Abstract
Grainyhead genes are involved in wound healing and developmental neural tube closure. In light of the high degree of similarity between the epithelial–mesenchymal transitions (EMT) occurring in wound-healing processes and the cancer stem cell–like compartment of tumors, including TGF-β dependence, we investigated the role of the Grainyhead gene, Grainyhead-like-2 (GRHL2) in oncogenic EMT. GRHL2 was downregulated specifically in the claudin-low subclass breast tumors and in basal-B subclass breast cancer cell lines. GRHL2 suppressed TGF-β–induced, Twist-induced or spontaneous EMT, enhanced anoikis sensitivity, and suppressed mammosphere generation in mammary epithelial cells. These effects were mediated in part by suppression of ZEB1 expression via direct repression of the ZEB1 promoter. GRHL2 also inhibited Smad-mediated transcription and it upregulated mir-200b/c as well as the TGF-β receptor antagonist, BMP2. Finally, ectopic expression of GRHL2 in MDA-MB-231 breast cancer cells triggered an MET and restored sensitivity to anoikis. Taken together, our findings define a major role for GRHL2 in the suppression of oncogenic EMT in breast cancer cells. Cancer Res; 72(9); 2440–53. ©2012 AACR.
Introduction
The oncogenic epithelial–mesenchymal transition (EMT) is a transcriptional reprogramming event that endows restricted subclasses of tumor cells with enhanced metastatic and stem cell–like properties. These properties include increased migration/invasion, chemo- and radiation resistance, anoikis resistance, and extraordinary tumor initiation frequency, a cancer stem cell–like capability (1, 2). In breast cancer, 2 subclasses of tumors, “claudin-low” and “metaplastic”, exhibit frank EMT-like gene expression signatures. These are among the most aggressive and least treatment-responsive tumors (3).
In general, TGF-β signaling can either promote or suppress tumorigenicity and progression (4, 5). It can suppress tumors through Smad-mediated induction of cyclin-dependent kinase inhibitor genes such as p15, and inactivating mutations in this pathway occur in certain tumors. In other contexts, however, TGF-β can promote tumor progression by supporting oncogenic EMT induction, through both transcriptional and nontranscriptional mechanisms. For example, in claudin-low and metaplastic mammary tumors—as well as in the stem cell–like CD44high/CD24low tumor cell subpopulation fractionated from breast tumors of other subclasses—TGF-β pathway components are strikingly upregulated and, indeed, TGF-β receptor kinase inhibitors partially revert the EMT-related gene expression profile (3, 6). This indicates a protumorigenic role of the TGF-β pathway in this context.
Cell culture models confirm these conclusions. An extensively characterized SV40 large T/hTERT-immortalized mammary epithelial cell line, HMLE, exhibits spontaneous EMT in a subpopulation of CD44high/CD24low cells (7). This EMT is accompanied by upregulation of autocrine TGF-β signaling, mainly through downregulation of antagonists such as BMP2/4. Reversion to an epithelial phenotype could be achieved by inhibition of this signaling pathway. Moreover, even Twist-induced EMT in this system was partially dependent upon autocrine TGF-β signaling, showing its functional significance in multiple contexts.
The similarities between oncogenic and wound-healing–related EMT, including TGF-β dependence, have been documented (8–10). In this light, we hypothesized that perhaps other “dedicated” wound-healing genes might prove important in oncogenic EMT. Grainyhead family genes have been shown to play an important role in wound healing, epidermal integrity, and the mechanistically related process of embryonic neural tube closure (11–14). Relevant, albeit limited, Grainyhead family target genes identified so far include E-cadherin, claudin-4, desmoglein-1, transglutaminase-1, rho-GEF19, several Zelda-targeted genes expressed during the maternal–zygotic transition (MZT) in Drosophila, and telomerase (11, 12, 14–17). With regard to cancer, Grainyhead-like-2 (GRHL2) gene amplification has been noted in several tumor types, including breast cancer, and suppressed death receptor expression, conferring resistance to apoptosis mediated by the corresponding ligands, indicating that GRHL2 was a potential oncogene (18). On the other hand, GRHL3 was recently shown to suppress squamous cell carcinoma, due to its activation of PTEN expression (19).
Inspired by its role in wound healing, we hypothesized and report here that GRHL2 suppressed EMT mediated by the TGF-β signaling pathway. Consistent with this effect, GRHL2 was downregulated specifically in EMT-dependent mammary tumors (claudin low, metaplastic) and cell lines (basal-B). ZEB1 was found to be required for EMT and was a direct target for repression by GRHL2. GRHL2 also enhanced anoikis sensitivity. These data suggest an EMT-suppressive function of GRHL2 that is downregulated in the context of TGF-β/EMT-driven tumor types.
