Circular RNAs in non-alcoholic fatty liver disease: Functions and clinical significance

ABSTRACT

Nonalcoholic fatty liver disease (NAFLD), which affects approximately 25% of the global population, is an urgent health issue leading to various metabolic comorbidities. Circular RNAs (circRNAs), covalently closed RNA molecules, are characterized by ubiquity, diversity, stability, and conservatism. Indeed, they participate in various biological processes via distinct mechanisms that could modify the natural history of NAFLD. In this review, we briefly introduce the biogenesis, characteristics, and biological functions of circRNAs. Furthermore, we summarize circRNAs expression profiles in NAFLD by intersecting seven sequencing data sets and describe the cellular roles of circRNAs and their potential advantages as biomarkers of NAFLD. In addition, we emphatically discuss the exosomal non-coding RNA sorting mechanisms and possible functions in recipient cells. Finally, we extensively discuss the potential application of targeting disease-related circRNAs and competing endogenous RNA networks through gain-of-function and loss-of-function approaches in targeted therapy of NAFLD.

KEYWORDS:

• Nonalcoholic fatty liver disease
• circular RNAs
• function
• sorting mechanism
• exosomal circRNA

Introduction

Non-alcoholic fatty liver disease (NAFLD) is currently the leading cause of chronic liver diseases globally, affecting 25% of the global population and approximately 70–80% of obese and diabetic individuals [1]. Most NAFLD individuals are mostly asymptomatic during its early stages, such as hepatic steatosis, often described as a ‘benign condition’, and do not usually progress to more severe disease. However, its active phenotype, namely non-alcoholic steatohepatitis (NASH), may potentially progress into advanced liver diseases, such as cirrhosis or even hepatocellular carcinoma (HCC) [1–3]. NAFLD is associated with metabolic syndrome, which leads to obesity, insulin resistance (IR), and hyperlipidaemia. Complications and comorbidities linked to NAFLD severely strain healthcare and economic systems [4,5]. The pathogenesis of NAFLD remains controversial, with the traditional ‘two-hit’ hypothesis replaced by the ‘multiple parallel hits’ hypothesis [6,7]. Although the mechanisms underlying NAFLD have not yet been fully elucidated, studies are underway to identify potential biomarkers that enable early diagnosis of those at high risk for NAFLD and help develop suitable therapeutic strategies for individuals with advanced NASH.

Although our understanding of NAFLD/NASH has progressed substantially over the past two decades, some dissatisfaction regarding terminology, such as the use of ‘non-alcoholic’, which overemphasizes ‘alcohol’ and underemphasizes metabolic risk factors, remains. To remedy this issue, a name change from NAFLD to metabolic-associated fatty liver disease (MAFLD) has been proposed to describe liver diseases associated with known metabolic dysfunction [8]. Criteria proposed for the diagnosis of MAFLD are based on histological (biopsy), imaging, or blood biomarker-related evidence of fat accumulation in the liver (hepatic steatosis), accompanied by one of the following three features: (i) overweight/obesity; (ii) type 2 diabetes mellitus (T2DM); or (iii) normal weight/lean with symptoms of metabolic dysregulation [9]. However, since all published studies are based on ‘NAFLD’ terminology, the present study will continue to utilize this term.

Non-coding RNAs (ncRNAs), such as microRNAs (miRNAs), circular RNAs (circRNA), and long non-coding RNAs (lncRNAs), are found in almost all tissues and biofluids of different species affected by multiple diseases demonstrating their biological importance [10,11]. Recent studies have stressed that ‘non-coding’ does not necessarily indicate ‘non-essential’, implying that ncRNAs are essential components of complex regulatory mechanisms in multiple diseases. Some studies have suggested that ncRNAs, which are involved in the initiation, progression, and remission of hepatic steatosis, may be utilized as non-invasive biomarkers for early risk assessment and clinic intervention in NAFLD [12]. CircRNAs, a novel class of endogenous ncRNAs, are characterized by covalently closed single-stranded loop structures without a 5′ cap or a 3′ poly (A) tail [13]. The unique structure of circRNAs imparts a longer half-life than linear RNAs [14]. Moreover, type- and tissue-specific expression patterns allow them to be considered potential candidates for diagnostic biomarkers and therapeutic targets [15]. A growing number of circRNAs are reportedly involved in multiple biological processes governing diseases, such as cancers, cardiovascular diseases, neurological disorders, and autoimmune diseases [13,16–19]. However, literature on the regulatory effects and clinical application of circRNAs in NAFLD remains scant.

To demonstrate the regulatory roles played by circRNAs in NAFLD, we summarized the currently available literature on circRNAs and NAFLD. Firstly, we briefly introduce the biogenesis, characteristics, and biological function of circRNAs. Secondly, we summarize the expression profiles of circRNAs in NAFLD using all published data via a Venn map. Next, we highlight the emerging role of circRNAs in the pathogenesis of NAFLD based on potential pathogenic factors, including IR, inflammation, oxidative stress, etc. Finally, we highlighted the sorting mechanism of exosome-derived ncRNAs and their functions in recipient cells and circRNA-based targeting approaches. These findings indicate that, in light of their stable characteristics, ubiquitous presence, and vigorous functioning, circRNAs and exosomal circRNAs may be exploited as reliable circulating biomarkers that enable non-invasive diagnosis of different stages of NAFLD, and also as promising therapeutic targets in NAFLD and NASH.

