Identification and characterization of circPDE5A
To investigate the potential role of circRNA in prostate cancer, we firstly performed a circRNA array in five paired samples of PCa by Arraystar Human circRNA array (Supplementary Table S1). A total of 27 dysregulated circRNAs were identified in PCa tissues (Fig. 1A). Among these circRNAs, 19 circRNAs could be found in the circbase database. However, only 8 of 19 circRNAs could be amplified in cDNA using the specific primers. Next, we examined the expression of these circRNAs in 15 paired prostate cancer and adjacent normal tissues. The results showed that only hsa_circ_0002474, derived from PDE5A, designated as circPDE5A, was downregulated in PCa tissues compared to adjacent normal tissues (Supplementary Fig. 1A). We then confirmed the expression of circPDE5A in 50 paired samples of PCa by qRT-PCR. The results demonstrated that hsa_circ_0002474 was significantly downregulated in prostate cancer compared with adjacent normal tissues (Fig. 1B). As shown in Fig. 1C, the expression of circPDE5A in patients with PCa was negatively correlated with Gleason score. Moreover, circPDE5A was downregulated in PCa cell lines compared with normal prostate epithelial cell line (Fig. 1D). These results reveal that the differential expression of circPDE5A might play a role in PCa progression.
circPDE5A is a 679-nt circRNA generated from the exon 2 to 3 of the PDE5A gene, and the junction site was confirmed using sanger sequencing (Fig. 1E). RT-PCR analysis showed that the divergent primer could amplify circPDE5A in cDNA reverse transcribed by total RNA, but not genome DNA (Fig. 1F). After treatment of actinomycin D, circPDE5A demonstrated more stability compared to PDE5A mRNA (Fig. 1G). RNase R was utilized to detect the stability of circPDE5A to examine the circular characterization of circPDE5A further. The results revealed that circPDE5A exhibited high resistance to RNase R digestion, whereas the liner RNA of PDE5A mRNA was mostly degraded (Fig. 1H). Next, both nuclear and cytoplasmic fractionation and FISH assay demonstrated that circPDE5A was localized in both the cytoplasm and nucleus (Fig. 1I, J). These results suggest the circularity and localization of circPDE5A in PCa cells.
circPDE5A restrains prostate cancer cells metastasis both in vivo and in vitro
To investigate the role of circPDE5A in PCa progression, we designed specific small-interfering RNAs (siRNAs) which targeting back-splicing junction sites of circPDE5A. qRT-PCR analysis showed that circPDE5A siRNAs specifically reduced the circPDE5A expression, while having little effect on PDE5A mRNA expression (Fig. 2A). Meanwhile, we stably overexpressed circPDE5A in C4-2B and 22Rv-1 cell lines using lentivirus plasmids, and the overexpression efficiency was evaluated (Supplementary Fig. 2A). Next, we investigated whether circPDE5A influenced PCa cells proliferation. The CCK-8 assay revealed that neither downregulation nor overexpression of circPDE5A had little effect on PCa cells proliferation (Supplementary Fig. 2B, C). Then, a transwell assay was performed to explore the ability of circPDE5A on migration and invasion. The results suggested that circPDE5A knockdown significantly promoted migration and invasion of PCa cells (Fig. 2B), while circPDE5A overexpression evidently inhibited migration and invasion of PCa cells (Fig. 2C). These results demonstrate that circPDE5A restrains PCa cells metastasis.
To further confirm the role of circPDE5A in vivo, xenograft tumor animal assay was performed using 22Rv-1 cells stably transfected with circPDE5A or vector. The result showed that circPDE5A overexpression has little effect on prostate cancer cells proliferation in vivo (Supplementary Fig. 2D), consistent with the in vitro assay. Next, we utilized a nude mice tail vein metastasis model to investigate the role of circPDE5A on metastasis of PCa cells in vivo. circPDE5A stably knockdown or overexpression 22Rv-1 cells were injected into the tail vein of nude mice. We found that 22Rv-1 cells with circPDE5A knockdown formed metastasis foci in 62.5% (5/8) mice in a period of 6 weeks after injection, while 22Rv-1 control cells only formed metastasis foci in 25% (2/8) mice (Fig. 2D). However, 22Rv-1 cells with circPDE5A overexpression formed fewer metastasis foci (2/8) than the control group (4/8) (Fig. 2E). Overall, these findings reveal that circPDE5A restrains metastasis of PCa cells in vivo.
