Some ecircRNAs are expressed stably and are rich in contents. They have miRNAs binding sites, which can compete with mRNA and thus regulate gene expression [1, 3]. The most typical example is ciRS-7 [17], which is derived from cerebellar degeneration related 1 (CDR1). antisense transcript and is mainly found in the brains of humans and mice [18]. It contains over 60 conserved miR-7 binding site [19]. ciRS-7 was highly expressed in the cytoplasm and can bind up to 20,000 miR-7 molecules per cell [19], which might affect the binding of miR-7 to its target mRNA. Reduced expression of ciRS-7 resulted in decreased expression of mRNAs which contain miR-7 binding sites, which confirmed that mRNAs compete with ciRS-7 for miR-7 binding sites [18, 19]. In addition, the function of ciRS-7 was conserved, and ciRS-7 expression in zebrafish affected the development of midbrain, whose effect was consistent with miR-7 knockdown [1, 19]. However, a recent article published in Science pointed out that after ciRS-7 knockout in mice, expression level of miR-7 in the cerebral cortex, hippocampus, cerebellum, and olfactory bulb significantly reduced, while the downstream target gene Fos was upregulated. Thus, it can be seen that the mechanism related to miRNA sponge function remains to be explored [20]. In addition, sex-determining region Y of testicular-specific circRNA (circSRY) contains 16 miR-138 binding sites in the mouse brain [18]. circZNF91, originating from zinc finger protein gene contains 24 miR-23 binding sites and 39 miR-296 binding sites [21]. Generally speaking, circRNA derived from genes containing repeat coding regions contains a series of conserved miRNA binding sites [22].

Different from ecircRNAs, ciRNAs and EIciRNAs mainly exist in the nucleus. However, its mechanism of transportation and retention in the nucleus is unclear. In human cells, ciRNAs localize near the host gene and combine with RNA Pol II during transcription, which increased the level of transcription through a mechanism of cis-regulation [23]. Studies have found that knock-down of ciRNAs would result in decreased expression of parental mRNA. Similarly, EIciRNAs can also interact with RNA Pol II. Knocking down EIciRNAs can also reduce the mRNA expression level of their parental gene. EIciRNAs associate with U1 small nuclear ribonucleoprotein (snRNP) through characteristic RNA-RNA interaction between EIciRNA and U1 small nuclear RNA (snRNA). Then the EIciRNA-U1 snRNP complexes may further communicate with the Pol II transcription complex at the promoters of parental genes, thereby increasing gene expression [12, 13]. circNOL10 was reported to promote transcription factor (TF) sex comb on midleg-like 1 (SCML1) expression by inhibiting ubiquitination [24].

circRNAs were originally considered as a special type of non-coding RNAs, but several researches have pointed out that circRNAs can be translated into proteins. In 1995, it was reported that eukaryotic ribosomes can start translation on circRNAs when the RNAs contain internal ribosome entry site (IRES) elements. Recently, a circRNA derived from the muscleblind locus was suggested to encode a protein in fly head when extracted by mass spectrometry [25]. CircZNF609 contained a 753-nt open reading frame (ORF), which was similar to linear transcripts, starting from AUG of the host gene and ending with a stop codon [26]. CircZNF609 interacted with polysomes, and then translated into a protein in a splicing-dependent and cap-independent manner [27]. Besides, circFBXW7 was highly expressed in the normal human brain. The spanning junction ORF in circ-FBXW7 driven by internal ribosome entry site encoded a novel protein, FBXW7-185aa, which is a functional protein. Increased expression of FBXW7-185aa in cancer cells was found to inhibit proliferation and cell cycle acceleration, while downregulation of FBXW7-185aa promoted malignant phenotypes both in vivo and in vitro. FBXW7185aa could also decrease the half-life of c-Myc by inhibiting USP28-induced c-Myc stabilization. Findings above suggested that circ-FBXW7 and FBXW7-185aa have potential prognostic significance in brain tumor [28]. Another novel tumor suppressor protein, SHPRH-146aa, consistent with the full-length protein and behaving as a protective molecule to decrease degradation, was derived from circ-SHPRH driven by IRES elements [29]. Besides, circ$\beta$-catenin could produce a novel 370-amino acids $\beta$-catenin isoform which shared the same start codon with the linear mRNA transcript but ended at a new stop codon when cyclizing. Silencing of circ$\beta$-catenin significantly suppressed malignant phenotypes both in vivo and in vitro, while knockdown reduced the protein expression level of $\beta$-catenin without influencing its mRNA level which further validates the possibility of circRNAs translation [30]. In addition, poly-adenosine or poly-thymidine in 3${}^{\prime }$ UTR could inhibit circRNAs translation [31] and m6A motif was found to initiate translation in endogenous circRNAs with translation potential in cellular responses to environmental stress [32]. The correlated mechanism is shown in Fig. 1.

