Kinetics of Metastatic Breast Cancer Cell Trafficking in Bone

Purpose:In vivo studies have focused on the latter stages of the bone metastatic process (osteolysis), whereas little is known about earlier events, e.g., arrival, localization, and initial colonization. Defining these initial steps may potentially identify the critical points susceptible to therapeutic intervention.

Experimental Design: MDA-MB-435 human breast cancer cells engineered with green fluorescent protein were injected into the cardiac left ventricle of athymic mice. Femurs were analyzed by fluorescence microscopy, immunohistochemistry, real-time PCR, flow cytometry, and histomorphometry at times ranging from 1 hour to 6 weeks.

Results: Single cells were found in distal metaphyses at 1 hour postinjection and remained as single cells up to 72 hours. Diaphyseal arrest occurred rarely and few cells remained there after 24 hours. At 1 week, numerous foci (2-10 cells) were observed, mostly adjacent to osteoblast-like cells. By 2 weeks, fewer but larger foci (≥50 cells) were seen. Most bones had a single large mass at 4 weeks (originating from a colony or coalescing foci) which extended into the diaphysis by 4 to 6 weeks. Little change (<20%) in osteoblast or osteoclast numbers was observed at 2 weeks, but at 4 to 6 weeks, osteoblasts were dramatically reduced (8% of control), whereas osteoclasts were reduced modestly (to ∼60% of control).

Conclusions: Early arrest in metaphysis and minimal retention in diaphysis highlight the importance of the local milieu in determining metastatic potential. These results extend the Seed and Soil hypothesis by demonstrating both intertissue and intratissue differences governing metastatic location. Ours is the first in vivo evidence that tumor cells influence not only osteoclasts, as widely believed, but also eliminate functional osteoblasts, thereby restructuring the bone microenvironment to favor osteolysis. The data may also explain why patients receiving bisphosphonates fail to heal bone despite inhibiting resorption, implying that concurrent strategies that restore osteoblast function are needed to effectively treat osteolytic bone metastases.

Breast cancer has a remarkable predilection to colonize bone, with an incidence between 70% and 85% in patients (1–3). At the time of death, metastatic bone disease accounts for the bulk of tumor burden (4). For women with bone metastases, the complications—severe, often intractable pain, pathologic fractures, and hypercalcemia—are catastrophic. Despite its obvious clinical importance, very little is understood about the fundamental mechanisms responsible for breast cancer metastasis to bone. Research progress has been hampered by the dearth of, and technical difficulties inherent in, the current models.

Most models of metastasis poorly recapitulate the pathogenesis of breast cancer. The ideal model would involve dissemination from an orthotopic site (i.e., mammary fat pad), colonization, and osteolysis. None of the currently available human breast xenograft models spread to bone following orthotopic implantation and only one murine model metastasizes to bone from the mammary fat pad (5). Furthermore, most human cell lines do not metastasize to bone in mice regardless of route of injection. The most commonly used model of breast cancer metastasis to bone involves injection of tumor cells into the arterial circulation via the left ventricle of the heart (4, 6–8). This route of injection minimizes first-pass filtration through pulmonary capillaries, thereby allowing more cells to reach the bone.

Current methods to detect bone metastases are insufficiently sensitive (e.g., radiography) or are impractical for adequately statistically powered experiments because of costs or labor-intensiveness. Radiography can detect osteolytic lesions only after more than half of the calcified bone matrix has been degraded (9). Microcomputerized tomography is not widely available, but is likewise of insufficient resolution to recognize single tumor cells. Serial sectioning (which would be required to locate rare single cells) is cost-prohibitive, except for small studies. As a result, experiments have been limited to late events of metastatic bone disease, such as osteolysis. Therefore, antecedent events (i.e., arrival, lodging, intraosseous trafficking, and colonization) have not been studied except by inference.

To overcome some of the technical limitations, more sensitive methods using reporter molecules, such as luciferase (10) or β-galactosidase (LacZ) have recently been described (11–13). Luciferase, although it allows for in situ detection of tumor cells in the bone, does not allow for microscopic localization of the cells. Because luminescence depends on a fully viable cell, use of luciferase is limited ex vivo. β-Galactosidase is excellent for studies at the histologic level but cannot be used for studies involving intact bone unless the lesions are macroscopic. Diffusion or distribution of substrate into bone is also a complication.

