Comprehensive Characterization of Androgen-Responsive circRNAs in Prostate Cancer

The androgen receptor (AR) signaling pathway plays an important role in the initiation and progression of prostate cancer. Circular RNAs (circRNAs), the novel noncoding RNAs without 5′ to 3′ polarity or 3′ poly (A), play an important role in multiple diseases. However, the potential roles of androgen-responsive circRNAs in prostate cancer remain unclear. In this study, we identified 3237 androgen-responsive circRNAs and 1954 androgen-responsive mRNAs after dihydrotestosterone (DHT) stimulation using microarray. Among them, the expression of 1296 androgen-responsive circRNAs was consistent with that of their parent genes, and we thought AR might regulate the expression of these circRNAs at the transcriptional level. In addition, 1941 circRNAs expression was not consistent with their parent genes, and we speculated that AR may regulate the expression of those circRNAs at the posttranscriptional level through affecting alternative splicing. Analyzing the androgen-responsive circRNAs regulated at the posttranscriptional level, we identified two key RNA binding proteins (RBPs), WTAP and TNRC6, using the circInteractome database, which may play important role in the biogenesis of androgen-responsive circRNAs. Furthermore, we explored the potential biological functions and predicted the molecular mechanisms of two dysregulated circRNAs (circNFIA and circZNF561) in prostate cancer. In this study, we revealed that circNFIA was upregulated in prostate cancer tissues and plasma samples from patients with prostate cancer; circNFIA may play an oncogenic role in prostate cancer. In contrast, circZNF561 was downregulated and may act as a tumor suppressor in prostate cancer. Our results suggest that androgen-responsive circRNAs might regulate the progression of prostate cancer and could be novel diagnostic biomarkers.


Introduction
Prostate cancer is the second most frequently diagnosed cancer in men worldwide, and 1.4 million estimated cases were diagnosed in 2020 [1]. The incidence rates of prostate cancer increased in most countries, such as Sweden, Thailand, China, and Lithuania [2]. Prostate cancer is hormone-responsive, and castration leads to tumor regression in prostate cancer patients [3]. Androgen deprivation therapy (ADT) is the main treatment for advanced and metastatic prostate cancer [4]. However, almost all patients develop castration-resistant prostate cancer (CRPC) approximately 2-3 years after ADT, at which time, the serum PSA levels increase again and the disease rapidly worsens [5,6].
The androgen receptor (AR) signaling pathway plays a vital role in the diagnosis and treatment of prostate cancer. CRPC, previously defined as hormone-independent prostate cancer, is now known to still be androgen dependent. AR amplification [7], AR mutations [8], mutations in coactivators or corepressors [9][10][11], androgen-independent AR Life 2021, 11, 1096 2 of 17 activation [12], and alternative androgen production [13,14] contribute to the progression of CRPC. Understanding the role of the AR signaling pathway in the progression of prostate cancer is important to develop future therapies.
With advances in high-throughput sequencing technologies and bioinformatics, a large number of circRNAs have been gradually discovered in mammalian cells [15]. CircRNAs, a novel type of noncoding RNAs with a covalent closed-loop structure, are derived from pre-mRNAs via back-splicing and have no 5 cap or 3 poly (A) [16,17]. CircRNAs are more stable and protected against RNA exonucleases in vivo. Accordingly, recent studies have suggested that circRNAs could be promising biomarkers in multiple diseases [18]. In addition, an increasing number of studies have found that circRNAs play important roles in diseases, such as diabetes [19], cardiovascular disease [20], chronic inflammatory diseases [21], and neurological diseases [22], especially in a variety of tumors. For example, circSMARCA5 acts as a tumor suppressor in liver cancer [23], malignant glioma [24], nonsmall cell lung cancer [25], and plays an oncogenic role in prostate cancer [26]. CircRNAs act as microRNA sponges to regulate the development and progression of disease. For example, CDR1as, also called ciRS-7, was reported to have approximately 70 conserved miR-7 binding sites and to act as a potential miR-7 sponge to contribute to several biological processes [27,28]. Recently, our group also found that circFOXO3 could promote prostate cancer progression by competitive binding of miR-29a-3p [29]. However, the biogenesis and biological functions of androgen-responsive circRNAs in prostate cancer still need to be further explored.
In this study, we identified circRNAs as diagnostic and therapeutic markers and explored the potential functions of circRNAs in prostate cancer. We identified a set of androgen-responsive circRNAs and two important regulatory factors, WTAP and TNRC6, which may involve in circRNA biogenesis. In addition, we explored the potential biological functions and predicted the molecular mechanism of two androgen-responsive circRNAs (circNFIA and circZNF561), which may act as promising biomarkers for the diagnosis and treatment of prostate cancer.

