A molecular landscape of quiescence and proliferation highlights the role of Pten in mammary gland acinogenesis

The mammary gland is unparalleled with regard to development and differentiation. A rudimentary epithelial tree at birth, the mammary gland further branches in puberty and will only fully grow and differentiate during pregnancy and lactation. Upon weaning, the mammary gland goes through involution and massive remodelling, returning to a histological architecture similar to the one found before pregnancy (Sternlicht et al., 2006). This cycle repeats as many times as pregnancy, lactation and weaning occur in a female mammal lifetime (Inman et al., 2015; Sternlicht et al., 2006). All these events are coordinated by local and systemic stimuli (Sternlicht et al., 2006). Therefore, the developing mammary gland offers an invaluable tool for the study of spatiotemporal control of cell behaviour by the cell microenvironment.

As a complex and widely diverse mesh of proteins and soluble factors, the extracellular matrix (ECM) not only confers physical support for cells, but also provides biochemical and mechanical cues perceived by the cells (Nelson and Bissell, 2006; Sternlicht et al., 2006), which can respond accordingly by altering cell phenotype and, in turn, remodelling the ECM in a dynamic and reciprocal process (Bissell et al., 1982). Cells attach to specific ECM molecules via surface receptors, including integrins and the dystroglycan receptor (encoded by Dag1) (Cloutier et al., 2019; Pozzi and Zent, 2011). Intracellularly, these receptors are physically linked to a range of cytoskeleton and signalling proteins; hence, they mechanically and biochemically connect cells to insoluble extracellular cues (Nelson and Bissell, 2006; Pozzi and Zent, 2011).

Thus, the composition and mechanical properties of the ECM, along with the receptors and downstream effectors expressed by the cells, will dictate cell behaviour, modulating processes such as proliferation, quiescence, differentiation, motility and polarity (Bissell et al., 1982; Nelson and Bissell, 2006).

Epithelial cells reside on top of a specialized ECM compartment, the basement membrane (BM), the formation and microarchitecture of which is nucleated and stabilized mainly by laminins (LeBleu et al., 2007). The BM physically separates epithelial cells from the surrounding stroma, in which the ECM is mainly composed of type I collagen and fibronectin (LeBleu et al., 2007). An intact BM is essential to sustain epithelial homeostasis (Gaiko-Shcherbak et al., 2015; Khalilgharibi and Mao, 2021).

During pubertal mammary epithelial development, the BM is thicker in ducts than in terminal end buds (TEBs), (Sternlicht et al., 2006) and this has been extensively linked to the spatiotemporal regulation of mammary gland morphogenesis in vivo and in cell culture models such as ECM overlays and three-dimensional (3D) cell cultures, where cells retain the ability to respond to laminin-triggered quiescence and differentiation, even in the presence of mitogens and growth factors (Bhat et al., 2016; Fiore et al., 2017; Inman et al., 2015; Spencer et al., 2011).

In this work, we provide a molecular landscape of key regulators of laminin-rich ECM (lrECM)-triggered mammary epithelial cell quiescence and differentiation, which we found involves multiple layers of regulation at the mRNA and protein level. Interestingly, cell cycle arrest was achieved despite the active status of Fak (also known as PTK2), Src and phosphoinositide 3-kinases (PI3Ks), indicating the existence of a ‘disconnecting node’ between upstream and downstream proliferative signalling in laminin-induced quiescence. Our data indicate that Pten, a tumour suppressor that counteracts PI3K activity, fulfils this role: Pten was upregulated in lrECM-treated mammary epithelial cells, and Pten inhibition hindered quiescence by restoring activity of the Akt, mTORC1 and mTORC2 pathways, as well as mitogen-activated protein kinase (MAPK) signalling via Mek [here referring collectively to Mek1 (MAP2K1) and Mek2 (MAP2K2)] and Erk [here referring collectively to Erk1 (MAPK3) and Erk2 (MAPK1)]. In the developing mammary gland, the levels and tissue distribution of Pten were positively correlated to the degree of laminin staining and the presence of lumen. Furthermore, Pten inhibition in mammary 3D acini increased cell proliferation, disrupted polarity and led to loss of the lumen in both forming and mature structures. Taken together, these data highlight the core role of Pten in mammary gland quiescence regulation, with Pten downregulation being necessary for epithelial cell growth and Pten upregulation sustaining quiescence and tissue architecture.

When the BM that envelopes the mammary epithelia is mimicked in culture systems using lrECM, either on a two-dimensional ECM overlay or in 3D assays, primary epithelial cells and some established non-malignant epithelial cell lines, such as murine EpH4 cells, exit the cell cycle (Mroue and Bissell, 2012), thus representing a physiological and suitable model for the study of microenvironmentally regulated cellular quiescence.

We further characterized this model using flow cytometry, cell cycle fluorescence probes and quantitative image analysis. As previously reported, lrECM overlay on two-dimensional cultures triggered remarkable morphological alterations and G1 arrest in EpH4 cells (Mroue and Bissell, 2012; Spencer et al., 2011; Xu et al., 2010), reducing the overall cell number without affecting cell viability (Fig. 1A–F; Fig. S1). LrECM-induced G1 arrest was consistently observed either by analysing DNA content via flow cytometry (Fig. 1D) or via fluorescence microscopy of EpH4 cells expressing the FUCCI reporter system (EpH4-ES-FUCCI) (Fig. 1F). The FUCCI system is based on the cyclic degradation of fluorescent probes labelling Cdt1 (mCherry) and geminin (mCitrine). Cdt1 accumulates during G1 and is degraded in early S phase, whereas geminin builds up over the S, G2 and M phases (S/G2/M); as such, FUCCI cells express mCherry during G1 and mCitrine during S/G2/M, allowing for monitoring of the cell cycle in live cells (Zielke and Edgar, 2015).

LrECM-induced quiescence was also accompanied by a decrease in the activity of the cyclin-dependent kinases (Cdks) Cdk4 and Cdk2 (Fig. 1G,H). Cdk2 and Cdk4 activities were evaluated using specific fluorescent reporters and fluorescence microscopy (Spencer et al., 2013; Yang et al., 2020). In this system, cells express a nuclear marker (histone H2B–mTurquoise) and probes for Cdk2 (DHB–mVenus) and Cdk4 (mCherry–Cdk4KTR; this probe also detects Cdk6 activity, but in mammary cells, Cdk4 is the predominant kinase) activities. Upon phosphorylation by Cdk2, or by Cdk4 or Cdk6, the respective fluorescent probe is translocated from the nucleus to the cytoplasm. Therefore, high Cdk activity results in the probe being translocated to the cytoplasm, and thus, Cdk activity can be estimated by the ratio of the cytoplasmic and nuclear mean fluorescence signal of a given probe (Spencer et al., 2013; Yang et al., 2020).

Quiescent cells were smaller than freely cycling cells, even when only cells in G1 were considered (Fig. 1I). This was consistent when size was estimated via flow cytometry (Fig. 1I, left) or via microscopy (Fig. 1I, right). Despite these differences in estimated cell size between quiescent and cycling G1 cells, the correlations between nuclear area and Cdk activity, for both Cdk4 and Cdk2, in freely cycling and lrECM-treated cells were weak (Fig. S2), and thus, discrimination between cycling G1 cells and G1-arrested cells based solely on nuclear area was not feasible. A possible explanation for this observation is that newly born cells would be small and would present low Cdk activity, but, unlike quiescent cells, their cell size and Cdk activity would build up over time (Rhind, 2021; Spencer et al., 2013), whereas quiescent cells would remain small and with low Cdk activity.

Many molecular circuits could be involved in the acquisition and maintenance of the quiescent phenotype, and this probably accounts for the morphological and functional diversity amongst quiescent cells in nature (O'Farrell et al., 2011). We sought to delineate a landscape of key molecules involved in cell cycle progression that are regulated during laminin-induced quiescence in mammary epithelial cells.

In addition to becoming quiescent in the presence of lrECM, upon concomitant exposure to prolactin, EpH4 cells become functionally differentiated for lactogenesis and produce milk-related proteins (Xu et al., 2009). Thus, the EpH4 model is largely used for the study of cellular and molecular mechanisms underlying mammary gland functional differentiation. In this sense, untreated (proliferating) cells could be considered a parallel for the proliferative TEBs found in the developing mammary gland, whereas lrECM-treated cells would be comparable to the quiescent cells found in the ducts and immature alveoli. Similarly, cells treated with both lrECM and prolactin would represent cells found in the alveoli of a fully mature, milk-producing mammary gland.

