Adaptation of Energy Metabolism in Breast Cancer Brain Metastases

Brain metastases are the most feared complication in breast cancer. Nearly 20% of patients with advanced breast cancer are eventually diagnosed with brain lesions, making breast tumors the main source of metastatic brain disease in women (1–3). The incidence of brain metastases increases as patients live longer in response to improved cancer therapy, but no current regimen significantly affects breast cancer brain metastases. Palliative treatment extends survival in most cases for only a few weeks or months, but often with severe impact on the quality of life (3–5). Therefore, it is imperative to gain a better understanding of the nature and functionality of breast cancer cells that cause brain metastases for development of effective regimens to prevent and control this stage of the disease.

In the past, much attention focused on invasive properties of metastatic cancer cells and their ability to attract new blood vessels (6–8). However, intrinsic mechanisms allowing metastatic tumor cells to survive and proliferate in preferred target organs are poorly understood. Thus, the goal of this study was to analyze protein expression profiles of tumor cells from breast cancer brain metastases and gain insight into cellular properties that promote tumor cell survival in the unique brain microenvironment. Metastatic breast cancer cells colonize the brain mainly from the bloodstream (9). We therefore isolated circulating tumor cells from a stage IV breast cancer patient, reintroduced the cells into the bloodstream of immunodeficient mice, and recovered tumor cells from the brain, or long bone for comparison. To define determinants of breast cancer cell growth in the brain, we examined the protein expression profiles of the parental cell line and its brain or bone homing variants by multidimensional proteomic analysis, MudPIT (10). More than 300 proteins were found uniquely regulated in the brain metastatic cells. Most of these proteins are involved in cellular metabolism and cell stress response. Our proteomic results, transcriptional validation, and cell function analyses in vitro indicate that brain metastatic cells use enhanced mitochondrial respiratory pathways for energy production and antioxidant defense mechanisms. The metabolic changes identified in the brain metastatic cells may reflect a predisposition or adaptation of the tumor cells to the brain microenvironment where a constant high energy demand is met almost entirely by glucose oxidation (11–14). Our results provide new evidence that brain metastatic breast cancer cells use energy metabolism pathways that are distinct from anaerobic glycolysis, which is predominant in most cancer cells in oxygen-poor tumor microenvironments. Furthermore, the redox state of brain metastatic breast cancer cells may provide a link between their energy metabolism and gene regulation. Our results show the significance of understanding the metabolic state of metastatic cells and support the rationale of targeting tumor energy metabolism as a potential therapy for breast cancer brain metastasis.

Real-time PCR primers. Real-time PCR was done with the following gene-specific primers:

Breast cancer cell model. BCM2 Parent cells were established from blood of a stage IV breast cancer patient with widespread metastasis and cultured in EMEM with 10% fetal bovine serum (FBS) after isolation with antiepithelial immunomagnetic beads (monoclonal antibody BerEP4; Dynal, Lake Success, NY). BCM2 parent cells (2.5 × 10⁵) were i.v. injected into 5- to 6-week-old female severe combined immunodeficient (SCID) mice (C.B17/lcTac-Prkdc scid) and tumor cells were recovered from the brain (BCM2 BrainG1) or femur (BCM2 Bone) 6 weeks later. BCM2 BrainG2 cells were from a brain lesion after i.v. injection of BCM2 BrainG1 cells. For in vivo tracking, the cell lines were transduced with firefly luciferase (F-luc) in a lentiviral vector system with cytomegalovirus promoter, injected i.v. into female SCID mice and followed by repeated noninvasive bioluminescence imaging with an IVIS 200 system (Xenogen, Alameda, CA). For ex vivo organ imaging, mice were injected i.p. with 100 mg/kg d-luciferin 5 min before necropsy and the excised organs were incubated with 100 μg/mL d-luciferin (5 min). Bioluminescence signal was quantified as photons per second per square centimeter in defined regions of interest using Living Image software (Xenogen). Where appropriate, signal was normalized to the bioluminescence expression of each cell variant. Background luminescence in vivo was 1 × 10⁵ to 2 × 10⁵ photons/s. To measure tumor cell growth in the brain directly, 1 × 10⁴ tumor cells in 2 μL were implanted into the forebrain of female SCID mice (2 mm lateral, 1 mm anterior to bregma, 3-mm depth from dura). All animal work was in accordance with The Scripps Research Institute Animal Resources (Association for Assessment and Accreditation of Laboratory Animal Care accredited).