Materials and Methods
Cell lines
HMLE, HMLE + Twist-ER, and HMLE + Ras cells (HMLER) were generous contributions from R. Weinberg (The Whitehead Institute, Cambridge, MA). HMLE and HMLE + Twist-ER were grown in a 1:1 mixture of Mammary Epithelial Growth Medium (MEGM; Lonza) and [Dulbecco's Modified Eagle's Medium (DMEM): Ham's F-12 (Gibco) + 5% horse serum + 1× penicillin-streptomycin-glutamine (PSG) + 10 μg/mL insulin, 10 ng/mL EGF, 0.5 μg/mL hydrocortisone], where indicated, 4-hydroxytamixofen (10 ng/mL) was added to the HMLE + Twist-ER cells to activate the Twist-ER protein. HMLER cells were grown in MEGM. MDA-MB-231LN were provided by E. Pugacheva (West Virginia University) and were grown in advanced DMEM (Invitrogen) + 10% FBS + 1× penicillin–streptomycin–glutamine (PSG).
Generation of stable cell lines by retroviral transduction
Human GRHL2 was amplified from a template purchased from Open Biosystems (MHS4426-99625903) and subcloned by standard molecular biology methods into the pMIG or MSCV-IRES-puro retrovirus (Addgene; XhoI site). Retroviruses were packaged and amplified in GP2 + 293T cells by transfection of 4.5 μg of retroviral plasmid and 2.5 μg of pCMV-VSV-G per 60 mm2 dish of cells, with Mirus TransIT reagent. Viral supernatants were collected 48 hours later, filtered through 0.45-micron filters (Whatman) and 0.6 mL of supernatant was used to infect 1 well of a 6-well dish of target cells by centrifugation at 1,400 rpm for 1 hour followed by 6 hours to overnight incubation. Infected cells were either selected for puromycin (2 μg/mL for HMLE, 0.5 μg/mL for MDA-MB-231) or flow sorted for GFP, followed by Western blot analysis to verify expression.
Generation of stable cell knockdown cell lines by lentiviral transduction and transient knockdown (siRNA)
GRHL2, ZEB1, and scrambled control short hairpin RNA (shRNA) were purchased from Open Biosystems in the pGIPZ vector. (catalog numbers: shGRHL2a: RHS4430-99291384; shGRHL2b: RHS4430-99614394 and shZEB1a: V3LHS_356186; shZEB1b: 3 V3LHS_356187; the latter were also subcloned into vector pTRIPz using Mlu1/Xho fragments). SiZEB1 Smartpool was from Dharmacon catalog number (L-006564-01-0005) and was transfected transiently using Lipofectamine RNAi-max (Invitrogen).
Mammosphere assay
Mammospheres were seeded at 1 × 104 cells per well of a 6-well Ultra Low Cluster Plate (Corning) and grown for 7 to 10 days in the appropriate growth media + 0.5% methylcellulose (MC). Wells were fed every third day with 1 mL media + MC. Total mammospheres per well were counted and the size cutoff value was set at 150 mm in diameter; the same cells were plated at 2 × 105 cells per well of a 6-well plate and the number of cells was counted each day for 4 days, to measure normal growth rate. Error bars represent the SD of replicates.
Anoikis assays
Cells were dissociated using TrypLE Express (Invitrogen), counted and a fixed amount 1 × 105 to 2 × 105 cells per well of a 6-well Ultra-Low Attachment Cluster Dish (Costar) were suspended in normal growth media + 0.5% MC for the indicated for 6 to 24 hours to induce anoikis. For cell death ELISA (CDE) analysis of apoptotic DNA fragmentation, the cells were collected in 3 volumes of media and then spun down at 1,500 rpm for 3 minutes. The pellet was then washed with D-PBS (Invitrogen) transferred to a microfuge tube, pelleted at 7,000 rpm for 15 seconds, and lysed in PBS + 0.5% Triton X-100 + 10 mmol/L EDTA (note: the lysis buffer with the Cell Death ELISA kit from Roche was found to lyse cells inadequately). Lysates were incubated on ice for 15 minutes with occasional mixing and then centrifuged at 13,000 rpm for 12 minutes. The supernatants were subjected to the CDE according to the manual provided with the kit (Roche). Alternatively, percentage of cell death was determined by a Trypan blue exclusion assay, wherein cells were suspended in same manner but collected and resuspended in Accumax (Innovative Cell Technology) to ensure a single-cell suspension. After brief incubation (2–3 minutes), Trypan blue was added to this solution and the percentage of dead cells were immediately counted on a hemocytometer. All samples were analyzed in either duplicate or triplicate, and time-zero cell death values were subtracted from the data presented here.
