Circular RNAs in tumor biology and immunology: molecular mechanisms and therapeutic implications

Abstract

Cancer immunotherapy has revolutionized oncology, yet its efficacy remains frequently constrained by the immunosuppressive tumor microenvironment (TME). Circular RNAs (circRNAs) are covalently closed RNA molecules generated through back-splicing. Their lack of free ends confers strong resistance to exonuclease degradation and high molecular stability. Within tumor cells, circRNAs govern multiple cancer hallmarks via diverse regulatory mechanisms. Beyond these cell-intrinsic functions, circRNAs can be secreted as intercellular messengers that enter and reshape the TME by modulating antitumor immunity. Clinically, the high stability of circRNAs in body fluids highlights their potential as non-invasive liquid biopsy biomarkers for cancer diagnosis and prognosis. Furthermore, beyond targeting endogenous circRNAs, engineered circRNAs encoding tumor antigens or immunomodulatory proteins represent a promising platform for cancer immunotherapy. In this article, we briefly summarize recent advances in the understanding of circRNAs in tumor biology and immunology.

Introduction

Cancer remains a global health challenge, characterized by a complex etiology and high mortality rates [1]. Traditional therapies often face limitations due to tumor heterogeneity, metastasis, and the development of resistance. In this context, cancer immunotherapy has emerged as a paradigm-shifting approach that aims to harness the host immune system to recognize and eliminate tumor cells [2–4]. However, the efficacy of immunotherapy is frequently constrained by the immunosuppressive TME [5, 6].

The TME is recognized not merely as a passive stroma but as a highly integrated, multicellular ecosystem in which antitumor immunity acts as both a critical component and a key regulator [7]. Effective antitumor immune response requires close cooperation and extensive crosstalk between the innate and adaptive compartments [8]. The innate compartment, comprising primarily dendritic cells (DCs), macrophages, natural killer (NK) cells, and myeloid-derived suppressor cells (MDSCs), acts as the frontline immune sensors and gatekeepers. Meanwhile, the adaptive compartment, driven mainly by CD8⁺ cytotoxic T cells and CD4⁺ helper T cells, mediates targeted and durable tumor clearance. Successful tumor control depends on a series of systemic antitumor immune events, including efficient release and presentation of tumor antigens, robust innate immune sensing, and effective activation and infiltration of effector cells [9]. Tumors actively dismantle these antitumor immune responses. By subverting antigen presentation, enforcing effector cell exhaustion, or orchestrating the recruitment of immunosuppressive populations, tumors establish a tolerogenic niche that fuels malignant progression and dictates therapy resistance [10]. Elucidating the molecular mediators that sustain the integrity of this immune ecosystem is therefore essential for developing next-generation cancer therapies.

Cancer progression and immune evasion are complex processes driven by a dynamic interplay of genetic, transcriptomic, and epigenetic alterations [11–13]. Within this intricate landscape, circular RNAs (circRNAs) have emerged as key transcriptional and post-transcriptional regulators. Unlike linear RNAs, circRNAs are generated via back-splicing, during which a downstream 5’ splice site covalently links to an upstream 3’ splice site [14]. This produces a closed loop lacking a 5’ cap and a 3’ polyadenylated tail, enabling circRNAs to resist degradation by RNA exonucleases (e.g., RNase R) and conferring substantially longer half-lives than their linear counterparts [15]. Compelling evidence demonstrates that circRNAs intrinsically govern multiple cancer hallmarks within tumor cells, including proliferation, apoptosis, metastasis, metabolism, angiogenesis, and genomic instability [16]. Beyond these cell-autonomous roles, emerging evidence highlights that circRNAs also extrinsically reshape the TME by modulating antitumor immunity. Notably, both circRNAs and their encoded cryptic peptides can be secreted into the extracellular space, where they act as potent intercellular messengers to regulate surrounding immune cells.

By orchestrating the complex crosstalk between tumor cells and surrounding immune populations, circRNAs have emerged as pivotal gatekeepers of the tumor-extrinsic immune ecosystem. In this review, we first summarize the biogenesis, degradation, distribution, and biological functions of circRNAs. We then systematically delineate the emerging paradigm of circRNAs as critical regulators of antitumor immunity, mapping their dynamic involvement across distinct stages of the antitumor immune responses, including antigen presentation, innate immune sensing, effector cell activation, and immune evasion. Furthermore, we discuss the profound clinical translational potential of circRNAs, highlighting their utility as highly stable, non-invasive liquid biopsy biomarkers and their promise in the development of next-generation targeted immunotherapeutics.

Overview of circRNAs in gene expression

Biogenesis of circRNAs

Based on their specific genomic origins and sequence composition, circRNAs can be classified into three major categories: exonic circRNAs (EcRNAs), exon-intron circRNAs (EIciRNAs), and intron-derived circRNAs [17–19]. Intron-derived circRNAs are further subdivided into circular intronic RNAs (ciRNAs), derived from pre-mRNAs, and tRNA intronic circular RNAs (tricRNAs), derived from pre-tRNAs [20, 21].

The back-splicing of circRNAs is elaborately regulated by the interplay between cis-acting elements and trans-acting factors [22]. Complementary sequences, such as Alu repeats located in flanking introns, facilitate back-splicing by base-pairing to form double-stranded RNA (dsRNA) hairpin loops (Fig. 1A) [23]. Meanwhile, RNA-binding proteins (RBPs) can physically interplay with these intronic structures to either stabilize or dismantle them. For instance, RBPs such as Muscleblind (MBL) and Quaking (QKI) specifically recognize and bind to motifs within flanking introns. Through protein-protein interaction, they act as molecular bridges that further pull the splice sites together and strongly facilitate circularization (Fig. 1A) [24, 25]. Conversely, this interplay can be highly antagonistic, as certain suppressive RBPs directly disrupt base-pairing to prevent loop formation. For instance, ADAR1 introduces adenosine-to-inosine (A-to-I) editing to chemically abolish sequence complementarity [26], whereas RNA helicases like DHX9 unwind double-stranded RNA stems [27]. Both mechanisms halt circRNA production.

Fig. 1.

