Mice

WAP-Cre mice (Jax#008735) [23] and Rosa26-lox-Stop-lox-YFP mice [24] were backcrossed up to four generations on an FVB background. All of the mice used in this study contain one allele each of the WAP-Cre and Rosa-lsl-YFP transgenes. For timed matings, female mice were placed in the cage of a male after 5 p.m. and checked for vaginal plugs at 9 a.m. the following morning (day 0). Mice were euthanized by carbon dioxide inhalation and immediately dissected for thoracic (#3), abdominal (#4), and inguinal (#5) mammary glands at the indicated time points. All animal protocols were approved by the SingHealth Institute for Animal Care and Use Committee.

Imaging and cryosectioning of whole-mounted mammary glands

Thoracic glands were whole mount-photographed on an Olympus SZX-12 fluorescent stereoscopic microscope with green fluorescent protein (GFP) filter (Olympus, Tokyo, Japan), and images were acquired by DP2-BSW software through an Olympus DP72 CCD detector. One #3 gland was subsequently fixed overnight in 4% buffered formaldehyde (ICM Pharma, Singapore) for paraffin embedding. The other #3 gland was fixed 1.5 hours in 2% formaldehyde in phosphate-buffered saline, prior to embedding in Tissue Tek OCT (Sakura Finetek, Tokyo, Japan) for cryopreservation. Cryosections (15 to 30 μm) were cut on a Leica CM1950 cryostat onto SuperFrost Plus-coated slides (Menzel-Gläser, Braunschweig, Germany), and stained with Rhodamine-conjugated phalloidin (Molecular Probes, now part of Invitrogen Corporation, Carlsbad, CA, USA) in accordance with the instructions of the manufacturer. Sections were mounted in Vectashield fluorescence mounting media (Vector Laboratories, Burlingame, CA, USA), and images were acquired on a Zeiss 710 confocal microscope (Carl Zeiss, Jena, Germany) with a pinhole aperture of 1 airy unit.

Indirect immunofluorescence

Fixed #3 mammary glands were processed and embedded in paraffin wax. Paraffin sections of 5 μm were prepared and subjected to 1 mM disodium-ethylenediaminetetraacetic acid (EDTA) antigen retrieval as described previously [21]. Primary antibodies used for immunofluorescence are the following: cytokeratin-8 (CK8) (TROMA-I, rat, 1:100; Developmental Studies Hybridoma Bank, Iowa City, IA, USA), estrogen receptor (NCL-ER-6 F11, mouse, 1:100; Novocastra, which is part of Leica, Wetzlar, Germany), GFP (600-401-215, rabbit, 1:100; Rockland Immunochemicals Inc., Gilbertsville, PA, USA), GFP (600-141-215, goat coupled to Dylight-488, 1:300; Rockland Immunochemicals Inc.), progesterone receptor (MAB9785, rabbit, 1:400; Abnova, Taipei, Taiwan), and smooth muscle actin (SMA) (A2547, mouse, 1:1,000; Sigma-Aldrich, St. Louis, MO, USA). Secondary antibodies used at 1:400 dilution are from Invitrogen Corporation: Alexa488-coupled goat anti-mouse (A11029), Alexa488-coupled goat anti-rabbit (A11034), Alexa568-coupled goat anti-mouse (A11031), and Alexa568-coupled goat anti-rabbit (A11036). Additionally, CF633nm-coupled donkey anti-rat (20137–1; Biotium, Hayward, CA, USA) was used at 1:400 dilution.

Isolation of primary mammary epithelial cells for fluorescence-activated cell sorting, RNA, and genomic DNA

Abdominal and inguinal glands were pooled and processed either for single mammary epithelial cell isolation for FACS analysis as previously described [21] or else processed just to the organoid stage for purposes of harvesting RNA and genomic DNA. At this point, three quarters of the organoid suspension was pelleted for lysis in PureZOL RNA isolation reagent (Bio-Rad Laboratories, Inc., Hercules, CA, USA), and total RNA then was prepared in accordance with the instructions of the manufacturer. The remaining quarter was pelleted and resuspended in 0.1 M Tris-Cl pH 8/0.2 M NaCl/5 mM EDTA/0.2% SDS/0.2 mg/mL Proteinase K for overnight digestion at 50°C. Genomic DNA was precipitated out from the digest with 1 volume isopranol, washed in 70% ethanol, and resuspended in 100 μL of water. A hundred nanograms of genomic DNA was amplified per sample, 35 cycles with Taq polymerase (Fermentas, Vilnius, Lithuania).

