AC1MMYR2, an Inhibitor of Dicer-Mediated Biogenesis of Oncomir miR-21, Reverses Epithelial–Mesenchymal Transition and Suppresses Tumor Growth and Progression

MicroRNAs (miRNA) are a class of evolutionarily conserved, small noncoding RNAs that regulate fundamental cellular processes through the modulation of gene expression (1, 2). The biogenesis of miRNAs begins with the transcription of primary transcripts (pri-miRNA) that undergo further processing within the nucleus, resulting in a hairpin intermediate of approximately 70 nucleotides (precursor miRNA, pre-miRNA). The pre-miRNA is then transported out of the nucleus to the cytoplasm and Dicer, an RNaseIII, removes the “terminal loop region,” or preelement, to yield a 22 nt mature miRNA duplex. After duplex separation, the guide strand or mature miRNA, is incorporated into an RNA-induced silencing complex that participates in mRNA degradation and/or translational suppression (3, 4).

Tremendous progress has been made during the last few years to identify novel, tumor-associated miRNAs, which are known to modulate many biologic pathways related to cancer initiation, progression, metastasis, and therapy resistance (5–7). The aberrant expression of miRNAs is deemed to be a notable molecular hallmark of human diseases. miR-21 was one of the first miRNAs detected in the human genome and it has been validated to be overexpressed in many different types of human cancers, such as glioblastomas (8), gastric tumors (9), breast cancers (10), and colon cancers (11). Thus, targeting miR-21 and modulating its activity is emerging as a promising therapeutic option and opens a new mode of cancer therapy.

To date, biochemically based tools for the sequence-specific inhibition of miRNA function used in preclinical and clinical studies include antisense oligonucleotides, locked nucleic acids, and antagomirs (12). Although satisfactory results have been reported, there are challenges implicated in the delivery of these non–small-molecule agents and their pharmacodynamic and pharmacokinetic properties are not ideal for the treatment applications (13). On the basis of these observations, the discovery of novel and potent small molecules targeting miRNA is needed, and one must take into account both the three-dimensional (3D) structure of miRNAs and the thermodynamics of miRNA–small-molecule interactions (14).

Recently, in silico 3D structure prediction has undergone significant advances due to the availability of new experimental data and enhanced computer power and modeling methodologies (15). MC-Fold/MC-Sym, an RNA-structure prediction program, has been successfully used to accurately predict the double-helix region of several pre-miRNAs, such as let-7c, miR-19, and miR-29a (16). Here, we propose a new high-throughput screening method for small molecules targeting miR-21 based on the 3D structure of the Dicer binding site on pre-miR-21 by integrating the MC-Fold/MC-Sym and AutoDock programs (17).

By screening a small chemical compound library, we identified a specific small-molecule inhibitor of miR-21, AC1MMYR2, which directly blocked Dicers' ability to process pre-miR-21 to mature miR-21. Furthermore, we determined that AC1MMYR2 was capable of suppressing tumor growth and invasion partly via the upregulation of miR-21 functional targets and the reversal of epithelial–mesenchymal transition (EMT) in epithelial tumor cells in vitro as well as in orthotopic nude mouse models.

The MC-Fold/MC-Sym pipeline is a web-hosted service for RNA secondary and tertiary structure prediction. The pipeline consists of uploading RNA sequence to MC-Fold, which outputs secondary structures that are directly input to MC-Sym, which outputs tertiary structures. Pre-miRNA sequences were obtained from the miRBase database (http://microrna.sanger.ac.uk/sequences/). The hairpin loop of the pre-miRNA was chosen to predict the 3D structure using the MC-Fold/MC-Sym pipeline described previously. Energy optimization was further conducted on the 3D model using the TINKER Molecular Modeling Package (http://dasher.wustl.edu/tinker/).

For docking with AutoDock, polar hydrogen atoms, Kollman united charges, and solvent parameters were applied to the receptor using pmol2q script (http://www.sourcefiles.org/Scientific/Biology/Proteins/pmol2q_2.3.0.tar.gz). This script converts the .pdb file format of the protein template to the .pdbqt file format that is compatible with AutoDock program version 4 (http://autodock.scripps.edu/).

