Cells were pelleted by centrifuging pleural fluid diluted with RPMI medium, and the pellet was washed twice with phosphate-buffered saline (PBS), before suspension in RPMI medium without serum. Cells were then collected from the interface after centrifuging over a Ficoll density gradient. Work with pleural effusions was carried out with Central Office for Research Ethics (COREC) approval number 05/Q0704/63.

Single-cell suspensions of cell lines or cells isolated from pleural effusions were suspended at a density of 40,000 cells per millilitre in Dulbecco's modified Eagle's medium/F-12 containing 5 mg/mL insulin, 0.5 mg/mL hydrocortisone, 2% B27 (Invitrogen Ltd., Paisley, Scotland), and 20 ng/mL epidermal growth factor and seeded into six-well plates (2.5 mL per plate) or T80 tissue culture flasks (10 mL per flask) coated with 1.2% polyhema. Cultures were fed weekly and passaged every 2 weeks. Mammospheres were measured using Zeiss Axiovision software (Carl Zeiss, Jena, Germany). When passaged, mammospheres were harvested, incubated with trypsin for 3 minutes at 37°C, and dispersed by pipetting with a 23-gauge needle. After checking for single cells, the cells were pelleted and suspended in mammosphere culture medium to 40,000 cells per millilitre before replating in non-adherent plates or flasks.

Disaggregated mammospheres were seeded on glass coverslips in mammosphere medium supplemented with 1% foetal calf serum (FCS). Cells were allowed to adhere and differentiate for 5 days before fixing and staining.

Parental cells and differentiated mammosphere-derived cells grown on glass coverslips or mammospheres in suspension were fixed in 4% paraformaldehyde for 10 minutes, washed in PBS, and then permeabilised in 0.1% triton for 5 minutes. After further washing in PBS, cells were incubated for 1 hour at room temperature with neat primary antibody SWA11 (a kind gift from Peter Altevogt) 13, cytokeratin 14 (CK14) 14, CK19 15, and HMFG2 1617 supernatants. Following a wash in PBS, the cells were incubated for 1 hour at room temperature in secondary antibody Alexa 488-conjugated rabbit anti-mouse (Molecular Probes, now part of Invitrogen Corporation, Carlsbad, CA, USA) diluted 1:500 in PBS. Cells were washed with PBS before coverslips were air-died and mammospheres stained in suspension were spun-down with all supernatant removed. Coverslips and cells were then stained and mounted with 10 mg/mL DAPI (4,6-diamidino-2-phenylindole dihydrochloride) (Sigma-Aldrich, Poole, UK) in aqueous mountant (Dako 2972; DakoCytomation, Glostrup, Denmark) and viewed under a Zeiss fluorescence microscope (Carl Zeiss).

For the detection of CD24^low/- CD44⁺ populations in uncultured pleural effusions, cells were stained in 96-well plates in a volume of 50 μL with 2 μL per well of each monoclonal antibody: CD24.PE (ML5), CD44.FITC (G44-26), and ESA.PcPCy5 (BD Biosciences, San Jose, CA, USA). Isotype-matched labelled controls were also used in the analysis. Cells were labelled on ice for 30 minutes and washed twice before analysis in the cytometer.

For comparison of ML5 and SWA11 13 antibodies for detection of CD24 on uncultured pleural effusion cells, unconjugated antibodies SWA11 (antibody supernatant, generous gift from Peter Altevogt) and ML5 (BD Biosciences) were used at 1:30 dilution and neat, respectively, and detected by a fluorescein isothiocyanate-conjugated rabbit anti-mouse secondary antibody diluted 1:100 (DakoCytomation). Incubation was carried out for 45 minutes on ice per antibody with two washes after each incubation. PBS supplemented with 0.5% bovine serum albumin was used for antibody dilution, washes, and resuspension of the cells. Cells were analysed on an Epics XL (Beckman Coulter, Fullerton, CA, USA).

Female SCID (CB17/ICR-Prkdc SCID/crl) mice, 7 weeks old, were intraperitonealy injected with 0.2 mg etoposide. Seven days later, cells resuspended in matrigel/RPMI were injected subcutaneously in the flank region. Mice were examined every 2 days for the appearance of a tumour, which was measured with callipers. When tumours reached 1.4 cm², they were taken and placed in 4% paraformaldehyde. All animal work was performed under Home Office guidelines and under project licence number 70/4701.

Xenograft tumours were fixed in formal saline and processed to paraffin wax, and 3-μm sections were cut using a microtome. After sections were air-dried overnight, they were dewaxed in xylene and dehydrated in alcohol. The sections were stained with the following antibodies: anti-keratin 14 (Biogenics, Napa, CA, USA), anti-keratin 19 (BA17 15; DakoCytomation), HMFG1 1617, HMFG2 1617, and SWA11 13 using the DakoCytomation REAL™ EnVision™ detection system according to the manufacturer's instructions. When staining with anti-CK14 and anti-CK19, antigen retrieval was performed by pressure-cooking for 2 minutes in citrate buffer (pH 6.0) (DakoCytomation).

Single-cell suspensions derived from pleural effusions from 27 breast cancer patients or ascites from 5 breast cancer patients were placed in mammosphere culture. Most of the samples had been frozen for more than 20 years, although five cultures were from fresh specimens of pleural effusions collected in 2006 (designated 06 followed by the PE number in Table 1). Twenty out of 27 pleural effusion samples (74%) produced viable mammospheres (20 to 100 μm) that could be cultured past passage 2 (Table 1). In many cases, these could be passaged further. However, no ascites samples produced viable mammospheres beyond a second passage (0/5). Figure 1a shows mammospheres cultured from four samples, and for comparison, the mammospheres produced by three breast cancer cell lines are shown in Figure 1c. Only the two epithelial cell lines produced discrete relatively large mammospheres, which could be passaged. The fibroblastic-like cell line MDA-MB-231 formed small loosely adherent structures, which, however, did survive passaging.

