CircRNA as an Achilles heel of cancer: characterization, biomarker and therapeutic modalities

Circular RNAs (circRNAs) are a class of endogenous noncoding RNAs characterized by their lack of 5′ caps and 3′ poly(A) tails. These molecules have garnered substantial attention from the scientific community. A wide range of circRNA types has been found to be expressed in various tissues of the human body, exhibiting unique characteristics such as high abundance, remarkable stability, and tissue-specific expression patterns. These attributes, along with their detectability in liquid biopsy samples such as plasma, position circRNAs an ideal choice as cancer diagnostic and prognostic biomarkers. Additionally, several studies have reported that the functions of circRNAs are associated with tumor proliferation, metastasis, and drug resistance. They achieve this through various mechanisms, including modulation of parental gene expression, regulation of gene transcription, acting as microRNA (miRNA) sponges, and encoding functional proteins. In recent years, a large number of studies have focused on synthesizing circRNAs in vitro and delivering them to tumor tissue to exert its effects in inhibit tumor progression. Herein, we briefly discuss the biogenesis, characteristics, functions, and detection of circRNAs, emphasizing their clinical potential as biomarkers for cancer diagnosis and prognosis. We also provide an overview the recent techniques for synthesizing circRNAs and delivery strategies, and outline the application of engineered circRNAs in clinical cancer therapy.

CircRNAs are a novel class of RNAs characterized by covalently closed structures without 5′ caps and 3′ poly(A) tails. The first circRNA was discovered in viruses several decades ago but they initially received little attention [1, 2]. Although a handful of circRNAs were found afterwards, they were considered errors in RNA splicing with limited function [1, 3, 4]. With the advancements of RNA deep sequencing technology and dedicated bioinformatics algorithms, an increasing number of circRNAs have been reported in eukaryotic cells, especially in the mammalian brain [5, 6]. CircRNAs have been implicated in various diseases, including neurodegenerative diseases, atherosclerosis and cardiovascular diseases, especially cancer [7,8,9,10,11]. Studies have shown that circRNA exhibit aberrant expression patterns in multiple cancers [12,13,14]. Research has also revealed the high abundance, remarkable stability, and tissue-specific expression patterns of circRNA in vivo, which can be detected and quantified using RT‒qPCR, ddPCR, Northern Blotting, In Situ Hybridization and NanoString nCounter, making circRNAs an ideal diagnostic and prognostic biomarker for cancer [15,16,17]. Memczak et al. and Hansen et al. proposed the function of ciRS-7 as a miRNA sponge regulating gene expression in 2013 [18, 19], which garnered significant interest. Furthermore, subsequent research has uncovered additional functions of circRNAs, such as modulating parental gene expression, acting as miRNA sponges and encoding functional proteins [20, 21]. Certain circRNAs have been shown to play roles in initiation, progression and metastasis through their functions [18, 19, 22, 23]. Thus, circRNA-based cancer therapies have been extensively investigated. Linear RNAs are circularized in vitro and administered to specific tissues of patients, exerting an inhibitory effect on tumors. However, the development of engineered circRNA therapies faces challenges, including the need for highly efficient and high-throughput circRNA synthesis methods and effective delivery systems for large circRNAs. Therefore, optimizing circRNA synthesis methods and developing reliable delivery vehicles are essential for advancing the application of engineered circRNAs in cancer therapeutics.

In this review, we summarize the current understanding of circRNAs, including their biogenesis, characteristics, function and detection, as well as highlight their enormous potential as biomarkers for diagnosis and prognosis in cancer. Moreover, we offer an overview of the current knowledge regarding circRNA composition and delivery methods, and highlight the clinical application of circRNAs in cancer therapy.

Based on their biogenesis, circRNAs can be classified into three primary categories [24]: exon-only circular RNAs (EcRNAs) [25], intron-only circular RNAs (ciRNAs) [21] and exon‒intron RNAs (EIciRNAs) [20]. Despite their diversity, circRNAs are generated from precursor mRNAs (pre-mRNAs) through back-splicing [26]. The mechanisms underlying circRNA biogenesis have not been completely elucidated. Back-splicing involves the 3’ splice donor site joining with the downstream 5′ splice site, resulting in a closed-loop structure with a 3′-5′ junction [27, 28]. Then, all or part of the introns are removed, and the rest of the sequences connect to form the three types of circRNAs. Three biogenesis models of circRNAs have been proposed: intron-pairing-driven circularization, RNA-binding protein (RBP)-induced circularization and lariat-driven circularization (Fig. 1A). In intron-pairing-driven circularization, introns on either side of exons are essential elements. The upstream introns are base-paired with the downstream introns, forming a hairpin and promoting the formation of circRNAs. These reverse complementary sequences include ALU repeats and other nonrepetitive elements [16, 28]. Additionally, certain RNA-binding proteins can interact with specific intron motifs to introns to form hairpins. In the lariat intermediate model, canonical splicing removes exons and introns from the pre-mRNA, creating a lariat structure. The intronic sequences are then removed in the internal splicing to form an EcRNA [29]. Intronic lariats that escape debranching can form ciRNAs [21]. Moreover, a small number of circRNAs originate from tRNAs that are spliced from precursor tRNAs (pretRNAs) [30]. During tRNA maturation, the tRNA splicing endonuclease complex cleaves pretRNAs, and a single enzyme called RtcB joins the exon halves and introns to form a tRNA intronic circular RNA (tricRNA) [31, 32].

