Human breast cancer cell lines contain stem-like cells that self-renew, give rise to phenotypically diverse progeny and survive chemotherapy

SUM cell lines were obtained from Dr Stephen Ethier (Kramanos Institute, MI, USA) and are commercially available (Asterand, Detroit, MI); the MDA.MB.231, MCF7, and MCF10A cell lines were purchased from American Type Culture Collection (Manassas, VA, USA). Human mammary epithelial cells immortalized with the catalytic subunit of telomerase [18] were cultured in Dulbecco's modified Eagle's medium-F12 (1:1) with 5% calf serum (Hyclone, Logan, UT), insulin (5 μg/ml), hydrocortisone (1 μg/ml), and epidermal growth factor (10 ng/ml; Sigma, St. Louis, MO). MCF10A and MCF7 cells were cultured in Dulbecco's modified Eagle's medium with 10% calf serum. MDA.MB.231 cells were cultured in RPMI 1640 with 10% calf serum. SUM225CWR, SUM149PT, and SUM159PT cells were cultured in Ham's F12 with 5% calf serum, insulin (5 μg/ml), and hydrocortisone (1 μg/ml), whereas SUM1315MO2 were in Ham's F12 with 5% calf serum, insulin (5 μg/ml), and epidermal growth factor (10 ng/ml). All cell lines were grown at 37°C and 5% carbon dioxide.

Nonconfluent cultures were trypsinized into single cell suspension, counted, washed with phosphate-buffered saline (PBS), and stained with antibodies specific for human cell surface markers: ESA-FITC (Biomeda), CD24-PE, and CD44-APC (BD Pharmingen, San Jose, CA). A total of 250,000 cells were incubated with antibodies for 15 minutes at room temperature. Unbound antibody was washed off and cells were analyzed no longer than 1 hr post staining on a BD Facscalibur.

A total of 5,600 single cells/cm² were plated on super-low adherence plates (Corning, Lowell, MA). Every 3 days, after cells pelleting and resuspending in fresh media, spheres visible by eye (no magnification) were counted. Pictures were taken to assess the ratio of spheres to aggregates of cells. To correct for differing growth rates of the cell lines, the average tumorsphere number on day 12 was divided by a proliferation constant for the cell line. To quantify proliferation, cells were seeded on day 0 at 50,000 cells/well in a six-well dish in duplicate. Cells were fixed and stained with 0.1% crystal violet and quantified by extraction in 10% acetic acid and spectrophotometer reading at optical density 600 nm. Spheres were fractionated by size using 100 μm cell strainers (BD Pharmingen) and trypsinized into single cell suspension for flow cytometric analysis, as described above.

Nonconfluent cultures were pulsed with 10 μmol/l bromodeoxyuridine (5-bromo-2-deoxyuridine [BrdU]) for 10 days, being split and re-fed so that a total of 50 μmol/l BrdU was given to no more than 5 × 10⁶ cells. Chased cells were analyzed at day 0, 4, 6, and 8 days for BrdU label retention. For these pulse-chase studies, at day 0 totals of 10,000, 5,000, or 1,000 BrdU-labeled cells were added per well of a four-well chamber slide and fixed at 4, 6, or 8 days after chase, respectively. Cells chased with chemotherapy were analyzed at day 6. For immunofluorescence, cells were prepared similarly to those described in Shinin and coworkers (2006) using 4% paraformaldehyde and 0.1% saponin in PBS as a fixative; 0.25% Triton-X and 0.1% bovine serum albumin in PBS as a permeabilization buffer; and 0.02 U/μl of DNase (NE Biolabs, Ipswich, MA). Antibodies used were mouse monoclonal anti-BrdU (BD Pharmingen), rabbit polyclonal anti-Ki67 (Abcam, Cambridge, MA), and anti-mouse Alexa 488 (Molecular probes). Slides were mounted in Vecta-shield mounting media with DAPI (4',6 diamidino-2-phenylindole; Vector Laboratories, Burlingame, CA, USA). For flow cytometry, mouse monoclonal anti-EpCAM (Abcam), anti-mouse PerCP, and the BrdU-FITC set of antibodies was used (BD Pharmingen).

