Introduction

Polycystic ovary syndrome (PCOS) is the most common and heterogeneous endocrine disorder in women of reproductive age. Depending on different criteria and populations, the prevalence of PCOS ranges from 6 to 8% with the NIH criteria, and up to 20% with the Rotterdam criteria [1, 2]. Further, it accounts for approximately 75% of anovulatory infertility [3]. Women with PCOS suffer multiple short-term and long-term complications, including reproductive (hirsutism, amenorrhea, anovulation, infertility, and pregnancy complications), metabolic (dyslipidemia, type 2 diabetes mellitus, and cardiovascular risk factors) and psychological (depression, anxiety, body image, and quality of life) disorders [4].

Circular RNAs (circRNAs), a class of noncoding RNAs, were named for their covalently closed loop structures with the absence of a canonical 5′ cap and 3′ poly A tail. Recently, they were found to be stably conserved in diverse types of eukaryotes; they are resistant to RNAse R, and often exhibit spatiotemporal specific expression [5]. By functioning as microRNA (miRNA) sponges, modifiers of alternative splicing and regulators of parental gene expression, circRNAs can regulate the progression of many diseases at the transcriptional and posttranscriptional level [6]. Many studies have revealed the potential close relationship between circRNAs and the initiation and development of many diseases and physiological processes, especially in tumors [7,8,9,10]. More recently, circRNA expression profiling of preimplantation embryos and human granulosa cells during maternal aging was investigated, showing a promising correlation with the development of oocytes and embryos [11, 12].

The aim of this study was to investigate the circRNA expression profile of CCs isolated from the cumulus–oocyte complex (COC) of PCOS and normal control patients undergoing controlled ovarian stimulation cycles by performing circRNA microarray analysis and to assess their potential significance in the pathogenesis of PCOS, which may serve as novel biomarkers for the accurate diagnosis and effective therapeutic intervention of PCOS.

Materials and methods

Subjects and protocol

Participants were scheduled for in vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI) at the Center for Reproductive Medicine of Yantai Yuhuangding Hospital affiliated to Qingdao University (Shandong, People’s Republic of China) from May to November 2017. To investigate the expression profiles of novel circRNAs, CCs from 6 PCOS and 6 normal control participants were analyzed with microarray. qRT-PCR was conducted in an independent cohort (PCOS group, n = 25; control group, n = 25) to validate the microarray results. All participants in the PCOS group (n = 25) were included under the Rotterdam two of three criteria [13]: oligo-ovulation or anovulation; polycystic ovarian morphology (PCOM) on ultrasonography; and hyperandrogenism with the exclusion of other etiologies such as congenital adrenal hyperplasia, Cushing’s syndrome and androgen-secreting tumors. Participants in the non-PCOS group (n = 25) were those who sought assisted reproductive techniques primarily due to male factors or problems with fallopian tubes (salpingemphraxis or salpingectomy on both sides); patients with endometriosis, thyroid disease, abnormal endocrine condition and other systematic diseases were all excluded. Basal serum hormone levels including follicular stimulating hormone (FSH), luteinizing hormone (LH), estradiol (E2), testosterone (T) and progesterone (P) were measured on day 2 of the menstrual cycle before ovarian stimulation. Information regarding anti-Müllerian hormone (AMH), inhibin B (INHB) and 25-(OH)D3 was also collected. All participants provided written informed consent prior to study commencement.

Each participant in both groups underwent a conventional agonist protocol as previously described by Huang et al. [14]. Briefly, the gonadotropin-releasing hormone agonist (GnRH-a) triptorelin acetate (0.05 mg/day, diphereline; Ipsen Pharma Biotech, France) was applied subcutaneously during the mid-luteal phase. Once pituitary function was fully downregulated, as evidenced by a serum LH level < 3.0 ng/mL and serum E2 level < 30 pg/mL, the patients would subcutaneously receive recombinant FSH (150–187.5 IU, Puregon, follitropin beta; Merck & Co., Inc. USA) for controlled ovarian stimulation. When two or more follicles were ≥ 18 mm in diameter and the serum E2 level was greater than 300 pg/mL per dominant follicle, participants were supposed to receive 250 μg hCG (Profasi; Merck Serono, Switzerland) to facilitate ovulation. The clinical characteristics of all participants in the two groups are shown in Table 1.

Table 1 Comparison of clinical characteristics and hormone levels
Full size table

Cumulus cell isolation

COC retrieval was performed by transvaginal puncture under the guidance of ultrasonography 36 h after hCG administration. CCs surrounding each single oocyte were carefully isolated by mechanical stripping with syringe needles, and CCs from all oocytes per patient were pooled together as one sample. After washing in phosphate-buffered saline (HyClone, GE Healthcare Life Sciences, USA) three times, CCs were then stored at − 80 °C until RNA extraction.