Three-dimensional culture
Three-dimensional (3D) Matrigel culture methods were adapted from (20). To summarize, 2.5 × 103 cells per well were seeded onto 8-well chamber slides (Labtek) where 45 μL Matrigel (Trevigen) had been evenly distributed. The cells were overlaid with the appropriate growth media (see above) + 2.5% Matrigel. After 6 days in culture 3D migration/invasion was quantified by counting the number of structures that had formed protrusions versus those that grew as lobular structures defined by their contact with matrix (see images for clarification). At least 200 structures were counted per experiment; error bars represent the SD across 3 samples.
Microarray methods
RNA isolated by RNeasy Plus kit (Qiagen) was quantified using Nanodrop (Fisher Scientific). The RNA quality was check on Bioanalyzer (Agilent). Two hundred fifty nanograms of each RNA sample with an RNA integrity number (RIN) value greater than 7 was processed by the Ambion WT Expression Kit according to the manufacturer's instructions. The typical yield from the reaction was 6 to 9 micrograms of cDNA. The required amount of cDNA (5.5 μg) was processed for fragmentation and biotin labeling by the GeneChip WT Terminal Labeling Kit (Affymetrix). The efficiency of fragmentation reaction was checked via Agilent Bioanalyzer. The entire reaction of fragmented and biotin-labeled cDNA (50 μL) with added hybridization controls was hybridized to the human GeneChip 1.0 ST Exon Arrays (Affymetrix) at 45°C for 17 hours in GeneChip Hybridization Oven 640 (Affymetrix). Human GeneChip 1.0 ST Exon Arrays were stained with FS 450_0001 protocol in Affymetrix GeneChip Fluidics Station 450. Briefly, biotin-labeled cDNA was reacted using 2 rounds of washes with a solution containing a streptavidin-phycoerythrin complex, with an intermediate treatment of biotin-labeled antistreptadvidin antibody to amplify the signal. Phycoerythrin labeling was detected within the Affymetrix GeneChip Scanner 3000 7G plus using 532 nm light and detected by a photomultiplier tube. Expression Console software (Affymetrix) was used to check quality controls of hybridized chips. All chips that passed quality controls were RMA normalized with Expression Console software. The microarray data were deposited into the NCBI GEO database as accession number GSE36081.
To examine the extent to which GRHL2 affects the propensity to undergo epithelial–mesenchymal transition (EMT), we compared the relative expression of genes in an identified EMT signature (21) to the relative expression of genes in cells with constitutive GRHL2 expression. Specifically, we obtained the expression profile of the 251 Core EMT signature genes from Supplementary Table S1 (21) and computed the mean log ratio of the relative expression. We restricted the genes to those that appeared on our array platform and computed the Pearson correlation coefficient of those genes to the log expression ratio of GRHL2-regulated genes compared with the control.
Reporter assays
HMLE were transiently transfected using Lipofectamine 2000 (Invitrogen) at a 1 μg DNA:2 μL Lipofectamine ratio. A total of 1.5 μg DNA per well of a 12-well plate was the maximum amount of DNA found to be tolerable. Transfection mixtures were incubated for 20 minutes in 200 μL Opti-MEM (Invitrogen) and then added to the cells in normal growth media lacking antibiotics. Cells were incubated for 4 hours then refed with normal growth media. Lysates were made by washing the cells once with PBS then lysing in 1× Cell Culture Lysis Buffer (Promega). Lysates were centrifuged at 13,000 rpm for 10 minutes, and the supernatants were assayed for luciferase and β-galactosidase activity as internal control. Luciferase assay reagent was obtained from Promega and the β-galactosidase 2× assay reagent was 200 mmol/L sodium phosphate (pH 7.3), 2 mmol/L MgCl2, 100 mmol/L 2-mercaptoethanol, and 1.33 mg/mL o-nitrophenylgalactoside. The Smad reporter construct 3TP-Lux was from Addgene. The ZEB1 promoter-luciferase construct in pGL3 (22) was kindly provided by Antonio Garcia de Herreros (Universitat Pompeu Fabra, Barcelona, Spain). CMV-LacZ or TK-LacZ were used as internal controls. The GRHL2 clone was purchased from Open biosystems, catalog no MHS4426-99625903, and the coding sequence was cloned into the XhoI site of pcDNA3.1+. Subfragments of the ZEB1 promoter were generated and cloned into pGL3-promoter (MluI-BglII) using the following primers: fragment 1: ZPfr1-f: Ttaat ACGCGT CCTTAAGGTCCTGCACGGCG, ZPfr1-r: tatat agatct AAGTTCCGCTTGCCAGCAGC; fragment 2: ZPfr2-f: TtaatACGCGT CTAGCCTCTCTTTCAATCCA, ZPfr2-r: tatatagatctTCCGCCCCCCGCACCCCGGGGC; fragment 3: ZPfr3-f: TtaatACGCGTCACGCGAGGCGTGGGACTGA, ZPfr3-r: tatat agatct GGATGCCGGGAAACCGTAGG; fragment 4: ZPfr4-f: Ttaat ACGCGT GCCTCCCTCTCCCCACCACA, ZPfr4-r: tatat agatct ACCGCACCTGGTTTACGACA; fragment 5: ZPfr5-f: TtaatACGCGT GCGGACCGGGTGTGGGAGGC, ZPfr5-r: tatat agatct TACCTGACCCGCGCAGCCCG; fragment 6: ZPfr6-f: Ttaat ACGCGT GCCTCTCTCCGGTCGCCGCG, ZPfr6-r: tatatagatct GGGGCAGGGAGGGATCTGGC, fragment 7: ZPfr7-f: Ttaat ACGCGT GGGCAGCCGCGGCGGGTGTG, ZPfr7-r: tatat agatct ACCGTGGGCACTGCTGAATT.
