All studies were conducted in an AAALAC accredited facility in compliance with the PHS Guidelines for the Care and Use of Animals in Research. Naïve 6-to-8 week old female scid/NCr (BALBc) mice from the NCI Animal Production Program (Frederick, MD) were used as transplant recipients. Autoclaved feed and hyper-chlorinated water were provided ad libitum. The Brca1^{Co/CoΔexon11}, p53^+/-, MMTV-Cre mouse mammary tumors 21 were dissected and cell suspensions prepared as described by Varticovski and coworkers 22. In summary, tumors were excised, dissected, mechanically dissociated and forced through a 40 uM mesh. Viable cells were either frozen in freeze down media, or plated at low density in 100 or 150 mm plates in RPMI supplemented with Pen/Strep, glutamine, and 2% FCS for selection of cell lines. Non-adherent cells were removed after 48 hours and surviving clones were isolated using cloning cylinders. For implantation of cell suspensions into naïve recipients, the mouse fat pad #4 of SCID mice was visualized through a small skin incision just anterior to the rear leg, and one million cells were injected in 50μ l of RPMI-1640. The incision was closed with a sterile wound clip, which was removed in 7 days.

Tumor size was determined biweekly using caliper measurements (millimeters) in two perpendicular dimensions (length and width). Tumor weights (milligrams) were calculated using the formula for a prolate ellipsoid and assuming a specific gravity of 1.0 g/cm³ 23. Mice were killed when tumors reached an approximate 1 g of wet weight. Tumors were divided into the following fragments: a portion was used to prepare cell suspensions and further passages in vivo, the remaining was frozen or fixed in 10% neutral-buffered formalin.

Cells derived from Brca1 mammary tumors were grown at 37°C in 5% CO₂ in RPMI-1640 media supplemented with increasing concentrations of FBS up to 10%, pen/strep, and glutamine. Over 40 clones were isolated. Sixteen cell lines were developed from 5 original primary tumors and maintained in culture for up to 50 passages in RPMI1640 supplemented with 10%FCS.

Brca1 primary tumors and cell lines were genotyped as described by PCR amplification of sequences specific for the null Brca1 allele 20, conditional Brca1 (Brca1^Co/Co) allele 1321, wild-type p53 allele 20, and CRE 2024.

Total RNA was isolated from 5 spontaneous (original, 0) Brca1 mammary tumors, 4 tumors from first-passage transplantation into naïve recipients (first-passage, 1), 3 tumors from second-passage transplantation into naïve recipients (second-passage, 2), and 7 representative original tumor-derived cell lines. Normal virgin mammary glands from three #4 glands of 6–8 wk old C57Bl6 or SCID mice were included in the microarray analysis. Total RNA was isolated using TRIzol reagent (Invitrogen, Carlsbad, CA) according to the manufacturer's instructions, followed by DNAse treatment and RNA clean-up (RNeasy Mini Kit, QIAGEN, Valencia, CA). RNA integrity was determined using the RNA 6000 Nano LabChip Kit on Agilent 2100 Bioanalyzer (Agilent, Santa Clara, CA). RNA was amplified and hybridized overnight to the Agilent Mouse Oligo Microarrays 22 K (G4121AorB) against a common reference total RNA 17. The co-hybridized samples and reference RNA were washed and scanned on an Axon GenePix 4000B scanner (Molecular Devices, Sunnyvale, CA), analyzed using GenePix 4.1 software, and uploaded to the UNC microarray database where Lowess normalization was performed 25. Genes were filtered by requiring the Lowess-normalized intensity values in both channels to be > 30. The log₂ ratio of Cy5/Cy3 was then reported for each gene. Hierarchical clustering and Gene Set Expression Comparison were performed using BRB Array Tools software developed by the Biometric Research Branch of the NCI 26. Analysis was restricted to probes that reported values in 75% or more of the samples for comparative analysis or probes that in addition had a minimum 2.5 – fold expression change in either direction from the median value in at least 20% of samples, for hierarchical clustering (defined as most variable genes).

A comparative analysis of previously published microarray expression data of Brca1-deficient tumors on the same array platform 17 was performed using Gene Cluster 3.0. To this end, we compared gene expression from 22 arrays encompassing the 5 original Brca1 tumors from our study, with 7 tumors from Brca1^+/-, p53^+/- irradiated (IR, Kohler) mice, and 10 tumors from Brca1^Co/Co, p53^+/-, MMTV-Cre (Furth) from a previous study 17. Genes [20,988] were filtered for 80% present (to remove genes that have missing values in greater than 20% of the arrays) and Standard Deviation (SD Gene Vector) >= 1.5 (to remove genes that have standard deviations of observed values less than 1.5, thus selecting for more variable genes across experiments). This filter yielded 339 genes that were used for Hierarchical Clustering. The dendrograms were visualized with Java TreeView.

