A Link between mir-100 and FRAP1/mTOR in Clear Cell Ovarian Cancer

Abstract

MicroRNAs (miRNAs) are small noncoding RNAs that direct gene regulation through translational repression and degradation of complementary mRNA. Although miRNAs have been implicated as oncogenes and tumor suppressors in a variety of human cancers, functional roles for individual miRNAs have not been described in clear cell ovarian carcinoma, an aggressive and chemoresistant subtype of ovarian cancer. We performed deep sequencing to comprehensively profile miRNA expression in 10 human clear cell ovarian cancer cell lines compared with normal ovarian surface epithelial cultures and discovered 54 miRNAs that were aberrantly expressed. Because of the critical roles of the phosphatidylinositol 3-kinase/v-akt murine thymoma viral oncogene homolog 1/mammalian target of rapamycin (mTOR) pathway in clear cell ovarian cancer, we focused on mir-100, a putative tumor suppressor that was the most down-regulated miRNA in our cancer cell lines, and its up-regulated target, FRAP1/mTOR. Overexpression of mir-100 inhibited mTOR signaling and enhanced sensitivity to the rapamycin analog RAD001 (everolimus), confirming the key relationship between mir-100 and the mTOR pathway. Furthermore, overexpression of the putative tumor suppressor mir-22 repressed the EVI1 oncogene, which is known to suppress apoptosis by stimulating phosphatidylinositol 3-kinase/v-akt murine thymoma viral oncogene homolog 1 signaling. In addition to these specific effects, reversing the expression of mir-22 and the putative oncogene mir-182 had widespread effects on target and nontarget gene populations that ultimately caused a global shift in the cancer gene signature toward a more normal state. Our experiments have revealed strong candidate miRNAs and their target genes that may contribute to the pathogenesis of clear cell ovarian cancer, thereby highlighting alternative therapeutic strategies for the treatment of this deadly cancer.

Mir-100 is a candidate tumor suppressor microRNA that inhibits mTOR signaling and enhances sensitivity to the rapamycin analog RAD001 (everolimus) in clear cell ovarian cancer.

Ovarian cancer is the fifth leading cause of cancer death in women and the most lethal gynecologic malignancy in the Western world (1). Approximately 90% of ovarian cancers are epithelial (carcinomas) and are thought to arise via transformation of the ovarian surface epithelium (OSE), the single layer of squamous-to-cuboidal cells that constitutes the ovarian surface (2). The four major histological subtypes of ovarian carcinoma are serous (50% of ovarian cancers), endometrioid (20%), mucinous (10%), and clear cell (5%) (3,4,5). Each subtype is a genetically distinct disease with unique pathogenesis and different susceptibility to chemotherapy, yet the regulatory mechanisms underlying this heterogeneity are not well understood (5,6). Hence, the clinical management of ovarian carcinoma has remained relatively uniform across subtypes, thereby limiting progress in reducing mortality.

Clear cell ovarian cancers are characterized by cuboidal or hobnail epithelial cells with abundant glycogen-rich clear cytoplasm (7,8). Nearly 70% of clear cell ovarian carcinomas are diagnosed at an early stage (I and II), in contrast to 20% of serous ovarian carcinomas (9). Nonetheless, advanced stage clear cell disease has the worst prognosis of any subtype, with 5-yr disease-specific survival rates of 31.5% for stage III and 17.5% for stage IV cancer (9). These sobering statistics in part reflect resistance to platinum-based chemotherapy, and women with clear cell ovarian cancer also have a higher incidence of thromboembolic vascular complications and hypercalcemia (8,10,11,12). Similar to ovarian carcinomas of the endometrioid subtype, clear cell tumors are strongly associated with endometriosis (13,14,15).

The phosphatidylinositol 3-kinase (PI3K)/v-akt murine thymoma viral oncogene homolog 1 (AKT1)/mammalian target of rapamycin (FRAP1/mTOR) pathway is a key signaling system in cancer and multiple tissues including the ovary (16,17). The phosphorylation of phosphatidylinositol (4,5)-bisphosphate (PIP2) to phosphatidylinositol (3,4,5)-triphosphate (PIP3) by PI3K activates the pathway and downstream AKT signaling, whereas phosphatase and tensin homolog (PTEN) is an antagonist of this signaling pathway by reversing the phosphorylation status of PIP3 to PIP2. Constitutive activation of PI3K/AKT signaling is a distinguishing feature of clear cell and endometrioid ovarian carcinomas (5,7,16). Whereas endometrioid tumors demonstrate inactivating mutations in PTEN, gain of function mutations in PIK3CA, encoding the PI3K catalytic subunit, are more frequent in clear cell tumors than in any other subtype (18). The serine/threonine kinase mTOR is a principal downstream effector of AKT activation. Treatment of multiple ovarian cancer mouse models, including xenograft models of the clear cell subtype, with the mTOR inhibitor and rapamycin analog RAD001 (everolimus) delayed tumor development and/or dramatically reduced tumor burden, indicating the important therapeutic implications of specific inhibition of mTOR in vivo (19,20,21).

The discovery of miRNAs has added another layer of complexity to gene regulation in cancer. miRNAs are 21- to 23-nucleotide single-stranded noncoding RNAs that direct gene regulation through translational repression and degradation of complementary mRNA (22). Although target pairing in mammalian cells is often imperfect, the specificity of interaction is especially dependent on complementarity to nucleotides 2–8 of the mature miRNA, known as the seed sequence (22). Each miRNA is predicted to target hundreds of genes, and each transcript may interact with multiple miRNAs, underscoring the potential for broad, intricate regulatory networks.

Altered miRNA expression has been observed in a variety of human cancers, and miRNA signatures accurately reflect the developmental lineages and differentiation states of tumors (23,24). Several microarray profiling studies have found differentially expressed miRNAs between epithelial ovarian tumors and normal ovary, immortalized OSE cell lines, or primary OSE cultures, in addition to uncovering miRNAs that may predict chemoresponsiveness and prognosis (25,26,27,28,29,30,31). However, only a few miRNAs have been ascribed specific functional roles as oncogenes or tumor suppressors, mainly because identifying bona fide gene targets is a challenging task (reviewed in Ref. 32). For instance, TP53 inactivation is frequent in high-grade ovarian carcinomas and may subsequently decrease TP53-dependent miRNAs such as mir-34, leading to derepression of BCL2, CDK4/6, CCNE2, and MET (33). Reduced let-7i expression promotes chemoresistance (30), consistent with data indicating that let-7 family members are master regulators of cell proliferation and differentiation by suppressing KRAS, HRAS, MYC, and HMGA2 (34,35,36,37). Overexpression of mir-214 likewise enhances chemoresistance by targeting PTEN (27), whereas loss of function of mir-199a-5p up-regulates IκB kinase β and stimulates nuclear factor-κB signaling, possibly cultivating a protumor microenvironment (38).

These observations support the hypothesis that miRNAs contribute to the pathogenesis of ovarian cancer, but this work has predominantly centered on the most common serous subtype. Here, we performed deep sequencing to comprehensively profile and compare miRNA expression between an extensive panel of human clear cell ovarian cancer cell lines and short-term primary cultures of human normal ovarian surface epithelium. Putative target genes of differentially expressed miRNAs were identified by integrating in silico prediction algorithms with mRNA signatures. We then selected the top candidate miRNA, mir-100, which targets mTOR, as well as mir-22, mir-182, and mir-30a, and investigated the specific and global effects of each miRNA on the gene expression pattern of clear cell ovarian cancer, simultaneously performing preliminary validation for a multitude of presumptive target genes. Our experiments reveal influential roles for specific miRNAs in determining the clear cell gene signature and highlight candidate miRNA-mRNA interactions that might drive the formation or progression of this aggressive subtype of ovarian carcinoma.

