Podoplanin regulates mammary stem cell function and tumorigenesis by potentiating Wnt/β-catenin signaling

Although mammary development initiates during embryogenesis the majority occurs postnatally. During puberty, the mammary ducts elongate and ramify extensively, generating a ductal network in sexually mature females. Pregnancy is characterized by ductal side-branching and alveoli formation. Lactational differentiation is followed by involution at weaning. Systemic hormonal cues and various local stimuli, including growth factors, cell-cell and cell-matrix interactions, control the morphogenesis and remodeling of the postnatal mammary gland (Macias and Hinck, 2012; Glukhova and Streuli, 2013).

The mammary epithelium is organized as a bilayer, with an outer layer of basal myoepithelial cells and an inner layer of luminal cells. During lactation, the luminal cells produce milk, whereas the myoepithelial cells are contractile and serve for milk expulsion. Basal myoepithelial cells express basal-specific keratins (including K5/14), P-cadherin, smooth muscle-specific contractile proteins, and the transcription factors ΔNp63 (an isoform of Trp63) and Slug/Snail2, which are essential for the maintenance of basal cell identity (Yalcin-Ozuysal et al., 2010; Guo et al., 2012). The luminal cell layer is characterized by the expression of K8/18. It includes a subset of hormone-sensing cells that express estrogen, progesterone and prolactin receptors (ER, PR and PrlR, respectively) and produce local mediators involved in the paracrine control of basal and luminal cell function (Brisken and Ataca, 2015).

It is established that both mammary lineages, basal and luminal, originate from a common embryonic stem cell (SC) expressing basal keratins (van Keymeulen et al., 2011; Moumen et al., 2012). In the postnatal mammary gland, multipotent SCs able to repopulate the entire epithelium upon transplantation have been localized to the basal compartment (Visvader and Stingl, 2014). Data from lineage-tracing studies have revealed the existence of basal and luminal lineage-restricted SCs (van Keymeulen et al., 2011, 2017; Prater et al., 2014; Rios et al., 2014). The precise molecular characteristics of multipotent and lineage-restricted SCs remain unknown, and their respective contributions to mammary bilayer development and homeostasis after birth are still a matter of debate (Lloyd-Lewis et al., 2017). Nevertheless, in recent years, considerable interest has focused on these cell subsets, from which certain breast cancers are thought to originate, particularly cancers of the triple-negative subtype (TNBC), lacking ER, PR and amplified HER2 (or ERBB2), often associated with a poor prognosis (Visvader and Stingl, 2014; Skibinski and Kuperwasser, 2015).

Many studies have shown that canonical Wnt/β-catenin (β-cat) signaling is essential for normal mammary development (Yu et al., 2016). In addition, this pathway is frequently dysregulated in TNBCs (Pohl et al., 2017). Wnt/β-cat signaling has been shown to play a major role in controlling the expansion of the basal cell population during postnatal mammary development (Teulière et al., 2005; Zeng and Nusse, 2010; Macias et al., 2011; van Amerongen et al., 2012; Cai et al., 2014; Rajaram et al., 2015). Basal cells display a complex Wnt receptor machinery, including Fzd7, Lrp5/6 and the R-spondin (Rspo) receptors Lgr4/5/6, known to modulate Wnt/β-cat signal strength (Badders et al., 2009; de Visser et al., 2012; Wang et al., 2013; Chakrabarti et al., 2014; Blaas et al., 2016; Driehuis and Clevers, 2017). Luminal cells have been identified as a major source of Wnt-associated ligands. In particular, they produce Wnt4 and Rspo1, two major regulators of paracrine Wnt/β-cat activation in basal cells (Cai et al., 2014; Rajaram et al., 2015).

Comparative transcriptome analyses of basal and luminal cells isolated from adult mouse and human mammary glands indicated that podoplanin (Pdpn) was among the top-ranking genes characterizing the basal cell signature (Lim et al., 2010). However, its functional importance remains unknown. Pdpn is a small mucin-type transmembrane protein composed of a glycosylated extracellular domain, a transmembrane region and a short cytoplasmic tail devoid of enzymatic activity (Renart et al., 2015). Widely used as a marker of lymphatic endothelial cells, Pdpn is also displayed by various other cell types, including certain epithelial cells, and is overexpressed in human carcinomas of various tissue origin (Schacht et al., 2005; Wicki and Christofori, 2007; Ugorski et al., 2016; Suzuki-Inoue et al., 2017).

Pdpn null mice die before or shortly after birth, exhibiting defects in lung organogenesis, cardiac function and blood-lymph separation (Ramirez et al., 2003; Schacht et al., 2003; Mahtab et al., 2009). Pdpn is therefore crucial for the early development of several tissues. Most of the data concerning the physiological function of Pdpn come from studies on the immune system, focusing on the ability of the extracellular domain of Pdpn to bind the C-type lectin Clec2 (or Clec1b) (Suzuki-Inoue et al., 2007, 2017). Heterotypic signaling from Pdpn- to Clec2-expressing immune cells is crucial for platelet activation, blood-lymph separation and dendritic cell migration (Astarita et al., 2012; Suzuki-Inoue et al., 2017).

Pdpn function in epithelial cells has been mostly investigated in culture. The ectopic expression of Pdpn in various epithelial cell lines promotes cell motility, modifying actin cytoskeleton organization (Martin-Villar et al., 2006; Wicki et al., 2006; Cueni et al., 2010; Asai et al., 2016). At the molecular level, Pdpn has been shown to interact, via its intracellular domain, with the membrane cytoskeleton linkers ezrin and moesin (members of the ERM family). Moreover, Pdpn has been reported to modulate Rho GTPase activity in fibroblasts and epithelial cells (Martin-Villar et al., 2006; Wicki et al., 2006; Cueni et al., 2010; Acton et al., 2014; Astarita et al., 2015; Asai et al., 2016).

