BT474, MCF-7, ZR75-1, MDA-MB-468, and MDA-MB- 231 cells were obtained from the American Type Culture Collection (Manassas, VA). BT474 (HTB-20) and MCF-7 (HTB-22) cells were grown in minimal essential medium supplemented with 10% fetal bovine serum and 10 μg/ml insulin. ZR75-1 cells (CRL-1500) were grown in RPMI 1640 medium supplemented with 10% fetal bovine serum. MDA-MB-468 cells (HTB-132) were grown in Leibovitz’s L-15 medium supplemented with 10% fetal bovine serum. The MDA-MB-231 cell line HTB-26 was grown in Earle’s α-minimal essential medium with glutamine and nucleosides (Mediatech, Herndon, VA) supplemented with 10% fetal bovine serum, 1% nonessential amino acids (Gibco, Invitrogen, Carlsbad, CA), and 50 μg/ml gentamicin (Gibco, Invitrogen). All cells were grown in a humidified 37°C incubator in the presence of 5% carbon dioxide except for the MDA-MB-468 cell line, which was grown in 100% air (0% CO₂). At confluence, cells were subcultured at a 1:4 ratio. MDA-MB-231 cells were stably transfected with expression vectors using GeneJuice (Novagen) and selected with the appropriate resistance antibiotic (200 μg/ml zeocin or 200 μg/ml hygromycin; Invitrogen). Notch2 and Notch4 coexpression was achieved in the MDA-MB-231 cell line by transfecting the hNotch2ICD construct into the established MDA-MB-231 mN4ICD stable line and selecting with zeocin in the presence of maintenance levels of hygromycin. The in vitro and in vivo experiments were performed using at least three stable cell populations from individual transfections. For knockdown experiments, short hairpin RNA (shRNA) retroviral vectors targeting Notch2 or a nontargeting vector for control were obtained from OpenBiosystems (V2HS_135987; Huntsville, AL). Negative control for transfection was achieved by using the shRNA nonsilencing control vector (RHS_1707), which contains no homology to known mammalian genes. Stable lines were selected with 0.75 μg/ml puromycin. For adenoviral transductions, cells were transduced in serum-free medium with 200 pfu/cell for 4 hours using a LacZ or green fluorescent protein viral construct as a control. For growth curves, cells were plated in complete medium at a concentration of 15,000 to 30,000 cells/cm² in 24-well plates and counted on day 1 and then every other day after plating using a Coulter counter. Growth curves were performed with each group measured in quadruplicate, with two counts performed in each well. In some cases, γ-secretase inhibitor XXI (Calbiochem) was added every second day on comparison to a dimethyl sulfoxide control as indicated. For assessment of proliferation, cells were incubated in 10 mmol/L bromodeoxyuridine (BrdU) for 4 hours before fixation in 4% paraformaldehyde and immunostaining or were analyzed for cell cycle phases by flow cytometry following 7-amino-actinomycin D incorporation. Data shown are representative from a minimum of three independent repeats of each experiment. For clonal growth experiments, cells were plated by serial dilution at 100 and 50 cells/well in all six wells of a six-well plate. Two weeks later, cells were washed two times with phosphate-buffered saline (PBS), fixed with methanol, and stained for 10 minutes with toluidine blue (Sigma, St. Louis, MO). Using Scion Image analysis software, a picture of each individual well was taken, the number of colonies were counted, and the total area covered by colonies was calculated. Shown are representative data collected from three independent experiments. Statistical analysis was performed by Student’s t-test or analysis of variance analysis, as appropriate, and differences were considered statistically significant with P < 0.05. For soft agar assays, a layer of 4 ml of 0.8% low-melting temperature agarose (SeaKem) dissolved in MDA-MB-231 growth medium was added to 60-mm dishes and then overlaid with a suspension of cells in 6 ml of 0.4% low-melting temperature agarose. After 21 days, the dishes were stained with 0.1 mg/ml p-iodonitrotetrazolium (Sigma) in PBS overnight. The next day, the colonies were counted using a dissecting microscope and pictures taken with a Zeiss AxioCam camera (Carl Zeiss GmbH, Jena, Germany).