Total RNA extraction and reverse transcription

Total RNA was extracted from cumulus cells using Trizol reagent (Takara, Japan) according to the manufacturer’s protocol. OD260/280 readings from a spectrophotometer (NanoDrop ND-1000) were used to determine the purity and concentration of sample RNA. RNA integrity was also determined by 1% formaldehyde denaturing gel electrophoresis. A total of 1000 ng RNA was added to the 20 μl reverse transcription system depending on the concentration of each sample, and random primers were used in the following step according to the recommendations of the QuantiTect Reverse Transcription Kit (QIAGEN GmbH, Hilden, Germany).

CircRNA microarray and data analysis

Human CircRNA Array v2 (CapitalBio Corporation, Beijing, People’s Republic of China) with four identical arrays per slide (4 × 180 K format) was applied to detect circRNAs with differential expression in cumulus cells between PCOS patients and normal controls. Each array contained probes interrogating approximately 170,340 human circRNAs. Those circRNA target sequences are all from Circbase, Deepbase and Rybak-Wolf2015 [15]. Array hybridization and sample labeling were processed according to the manufacturer’s protocol.

Using GeneSpring software V13.0 (Agilent), the circRNA array data summarization, normalization and quality control were obtained. The data were all Log2-transformed and median-centered by genes using the Adjust Data function of CLUSTER 3.0 software (Stanford University, Stanford, CA, USA), and further analyzed by applying hierarchical clustering with average linkage. Threshold values of ≥ 2 and ≤ − 2-fold change and a t test p value of 0.05 was regarded as proper for selecting the differentially expressed genes. Using Java Treeview (Stanford University School of Medicine, Stanford, CA, USA), we also performed tree visualization. The microarray data were analyzed with the help of CapitalBio Corporation (Beijing, People’s Republic of China). Detailed clinical characteristics of the 12 patients in the microarray pilot group can be found in the supplementary data.

Quantitative real-time PCR (qRT-PCR)

Given the unique covalently closed loop structure of circRNAs, divergent primers rather than canonical convergent primers were designed through Primer-BLAST (NCBI, USA) to target the back-splicing sites, thus distinguishing circRNAs from their linear counterparts. Sequences of all circRNAs are available in certain sources (Circbase, Deepbase or Rybak-Wolf2015). Primers were synthesized by Sangon Biotech (Shanghai, China), and β-actin was chosen as the endogenous control. qRT-PCR was performed using the QuantiNova SYBR Green PCR Kit (QIAGEN GmbH, Hilden, Germany) and monitored with the StepOnePlus Real-Time PCR System (Applied Biosystems, Life Technologies, Waltham, MA, USA). The reactions were initiated in a 96-well optical plate at 95 °C for 2 min, followed by 40 cycles of 95 °C for 5 s and 60 °C for 10 s; each reaction was run in triplicate. The Ct values were determined with the fixed default settings (threshold 0.2, baseline start cycle: 3, end cycle: 15).

Statistical analysis

Data in this study were represented as the mean ± SD for quantitative variables. The Gaussian distribution of the continuous variables was tested by the Kolmogorov–Smirnov statistic. Student’s t test for quantitative data with a Gaussian distribution or Mann–Whitney U test for quantitative data with a non-Gaussian distribution were used to evaluate the differences between groups. Cytoscape (version 3.2.1) was used to visualize the predictive network of circRNA/miRNA/mRNA. Statistical analysis was performed with GraphPad Version 6 (GraphPad Software Inc., San Diego, CA, USA). A p value < 0.05 was considered to be statistically significant. The relative expression level of each circRNA was presented using the 2−ΔΔCt method.

Results

Expression profiles of circRNAs in CCs of PCOS patients

Hierarchical clustering and volcano plot revealed significant circRNA expression differences between two groups (Fig. 1); in total, 286 circRNAs were identified, including 167 that were upregulated and 119 that were downregulated in the PCOS group. To test the genuineness of the differentially expressed circRNAs revealed by microarray and to seek a potential novel biomarker for PCOS, we further validated six candidates (details are shown in Table 2) selected using a stricter screening criteria: fold change > 3.0, p < 0.02 and their relative raw intensity.

