m⁶A-Mediated Biogenesis of circDDIT4 Inhibits Prostate Cancer Progression by Sequestrating ELAVL1/HuR

Circular RNAs (circRNA) are a newly emerging class of endogenous non-coding RNAs that originate from parental linear genes via RNA polymerase II transcription and harbor covalently closed circular structures without poly(A) tails and 5′-3′ polarity (1). These circRNAs have been identified in various cell types and species, mainly as backspliced exons. Backspliced circularization requires both canonical splice signals and the canonical spliceosome machinery (2, 3). In addition, a majority of highly expressed circRNAs come from internal exons of pre-mRNAs and usually contain multiple exons, indicating that backsplicing is commonly coupled with canonical splicing (4). As a result, many circRNAs do not include the last exon, which typically contains the 3′-untranslated region (UTR) end.

Prostate cancer is the second most common cancer in men worldwide, accounting for a significant number of cancer-related deaths (5). Research has shown that certain circRNAs, such as circFOXO3 (6, 7), circ005276 (8), circAMOTL1 L (9), and circ-51217 (10), play crucial biological roles in prostate cancer progression. circRNAs act as transcriptional regulators to regulate gene expression (11) and often function as miRNA sponges to fine-tune the miRNA-mRNA regulatory axis (12). Yet, the precise molecular mechanisms underlying circRNA-mediated regulation of prostate cancer tumorigenesis and progression remain poorly understood.

N⁶-methyladenosine (m⁶A) is a widely observed RNA modification in eukaryotic cells. Several research studies have highlighted the dynamic and reversible nature of m⁶A modification, which plays a crucial role in RNA transcription, processing, splicing, degradation, and translation (13, 14). Disruption of m⁶A homeostasis is associated with varied diseases, especially metabolic disorders and cancers (15). m⁶A modifications are significantly present in various circRNAs as well (16). Wilms tumor 1 associated protein (WTAP), which is a splicing factor linked to WT1 protein, is also a component of the m⁶A methyltransferase complex. As m⁶A writers, WTAP aids in localizing methyltransferase-like 3 (METTL3) and methyltransferase-like 14 (METTL14) in nuclear speckles (17), Furthermore, it participates in binding to the 3′-UTR to create increased mRNA stability and acts as spliceosomes to control alternative splicing of pre-mRNA (18, 19). However, the effects of m⁶A modification on circRNA circularization remain an unknown area of study.

The RNA-binding protein ELAVL1, also known as human antigen R (HuR), is a ubiquitous protein that typically binds to AU-rich elements (ARE) in the 3′-UTR of target genes, enhancing RNA stability and increasing the half-life of target gene mRNAs (20, 21). However, the binding of ELAVL1 to circRNAs and the functional role of these interactions have been sparsely documented. Anoctamin 7 (ANO7), originally named NGEP (a new gene expressed in the prostate), is almost exclusively expressed in prostate epithelial cells (22). Recent studies indicate that ANO7 expression increases in more malignant prostate cancer and independently predicts prostate cancer prognosis, suggesting that it may serve as a biomarker for routine prostate cancer diagnosis (23, 24).

Previously, we identified circDDIT4 (Hsa_circ_0018744) as androgen-repressed and found that its expression was downregulated in prostate cancer tissues and serum samples. We have previously identified an androgen-repressed circDDIT4 (Hsa_circ_0018744; ref. 25). Unlike other circRNA types, circDDIT4 is derived from exons 2 and 3 of the DNA damage-inducible transcript 4 (DDIT4) gene containing three exons, which gives it a distinct biogenesis process, apart from backspliced circularization. In addition, the 3′-UTR can act as a critical competing endogenous RNA (ceRNA) by sponging miRNAs (26), but its function as an RNA-binding protein (RBP) sponge has been poorly understood. In this study, we provide compelling evidence that circDDIT4 acts as an RBP sponge at the 3′-UTR to downregulate the expression of ANO7, thus exerting an antioncogenic role in prostate cancer. Furthermore, we explored how WTAP-dependent m⁶A modification at the 3′-UTR impacts the formation of circDDIT4, shedding light on the regulatory mechanisms involved.

However, the biological function of circRNAs as RBP sponges is poorly understood. Here, we provide compelling evidence that circDDIT4 acts as an RBP sponge at the 3′-UTR to downregulate the expression of ANO7, thereby exerting an antioncogenic role in prostate cancer. Furthermore, we explored the impact of WTAP-dependent m⁶A modification at the 3′-UTR on ring formation of circDDIT4.