Biogenesis of different types of circRNAs. A The back-splicing of EcRNAs and ElciRNAs is regulated by the interplay between cis-acting elements and trans-acting factors, with RBPs facilitating/disrupting circularization via dimerization. B Lariat-mediated biogenesis of circRNAs forms ElciRNAs and ciRNAs through exon skipping and intron removal. C Biogenesis of tricRNAs is processed through splicing of pre-tRNA

Certain circRNAs are instead generated through lariat-driven circularization or during the maturation of pre-tRNAs. A process termed exon skipping gives rise to EIciRNAs (Fig. 1B), whereas completely intronic circRNAs (ciRNAs) can originate from the lariat-driven intron removal process (Fig. 1B) [19, 28]. In addition, a minor subset, termed tricRNAs, originate from intron-containing pre-tRNAs (Fig. 1C). During pre-tRNA maturation, the tRNA splicing endonuclease (TSEN) complex cleaves the intron at a canonical bulge-helix-bulge (BHB) motif, and the released intron termini are ligated by RtcB ligase to form a stable circular RNA [21].

Degradation of circRNAs

To maintain proper cellular homeostasis, circRNA levels are strictly governed by a highly dynamic balance between their biogenesis and turnover. However, the full spectrum of degradation mechanisms and their pathophysiological implications remain to be further elucidated [29].

Because of their resistance to classical exonucleases, the degradation of circRNAs is initiated by specific endonucleases that open the loop before subsequent exonucleolytic clearance. Recent advances have unraveled several distinct endonucleolytic decay pathways. The miRNA-loaded Argonaute 2 (AGO2) complex can cleave highly complementary circRNAs (e.g., miR-671 directing the cleavage of CDR1as) (Fig. 2A) [30]. Additionally, N⁶-methyladenosine (m⁶A) modifications deposited on circRNAs are recognized by the YTHDF2 reader protein, which recruits the RNase P/MRP complex to trigger circRNA degradation (Fig. 2B) [31].

Fig. 2.

Different degradation mechanisms of circRNAs. A The miRNA-loaded AGO2 complex cleaves highly complementary circRNAs via endonucleolytic activity. B YTHDF2 recognizes m⁶A-modified circRNAs and recruits the RNase P/MRP complex to initiate degradation. C The endoribonuclease DIS3 mediates degradation of a broad spectrum of circRNAs. D Highly structured circRNAs are degraded by the UPF1-G3BP1 complex via the SRD pathway. E mecciRNAs are turned over within mitochondria by the SUPV3L1-ELAC2 complex. F Under cellular stress, RNase L is activated to globally degrade circRNAs. G During aging, RNase K (RNK-1/RNASEK) collaborates with HSP90 in stress granules to clear toxic circRNAs

Beyond these pathways, a recent study identifies the DIS3-mediated degradation pathway (Fig. 2C) [32]. DIS3 is an endoribonuclease that associates with the nuclear RNA exosome and is also partially detected in the cytoplasm, where it mediates the degradation of circRNAs. DIS3 depletion leads to upregulation of over 60% of circRNAs, with minimal effects on their linear cognate transcripts, indicating a more general and specialized role for DIS3 in circRNA degradation. Furthermore, a novel structure-mediated RNA decay (SRD) pathway has been identified, wherein highly structured circRNAs are selectively unwound and degraded by the UPF1-G3BP1 complex (Fig. 2D) [33].

Recently identified mitochondrial genome-derived circRNAs, termed mitochondria-encoded circRNAs (mecciRNAs) [34], differ from nuclear genome-encoded circRNAs in their degradation mechanism. Nuclear genome-encoded circRNAs are predominantly degraded by cytoplasmic machinery, whereas mecciRNAs are rapidly turned over within mitochondria by a dedicated, evolutionarily conserved complex composed of the RNA helicase SUPV3L1 and the endoribonuclease ELAC2 (Fig. 2E) [35]. In heart failure, accelerated mecciRNA degradation disrupts the regulation of the mitochondrial permeability transition pore (mPTP) and mitochondrial reactive oxygen species (ROS) homeostasis, thereby driving cardiomyocyte damage.

Under certain stress conditions, such as viral infection and aging, specialized circRNA degradation pathways are engaged to rapidly remodel the circRNA landscape, thereby preventing pathological progression. Upon viral infection, the innate immune sensor oligoadenylate synthetase (OAS) is activated by pathogenic dsRNA, which in turn activates the latent endoribonuclease RNase L [36]. Once active, RNase L globally and rapidly degrades circRNAs (Fig. 2F), a process functionally essential for protein kinase R (PKR) activation in early innate immune responses. Dysregulation of this pathway is linked to autoimmune diseases such as systemic lupus erythematosus (SLE). During aging, the endoribonuclease RNase K (RNK-1/RNASEK) collaborates with HSP90 in stress granules to cleave and prevent toxic accumulation of circRNAs (Fig. 2G) [37]. Aberrant loss of this clearance pathway may result in stress granule persistence and accelerate aging across species.

Subcellular distribution and biological functions of circRNAs

While circRNA biogenesis occurs in the nucleus, the vast majority of these transcripts localize to the cytoplasm [38]. The distinct subcellular distributions of circRNAs imply their involvement in regulating gene expression at multiple levels.

Nuclear functions

In the nucleus, circRNAs primarily regulate gene expression by interacting with DNA and the transcription machinery. Certain circRNAs can hybridize with their cognate genomic DNA loci to form RNA-DNA hybrids, termed R-loops, thereby modulating alternative splicing (Fig. 3A) or RNA polymerase II (Pol II)-mediated transcription (Fig. 3B) of their parental transcripts. For instance, in Arabidopsis thaliana, a circRNA derived from exon 6 of the SEPALLATA3 gene forms an R-loop at its cognate DNA locus [39]. This R-loop formation promotes the recruitment of splicing factors, thereby facilitating the production of the exon 6-skipped alternative splicing variant SEP3.3 and driving floral homeotic phenotypes. Similarly, in human cells, circPOLR2B forms an R-loop with its parental gene POLR2B, blocking Pol II activity to negatively regulate its transcription [40]. Furthermore, certain circRNAs can modulate gene expressions by interacting with small nuclear ribonucleoprotein U1 (U1 snRNP), which further recruits Pol II to the promoters of their parental genes (Fig. 3B). For example, circEIF3J and circPAIP2 predominantly localize to the nucleus and interact with U1 snRNP [41]. Knockdown of circEIF3J or circPAIP2 significantly reduces transcription of their corresponding parental genes.

Fig. 3.

