Phenotypic and functional characterisation of the luminal cell hierarchy of the mammary gland

All primary human material was derived from 11 reduction mammoplasties at Addenbrooke's Hospital, Cambridge, UK, under full informed consent and in accordance with the National Research Ethics Service, Cambridgeshire 2 Research Ethics Committee approval (08/H0308/178) as part of the Adult Breast Stem Cell Study. All tissue donors had no previous history of cancer and were premenopausal (ages 18 to 46). Mammary tissue was dissociated to single cell suspensions as previously described [19].

The number 3 and/or number 4 mammary glands were dissected from 10-week-old to 14-week-old virgin or 20-day pregnant female C57BL6/J, C57BL6/J.CBA-Tg (ACTbEGFP) and FVB mice and were dissociated in DMEM/F12 (with 2.5 mM L-glutamine and 15 mM HEPES; Gibco, Paisley, Renfrewshire, UK) supplemented with 1 mg/ml collagenase (Roche, Burgess Hill, West Sussex, UK), 100 U/ml hyaluronidase (Sigma, Poole, Dorset, UK) and 50 μg/ml gentamicin (Gibco) for 14 to 16 hours at 37°C. The mammary glands were then processed to single cells as previously described [8]. In some experiments, 8-week-old C57BL/6J mice were ovariectomised or sham-operated 3 weeks prior to collection of mammary tissue.

Single cell suspensions of human mammary cells were treated to detect the enzyme activity of aldehyde dehydrogenase (ALDH) using the Aldefluor Kit (StemCell Technologies, Grenoble, Rhône-Alpes, France) as per the manufacturer's instructions. The cells were then preblocked with 10% normal rat serum (Sigma) and incubated with the following primary antibodies (Table S1 in Additional file 1): CD31-PE/Cy7, CD45-PE/Cy7, epithelial cell adhesion molecule (EpCAM)-PE, CD49f-Alexa Fluor (AF) 647 or CD49f-Pacific Blue, ERBB3-biotin, CD44-AF647, MUC1-AF647, and CD24-AF647. Hank's balanced salt solution supplemented with 2% FBS (Gibco) was used as the diluent for all preblock, antibody incubation and washing steps.

Mouse mammary cells were preblocked with 10% normal rat serum and then incubated with the following primary antibodies (Table S1 in Additional file 1): CD31-biotin, CD45-biotin, Ter119-biotin, BP-1-biotin, EpCAM-AF647, CD49f-AF488, or CD49f-Pacific Blue, CD49b-PE and Sca1-PE/Cy7. CD45, Ter119, CD31 and BP-1 were used to deplete contaminating haematopoietic cells, endothelial cells and a proportion of stromal cells, respectively (collectively termed Lin⁺ cells). Where required, single cell suspensions of mouse mammary cells were treated to detect the enzyme activity of ALDH using the Aldefluor Kit as per the manufacturer's instructions, and then cells were stained as above.

Biotin-conjugated antibodies were detected with streptavidin-APC-Cy7 (BioLegend, Bar Hill, Cambridgeshire, UK). Cells were then filtered through a 30 μm cell strainer and incubated with 4',6-diamidino-2-phenylindole (Invitrogen, Paisley, Renfrewshire, UK) or propidium iodide (Sigma). Human cells were sorted using an Influx (Becton Dickinson, Oxford, Oxfordshire, UK) and mouse cells were analysed using an LSRII (Becton Dickinson) and they were sorted on a FACSAria I (Becton Dickinson) or a MoFlo (Beckman Coulter, High Wycombe, Buckinghamshire, UK). The gating cascade is shown in Additional file 2. A cell recovery count was performed after each sort. Single-stained control cells were used to perform compensation manually. Gates were set in reference to negative controls stained with isotype antibodies conjugated to individual fluorochromes or to fluorescence-minus-one controls (omitting one reagent at a time). The ALDH⁺ gate was set in reference to control populations incubated with the ALDH inhibitor DEAB in addition to Aldefluor. Flow cytometry data were analysed using FlowJo™ software (Tree Star, Inc., Ashland, OR, USA).

