Integrated analysis of cancer stem cells-associated lncRNA-miRNA-mRNA network for ovarian cancer via microarray and Gene Expression Omnibus database

Background: Cancer stem cells (CSCs) are associated with the recurrence, metastasis and chemoresistance of epithelial ovarian cancer. Competing endogenous RNAs (CeRNAs) play an important role in maintenance of ovarian cancer stem cell-like cells (OCSCs) characteristics. To construct a ceRNA regulatory network for OCSCs, microarray technology and Gene Expression Omnibus (GEO) database had been used. Human serous epithelial ovarian carcinoma cell line COC1 cells were treated with cisplatin and paclitaxel then maintained in stem cell conditions for 6 days to obtain CD117+/CD133+ cells (OCSCs). We identi�ed the differentially expressed miRNAs (DEMs), lncRNA (DELs) and mRNA (DEGs) between OCSCs and COC1 by microarray and combined them with representative microarray pro�les in GEO Database. Results: According to the combination, 28 DEMs were identi�ed at �rst, and 452 DEGs were obtained combining with the predicted targets of these miRNAs and our mRNA microarray results. Up-regulated DEGs of them were signi�cantly enriched in ‘p53 signaling pathway’, ‘FoxO signaling pathway’ and ‘MicroRNAs in cancer’, whereas down-regulated DEGs were signi�cantly enriched in ‘Adherens junction’ and ‘Hepatitis C’ pathway. 29 transcripts of 17 lncRNAs should be the ceRNAs of 10 of these miRNAs according to bioinformatics predicted results and lncRNA microarray. Finally, we obtained ceRNA network with 10 DEMs, 21 DEGs, and 25 transcripts of 13 DELs which should play an important role in maintenance of OCSCs characteristics. LINC00665-miR-146a-5p-NRP2 should be one of ceRNA pathways of the network. The qPCR results indicated that the expression of miR-146a-5p in OCSCs was lower than that in COC1, and LINC00665 shows the opposite trend. These results were consistent with the results of microarray partially. When LINC00665 expression was up-regulated in COC1, the cell proliferation ability


