# Chapter 94. Circular RNAs, Back-Splicing, Intronic Circles, Translation Potential, and Functional Evidence

## Scope Note

This chapter explains circular RNAs as covalently closed RNA molecules produced by back-splicing, intron lariat processing, viral replication strategies, or engineering. It focuses on how circular RNAs are made, classified, localized, exported, stabilized, degraded, translated, and tested for function. The chapter treats circRNA claims conservatively because the field contains both well-supported examples and many association-level reports. A proposed circular RNA function is strongest when the circular molecule is directly detected across its back-splice junction, separated from its host linear RNA, perturbed without confounding the host gene, and linked to a reproducible molecular or cellular phenotype.

## Executive Summary

Circular RNAs are RNA molecules in which the 3′ and 5′ ends are covalently joined, so the molecule lacks the free ends that define a conventional linear transcript. Most cellular circRNAs discussed in animals are produced by back-splicing, a spliceosomal reaction in which a downstream splice donor is joined to an upstream splice acceptor. This creates a back-splice junction that is absent from the colinear pre-mRNA and can be used as the defining sequence feature for detection. Back-splicing competes with or accompanies ordinary splicing, and it is influenced by intronic complementarity, RNA-binding proteins, splice-site strength, transcription kinetics, chromatin state, and host-gene architecture.

Circular RNAs are not one biological class. Exonic circRNAs usually contain one or more exons and often accumulate in the cytoplasm. Circular intronic RNAs, or ciRNAs, derive from intron lariats that escape debranching and remain nuclear. Exon-intron circRNAs retain intronic sequence and are commonly discussed as nuclear regulatory molecules. Viral and viroid-like circular RNAs arise from replication and processing strategies that differ from spliceosomal back-splicing. Engineered circRNAs are deliberately circularized for protein production, RNA editing, antigen expression, or other applications. These categories differ in sequence composition, biogenesis, localization, turnover, and evidence standards.

The most reproducible biochemical property of many circRNAs is increased resistance to exonucleases that require RNA ends. Resistance to exonucleases does not mean immortality. circRNAs can be degraded by endonucleolytic cleavage, immune-sensing pathways, miRNA-guided slicing when a nearly perfect target site exists, ribosome-associated quality-control routes, or context-specific decay mechanisms. CircRNA localization and export also vary. Some circRNAs are nuclear because they retain intronic sequence, associate with chromatin, or are newly produced. Many exonic circRNAs are cytoplasmic, and export has been linked to RNA length, helicases, and nuclear export pathways, but a universal export code is not established.

Functional claims about circRNAs require more than differential expression. Some circRNAs bind many copies of a microRNA and can act as competitive endogenous RNAs, with CDR1as/ciRS-7 as the classic high-site example. Most circRNAs do not contain enough abundant, accessible, high-affinity miRNA sites to be assumed to function as miRNA sponges. Some circRNAs bind proteins, alter protein localization or activity, scaffold complexes, or affect transcription near their host loci. Some circRNAs can be translated by cap-independent mechanisms, including internal ribosome entry site-like elements, m6A-associated initiation, engineered initiation cassettes, or rolling-circle translation when stop codons are absent. Translation evidence is strongest when ribosome association is combined with junction-spanning peptide detection, initiation and stop-codon logic, perturbation of the circular RNA itself, and controls for linear contaminants.

The field has a high artifact burden. Reverse transcription can template-switch; repetitive sequences can mis-map; gene duplications and trans-splicing can mimic back-splice junctions; RNase R can enrich circular RNAs but also deplete or spare linear RNAs unevenly; and circRNA knockdown can unintentionally perturb host-gene mRNA. A rigorous circRNA study therefore validates the back-splice junction, demonstrates circularity by multiple orthogonal tests, quantifies the circular isoform separately from host linear transcripts, evaluates RNase R bias, and uses perturbations that distinguish the circular molecule from the pre-mRNA or mRNA that produced it.