Fig. 1
figure 1

Microarray results. a Cluster map of the circRNA microarray profiles in PCOS patients and control individuals. The expression of circRNAs is hierarchically clustered on the y-axis, numbers marked P and N on the x-axis represent different CC samples from PCOS patients and normal control individuals. b A volcano plot was constructed using fold-change values and p values. The vertical lines correspond to 2.0-fold up- and downregulation between CC samples from PCOS patients and normal control individuals, and the horizontal line represents a p value of 0.05. Red and green signs in the two figures represent statistically significant up- and downregulated circRNA expression in the PCOS group, respectively

Full size image
Table 2 Profiles of six candidate circRNAs revealed by microarray
Full size table

Gene ontology (GO) enrichment and KEGG pathway findings

Investigations of circRNA biogenesis have shown that circRNA and its linear isoform originate from the same pre-mRNA but are derived following different subsequent processing patterns [16]. The circulation of a given exon is mainly promoted by the flanking intronic sequences and back splicing, while canonical splicing contributes to the formation of its linear counterpart. General splicing factors are supposed to determine the balance between circRNA biogenesis and canonical splicing [16]. Further, circRNAs were reported to have potential regulatory effects on the expression of parental genes [17, 18]. Given the intricate relationship among circRNAs, their host gene and pre-mRNAs, GO enrichment and KEGG Pathway analysis (Fig. 2) were conducted in our study to better elucidate the novel role of dysregulated circRNAs in the pathogenesis of PCOS. We found that the most significantly enriched GO term in the biological process was organelle fission (GO: 0048285, p = 1.978e − 14) (Fig. 2a); the most significantly enriched GO term in the cellular component was condensed chromosome outer kinetochore (GO: 0000940, p = 4.798e − 09) (Fig. 2b); and the most significantly enriched GO term in the molecular function was cytoskeletal protein binding (GO: 0008092, p = 1.140e − 06) (Fig. 2c). Among the top ten most significantly enriched KEGG pathways (Fig. 2d), cell cycle (hsa04110), oocyte meiosis (hsa04114) and progesterone-mediated oocyte maturation (hsa04914) are known to be involved in oocyte development. Furthermore, it had been reported that excess intraovarian androgen in PCOS might promote hyperfolliculogenesis mainly via the Foxo (hsa04068) and phosphatidylinositol signaling pathway (hsa04070) [19]. Finally, abnormal glycerophospholipid metabolism (hsa00564) has been shown to play a profound role in the pathogenesis of PCOS and complications of insulin resistance [20].

Fig. 2
figure 2

GO enrichment and KEGG pathway analysis of dysregulated circRNAs gene symbols revealed by microarray. a Most significantly enriched GO (-lg p value) terms according to biological process. b Most significantly enriched GO (-lg p value) terms according to cellular component. c Most significantly enriched GO (-lg p value) terms according to molecular function. d Top ten significantly enriched KEGG pathways according to dysregulated circRNA gene symbols

Full size image

qRT-PCR validation

Three pairs of primers were designed for each circRNA, but only one resulted in a single peak (Fig. 3) in the melting curve used (Table 3). The results are shown in Fig. 4; the expression levels of hsa_circ_0043533 (p < 0.01) and hsa_circ_0074576 (p < 0.05) were significantly higher in the PCOS group while the expression level of hsa_circ_0097636 (p < 0.01) was significantly lower versus the non-PCOS group, which was consistent with the microarray data. Finally, hsa_circ_0118530, hsa_circ_0020491, and hsa_circ_0043532 showed no significant differences between the two groups.

Fig. 3
figure 3

Melt curves of six candidate circRNAs’ qRT-PCR results, which all showed a single peak

Full size image
Table 3 Paired primer sequences for the internal control and candidate circRNAs
Full size table
Fig. 4
figure 4

Relative expression levels of six candidate circRNAs quantified by qRT-PCR between the PCOS (n = 25) and non-PCOS (n = 25) groups. The expression levels of hsa_circ_0043533 (c, *p < 0.05) and hsa_circ_0043532 (e, **p < 0.01) were significantly higher in the PCOS group, and the expression level of hsa_circ_0097636 (d, **p < 0.01) was significantly lower versus the non-PCOS group. hsa_circ_0118530 (a, p = 0.4157), hsa_circ_0020491 (b, p = 0.1536) and hsa_circ_0074576 (f, p = 0.4440) showed no significant differences between the two groups

Full size image

Diagnostic value of novel circRNAs in PCOS

A receiver operating characteristic (ROC) curve was performed to determine the value of three novel circRNAs in distinguishing PCOS patients from normal control individuals (Fig. 5). The area under the ROC curve (AUC) of hsa_circ_0043533 was 0.709 (95% CI 0.566–0.851), and the sensitivity and specificity were 60 and 72%, respectively. Similarly, the AUC of hsa_circ_0043532 was 0.718 (95% CI 0.573–0.864), and the sensitivity and specificity were 56 and 88%, respectively. The AUC of hsa_circ_0097636 was 0.738 (95% CI 0.599–0.877), and the sensitivity and specificity were 52 and 96%, respectively. Intriguingly, when the serum testosterone (T) level was applied in combination with hsa_circ_0097636 using binary logistic regression analysis, the AUC increased significantly to 0.893 (95% CI 0.807–0.978), with 68% sensitivity and 100% specificity.