Human Prostate Cancer Samples
Prostate cancer patient tissue samples and matched adjacent normal samples were obtained from patients who underwent radical prostatectomy at Fudan University Shanghai Cancer Center. Serum samples from newly diagnosed prostate cancer patients (n = 26) and healthy donors (n = 19) who provided informed consent were obtained at Center named above. These patients did not receive any preoperative treatment. The study was approved by the Research Ethics Committee of Fudan University Shanghai Cancer Center. All the samples were collected and used for gene expression analysis by qPCR.

Microarray Analysis
LNCaP cells were starved for 3 days and treated with 100 nM DHT for 8 h. Then, the cells were collected, and microarray analysis was performed as previously described [30] (Shanghai Biotechnology Corporation, Shanghai, China). In brief, total RNA was amplified, labeled with a Low Input Quick Amp Labeling Kit and then purified using RNeasy mini kit. Each slide was hybridized with 1.65 µg Cy3-labeled RNA for 17 h and then scanned by an Agilent microarray scanner (Agilent technologies, Santa Clara, CA, USA). The data were extracted using Feature Extraction software 10.7 (Agilent Technologies). The raw data were normalized by the quantile algorithm and limma package in R (Version 3.6.2).

Cell Culture and Treatment
LNCaP, DU145, 22Rv1, PC-3, and WPMY-1 cells were purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA). All the cells were cultured in RPMI-1640 or DMEM medium supplemented with 10% fetal bovine serum (ExCell Bio, Shanghai, China). The cells were maintained in an incubator at 37 • C containing 5% CO 2 . The androgen treatment assays were performed as described previously [31]. Briefly, LNCaP cells were starved for 3 days and then treated with DHT. The cells were collected and used in the following detection.

RNA extraction and qPCR
Total RNA was isolated from homogenized prostate cancer tissue or cultured cells using MagZol Reagent (Magen, Guangzhou, China) as previously described [29]. Total RNA was reverse transcribed using the PrimeScript ™ RT reagent Kit (Takara, Japan) according to the manufacturer's protocol. The relative expression of candidate genes was determined using AceQ qPCR SYBR Green Master Mix (Vazyme, Shanghai, China) on LightCycler ® 480II (Roche, Basel, Switzerland). The primers used for the qPCR analysis are listed in Supplementary Table S1.

Transfection
Cells were seeded in 6-well plates until cell confluence reached 60%-80%. Liposomal cocktails with siRNA (GenePharma, Shanghai, China) (50 nM final) were generated with HilyMax (Dojindo Laboratories, Kumamoto, Japan) in Opti-MEM (Invitrogen, Carlsbad, CA, USA) according to the protocol. The transfected cells were incubated for 48 h and then used for gene expression analysis or other experiments. The sequences of the specific siRNAs are shown in Supplementary Table S1.

Cell Apoptosis
Cells were cultured in 6-well plates after transfection. Cell apoptosis was measured using the Annexin V-FITC apoptosis kit (Dojindo Laboratories, Kumamoto, Japan). After incubation at room temperature for 15 min, cell apoptosis was analyzed on a FACSCalibur flow cytometer (Becton, Dickinson and Company, Franklin Lakes, NJ, USA).

Cell Proliferation
Cell Counting Kit-8 (Dojindo Laboratories, Kumamoto, Japan) was used to assess cell proliferation according to the manufacturer's protocol. Briefly, transfected cells were seeded in 96-well plates and cultured in 100 µL medium. 10 µL of CCK-8 solution was added to each well at 0, 24, 48, and 72 h and further incubated for 2 h to allow the colorimetric reaction to occur. The absorbance at 450 nm was measured by a Microplate Reader ELx808 (Bio-Tek, Winooski, VT, USA) to determine cell viability.

Cell Cycle
Transfected cells were cultured in 6-well plates for 48 h. Then, the cells were collected and treated with Triton X-100 (0.03%, Sigma, Darmstadt, Germany) and propidium iodide (PI, 50 ng/mL, Beyotime, Shanghai, China) for 15 min. Cell cycle analysis was performed by a FACSCalibur flow cytometer (Becton, Dickinson and Company, Franklin Lakes, NJ, USA).

