All-trans-retinoic Acid Modulates the Plasticity and Inhibits the Motility of Breast Cancer Cells

Background: All-trans-retinoic acid is a promising therapeutic agent in breast cancer. Results: All-trans-retinoic acid modulates mammary tumor cell epithelial-to-mesenchymal-transition via the TGFβ and NOTCH pathways. Conclusion: The present study unveils a new aspect of all-trans-retinoic acid activity (i.e. regulation of phenotypic cell plasticity). Significance: Our results indicate that all-trans-retinoic acid is endowed with anti-metastatic properties that could be exploited at the therapeutic level. All-trans-retinoic acid (ATRA) is a natural compound proposed for the treatment/chemoprevention of breast cancer. Increasing evidence indicates that aberrant regulation of epithelial-to-mesenchymal transition (EMT) is a determinant of the cancer cell invasive and metastatic behavior. The effects of ATRA on EMT are largely unknown. In HER2-positive SKBR3 and UACC812 cells, showing co-amplification of the ERBB2 and RARA genes, ATRA activates a RARα-dependent epithelial differentiation program. In SKBR3 cells, this causes the formation/reorganization of adherens and tight junctions. Epithelial differentiation and augmented cell-cell contacts underlie the anti-migratory action exerted by the retinoid in cells exposed to the EMT-inducing factors EGF and heregulin-β1. Down-regulation of NOTCH1, an emerging EMT modulator, is involved in the inhibition of motility by ATRA. Indeed, the retinoid blocks NOTCH1 up-regulation by EGF and/or heregulin-β1. Pharmacological inhibition of γ-secretase and NOTCH1 processing also abrogates SKBR3 cell migration. Stimulation of TGFβ contributes to the anti-migratory effect of ATRA. The retinoid switches TGFβ from an EMT-inducing and pro-migratory determinant to an anti-migratory mediator. Inhibition of the NOTCH1 pathway not only plays a role in the anti-migratory action of ATRA; it is relevant also for the anti-proliferative activity of the retinoid in HCC1599 breast cancer cells, which are addicted to NOTCH1 for growth/viability. This effect is enhanced by the combination of ATRA and the γ-secretase inhibitor N-(N-(3,5-difluorophenacetyl)-l-alanyl)-S-phenylglycine t-butyl ester, supporting the concept that the two compounds act at the transcriptional and post-translational levels along the NOTCH1 pathway.

All-trans-retinoic acid (ATRA) 2 and derived retinoids are potent antitumor agents (1), causing growth inhibition, differentiation, and secondary apoptosis. Retinoids bind to specific nuclear receptors (RARs), which function as transcriptional regulators controlling the expression of numerous genes. RARs consist of three subtypes (␣, ␤, and ␥), which act by heterodimerizing with the retinoid X receptors (2). ATRA controls diverse developmental processes and the homeostasis of numerous organs, including the mammary gland (3)(4)(5). In this organ, the retinoid induces lactogenic differentiation of breast cancer cells and triggers an epithelial-like phenotype via cadherin-mediated junction formation (6,7).
Phenotypic plasticity (i.e. the ability of cells to reversibly change phenotype) modulates tumor growth and dissemination (8). Epithelial to mesenchymal transition (EMT) drives polarized, non-motile epithelial cells to acquire highly migratory and fibroblastoid-like features, and it is crucial in normal embryonic development, tissue remodeling, and wound healing (9,10). Increasing evidence supports a role for EMT in tumor invasion and metastatic spread. EMT causes loss of apical-basal polarity, disintegration of tight/adherens junction, and cytoskeletal changes. These structural features are associated with the acquisition of a motile and invasive phenotype (9). Accordingly, many signal transduction pathways, such as TGF␤, NOTCH, and WNT, that are involved in physiological EMT are activated in tumorigenesis and contribute to disease progression (11). The molecular mechanisms underlying EMT are controlled by transcription factors, such as SNAIL, SLUG, TWIST, and ZEB1, as well as specific miRNAs acting in regulatory feedback loops (12).
NOTCH signaling is an evolutionarily conserved pathway involved in development, stem cell self-renewal, and tissue differentiation (13,14). NOTCH activation requires ligand binding and proteolytic cleavage by ADAM/TACE metalloproteases and ␥-secretase. This results in intracellular release of the NOTCH intracellular domain (NICD) from the inner cell membrane. NICD migrates to the nucleus, where it associates with a number of transcription factors. NOTCH activation triggers mesenchymal transformation of breast cancer epithelial cells, particularly via TGF␤ (15,16). For instance, TGF␤ upregulates NOTCH ligands (17), and TGF␤-induced EMT is blocked by pharmacological inhibition of NOTCH (17).
In this study, we exploit a cellular model of breast cancer exquisitely sensitive to the anti-proliferative action of ATRA to demonstrate that the retinoid influences cell plasticity. The retinoid modulates the process of EMT induced by EGF or heregulin-␤1 (Herg), and it inhibits cell migration. We also identify NOTCH1 as a major molecular determinant of ATRA antimigratory action.
Three-dimensional Cell Cultures-Experiments involving three-dimensional cell cultures were carried out according to the protocol of the Muthuswamy laboratory (available on the Muthuswamy laboratory Web site). Briefly, 8-well plates (-slide 8-well, Ibidi) were filled with 70 l of phenol red/ growth factor-free Matrigel (BD Biosciences) and incubated for 20 min at 37°C to solidify the matrix. SKBR3 cells were harvested in trypsin-EDTA and resuspended in culture medium (DMEM/F-12, supplemented with 5% charcoal-stripped FBS) to generate a cell suspension of 2,500 cells/100 l. One volume of 5% Matrigel was combined with one volume of cell suspension, and 400 l of the mixture were seeded in culture wells. Cells were incubated at 37°C in 5% CO 2 for 5 days (the optimal time necessary to grow non-polarized aggregates typical of the tumor phenotype) before treatment. Medium and chemicals were replaced every 3 days.