We used a conditional gene deletion approach to investigate the role of Pdpn in mammary development and tumorigenesis. Our data reveal that Pdpn participates in the control of basal SC function through positive regulation of the Wnt/β-cat signaling pathway. Moreover, using a mouse model of β-cat-induced TNBC, we found that Pdpn loss limited tumor-initiating cell expansion, attenuating mammary tumorigenesis.

Pdpn was present throughout the mammary bud on embryonic day (E) 15, and was strongly expressed at cell-cell contacts (Fig. 1A). After birth, Pdpn expression was restricted to the basal cell layer. Ductal myoepithelial cells and the basal cap cells of the terminal end buds (TEBs), specialized structures driving ductal growth and branching during puberty, expressed Pdpn, as did ductal and alveolar myoepithelial cells in pregnant mouse mammary glands (Fig. 1B,C). Of note, Pdpn was concentrated at the apical and lateral surfaces of myoepithelial cells in ducts and alveoli and colocalized with phospho-ezrin/moesin/radixin (p-ERM) (Fig. 1B,C). Basal cells, unlike luminal cells, expressed moesin rather than ezrin (Fig. S1A). Differentiated myoepithelial cells on day 4 of lactation stained negative for Pdpn (Fig. 1D).

To complement immunohistological data, we performed flow cytometry analyses. Mammary cells were isolated at representative stages of postnatal development, including puberty, maturity, early and late gestation, and stained for CD45 (or Ptprc), CD31 (or Pecam1), CD24 and Pdpn. At each stage of development, two distinct populations, Pdpn⁺ and Pdpn⁻, were detected within the pool of CD31/45⁻ CD24⁺ epithelial cells (Fig. 1E). An analysis of gene expression showed that the Pdpn⁺ population consisted of cells expressing the basal-specific markers Trp63 (p63) and Acta2 [alpha 2 smooth muscle actin (SMA)], whereas the Pdpn⁻ population comprised K18⁺ luminal cells (Fig. 1F). These data demonstrate that Pdpn is a reliable and specific surface marker of mammary basal cells in flow cytometry experiments. The Pdpn receptor Clec2 was not expressed in the CD24⁺ basal and luminal cell populations at either the protein (Fig. S1B) or mRNA (data not shown) level, indicating that Pdpn-Clec2 interactions do not occur in the mammary bilayer.

To ascertain that the Pdpn⁺ cell population contained the multipotent SC subset, we compared the regenerative potential of the Pdpn⁺ and Pdpn⁻ epithelial cell populations isolated from adult virgin mammary glands in limiting-dilution transplantation assays. Repopulating activity, characterized by ductal outgrowths in virgin hosts and alveolar development in pregnant recipients, was restricted to the Pdpn⁺ cell fraction (Fig. 1G, Fig. S1C), showing that Pdpn marked multipotent SCs.

To investigate the role of Pdpn in mammary development and SC function, we generated K5Cre;Pdpn^F/F mice, in which Pdpn was deleted specifically in K5-expressing epithelial cells. Flow cytometry and immunohistofluorescence analyses showed Pdpn to be absent from the entire basal cell compartment of adult K5Cre;Pdpn^F/F mammary glands (Fig. 2A, Fig. S2A). As expected, the non-targeted CD24⁻ mammary stromal cells expressed Pdpn (Fig. 2A).

We monitored Pdpn deletion from the mutant epithelium by crossing K5Cre;Pdpn^F/F mice with the Rosa26-lacZ reporter mouse strain (R26) and analyzing lacZ activity in 5-month-old virgin mouse mammary glands. As in the control K5Cre;R26 mammary epithelium, both basal and luminal cells were lacZ positive in mutant K5Cre;Pdpn^F/F;R26 glands (Fig. 2B), indicating that the entire mutant epithelium was derived from Pdpn null multipotent SCs.

The K5Cre;Pdpn^F/F mice were healthy, fertile and displayed normal postnatal development. We did not observe any deleterious phenotype in skin, a tissue targeted by the K5 promoter (Ramirez et al., 2004). Mutant dams were able to feed their pups, indicating that Pdpn loss did not compromise lactation.

To evaluate the role of Pdpn in mammary morphogenesis during puberty, we performed whole-mount and histological analyses on glands from 6-week-old K5Cre;Pdpn^F/F mutant females and their control Pdpn^F/F littermates. Overall, ductal elongation and branching were similar in mutant and control glands (Fig. 2C, Fig. S2B). Pdpn-deficient ducts and TEBs had a normal organization, with inner luminal cells and outer K5- and SMA-expressing myoepithelial cells (Fig. S2C). As in the control epithelium, a large proportion of luminal cells in mutant ducts and TEBs expressed nuclear PR (Fig. S2C). Flow cytometry data showed that the proportion of basal cells was unaffected in the mutant epithelium (Fig. S2D). However, ex vivo mammosphere-formation assays, used to assess stem/progenitor cell activity (Spike et al., 2012; Chiche et al., 2013), revealed that purified Pdpn null basal cells were significantly less clonogenic than control cells (Fig. 2D). Thus, although Pdpn loss had no measurable effect on mammary ductal morphogenesis during puberty, it caused a decrease in basal SC content.

We then compared the mammary glands of adult virgin mutant and control mice. Whole-mount analyses revealed a lower degree of mammary branching in K5Cre;Pdpn^F/F mutant females than in their control littermates (Fig. 2E). Serum hormone levels were comparable in mutant and control adult virgin mice (Fig. S2E). Mutant mammary ducts appeared to be normally organized; however, flow cytometry data revealed that the percentage of basal cells was significantly lower in the Pdpn-deficient than in the control epithelium (Fig. 2F, Fig. S2F). Expression levels of the integrin α6 (Itga6, or CD49f) and β1 (Itgb1, or CD29) chains were identical in control and mutant basal cells (Fig. 2F, Fig. S2G).