Biogenesis and functions of circRNAs. A Lariat-driven circularization: after canonical splicing occurs the skipped exons and introns are spliced from the pre-mRNA to form a lariat structure. The intronic sequences are then removed in the internal splicing to form an EcRNA. Intronic lariats that escape debranching can form ciRNAs; RNA-binding protein (RBP) induced circularization: RNA-binding proteins can also bind to specific intron motifs to introns to form hairpins; intron-pairing-driven circularization: the upstream introns are base-paired with the downstream introns, forming a hairpin and promoting the formation of circRNAs. These reverse complementary sequences include ALU repeats and other nonrepetitive elements. B circRNA functions as a transcription regulator. EIciRNAs can interact with U1 small nuclear ribonucleoproteins and form the EIciRNA-U1snBNP complex, which associates with polymerase II at the parental gene promoters to enhance their expression. C circRNA functions as a miRNA sponge. CircRNAs with complementary binding sites can pairing with miRNA response elements of mRNAs and compete for miRNAs and sequester miRNAs away from their target mRNAs, thereby regulating their translation. D circRNAs are translated into proteins. Endogenous circRNAs with internal ribosome entry sites (IRESs) have been found to be translated by cap-independent translation initiation mechanisms. E circRNAs as biomarkers. CircRNA is expressed in tumors and can be secreted outside cells, with stable high expression in body fluids like plasma and urine. Their abundance has supported researchers in detecting them in biological liquids

Full size image

Stability

CircRNAs are exceptionally stable molecules with higher stability than linear transcripts [33]. With the structure of a covalently closed loop structure without the 5′ cap and 3′-poly tail, circRNAs demonstrate increased resistance to RNase-mediated cleavage [34, 35]. For instance, EcRNAs are stable in cells with a half-life of 48 h, whereas the average half-life for mRNA is 10 h [35]. High stability can result in the accumulation of circRNAs during various physiological processes [15, 36]. The enhanced stability of circRNAs facilitates their detection and quantification, often at higher levels than their linear counterparts. This makes circRNAs particularly suitable as biomarkers for disease diagnosis.

Abundance

With the extensive application of RNA sequencing technology, circRNAs have been found to be abundant and widespread [37]. Studies have shown that circRNAs are highly prevalent in human tissues, particularly in the human brain [38]. More than 25,000 different circRNAs have been found in human fibroblasts, which are ten times more abundant than their linear counterparts [16, 23, 39]. Alternative circularization between different flanking complementary introns results in a gene forming many circRNA isoforms, thereby contributing to the overall diversity and abundance of circRNAs [28].

Temporal and spatial expression specificity

The expression patterns of circRNAs are spatiotemporally specific, varying at different stages of growth, development and tumor progression. CircRNA expression patterns are strictly regulated in eukaryotic cells and correlated with neural development [6, 36]. Moreover, circRNAs play a crucial role in several hallmarks of tumor progression. For instance, at the origin of liver cancer, circ-ZNF609 enhances the stemness of hepatocellular carcinoma (HCC) cells and promotes carcinogenesis [40]. Conversely, circMEG3 is downregulated during the invasion and metastasis of HCC, reducing its inhibition of telomerase activity and promoting cell replicative immortality and tumor metastasis [41]. Similarly, circRNAs exhibit differential expression in tumor tissues and adjacent non-malignant tissues [42, 43]. An analysis was performed to illustrate the circRNA profile of prostate adenocarcinoma and normal cells, bone and osteosarcoma, colon and colorectal adenocarcinoma, and kidney and renal cell carcinomas [44]. The results showed that the total circRNA abundance was lower than that in the normal tissues. Differential expression analysis revealed 34 upregulated circRNAs with statistical significance. These distinct expression patterns underscore the potential of circRNAs as biomarkers for cancer.

The aforementioned characteristics make circRNA an ideal tumor diagnostic and prognostic markers. In clinical practice, the expression levels of specific circRNAs can be measured to determine the presence of a tumor or assess the degree of tumor progression.

A large number of circRNAs have been identified by genome-wide analysis, and their unique characteristics suggest that they may probably play an essential role in biological processes. To date, sevaral possible functions have been proposed that have broadened our understanding of circRNAs, such as transcriptional regulation, miRNA sponges, and translation into proteins. However, additional functions of circRNAs remain to be discovered.

Transcriptional regulator

Although most circRNAs are located in the cytoplasm, ciRNAs and EIciRNAs are restricted to the nucleus and function as transcriptional regulators [20] (Fig. 1B). For instance, EIciRNAs can interact with U1 small nuclear ribonucleoproteins to form the EIciRNA-U1snBNP complex, which associates with polymerase II at the parental gene promoters to enhance their expression. Circ-EIF3J and circ-PAIP2 enhance the expression of their parental genes in HeLa and HEK293 cells by promoting the elongating activity of polymerase II [20]. Moreover, circSEP3, located in exon 6 of SEP3, can bind to its cognate DNA locus and form an RNA‒DNA hybrid, resulting in transcriptional pausing. This pausing subsequently recruits splicing factors, leading to the formation of alternatively spliced SEP3 mRNA with exon skipping [45]. Therefore, these intron circRNAs function as transcriptional regulators of parental gene transcription.

MiRNA sponge

MicroRNA (miRNA) sponges are considered one of the most critical functions of circRNAs. MiRNAs are thought to be key regulators of gene expression by base pairing with miRNA response elements of mRNAs and regulating their translation in physiological and pathological processes [46, 47]. CircRNAs with complementary binding sites can sequester miRNAs away from their target mRNAs, thereby relieving the suppression of miRNA-underlying genes (Fig. 1C). For example, ciRS-7 is a circRNA that contains more than 60 conserved miR-7 binding sites, enabling it to stably bind to miR-7. This interaction suppression of the biological activity and function of miR-7 and elevated expression of miR-7 targets [48]. Moreover, studies have showed that there is no one-to-one correspondence between miRNAs and circRNAs. Single circRNAs can bind to multiple miRNAs and perform different functions. For example, circHIPK3 can sponge miR-558 to suppress heparanase expression in bladder cancer (BCa) and can also bind to miR-124-3p to promote lung cancer [49, 50].

Translation into proteins

CircRNAs were long considered to be untranslatable due to the absence of elements necessary for translation, such as the poly(A) tail and 5′ cap. However, endogenous circRNAs with internal ribosome entry sites (IRESs) have been found to be translated via cap-independent translation initiation mechanisms (Fig. 1D) [51]. Additionally, N6-methyladenosine (m6A) modification can also drive the translation of circRNAs [52]. In the former mechanism, the ribosome can directly bind to the IRES to initiate the translation of circRNAs. Circ-ZNF609 with IRES can undergo cap-independent translation, whereas circARHGAP35 can undergo m6A-dependent translation. Moreover, these two methods can cooperate to improve the efficiency of translation [53]. Yang et al. suggested that circRNA translation increased under heat shock conditions, indicating that circRNA-encoded proteins may play a role in stress conditions [54]. While most functions of peptides translated from circRNAs remain unknown, and some peptides lack essential functional domains, further studies are needed to elucidate the detailed mechanisms of circRNA translation and the roles of circRNA-encoded proteins.

