Insights into circular RNAs: Biogenesis, function and their regulatory roles in cardiovascular disease

Abstract As a distinctive member of the noncoding RNA family, circular RNAs (circRNAs) are generated from single‐stranded, covalently closed structures and are ubiquitous in mammalian cells and tissues. Due to its atypical circular architecture, it was conventionally deemed insignificant dark matter for a prolonged duration. Nevertheless, studies conducted over the last decade have demonstrated that this abundant, structurally stable and tissue‐specific RNA has been increasingly relevant in diverse diseases, including cancer, neurological disorders, diabetes mellitus and cardiovascular diseases (CVDs). Therefore, regulatory pathways controlled by circRNAs are widely involved in the occurrence and pathological processes of CVDs through their function as miRNA sponges, protein sponges and protein scaffolds. To better understand the role of circRNAs and their complex regulatory networks in CVDs, we summarize current knowledge of their biogenesis and function and the latest research on circRNAs in CVDs, with the hope of paving the way for the identification of promising biomarkers and therapeutic strategies for CVDs.


| INTRODUC TI ON
Cardiovascular diseases (CVDs) remain the leading cause of death and rising health expenditure globally. 1 The overall prevalence of cardiovascular disease has nearly doubled from 271 million in 1990 to 550 million in 2019, and cardiovascular deaths have also surged by 6.5 million in those decades. 2,3 Meanwhile, the social burden caused by the morbidity and disability of CVDs cannot be overlooked. 4 The globalization trend of aging and population growth has resulted in increased morbidity and mortality of CVDs in the elderly. 5 Therefore, the identification of novel predictive biomarkers and therapeutic targets for CVDs is of paramount importance.
Of the entire human genome, only about 2% can code for proteins. Most of the remaining noncoding genomes that do not code for any proteins have now been demonstrated to be critical regulators of CVDs. [6][7][8][9] As a particular class of noncoding RNA, circular RNA (circRNA), initially discovered in plant viroids in 1976, has a single-stranded, covalently closed RNA structure. 10 Unlike linear RNAs, circRNAs are generated by covalent reverse splicing of pre-mRNAs and are considered by-products of linear RNAs. In the past few years, circRNAs were rarely detected owing to the lack of canonical 5′ caps and 3′ poly-A structures. 11 With the advent of highthroughput sequencing and circRNA-specific computational tools, the enigma of circRNA was gradually unravelled and became a hotpot in the noncoding RNA family. [12][13][14][15] Significantly, emerging studies have supported the vital role of circRNAs in CVDs, such as atherosclerosis (AS), myocardial infarction (MI), heart failure (HF), diabetic cardiomyopathy (DCM), hypertension, cardiac hypertrophy and fibrosis. A comprehensive summary of these dysregulated circRNAs would assist in elucidating the association between circRNAs and the occurrence and progression of CVDs.
Here, we review the biogenesis, structural characteristics and biological functions of circRNAs, summarize the latest research progress of circRNAs in the pathogenesis of CVDs and discuss the application prospects of circRNAs in the prevention, diagnosis and treatment of CVDs.

