Emerging Roles of Plant Circular RNAs

Based on their genomic origin, circRNAs are categorized into three types: exonic, intronic and intergenic. They have independent modes of biogenesis, but all related to the splicing mechanism of eukaryotes. Typical eukaryotic genes contain exons and introns that are spliced out from primary messenger RNAs (mRNAs) by the spliceosome machinery. A conventional collinear splicing event is fulfilled in two sequential transesterification reactions. First, the 2′-hydroxyl end of the adenosine at the intron branch-point attacks the 5′ nucleotide (i.e. the 5′ splice site) of the excised intron to form a lariat by a 2′-5′ phosphodiester bond. This reaction generates a free 3′-hydroxyl end on the 5′ exon, which then attacks the first nucleotide immediately after the excised intron (i.e. the 3′ splice site) to join together two exons in a 5′ to 3′ orientation and release the intronic lariat consisting of a loop and a 3′ tail. The intronic lariat without the 3′ tail is a perfect circular RNA, i.e. intronic circRNA. Exonic circRNA is generated by a backsplice event, in which the free 3′-hydroxyl end of an exon (i.e. “tail”) released at the first reaction of a conventional collinear splicing event joins to the 5′ “head” of an exon that is normally upstream to form a “tail to head” RNA structure, i.e. 5′ exon is shuffled downstream of 3′ exon. When the 3′-tail and the 5′-head are from the same exon, a single-exon circRNA is generated, usually by a cotranscriptional mechanism. When the 5′-head is from another upstream exon, an intermediate lariat containing exons and intron(s) is formed. In this case, a circRNA with multiple exons can be produced after removing the intron(s) posttranscriptionally by the conventional, i.e. linear, splicing mechanism. Most backsplicing events are cis-ones that occur within the same or different exons of a single gene, but trans-backsplicing events are also quite common. For instance, we found that ~13% and ~34% of the identified circRNAs in Arabidopsis thaliana and rice (Oryza sativa), respectively, are potential products of trans-backsplicing events ¹¹. Involvement of spliceosome in generation of circRNAs suggests that 1) circRNAs should contain canonical splice signals, and 2) circRNA production competes with linear pre-mRNA splicing. Studies have shown that exon circularization requires canonical splice signals ¹², and that majority (70%) of circularization events use canonical splice signals in animals and humans ¹³. However, we found that the vast majority of rice circRNAs (92.7%) were flanked by a diverse set of non-GT/AG canonical splicing signals, suggesting that plants may use alternative mechanism(s) for circRNA biogenesis ¹⁴. It has been reported that production of circRNAs has a positive or negative effect on expression of their parental gene ^{15, 16}. A recent study reported that circRNA production is enhanced when core spliceosomal components are depleted and/or canonical linear mRNA splicing events are inhibited ¹⁷. These observations suggest that the relationship between circRNAs and their parental genes is quite complicated.

Several characteristics of animal circRNAs, such as longer flanking introns, low expression levels and tissue-specific expression profiles, are shared by plant circRNAs ^{11, 15, 18, 19, 20, 21}, but plant circRNAs have their own unique features. A number of cis-elements and trans-factors, particularly the unique features and structures and specific binding sites of splicing factor of the flanking intronic sequences of circRNAs, have been shown to be important for regulation of biogenesis of circRNAs in animals and humans ^{15, 22, 23}. Several transcriptome-wide analyses have shown that repetitive and reverse complementary sequences, for example Alu repeats, are enriched in the flanking introns of circularized exons in animals and significantly correlated with exon circularization ^{24, 25, 26}. Compared to animals, plant exonic circRNAs are less likely to be generated from exon(s) flanked by introns containing repetitive and/or reverse complementary sequences ^{11, 19, 21, 27}. For instance, of the 13 validated plant circRNAs, only two (~15%) contain >15-bp reverse complementary sequences in their flanking introns 11. In cotton (Gossypium spp.), despite circRNAs seem to have more repeat sequences in their flanking introns than linear genes, only ~10% of exonic circRNAs are associated with reverse complementary intronic sequences ²⁸. Nevertheless, short (4-11 bp) complementary sequences found near the splice sites might be important for circRNA production in Arabidopsis and some trans-spliced circRNAs found in chloroplast, an organelle enriched for circRNAs, have longer complementary sequences (>20 bp) near the splice sites ²⁷. A recent study in maize (Zea mays) found that LLERCPs (reverse complementary pairs of LINE1-like elements) are significantly enriched in the 35-kb, particularly in the 5-kb, flanking regions of circRNAs ²⁰. The study also found that circRNAs with LLERCPs have an expression level significantly higher than those without LLERCPs nearby, indicating LLERCPs could reinforce the expression of circRNAs, although the number of LLERCPs seem not to be related to the expression level of circRNAs ²⁰. Because LLERCPs were found in a relatively large flanking region of circRNAs, it is of interest to know how they are related to circRNA biogenesis. It is also of interest to know whether repeat sequences located at the flanking introns of circRNAs are associated with genome complexity so that large and polyploid genomes tend to have more repeat sequences in their flanking introns of circRNAs.