Results

miRNAs are aberrantly expressed in clear cell ovarian cancer

We used Illumina next-generation sequencing technology to exhaustively identify the repertoire of approximately 18- to 30-nucleotide small RNAs expressed in human clear cell ovarian cancer cell lines and short-term primary cultures of human normal ovarian surface epithelium (NOSE). After passing the raw sequence data through a series of quality control filters (see Materials and Methods), on average we generated 3.8 million usable small RNA reads for each cancer specimen and 1.7 million usable small RNA reads for each NOSE culture. Approximately 50–70% of the reads for each sample mapped to known mature miRNAs in miRBase release 11.0, with 305 miRNAs detected in cancer, 227 miRNAs detected in NOSE (Table 1), and a total of 338 miRNAs detected in at least one sample. By comparing the miRNA signatures for clear cell ovarian cancer and NOSE, we found that 21 miRNAs were up-regulated and 33 miRNAs were down-regulated in cancer (P < 0.01, fold change >2) (Fig. 1A and supplemental data S2 published on The Endocrine Society's Journals Online web site at http://mend.endojournals.org).

Table 1.

Thirty most abundant miRNAs in NOSE and their average relative expression

miRNA	% miRNA population
hsa-let-7f	38.57
hsa-let-7a	20.12
hsa-mir-21	5.38
hsa-let-7g	4.87
hsa-let-7b	3.48
hsa-mir-103	2.85
hsa-mir-140-3p (−6.6)	2.83
hsa-let-7e	2.38
hsa-mir-31	2.33
hsa-mir-29a	2.06
hsa-let-7d	1.65
hsa-mir-320a	1.13
hsa-mir-191	1.09
hsa-mir-221	0.74
hsa-mir-100 (−117.7)	0.73
hsa-let-7i (+4.3)	0.67
hsa-mir-222	0.63
hsa-let-7c	0.52
hsa-mir-107	0.52
hsa-mir-152	0.52
hsa-mir-181a	0.48
hsa-mir-30a (+8.1)	0.39
hsa-mir-101	0.38
hsa-mir-423–5p	0.36
hsa-mir-130a	0.33
hsa-mir-22 (−7.0)	0.31
hsa-mir-24	0.28
hsa-mir-99b (−2.9)	0.27
hsa-mir-26a	0.23
hsa-mir-378	0.20

miRNAs in bold were aberrantly expressed in clear cell ovarian cancer, and their fold change in cancer vs. normal is in parentheses. 

Figure 1.

Expression profiling of miRNAs and genes in human clear cell ovarian cancer cell lines and NOSE cultures. A, Heat map representation of miRNAs overexpressed (yellow) and underexpressed (blue) in clear cell ovarian cancer compared with NOSE (P < 0.01, fold change >2). Rows, miRNAs; columns, Profiled samples. Two candidate oncogenic miRNAs, mir-182 and mir-30a, and two candidate tumor suppressor miRNAs, mir-22 and mir-100, are labeled. B, Heat map representation of transcripts overexpressed (yellow) and underexpressed (blue) in cancer vs. normal (P < 0.01, fold change >2). Rows, Transcripts; columns, Profiled samples. C, Numbers of predicted miRNA-mRNA functional pairs for each algorithm and intersection of algorithms based on anticorrelated expression in clear cell ovarian cancer.

The significant association between clear cell ovarian carcinoma and endometriosis has led to the hypothesis that endometriosis may be a direct precursor lesion for this subtype of ovarian cancer (13,14,15). Hence, we considered the possibility that in addition to NOSE, endometrium may be an appropriate control group for our study. We sequenced the miRNA repertoire of human endometrium and compared it with the miRNA signature of our clear cell lines. Interestingly, we found that 13 of the 21 miRNAs up-regulated and 26 of the 33 miRNAs down-regulated in clear cell ovarian cancer vs. NOSE were significantly misregulated in the same direction when we compared clear cell ovarian cancer with endometrium (supplemental data S1 and S7). Notably, the set of overlapping miRNAs included four miRNAs of particular interest for further analyses: mir-100, mir-22, mir-182, and mir-30a (described below). Thus, using either NOSE or endometrium as our control group provided a similar baseline for the identification of differentially expressed miRNAs in clear cell ovarian cancer.

Genes differentially expressed in clear cell ovarian cancer are predicted targets of altered miRNAs

One mechanism of miRNA-mediated gene silencing is through mRNA destabilization. Therefore, we explored the functional consequences of miRNA alterations in clear cell ovarian cancer by generating gene expression profiles of our clear cell lines and NOSE cultures. There were 1160 genes (1345 probes) up-regulated and 1366 genes (1709 probes) down-regulated in cancer compared with normal (P < 0.01, fold change >2) (Fig. 1B and supplemental data S3). To identify putative miRNA-mRNA functional pairs, we integrated miRNA and mRNA profiles using multiple algorithms for miRNA target prediction, namely TargetScan, PicTar, and miRanda. For genes that were up-regulated in clear cell ovarian cancer compared with NOSE, we searched for targeting miRNAs that were down-regulated, and vice versa. This approach allowed us to focus on interactions that might be especially relevant to our model system. The overall numbers of miRNA up-mRNA down and miRNA down-mRNA up prediction pairs for each algorithm and intersection of algorithms are illustrated in Fig. 1C, with the specific pairs provided as supplemental data (S4 and S5).

mir-100 overexpression represses mTOR and fibroblast growth factor receptor (FGFR)3, and mir-22 overexpression represses other key oncogenes in clear cell ovarian cancer cells

Our bioinformatic analyses resulted in a large collection of putative miRNA-mRNA interactions, leading us to hypothesize that changes in miRNA expression may regulate the gene signature of clear cell ovarian cancer. A subset of up-regulated miRNAs might function as oncogenes by silencing target tumor suppressors, and a subset of down-regulated miRNAs might be tumor suppressors by silencing target oncogenes. According to this hypothesis, if we reverse the expression of candidate oncogenic or tumor suppressor miRNAs in the cancer cell lines, then we should cause a concomitant reversal in the cancer gene signature and partially restore gene expression to a more normal state. These experiments would also help to differentiate functional miRNA-mRNA predictions from those that are purely correlative.

Of the miRNAs that were down-regulated in cancer compared with normal, we selected two candidates to overexpress in our clear cell lines: mir-100 (118-fold decrease) and mir-22 (7-fold decrease). mir-100 was abundant in NOSE cultures and nearly absent in eight of ten clear cell lines, making it the most differentially expressed miRNA. Several groups have found lower levels of mir-100 in ovarian cancer (25,26,27,31), and a recent study reported decreased expression of mir-22 in serous, endometrioid, and clear cell ovarian carcinomas compared with primary NOSE cultures (31). Using miRNA mimics, we overexpressed mir-100 in OVSAYO cells and mir-22 in ES-2 cells. After transfection, we compared gene expression profiles between overexpression cells and cells transfected with a negative control mimic. If down-regulation of these miRNAs was responsible for derepressing and inducing genes in clear cell ovarian cancer, then overexpression of these miRNAs should silence functional targets, including genes that were up-regulated in clear cell ovarian cancer compared with NOSE.