Fig. 1.

Three possible mechanisms that circRNAs exert functions. (A). Function as miRNA sponges. circRNAs could interact with miRNAs and further relieve the suppression of target genes by miRNAs. (B). Gene transcription regulation. circRNAs have binding sites of RNA binding protein. They could interact with U1 snRNP, RNA Pol II or transcription factors and then regulate gene transcription. (C). Encode protein or peptide directly. circRNAs might transcript into peptides or proteins and proteins play a role directly.

Considering circRNAs are stable and expressions are cell-specific, selectively abundant and varying in different diseases, circRNAs are expected to have the potential to be a biomarker [33]. circRNAs were identified in several kinds of fluids, including exosomes [34], saliva [35], and blood [36]. Actually, circRNAs were suggested to behave as biomarkers in a variety of diseases, including cancer [34], systemic lupus erythematosus [37], rheumatoid arthritis [38], diabetes [39] and tuberculosis [40]. In addition, in the condition of nervous system diseases, such as Alzheimer’s disease (AD) [41], stroke [42] and amyotrophic lateral sclerosis (ALS) [43], blood-brain barriers are easy to be damaged, allowing free or exosome-encapsulated circRNAs [44] to release from the brain and be measured peripherally. Therefore, in neurological diseases, brain-specific circRNAs might be released into the circulatory system, so that the level of circRNAs in the blood can be used to monitor disease progression or achieve therapeutic purposes.