A total of 69 prostate cancer samples and their corresponding adjacent normal prostate tissues were obtained from patients who provided informed consent at Fudan University Shanghai Cancer Center (FUSCC). In addition, blood samples were collected from 26 of these patients with prostate cancer and 19 healthy donors who also gave their informed consent at the center. None of the patients had undergone any preoperative treatment, and all healthy volunteers had no prior history of cancer. The serum was isolated by centrifuging the samples at 3,000 × g for 10 minutes at 4°C. This study was in accordance with the recommendations of the Research Ethics Committee of FUSCC according to the provisions of the Declaration of Helsinki (as revised in Fortaleza, Brazil, October 2013). The protocol was approved by the Research Ethics Committee of FUSCC. Written informed consent for the use of clinical data was obtained from all the patients recruited in this study.

WPMY-1, LNCaP, 22Rv1, DU145, C4-2B, PC-3, C4-2, and HEK293T cells were procured from the Cell Bank of the Chinese Academy of Sciences and thoroughly authenticated through various tests, including Mycoplasma detection, DNA fingerprinting, isozyme detection, and cell vitality detection. WPMY-1, C4-2B, and HEK293T cells were cultured in DMEM (HyClone, SH3002), while LNCaP, 22Rv1, DU145, and PC-3 cells were maintained in RPMI1640 medium (HyClone, SH30809). FBS (BI, 04-001-1) was added to the medium before use, and all the cell lines were kept in a 5% CO₂ incubator at 37°C.

To construct circRNA plasmids [circDDIT4-pLCDH (circDDIT4-OE), circDDIT4-mut464/815-pLCDH], the pLCDH-ciR vector, containing a circular forming sequence, was utilized. For studying 3′-UTR–mediated circularization, the following plasmids were constructed using the linear expression vector pcDNA3.1: circDDIT4-pcDNA3.1, del-1/3-3′-UTR-pcDNA3.1, del-2/3-3′-UTR-pcDNA3.1, and del-all-3′-UTR-pcDNA3.1. Plasmid transfection was performed using HilyMax (Dojindo, H357) as per the manufacturer's instructions. circRNA plasmids [circDDIT4-pLCDH (circDDIT4-OE), circDDIT4-mut464/815-pLCDH]. The linear expression vector pcDNA3.1 was used to construct the following plasmids to study the 3′-UTR–mediated circularization: circDDIT4-pcDNA3.1, del-1/3-3′-UTR-pcDNA3.1, del-2/3-3′-UTR-pcDNA3.1, and del-all-3′-UTR-pcDNA3.1. Plasmid transfection was carried out using HilyMax (Dojindo, H357) according to the manufacturer's protocol.

GenePharma provided all siRNA oligonucleotides. Prostate cancer cells were transfected with either siRNA or a negative control (NC) at a final concentration of 50 nmol/L using HilyMax. Supplementary Table S1 provides the list of siRNA sequences used in the experiment.

The MagZol Reagent (Magen, R4801-02) was employed to lyse prostate cancer tissues and cultured cells, followed by total RNA extraction. The PrimeScript RT Reagent Kit (Takara, RR036A) was used for reverse transcription as per the manufacturer's instructions. Gene expression analysis was performed using the AceQ qPCR SYBR Green Master Mix (Vazyme, Q111-01) in triplicate on a LightCycler 480II (Roche) instrument. The β-actin gene was used for normalization of gene expression levels. The relative expression level of each gene was calculated by the 2^−ΔΔCt method. Supplementary Table S2 provides the list of primers used for qRT-PCR.

RNA extraction from 1 × 10⁷ DU145 cells was done following the manufacturer's instructions. Transcriptome sequencing library preparation and sequencing were carried out on an Illumina HiSeq X Ten platform at APExBIO Technology LLC to generate 100 bp paired-end reads. Gene expression levels were quantified using HTSeq v0.6.0 (RRID: SCR_005514) to count the number of reads mapped to each gene, and fragments per kilobase of transcript per million fragments mapped values were calculated for each gene.

Cells in culture were lysed using RIPA buffer (Magen) to extract the total protein, which was further separated using SDS-PAGE. The proteins were then transferred onto nitrocellulose membranes and incubated with antibodies against WTAP (Abcam, ab195380, RRID: AB_2868572) and ELAVL1/HuR (Santa Cruz Biotechnology, sc-52611, RRID: AB_675862). Horseradish peroxidase (HRP)-conjugated goat anti-rabbit IgG or HRP-conjugated goat anti-mouse IgG were employed as secondary antibodies. In addition, HRP-FLAG (Sigma-Aldrich, A8592, RRID: AB_439702) and HRP-Myc (Abcam, ab32, RRID: AB_303599) were also used.

To immunoprecipitate the FLAG-tagged proteins expressed ectopically, transfected cells were lysed in BC100 buffer 24 hours posttransfection. Whole-cell lysates were then subjected to immunoprecipitation (IP) using monoclonal anti-FLAG antibody-conjugated M2 agarose beads (Sigma-Aldrich, A8592) overnight at 4°C. After three washes with lysis buffer followed by two washes with BC100 buffer, the bound proteins were eluted using FLAG-peptide (Sigma-Aldrich, M8823) in BC100 for 3 hours at 4°C and resolved by SDS-PAGE. For endogenous protein IP, cells were lysed with 1× cell lysis buffer (Cell Signaling Technology, 9803). The supernatant was precleared with protein A/G beads (Yeasen, 36403), and then incubated with the designated antibody and protein A/G beads overnight at 4°C. The beads were washed five times with lysis buffer, resuspended in sample buffer and subjected to SDS-PAGE analysis.