Construction of ceRNA Network
The circBank (http://www.circbank.cn/ (accessed on 18 September 2021)) was used to predict miRNA binding sites. We found 59 miRNAs targeted to circNFIA and 26 miRNAs targeted to circZNF561. The networks were drawn using Cytoscape 3.0. Following this, we identified the target genes of miRNAs interacting with circNFIA or circZNF561 using the miRDB database. The target score ≥ 90 was chosen to construct circRNA-miRNA-mRNA networks using Cytoscape 3.0.

Statistical Analysis
The numerical data are presented as the mean ± standard deviation (SD) of at least three determinations. Statistical comparisons between groups of normalized data were performed using t-tests or Mann-Whitney U-tests according to the test conditions. A p < 0.05 was considered statistically significant with a 95% confidence level.

Identification of Androgen-Responsive circRNAs in Prostate Cancer
To identify the differentially expressed mRNAs and circRNAs after DHT stimulation, we performed high-throughput microarray analysis and the expression of 88,751 circRNAs and 18,854 mRNAs was detected. Of these, 3237 circRNAs and 1954 mRNAs were differentially expressed. As shown in Figure 1A,B, 2295 circRNAs and 969 mRNAs were upregulated and 942 circRNAs and 985 mRNAs were downregulated after DHT stimulation. Interestingly, the number of upregulated circRNAs was approximately 2.5 times that of downregulated circRNAs, while the number of upregulated mRNAs was close to that of downregulated mRNAs. The result suggests that the AR signaling pathway may have differences regulating the expression of mRNAs and circRNAs and circRNAs may not be meaningless splicing byproducts.

Statistical Analysis
The numerical data are presented as the mean ± standard deviation (SD) of at least three determinations. Statistical comparisons between groups of normalized data were performed using t-tests or Mann-Whitney U-tests according to the test conditions. A p < 0.05 was considered statistically significant with a 95% confidence level.

Identification of Androgen-Responsive circRNAs in Prostate Cancer
To identify the differentially expressed mRNAs and circRNAs after DHT stimulation, we performed high-throughput microarray analysis and the expression of 88,751 circRNAs and 18,854 mRNAs was detected. Of these, 3237 circRNAs and 1954 mRNAs were differentially expressed. As shown in Figure 1A,B, 2295 circRNAs and 969 mRNAs were upregulated and 942 circRNAs and 985 mRNAs were downregulated after DHT stimulation. Interestingly, the number of upregulated circRNAs was approximately 2.5 times that of downregulated circRNAs, while the number of upregulated mRNAs was close to that of downregulated mRNAs. The result suggests that the AR signaling pathway may have differences regulating the expression of mRNAs and circRNAs and circR-NAs may not be meaningless splicing byproducts.
We also analyzed the distribution of the circRNAs on the human chromosomes and we found that the number of upregulated circRNAs derived from each chromosome was greater than the number of downregulated circRNAs ( Figure 1C). In addition, most circR-NAs were derived from exons (CDS region, 3′UTR, and 5′UTR) and a few circRNAs were derived from introns, intergenic regions, and antisense strands ( Figure 1D). We also analyzed the distribution of the circRNAs on the human chromosomes and we found that the number of upregulated circRNAs derived from each chromosome was greater than the number of downregulated circRNAs ( Figure 1C). In addition, most circRNAs were derived from exons (CDS region, 3 UTR, and 5 UTR) and a few circRNAs were derived from introns, intergenic regions, and antisense strands ( Figure 1D).

