Current research implies the presence of a population of cells in breast tumors with tumor-initiating capabilities. A number of cell surface markers have been used to sort for breast cancer cells with tumor-initiating capacity, including CD44 , CD133 , ALDH [31, 38], CD90 [39, 40], and CD117 (KIT) . However, attempts to eliminate artifact may result in the biased selection of a small sub-population of cells present in disaggregated tissue. Presented in this study is an efficacious model for the in vitro isolation of tumorspheres, containing breast tumor-initiating cells, from human breast core biopsies. Injection of isolated tumorspheres into the mammary fat pad of NUDE mice resulted in formation and maintenance of small, palpable tumors that exhibited elevated proliferation and apoptosis. Within 8 months post-injection, widespread metastasis to mouse organs occurred most notably to liver, lung, and kidney. The cells within the tumors present in the mammary fat pad displayed heterogeneous expression of a range of markers for epithelial and mesenchymal differentiation that included cytokeratins, E-cadherin, β-catenin, fibronectin, as well as Her2/neu. Histological heterogeneity was observed both within individual tumors, and between tumors formed from tumorspheres isolated from different patient samples. Whereas primary tumors did not exhibit E-cadherin staining in cell membranes, a subset of tumor cells that had metastasized to the lung and the liver exhibited a re-distribution of E-cadherin to the membrane. These data describe a novel hetero-transplantation tumor model that exhibits a metastatic profile that recapitulates the development of metastasis in human breast cancer patients.
Tumorspheres were successfully isolated from all breast core biopsies including samples 1-3, which were not included in the data presented due to the lack of tumor formation and development of thymic lymphoma in NOD/SCID mice (as described in the Results section). Samples 5-9 tumorspheres were derived from patients with invasive ductal carcinoma (IDC), similar to other reports investigating the tumorigenic potential of breast cancer cells isolated as tumorspheres in vitro [16, 33]. Although FACS has been used to isolate tumorigenic breast cancer cells from other breast cancer subtypes such as inflammatory breast cancer and lobular carcinoma , there are no reports demonstrating the successful isolation of tumor-initiating cells as tumorspheres from these breast cancer subtypes. When the host was changed to NUDE mice, the methods described herein reproducibly resulted in tumorsphere formation in vitro, and in tumor formation for all samples derived from primary breast core biopsies irrespective of ER/PR/Her2 status, tumor grade or stage. Tumorsphere number and size varied between samples, however differences could not be correlated with ER/PR/Her2 status, tumor stage or grade.
Tumorspheres isolated under select culture conditions likely contain a heterogeneous population of cells most of which will not likely have tumor-initiating capabilities. Therefore tumor formation upon the injection of a minimal number of cells that were not selected on the basis of epithelial origin or differentiation potential indicated the presence of tumor-initiating breast cancer cells within the tumorspheres. The injection of non-disaggregated tumorspheres, as opposed to single cell suspensions, establishes a novel method for the investigation of tumorigenicity in vivo. The absence of mechanical and/or enzymatic stress during preparation of the cells for injection improved the tumor cell viability. Therefore the injection of non-dissociated tumorspheres, in combination with the immune-deficient mouse model employed (NUDE mouse strain), contributed to the high tumor engraftment rates observed. However, future studies should determine differences in tumor formation and metastatic potential between single cell suspensions and non-disaggregated tumorspheres. Differences detected may reveal the effects of cell selection bias based on methodologies employed when preparing cells for injection.
The small primary tumor volume observed in the mammary fat pad suggested a low level of tumor cell proliferation. However the number of proliferating cells detected by Ki67 staining was high in all tumor samples. This high proliferation was offset by a number of cells undergoing apoptosis. This equilibrium between proliferation and apoptosis provided an explanation for the formation and long-term propagation of the small, palpable tumors, indicating the tumors were persisting in a state of tumor dormancy. However it also raised the question as to why the high level of proliferation observed in the tumors did not eventually overcome the apoptosis and result in larger tumor volumes with time. One possible explanation could be the inability of tumors to initiate neoangiogenesis that was manifest in high rates of apoptosis and proliferation that culminated with tumor dormancy [42, 43]. Failure of the injected cells to efficiently recruit mouse stromal cells/endothelial cells would lead to nutrient and oxygen deprivation that would result in high levels of apoptosis to compensate for a high proliferation of the tumor cells. Alternatively, this disruption in homeostasis may occur as a later event during tumor progression, upon sufficient acquisition of somatic mutations within the differentiated progeny. Nonetheless, the metastatic phenotype demonstrated in this study implies that metastasis is an early event in tumorigenesis; conferring implications that challenge the linearity of breast tumor progression.