Cell Growth Assays-Cell growth was determined with the sulforhodamine assay (19) or by the CellTiter-Glo luminescent cell viability assay (Promega) according to the manufacturer's instructions. Each cell line was exposed to increasing concentrations of ATRA (0.001, 0.01, 0.1, 1.0, and 10.0 M) for 3, 6, or 9 days. In the latter case, the culture medium was replaced with fresh medium once every 3 days. The percentage growth inhibition (% GI) induced by ATRA concentration i was calculated only for concentrations for which T i (6) Ն C (3), according to the formula, % GI(i) ϭ 100 ϫ (T i (6) Ϫ T i (3))/(C(6) Ϫ C(3)), where T i (3) represents the absorbance at day 3 of treated samples for a drug concentration i; C (3) is the absorbance at day 3 of control samples for a drug concentration i; T i (6) is the absorbance at day 6 of treated samples for a drug concentration i; and C (6) is the absorbance at day 6 of control samples for a drug concentration i. The % GI(i) values were fitted for ATRA concentrations with a Hill function. GI50 was calculated from the best fit function if at least three data points with % GI between 5 and 95% could be fitted and the r 2 of the fit was Ͼ0.8; otherwise, the value was determined by linear interpolation.
Immunofluorescence and Time Lapse Microscopy-For indirect immunofluorescence microscopy, cells were fixed with 3% paraformaldehyde and permeabilized with 0.1% Triton X-100 in PBS for 5 min. Fixed cells were incubated for 45 min with anti-VE-cadherin (BV9) (20), anti-␤-catenin (BD Biosciences), anti-occludin (Zymed Laboratories Inc.), anti-ZO-1 (Zymed Laboratories Inc.), and anti-SNAIL (Cell Signaling) antibodies. For NOTCH1 immunofluorescence, cells were fixed and permeabilized with methanol for 5 min on ice and stained with anti-NOTCH1 antibodies (Cell Signaling). Positive cells were identified following incubation with 488 or 543 Alexa-conjugated secondary antibodies (Invitrogen) for 30 min. Threedimensional immunofluorescence studies were performed according to a standard protocol (available on the Muthuswamy laboratory Web site). Fluorescence microscopy studies were performed with an IX71 apparatus (Olympus), and images were recorded using a CCD camera and the Cell ∧ F imaging software (Olympus). Confocal laser microscopy experiments were conducted with a IX81 microscope coupled to an Olympus FluorView 300 laser-scanning confocal microscope (Olympus).
Transfections and Western Blots-For the experiments involving NOTCH1 overexpression, SKBR3 cells were seeded in 6-well plates and transfected at 30 -40% confluence with the mammalian expression plasmids pcDNA3HAN1⌬E FLAG (hNOTCH1) (1 g) and pEGFPN1 (100 ng) using Lipofectamine 2000 (Life Technologies, Inc.) according to the manufacturer's instructions. Twenty-four hours later, cells were split and reseeded to normalize for the transfection efficiency.
For the experiments involving SMAD3 silencing, SKBR3 cells were co-transfected with a 60 nM concentration of a validated SMAD3 siRNA (HSS106252, Life Technologies) (6) or an appropriate control siRNA (stealth RNAi siRNA negative control HiGC, 12935400) and the normalization plasmid, pEG-FPN1 (300 ng), using Lipofectamine 3000 (Life Technologies) according to the manufacturer's instructions. Forty-eight hours following transfection, cells were subjected to cell motility assays using Boyden chambers.
Subcellular Fractionation-Triton X-100-soluble and -insoluble fractions were obtained as described (22). Briefly, cells were incubated with Triton lysis buffer (1% Triton X-100, 0.3 M sucrose, 25 mM HEPES, pH 7.4, 100 mM NaCl, 4.7 mM KCl, 1.2 mM KH 2 PO 4 , 1.2 mM MgCl 2 , protease inhibitors-complete; Roche Applied Science) on a rocking platform for 15 min. After centrifugation for 15 min at 10,000 ϫ g, the supernatant (Triton-soluble fraction) was collected. The Triton-insoluble fraction was resuspended in SDS lysis buffer. Both fractions were reconstituted in equal volumes, and 15 l of each fraction were analyzed by Western blot.
Cell Motility Assays-Migration assays were performed using Boyden chambers with insets coated on the bottom surface with fibronectin (8.0-m pore size; BD Biosciences). Subconfluent cell cultures were serum-starved for 24 h prior to harvesting with trypsin-EDTA and seeding (100,000 cells/well) in serum-free medium (DMEM/F-12, 1% BSA) supplemented with the indicated growth factor and/or chemical compound. The lower compartment of the chamber was filled with DMEM/F-12 medium supplemented with 5% charcoalstripped FBS as chemoattractant and with the indicated growth factor and/or chemical compound. Migrated cells were washed twice with PBS and fixed with 100% ice-cold methanol for 10 min. After methanol removal, cells were stained with crystal violet (0.5% in 25% methanol) for 10 min and washed several times with water. The number of migrated cells was determined by counting three replicates of six independent visual fields per well at ϫ20 magnification under an inverted microscope (IX71, Olympus).
For wound closure assays, cells were allowed to reach confluence before serum starvation for 24 h. The monolayers were scraped with a plastic pipette tip and replenished with fresh serum-free medium supplemented with vehicle (ethanol), EGF (50 ng/ml), Herg (50 ng/ml), ATRA (1 M), and combinations thereof. Phase contrast images were obtained 24 h after wounding.
RNA Extraction and Gene Expression Analysis by qRT-PCR-Total RNA was isolated with the miRNeasy minikit (Qiagen, Hilden, Germany). GeneAmp RNA PCR core kits (Applied Biosystems) were used to perform reverse transcription reactions according to the manufacturer's instructions.