| B IOG ENE S IS
In eukaryotes, spliceosomes typically snip off introns in messenger RNA (mRNA) to produce linear RNA. In contrast to this canonical splicing, circRNA undergoes a distinct form of splicing called backsplicing, whereby the downstream 3′ splice site is connected to the upstream 5′ splice site through a 3′-5′ phosphodiester bond and forms a covalently linked closed loop ( Figure 1). [16][17][18] This process is regulated by RNA-binding proteins (RBPs), transcription factors, cis-acting elements and trans-acting factors, and there exists a competitive relationship between canonical splicing and back-splicing. 19,20 F I G U R E 1 The biogenesis models of circRNAs.
Jeck proposed two back-splice formation models of circRNA in 2013 12 : the lariat-driven circularization model (also called the exon-skipping model) and the intron-paring-driven model. In the first model, exon skipping generates a functional transcript and forms a lariat intermediate composed of the skipped exons and introns. This lasso can undergo secondary cleavage to release exonintron circular RNAs (EIciRNAs) containing both introns and exons, exonic circRNAs (EcRNAs) containing only exons or intron-only circRNAs (ciRNAs). 21,22 When the 7-nt GU-rich base sequence and the 11-nt C-rich base occur near the 5′ splice site and branch site sequence, those motifs can induce introns to form lariat structures to avoid degradation by branching enzymes, resulting in stable ciRNAs. 23 If we regard the first model as a passive manner of circNRA formation, the second model is more like an active process of direct splicing to form circRNAs after intron pairing. In the intron pairingdriven circularization model, direct base pairing (such as Alu repetitive elements) in long flanking introns can generate the formation of secondary structures in the pre-mRNA, which is crucial for the subsequent connection between downstream splice donors and upstream splice acceptors. 11,24 Several studies have demonstrated that some RBPs can recognize and bind specific sequences within flanking introns, bring two splicing sites close enough to form a loop and then enhance circRNA biogenesis. 25,26 For example, the alternative splicing factor Quaking (QKI) and Muscle blind protein (MBL) can promote the efficiency of circRNA production through protein-toprotein interactions or self-dimerization. 16,26 Additionally, downregulation of MBL leads to a significant decrease in circMbl production, which further illustrates the crucial regulatory role of RBPs in cir-cRNA cyclization.
Furthermore, a small fraction of circRNAs originate from pretransfer RNA (pre-tRNA). The intronic elements of tRNAs must be removed to form functional RNAs. In this process, the tRNA splicing endonuclease (TSEN) complex recognizes the bulge-helix-bulge (BHB) motif and cleaving within the anticodon loop of pre-tRNA.
Subsequently, the cleaved introns are joined together by RtcBlike proteins to form tRNA intronic circular RNAs (tricRNAs). 27,28 Moreover, a recent study identified a novel type of circular transcript: read-through circRNA (rt-circRNA), which consists of two adjacent genes on the same strand and accounts for a small portion of all circRNAs. 29 Within the nucleus, pre-mRNA removes introns by canonical splicing to form linear mRNA. Linear splicing produces circRNAs through two back-splice formation models: the lariat-driven circularization model and the intron-paring-driven model. These two ways can eventually form ciRNA, EcRNA or EIciRNA. Furthermore, two special circRNAs, tricRNAs and rt-circRNA, can be formed through pre-tRNA cleavage and two adjacent genes. mRNA: messenger RNA; EIciRNAs: exon-intron circRNA; EcRNAs: exonic circRNAs; ciRNAs: intronic circRNA; tricRNAs: tRNA intronic circular RNAs; rt-circRNA: read-through circRNA; pre-tRNA: pre-transfer RNA; RBPs: RNAbinding proteins; BHB: bulge-helix-bulge.

| CHAR AC TERIS TIC S
CircRNAs are abundantly present in a variety of organisms such as Drosophila, nematodes, plants and mammals. In mammals, including humans, mice and rats, circRNAs are highly conserved. 30,31 It has been reported that approximately 9% of cardiac-expressed genes can produce circRNAs. 13,30,32 A recent RNA sequencing analysis has revealed that about 30% circRNAs in the heart are conserved between mice and rats, and approximately 10% are conserved among the three species. 33 Moreover, circRNAs can avoid degradation by PNase R enzyme and other exonucleases due to their unique circular structure and maintain a stable structure. 34 Hence, circRNA has a longer half-life than linear RNA, with an average half-life of about five times that of mRNA. 12,35 Additionally, the expression of circRNAs is also tissue-specific, and it can be expressed in normal tissues of the colon, heart, kidney, liver, lung, stomach and human gland. 31,36,37 Meanwhile, the abundance of circRNAs is negatively correlated with the proliferative capacity of cells. 36 For instance, low proliferative cells such as cardiomyocytes usually have higher expression levels of circRNAs.
These unique features make circRNAs have great potential to be ideal biomarkers for CVDs.

| FUN C TION
Initially, circRNAs were considered noncoding RNA molecules with regulatory functions. Nevertheless, recent studies have unveiled several additional roles of circRNAs, as depicted in Figure 2. By regulating mRNA and protein expression, circRNAs indirectly regulate target genes. Moreover, circRNAs can act as protein scaffolds, thereby facilitating protein-protein interactions. Furthermore, a subset of circRNAs has been shown to mediate cap-independent translation, while circRNAs located in the nucleus can also regulate the transcription of parental genes. In the following module, we will systematically review the biological functions of circRNAs.