Biogenesis, characteristics, and biological functions of circRNAs

CircRNAs, first discovered in viroids by Sanger et al. in 1976 [20], were previously considered accidental by-products with low abundance and little functional potential, or ‘splicing noise’ generated by aberrant splicing events [21]. With the advent of high-throughput RNA-sequencing and circRNA-specific bioinformatic algorithms, a large number of circRNAs have been successfully identified in various cell lines and species and thereby considered crucial by researchers involved in various biological disciplines [15,22,23]. In recent years, considerable progress has been made to understand the biogenesis and identification of circRNAs. Unlike mRNAs and lncRNAs, which undergo canonical splicing, circRNAs are primarily generated from exonic and/or intronic sequences of primary transcripts via back-splicing [24] (Figure 1). Backsplicing requires spliceosomal machinery, wherein circRNA production depends on both cis-regulatory elements and trans-acting factors, such as inverted-repeat Alu elements or dimerized RNA-binding proteins (RBPs) [24,25]. RBPs generally regulate back splicing via two functional mechanisms: (i) they facilitate back splicing by binding to intronic complementary sequences (ICSs) and stabilizing the transiently formed intronic RNA pairs flanking circRNA-forming exons, such as adenosine deaminase 1 acting on RNA (ADAR1) and DHX9, containing double-stranded RNA-binding domains [26–28] and (ii) they promote back splicing by directly bridging distal splice sites [28]. A classic example of these RBPs are QKI [29].

Figure 1. The schematic of existing circRNA formation models. Circular RNAs (circRnas) are primarily produced through back-splicing of exons (red arrow): a. intron pairing-driven circularization is guided by direct base pairing of the introns flanking complementary sequences or inverted repeats; b. RNA-binding protein (RBP)-driven circularization is produced through back-splicing of exons, which is mediated by RBPs that recognize intronic complementary sequences (ICSs) in flanking introns of circularized exons. In lariat-driven circularization, circRnas are formed during linear splicing: c. lariat-driven circularization facilitates the formation of an exon-containing lariat, which originates during an exon-skipping event (purple arrow); d. lariat-driven circularization facilitates the formation of intronic lariats, which are formed when an intron is removed during precursor mRNA (pre-mRNA) splicing (blue arrow). e. TircRNAs are formed when an intron is removed during precursor tRNA (pre-tRNA) splicing. EcircRNAs, exonic circRnas; eiciRNA, exon-intron circRnas; ciRnas, intronic circRnas; TircRNAs, tRNA intronic circRNA.

Currently discovered circRNAs may be sorted into four categories: exonic circRNAs (ecircRNAs); intronic circRNAs (ciRNAs); exon-intron circRNAs (eiciRNA); and intergenic circRNAs [13,30]. Most circRNAs that consist of exons are predominantly located in the cytoplasm, while a small portion of circRNAs consisting of introns is located in the nucleus. In addition, tRNA intronic circRNA (tricRNA) is another subtype of circRNAs produced during tRNA splicing [31]. Although generally expressed at low levels, a subset of circRNAs has been reported to be expressed as highly as, or even higher, than their linear counterparts, their expression being independent of related linear isoforms [14,15]. They generally exhibit diverse expression patterns across cell types and tissues of mammals [32–34]. Furthermore, most circRNA sequences are highly conserved. Approximately 15,000 human circRNAs sequences have been detected in parallel mouse genomes [33]. Another essential characteristic of circRNAs is that they cannot be degraded by RNase R exonuclease and are therefore more stable than linear RNAs [35]. The median half-life of circRNAs (18.8–23.7 h) was reportedly at least 2.5 times longer than their cognate linear RNAs (4.0–7.4 h) [36].

Emerging evidence suggests that circRNAs participate in various pathophysiological processes and act via diverse mechanisms [37]. In the nucleus, circRNAs regulate (i) parental gene expression by interacting with the small nuclear ribonucleoprotein U1 and enhancing RNA polymerase II activity [38], (ii) the splicing of its cognate mRNA via R-loop formation or by modulating the balance between canonical splicing and backsplicing [39,40], and (iii) centromeric chromatin loops occurrence by forming R-loops [41]. In the cytoplasm, circRNAs function as (i) miRNA ‘sponges’, by binding miRNAs and preventing them from interacting with target mRNAs [42], (ii) protein decoys to modulate their activities [43], (iii) protein scaffolds to facilitate contact between proteins [43], (iv)protein recruiters to recruit specific proteins to particular loci or subcellular compartments [37,43], and (v) translation templates for synthesizing polypeptides in the internal ribosome entry sites (IRES) in an- or N6-methyladenosine (m6A) – mediated cap-independent manner [44–46].

CircRNAs expression profiles in NAFLD

A growing body of evidence suggests that aberrant expression of circRNAs is linked to the occurrence and development of liver diseases, including HCC, liver regeneration, and hepatic steatosis [47–49]. Accordingly, increasing numbers of circRNAs have been identified as unconventionally expressed in NAFLD and functional as predictive biomarkers and potential therapeutic targets.

Animal and cell models, which play vital roles in elucidating pathophysiological mechanisms underlying NAFLD, have been used to investigate circRNA profiles in NAFLD. Currently, a total of seven circRNAs expression profiles, including three high-fat diets (HFD) models, three methionine and choline-deficient (MCD) models, and one HepG2 cell line model, have been recognized in the NAFLD research field. In mouse HFD-induced NAFLD models, Yuan, Guo, and Li et al. identified 93, 396, and 289 differentially expressed circRNAs, using RNA sequencing or microarray analysis, respectively [50–52]. Accordingly, Ou et al., Jin et al., and Zhu et al. identified 450, 132, and 4843 differentially expressed circRNAs in mouse MCD-induced NASH models, respectively [53–55]. Moreover, in the free fatty acid-induced HepG2-based NAFLD cell model, Guo et al. confirmed the presence of 357 differentially expressed circRNAs [49].

