The Gatekeeper Effect of Epithelial-Mesenchymal Transition Regulates the Frequency of Breast Cancer Metastasis¹

EMT⁴ is a variant of transdifferentiation and a well-recognized mechanism for dispersing cell lineages in vertebrate embryos (1). Examples of EMT can be seen during gastrulation, after decondensation of somites and in the formation of the midline palate (2). The fate maps for these cell transitions are hierarchical, timed, and depend on genetic programs controlled by morphogenic cues (3, 4). Recent observations also suggest EMT is a physiological choice for mature epithelium seeking a new phenotype as a fibroblast (5, 6).

The biochemical basis for EMT is grounded in cytokine signaling and the molecular reprogramming of tissue epithelium. Metalloproteinases (7, 8) or membrane assembly inhibitors (9) initiate the process by dismantling local basement membrane. Transforming growth factor β, epidermal growth factor, or fibroblast growth factor 2 facilitate EMT by binding epithelial receptors with ligand-inducible intrinsic kinase activity (10, 11). The activation of Ras and c-Src pathways and a shift in the balance of small GTPase activity (12, 13, 14) provides important transcriptional signals (4, 15) for cytoskeletal remodeling associated with the expression of FSP1. Transitional cells subsequently reorganize their F-actin stress fibers and form pseudopodia to facilitate directional movement (11).

FSP1 is a lineage marker that uniquely identifies fibroblasts or epithelium undergoing EMT during tissue fibrogenesis. Support for this observation comes from showing FSP1 expression in cultured epithelium during EMT stimulated by cytokine exposure (11), histological evidence suggesting that epithelial units expressing FSP1 disaggregate as injured tissues respond to inflammation during the early stages of fibrogenesis (5), as well as direct observation of EMT in transgenic mice with marked epithelium (6). Dividing fibroblasts exposed to nucleoside analogues are also selectively eliminated in transgenic mice expressing thymidine kinase under control of the FSP1 promoter (16).

FSP1 is identical to S100A4, a member of the S100 family of cytoplasmic proteins (17). Members of this family have been implicated in cytoskeletal-membrane interactions, calcium signal transduction, and cellular growth and differentiation (18) Although the precise function of FSP1/S100A4 is not entirely clear, its interaction with cytoskeletal moieties and its early role in EMT suggests that FSP1 may fashion mesenchymal cell shape and enable motility (11). Furthermore, FSP1/S100A4 can act as an angiogenic factor (19), and when Ca²⁺ dimerizes FSP1/S100A4, the dimer binds the COOH-terminal of p53 and alters its function (20, 21). The expression of FSP1/S100A4 indicates the presence of a molecular program determining mesenchymal phenotype.

Although stromal fibroblasts in breast tissue engage and influence the development of mammary epithelium (12, 22), ductal epithelium put in culture under cytokine pressure can also undergo EMT (23, 24). EMT events create ancestral relationships between local cells and may explain the genetic observation in breast tissue that ductal epithelia and stromal fibroblasts share p53 mutations (25) or other losses of heterozygosity (26). This complicated relationship of cell-cell interactions (27) or morphological transitions (28, 29, 30) reappears later when mammary epithelia form tumors. When breast carcinomas metastasize, they also must alter their phenotype to facilitate movement (31). This transition of mammary cancer cells aids in the spread of tumors (30), perhaps by exaptation (32) of the cellular machinery normally used for producing fibroblasts through EMT. The evidence that ductal carcinomas undergo EMT in vivo, however, continues in uncertainty because profiles of surrogate markers used thus far have not been highly specific for an invasive phenotype (33). One marker that appears to overcome this problem in humans is FSP1/S1004 (34, 35, 36), and we favorably demonstrate its functionality in metastatic mammary cancer in PyV-mT mice.