CircRNAs can intervene at various stages of tumor development through the aforementioned functions, and through artificially modulating of circRNAs, the goal of treating tumors can be achieved. CircRNA-based therapeutics offer a potential avenue for cancer treatment, as circRNAs serving as miRNA sponges could inhibit oncogenic circRNAs. Several circRNAs have been identified as potential therapeutic targets. Overexpression of circRNAs that inhibit tumor growth or silencing of circRNAs that play a positive role during cancer development can exert opposite effects, thereby suppressing tumor progression.

The National Cancer Institute has defined biomarkers as “a biological molecule found in blood, other body fluids, or tissues that is a sign of a normal or abnormal process, or of a condition or disease,” outlining several essential characteristics as biomarkers (Fig. 1E). An ideal biomarker should be specific, sensitive, detectable and predictable in biofluids such as plasma, saliva, urine or even in exosomes [55]. As mentioned earlier, circRNA fulfill these criteria. Numerous studies have confirmed that many different circRNAs are aberrantly expressed in tumors compared to the surrounding normal tissue and are associated with clinical tumor characteristics, such as location, tumor size and TNM stage [56,57,58,59]. To date, many circRNAs have been identified as potential cancer diagnostic and prognostic biomarkers, both individually and synergistically with known cancer markers [60]. These circRNAs have been found to be stably present at high levels not only in tissues and cells but also in body fluids, such as plasma, serum, exosomes, and urine [61]. Their abundance facilitates their detection in detecting them in biological liquids. Because most circRNAs can be detected in biofluids, liquid biopsy has emerged as a revolutionary tool for cancer management, diagnosis and treatment guidance. Liquid biopsy is a real-time, non-invasive method for early tumor detection, comprehensive tumor characterization beyond localized malignancies, and dynamic monitoring of disease progression [62] (Fig. 2). Liquid biopsy be combined with tissue biopsy or may even replace it in the future [63]. Clinical studies have applied circRNAs as diagnostic and prognostic markers for tumors (Table 1) (https://clinicaltrials.gov/), most of which are still in the “early adopter” phase. Extensive research is needed to establish circRNA expression patterns in diverse cancer populations and identify reliable diagnostic or prognostic biomarkers.

Detection and quantification of circRNA. A RT-qPCR. Divergent and convergent primer pair was applied to reverse transcription and quantitative PCR (RT‒qPCR) to detect circRNA and linear RNA. The fluorescence signals are detected, and circRNA levels are quantified by cycle quantification (Cq) values. PCR products can be verified by melting curves and gel electrophoresis. B ddPCR. Droplet digital PCR is carried out in droplets, and the absolute concentration of circRNA is determined based on the proportion of nonfluorescent partitions. C Northern blotting. The probes target the BSJ of the circular and universal sites of the linear transcript. RNAs were separated by agarose gel electrophoresis, and luminescence was detected. D In Situ Hybridization. In situ hybridization can provide the precise spatial position of circRNA in cells and tissues. E NanoString nCounter. NanoString nCounter is based on biotinylated capture probes and reporter probes. Probes span the BSJ of the circRNAs and can generate a signal when binding to the circRNA without any signal in its corresponding linear RNA. Probes and RNA samples are fixed together, and RNA and its corresponding probes can hybridize to form a complex. The complexes can be detected and counted by a charge-coupled device camera

Full size image

Detection and quantification of circRNA

Studies have demonstrated the aberrant expression of circRNAs in the cytoplasm and exosomes of various cancer cells. Therefore, early detection, validation, and quantification of abnormally expressed circRNAs could enable earlier diagnosis and treatment of cancer. Here, we discuss several methods for detecting the expression levels of circRNAs and the advantages of these different methods.

RT‒qPCR

Considering their sensitivity, specificity, speed and cost-effectiveness, RT‒qPCR-based techniques are the most widely utilized methods for the detecting and quantifying circRNAs in clinical laboratories [64]. CircRNAs in tissues or biofluids can be extracted using an RNA preparation protocol. Following RNase R treatment to degrade coexpressed linear RNAs, circRNAs were enriched and isolated [65]. Because of tissue and development specificity, primers for reverse transcription were designed to target the specific circRNA of interest. During reverse transcription, complementary DNA (cDNA) is synthesized from the circRNA, containing exon-exon junction sequences that can be specifically amplified by the primers. DNA without the junction sequence is not amplified. Fluorescence signals were then detected to quantify circRNA levels. RT‒qPCR assays with RNase R play a significant role at the medium-throughput scale and can be quantified by real-time PCR (Fig. 2A). However, for the circRNAs derived from a single locus, a single primer can amplify several different amplicons. It is imperative to verify the amplicons such as the melting curves.

Nevertheless, a major source of error in RT-qPCR is the reverse transcription step [66]. Amplification errors might occur in this process, such as template switching [67] and rolling circle amplification [68], resulting in false-positive error and overestimation of circRNA concentration by 34–55% [69]. Furthermore, these errors can be avoided by reverse transcription-droplet digital polymerase chain reaction (RT-ddPCR) and hybridization techniques.

ddPCR

Drolet digital PCR is a new generation of traditional quantitative polymerase chain reaction (qPCR) and is used for the absolute quantification of target nucleic acids. To address the challenge of rolling circle amplification, Chen et al. developed a method based on RT-qPCR [70]. By employing a Poisson distribution-based algorithm, this approach eliminates the effects of rolling circle amplification. Similar to RT-qPCR, ddPCR uses the same primers, probes and reagents as qPCR. RNA was isolated and enriched from fluids, but unlike RT-qPCR, RNA was not digested by RNase R. Reverse transcription is performed, followed by degradation of the RNA strand in the DNA-RNA duplex by RNase R. Subsequently, ddPCR is conducted, where cDNAs are encapsulated within droplets, and thermal cycling is initiated. Almost all the droplets contained one or fewer copies of the DNA. After the PCR routine, concentrations were determined based on the proportion of nonfluorescent partitions by Poisson distribution [71] (Fig. 2B). This method shows better performance in the accurate quantification of trace amounts of nucleotides, with higher sensitivity compared to RT‒qPCR [72]. Its sensitivity makes it suitable for detecting targets with low copy numbers without preamplification steps. Additionally, while the qPCR method relies on fluorescent dye intensity proportional to PCR amplicons and requires a standard curve for quantification, ddPCR does not rely on standard curves, thereby avoiding associated errors [73]. Hu et al. utilized this assay to demonstrate that upregulated expression of circGSK3β was positively associated with advanced stage and poor outcome of esophageal squamous cell carcinoma (ESCC) and identified that the circGSK3β expression level in plasma is a biomarker for the detection of ESCC [58]. However, ddPCR requires automation and a cost reduction for widespread adoption. If accepted by a clinical laboratory, the future of this is promising.

