|Home | About | Journals | Submit | Contact Us | Français|
Epithelial-mesenchymal transition (EMT) is implicated in various pathological processes within the prostate, including benign prostate hyperplasia (BPH) and prostate cancer progression. However, an ordered sequence of signaling events initiating carcinoma-associated EMT has not been established. In a model of transforming growth factor β (TGFβ)-induced prostatic EMT, SLUG is the dominant regulator of EMT initiation in vitro and in vivo, as demonstrated by the inhibition of EMT following Slug depletion. In contrast, SNAIL depletion was significantly less rate limiting. TGFβ-stimulated KLF4 degradation is required for SLUG induction. Expression of a degradation-resistant KLF4 mutant inhibited EMT, and furthermore, depletion of Klf4 was sufficient to initiate SLUG-dependent EMT. We show that KLF4 and another epithelial determinant, FOXA1, are direct transcriptional inhibitors of SLUG expression in mouse and human prostate cancer cells. Furthermore, self-reinforcing regulatory loops for SLUG-KLF4 and SLUG-FOXA1 lead to SLUG-dependent binding of polycomb repressive complexes to the Klf4 and Foxa1 promoters, silencing transcription and consolidating mesenchymal commitment. Analysis of tissue arrays demonstrated decreased KLF4 and increased SLUG expression in advanced-stage primary prostate cancer, substantiating the involvement of the EMT signaling events described in model systems.
Epithelial-mesenchymal transition (EMT) is a developmental regulatory program that mediates plasticity between epithelial and mesenchymal phenotypes in normal and pathological settings (32). An irreversible EMT program is a major driver of pathological fibrosis. A reversible EMT program that encompasses a partial EMT phenotype has been implicated as a common mechanism associated with the ability of epithelial cancers to invade, resist apoptosis, and disseminate (25).
EMT plays an important role in various pathological processes within the prostate. There is evidence that benign prostatic hyperplasia (BPH), an enlargement of the prostate characterized by increased stromal and glandular elements, contains the accumulation of mesenchymal-like cells derived from prostate epithelium (4). In addition, EMT is implicated in prostate cancer progression. The presence of individual invasive prostate cancer cells within high-Gleason-grade tumors is a morphological indication that an EMT has initiated. High Gleason scores predict a poor prognosis associated with the development of metastatic disease that is resistant to androgen deprivation therapy (14). Consistent with the occurrence of EMT in progressive prostate cancers, analysis of prostate cancer tissue arrays has shown that loss of the epithelial marker, E-cadherin (CDH1), and gain of the mesenchymal marker, N-cadherin (CDH2), are associated with multiple endpoints of progression and cancer-associated death (16, 18). To begin mechanistically defining pathways leading to invasion and progression, we address here the circuitry of multiple components within the program leading to prostate cancer EMT.
A set of transcription factors that included SNAI1, SNAI2 (SLUG), TWIST, and ZEB1/2 were initially identified as regulating epithelial-mesenchymal plasticity in embryonic morphogenesis and subsequently as suppressing CDH1 expression associated with various forms of EMT (32). However, much of the existing evidence regarding EMT and cancer progression is built upon ectopic expression of EMT regulators or correlative analyses of stable phenotypes in various cell lines or clinical samples. These static approaches have not defined the order or interdependence of EMT regulators present during the initiation of EMT.
During cancer progression, a variety of signaling pathways, originating in the carcinoma cells or the microenvironment, lead in a context-dependent manner to the expression of various combinations of the EMT transcription factors mentioned above. Members of the transforming growth factor β (TGFβ) family are principal extracellular inducers of EMT (37). Although TGFβ functions as a tumor suppressor of the prostate, it can also promote invasion and malignant progression of advanced disease (5, 19). In late-stage prostate cancer, TGFβ expression is elevated in prostate tumors and the circulation, with expression levels negatively correlated with patient prognosis (2, 36).
We use here a mouse model of progressive prostate cancer, initiated by probasin-driven cis-acting replication element (CRE) deletion in the prostate epithelium of the tumor suppressors Pten and TP53 (11, 23). PTEN and TP53 were recently shown to be two of the most common copy number deletions that occur in prostate cancer (31). The prostate Pten/TP53 deletion model is characterized by stem cell features and EMT. Increased stem/progenitor activity is apparent as expanded progenitor self-renewal activity in vitro and as histologically diverse tumor formation (1, 23). In addition, fully penetrant adenocarcinoma (AC) development at 4 months of age progresses by EMT to lethal sarcomatoid carcinoma (SC) by 7 months of age. We have derived a panel of clonal epithelial cell lines from Pb-Cre Ptenfl/fl; Tp53fl/fl tumors that give rise to epithelial, mesenchymal, or mixed-phenotype tumors (23). The isolation of adenocarcinoma 3 (AC3), a clonal cell line capable of undergoing TGFβ-induced EMT in vitro, has allowed us to define the kinetics and regulation of EMT initiation and mesenchymal commitment. For a number of Pb-Cre Ptenfl/fl; Tp53fl/fl clonal epithelial cell lines, sarcomatoid carcinoma formation is observed in vivo (23), but mesenchymal commitment does not occur in vitro. Thus, the AC3 clone appears to recapitulate a common phenotype that is selected during tumorigenesis.
