An epithelial marker promoter induction screen identifies histone deacetylase inhibitors to restore epithelial differentiation and abolishes anchorage independence growth in cancers

Epithelial–mesenchymal transition (EMT), a crucial mechanism in development, mediates aggressiveness during carcinoma progression and therapeutic refractoriness. The reversibility of EMT makes it an attractive strategy in designing novel therapeutic approaches. Therefore, drug discovery pipelines for EMT reversal are in need to discover emerging classes of compounds. Here, we outline a pre-clinical drug screening platform for EMT reversal that consists of three phases of drug discovery and validation. From the Phase 1 epithelial marker promoter induction (EpI) screen on a library consisting of compounds being approved by Food and Drug Administration (FDA), Vorinostat (SAHA), a histone deacetylase inhibitor (HDACi), is identified to exert EMT reversal effects by restoring the expression of an epithelial marker, E-cadherin. An expanded screen on 41 HDACi further identifies 28 compounds, such as class I-specific HDACi Mocetinosat, Entinostat and CI994, to restore E-cadherin and ErbB3 expressions in ovarian, pancreatic and bladder carcinoma cells. Mocetinostat is the most potent HDACi to restore epithelial differentiation with the lowest concentration required for 50% induction of epithelial promoter activity (EpIC-50).The HDACi exerts paradoxical effects on EMT transcriptional factors such as SNAI and ZEB family and the effects are context-dependent in epithelial- and mesenchymal-like cells. In vitro functional studies further show that HDACi induced significant increase in anoikis and decrease in spheroid formation in ovarian and bladder carcinoma cells with mesenchymal features. This study demonstrates a robust drug screening pipeline for the discovery of compounds capable of restoring epithelial differentiation that lead to significant functional lethality.


INTRODUCTION
Epithelial-mesenchymal transition (EMT) is a gradual process whereby epithelial cells lose their epithelial features, enter into the hybrid intermediate state while gaining some mesenchymal features, and ultimately transdifferentiate into mesenchymal cells. 1 This process is reversible in nature with the hybrid intermediate cells being shown to be in the most plastic state. 2 As EMT has been implicated as one of the contributing mechanisms to the aggressiveness of carcinoma during disease progression, cancer stemness and chemoresistance, 3 the possibility to reverse the aggressiveness by reversing EMT and restoring the epithelial differentiation has emerged to be an appealing strategy in cancer treatment. 4 The main molecular mechanism for EMT is mediated by several key transcription factors (TF) to regulate their downstream targets at the transcriptional, translational and post-translational levels that are associated with transdifferentiation. 5 Upstream to this, several signaling pathways responding to the external cues are crucial to mediate the convergence of the signals to the main transcriptional EMT factors. Therefore, these signaling pathways such as transforming growth factor-β (TGFβ), hepatocyte growth factor, insulin-like growth factor 1, epidermal growth factor (EGF), fibroblast growth factor and platelet-derived growth factor pathways have been suggested to be the desirable targets against EMT. 5,6 EMT reversal and restoring epithelial differentiation has been shown to achieve by kinase inhibitors, such as the TGFβ receptor types I and II inhibitor LY2109761, 7 the Src-kinase inhibitor saracatinib (AZD0530) 8 and the triple angiokinase inhibitor nintedanib (BIBF1120), 9 which acts to upregulate E-cadherin expression both in vitro and in vivo. However, most of these kinase inhibitors have not been demonstrated to exhibit anti-EMT-related effects in the clinical settings except for nintedanib (BIBF1120) which has been shown to resolve idiopathic pulmonary fibrosis by inhibiting the TGFβ pathway. 10 Therefore, there is a need to explore other classes of compounds.
