MicroRNA-874–mediated inhibition of the major G1/S phase cyclin, CCNE1, is lost in osteosarcomas

The tumor microenvironment is characterized by nutrient-deprived conditions in which the cancer cells have to adapt for survival. Serum starvation resembles the growth factor deprivation characteristic of the poorly vascularized tumor microenvironment and has aided in the discovery of key growth regulatory genes and microRNAs (miRNAs) that have a role in the oncogenic transformation. We report here that miR-874 down-regulates the major G1/S phase cyclin, cyclin E1 (CCNE1), during serum starvation. Because the adaptation of cancer cells to the tumor microenvironment is vital for subsequent oncogenesis, we tested for miR-874 and CCNE1 interdependence in osteosarcoma cells. We observed that miR-874 inhibits CCNE1 expression in primary osteoblasts, but in aggressive osteosarcomas, miR-874 is down-regulated, leading to elevated CCNE1 expression and appearance of cancer-associated phenotypes. We established that loss of miR-874–mediated control of cyclin E1 is a general feature of osteosarcomas. The down-regulation of CCNE1 by miR-874 is independent of E2F transcription factors. Restoration of miR-874 expression impeded S phase progression, suppressing aggressive growth phenotypes, such as cell invasion, migration, and xenograft tumors, in nude mice. In summary, we report that miR-874 inhibits CCNE1 expression during growth factor deprivation and that miR-874 down-regulation in osteosarcomas leads to CCNE1 up-regulation and more aggressive growth phenotypes.

The poorly vascularized tumor microenvironment is associated with conditions of hypoxia and growth factor deprivation to which the cancer cells adapt for subsequent tumor progression and metastasis (1,2). Serum withdrawal has been used to mimic growth factor deficiency, which results in stalling of the cell cycle progression (3)(4)(5). Movement of the cells from one phase to the next is regulated by cyclins and their catalytic subunits, the cyclin-dependent kinases (CDKs), 2 which drive the transitions between the different phases (6). It has been known that one of the reasons for cell cycle stalling during growth factor deprivation is the decrease of cyclin E, and independent mechanisms have been proposed to curtail cyclin E-CDK activity during serum starvation; protein kinase C associates with cyclin E and recruits it to the perinuclear zone during serum starvation, which is reversed when cells enter the cell cycle (7)(8)(9).
The role of non-coding RNAs in cell cycle regulation during serum starvation has been investigated previously; microarray analysis shows that a large number of microRNAs are significantly and sustainably altered during serum starvation (10). The levels of miR-16 increase during arrest in G 0 phase and decrease upon release into the cell cycle, suggesting that the rapid up-regulation of miR-16 induces cell cycle arrest (11). Similarly, it has been reported that miR-34a targets Sirt1 in response to oxidative and genotoxic stress, exacerbating hypoxia and serum starvation-induced apoptosis during serum starvation (12). Similar miRNA expression patterns were observed during stalling of cell growth either by contact inhibition or serum starvation, suggesting that common growth pathways are targeted (13). Thus, there is accumulating evidence that miRNAs modulate major physiological pathways during stress, playing a key role in adaptation of cancer cells. miRNAs from multiple families (miR-16, miR-7, miR-132, and miR-144) have been identified that target cyclin E, with their down-regulation being linked to different cancers (14 -17). Although it is widely accepted that miRNA-mediated cyclin E regulation has a vital role in oncogenesis, the role of cyclin E in regulating the cell cycle in the tumor microenvironment remains elusive. Moreover, no miRNA has been identified that targets CCNE1 during serum starvation. In recent years, it has been realized that long non-coding RNAs (lncRNAs) play a crucial role in tumorigenesis through their interactions with DNA, protein, and RNA. Functional annotation of mammalian cDNA has identified more that 25,000 RNAs that have a 5Ј cap and splicing and polyadenylation sites but no open reading frame (18). Deregulated expression of several lncRNAs has been associated with cancers. One such oncogenic lncRNA is HOTAIR, up-regulation of which indicates poor prognosis and enhanced metastasis in breast cancer (19). One of the mechanisms of action of lncRNA is to titrate away specific miRNAs, thereby relieving the repression of miRNA target genes. It has been shown that long non-coding RNA, PTENP1, acts as competing endogenous RNA (ceRNA) to sponge miR-106b and miR-93 (20). It has been reported that lncRNA XLOC_008466 is up-regulated in human non-small cell lung cancer, where it functions as a ceRNA, directly binding to miR-874, although the target gene that affects cell proliferation remained unidentified (21).
Serum withdrawal resembles the growth factor deprivation characteristic of the poorly vascularized tumor microenvironment and has aided in the discovery of key growth modulatory genes and miRNAs that have a role in the oncogenic transformation (22). In the present study, we investigated the miRNA-mRNA regulatory networks in contrasting physiological growth states, cell cycle entry and exit, to draw a correlation between them and tested these regulatory networks during osteosarcoma oncogenesis. We observed that miR-874 targets CCNE1 during serum starvation, whereas in growth factor-sufficient conditions, miR-874 is downregulated, resulting in CCNE1 prevalence and cell growth. We observed that the miR-874 regulation of CCNE1 is lost in osteosarcoma, leading to aggressive cancer-related phenotypes.