Background
Epithelial ovarian cancer is the most lethal gynecologic malignancy GD Aletti, MM Gallenberg, WA Cliby, A Jatoi and LC Hartmann [2], and it's the fourth most common cause of death from cancer in 40-59 years old women [59].According to statistics, it's predicted that the increase of incidence should be 22530 in the US in 2019, which accounts for 2.5% of all estimated new tumor cases in women, and 13980 of deaths should occur, accounting for almost 5% of the total of cancer deaths in women [60].The most common therapy is the optimal cytoreductive surgery combined with 6-8 cycles of platinum-based chemotherapy.However, after optimal cytoreductive surgery performance, many patients emerged secondary chemoresistance within one year with chemotherapy, resulting in a recurrence rate of serous ovarian cancer of up to 80% [30].
It is reported that many mechanisms play important role in regulation of chemoresistance in ovarian cancer chemotherapy, such as enhancement of repair of DNA damage, reduction of drug accumulation in cells, loss of drug function and inhibition of apoptosis.While cancer stem cells (CSCs) theory gives a new understanding in the development of drug-resistant tumors [12].CSCs are one unique subpopulation of cells with stem cell proprieties in tumors.This kind of cells remain in a dormant and quiescent state, therefore, exposure to a drug that targets divising cells at DNA replication and mitosis could not reduce the number of CSCs e ciently [42,64].On the contrary, chemotherapy drugs may increase the ratio of CSCs in tumor tissues because of its inhibitory effect on differentiated tumor cells [30].And the expression of drug membrane transporters such as ABCG2 make it di cult for chemotherapy drugs to accumulate in CSCs [40].CSCs also can expand the pool of stem cells through symmetric division, and differentiate cancer cell causing tumor recurrence after treatment through asymmetric division, during which stem cells give rise to daughter cells of different fates, proliferative potential, size, or other characteristics [43].Sometimes these two kinds of division can be transformed into each other.Some studies have suggested that asymmetric division was associated with CSC in early and highly differentiated cancers, while inhibition of it was associated with advanced and highly proliferative cancers [4].In summary, CSCs are a kind of 'self-renewing cells within a tumor that can cause the heterogeneous lineages of cancer cells that comprise the tumor' [8].The existence of CSCs means that there is a small population of cells with stem cell properties proliferating slowly and carrying primary resistance to chemotherapy in tumor tissues and these cells can be selected during the period of chemotherapy and pass the drug resistant property to their daughter cells [53].It is also one of the important reasons for ovarian cancer recurrence after chemotherapy.Therefore, CSCs theory, as a way to result in CSCs-dependent cancer recurrence during the treatment of ovarian cancer, is worthy of further study.
There are many methods for CSCs isolation, such as identifying SP cells from ovarian cancer cells by e ux of Hoechst 33342 dye which could be blocked by exposure to verapamil [62], or isolating CSCs by detecting the transmembrane pluripotent marker protein such as CD44+, CD133+ [3,31] and CD117+ [71] at the use of ow cytometry.CD133 has been used as a means of detecting CSCs in many malignancies, including epithelial ovarian cancer (EOC) [54].In ovarian cancer, CD133 + cancer cells have stronger tumorigenicity than CD133cancer cells and are more likely to develop chemoresistance [29].While CD117, which is also known as c-Kit, is commonly used to identify CSCs among EOC cells [70].CD117 is tyrosine kinase type receptor, which is involved in cell differentiation, proliferation, apoptosis, migration and other biological processes.It has been suggested that human ovarian cancer cells with CD 117 + phenotype had properties similar to those of CSCs, such as self-renewing, differentiation, high tumorigenicity and chemoresistance [39].In some studies, CSCs were obtained by recognizing CD117+/CD133 + cells.The CD117+/CD133 + CSCs subpopulation of HO8910, which was the human EOC cell line, were identi ed to be more likely to develop resistance to conventional chemotherapy agents including cisplatin, doxorubicin, and mitoxantrone[68].Li Ma, et al had obtained CD117+/CD133 + cells identi ed stem cell properties with cisplatin and paclitaxel selection and under the stem cell culture condition from serous ovarian cancer cell line SKOV3 [40].According to their protocol, we selected CD117+/CD133 + cells from serous ovarian cancer cell line COC1 for our further study.
Competitive endogenous RNA (ceRNA) theory indicates that speci c RNAs can isolate miRNAs to bind to target genes, impair miRNAs activity, and thus up-regulate miRNA target genes expression.In addition to being involved in the occurrence and development of tumors, this regulation mechanism plays an important role in the stem characteristics maintenance of CSCs.For example, miR-193a targeted E2F6 and c-Kit, and there was an estrogen-mediated E2F6 ceRNA network which epigenetically and competitively inhibited miR-193a activity, up-regulated OCSCs marker c-Kit expression, and promoted ovarian cancer stemness and tumorgenesis [6].The latest study also found that as an oncogenic lncRNA in cervical cancer, LINC00337 affected the expression of KLF5 and maintained the CSC-like properties by down-regulating miR-145 [23].At present, there are many studies on the involvement of lncRNAs as ceRNAs in the occurrence and development of ovarian cancer, but fewer in the cancer stem cell characteristics regulation of ovarian cancer.There even has been no report on the constructed lncRNA-miRNA-mRNA interaction network that were associated with stemness and progression of ovarian cancer.
Herein, we analyzed the differentially expressed miRNAs, lncRNAs and mRNAs by microarray, and obtained the key differentially expressed miRNAs (DEMs) by combining with representative microarray pro les in GEO Database.The predicted target genes of these DEMs (identi ed by PicTar [5], Targetscan [1], DIANA [55] and starbase [32]) and the differentially expressed mRNAs (DEGs) of our microarray and GEO database were combined to identify key DEGs that were differentially expressed and had the same trend in all these datasets.Protein-protein interactions (PPI) of these DEGs were constructed by STRING [61] and Cytoscape[58] to nd hub genes in OCSCs.Differentially expressed lncRNAs (DELs) that may interact with these DEMs were obtained by combining the predicted results of lncBase [47] and starbase with our lncRNA microarray results.Finally, we obtained the ceRNAs network constructed by miRNA, lncRNA and mRNA in OCSCs.MiR-146a-5p and LINC00665 of this network were selected for partial validation.Through bioinformatics methods, we constructed ceRNAs network that might be involved in the stem characteristics maintenance of OCSCs.The study of OCSCs can help to reveal the CSCs related way in which ovarian cancer develops chemoresistance and provide a theoretical basis for the research on reducing or reversing chemoresistance during the treatment of ovarian cancer.

