Control of PD-L1 expression by miR-140/142/340/383 and oncogenic activation of the OCT4–miR-18a pathway in cervical cancer

PD-L1, a key inhibitory immune receptor, has crucial functions in cancer immune evasion, but whether PD-L1 promotes the malignant properties of cervical cancer (CC) cells and the mechanism by which PD-L1 is regulated in CC remains unclear. We report that PD-L1 is overexpressed in CC, and shRNA-mediated PD-L1 depletion suppresses the proliferation, invasion, and tumorigenesis of CC cells. Loss of miR-140/142/340/383 contributes to PD-L1 upregulation. miR-18a enhances PD-L1 levels by targeting PTEN, WNK2 (ERK1/2 pathway inhibitor), and SOX6 (Wnt/β-catenin pathway inhibitor and p53 pathway activator) to activate the PI3K/AKT, MEK/ERK, and Wnt/β-catenin pathways and inhibit the p53 pathway, and miR-18a also directly suppresses the expression of the tumor suppressors BTG3 and RBSP3 (CTDSPL). miR-18a overexpression in CC cells is triggered by OCT4 overexpression. Our data implicate PD-L1 as a novel oncoprotein and indicate that miR-140/142/340/383 and miR-18a are key upstream regulators of PD-L1 and potential targets for CC treatment.


Introduction
Cervical cancer (CC) is the fourth most common malignancy in women and the fourth leading cause of cancerrelated deaths among women worldwide [1,2]. Tumors suppress the host immune system by upregulating programmed death ligand 1 (PD-L1) that binds to programmed death-1 on T cells, resulting in inhibitory checkpoint signaling that inhibits T cell expansion and function [3][4][5]. Overexpression of PD-L1 has been found in human cancers, including CC and pancreatic cancer [6][7][8]. In addition to mediating T cell suppression, recent studies have shown the critical roles of PD-L1 in promoting cancer cell growth and invasion [9][10][11]. However, the exact biological function of PD-L1 in CC remains unclear.
We hypothesize that PD-L1 not only promotes tumor immune escape, it also enhances the malignant properties of CC cells. In the present study, we found that PD-L1 is overexpressed in CC and is an important promoter of CC cell proliferation and invasion. We also identify two novel mechanisms, including a miR-140/142/340/383-PD-L1 axis and an OCT4-miR-18a-PTEN/WNK2/SOX6 axis, that are responsible for the upregulation of oncoprotein PD-L1 in CC, suggesting that targeting PD-L1 by introducing miR-140/miR-142/miR-340/miR-383 or silencing of miR-18a might represent a therapeutic option to repress the metastatic phenotypes of CC cells and simultaneously reverse the immunosuppressive CC microenvironment.

