Circular RNAs are differentially expressed in prostate cancer and are potentially associated with resistance to enzalutamide

Most forms of castration-resistant prostate cancer (CRPC) are dependent on the androgen receptor (AR) for survival. While, enzalutamide provides a substantial survival benefit, it is not curative and many patients develop resistance to therapy. Although not yet fully understood, resistance can develop through a number of mechanisms, such as AR copy number gain, the generation of splice variants such as AR-V7 and mutations within the ligand binding domain (LBD) of the AR. circular RNAs (circRNAs) are a novel type of non-coding RNA, which can regulate the function of miRNA, and may play a key role in the development of drug resistance. circRNAs are highly resistant to degradation, are detectable in plasma and, therefore may serve a role as clinical biomarkers. In this study, AR-V7 expression was assessed in an isogenic model of enzalutamide resistance. The model consisted of age matched control cells and two sub-line clones displaying varied resistance to enzalutamide. circRNA profiling was performed on the panel using a high throughout microarray assay. Bioinformatic analysis identified a number of differentially expressed circRNAs and predicted five miRNA binding sites for each circRNA. miRNAs were stratified based on known associations with prostate cancer, and targets were validated using qPCR. Overall, circRNAs were more often down regulated in resistant cell lines compared with control (588 vs. 278). Of particular interest was hsa_circ_0004870, which was down-regulated in enzalutamide resistant cells (p ≤ 0.05, vs. sensitive cells), decreased in cells that highly express AR (p ≤ 0.01, vs. AR negative), and decreased in malignant cells (p ≤ 0.01, vs. benign). The associated parental gene was identified as RBM39, a member of the U2AF65 family of proteins. Both genes were down-regulated in resistant cells (p < 0.05, vs. sensitive cells). This is one of the first studies to profile and demonstrate discrete circRNA expression patterns in an enzalutamide resistant cell line model of prostate cancer. Our data suggests that hsa_circ_0004870, through RBM39, may play a critical role in the development of enzalutamide resistance in CRPC.

Enzalutamide is a targeted AR inhibitor that competitively binds to the ligand-binding domain (LBD) of the AR 4 . It inhibits AR translocation, recruitment of AR cofactors, and AR binding to DNA 4 . In previous phase 3 studies, enzalutamide prolonged overall and progression-free survival in patients who were chemotherapy naïve 5 , and in those who had previously received chemotherapy 4 . As a result, therapy with second generation anti-androgens has become recognised as a standard of care for advanced PCa 4,5 . Nevertheless, approximately 20 to 40% of patients will present with intrinsic resistance to enzalutamide as determined by sustained elevated prostate-specific antigen (PSA) levels and radiological or clinical progression 6 . Furthermore, patients who have an initial objective response will eventually develop secondary resistance 6 . While the exact mechanisms of enzalutamide resistance are yet to be fully understood, it appears that AR gene amplification emerges during treatment with ADT and facilitates tumour growth in low androgen concentrations 7 . Additionally, expression of the AR splice variant-7 (AR-V7), which is a truncated form of the AR lacking the ligand-binding domain 8 , has been shown to be associated with resistance to enzalutamide 6,[9][10][11] . A number of mutations have also been identified in the AR in patients who are resistant to enzalutamide, such as F876L and may contribute to resistance 12,13 .
With the advances in experimental technology and bioinformatics, our understanding of RNA families has improved, as well as our general understanding of the importance of RNA associated interactions and subcellular locations 14,15 . One type of RNA family is non-coding RNA (ncRNA). ncRNA comprises of several different classes, including microRNAs (miRNAs) and long non-coding RNAs (lncRNAs), both of which are areas of active investigation in PCa 16 . A recently discovered novel ncRNA, called circular RNA (circRNA), may play an important role in cancer initiation, development, and progression [17][18][19] . circRNAs are RNA molecules with covalently joined 3′-and 5′-ends formed by back-splice events, thus presenting as closed continuous loops, which makes them highly stable 20,21 . They typically comprise of one to several coding exons of otherwise linear messenger RNAs (mRNAs) and range between a few hundreds and thousands of nucleotides in length 22 . Their high abundance, stability and evolutionary conservation between species suggest that they may have an important biological regulatory role 19 . circRNAs have been identified in a number of cancers including PCa 23 , suggesting a potential role as a biomarker or therapeutic target. Although, their role in cancer has yet to be fully elucidated, recent research suggests they can bind RNA-binding proteins (RBPs), translate peptides 24 and confer resistance to therapy 25 . miRNAs have previously been shown to affect a wide array of biological processes and have an important role in regulating gene expression in cancer, where they act through downstream tumour-suppressive mRNAs 26 . It has been proposed that circRNAs can act as a miRNA 'sponge' thereby modifying miRNA activity through sequestration, thus altering mRNA target gene expression (34). circRNAs are extremely stable and resistant to RNA degradation, and as such they have the potential to translate into clinically useful blood based 'liquid biopsies' to detect early stage disease and monitor treatment response in real time. The goal of this study was to determine if circRNAs were differentially expressed in enzalutamide resistant cells, and to examine the circRNA-mRNA network involved in the development of drug resistance.