We compared proliferating, quiescent and lactogenic EpH4 cells with respect to some key cell cycle regulators, initially at the transcriptional level (Fig. 2). The mRNA levels of several cell cycle regulators dramatically changed upon lrECM treatment, regardless of co-treatment with prolactin (Fig. 2A; Fig. S3). Confirming the functional differentiation state towards a lactogenic programme (Xu et al., 2009), only exposure to both lrECM and prolactin triggered the expression of Csn2, the gene encoding β-casein (Fig. 2A). The mRNA expression of most positive regulators of the cell cycle (such as cyclins and Cdks) plummeted upon lrECM treatment (Fig. 2A), but surprisingly, so did the expression levels of some negative regulators of the cell cycle (such as Trp53, Cdkn1a, Cdkn2c and Cdkn2d). On the other hand, the mRNA levels of Pten, Dag1 and Cdkn1b were unchanged across all treatments (Fig. 2A). These data confirm some previous findings indicating that upon laminin-induced quiescence acquisition there is a general repression of transcription (Spencer et al., 2011; Fiore et al., 2017). But how are these cells arrested if, apparently, expression of negative regulators of the cell cycle is either unchanged or also downregulated? The lack of alteration or even the decrease observed in mRNA levels of cell cycle negative regulators might indicate that these molecules are most likely to be controlled at the protein level during lrECM-induced quiescence (Koepp, 2014); therefore, we assessed the protein levels of a range of known positive and negative regulators of the cell cycle by western blotting (Fig. 2B–F).

LrECM-induced quiescent cells displayed dramatic changes in protein expression and/or phosphorylation levels of key cell cycle regulators (Fig. 2B–F). Again, lrECM-treated cells exhibited similar protein expression profiles concerning cell cycle regulators, regardless of prolactin supplementation (Fig. 2B,F), and likewise, cells treated only with prolactin exhibited a protein expression profile similar to that of untreated cells (Fig. 2B,E).

Quiescent cells displayed decreased protein levels of positive regulators of proliferation, such cyclins and Cdks (Fig. 2B–D), matching their transcriptional repression (Fig. 2A), which resulted in the near absence of hyperphosphorylated retinoblastoma protein (Rb, also known as RB1) in lrECM-treated quiescent cells (Fig. 2B–D). Quiescent cells also had increased protein levels of some negative regulators of the cell cycle, such as Pten (Worby and Dixon, 2014), p27 (also known as CDKN1B) (Sherr and Roberts, 1999) and the dystroglycan receptor (Sgambato et al., 2004), which showed no alterations at the transcript level (Fig. 2A). The balance between cyclins and cyclin-dependent kinase inhibitors (CKIs) is believed to be one of the components that dictates cell cycle progression (Gérard and Goldbeter, 2012; Yang et al., 2017); thus, the decreased activity of Cdk4 and Cdk2 observed in lrECM-treated cells (Fig. 1G,H) could be, at least in part, attributed to the lower levels of cyclins and Cdks and the increased levels of p27 found in quiescent cells (Fig. 2B–D).

LrECM treatment attenuated many proliferative signalling pathways, such as the MAPK pathway (Fiore et al., 2017), as evidenced by reduced levels of phosphorylated Mek and phosphorylated Erk, and the PI3K–Akt–mTORC1 and mTORC2 (mTORC1/2) pathways (Fiore et al., 2017), as evidenced by reduced levels of phosphorylated Akt kinase (herein referring collectively to Akt1, Akt2 and Akt3) and phosphorylated S6 (also known as RPS6) (Fig. 2B–D). Surprisingly though, lrECM-treated quiescent cells displayed increased or unchanged levels of active (phosphorylated) Fak, Src and PI3K (Fig. 2B–D). However, downstream signalling pathways triggered by these kinases were widely downregulated in the presence of lrECM (Fig. 2B–D). These included the PI3K–Akt–mTORC1/2 axis and the MAPK pathway (Fig. 2B–D). Fak, Src and PI3K are components of adhesion complexes and are activated upon integrin binding to ECM molecules (Glukhova and Streuli, 2013). Fak, Src and PI3K also take part in the control of mitogenic signalling triggered by the binding of growth factors to receptor tyrosine kinases (RTKs) (Streuli and Akhtar, 2009). There is significant cross-activation between Fak, Src and PI3K, and although each of them has specific functions in cell behaviour, their activation usually triggers Akt–mTORC1/2 and MAPK signalling (Moreno-Layseca and Streuli, 2014), which places them as positive regulators of cell proliferation.

Decreased phosphorylation levels of Akt and Mek–Erk have been previously reported in EpH4 acini (Janda et al., 2002; Tognon et al., 2011) and also in non-malignant human mammary epithelial S1 (Beliveau et al., 2010; Fiore et al., 2017; Onodera et al., 2014) and MCF 10A (Debnath et al., 2003; Watt et al., 2017) cells undergoing laminin-induced quiescence.

The Ras–Raf–Mek–Erk axis is the most crucial MAPK signalling pathway for the positive regulation of cell proliferation in epithelial cells (Sun et al., 2015). It is manly triggered by the binding of mitogens and growth factors to RTKs, but it is also modulated by the ECM (Moreno-Layseca and Streuli, 2014; Streuli and Akhtar, 2009; Sun et al., 2015). Complementarily, mTOR controls cell growth by promoting anabolism and suppressing catabolism, and its function might be different depending on upstream signalling (Kim and Guan, 2019). Consistent with the smaller cell size observed in quiescent cells (Fig. 1I), we observed decreased levels of phosphorylation of S6 (at S240 and/or S244) and Akt (at S473), reflecting the activities of mTORC1 and mTORC2, respectively.

Another feature of quiescent cells that captured our attention was the presence of phosphorylated PI3K p55 only in lrECM-treated cells (Fig. 2B). The PI3K complex is comprised of catalytic (p110) and regulatory (p85 or its alternative spliced versions p50 and p55) subunits. Specific functions for different PI3K isoforms in mammary gland development have been described, including a role of p110γ (encoded by Pik3cg) in lumen formation (Peng et al., 2015) and for p55α and/or p50α (encoded by Pik3r1) in cell death induction during mammary gland involution (Pensa et al., 2014). Thus, phosphorylated PI3K p55 might have functions beyond the regulation of proliferation in lrECM-induced quiescence, such as suppression of apoptosis when mammary epithelial cells are exposed to a reconstituted BM (Boudreau et al., 1995).

The data on protein expression prompted us to assess the contribution of different signalling pathways to the cell cycle arrest upon lrECM treatment by using commercially available inhibitors (Fig. 3; Table S3).

As expected, treatment with lrECM increased the percentage of cells in G1 (quiescent), and treatment with pharmacological inhibitors of intermediate and downstream positive regulators of the cell cycle (such as mTORC1/2, Akt, Mek, Cdk4 and Cdk2) mostly decreased cell cycle progression and, in general, increased the percentage of Eph4-ES-FUCCI cells arrested in G1 in both control and lrECM conditions (Fig. 3A and B, respectively).

Even though both cyclinE1–Cdk2 and cyclinD1–Cdk4 were less active and had lower expression levels in quiescent cells (Figs 1G,H, 2A–D), the effects of Cdk2 inhibition on cell cycle and cell number were more prominent than those observed for Cdk4 inhibition (Fig. 3B). Cells can often bypass loss of cyclins or Cdks and progress through the cell cycle by activating alternative cyclin–Cdk complexes (Malumbres and Barbacid, 2009; Shapiro, 2006; Whittaker et al., 2017). Particularly, mature cyclinD–Cdk4 complexes are refractory to palbociclib, the Cdk4 inhibitor used in this work. Palbociclib is only functional if binding to Cdk4 by itself, before Cdk4 complexes with cyclinD, and therefore inhibits cell cycle progression by preventing the assembly of new cyclin D–Cdk4 complexes and their activity, and also by indirectly inhibiting Cdk2 (Guiley et al., 2019). Furthermore, the smooth-edged and compact morphology of cell clusters treated with Cdk2 inhibitor (Fig. 3A, Cdk2i) closely resembled those formed by quiescent lrECM-treated cells (Fig. 3B, lrECM), which was not the case for cells treated with the Cdk4 inhibitor (Fig. 3A, Cdk4i). These data could be indicative that Cdk2 and Cdk4 have distinct roles in lrECM-induced quiescence.

