and invasion in mammary tumors. The use of either CSF-1 or EGF in the collection needle may mimic a process in which CSF-1 secreted by carcinoma cells leads to the activation of macrophages to secrete EGF receptor ligands, leading to stimulation of carcinoma cell movement.
The paracrine loop described in this study may be related to malignancy because (1) few cells are collected into needles placed into normal mouse mammary fat pads; (2) the collection of cells from Csf1op/Csf1op/PyMT mice is carcinoma stage specific and follows the delay in progression to malignancy because of the absence of endogenous CSF-1 and a low density of tissue macrophages (5) ; and (3) in recent studies involving one of us (E.R.S), it was shown that mouse CSF-1 antisense administered to nude mice bearing human colon cancer xenografts decreased CSF-1 protein expression and increased mouse survival (60) . More recently, the same group has shown that mouse CSF-1 blockade by antisense oligonucleotides or small interfering RNAs suppressed the growth of human mammary tumor xenografts in nude mice and improved mouse survival. These treatments also suppressed host macrophage infiltration within tumors (61) .
The novelty of our results is the direct demonstration of the existence of a robust and self-propagating paracrine loop in mammary tumors and demonstration of the mechanism by which macrophages enhance carcinoma cell migration by completing this paracrine loop. Macrophages have been hypothesized to play a role in tumor rejection and increased malignancy (3 , 4 , 62) . These opposing hypothetical roles have kept the importance of tumor-associated macrophages in tumor invasion controversial. Our results define a role for macrophages in enhancing cell migration that could contribute to invasion and metastasis. They also suggest a model to explain the requirement for CSF-1 in invasion and progression to metastasis seen in studies with CSF-1–deficient mice (5) .
The chemotaxis of cells that move slowly compared with the rate of diffusion of chemoattractant generally requires the renewed propagation of the chemotactic signal from cell to cell to retain a steep gradient of chemoattractant near each responding cell (63) . The classic example of this type of chemotaxis is that exhibited by Dictyostelium amoebae during mound formation, in which large fields of cells are attracted by the relay of cyclic AMP from cell to cell throughout a large aggregation field (64) . The result is the recruitment of hundreds of thousands of cells from 1 million μm2 of area during morphogenesis, a scale of cell collection that would not be possible by simple diffusion from a point source. In the absence of the ability to relay the chemotactic signal, only cells immediately adjacent to the founder cell, the cell that initially secretes cyclic AMP, would respond, and mound formation would fail. We propose that autocrine and paracrine loops exist in tumors that achieve the same relayed chemotaxis effect by recruiting cells from volumes of the tumor that are vast compared with that possible by simple diffusion of chemoattractant alone. This hypothesis predicts that every malignant tumor has a well-developed aggregation field using defined chemoattractants to drive autocrine and/or paracrine loops, resulting in the accumulation of cells around the initiating chemotactic signal. It is possible that autocrine and paracrine loops will be tumor type specific and will operate not only in the primary tumor but also in secondary and tertiary metastatic tumors. The ability to disrupt autocrine- and paracrine-based relayed chemotaxis raises the possibility that the discovery of self-propagating chemotaxis loops in tumors will provide new therapeutic targets to specifically inhibit invasion and metastasis in primary tumors and metastatic tumors derived from them.
We thank Howard Berg and Benjamin Segall for help in modeling diffusion from the needle tip, and the Analytical Imaging Facility at AECOM.
Grant support: CA100324 from the National Cancer Institute.
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.
Note: Supplementary data for this article can be found at Cancer Research Online (http://cancerres.aacrjournals.org).
Requests for reprints: Jeffrey Wyckoff or John Condeelis, Anatomy and Structural Biology, Analytical Imaging Facility, Albert Einstein College of Medicine, Jack and Pearl Resnick Campus, 1300 Morris Park Avenue, Bronx, NY 10461. Phone: 718-430-3348; E-mail:or
↵3 Berg, personal communication.
- Received April 26, 2004.
- Revision received July 14, 2004.
- Accepted July 30, 2004.