All animal work was approved by Cambridge Research Institute Animal Ethics Committee and the Home Office. Renal capsule experiments were carried out on 10-week-old female NOD/SCID IL2Rγc^-/- (NSG) mice as previously described [19] with the modification that in some experiments the collagen gels were supplemented with 20% growth factor-reduced Matrigel (BD Biosciences, Oxford, Oxfordshire, UK). A sialastic pellet containing 2 mg 17β-oestradiol and 4 mg progesterone was implanted subcutaneously in recipient mice when human cells were being transplanted [20]. In some experiments, the hormone pellets were surgically excised 5 weeks post surgery and the mice mated. To recover renal gels, recipient mice were killed and the retrieved gels were fixed in 4% paraformaldehyde for 1 hour before being processed into paraffin. On occasion, gels were dissociated for 4 to 5 hours at 37°C in Mouse EpiCult-B™ media (StemCell Technologies) supplemented with 5% FBS, 600 U/ml collagenase and 200 U/ml hyaluronidase. After digestion, cells were washed in Hank's balanced salt solution supplemented with 2% FBS, trypsinised for 5 minutes with gentle pipetting and injected into the cleared mammary fat pads of NSG mice as described below.

For the mouse mammary repopulating unit (MRU) assays, donor cells were suspended in 65% Hank's balanced salt solution supplemented with 2% FBS additionally supplemented with 25% growth factor-reduced Matrigel and 10% trypan blue solution (0.4%; Sigma), such that a 10 μl injection volume contained the desired cell dose. The endogenous mammary epithelium in the inguinal glands of 3-week-old female C57BL6/J or NSG mice was cleared and cells were injected into cleared fat pads as previously described [21]. The mice were mated 3 weeks after surgery and the number 4 glands were removed during pregnancy and fixed in Carnoys fixative and stained with carmine alum. An outgrowth was scored positive if it contained both lobular and ductal elements. MRU frequencies were calculated using the Extreme Limiting Dilution Analysis tool [22]). In some experiments, mice were kept in a virgin state for a total of 10 weeks and the number 4 glands removed for analysis by flow cytometry or histology.

Flow-sorted human mammary cells were seeded into 60 mm plates with 2.5 × 10⁵ irradiated NIH-3T3 feeder cells. The cultures were maintained in Human EpiCult-B (StemCell Technologies) supplemented with 5% FBS (StemCell Technologies) and 50 μg/ml gentamicin for 24 to 48 hours and then the media changed to serum-free conditions and maintained for an additional 10 to 12 days. Flow-sorted mouse cells were cultured in Mouse EpiCult-B and 50 μg/ml gentamicin in the presence of irradiated feeders for 5 to 7 days. At the end of the assays, the colonies were fixed with acetone:methanol (1:1), stained with Giemsa (Fisher Scientific, Cramlington, Northumberland, UK) and enumerated under a microscope.

In some experiments, the sorted cells were seeded within growth factor-reduced Matrigel and cultured in the presence of Human or Mouse EpiCult-B and irradiated feeders for 14 to 21 days. In some experiments, the culture media were changed after 7 days into differentiation media (DMEM/F12 with Glutamax (Gibco) supplemented with 10% FCS, 1 μM dexamethasone, 5 μg/ml insulin and 5 μg/ml prolactin) to induce lactogenic differentiation of the mammary cells, and the cultures were maintained for a further 7 to 14 days. At the end of the assay, the gels were then fixed in 4% paraformaldehyde and embedded in paraffin for sectioning and immunostaining.

Sorted cells were allowed to adhere to μ-Slide eight-well chamber poly-lysine-coated slides (Ibidi, Uddingston Glasgow, UK) for 15 minutes before fixation in 4% paraformaldehyde. Cells were blocked in 10% normal goat serum (Sigma) for 1 hour and stained with primary antibodies (Table S1 in Additional file 1) overnight at 4°C. Goat anti-mouse or anti-rabbit antibody conjugated to either AF488 or AF555 (Invitrogen) was used to detect primary antibodies. IgG antibodies at the same concentration as the primaries were used as isotype controls. Slides were stained with 4',6-diamidino-2-phenylindole to visualise the nuclei. Paraffin-embedded renal gels and Matrigel cultures were sectioned at 4 μm, deparaffinised and boiled in pH 6.0 citrate buffer. The sections were stained as above. Where required, a Mouse on Mouse (Vector Labs, Peterborough, Cambridgeshire, UK) preblocking kit was used as per the manufacturer's instructions.