Total RNA was extracted using MagZol Reagent and its concentration and purity were measured using Nanodrop 2000. Agarose gel electrophoresis was performed to confirm RNA structural integrity. Northern blot analysis was conducted as described previously, with 30 μg of total RNA denatured in formaldehyde and electrophoresed in a 1% agarose-formaldehyde gel. The RNAs were transferred onto a Hybond-N+ nylon membrane and hybridized with biotin-labeled DNA probes (Supplementary Table S4). A chemiluminescent biotin-labeled nucleic acid detection kit (Beyotime, D3308) was employed to visualize the bound RNAs.

A biotin-coupled probe was designed on the basis of the junction of circDDIT4 (Supplementary Table S4) to enable its detection. To extract cellular proteins, 0.1% NP-40 supplemented with fresh 100 U RNase inhibitor (Yeasen, 10603), 1 mmol/L DTT, ethylene diamine tetraacetic acid (EDTA)-free protease inhibitor cocktail (Roche, 04693132001), and phenylmethylsulfonylfluoride (PMSF) were used, followed by incubation with 3 μg of the biotin-coupled probes at 4°C for 2 hours. Subsequently, 30 μL of streptavidin-conjugated magnetic beads (Invitrogen, 11205D) were added to the cell lysate and incubated at 4°C for another hour. The proteins retrieved from this process were detected using mass spectrometry or Western blotting, as described previously.

Cellular proteins were extracted using 0.1% NP-40 supplemented with fresh 100 U RNase inhibitor, EDTA-free protease inhibitor cocktail, and PMSF. The extracted proteins were then incubated with 4 μg anti-WTAP/anti-ELAVL1 antibody or IgG at 4°C overnight. Afterward, 60 μL of protein A and protein G agarose beads (Millipore, 16-157/16-201) were added to the cell lysate and incubated at 4°C for 1 hour. Finally, the retrieved RNA was detected using qRT-PCR, as described above.

Total RNA was isolated using the MagZol reagent according to the manufacturer's protocol. The RNA was then fragmented in a 20 μL fragmentation buffer at 94°C for 5 minutes. Fragmented RNA (∼100 nt) was tumbled with 10 μg of anti-m⁶A antibody (Millipore, ABE572-I, RRID: AB_2892214) in 1 mL IP buffer at 4°C for 2 hours and subsequently immunoprecipitated with the same antibody in 1 mL IP buffer for another 2 hours at 4°C with rotation. The m⁶A RNA was washed thrice in 500 μL IP buffer and twice in 500 μL wash buffer for 5 minutes each at 4°C. Two hundred microliters of protein A and protein G agarose beads (Millipore, 16-157/16-201) were prepared by blocking with 0.5 mg/mL BSA at 4°C for 2 hours. The RNA-antibody IP reaction mixture was then incubated with blocked beads for an additional 2 hours at 4°C. The beads were washed thrice with 1× IP buffer and thrice with 1× wash buffer. The m⁶A-antibody immunoprecipitated RNA fragments were separated from the beads in 200 μL elution buffer at 50°C for 30 minutes. RNA was extracted by phenol-chloroform and ethanol precipitation and detected using qRT-PCR. Refer to Supplementary Table S3 for the primers used in methylated RNA immunoprecipitation (MeRIP)-qPCR.

Cell proliferation was assessed using CCK-8 (Dojindo, CK04), as previously described in our report (27). Briefly, transfected cells were maintained in 96-well plates at a density of 5,000 cells per well. At 0, 24, 48, and 72 hours posttreatment, 10 μL of CCK-8 was added to each well and then incubated for 2 hours at 37°C. The optical density was measured at 450 nm using an ELx808 microplate reader (Bio-Tek).

DU145 and LNCaP cells were cultured in 6-well plates, and then transfected with siRNA or plasmids using HilyMax transfection reagent. After 4 hours of transfection, the cells were harvested and the cell suspension was diluted to a concentration of 1,000–2,500 cells/mL. Next, approximately 100–250 cells/well were seeded in a 6-well plate, and this process was repeated thrice for each group. The plate was then placed inside an incubator for 10 days. After the incubation period, the cells were washed twice with 1× PBS. A volume of 500 μL of 4% paraformaldehyde was then added to the wells, and the cells were fixed at room temperature for 15 minutes. To stain the cells, 500 μL of diluted crystal violet staining solution was added to each well and incubated at room temperature for 15 minutes. Finally, the cells were visualized and photographed.