We show here that depletion of Slug but not Snail inhibits TGFβ-initiated prostatic EMT. The expression of KLF4, an epithelium-determining transcription factor, decreases with progression in various epithelial cancers, including prostate cancer (3, 28, 33, 35). Therefore, we investigated the relationship of SLUG, KLF4, and FOXA1, another epithelial transcription factor important for prostate function (15, 22). Interestingly, TGFβ-dependent KLF4 loss and SLUG increases interact in a self-reinforcing loop to establish prostatic EMT and associated phenotypic properties such as invasiveness and sarcomatoid carcinoma transdifferentiation. The data presented here address the ordered regulatory interactions of transcription factors that initiate EMT.
PrEGM media with supplements were from Lonza). BD Matrigel was purchased from BD Biosciences. TGFβ1 was from R&D Systems, and TGFβR1 inhibitor (SB431542) was from Sigma. pLenti6-IRES-GFP-HA-Klf4-2D was constructed as described previously (17). Klf4 and Foxa1 short inhibitory RNA (siRNA) were from Thermo Scientific. All siRNA sequences are listed in Table S1 in the supplemental material. The Snai2 (Slug) knockdown shRNA vector was from Thermo OpenBiosystem. Red fluorescent protein (RFP) reporter vectors were constructed using a Clone-it enzyme-free lentivectors kit (System Biosciences). All primers used for these constructs are listed in Table S2 in the supplemental material. The positions of the FOXA1 and KLF4 binding sites, F1 to F5 and K1 to K5, located upstream of Slug on chromosome 16 are as follows: for F1, 16:14704338; for F2, 16:14704440; for F3, 16:14704601; for F4, 16:14704667; for F5, 16:14704970; for K1, 16:14704904; for K2, 16:14705467; for K3, 16:14705625; for K4, 16:14705638; and for K5, 16:14705930. The positions of E boxes EK1 to EK9 and EF1 to EF2, located upstream of mouse Klf4 and Foxa1, are as follows, with the chromosome indicated first: for EK1, 4:55546869; for EK2, 4:55546807; for EK3, 4:55545842; for EK4, 4:55545743; for EK5, 4:55545244; for EK6, 4:55544831; for EK7, 4:55543556; for EK8, 4:55543278; for EK9, 4:55543182; for EF1, 12:58649540; and for EF2, 12:58646605. E boxes and FOXA1 and KLF4 binding site mutations were made using a site-directed mutagenesis system kit (Invitrogen). All constructs were verified by DNA sequence analysis. Transient transfections were carried out using Lipofectamine 2000 (Invitrogen).
AC1 and AC3 cell lines were isolated from prostate tumors of 4-month-old animals and sarcomatoid carcinoma 1 (SC1) cells from a PbCre4+; Ptenfl/fl; TP53fl/f; lLuc+ prostate tumor of a 7-month-old animal. Cell lines were established from isolated colonies and recloned using limiting dilution as described previously (23). AC1 and AC3 cell lines were cultured in prostate epithelial growth medium (PrEGM), and SC1 cells were cultured in PrEGM supplemented with 10% fetal calf serum (FCS). DU145 and PC3 cell lines were from ATCC and cultured in RPMI 1640 media supplemented with 10% FCS. TGFβ1 measurements by enzyme-linked immunosorbent assay (ELISA) were performed as described previously (30).
Cells were stained with anti-CD24-PE and anti-EpCAM-APC antibodies (BD Biosciences) for 30 min at 4°C. A volume of 7-aminoactinomycin D (Sigma) (100 μg/ml) was added prior to analysis. Fluorescence-activated cytometry (FAC) analysis and cell sorting using FACSDiva software were performed as described previously (1, 23). Promoter functional analysis using FAC and relative median fluorescence intensity (MFI) value determinations were performed as described previously (13).