Freshly sorted cells were pelleted and the supernatant removed. RNA was extracted using the PicoPure™ RNA extraction kit (Applied Biosystems, Paisley, Renfrewshire, UK) as per the manufacturer's instructions, and samples were treated with DNase using the RNase-free DNase Set (Qiagen, Crawley, West Sussex, UK). For mouse quantitative RT-PCR analysis, RNA from sorted cells of five independent experiments was collected. For human quantitative RT-PCR analysis, RNA from sorted cells of six different human breast specimens was collected. cDNA was generated using 100 ng RNA and random hexamers in a 20 μl reaction using SuperScript III (Invitrogen) according to the manufacturer's instructions. cDNA was diluted 1/10 and 1 μl was used in a 10 μl volume reaction with 2× SYBR Green Fast PCR Master Mix (Applied Biosciences) and 1 μl of 5 μM forward and reverse primers (Table S2 in Additional file 1) and H₂O. The real-time PCR reactions for each sample were performed in triplicate with an ABI 7900 Real Time PCR system under the following conditions: 95°C for 20 seconds followed by 40 cycles of 95°C for 1 second and 60°C for 20 seconds, followed by a dissociation run to obtain melt profiles of the amplicons. A no-template control (no cDNA) was used as a control for all primers, also performed in triplicate. Results were analysed with the delta-delta method normalised to two housekeeping genes (Ppia and Rpl13a or UBC and TBP for mouse and human samples, respectively) and compared with a comparator sample (nonclonogenic luminal (NCL) cells).

Total RNA was purified from freshly sorted cell populations using the PicoPure™ RNA extraction kit. Up to 250 ng RNA was labelled according to the standard Illumina gene expression array protocols with the Ambion TotalPrep 96 kit (4397949; Ambion, Paisley, Renfrewshire, UK). Biotinylated complementary RNA was quality controlled using Agilent Bioanalyser and quantified by spectrophotometry (Nanodrop, Ringmer, East Sussex, UK), and 750 ng cRNA was hybridised to Illumina Mouse6 or HumanHT12v4 BeadChips and washed, stained and scanned according to the standard protocol (WGGX DirectHyb Assay Guide 11286331 RevA; Illumina, Saffron Walden, Essex, UK). Arrays were scanned on an Illumina BeadArray scanner, and data were processed using the Bioconductor beadarray package [23]. (Further information can be found in Additional file 3.) All data files can be accessed via the Gene Expression Omnibus [GEO:GSE35399].

Data presented are the mean of multiple independent experiments and the standard error of the mean. One-way analysis of variance was used to test multiple groups followed by Tukey's post test to test significant differences between pairs of results. Comparisons between just two groups were analysed by Student's t test. Significance was set at * = P < 0.05, ** = P < 0.01 or *** = P < 0.0001.

To test the hypothesis that the LP population is a heterogeneous population, we dissociated mammary glands from 10-week-old virgin C57BL6/J females and analysed the liberated cells using flow cytometry to detect EpCAM, and CD49f (α₆-integrin; Figure 1A and Additional file 2A). We used EpCAM rather than the previously described CD24 [8, 9] since the use of EpCAM permits greater resolution of the luminal and basal cell subpopulations. We also used CD49b (α₂-integrin) instead of CD61 (β₃-integrin) to identify LPs. We observed that CD49b⁺ was a more selective marker of LPs than the previously reported CD61 [26] since up to 47% of progenitors are of CD61^- phenotype (Additional file 4A,B).

As shown in Figure 1A, the luminal cell compartment can be subdivided into three distinct subpopulations on the basis of expression of CD49b and Sca1. We observed a fourth population (Sca1^-CD49b^-) during pregnancy and remnants of this population are maintained throughout the involution and post-involution stages (region gates R9 and R7 in Figure 1B; see Additional file 4C). Flow sorting the three subpopulations from the virgin gland and seeding them into two-dimensional colony-forming cell (CFC) assays reveals that progenitor activity is restricted to the Sca1^-CD49b⁺ and Sca1⁺CD49b⁺ subpopulations (Figure 1C). Cloning efficiency was observed to be approximately 25% and 40% for Sca1⁺CD49b⁺ and Sca1^-CD49b⁺ progenitors, respectively. However, we suspect that these cell populations may be pure progenitor cells, since flow-associated toxicity is calculated to be as high as 75% (Additional file 4D).