PD-L1 is aberrantly expressed in primary CC samples and CC cell lines
We evaluated PD-L1 expression using immunohistochemical (IHC) analysis of 23 primary CC and paired adjacent normal tissue specimens. A strong PD-L1 staining was observed in CC samples (Fig. 1a). 78% of the cancer tissues displayed strong PD-L1 expression, whereas most adjacent normal samples (74%) showed no or weak PD-L1 expression (P < 0.01, Fig. 1b). The staining intensity of PD-L1 in CC was significantly higher than that in adjacent normal samples (2.61 ± 0.19 vs. 1.0 ± 0.14, P < 0.01). Higher PD-L1 expression levels were associated with a higher incidence of lymphovascular space invasion (Fig. 1c).
We next assessed PD-L1 expression in a panel of CC cell lines and the immortalized human endometrial epithelial EM cell line using immunoblotting. Compared with EM cells that do not have significant PD-L1 expression, CC cell lines overexpressed PD-L1 ( Supplementary Fig. 1a). The highest level of PD-L1 was observed in SiHa cells, which have a high capacity to invade in a Matrigel cell invasion assays [23] (Supplementary Fig. 1a), suggesting that PD-L1 may function as an oncoprotein in CC. Fig. 1 PD-L1 is upregulated in cervical cancer patients and is required for tumor growth and invasion. a Representative immunohistochemistry (IHC) image of PD-L1 in 23 paired human cervical cancer (CC) and adjacent normal cervical specimens (scale bar: 50 µm). b Percentage of cases that exhibit different PD-L1 protein expression levels (no staining, weak, or strong) in 23 paired human CC tissues and adjacent normal tissues (P < 0.01). c Representative IHC images of PD-L1 (left panel) and scatter plots showing the IHC scores for PD-L1 protein in patients with or without lymphovascular space invasion (LVSI, right panel). Scale bar: 50 µm. d Growth curves of PD-L1knockdown SiHa cells and CaSki cells with ectopic PD-L1 expression (n = 6). e Representative images (left panel) and data quantification (right panel) of colony formation of PD-L1-knockdown SiHa cells and CaSki cells with ectopic PD-L1 expression (n = 6). f Representative images (left panel) and data quantification (right panel) of Matrigel invasion assays of PD-L1-knockdown SiHa cells and CaSki cells with ectopic PD-L1 expression (n = 6). g, h Growth curves and quantification of weight of tumors that developed in nude mice injected subcutaneously with PD-L1-knockdown SiHa cells (g) or PD-L1overexpressing CaSki cells (h) (n = 8 per group). i Representative IHC images of Ki-67 in tumors from the mice described in g, h. Scale bar: 50 µm. *P < 0.05, **P < 0.01

PD-L1 promotes CC cell proliferation and invasion
To examine the contribution of endogenous PD-L1 to the proliferative and invasive phenotype of CC cells, we performed PD-L1 loss-of-function and gain-of-function experiments in CC cell lines ( Supplementary Fig. S1a). Short hairpin RNA-mediated knockdown of PD-L1 decreased SiHa cell proliferation and invasion, and stable overexpression of PD-L1 in CaSki cells promoted their proliferation and invasion ( Fig. 1d-f).
We next explored the biological function of PD-L1 in CC tumorigenesis in vivo by injecting PD-L1-knockdown SiHa cells or PD-L1-overexpressing CaSki cells into nude mice. PD-L1-knockdown SiHa cells generated smaller tumors, and PD-L1-overexpressing CaSki cells formed larger tumors than control cells (Fig. 1g, h). PD-L1-depleted SiHa tumors exhibited downregulation of Ki-67, but PD-L1-overexpressing CaSki tumors showed upregulation of Ki-67 (Fig. 1i). To examine whether the cytoplasmic domain of PD-L1 plays a critical role in promoting CC cell invasion and proliferation, we overexpressed either wild-type (WT) PD-L1 or cytoplasmic domain-truncated PD-L1 and observed that CaSki cells overexpressing WT PD-L1, but not cytoplasmic domain-truncated PD-L1, exhibited significantly higher proliferation and invasion than the control cells ( Supplementary Fig. S1a), suggesting that PD-L1 induces CC cell proliferation and invasion.

Identification of miRNAs that regulate PD-L1 expression
To identify miRNAs that suppress CC invasion and metastasis, we isolated highly invasive subpopulations of SiHa cells using Matrigel Invasion Chambers [24]. In contrast to parental SiHa cells, the majority of invasive SiHa cells exhibited a spindle-shaped and mesenchymallike morphology ( Supplementary Fig. S2a). Compared with the parental cells, the invasive SiHa sublines showed enhanced invasion abilities ( Supplementary Fig. S2a).
Then, a miRNA microarray assay was performed, and the expression profiles of invasive SiHa cells and their parental cells were determined ( Fig. S2b; Supplementary Table S1). We merged the obtained list of downregulated miRNAs in highly invasive SiHa cells obtained by microarray assay with miRNAs predicted by an in silico analysis, and identified miR-140/142/340/383 as potential regulators of PD-L1 in CC cells ( Supplementary Fig. S1c). Similarly, we defined miR-18a, which was upregulated in highly invasive SiHa cells (Supplementary Figs. S1c, S2b), and may act as the suppressor of PTEN, an upstream repressor of PD-L1 [3].
We tested whether PD-L1 mRNA expression is regulated by these identified miRNAs.