Fluorescent molecules, like enhanced green fluorescent protein (GFP), have also been employed with some success in the early detection of bone metastasis (14–16). We recently used the GFP-tagged MDA-MB-435 metastatic human breast cancer cell line to reveal formation of osteolytic bone lesions following intracardiac injection in athymic mice (15). Like luciferase, GFP can be used to detect lesions in situ, even though the limits of detection are restrictive (∼0.5-1 mm). During experiments designed for other purposes, we detected single tumor cells in bone within minutes postinjection. Because to the best of our knowledge, no one had ever systematically studied the earliest tumor cell-bone interactions (except by serendipitous histologic sections), we decided to use the power of GFP to begin addressing the early events associated with breast tumor cells that have already disseminated to bone.

It has long been recognized that, once cells arrive in the bone, they alter homeostasis. Turnover of the skeleton is dynamic and continuous throughout embryonic development and adulthood. Calcified bone matrix turns over completely, on average, every decade (17, 18). Calcified matrix remodeling involves an interplay between osteoblasts (bone forming cells) and osteoclasts (bone resorbing cells). Altering the balance of activities results either in excessive bone deposition (osteopetrosis) or bone loss (osteoporosis). Although larger individual bone lesions contain regions that are both osteopetrotic and osteoporotic, most breast cancer bone metastases are not osteolytic. The current paradigm suggests that tumor cells influence osteoclast activity (4, 19). Using the GFP model of breast cancer metastasis to bone, we sought to identify key tumor cell–bone cell interactions (and the timing of those interactions) that occur during the pathogenesis of bone metastasis.

Cell lines and culture. Metastatic human breast carcinoma cell line, MDA-MB-435 (MDA-435), a generous gift from Dr. Janet Price (University of Texas M.D. Anderson Cancer Center, Houston, TX), was stably transfected with pEGFP-N1 (BD Biosciences Clontech, Palo Alto, CA) by electroporation (Bio-Rad Model GenePulser, Hercules, CA; 220 V, 960 μ Fd, ∞Ω) or transduced with a HIV type 1-based, lentiviral vector system constitutively expressing enhanced GFP (20, 21). For the lentivirus, the GFP coding sequence was inserted into the vector 5′ of the internal ribosome entry site and puromycin sequences, each of which were under control of the early cytomegalovirus promoter. Infectious stock were prepared by transfection of 293T cells and used at a multiplicity of infection of ∼10.

The origin of MDA-MB-435 has been questioned because the cells express melanoma-associated genes in cDNA microarray experiments. However, the patient was reported only to have a breast carcinoma. Because MDA-MB-435 cells express milk proteins (22), it is most simple to conclude that the cells are poorly differentiated breast carcinoma.

Parental cells were cultured in a mixture (1:1 vol/vol) of DMEM and Ham's F12 media (DMEM/F12; Invitrogen, Carlsbad, CA) supplemented with 2 mmol/L l-glutamine, 1 mmol/L sodium pyruvate, 0.02 mmol/L nonessential amino acids, 5% fetal bovine serum (Atlanta Biologicals, Norcross, GA), without antibiotics or antimycotics (cDME/F12). All cultures were confirmed to be negative for Mycoplasma spp. infection using a PCR-based test (TaKaRa, Shiga, Japan).

GFP-expressing cells were grown in cDME/F12 plus G418 (Geneticin, 500 μg/mL, Invitrogen) or puromycin (500 μg/mL, Fisher Scientific, Hampton, NH). The brightest 15% (lentiviral) or 25% (pEGFP) fluorescing cells were sorted using either Coulter EPICS V cell sorter (Beckman-Coulter, Fullerton, CA) or a BD FACSaria cell sorter (BD Biosciences Immunocytometry Systems, San Jose, CA).

Intracardiac injections. Cells at 80% to 90% confluency were detached using a mixture of 0.5 mmol/L EDTA and 0.05% trypsin in Ca⁺²-, Mg⁺²-, and NaHCO₃-free HBSS. Viable cells were counted using a hemacytometer and resuspended at a final concentration of 1.5 × 10⁶ cells/mL in ice-cold HBSS. Cells were not used unless viability was >95%, but was usually >98%. Female athymic mice ages between 4 and 6 weeks (Harlan Sprague-Dawley, Indianapolis, IN) were anesthetized by i.m. administration of a mixture of ketamine-HCl (129 mg/kg), and xylazine (4 mg/kg). Cells (3 × 10⁵ in 0.2 mL) were injected into the left ventricle of the heart between the third and fourth or between the fourth and fifth intracostal space. The presence of bright red, as opposed to burgundy, colored blood prior to and at the end of each inoculation confirmed injection of the entire volume into the arterial system. Mice were necropsied at 1, 2, 4, 8, 24, 48, and 72 hours and 1, 2, 4, and 6 weeks postinoculation following anesthesia with ketamine/xylazine and euthanasia by cervical dislocation. At least two independent experiments were done with 5 to 12 mice per experimental group. Not all time points were collected for every experiment.

