Programmed Cell Death 4 (PDCD4) Is an Important Functional Target of the MicroRNA miR-21 in Breast Cancer Cells*

MicroRNAs are emerging as important regulators of cancer-related processes. The miR-21 microRNA is overexpressed in a wide variety of cancers and has been causally linked to cellular proliferation, apoptosis, and migration. Inhibition of mir-21 in MCF-7 breast cancer cells causes reduced cell growth. Using array expression analysis of MCF-7 cells depleted of miR-21, we have identified mRNA targets of mir-21 and have shown a link between miR-21 and the p53 tumor suppressor protein. We furthermore found that the tumor suppressor protein Programmed Cell Death 4 (PDCD4) is regulated by miR-21 and demonstrated that PDCD4 is a functionally important target for miR-21 in breast cancer cells.

Since their discovery (1)(2)(3)(4), microRNAs (miRNAs) 3 have emerged as integrated and important post-transcriptional regulators of gene expression in animals and plants (5,6). In animals, miRNAs bind to partly complementary sequence motifs present predominantly within the 3Ј-untranslated regions (UTRs) and mediate translational repression, sometimes involving degradation of the target mRNA (7)(8)(9). miRNAs have been found implicated in a multitude of cellular processes including proliferation, differentiation, migration, and apoptosis (10 -12). Accordingly, aberrant miRNA expression has been linked to diseases, including cancer (13)(14)(15). Evidence for the causal involvement of miRNAs in cancer comes from several sources. First, mapping of fragile sites and chromosomal regions lost or amplified in cancer displays an intriguing overlap with the localization of miRNA genes (16,17), and detailed mapping studies have demonstrated loss of the miR-15/miR-16 genes in chronic lymphocytic leukemia and of the miR-34a gene in neuroblastomas (18,19). Reintroduction of these "tumor suppressor" miRNAs stalls the proliferation of cancer cells or induces apoptosis (18, 20 -22). Second, genetic studies in model organisms have led to the identification of miRNAs with relevance for human cancer. This is exemplified by the analysis of let-7, which in Caenorhabditis elegans targets let-60/ RAS and in humans is lost in some lung cancers, leading to overexpression of N-RAS (23). Third, forward genetic studies have demonstrated a role for the miR-17-92 cluster in lymphoma development in mice predisposed to cancer (24), and screenings of miRNA expression libraries in cell culture models of cancer have identified miR-372/373 and miR-221/222 as cancer-promoting miRNAs via repression of the tumor suppressor proteins LATS2 and p27, respectively (25,26).
MicroRNA profiling studies of human tumors have shown cancer type-specific deregulation of miRNA expression and have identified a number of miRNAs with putative tumor suppressor or oncogenic functions (13,15). Interestingly, miR-21 stands out as the miRNA most often found overexpressed in solid tumors (27), and increased levels of miR-21 have been found in very diverse cancer types including glioblastoma, breast, liver, and pancreatic cancers (11,(27)(28)(29)(30). Furthermore, causal links between miR-21 expression and cancer-related processes such as proliferation, migration, apoptosis, and tumor growth have been demonstrated in human hepatocellular and breast cancer cells (11,31).
Previously identified targets for miR-21 include the tumor suppressors tropomyosin 1 in breast cancer cells (32) and phosphatase and tensin homolog (PTEN) in hepatocellular carcinomas (11,33). The widespread occurrence of miR-21 overexpression in cancer and the relatively few experimentally defined targets prompted us to perform expression array analyses of breast cancer cells transfected with miR-21 inhibitors to identify and functionally validate additional targets of miR-21.