## Concept Inventory

- **Circular RNA:**, abbreviated circRNA, is an RNA molecule with a covalently closed continuous backbone. In cellular spliceosomal circRNAs, the defining sequence feature is usually a back-splice junction rather than a free 5′ cap or 3′ poly(A) tail.
- **Back-splicing:** a spliceosomal reaction in which a downstream 5′ splice donor is joined to an upstream 3′ splice acceptor. The resulting junction reverses the colinear order of genomic exons.
- **Back-splice junction:**, abbreviated BSJ, is the sequence that crosses the circularizing splice junction. Junction-spanning reads or PCR products are central evidence for a candidate circRNA, but they must be interpreted with mapping and reverse-transcription controls.
- **Exonic circRNA:** a circRNA composed mainly or entirely of exonic sequence. Many known exonic circRNAs are cytoplasmic, though localization depends on the molecule and cell type.
- **Circular intronic RNA:**, abbreviated ciRNA, is an intron-derived circular RNA produced when a lariat intron escapes normal debranching and trimming. Many ciRNAs are nuclear and can remain near their host gene.
- **Exon-intron circRNA:**, abbreviated EIciRNA, is a circular RNA that contains exonic sequence and retained intronic sequence. EIciRNAs are often discussed as nuclear circRNAs with potential transcriptional regulatory roles.
- **RNase R enrichment:** a common assay in which a 3′-to-5′ exoribonuclease is used to digest many linear RNAs while sparing many circular RNAs. RNase R resistance supports circularity but is not definitive by itself.
- **Divergent primers:** PCR primers oriented away from one another on a linear transcript model but toward each other across a circular junction after reverse transcription. They help amplify candidate BSJs from cDNA.
- **miRNA sponge:** a competitive RNA molecule that binds a microRNA enough to reduce the microRNA's access to other targets. This function requires stoichiometric plausibility, accessible binding sites, and phenotypic rescue evidence.
- **CircRNA translation:** ribosome-mediated protein or peptide synthesis from a circular RNA template. Because circRNAs lack a conventional 5′ cap, translation usually requires cap-independent initiation or engineered design.
- **Rolling-circle translation:** repeated ribosome traversal around a circular RNA when the template lacks an in-frame stop codon or is engineered for continuous translation. The product can be a repeating polypeptide.

## What to Know Before Reading This Chapter

The reader should know the basic architecture of eukaryotic pre-mRNA splicing: exons, introns, 5′ splice donors, 3′ splice acceptors, branch points, lariat intermediates, and spliceosome-mediated exon ligation. Ordinary splicing joins an upstream exon to a downstream exon in genomic order. Back-splicing uses the same general splice-site chemistry but joins the ends in the opposite order, generating a circular exon arrangement.

The reader should also separate three ideas that are often merged. First, circularity is a chemical topology: the RNA backbone is closed. Second, a back-splice junction is a sequence signature: it indicates that a downstream donor has been joined to an upstream acceptor. Third, function is a biological claim: the circRNA must cause a measurable effect that is not explained by its host gene, linear mRNA, or experimental artifact. A molecule can be circular and abundant without having a demonstrated regulatory role.

Finally, circRNA nomenclature is not completely standardized. Many circRNAs are named after the host gene plus an isoform label, but the same host gene can produce multiple circular isoforms. When possible, a circRNA should be specified by species, genome annotation, host gene, exon or intron coordinates, strand, BSJ sequence, and database accession. Host-gene names alone are often too ambiguous for mechanistic claims.

## 94.1. Back-splicing mechanisms and circRNA biogenesis

Back-splicing is the central mechanism for most exonic circRNAs in animals. In a pre-mRNA, a downstream splice donor and an upstream splice acceptor are brought into proximity. The spliceosome then ligates these splice sites in a non-colinear order, producing a circular RNA and leaving the remaining transcript to be processed, degraded, or spliced in other ways. The reaction uses canonical splice-site logic, but the topology is different: the product contains a junction where the downstream exon end precedes the upstream exon start.

![Figure 94.1. Back-Splicing Routes to Exonic circRNAs](../assets/figures/chapter1089_figure1.png)

**Figure 94.1. Back-Splicing Routes to Exonic circRNAs.** Back-splicing uses spliceosomal logic but produces a non-colinear circular product.

Two broad models explain how splice sites are brought close enough to favor back-splicing. In an intron-pairing model, complementary sequences in the flanking introns base-pair and loop the intervening exons into a circularization-competent configuration. Repetitive elements, especially inverted repeat elements in long introns, can support this architecture in primates and other genomes with abundant repeats. In an RNA-binding-protein model, proteins bind motifs in the flanking introns or exons and dimerize, multimerize, or recruit spliceosomal components in a way that favors circularization. These models are not mutually exclusive. A single locus can use both intronic complementarity and protein-mediated bridging, and different circRNA isoforms from the same host gene can be favored by different local features.

Back-splicing is coupled to ordinary gene expression. A gene must be transcribed before a circRNA can be produced, so circRNA abundance often correlates with host-gene transcription. That correlation is not enough to infer a specific circRNA regulatory program. Back-splicing competes with canonical splicing for splice sites in some contexts, but it can also occur from transcripts that would otherwise be unproductive or from intronic architectures that do not strongly reduce mature mRNA production. Transcription elongation rate, spliceosome availability, chromatin marks, and cell-state-specific RNA-binding proteins can shift the balance between linear and circular isoforms.