By extracting data from the original article as well as from supplementary data, we summarized the commonly identified circRNAs from all seven published articles, using two different criteria as follows: (i) since the terminology of circRNAs varies across different studies, circRNA ID codes in various databases were both transformed into gene names, the most frequently used method in circRNAs reports (Supplementary Figure S1A); and (ii) Aligent data were summarized using original Aligent nomenclature (Supplementary Figure S1B). As a result, 43 circRNAs were identified via at least two different expression profiles (Supplementary Table S1). Thus, subsequent experimental validation of these circRNAs was needed to confirm their biological functions and usefulness as serum biomarkers. Notably, the expression profile of the NAFLD cell line model (Guo 2017 [49] did not intersect with those of the other six animal models. Since most circRNAs sequences are highly conserved, focusing on the circRNAs that are conserved across species in animal sequence profiles may allow further opportunities for conducting clinical trials from bench to bed. Simultaneously, choosing conserved circRNAs from the human sequence profile was vital for investigating the underlying mechanisms and regulatory roles of circRNAs via cell line- and animal- experiments based on their identical cross-species characteristics of nucleotide sequences, biological syntheses, and functions. In NAFLD patients, a large number of differentially expressed circRNAs were identified via bioinformatics analysis, some of which may be related to the occurrence and development of NASH cirrhosis by specific regulatory networks, such as circRNAs-miRNAs-mRNA [56,57]. Currently, sequence data is derived mainly from animal and cell models of NAFLD, which are limited sources; thus, future studies integrating more sequencing data based on human specimens will accurately screen out the essential functional circRNAs in NAFLD.

Role of circRNAs in the pathogenesis of NAFLD

The complex pathogenesis of NAFLD remains unclear. The traditional ‘two-hit’ hypothesis has been eschewed in favour of the ‘multiple parallel hits’ hypothesis, which proposes that different pathogenesis promotes liver injury in parallel and not consecutively [6]. These pathogenic drivers include insulin resistance (IR), lipotoxicity, oxidative stress, endoplasmic reticulum stress, mitochondrial dysfunction, adipose tissue dysfunction, inflammatory cytokines, altered regulation of innate immunity, the gut – liver axis, and genetic and epigenetic factors, among others [58–61]. Experimental data have increasingly demonstrated that circRNAs play essential roles in the pathogenesis of NAFLD (Figure 2 and Figure 3). We summarized studies according to the pathogenic drivers based on circRNAs in NAFLD.

Figure 2. Schematic representation of circRnas in the pathogenesis of NAFLD. Multiple pathogenic drivers, such as insulin resistance, hepatic steatosis, oxidative stress, and inflammation, are interrelated and promote NAFLD progress to hepatic fibrosis and hepatocellular carcinoma in parallel and consistently. CircRNAs can exert diverse roles in the pathogenesis of NAFLD.

Figure 3. In vitro or in vivo experiments validation of functional mechanisms of predicted circRNAs. NAFLD, non-alcoholic fatty liver disease; CPEB1, cytoplasmic polyadenylation element-binding protein 1; PTEN, phosphatase and tensin homolog; AMPK, AMP-activated protein kinase; mTOR, mammalian target of rapamycin; PPARα, peroxisome proliferator-activated receptor α; CPT2, carnitine palmitoyltransferase 2; ACBD3, acyl-CoA binding domain containing 3; CPT1, carnitine palmitoyl transferase 1; ABCA1, ATP-binding cassette transporter A1; ROCK1, rho-kinase 1; AMPK, AMP-activated kinase; JAK2, Janus kinase 2; STAT5, signal transducer and activator of transcription 5; GPX4, glutathione peroxidase 4; CypD, cyclophilin D; mPTP, mitochondrial permeability transition pore; mROS, mitochondrial reactive oxygen species.

IR

IR, defined as a metabolic effect seen in target tissues because of insufficient insulin, involves glucose utilization in skeletal muscle, suppression of hepatic glucose production, and suppression of lipolysis in fat tissue [62]. IR is a common feature of NAFLD that contributes to its pathogenesis by stimulating adipogenesis and lipotoxicity [63]. CircHIPK3 levels were increased in response to oleate stimulation, and knockdown of circHIPK3 expression markedly inhibited adipose deposition and TG accumulation, and reduced cellular glucose content in HepG2 cells [64]. Furthermore, functional studies have indicated that circHIPK3 exerts its regulatory effect by sponging miR192-5p and regulating the expression levels of the transcription factor forkhead box O1 (FOXO1) [64]. Previous studies have demonstrated that FOXO1 integrates insulin signalling with mitochondrial function, and inhibition of FOXO1 expression improves hepatic metabolism during IR and the metabolic syndrome [65]. Thus, the circHIPK3–miR-192-5p – FOXO1 signalling pathway may be involved in the development of adipogenesis, IR, and hepatic steatosis [64]. CircSAMD4A was found to be significantly upregulated in obese patients, and was associated with high body mass index (BMI) and poor prognoses. Mechanistically, circSAMD4A knockdown increased insulin sensitivity, glucose tolerance, and energy expenditure by binding to miR-138-5p and regulating EZH2 expression [66].

Hepatic steatosis

Hepatic fat accumulation has been characterized as the uptake of fatty acids and de novo lipogenesis surpassing of fatty acid oxidation and exportation, resulting in hepatocyte steatosis caused by the imbalance between lipid acquisition and disposal [67]. Hepatic steatosis is a hallmark of NAFLD and a significant risk factor for conditions predisposing to NASH [68]. Guo et al. conducted a major survey, reporting that circRNA_0046367 expression was downregulated during hepatocellular steatosis in vivo and in vitro and its normalization abolished the inhibitory effects exerted by miR-34a on peroxisome proliferator-activated receptor α (PPARα) by blocking the miRNA/mRNA interaction with miRNA response elements [69]. Subsequent PPARα restoration improved hepatocellular steatosis by regulating the expressions of the genes associated with lipid metabolism, including carnitine palmitoyltransferase 2 and acyl-CoA binding domain containing 3 [69]. Finally, hepatotoxicity related to lipid peroxidation was resultantly resolved [69]. The same team also reported that another circRNA, circRNA_0046366, maintained PPARα levels by acting as a miR-34a sponge, which abolished the inhibitory effect exerted on PPARα by miR-34a⁷⁰. The restoration of PPARα expression further improved the expression of TG-specific lipolytic genes [i.e. carnitine palmitoyltransferase 1A (CPT1A) and solute-carrier family 27A (SLC27A)], thereby decreasing TG content and ameliorating hepatocellular steatosis [70]. The expression of another circRNA, circ_0057558, was upregulated in both in vivo and in vitro NAFLD models, which acted as a miR-206 sponge, facilitating lipogenesis and TG secretion by derepressing Rho-kinase 1 (ROCK1)/AMP-activated kinase (AMPK) signalling [71]. The JAK/STAT pathway plays a vital role in NAFLD, with its aberrant expression leading to abnormal lipid metabolism. CircScd1 was expressed at low levels in NAFLD tissues; and the aberrant expression of circScd1 affected the extent of hepatocellular lipidosis in NAFLD and promoted steatosis via the JAK2/STAT5 pathway [52]. The level of hsa_circH19, an independent risk factor for MetS, was closely associated with ectopic fat deposition in visceral organs [72]. Furthermore, hsa_circH19 knockdown increased the expression levels of adipogenic genes, which were accompanied by lipid accumulation. Mechanistically, the depletion of hsa_circH19 levels triggered the translocation of sterol-regulatory element binding proteins (SREBP1) from the cytoplasm to the nucleus via interaction with polypyrimidine tract-binding protein 1 [72].