Because genetic background can affect tumor latency in PyV-mT mice (37), we controlled for potential variability in two ways: first, we backcrossed our transgenic lines to BALB/c (greater than F6 generations) and used only F1 crosses to PyV-mT mice or their littermate controls. Second, we timed most experiments to conclude when primary tumor burden reached ∼25% (primary nodule weights/total body weight × 100%). Surface metastases were counted on whole lungs using a dissecting microscope, and micrometastases (clumps of ∼10 cells that were GFP⁺, casein⁺) were determined from 5μ paraffin tissue sections processed for immunohistochemistry or standard histology.

Three genetically engineered strains of mice were used as crosses to PyV-mT tumor mice (Fig. 1 a), including FSP1.ΔTK-transgenic mice expressing ΔTK under the control of the FSP1 promoter (16), FSP1^+/+.GFP mice, which express enhanced GFP under the control of the FSP1 promoter (6), and a targeted mutant mouse in which the endogenous FSP1 gene was replaced with GFP (FSP1^GFP/GFP). The FSP1^GFP/GFP knockin mouse was created by taking a 3.5-kb fragment (from BAC 472-F1/Research Genetics) from the EcoRI site at the end of the FSP1-coding region and subcloning it into the EcoRI site at the 3′ end of the GFP-coding sequence in the FSP1^+/+.GFP-transgenic construct (6), providing a knockin construct with 3.4 kb of the 5′ FSP1 promoter, the EGFPn-1 backbone, and 3.5-kb fragment from the 3′ flanking region of the FSP1 gene. The linearized plasmid was transfected into Tl-1 stem cells and injections into 3.5-day-old C57B1/6 embryos yielded 12 chimeric pups (Vanderbilt Transgenic/ES cells Shared Resource). Three lines with successful transmission by Southern blot (38) were then characterized for use. PCR primers distinguishing mutant from wild-type amplicons were used for thermocycler genotyping. The amplification profiles for ΔTK, GFP, and FSP1 were reported previously (6, 16). Transgenic protocols were approved by the Institutional Animal Care and Use Committee at Vanderbilt University.

PyV-mT mice × FSP1.ΔTK F1 mice were used to test the effectiveness of GCV-induced cell death in dividing metastatic cancer cells that express FSP1.ΔTK (16). GCV, after preliminary testing, was administered at a dose of 75 mg/kg/day by i.p. injection. A control group was given injections of normal saline. To mitigate side effects of the drug on eating, mice were dosed for 7 days and then allowed a recovery time of 7 days before repeating this cycle. Dosing started when the mice reached 20 g of body weight at or near 40 days of age. Dosing covered 4½ cycles (63 days), and each group was sacrificed after the last day of dosing in the fourth cycle with a lethal dose of isoflourane. They were then perfused with 20 ml of ice-cold PBS injected through the heart followed by 20 ml of ice-cold PBS with 30% sucrose. Liver, lung, and breast tissues were put into 4% paraformaldehyde in PBS to fix overnight (6).

Tumors dissected from PyV-mT mice × FSP1^+/+.GFP F1 mice were digested in RPMI 1640, 0.1 mg/ml hyaluronidase V, 3 mg/ml collagenase IV, and 0.2 mg/ml DNase II at room temperature for 4 h, washed three times with DMEM/ Ham’s F-12 complex media, 10% fetal bovine serum, 100 units of penicillin, 100 μg/ml streptomycin, 292 μg/ml glutamine, 1 mm sodium pyruvate, 10 mm HEPES, 0.1 mm MEM nonessential amino acids, and filtered with 100- and 40-μm filters. Sorting of the harvested cells was performed using a Becton Dickinson FACStarPlus cell sorter with standard filters (6). Populations of GFP⁻ and GFP⁺ tumor cells were collected and transferred into PyV-mT⁻ × GFP⁻ F1 littermates. Each recipient was injected through the tail vein with 2 × 10⁶ sorted cells, and 14 days later, liver and lung tissues were harvested for evaluation as above. Micrometastases were counted at ×200/×400 under fluorescent microscopy using 25 random fields (or random interlobular fields) for lung and whole sections from liver, using a single coded section/animal. Fields were converted into mm² surface area.

