Non-Specific Blocking of miR-17-5p Guide Strand in Triple Negative Breast Cancer Cells by Amplifying Passenger Strand Activity

Conventional wisdom holds that only one of the two strands in a micro ribonucleic acid (miRNA) precursor duplex is selected as the active miRNA guide strand. The complementary miRNA passenger strand, however, is thought to be inactive. High levels of the oncogenic miRNA (oncomiR) guide strand called miR-17-5p is overexpressed in triple negative breast cancer (TNBC) and can inhibit ribosomal translation of tumor suppressor gene mRNAs, such as programmed cell death 4 (PDCD4) or phosphatase and tensin homolog (PTEN). We hypothesized that knocking down the oncogenic microRNA (oncomiR) miR-17-5p might restore the expression levels of PDCD4 and PTEN tumor suppressor proteins, illustrating a route to oligonucleotide therapy of TNBC. Contrary to conventional wisdom, antisense knockdown of oncomiR miR-17-5p guide strand reduced PDCD4 and PTEN proteins by 1.8±0.3 fold in human TNBC cells instead of raising them. Bioinformatics analysis and folding energy calculations revealed that mRNA targets of miR-17-5p guide strand, such as PDCD4 and PTEN, could also be regulated by miR-17-3p passenger strand. Due to high sequence homology between the antisense molecules and miR-17-3p passenger strand, as well as the excess binding sites for the passenger strand on the 3’UTR of PDCD4 and PTEN mRNAs, introducing a miR-17-3p DNA-LNA mimic to knock down miR-17-5p reduced PDCD4 and PTEN protein expression instead of raising them. Our results imply that therapeutic antisense sequences against miRNAs should be designed to target the miRNA strand with the greatest number of putative binding sites in the target mRNAs, while minimizing affinity for the minor strand.


Introduction
New cases of aggressive breast cancer are predicted to occur in 232,340 U.S. women in 2014, and to kill 39,620 [1]. Triple-negative breast cancers (TNBC) lack estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (ErbB2, Her2), and represent 16% of cases (2). Without specific molecular targets, conventional chemotherapy of TNBC yields modest clinical outcomes [2]. Thus, the 16% of breast cancer patients who suffer from TNBC have worse prognoses than other subtypes of breast cancer [3] [2].
TNBC clearly needs new molecular therapies that specifically target genes promoting cancer cell survival. A variety of oncogenic micro ribonucleic acids (oncomiRs) are overexpressed in TNBC, and are being studied intensively as targets for complementary oligonucleotide therapy [4]. OncomiRs are non-protein-coding RNAs of 18-25 nucleotides (nt) that form base pairs with specific sequences in mRNAs. They inhibit translation of mRNAs sterically or by inducing mRNA degradation by Ago2 [5,6]. Biogenesis of all miRNAs initiates in the nucleus, where primary miR-NAs are transcribed by either RNA polymerase II or RNA polymerase III. Primary miRNA transcripts are then processed by Drosha and its cofactor DGCR8 to produce shorter precursor miRNA hairpins of~70 nt [7]. Pre-miRNA hairpins are exported to the cytoplasm by exportin 5, then cleaved by Dicer to yield double-stranded miRNAs. The guide strand of the double-stranded miRNA is thought to exhibit weak hydrogen bonding at its 5' end, favoring its binding to Ago2 in an RNA-induced silencing complex (RISC), allowing the guide strand to be active against complementary mRNAs. The passenger strand is thought to be inactive, dissociated, and degraded [8].
Specific oncomiR target recognition is predominantly defined by Watson-Crick base pairing that occurs between the seed region (nucleotide 2 to 8 from the 5' end of the oncomiR guide strand) and the 3'-untranslated regions (3'UTR) of target mRNAs. Translational repression by oncomiRs can be achieved by perfect complementarity between oncomiRs and the 3'UTR of mRNAs mediated by RISC, leading to mRNA degradation. Alternately, the translation of mRNA is sterically inhibited through imperfect oncomiR-mRNA recognition [9]. On average, each oncomiR has hundreds of possible mRNA targets [10]. As a result, complementary oligonucleotide therapy against one oncomiR could impact a broad panel of genes.
