TGF-β regulates Sca-1 expression and plasticity of pre-neoplastic mammary epithelial stem cells

The epithelial-mesenchymal plasticity, in tight association with stemness, contributes to the mammary gland homeostasis, evolution of early neoplastic lesions and cancer dissemination. Focused on cell surfaceome, we used mouse models of pre-neoplastic mammary epithelial and cancer stem cells to reveal the connection between cell surface markers and distinct cell phenotypes. We mechanistically dissected the TGF-β family-driven regulation of Sca-1, one of the most commonly used adult stem cell markers. We further provided evidence that TGF-β disrupts the lineage commitment and promotes the accumulation of tumor-initiating cells in pre-neoplastic cells.

Animal experiments. The colony of severe combined immunodeficient animals (Crl:SHO-Prkdc scid Hr Hr ) was acquired from Charles River (Sulzfeld, Germany) and maintained according to the ARRIVE guidelines. For orthotopic mammary fat pad injections, suspension of 7.5 × 10 5 (Comma-Dβ), 5 × 10 4 (ANV2), or 2.5 × 10 5 (MMC) cells in 50 μL PBS were injected into fourth mammary gland of 6 weeks old females. Tumor dimensions were measured with the calibrated digital caliper (VWR, Radnor, PA, United States). Mice were sacrificed when the tumor size reached the end-point specified in the animal protocol, the date was recorded and use for Mantel-Cox logrank analysis (overall survival). Experiments were terminated twelve weeks after tumor cell injection. For implantations into cleared mammary fat pad, 1 × 10 5 cells in 50 μL PBS were implanted into fourth mammary gland of 6 weeks old females, with mammary fat pad cleared at the age of 3 weeks using sparing procedure, as described previously 15 . Animal experiments were approved by the Academy of Sciences of the Czech Republic (AVCR 2015/13); supervised by the local ethical committee and performed by the certified individuals (JR, OV, KS).
Antibody-based cell surface screening, and data analysis. The high-throughput surface profiling and data analysis were performed essentially as described in 16 with minor modifications specified below, as different cell lines and kit components were used in this study. Epithelial cell lines were incubated in 1.35 mM EDTA solution in PBS prior to trypsinization to allow the non-enzymatic weakening of the cell junctions prior to trypsinization (10 min for MMC and cE2 cells). Mesenchymal cell lines (ANV2, E2, RM-1 and UGSM-2 cells) were only briefly washed with EDTA solution, as prolonged incubation would lead to complete cell detachment. Cell lines were then harvested with mild 0.05% trypsin/0.02% EDTA solution (5 min at 37 °C, GE Healthcare, Little Chalfont, United Kingdom, cat. L15-004). Next, the cell suspensions were processed as described previously, except that 4.2 × 10 7 cells from each, re-counted cell line were then pooled. The pool of cells was washed with PBS and stained with LIVE/DEAD Green Fixable Dead Cell Stain diluted 1:1,000 in PBS for 15 min at 4 °C (1 mL staining solution per 1 × 10 7 cells; Molecular Probes, TSF). Next, cells were washed, resuspended in 25.2 mL Cell Stain buffer and filtered via 70 μm cell strainer to remove large cell aggregates. Filtered cell suspension (75 μL/well, equal to pool of 0.75 × 10 6 cells/well) was dispensed into LEGENDScreen Mouse Phycoerythrin (PE) Kit 96 well plates (cat. 700005; Biolegend, San Diego, California, United States), each well was already containing 25 μL of single, validated and pre-titrated antibody conjugated with PE. Plates were reconstituted with deionized water, and cells were then stained and fixed as recommended by the manufacturer. Data acquisition and analysis were described previously. Since we further focused on mammary gland biology, results from RM-1 17 , UGSM-2 18 , cE2 and E2 19 prostate-derived cell lines were not presented in this manuscript. However, they are included in the Supplementary Table 1  Instrumentation, flow cytometry analysis, and cell sorting. Flow cytometry was performed as described in 16 Smad6 Ly6a * * *  Analysis of cell cycle, EdU incorporation assay, apoptosis, and cell death. For multiparametric analysis of surface marker expression and proliferation, cells were treated with 10 μM EdU two hours prior to harvesting. Cells were then trypsinized, stained for surface markers, fixed in 4% formaldehyde, permeabilized with 0.15% Triton X-100. Click-iT reaction was performed with Click-iT EdU Alexa Fluor 488 Flow Cytometry Kit, and cell cycle was analyzed with FxCycle Violet (Molecular Probes, TFS) according to manufacturer's recommendation. For analysis of cell death, cells (including the floating fraction) were collected, stained for surface Sca-1 as described above, followed by staining with Annexin V-FITC conjugate (#ANXV-FT100, Apronex, Prague, Czech Republic) in annexin V staining buffer, as recommended. Propidium iodide was added to the suspension one minute before analysis. Analysis was performed on SP-6800 Spectral Analyzer.