Tissue sections were stained with sheep anticasein antibody (specific for α, β, and κ bovine caseins; 1:100 dilution; Immunology Consultants, Newberg, OR) followed by rabbit antisheep IgG antibody (1:300 dilution) conjugated to Texas Red (6). Rabbit anti-FSP1 antibody (1:500 dilution) was developed with FITC goat antirabbit IgG (5, 6). Microscopic photography was performed using a Zeiss LSM 410 laser-scanning confocal microscope, with 488-nm excitation for FITC and 543 or 568 nm excitation for Texas Red (Vanderbilt Cell Imaging Shared Resource). Single-depth images were taken of one section/animal at ×400–630 magnification and processed in Photoshop 7.0 (Adobe, San Jose, CA) and Canvas 7.0 (Deneba, Miami, FL).

An ANOVA using either a Wilcoxon rank-sum or Student’s t test to identify significant differences. P < 0.05 was accepted as significant.

Virgin female PyV-mT mice are genetically engineered to express the polyomavirus middle T antigen under the control of the MMTV LTR in mammary epithelium (37). These mice develop synchronous, multifocal mammary carcinomas (37, 39) with a median latency of 80 days on a BALB/c background. The mechanism of oncogenesis in PyV-mT mice is mediated by intracellular signaling through c-Src-dependent tumor induction, Ras-induced Shc-Grb2-dependent growth, and phosphatidylinositol 3′-kinase promotion of tumor progression (40). When PyV-mT mice are crossed with mice expressing thymidine kinase (FSP1.ΔTK; Ref. 16) or GFP (FSP1^+/+.GFP; Ref. 6) under the control of the murine FSP1 promoter (Fig. 1, a and b), mammary tumors became evident at 75 days and continue to enlarge to 35% of body weight ∼95 days after birth (Fig. 1,c). There is a preference in these mice for lung (and liver) metastases, and the number of surface metastases in lung increases with age or extent of tumor burden. Primary tumor histology (Fig. 1, d and e) in these two groups of intercrossed mice is similar to what has been described previously (39). Casein milk proteins are expressed in the cytoplasm of preneoplastic mammary epithelia from virgin mice infected with MMTV-LTR (41). Some PyV-mT tumor cells also express increases in transcripts encoding the β-casein (37), which we detected indirectly using antibodies against all casein milk proteins. Primary carcinoma nodules from PyV-mT × FSP1.ΔTK F1 mice demonstrate weak staining for casein (lo) within the primary tumor mass (Fig. 1,f), which is consistent with a limited presence of caseins in preneoplastic lesions after MMTV-LTR infection (41). In the internodular spaces, however, loose tumor cells strongly express casein (hi). When double stained for casein (red) and FSP1/S100A4 (green), the merged image coexpresses both markers (yellow). FSP1⁺ fibroblasts surrounding the nodular capsules remain green (Fig. 1 g). We suggest these double-positive cells found between primary nodules have undergone EMT and escaped the primary tumor nodule to metastasize.

A cohort of PyV-mT × FSP1^+/+.GFP F1 mice were sacrificed between 75 and 95 days, and macroscopic metastases counted on the surface of the lung (Fig. 2, a and b). At a microscopic level, FSP1⁺ (GFP⁺), casein⁻ fibroblasts were observed in normal lung around alveolar interstitial spaces (Fig. 2,c) and in metastatic lung (Fig. 2,d; arrows). Metastatic lung also contained yellow casein^hi, GFP⁺ tumor cells that appear to attach and traffic through lung blood vessels (Fig. 2,d; arrowheads). Early tumor expansion of metastatic clusters were double positive for casein and GFP, but as the nodular shape of the carcinoma grew, the metastatic tumor cells within the nodule shifted from casein^hi to casein^lo and became GFP⁻ (Fig. 2, e and f). This progressive attenuation of casein staining with increasing tumor size is demonstrated in Fig. 2, g–i. The regression to casein^lo in secondary nodules is consistent with metastatic cells forming various clusters that enlarge in reflection of the primary tumor nodule (30).