Cell labeling, flow cytometric analysis, and fluorescence-activated cell sorting

Fluorochrome-conjugated antibodies were titrated on primary mammary epithelial cells to ensure maximal positive signal-to-background fluorescence ratio. Anti-mouse or anti-rat compensation beads (BD 552843 and 552845, respectively) or both were used for single-stain antibody controls. Compensation controls also included two cellular samples: unstained cells and cells stained with 4′,6-diamidino-2-phenylindole (DAPI) (D8417; Sigma-Aldrich). Cells were incubated with antibodies on ice for 45 minutes with agitation every 15 minutes. Samples were then washed with twice the sample volume and resuspended in L15 (Gibco-Life Technologies, now part of Invitrogen Corporation) with 6% fetal calf serum (FCS) (HyClone, Logan, UT, USA) and 200 ng/mL of DAPI, except non-DAPI compensation controls. All multiple-labeled samples were gated on FSC-A versus SSC-A and doublet discrimination (FSC-H versus FSC-W and SSC-H versus SSC-W) and DAPI negativity. Samples contained anti-CD45 to exclude lymphocytes from analysis. Cells were analyzed and sorted on a BD FACS-Aria II containing 355 nm UV, 488 nm blue, 561 nm yellow-green, and 633 nm red lasers (BD Biosciences, San Jose, CA, USA). Specific antibodies used and gating strategy are detailed in Additional file 1.

Synthesis of cDNA and quantitative polymerase chain reaction analysis

For analysis of transcript levels by quantitative polymerase chain reaction (qPCR) in FACS-sorted populations, cells were sorted directly into lysis buffer (10 IU RNase inhibitor (Invitrogen Corporation), 2 mM dithiothreitol (DTT), 0.15% Tween-20 (Bio-Rad Laboratories, Inc.) in 12 μL of nuclease-free water) in PCR tubes by using a direct reverse transcription (RT) method described by Ho and colleagues [25]. Five hundred cells were sorted into each tube, and RT was performed by using Superscript VILO (Invitrogen Corporation) in accordance with the protocol of the manufacturer. In the case of PureZOL-purified total RNA harvested from epithelial organoids, cDNA was prepared by using the Bio-Rad iSelect kit (Bio-Rad Laboratories, Inc.) and a pool of gene specific RT primers (each at 0.42 mM final concentration), comprising the reverse primers for each of the amplified genes: WAP, Cre transgene, YFP transgene, Elf5, and hypoxanthine-guanine phosphoribosyltransferase (HPRT). Primers were designed that span introns to exclude the detection of genomic DNA and selected for optimum melt curve and amplification profiles. The primers to detect the WAP-Cre transgene expression were designed from the Cre sequence and rabbit β-globin polyA sequence which spans the β-globin intron that is spliced out in MECs (data not shown). All primer sequences are listed in Additional file 2. In the case of the intronless YFP transcript, a no-reverse-transcriptase (NRT) control reaction was run in parallel with the normal RT reaction, and qPCR background signal from the NRT reaction (due to genomic DNA contamination in those samples with low RNA yields) was subtracted from the RT reaction quantification for YFP in the reported results. qPCR was performed by using Sso Fast Evagreen supermix reagent (Bio-Rad Laboratories, Inc.) in accordance with the protocol of the manufacturer. To assess genomic recombination of the Rosa-lsl-YFP locus, primer sequences flanking the loxP sites were designed such that the recombined locus could be amplified from organoid genomic DNA as a 578-bp band (Additional file 3), whereas the unrecombined locus would generate a 3,244-bp band (data not shown).