DNA Extraction and Gene Copy Number Assessment by qRT-PCR-DNA was prepared with the DNeasy blood and tissue kit (Qiagen). RARA gene copy number was determined by qRT-PCR, using the 7300 System (SDS software version 1.3, Applied Biosystems, Foster City, CA). Predesigned TaqMan copy number assays Hs01079964_cn (RARA) were from Applied Biosystems. The ribonuclease P RNA component H1 gene (RPPH1) was used as a reference control. Samples were analyzed in triplicate. Human genomic DNA (catalog no. G1471) (Promega, Madison, WI) was used as a calibrator sample in every assay. Analysis of gene copy number was carried out using the comparative Ct method (23). The copy number of ERBB2 and RARA genes in normal human genomic DNA was set as 2.0, and a copy number of 3.0 was considered as the threshold value for the gene amplification calls.
Whole Genome Gene Expression Microarrays-Gene expression data for SKBR3 cells treated with ATRA (100 nM) were previously deposited (Arrayexpress E-MEXP-3192). For enrichment analysis, we used genes whose expression was significantly altered by ATRA either at 12 or at 48 h (paired t test, p Ͻ 0.05) and the Metacore platform (see the Thomson Reuters Web site).

ATRA Induces Adherens and Tight Junctions in SKBR3
Cells-The HER2 ϩ SKBR3 cell line is highly responsive to the anti-proliferative activity of ATRA, because it is characterized by amplification of the gene coding for RAR␣ (6). This luminal cell line does not express typical epithelial or mesenchymal biomarkers, and it is an excellent model to study the action of ATRA on EMT (24 -26), a process activated by exposure to EGF (27,28) and Herg (29). In SKBR3 cells, ATRA increased the amounts of VE-cadherin, ␣-catenin, and ␤-catenin (Fig. 1A), which participate in the formation of intercellular contacts (30,31). ATRA-dependent induction involves both the Triton X-100-soluble and -insoluble pools of the three proteins. Increases in the Triton X-100-insoluble fraction are consistent with relocalization of VE-cadherin and ␤-catenin to adherens junctions, which were visible only in SKBR3 cells exposed to ATRA (Fig. 1B). VE-cadherin is the only cadherin induced by ATRA in SKBR3 cells (6). In ATRA-induced adherens junctions, VE-cadherin is likely to substitute for E-cadherin, the major adhesion protein observed in the mammary gland epithelium. In fact, SKBR3 cells do not express E-cadherin, because the corresponding gene is deleted (32). In SKBR3 cells, ATRA exposure led also to reorganization of tight junctions and relocalization of occludin as well as ZO-1 to these structures (Fig. 1B).
Taken together, our results demonstrate that ATRA increases the number of cell-to-cell contacts. The process may involve activation of a program causing cancer cells to progress along the mammary epithelial differentiation pathway (6).

ATRA Stimulates the Formation of Organized Structures Reminiscent of a Luminal Epithelium in Three-dimensional
Cultures of SKBR3 Cells-We used three-dimensional cultures of SKBR3 cells to evaluate whether ATRA-dependent induction of adherens and tight junctions is indeed part of an epithelial differentiation program set in motion by the retinoid. Normal mammary epithelial cells grown in three-dimensional Matrigel cultures respond to microenvironmental stimuli by reducing their proliferation and by differentiating into mammary acinar structures (spheres). These responses are lost in three-dimensional cultures of breast cancer cells (33). In threedimensional Matrigel cultures, untreated or vehicle-treated SKBR3 cells are organized in grapelike clusters (34) with no apical/basal polarity, as indicated by triple-fluorescence staining with the plasma membrane protein HER2, the Golgi protein GM-130, and the nuclear dye Hoechst ( Fig. 2A). Upon challenge with ATRA, the grapelike clusters turn into hollow spheres, which are reminiscent of a normal mammary gland epithelium. These spheres are polarized with HER2 displaying an apico-lateral distribution and the Golgi apparatus oriented toward the lumen. The spherical structures are characterized by cell-to-cell contacts that are maintained by VE-cadherin (Fig. 2B). In fact, anti-VE-cadherin antibodies disrupt cellular interactions and consequently sphere integrity (Fig. 2C).
In ATRA-treated cultures, propidium iodide staining demonstrates cell death in undifferentiated grape-like clusters (Fig.  3A). In contrast, no appreciable cell death was evident in ATRA-induced spheres. The observation suggests that spheres consist of viable cells organized into a luminal epithelium. No cell death or caspase activation was detectable in the lumen of spheres (Fig. 3B). Thus, the cavity is likely to result from a structural reorganization of the growing cells rather than an apoptotic process involving the inner cell layer.
ATRA Modulates Plasticity in HER2 ϩ Breast Cancer Cells via RAR␣ Independently of Growth Inhibition-We tested whether ATRA-dependent differentiation leading to the generation of organized luminal structures is associated with a growth-inhibitory response. To this purpose, we used a small and homoge- neous panel of cell lines belonging to the same HER2 ϩ breast cancer subtype as the SKBR3 counterpart (24,25). The five selected cell lines are characterized by resistance or sensitivity to the anti-proliferative action of ATRA (Table 1). MDAMB453 and HCC1954 cells, which showed relative resistance to ATRA, were equally unresponsive to the retinoid in three-dimensional cultures (Fig. 4A). Similar to the SKBR3 counterparts, ATRAsensitive UACC812 cells underwent epithelial differentiation into hollow spherical structures upon challenge with the retinoid (Fig. 4A). In UACC812 spheres, cells were polarized with nuclei located at the periphery (Fig. 4B). These spheres resemble the prototypical epithelial ducts described for three-dimensional cultures of Madin-Darby canine kidney cells (35). The data obtained in our cell lines suggest that the process of growth inhibition set in motion by ATRA is generally accompanied by epithelial differentiation. However, sensitivity to growth inhibition by ATRA in HER2 ϩ breast cancer cells is not sufficient to induce morphological effects. Indeed, three-dimensional cultures of ATRA-sensitive MDAMB361 cells (Fig. 4A) did not show the same retinoid-dependent morphological changes observed in the SKBR3 and UACC812 counterparts, indicating that epithelial differentiation is not a direct consequence of growth inhibition.