Many studies have reported circRNAs microarrays or sequencing results which are related with IS. After validation by qRT-PCR, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses were performed. Several researches were conducted in patients and healthy controls while the others used rodent models or cells. In human sample, the ratio of serum hsa-circRNA-284 and miR-221 was found to notably increase in the patients of IS within 5 days (P = 0.0002), which had a biomarker potential of carotid plaque rupture and stroke [49]. Ostolaza et al. found that a set of 60 circRNAs were increased in atherotrombotic compared with cardioembolic strokes. Differential expression of hsa-circRNA-102488 which is originated from UBA52 gene, was further validated. This suggests that circRNA might be a new source of biomarker for IS etiology [50]. Moreover, circFUNDC1, circPDS5B and circCDC14A were increased in IS patients versus healthy controls and were reported to serve as diagnostic and prognostic biomarkers [51]. In addition, 373 circRNAs were increased, while 148 were decreased when compared the peripheral blood mononuclear cells (PBMCs) from IS patients versus healthy controls. Thirteen candidate circRNAs were verified by qRT-PCR. Bioinformatics analysis showed that aberrantly expressed circRNAs were associated with inflammation and immunity which were involved in IS pathogenesis and had the potential to be diagnostic biomarker [52]. While using rodent models, 128, 198, and 789 circRNAs were found to be significantly changed at 5 min, 3 h, and 24 h after IS in mice blood sample. After validation by IS patient blood sample, hsa-circ-0090002 and hsa-circ-0039457 were reported to be differentially downregulated [53]. Besides, after 45-minute-transient middle cerebral artery occlusion (MCAO), circRNA microarrays revealed that 1027 circRNAs were significantly changed in 48 hours after reperfusion in the ischemic brain versus the sham group in which 914 circRNAs were notably increased, while the other 113 were notably decreased. The expression level of mmu-circRNA-40001, mmu-circRNA-013120, and mmu-circRNA-40806 was further validated by qRT-PCR. After GO and KEGG analyses, the Rap1 and Hippo signaling pathway which are involved in cell survival and death regulation was found to be the most enriched [54]. In another microarray analysis, 283 circRNAs were changed ($>$2-fold) at least at one of the reperfusion time points, whereas 16 were changed at all 3-time points (6 h, 12 h and 24 h) of reperfusion after transient MCAO studied versus sham. Then bioinformatics analysis showed that binding to proteins, ions and nucleic acids, cell communication, metabolic process, and biological regulation are leading biological and molecular functions regulated by circRNAs change after transient MCAO [55]. Duan et al. suggested that totally 14,694 circRNAs from 6 brain tissues were detected differentially expressed by HTS of which 87 showed a significant fold change $>$2 (P 0.05). rno-circRNA-17737, rno-circRNA-8828 and rno-circRNA-14479 were significantly upregulated, while rno-circRNA-1059, rno-circRNA-9967 and rno-circRNA-6952 were significant downregulated when validated [56]. Zhang et al. reported that 16 and 28 circRNAs showed significant increase and decrease respectively in rat ischemia-reperfusion model by HTS. In in vitro experiments, levels of the hsa-circ-camk4 were elevated in SH-SY5Y cells exposed to oxygen-glucose deprivation/reoxygenation (OGD/R) treatment. Overexpression of hsa-circcamk4 in SH-SY5Y cells significantly increased cell death rate after OGD/R. Bioinformatics analysis showed that circcamk4 was involved in the apoptosis signaling pathways, glutamatergic synapse pathway and MAPK signaling pathway, all of which are known to be included in brain injury after ischemia-reperfusion (I/R). This suggested that hsa-circ-camk4 may have been involved in in progression of cerebral I/R injury [57]. In parallel, 3 circRNAs were significantly increased, and 12 were decreased in HT22 cells with OGD/R compared with normal controls. Furthermore, upregulation of mmu-circRNA-015947 was validated by qRT-PCR. Bioinformatics analysis revealed that mmu-circRNA-015947 could interact with mmu-miR-188-3p, mmu-miR-329-5p, mmu-miR-3057-3p, mmu-miR-5098 and mmu-miR-683 and thus increase target gene expression. KEGG pathway analysis predicted that mmu-circRNA-015947 might be involved in apoptosis, metabolism and immune-related pathways, which are known to play a role in cerebral I/R injury [58]. The detailed information summary is shown in Table 1[49, 50, 51, 52, 53, 54, 55, 56, 57, 58]. Besides, circHECTD1 expression was upregulated in IS recurrence patients versus non-recurrence patients (160 : 160). Further, ROC curve analysis indicated that circHECTD1 expression predicted higher risk of IS recurrence. CircHECTD1 also correlated with higher disease risk, disease severity and inflammation [59]. Similar results were found in peripheral blood mononuclear cell circDLGAP4. CircDLGAP4 is decreased and negatively related with severity, inflammatory cytokine expression such as TNF-$\alpha$, IL-6, IL-8 as well as IL-22 and miR-143 expression in IS patients [60]. Moreover, in vitro, silencing of circHIPK2 stimulated neural stem cells (NSCs) specifically differentiated into neurons but made no difference to the differentiation to astrocytes. In vivo, microinjection of si-circHIPK2-NSCs could migrate to the ischemic area after stroke induction. Si-circHIPK2-NSCs increased neuronal plasticity in the ischemic brain, led to long-lasting neuroprotection, and significantly reduced functional dysfunction which exhibits a promising therapeutic strategy to neuroprotection [61]. The findings suggest that circRNAs might be a biomarker for IS diagnosis and involved in IS pathogenesis but the concise mechanism needs further investigation.