Genetic labeling of parity-identified mammary epithelial cells

To investigate the cellular identity and lineage potential of PI-MECs in intact mammary glands, we chose to label them with a fluorescent reporter. The labeling of PI-MECs was first described by Wagner and colleagues [15, 23] and makes use of the Cre/lox system to permanently label cells by genetic recombination. They created a transgenic mouse line expressing the Cre recombinase under the control of the WAP gene promoter [23]. We crossed these with a reporter strain that contains the YFP gene separated from the constitutive Rosa26 promoter by a stop sequence flanked by loxP sites (Rosa26-lox-stop-lox-YFP, or Rosa-lsl-YFP) [24]. Transcriptional activation of the WAP promoter induces expression of Cre recombinase [16], which excises the transcriptional stop sequence in the reporter construct under control of the constitutively active Rosa26 promoter (Figure 1). From that point onwards, the cell and all of its progeny permanently express the YFP reporter gene.

We crossed WAP-Cre mice with the Rosa-lsl-YFP strain and collected mammary glands from double transgenic female mice before, during, and after pregnancy (Figure 2). Thoracic mammary glands were examined by wholemount fluorescence microscopy (Additional file 4) and by cryosections (Figure 2) to confirm the proper functioning of the PI-MEC labeling system. The cryosections were counterstained with phalloidin to visualize the alveolar structures independent of their YFP status. At day 7 of the first pregnancy, we detected no evident YFP expression by wholemount analysis (Additional file 4A) and very few YFP-positive (YFP^pos) cells on cryosections (Figure 2A, see Additional file 5 for close-ups of individual cells). In only one mouse out of five harvested at this stage did we observe extensive YFP labeling of epithelial structures (later determined by FACS to represent approximately 6% of epithelial cells; Additional file 6), which may be attributable to minor alveolar development as part of normal estrus cycling, as previously described [15]. At 14 days of pregnancy, we detected unambiguous activation of the YFP reporter gene (Figure 2B), with about half of alveoli containing at least one YFP-expressing cell and a small percentage of alveoli consisting of more than 90% YFP^pos cells. This timing precedes reports of labeling using the LacZ reporter, which was undetectable at 14 days and started at only 15 days [11] or 18 days of pregnancy [15]. This could be due to more efficient floxing of the YFP reporter compared with the LacZ reporter or a difference in the sensitivity of detection of the labeled cells. We could detect WAP mRNA, and in parallel Cre mRNA, in primary MEC preparations at day 7 of the first pregnancy (Figure 2G). Consistent with the observed Cre expression and appearance of more YFP^pos cells, we could detect recombination of the genomic Rosa26 locus by PCR at pregnancy day 7 (Additional file 3) which is followed at day 14 by detectable accumulation of mRNA expression for YFP (Figure 2G). As expected, WAP (and Cre) expression reached maximum levels during lactation. Expression of the alveolar specification gene Elf5 [26] reflects the increase in the number of alveolar cells generated during pregnancy and precedes the appearance of YFP, which is dependent on the accumulation of sufficiently high levels of the Cre recombinase in the individual cells (Figure 2G).

At 3 days post-partum, virtually every alveolus contained only YFP-expressing cells (Figure 2C, see also Additional file 5C), consistent with complete recombination of the Rosa-lsl-YFP locus at this stage and underscoring the efficiency of the WAP-Cre/YFP labeling system. Larger ducts of glands at 3 days lactation often remained negative for YFP, in contrast to secondary ducts or side branches that contain multiple YFP^pos cells (Figure 2C, top panel and Additional file 7). Six weeks after weaning, the alveolar structures had been cleared by involution. Only the ductal network and regressed terminal branches remained in the post-involution glands, and YFP^pos cells were readily detectable in involuted mammary epithelium (Figure 2D). These YFP^pos cells are referred to as PI-MECs, cells present in parous tissue that survived involution. In contrast, a large proportion of cells that were initially labeled during pregnancy terminally differentiated and were removed. Thus, not all YFP^pos cells that are present during pregnancy and lactation will become PI-MECs. In parous epithelium, WAP and Cre expression dropped by several orders of magnitude, consistent with loss of detectable WAP protein at this stage [19]. Notably, the baseline WAP mRNA expression remains higher after involution than in virgin mice (Figure 2G), consistent with a recent study from the Bentires-Alj laboratory [27]. Overall, these results were concordant with observations with previously published LacZ and GFP reporters [11, 15, 22, 23, 28].