The National Cancer Institute (NCI) diversity dataset (http://dtp.nci.nih.gov/branches/dscb/diversity_explanation.html) contains 1,990 chemical structures. The ligand .pdbq files compatible with AutoDock program version 4 were prepared from the .pdb files using the prepare_ligand.py script.

High-throughput docking-based virtual screening was conducted using AutoDock program version 4. The rotational bonds of the ligands were treated as flexible, whereas the receptor was kept rigid. Grid boxes were fixed around the Dicer-binding site as the grid box center. Protein–ligand interactions were analyzed and visualized using Jmol (http://jmol.sourceforge.net/).

Human epithelial cancer cell lines (U87 and LN229 glioblastoma cells, MCF-7 and MDA-MB-231 breast cancer cells, and SGC7901 gastric cancer cells) were purchased from China Academia Sinica Cell Repository (Shanghai, PR China). The cell lines were tested 1 month before the experiments for authentication with the methods including morphologic analysis, growth curve analysis, and mycoplasma detection, which were conducted according to the American Type Culture Collection cell line verification test recommendations. Genotyping of the cells used are: U87 (PTEN del/P53 wt), LN229 (PTEN wt/P53 wt), MCF-7 (PTEN wt/P53 wt), MDA-MB-231(PTEN wt/P53 mut), and SGC7901 (P53 mut). The cells of MDA-MB-231 were maintained in RPMI medium 1640 (Gibco), whereas the others were maintained in Dulbecco's Modified Eagle Medium (DMEM, Gibco) supplemented with 10% FBS (Gibco), 2 mmol/L glutamine (Sigma), 100 μg/mL penicillin (Sigma), 100 μg/mL streptomycin (Sigma), and incubated at 37°C in 5% CO₂. AC1MMYR2 was obtained from the NCI diversity dataset and diluted to the indicated concentrations in cell culture media. Dicer siRNA sequence, As-miR-21 oligonucleotide, miR-21 mimics, and pre-miR-21 wild-type and mutant plasmids were purchased from GenePharma Co., Ltd. and transfections were conducted as previous described (18).

Real-time PCR (RT-PCR) and Western blot analyses were conducted according to the manufacturers' instructions as previously described (19). For Western blot analyses, the band density of specific proteins was quantified after normalization with the density of GAPDH.

FISH detection of miR-21 was conducted as previously described (20). Immunofluoresence staining was conducted using antibodies against E-cadherin, N-cadherin, β-catenin, and ZEB1/2 (1:100 dilutions; Cell Signaling Technology), and the cells were visualized using FV-1000 laser scanning confocal microscopes. The assays for colony formation, cell cycle, apoptosis, and Transwell were reported previously (21). Caspase-3/7 activity was measured using Caspase-Glo 3/7 reagent (Promega).

BALB/c-A nude mice at 4 weeks of age were purchased from the Animal Center at the Cancer Institute at Chinese Academy of Medical Science (Beijing, PR China). To establish intracranial gliomas, 0.5 × 10⁵ U87 glioblastoma cells transduced with luciferase lentivirus were implanted stereotactically (22). Seven days postimplantation, the mice were randomly assigned to two groups (n = 6 per group). The dimethyl sulfoxide (DMSO) group [100 μL DMSO/DMEM (1:1)] and the AC1MMYR2 group [AC1MMYR2 (25 mg/kg) dissolved in 100 μL DMSO/DMEM (1:1)] were intraperitoneally injected every 2 days during the survival period. Mice were imaged for Fluc activity using bioluminescence imaging on days 0 (the first day of drug administration), 7, 14, 21, 28, 35, and 47. In addition, the body weight and overall survival time of mice were also monitored. For MCF-7 and MDA-MB-231 orthotopic breast cancer models, 5 × 10⁶ MCF-7 or MDA-MB-231 tumor cells transduced with luciferase lentivirus were injected into mammary fat pads of each nude mouse. Seven days after implantation, AC1MMYR2 (25 mg/kg) or DMSO was administered by intraperitoneal injection (n = 6 per group) every 3 days for 35 days, mice were imaged for Fluc activity on days 0, 7, 14, 21, 28, and 35, and the tumor volume was calculated with a caliper using the formula: volume = length × width². At the end of the experiment on day 35, the mice were killed, tumor weight was measured, and mice hearts, lungs, livers, spleens, and kidneys were removed for bioluminescent imaging.

