Notch3 marks clonogenic mammary luminal progenitor cells in vivo

Mammary epithelial cells form a network of branching ducts composed of a luminal and a basal myoepithelial layer. Luminal cells are subdivided into ductal cells, which line the mammary lumen, and alveolar cells, which give rise to alveolar units at pregnancy. The mammary gland shows remarkable plasticity, giving rise to a milk-producing organ at pregnancy (Oakes et al., 2006), which reverts to a virginlike state by responding to specific apoptotic signals during involution, when milk production is no longer needed (Watson, 2006). Central to this morphogenetic cycle are thought to be mammary stem cells (Williams and Daniel, 1983; Smith and Chepko, 2001), initially suggested to reside in the basal layer of the mammary duct (Shackleton et al., 2006; Stingl et al., 2006). More recent studies afforded new insights into the morphogenesis of the mammary epithelium, revealing hierarchical cell lineage relationships. A subset of luminal cells with specific regenerative capacity were recently uncovered (Sleeman et al., 2007; Regan et al., 2012; Shehata et al., 2012; Šale et al., 2013), and elegant in vivo lineage-tracing experiments confirmed the existence of two distinct adult stem cell populations in the mouse mammary gland, one for each compartment of the ductal bilayer, which give rise only to their respective compartment of origin (Van Keymeulen et al., 2011).

Breast cancer is the most common tumor in women, and the vast majority of breast tumors, particularly the poorly differentiated ones, are thought to arise from luminal progenitors (Prat and Perou, 2009; Molyneux et al., 2010). Thus, elucidating the lineage hierarchies that contribute to the development of the mammary gland is not only essential in understanding the morphogenesis of the gland but also in gaining insights into the cellular origins of mammary tumors, knowledge that may have therapeutic implications. Yet, the high cellular heterogeneity of the mammary epithelium and, most importantly, the lack of robust markers, have made the analyses of lineages quite challenging.

Notch pathway activity is known to be associated with stem cells in many tissues, such that it has been successfully used to trace lineages in the intestine and, more recently, in the mammary gland (Fre et al., 2011; Šale et al., 2013). This pathway controls cell differentiation and proliferation decisions throughout development and adult homeostasis, by linking the fate of one cell to that of its neighbors. Cell-to-cell communication is achieved through the interaction of the Notch receptor expressed on one cell with membrane-bound ligands expressed on adjacent cells. All four Notch receptor paralogues, Notch1, 2, 3, and 4, have been shown to be expressed in the developing mammary gland (Bouras et al., 2008; Raouf et al., 2008; Raafat et al., 2011). Given the association of Notch with stem and progenitor cell populations, the expression of each paralogue may be used as a lineage marker. We have thus generated new transgenic mice that allowed us to use an in vivo lineage tracing approach to follow Notch-related cell lineages (Fre et al., 2011; Šale et al., 2013). We recently reported the tracing of Notch2 lineages in the mammary gland uncovering two hitherto unidentified epithelial lineages, revising significantly our understanding of mammary morphogenesis.

In this study, we focused on Notch3, a receptor that has been recently associated with the highly aggressive “triple negative” breast cancer (Yamaguchi et al., 2008; Turner et al., 2010; Speiser et al., 2013) and, of relevance for the present study, was found to be involved in the maintenance of stem cell quiescence in several tissues (Kuang et al., 2007; Kitamoto and Hanaoka, 2010; Kent et al., 2011; Mourikis et al., 2012; Alunni et al., 2013). Our analysis reveals that Notch3 is expressed in highly clonogenic luminal progenitor cells that are mostly found in a nonproliferative state but can get activated during mammary gland development to produce luminal daughter cells that can survive multiple cycles of pregnancy and involution. Our results using gain-of-function Notch3 mutant mice suggest that the proliferation of these cells is controlled by Notch3 activity.