A screen identifies miRNA-mRNA regulatory networks during serum starvation
To identify the miRNAs that are differentially regulated during growth factor deficiency, we followed a systematic approach as described in the legend to Fig. 1A. Asynchronous HCT116 cells were grown in serum-free medium for 48 h, after which cells were resupplemented with 10% fetal bovine serum and further grown for 12 h. The asynchronous, serum-starved, and serumresupplemented cells were harvested for evaluating RNA and protein levels as well as cell cycle analysis; flow cytometry analysis established that serum-starved cells were arrested in G 0 /G 1 phase, indicating cell cycle exit, whereas after serum restimulation, cells progressed into the cell cycle (Fig. 1B). Serum withdrawal led to an increase in the levels of CDK inhibitors, p21 and p27, which were down-regulated as cells progressed into the cell cycle after serum restimulation (Fig. 1C). We performed global miRNA expression profiling of asynchronous, serumstarved, and serum-resupplemented cells from two biological replicates using miRCURY LNA TM microRNA Array (7th generation), which is based on miRBase version 19.0. High concordance among the biological replicates demonstrated that the individual sample sets had low variability, as seen in the heat map (Fig. 1D). Of a total of 567 miRNAs that were analyzed, we identified 51 differentially expressed miRNAs in asynchronous, serum-starved, and serum-resupplemented HCT116 cells with a Ͼ2-fold change using a significance analysis of microarray (SAM) plot (Fig. 1E). A majority of these differentially regulated miRNAs were up-regulated following serum withdrawal (48) and down-regulated upon serum restimulation (44), whereas only three miRNAs exhibited an inverse expression pattern. The quantitative real-time PCR analysis confirmed differential expression of 13 miRNAs: miR-663a, miR-874-3p (miR-874), miR-1469, miR-3178, miR-3195, miR-3196, miR-3687, miR-4279, miR-4449, miR-4488, miR-4497, miR-4508, and miR-4516. Each of these miRNAs was found to be up-regulated in serum-starved cells in comparison with asynchronously growing cells, whereas the expression levels of these miRNAs were down-regulated in serum-resupplemented cells in comparison with serum-starved cells (Fig. 1F).
A repertoire of E2F-regulated genes primarily involved in DNA replication and cell cycle is known to be down-regulated in the G 0 phase, and by using qRT-PCR we determined the levels of key cell cycle genes in asynchronous, serum-starved, and serum-resupplemented cells. The expression levels of major cell cycle and replication genes showed an inverse correlation to that of the differentially regulated miRNAs; serum deprivation resulted in down-regulation of the genes, and upon serum restimulation, the normal expression levels were restored (Fig. 1G). Immunoblotting confirmed the down-regulation of key cell cycle and replication proteins (Fig. 1H). Next, we sought to identify the putative targets of the serum starvation-induced miRNAs based on the miRNA target prediction algorithms, TargetScan, miRanda, miRWalk, and RNA22 (Fig. 1I). While deciding on the candidate miRNA to be pursued for further investigation, our decision was based on miRBase 21, which identifies "high-confidence" miRNAs based on several criteria, such as sufficient reads for each miRNA arm in the miRBase-collated deep sequencing data set, processing pattern of the mature sequences from the hairpin precursor form, and conservation of miRNA sequences across species, the methodology of which has been described in detail (23). Among the 13 validated miRNAs identified in our screen, only miR-874 was included in the high-confidence miRNA subset; therefore, we focused on this miRNA for further investigation. miR-874 is well conserved across mammals ranging from humans to rhesus monkeys to mice and has been detected in deep sequencing from multiple tissues: squamous cells, embryonic stem cells, cerebellum, heart, kidney, melanomas, etc. Another major factor in deciding on the candidate miRNA was the size; it has been reported that several miRNAs in the range of 17-19 nucleotides are possibly by-products of other non-coding RNAs like t-RNAs and sno-RNAs, originating from pseudo-precursor miRNA genes that lack the typical pre-miRNA hairpin structure. In our study, whereas miR-874 was 22 nt long, a majority of other miRNAs (miR-3178, miR-3195, miR-3196, miR-4279, miR-4488, miR-4497, miR-4508, and miR-4516) were found to be shorter than 19 nt. Therefore, for further evaluations, we followed up with miR-874 that fulfilled the basic acceptance criteria for a high-confidence miRNA. To evaluate whether the predicted genes are targets of miR-874, we transfected doublestranded mimic of miR-874, which increased its intracellular levels ( Fig. 1J). miR-874 overexpression led to a significant decrease in the transcript levels of cyclin E1, whereas the mRNA levels of the other predicted targets, Cdc25a, Mcm8, and Psf2, remained unaffected, thereby suggesting that CCNE1 is a target of miR-874 (Fig. 1K). The levels of CCNE2 did not decrease after miR-874 expression, indicating that it specifically targets CCNE1 for cell cycle regulation (Fig. 1K). Thus, our screen identified miR-874 as a serum starvation-induced miRNA.

miR-874 down-regulates cyclin E1
We have observed that the levels of miR-874 increase during serum starvation, whereas those of CCNE1 decrease, and this pattern is reversed during serum restimulation. We wanted to establish whether miR-874 was responsible for the oscillation in CCNE1 levels; therefore, we restored miR-874 levels during miR-874 regulates CCNE1 serum restimulation and assayed the effect on CCNE1 levels. We transfected miR-874 twice in serum-starved cells followed by serum stimulation to allow them to enter the cell cycle ( Fig.  2A, i). We observed that the presence of miR-874 prevented the cyclin E1 increase during cell cycle reentry ( Fig. 2A, ii and iii). However, unlike CCNE1, the increase in the levels of CCNE2 during cell cycle reentry was not affected by miR-874 overexpression ( Fig. 2A, ii). We observed that although the levels of

miR-874 regulates CCNE1
CCNE1 mRNA increase significantly in the 12-h period of serum resupplementation, the increase in protein levels was not proportionate to the increase in mRNA. We believe that although fresh mRNA is synthesized during serum restimulation, it is not completely translated in the 12-h period, as it has been previously reported that levels of many proteins, including cyclin E, are only partially restored during 12-16 h of serum restimulation (24 -26). Thus, we conclude that miR-874 specifically targets CCNE1, and its down-regulation permits CCNE1 increase during cell cycle reentry.

Inverse correlation between CCNE1 and miR-874 expression is conserved in human osteosarcoma
Next, we wanted to test whether miR-874 -mediated CCNE1 regulation is preserved in osteosarcoma, a condition where CCNE1 deregulation is well established (27). We compared the endogenous levels of miR-874 and CCNE1 in human normal osteoblastic cell line hFOB1. 19 and two other human osteosarcoma cell lines that differ in their aggressiveness: U2OS (rapidly proliferating, highly invasive, and migratory) and KPD (slowly proliferating, less invasive, and migratory) (Fig. 2B, i) (28). The levels of CCNE1 mRNA and protein were significantly higher in U2OS in comparison with both KPD and hFOB1.19 (Fig. 2B, ii and iii). The expression of miR-874 in all three of the cell lines was inversely correlated to CCNE1 levels (i.e. miR-874 was considerably low in U2OS as compared with KPD and hFOB1.19) (Fig. 2B, iv). This indicates an inverse correlation between CCNE1 and miR-874 expression in human osteosarcoma. We also established that the oscillation of miR-874 and CCNE1 during cell cycle reentry and exit is conserved in osteosarcoma cell lines (Fig. 2C, i-iv).