Traditional xenograft models derived from established breast cancer cell lines (e.g. MCF-7, MDA-MB-231) display fairly homogeneous organization, morphology, and protein expression patterns. In contrast, characterization of the tumors derived from primary tumorspheres presented here revealed heterogeneity among samples in morphology and marker expression. Because the tumorspheres exhibited stem-like properties, it might be expected that tumors in mice derived from the tumorspheres would exhibit similar morphology as the patient tumors. Because the tumorspheres exhibited stem-like properties, it might be expected that tumors in mice derived from the tumorspheres would exhibit similar morphology as the patient tumors. Although the human breast tumor samples ranged from grade 1-3, the tumors in the mammary fat pad did not exhibit all of the morphological characteristics of the patient tumors, such as tubule formation and ER and Her2 expression. This discrepancy is likely a result of the microenvironment within the mammary fat pad that is mostly devoid of the stromal and cellular components present in human mammary tissue, and instead is composed predominantly of adipose tissue. It is possible that the inclusion of human stromal/cellular components with the tumorspheres would more accurately recapitulate the microenvironment within human mammary tissue resulting in tumors that might exhibit tubule formation, ER/PR expression and other characteristics of low grade, more differentiated. In general, the edges of the tumors were lined with dense areas of cells that were also observed within the tumors as a dense ring of cells surrounding a more diffuse distribution of tumor cells (Figure 2). Although the expression patterns of most markers used in the characterization of the tumors did not correspond to this evident cellular organization, the expression of β-catenin did correlate with this organization displaying strong expression in the ring of cells encapsulating the diffuse population of cancer cells. All tumors in the mammary fat pad were negative for ERα despite the varied ERα status of the patient samples. During an EMT in breast cancer, E-cadherin expression in the membrane is reported to be lost or re-localized to the cytoplasm [29, 44]. Consistent with a mesenchymal phenotype, in the primary tumors E-cadherin and β-catenin expression was localized in the cytoplasm and nucleus (Figure 3). Given that loss of E-cadherin expression in breast cancer is associated with an EMT and with the metastatic process, the unique expression pattern of E-cadherin in the present tumor model that exhibits a metastatic phenotype warrants further study.
Human metastatic cancer cells were detected by PCR at 3 months post-injection of tumorspheres into the mammary fat pad within the lung, liver, kidney, brain and femur (Figure 4). However the development of large metastatic lesions was observed by H+E staining in the lung, kidney and liver only. Comparison of the organ-specific percent metastasis to the metastatic burden within each organ revealed interesting results. Although sample 7 demonstrated the lowest percent metastasis overall and within the organs analyzed, the percent metastatic burden within each organ was comparable to or more often higher on average as compared to the other samples. In contrast, sample 9 demonstrated the highest percent metastasis overall and within the organs analyzed, however the percent metastatic burden within each organ was minimal compared to the other samples. These conflicting results elucidate a potential shortcoming in the current methods employed for predicting metastatic disease. The ability of cancer cells to metastasize from the primary site to distant sites may not accurately reflect the actual metastatic potential of the cells. Rather, understanding and predicting the adaptation of the cells that have arrived at the distant site of metastasis, may more accurately determine the potential for the development of overt metastatic disease. Samples 5 and 9, the only tumor samples derived from patient biopsies with clinical diagnosis of triple negative (ER-/PR-/Her2-), demonstrated the highest average metastatic burden within the organs. This observation may imply a growth advantage at distant sites for metastatic cells derived from triple negative tumors, although additional samples would be needed to demonstrate statistical significance. Despite the detection of metastasis by H+E staining in 60%-100% of brains analyzed, the metastatic burden was on average between only 1-3%. The blood-brain barrier (BBB) consists of tight junctions and adherens junctions between the brain endothelial cells, restricting the passage of substances from the bloodstream into the brain . Impedance of entry into the brain by the BBB may contribute to the apparent extended dormancy period of the metastatic cells present in the brain, in combination with other influencing factors. Overall, the long duration from the time of detection of cancer cells in the organs (by PCR and/or histological staining) to the development of visual macrometastasis in mouse organs was similar to the delay in development of measurable metastasis in breast cancer patients following diagnosis of primary breast tumors [46, 47]. The extended latency between the injection of tumorspheres and the development of macroscopic metastatic lesions is a limitation of this model. The injection of a higher number of cells may increase the time to tumor formation and the tumor volume in the mammary fat pad, however we speculate that the observed metastatic latency would not be affected. The organ microenvironment and intrinsic properties of the disseminated cells likely predominantly contribute to the observed metastatic latency. However secreted factors from the primary tumor within the mammary fat pad could also influence metastatic progression, therefore an increase in tumor volume may affect the proliferative state of disseminated tumor cells at distant sites. Future studies investigating the effects of larger tumor volumes on metastatic latency, possibly by the injection of increasing numbers of cells into the mammary fat pad, may elucidate the influence of secreted factors from the primary tumor on disseminated cancer cells. The detection of micrometastases in an array of organs with the development of macrometastases in only a select few of those organs suggests that mechanical/stochastic forces may permit entry of cancer cells into a range of organs; however metastatic cells will only survive and develop into overt lesions when present within a permissive, conditioned metastatic niche .