The paraffin-embedded tissue sections were used for hematoxylin and eosin (H&E) staining, TUNEL staining, and immunohistochemistry analysis as previous described (21). For immunohistochemistry analysis, the sections were incubated with primary antibodies (1:100 dilutions) overnight at 4°C, followed by a biotin-labeled secondary antibody (1:100 dilutions) for 1 hour at 37°C and then incubated with ABC-peroxidase and 3,3′-diaminobenzidine, counterstained with hematoxylin and visualized using a light microscope.

SPSS 16.0 (SPSS) was used for all calculations. All data are represented by the mean ± SD. Statistical significance was determined at P < 0.05.

To screen small-molecule inhibitors of miR-21, we proposed a high-throughput screening method for small molecules targeting miR-21 based on the 3D structure of the Dicer binding site on pre-miR-21. A schematic representation of our workflow is illustrated in Supplementary Fig. S1A. The sequence of the pre-miR-21 hairpin loop (Dicer binding site on pre-miR-21) was generated from the miRBase database and then input to the MC-Fold/MC-Sym pipeline to construct a 3D model (Supplementary Fig. S1B and S1C). For this model, we analyzed energy optimization using TINKER. Next, we conducted high-throughput molecular docking for pre-miR-21 against the 1,990 NCI/diversity compounds using the AutoDock program (Supplementary Table S1). The specific docking process can be visualized in Supplementary Video S1. Here, our dockings revealed 48 compounds with high-binding affinity (indicated by DeltaG, Supplementary Table S2). The top 5 were further assessed for miR-21 inhibition efficacy in 4 human epithelial cancer cells (U87 and LN229 glioblastomas cells, MCF-7 breast cancer cells, and SGC7901 gastric cancer cells). Both the compound AC1MMYR2 (also known as NSC211332) and the As-miR-21 presented strong inhibition of miR-21 expression in all cancer cells (Supplementary Fig. S2). Figure 1A and B display the chemical structure of AC1MMYR2 (2,4-diamino-1, 3-diazinane-5-carbonitrile) and its highest affinity with the best docking pose.

To evaluate the efficacy of AC1MMYR2 treatment, miR-21 expression was tested using RT-PCR. The compound inhibited mature miR-21 generation in a dose-dependent manner after 24 hours of treatment (Fig. 1C). Furthermore, AC1MMYR2 (30 μmol/L) caused a time-dependent reduction of mature miR-21, with an approximate 50% inhibition at 6 hours of treatment in all cancer cells (Fig. 1D). These data indicate that AC1MMYR2 could inhibit mature miR-21 generation in a time- and dose-dependent manner.

Next, we examined whether AC1MMYR2 could impact miR-21 functional targets. Several miR-21 targets have been evidenced, such as PTEN, PDCD4, and RECK. As shown in Fig. 1E, PTEN (U87 excluded for its deletion phenotype), PDCD4, and RECK expression levels were significantly induced on the basis of the Western blot assay. Taken together, AC1MMYR2 could repress mature miR-21 generation, thereby modulating the level of miR-21 targets.