Northern blotting

PCR assays using RNase R offer a straightforward and rapid methods for circRNA detection. Nevertheless, there are some limitations to PCR assays. Linear RNAs with stable structures may not be fully digested by RNase R, and large circRNAs may not be absolutely resistant to RNase R [74]. The northern blot hybridization method has the advantage of detecting circRNAs without the digestion step and reversing the transcription step. In this method, hybridization begins with labeled probes binding to the complementary back-splicing junction of circRNAs. Following hybridization, denatured RNAs are separated by agarose gel electrophoresis and transferred to a nitrocellulose membrane via capillary blotting [75]. Circular and linear RNA can be detected simultaneously because of their different migration rates in the gel. Luminescence detection of the hybridized probes enables profiling of the hybrid signals (Fig. 2C). The probe design is the most crucial step in the entire experiment. To specifically detect a circRNA, the probes designed have two options: a short probe spanning the back splice junction (BSJ) or an exonic probe directed against a region within the circRNA. Despite its advantages, the northern blotting is more laborious and sensitive, make it more suitable for low-throughput identification of putative circRNAs. Many attempts have been made to save time and improve sensitivity. For example, the digoxigenin-labeled probe method can increase sensitivity and safety and shorten time [76]. Nevertheless, northern blotting still suffers from several drawbacks, such as the requirement for good-quality RNA samples and its limited accuracy for the quantitation of circRNAs.

In situ hybridization (ISH)

ISH assays are used to detect specific RNAs within cells and localize them at the subcellular level [77]. Fixed cells and tissues were exposed to labeled DNA probes at high concentrations to form stable DNA‒RNA strands. DNA probes with fluorescence signals can then be visualized and profiled by microscopy (Fig. 2D). It is challenging to design probes that are restricted to the BSJ sites and are more difficult to detect because of the lower sensitivity, RNA degradation, competition for secondary structures and so on. There are two main types of probes used for circRNA detection: one is binding on both sides of BSJ [6] and another is a probe set covering the entire length of the RNAs [78]. Probes that complement both sides of the BSJ sites contain approximately 20 nucleotides and can detect circRNAs even in the presence of their linear counterparts [19]. BaseScope is a method that can specifically detect circRNA [79]. After binding to BSJ, the signal can be amplified, marking the localization of circRNA. This method is compassionate, but has a low-throughput. The probe sets covering the entire length of the RNAs contain 30–50 individual 20-nucleotide singly labeled probes [78] and could bind different parts of the target RNA. Multiple probes nave been designed for this study. Each probe was labeled with a unique fluorophore and pooled before detection [77]. However, this method is limited to detecting circRNAs in the absence of their linear counterparts. ISH can provide the precise spatial position of circRNAs within cells and tissues, but is limited in its ability to detect multiple circRNAs simultaneously, which restricts its use in routine clinical laboratories.

NanoString nCounter

The NanoString nCounter is an emerging technology for quantifying circRNAs. This method utilizes a combination of capture probe and reporter probe, both of which can specifically bind sequences of different RNAs, such as the BSJ of circRNAs (Fig. 2E) [80]. The identification of different transcripts depends on the arrangement and combination of a string of fluorophores in the reporter probes. In this method, probes and RNA samples are fixed together, and RNA and its corresponding probes hybridize to form a complex. After hybridization, unbound probes are washed away, and the remaining complexes are detected and quantified using a charge-coupled device (CCD) camera. This method is suitable for high-throughput detection and can detect multiple circRNAs simultaneously. It has no high requirements for sample quality, and this method can work well, although the RNAs are formalin-fixed, paraffin-embedded samples [81]. However, each probe needs to be specifically designed for different circRNAs to ensure accurate detection.

CircRNA as a diagnostic biomarker

Biomarkers, such as AFP and CEA, have long been used for the diagnosis. However, conventional biomarkers often suffer from low specificity and sensitivity, and the absence of early symptoms in some cancers can lead to diagnosis at an advanced stage, thereby missing the optimal window for treatment [82]. The aforementioned characteristics of circRNAs address these limitations and they have potential as biomarkers for early detection. For instance, Xu et al. constructed a panel of five circRNAs using a liquid biopsy assay, which could robustly distinguish between patients with early-stage pancreatic ductal adenocarcinoma (PDAC) and those without diseases [83] (Table 2). The panel achieved an AUC of 0.81 for identifying early PDAC in the validation cohort, which was the highest AUC value when compared with other gastrointestinal cancers, underscoring its specificity for PDAC. Wei et al. identified a plasma circRNA panel containing three circRNAs that demonstrated higher accuracy than AFP in distinguishing individuals with early stages HCC from healthy individuals [84]. Logistic regression analysis indicated that it was independent markers for discrimination of early HCC tissues. In non-small cell lung cancer (NSCLC), hsa_circ_0014130 in plasma is associated with clinicopathological outcome [85]. Elevated levels of hsa_circ_0014130 in tumor accurately identify NSCLC tissue and the adjacent non-cancerous tissues with the sensitivity and specificity of 87.0% and 84.8%, respectively. Moreover, Wang et al. discovered that the abnormal expression of different circRNAs can not only detect NSCLC in the early stage, but also distinguish between lung adenocarcinoma (LUAD) and squamous cell carcinoma [86]. Using qPCR validation, four circRNAs were identified as potential NSCLC biomarkers. Hsa_circ_0077837 and hsa_circ_0001821 showed diagnostic potential, with AUCs of 0.921 and 0.863, respectively, for distinguishing NSCLC from healthy tissues. Moreover, the upregulation of hsa_circ_00010739 (AUC = 0.919) and downregulation of hsa_circ_0001495 (the AUC was 0.919) could predict pathological subtyping between LUAD and squamous cell carcinoma. In osteosarcoma, hsa_circ_0081001 emerged as an independent diagnostic factor, and its expression level can dynamically monitor changes in the illness of patients [87]. Serum samples were collected from five newly diagnosed patients at each treatment stage, revealing that RNA levels in chemoresistant patients increased before surgery, decreased post-surgery, and rose again after postoperative chemotherapy, while chemosensitive patients consistently maintained low levels throughout treatment. The ROC curve indicated that hsa_circ_0081001 (AUC = 0.898) outperformed traditional markers such as ALP (AUC = 0.673) and LDH (AUC = 0.80) in diagnosing overall survival (OS). In addition, the combination of circRNAs with traditional biomarkers can increase sensitivity and specificity. In gastrointestinal tumors, circCCDC66 promotes cancer cell proliferation, migration and presents differential expression pattern in colorectal cancer tissues, showing a differential expression pattern in colorectal cancer tissues [59]. The study demonstrated that the expression level of circCCDC66 was a good predictive biomarker for colorectal cancer detection. Moreover, circRNAs in exosomes provide a new way for cancer cells to communicate with other cells [88], exosome-derived circRNAs can also be novel diagnostic biomarkers. Tang et al. found that circRNAs encapsulated in exosomes demonstrated comparable detection power for gastric cancer (GC) as plasma circRNAs [89]. CircKIAA-1244 can be a potential biomarker for cancer diagnosis, with univariate and multivariate analyses suggesting its independence as a diagnostic indicator for GC. The AUC for distinguishing GC plasma and healthy plasma is 0.748 with the sensitivity of 77.42% and specificity of 68.00%. In addition, numerous other studies have also demonstrated the diagnostic potential of circRNA (Table 2).