circPDE5A binds to WTAP and regulates its m6A methylation activity
We next explored the detailed mechanism of circPDE5A in PCa metastasis. Since many circRNAs were reported to function by serving as “miRNA sponges”, we performed AGO2-RIP assay to examined whether circPDE5A could sponge miRNAs. However, the results showed that circPDE5A could not bind to AGO2 protein (Supplementary Fig. 3A), indicating that circPDE5A may not function as “miRNA sponges”. Next, we hypothesized whether circPDE5A exerted its roles by binding to the functional proteins. RNA-pulldown assay followed by qRT-PCR analysis revealed that circPDE5A probe could specifically bind to circPDE5A (Supplementary Fig. 3B). Then, the protein extraction in the RNA pulldown assay was separated by SDS-PAGE, and followed by silver staining (Fig. 3A), demonstrating that circPDE5A could specifically bind to many proteins. Moreover, RNA pulldown assay followed by mass spectrometry was performed to detect the circPDE5A binding proteins, and a total of 123 proteins were identified. Among these proteins, WTAP attracted our attention. WTAP was a key component of the m6A methyltransferase. Since the function of circRNAs in m6A modification remains elusive [24]. We, therefore examined the role of circPDE5A/WTAP complex in prostate cancer. We observed that circPDE5A bind directly to WTAP, but not to other m6A methyltransferases and demethylases METTL3, METTL14, FTO, and ALKBH5 (Fig. 3B). The RIP-qPCR assay was then used to confirmed the binding of circPDE5A and WTAP (Fig. 3C). Moreover, FISH and immunofluorescence assay was performed to confirm that circPDE5A was colocalized with WTAP protein in cells (Fig. 3D). Based on the interaction between circPDE5A and WTAP, we tested whether WTAP affect circPDE5A expression. However, the qRT-PCR analysis revealed that WTAP had little effect on circPDE5A expression (Fig. 3E and Supplementary Fig. 3C). Furthermore, changing circPDE5A expression did not alter the WTAP expression and localization (Fig. 3F–H, Supplementary Fig. 3D–F). Since WTAP was a well-known N6-methyadenosine methyltransferase which promoted m6A modification by recruiting METTL3 and METTL14 [25]. We speculated that the interaction between circPDE5A and WTAP could affect WTAP m6A activity. Therefore, we detected m6A levels in circPDE5A overexpression or knockdown PCa cells. The dot blot assay demonstrated that circPDE5A overexpression decreased the global m6A level, while circPDE5A knockdown increased the global m6A level in PCa cells (Fig. 3I and Supplementary Fig. 3G). The m6A immunofluorescence assay also received the same results (Fig. 3J and Supplementary Fig. 3H). Next, we explored how circPDE5A interfered the WTAP-dependent m6A modification in PCa. The CO-IP assay results showed that circPDE5A overexpression decreased the interaction between METTL3 and METTL14, while silencing of circPDE5A increased the interaction between METTL3 and METTL14 (Fig. 3K), suggesting that circPDE5A may block WTAP m6A activity via interfering the formation of METTL3-METTL14-WTAP complex. Meanwhile, the transwell assay revealed that silencing of WTAP inhibited PCa migration and invasion, at the same time, overexpression of WTAP promoted PCa migration and invasion (Fig. 3L, M, Supplementary Fig. 3I, 3 J), which was consistent with our previous findings. These results reveal that circPDE5A inhibits PCa metastasis via interacting with WTAP and blocks its m6A activity.