The HER2 ϩ cell lines considered expressed significant levels of RAR␣3 and RAR␥1, the major RAR␣ and RAR␥ transcripts (Fig. 5A, left). With the exception of UACC812 cells, which showed barely detectable amounts of the RAR␤1 transcript, no significant expression of the main RAR␤ mRNA was observed in the other cell lines. In all of the cell lines, RAR␣3 was the most abundant RAR form measured, and the highest levels of the transcript were observed in SKBR3 and UACC812 cells. This is due to the fact that the two cell types are characterized by amplification of the RARA gene, which is part of the HER2 amplicon (Fig. 5A, right) (6). This observation suggests that RAR␣3 is involved in the formation of ATRA-induced spheric structures. Because ATRA is a pan-RAR agonist, binding and activating RAR␣, RAR␤, and RAR␥ with the same affinity and efficiency, we compared the action of the RAR␣ agonist AM580, the RAR␤ agonist BMS641, and the RAR␥ agonist CD437 on sphere formation in SKBR3 and UACC812 three- dimensional cultures (Fig. 5B). In both cell lines, AM580 is capable of recapitulating the effects of ATRA, whereas BMS641 and CD437 do not induce spheres. It must be noted that BMS961, another RAR␥ agonist, is equally ineffective as CD437 in inducing spheres (data not shown). Taken together, our results indicate that activation of RAR␣ is necessary and sufficient to trigger the differentiation process leading to the formation of spheres. ATRA Modulates EMT and Inhibits SKBR3 Cell Migration-The data obtained in breast cancer cell lines indicate that the differentiation program set in motion by ATRA leads to an increase in cell-to-cell interactions. This process is a key regulator of cancer cell motility, invasiveness, and metastatic spread. In cell lines characterized by plasticity, such as SKBR3, the migratory potential is augmented by acquisition of a mesenchymal phenotype (9) via the reversible process of EMT.
To induce EMT in HER2 ϩ SKBR3 cells, we used EGF and Herg (29,36), which activate the process through the HER1/ HER2 and HER3/HER2 pathways, respectively (27). SKBR3 cells treated with EGF or Herg for 24 h acquired an elongated/ fibroblast-like phenotype characteristic of mesenchymal cells (Fig. 6A). ATRA suppressed the EMT-associated morphological changes afforded by EGF and Herg. This was paralleled by induction and reorganization of adherens and tight junctions, as demonstrated by the formation of ␤-cateninand ZO-1-containing complexes in cell-to-cell contacts (Fig. 6B). To test whether these actions translate into inhibitory effects on the directional migration of SKBR3 cells, we used Boyden chambers. The studies were complemented by an analysis of the morphological traits of migrating cells performed with the use of wound assays (37). As expected, EGF and Herg caused a dramatic increase in SKBR3 cell migration (Fig. 6C), which was due to an increase in single-cell motility (Fig. 6D). In fact, cells stained for the occludin marker did not show the appearance of tight junctions and an increase in cell-to-cell contacts (Fig. 6D). In Boyden chamber assays, ATRA inhibited the dramatic increase in SKBR3 cell migration caused by EGF and Herg (Fig.  6C). This is likely to be the result of an ATRA-dependent increase in intercellular contacts, as indicated by the relocaliza- tion of occludin to tight junctions (Fig. 6D), which causes a shift from single to collective cell migration (37).
To identify the specific RAR isoform mediating the anti-migratory action, we compared the effects of ATRA, AM580, BMS641, and CD437 on the migratory behavior of SKBR3 cells using Boyden chamber assays (Fig. 6E). The RAR␣ agonist AM580 is the only compound capable of mimicking the antimotility action of ATRA. This effect is likely to result from activation of the same differentiation program set in motion by ATRA and AM580. In fact, AM580 was the sole RAR agonist that inhibited EGF-and Herg-activated EMT (data not shown). In addition and unlike BMS641 or CD437, exposure of SKBR3 cells to AM580 caused the same reorganization of tight junctions observed with ATRA (data not shown).
In conclusion, the results indicate that the epithelial differentiation process triggered by ATRA via RAR␣ activation in retinoid-sensitive and plastic SKBR3 cells is capable of counteracting the EMT responses ignited by two established EMTinducing agents, like EGF and Herg. EMT inhibition is likely to be a major contributor to the anti-migratory action of ATRA determined in our in vitro models.
ATRA Suppresses the Induction of NOTCH1 and SNAIL Afforded by EMT Activation in SKBR3 Cells-To clarify the molecular mechanisms underlying the inhibitory effects of ATRA on migration, we considered SNAIL, SLUG, NOTCH1, FOXC2, and TWIST, a series of EMT-associated transcription factors. Using qRT-PCR, we evaluated the effects of EGF, Herg, and ATRA alone and in combination on the expression of the corresponding mRNAs.