Protein fractionation and preparation. BCM2 Parent, BCM2 Bone, and BCM2 BrainG1 cells were seeded at the same density and harvested at ∼80% confluency. Whole-cell protein extracts were prepared with the TotalProteinExtraction kit (BioChain, Hayward, CA) and protein concentration was determined by bicinchoninic acid (BCA) assay (Pierce, Rockford, IL). For each cell line, 1 mg of total lysate was resuspended in starting buffer following the buffer exchange method of the ProteomeLab PF2D kit from Beckman Coulter (Fullerton, CA). Cell lysates were resolved based on their isoelectric point (pI) using the method of Beckman Coulter (15) and fractions collected in intervals of 0.3 pH unit using the chromatofocusing separation of the ProteomeLab PF2D system. Eleven fractions were generated for each cell line, and each fraction was precipitated with trichloroacetic acid/acetone before in-solution digest with trypsin. Protein pellets from each fraction were resuspended in trypsin digestion buffer (50 mmol/L ammonium bicarbonate + 0.1% Rapigest; Waters Corp., Milford, MA) and digested with trypsin overnight at 37°C.

Multidimensional chromatography and tandem mass spectrometry. Peptide mixtures were resolved by strong cation exchange liquid chromatography upstream of reverse-phase liquid chromatography as described (16). Eluted peptides were electrosprayed directly into an LTQ ion trap mass spectrometer equipped with a nano-liquid chromatography electrospray ionization source (ThermoFinnigan, San Jose, CA). Full mass spectra were recorded over a 400 to 1,600 m/z range, followed by three tandem mass spectrometry (MS/MS) events sequentially generated in a data-dependent manner on the first, second, and third most intense ions selected from the full MS spectrum (at 35% collision energy). Mass spectrometer scan functions and high-performance liquid chromatography solvent gradients were controlled by the Xcalibur data system (ThermoFinnigan).

Interpretation of MS/MS data sets. SEQUEST (17) was used to match MS/MS spectra. The validity of peptide/spectrum matches was assessed using SEQUEST-defined parameters, the cross-correlation score (X_Corr) and normalized difference in cross-correlation scores (ΔCn).

Distribution of X_Corr and ΔCn values for (a) direct and (b) decoy database hits was obtained, and the two subsets were separated by quadratic discriminant analysis. Full separation of the direct and decoy subsets is not generally possible. Therefore, the discriminant score was set such that a false-positive rate of 5% was determined based on the number of accepted decoy database peptides. This procedure was independently done on data subsets for charge states +1, +2, and +3. In addition, spectra were only retained if they had a ΔCn of at least 0.08 and minimum X_Corr of 1.8 for +1, 2.5 for +2, and 3.5 for +3 spectra. In addition, the minimum sequence length was seven amino acid residues. DTASelect (18) was used to select and sort peptide/spectrum matches passing this set of criteria. The human protein database used was downloaded from the International Protein Index (IPI) version 3.12 in November 2005 and reverse protein sequences of the IPI database were used as the decoy database. Peptide hits from multiple runs were compared using CONTRAST (18). Proteins were considered detected if they were identified by at least half tryptic status and more than two peptides. A semiquantitative comparison of protein expression was obtained based on the ratio of spectra counts (the number of spectra collected per protein; refs. 16, 19).