Characteristics of Androgen-Responsive circRNA Expression
Both mRNAs and circRNAs are derived from pre-mRNA by alternative splicing. We analyzed the expression of circRNAs and their parent genes in the microarray data. The results showed that only 51.99% of the 2087 circRNAs derived from androgen-upregulated mRNAs were upregulated and 18.22% of 1158 circRNAs derived from androgen-downregulated mRNAs were downregulated ( Figure 3A,B). The results showed that only part of the circRNAs derived from androgen-induced or -reduced genes were upregulated or downregulated after DHT treatment, which suggested that the circRNAs and their parent genes had different regulated patterns.
According to the expression of mRNAs and circRNAs after DHT treatment, we divided androgen-responsive circRNAs into six groups ( Figure 3D). We focused on the circRNAs in groups a and d that were regulated by androgen at the transcriptional level and the circRNAs in groups b and e that were regulated by androgen at the posttranscriptional level. Among them, the expression of hsa-circ-0076151, hsa-circ-0001055, hsa-circ-0005035, hsa-circ-0005079, and hsa-circ-0073627 was increased after DHT treatment (group a) (Figure 2), while hsa-circ-0008326 expression was decreased after DHT treatment (group d) (Figure 2). According to the expression of mRNAs and circRNAs after DHT treatment, we divided androgen-responsive circRNAs into six groups ( Figure 3D). We focused on the circRNAs in groups a and d that were regulated by androgen at the transcriptional level and the circRNAs in groups b and e that were regulated by androgen at the posttranscriptional level. Among them, the expression of hsa-circ-0076151, hsa-circ-0001055, hsa-circ-0005035, hsa-circ-0005079, and hsa-circ-0073627 was increased after DHT treatment (group a) (Figure 2), while hsa-circ-0008326 expression was decreased after DHT treatment (group d) (Figure 2).

Androgen Regulated circRNAs at tRanscriptional Level
To further verify the expression levels of circRNAs and their parent genes, we detected the expression of the above six circRNAs (in groups a and d) and their parent genes at a series of time points after DHT treatment ( Figure 4A). The results showed that the expression level of these six circRNAs was basically the same as that of their parent genes. The expression of hsa-circ-0076151, hsa-circ-0001055, hsa-circ-0005035, hsa-circ-0005079, hsacirc-0073627, and their parent genes FKBP5, AFF3, IGF1R, MBOAT2, KCNN2 was increased after DHT treatment, while hsa-circ-0008326 and RFX3 expression was decreased.

Androgen Posttranscriptional Regulated circRNA Expression
We identified a large number of differentially expressed circRNAs, but their host genes were not differentially expressed (group b and group e). We speculated that the androgen may affect the alternative splicing and regulate these circRNAs expression in the posttranscriptional level. To further explore the regulation of circRNA expression, we To test the hypothesis that the expression of these six circRNAs were transcriptionally regulated by androgen, we predicted the AREs within the upstream 10 kb from the transcription start site using the hTFtarget database. As shown in Figure 4B, we found many AR binding sites in those genes' promoters. Furthermore, a ChIP-qPCR assay was performed using anti-AR antibodies to pull down DNA fragments with AREs in LNCaP cells with or without DHT stimulation. Consistently, AR was directly associated with the FKBP5, AFF3, RFX3, IGF1R, MBOAT2, and KCNN2 promoters after DHT treatment ( Figure 4C). The results showed that AR might directly regulate the expression of FKBP5, AFF3, RFX3, IGF1R, MBOAT2, KCNN2, and their corresponding circRNAs.

Androgen Posttranscriptional Regulated circRNA Expression
We identified a large number of differentially expressed circRNAs, but their host genes were not differentially expressed (group b and group e). We speculated that the of each RBP. The higher the µ value was, the more likely it was that the RBPs interacted with androgen-induced circRNAs. Conversely, the smaller the µ value was, the more likely it was that the RBPs interacted with androgen-reduced circRNAs ( Figure 5A). The results showed that WTAP, TAF15, FOX2, and RBPMS potentially regulated the androgen-downregulated circRNAs, while TNRC6, C17ORF85, AUF1, FXR1, MOV10, FXR2, and ZC3H7B potentially regulated the androgen-upregulated circRNAs. Among those RBPs, WTAP, and TNRC6 were the most significant ( Figure 5A). Next, we detected the expression of circRNAs potentially associated with WTAP or TNRC6 after knockdown of WTAP or TNRC6, respectively. As shown in Figure 5B, after knockdown of WTAP, the expression of hsa_circ_0012079, hsa_circ_0087714, and hsa_circ_0018744 was significantly decreased. Similarly, after knockdown of TNRC6, the expression of hsa_circ_0090744, hsa_circ_0076815, hsa_circ_0049519, hsa_circ_0014100, and hsa_circ_0061590 was significantly decreased ( Figure 5C). We initially revealed that WTAP and TNRC6 may be involved in those circRNA's biogenesis. ) of each RBP. The higher the μ value was, the more likely it was that the RBPs interacted with androgen-induced circRNAs. Conversely, the smaller the μ value was, the more likely it was that the RBPs interacted with androgen-reduced circRNAs ( Figure 5A). The results showed that WTAP, TAF15, FOX2, and RBPMS potentially regulated the androgen-downregulated circRNAs, while TNRC6, C17ORF85, AUF1, FXR1, MOV10, FXR2, and ZC3H7B potentially regulated the androgen-upregulated circRNAs. Among those RBPs, WTAP, and TNRC6 were the most significant ( Figure 5A). Next, we detected the expression of circRNAs potentially associated with WTAP or TNRC6 after knockdown of WTAP or TNRC6, respectively. As shown in Figure 5B, after knockdown of WTAP, the expression of hsa_circ_0012079, hsa_circ_0087714, and hsa_circ_0018744 was significantly decreased. Similarly, after knockdown of TNRC6, the expression of hsa_circ_0090744, hsa_circ_0076815, hsa_circ_0049519, hsa_circ_0014100, and hsa_circ_0061590 was significantly decreased ( Figure 5C). We initially revealed that WTAP and TNRC6 may be involved in those circRNA's biogenesis.