Despite most fundamental features of plant circRNAs are evolutionary conserved in different plant species, some are only observed in certain plants. The majority exonic circRNAs in rice and Arabidopsis contain 1-4 exons ^{11, 19, 21}. Large parental genes with multiple shorter exons are preferentially circularized ^{19, 20}. Multiple circRNAs can be generated from a single parental gene through alternative backsplicing and circularization ^{11, 14, 19, 29}. In rice, ~6% of exonic circRNAs had their expression levels significantly positively correlated with the expression levels of their parental genes, and no negatively correlated pair was found ¹¹. In kiwifruit (Actinidia spp.), both exonic and intronic circRNAs were significantly positively correlated to parental genes ³⁰. This is different from the findings observed in maize and barley, in which no correlation between the accumulation of circRNAs and their corresponding parental genes was found ^{20, 31}.

Parental genes of over 700 exonic circRNAs (~15% of Arabidopsis circRNAs) are orthologues between rice and Arabidopsis ¹¹. Approximately 34% and 55% of circRNA-producing soybean genes are conserved orthologs in Arabidopsis and rice, respectively ²¹. These results suggest conservation of circRNAs in plants. In cotton (G. hirsutum), ~15% of circRNAs are conserved in either Arabidopsis or rice, and ~7% are conserved in both Arabidopsis and rice ²⁸. However, of the 2804 maize circRNAs, only 47 (<2%) are conserved between maize and rice, and three are conserved between maize and Arabidopsis ²⁰. In view of the circRNA feature of tissue-specific expression, this low conservation of circRNAs could be related to different tissues used in different plants, but it could also be contributed by use of different criteria. Conservation of plant circRNAs demands more robust investigations by applying unified criteria to circRNAs from multiple plant species.

Soybean (Glycine max) is a partially diploidized tetraploid and about three quarters of its genes have paralogous copies. It was found that ~83% of soybean circRNAs are generated from paralogous genes but from different positions although paralogous gene pairs usually have high sequence similarity. In addition, paralogous circRNAs showed tissue-specific expression pattern ²¹. In cotton, ~20% of circRNAs are conserved between G. hirsutum and its two potential ancestors (G. arboreum and G. raimondii), and ~44% of circRNAs are unique to each individual species ²⁸. These results suggest the presence of paralog- and species-specific circRNAs. Identification and characterization of homoeologous circRNAs from other polyploids will further shed lights on whether these features are conserved in polyploid species.

Presence of circRNAs in plants was first demonstrated in 2014 ³². We then first did genome-wide identification of circRNAs in Arabidopsis and rice and found a large number of circRNAs in both species, demonstrated widespread of circRNAs in plants ¹¹. Since then, a number of circRNA identifications have been carried out in other plant species, including wheat (Triticum aestivum), barley (Hordeum vulgare), maize, tomato (Solanum lycopersicum), potato (Solanum tuberosum), soybean, cotton and kiwifruit (Table 1).

Dozens to thousands of circRNAs were found in these species. The vast majority of these are nuclear circRNAs, but circRNAs derived from chloroplasts and mitochondria have also been found. For example, 6% and 1% of the 803 Arabidopsis circRNAs identified by Sun et al. (2016)²⁷ are derived from chloroplast and mitochondrion genes, respectively.

Most de novo identification studies used ribosome RNA depleted and RNase R ³³ treated total RNA for creation of sequencing libraries with an aim to enrich circRNA species. CircRNAs were also successfully identified in publically available RNA-seq datasets that were generated using poly(A) selected mRNAs ^{14, 34, 35}. Regarding the algorithms used in circRNA identification, almost all studies adopted the pipelines or programs developed for animals and humans. That is one of the potential reasons contributing to high false positive rate of the identified plant circRNAs, particularly those from non-coding genomic regions, such as introns and intergenic regions. For example, in our experiments, 10 of the 18 exonic circRNAs were confirmed by sequencing PCR products amplified by divergent primers, but none of the 30 predicted non-exonic circRNAs was confirmed ¹¹. Some studies applied multiple programs to the same set of data for circRNA prediction ^{20, 29}. Similar to what has been previously reported ³⁶, poor overlapping was evident. For instance, <21% of the maize circRNAs could be simultaneously identified by CIRI, CIRC_FINDER and CIRCexplorer2 ²⁰, suggesting poor accuracy and/or sensitivity when using the programs designed for animals and humans in plants. To improve the accuracy and sensitivity of plant circRNA prediction, we developed a program termed PcircRNA_finder based on the features of plant genome and transcriptome. Based on the testing results using simulated and real RNA-seq datasets, PcircRNA_finder is much more sensitive and effective than find_circ and CIRCexplorer ³⁷.