CircRNA as a prognostic biomarker

Accurate cancer prognostication can provide doctors with a reference for follow-up treatment and guide treatment decisions for patients, potentially improving patient outcomes. Numerous single circRNAs or circRNA panels have been proven to have superior prognostic power (Table 3) [60, 90,91,92]. Among these, ciRS-7 and circ ITCH may be the most well-known circRNA among these circRNAs and has been associated with poor prognosis in various cancers, including NSCLC [93], colorectal cancer (CRC) [60], GC [94] and other cancers. CiRS-7 is highly abundant in many cancers and functions as an oncogene via the ciRS-7/miR-7 axis [95]. MiR-7 plays a suppressive role in tumor progression and can inhibit the proliferation and migration of cancer cells through various mechanisms [96]. As ciRS-7 acts as a sponge for miR-7, its high expression indicates a poor prognosis. To validate the prognostic value of the ciRS-7, Tian et al. [90] conducted a meta-analysis involving 1714 patients from multiple nations, and their ciRS-7 expression levels were measured using RT‒qPCR. The results showed that upregulated ciRS-7 expression was significantly correlated with worse OS and higher tumor stage in NSCLC and CRC [93]. Another important circRNA that has been proven to have a prognostic role is circ ITCH [97]. Circ ITCH functions as a well-known tumor suppressor in multiple cancers via various pathways. In BCa, a positive correlation exists between tumor histological grade and circ-ITCH expression. Patients with higher circ-ITCH expression had better OS than those with lower expression levels. Circ-ITCH can inhibit BCa progression, including proliferation, invasion and tumorigenesis, in vivo and in vitro through the circ-ITCH/miR-17, miR-224/p21, and PTEN signaling regulatory networks. Luo et al. found that circ-ITCH expression in ovarian cancer was reduced than that in paired adjacent tissues [91]. Univariate Cox regression analysis indicated a positive correlation between circ-ITCH levels and prolonged OS, while multivariate analysis identified it as an independent predictor of favorable OS. Moreover, other circRNAs were also found to be a potential prognostic biomarker for the clinical outcome. In PDAC, high expression levels of circ-PDE8A and its linear counterpart were noted [98]. Upregulation of circ-PDE8A was associated with lymphatic invasion (p = 0.014) and higher TNM stages (p = 0.005), while linear-PDE8A was not significantly associated with patient survival. In gastric cancer, circPVRL3 was downregulated, and this downregulation promoted cell proliferation and migration [99]. Reduced expression of circPVRL3 was negatively correlated with survival time, suggesting it may be an ideal biomarker for predicting medical prognosis.

CircRNAs play vital roles in cancer development such as proliferation, invasion and metastasis through various pathways (Fig. 3), with their expression levels dysregulated in cancer cells [100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119]. CircRNAs regulate tumorigenesis through specific signaling axes. Firstly, in the circRNA-miRNA-mRNA axis, circRNAs function as miRNA sponges, sequestering miRNAs and preventing them from repressing their target mRNAs, leading to oncogene upregulation and enhanced tumor proliferation. Secondly, the circRNA-RBP axis involves circRNAs binding to RNA-binding proteins, modulating their activity and influencing processes such as cell cycle progression. Thirdly, within the circRNA-mRNA axis, circRNAs can directly interact with mRNAs to regulate their stability and translation, affecting gene expression relevant to cancer. Lastly, the circRNA-protein axis sees circRNAs encoding functional proteins that can contribute to tumorigenesis. These diverse signaling pathways underscore the multifaceted roles of circRNAs in cancer biology, highlighting their potential as therapeutic targets and biomarkers. These findings open new avenues for treating cancer treatment, suggesting that regulating the levels of aberrantly expressed circRNAs may be an approach to cure cancer. Studies have demonstrated the potential of circRNAs as novel therapeutic agents [120, 121]. Synthetic and engineered circRNAs have been shown to have superior stability and strong translation in eukaryotic cells [122, 123]. The synthesis, purification and delivery of engineered circRNAs is attracting researchers' attention.

