Research shows that direct back-splicing is the main mechanism of circRNA formation (Pan et al., 2020). During this process, a circular structure is formed by looping the intron sequences flanking the downstream splice-donor site with the upstream splice-acceptor site. Base pairing of two introns via short repeat sequences is one important pattern to bring these sites into close proximity (Kristensen et al., 2019). One report by Liang and Wilusz (2014) found that the Alu family played an important role in the circularization of some exons, and only a small portion of them was sufficient to form circRNAs. Interestingly, they found that longer repeat parts were not necessarily better for circularization; in contrast, some may even play an inhibitory role (Liang and Wilusz, 2014). RNA-binding proteins (RBPs) that bind to specific motifs in the flanking introns can also facilitate the formation of circRNAs during direct back-splicing. For example, Quaking changed during the epithelial–mesenchymal transition could specifically bind with the motif in introns and positively regulate the abundance of circRNAs (Conn et al., 2015). Errichelli et al. (2017) reported that another RBP, FUS, a key protein linked to the pathogenesis of amyotrophic lateral sclerosis and frontotemporal dementia, regulated circRNA biogenesis by binding the introns flanking the back-splicing junctions. After forming the circular structure, some introns are then removed, and exonic circRNAs (ecircRNAs) or exon-intron circRNAs (EIciRNAs) are generated (Figure 1A).

During this process, an exon “skips” from the final mRNA product, and sequences between its upper and downstream exons can form an exon-intron lariat structure. Then, the lariat undergoes internal splicing and finally generates a circRNA (Petkovic and Muller, 2015) (Figure 1B). This phenomenon is common in lower eukaryotes. For example, in a type of yeast named Schizosaccharomyces pombe, mrps16 gene lacks any base pairing between the regions flanking the circularized exon, which means that the method to form mrps16 circRNA is quite likely to be different from direct back-splicing. Through the quantitative analysis of isoforms, it has been demonstrated that exon skipping is the main mechanism of circRNA biogenesis in mrps16 (Barrett et al., 2015). Kelly et al. (2015) also verified exon skipping in primary human endothelial cells. They used a custom pipeline to characterize circRNAs derived from coding exons and found that circularization of exons is widespread and correlates with exon skipping (Kelly et al., 2015).

CiRNA can be derived from an intron lariat, which is quite different from back-splicing and exon skipping. An intron lariat is produced due to the failure of debranching, and the formation of a ciRNA is related to specific consensus motifs, which contain key RNA elements near the 5’ splice site and the branchpoint (Figure 1C). These critical elements can produce a marked effect through the promotion of inefficient debranching, making it possible to generate a ciRNA (Zhang et al., 2013).

Previous GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses of circRNAs in the brain revealed that many of them were closely related to the regulation of synaptic plasticity (Bu et al., 2019; Mahmoudi et al., 2019; Filippenkov Ivan et al., 2021), which was supported by some in vitro and in vivo experiments. For example, a circRNA called circHomer 1 derived from the precursor RNA of Homer scaffold protein 1 is thought to modulate the expression of many genes involved in synaptic plasticity and psychiatric disease (Zimmerman et al., 2020). circGRIA1, a conserved circRNA isoform derived from the AMPA receptor subunit Gria1, shows an age-related and male-specific increase in the prefrontal cortex and hippocampus of rhesus macaques. Knockdown of circGRIA1 in hippocampal neurons promotes synaptogenesis and attenuates age-induced decreases in spontaneous mEPSCs both in amplitude and frequency (Xu et al., 2020). Wang et al. (2020a) showed that circHIPK2 silencing enhanced NSC-differentiated neurons and promoted neuronal plasticity in vitro by changing the formation of synapses, thereby helping ischemic brain tissue recover more quickly after stroke. Additionally, Nrf2, a neuroprotective agent, can positively regulate circ-Vps41 expression to increase the number of synaptic vesicles and dendritic spines in the hippocampus and improve impaired memory (Zhang et al., 2022). All of these studies suggest that circRNAs are involved in neuronal plasticity regulation. Our lab, using a fear conditioning model, first screened for the circRNAs that play a role in memory formation, extinction and erasure through the RNA-seq method. We also constructed circRNA-miRNA-mRNA networks in different memory stages according to the competing endogenous RNAs (ceRNA) theory. Through these analyses, we further proved that circRNAs play a role in cognition under normal physiological conditions, and the circRNAs involved in different memory processes vary greatly (Wang, 2021). In the following, more works need to be done to clarify the key circRNAs involved in the specific processes of memory, including fear memory and spatial memory, and their potential mechanisms.