Although widespread skeletal metastases develop after intracardiac injection (15, 23), the experiments reported here focused exclusively on the femur, a common site for metastasis that is easily accessible. The femurs were removed and examined by low magnification (×2-10) fluorescence stereomicroscopy and histologic and histomorphometric analyses (24, 25). Some femurs were divided into proximal and distal metaphyses plus cortical shaft (diaphysis) from which the marrow was collected and cells examined by flow cytometry or quantitative real-time PCR. Corroborating experiments were done with the contralateral femur to assure that there was no bias for sidedness.

Mice were maintained under the guidelines of the NIH, the University of Alabama at Birmingham, and the Pennsylvania State University. All protocols were approved and monitored by the appropriate Institutional Animal Care and Use Committees.

Fluorescence microscopy. To visualize metastases derived from the GFP-tagged cell lines, whole femurs (dissected free of soft tissue using a no. 11 scalpel blade with gauze used to grip and remove tissue remnants) were placed into Petri dishes containing ice-cold Ca⁺²- and Mg⁺²-free Dulbecco's PBS and examined by fluorescence microscopy using a Leica MZFLIII dissecting microscope with ×0.5 objective and GFP fluorescence filters (λ_excitation = 480 ± 20 nm; λ_emission, 510 nm barrier; Leica, Deerfield, IL). Photomicrographs were collected using a MagnaFire digital camera (Optronics, Goleta, CA), and ImagePro Plus 5.1 software (Media Cybernetics, Silver Spring, MD).

Bone fixation and storage. Intact, dissected femurs from individual mice were placed in 25 mL glass scintillation vials and fixed in freshly prepared 4% paraformaldehyde in Ca⁺²- and Mg⁺²-free Dulbecco's PBS or in periodate-lysine-paraformaldehyde solution (26) at 4°C for 24 to 48 hours. GFP fluorescence was difficult to maintain in fixed tissues and bone sections; however, we were able to overcome this limitation by maintaining the samples at 4°C (27). Bones destined for histologic sectioning were subsequently removed and decalcified in 0.5 mol/L EDTA in Ca⁺²- and Mg⁺²-free Dulbecco's PBS.

Bone histomorphometry. Bones were dehydrated in increasing concentrations of ethanol and embedded in a mixture of 80:20 methyl methacrylate and dibutylphthalate. Serial coronal sections (5 μm) were obtained using a Leica 2265 microtome. The distal ends of femurs (spongiosa) were analyzed. Sections were first stained with Sanderson's rapid bone stain for 2 minutes. Once tumor cells were identified, subsequent sections were stained with Goldner's trichrome and tartrate-resistant acid phosphatase (TRAP). Histomorphometry was done at the University of Alabama at Birmingham Center for Metabolic Bone Disease Histomorphometry and Molecular Analysis Core Facility by the method of Parfitt et al. (24, 25) using Bioquant image analysis software (R&M Biometrics, Nashville, TN).

Immunohistochemistry. Paraffin-embedded samples were sectioned (5 μm, coronal or sagittal), deparaffinized, and rehydrated before antigen retrieval by microwaving for ∼8 minutes at full power (700 W) in a 10 mmol/L citrate buffer (pH 6). Samples were boiled for 5 minutes in the microwave oven. Endogenous peroxidase activity was blocked by treatment with 3% hydrogen peroxide for 5 minutes. Sections were blocked with 1% goat serum for 1 hour. Slides were incubated with primary rabbit polyclonal anti-GFP IgG (1:250; Molecular Probes, Eugene, OR) for 1 hour, followed by secondary biotinylated anti-rabbit antibody (TITRE; Level 2 Ultra Streptavidin Detection System, Signet Labs, Dedham, MA). Detection was achieved using Biogenex liquid DAB kit (Biogenex, San Ramon, CA) and slides were counterstained using hematoxylin. GFP-positive tumor samples served as positive controls. Negative controls were done by omitting the primary antibody.