Previously, gene expression profiling of sorted mouse mammary epithelial cells identified CD14, a co-receptor for bacterial lipopolysaccharide, as being highly enriched in the luminal cell population [8]. When we examined the distribution of CD14 expression among the luminal population, we observe that approximately 86% of LP cells express this protein, whereas CD14^- luminal cells are relatively deficient in CFCs (Additional file 4E,F,G). Other studies have reported that c-Kit expression identifies both ER^- and ER⁺ LPs [15] and progenitors that are primed to generate progeny that can synthesise milk proteins [16]. When examining c-Kit expression in luminal cells, we observed variation between mouse strains, with c-Kit expression localised to a minority of Sca1⁺ luminal cells, and only in FVB mice (Additional file 4H). However, we were unable to observe any significant expression of c-Kit among the luminal cells isolated from C57BL6/J mice, even when using the same c-kit antibody clone (2B8) that was previously used [15]. The c-kit expression levels in FVB mice in our study were quite low when compared with the other studies; one explanation for this discrepancy may be the use of different tissue dissociation protocols.

Immunostaining of sorted populations reveals that both the Sca1⁺CD49b^- and Sca1⁺CD49b⁺ cells express high levels of luminal differentiation markers such as ER and keratin (Krt)18 compared with the Sca1^-CD49b⁺ cells (Figure 1D,E). These results demonstrate that there are two functionally distinct types of ER cells in the mammary gland; the vast majority are relatively mature with little proliferation capacity, but a small population representing ~9% of all luminal cells are ER⁺ progenitors. The ER⁺ progenitors (Sca1⁺CD49b⁺) express higher transcript levels of luminal differentiation transcripts such as ER, FoxA1 and Gata3 and lower levels of Krt5 and Krt14 when compared with Sca1^-CD49b⁺ cells (Figure 1F). Immunostaining of Sca1^-CD49b⁺ (ER^-) cells demonstrates that these cells express lower levels of Krt18 and low but detectable levels of the basal cell-specific Krt5 (Figure 1D,E). These cells express no to low levels of ER, which is in agreement with a previously published report [17]. This intermediate level of expression for both luminal and basal cell markers suggests that these cells are a progenitor cell intermediate between the basal stem cells and the more differentiated ER⁺ LPs.

The ER^- progenitor subpopulation also has relatively higher levels of milk protein transcripts in the virgin state compared with the ER⁺ progenitor cells, including α-lactalbumin (Lalba) and milk fat globule-epidermal growth factor 8 (Mfg-e8), an observation consistent with previous reports [17]. These cells also express high levels of Elf5 and Lmo4, both of which have been involved with specifying alveolar cell fate (Figure 1F) [18, 27]. These results suggest that ER^- cells probably represent alveolar progenitors and are primed for milk production.

ALDH is an enzyme family previously reported to identify human mammary stem cells [28]. When we examined ALDH among the mammary cell populations, we observed that all of the ER^- and a subset of the ER⁺ LPs show high levels of enzyme activity, whereas the basal cells (EpCAM^loCD49f^hi) and NCL (Sca1⁺CD49b^-) ER⁺ cells show low or absent activity (Additional file 4I).