Oxidative stress

Oxidative stress refers to an imbalance between the production of reactive oxygen species (ROS) and the scavenging capacity of an antioxidant system. ROS are primarily generated in mitochondria; a hepatic lipid overload may induce the overproduction of ROS, leading to mitochondrial dysfunction and oxidative stress and thereby initiating the pathological mechanisms of NASH [73,74]. CircRNAs mediate the production of ROS and promote ROS-induced cell death, apoptosis, and inflammation [16]. Hsa_circ_0048179 levels were downregulated in an oleate/palmitate-induced NAFLD model in vitro, accompanied by decreased levels of the antioxidant enzyme glutathione peroxidase 4 (GPX4). GPX4, a phospholipid hydroperoxide, protects cells against membrane lipid peroxidation [75]. Analysis of its mechanism revealed that hsa_circ_0048179 overexpression upregulated GPX4 levels by acting as a competitive sponge of miR-188-3p, thereby attenuating oleate/palmitate-induced ROS generation, mitochondrial dysfunction, lipid accumulation, and steatosis in HepG2 cells [75]. L02 cells exhibited substantially increased circRNA-4099 levels in response to hydrogen peroxide (H2O2)-induced oxidative stress, and abundant circRNA-4099 further stimulated the activation of the keap1/Nrf2 and p38 MAPK cascade by inhibiting miR-706 expression, thereby promoting apoptosis, ROS generation, and cell fibrosis [76]. Additionally, mitochondria-located circRNA SCAR in NASH fibroblasts can directly bind to ATP5B of ATP synthase in the mitochondrial permeability transition pore (mPTP) complex and block cyclophilin D (CypD)-mPTP interaction, inhibiting ROS generation, mitochondrial ROS output, and fibroblast activation [77]. Additionally, the in vivo delivery of SCAR to liver fibroblasts using mitochondria-targeting nanoparticles mitigated HFD-induced IR and cirrhosis [77]. Furthermore, clinical analysis showed that the downregulation of SCAR expression was associated with steatosis-to-NASH progression [77].

Inflammation

Inflammation is a physiological response to tissue injury or infection that produces various inflammatory mediators, such as cytokines, chemokines, and eicosanoids, which coordinate cellular defence mechanisms and tissue repair [78]. The hepatic inflammatory response is crucial in disease progression and transition from NAFLD to NASH [79]. CircRNA_002581 expression was significantly upregulated in NASH cell- and mouse- models. CircRNA_002581 knockdown markedly attenuated hepatic lipid accumulation, oxidative stress, and inflammation (as evidenced by decreased tumour necrosis factor α (TNFα) and interleukin-6 expression) [80]. Mechanistically, circRNA_002581 overexpression significantly relieved the inhibitory effects of miR-122 on its target CPEB1 by sponging miR-122 [80]. Furthermore, the CPEB1–PTEN – AMPK–mTOR pathway was found to be linked to autophagy and circRNA_002581 knockdown – mediated NASH alleviation (i.e. the circRNA_002581–miR-122–CPEB1 axis actively participates in the pathogenesis of NASH via the PTEN – AMPK–mTOR pathway-related autophagy suppression [80]. Furthermore, adipose inflammation, which is driven by activated macrophages, is crucial for the development of NASH [81]. Zhang et al. found that circARF3 acts as an endogenous miR-103 sponge to enhance TRAF3 levels, alleviating adipose inflammation by promoting mitophagy. Thus, the circARF3–miR103–TRAF3 axis may regulate adipose inflammation via mitophagy [82]. Accordingly, the function of circRNAs in NAFLD represents a new area of research, and further in-depth analyses of circRNAs may provide a more comprehensive understanding of NASH pathogenesis.

Hepatic fibrosis

Hepatic fibrosis occurs in response to chronic liver injuries, including viral hepatitis, alcohol consumption, and NAFLD, and its continuous deterioration can gradually develop into cirrhosis and even liver cancer [83–85]. Activation of hepatic stellate cells (HSCs) represents a pivotal event in fibrosis. Emerging evidence suggests a link between circRNAs and fibrosis; changes in the expression profile of circRNAs may promote or inhibit hepatic fibrosis by regulating HSC activation and proliferation. Correspondingly, circRNAs act as promotors of hepatic fibrosis. Circ-PWWP2A expression was upregulated in both TGF-β– and LPS-activated HSCs and in mouse fibrotic liver tissues, causing the augmented proliferation and activation of HSCs via sponging of miR-203 and miR-223 [86]. Similarly, upregulated circRSF1 expression in irradiated L×2cells promoted the inflammatory and fibrotic phenotypes of irradiated HSCs by counteracting the miR-146a-5p – mediated repression of Ras-related C3 botulinum toxin substrate 1 (RAC1), which was shown to promote HSCs activation, suggesting the role of circRSF1 in promoting hepatic fibrosis [87]. Moreover, circRNAs also function as inhibitors of hepatic fibrosis. CircFBXW4 expression was downregulated in liver fibrogenesis, whereas its overexpression inhibited HSCs activation and proliferation by targeting miR-18b-3p to regulate FBXW7 expression [88]. The inhibitory effects of another circRNA, circCREBBP, on liver fibrosis has also been observed. Mechanically, circCREBBP sponges miR-1291 to promote left-right determinant cluster 2 (LEFTY2) expression, suppressing the activation and proliferation of LX-2 cells and alleviating liver fibrosis injury in mice [89].