Distribution and biological functions of circRNAs. A In the nucleus, circRNAs form R-loops with their cognate genomic DNA to modulate alternative splicing of their parental transcripts. B Nuclear circRNAs regulate Pol II-mediated transcription either by forming R-loops that blocks elongation or by interacting with U1 snRNP to recruit Pol II to gene promoters. C In the cytoplasm, circRNAs act as miRNA sponges by sequestering miRNAs, thereby relieving the repression of target mRNAs. D Cytoplasmic circRNAs serve as protein scaffolds or decoys to modulate protein stability, mRNA stability, and translation. They also drive liquid-liquid phase separation and localize to membrane-bound organelles, such as mitochondria and the ER, to execute specialized functions. E A subset of circRNAs harboring IRES or m⁶A modifications can be translated into functional peptides that act as protein decoys or immunogenic neoantigens

Cytoplasmic functions

In the cytoplasm, circRNAs primarily modulate gene expression by interacting with RNA and protein. A well-characterized RNA-mediated function of circRNAs is acting as competitive endogenous RNAs (ceRNAs), or miRNA sponges (Fig. 3C). A classic example is CDR1as, which sequesters miR-7 to alleviate the repression of its target genes [42, 43]. Additionally, circRNAs regulate post-transcriptional networks by serving as protein scaffolds or decoys, thereby modulating protein stability, mRNA stability, and the translational output of mRNA (Fig. 3D).

At the protein level, circRNAs can modulate the stability of proteins themselves through scaffold-assisted ubiquitination or decoy-mediated protection from degradation (Fig. 3D, left). For instance, circFOXO3 forms a ternary complex with p53 and MDM2, which enhances MDM2-mediated ubiquitination of p53 while simultaneously protecting FOXO3 from MDM2-induced degradation [44]. CircNEIL3 binds the oncoprotein YBX1 and promotes its Nedd4L-mediated proteasomal degradation [45]. Moreover, circLNF acts as a protein decoy that interacts with IGF2BP family members, shielding them from TRIM25-mediated ubiquitination and proteasomal degradation, thus sustaining oncogenic IGF2BP activity [46].

At the RNA level, circRNAs modulate the stability and translation of target mRNAs primarily through interactions with RBPs (Fig. 3D, right). TGFβ-induced circITGB6 functions as a scaffold that directly binds IGF2BP3, reinforcing the stabilization of PDPN mRNA, which drives downstream epithelial-mesenchymal transition (EMT) and colorectal cancer (CRC) metastasis [47]. Furthermore, circular RNA BISC (hsa_circ_0001573) selectively recognizes the KH3/KH4 domains of IGF2BP2, thereby blocking IGF2BP2 from binding to c-Myc mRNA and suppressing its translation. This overcomes bromodomain-and-extraterminal-domain (BET) inhibitor resistance in triple-negative breast cancer (TNBC) [48].

Compartment-specific functions

Importantly, the cytoplasmic environment is not homogeneous but is partitioned into functionally distinct subcellular compartments, such as membraneless organelles formed by liquid-liquid phase separation (LLPS) that concentrate specific biomolecules and drive efficient biochemical reactions. Within LLPS-driven condensates, circRNAs serve as architectural scaffolds that promote condensate formation and functionally localize within them to orchestrate rapid stress responses. For instance, circVAMP3 directly binds CAPRIN1 and acts as a scaffold that drives CAPRIN1-G3BP1 phase separation to promote the formation of cytoplasmic stress granules [49]. CircVAMP3 is itself enriched within these stress granules, where it functions to suppress c-Myc translation and inhibit hepatocellular carcinoma (HCC) progression.

CircRNAs can also dynamically localize to membrane-bound organelles such as mitochondria and the endoplasmic reticulum (ER) to maintain specialized metabolic and signaling microenvironments. Within mitochondria, both nuclear genome-derived circRNAs and mecciRNAs have been identified. For example, circPUM1, derived from back-splicing of the nuclear genome PUM1 pre-mRNA, localizes to mitochondria and acts as a protein scaffold to enhance oxidative phosphorylation. Its depletion leads to pyroptosis in esophageal squamous cell carcinoma (ESCC) cells [50]. MecciRNAs such as mcPGK1 regulate metabolic reprogramming by inhibiting mitochondrial oxidative phosphorylation and promoting glycolysis, thereby driving liver tumor-initiating cell self-renewal [51]. CircFAM13B localizes to the ER and acts as a tumor suppressor in nasopharyngeal carcinoma by binding to RBM3 and promoting the downregulation of key ER stress molecules such as XBP1s, thereby inhibiting lymphangiogenesis and lymphatic metastasis [52]. Beyond these findings, the additional distribution and function of circRNAs in other cytoplasmic compartments, such as processing bodies (P-bodies), require further elucidation.

CircRNA-encoded peptides

Remarkably, bypassing classic cap-dependent translation, a subset of cytoplasmic circRNAs harboring internal ribosome entry sites (IRES) or m⁶A modifications can be directly translated into functional peptides, adding a novel dimension to the known proteome (Fig. 3E) [53, 54]. These circRNA-derived cryptic peptides mainly act as protein decoys and immunogenic antigens. For instance, circ-FBXW7 encodes a small peptide that competitively binds to the deubiquitinating enzyme USP28, thereby accelerating the degradation of the c-Myc oncoprotein [55]. Beyond modulating intracellular signaling, some circRNA-translated peptides can be presented on the cell surface as cryptic neoantigens. CircFAM53B translates into specific immunogenic peptides that elicit robust antitumor T-cell responses [56]. Nevertheless, the global prevalence, regulatory mechanisms, and pathophysiological roles of circRNA-derived peptides remain elusive, warranting further investigation.

Regulation of circRNAs in tumor biology

The precise equilibrium between circRNA biogenesis and degradation can be disrupted by pathological stimuli, including aberrant splicing factor expression, chromosomal translocations, or dysfunctional decay machinery. Such dysregulation contributes to a broad spectrum of human diseases, ranging from neurodegenerative pathologies and cardiovascular disorders to cancer [57]. Emerging evidence indicates that aberrantly expressed circRNAs promote tumor progression. By modulating essential intracellular signaling networks, these dysregulated circRNAs align closely with multiple cancer hallmarks, including proliferation, cell death, invasion and metastasis, metabolic reprogramming, angiogenesis, and genome instability.

Regulation of circRNAs in tumor cell proliferation

Uncontrolled cell proliferation is a fundamental trait of tumor cells, driven by the dysfunction of interconnected oncogenic regulators [58]. At the core of this regulatory network, cyclin-dependent kinases (CDKs) govern cell-cycle progression and are frequent targets of circRNA-mediated dysregulation in cancer (Fig. 4A).

Fig. 4.

