Nine cases of basal breast cancers with associated clinical and pathological data were obtained from the archives of the Department of Pathology, University of Massachusetts Medical School. Analysis of anonymous discarded tissue with no patient identifiers was approved by and in compliance with Institutional Review Board guidelines.

Tissue sections (5 μm) were cut from paraffin blocks, deparaffinized in xylene, rehydrated, and baked overnight at 60°C. Slides were quenched for endogenous peroxidase with 3% H₂O₂ in methanol for 20 minutes, and were processed for antigen retrieval by pressure cooking in 9 mM sodium citrate, pH 6.0, for 20 minutes. Slides were washed in PBS, and incubated overnight at 4°C with a rabbit antibody to the Notch-1 intracellular domain (NIC) or control IgG, were rinsed, and were further incubated with a biotinylated anti-rabbit IgG for 10 minutes at 22°C. After addition of streptavidin-conjugated horseradish peroxidase, the slides were incubated with 3',3'-diamino-benzidine for 3 to 10 minutes, and were counterstained with hematoxylin, as described previously 23.

The log₂ Cy5/Cy3 ratios of 232 cases of human breast cancer and their associated clinical data were downloaded from the University of North Carolina Microarray Database 2425. Only genes where the Lowess normalized intensity values in both channels were > 30 and data existed in > 70% samples were included for analysis. The gene set was further filtered to include only genes with Pearson's correlation coefficient > 0.58 with Notch-1 (n = 101).

Two-way hierarchical clustering was performed using Cluster v3 26, and the results were displayed using JavaTreeview 27. Analysis of overall survival (log-rank test) was carried out using JMP 6.0 28 (SAS Institute, Cary, NC, USA) on the subset of breast cancer patients in this cohort with available clinical data (n = 125). Data were plotted for each quartile of normalized Notch-1 log₂ ratios, from highest (first quartile) to lowest (fourth quartile). The breast cancer patients were further divided into basal (n = 35) or nonbasal (n = 88) subgroups and were analyzed for overall survival (log-rank test) using JMP 6.0 28.

We reviewed Oncomine 2930 for independent human breast cancer microarray datasets comparing estrogen receptor (ER)-negative and ER-positive tumors. Databases from 15 studies were found to contain Notch-1 and survivin relative expression data. The patient characteristics and analyses performed in each study are summarized in Table 1. Descriptive statistics including the mean, standard error, and a two-tailed unpaired t test were calculated for the comparisons between ER-positive and ER-negative samples within each study. Separately for ER-negative and ER-positive samples, a Pearson's correlation coefficient (r) was calculated for each study to measure levels of pair-wise co-expression between Notch-1, survivin, and keratin-5.

The 95% confidence interval for r was calculated based on Fisher's Z transformation 31. In one study a Fisher's Z transformation could not be performed for ER-negative samples (n = 3) 32, and an approximate variance for a Pearson's correlation coefficient was used to derive its 95% confidence interval. To summarize ER-specific results from the individual studies, Fisher's Z transformation and its variance were used in pooling correlation from different studies. The weighted average of Fisher's Z transformation and its 95% confidence interval were first estimated based on a fixed-effect model, taking into account the variance associated with each study. The ER-specific pooled estimate of Pearson's correlation coefficient and its 95% confidence interval were then derived from the estimates based on the Fisher's Z transformation. We applied a random-effect model for meta-analysis 33 to evaluate whether levels of co-expressions between Notch-1, survivin, and keratin-5 differ between ER-negative and ER-positive samples among the different studies.

The breast adenocarcinoma MDA-MB-231 cell line was obtained from the American Type Culture Collection (Manassas, VA, USA), and was maintained in culture as recommended by the supplier. The cDNA encoding activated NIC was characterized previously 34. MDA-MB-231 cells were transfected with control plasmid cDNA or NIC cDNA (2 μg) using 6 μl LipofectAmine (Invitrogen, Carlsbad, CA, USA) in Opti-Mem medium (1 ml) (Gibco, Carlsbad, CA, USA). The media was changed after 5 hours, and cells were harvested after 24 hours.

The peptidyl γ-secretase inhibitor z-Leu-Leu-Nle-CHO was purchased from Calbiochem (San Diego, CA, USA), and has been characterized previously 35. For gene silencing experiments by siRNA, MDA-MB-231 cells were transfected with double-stranded RNA oligonucleotide directed to Notch-1 (pool of three siRNA; Santa Cruz, Santa Cruz, CA, USA), survivin, or control nontargeted sequences using 10 μl HiPerfect (Gibco). Cells under the various conditions were harvested after 48 hours, and were analyzed by western blotting.

Recent studies have shown that Notch activation results in increased expression of survivin in basal breast cancer cell lines 35. To determine whether a similar association occurs in vivo, we examined by immunohistochemistry a panel of basal breast cancer cases for expression of activated Notch-1 (NIC) and survivin. The average age of the nine patients was 52.3 ± 6.1 years. All cases were grade 3 tumors with negative protein expression of ER, progesterone receptor, and HER-2, and positive protein expression of keratin 5/6, as assessed by immunohistochemistry. Activated Notch-1 was abundantly expressed in all cases examined of basal breast cancer, and was localized to both the cytosol and nuclei of tumor cells (Figure 1). Survivin was also strongly expressed in all basal breast cancer cases, and was similarly localized to the nuclei and cytosol of the tumor cell population (Figure 1).