The selectivity and specificity of AC1MMYR2 interruption miR-21 generation was further assessed by measuring intracellular miRNA levels using RT-PCR (Fig. 1F). We randomly chose 11 miRNAs: miR-23b, miR-566, miR-27b, miR-222, miR-524-5p, let-7i, miR-218, miR-1280, miR-200a/b, and miR-181d. After 6 hours of treatment, the expression levels of 11 miRNAs were virtually unchanged in the 4 cell lines. Twenty-four hours later, the expression levels of miR-23b, miR-566, miR-27b, miR-222, and miR-524-5p were significantly reduced, whereas the expression levels of let-7i, miR-218, miR-1280, and miR-200a/b were induced. Furthermore, miR-181d expression in U87, LN229, and MCF-7 cells increased, but the level in SGC7901 cells decreased. These results provided evidence that AC1MMYR2 is a specific miR-21 inhibitor, whereas the alteration of 11 miRNAs indirectly triggered by AC1MMYR2 requires further investigation.

Because the identification of small-molecule inhibitors targeting miR-21 was based on the 3D structure of the Dicer binding site on pre-miR-21, we evaluated whether the pre-miR-21 expression was changed. As expected, the level of pre-miR-21 increased in a time-dependent manner in all cell lines (Fig. 2A). Notably, after 6 hours of treatment, AC1MMYR2 significantly induced pre-miR-21 expression by approximately three-fold over the DMSO (0 hour) group. These results indicated that AC1MMYR2 might directly interact with pre-miR-21 at the binding site of Dicer to prevent the cleavage of pre-miR-21 to the mature miRNA. However, as displayed in Fig. 2B, AC1MMYR2 treatment resulted in the time-dependent reduction of pri-miR-21. After 6 hours of treatment, the compound greatly repressed pri-miR-21 expression by approximately 50% relative to the DMSO group.

PTRE-miR-21(FL) plasmid expresses a full-length miR-21 pri-miRNA transcript and serves as a precursor for the mature miR-21(23). Meanwhile, one key step of mature miR-21 biogenesis from precursors is Dicer removing the terminal loop region (3, 4). Thus, we conclude that pTRE-miR-21(FL) plasmid induces mature miR-21 via Dicer processing. To further determine the effect of AC1MMYR2 on pri-miR-21 and pre-miR-21 expression, we transfected the miR-21 mimics and PTRE-miR-21(FL) plasmid after the compound treatment in U87 and MCF-7 cells, respectively (Fig. 2C). This indicated that miR-21 mimics increased the mature miR-21 levels, whereas the PTRE-miR-21(FL) plasmid only induced pri-miR-21 and pre-miR-21 expression levels. These data confirmed that AC1MMYR2 blocked Dicer processing at the binding site of pre-miR-21 to repress the mature miR-21 generation.

To verify the potential site of Dicer binding on pre-miR-21 that AC1MMYR2 acts on, we constructed 3 pre-miR-21 point mutation plasmids, MUT1 (base “a” was deleted), MUT2 (base “c” was switched to “t”) and MUT3 (base “c” was switched to “g”; Fig. 2D), and a pre-miR-21 wild-type plasmid (WT). After transfection with the 4 plasmids in U87 and MCF-7 cells, pre-miR-21 expressions were all upregulated in the transfected groups, compared with control. Moreover, there was no statistic significance between MUTs and wild-type groups in pre-miR-21 overexpression. However, overexpression of mature-miR-21 was detected merely in wild-type group; there were no significant changes in MUTs groups in comparison with control (Fig. 2E). Therefore, we concluded that the point mutations at Dicer-binding site on pre-miR-21 disturbed Dicer processing pre-miR-21 to mature miR-21.

To analysis whether these point mutations also influenced AC1MMYR2 action, we created the 3D structure of pre-miR-21 hairpin loop with such mutations. After AC1MMYR2 respectively docking with these 3D models (Fig. 2F), the estimated free energy of binding was −5.15 kcal/mol, −4.94 kcal/mol, and −5.84 kcal/mol, whereas the binding affinity of wild-type was −12.04 kcal/mol (Supplementary Table S2). Consequently, we confirmed that point mutations at Dicer binding site might attenuate AC1MMYR2 binding affinity and thus weaken its activity. Therefore, AC1MMYR2 probable specifically blocked the Dicer binding site on pre-miR-21. Besides, we proved that AC1MMYR2 selectively targeted tumor cells with miR-21 high expression (Supplementary Fig. S3), which also showed the specificity of AC1MMYR2 towards miR-21 to some extent.