Regulation of tumor cell proliferation and death by circRNAs. A CircRNAs modulate tumor cell proliferation through multiple mechanisms. These include acting as protein scaffolds to form inhibitory complexes with CDKs and cell-cycle inhibitors, as well as functioning as miRNA sponges or protein scaffolds to upregulate CDKs, c-Myc, or upstream growth factor receptors (e.g., EGFR), thereby driving cell-cycle progression. B CircRNAs intercept diverse forms of RCD, including apoptosis, pyroptosis, ferroptosis, cuproptosis, and autophagy-dependent cell death, through diverse biological functions

CircFOXO3, a tumor-suppressive circRNA frequently downregulated in tumor cells, directly binds to both CDK2 and the cell cycle inhibitor p21 to form a ternary complex. By preventing CDK2 from interacting with cyclin A and cyclin E, this scaffolding interaction arrests the cell cycle at the G1/S transition and suppresses tumor growth [59]. In contrast, certain oncogenic circRNAs promote tumor cell proliferation by upregulating CDK activity. In bladder cancer, hsa_circ_0005320 interacts with and upregulates IGF2BP3, which in turn binds to and stabilizes CDK2 mRNA, enhancing CDK2 protein expression and accelerating cell-cycle progression [60]. CircHIPK3 is aberrantly overexpressed in gallbladder cancer and directly sponges miR-124 to upregulate its downstream target CDK6, thereby driving cell-cycle progression [61].

c-Myc, a key transcription factor that drives the expression of cyclins and CDKs while suppressing CDK inhibitors, represents another major node through which circRNAs modulate tumor proliferation. Hsa_circ_0000467 acts as a molecular scaffold that simultaneously binds the translation initiation factor eIF4A3 and c-Myc mRNA, forming a ternary complex that enhances c-Myc protein translation and promotes proliferation in CRC [62]. Moreover, circ-FBXW7 encodes a small peptide, FBXW7-185aa, that competitively binds to the deubiquitinating enzyme USP28, thereby preventing USP28 from stabilizing c-Myc and accelerating c-Myc degradation [55]. Circ-FBXW7 is highly expressed in the normal brain but significantly downregulated in glioblastoma, where its expression positively correlates with patient overall survival. Upregulation of its encoded peptide FBXW7-185aa inhibits cancer cell proliferation and cell cycle progression.

CircRNAs also sustain proliferative signaling by targeting upstream growth factor receptor pathways, such as epidermal growth factor receptor (EGFR). For instance, CDR1as is overexpressed across multiple cancer types and promotes EGFR expression by sequestering miR-7 in CRC and by sponging miR-1299 in ESCC, thereby sustaining continuous mitogenic signaling and facilitating tumor proliferation [63, 64].

Regulation of circRNAs in tumor cell death

Cell death is broadly classified into accidental cell death (ACD) and regulated cell death (RCD), the latter of which encompasses multiple molecularly defined subroutines, including but not limited to apoptosis, pyroptosis, ferroptosis, cuproptosis, and autophagy-dependent cell death [65]. Cancer cells must evade multiple forms of RCD to sustain survival and malignant progression [66]. Each of these death modalities is governed by distinct molecular machinery, and circRNAs have emerged as pleiotropic regulators capable of intercepting these cell death pathways (Fig. 4B).

Apoptosis, the most extensively characterized form of RCD, is principally executed through the BCL2-regulated mitochondrial pathway and survival signaling cascades such as the PI3K/AKT axis [67]. CircRNAs can drive tumor cell evasion of apoptosis by modulating these pathways. For example, hsa_circ_0014130, which is significantly upregulated in non-small cell lung cancer (NSCLC), functions as a competitive sponge for miR-136-5p, derepressing BCL2 and tipping the balance toward apoptotic evasion [68]. Similarly, circ-BIRC6 is overexpressed in HCC and sequesters miR-3918, a known inhibitor of BCL2, resulting in BCL2 upregulation that suppresses HCC cell apoptosis [69]. Additionally, circAKT3 acts as a miRNA sponge that alleviates miR-198-mediated suppression of PIK3R1, thereby hyperactivating PI3K/AKT signaling to blunt pro-apoptotic effectors and drive apoptosis evasion [70].

Pyroptosis is a lytic and pro-inflammatory form of RCD mediated by gasdermin family members, such as GSDME and GSDMD, whose pore-forming activity is triggered by NLRP3 inflammasome-dependent and -independent pathways [71]. In this arena, circPDIA3 directly binds the GSDME-C domain to block pyroptotic pore formation, thereby restraining pyroptosis and inducing chemoresistance in CRC [72]. Conversely, the circMAP3K13-encoded novel protein, MAP3K13-232aa, binds IKKα to activate NF-κB signaling and upregulate NLRP3, thereby inhibiting cisplatin-induced pyroptosis in gastric cancer (GC) [73].

Ferroptosis is an iron-dependent, lipid peroxidation-driven form of RCD that critically shapes tumor evolution and therapy resistance [74]. Tumor cells engage multiple defense systems to survive this RCD, primarily through the classic solute carrier family 7 member 11 (SLC7A11) and glutathione peroxidase 4 (GPX4) antioxidant axis, iron storage proteins such as ferritin heavy chain 1 (FTH1), and GPX4-independent mechanism such as the GTP cyclohydrolase 1 (GCH1) pathway. Tumors strategically deploy circRNA networks to intercept specific ferroptotic checkpoints. For instance, circPIAS1 is upregulated in HCC to activate the NUPR1/FTH1 axis by sponging miR-455-3p, which limits intracellular labile iron availability and inhibits tumor cell ferroptosis [75]. Conversely, specific circRNAs function as tumor suppressors. CircLRFN5, downregulated in glioblastoma and associated with poor patient prognosis, restrains malignant progression by modulating the PRRX2/GCH1 signaling cascade, triggering ferroptosis via GPX4-independent mechanisms [76].

Cuproptosis is a recently identified copper-triggered form of RCD. Mechanistically, excess intracellular copper is reduced from Cu²⁺ to its more toxic Cu⁺ form by the mitochondrial ferredoxin FDX1, a key upstream regulator of this pathway [77]. The accumulated Cu⁺ then binds to lipoylated tricarboxylic acid (TCA) cycle enzymes, inducing their aberrant oligomerization and subsequent loss of iron-sulfur cluster proteins, which ultimately triggers proteotoxic stress and cell death. CircRNAs can modulate this pathway at multiple nodes. For instance, p53-induced circFRMD4A binds and inactivates pyruvate kinase M2 (PKM2), thereby reducing glycolytic flux and rerouting carbon metabolites toward the TCA cycle. This promotes mitochondrial respiration and sensitizes cancer cells to elesclomol-induced cuproptosis [78]. Conversely, circKIAA1797 inhibits cuproptosis by binding to FDX1 mRNA and reducing its stability, thereby suppressing FDX1 expression and downstream lipoylation-dependent protein aggregation to drive lung cancer progression [79].