Inhibition of Src, PI3K, mTORC1/2 and Akt all gave rise to small, G1-arrested cells, whereas Fak inhibition arrested cells in G2 (Fig. 3A,B). G2 arrest upon Fak inhibition was evidenced by increased mCitrine signal along with decreased cell number (Fig. 3A,B), and this is in agreement with previous reports (Aboubakar Nana et al., 2019; Cabrita et al., 2011). Inhibition of the Jak–Stat pathway or Ilk had no or only minor effects on the cell cycle (Fig. 3A,B). This is consistent with previous reports that have proposed a role for Jak–Stat signalling in milk production rather than proliferation (Xu et al., 2009). Similarly, Fak, Akt and Ilk signalling have all been linked to cell survival and/or polarity (Keely, 2011; Lee and Streuli, 2014), whereas Src and PI3K–Akt–mTORC1/2 signalling are involved in the control of cell size (Cohen, 2005; Kim and Guan, 2019).

Another key observation from this assay was that, despite the high levels of phosphorylated Fak, Src and PI3K proteoforms found in quiescent cells (Fig. 2), inhibition of these molecules changed cell proliferation (arresting cells in G1 for Src and PI3K inhibition and in G2 for Fak inhibition) (Fig. 3). G1 arrest upon Src and PI3K inhibition (Chu et al., 2007; Liu et al., 2004), as well as G2 arrest in response to Fak inhibition (Aboubakar Nana et al., 2019; Cabrita et al., 2011) have been previously reported. Therefore, the activity of these signalling pathways, as a default, contributes to proliferation in mammary epithelial cells.

How could Fak, Src and PI3K be active, but their downstream proliferative pathways be downregulated in quiescent lrECM-treated cells? There must be a ‘disconnecting point’ between upstream and downstream proliferative signalling triggered lrECM treatment.

Pten, which was found to be upregulated in lrECM-treated cells (Fig. 2B–D) could fulfil this role: its lipid phosphatase function counterbalances PI3K activity by converting phosphatidylinositol (3,4,5)-trisphosphate (PIP₃) to phosphatidylinositol (4,5)-bisphosphate (PIP₂) (Worby and Dixon, 2014), whereas its protein phosphatase activity can target Fak (Tamura et al., 1999). Strikingly, Pten inhibition led to increased proliferation, increased cell number, a reduced percentage of cells in G1 and an increased percentage of cells in S/G2/M (Fig. 3). Alterations induced by Pten inhibition were clearer in lrECM-treated cells, in which the distribution of cells across the phases of the cell cycle resembled that observed for freely cycling (control) cells (Fig. 3A,B).

Furthermore, inhibition of Pten increased the levels of active Fak (phospho-Fak^Y397) and Src (phospho-Src^Y416), and partially restored PI3K (phospho-Akt^T308), MAPK (phospho-Erk), mTORC1 (phospho-S6) and mTORC2 (phospho-Akt^S473) signalling in lrECM-treated cells (Fig. 4). Treatment with Pten inhibitor in the absence of lrECM had overall milder effects in further increasing phosphorylated levels of Fak, Akt, Erk and S6 (Fig. 4), which is consistent with the notion that sub confluent freely cycling (control) cells in culture are in exponential growth, near the maximum of their proliferation rates (https://www.atcc.org/resources/culture-guides/animal-cell-culture-guide#Cell).

These data, in combination with the protein expression analysis (Fig. 2B–D), indicated that Pten could act as a disrupting node for Fak–Src–PI3K proliferative signalling in mammary epithelial cells, especially upon quiescence induction by a laminin-rich microenvironment.

Our data obtained for mammary epithelial cells pinpointed Pten as being upregulated upon lrECM treatment and as an important regulator of laminin-induced quiescence. To investigate whether this holds true in vivo, we assessed the correlation between laminin deposition and Pten staining in developing murine mammary glands (Fig. 5). Ducts, which are laminin-rich, mostly quiescent epithelial structures in the tubuloalveolar mammary network (Sternlicht, 2005) had higher levels of Pten (Fig. 5B,C). TEBs, which are the laminin-poor, invasive and proliferative tip of the growing epithelia in the mammary gland (Sternlicht, 2005) had overall lower levels and an unclear pattern of Pten staining (Fig. 5B,C). Furthermore, Pten intensity was found to be positively correlated with the degree of laminin deposition (Fig. 5B,D).

One feature that became clear while analysing the images of the developing mammary glands was that Pten presented distinct staining patterns in ducts and TEBs, and amongst specific regions or cell types. In the ducts, Pten showed strong apicolateral staining in the membrane of luminal cells (Fig. 5B, top); this pattern was only seen in cells surrounding the nascent lumen in TEBs (Fig. 5B, bottom).

Once we observed a predominance of apicolateral Pten staining, particularly in polarized luminal epithelial cells in the duct, we next stained developing mammary glands for Pten and F-actin, and imaged ducts and TEBs optically sliced at different focal planes using confocal and super-resolution microscopy (Fig. 6). In ducts, Pten appeared in puncta at the basal region (portion of the cell laying on the BM) and at the membrane along with the actin belt in the apical region (facing the lumen) (Fig. 6A). In TEBs (Fig. 6B), which mostly lack polarization, there was not a clear pattern of Pten staining. The exceptions were cap cells, which are found at the front edge of TEBs, where Pten was mainly found in the nuclei (Fig. 6B, middle row), and cells surrounding the nascent lumen, which presented a staining pattern of Pten and F-actin similar to that observed in the apical region of ducts (Fig. 6B, top row).

Cells near the nascent lumen seemed to present higher overall levels of Pten in comparison to the majority of the cells within the TEB, namely body cells (Fig. 6C). Furthermore, Pten was found predominantly in the cytoplasm in luminal cells in the ducts and in cells near the nascent lumen in TEBs (Fig. 6D), whereas body cells in the TEB often presented a more even distribution of Pten between the cytoplasmic and nuclear compartments (Fig. 6D). Basal cells in ducts also presented a more even distribution of Pten across both compartments, whereas basal (cap) cells in TEBs most often presented a predominance of nuclear Pten (Fig. 6D).

The presence of Pten in the membrane is directly related to its phosphatase activity (Brito et al., 2015) and might therefore indicate not only that Pten activity is higher in ducts and near the nascent lumen, but also that this activity is predominantly polarized to the apical surface. The fact that cap cells, which are believed to be multipotent mammary stem cells (Paine and Lewis, 2017; Visvader and Stingl, 2014), presented high Pten levels in the nuclei might be related to the contribution of Pten to genomic stability (Ho et al., 2020), which would be particularly important for progenitor cells, as mutations in these cells would be passed on to all their descendant cells.

The distinct levels and cell and tissue localization of Pten in ducts and in TEBs prompted us to use 3D acini cell culture models to evaluate the contribution of Pten to inducing and sustaining cell cycle arrest, cell polarity and lumen formation. Upon lrECM exposure in 3D, EpH4 cells give rise to quiescent polarized acini with a central lumen, recapitulating many cellular and molecular events observed during mammary gland morphogenesis in vivo (Xu et al., 2009).

To better understand the role of Pten in the interplay between quiescence acquisition and maintenance, as well for establishing and sustaining apicobasal polarity and tissue architecture, we pharmacologically inhibited Pten in 3D epithelial acini at two distinct stages: during development and in mature structures (Fig. 7A,F).

In both cases, Pten inhibition led to increased proliferation, which was assessed via the expression of FUCCI sensors (Fig. 7B,C) or by Ki-67 (MKI67) staining (Fig. 7G,H). These data indicate that in mammary epithelial cells, Pten activity is necessary to induce quiescence during development and to sustain quiescence in mature structures (Fig. 7B,C,G,H). Structures treated earlier (and therefore for longer) with Pten inhibitor were larger and less spherical than those that were exposed to the inhibitor later, although in both conditions the inhibitor-treated structures were larger and more irregular than those observed in control conditions (Fig. 7B,D,E).

