Aldehyde dehydrogenase and estrogen receptor define a hierarchy of cellular differentiation in the normal human mammary epithelium

Normal breast tissue was obtained with informed consent from patients undergoing mammoplasty for aesthetic or prophylactic reasons, under protocols approved by the IRB and by Guy’s Research Ethics Committee, in compliance with the Human Tissue Act. The tissue was processed as previously described [12]. To generate a single-cell suspension for the in vivo studies, a shorter 6-hour collagenase digestion was used. Pieces of tissue were fixed in formalin for 24 to 48 hours before being processed and embedded in paraffin.

The ALDEFLUOR kit (StemCell Technologies Vancouver, Canada) was used according to manufacturer’s protocol. Cells were sorted by using a FACS Aria II (BD Biosciences, San Jose, CA) with 130-μm nozzle. Cell viability was assessed with LIVE/DEAD Fixable Violet Dead Cell Stain (Life Technologies, Carlsbad, CA). Sorted cells were cytospun onto glass slides for immunofluorescent analysis.

Cells were fixed with methanol for 20 minutes, washed with PBS, treated with 0.1% Triton X-100 for 5 minutes, and incubated in blocking buffer (PBS with 2% BSA) for 1 hour and subsequently stained with primary antibodies against ALDH1A1 (BD Biosciences, clone 44/ALDH, 1:50) and ALDH1A3 (Santa Cruz Biotechnology, Dallas, TX, clone C-13, 1:200) followed by secondary antibodies anti-mouse AlexaFluor-488 and anti-goat AlexaFluor-555 (Life Technologies, 1:500). Antibody incubations were done for 1 hour in blocking buffer. Nuclei were counterstained with DAPI.

Paraffin-embedded sections (3 μm) of primary or xenotransplanted normal breast epithelium were deparaffinized in xylene and rehydrated in graded alcohol. Antigen retrieval was achieved by heating slides in citrate buffer (Dako Glostrup, Denmark) according to recommendations. Sections were blocked with 10% donkey serum for 1 hour before incubation with primary antibodies in 10% donkey serum for 1 hour at RT. Antibodies used were ALDH1A1 and ALDH1A3 as above, ER raised in either rabbit (Novocastra (Leica) Wetzlar, Germany, 1:100) or in mouse (Dako, 1:100), SMA (Novocastra, 1:100), CK18 (Novocastra, 1:20), Ki67 (Dako, 1:100), MCM2 (Novocastra, 1:50), and RARα (Abcam Cambridge, UK, 1:250). Primary antibodies were detected either with fluorescent secondary antibodies (from Life Technologies, conjugated with AlexaFluor-488 or -555) or enzymatically by using Peroxidase Histostain-Plus Kit (Zymed South San Francisco, CA) or EnVision G2 Doublestain System (Dako), according to the manufacturer’s protocols. Nuclei were counterstained with DAPI and hematoxylin, respectively. For assessing the percentage of ALDH1A1+ and ALDH1A3+ cells detected by IHC, double immunostained tissue sections of normal breast (four different patients) were scanned by using the Hamamatsu Nanozoomer and analyzed by using Digital Images HUB (SlidePath system, Leica). Low-magnification images were used to delineate areas of epithelium. ALDH1A1+ cells, ALDH1A3+ cells, and total number of cells in each nonoverlapping area were counted. Each tissue section contained between 22,000 and 36,500 cells.

ALDEFLUOR-sorted cells were fixed in methanol and stained with antibody against ER (Thermo Scientific, Waltham, MA,, 1:100) followed by FITC conjugated anti-rabbit secondary (Jackson Laboratory West Grove, PA, 1:250). Antibody stainings were done for 20 minutes on ice in Hanks Balanced Salt Solution (HBSS, Gibco Life Technologies) supplemented with 2% FBS. Staining for viability was done by using 1 μg/ml propidium iodide (PI, Sigma St Louis, MO) for 5 minutes.

