The novel roles of circRNAs in human cancer

CircRNAs are covalently closed, single-stranded circular transcripts with no 5′ caps and 3′ poly(A) tails; this structural characteristic makes circRNAs resistant to the digestion of ribonucleases, such as RNase R and exonuclease, and confers a longer half-life than that of linear mRNAs [12]. In addition, most circRNAs are evolutionarily conserved across species [13]. CircRNAs are often expressed at low levels [14‐16], implying the possibility that they may act as ‘splicing noise’ with little functional potential. However, multiple circRNAs detected by deep sequencing have been experimentally shown to be expressed more abundantly than their linear counterparts, sometimes even more than 10 times [16, 17]. Most circRNAs are often located in the cytoplasm, consisting of exons, while a small part of circRNAs consisting of introns are located in the nucleus [18], and they are generally expressed in cell type-specific and tissue-specific manners [19].

CircRNAs are produced from precursor mRNA (pre-mRNA), and they are transcribed by RNA polymerase II [20]. The currently discovered circRNAs can be simply sorted into three types according to their different composition and cycling mechanisms: exonic circRNAs, intronic circRNAs and exon-intron circRNAs (EIciRNA). At present, the maturation mechanism of circRNAs is not fully understood. It is inferred that exonic circular RNA is formed by backsplicing [1]. There are currently three hypothetical models explaining the formation of exonic circRNAs: lariat-driven circularization, intron-pairing-driven circularization and RNA binding protein (RBP) mediated circularization [14] (Fig. 1). In the process of forming exonic circRNAs, partial RNA folding occurs during pre-mRNA transcription, and the exon skips along with folding of the RNA. These structural changes result in the formation of specific regions, called lariat structures, in which originally non-adjacent exons become close to each other along with their introns. CircRNA is then formed after the intron sequence is removed by splicing within the lariat structure. This model is defined as lariat-driven circularization. Due to the presence of reverse complement sequences in introns on both sides of pre-mRNA, the complementary pairing of introns on both sides mediates the formation of circRNA. This model is defined as intron-pairing-driven circularization. Additionally, some RNA binding proteins are found to be critical in the formation of circRNAs. The highly conserved RNA-editing enzyme ADAR can bind double-stranded RNAs by targeting double-stranded ALU repeats in human cells [21‐23]. ADAR1 antagonizes circRNA biogenesis through A-to-I editing of intron pairs flanking circularized exons, thus diminishing the complementarity and stability of these intron pair interactions [9, 23, 24]. DHX9, an abundant nuclear RNA helicase, has a unique domain organization that resembles ADAR. Silencing DHX9 leads to increased circRNA production through unwinding RNA pairs flanking circularized exons in general. Interestingly, there is a conserved RNA-independent interaction between ADAR (p150) and DHX9 in mouse and human cells, and co-depletion of DHX9 and ADAR can even promote more circRNA production [25]. Errichelli et al. revealed that FUS regulates circRNA formation by binding introns flanking back-splicing junctions in mouse embryonic stem cell (ESC)-derived motor neurons [26]. Heterogeneous nuclear ribonucleoprotein L (HNRNPL) was found to directly regulate the alternative splicing of multiple RNAs. A recent study found that HNRNPL was also involved in the regulation of circRNAs in prostate cancer [27]. Quaking (QKI), as a kind of alternative splicing factor, promotes circRNA generation by binding to its intronic binding motifs. The addition of a synthetic QKI binding motif into introns was sufficient to induce de novo circRNA formation [28]. With respect to intronic circRNAs, it is currently believed that some introns form lariat structures during splicing, but most will be degraded rapidly by debranching, and only some containing the essential nucleic acid sequences, such as a seven nucleotide GU-rich motif near the 5′ splicing site and an eleven nucleotide C-rich motif near the branch point, will not be debranched after splicing, thus forming intronic circRNA [29]. EIciRNAs are composed of exons and introns. In the formation of exonic circRNAs, introns that surround the exons are usually spliced out, however, in some cases, they are retained and are thus named EIciRNA [30].

With the development of circRNA field, many databases about circRNAs have been built to facilitates the analysis of circRNAs (Table 1). The Circbase, CIRC pedia v2 and Deepbase 2.0 databases contain numerous circRNAs about different species with detailed information [61‐63]. The CircRNADb database contains genomic information, exon splicing, genome sequence, IRES, ORF and references about 32,914 human exonic circRNAs [64]. The Circnet and Starbase v2.0 databases offer researchers the circRNA-miRNA-gene regulatory networks with the application of bioinformatics [65, 66]. The CSCD and CircInteractome databases can be used as tools to predict miRNA response elements(MRE)and RBP [67, 68]. The CirclncRNAnet database provides researchers an easy way to analysis sequencing results [69]. The ExoRBase database provides 58,330 circRNAs existed in human blood exosomes [70]. The circRNADisease database records experimentally verified circRNAs in mutiple diseases [71].

To date, there has been multiple functional cancer-associated circRNAs screened and identified by different research groups (Table 2). These circRNAs play as oncogenes or tumor suppressors in multiple cancers and affect cancer phenotype via diverse ways.

With the development of sequencing technology and bioinformatics, an increasing number of circRNAs are gradually being discovered and have become hotspots in the field of RNA. However, there are still quite a few challenges that need to be addressed. First, there is no uniform naming convention for circRNAs at present, which means researchers in different fields use different names when studying the same circRNA molecules, thus causing confusion. For example, cirs-7 indicates that this circular RNA acts as a sponge for miR-7, and CDR1as refers to a circRNA that is related to the CDR1 gene, but both actually refer to the same circRNA. Second, the current databases of circRNA are still not perfect. Although researchers can predict the function of circRNA by using multiple databases, there is no circRNA database related to tumour prognosis, which makes it difficult to screen functional circRNA after RNA-seq. In addition, although sponging miRNAs is the most classical functional model of circRNA, whether this effect between miRNA and circRNA is universal is questioned. It has been reported that, based on the results of bioinformatics and experimental verification, the circRNA-miRNA functional model has a negative result [81]. Most of the circRNAs currently studied have fewer binding sites for miRNAs, compared with the relationship between CDR1as and miR-7, which greatly reduces the possibility of circRNAs binding to miRNAs. Therefore, the sponge functional model of circRNA and miRNA needs further research. Also, compared with other members of the RNA family, the mechanisms and functional models of the currently reported circRNAs are not diverse, which needs to be further explored, and there are problems for beginners in that the experimental techniques are relatively difficult.