Colony formation assays showed that adult Pdpn null basal cells were half as clonogenic as control cells and that the mutant luminal cell population was less clonogenic than the control population, indicating a smaller progenitor content in both compartments (Fig. 2G,H). We explored the functional importance of Pdpn for SC activity further, by comparing the regenerative potential of control and mutant basal cells isolated from mammary glands of adult virgin mice in limiting-dilution transplantation assays. When transplanted in large numbers, Pdpn null basal cells were able to fully colonize cleared fat pads, giving rise to well-organized mammary ducts (Fig. S3A). However, their SC content was reduced by a factor of 2.8, compared with the control basal cell population (Table 1).

We next investigated the long-term regenerative potential of Pdpn-deficient mammary epithelium by serially transplanting epithelial fragments. Pdpn deletion was confirmed by qPCR in mutant basal cells isolated from grafted tissues (Fig. S3B). Basal cells isolated from the primary, secondary and tertiary outgrowths were assessed for their clonogenic capacities. Unlike their control counterparts, mutant basal cells displayed a marked decrease in clonogenic ability after one round of transplantation (Fig. S3C). Consistent with these findings, whole-mount analyses of tertiary outgrowths showed that Pdpn-deficient epithelial fragments were less competent than control grafts for the production of highly ramified ductal structures occupying more than 25% of the cleared fat pad (Fig. 2I).

Altogether, these data indicated that Pdpn deletion affected SC activity in the mammary basal and luminal compartments and restricted the developmental capacity of the mammary epithelium over the long term.

To analyze the molecular alterations induced by Pdpn loss, we first compared by qPCR the expression of a panel of lineage-specific genes in mammary basal cells purified from control and K5Cre;Pdpn^F/F adult virgin mice. As expected, the six analyzed mutant cell preparations were devoid of Pdpn expression (Fig. S4A). Mutant and control basal cells displayed similar levels of expression for essential basal-specific genes including keratin 5 (Krt5), keratin 14 (Krt14), Cdh3 (P-cadherin), the transcription factors Trp63 and Snai2, and the smooth muscle-specific genes Acta2, calponin 1 (Cnn1) and Myh11 (smooth muscle-specific myosin heavy chain) (Fig. 3A). Basal/myoepithelial lineage specification therefore appears to occur normally in the absence of Pdpn.

Interestingly, we found Lgr5 more strongly expressed in Pdpn null than in control basal cells (Fig. 3A). Fzd7 was also upregulated but exhibited variable levels of induction among the mutant samples analyzed. As these two genes are important components of the Wnt signaling pathway, we next compared the expression of several Wnt-related genes in control and mutant basal cells (http://web.stanford.edu/group/nusselab/cgi-bin/wnt/; Rodilla et al., 2009; Moumen et al., 2012; van Amerongen et al., 2012; Meier-Abt et al., 2014; Wang et al., 2015; Yu et al., 2016). Expression of the established Wnt/β-cat target genes Axin2, Myc, Id2, Cdh1 (E-cadherin) and protein C receptor (Procr) and the Wnt pathway-associated genes Lrp5, Lrp6, Lgr4 and Lgr6 was not significantly modulated in Pdpn null basal cells (Fig. S4B). However, cyclin D1 (Ccnd1), keratin 15 (Krt15), versican (Vcan) and Jag1 were less strongly expressed in mutant than in control basal cells (Fig. 3B). Notably, hierarchical clustering of the qPCR data, including those for Lgr5, Fzd7, Ccnd1, Krt15, Vcan and Jag1 expression, clearly separated Pdpn null from control basal cells (Fig. 3B).

We could not analyze Lgr5 and Fzd7 expression at the protein level because no validated antibodies against these receptors in mouse tissues or isolated cells are currently available. However, consistent with the qPCR data, immunofluorescence labeling showed that, unlike K5 and K14, K15 was poorly expressed in the basal cell layer of adult virgin Pdpn-deficient glands, whereas, as previously reported (Meier-Abt et al., 2014), the vast majority of control basal myoepithelial cells strongly expressed K15 (Fig. S4C).

Collectively, our data revealed that Pdpn loss affected the expression of Wnt signaling components, including Wnt receptors and Wnt/β-cat target genes, in mammary basal cells. The low level of Ccnd1 in Pdpn null basal cells is suggestive of impaired cell cycle progression in vivo.

We next compared the developmental potential of control and Pdpn null basal cells in mammosphere formation assays. In agreement with the data from colony-formation assays (Fig. 2G), Pdpn null basal cells isolated from adult virgin glands generated half as many primary spheres as control basal cells (Fig. 3C). Moreover, after three serial passages almost no spheres had formed in the Pdpn-deficient samples (Fig. 3C), indicating a long-term impairment of basal SC maintenance.

The primary spheres derived from mutant basal cells were smaller than those derived from control basal cells (Fig. 3C, Fig. S4D) and contained more cells expressing K5 (Fig. 3D). The number of metabolically active cells, as measured by the CellTiter-Glo assay, was smaller in mutant spheres (Fig. 3E), indicating lower levels of cell survival and/or proliferation. Both Trp63 and Lgr5 were expressed more strongly in mutant than in control spheres, whereas transcript levels for the luminal-specific gene keratin 18 (Krt18) were identical (Fig. 3F). Double immunofluorescence labeling consistently revealed that 90±5% of the K5⁺ cells in mutant spheres coexpressed the luminal marker K8 (Fig. 3G).

Thus, in 3D cultures, Pdpn null basal SCs displayed an impairment of growth and self-renewal relative to control basal cells and they generated more uncommitted K5⁺/K8⁺ cells, indicating an alteration in their developmental potential.