Hepatocellular carcinoma (HCC)

HCC is one of the most common malignancies worldwide, and its steady increase is concurrent with the growing NAFLD epidemic [90]. In many countries, NAFLD is projected to become the leading cause of HCC, replacing viral- and alcohol-related liver disease [91]. Accumulating evidence indicates that circRNAs are closely related to HCC tumorigenesis and progression. Circβ-catenin encodes a novel 370-amino acid β-catenin isoform, which protects β-catenin from GSK3β-mediated degradation and subsequently activates the Wnt/β-catenin pathway, thus promoting tumour growth [92]. Downregulation of cSMARCA5 expression was associated with HCC progression, and it promoted the expression of the tumour suppressor TIMP3 by sponging miR-17-3p/miR-181b-5p, thereby inhibiting HCC invasion and metastasis [47]. Additionally, some circRNAs also participated in the process of HCC drug resistance. CircSORE mediated sorafenib resistance by stabilizing YBX1. It was transmitted by exosomes, enabling the spread of sorafenib resistance [93]. It also appears to be a promising biomarker in liquid biopsy for the detection of early-stage HCC.

In addition, using a series of bioinformatics-based predictions, microarray analyses, and qRT-PCR validation, seven qRT-PCR verified circRNA – miRNA–mRNA pathways, namely circRNA_002581–miR-122–SLC1A5, circRNA_002581–miR-122–PLP2, circRNA_007585–miR-326–UCP2, circRNA_021412–miR-1972–LPIN1, circRNA_29981–miR-181b – SIRTI, has_circ_0001453–miR-27b-3p – PEG10, and has_circ_0000313–miR-6512-3p – PEG10 were predicted and constructed, which may play essential roles in lipid metabolism, hepatocellular steatosis, and liver fibrosis [49,53,54,57]. However, the specific mechanisms underlying these pathways need extensive exploration. The circRNA-miRNA-mRNA signalling cascade may be a good topic of research for further NAFLD studies (Supplementary Figure S2).

CircRNAs as novel biomarkers in NAFLD

The proportion of individuals with NAFLD is reaching epidemic levels worldwide, imposing considerable economic and health burdens and diminishing the quality of life. Liver biopsy remains the gold standard for diagnosing NAFLD. A biopsy is difficult because of its well-known limitations, including invasiveness, high cost, poor acceptability, sampling errors, and inter- and intra-observer variability, rending it unsuitable for screening populations at risk [94,95]. Therefore, efforts to identify accurate non-invasive biomarkers for high-risk NAFLD individuals must be expedited. Moreover, NASH, accompanied by profound hepatocyte damage, liver inflammation, and fibrosis, is an important determinant of liver-related morbidity and mortality and is a critical determinant of long-term prognoses in NAFLD [96–98]. Therefore, identifying patients at a high risk for NASH and advanced fibrosis is imperative for NAFLD management.

Blood-based biomarkers and clinical prediction rules

Recently identified hepatokines, including fetuin A, fetuin B, retinol-binding protein 4, and selenoprotein P, as well as several novel adipose tissue – derived and macrophage – derived proteins, including Gremlin 1 and Fibrinogen-like protein 2, which are associated with hepatic lipid metabolism and hepatocyte injury, have been developed to identify NAFLD individuals [99–101]. Meanwhile, an increasing number of blood-based inflammatory variables, including Keratin 18, soluble Fas, C-methionine, IL-1Ra, and fibrosis indicators containing terminal peptides of procollagen III (Pro-C3), have been developed to identify NASH and advanced fibrosis [102–107]. Although tissue-derived protein-based biomarkers hold promise, these are beset by certain limitations, including transcriptional reprogramming during disease progression, uncertainties associated with transport mechanisms, and increased susceptibility to clinical features (diabetes, hypertension, BMI, etc.) [99,108]. Given the low diagnostic efficacy of single indexes, several non-invasive diagnostic models, including the fatty liver index (FLI), hepatic steatosis index (HSI), and NAFLD liver fat score (LFS) have been investigated for their ability to predict NAFLD [109–111]. Scoring systems, including the NAFLD fibrosis score, FIB-4 index, enhanced liver fibrosis (ELF) test, and liver stiffness measurement via vibration-controlled transient elastography (LSM-VCTE) with enhanced diagnostic accuracy, have been established to diagnose patients at increased risk for NASH and advanced fibrosis [112,113]. However, uncertainties regarding the best diagnostic cut-off value, difficulty distinguishing between NASH and simple steatosis, and susceptibility to clinical features have limited the widespread application of these systems in clinical practice. A larger clinical cohort is needed to validate these scoring systems, especially among different ethnic groups or non-bariatric surgery populations.