Confluent plates were split 1:4 and treated the following day with a 1:100 dilution of 0.1 mol/l 5-fluorouracil (5-FU) in dimethyl sulfoxide (DMSO) for a final concentration of 1 mmol/l 5-FU, or 1 μm paclitaxel in DMSO for a final concentration of 10 nmol/l Taxol (Sigma). Placebo control plates received 1:100 DMSO. Cells were fed and re-treated at 3 days and analyzed at 6 days.

Breast cancers are broadly categorized into two classes: those that express luminal keratins (luminal-type) and those that express stratified epithelial keratins (basal-type) [11, 12, 19, 20]. Recent work has characterized the cellular heterogeneity of human breast tissues based on cell surface expression of CD44 and CD24, using epithelial cells from primary reduction mammoplasty tissues, pleural effusions, and primary tumors [21]. CD44⁺/CD24^- cells exhibit features of basal cells and express genes that are involved in motility, whereas CD24⁺/CD44^- cells exhibit features of more differentiated luminal epithelial cells and express genes involved in hormone responses.

To examine whether breast cancer cell lines adhere to a similar molecular definition, we examined eight different human breast cell lines with differing tumorigenic, invasive, and metastatic potentials (Additional file 1). Using flow cytometry, cell lines could be could be grouped into three classes based on CD44 and CD24 expression profiles (Figure 1a and Additional file 2). Class 1 (luminal) includes the MCF7 and SUM225 lines that were mainly CD24⁺ and exhibit a more differentiated morphology and luminal cytokeratin expression, which is consistent with their luminal-type classification (Figure 1a; data not shown) [8, 21]. Class 2 (basal/mesenchymal) consists of the MDA.MB.231, SUM159, and SUM1315 lines that are more than 90% CD44⁺/CD24^- cells and exhibit a spindle-like appearance, consistent with their basal-type classification. Finally, Class 3 (mixed) consists of the SUM149 and human mammary epithelial cell lines, both of which exhibit two distinct CD44/CD24 populations and contain distinct populations of cells with either basal-like or luminal-like features. These data suggest that cultured breast cell lines maintain a similar cellular definition as cells in primary human tissues.

Cells from primary human breast tumors with a CD44⁺/CD24^-/low phenotype enrich for tumor formation in vivo [2]. Because basal cell lines comprise more than 90% CD44⁺/CD24^- cells, we reasoned these lines should be more tumorigenic than the mixed cell line, SUM149, which contains an average of 18-fold fewer CD44⁺/CD24^- cells (Figure 1b). To compare quantitatively the tumorigenicity of the cell lines, we performed limiting dilution injection experiments. SUM149, SUM1315, and SUM159 cell lines were injected into NOD/SCID mice in a 10-fold dilution series from 1 × 10⁶ to 1 × 10³ cells per mammary gland and the mice were monitored for tumor growth (Figure 1c). Despite containing more than 90% CD44⁺/CD24^- cells, the SUM1315 and SUM159 lines form tumors only 50% or 60% of the time with 10,000 cells injected, respectively. Surprisingly, SUM149 line, which contains only about 5% CD44⁺/CD24^- cells, exhibits a tumorigenic potential similar to that of the SUM159 and SUM1315 lines. Collectively, these data suggest that, analogous to primary breast cancers [21], a basal cellular phenotype rather than tumorigenicity consistently correlates with the percentage of CD44⁺/CD24^- cells in cell lines.