To determine whether mammosphere-cultured cells from pleural effusions taken from late-stage breast cancer patients could differentiate into multiple lineages, cells were placed in differentiating conditions and then stained for lineage markers. In the normal breast, different keratins predominate in the different lineages, with CK18 and CK19 being expressed in the luminal cells and CK5 and CK14 in basal/myoepithelial cells. Mammospheres were disrupted, and single cells were plated on glass coverslips in medium supplemented with 1% FCS and after 5 days were stained with antibodies to keratins 5, 14, and 19 and to the luminal cell marker MUC1. While cells from uncultured pleural effusion samples were not viable under these culture conditions, pleural effusion cells that had been cultured first as mammospheres survived. Figure 1b shows that differentiated PE14 mammosphere-derived cells showed higher expression of CK5, CK14, and CK19 compared with undifferentiated mammospheres (Figure 1b). The MUC1 membrane mucin expressed by luminal epithelial cells and by the bulk population of most breast tumours was also not expressed in mammospheres, but expression was induced after transferring the disrupted mammospheres to adherent plates in the presence of serum (Figure 1b).

A CD44⁺ CD24^low/- population was isolated from breast tumours and pleural effusions by Al-Hajj and colleagues 12 and these cells were found to be enriched with tumorigenic cells. In these experiments, cells were sometimes gated for epithelial specific antigen-positive (ESA⁺) cells, and there was a suggestion that the ESA⁺ subset of CD44⁺/CD24^low/- cells were more tumorigenic. Using the same CD24 antibody used by Al Hajj and colleagues 12 (ML5) and the same CD44 antibody, we found that only some pleural effusion samples yielded subfractions of cells and these could be detected whether or not the cells were first gated as ESA⁺ (data not shown). We therefore proceeded to analyse the samples without gating for ESA. We also checked whether there was any difference in staining for CD24 using the SWA11 antibody, which is reactive with all glycoforms of CD24, and ML5, the antibody used by Ponti and colleagues 10 and Al-Hajj and colleagues 12. Figure 2 shows that, for six pleural effusion samples, we found that SWA11 did not give a stronger signal or detect more CD24⁺ cells. All the samples were therefore analysed using the phycoerythrin-conjugated ML5 antibody (see Materials and methods).

Figure 3 records the fluorescence-activated cell sorting profiles of uncultured cells stained with antibodies to CD24 and CD44 for 21 of the 27 samples of pleural effusions. The data show that in several samples two distinct populations can be recognised (for example, PE06/8, PE88, and PE70), whereas with other samples this is not the case. Moreover, CD44⁺/CD24^low/- cells were often within a subpopulation and existed only as a discrete defined subset in some samples (for example, PE23 and PE75). Inserting the cutoff values from the control antibody analysis, the cells in the bottom righthand quadrant were considered as CD44⁺/CD24^low/-, and the percentage for each sample is listed in Table 1 and Figure 3. Comparing the ability to form mammospheres and the size of the mammospheres produced with the presence of a CD44⁺/CD24^low/- population as defined in this way showed no obvious correlation with the ability of the cells to form mammospheres or with the size of the mammosphere produced (Table 1).

The flow profile of the PE14 sample was particularly striking as surface expression of CD24 and CD44 by the uncultured cells was uniformly negative (Figure 3). Moreover, mammospheres from this sample were large and could be extensively passaged. Staining of PE14 mammospheres for CD24 with either the SWA11 or ML5 antibody showed that CD24 was low or absent (Figure 4a), as was also observed by Ponti and colleagues 10 in mammospheres cultured from recurrent breast cancer or MCF7 cells. As with the lineage markers, expression was increased upon differentiation (Figure 4b).

To assess tumorigenic potential, cells from mammospheres from selected samples (governed by availability) were injected into SCID mice (Table 2). The mammospheres were gently disrupted by pipetting and resuspended in matrigel solution. Cells were injected into the flank of mice that had been previously injected intraperitonealy with etoposide solution, but the mice did not receive oestrogen implants. Unsorted/uncultured cells were injected as a control. The number of cells available for injection restricted the number of mice and the number of cells that could be injected. To obtain the data shown in Table 2, 5,000 cells from mammospheres or uncultured cells from eight samples were injected per mouse. Whereas four of the samples tested in this way produced tumours, uncultured cells from these samples did not (Figures 5a and 5b and Table 2). Figures 5a and 5b show the tumour development in mice injected with 5,000 cells from mammospheres grown from samples PE14 and PE66. Cells from PE14 mammospheres were extremely tumorigenic and cell dilution experiments showed that tumours could be produced from as few as 500 cells (Figure 5c).

Table 2 shows that of the four samples that induced tumours, all produced mammospheres larger than 50 μm; however, tumorigenicity did not correlate with the percentage of CD44⁺/CD24^low/- cells in the original population. Uncultured cells from the fourth sample, PE14, showed the unusual feature of expressing no surface CD24 or CD44. Moreover, tumours isolated from the mouse did not express these components either, nor did they express the lineage markers MUC1 or CK14 (Figure 5d) in accordance with the pathology of the tumours, which showed no indication of differentiation but a high percentage of mitoses. Weak keratin 19 expression was detected (Figure 5d), and while this is a luminal marker, it has been reported to be expressed by stem cells 18.