CircRNAs play important roles in cancer through different mechanisms. CircRNAs regulate tumorigenesis through specific signaling axes. Firstly, in the circRNA-miRNA-mRNA axis, circRNAs function as miRNA sponges, sequestering miRNAs and preventing them from repressing their target mRNAs, leading to oncogene upregulation and enhanced tumor proliferation. Secondly, the circRNA-RBP axis involves circRNAs binding to RNA-binding proteins, modulating their activity and influencing processes such as cell cycle progression. Thirdly, within the circRNA-mRNA axis, circRNAs can directly interact with mRNAs to regulate their stability and translation, affecting gene expression relevant to cancer. Lastly, the circRNA-protein axis sees circRNAs encoding functional proteins that can contribute to tumorigenesis. These diverse signaling pathways underscore the multifaceted roles of circRNAs in cancer biology, highlighting their potential as therapeutic targets and biomarkers

Full size image

Synthesis of therapeutic circRNA

Chemical molecule circularization strategy

The circularization of RNA involves an intramolecular reaction between the 3′ and 5′ terminal nucleotides of linear RNA. Various molecules, such as amines and thiols, have been proven to facilitate RNA cyclization in vitro [124]. The phosphate and hydroxyl groups at the 5′ or 3′ terminus of linear RNA were replaced by other functional groups for cyclization. Ethyl-3-(3′-dimethylaminopropyl)-carbodiimide (EDC) and cyanogen bromide (BrCN) are the most commonly used reagents. However, these chemical strategies have several drawbacks. One significant issue is the formation of linear products through intermolecular ligation, which are the main byproducts. Reducing the concentration of linear RNA could reduce the occurrence of intermolecular reactions, but this approach negatively impacts the efficiency of cyclization and is unsuitable for large-scale applications. Oligonucleotide splints that can bring the two ends closer together can promote intramolecular ligation [125]. Moreover, the formation of 2′,5′-phosphodiester bonds at the ligation junction, rather than the desired 3′,5′-phosphodiester bonds, can also leads to the generation of byproducts (Fig. 4A).

The synthesis, purification and delivery of engineered circRNAs. A Chemical molecule circularization strategy. Under the action of EDC and BrCN, the phosphate and hydroxyl groups at the 5′ or 3′ end of linear RNA are replaced by other functional groups to form a ring. B Enzymatic circularization strategy.T4 DNA ligase form phosphodiester bonds at both ends of linear RNA and cyclize at the junction sites of double-stranded RNA. C Ribozyme circularization strategy. In Group I PIE method, the 5′ half of the intron is seated at the tail of the exon, and the 3′ half is positioned at the head of the exon. The 5′ half of the intron is moved to the tail of the exon, and the remaining 3′ half is transferred to the head of the exon by self-splicing. After two transesterifications at the defined splice sites, circRNAs are formed from a 3′–5′ phosphodiester bond. Group II introns, which can connect the 5′ and 3′ ends of an exon through inverse splicing reactions and form 2′,5′-phosphodiester bonds at the ligation site. D Nanoparticles delivery. Nanoparticles can encapsulate circRNAs and target specific cancer cells. Then, the membrane ruptures, and the circRNA is released into the cytosol. E Exosomes delivery. Exosomes promote targeted cells endocytosis and then release circRNAs to the cytosol by endogenous mechanism. F Viral vector delivery. Viral vectors have the inherent capacity to enter cells and can effectively deliver their DNA or RNA cargo into the nucleus, with great efficiency

Full size image

Enzymatic circularization strategy

The enzymatic strategy for RNA circularization utilizes various nucleotide ligases to form phosphodiester bonds at both ends of linear RNA, thereby achieve cyclization [126, 127]. Commonly used enzymes include T4 DNA ligase, T4 RNA ligase 1 and T4 RNA ligase 2, all of which are from the bacteriophage T4 [127, 128]. For enzymatic ligation to occur, the linear RNA must have a 3′-OH and a 5′-monophosphate; otherwise, the 5' end must be phosphorylated with enzymes before the cyclization reaction can proceed. The function of T4 DNA ligase is to form connections and cyclize at the junction sites of double-stranded RNA, which requires the two ends of the substrate RNA connection to be perfectly complementary. The substrate of T4 RNA ligase 1 is single-stranded RNA, which can form 3′,5′-phosphodiester bonds in RNA and form circRNAs. The secondary structure of RNA is the key factor affecting cyclization efficiency, and T4 RNA ligase 1 requires the 3′ and 5′ termini to be in close proximity. Splints can effectively perform this function and bring the two ends of linear RNA into close proximity. T4 RNA ligase 2 is effective in cyclizing both single-stranded and double-stranded RNA, though it is more efficient at joining nicks in double-stranded RNA. Similar to chemical strategies, intermolecular reactions are the most important side reactions. The generation of byproducts can be minimized through helper oligonucleotide splints. However, enzymatic methods struggle with the ligation of large RNA molecules. In such cases, a permuted intron–exon self-splicing system (PIE strategy) may offer a more suitable approach (Fig. 4B).

Ribozyme circularization strategy

The chemical and enzymatic methods mentioned above are suitable for synthesizing smaller circRNAs. However, for larger molecules, alternative approaches are required. The ribozyme circularization strategy is effective for producing larger circRNAs [123]. The Group I Permuted Intron–Exon (PIE) method is the most commonly used ribozyme strategy. The 5′ half of the intron is seated at the tail of the exon, and the 3′ half is positioned at the head of the exon. Through self-splicing, the 5′ half of the intron is moved to the tail of the exon, and the remaining 3′ half is transferred to the head of the exon. After two transesterifications reactions at the defined splice sites, circRNAs are formed from a 3′–5′ phosphodiester bond. Group II introns, which can connect the 5′ and 3′ ends of an exon through inverse splicing reactions, can also be used for circRNA synthesis. However, group II introns form 2′,5′-phosphodiester bonds at the ligation site. Owing to the relatively simple reaction conditions and high efficiency of this method, the PIE method is the most widely applied method. However, there are still some limitations. First, the Group I intron method leaves a splicing scar in the ligation, which can induce immunogenicity [129]. Second, the RNA secondary structure could influence the efficiency of circularization and lead to the generation of byproducts (Fig. 4C).

Delivery of circRNA

The delivery of synthetic circRNAs primarily depends on viral and nonviral delivery systems (Fig. 4D). The unique terminus-free structure makes them resist degradation by exonucleases during their delivery at the cellular level. Additionally, their compact structure enhances the capacity of the delivery carriers.

