Paracrine Met signaling triggers epithelial–mesenchymal transition in mammary luminal progenitors, affecting their fate

HGF/Met signaling has recently been associated with basal-type breast cancers, which are thought to originate from progenitor cells residing in the luminal compartment of the mammary epithelium. We found that ICAM-1 efficiently marks mammary luminal progenitors comprising hormone receptor-positive and receptor-negative cells, presumably ductal and alveolar progenitors. Both cell populations strongly express Met, while HGF is produced by stromal and basal myoepithelial cells. We show that persistent HGF treatment stimulates the clonogenic activity of ICAM1-positive luminal progenitors, controlling their survival and proliferation, and leads to the expression of basal cell characteristics, including stem cell potential. This is accompanied by the induction of Snai1 and Snai2, two major transcription factors triggering epithelial–mesenchymal transition, the repression of the luminal-regulatory genes Elf5 and Hey1, and claudin down-regulation. Our data strongly indicate that paracrine Met signaling can control the function of luminal progenitors and modulate their fate during mammary development and tumorigenesis. DOI: http://dx.doi.org/10.7554/eLife.06104.001


Introduction
The postnatal development of the mammary gland comprises two distinct morphogenetic events: the growth and branching of epithelial ducts during puberty and the lobulo-alveolar development during pregnancy. The mammary epithelium is embedded in a fatty connective tissue and organized as a bilayer, with a basal layer of myoepithelial cells and a luminal epithelial layer. During lactation, the luminal cells produce milk, whereas myoepithelial cells serve for milk expulsion.
The adult mammary basal compartment harbors multipotent stem cells able to fully regenerate the gland upon transplantation Stingl et al., 2006). The luminal cell layer contains clonogenic cells, the luminal progenitors (Asselin-Labat et al., 2007;Sleeman et al., 2007). This population is heterogeneous, consisting of hormone receptor-positive and hormone receptornegative clonogenic cells that are considered to be functionally distinct ductal and alveolar progenitors (Beleut et al., 2010;Regan et al., 2012;Shehata et al., 2012;Fu et al., 2014). Considerable interest has recently focused on luminal progenitors. According to lineage-tracing studies, these cells are able to drive the expansion of the luminal cell population in mammary ducts and alveoli during postnatal development (Van Keymeulen et al., 2011;Fu et al., 2014;Rios et al., 2014). Moreover, luminal progenitors are thought to be at the origin of the triple-negative, basal-like, Brca1-associated breast cancers, indicating that they display phenotypic plasticity and an ability to upregulate basal markers (Lim et al., 2009;Molyneux et al., 2010;Proia et al., 2011). Of note, ectopic expression of Snail2 endows luminal progenitors with basal cell features (Guo et al., 2012), supporting an important role of EMT in the regulation of luminal progenitor plasticity.
The Met tyrosine kinase receptor and its major ligand, HGF, are known to control numerous epithelial cell functions and trigger EMT Gherardi et al., 2012). Met signaling has been implicated in mammary development and tumorigenesis. Although not investigated in detail yet, Met-deficient mammary glands of adult MMTV-Cre:Met f/f mice showed defects in branching morphogenesis (Garner et al., 2011). Aberrant Met activation occurs in breast cancers, in particular in the triple-negative, basal-like subtype (Gastaldi et al., 2010;Ponzo and Park, 2010). Expression of activated Met or over-expression of HGF in mouse mammary gland induce tumors, including those of the basal-like subtype (Graveel et al., 2009;Ponzo et al., 2009;Holland et al., 2013;Knight et al., 2013). Several in vitro studies performed with mammary organoids or immortalized cell lines have implicated HGF/Met signaling in the regulation of mammary epithelial cell growth, morphogenesis, and differentiation (Soriano et al., 1995;Yang et al., 1995;Haslam et al., 2008;Lee et al., 2011). eLife digest Throughout the life of a female mammal, the mammary glands undergo different phases of development to prepare for, and adapt to, feeding offspring. Luminal cells line the inside of branch-like structures throughout the mammary gland and are responsible for producing milk. When the mammary gland grows, new luminal cells develop from a kind of cell called luminal progenitor cells. However, these progenitor cells are also thought to be the source of certain types of breast cancer.
Recently, it has been suggested that luminal progenitor cells display a receptor protein called Met on their surface. When Met and 'co-receptor' proteins bind to a molecule called HGF, this triggers a cascade of signals that can cause certain cells to change their properties. This is known as the epithelial-mesenchymal transition. Although this transition is important for new tissues to develop, it can also result in cancerous tumors forming if it is not correctly controlled. Luminal cells do not produce HGF themselves, which suggests that Met signaling in these cells is triggered by the HGF released from neighboring cells. However, neither the mechanisms behind this signaling nor the effects of signaling on the luminal progenitor cells are well understood.
Di-Cicco et al. set out to identify where Met, its co-receptors and HGF are located in the mouse mammary gland during different phases of development. This revealed that one of the coreceptors-called ICAM-1-can be used as a marker to identify certain types of luminal progenitor cell. Di-Cicco et al. found that these progenitor cells display Met on their surface, and other types of mammary cell-called stromal cells and myoepithelial cells-produce HGF.
When exposed to HGF, luminal progenitor cells grown in culture in the laboratory proliferated and went through the epithelial-mesenchymal transition. These findings suggest that myoepithelial and stromal cells regulate luminal progenitor cells by producing HGF to activate Met signaling in these cells. Such interactions could be of great importance during mammary development and tumorigenesis. The next big challenge will be to determine the circumstances under which luminal progenitor cells stimulated by HGF can give rise to breast cancers. This work will allow us to better define the cell population that should be targeted by anti-cancer drugs.
One recent work indicated that luminal progenitors express Met and that, when forced to express HGF, display an enhanced in vivo regenerative potential, a property attributed to basal-type stem cells (Gastaldi et al., 2013). However, in situ, mammary luminal cells do not express HGF, suggesting that Met signaling in mammary epithelial cells is activated in a paracrine manner. HGF-mediated paracrine and autocrine Met stimulation can trigger distinct biological responses in epithelial cells (Mai et al., 2014). How Met-expressing luminal progenitors respond to the physiological paracrine action of HGF has not yet been investigated at the cellular and molecular levels.
Met activation requires adhesion molecules as co-receptors. Two co-receptors for Met have been identified in hepatocytes: CD44v6, the CD44 isoform containing the exon v6 and ICAM-1 (CD54), a member of the immunoglobulin superfamily of cell adhesion molecules (Orian-Rousseau et al., 2002;Olaku et al., 2011). In this study, we investigated the expression of ICAM-1, CD44v6, HGF, and Met in the mouse mammary epithelium. We found that ICAM-1 is a useful marker for obtaining highly enriched preparations of clonogenic luminal progenitors that express Met, and that HGF is produced by both stromal and myoepithelial cells. Using ICAM-1 as a surface marker, we further characterized Met-expressing luminal progenitors and analyzed their cellular and molecular responses to HGF. Our data suggest that HGF can act as a paracrine regulator of luminal progenitor function and fate during mammary development and tumorigenesis.