For each miRNA, global differences in gene expression between overexpression and control cells were subtle. Therefore, to test for enrichment of predicted miRNA targets, we used rank-based methods such as Gene Set Enrichment Analysis (GSEA) and the classical Spearman's rank sum statistic, both of which can detect modest but coordinate expression changes in groups of functionally related genes (39,40). For each dataset, we sorted all profiled genes in descending order of fold change in overexpression vs. control (i.e. gene most induced after overexpression at the top and gene most repressed at the bottom). Within this ranked list, we used enrichment analysis to capture the positions of predicted miRNA target genes from miRanda, PicTar, and TargetScan. For each of these gene sets, we determined whether the genes as a group were at or near the top of the ranked list (enrichment), at or near the bottom of the ranked list (antienrichment), or randomly distributed (no enrichment).

Using enrichment analysis, the entire sets of mir-100 and mir-22 target predictions were down-regulated as a group (antienrichment) after mir-100 and mir-22 overexpression, respectively (Fig. 2, A–C). For each miRNA, the subset of predictions that were elevated in clear cell ovarian cancer compared with NOSE were also repressed and significantly antienriched. These enrichment patterns were visually evident by the negative enrichment score curves and the greater density of miRNA target predictions near the bottom vs. the top of the ranked lists. Whereas miRanda, PicTar, and TargetScan each predicted targets that were down-regulated after miRNA overexpression, it was not possible to compare their performance using the described analytical approaches because different numbers of genes were predicted by each algorithm, with more genes providing greater power over fewer genes.

Figure 2.

Enrichment analysis of gene expression profiles after modulation of specific miRNAs in clear cell ovarian cancer cell lines. A, GSEA after mir-100 overexpression in OVSAYO cells. All genes on the microarray were ranked by fold change (overexpression/control), and the GSEA algorithm was used to identify significant patterns in gene sets of interest. For each gene set, vertical bars along the x-axis of the GSEA plot represent the positions of genes within the ranked list (i.e. their fold change). Negative GSEA enrichment score curve indicates antienrichment (down-regulation), and positive curve denotes enrichment (up-regulation). NES, Normalized enrichment score. Because there were very few genes overexpressed in clear cell ovarian cancer (OvCa up) that were also predicted targets of mir-100, we performed GSEA on all mir-100 target predictions by miRanda (green), PicTar (blue), or TargetScan (red), regardless of their expression pattern in ovarian cancer. B, GSEA after mir-22 overexpression in ES-2 cells to examine OvCa up genes that were also predicted mir-22 targets by PicTar or TargetScan. There were no mir-22 targets predicted by miRanda (September 2008 release, MIMAT0000077). C, Spearman's rank sum analysis after overexpression of mir-100 (top) or mir-22 (bottom). For each miRNA, the gene sets of interest included all predicted targets by miRanda, PicTar, or TargetScan; all genes overexpressed (OvCa up) or underexpressed (OvCa down) in clear cell ovarian cancer; and OvCa up genes that were also predicted miRNA targets by miRanda, PicTar, or TargetScan. Numbers of genes in each set are indicated in parentheses. Negative rank sum statistic indicates antienrichment (down-regulation), and positive statistic denotes enrichment (up-regulation). Colors correspond to the sets analyzed by GSEA in panels A and B. D, GSEA after knockdown of mir-182 in ES-2 cells to examine OvCa down genes that were also predicted mir-182 targets. E, GSEA after knockdown of mir-30a in OVSAYO cells. F, Spearman's rank sum analysis after knockdown of mir-182 (top) or mir-30a (bottom). Gene sets of interest were similar to panel C but instead included OvCa down genes that were also predicted miRNA targets. Colors correspond to the sets analyzed by GSEA in panels E and F. Two-sided significance levels: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. ctrl, Control; expt, experimental.

Of the 1160 genes overexpressed in clear cell ovarian cancer, those predicted to be targeted by mir-100 and mir-22 are listed in Tables 2 and 3, respectively. Overexpression of mir-100 down-regulated nine of 14 putative targets, whereas overexpression of mir-22 down-regulated 22 of 27 target predictions. Several of these interactions were independently validated by quantitative real-time PCR (QPCR). In support of our hypothesis that mir-100 and mir-22 are tumor suppressors in clear cell ovarian cancer, we confirmed significant down-regulation of multiple target oncogenes after miRNA overexpression, including the mir-100 targets FRAP1/mTOR and FGFR3 (Fig. 3A), and the mir-22 targets ARRB1, CLIP2, EVI1, and FRAT2 (Fig. 3B). Interestingly, we also found repression of the mir-100 target EIF2C2/AGO2 and the mir-22 target EDC3, two components of the miRNA machinery that were overexpressed in clear cell ovarian cancer. It was important that we included the mock transfection group during QPCR validation because in rare instances we observed unexpected effects after treatment with negative control mimic, specifically, increased FGFR3 and decreased EIF2C2/AGO2. Of all the predicted mir-100 or mir-22 target predictions that were overexpressed in clear cell ovarian cancer, EIF2C2/AGO2 was the only gene also predicted to be silenced by the negative control mimic based on its seed sequence, in agreement with our findings.

Table 2.

All mir-100 predicted target genes that were overexpressed in clear cell ovarian cancer

Gene	Description	Microarray fold change (clear cell vs. NOSE)	Microarray fold change (mir-100 overexpression)	Algorithm
FGFR3	Fibroblast growth factor receptor 3	+13.10	−2.62	T, P, M
FRAP1/mTORa	FK506-binding protein 12-rapamycin associated protein 1	+1.92	−1.85	T, P, M
CTDSPL	CTD (carboxy-terminal domain, RNA polymerase II, polypeptide A) small phosphatase-like	+3.86	−1.62	T, P, M
TRIB1	Tribbles homolog 1 (Drosophila)	+3.47	−1.28	T
CMTM6	CKLF-like MARVEL transmembrane domain containing 6	+3.12	−1.15	M
HLTF	Helicase-like transcription factor	+2.09	−1.12	M
EIF2C2/AGO2	Eukaryotic translation initiation factor 2C, 2	+3.29	−1.07	T, P, M
IMPDH1	IMP (inosine monophosphate) dehydrogenase 1	+2.31	−1.07	M
RAVER2	Ribonucleoprotein, PTB-binding 2	+2.44	−1.06	T, P, M
CCDC21	Coiled-coil domain containing 21	+2.30	+1.04	T
KAZALD1	Kazal-type serine peptidase inhibitor domain 1	+4.32	+1.06	M
E2F2	E2F transcription factor 2	+6.76	+1.07	M
CDC25A	Cell division cycle 25 homolog A (S. pombe)	+3.06	+1.61	M
PLAGL2	Pleiomorphic adenoma gene-like 2	+3.82	+1.63	M

Nine of 14 genes were down-regulated after mir-100 overexpression. Prediction algorithms: T, TargetScan; P, PicTar; M, miRanda. 

^aAlthough FRAP1/mTOR was less than 2-fold up-regulated in clear cell vs. NOSE, it was included in our analysis based on previous work validating the gene as a bona fide mir-100 target (44). 

Table 3.