Because the percentage of CD44⁺/CD24^- cells in breast cancer cell lines does not predict tumorigenic potential, we next considered whether adding ESA to the repertoire of cell surface markers could enrich for tumor initiation capacity, as was previously demonstrated [2]. ESA and CD24 are coexpressed in malignant breast cell lines, such that ESA expression can substitute for CD24 expression in classifying the lines into luminal, basal, and mixed phenotypes (Figure 2a–c). The luminal lines, MCF7 and SUM225, exhibit two populations of ESA-expressing cells; the majority express high levels of ESA, as indicated by immunofluorescence and flow cytometry (ESA⁺), whereas a smaller subset of cells express a lower level (ESA^low; Figure 2a). Conversely, the basal lines SUM159, SUM1315, and MDA.MB.231 contain a large population of ESA-negative (ESA^-) cells and a smaller subset of cells expressing low levels ESA (ESA^low; Figure 2b). Finally, mixed cell lines such as SUM149, which contains both basal and luminal cell types, exhibit three ESA populations: those that are negative for ESA (ESA^-) and two populations of ESA-positive cells, namely ESA⁺ and ESA^low cells (Figure 2c).

Using flow cytometry, CD44⁺/CD24^-/ESA⁺ cells were detected in all of the lines examined (Figure 2d). The percentage of CD44⁺/CD24^-/ESA⁺ cells ranges from 0.01% to 0.5% in luminal lines, such as MCF7 and SUM225, to about 2.5% in the basal and mixed cell lines SUM149, SUM159, SUM1315, and MDA.MB.231 (Figure 2d). Most importantly, there is a clear correlation (P < 0.0001) between the percentage of CD44⁺/CD24^-/ESA⁺ cells in a cell line and the tumorigenic and aggressive phenotype of the line in vivo (Additional file 1 and Figure 2d).

We hypothesized that CD44⁺/CD24^-/low/ESA⁺ cells are enriched for tumorigenic potential as compared with the bulk of the cell line. Therefore, SUM159 and SUM1315 lines were sorted for CD44⁺/CD24^-/low/ESA⁺ and CD44⁺/CD24^-/ESA^- cells (Figure 3a), whereas the SUM149 cell line was sorted for CD44⁺/CD24^-/low/ESA⁺ and CD44⁺/CD24⁺/ESA⁺ cells (Figure 3b); all sorted fractions were injected into NOD/SCID mice as a 10-fold dilution series from 1 × 10⁶ to 1 × 10² cells per mammary gland.

With as few as 100 cells, CD44⁺/CD24^-/low/ESA⁺ subpopulations from SUM159 and SUM149 cell lines could form tumors in NOD/SCID mice, whereas unsorted, CD44⁺/CD24^-/ESA^-, or CD44⁺/CD24⁺/ESA⁺ fractions were unable to do so at the same dilution (Table 1). Similar results were seen with 1,000 cells injected for the SUM1315 line. Histological examination of tissue sections reveals that unsorted SUM159 cells form highly invasive spindle-cell carcinomas, and this spindle-cell morphology is recapitulated in SUM159 tumors derived from 1 × 10³ CD44⁺/CD24^-/low/ESA⁺ cells (Figure 3c). Tumors could from with as many as 1 × 10⁶ SUM159 CD44⁺/CD24^-/ESA^- cells, but these tumors lacked the morphology of the parentally derived tumors in all sections investigated (Figure 3c). The tumor histology of unsorted SUM149 cells resembles high-grade carcinomas that incorporate mouse stroma (Figure 3c). This morphology is only recapitulated in tumors derived from the CD44⁺/CD24^-/low/ESA⁺ sorted cells (Figure 3c). Examination of the injection sites where CD44⁺/CD24⁺/ESA⁺ and CD44⁺/CD24^-//ESA^- cells were unable to form tumors revealed viable cells and Matrigel at the site of injection (data not shown), confirming that lack of tumor growth from these cells was not due to clearing or cell death.