DU145, PC-3, and LNCaP cells were cultured in 6-well plates and transfected with siRNA or plasmids using HilyMax transfection reagent. Following a 48-hour incubation period, the cells were treated with the FITC-Annexin V Apoptosis Detection Kit (Dojindo, AD10) for 15 minutes at room temperature. Apoptosis was then quantified by utilizing a FACSCalibur flow cytometer (BD Biosciences).

A total of 10⁵ cells at a concentration of 100 μL were seeded into the upper chamber of a 24-well transwell (Corning, 353097), while 700 μL of medium containing 10% FBS was added to the lower chamber. After incubation for 24 hours, the cells were fixed with 4% paraformaldehyde for 15 minutes. The cells on the upper membrane surface were removed with a cotton swab, and the remaining cells were stained with 0.1% crystal violet and counted randomly in five different fields using an optical microscope (Olympus).

DU145 cells were cultivated in 6-well plates and transfected with siRNA or plasmids utilizing HilyMax transfection reagent. After 4 hours of transfection, the cells were cleaned with 1× PBS. Three smooth and vertical lines were created in the hole using a 1 mL pipette tip. After streaking, the cells were washed again with 1× PBS, and serum-free medium was added. Images were captured under a microscope at 0 hour. The 6-well plate was returned to the incubator for further cultivation, removed after 24 and 48 hours of incubation, and imaged again under a microscope.

To construct stably overexpressed circDDIT4 and control C4-2 cells, the targeted overexpressed circDDIT4 was inserted into the pZXE-LvSC01 vector containing puromycin. The recombinant vector was named pZXE-circDDIT4 or pZXE-NC and then transfected into HEK 293T cells. After 48 hours, the medium from packaging cells was collected, centrifuged at 4,000 × g for 10 minutes at room temperature to discard cell debris, and filtered through a 0.45-mm filter. This process resulted in the production of lentiviral stocks with a titer of more than 1 × 10⁸. Subsequently, C4-2 cells were transfected with Lenti-pZXE-circDDIT4 or Lenti-pZXE-NC. After 48 hours, the cells were selected with puromycin for 4 weeks.

Male nu/nu immunodeficient mice ages 3–4 weeks were purchased from the Shanghai Experimental Animal Center of Chinese Academy of Sciences (Shanghai, P.R. China), were bred and housed in our institutional pathogen-free mouse facilities. The use of mice was approved by the Institutional Review Board of Fudan University (Shanghai, P.R. China). Ten mice were randomly divided into two groups, and C4-2-pZXE-circDDIT4 or C4-2-pZXE-NC cell suspension was concentrated to 5 × 10⁶ cells/100 μL PBS and then injected into the flanks of the nude mice. After 20 days, the mice were euthanized, and the xenograft tumors were harvested, weighed, and photographed. The ethical approval for the study was obtained from the Animal Care Committee of Fudan University (Shanghai, P.R. China). All procedures for animal care and animal experiments were carried out in accordance with the guidelines of the Care and Use of Laboratory Animals proposed by the Animal Care Committee of Fudan University (Shanghai, P.R. China). The protocol was approved by the Science and Technology Commission of Shanghai Municipality [permit number: SYXK (hu) 2020-0032].

The presented data are expressed as mean ± SD from experiments conducted with a minimum of three replicates. Student t test was employed to determine significant differences between two groups based on the test conditions. In situations where there were multiple groups, one-way ANOVA was used to identify significant differences. Statistical significance was set at P < 0.05.

The data generated in this study are available upon request from the corresponding author.

In our previous study, we observed that DHT stimulation repressed the expression of circRNA circDDIT (25). However, the biological functions of circDDIT4 remain unknown. To investigate further, we conducted qRT-PCR assays to determine the levels of circDDIT4 expression in 69 prostate cancer tissues and corresponding adjacent normal tissues. The results showed that circDDIT4 was downregulated in prostate cancer samples compared with the adjacent normal tissues (Fig. 1A). Similarly, the expression levels of circDDIT4 in prostate cancer serum samples were significantly lower than those in serum samples from healthy donors (Fig. 1B). We also found that circDDIT4 was downregulated in prostate cancer cell lines when compared with normal human prostate epithelial WPMY-1 cells (Fig. 1C). These findings suggest that the expression of circDDIT4 is reduced in prostate cancer.

circDDIT4 is produced via splicing of the second and third exons of its parental DDIT4 pre-mRNA, which contains the coding sequence (CDS) region and the complete 3′-UTR of the parental gene and has a total length of 1,610 bp. To verify the sequence of the circDDIT4 linker position, we designed PCR primers based on the sequences on both sides of the circRNA junction (Fig. 1D). A Northern blot analysis confirmed that the length of circDDIT4 is consistent with the report in the circBank database (http://www.circbank.cn; Fig. 1E). In addition, nucleocytoplasmic separation assays indicated that circDDIT4 is mainly localized in the cytoplasm (Supplementary Fig. S1A).