The mammary epithelium of adult virgin K5Cre;Pdpn^F/F mice displayed an increase in the proportion of luminal cells, as a corollary to basal cell depletion (Fig. 2F). However, the mutant luminal cell population was less clonogenic than the control population (Fig. 2H), indicating a reduced progenitor content possibly accompanied by an increase in the non-clonogenic PR-positive subset. Such perturbations may impact the overall production of luminal-specific mediators, in particular Wnt4 and Rspo1 that act synergistically to control the Wnt-responding basal cells (Cai et al., 2014; Rajaram et al., 2015).

To directly investigate the role of Pdpn in the response of mammary basal cells to Wnt signaling, we took advantage of the mouse mammary cell line BC44, characterized in previous studies (Deugnier et al., 2002). Parental BC44 cells display basal progenitor features but lack Pdpn expression. We therefore established stable derivatives expressing full-length Pdpn (BC44-Pdpn). BC44 cells transfected with an empty vector were used as controls. Flow cytometry analysis showed that more than 90% of the BC44-Pdpn cells expressed Pdpn, whereas Pdpn was completely absent from control cells (Fig. 4A). Surface levels of Pdpn expression were similar in BC44-Pdpn cells and basal cells freshly isolated from control mammary epithelium (Fig. 2A, Fig. 4A). Like parental BC44 cells, the transfectants stained positive for the basal-specific markers Itga6, K5 and Trp63 (Fig. 4A,B). Moreover, they strongly expressed basal-specific components of the Wnt signalosome, including Fzd7, Lrp5/6 and the Rspo receptor Lgr4 (Fig. S5A,B).

We compared the Wnt response of BC44 cell transfectants by first stimulating them with Wnt3a, a Wnt ligand known to activate Wnt/β-cat signaling in mammary basal cells, inducing the expression of target genes such as Axin2 (Zeng and Nusse, 2010). Dose-response assays showed that BC44 control cells significantly upregulated Axin2 upon stimulation with 40 ng/ml Wnt3a (Fig. S5C). Strikingly, Wnt3a treatment induced much higher levels of Axin2 expression in BC44-Pdpn than in control cells (Fig. 4C). A similar differential response was observed after cotreatment with Wnt3a and Rspo1, Rspo1 potentiating the Wnt signal as expected (Fig. 4D). Consistent with the gain-of-function studies in BC44 cells, Wnt3a/Rspo1 cotreatment induced higher levels of Axin2 expression in Pdpn-expressing than in Pdpn null basal cells isolated from control and mutant adult virgin glands. respectively (Fig. 4E).

As an additional readout for Wnt/β-cat activation, we performed western blot analysis to assess the levels of activated β-cat in BC44 transfectants. In agreement with the data of the Axin2 induction assays, BC44-Pdpn cells treated with Wnt3a contained larger amounts of active β-cat than stimulated control cells (Fig. 4F). Moreover, following stimulation with Wnt3a, β-cat was detected in the nuclei of multiple BC44-Pdpn cells, whereas control cells rarely displayed β-cat-containing nuclei (Fig. 4G). Unstimulated BC44-Pdpn and control cells contained similar low amounts of active β-cat and did not display nuclear β-cat (Fig. 4F,G), indicating that forced expression of Pdpn was not accompanied by an intrinsic activation of Wnt/β-cat signaling.

Finally, we performed TOPFlash reporter assays in BC44-Pdpn and control cells transiently transfected with a construct encoding a constitutively active N-terminally truncated β-cat (ΔNβcat). The induction of TOPFlash reporter activity in BC44-Pdpn cells was three times as strong as that in control cells (Fig. 4H). Thus, collectively, our data strongly indicate that Pdpn can potentiate Wnt/β-cat signaling events in mammary basal SCs.

There is a large body of data supporting a crucial role for Wnt/β-cat signaling in TNBCs (Pohl et al., 2017). We therefore investigated the possible role of Pdpn in TNBCs, using a mouse model of tumorigenesis established in our previous studies. K5ΔNβcat mice express ΔNβcat in the mammary basal cell layer and develop triple-negative basal-like mammary tumors (Teuliere et al., 2005; Moumen et al., 2013). Immunofluorescence labeling and flow cytometry analysis revealed that Pdpn was expressed in the K5ΔNβcat tumors, both in the CD24⁺ epithelial and the CD24⁻ stromal cell compartments (Fig. 5A,B). A large fraction of tumor cells contained K5 and coexpressed Pdpn (Fig. 5A). Bright Pdpn⁺ cells were often detected at the edge of the tumor (Fig. 5A).

To study the contribution of Pdpn to tumor formation, we crossed K5ΔNβcat mice with K5Cre;Pdpn^F/F mice and compared tumors developed in the presence (K5ΔNβcat;K5Cre⁻;Pdpn^F/F control mice) and absence (K5ΔNβcat;K5Cre⁺;Pdpn^F/F mutant mice) of Pdpn. The absence of Pdpn in the CD24⁺ cell population of the tumors developed by mutant mice was confirmed by flow cytometry and qPCR (Fig. 5B,C). Expression of ΔNβcat was not affected by Pdpn loss (Fig. 5C). Control and mutant tumor cells expressed similar high levels of the basal-specific genes Krt5, Cdh3, Trp63 and Snai2, whereas they poorly expressed the luminal-specific gene Krt18 (Fig. 5C,D). Interestingly, E-cadherin levels were higher and vimentin levels lower in Pdpn-deficient tumors, for both mRNA and protein (Fig. 5D,E). Snai2 expression was unaffected, but the expression of Snai1 and Twist1, two master epithelial-to-mesenchymal transition (EMT)-inducing transcription factors (Nieto et al., 2016), was weaker in the absence of Pdpn (Fig. 5D).