Although tumors will form if more than 100,000 CD44⁺/CD24^-/low/ESA^- and CD44⁺/CD24⁺/ESA⁺ cells are injected, only tumors from CD44⁺/CD24^-/low/ESA⁺ cells, even with as few as 100 cells, phenocopy parentally derived tumors. Therefore, cells from the parental cell line that preferentially seed tumors segregate with the CD44⁺/CD24^-/low/ESA⁺ subpopulation. Collectively, these data show that breast cancer cell lines contain cells with the capacity to form large tumors from few seeding cells. Moreover, these resulting tumors recapitulate the cellular heterogeneity of tumors derived from the unsorted parental cell line. Hence, similar to the case in primary tumors, these tumorigenic breast cancer stem-like cells are enriched 100-fold in the CD44⁺/CD24^-/low/ESA⁺ fraction across multiple cell lines. (Additional file 3).

The ability to both self-renew and differentiate into heterogeneous cell types is the definition of a stem cell that is thought to be functionally mimicked by cancer stem cells [7, 22‐25]. The only way to distinguish between highly proliferative progenitors and stem cells is the ability of the latter to grow indefinitely. However, because established cancer cell lines, by definition, have the ability to grow indefinitely, serial in vivo passaging may not necessarily accurately assess self-renewal of stem-like cells. Because serial colony-forming unit (CFU) assays have been used to assay stem cell self-renewal in other systems [26], we examined whether sorted CD44⁺/CD24^-/ESA⁺ cells contain more serial CFUs than other subpopulations. To this end, SUM159 and SUM149 cells were sorted into CD44⁺/CD24^-/ESA⁺, CD44⁺/CD24^-/ESA^-, and CD44⁺/CD24⁺/ESA⁺ populations and seeded at low density. Six days later, colonies larger than 30 cells were quantified in order to determine which fraction of cells exhibits the greatest CFU activity.

After initial seeding, all sorted phenotypes from the SUM159 line exhibit similar plating efficiency and colony formation, although unsorted, CD44⁺/CD24^-/ESA⁺, and CD44⁺/CD24^-/ESA^- populations formed more colonies than did the CD44⁺/CD24⁺/ESA⁺ cells (P < 0.0003; Figure 4a, blue bars). The day 6 colonies were trypsinized and replated at the same low density for two additional rounds in order to assay for self-renewal and clonogenic potential of the cells that seeded the initial colonies. After the second round of passaging, the CD44⁺/CD24⁺/ESA⁺ sorted cells exhibit the most depleted clonogenic potential among all the populations (P < 0.0001; Figure 4a, red bars). Upon the third serial replating, unsorted (P = 0.063) and CD44⁺/CD24^-/ESA⁺ sorted (P = 0.001) cells continue to enrich for CFUs, whereas the CD44⁺/CD24⁺/ESA⁺ and CD44⁺/CD24^-/ESA^- sorted cells exhibit evidence of exhaustion at the third passage (Figire 4a, yellow bars versus red bars). Similar results were seen upon serially replating sorted SUM149 cells (data not shown).

To assess the ability of the various populations to differentiate and reconstitute the parental cell line, SUM159 and SUM149 cells were sorted on the basis of CD44, CD24 and ESA subpopulations (Figure 4b), plated, and allowed to expand in vitro for 12 days. Expanded cultures were then analyzed by flow cytometry for reconstitution and differentiation (Figure 4c). The cultures seeded with CD44⁺/CD24^-/ESA⁺ enriched cells expand back into cultures nearly identical to the parental cell line and unsorted cultures that were seeded in parallel (Figure 4c). For the SUM149 cell line, this reconstitution and differentiation represents a change from nearly 100% CD24^- to 5.7% CD24^-, whereas for the SUM159 cell line reconstitution and differentiation represents a change from nearly 100% ESA⁺ to 1.9% ESA⁺. In contrast, cultures derived from CD44⁺/CD24⁺/ESA⁺ cells from either cell line remained 98% CD44⁺/CD24⁺/ESA⁺.