To conditionally label Notch3-expressing cells in vivo, we relied on our reporter strain Notch3-CreERT2^SAT in which a CreERT2 cassette is knocked in into the first exon of the Notch3 locus (Fre et al., 2011). To follow cell lineages, we crossed them to a double-fluorescent reporter line, R26^mTmG (Muzumdar et al., 2007). In Notch3-CreERT2^SAT/R26^mTmG mice, membrane-bound GFP (mG) marks cells in which Cre-mediated recombination has occurred after hydroxytamoxifen induction (4-hydroxytamoxifen [4-OHT]). Notch3-CreERT2^SAT/R26^mTmG bigenic mice were injected with a single dose of 4-OHT (1 mg/20 g of body weight) at different developmental stages. The analysis of mice 24 h after Cre induction allowed us to detect the cells that express the Notch3 receptor but not their progeny. Even though the CreERT2 system is intrinsically mosaic, preventing us to target all Notch3-expressing cells, the clear enrichment in Notch3 expression we found in GFP⁺-sorted cells (Fig. S1 B) indicates that these cells represent a subpopulation of mammary luminal cells that expresses the Notch3 receptor. At this time point, we detected histologically rare marked cells along the mammary ducts, representing 0.4 ± 0.03% of the total luminal compartment (Fig. S1 A). All of the Notch3-expressing cells at all developmental stages are marked by the luminal marker CK8 (cytokeratin 8) and do not express basal CK5 (cytokeratin 5; Fig. 1, A–J), confirming and extending previous expression studies (Bouras et al., 2008; Raafat et al., 2011). In addition, we observed that GFP-labeled cells consistently show strong nuclear levels of the polycomb group transcriptional repressor Bmi1 (Fig. S1, C and D), commonly associated with uncommitted progenitor cells in diverse tissues (Lessard and Sauvageau, 2003; Molofsky et al., 2003; Park et al., 2003).

The expression of membrane-tethered GFP in the R26^mTmG reporter mouse allows visualization of cellular protrusions otherwise indistinguishable from neighboring cells. By confocal microscopy, we observed that several GFP⁺ cells extended long cellular protrusions that crossed the myoepithelial layer and came in close proximity with the basement membrane (Fig. S1, E and F). Similar protrusions have been previously described and suggested to be a characteristic of progenitor cells (Smith and Chepko, 2001; Oakes et al., 2006). Given that the ligands for Notch receptors appear to be differentially localized between the stroma and the myoepithelial layer of the mammary gland (Xu et al., 2012), it is conceivable that these protrusions may serve the purpose of achieving selective receptor activation and signaling by providing access to local cues from the microenvironment.

To further characterize the Notch3-expessing luminal cells in vivo, we induced Notch3-CreERT2^SAT/R26^mTmG mice with a single dose of 4-OHT at different developmental stages and followed their lineages at distinct time points after induction (Fig. S2). Regardless of the developmental stage of the mice at induction, Notch3-labeled cells invariably give rise to clusters of luminal cells. When mice were induced before puberty, at 3 wk of age, only single GFP⁺ luminal cells could be detected 24 h after 4-OHT administration (Fig. 1, A and B; and Fig. 2 A). 3 wk after this prepubertal induction, the labeled cells expanded, generating small clones of GFP⁺ cells (Fig. 2 B). An additional chase of 3 wk to adulthood (6-wk chase) resulted in further expansion of these clones (Fig. 2 C). The same type of clonal expansion was observed when mice were induced during puberty, at 6 wk of age (unpublished data). Induction of adult mice, after completion of mammary gland development, followed by a chase of 3 mo without going through pregnancy, also generated clusters of GFP⁺ luminal progeny (Fig. 2 D). Thus, Notch3-expressing cells persist for a long time, eventually giving rise to luminal lineages during adult homeostasis. Of note, the GFP⁺ progeny of single Notch3-expressing cells is heterogeneous for the expression of progesterone receptor (PR; Fig. 2, E and F) and estrogen receptor α (ER-α; Fig. 2, G and H), indicating that one cell can give rise to different lineages within the same clone. The analysis of the expression of these hormone receptors 24 h after induction shows that GFP⁺ cells are also heterogeneous in their expression of ER-α and PR (Fig. S3). To examine whether all GFP-labeled cells have progenitor properties, we quantified the rate of their clonal expansion during puberty. Mice were induced at four weeks of age and were analyzed every week (n = 3 mice per time point) until the age of 8 wk (Fig. S2). This analysis revealed that 70% of Notch3-expressing luminal cells expand and give rise to clones, but 30% remain as single cells over the course of pubertal development, which lasts 4 wk (Fig. 2 I and Table S1). Conversely, the quantification of clonal expansion of CK8-expressing luminal cells, in the same developmental window, indicated that only 40% of luminal cells give rise to clones, as previously shown (Van Keymeulen et al., 2011). This important difference supports the notion that, unlike the ubiquitous luminal marker CK8, Notch3-CreERT2^SAT mice may target a specific luminal cell subpopulation with progenitor properties.