Exonic circRNAs are the focus of most studies in both animals and plants; however, the proportion of exonic circRNAs is significantly different in different plant species. In Arabidopsis and tomato, ~86% of circRNAs were derived from exon(s) ^{11, 38}. In rice, only about half identified circRNAs are derived from exon(s) ^{11, 19}. In cotton, ~60% (G. arboreum), ~84% (G. raimondii) and ~46% (G. hirsutum) of circRNAs are exonic ²⁸. In wheat, soybean and kiwifruit, only ~7%, ~16% and ~21% are exonic circRNAs, respectively, the majority are intergenic circRNAs ^{21, 30, 39}. One of the future tasks would be to find the reason(s) behind the discrepancy. Is it due to 1) species difference, 2) poor annotation of genome sequences, or 3) only a small portion of the circRNA repertories have been revealed so the circRNAs identified are biased?

Assembling full-length circRNAs is important for understanding the internal structures, alternative circularization and biogenesis of circRNAs. Considering that nearly all circRNA identification tools can only find the backsplicing site of candidate circRNAs, we developed a pipeline, circseq_cup, for assembling full-length circRNAs in plants 14. The core of the circseq_cup algorithm is to identify and align back-spliced junction reads and their corresponding paired reads. Using this principle, we successfully assembled ~3,000 full-length circRNAs in rice. Similar rationale has also been used by the CIRI-AS algorithm for detection of the internal structures of circRNAs in human and fruit fly ⁴⁰. Based on the assembly rationale of circseq_cup and CIRI-AS, theoretically if the fragments used in sequencing library preparation are large enough and/or the length of RNA-seq reads are long enough, circRNAs of any full-lengths could be assembled and identified. To support this, we found that, for RNA-seq data with an insert size of 300-bp and 400-bp, the longest full-length circRNAs assembled in rice were 488 and 680 bp, respectively, although the majority of full-length circRNAs in rice are 150-180 bp long ¹⁴. The study by Gao et al. (2016)⁴⁰ also found that a large proportion of alternative spliced circRNA exons can hardly be detected in linear mRNAs and are enriched with binding sites of splicing factors distinct from those enriched in linear mRNAs, suggesting that algorithms based solely on backsplicing junctions may not be enough for identification of all types of circRNAs.

Given the low expression levels of most circRNAs, their tissue-specific expression profile and their complexity of alternative splicing, the number of circRNAs identified so far are likely to be just the tip of the iceberg. To enhance identification of plant circRNAs, particularly full-length circRNAs, to improve the prediction accuracy and to better understand plant circRNA biogenesis, it is of great importance to develop robust novel algorithms incorporating the characteristics of plant genomes and circRNAs, and to adopt new sequencing technologies generating long reads or full-length transcripts so that the internal structures, alternative splicing events and contribution of trans-backsplicing can be thoroughly evaluated.

In animals and humans, circRNAs have been shown to be involved in a wide range of biological processes through various molecular mechanisms ^{9, 41}. They can serve as miRNA sponges to attenuate miRNA-mediated regulation ^{42, 43}, regulate the functions of RNA-binding proteins by direct interaction in a way as they modulate miRNA activity ^{15, 44}, and regulate the transcription of their parental genes ^{15, 16}. Given the potential large repertory of circRNAs, their origin from genes with important biological roles, and the relationship between circRNAs and their parental genes, it is not surprise to uncover diverse biological functions of plant circRNAs. A recent study has demonstrated a role of circRNAs in development of Arabidopsis floral organs through a novel molecular mechanism ⁴⁵, but until now, most plant circRNA related studies were still at the preliminary level with a focus on prediction of potential circRNAs acting as miRNA sponges, analysis of the expression correlation between circRNAs and their parental genes and the co-expression networks involving circRNAs.