The concept of EMT reversal is similar to the differentiation therapy 11 that involves re-programming of the cancer cells 12 from the mesenchymal to epithelial trait. With EMT being a crucial differentiation-based developmental model in cancers, the identification of targetable pathways to re-program the mesenchymal trait would be very promising. Differentiation therapy with all-trans retinoic acid has been developed to treat acute myeloid leukemia. 13 Over the years, several pharmaceuticals and natural compounds have also been shown to re-program the differentiation pathways in leukemia cells. 14 Increasing evidences have shown that the differentiation therapy in solid tumors is possible. 15 The cancer stem cell (CSC) concept have further provided the theoretical and practical grounds to develop the differentiation therapy in solid tumors such as breast and renal cancers. 11,16 During the differentiation of stem cells, epigenetic regulations are the key governing mechanism and thus pose as an appealing therapeutic target for differentiation therapy in CSC. 17 The implication of EMT and CSC 18 thus makes epigenetic modifiers a promising class of compounds for EMT reversal and restoring epithelial differentiation.
In this study, we describe the discovery of histone deacetylase (HDAC) inhibitors from a 3-phase drug screening pipeline for restoring epithelial differentiation. We demonstrate that these HDAC inhibitors (HDACi) induce different effects in ovarian cancer cells with different EMT statuses. The EMT reversal effect of restoring E-cadherin ErbB3 expressions by HDACi is also validated in non-ovarian cancer cells such as pancreatic and bladder cancers. Restoration of epithelial differentiation by these HDACi has a functional relevance in overcoming anoikis resistance and anchorage independence growth.

An epithelial marker promoter induction screen identifies EMT reversal agents
The EMT reversal application is based on re-differentiating cancers along an EMT spectrum which is quantitatively defined by continuous EMT scores. 19 Therefore, establishing a robust drug discovery pipeline based on the reversibility of EMT is mandatory. We have established a pre-clinical drug discovery pipeline ( Figure 1a) for EMT reversal by using the re-expression of an epithelial differentiation marker, E-cadherin, as the readout. We have previously demonstrated that a short version of the E-cadherin promoter region containing the E-box sequences can be used to reflect an increase in CDH1 promoter activity upon EMT reversal. 9 The discovery pipeline starts from the Phase 1 screening of a Food and Drug Administration (FDA)-approved drug library in a human ovarian cancer cell line harboring an intermediate EMT score, SKOV3, transiently transfected with the pGL3 luciferase plasmid containing the short promoter region of E-cadherin ( Figure 1b, Supplementary Materials). This screening platform is therefore referred as the epithelial marker promoter induction (EpI) screen. The Phase 2 screen expands to compounds belonging to the same class of action as the shortlisted compounds from Phase 1. These compounds would be subjected to another round of EpI screen in SKOV3. Compounds selected from Phase 2 are further tested in various cancer cell lines annotated with similar EMT scores to SKOV3 including, MDA-MB-231 (breast), T24 (bladder), A549 (lung) and Mia-Paca2 (pancreas). The dose-dependent dynamic ranges of the compounds to induce re-expression of E-cadherin are determined in vitro. Compounds with good linear dynamic ranges in re-expressing E-cadherin without apparent cytotoxicity will then be tested in the Phase 3 functional validation.  promoter activities (Figure 1c) in SKOV3. Among them, Vorinostat (SAHA) showed no apparent cytotoxicity represented by the Renilla readout at the screening concentration of 5 μM. This was further validated by the MTS assay (Supplementary Figure 1). We subsequently expanded the screen to include other compounds belonging to the same class of action as Vorinostat, which is the histone deacetylase inhibitor (HDACi). Among 41 HDACi screened, 28 showed enhanced E-cadherin promoter activities of at least twofold (Figures 1d and e). Interestingly, in the order of fold changes of E-cadherin promoter activities, the top ranking compounds (CUDC-907, Mocetinostat, 4SC-202, CI994, Scriptaid and Entinostat) all have selectivity towards the class I HDAC (Figure 1e). In particular, two compounds with dual inhibition of HDAC and kinase activities, CUDC-907 and CUDC-101, also emerged from this screen. This is consistent with the findings that CUDC-101 prevents in vitro and in vivo aggressiveness via suppressing EMT. 20, 21 We previously reported that a triple angiokinase inhibitor, Nintedanib, could restore E-cadherin expression. 9 Therefore, the class of angiogenesis inhibitors was also subjected to the Phase 1 EpI screen. Among 28 compounds screened, 10 showed enhanced E-cadherin promoter activities of at least twofold ( Supplementary Figures 2A and B). Of note, these angiogenesis inhibitors were all less potent than the selective class I HDACi in EpI.

