ERBB2 Overexpression Establishes ERBB3-Dependent Hypersensitivity of Breast Cancer Cells to Withaferin A

Overexpression of the receptor tyrosine kinase ERBB2 (ErbB2, HER2) occurs in various solid tumors but most prominently in breast cancer, with approximately one third of patients displaying high ERBB2 levels due to gene amplification. Inconsistent response rates to therapeutic antibodies and ERBB family–directed kinase inhibitors as well as disease reoccurrence remain a problem for a large percentage of patients. ERBB3, the most potent heterodimerization partner of ERBB2, has emerged as a critical factor in both the initial oncogenicity as well as the emergence of resistance to ERBB2-directed therapies (1). When targeted for inhibition, residual levels of active ERBB2 can drive the elevation of both ERBB3 and pERBB3 (1, 2). Impaired in its own kinase activity (3), ERBB3 acts as an allosteric activator of its heterodimerization partners (4), and elevated levels of ERBB3 are correlated with progression of several solid tumors (5–10).

With its six docking sites for activated PI3K, much of the ERBB2-amplifying phenotype of ERBB3 relates to its exceptionally potent activation of the PI3K/AKT pathway (11, 12). This has prompted an intense search for therapeutics that target either ERBB3 directly, such as mAbs, or its downstream signals, especially PI3K/AKT. Given the pronounced supporter role of ERBB3, treatment approaches will almost certainly involve a combination with drugs targeting the primary oncoprotein, such as ERBB2. For such approaches, the ability to complement the primary treatments at low intrinsic toxicity is highly desirable. Although often limited in their utility as stand-alone therapeutics, natural compounds with known medicinal utility can meet this requirement. In fact, the very same complex mode of action that is often associated with their limitations as classic therapeutics may prove beneficial for their use as a sensitizers or supplementary treatment. Beyond their direct use, chemically well characterized natural products, such as Withaferin A (WA), can provide lead compounds for subsequent structure–activity relationship–based drug development.

WA is a steroidal lactone isolated from winter cherry (Withania somnifera). The purified compound was first shown to suppress Ehrlich ascites carcinoma in treated mice and increase disease-free survival when administered (continuously) posttreatment (13, 14). Since then, WA has been studied extensively as a prototype withanolide for anticancer treatment (reviewed in ref. 15). In addition to its broad and likely multicausal anticancer activity, observed generally in the high micromolar concentration range, WA can exhibit selective and cell type-specific functions at lower concentrations. It inhibits human umbilical vein endothelial cell proliferation and exerts potent antiangiogenic activity in FGF-2 Matrigel mouse models of angiogenesis with an IC₅₀ of 12 nmol/L (16). This antiangiogenic property is linked to the covalent and degradation-enhancing modification of vimentin by WA (17). The complex mode of action of WA is a limiting factor in the utilization as a stand-alone treatment. On the other hand, the broader range of responses across cancers, low toxicity, and the potential to show more potent properties in a cell type-specific setting suggest that WA may be well suited as an amplifier or supplementary treatment alongside conventional drugs.

For breast cancer cells, a previous study documented sensitivity of triple-negative MDA-MB-231 cells and ER⁺/ERBB2⁻ MCF7 to high micromolar concentrations of WA (18). ERBB2 overexpression defines a clearly discernable group of breast cancers, both in terms of expression profile, phenotype, and treatment options. We therefore evaluated whether ERBB2-amplified cancer cells share sensitivity to WA or exhibit a distinct response. Our analysis demonstrates that ERBB2 overexpression confers WA hypersensitivity in a manner that is correlated to the cell's level of dependency on synergistic signaling by ERBB3 and ERBB2.

The effects of WA were tested on several naturally occurring breast cancer cell lines carrying different ERBB2 and ERBB3 levels, including MCF7, SKBr3, BT474, and MDA-MB-231. MCF7-B2 cells expressing moderately elevated ERBB2 were generated by retrovirally transducing MCF7 with human ERBB2 cDNA. All cell lines were maintained in 1× RPMI1640 medium supplemented with 10% FBS, 2 mmol/L l-glutamine, and 0.05 mg/mL gentamicin. Untransfected base cell lines were confirmed within the last 6 months through comparative analysis against data profiles (LabCorp). Lines derived from these validated lines by transfection were not confirmed independently.