miR-18a promotes CC cell proliferation and invasion
We investigated the biological role of miR-18a in CC cells, and found that downregulating miR-18a levels in SiHa cells decreased proliferation and invasion, whereas overexpressing miR-18a in ME-180 and CaSki cells increased proliferation and invasion ( Supplementary Fig. S4a, b). To further explore the cellular function of miR-18a, we generated miR-18a-knockout SiHa cell lines using the CRISPR/Cas9 gene editing system by transfecting the lenti-CRISPR/Cas9-miR-18a vectors containing the individual sgRNAs into SiHa cells ( Supplementary Fig. S4c, d). The T7EN1 assay [25] showed the cleavage bands, suggesting that CRISPR/Cas9 induced mutations at the miR-18a locus ( Supplementary Fig. S4e). Among the miR-18a-knockout clones, we identified two clones that carried a 4-bp deletion or a 10-bp deletion ( Supplementary Fig. S4f). Deletion of 4 nucleotides significantly reduced and deletion of 10 nucleotides dramatically reduced (by more than 90%) the expression of mature miR-18a in SiHa cells ( Supplementary  Fig. S4g). miR-18a knockout significantly repressed CC cell proliferation and invasion ( Supplementary Fig. S4h, i).
To determine the effects of PD-L1 disruption in vivo, nude mice were subcutaneously injected with SiHa cells with PD-L1-knockout or control cells. Mice injected with PD-L1knockout cells developed smaller subcutaneous tumors than those injected with control cells ( Supplementary Fig. S4j, k), indicating that miR-18a promotes a metastatic phenotype in CC cells.
WNK2 negatively regulates the MEK/ERK pathway in CC cells [26]. We explored the role of the WNK2/MEK/ ERK pathway in regulating miR-18a-induced PD-L1 expression. Overexpression of miR-18a increased ERK1/2 phosphorylation and upregulated PD-L1 expression; however, forced WNK2 expression blocked miR-18a-induced MEK/ERK pathway activation and reduced PD-L1 expression ( Supplementary Fig. S7d), suggesting that miR-18a induces PD-L1 levels by targeting WNK2 to activate the MEK/ERK pathway. To further support these findings, we treated CC cells with the MEK inhibitor PD0325901. Inhibition of the MEK/ERK pathway with PD0325901 resulted in decreased PD-L1 expression and attenuated miR-18a-enhanced CC cell proliferation and invasion ( Supplementary Fig. S7e-g). These data suggest that miR-18a-mediated PTEN and WNK2 inhibition and subsequent activation of the PI3K/AKT and MEK/ERK pathways contributes to PD-L1 upregulation in CC cells.
We next determined whether SOX6 downregulates PD-L1 expression by repressing Wnt/β-catenin pathway activity. Western blotting analysis and TOP/FOP-Flash reporter assays revealed that overexpressing SOX6 inhibited the induction of active β-catenin levels and Wnt/β-catenin activity by miR-18a and downregulated cyclin D1 and PD-L1 induced by miR-18a overexpression in CC cells (Fig.  4b-d). Co-immunoprecipitation assay [29] showed that βcatenin coimmunoprecipitated with SOX6 in nuclear extracts from CC cells (Fig. 4e), demonstrating that SOX6 represses Wnt signaling and PD-L1 expression by binding to β-catenin.
Another possible mechanism by which SOX6 may antagonize Wnt/β-catenin activity is by repressing WISP3, an activator of this pathway [30]. Transient overexpression of SOX6 reduced endogenous WISP3 levels in CC cells (Fig. 4f, g). We identified the SOX6-binding motif within the WISP3 promoter (Fig. 4h). Chromatin immunoprecipitation (ChIP)-qPCR assays confirmed chromatin occupancy of SOX6 at the promoter regions of WISP3 and c-Myc [27] (Fig. 4i). We co-expressed a SOX6 expression vector with the WT WISP3 promoter or a version harboring a mutation at the predicted SOX6 binding site in CC cells. Forced SOX6 expression suppressed the activity of the WT, but not the mutant, WISP3 promoter compared with the controls (Fig. 4j). The inhibition of Wnt/β-catenin activity, PD-L1 expression, and CC cell proliferation and invasion by SOX6 were reversed by ectopic WISP3 expression ( Fig. 4k; Supplementary Fig. S9), suggesting that SOX6 increases PD-L1 expression by inhibiting WISP3.