circPDE5A inhibits WTAP-mediated m6A modification of EIF3C mRNA and restrains its translation
To figure out the molecular mechanism in the regulatory effect of circPDE5A on m6A modification. We performed m6A methylated RNA immunoprecipitation sequencing (MeRIP-seq) in vector, circPDE5A overexpression, control and circPDE5A knockdown groups (Fig. 4A, B). To narrow down the scope of downstream genes, we integrated the circPDE5A overexpression and circPDE5A knockdown data (criteria: m6A peak |log2FC| > 1.5). A total of 31 m6A peak dysregulated genes is shown in the Venn diagram (Fig. 4C). Among these genes, 16 mRNAs were selected as the candidate downstream targets that might participated in the cancer progression. Then, we performed MeRIP-qPCR to evaluate the effect of circPDE5A on these candidate targets. The results showed that only the m6A modification of EIF3C was upregulated after circPDE5A knockdown and downregulated after circPDE5A overexpression (Supplementary Fig. 4A). These data suggested that EIF3C was the potential target of circPDE5A/WTAP complex (Fig. 4D). Next, we explored whether circPDE5A regulated EIF3C expression. The m6A modification and protein expression of EIF3C were elevated, while the EIF3C mRNA level has little change in circPDE5A knockdown PCa cells compared to scramble cells (Fig. 4E and Supplementary Fig. 4B). By contrast, the m6A modification and protein expression of EIF3C were decreased, while the EIF3C mRNA level had little change in circPDE5A overexpression PCa cells compared to vector cells (Fig. 4F and Supplementary Fig. 4C). Next, we explored why circPDE5A altered the EIF3C protein level but not the mRNA level. Since it is reported that the m6A modification of mRNA could change its cellular distribution [26], we conducted nuclear and cytoplasmic fractionation, and the results suggested that circPDE5A overexpression did not alter the EIF3C mRNA cellular distribution (Fig. 4G and Supplementary Fig. 4D). Similarly, overexpression or knockdown of circPDE5A did not change the EIF3C mRNA stability (Fig. 4H and Supplementary Fig. 4E, F). Then, protein translation inhibitor cycloheximide (CHX) was added in circPDE5A knockdown or overexpression PCa cells. The results revealed that neither knockdown nor overexpression of circPDE5A had little effect on EIF3C protein stability, excluding the possibility that circPDE5A affects EIF3C protein stability (Fig. 4I and Supplementary Fig. 4G). Finally, polysome profiling demonstrated that overexpression of circPDE5A considerably decreased the enrichment of EIF3C mRNA in the polysome fractions but increased in the non-polysome fractions (Fig. 4J), while knockdown of circPDE5A increased the enrichment of EIF3C mRNA in the polysome fractions (Supplementary Fig. 4H), supporting that circPDE5A regulated EIF3C protein expression via interfering with its translation.
Given that circPDE5A regulated EIF3C translation via altering its m6A modification, we first verified whether WTAP played an indispensable role in this process. MeRIP-qPCR assay suggested that silencing WTAP decreased the EIF3C m6A modification significantly, while overexpression of WTAP increased the EIF3C m6A modification in PCa cells (Fig. 5A, B). Next, WTAP-RIP-qPCR analysis showed that silencing circPDE5A enhanced the binding capacity between WTAP and EIF3C mRNA (Fig. 5C). However, circPDE5A overexpression impaired this interaction (Fig. 5D). Meanwhile, we found that WTAP overexpression could reverse circPDE5A induced EIF3C downregulation in protein levels rather than mRNA levels (Fig. 5E and Supplementary Fig. 5A). These results suggested that circPDE5A regulated EIF3C m6A modification via interfering with the interaction between WTAP and EIF3C mRNA.