The miR-17~92 cluster is one of the most studied of the oncomiR groups that play important roles in cancer development. miRNAs from this cluster are generally up-regulated in various cancers, including breast, lung, colon, pancreas, prostate, and gastric cancer [11,12]. Caloric restriction (CR) and ionizing radiation (IR) down-regulate members of the miR-17~92 cluster in TNBC models, decreasing their metastatic activities by suppressing extracellular matrix (ECM) mRNAs that exhibit miR-17-5p binding sites [13]. Among the seven members of the miR-17~92 cluster, the guide strand miR-17-5p is predominantly responsible for promoting migration and invasion of metastatic cancer cells, targeting the mRNAs of tumor suppressor genes, such as PDCD4 (programmed cell death 4) and PTEN (phosphatase and tensin homolog) [14]. Thus, miR-17-5p is considered to be an oncogenic miRNA, or oncomiR. The tumor suppressor proteins PDCD4 and PTEN are usually depressed in TNBC, associated with elevated oncomiR levels [15][16][17][18][19].
The oncomiR miR-17-5p is significantly up-regulated in mesenchymal MDA-MB-231 TNBC cells compared to the noninvasive luminal MCF7 cells, and contributes to the invasiveness and migratory behavior of TNBC [20]. Most TNBCs are basal-like [21,22], and transcription profiling has suggested that most basal-like TNBC cells have molecular properties of mesenchymal tumors [2]. Using MDA-MB-231 cells as a mesenchymal TNBC cell model, we hypothesized that knocking down miR-17-5p might restore the expression levels of PDCD4 and PTEN tumor suppressor proteins, illustrating a route to oligonucleotide therapy of TNBC.

OncomiR Target Prediction
Three different oncomiR target prediction algorithms, rna22, TargetScan, and Miranda [23][24][25][26] were used to search for oncomiRs that might target PDCD4 and PTEN mRNAs. These 3 databases were chosen because of the diverse algorithms they employ to find targets. rna22 uses non-canonical seed pairing, which allows mismatches in the miRNA seed:mRNA interaction. It utilizes pattern discovery to find matching sequence patterns from a miRNA:mRNA set, then rank predicted miRNA:mRNA pairs based on free energy calculations [27]. rna22 allows G-U wobble base pairs and mismatches in the seed region. Miranda is also less stringent about seed matching by allowing G-U wobble base pairs. Although it considers matching sequences along the entire miRNA:mRNA pair, the final prediction favors matching in the seed region. TargetScan relies heavily on conservation of 3'UTR interactions, followed by fully complementary seed regions that are conserved. It does not allow mismatches or G-U wobble base pairs, although it does consider 3' compensatory regions.

Cell Line and Cell Culture
The human mesenchymal TNBC cell line MDA-MB-231 was obtained from American Type Culture Collection (ATCC). MDA-MB-231 cells were characterized as basal B subtype TNBC, with mesenchymal features that allow them to migrate readily and degrades their ability to adhere and polarize [28]. MDA-MB-231 cells were maintained in L-15 medium (ATCC) containing 10% fetal bovine serum (FBS) and 100 U/mL of penicillin/streptomycin (Invitrogen) in a humidified incubator at 37°C without added CO 2 .