The Expression of circNFIA and circZNF561 in Prostate Cancer
To explore the biological roles of androgen-responsive circRNAs in prostate cancer, we identified differentially expressed circRNAs in 30 pairs of prostate cancer tissue samples and corresponding adjacent normal tissues by qPCR. Two circRNAs, hsa-circ-0012755 (also called circNFIA) and hsa-circ-0049154 (also called circZNF561), were significantly dysregulated in prostate cancer. The results showed that circNFIA expression was increased and circZNF561 expression was decreased in prostate cancer tissue samples compared to adjacent normal tissues ( Figure 6A,B). Furthermore, we also detected the expression of circNFIA and circZNF561 in prostate cancer cells and plasma samples. As shown in Figure 6C,D, the expression of circNFIA was significantly upregulated in prostate cancer cell lines (LNCaP, LNCaP-AI, DU145, 22Rv1, and PC-3 cells), while circZNF561 was downregulated in prostate cancer cells. The expression of circNFIA and circZNF561 in plasma samples was further detected by qPCR. Consistent with the above results, circNFIA was significantly overexpressed in the plasma samples of prostate cancer patients, while the expression of circZNF561 was decreased ( Figure 6E,F). These results suggest that circNFIA and circZNF561 may play roles in the progression of prostate cancer.

The Biological Functions of circNFIA and circZNF561 in Prostate Cancer.
Next, we knocked down circNFIA and circZNF561 expression in the DU145 and LNCaP cell lines and explored their possible biological role in prostate cancer. As shown in Figure 7A,B, sicircNFIA and sicircZNF561 significantly decreased the expression of