Adherent cells were fixed in 4% paraformaldehyde (PFA)–phosphate-buffered saline (PBS) for 10 min, followed by permeabilization with 0.5% Triton X-100–PBS for 2 min. Nonspecific sites were blocked by incubation in 2% bovine serum albumin (BSA)–PBS for 30 min. Cells were then incubated overnight at 4°C with the specified antibodies in 2% BSA–PBS. Antibodies and stains used in this study are listed in Tables S3 in the supplemental material. Cells were washed with PBS containing 0.1% Tween 20, incubated with Alexa-488 and/or -568-conjugated IgG–2% BSA for 30 min at room temperature, and finally washed and mounted using the antifade reagent Fluoro-gel II with DAPI (4′,6-diamidino-2-phenylindole). Fluorescent signals and bright-field images were captured using an inverted and/or upright fluorescent Zeiss Axioplan microscope. For cytospun cells, single-cell suspensions were washed twice with PBS, and 1 × 104 cells were deposited on glass slides in PBS by centrifugation at 1,000 rpm for 2 min using a cytospin system (Thermo Shandon, Pittsburgh, PA). Cells were fixed and stained as described above. Using a 63× objective and an upright fluorescent Zeiss Axioplan microscope, at least 300 cells were manually quantified and all numbers were plotted as percentages of total cells counted.
For histological analysis, the urogenital tracts of male mice were removed, fixed in 4% paraformaldehyde–PBS overnight, transferred to 70% ethanol, cut, and embedded in paraffin. Hematoxylin and eosin (H&E) staining was performed (Histoserv Inc.). Immunohistochemistry (IHC) was performed using the antibodies and antigen retrieval methods indicated in Table S4 in the supplemental material. In general, unstained sections were deparaffinized and rehydrated. Antigen retrieval was performed using target antigen retrieval solution (Dako) and steaming at 100°C for 15 min. For IHC, endogenous peroxidase was blocked using a 3% hydrogen peroxide solution. All sections were blocked with Cyto Q Background Buster reagent (Innovex BioSciences). Primary antibodies were incubated overnight at 4°C in antibody diluent with background-reducing components (Dako). Incubation with a secondary anti-mouse/rabbit antibody (Dako) (1:400 horseradish peroxidase [HRP] labeled) was performed at room temperature for 30 min and bound peroxidase detected using an ABC peroxidase kit (Vector Laboratories) and 3,3′-diaminobenzidine tetrahydrochloride (DAB; Dako). All IHC slides were counterstained with hematoxylin. For histomorphometric analysis of tissue sections, 10 bright-field microscopic images were collected in each core using ×200 magnification and an Axioplan microscopy system (Zeiss, Thornwood, NY). Slug- or KLF4-positive tumor cells and total tumor cells were quantified using AxioVision software (Zeiss). The percentage of Slug- or KLF4-positive cells in each core represents an average of 10 measurements.
Total RNA was isolated using a mirVana PARIS RNA isolation system (Ambion). Reverse transcription (RT) of cDNA and PCR were performed as described previously (1, 23). All reactions were normalized to mouse glyceraldehyde-3-phosphate dehydrogenase Gapdh and run in triplicate using the primers listed in Table S5 in the supplemental material. For microRNA analysis, microRNA was isolated using a mirVana Paris RNA isolation system (Ambion). MicroRNA reverse transcriptation and PCRs were performed using a TaqMan MicroRNA assay kit (Applied Biosystems). All values were normalized to a mouse snoRNA202 endogenous control (Applied Biosystems).
Cells were lysed with radioimmunoprecipitation assay (RIPA) buffer containing complete protease inhibitors (Roche) plus the phosphatase inhibitors 25 mM β-glycerophosphate, 10 mM sodium fluoride, and 1 mM sodium vanadate. Primary antibodies were incubated overnight at 4°C using the dilutions listed in Table S6 in the supplemental material, and analysis was performed as described previously (1, 23).
Chromatin immunoprecipitation (ChIP) assays were performed using an EZ magna ChIP A kit (Millipore) with a modified protocol. Cultured cells (1 × 107) were cross-linked with 1% formaldehyde at room temperature for 15 min. The fixation was quenched with glycine, and cells were washed twice with cold PBS containing complete protease inhibitor (Roche). Cell pellets were resuspended in cell lysis buffer and incubated on ice for 15 min. Nuclei were collected by centrifugation at 10,000 rpm at 4°C for 10 min and resuspended in nucleus lysis buffer. Chromatin was sheared using a sonicator (Branson Sonifier 250; Branson, Teltow, Germany) with a microtip in a 20-s burst followed by 1 min of cooling on ice for a total sonication time of 5 min per sample. The procedure results in DNA fragment sizes of 100 to 300 bp. Sheared chromatin was divided to perform immunoprecipitation with rabbit IgG antibody (Santa Cruz Biotechnology) or primary antibody at 4°C overnight. Immunoprecipitation, washing, elution, reverse cross-linking, and DNA purification steps were performed according to Millipore's protocol. Quantitative PCR was performed in triplicate with 2 μl of eluted chromatin. ChIP antibodies and PCR primers are listed in Table S7 in the supplemental material. Predictions for transcription factor binding sites within promoter regions were adopted from the AliBaba 2.1 program.