Cohorts of Pdpn mutant females and their control littermates were monitored for mammary tumor formation until the age of 15 months. Tumor onset was slightly but significantly delayed in the absence of Pdpn (Fig. 5F). Moreover, the number of mammary tumors per mouse was markedly smaller in mutant females (Fig. 5F), suggesting a possible effect on the tumor-initiating cell (TIC) pool. We tested this hypothesis by assessing the ability of purified CD24⁺ tumor cells to form spheres in 3D culture. These assays revealed a significantly reduced tumorsphere-forming cell content in Pdpn-deficient as compared with Pdpn-proficient tumors (Fig. 5G). Moreover, mutant primary tumorspheres were smaller than control spheres (Fig. 5G, Fig. S5D) and their ability to generate secondary spheres was severely impaired (Fig. S5E).

Thus, Pdpn deletion attenuated the formation of β-cat-induced mammary tumors and caused TIC depletion. In addition, Pdpn-deficient tumors displayed molecular features associated with a mesenchymal-to-epithelial cell transition (MET) program.

Our study uncovers a role for Pdpn in mammary SC function and tumorigenesis. In particular, we report that Pdpn (1) is a specific marker of the basal cell layer, including multipotent SCs, (2) participates in the control of basal SC activity, and (3) favors mammary tumorigenesis in a model of TNBC. Mechanistically, Pdpn was found to potentiate Wnt/β-cat signaling in basal SCs.

We found that Pdpn was exclusively displayed by basal cells in pubescent or sexually mature virgin and pregnant mice. The cap cells of TEBs, ductal and alveolar myoepithelial cells displayed intense staining for Pdpn, whereas luminal cells were negative. This non-overlapping pattern of expression makes Pdpn a robust surface marker for separating basal and luminal cells by flow cytometry. Importantly, Pdpn labels adult multipotent SCs residing in the basal compartment.

Recent studies have identified Pdpn as a regulator of fibroblastic reticular cell contractility, controlling the acto-myosin cytoskeleton through Rho GTPase activation (Acton et al., 2014; Astarita et al., 2015). Pdpn does not appear to be essential for the contractile function of myoepithelial cells, as it is absent from the lactating gland. Conceivably, the functional importance of Pdpn for acto-myosin contractility depends on cell type.

Although we cannot exclude the existence of specific Pdpn receptors in the mammary bilayer, neither basal nor luminal cells expressed Clec2, the partner of Pdpn in immune cells (Suzuki-Inoue et al., 2007, 2017; Astarita et al., 2012). Nonetheless, Pdpn was concentrated at basal-to-basal and basal-to-luminal cell contacts, suggesting a role in cell-cell communication processes and the existence of regulatory mechanisms governing its polarized distribution. Interestingly, Pdpn colocalizes with p-ERM, further indicating that its cytoplasmic tail might transmit signals into basal cells, as reported for immune and epithelial cells of various origins (Martin-Villar et al., 2006; Acton et al., 2014; Astarita et al., 2015).

We found that the K5Cre-driven embryonic deletion of Pdpn affected mammary SC activity in the virgin gland. Pdpn loss caused a depletion of basal SCs and impaired ex vivo growth and self-renewal potential. It also resulted in a smaller proportion of clonogenic luminal progenitors. Consistent with epithelium-intrinsic defects, Pdpn-deficient epithelial fragments displayed a limited potential for development upon serial transplantation.

In line with the diminished basal SC activity, the basal cell fraction was smaller in the adult virgin mutant epithelium. These alterations were accompanied by a decreased ductal branching complexity. Postnatal mammary development is locally regulated by a complex molecular crosstalk between basal and luminal cells, including direct intercellular and paracrine interactions (Macias and Hinck, 2012; Brisken and Ataca, 2015). Basal cells are known to produce soluble growth factors that regulate the luminal progenitor population and are involved in controlling branching morphogenesis (Macias et al., 2011; Forster et al., 2014; Di-Cicco et al., 2015). In turn, hormone-induced paracrine signals from luminal to basal cells, particularly those mediated by Wnt ligands, play an important role in the expansion of basal SCs and branch formation (Yu et al., 2016). Notably, Pdpn is localized at the basal-luminal interface, where paracrine interactions take place, and could thereby contribute to the Wnt-mediated mechanisms controlling basal SC expansion and mammary morphogenesis (Fig. 6).

The expression of several Wnt-associated genes was altered in Pdpn null basal cells. In particular, we observed decreased levels of Ccnd1, Krt15, Vcan and Jag1, indicating an attenuation of Wnt/β-cat signaling in Pdpn null basal cells and providing a molecular basis for their reduced SC activity. Consistently, freshly isolated Pdpn null basal cells displayed lower levels of Axin2 induction than control cells following cotreatment with Wnt3a and Rspo1.

Wnt signaling events, triggered by Wnt/Fzd and Rspo/Lgr couples in mammary basal cells, are highly complex and remain poorly deciphered (Yu et al., 2016). Distinct Wnt-associated cell populations have been identified in the basal layer of the postnatal gland, including, in particular, minor subsets consisting of Axin2⁺, Procr⁺ and Lgr5⁺ cells (Zeng and Nusse, 2010; van Amerongen et al., 2012; de Visser et al., 2012; Wang et al., 2015). In addition, Lgr4 is widely expressed in the basal cell layer (Wang et al., 2013). The Axin2⁺ and Procr⁺ cell subsets contain Wnt/β-cat-responsive SCs, whereas Lgr5⁺ cells do not display the hallmarks of activated Wnt/β-cat signaling (Zeng and Nusse, 2010; Wang et al., 2015; Fu et al., 2017). Most adult SCs belong to the Lgr5⁻ cell population, which accounts for 90% of the basal compartment (Rios et al., 2014; Wang et al., 2015; Trejo et al., 2017). However, according to recent studies, the Lgr5⁺ cell subset includes a pool of quiescent multipotent SCs that may have persisted from the fetal gland (Fu et al., 2017; Trejo et al., 2017).