Previous studies had reported that YTHDF1 and IGF2BP1, two m6A readers, could regulated protein translation via recognizing the m6A modification of mRNA [27, 28]. Thus, western blotting assay was used to explored which reader influenced the expression of EIF3C. The result showed that silencing of YTHDF1, but not IGF2BP1, decreased the protein level of EIF3C significantly (Supplementary Fig. 5B, C), revealing that YTHDF1 might a key reader regulating EIF3C mRNA translation. The YTHDF1-RIP-qPCR assay was then used to prove that YTHDF1 could bind more EIF3C mRNA compared to the IgG control group (Fig. 5F). Meanwhile, the binding of EIF3C and YTHDF1 also was confirmed by the CLIP-seq data (http://starbase.sysu.edu.cn/index.php) (Supplementary Fig. 5D). In addition, the YTHDF1-RIP-qPCR assay showed that silencing of circPDE5A enhanced the binding capacity between YTHDF1 and EIF3C (Fig. 5F and Supplementary Fig. 5E). At the same time, circPDE5A overexpression decreased the YTHDF1 and EIF3C binding capacity (Fig. 5G and Supplementary Fig. 5F), suggesting that the binding capacity between YTHDF1 and EIF3C mRNA depending on the m6A modification level of EIF3C mRNA. Also, WB analysis demonstrated that silencing of YTHDF1 abrogated the promoting effect of circPDE5A knockdown in the EIF3C protein level (Fig. 5H). These data demonstrate that circPDE5A regulated EIF3C protein expression in a YTHDF1-dependent manner.
EIF3C promotes prostate cancer cells metastasis through MAPK pathway
EIF3C was one of the subunits of EIF3, playing vital roles in translation initiation [29]. Previous studies had revealed that EIF3C was inevitable for tumor progression in many cancer types, including lung cancer, glioma, and ovarian cancer [30,31,32]. However, the specific role of EIF3C in prostate cancer is still elusive. We found that the protein level EIF3C was upregulated in PCa tissues compared to adjacent normal tissues (Fig. 6A, B). To further investigated the role of EIF3C in PCa progression, we first manipulated the expression of EIF3C in C4-2B and 22Rv-1 cells, and the knockdown or overexpression efficiency was confirmed (Fig. 6C, D, Supplementary Fig. 6A, B). The transwell assay in C4-2B and 22Rv-1 cells demonstrated that silencing of EIF3C significantly restrained the migration and invasion ability of PCa cells (Fig. 6E). In contrast, overexpression of EIF3C promoted the ability of migration and invasion of PCa cells (Fig. 6F). Previous studies revealed that EIF3C promoted cancer progression via regulating MAPK pathway [33,34,35]. We, therefore, examined the p-mTOR, p-AKT, and p-P38 levels in PCa cells after EIF3C knockdown or overexpression. The WB analysis demonstrated that overexpression of EIF3C elevated the phosphorylation level of mTOR, AKT, and P38 in C4-2B and 22Rv-1 cells (Fig. 6G). However, the phosphorylation level of mTOR, AKT and P38 in PCa cells was decreased after EIF3C knockdown (Fig. 6H). These results reveal that EIF3C participates in PCa progression via the MAPK pathway.
circPDE5A restrains prostate cancer metastasis via EIF3C
Since circPDE5A regulated EIF3C protein levels in PCa tissues, we first examined whether circPDE5A also regulated the MAPK pathway. The WB analysis showed that overexpression of circPDE5A decreased the p-mTOR, p-AKT and p-P38 levels in PCa cells while silencing circPDE5A elevated the p-mTOR, p-AKT, and p-P38 levels in PCa cells (Fig. 7A, B). We then overexpressed the EIF3C level in circPDE5A overexpression PCa cells, and the WB analysis showed that the EIF3C expression was restored in PCa cells (Fig. 7C). The transwell assay revealed that circPDE5A overexpression inhibited migration and invasion of PCa cells, while this inhibition effect could be reversed via overexpression of EIF3C (Fig. 7D and Supplementary Fig. 6C). Similarly, silencing EIF3C could rescue the promoting effect when circPDE5A knockdown in PCa cells (Fig. 7F and Supplementary Fig. 6D). The WB analysis also demonstrated that the inhibitory effect of circPDE5A on the MAPK pathway could be reversed by elevating the EIF3C expression in PCa cells (Fig. 7G) and the elevated level of MAPK pathway markers caused by circPDE5A knockdown could be restored by silencing of EIF3C (Fig. 7H). These data reveal that EIF3C is a critical downstream target of circPDE5A in PCa.