Results and Discussion
AR-V7 is elevated in enzalutamide resistant cells. While it is known that the resistant cell lines used in this study harbour increased F876L, which is an agonist-switch mutation resulting in increased resistance to enzalutamide 12 , there is no information on it's association with AR-V7 levels. We detected AR-FL and AR-V7 expression in the cell line model using a standard curve qPCR method. While, AR-FL copy number was consistent across the panel (Fig. 1A), AR-V7 copy number varied depending on enzalutamide resistance status (Fig. 1B). AR-V7 was significantly elevated in LNCaP clone 1 (highly resistant) compared with LNCaP control (sensitive) (p ≤ 0.001). AR-V7 was also higher in LNCaP clone 1 compared with LNCaP clone 9 (moderately resistant) (p ≤ 0.001). The expression of AR-FL (Fig. 1C) and AR-V7 ( Fig. 1D) was confirmed using RNA in situ hybridisation (RISH) (BaseScope ™ ). Qualitatively, AR-V7 expression varied across cell lines, with the highest expression in clone 1 and no expression detected in the control cell line (Fig. 1D). Enhanced levels of AR-V7 are associated with increased drug resistance 6,27 . circRNA screening identified differentially expressed profiles within an isogenic model of enzalutamide resistance. To determine differential expression of circRNAs within the dug sensitive (control) and resistant clones (clone 1 and clone 9), cell lines were screened for circRNA expression using a circRNA 2.0 microarray (Arraystar), which covers 13,617 circRNAs. In total, 930 circRNAs were classified as present across the panel of three cell lines. These target circRNAs were used for further differential analysis. The fold change (FC) for each circRNA between two groups (control vs. combined clone1/9) was computed. A student's paired t test was then used to identify significantly altered circRNAs. The false discovery rate (FDR) was applied to determine the threshold of p value. circRNAs with FC ≥ 1.5 and p < 0.05 were considered to be significantly differentially expressed. Grouped analysis (control vs. combined clone1/9) of detected circRNAs according to FC was performed. Overall, more circRNAs were significantly down-regulated in the enzalutamide resistant cell lines compared with the control. There were 278 circRNAs significantly up-regulated (p < 0.05, control vs. combined clone1/9) and 588 circRNAs that were significantly down-regulated (p < 0.05, control vs. combined clone1/9). Data is presented as a heat map in Fig. 2. A complete list of circRNAs is accessible through the GEO Series accession number GSE118959 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE118959) 28 and is provided in Supplementary Table 1.
The circRNA profile is further altered depending on the extent of enzalutamide resistance.
Differential circRNA expression was also evident depending upon the extent of enzalutamide resistance (Fig. 3).
www.nature.com/scientificreports www.nature.com/scientificreports/ Clone 1 vs. control. In clone 1, we identified 230 up-regulated circRNAs (p < 0.05, vs. control), and 465 that were down-regulated (p < 0.05, vs. control). Thus, indicating the changing levels of circRNAs as enzalutamide resistance develops and levels of AR-V7 increases. Data is shown as a scatterplot and associated heatmap in Fig. 3A,C. A complete list of circRNAs is provided in Supplementary Table 2.
Clone 9 vs. control. In terms of clone 9, we discovered 60 up-regulated circRNAs (p < 0.05, vs. control), and 175 that were down-regulated (p < 0.05, vs. control). Data is shown as a scatterplot and associated heatmap in Fig. 3B,D. A complete list of circRNAs is provided in Supplementary Table 3. A Venn diagram is provided to show the overlap and different levels of expression between clone 1 and clone 9 with control ( Fig. 3E). This Venn diagram display 585 circRNAs that were differentially expressed clone 1 vs. control but not clone 9 (shown in red) and 125 differentially expressed circRNAs between clone 9 vs. control but not clone 1 (shown in light green). There were 111 differentially expressed circRNAs common to both clone 1 and clone 9 vs. control (dark green).