The ER reporter system, a generous gift from Dr. Polyak, Dana Farber Cancer Institute, was used to sort ER+ cells from normal mammary epithelium and was tested for specificity and sensitivity on ER+ and ER– breast cancer cell lines (MCF7 and MDA-MB-231). It consists of an adenovirus encoding GFP, under the control of 25 estrogen-responsive elements (EREs) and the rat progesterone promoter [14]. Cells were grown in phenol red-free, charcoal-stripped serum containing medium for 8 hours, after which the adenovirus was added in serum-free phenol red-free medium for an additional 15 hours. The medium was then removed and replaced with medium containing serum and β-estradiol (10⁻⁹M) for 48 hours. GFP expression was monitored microscopically. Flow-cytometry analysis and sorting was performed. Previous in vivo transplantation studies from the Werb [15] and Visvader [16] groups showed that this short in vitro cultivation and viral cell marking do not affect the potential and cell fate of human and mouse mammary epithelial cells [15, 16]. An aliquot of the sorted cells was analyzed microscopically for GFP expression and immunostained for ER to test the quality of the sorting. Cells were permeabilized with ice-cold methanol, and stained in suspension with ER and secondary antibodies, as described earlier.

Xenotransplantation of normal mammary epithelial cells sorted for ALDEFLUOR activity or with the ER reporter in humanized cleared mammary fat pads was performed as previously described [12, 17]. For assessing proliferation in vivo of ER+ cells, sorted cells were mixed with human mammary fibroblasts in equal ratios, suspended in gelatinous protein (Matrigel, BD Biosciences), and implanted in the cleared and humanized 4th inguinal mammary fat pads (25,000 to 800,000 cells/fat pad). Estrogen pellets (90-day release, 0.5 mg/pellet; Innovative Research of America Sarasota, FL) were implanted subcutaneously at the time of the clearing. Animals were killed after 2 months, and fat pads were fixed in formalin and embedded in paraffin for histologic analysis. The animal studies were performed under protocols approved by UCUCA and the Institutional Committees on Animal Welfare of the United Kingdom Home Office.

Primary human mammary epithelial cells sorted by using the ER reporter were placed on collagen-coated plates at a density of 2,000 viable cells/10-cm-diameter dish. Cells were grown in phenol red-free DMEM-F12 supplemented with 5% charcoal-stripped serum, 5 μg/ml insulin, 1 μg/ml hydrocortisone, 10 ng/ml EGF, 10 ng/ml cholera toxin, and 20 μg/ml gentamycin, and 1×antibiotic-antimycotic. Cells were fixed and immunostained after 12 days.

Mammosphere culture was performed as previously described [18]. In brief, primary human mammary epithelial cells were plated in ultra-low attachment plates (Corning, Corning, NY) or plates coated with 1% agarose, at a density of 20,000 viable cells/ml in primary culture and 5,000 cells/ml in subsequent passages. Medium used was mammary epithelial basal medium (MEBM, Lonza, Basel, Switzerland) supplemented with 20 ng/ml EGF, 1 μg/ml hydrocortisone, 50 μg/ml insulin, 100 μM 2-mercaptoethanol, 20 μg/ml gentamycin, and 1× antibiotic-antimycotic. The medium did not contain B27 or phenol red. Counting of mammospheres was done manually after 7 to 10 days in six-well plates under a light microscope in at least three wells for each condition. All primary cultures were incubated at 37°C, 10% CO₂.

After 7 to 10 days in culture, mammospheres were gently centrifuged at 200 rpm for 2 minutes at RT. The pellet was then resuspended in formalin and allowed to fix for 1 hour at RT. After fixation, mammospheres were centrifuged, and part of the supernatant was removed. An equal volume of melted 4% agarose (Gibco electrophoresis grade) in distilled water was then added to the formalin-fixed pellet (to a final agarose concentration of 2%). The volume was aspirated in a 1-ml syringe on which the top had been cut off and allowed to cool for 30 minutes at RT before the gel was expelled by using the syringe plunger. The gel was wrapped in lens tissue and paraffin/wax processed after routine tissue-processing procedures. Staining of sectioned mammospheres was done as described for paraffin sections.