OCT4 transactivates miR-18a in CC cells
Using the TRANSFAC database, we detected the binding sites for the transcriptional factors such as OCT4 and E2F1 [31], and found a positive correlation between OCT4 and miR-18a expression in CC cell lines ( Supplementary Fig.  S11a, b), indicating that OCT4 might positively regulate miR-18a expression. Transient overexpression of OCT4 increased miR-18a and PD-L1 levels, and siRNA-mediated silencing of OCT4 reduced miR-18a and PD-L1 expression ( Supplementary Fig. S11c, d). ChIP-qPCR assays showed the binding of OCT4 to the promoter regions of miR-18a and miR-125b [32] in CC cells (Supplementary Fig. S11e). The luciferase reporter assays revealed that the luciferase activity of the WT, but not the mutant, miR-18a promoter was increased by ectopic OCT4 expression but decreased by OCT4 depletion (Supplementary Fig. S11f). Overexpression of OCT4 increased CC cell proliferation and invasion, but inhibition of miR-18a with an anti-miRNA inhibitor blocked these OCT4-mediated effects (Supplementary Fig. S11g). Knockdown of OCT4 inhibited CC proliferation and invasion, and these suppressive effects were partially rescued by restoring miR-18a levels (Supplementary Fig. S11h). Thus, OCT4 transactivates and upregulates PD-L1 expression, thereby promoting CC cell proliferation and invasion.
Prognostic impacts of PD-L1, OCT4, WNK2, SOX6, BTG3, and RBSP3 in CC By analyzing web-based microarray database (GENT, MethHC, and ONCOMINE) that provide gene or miRNA expression patterns across normal and CC tissues, we found that OCT4, miR-18a, and PD-L1 expression levels were significantly upregulated, whereas WNK2, SOX6, BTG3, and RBSP3 levels were significantly downregulated in CC tissues compared with normal cervical tissues (Supplementary Fig. S12a). Moreover, we evaluated the prognostic impacts of these genes on patient outcomes. Sixty CC patients were classified into two groups using the median expression value as the cutoff. Patients with higher levels of PD-L1 and OCT4 or lower levels of WNK2, SOX6, BTG3, and RBSP3 had much shorter overall survival (Supplementary Fig. S12b). Lastly, we investigated whether elevated miR-18a expression is associated with OCT4 upregulation and with PTEN, WNK2, SOX6, BTG3, and RBSP3 downregulation. qRT-PCR analysis revealed a positive correlation between the expression of miR-18a and OCT4, and an inverse correlation of between miR-18a and  4). e Coimmunoprecipitation assays reveal associations between endogenous SOX6 and β-catenin in CC cells. Cell lysates are labeled as input. IgG was used as the control (n = 3). f, g qRT-PCR (f) and western blotting (g) of WISP3 expression in CaSki and SiHa cells transfected with the SOX6 expression vector or control vector (n = 3). h Schematic representation of the human WISP3 gene promoter. The predicted SOX6 binding site is shown in red, and the primer sequences for ChIP-qPCR assays are underlined. i SOX6 occupancy at its potential binding site in the WISP3 gene promoter was analyzed using ChIP-qPCR assays. c-Myc, a known SOX6 target, was used as a positive control (n = 3). j CaSki and SiHa cells were transfected with the wild-type (WT) or mutant (MU) WISP3-promoter luciferase plasmid with or without the SOX6 expression vector. Relative luciferase activity was assessed (n = 4). k Western blot analysis of SOX6, WISP3, active βcatenin, and PD-L1 in CaSki and SiHa cells transfected with SOX6 expression vector or control vector and WISP3 expression vector or the appropriate control (n = 3). l Western blot analysis of p53, p21, and PD-L1 in CaSki and SiHa cells transfected with the SOX6 expression vector or control vector in the presence or absence of p53 siRNA (n = 3). m qRT-PCR analysis of miR-34a expression in CaSki and SiHa cells transfected with the SOX6 expression vector in the presence or absence of p53 siRNA (n = 3). n Western blot analysis of PD-L1 in CaSki and SiHa cells transfected with the miR-34a mimic or its negative control (n = 3). **P < 0.01, ***P < 0.001 PTEN, WNK2, SOX6, BTG3, or RBSP3 levels in CC samples ( Supplementary Fig. S13c). Altogether, these data support the utilization of PD-L1, OCT4, WNK2, SOX6, BTG3, and RBSP3 levels in clinical outcome predictions of CC.