Cells from each cell line were grown to 70% confluence, scraped or trypsinized, stained with PE Anti-mouse CD44 (BD Pharmingen, San Jose, CA), APC Anti-mouse CD24 (Biolegend, San Diego, CA), FITC Anti-mouse CD44 (Southern Biotech, Birmingham, AL), or APC Anti-mouse CD117 (Biolegend). Rat IgG (CHEMICON, Billerica, MA) was used as the isotype control according to manufacturer's instructions. For analysis of Cytokeratins, cells were washed in ice-cold PBS, fixed in 70% ethanol, and stained with either Cytokeratin 5 (Covance, Berkeley, CA) or Cytokeratin 18 (Abcam, Cambridge, MA), and counter-stained with Alexa 488-conjugated anti-rabbit or anti-mouse IgG, respectively. Cells were analyzed by flow cytometry using FACScalibur or LSR II (BD Biosciences, San Jose, CA). Data were collected with Cell Quest Pro software (BD) from no fewer than 10,000 cells and FACSDiVa software (BD) from no fewer than 30,000 cells, respectively.

Paraffin sections of original and transplanted tumors were stained for H&E for histological analysis. Histology of original tumors was compared to transplanted tumors derived from cell suspensions. Immunofluorescent staining was performed for smooth muscle actin (SMA) using mouse A2537 antibody (Sigma, St Louis, MO) at 1:1000 dilution and vimentin, using guinea pig anti-Vimentin antibody (RDI, Concord, MA) at 1:400 dilution. Slides were de-paraffinized and hydrated through a series of xylenes and graded ethanol steps. Heat-mediated epitope retrieval was performed in boiling citrate buffer (pH 6.0) for 15 min. Samples were then cooled to room temperature for 30 min. Secondary antibodies for immunofluorescence were conjugated with Alexa Fluor-488 or -594 fluorophores (1:200, Molecular Probes, Invitrogen, Carlsbad, CA).

Cell lines were plated in chamber glass slides, allowed to attach overnight, stained with APC Anti-mouse CD117 (c-kit) from Biolegend, Cytokeratins 5 and 18, and counter stained with Alexa 488-conjugated anti-rabbit or anti-mouse IgG, respectively.

The analysis of protein levels was performed essentially as described 27. Membranes were blocked at room temperature for 1 h I nTBS/0.05% Tween-20 (TBS-T) containing 5% non-fat dry milk (monoclonal antibodies) or 5% BSA (polyclonal antibodies) and exposed to primary antibodies overnight at 4°C in blocking solution. The following commercial antibodies were used: Keratin 5 (Covance); Keratin 18 (Abcam); CD117 (Abgent; San Diego, CA), p53(S15), PDGF Receptor (Anti-CD140a) (Chemicon), Vimentin (BD Pharmingen), and Actin (Ab-1) (Oncogene Research Products, Cambridge, MA). Membranes were washed twice in TBS-T and incubated with species-specific HRP-labeled secondary antibodies in TBS-T for 1 h. Finally, the membranes were washed in TBS-T three times and the proteins were visualized using ECL Western blotting detection reagents (Amersham, Piscataway, NJ).

The overall similarity of gene expression profiles among five original Brca1 mammary tumors was assessed by pairwise correlations of microarray log-ratios based on 15,781 probes with reported values in at least 75% of samples. The correlation matrix showed Pearson correlation coefficients for pairwise comparisons of 0.66–0.82 (Figure 1A). These correlation coefficients are lower than those obtained for mouse mammary tumors from MMTV-PyMT (0.79–0.94) or MMTV-Wnt1 (0.87–0.95) mouse tumors using a similar analysis 22 (Additional file 1A). In spite of having a similar size at the time of collection (approximately 1 gm of wet weight), heterogeneity of Brca1 mammary tumors was also evident upon pathological examination of morphology, immunohistochemistry, and immunofluorescence analysis of tumor markers. Although all tumors were adenocarcinomas, the original tumor 0_A1 had areas of whorls and clusters of spindle-shaped tumor cells in a myxomatous stroma, features suggestive of epithelial-to-mesenchymal transition (EMT) (Additional file 2A). Consistently, this tumor was also positive for protein expression of a mesenchymal marker, vimentin (Additional file 2B). No evidence of EMT was found in any of the four other original tumors. One of the other tumors was an adenocarcinoma with features of squamous carcinoma (Additional file 2A).