Fig. 5
figure 5

ROC curves revealed that serum testosterone (T) level combined with hsa_circ_0097636 enhanced the efficacy of distinguishing PCOS patients from normal control individuals. The AUC of hsa_circ_0097636, hsa_circ_0043533, and hsa_circ_0043532 alone was 0.738 (95% CI 0.599–0.877), 0.709 (95% CI 0.566–0.851), and 0.718 (95% CI 0.573–0.864), respectively. The AUC of combination model (hsa_circ_0097636 + T) increased significantly to 0.893 (95% CI 0.807–0.978)

Full size image

Prediction and annotation of the ceRNA network using the three validated circRNAs potentially involved in PCOS

To investigate the biological role of the three validated circRNAs, we used bioinformatic tools to predict their potential targets and visualize the interactive network (Fig. 6). The most important and widely recognized function of circRNAs is acting as an miRNA sponge; therefore, we first manually selected all miRNAs which have been reported to have a role in the pathogenesis of PCOS and then tested whether they have potential binding sites for the three dysregulated circRNAs by miRanda (34.236.212.39/microrna/home.do) and TargetScan (http://www.targetscan.org/vert_71/). We also predicted miRNA potential binding with the input genes involved in the six significantly enriched KEGG pathways related to PCOS. Remarkably, the coprediction revealed that four input genes (FOXO3, DGKI, PCYT1B, and CCNB1) have binding sites for the three validated circRNAs’ potential target miRNAs.

Fig. 6
figure 6

Mapping of the predictive interaction network of circRNA/miRNA/mRNA and six significantly enriched KEGG pathways related to PCOS. White rectangles represent KEGG pathways and octagons indicate input genes involved in the pathways: those in yellow are predicted to have binding sites for miRNAs by TargetScan

Full size image

Discussion

PCOS is widely considered an intricate trait resulting from the intrinsic interaction of multiple inherited and environmental factors that first manifests at adolescence on most occasions [21]. To date, many signaling cascades have been proposed in the pathogenesis of PCOS and hold great promise as potential targets for effective treatment strategies, including the MAPK-ERK [14], Wnt/β-catenin [22] and PI3 K-Akt [23] pathways. Although the exact etiology of PCOS remains obscure, its clinical manifestations are also highly heterogeneous, including abnormal folliculogenesis, aberrant steroidogenesis, prominent functional ovarian hyperandrogenism, and adipocyte dysfunction are the major features.

Although the first evidence was in the early 1970s in higher plants, the history of circRNA research is relatively short due to the lack of useful techniques and tools. Only when RNA deep-sequencing technology and bioinformatics became highly developed over the past decade did scientists realize that circRNAs are far more than simple byproducts from mis-splicing during transcription but are rather a sort of ubiquitous, conserved, abundant and stable type of RNA molecule with great regulatory potency [5]. Until now, several roles for circRNAs at the transcriptional and posttranscriptional level have been proposed and investigated [24]. The most cardinal and well-studied function is as an miRNA sponge—a human circRNA, antisense to the cerebellar degeneration-related protein 1 transcript (CDR1as), was found to harbor 63 miR-7 in neuronal tissues. Similar to knocking down miR-7, midbrain development in zebrafish was impaired by the expression of human CDR1as [25]. In most studies, the expression of a certain circRNA normally downregulates its target miRNA and subsequently improves the expression of specific mRNAs and proteins downstream. For instance, upregulated circHIPK3 in diabetic retinas increased the expression of VEGF-C, FZD4 and WNT2 by sequestering the activity of miR-30a-3p through a “sponge mechanism”, which led to endothelial proliferation and vascular dysfunction [9]. A heart-related circRNA termed HRCR by Wang et al. was reported to act as an endogenous miR-223 sponge, which negatively regulated miR-223 function in cardiac hypertrophy [26]. Strikingly, a novel and opposing effect of circRNA was recently delineated [27] when Cdr1as-knockout animals displayed a prominent neuropsychiatric phenotype with destabilized mature miR-7 and enhanced expression of Fos and Nr4a3 in neurons, direct targets of miR-7. These findings may suggest more intricate and profound functions of circRNAs in the competing endogenous RNA (ceRNA) networks, which still requires in-depth investigation.

