FGSCs were cultured according to the previously method (Zou et al., 2009; Zhang et al., 2016; Li et al., 2017b). FGSCs passages were performed at a ratio of 1:4 and intervals of 4–7 days.

We used pLCDH-ciR to construct the circGFRα1 overexpression vector to perform transcript circularization. An EcoRI restriction enzyme site was discovered within an endogenous flanking sequence located in the front circular frame. On the other hand, a BamHI site was detected within a partially inverted upstream sequence located in the back circular frame. We cloned the fragment amplified into the vector between the front and the back circular frames. We also constructed a control vector that contained only a non-sense insert between the front and back circular frames in the absence of circGFRα1-encoding cDNA. Lentiviral vectors (pGMLV-SC5) loaded with anti-circGFRα1 shRNA and the negative control shRNA (nc-shRNA) with EGFP were provided by Shanghai Genomeditech Company Ltd. In the meantime, we used an irrelevant, scrambled shRNA that was not matched with the mouse genome sequence as the control.

For constructing the METTL14-knockdown lentiviral vectors, we applied molecular biological approaches to insert an interfering fragment in the U6 promoter downstream into the lentiviral vector (pLKD-CMV-G and PR-U6-shRNA). We selected four or more independent siRNAs to examine the target knockdown efficiency. Finally, we selected the optimal siRNA target (targetSeq: GCTAAAGGATGAGTTAAT). To construct METTL14 overexpression vectors, we inserted the candidate gene cDNAs into the BamHI and EcoRI restriction sites in an overexpression plasmid (pHBLV-CMVIE-ZsGreen-T2A-puro).

We used 293T cells to prepare lentiviruses according to a method described previously (Shi et al., 2004). For lentivirus infection, the FGSCs were incubated with 1:1 mixture of culture medium and lentivirus (titer: 2.5 × 10⁸). After 12–16 h of infection, the culture medium containing lentivirus particles was sucked out, and the culture medium was added to the culture plate for further culture. Subsequently, the FGSCs were screened by 100 ng/ml puromycin.

The RNA-FISH procedure for circGFRα1 was performed with the RNA-FISH kit of GenePharma Inc. (Shanghai, China), according to the manufacturer’s instructions. Briefly, FGSCs were cultured in 48-wells overnight at 37°C and 5% CO2. Next day, cells were rinsed in chilled phosphate-buffered saline (PBS) two times. Cells were fixed with 4% paraformaldehyde in the room temperature for 30 min. Then 0.1% buffer A was added and incubated in the room temperature for 30 min. After that, the cells were washed twice with PBS for 5 min. The RNA-FISH probes targeting circGFRα1 were synthesized by GenePharma Inc. (Shanghai, China). Those probes were mixed well with buffer E to a final concentration 2 μM, then to hybridize with cell’s genes and incubate overnight at 37°C. The cells were washed twice with buffer C for 5 min. This was followed by rinsing and staining with DAPI (1:1,000 dilution; Sigma)-containing PBS at room temperature for 5 min. The images were acquired with a fluorescence microscope (Leica, United States).

We cultured FGSCs in each well (containing 200 μL culture medium) of 96-well plates at a density of 5,000 cells/well. At a confluence of 70%–80%, we added 20 μL of the CCK8 reagent into each well of the plate and incubated it for another 2 h at 37°C and 5% CO₂. Then, we used a microplate reader to measure the absorbance (OD) value at 450 nm.

We cultured FGSCs to 80% confluence and added 50 μM of the EdU reagent into each well, after which the plate was incubated for another 2 h. Thereafter, cells were fixed with 4% PFA for 30 min under ambient temperature and neutralized for 5 min by using 2 mg/mL glycine. Then, the cells were punched with 0.5% Triton X-100, stained with the 1 × Apollo staining solution, incubated for 30 min, and finally washed thrice by using PBS supplemented with 0.5% Triton X-100. Finally, cell nuclei were dyed using 1 × Hoechst 33342. A Leica fluorescence microscope was used to capture images.

Trizol reagent was used to extract total RNA from FGSCs. After denaturing at 95°C, the RNAs were immediately chilled on ice. Then, they were dropped onto a Hybond-N + membrane. The membrane was cross-linked with a UV cross-linker, followed blocking with 5% skim milk, and overnight incubation with the m6A-specific antibody (1: 1,000) at 4°C. Thereafter, the membrane was rinsed by TBST for 10 min, followed by another 1 h of incubation with a secondary antibody at an ambient temperature. Finally, Tanon 4600SF was used to scan the dots.