Western blot analysis. Whole-cell protein extracts were prepared with the TotalProteinExtraction kit (BioChain) and protein concentrations determined by BCA assay (Pierce). Total cell lysates (50 μg) were used to compare protein expression. Immunoblotting assays were carried out by standard procedures using AMP-activated protein kinase (AMPK)-α and phospho-AMPK-α(Thr¹⁷²) antibodies from Cell Signaling Technology (Beverly, MA); anti–β-actin was used for equal loading control (Sigma, St. Louis, MO). Bands were detected using horseradish peroxidase–labeled secondary antibodies and enhanced chemiluminescence (Amersham Pharmacia, Piscataway, NJ).

Determination of cellular ATP. Cellular ATP levels were measured in 96-well plates with the CellTiter Glo luminescence ATP assay (Promega) according to the manufacturer's instructions. Cells (2.5 × 10⁴ per well) in growth medium containing 10% FBS were seeded in quadruplicates and the assays repeated thrice with similar results. The cells in this assay were not genetically tagged with luciferase. ATP-dependent luminescence was related to an ATP standard curve ranging from 10 nmol/L to 1 μmol/L.

Real-time quantitative PCR. Changes in mRNA expression of identified metabolic enzymes were examined by real-time PCR after reverse transcription of 400-ng RNA from each sample with SuperScript II reverse transcriptase (Invitrogen Life Technologies, Inc., Carlsbad, CA). cDNA was diluted 20-fold before PCR amplification using a 2720 Thermal Cycler and SYBR Green mix (Applied Biosystems, Foster City, CA). A typical protocol involved 10-min denaturation at 95°C, 40 cycles with denaturation at 95°C for 20 s, annealing at 60°C for 20 s, and extension at 72°C for 20 s, and final elongation at 72°C for 5 min. Melting curve analysis verified that all primers yielded a single PCR product. Priming specificity for human genes was established with mouse brain cDNA. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) primers were used for mRNA normalization and quantitative real-time PCR was done twice in triplicates for each gene.

Cytotoxicity assay with bortezomib and 2-deoxy-d-glucose. Cells (1 × 10⁴ per well in 96-well plates) were seeded in growth medium and allowed to attach overnight before three gentle washes with PBS and addition of bortezomib or 2-deoxy-d-glucose (2DG). Bortezomib (VELCADE, Millenium Pharmaceuticals, Inc., Cambridge, MA) was reconstituted in DMSO and 2DG in water. Bortezomib dilutions from 40 μmol/L to 0.04 μmol/L were made with growth medium containing 2% FBS, and 2DG ranging from 20 mmol/L to 0.31 mmol/L with growth medium containing 2% dialyzed FBS (Invitrogen). Drug-treated cells were compared with cells treated with the respective diluent medium. Cytotoxic effects of each drug were assessed after 24 h using the Cell Counting-8 viability assay (Dojindo Molecular Technologies, Inc., Gaithersburg, MD). Each experiment was repeated twice.

Determination of total glutathione. Total glutathione was determined in cell lysates as described (20). Cell pellets were resuspended in 2 volumes of PBS, lysed by freezing and thawing in liquid nitrogen, sonicated twice for 10 s, and centrifuged at 11,800 × g at 4°C for 10 min. Ten microliters of supernatant were analyzed in 0.5-mL phosphate buffer [143 mmol/L sodium phosphate, 6 mmol/L EDTA (pH 7.5) containing 0.6 mmol/L 5,5′-dithiobis(2-nitrobenzoic acid), 0.5 units/mL glutathione reductase, and 0.3 mmol/L NADPH]. Glutathione reductase was produced recombinantly as described (21). Reduction of 5,5′-dithiobis(2-nitrobenzoic acid) was measured at 25°C by light absorbance at 412 nm. Glutathione contents were determined against a calibration curve and corrected for total protein content. Protein concentration of cell extracts was determined by the Bradford method.