RNA isolation, reverse transcription, and qPCR analysis. Total RNA from either cell culture or freshly sorted cells was extracted using High Pure RNA Isolation Kit (Roche) and quantified on BioSpectrometer (Eppendorf, Hamburg, Germany). cDNA synthesis was performed with 0.2-2 μg of total RNA using High-Capacity RNA-to-cDNA Kit (Applied Biosystems, TFS). mRNA levels were measured with gene-specific primers using Roche LightCycler 480 master mix, probes and thermocycler system (Roche, Basel, Switzerland) and relative expression levels were normalized to the reference gene Tbp. Gene expression assays used for qPCR analysis are listed in the Supplementary Table 4.

Sca-1 is differentially expressed on the cell surface of mammary cancer stem cells.
Acquisition of cancer stem cell phenotype is often accompanied by a dramatic and complex remodeling of the cell membrane and other cellular components. To gain deeper insight into the changes of cell surface antigens accompanying such epithelial-mesenchymal plasticity, we performed multiplexed, high-throughput cell surface profiling (Fig. S1A-B and Supplementary Table 1 and 2, for technical details see 16 ). We identified nine surface antigens that were upregulated on the cell surface of ANV2 cells (Fig. 1A). These HER2 antigen-negative variants (ANV) are exhibiting stem-like properties and were derived from spontaneously relapsed and immune-edited tumors after injection of purely epithelial murine mammary cancer cells (MMC) into the syngeneic hosts (Fig. S2A-E and 13,22 ). Stem cell antigen-1 (Sca-1 or Ly6A/E) was present exclusively on the surface of ANV2 cells. In order to verify the observed differences in the cytokine-induced model of EMT, we exposed MMC and ANV2 cells to several TGF-β family ligands. To our surprise, TGF-β1 mediated down-regulation of Sca-1 expression in ANV2 cells (Fig. 1B). In our experimental settings, Sca-1 marked mesenchymal cancer cells with stem-like properties and its expression was down-regulated in the presence of TGF-β1. Despite that Sca-1 is one of the most commonly used markers for adult murine stem cells, its contribution to stemness is not yet understood in many tissue types. We overexpressed Sca-1 in epithelial MMC cells that do not display stem-like properties and assessed their phenotype and behavior in vitro and in vivo. Cells overexpressing ectopic Sca-1 did not demonstrate increased ABC transporter or ALDH activity, elevated spheroid formation capacity under standard culture conditions ( Fig. 2A-E), or enhanced tumor growth (Fig. 2F). Sca-1 itself is thus not sufficient to induce stem cell phenotype in mammary epithelial cancer cells.

TGF-β affects the differentiation state of mammary epithelial cells.