EXPERIMENTAL PROCEDURES
Antibodies and Western Blot Analysis-MCF-7 cells were seeded in 6-well plates and transfected the following 2 days. The cells were harvested 5 days after the first transfection, washed once in phosphate-buffered saline, and lysed in radioimmune precipitation buffer (150 mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS, 50 mM Tris-HCl, pH 8, 2 mM EDTA) containing 1 mM dithiothreitol, 1 mM Pefabloc (Roche Applied Science), 1ϫ Complete Mini protease inhibitor mixture (Roche Applied Science), 1 mM NaVO3, 10 mM NaF, 10 mM pyrophosphate, and 50 mM ␤-glycerophosphate. 15 g of pro-tein/lane was separated on a 4 -12% NuPAGE Bis-Tris gel (Invitrogen) and transferred to a nitrocellulose membrane. The PDCD4 antibody was kindly provided by Dr. Iwata Ozaki, Japan. The p53 and CDK6 antibodies were purchased from Santa Cruz Biotechnology. The cofilin 2 antibody was purchased from Cell Signaling and the vinculin antibody from Sigma-Aldrich.
miRNA Precursors, Anti-miRNA Oligonucleotides, and siRNAs-The locked nucleic acid (LNA)-modified oligonucleotide inhibitors used for miRNA knock down were purchased from Exiqon. The miRNA precursor hairpins were purchased from Ambion. The PDCD4 SMARTpool siRNA was purchased from Dharmacon.
Growth Curves and Viability Assays-MCF-7 cells were seeded in 24-well plates and transfected the following day with 50 nM LNA or siRNA using Lipofectamine 2000. Cells were fixed at indicated time points in 4% paraformaldehyde, stained in a 0.1% crystal violet solution, and resuspended in 10% acetic acid. Sample absorbance was measured at 620 nm. For cell viability assays, MCF-7 cells were seeded in 24-well plates and transfected the following day with 50 nM LNA or siRNA using Lipofectamine 2000. After 5-6 days of growth, cells were incubated for 4 h at 37°C with 0.5 mg/ml 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (Sigma-Aldrich). After incubation, supernatants were discarded and formazan crystals were dissolved in a 10% formic acid, 90% isopropanol solution. Sample absorbance was measured at 570 nm with a reference wavelength of 690 nm.
Affymetrix Microarray-MCF-7 wells were seeded in 10-cm plates in Dulbecco's modified Eagle's medium containing 10% fetal calf serum and triplicate independent transfections performed the following day with either 50 nM LNA-miR-21 or scrambled control LNA using Lipofectamine 2000. Total RNA was harvested 24 h post-transfection using TRIzol reagent. Affymetrix microarray analysis (HG-U133 Plus 2.0 human) was performed at the Microarray Center, Rigshospitalet, Copenhagen University Hospital. Briefly, 2 g of total RNA was used to synthesize double-stranded cDNA using Superscript Choice System (Invitrogen) with an oligo(dT) primer containing a T7 RNA polymerase promoter. The cDNA was subsequently used as template for an "in vitro" transcription reaction generating biotin-labeled antisense cRNA (BioArrayTM High Yield RNA Transcript Labeling kit; Enzo Diagnostics, Farmingdale, NY).
After fragmentation at 94°C for 35 min in 40 mM Tris, 30 mM MgOAc, 10 mM KOAc, samples were hybridized for 16 h to Affymetrix HG-U133 2.0 human arrays (Affymetrix, Santa Clara, CA). The arrays were washed and stained with phycoerythrin-conjugated streptavidin (SAPE), and the arrays were scanned in the Affymetrix GeneArray 2500 scanner, exactly as described in the Affymetrix GeneChip protocol.
Bioinformatic Analyses-The expression data were processed using the "affy" package in BioConductor. Probe set intensities were summarized using the Robust Multichip Average method and then transformed to generalized log values (approaches the natural logarithm for high values) with the variance stable VSN method (35). In a final step, the six arrays were qspline-normalized. Differentially expressed genes were selected by a t test (p Ͻ 0.05), and two sets of probe sets were defined, the up and downregulated, by additionally requiring reasonable average expression intensities (Ͼ2.87, 1st quartile) and high -fold changes (Ͼ Ϯ 0.168, mean Ϯ 2 ϫ S.D.) (supplemental Table  S1, a and b). A third set of probe sets that do not change from control to experiment was defined by selecting probe sets with stable expression intensities (p Ͼ 0.95 and expression intensity S.D. Ͻ 0.06) and reasonable expression (Ͼ2.87) (supplemental Table S1c). The up, down, and no-change sets comprised 339, 258, and 195 probe sets, respectively.