| miRNA S P ONG E
MicroRNAs (miRNAs) are a class of endogenous ~23 nucleotide RNAs that regulate post-transcriptional silencing of target genes by pairing to the mRNAs of protein-coding genes. 38 As competing endogenous RNAs (ceRNAs), circRNAs can act as decoys for miRNA by binding to miRNA response elements (MREs) and subsequently hindering miRNA activity, thus regulating mRNA expression. 39,40 This miRNA sponge function of circRNAs has been demonstrated in several studies. For example, the cerebellar degeneration-related protein 1 antisense (Cdr1as) contains 63 conserved binding sites for the miRNA miR-7. 13,41 In neural tissues, CDR1as can bind to miR-7 and increase miR-7 target mRNA levels by inhibiting miR-7 activity. 13 Similarly, by acting as a sponge for miR-7, CDR1as promotes MI by targeting miR-7 and its downstream factors, SP1 and polymerase (PARP). 42 Additionally, circCHFR regulates cell growth, migration and inflammation in oxidized low-density lipoprotein (ox-LDL)-treated human vascular smooth muscle cells (VSMCs) by functioning as a sponge of miR-214-3p. 43 Furthermore, circUSP36 aggravates endothelial cell dysfunction in AS caused by ox-LDL by competitively binding to miR-637. 44 These researchers indicate the crucial role of circRNAs in the pathological development of CVDs via the miRNA sponge effect.

| PROTEIN S P ONG E
In addition to their function as miRNA sponges, researchers have demonstrated that circRNAs can bind to specific protein binding sites, thereby playing the role of protein sponges. RBPs are a class of proteins involved in gene transcription and translation. 45 Combining circRNAs to form circRNA-RBP complexes can regulate the life cycle of circRNAs, protein biogenesis and transcription of parental genes. 16,[46][47][48] For instance, circMbl and its intronic sequences flanking the second exon contain conserved MBL binding sites, and MBL can induce the circularization of circMbl through these binding sites. 16 Similarly, circPABPN1 suppressed the binding of HuR to PABPN1 mRNA and the translation level of PABPN1 by binding to HuR in human cervical carcinoma HeLa cells. 46 In addition, circular antisense noncoding RNA in the INK4 locus (circANRIL) regulates ribosome biogenesis and promotes atheroprotection by binding to the lysine-rich domain of pescadillo zebrafish homologue 1 (PES1). 47,48 Compared with linear RNA, the tertiary structure of circRNAs has more extraordinary protein binding ability, which may be an essential structural foundation for the interaction between RBPs and circRNAs. 49

| PROTEIN SC AFFOLD
Some circRNAs can act as protein scaffolds to regulate gene expression and protein function by promoting the interaction of two or more proteins. 50 For example, circ-forkhead box protein O3 (circ-FOXO3) can act as a protein scaffold of MDM2 and p53, facilitating MDM2-induced p53 ubiquitination and subsequent degradation, thereby promoting tumour apoptosis. 51 Moreover, circ-Foxo3 can interact with cyclin-dependent kinase 2 (CKD2) and the CDK2 inhibitor p21 to form a circ-FoxO3-p21-CDK2 ternary complex and block cell cycle progression. 52 Circ-Amotl1 can serve as a scaffold for AKT and phosphoinositide-dependent kinases (PDK), activate AKT phosphorylation and pAKT nuclear translocation, enhance cell proliferation and survival and protect against doxorubicin-induced cardiomyopathy. 53

| PROTEIN TR AN S L ATI ON
Generally, translation initiation in eukaryotes relies on the cap structure at the 5′ end of the mRNA. Due to the lack of the 5′ cap, circR-NAs have long been considered to have no translational function. In  [55][56][57] Recent studies have shown that the translation of circRNAs is mainly mediated by two cap-independent translation mechanisms, IRES and m6A. IRESs are sequences located in the 5′ UTRs of mRNAs that can directly recruit the 40S subunit of the ribosome and initiation factors on RNA to initiate translation. 58 The m6A motif promotes protein translation under the combined action of YTHDF3 and initiation factor eIF4G2, and a single m6A site is sufficient to drive translation initiation. 59 Notably, the two mechanisms of circRNA-protein translation function are not independent of each other. High m6A methylation levels were detected in circZNF609, which translates protein via the IRES pathway. 60 Nevertheless, further research is necessary to determine the specific relationship between these two mechanisms.