CircRNA-based biomarkers

The epigenetic regulatory role of ncRNAs in NAFLD is being recognized rapidly. Thus, circRNAs appear to present a tremendous potential as extremely effective biomarkers of NAFLD for the following reasons: First, circRNAs are abundantly expressed in a cell type- and tissue-specific manner; for example, liver-specific circRNAs that represent liver pathologic damage caused by NASH and fibrosis may be isolated in serum or plasma [32]. Second, circRNAs are resistant to degradation by RNase R exonuclease and are more stable than their cognate linear transcripts, suggesting that circRNAs have an inherent advantage as serum markers that can be used for diagnosing and monitoring diseases [35]. Finally, circRNAs are primarily localized in the cytoplasm and are easily detected in the peripheral circulation via polymerase chain reactions, thus providing a much higher sensitivity than that of protein biomarkers. The expressions of cytoplasm-localized circRNAs, such as circRNA_0046367, circRNA_0046366, hsa_circ_0048179, circScd1, hsa_circRNA_021412, and circRNA_0001805 were significantly downregulated in NAFLD cell and animal models as well as in NAFLD patients. Restoring these circRNAs ameliorated oxidative stress, hepatic steatosis, mitochondrial dysfunction, and disease severity [49,52,69,70,75,114], suggesting that they could distinguish patients with NAFLD from those with abnormal liver function. Downregulation of the expression of another mitochondrial-localized circRNA, SCAR, in liver fibroblasts of NASH patients was associated with ROS generation, fibroblast activation, and steatosis-to-NASH progression [77], which may be sufficient to distinguish between patients with NAFLD and those with NASH. However, studies have rarely reported circRNAs that distinguishing between the different stages of NAFLD, such as NAFL, NASH, NASH-related fibrosis, and NASH-related HCC. Although the concept of circRNAs as biomarkers for NAFLD and NASH is conceptually appealing, further studies are needed to identify the specific circRNAs that can serve as biomarkers in NAFLD.

Exosomal ncRNA sorting mechanisms and functional mystery

The biological function and mechanism of exosomal-derived circRNAs have been reported in multiple studies on T2DM, the tumour microenvironment, and inflammatory diseases [115,116]. Given the critical role played by protein- or miRNA-enriched exosomes in the pathogenesis and progression of NAFLD as well as the versatility, individuality, and stability of circRNAs, it is reasonable to infer that exosomal circRNAs may also represent a novel frontier in NAFLD research, thereby providing us with new diagnostic concepts. Exosomes are nano-sized membrane-bound vesicles, whose biogenesis and secretion of which are primarily regulated by the endosomal sorting complex required for transport (ESCRT) and the RAB family of small GTPase proteins [117]. The relative proportion of exosomes produced and released varies depending on factors such as cell type, environmental conditions, and other stimuli [118]. Recent studies have indicated that exosomes serve as novel mediators of long-distance cell – cell communications by transferring various bioactive cargos, such as proteins and RNAs, from their parent cells to distant target cells to induce phenotypic changes in the latter [119,120]. CircRNAs could be packaged and secreted in exosomes, which can carry circRNAs to transfer biological information. For example, circ-133 is enriched in the plasma exosomes of colorectal cancer patients, and exosomal circ-133 derived from hypoxic cells was transported into normoxic cancer cells, promoting cancer metastasis [121]. The sorting mechanism of exosomal ncRNAs (ex-ncRNAs) and their physiological functions in recipient cells are attractive and promising as a prospective research field.

Exosomal ncRNA sorting mechanisms

ncRNAs are selectively integrated into exosomes via two mechanisms: exo-motif recognition by RBPs and miRNA-circRNA reciprocal transportation. Regarding exo-motif recognition by RBPs, ncRNAs are most likely transported into exosomes based on specific exo-motifs in their nucleotide sequence including the recently proven ‘GGAG/CCCU’, ‘GGAG/CCCU’, and ‘5ʹ-GMWGVWGRAG-3ʹ’ for miRNAs, lncRNAs, and circRNAs, respectively [122–125]. Exosomal-sorting RBPs, including heterogeneous nuclear ribonucleoprotein A2B1 (hnRNPA2B1), argonautes, alyref, and fus, specifically recognize and load exo-motif ncRNAs into exosomes [122,126]. Regarding miRNA-circRNA reciprocal transportation, miRNAs mediate circRNA secretion into exosomes and vice versa. For example, miR-7 overexpression in HEK293T cells reduced the level of CDR1as circRNA in exosomes [127]. In contrast, CDR1as circRNA may also stabilize and transport miRNA-7 in neurons [128]. This led to the following seminaries: (i) exosomal-sorting RBPs have been proven to recognize a single exo-motif from a single ncRNA and sort a single ncRNA into an exosome [122,129] or (ii) exosomal-sorting RBPs recognize multiple exo-motifs separately from miRNAs, circRNAs, lncRNAs, and thus sort them into exosomes [124] (iii) hypothetically, exosomal-sorting RBPs could recognize a single motif from one miRNA/circRNA/lncRNA, consequentially ‘dragging’ or ‘hijacking’ the miRNA-circRNA-lncRNA binding complex and sort these into an exosome. These hypotheses stem from the observations that indicate inverse correlation of several liver-specific protective miRNAs in the serum and liver, as noticed during our previous meta-analysis, which showed that hepatic miRNA-122 levels in NAFLD were decreasing, whereas miRNA-122 levels in the sera were continually increasing from NAFL to NASH [130]. We inferred that miRNA-122 was possibly ‘hijacked’ from other circRNAs and ‘dragged’ into sera with increased stability via circRNA bonding. Therefore, the ncRNA-exosomal-sorting biologic process appears complicated and may need further investigation (Figure 4).

Figure 4. Possible sorting mechanisms and functional pathways for exosomal ncRnas. ncRnas are sorted into exosomes via two potential routes: (i) the RNA motif and RBP-mediated pathway: ncRnas with specific ex-motifs are recognized by some RBPs and packaged into exosomes via the formation of an ncRNA-RBP binary, or ternary, complex. (ii) the ncRNA-related pathway: changes in miRNA levels cause circRNA to be packed into exosomes and vice versa. After entering a recipient cell, exosomal ncRnas (i) are separated and released to perform functions specific to each ncRNA or (ii) remain in the circRnas-miRNA or exosomal-sorting-RBPs binding complex without participating in any biological function. RBPs, RNA-binding proteins; ex-ncRnas, exosomal ncRNA; MVBs, multivesicular bodies; AGO, argonaute; ESCRT, endosomal sorting complex required for transport; miRisc, miRNA-induced silencing complex.