WA (ChromaDex Inc.) was prepared in 100% ethanol for experiments. Lapatinib (LC Laboratories) was prepared in 100% DMSO. The ERBB3-GFP construct was generated by cloning ERBB3 cDNA, including a C-terminal GFP sequence into a pFlag-myc-CMV-19 vector. The m-Cherry construct was purchased directly from Clontech. Western blot antibodies against ERK 1/2 MAPK (9102), AMPK (2532), GAPDH (8884), and phospho-ERBB3 (4791) were purchased from Cell Signaling Technology. ERBB4 (3412) and phosphor-P38 (1229) were products of Epitomics. ERBB2 (ab8054) and phosphor-ERBB2 (ab53290) antibodies were obtained from Abcam. ERBB3 antibody (sc-285) was purchased from Santa Cruz Biotechnology.

Various methods were used to detect cell proliferation or viability in response to WA treatment. In a subset of experiments, MTT was used as an indicator of cell viability. Tumor cells were seeded in 96-well plates at a density of 5 × 10⁴ cells per well. After overnight adherence, the cells were treated with increasing concentrations of WA as indicated. After 48-hour incubation, the cells were washed in PBS and new phenol red negative media were added containing 20 μL of MTT (5 mg/mL stock). Cells were incubated at 37°C for 1.5 hours. The media were removed and DMSO (150 μL) was added to solubilize the formazan salt formed. Quantification of the formazan salt formed was determined by measuring OD at 560 and 690 nmol/L reference using a microplate reader. Background control consisted of media containing MTT without cells.

In drug combination assays with WA and lapatinib, proliferation was assessed using Alamar Blue. BT474 and SKBr3 were seeded in 96-well plates at a starting quantity at 3 × 10³ and 6 × 10³ per well, respectively, with the presence of WA and lapatinib of various concentration indicated. Alamar Blue reagent at 10% culture volume was added to the cells 48 hours after inoculation. After 4-hour incubation, absorbance at 570 nm was measured by using a microplate reader. Spare wells on each plate without cell inoculation but receiving medium and Alamar Blue were used as controls for background subtraction.

Cells were plated overnight at a density of 5 × 10⁵ in 6-well plates and treated next day with 10 μmol/L WA at indicated time points, followed by Fc receptor blocking using CD16/32 antibody (eBioscience) for 20 minutes on ice to prevent nonspecific antibody binding. Cells were then washed and incubated with anti-neu (ERBB2) antibody directed toward the extracellular domain of human ERBB2 (Santa Cruz Biotechnology) for 30 minutes on ice at a 1:50 antibody dilution. Cells were subsequently washed and incubated with goat anti-mouse Alexa Fluor 488 secondary antibody (Molecular Probes) for 30 minutes on ice at a 1:200 antibody dilution. Finally, cells were washed and fixed in a final concentration of 1% p-formaldehyde. Cells were assessed on LSRII flow cytometer and analyzed using FlowJo software (Tree Star Inc.). Analysis was performed on 10,000 events per treatment sample.

Cells were seeded in 6-well plates with starting quantity of 3 × 10⁵. The next day, cells were treated with 10 μmol/L WA or vehicle. After 4 hours of drug treatment, cell lysates were collected by either using mild lysis buffer (20 mmol/L Tris, 137 mmol/L NaCl, 5 mmol/L EDTA, 1% Triton X-100, and 10% glycerol) or SDS sample buffer. Immunoblots were performed using antibodies specific against proteins listed in the Result section. For comparison across cell lines, equal amount of total lysate protein (estimated by BCA assay) was loaded.

SKBr3 cells (2 × 10⁵) were transfected with ERBB3-GFP and mCherry constructs using lipofectamine 2000 (11668019, Life Technologies). Cells were treated with 1 μmol/L of WA on the day of transfection (48-hour time point) and 24, 38, or 43 hours afterward. After 48 hours, cells were stained for ERBB2 surface levels using monoclonal anti-ERBB2 ECD antibody (ER23, Santa Cruz Biotechnology) and Alexa Fluor 647 secondary antibody (Molecular Probes). Fixable viability dye eFluor450 was used to distinguish live versus dead cells according to the manufacturer's instructions (eBioscience). Cells were fixed using a final concentration of 1% p-formaldehyde and assessed on the LSR II flow cytometer. Analysis was done on 50,000 events using FlowJo software (Treestar Inc.).