| TR ANSCRIP TIONAL REG UL ATION OF PARENTAL G ENE S
CiRNAs and EIciRNAs are primarily located in the nucleus, can function as transcriptional regulators to enhance the expression of parental genes in a cis-acting manner. For example, ci-ankrd52 can accumulate at its sites of transcription and then interact with RNA Pol II, acting as a cis-regulator in its parental gene ANKRD52. 23 Besides, EIciRNA circEIF3J and circPAIP2 can form an EIciRNA-U1 snRNP complex through specific RNA-RNA interactions between U1 snRNA and EIciRNA, which subsequently interact with the promoter of the RNA pol II transcription complex and enhance the expression of their parental genes EIF3J and PAIP2. Meanwhile, knockdown of circEIF3J and circPAIP2 resulted in decreased transcript levels of EIF3J and PAIP2, respectively. 61 In the nucleus, circRNAs and U1 snRNP can form an EIciRNA-U1 snRNP complex and bind to the promoter to enhance the expression of parental genes. In the cytoplasm, circRNA can function as a miRNA sponge, protein sponge and protein scaffold. A small fraction of RNA has been demonstrated to accomplish protein transcription cap independently.

| THE ROLE IN C ARD I OVA SCUL AR DISE A SE S
It has been less than a decade since the first study focusing on the biological role of circRNAs in CVDs. 62 Therefore, the relationship between circRNAs and CVDs is still a burgeoning area of research.
While still in its infancy, the role of circRNAs in the occurrence and progression of CVDs has attracted much attention from researchers, and the number of related studies is rapidly increasing. In this section, we present a detailed summary of CVD-related circRNAs and their functions to support the evaluation of circRNAs as potential biomarkers and therapeutic targets for CVDs.

| ATHEROSCLEROS IS
Atherosclerosis (AS) is a chronic inflammatory disorder of the arterial vessel walls and the primary pathological basis of CVDs. Endothelial cells (ECs) and vascular smooth muscle cells (VSMCs) are crucial components of the coronary intima and media, respectively. Abnormal proliferation, migration and invasion of these two cells play a vital role in the pathogenetic process of AS. 63,64 Growing evidence suggests that circRNAs regulate cell proliferation and invasion via targeted miRNAs, thereby participating in AS progression (Table 1) Circ-SATB2 and circ_0029589 can target STIM1 via miR-939 and miR-214-3p, respectively, to regulate the proliferation, migration and invasion of VSMCs. 67,68 Furthermore, circANRIL can induce apoptosis, inhibit proliferation and confer atheroprotection by generating nucleolar stress and p53 activation through binding to pescadillo homologue 1 (PES1). 47 It is worth noting that circRNAs can also affect AS progression in other ways. Treating human aortic endothelial cells (HAECs) with ox-LDL increases the expression of circ_0090231, which was also identified as a sponge for miR-635. 69 Furthermore, knockdown of circ_0090231 inhibits cell apoptosis through the circ_0090231/miR-635/NLRP3 axis, thereby inhibiting AS development. Hsa_circ_0030042 inhibits abnormal autophagy by blocking its recruitment to beclin1 and FOXO1 mRNA and protects advanced atherosclerotic plaque stability by targeting eukaryotic initiation factor 4A-III (eIF4A3). 70 CircDENND1B can inhibit foam cell formation by promoting the expression of ATP binding cassette subfamily A member 1 (Abca1) and participating in the anti-atherosclerotic effect of IL-1β monoclonal antibody (IL-1β mAb) by promoting cholesterol efflux through the miR-17-5p/Abca1 pathway. 71 Future studies will likely uncover additional ways in which circRNAs contribute to AS progression.