Lentiviral vectors (pGIPZ) expressing shRNAs against ALDH1A1 and ALDH1A3 (Applied Biosystems, Life Technologies) were used to downregulate the expression of these genes. The vectors were transfected into the 293T packaging cell line by using calcium phosphate precipitate. After 36 hours, supernatants expressing lentivirus were collected, concentrated by using Amicon Ultra Centrifugal Filter Units, 100 kDa (Millipore Billerica, MA), and used to infect normal human mammary epithelial cells in adherent conditions, in serum-free medium. Two days of infection were performed (two cycles each day), at the end of which, cells were checked for GFP expression with fluorescence microscopy. Knockdown (KD) efficiency was determined by using Western blot.

In vitro 3D culture of human mammary epithelial cells was performed as previously described [19]. In brief, single-cell suspensions were aggregated overnight in MEBM in 24-well ultra-low-attachment plates at 100,000 cells/well. The next day, TC-treated 24-well plates were coated with 100 μl Matrigel diluted 2:1 in serum-free F12 media (Life Technologies) and allowed to solidify. Approximately 130 aggregates were collected by gentle centrifugation, resuspended in 300 μl Matrigel, diluted 2:1 in F12 containing 5% FBS, insulin (5 μg/ml), hydrocortisone (1 μg/ml), EGF (10 ng/ml), cholera toxin (10 ng/ml), gentamycin (20 μg/ml), and 1× antibiotic-antimycotic, and plated on top of the 100 μl Matrigel in the 24-well plates. Once solidified, 300 μl of F12 with FBS was added on top. Medium was changed every other day.

To identify in situ the ALDH+ human mammary epithelial cell population previously shown to be associated with stem/progenitor properties in functional assays, we sought to identify the ALDH isoforms associated with this enzymatic activity. To this end, we used the ALDEFLUOR assay and flow-activated cell sorting (FACS), as previously described [12]. We separated ALDEFLUOR+ and ALDEFLUOR– mammary epithelial cells from three different mammoplasty samples. We subsequently immunostained cytospins from these cell populations for ALDH isoforms. We found that the majority of ALDEFLUOR+ cells expressed either ALDH1A1 (38% ± 17%, mean ± SD) or ALDH1A3 (25% ± 5.7%, mean ± SD) (Figure 1A). No expression of ALDH1A1 and only very-low-intensity levels of ALDH1A3 were detected in a minority of ALDEFLUOR– cells (6% ± 2.5%, mean ± SD). The specificity of antibodies for the ALDH isoforms was tested after knockdown of ALDH1A1 and ALDH1A3, respectively, in cell lines (Additional file 1). The representation of ALDH1A1 and ALDH1A3 cells (cumulated percentages) was similar to the representation of the ALDH+ cell population detected by ALDEFLUOR in three out of four samples (Additional file 2), which further supported that these are the isoforms primarily responsible for the activity, although other isoforms are likely to contribute as well. Consistent with our findings, these two isoforms were found to be significantly elevated in mRNA levels when the ALDEFLUOR+ population was compared with the ALDEFLUOR– fraction in breast cancer cell lines [20]. Both ALDH1A1 and ALDH1A3 have been associated with poor prognosis in breast cancer [12, 21‐23]. ALDH1A3 was associated with primitive luminal progenitors in normal breast epithelium [24].

IHC analysis of multiple sections of normal breast from six different patients showed that ALDH1A1 and ALDH1A3 were never coexpressed in the same cells in the breast epithelium (Figure 1B, C). High levels of ALDH1A3 were present mostly in larger ducts, and low levels of ALDH1A3 in the smaller extra- and intralobular ductules (Figure 1D, E and Additional file 3D, E). ALDH1A1 was seen in intralobular terminal ductules or at branching points (Figure 1F, G and Additional file 3F-I). Several studies failed to identify an ALDH1A1+ mammary epithelial cell population [24‐26]. However, our findings, as well as those from several other groups [27‐29], demonstrate unequivocally the existence of this cell population. Considerable differences in technical approach may account for these contradictory results. We henceforth refer to ALDH enzymatically active cells identified by the ALDEFLUOR assay as ALDE+ cells, and to cells identified as positive by immunostaining as ALDH+ (if referring to both isoforms) or ALDH1A1+ and ALDH1A3+, according to the isoform investigated.