Trizol reagent was used to extract total cellular RNA from FGSCs that was quantified using Nanodrop Lite. cDNA was prepared by reverse transcribing the RNA (1,000 ng) by using a reverse transcription kit in the 20-μL system. In this qRT-PCR assay, Taq DNA polymerase was used with SYBR Premix Ex Taq in the Applied Biosystems^® 7500 Real-Time PCR system. The 2^–ΔΔCt approach was used for data analysis.

Total RNA was extracted from FGSCs by using the Trizol reagent and quantified using Nanodrop Lite. We bound 1.25 μg of the anti-m6A antibody onto protein A/G magnetic beads dissolved in the IP buffer (consisting of 140 mM NaCl, 20 mM Tris pH 7.5, 2 mM EDTA, and 1% NP-40) 1 h in advance. Thereafter, we incubated the resultant antibody-bound protein A/G beads with the RNA sample at 4°C for 2 h. Then, the obtained samples were rinsed twice by using low-salt wash buffer (composed of 5 mM EDTA and 10 mM Tris; pH 7.5), followed by washing with a high-salt wash buffer (composed of 1 M NaCl, 20 mM Tris pH 7.5, 0.5% sodium deoxycholate, 1% NP-40, 1 mM EDTA and 0.1% SDS) twice and then with the RIPA buffer (150 mM NaCl, 20 mM Tris pH 7.5, 0.5% sodium deoxycholate, 1% NP-40, 1 mM EDTA, and 0.1% SDS) twice. For all the samples, their wash solutions were harvested and considered the EDTA, an fraction. In addition, the beads were incubated with 50 μL N⁶-methyladenosine 5-monophosphate sodium salt (20 mM) at 4°C for 1 h to elute the RNA. After precipitation with ethanol, cDNA was prepared by reverse transcribing the RNA in the input, unbound and m6A-bound fractions by using Superscript III random hexamers. Subsequently, qRT-PCR was performed to determine the m6A-containing transcript levels compared with the Rplp0 level. For Rplp0, the primer sequences were GATGGGCAACTGTACCTGACTG and CTGGGCTCCTCTTGGAATG.

We inserted the mutant (circGFRα1-MUT) and wild-type (circGFRα1-WT) circGFRα1 miRNA-binding site sequences into SacI and KpnI sites in the pGL3 promoter vector. Thereafter, we cultured the cells in 24-well plates and used Lipofectamine 3000 to transfect 5 ng of the pRL-SV40 Renilla luciferase vector and 80 ng plasmid, together with 50 nM of the negative control or miR-449 mimics into the cells. After 48 h, we harvested and examined the cells by performing the dual-luciferase assay according to specific protocols. Each independent experiment was performed thrice.

The significance of difference in graphs was assessed using Student’s t-test unless specified otherwise. The normally distributed two-sample equal variance was used in this study. Our researchers were aware of sample grouping in each experiment. P value of <0.05 indicated statistical significance. Graphs as well as error bars are represented as means ± SEM unless specified otherwise. The R statistical environment and GraphPad Prism 4.0 were used for performing statistical analysis.

To identify circRNAs involved in FGSC formation, we reanalyzed our previous circRNA expression data (Li et al., 2019) and found that circRNA_12447 (chr19: 58263912–58270163) is upregulated in FGSCs that possibly leads to FGSC differentiation and self-renewal. As discovered using the mouse reference genome (mm10), circRNA_12447 originated from exons 6–8 at the locus of GDNF family receptor alpha 1 (GFRα1) (Figure 1A) and was thus referred to as circGFRα1. For investigating the expression profile of circGFRα1 and verifying its circular shape, RT-PCR was performed using divergent primers, through which the expression of circGFRα1 in FGSCs was detected (Figure 1B). To confirm the circular shape of circGFRα1, RNase R, a 3′–5′ exoribonuclease with high process ability and no effect on circRNAs, was used. circGFRα1 developed resistance to RNase R exposure compared with the control linear gene GFRα1 (Figure 1C). In addition, RT-PCR products obtained after amplification with divergent primers were subjected to Sanger sequencing, which confirmed that circGFRα1 contains the back-spliced junction (Figure 1D). After exposure of circGFRα1 and the control linear gene to actinomycin D, a transcription inhibitor, our qRT-PCR results revealed that half-life of circGFRα1 is over 12 h, whereas that of the related linear transcript was only 3 h (Figure 1E), indicating higher stability of circGFRα1 in FGSCs. Furthermore, we performed fluorescence in situ hybridization assays that revealed that circGFRα1 is mostly located in the cytoplasm (Figure 1F). Taken together, the findings indicate the stable and rich expression of circGFRα1 in FGSCs.