Preparation of CD117+/CD133 + ovarian cancer cells and detection of stem cell characteristics
Cells treated with cisplatin (40 mol/L) and paclitaxel (10 mol/L) were cultured under stem cell conditions for 6 days to form cell spheres.Western blot analysis of sphere cells showed that the expression of SOX2, OCT4 and NANOG were higher than those of COC1, and the expression of ALDH1 and LGR5, which were considered to be OCSCs-speci c stem cell markers, were also increased.Flow cytometry detection showed that the percentage of CD117+/CD133 + cells in sphere was signi cantly higher (78.49%) than that in COC1 (27.78%).Cell cycle detection showed that the percentage of M stage of cells was signi cantly higher (84.9-95.6%)than that in COC1 (3.2%).Therefore, CD117+/CD133 + cells were considered to be ovarian cancer stem cell-like cells (OCSCs).(Fig. 1)

MiRNA-target regulatory network was established by combining miRNA microarray and GEO datasets results
The results of miRNA microarray were combined with GSE107155-SKOV3 and GSE107155-Kuramochi to nd common DEMs.28 DEMs were identi ed.In our miRNA microarray results, 15 of them up-regulated and 13 down-regulated in OCSCs.In GSE107155-SKOV3 dataset, 24 of them showed increased expression, while only 4 DEMs showed decreased expression.While the results of GSE107155-Kuramochi dataset suggested that all of these 28 DEMs were up-regulated.According to our miRNA microarray results, miR-1268a was the most signi cantly up-regulated miRNA and miR-324-5p was the most signi cantly down-regulated one.MiR-630 was the most signi cantly upregulated miRNA in GSE107155-SKOV3 and -Kuramochi datasets.MiR-221-3p was the most signi cantly downregulated one in GSE107155 SKOV3 dataset.(Additional le 1) After analysis by PicTar, Targetscan, DIANA and Starbase, potential target genes of these DEMs were combined with the mRNA microarray results to nd common DEGs.Finally, 285 up-regulated and 167 down-regulated DEGs were identi ed.(Table 2) These DEGs were the predicted target genes for 24 of the 28 DEMs.Using STRING and cytoHubba, we had obtained the top 10 key genes of these DEGs, ranked by Maximal Clique Centrality (MCC) degree.(Table 3) According to MCODE analysis, there were 16 clusters, among which the most signi cant cluster included 11 nodes and 55 edges.The top 10 key genes were all included in this cluster.UNKL, GNAQ, NMD3, CPD, YPEL5, GAPVD1, VIM, EIF4EBP2, RND3, CALB2, STX16, NHLRC3, TNFRSF21 and OAS2 were seed genes of these clusters.(Fig. 3) LncBase and Starbase were applied to predict the lncRNA transcripts that could bind to the 24 DEMs mentioned above.Combined with our lncRNA Microarray result, we found that 29 transcripts of 17 lncRNA could bind to 14 of the 24 DEMs.(Additional le 2) 22 transcripts were up-regulated and 7 transcripts were down-regulated in OCSCs compared with common ovarian cancer cells according to our lncRNA microarray analysis.