We next examined the phenotype of the cell colonies that expanded after sorting and replating. Consistent with the fluorescence-activated cell sorting data (Figure 4c), CD44⁺/CD24^-/ESA⁺ enriched cells formed heterogeneous mixed colonies that best recapitulated the morphological phenotype of the parental cell line (Figure 4d). In contrast, colonies derived from CD44⁺/CD24⁺/ESA⁺ cells exhibited a luminal type phenotype, whereas CD44⁺/CD24^-/ESA^- colonies exhibited a basal type phenotype regardless of the cell line from which the cells were sorted from. These data further support the finding (Figure 1) that a basal or luminal cellular phenotype is consistent with CD44⁺/CD24^-/ESA^- or CD44⁺/CD24⁺/ESA⁺ cells. It is worth noting that CD44⁺/CD24^-/ESA^- basal cells from the SUM149 line have a capacity to generate CD44⁺/CD24⁺/ESA⁺ cells, but this differentiation is significantly less efficient than that of the stem-like cells.

The operational definition of stem cell activity is the ability to self-renew and reconstitute a phenotypically diverse population of cells. Because cancer cell lines are generally less heterogeneous than freshly isolated primary tumor cells and have the ability to grow indefinitely, in principle this property may not be maintained in established cell lines. However, these data collectively demonstrate that human breast cancer cell lines maintain a cellular hierarchy, such that a small fraction of cells enriched in the CD44⁺CD24^-ESA⁺ subpopulation can self-renew (as indicated by increasing serial CFUs) and can reconstitute the differentiation spectrum of the parental cell line.

Suspension sphere cultures have been used to enrich for and characterize stem/progenitor cells from primary breast and glioblastomic tissues [24, 27, 28]. This process is presumed to simulate the events of tissue regeneration and maintenance from cells that survive suspension conditions. In this process, an initial phase of symmetric expansion of the seeding stem cells precedes a phase of asymmetric division, which gives rise to the differentiated progeny that comprise the sphere bulk. To evaluate whether tumorsphere formation indeed mirrors tissue regeneration in our breast cancer cell lines, we assayed the percentages of CD44⁺/CD24^-/ESA⁺ cells in maturing tumorspheres. We reasoned that if each sphere is seeded and maintained by a small population of CD44⁺/CD24^-/ESA⁺ stem cells, then smaller spheres will contain an over-representation of CD44⁺/CD24^-/ESA⁺ compared with large spheres or nearly confluent adherent cultures, both of which will contain more differentiated progeny cells.

We used cell strainers to separate spheres larger than 100 μm from smaller spheres (Figure 5a). Collected tumorspheres were dissociated into single cell suspension and analyzed by flow cytometry for the percentage of CD44⁺/CD24^-/ESA⁺ cells. SUM159 tumorspheres less than 100 μm contained sixfold more CD44⁺/CD24^-/ESA⁺ cells, whereas larger tumorspheres contained a similar percentage of CD44⁺/CD24^-/ESA⁺ cells to that of a nearly confluent adherent culture (Figure 5b). Comparable fold increases were also observed for the smaller SUM1315 and SUM149 tumorspheres compared with matched adherent cultures (Additional file 4 [part A]).

Next, to evaluate whether tumorsphere formation is related to the percentage of CD44⁺/CD24^-/ESA⁺ cells, the number of spheres formed by each cell line was quantified. The absolute number of tumorspheres that develop increases over 12 days (Figure 5c), suggesting that the proliferation rate of the cell line contributes to the number of observable spheres. Because the growth rate of suspension cultures is similar to that of matched adherent cultures (Additional file 4 [part B]), and the proliferation rates among the cell lines vary (Additional file 4 [part C]), tumorsphere formation was normalized to the proliferation rate of each cell line (Figure 5d). In doing so, it is clear that tumorsphere formation correlates with the percentage of CD44⁺/CD24^-/ESA⁺ cells in the cell line (Figures 5d and 2d), and thus also correlates with tumorigenicity.