Some femurs were fixed in a solution of 2% paraformaldehyde containing 0.075 mol/L lysine and 0.01 mol/L sodium periodate (pH 7.4), 4°C for 24 hours in an attempt to maintain alkaline phosphatase activity (26). Although the alkaline phosphatase activity was not well preserved, fluorescence was maintained; fluorescent MDA-435^GFP cells could be observed in femurs taken throughout the time course. Following decalcification as described above, the bones were embedded in paraffin. Paraffin-embedded bones were sectioned lengthwise into 10 μm sections and several sections throughout the bone were analyzed. The sections were deparaffinized, rehydrated, and stained for apoptotic cells with a modified terminal nucleotidyl transferase–mediated nick end labeling (TUNEL) procedure using Cy-5 rather than FITC-labeled dUTP (28). Bone sections were first scanned at ×20 magnification using a fluorescence confocal microscope. Areas in which GFP-positive cells were detected were further analyzed at ×40 magnification with both fluorescence and phase microscopy. Fluorescent images were captured at two wavelengths λ_excitation = 480 ± 20 nm (λ_emission, 520 nm for GFP) and λ_excitation = 647 nm (λ_emission, 670 nm for Cy-5). A comparison of the numbers of breast cancer cells detected by fluorescence microscopy versus use of anti-GFP gave essentially the same trends (data not shown).

TRAP-positive cells were determined in the femurs of mice at various times following inoculation with metastatic breast cancer cells. Two to eight sections from two to four bones per time period were stained for the presence of TRAP by immunohistochemistry (Sigma-Aldrich, St. Louis, MO). After staining, the sections were viewed with a fluorescent light microscope at ×20 magnification. Three images (1,349 pixels/500 μm) from the distal end and three images from the proximal end were collected and converted into JPEG format. The number of TRAP-positive cells was counted in each image. The Image J program (NIH) was used to calculate the bone area in each field.

Some decalcified femurs were cryosectioned and stained for alkaline phosphatase activity (Sigma-Aldrich). The sections were examined with a light microscope, the images digitally collected, and analyzed for the amount of alkaline phosphatase stain per area of bone at the growth plate and in the trabecular region. The data were calculated as ratio of the mm² of the alkaline phosphatase stain to mm² of bone.

Flow cytometric and DNA analysis. Femurs were removed from five to six mice at each time and cut into the distal and proximal metaphyses and the diaphysis. Bone marrow was flushed from these regions with a 1 mL tuberculin syringe fitted with a 26-gauge needle. The marrow in the center of the diaphysis was collected separately from the endosteal marrow close to the cortical bone as previously described (29). For flow cytometry, the RBC were lysed with ACK solution (15 mmol/L NH₄Cl, 1 mmol/L KHCO₃, 0.1 mmol/L Na₂ EDTA) and the remaining cells were fixed with 2% paraformaldehyde. Samples were stored at 4°C until they were analyzed by flow cytometry (Coulter XL-MCL) using standardized fluorescent beads (10 μm; Sphero AccuCount Rainbow Fluorescent particles, Spherotech, Libertyville, IL) to estimate the total number of cancer cells present. Standard curves were also generated by adding known numbers of MDA-MB-435 cells to mouse bone marrow cells. The samples of the mixtures of cells were prepared in the same way as the experimental samples. A background value of 200 cells was determined from the data obtained from the control animals in which no GFP-positive cells were present.

For DNA analysis, marrow was centrifuged and frozen in Ca⁺²- and Mg⁺²-free Dulbecco's PBS. At a later time, DNA was prepared from the samples with a DNeasy kit (Qiagen, Valencia, CA). The DNA was subjected to quantitative real-time PCR (Nucleic Acid Facility, Penn State, University Park, PA) using primers to detect the HERVK gene (human endogenous retrovirus, group K), a gene found in the human but not in the mouse genome (30). To establish a standard curve, MDA-435^GFP cells were counted, diluted, and added to preparations of mouse bone marrow cells. DNA was isolated from these samples and treated as the experimental samples for PCR. Although one cell could be detected in the standard curve samples, a more conservative cutoff of 150 cells was used due to practical considerations, i.e., cell extract volumes and variations in the amount of mouse DNA present in each sample.

Statistics. Each series of injections involved between 5 and 15 mice per experimental group or time. Femurs were apportioned for various subsequent analyses. Comparisons between groups were done by one-way ANOVA with Student-Neumann-Kuels or Tukey's post-tests. Statistical significance was defined as a probability P ≤ 0.05.

Kinetics of MDA-435^GFP tumor cells trafficking in the bone. Numerous solitary fluorescent cells could be visualized in the intact bone (i.e., the bone is not cut, but has been stripped of surrounding muscle) 1 hour following intracardiac injection using fluorescence microscopy (Fig. 1A1). The majority (>90%) of cells were found in the metaphyseal regions, not in the diaphysis, by fluorescence. Routine identification of single tumor cells using H&E stain, although possible, proved difficult. Even with evidence that tumor cells were present in the bone (i.e., by fluorescence), we could not always unambiguously identify single tumor cells in histologic sections stained by H&E.