Alternative circularization creates additional complexity. One host gene can produce many circRNAs through alternative back-splice donors, alternative acceptors, exon skipping, retained introns, or nested circular isoforms. A single short-read BSJ call usually identifies the circular junction but not the full internal exon composition. Long-read sequencing, targeted amplification, and junction-spanning plus internal-read reconstruction can help distinguish isoforms, although each method has its own error profile. For functional experiments, this matters because two circular isoforms from the same host gene may differ in miRNA sites, protein-binding motifs, open reading frames, localization signals, or stability.

Intronic circles follow a related but distinct route. During normal splicing, introns are released as lariat molecules with a 2′-to-5′ linkage at the branch point. Most lariats are debranched and degraded. Some intron lariats evade debranching because sequence motifs near the 5′ splice site and branch point, secondary structure, or protein association stabilize the lariat, after which trimming can produce a circular intronic RNA. These ciRNAs are not back-spliced exon circles; they are intron-derived circles whose circular bond comes from the lariat branch. The distinction matters because ciRNAs often remain nuclear and can be linked to transcriptional regulation near their parent loci.

Biogenesis should not be treated as proof of function. A circRNA can arise because splicing geometry makes back-splicing possible, because the host gene is highly expressed, or because degradation removes competing linear products. The field therefore distinguishes biogenesis claims from functional claims. A biogenesis claim asks how the RNA is made. A functional claim asks whether the circular molecule changes a cellular process, and by what direct mechanism. The evidence standards for those claims differ.

> **Box 94.1. Circularity Is Topology, Not Function**
>
> Circularity is a statement about RNA backbone topology: the molecule lacks free 5′ and 3′ ends. It is not a statement about purpose. A back-splice junction supports the existence of a candidate exonic circle, but it does not by itself define the full isoform, absolute abundance, localization, binding partners, or biological effect. A useful analysis asks three separate questions. First, is the molecule truly circular rather than a mapping, reverse-transcription, or genomic artifact? Second, which circular isoform is present, in which compartment, and at what copy number relative to plausible targets? Third, does selective loss or restoration of the circular molecule change a molecular or cellular phenotype while the host linear RNA remains controlled? Many circRNAs may be stable byproducts of gene expression. The evidentiary burden rises when a study moves from cataloging a circRNA to claiming regulation, disease causality, or therapeutic action.

## 94.2. Exonic, intronic, exon-intron, viral, and engineered circular RNAs

Exonic circRNAs are the best-known class in animal transcriptomes. They are composed mostly of annotated exons and often lack retained introns. Many exonic circRNAs accumulate in the cytoplasm, where they may bind miRNAs, bind proteins, associate with ribosomes, or persist as stable byproducts. The cytoplasmic bias of many exonic circRNAs is useful but not universal. Some exonic circRNAs have nuclear pools, some are cell-type restricted, and some are present at levels far below their host linear mRNAs.

Circular intronic RNAs, or ciRNAs, are generated from intron lariats that resist debranching. Their sequences are intronic, and their biology is often nuclear. The classic model is that selected ciRNAs remain near their sites of transcription and can modulate transcriptional activity of host or neighboring genes. The evidence for a ciRNA should therefore include intronic sequence, lariat-derived circular topology, resistance or sensitivity patterns consistent with the proposed structure, and nuclear localization. A BSJ-style exon-exon junction is not the expected defining feature of a ciRNA.

Exon-intron circRNAs contain both exonic sequence and retained intronic sequence. They are sometimes abbreviated EIciRNAs. Because retained introns can carry nuclear retention features and protein-binding motifs, EIciRNAs are often discussed as nuclear regulatory molecules that interact with transcription machinery or small nuclear ribonucleoproteins. The category is useful, but the same caution applies: a retained intron in a circular RNA does not automatically prove transcriptional regulation. The circular isoform must be detected, localized, perturbed, and linked to a specific molecular output.

**Table 94.1. Major Classes of Circular RNAs.** Circular RNAs share topology but not mechanism, localization, or evidence standards.