Functional enrichment analysis, establishment of protein-protein interactions (PPI) network and and identi cation of hub genes in OCSCs
Microarray analysis of DEGs was combined with GSE80373 and GSE28799 datasets.As a result, 105 differentially expressed genes were identi ed in all the three datasets, including 72 up-regulated genes and 33 down-regulated genes in OCSCs compared with ovarian cancer cells.(Fig. 4) GO functional enrichment analysis on these genes indicated that these DEGs were signi cantly enriched in 'negative regulation of cell proliferation' (BP, GO:0008285), 'cytosol' (CC, GO:0005829) and 'ferroxidase activity' (MF, GO:0004322).72 up-regulated genes were signi cantly enriched in 'doxorubicin metabolic process' (BP, GO:0044597), 'extracellular exosome' (CC, GO:0070062) and 'geranylgeranyl reductase activity' (MF, GO:0045550).33 down-regulated genes were signi cantly enriched in 'intracellular ribonucleoprotein complex' (CC, GO:0030529).Although they were enriched in 'positive regulation of cell growth' (BP, GO:0030307) and 'poly(A) RNA binding' (MF, GO:0044822), but the enrichments were not signi cant (P > 0.05).(Fig. 4) KEGG pathway analysis indicated that these genes were signi cantly enriched in 'FoxO signaling pathway'.72 up-regulated genes were signi cantly enriched in 'FoxO signaling pathway', 'Pentose and glucuronate interconversions', 'Mineral absorption' and 'Glutathione metabolism'.There were 2 KEGG pathways which the 33 down-expression genes were enriched in, such as 'RNA transport' and 'Ras signaling pathway', while the P value of the latter one is > 0.05.This might be due to the relatively small number of down-regulated genes.(Fig. 4) Using STRING and cytoHubba, we had obtained the top 10 key genes of these DEGs, ranked by MCC degree.They were CAT, EGFR, GSR, EEF1A1, IDH1, CLU, TGFB1, AKR1B1, MT1X, MT1E.(Table 4, Fig. 4) According to MCODE analysis, there were 3 clusters,including 3 nodes and 3 edges, 10 nodes and 13 edges and 9 nodes and 11edges respectively.MT1E and FBXO32 were seed genes of these clusters.The GSR, EEF1A1, IDH1, CLU, TGFB1, AKR1B1, MT1X and MT1E of top 10 key genes were included in these clusters.(Fig. 4) up-regulated and 10 down-regulated DEGs.Finally, these DEGs were predicted as target genes for 10 of the 24 DEMs.Among them, the up-regulated miRNA were miR-1287-5p, miR-193b-3p, miR-423-5p and miR-374b-5p, and the down-regulated miRNA were miR-425-5p, miR-96-5p, miR-26a-5p, miR-30e-5p, miR-183-5p and miR-146a-5p.A total of 25 transcripts of 13 lncRNA were predicted as the ceRNA of these miRNAs, including 21 up-regulated transcripts and 4 down-regulated ones.(Table 5, Additional le 3) Fig. 5 showed the obtained ceRNA regulatory network.Through Kaplan-Meier (Km) curve analysis, there were 5 miRNAs, 12 mRNAs and 5 lncRNAs of which the differential expressions were signi cantly related to survival probability in ovarian cancer.Among these genes, the most signi cantly related lncRNA was LINC00665, of which the expression was up-regulated in OCSCs.According to the ceRNA network, one of LINC00665 transcripts was considered to be the ceRNA of miR-146a-5p.In this study, its expression showed a downward trend in OCSCs.NRP2 was one of its predicted target genes, which was the most signi cantly up-regulated mRNA of in the OCSCs ceRNA network.Km curve analysis showed that ovarian cancer patients with reduced mir-146a-5p expression had shorter survival time, while patients with reduced LINC00665 and NRP2 expression tended to have longer survival time.QPCR con rmed the microarray results that miR-146a-5p expression in OCSCs was lower than that in COC1, while LINC00665 expression was higher than that in COC1.(P < 0.05) LINC00665 was further overexpressed in COC1, and it was found that the proliferation ability of LINC00665 + COC1 was increased, apoptosis was decreased, and percentage of G2/M cells was increased.(Fig. 6, P < 0.05) In future studies, we need to further clarify the roles of LINC00665, miR-146a-5p and NRP2 in OCSCs and the regulatory relationships among them.(Fig. 6 and Additional le 4) The procedure of our study was shown in Fig. 7.