ICAM-1 expression discriminates the basal/myoepithelial cell compartment and two distinct populations of luminal cells
We examined ICAM-1 expression at representative stages of mammary gland development, including puberty, maturity, early, and late gestation. Freshly isolated mammary cells, double-stained for CD24 and ICAM-1, were analyzed by flow cytometry ( Figure 1A, Figure 1-figure supplement 1A). From puberty to late gestation, ICAM-1 was differentially expressed in the mammary epithelium. Staining for this marker identified three epithelial populations, negative (ICAM1-neg) or displaying low (ICAM1-low) or high (ICAM1-hi) levels of expression ( Figure 1A, Figure 1-figure supplement 1B). Gene expression analysis by qPCR showed that, at all stages of development, K5-expressing basal myoepithelial cells were located in the ICAM1-hi population, while K8-positive luminal cells were distributed in ICAM1-neg and ICAM1-low-cell subsets ( Figure 1B). Thus, ICAM-1 can be used to identify the basal cell compartment, and separate two populations of luminal cells, referred to hereafter as Lu-pos and Lu-neg cells.
Throughout development, the basal myoepithelial cell population constitutively expressed ICAM-1 at high level ( Figure 1A). By contrast, the luminal cells displayed temporal changes in ICAM-1 expression ( Figure 1A, Figure 1-figure supplement 1C). In virgin mice, most of the luminal cells were negative for ICAM-1 at puberty, whereas 25-30% of the luminal cells expressed ICAM-1 in the mature gland of 11 to 14-week-old cycling females. Notably, the Lu-pos cell subset was amplified early in gestation (P-8d), during the expansion of the alveolar buds. Later, at P-16d, when the alveoli were well developed, almost the entire luminal compartment was positive for ICAM-1. Thus, in the luminal cell population, ICAM-1 expression is developmentally regulated and correlates temporally with changes in hormone signaling.

ICAM-1 expression delineates a luminal cell population highly enriched in clonogenic progenitors
We investigated the functional properties of the luminal cell populations separated on the basis of ICAM-1 expression, by plating purified Lu-pos and Lu-neg cells at low density and assessing their ability to form colonies in two-dimensional cultures. In adult virgin mice, almost all the colony-forming activity was associated with the Lu-pos cell subset, demonstrating that ICAM-1 expression delineates a luminal population highly enriched in clonogenic progenitors ( Figure 1C). Notably, ICAM-1 efficiently marked mammary luminal progenitors from adult virgin Balb/c ( Figure 1C As expected from previous findings (Selbert et al., 1998;Molyneux et al., 2010), LacZ activity, as assessed by X-gal staining, was undetectable in basal myoepithelial cells and present in a subset of luminal cells in the mammary gland of adult virgin Blg-Cre; R26 females ( Figure 1-figure supplement 2C). An analysis of Cre expression by qPCR indicated that Blg (β-Lactoglobulin) drove recombination  We next compared the molecular characteristics of the luminal cell populations defined by ICAM-1 by analyzing the expression of a panel of genes by qPCR ( Figure 1D, Figure 1-figure supplement 2F). The non-clonogenic Lu-neg population had high level of transcripts for the hormone receptors, ERα, PR, and PrlR, and for genes encoding secreted hormonal mediators implicated in the control of mammary development such as amphiregulin (encoded by Areg), RankL (encoded by Tnfsf11) and Wnt4. These genes, along with Ly6a (encoding Sca-1), are characteristic of mature mammary luminal cells (Mulac-Jericevic et al., 2003;Kendrick et al., 2008;Cai et al., 2014). The clonogenic Lu-pos population expressed Mki67, a marker of cell proliferation, more strongly than the Lu-neg fraction. It exhibited high levels of expression for genes encoding essential regulators of mammary development including Elf5, the Notch effector, Hey1, RANK (encoded by Tnfrsf11a), and the local mediator R-spondin1 (encoded by Rspo1) (Fata et al., 2000;Bouras et al., 2008;Oakes et al., 2008;Cai et al., 2014). Lu-pos cells also over-expressed Kit and Itga2, two markers previously used to obtain mammary cell populations enriched in luminal progenitors (Regan et al., 2012;Shehata et al., 2012).
The Lu-neg and Lu-pos cell populations isolated from the mammary glands of females in early pregnancy had functional properties and gene expression patterns similar to those purified from adult virgin glands ( Figure 1D, Figure 2-figure supplement 1A). ICAM-1 thus appeared to be a robust surface marker for the enrichment in mammary luminal progenitors from adult virgin and early pregnant females.