The probe sets were subsequently mapped to Ensembl transcripts (version 45) using the mappings provided at BioMart. Probe sets that mapped to two different Ensembl genes were discarded. Transcripts with 3Ј-UTR sequences shorter than 50 nucleotides and copies of transcript isoforms of the same gene with matching 3Ј-UTR sequences were discarded. The up, down, and no-change sets comprised 402, 335, and 262 transcripts, corresponding to 272, 215, and 164 genes. For the analysis of seed site enrichment, miRNA sequences from miRBase version 9.2 were used (36). The 3Ј-UTRs of the transcripts were scanned for matching 6-, 7-1A, 7-, and 8-mer (perfect 8-nucleotide match) miRNA seed sites (37). In the definition used, seed sites are contained in each other, that is, a given 6-mer site always also corresponds to a 7 mer and 8 mer. Note that this definition is different from the seed sites reported at TargetScan. The difference in the fraction of transcripts having seed sites in the up, down, and no-change sets was evaluated using one-sided p values from Fisher's exact test. When comparing all miRBase miRNAs, the -fold enrichment for a given miRNA is calculated by dividing the fraction of transcripts with a 7-mer seed site in the up set with the same fraction in the down (or nochange) set.
In an unbiased word analysis, all words of length 7 were investigated for over-representation in the up versus the down sets. The words were ranked by p values from Fishers exact test.

Suppression of MCF-7 Cell
Growth by Inhibition of miR-21-Several studies have demonstrated that miR-21 is an oncogenic miRNA with anti-apoptotic potential. Inhibition of miR-21 leads to growth suppression and apoptosis in glioblastoma and breast cancer cell lines, and loss of miR-21 can inhibit MCF-7 cell-derived tumor growth in vivo (30,31,33). MCF-7 cells express substantial amounts of miR-21 (supplemental Fig. S1), and consistent with previous findings, we observed a dose-dependent suppression of MCF-7 cell growth upon inhibition of miR-21 with an LNA-derived oligonucleotide inhibitor (Fig. 1A). Co-transfection of the LNA inhibitor with a luciferase reporter containing perfect complementarity to the mature miR-21 sequence (pmiR-21-luc) results in marked de-repression of luciferase activity, demonstrating a highly effective inhibition of endogenous miR-21 mediated by the LNA inhibitor (supplemental Fig. S2). The underlying mechanism of the role of miR-21 in tumorigenesis remains unclear, as only few targets for this miRNA have been experimentally verified (11,32). Meng et al. (11) have recently shown that the tumor suppressor PTEN is a direct functional target of miR-21 in human hepatocellular cancer cell lines. Given the importance of PTEN in regulating the PI3K/AKT pathway and the frequency of PTEN mutations or silencing in a variety of cancers (38), this constitutes an appealing explanation for the overexpression of miR-21 observed in many cancer types (27,28). To investigate the role of the PTEN-miR-21 interaction in breast cancer cells, we transfected MCF-7 cells with a miR-21 precursor, a miR-21 inhibitor, and appropriate controls. Interestingly, these treatments caused only subtle changes in PTEN protein levels (supplemental Fig. S3), suggesting that cell-and tissue type-specific differences may result in different functional miR-21 targets.