The primers used to mutate the fragment 4 construct (using Stratagene Quickchange II XL kit) were: Mut-f, gtaaagccgggagtgtcgtcccacaggtgcggtagatctgcg; and Mut-r, cgcagatctaccgcacctgtgggacgacactcccggctttac.
Immunofluorescence
For Smad2 localization, TGF-β (5 ng/mL) was added for 6 hours and the coverslips were fixed with 4% paraformaldehyde (PFA) in PBS for 10 minutes. PFA was quenched with 100 mmol/L glycine in PBS. Cells were permeabilized with 0.2% Triton X-100 in PBS at 4 degrees for 10 minutes, washed twice with PBS, and blocked for 1 hour in: PBS + 10% goat serum + 0.1% Tween-20 + 0.1% BSA. Primary and secondary antibodies were diluted in blocking buffer. Primary antibodies were as follows: SMAD2; Cell Signaling, rb, 1:200. Secondary: rb Alexa 555 (red); Molecular Probes, 1:1,000. Mounting media: Prolong Gold w/DAPI (Invitrogen).
For E-cadherin, vimentin, and GRHL2 the cells were fixed in 100% methanol at −20°C for at least 1 hour. They were then washed twice with PBS and blocked as earlier. E-cadherin, ms (BD Biosciences), 1:200; Vimentin, rb (Cell Signalling), 1:200; GRHL2, rb (Sigma), 1:200. Secondaries used were anti-mouse Alexa 555 (red) or anti-rabbit Alexa 488 (green) or A555 (red); Molecular Probes, diluted 1:1,000. Coverslips were mounted in Prolong Gold as above. Images were produced with the Axiovert 200M microscope, AxioCam MRM camera, and Axio Vision 4.3.1 software (Zeiss).
Chromatin immunoprecipitation
A total of five 100-mm2 dishes of 4-hydroxytamoxifen (4-OHT)–induced HMLE + Twist-ER (+GRHL2, in some experiments) were each fixed in 1.2 mL 10% electron microscopy grade PFA for 10 minutes. Following quenching with glycine, chromatin immunoprecipitation (ChIP) was conducted exactly as described previously (23) with the following antibodies, 3.3 μg each: GRHL2 (Sigma); Histone H3 (Cell Signaling), or nonimmune rabbit IgG (Pierce). ChIP-derived DNA was analyzed by PCR using the following primer sets: ZEBf6, GCGAGGCGTGGGACTGATGG; ZEBr6, AAAGTTGGAGGCTCGGCGGC; ZEBf10, CTGCACGGCGATGACCGCT; ZEBr10, TTCCGCTTGCCAGCAGCCTC; GAPDH-f, ATGGTTGCCACTGGGGATCT; and GAPDH-r, TGCCAAAGCCTAGGGGAAGA. For quantitative PCR analysis, ChIP DNAs were analyzed using 2× Power-Sybr Green Master Mix (Applied Biosystems) and signals were calculated relative to input DNA using the “ΔΔCt” method; a correction of 1.5× was applied to adjust for the higher efficiency of glyceraldehyde-3-phosphate dehydrogenase (GAPDH) amplification compared with ZEB1 amplification obtained in a control reaction using lysates from GRHL2-null cells. The primer set used was F8, GCCGCCGAGCCTCCAACTTT; and R8, TGCTAGGGACCGGGCGGTTT.