Class I HDACi exhibits dose-dependent induction of E-cadherin and ErbB3 expression
We have previously shown that the degree of EMT reversal could be represented by determining the concentration required for 50% induction of E-cadherin (E-cad) promoter activity. 9 The EpIC-50 of Vorinostat, Mocetinostat, CI994 and Entinostat in SKOV3 was determined to be 1.24 μM, 147.8 nM, 2.37 μM and 568.3 nM, respectively (Figure 2a). A HDAC6-specific inhibitor, ACY-1215, though showing 7.8-fold of increase in EpI at the screening concentration of 5 μM (Figure 1e), its dose-response curve only started to display linearity at much higher concentrations with an EpIC-50 reaching 47 μM (Figure 2b). It is known that HDAC6-specific inhibitors like ACY-1215 would display cross selectivity to class I HDAC at high concentrations. Therefore, our data indicates that the EpI effect is specifically attributed to the inhibition of class I HDAC.
The linearity of EpIC was concordant with the dose dependency increase in transcript expressions of E-cadherin (CDH1) upon treating SKOV3 with 500 nM and 5 μM of Vorinostat ( Figure 2c). However, this trend was only evident in mesenchymal-like but not in epithelial-like cell lines. By using four ovarian cancer cell lines representing different EMT phenotypes, 8  the epithelial lines PEO1 and OVCA429 below twofold, suggesting that the regulation of CDH1 transcript level might have reached equilibrium in these epithelial-like lines. We further tested whether the EpI effect could be generalized to other epithelial genes. Another epithelial-specific gene, ERBB3, a direct transcriptional target of the epithelial gatekeeper GRHL2, 22 has emerged as a promising therapeutic target in cancer. We generated various clones of the ERBB3 promoter to cover two GRHL2-binding sites and five E-box sites close to the transcription start site of ERBB3 (Figure 3a). We utilized Mocetinostat to test which promoter clones would show similar activity as the E-cad promoter. ERBB3-L, the clone of 1.2-kb sequence spanning the two upstream GRHL2-binding sites, one upstream E-box site and one downstream E-box site to transcription start site showed similar fold of increase of promoter activity to E-cad (Figure 3b). The ERBB3-L promoter activities also showed a dose-dependent induction following HDACi treatments. The EpIC-50 concentrations based on ERBB3 promoter activities in SKOV3 were determined to be 318.5 nM, 77.7 nM, 365.2 nM and 3.40 μM for Vorinostat, Mocetinostat, Entinostat and CI994, respectively (Figure 3c), and were consistently lower than those for E-cad promoter. Similar to the results of E-cad promoter activities, Mocetinostat showed the lowest EpIC-50 for the ERBB3 promoter. This suggested that epithelial genes might have different thresholds for epigenetic regulations to induce transcriptions. Similar to CDH1, ERBB3 expressions were significantly increased by these HDACi in the mesenchymal lines, with Mocetinostat being the most potent HDACi (Figures 3d-g and 2g). Collectively, our data suggested that by using epithelial gene promoters as readouts, the restoration of epithelial differentiation can be achieved by the EpI screen strategy.