Whereas cytoplasmic and nuclear localization of E-cadherin and β-catenin was observed in the primary tumors, E-cadherin and β-catenin were re-expressed in the membrane in a subset of metastatic cells in the lung suggesting a possible reversion back to an epithelial phenotype (a mesenchymal to epithelial transition) [36, 48]. E-cadherin and β-catenin expression was predominantly observed in the membrane of metastatic cells proximal to hepatocytes in the liver. Furthermore, hepatocytes proximal to metastatic cancer cells demonstrated stronger E-cadherin expression as compared to hepatocytes not within close proximity to metastatic cancer cells. Chao et al. demonstrated re-expression of E-cadherin in the membrane of MDA-MB-231 breast carcinoma cells in vivo and in vitro when in close proximity to hepatocytes . Additionally, in prostate cancer models of metastasis to the liver, E-cadherin was shown to accumulate at the interface with hepatocytes [50, 51]. These findings further support the importance of crosstalk between the cancer cells and native cells present within the organ. Fibronectin, a component of the extracellular matrix (ECM), plays an important role in cell migration, adhesion, maintenance of cell shape and wound healing. Fibronectin-dependent signaling has been linked to cancer cell dormancy, involved in the quiescent to proliferative switch [52, 53]. Many of the smaller metastatic lesions in the liver did not demonstrate expression of fibronectin, however fibronectin was found variably expressed within the larger metastatic lesions. This observation could implicate fibronectin signaling as a possible mechanism involved in the development of macrometastatic lesions within the liver. The metastatic lesions within the liver were negative for ERα in all samples, consistent with the lack of expression within primary tumors. However in samples 6 and 8, metastatic lesions in the lungs of mice bearing primary tumors demonstrated variable re-expression of ERα, whereas the normal lung tissue did not exhibit expression of ERα. Interestingly, samples 6 and 8 were derived from patient biopsies clinically diagnosed as ERα positive. Previous clinical studies have reported differences in the expression of ERα between the primary patient tumor and the metastatic sites [54, 55] although the underlying mechanisms to explain the altered expression are unknown. These results reinforce the importance of the microenvironment within the metastatic niche for influence on the differentiation state of the metastasized tumor cells and the expression of clinically relevant markers. The heterogeneity of the metastatic cells within the lung and the liver implied plasticity, whether innate or induced, during the metastatic process. Collectively, these findings demonstrate the complex integration of extrinsic and intrinsic factors affecting the behavior and phenotype of breast cancer cells during the metastatic process.
The present study data demonstrates a novel model for the study of human breast cancer breast cancer metastasis using samples obtained directly from patient biopsies. The metastatic phenotype demonstrated upon injection of tumorspheres into the mammary fat pad permits the study of all steps within the metastatic process. In particular, the development of macrometastatic lesions with organ-specific phenotypic distinctions provides a superior model for the investigation of organ-specific effects on metastatic cancer cell survival and growth. This model accurately recapitulates the metastatic latency observed clinically, permitting the development of therapeutics that target metastatic cells during dormancy prior to activation. Experimental manipulation of the organ-specific microenvironment could reveal molecular targets that are clinically accessible and biologically relevant. Furthermore, this model can be used to develop improved methods for the detection of micrometastatic cells and the prediction of metastatic disease.