CCNE1 is a direct target of miR-874
For evaluating whether CCNE1 is a direct target of miR-874, we mutated the binding site of miR-874 in CCNE1 (Fig. 3A). The entire 3Ј-UTR of CCNE1 carrying either the WT or mutated binding site for miR-874 was cloned into pmirGLO Dual-Luciferase vector. When the wild-type CCNE1 3Ј-UTR was introduced into U2OS cells along with miR-874 mimic or positive control miR-16 mimic, it exhibited a significant reduction in the luciferase activity, but the mutation in the binding site at the 3Ј-UTR abrogated the inhibition on luciferase activity caused by miR-874 overexpression (Fig. 3B). Thus, miR-874 inhibits CCNE1 expression through direct interaction with its 3Ј-UTR. We have determined the effect of miR-874 overexpression on CCNE1 levels, and next we assessed the effect of loss of function of miR-874 on CCNE1 levels. We transfected U2OS cells with mimic of miR-874 or single-stranded antisense inhibitor of miR-874, alone or in combination, and evaluated the effect on CCNE1 expression. As expected, we observed a marked decrease in CCNE1 levels as a result of miR-874 overexpression, whereas CCNE1 protein increased marginally upon miR-874 inhibition (Fig. 3, C and D). However, transfection of miR-874 inhibitor reversed the CCNE1 down-regulation caused by miR-874 mimic expression, confirming that miR-874 regulates endogenous CCNE1 levels. Thus, miR-874 represses CCNE1 by binding to a site in its 3Ј-UTR.

miR-874 -mediated down-regulation of CCNE1 is independent of E2F transcription factors
It has been well established that the promoter for the CCNE1 gene harbors multiple binding sites for E2F transcription factors, and they regulate the expression of endogenous cyclin E (29,30). Thus, it is possible that the decrease of CCNE1 levels after miR-874 overexpression is due to the down-regulation of E2F transcription factors. We evaluated the levels of E2F1, E2F2, and E2F3 after miR-874 overexpression in asynchronously growing U2OS cells. We observed that although the CCNE1 mRNA levels were down-regulated, the levels of E2F transcription factors did not alter significantly (Fig. 4A). We also evaluated the levels of E2F1, E2F2, and E2F3 during serum starvation, and as expected, the levels of all three of the E2F transcription factors were down-regulated in serum-deficient cells (Fig. 4B). In addition, miR-874 overexpression during serum resupplementation did not significantly alter the endogenous levels of E2F1, E2F2, and E2F3 mRNA (Fig. 4B). There Figure 1. A screen to identify differentially regulated miRNAs during serum starvation. A, schematic outline of the strategy employed for identification of differentially regulated miRNAs upon serum starvation and serum resupplementation. B, HCT116 cells were serum-starved for 48 h, followed by resupplementation with fetal bovine serum for 12 h. Flow cytometry of propidium iodide-stained DNA demonstrates that serum starvation for 48 h led to an increase in the G 0 /G 1 population, indicating cell cycle exit, whereas serum stimulation for 12 h resulted in cell cycle reentry, seen as a clear accumulation in the S and G 2 /M phases. C, immunoblot shows induction of CDK inhibitors p21 and p27 after serum withdrawal and down-regulation of these marker proteins after serum stimulation. LC, loading control, a nonspecific band that displays equal protein load in different lanes. D, heat map representation of expression of most significantly differentially regulated miRNAs in asynchronous, serum-starved, and serum-resupplemented HCT116. Each column corresponds to an individual sample, whereas each row represents an individual miRNA. Relative expression is represented as a colorgram (green, high expression; red, low expression). E, SAM plots of two comparison pairs: serum-starved versus asynchronous and serum-resupplemented versus serum-starved. The two parallel dashed lines represent the cut-off threshold specified by the false discovery rate, thus displaying the total number of up-regulated (red dots) and down-regulated (green dots) miRNAs for each plot. The SAM plot identified 48 up-regulated and three down-regulated miRNAs during serum starvation and three up-regulated and 44 down-regulated miRNAs during serum resupplementation. F, the relative expression of 51 top differentially regulated miRNAs in asynchronous, serumstarved, and serum-resupplemented HCT116 was confirmed using a quantitative RT-PCR assay, where the expression of miRNAs was normalized with an endogenous control, RNU48. G, the mRNA expressions of selected cell cycle and DNA replication genes were individually determined using qRT-PCR in asynchronous, serum-starved, and serum-resupplemented HCT116 cells, where the expression of genes was normalized with an endogenous control, ␤ 2 -microglobulin. H, immunoblot of the same samples as described in C shows differential expression of some of the activator genes involved in the cell cycle pathway. I, computational integration of differentially regulated miRNA-mRNA pairs using four miRNA target prediction algorithms: miRanda, miRWalk, RNA22, and TargetScan. Putative targets of each miRNA are represented by different color codes as indicated based on the number of algorithms predicting a binding site. J, HCT116 cells transfected on 3 consecutive days with miR-874 mimic or negative control mimic (NC) were harvested 24 h after the last transfection, and the -fold change in miRNA expression (ϫ1000) is represented. K, effect of miR-874 overexpression on the mRNA levels of putative targets by qRT-PCR. The -fold change with respect to negative mimic transfected control is shown. Data are represented as the mean of two biological replicates Ϯ S.D. (error bars).