At day 7 of the second pregnancy, alveoli emerged from the involuted ductal network. In contrast to the first pregnancy, some of these contained mostly YFP^pos cells, whereas others were unlabeled (Figure 2E). The majority of these YFP^pos cells are likely progeny generated by PI-MECs (this will be addressed later). By day 14 of the second pregnancy, most alveoli were completely labeled by YFP (Figure 2F) but with a marked proportion of alveoli that were still partially labeled or even unlabeled. This suggests that, in addition to PI-MEC-derived alveoli which are already YFP-labeled, some alveoli were derived from unlabeled progenitors that progressively undergo YFP recombination and activation, perhaps at a somewhat accelerated pace compared with the first pregnancy (compare Figure 2B and F). The increase in cells with a de novo recombined YFP reporter coincides with the renewed induction of WAP (and in parallel Cre) during the second pregnancy (Figure 2G, note the logarithmic scale).

Parity-identified mammary epithelial cells are luminal cells that express markers of alveolar progenitor cells

After validation of the reporter system, we set out to characterize the cellular identity of PI-MECs by FACS analysis by using cell surface markers for the various epithelial populations of the mammary gland. Six weeks after weaning, involuted mammary glands were processed to single cells, and luminal and basal MEC populations were identified by staining for CD24 and α6-integrin (CD49f) [21, 29], after exclusion of doublets, dead cells, and lymphocytes (Figure 3A and Additional file 1B). Analysis of four individual WAP-cre;Rosa-lsl-YFP animals showed that the YFP^pos cells (PI-MECs) fall squarely within the luminal gate (Figure 3B). Plotting the luminal and basal population on separate histograms for YFP further highlights the lack of PI-MECs in the basal population (Figure 3C). Within the luminal population, PI-MECs represent roughly half of the population (Figure 3C and D). Immunofluorescence staining on sections of involuted mammary glands confirmed the exclusive localization of PI-MECs to the luminal layer (Figure 3E and F). Cells identified with an antibody recognizing YFP also expressed the luminal cell-specific marker CK8 but never the basal cell-specific marker SMA. In contrast to the previously reported basal identity of cultured PI-MECs [22], these data definitively establish PI-MECs as a luminal cell type in intact mammary glands.

To further evaluate the cellular identity of PI-MECs within the luminal population, we separated the luminal cells into hormone-sensing cells (Sca1^hiCD49b^lo) and estrogen receptor (ER)-negative cells (Sca1^loCD49b^hi) [21, 29] (Figure 4A). The ER-negative Sca1^loCD49b^hi population contains most of the progenitor activity in the luminal population as measured by colony-forming ability [29] and likely contains progenitor cells in different stages of lineage commitment. For clarity, and based on our results described below and the enrichment for markers such as beta-casein and Elf5 [21, 30], we refer to the ER-negative Sca1^loCD49b^hi population as ‘alveolar progenitor cells’.

FACS analysis of MECs isolated from involuted mammary glands and stained for Sca1 and CD49b showed that about a quarter of luminal cells fell within the gate for hormone-sensing cells, whereas the remaining three quarters were found within the alveolar progenitor gate (Figure 4B). YFP-negative (YFP^neg) cells belonged to both cell types, but the vast majority of YFP^pos cells were part of the alveolar progenitor population (Figure 4C). Only around 6% of YFP^pos cells were part of the hormone-sensing cell population (Figure 4D). This rare population expresses ERα mRNA to the same extent as the collective hormone-sensing cell population (Additional file 8), suggesting that their FACS profile truly reflects a hormone-sensing identity of these cells. Notably, even though almost all YFP^pos cells belonged to the alveolar progenitor population, about 40% of the alveolar progenitor population in the involuted mammary gland lacked YFP expression (Figure 4D). Since secretory alveoli were close to 100% YFP^pos at lactation (Figure 2C), this YFP^neg population may represent alveolar progenitors that resided in the primary ducts, where most cells are YFP^neg even during lactation (Additional file 7). In summary, there appear to be three major populations of luminal cells in involuted mammary epithelium: (a) YFP^neg hormone-sensing cells, (b) YFP^neg alveolar progenitor cells, and (c) YFP^pos alveolar progenitor cells (PI-MECs). Immunofluorescence analysis on tissue sections confirmed the presence of all of these three types of cells and showed their juxtaposition within the luminal layer of ducts in involuted mammary glands (Figure 4E).