To assess the stability of circDDIT4, we treated total RNA with RNase R in vitro and performed qRT-PCR to detect levels of DDIT4 mRNA and circDDIT4. Our data revealed that RNase R treatment led to significant degradation of DDIT4 mRNA but had a minimal effect on circDDIT4 (Fig. 1F). Moreover, we employed actinomycin D to measure RNA turnover, which is a well-known drug used to evaluate mRNA stability and inhibit the synthesis of new RNA. Our results showed that linear DDIT4 mRNA was significantly reduced with prolonged treatment with actinomycin D in LNCaP cells. However, circDDIT4 was relatively stable, and its half-life was longer than that of the parental gene (Fig. 1G). These findings are consistent with previous reports indicating that circRNA can resist exonuclease cutting and has a more stable existence with a longer half-life (28, 29).

To explore the biological roles of circDDIT4, we conducted functional studies in prostate cancer cell lines. First, we used an overexpression construct (circDDIT4-OE) and siRNAs specifically targeting the junction of circDDIT4. These experiments increased or decreased the expression of circDDIT4, respectively, but had no obvious effect on the expression of the parental DDIT4 gene at the mRNA and protein levels (Supplementary Fig. S2A–S2C). We then investigated the impact of circDDIT4 on the apoptosis of LNCaP, DU145, and PC-3 cells and discovered that either overexpression or knockdown of circDDIT4 significantly increased or decreased the apoptotic rates of prostate cancer cells, respectively (Fig. 2A). Colony formation assays showed that circDDIT4 reduced the growth of prostate cancer cells (Fig. 2B; Supplementary Fig. S2D). Moreover, wound healing and transwell assays demonstrated that circDDIT4 reduced the migration of prostate cancer cells (Fig. 2C; Supplementary Fig. S2E). Our in vivo experiments also revealed reduced tumor growth derived from circDDIT4-overexpressing C4-2 cells (Fig. 2D). Taken together, these results suggest that circDDIT4 exerts tumor suppressor activity in prostate cancer.

Next, we investigated the mechanisms underlying the tumor suppressor effect of circDDIT4. Naoko and colleagues previously reported that a subset of circRNAs containing open reading frames (ORF) may translate proteins (30), and in the case of circDDIT4, the complete ORF of DDIT4 is present. To test whether circDDIT4 encodes proteins, we constructed a plasmid (Supplementary Fig. S3A) to evaluate its coding ability. However, the results showed that the cyclized fragment (circDDIT4-FLAG) did not encode proteins or peptides (Supplementary Fig. S3B), ruling out the possibility that the tumor suppressor effect of circDDIT4 is due to protein translation.

We confirmed that circDDIT4 cannot translate proteins, and neither overexpression nor knockdown of circDDIT4 affected the mRNA or protein levels of the parental DDIT4 gene. This suggests that circDDIT4 does not act as a ceRNA to DDIT4. While bioinformatic predictions suggested that circDDIT4 may have 11 potential miRNA binding sites, further analysis revealed few binding sites and no positive regulation of miRNA target genes (Supplementary Fig. S3C and S3D; Supplementary Table S5). Therefore, we hypothesized that circDDIT4 exerts its tumor suppressor role by directly regulating the functions of its interacting proteins rather than acting as a miRNA sponge. To validate this, we performed biotin-labeled RNA pulldown assays followed by proteomic mass spectrometry analysis in LNCaP cells to identify potential binding partners of circDDIT4-3′-UTR. We identified 95 circDDIT4-3′-UTR binding proteins, five of which were established RBPs from the RBPDB database (http://rbpdb.ccbr.utoronto.ca), namely HNRNPD, SART3, ELAVL1, FUBP2, and PUF60. The binding capacities of these five RBPs to circDDIT4-3′-UTR were confirmed by RNA pulldown and Western blot analyses (Fig. 3A and B). We focused on ELAVL1 as a potential RBP for circDDIT4, given that it showed the strongest combination with circDDIT4. The decrease in the amount of ELAVL1 bound by biotin-circDDIT4-3′-UTR with increasing amounts of unlabeled circDDIT4-3′-UTR further supports this conclusion (Fig. 3C). The biotin-labeled circDDIT4-probe also validated the physical interaction of circDDIT4 with ELAVL1 (Fig. 3D). The RNA immunoprecipitation (RIP) results for ELAVL1 showed significant enrichment of both circDDIT4 and DDIT4 mRNA in prostate cancer cells (Fig. 3E), further supporting our hypothesis that ELAVL1 is a potential RBP for circDDIT4. In addition, following knockdown of ELAVL1, we observed a decrease in DDIT4 expression, but not in circDDIT4 expression, suggesting that DDIT4 stability requires ELAVL1 binding (Fig. 3F). The results of our study provide strong evidence for a specific and direct interaction between circDDIT4-3′-UTR and ELAVL1.