| Class | Biogenesis route | Defining sequence feature | Typical localization | Common evidence | Key caveat |
| --- | --- | --- | --- | --- | --- |
| **Exonic circRNA** | Spliceosomal back-splicing joins a downstream donor to an upstream acceptor, often aided by intron pairing or RNA-binding proteins. | Exon-exon back-splice junction with mostly annotated exonic sequence. | Often cytoplasmic in animal cells, with molecule- and cell-type-specific exceptions. | BSJ reads, divergent RT-PCR with Sanger sequencing, RNase R enrichment, northern or long-read support. | A BSJ call does not define full isoform structure or prove function. |
| **ciRNA** | Intron lariat escapes debranching, is trimmed, and remains as an intron-derived circle. | Intronic sequence with lariat-derived circular topology rather than an exon-exon BSJ. | Usually nuclear and often near the host transcription locus. | Intron-specific detection, circularity or lariat-branch support, nuclear fractionation, host-gene transcription assays. | Exonic BSJ criteria are inappropriate; distinguish ciRNAs from intron fragments and unprocessed lariats. |
| **EIciRNA** | Back-splicing preserves exonic sequence while retaining one or more intronic segments. | Exonic BSJ plus retained intronic sequence. | Often nuclear or chromatin-associated in reported examples. | BSJ validation, intron-retention evidence, nuclear localization, interaction or transcription-output assays. | Retained introns do not by themselves prove transcriptional regulation. |
| **Viral or viroid-like circular RNA** | Viral or viroid replication, processing, noncanonical ligation, or host-splicing route depending on pathogen and cell context. | Circular genome, antigenome, replication intermediate, or viral-transcript junction rather than a standard host exon BSJ. | Infection-context dependent; may follow viral replication compartments or host-cell RNA handling. | Strand-specific junction detection, nuclease or topology assays, replication-intermediate evidence, infection controls. | Do not infer spliceosomal back-splicing; cite pathogen-specific replication or host-splicing evidence. |
| **Engineered circRNA** | Designed circularization by permuted intron-exon systems, tRNA splicing, ribozymes, enzymatic ligation, or splint ligation. | Designed circularization junction with payload, initiation element, guide region, or other engineered module. | Defined by delivery and design; commonly intended for cytoplasmic protein expression or RNA-guided activity. | Purity and topology QC, linear-contaminant assays, RNase R or orthogonal circularity tests, protein or activity readout. | Synthetic performance depends on purity, immune sensing, delivery, and design; it does not generalize to endogenous circles. |

Viral and viroid-like circular RNAs broaden the definition beyond host spliceosomal products. Some plant pathogens, viroids, and viral replication intermediates use circular RNA genomes or rolling-circle replication. Certain viral infections can also induce host circRNA changes, and viral genomes or transcripts can produce circular RNA species. These molecules should not be folded into the same mechanism as exon back-splicing unless splice-site and host-processing evidence supports that mechanism. KSHV and other gamma-herpesvirus studies illustrate that viral circRNAs can be detected and regulated in infected cells, whereas viroid and viroid-like RNAs illustrate circular RNA genomes copied through host-dependent replication cycles. Viral circular RNAs may therefore be replication products, processing products, noncanonical ligation products, or host-splicing products depending on the virus and cell type.

Engineered circRNAs are designed molecules rather than transcriptome discoveries. They can be circularized by permuted intron-exon systems, ribozymes, enzymatic ligation, splint ligation, or in vitro transcription workflows followed by circularization and purification. Engineering changes the evidence logic. The question is no longer whether the cell naturally makes the molecule, but whether the synthetic product is truly circular, sufficiently pure, translationally competent or regulatory as designed, and biologically safe in the intended context. Engineered circRNAs are attractive because closed topology can improve exonuclease resistance and protein-expression durability, but immune sensing, purification from linear contaminants, innate immune activation, and delivery remain central design constraints.

Classification should therefore be molecular, not merely historical. A useful circRNA annotation states whether the circle is exonic, intronic, exon-intron, viral, or engineered; whether the circular bond is a splice junction, lariat branch, ribozyme-ligated junction, enzymatic ligation junction, or replication intermediate; and whether the molecule is endogenous or introduced. These distinctions prevent the field from overgeneralizing rules from one class to another.

## 94.3. circRNA stability, localization, export, and turnover

Circular topology changes RNA stability because many RNA decay pathways begin at a 5′ or 3′ end. A closed RNA lacks those free ends, so exonucleases that require an end cannot simply chew through the molecule. This explains why many circRNAs are enriched after RNase R treatment and why some circRNAs accumulate in slowly dividing or postmitotic tissues such as brain. The mechanistic principle is real, but it is often overstated. Circular RNAs are resistant to many exonucleases, not resistant to all decay.

CircRNA stability also depends on sequence, structure, binding proteins, modifications, translation, and cell state. An endonuclease can cut a circular RNA internally and convert it into linear fragments that decay. A miRNA with nearly perfect complementarity can guide Argonaute-mediated slicing, as in highly complementary miRNA-circRNA interactions. Immune pathways can detect certain circRNAs depending on sequence, modification, protein coating, and purity. Translation can expose a circRNA to ribosome-associated quality-control mechanisms. Cellular stress can change protein binding and localization. Stability is therefore a measured half-life in a defined condition, not an intrinsic category label.