The fact that circRNAs are the most resistant of the RNA species to degradation, they are ideal for postmortem studies. The expression profiles of most circRNAs in autism have been elucidated. Chen et al. (2020) measured circRNA expression levels in the postmortem cerebral cortex of ASD and non-ASD controls and identified 60 circRNAs (22 upregulated and 38 downregulated) involved in ASD. Putative targets of circRNA-microRNA-mRNA axes were enriched in ASD risk genes and genes encoding inhibitory postsynaptic density proteins. By comparing the expression of circRNA in the brains of autistic patients and mice, it was found that the expression of circRNA is highly conserved between humans and mice. In addition, Gasparini et al. (2020) studied the expression profile of circRNA in the hippocampus of BTBR mice, a fully characterized ASD model, and age-matched C57BL/6J mice. In this study, 29 circRNAs were significantly downregulated and 12 were upregulated in BTBR mice. Using another ASD mouse model induced by valproate acid, 1,059 differentially expressed circRNAs (477 upregulated and 582 downregulated) were screened by RNA sequencing. Module analysis further showed that these dysregulated circRNAs correlated with autism-related pathways, including the TGF-beta signaling pathway, Notch signaling pathway, and long-term depression (Wang et al., 2020b).

Regarding the underlying mechanism of these candidate circRNAs in ASD, the ceRNA pattern was proposed in these studies. In vitro experiments confirmed that upregulation of circARID1A could increase some ASD risk genes, such as SFARI or AutismKB, by directly sponging miR-204-3p (Chen et al., 2020).

At present, there are relatively few experimental studies on the effect of circRNA on ASD, and most of them are descriptive studies. More experimental and clinical correlation studies need to be done in the future. Therefore, there is still a long way to go for ASD diagnosis and treatment using circRNA.

Given the multiple roles of circRNAs in the regulation of neurotransmitters and synaptic plasticity, the possible involvement of circRNAs in depression has received more attention. Studies have shown that compared with normal subjects, there are changes in the expression of circRNAs in patients with depression, which may be related to the formation mechanism of depression.

Cui et al. (2016) analyzed the expression profiles of circRNAs in peripheral blood mononuclear cells (PBMCs) of patients with depression and healthy controls and identified four differentially expressed circRNAs: hsa_circRNA_002143, hsa_circRNA_103636, hsa_circRNA_100679, and hsa_circRNA_104953. After antidepressant treatment, it was found that only the downregulated hsa_circRNA_103636 had obvious changes, suggesting that hsa_circRNA_103636 may have potential diagnostic value for depression (Cui et al., 2016). Jiang et al. (2017) analyzed by microarray the plasma samples of 7 patients with type 2 diabetes mellitus (T2DM) and 7 patients with T2DM and depression and identified 183 circRNAs that were upregulated and 64 circRNAs that were downregulated. Zhang et al. (2018) comprehensively analyzed the circRNA spectrum of chronic unpredictable mild stress (CUMS) mice with typical depressive symptoms and found that the expression levels of mmu_circ_0001355 and 0001788 were slightly lowered, while the expression of mmu_circ_0001223 was greatly inhibited in the ventral medial prefrontal cortex (vmPFC) of CUMS mice. In addition, they also confirmed that total saponins from the leaves of Panax notoginseng saponins (PNS) can inhibit the development of depression, and deep sequencing shows that a large number of circRNAs are differentially expressed in the vmPFC and hippocampus of CUMS mice treated with PNS (Zhang et al., 2018). Ketamine has an obvious antidepressant effect (Berman et al., 2000). Mao et al. (2020) found five differentially expressed circRNAs in the rat hippocampus after ketamine injection and verified the upregulated rno_circRNA_014900 and downregulated rno_circRNA_005442 by qRT-PCR. Maternal separation (MS) is a stable and reliable method to simulate depression-like behavior. Zheng et al. (2019) examined the whole genome transcriptome of the prefrontal cortex in male rats with MS and found that 20 circRNAs were significantly overexpressed and 17 circRNAs were significantly underexpressed.