FOXO4 regulates circPDE5A expression in prostate cancer
The biogenesis of circRNA can be regulated in both the transcriptional and posttranscriptional levels [36, 37]. Since circPDE5A was derived from the PDE5A pre-mRNA, we first analyzed the expression level of PDE5A mRNA in the TCGA database and in PCa tissues. The results revealed that the expression of PDE5A mRNA was both downregulated both in our PCa cohort and TCGA database (Supplementary Fig. 7A, B). Since circPDE5A and PDE5A mRNA were both downregulated in PCa tissues, we hypothesized that circPDE5A and PDE5A mRNA might be regulated at the transcriptional level. So, we hypothesized if transcriptional factors, which can initiate and regulate the transcription of genes, regulated circPDE5A expression in PCa. PROMO and JASPAR databases were used to predict the transcriptional factors which might regulate PDE5A gene expression (Fig. 8A). We found six transcriptional factors (including FOXO4, CEBPB, FOXP3, SP1, CEBPA, and STAT5a) with a high possibility of binding to the promoter region (− 2000 bp to TSS) of the PDE5A gene. Since the expression level of transcriptional factors should be positively or negatively correlated with circPDE5A in PCa, FOXO4, CEBPB, and STAT5a were selected as the candidate transcriptional factors that might regulate circPDE5A expression according to the correlation analysis results in 30 paired PCa tissues (Supplementary Fig. 7C). We, therefore, evaluated whether FOXO4, CEBPB, or STAT5a regulated circPDE5A expression in PCa cells. The results showed that knockdown of FOXO4, rather than CEBPB or STAT5a, could significantly downregulate the expression of circPDE5A (Supplementary Fig. 7D), suggesting that FOXO4 may be the only transcriptional factor that regulates circPDE5A expression. Then, we evaluated whether FOXO4 regulates both circPDE5A and PDE5A mRNA expression. The FOXO4 knockdown and overexpression efficiency was detected and shown in Supplementary Fig. 7E, F. The RT-qPCR assay showed that silencing of FOXO4 decreased the expression of circPDE5A and PDE5A mRNA significantly (Fig. 8B and Supplementary Fig. 7G), while FOXO4 overexpression increased circPDE5A and PDE5A mRNA expression (Fig. 8C and Supplementary Fig. 7H). Furthermore, by analyzing the promoter sequence of PDE5A, three potential FOXO4-binding motifs were found (Fig. 8D). CHIP-qPCR analysis in 22Rv-1 cells using the FOXO4 antibody showed that FOXO4 could bind directly to the PDE5A promoter (Fig. 8D). The dual-luciferase reporter assay revealed silencing of FOXO4 inhibited PDE5A promoter activity in C4-2B and 22Rv-1 cells (Fig. 8E). We then analyzed the FOXO4 expression in the SRRSH PCa cohort and TCGA database; the results showed that FOXO4 was downregulated in PCa tissues compared to paired normal specimens (Fig. 8F and Supplementary Fig. 7I). Correlation analysis of circPDE5A and FOXO4 in the PCa cohort demonstrated the positive correlation between the circPDE5A and FOXO4 (Fig. 8G). Collectively, these data show that FOXO4 transcriptionally regulated circPDE5A in PCa cells.
eIF4A3 regulates the expression of circPDE5A in prostate cancer
Previous studies showed that some RNA binding proteins could regulate the generation of circRNAs via binding to the circRNAs flanking intron regions [38, 39]. Using the online database Circinteractome (https://circinteractome.nia.nih.gov/), we found that eIF4A3 could specifically bind to the flanking regions of PDE5A (Supplementary Fig. 8A). We firstly verified the overexpression or knockdown efficiency of eIF4A3 in C4-2B and 22Rv-1 cells (Supplementary Fig. 8B). The RIP-qPCR assay confirmed that eIF4A3 could directly bind to the flanking region of PDE5A (Supplementary Fig. 8C). Moreover, overexpression of eIF4A3 increased circPDE5A expression, while silencing of eIF4A3 reduced the circPDE5A expression in C4-2B and 22Rv-1 cells significantly (Supplementary Fig. 8D, E). These results demonstrate that eIF4A3 regulates circPDE5A biogenesis in PCa.