Tissue-specific expression results revealed that circGFRα1 is highly expressed in the mouse ovary (Supplementary Figure 1A). Moreover, the circGFRα1 level was found to be stage-specific during the development of mouse FGSCs; a high level of transcript was present in FGSCs, whereas significantly lower levels were detected in germinal vesicle (GV)-stage oocytes (P < 0.05) and metaphase II (MII)-stage oocytes (Supplementary Figure 1B, P < 0.001). To examine the effect of circGFRα1 on FGSC development, we regulated its expression through RNA interference or overexpression by inducing lentivirus infection (Figure 2A). As expected, circGFRα1 overexpression was found to significantly upregulate its level (Figure 2B, P < 0.001). However, relative to negative controls, its expression level was significantly decreased in shRNA-loaded lentivirus-infected FGSCs, which specifically bind to the circGFRα1 junction site (Figure 2C, P < 0.001). Then, we examined the effects of circGFRα1 overexpression and knockdown on FGSC proliferation through CCK-8 and EdU incorporation assays, respectively. According to CCK8 assay results, the OD values of FGSCs with circGFRα1 overexpression markedly increased relative to controls, whereas those of circGFRα1-knockdown FGSCs evidently decreased compared with those of controls (Figure 2D, P < 0.001). Moreover, the EdU assay results indicated that EdU-positive FGSCs with circGFRα1 overexpression that were transfected with lentivirus were markedly more in number than controls (Figure 2E, P < 0.001). However, transfection with circGFRα1-knockdown lentivirus markedly decreased the number of EdU-positive FGSCs compared with that of controls (Figure 2E, P < 0.001). Thus, we found that genes, including Akt, Bcl6b, Lhx, and Etv5, associated with self-renewal with the highest responsiveness to GDNF signaling within FGSCs are significantly upregulated (Figure 2F). Meanwhile, we found that the expression levels of genes associated with FGSC self-renewal were significantly downregulated (Figure 2F), whereas the expression levels of genes associated with differentiation, such as Stra8, Sypc3, and Kit, were extremely low and not affected by circGFRα1 knockdown (Figure 2G, P > 0.05). In summary, these findings suggest that the overexpression of circGFRα1 promoted the self-renewal and maintenance of FGSCs, while its knockdown impaired these characteristics.

To explore the potential of circGFRA1 as a miRNA sponge, we performed the RNA hybrid analysis¹. From our results, we predicted that certain miRNA-binding sites are present in circGFRα1. Of the estimated miRNAs, miR-449, which was also predicted to target the GFRα1 gene based on TargetScan and miRanda, was identified. Figure 3A shows the miR-449 seed region nucleotides (denoted in red). Later, CCK-8 and EdU assays were performed for examining the biological effects of miR-449 on FGSCs. We found that FGSCs transfected with miR-449 inhibitors have enhanced proliferation capacity, whereas those transfected with miR-449 mimics have markedly reduced proliferation capacity (Figures 3B,C). These findings indicate the role of miR-449 in the regulation of FGSC proliferation.

To verify whether the circGFRα1 transcript interacts with miR-449, the luciferase reported gene assay was performed using the circGFRα1-fused reporter gene (pGL3-circ GFRα1). In addition, a construct containing a non-specific circGFRα1 sequence (pGL3-circEGFR-MUT) and a wild-type construct (pGL3-circGFRα1-WT) were developed (Figure 3D). Then, miR-449 mimic and pGL3-circGFRα1-WT were co-transfected into the cells, which markedly reduced the luciferase activity in these cells compared with that in cells co-transfected with control miRNA and pGL3-circGFRα1-WT or co-transfected with mimic and pGL3-circGFRα1-MUT (Figure 3E). These findings indicated that miR-449 directly binds to circGFRα1 and adversely targets the latter. Furthermore, to verify whether circGFRα1 directly binds to miR-449, RIP assays were performed using control IgG or anti-AGO2 antibodies. The qRT-PCR assay was used to analyze miR-449 and circGFRα1. We found that anti-AGO2 antibodies markedly downregulate miR-449 and circGFRα1 compared with control IgG (Figure 3F), which suggests that miR-449 directly binds to circGFRα1 in the presence of AGO2.