MCF7 cells were transfected with Escherichia coli BirA (biotinyl transferase) and either bio-tagged ERBB3 and HA-CDC37 or bio-tagged His-CDC37. The bio-tag encodes a biotinylation substrate peptide (GLNDIFEAQKIEWHE) for BirA. Forty-eight hours after the transfection, lysates were collected in mild lysis buffer (20 mmol/L Tris, 137 mmol/L NaCl, 5 mmol/L EDTA, 1% Triton X-100, and 10% glycerol), supernatant cleared by centrifugation and incubated for 1 hour with neutravidin agarose beads (Pierce, 29202). The beads were washed three times with lysis buffer, and associated proteins were analyzed by SDS-PAGE and Western blots.

GraphPad Prism software was used to perform all statistical analyses. At least two independent experiments were performed and mean ± SD was calculated. Statistical differences between control and experimental groups were determined by an unpaired Student t test. Significance is denoted as *, P < 0.05; **, P < 0.01; ***, P < 0.001 in some experiments. The IC₅₀ value for cytotoxicity was accessed by nonlinear curve regression analysis.

Clinically, the overexpression of ERBB2, and neighboring genes such as GRB7, is the result of an amplification of a portion of chromosome 17. Well-established model systems for this scenario are SKBr3 cells (ER⁻) and BT474 (ER⁻ or marginally positive). Especially, BT474 is frequently used as a model for ERBB2 overexpression and drug resistance. MCF7 cells are a widely used ER⁺ line with modest and ligand-responsive levels of both ERBB2 and ERBB3. ERBB2-transfected MCF7 cells are often used as an isogenic model to study the impact of ERBB2 overexpression in isolation. However, it is important to keep in mind that this engineered ERBB2 overexpression does not reflect the full chromosomal amplification phenotype, artificially enforces ERBB2 expression in an originally ER⁺ cell line, and often does not recapitulate anticipated drug sensitivity (e.g., Herceptin/trastuzumab).

Within this set of cell lines, supplemented by the triple-negative line MDA-MB-231, pronounced differences in sensitivity manifested themselves toward WA (Fig. 1). For MCF7 cells, proliferation was inhibited less than 20% at WA concentrations up to 5 μmol/L. Over this concentration range, MBA-MD-231 exhibited a slightly higher sensitivity in the micromolar range, consistent with previous reports (18). A much more pronounced response occurred in the two ERBB2-overexpressing lines, SKBr3 and BT474 (Fig. 1). Non-ERBB2–overexpressing lines generally do not tolerate transient overexpression of ERBB2. To evaluate the extent to which elevated ERBB2 expression drives WA sensitivity, we used an MCF7 line with modest but stable overexpression of ERBB2 (MCF7-B2). Compared with MCF7, even the modest 3- to 4-fold overexpression of total ERBB2 significantly increases the WA sensitivity of the isogenic MCF7-B2.

With elevated ERBB2 levels differentiating MCF7 and MCF7-B2 cells, we evaluated whether the receptor levels were sensitive to the addition of WA (Fig. 2A). When challenged with a high concentration of WA, the immediate (4 hours) response of MCF7-B2 cells is enhanced downregulation of ERBB2, resulting in the same final receptor levels observed in WA-treated MCF7 cells. (Fig. 2A). An analysis of cell-surface accessible receptors by FACS (Fig. 2B) qualitatively confirms the enhanced downregulation of overexpressed ERBB2 with little additional decrease at 8 hours of treatment.