Pdpn null basal cells contained higher levels of Lgr5 transcript than control cells. It remains unclear whether loss of Pdpn results in Lgr5⁺ cell enrichment. However, this would be consistent with the lower proliferation activity of Pdpn null basal cells and their propensity to generate spheres enriched in uncommitted K5⁺/K8⁺ cells, a phenotype characteristic of fetal mammary SCs (Spike et al., 2012).

To gain mechanistic insights into Pdpn function out of the complexity of the in vivo context, we used the previously established mammary basal cell line BC44 (Deugnier et al., 2002). These cells express the Wnt signalosome components Fzd7, Lrp5/6 and Lgr4, but are devoid of Pdpn. Notably, the forced expression of Pdpn in BC44 cells strongly enhanced early Wnt/β-cat signaling events triggered by Wnt3a with or without Rspo1, as demonstrated by the nuclear localization of β-cat, and the higher levels of active β-cat and Axin2 induction. Moreover, Pdpn enhanced the induction of TOPFlash reporter activity by ΔNβcat. The Wnt/β-cat pathway being tightly regulated from the cell surface to the nucleus (Driehuis and Clevers, 2017), the level at which Pdpn contributes to the signaling cascade remains to be precisely determined (Fig. 6). However, data from the TOPFlash reporter assay suggest that Pdpn might contribute to the control of cytoplasmic/nuclear signaling events. Interestingly, Rho GTPase signaling, a pathway modulated by Pdpn in certain epithelial cells, has been reported to regulate the nuclear accumulation of β-cat (Schlessinger et al., 2009).

The different functional domains of Pdpn may contribute to various steps in Wnt signal transduction. The transmembrane and cytoplasmic parts of Pdpn have been implicated in targeting of the protein to lipid rafts, specialized membrane domains potentially involved in Wnt signalosome activation (Renart et al., 2015; Özhan et al., 2013). The cytoplasmic association of Pdpn with ERM proteins and cytoskeleton may be required for Wnt/β-cat activation, as recently described for CD44, a cell adhesion molecule that, like Pdpn, appears to potentiate this pathway in epithelial cells (Schmitt et al., 2015). Moreover, Pdpn can interact with CD44 (Martin-Villar et al., 2010). As mammary basal cells express CD44 (Louderbough et al., 2011), a molecular cooperation with Pdpn is possible.

Pdpn overexpression has been documented in various types of carcinomas and is associated with faster tumor progression and invasiveness in models of pancreatic and skin tumors (Wicki et al., 2006; Renart et al., 2015; Suzuki-Inoue et al., 2017). We found that Pdpn was strongly expressed in a mouse model of β-cat-induced TNBCs. In this context, the loss of Pdpn resulted in fewer mammary tumors and in an impairment of tumorsphere formation in culture, consistent with probable TIC depletion. As mammary tumors in K5ΔNβcat mice originate from a dysregulated amplification of basal SCs (Moumen et al., 2013), attenuated tumorigenesis in the absence of Pdpn might be due, in part, to the depletion of basal SCs, the population targeted for oncogenic transformation by β-cat.

Pdpn expression in epithelial cells has been found to favor the acquisition of mesenchymal hallmarks, evoking activation of an EMT program (Wicki and Christofori, 2007; Renart et al., 2015). EMT is viewed as a dynamic and reversible process, comprising multiple transitional cell states between the epithelial and mesenchymal phenotypes (Nieto et al., 2016). Interestingly, we found that the β-cat-induced tumors that developed in the absence of Pdpn displayed features of a MET program. In particular, they presented increased amounts of E-cadherin, lower levels of vimentin and reduced expression of Snai1 and Twist1, as compared with Pdpn-proficient control tumors. Canonical Wnt signaling is closely connected to EMT processes, characterized by the downregulation of E-cadherin expression via the induction of members of the Snail and Twist families (Heuberger and Birchmeier, 2010). Thus, in our model of TNBCs, Pdpn probably favored EMT features by potentiating Wnt/β-cat signaling. It is interesting to mention that expression of Snail1, rather than its paralog Snail2/Slug, has been associated with EMT activation in mammary tumors (Ye et al., 2015).

Overall, our study reveals that Pdpn is specifically expressed by the mammary basal cell layer and participates in the regulation of mammary SC function and tumorigenesis by potentiating Wnt/β-cat signaling. The conserved expression of Pdpn between mouse and human mammary tissue strongly suggests a conserved molecular function. Interestingly, Pdpn is an unfavorable prognostic marker for invasive, ER-negative, ductal breast cancers. Pdpn is largely expressed by cancer-associated fibroblasts but is also present in the tumor cell compartment in a restricted number of cases (Pula et al., 2011; Schoppmann et al., 2012). It would be of interest to further evaluate the clinical importance of Pdpn and investigate whether its expression in the myoepithelium can serve as a predictive marker for progression from in situ to invasive breast cancer, as this cell layer is thought to display tumor-suppressive function (Russell et al., 2015).

K5Cre transgenic mice, expressing Cre recombinase under the control of the bovine keratin 5 (K5) promoter, were kindly provided by Dr J. Jorcano (Ramirez et al., 2004) and the Rosa26-lacZ reporter strain by Dr P. Soriano (Soriano et al., 1999). K5ΔNβcat mice were described previously (Teuliere et al., 2005). Pdpn^F/F mice were generated by Ozgene (Bentley DC, Australia). LoxP sites, flanking Pdpn exon 1 including the starting codon, were introduced through homologous recombination in C57BL/6 mouse embryonic stem cells. Pdpn^F/F mice were mated in a 129/SV×C57BL/6 mixed genetic background with either K5Cre or K5Cre;Rosa26-lacZ or K5ΔNβcat mice. Age-matched Pdpn^F/F or Pdpn^F/F;K5ΔNβcat littermates were used as controls. Hormone serum levels were quantified by Oniris Laboratory (LDHVet, Nantes, France) by ELISA. Mice carrying tumors were sacrificed when at least one palpable tumor (1 cm³) was detected and all glands were analyzed for the presence of lesions. The care and use of animals were conducted in accordance with the European and National Regulations for the Protection of Vertebrate Animals used for Experimental and other Scientific Purposes (facility license C750517/18). All experimental procedures were ethically approved (ethical approval 02265.02).