Nanoparticles

Nanoparticles are nonviral delivery carriers composed of phospholipids, cholesterol and ionizable lipids. They are engineered to deliver different molecules for different applications such as imaging and therapeutic targets [130, 131]. Their sizes and compositions can be tailored for specific uses. Owing to its low production cost, it can be easily produced, transported, stored, and transported in large quantities, making them an effective carrier for circRNA [132]. In cancer therapy, nanoparticles can encapsulate circRNAs and target specific cancer cells. Upon reaching the target, the membrane ruptures, releasing the circRNA into the cytosol where it can exert its function. For example, a study demonstrated that delivering nanoparticles encapsulating siRNA targeting circDnmt1 significantly suppressed breast cancer tumor growth and extended the lifespan of mice [133]. Additional studies have shown the therapeutic potentials of nanoparticles. Nanoparticles that deliver circFoxo3 plasmids increase tumor cell apoptosis and delivered circEHMT1 plasmids inhibit lung metastasis of breast cancer [134, 135].

Exosomes

Exosomes are vesicles secreted by a range of cell types, such as T cells and cancer cells, carrying diverse molecules such as circRNAs and playing a crucial role in intercellular communication. Characterized by a lipid bilayer membrane with embedded signal receptors, exosomes isolated from cells exhibit innate bioactivity and biocompatibility, functioning effectively without further modification [132, 136]. Similar to nanoparticle, exosomes could protect their content from degradation, promote targeted cells endocytosis and release circRNAs to the cytosol by endogenous mechanism. Therefore, exosomes are recognized as ideal delivery vehicles for circRNAs. Currently, researchers have designed an exosome delivery system that could deliver mRNA vaccination [137]. As for cancers, there are almost tenfold more exosomes in tumor cells than in other cells. Thus, extracellular vesicles seem to be ideal tumor RNA carriers [138]. Wang et al. [139] found that ciRS-122 expression is a primary cause of oxaliplatin resistance in colorectal cancer (CRC), with cancer cells secreting exosomes containing ciRS-122 to spread drug resistance. Isolated exosomes from cancer cells were injected isolated exosomes into chemotherapy-sensitive mice and tumor volume did not shrink significantly after treatment with oxaliplatin. This study also used the natural carrier system to deliver siRNA targeting ciRS-122 in mice, suppressing its expression and enhancing chemotherapy sensitivity. In summary, the biocompatibility, stability in circulation, and permeability of biological barriers make exosomes ideal circRNA carriers for cancer therapy.

Viral vectors

Viral vectors, particularly adeno-associated viruses (AAV) and adenoviruses (Ads), are highly efficient in delivering circular RNA (circRNA) into various cell types, including cancer cells [140]. They enable stable and long-term expression of circRNA, with lentiviruses integrating into the host genome for prolonged expression and AAV providing stable episomal expression. Viral vectors also offer tissue-specific delivery through the selection of different serotypes or specific promoters, high transfection efficiency compared to physical or chemical methods, and precise regulation of circRNA expression using inducible elements [141, 142]. These features make viral vectors a crucial tool for delivering circRNA in basic research and therapeutic applications. In a previous study, a lentiviral vector was used to deliver circZKSCAN1 successfully. Moreover, the results showed that overexpression of circRNA repressed HCC growth in vivo and in vitro [143]. However, a disadvantage is that the vector may produce some unnecessary linear byproducts.

The antitumor application of circRNA

Gene silencing therapy

The positive roles of some circRNAs in tumors make them potential targets for cancer therapy. Several strategies have been developed to exploit this potential. One common approach is to use RNA interference (RNAi), which includes small interfering RNA (siRNA) and short hairpin RNA (shRNA), which target the back-spliced sequence of circRNA to induce cleavage. SiRNAs are long dsRNAs that target circRNAs via complementary pairing, leading to their incorporation into the RNA-induced silencing complex (RISC) for cleavage [144]. shRNAs, which have a loop and base-paired stems, can be converted into siRNAs. By specifically targeting the back-spliced sequence of circRNA, the circRNA can be knocked out, but does not involve its corresponding linear mRNA. In HCC, siRNA and shRNA targeting the specific sequence of the oncogenic circRNA Crd1as can inhibit its expression, thereby suppressing HCC cell proliferation and invasion [145]. Similarly, in lung adenocarcinoma (LUAD), Qiu et al. found that circPRKCI, which promotes tumor growth and can be silenced by E2F7, becomes a therapeutic target when transfected with small interfering circPRKCI, resulting in reduced tumor size. These findings highlight the therapeutic potential of targeting circRNAs in cancer treatment [146]. However, this strategy has some limitations, such as degradation by nucleases, lack of cell specificity, and off-target effects [131, 147]. To mitigate these issues, researchers have explored the use of circular sense RNA and its complementary linear RNA, which can reduce off-target effects due to their unique circular structure. Linear antisense RNA can form a complex with RISC, and caged circular siRNA has been designed based on this complex [148]. In a xenograft tumor model with U87-GFP cells in mice, the complex was found to induce GFP gene silencing, demonstrating its potential [149]. However, unpredictable off-target pathological effects are the main obstacle for siRNA to become a viable treatment. Since circRNAs may play roles in multiple tissues and diseases, it is crucial to understand their effects throughout the body and design siRNA specifically to minimize off-target effects before clinical application.

Gene editing therapy

The CRISPR-Cas9 system provides a precise method for genome editing, including the targeted disruption of circRNAs by interfering with the pairing of introns flanking the circularized exons [150, 151]. The CRISPR/Cas system has been proven to knockdown circRNAs with greater specificity than RNAi [152]. Zheng et al. utilized the CRISPR/Cas9 technique for the knockdown of circHIPK3 expression, resulting in suppressed cell proliferation in various human and cancer cell lines [23]. The first human trial of CRISPR-Cas9 was conducted by Cyranoski et al., where metastatic non-small cell lung cancer (NSCLC) patients who had not responded to chemotherapy, radiation, or other therapies were treated with CRISPR gene editing [153]. The trial used CRISPR-Cas9 to target the PD-1 gene in T cells, with recent results indicating the feasibility of this approach for patients with advanced lung cancer [154]. Meanwhile, clinical trials are ongoing for CRISPR-mediated PD-1 gene knockout in other cancers, including hepatocellular carcinoma (HCC) and advanced esophageal cancer [155]. In the context of HPV-related cervical cancer, where HPV E6 and E7 proteins play crucial roles in malignant transformation and maintaining the malignant phenotype, CRISPR/Cas9-mediated knockdown of these genes led to the death of HPV-16-positive cervical cancer cells and inhibited tumor growth in a nude mouse model [156].