Autophagy is a lysosome-mediated degradation process whose homeostatic balance between metabolic recycling and lethal self-digestion critically determines tumor cell fate [80, 81]. In tumor biology, circRNAs can modulate autophagy-dependent cell death (ADCD) in a context-dependent manner, primarily through key signaling intermediates such as GLUD1. CircTICRR, which is aberrantly upregulated in cervical cancer, directly binds the RBP HuR, thereby stabilizing GLUD1 mRNA [82]. The accumulation of GLUD1 protein restrains autophagic flux and suppresses tumor cell ADCD. Conversely, circBCL2, derived from its host gene BCL2, acts as a tumor suppressor by sponging miRNA-198 to promote AMBRA1 mRNA expression, which induces ADCD in castration-resistant prostate cancer (CRPC) [83].

Regulation of circRNAs in tumor cell invasion and metastasis

The invasion and metastatic cascade of tumor cells is frequently initiated by EMT, a process where epithelial cells lose adhesion and acquire migratory capabilities. EMT is driven principally by two upstream signaling pathways, TGF-β and Wnt/β-catenin, whose activation induces downstream EMT-inducing transcription factors (EMT-TFs), including SNAIL, SLUG, TWIST1, and ZEB1/2. Emerging evidence indicates that circRNAs serve as master regulators of EMT by strategically intercepting its signaling and transcriptional networks [84].

TGF-β pathway

The TGF-β pathway is a principal inducer of EMT, operating through ligand-mediated activation of TGF-β receptor complexes that phosphorylate downstream SMAD transcription factors to drive EMT gene expression. CircRNAs can potentiate this pathway by acting as upstream signal amplifiers. CircTGFBR2 [3–6], a circRNA derived from the TGF-β receptor 2 pre-mRNA (TGFBR2), functions as a molecular scaffold that simultaneously binds both the RBP IGF2BP3 and TGFBR1 mRNA, thereby enhancing TGFBR1 mRNA stability and hyperactivating the TGF-β/SMAD signaling cascade to drive breast cancer EMT and metastasis [85].

Following activation of the TGF-β signaling pathway, circITGB6 can be induced and promotes the EMT process. Its expression is significantly upregulated in metastatic cancer samples and is tightly linked to poor prognosis in CRC patients. Mechanistically, circITGB6 acts as a key TGF-β effector molecule that directly interacts with IGF2BP3 and enhances the mRNA stability of the EMT-promoting gene PDPN, thereby robustly driving the EMT program and facilitating tumor cell invasion [47]. Conversely, circNEIL3 was identified as a TGF-β-repressed and metastasis-suppressive circRNA. It is significantly downregulated in metastatic tumor lesions and functions as a potent metastasis suppressor. CircNEIL3 directly bind the oncoprotein YBX1 and scaffold the E3 ubiquitin ligase Nedd4L to form a ternary complex, which facilitates the proteasomal degradation of YBX1 and inhibits tumor metastasis [45].

Wnt/β-catenin pathway

The Wnt/β-catenin pathway represents another major EMT-driving axis. Upon Wnt ligand binding, β-catenin escapes degradation, translocates to the nucleus, and cooperates with TCF/LEF transcription factors to activate pro-EMT gene programs. CircRNAs intercept this pathway through both miRNA-sponging and protein-scaffolding mechanisms. For instance, circNIPBL directly sponges miR-16-2-3p to upregulate Wnt5a, a non-canonical Wnt ligand that nevertheless activates β-catenin signaling. This activates ZEB1, a downstream EMT-TF that positively promotes further circNIPBL biogenesis and thereby sustaining bladder cancer metastasis [86]. Moreover, circ-CTNNB1 binds the RNA helicase DDX3 to facilitate the transactivation of YY1, which in turn promotes β-catenin signaling and drives tumor metastasis [87].

Regulation of circRNAs in tumor cell metabolism

Dysregulated circRNAs are increasingly recognized as metabolic modulators that fundamentally rewire tumor cell energy metabolism. Tumor metabolic reprogramming encompasses multiple interconnected dimensions, such as enhanced aerobic glycolysis (the Warburg effect), dysregulated lipid metabolism, reprogrammed amino acid metabolism, and altered TCA cycle and oxidative phosphorylation (OXPHOS) [88]. Each dimension is governed by distinct metabolic sensors and regulated by circRNAs.

Aerobic glycolysis, the fundamental metabolic hallmark of tumor cells characterized by accelerated glucose uptake and lactate production even in the presence of oxygen, is driven by AMP-activated protein kinase (AMPK) and glycolytic enzymes such as hexokinase 2 (HK2), PKM2, α-enolase (ENO1) and aldolase A (ALDOA). For instance, circPETH, a circRNA transported via extracellular vesicles from tumor-associated macrophages (TAMs) to HCC cells, encodes a peptide called circPETH-147aa [89]. CircPETH-147aa promotes PKM2-catalyzed phosphorylation of ALDOA, thereby enhancing ALDOA enzymatic activity and accelerating glycolysis and metastasis in recipient HCC cells. Moreover, circUBE2G1 encodes a small peptide, circUBE2G1-99aa, that directly binds to the glycolytic enzyme ENO1, inhibiting the ENO1/PI3K/AKT signaling axis and thereby suppressing GC proliferation [90].

Lipid metabolism reprogramming in cancer involves enhanced de novo fatty acid synthesis and dysregulated cholesterol homeostasis. Fatty acid synthesis is primarily driven by acetyl-CoA carboxylase 1 (ACC1), fatty acid synthase (FASN) and ATP-citrate lyase (ACLY). CircACC1, derived from ACC1, is transcriptionally upregulated and facilitates tumor survival under metabolic stress in CRC [91]. Moreover, circUTRN directly binds ACC1 and impairs its polymerization through both phosphorylation-dependent and -independent mechanisms, thereby disrupting de novo fatty acid synthesis and inhibiting prostate cancer proliferation [92]. By contrast, circSpdyA promotes fatty acid synthesis through an alternative mechanism. It encodes a 127-amino-acid peptide that directly binds FASN to enhance de novo lipogenesis in breast cancer [93]. In cholesterol metabolism, driven by scavenger receptors like SCARB1, circABCA1 forms a ternary complex with IGF2BP3 and SCARB1 mRNA, stabilizing the latter to enhance cholesterol uptake while suppressing its efflux. This drives cholesterol accumulation and promotes tumor progression in ccRCC [94]. Furthermore, circMBOAT2 promotes intrahepatic cholangiocarcinoma progression by stabilizing PTBP1 to facilitate FASN mRNA cytoplasmic export, thereby enhancing lipid synthesis [95].