A confluent culture of cells in a 75-cm2 tissue culture flask was trypsinized using 1 ml of trypsin solution and quenched with 11 ml of RPMI 1640 media supplemented with 10% fetal calf serum (FCS). Following 2 h of recovery, cells were resuspended to a concentration of 106 cells/ml in RPMI 1640 media supplemented with 10% FCS. Cells that invaded the Matrigel-coated transwells in response to 10% FCS after 4, 12, and 24 h were fixed and counted.
Animal work was performed in accordance with a protocol approved by the NIH Animal Care and Use Committee. Male nude mice (NCI, Frederick, MD) (6 to 7 weeks of age) were injected subcutaneously with 1 × 106 tumor cells in 50% Matrigel. Tumor size was measured weekly with calipers, and tumor volume was calculated using the following formula: tumor volume = (4/3π) × (L/2) (W/2)2, where L = length and W = width. Results represent means ± standard errors of the means (SEM) of the data determined for each experimental group. Tumors were radiographed ex vivo using an MX-20 Faxitron X-ray system (Faxitron).
All data were presented as means ± SEM. Statistical calculations were performed with GraphPad Prism analysis tools. Differences between individual groups were determined by one-way or two-way analysis of variance (ANOVA). Bonferroni's posttest was used for comparisons among 3 or more groups. P values of less than 0.05 were considered statistically significant.
Tumors of epithelial adenocarcinoma and tumors of mesenchymal sarcomatoid carcinoma were obtained for histology from prostates of 4- and 7-month-old PbCre-4; Ptenfl/fl; Tp53fl/fl animals, respectively. As shown in Fig. 1A and B, adenocarcinomas are characterized by Cdh1 and Krt18 (encoding cytokeratin 18 [CK18]), and sarcomatoid carcinomas express decreased levels of Cdh1/Krt18 and elevated levels of Vim, Cdh2, and SLUG. Clonal cell lines were established from the two types of tumors, and adenocarcinoma 1 (AC1) and AC3 and sarcomatoid carcinoma 1 (SC1) were characterized for morphology and lineage markers (Fig. 1C). FAC analysis revealed CD24-positive (CD24+) and EpCAM+ (E+) cells in the AC1 epithelial clone but a mixture of CD24+, E+, and E− cells in the AC3 cell line (Fig. 1C). The AC3 cell line was of particular interest, because it appeared to undergo spontaneous EMT in vitro, as shown by the presence of cells with both epithelial and mesenchymal phenotypes. A clonal origin that led to the development of both lineages was shown by the presence of CK8+ and VIM+ progeny in AC3 colonies that developed from single cells plated in individual wells (Fig. 1D). Sorted EpCAM-positive (E+) cells expressed markers characteristic of epithelial cells (Cdh1, Krt 18, and Foxa1), and EpCAM-negative (E−) cells expressed markers characteristic of mesenchymal cells (Vim, Slug, Snail, Zeb1, Zeb2) (Fig. 1C and E). With increasing passages, sorted AC3E+ cells produced a fraction of cells expressing mesenchymal markers (Fig. 1F). Upon subjecting AC3E− cells to passage, no epithelial lineage cells were observed, suggesting irreversible, or type 2, EMT.
To determine whether autocrine TGFβ, a common mediator of EMT in carcinomas, could be responsible for EMT in AC3 cells, TGFβ signaling was assessed. TGFβ was present in AC3 supernatants (Fig. 2A). Blocking TGFβ signaling with SB431542, a small-molecule ALK5 (TGFβ type I receptor) kinase inhibitor, prevented the generation of mesenchymal cells in AC3E+ cultures, as determined by cell surface EpCAM expression (Fig. 2B) and by the expression of other marker proteins (Fig. 2C). Conversely, TGFβ treatment of AC3 cultures increased the number of E− cells (Fig. 2B). We conclude that spontaneous EMT in AC3 cells results from autocrine TGFβ production and engagement of TGFβ receptors. We observed that TGFβ treatment of AC3E+ cells, in addition to stimulating EMT, induced the polycomb repressor complex (PRC) proteins SUZ12 and BMI1 but not EZH2 (Fig. 2D). In contrast to AC3 cells, TGFβ treatment did not lead to morphological transformation of AC1 (Fig. 2E). This is consistent with a previous study that found relatively rare TGFβ-induced morphological transformation in culture for a variety of human and mouse epithelial cell lines, suggesting infrequent genetic selection for the ability to undergo EMT in vitro (7).