The effects of circRNAs on depression by miRNA sponging are the most studied. As an important posttranscriptional regulator, miRNAs decrease the expression of target genes by binding to their 3′-UTR, which could indirectly be regulated by circRNAs. By using the bioinformatics program RNA hybrid, Huang et al. (2017) found that circHIPK2 contained a target site of miRNA-124. Double ISH showed that circHIPK2 and miRNA-124 colocalize in primary mouse astrocytes, and the expression of miRNA-124 and its downstream target gene sigma non-opioid intracellular receptor 1 are both regulated by circHIPK2 in a ceRNA-dependent manner, thus subsequently affecting the activation of astrocytes (Huang et al., 2017). Due to the important roles of activated astrocytes in depression, circHIPK2 is believed to participate in the formation and development of depression. Zhang et al. (2019) found that the expression level of circHIPK2 in chronic unpredictable stress (CUS) mice was significantly higher than that in the control group. Fecal microbiota transplantation from NLRP3 KO mice significantly alleviated astrocyte dysfunction in CUS-treated recipient mice via inhibition of circHIPK2 expression (Zhang et al., 2019). In addition to discovering the role of circRNAs in astrocytes, studies confirmed another circRNA, circDYM, which is involved in MDD pathogenesis and treatment by sponging miR-9 and regulating the activation of microglia (Zhang Y. et al., 2020). Moreover, circDYM delivered by extracellular vesicles (RVG-circDYM-EVs) significantly increased the expression of circDYM in the brain, inhibited microglial activation, and thus alleviated CUS-induced depression-like behavior (Yu et al., 2022). RVG-circDYM-Evs can easily overcome the obstacles of the blood–brain barrier but also enhance its integrity and inhibit the infiltration of peripheral immune cells (including lymphocytes and myeloid cells) into the brain of CUS mice. This indicates that extracellular vesicle-mediated circDYM delivery is expected to be applied in clinical depression treatment. In addition, Fan et al. (2022) found that the expression of circANKS1B in the dentate gyrus area of CUS rats decreased significantly. This change could decrease the sponging of miR-146a-5p, a microRNA derived from microglial cells, and its downstream target gene KLF4, thus participating in the occurrence of depressive-like behaviors (Fan et al., 2022). These discoveries provide novel insights regarding the specific contribution of circRNAs in the pathogenesis and treatment of depression.

Increasing evidence suggests that the RNA demethylase ALKBH5 is related to MDD (Du et al., 2015). Huang et al. (2020) found that the expression of circSTAG1 in the hippocampus of CUS mice decreased, and the expression of circSTAG1 in the plasma and whole blood of patients with depression decreased significantly. Furthermore, it was found that circSTAG1 had a stronger affinity for ALKBH5 than STAG1 mRNA, especially in astrocytes. When the expression of circSTAG1 was downregulated, ALKBH5 could be released, which reduced the methylation of FAAH and then enhanced the stability of FAAH mRNA, resulting in the dysfunction of astrocytes and the occurrence of depressive-like behaviors (Huang et al., 2020).