| MYOC ARDIAL INFARC TION/ ISCHEMIA REPERFUS I ON INJ URY
Myocardial infarction (MI) is a clinical syndrome resulting from a disturbance in the equilibrium between myocardial oxygen supply and demand. 73 Typically, the coronary arteries perform an essential function by providing blood and nutritional sustenance to the heart. However, when an atherosclerotic plaque in a coronary artery ruptures and haemorrhages, it can result in stenosis and even occlusion of the artery. Prolonged ischemia and hypoxia can trigger a cascade of events such as necrocytosis, apoptosis, and the inflammatory response of cardiomyocytes, ultimately culminating in heart failure and cardiac remodelling. [74][75][76][77] Recent research has substantiated that several circRNAs engage in multiple pathological phases of MI through diverse pathways (Table 2).
Vausort et al. were the pioneers in demonstrating that myocardial infarction-associated circular RNA (MICRA) expression levels were lower in blood samples from MI patients than in healthy volunteers.
Besides, patients with low MICRA levels were at relatively higher risk for left ventricular (LV) dysfunction 3-4 months after MI. 78 The association of MICRA with the degree of LV dysfunction after acute MI and the inverse relationship between MICRA levels and ejection fraction were subsequently reconfirmed in another study by the same research group. Thus, it was suggested that MICRA could be helpful in the risk stratification of MI patients. 79 Furthermore, three subsequent studies confirmed that circular tetratricopeptide repeat domain 3 (circ-Ttc3), circRNA 010567 and circSNRK could sponge specific miRNAs and play a vital protective role in MI by inhibiting rat cardiomyocyte apoptosis. [81][82][83] In contrast, several other cir-cRNAs, including ACAP2, circ_0068655, circSAMD4A and MFACR, promote cardiomyocyte apoptosis through their interaction with miRNAs. [84][85][86][87][88] Furthermore, Geng et al. demonstrated that the Cdr1as/miR-7 pathway in mouse brains also exists in cardiomyocytes. Meanwhile, Cdr1as overexpression was proven to increase cardiac infarct size in MI mice by inhibiting miR-7 and its downstream targets PARP and SP1, as well as aggravating cardiomyocyte apoptosis under hypoxia treatment. 42 However, in a recent study, Cdr1as was found to reduce MI size and improve cardiac function after MI. 80 This study found that the expression of Cdr1as in the AMI region of domestic pig hearts was significantly increased, and there was a significant positive correlation between Cdr1as and LV, as well as right ventricular ejection fraction (LVEF) and LV stroke volume, and negatively correlated with infarct size. The discrepant results between the two studies may be attributed to differences in the experimental species and AMI intervention. Overall, existing studies have confirmed that circRNAs regulate MI progression, but the exact role of CDr1as remains controversial and deserves further investigation in future studies.
Opening a blocked coronary artery and promptly restoring blood perfusion to myocardial cells, known as reperfusion, is the primary method of myocardial cell resuscitation in clinical practice. However, reperfusion is a double-edged sword. While saving dying cardiomyocytes, calcium load and a large number of reactive oxygen species can also aggravate cardiomyocyte damage. 98 Inhibition of autophagy during ischemia/reperfusion (I/R) is beneficial for protecting cardiomyocytes. 99 96 In addition to autophagy, a recent study showed that circRNAs could regulate I/R injury through pyroptosis. 97 In contrast to the sham group, the protein expression of pyroptosis- Similarly, circSamd4 reduces mitochondrial oxidative stress and promotes cardiomyocyte proliferation, thereby reducing the area of fibrosis and promoting cardiac repair after MI. 93 These studies collectively suggest that circRNAs may be potential targets for improving MI prognosis.