The luminal progenitors identified by ICAM-1 express Met whereas myoepithelial and stromal cells produce HGF
We then used q-PCR to examine the expression of Met and its major cytoplasmic effector, Gab1, in the mammary cell populations separated by ICAM-1. In adult virgin mice, Met transcript levels were five and ten times higher in the Lu-pos population than in the Lu-neg and myoepithelial cells, respectively; Gab1, like Met, was strongly expressed only in the Lu-pos fraction ( Figure 2A). Similarly, the clonogenic Lu-pos population isolated from the mammary glands of early pregnant females was greatly enriched in Met-expressing cells (Figure 3-figure supplement 1B). Thus, in adult virgin females and during early pregnancy, the target cells for Met signaling mostly belonged to the luminal progenitor-enriched population defined by ICAM-1 expression.
As expected (Niranjan et al., 1995;Yang et al., 1995), HGF, the major Met ligand, was strongly expressed in the CD24-negative Pdgfr-expressing stromal cell population ( Figure 2B). In addition, HGF was expressed at high level in the mammary basal cell population of mature virgin and early pregnant mice (Figure 2A As both ICAM-1 and CD44v6 may serve as co-receptors for Met (Olaku et al., 2011), we analyzed the expression of CD44v6 in freshly isolated Lu-pos and Lu-neg cells. Most cells of both populations were stained with the anti-CD44v6 antibody ( Figure 2-figure supplement 1C). Thus, CD44v6 is broadly expressed in the luminal compartment and cannot, therefore, be used for the enrichment in Met-expressing progenitors.

HGF stimulates the clonogenic activity of luminal progenitors and promotes acquisition of basal-specific traits
We used ICAM-1 to isolate Met-expressing luminal progenitors and investigated their response to the paracrine action of HGF. Purified Lu-pos cells were cultured in suspension in the presence of 2% Matrigel as previously described (Spike et al., 2012;Chiche et al., 2013). We found that HGF-treated Lu-pos cells formed enlarged spheres after 10 days in culture ( Figure 2C). These spheres had multiple lumens, with an overall lumen space larger than that of the untreated spheres ( Figure 2C).
Quantitative analysis revealed that HGF-treated cultures of purified Lu-pos cells contained twice as many spheres as non-stimulated cultures ( Figure 2D), suggesting a role for HGF in favoring cell survival and/or proliferation. Measurements of caspase activity showed that HGF-treated cells had lower levels of activated caspase-3 and caspase-7 than untreated cells after 2 days in culture ( Figure 2E). BrdU-incorporation assays performed after 1, 2, and 4 days in culture indicated that HGF treatment increased cell proliferation, particularly on day 2 ( Figure 2F). Thus, in 3D cultures, HGF stimulated the clonogenic activity of ICAM1-expressing luminal progenitors, by promoting cell survival and proliferation early in culture.
We then investigated the long-term effects of HGF on purified luminal progenitors. Using q-PCR, we analyzed proliferation and lineage-specific gene expression in Lu-pos cells cultured in the presence or absence of HGF for 9-10 days ( Figure 3A). Mki67 was down-regulated in HGF-stimulated cultures whereas Cdkn1a, which encodes p21, a negative regulator of the cell cycle (Abbas and Dutta, 2009), was upregulated by a factor of 2.5. The basal-specific keratin, Krt5, undetectable in untreated spheres, was clearly expressed in HGF-stimulated spheres, whereas no significant decrease was observed in expression of the luminal-specific keratin, Krt18. Thus, in the long-term, HGF/Met signaling biased luminal progenitors toward a basal cell fate while attenuating cell cycle progression.
Consistent with the q-PCR data, immunofluorescence labeling showed that the percentages of K5-positive cells in spheres stimulated with HGF for 9-10 days were six times higher that that of untreated cultures ( Figure 3B). Numerous K5-expressing cells stained positive for K8 ( Figure 3B), and none contained the myoepithelial-specific protein α-SMA (Figure 3-figure supplement 1). Interestingly, BrdU incorporation assays showed that, unlike K5-negative luminal cells, K5-expressing cells did not cycle following HGF stimulation ( Figure 3C).
Thus, paracrine Met activation controls the survival and proliferation of ICAM1-expressing luminal progenitors and favors the acquisition of basal-specific keratin.