Exosomal ncRNA functional mystery

The functional activity of ex-ncRNAs following the entry of the miRNA-circRNA-lncRNA binding complex into the recipient cell requires exploration. Two different processes have been proposed regarding ex-ncRNA’ activity in recipient cells: (i) these ncRNAs are functional, primarily as separate ex-ncRNA forms, such as separated ex-miRNA or ex-circRNA, which elicit physiological responses in recipient cells. Recent studies have reported that circ -0,051,443, which is transported from normal cells to HCC cells via exosomes, suppresses malignant biological behaviour patterns by promoting cell apoptosis and arresting the cell cycle [131] (ii) these ncRNAs are unfunctional as they exist as dimer or trimer complexes after being sorted out of donor cells in two different hypothetical scenarios: first, miRNA becomes unfunctional because of its fixed bond formation with circRNAs after being sorted-out of donor cells [132] second, circRNA also becomes unfunctional because of its settled bond formation with exosomal sorting RBPs, such as A2B1 or Syncrip, instead of switching to silencing complex functional RBPs, such as AGO2. Therefore, we inferred that detectable ncRNAs in sera or exosomes do not always indicate that these would become functional following entry into recipient cells (Figure 4).

Considered together, a comprehensive understanding of the mechanisms underlying the sorting of circRNA, miRNA, and RBP assemblies into exosomes as well as disassembling the circRNA-miRNA-RBP binding complex into separate and functional units or its continuance as a bonding complex without function remains elusive and need further investigation.

Therapeutic potential of circRNAs in NAFLD

Currently, NAFLD-approved pharmacological therapies are unavailable. Vitamin E and pioglitazone are the most used medications in current clinical practice, however, their use is restricted to specific NASH patients, and potential risks limit their widespread use. Therefore, investigating these agents’ long-term efficacy and safety will be critical for NAFLD management. Accumulated evidence has shown that a subset of relatively abundant circRNAs plays essential cellular roles in the development and progression of NAFLD, piquing substantial interest in developing circRNA-targeted therapies.

CircRNA-targeted therapeutics

Precise interference via the targeting of unique back splice junctions of circRNAs may alleviate disease manifestation. Several approaches have been developed to study and target circRNAs including circRNA expression vectors, RNA interference (RNAi) – mediated circRNA knockdown, nanoparticle- and exosome-based delivery vehicles, antisense oligonucleotides (ASOs), conditional circRNA knockdown, and CRISPR-Cas9/13-mediated circRNA knockdown (Figure 5). Lentiviral/adenoviral vectors, expression plasmids, small interfering RNAs (siRNAs), and short hairpin RNAs (shRNAs) are currently the most convenient methods for the overexpression and knockdown of circRNAs. Lentivirus-mediated hsa_circ_004817 overexpression reduced oleate/palmitate-induced lipid deposition and mitochondrial dysfunction in HepG2 cells [75]. siRNA- or shRNA-mediated circRNA_002581 knockdown effectively attenuated lipid droplet accumulation and pro-inflammatory cytokine expression in NASH cell and animal models [80]. However, these molecules have many limitations, such as instability, lack of cell specificity, low rate of intracellular entry, immunogenicity, and other off-target effects [133]. Nanoparticles or exosomes as delivery systems can improve the feasibility of circRNA-based therapeutics. The novel GA-RM/GZ/PL nano drug system – mediated circRNA_0001805 transport to the liver, significantly inhibited lipid accumulation and inflammation in NAFLD [114]. However, their safe use in the clinical setting is still being determined. Exosomes are likely more biocompatible than nanoparticles; however, their isolation and purification methods are challenging, and their ex-circRNA sorting mechanisms are complicated. In addition, ASOs could bind to circRNAs via Watson – Crick base pairing to induce targeted circRNA cleavage; however, the long-term toxicology and promiscuous off-target effects cannot be ignored [134]. The cre-lox conditional circRNA knockdown system was recently used to knockdown circRNAs in specific cells. Cre-dependent shRNA targeting cZNF532 was intravenously administered to PDGFR-β-cre mice to confer specific cZNF532 knockdown in pericytes [135]. Moreover, CRISPR technology and Cas9/Cas13 systems have shown promise in knocking down circRNAs with high specificity and efficiency. CRISPR/Cas9 disrupted the inverted complementary sequences of flanking introns by a guide RNA (gRNA) to target and cleave circRNAs; however, the targeted intron sequence is difficult to determine, and off-target effects are hard to avoid. Encouragingly, CRISPR/Cas13 can directly target the back-splice junction of circRNAs without any impact on related mRNAs, and it can discriminate circRNAs from their linear cognates [136,137]. However, these approaches are still in the early stages of development, and many obstacles remain. Further in vivo investigations to identify toxic and other side effects following the long‐term use of these systems are required before clinical application.

Figure 5. Strategies used to target circRnas. a siRNa/shRNA targeting the back-splice junction of circRnas induces circRnas cleavage. b CircRNAs expression plasmid leads to circRnas overexpression. C gold nanoparticle-mediated delivery of circRnas. d exosome-mediated delivery of circRnas. e antisense oligonucleotides (ASO) bind to targeted circRnas through complementary pairing to induce circRnas cleavage. f cre-lox system-based conditional circRNA knockdown to induce circRNA cleavage. g CRISPR/Cas9 disrupting the flanking intronic complementary sequence induces circRNA cleavage. h CRISPR/Cas13 directly targeting the back-splice junction of circRnas induces circRNA cleavage.

CircRNA-based combination therapies

Interactions among ncRNAs, such as circRNA – miRNA and lncRNA – circRNA–associated ceRNA networks, may give rise to new concepts on targeted therapy. Lin et al. revealed that the inhibition of circRNF111 impaired insulin sensitivity and accelerated lipid accumulation via the circRAF111–miR-143-3p – IGF2R cascade. However, co-insertion of si-circRNF111 and miR-143-3p antagomir or IGF2R overexpression plasmid into a cell- or animal- models partially reversed this effect [138]. Besides, the traditional Chinese medicine Qianggan extract relieved NASH by modulating the lncRNA/circRNA immune ceRNA network [55]. Therefore, it is essential to explore the interactome in these cascades for comprehensive therapeutic targeting However, current research is limited to the cellular and animal levels; no clinical trials to test circRNA-based therapies for NAFLD have been registered to date. Over the previous two decades, therapeutic agents based on mRNAs and miRNAs have reached phase III clinical trials and have even been commercialized. siRNA nanoparticles previously approved for treatment are currently undergoing clinical trials, indicating that circRNA-targeted and/or ceRNA network-based therapeutic strategies for NAFLD may begin a new chapter.