We have optimized a direct lysis method for qPCR analysis of limited numbers of cells [25] and used this method to validate the molecular identity of the cell populations identified by FACS. Estrogen receptor alpha (ERα) and progesterone receptor (PR) expression was largely limited to hormone-sensing cells (Figure 4F), whereas both YFP^pos and YFP^neg alveolar progenitor populations exclusively expressed the alveolar marker genes Elf5 and β-Casein [26] (Figure 4G). PI-MECs thus clearly belong to the luminal alveolar lineage within the involuted mammary epithelium.

Initially, we performed a microarray on YFP^pos and YFP^neg luminal cells in order to identify unique cell surface markers of PI-MECs to facilitate future studies of this cell population without the need for generating parous double transgenic mice. However, the presence of a significant proportion of YFP^neg alveolar progenitor cells that express Elf5 and β-casein to the same extent as the YFP^pos cells explains why we were unable to identify unique markers for PI-MECs. Apart from YFP, which was enriched approximately 30-fold and validates the sorting procedure, no other transcripts were significantly enriched in the YFP^pos population (data not shown). This suggests that the YFP^pos and YFP^neg alveolar progenitor populations share their transcriptome profile beyond Elf5 and β-Casein.

Taken together, our data show that in involuted mammary epithelium apart from PI-MECs there exists an unlabeled alveolar progenitor population with a similar transcriptional and cell surface profile. To address the question of whether this unlabeled alveolar progenitor population is functionally equivalent to PI-MECs, we examined the relative contribution of these populations toward lobuloalveologenesis in a second pregnancy (Figure 5).

An unlabeled pool of alveolar progenitor cells is equipotent to parity-identified mammary epithelial cells

To evaluate whether PI-MECs or the unlabeled alveolar progenitor cells were the main source for the development of new alveoli, we analyzed WAP-Cre;Rosa-lsl-YFP females at day 7 of their second pregnancy. In typical lineage-tracing techniques, Cre is activated for a short time period following administration of a chemical inducer [31]. In contrast in PI-MEC labeling, Cre expression is controlled by WAP promoter activity and therefore the progeny of PI-MECs can be traced only during developmental time periods when WAP is not expressed. We chose the 7-day second pregnancy time point because alveologenesis is already apparent but WAP expression is not significantly induced yet ([15, 16] and Figure 2G). Thus, YFP^pos alveoli at this time point likely originate from PI-MECs and are not the result of de novo activation of the reporter, which occurs again later in the second pregnancy (Figure 2F).

In involuted glands, roughly half of the alveolar progenitor population was YFP^pos (Figure 4D). Strikingly, at day 7 of the second pregnancy, newly forming alveoli stained for YFP either almost completely or not at all (Figure 2E and Figure 5A-C, more examples in Additional file 9). Therefore, besides the PI-MECs that give rise to YFP^pos alveoli, there is a significant contribution of unlabeled cells that can form morphologically indistinguishable alveoli. Notably, the number of alveoli derived from PI-MECs or unlabeled alveolar progenitor cells varied widely between animals. To quantify this effect, a cryosection of an entire mammary gland from each of the three mice was scanned on a confocal microscope and reconstituted digitally (see Additional file 10 for an example). Each individual alveolus from a section was then scored into one of three categories: YFP^neg, YFP^pos, or partially labeled by YFP. To be considered YFP^pos, at least 90% of the cells in the alveolus must express YFP, since we had noticed that alveoli frequently contain a few ER-positive hormone-sensing cells which are YFP^neg (Figure 6). In all three 7-day pregnant samples, the vast majority of alveoli were scored as either YFP^pos or YFP^neg (Figure 5A-D and Additional file 11). This all-or-nothing distribution suggests a clonal contribution from either an unlabeled alveolar progenitor or a PI-MEC rather than a mixture of alveolar progenitors contributing to the same alveolus. The small proportion of alveoli that had a mixture of YFP^pos and YFP^neg cells could signify the mixed contribution of a labeled and an unlabeled progenitor, but these partially labeled alveoli could also reflect early de novo activation of the reporter. Indeed, reporter activation due to induction of WAP expression at 14 days of the first pregnancy shows a similar stochastic pattern and the majority of alveoli that become labeled at this point belong to the YFP^partial category (Figure 5E and Additional file 11). Together with the radically different ratio of partial-versus-totally labeled alveoli (1:3 at day 7 of the second pregnancy compared with 5:1 at day 14 of the first pregnancy), these data fit best with a model whereby at day 7 of the second pregnancy the frequency of de novo floxing is still low and the majority of YFP^pos cells are generated by PI-MECs.