To generate luciferase constructs containing individual predicted 3'UTR binding sites, we purchased the DNA sequence of each binding site, along with its flanking sequence region, as sense and antisense primers (Fisher Scientific). The primers were then annealed and cloned into the pMir-Report luciferase vector between the SpeI and HindIII restriction sites. The sequence of each primer used to generate 3'UTR fragments for cloning is shown in S1 Fig. Luciferase assay 8×10 4 MDA-MB-231 cells were seeded in 24-well plates in complete medium without antibiotics the day before transfection. Cells were transfected with 400 ng of each of the reporter constructs, 100 ng of pRL-TK Renilla luciferase internal control vector (E2241, Promega), and with either miR-17-5p mimic or miR-17-3p mimic using Lipofectamine 2000 (Invitrogen, CA). In parallel, transfections without the mimics were performed. After 24 hours, cells were lysed and 50 μL of each lysate was transferred to 96-well plates. Dual-Glo reporter assays were performed according to the manufacturer's protocol (E2920, Promega). Luminescence intensity for each sample was measured with a Veritas Microplate Luminometer (Turner BioSystems), and each value from pMir-Report firefly luciferase was normalized by pRK-TK Renilla luciferase.

Statistical Analysis
All experimental measurements were performed independently at least three times. Significance was assessed by Student's t-test.

Results
PDCD4 mRNA is a potential target for miR-17-5p Using rna22, TargetScan and miRanda, we searched for oncomiR targets in PDCD4 and PTEN mRNAs. rna22 identified miR-17-5p as a potential gene regulator through its interaction with one binding site in the 3'UTR of PDCD4 mRNA (Fig 1). Although rna22 is the only algorithm that predicted a binding site for miR-17-5p in the 3'UTR of PDCD4 mRNA, the predicted 23 bp oncomiR:mRNA duplex is stable, containing 17 complementary basepairs and an Mfold predicted folding energy ΔG°of -24.5 kcal/mol at 37°C (Fig 1).

miR-17-5p Passenger Strand Activity
To determine whether the complementary DNA-LNA chimeras were working as predicted, we examined another direct target of miR-17-5p, PTEN mRNA [14]. Both miRanda and TargetScan predicted one binding site for miR-17-5p in the 3'UTR of PTEN mRNA (Fig 1). Surprisingly, we found that knocking down miR-17-5p decreased PTEN protein level by 1.8±0.3 fold (Fig 3C). This result conflicted with the expectation that miR-17-5p target expression would increase when miR-17-5p was knocked down. These results implied that mature miR-17-5p is a gene regulator of PDCD4 and PTEN mRNA translation. However, anti-miR-17-5p resulted in down-regulation of both PDCD4 and PTEN mRNA translation, rather than stimulation.
miR-17-3p passenger strand is a potential inhibitor of PDCD4 and PTEN mRNAs, as well as miR-17-5p To understand the unexpected results in Fig 3B, we examined miR-17 in miRBase. In the pre-miRNA, miR-17-5p was predicted to hybridize with its passenger strand miR-17-3p to form a  hairpin ( Fig 3A). Most of the miR-17-3p is fully complementary to its guide strand miR-17-5p. Since anti-miR-17-5p is fully complementary to miR-17-5p, its sequence is highly homologous to miR-17-3p (Fig 3B). Therefore, we speculated that anti-miR-17-5p DNA-LNA chimera could act as a miR-17-3p mimic, binding to miR-17-3p target sites in the 3'UTR of PDCD4 and PTEN mRNAs. To test this hypothesis, rna22, TargetScan and miRanda were used to identify potential binding sites for miR-17-3p in the 3'UTR of PDCD4 and PTEN. Four potential binding sites for miR-17-3p were noted in the PDCD4 3'UTR (Fig 4), while the PTEN 3'UTR had six (Fig 5). For each predicted oncomiR:mRNA interaction, the CUGCA motif within the seed region of miR-17-3p is complementary to the corresponding 3'UTR binding sites. This same sequence motif exists in anti-miR-17-5p ( Fig 3A). Thus, considering seed pairing as one of the important factors in miRNA:target recognition, all of the miR-17-3p target sites in the 3'UTRs of PDCD4 and PTEN could be putative binding sites for anti-miR-17-5p.