In previous studies 1728 Brca1 loss was found to give rise to mammary tumors with a primarily basal-like phenotype. Two models with Brca1 deficiency were featured in a large study that used multiple other mouse mammary tumor models 17, and included 7 mammary tumors that arose after irradiation of Brca1^+/-, p53^+/- mice (designated Brca1^+/-, p53^+/- IR-Koller) and 10 tumors from Brca1^Co/Co, p53^+/-, MMTV-Cre strain, equivalent to the tumors used in our study (designated Brca1-Furth). Since the same array platform and common reference RNA had been used for those studies, we performed an analysis of all 22 samples as described in Methods and found 339 that distinguished these samples. Unsupervised Hierarchical Clustering using these 339 most variable genes (Additional file 3) separated the tumors in 3 main clusters (Figure 1B). All seven Brca1^+/-, p53^+/- IR-Koller and 3/10 Brca1-Furth tumors showed a basal-type phenotype, previously segregated in Group IV, and characterized by expression of Keratin 5 17. Four of the original Brca1 tumors from our study clustered with those four Brca1-Furth tumors that had mixed basal and luminal features. An additional cluster was formed by the original Brca1 tumor A1 from our study and three of Brca1-Furth tumors previously segregated in Group II tumors of spindloid morphology 17. This cluster showed mesenchymal-like features that included lack of epithelial markers, such as keratins. Moreover, the pattern of expression of keratins across all 339 filtered genes further defined separation of all 22 tumors into the same 3 groups (Figure 1C). Thus, although the majority of Brca1-deficient tumors are basal-like, some have mixed basal and luminal characteristics, and a third group has features consistent with EMT.

Given the apparent heterogeneity of Brca1 mammary tumors, we examined the extent to which serial transplantations of individual tumors affect gene expression. Cell suspension from one original (0_A) tumor was transplanted into naïve recipients. Unsupervised Hierarchical Clustering of 5 original Brca1 mammary tumors (0), 4 first-passage (1) and 3 second-passage (2) tumors from 0_A tumor were performed based on 416 most variable probes (Additional file 4). This analysis showed that the transplanted tumors segregated into a single branch closer to the tumor of origin, and further subdivided into two groups according to passage number (Figure 1D). As suggested by the Pearson correlation analysis, evidence that all original tumors were heterogeneous was indicated by the long distance of each sample from the immediate tree node above. Gene expression differences between 0_A tumor and the average of gene expression in all transplanted tumors showed 76 upregulated and 79 downregulated genes whose expression changed at least 3-fold (Additional file 4). However, comparison of gene expression between the first and second passage tumors showed no genes with significantly different expression (p = 0.2). Thus, serial transplantation of Brca1 mammary tumors resulted in minimal gene expression profile changes.

Because the Brca1 original tumors used for our studies were generated in mice bred in a mixed strain background composed mostly of C57Bl6 21, and cell suspensions were transplanted into immunosuppressed SCID mice, we hypothesized that some of the genetic differences between the original and transplanted tumors could be associated with differences in the genetic background. Thus, we re-analyzed tumor samples with normal virgin mammary glands from C57Bl6 and SCID mice using the 416 most variable probes defined above (Additional file 5). This analysis revealed that several immune response genes, including Histocompatibility 2, class II antigen A, alpha (H2-aA), Histocompatibility 2, class II, locus Mb1 (H2-DMb1) and Histocompatibility 2, class II antigen A, beta 1 (H2-Ab1), as well as Lymphotoxin B (Ltb), and Interleukin 15 (Il15) were downregulated in both, the transplants and normal SCID mammary gland as compared to original tumors and normal C57Bl mammary gland. These data are consistent with immunodeficient genotype of SCID mice, and confirms that many of the differences in gene expression between the original and transplanted tumors are due to differences in genetic background.

The final analysis between the original and transplanted tumors showed that several of the upregulated genes in the transplanted tumors were associated with growth, metastasis, and cancer stem cells, such as Wnt inhibitory factor 1 (Wif1), Notch gene homolog 4 (Notch4), Hormonally-regulated Neu-associated kinase (Hunk), Sprouty homolog 1 (Spry1), Twist gene homolog 1 (Twist1), and Aldehyde dehydrogenase 3 family, member B1 (Aldh3b1). In addition, apoptosis-associated Phorbol-12-myristate-13-acetate-induced protein 1 (Pmaip1) was among the genes downregulated in transplanted tumors. These data are consistent with the accelerated growth observed in serially transplanted tumors 22.