The Biological Functions of circNFIA and circZNF561 in Prostate Cancer
Next, we knocked down circNFIA and circZNF561 expression in the DU145 and LNCaP cell lines and explored their possible biological role in prostate cancer. As shown in Figure 7A,B, sicircNFIA and sicircZNF561 significantly decreased the expression of circNFIA and circZNF561. The CCK-8 assay showed that knockdown of circNFIA significantly inhibited cell proliferation and sicircZNF561 promoted proliferation in prostate cancer ( Figure 7C-F). cancer ( Figure 7C-F).
In addition, we also tested the effect of knocking down circNFIA and circZNF561 on the cell cycle. The result showed that knocking down circNFIA increased the percentage of cells in the G1 phase (49.5 to 66.62% in DU145, 60.94 to 67.86% in LNCaP) and decreased the percentage of cells in the S phase (39.5 to 27.72% in DU145, 25.49 to 21.53% in LNCaP) to arrest cell cycle progression ( Figure 8A,B). In contrast, sicircZNF561 decreased the percentage of cells in the G1 phase and increased the percentage of cells in the S phase, these results suggested sicircZNF561 could promote the cell cycle ( Figure 8C,D).
Similarly, apoptosis assays showed that sicircNFIA significantly promoted cell apoptosis in DU145 and LNCaP cells. Compared with the control, apoptotic cells increased from 6.79% (DU145) and 7.07% (LNCaP) to 8.46% (DU145) and 7.85% (LNCaP) ( Figure  9A,B). After knocking down circZNF561, we detected the percentage of apoptosis cells was decreased ( Figure 9C,D). Taken together, we found that circNFIA may play an oncogenic role and circZNF561 may act as a tumor suppressor in prostate cancer. These results indicated that circNFIA and circZNF561 can be used as candidate tumor biomarkers.   circZNF561 (B left). The efficiency of sicircNFIA (A right) and circZNF561 (B right) was confirmed by qPCR. (C-F) Cell proliferation analysis was performed with a CCK-8 assay in LNCaP and DU145 cells after knockdown of circNFIA and circZNF561. The p-value was calculated by the unpaired two-tailed Student's t-test. * p < 0.05; ** p < 0.01; *** p < 0.001 (circRNAs).
In addition, we also tested the effect of knocking down circNFIA and circZNF561 on the cell cycle. The result showed that knocking down circNFIA increased the percentage of cells in the G1 phase (49.5 to 66.62% in DU145, 60.94 to 67.86% in LNCaP) and decreased the percentage of cells in the S phase (39.5 to 27.72% in DU145, 25.49 to 21.53% in LNCaP) to arrest cell cycle progression ( Figure 8A,B). In contrast, sicircZNF561 decreased the percentage of cells in the G1 phase and increased the percentage of cells in the S phase, these results suggested sicircZNF561 could promote the cell cycle ( Figure 8C,D). and circZNF561. The p-value was calculated by the unpaired two-tailed Student's t-test. *, p < 0.05; **, p < 0.01; ***, p < 0.001 (circRNAs).  Similarly, apoptosis assays showed that sicircNFIA significantly promoted cell apoptosis in DU145 and LNCaP cells. Compared with the control, apoptotic cells increased from 6.79% (DU145) and 7.07% (LNCaP) to 8.46% (DU145) and 7.85% (LNCaP) (Figure 9A,B).
After knocking down circZNF561, we detected the percentage of apoptosis cells was decreased ( Figure 9C,D). Taken together, we found that circNFIA may play an oncogenic role and circZNF561 may act as a tumor suppressor in prostate cancer. These results indicated that circNFIA and circZNF561 can be used as candidate tumor biomarkers.

Construction of a circRNA-miRNA ceRNA Network
As mentioned in previous reports, endogenous circRNAs could act as efficient mi-croRNA sponges to regulate the expression of protein-coding genes [29,32]. To explore the mechanism by which circNFIA and circZNF561 regulated prostate cancer progression, the putative candidate miRNAs binding to circNFIA or circZNF561 were predicted using circBank. As shown in Figure 10A,E, we found 59 miRNAs targeted to circNFIA and 26 miRNAs targeted to circZNF561. Next, we constructed a circRNA-mediated competing The data are presented as the mean ± SD (n = 3). The p-value was calculated by the unpaired two-tailed Student's t-test. * p < 0.05; ** p < 0.01; *** p < 0.001.

Construction of a circRNA-miRNA ceRNA Network
As mentioned in previous reports, endogenous circRNAs could act as efficient mi-croRNA sponges to regulate the expression of protein-coding genes [29,32]. To explore the mechanism by which circNFIA and circZNF561 regulated prostate cancer progression, the putative candidate miRNAs binding to circNFIA or circZNF561 were predicted using circBank. As shown in Figure 10A,E, we found 59 miRNAs targeted to circNFIA and 26 miRNAs targeted to circZNF561. Next, we constructed a circRNA-mediated competing endogenous RNA (ceRNA) network. Using the miRDB database, we identified the target genes of miRNAs interacting with circNFIA or circZNF561. The networks were drawn using Cytoscape 3.0 ( Figure 10B  Moreover, we performed Gene Ontology (GO) and KEGG pathway analysis of potential target genes in the network using the DAVID Bioinformatics Database (Figure 10C,D,G,H). Our results showed that target genes within the circNFIA-related network were mainly enriched in the cGMP-PKG, Rap1, Ras, PI3K-Akt and MAPK signaling pathways and were associated with regulating cell migration, cell-cell adhesion and apoptotic processes. (Figure 10C,D). The circZNF561-related network was mainly enriched in the Rap1, Ras, MAPK, and PI3K-Akt signaling pathways and was associated with regulating protein ubiquitination, GTPase activity, apoptotic processes, and cell proliferation ( Figure 10G,H).