To explore the potential of circGFRα1 as a ceRNA for sequestering miR-449 and upregulating GFRα1 expression and activation of the GDNF signaling pathway. We firstly showed that the circGFRα1 expression is comparable to the GDNF signal after removal and replenishment of GDNF; its expression decreased 18 h after GDNF removal but increased after GDNF replenishment (Figure 4A). Moreover, TargetScan was used to identify the miR-449 putative target genes, which predicted GFRα1 (Figure 4B). To confirm this prediction, luciferase assays were performed using a GFRα1-fused reporter gene. Co-transfection of miR-449 mimic with the GFRα1 UTR significantly reduced the luciferase activity compared with that in control miRNA co-transfected with the GFRα1 UTR or in mimic co-transfected with the GFRα1 UTR (Figure 4C). These findings revealed that miR-449 directly binds to the GFRα1 UTR and adversely targets the latter. To verify whether the GFRα1 UTR directly binds to miR-449, RIP assays were performed using control IgG and anti-AGO2 antibodies; miR-449 and the GFRα1 UTR were analyzed through qRT-PCR. We found that anti-AGO2 antibodies markedly downregulate the expression of miR-449 and the GFRα1 UTR compared with control IgG (Figure 4D), indicating that the GFRα1 UTR directly binds to circGFRα1 depending on the presence of AGO2. To examine whether circGFRα1 regulates the GFRα1 level by interacting with miR-449, we detected GFRα1 expression in circGFRα1-deleted or circGFRα1-overexpressed FGSCs. As shown in Figures 4E,F, circGFRα1 overexpression increased the GFRα1 mRNA and protein expression, whereas circGFRα1 silencing reduced the GFRα1 mRNA and protein expression. Furthermore, the miR-449 inhibitor was co-transfected with si-circGFRα1 in FGSCs, which reversed the repression (Figure 4G). Taken together, the findings support the assumption that circGFRα1 modulates the GFRα1 level by directly interacting with miR-449.

As circGFRα1 functioned as a ceRNA, it might be exported to the cytoplasm in an m⁶A-dependent manner. To study whether m⁶A modification occurs in circGFRα1, we first predicted the m⁶A sites using an online bioinformatic tool, m6Avar², and found three RRACU m⁶A sequence motifs located in circGFRα1. Next, we performed methylated RNA immunoprecipitation (Me-RIP)–qPCR assays and found that the m⁶A level of circGFRα1 in FGSCs was very high (Figure 5A). We then performed shRNA-mediated silencing of METTL14, a core component of the m⁶A methylase complex, and found that downregulation of METTL14 resulted in decreases in the m⁶A levels of both total RNA and circGFRα1 (Figures 5B,C). The results of METTL14 knockdown were confirmed by western blotting (Supplementary Figure 2A). We then investigated whether m⁶A modification could affect the RNA metabolism of circGFRα1. Knockdown of METTL14 did not lead to a change in the expression of circGFRα1 (Supplementary Figure 2B).

In addition, we performed FISH and cytoplasmic and nuclear mRNA fractionation experiments, and found that silencing of METTL14 significantly increased the nuclear circGFRα1 content (Figures 5D,E, p < 0.001). We also found that, when circGFRα1 was overexpressed, both nuclear and cytoplasmic circGFRα1 levels were increased, particularly in the cytoplasmic fraction (Figure 5F, p < 0.001). By browsing the junction sequence in circGFRα1, we identified that the GGACU motif was a putative m⁶A motif. Once we mutated the GGACU m⁶A motif in a circGFRα1-overexpressing construct, although both nuclear and cytoplasmic circGFRα1 levels were increased, the main increase occurred in the nuclear fraction (Figure 5G, p < 0.001). These findings indicated that m⁶A modification of circGFRα1 facilitated circGFRα1 export from the nucleus to the cytoplasm in an m⁶A-dependent manner and that m⁶A modification of circGFRα1 are important for FGSC development.