In SKBR3 cells, SLUG, FOXC2, and TWIST are unlikely to play a role in the EMT process triggered by EGF or Herg and inhibited by ATRA. In fact, TWIST was never detectable in any of the experimental conditions considered (data not shown). In addition, SLUG mRNA expression was left unaffected by EGF, Herg, and ATRA alone or in combination (Fig. 7A). Finally, the FOXC2 transcript was up-regulated only by the retinoid. In contrast, SNAIL and NOTCH1 are potential mediators of ATRA-dependent EMT inhibition. Indeed, EGF and Herg up-  regulate SNAIL and NOTCH1 mRNA. Challenge of SKBR3 cells with ATRA left the basal expression levels of SNAIL and NOTCH1 mRNAs unaltered, whereas it suppressed induction of the two transcripts by EGF and Herg (Fig. 7A). The mRNA results were confirmed at the protein level (Fig. 7B). The immunofluorescence analyses conducted on SNAIL indicate that EGF-dependent induction leads to relocalization of the transcription factor into the nucleus of SKBR3 cells (Fig. 7C). Consistent with the suppressive effect of the retinoid on SNAIL induction, the transcription factor was undetectable in the nuclei of cells treated with ATRA and EGF.

ATRA-dependent Suppression of EMT-associated NOTCH1 Induction Is Accompanied by Reduction of the Derived NICD
Cleavage Product-Given the predicted significance in breast cancer progression and the role played in the metastatic process of this tumor (38 -40), we studied the regulation of NOTCH1 by ATRA in more detail.
In a first set of experiments, we evaluated whether NOTCH1 down-regulation by ATRA is the result of a transcriptional effect and performed transfection studies with a luciferase construct controlled by the NOTCH1 promoter (41). EGF stimulated luciferase activity in SKBR3 cells transfected with the construct, and ATRA counteracted this effect (data not shown). These results support the concept that NOTCH1 down-regulation by the retinoid is the result of a transcriptional effect.
In a second set of experiments, we evaluated whether posttranscriptional effects contribute to NOTCH1 down-regulation by ATRA. In fact, the NOTCH1 gene codes for a cell membrane-bound precursor, which can be processed into NICD via two sequential proteolytic steps involving ADAM/TACE and ␥-secretase. NICD consists of the intracellular carboxyl-terminal portion of the protein, which translocates to the nucleus, where it acts as a transcription factor (14). Thus, we evaluated whether ATRA exerts any effect on NOTCH1 proteolytic cleavage following EMT activation by EGF. The total amounts of NOTCH1 and NICD were determined by Western blot analyses with an antibody recognizing the C-terminal portion of the protein (Fig. 8A). As expected, treatment of SKBR3 cells with EGF caused an increase in NOTCH1 ϩ NICD, which was suppressed by ATRA. To determine the amounts of NICD specifically, we used an antibody targeting Val-1744 (42). The data indicate that the surge of NICD in SKBR3 cells exposed to EGF is also blocked by ATRA. This demonstrates that ATRA-dependent down-regulation of the NOTCH1 transcript translates into a decrease in the corresponding precursor and the ␥-secretase processed product. To further evaluate the effects of EGF and/or ATRA in SKBR3 cells, we performed confocal immunofluorescence studies in conditions allowing specific visualization of membrane-associated NOTCH1. EGF caused a remarkable increase in the amounts of NOTCH1 (Fig. 8B). In contrast, ATRA did not alter the basal levels of NOTCH1 in conditions causing the expected increase in tight junction-associated ␤-catenin (see also Fig. 1B). Additionally, the retinoid suppressed EGF-dependent up-regulation of membrane-bound NOTCH1. Incidentally, it is interesting to note that ATRAinduced ␤-catenin does not co-localize with NOTCH1, indicating that the two proteins are located in different domains of the plasma membrane.
We evaluated whether the ATRA-dependent decrease in NICD levels is simply consequent to down-regulation of the precursor mRNA/protein or if it is also due to inhibition of ␥-secretase-dependent NOTCH1 cleavage. To this purpose, we overexpressed a membrane-tethered, ligand-independent NOTCH1 derivative, which is constitutively cleaved into NICD (43). In this construct, NOTCH1 was tagged with an N-terminal HA epitope located upstream of the ␥-secretase cleavage site and a C-terminal FLAG epitope (Fig. 8C). Western blots performed with anti-HA antibodies demonstrated similar steady-state levels of the NOTCH1 precursor protein in SKBR3 cells treated with vehicle, the ␥-secretase inhibitor DAPT (44), ATRA, and ATRA ϩ DAPT (Fig. 8C). In these experimental conditions, DAPT was fully active, as demonstrated by the results obtained with the anti-NICD antibody. Finally and more importantly, ATRA exerted no effect on the FLAG-tagged NOTCH1 protein, whose levels were increased by DAPT as predicted. Taken together, these results demonstrate that ATRA does not affect ␥-secretase cleavage of NOTCH1. Thus, suppression of EGF-dependent NICD induction by ATRA is solely the result of a transcriptional effect causing a reduction in the levels of the NOTCH1 mRNA and protein.
NOTCH1 and TGF␤ Mediate the Anti-migratory Activity of ATRA in SKBR3 Cells-Because NOTCH1 stimulates migration in breast cancer cells (45), retinoid-dependent inhibition of NOTCH1 expression may contribute to the anti-migratory action of ATRA in SKBR3 cells challenged with EGF. To test this hypothesis, we evaluated the effect of DAPT on the promigratory action of EGF. In conditions of NOTCH1 proteolytic cleavage inhibition (Fig. 9, right), treatment of SKBR3 cells with DAPT resulted in a complete suppression of the motile response activated by the growth factor (Fig. 9,  left and middle). These results demonstrate that processing of NOTCH1 into NICD is necessary for the pro-migratory activity of EGF. In addition, they are consistent with the idea that NOTCH1 down-regulation is involved in ATRA antimotility action.