Discussion
Failure of ovarian cancer treatment was often associated with the development of chemoresistance, including refractoriness (inherent resistance) and resistance (recurrence occurred within 6 months after response to complete treatment) [44], which was more common in ovarian cancer chemotherapy.It is of great signi cance to study OCSCs to clarify the formation and reversal of chemoresistance in ovarian cancer.Therefore, the study on the stem characteristic maintenance of OCSCs is expected to partially reveal the causes of ovarian cancer drug resistance and help nd a way to inhibit or reverse the chemoresistance of ovarian cancer through the OCSCs pathway.The application of microarray technology and GEO database helps us to screen differential genes in OCSCs and to further nd important genes related to maintain stem characteristics, so as to identify important regulatory genes that may in uence drug resistance in ovarian cancer.
In order to construct important ceRNA regulatory networks, CD117+/CD133 + cells were rst screened in the sphere from ovarian cancer cells COC1.The elevation of stem cell markers revealed that this group of cells had maintained stem cell conditions.The DEMs, DELs and DEGs between CD117+/CD133 + cells (OCSCs) and COC1 were further identi ed by microarray technology.Combining the microarray results with GEO datasets, we found that the target DEGs of DEMs of OCSCs were enriched in cancer related pathway, such as 'Ras signaling pathway', 'p53 signaling pathway', 'AMPK signaling pathway' and 'FoxO signaling pathway'.It was suggested that these DEMs were involved in the regulation of the occurrence and development of OCSCs.While the DEGs of OCSCs were enriched in cancer related pathway, such as 'FoxO signaling pathway', 'RNA transport', 'Pentose and glucuronate interconversions', 'Mineral absorption' and 'Glutathione metabolism'.It was also suggested that these DEGs were involved in the regulation of the occurrence and development of OCSCs by metabolic pathways.Combining with these two groups of DEGs had contributed to narrow the range of important mRNAs that regulate the characteristics of OCSCs.Meanwhile, lncRNAs that might be ceRNAs of these DEMs were found by bioinformatics analysis combined with lncRNA microarray result.Finally, 10 key (4 up-regulated and 6 down-regulated miRNAs) miRNAs and 21 predicted target mRNAs (11 up-regulated and 10 downregulated mRNAs) were identi ed. 25 transcripts of 13 lncRNAs (21 up-regulated and 4 down-regulated transcripts) should be ceRNAs of these miRNAs.According to Km curve analysis, we found that the 5 DEMs, 12 DEGs and 5 DELs were signi cantly related to survival probability in ovarian cancer.
We selected the lncRNA with the highest correlation with survival probability, LINC00665, and its related miRNA,miR-146a-5p for preliminary study.QPCR con rmed microarray results that the expression of LINC00665 was increased and miR-146a-5p was decreased in OCSCs compared with COC1.Further regulation of the expression of LINC00665 indicated that LINC00665 was associated with cell proliferation and apoptosis in ovarian cancer cells.The target genes of MiR-146a-5p in our ceRNA network included NRP2, CASK and HNRNPD, and the differential expression of these target genes was statistically signi cant with survival probability.However, the expression trend of HNRNPD was the same as that of miR-146a-5p, while among the three genes, the expression difference of NRP2 mRNA was the largest in our microarray result.Therefore, we focused on analyzing the possible roles of miR-146a-5p, LINC00665 and NRP2 in the regulation of OCSCs stem cell characteristics in the following sections.
MiR-146a-5p is one of the anti-in ammatory miRNAs, which is considered to be a therapeutic target for in ammation-related diseases such as perinatal cardiomyopathy [22] and obesity [51].It acts as a single miRNA regulatory factor inhibiting the function of regulatory T (Treg) cells, maintaining immune homeostasis of Treg cells, and its absence leads to IFNγ-dependent immune-mediated fatal lesions or tumors in a variety of organs [38,41].Its rs2910164 SNP genotype may also in uence the age of onset of cancer and is associated with early onset of cancer [48].The abnormal expression of miRNA-146a in CSCs has been found in many studies.And its effects in different tumors might be completely opposite [56].As an exosomal miRNA in colorectal cancer, miR-146a-5p could target NUMB to activate wnt signaling pathway to generate and maintain cancer stem cells characteristics for promotion of tumor formation [7,26].While in cervical cancer and breast cancer CSCs, miR-146a-5p played a completely opposite role [14,34].It has been indicated that the expression of miR-146a-5p in tumors was also related to chemoresistance and metastasis of tumor cells [56].MiR-146a-5p could enhance the induction effect of platinum on apoptosis so as to prohibit cell proliferation in EOC cells [11,33].Patients with lower expression of miR-146a-5p had reduced survival probability, shortened progression-free survival (PFS), and were prone to platinum chemoresistance [66].At present, there is no study on miR-146a-5p in OCSCs, but the miR-146a-5p derived from mesenchymal stem cells exosome has been reported to be able to enhance the sensitivity of ovarian cancer cells to docetaxel and taxane [50].According to our study, miR-146a-5p might play an inhibitory role in OCSCs.
