The mechanism and detection of alternative splicing events in circular RNAs

Direct back-splicing

Complementary sequences across long flanking introns (Liang & Wilusz, 2014; Zhang et al., 2014) or the dimerization of RBPs (Ashwal-Fluss et al., 2014; Conn et al., 2015; Li et al., 2017) can facilitate ‘direct back-splicing’ by bridging the splice donor site into close proximity to the acceptor site (Fig. 1A) (Chen & Yang, 2015). Therefore, the branch point upstream of a circularized exon is able to attack the downstream splice donor site, resulting in a pre-mRNA intermediate containing a 2′, 5′-phosphodiester linkage. Subsequently, the upstream splice acceptor is attacked by the free 3′ hydroxyl group of the prospectively circRNA to form a circRNA. It is worthwhile noting that one set of double-stranded RBPs, Adenosine deaminase acting on RNA (ADARs), suppress circRNA expression by destabilizing intron pairing interactions (Ivanov et al., 2015).

Lariat intermediate

Interestingly, inverted repeated elements flanking the circle-forming exons are widespread in mammals but rare in lower eukaryotes. Lariat intermediate mechanism has been proposed in the literature accounting for producing circRNAs (Fig. 1B, left). Using fission yeast (Schizosaccharomyces pombe) as a model system, researchers found that mrps16 gene produced a circRNA by forming an exon-containing lariat firstly (Barrett, Wang & Salzman, 2015). In this model, canonical splicing appears first, forming a lariat precursor (Eger et al., 2018). And then internal back-splicing take place, which result in a double lariat molecular and a circRNA.

According to this approach, alternative exons or introns are excised from the pre-mRNA as exon-intron(s)-exon intermediate molecules. Lariat structure is not permanently stable for the reason that the 2′to 5′phosphodiester bond will be recognized and debranched specifically to linear form by debranching endonucleases (Hesselberth, 2013). However, some lariats are relatively stable and may probably escape from debranching to forming intronic circRNAs (ciRNAs), when contain consensus RNA sequences (a 7-nt GU-rich sequence near the 5′splice site and an 11-nt C-rich sequence near the branch point). These ciRNAs are distinguished from exonic circRNAs by a 2′to 5′phosphodiester bond (Fig. 1B, right) (Lasda & Parker, 2014; Zhang et al., 2013).

Competition of reverse complementary sequences

Considering that most of circRNAs are generated after their parent genes have been transcribed completely, the majority of circRNA isoforms may occur post-transcriptionally (Ashwal-Fluss et al., 2014; Zhang et al., 2016b). Reverse complementary sequences flanking the circularized exons, such as abundant Alu elements (Jeck & Sharpless, 2014), highly conserved mammalian-wide interspersed repeat (MIR) sequence (Yoshimoto et al., 2019) or other non-repetitive complementary sequence (Memczak et al., 2013), are efficient to enhance exon circularization by forming paired duplex structures (Liang & Wilusz, 2014). Shorter sequences as long as 30 to 40 nucleotides are even able to promote circRNA biogenesis (Liang & Wilusz, 2014). Once the intronic complementary sequences across the circularized exons were disrupted, no circRNA could be detected at the examined locus (Zhang et al., 2016b).

Theoretically, a series of inverted repeated RNA pairs form different RNA duplexes resulted in multiple circRNA isoforms (Figs. 3B, 3C). Moreover, an individual intron can also form RNA pairing which promotes canonical splicing to linear RNA formation (Fig. 3A). It means that the competition of RNA pairing leads to different splicing products (Ashwal-Fluss et al., 2014; Wilusz, 2018). Taken together, the endogenous conditions for circRNAs generation are very complex, since the number of repetitive elements, the distance between them, and their degree of complementarity all affect the splicing outcome (Zhang et al., 2014).

RNA binding proteins

Various AS events were observed despite of the presumably identical of paired complementary sequences among all tested human samples (Zhang et al., 2016a), which indicated that the regulation of AS is more complicated than we found. Previous study showed that the RNA pairing was influenced by RBPs as well (Ivanov et al., 2015). These proteins bind to specific intron motifs firstly and then bring the donor site closer to the acceptor site (Ashwal-Fluss et al., 2014; Conn et al., 2015). In contrast, negative RBPs will suppress circRNA formation by destabilizing RNA pairing interactions, like ADAR1 (Ivanov et al., 2015; Rybak-Wolf et al., 2015). Differentially expressed RBPs have been reported to mediate pre-mRNA AS in various cell lines, for example, SR and SR-related proteins typically activate splicing (Nilsen & Graveley 2010). By contrast, hnRNPs are generally thought to repress splicing. The activity of these proteins can be regulated through signaling networks in specific tissues and cell-types (McManus & Graveley, 2011). Hence, we hypothesize that AS in circRNAs may be regulated by similar mechanism.

Others

It is well known that pre-mRNA processing is closely correlated with polymerase II (Pol II) transcription, and the transcription elongation rate has obvious effect on the occurrence of splicing events (Bentley, 2014). Flies with decreased elongation capacity of RNA pol II, produced a significantly lower number of circRNAs (Ashwal-Fluss et al., 2014). By applying 4-thiouridine (4sU) to metabolic tagging of nascent RNAs, Zhang et al. discovered that circRNA-producing genes had higher average transcription elongation rate (Zhang et al., 2016b). In summary, the positive correlation between nascent circRNA generation and Pol II elongation speed indicates that fast elongation favors RNA folding across flanking introns to form circRNAs rather than within introns (Bentley, 2014; Zhang et al., 2016b).

