Coding potential of circRNAs: new discoveries and challenges

Molecular structure

Analogous to similar ncRNAs

Recent studies demonstrate that many lncRNAs are able to translate into functional polypeptides. In 2013, Magny et al. (2013) found a putative noncoding RNA 003 in 2L (pncr003:2L), including two potentially functional smORFs in the fly’s heart, which could translate into bioactive peptides and synergistically regulate cardiac calcium uptake. In 2015, Anderson et al. (2015) discovered an annotated lncRNA that translates for a conserved micropeptide- myoregulin (MLN) that functions as a regulator of skeletal muscle physiology. One year afterward, Nelson et al. (2016) found that a peptide (named dwarf open reading frame (DWORF)) is encoded by a putative lncRNA. This peptide is mutually exclusive with the other three inhibitors (phospholamban, sarcolipin, and myoregulin) to competitively combine with the SEARC pump to adjust the reuptake of the Ca2⁺ in muscle. Then, Matsumoto et al. (2017) identified a functional novel polypeptide encoded by a lncRNA. This peptide can negatively regulate mTORC1 activation by interacting with the lysosomal v-ATPase in late endosome/lysosome. With deep research, increasingly more lncRNAs with the capacity of translating proteins (peptides) will be explored. As a special type of lncRNAs, we have reason to speculate that the biological significance of coding ability of circRNAs is still to be uncovered.

Early exploration findings

The first indications of a translational role for circRNAs emerged from studies of virus nucleic acids. One of the first observations of a circRNA behaving as a translational template was made with the single-stranded circular RNA genome of the hepatitis δ virus, a satellite virus of the hepatitis B virus; encapsulation of the former by hepatitis B virions was found to result in the production of a single viral protein of 122 amino acids, in a noncanonical manner (Kos et al., 1986). In 1995, Chen & Sarnow (1995) demonstrated that synthetic circRNAs containing IRES elements were able to correctly translate into polypeptides in rabbit reticulocyte lysate, but those without IRES could not. Furthermore, they speculated that this type of RNA can translate along the RNA circles for multiple consecutive rounds. In 1998, Perriman et al. used plasmids for creating RNA cyclase ribozymes to produce desired circular RNAs that were inserted into the green fluorescent protein (GFP) ORF (finite GFP encoding) and stop codon-devoid GFP reading frame (infinite GFP encoding) (Perriman & Ares Jr, 1998). The authors showed that both circRNAs can directly translate along with GFP in Escherichia coli strains and in the meantime, the infinite GFP-encoding RNA could be translated into an extremely long repeating poly-GFP. These findings validated Chen’s previous prediction in 1995 (Chen & Sarnow, 1995). In 1999, Li & Lytton (1999) observed that a circRNA containing NCX1 exon 2 might translate for a protein. It is a pity that they could not detect a protein corresponding exactly to what they predicted from the circular transcript; however, when the circRNAs were made into linear RNAs and transfected into HEK-293 cells, the linear versions of circRNAs were shown to result in the proteins of the expected size of ∼70 kDa, and the transfected cells possessed Na/Ca exchange activity.

Over a decade later, Wang & Wang (2015) reported on their construction of an efficient back-splicing circRNA, which could be translated into functional GFP proteins in human and Drosophila cell lines. Furthermore, due to the nuclease resistance characteristics of circRNAs, when the cell was transfected with circRNA, protein production was prolonged for several days. In the same year, Abe et al. (2015) provided evidence that circRNAs were translated into infinite FLAG proteins in rabbit reticulocyte lysate and HeLa cells with an infinite ORF in the absence of any particular translation initiation element such as a poly-A tail, internal ribosome entry, or a cap structure. This series of experiments proves that artificial circRNAs with stop codon mutations have a rolling circle amplification (RCA) mechanism to code for long repeating poly proteins. In 2014, AbouHaidar et al. (2014) reported a small new virusoid with covalently closed circular (CCC) RNA (220 nt) associated with rice yellow mottle virus that could translate into a 16-kDa highly basic protein. This example is the only one that codes proteins among all known viroids and virusoids. This unique natural supercompact “nano genome” even overlaps its initiation and termination codons to UGAUGA (AbouHaidar et al., 2014).

Nevertheless, all these scattered reports, however, are limited to viruses, bacteria, or synthetic circRNA (Granados-Riveron & Aquino-Jarquin, 2016) (Table 1), and the translation ability of endogenous circRNAs still requires further exploration.

Solid evidence for endogenous circRNA direct translation

In 2013, Jeck et al. (2013) reported that circRNAs are abundant, conserved and associated with ALU repeats, but there are no detectable levels of exonic circRNAs in the ribosome-bound fraction (via ribosome profiling). One year later, Dudekula et al. (2016) raised doubts about this conclusion when they reported their findings from a bioinformatic analysis; IRES regions in circRNAs represented predicted binding sites for RNA binding proteins, including some known to modulate IRES-driven translation.