Localization is one of the strongest clues to possible mechanism. A nuclear circRNA is better positioned to influence transcription, chromatin, splicing, or nuclear protein availability. A cytoplasmic circRNA is better positioned to affect miRNA activity, cytoplasmic protein complexes, translation, or decay pathways. Exonic circRNAs are often cytoplasmic, whereas ciRNAs and EIciRNAs are often nuclear. This pattern is useful for teaching, but individual examples must be assayed. Fractionation can be contaminated, imaging probes can cross-react with linear host transcripts, and short reads do not directly report subcellular position.

![Figure 94.2. Localization, Export, and Turnover States of circRNAs](../assets/figures/chapter1089_figure2.png)

**Figure 94.2. Localization, Export, and Turnover States of circRNAs.** Circular topology increases exonuclease resistance but does not eliminate regulated turnover.

Export from nucleus to cytoplasm is still an active area. Evidence has connected circRNA export to RNA length, helicase activity, and general mRNA export components, but there is no single export signal that explains all circRNAs. The circular molecule must be recognized as an RNA-protein particle that can leave the nucleus despite lacking a normal cap-binding and poly(A)-binding export architecture. Some circles may piggyback on exon-junction complexes, RNA-binding proteins, or sequence-structure features. Others may remain nuclear because retained introns, chromatin association, or inefficient export keep them near the production site.

Turnover can be selective. A circRNA that is stable under basal conditions may be rapidly degraded after viral infection, innate immune activation, cell-cycle change, or differentiation. Some reports connect circRNAs to endonucleases, RNA modification-dependent decay, or nonsense-mediated decay-like pathways when circRNAs engage translation in ways that resemble aberrant templates. These mechanisms are important because they break the simple model in which circRNA abundance only reflects production plus passive stability. A cell can regulate circRNA levels after biogenesis.

The practical consequence is that circRNA abundance cannot be interpreted without considering production and decay together. High abundance may reflect strong host-gene transcription, efficient back-splicing, slow degradation, slow cell division, or enrichment after experimental treatment. Low abundance may reflect weak production, rapid turnover, poor export, tissue specificity, or detection failure. A mechanistic study should measure host-gene linear RNA, circular RNA, localization, and stability when the proposed function depends on abundance or compartment.

## 94.4. Translation potential, miRNA binding, protein binding, and functions

### Translation Potential and Peptide Products

Circular RNAs lack the canonical 5′ cap used by most eukaryotic mRNAs for translation initiation. They also lack a conventional 3′ poly(A) tail. For that reason, early circRNA discussions often treated circRNAs as noncoding by default. That default is no longer safe. Some endogenous and engineered circRNAs can associate with ribosomes and support peptide or protein synthesis through cap-independent mechanisms. The correct conclusion is not that all circRNAs are translated, but that circular topology does not preclude translation.

Cap-independent translation can be driven by internal ribosome entry site-like elements, short initiation elements, RNA modifications such as N6-methyladenosine in some contexts, or engineered untranslated regions designed for ribosome recruitment. A circular template also changes reading-frame logic. If an open reading frame contains a stop codon, translation can produce a discrete peptide. If a template lacks an in-frame stop codon and permits continuous ribosome transit, rolling-circle translation can produce a repetitive polypeptide. Engineered systems exploit this property for potent protein expression, but endogenous rolling-circle translation claims require stringent controls.

![Figure 94.3. CircRNA Translation Evidence Ladder](../assets/figures/chapter1089_figure3.png)

**Figure 94.3. CircRNA Translation Evidence Ladder.** Ribosome association is weaker evidence than circular-specific peptide and genetic separation evidence.

Evidence for circRNA translation should be layered. Ribosome profiling can show ribosome-protected fragments over a circRNA, but short reads can mis-map to the host linear mRNA and may not cross the back-splice junction. Polysome association suggests translational engagement, but bound RNA can co-sediment without productive elongation. Reporter assays can show that a sequence supports translation, but reporters may create artificial context. Mass spectrometry can detect a peptide, but the strongest peptide evidence is junction-spanning or otherwise unique to the circular template. Mutating the circular RNA's start codon, initiation element, or junction without changing the host gene can test causality.