Following TGFβ addition to sorted AC3E+ cells, a kinetic analysis of EMT regulators demonstrated increased Slug (Snai2), Snail (Snai1), and Zeb2 expression, while Twist1 and Zeb1 were minimally expressed (Fig. 3A). Slug expression appeared to increase slightly more rapidly than that of the others. Depletion of Slug in AC3E+ cells by the use of lentivirus-encoded small hairpin RNA (shRNA) (referred to here as AC3E+SKD) resulted in inhibition of spontaneous and exogenous TGFβ-driven EMT, as shown by continued expression of EpCAM (Fig. 3B) as well as other epithelial but not mesenchymal markers (Fig. 3C). The specificity of Slug depletion was shown by the reconstitution of the mesenchymal AC3 phenotype following Slug reexpression in Slug-depleted cells (Fig. 3C). Depletion of Slug expression inhibited TGFβ-induced Snail and Zeb1/2 expression, consistent with increased Slug expression being necessary for initiation and/or maintenance of EMT (Fig. 3D). In contrast, in serially passaged AC3E− cells, Slug depletion induced some Cdh1 but did not inhibit other EMT regulators such as Snail and Zeb2, suggesting that SLUG regulated the initiation of EMT but not maintenance of established mesenchymal commitment (Fig. 3E).
In contrast to depletion of Slug, Snail-depleted AC3E+ cells (referred to here as AC3E+SnKD) still underwent TGFβ-induced EMT, albeit less efficiently than control AC3 cells, as determined by the increase in levels of EpCAM-negative cells (Fig. 3B). Following TGFβ treatment of AC3E+SnKD and AC3E+ cells, the epithelial (E+) and mesenchymal (E−) populations were isolated and analyzed (Fig. 3F and G). The E− fraction of TGFβ-treated Snail depleted cells showed loss of CDH1 and increased VIM, demonstrating EMT. Importantly, Snail RNA and SNAIL protein were expressed at minimal levels in both the epithelial and mesenchymal fractions of TGFβ-treated AC3E+SnKD, demonstrating efficient Snail depletion (Fig. 3F and G). Despite reduced SNAIL levels, the patterns and levels of Slug expression in epithelial and mesenchymal fractions for TGFβ-treated AC3E+ and AC3E+SnKD cells were similar. These data suggest that SLUG expression occurs independently of SNAIL but that Snail induction requires the initiation of EMT, which is dependent on the presence of SLUG. SLUG is an essential regulator of EMT, whereas analysis of several markers (Fig. 3F and G) suggests that SNAIL is less rate limiting.
KLF4 and FOXA1 are transcription factors that have been shown to be mediators of epithelial determination in induced pluripotent stem (iPS) cells and pancreatic carcinomas, respectively (20, 26, 29). Consistent with this, we observed inverse relationships of SLUG to KLF4 and FOXA1 levels in AC1, AC3E+, AC3E−, and SC1 cells (Fig. 4A). Following TGFβ treatment of AC3E+ cells, we observed rapid loss of Klf4 RNA and gradual loss of Foxa1 RNA (Fig. 4B). In addition, levels of KLF4 and FOXA1 proteins were significantly reduced after 24 and 72 h, respectively (Fig. 4C). Interestingly, the rapid TGFβ-dependent loss of KLF4 occurred in SLUG-depleted AC3 cells, suggesting that KLF4 loss was independent of SLUG accumulation. Consistent with the slower kinetics of its TGFβ-induced loss, FOXA1 loss appeared to be dependent upon initiation of EMT, as evidenced by continued FOXA1 expression in Slug-depleted cells (Fig. 4B and C). Increased Slug expression beginning between 12 and 24 h after TGFβ treatment approximately correlated with loss of Klf4 RNA and protein (Fig. 4D). TGFβ-dependent degradation of KLF4 has been previously demonstrated to occur through Cdh1-anaphase promoting complex-dependent ubiquitylation of KLF4 at two destruction box sites (17). To determine whether KLF4 degradation was necessary for TGFβ-induced increases in Slug levels, a KLF4 mutant with deleted destruction boxes (KLF4ΔD) was permanently introduced into AC3 cells. The expression of KLF4ΔD inhibited TGFβ-induced EMT, as assayed by the absence of EpCAM-negative cells (Fig. 4E) and the absence of SLUG protein expression (Fig. 4F). Consistent with this, in PC3 human prostate cancer cells, which grow as loosely organized spindle-shaped cells, infection with KLF4ΔD lentivirus resulted in an mesenchymal to epithelial (MET) process whereby compacted epithelial colonies were formed (Fig. 4G).
To determine whether depletion of Klf4 was sufficient to initiate EMT, AC3E+ cells were treated with SB431542 to inhibit spontaneous EMT and subsequently transfected with siRNA directed against Klf4 (Fig. 4H). Klf4 depletion initiated EMT, as shown by the accumulation of EpCAM-negative cells (Fig. 4H) and by increased mesenchymal and decreased epithelial RNA markers (Fig. 4I). Because EMT occurred in the presence of SB431542, these data imply that Klf4 depletion was sufficient to initiate EMT, although, as a result of autocrine TGFβ secretion, prior TGFβ priming might have occurred. Unexpectedly, depletion of Foxa1 led to the accumulation of mesenchymal cells (Fig. 4H), most likely because forced loss of FOXA1 led to increased Slug levels (Fig. 4J). Neither Klf4 nor Foxa1 depletion led to EMT in the absence of Slug, confirming that SLUG accumulation is required for EMT (Fig. 4H). We conclude that TGFβ-induced loss of KLF4 is necessary for increased Slug expression and EMT in AC3E+ cells.