miR-874 regulates CCNE1
was a 30% reduction in the levels of E2F3 after miR-874 mimic transfection; however, because the expression of CCNE1 is known to be controlled by all three transcription factors, it is unlikely to result in major CCNE1 down-regulation. Thus, the decrease in CCNE1 levels caused by miR-874 is not due to down-regulation of E2F transcription factors. Transcription factor-binding site prediction algorithms identify 16 putative E2F consensus-binding sites in the 10-kb region surrounding

miR-874 regulates CCNE1
the genomic loci from where miR-874 is expressed (chromosome 5q31.2). We wanted to assay whether E2F transcription factors regulate the expression of primary transcript of miR-874. Therefore, we evaluated the levels of primary miRNA (pri-miR-874) upon E2F1 or E2F2 depletion by RT-PCR with primers that amplify the 201-nt region around the mature miR-874.
We observed that E2F1 or E2F2 siRNA effectively reduced the levels of targeted gene, but the levels of pri-miR-874 were not significantly reduced (Fig. 4C, i and ii). The results indicate that although miR-874 harbors several E2F consensus-binding sites, its expression may not be regulated in an E2F-dependent manner. . miR-874 -mediated down-regulation of CCNE1 is independent of E2F transcription factors. A, U2OS cells were transfected on 3 consecutive days with negative control mimic or miR-874 mimic followed by evaluation of CCNE1, E2F1, E2F2, and E2F3 transcripts. Ectopic miR-874 mimic expression resulted in decreased CCNE1 mRNA levels, but the levels of E2F1, E2F2, and E2F3 mRNA were not suppressed, demonstrating that miR-874 down-regulation of CCNE1 is independent of E2F transcription factors. The numbers indicate levels of individual transcripts relative to negative control mimic-transfected cells. B, asynchronously growing U2OS cells were serum-starved for 48 h, followed by evaluation of E2F1, E2F2, and E2F3 transcripts. As reported previously, E2F transcription factors were down-regulated after serum starvation in comparison with asynchronously growing cells (compare asynchronous to serumstarved). During serum starvation, the cells were transfected twice with miR-874 mimic or negative control mimic, replenished with FBS, and further grown for 12 h, following which cells were harvested. The levels of E2F1, E2F2, and E2F3 mRNA after miR-874 mimic transfection have been expressed relative to negative control mimic-transfected serum-resupplemented cells. E2F1, E2F2, and E2F3 mRNA levels were not significantly different in negative control mimic-and miR-874 mimic-transfected serum-resupplemented cells. C (i and ii), E2F transcription factors do not regulate miR-874 at the transcriptional level. U2OS cells were transfected on 3 consecutive days with control GL2, E2F1, or E2F2 siRNA as indicated, followed by the evaluation of E2F1, E2F2, and pri-miR-874 transcripts.

XLOC_008466 -miR-874 -CCNE1 ceRNA network exists in osteosarcomas
It has been reported that lncRNA XLOC_008466 functions as a competing endogenous RNA binding to miR-874. We wanted to assay whether the XLOC_008466 -miR-874 network affects CCNE1 expression in osteosarcomas (Fig. 4D). We depleted the endogenous levels of XLOC_008466 by transfecting a siRNA duplex that targets its 3Ј-end. Transfection of siRNA targeting XLOC_008466 decreased its endogenous levels to about half that of control cells (Fig. 4D). The decrease in XLOC_008466 levels resulted in a significant increase in miR-874 levels, establishing that XLOC_008466 functions as a natural sponge for miR-874. Concomitantly, up-regulation of miR-874 due to XLOC_008466 inhibition resulted in CCNE1 down-regulation. Thus, the XLOC_ 008466 -miR-874 -CCNE1 network exists in osteosarcoma, and it is likely that a balance of these antagonistic factors influences the progression of osteosarcoma.

miR-874 overexpression impedes S phase progression
Having identified CCNE1 as a direct target of miR-874, we next investigated the effect of miR-874 on cell cycle progression, particularly on S phase progression. U2OS cells transfected on 3 consecutive days with miR-874 mimic were either grown asynchronously or synchronized at the G 1 /S boundary by hydroxyurea treatment for 20 h and were subsequently released into S phase by further growing the cells in drug-free medium for 3 h. Cells at the three indicated time points were subjected to a short BrdU pulse of 20 min and were subsequently harvested for BrdU immunofluorescence. We observed that the BrdU incorporation in asynchronously growing cells transfected with miR-874 was reduced to 23% as compared with 49% in control cells (Fig. 5A, i and iii). When BrdU incorporation was measured distinctly upon S phase enrichment (HU ϩ 3 h), we observed that in comparison with 61% control cells that were BrdU-positive, only 15% of the miR-874transfected cells incorporated BrdU (Fig. 5A, ii and iii). Thus, miR-874 expression impedes S phase progression.

Loss of miR-874 -mediated control of cyclin E1 is a general feature of osteosarcomas
Next, we tested the miR-874 -CCNE1 regulation in HOS and MG-63 osteosarcoma cell lines. MG-63 is an osteosarcoma cell line that displays a high proliferation rate and clonogenic ability under anchorage-independent conditions and shows Ͼ90% tumor formation in in vivo tumorigenicity assays (28,31,32). HOS is a highly tumorigenic osteosarcoma cell line that displays high invasion and migration potential as well as high proliferation and clonogenic ability (28,33). Considered as a highly aggressive cancer cell line, HOS is utilized as a control for assaying tumorigenic properties. First, we tested whether HOS and MG-63 display an inverse pattern of CCNE1 and miR-874 expression in comparison with human normal osteoblastic cell line hFOB1. 19. We noted that the mRNA levels of CCNE1 were significantly higher in HOS and MG-63 in comparison with hFOB1.19 (Fig. 5B). On the other hand, the levels of miR-874 were considerably low in HOS and MG-63 as compared with hFOB1. 19 and, thus, inversely correlated to CCNE1 levels. We observed that miR-874 overexpression led to a significant decrease in the transcript levels of cyclin E1 in HOS and MG-63 cells (Fig. 5C). Thus, the high levels of CCNE1 observed in aggressive osteosarcomas were reversed by the restoration of miR-874. These results establish that the loss of miR-874mediated regulation of cyclin E1 is a characteristic feature of osteosarcomas. Moreover, ectopic expression of miR-874 in both HOS and MG-63 led to a marked decrease in the viable cell count (Fig. 5D, i and ii). Our results demonstrate that the restoration of miR-874 control of CCNE1 leads to suppression of osteosarcoma cell proliferation. Thus, by utilizing multiple osteosarcoma cell lines, we demonstrate that miR-874 is down-regulated in osteosarcomas, which results in elevated CCNE1 levels, and miR-874 restoration reverses high CCNE1 expression and the high rate of proliferation of aggressive osteosarcoma.