A number of potential plant circRNAs have been bioinformatically predicted to have miRNA sponge capacity because the presence of miRNA binding sites ^{38, 39}. The number of circRNAs with potential miRNA binding site(s) and the number of potential miRNA binding site(s) in each individual circRNAs vary significantly in different studies and plant species. We found only a small proportion of circRNAs (6.6% and 5.0% in rice and Arabidopsis, respectively) with the potential capacity acting as target mimics of miRNAs ¹¹. In another rice study, although 235 (~17%) of the 1356 exonic circRNAs were found to contain putative miRNA binding sites, only 31 have two or more miRNA-binding sites, which is significantly less than that reported for the human circRNA CDR1as ^{19, 43}. Similarly in maize, circRNAs have an average of 1.33 miRNA binding sites but the number of circRNAs with miRNA binding site is very low (only 15 out of 2804) ²⁰. These results do not provide strong evidence for the enrichment of miRNA binding sites in plant circRNAs.

To verify experimentally whether circRNAs can function as miRNA sponges, transgenic rice plants overexpressing Os08circ16564, a rice circRNA predicted to contain binding sites of miR172, was generated ¹⁹. Overexpressing the circRNA reduced the expression level of its parental gene but without affecting the expression level of miR172, a miRNA shown to be a crucial regulator of spikelet and floral organ development in rice ^{46, 47}. Consequently, no visible phenotypic changes in spikelet or floral organs were observed in the overexpressing rice plants ¹⁹. These results do not support Os08circ16564 functioning as a miR172 sponge.

In fact, precautions should be taken when predicting miRNA binding site(s) of circRNAs. Unless the miRNA binding site is located at the newly formed backsplicing junctions or the circRNA-forming exon(s) are skipped during splicing of the corresponding linear mRNA, the predicted binding site exists in both the circRNA and its related linear transcripts, and its belonging is unable to be distinguished bioinformatically. Sophisticated experiments are thus required to evaluate separately the expression changes of circRNA and its related linear transcripts, and to assess their relevant biological consequences.

CircRNAs are able to bind to RNA polymerase II complex by interacting with the U1 snRNA at their parental loci to regulate the transcription of their parental genes. It has been proposed that circRNAs localize or position the U1 snRNA so that U1 can stimulate RNA polymerase II activity ¹⁶. A recent study in Arabidopsis revealed a new mechanism, by which circRNA regulates the transcription of its parental gene ⁴⁵. It was found that circRNA could bind to the DNA of its host gene through R-loop, a structure formed by hybridization of RNA and DNA, and control linear alternative splicing, and consequently the composition and expression levels of alternatively spliced variants. The Arabidopsis SEPALLATA3 (SEP3) gene, an important MADS-box transcription factor involved in floral organ development, is alternatively spliced under low temperature conditions, producing a splicing variant (SEP3.3) without the 6th exon and a circRNA derived from the skipped exon. Transgenic Arabidopsis plants (Exon6-circRNA) overexpressing the circRNA derived from the 6th exon of SEP3 were generated to investigate the potential functions of the circRNA. It was found that experimental overexpression of the Exon6-circRNA led to an increased level of SEP3.3 and caused mutant floral phenotypes (reduced stamen number and increased petal number) comparable with that observed in the transgenic plants overexpressing SEP3.3. Dot blot experiments showed that Exon6-circRNA hybridized to DNA of the host locus by forming R-loops ⁴⁵. Together, these results suggest the ability of the skipped exon derived circRNA in stimulating exon skipping in its parental mRNA, probably by increasing pausing of RNA polymerase II due to the formation of stable R-loops. Whether this is a general function of many circRNAs in eukaryotes remains an open question.

The study by Conn et al. (2017)⁴⁵ provided evidence for the association between production of circRNAs and exon skipping in their host mRNAs. The same study also detected reliable exon-skipping events in 11 out of the 12 most abundant exonic circRNAs identified in Arabidopsis leaf RNAs 11. These results suggest the possibility of using the abundantly expressed circRNAs as biomarkers for prediction of alternative splicing variants with skipped exon(s) ^{45, 48}.

In tomato, circRNAs derived from several genes associated with fruit ripening and coloring, including Phytoene Synthase 1 (PSY1) and Phytoene Desaturase (PDS), key genes involved in carotenoid biosynthesis, were identified 29. The expression levels of both PSY1 and PDS and their circRNAs (PSY1-circ1 and PDS-circ1) are gradually increased during the period of fruit ripening (color changed from green to red). When PSY1-circ1 or PDS-circ1 was overexpressed, transgenic fruits showed yellow instead of red. In these yellow fruits, the expression levels of PSY1 and PDS are significantly decreased compared to that of wild-type. Consistent with this down-regulated levels of PSY1 and PDS, the contents of lycopene and β-carotene in the yellow transgenic fruits are significantly lower than those in wild-type 29. These results suggest that over-production of PSY1-circ1 or PDS-circ1 has a negative effect on transcription and/or accumulation of its parental gene, and subsequently reduces accumulation of red pigments. The molecular mechanism underlying the circRNA over-production induced repression of parental gene is yet to be uncovered.