To analyze whether KLF4 and FOXA1 directly regulate Slug transcription, potential KLF4 and FOXA1 binding sites were mapped in the Slug promoter/enhancer shown in Fig. 5A. In AC3E+ cells, ChIP analysis identified two KLF4 (K1 and K2) (Fig. 5B)- and two FOXA1 (F1 and F5) (Fig. 5C)-bound regions within the Slug promoter. KLF4 and FOXA1 ChIP performed on AC3E+ cells treated with SB431542 demonstrated binding to the K1, K2, F1, and F5 binding sites that was enhanced more than 5-fold compared to that seen with TGFβ-treated AC3E+ cells or AC3E− cells (Fig. 5D). These data are consistent with KLF4 and FOXA1 acting as transcriptional repressors of Slug in epithelial cells. To determine whether KLF4 and/or FOXA1 were able to directly repress Slug transcription, the Slug upstream region between −1721 and +201 was cloned in front of an RFP reporter and is referred to here as Slug regulatory sequence-RFP (SRS-RFP). Following TGFβ treatment of SRS-RFP-transfected AC3E+ cells, significant RFP accumulation was seen between 24 and 72 h, demonstrating TGFβ-dependent activation of the Slug promoter/enhancer (Fig. 5E). SRS-RFP was cotransfected with expression vectors carrying Klf4 and/or Foxa1 into AC3E− cells. As shown in Fig. 5F, SRS-RFP was highly expressed in AC3E− cells, and the Slug promoter-enhancer activity was inhibited by ectopically expressed KLF4 and FOXA1. Conversely, the effect upon SRS-RFP activity of depleting Klf4 or Foxa1 in AC3E+ cells was assessed in the epithelial fraction by analyzing RFP in EpCAM+ cells (Fig. 5G). Increased RFP in the EpCAM+ fraction showed that the Slug transcriptional reporter activity increased in epithelial cells depleted for Klf4 or Foxa1. Consistent with this, mutation of KLF4 or FOXA1 binding sites within the Slug promoter construct also led to increased reporter activity in AC3E+ transfected cells (Fig. 5H). Finally, individual wild-type or mutated KLF4 (K2) or FOXA1 (F1) or adjacent KLF4/FOXA1 (K1/F5) binding elements were cloned into a promoter-RFP reporter and transfected into AC3E+ cells. The presence of K2, F1, or K1/F5 elements led to decreased RFP accumulation, and treatment with TGFβ led to increased expression of all 3 reporter constructs (Fig. 5I). The presence of mutant binding elements mimicked the empty vector. Taken together, these data demonstrate that KLF4 and FOXA1 binding sites in the Slug promoter function to suppress transcription, the inhibitory role of which is reversed following exposure to TGFβ.
Reprogramming the epithelial-to-mesenchymal phenotype is expected to involve epigenetic inhibition of epithelial genes and, potentially, positive reinforcement of mesenchymal gene expression. Because SLUG accumulation is central to EMT in AC3 cells, we asked whether SLUG is involved in inhibiting epithelium-determining genes, specifically Klf4 and Foxa1. E boxes in the Klf4 and Foxa1 upstream regions were mapped (Fig. 6A and B). SLUG-specific ChIP analysis of ACE+ cells treated with TGFβ established binding to 4 of 9 chromatin sites in the Klf4 promoter/enhancer (EK2, EK3, EK7, and EK8) (Fig. 6C) and 2 of 2 sites in the FOXA1 promoter/enhancer (EF1 and EF2) (Fig. 6D). Parallel ChIP analyses performed with antibodies directed against polycomb repressive complex (PRC) proteins SUZ12, EZH2, and BMI1 and against the trimethylated lysine 27 modification of histone H3 identified a binding distribution identical to that seen with SLUG. Further ChIP analyses of the identified SLUG binding regions in AC3E+ cells treated with SB431542 showed no binding of SLUG or PRC proteins to specific E boxes in Klf4 (Fig. 6E) or Foxa1 (Fig. 6F), verifying that both SLUG binding and PRC binding correlated with the inhibition of KLF4 or FOXA1 expression observed in mesenchymal but not epithelial phenotypes. In addition, there was no binding of PRC complexes in Slug-depleted cells and significantly reduced trimethylated lysine 27 histone H3, demonstrating that PRC recruitment and inhibitory histone marks were dependent upon SLUG expression (Fig. 6E and F). Thus, we demonstrate here an apparently self-reinforcing regulatory circuit leading to mesenchymal commitment. TGFβ treatment stimulates loss of KLF4 and increased SUZ12 and BMI1 expression. Loss of KLF4 results in increased SLUG expression, leading to SLUG-dependent PRC binding to and inhibition of Klf4 and Foxa1 promoters.