Emerging studies have shown that circRNAs may be involved in the onset and progression of depression, so they have become a promising new field in the diagnosis of depression. Reports have shown that hsa_circRNA_103636 (Cui et al., 2016), hsa_circ_0126218 (Bu et al., 2021), circFKBP8 (Shi et al., 2021), circMBNL1 (Shi et al., 2021), and circDYM (Song et al., 2020) might be promising MDD biomarkers, which may further improve the early detection, effective diagnosis, and convenient monitoring of MDD. However, at present, the diagnostic accuracy and specificity of these circRNAs still need to be verified by multicenter and large samples. Similarly, both in vivo and in vitro experiments indicated that some circRNAs, including circHIPK2 (Zhang et al., 2019), circDYM (Zhang Y. et al., 2020; Yu et al., 2022), circSTAG1 (Huang et al., 2020), and mmu_circ_0001223 (Zhang et al., 2018) might be drug targets for depression treatment, although there are still few effective anti-depression drugs targeting these targets. Recently, the transportation of exogenous circRNAs or their interfering RNAs into the brain by extracellular vesicles has achieved the same effects as local brain region microinjection on cognitive-related diseases (Yu et al., 2022). Since it is a much easier way to change the expression of circRNAs in the brain, it is expected to be used in clinical practice in the future. However, before this clinical application, more research is needed, especially regarding the safety, efficacy, and cost of circRNA delivery systems.

Aging is a degenerative change of the body due to the increase in age. In the nervous system, it is mainly reflected in the structural degeneration and the decline of physiological functions, such as cognitive impairment and reduced stress response ability, which lead to the occurrence of age-related diseases (Cai et al., 2019). Studies have shown that circRNAs seem to accumulate specifically in the brains of various aging organisms, including Drosophila, mice, and humans (Knupp and Miura, 2018; Xu et al., 2021) and are involved in cell aging and survival, thus regulating aging-related disorders (Cai et al., 2019). However, the mechanisms accounting for the increased levels of circRNAs during aging are still unclear. A special ring structure resistant to degradation by exoribonucleases, the postmitotic nature of neurons, global changes in splicing patterns, and all possible circRNA turnover regulators during aging in the brain are mostly involved in the increased levels of circRNAs (Jeck et al., 2013; Deng et al., 2014; Enuka et al., 2016; Errichelli et al., 2017; Knupp and Miura, 2018).

The high stability and specificity of circRNA expression make circRNA a potential biomarker of human disease and one of the methods for disease diagnosis and prognosis. In particular, some circRNAs accumulated rapidly during brain aging compared with their corresponding mRNAs, suggesting that circRNAs may be sensitive biomarkers for aging. For example, in the analysis of human peripheral blood, the levels of hsa_circ_1305 and hsa_circ_722 in the elderly are higher than those in the young (Dluzen et al., 2018). CircDEF6, circFOXO3, circEP300, and circFNDC3B are confirmed to be related to the human aging phenotype (Haque et al., 2020). These screened circRNAs are helpful for determining the physical condition and initiating early prevention of some aging-related diseases, such as Alzheimer’s disease and Parkinson’s disease. Due to length limitations, this review will mainly focus on the relationship between circRNAs and AD.

AD is a neurodegenerative disease with insidious onset and slow progression and is also the leading cause of cognitive dysfunction in the aging population. To date, the neuropathological features of AD are thought to involve excessive amyloid β (Aβ) aggravation, hyperphosphorylation of Tau-induced neurofibrillary tangles (NFTs), and neuronal damage (Jeong, 2017; Ma et al., 2017; Reiss et al., 2018; Naseri et al., 2019). Excessive Aβ causes neuronal inflammation, oxidative stress, and Ca²⁺ overload, which may lead to neuronal damage and subsequent cognitive impairment (Katsumoto et al., 2018; Tiwari et al., 2019). It is helpful for the diagnosis and treatment of AD to understand how related molecules participate in these processes. Due to the rapid development of high-throughput RNA-seq technologies, the dynamic expression patterns and functions of circRNAs in AD have been characterized recently in a large number of studies (Dube et al., 2019; Cervera-Carles et al., 2020; Cochran et al., 2021).