Cancer cells with similar EMT status show comparable EpI responses induced by class I HDACi
We next explored whether the EpI screen could be applied to cancer cells other than ovarian cancer. Cancer cell lines with similar EMT scores 19  However, given that MDA-MB-231 did not showed EpI responses, the restoration of CDH1 and ERBB3 expression in mesenchymallike cells by HDACi is therefore still context dependent. Collectively, our data suggested that the EpI screen strategy is an efficient and robust method to identify targets for restoring the epithelial differentiation.  (Figures 5a and c); the mesenchymal-like OVCA429_shGRHL2 showed a huge induction of CDH1 expression but with reciprocal changes between SNAI1, SNAI2 with the ZEB family and TWIST1 (Figures 5b and d). Similar to ovarian cancer cells, the class I HDACi induced significant SNAI family expressions in Mia-Paca2 and T24 cells (Figures 5e and f). These results also indicate that the epigenetic regulations of the EMT TFs are different between the epithelial and mesenchymal cells. Of note, the folds of induction of CDH1 expression were significantly higher than those of the reduction of EMT TFs. This suggested that the restoration of CDH1 expression by HDACi was a direct epigenetic regulation to the E-cadherin gene rather an indirect consequence to the downregulation of the EMT TFs.
Class I HDACi inhibits anchorage independence growth in mesenchymal ovarian and bladder carcinoma cells EMT reversal and restoration of E-cadherin expression are associated with increased anoikis. 8,9,23 We subsequently utilized the ovarian carcinoma SKOV3 and bladder carcinoma T24 lines for functional studies to test if these class I HDACi affect the anoikis resistance in vitro. The percentage of the Annexin V-positive populations at 48 h in ultra-low attachment cultures was applied as the indication of cells entering apoptosis. The HDACi-treated T24 cells showed significant increase in the cell fractions entering the early anoikis phase (Annexin V high /PI low populations; Figure 6a). This anoikis-inducing effect followed dose dependency of HDACi whereas increased percentage of Annexin V-positive cells was found in the 5 μM compared with the 500 nM treated group (Figure 6a; Supplementary Figure 3). In the isogenic EMT model, similar trends were also evident in the mesenchymal OVCA429 shGRHL2 cells that all three class I HDACi significantly enhanced anoikis (Figure 6b). Of note, the epithelial OVCA429 shLUC line consistently showed higher anoikis fraction compared with the mesenchymal OVCA429 shGRHL2 line. In addition, the HDACi-treated cancer cells showed decreased spheroid forming efficiencies. Consistent with the anoikis-inducing effect, Mocetinostat-treated cells showed a significant reduction of spheroid formation at 500 nM in both SKOV3 and T24 (Figure 6c). The effect was even more prominent at 5 μM of Mocetinostat and Entinostat (Figures 6c and d). In conclusion, the HDACi abolished the anchorage independence growth and overcome the anoikis resistance of mesenchymal-like cancer cells while restoring the epithelial differentiation.

DISCUSSION
Previously, we have proposed two strategies to ascertain the utility of EMT reversal in carcinoma: the first, assessing EMT-related functions, such as the induction of colony compaction; the second, inducing re-differentiation of mesenchymal-like cells by the upregulation of epithelial markers, such as E-cadherin. 9 In this study, we utilized the second approach and established a 3-phase drug discovery pipeline and identified HDACi to be a potent class of compounds to restore epithelial differentiation.
Increasing reports have demonstrated HDACi to be a promising class of anticancer agents. They act as epigenetic modifiers, by promoting histone hyperacetylation, to influence chromatin remodeling and eventually, gene expression. Being structurally diverse, HDACi is classified into hydroxamic acids, carboxylic acids, benzamides, cyclic peptides and short fatty acids. 24 Two FDAapproved HDACi, Vorinostat and Romidepsin, are currently in use for cutaneous T-cell lymphoma treatment. [25][26][27][28] Several other HDACi, including Entinostat and Mocetinostat, are under investigation in different phases of clinical trials. [29][30][31] The mechanisms of epigenetic modifiers were mainly focused on their anti-proliferative, pro-apoptotic and anti-angiogenic effects. 32,33 However, the role of HDACi in EMT remains unclear. In fact, contradicting data, either promoting EMT or inhibiting EMT by HDACi, have been reported.