To test whether parallel TGFβ, Klf4, Foxa1, and Slug regulation occurs in human cells, PC3 and DU145 cells were used. Following TGFβ treatment for 7 days, decreased KLF4 and FOXA1 protein expression was observed (Fig. 7A). We determined whether depletion of Klf4 was sufficient to initiate EMT in human prostate cancer cells as seen with AC3 cells. Klf4 depletion initiated EMT, as shown by increased mesenchymal and decreased epithelial protein marker levels (Fig. 7B and C) and by morphological changes (Fig. 7C). In addition, loss of KLF4 led to increased invasive function (Fig. 7D). To analyze whether KLF4 and FOXA1 directly regulate Slug transcription, potential KLF4 and FOXA1 binding sites were mapped in the human Snai2 (Slug) promoter/enhancer as shown in Fig. 7E. In PC3 and DU145 cells, ChIP analysis identified two KLF4 (hK1 and hK2)- and two FOXA1 (hF1 and hF2)-bound regions (Fig. 7F). Following treatment with TGFβ, ChIP analyses demonstrated loss of binding to the hK1, hK2, hF1, and hF2 sites (Fig. 7F). These data are consistent with KLF4 and FOXA1 acting as transcriptional repressors of human Slug in prostate cancer cells. To determine whether KLF4 or FOXA1 or both are able to directly repress Slug transcription in human cells, a human Slug upstream region (between −4153 and +1) reporter was constructed (referred to here as hSRS-RFP). hSRS-RFP lentivirus was coinfected with expression vectors carrying Klf4 and/or Foxa1 in DU145 and PC3 cells, resulting in suppression of Slug promoter-enhancer activity (Fig. 7G), consistent with AC3 cell results. Taken together, these data demonstrate that loss of KLF4, an inhibitor of human Slug transcription, is sufficient to enhance EMT and invasiveness in human prostate cancer cells.
We have addressed the influence of EMT regulators on tumorigenesis. AC3E+ cells injected subcutaneously formed tumors with mixed epithelial and mesenchymal histologies (Fig. 8A). AC3E+ cells are not unusual in this respect, in that other Pb-Cre Ptenfl/fl; Tp53fl/fl clonal cell lines, including AC1, have been observed to form tumors that lost epithelial markers (CDH1 and CK8) and gained SLUG+ mesenchymal components over time (Fig. 8B) (23). AC3 tumors generally demonstrated areas of chondrosarcomas and bone-producing osteosarcomas, implying differentiation of the mesenchymal component (Fig. 8A and C). Slug-depleted (SKD) AC3E+ cells produced adenocarcinomas with high KLF4 expression and produced bone elements only rarely, consistent with a loss of EMT potential (Fig. 8). These data imply that SLUG expression is not required for the growth of prostate epithelial tumors but is necessary for EMT and sarcomatoid carcinoma development in vivo.
To determine whether KLF4 and/or SLUG protein levels correlate with prostate cancer progression, we quantified KLF4 and SLUG with immunohistochemical staining of primary prostate tumor tissue arrays (Fig. 9A). As shown in Fig. 9B, the percentage of KLF4+ tumor cells decreased and the percentage of SLUG+ tumor cells increased in pathologically graded stage III and stage IV tumors (defined as tumors in which invasive tumor cells had spread beyond the prostate). In addition, for individual samples, decreased KLF4 was significantly (P ≤ 0.0001) correlated with increased SLUG. These data support a linked role for KLF4 loss and increased SLUG levels in prostate cancer invasiveness and progression. A model showing the regulatory interactions of Slug, Klf4, and Foxa1 relative to the development of EMT is shown in Fig. 9C.
We describe here an ordered, self-reinforcing, TGFβ-initiated signaling circuitry leading to EMT in mouse and human prostate cancer (Fig. 9C). TGFβ-dependent loss of KLF4 was sufficient to initiate Slug induction and EMT (Fig. 4). KLF4 and FOXA1 are epithelium-determining transcription factors that are shown here to directly suppress Slug transcription (Fig. 5). Slug appears to be the “gatekeeper” of EMT in this model, since depletion of Slug not only inhibits EMT but also inhibits expression of a variety of other EMT regulators such as Zeb2 and Snai1 (Fig. 3). In contrast, EMT occurs, although less efficiently, following Snai1 depletion. SLUG is expressed in normal prostate basal cells, suggesting that SLUG expression in tumor cells may represent a developmentally acquired competency. Finally, for selected E boxes in the promoters of the epithelial genes Klf4 and Foxa1, there exists SLUG-dependent PRC1/2 binding and transcriptional inhibition, strengthening continued Slug expression and commitment to the mesenchymal lineage (Fig. 6).