circRNAs easily accumulate in the aging brain. Of the numerous changes in circRNAs, which ones are only related to aging and which ones are related to the development of AD needs to be clarified. Therefore, microarray or RNA-sequencing data from different brain regions of AD and age-matched healthy persons, including the frontotemporal lobar, entorhinal cortex, parietal cortex, posterior cingulate, hippocampus, and parahippocampal gyrus, were fully analyzed (Sekar et al., 2018; Dube et al., 2019; Cervera-Carles et al., 2020; Lo et al., 2020; Cochran et al., 2021; Urdanoz-Casado et al., 2021) (Table 1). Many AD-related circRNAs have been confirmed, and most of them are enriched in neurotrophic signaling, Aβ clearance, neuroinflammation, and oxidative stress. In addition to the brain parenchyma described above, abnormal changes in circRNAs in the cerebrospinal fluid and peripheral blood of AD patients have also been observed (Li et al., 2020a,b; Liu et al., 2020) (Table 1). Although it remains unknown whether the occurrence of AD is caused by the changed circRNAs in the brain, they may serve as biomarkers for the prediction or diagnosis of AD. Additionally, differential expression profiles of circRNAs in some AD cellular or animal models, including Amyloid precursor protein (APP)/PS1, Tg2576, 5xFAD, SAMP8, or Aβ administration, were constructed (Zhang et al., 2017, 2021; Wang et al., 2018; Lee et al., 2019; Ma et al., 2019; Nam et al., 2019) (Table 1). However, few dysregulated circRNAs have been found in common in both AD human and animal models according to these studies.

To date, the pathogenesis of AD is still unclear. The cascade of pathological reactions caused by amyloid aggregation is still the focus of AD research. Recently, the effects of circRNAs on AD pathology have mainly been explored in terms of Aβ production and clearance, neuroinflammation, and oxidative stress.

Excessive deposition of Aβ in the cerebral cortex accelerates the development of AD (Alcendor, 2020). Aβ protein comes from the cleavage of APP by β-secretase and γ-secretase, the clearance of which is mainly dependent on the function of autophagy in neural cells (Baranello et al., 2015; Chen et al., 2017; Xin et al., 2018). GO enrichment analysis of some RNA sequencing data showed that many dysregulated circRNAs in AD were involved in the regulation of Aβ production and clearance. CiRS-7, a circRNA highly expressed in the human brain, has been proven to be downregulated in the brains of AD patients (Lu and Xu, 2016). Using HEK293T and SH-SY5Y cells, Shi et al. (2017) found that dysregulated ciRS-7 promoted the degradation of APP and beta-site amyloid precursor protein cleaving enzyme 1(BACE1) via the proteasome and lysosome in an NF-κB-dependent manner, which finally reduced the generation of Aβ. CircCwc27 could also be directly bound with purine-rich element-binding protein A, which suppressed the expression of a cluster of AD genes (Song et al., 2022). Interestingly, while studying the mechanism of Aβ production, a circRNA that can encode a polypeptide was found. This circRNA could be efficiently translated into a novel Aβ-containing Aβ175 polypeptide, which will be processed into Aβ (Mo et al., 2020). These results indicated an alternative pathway of Aβ production induced by a circRNA. In terms of Aβ clearance, the role of ciRS-7 was also referred to. It showed that ciRS-7 could sponge miR-7 and affect the expression of its target ubiquitin protein ligase UBE2A, a protein essential in the clearance of amyloid peptides in AD (Zhao et al., 2016). Moreover, circRNAs can change autophagy ability and indirectly regulate Aβ clearance and reduce the phosphorylation of Tau (Inoue et al., 2012; Lonskaya et al., 2014; Diling et al., 2019). For example, circRNA NF1-419 enhanced autophagy by binding dynamin-1 and adaptor protein 2 B1 in AD-like mice and reduced the expression of AD marker proteins, including Tau, p-Tau, Aβ, and APOE (Diling et al., 2019).