Although ERBB2 is a strong contributor to cell proliferation, its oncogenic potency is very dependent on the available complement of heterodimerizing ERBB family members, especially ERBB3 (19). ERBB3 conveys ligand binding, the ability to crossactivate ERBB2 allosterically (4), and complements the MAPK-centered signal transduction of ERBB2 through its exceptionally strong activation of the PI3K/AKT pathway. SKBr3 and BT474 are broadly used models for the study of ERBB2 gene–amplified cancers with different levels of ERBB3 contributions. Figure 3A shows the immediate (4 hours) response to 10 μmol/L WA for both ERBB2 and ERBB3 in terms of receptor levels, phosphorylation states, and downstream signaling partners. ERBB2 levels, determined by fluorescence-based Western blot analysis (Licor), are shown at two intensity settings to convey the large (>100-fold) difference in receptor expression between SKBr3 and MCF7. At 10 μmol/L, ERBB2 downregulation occurs across all cell lines but is more pronounced for ERBB2-amplified lines. Likewise, WA treatment reduces levels of ERBB3, completely eliminates pERBB3, and greatly reduces pERBB2. Despite qualitatively comparable responses at the receptor level, pronounced differences in the immediate drug response surface for MAPK and AKT. For BT474 and SKBr3 cells, but not MCF7, both total and pAKT are effectively lost. For MAPK (ERK1/2) on the other hand, neither the levels nor the ERBB2-driven steady-state levels of pMAPK are reduced. Instead, all three lines show, to varying degrees, a WA-induced increase in pMAPK. Studies on the response of triple-negative MDA-MB-231 had reported a WA-induced increase in the phosphorylation of residual ERBB2 and associated MAPK activation (ref. 18; confirmed in Supplementary Fig. S1). MAPK activation in a proliferative signaling context is both variable and very timing sensitive. Indeed, in BT474 cells, MAPK phosphorylation shows a very pronounced early peak while constitutive receptor phosphorylation becomes subject to attenuation (Supplementary Fig. S2). Hence, the immediate response of the ERBB2/MAPK signaling axis to a short-term stimulus with WA can be reconciled with forced activation, and this explanation has been advanced as one possibility for the impact of WA on triple-negative cancers lines (18). However, ligand-like activation would not explain the loss of AKT, previously observed independently in prostate cancer lines (20).

A dose dependency of the immediate response to WA (4 hours) in BT474 shows that pMAPK peaks at a WA concentration at which the loss of pAKT and pERBB3 has leveled out (Fig. 3B). At higher WA concentrations, a loss of AKT occurs. By contrast, the loss of ERBB3 and pERBB3 occurs at low WA concentrations and is associated with a loss of constitutively phosphorylated ERBB2. This indicates three different concentration ranges for the loss of AKT (3–10 μmol/L), opposing changes for pMAPK and pAKT (1–3 μmol/L), and the loss of ERBB3, pERBB3, and pERBB2 (<0.3 μmol/L).

In our isogenic MCF7 model system, increased WA sensitivity correlates with elevated levels of ERBB2 (Fig. 1). Efficient ligand responsiveness in MCF7 is assured by matched levels of ERBB3 to ERBB2, and this ratio is largely maintained in MCF7-B2 cells. For BT474 and a panel of ERBB2-amplified cancers, the sensitivity to the therapeutic antibody Herceptin (trastuzumab) is not derived from interference with ERBB2 autophosphorylation but instead correlates directly with its ability to block the constitutive activation of ERBB3 by amplified ERBB2 (21). SKBr3 cells rely far less on this supporting action of ERBB3 (Fig. 3a). This translates into a higher sensitivity of SKBr3 cells to EGFR, ERBB2, and ERBB4-targeting kinase inhibitors, such as lapatinib (Fig. 4), while BT474 displays reduced sensitivity. At clinically relevant concentrations of lapatinib (>1 μmol/L), both WA (500 nmol/L) and lapatinib contribute to the increased inhibition, but their impact is less than additive. This behavior is consistent with a model in which both inhibitors target components of the same signaling pathway. At lower lapatinib concentrations, inhibition by WA is dominant. To the extent that lapatinib alone does show inhibition, contributions by WA are not additive. This inhibition profile suggests that the ability of WA to target ERBB2 and ERBB3 signaling as a functional unit should not be subject to the limitations faced by ERBB2 kinase inhibitors in isolation.

As different ERBB2 amplified cancer subtypes, a direct comparison of SKBr3 and BT474 is difficult. We therefore tested the impact of WA in an artificial, isogenic SKBr3 model carrying additional recombinant ERBB3 (Fig. 5). As contributions by EGFR or ERBB4 in SKBr3 are minimal, certainly relative to the more than 1,000-fold excess of ERBB2, this approach has also been used to establish a model system for ERBB3-mediated resistance to ERBB2-targeted therapy (2). Receptor levels and viability were determined using FACS analysis at a constant concentration of 1 μmol/L WA. This concentration emphasizes the combined impact on ERBB2/ERBB3 receptors over the direct impact on MAPK and AKT at higher WA concentrations. WA was added at different times, but all cells were analyzed at the same time posttransfection. This approach ensures that all WA-unrelated aspects, such as gradually decreasing transient expression of ERBB3-GFP, are incorporated into the 0-hour (treatment) control. Total cell numbers were significantly reduced after 48 hours of WA treatment, consistent with earlier growth assays (Fig. 1). With comparable counting events, the FACS analysis therefore does not track decreases in absolute numbers of cells. Instead, it determines the rate at which the ratio of specific cell populations changes over time. Cells with low but detectable ERBB3 levels (ERBB2^H/ERBB3^L) effectively represent the baseline for SKBr3 sensitivity to WA. Consistent with the dose–response curve in Fig. 3B, no shift toward lower cell-surface levels of ERBB2 was observed (Fig. 5A–C). This reduced contribution of receptor downregulation over the course of the 48-hour assays is also evident by Western blot analysis of the entire population (Fig. 5D). Consistent with Fig. 3, ERBB3 displays elevated sensitivity, but this only becomes apparent at 48 hours. When expressed in terms of receptor ratios, the ERBB2^H/ERBB3^H population is reduced at 24 hours and has all but disappeared at 48 hours (Fig. 5E).