Overall, the relative contribution of PI-MECs to alveologenesis, as measured by the fraction of total YFP^pos alveoli, varied from 4% to 79%. A similar variation was found when four independent animals were analyzed by FACS (27% to 77%, Figure 6H). It is currently unclear whether this variation is stochastic or whether certain conditions favor one population over the other. It should be noted, however, that the females in Figure 5A and B, which show the most divergence in PI-MEC contribution, are littermates who shared nearly identical histories. The proportion of YFP^pos alveoli was roughly similar in the right and left mammary glands of each individual mouse (data not shown). This suggests that the relative contribution of PI-MECs and unlabeled alveolar progenitors to alveologenesis is due to systemic regulation rather than independent gland-autonomous effects.

In summary, even though PI-MECs contribute significantly to the generation of new alveoli, a substantial proportion of newly developing alveoli are unlabeled. Therefore, the unlabeled alveolar progenitor cells that are present after involution not only are very similar to PI-MECs in their transcriptome and cell surface markers but also have the ability to contribute to alveologenesis to the same extent as PI-MECs.

Basal and hormone-sensing cells of developing alveoli are not generated by parity-identified mammary epithelial cells

At low magnification, clusters of alveoli appeared to be clonally derived from PI-MECs (for example, Figure 5B and Additional file 9); however, at a higher magnification, it became apparent that even though the majority of cells in these alveoli were YFP^pos, they consistently contained a small fraction of YFP^neg cells. To investigate the cellular identity of these YFP^neg cells, we stained paraffin sections with antibodies against YFP, ER, and CK8. Figure 6A shows a typical example from the 7-day second pregnancy time point, whereby all luminal cells in an alveolus are YFP^pos, except the few cells that express the ER (in red) (more examples in Additional file 12). CK8-negative basal cells are also YFP^neg, like hormone-sensing cells (Figure 6A). The small proportion of alveoli that are almost fully recombined at 14 days of the first pregnancy also contain a low proportion of YFP^neg cells (Figure 6B). We noted that the signal intensity for ER is already reduced by 7 days of pregnancy and becomes virtually undetectable at 14 days, and therefore we used an antibody against progesterone receptor for sections from 14-day pregnant animals. Again, the majority of YFP^neg cells are hormone-sensing cells (Figure 6B), similar to the 7-day second pregnancy time point (Figure 6A). Thus, even though all luminal alveolar cells derive from a common progenitor cell (based on the clonal appearance at 7 days of the second pregnancy), developing alveoli contain cells from different lineages; PI-MECs are able to contribute all ER-negative luminal cells to a cluster of alveoli, but the hormone-sensing and basal cells are derived from different lineages (see Figure 7 for a schematic representation of these observations).

Since PI-MECs can give rise to all mammary epithelial cell types in transplantation experiments [11, 15, 22], we further evaluated their lineage potential in vivo during a normal pregnancy by FACS analysis (Figure 6C). Separating the mammary epithelial population of WAP-Cre;Rosa-lsl-YFP mice at 7 days of the second pregnancy into YFP^neg and YFP^pos subpopulations showed that the majority of YFP^pos cells remained restricted to the luminal cell population (Figure 6D). Only 0.29% ± 0.03% of PI-MEC-derived cells were found in the basal cell compartment, indicating that PI-MECs cannot be the source of the expansion of the basal layer during pregnancy. A closer examination of the luminal subpopulations showed that during pregnancy the hormone-sensing cells lose their CD49b and especially their Sca1 expression (Figure 6E versus Figure 4A), consistent with a previous report [29]. The cellular identity of the FACS populations was again validated by qPCR. Even during pregnancy, more than 98% of PI-MECs remained firmly within the luminal alveolar progenitor population (Figure 6F and G). These data show that, as a rule, PI-MECs remain restricted to the luminal alveolar lineage during pregnancy and do not contribute significantly to the other lineages.