Discussion
ADT is the main treatment for prostate cancer, especially advanced and metastatic prostate cancer. With AR mutation or amplification, a low concentration of androgen can still activate the AR signaling pathway in CRPC, causing rapid worsening of prostate cancer in patients. Therefore, our group and others focused on exploring the important regulator of the AR signaling pathway in prostate cancer. Androgen-responsive noncoding RNAs, including miRNAs [33] and lncRNAs [31], have been reported, while androgen-responsive circRNAs have rarely been studied. In this study, we identified androgen-responsive circRNAs through high-throughput microarray analysis. The expression of six circRNAs (hsa-circ-0076151, hsa-circ-0001055, hsa-circ-0005035, hsa-circ-0005079, hsa-circ-0008326, and hsa-circ-0073627) regulated at the transcriptional level was validated by qPCR. Importantly, ChIP-seq data showed that the AR peaks in these circRNA loci significantly increased after DHT treatment, suggesting that these circRNAs were direct targets of AR.
Recent studies have reported that the biogenesis of circRNAs is mainly regulated by cis-elements or trans-factors, that is, through complementary base pairing on the flanking sequence [15,16] or binding to RBPs special motifs [34,35]. In this study, we identified RBPs involved in the biogenesis of circRNA using CircInteractome. We found that WTAP mainly participated in the formation of androgen-reduced circRNAs, while TNRC6 mainly participated in androgen-induced circRNAs. To further explore the importance of WTAP/TNRC6 in the biogenesis of circRNAs, we knocked down WTAP or TNRC6 and detected candidate circRNAs expression. The expression of hsa_circ_0012079, hsa_circ_0087714, and hsa_circ_0018744 was significantly decreased after WTAP knockdown. Similarly, the expression of hsa_circ_0090744, hsa_circ_0076815, hsa_circ_0049519, hsa_circ_0014100, and hsa_circ_0061590 was significantly decreased after TNRC6 knockdown. These results indicated that WTAP and TNRC6 might both involve in circRNA processing.
CircRNAs were more stable and tissue-specific than linear RNAs. Recent studies have reported that circRNAs could act as promising noninvasive markers for cancer diagnosis and prognosis [36,37]. In addition, abnormal expression of circRNAs has been observed in multiple types of cancer. In this study, we found that two circRNAs, circNFIA and circZNF561, were significantly differentially expressed in prostate cancer cells, tissue samples and plasma samples from patients with prostate cancer. Our results showed that circNFIA was significantly upregulated in prostate cancer cells, prostate cancer tissues, and plasma samples, while circZNF561 was downregulated in prostate cancer cells, tissues and plasma samples. Therefore, circNFIA and circZNF561 are potential diagnostic markers in prostate cancer.
Many studies have reported that circRNA expression is related to the etiology and progression of cancer and is involved in regulating the key signaling pathway of cancers [38]. To characterize the biological roles of circNFIA and circZNF561, we knocked down circNFIA and circZNF561 and explored the effect on cell proliferation, cell cycle, and cell apoptosis. In this study, we revealed that circNFIA may play an oncogenic role, while circZNF561 may act as a tumor suppressor in prostate cancer.
As mentioned in previous reports, circRNAs can act as miRNA sponges to regulate target genes and affect tumor progression. Wei et al. reported that CDR1as inhibited the tumor suppressor miR-7 to promote cell proliferation in NSCLC [39]. Zhang et al. reported that circFOXO3 can act as a miR-138-5p and miR-432-5p sponge to promote the expression of NFAT5 and promote the progression of glioblastoma [40]. CircFOXO3 could also promote the occurrence and development of prostate cancer by competitively binding miR-29a-3p [29]. To explore the potential mechanism by which circNFIA and circZNF561 played a role in prostate cancer, we constructed circNFIAor circZNF561-mediated ceRNA networks by using the circBank and miRDB databases. We demonstrate that this study provides useful information for exploring potential therapeutic and prognostic biomarkers for prostate cancer.

Conclusions
In conclusion, we performed a androgen-responsive circRNAs analysis in prostate cancer patient tissues and identified two important regulatory factors, WTAP and TNRC6, which may involve in circRNA biogenesis. In addition, we explored the potential biological functions and predicted the molecular mechanism of two androgen-responsive circRNAs (circNFIA and circZNF561) by computational analysis and experimental verification, which may act as promising biomarkers for the diagnosis and treatment of prostate cancer.