Establishing a tumor cell model for breast cancer brain metastasis. It is generally accepted that metastatic breast cancer cells reach the brain primarily from the bloodstream (4). We therefore reasoned that blood-borne tumor cells could serve as a source for brain homing breast cancer cells and as a control against which cells derived from brain metastases can be compared with defined determinants of metastatic growth in the brain. We thus isolated circulating tumor cells from a stage IV breast cancer patient, established the tumor cells in culture, reintroduced them into the bloodstream of immunodeficient mice by tail-vein injection, and recovered tumor cells that had colonized the brain, or long bone for comparison. The cell line representing the circulating tumor cells is named BCM2 Parent (22), and their derivatives from bone lesion or brain lesions BCM2 Bone or BCM2 BrainG1, respectively. To enrich for a brain metastatic phenotype, BCM2 BrainG1 cells were subjected to an additional round of in vivo selection and the resulting cells were named BCM2 BrainG2. To establish the brain homing properties of these cells, they were genetically tagged with F-luc and their fate and distribution followed in female SCID mice after tail-vein injection by noninvasive bioluminescence imaging. BCM2 Parent cells colonized the brain and all other major target organs of breast cancer metastasis as seen in the clinic (Fig. 1A). A similar spectrum of organ colonization was observed for the in vivo selected cell variants derived from bone (BCM2 Bone) or brain metastases (BCM2 BrainG2), indicating a pluripotential of these cells on injection into the tail vein. Despite the fact that this route favors lung colonization, because the pulmonary vasculature is the first capillary bed that the cells encounter, the incidence of brain metastases in mice injected with the brain lesion-derived cells (83%) was significantly higher than that seen for BCM2 Parent (22%) or bone-derived cells (20%; Fig. 1B). Furthermore, tumor burden in the brain caused by the brain metastatic cells was >36-fold larger than that induced by the parental cells, and >17-fold larger than that caused by the bone metastatic cells. Measurements were based on total photon flux in the excised brains (Fig. 1C and D). Thus, BCM2 Parent cells represent a circulating breast cancer cell population that can efficiently colonize the brain from the bloodstream. Importantly, brain homing descendants of these cells show an increased propensity to survive and proliferate within the brain microenvironment. These results indicate a stable change in the functional phenotype of brain metastatic breast cancer cells and provide the basis for a molecular characterization of this specialized cell type.

Establishing a molecular profile of brain metastatic breast cancer cells. To define determinants of metastatic growth of breast cancer cells in the brain, we examined the protein expression profiles of the parental cell line and its brain or bone homing descendants by shotgun proteomic analyses. Total protein lysates were fractionated by chromatofocusing using the PF2D ProteomeLab System (Beckman Coulter) and the fractions analyzed by multidimensional protein identification technology, MudPIT, to establish comprehensive protein profiles of BCM2 Parent, BCM2 BrainG1, and BCM2 Bone cells. More than 300 identified proteins were found ≥2-fold up-regulated or down-regulated in the brain metastatic cells compared with the parental cells and bone metastatic variant. Functional classification of the proteins found differentially expressed in the brain metastatic cells revealed that 50% belong to the functional category of cell metabolism. Proteins involved in cell stress response comprise the second largest category (8%). Given the strong predominance of alterations in the metabolic protein profile of the brain metastatic cells, we focused our attention on proteins involved in cell metabolism and identified 63 proteins, whose differential expression particularly marks the brain-derived cells. These proteins were further clustered into three subcategories: glucose oxidation, fatty acid oxidation, and cellular redox-active proteins (Table 1).

Brain metastatic cells show enhanced mitochondrial respiratory pathways for energy production. Combining our large-scale expression data and knowledge of known metabolic networks, we were able to construct a metabolic profile of brain metastatic cells based on the differentially expressed proteins listed in Table 1. Proteins up-regulated in the brain metastatic cells indicate three major changes in the energy metabolism of these cells: enhanced glycolysis coupled to increased activity of the tricarboxylic acid (TCA) cycle, increased β-oxidation of fatty acids, and an elevated pentose phosphate pathway. Major up-regulated proteins indicating these changes in the context of their energy metabolism pathways are highlighted in Fig. 2.