In our study, it had been found that a transcript of LINC00665 could be a ceRNA to participate in the regulation of miR-146a-5p on target genes.Studies on LINC00665 have been increasing in recent years.It has been considered to play an important regulatory role in various tumors.In 2018, Dong-Yue Wen et al.
found an abnormally elevated LINC00665 expressed in hepatocellular carcinoma (HCC) data in the Cancer Genome Atlas (TCGA) database, which could interact with the cell cycle regulatory proteins to promote the development and progression of HCC [65].Following studies suggested that LINC00665, as the ceRNA of multiple miRNAs, could play an important role in the regulation of cell biological behaviors and chemoresistance in various malignant tumors [10,37,49,57,69].In high-grade serous ovarian cancer, a ceRNA regulatory network correlated with lymphocyte in ltration had been established by using TCGA, GTEx and GEO datasets, and it was suggested that LINC00665 was positively correlated with lymphocyte in ltration [67].LINC00665 also plays a role in tumor regulation by maintaining protein stability.For example, by activating and maintaining PKR stability, LINC00665 could promote hepatocellular carcinoma progression [13].It also could promote lung adenocarcinoma angiogenesis by stabilizing YB-1 protein [9].Encoding small peptides is another way in which LINC00665 participates in the regulation of tumor progression [21].LINC00665 could also participate in the regulation of malignant biological behaviors of tumor cells by STAU1-mediated mRNA Degradation (SMD), which was an important post-transcriptional regulation of mRNA stability [52].In conclusion, LINC00665 has a clear regulatory role in many tumors, but there has been no study on its in uence on the stem cell characteristics of CSCs, including OCSCs.In addition, we found that not all transcripts of LINC00665 showed an upward trend in the constructed ceRNA regulatory network.Therefore, the function of LINC00665 is worthy of further study.
NRP2 protein, a member of the NRPs family, is related to the development of the nervous system.It has the function of affecting signal receptors as a co-receptor.It is a transmembrane glycoprotein, which contains four extracellular domains capable of binding to ligands and one short cytoplasmic domain.
Many microarray analyses revealed that NRP2 might be a hub gene for the development of a variety of tumors [36,72,73].Its function was either dependent or independent of the vascular endothelial growth factor receptor (VEGFR).Vascular endothelial growth factors (VEGFs) have the function of receptor tyrosine kinases activity to promote angiogenesis and increase vascular permeability by binding with VEGFR, and are involved in maintaining the stem cell characteristics of CSCs in many tumors [28,35].
While the co-receptor NRP2 can form a complex with VEGFR to enhance its a nity with VEGF and enhance tumor-promoting effect of VEGF [46].In addition, NRP2 can bind to integrins to regulate CSCs generation and stem characteristics maintenance through VEGF/NRP2 signaling [17,18,20].At present, most of anti-VEGF drugs (such as bevacizumab), which only block the binding of VEGF and VEGFR but not NRP2, cannot inhibit CSCs effectively, but promote the enrichment [16].Therefore, in addition to anti-VEGF therapy, combining anti-VEGF/NRP2 therapy may enhance the bene ts of anti-VEGF therapy alone [19].NRP2 can also act without VEGF signaling in CSCs.It has been indicated that NRP2 could bind to PDZ protein through PDZ domain in its cytoplasmic part, participating in receptor transport and signal transduction in tumor development regulation M Katoh [27].Other studies have shown that NRP2b, an alternative splicing isomer of NRP2, could bind to PTEN to increase the resistance of non-small-cell lung cancer (NSCLC) cells to EGFR inhibitor ge tinib [15].Although NRP2 wasn't the hub genes of PPI networks according to our analysis, but the up-regulated trend was the most obvious among mRNA in the ceRNA network in OCSCs.Km curve suggested that the survival probability of patients with high NRP2 expression decreased faster than that of patients with low expression.All the above suggested the importance of NRP2 in tumor development.According to our study, NRP2 might play an important role in OCSCs.Further study of the ceRNA mechanism regulating NRP2 expression in OCSCs can better reveal the mechanism of stem cell characteristics maintenance in OCSCs.
CeRNA is an important post-transcriptional regulation mode of genes.At present, studies on ceRNA regulation in various CSCs have been carried out extensively.However, as one of cancer stem cell-related tumors, the establishment of ceRNA regulatory network in ovarian cancer hasn't been reported yet.
According to our study, a ceRNA regulatory network of OCSCs has been initially established by microarray analysis and bioinformatics research, but further clinical and molecular biology studies are required.In the ceRNA network constructed, it was indicated that the protein encoding genes had less direct interaction with each other according to PPI constructed by STRING and Cytoscape.However, through the construction of ceRNA network, these protein encoding genes with no direct interactions were connected indirectly by lncRNA-miRNA-mRNA network.Our further studies are expected to nd out the important sites and regulatory ways to maintain the characteristics of stem cells in OCSCs, providing new theoretical basis and experimental guarantee for ovarian cancer research.