Identification and Validation of miR-21 Targets by Microarray Analysis-To identify targets of miR-21 that can explain the proliferation defect observed upon miR-21 inhibition in breast cancers cells, we performed microarray expression analyses using total RNA harvested from MCF-7 cells 24 h after transfection with either LNA-miR-21 or a scrambled control LNA not affecting cellular proliferation (Fig. 1B). We reasoned that cellular mRNAs subjected to increased degradation due to binding of endogenous miR-21 would be up-regulated upon miR-21 inhibition and that specific inhibition of the endoge-

miR-21 Targets PDCD4 in Human Breast Cancer Cells
nous miR-21 may cause fewer off-target effects than transfection with exogenous miRNA. Following normalization and statistical analysis we found 737 transcripts with significantly different expression between the LNA-miR-21-transfected cells and the controls, of which 402 (55%) were up-regulated and 335 (45%) down-regulated upon miR-21 inhibition ( Fig. 2A and supplemental Table S1). To verify the array analysis, 18 up-regulated mRNAs were validated by quantitative reverse transcription PCR analysis of RNA from an independent experiment. Ten of the genes chosen for validation contained at least one miR-21 7-mer seed match, two contained a 6-mer, and six did not contain miR-21 seed matches in their 3Ј-UTR. All 18 genes showed, to varying degrees, increased mRNA expression levels upon miR-21 inhibition (Fig. 2B). It is unclear whether the mRNAs without matches to the miR-21 seed sequence are direct targets or whether their regulation is a result of secondary effects.
To get a qualitative assessment of the data, we analyzed for the presence of different types of miR-21 seed matches among the genes regulated by miR-21 inhibition. Notably, we found very significant overrepresentations of all miR-21 seed match categories among the genes up-regulated by miR-21 inhibition relative to gene sets that exhibited no change or down-regulated expression, strongly suggesting that inhibition of endogenous miRNAs can be used to identify bona fide targets (Fig. 2C). In effect, the motif complementary to a 7-mer miR-21 seed sequence (or miR-590, holding the exact same seed sequence) is by far the most prevalent motif when analyzing the 3Ј-UTRs of the up-regulated transcripts against all seed sequences present in miRBase (36) (Fig. 2D). In addition, an unbiased analysis for the frequency of all possible 7-mer sequence motifs, regardless of whether these match known mi-RNAs, shows that the miR-21 complementary motif is very highly enriched (p Ͻ 4 ϫ 10 Ϫ12 ) and represents the most frequently occurring sequence motif in the 3Ј-UTRs of the up-regulated versus the down-regulated transcripts. Hence, the data strongly suggest that expression array analysis following inhibition of endogenous miRNAs is a strong tool to identify miRNA targets subjected to increased mRNA degradation.
Potential Involvement of p53 in the miR-21 Pathway-Among the transcripts up-regulated upon miR-21 inhibition we noticed the presence of several mRNAs known to be regulated by the p53 tumor suppressor, including FAM3C, ACTA2, APAF1, BTG2, FAS, CDKN1A (p21), and SESN1. We validated the up-regulation of these p53-regulated mRNAs by quantitative reverse transcription PCR (Fig. 2B). A connection between miR-21 and p53 was further substantiated using the Ingenuity

miR-21 Targets PDCD4 in Human Breast Cancer Cells
Systems pathway analysis program (data not shown). Given the pivotal importance of p53 in protecting cells from cancerpromoting events such as genomic instability and aberrant oncogene activation (39) and the high frequency of miR-21 overexpression found in a broad variety of tumors, we speculated that pathways affected by miR-21 and p53 could be interconnected. To address the importance of p53 in mediating the proliferation effects observed upon miR-21 inhibition, we developed a stable p53 knockdown cell line (MCF-7 shp53) by retroviral transduction with a pRetroSuper short hairpin RNA construct directed against p53 (Fig. 3A). We did not observe any difference in the proliferative capacity of MCF-7 shp53 relative to MCF-7 EV (not shown). Importantly, the MCF-7 shp53 cells were significantly less sensitive to the growth inhibitory effect of LNA-miR-21 (p Ͻ 0.001), relative to control MCF-7 cells transduced with an empty vector (Fig. 3B). Although we did not observe a significant change in p53 protein levels following miR-21 inhibition or overexpression (supplemental Fig. S4), the data suggest that miR-21 antagonizes the p53 pathway by inhibiting expression of p53-regulated genes, demonstrating an important functional link between these two signaling pathways.