miR-874 inhibits tumorigenic phenotypes in osteosarcomas
To evaluate the effect of miR-874 on in vitro cell survival, we transfected miR-874 mimic, followed by ␥-irradiation and colony count determination at 11 days. miR-874 restoration negatively affected the clonogenic cell survival, with at least 50% inhibition in the colony formation capacity in non-irradiated as well as ␥-irradiated samples (Fig. 6A, i and ii). To assess the role of miR-874 in wound healing, a key determinant of malignancy, we performed a wound-healing assay in U2OS cells after transfection with miR-874 mimic. Confluent monolayers were scratched, followed by visualization of migrating cells into the wounded area. We observed that ectopic expression of miR-874 dramatically reduced the wound healing capacity of U2OS cells (Fig. 6B, i and ii). Collectively, our results indicate that miR-874 overexpression substantially inhibits cell movement and cell survival in osteosarcoma in vitro. We utilized in vitro transwell migration and invasion assays to investigate the effects of miR-874 on cell migration and invasion ability. We observed that the cell migration ability was suppressed by miR-874 overexpression in U2OS cells (Fig. 6C, i and ii). In addition, ectopic expression of miR-874 inhibited the invasion of U2OS cells through Matrigel (Fig. 6D, i and ii). Thus, by utilizing independent assays, we demonstrate that miR-874 overexpression inhibits tumorigenicity of osteosarcomas.

Restoration of CCNE1 suppresses the miR-874 -mediated inhibition of S phase progression
Our data demonstrate that miR-874 inhibits cell proliferation and impedes S phase progression by targeting CCNE1. To further illustrate that the inhibitory effect of miR-874 on cell proliferation was primarily due to down-regulation of CCNE1, we utilized a retroviral vector to express the coding region of CCNE1, which lacks the site for miR-874 binding. U2OS cells stably expressing CCNE1 were transfected on 3 consecutive days with miR-874, followed by incubation with nucleotide analog, BrdU, to evaluate the rate of DNA synthesis. Endogenous CCNE1 levels were decreased in all samples transfected with miR-874 mimic, whereas the exogenously expressed CCNE1 expression remained unaffected by miR-874 overexpression (Fig. 7A). As shown previously, transfection of miR-miR-874 regulates CCNE1 trol cells, but co-expression of miR-874 -resistant CCNE1 partially reverses the suppression of loss of cell viability, demonstrating that the effect of miR-874 on cell viability was primarily due to down-regulation of CCNE1 (Fig. 7B). Similarly, the BrdU-positive population that was significantly decreased after miR-874 overexpression was restored after co-expression of miR-874 -resistant CCNE1 (Fig. 7C). Collectively, these results prove that the anti-proliferative activity of miR-874 is primarily due to down-regulation of CCNE1.

miR-874 regulates CCNE1 miR-874 suppresses tumor formation and progression in nude mice
To explore the anti-tumorigenic activity of miR-874 in vivo, nude mice were injected with U2OS cells stably expressing miR-874. However, tumor formation was not seen in nude mice up to 45 days, possibly due to poor tumorigenicity of the cell line; thus, we carried out the in vivo functional study using HCT116-derived tumors in nude mice (28). We constructed a recombinant lentiviral vector stably expressing miR-874 (pLKO.1 miR-874) in HCT116. qRT-PCR confirmed a decrease in the expression level of CCNE1 in pLK0.1 miR-874 as compared with pLK0.1 control (Fig. 8A). HCT116 cells stably expressing either pLK0.1 control vector or pLK0.1 miR-874 were injected subcutaneously into the right flank of nude mice between 4 and 6 weeks of age. Tumor volume was measured every second day, and mice were sacrificed after 17 days, followed by excision of tumors. Notably, miR-874 -treated mice showed significant inhibition of tumor growth as compared with the control group (Fig. 8, B-E). The levels of cyclin E1 were significantly reduced in tumors excised from miR-874 -treated mice, indicating that the suppression of tumor induction and progression in vivo by miR-874 is primarily due to down-regulation of CCNE1 (Fig. 8F).

Discussion
Alteration in the expression levels of miRNAs and potential target genes are well characterized for several human cancers, but the regulatory circuits cannot be simply established, as multiple miRNAs could possibly target a gene and multiple genes could be potentially targeted by a miRNA. By analyzing the expression of miRNA and potential target mRNAs in contrasting physiological states, as we have done during cell cycle exit and cell cycle reentry, an interrelationship between them could be established. We investigated the miRNA-mRNA regulatory networks functional during serum starvation, which has been used to mimic growth factor deficiency in the tumor microenvironment, and tested them in osteosarcoma oncogenesis (4,34). miR-874 has been reported to be down-regulated in multiple cancers (breast cancer, gastric cancer, and head and neck squamous cell carcinoma), with its targets including CDK9, STAT3, and HDAC1 (34 -37). On the other hand, cyclin E has been known to be overexpressed in multiple malignancies, including carcinomas, lymphomas, leukemias, sarcomas, and adrenocortical tumors (38 -41). Although the dysregulation of miR-874 and cyclin E expression during oncogenesis was individually reported before our study, their relation was not established. We discovered the miR-874 -cyclin E1 regulation dur-ing growth factor deprivation and assayed it in human osteosarcoma cell lines that differ in their aggressiveness: KPD (slowly proliferating and less invasive cell line) and U2OS (rapidly proliferating and highly invasive cell line) (28). Comparative analysis of osteosarcoma cell lines with contrasting cancerrelated phenotypes offers an alternative to specifically address the molecular changes during progression of oncogenesis. Although the osteosarcoma cell lines of varying aggressiveness are not isogenically matched, nonetheless, we have attempted to correlate miRNA expression with aggressive growth phenotypes. Comparisons were also made with a primary osteoblast cell line, but because the osteoblasts have progressed into a specific differentiation lineage, the differences in gene expression would also be likely to result from its differentiation commitment (42). Our work clearly indicates a strong relationship of CCNE1 and miR-874 expression with the aggressiveness of the cancer. Overexpression of miR-874 led to a decrease in the transcript levels of cyclin E1 in U2OS, HOS, and MG-63 cells, establishing that the loss of miR-874 -mediated control of cyclin E1 is a general feature of osteosarcomas. The effect of CCNE1 repression due to miR-874 overexpression was apparent on the S phase progression, whereas co-expression of miR-874 -resistant CCNE1 largely reversed the effect. Unlike the miR-16 family, which targets multiple cell cycle genes, miR-874 specifically targets cyclin E1, exemplifying the diverse nature of cell cycle control and its subversion during oncogenesis (43).
It has been reported that expression of miR-874 was negatively correlated with TNM (tumor, node, and metastasis) stage in cancer patients (44). Significantly, patients with high levels of miR-874 demonstrate better survival; data analysis from the Cancer Genome Atlas reveals that bladder urothelial carcinoma patients with high levels of miR-874 displayed an overall 50% survival rate over a period of 3000 days in comparison with a 25% survival rate of patients exhibiting low miR-874 expression (45). Although the survival data are not available for osteosarcomas, for other cancers, such as hepatocellular carcinoma and pancreas adenocarcinoma, patients with high levels of miR-874 have a significantly better survival rate in comparison with patients with low levels of miR-874. The accumulating information on miR-874 (poor expression in cancer tissues, inverse correlation with cancer metastasis, and, most importantly, the reversal of aggressive cancer phenotypes upon its restoration) clearly establishes that miR-874 is an important molecule during tumorigenesis. The significance of miR-874 in human cancers is also apparent from the multiple levels of regulation that it is subjected to during carcinogenesis. XLOC_ 008466, a long non-coding RNA, functions as a competing endog-