Conclusion
The ceRNA network we constructed may be involved in the stem cell characteristics maintenance of OCSCs and provide directions for further OCSCs research in the future, so as to assist the development and treatment of ovarian cancer.

Cell culture and drug screening
Human serous epithelial ovarian carcinoma cell line COC1 was bought from China Center for Type Culture Collection (CCTCC).The cells were seeded in 6-well plate and cultured in the 37 ℃, 5%CO2 incubator, feeding with RPMI-1640 medium (Gibco, 31800-014, US) supplement with 10% fetal bovine serum (Hyclone, SH30084.03,US).Centrifuged and then changed the medium after 24 h culture.The cells were feeded with the same medium above until they grew to 80% con uence.After treated with 40 µmol/L cisplatin and 10 µmol/L paclitaxel, the cells were cultured for another 5 days [63].Then passed the cells after centrifuging and maintained them under stem cell conditions, which was composed of RPMI-1640 medium (Gibco, 31800-014, US), 5 µg/ml recombinant human insulin (Solarbio, 11061-68-0, China), 10 ng/ml recombinant human epithelial growth factor (EGF, Sino Biological Inc, 10605-HNAE, China), 10 ng/ml basic broblast growth factor (bFGF, Sino Biological Inc, 10014-HNAE, China) and 12 ng/ml leukemia inhibitory factor (LIF, Cloud-Clone Corp, RPA085Hu01, US) [40].Changed the medium every other day by centrifuging at 1000 rpm for 5 min, and then collected the cells after 6 days for further gene expression and protein detection.

Sphere formation assay
The oating sphere cells were dissociated by incubation with 0.02% trypsin-EDTA for 1-2 min at 378C and 100 cells per well were plated in 96-well culture dishes in 200 ml of growth medium; 25 ml of medium per well was added every 2 days.The number of dissociated spherical cells for each well was evaluated after 7-day culture.

Flow cytometric analysis
The treated cells were centrifuged at 1000 rpm for 5 min and then yielded.Wash the cells two times by phosphate buffered solution (PBS, Doublehelix, P10033, China) and collected the cells 1000rmp at 1000 rpm for 5 min.After centrifugation, the supernatant was discarded.The grouped 1 × 10 6 cells were resuspended with 100 µl PBS supplement with anti-CD133 (eBioscience, 12-1339-41, US), anti-CD117 (eBioscience, 11-1178-41, US) and control antibody, respectively.The percentage of CD133+/CD117 + cells was performed by ow cytometric analysis after incubated in dark.
4. RNA extraction and real-time quantitative PCR (qPCR) Total RNA was extracted according to the protocol of total RNA kit (BioTeke, RP1201, China).The concentration of the RNA extracted was calculated, then 1 µg of total RNA was added into 19 µl reverse transcription mixture for reverse transcription polymerase chain reaction (RT-PCR) according to the protocol of PrimeScript™ RT reagent Kit (Perfect Real Time) (Takara, RR037Q, China).The gene expression was detected according to the protocol of SYBR Green (Solarbio, SY1020, China) method by using the ExicyclerTM 96 uorescent quantitative machine (BIONEER, Korea).The primers sequences are showed in representative microarray pro les in GEO Database (GSE107155-SKOV3 and -Kuramochi).PicTar, Targetscan, DIANA and Starbase were used to predict the target genes of these DEMs.The results of each group were analyzed to nd the common genes as the DEMs prediction target genes dataset in OCSCs.
The obtained target genes were further combined with our microarray results to identify a set of miRNA target DEGs.At the same time, we identifed the common DEGs by combining our microarray pro les with GEO Database (GSE80373 and GSE28799).The obtained DEGs were further combined with the miRNA target DEGs to identify common DEGs in both sets.Differentially expressed lncRNAs (DELs) that may interact with DEMs were obtained by lncBase [47] and starbase, and combined with our lncRNA microarray results.
GO and KEGG analyses were performed for each group of DEGs [24,25].GO includes three parts: Biological Process (BP), Cellular Component (CC) and Molecular Function (MF).The number of DEGs included in each GO entry was counted, and the signi cance of enrichment of DEGs in each GO entry was calculated by statistical test.P < 0.05 indicated that the DEGs were enriched in this GO entry.Pathway analysis of DEGs was carried out using KEGG database and statistical test was used to calculate the signi cance of enrichment of DEGs in each pathway entry.P < 0.05 indicated that the DEGs were enriched in this pathway.The relationship between the expression and survival probability of key genes was analyzed by Km Curve [45].Finally, we analysed the ceRNA network constructed by miRNA, lncRNA and mRNA in OCSCs by STRING and Cytoscape.

Statistical analysis
Data analysis was performed using SPSS 16.0 (IBM) statistical software.Numerical data were presented as mean ± standard deviation (SD), and categorical data were presented as percentage.To compare the The     The nal ceRNA regulatory network in OCSCs.
sorting of ovarian cancer stem cell-like cells (CSCs) and the detection results of stem cell markers: (A) sphere formation in COC1, (B) detection of differences in expression of stem cell markers in COC1 after treatment by Western blot.Detection of the percentage of CD117+/CD133+ cells in COC1 after treatment by ow cytometry: (C) a, COC1, b, CD117+/CD133+ cells; (D) a, COC1, b, CD117+ cells, c, CD133+ cells.