Associated circRNA parental genes are involved in pro-oncogenic activities. The top 5 circR-NAs, ranked by FC, are shown in Table 1 for clone 1 vs. control; and in Table 2 for clone 9 vs. control. The parental genes of differentially expressed circRNAs were obtained from circBASE database (www.circbase.org) 29 . hsa_circ_0001275 was up-regulated in clone 1 vs. control (p = 0.047). The associated parental gene is PLCL2. Previously, PLCL2 (Inactive phospholipase C-like protein 2) was identified as part of a 23-gene signature, which predicted metastatic-lethal PCa outcomes in men diagnosed with clinically localised PCa 30 . hsa_circ_0022392 was down-regulated clone 1 vs. control (p = 0.0002) and is associated with the gene FADS2 (Fatty acid desaturase 2), which may have a role to play in cancer development 31 . In clone 9, hsa_circ_0045697 is up-regulated (p = 0.029, vs. control) and is associated with the oncogene ITGB4 (Integrin Subunit Beta 4). Studies have shown that ITGB4 promotes prostate tumourigenesis 32 . Further information is outlined in Tables 1 and 2. www.nature.com/scientificreports www.nature.com/scientificreports/ miRNAs are associated with circRNAs. circRNAs contain multiple sites called miRNA response elements (MREs) which are miRNA binding sites found on circRNAs 18 . circRNAs can bind up to five different miRNAs. For this study, the miRNAs were predicted using Targetscan 33 and miRanda 34 bioinformatic platforms. This bioinformatics approach determined which probable miRNA was associated with each circRNA. For each identified circRNA, the top five most likely miRNA binding sites were predicted. The circRNAs were then filtered according to miRNAs that were strongly associated with PCa in the literature (Table 3), thus producing a list of ten relevant up-regulated and down-regulated circRNAs for validation (Table 4) (41). Further information relating to corresponding parental gene, MREs, miRNAs, and associated miRNA function is outlined in Table 4.  www.nature.com/scientificreports www.nature.com/scientificreports/ Validation of circRNAs. Custom designed outward facing primers were designed for use with qPCR for selected circRNAs (Table 5). hsa_circ_0001721 was significantly up-regulated in clone 1 vs. control (p ≤ 0.05), which corresponded to the array data (Fig. 4A). Similarly, hsa_circ_0001721 was significantly up-regulated in the more resistant clone 1 vs. clone 9 (p ≤ 0.05) (Fig. 4A). hsa_circ_0001721 is an exonic circRNA, located on chromosome 7 and is associated with the gene CDK14 35,36 . hsa_circ_0004870 was significantly down-regulated in clones 1 and 9 vs. control (p ≤ 0.05) (Fig. 4B). hsa_circ_0004870 is an exonic circRNA located on chromosome 20 and is associated with the gene RBM39 37 .

hsa_circ_0004870 may have a role in splicing via U2AF65. Previous studies have demonstrated that
circRNAs are down-regulated in cancer 17 , therefore we selected hsa_circ_0004870 for further investigation. We confirmed that hsa_circ_0004870 was down-regulated in LNCaP (p ≤ 0.01) compared with the benign prostatic hyperplasia line, BPH1 (Fig. 5A). Similarly, hsa_circ_0004870 was down-regulated in the AR positive 22Rv1 cell line (p ≤ 0.01) compared with the AR independent line, DU145 (Fig. 5B). The coordinates for hsa_circ_0004870 (chr20:34,302,106-34,313,077), correspond to the gene RBM39 on the UCSC Genome Browser, thus identifying this as the parental gene. RBM39 is a serine/arginine-rich RNA-binding protein thought to activate or inhibit the alternative splicing of specific mRNA by interacting with the spliceosomal components within splice sites 37 . RBM39 was significantly down-regulated in the resistant clones 1 (p ≤ 0.0001) and clone 9 (p ≤ 0.0001) compared with control (Fig. 6A). RBM39 encodes a member of the U2AF65 family of proteins and it has previously been shown that, U2AF65 leads to expression of AR-V7 via the lncRNA, PCGEM1, binding to AR pre-mRNA 38 . We confirmed expression of U2AF65 in the cell line panel, which was significantly down-regulated in the clone 1 (p ≤ 0.05) (Fig. 6B). Our data has shown that RBM39 and U2AF65 are down-regulated in clone 1, which has the highest expression of AR-V7. This may be due, in part, to high turnover of mRNA, or circRNA regulation of alternate pathways. This data suggests that the deregulation of hsa_circ_0004870 may be associated with the development of drug resistance through the regulation of AR-V7.