OncomiR:mRNA duplex structures occupy A-form helices
Our molecular dynamics calculations predicted stable A-form duplexes for all passenger strand: mRNA targets, as well as for guide strands, despite the mismatches and bulges that appear so distorted in the Mfold presentation. The example of miR-17-3p bound to a PTEN 3'UTR site (Fig 9) illustrates the realization that miRNA:mRNA: duplexes could be accommodated in the substrate groove of Ago2, in agreement with an earlier simulation of an 11mer duplex bound to Thermus thermophilus Ago [37]. An animated mpg file showing 25 nsec of simulation at 300°K in explicit water with 100 mM NaCl, at pH 7.0, can be viewed in S1 Movie. These results indicated that miRNA inhibitors composed of LNA-DNA could also be bound to Ago when they form duplexes with target miRNAs, since LNA:RNA will pre-organize the helices in A-form.

Inhibition of the miR-17-3p passenger strand maintained PDCD4 and PTEN protein levels
To determine if the passenger strand was involved in the contradictory results above, we knocked down endogenous miR-17-3p with anti-miR-17-3p and analyzed PDCD4 and PTEN protein expression levels. In contrast to miR-17-5p knockdown (Fig 3C), miR-17-3p knockdown showed no significant changes in PDCD4 or PTEN protein levels (Fig 3D). The maintained protein levels of PDCD4 and PTEN could be a comprehensive outcome of both miR-17-5p and miR-17-3p binding to the PDCD4 and PTEN 3'UTRs. The static result is plausible, because there are more potential binding sites for miR-17-3p on the 3'UTR of PDCD4 and PTEN mRNAs compared to that of miR-17-5p (Fig 4 and Fig 5), although anti-miR-17-3p could act as a miR-17-5p mimic (Fig 3A and Fig 8). In addition, qPCR results showed that the endogenous level of miR-17-5p was about ninety times higher than miR-17-3p in MDA-MB-231 cells (S2A Fig). Thus, knocking down the low level of miR-17-3p passenger strand might not have a noticeable effect.
Both miR-17-5p and miR-17-3p directly affect the translation of PDCD4 and PTEN mRNAs To test the hypothesis that the translation of PDCD4 and PTEN mRNAs are directly inhibited by miR-17-5p and miR-17-3p, we transfected MDA-MB-231 cells with either miR-17-5p mimic or miR-17-3p mimic, and measured the protein levels of PDCD4 and PTEN 48 hr posttransfection. Exogenous miR-17-5p mimic lowered both PDCD4 and PTEN protein levels miR-17-5p Passenger Strand Activity ( Fig 12A). Transfection with miR-17-3p mimic lowered PTEN protein level, but not PDCD4 protein level (Fig 12A). This could be explained by the low endogenous expression level of miR-17-3p passenger strand compared to miR-17-5p guide strand (S2A Fig). Indeed, qPCR showed that transfection with miR-17-5p guide strand mimic increased the measured level of miR Both miR-17-5p and miR-17-3p directly interact with the 3'UTR of PDCD4 and PTEN mRNAs To study the mechanism of post-transcriptional regulation of PDCD4 and PTEN mRNAs by miR-17-5p and miR-17-3p, we cloned individual binding sites for miR-17-5p or miR-17-3p (Fig 1, Fig 3 and Fig 4) in the 3'UTR of PDCD4 or PTEN mRNAs into luciferase reporter vectors right after the luciferase gene. Since PDCD4 has not been reported to be a target for miR-17-5p or miR-17-3p, we also cloned the whole 3'UTR of PDCD4 mRNA into the luciferase reporter vector. The resulting luciferase constructs were expected to express luciferase mRNAs  containing the miRNA binding sites as their 3'UTR. MDA-MB-231 cells were co-transfected with each luciferase construct, miR-17-5p or miR-17-3p mimic, and a pRL-TK Renilla luciferase vector served as a transfection efficiency control. In parallel, the same transfection was also done without the addition of mimic for each luciferase construct to obtain the baseline luciferase signal. The luciferase light units from each construct were first normalized to that of Renilla luciferase vector, then further normalized to the corresponding baseline signal without the mimic. The vector containing the luciferase gene, and the restriction sites in the luciferase 3'UTR, was used as a negative control.