BC44 cells, established from the mammary mouse epithelial cell line HC11, were grown in RPMI 1640 medium (Gibco Life Technologies) supplemented with 10% fetal bovine serum (FBS; Gibco Life Technologies), 2 mM L-glutamine, 5 μg/ml bovine insulin (Sigma-Aldrich), and penicillin-streptomycin (Gibco Life Technologies), as described (Deugnier et al., 2002). Cells were routinely checked for mycoplasma contamination by Hoechst staining.

Thoracic and inguinal mammary glands from three to six pubertal (6-week-old) or virgin (16- to 25-week-old) mice were pooled for the preparation of a single-cell suspension suitable for flow cytometry, as described in detail elsewhere (Di Cicco et al., 2015). Briefly, minced tissues were transferred to a digestion solution containing 3 mg/ml collagenase (Roche), 100 units/ml hyaluronidase (Sigma-Aldrich) in CO₂-independent medium (Gibco Life Technologies) completed with 5% FBS (Lonza) and 2 mM L-glutamine (Sigma-Aldrich), and incubated for 90 min at 37°C with shaking. Pellets of digested samples were centrifuged (450 g) and successively treated at 37°C with solutions of 0.25% trypsin (Gibco Life Technologies)/0.1% versen (Biochrom) for 1 min, 5 mg/ml dispase II (Roche)/0.1 mg/ml DNaseI (Sigma-Aldrich) for 5 min. Pellets were treated with a cold ammonium chloride solution (Stem Cell Technologies) and filtered through a nylon mesh cell strainer with 40 mm pores (Fisher Scientific) before immunolabeling. The same procedure was applied to mammary tumors with an enzymatic dissociation time extended to 2 h.

Freshly isolated mammary cells or BC44 cells were incubated at 4°C for 20 min with the following antibodies: anti-CD24-BViolet421 (clone M1/69; 101826; 1/50), anti-CD49f-PeCy7 (clone GoH3; 313622; 1/50), anti-CD45-APC (clone 30-F11; 103112; 1/100), anti-CD31-APC (clone MEC13.3; 102510; 1/100), anti-CD54-PE (clone YN1/1.7.4; 116107; 1/50), anti-Pdpn-PE (clone 8.1.1; 127407; 1/50) or anti-Clec2-PE (clone 17D9; MCA5700PE; 1/30); all antibodies were from BioLegend, except anti-Clec2 (Bio-Rad). Labeled cells were analyzed and sorted out using either a FACSVantage flow cytometer (BD Biosciences) or a MoFlo Astrios cell sorter (Beckman Coulter). Data were analyzed using FlowJo software. Sorted cell population purity was at least 95%.

For 2D clonogenic assays, sorted basal or luminal cells were plated on irradiated 3T3 cell feeders in 24-well plates at a density of 2000 or 500 cells per well, respectively. Basal cells were grown in DMEM/F12 medium supplemented with 1% FBS, 2% B27 (Gibco Life Technologies), 5 µg/ml insulin (Sigma-Aldrich) and 10 ng/ml EGF (Invitrogen, Gibco Life Technologies), whereas luminal cells were cultured in DMEM/F12 medium supplemented with 10% FBS, 5 µg/ml insulin, 10 ng/ml EGF and 100 ng/ml cholera toxin (ICN Biochemicals) for 7-8 days, as previously described (Moumen et al., 2012; Chiche et al., 2013).

For mammosphere 3D culture, freshly isolated mammary basal cells or CD24-positive cells from mammary tumors were seeded on ultralow-adherence 24-well plates (Corning) at a density of 5000 or 20,000 cells per well, respectively, in DMEM/F12 medium supplemented with 2% B27, 20 ng/ml EGF, 20 ng/ml bFGF (FGF2; Gibco Life Technologies), 4 μg/ml heparin (Sigma-Aldrich), 10 μg/ml insulin and 2% Matrigel (BD Pharmingen), as described (Spike et al., 2012; Chiche et al., 2013). For second-generation sphere assays, mammospheres were dissociated for 10 min with 0.05% trypsin (Gibco Life Technologies) and reseeded as described above. ImageJ software (NIH) was used to count colonies and mammospheres and quantify their size in pixels. When specified, isolated mammary basal cells cultured for 24 h in the mammosphere condition were treated once with 10 ng/ml mouse recombinant Wnt3a (R&D Systems) or cotreated with Wnt3a and 50 ng/ml mouse recombinant R-spondin 1 (R&D Systems) for 6 h.

Dissected mammary fat pads were spread onto glass slides, fixed in methacarn (1/3/6 mixture of acetic acid/chloroform/methanol) overnight at room temperature and stained with carmine alum (Stem Cell Technologies), as described (Chiche et al., 2013) or fixed in 4% paraformaldehyde overnight at 4°C. ImageJ was used to determine the fat pad filling percentages. For whole-mount X-gal staining, mammary glands were fixed in 2.5% paraformaldehyde in PBS (pH 7.5) for 1 h at 4°C, and stained overnight at 30°C with X-gal staining solution [1.5 mg/ml X-gal, 10 mM K₃Fe(CN)₆, 10 mM K₄Fe(CN)₆, 2 mM MgCl₂, 0.01% Na deoxycholate, 0.02% Tergitol-NP40 in PBS]. For histological analyses, fixed glands were embedded in paraffin, and 6 μm-thick sections were cut, dewaxed and stained with Hematoxylin-Eosin or counterstained with Fast Red for X-gal-stained glands.