Conclusion
circRNAs have been identified in a number of different cancers (90), suggesting a potential role as a biomarker or therapeutic target. Although, their role in cancer has yet to be fully elucidated, recent research suggests they can act as miRNA sponges (170), bind RNA-binding proteins (RBPs), translate peptides (83) and may confer  www.nature.com/scientificreports www.nature.com/scientificreports/ resistance to therapy (192). In this study, we report for the first time, to the best of our knowledge, circRNA expression profiles associated with enzalutamide resistant PCa. Our findings indicate that circRNAs may potentially represent valuable prognostic and diagnostic biomarkers in the real time monitoring of treatment response to enzalutamide. Given that other studies have shown circRNAs to be abundant, highly stable, and detectable in human saliva, tissue and blood samples 22,39 , their potential as liquid based biopsy markers is evident, in addition to their capacity to serve as a predictive marker in this disease.

Methods
Cell lines. The isogenic enzalutamide resistance LNCaP model was gifted from Novartis 12 . The panel consisted of an aged match control cell line (drug sensitive), and two sub-lines termed clone 1 and clone 9. Clone 1 was most resistance to the drug, with clone 9 displaying moderate resistance. Cells were cultured in RPMI-   www.nature.com/scientificreports www.nature.com/scientificreports/ circRNA microarray. Cell line (from three independent biological replicates) analysis was performed using the Arraystar Human circRNA Array version 2.0 (Arraystar, Rockville, MD, USA). The sample preparation and microarray hybridization were performed according to manufacturer's instructions. Briefly, total RNA was digested with RNAse R (Epicentre, Illumina, San Diego, CA, USA) to remove linear RNAs and enrich for circR-NAs. The enriched circRNAs were amplified and transcribed into fluorescent cRNA utilizing a random priming method Arraystar Super RNA Labelling Kit (Arraystar). The labelled cRNAs were hybridized onto the Arraystar Human circRNA Array V2 (8 × 15 K). The array slides were washed and scanned on the Agilent Scanner G2505C. Agilent Feature Extraction software (version 11.0.1.1) was used to analyse acquired array images.
Microarray data analysis. Quantile normalization and subsequent data processing were executed using R software package 40 . CircRNAs with at least 4 out of 8 samples that were flagged as present or marginal (an attribute that denotes the quality of the entities) were considered to be target circRNAs according to GeneSpring software's definitions and instructions. CircRNA and miRNA interactions were predicted with the Arraystar's miRNA target prediction software based on TargetScan 33 and miRanda 34 . These target circRNAs were used for further differential analysis.  Table 5. Primers used in this study. www.nature.com/scientificreports www.nature.com/scientificreports/ Quantitative real-time PCR. cDNA was synthesized from 1 µg RNA using a High Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific). The qPCR analyses were performed on a 7500 Real-Time PCR System using SYBR ™ Green (Thermo Fisher Scientific). Primers are outlined in Table 5. GAPDH was used as reference gene. The relative expression and fold change of each gene was calculated using the delta delta Ct method.
RNA in situ hybridisation. The BaseScope ™ (Advanced Cell Diagnostics, CA, USA) assays were performed manually according to the manufacturer's instructions. This method allows the detection of exon junctions and the analysis of splice variants. Briefly, the BaseScope ™ assay procedure included the following steps: FFPE sections were deparaffinised and treated sequentially with specific pre-treatments to allow for target probe access. Target probes were added onto the slides and incubated in the HybEZ oven (Advanced Cell Diagnostics) for 2 h at 40 °C to allow probe hybridization to RNA targets. The slides were washed and incubated with a series of signal amplification solutions. The signal was amplified using a multi-step process, and detected using a red chromogenic substrate (10 min at room temperature). The slides were counterstained with haematoxylin and mounted with Cytoseal mounting medium (Richard-Allan Scientific, CA, USA).