The expression of Notch2 transcripts in human breast tumor tissues was compared using the Oncomine meta-analysis of cancer gene microarray meta-analysis public database.^25,26 Student’s t-test was used for analyzing differences between published datasets on the database.

Cells were lysed with HNTG buffer [20 mmol/L 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid, pH 7.4, 150 mmol/L NaCl, 10% glycerol, 1% Triton X-100, 1.5 mmol/L MgCl₂, 1.0 mmol/L ethylenediamine tetraacetic acid, protease inhibitor cocktail (Roche), 200 μmol/L NaVO₄, 1 mmol/L NaF, and 5 mmol/L β-glycerol phosphate] and cleared of insoluble material by centrifugation for 10 minutes at 14,000 rpm at 4°C. Protein concentration was determined by the bicinchoninic acid method, and 100 μg of protein was loaded. Lysates were subjected to sodium dodecyl sulfate-polyacrylamide gel electrophoresis followed by electrophoretic transfer to nitrocellulose (Schleicher & Schuell, Keene, NH) and immunoblotting with the indicated antibodies. The following antibodies were used for immunoblot analysis: anti-V5 (1:5000; Invitrogen) and anti-HA (1:1000; Covance Research Products, Princeton, NJ) followed by horseradish peroxidase-conjugated secondary antibodies (Bio-Rad, Hercules, CA). Bound antibodies were visualized by chemiluminescence (West Pico SuperSignal; Pierce, Rockford, IL).

All protocols involving mice were evaluated and approved by our Institutional Animal Care and Use Committee and performed under veterinary supervision. NCr homozygous nude mice (Taconic Farms, Germantown, NY) at 5 to 6 weeks of age were injected subcutaneously in the right flank, or in the mammary fat pad, with 2.5 × 10⁶ stably transfected MDA-MB-231 populations. Tumor growth was monitored by palpation, and the onset when tumors were detectable was noted. Tumor size was measured with calipers, and tumor volume was calculated assuming the shape as ellipsoid. Representative data were obtained from five mice per experimental group, and the entire experiment was repeated in three independent trials. Before collection, mice were injected intraperitoneally with 200 μl of 80 mmol/L BrdU solution at 15 hours and 1 hour before collection. Individual tumors were split for fixation in 4% paraformaldehyde and flash freezing in liquid nitrogen and then used for histology and immunostaining or RNA and protein collection, respectively.

BrdU immunostaining was performed using a monoclonal anti-BrdU antibody (MP Biomedicals, Irvine, CA). Following fixation, cells were treated with 0.3% H₂O₂ in methanol at room temperature for 20 minutes, followed by treatment with 20 μg/ml proteinase K in 50 mmol/L Tris/5 mmol/L ethylenediamine tetraacetic acid for 7 minutes at room temperature. Immediately following proteinase K treatment, cells were washed in 0.4% glycine-PBS and then incubated in 1.5 N HCl for 15 minutes at 37°C. Cells were then washed in 0.1 mol/L borax buffer and immunostained with a 1:100 dilution of anti-BrdU followed by a biotinylated anti-mouse antibody and the ABC Elite reagent. The antigen was detected using diaminobenzidine as the color substrate. Ten random fields of cells were captured for each sample, and the percentage of BrdU-labeled cells was determined by counts of labeled/total cells in a blinded manner. Tumor sections were labeled with biotin dUTP using terminal deoxynucleotidyl transferase (TdT) to detect DNA fragmentation. Following 0.3% H₂O₂ treatment and proteinase K antigen retrieval, tumors were incubated for 1 hour at 37°C in TdT reaction solution [TdT, 0.25 U/μl, biotin-dUTP 0.4 nmol/ml in TdT buffer (30 mmol/L Tris-base pH 7.2, 140 mmol/L sodium cacodylate, and 1 mmol/L cobalt chloride)]. Incubation in TdT reaction termination buffer (300 mmol/L NaCl and 30 mmol/L sodium citrate) quenched TdT activity. Antigen was detected using the ABC Elite reagent and diaminobenzene as the color substrate. Quantitation was done as described for BrdU immunostaining. Immunostaining for endothelial cells was performed with both anti-platelet endothelial cell adhesion molecule (PECAM) antibodies (BD Biosciences, San Jose, CA) and the anti-endothelial antigen MECA-32 (BD Biosciences) with similar results. Anti-PECAM staining was performed with a biotinyltyramide amplification reagent (Perkin Elmer, Waltham, MA), using diaminobenzidine as the color substrate. Anti-mouse MECA-32 and LYVE-1 antibodies were obtained from R&D Systems (Minneapolis, MN).