Compared to circRNAs, studies on miRNAs have been relatively well established. Studies to date have unveiled numerous miRNAs with significant regulatory effects in circulating blood, follicular fluid and granulosa cells of patients with PCOS. For example, miR-93 is upregulated in the adipose tissue of PCOS patients, and can downregulate the expression of insulin-sensitive glucose transporter isoform 4 (GLUT-4) through direct targeting of the 3′ UTR of GLUT-4, leading to insulin resistance [28]. miR-145 suppresses the proliferation of isolated granulosa cells from PCOS patients by targeting insulin receptor substrate 1 (IRS1) and subsequently inhibiting the MAPK/ERK signaling pathway [29]. miR-320a was found to be significantly downregulated in CCs from patients with PCOS, which impaired its modulation of CYP11A1 and CYP19A1 in the process of steroidogenesis thus contributing to the development of estrogen deficiency in CCs [30].

In light of the unique property of circRNAs as “miRNA sponges” corroborated by previous studies, we first detected the expression of novel circRNAs in the CCs of patients with PCOS, and three (hsa_circ_0043533, hsa_circ_0097636, and hsa_circ_0043532) of six were found to be dysregulated through qRT-PCR. Spearman’s rank correlation (Table 4) indicated that the serum T level positively correlated with both the expression of hsa_circ_0043533 and hsa_circ_0097636 in the PCOS group. It has been commonly acknowledged that most PCOS is due to hyperandrogenism arising from dysregulation of steroidogenesis in theca cells that are oversensitive to LH, while a previous study [31] found that female androgen-receptor knockout mice had crucial reproductive defects, indicating androgen’s vital role in the balance between follicle growth and follicular atresia. In this study, we also predicted circRNA–miRNA interactions (Fig. 6) by miRanda and TargetScan. Among the numerous potential targeting miRNAs, surprisingly, we found that miR-125b and miR-30c had potential binding sites for hsa_circ_0097636, miR-132 for hsa_circ_0043533, and hsa_circ_0043532. All of these predicted miRNAs have been previously studied in PCOS [31,32,33,34]. miRNA-125b has been reported to be an anti-apoptotic miRNA and one of the most plentiful miRNAs in the ovary among different species [35]. It can be upregulated by excess androgen and subsequently prevent the ovarian follicle from apoptosis and atresia [31], which may also explain the hyperfolliculogenesis in PCOS patients with hyperandrogenemia. miR-30c was observed to be upregulated in the serum of PCOS patients and rat granulosa cells following FSH exposure [32, 36]. The upregulated expression of miRNA-125b and miR-30c described in previous studies and the downregulation of has_circ_0097636 validated in our work may together be involved in the onset of PCOS. Likewise, through genome-scale screening, miR-132 was identified to inhibit the release of all three steroid hormones (P, E2 and T) in human ovarian cells [34]. Hence, we speculate that upregulated hsa_circ_0043533 and hsa_circ_0043532 likely play a significant role in altering steroidogenesis in PCOS by targeting miRNA-132, but further studies are warranted to explore whether our speculation is true and to reveal the exact underlying mechanisms. To summarize, these three dysregulated circRNAs may have potential significance in the abnormal hyperfolliculogenesis and steroidogenesis in PCOS patients by targeting circRNA–miRNA interactions.

Table 4 Spearman’s rank correlation coefficients of the expression of three differentially expressed circRNAs and paticipants’ partial clinical characteristics
Full size table

Additionally, ROC curves revealed that a combination model using the serum T level and hsa_circ_0097636 significantly enhanced the efficacy of distinguishing PCOS patients from normal control individuals; the AUC was 0.893 (95% CI 0.807–0.978) with 68% sensitivity and 100% specificity. In a previous study, many miRNAs including miR-146a, miR-93 and let-7b were proposed as valuable biomarkers for PCOS [32, 37, 38]. Considering its high cytoplasmic stability, we speculate that using a certain circRNA, such as hsa_circ_0097636, together with the serum T level as the screening criteria could be of great diagnostic value for PCOS.

To the best of our knowledge, this study is the first to detect novel circRNA expression in the cumulus cells of patients with PCOS. We validated several dysregulated circRNAs, which may offer great promise to better understanding the pathogenesis and to develop effective therapeutic interventions for PCOS patients. Due to the limited size of the samples in the present study, a larger population and the specific mechanisms underlying circRNA involvement are required for further investigation.