Mammary tissue sections were dewaxed, processed for acidic antigen retrieval, incubated overnight at 4°C with primary antibodies, and then at room temperature with secondary antibodies for 2 h.

Prior to immunostaining, freshly isolated cells from mouse mammary glands were cyto-centrifuged onto slides and fixed in cold methanol for 10 min. BC44 cells were cultured onto glass slides for 24 h and then fixed in cold methanol, or in paraformaldehyde for 10 min at room temperature, and treated with 0.5% Triton X-100 for 5 min before immunostaining. Then, fixed cells were incubated with primary antibodies at room temperature for 2 h, with secondary antibodies for 1 h and mounted in Prolong Gold antifade reagent with DAPI (Invitrogen, Gibco Life Technologies).

The following primary antibodies were used: anti-Pdpn (PA2.26; Gandarillas et al., 1997; 1/200), anti-K5 (BioLegend, 905501; 1/1000) and anti-K8 (BioLegend, 904801; 1/100), anti-p63 (Abcam, ab735; 1/50), anti-pan-keratin (Dako, ZO622; 1/100), anti-SMA Cy3-conjugated (Sigma-Aldrich, C6198; 1/200), anti-PR (Santa Cruz, sc-7208; 1/200), anti-p-ERM (Cell Signaling Technologies, 3149; 1/100) and anti-total β-cat (Cell Signaling Technologies, 9587; 1/250).

AlexaFluor 488- or 594-conjugated secondary antibodies were from Molecular Probes (Invitrogen). Image acquisition was performed using a Leica DM 6000B microscope and MetaMorph software (Molecular Devices).

Isolated basal cells or epithelial fragments from adult mammary tissues were transplanted into the inguinal fat pads of 3-week-old BALB/c-Nude females (Charles River) cleared of endogenous epithelium as described (Moumen et al., 2012; Chiche et al., 2013). Primary outgrowths were collected after 6-10 weeks and, when specified, used for serial transplantation assays. Outgrowths were either pooled to isolate mammary cell populations or individually treated for histological analyses, as described above. Repopulating unit frequency was calculated with Extreme Limiting Dilution Analysis software (http://bioinf.wehi.edu.au/software/elda/).

Purified RNA was reverse-transcribed using MMLV H(−) Point reverse transcriptase (Promega), and quantitative (q) PCR was performed by monitoring, in real time, the increase in fluorescence using the QuantiNova SYBR Green PCR Kit (Qiagen) on a LightCycler 480 real-time PCR system (Roche). The values obtained were normalized to Gapdh levels. The primers used for qPCR analysis were purchased from SABiosciences/Qiagen or designed using Oligo 6.8 software (Molecular Biology Insights) and synthesized by Eurogentec. Primers are listed in Table S1.

Protein extracts from isolated tumor cells or BC44 cells were prepared in Laemmli or RIPA buffer, respectively. The following primary antibodies were used for immunoblotting: monoclonal rat anti E-cadherin (clone ECCD-2; Thermo Fisher Scientific, 13-1900; 1/1000), monoclonal mouse anti-vimentin (clone V13.2; Sigma-Aldrich, SAB4200716; 1/1000), anti-β-actin (clone A2228; Sigma-Aldrich, A2228; 1/20,000), anti-active β-catenin (Ser33/37/Thr41; Cell Signaling Technology, 4270; 1/1000) and anti-total β-catenin (clone 14/β-catenin; BD Transduction Laboratories, 610154; 1/10,000).

Stable BC44 transfectants were obtained using Lipofectamine 3000 reagent (Thermo Fisher Scientific). Cells were transfected with pcDNA3.1 empty vector (Thermo Fisher Scientific) or pcDNA3.1-Pdpn full-length, kindly provided by Dr S. Acton (Acton et al., 2014). Transfected cells were collected after geneticin selection (Sigma-Aldrich, 600 μg/ml). The pool of cells expressing Pdpn was then isolated using a FACSAria (BD Biosciences) and further cultured in the presence of geneticin.

Firefly/Renilla luciferase transient transfections were performed using GeneJuice transfection reagent (EMD Millipore), following the manufacturer's instructions (3 µl reagent/µg plasmid DNA). Cells were plated into 12-well dishes at a density of 1.2×10⁵ cells/well. Twenty-four hours later, cells were transfected with 500 ng/well TOPFlash reporter plasmid and 250 ng/well pCGN-ΔNβcat plasmid, kindly provided by Dr A. Ben-Ze'ev (Teuliere et al., 2004). TK-Renilla plasmid was used to monitor transfection efficiency (125 ng/well; Promega). Dual-Glo luciferase (Promega) assay was performed 48 h after the beginning of the transfection procedure, using a FLUOstar OPTIMA microplate reader (BMG Labtech). Values obtained for firefly luciferase were normalized to Renilla luciferase activity.

P-values were determined using Student’s t-test with two-tailed distribution and Welch's correction, assuming both populations have unequal variance. When specified, a Pearson's Chi-square test was applied. For survival curves, a log-rank (Mantel–Cox) test was used. All statistical analyses were performed using GraphPad Prism v6 software.

We are particularly grateful to the personnel of the Animal Facility (Sonia Jannet, Isabelle Grandjean) and the Flow Cytometry Core Facility (Annick Viguier, Sophie Grondin and Zosia Maciorowski) of the Institut Curie. We sincerely thank Nancy Tamir-Geddis and Evens Bousiquot for participating in the work during their internships, Pierre de la Grange (GenoSplice, France) for generating heatmaps, Sophie Acton (MRC Laboratory for Molecular Cell Biology, London, UK) for providing Pdpn constructs, and Mathilde Romagnoli for helpful discussions.