Slower cell cycle kinetics and the ability to retain DNA label are characteristics of normal tissue stem cells that have been assayed by [³H]deoxythymidine or BrdU incorporation [29, 30]. To determine whether cancer stem-like cells in cell lines share this attribute, asynchronous SUM159 and MDA.MB.231 cells were cultured with BrdU for 10 days, allowing incorporation of BrdU label into DNA of all cells (Additional file 5 [part A]). Cells were then replated at low densities and chased with non-BrdU containing media to examine label retention at 4, 6, or 8 days after chase. After 6 days of chase, 7% of cells stained for partial BrdU label retention, seen as speckles in the nucleus, indicating that at least one round of DNA replication and division had taken place (Figure 6a,b). In contrast, only 0.9% of cells retained complete BrdU labeling, indicating a failure to divide during the chase period (Figure 6a,b). By flow cytometry, 90% of the CD44⁺/CD24^-/ESA⁺ cells contain BrdU label 6 days after chase, whereas again only 7% of the total population is BrdU⁺ using the same gate (Figure 6c and data not shown). Furthermore, cells containing the highest 1% of BrdU label, corresponding to complete BrdU label retention and therefore a lack of division during the chase period, are comprised of only 30% CD44⁺/CD24^-/ESA⁺ cells (Additional file 5 [part B]). Cultures were immunostained for BrdU and the proliferation marker Ki67 at 6 days after chase to examine the proliferation rate of the cells. As expected, nearly all of the cells in the cultures stained positive for Ki67, indicating that most cells are in S-phase. Interestingly, 80% of the BrdU⁺ cells also stained positive for Ki67 (Figure 6d [indicated by i]), further supporting the notion that BrdU label retention is not specific only for quiescent cells. These results indicate that CD44⁺/CD24^-/ESA⁺ cells do cycle but at a slower rate than the bulk of the cell line.

Given that breast cancer cell lines contain cells that exhibit properties of cancer stem-like cells, and cancer-initiating cells in primary human leukemia and glioblastoma are resistant to chemotherapy [31, 32], we sought to determine whether cell line-derived CD44⁺/CD24^-/ESA⁺ cells would also preferentially survive treatment with chemotherapeutic agents. To this end, nonconfluent adherent cultures of SUM159, SUM1315, and MDA.MB.231 cells were treated for 6 days with high dose 10 nmol/l paclitaxel (Taxol) or 1 mmol/l 5-FU (Additional file 6 [part A]). Of the surviving cells, CD44⁺/CD24^-/ESA⁺ cells were enriched 5-fold to 30-fold in chemotherapy-treated cultures as compared with placebo-treated control cultures (Figure 7a). As expected for cell cycle inhibitory drugs, the cell line with the highest proliferation rate, MDA.MB.231, exhibited the greatest cytotoxicity (Additional files 4 [part C] and 6 [part B]), but of the 1% surviving cells after Taxol or 5-FU treatment an average of 30% or 35% were CD44⁺/CD24^-/ESA⁺, respectively (Figure 7a).

The observation that 90% of CD44⁺/CD24^-/ESA⁺ cells retain BrdU label in an untreated asynchronously diving culture suggests that one mechanism by which these cells preferentially survive cell cycle inhibitory agents could be that they divide more slowly or remain quiescent compared with the bulk of the line. To examine this possibility, BrdU-labeled cultures were chased for 6 days with media containing 10 nmol/l Taxol or 1 mmol/l 5-FU. In contrast to cells chased with plain media, nearly all of the cells with intact nuclei following chemotherapy chase were BrdU⁺, yet 30% to 70% of these cells exhibited the speckled pattern of BrdU staining, consistent with the notion that they had undergone cell division. Hence, although cells that survive chemotherapy do cycle more slowly, they continue to replicate, which suggests that their resistance is not due solely to quiescence (Figure 7b and Additional file 6 [part C]). Taken together, these data show that CD44⁺/CD24^-/ESA⁺ cells in breast cancer lines exhibit increased resistance to chemotherapy, providing evidence for a mechanism of post-treatment recurrence.