The network of NOTCH-interacting genes and proteins is complex (Fig. 10), and it contains many growth factors, including TGF␤, EGF, and Herg (Neuregulin1 in Fig. 10). TGF␤ is of particular interest, because it is a well known mediator of ATRA activity in breast cancer cells and it is an EMT inducer (9). The TGF␤ pathway contains elements that lie both upstream and downstream of NOTCH1 activation. Pathwaymap and process-network analyses of the genes differentially expressed in SKBR3 cells challenged with ATRA (6) demonstrated a significant enrichment of both the NOTCH1 and TGF␤ signaling pathways ( Table 2 and Table 3). Given the role played by both NOTCH1 and TGF␤ in EMT, it is not surprising that this is the most enriched process-network modulated by ATRA in our experimental model. Interestingly, many of the genes regulated by ATRA are indeed part of the TGF␤ pathway (Fig. 10). In particular, ATRA down-regulated TGF␤3, and it up-regulated GDF15, TGF␤RII, and SMAD3 (Fig. 11, A and B,  right). Thus, ATRA acts at multiple levels along the TGF␤ signal transduction pathway, controlling the steady-state levels of two ligands (TGF␤3 and GDF15), a type II receptor (TGF␤RII), and one of the major intracellular signal transducers (SMAD3). In a related series of experiments, we evaluated whether EGF exerts any effect on the members of the TGF␤ pathway regulated by ATRA. These analyses were conducted both in the absence and in the presence of ATRA. EGF regulated the TGF␤3 and GDF15 mRNAs and the SMAD3 protein (Fig. 11, A  and B). Thus, the growth factor interferes with the TGF␤ pathway in SKBR3 cells. Interestingly, both EGF and ATRA downregulate TGF␤3 and induce GDF15 mRNAs, despite the opposite effects exerted by EGF and ATRA on EMT. EGF-dependent down-regulation of TGF␤3 is not enhanced by the addition of ATRA, whereas EGF-dependent induction of GDF15 is enhanced by the retinoid. Conversely and in line with the relative action on EMT, up-regulation of the TGF␤RII transcript by ATRA is suppressed by EGF.
To directly define the functional involvement of the TGF␤ pathway in the NOTCH1-dependent anti-migratory effects exerted by ATRA, SKBR3 cells were exposed to EGF and subjected to Boyden chamber assays using the TGF␤RI inhibitor, SB431542. Consistent with the involvement of TGF␤ in EMT (46), SB431542 reduced the pro-migratory action of EGF (Fig.  11B, left and middle). Paradoxically, however, TGF␤RI inhibition causes a partial and reproducible reversion of the antimotility action of ATRA observed in cells exposed to EGF.
Binding of TGF␤ receptors results in the activation of the transcription factor, SMAD3, which plays a central role in this signal transduction pathway (46,47). To confirm and extend the data obtained with SB431542, we silenced SMAD3 (6) with a validated siRNA (Fig. 11C, left) and evaluated the effects of this knockdown on the migratory behavior of SKBR3 cells exposed to EGF, ATRA, and ATRA ϩ EGF. Down-regulation of SMAD3 and consequent downstream inhibition of the TGF␤ signal transduction pathway inhibit the migration of SKBR3 cells challenged with EGF. In contrast and similar to what is observed with SB431542, SMAD3 knockdown causes an increase in the number of migrated cells observed following challenge with ATRA ϩ EGF. Taken together, the data indicate  that ATRA causes a switch of the TGF␤ pathway from promigratory to anti-migratory.
ATRA Controls the Phosphorylation/Activation of SMAD3-ATRA is likely to modulate the levels and the activation state of SMAD3 via TGF␤. In addition, SMAD3 may contribute to NOTCH1 down-regulation by ATRA in SKBR3 cells exposed to EGF. Both ATRA and EGF increase the amounts of the SMAD3 protein. SMAD3 induction was enhanced by the combination of ATRA and EGF (Fig. 11B, right). Unlike EGF, ATRA not only increases SMAD3 expression; it also induces activation of the transcription factor via phosphorylation of its carboxyl-terminal domain at Ser-423 and Ser-425. Ser-423 and Ser-425 phosphorylation by ATRA is mediated by TGF␤, as the specific TGF␤RI inhibitor, SB431542, abrogates it. Consistent with the reported relevance of these SMAD3 phosphorylation sites for the growth-inhibitory action of TGF␤ (47,48), ATRA-dependent phosphorylation is counteracted by a growth factor like EGF. These data indicate that both ATRA and EGF modulate the levels of SMAD3,

TABLE 2 Pathway enrichment analysis in SKBR3 cells treated with ATRA
We performed enrichment analysis of the genes regulated by ATRA using Metacore-annotated pathway maps (see the Thomson Reuters Web site). Each pathway map represents a multistep canonical pathway backed by "small experiment" evidence in the literature. The pathways of particular relevance for our study are indicated in boldface type. FDR, false discovery rate.

No. Pathway maps
Total genes ATRA-modulated p FDR

TABLE 3 Process/network enrichment analysis in SKBR3 cells treated with ATRA
We performed enrichment analysis of the genes regulated by ATRA using Metacore-annotated process networks (see the Thomson Reuters Web site). The content of these cellular and molecular processes is defined and annotated by Thomson Reuters scientists. Each process represents a preset network of protein interactions characteristic for the process. The processes of particular relevance for our study are indicated in boldface type.

No. Process networks
Total genes ATRA-modulated p FDR  although the two stimuli exert differential effects on the activation status of the TGF␤ mediator. With respect to this last point, it is interesting to note that EGF phosphorylates SMAD3 in the linker domain on sites other than Ser-423 and Ser-425, and this phosphorylation site has been associated with the pro-invasive action of theTGF␤pathway (47,48).Weproposethatdifferentialphosphorylation of SMAD3 by ATRA and EGF may explain the paradoxical effects on cell migration exerted by activation of the TGF␤ pathway in SKBR3 cells.
Because the anti-migratory action of the TGF␤ pathway activated by ATRA involves NOTCH1, the effects of SB431542 on NICD levels were evaluated in SKBR3 cells exposed to EGF, ATRA, and ATRA ϩ EGF. SB431542 did not exert any significant and reproducible effect on the levels of NICD observed in any of the experimental conditions considered (Fig. 11B, right). Similarly, SB431542 did not affect the levels of the NOTCH1 mRNA (data not shown). Taken together, our results support the idea that ATRA-dependent induction of the TGF␤ pathway does not interfere with NOTCH1 activation.