> **Box 94.2. Four Labels for CircRNA Translation Evidence**
>
> Use four labels when evaluating a circRNA translation claim. **Sequence-compatible** means the circle contains an open reading frame and a plausible cap-independent initiation element, but no translation has been shown. **Ribosome-associated** means ribosome profiling, polysome fractionation, or a related assay places the RNA near translation machinery; mapping to host linear mRNA and nonproductive co-sedimentation remain concerns. **Peptide-supported** means mass spectrometry, epitope tagging, or another protein-level assay detects a product that is unique to the circular template, ideally through a back-splice-junction-spanning peptide or a reading frame unavailable to the linear transcript. **Functionally coding** means the phenotype tracks with peptide production: start-codon, initiation-element, or frame mutations preserve the RNA but remove the peptide effect, and a coding-competent circle or peptide restores the activity. Reporter assays test sequence sufficiency; they do not automatically prove endogenous translation.

The field should distinguish translation potential from demonstrated biological function. A circRNA may be translated at low levels without a physiologically important peptide product. Conversely, a low-abundance peptide could be biologically potent if it acts in a signaling or dominant-negative manner. Functional evidence requires showing that the peptide, not only the RNA, changes a phenotype. Rescue with peptide expression, loss of peptide after start-codon mutation, and separation of RNA-binding effects from peptide-coding effects are especially informative.

Engineered circRNA translation is more advanced as a design problem than many endogenous claims are as biology. Synthetic circRNAs can be purified, modified, sequence-optimized, and delivered to cells. Their translation can be measured directly by encoded protein output. For therapeutic use, however, the same standards that apply to mRNA medicines also matter: purity, innate immune activation, delivery, dose, tissue distribution, persistence, protein quality, and safety. Circular topology may improve duration, but it does not remove the need to understand RNA sensing, translation control, or degradation.

### miRNA Binding, Protein Binding, and Functional Evidence

The most famous circRNA function is miRNA binding. CDR1as, also called ciRS-7, contains many binding sites for miR-7 and became the archetype of a circRNA that can sequester a microRNA. This example is real enough to be pedagogically useful, but it created an overgeneralization problem. Many circRNAs have one or a few predicted miRNA sites. A predicted site is not a sponge. To affect miRNA regulation, a circRNA must be present at sufficient copy number, expose sites in an accessible structure, bind the relevant miRNA-loaded Argonaute complex in the right cell type, and change repression of other targets in a way that can be rescued or phenocopied.

Protein binding may be a broader functional route for many circRNAs. A circRNA can provide a binding platform for RNA-binding proteins, alter protein localization, compete with other RNAs for protein binding, scaffold multiprotein complexes, or modulate enzymatic activity. Protein-binding claims require direct evidence because RNA pull-downs are prone to abundant-protein contamination and indirect association. CLIP-family methods, orthogonal pull-downs, mutational mapping of binding motifs, proximity labeling, and reciprocal perturbations can strengthen the inference.

Some circRNAs may regulate transcription or splicing near their host genes. Nuclear ciRNAs and EIciRNAs have been proposed to interact with transcription machinery or spliceosomal components, supporting local regulation of host-gene expression. These models are plausible because nuclear localization and nascent-transcript proximity create high local concentration. They remain context-specific. If knocking down a circRNA changes host-gene mRNA, the experiment must distinguish loss of the circular molecule from disruption of the pre-mRNA, splice sites, promoter activity, intronic enhancers, or linear isoforms.

**Table 94.2. Functional Claim Standards for circRNAs.** Differential expression or predicted binding is weaker than circular-specific perturbation and rescue.

| Proposed function | Minimal evidence | Strong evidence | Common artifact | Best rescue test |
| --- | --- | --- | --- | --- |
| **miRNA sponge** | Predicted sites plus co-expression with the miRNA and evidence of AGO-associated binding. | Stoichiometric abundance, accessible multi-site binding, target derepression, site mutation, and circular-specific perturbation. | Treating seed matches or disease-associated expression as functional sponging. | Restore phenotype with a knockdown-resistant circle; lose rescue when miRNA sites are mutated without disrupting circularity. |
| **Protein scaffold or decoy** | Pull-down, RIP, CLIP, or proximity signal for a candidate protein in the relevant compartment. | Reciprocal binding evidence, motif mapping, stoichiometry, protein localization or activity change, and host-linear controls. | Abundant-protein contamination, indirect bridges, or recovery of linear host RNA. | Rescue with an intact binding-motif circle; motif-dead circle should remain circular but fail to rescue. |
| **Transcriptional regulator** | Nuclear ciRNA or EIciRNA plus altered nascent or steady-state transcription after circRNA-directed perturbation. | Locus-specific transcription, chromatin, or Pol II readout separated from pre-mRNA, splice-site, and promoter effects. | Genomic deletion or splice-site editing changes the host gene independently of the circRNA. | Restore local transcription with a circular isoform or allele-specific design that avoids changing host linear splicing. |
| **Translated peptide** | ORF and initiation element prediction plus ribosome profiling, polysome association, or reporter activity. | Junction-specific peptide evidence, initiation-site control, circular-specific knockdown, and separation of RNA and peptide effects. | Ribo-seq mis-mapping, artificial reporter context, or translation from linear contaminants. | Peptide expression rescues the coding phenotype; start-codon or initiation-mutant circle preserves RNA but loses peptide function. |
| **Biomarker** | Reproducible disease or state association measured by a BSJ-aware assay with host-linear controls. | Independent cohort, targeted validation, preanalytical robustness, and performance beyond cell-composition or host-gene confounders. | RNase R bias, library-composition shifts, sample handling, or host-gene expression masquerading as circRNA specificity. | Rescue is not applicable; use blinded cohort validation and an orthogonal assay. |
| **Engineered expression platform** | Purified circular product gives protein expression or designed RNA activity after delivery. | Topology and purity QC, linear-contaminant depletion, dose-response, duration, innate-immune readout, and protein-quality evidence. | Linear or nicked RNA contaminants, immune stimulation, or delivery differences explain output. | Compare matched circular, linear, nicked, and initiation-dead constructs; rescue output only with clean active circle. |