As suggested, Dicer is a central enzyme in miRNA processing (24). To further investigate whether the AC1MMYR2 specific interruption of Dicer processing correlated with its expression, we reduced Dicer levels using a siRNA sequence in U87 and MCF-7 cells. Dicer expression was greatly reduced after transfection relative to the control and scramble (Fig. 3A). Meanwhile, cells transfected with the Dicer siRNA sequence alone reduced the mature miR-21 expression; however, AC1MMYR2 retreatment after transfection did not induce a more significant decrease (Fig. 3B). Confocal microscopy also displayed miR-21 staining to be attenuated in the cytoplasm of siRNA-transfected cells, whereas no obvious change occurred after compound retreatment (Fig. 3C). Taken together, these data indicate that AC1MMYR2 functions are dependent on Dicer expression. The compound could not effectively repress mature miR-21 generation when Dicer levels were low and the underlying mechanism demands intensive study.

Excessive epithelial cell proliferation is a hallmark of the initiation and early growth of primary epithelial cancers (25). After confirming the target of AC1MMYR2 action, we went on to test the compound effect on tumor cell biologic characteristics, especially proliferation and apoptosis. FISH assessment displayed that miR-21 expression was remarkably decreased after AC1MMYR2 treatment in U87, MCF-7 (Fig. 4A), LN229, and SGC7901 cells (Supplementary Fig. S4A). Moreover, reduced miR-21 resulted in decreased colony formation rates (Fig. 4B and Supplementary Fig. S4B) and flow cytometry analysis showed that the cell cycle was blocked in the G₀–G₁ phase (Fig. 4C and Supplementary Fig. S4C).

GO analysis for the biologic processes was conducted on the 534 genes whose expression was significantly altered in response to 6-hour treatment with AC1MMYR2 in U87 glioma cells (Supplementary Table S3). “Regulation of cell proliferation” was the sixth term on the top 20 most statistically significant (by P value) GO categories (Supplementary Table S4). Meanwhile, KEGG pathway analysis revealed that there were 14 different pathways corresponding to the target genes, of which “pathways in cancer” was most significant (P = 5.7E-4, Supplementary Table S5).

Annexin V/propidium iodide (PI) staining was conducted to quantify the apoptosis induced by the compound. Compared with the DMSO group, the AC1MMYR2 treatment caused more apoptosis and the apoptosis rates were approximately 15% (Fig. 4D and Supplementary Fig. S4D). Specifically, the cells that respond to AC1MMYR2 also had higher caspase 3/7 activities (Fig. 4E and Supplementary Fig. S4E). On the basis of the above findings, we conclude that AC1MMYR2, as a miR-21 inhibitor, is potent at suppressing tumor cell growth and survival.

Migration and invasion are crucial biologic characteristics for tumor cell metastasis (26) and one mission-critical step in the metastatic cascade is the process of EMT (27). To investigate whether AC1MMYR2 could impact the migration and invasiveness of tumor cells, Transwell assays and EMT-associated markers expression levels were detected. As shown in Fig. 5A and Supplementary Fig. S5A, fewer AC1MMYR2-treated tumor cells invaded across the precoated Matrigel membrane compared with the DMSO-treated cells. In addition, Western blot analysis indicated that E-cadherin expression was increased, whereas N-cadherin, β-catenin, ZEB1, ZEB2, and MMP9 expression levels were reduced after compound treatment (Fig. 5B and Supplementary Fig. S5B).

Importantly, because the activities of transcriptional factors are usually associated with their subcellular location, we further assessed the distribution of the proteins mentioned above by confocal microscopy. As displayed in Fig. 5C and Supplementary Fig. S5C, E-cadherin plasmalemmal expression was upregulated after AC1MMYR2 treatment, accompanied with a decrease in N-cadherin plasmalemmal expression. Moreover, in the DMSO group, β-catenin expression was mainly localized to the nucleus; however, after compound treatment, nuclear β-catenin expression was reduced and translocated to the cytoplasm. ZEB1 and ZEB2 nuclear expression levels were also reduced in the compound-treated group.