The differences between individual Brca1 tumors prompted us to generate and characterize multiple cell lines from each original mammary tumor. To ensure that all 16 cell lines derived from 5 original tumors were free of non-mammary cellular components, we genotyped these cells for Brca1^Co and p53 alleles. We confirmed that all cell lines contained the recombined Brca1^Co allele and, as shown for primary mammary tumors in this model, lost the wild type p53 allele 20 (Additional file 6). Microarray expression analysis of 7 representative cell lines from 3 original Brca1 tumors were compared to the individual tumors of origin. Cell lines B1.1, B1.2, and B1.15 were generated from the original tumor 0_B1, and the cell line RP3 was derived from 0_RP tumor. Cell lines A1.1, A1.8 and A1.10 were derived from the original tumor 0_A1. This analysis revealed 1,660 most variable genes. Unsupervised Hierarchical Clustering of all 12 samples using these 1,660 genes showed that the cell lines were discriminated from the original tumors (Figure 2A). However, all cell lines derived from the same tumor clustered together in the same sub-branch of the tree, indicating that molecular features of the original tumor are preserved in each set of cell lines derived from that tumor. Some of the differences in gene expression between original tumors and cell lines included downregulation of c-kit (CD117) and further downregulation Cytokeratins 23, 19, 18, 14, 13 and 8, in cell lines. Vimentin, Platelet-Derived Growth Factor receptor alpha (Pdgfr α) and beta (Pdgfr β) were among genes that were upregulated in cell lines consistent with adaptation to tissue culture conditions.

To further assess the conservation of features from original tumors to cell lines, we applied the Gene Set Expression Comparison tool. This tool analyzes pre-defined gene sets for differential expression among pre-defined classes, and indicates which gene set(s) contain more differentially expressed genes than expected by chance. A gene set representing differences in gene expression between the two original tumors 0_A1 and 0_B1 was generated that contained 914 probes either up- or down-regulated by at least 4-fold. This gene set was enriched in genes that were also differentially expressed among the cell lines, with high significance (p < 1e^-07). Consistently, Unsupervised Hierarchical Clustering of those original tumors and cell lines based on the 914 differentially expressed probes showed that the cell lines segregated with each tumor of origin (Figure 2B). Genes with differences in expression that were maintained in the original 0_A1 tumor and in cell lines derived from this tumor as compared to 0_B1 tumor and its respective cell lines included loss of CD24a, indicating a possible enrichment in putative cancer stem cells 2930. These data indicate that although the original tumors are heterogeneous, the groups of cell lines derived from each one of them are similarly different from each other.

We examined the expression of several gene products in all cell lines by Flow Cytometry and Western Blot. Although Cytokeratin 18 was downregulated in all cell lines by expression profiling, the protein was still detectable by FACS (Figure 2C) and Western Blot (Additional file 7). Upregulation of Pdgfr α which was detected by microarray analysis, was confirmed by FACS (Figure 2C) and Western blot (Additional file 7), while only a low level of Pdgfr β protein was detected in some cell lines. Variable expression of Vimentin and CD117 (c-kit) and a consistent loss of Cytokeratin 5 in all cell lines was observed by Western blot analysis (Additional file 7) which correlated with gene expression. Interestingly, cell lines derived from anterior and posterior tumors differed in expression of CD117 (c-kit) with protein expression detected only in cell lines derived from tumors arising in posterior mammary glands (Additional file 8).

Recent studies identified putative human breast cancer stem cells by CD44⁺/CD24^low cell surface markers that correspond to normal mammary gland-initiating cells in the mouse model 2930. Because we found a decrease in CD24a gene expression in the original 0_A1 tumor and in all cell lines derived from this tumor, we examined whether these cells contain a distinct CD44+/24^-/low population. We performed flow cytometry on multiple cell lines representative of 0_A1 and all 5 original tumors using CD44 and CD24 markers (Figure 3A). Remarkably, the A1.1 cell line which is derived from 0_A1 tumor showed up to 2.6% CD44⁺/CD24^-/low cells. Analysis of additional cell lines showed that all cell lines derived from 0_A1 tumor were enriched in CD44⁺/CD24^-/low population with 1.32–5% (Figure 3A and data not shown). In contrast, this cell population was less than 1% in all other cell lines derived from other tumors (Figure 3C). Specifically, B1.15 had 0.39%, P2.1 0.60%, and P3.17 0.46% of CD44⁺/CD24^-/low cells. Thus, enrichment in cells with markers consistent with human cancer-initiating cells is preserved in the cell lines derived from the tumor that has features of EMT. In contrast to small CD44+/24^-/low population, 50–80% of all cell lines had CD24+ or CD44+/24+ fractions, which makes these markers unlikely to represent a stem cell-enriched population. There have been no previous studies implicating CD44+24^-/low population as cancer stem cells in mouse models. Our subsequent studies confirmed stem cell nature of this population by their capacity of reconstituting tumors in mice, drug resistance, and expression of many genes associated with normal stem cell biology 31.