Noncounterstained PECAM sections (five tumors per condition) were quantified for vessel area. Four or five pictures of comparable regions of each tumor were taken and quantified in a blinded fashion. Using Photoshop 7.0 (Adobe Systems, Mountain View, CA), the vessels were outlined in a transparent layer and filled in with black. The outlined vessel image was opened in Scion Image, converted to binary, thresholded, and the area of black pixels measured. Shown are average percentage of vessel area per tumor area, and results were analyzed by Student’s t-test to determine statistical significance.

Total RNA was collected using TRI Reagent (Sigma) following the manufacturer’s protocol. RNA was reverse-transcribed using random hexamers in the presence of avian myeloblastosis virus reverse transcriptase to make cDNA. Successful cDNA production was verified using primers against glyceraldehyde-3-phosphate dehydrogenase or β-actin. Real-time PCR was performed using SYBR Green for Notch2, HES1, HRT1, and β-actin as the housekeeping gene. cDNA concentration in the samples was adjusted to 50 ng/μl based on the β-actin cDNA content. cDNA (1 μl) and specific primers were added to the SYBR Green PCR master mix (Qiagen, Valencia, CA), and amplification was performed in a Bio-Rad iCycler machine. Shown are relative expression ratios of Notch2, HES1, and HRT1 target genes. For PCR amplification of angiogenic factors in tumors, 25 ng of tumor cDNA was used with human primers and 100 ng of tumor cDNA with mouse primers. Primers used for detection of transcripts for Notch receptors and angiogenic factors can be found in Table 1.

Detection of human angiogenic factors in cells was performed using the TransSignal Angiogenesis Antibody Array (Panomics, Fremont, CA) following the manufacturer’s instructions. MDA-MB-231 cells were cultured for 24 hours in serum-free medium, medium collected and filtered to remove cellular debris, and 2 ml of undiluted medium was immediately used for the assay.

Previous studies detected expression of Notch in the MDA-MB-231 cells,²³ and we confirmed detectable protein levels of Notch1, Notch2, Notch4, and the Jagged1 ligand (Figure 2A). To establish a gain-of-function model, the hNotch2 or mNotch4 pathways were stably activated in MDA-MB-231 cells, and NotchICD production was confirmed by Western blotting (Figure 2B). Receptor activity was further validated using a CBF-1 response element as well as HES1 and HRT1 promoter reporters to assess activated Notch signaling (Figure 2C). HES/HRT proteins, in addition to being well-known downstream Notch targets,²⁸ may participate in the regulation of breast cancer cell growth.²⁹ Although mNotch4ICD activated CBF-1 reporter and increased HES1 and HRT1 promoter activity, hNotch2ICD activated the CBF1 reporter and increased HES1 promoter activity but did not activate the HRT1 promoter. Quantification of HES1 and HRT steady-state transcript detected HES1 expressed under all conditions, whereas HRT1 was only detected in control and mN4ICD cells (Figure 2D). It is thus possible that hNotch2 signaling has an effect in the repression of HRT1 transcript. Our data show that although NotchICD expression strongly transactivates CBF-1 and HES reporter activity, there is a significant level of HES1 expressed in the control MDA-MB-231 cells, suggesting that endogenous Notch signaling may be active. In addition, treatment of cells with γ-secretase inhibitor decreased cell number in a dose-dependent manner (Figure 2E). Because many genetic studies show that the Notch pathway is very dosage-sensitive, we also generated a loss-of-function model for Notch2 using Notch2 shRNA (Figure 2F) or a nontargeting shRNA. This approach was designed to determine the function of endogenous hNotch2 signaling on tumor cell phenotype. We were able to obtain stable populations of cells with ~50% inhibition of Notch2 transcript for further study.