Oxidative metabolism. Instead of generating energy through anaerobic glycolysis, which is often used by fast proliferating cancer cells including BCM2 Parent cells, BCM2 brain-derived cells seem to use primarily aerobic glycolysis, coupled to the TCA cycle and oxidative phosphorylation, to generate energy for cell growth. We found several lines of evidence supporting this hypothesis. First, we detected considerable up-regulation of proteins involved in glycolysis, TCA cycle (marked in gray italics in Fig. 2), and oxidative phosphorylation based on our proteomic analyses (Table 1). Second, we found corresponding increase in the transcription of genes encoding proteins that are involved in oxidative energy metabolism. Quantitative real-time PCR analyses confirmed a trend in gene expression changes consistent with our proteomic data and revealed a specific up-regulation of proteins involved in glycolysis and oxidative metabolism in the brain metastatic cells (Fig. 3). The up-regulated proteins include the glycolytic enzyme triose phosphate isomerase (TPI) and phosphoenolpyruvate carboxylase (PCK2), which can enhance TCA flux via cataplerosis. Also up-regulated were the TCA cycle enzymes, aconitate hydratase (ACO2), isocitric dehydrogenase (IDH3A), and mitochondrial malate dehydrogenase (MDH2; Fig. 3A). Furthermore, we detected increased expression of transcriptional regulators that promote oxidative metabolism, including members of the poly(ADP-ribose) polymerase-γ coactivator-1 (PGC-1) family of transcriptional coactivators, PGC-1α and PGC-1β (23), and the nuclear receptor ERRα (Fig. 3A). ERRα is a downstream effector of PGC-1α in the regulation of mitochondrial energy metabolism and enables PGC-1α to induce target genes such as IDH3A (24). The coordinated up-regulation of all three regulators, PGC-1α, PGC-1β, and ERRα, suggests that the brain metastatic cells have undergone a transcriptional switch that supports expression of oxidative metabolism pathways. Interestingly, PGC-1α, but not PGC-1β or ERRα, was elevated in the bone metastatic cells, suggesting that PGC-1α expression alone was not sufficient to activate the expression of downstream targets such as ACO2 and IDH3A. ERRα is also a regulator of other mitochondrial energy transduction pathways, including fatty acid oxidation and oxidative phosphorylation (24–26). Consistent with this role, our proteomic results showed up-regulation of several enzymes involved in fatty acid oxidation, as well as components of the electron transfer chain (Table 1). Similar to the enzymes in the TCA cycle, we found increased mRNA levels for fatty acid oxidation genes, such as β-ketothiolase (ACAA2) in the brain metastatic cells (Fig. 3B). To evaluate an in vivo relevance of our gene and protein expression results obtained with in vitro cultured BCM2 cell variants, we propagated the brain homing metastatic cells in the mouse brain and analyzed gene expression directly in the brain lesions. Consistent with the results in Fig. 3A, to C, several of the genes up-regulated in the cultured brain metastatic cells were found at similarly elevated levels in the brain lesions (Fig. 3D). This finding indicates that the changes in gene regulation seen in the brain homing cells can be maintained when the cells are expanded culture.

In addition to enzymes and transcriptional regulators, our proteomic analyses identified at least one signaling protein important for energy homeostasis, the AMPK, up-regulated in brain metastatic cells (Table 1). AMPK is a key regulator and has been implicated in the control of PGC-1α in muscle (27). We found increased levels of AMPK protein expression and activation, based on Thr¹⁷² phosphorylation, in the brain metastatic cells (Fig. 4A). This result supports a possible functional link between AMPK, PGC-1α expression, and the altered energy metabolism in the brain homing cells. Numerous studies have established AMPK as a key regulator of energy homeostasis within the cell. On activation, AMPK switches off ATP consuming biosynthetic pathways (e.g., fatty acid synthesis) and turns on ATP generating metabolic pathways (e.g., fatty acid oxidation and glycolysis) to preserve ATP levels for cell survival (28, 29). The role of fatty oxidation in the altered energy metabolism of brain metastatic breast cancer cells is less clear and may have been influenced by the culture conditions to which the cells were exposed. However, because one of the metabolites of β-ketothiolase is acetyl-CoA, a substrate for the TCA cycle, it is possible that fatty acid oxidation is switched on by AMPK at an increased rate in the brain metastatic cells to further support the enhanced mitochondrial respiratory chain pathways for energy production. In addition to acute effects of AMPK activation, long-term effects can influence the overall regulation of energy metabolism, including increased mitochondrial enzyme content and mitochondrial biogenesis (27, 28, 30, 31). Thus, AMPK can contribute to the maintenance of high ATP levels, which may stay remarkably stable for high energy–dependent molecular activities (32). Consistent with increased glycolysis and oxidative phosphorylation in the brain metastatic cells, indicated by our proteomic and transcriptional analyses, we found elevated levels of cellular ATP in these cells (Fig. 4B). It is possible that long-term effects promoted by constitutive increase in AMPK protein and activation support adaptation of the brain metastatic cells to energy pathways predominant in the brain and contribute to the growth advantage of tumor cells in the brain microenvironment.