Amino acid metabolism modulates tumor progression by regulating aspartate metabolism, serine/glycine metabolism, and glutaminolysis/branched-chain amino acid (BCAA) catabolism. Serine hydroxymethyltransferase 2 (SHMT2) and argininosuccinate synthetase 1 (ASS1) control serine/glycine and aspartate metabolism, respectively, while glutaminase (GLS) and branched-chain amino acid transaminase 2 (BCAT2) regulate glutaminolysis/BCAA catabolism. In the aspartate metabolism axis, circRAPGEF1 is enriched in liver cancer stem cells (LCSCs) and competitively binds IGF2BP3 to destabilize ASS1 mRNA, driving aspartate accumulation to fuel HCC stemness properties and conferring sorafenib resistance [96]. In the serine/glycine metabolism axis, circMYH9 promotes serine/glycine metabolism and enhances glutathione recycling to suppress ROS accumulation, which promotes CRC proliferation [97]. Moreover, circHECTD1 promotes GC progression by activating β-catenin/c-Myc signaling to enhance glutaminolysis [98], and hsa_circ_0126925, a gut microbiota-responsive circRNA, targets BCAT2 to maintain BCAA metabolic homeostasis and promote CRC progression [99].

The TCA cycle and OXPHOS require moderate mitochondrial respiration to maintain homeostasis, with circRNAs modulating this equilibrium. CircZFR is induced by glucose deprivation and significantly upregulated in lung adenocarcinoma (LUAD) tissues, where it enhances OXPHOS by regulating alternative splicing of mitochondrial respiratory chain-related genes, thereby enabling tumor cells to adapt to energy stress [100]. In thyroid cancer, circPSD3 sponges miR-338-5p to upregulate SUCLG2, a key TCA cycle enzyme that maintains α-ketoglutarate-to-succinate flux and mitochondrial respiration, thereby driving tumor proliferation and suppressing apoptosis [101].

Regulation of circRNAs in angiogenesis

Solid tumors require the formation of new blood vessels, termed angiogenesis, to supply nutrients and oxygen [102]. This process is primarily orchestrated by hypoxia-inducible factor 1α (HIF-1α), a master transcription factor that accumulates under hypoxic microenvironments to drive vascular endothelial growth factor (VEGF) expression. CircRNAs actively participate in this angiogenic switch by serving as pivotal upstream regulators of the HIF-1α/VEGF axis.

Under hypoxic stress, hsa_circ_0010729 is significantly upregulated in human umbilical vein endothelial cells (HUVECs). It functions as a ceRNA by specifically sponging miR-186, thereby preventing miR-186-mediated degradation of HIF-1α mRNA. This leads to HIF-1α protein accumulation and subsequent transcriptional activation of VEGF, which stimulates endothelial cell proliferation and migration while suppressing apoptosis, ultimately promoting vascular network formation in tumors [103]. Similarly, circ_0030998 is overexpressed in CRC tissues and acts as an oncogenic sponge for miR-567, thereby upregulating VEGFA expression and promoting endothelial cell proliferation in vitro and in vivo [104]. Furthermore, intercellular communication via exosomal circRNAs adds another layer of angiogenic regulation. Circ_0084043, an exosomal circRNA secreted by cancer-associated fibroblasts (CAFs) in CRC, is horizontally transferred to endothelial cells. In the recipient cells, it sponges miR-140-3p to derepress the HIF-1α/VEGF signaling axis, thereby enhancing endothelial cell proliferation, migration, and vascular formation [105]. Similarly, circPUM1 is released from ovarian cancer cells via exosomes and delivered to vascular endothelial cells, where circPUM1 sequesters miR-607 to upregulate VEGFA, thereby boosting endothelial cell viability, invasion, migration, and angiogenic capacity [106].

Regulation of circRNAs in tumor cell genomic stability

Tumor cells frequently experience high levels of endogenous DNA double-strand breaks (DSBs) driven by oncogene-induced replication stress [107]. Although this genomic instability fuels malignant evolution, tumor cells must also maintain a functional DNA damage repair (DDR) network to prevent catastrophic cell death. Emerging evidence reveals that endogenous circRNAs directly and actively dictate tumor cell genomic stability, primarily through modulating DDR machinery and factor expression, as well as regulating chromatin organization.

CircRNAs can bind to core DNA damage sensors, such as the H2A.X variant histone (H2AX), to orchestrate or disrupt repair complex assembly. For instance, circ-Ccnb1 functions as a context-dependent tumor suppressor by sensing p53 status. In breast cancer cells with wild-type p53, it scaffolds H2AX and p53 to sequester Bclaf1, thereby supporting DNA repair and genomic stability. Upon p53 mutation, this complex switches to a pro-death circ-Ccnb1-H2AX-Bclaf1 interaction that eliminates malignant cells. Consequently, the frequent loss of circ-Ccnb1 in breast cancer abolishes this dual safeguard, allowing the accumulation of unrepaired DNA lesions and driving tumor progression [108].

Beyond scaffolding DDR sensors, circRNAs can modulate genomic stability through direct regulation of core DNA repair factor expression via RNA-RNA interactions. CircHIPK3, a widely studied circRNA deregulated across diverse tumor types, was recently shown to regulate BRCA1, an essential factor in homologous recombination (HR) repair, through a miRNA-independent mechanism [109]. Specifically, the back-splicing junction (BSJ) sequence of circHIPK3 directly base-pairs with the last coding exon of BRCA1 mRNA, which sustains BRCA1 protein levels and maintains HR repair competence. Disruption of this RNA-RNA interaction triggers BRCA1 downregulation, leading to DNA damage accumulation in tumor cells.

At the level of chromatin organization, circRNAs can directly regulate DDR pathway activation as integral chromatin-associated components. A recently identified subset of chromatin-associated circRNAs (cacircRNAs) was found to be functionally involved in the non-homologous end joining (NHEJ) pathway. Among these, ciCRLF3 [2] is aberrantly upregulated in MLL-rearranged leukemia, a clinically fatal malignancy with low mutation burden, and its elevated expression correlates with poor prognosis [110]. Mechanistically, ciCRLF3 [2] is a previously unrecognized component of the NHEJ complex that directly recruits NHEJ regulators to DNA lesions, supporting abundant DNA repair in leukemia cells.

Regulation of circRNAs in antitumor immunity

CircRNAs can facilitate tumor progression by modulating antitumor immunity. In addition to remodeling the TME via tumor-intrinsic reprogramming, recent studies reveal that circRNAs and their encoded peptides can be secreted into the TME as long-distance messengers that modulate intercellular crosstalk and antitumor immune responses.