All mir-22 predicted target genes that were overexpressed in clear cell ovarian cancer

Gene	Description	Microarray fold change (clear cell vs. NOSE)	Microarray fold change (mir-22 overexpression)	Algorithm
EDC3	Enhancer of mRNA decapping 3 homolog (S. cerevisiae)	+2.03	−2.22	T, P
FRAT2	Frequently rearranged in advanced T-cell lymphomas 2	+3.23	−1.70	T
EVI1	Ecotropic viral integration site 1	+22.41	−1.59	T
ARID3B	AT-rich interactive domain 3B (BRIGHT-like)	+4.17	−1.58	T, P
CHD7	Chromodomain helicase DNA-binding protein 7	+2.35	−1.58	T, P
CLIP2	CAP-GLY domain containing linker protein 2	+2.64	−1.48	P
AGBL5	ATP/GTP binding protein-like 5	+2.09	−1.44	T, P
FBX046	F-box protein 46	+2.23	−1.43	T
ZCCHC14	Zinc finger, CCHC domain containing 14	+2.14	−1.42	T
PDSS1	Prenyl (decaprenyl) diphosphate synthase, subunit 1	+2.05	−1.40	T
FAM125B	Family with sequence similarity 125, member B	+2.55	−1.37	T
BIN1	Bridging integrator 1	+4.75	−1.26	P
CYTH3	Cytohesin 3	+2.12	−1.26	T, P
MTHFD2	Methylenetetrahydrofolate dehydrogenase (NADP+ dependent) 2, methenyltetrahydrofolate cyclohydrolase	+4.34	−1.25	T, P
POGK	Pogo-transposable element with KRAB domain	+2.41	−1.19	T, P
CCDC47	Coiled-coil domain containing 47	+2.07	−1.18	T, P
MYST3	MYST histone acetyltransferase (monocytic leukemia) 3	+2.09	−1.17	T
ARRB1	Arrestin, β 1	+2.25	−1.09	T
RGS14	Regulator of G protein signaling 14	+2.09	−1.09	T
ACVR1B	Activin A receptor, type IB	+4.37	−1.07	T
ZNF618	Zinc finger protein 618	+2.59	−1.06	T
PPP1R9A	Protein phosphatase 1, regulatory (inhibitor) subunit 9A	+2.40	−1.05	T
MPHOSPH9	M-phase phosphoprotein 9	+2.35	1.00	P
ANKRD57	Ankyrin repeat domain 57	+3.21	+1.02	P
ERBB3	v-erb-b2 Erythroblastic leukemia viral oncogene homolog 3 (avian)	+13.27	+1.08	T, P
PLAGL2	Pleiomorphic adenoma gene-like 2	+3.82	+1.11	T, P
IGF2BP3	IGF 2 mRNA-binding protein 3	+8.48	+1.19	P

Of 27 genes, 22 were down-regulated after mir-22 overexpression. Prediction algorithms: T, TargetScan; P, PicTar; M, miRanda. 

Figure 3.

QPCR analysis of predicted target genes after miRNA overexpression. A, Relative quantity (RQ) of mir-100 predicted targets after mir-100 overexpression in OVSAYO cells. For each gene, groups not connected by the same letter are significantly different (P < 0.05). B, Levels of mir-22-predicted targets after mir-22 overexpression in ES-2 cells.

mir-100 inhibits mTOR signaling and enhances sensitivity to RAD001

Because of the quantitative nature of this work, our experiments focused on changes in mRNA expression after miRNA modulation. Studies from the Bartel and Rajewsky laboratories (41,42) have shown that, in parallel with mRNA destabilization, miRNAs regulate widespread repression of protein translation. To confirm mir-100 regulation of mTOR, we used a lentiviral system to overexpress mir-100 in OVSAYO cells and observed decreased mTOR protein levels as compared with cells transduced with a negative control vector or mock transduced cells (Fig. 4A). Furthermore, we observed decreased levels of the phosphorylated isoforms of 4EBP1 and p70 S6 kinase, two direct substrates of mTOR kinase activity that mediate mTOR signaling (Fig. 4A). Thus, mir-100 represses mTOR mRNA and protein levels and inhibits the mTOR pathway in clear cell ovarian cancer cells.

Figure 4.

Effect of mir-100 on mTOR signaling and RAD001 sensitivity in clear cell ovarian cancer cells. A, Protein levels of mTOR and phosphorylated isoforms of 4EBP1 and p70 S6 kinase were examined by Western blot on OVSAYO cell lysates after transduction of a lentiviral vector expressing mir-100 or a nontargeting sequence (negative control). β-actin levels served as a loading control. B, Viability of OVSAYO cells treated with varying concentrations of RAD001 after stable transduction with mir-100 or a negative control vector. Student's t test: *, P < 0.001; **, P < 0.0001; ***, P < 0.00001.

To determine the physiological significance of these alterations, we examined the effect of mir-100 on ovarian cancer cell viability in vitro. Under normal steady-state culture conditions, there was no difference in viability between OVSAYO cells stably transduced with mir-100 or a negative control vector (data not shown). However, addition of the mTOR inhibitor RAD001 (everolimus) to the culture media revealed the potent biological effect of mir-100. RAD001 is a rapamycin analog that has recently been shown to decrease tumor burden in a xenograft model of clear cell ovarian cancer (21). OVSAYO cells expressing mir-100 were significantly more vulnerable to RAD001 across a wide range of concentrations (Fig. 4B). Hence, our experiments indicate that by repressing the mTOR pathway, mir-100 potentiates the chemotherapeutic effect of RAD001 on clear cell ovarian cancer cells.

Knockdown of mir-182 and mir-30a induces predicted gene targets

In a complementary fashion, we tested the hypothesis that higher expression of specific miRNAs contributes to the clear cell ovarian cancer gene signature. Of the miRNAs that were up-regulated in clear cell ovarian cancer compared with NOSE, we selected two candidates to knockdown in our clear cell lines: mir-182 (8.6-fold increase) and mir-30a (8.1-fold increase). Previous studies have identified higher levels of these miRNAs or their seed-matched family members in ovarian cancer compared with primary NOSE cultures (31) and immortalized OSE cultures (28). We used miRNA hairpin inhibitors to knock down the activity of mir-182 in ES-2 cells and mir-30a in OVSAYO cells and then compared their gene expression profiles to cells transfected with a negative control inhibitor. If up-regulation of these miRNAs was responsible for silencing genes in clear cell ovarian cancer, then knockdown of these miRNAs should derepress and induce expression of bona fide miRNA targets, including genes that were down-regulated in cancer compared with NOSE.

By GSEA, we found that mir-182 knockdown caused up-regulation of mir-182-predicted targets that were underexpressed in clear cell ovarian cancer, no matter whether the targets were defined using miRanda, PicTar, or TargetScan (Fig. 2D and Table 4). Similarly, mir-30a knockdown induced mir-30a presumptive targets that were down-regulated in cancer, although in this case TargetScan predictions showed the best enrichment (Fig. 2E and Table 5). Using Spearman's rank sum statistic to analyze the knockdown results, we again observed significant enrichment for mir-182 and mir-30a predicted targets that were underexpressed in clear cell ovarian cancer (Fig. 2F). Furthermore, we found that the entire sets of mir-182 and mir-30a presumptive target genes, regardless of their expression patterns in cancer compared with normal, were also enriched after knockdown of mir-182 and mir-30a, respectively (Fig. 2F).

Table 4.