When we analyzed Notch3 lineages during pregnancy, we found that they contributed to the formation of alveolar buds, independent of the age of induction (Fig. 2 J). However, both GFP-positive and -negative luminal cells are found in alveoli, reflecting the polyclonal origin of these structures. To further examine whether Notch3-expressing cells were able to persist during mammary gland remodeling in vivo, we analyzed mice that had undergone three successive pregnancies, lactations, and involutions and found that Notch3-derived luminal lineages survived cell death during involution, suggesting that they derived from single GFP⁺ progenitors capable of self-renewal (Fig. 2, K and L).

To investigate the clonogenic capacity of Notch3-expressing cells, we isolated by FACS GFP⁺ cells from 6-wk-old females, after a 24-h induction with 4-OHT (Fig. S1 A), and seeded them on a feeder layer of irradiated fibroblasts. When we compared their in vitro colony-forming potential to that of GFP⁻ luminal cells, we observed that GFP⁺ cells formed clones with a 25% higher frequency (Fig. 3 A). To further characterize the GFP-marked cells, we tested whether they express CD49b (α-integrin), a recently established luminal progenitor marker (Shehata et al., 2012). After the protocol of Shehata et al. (2012), we subdivided the luminal compartment in three cell populations, based on their expression of CD49b and Sca1 (CD49b⁻/Sca1⁺, Cd49b⁺/Sca1⁺, and CD49b⁺/Sca1⁻). The distribution of GFP⁺ cells in different luminal fractions using these two markers did not correlate Notch3 expression with a specific luminal subset (Fig. 3 B). Accordingly, quantitative RT-PCR (qRT-PCR) showed Notch3 expression in all three luminal subpopulations (Fig. 3 C). We then probed the colony-forming capacity of CD49b⁺ and CD49b⁻ luminal cells and found that GFP⁺ cells within the clonogenic CD49b⁺ population had a 75% increased cloning efficiency compared with CD49b⁺/GFP⁻ luminal cells (Fig. 3 D). We confirmed that CD49b⁻/Sca1⁺ cells are nonclonogenic in this assay (Fig. 3 D, CD49b⁻), as previously reported (Shehata et al., 2012). On the basis of this analysis, we propose that Notch3 is expressed in luminal progenitor cells with high clonogenic potential.

To determine the proportion of Notch3-expressing cells that are actively cycling, we asked what percentage of GFP⁺ cells is marked by the proliferation marker Ki67. Surprisingly, the vast majority of Notch3-expressing cells were negative for Ki67 at different developmental stages (prepuberty [Fig. 4 A], puberty [Fig. 4 B], and adulthood [Fig. 4 C]). The quantification of cycling GFP⁺ cells at puberty showed that 98% of them were Ki67 negative (Fig. 4, A–D and G). The difference with total luminal cells was more prominent when we quantified Ki67 expression of GFP-labeled cells within the terminal end buds (TEBs; Fig. 4, D and H), where normally ∼50% of total luminal cells are proliferative. On the other hand, when we examined GFP-marked clones derived from Notch3-expressing cells 3 and 6 wk after induction at puberty, ∼12 and 20%, respectively, of these cells expressed Ki67 (Fig. 4, E–G), indicating that Notch3-expressing cells are not terminally differentiated but get activated asynchronously to proliferate and produce the observed clonal expansion of GFP⁺ cells over time (Fig. 2 I).

To better define the apparent quiescent state of Notch3-expressing cells, we analyzed their capacity to incorporate BrdU. Pubertal Notch3-CreERT2^SAT/R26^mTmG mice were injected with 4-OHT and with 50 µg/ml BrdU on day 0. Mice were then kept on continuous BrdU feeding for either a consecutive 3 d and sacrificed immediately after (short pulse; Fig. 4, I–K) or for 2 wk and sacrificed 3 wk thereafter (long incorporation; Fig. 4, I , L, and M). Although the majority of GFP⁺ cells was negative for BrdU (98% in ducts and 65% in TEBs) after the short BrdU pulse (Fig. 4, J, K, and N), both BrdU-negative (Fig. 4 L) and BrdU-positive (Fig. 4 M) clones were observed after the long incorporation chase, further indicating that these cells are not terminally differentiated and that they can enter the cell cycle asynchronously. Importantly, GFP⁺ cells show elevated expression levels of the cell cycle inhibitors p21/WAF1 and Ink4/Arf compared with GFP⁻ luminal cells (Fig. 4 O), further corroborating their nonproliferative state. Our results reveal that Notch3 gene expression marks, in vivo, a luminal progenitor cell population that is maintained in a transiently nonproliferative state but can reenter the cell cycle, responding to yet unknown cues, to expand and give rise to luminal daughter cells.