Discussion
The present study shows that PD-L1 functions as a powerful cancer driver by stimulating cancer proliferation and invasion in vitro and facilitating cancer growth in vivo, providing a new insight into the involvement of PD-L1 in CC metastasis. Our results highlighted the crucial roles of miR-140/142/340/383 downregulation and miR-18a upregulation in elevating PD-L1 expression and promoting CC progression. OCT4 induces the transcription of miR-18a, which targets multiple tumor suppressors, enhances PI3K/ AKT, MEK/ERK, and Wnt/β-catenin pathway activity, and attenuates p53 signaling, finally leading to PD-L1 upregulation (Fig. 5).
miR-18a, a member of the miR-17-92 cluster, is frequently overexpressed in human cancers and associated with cancer progression [38]. A previous study reported miR-18a upregulation in CC samples compared with normal cervical samples [39]. We showed here that miR-18a stimulates the metastatic ability of CC cells, and verified therapeutic effectiveness of CRISPR/Cas9-mediated miR-18a disruption in a xenograft mouse model of CC, suggesting that miR-18a-targeted therapy is a potential approach for treating patients with CC. The low miR-18a expression in normal cervical tissues may indicate that miR-18a-targeted therapy will be characterized by high specificity and limited toxicity.
Several mechanisms have been shown to sustain miR-18a expression in cancer cells. Activation of the transcription factors STAT3 and E2F1 induces miR-18a expression through direct binding [30,40], highlighting the fact that Once miR-18a expression is triggered by OCT4 overexpression, miR-18a enhances PD-L1 levels by targeting PTEN, WNK2 (ERK1/ 2 pathway inhibitor), and SOX6 (Wnt/β-catenin pathway inhibitor and p53 pathway activator) to activate the PI3K/ AKT, MEK/ERK, and Wnt/βcatenin pathways and inhibit the p53 pathway, therefore promoting tumor growth and invasion our findings on OCT4 do not represent the only mechanism responsible for miR-18a overexpression in CC. Interestingly, BTG3, a miR-18a target gene identified in this study, can directly bind and inhibit E2F1 [41], indicating the presence of a miR-18a-BTG3-E2F1 feedback loop that leads to persistent miR-18a overexpression in CC cells.
This study uncovers two previously unappreciated mechanisms contributing to PD-L1 overexpression in CC, the miR-140/142/340/383-PD-L1 axis, and the OCT4-miR-18a-PTEN/WNK2/SOX6 oncogenic axis, thereby providing insights into epigenetic/genetic mechanisms underlying the pathogenesis of CC, and highlighting the clinical potential of miRNA-based CC therapy.