Top 25 mir-182 predicted target genes that were underexpressed in clear cell ovarian cancer and up-regulated after mir-182 knockdown

Gene	Description	Microarray fold change (clear cell vs. NOSE)	Microarray fold change (mir-182 knockdown)	Algorithm
CAMSAP1L1	Calmodulin regulated spectrin-associated protein 1-like 1	−2.19	+1.15	T, P, M
SNAP23	Synaptosomal-associated protein, 23 kDa	−4.36	+1.14	T, P, M
SP3	Sp3 transcription factor	−2.07	+1.14	T, P, M
CD164	CD164 molecule, sialomucin	−3.03	+1.13	T
ATXN1	Ataxin 1	−4.98	+1.12	T, P, M
PLCG1	Phospholipase C, γ 1	−2.44	+1.12	T, P
NAMPT	Nicotinamide phosphoribosyltransferase	−5.11	+1.11	T, P, M
BNC2	Basonuclin 2	−3.22	+1.11	T, P, M
EIF5A2	Eukaryotic translation initiation factor 5A2	−2.84	+1.11	T, P, M
ZHX1	Zinc fingers and homeoboxes 1	−2.16	+1.11	T, P
EFNB2	Ephrin-B2	−4.16	+1.10	T, P, M
ACVR1	Activin A receptor, type I	−2.32	+1.10	T, P, M
EVI5	Ecotropic viral integration site 5	−2.21	+1.10	T, P, M
PLOD2	Procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2	−26.14	+1.09	T, P
MBNL2	Muscleblind-like 2 (Drosophila)	−4.41	+1.09	T, P, M
SEPT7	Septin 7	−3.28	+1.09	T, P
DCUN1D4	DCN1, defective in cullin neddylation 1, domain containing 4 (S. cerevisiae)	−2.94	+1.09	T, P, M
AHR	Aryl hydrocarbon receptor	−2.54	+1.09	T
SMAD7	SMAD family member 7	−2.34	+1.09	T, P
YWHAG	Tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein, γ polypeptide	−3.59	+1.08	T, P, M
SBDS	Shwachman-Bodian-Diamond syndrome	−3.20	+1.08	T
ABL2	v-abl Abelson murine leukemia viral oncogene homolog 2	−2.69	+1.08	T
ADAM10	ADAM metallopeptidase domain 10	−2.31	+1.08	T, P, M
PDGFRA	Platelet-derived growth factor receptor, α polypeptide	−13.38	+1.07	T, P, M
GPRASP2	G protein-coupled receptor associated sorting protein 2	−2.14	+1.07	T, P, M

The table lists only those targets that were at least predicted by TargetScan. Prediction algorithms: T, TargetScan; P, PicTar; M, miRanda. 

Table 5.

Top 25 mir-30a predicted target genes that were underexpressed in clear cell ovarian cancer and up-regulated after mir-30a knockdown

Gene	Description	Microarray fold change (clear cell vs. NOSE)	Microarray fold change (mir-30a knockdown)	Algorithm
AVEN	Apoptosis, caspase activation inhibitor	−3.01	+1.19	T
DEXI	Dexamethasone-induced transcript	−2.26	+1.18	T, P
SCAMP1	Secretory carrier membrane protein 1	−2.45	+1.17	T, M
GALNT1	UDP-N-acetyl-α-d-galactosamine:polypeptide N-acetylgalactosaminyltransferase 1 (GalNAc-T1)	−2.42	+1.16	T, P, M
GJA1	Gap junction protein, α 1, 43 kDa	−7.59	+1.15	T, P, M
ACVR1	Activin A receptor, type I	−2.32	+1.15	T, P, M
P4HA2	Prolyl 4-hydroxylase, α polypeptide II	−5.34	+1.14	T, P, M
MARCKS	Myristoylated alanine-rich protein kinase C substrate	−7.93	+1.13	T, P
MBNL2	Muscleblind-like 2 (Drosophila)	−4.41	+1.13	T, M
DACT1	Dapper, antagonist of β-catenin, homolog 1 (Xenopus laevis)	−2.72	+1.13	T, M
EXTL2	Exostoses (multiple)-like 2	−2.48	+1.12	T, P, M
MXRA5	Matrix-remodeling associated 5	−18.47	+1.11	T
PDE5A	Phosphodiesterase 5A, cGMP-specific	−13.27	+1.11	T
ATXN1	Ataxin 1	−4.98	+1.11	T, P, M
FKBP3	FK506 binding protein 3, 25 kDa	−2.47	+1.11	T, M
SSR3	Signal sequence receptor, γ (translocon-associated protein γ)	−2.35	+1.11	T, M
DSTYK	Dual serine/threonine and tyrosine protein kinase	−2.31	+1.11	T, P
S100PBP	S100P-binding protein	−2.11	+1.11	T, M
C14orf129	Chromosome 14 open reading frame 129	−2.01	+1.11	T, M
NTNG1	Netrin G1	−6.37	+1.10	T, M
BNC1	Basonuclin 1	−3.91	+1.10	T, P, M
FRMD4A	FERM domain containing 4A	−2.47	+1.10	T
NT5E	5′-Nucleotidase, ecto (CD73)	−2.39	+1.10	T
SEC23A	Sec23 homolog A (S. cerevisiae)	−2.03	+1.10	T, P, M
CALD1	Caldesmon 1	−6.46	+1.09	T, M

The table lists only those targets that were at least predicted by TargetScan. Prediction algorithms: T, TargetScan; P, PicTar; M, miRanda. 

mir-22 overexpression and mir-182 knockdown shift the global gene expression pattern of clear cell ovarian cancer toward a more normal state

Using enrichment analysis, we also investigated the global effects of specific miRNAs on the clear cell ovarian cancer gene signature. Surprisingly, the complete set of 1366 genes down-regulated in clear cell ovarian cancer compared with NOSE was significantly enriched and up-regulated after overexpressing mir-22 (Fig. 2C) or inhibiting the activity of either mir-182 or mir-30a (Fig. 2F), irrespective of whether these genes were predicted targets. Although mir-100 overexpression and mir-30a knockdown up-regulated the group of 1160 genes already overexpressed in cancer vs. normal, mir-22 overexpression and mir-182 knockdown caused significant antienrichment and down-regulation of this gene set. Thus, mir-22 overexpression and mir-182 knockdown attenuated the global gene signature of clear cell ovarian cancer, both through direct effects on predicted targets and widespread secondary effects on nontarget genes that are aberrantly expressed.

Discussion

In the present study, we identified putative oncogenic and tumor suppressor miRNAs in human clear cell ovarian cancer, the subtype with the worst advanced stage prognosis. We used next-generation sequencing technology to comprehensively profile miRNA expression in a broad spectrum of clear cell ovarian cancer cell lines. Although there is no perfect control group for ovarian cancer profiling studies (43), there are considerable drawbacks to using either whole ovary (e.g. contamination from other cell types) or immortalized OSE cultures (e.g. artifacts from immortalization). Selecting an appropriate baseline is further complicated by evidence of a strong association between endometriosis and ovarian carcinomas of the clear cell and endometrioid subtypes (13,14,15). In one meta-analysis, approximately 30% of clear cell ovarian cancers demonstrated concurrent endometriosis in the same ovary (13). The pathological basis for this correlation remains unknown, although several hypotheses have been proposed. In some cases, foci of endometriosis may directly develop into clear cell ovarian carcinoma; histological proof of this transition has been reported, albeit rarely, perhaps because it is difficult to capture (13,14,15). Another potential scenario is that the presence of ovarian endometriosis may create a microenvironment that fosters the malignant transformation of OSE cells (15). It is also conceivable that endometriosis and clear cell ovarian cancers develop as independent lesions that coexist because of shared risk factors or common mechanisms of pathogenesis (13,14,15). To account for the possibility that clear cell ovarian carcinomas may arise from OSE cells or from endometriosis, we compared clear cell ovarian cancer cell lines both to short-term primary cultures of human NOSE and to human endometrium.