Importantly, this secretion occurs primarily via exosomes. Exosomes are nano-sized extracellular vesicles (30–150 nm) that serve as key vehicles for intercellular communication [111]. The sorting of circRNAs into exosomes is not a passive leakage but a highly selective and active process driven by specific RBPs (e.g., AGO2, ESRP1, hnRNPA2B1, KHDRBS3) [112–114]. This process can be further enhanced by m⁶A methylation [115] and hypoxia [116].

Once internalized by recipient immune cells, these exosomal circRNAs can systemically regulate antitumor immunity. This regulation begins at the earliest phase of immune surveillance, where circRNAs provide cryptic neoantigens for antigen presentation and fine-tune innate immune sensing. Subsequently, circRNAs modulate the magnitude of effector T and NK cell activation. Notably, tumors frequently hijack circRNA networks to drive immune evasion, primarily by promoting immunosuppressive macrophage polarization, inducing T and NK cell exhaustion, and expanding populations of immunosuppressive cells.

Regulation of circRNAs in antigen presentation and immune sensing

To successfully initiate the antitumor immune response, the immune system requires specific signals to optimally activate antigen-presenting cells (APCs) and recruit effector cells like CD8⁺ T cells. Within the TME, circRNAs intervene at this stage both by serving as a source of novel antigens and by modulating innate immune sensing.

Regulation of circRNAs in antigen presentation

The activation of effector T cells relies on the recognition of antigenic peptides presented by major histocompatibility complex (MHC) class I/II molecules (known as HLA-I/II in human). Such presentation occurs directly on tumor cells or via cross-presentation by professional APCs, primarily DCs [117]. Tumor frequently evades this surveillance by downregulating HLA expression or suppressing DC function [9]. CircRNAs can offer cryptic antigens that are particularly valuable for immunologically “cold” tumors lacking sufficient DNA mutations to generate canonical neoantigens.

A recent study utilizing HLA-I peptidomics and ribosome sequencing in breast cancer identified cryptic peptides noncanonically translated from tumor-specific circFAM53B. Notably, these cryptic peptides bind to both HLA-I and HLA-II, thereby simultaneously activating CD8⁺ and CD4⁺ T cells [56] (Fig. 5A). Efficient cross-presentation of these cryptic antigens by DCs further drives antigen-specific T cell priming and tumor infiltration, leading to effective tumor control in vivo. Clinically, circFAM53B expression and its encoded peptides strongly correlate with increased CD8⁺ T cell infiltration and significantly better survival in breast cancer and melanoma patients. These findings establish circFAM53B-derived cryptic peptides as a prognostic biomarker and a vaccination target.

Fig. 5.

Regulation of antigen presentation and innate immune sensing by circRNAs. A CircRNA-encoded cryptic peptides are non-canonically translated and presented by both HLA-I and HLA-II molecules on tumor cells. These peptides can be cross-presented by DCs to prime CD8⁺ and CD4⁺ T cells, driving antigen-specific T cell infiltration and tumor control. B CircRNAs modulate innate immune sensing primarily through RIG-I pathway-mediated and chemokine network-mediated T cell recruitment and infiltration

Regulation of circRNAs in innate immune sensing

Effective antitumor immunity requires innate signals to trigger inflammation and specific chemokine gradients to guide effector T cells into the tumor bed [118]. CircRNAs act as critical molecular switches linking these two interconnected processes. They regulate initial immune sensing primarily via the retinoic acid-inducible gene I (RIG-I) pathway and also directly modulate the chemokine networks essential for T cell infiltration (Fig. 5B).

RIG-I is a cytosolic pattern recognition receptor that typically surveils double-stranded or structurally aberrant RNAs. Upon ligand binding, RIG-I undergoes K63-linked polyubiquitination, translocates to mitochondria, and subsequently activates the MAVS/TBK1/IRF3 signaling cascade to drive the transcription of interferons (IFNs) and pro-inflammatory chemokines such as C-X-C motif chemokine ligand 10 (CXCL10) [119]. While transient RIG-I activation promotes antitumor immunity by recruiting effector T cells, chronic or dysregulated activation can exacerbate inflammation-driven tumorigenesis [120].

In NSCLC, circNDUFB2 is downregulated in tumor tissues and functions as a tumor-suppressive circRNA [121]. It is recognized by RIG-I, which activates the RIG-I-MAVS signaling cascade to promote the secretion of type I IFNs and the CXCL10, thereby robustly recruiting immune cells into the TME. Similarly, in osteosarcoma, circKEAP1 interacts with RIG-I to activate antitumor immunity via the IFN-γ pathway, and its expression is associated with favorable patient prognosis, positioning it as a tumor suppressor [122].

By contrast, a group of circRNAs function as negative regulators of this pathway. In soft tissue sarcoma, circCsnk1g3 and circAnkib1 physically bind RIG-I but act as inhibitors that block RIG-I-mediated pathway activation, thereby suppressing the expression of IFN-related genes and pro-inflammatory factors in sarcoma cells and promoting tumor growth [123]. Moreover, circRIG-I, whose expression is associated with frameshift germline mutations of the RIG-I gene, is frequently upregulated in colon cancers and predicts poor patient outcome [120]. Mechanistically, circRIG-I interacts with DDX3X, which in turn stimulates the MAVS/TBK1/IRF3 signaling cascade, thereby aggravating chronic inflammation-driven tumorigenesis.

Certain circRNAs can directly disrupt chemokine gradients, such as CXCL10, CXCL11 and CXCL12, that are required for effective T cell infiltration. In NSCLC, the tumor-intrinsic oncogenic circFNDC3B is significantly upregulated [124]. It binds the transcription factor TFII-I to form an RNA-protein complex that disrupts interaction between TFII-I and STAT1, thereby abrogating STAT1-dependent transcriptional activation of CXCL10 and CXCL11, which impairs CD8⁺ T cell infiltration and antitumor immunity. Furthermore, in pancreatic ductal adenocarcinoma (PDAC), CAF-secreted circKLHL24 drives excessive CXCL12 secretion by tumor cells through sponging miR-615-5p to upregulate CXCL12 mRNA [125]. The resulting CXCL12 overexpression creates a chemical barrier that limits T cell entry.

Regulation of circRNAs in effector cell activation

Once effector cells, such as CD8⁺ T cells and NK cells, successfully infiltrate the tumor bed, their ability to eradicate tumor cells relies on robust cellular activation. Tumor cells actively exploit circRNA networks to suppress these critical activation pathways, thereby promoting cancer progression.