Western blotting
SDS-PAGE was conducted using 4% to 20% gradient Tris Glycine gels (Invitrogen). Proteins were immobilized by electophoretically transferring them to a polyvinylidene difluoride filter (Immobilon) in 5% MeOH containing Tris-Glycine transfer buffer. Filters were blocked in PBS + 5% nonfat milk, primary antibodies were incubated in PBS + 0.1% Tween-20 + 5% nonfat milk, secondary antibodies were incubated in PBS + 0.1% Tween-20 + 5% milk + 0.01% SDS. Primary antibodies were typically incubated for 2 hours to overnight, secondary antibodies were incubated for 1 hour. Primaries used were: E-cadherin, ms (BD Biosciences); Vimentin, ms [Santa Cruz Bio-Tech (SCBT)]; N-cadherin, ms (BD Biosciences); CD44 (HCAM), ms (SCBT); ESRP1/2, ms (Contributed by Russ Carstens, University of Pennsylvania, Philadelphia, PA); Actin, ms (Millipore); Akt, rb or ms (Cell Signaling); GRHL2, rb (Sigma); Zeb1, rb (Sigma) or rb (Cell Signaling); Ankyrin-G, rb (S.M.F. custom generated; ref. 23); total-Smad2/3,ms (BD Biosciences); phospho-Smad2/3, rb (Cell Signaling); NF2 (merlin), rb (SCBT). Secondary antibodies for chemiluminescence were either anti-mouse or anti-rabbit, conjugated to horseradish peroxidase (Bio-Rad) and were used at 1:3,000 dilution; Western blot analysis were developed using (ECL-West Pico; Pierce). Fluorescently tagged secondary antibodies were mouse IRDye 800CW or rabbit IRDye 680LT (Li-Cor) used at 10:000 and detected using the Odyssey Infrared Imaging System (Li-Cor). Fluorescent images were converted to gray scale.
Other bioinformatics methods
We obtained GRHL2 expression data from NCBI Gene Expression Omnibus (accession number GSE10885; ref. 3). We compared the log expression of GRHL2 among the tumor types basal, claudin, Her2, luminal, and normal. We carried out a Tukey HSD test to determine which comparisons among these groups showed statistically significant differences; family-wise 95% confidence intervals for the log expression were plotted. Statistical analyses were conducted with R version 2.13.1 (www.r-project.org).
Results
Loss of GRHL2 expression is associated with a mesenchymal phenotype
On the basis of the specific expression of GRHL2 in mouse epithelia (12–14, 24), we investigated its potential regulation during EMT. Using HMLE cells that express a Twist-ER fusion—a previously characterized model of inducible EMT (21)—we analyzed the levels of GRHL2 over a time course of Twist induction by tamoxifen and found that, indeed, GRHL2 protein was downregulated during EMT with kinetics similar to the loss of E-cadherin and gain of N-cadherin (Fig. 1A). Consistent with this finding, spontaneous, stable EMT in a subpopulation of HMLE cells (“mesenchymal subpopulation”) obtained by sorting for CD44high phenotype, described previously (7), also displayed low GRHL2 expression (Fig. 1B).
Breast cancer cell lines have been classified by gene expression profiles. One particular subclass, basal-B, was characterized by a mesenchymal profile. GRHL2 was dramatically downregulated, specifically in this subclass (Fig. 1C). Analogously, the “claudin-low” subclass of mammary tumor samples—also characterized by a mesenchymal gene expression profile and poor prognosis—expressed substantially lower levels of GRHL2 than other subclasses, which, in fact, showed a modest upregulation, compared with normal breast tissue (Fig. 1D, and see Discussion). GRHL2 was also downregulated dramatically in a different tumor type characterized by EMT, clear cell renal carcinoma (Supplementary Fig. S1). Moreover, chemoresistant subpopulations of primary breast tumor cells obtained after chemotherapy of patients by sorting for CD44high/CD24low marker expression or mammosphere generation (25) expressed decreased levels of GRHL2 as well (Fig. 1E).
These data suggested that GRHL2 loss is a widespread characteristic of both primary and cultured tumor cells that have undergone EMT and acquired a tumor-initiating phenotype, informed the hypothesis that GRHL2 downregulation was functionally important for EMT.
GRHL2 is an EMT suppressor
The spontaneously occurring, CD44highCD24low mesenchymal subpopulation (MSP) cells within the heterogeneous HMLE cell line are characterized by EMT, attributed to autocrine TGF-β signaling pathway activation (7). Infection of HMLE cells with a GRHL2 expression construct and selection for the infected cells using a GFP marker caused the disappearance of the CD44high subpopulation (Fig. 2A) within a few days after infection, suggesting a conversion effect rather than selective growth (as shown later).
To further characterize this phenomenon, we isolated MSP cells from the HMLE cell line by flow sorting and infected these with the GRHL2 retrovirus. On the basis of E-cadherin immunofluorescence and Western blotting for epithelial and mesenchymal markers, GRHL2 reverted MSP back to an epithelial phenotype (Fig. 2B). Anoikis resistance and the ability to form mammospheres are key characteristics associated with EMT in HMLE cells. GRHL2 expression in the MSP cells restored anoikis sensitivity and reduced mammosphere formation dramatically, without affecting adherent cell growth (Fig. 2C).