FSP1.ΔTK-transgenic mice selectively express thymidine kinase in fibroblasts under the control of the FSP1/S100A4 promoter. Because S100A4 expression is highly correlated with more aggressive and invasive breast cancer (35), we crossed FSP1.ΔTK mice with PyV-mT mice to determine whether metastatic breast cancer cells selectively expressing the FSP1/S100A4 gene are susceptible to conditional apoptosis. Female PyV-mT × FSP1.ΔTK⁺ and PyV-mT × FSP1.ΔTK⁻ F1 mice were assigned to one of three treatment groups: PyV-mT × FSP1.ΔTK⁻ mice given i.p. injection of saline or the nucleoside analogue GCV (75 mg/kg/day) every other week beginning 40 days after birth; and PyV-mT × FSP1.ΔTK⁺ mice given GCV using the same protocol. Conditional exposure to GCV in cells expressing thymidine kinase induces DNA chain termination and cell death during mitosis (16). Treatment with GCV significantly attenuated (P ≤ 0.01) the number of gross lung surface metastases in PyV-mT × FSP1.ΔTK⁺ mice when compared with PyV-mT × FSP1.ΔTK⁻ mice, an 86% decrease in the number of metastases (Fig. 2 j). Primary tumor nodules among treatment groups did not change their weights at harvest, suggesting the thymidine kinase effect on cell depletion was largely directed toward dividing FSP1⁺ tumor cells with metastatic potential.

It has been known for some time that breast tumor cells with metastatic potential express greater amounts of S100A4 in culture (42, 43) or on histopathology (35, 43) than similar tumor cells with less ability to metastasize. We therefore asked whether the expression of FSP1/S100A4 enables metastatic disease by trying to identify the cell type responsible. Murine FSP1 is only expressed in fibroblasts or epithelia undergoing EMT (5), and FSP1^+/+.GFP mice selectively express GFP in normal tissue fibroblasts under the control of the murine FSP1/S100A4 promoter (6).

To determine whether mammary tumor cells expressing FSP1/S10A4 produce metastatic disease more quickly than cells which do not, we prepared cell suspensions from primary tumor nodules harvested from female PyV-mT × FSP1^+/+.GFP F1 mice, and FACS sorted this cell suspension into two pools that were either GFP⁺ (also casein^hi) or GFP⁻ (also casein^lo; Fig. 3,a). A total of 2 × 10⁶ cells from each pool were then separately injected by tail vein into female PyV-mT⁻ × FSP1^+/+ (GFP⁻) F1 littermate recipients to determine the number of microscopic tumors that metastasize to lungs or liver after 14 days. Sorted cells placed in early culture for staining revealed phenotypic differences based on sorting characteristics; casein^hi, GFP⁺ tumor cells, in addition to coexpressing increased casein and FSP1/S100A4, also had more filamentous pseudopodia as cytoplasmic fringe compared with casein^lo, GFP⁻ tumor cells (Fig. 3,b). Approximately 4–5% of these GFP⁺ cells were identified as casein⁻ fibroblasts (from four experiments; data not shown). Micrometastases were counted/mm² in random fields of sectioned lung and in whole liver sections from each group of injected PyV-mT⁻ × FSP1^+/+ F1 littermate recipients. The recipient mice receiving casein^hi, GFP⁺ cells had significantly (P ≤ 0.01) more metastases in lung and liver than the group receiving casein^lo, GFP⁻ tumor cells (Fig. 3 c). This observations indicates that the FSP1/S100A4 promoter is selectively active in murine tumor cells relocating to metastatic sites.