Vorinostat or Trichostatin A treatment was found to induce EMT through upregulation of ZEB1, ZEB2, SNAI1 and SNAI2 expression, with associated increased expression of mesenchymal markers, N-cadherin and vimentin, in prostate cancer cells. 34 Similar results were obtained in human nasopharyngeal carcinoma cells (CNE2), colon cancer cells (LoVo) and liver carcinoma cells (HepG2). These cancer cells demonstrated increased SNAI1 expression and metastasis, coupled with a fibroblast-like morphology of mesenchymal cells, after Vorinostat or Sodium butyrate (NaB) treatment. 35 On the contrary, treatment with either Vorinostat or Trichostatin A, squamous cell carcinoma of the head and neck (SCCHN) cells, induced E-cadherin expression and suppressed TGFβ expression, an indication of EMT inhibition. 36 Recently, several epigenetic modifiers have indeed been reported to reverse the aggressiveness associated with EMT in solid tumors. [37][38][39][40] Meidhof et al. 39 reported that Mocetinostat reversed the gemcitabine resistance in pancreatic cancer via suppressing the EMT TF ZEB1. However, based on our results, Mocetinostat increased ZEB1 expression significantly in epithelial-like lines PEO1 and OVCA429, correlating with increased resistance to HDACi treatment. On the other hand, mesenchymal-like cell lines, SKOV3 and Hey, demonstrated suppressed ZEB1 expression after Mocetinostat treatment, correlating with increased sensitivity to HDACi treatment. Interestingly, epithelial-like ovarian cancer cells had been shown to be more resistant to cisplatin than mesenchymal-like ovarian cancer cells. 41 Thus, the intrinsic EMT states of cancer cells might dictate their difference in drug sensitivity to both HDACi and conventional chemotherapeutics. Indeed, in our hands, prior short-term exposure to HDACi did not sensitize SKOV3 cells to paclitaxel or cisplatin (Supplementary Figure 4). Therefore, the combinatory usage of HDACi with other cytotoxic agents would need to be carefully investigated for the optimized dosing schedules.
We also observed that MDA-MB-231 has the highest EpIC-50 among all tested cell lines. This is supported by the work done by Mocetinostat has the specificity against HDAC1, 2, 3, 11; Entinostat and CI994 are both specific to HDAC1 and 3. HDAC1, HDAC2 and HDAC3 have all been shown to regulate the expression of epithelial genes and EMT. The suppression of E-cadherin is mediated by a SNAIL/HDAC1/HDAC2 repressor complex. 43,44 HDAC3, by interacting with hypoxia-induced WDR5, serves as an essential corepressor to repress epithelial gene expression. 45 Intriguingly, these HDACi all significantly induce SNAI1 expressions in both epithelial-and mesenchymal-like lines. Being a strong transcriptional repressor to CDH1, the SNAI1 induced by HDACi does not decrease the CDH1 expression in these lines. On the contrary, the CDH1 expression is greatly enhanced in the mesenchymal-like lines. These data suggested that class I HDACs are pivotal in the downregulation of epithelial genes during EMT. Therefore, the class I HDACs, even at the presence of a strong EMT TF such as SNAI1, are a very promising class of target for restoring epithelial differentiation.
In addition to using the E-cadherin promoter as the readout, we also show that promoters of other epithelial genes, such as ERBB3, can also be utilized for EpI screen. ErbB3 belongs to the epidermal growth factor receptor family. It has emerged as a promising therapeutic target in cancer by using monoclonal antibodies. 46 Vorinostat was reported to revert the mesenchymal phenotype and enhance the antitumor effect of gefitinib by inducing both E-cadherin and ErbB3 expressions. 47 In our hands, the class I HDACi shows even lower EpIC-50 in inducing the ERBB3 promoter activities than that in CDH1. This suggests that restoring ErbB3 expressions might be easier to achieve to overcome the epigenetic barriers. Our group has previously shown that ERBB3 is a direct transcriptional target of an epithelial-specific TF, GRHL2 22 , and the ErbB3 expression and its phosphorylation are significantly decreased along an EMT spectrum. 8 Therefore, sensitizing the mesenchymal-like tumors expressing low ErbB3 with HDACi would be a promising strategy. Recently, synergistic antitumor activity of HDACi and anti-ErbB3 antibody has been reported in NSCLC primary cultures. 48 In summary, we report here a drug screening pipeline for the discovery of classes of compounds to restore epithelial differentiation. We found that HDAC inhibitors achieved the EMT reversal and epithelial differentiation effect by restoring E-cadherin and ErbB3 expressions in a panel of cancer cells with an intermediate EMT state. The identified HDACi also showed functional relevance in reversing anoikis resistance and abolishing spheroid formation. Future combination of HDACi with agents targeting ErbB3 signaling is promising. were maintained in RPMI11640 (Biowest SAS, Nuaillé, France), supplemented with 10% (v/v) fetal bovine serum (Biowest SAS, Nuaillé, France) and 2 mM NaPy (Sigma-Aldrich, St. Louis, MO, USA). All cells were cultured at 37°C in a humidified atmosphere containing 5% CO 2 and 95% air.