Figure 1 The
Figure 1

Table 2
Obtained potential target DEGs by combining mRNA microarray results with PicTar, Targetscan, DIANA and Starbase predicted results of targets of the 28 DEMs (Only the top 25 DEGs with the greatest expression difference were listed in each group.All the DEGs listed in the table showed signi cant differences, P < 0.05.And the fold changes of each gene was more than 2.) GO functional enrichment analysis was performed on these 452 DEGs.It was indicated that these genes were signi cantly enriched in 'regulation of GTPase activity' (BP, GO: 0043087), 'cytosol' (CC, GO: 0005829) and 'protein binding' (MF, GO: 0005515).285up-regulatedgenesweresigni cantly enriched in 'cell migration' (BP, GO: 0016477), 'nucleus' (CC, GO: 0005634) and 'ATP binding' (MF, GO: 0005524), et al. 167 down-regulated genes were signi cantly enriched in 'positive regulation of transcription from RNA polymerase II promoter' (BP, GO: 0045944), 'nucleoplasm' (CC, GO: 0005654) and 'protein binding' (MF, GO: 0005515), et al. (Fig.2) KEGG pathway analysis found that these genes were signi cantly enriched in 'Salivary secretion', 'Ras signaling pathway', 'p53 signaling pathway' and 'AMPK signaling pathway', et al.The tumor-related pathways in which 285 up-regulated genes were signi cantly enriched included 'p53 signaling pathway', 'FoxO signaling pathway' and 'MicroRNAs in cancer', et al.There were 2 KEGG pathways which the 176 down-expression genes signi cantly riched in, such as 'Adherens junction' and 'Hepatitis C', but for the other pathways, 'Notch signaling pathway' and 'Endometrial cancer', the P values were > 0.05.(Fig.2)

Table 3
The top 10 key genes of PPI network of the target DEGs, ranked by Maximal Clique Centrality (MCC) degree 3. MiRNA-lncRNA regulatory network was established by combining miRNA, lncRNA microarray, lncBase and starbase results

Table 4
The top 10 key genes of the PPI network of the 105 DEGs, ranked by Maximal Clique Centrality (MCC) degree 5. Construction of ceRNA regulatory network of OCSCs21 DEGs were identi ed after combination of the 452 DEMs target genes and the 105 DEGs, including 11

Table 5
MiRNAs, mRNAs and lncRNAs of the ceRNA network of OCSCs

Table 1 .
The reaction cycle is: 94 ℃, 10 min, 94 ℃, 10 s, 60 ℃, 20 s, 72 ℃, 30 s, repeated 40 cycles, 72 ℃, 2 min 30 s, 40 ℃, 5 min 30 s, 60 ℃ to 94 ℃, 1.0 ℃ per 1 s, 25 ℃, 1 min.The data were analyzed by 2 −ΔΔCT method and the CT (threshold cycle) value means the number of cycles after which a signal of each sample could be detected.And β-actin was used as internal reference control for normalization.China) was used to lyse COC1 and spheroid cells at 4 ℃ for 5 min.The debris was removed by centrifugation at 12000 rpm, 4 ℃ for 10 min and the supernatant protein extract was analyzed for western blot.The quantitative determination of protein was made according to the manufacturer's instructions (BIOTEK, ELX-800, US).The protein sample diluted by 5 × Loading Buffer and PBS was boiled in boiling water for 5 min.The volume of each sample was 20 µl containing 40 µg protein.SDS-polyacrylamide gel electrophoresis (SDS-PAGE) was performed after protein loaded at 80 V for 2.5 h.Then the protein was transferred to a polyvinylidene uoride (PVDF, Millipore, IPVH00010, US) membrane by electroblotting.The transferred PVDF was put into the 5% (w/v) milk diluted by TBST buffer supplemented with 0.005% Tween 20 for 1 h to block nonspeci c reaction.The membrane was incubated with rabbit anti-human OCT4 antibody (1:500, Abcam, ab18976, UK), rabbit anti-human SOX2 antibody (1:500, Abcam, ab97959, UK), rabbit anti-human NANOG antibody (1:500, Abcam, ab80892, UK), Design ID:076500) was used in this study for analysis of DGLs and DEGs.Experiments were performed according to the instruction of the manufacturer.The criteria for differentially expressed miRNA is |Score (d)| ≥2.We iden ed the DEMs by combining with