The phenotype of adoptively transferred tumor cells that traffic to the lung change over time. Normal lung from PyV-mT⁻ × FSP1^+/+ F1 littermates used as recipients do not contain any GFP or casein staining (Fig. 4,a). Fourteen days after adoptive transfer, the casein^hi, GFP⁺ tumor cells populate in numerous clumps between air sacs (Fig. 4,b) and spread out around blood vessels (Fig. 4,c). As the nodules enlarge, the original double positivity in transferred casein^hi, GFP⁺ tumor cells begins to attenuate; that is, cells expressing casein become a mixture of casein^hi or casein^lo-expressing cells with varying degrees of FSP1/S100A4 coexpression (Fig. 4, d and e). Eventually, the nodules chiefly comprise casein^lo, GFP⁻ tumor cells (Fig. 4,f). Transferred casein^lo, GFP⁻ tumor cells produced fewer nodules (Fig. 4,g) but those that established themselves in the lung grew in size (Fig. 4,h) and eventually had the same phenotypic features as the casein^lo, GFP⁻ nodules from transferred casein^hi, GFP⁺ tumor cells (Fig. 4 f). More metastases were also seen in liver after the transfer of casein^hi, GFP⁺ tumor cells, particularly in the subcapsular tissue plane (data not shown). Finally, our observation that single cell suspensions of casein^lo, GFP⁻ (FSP1⁻) tumor cells can metastasize, albeit at a reduced rate, suggests our tissue disaggregation protocols confers some temporary relocation properties, after injection, to otherwise nonmigrating cells.

Two previous studies looked at overexpression of S100A4 protein in murine mammary cancer. In the first study, the expression of S100A4 in multiparous MMTV-neu-transgenic mice produced a slightly reduced breast tumor latency and a more invasive appearance (44). In a second report, mice transgenic for S100A4 placed under the control of MMTV increased the number of secondary tumors in the lungs of GRS/A mice (45). To better determine whether the S100A4 protein is simply a marker for metastatic disease or has a causal role in facilitating cell dispersion, we crossed PyV-mT tumor mice with mice carrying null alleles for FSP1 (FSP1^GFP/GFP) and followed female progeny and their littermate controls for development of primary mammary tumors and lung metastases.

FSP1^GFP/GFP mice were constructed by replacing parts of the second and third exons of FSP1/S100A4 with in-frame sequences encoding GFP (Fig. 1,a) so that the fibroblast lineage would be absent FSP1 but marked green (GFP⁺) on fluorescent microscopy in null and heterozygotic progeny. Heterozygotic knockin mice were intercrossed and subsequent genotyped by Southern blot or PCR amplicons (Fig. 5,a). When fibroblasts isolated from the FSP1^GFP/GFP and FSP1^+/GFP genotypes were stained with antibody to FSP1, FSP1^GFP/GFP mice demonstrated complete absence of a cellular reaction product (Fig. 5, b and c). FSP1^GFP/GFP mice produce normal size litters and age well; naïve tissues from brain, heart, liver, lung, kidney, spleen, prostate, and skin appear similar to FSP1^+/+ littermates or FSP1^+/+.GFP mice on review of histology. GFP⁺ fibroblasts from FSP1^GFP/GFP progeny are apparent in the predicted interstitial regions of epithelial tissues (data not shown).

When primary mammary tumors were examined at ∼26% primary tumor burden in female PyV-mT × FSP1^GFP/GFP mice, we observed a mixture of GFP⁺, casein⁻ fibroblasts and casein^hi, GFP⁺ tumor cells between adjacent nodules of primary mammary tumor (Fig. 5, d and e), much like that observed in Fig. 1,g, only to a lesser degree. Both in normal tissues (data not shown) and under the stress of oncogenesis, there appears to be a GFP⁺ fibroblast lineage in FSP1^GFP/GFP mice. With an apparently normal gross phenotype, we nevertheless observed that female PyV-mT × FSP1^GFP/GFP and PyV-mT × FSP1^+/GFP mice produced significantly less lung metastases when compared with PyV-mT × FSP1^+/+.GFP females (P ≤ 0.04 and P ≤ 0.01, respectively; Fig. 5,f), a 60% decrease in the number of metastases. Because PyV-mT × FSP1^GFP/GFP females had fewer numbers of lung nodules, we interpret the absence of FSP1/S100A4 as a conditional effect on metastatic latency. Confirmation that the number of casein^hi, GFP⁺ internodular tumor cells/high-power field in PyV-mT × FSP1^GFP/GFP and PyV-mT × FSP1^+/GFP females were significantly reduced when compared with PyV-mT × FSP1^+/+ or PyV-mT × FSP1^+/+.GFP mice (P ≤ 0.01 and P ≤ 0.01, respectively; Fig. 5 g) supports this hypothesis. These reduced numbers of casein^hi, GFP⁺ tumor cells between primary nodules of PyV-mT × FSP1^{+/GFP or GFP/GFP} mice are a reflection of a reduced metastatic phenotype.