miR-874 regulates CCNE1
enous RNA to miR-874, buffering it away from the cellular milieu; its overexpression results in the reversal of miR-874 function of inhibiting cell proliferation and invasion, and the targets of miR-874 are derepressed with XLOC_008466 expression (21). The same seed sequence of miR-874 that binds to the UTR region of its target genes is bound by XLOC_008466. It thus displays a level of complexity in regulation of growth-inhibitory miRNAs similar to that of miR-874 and thus presents the challenge of fully understanding the regulation of growth-modulating non-coding RNAs. Using a rat model of pregnancy and lactation, it has been shown that miR-874 increases osteoblastogenesis and mineralization during the weaning period, the stage that makes up for the maternal bone loss (46). It has been reported that miR-874 is significantly high during this period, when it induces osteoblast differentiation by inhibiting HDAC1 expression. In our study, we have shown that CCNE1 is targeted by miR-874, which may possibly contribute to the osteoblast differentiation reported in the above study (46). What would be required is an analysis of the levels of CCNE1 during the weaning period as well as testing of the role of CCNE1 in miR-874 expression-induced osteoblastogenesis. Recent studies have reported that the higher levels of miR-874 can serve as a marker identifying mild congenital disorder (47). It is possible that miR-874 -induced inhibition of cell proliferation contributes to the neurodegenerative processes. Understanding the biology of miR-874, as we have done in the present study, would help in addressing these questions in the future. In summary, we have analyzed the miRNA-mRNA regulatory networks during cell cycle entry and exit, two contrasting physiological growth states, and tested these regulatory networks during osteosarcoma oncogenesis. We discovered that miR-874 inhibits CCNE1 during growth factor deprivation and is down-regulated in osteosarcomas, leading to aggressive growth phenotypes (model described in Fig. 8G).

Cell culture, irradiation, cell synchronization, and cloning
U2OS, HOS, MG-63, and KPD (human osteosarcoma cell lines); HEK293T (human embryonic kidney cells with SV40 large T antigen cell line); and HCT116 (human colorectal cell line) were maintained in DMEM supplemented with 10% FBS along with 1% of 100 units/ml antibiotic and antimycotic solution at 37°C in a humidified atmosphere with 5% CO 2 . HFOB1.19 (human fetal osteoblastic cell line) was maintained in a 1:1 mixture of Ham's F-12 medium and DMEM supplemented with 10% FBS and 1% of 100 units/ml antibiotic and antimycotic solution at 34°C in a humidified atmosphere containing 5% CO 2 . For ionizing radiation treatment, cells were exposed to either 2 or 4 grays of ␥-radiation and were harvested 11 days post-treatment for a clonogenic survival assay. For serum starvation, cells were seeded at ϳ50% confluence in complete medium, and once cells adhered and attained morphology, the medium was removed, followed by two 1ϫ PBS washes, and the cells were cultured in medium completely devoid of FBS for 48 h. For arresting the cells at the G 1 /S transition, cells were grown in DMEM supplemented with 2 mM hydroxyurea for 20 h, after which the cells were harvested for immunofluorescence. For cloning of human CCNE1 coding sequence, CCNE1 cDNA was amplified by PCR and cloned into murine leukemia virus long terminal repeat-driven plasmid, pMX-puro-3NLS-GST-HA. HEK293T cells were transfected with pMX-puro-CCNE1 along with helper plasmids expressing the viral VSV-G envelope protein as well as the Gag and Pol proteins to generate viral particles. To obtain a stable U2OS cell line expressing CCNE1, U2OS cells were infected with the viral particles and selected with 1 g/ml puromycin after 48 h.

Cell cycle analysis and flow cytometry
For cell cycle analysis, the cells were harvested and fixed with 70% ethanol at 4°C for 1 h. Following fixation, the cells were washed with 1ϫ PBS, and the cell pellet was resuspended in 1ϫ PBS with 0.1% Triton X-100, 20 mg/ml RNase A, and 70 mg/ml propidium iodide, and then the stained cells were analyzed by flow cytometry. The flow cytometry data were acquired on a BD Biosciences FACSCalibur machine by Cell Quest Pro software. Cell cycle distribution was evaluated by the Dean/Jett/Fox method using FlowJo software. To study the BrdU incorporation, cells were cultured in medium containing 100 M BrdU (BD Biosciences) for 20 min, before harvesting. After fixation, Figure 6. miR-874 overexpression inhibits tumorigenic phenotypes in osteosarcomas. A (i and ii), clonogenic cell survival assay to evaluate the effect of miR-874 expression. U2OS cells were transfected on 3 consecutive days with either negative control mimic or miR-874 mimic, after which they were ␥-irradiated, allowed to grow for 11 days, and stained with crystal violet, after which the colonies were counted. ii, quantification of i, which demonstrates that the colony-forming ability is significantly decreased upon miR-874 overexpression. B (i and ii), wound-healing assay to evaluate the effect of miR-874 expression on cell migration. U2OS cells were transfected on 2 consecutive days with either negative control mimic or miR-874 mimic and grown to confluence, after which a wound was created using a micropipette tip. The extent of wound healing was monitored every 24 h. ii, quantification of i, which demonstrates that the wound-healing capability is profoundly reduced upon miR-   T1  T2  T3  T4  T5  T6  T7  T8  T9  T10 T11 T12   1