ELAVL1 is known to bind to AREs in the 3′-UTR of target genes to regulate mRNA stability (21). We hypothesized that circDDIT4 may competitively bind to ELAVL1 as an RBP sponge, reducing the stability of ELAVL1 target mRNAs. RBPmap analysis (http://rbpmap.technion.ac.il) revealed 20 potential ELAVL1 binding sites in the 848 bp circDDIT4-3′-UTR sequence, mainly concentrated near 460 and 810 bp (Supplementary Fig. S3E). Three mutant 3′-UTRs were designed on the basis of the sequence near these two regions (Fig. 3G), and RNA pulldown assays showed a decreased amount of ELAVL1 binding to the mutant circDDIT4-3′-UTR compared with the wild-type. Mutation of 464/815 bp significantly inhibited the interaction between ELAVL1 and circDDIT4 (Fig. 3H). RIP-qPCR assays showed that overexpression of circDDIT4 significantly increased the amount of ELAVL1 bound circRNA, but overexpression of circDDIT4-mut 464/815 had no significant effect on the interaction, further validating the binding authenticity (Fig. 3I).

ELAVL1 was found to be significantly overexpressed in prostate cancer, according to RNA-sequencing (RNA-seq) data in The Cancer Genome Atlas (TCGA) database (Supplementary Fig. S4A). Knockdown of ELAVL1 inhibited proliferation and promoted apoptosis of prostate cancer cells (Supplementary Fig. S4B and S4C), suggesting that ELAVL1 acts as an oncogene in prostate cancer. To investigate whether circDDIT4 exerts a tumor suppressor effect through its interaction with ELAVL1, we evaluated its effects on cell proliferation, colony formation, and apoptosis after overexpression of circDDIT4 or circDDIT4-mut 464/815. The results indicated that circDDIT4-mut 464/815 almost completely eliminated the tumor suppressor activity of the wild-type (Fig. 4A–C). Thus, these findings suggest that circDDIT4 may play a crucial role in suppressing prostate cancer by acting as a sponge for ELAVL1.

To identify downstream target genes coregulated by circDDIT4 and ELAVL1, we performed RNA-seq to identify transcriptome profiles following overexpression of circDDIT4-OE and circDDIT4-mut 464/815, as well as knockdown of ELAVL1. Venn diagrams of the differentially expressed mRNAs in the three datasets showed a high overlap, with the majority of the differentially expressed mRNAs being downregulated. Furthermore, there was a positive correlation in gene expression between the datasets, indicating a similar pattern of gene regulation (Fig. 5A; Supplementary Table S6). These findings support our hypothesis that circDDIT4 reduces the stability of ELAVL1 target mRNA through sequestration of ELAVL1, rather than acting solely as a miRNA sponge. Moreover, circDDIT4-mut 464/815 only partially inhibited the activity of circDDIT4, further highlighting the importance of the circDDIT4–ELAVL1 interaction in regulating downstream target genes.

We identified 20 candidate target genes that were coregulated by circDDIT4-OE and si-ELAVL1, but not by circDDIT4-mut 464/815 (Fig. 5B). To determine which of these interacted with ELAVL1, we performed RIP-qPCR assays and found that ANO7, CLK1, LGALS4, and PMAIP1 all showed interaction with ELAVL1 (Supplementary Fig. S5A). The expression of these genes was measured after overexpression of circDDIT4-OE or circDDIT4-mut 464/815, or knockdown of ELAVL1 (Supplementary Fig. S5B and S5C). Among them, ANO7 was significantly downregulated by transfection with circDDIT4-OE and si-ELAVL1. Therefore, we selected ANO7 for further investigation. ANO7 is known to be specifically expressed in prostate tissues and has been suggested as a potential biomarker for predicting aggressiveness in prostate cancer (23). Analysis of TCGA and GSE6919 datasets revealed that ANO7 mRNA expression was significantly increased in prostate cancer (Supplementary Fig. S5D and S5E). Knockdown of ANO7 partially restored the cancer-promoting effect of circDDIT4 knockdown in prostate cancer cells (Fig. 5C–E; Supplementary Fig. S5F and S5G), indicating that ANO7 is a downstream target regulated by the circDDIT4-ELAVL1 axis.

As circDDIT4 comprises the last exon and the complete 3′-UTR, it is likely that its biogenesis differs from circRNAs derived from reverse splicing exons. To investigate whether the 3′-UTR is essential for loop formation, we generated circRNA expression plasmids containing the complete circDDIT4 3′-UTR or different combinations of truncated 3′-UTRs using the linear expression vector pcDNA3.1 (circDDIT4-pcDNA3.1, del-1/3-3′-UTR-pcDNA3.1, del-2/3-3′-UTR-pcDNA3.1, del-all-3′-UTR-pcDNA3.1). Following transfection, we examined the expression levels of circDDIT4 and observed that the full-length 3′-UTR was necessary for looping, while truncated 3′-UTRs nearly abolished looping (Fig. 6A).