Cell culture, reagents, and transient transfection
The CC cell lines (CaSki, ME-180, and SiHa) and pancreatic cancer cell lines (AsPC-1 and PaCa-2) were purchased from the ATCC. The immortalized human endometrial epithelial EM cell line was kindly provided by Satoru Kyo (Shimane University, Japan). Cells were cultured under the recommended conditions. Highly invasive SiHa cells were selected using Matrigel invasion chambers as previously described [24]. These cell lines were routinely tested by PCR for mycoplasma contamination. The MEK inhibitor PD0325901 and the Wnt/beta-catenin inhibitor XAV939 were obtained from Sigma-Aldrich. CC cells were treated with PD0325901 (1 µM) for 48 h or with XAV939 (4 or 8 µM) for 16 h. The miRNA mimics, anti-miRNA inhibitors, siRNAs against PD-L1, p53, and OCT4 and negative control siRNA were obtained from Ambion (Austin, TX). Transient transfection was accomplished with using Lipofectamine 3000 (Invitrogen, Carlsbad, CA) according to the manufacturer's instructions. The cDNA vectors encoding human PD-L1, SOX6, WNK2, BTG3, RBSP3, WISP3, or OCT4 were from OriGene (Rockville, MD). To generate stable cell lines overexpressing PD-L1, CaSki cells were transfected with a human PD-L1 cDNA expression vector (Origene) using Lipofectamine 3000 (Invitrogen). The respective control vector was used to generate control cell lines. After 24 h, stable transfectants were selected in medium containing 0.5 mg/ml G418 (Sigma-Aldrich). The lentiviral vector harboring shRNA against human PD-L1 and the appropriate control were purchased from the shRNA/ORF Core Facility (MD Anderson Cancer Center). SiHa cells with stable PD-L1 knockdown were generated as previously reported [42]. The expression vectors harboring either full-length or cytoplasmic domain-truncated PD-L1 gene were constructed and stable cell lines were established as previously reported [43].

CC specimens and IHC staining
All work related to human tissues was performed under protocols approved by the Institutional Review Board at Sun Yat-Sen University Cancer Center, China. Informed consent was obtained from all patients. Twenty-three paired human CC tissues and adjacent non-tumor tissues were collected at Sun Yat-Sen University Cancer Center. IHC staining was performed using antibodies against PD-L1 (clone E1L3N, Cell Signaling) as previously described [44]. Non-specific IgG was used as a negative control. A semiquantitative staining score ranging from 0 to 8 was calculated for each specimen by combining the staining intensity (0: negative; 1: low; 2: moderate; and 3: high) and the percentage of positive cells (0: negative; 1+: 0-1%; 2+: 1-10%; 3+: 11-33%; 4+: 34-66%; and 5+: 67-100%). Further, the degree of PD-L1 expression was quantified as follows: Negative expression (0 score), weak expression (1-3 scores), or strong expression (4-8 scores). In addition, a total of 60 pairs of snap-frozen primary CC specimens and adjacent non-tumor cervical tissues were processed for qRT-PCR analysis. Written informed consent was obtained from all patients.

qRT-PCR analysis of miRNA and mRNAs
Total RNA was isolated using TRIzol (Invitrogen). qRT-PCR analysis of miRNA or mRNA was performed as previously reporetd [45,46].
Cell proliferation assay, colony formation assay, and Matrigel invasion assay Cell proliferation was determined using the Cell Counting Kit-8 (Dojindo, Kumamoto, Japan) according to the manufacturer's protocol. In the colony formation assay, 500 cells were added to each well of a 6-well culture plate. After 14 days of culture, cells were fixed using 10% formalin and then stained using 10% Giemsa. Colony formation efficiency was calculated as follows: Plate colony formation efficiency (%) = (number of colonies/number of cells inoculated) × 100. Cell invasion was monitored and analyzed as described previously [45,46].