In 2017, it was finally proved that endogenous circRNAs are capable of directly translating into proteins. By using ribosome footprinting and immunoprecipitation of Drosophila brain tissues, Pamudurti et al. demonstrated that circRNA sequences could be bound by ribosomes including the termination codon (Pamudurti et al., 2017). They focused on circ-Mbl from the Mbl gene among all of the ribo-circRNAs and repeatedly verified that circ-Mbl could translate into protein. Through the construction of an overexpression vector, the substitution of the ORF with a split Cherry molecule and target mass spectrometry from the Drosophila brain circ-Mbl was immunoprecipitated. In the same year, through a screening study of circRNAs related to human, mouse (C2,C12) and a Duchenne muscular dystrophy disease model, Legnini et al. reported that the circ-ZNF609 combined with ribosomes and that its encoded protein was suggested to be involved in the myoblast growth process; however, the circ-ZNF609 was found to be translated at almost two orders of magnitude lower efficiency than that of the linear form (Legnini et al., 2017).

Thereafter, Yang et al. (2017b) explored circRNA translation ability by the same approach and found that control sequences without IRES were also capable of translating the target protein. These unexpected circRNA translation events were initiated by eIF4G2 and eIF3A and associated with the m⁶A modification. When the m⁶A modifications were “erased”, the target protein translation activity was substantially affected, to the point that it completely disappeared. Ribosome spectrum analysis confirmed that a multitude of endogenous circRNAs were bound by ribosomes, but whether these circRNAs harbored any IRESs was not examined. Finally, high-throughput sequencing analysis determined that approximately 13% of the total circRNAs carried the m⁶A modification. Months later, another independent study showed that circRNAs carry extensive m⁶A modifications and are expressed in cell type-specific patterns (Zhou et al., 2017). The writing and reading machinery of these m⁶A modifications were found to be similar to those of mRNAs (i.e., involving the METTL3/14 and YTH proteins) but were distinctive in their location patterns; the data also suggested that the m⁶A modification did not appear to promote degradation of circRNAs as it does for mRNAs. Ultimately, interpretation of these findings indicates that switching the state of m⁶A modifications may allow for functional control of circRNAs.

In 2018, Yang et al. (2018) reported that the circ-FBXW7 can translate for a new protein FBXW7-185aa during glioma tumorigenesis. Intriguingly, this protein cooperates with FBXW7, which is encoded in their parental genes, to control c-Myc stability and repress cell cycle acceleration and the consequent proliferation. This is the first study to provide definitive evidence of protein translation via circRNA synergy with the protein expression by parental genes and joint function of the proteins. Zhang et al. (2018a) further reported that circ-SHPRH, a circRNA containing an IRES-driven ORF, translates into a functional protein. For this process, circ-SHPRH utilizes overlapping genetic codes to create a UGA stop codon, causing translation of the SHPRH-146aa protein. The translated SHPRH-146aa functions as a protector of the full-length SHPRH protein, guarding against degradation by the ubiquitin proteasome and consequently inhibiting cell proliferation and tumorigenicity in human glioblastoma. In the same year, Zhang et al. (2018b) found that the 1084 nt CircPINTexon2 which generated by the circularization of exon 2 of LINC-PINT encodes an 87-aa peptide. This peptide (PINT87aa) suppresses glioblastoma cell proliferation in vitro and in vivo by directly interacting with polymerase associated factor complex (PAF1c) and inhibiting the transcriptional elongation of multiple oncogenes.

In 2019, Zheng et al. (2019) reported an upregulated circRNA (circPPP1R12A) in colon cancer tissues that could translate a 73-aa protein (circPPP1R12A-73aa). The circPPP1R12A-73aa promotes the proliferation, migration and invasion abilities of colon cancer via activating hippo-YAP signaling pathway. In addition, Liang et al. (2019) reported circ β-catenin originated from β-catenin gene locus could promote tumorigenesis. Knockdown of circ β-catenin repressed liver cancer cell growth and migration in vitro and in vivo by inhibiting Wnt/ β-catenin pathway. In terms of mechanism, circ β-catenin encoded a novel protein (β-catenin-370aa) which shared homologous N-terminus sequence with wild type β-catenin, but it contained a new C-terminus with 9 specific amino acids. The β-catenin-370aa might function as a decoy for GSK3 β, leading to escape from GSK3 β-induced β-catenin degradation (Liang et al., 2019).

Interestingly, Zhao et al. (2019) in the same year identified a virus-derived circRNA, HPV16-circE7, which could translate E7 oncoprotein. By constructing various mutant vectors (such as circE7_noATG) for comparison, the authors found that circE7 can also provide the template for E7 oncoprotein translation. Further studies have found that the initiation of circE7 translation may be related to m6A modification and capable of generating the E7 oncoprotein in a heat-shock regulated manner. Moreover, HPV16 circE7 is essential for the transformed growth of CaSki cervical carcinoma cells and could be regulated by keratinocyte differentiation (Zhao et al., 2019).

In 2020, Weigelt et al. (2020) showed that circSfl was highly upregulated in all tissues by next-generation sequencing of wild-type and mutant flies. However, circSfl is lack of enrichment of miRNA binding sites in loop, which makes it unlikely acts as a miRNA sponge. Further study verified that circSfl is translated into a small protein that shares the N terminus with full-length Sfl protein. Furthermore, the protein encoded by circSfl and the protein encoded from the linear Sfl transcripts can positively extend the lifespan of female flies (Weigelt et al., 2020), indicating the unique role of circSfl in fly life.

Detailed information for the published coding circRNAs is summarized in Table 1.