Epithelia develop into clonal tumors through a series of genetic or epigenetic events (46). Phenotypic heterogeneity among these clones is expected because sequential mutagenesis distributes new traits unevenly among subsequent tumor progeny (47). Thus, if only a small fraction of cancer cells achieve the capacity to metastasize, one can rightfully ask if this trait is preordained by earlier mutations in a multistep process? The answer appears to be no. Although it has been relatively easy to recognize mutations or viral components that favor cell cycle growth or anomalous differentiation, it has been difficult to identify simple mutations that predict leaving (46). Cells leaving a primary tumor nodule as metastases depend on a number of biochemical and signaling events emblematic of a molecular program similar to EMT (3, 48, 49). These transitions are conditional, and any tumor cell, in theory, is capable of migrating if exposed to an appropriate mix of morphogenic cues after basement membrane disruption (50, 51) and local cytokine stimulation (2, 3).

We propose such transitions in tumor cells are the result of molecular exaptation. Molecular exaptation is a form of economy by which cells reuse known physiological processes to provide new functions (32). In the present circumstance, epithelia that normally use FSP1-directed EMT to become fibroblasts rely on the same molecular program to metastasize when they convert from in situ to invasive tumor cells. These transitions are reversible in a new environment (30), and release from exaptation permits the development of secondary tumor nodules. This reversal of EMT may be comparable with the physiological process of mesenchymal-epithelial transition (52).

An increase in metastatic latency in FSP1^GFP/GFP and FSP1^+/GFP haploinsufficient mice suggests a role for FSP1 in enabling motility. This may result from FSP1’s role in the reorganization of cytoskeletal components (11) or from FSP1’s ability to bind the COOH-terminal and physiologically sequesters wild-type p53 (21) with attendant effects on cell division, senescence, or apoptosis in breast cancer (53). p53 activity also attenuates cell motility (54), and reductions in levels of FSP1 may link p53 to increases in metastatic latency. In vitro studies of FSP1/S100A4 on p53 function in tumor cells, however, suggests a more complicated picture that may depend on a changing cellular environment or the mutational status of p53 (20). Further study p53 will be needed to clarify mechanism.

Finally, our results directly link the expression of FSP1 to the development of metastatic potential in primary tumors. Beyond that, we have established that FSP1 is not simply a marker of latency but has a role in the process. We are now in a position to investigate the role of FSP1 at the cellular or gene level. The phenotypic shift of EMT is transcriptionally driven (4, 55), involving a predictable transcriptome for program proteins, including FSP1, c-myc, c-Fos, H-ras, Slap, Snail, transforming growth factor β, fibroblast growth factors 1, 2, and 8, vimentin, α-smooth muscle actin, aggrecan, collagen types I and III, thrombospondin I, and matrix metalloproteinases 2 and 9, to name a few. The regulating cis-acting element, FTS-1, in the FSP1 promoter (55) also commonly appears as an element for the above genes related to EMT. The coexpression of FSP1 with casein proteins in metastatic mammary cells may also be related to an FTS-1-driven expression (data not shown). Using this cohort of markers, we can now evaluate the comparison of tumor progression with EMT and gain a greater understanding of mechanism of progression and its control.

We thank Drs. Hal Moses, Lynn Matrisian, Carlos Arteaga, Ray DuBois, and Jennifer Pietenpol in the Vanderbilt-Ingram Cancer Center for comments on earlier versions of this manuscript.