Based on these observations, we were prompted to ask whether a functional correlation exists between levels of Notch3 activity and the proliferative state of GFP-labeled cells. To examine this, we expressed a constitutively active Notch3 receptor in Notch3-expressing cells and their progeny, by crossing Notch3-CreERT2^SAT/R26^mTmG mice to a gain-of-function mouse line, in which the intracellular form of Notch3 (N3IC) is conditionally inserted into the ROSA26 locus (R26-N3IC^SAT; Fig. 5 A). We then quantified the clonal expansion in triple transgenic mice Notch3-CreERT2^SAT/R26^mTmG/R26-N3IC^SAT after a 4-wk chase. We found that in mice expressing N3IC, the number of clones generated was reduced by almost 50% compared with the clones generated in control Notch3-CreERT2^SAT/R26^mTmG mice (Fig. 5 B and Table S1). The reduced ability of these cells to generate marked progeny paralleled an increase in the number of single cells found 4 wk after induction (Fig. 5 B, blue histogram). No overt phenotype was observed, probably because Notch3-CreERT2^SAT mice target very few cells (0.1% of mammary epithelial cells). Expression of the N3IC transgene was evident both by the presence of the floxed allele in genomic DNA extracts in the presence of 4-OHT (Fig. 5 C, lanes 3 and 5) and by the elevated expression of intracellular Notch3 without significant changes in extracellular Notch3 levels, as detected by qRT-PCR using appropriate oligonucleotides (oligos; Fig. 5 D). Importantly, we observed that N3IC-expressing cells show increased expression of the cell cycle inhibitors Ink4/Arf and the cyclin-dependent kinase inhibitor p21/Waf1 (Fig. 5 E), which are also strongly expressed in cells targeted by Notch3-CreERT2^SAT (Fig. 4 O), further confirming their nonproliferative state. These results are consistent with the notion that sustained Notch3 activation can keep luminal cells in a transiently quiescent state. Consequently, we propose that luminal progenitors need to down-regulate Notch3 signaling to ensure luminal lineage expansion, as illustrated in the model depicted in Fig. 5 F. In this context, it is worth considering that previous studies in the muscle and in cell cultures suggested a functional role for Notch3 signaling and Hes1, a canonical Notch target gene, in maintaining cellular quiescence (Kuang et al., 2007; Sang and Coller, 2009; Kitamoto and Hanaoka, 2010; Kent et al., 2011; Mourikis et al., 2012; Alunni et al., 2013).

The role of Notch signals in oncogenesis has often been attributed to their positive influence on cell proliferation (Kiaris et al., 2004; Fre et al., 2009; Artavanis-Tsakonas and Muskavitch, 2010). Such a role presents an apparent contradiction with the results we obtained in this study because Notch3 activity seems to keep cells in a nonproliferative, resting state in the normal mammary gland. However, it is conceivable that in pathological conditions, Notch3 may contribute to increased survival of a cellular population, by protecting it against drugs that target highly proliferative cells.

The generation of Notch3-CreERT2^SAT knockin mice has been previously described (Fre et al., 2011). The targeting vector included a genomic fragment covering the promoter region, exon1, and exon2 of the Notch3 locus. Exon1, between the translation start site and the end of exon1, was replaced with a cassette containing the CreERT2 ORF, a polyadenylation signal and a flippase recognition target–flanked neomycin resistance gene. This results in expression of tamoxifen-inducible CreERT2 recombinase in all cells in which the promoter of Notch3 is active. These mice were crossed to a double-fluorescent reporter mouse strain R26^mTmG (Muzumdar et al., 2007), which was obtained from Jackson ImmunoResearch Laboratories, Inc. To generate R26-N3IC^SAT mice, the mouse the N3IC cDNA (amino acids 1,668–2,318) was amplified by PCR, sequenced, and cloned directly into a ROSA26 targeting vector. This targeting construct included a lox-STOP-lox cassette at the 5′ end of the N3IC sequence and a polyadenylation sequence at the 3′ end. This construct was electroporated into embryonic stem cells (C57BL/5NTac). All experiments and procedures involving animals were in strict accordance with the French and European legislations for the Protection of Vertebrate Animals used for Experimental and other Scientific Purposes and were approved by the Departmental Direction of Populations Protection (approval number: B75-05-18). Husbandry and supply of animals as well as maintenance and care of the animals in exempt of pathogen species environments before and during experiments fully satisfies the animal’s needs and welfare. Suffering of the animals has been kept to a minimum: no procedures inflicting pain have been performed.