Differentiated EpH4 acini express apicobasal markers such as ZO-1 (apical marker; also known as TJP1) and integrin α6 (basal marker; also known as ITGA6) (Fig. 7G, left). Pten inhibition, either during development or in mature acini, unleashed proliferation and disrupted tissue architecture, as evidenced by the presence of irregular structures that lacked a lumen and had widespread membrane staining for ZO-1 and integrin α6 (Fig. 7B–D,G–I), consistent with a role of Pten in inducing and sustaining quiescence and tissue architecture.

By using a relevant cell culture model where, even in the presence of growth factors, exposure to a BM surrogate (Kleinman and Martin, 2005) triggers quiescence and differentiation in mammary epithelial cells, akin to what occurs in vivo (Fiore et al., 2017; Paine and Lewis, 2017; Sternlicht et al., 2006; Xu et al., 2010), we have provided a snapshot of the molecular landscape of cell cycle regulators in lrECM-induced quiescent cells, which in combination with microscopy data derived from fluorescent cell cycle indicators and perturbations made using specific inhibitors, allowed us to investigate in detail the signalling networks that might lay the molecular foundation for specific traits of quiescent cells.

For example, the complexity of alterations found in this molecular landscape helps to explain why overexpressing a CKI or downregulating cyclins and/or Cdks might lead to irreversible cell cycle arrest, not fully recapitulating the transient state of quiescence (Lim and Kaldis, 2012; Sang et al., 2008).

Also, in order to proliferate, cells need to grow throughout G1 until they achieve a specific target size (Gokhale and Shingleton, 2015; Lloyd, 2013; Vollmer et al., 2017). Thus, although cell growth and cell proliferation can be separable events, there is a high degree of coupling between cell size and cell cycle progression (Kafri et al., 2013). Mechanisms underlying cell size control are poorly understood (Lloyd, 2013), but mTOR is a clear positive regulator of cell growth (Kim and Guan, 2019). The reduced size observed in quiescent cells, along with reduced mTOR signalling upon lrECM treatment and induction of cell cycle arrest by mTORC1 and mTORC1/2 inhibitors, combined with previous findings (Kafri et al., 2013; Kim and Guan, 2019), indicate that reduced cell growth might be one of the mechanisms behind cell cycle arrest in BM-induced quiescence.

Another intriguing observation was the differential effect of Cdk2 and Cdk4 inhibition in cell proliferation and morphology, together with the interplay between cyclins, Cdks and CKIs during lrECM-induced quiescence. A few things need to be considered to better explore these data. First, in the classical model of cell cycle progression, in response to mitogenic stimuli during G1, cyclinD–Cdk4 complexes phosphorylate and inactivate Rb, which releases E2F transcription factors that, in turn, switch on the expression of a range of positive regulators of the cell cycle, including cyclinE, which in complex with Cdk2 further phosphorylates and inhibits Rb. This process culminates in Rb hyperphosphorylation, at the ‘restriction point’ of the cell cycle, after which cells are committed to proliferation and will progress through cell cycle regardless of mitogenic stimuli (Blagosklonny and Pardee, 2002; Gérard and Goldbeter, 2012; Pardee, 1974). It has been recently demonstrated, however, that molecular events occurring much earlier than Rb hyperphosphorylation, including the degree of Cdk2 activity in newly born cells, are the determinant factors behind cell cycle commitment (Spencer et al., 2013). Second, although the CKIs p21 (also known as CDKN1A) and p27 supress Cdk2 activity, they are required for the assembly of cyclinD–Cdk4 complexes. In this context, p21 is generally an inhibitor or a poor activator, whereas p27, depending on its phosphorylation status, can be either an inhibitor or a strong activator for cyclinD–Cdk4 complexes (Guiley et al., 2019; Ray et al., 2009). Thus, in addition to kinase activity, cyclinD–Cdk4–p27 complexes would contribute to cell cycle progression by titrating p27 away from Cdk2 complexes (Guiley et al., 2019; Ray et al., 2009).

Although a much more in-depth analysis of these mechanisms would be necessary to draw any conclusions, it is tempting to speculate that during lrECM-induced cell cycle arrest, inhibition of Cdk2 activity could be a core event for acquiring and sustaining the quiescent phenotype. LrECM-induced inhibition of Cdk2 activity, due to both reduced levels of cyclinE and Cdk2, along with increased p27, in a context where cyclinD1 and Cdk4 were also reduced, would allow for cell cycle arrest even in the presence of mitogens. Which of these events comes first, how this is regulated at the molecular level and whether this stands true for lrECM-induced quiescence would need extensive investigation and is beyond the scope of the present work. Moreover, although sufficient to map the overall molecular landscape of cells induced to quiescence by the ECM, our current bulk analysis relying on RT-qPCR and western blotting might not have completely uncovered important aspects of the decision signalling network made at the single-cell level. By employing single-cell level methodologies in future investigations, we could gain enhanced granularity, thereby elucidating cell subpopulations and distinct cellular dynamics concerning Cdks and other key signalling molecules.

The apparent discrepancy between the activation status of some upstream (Fak, Src and PI3K) and downstream (Akt, mTORC1/2 and MAPK) regulators of cell cycle progression could also be consistent with the idea that proliferative signals from the ECM are sensed by quiescent cells, but that proliferative signalling effectors are actively shut down by laminin-triggered alterations. Depending on spatiotemporal regulation, signalling integration and subcellular localization, Fak, Src and PI3K are involved in a range of cellular processes other than proliferation, such as cell survival (Aksamitiene et al., 2012), invasion (Levental et al., 2009), migration (Gehler et al., 2013), polarity (Liu et al., 2004) and lumen formation (Peng et al., 2015). Shutting down the proliferative signalling arms of these pathways might open space to the rise of other signalling routes and, for instance, result in a shift in cell phenotype towards differentiation. Our findings indicate that Pten might, at least in part, fulfil this role. Pten was upregulated at the protein level in quiescent cells (Fig. 2), and its inhibition led to increased proliferation, restored Akt–mTORC1/2 and MAPK signalling in the presence of quiescence-inducing signals from lrECM in mammary epithelial cells, and disrupted quiescence, polarity and tissue architecture in 3D acini (Figs 3, 4 and 7).

Pten is a well-known tumour suppressor, and its role in restraining cell proliferation has been extensively demonstrated in a wide range of tissues and cell types (Lee et al., 2018). The PI3K pathway is one of the most mutated pathways in cancer, with Pten figuring, in fact, as the most frequently altered PI3K pathway member (Millis et al., 2016). Pten loss has been associated with increased proliferation due to upregulation of PI3K–Akt–mTORC1/2 and MAPK signalling pathways (Gu et al., 1998; Worby and Dixon, 2014), but even discreet alterations in Pten levels can trigger dramatic changes in cell behaviour and increase the risk of developing cancer (Lee et al., 2018).

The mechanisms by which laminin upregulates Pten and whether this regulation follows the rules of dynamic reciprocity (Bissell et al., 1982) remain unclear, but, in 3D culture of mammary epithelial cells, Pten levels have been linked to increased dystroglycan receptor expression (Muschler et al., 2002), as well as to E-cadherin (CDH1) expression and localization (Fournier et al., 2008). Tissue architecture integrity and the levels and localization of Pten have been pointed out to exist in dynamic reciprocity (Fournier et al., 2008). We did find increased levels of dystroglycan receptor upon lrECM-induced quiescence (Fig. 2A,D,E). The dystroglycan receptor has two subunits encoded by a single transcript (Moore and Winder, 2012). The extracellular subunit α-dystroglycan is subjected to posttranslational modifications through glycosylation, which confer its ability to bind laminin (Cloutier et al., 2019; Pozzi and Zent, 2011). The intracellular subunit β-dystroglycan usually binds to α-dystroglycan and to cytoskeleton proteins, although it can also be found in the nucleus (Cloutier et al., 2019; Gracida-Jiménez et al., 2017). Dystroglycan receptors are associated with attenuation of integrin signalling (Ferletta et al., 2003), cell cycle arrest (Sgambato et al., 2004) and polarity (Muschler et al., 2002) in mammary epithelial cells treated with lrECM. Therefore, increased expression of Pten and dystroglycan receptor might cooperate to trigger and sustain quiescence in growth-suppressive microenvironments.