Lim et al. [25] showed that stem/progenitor cells from the normal human breast identified by flow cytometry as CD49f-high/EpCAM-/Lin- do not express ER. In another study [30], Eirew et al. demonstrated that mammary epithelial cells bearing the same markers generated in vivo outgrowths that contained ER+ cells. Taken together, these two studies suggest that human mammary stem/progenitor cells are ER– but generate ER+ cells. Similar results were obtained by the Kuperwasser group by using the CD10+/EpCAM– phenotype to identify stem/progenitor cells [31]. We tested whether this conclusion holds true in our experimental approach, which uses ALDH-based markers.

IHC analysis of normal breast epithelium from 11 patients (five to 17 sections examined per sample) showed that neither ALDH1A1 nor ALDH1A3 co-localizes with ER (Figure 1D-G and Additional file 3D-I). In three of the 11 mammoplasty samples analyzed, however, very low levels of ER were detected in a small minority of ALDH1A1+ cells (2.4% ± 0.5% (mean ± SD) ER-low cells in the ALDH1A1+ population, 0.36% ± 0.2 in the total population) (Additional file 3F-J). These may represent a transient, very rare cell population detectable only in a subset of mammoplasty samples. Flow-cytometry analysis of six mammoplasty samples showed no ER expression in ALDE+ cells (Additional file 3A). This was further supported by immunostaining for ER on ALDE-sorted cells (Additional file 3C). These results are consistent with our previous studies in BRCA1 healthy carriers [32].

We previously showed that ALDE+, but not ALDE– cells, from the normal breast epithelium generate outgrowths in immunodeficient mice [12]. To test whether ALDE+/ER– cells generate ER+ cells in vivo, we sorted ALDE+ cells from four normal mammoplasty samples and implanted 25,000 cells into humanized cleared mammary fat pads of NOD/scid mice. This cell number was chosen to increase the frequency of rare progeny and optimize their detection, based on our previous findings regarding ability of ALDE+ cells to generate outgrowths [12]. As expected, ductal outgrowths were observed in all the transplanted fat pads. Figure 2A and B shows a representative H&E-stained section through a ductal outgrowth generated by ALDE+ cells in vivo. Consistent with what we previously described [12], ducts contained two cell layers, an outer layer that stained positive for smooth muscle actin (SMA), a marker of the myoepithelial lineage (Figure 2C, D) and an inner layer that stained positive for the luminal lineage marker cytokeratin (CK) 18 (Figure 2E). ER+ cells were present in outgrowths from all four mammoplasty samples analyzed (Figure 2F). Similar results were obtained in vitro. Mammospheres generated by ALDE+/ER– cells contained ALDH1A1+ and ER+ cells (Figure 2G, H).

We enquired subsequently whether, in our experimental system, ER+ cells are only “sensory” cells or if these cells also possess progenitor potential, as has been indicated by studies in mice using surrogate markers that enrich for ER+ cells [26, 33]. ER+ cells from normal breast epithelium were identified and separated with FACS by using a reporter system built on estrogen responsive elements (ERE) and the rat progesterone promoter (Additional file 4). When plated at clonogenic densities both ER+ and ER– sorted cells generated colonies in adherent conditions (Figure 3A-C, E-G, R), but only ER– cells formed mammospheres in suspension culture (Figure 3D, H-I, S). These mammospheres contained ER+ cells, demonstrating that ER+ cells arise from ER– progenitors. In adherent culture on a collagen substratum, ER+ cells generated progeny bearing only luminal markers (CK18, MUC1, EpCAM) (Figure 3A-C, R) while ER– cells or cells sorted on viability alone generated mixed luminal-myoepithelial colonies containing both CK14+ and CK18+ cells as well as pure luminal (EpCAM+) and myoepithelial (CD10+) colonies (Figure 3E-G, R). The ER– cell population contained bipotent progenitor cells, which generated cell colonies displaying mixed-lineage markers. The ER– cell population and cells sorted on viability alone served as controls, showing that the mammary epithelial cells used in these experiments had the capacity to generate both myoepithelial and luminal cells. The restricted potential of ER+ cells was not due to the culture conditions or the manipulation of cells during sorting.