The Anti-proliferative Action of ATRA Is Mediated by NOTCH1 in HCC1599 Breast Cancer Cells-The relevance of NOTCH1 for the anti-tumor action of ATRA may not be limited to the control of breast cancer cell migratory behavior and invasiveness and may extend to other processes, such as cell growth. However, given the very low basal expression level of NICD, SKBR3 cells are not suitable to define whether NOTCH1 inhibition contributes to the anti-proliferative action of ATRA. For this reason, we turned to HCC1599 breast cancer cells, which are characterized by addiction to NOTCH1 in terms of growth (49). This cell line is exquisitely sensitive to the anti-proliferative action of the retinoid (GI50 ϭ 30 nM). With the highest concentrations of ATRA, growth inhibition was already evident after 3 days, although the maximal anti-proliferative effects were observed after 6 days of treatment (Fig. 12A). Maximal growth inhibition was preceded by down-regulation of the NOTCH1 mRNA (Fig.  12B) and the NICD protein (Fig. 12C). In fact, complete disappearance of NICD was observed after 4 days of treatment with ATRA. In the same experimental conditions, DAPT did not alter NOTCH1 mRNA levels (Fig. 12B), whereas it caused an early and long-lasting suppression of NOTCH1 processing into NICD (Fig. 12C). Co-treatment of HCC1599 cells with ATRA and DAPT enhanced the growth-inhibitory effects exerted by the single components of the combination (Fig. 12D). In this cellular model, ATRA-and DAPT-induced growth arrest is accompanied by down-regulation of the NOTCH1 target genes, c-MYC, CCND1 (cyclin D1), HEY2 (Hes-related family basic helix-loop-helix transcription factor with YRPW motif 2), and HES4 (Hes family basic helixloop-helix transcription factor 4) (Fig. 12E). In this case too, combined treatment of HCC1599 cells with ATRA and DAPT results in enhanced down-regulation of three (c-MYC, CCND1, and HES4) of the four NOTCH1-regulated genes (13,14). The additive/synergistic anti-proliferative and gene regulatory effects observed with the combination are consistent with the idea that ATRA and DAPT act at two different points along the NOTCH1 activation pathway (i.e. gene transcription and post-translational processing). In addition, our results indicate the existence of a cross-talk between the retinoid and the NOTCH1 signaling pathways, not only in terms of cell motility but also as far as cell growth is concerned.

Discussion
The paradigmatic retinoid, ATRA, is the first example of a clinically useful cyto-differentiating agent; it is approved for the treatment of acute promyelocytic leukemia, a rare form of acute myelogenous leukemia in the adult (50). ATRA is proposed as an antitumor agent not only in the context of hematological malignancies but also for the chemoprevention and treatment of breast cancer (1). Unlike classic chemotherapeutics, which are direct cytotoxic agents, the antileukemic activity of ATRA in APL blasts is primarily the result of anti-proliferative and cyto-differentiating effects. Similar mechanisms are likely to be at the basis of the antitumor action of ATRA in breast cancer, because one of the physiological roles of vitamin A is the control of epithelial homeostasis in the mammary gland (51,52).
In this report, we demonstrate that ATRA exerts a strong anti-migratory action in the SKBR3 cell line, which is representative of a particular subtype of HER2 ϩ breast cancer characterized by unique sensitivity to the anti-tumor action of the retinoid, because of an amplification of the RARA locus, coding for RAR␣ (6). Our results indicate that ATRA is endowed with anti-migratory properties, which may contribute to the overall anti-tumor activity of this compound in breast cancer, influencing the tissue invasion and metastatic processes. The anti-migratory action of ATRA is observed in conditions in which the process is stimulated by well known inducers of EMT, like the growth factors EGF and Herg. Collectively, our data are consistent with the idea that the anti-migratory action of ATRA is largely the result of regulatory effects on the process of EMT, which is associated with the invasive and metastatic behavior of the cancer cell (53), and it is considered to play a crucial role in tumor progression (9).
In the plastic SKBR3 cell line, the action of ATRA is likely to be secondary to the activation of a mammary epithelial differentiation program. Activation of the program causes an increase in cell-to-cell contacts via the formation of adherens junctions and the reorganization of tight junctions. These phenomena are at the basis of the anti-migratory action of ATRA. Consistent with the activation of an epithelial differentiation program, three-dimensional cultures of SKBR3 cells exposed to ATRA give rise to polarized structures reminiscent of mammary gland acini (33). The phenomenon is not a general feature of cell lines sensitive to the anti-proliferative effects of ATRA, as indicated by the results obtained on a small and homogeneous panel of HER2 ϩ breast cancer cell lines endowed with different sensitivity to the ATRA anti-proliferative action. It is interesting to notice that SKBR3 and UACC812 are the only HER2 ϩ cell lines of the panel undergoing differentiation into polarized epithelial structures in our three-dimensional culture conditions. Both cell lines are characterized by amplification of the RARA gene encoding RAR␣, which we identified as the receptor responsible for the differentiating action of ATRA (6). Our data are of relevance for a recent study, demonstrating that unliganded RAR␣ triggers cell migration via EMT induction in a model of MCF10A engineered for the overexpression of the retinoid receptor and HER2 (54). Thus, the presence of pharmacological concentrations of ATRA modifies the activity of RAR␣ in the processes of EMT and cell migration.