Functional perturbation is the weak point in many circRNA studies. Small interfering RNAs or antisense oligonucleotides directed across the BSJ can selectively reduce a circular RNA, but off-target effects, nuclear versus cytoplasmic accessibility, and incomplete depletion must be measured. CRISPR deletion of flanking intronic repeats can reduce circularization, but it also changes genomic regulatory sequence and may alter host-gene splicing. Splice-site mutation can block circularization, but it may change linear mRNA isoforms. Cas13-based approaches can screen circular RNAs more directly, although guide specificity, RNA accessibility, and collateral or indirect effects still require validation.

> **Box 94.3. Perturb the Circle, Not the Host Gene**
>
> Most endogenous circRNAs are embedded in genes that also produce linear pre-mRNAs and mature mRNAs. A perturbation can therefore change the host gene while appearing to change the circle. Back-splice-junction antisense oligonucleotides or small interfering RNAs are conceptually targeted, but the experiment still requires measurements of circular depletion, host linear isoforms, off-target transcripts, and dose-dependent toxicity. Genomic deletion of flanking repeats or splice sites can test circularization requirements, yet the same edit may remove enhancers, alter chromatin, or redirect canonical splicing. A strong design combines circular-specific knockdown with host-linear controls, rescue by a knockdown-resistant circular RNA, and mutation of the proposed miRNA site, protein-binding motif, or start codon while preserving circularization. For nuclear ciRNA or EIciRNA claims, locus-specific and allele-specific readouts are especially important because local transcription effects can be mistaken for direct circRNA function.

The strongest functional evidence uses separation-of-function designs. A study may reduce the circRNA without changing the linear mRNA, restore the circular molecule with an expression construct resistant to the knockdown reagent, mutate the proposed miRNA or protein-binding site while preserving circularity, and show that the phenotype tracks with the circular isoform rather than host-gene expression. For translation claims, the same logic separates RNA function from peptide function. For nuclear regulatory claims, allele-specific or locus-specific experiments can separate local transcription effects from global abundance effects.

CircRNAs are often discussed in cancer, development, infection, muscle, and neuronal biology because many circRNAs are tissue-specific, disease-associated, or stable in clinical samples. Association studies can be valuable biomarkers, but biomarker value is different from mechanistic function. A circRNA can classify tumor state without driving the tumor. A circRNA can change during infection without being a viral defense factor. Disease chapters should therefore label circRNAs as markers, candidate regulators, validated regulators, or therapeutic targets according to evidence level rather than according to differential expression alone.

## 94.5. Detection artifacts, RNase R biases, and validation standards

CircRNA discovery depends heavily on detecting reads that cross a back-splice junction. This is powerful because a BSJ is not expected in a normal colinear transcript. It is also vulnerable to artifacts. Reverse transcriptases can template-switch. PCR can recombine products. Repetitive genomic regions can mis-map. Paralogous genes and pseudogenes can create ambiguous alignments. Trans-splicing, genomic rearrangement, and tandem duplications can mimic non-colinear exon order. Comparative analyses of circRNA callers show that software choice, filtering, simulated ground truth, and tool combination materially change sensitivity and false-positive behavior. A computational circRNA call is therefore a candidate, not final proof.

RNase R enrichment is useful but imperfect. RNase R digests many linear RNAs from a free 3′ end, enriching many circles. However, structured linear RNAs, protein-protected RNAs, some lariat or highly modified RNAs, and incomplete digestion products can resist RNase R. Some circRNAs are sensitive to RNase R depending on structure, nicks, purification conditions, or sequence context. RNase R treatment can also introduce abundance bias and distort comparisons between circRNAs. A molecule enriched after RNase R is more likely to be circular, but RNase R resistance is not sufficient evidence by itself.