Previous studies have demonstrated that mammary stem cells are localised within the basal cell compartment of the mouse mammary epithelium since MRU-enriched populations have a basal cell-signature [8, 9]. However, a recent report has challenged the notion that basal cells are the most potent stem cells since a subpopulation of MRUs expressing high levels of the luminal cell differentiation marker CD24 can be also be detected [29]. To investigate this further, we double sorted the three luminal cell subpopulations to ensure purity and minimise contamination from other cell types (Additional file 5) and transplanted them into 25% Matrigel at limiting dilutions into cleared mammary fat pads of recipient mice. As shown in Figure 2A, MRUs are present within the luminal population, albeit at exceedingly low frequencies. No robust MRUs could be detected in the NCL cells, although occasional small ductal-lobular structures could be detected at very low frequencies (Figure 2A,B). Outgrowths derived from these LPs are morphologically normal and contain all of the subpopulations as those derived from basal cells, although a skewing in the distribution of luminal and basal cell populations towards the basal cells is observed when analysed by flow cytometry (Figure 2B; see Additional file 6A). Secondary transplantations reveal that four of the six primary outgrowths contained >5 MRUs, indicating that some of the luminal MRUs are potent and have extensive self-renewal capacity and can generate normal glands (Additional file 6B,C). However, when the distribution of MRUs among all of the mammary cell populations is calculated, we observe that approximately 99% of all MRUs present in a mouse mammary gland are localised within the basal cell population (Figure 2A).

Van Keymeulen and colleagues previously demonstrated that flow-sorted luminal cells can contribute to the luminal epithelium upon transplantation into cleared mammary fat pads, but only when co-transplanted with an approximate equivalent number of basal cells [10]. To determine which luminal cell population has this in vivo engrafting potential, double-sorted GFP⁺ luminal populations (ER^- progenitors, ER⁺ progenitors and NCL cells) were mixed with equal numbers of wild-type total mammary epithelial cells such that the final ratio of GFP⁺ marked luminal cells to wildtype basal cells was approximately 2:1 (Additional file 7). These cell mixtures were then transplanted into cleared mammary fat pads of NSG mice. Outgrowths containing GFP⁺ cells could be obtained for all mammary luminal cell populations, although, like the transplants described in Figure 2A, the frequency of this event was rare since only 1 in 30,000 to 340,000 sorted GFP⁺ luminal cells could engraft (Additional file 7). Unlike the previous report by Van Keymeulen and colleagues, the outgrowths generated in these experiments were not lineage-restricted since the engraftments contained both basal and luminal cells that expressed GFP (Additional file 8). Interestingly, when GFP⁺ cells engrafted, no engraftment by the co-injected wildtype basal and luminal cells was observed, even when nonlimiting numbers of basal cells were transplanted (Additional file 7). Further work is required to reconcile these two studies.

In an attempt to establish parent-progeny relationships between the different populations described in Figure 1A, we sorted the different luminal cell populations and seeded them into in vitro and in vivo assays. When ER^- and ER⁺ progenitor cells are seeded into three-dimensional Matrigel cultures and maintained for 3 weeks, we observed that the ER^- progenitors generated translucent alveolar-like structures that contained eosinophilic material in the lumen, whereas the ER⁺ progenitors generated alveolar-like outgrowths that were optically dense without any deposits (Additional file 6D). These results demonstrate that these two types of progenitors are functionally distinct. However, when these colonies were examined for expression of lineage markers such as Muc1 and p63, we observed that both progenitors can generate colonies that contain both luminal and basal cell lineages (Additional file 6D). When double-sorted ER^- and ER⁺ LPs were seeded within collagen/Matrigel gels or in 100% collagen gels and transplanted under the renal capsule of NSG mice, both populations generated outgrowths that contained both luminal (Muc1⁺ER⁺) and myoepithelial (Sma⁺p63⁺) cells in virgin mice and β-casein⁺ cells in pregnant mice (Figure 2C; see Additional file 9A). These outgrowths in virgin mice also contained MRUs since cells dissociated from these renal grafts could engraft multiple cleared mammary fat pads (Additional file 9B).

As a further check for high-fidelity sorting, donor cells (GFP^- or GFP⁺) were mixed with genetically tagged (GFP⁺ or GFP^-) cells (Additional file 9C) and the genotype of the resultant outgrowths was checked by immunohistochemistry for GFP (Additional file 9D). The outgrowths generated were positive for both luminal and basal markers and expressed the appropriate genotype (Additional file 9D,E). These results demonstrate that ER⁺ LPs have the potential to dedifferentiate to MRUs.

Both luminal and basal cells, and particularly mammary stem cells, were previously reported to be susceptible to withdrawal of oestrogen and progesterone [30]. To investigate the effects of oestrogen and progesterone withdrawal on the different subtypes of mammary luminal cells, 8-week-old C57BL6/J were ovariectomised and the change in the number of different mammary cell types was determined 3 weeks later.