To facilitate detection of solitary tumor cells by histology and to determine their positions within the trabeculae, cellular location was estimated in two dimensions using fluorescence. Then, serial histologic sections were cut from the regions exhibiting fluorescence. Although this manipulation increased the odds of finding sections containing single cells, it was not always possible to detect cells in every 5 to 10 μm section stained using anti-GFP antisera. Nonetheless, as implied by the fluorescence data, most tumor cells were located in the primary spongiosa of the metaphysis of the distal femur (Fig. 1B and E). Although fluorescent cells were not frequently detected in the femoral head at 1 hour postinjection, some GFP-positive cells were detected by immunohistochemistry (data not shown). There was no evidence of unusual inflammation or immune cell infiltration at the sites of tumor cell arrest or colonization.

In a third, parallel approach, we detected cells without visualization constraints or sampling errors associated with sectioning, instead quantifying tumor cells in various marrow compartments using flow cytometry or real-time PCR. Separation of the marrow from metaphyses and diaphysis followed by flow cytometry or real-time PCR to detect MDA-435^GFP cells revealed a slightly different pattern of tumor cell distribution within the bone at the early times, but an entirely consistent pattern of distribution at the later times (Fig. 2). These methods were limited because precise separation of the diaphysis from the metaphysis was not consistent. Therefore, cancer cells at the interface between bone regions could not be localized with certainty. Histologic sections confirmed precise localization (Fig. 1). As a whole, the various methods to assess localization were largely confirmatory. In addition, we also determined that cancer cells in the diaphysis were mostly located next to the endosteal bone rather than in the marrow in the center of the shaft. By 4 to 6 weeks, >95% of tumor cells were found in the endosteal marrow (Fig. 5).

Of the 3 × 10⁵ cells injected per mouse, a small fraction were detectable in the femurs (Table 1), as expected, because cells are distributed throughout the body following injection into the arterial circulation. Flow cytometry and real-time PCR were used to quantify the number of cells present. Cells were flushed from the marrow space in the metaphyseal and diaphyseal regions and examined by flow cytometry to detect GFP-tagged cells (Fig. 2). In addition, DNA isolated from the bone marrow was analyzed for the presence of human DNA by real-time PCR using a human-specific primer/probe set (Fig. 2). Regardless of the method, seeding of the femurs with breast cancer cells was rare (44 ± 6 cells; 0.01% per femur; Table 1). As a result, the absolute numbers were highly variable between mice and between techniques. Thus, the total number of single cells identified was not sufficient to perform statistical analyses with adequate power. Nonetheless, we did observe several consistent changes. First, solitary cells in the diaphysis were seldom detected beyond 24 hours in any of the mice, until metaphyseal lesions had apparently extended into the diaphyses at later times (Figs. 1G and 2). Second, the number of fluorescent foci (i.e., cell masses) detected by fluorescent stereomicroscopy decreased progressively beginning from 1 to 72 hours. This result is consistent with the clearance of disseminated tumor cells from other organs (31, 32). Third, single cells persisted in the femur for up to 72 hours. In general, evidence of cell division prior to 72 hours postinoculation was infrequent.

Whereas initial proliferation of arrested cells was delayed, distinct metastatic foci (5 ± 1) were easily detected by fluorescence microscopy of the intact bone (Fig. 1A2) as well as immunohistochemistry in the femur at 1 week (Fig. 1B2) and fluorescence microscopy of bone sections (Fig. 1E). Most of the foci were small and consisted of <10 cells (i.e., only three to four cell divisions). Although most metastatic lesions were localized at the distal end, a fraction of the bones had fluorescent foci growing at the proximal ends as well.

By 2 weeks, larger but fewer foci (2.8 ± 0.5) were detected at the distal end (Fig. 1A3 and B3). The foci were comprised of clusters of ∼50 cells. These progressively increased in size and decreased in number to an average of one metastatic focus at 4 weeks, presumably by coalescence (Fig. 1A4, B4, and G). By 6 weeks, tumor cells directly extended into the diaphysis, and in some cases, the whole medullary canal was occupied by tumor. Histomorphometry of the lesions revealed loss of most trabecular bone by 4 to 6 weeks (compare Fig. 1D and Fig. 1C).