Of the 54 miRNAs that were aberrantly expressed in clear cell ovarian cancer vs. NOSE, 39 miRNAs were misregulated in the same direction when we compared clear cell ovarian cancer to endometrium, demonstrating that these two control groups provided a similar baseline for comparison. Many of these miRNAs had predicted gene targets with anticorrelated expression patterns, including two potential tumor suppressors (mir-100, mir-22) that were down-regulated in cancer and two potential oncogenes (mir-182, mir-30a) that were up-regulated. Despite the characterization of several gene targets for these miRNAs in other model systems, and the broad evidence implicating these miRNAs in ovarian cancer, no strong candidate target genes in ovarian cancer had been identified until now. We found that overexpression of mir-100 and mir-22 caused antienrichment (down-regulation) of predicted targets that were overexpressed in clear cell ovarian cancer. On the other hand, knockdown of mir-182 and mir-30a caused enrichment (up-regulation) of predicted targets that were underexpressed in clear cell ovarian cancer.

The PI3K/AKT signaling pathway regulates survival, proliferation, and growth in multiple tissue types including the ovary and is frequently activated in clear cell ovarian carcinoma. Among the myriad downstream effectors of PI3K/AKT, FRAP1/mTOR is a serine/threonine kinase that regulates protein synthesis and cell growth (16). In concordance with previous work demonstrating that mTOR is a bona fide target of mir-100 (44), we found that up-regulating mir-100 activity inhibited mTOR signaling in clear cell ovarian cancer, as evidenced by lower mTOR mRNA and protein levels and reductions in the phosphorylated isoforms of the mTOR substrates 4EBP1 and p70 S6 kinase. The functional consequence of this repression was enhanced sensitivity of mir-100-expressing clear cell ovarian cancer cells to the rapamycin analog RAD001 (everolimus). RAD001 is an mTOR inhibitor that has recently been shown to decrease tumor growth in a xenograft model of clear cell ovarian cancer (21). In agreement with evidence suggesting that cisplatin resistance in clear cell ovarian cancer cells correlates with increased activation of AKT and mTOR, RAD001 efficacy was greater in cisplatin-resistant than cisplatin-sensitive disease (21). Whether the same pattern holds true for combined treatment with RAD001 and mir-100 remains to be determined. It is tempting to speculate that the loss of mir-100 in clear cell ovarian cancer may contribute to cisplatin resistance by relieving inhibition of mTOR signaling. Indeed, mir-100 was abundant in NOSE cultures and nearly absent in eight of 10 clear cell lines, making it the most differentially expressed miRNA with an overall 118-fold decrease.

In addition to mTOR regulation, mir-100 overexpression repressed FGFR3 levels. No roles for FGFR3 have previously been described in ovarian cancer, although FGFR3 activation has protumor activity in bladder carcinoma and multiple myeloma (45). Interestingly, FGFR signaling induces recruitment and activation of PI3K (46), suggesting that mir-100 may exert tumor suppressor activity through the coordinated repression of multiple components of the PI3K/AKT/mTOR signaling pathway.

The mir-22 primary transcript is located in the 5′- untranslated region of an open reading frame (C17orf91) on chromosome 17p13.3, and its expression is driven by the promoter for this gene (47). This locus has been defined as a minimal region of loss of heterozygosity in hepatocellular carcinoma (48), and reduced mir-22 expression secondary to loss of 17p has been reported in multiple myeloma (49). The MYC oncogene represses mir-22 expression in immortalized B cells, and overexpression of mir-22 decreased the growth of lymphoma cells in a xenograft mouse model (47). Despite these multiple lines of evidence identifying mir-22 as a candidate tumor suppressor, the only previously described mir-22 targets are ESR1 in breast cancer cells (50) and BMP7 and PPARA in chondrocytes (51).

We observed a 7-fold decrease in mir-22 expression in clear cell ovarian cancer, in part explained by the 4-fold decline in C17orf91 levels. Overexpression of mir-22 repressed multiple target oncogenes that were higher in clear cell vs. normal, including ARRB1, CLIP2, FRAT2, and EVI1. ARRB1 promotes invasion and metastasis of ovarian cancer cells by activating β-catenin signaling downstream of endothelin receptor type A (52). CLIP2 is a cytoplasmic linker protein that is amplified in colorectal carcinoma and glioblastoma (53,54), and FRAT2 is an activator of canonical WNT signaling that is up-regulated in gastric cancer (55).

The mir-22 target EVI1 is a zinc finger transcription factor involved in myeloid leukemia and myelodysplastic syndrome (56), consistent with the link between mir-22 and hematopoietic malignancies. Previous studies have detected amplification of the EVI1 locus (3q26.2) in ovarian cancers (57,58), and we observed 22.4-fold up-regulation of EVI1 in our clear cell lines, an expression level possibly exacerbated by loss of mir-22. High EVI1 expression correlates with poor prognosis in patients with advanced stage serous ovarian carcinoma (58). Accordingly, EVI1 overexpression enhanced the proliferation and motility of immortalized OSE cells, with similar effects in ovarian cancer cells (58). These effects may be due, in part, to loss of TGF-β responsiveness, as EVI1 represses TGF-β signaling by inhibiting SMAD family member 3 (59). Furthermore, EVI1 suppresses apoptosis by stimulating PI3K/AKT signaling (60). Thus, as illustrated in Fig. 5, both mir-100 and mir-22 are candidate tumor suppressor miRNAs in clear cell ovarian cancer that influence PI3K/AKT/mTOR signaling through direct and indirect effects at multiple points in the pathway.

Figure 5.

mir-100 and mir-22 target known oncogenes that influence the PI3K/AKT/mTOR pathway. PI3K is comprised of the p85 regulatory subunit (PIK3R1) and the p110 catalytic subunit (PIK3CA). PI3K activation leads to the synthesis of the second messenger PIP3, which recruits AKT to the cell membrane for subsequent phosphorylation and activation. AKT activation has multiple biological consequences including the regulation of cell growth, proliferation, and survival by activating a key downstream effector, the serine/threonine kinase mTOR. The PI3K/AKT/mTOR pathway is frequently activated in clear cell ovarian cancer, usually secondary to gain of function mutations in PIK3CA. We found that mTOR was up-regulated in clear cell ovarian cancer cell lines and repressed by mir-100 on the mRNA and protein levels. mir-100 overexpression also decreased levels of the phosphorylated isoforms of 4EBP1 and p70 S6 kinase, two direct substrates of mTOR kinase activity that mediate mTOR signaling. Two other target oncogenes that were overexpressed in clear cell ovarian cancer also impinge on this pathway. Signaling through FGFRs, such as the mir-100 target FGFR3, induces recruitment and activation of PI3K, and the mir-22 target EVI1 promotes cell survival by stimulating PI3K/AKT signaling. Thus, mir-100 and mir-22 exert tumor suppressor activity through the coordinated repression of multiple components of the PI3K/AKT/mTOR pathway.

Two effectors of miRNA-mediated silencing were also up-regulated in clear cell ovarian cancer, namely the mir-100 responsive target EIF2C2/AGO2 and the mir-22 responsive target EDC3. AGO2 is a core protein of the RNA-induced silencing complex that catalyzes mRNA cleavage and translational repression (61), and previous work has demonstrated amplification of the AGO2 locus in serous ovarian carcinoma (62). EDC3 functions as a scaffold for P-body assembly and enhancer of decapping that promotes mRNA decay (63,64). Except for a 2.1-fold decrease in DICER1 levels, we did not observe significant changes in other components of the miRNA machinery (e.g. RNASEN/DROSHA, DGCR8, TARBP2, and XPO5), making it challenging to predict the overall impact on miRNA function. One study found that low DICER1 expression was an independent predictor of shorter survival in patients with ovarian carcinoma (65). However, whether DICER1 is a robust prognostic factor remains uncertain because other groups have reported a lack of correlation with patient outcome (66).