The NCL cells were acutely sensitive to loss of oestrogen since the size of these populations decreased by 72%, with the proportion of these cells to total epithelial cells decreasing in ovariectomised mice when compared with sham-operated mice (Figure 3A,B,C). The basal, ER^- progenitor and ER⁺ progenitor subpopulations were all mildly sensitive to the withdrawal of oestrogen, decreasing by 46%, 47% and 37%, respectively (Figure 3A,B); no statistically significant effect was observed with any of the luminal populations. When the CFC content from ovariectomised and sham-operated mice was analysed, the ER^- progenitor population in ovariectomised mice contained 31% fewer progenitors than control mice, and the colonies that were generated were smaller (Figure 3D,E). The number and size of the resultant colonies derived from ER⁺ progenitors isolated from ovariectomised mice were marginally smaller when compared with controls, although these differences were not statistically significant. No effect of ovariectomy was observed on any of the non-epithelial (EpCAM^-CD49f^- and EpCAM^-CD49f⁺) cell populations.

These results demonstrate that all populations of mammary epithelial cells are sensitive to loss of steroid hormones, but the ER⁺ LPs are only mildly affected and have a survival advantage when compared with the other epithelial cell populations.

Previous studies have demonstrated that the luminal compartment in the human mammary epithelium can be divided into a luminal-restricted progenitor population (EpCAM⁺CD49f⁺) and mature NCL (EpCAM⁺CD49f ^- ) cells that express high levels of ER [31, 32]. To test the hypothesis that this LP cell compartment is heterogeneous, as in the mouse, we screened the expression of a variety of markers in 11 reduction mammoplasty samples using flow cytometry. We observed that the differential expression of ALDH and ERBB3 was able to resolve the LP population into not two but three subpopulations of cells: ALDH⁺ERBB3⁺ (ALDH⁺), ALDH^-ERBB3⁺ (ALDH^-) and ALDH^-ERBB3^- (ERBB3^-) (Figure 4A; see Additional file 2B,C - the threshold between ALDH^- and ALDH⁺ is set with reference to a control population incubated in the presence of ALDH inhibitor DEAB, whilst ERBB3 gating was determined using FMO controls). The proportion of these subtypes of cells is very variable between different donors, especially the ERBB3^- subpopulation, which can range in frequency from 2 to 65% of the total LP population (Figure 4B). However, only one-quarter of all patients have a distinctly identifiable ERBB3^- population (Figure 4A, middle and lower panels). This population appears to have no correlations with age, although other clinical parameters such as parity, menstrual stage and oral contraceptive use are not known.

All three subpopulations express the luminal-specific KRT8; they also express KRT5 but not KRT14 (Figure 4C). These latter two keratins have historically been considered specific for basal cells, although a recent report has challenged this notion [33]. Gene expression analysis by quantitative RT-PCR confirms that ALDH expression is highest in the ALDH⁺ population and lowest in the NCL cells (Figure 4D). When we examine the distribution of the luminal differentiation markers MUC1, AR and FOXA1 among the luminal cell populations, we consistently observe that the ERBB3^- population displays the lowest levels of luminal differentiation, followed by ALDH⁺, with ALDH^- and NCL cells displaying the highest levels of luminal differentiation (Figure 4D) - with the exception being that ALDH⁺ cells express the highest levels of MUC1 protein (Additional file 10A). Transcripts for ER and GATA3 are not significantly different in the NCL and ALDH^- subpopulations when compared with the ERBB3^- and ALDH⁺ subpopulations. The ERBB3^- LPs express the highest levels of the basal-specific genes KRT14, myosin light chain kinase and snail homolog 2a and lower levels of the luminal marker KRT19 when compared with the other LP populations. Interestingly, the ALDH⁺ subpopulation expresses the highest levels of transcripts for ELF5, MFG-E8 and LFT (lactoferrin), thereby suggesting that these cells are primed for milk production (Figure 4D).