Mice were genotyped by PCR using the following primers: Notch3-CreERT2^SAT 1228_23, 5′-ATAGGAACTTCAAAATGTCGCG-3′, and 1239_22, 5′-CCCAGCTGCTGCATCTCTGC-3′ (187 bp); Notch3 wild type 1239_21, 5′-TCGTCGCCTGATGGCCTTGC-3′, and 1239_22, 5′-CCCAGCTGCTGCATCTCTGC-3′ (244 bp); R26-N3IC^SAT 1336_5, 5′-GCATGGTGGAAGAGCTCATC-3′, and 1336_6, 5′-GTCCAAGTGATCTGTGATCTCC-3′ (257 bp); ROSA26 wild type 1260_1, 5′-GAGACTCTGGCTACTCATCC-3′, and 1260_2, 5′-CCTTCAGCAAGAGCTGGGGAC-3′ (585 bp); R26^mTmG Rosa1, 5′-AAAGTCGCTCTGAGTTGTTAT-3′; Pcag-r105, 5′-GTCGTTGGGCGGTCAG-3′ (250 bp); and N3IC floxed forward, 5′-CCTCCTGGCTTCTGAGGAC-3′, reverse (unfloxed), 5′-CTCGTCCTGCAGTTCATTCA-3′ (504 bp), and reverse (floxed), 5′-ACCTCCCCCATCAGACTCTC-3′ (432 bp).

Notch3-CreERT2^SAT/R26^mTmG and control Notch3-CreERT2^SAT females ranging from 3 to 9 wk of age were induced with a single i.p. injection of 4-OHT (1 mg/20 g mouse body weight; H7904; Sigma-Aldrich). For expression analysis, animals were sacrificed 24 h after induction. For lineage-tracing experiments, animals were sacrificed at different time points ranging from 48 h to several months after induction. For each time point, four mammary glands of at least three mice were analyzed. No GFP expression was observed in noninduced transgenic mice.

Freshly dissected mammary glands from Notch3-CreERT2^SAT/R26^mTmG females were fixed at room temperature with 4% PFA in PBS for 2 h and either embedded in optimal cutting temperature medium (VWR International) or embedded in paraffin. Samples were sectioned at 10 or 20 µm. For immunofluorescence staining, frozen sections were permeabilized with 0.2% Triton X-100–PBS for 45 min and blocked for 1 h with 5% FBS and 2% BSA. The following primary antibodies were used: chicken anti-GFP (1:1,000; ab13970; Abcam), rabbit anti-CK5 (1:1,000; PRB-160P; Covance), rabbit anti-PR (1:300; sc-7208; Santa Cruz Biotechnology, Inc.), and rabbit anti-Ki67 (1:300; ab15580; Abcam). Paraffin-embedded sections were rehydrated through a gradient of ethanol. Subsequently, antigen retrieval was achieved by boiling in 10 mM citrate buffer (20 min) for all antibodies. Staining was performed as mentioned previously for frozen sections. The following primary antibodies were used: mouse anti-CK8 (1:200; MMS-162P; Covance), mouse anti–ER-α (1:300; M7047; Dako), mouse anti-BrdU (1:200; 347580; BD), mouse anti–BMI-1 (1:100; 39993; Active Motif), mouse anti–α-smooth muscle actin (1:500; A2547; Sigma-Aldrich), and rabbit anti–Collagen type IV (1:80; AB756P; EMD Millipore). Secondary antibodies were incubated in blocking buffer for 1–2 h at room temperature. The following secondary antibodies were used: anti–chicken Alexa Fluor 488 (1:1,000; A-11039; Invitrogen), anti–rabbit Alexa Fluor 633 (1:1,000; A-21071; Invitrogen), anti–mouse Alexa Fluor 633 (1:1,000; A-21202; Invitrogen), and goat anti–mouse Cy3 (1:500; 115–165-003; Beckman Coulter). Nuclei were stained with DAPI. Analysis and counting of positively stained cells were performed on at least two sections for each of three different parts of at least two mammary glands per animal.