Putative Targets Are Directly Regulated by miR-21-To substantiate that miR-21 is a direct regulator of the up-regulated transcripts, we selected six target genes containing potential miR-21 binding sites within their 3Ј-UTRs for further validation by luciferase reporter assays. 400 -500-base pair fragments of the 3Ј-UTRs were cloned into a modified pGL3 vector, downstream of the luciferase gene. Upon co-transfection in HEK293 cells, miR-21 significantly repressed the expression of all six constructs relative to a lin-4 control (p Ͻ 0.001) (Fig. 4A). The empty pGL3 vector was not significantly affected by miR-21, highlighting the importance of the 3Ј-UTR regions in mediating this regulation. In addition, we tested the effect of the miR-21 inhibitor on all constructs in MCF-7 cells. Only one of the six constructs, pGL3-PDCD4, showed a significant increase in luciferase expression upon LNA-miR-21 treatment relative to scrambled LNA (p Ͻ 0.001), suggesting that additional mechanisms control the expression of the remaining constructs (Fig. 4B).
To investigate miR-21 regulation of endogenous target proteins, three targets for which functional antibodies could be obtained were confirmed by Western blot analysis. Upon transfection with the miR-21 precursor, PDCD4, CDK6, and cofilin 2 protein levels were all reduced relative to the lin-4 control. In addition, miR-21 inhibition led to a corresponding increase in endogenous protein levels relative to the effect of the scrambled control (Fig. 4C).
Depletion of PDCD4 Abrogates the LNA-miR-21-mediated Phenotype in MCF-7 Cells-PDCD4 is a tumor suppressor known to be up-regulated during apoptosis (40) and down-regulated in several cancer forms (41)(42)(43). The predicted interaction between the PDCD4 3Ј-UTR and miR-21 is illustrated in Fig. 5A. To further substantiate a direct regulation of pGL3-PDCD4 by miR-21 we introduced a single (pGL3-PDCD4MUT1) or double (pGL3-PDCD4MUT2) point mutation in the seed sequence of pGL3-PDCD4. Whereas miR-21 caused only a slight regulation of pGL3-PDCD4MUT1, pGL3-PDCD4MUT2 remained unaffected by miR-21, suggesting a direct interaction between miR-21 and PDCD4 mediated through the seed region (Fig. 5B). Given the evidence presented above of PDCD4 regulation by miR-21 at both the RNA and the protein levels and considering the reported tumor suppressor activity of PDCD4, we speculated that PDCD4 could be a functionally important target of miR-21. To investigate the biological importance of PDCD4 as a target of miR-21, we depleted MCF-7 cells of PDCD4 protein by siRNA and assayed the effect of miR-21 inhibition (Fig. 5, C and D). Although PDCD4 depletion itself had no effect on cellular proliferation (supplemental Fig. S5), it significantly alleviated the anti-proliferative effect of miR-21 inhibition from 58 to 87% of control levels (p Ͻ 0.003) (Fig. 5D). The data therefore suggest an essential role for PDCD4 as a mediator of the biological effects of miR-21 in breast cancer cells.