Although the BM is deposited mainly by basal cells, cells across the entire epithelial structure are under the influence of cues triggered by it, in both physiological and pathological conditions (Allinen et al., 2004; Gudjonsson et al., 2002). In this sense, correlations between BM thickness and some regulators of proliferation and migration, including nuclear actin (Spencer et al., 2011) and nuclear galectin-1 (Bhat et al., 2016), have been reported in the developing mammary gland, and our data indicate that Pten might also fall into this category. Pten displayed differential levels and staining patterns in TEBs versus ducts of the developing mammary gland in vivo, which correlated with the degree of laminin deposition in these structures (Figs 5, 6). Myoepithelial cells are organized as a network that does not cover luminal cells completely (Gieniec and Davis, 2022), allowing for luminal cell contact with the BM in ducts, which could be linked to the overall higher levels of Pten observed in ducts in comparison to levels in TEBs. Interestingly, cells surrounding the nascent lumen in TEBs exhibited apicolateral Pten staining, the same pattern found in differentiated luminal cells in the ducts (Figs 5, 6). These data might indicate that in TEBs, cells surrounding the lumen already show some degree of polarization and maybe even differentiation, whereas cells in other areas do not. It has been shown that even though the TEB is a highly proliferative structure within the developing mammary epithelia, proliferation is not observed in luminal epithelial cells surrounding the nascent lumen (Ewald et al., 2012), and our data indicate that Pten might have a role in this phenomenon. These cells near the nascent lumen in TEBs clearly are not in contact with the BM; thus, other cues, such as signalling through cell–cell adhesions, might be involved in their differentiation programme (Runswick et al., 2001). Based on these observations, it is tempting to speculate that mammary gland differentiation involves two parallel and converging events: one occurring at the luminal interface and one coming from the basal surface, both involving Pten upregulation, with Pten polarization to the apicolateral region possibly being involved in commitment to the luminal lineage.

Pten in the apicolateral membrane, either in ducts or in cells surrounding the lumen in TEBs, strongly colocalized with the actin belt (Fig. 6), resembling the staining of other classical apicolateral markers of polarized cells, such as ZO-1 and E-cadherin (Laprise et al., 2002; Weaver et al., 1997). Whether cell differentiation precedes or follows Pten apicolateral polarization remains to be experimentally demonstrated. Nevertheless, Pten most likely acts on the onset of polarity acquisition rather than being a late factor recruited to mature junctional complexes.

It has been shown that Par3 (also known as PARD3), which has a core role in the establishment of cell polarity, and E-cadherin, a key regulator of epithelial tissue integrity, are both able to recruit Pten to cell–cell junctions, which is required for cell differentiation (Feng et al., 2008; Fournier et al., 2009). Polarization of Pten to the apicolateral membrane might contribute for the establishment of lipid domains (Comer and Parent, 2007; Pinal et al., 2006) and for the compartmentalization of PI3K activity in quiescent, differentiated cells (Fournier et al., 2009). In fact, PIP₂, the lipid product of Pten phosphatase activity, has been proposed to be an apical marker itself (Comer and Parent, 2007; Martin-Belmonte et al., 2007). EpH4 acini display active PI3K polarized to the basolateral membrane, and activation of the PI3K–Rac1 axis is required for complete differentiation and milk protein synthesis (Xu et al., 2009, 2010), thus, polarized activity of PI3K to the basal surface might trigger survival and differentiation, rather than proliferation.

Our data support a role for Pten not only in proliferation inhibition, but also in lumen assembly and maintenance in the developing mammary gland, and they are consistent with previous findings (Berglund et al., 2013; Martin-Belmonte et al., 2007). Apicolateral Pten has been reported in several 3D polarized epithelial structures, such as Madin–Darby canine kidney (MDCK) cell and T84 (colon) cell cysts (Martin-Belmonte et al., 2007), and the chick epiblast (Leslie et al., 2007), as well as in human and murine mammary epithelial cells (Berglund et al., 2013; Fournier et al., 2008); however, to our knowledge, the results we present here are the first report of apicolateral Pten in the developing mammary gland in vivo.

Facing these results, two old but not fully understood questions come to mind. What comes first: polarity or cell cycle arrest? And at what point does Pten come into play? Cell polarity and cell cycle arrest are interrelated, PI3K-modulated, but molecularly distinct events, and inhibition of the PI3K downstream effectors Akt and Rac1 has been linked to proliferation arrest and polarity, respectively (Liu et al., 2004). Similarly, by using 3D acini, Leslie and collaborators (Berglund et al., 2013) have shown that Pten requires both lipid phosphatase and protein phosphatase activities to control lumen formation through a mechanism that does not correlate with its ability to control Akt. Interestingly, in murine NMuMG acini, despite displaying increased cell proliferation, Pten knockdown or overexpression of p110α (PIK3CA) oncogenic mutants disrupts lumen formation, whereas a constitutively activated Akt does not (Berglund et al., 2013). These data further support the notion that different pathways downstream of PI3K govern the regulation of lumen formation and proliferation, and that Pten might be a key regulator of both. In this sense, our data in 3D acini models provide further evidence that Pten simultaneously coordinates proliferation and normal tissue architecture during acini development, and sustains quiescence and tissue architecture in mature, quiescent acini (Fig. 8).

Taken together, the data we present here provide evidence that laminin, found in the BM, induces quiescence and differentiation in mammary epithelial cells by differentially modulating the expression and activity of cell cycle regulators at the transcriptional and protein levels. This panorama, along with the extensive previous work done by many others, provides some hints of the molecules, pathways and mechanisms involved in microenvironmental control of quiescence in higher organisms, which could be explored in studies to come. The notion that lrECM-triggered quiescence involves several pathways and has many layers of regulation provides an explanation for the robustness of this system in suppressing epithelial cell proliferation, which could ultimately be linked to cancer incidence being lower than expected (Bissell and Hines, 2011) and even for the need of more than one perturbation to bypass microenvironmentally induced suppression of cancer development (Knudson, 2001). In this sense, disruptions in master regulators such as Pten, which regulate several signalling pathways and phenotypical features required for tissue homeostasis and tumour suppression, could indeed be drivers of malignant transformation.

Primary and secondary antibodies used in this work are listed in Table S1 and Table S2, respectively. Table S3 shows information on the pharmacological inhibitors used. Additional reagents used are listed in Table S4 and plasmids are listed in Table S6.

Murine mammary epithelial cells EpH4 were a gift from Mina J. Bissell (Lawrence Berkeley National Laboratory, Berkeley, CA, USA) and were routinely tested for mycoplasma contamination and verified to express specific markers for mammary epithelial cells. The parental EpH4 and the EpH4-ES-FUCCI and EpH4-DHB-mVen-mCh-Cdk4.KTR-H2B-mTurq (expressing DHB–mVenus, mCherry–Cdk4KTR and histone H2B–mTurquoise) sublines were routinely grown in DMEM/F12 containing 2% FBS, 5 µg/ml insulin and 50 µg/ml gentamycin at 37°C and 5% CO₂ in a humidified incubator. Cells were passaged when necessary (every 2–3 days) using trypsin-EDTA (0.05%). The cell sublines generated in this work can be provided upon request.

EpH4 cell quiescence induction and lactogenesis differentiation upon overlay treatment with lrECM has been described previously (Bridgewater et al., 2017; Spencer et al., 2011). Briefly, cells were seeded at 2000 cells/cm² (unless otherwise specified) in uncoated plastic wells or culture dishes. The following day (day 2), the medium was changed for GIH medium (DMEM/F12, 50 µg/ml gentamycin, 5 µg/ml insulin, 1 µg/ml hydrocortisone). Then, 24 h later (day 3), cells were treated or not with 2% growth factor-reduced Matrigel (lrECM) and/or 3 µg/ml prolactin (Prl and lrECM+Prl conditions), and where appropriate, with specific inhibitors. Cells were harvested or imaged 24–48 h post treatment (day 4 or day 5, depending on the assay). Fig. 1A depicts a general workflow for the lrECM overlay model used here.