Nevertheless, those factors mentioned above still cannot adequately explain the widespread nature of AS in circRNAs. Additional elements that are responsible for alternative circularization await identification. It is helpful to discover the rules of AS in circRNAs by large-scale identification of full-length circRNAs. Currently, lots of endeavors are focused on the internal structure of circRNAs which further promote their functional characterization and evolutionary analyses.

CircRNA enrichment methods

Current studies show that circRNAs abundance is approximately ≤10% of their corresponding linear RNA (Salzman et al., 2013), and the estimation of circRNA isoforms with lower expression level probably be biased. Therefore, circRNA enrichment is needed prior to RNA library construction and sequencing. Due to the great variability length of circRNAs, it is hard to separate them from other RNA species by size or electrophoretic mobility. Owing to the covalently closed structure, most circRNAs show higher tolerance to Ribonuclease R (RNase R) digestion in comparison with linear RNAs. Nevertheless, utilizing RNase R alone is not sufficient to remove RNAs lacking 3′overhangs. Therefore, researchers pretreated the RNA samples by combining ribosome RNA (rRNA) depletion with RNase R treatment (Jeck & Sharpless, 2014) or depletion of poly(A)+ RNA to ensure that sequencing reads are generated from bona fide circRNAs.

Whereas, some linear RNAs without sufficiently long single-stranded 3′overhangs, such as small noncoding RNAs, are naturally resistant to RNase R (Vincent & Deutscher, 2006). Besides, many polyadenylated mRNAs with complex structures, especially G-quadruplex (G4) structures, are also poor substrates for RNase R, as the enzyme stalls in the body of these transcripts (Panda et al., 2017; Xiao & Wilusz, 2019). Considering that the expression of circRNAs is often lower than the linear counterparts, even a small amount of undigested linear RNAs may surpass the abundant of cognate circRNAs (Panda et al., 2017). To resolve this major obstacle, Panda et al. raised another approach to isolate high-purity circRNA termed RNase R treatment followed by Polyadenylation and poly(A)+ RNA Depletion (RPAD) (Fig. 4A) (Panda et al., 2017; Pandey et al., 2019). After RNase R treatment, small RNAs were decreased to 10–50% of the remaining by polyadenylation and poly(A)+ RNA depletion. In an alternative manner, Xiao et al. found that RNase R could digest linear RNAs more efficiently by using Li+ instead of K+ (which stabilizes G4s) in the reaction buffer (Xiao & Wilusz, 2019). Drawing inspiration from the RPAD method, purer circRNAs can be isolated by coupling A-tailing with RNase R in LiCl-containing buffer (Fig. 4B) (Xiao & Wilusz, 2019).

Compared with conventional models of circRNA enrichment using rRNA depletion kits of model organisms, the other two methods can be used in non-model organisms for efficient removal of linear RNAs regardless of the RNA sequences. A noteworthy drawback of RPAD is that quantification of circRNA abundance may not be strictly accurate due to involvement of many enzymatic and RNA isolation steps. Furthermore, according to specific experiment purpose, RPAD protocol was modified via adding rRNA depletion step before RNase R treatment (Rahimi et al., 2019). It is worth noticing that extensive purification steps will cause significant biases on circRNA abundance.

Genome-wide annotation of full-length circRNAs

At present, a combination of RNA-seq analysis with bioinformatic survey has been widely used for large-scale determination of circRNAs. Moreover, various algorithms have been raised to analyse sequencing information generated from different platforms.

Experimental validation of full-length circRNAs by rolling circle amplification

Compared with bioinformatics reconstruction, experimental approaches provide precise sequence of specific circRNAs most intuitively. Hence, follow-up experimental validation for assembled circRNAs may then be applied to further strengthen the case for existence of many circRNAs. As reverse transcriptase has potential strand displacement activity (Kelleher, 1998), it allows the displacement of any complementary sequence hybridized downstream to produce many copies of the same template (Acevedo, Brodsky & Andino, 2014).

Owing to the unique structure of circRNAs, multiple rounds of reverse transcription could produce double or triple sized products (Barrett, Wang & Salzman, 2015; Danan et al., 2012; Starostina et al., 2004). You et al. (2015) for the first time, detected the full-length of circRNAs by sequencing of rolling circle cDNA products on PacBio platform. Through creating circular-derived PCR products with divergent primers, Hirsch et al. combined ONT sequencing with a PCR-based approach to gain insight into the internal structure of circNPM1 (Hirsch et al., 2017). Moreover, Das et al. presented a novel method, named circRNA-Rolling Circle Amplification (circRNA-RCA), aiming to distinguish circRNA variants containing the same BSJ site but distinct in internal sequence (Fig. 5). In this method, a forward primer spanning the BSJ sequence and a reverse primer exactly upstream of the forward primer were used to produce tandem-repeat cDNA amplicons with the supplementation of RNase H-minus reverse transcriptase (Das et al., 2019). However, the main drawback of these methods is that only a handful of circRNAs can be examined and the sensitivity and /or specificity is not satisfactory.