Limitations of circRNA-based therapies

Several outstanding issues may need to be resolved before applying circRNA-based therapies to clinical practice. First, the mechanisms and pathophysiological roles of circRNAs in NAFLD remain largely unknown. Exploring circRNA-dependent ncRNAs, RBPs, and other mechanisms could drive the development of combination therapies, similar to the previous cocktails of highly active antiretroviral therapies (HAART), which marked eased the HIV burden. One classic example originates from a previous study on CDR1as and miR-7: ex-circRNA retained biological activity because of CDR1as exosomes abrogating miR-7–induced growth suppression in receipt cells, emphasizing the importance of combining miRNAs and circRNAs in future drug-development [127]. Simply put, combining circRNA-related anti-miRNAs or anti-lncRNAs may maximize the therapeutic effect of circRNAs. Second, the definitive identification of cellular localization and functional compartment of circRNAs is warranted to prevent adverse effects on off-target tissues or cells. Zhang et al. reported that exo-circ-DB secreted by adipocytes promoted HCC growth and reduced DNA damage [139]. Therefore, accurately identifying related cells or tissues secreting circRNAs may help exploit appropriate circRNA-based targeted therapy strategies. Accordingly, targeting these cells or tissues may enhance our understanding of precision therapy. Third, employing safe and convenient delivery mechanisms is crucial. Newly developed drug delivery systems, including ‘metal-organic framework – based stimuli-responsive systems’ and “in situ bio-self-assembled DNA-Au nanostructures loaded exosomes’, display low toxicity, enhanced targeting capacity, and high therapeutic efficacy hold promise as tissue-specific drug delivery agents [140,141]. Encouragingly, Jaffrey et al. used Twister-optimized RNA for stable overexpression to achieve rapid RNA circularization and the highly efficient expression of circRNAs, opening a new approach to circRNA overexpression [142]. Therefore, resolving these issues may enhance the capability of researchers to develop new therapeutic strategies against NAFLD.

Limitations and future perspectives

CircRNAs have become a new research hotspot in the field of NAFLD in recent years. Many circRNAs with important physiological and clinical significance have been identified. Given their stability, tissue specificity, and functional activity, circRNAs are promising to as a non-invasive diagnostic tool and therapeutic target for NAFLD. However, most current studies are limited to the pre-clinical phase, and the approach to clinical translation will soon be revealed. Several challenges and innovations remain to be overcome and updated in future investigations. First, whether a circRNA with relatively low abundance can achieve measurable effects remains controversial. CircRNAs are generally expressed at lower levels than their linear counterparts, suggesting that regulatory circRNAs possess unique subcellular localizations according to function instead of even distribution throughout the cell [143]. Therefore, it is indispensable to assess the subcellular localization of specific circRNAs and their possible nucleoplasmic shuttle mechanism. A recent study revealed that UAP56/URH49 are associated with circRNA localization and control the efficiency of nuclear export by sensing the lengths of mature circRNAs; it will be of interest to further investigate whether circRNAs are subjected to the same nuclear export mechanism as mRNAs [144]. Second, the relationship between epigenetic modification and circRNA functions remains largely undetermined. As one of the most abundant RNA modifications, m6A has been reported to regulate circRNA translation and degradation. Moreover, m6A-modified circRNAs could inhibit innate immunity and promote tumour progression and metastasis [145–147]. Other effects of m6A on circRNAs, such as the splicing process and subcellular localization, and its biological functions in other diseases could be further investigated. In addition, whether m6A-modified circRNAs can be utilized as diagnostic biomarkers and therapeutic targets also needs to be explored. Finally, whether highly stable circRNAs could be developed into a safe and effective vaccine platform, similar to the mRNA vaccine technology, remains unclear. CircRNAs have been reported to participate in different immune responses in multiple immune diseases [148]. More recently, Qu L et al. developed a circRNA vaccine that elicited potent neutralizing antibodies and T cell immune responses, effectively protecting against SARS-CoV-2 and its emerging variants. Compared with the mRNA vaccine, the circRNA vaccine enabled more durable immunogens in a greater amount [149]. Therefore, circRNAs hold the potential to become an effective and safe platform for vaccination against viral infections.

Conclusion

Our understanding of circRNAs and exosomal circRNAs has advanced substantially in recent years, providing new perspectives on diagnosing and treating liver diseases. However, many aspects of circRNA regulatory mechanisms underlying NAFLD, a class of highly heterogeneous and multifactorial metabolic disorders, and exosomal ncRNA sorting mechanisms remain obscure, elaborating the need for continued exploration. Consequently, many difficulties and challenges need to be addressed before applying circRNA-based diagnostics and therapeutic strategies against NAFLD at the clinical level, thus further basic experiments and clinical studies may be warranted.

Author contribution

Hong Tang conceived the idea of the review article. Qingmin Zeng and Chang-Hai Liu performed the literature search and drafted the manuscript. Dongbo Wu and Wei Jiang researched data and draw the Figs and Table. Javier Ampuero, Lingyun Zhou, Hong Li, Lang Bai, and Manuel Romero-Gómez edited and revised the manuscript. All authors approved the final version of the manuscript.

Acknowledgments

Not applicable.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15476286.2023.2290769.

Additional information

Funding

This work was supported by the 1.3.5 project for disciplines of excellence, West China Hospital, Sichuan University (No. ZYGD20009); Sichuan Science and Technological Program (No. 2022YFS0338); Post-Doctor Research Project of West China Hospital of Sichuan University (2020HXBH079); Chengdu Science and Technology innovation project (2021-YF05-00800-SN); National Natural Science Foundation of China (No. 81900512 and No. 81802468).