We explored the effect of TGF-β1-mediated Sca-1 down-regulation in the context of pre-neoplastic mammary epithelial cells. The Comma-Dβ cell line is derived from the normal mammary gland of mid-pregnant mice and serves as a pre- The plot shows changes in gene expression of Sca-1 mRNA (Ly6a), regulator of basal phenotype Snai2, and regulator of luminal phenotype Gata3 in sorted subpopulations of Comma-Dβ cells, exposed to exposed to vehicle (PBS) or TGF-β1 (1 ng/mL) for 96 h directly after sorting. Results are presented as mean ± SD (paired t test or ratio-paired t test, *P < 0.05). (D) The plot shows single mouse tracks for tumor growth of Comma-Dβ cells, measured with calibrated digital caliper. Results are from two independent experiments, n = 9 mice per group. Cells were pretreated prior to implantation with vehicle or 1 ng/mL TGF-β1 for 72 h. (E) The plot shows % of mice without relapsing tumors and is related to panel (D; Mantel-Cox test). (F) Representative images showing tumor morphology (H&E) and Cytokeratin 5 staining in control and TGF-β1-pre-treated tumors. Refer to Supplementary Fig. S6D for a full panel of stainings and additional controls (scale = 200 μm). (G) Plot shows quantification of tumorigenic potential of non-sorted Comma-Dβ cells exposed to vehicle or TGF-β1 (1 ng/ mL) for 72 h prior to implantation, and of Sca-1 − and Sca-1 + Comma-Dβ cells sorted from vehicle-or TGF-β1treated Comma-Dβ cell line. Number or relapsing tumors is shown above corresponding error bars. Data were collected at week 8 post-implantation in two independent experiments and are visualized as mean ± SEM (n = 6 per group, Mann-Whitney u test). www.nature.com/scientificreports/ neoplastic cell line model for studying mammary gland plasticity 23,24 . Comma-Dβ cells are known for their heterogeneous expression of Sca-1: Sca-1 + subpopulation is enriched in mammary progenitors 23 . We first extensively characterized both the Sca-1 − and Sca-1 + subpopulations of these cells, confirming that Sca-1 − fraction resembled the luminal-like mammary epithelial cells, while the Sca-1 + fraction showed increased expression of basal-like markers (Snai2, Twist2, Cd49f; Fig. 3A and S3A-B). Interestingly, the other members of the LY6 family -Ly6c1, Ly6c2, Ly6e-showed similar expression pattern to that of Sca-1 (Fig. S3A-D), suggesting their potential in co-maintenance of the progenitor functions and presence of multiple other subpopulations. The Comma-Dβ cells responded to TGF-β1 by almost complete surface ablation of Sca-1, in a dose-dependent manner ( Fig. 3B and S4A). We analyzed the TGF-β-induced expression changes in both subpopulations, as well as the effect of TGF-β on cell death and proliferation. Despite that TGF-β1-mediated loss of Sca-1 was not accompanied by the epithelial-to-mesenchymal transition ( Fig. 3C and S3E), it resulted in a down-regulation of transcription factors Slug/Snai2 25 and Gata3 26 , responsible for lineage commitment (Fig. 3D). Similar to TGF-β1, the bone morphogenetic proteins BMP4 and BMP7 also efficiently repressed Sca-1 expression (Fig. S4B-C). Pharmacologic inhibition of TGF-β, but not that of BMP, led to the increase in Sca-1 expression, independently of a pharmacophore (Fig. S4D-E). Simultaneous kinetic analysis of cell surface Sca-1, DNA synthesis and cell cycle progression further revealed that the complete loss of surface Sca-1 in response to TGF-β1 appeared after 72 h (Fig. S5A-B). The Sca-1 − and Sca-1 + subpopulations had distinct cell cycle profiles even without perturbation (while 44% of Sca-1 − cell is in G1/0 and 51% in G2/M; 61% of Sca-1 + cell is in G1/0 and 36% in G2/M). Modulation of the TGF-β pathway did not alter the cell cycle pattern and neither did induce apoptosis (Fig. S5C-D). Based on these results we conclude that the observed TGF-β1-induced phenotypic changes in Comma-Dβ cells are not the result of aberrant proliferation or induction of cell death.
The short-term exposure of pre-neoplastic mammary epithelial cells to TGF-β enriches for tumor-initiating cells. The TGF-β1-induced perturbations in lineage-specific genes suggested that the TGF-β signaling affects the differentiation state of Comma-Dβ cells. The unexpected, concomitant decrease of master basal-(Snai2 and Sox9) and luminal-lineage regulators (Gata3, see Fig. 2D) led us to further dissect the phenotypic changes in both cell lineages. We sorted the Sca-1 −/+ fractions and exposed them to TGF-β1 ( Fig. S6A-C). As expected, the Sca-1 and Slug mRNAs were down-regulated in sorted basal-like Sca-1 + fractions, while the Gata3 mRNA was decreased in sorted luminal-like Sca-1 − fraction ( Fig. 4A-C).