Pentose phosphate pathway. Besides enhanced mitochondrial respiratory chain pathways, we also detected an up-regulation in the brain metastatic cells of enzymes involved in the oxidative phase of the pentose phosphate pathway (Figs. 2 and 3B). The pentose phosphate pathway serves to generate NADPH and pentose (five-carbon) sugars. This pathway consists of two distinct phases: oxidative and nonoxidative. The oxidative phase generates NADPH and is one of three major ways in which the body creates reducing molecules to prevent oxidative stress (33). We found two key enzymes of the oxidative phase up-regulated in the brain metastatic cells: glucose-6-phosphate dehydrogenase, a NADPH producing enzyme, and 6-phosphogluconolactonase (Table 1). Oxidative stress within cells is controlled primarily by the action of the tripeptide glutathione (GSH), which can reduce peroxides to maintain a reducing milieu within the cell. Regeneration of reduced glutathione by flavoenzyme glutathione reductase uses NADPH as a source of reducing equivalents (34). Consistent with the known metabolic network connecting the pentose phosphate pathway and the glutathione system, we found an up-regulation of glutathione-dependent enzymes in the brain metastatic cells. Prominent examples are glutathione reductase and glutathione S-transferase P. In addition, catalase, another important antioxidant enzyme, was found up-regulated (Table 1; Fig. 2). We believe that the NADPH-producing phase of the pentose phosphate pathway, in combination with the induction of antioxidant enzymes, is crucial in the brain metastatic cells to support the detoxification of oxidative stress. Our results suggest that the brain metastatic cells have an enhanced mitochondrial respiratory metabolism, which may lead to an increased production of reactive oxygen species. We therefore hypothesize that enhanced antioxidative defense mechanisms are crucial in the cells to maintain reduced GSH and to minimize oxidative stress caused by reactive oxygen species in the brain microenvironment.

Brain metastatic cells show reduced susceptibility to 2DG-induced cytotoxicity. It is well established that many cancer cells exhibit increased rates of glycolysis and reduced rates of respiration (35, 36). Fast proliferating cancer cells adopt anaerobic glycolysis to support rapid growth, and this renders the cells sensitive to glucose deprivation (37, 38). Glucose deprivation causes oxidative stress in cancer cells, and a synthetic glucose analogue, 2DG, mimics the effect of glucose deprivation. 2DG has been used to target cancer cells by direct inhibition of glycolysis. Recently, 2DG has been shown to cause selective cytotoxicity in tumor tissue by mechanisms similar to those triggered by glucose deprivation (39, 40). To further confirm the metabolic differences between the originally circulating parental breast cancer cells and their bone- or brain lesion–derived metastatic variants, we studied the effect of 2DG on the growth rate of these cells. We found that the brain metastatic cells, BCM2 BrainG1 and BrainG2, were 2-fold less sensitive to 2DG treatment than the parental and bone metastatic breast cancer cells (Fig. 5A). Pyruvate, a metabolite of glucose provided in the culture medium, may allow the cells to bypass the requirement for glucose and can direct the energy production via oxidative metabolism. It is also possible that the enhanced antioxidant capacity seen in the brain metastatic cells contributes to the partial resistance of these cells to 2DG. Ample evidence suggests that glucose deprivation is sufficient to induce cytotoxicity in cancer cells through metabolic oxidative stress (41, 42). Glucose deprivation can lead to a lack of NADPH and pyruvate and, consequently, to an impairment of the hydroperoxide metabolism (42). Thus, the enhanced pentose phosphate pathway and glutathione system detected in the brain metastatic cells could help the cells to quickly correct any disturbance of the cellular pro-oxidant and antioxidant balance, whereas the parental and bone metastatic breast cancer cells apparently lack this flexibility.