Importantly, the EMT circuitry discovered in the AC3 cell line appears relevant for human prostate cancer cell lines and clinical prostate cancers (Fig. 7 and and9).9). In tissue arrays, decreased KLF4 and increased SLUG staining in primary prostate cancer relative to that seen with normal glands was observed in grade III and IV metastatic tumors that had demonstrated progression to metastasis. In addition, for individual samples, decreased KLF4 was significantly correlated with increased SLUG. The results of a recent gene-profiling meta-analysis and tissue array study concluding that KLF4 is downregulated in prostate cancer tissue with metastases support the KLF4 data presented here (33).
TGFβ is a common inducer of EMT (37). Although many changes in gene expression occur in association with EMT, the transcriptional mechanisms that lead to EMT initiation have not been clearly established (32). We show that TGFβ treatment resulted in rapid loss of KLF4, leading to induction of Slug, the EMT regulatory transcription factor. KLF4 depletion was sufficient to initiate Slug-dependent EMT; in contrast, the inhibition of EMT occurred following stabilization of KLF4 protein secondary to inhibiting KLF4 ubiquitylation and proteosomal degradation (Fig. 4). Previous studies have demonstrated that TGFβ-dependent proteosomal degradation of KLF4 occurs in the absence of new protein synthesis, suggesting a direct effect (17). Thus, KLF4 loss provides a mechanistic link between TGFβ action and the known EMT regulator SLUG.
TGFβ-dependent transcriptional inducers of Slug have been described. In MDCK cells, TGFβ-initiated myocardin-related transcription factor (MRTF)-SMAD3 complex regulation was previously demonstrated (24). In Panc1 cells, TGFβ induced SMAD3-IκB kinase 1 complex (SMAD3-IKK1) binding to the SLUG promoter, and depletion of IKK1 inhibited SLUG expression (6). Taking these data together, TGFβ-initiated Slug transcription appears to be regulated in a context-specific manner by repressors and inducers, at least some of which are dependent on the presence of SMAD. Interestingly, KLF4 has been shown to act as an SMAD corepressor in reporter assays (17).
KLF4 is a pleiotropic regulator that influences growth, epithelial differentiation, and invasion in normal and tumor cells (20, 26). KLF4 can function as either an inductive or inhibitory transcription factor acting upon, for example, various cell cycle regulatory genes or differentiation genes, including CDH1 (9, 10, 27, 38). Loss of KLF4 in various cancers, including colon, breast, and gastric cancers, is consistent with tumor-suppressive functions (3, 28, 35). Ectopic overexpression of KLF4 inhibits growth and invasiveness of tumor cell lines, including prostate cancer (12, 33, 35, 39). The pleiotropic potential of KLF4 suggests that it may be an integrator of growth- and lineage-dependent differentiation functions.
Similarly to KLF4, FOXA1 is an epithelial determinant and, as shown here, functions as a direct transcriptional repressor of Slug. In turn, Foxa1 expression is transcriptionally inhibited by increased SLUG binding, which occurs in combination with PRC1/2 complexes. In the prostate, FOXA1 is necessary for luminal ductal epithelial differentiation and functions as a required cofactor for androgen receptor binding to chromatin (15, 22). In addition, FOXA1 is a transcriptional inducer of CDH1, a positive regulator of the epithelial phenotype (21, 38). Unlike those of Klf4, Foxa1 mRNA and protein levels decline gradually following TGFβ treatment, and Foxa1 downregulation is dependent on the presence of Slug. Thus, loss of FOXA1 most likely plays a role in the later phases of transitioning to the mesenchymal state. These data further suggest that FOXA1 and KLF4 may functionally synergize in regulating partial or full EMT.
We anticipate that the EMT circuitry described here may be particularly relevant to TP53-deficient tumors. Deletion of TP53 is one of the most common copy number variations observed in human prostate cancer (31). Basal and induced SLUG levels are expected to increase in TP53-deficient cells. TP53 induces SLUG degradation by upregulating MDM2 expression and participating in a TP53-MDM2-SLUG complex that facilitates MDM2-mediated SLUG degradation (34). Loss of Tp53 in certain epithelial tumor models is associated with a propensity to form sarcomatoid carcinomas (8).
In conclusion, we have used a genetically defined Pten/TP53 null mouse model of prostate cancer and human prostate cancer cells to show that EMT initiated by TGFβ proceeds by a self-reinforcing feedback loop between decreased KLF4 and increased SLUG levels. The data suggest that SLUG and KLF4 should be explored further as potential diagnostic and prognostic markers of invasive prostate cancer. We anticipate that this defined circuitry will provide insight for investigating the regulation of EMT phenotypes that are proposed to function during tumor progression.
We acknowledge the support of the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research.
We indicate no potential conflict of interest.
Published ahead of print 27 December 2011
Supplemental material for this article may be found at http://mcb.asm.org/.