Tumors were collected for histological analysis and immunostaining. Corresponding to their growth characteristics, we found that the control and hNotch2ICD tumors had necrotic central regions that were filled with inflammatory cells and regions of hemorrhage (Figure 6). Similar features were rare in mNotch4ICD tumors, which had high cell density within the periphery and the core of the tumor. Trichrome staining also demonstrated that hNotch2ICD tumors were fibrotic and had a strong stromal reaction in the tumor, with regions of highly organized vascular networks and fibroblast infiltration between tumor lobes. On the other hand, the mNotch4ICD tumors were more encapsulated and dense, with less apparent collagen content compared with the control tumors.

Our discovery that Notch2 signaling in the MDA-MB-231 system leads to tumor inhibition may be tissue type-selective. Previous studies have shown that Notch2 is oncogenic in thymic lymphoma⁴⁵ and that, in embryonal brain tumor cell lines, Notch2 increases tumorigenicity, whereas Notch1 suppresses tumorigenicity.⁴⁶ Furthermore, Notch2 signaling led to growth arrest of small cell lung cancer cells⁴⁷ but was transforming in rat kidney cells.¹⁸ These collective data imply that the cellular context of Notch2 signaling is critical for tumorigenic outcome, as is apparently the case for all Notch receptors.¹⁹ In addition, other oncogenic stimuli such as activated Wnt signaling,⁴⁸ activated Ras,⁴⁹ and myc activity⁵⁰ may modify the effects of Notch signaling on breast cancer phenotype. If substantiated by further studies, the balance of signaling through independent Notch receptors will be a critical consideration in designing Notch inhibitors as a therapeutic strategy for breast cancer.

One of our findings was that the Notch effector HRT1 is repressed selectively in the hNotch2ICD cultures. This is of interest, because the HES/HRT families have been shown to target cell cycle regulators. Notch regulates tumor cell proliferation through regulation of cyclin D1 and the cdk inhibitors p27^kip and p21^WAF1/Cip1,⁵¹ and we have observed that p27^kip is a direct transcriptional target of HRT2.⁵² Sriuranpong et al⁴⁸ have demonstrated that Notch1 and Notch2 activity in small cell lung cancer cells results in significant growth arrest and apoptosis. The inhibition of these cancer cells was a result of up-regulation of p27^kip concomitant with G₁ cell cycle inhibition. Although we did address the regulation of these cyclin-dependent kinase inhibitors and cyclin-dependent kinases in our model, we found that none of our cells or tumors express detectable p27^kip, as previously reported.⁵¹ However, p21^WAF1/Cip1 was repressed by mNotch4ICD, and this may contribute to cell cycle progression in these cells. Interestingly, the Notch2-activated cells forced a G₂ arrest when this pathway was activated concomitantly with Notch4 activation, which is a similar effect of chemotherapeutic drugs on breast cancer cells. However, unlike some of these drug mechanisms,⁵³ Notch signaling is not mediated by extracellular signal-regulated kinase phosphorylation, although loss of AKT phosphorylation in the Notch2ICD/Notch4ICD compared with Notch2ICD alone may contribute to an AKT-dependent cascade. The caveat to these experiments and interpreting their connection to the in vivo phenotype is that dynamic changes in cell cycle components occur during tumor progression in the host microenvironment that cannot be mimicked in vitro. However, our results provide evidence that activation of multiple Notch pathways leads to phenotypes distinct from either pathway alone and that the mechanisms may not be limited to regulation of cell cycle components.