| C ARDIAC HYPERTROPHY/C ARDIAC FIB ROS IS/HE ART FAILURE
Both cardiac hypertrophy and fibrosis are closely related to heart failure. When the heart is exposed to various physiological or pathological stimuli, it undergoes adaptive hypertrophy to maintain its normal function. However, persistent cardiac hypertrophy led to cardiac dysfunction and gradually progressed to heart failure. Cardiac fibrosis is one of the most frequent causes of cardiac dysfunction.
Excessive deposition of extracellular matrix (ECM) disorganizes the structure of cardiomyocytes and reduces the compliance of cardiac tissue, which further accelerates the progression of heart failure.
Several studies have demonstrated that circRNAs are dysregulated in cardiac hypertrophy ( CircRNAs play an active role in myocardial fibrosis ( Table 3).
of cardiac fibrosis, respectively. Further research is needed to determine the specific functions of these circRNAs in cardiac fibrosis.
Additionally, excitingly, several studies have revealed the direct involvement of circRNAs in regulating heart failure (HF) and their potential as therapeutic targets (Table 3). For instance, CDR1as, a research hotspot, has been confirmed to be upregulated in the plasma of chronic HF patients and positively correlates with NYHA class in HF patients. 109 Functional experiments in human cardiomyocytes confirmed that CDR1as could regulate the occurrence and progression of HF through the miR-135a/heme oxygenase 1 and miR-135b/ heme oxygenase 1 pathways. Sun et al. 117  Notably, the patients enrolled in this study had no other comorbidities.
Furthermore, two studies focused on whether circRNAs exhibited differential expression levels in various stages of AF. 127,128 In the first study, 83 (48 upregulated and 35 downregulated) and 99 (58 upregulated and 41 downregulated) circRNAs with significantly different expression levels were identified in paroxysmal AF and persistent AF compared with the control group, respectively. 127 In the second study, the increased number of circRNA species detected was accompanied by miRNA downregulation during the transition from paroxysmal to permanent AF. 128  In a case-control study, Bao et al. 130 found that hsa_circ_0037911 was significantly elevated in blood samples from patients with essential hypertension (EH) compared with the non-EH group. They also observed higher levels of hsa_circ_0037911 in the EH group in patients who were male, smoked and drank alcohol, which are risk factors for EH. In another case-control study, He et al. 131   Although research on circRNAs in CVDs is just emerging when compared with other research fields, such as cancer and neurological disorders, multiple studies have confirmed that circRNAs with stable, highly conserved and longer half-lives have bright prospects as candidate biomarkers and treatment targets for CVDs.

| FUTURE PER S PEC TIVE S AND CON CLUS IONS
First, circRNA has a long half-life and can be stably expressed in blood, making it easy to detect and an ideal diagnostic biomarker.
Second, previous studies have confirmed that some circRNAs are dysregulated in various CVDs and even in different stages of the same disease, which helps to reflect the occurrence and dynamic progression of CVD. Third, artificial circRNA sponges (circmiRs) have been shown to attenuate the progression of stress overloadinduced cardiac hypertrophy and HF in mice by exploiting the function of miRNA sponges of circRNAs. 114 Therefore, using exogenous circRNAs as therapeutic miRNA antagonists for the treatment of CVD is also a research field worthy of active exploration and has excellent prospects. Finally, the added value of circRNAs for the prognosis of CVD patients has been confirmed. 78,79 Future large-scale, multicentre studies are helpful to further prove the feasibility of translation from bench to clinic. However, challenges remain in applying circRNAs for diagnosing and treating CVDs Compared with traditional biomarkers, the detection cycle of circRNAs is longer due to the relatively imperfect detection technology. How to improve the detection efficiency of circRNAs through technological innovation is also a question worth considering. Additionally, the current research on circRNAs in hypertension, DCM and arrhythmia other than AF is still in its infancy.
Therefore, detailed and meticulous studies to clarify the role of circRNAs in these diseases may lead to breakthrough progress. In addition, numerous studies have used bioinformatics methods to construct a circRNA-miRNA-mRNA interaction network, the specific mechanism still requires further research and verification.
Overall, although exploring circRNAs is only the tip of the iceberg, existing studies have shown that circRNAs have tremendous potential in preventing, diagnosing and treating CVDs. With the continuous progress of scientific technology and detection methods, we have reason to believe that gaining insight into the exact regulatory mechanism, complete regulatory network and specific clinical significance of circRNAs will boost our knowledge of circRNAs in the occurrence and development of CVDs and develop novel, circRNAdependent methods for the diagnosis, treatment and prognosis of cardiovascular diseases.

AUTH O R CO NTR I B UTI O N S
Chen Ding: Writing -original draft (lead). Yafeng Zhou: Funding acquisition (lead); writing -review and editing (lead). in the study design, data collection, analysis, decision to publish or preparation of the manuscript.

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors confirm that there are no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing is not applicable to this article as no new data were created or analyzed in this study.