Tumor cell modification of the bone microenvironment. Histomorphometric analysis further showed specific modification of the microenvironment when tumor cells were present (Fig. 3). We had previously shown that MDA-435^GFP cells form radiographically detectable osteolytic lesions within 4 to 6 weeks following intracardiac injection (15). That finding was corroborated by histomorphometry showing that calcified bone volume decreased as tumor volume increased (Fig. 3A). A decrease of 97% in the ratio of osteoid surface to bone surface at the 4-week point to osteoblast loss or loss-of-function as a major contributor to the decrease in calcified bone volume. This finding is consistent with previous work showing that MDA-MB-435 and MDA-MB-231 cells induce osteoblast apoptosis (33) and retard osteoblast differentiation (34) in vitro.

Importantly, osteoblast number per trabecular bone surface area in the metastatic lesions decreased with time (Fig. 3B). Bones from uninjected, age-matched mice served as negative controls. By 2 weeks, the number of osteoblasts decreased by ∼20% in tumor-bearing mice. By 4 weeks, however, the decrease was more dramatic (∼92% decrease; Fig. 3B). The decrease in osteoblast number was accompanied by an increase in the number of apoptotic osteoblasts observed by TUNEL (Figs. 3D and G, and 4A). Apoptotic osteoblasts were found at both proximal and distal ends of the femur, but little change in the number of apoptotic osteoblasts was observed in the diaphysis until tumor cells were routinely observed in that portion of the bone (Fig. 4A). This finding suggested that osteoblast apoptosis was occurring mostly when tumor cells were present. Supporting this hypothesis, TUNEL-positive osteoblasts were found almost exclusively in the presence (≤50 μm) of GFP-positive breast cancer cells (Figs. 3G and 4B). Previous (33) and current experiments (data not shown) have shown almost no tumor cell apoptosis when adjacent to osteoblasts. Bone sections stained for alkaline phosphatase, a marker for osteoblasts, showed a dramatic decrease as tumor burden increased (Fig. 3H-J). The ratio of the area occupied by alkaline phosphatase-positive cells per area of trabecular bone in the femurs of cancer-bearing mice was significantly lower compared with healthy mice (2.0 ± 0.6 versus 9.6 ± 1.7, respectively; P ≤ 0.0001). In contrast, alkaline phosphatase activity in chondrocytes located in the growth plate was not significantly different (control, 25.9 ± 4.7; tumor bearing, 24.1 ± 5.0).

The number of osteoclasts remained unchanged at 2 weeks, but unexpectedly, a consistent decrease (∼35-40% or to 60% of control) in osteoclast number was observed at later times (Fig. 3C). This difference was evident by both quantitative histomorphometric analysis (Fig. 3C) and TRAP staining (Fig. 3E and F). The absolute number of both cell populations decreased so that the ratio of osteoblasts to osteoclasts decreased from an average of 40 to only 4 by 4 weeks.

Bone is the most common site for metastases from breast carcinomas and their sequela account for approximately two-thirds of the costs associated with treating women with the disease (4, 36). As with most metastases, symptoms occur relatively late in disease progression. Whereas prevention of metastases altogether is ideal, restriction of progression to an asymptomatic state would improve clinical management of breast cancer. Likewise, repair of already existing lesions would benefit patients whose disease progression has been halted. As a result, understanding the antecedent steps for bone metastasis and osteolysis should provide insights for developing therapeutic interventions.

The results presented here are, to the best of our knowledge, the first to describe the behavior of breast cancer cells at the earliest times after they have arrived in the femur. Although the femur may not represent the behavior of tumor cells in all bones, it is a common site of secondary colonization both in patients with breast cancer and in experimental models. Therefore, we considered it an appropriate site for studying the process.

Single tumor cells were detected in the femur 1 hour after introduction into the arterial circulation (Fig. 5B). It is noteworthy that, even at this early time, tumor cells arrested primarily in the metaphyses rather than diaphyses. Although it is possible that differential expression of adhesion molecules may be found in metaphyseal versus diaphyseal bone vasculature (reviewed in ref. 37), anatomic and physiologic mechanisms are also likely to be important factors. Approximately 90% of the blood flow goes to the metaphyseal regions whereas a smaller fraction is found in the diaphyses (reviewed in ref. 38). Additionally, blood flow in the diaphysis is still largely vessel-based, whereas in the metaphyses, it is more sinusoidal. In the sinusoids, the rate of blood flow is <10% of that found in capillaries or other vessels (reviewed in ref. 37). Because of the sluggish blood flow, weaker adhesion molecules would not be subject to negative selection as when cells experience stronger sheer forces. However, arrest is not the only variable involved. The relatively rare cells initially seeding the diaphysis fail to remain there for prolonged periods. It is also possible that cancer cells follow a gradient of growth factors or cytokines. We have preliminary evidence⁷

that several cytokines are found at much greater concentrations in the metaphysis compared with the diaphysis. The limitations of the present study cannot discriminate between loss of tumor cells to immune killing, apoptosis, sheer forces, etc., versus migration of tumor cells from the diaphysis to the metaphysis.