Luminal progenitors activate EMT program and repress luminal-specific regulatory genes upon stimulation by HGF
To prove that K5-expressing cells in HGF-treated spheres originated from luminal progenitors, we isolated Lu-pos cells from Blg-Cre; R26 mouse mammary glands and stimulated them with HGF for 9-10 days. As in Balb/c mice, Met was strongly expressed in the Blg-Cre; R26 luminal progenitor population purified with ICAM-1 ( We investigated the molecular events induced by the paracrine action of HGF in luminal progenitors, by analyzing the expression of basal-specific and luminal-specific regulatory genes in untreated and HGF-treated spheres derived from purified Lu-pos cells ( Figure 4B). Upon stimulation with HGF, the expression of essential regulators of the luminal cell fate such as Elf5, Hey1, and Gata3 was repressed, whereas that of several basal-specific genes increased markedly. These genes included Cdh3 (encoding P-cadherin), Trp63, and the EMT inducer Snai2. Other genes associated with EMT program-Snai1, Sox9, and Cdh2 (encoding N-cadherin)-were also upregulated. In agreement with gene expression data, immunodetection studies revealed that HGF-treated spheres contained  Snai2 and Snai1 can directly repress the transcription of Cdh1 (encoding E-cadherin) and claudins (reviewed in Thiery et al., 2009;Gonzalez and Medici, 2014). Noticeably, Cdh1 expression was not significantly reduced upon HGF treatment, whereas claudin-1 and claudin-3 transcript levels decreased strongly ( Figure 4B). Consistent with the q-PCR data, immunofluorescence staining revealed E-cadherin in the vast majority of cell-cell contacts in untreated and HGF-treated spheres ( Figure 5A). Notably, numerous Snail2-positive cells present in the HGF-stimulated spheres displayed E-cadherin at their surface ( Figure 5B). Claudin-1 distribution was more heterogeneous than that of E-cadherin, but it was clearly altered following HGF treatment. Untreated spheres contained a majority of cells with claudin-1 at their junctions, whereas HGF-treated spheres displayed large cell areas lacking claudin-1 expression ( Figure 5C).
Thus, the persistent stimulation of Met signaling by exogenous HGF in luminal progenitors repressed luminal-regulatory gene expression, upregulated basal-specific markers, and triggered an EMT program, including Snai2 and Snai1 expression and decrease in claudin levels.

Persistent stimulation by HGF is required to sustain effects on luminal progenitors
To examine whether the HGF/Met signaling effects on luminal progenitors persist after HGF withdrawal, primary HGF-treated spheres obtained from purified Lu-pos cells were dissociated and then cultured either with or without HGF for 10 additional days. Both clonogenic activity and level of expression of the basal-specific keratin Krt5 were much lower in secondary mammospheres deprived of HGF ( Figure 6A,B). Consistently, immunofluorescence stainings revealed that, like untreated control cultures, secondary mammospheres deprived of HGF contained less than 1% K5-positive cells ( Figure 6C,D). HGF-treated secondary mammospheres contained twice more K5-positive cells than HGF-treated primary spheres. Many K5-positive cells co-expressed K8, as in primary spheres,  Figure 6C). Experiments performed with mammary luminal progenitors purified from BlgCre; R26 mice confirmed that, in HGF-treated secondary mammospheres, K5-expressing cells were derived from LacZ-positive luminal cells ( Figure 6E). Gene expression analysis showed that Trp63, Snai2, and Cdkn1a were upregulated whereas Hey1, Elf5, and Mki67 were downregulated only in HGF-treated secondary mammospheres ( Figure 6F).
Thus, the continuous activation of HGF/Met signaling is required to sustain the response of luminal progenitors in vitro and leads to the accumulation of cells with basal characteristics.

HGF stimulation confers regenerative potential on luminal progenitors
To compare the stem cell activity of untreated and HGF-treated luminal progenitors, we transplanted either single-cell suspensions or intact spheres into cleared mammary fat pads of recipient mice (Figure 7). Untreated cells isolated from primary spheres failed to repopulate the fat pad or gave rise to very limited outgrowths. In contrast, HGF-treated cultures formed ductal-type outgrowths in seven of the eight transplanted fat pads ( Figure 7A, Supplementary file 1). Serial passages of luminal progenitor-derived mammospheres stimulated by HGF led to an important enrichment in K5-positive cells ( Figure 6C); we therefore compared the regenerative potential of secondary spheres from untreated and HGF-stimulated cultures. Strikingly, these transplantation assays showed that only HGF-treated spheres were able to form outgrowths ( Figure 7B). 6-week-old outgrowths were composed of well-organized ducts and growing buds comprising K8-positive luminal cells and basally located α-SMA-expressing myoepithelial cells ( Figure 7C,D). These data demonstrate that paracrine Met activation induced multipotent stem cell properties in luminal progenitors.

Met-expressing luminal progenitors comprise hormone receptorpositive and hormone receptor-negative cells
Differential Sca-1 expression in the luminal compartment has been reported to delineate populations enriched in hormone receptor-positive and hormone receptor-negative cells, both containing clonogenic progenitors (Sleeman et al., 2007;Regan et al., 2012). We characterized Met-expressing luminal progenitors further, by analyzing ICAM-1 and Sca-1 distribution in the luminal compartment at different developmental stages by flow cytometry ( Figure 8A, Figure 8-figure supplement 1A).
Only Lu4, the major clonogenic subset devoid of HR-positive cells, expressed the milk protein β-casein ( Figure 8C). It also strongly expressed Elf5 and Tnfrsf11a, two crucial regulators of alveologenesis (Fata et al., 2000;Oakes et al., 2008). Interestingly, flow cytometry analysis on mammary epithelial cells isolated at different stages of development showed that Lu4 was absent at puberty; this subset appeared in the glands of sexually mature cycling females and displayed massive expansion early in pregnancy ( Figure 8A). By contrast, Lu2 and Lu3 were detected at all stages of development. As in mature virgin glands, Lu2 and Lu4 were highly clonogenic during early pregnancy, whereas Lu3 was not (Figure 8-figure supplement 1C). The gene expression patterns of Lu1, Lu2, Lu3, and Lu4 cells isolated from mature virgin and early pregnant mice were similar (Figure 8-figure supplement 1D,E).
The q-PCR analysis showed that Met was strongly expressed in the clonogenic populations, Lu2 and Lu4 ( Figure 8D). These populations responded similarly to HGF treatment in 3D cultures. On stimulation, both produced larger spheres, displaying an upregulation of basal-specific genes (Snai2, Trp63, and Cdh3) and a down-regulation of luminal-specific regulatory genes (Gata3, Elf5, and Hey1) ( Figure 8E,F).
These data show that luminal progenitors defined by ICAM-1 expression include hormone receptor-positive and hormone receptor-negative cells, all of which are potential targets of HGF/Met signaling.