The mir-182 locus is amplified in ovarian cancer, and overexpression of mir-182 in an ovarian cancer cell line reportedly enhanced tumor growth in vivo (66). Overexpression of mir-182 also promotes melanoma metastasis by directly repressing the transcription factors FOXO3 and MITF (67). Neither of these genes was down-regulated in clear cell ovarian cancer, highlighting the fact that functional miRNA-mRNA interactions are tissue-specific.

Although mir-30 family members are elevated in serous, endometrioid, and clear cell ovarian carcinomas compared with NOSE, it was reported that expression in the clear cell subtype is highest overall (31). Two noteworthy mir-30a-responsive target genes that emerged from our experiments are AVEN and GALNT1. Whereas AVEN is traditionally known as an apoptotic inhibitor, recent evidence indicates that AVEN also impedes cell cycle progression at the G2/M checkpoint by activating ataxia telangiectasia mutated (ATM) in response to DNA damage (68). These data suggest that AVEN has the dual potential to be an oncogene or tumor suppressor, the latter function consistent with its underexpression in clear cell ovarian cancer and up-regulation after mir-30a knockdown. GALNT1 is a glycosyltransferase that initiates O-linked glycosylation in the Golgi apparatus. Single-nucleotide polymorphisms in GALNT1 are associated with the risk of ovarian cancer (69), in agreement with data indicating altered glycosylation of proteins in ovarian cancer (70).

One of our most compelling findings was that reversing the expression of mir-22 (overexpression) or mir-182 (knockdown) in the clear cell cultures caused a global shift in the cancer gene signature such that genes overexpressed in clear cell ovarian cancer (OvCa up) were down-regulated, and genes underexpressed in clear cell ovarian cancer (OvCa down) were up-regulated, regardless of whether the genes were predicted to be direct targets of these miRNAs. Knockdown of mir-30a caused up-regulation of OvCa down genes, but also further up-regulated OvCa up genes, similar to the effects of mir-100 overexpression. Hence, the capacity to drive the gene signature of clear cell ovarian cancer toward a more normal state is not a universal phenomenon but instead a function restricted to specific miRNAs. So far we have tested a handful of the 54 miRNAs that were differentially expressed, and there may be other miRNAs with similar effects. In the future, it will be critical to determine whether combined modulation of mir-22 and mir-182, as well as other miRNAs altered in clear cell ovarian cancer, will be even more effective in reversing the gene signature and phenotype of cancer cells to a more normal state.

In conclusion, our experiments have revealed strong candidate oncogenic and tumor suppressor miRNAs and their target genes in human clear cell ovarian cancer. In particular, mir-22 and mir-182 are influential nodes with global control over the cancer gene signature. These miRNAs may be ideal targets for the development of novel therapies or as sensitizing agents for existing chemotherapy regimens and small molecule drugs. The latter notion is especially tangible considering that several miRNA-regulated genes uncovered in our study are linked to activation of PI3K/AKT/mTOR signaling, a pathway for which multiple small molecule inhibitors are currently available or in clinical trials (71). The novel finding that FGFR3 is overexpressed in clear cell ovarian cancer and is regulated by mir-100 likewise holds promise for combined therapy in light of recent work demonstrating that antibody-based targeting of FGFR3 has potent antitumor activity (45).

Materials and Methods

Primary culture of human NOSE cells

Permission to perform all experiments was obtained from the Institutional Review Board for Baylor College of Medicine and its affiliated institutions (H-15188, H-22124, H-22119). After obtaining informed consent from each study participant, primary cultures of NOSE were grown and expanded as previously described (72). Briefly, brushings of the ovarian surface were performed in pre- and postmenopausal women undergoing clinically indicated surgery. Women with evidence of endometriosis, cancer, or inflammatory conditions at the time of surgery were excluded from participation. Cells were cultured in 1:1 MCDB105/M199 (Sigma Chemical Co., St. Louis, MO) with 10% heat-inactivated fetal bovine serum, 1% penicillin-streptomycin solution (Invitrogen, Carlsbad, CA), and 5 ng/ml epidermal growth factor (Invitrogen). The epithelial origin of cultured cells was confirmed using immunohistochemistry to examine the expression of vimentin and cytokeratins. Only cultures containing more than 90% epithelial cells were used.

Clear cell ovarian cancer cell lines

ES-2^a and TOV-21G^b were obtained from the American Type Culture Collection (Manassas, VA). RMG-I^c (73) and OVMANA^d (74) were obtained from the Health Science Research Resources Bank. The other cell lines were kindly provided as follows: OVISE^d (75), OVAS^d (76), OVTOKO^d (75), and OVSAYO^d (74) from Dr. Hiroshi Minaguchi (Yokohama City University, Yokohama, Japan); KK^d (77,78) from Dr. Yoshihiro Kikuchi (National Defense Medical College, Tokorozawa, Japan); and SMOV-2^d (79) from Dr. Tomohiro Iida (St. Marianna University, Kawasaki, Japan). Cells were cultured in ^aDMEM/F12 (Invitrogen), ^b1:1 MCDB105/M199 (Sigma), ^cHam's F-12 (Invitrogen), or ^dRPMI 1640 (Invitrogen) with 10% heat-inactivated fetal bovine serum and 1% penicillin-streptomycin solution (Invitrogen).

Gene expression profiling and small RNA sequencing

Total RNA was extracted from human NOSE cultures (n = 6) and clear cell ovarian cancer cell lines (n = 10) using the mirVana miRNA Isolation Kit (Applied Biosystems, Foster City, CA). RNA quality and the presence of small RNAs were inspected on a 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). Gene expression profiles were generated on HumanWG-6 version 3.0 BeadChips (Illumina) at Texas Children's Cancer Center Genomics and Proteomics Core Laboratory. The array platform consists of 48,803 transcript probes representing 24,384 unique named genes. Expression data were quantile normalized by applying a monotone linear spline to each array that mapped quantiles 0.01 to 0.99 in increments of 0.01 exactly to the corresponding quantiles of the standard using software kindly provided by Kerby Shedden (University of Michigan). All array datasets described in this study have been deposited into the Gene Expression Omnibus (GEO accession number GSE16574).

Small RNA library construction was performed using the DGE-Small RNA Sample Prep Kit (Illumina) (n = 4 NOSE cultures; n =10 clear cell lines). Purified cDNA was quantified with the Quant-iT PicoGreen dsDNA Kit (Invitrogen) and diluted to 3 pm for sequencing on the Illumina 1G Genome Analyzer (University of Houston). The raw sequence data were passed through a series of quality-control filters to discard reads without a 3′-adapter, reads with copy number less than 4, reads that were less than 10 nucleotides, and reads with more than 10 consecutive identical nucleotides. The remaining usable sequence reads were aligned to human miRNA hairpin precursor sequences in miRBase (release 11.0; http://microrna.sanger.ac.uk/) (80), with up to three mismatches tolerated in an alignment. If a read aligned equally well to multiple precursors, its copy number was equally apportioned. For each mature miRNA sequence, we counted the number of reads that were exact matches, allowing a four-nucleotide extension into the precursor sequence on each side of the mature miRNA. Each miRNA profile was scale normalized using the total number of usable sequence reads after which 40 units were added to each miRNA expression value.

Differentially expressed genes and miRNAs were identified by t test on log-transformed data and fold change (ratio of averages of clear cell ovarian cancer compared with NOSE).