In human, an intronic circRNA, ci-ankrd⁵², is associated with the elongating form of RNA polymerase II at its host locus and acts as a positive regulator of the transcription speed of its host gene. Knock-down ci-ankrd⁵² reduces the expression level of its host gene. The stimulating role of ci-ankrd^[52 could be due to its direct cis-interaction with its host gene or decreased rate of including downstream introns with premature stop codons in the mature linear host mRNA, which is known to reduce mRNA abundance ⁴⁴. It has also been speculated that intronic circRNAs may potentially have effects on the expression of whole transcriptome because ci-ankrd⁵² can bind to genetic loci in addition to their host loci ⁴⁴. A recent study in Arabidopsis provided evidence for such a mechanism ⁴⁹.

miRNAs are small non-coding RNA molecules and play an important role in plant development and stress tolerance by regulating the expression levels of their target genes via mRNA cleavage and/or translational repression. Biogenesis of miRNAs involves dicing of primary miRNAs (pri-miRNAs) by the DCL1/HYL1 complex. Intron lariats are by-products of linear mRNA splicing and are usually quickly degraded by the RNA debranching enzyme 1 (DRB1). This biological process is critical because dysfunctional DRB1 results in embryo lethal in both animals and plants. Arabidopsis non-lethal mutant dbr1-2 accumulated a large number of RNase R resistant circular intron lariat RNAs, i.e. intronic circRNAs, and showed global reduction in miRNA accumulation due to intronic circRNA-mediated sequestration of the DCL1/HYL1 complex, which competitively inhibited the binding of the complex with pri-miRNAs and reduced accumulation of miRNAs. This observation was further confirmed by overexpression of an intronic circRNA (lariat41) derived from the first intron of At5g37720 ⁴⁹. Arabidopsis plants overexpressing the circRNA (lariat41-OE) showed curly leaves, late flowering and lower fertility ⁵⁰, similar to phenotypes observed in some miRNA mutants. In addition, the expression level of FT, a key flowering time regulator, was significantly reduced in the lariat41-OE plants, which could also contribute to the late flowering phenotype. These results suggest that intronic circRNAs are normally by-products of splicing and degraded by DBR1, which acts as a safeguard to ensure the proper biological functions associated with the miRNA pathways, and that unusual accumulation of intronic circRNAs could be potentially used as biomarkers for certain functional defaults. These speculations and the roles of intronic circRNAs are waiting to be confirmed because visible mutant phenotypes were only observed in plants overexpressing lariat41 and not in plants overexpressing one of the other 16 intronic circRNAs that were significantly enriched in the dbr1-2 mutant ⁵⁰.

Similar to enrichment of intronic circRNAs in the dbr1-2 mutant 49, differential expression of circRNAs under certain biotic or abiotic stress conditions have been investigated in several plants. CircRNAs specifically expressed at a specific developmental time point/stage or induced/repressed by certain stresses could have their roles related to certain development events at that time point/stage or in fighting against the stresses. In other words, differentially expressed circRNAs may act as important functional regulators involved in developmental-/stress-specific biological processes in plants.

We identified 27 rice exonic circRNAs differentially expressed under phosphate-sufficient or -starvation conditions 11. In response to cold and heat treatment, 163 and 1583 circRNAs were differentially expressed in tomato and Arabidopsis, respectively ^{51, 52}. Under heat stress conditions, the length of circRNAs increased probably due to the increased number of exons involved in circularization, suggesting that heat stress changed not only the quantity but also the quality (i.e. alternative backsplicing and circularization) of circRNAs ⁵¹. In wheat, 62 out of the 88 identified circRNAs were found to be differentially expressed in the seedlings under dehydration stress conditions ³⁹. CircRNAs responsive to biotic stresses have also been reported. For instance, 584 circRNAs responsive to infection of Pseudomonas syringae pv. actinidiae (Psa), the agent for the bacterial canker disease, were found in kiwifruit ³⁰. In another study in potato, 429 circRNAs were found to be differentially expressed in response to infection of Pectobacterium carotovorum subsp. brasiliense (Pcb), the most important causal agent of potato blackleg and soft rot globally. Upon Pcb challenge, some circRNAs were down-regulated in cultivar susceptible to the disease but up-regulated in cultivar resistant to the disease, or vice-versa, suggesting a potential role of these circRNAs in disease response ³⁵. Despite differential expression of certain circRNAs in responses to various biotic or abiotic stresses, no such plant circRNA has been experimentally demonstrated to be functional.