To quantify the clonal expansion on Notch3-expressing cells, we developed a protocol to remove the stroma surrounding the mammary gland by enzymatic digestion and visualize the remaining intact mammary ducts through compiled confocal z stacks. Freshly dissected mammary glands were incubated rocking in digestion medium containing 1,200 U/ml collagenase and 400 U/ml hyaluronidase for 2 h at 37°C. Tubes were centrifuged at 1,500 rpm for 3 min. The resuspended pellet was fixed for 20 min in 3 ml of 4% PFA in PBS at 4°C, in a 15-ml tube previously coated with FBS for 2 h. The fixed tissue was washed three times in PBS by pulse centrifuging at 1,500 rpm to remove single cells and muscle tissue. The tissue was then imaged on a confocal microscope on a cell culture dish (Fluorodish; World Precision Instruments). At least three mice per time point were analyzed and counted.

BrdU (Sigma-Aldrich) was resuspended in PBS at a concentration of 5 mg/ml. Mice were injected with 50 mg BrdU/kg of body weight. BrdU was added to the drinking water at a concentration of 1 mg/ml supplemented with 2.5% wt/vol sucrose. For the short pulse, 5-wk-old mice were induced with 1 mg/20 g 4-OHT and injected with BrdU at day 0. BrdU was added to the drinking water, and mice were analyzed 3 d later. For the long incorporation, 5-wk-old mice were induced with 1 mg/20 g 4-OHT and injected with BrdU at day 0. BrdU was added to the drinking water and renewed every 2 d for 2 wk, after which mice were chased for 3 wk in normal drinking water. In each case, four mammary glands from two mice were analyzed.

Bigenic Notch3-CreERT2^SAT/R26^mTmG females were induced with 4-OHT at 6 wk of age and sacrificed after 48 h. Single-cell suspensions from mammary glands were prepared essentially as previously described (Shackleton et al., 2006). Mammary glands were dissected from 6-wk-old female mice and mechanically dissociated with a scalpel and razor blade. The tissue was placed in culture medium (DMEM/F12 1:1 with l-glutamine supplemented with 500 ng/ml hydrocortisone, 10 ng/ml EGF, 5 µg/ml insulin, and 20 ng/ml cholera toxin) with 600 U/ml collagenase (Sigma-Aldrich) and 200 U/ml hyaluronidase (Sigma-Aldrich) and incubated for 1 h at 37°C. The resulting suspension was then treated with 0.25% trypsin-EDTA for 2 min and quenched with an equal volume of culture medium containing 10% FBS, 5 mg/ml dispase (Roche), and 0.1 mg/ml DNase (Sigma-Aldrich) for 3 min and then with 0.64% NH₄Cl for 3 min and finally filtrated through a 40-µm mesh. Dissociated cells were then incubated with antibodies in PBS (Ca/Mg²⁺ free), 1% BSA, 5 mM EDTA, 25 mM Hepes, pH 7.0, and 10 U/ml DNase for 20 min at a density of 40 × 10⁶ cells/ml in the dark and on ice. The following antibodies were used: CD45-Allophycocyanin (APC) at 1:60 (clone 30-F11; BioLegend), CD31-APC at 1:60 (clone MEC13.3; BioLegend), Ter-119-APC at 1:60 (clone TER-119; BioLegend), CD24-phycoerythrin/Cy7 (Cyanin7) at 1:60 (clone M1/69; BioLegend), CD29–Alexa Fluor 700 at 1:60 (clone HMβ1-1; BioLegend), Sca-1–PerCP/Cy 5.5 at 1:60 (clone D7; BioLegend), and CD49b-phycoerythrin at 1:60 (clone HM-α; BioLegend). Debris and doublets were excluded by sequential gating on forward scatter area versus side scatter area followed by side scatter width versus side scatter area. DAPI and CD45, CD31, and ER-119 (Lin) staining was used to exclude nonviable cells, immune cells, endothelial cells, and erythrocytes, respectively. Cell analysis was performed on a FACS flow cytometer (LSR II; BD), and cell sorting was performed on a FACSAria (BD). Data were analyzed with the FlowJo software (TreeStar, Inc.).