Prediction of potential m⁶A modification sites on the DDIT4 pre-mRNA flanking sequence of the loop junction site, including intron 1, exon 2, and the 3′end of exon 3, yielded six sites (25).To investigate the role of WTAP in regulating circDDIT4 ring formation, we produced circDDIT4-5′ flanking and circDDIT4-3′ UTRs via in vitro transcription. RNA pulldown assays demonstrated that both RNA fragments could bind to WTAP (Fig. 6B). Similarly, RIP-qPCR assay results revealed that WTAP could interact with both the circDDIT4-5′ flanking and circDDIT4-3′-UTR regions (Fig. 6C). These findings suggest that WTAP may play a role in circDDIT4 ring formation. We used the STRING database to query the WTAP-interacting protein network (Supplementary Fig. 6SA) and found that METTL3, METTL14, and WTAP formed the m⁶A methyltransferase complex involved in regulating m⁶A methylation modifications (Supplementary Fig. S6B). Co-IP experiments confirmed the interaction between WTAP, METTL3, and METTL14 (Fig. 6D and E). Therefore, we speculate that WTAP modulates m⁶A modifications, thereby affecting pre-mRNA splicing and participating in the regulation of circRNA ring formation.

Through bioinformatic analysis, we identified six potential m⁶A modification sites in the circDDIT4–5′ flanking and circDDIT4-3′-UTR regions (Supplementary Fig. S6C). To verify these predictions, we designed several sets of specific primers for MeRIP-qPCR experiments. The results showed that the RNA fragments with candidate m⁶A modification sites had enriched m⁶A signals, and the fragments containing more modification sites exhibited stronger m⁶A antibody binding capacity (Fig. 6F). Moreover, knockdown of WTAP led to a decreased m⁶A modification on circDDIT4, resulting in fewer RNA fragments being enriched by the m⁶A antibody (Fig. 6G). The data suggest that m⁶A modification indeed exists in the circDDIT4-5′ flanking and circDDIT4-3′-UTR regions.

WTAP, METTL3, and METTL14 are essential components of the m⁶A methyltransferase complex, while FTO and ALKBH5 act as demethylases. To investigate the effects of these m⁶A modification–related regulatory proteins on circDDIT4 ring formation, we knocked down or overexpressed them. Our results showed that overexpression of METTL3 or METTL14 in prostate cancer cells promoted circDDIT4 ring formation (Fig. 6H). Conversely, knockdown of METTL3 or METTL14 by siRNAs reduced circDDIT4 ring formation but did not affect parental DDIT4 gene expression (Fig. 6I). In addition, knockdown of FTO expression (but not ALKBH5) increased the amount of circDDIT4 looping (Fig. 6J). Notably, m⁶A modification did not impact the stability of circDDIT4 (Supplementary Fig. S6D). In conclusion, our findings suggest that m⁶A modification promotes circDDIT4 circularization, which is mediated by the methyltransferase complex consisting of WTAP/METTL3/METTL14, while FTO decreases circDDIT4 levels through demethylation.

To investigate the relationship between METTL3/METTL14 and prostate cancer, we examined their expression levels in TCGA dataset. Our analysis revealed that METTL3 and METTL14 were downregulated in prostate cancer tissues (Supplementary Fig. S6E). Next, we conducted rescue experiments to determine whether m⁶A modification affects prostate cancer cell phenotype via regulation of circDDIT4 biogenesis. Our results demonstrated that knockdown of METTL3/14 inhibited prostate cancer cell proliferation and promoted apoptosis. However, overexpression of circDDIT4 reversed the inhibitory effect of METTL3/14 knockdown on cell proliferation and the promotion of apoptosis (Supplementary Fig. S6F–S6I). These findings suggest that decreased expression of METTL3 or METTL14 in prostate cancer could lead to reduced m⁶A modification and subsequently inhibit circDDIT4 expression, thereby affecting prostate cancer progression.

In our previous study, we identified a group of androgen-responsive circRNAs in prostate cancer and suggested the involvement of RBPs such as WTAP and TNRC6 in circRNA biogenesis (25). To investigate the roles of specific circRNAs in prostate cancer, we selected several candidates for further validation. Here, we discovered that circDDIT4 is downregulated in prostate cancer and functions as a tumor suppressor in disease progression. Notably, analysis of different prostate cancer cells revealed that circDDIT4 expression is regulated in an androgen-independent manner. Thus, we delved deeper to understand the underlying mechanism of its biogenesis and found evidence suggesting that m⁶A modification plays a role in mediating the biogenesis of circDDIT4.