ICAM-1 can serve to localize luminal progenitors in the mammary epithelium
To complement the flow cytometry data, we analyzed ICAM-1 localization on histological sections of mammary gland. Immunohistochemical studies confirmed that in the pubertal gland, basal myoepithelial rather than luminal cells expressed ICAM-1 while at late pregnancy, the whole epithelium was positive ( Figure 9A). At the onset of lactation, ICAM-1 was down-regulated both in ducts and alveoli ( Figure 9A).  At early pregnancy, myoepithelial cells lining large ducts strongly expressed ICAM-1, whereas luminal cells were devoid of ICAM-1 ( Figure 9C). In contrast, in the developing alveoli and small ducts, the luminal layer, presumably enriched in alveolar progenitors, contained clusters of cells positive for ICAM-1 ( Figure 9D). Consistently with qPCR data, ER-positive cells were mostly ICAM1-negative ( Figure 9E). Notably, in adult virgin and early pregnant mice, ICAM-1 expression was restricted to cell-cell contacts, basal-to-basal, basal-to-luminal, and luminal-to-luminal ( Figure 9B-E).

Loss of ICAM-1 does not impede the response of luminal progenitors to HGF
To analyze a possible contribution of ICAM-1 in mammary development and HGF/Met signaling, we investigated the mammary phenotype of Icam1-deficient (Icam1-KO) mice. The Icam1-KO females are viable, fertile, and able to feed their pups (Xu et al., 1994). Whole-mount analysis of mammary glands from wild-type and Icam1-KO adult virgin and pregnant mice indicated that lack of ICAM-1 did not induce visible defects in mammary morphogenesis (Figure 9-figure supplement 1A). Flow cytometry analysis showed that the percentages of basal and luminal cells were not significantly affected in the Icam1-KO epithelium of adult virgin mice (Figure 9-figure supplement 1B).
For HGF stimulation assays, we isolated Icam1-KO mammary epithelial cells and used the Sca-1negative luminal cell population enriched in clonogenic progenitors (Figure 9-figure supplement 1C). We found that Icam1-KO Sca-1-negative luminal cells responded to HGF by increased clonogenicity and production of large spheres (Figure 9-figure supplement 1D,E). In addition, as previously observed in the presence of ICAM-1, HGF treatment led to the induction of the basal-specific genes Krt5, Trp63, and Snai2, with a concomitant decrease in the regulators of the luminal lineage, Elf5, Gata3, and Hey1 ( Figure 9-figure supplement 1F,G). Thus, ICAM-1 is not mandatory for mammary development and HGF/Met signaling in luminal progenitors.

Discussion
We report here that ICAM-1 is a new surface marker for the enrichment of mouse mammary luminal progenitors. Luminal progenitor population identified by ICAM-1 labeling is heterogeneous, comprising HR-positive and HR-negative cells, potentially ductal and alveolar progenitors, both strongly expressing Met. We show that paracrine Met activation stimulates the clonogenic activity of ICAM1-expressing luminal progenitors, controlling their survival and proliferation, and promotes a luminal-to-basal switch while triggering EMT program ( Figure 10). EMT was accompanied by acquisition of regenerative potential, a property restricted to multipotent stem cells. Our data suggest that paracrine HGF signaling can affect the fate of luminal progenitors during mammary development and tumorigenesis.