Out of all the hormone-sensing cells analyzed by FACS at the 7-day second pregnancy time point, only 1.7% ± 1.2% were YFP^pos (Figure 6H). Even though the percentage of YFP^pos cells in the hormone-sensing cell gate was small, it is not noise in the FACS data, because rare cells that were positive for both ER and YFP were detectable by confocal microscopy of tissue sections from pregnant mice (Figure 6I).

Taking these findings as a whole, we conclude that, in contrast to the clear multi-lineage potential of PI-MECs in transplantation assays, the lineage potential of PI-MECs in vivo is almost entirely restricted to luminal ER-negative cells, even during pregnancy.

Parity-identified mammary epithelial cells are restricted luminal estrogen receptor-negative progenitors

We show that PI-MECs are a strictly luminal population, both by FACS analysis (including qPCR validation of the different populations) and by confocal microscopy on tissue sections from primiparous and pregnant mice. Matulka and colleagues [22] reported that PI-MECs were located at the tip of the basal cloud, the same position as the mammary stem cells. They used a GFP reporter strain to label the PI-MECs and made use of the same cell surface markers (CD24 and CD49f), but an important difference is that the cells were cultured for 3 days before FACS analysis. Based on the reproducibility of our results in all animals examined and the agreement between different methods, we conclude that PI-MECs are a luminal cell type.

In line with the labeling based on expression of one of the milk genes (WAP), we found that PI-MECs have an alveolar identity, based on cell surface expression (Sca1^loCD49f^hi) and gene expression (Elf5, β-casein). Consistent with previous reports, PI-MECs contribute significantly to alveolar development in pregnancy, and we therefore designate them luminal alveolar progenitor cells.

Unexpectedly, we noted that, at least in primiparous mice, there is a population of YFP^neg alveolar progenitor cells that displays characteristics similar to those of PI-MECs. Possibly, these cells resided in the larger ducts during lactation where hardly any cells were YFP-labeled, in contrast to the almost complete labeling of the secretory cells in the alveoli (Additional file 7). Interestingly, there seemed to be a gradient of reporter activation in the ducts because we detected quite a number of YFP^pos cells in the smaller ducts of lactating glands. This raises the question of whether PI-MECs might be part of an alveolar progenitor reservoir that is located in the ducts rather than representing fully committed secretory cells that dedifferentiated and reintegrated into the ducts during involution, but a different experimental system is required to address this.

The use of the WAP promoter to label PI-MECs creates a technical challenge to investigate PI-MEC progeny, because new YFP^pos cells could be either daughter cells of PI-MECs or the result of activation of the WAP promoter. Even though we found a general increase in baseline activity of the WAP promoter in parous MECs (similar to [27]) and in early pregnancy, our data indicated that de novo floxing activity was low at day 7 of the second pregnancy and we used this time point to analyze the lineage potential of PI-MECs. We used both FACS and confocal microscopy to analyze the progeny generated by the YFP-labeled alveolar progenitors in early pregnancy and found that PI-MECs can produce all the steroid receptor-negative luminal cells of an alveolus; however, as a rule, they do not generate basal or hormone-sensing cells of alveoli (Figure 7A). The more restricted lineage potential observed in recent lineage-tracing strategies based on intact mammary glands compared with previous transplantation studies has prompted the revision of current models of the mammary stem cell hierarchy [10, 32]. Previously, PI-MECs have been designated (limited) stem cells based on their contribution to multiple lineages in cleared fat pad transplantations [11, 14, 32]. The data presented here warrant a revision of the place of PI-MECs in the hierarchy: strictly into the luminal lineage (Figure 7B). Note that our observation that PI-MECs generate luminal but not basal cells is consistent with a study showing that in intact mammary glands basal cells maintain the basal layer and luminal cells maintain the luminal layer [10].