Several studies have previously shown that DDIT4 promotes proliferation and tumorigenesis in both prostate cancer and gastric cancer (31, 32). In contrast, our study reveals a tumor suppressor role for circDDIT4 in prostate cancer. It is important to note that overexpression or knockdown of circDDIT4 did not impact the levels of DDIT4, suggesting that circDDIT4’s function is not dependent on the modulation of its parental gene, DDIT4.

While there are numerous reports on non-coding RNA with miRNA sponge function (12, 33), few have explored their role as an RBP sponge (34). ELAVL1 is an established posttranscriptional regulator that promotes cancer progression and is often overexpressed in prostate cancer (35). Existing research has found that circPABPN1 binds to ELAVL1 and prevents its binding to PABPN1 mRNA and lowers PABPN1 translation (36). Another study showed that circAGO2 physically interacts with ELAVL1 to facilitate its activation and enrichment on the 3′-UTR of target genes, which reduces AGO2 binding and represses AGO2/miRNA-mediated gene silencing (37). However, further study is required to fully understand the effect of circRNAs on ELAVL1.

In our study, we provide strong evidence that circDDIT4, which contains the CDS region and 3′-UTR of DDIT4 mRNA, can function as an RBP sponge. The long circDDIT4-3′-UTR contains multiple binding sites for ELAVL1, allowing it to act as a sponge and sequester ELAVL1, thereby downregulating its downstream target genes. Our findings suggest that the sponge effect of circRNAs on ELAVL1 is more pronounced than that of linear DDIT4 mRNA.

ANO7 is a promising biomarker in predicting prostate cancer aggressiveness (23, 24) and our study shows that circDDIT4 can act as an RBP sponge to downregulate the expression of both ELAVL1 and ANO7, coregulated downstream target genes. Furthermore, we found that circDDIT4 was downregulated, while ELAVL1 and ANO7 were significantly upregulated in prostate cancer. Knockdown of ANO7 partially restored the cancer-promoting effect of circDDIT4 knockdown in prostate cancer cells, further supporting circDDIT4 as a potential biomarker or therapeutic target for prostate cancer.

circRNAs can be classified into four categories based on their biogenesis patterns, all of which involve intron-exon backsplicing, where downstream 3′ splice donors are covalently linked to upstream 5′ splice acceptors in reverse order (38). In addition, studies have suggested that m⁶A modification plays a crucial role in the backsplicing of a specific subset of circRNAs (39).

We propose a new mode of circRNA biogenesis that involves the back connection of the 3′-UTR end and upstream 5′ splice acceptor site, which is distinct from intron-exon backsplicing. One example of this mode is circDDIT4. Our findings indicate that WTAP binds to the loop flanking elements of circDDIT4 and promotes its formation (25). WTAP is a splicing factor related to the WT1 protein, which binds to 3′-UTRs and contributes to increased mRNA stability (18, 19). In addition, WTAP is a component of the m⁶A methyltransferase complex, which plays an important role in m⁶A modification as a “writer” alongside METTL3 and METTL14 (40). Previous studies have shown that m⁶A modification regulates circRNA translation (16, 39) and degradation (41). We suggest that WTAP may facilitate the loop formation of circDDIT4 by mediating m⁶A modification in its flanking sequence. Our MeRIP assay data confirm the presence of multiple m⁶A modification sites in the loop-forming flanking sequence of circDDIT4, and the expression of circDDIT4 is regulated by components of the m⁶A modification machinery, indicating that m⁶A modification is involved in the biogenesis of circDDIT4.

In summary, we propose a model in which m⁶A modification mediates the biogenesis of circDDIT4, which binds to ELAVL1 and acts as an RBP sponge to downregulate the expression of ELAVL1 target mRNAs, including ANO7, thus exerting a tumor suppressor effect on the progression of prostate cancer (Fig. 7).

No disclosures were reported.

Z. Kong: Formal analysis, validation, methodology, writing–original draft. Y. Lu: Methodology, writing–original draft. Y. Yang: Software, formal analysis. K. Chang: Software, validation. Y. Lin: Resources, visualization. Y. Huang: Resources, data curation, project administration. C. Wang: Supervision, investigation. L. Zhang: Visualization. W. Xu: Supervision, funding acquisition, investigation. S. Zhao: Conceptualization, funding acquisition, writing–review and editing. Y. Li: Conceptualization, funding acquisition, writing–review and editing.

This work was supported by grants from the National Key Research and Development Program of China (no. 2018YFA0800300 to S. Zhao, 2018YFA0801300 to W. Xu), the National Natural Science Foundation of China (no. 31821002, 31930062 to S. Zhao, 31871432 to W. Xu), the Shanghai Science and Technology Development Foundation (20ZR1404500 to Y. Li). We thank Dr. Jun Jiang for advice on bioinformatic analysis of RNA-seq data.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.