The earliest arriving tumor cells were mostly located in close proximity to osteoblasts and the bone-lining cells (Fig. 1B). In general, although the animal-to-animal variability was considerable, the majority of cells tended to be in the endosteal marrow, rather than in central marrow (Fig. 5G), suggesting that traversal from the sinusoids to trabecular space occurs relatively rapidly. This pattern is similar to that reported for hematopoietic precursors in bone marrow (29). Regardless of intra-osseous location, the trend was for tumor cell numbers to decline during the first 72 hours after arrival (Fig. 5C and D).

It was somewhat surprising to find that most tumor cells had not begun to proliferate within 72 hours because proliferation typically begins much sooner (i.e., <48 hours) in lungs and orthotopic sites.⁸

The period of quiescence or dormancy could be relevant because recent clinical studies have shown that 20% to 100% of women with breast cancer have evidence of disseminated tumor cells in the skeleton (35, 39, 40). Disseminated occult tumor cells within bone are thought to be in a sanctuary from which tertiary metastases could form and are thought to be responsible for late (sometimes months or years) recurrences. Based on abundant evidence from multiple experimental models, the fraction of disseminated cells progressing to overt metastasis is small. However, the presence of disseminated cells predicts poor prognosis in many tumor types, emphasizing that their presence at non-orthotopic sites should not be ignored (39, 40).

What triggers the conversion from dormant to proliferative cells remains unknown and is the subject of intense current investigation. The data presented here do not clarify the mechanistic issue, except to note that there is a substantial delay (72 hours) before the breast carcinoma cells begin to divide. Perhaps, during that time, tumor cells are altering the microenvironment. At our limits of detection, we could not tell whether there was balanced division (i.e., one of two of the progeny died or were eliminated) or whether tumor cells were adapting to the bone microenvironment prior to initiating growth. Upon arriving at a secondary site, tumor cells may remain dormant, undergo a limited number of cell divisions, or continue proliferating to form an overt mass. Presumably, the decisions are based on the response to signal(s) from the local microenvironment. There are self-evident cues that the tissue environment controls the development of bone metastasis based on the patterns of metastasis observed—e.g., metaphyseal > diaphyseal; distal > proximal; endosteal marrow > central marrow. The findings support and extend Stephen Paget's Seed and Soil hypothesis, by which he explains the predilection of breast carcinomas to colonize bone over other tissues (41). In short, Paget posited that tumor cells (seeds) were best suited for growth only in certain tissues (soils). The data presented here indicate that the soil may differ even within individual bones. Indeed the patterns of clinical bone metastases show marked preference for proximal and trabecular bone compared with distal or cortical bone (4). Surprisingly, the metastases arising from injection of MDA-435^GFP cells showed the opposite pattern—a slight, but consistent, penchant for distal, compared with proximal, femur. The pattern was consistent regardless of the method used to detect tumor cells. Hence, trivial experimental variables cannot explain why, in this model, metastases develops in the distal femur with greater frequency. One possibility invokes anatomic differences between mice and humans. Mice have extremely different gaits and mechanical pressures on their joints. Because tumor cells are generally predisposed to colonize injured tissues, biomechanical stresses on the knees in mice could, in part, explain the difference in arrest and colonization patterns between the species.

Notwithstanding preferential arrest/adhesion and growth in the metaphyses, metastatic breast carcinoma cells could still survive and grow in the diaphyses when the metastatic lesions are large enough. This growth pattern may suggest that tumor cells renovate the bone microenvironment sufficiently to reduce (or eliminate?) negative signals or that they can induce the production of positive factors from local, nontumor cells. Whether neoplastic cells have surpassed a threshold number, or whether some negative influence on tumor cells has been overcome, the incapacity to colonize diaphyses is not absolute.