EpH4 cells were subjected or not to lrECM overlay treatment for 48 h. Cells were then detached and pelleted by centrifugation, washed with cold PBS, fixed in 70% cold ethanol and stored for 24 h at −20°C. Fixed cells were incubated with RNase A (0.2 mg/ml) at 37°C for 30 min and stained with propidium iodide (PI) (1 μg/ml). DNA content of cells (20,000 events) was analysed using an ACCURI C6 flow cytometer. Data were plotted using GraphPad Prism, and statistical differences between treatments were assessed by two-tailed unpaired Student's t-test. P<0.05 was considered significant.

EpH4 cells were subjected or not to lrECM overlay treatment for 48 h. Cells were then stained with the DNA-binding dyes PI (5 μg/ml) and Hoechst 33342 (5 μg/ml) for 15 min. Cells were then imaged on a customized TissueFAXS i-Fluo system (TissueGnostics) mounted on a Zeiss AxioObserver 7 microscope (Zeiss), using a 20× Plan-Neofluar (NA 0.5) objective and an ORCA Flash 4.0 v3 camera (Hamamatsu). 8×8 adjacent microscope fields of view were acquired per well using automated autofocus and image acquisition settings. Thousands of cells per sample were analysed using a customized pipeline on StrataQuest software (TissueGnostics), which can be made available upon request. Details of parameters used for nuclei segmentation and detection on StrataQuest were as described previously (Russo et al., 2021). Individual nuclei for all cells were detected in the Hoechst channel, which was used to build a nuclear mask where the PI signal (dead cells) was detected. The relative cell density (number of cells based on the Hoescht signal/total area imaged) was calculated for control (normalized to 1) or lrECM-treated cells. The percentage of dead cells (PI-positive cells) was calculated in relation to the total number of cells (Hoechst-positive cells). Data were plotted using GraphPad Prism and statistical differences between treatments were assessed by two-tailed unpaired Student's t-test. P<0.05 was considered significant.

For these experiments, EpH4 cells constitutively expressing the fluorescence cell cycle indicator FUCCI (Sakaue-Sawano et al., 2008) were used. EpH4-ES-FUCCI cells were generated by transfecting EpH4 cells with the ES-FUCCI plasmid (Table S6), followed by hygromycin B selection (500 µg/ml, 10 days) and aseptic cell sorting of fluorescent cells.

EpH4-ES-FUCCI cells were seeded into 96-well plates and subjected to the lrECM overlay assay (control and lrECM). At 48 h post treatment, cells were incubated with Hoechst 33342 (5 μg/ml) for 15 min and imaged on a Leica DMi8 wide-field fluorescence microscope (E-signal lab, University of São Paulo) using the LasX Navigator Application (Leica Microsystems). The DMi8 microscope was coupled with an incubation system that allowed live-cell imaging at 37°C and 5% CO₂. Each well was imaged in 3×3 adjacent microscope fields, using the 10× objective and specific fluorescence filters (DAPI for Hoescht, Txr for mCherry and YFP for mCitrine). Images were analysed on StrataQuest software (TissueGnostics) using a customized pipeline, which can be provided upon request. Details of parameters used for nuclei segmentation and detection on StrataQuest were as described previously (Russo et al., 2021). In brief, nuclei were detected in the Hoechst channel and a mask was created, allowing the measurement of the mean fluorescent intensity signal of the FUCCI probes (Cdt1–mCherry and geminin–mCitrine) for thousands of cells per sample. The percentage of mCherry-positive cells (cells in G1), mCitrine-positive cells or double-positive cells (cells in S/G2/M), or double-negative cells (cells that had just finished M) was determined for each well. Data were plotted using GraphPad Prism, and statistical differences between treatments for each cell cycle phase were assessed by two-tailed unpaired Student's t-test. P<0.05 was considered significant. These experiments were performed in triplicate wells in three biological replicates.

Cdk2 and Cdk4 activities were evaluated using fluorescent reporters (Spencer et al., 2013; Yang et al., 2020). EpH4-DHB-mVen-mCh-Cdk4.KTR-H2B-mTurq cells were generated by simultaneous infection of EpH4 cells with lentiviruses containing either DHB-mVen-mCh-Cdk4.KTR or the H2B-mTurq constructs in the presence of 8 µg/ml polybrene. Lentiviruses were previously packaged by HEK293 FT cells (gift from Mina J. Bissell, Lawrence Berkeley National Laboratory, Berkeley, CA, USA) transfected with polyethylenimine (PEI), the packaging plasmids psPAX2 and pMD2.G and pLenti-DHB-mVenus-p2a-mCherry-Cdk4KTR or CSII-EF1-H2B-mTurquoise (Table S6), following instructions from Addgene. Cells were then subjected to aseptic sorting for selection of double-infected cells (expressing mVenus and mCherry, as well as mTurquoise).

EpH4-DHB-mVen-mCh-Cdk4.KTR-H2B-mTurq cells were plated into 96-well plates and subjected to lrECM overlay assay (control and lrECM). At 48 h post treatment, cells were imaged on a Leica DMi8 wide-field fluorescence microscope (E-signal lab, University of São Paulo) using the LasX Navigator Application (Leica Microsystems). Each well was imaged in 3×3 adjacent fields of view, using the 10× objective and specific fluorescence filters (CFP for mTurquoise, Txr for mCherry and YFP for mVenus). Cdk2 and Cdk4 activities were calculated based on the guidelines from Spencer et al. (2013) and Yang et al. (2020), the developers of these specific probes. Images were analysed on StrataQuest software (TissueGnostics) using a customized pipeline, also available upon request. Details of parameters used for nuclei segmentation and detection on StrataQuest were as described previously (Russo et al., 2021). In brief, nuclei were detected in the mTurq channel and a nuclear mask was created. In order to measure the cytoplasmic intensity of Cdk2 and Cdk4 probes, another mask, corresponding to a cytoplasmic ring was created (2 µm around the nucleus, or less if touching another cytoplasmic ring or nucleus, starting 0.5 µm away from the nucleus). The mean intensity of fluorescence signals (mVenus for Cdk2 and mCherry for Cdk4) were obtained for the nucleus and for the cytoplasmic ring for thousands of individual cells per sample. For each cell, Cdk2 activity was calculated as the ratio of the mean fluorescent intensity of mVenus in the cytoplasmic ring and in the nucleus. Because Cdk2 can to some extent phosphorylate the Cdk4 probe (Yang et al., 2020), Cdk4 activity was calculated as the cytoplasmic:nuclear ratio of the mean fluorescent intensity of mCherry, minus the Cdk2 activity in that cell, multiplied by a correction factor that was previously calculated by Yang et al. (2020): Cdk4 activity=[(mCherry cytoplasmic/mCherry nuclear) – Cdk2 activity]×0.05. Data were plotted as a ‘Superplot’ (Lord et al., 2020) using GraphPad Prism, and statistical differences between treatments were assessed by two-tailed unpaired Student's t-test. P<0.05 was considered significant. These experiments were performed in triplicate wells in three biological replicates.

EpH4 cells were subjected to lrECM overlay assay for 48 h, and cell size was estimated according to the forward scatter area (FSC-A) determined using a BD Accuri C6 flow cytometer. As freely cycling cells (control) would have significantly more cells in S/G2/M (which are larger in size) than quiescent (lrECM-treated) cells, in order to do a fair comparison, both control and lrECM-treated cells were gated in G1 according to DNA content (assessed using PI staining, see above).

Similarly, EpH4-FUCCI cells were subjected to lrECM overlay assay for 48 h, counterstained with Hoechst 33342 (5 µg/ml, 15 min) and imaged (as described in the ‘FUCCI-based cell cycle analysis’ section above). Cell size was estimated according to the area of the nucleus (µm²). Images were analysed on StrataQuest software (TissueGnostics), and once again, only cells in G1 (mCherry positive) were considered for this analysis.

Data were plotted using GraphPad Prism, and statistical differences between treatments were assessed by two-tailed unpaired Student's t-test. P<0.05 was considered significant. These experiments were performed in three biological replicates.

For the analysis of the correlation between Cdk activity and nuclear area (Fig. S2), the data regarding Cdk4, Cdk2 and nuclear area were obtained from the experiments shown in Fig. 1G,H, plotted using the Python package Seaborn (version 0.12.2), and the Pearson's correlation r value was calculated using GraphPad Prism.