![Figure 94.4. circRNA Validation Workflow and Artifact Filters](../assets/figures/chapter1089_figure4.png)

**Figure 94.4. circRNA Validation Workflow and Artifact Filters.** A computational BSJ call is a candidate until orthogonal evidence supports circularity and function.

Validation should use orthogonal assays. Divergent primers can amplify a BSJ from cDNA, but the same primers should not amplify genomic DNA, and the PCR product should be sequenced. Convergent primers can measure host linear transcripts. Northern blotting can show a band consistent with circular size and RNase R resistance, although sensitivity can be limiting. RNase H cleavage with probes that linearize a specific circle can shift mobility. Long-read sequencing can identify full-length circular isoforms, but reverse-transcription-based long reads still require artifact controls. Direct RNA nanopore sequencing avoids reverse transcription but has its own challenges for circular templates, adapters, and base-calling.

Quantification requires attention to denominators. BSJ read counts are not directly comparable to linear exon read counts because the effective junction length, library construction, read length, mapping algorithm, and depletion strategy differ. A circRNA-to-linear ratio can be informative if measured consistently, but it is not a universal molecule count. Absolute copy-number estimates, spike-ins, or targeted assays are better when stoichiometry matters, such as miRNA sponge claims.

The validation standard should scale with the claim. For a catalog of candidate circRNAs, junction reads plus filtering may be acceptable if the dataset is labeled as discovery. For a chapter-level mechanistic claim, the circular isoform should be validated by BSJ sequencing, RNase R or another circularity assay, host-linear controls, isoform reconstruction when relevant, localization evidence when mechanism requires compartment specificity, and perturbation-rescue evidence when function is claimed. For therapeutic or engineered circRNAs, purity from linear precursors and nicked products is a safety and efficacy issue, not merely a technical detail.

Direct method comparisons make this validation standard practical rather than decorative. The 2016 comparison of circRNA prediction tools established that callers disagree and that simulated or spike-in-like validation cannot be replaced by read-count thresholds alone. A 2026 comparative analysis of circular-DNA and circRNA tools reached the same operational lesson for tool combinations: agreement can improve confidence, but it does not replace orthogonal RNA evidence. Detection-method claims in this chapter therefore use caller comparisons, RNase R caveats, BSJ sequencing, host-linear controls, and topology assays as separate evidentiary layers rather than treating any single assay as decisive.

## Recent Consensus

The current consensus is that circRNAs are a real and diverse part of many transcriptomes, especially in animals, and that back-splicing is a major route for endogenous exonic circRNA production. The field also agrees that circular topology can increase exonuclease resistance and that some circRNAs have validated molecular functions. Translation of circular RNAs is now a legitimate topic rather than an anomaly, especially for engineered circRNAs and selected endogenous candidates.

The consensus is equally clear about caution. Most annotated circRNAs do not yet have demonstrated functions. Many disease-associated circRNAs remain biomarkers or candidates. Many miRNA sponge claims are under-supported because they lack stoichiometric evidence. Many translation claims require stronger peptide-level validation. Detection artifacts and perturbation confounders are common enough that validation standards should be explicit in every circRNA study.

## Open Questions, Controversies, Deprecated Models, and Common Misconceptions

Open questions:

- How many circRNAs are selected functional molecules rather than tolerated byproducts of splicing? The answer is unlikely to be one number for all organisms or tissues. Neurons, muscle, immune cells, cancers, and engineered systems may have different balances of production, stability, function, and noise.
- What determines how circRNA export and turnover are specified? Known examples implicate helicases, length, structure, protein binding, immune sensing, endonucleolytic cleavage, and translation-associated decay, but no single rule explains all circRNAs. A useful future model will probably integrate biogenesis, RNP assembly, localization, and degradation rather than assigning each circRNA to one isolated pathway.

Controversies:

- A third controversy concerns translation. Some circular RNAs clearly can be translated, but ribosome profiling alone is not enough to prove productive translation from an endogenous circRNA. The field still has too few examples with unique peptide detection, genetic separation of RNA and peptide functions, and physiological relevance.

Common misconceptions:

- "A BSJ read is not automatically a validated circRNA." RNase R resistance is not definitive proof of circularity. A predicted miRNA site is not a sponge. Differential expression in cancer is not evidence of driver function. A host-gene phenotype is not automatically a circRNA phenotype. Circular topology increases resistance to some decay enzymes but does not make an RNA permanent.