The accelerated loss of the (ERBB2^H/ERBB3^H) population could represent an actual loss of viable or dividing cells, an accelerated decrease in ERBB3 receptor levels, thereby converting (ERBB2^H/ERBB3^H) to (ERBB2^H/ERBB3^L) cells, or both. To address this question, we combined receptor-based sorting with a measurement of the percentage of dead cells, as determined by eFluoro450 exclusion (Fig. 5F). This analysis indicates a faster increase of dead cells within the (ERBB2^H/ERBB3^H) population. With a difference of 2.7% (10.5% for ERBB2^H/ERBB3^H vs. 7.8%) for the total increase of dead cells at 48 hours, this difference is relatively small but compares with a difference of 0.5% between control data (0 hour) in the same analysis. Combined, these data show that against a background of ERBB2 overexpression, ERBB3 receptors are decimated by WA addition while cells maintaining high ERBB3 die at a faster rate.

The 40-hour time point for ERBB3-transfected SKBr3, as well as the 4-hour time points for BT474, suggested that ERBB3 may in fact be the most WA-sensitive component in the ERBB2/ERBB3 signaling unit. We therefore evaluate receptor levels in BT474 cells after one and two days of treatment with 1 μmol/L WA (Fig. 6A). While an overall decrease in protein levels (and cell viability) is apparent at day two, ERBB3 is already depleted after one day of treatment. Within the ERBB receptor family, ERBB3 stands out by its relatively fast constitutive turnover (22), while ERBB2 displays an exceptionally slow turnover, largely due to a high rate of constitutive recycling (23). However, both receptors require extensive assistance during maturation, and both are clients of the kinase-specific HSP90 cochaperone CDC37 (24). Previous studies had suggested that the CDC37–HSP90 complex is a target of WA (25–27). We therefore tested whether the ERBB3–CDC37–HSP90 complex is sensitive to disruption by WA (Fig. 6B). Indeed, we found that coprecipitation of the ternary complex is inhibited by WA in a dose-dependent manner. The affinity precipitation of biotin-tagged ERBB3 recovers both HSP90 and CDC37. Treatment with the HSP90 inhibitor geldanamycin disrupts the interaction of ERBB3 and HSP90 (Fig. 6C). The partial sensitivity of CDC37 coprecipitation to geldanamycin reflects the increased abundance of the ERBB3–CDC37 binary intermediate when both components are overexpressed (24). When biotin-tagged CDC37 is affinity precipitated instead, the majority of the overexpressed cochaperone is indeed in a geldanamycin-insensitive state but remains sensitive to WA, suggesting that WA targets CDC37 in a manner that destabilizes both the binary and tertiary complex of ERBB3 with CDC37 and HSP90.

Previous work had demonstrated that the triple-negative line MDA-MB-231 as well as the ERBB2-negative MCF7 line are sensitive to WA (18). We now demonstrate that this response does not only extend to ERBB2 overexpression, but ERBB2-positive lines exhibit heightened sensitivity. This sensitivity is further enhanced on the basis of ERBB3 status. The ability to target the ERBB2/ERBB3 signaling unit, even at low WA concentrations (500 nmol/L), is also underscored by the dominance of WA over partial inhibition by lapatinib (Fig. 4). More direct evidence that WA targets the ERBB2/ERBB3 signaling unit comes from the elevated WA sensitivity of MCF7-B2 cells compared with its isogenic control. Against a background of matched and highly ligand-responsive levels of ERBB2 and ERBB3 in MCF7 (28, 29), even a modest overexpression of ERBB2 establishes a pronounced shift in WA sensitivity. WA undoubtedly has a range of additional targets, especially at higher concentrations. As previously reported for prostate cancer cells, AKT levels are sensitive to high concentrations of WA. In addition, higher concentrations of WA exert a ligand-like effect and MAPK activation with a temporal profile that is typical for growth factor receptor–initiated signaling. Indeed, a previous report linked the sensitivity of the triple-negative line MDA-MB-231 to “out of context” activation of MAPK (18). These responses are present in ERBB2-amplified cells, but enhanced sensitivity arises when the ERBB2/ERBB3 signaling unit itself is being targeted.