ICAM-1 is a new marker for the enrichment of clonogenic luminal progenitors
Our study provides the first evidence showing that ICAM-1 is differentially distributed, both spatially and temporally, in the mouse mammary epithelium. In mature virgin mice and early in pregnancy, the luminal cell population comprises ICAM1-negative and ICAM1-positive cell fractions, highly enriched in hormone-sensing cells and clonogenic progenitors, respectively. The luminal progenitors identified by ICAM-1 staining strongly express Elf5, Hey1, Tnfrsf11a, and Rspo1 genes encoding critical regulators of mammary development (Fata et al., 2000;Bouras et al., 2008;Lee et al., 2011;Cai et al., 2014).
Several recent studies have indicated that the luminal progenitor population includes cells with different phenotypes (Oliver et al., 2012;Regan et al., 2012;Shehata et al., 2012;Lafkas et al., 2013;Sale et al., 2013;Rodilla et al., 2015). The differential expression of Sca-1 and CD61 was first used to obtain partial enrichment in luminal progenitor cells (Asselin-Labat et al., 2007;Sleeman et al., 2007;Rios et al., 2014). Subsequently, the combined use of Sca-1 and c-kit or CD49b revealed that the luminal progenitors in the mature virgin gland belong to two distinct populations, one consisting essentially of HR-positive cells and the other mostly of HR-negative cells (Regan et al., 2012;Shehata et al., 2012). We found that in virgin glands and during early pregnancy, combined ICAM-1 and Sca-1 expression discriminated these two major luminal clonogenic cell subsets. ICAM-1 clearly appeared more efficient than Sca-1 for separating non-clonogenic and clonogenic luminal populations. Importantly, ICAM-1 marked the luminal progenitor population in different mouse strains, including C57Bl6/J, and in mixed backgrounds, such as that of BlgCre; R26 mice, whereas neither CD61 nor c-kit can be used for luminal progenitor enrichment from C57Bl6/J mouse mammary glands (Visvader and Stingl, 2014). Furthermore, differential ICAM-1 expression within the Sca-1negative luminal cell population separated a poorly clonogenic ICAM1-negative subset enriched in HR-positive cells (Lu3) from a major clonogenic cell fraction positive for ICAM-1 and lacking HR expression (Lu4). Such separation was not possible with CD49b, the most recently described marker for characterizing luminal cell subsets (Shehata et al., 2012;De Silva et al., 2015).
The clonogenic luminal cell population is thought to comprise distinct ductal-and alveolarrestricted progenitors, as yet not fully characterized (Regan et al., 2012;Shehata et al., 2012). Cells committed to the alveolar fate are expected to expand in adult females particularly at the onset of pregnancy, to express Tnfrsf11a and Elf5 strongly, and most probably, to lack ER and PR (Fata et al., 2000;Lee et al., 2011;Fu et al., 2014;Rodilla et al., 2015). Interestingly, the Sca-1-negative ICAM1-positive clonogenic cell population (Lu4) absent from prepubertal mammary epithelium fulfills these criteria. In addition, these cells have high levels of transcript for the milk protein β-casein. By contrast, the Sca-1-and ICAM1-positive clonogenic subset containing HR-expressing cells (Lu2) present throughout mammary development may preferentially contain ductal-restricted progenitors.
Thus, ICAM-1 is a new robust marker for separating basal and luminal cells and analyzing the heterogeneity of the luminal cell compartment in the adult mammary gland using flow cytometry. In addition, ICAM-1 can be used to localize and further characterize luminal progenitors in situ.

HR-negative and HR-positive luminal progenitors are potential targets of HGF/Met signaling
The target cells for Met signaling in the mammary epithelium remain poorly characterized. A recent work reported the expression of Met predominantly in the Sca-1-negative luminal progenitorenriched population (Gastaldi et al., 2013). In agreement with these data, we found high-Met transcript levels in the Sca-1-negative ICAM1-positive clonogenic cell population (Lu4). Additionally, we observed strong Met expression in the Sca-1-positive ICAM1-positive luminal progenitors (Lu2). Thus, the two major clonogenic populations of luminal progenitors, enriched in HR-negative and HRpositive cells, are potential targets of HGF/Met signaling. Met was also expressed in the Sca-1negative ICAM1-negative cell subset (Lu3), which was weakly clonogenic and enriched in HR-positive cells. Our data thus highlight the phenotypic diversity of Met-expressing cells in the luminal population. Consistently, a knockin mouse model with a mutationally activated Met has been reported to develop various types of mammary tumors, including ER-positive and ER-negative tumors (Graveel et al., 2009).
We found that ICAM1-expressing luminal progenitors responded to HGF stimulation by basal marker upregulation. Importantly, using the Blg-Cre; R26 reporter mouse, we demonstrated that these basal marker-expressing cells were of luminal origin. Triple-negative basal-like breast cancers are thought to originate from luminal progenitors aberrantly expressing basal-specific markers (Lim et al., 2009;Molyneux et al., 2010;Proia et al., 2011). The highest levels of Met have been associated with the basal subtype of breast cancers (Garcia et al., 2007;Graveel et al., 2009;Ponzo et al., 2009;Gastaldi et al., 2010;Knight et al., 2013). Moreover, amplifications of the Met locus and Met overexpression occur in Brca1/p53-deficient mouse tumors, a model of basal-like breast cancers (Smolen et al., 2006;Gastaldi et al., 2013). Thus, consistent with these reports, our results strongly indicate that Met signaling may contribute to the phenotypic characteristics of certain basal-like breast carcinomas.

ICAM-1 is not mandatory for HGF/Met signaling in luminal progenitors
We found that in the mouse mammary epithelium, Met-expressing luminal progenitors displayed both ICAM-1 and CD44v6 at their surface. In mouse hepatocytes, CD44v6 acts as main co-receptor for Met, however, ICAM-1 can trigger Met activation in the absence of CD44 (Orian-Rousseau et al., 2002;Olaku et al., 2011). Our data showed that loss of ICAM-1 did not severely affect mammary development, HGF/Met signaling, and basal cell differentiation process in luminal progenitors. This suggests that CD44v6 rather than ICAM-1 serve as main co-receptor for Met in the mouse mammary epithelium and that identical compensation mechanisms may be involved in Met activation in liver and mammary gland. Distribution of Met and its co-receptors in the human mammary gland remains poorly defined; however, strong expression of Met, ICAM-1, and CD44v6 has been associated with triple-negative breast cancers and correlated with poor prognosis (Charpin et al., 2009;Gastaldi et al., 2010;Ponzo and Park, 2010;Schroder et al., 2011). Moreover, ICAM-1 has recently been identified as a molecular target for triple-negative breast cancers (Guo et al., 2014).