It has previously been reported that human breast cancer stem cells often have an EpCAM⁺CD44⁺CD24^-/low phenotype [34]. To understand the normal cellular context of this signature, we used flow sorting to determine the distribution of CD24 and CD44 among the different human mammary epithelial cell populations described in Figure 4A. Our results demonstrate that none of the luminal cell populations have this phenotype and that this phenotype is restricted only to the basal cell population (Additional file 10B,C).

To interrogate the proliferative capacities of the three different LP populations, purified cells from each population were seeded into CFC assays. Results demonstrate that the ALDH⁺ subpopulation had the highest cloning efficiencies and contained the highest proportion of CFCs (Figure 4E). Surprisingly the ERBB3^- population had a very low cloning efficiency and contained an almost undetectable number of progenitor cells, which was unexpected since these cells have a relatively undifferentiated phenotype.

These results demonstrate that the ALDH⁺ subpopulation in the human mammary gland is analogous to the ER^- population in the mouse because both populations contain the highest proportion of progenitors and express high levels of ALDH1a3 and alveolar-associated genes (Figures 1C,F and 4D,E; see Additional file 4F).

To further characterise the growth and differentiation potential of the three LPs, we sorted these cells and seeded them into collagen gels that were then transplanted under the renal capsule of female NSG mice. All three subpopulations have the ability to generate hollow acinar multilayered structures (Figure 4F), albeit with vastly different efficiencies since the ERBB3^- subpopulation generated very few outgrowths. All three LPs gave rise to engraftments that contained both luminal (MUC1⁺GATA3⁺) and basal (p63⁺SMA⁺) cells (Figure 4F). Some engraftments from all populations generated single-layered structures that only contained luminal cells (Additional file 11A). Both ALDH^- and ERBB3^- progenitors could generate both ER⁺ and ER^- cells (Figure 4F). The ALDH⁺ cells, despite being able to generate GATA3⁺ cells, were unable to generate ER⁺ cells during the initial 5-week assay. However, ER⁺ progeny could be detected when the assay was extended for an additional 3-week period, thereby suggesting that ALDH⁺ cells are a primitive progenitor cell that needs additional time to generate all cell lineages (Additional file 11B). A similar pattern of expression was observed when the grafts were examined for expression of prolactin-induced protein, a protein whose expression occurs in the majority of ER⁺ breast tumours [35]. Both ALDH^- and ERBB3^- progenitors could generate prolactin-induced protein positive progeny, whereas ALDH⁺ cells were unable to do so during the 5-week assay (Figure 4F).

The LP population has previously been shown to have a gene expression signature resembling that of basal-like breast tumours, while the NCL cells resemble Luminal A/B tumours [32]. We hypothesised that subdividing the LP population would identify a closer relationship between the different types of mammary epithelial cells and the different breast cancer subtypes.

To test this hypothesis we sorted six different freshly isolated mammary cell populations (NCL, ALDH^-, ALDH⁺, ERBB3^-, basal and stromal) isolated from up to 11 mammoplasty samples and obtained gene expression profiles of these cells (Additional files 12 and 13). As expected, all three LP populations had gene profiles more similar to basal-like breast cancers than the other sorted breast cell populations (Figure 5A; see Additional file 14). Although the gene signature from the ALDH^- population most strongly correlates with basal-like cancers, it also has some correlations with the Luminal A and B signatures (Figure 5A). When we created a decision tree for the different luminal subpopulations, however, we observed that the gene signature of the ALDH⁺ population, and not the ALDH^- or ERBB3^- subpopulations, had the highest correlation with those obtained from basal-like breast tumours (Figure 5B). Consistent with what was previously reported, the gene expression signature of the NCL cells resembled those obtained from Luminal A/B breast cancer subtypes and the stromal cells resembled the claudin^low subtype (Figure 5A) [32]. To obtain a broader picture of the molecular characteristics of the different mouse mammary epithelial cell subpopulations, gene expression profiles of these cells were obtained. Results showed that the gene signatures of each luminal cell population are unique and distinct from basal cells (Figure 5C; see Additional file 15). Similar observations are seen when the microarray expression profiles of the purified mouse mammary cell populations are compared with those obtained from human breast tumours since the ER^- LPs have a gene expression profile that most resembles that of human basal-like breast tumours and the NCL ER⁺ cells resembling Luminal A/B tumours (Figure 5D).