DISCUSSION
Over the past few years the vast potential of miRNAs as regulators of cancer-related signaling pathways has fully emerged (10,14). Understanding the connections between miRNAs deregulated in cancer and cellular signaling pathways involved in cancer has been hampered by our limited knowledge of miRNA target recognition. Although several studies have demonstrated a central role of the miRNA seed region in target binding (13,37), additional binding requirements and constraints likely exist. In addition, studies have reported functional miRNA binding without perfect complementarity to the

miR-21 Targets PDCD4 in Human Breast Cancer Cells
seed region (44,45). Hence, there is a need for experimental approaches to target identification in order to gain knowledge of the mechanisms and modalities of miRNA target recognition. A number of studies have identified putative miRNA targets by microarray analysis of total RNA following transfections with miRNA duplexes (7,37,46). Reasoning that high concentrations of exogenous miRNAs could lead to the identification of false positives, especially in cells where the miRNA is not highly expressed, we took the opposite approach and inhibited an endogenous miRNA to relieve mRNA targets from the increased degradation observed for some mRNAs upon miRNA binding (7,9,47). A similar approach was previously used by Stoffel and coworkers (48) to identify targets of miR-122 in the mouse liver. Although microarray analysis following miRNA inhibition or overexpression is a relatively simple and robust method for target identification, this approach can, per definition, not identify mRNAs subjected exclusively to translational repression (49,50). Therefore, the development of new genome-wide methodologies is urgently needed to unravel the true importance of miRNA regulation.
Given the indications that miR-21 acts as an oncogene in a variety of tumor types, and the limited knowledge of miR-21 targets, our aim was to identify functionally relevant miR-21 targets in MCF-7 breast cancer cells. Bioinformatics analyses demonstrate a very significant over-representation of miR-21 complementary motifs among the transcripts up-regulated by miR-21 inhibition, demonstrating the validity of the experimental approach. We subsequently verified a direct responsiveness to miR-21 for a subset of the putative target mRNAs in heterologous reporter assays. Interestingly, among the six 3Ј-UTR sequences tested in luciferase assays, only the PDCD4 sequence responded to miR-21 inhibition.
The tumor suppressor PDCD4 was originally characterized as an inhibitor of cellular transformation in a mouse cell culture model (51). PDCD4 expression is down-regulated or lost in several tumor types (52,53), and ectopic expression of Pdcd4 reduces tumor formation in a mouse skin cancer model (54). Consequently, PDCD4 has been indicated by several as a promising molecular target in cancer treatment (55)(56)(57). At the molecular level, PDCD4 binds and inhibits the translation initiation factor eukaryotic initiation factor 4a, thereby impacting on protein translation (58,59). In addition, PDCD4 has been found to inhibit AP-1-mediated trans-activation (51) and to induce expression of the cyclin-dependent kinase inhibitor p21  (53). As a result, loss of PDCD4 confers growth advantages to the cells by several means and thereby facilitates the development of cancer. We demonstrate here that PDCD4 is directly regulated by the "oncomiR" miR-21. This is evident at the level of PDCD4 mRNA as well as protein where endogenous PDCD4 protein level is ϳ3.5-fold up-regulated by miR-21 inhibition. Importantly, depletion of PDCD4 by siRNA transfection partly rescues the reduced cellular proliferation observed upon miR-21 inhibition in MCF-7 cells, demonstrating that PDCD4 is an important functional target of miR-21 in this model.

miR-21 Targets PDCD4 in Human Breast Cancer Cells
A recent report demonstrated that the stability of PDCD4 is controlled by the mTOR pathway since PDCD4 during mitogen stimulation is phosphorylated by the S6K1 kinase, which marks it for degradation by the proteasome (60). That miR-21 via repression of PDCD4 affects the PI3K/AKT/mTOR pathway downstream of mTOR in breast cancer cells is interesting, given the evidence that miR-21 in hepatocellular carcinoma cells regulate the PI3K antagonist PTEN (11). Hence, in different cell types miR-21 may target different negative regulators of the PI3K/AKT/mTOR survival pathway. In addition, knock down of the tumor suppressor protein p53 partly abrogated the proliferation decrease observed in MCF-7 cells following inhibition of miR-21. This suggests a functional link between miR-21, the miRNA most frequently found overexpressed in cancer (27), and the tumor suppressor pathway most often found mutated or otherwise obstructed in cancer (39). There is accumulating evidence of extensive cross-talk between the p53 and the PI3K/AKT/mTOR pathways (61), and our data may reflect such cross-coordination between important anti-cancer networks.