In order to determine the contribution of short-term TGF-β1 exposure to the tumorigenicity of these preneoplastic cells, we exposed them to TGF-β1 prior to implantation into the mammary fat pad. Most of the tumors showed 4-6 weeks latency and grew rather slowly. However, a portion of tumors derived from TGF-β1 pre-treated cells (4/9) escaped this latency and formed relapsing tumors (Fig. 4D-E). These relapsed tumors were typical by high cell density, weak Cytokeratin 5 immunoreactivity 27 and low content of collagen-containing extracellular matrix ( Fig. 4F and S6D). Further sorting and in vivo implantation of TGF-β1 pre-treated cells showed, that the Sca-1-fraction of Comma-Dβ cells contained tumorigenic clones, and this was further enhanced with TGF-β1 (Fig. 4G). On the contrary, pre-treatment of bona fide cancer-stem like cell line ANV2 with TGF-β1 resulted in delayed tumor growth, reduced tumorigenicity (100% in vehicle vs. 87% in TGF-β1 pre-treated group) and prolonged overall survival (Fig. S5E-F). These observations confirm that TGF-β can be promoting tumor growth in pre-neoplastic cells while being tumor suppressive in cancer cells with functional TGF-β signaling. Since it was uncertain whether the endogenous mammary epithelium itself supports the sudden growth of relapsing tumors, we injected the Comma-Dβ cells into the cleared mammary gland. Both the vehicle and TGF-β1-pretreated cells showed similar tumorigenic potential in the cleared mammary fat pad (Fig. S6G-H). However, the TGF-β1-pre-treated cells failed to form relapsing tumors in the lack of endogenous epithelium. The endogenous epithelium was thus necessary for this relapsing behavior of luminal-like Sca-1 − cells. The de-differentiation of pre-neoplastic mammary epithelial cells in response to TGF-β1 is accompanied by the enrichment or preselection of cells with increased tumor-initiating capacity.
Sca-1 is the target gene of the TGF-β signaling pathway. We further evaluated the molecular basis of Sca-1 repression through TGF-β. Pre-treatment of cells with galunisertib, an ALK5/TGFBR1 inhibitor, before the addition of TGF-β1 rescued Sca-1 repression (Fig. 5A and S7A). This indicated an ALK5-dependent mecha- www.nature.com/scientificreports/ nism. Silencing of ALK5 downstream signal transducers Smad2, Smad3, and Smad4 resulted in an increased number of Sca-1 + cells ( Fig. 5B and S7B). However, both Smad2 and Smad3 were dispensable for Sca-1 repression upon the exposure to TGF-β1 (Fig. 5C). Interestingly, while the inhibition of ALK5 did not stimulate the Sca-1 expression in Sca-1-negative MMC cells ( Fig. 5D and S7C), the silencing of Smad4 generated a subpopulation of Sca-1 + cells (Fig. 5E). TGF-β ligands are important components of cell secretome that can potentially affect the lineage commitment in an autocrine or paracrine manner. We focused on their role in the maintenance of basal-to-luminal (Sca-1 +to-Sca-1 − ) equilibrium in the Comma-Dβ cell line. The inhibition of protein transport from the Golgi apparatus outside of the cell with a sublethal dose of monensin did not disrupt this equilibrium (Fig. 6A). We further specifically neutralized several ligands-of-interest using Type II receptor Fc-chimeras ( Fig. 6B and S8A-D). Such inhibition of TGF-β ligands did not affect the ratio of basal-and luminal-like cells and the Sca-1 levels (Fig. S7B), except for the up-regulation of Sca-1 in ANV5 cells, that produce high levels of active TGF-β1 (Fig. S8E), upon exposure to TGFBR2-Fc. The conditioned media from ANVs also showed high activation of Smad2/3 (TGF-β1 and Activin) reporters and moderate activation of BMP reporter (Fig. 6C-E). These results confirmed that the components of the TGF-β signaling pathway modulate the expression of Sca-1 in a cell type-specific manner: while the basal-to-luminal ratio in Comma-Dβ was not affected by its own TGF-β family ligands, the TGF-β1 secreted by ANV cancer stem cells changed the expression of its target gene Sca-1 in the autocrine fashion.