Prediction of miRNA-mRNA functional interactions

miRNAs predicted to target differentially expressed mRNAs were identified using TargetScanHuman (release 5.0; http://www.targetscan.org/) (81), PicTar (http://pictar.mdc-berlin.de/) (82), and miRanda (September 2008; http://www.microrna.org/) (83). For mRNAs that were up-regulated in clear cell ovarian cancer compared with NOSE, we searched for targeting miRNAs that were down-regulated, and vice versa. Retrieval of putative miRNA-mRNA pairs was facilitated by SigTerms software (http://sigterms.sourceforge.net/) (84).

miRNA knockdown and overexpression

Transfection conditions for each clear cell ovarian cancer cell line were optimized for maximum efficiency in preliminary experiments using Dy547-labeled miRIDIAN hairpin inhibitor and mimic transfection controls (Dharmacon, Lafayette, CO). Selection of individual cell lines for knockdown experiments was based on significant up-regulation of each miRNA in those cells compared with NOSE, whereas choice of cell lines for overexpression experiments was based on significant down-regulation of each miRNA. Because the specificity of miRNA silencing is determined by the seed sequence, we examined the selected cell lines for aberrant expression of seed-matched family members that might hamper knockdown or overexpression efficiency, and simultaneously targeted those miRNAs.

ES-2 cells were transfected in six-well plates (2.5 × 10⁵ cells per well) using 10 μl X-tremeGENE siRNA Transfection Reagent (Roche, Indianapolis, IN) and 2 μg hsa-mir-182 hairpin inhibitor (50 nm final concentration) or 2 μg hsa-mir-22 mimic (65 nm) (Dharmacon). OVSAYO cells were transfected in six-well plates (3.0 × 10⁵ cells per well) using 10 μl Lipofectamine 2000 (Invitrogen) and 2 μg hsa-mir-100 mimic (60 nm), or 3 μl Lipofectamine 2000 and 3 μg total hairpin inhibitor (66 nm). Knockdown of hsa-mir-30 family miRNAs in OVSAYO cells was achieved using a mix of 2 μg hsa-mir-30a inhibitor (44 nm) and 1 μg hsa-mir-30d inhibitor (22 nm). Control groups were cells treated with transfection reagent alone (mock transfection), and cells transfected with miRIDIAN hairpin inhibitor negative control no. 1 or mimic negative control no. 2, which are based on Caenorhabditis elegans miRNAs that have minimal sequence identity to human miRNAs (Dharmacon).

Cells were harvested 48 h after transfection, RNA was prepared using the RNeasy Mini Kit with on-column DNase treatment (QIAGEN, Chatsworth, CA), and RNA quality was inspected on a 2100 Bioanalyzer (Agilent). Gene expression profiling was performed as described above (n = 3 for each inhibitor or mimic; n = 3 for each negative control inhibitor or mimic), and the data were evaluated using Spearman's rank sum analysis and GSEA (39). GSEA was executed using public software from the Broad Institute (http://www.broad.mit.edu/gsea/) with the following settings: classic enrichment statistic; permutation by gene set with 1000 permutations; and ratio of classes used for ranking genes. For both Spearman's rank sum analysis and GSEA, the entire set of genes represented in the given functional array dataset were ranked by ratio of experimental/control. When more than one array probe corresponded to a given gene, the probe with the greatest variation between experimental and control groups was used to represent the gene.

QPCR

QPCR validation of microarray data was performed on samples independent of those used in microarray experiments (n = 5 for each transfection condition). Whereas our microarray experiments compared gene expression in cells treated with miRNA inhibitor or mimic relative to negative control, our QPCR experiments measured differences in these groups compared with mock transfected cells. This allowed us to examine for unexpected effects after treatment with negative control mimic, which has a seed sequence (UGUACUA) predicted to target 92 human genes (TargetScan).

Total RNA (500 ng) was reverse transcribed in a 50 μl reaction using 250 U Superscript III reverse transcriptase and random hexamers (Invitrogen). cDNA was diluted 50-fold and 5 μl was used for each QPCR reaction. Custom primers were designed using Primer Express software (Applied Biosystems) and primer sequences are listed in supplemental data (S6). QPCR was performed on an ABI Prism 7500 Sequence Detection System using SYBR Green PCR Master Mix (Applied Biosystems) in a 20 μl reaction and human β-actin (ACTB) as an endogenous control. Reaction conditions were 2 min at 50 C, 10 min at 95 C, 40 cycles of 15 sec at 95 C (denaturation), and 1 min at 60 C (annealing/extension). Each sample was analyzed in triplicate, and a nontemplate control (nuclease-free water) was included for each primer set. The relative quantity of transcript was calculated using the 2^−ΔΔC_T method with the average ΔC_T of the mock transfection group as the calibrator (85). Data were plotted as mean ± sem, and statistical differences were tested using one-way ANOVA followed by Tukey's honestly significant difference test, with P < 0.05 considered statistically significant.

Lentivirus production, transduction, and immunoblot analysis

CMV-TurboRFP-mir100-IRES-puroR vector (Open Biosystems HMR4842–99855762) was packaged and transduced into OVSAYO cells. A similar vector with nontargeting sequence served as the negative control. Briefly, mir-100 or control vector, as well as helper virus DNA plasmids carrying gag, pol, VSVG and tat, were transfected into 293T cells using Lipofectamine 2000 (Invitrogen). Cell medium containing virus was syringe filtered (0.45 μm) 48 h after transfection and added to OVSAYO cells at an appropriate dilution in the presence of polybrene (8 μg/ml). The virus was replaced with fresh medium 6 h after infection. Transduction efficiency was measured by examining the cells under fluorescence microscope. Puromycin (1 μg/ml) was added to generate stable cells.

Ninety-six hours after infection, cells were washed once with cold PBS and collected. Total lysates were generated in sodium dodecyl sulfate lysis buffer (50 mm Tris-HCl, pH 6.8; 2% sodium dodecyl sulfate) by heating at 90–100 C for 5 min and then clearing by centrifugation. Protein (20 μg) was electrophoresed through a NuPage 3–8% Tris-acetate gel (Invitrogen) under reducing conditions, transferred to a polyvinylidene difluoride membrane, and incubated with antibodies against mTOR (no. 2972; Cell Signaling Technology, Danvers, MA), phospho-4EBP1 (Thr37/46; Cell Signaling 2855), phospho-p70S6K (Ser371; Cell Signaling 9208), or β-actin (Sigma A5441). Immune complexes were detected with horseradish peroxidase-conjugated secondary antibodies and the SuperSignal West Pico Chemiluminescent Substrate Kit (Pierce Biotechnology, Rockford, IL).

Cell viability assay

The CellTiter-Glo Luminescent Cell Viability Assay (Promega Corp., Madison, WI) was used to analyze the effect of the mTOR inhibitor and rapamycin analog RAD001 (everolimus; Sigma 07741) on cell viability according to the manufacturer's protocol. OVSAYO cells stably transduced with mir-100 or control vector were cultured overnight in 96-well plates (5 × 10³ cells per well). Cell viability was measured after the addition of RAD001 (in the presence of 4% fetal bovine serum) at the indicated concentrations for 72 h (n = 3 wells per concentration). The medium containing RAD001 was changed every 24 h. For each RAD001 concentration, cell viability was expressed as a fraction of the untreated mir-100 or control group. Data were plotted as mean ± sem and statistical differences were tested using Student's t test, with P < 0.05 considered statistically significant. Note that for all data points the sem was too small to be visible on the graph.