Myoepithelial and stromal cells may both control luminal progenitor function in a paracrine manner
Stromal cells are usually considered to be the only source of HGF in the mammary gland (Niranjan et al., 1995;Yang et al., 1995;Gastaldi et al., 2013). However, we show here that mammary myoepithelial cells also produce HGF. It is worth mentioning that in the postnatal mammary epithelium, ductal luminal cells are in direct contact with myoepithelial rather than stromal cells. In alveoli, the myoepithelial cell layer is discontinuous, so luminal cells may also come into contact with the basement membrane (Moumen et al., 2011). HGF produced by both myoepithelial and stromal cells may therefore contribute to the control of the luminal progenitor activity, favoring either their normal expansion during development or aberrant amplification in cancers. Paracrine basal-to-luminal signaling mediated by the Nrg1-Erbb4 ligand-receptor pair has recently been implicated in the regulation of luminal cell maturation during lobulo-alveolar development (Forster et al., 2014).
We found that HGF had multiple effects on Met-expressing luminal progenitors. Early in culture, HGF promoted cell survival and proliferation. Consistent with its role in tubulogenesis (Rosario and Birchmeier, 2003), HGF then favored lumen formation. Finally, HGF treatment affected the fate of luminal progenitors, leading to the accumulation of a cell population expressing basal markers, including cells co-expressing basal and luminal keratins. Such dual luminal/basal cells are seldom in the adult mammary epithelium but are abundant in fetal mammary rudiments (Spike et al., 2012). The accumulation of cells co-expressing basal and luminal keratins has been observed in the mammary epithelium of Elf5-null mice, which have high-Snail2 levels and display EMT-like phenotypic changes (Chakrabarti et al., 2012a(Chakrabarti et al., , 2012b.

HGF triggers EMT in luminal progenitors and modulates the expression of antagonistic regulators of luminal and basal fates
Several regulators of the balance between mammary luminal and basal phenotypes have recently been identified. Notch signaling specifies a luminal cell fate, whereas ΔNp63 is required for the maintenance of basal cell characteristics (Bouras et al., 2008;Yalcin-Ozuysal et al., 2010). Forced expression of Snail2 endows mammary luminal progenitor cells with molecular characteristics and functional properties of basal-type stem cells, whereas Elf5, through the direct repression of Snai2, promotes luminal cell differentiation (Guo et al., 2012;Chakrabarti et al., 2012aChakrabarti et al., , 2012b. Consistent with these data, we found that HGF induced the expression of Trp63 whilst repressing that of Hey1, a major Notch effector, in luminal progenitors. Concomitantly, we observed important hallmarks of EMT, including an increase in Snai1 and Snai2 levels, Elf5 repression and claudin down-regulation. The transcription of P-cadherin and N-cadherin was induced upon HGF treatment, but E-cadherin expression was not repressed, suggesting a partial, reversible EMT process (Nieto and Cano, 2012;Gonzalez and Medici, 2014). Consistently, alterations of the luminal progenitor phenotype did not persist in vitro after HGF withdrawal. Altogether, these data suggest that microenvironmental signals, such as HGF, may modulate the phenotype of luminal progenitors, by affecting the expression of antagonistic regulators of the basal and luminal fates.
In addition to their role in triggering EMT, transcription factors of the Snail family have been reported to attenuate cell cycle progression by increasing p21 expression (Vega et al., 2004). At late time points, Cdkn1a, Snai1, and Snai2 were upregulated in HGF-treated cultures of luminal progenitors, whereas Mki67 was down-regulated. These changes were accompanied by the accumulation of K5-expressing quiescent cells. How p21 and its major upstream regulator p53 are involved in the response of luminal progenitors to HGF remains to be unraveled. A recent analysis of transgenic mouse models showed that Met can synergize with p53 loss to promote the formation of triple-negative mammary tumors with a claudin-low, EMT-type molecular signature (Knight et al., 2013). Furthermore, the loss of p53 from the mammary epithelium leads to the expansion of clonogenic mammary luminal progenitors (Chiche et al., 2013).
In brief, our data provide new insights into the cellular and molecular mechanisms underlying the phenotypic plasticity of luminal progenitors. They also suggest that mammary basal and stromal cells may both be involved in controlling luminal progenitor function via the paracrine activation of Met during mammary development and tumorigenesis.

Materials and methods
Mouse strains and transgenic mice BALB/cByJ JAX, C57BL/6, and BALB/c-Nude JAX females were purchased from Charles Rivers (L'arbresle, France). Transgenic mice expressing the Cre recombinase under the control of the β-Lactoglobulin promoter (Blg-Cre) were from the Jackson Laboratories (Sacramento, CA, USA) and purchased from Charles Rivers. Rosa26-LacZ reporter strain was provided by Soriano (1999). Icam1deficient mice (B6.129S4-Icam1 tm1Jcgr /J) were from the Jackson Laboratories. The mature virgin females used in the experiments were 11-25 weeks old. The care and use of animals used here was strictly applying European and National Regulation for the Protection of Vertebrate Animals used for Experimental and other Scientific Purposes in force (facility licence #C75-05-18). It complies also with internationally established principles of replacement, reduction, and refinement in accordance with Guide for the Care and Use of Laboratory animals (NRC, 2011). All experimental procedures were ethically approved (Ethical approval 02265.01/CEEA-IC 118).