Our in vitro study suggested that AC1MMYR2 was a potent miR-21 inhibitor that suppressed tumor cell growth and invasion. To further confirm this, a proof-of-principle experiment was used, using an orthotopic glioma model, two orthotopic breast cancer models, and an AC1MMYR2-mediated antitumor therapeutic approach.

For U87 glioma intracranial model, bioluminescence imaging showed tumor growth stasis in the AC1MMYR2-treated group compared with the DMSO-treated tumors and on day 14, a statistically significant difference in tumor volume appeared between the 2 groups (Fig. 6A and B). The body weights of the mice in the treatment group increased constantly during their overall survival time, whereas the DMSO-treated mice lost in weight (Fig. 6C). Only 2 mice in the AC1MMYR2 treatment group died on day 47, whereas there were no mice alive in the DMSO treatment group (Fig. 6D). Furthermore, H&E staining displayed that the tumors formed by DMSO-treated cells exhibited extensive branch-like growing patterns and vessels that spread into the surrounding tissue. In contrast, AC1MMYR2-treated cells formed oval-shaped tumors with smooth margins and a noninvasive front, further validating that the invasive behavior of glioma could be suppressed by AC1MMYR2. Meanwhile, miR-21 expression was markedly decreased, but apoptotic nuclei were increased in the compound treatment group (Fig. 6E).

We also stained for an astrocyte differentiation marker, GFAP, existed during the process of differentiation (28). A significant increase in GFAP expression was observed in cells treated with AC1MMYR2, which indicated that the malignancy of glioblastomas was attenuated by AC1MMYR2 to a certain extent. Besides, the expressions of PCNA, Ki67, cyclin D1, β-catenin, ZEB1/2, and MMP9 decreased, whereas RECK, PDCD4, and E-cadherin expression levels were simultaneously increased in AC1MMYR2-treated group relative to the DMSO-treated group (Fig. 6F). Furthermore, H&E staining of multiple tissues showed no discernible toxicity of AC1MMYR2, including in the liver and kidney (Supplementary Fig. S6).

As previously reported, MDA-MB-231 breast cancer cells are highly metastatic compared with MCF-7 cells (29). AC1MMYR2 also inhibited MDA-MB-231 cells proliferation and invasion in vitro (Supplementary Fig. S7–S9). To further monitor treatment efficacy in live animals, we constructed not only a MCF-7 orthotopic breast cancer model but also a highly metastatic MDA-MB-231 orthotopic model. Bioluminescence imaging showed tumor growth stasis in the AC1MMYR2-treated group compared with the DMSO-treated tumors and on day 7, a statistically significant difference appeared in tumor volume (Fig. 7A–C). At the termination of the study, the tumor weights were significantly different between the 2 groups (Fig. 7D) and we also found lung metastasis in MDA-MB-231 orthotopic model but no organ metastasis in MCF-7 orthotopic model. AC1MMYR2 treatment significantly reduced the MDA-MB-231 cells lung metastasis (Fig. 7E). Meanwhile, tumors derived from DMSO-treated groups revealed more nuclei and darker chromatin staining compared with the AC1MMYR2 treatment group, as determined by H&E staining. The expression levels of PTEN, RECK, PDCD4 and E-cadherin, and apoptotic nuclei were increased, whereas miR-21, PCNA, cyclin D1, Ki67, N-cadherin, β-catenin, ZEB1/2, and MMP9 expressions were significantly reduced in tumor specimens from the compound treatment group (Supplementary Fig. S10A and S10B).

In summary, our in vivo preclinical efficacy trials showed a remarkable antitumor effect of AC1MMYR2 in both glioblastoma and breast cancer orthotopic models.