It is widely accepted that metastatic breast carcinoma cells manipulate the bone microenvironment to induce osteolysis. In particular, tumor cell activation of osteoclasts via the “vicious cycle” of secretion of PTHrP or RANK ligand has been implicated in bone resorption (4, 19, 42). However, we previously hypothesized that the balance of bone deposition and resorption could be altered as well by reducing osteoblast number or differentiation and/or activity (33). In vitro coculture of metastatic human breast carcinoma cells or their conditioned medium with human osteoblasts resulted in apoptosis of the osteoblasts (33). The in vivo data presented here extend the previous findings by demonstrating that osteoblasts in the regions colonized by MDA-435^GFP are also eliminated.

It is possible that changes in the ratio of osteoblasts to osteoclasts in tumor-infiltrated bone tilts the balance in favor of increasing osteoclastic activity, thereby promoting osteolysis. The dramatic decrease in osteoblasts observed by 4 weeks prompted closer examination of osteoblast apoptosis at multiple times using TUNEL. The number of apoptotic osteoblasts progressively increased until later times, perhaps due to the already decreased number of osteoblasts. Interestingly, at earlier times (<4 weeks) at least one tumor cell was in direct contact (≤50 μm) with each TUNEL-positive osteoblast (Figs. 3E and 4B). Although this finding does not prove that tumor cells directly induce apoptosis, the data are consistent with this hypothesis. We also found with in vitro studies that breast cancer cell–conditioned medium prevented osteoblasts from differentiating as evidenced by lack of production of alkaline phosphatase, osteocalcin, and bone siaoloprotein (34). In this current study, the lack of alkaline phosphatase activity (Fig. 3H and I) may be due, in part, to failure of preosteoblasts to differentiate as well as apoptosis of mature osteoblasts. In either case, the outcome is the same, lack of functional osteoblasts.

Ours is the first in vivo evidence that tumor cells influence not only osteoclasts, as widely believed and showed, but also osteoblasts. The findings may explain, in part, the failure of bisphosphonate-treated patients to repair osteolytic bone lesions (43)—i.e., if there are no osteoblasts to reconstitute the bone, the lesions will remain.

It was surprising that osteoclast numbers were dramatically reduced in late-stage bone metastasis. Perhaps it should not have been because Orr, Mundy, and colleagues previously described that tumor cells themselves (i.e., in the absence of osteoclasts) could resorb bone (44, 45). The published data, although somewhat controversial, and the evidence presented here leave open the possibility that further progression of osteoclast-initiated osteolysis is possible. Furthermore, previous publications have shown significantly diminished osteoblast and osteoclast numbers in late stage breast carcinoma metastasis to bone (46–48). Based on heterogeneity among tumors for multiple variables, molecular mechanisms of bone metastasis may vary while yielding the same end point, osteolysis.

A potential criticism of the reported work is that the findings are based on a single cell line. Given the heterogeneity of tumors and the redundant mechanisms from which each can choose to accomplish a given task, we are careful not to overgeneralize. Nonetheless, most key observations reported here have been replicated in the MDA-MB-231 breast carcinoma model of bone metastasis. MDA-MB-231 cells form osteolytic lesions with similar distribution as those found in MDA-MB-435. With expansion of bone lesions, osteoblast numbers decrease in MDA-MB-231 as well.⁹

Assuming that the essential elements of osteolytic metastasis are observed in multiple breast carcinoma models, the findings reported here have significant implications with regard to control of bone metastasis in the clinic.

Foremost, the osteoblast is key. Whereas tumor cells could initiate growth prior to osteolysis, one of the earliest observed changes is the elimination of bone-forming cells. Even if bone resorption is controlled exogenously (i.e., by treatment with bisphosphamates), repair of defects is not possible. Because the structural integrity of the skeleton is critical to survival and quality of life, comprehensive treatment needs to restore bone matrix as well as limit osteolysis. By studying the trafficking of tumor cells within the bone and the effect of their presence on normal bone physiology, insights regarding how to improve control of bone metastasis will be forthcoming. Since the submission of this article, a report (49) describing a role for Dickkopf-1 in a Wnt-mediated pathway in prostate cancer cell osteoblastic lesions was published. Whether breast cancer–mediated osteolysis via impairment of osteoblast function is regulated by Dickkopf-1 or other components of the Wnt signaling pathway remains to be determined.

We are indebted to the superb technical assistance from Virginia R. Gilman as well as Patti Lott at the University of Alabama at Birmingham, Center for Metabolic Bone Disease Histomorphometry Core Facility, the University of Alabama at Birmingham Breast Specialized Programs of Research Excellence Tissue Core Facility, Deborah Grove and the Penn State Nucleic Acid Core Facility, and the Penn State Electron Microscopy and Histology Core Facility. We appreciate the critical reading of the manuscript and helpful suggestions from Dr. Tom Clemens.