FACS-sorted single mammary epithelial cells were seeded on a feeder layer of irradiated NIH-3T3 fibroblasts. Fibroblasts were grown in DMEM medium (Invitrogen) supplemented with 10% FBS (Invitrogen) and 1% vol/vol penicillin/streptomycin (Invitrogen). The day before the experiment, fibroblasts were irradiated at 20 Gray and, after 3 h in culture, were seeded in 24-well plates at a density of 5,000 cells per well. Isolated epithelial cells were seeded on the feeder layer at a density of 300 cells/well and grown in DMEM/F12 1:1 (Invitrogen) supplemented with 10% FBS, 1% vol/vol penicillin/streptomycin (Invitrogen), 5 µg/ml insulin (Sigma-Aldrich), 10 ng/ml EGF (Sigma-Aldrich), and 10 ng/ml cholera toxin (Sigma-Aldrich). The growth of colonies was monitored daily, and the experiment was stopped after 6 d in culture. Colonies were stained with hematoxylin and counted. The experiment was repeated three times counting ≥12 wells per condition.

DNA was extracted using the DNeasy Blood & Tissue kit (QIAGEN) according to the manufacturer’s instructions. RNA was extracted from FACS-sorted cells using the RNeasy Mini kit (QIAGEN).

Reverse transcription (Superscript III; Invitrogen) was used for cDNA synthesis. The primers used for RT-PCR analyses were Arf forward, 5′-CGCAGGTTCTTGGTCACTGTGAGG-3′, and reverse, 5′-TGCCCATCATCATCACCTGGTCC-3′; p21 forward, 5′-TATTTAAGCCCCTCCCAACC-3′, and reverse, 5′-AGCTGGCCTTAGAGGTGACA-3′; N3EC forward, 5′-TTGTCTGGATGGAAGCCCATG-3′, and reverse, 5′-ACTGAACTCTGGCAAACGCCT-3′; and N3IC forward, 5′-TGAACAACGTGGAGGCTACC-3′, and reverse, 5′-GCAGCCTGTCCAAGTGATCT-3′. qRT-PCR was performed in ViiA 7 by Life Technologies obtained from Applied Biosystems, using SYBR Green I Master kit (Roche).

Stained sections were mounted in imaging medium (Aqua-Poly/Mount; Polysciences, Inc.) and imaged on an inverted confocal spinning disk (CSU-X1 [Yokogawa Corporation of America and Roper Scientific] mounted on an inverted microscope [Eclipse-Ti; Nikon]) equipped with a camera (CoolSNAP HQ2; Photometrics) and a 405-, a 491-, a 561-, and a 633-nm laser. Images were taken with objectives ranging from 20 to 100×, with a numerical aperture between 0.3 and 1.4, depending on the objective used. Raw images were acquired with the MetaMorph software (Molecular Devices) and analyzed using ImageJ (National Institutes of Health).

Fig. S1 shows the characterization of Notch3-expressing cells. Fig. S2 shows a schematic representation of the timeline of 4-OHT induction and analyses for lineage-tracing experiments, presented in Fig. 2. Fig. S3 shows ER-α/PR expression in GFP⁺ cells 24 h after induction. Table S1 gives raw data of the number and percentage of clones counted for each experimental condition in triplicates, for the analysis of clonal expansion presented in Fig. 2 (wild-type mice) and in Fig. 5 (Notch3 gain-of-function mice).

We are extremely grateful to Prof. Spyros Artavanis-Tsakonas for generously sharing the Notch3-CreERT2^SAT mice with us, for his support, and his precious and critical scientific advice. The authors would like to acknowledge the Cell and Tissue Imaging Platform of the Genetics and Developmental Biology Department (UMR3215/U934) of Institut Curie for help with light microscopy. We are particularly grateful to S. Boissel, C. Daviaud, and M. Garcia for precious help in the maintenance and care of the transgenic animals and to the Flow Cytometry Platform of Institut Curie. We also wish to thank S. Sale, A.S. Kaanta, A. Louvi, and the lab of M. Glukhova for constructive discussions, scientific insights, and technical advice.

This research was in part supported by the Association pour la Recherche contre le Cancer to S. Fre and by Labex DEEP (Développement, Epigénèse, Epigénétique et Potentiel) 11-LBX-0044.