These data indicated that GRHL2 reverted the spontaneous EMT and accompanying tumor-initiating cell characteristics of MSP cells. To test the effect of GRHL2 in other scenarios of EMT, we expressed it constitutively in HMLE + Twist-ER cells and in the prototypical EMT-like/triple-negative breast cancer cell line, MDA-MB-231, where it caused dramatic reversion of EMT and anoikis resistance in both cases (Fig. 2D and E), indicating a surprisingly broad specificity for this effect.
GRHL2 suppresses TGF-β–induced EMT
MSP cells and Twist-expressing HMLE cells rely on autocrine TGF-β signaling for their maintenance of mesenchymal and tumor-initiating properties (7), suggesting that GRHL2 could be suppressing EMT through this common pathway. We confirmed that Twist-mediated EMT and acquisition of anoikis resistance were TGF-β–dependent by using LY364947, a specific inhibitor of the TGF-βR1 kinase activity (Supplementary Fig. S2). Because this inhibitor mimicked the effects of ectopic GRHL2 in some respects, we tested the effect of GRHL2 on TGF-β–induced EMT.
TGF-β alone was previously reported to be insufficient for EMT induction in HMLE, a process requiring activation of multiple pathways (7). When GRHL2 was partially depleted by 2 distinct shRNAs, TGF-β alone induced EMT more efficiently than in cells with control shRNA (Fig. 3A). GRHL2 knockdown facilitated several activities of TGF-β: induction of a mesenchymal morphology, downregulation of epithelial-specific genes (E-cadherin, ESRP1 and ankyrin-G—an epithelial cytoskeletal protein that regulates anoikis; ref. 23), and upregulation of vimentin as well as the tumor-initiating cell marker, CD44S; surprisingly, TGF-β induced N-cadherin partly independently of GRHL2 expression. Coincident with this facilitation of EMT, GRHL2 knockdown also permitted TGF-β to confer a mammosphere-generating, anoikis-resistant state (Fig. 3A and B). GRHL2 also suppressed another feature of EMT, the formation of large protrusive structures during the growth of colonies in 3D Matrigel culture, indicative of invasive potential (Supplementary Fig. S3).
These results suggested that signaling downstream of the TGF-β receptor might be suppressed by GRHL2. We focused on Smad-mediated transcription, as it plays an important, although not exclusive role in the response to exogenous TGF-β–mediated EMT (4). Using a well-characterized reporter assay (3TP-lux; ref. 4), GRHL2 knockdown stimulated TGF-β/Smad-mediated transcription by approximately 4.6× relative to control cells (Fig. 3C). Correspondingly, GRHL2 knockdown promoted the induction of the TGF-β/Smad target genes, CTGF (26), and ZEB1 (27), by exogenous TGF-β. Surprisingly, there was no discernable effect of GRHL2 on either the phosphorylation or nuclear translocation of Smad2, suggesting that other mechanisms of inhibition were operative (see Discussion). These results indicated that GRHL2 inhibited Smad-mediated transcription in response to exogenous TGF-β.
In HMLE cells expressing low levels of activated K-ras, (HMLER), GRHL2 knockdown sufficed to induce EMT—that is, even without exogenous TGF-β based on the criteria used earlier (Fig. 3D). This EMT was clearly dependent upon autocrine TGF-β signaling, in that LY364947 reversed it (Supplementary Fig. S4). The TGF-β signaling antagonists BMP2 and -4 were previously shown to be downregulated in MSP cells relative to normal HMLE, which promoted autocrine TGF-β signaling (7). Interestingly, BMP2 expression was activated by GRHL2 (Fig. 3E), consistent with the idea that GRHL2 suppressed not only TGF-β signaling in response to exogenous ligands but also (by comparison with reported data on the same cell lines; ref. 7) autocrine signaling.
GRHL2 represses ZEB-1 expression
Suppression of EMT by GRHL2 could occur by a diversity of mechanisms. To elucidate one or more of these in an unbiased manner, we conducted a microarray-based gene expression profiling comparing the HMLE-Twist cells with or without GRHL2 expression. This analysis revealed that genes regulated by GRHL2 (GEO database accession number GSE36081) correlated negatively (R = −0.7508) with genes regulated during EMT (by multiple transcription factors) in the HMLE system, validating the EMT-suppressive effect of GRHL2 (Fig. 4A). The genes regulated by GRHL2 included markers of epithelial versus mesenchymal phenotypes, several of which were ZEB1 target genes; transcription factors implicated in the control of EMT were also noted.
Interestingly, one of the major downregulated GRHL2 target genes was the E-cadherin repressor/EMT inducer ZEB1, as shown by RT-PCR (Fig. 3E) and Western blotting in MSP cells (Fig. 2B), MDA-MB-231 cells (Fig. 2E), HMLE + shGRHL2 cells with TGF-β (Fig. 3A), HMLER + shGRHL2 cells (Fig. 3D) and HMLE + Twist-ER cells (Fig. 2D). Functional consequences of ZEB1 downregulation (28) such as the upregulation of mir-200b/c (Fig. 4B) and ESRP1 (Fig. 3E) were also evident. The downregulation of ZEB1 by GRHL2 was investigated further as a potential mechanism for suppression of EMT.