We next analyzed the expression of Notch-1 mRNA in an established breast cancer patient cohort. Supervised hierarchical clustering of 232 cases of human breast cancer 25, using intrinsic gene analysis, revealed that higher expression of Notch-1 segregated with basal breast cancer. Other known markers of the disease, including keratin-5, keratin-14, and kit, were also highly correlated with Notch-1 expression (r ≥ 0.58) in this cohort (Figure 2).

One hundred and twenty-five patients with associated clinical outcome data were further analyzed. When stratified according to levels of Notch-1 log₂ transcript ratios, tumors with the highest quartile of Notch-1 gene expression (first quartile) exhibited abbreviated overall survival with a median survival of 27 months compared with the other groups (P < 0.001 via the log-rank test) (Figure 3a). Seventy-two percent of tumors in this first quartile (23/32 tumors) were classified as basal breast cancer, and the overall survival of these patients was approximately 50% lower of that of the remaining population (P < 0.02). Conversely, reduced levels of Notch-1 (second to fourth quartiles) were associated with better overall survival (Figure 3a). The percentage of basal breast cancers in these groups was 17% (second quartile, 5/30 tumors), 20% (third quartile, 6/30 tumors), and 10% (fourth quartile, 3/31 tumors), respectively.

The tumors were then segregated based on basal or nonbasal subgroup status and were analyzed for overall survival. In the nonbasal breast tumors, the expression of Notch-1 was not associated with significant differences in overall survival (P = 0.2734) (Figure 3b). In contrast, Notch-1 levels dictated overall survival in basal breast tumors (Figure 3c). In this basal subgroup, increased Notch-1 expression (fourth quartile) greatly reduced overall survival to the median of 26 months (Figure 3c). Comparatively, low levels of Notch-1 (first to third quartiles) demonstrated improved overall survival (Figure 3c).

We next carried out a meta-analysis of published microarray datasets to identify genes associated with Notch and potentially implicated in the molecular pathogenesis of basal breast cancer. Based on our recent data 35, we focused on survivin – a mitotic regulator and cell death inhibitor overexpressed in breast cancer 321, and associated with unfavorable outcome 2 – and keratin-5 – a marker of basal epithelium, often linked to a progenitor/stem cell phenotype 13.

Fifteen microarray datasets, mostly employing Affymetrix technology, published between 2002 and 2007 met the search criteria (Table 1). The overall median age of patients was 55.2 years. The breast tumors examined were typically < 5 cm, encompassing all grades, and included lymph node-positive and lymph node-negative disease. In one study, separate databases for lobular and ductal breast cancer were examined 36 – bringing the datasets analyzed to a total of 16. Two studies did not contain downloadable keratin-5 expression data 3637, and one study contained 68 out of 96 samples with Notch-1 and survivin data and 19 samples with keratin-5 data 38. Table 2 presents the descriptive statistics of each cohort with respect to Notch-1, survivin, and keratin-5 relative expression.

Analysis of 507 ER-negative and 1,356 ER-positive breast cancer patients revealed that keratin-5 associated with ER-negative breast cancers (Figure 4a) in seven out of 13 datasets, and that Notch-1 associated with ER-negative breast cancers in nine out of 16 datasets (Table 2). Pooled estimates of Pearson's correlation coefficient between Notch-1/keratin-5 were 0.3315 and 0.2043 for ER-negative and ER-positive breast cancers, respectively (P = 0.04) (Figure 4a). Similarly, survivin and keratin-5 co-segregated in ER-negative breast cancer, with a pooled estimate of Pearson's correlation coefficient of 0.1314 for ER-negative breast cancer and of -0.2408 for ER-positive breast cancer (P < 0.0001) (Figure 4b). A negative correlation exists between survivin and keratin-5 in ER-positive breast cancers most probably because other transcriptional and nontranscriptional mechanisms are likely to control survivin expression in nonbasal cancers (Figure 4b).

Analysis of 604 ER-negative and 1,463 ER-positive breast cancer patients revealed that survivin segregated with ER-negative tumors (two-tailed P < 0.05) in 12 out of 16 cohorts (Table 2). The Pearson's correlation coefficients between Notch-1 and survivin were 0.1804 and -0.0674 for ER-negative and ER-positive breast cancers, respectively (P < 0.0001) (Figure 4c).

Consistent with the model presented above, recent studies have shown that survivin may function as a direct transcriptional target of Notch-1, thus controlling mitotic transition and resistance to apoptosis in breast cancer 35. In agreement with these data, transfection of ER-negative breast cancer MDA-MB-231 cells with NIC resulted in increased survivin expression, as determined by western blotting, whereas acute siRNA silencing of Notch was associated with reduced survivin levels and induction of apoptosis (data not shown). Similarly, inhibition of Notch signaling by a pharmacologic inhibitor of γ-secretase suppressed survivin gene expression (data not shown), validating the identity of survivin as a direct transcriptional target of Notch in breast cancer cells 35.