miR-874 regulates CCNE1
cells were treated with 2 N HCl for 15-20 min to denature the DNA, followed by a neutralization step of 5 min at room temperature with 0.1 M sodium tetraborate (pH 8.5). Cells were then washed with a blocking solution comprising 3% BSA in PBS containing 0.1% Triton X-100 followed by incubation with mouse anti-BrdU antibody (dilution 1:10 in blocking solution) conjugated to FITC for 1 h. After antibody staining, cells were washed with 1ϫ PBS, and DNA was stained with propidium iodide and run on a FACS machine as described above.

RNA extraction and quantitative real-time PCR
Total RNA was extracted from cells using TRIzol reagent (Invitrogen) and reverse transcribed into cDNA using Moloney murine leukemia virus reverse transcriptase (Invitrogen). For miRNA analysis, total RNA was poly(A)-tailed using Escherichia coli poly(A) polymerase (New England Biolabs) before reverse transcription as described previously (48). The qRT-PCRs were carried out in duplicates in a 10-l volume for the expression analysis. The reaction mixture contained SYBR Select master mix (1ϫ; Applied Biosystems), cDNA template, and forward and reverse gene-or miRNA-specific primers (0.1 M each). The target sequence amplification temperature profile followed was as follows: initial denaturation for 10 min at 95°C, followed by 40 cycles of 10 s at 95°C and amplification for 30 s at annealing temperature of 60°C. Finally, a melt curve analysis was carried out at a temperature range of 60 -95°C for 20 min. The small nucleolar RNA, RNU48, and BMG (␤-microglobulin) were used as internal controls for miRNA and mRNA quantification, respectively. Results were calculated using the ⌬⌬Ct method to determine the -fold change in expression between the experimental and control groups.

Indirect immunofluorescence and microscopy
To study the BrdU incorporation, U2OS cells grown on coverslips were cultured in medium containing 100 M BrdU for 20 min before harvesting. Cells were fixed with 4% formaldehyde for 10 min, followed by denaturation with 2 N HCl for 15-20 min and neutralization with 0.1 M sodium tetraborate (pH 8.5) for 2 min. Cells were then blocked with 3% BSA for 30 min and stained with mouse anti-BrdU antibody (BD Biosciences) for 1 h, followed by incubation with secondary antibody for 1 h. Finally, the cells were visualized under the microscope after mounting with Vectashield mounting reagent containing DAPI, which stains the nucleus. The secondary antibody used was conjugated to Alexa Fluor 488 and purchased from Invitrogen.

Immunoblotting and antibodies
For Western blotting, the whole-cell lysates from cells of almost equal confluence were prepared in a proportionate volume of Laemmli buffer and denatured at 95°C, followed by SDS-PAGE. The gel was transferred onto a nitrocellulose membrane, blocked with 3% BSA prepared in 1ϫ TBST. The membrane was then incubated with the appropriate antibody, washed, and probed with horseradish peroxidase-conjugated secondary antibody. Enhanced chemiluminescence was used to visualize the protein bands. Quantity One software was utilized to evaluate the levels of specific proteins, which were expressed after normalization with the protein-loading control. The following antibodies were used for Western blotting. Antibodies against CCNE1, CDK2, ␤-actin, CCNB1, and CDC25A were obtained from Santa Cruz Biotechnology, Inc. Antibodies against p21 and p27 were procured from Cell Signaling Technology. Antibody used to detect the HA epitope was purchased from Sigma.

miRNA microarray
Total RNA was isolated by the TRIzol method, and samples were outsourced to Exiqon for miRNA microarray analysis. Microarray analysis was conducted as dual-channel Hy3/Hy5 experiments on Exiqon's miRCURY LNA microRNA Array 7th generation, comprising 3100 capture probes, complementary to human, mouse, and rat, and their related viral sequences from the version 19.0 release of miRBase. A 500-ng RNA sample was labeled with an Hy3 fluorophore along with Hy5-labeled reference RNA sample (Exiqon, Vedbaek, Denmark) using Exiqon's miRCURY LNA microRNA HiPower labeling kit with the use of synthetic spike controls, Spike-in microRNA Kit version 2 (Exiqon), according to the manufacturer's protocol. Hybridization of labeled RNA to the array was performed in SureHyb chambers (Agilent Technologies) for 16 h at 56°C. Slides were washed according to the manufacturer's instructions and scanned at 10 m resolution using an Agilent G2505C DNA microarray scanner. Raw data were generated using Imagene version 9.0 software (BioDiscovery, Inc.), using an feature extraction protocol available on demand from Exiqon. These raw data have been submitted to the GEO database under accession number GSE103436.

miRNA target prediction
The putative miRNA targets were predicted using the following algorithms: TargetScan, miRanda, RNA22, and miRWalk. Tumor volume was measured on alternate days, and on the 17th day, mice were sacrificed, tumors were excised, and the tumor weight was determined as shown in C. D, representative photographs of nude mice injected with HCT116 cells expressing pLK0.1 control or pLK0.1 miR-874. Arrowheads point to the tumor. E, macroscopic images of the tumors on the 17th day revealed that the tumors formed in miR-874 -treated mice were significantly smaller than control mice. T1-T6 and T7-T12, tumors excised from control and miR-874 -expressing mice, respectively. F, immunoblot of the excised tumors revealed that cyclin E1 protein levels were low in the miR-874 -treated group as compared with the control group. The numbers indicate levels of cyclin E1 relative to control tumor T1. Data are represented as the mean Ϯ S.D. (error bars) of two biological replicates. G, model depicting miR-874 -mediated regulation of CCNE1 during cell cycle phases and osteosarcomas. During growth factor deprivation, miR-874 levels are induced, resulting in down-regulation of CCNE1 and cell cycle exit. However, during serum restimulation, miR-874 expression is decreased, leading to derepression of CCNE1 levels and resulting cell cycle reentry. An analogous miR-874 -CCNE1 regulation exists in osteosarcomas; primary osteoblasts or less aggressive osteosarcomas display a regulation that is similar to serum-starved cells, where high levels of miR-874 inhibit the expression of CCNE1, resulting in poor tumorigenicity. Aggressive osteosarcomas exhibit a miR-874 -CCNE1 regulation observed during cell cycle reentry (i.e. miR-874 is down-regulated, resulting in high CCNE1 levels and development of cancer-related phenotypes, such as increased migration and invasion).