Preparation of mammary epithelial cells
Single cells were prepared from 2 to 8 inguinal mammary glands. Glands taken from virgin or pregnant females were minced, first with scissors and then with scalpels. Minced tissues were transferred to a digestion solution containing 3 mg/ml collagenase (Roche Diagnostics, Meylan, France), 100 units/ml hyaluronidase (Sigma-Aldrich, Saint Louis, MO, USA) in CO2-independent medium (Gibco, Life Technologies, St Aubin, France) completed with 5% fetal bovine serum (FBS, Lonza, Basel, Switzerland) and 2 mM L-glutamine (Sigma-Aldrich), and incubated for 90 min at 37˚C with shaking (150 rpm). Digested samples were centrifuged at 450×g for 5 min and the supernatant eliminated. Pellets were washed once with CO2-independent medium, treated with a prewarmed 0.25% trypsin/0.1% versen (Life Technologies; Biochrom, Berlin, Germany) solution in Phosphate Buffered Saline (PBS) for 1 min and rinsed with CO2-independent medium containing 5% FBS. Pellets were then resuspended in a solution of 5 mg/ml dispase II (Roche) in CO2-independent medium containing 5% FBS and DNAseI (Sigma-Aldrich) was added to a final concentration of 0.1 mg/ml. After 5 min incubation at 37˚C, cells were rinsed once in CO2-independent medium containing 5% FBS and the pellets were treated with a cold ammonium chloride solution (Stem Cell Technologies, Grenoble, France). Cell suspensions were centrifuged, resuspended in CO2-independent medium, and filtered through a nylon mesh cell strainer with 40-μm pores (Fisher Scientific, Illkirch, France) before immunolabeling.
Caspase-3/7 activity was assessed in cells harvested 1 and 2 days after plating them in mammosphere medium, using the Caspase-Glo 3/7 Assay System (Promega, Madison, WI, USA) according to the manufacturer's protocol.

Immunofluorescence and immunohistochemical stainings
Freshly isolated cells were cytocentrifuged onto slides and fixed in cold methanol for 10 min. Cells were incubated with primary antibodies at 37˚C for 1 hr, with secondary antibodies at room temperature for 1 hr and mounted in Prolong Gold antifade reagent with DAPI (Invitrogen, Life Technologies). Mammospheres were resuspended in 50 μl Matrigel and incubated at 37˚C for 2 hr prior to embedding either in OCT (Tissue-Tek, Sakura Finetek Europe, Leiden, Netherlands) for cryosections or in paraffin. Cryosections were post-fixed with 4% paraformaldehyde for 10 min and treated with 0.5% triton X-100 for 5 min before immunostaining. Spheres were fixed in Methacarn (1:3:6 mixture of acetic acid:chloroform:methanol) before embedding in paraffin. Sections were dewaxed, processed for acidic antigen retrieval, and immunolabeled, as described elsewhere (Chiche et al., 2013). For cell proliferation assays, mammospheres were incubated with BrdU (5 μg/ml; Sigma-Aldrich) for 1 hr before being fixed and processed.

X-gal staining
Mammospheres established from Blg-Cre; R26 mammary cells were dissociated and isolated cells were pre-fixed with 2.5% paraformaldehyde for 5 min at 4˚C and X-gal stained in suspension at 37˚C overnight. Cells were cytocentrifuged, post-fixed with 4% paraformaldehyde for 10 min at room temperature, and processed either for Fast Red staining or immunostaining.
For whole-mount X-gal staining, mammary glands from Blg-Cre; R26 females were fixed in 2.5% paraformaldehyde for 1 hr at 4˚C and stained overnight at 30˚C. Glands were embedded in paraffin and sections counterstained with Fast Red.

Transplantation assays and whole-mount analysis
Isolated cells derived from primary spheres or intact secondary spheres were transplanted into the inguinal fat pads of 3-week-old BALB/c-Nude JAX females cleared of endogenous epithelium as described elsewhere (Deome et al., 1959;Moumen et al., 2012;Chiche et al., 2013). Untreated and HGF-treated cells or spheres were resuspended in 10 μl of 25% growth factor-reduced Matrigel before being grafted in the contralateral fat pads of host mice. For secondary sphere transplantations, 5000 ICAM1-positive luminal progenitors per well were plated, grown with or without HGF for 14 days, then, dissociated and replated at 5000 cells per well as for primary spheres. After 15 days in culture, the content of each culture well was grafted in a separate mammary fat pad. To analyze the outgrowths, dissected mammary fat pads were spread onto glass slides, fixed in Methacarn, and stained with carmine alum (Stem Cell Technologies), as described elsewhere (Moumen et al., 2012;Chiche et al., 2013).

Reverse transcription-polymerase chain reaction
RNA was reverse-transcribed with MMLV H(−) Point reverse transcriptase (Promega, Madison, WI, USA), and quantitative PCR (qPCR) was performed by monitoring, in real time, the increase in fluorescence of the SYBR Green dye on an LightCycler 480 Real-Time PCR System (Roche Applied Science, Basel, Switzerland). The primers used for qPCR analysis were purchased from SABiosciences/ Qiagen (Hilden, Germany) or designed using Oligo 6.8 software and synthesized by Eurogentec (Seraing, Belgium) (Supplementary file 2).
Statistical analysis of the data p values were determined using Student's test with two-tailed distribution and unequal variance. data, Drafting or revising the article; M-AD, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article Ethics Animal experimentation: The care and use of animals used here was strictly applying European and National Regulation for the Protection of Vertebrate Animals used for Experimental and other Scientific Purposes in force (facility licence #C75-05-18). It complies also with internationally established principles of replacement, reduction and refinement in accordance with Guide for the Care and Use of Laboratory animals (NRC 2011). All experimental procedures were ethically approved (Ethical approval 02265.01/CEEA-IC 118).