EpH4 cells were subjected to lrECM overlay and/or prolactin treatment for 48 h. Cells were then harvested for RNA extraction, reverse transcription and RT-qPCR. This experiment was performed in three biological replicates.

Gene expression was assessed using RT-qPCR, following the MIQE guidelines (Bustin et al., 2009). Briefly, total RNA was extracted using an RNeasy kit, quantified and treated with a Turbo DNAse kit according to the manufacturer's instructions. DNAse-treated RNA was reserve transcribed into cDNA using the Superscript II reverse transcriptase, oligoDT and random hexamer primers (Thermo Fisher Scientific), following the manufacturer's guidelines.

RT-qPCR was performed in triplicates using 5 ng of cDNA, Power SYBR Green PCR Master Mix and specific primers (listed in Table S5) into a 10 µl reaction. The reactions were run in the AB-7500 real-time thermocycler (Applied Biosystems) using standard settings. B2m was used as the endogenous control, and mRNA fold changes (FCs) were calculated according to the Pfaffl method (Pfaffl, 2001).

Statistical analysis was carried out on GraphPad Prism using one-way ANOVA followed by Dunnett's multiple comparison test (comparing all samples to the control; P<0.05 was considered significant). Data were plotted as log₂FC and –log₁₀P-value using the online software Morpheus (Broad Institute; https://software.broadinstitute.org/morpheus/).

EpH4 cells were subjected to lrECM overlay and/or prolactin treatment for 48 h. Cells were then lysed with boiling 1× Laemmli buffer (62.5 µM Tris, 2% SDS, 10% glycerol), and the obtained lysate was further incubated at 100°C for 15 min and stored at −20°C. Protein was quantified using the DC Protein Assay kit (Bio-Rad) according to the manufacturer's instructions.

The levels of specific proteins were then assessed by western blotting. For loading the gel, 1 µl of 0.15% Bromophenol Blue in 2-mercaptoethanol was added to 20 µg of protein in 1× Laemmli buffer for a total volume of 30 µl. Samples were boiled at 100°C for 10 min and run in on Tris–glycine polyacrylamide gels (8%, 10%, 12% or 15%, depending on the molecular mass of the protein of interest). Resolved proteins were transferred overnight onto a PVDF membrane (0.22 µm; Millipore) followed by 30 min incubation in blocking buffer (TBS, 0.2% Tween-20 with 5% BSA). Membranes were incubated overnight in blocking buffer containing primary antibodies (listed in Table S1), followed by incubation with the appropriate HRP-conjugated second antibody (listed in Table S2) for 1 h at room temperature. HRP was detected by SuperSignal West Dura Extended Duration Substrate (Thermo Fisher Scientific) and the chemiluminescence signal was captured with a ChemiDoc MP Imaging System (Bio-Rad).

When necessary, membranes were subjected to a stripping protocol to remove antibodies and allow the reprobing with antibodies against another protein (this was used particularly to assess the total levels of a given protein after assessing its phosphorylated levels). Stripping consisted of 2×10 min in distilled water, followed by 1 min in stripping buffer (25 mM glycine, 1% SDS, pH 2) and 4×5 min in TBS containing 0.2% Tween-20. Membranes were then blocked for 30 min and subjected to standard primary antibody and HRP-conjugated secondary antibody incubation and detection.

For the relative quantification of protein levels, the optical density of each band was determined using ImageJ (NIH, Bethesda, MD, USA). The endogenous control (lamin B or α-tubulin, depending on the experiment) was also quantified, and the relative expression was determined as the ratio between target protein level and the endogenous control. For phosphorylated proteins, we used the ratio of phosphorylated protein to total protein to endogenous control. Phospho-Rb and phospho-PI3K (p55) were normalized only against the endogenous control, as we could not assess the total levels of these proteins.

Data were analysed using GraphPad Prism. Fold change differences of a given protein between two groups were evaluated by two-tailed unpaired Student's t-test (P<0.05 was considered significant). The obtained data were compiled and displayed as volcano plots (log₂FC versus –log₁₀P-value). Comparisons between multiple groups were performed by one-way ANOVA followed by Tukey post-hoc test (P<0.05 was considered significant). This experiment has at least three biological replicates for each protein and condition analysed. Images of uncropped membranes can be found in Figs S4 and S5.

EpH4-ES-FUCCI cells were seeded into 96-well plates (Corning) at a density of 2000 cells/well. The following day, the medium was changed to GIH without Phenol Red. Then, 24 h later, cells were treated or not with 2% Matrigel and/or specific inhibitors (see Table S3 for details and concentrations). At 24 h post treatment, cells were incubated with Hoechst 33342, imaged and analysed as described in the ‘FUCCI-based cell cycle analysis’ section above. The relative cell density (as described in the ‘Cell density and live–dead assay’ section above) was also determined for each well.

Data were analysed using GraphPad Prism. Relative cell density (normalized to the density of control or lrECM untreated cells) was analysed by one-way ANOVA followed by multiple comparisons (all conditions versus control) using the two-stage linear step-up procedure of Benjamini, Krieger and Yekutieli (q<0.05 was considered significant). Differences in cell cycle based on the FUCCI probe expression were assessed by two-way ANOVA followed by multiple comparisons (all conditions versus control, for G1, S/G2/M and just finished M/early G1) using the two-stage linear step-up procedure of Benjamini, Krieger and Yekutieli (q<0.05 was considered significant). These experiments were performed at least two times with two technical replicates (two wells/condition).

EpH4 cells were subjected to lrECM overlay assay with or without concomitant treatment with 2.5 µM bpV(pic) Pten inhibitor [control, Pten inhibitor (Pteni), lrECM and lrECM+Pteni]. At 24 h post treatment, cells were harvested for protein extraction and western blot analysis of protein expression. Data were analysed on GraphPad Prism using one-way ANOVA followed by Tukey's multiple comparisons test. These experiments were performed at least three times.

The mammary fat pads were harvested from 5-week-old female Balb/C mice obtained from Rede de Bioterios – University of São Paulo, and the experimental procedures involved in this work were approved by the Ethics Committee for Animal Welfare from the Institute of Chemistry, University of São Paulo (ethics number #218/2022).

Mammary fat pads were embedded in Tissue-Tek O. C. T., snap-frozen and stored at −80°C. The samples were sliced at a thickness of 25 µm using a cryostat (Leica). Slices were collected in StarFrost slides, dried briefly and were stored at −80°C until used.

For immunofluorescence staining, slices were thawed, fixed in 4% paraformaldehyde (PFA) for 10 min and quenched in PBS containing 0.3 M glycine. Antigen retrieval was performed by applying boiling Tris-EDTA (10 mM Tris, 1 mM EDTA, pH 9) directly onto the slides thrice with 5 min interval. Slices were then permeabilized in PBS containing 0.5% Triton X-100 (PBS-T) for 20 min, blocked in undiluted Odyssey Blocking buffer (LI-COR) containing 10% goat serum for 1 h at room temperature and incubated with primary antibodies (anti-Pten, anti-laminin-111; Table S1) overnight at 4°C. The following day, slides were washed and incubated with Alexa Fluor-conjugated secondary antibodies (Table S2) and, when pertinent, with Phalloidin–Alexa Fluor 647 (1:50) for 45 min at room temperature. Slides were counterstained with DAPI (0.5 µg/ml), mounted with Prolong Diamond antifade reagent, sealed and left overnight at 4°C to cure.

Epithelial structures of interest (TEBs and ducts) were either imaged with a Leica DMi8 wide-field fluorescence microscope (E-signal lab, University of São Paulo) or a Zeiss LSM880 Airyscan (INFABIC – National Institute of Science and Technology on Photonics Applied to Cell Biology at the State University of Campinas). Several ducts and TEBs from mammary glands obtained from at least three mice were used for these analyses.