Luciferase reporter assay
To construct a luciferase reporter plasmid, 3Ј-UTR fulllength fragment of CCNE1 that hosts the predicted miR-874binding site or mutated site was cloned downstream to the firefly luciferase into pmirGLO Dual-Luciferase miRNA target expression vector (Promega). For the luciferase reporter assay, U2OS cells were co-transfected with 750 ng of luciferase reporter plasmid harboring the 3Ј-UTR of the gene along with 50 nM miRNA mimic using Lipofectamine 2000 reagent. After 24 h of transfection, cells were harvested, and luciferase activity was measured in a Lumicount Luminometer (Packard) using the Dual-Luciferase reporter assay kit (Promega) according to the manufacturer's instructions. Firefly luciferase activity was normalized to Renilla luciferase activity.

Cell proliferation assays
For MTT cell proliferation assay, 30,000 cells in 500 l of medium/well were seeded in triplicates in 24-well cell culture dishes. The MTT substrate, thiazolyl blue tetrazolium bromide, was added to cells in culture at a final concentration of 0.5 mg/ml and incubated at 37°C. After 3-4 h, the cells were resuspended in 500 l of DMSO and shaken for 15 min. The quantity of formazan was measured by recording changes in absorbance at 570 and 630 nm (reference wavelength) using a microplate reader (BioTek PowerWave XS). For cell viability count, the trypan blue exclusion method was utilized where U2OS cells were transfected with miRNA mimic for 3 consecutive days, and 24 h after the last transfection, cells were collected and dissolved in 1 ml of 1ϫ PBS, and 20 l of cell suspension was stained with an equal volume of 0.4% trypan blue. Viable cells, which excluded trypan blue dye, were counted in quadruplicate using a glass hemocytometer.

Wound-healing assay
U2OS cells seeded in 6-well plates with an approximate confluence of 30 -40% were transfected with 50 nM miRNA mimic using Lipofectamine 2000 reagent for 2 consecutive days and were cultured until confluence. A wound was then created by manually scraping the cell monolayer with a 200-l pipette tip. The cultures were washed twice with 1ϫ PBS and supplemented with fresh medium. Cell movement into the wound was observed at four preselected time points (0, 24, 36, and 48 h) in eight randomly selected microscopic fields for each condition and time point. Images were acquired with a Nikon TE2000-S inverted fluorescence microscope. The distance traveled by the cells was determined by measuring the wound width at different time points, and the experiment was performed in triplicates.

Clonogenic assay
U2OS cells transfected with miRNA mimic were counted, and 1000 cells were seeded in a 6-well culture dish in triplicates. Following 2 h of seeding, cells were irradiated with 0, 2, or 4 grays of ␥-radiation. After 11-12 days of incubation, plates were gently washed with 1ϫ PBS and stained with 0.1% of crystal violet. Colonies with over 50 cells were manually counted.

Invasion and migration assays
U2OS cells (3 ϫ 10 4 ) resuspended in 250 l of serum-free medium were seeded into the upper well of the Matrigel-coated membrane of a transwell chamber (8-m pore size, Corning), for assaying cell invasion. For migration assays, the cells were seeded into the upper well of a millicell hanging cell culture insert of a transwell chamber (8-m pore size; Millipore). Serum-containing medium was added to the lower chamber, and cells were incubated at 37°C for 24 h. Subsequently, cells in the upper chamber were removed, and the cells migrating to or invading the bottom of the membrane were fixed with cold methanol and visualized under the microscope after mounting with Vectashield mounting reagent containing DAPI that stains the nucleus. Five random fields of each membrane were photographed and counted for statistical analysis.

Generation of lentivirus and infection into mammalian cell lines
For lentivirus preparation, lentiviral vector pLKO.1 containing mature miRNA sequences was co-transfected with packaging vector pMD2.G and envelope vector psPAX2 at a 4:3:1 ratio using Lipofectamine 2000 in HEK293T cells. Supernatant was collected after 48 h of transfection, pooled, and filtered using a 0.45-m filter. The filtered supernatant was used along with 1 g/ml Polybrene for infecting HCT116 cells. 24 h postinfection, cells were grown in medium containing 1 g/ml puromycin.

Tumorigenicity in vivo
A total of 12 female nude mice, between 6 and 8 weeks of age were randomly divided into two groups. Five million HCT116 cells stably expressing either pLKO.1 control or pLKO.1 miR-874 were suspended in 100 l of PBS along with 10% Matrigel and injected subcutaneously into the right flank of each nude mouse. Tumor growth was measured every alternate day using a vernier caliper, and the mice were euthanized after 17 days. The tumor volume was calculated using the formula, tumor volume ϭ (width 2 ϫ length)/2. Next, the tumors were extracted and weighed. For Western blotting, lysis buffer was added to the tumors in a 1:10 (w/v) ratio, followed by homogenization and sonication until a clear lysate was formed. Next, the lysates were centrifuged at 13,000 rpm for 30 min, and the clear supernatant was collected. The protein amount for each sample was estimated, and 20 g of each sample was mixed with 3ϫ Laemmli buffer and processed for SDS-PAGE.

Statistical analysis
The results are presented as mean Ϯ S.D. and were analyzed with Student's t test. A p value of Ͻ0.05 was considered significant unless noted otherwise.
Author contributions-T. G. and A. V. conceived the project, conducted the experiments, analyzed the data, and wrote the manuscript. P. K., M. K., V. K., R. S., R. D., P. P., M. M. K., and S. S. contributed to the experiments of this work.