To determine whether the ZEB1 gene could be a direct target for GRHL2, we cotransfected the previously characterized ZEB1 promoter (22) together with GRHL2 into the MSP cells (which express low endogenous GRHL2). This revealed a marked repression of the ZEB1 promoter by GRHL2, as did the converse experiment, transfection of the ZEB1 promoter into cells with or without knockdown of endogenous GRHL2 (Fig. 4C).
Inspection of the approximately 1 kb of promoter sequence that was GRHL2 responsive revealed several potential binding sites for Grainyhead proteins. We tested approximately 200-bp nested fragments of the ZEB1 upstream region, in the context of an SV40 promoter, for repression by GRHL2, and identified one fragment (fragment 4) that was highly repressed. This fragment contained a consensus GRHL2-binding site and also carried a strong enhancer; the repression by GRHL2 was completely eliminated by a 4-base mutation of this consensus site (Fig. 4D). To determine whether the ZEB1 promoter was a direct target for repression by GRHL2, ChIP analysis was conducted, showing a strong enrichment of PCR signal using GRHL2 antibody, with respect to nonimmune IgG or a primer set (using GRHL2 antibody) representing an unrelated region of the genome (Fig. 4E). These results indicated that GRHL2 repressed ZEB1 expression and interacted directly with the ZEB1 promoter.
Suppression of ZEB1 is critical for the suppression of EMT by GRHL2
ZEB1 plays a critical role in EMT in response to various stimuli including TGF-β (29–33), informing the hypothesis that GRHL2 suppressed EMT, at least in part, by repressing ZEB1 expression. To test this, ZEB1 was expressed ectopically, using a doxycycline-inducible promoter, in the HMLE + Twist-ER + GRHL2 cells. By the criteria of morphology, expression of epithelial and mesenchymal markers, and anoikis resistance, ZEB1 restored EMT that had previously been blocked by GRHL2 expression (Fig. 5A). Analogous effects of ZEB1 expression were also observed in MSP cells that had been reverted to an epithelial phenotype by stable GRHL2 expression (Supplementary Fig. S5). Conversely, in the HMLE cells where GRHL2 knockdown predisposed the cells toward TGF-β–induced EMT, ZEB1 knockdown blocked this induction (Fig. 5B). Similarly, EMT that was induced by GRHL2 knockdown in HMLER cells was reversed by ZEB1 knockdown (Supplementary Fig. S6). These results indicated that the repression of ZEB1 was a key mechanism by which GRHL2 suppressed EMT.
Discussion
Mammalian GRHL2 is a transcription factor that plays important role in epidermal junctions, in part due to activation of target genes including claudin-4 and E-cadherin. Consistent with this role, the Drosophila Grainyhead gene is among the first transcription factors used in the MZT during embryonic development (16), and the 3 mammalian Grainyhead genes are critical for embryonic and adult wound healing (11–14). In light of the fact that wound healing is orchestrated in part by TGF-β signaling (9), the suppressive effect of GRHL2 on this pathway suggests that GRHL2 may contribute to the resolution phase of wound healing, wherein transient EMT-like cell conversions in keratinocytes are instructed to reverse. The suppressive effect of GRHL2 on oncogenic EMT may be understood by analogy to this function, given the similarities between the 2 contexts of EMT (8).
The significance of mammalian Grainyhead proteins in cancer is emerging. GRHL3 was recently shown to function as a tumor suppressor in squamous cell carcinoma, acting, at least in part, as a direct activator of PTEN expression (19); EMT-related issues were not examined, however. The GRHL2 gene shows frequent amplification in unclassified breast tumor samples, and has been proposed as a potential oncogene in breast cancer, due, in part, to its suppression of death receptor expression (18). Consistent with this, modest upregulation of GRHL2 mRNA was observed in luminal A, B, and HER2-positive tumor types (Fig. 1D; although it is unclear whether this is an artifactual result of expansion of the epithelial cell compartment relative to normal mammary gland during tumor outgrowth). In contrast, our results show that GRHL2 is downregulated in EMT models and EMT-driven tumor subclasses, and that it suppresses TGF-β–induced ZEB1 expression; in light of the established protumorigenic potential of ZEB1, this result predicts that GRHL2 will specifically suppress EMT-like tumors (34, 35).