Compared to vector control, miR-17-5p mimic lowered the luciferase activity of both predicted PDCD4 binding sites and the entire PDCD4 3'UTR (Fig 12B), suggesting that miR-17-5p can directly target PDCD4 mRNA. Similarly, miR-17-5p mimic lowered the luciferase activity of the vectors containing the binding sites from the PTEN 3'UTR ( Fig 12C). Exogenous miR-17-3p mimic lowered luciferase activities from vector constructs harboring each of the four predicted binding sites, as well as the construct containing the whole 3'UTR of PDCD4 mRNA (Fig 12D), indicating that both miR-17-5p guide strand and miR-17-3p passenger strand could directly target PDCD4 mRNA. miR-17-3p mimic lowered luciferase activities of four out of six predicted binding sites in the 3'UTR of PTEN mRNA (Fig 12E).
Anti-miR-17-5p DNA-LNA inhibitors act as miR-17-3p mimics, and reduced the translation of PDCD4 and PTEN mRNAs To investigate whether anti-miR-17-5p DNA-LNA could mimic miR-17-3p by binding to the predicted sites for miR-17-3p in the 3'UTR of PDCD4 and PTEN, we carried out the luciferase experiments by co-transfecting MDA-MB-231 cells with luciferase constructs and anti-miR-17-5p as described above. Compared to vector control, anti-miR-17-5p lowered the activity of luciferase vectors containing two out of four PDCD4 3'UTR binding sites predicted for miR-17-3p (Fig 13A). However, it did not affect luciferase activity from the vector containing the entire 3'UTR of PDCD4 mRNA (Fig 13A). Since the whole PDCD4 3'UTR is much longer than the individual binding sites, the secondary and tertiary structure of the entire 3'UTR might be changed after being cloned after the luciferase gene. Anti-miR-17-5p also lowered luciferase activities of constructs harboring four out of six PTEN 3'UTR binding sites predicted for miR-17-3p (Fig 13B). These results implied that artificial anti-miR-17-5p DNA-LNA against miR-

Discussion
In our study, we discovered that in MDA-MB-231 TNBC cells, anti-miR-17-5p DNA-LNA chimera knocked down endogenous miR-17-5p, but surprisingly decreased the protein levels coded by their potential targets PDCD4 and PTEN mRNAs, rather than elevating them. In contrast, anti-miR-17-3p knocked down endogenous miR-17-3p, but maintained PDCD4 and miR-17-5p Passenger Strand Activity PTEN protein expression (Fig 3C). We ascribe this result to limited binding sites for the miR-17-5p guide strand relative to the miR-17-3p passenger strand.
Through mRNA sequence analysis, we found four putative binding sites for miR-17-3p relative to one for miR-17-5p in the 3'UTR of PDCD4 mRNA. Thus, while the anti-miR-17-5p could silence miR-17-5p and theoretically alleviate the translational inhibition of PDCD4 mRNA, as a miR-17-3p mimic it apparently repressed the translation of PDCD4 mRNA by binding to multiple 3'UTR target sites for miR-17-3p (Fig 13 and Fig 14). Although the endogenous passenger strand miR-17-3p had only modest effects in modulating PDCD4 and PTEN post-transcriptionally (Fig 3D and Fig 12A), the passenger strand mimicking anti-miR-17-5p could target PDCD4 mRNA more effectively than the miR-17-5p guide strand. This phenomenon could be explained by the excess amount of highly stable anti-miR inhibitors introduced into the cells, which could be associated with Ago, and form strong interactions with the Values are the average of three blots ± s.d. after normalization to β-actin and to control/treatment group. Each blot was subjected to gamma setting adjustments. B-E: Luciferase activity after cotransfecting MDA-MB-231 cells with indicated luciferase reporter constructs in the presence or absence of miRNA mimic. All luciferase signals from pMirreport firefly luciferase vectors are normalized to signals from pRL-TK Renilla luciferase vector. The ratio of normalized signal in the presence of mimic to signal in the absence of mimic for each construct is then calculated. The pMir-report luciferase vector is used as negative control. Results represent fold changes of the above ratio relative to vector control. Values are the average of at least three measurements ± s.e.m * indicates p<0.05, ** indicates p<0.01.