The present study also provides insights into the molecular mechanisms at the basis of the modulating effects exerted by ATRA on EMT and cell migration. Modulation of the NOTCH1 signal transduction pathway seems to play a major role in the anti-motility responses activated by ATRA in the SKBR3 model. This is consistent with the well known regulatory action of the protein on EMT in breast cancer cells (15,17). NOTCH1 expression levels are induced by EGF and Herg in SKBR3 cells, and ATRA counteracts this effect. ATRA-dependent inhibition of NOTCH1 expression is the result of a transcriptional action and does not involve modulation of the posttranslational activation step carried out by ␥-secretase. In fact, ATRA does not affect ␥-secretase-dependent cleavage of the NOTCH1 precursor into the transcriptionally active NICD form of the protein, although we cannot exclude the possibility that the retinoid acts on other post-transcriptional regulatory checkpoints (13). The results obtained with the ␥-secretase inhibitor, DAPT, which suppresses the migration of SKBR3 cells challenged with EGF, strongly support the functional relevance of the NOTCH1 pathway in the anti-motility action of ATRA. This is in agreement with a previous report demonstrating that NOTCH1 activation favors the motility and invasive ability of epithelial cancer cells (45).
The TGF␤ pathway is a second signal transduction system that is of fundamental importance for the anti-migratory action of ATRA in our cellular model. TGF␤ is generally considered to be an EMT inducer (55); hence, activation of the corresponding pathway would be expected to increase the migratory potential of breast cancer cells. In contrast, the data obtained in SKBR3 cells exposed to ATRA and EGF indicate that the picture is much more complex. Challenge of SKBR3 cells with ATRA activates the TGF␤ pathway, as indicated by increased phosphorylation of the intracellular SMAD3 signal transducer. Nevertheless, the effects observed upon suppression of the TGF␤ pathway with SB431542 and SMAD3 silencing are paradoxical. Inhibition of TGF␤ signaling reduces the EGF-induced migratory behavior of SKBR3 cells, whereas it stimulates motility in ATRA ϩ EGF-treated cells. This is consistent with a dual role of the TGF␤ pathway in EMT, which is dependent on the exposure of the neoplastic cell to pharmacological concentrations of ATRA. In the absence of ATRA, activation of the TGF␤ pathway exerts the expected pro-migratory action. In the presence of ATRA, a functional switch of the TGF␤ pathway to an antimigratorydeterminantisobserved.Interestingly,ATRA-dependent activation of the TGF␤ pathway in SKBR3 cells results in the phosphorylation of Ser-423 and Ser-425 of SMAD3, whereas EGF-dependent activation of the pathway does not result in the phosphorylation of the same residues. EGF inhibits the phosphorylation of Ser-423 and Ser-425. This differential effect on SMAD3 phosphorylation/activation may contribute to the switch of TGF␤ action on EMT observed in the absence and presence of ATRA.
The TGF␤ and NOTCH pathways are highly integrated, which raises the question of whether ATRA-dependent downregulation of NOTCH1 influences the activation state of the TGF␤ signal transduction system or if the opposite is true. In our experimental conditions, inhibition of TGF␤RI by SB431542 in ATRA-treated SKBR3 cells affects neither the amounts of NICD protein nor the levels of the NOTCH1 mRNA. The observation suggests that stimulation of the TGF␤ pathway by ATRA lies downstream of NOTCH1 activation. At present, the elements of the TGF␤ pathway responsible for the paradoxical effects of the growth factor on the action of ATRA on cell motility are unknown.
The action of ATRA on the NOTCH pathway is of relevance not only for the anti-migratory but also for the anti-proliferative responses triggered by ATRA, as indicated by the results obtained in the HCC1599 cell line. This cell line is representative of NOTCH1-addicted breast cancer cells, because it is characterized by high levels and constitutive activation of the protein. In addition, the proliferation of HCC1599 cells in culture is dependent on NOTCH1 (49,56). Interestingly, this model is extremely sensitive to ATRA, despite the basal phenotype and estrogen receptor negativity, which are determinants of retinoid resistance in breast cancer cells (57). In HCC1599 cells, our data support the concept that down-regulation of NOTCH1 contributes to the growth-inhibitory action of ATRA. In fact, combined treatment of the cell line with ATRA and the ␥-secretase inhibitor, DAPT, causes a synergistic growth-suppressing effect. Interestingly, the anti-proliferative action of ATRA and DAPT in NOTCH1-addicted cells is associated with a synergistic down-regulation of NOTCH1 target genes. Besides their interest for the molecular mechanisms underlying the action of ATRA, these data suggest further studies aimed at evaluating whether NOTCH1 activation represents a biomarker of ATRA sensitivity in the context of triple negative breast cancer.
The present study unveils a new aspect of ATRA activity on breast cancer epithelial cells (i.e. regulation of phenotypic cell plasticity). Our study supports and revives the long-standing and clinically unexploited idea that the paradigmatic retinoid, ATRA, is endowed with anti-metastatic properties in breast and other tumors (58 -60). The idea needs to be strengthened by further in vivo studies in appropriate models of metastatic spread. In the context of breast cancer, one of the main obstacles is represented by the scarcity of suitable retinoid-sensitive human mammary tumor cell lines that can be xenografted in immunodeficient mice and used to study the metastatic process. Nevertheless, it is worthwhile mentioning that preliminary results obtained with subcutaneous xenografts of HCC1599 cells in SCID mice are consistent with the anti-metastatic potential of ATRA. In fact, intraperitoneal administration of ATRA at dosages reducing the primary tumor mass causes a significant down-regulation of the NOTCH1, SNAIL, and SLUG mRNAs. 3 These in vivo data are in line with the down-regulation of NOTCH1 and SNAIL observed in SKBR3 cells and the recently reported role of SLUG as a mediator of the EMT and metastatic processes activated by NOTCH1 in xenografts of retinoid-insensitive MDAMB231 cells (61). The antimetastatic properties of ATRA can be exploited at the clinical level for the design of retinoid-based therapeutic strategies aimed at the treatment or chemoprevention of breast cancer.