When it comes to the mechanism research, some studies indicate that circRNAs are involved in the process of IS by interacting with miRNAs and subsequent downstream target genes. CircDLGAP4 was found to be downregulated in IS patients and tMCAO mice and functioned as miR-143 sponge to repress miR-143 activity, which inhibited the expression of homologous to the E6-AP C-terminal (HECT) domain E3 ubiquitin protein ligase 1 (HECTD1). Overexpression of circDLGAP4 enhanced tight junction protein expression and decreased mesenchymal cell marker expression which inhibited endothelial-mesenchymal transition (EMT). and would lead to blood brain barrier (BBB). protection [62]. In parallel, circHECTD1 was upregulated in tMCAO mice and functioned as an endogenous miR-142 sponge to repress miR-142 activity, resulting in the inhibition of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) inducible poly (ADP-ribose) polymerase (TIPARP) expression. The knockdown of circHECTD1 reduced infarct volumes and attenuated neurological deficits in transient MCAO mice through regulation of astrocyte activation via inhibition of autophagy [63]. Another associated study illustrated that circTLK1 functioned as an endogenous miR-335-3p sponge to repress miR-335-3p activity, which resulted in the increasing expression of TIPARP and a subsequent exacerbation of neuronal injury. The knockdown of circTLK1 significantly decreased infarction area, ameliorated neuronal injury, and improved neurological dysfunction [64]. Results are consistent with and complement each other, which provide new potential therapeutic targets for IS patients.

In ICH model, SD rats were injected with autologous artery blood. 111,1145,1751 circRNAs were upregulated while 47,732,1329 were downregulated in the cerebral cortex of ICH rats at 6 h, 12 h and 24 h after ICH induction compared with sham group (fold change $\ge$ 1.5, P-value $\le$ 0.05). 93 were up-regulated and 20 were down-regulated at all three time points. The expression level of 6 circRNAs were further validated using qRT-PCR while 2 up-regulated circRNAs (rno_circRNA_011054 and rno_circRNA_005098) and 1 down-regulated circRNA (rno_circRNA_012556) were statistically altered. After GO and KEGG analyses, parent genes showed transition from protein complex assembly, cell-cell adhesion and cyclic adenosine monophosphate (cAMP) signaling pathway at 6 h to intracellular signal transduction, protein phosphorylation and glutamatergic synapse at 12 h and 24 h. Enrichment analyses of targeted mRNAs indicated transcriptional regulations and pathways including Ras-associated protein 1 (Rap1), mitogen-activated protein kinase (MAPK), phosphatidylinositol 3 kinase/protein kinase B (PI3K-Akt), tumor necrosis factor (TNF) and Wingless/Integrated (Wnt) signaling pathways in cancer [65]. Findings above suggest an important role of circRNAs in ICH and provide potential mechanism direction for further basic research and a promising target for ICH treatment.

circRNAs also participated in the pathogenesis of other neurological diseases, such as AD, Parkinson’s disease (PD), motor neuron disease (MND) and demyelination diseases. In AD, many researches also focused on expression profiles [66, 67, 68, 69, 70, 71, 72]. Consistent with result above, majority of AD-associated circRNAs are independent from changes in cognate linear mRNA expression [66]. circRNAs are found to play roles in AD [20, 73, 74, 75, 76, 77], either by interacting with proteins [73] or miRNAs [74, 75]. In PD, circDLGAP4 exerted neuroprotective effects via miR-134-5p/CREB pathway [78] while circzip-2 possible sponged miR-60-3p [79]. Similar to other diseases, expression profile was detected by RNA-sequencing in different brain regions of PD mouse model, not comparison between case group and control group [80]. circRNAs are also suggested to behave as biomarkers in ALS [81]. Moreover, in demyelinating diseases, circRNAs either in exosomes [82] or in peripheral blood leucocytes [83] could be used as biomarkers. The expression level of ecircRNAs were significantly decreased in peripheral blood mononuclear cells of relapsing-remitting multiple sclerosis patients compared to controls [84].