Both, the immediate (4 hours) response to WA as well as the receptor levels after one and two days of WA exposure point toward ERBB3 as the most sensitive component in ERBB2/ERBB3 signaling. While ERBB2 receptor levels are not decreasing, unless challenged with high concentrations of WA, the loss of both pERBB3 and ERBB3 impacts the activation status of ERBB2. This speaks to the potency of ERBB3 as an enhancer of ERBB2 signaling. This involvement of ERBB3 is likely to reflect noncanonical modes of interaction and activation in which ERBB3 is as much a substrate for ERBB2 as it serves as a scaffold to facilitate ERBB2 autophosphorylation (21, 28). As is the case for the inhibition of elevated ERBB2 signaling in general, the loss of ERBB3 creates a hypersensitivity to WA that suggests an “addiction” to this amplification effect.

What sets the WA sensitivity of ERBB3 apart from ERBB2 or AKT is the impact on ERBB3 levels at low concentrations of WA. Compared with ERBB2, ERBB3 shows faster constitutive turnover. With CDC37/HSP90 as a target of WA, we investigated whether the complex formation between ERBB3, CDC37, and HSP90 at the initiation of the kinase maturation and quality control process is sensitive to WA. We found that WA indeed disrupts the ERBB3–CDC37 complex and subsequently disrupts or blocks progression to the more stable ternary complex with HSP90. Although WA may target ERBB3 at additional levels, this interference with protein maturation is consistent with the observed rapid loss of ERBB3. In addition to differences in the turnover of kinases, differences also exist in their CDC37 “client” status and stability of interactions. With ERBB2, ERBB3, and to a lesser degree, AKT as clients of CDC37 at early stages in their biosynthesis, ERK1/2 is a weak client at best, explaining the relative lack of WA sensitivity.

Through its ability to target ERBB2 and ERBB3 as a signaling unit at low concentrations of WA, and combined with its low toxicity, WA or derivatives thereof may provide a promising route as supplements to other forms of treatment. WA does not only target ERBB2-overexpressing cancers at concentrations below its previously reported anticancer properties; it also provides a second line of attack in cancers where ERBB3 contributes to resistance or overall aggressiveness. Compared with the additional targeting PI3K/AKT signaling (30), WA may achieve the objective of exploiting ERBB3 codependency and enhance ERBB2 treatment with lower intrinsic toxicity.

As a drug, unmodified WA has clear limitations, such as its stability in serum. However, as a potency enhancer of existing anti-ERBB2 therapeutics, it may not even be desirable to deviate from the natural compound with its low toxicity. On the other hand, studies that demonstrate its efficacy may also spur the development of better-suited derivatives. In the structure–activity relationship evaluation of these derivatives, it will be important to keep in mind that the dominant, efficacy-enhancing modes of actions for WA vary in different cancer settings. This may well translate into WA derivatives that are optimized for different cancer targets. Knowledge of the distinguishing parameters that characterize the WA response will be critical in such pursuits.

No potential conflicts of interest were disclosed.

Conception and design: S. Andreansky, R. Landgraf

Development of methodology: W. Liu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): W. Liu, A.R. Barnette

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W. Liu, A.R. Barnette, S. Andreansky, R. Landgraf

Writing, review, and/or revision of the manuscript: W. Liu, A.R. Barnette, S. Andreansky, R. Landgraf

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Andreansky

Study supervision: S. Andreansky, R. Landgraf

The authors thank the Sylvester Comprehensive Cancer Center for their sponsorship of core facilities, specifically fluorescence-assisted cell sorting.

This work was supported by the NIH/NCI (R01-CA98881/to R. Landgraf), the Braman Family Breast Cancer Institute (to R. Landgraf), NIH/NCCAM (F31AT007134-01A1/to A.R. Barnette), and the American Cancer Society (MRSG-08-064-02-LIB/to S. Adreansky).

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.