Tumor xenograft experiments
All experiments involving mice were performed in accordance with the guidelines of the Animal Care and Use Committee of Sun Yat-Sen University Cancer Center. Fiveweek-old female nude mice (BALB/c) were purchased from Shanghai Laboratory Animal Center (Chinese Academy of Sciences, Shanghai, China). SiHa cells (1 × 10 6 ) were suspended in phosphate buffered saline (100 µl) and then injected subcutaneously into mice. Tumor size was monitored using digital calipers, and tumor volume was calculated as length × width 2 x 0.5. All the mice were euthanized at day 35 post-inoculation. The excised tumors were embedded in paraffin and then subjected to IHC analysis of Ki-67 (ab15580, Abcam).

Microarray experiment
miRNA expression profiling was performed using Superprint G3 Human GE 8x60k Microarrays (Agilent Technologies) as previously described [44]. Microarray expression data were imported into GeneSpring software (Agilent Technologies). Differentially expressed miRNAs and mRNAs (P < 0.05 and fold change >2.0) were selected for further analysis. Microarray data was shown in Supplementary Table S1.

Luciferase activity assay
The 3′-UTR of PD-L1 was amplified from human genomic DNA and cloned downstream of the luciferase gene in the pMIR-reporter luciferase vector (Ambion, Austin, TX) as previously reported [48]. To construct the WISP3 or pri-miR-18a promoter luciferase vector, the promoter region encompassing the SOX6 or OCT4 binding site was amplified from human genomic DNA and cloned into the MluI/ BglII sites of the pGL3 vector (Promega). The potential miR-140/142/340/383 binding sites in the 3′-UTR of PD-L1 and the potential SOX6 and OCT4 binding sites in the WISP3 and pri-miR-18a promoter regions were mutated using a QuickChange Site-Directed Mutagenesis Kit (Stratagene, La Jolla, CA). ImiRP was used for mutating miRNA binding sites that are located in the 3′-UTR of target genes [49]. Luciferase activity was measured 48 h after transfection using a Dual Luciferase Assay Kit (Promega).

TOP-Flash/FOP-Flash reporter assay
The TOP-Flash reporter and pTK-RL plasmids were cotransfected into CC cells (5 × 10 4 ) in 24-well plates with the miR-18a mimic and a SOX6 expression vector, and the activity of the Firefly and Renilla luciferase reporters was determined at 48 h after transfection using a Dual Luciferase Assay Kit (Promega). TOP-Flash reporter activity is presented as the relative ratio of Firefly luciferase activity to Renilla luciferase activity.

ChIP-qPCR assay
Cellular chromatin was immunoprecipitated with an antibody against SOX6 (ab30455, Abcam), OCT4 (ab19857, Abcam), or IgG (control, Santa Cruz Biotechnology) using the Pierce Agarose ChIP Kit (Pierce). The amount of immunoprecipitated DNA was quantified using Takara SYBR Premix Ex Taq II (Takara). The results are presented as the fold enrichment over the IgG control. The human c-Myc [27] and miR-125b [32] promoters were used as positive controls for SOX6 and OCT4 binding, respectively.

Statistical analysis
Statistical analyses were performed using SPSS version 22.0 (SPSS Inc., Chicago, IL, USA). All results are expressed as mean ± s.d. All experiments were performed at least three times. Statistical differences between categorical data were evaluated by Fisher's exact test. Comparisons between groups of continuous variables were performed using Student's t-test (2 groups) or one-way ANOVA test (>2 groups). The Wilcoxon signed-rank test was used to compare two groups of paired nonparametric data. Differences in mRNA or miRNA expression between CC tissues and normal cervical tissues were evaluated using the nonparametric Mann-Whitney U test. Results with two-tailed P-values < 0.05 were judged to be statistically significant.