Brain metastatic cells show enhanced resistance to oxidative damage related to the cellular glutathione status. To further explore the hypothesis that an enhanced redox system in the brain metastatic cells can provide protection against oxidative stress, we studied the stress response of the parental breast cancer cells and their bone or brain metastatic variants to oxidative damage induced by bortezomib. Bortezomib, or PS341, is a specific inhibitor of proteasome activity that targets the protein degradation system. The ubiquitin-proteasome system is responsible for intracellular proteolysis, particularly the degradation of short-lived or oxidized proteins (43). Ample evidence suggests that proteasome inhibition induces mitochondrial dysfunction, increases generation of reactive oxygen species, elevates RNA and DNA oxidation, and promotes protein oxidation (44). We exposed the parental breast cancer cells, as well as their bone- and brain-derived metastatic variants, to increasing concentrations of bortezomib and found that the brain metastatic cells (BCM2 BrainG1 and BrainG2) were >60-fold less sensitive to this treatment than the parental cells (Fig. 5B). Furthermore, we measured the cellular content of total glutathione as an indicator of the cellular redox state under normal culture conditions and in the presence of bortezomib. Under normal growth conditions, which include 10% FBS in the medium, BCM2 Parent cells had the highest cellular glutathione levels. The difference in glutathione between the parental and brain metastatic cells under normal culture conditions might be related to the slower in vitro growth rate of BCM2 BrainG1 and BrainG2 cells, which we consistently observed (data not shown). However, in the presence of bortezomib, the glutathione levels in the brain metastatic cells notably increased (22% in BCM2 BrainG1 cells and 13% in BrainG2 cells) whereas GSH decreased (−13%) in the parental cells (Fig. 5C). This finding is consistent with the reduced sensitivity of the brain metastatic cells to this drug and the higher sensitivity of BCM2 Parent cells, where glutathione is depleted in response to bortezomib (Fig. 5B). This depletion likely reflects enhanced oxidation of glutathione to GSSG, followed by GSSG export. Together, these results support our hypothesis that brain metastatic cells can up-regulate and rapidly adapt their cellular antioxidant defense to protect against oxidative stress. Up-regulation of NADPH production coupled with enhanced glutathione reductase activity allows for the detoxification of higher reactive oxygen species and GSSG fluxes, thus maintaining high GSH levels. The observed changes in energy metabolism and cellular stress response may thus confer a growth advantage to brain metastatic breast cancer cells, allowing them to thrive in the microenvironment of the brain tissue.