doi:10.1371/journal.pone.0142574.g012 miR-17-5p Passenger Strand Activity targeting miRNAs. Moreover, the anti-miR inhibitors could be selected by Ago to perform functions of a miRNA due to their pre-organized A-form structure, and could target mRNAs with great side effects through binding to multiple 3'UTR sites.  On the other hand, anti-miR-21-5p did not constitute a passenger strand mimic that competes with the guide strand for binding to target mRNAs. Thus, knocking down miR-21-5p consistently increased PDCD4 protein levels ( Fig 3E) and PDCD4 mRNA levels ( Fig 11A) at 12 hr and 48 hr after transfection by relieving translational inhibition of PDCD4 mRNA by miR-21-5p (Fig 14).
These results suggest that a miRNA inhibitor designed to bind to the mature guide strand could potentially constitute a mimic of the passenger strand. As a result, an oncomiR guide strand inhibitor could fail if passenger strand binding sites exist in the 3'UTR of target mRNAs. This effect was not apparent in an earlier study in lymphocytes that utilized a luciferase vector containing only a fragment of the PTEN 3'UTR bearing the single miR-17-5p target [14].
Recent literature has reported some examples of cancer cells with substantial activity exhibited by passenger strand species, including mRNA regulatory activities [39][40][41][42][43][44]. For example, in hepatocellular carcinoma, miR-17-5p reduced the translation of PTEN mRNA, while miR-17-3p directly targeted vimentin mRNA [45]. As a result, miR-17-5p and miR-17-3p cooperatively contribute to the development of hepatocellular carcinoma. The same investigators reported that both the guide strand and the passenger strand of miR-17 targeted sites in TIMP3 mRNA in prostate cancer cells, thus promoting proliferation and invasion [46].
Those observations in hepatocellular carcinoma and prostate cancer cells, and our own in TNBC cells, contradict the conventional model in which the passenger strands of miRNAs are dissociated and degraded.
Thus, when studying an oncomiR guide strand, one must consider that its passenger strand could also be functional. Furthermore, the expression patterns of passenger strand species vs. guide strand species should also be characterized at various cell cycle stages, since the two oncomiR strands are not always expressed equally as cell growth progresses [47,48]. Finally, when combining the effect of the two oncomiR strands, different scenarios of consequences must be investigated since the guide strand and the passenger strand miRNAs could theoretically work in synergy, in opposing fashion, or independently. Clearly, oncomiRs regulate genes via a fine-tuning process.
Based on the findings that both oncomiR strands could be functional, precaution must be taken when designing oncomiR knockdown sequences as potential therapeutic or diagnostic agents, since both the guide strand and the passenger strand of an oncomiR might target the same transcript. For research and therapeutic purposes, developing oncomiR inhibitors or mimics that modulate the functionality of either the guide strand or the passenger strand specifically and independently is valuable. Inhibitors specific to one strand or the other will diminish contradictory side effects associated with mimicking the passenger strand.
For future investigations, the activity of passenger strand species from other oncomiRs should also be explored in multiple TNBC cell models. In this study, only two genes were found to be regulated by miR-17 passenger strand. As each oncomiR could target hundreds of gene transcripts, one could expect a comparable number of mRNA targets for oncomiR passenger strand relative to their guide strand.
To characterize the functionality of oncomiR passenger strand species, a global RNA expression analysis will be essential in combination with oncomiR target prediction algorithms and molecular dynamics calculations of complex stability. Gene expression profiling in concordance with oncomiR guide and passenger strand expression will provide us with more powerful insight into the purpose of active passenger strand species.