Even though it is clear that PI-MECs are remarkably restricted to the luminal alveolar lineage, we observed a small percentage of labeled cells that belonged to the luminal ER⁺ population by FACS and in tissue sections of early pregnancy. It remains to be determined whether this reflects a low level of plasticity of PI-MECs in situ (Figure 7B) or whether hormone-sensing cells induce WAP expression in particular circumstances. Interestingly, the multipotency of PI-MECs in reconstitution assays suggests that the conditions of transplantation can unlock their lineage restriction. However, cells from the basal layer have the most robust long-term multipotent potential in transplantation assays [9, 33], whereas the reconstitution potential of PI-MECs or luminal populations identified by other methods is more limited [11, 33], suggesting that some lineage restriction remains even in transplantation assays. It will be interesting to investigate what factors control the lineage potential of these different cell populations both in intact mammary glands and in reconstitution assays and to what extent transplant conditions reflect circumstances relevant for tumorigenesis and metastasis.

Alveoli are formed by collaborating cell types

An alternative version of the mammary stem cell hierarchy has been proposed on the basis of transplantation experiments that show ductal-limited and lobule-limited mammary epithelial outgrowths [3]. Instead of a division into a basal and a luminal lineage as discussed above [32], this alternative model proposes that stem cells give rise to a ductal- and lobule-restricted lineage whereby the lobule-restricted progenitors are thought to give rise to both the basal and luminal cells of alveoli [6, 13]. We did not detect the activity of a bipotent lobule-restricted progenitor, because we did not observe any YFP^pos basal cells by histology, and less than 0.1% of basal cells were YFP^pos by FACS analysis and could be technical noise. Consistent with our observation that the basal cells in alveoli are derived from a different source are recent lineage-tracing experiments with the Axin2-promoter whereby Cre-ER was activated briefly before puberty. In those experiments, clonal Axin2-traced offspring contributed exclusively to the basal layer of multiple clusters of alveoli [34]. In contrast, when Cre activity was briefly induced in adult females, the authors found evidence that Axin2-traced progeny could contribute to both the luminal and basal layers of alveoli [34] (Figure 7B). Interestingly, after involution, Axin2-traced cells remained in the luminal layer, suggesting that PI-MECs could potentially be generated from Axin2-traced basal cells. The Axin2-lineage tracing experiments demonstrate that bipotent activity can be detected in intact mammary glands of adult mice; however, it is unclear how common this activity is. Our data show that PI-MECs make a significant, though variable, contribution to alveologenesis, but this does not include basal alveolar cells.

Moreover, PI-MECs generally do not generate the luminal ER-positive cells, even when all ER-negative luminal cells of an alveolus are derived from PI-MECs. Interestingly, lineage-tracing of cells that expressed Notch2 in puberty revealed two previously unrecognized cell types—small string cells and large L cells—and these cell types do not fit in current models of the stem cell hierarchy [35]. The cells from the Notch2-lineage are functionally required for alveologenesis, and the authors showed that each alveolus contains a small proportion of L cells. An inverse picture of what we have observed was found in Notch2-lineage tracing experiments in early pregnancy: alveoli in which only a single L cell is labeled and the remaining luminal alveolar cells are unlabeled and thus derived from a different source [35]. In addition, hormone-sensing cells play an important role in the early stages of alveologenesis due to their secretion of paracrine factors such as receptor activator of nuclear factor kappa-B ligand (RANKL) and insulin-like growth factor 2 (IGF-2) [21, 36–38]. Together, these observations warrant the investigation of steroid-receptor status of Notch2-lineage traced L cells, and if they prove to be ER-positive, they could be incorporated into the hormone-sensing lineage of the stem cell hierarchy.

Overall, the Notch2-lineage tracing data and the experiments presented here highlight that, as a rule, alveoli develop through collaboration of different cell types (Figure 7B-1) rather than being generated as clonal units from a single progenitor. In the case of cooperative growth, basal cells generate the basal cells of alveoli and luminal alveolar progenitors such as PI-MECs produce luminal ER-negative cells. Hormone-sensing cells may be recruited in small numbers from the ducts and are the likely instigators of alveolar outgrowth [20, 21]. The Axin2-lineage tracing data suggest that under certain circumstances alveolar luminal and basal cells can derive from the same source (Figure 7B-2). It will be interesting to determine whether this includes ER-positive cells and how these different modes for alveologenesis, collaboration versus clonal origin, are regulated.