These results can be reconciled in light of the diametrically opposed, context-dependent effects of TGF-β: growth arrest and tumor suppression in certain tumors versus tumor promotion in others (4). In breast cancer, fewer than 10% of patients have tumor types (claudin low, metaplastic) in which EMT/TGF-β contributes critically to tumor progression, whereas in the majority of tumors (other basal, luminal A, B, and HER2-positive subclasses)—which most transgenic mouse models emulate—TGF-β is tumor suppressive (3, 36). By targeting the TGF-β pathway, GRHL2 is predicted usually to act as an oncogene (i.e., in the most common subclasses of breast cancer) or, less frequently, as a tumor suppressor gene (i.e., in EMT-like subclasses).
The results here indicate that GRHL2 interferes with the response to TGF-β by at least 2 mechanisms, interference with Smad2/3–mediated transcriptional activation and direct repression of the ZEB1 promoter (diagrammed in Fig. 6). Consistent with previous observations in other systems (28, 37), ZEB1 was required for EMT in response to Twist, TGF-β, and spontaneous conversion. GRHL2 also upregulated mir-200b/c, consistent with a critical role of the established ZEB1/mir-200 feed-forward regulatory loop in EMT (28). The precise mechanism by which GRHL2 represses the ZEB1 promoter may relate to Grainyhead proteins' ability to repress transcription, by recruiting polycomb repression complex components or by interfering with the binding of a transactivator (16, 38).
The mechanism by which GRHL2 inhibits Smad-mediated transcription is unresolved at present. Previous work has shown that ZEB1 protein can bind to the Smad2/3 complex, enhancing transactivation (39); our preliminary observations indicated that this mechanism did not apply in our system. Smad2/3 nuclear versus cytoplasmic localization is regulated by phosphorylation as well as signaling from the Crumbs polarity complex through Hippo pathway components (26). These mechanisms were, however, unlikely to explain the suppression by GRHL2, because Smad phosphorylation and nuclear translocation were not apparently affected. Other nuclear proteins that affect Smad2/3 transactivation, such as TGIF, Ski, and Sno (4), remain to be tested in the context of GRHL2.
TGF-β–induced EMT is a highly restricted phenomenon in cell culture models, occurring in only a small number of epithelial cell lines (40). In fact, we observed that the mouse mammary epithelial cell line NMuMg, commonly used to study this phenomenon, has undetectable GRHL2 expression, whereas other mouse mammary lines that are unresponsive do express GRHL2 (Supplementary Fig. S7). These results are consistent with the previous finding that additional factors from the tumor microenvironment confer TGF-β responsiveness upon HMLE cells, suggesting that one or more of these factors could function by downregulating GRHL2 (7). More generally, the GRHL2 expression profile in breast cancer samples and cell lines indicate that GRHL2 is a general barrier to EMT. Accordingly, GRHL2 prevented TGF-β from conferring anoikis resistance, mammosphere formation (a tumor-initiating cell behavior), and invasive growth in 3D culture, predicting a tumor-suppressive effect in this context.
These results also suggest that GRHL2 may be a useful biomarker for tumors predicted to respond to TGF-β receptor inhibitory drugs currently in clinical trials (41): GRHL2-null tumors, being susceptible to the tumor-promoting effects of TGF-β, are predicted to respond specifically to this class of drugs, an approach that could improve their efficacy substantially.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were declared.
Authors' Contributions
Conception and design: B. Cieply, A.V. Ivanov, S.M. Frisch
Development of methodology: B. Cieply, A.V. Ivanov, S.M. Frisch
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B. Cieply, P. Riley, P.M. Pifer, J. Widmeyer, J. Addison, S.M. Frisch
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B. Cieply, P.M. Pifer, J. Addison, A.V. Ivanov, J. Denvir, S.M. Frisch
Writing, review, and/or revision of the manuscript: B. Cieply, P. Riley, P.M. Pifer, J. Denvir, S.M. Frisch
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B. Cieply, P. Riley, S.M. Frisch
Study supervision: B. Cieply, S.M. Frisch
Acknowledgments
The authors thank Kathy Brundage (flow cytometry), Kimmi Alonge, and the laboratories of Peter Stoilov, Mike Ruppert, Laura Gibson, Antonio Garcia de Herreros, Antonio Postigo, Richard Myers, and Kai-Schmidt-Ott for help, advice, and reagents and Wioletta Szeszel-Fedorowicz for carrying out the microarray experiment.
Grant Support
S.M. Frisch was supported by NIH grant R01CA123359. The flow cytometry core facility (Mary Babb Randolph Cancer Center) was supported by NIH grants RR020866 and P20 RR16440. J. Denvir was supported in part by the West Virginia IDEA Network of Biomedical Research Excellence (INBRE) P20 RR 016477-12 (Rankin, Gary, PI). A.V. Ivanov was supported by NIH grant P20 RR16440, American Cancer Society grant 122300-IRG-09-061-04-IRG, and Susan G. Komen for the Cure grant KG110350.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.