Images obtained with the Leica microscope were taken using the 63×/1.4NA oil objective in z-stacks and adjacent fields (to cover larger structures comprising whole ducts and/or TEBs) using the LasX Navigator Application (Leica microsystems) in three channels (DAPI for nuclei, FITC for Pten and TXR for laminin-111). Images were initially processed in the LasX software, where adjacent fields were merged and 3D-blind deconvoluted. Five sequential z-stacks were then combined to obtain the maximal projection. The resulting images were used to analyse the association between laminin content and Pten levels in ducts and TEBs using a customized pipeline on StrataQuest software (TissueGnostics). Details of parameters used for nuclei segmentation and detection on StrataQuest were as described previously (Russo et al., 2021), and for this specific analysis, nuclei detection and segmentation were manually corrected for each image. In brief, images of ducts or TEBs were analysed separately; nuclei were detected in the DAPI channel, and the mean intensity of Pten signal was obtained for each cell in an area comprised by the nucleus and the cytoplasm, which was estimated by growing outward from the nuclear area until neighbouring areas touched each other or until the fluorescence signal of Pten reached background levels (a similar approach to estimate cell boundaries was used for ADP ribose signal by Russo et al., 2021). Total laminin-111 fluorescence intensity (sum of intensity) was obtained for the entire field of interest (epithelial structure) and this value was divided by the total number of cells detected in that field (mean laminin-111 fluorescence). The average of the mean fluorescence intensity of Pten was plotted against the associated mean laminin-111 intensity for that field using GraphPad Prism, and data were analysed by linear regression and Pearson's correlation. Differences between Pten intensity in ducts versus TEBs were analysed by two-tailed unpaired Student's t-test.

Images obtained using the Zeiss LSM880 Airyscan were taken using the 63×/1.4NA oil objective, and fluorescence was captured in four channels using the following laser lines for excitation: 405 nm (DAPI), 488 nm (Pten, Alexa Fluor 488), 543 nm (laminin-111, Alexa Fluor 546) and 633 nm (Phalloidin, Alexa Fluor 647). The detection spectra were adjusted manually for each channel to avoid crosstalk between the different channels. Both confocal and zoomed (3.4 zoom factor) super-resolution (Airyscan) images were obtained for ducts and TEBs at different focal planes, corresponding to the basal domain (closer to the laminin-rich BM) and apical domain (closer to the lumen).

Further analyses regarding to the fluorescence levels of Pten in different regions of the epithelial structure (Fig. 6C,D) were performed on ImageJ using confocal images from three different mice (one duct and one TEB per mouse). Briefly, regions of interest (ROIs) corresponding to luminal or basal cells in ducts and to cells surrounding the nascent lumen, body cells or basal (cap) cells in TEBs were manually assigned to ducts and TEBs, based on the position of the BM or lumen in these structures. Initially, the mean fluorescence intensity for each ROI was obtained, and the mean intensity of all ROIs for the same region was calculated for each mouse. To calculate the cytoplasmic/nuclear ratio, we used the DAPI channel to build a mask that was used to obtain the mean fluorescence intensity of Pten in the nucleus and outside the nucleus (cytoplasm). We then calculated the cytoplasmic:nuclear ratio for each ROI and the average for each mouse. Data were plotted in GraphPad Prism. Data from luminal versus basal cells in ducts were compared by two-tailed paired Student's t-test (P<0.05 considered significant). In TEBs, comparisons between cells near the nascent lumen versus body cells versus basal cells were conducted by repeated measures one-way ANOVA followed by Tukey's post-hoc test (P<0.05 was considered significant).

EpH4-ES-FUCCI cells (5×10³ cells/cm²) were plated into 24-well plates on top of a bedding of Matrigel as previously described (Mroue and Bissell, 2012). Wells were treated with 2.5 µM Pten inhibitor bpV(pic), either on the day of seeding (day 0 and again at day 2 and day 4) or when quiescent 3D structures were already established (day 4). On day 5, 3D structures were incubated with Hoechst 33342 and imaged in z stacks using a 10× objective in a fluorescence microscope (Leica DMI8 wide-field microscope; imaging mCherry, mCitrine and Hoechst). A general workflow for the ‘on top’ 3D model is shown in Fig. 7A. Representative structures were then imaged at 40×. Image analysis was conducted using the LASX application (Leica). Briefly, images were deconvoluted and the maximum-intensity projection image was obtained. Structure size and roundness (0–1, where 1 represents a perfect circle) were obtained for individual structures, based on a mask built using the Hoechst channel. FUCCI-based cell cycle was calculated according to the mCherry and mCitrine areas in each field. These experiments were performed three times in duplicate wells, and three microscope fields per well at 10× were used for image analysis. Data were plotted in GraphPad Prism and analysed by one-way ANOVA followed by Dunnett's multiple comparisons test between all groups (Control, Pten inhibitor at day 0 and Pten inhibitor at day 4).

EpH4 cells were seeded in low-adherence polyHEMA-coated plates (30,000 cells/cm²), incubated for 48 h to allow for cell clusters to be formed and replated in medium supplemented or not with 4% Matrigel. Plates were coated in-house with polyHEMA at 0.25 mg/cm² (Xu et al., 2009). When indicated, cell clusters were treated with 2.5 µM Pten inhibitor bpV(pic) either on the day of seeding (day 0) or when mature acini were already present (day 5). On day 6, 3D structures were fixed and subjected to immunofluorescence staining for ZO-1 (apical marker), α6 integrin (basal marker), Ki-67 (proliferation marker) and counterstained with DAPI (nuclei). A general workflow for the 3D lrECM in polyHEMA plates is shown in Fig. 7F. Briefly, structures were fixed for 10 min by adding 32% PFA straight into the culture well (for a final concentration of 4%). Structures were then collected into microcentrifuge tubes, spun, resuspended and washed twice in PBS containing 0.3 M glycine (quenching), permeabilized with 0.5% Triton X-100 in PBS for 20 min and blocked in 10% goat serum in Odyssey blocking buffer. Structures were then incubated overnight with primary antibodies diluted in Odyssey blocking buffer, washed three times with PBS and incubated with Alexa Fluor-conjugated secondary antibodies for 45 min (details of primary and secondary antibodies can be found in Tables S1 and S2), washed twice in PBS, incubated with DAPI for 5 min and mounted on glass slides using Prolong Diamond. All centrifugation steps were performed at 200 g for 3 min, and all tubes and tips used were previously rinsed with 5% BSA in PBS. These experiments were performed in two biological replicates.

Proliferation was estimated by the ratio between the area of Ki-67-positive cells and total area (DAPI). Slides were scanned with a Leica DMi8 wide-field microscope at 10×/0.3 NA (200 fields per slide), using the LASX Navigator application (Leica). Image analysis was conducted using the LASX application (Leica), where the total area for Ki-67 and DAPI were obtained and used to calculate the ratio between Ki-67 (proliferative cells) and DAPI (all cells) for a given slide.

The presence of lumen was assessed by manually scanning and counting the number of structures with lumen and the total number structures, using a 40×/0.6 NA objective with a Leica DMi8 wide-field microscope. The percentage of structures with lumen was then calculated for each condition. At least 600 structures per slide were analysed. The data regarding proliferation and lumen were then plotted and analysed in GraphPad Prism using one-way ANOVA followed by Dunnett's multiple comparisons test.

Representative structures were imaged with a Leica SP8-STED FALCON microscope using the confocal mode, a 63×/1.4 NA oil objective and using the following laser lines for excitation: 405 nm (DAPI), 488 nm (Ki67, Alexa Fluor 488), 561 nm (integrin α6, Alexa Fluor 546), 633 nm (ZO-1, Alexa Fluor 647). The detection range for each emission spectrum was adjusted manually to prevent signal overlap.

The authors thank Hugo Armelin (Institute of Chemistry, University of São Paulo, Brazil), Patricia Gama (Institute of Biological Sciences, University of São Paulo, Brazil), William Festuccia (Institute of Biological Sciences, University of São Paulo, Brazil), Cyrus Ghajar (Fred Hutchinson Cancer Center, USA) and Sabrina Spencer (University of Colorado Boulder, USA) for the kind donation of antibodies, reagents and plasmids. We thank the National Institute of Science and Technology on Photonics Applied to Cell Biology (INFABIC) at the State University of Campinas for access to equipment and assistance; INFABIC is co-funded by FAPESP (#2014/50938-8) and CNPq (#465699/2014-6). The authors are also grateful to Nicolas Hoch and Deborah Schechtman for scientific discussions, advice on image analysis and insightful inputs.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.261178.reviewer-comments.pdf