Although the existing anticancer drugs show promise during early treatment, neoplasms of high malignancy and metastasis are still considered incurable. Therefore, there is an urgent need to develop potential and novel drugs for cancer treatment (30). Currently, miRNA-based gene therapy offers the theoretical appeal of targeting multiple gene networks and has garnered increasing attention (31). However, given the challenges involved in the use of nucleotide analogs, the development of natural small-molecule inhibitors targeting specific miRNAs and regulating their activities would be a promising approach (32). With the recent deeper understanding of miRNA structure and the thermodynamics of miRNA–small-molecule interactions, computer-aided drug design offers promise because it can improve the efficiency of the drug pipeline dramatically in a cost-effective way compared with traditional strategies for RNA-targeted lead identification (33). To our knowledge, Gumireddy and colleagues (34) reported the first use of small molecules as inhibitors of miRNA-21. They conducted a primary screen of more than 1,000 compounds, conducted structure–activity relationship analyses, and identified diazobenzene and its derivatives as effective inhibitors of miR-21 function. Recently, Bose and colleagues (35) screened 15 known aminoglycosides and determined that the tuberculosis drug, streptomycin, was also a small miR-21 inhibitor. However, both of these studies focused on the drug screening process and their chemical characteristics and did not further investigate antitumor efficacy, particularly in preclinical pharmacodynamic studies in vivo.

In this study, we constructed a 3D model of the pre-miR-21 hairpin loop and Dicer binding domain. By high-throughput molecular docking to the 1,990 NCI diversity compounds and further miR-21 inhibition analysis, AC1MMYR2 was identified as a specific miR-21 small-molecule inhibitor, which specifically blocked Dicer processing pre-miR-21 to mature-miR-21. Recently, a transcriptional regulatory circuit involving miR-21, PTEN-AKT pathway, and NF-kB-IL-6-STAT3 positive feedback loop was reported in cancers. Inhibition of miR-21 expression increases the PTEN expression, thereby reduces the AKT activity and subsequently inhibits NF-kB activity and interleukin (IL)-6 production. Moreover, STAT3, a transcription factor activated by IL-6, directly activates miR-21 (36). On the basis of these, we supposed that overexpression of PTEN induced by AC1MMYR2 inhibited the AKT activity and then suppressed NF-kB-IL-6-STAT3 positive feedback loop (PTEN and pSTAT3 expressions were presented in Fig. 1E), ultimately resulted in the inhibition of pri-miR-21 transcription. Thus, we established a positive feedback loop between STAT3 and pri-miR-21, which certainly required further investigation in future.

PTEN, PDCD4, and RECK, which are considered to be associated with tumor cell survival, proliferation, and invasion (37, 38, 20), were upregulated after AC1MMYR2 treatment. More importantly, altered expression levels of a set of miRNAs, such as miR-200a/b and miR-181d, were observed after AC1MMYR2 treatment. It is reported that miR-21 overexpression in epithelial cancers facilitates the initiation of EMT (39, 40). EMT is an evolutionarily conserved process that allows polarized epithelial cells to undergo multiple biochemical changes to adopt mesenchymal cell characteristics, such as enhanced migratory capacity, invasiveness, and elevated resistance to apoptosis (41). Several transcriptional factors, including Snail1/2, ZEB1/2, and Slug have been reported to be involved in EMT (42). On the basis of the preceding observations, we indicated that miR-200a directly interacted with β-catenin and ZEB1/2 to reverse EMT in U251 glioblastoma and SGC7901 gastric cancer cells (43) and that β-catenin and ZEB1 were also functional targets of miR-181d (unpublished data). In addition, PTEN inhibition of β-catenin activation via the downregulation of pAKT has been verified in our previous studies (8, 44). Thus, altered miRNAs and PTEN expression levels may impact the EMT process via the regulation of β-catenin, ZEB1, and ZEB2. To further confirm this, we detected epithelial and mesenchymal markers and EMT-related transcriptional factor expressions after compound treatment.