The loss of nerve cells is one of the typical features of AD, which is mainly caused by neuroinflammation and oxidative stress. Several studies have confirmed some roles of circRNAs in these processes. circ_0002945 was overexpressed in AD patient serum, Aβ-stimulated SK-N-SH cells and human primary neurons. Inhibition of circ_0002945 attenuated Aβ-induced cell apoptosis and endoplasmic reticulum stress through a novel circ_0002945/miR-431-5p/TNFAIP1 ceRNA network (Li G. et al., 2022). Similarly, using Aβ-treated SK-N-SH and SK-SY5Y cell lines, a study suggested that circRNA AXL increased neuronal injury and inflammation by targeting microRNA-328-mediated BACE1 in AD (Li Y. et al., 2022). The presence of Aβ and NFTs promotes microglial cell activation, leading to a persistent and robust release of inflammatory cytokines and chemokines and finally the death of surrounding neurons (Agrawal and Jha, 2020; Perea et al., 2020; Xu Y. et al., 2022). In a microglial cell line, BV2 cells, knockdown of circ_0005835 downregulated neuroinflammation by sponging miR-576-3p and changed the process of AD (Xu X. et al., 2022). In terms of oxidative stress, circLPAR1 and circ_0001588 have been proven to regulate oxidative stress in rat models and in vitro models of AD by sponging miRNAs (Zhu et al., 2020; Wu et al., 2021). In addition to the above indirect effects of circRNAs on neuronal morphology, circRNAs can directly regulate neuronal survival or neurite growth and branching in AD. CircTulp4 predominantly localizes in the nucleus, positively regulates hippocampal neurite growth and branching by changing the transcription of its parental gene, Tulp4, and might participate in the development of AD (Ma et al., 2021). It has been found that circ_0000950 decreased the expression of miR-103 and increased the expression of Ptgs 2 in AD patients, as well as promoted neuronal apoptosis, inhibited the growth of neuronal processes, and enhanced the level of inflammatory cytokines (Yang et al., 2019). The regulatory roles of circRNAs in neuroinflammation, oxidative stress, and neurite growth imply that circRNAs may function as important players in the pathogenesis of AD. Certainly, additional studies are still required to test the hypotheses in humans.

In view of the changes and biological roles of circRNAs in AD, as well as its high stability, it is expected to be used in clinical diagnosis and as a therapeutic target of drugs. Some studies have made useful attempts in this regard. Hsa_circ_0003391 was significantly downregulated in the peripheral blood of AD patients. Receiver operating characteristic curve analysis showed that hsa_circ_0003391 in peripheral blood or circ-AXL and circ-GPHN in cerebrospinal fluid had a relatively high area under the curve (AUC) value, suggesting a promising biomarker for AD diagnosis (Liu et al., 2020). Moreover, circPSEN1 was reported to be able to discriminate autosomal dominant AD from sporadic AD and controls with an AUC above 0.70. These results showed that circRNAs expression levels in body fluids are significantly correlated with both the occurrence and clinical severity of AD. Although there has been some progress in diagnosis, the evidence is still weak and needs to be further supported by large, multicenter studies (Chen et al., 2022). In addition, circRNAs, combined with other biomarkers and imaging tools to improve the diagnostic power, should also be considered.

Recently, direct intracerebral injection or in vitro cellular intervention and certain circRNAs have been confirmed to improve AD symptoms. In addition, the roles of circRNAs in AD drug treatment have also been reported, although disease-associated proteins are currently the main targets for AD treatment. Breviscapine, combined with bone marrow mesenchymal stem cell treatment, can reduce Aβ deposition and promote the degradation of APP and BAEC1 by regulating the circRNA ciRS-7 (Sun et al., 2023). PNS significantly inhibited the pathological progression of AD by upregulating five circRNAs or downregulating two circRNAs in AD mouse models (Huang et al., 2018b). Similarly, in human neuronal cells, berberine attenuated Aβ42-induced neuronal damage by regulating the circHDAC9/miR-142-5p axis (Zhang N. et al., 2020). Of course, the evidence from in vitro and animal experiments is still inadequate to prove the clinical efficacy of circRNAs as therapeutic drugs. In the future, more work should be done in AD patients, and targeting circRNAs may provide new ideas for the struggling development of AD drugs.