Compound library screens
A FDA-approved 853 drug library was purchased from Selleckchem (Houston, TX, USA) and used in Phase I EpI screen. Compounds, in the form of 10 mM in DMSO, were diluted to 1 mM and kept at − 20°C, to be used on day 3 drug treatment. Class I HDACi (10 mM in DMSO), used in Phase 2 EpI screen, were obtained from Inhibitor Library (Selleckchem, Houston, TX, USA). Selected class I HDACi-Vorinostat, Mocetinostat, Entinostat and CI994-were purchased from Selleckchem in the form of powder and reconstituted in DMSO to a concentration of 10 mM and kept at − 20°C.
Cells of interest were plated at 40-50% confluence per well, in complete media, into 96-well plates (no. 655094, Greiner). After 24 h, cells were transfected with 100 ng of pGL3-Ecad or vector control and 1.5 ng of pRL-CMV Renilla vector per well, using X-tremeGENE HP DNA transfection reagent (no. 6366236001, Roche, Mannheim, Germany) at 2 : 1 HP : DNA ratio. On day 3, cells were treated with 5 μM of drug (Selleckchem, final concentration) or a similar volume of DMSO (for control conditions, final concentrations: 0.05% v/v). After 24 h of drug treatment, Dual-Glo Luciferase Assay was performed according to the manufacturer's protocol (no. E2940, Promega, Madison, WI, USA).

EpIC-50 determination
All cells were plated and transfected as described above. On day 3, SKOV3 were treated with Vorinostat or CI994 (both at 1  Anoikis assay SKOV3 and T24 cells were seeded in 10 cm ultra-low attachment dish and treated with two different concentrations of 500 nM and 5 μM of HDACi (SAHA, Mocetinostat, Entinostat and CI994) at the time of seeding. After 48 h incubation, the cells were collected and trypsinized into single cell suspensions. The cells were then washed twice with PBS and counted using a hemocytometer. Equal numbers of cells were taken from control and HDACi, then transferred into a 15-ml Falcon tube (BD Biosciences, San Jose, CA, USA). The cells were re-suspended in 100 μl of 1 × Annexin V binding buffer. Annexin-Pacific Blue conjugate (5 μl; Sigma-Aldrich) and 1 μl of 100 μg/ml Propidium Iodide (Sigma-Aldrich) working solution were added and incubated for 15 min at room temperature. After the incubation period, 400 μl of 1 × annexin-binding buffer were mixed gently and kept the samples on ice. Data were acquired using a BD LSRII flow cytometer (BD Biosciences). A minimum of three independent experiments was performed to study Anoikis assay.
Spheroid forming assay SKOV3 and T24 cells were cultured in 2D culture dish up to 70% of confluency. The cells were trypsinized and counted using a hemocytometer. In each well of 96-well ultra-low attachment plates, 200 cells were seeded and treated with HDACi (SAHA, Mocetinostat, Entinostat and CI994) with two different concentrations of 500 nM and 5 μM at the time of seeding. The cells were incubated for 7 days and spheroids were manually counted at day 7. Each HDACi drug study has been performed with three independent experiments.