Vaccine therapy

Recently, tumor vaccines have emerged as a novel and promising strategy for cancer therapy. Cancer vaccines can stimulate antitumor responses by expressing tumor-associated antigens, leading to the recruitment of immune cells, activation of T cells, and ultimately the destruction of tumor cells and the formation of sustained immune memory [157]. Recent studies have demonstrated that circRNAs can be translated into small peptides or proteins [158, 159]. Engineered circRNAs often exhibit consistently high levels of protein expression, resulting in prolonged antigen exposure, sustained T-cell activity, and durable immunological memory [123]. A comparison of a circRNA SARS-CoV₂ RBD vaccine with an mRNA vaccine revealed that the circRNA vaccine exhibited higher and more durable expression levels [160]. Moreover, the high stability and low immunogenicity of circRNA vaccines perfectly compensate for the limitations of mRNAs, positioning them the next blockbuster of cancer therapy. Despite their promise, reports on circRNA vaccines in cancer therapy remain limited. A study published in 2022 demonstrated the positive role of circRNA vaccines in the prevention and treatment of melanoma [161]. In this study, circRNA was circularized in vitro using a Group I PIE system, purified via liquid chromatography, and encapsulated in lipid nanoparticles (LNPs) to form a stable vaccine. This vaccine was successfully delivered into cells and translated into proteins. Following intramuscular injection into mice, there was a notable increase in proinflammatory cytokines and enhanced antigen-specific T-cell responses, indicating effective induction of both innate and adaptive immune responses. In immune-desert orthotopic and metastatic melanoma models, the vaccine exhibited superior antitumor efficacy, effectively inhibiting tumor formation and blocking lung metastasis. The antitumor response induced by circRNA vaccines is mainly exerted by cytotoxic T cells rather than antibody-based B-cell responses. Therefore, to enhance therapeutic outcomes, researchers have combined circRNA vaccines with adoptive cell transfer therapy, showing improved persistence of engineered T cells and superior efficacy compared to monotherapy in targeting late-stage melanoma. This evidence underscores the potential of circRNA vaccines as a viable alternative in vaccination strategies and highlights their promising role in cancer therapy. However, the role of vaccines against other malignancies remains unclear. The function of circRNA vaccines in these cancers remains to be explored.

To date, therapies that target circRNAs have not yet been investigated in clinical studies. Several challenges must be addressed before these approaches can be applied in humans, and the development process is complex and time-consuming.

CircRNAs, initially regarded as mere byproducts of mRNA splicing, have emerged as a significant class of noncoding RNAs with advancements in biological and computational technologies. They are currently the subject of extensive research across various biological fields, revealing a multitude of functions including acting as miRNA sponges, regulating parental gene transcription, and serving as templates for protein translation. CircRNAs are implicated in a range of physiological and pathological processes, and recent evidence suggests their involvement in tumorigenesis, cancer cell proliferation, migration, and invasion, forming a critical network in cancer development. The stability, abundance, tissue and development specificity and detectability of circRNAs make them promising candidates for cancer diagnostics and prognostics. Therefore, many studies have applied circRNAs for the clinical diagnosis and treatment of cancer.

However, compared with other noncoding RNAs, our understanding of circRNAs is still shallow and is far from applying circRNAs in clinical practice. Future research should focus on several key areas. First, the precise mechanisms of circRNA formation, degradation and function remain unclear [162]. Further explorations are needed to fully elucidate and deepen our understanding of circRNAs. Second, we lacked a standardized methodology for identifying, validating and detecting clinical circRNAs. Reliable detection methods for circRNAs in bodily fluids are essential. Third, because most circRNAs are expressed at a low level and cannot be detected accurately, research should focus on improving the sensitivity and accuracy of circRNA detection. Finally, most circRNA studies were single-center and retrospective, with limited validation and clinical translation. Prospective validation and controlled clinical trials are necessary to establish circRNAs as reliable biomarkers.

Although the mechanisms and physiological and pathological effects of circRNAs in cancer are still controversial, circRNAs hold significant potential as diagnostic, prognostic and predictive biomarkers. This field has great translational potential in the future. However, before achieving this, we still need to overcome some knowledge challenge.

Not applicable.

GC:

Gastric cancer

NSCLC:

Non-small cell lung cancer

LUAD:

Lung adenocarcinoma

LUSC:

Lung squamous cell carcinoma

HCC:

Hepatocellular carcinoma

BCa:

Bladder cancer

BC:

Breast cancer

CRC:

Colorectal cancer

ESCC:

Esophageal squamous cell carcinoma

PDAC:

Pancreatic ductal adenocarcinoma

ROC curve:

Receiver operating characteristic curve

OS:

Overall survival

ALP:

Alkaline phosphatase

LDH:

Lactic dehydrogenase

AFP:

Alpha-fetoprotein

CEA:

Carcinoembryonic antigen

Not applicable.

This review was supported by the National Natural Science Foundation of China (No. 82072662), Shanghai Municipal Education Commission—Gaofeng Clinical Medicine Grant Support (2019142), Shanghai Three-year Action Plan to Promote Clinical Skills and Clinical Innovation in Municipal Hospitals (SHDC2020CR4022), and the 2021 Shanghai ‘Rising Stars of Medical Talent’ Youth Development Program: Outstanding Youth Medical Talents.

1. Jun Zhang, Zai Luo, Yang Zheng contributed equally to this work.

Authors and Affiliations

1. Department of Gastrointestinal Surgery, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, 100 Haining Road, Hongkou District, Shanghai, 200080, China

   Jun Zhang, Zai Luo, Yang Zheng, Zhengjun Qiu & Chen Huang

2. Department of Education, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, 650 Xinsongjiang Road, Songjiang District, Shanghai, 201600, China

   Mingyu Duan

Contributions

J.Z. and C.H. proposed concepts; J.Z. and Z.L. drafted the figures and wrote the original draft; YZ and ZQ supervised and provided important advice; J.Z, Z.L, Y.Z, M.D, Z.Q and C.H. reviewed the original draft and gave the final approval of the submitted version.

Corresponding authors

Correspondence to Zai Luo or Chen Huang.

Consent for publication

All authors of this article agree to publish.

Competing interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions