MiR-183/-96/-182 cluster is up-regulated in most breast cancers and increases cell proliferation and migration

The miR-183/-96/-182 cluster is a conserved polycistronic microRNA (miRNA) cluster which is highly expressed in most breast cancers. Although there are some sporadic reports which demonstrate the importance of each miRNA in this cluster in breast cancer, the biological roles of this cluster as a whole and its regulation mechanisms in breast cancer are still unclear. We compared the expression of this cluster in different cancer types, analyzed the regulation mechanism of this cluster, identified new target genes, and examined the impact of this cluster on breast cancer cells. The miRNA level was detected by LNA-based northern blot and Real-time PCR, and was also analyzed from TCGA dataset. Bioinformatics research and luciferase assay were applied to find the promoter regions and transcription factors. To investigate the biological effects of the miR-183/-96 /-182 cluster in breast cancer, we generated miR-96, miR-182 and miR-183 overexpression stable cell lines to check the overdose effects; we also used miR-Down™ antagomir for each miRNA as well as miR-183/-96 /-182 cluster sponge lentivirus to check the knockdown effects. Growth, migration, cell cycle profile and survival of these cells was then monitored by colony formation assay, MTT assay, cell wound healing assay, flow cytometry and microscopy. The target gene was validated by Real-time PCR, luciferase assay, Western blot and Phalloidin/DAPI counterstaining. The miR-183/-96/-182 cluster was highly expressed in most breast cancers, and its transcription is disordered in breast cancer. The miR-183/-96/-182 cluster was transcribed in the same pri-miRNA and its transcription was regulated by ZEB1 and HSF2. It increased breast cell growth by promoting more rapid completion of mitosis, promoted cell migration and was essential for cell survival. MiR-183 targeted the RAB21 mRNA directly in breast cancer. The miR-183/-96/-182 cluster is up-regulated in most breast cancer. It functions as an oncogene in breast cancer as it increases cell proliferation and migration.


Introduction
Breast cancer is a family of diseases that involve unregulated breast epithelial cell growth and division, which is caused by many different carcinogenic factors. The exact cause of breast cancer is unclear. Many risk factors may increase the chance of having breast cancer, such as endocrine disorders, genetic mutations and declines in immune function. However, unregulated mammary epithelial cell proliferation and apoptosis, which are caused by an accumulation of gene mutations and by dysregulated gene expression, is the essential reason for breast cancer. As numerous genes are predicted to be regulated by microRNA (miRNA), mammary tumorigenesis and metastasis is likely to be regulated by several tissuespecific miRNAs.
The miR-183/-96/-182 cluster is a highly conserved polycistronic miRNA cluster which was first identified by Dr Xu in sensory organs [1]. Members of this cluster are located within a 5-kb region on human chromosome 7q32.2 and are transcribed in the same direction from telomere to centromere. Previous studies showed that the miR-183/-96/-182 cluster is abnormally expressed in a variety of tumors and is directly involved in human cancers. But the role of this miRNA cluster in tumors is still unclear. It may function as an oncogene or tumor suppressor gene, depending on the type, location and stage of the cancer. We summarize its reported functions in cancers and its target genes in Table 1.
The miR-183/-96/-182 cluster has not yet been extensively studied in breast cancer. Forkhead box O (FOXO) proteins, which are a family of tumor suppressor transcription factors involved in cell growth, proliferation, differentiation, and longevity, are the main targets for this cluster in breast cancer. Both FOXO1 and FOXO3a are regulated by miR-96 and miR-182 [5,6]. It seems that this miRNA cluster functions as onco-microRNA in breast cancer. However, in 2010, Lowery et al. reported that miR-183 inhibits cell migration in breast cancer by repressing Ezrin, which plays a key role in cell-surface structure adhesion, migration, and organization [12]. These conflicting results may be ascribed to two reasons. One possibility is that these three miRNAs are transcribed or processed in different way and they function separately and differently; the other possibility is that this cluster plays different roles in different breast cancer types. In fact, the level of miR-183 was lower in estrogen receptor (ER)-positive breast tumors compared to ER-negative tumors, and higher in human epidermal growth factor receptor-2 (HER2)/neu-receptor-positive tumors compared to HER2/neu-receptor-negative tumors [12], suggesting the roles of miR-183 in different breast cancer cells are different.
Recently, attention has focused on the target genes of these miRNAs; however, little is known about the regulation mechanism of the miRNA cluster itself. Most miRNA genes are transcribed by RNA polymerase II [15], which means miRNA biogenesis is controlled elaborately through various regulatory pathways just as protein-coding mRNAs. Chromatin structure analysis, genomic and RNA sequence analysis and RNA polymerase II chromatin immuneprecipitation assays have been applied to predict the transcription start site (TSS) and promoter region of miRNAs [16][17][18][19], but few results have been confirmed by experiments. The Ozsolak [16], Wang [18], and Chien [19] laboratories predicted that the TSS of miR-183/-96/-182 was 5068 bp, 5200 bp and 5207 bp upstream of the miR-183 precursor, respectively. However, the promoter region of miR-183/-96/-182 and the transcription regulators remain unknown.
Here, we investigated the function of the miR-183/-96/-182 cluster in breast cancer. We found that the miR-183/-96/-182 cluster was highly expressed in most breast cancers. These three miRNAs were transcribed in the same pri-miRNA and this miRNA cluster was regulated by HSF2 and ZEB1. We also demonstrated that the miR-183/-96/-182 cluster functioned as an onco-miRNA in breast cancer. Overexpression of the miR-183/-96/-182 cluster increased the cell proliferation rate and promoted cell migration while inhibition of the miR-183/-96/-182 cluster decreased cell growth rate, and even induced cell death. MiR-183 targeted RAB21 directly in breast cancer and accumulated nucleus number aberration cells. Our results suggested that the miR-183/-96/-182 cluster plays an important role in tumorigenesis and in the migration of breast cancer cells.

Virion and cell line constructions
To establish the miRNA overexpression cell lines, partial mir-96, mir-182 and mir-183 pri-microRNA sequences flanked by EcoRI and AgeI restriction sites were inserted into the CMV promoter of lentivirus infectious virions pLKD-CMV-G&PR-U6-shRNA (Hpcoo3) (Additional file 1: Figure S1A). MCF-7 or T47D cells were infected with these viruses and selected under the pressure of 1 μg/ml puromycin (Invitrogen, San Diego, CA, USA). The green fluorescent protein (GFP) signal of the infected cells was detected under microscope (Additional file 1: Figure S1B), and the expression of the miR-183/-96/-182 cluster in each cell line was measured by reverse transcription (RT)-PCR (Additional file 1: Figure S1C).
To disrupt the activity of the miR-183/-96/-182 cluster, we generated miR-183/-96/-182 cluster sponge lentivirus virion. Basically, 10 copies each of complementary sequences to miR-183, miR-96 and miR-182, each with mismatches at positions 9 to 12 for improved stability [20,21], were introduced into the pLOV-CMV-eGFP-EF1a-PuroR lentivirus infective virion (Additional file 2: Figure S2). A moderate multiplicity of infection (MOI) of 1 was used for transduction. The infection efficiency and cell morphology were monitored under microscope every day. After 3 days of transduction, cells were collected for cell cycle analysis and RNAs were collected for realtime PCR.
To research the function of transcription factors, the coding sequences of HSF2 and ZEB1 flanked by XhoI and KpnI restriction sites were inserted into vector GV219. The plasmids were transfected into MCF-7 cells and the cells were selected with a culture medium containing 600 μg/ml G418-Geneticin (GenView, Galveston, TX, USA) for 2 months.

LNA-based Northern Blotting
Total RNAs were extracted from cancer samples with the mirVanaTM miR isolation kit and 10 μg of total RNA was used for each assay. All procedures followed manufacturer's instructions for the miRCURY LNA™ microRNA detection probes (Exiqon, Woburn, MA, USA). After fractionation by electrophoresis on a denaturing 12% polyacrylamide gel containing 8 M urea, RNAs were transferred to Nytran N membrane (Amersham Biosciences, Piscataway, NJ, USA) and fixed by UV crosslinking. Blots were prehybridized for 1 h at 45°C in PerfectHyb™ Plus Hybridization Buffer (Sigma, St Louis, MO, USA) and hybridized overnight at 45°C in hybridization buffer containing 0.1 nM probe, then washed twice for 30 minutes at 65°C in 0.1SSC/0.1% SDS. As the probes were 5′-DIG labeled, we detected the signal by PhototopeR-Star Kit (New England BioLabs Inc, Ipswich, MA, USA), and the densities were quantified by the Image J program. Because the miR-183, miR-96 and miR-182 sequences are similar, we tested the probe specificities before doing the experiments (Additional file 3: Figure S3). Mimic oligonucleotides were designed based on miRNA sequences registered in the miRBase Sequence Database (see Additional file 4: Table S1).

RT-PCR and real-time PCR
For mRNA RT-PCR and real-time PCR, total RNAs were extracted from cancer samples or cultured cells with Trigol (Dingguo, Co.) reagent. Primer sets were designed within the exon junction areas listed in Additional file 4: Table S2. For miRNA real-time PCR, miRNAs were extracted from cells using a mirVana miRNA isolation kit (Ambion, Austin, TX, USA). All primers, including the YRBIO™ miRNA qPCR Detection primer sets and U6 snRNA PCR primer set were purchased from Yingrun Biotechnology (Changsha, China).
In brief, mRNA and miRNA were reverse-transcribed with an M-MLV First Strand kit (Invitrogen). Then 50 ng cDNA was mixed with All-in-one™ qPCR Mix (Genecopoeia, Rockville, MD, USA) and the target gene primer set (final concentration: 1 μM for each primer) to produce a 20-μl reaction mixture. All real-time PCR experiments were carried out with an ABI Step One Plus Real-time PCR System (Applied Biosystems, Carlsbad, CA, USA). All real-time PCR reactions were done in triplicates, and the average ΔCT (Δ cycle threshold) for the triplicates was used in subsequent analysis.

Luciferase reporter assays
For promoter analysis, promoter region sequences or their mutants flanked by XhoI and KpnI restriction sites were inserted into the upstream region of luciferase reporter gene in pGL3-Basic vector (Promega). MCF-7 cells were transfected with 200 ng reporter construct and 1 μg GV219 vector with or without transcription factor sequence. Also, 40 ng of pRL-CMV-Renilla plasmid was transfected as an internal control.
For target analysis, 33 bp of RAB21 3′-UTRs including the seed sequence were flanked by XbaI and FseI restriction sites and inserted between the Luciferase coding sequence and SV40 polyadenylation element in pGL3-Promoter vector (Promega). HEK-293 cells were transfected with 200 ng reporter construct and 1 μg Hpcoo3 vector with or without partial pri-microRNA sequence of miR-183/-96/-182 cluster. Also, 40 ng of pRL-CMV-Renilla plasmid was transfected as an internal control.
The luciferase reporter assays (Promega) were performed 48h after transfection, and luciferase activity was determined with a GloMax 20/20 Luminometer (Promega). Relative luciferase activities were calculated as ratios of firefly to renilla luciferase activities.

Cell wound-healing assays
Cells were seeded on 6-well plates (5 × 10 5 cells per well) and incubated for 24 h. Adherent cell monolayers were scratched with a 10-μl pipette tip and cultured in 2 ml DMEM high-glucose medium without FBS or antibiotics. Cell migration was monitored under microscopy later.

Colony formation assays
The culture dish was covered by 2 ml bottom gel (0.5% basic agar in RPMI medium 1640 (Invitrogen) supplemented with 10% FBS and 1% penicillin/streptomycin) and 1.5 ml top gel (0.7% agar in RPMI-1640 medium supplemented with 10% FBS and 1% penicillin/streptomycin) mixed with 10,000 cells. Cells were incubated for 16 days and the colonies were stained with 0.5ml 0.005% crystal violet overnight followed by washing with PBS (Hyclone) three times. The pictures of cell colonies were taken by a digital camera.

Cell cycle analysis
Cells were digested with 0.05% trypsin (Thermo Scientific, Logan, UT, USA) for 2 minutes to dissociate them from the plates. After fixation in 70% pre-chilled (−20°C) ethanol in PBS at 4°C overnight, cells were treated with 10 μg/ml of RNase (Auragene, Co., Shenzhen, China) in PBS at 37°C for 2 h and stained with 50 μg/ml of propidium iodide (PI) (Sigma) for 5 minutes. Flow cytometry was conducted on a BD FACSCalibur flow cytometer (BD Biosciences, Franklin, IN, USA) and data were analyzed by ModFit LT software.

Statistical analysis
Data were expressed as means ± SD, and the statistical software SPSS 11.5 (IBM, Armonk, NY, USA) was used for analysis of variance (ANOVA) and analysis using Student's t-test. Statistical probability (P) in tables, figures, and figure legends are expressed as follows: *P <0.05, **P <0.01, *** P <0.001.

Results
The miR-183/-96/-182 cluster was highly expressed in most breast cancers Six different tumors and their normal adjacent tissues (NAT) were collected from the Hunan Tumor Hospital.
Breast cancer and liver cancer tumors were available from two patients, and other types of cancer were from one patient. The miRNAs were detected by LNA-based northern blot. We found that miR-96, miR-182 and miR-183 expression levels were dramatically higher in tumors compared to the normal adjacent tissues in breast, lung and liver cancers. MiR-96 was also expressed in thyroid and larynx cancers, but the expression differences between tumors and their normal adjacent tissues were not obvious. The expressions of these three miRNAs were undetectable in other carcinoma tissues ( Figure 1A). We then performed an analysis of miRNA expression data detected by either IlluminaGA_miRNASeq or IlluminaHi-seq_miRNASeq in breast invasive carcinoma from the TCGA dataset. From 102 matched pairs of samples (Additional file 5), we found the expression levels of miR-96, miR-182 and miR-183 in tumor samples were increased 8.4 (± 1.1)-fold, 4.2 ± (1.1)-fold and 7.5 ± (1.1)fold respectively compared to the matched normal samples ( Figure 1B, upper panel). Another interesting phenomenon was that the expression levels of miR-183 and miR-182 were highly correlated in normal samples (R 2 = 0.9127), but the correlation dropped dramatically in tumor samples (R 2 = 0.5475), which indicated that the transcription pattern was changed in breast cancer ( Figure 1B, lower panel).
Because breast cancer is a heterogeneous disease comprising different subtypes that vary significantly with regard to clinical features and molecular markers, we compared the miRNAs expression levels in different breast cancers based on their clinical features, surface markers and clinical stages. From 990 samples (Additional file 6), we found the expressions of miR-96 and miR-183 were lower in lobular carcinoma than in ductal carcinoma and other types of carcinoma, but the expression of miR-182 was not correlated with the clinical features ( Table 2). The levels of miR-96 and miR-183 were also lower in ER+ and PR+ cancers than in ER− and PR− cancers, but miR-182 was almost the same, even slightly higher in ER+ cancers. We did not find any correlation between the miR-183/-96/-182 cluster level and the HER2/neu receptor ( Table 3). The expression of miR-183/-96/-182 cluster was not correlated with clinical stages, as all the three miRNAs remained the same in all clinical stages (Table 4). To divide the breast cancer samples into different subtypes, the following surface markers were used: luminal A (ER+ and/or PR+, HER2−), luminal B (ER+ and/or PR+, HER2+), basal-like (ER−, PR−, HER2−), HER2-enriched (ER−, PR−, HER2+) [22]. We found miR-96 and miR-183 levels were higher in HER2-enriched breast cancers than other types. In basal-like breast cancers, miR-182 was lower but miR-183 was higher comparing to other types of breast cancer ( Table 5). All these data indicated that although miR-183/-96/-182 cluster was up-regulated in most breast cancers, its expression pattern was slightly different in different breast cancer subtypes.
To confirm our findings, we also compared the miR-NAs levels in different breast cancer cell lines based on their ER, PR and HER2/neu receptor status.  Figure 1C). Our data were similar to those reported by Riaz et al. [23], who also found that the highest expression of miR-96 was SK-BR-3 and the lowest expression of all these three miRNAs was MD-MBA-231 among these six breast cancer cell lines. We chose MCF-7 and T47D cells for further studies because their miR-183/-96/-182 clusters were highly expressed and they were easy to culture.
MiR-183/-96/-182 cluster was transcribed in the same pri-miRNA and was regulated by ZEB1 and HSF2 To study the regulation mechanism of the miR-183/-96/-182 cluster itself, we first analyzed the upstream sequence of the miR-183/-96/-182 cluster through the ENCODE project. We found a highly conserved region from 5054 bp to 9324 bp upstream of the human miR-183 precursor (Figure 2A, red box). The ENCODE project displayed the acetylation of histone H3 and the transcription factor chromatin immunoprecipitation (Chip) data to find the active regulatory elements. H3K27Ac histone marks were enriched in this region, which demonstrates that this region contains active regulatory elements. Transcription factor Chip data also showed that this region was easily pulled down with transcription factors. Altogether the information suggested that the promoter region and TSS of the miR-183/-96/-182 cluster is in this region.
Then, to check whether miR-183, miR-96 and miR-182 were transcribed in the same pri-miRNA or separately, we designed a series of primer pairs (Additional file 4: Table S3) to determine whether the corresponding regions of DNA were transcribed. Each primer pair spanned about 1600 bp and all the primer pairs divided the genomic DNA surrounding the miR-183/-96/-182 cluster (5352 bp upstream to 5893 downstream of human miR-183 precursor) into eight regions. From 5′-end to 3′-end, they were named Seq#1, Seq#2 … Seq#8 (their relative locations are showed in Figure 2B, upper panel). Total RNAs were extracted from MCF-7, T47D and MCF10A cell lines. MCF-7 genomic DNA was used as a positive control to check the efficiency of primer pairs. RT-PCR data showed that RNA were correctly transcribed from Seq#2 to Seq#7 (Seq#8 was a non-specific band because the size is incorrect) ( Figure 2B). This data indicated that miR-183, miR-96 and miR-182 were transcribed in the same pri-miRNA and the start site of this pri-microRNA was 5352 bp to 3991 bp upstream of the miR-183 precursor, and the transcript termination site was 289 bp to 1352 bp downstream of the miR-182 precursor. Several papers also predicted that the TSS of miR-183/-96/-182 was between 5068 bp and 5207 bp upstream of human miR-183 precursor [16,18,19]. We could not tell whether the transcription pattern was changed in cancer cells from this experiment because the PCR method is not linear.
Next we sought to determine how this pri-miRNA was regulated. To find the promoter region, we generated luciferase reporters with 1 kb, 2 kb, 3 kb and 4 kb DNA fragments within the conserved region (4263 bp to 8533 bp upstream of the mouse miR-183 precursor, corresponding to 5054 bp to 9324 bp upstream of the human miR-183 precursor. Figure 2A, red box), named upstream 1 kb, upstream 2 kb, upstream 3 kb and upstream 4 kb respectively. These luciferase assay results showed that the upstream 1 kb, upstream 2 kb and upstream 3 kb fragments increased luciferase activity approximately 30-fold compared with the empty vector. No significant difference was detected among upstream 1 kb, upstream 2 kb and upstream 3 kb. Upstream 4 kb increased luciferase activity 17-fold compared with the empty vector, which was much lower than the other three reporters ( Figure 2C). These data demonstrate that most active regulatory elements were located within 1 kb from the upstream of TSS region, and some repression elements were located between 3 kb and 4 kb from upstream of the TSS region.
To find the transcription factors regulating the miR-183/-96/-182 cluster, we used the online bioinformatics tools TFSEARCH to predict the transcription factor binding sites within 1 kb upstream from the TSS region of the miR-183/-96/-182 cluster. Four DNA sequences were predicted to be recognized by ZEB1, HSF2, ZEB1 and Sp1 respectively ( Figure 3A). We mutated the candidate transcription factor binding sites and performed the luciferase assay again. The luciferase activities of the HSF2 and the first ZEB1 mutant were significantly lower than upstream 1 kb (by about 50%), which suggested that these two sites were indeed transcription factor binding sites and that HSF2 and ZEB1 were two important transcription factors in cluster transcriptional regulation ( Figure 3B). Therefore, we cloned HSF2 and ZEB1 into the GV219 vector and co-transfected the transcription factors and the native or mutated upstream 1 kb luciferase reporters together into the MCF-7 cells. We found that HSF2 alone upregulated the luciferase activity of native upstream 1kb 1.9 (± 0.3)-fold, but had no effect on upstream 1 kb with a mutant HSF2 site. ZEB1 upregulated the luciferase activity of native upstream 1 kb 6.7 (± 0.7)fold, but had no effect on ZEB1 mutant upstream 1 kb reporter. There was no synergetic effect of these two genes, as co-transfection of the two genes only upregulated the luciferase activity of native upstream 1 kb 2.5 (± 0.2)-fold ( Figure 3C). The expression of each miRNA in the miR-183/-96/182 cluster in different breast cancer subtypes based on their surface markers: Patient number is indicated in the first column. Data are presented as mean ± SD. Statistical probability (P) was expressed as *P<0.05, ***P<0.001. ER, estrogen receptor; PR, progesterone receptor; HER, human epidermal growth factor receptor. To further confirm our results, we transfected the HSF2 and ZEB1 overexpression plasmids into MCF-7 cells, and then selected for stable cell lines with G418. Then we compared the expression levels of miR-96, miR-182 and miR-183 in stable overexpression cell lines with the control cell line, which was transfected with empty vector. Real-time PCR data showed that miR-96 and miR-183 were increased 2.7-to 3.8-fold compared to the control cell line, but miR-182 did not increase very much ( Figure 3D). We think the reason why miR-182 did not increase much is because miR-182 is far from the transcript regulation area. Although these three miRNAs are transcribed in the same pri-microRNA, the ending of this pri-microRNA is not always the same. Sometimes, it will end before miR-182 transcription. This result might explain why miR-182 only increased 4.2 (± 1.1)-fold in tumor samples, but miR-96 and miR-183 increased 8.4 (± 1.1)-and 7.5 (± 1.1)-fold in tumor samples. It could also explain why the expression levels of miR-183 and miR-182 correlated more strongly sin normal samples, but   the correlation dropped dramatically in tumor samples. Because the transcription of miR-183/-96/-182 was so fast in cancer, some pri-miRNA was not complete.

Up-regulation of the miR-183/-96/-182 cluster increased cell proliferation and migration and changed the cell cycle profile
To investigate the biological effects of miR-183/-96/-182 cluster up-regulation in the development and progression of breast cancer, we generated miR-96, miR-182 and miR-183 overexpression cell lines in both MCF-7 and T47D cells (Additional file 1: Figure S1). Using MTT assays, we observed that the growth rates of all overexpression cell lines were increased as compared with that of empty vector control or non-transfected cells in both MCF-7 and T47D cells ( Figure 4A). Furthermore, in colony formation assays, the increase of colony numbers in MCF-7 overexpression cell lines indicated that ectopically expression of the miR-183/-96/-182 cluster in MCF-7 cells significantly enhanced anchorage-independent growth ( Figure 4B). Furthermore, in both MCF-7 and T47D cells, in vitro wound-healing assays demonstrated that the migration abilities of miR-183, miR-96, and miR-182 overexpression cell lines were elevated, as the non-healed areas were smaller in overexpression cell lines than in control (empty vector) or nontransfected cells ( Figure 4C). To further explore the ability of the miR-183/-96/-182 cluster to promote cell proliferation, we analyzed the cell cycle profile of these overexpression cell lines. In both MCF-7 and T47D cells, flow cytometry results showed a small but significant decrease in the percentage of cells in the G2/M peak and a small but significant increase in the percentage of cells in the G1/G0 peak, the percentage of cells in the S phase was unaltered ( Figure 5). These data suggest that the miR-183/-96/-182 cluster increased the cell proliferation by promoting more rapid completion of mitosis.

Inhibition of miR-183/-96/-182 cluster miRNAs decreased cell proliferation, and even induced cell death
To explore the knockdown effects of miR-183/-96/-182 cluster miRNAs, we transfected the miR-Down™ antagomirs to the MCF-7 and T47D cells. First, we checked the knockdown efficiency and specificity of these antagomirs. Real-time PCR data showed that each antagomir knocked down its corresponding miRNA efficiently in both MCF-7 and T47D cells. MiR-182 antagomir also slightly decreased miR-96 expression, except that there were no crossreactions. The knockdown efficiency was higher in MCF-7 cells than in T47D cells, and miR-96 antagomir and miR-182 antagomir were more efficient than miR-183 antagomir ( Figure 6A). Then, we checked the cell growth rates, cell migrations and cell cycle profiles of these knockdown cells by MTT assay, cell wound-healing assay and cell cycle analysis. MTT assay data showed that knockdown of miR-96 and miR-182 decreased the cell growth rates significantly in both MCF-7 and T47D cells. The growth rate of miR-183 antagomir-treated cells also decreased slightly, but the decrease was not significant ( Figure 6B). In MCF-7 cells, the migration abilities of knockdown cells were all decreased although the decrease was not significant for miR-183 antagomir-treated cells. However, in T47D cells, only the miR-182 antagomir led to the decrease of migration; the migration ability of miR-96 and miR-183 antagomir-treated cells remained the same ( Figure 6C). Furthermore, cell cycle analysis demonstrated a significant increase in G2/M phase and a decrease in S phase for cells treated with miR-182 antagomir in MCF-7 cells. Knockdown of miR-96 also decreased the percentage of cells in S phase slightly in MCF-7 cells, but in T47D cells the cell cycle profiles were not changed except for a slight increase in G2/M phase after miR-182 antagomir treatment ( Figure 6D). We think the different behavior of MCF-7 and T47D cells after antagomir treatment was related to the knockdown efficiency. As the knockdown efficiency was higher in MCF-7 cells, the growth rate and To further examine the biological effect of the miR-183/-96/-182 cluster as a whole on breast cancer cells, we generated miR-183/-96/-182 cluster sponge lentivirus (Additional file 2: Figure S2), and infected T47D cells with this vector. First, we checked the inhibition efficiency of this virus by real-time PCR. Compared to the empty vector, the expressions of miR-183/-96/-182 cluster miRNAs were dropped to a half after sponge lentivirus transduction; and the expression of FOXO1, which was a generally acknowledged target gene of the miR-183/-96/-182 cluster, was increased about 2-fold after sponge lentivirus transduction ( Figure 6F upper panel). We found that T47D cells underwent cell death and apoptosis after transduction. Three days after transduction, the cells became round and detached ( Figure 6E). Cell cycle analysis showed an increase in the percentage of cells in the G2/M peak and pre-G1 peak and a decrease in the percentage of cells in the G1/G0 peak, indicating that inhibition of miR-183/-96/-182 induced G2/M arrest and apoptosis ( Figure 6F lower panel).
As MCF-7 and T47D cells are both luminal breast cancer, we also tested the miRNA knockdown effects in basal-like breast cancer cells, such as BT-20 (Basal A) and MDA-MB-231 (Basal B) cells. We found basal-like cells were more sensitive to the depletion of the miR-183/-96/-182 cluster than the luminal-like cells. MTT experiments showed BT-20 ceased proliferation and underwent cell death after knockdown of miR-96, miR-182 or miR-183 (Additional file 7: Figure S4A, B). MDA-MB-231 cells underwent cell death and apoptosis after transduction of miR-183/-96/-182 cluster sponge lentivirus. The cells became round and detached 3 days after transduction, and the cell cycle analysis showed that pre-G1 cells, which represented the apoptotic cells, were increased in knockdown cells (Additional file 7: Figure S4C, D).

MiR-183 targeted the RAB21 gene directly in breast cancer
To better understand the biological roles of the miR-183/-96/-182 cluster miRNAs in breast cancer, we compiled a list of putative target genes of the miR-183/-96/-182 cluster that were dysregulated in breast cancer by using three computational target prediction-algorithms: PicTar, TargetScan 5.1 and MicroCosm (Additional file 4: Table S2). As mammalian miRNAs regulate target genes predominantly by acting to decrease target mRNA levels [24], we first compared the mRNA levels of those genes between breast cancer sample and its NAT by real-time PCR with GAPDH used as an internal control. Eight genes out of twenty-five candidates showed significantly decreased expression in breast cancer ( Figure 7A). To validate the eight candidates, we checked their mRNA levels in the miR-96, miR-182 and miR-183 MCF-7 overexpression cell lines.
Compared to empty vector control cells, RAB21 was decreased in the miR-183 overexpression cell line; RAB40B was decreased in miR-96 and miR-183 overexpression cell lines and TNFSF11 was decreased in the miR-96 overexpression cell line ( Figure 7B). Because RAB21 was the predicted target gene of miR-183, we focused our efforts on this target. RAB21, which belongs to the Rab family of monomeric GTPases, plays a role in integrin internalization and recycling. As a result, the encoded protein is involved in cytokinesis during the mitosis. Loss of RAB21 in the tumor induces chromosome number aberrations and malignancy [25]. To further evaluate the role of miR-183 in regulating RAB21, we generated luciferase reporters with 33 bp of the predicted target regions from the 3′-UTR of RAB21, and co-transfected the reporter with miR-96, miR-182 and miR-183 overexpression vectors and empty vector. The luciferase assay results showed that miR-183 repressed luciferase activity dramatically in the reporter derived from the RAB21-targeted region compared with the empty vector, and miR-96 and miR-182 had no effect on the luciferase activity of the RAB21 reporter. As a negative control, the luciferase activity of cells containing the empty pGL3-Promoter vector was not affected by any miR-183/-96/-182 cluster miRNA ( Figure 7C). The protein levels of RAB21 were also determined in both MCF-7 and T47D overexpression cell lines with β-tubulin used as an internal control. The data showed that RAB21 protein was significantly decreased in the miR-183 overexpression cell lines, but not in miR-96 and miR-182 overexpression cell lines compared with the empty vector control cell lines and wild-type cells ( Figure 7D). As loss of RAB21 in the tumor would induce chromosome number aberrations, we checked the nucleus aberration in miR-183/-96/-182 cluster overexpression MCF-7 cells. Phalloidin and DAPI counterstaining results showed that the bi-and multinuclear cells were accumulated in miR-183 overexpressed cells but not in miR-96 and miR-182 overexpressed cells ( Figure 7E). All these data indicated that miR-183 targeted the RAB21 gene directly in breast cancer and induced aneuploidy.

Discussion
The MiR-183/-96/-182 cluster is a conserved polycistronic miRNA cluster that is highly expressed in several tumor types. Although it is well known that the expression level of this miRNA cluster is increased in breast cancer, its biological roles and the regulatory mechanisms governing MiR-183/-96/-182 expression in breast cancer are still unclear. Here, we report that miR-96, miR-182 and miR-183 expression levels are significantly higher in breast cancer compared to the NAT, and the transcription pattern of miR-183/-96/-182 is irregular in breast cancer as the correlation between miR-182 and miR-183 expression dropped dramatically in tumor samples. The expression of miR-183/-96/-182 is not upregulated in a specific breast cancer subtype. It is overexpressed in all kinds of breast cancer -ductal or lobular, luminal or basal, early-stage or late-stage -but there are some differences in their expression patterns. For example, miR-96 and miR-183 were lower in lobular carcinoma than in ductal carcinoma and other types of carcinoma. The levels of miR-96 and miR-183 were also lower in ER+ and PR+ cancers than in ER− and PR− cancers, but miR-182 was almost the same, even a little higher in ER+ cancers. Among the four different subtypes of breast cancer, miR-96 and miR-183 levels were higher in HER2-enriched breast cancers than other types; miR-182 was lower but miR-183 was higher in basal-like breast cancers than other types of breast cancer. We also compared the miRNA expression levels in different breast cancer cell lines based on their molecular markers. We found that miR-96 is only upregulated in SK-BR-3 and BT-20 cells, whereas miR-182 and miR-183 are upregulated in most of the breast cancer cell lines tested except for MDA-MB-231. Basically, the cell line data closely match the clinical analysis. MiR-96 and miR-183 levels are higher in HER2-enriched cell line SK-BR-3. MiR-96 is lower in ER+ and PR+ cancers than in ER− and PR− cancers. MiR-182 is higher in luminal breast cancer than basal breast cancer. MDA-MB-231 is the only exception. It is an ER− and PR− cancers, but its expression of miR-183/-96/-182 is low. Because MDA-MB-231 is a basal B/ claudin-low breast cancer cell line, which lacks common epithelial cell features and most closely resembles the mammary epithelial stem cell [26], we think its regulation of miR-183/-96/-182 is different to other breast cancer cell lines. Our data were similar to those reported by Riaz and colleagues. Based on their work, 51 human breast cancer cell lines were divided into two groups: the first major group included 33 cell lines, which was a luminal-like group; the second minor group included 18 cell lines, which was a basal-like group. Seventeen miRNAs, which included miR-182, showed significantly higher expression in the major cluster compared with the other miR-NAs. They also found that the expression of miR-183/-96/-182 is low in MDA-MB-231 cells [23]. Although the miR-183/-96/-182 cluster is transcribed in the same pri-miRNA, the expression profile of each miRNA varies between different cell lines, which indicate that their subsequent processing or stability are regulated in different ways. An interesting phenomenon is that from the 102 patient samples of TCGA dataset, miR-182 only increases 4.2 (± 1.1)-fold in tumor samples, but miR-96 and miR-183 increase 8.4 (± 1.1)-and 7.5 (± 1.1)-fold in tumor samples. The correlation between the expressions of miR-182 and miR-183 dropped dramatically in tumor samples. This phenomenon was also confirmed in HSF2 and ZEB1 overexpression cell lines, as the expressions of miR-96 and miR-183 were increased significantly, but not miR-182. We think it is because the transcription of miR-183/-96/-182 is very fast in cancer; some pri-miRNA is not complete and the transcription stalls before miR-182.
We also identified two transcriptional factors that regulate the transcription of the miR-183/-96/-182 cluster, ZEB1 and HSF2. ZEB1, which is a zinc finger transcription factor, is involved in the epithelial-mesenchymal transition and promotes metastasis in cancer [27,28]. Although most work has concentrated on the capacity of ZEB1 to repress gene expression, several groups demonstrated that ZEB1 can also activate transcription of downstream targets [28,29]. HSF2 binds heat shock promoter elements (HSE) and activates transcription. Although there is little evidence on the involvement of HSF2 in tumorigenesis, it can play a role indirectly by modulating HSF1 [30]. Previous studies also report that HSF2 regulates the proto-oncogene c-fos and may be involved in tumorigenesis [31]. Our findings show that ZEB1 and HSF2 activate the transcription of the miR-183/-96/-182 cluster, which gives us new insights into how ZEB1 and HSF2 enhance tumorigenesis.
The biological role of the miR-183/-96/-182 cluster in breast cancer is complicated. In our experience, this cluster functions more like an oncogene in breast cancer as it increases cancer cell proliferation and migration. Most previous and recent publications support this conclusion, especially for miR-182, which has been confirmed by many groups to induce breast cancer metastasis [6,[32][33][34]. Mir-96 is also proposed to be an onco-miRNA in breast cancer [5,6], but the role of miR-183 is more complex. It represses the expression of EGR1 and functions as an oncogene in breast cancer [35], but it also targets the Ezrin gene and inhibits cell migration in T47D cells [12]. Our results support a pro-oncogenic role for miR-183 in breast cancer, because upregulated expression of miR-183 by lentivirus in MCF-7 cells induces cell proliferation and migration. The effects of knockdown of miR-183/-96/-182 cluster are more complicated, and depend on the knockdown efficiency and specificity. We did not observe obvious changes after inhibition of miR-183, but we found a significant decrease in cell growth rates and S phase cell percentages in miR-96 and miR-182-inhibited cells. Two reasons can explain these results. First, the knockdown efficiency of miR-183 antagomir is lower than miR-96 and miR-182 antagomir. Second, miR-96 and miR-182 target FOXO1, but miR-183 does not [6]. MiR-96 and miR-182 might compensate partial functions of miR-183, but miR-183 cannot replace the function of miR-96 and miR-182 on inhibition of FOXO1.
Long-term inhibition of three miRNAs by sponge elements induced cell death and apoptosis in T47D cells, but we did not detect apoptosis with a single antagomir transfection. Inhibition of two or three of the cluster members at one time induced apoptosis, though some of them were not statistically significant (Additional file 8: Figure S5). These data indicate that these three miRNAs are redundant; they may be complimentary to each other. Knockdown of miR-183 had little effect on its own, but it had collaborative effects with the other two miRNAs.
We identified RAB21 as a target gene of miR-183 in both mRNA and protein levels, and also confirmed that overexpression of miR-183 induced accumulation of biand multinuclear cells. RAB21 is involved in the targeted trafficking of integrins via its association with integrin alpha tails. As a consequence, RAB21 regulates cell adhesion and migration [36]. In mitotic cells, integrin trafficking regulated by RAB21 is necessary for cytokinesis and cytokinesis failure will induce aneuploidy and oncogenic transformation [25,37]. This information may answer the question why miR-183 has dual effects in breast cancer. In some cases, repression of RAB21 results in decreased cell mobility, but in other cases, repression of RAB21 may lead to cytokinesis failure and aneuploidy. The 3′-UTR of RAB21 matches the seed sequence of miR-183, but not miR-96 nor miR-182. So, only miR-183 can inhibit the expression of RAB21. As the phenotype is similar no matter which of the three miRNAs is overexpressed in MCF-7 and T47D cells, RAB21 down regulation itself is not enough to explain the phenotype. Some other mechanisms are also involved in the regulation of cell proliferation and migration. For example, inhibition of FOXO1 by miR-96 and miR-182 will increase cell proliferation.
We identified two regulators (ZEB1 and HSF2) and one target gene (RAB21) for the miR-183/-96/-182 cluster in breast cancer cell lines. How do they work in clinical samples? We looked for correlation between miR-183/-96/-182 cluster miRNAs and their target/regulators by analysis of 508 clinical samples from TCGA data (Additional file 9). Because the correlations between miRNAs and their targets/regulators are not simply negative or positive correlations, we did not find any direct correlations between these miRNAs and the expressions of HSF2, ZEB1 and RAB21 based on the TCGA data analysis. But there were some interesting correlations between them in different subtypes. MiR-96 and miR-183 weree lower in ER+ and PR+ breast cancers than ER− and PR− breast cancers; in the meantime, their regulator, HSF2 level was lower and their target, RAB21 level, was higher in ER+ and PR+ breast cancers than ER− and PR− breast cancers (Additional file 4: Table S4). Subtype analysis also confirmed our findings. HSF2 level was high in basal breast cancers, which are miR-183-enriched breast cancers; and RAB21 level was low in HER2 and basal breast cancers, which are miR-96-and/or miR-183-enriched breast cancers (Additional file 4: Table S5). MiR-182 was not strongly correlated with the levels of HSF2 because its transcription is not controlled by HSF2 ( Figure 3D). There is still a complicated phenomenon that requires explanation, which is that the ZEB1 level was negatively correlated with miR-96 and miR-183 (Additional file 4: Table S4, S5). In MCF-7 cells, ZEB1 upregulates the expressions of miR-96 and miR-183 (Figure 3), and Graham et al. also report that ZEB1 is more expressed in ER/PR− breast cell lines than ER/PR+ breast cell lines [38]. However, in clinical samples, ZEB1 was enriched in ER/PR+ samples. Considering ZEB1 is a transcription factor that can either activate or repress its target genes, we think it functions differently in breast cancer cell lines and breast cancer patients. In patients, ZEB1 may repress the transcription of miR-183/-96/-182 cluster. This conclusion needs further work for confirmation, but nevertheless, ZEB1 plays an important role in the regulation of miR-183/-96/-182 cluster.

Conclusion
We found that the miR-183/-96/-182 cluster is highly expressed in most breast cancers, and its transcription is disordered in breast cancers. The miR-183/-96/-182 cluster is transcribed in the same pri-miRNA and its transcription is regulated by ZEB1 and HSF2. It increases breast cancer cell proliferation, promotes cell migration and is essential for cell survival. Also, miR-183 targets the RAB21 gene directly in breast cancer. In summary, the miR-183/-96/-182 cluster is upregulated in most breast cancers. It functions as