Discussion
TGF-β signaling plays a critical role in the development and homeostasis maintenance of mammary gland 28 . To assess the effect of TGF-β on Sca-1 expression in pathophysiology, we used the pre-neoplastic mammary epithelial cells Comma-Dβ. This cell line is known for its heterogeneity: the presence of the basal-like subpopulation The plot shows relative reporter activity in HEK293 cells, co-transfected with (SBE) 4 -luc and Renilla vector, starved for 4 h in serum-free medium and exposed to control treatment (DMSO, 1 ng/mL TGF-β1 or 1 μM RepSox) or conditioned media (Comma-Dβ, MMC, ANV2 or ANV5) for 2 h. Data are from two (controls) or four technical replicates (conditioned media). Three biological replicates of conditioned media per cell line were analyzed (*P < 0.05, t test). (D) The plot shows relative reporter activity in HEK293 cells, co-transfected with (CAGA) 12 -luc and Renilla vector, starved for 4 h in serum-free medium and exposed to control treatment (DMSO, 1 ng/mL Activin A or 1 μM SB431542) or conditioned media (Comma-Dβ, MMC, ANV2 or ANV5) for 2 h. Data are from two (controls) or four technical replicates (conditioned media). Three biological replicates of conditioned media per cell line were analyzed (*P < 0.05, t test). (E) The plot shows relative reporter activity in HEK293 cells, co-transfected with (BRE) 2 -luc and Renilla vector, starved for 4 h in serum-free medium and exposed to control treatment (DMSO, 10 ng/mL BMP4 or 1 μM K02288) or conditioned media (Comma-Dβ, MMC, ANV2 or ANV5) for 2 h. Data are from two (controls) or four technical replicates (conditioned media). Three biological replicates of conditioned media per cell line were analyzed (*P < 0.05, t test). www.nature.com/scientificreports/ that exhibits progenitor capability and expresses Sca-1, beside the luminal-like cells that are negative for Sca-1 expression 23 . We showed that the exposure of these cells to TGF-β induced the loss of Sca-1 expression and drove the de-differentiation of both cell lineages. In the normal mammary gland, TGF-β signals are exacerbated during involution and induce apoptosis of unnecessary epithelial cells in branched mammary gland 29 . Comma-Dβ cell line contains clones with mutant Tp53 27,30 , likely responsible for its immortality and pre-neoplastic properties of Sca-1 − subpopulation. Implantation of these cells, pre-treated with TGF-β, into mammary fat pad resulted in shorter latency of tumor growth and sudden relapses. This phenomenon was previously shown only for cells that were exposed to TGF-β for relatively long periods of time 31 . However, our results confirmed that even the transient, short-term exposure to low TGF-β concentrations sufficiently induces the accumulation of tumorinitiating cells through an unknown mechanism. A similar experiment using mesenchymal cancer-stem cells with functional TGF-β pathway showed that TGF-β has as well tumor-suppressive action. Altogether, we provide evidence that TGF-β can promote or suppress tumorigenesis in a context-dependent manner. Mechanistically, we proposed that Sca-1 was repressed by the endogenous TGF-β signaling and was reexpressed after the inhibition of ALK5, the Type I TGF-β receptor, or by the knock-down of its downstream signal transducers Smad2/3/4. This effect was independent of the ALK5 kinase activity in luminal-like, Her2overexpressing cancer cells and only the silencing of Smad4 led to the generation of Sca-1 positive cells. In contrary to myogenic cells, Smad2 and Smad3 were dispensable for Sca-1 repression in mammary epithelial and cancer cells, mediated by exogenous TGF-β, possibly specific for the mammary gland 32 . These results, thus, confirm that the regulation of Sca-1 expression on the cell surface of mammary epithelial and cancer cells is cell lineage-specific and governed by the components of the TGF-β family signaling.