In many cancer types, the loss of E-cadherin coincides with a gain of expression of the mesenchymal cadherin, N-cadherin. This “cadherin switch” is considered to be a hallmark of EMT (45). However, it was reported that epithelial cell adhesion complexes reorganized and cell proliferation was suppressed upon exogenous expression of the E-cadherin gene (46); simultaneously, the cells that passed through EMT lost their mesenchymal phenotype (41). Increased expression of E-cadherin was also well established to be an antagonist of invasion and metastasis (47). Sequestration of β-catenin in the cytoplasm was essential for the preservation of the epithelial features of cancer cells. When β-catenin relocalizes from the nucleus to the cytoplasm, it can reverse the migratory characteristics of the cell by reestablishing interactions with E-cadherin and forming cell–cell adhesions (48). Zinc finger E-box binding homeobox proteins, ZEB1 and ZEB2, which are well known transcriptional repressors of E-cadherin, are both implicated in EMT and tumorigenesis (49). Thus, cells overexpressing β-catenin, ZEB1, and ZEB2 cause them to be more susceptible to EMT. In our study, in combination with upregulated E-cadherin, AC1MMYR2 significantly reduced the nuclear expression levels of β-catenin and ZEB1/2, as well as the expression of MMP9, thereby reversing EMT in the epithelial cancers. This may be another underlying mechanism of AC1MMYR2 to facilitate the suppression of the tumor growth and invasion.

Moreover, it remains highly interesting that an astrocyte differentiation marker, GFAP, had its expression level induced in U87 intracranial samples after AC1MMYR2 treatment. A recent study observed the coexpression of GFAP/Slug and GFAP/Twist in mesenchymal areas of gliosarcoma and indicated that it may represent an initial step in glial–mesenchymal transition (50). On the basis of these observations, we considered that a potential association might exist between GFAP levels and EMT. Increased GFAP expression, with the simultaneous reduction of mesenchymal marker expression levels induced by AC1MMYR2, perhaps indicated a reacquisition of epithelial characteristics, which certainly requires further investigations.

In this study, we propose a novel high-throughput screening method for small molecules targeting miRNAs and identify a potent and specific miR-21 inhibitor, AC1MMYR2, which shows remarkable antitumor efficacy not only through an induction of common functional targets (PTEN, PDCD4, and RECK) but also the reversibility of epigenetic mechanisms (i.e., reversion of EMT through upregulation of miR-200a/b and miR-181d; Supplementary Fig. S11). Nevertheless, the underlying mechanism of AC1MMYR2 upregulation of miR-200a/b and miR-181d warrants further study. Taken together, our results showed the robust antitumor efficacy of AC1MMYR2 without any observable toxicity to normal tissues, which highlighted its promising use clinically. AC1MMYR2 compound derivatives might have potential as anticancer agents in epithelial cancers in the future.

No potential conflicts of interest were disclosed.

Conception and design: J. Zhang, C. Kang

Development of methodology: Z. Shi, L. Han, Y. Ren, C. Kang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Z. Shi, X. Qian, L. Chen, C. Kang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Z. Shi, J. Zhang, L. Han, K. Zhang, J. Liu, Y. Ren, M. Yang, A. Zhang, C. Kang

Writing, review, and/or revision of the manuscript: Z. Shi, J. Zhang, X. Qian, C. Kang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Kang

Study supervision: P. Pu, C. Kang

The authors thank Dr. Bryan R. Cullen, Department of Molecular Genetics & Microbiology and Center for Virology, Duke University Medical Center (Durham, North Carolina) for providing pTRE-miR-21(FL) plasmid and thank the members of Shanghai Sensichip Co Ltd., Shanghai 200433, PR China, for their great work with 3D structure prediction, molecular docking, and high-throughput virtual screening process of the compound.

This work was supported by the China National Natural Scientific Fund (81172406), the National High Technology Research and Development Program 863 (2012AA02A508), the Natural Science Foundation of Tianjin Municipal Science and Technology Commission (12JCZDJC24300), and the Research Fund for the Doctoral Program of Higher Education of China (20111202110004).

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.