Brain metastatic breast cancer cells have a growth advantage in the brain. Our experimental metastasis data and proteomic analyses indicate that the brain metastatic cells, selected in vivo for their ability to establish brain metastases, possess a phenotype distinct from the parental circulating tumor cells and their bone metastatic counterparts. The protein expression profile of the brain metastatic cells and its functional validation imply a predisposition or bioenergetic adaptation of the tumor cells to the energy metabolism of the brain, conferring an advantage for tumor cell survival and proliferation in the brain microenvironment. We found that i.v. injection of the brain lesion–derived breast cancer cells leads to a high incidence of brain metastasis significantly surpassing the parental cells and the bone metastatic variant (Fig. 1). However, this approach includes the possibility that the formation of brain metastases is a result of enhanced tumor cell survival in the bloodstream or an ability to cross the blood-brain barrier efficiently. To eliminate these factors and directly assess tumor cell survival and growth in the brain, we implanted the F-luc–tagged BCM2 BrainG2 versus BCM2 Bone cells into the forebrain of experimental mice by stereotactic injection. Tumor cell growth was followed by repeated noninvasive bioluminescence imaging over a period of 21 days. Based on the percent increase in photon flux over the signal of the original implant, we measured a significant enhancement in tumor cell growth for the brain metastatic cells compared with their bone-derived counterparts (Fig. 6A). After 21 days, the overall tumor burden in the brain caused by the brain metastatic cells was cumulatively ∼100-fold larger than that measured for the bone metastatic cells (Fig. 6B). Furthermore, we found that the brain metastatic cells spread to other areas of the central nervous system and frequently extended to the spine (Fig. 6C), similar to the clinical situation often seen in breast cancer patients with brain metastases. Thus, breast cancer cells from brain lesions can acquire a stable change in functional phenotype that confers a growth advantage in the central nervous system. Our proteomic data and cell function analyses suggest that this phenotype involves an adaptation in energy metabolism that promotes tumor cell growth in the brain.

An important question arising from this and other studies is whether tumor cell clones with a potential for seeding brain metastases already exist within certain primary tumors, or if the development of brain metastases is primarily based on the evolution of a tumor cell phenotype that supports brain colonization and expansion within the unique microenvironment of the brain. A number of studies suggest that gene expression signatures, indicating a metastatic potential or poor prognosis, can exist within primary tumors and be preserved throughout metastatic progression (45–49). Many of those genes are directly involved in cell proliferation, indicate the epithelial cell type from which the primary tumor derived, or contribute to overall cell motility and invasion. In addition, expression profiles associated with organ-specific breast cancer metastasis to bone or lung have been reported and comprise genes that are likely critical for tumor cell entry into those target organ microenvironments (50, 51). Notably, these and other studies also show that gene expression pattern, indicative of metastatic potential and target organ tropism, can be preserved after in vivo selected tumor cells are expanded in tissue culture (52).

Our study is based on originally circulating breast cancer cells that represent an advanced state of disease progression, as the cells were derived from a patient with active metastatic disease, years after the primary tumor had been removed. Strikingly, the majority of proteins, which we found differentially expressed in the brain metastatic variant of the originally circulating tumor cells, is involved in cell metabolism and may reflect either a predisposition or an adaptation of the tumor cells to the specific conditions in the brain tissue. In this context, it is important to emphasize that brain metastases have an exceptional long latency, with a median latent interval of 2 to 3 years after initial breast cancer diagnosis (53). Metastases to bone and other target organs generally appear much earlier and also occur with a significantly higher frequency (54). Our data are compatible with the concept that a subset of tumor cells evolves during malignant progression, perhaps from a preexisting cancer cell population, which is capable of gaining entry to the protected environment of the central nervous system. Importantly, these tumor cells can apparently differentiate along pathways that allow them to adapt and ultimately thrive in the brain microenvironment in response to signals from the brain tissue.

Grant support: NIH training grant T32 HL 07695 and later by the NIAID sub-contract grant UCSD/MCB0237059 (E.I. Chen); grant U19 AI063603-02 (J. Hewel); Susan G. Komen grant PDF0403205 (J.S. Krueger); the Deutsche Forschungsgemeinschaft (BE 1540-9/1; to K. Becker). Additional support came from NIH grant P41RR11823 (J.R. Yates III); NCI grants CA095458 and CA112287, and CBCRP grants 10YB-020 and 11IB-0077 (B. Felding-Habermann); NIH grant DK064951 (A. Kralli); and the Sam and Rose Stein Endowment Fund.

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.

The lentiviral expression system for breast cancer cell tagging with Firefly luciferase was developed and provided by Drs. Bruce Torbett and Mario Tschan of TSRI and the statistical analyses were supported by Dr. Jim Koziol of TSRI.