Additional evidence that ER+ cells can proliferate in culture comes from the colocalization of ER with proliferation marker Ki67 in mammosphere sections (see Additional file 5). In normal breast sections, the percentage of ER+ cells that express proliferation markers is very low [3, 34]. In culture, ER expression is downregulated within 4 to 5 days in standard 2D culture [8, 35] and within 7 days in Matrigel drip culture [35]. In mammosphere sections, ER expression can be detected after 10 to 12 days in culture (Additional file 5).

To investigate the ability of ER+ cells to proliferate in vivo, ER+, ER–, and unseparated live-cell populations were sorted from three different mammoplasty samples and implanted in cleared humanized fat pads of NOD/scid mice (25,000 to 800,000 cells/fat pad). These cell numbers were chosen based on previous findings [12, 17] regarding growth potential of mammary epithelial cells in xenografts, taking into account that no enrichment in stem/progenitor cells was expected. All the cell populations generated outgrowths (see Additional file 6), although not in all fat pads transplanted with 25,000 ER+ cells. Outgrowths from ER+ cells had only one layer of CK18+ luminal cells, which contained both ER+ and ER– cells (Figure 3J-M). Myoepithelial markers were not detected in any of the outgrowths from ER+ cells (Figure 3L and data not shown). ER– cells and unseparated cells from the same samples generated outgrowths in all implanted fat pads. These outgrowths were composed of two cell layers, consisting of CK18+ luminal cells surrounded by a layer of SMA+ myoepithelial cells (Figure 3N-P), indicating that the differentiation potential of the cells was not altered by experimental manipulations. These outgrowths contained a subset of ER+ cells (Figure 3Q).

Taken together, these data indicate that the ER– cell population contains a subset of cells that can generate ER+ cells. ER+ cells are able to proliferate and generate outgrowths that are restricted to differentiation along the luminal lineage.

To assess whether ALDH activity is necessary for the functionality of mammary stem/progenitor cells, we knocked down ALDH1A1 and ALDH1A3 in primary human mammary epithelial cells by using lentivirus-mediated shRNA interference (Figure 4A). We tested the effect of this downregulation on mammosphere formation. We found that ALDH1A1 knockdown (KD) significantly reduced the number of primary and secondary spheres formed in suspension culture compared with nontargeting (NT) control in five different mammoplasty samples (40% to 90% reduction). ALDH1A3 KD modestly decreased sphere formation (25% reduction) (Figure 4B and Additional file 7).

The biochemical function of ALDH is to convert retinol to retinoic acid (RA); this represents the only source of RA in tissues. Therefore, we tested whether treatment with RA could rescue the KD effect of ALDH1A1. Mammosphere formation in the ALDH1A1 KD cell cultures treated with 1 μm all-trans retinoic acid (ATRA) was restored to levels seen in control cultures (Figure 4C). ATRA treatment of control mammary epithelial cells in suspension culture did not affect mammosphere formation.

We also investigated the expression of retinoic acid receptor alpha (RARα) in the mammary epithelium and found that, consistent with previous reports [36, 37], RARα was ubiquitously expressed in the nuclei of mammary epithelial cells (see Additional file 8). It appears, therefore, that the effects of RA are controlled by the restricted expression of ALDH isoforms in small islands of epithelial or stromal cells.

We demonstrated that ER+ cells are generated by ALDE+ cells. We further investigated whether ALDH KD affected generation of ER+ cells by assessing representation of ER+ cells in mammospheres after silencing ALDH1A1 and ALDH1A3 with shRNAs. We detected significantly fewer ER+ cells in spheres derived from ALDH1A1 KD cells compared with spheres from nontargeting controls or ALDH1A3 KD cells (Figure 4D,E). Given that ER has been found to be a target of RA [38], our findings can be due to a direct effect on ER expression. Branching morphogenesis in 3D Matrigel culture was abolished by ALDH1A1 KD in three different mammoplasty samples (Figure 4F), although mammary epithelial cells proliferated in these conditions and formed spheres or structures with very limited branching. Interestingly, RA treatment of the Matrigel culture completely abolished branching morphogenesis of mammary epithelial cells, without blocking their proliferation. It is not clear what accounts for this effect and if it may be an in vitro artifact. Together, the effects of ALDH1A1 KD on mammosphere formation and branching morphogenesis demonstrate that this isoform has a functional role in regulating human mammary stem/progenitor cell proliferation and/or early differentiation.