Optimal expansion and tumor-killing capacity of CD8⁺ T cells strictly depend on T cell receptor (TCR) engagement [126]. Upon recognizing antigen-MHC complexes, the TCR triggers a rapid phosphorylation cascade involving key kinases such as LCK, ZAP70, and LAT to drive proliferation and cytokine release. Tumor-derived circRNAs can directly modulate this network by targeting these early signal transducers or downstream effectors like the tyrosine phosphatase SHP2, which dephosphorylates TCR signaling components. In GC, tumor-derived exosomal circMAN1A2, whose elevated expression correlates with poor survival in GC patients, is efficiently delivered into CD8⁺ T cells, where it binds to and stabilizes the splicing factor SFPQ. This suppresses the phosphorylation of LCK, ZAP70, and LAT, thus inhibiting T cell activation [127]. In NSCLC, tumor-derived exosomal circUSP7 sponges miR-934 to upregulate SHP2 in CD8⁺ T cells, which profoundly inhibits TCR-mediated proliferation and dampens the secretion of cytotoxic cytokines, including IFN-γ, TNF-α, Granzyme B, and Perforin, thereby promoting tumor progression [128].

NK cells play a crucial role in antitumor immunity by directly eliminating malignant cells and modulating broader immune responses [129, 130]. CircRNAs can modulate the activation receptors of NK cells, such as NKG2D. In HCC, tumor cells downregulate circARSP91, impairing the expression of UL16-binding protein-1 (ULBP1), an NKG2D ligand. This dampens NKG2D receptor engagement on surrounding NK cells and suppresses NK cell activation and cytotoxicity against tumor cells [131]. Transcriptomic analyses have revealed that circARSP91 expression positively correlates with NK cell infiltration and overall survival in HCC patients, suggesting its protective role against tumor progression.

Regulation of circRNAs in immune evasion

During the effector phase of antitumor immunity, the tumor establishes a complex immunosuppressive microenvironment to neutralize effector cells. Tumor-derived circRNAs drive this evasion by polarizing macrophage, inducing effector cell exhaustion, and recruiting suppressive cell populations. These multifaceted roles collectively foster a tolerant niche and enable immune escape.

Regulation of circRNAs in macrophage polarization

Within the TME, TAMs represent a critically important component and play extensive roles in regulating tumor progression [132]. Macrophages undergo polarization into distinct functional states, primarily the antitumor M1 phenotype or pro-tumor M2 phenotype, depending on specific microenvironmental cues and stimuli [133, 134]. Exosomal circRNAs critically regulate macrophage polarization toward the M2 phenotype through several canonical signaling axes, including the STAT3, PI3K/AKT and NF-κB pathways, thereby shaping the immunosuppressive TME, promoting tumor progression, and contributing to therapy resistance [135].

The STAT3 signaling pathway, frequently activated via upstream kinases such as JAK1, is a central driver of M2 macrophage polarization and immunosuppression in the TME. Multiple tumor-derived exosomal circRNAs have been shown to activate this axis via distinct mechanisms (Fig. 6A). For instance, exosomal circPOLQ is significantly upregulated in CRC tissues, and its high expression is associated with poorer patient prognosis. Mechanistically, circPOLQ enters macrophages via tumor-derived exosomes and sponges miR-379-3p, thereby activating the IL-10/STAT3 axis to drive M2 polarization [136]. Similarly, in renal cell carcinoma (RCC), exosomal circSAFB2 acts as a sponge for miR-620, relieving the inhibition on JAK1 and subsequently activating STAT3 signaling. By mediating M2 macrophage polarization through the miR-620/JAK1/STAT3 axis, circSAFB2 promotes RCC metastasis and contributes to immune escape [137].

Fig. 6.

Regulation of macrophage polarization toward the M2 phenotype by circRNAs. Tumor-derived exosomal circRNAs are transferred into macrophages and promote M2 polarization primarily through (A) STAT3, (B) PI3K/AKT, and (C) NF-κB signaling pathways. These pathways reprogram TAMs toward a pro-tumor M2 phenotype, shaping an immunosuppressive TME that facilitates tumor progression and therapy resistance

The PI3K/AKT pathway, whose activation is modulated by the tumor suppressor PTEN, is another critical regulator that promotes M2 polarization in TAMs (Fig. 6B). For instance, tumor-derived exosomal circFARSA promotes PTEN ubiquitination and degradation, which results in activation of the PI3K/AKT signaling cascade to enhance M2 macrophage polarization and promote tumor progression in NSCLC [138].

The NF-κB axis also potently mediates M2 polarization (Fig. 6C). Glioma-cell-derived exosomes enriched with circ-001422 are transferred to macrophages, where they induce M2 polarization by activating the STAT3/NF-κB pathway [139]. Higher levels of circ-001422 correlate with poorer prognosis in glioma patients, and its targeting may represent a potential therapeutic approach for this malignancy. In breast cancer, tumor-derived exosomal cSERPINE2 is transferred to TAMs, where it sponges miR-513a-5p, activates the NF-κB pathway and enhances IL-6 secretion. This, in turn, establishes a positive feedback loop that boosts cSERPINE2 biogenesis in tumor cells and further promotes TAM recruitment, ultimately fostering breast cancer progression [140].

Emerging evidence has revealed additional non-canonical mechanisms of circRNA-mediated M2 polarization, such as modulation of protein stability. For instance, circNEIL3 is highly upregulated in glioma tissues and correlates with malignant progression and poor prognosis. Packaged into exosomes, circNEIL3 is transferred to TAMs and stabilizes IGF2BP3, reprogramming macrophages toward an M2-like phenotype and promoting glioma progression [141].

Regulation of circRNAs in effector cell exhaustion

Even when effector cells successfully infiltrate the TME, their antitumor functions can be frequently neutralized. Sustained exposure to tumor antigens and immunosuppressive signals drives effector cells into a state of exhaustion [142]. Emerging evidence highlights that tumor-derived circRNAs drive this dysfunctional state through distinct mechanisms, including upregulating inhibitory immune checkpoints, epigenetically silencing effector genes, and reshaping the extracellular milieu to indirectly drive effector cells exhaustion.

CircRNAs can enforce inhibitory checkpoint expression on effector cells, a canonical hallmark of exhaustion (Fig. 7A). This dysfunctional state is primarily characterized by the overexpression of inhibitory immune checkpoints such as programmed cell death protein 1 (PD-1). Upon binding its ligands (PD-L1/PD-L2), PD-1 attenuates TCR-mediated proliferation and cytokine secretion of CD8⁺ T cells [143]. Blocking PD-1 can restore antitumor immunity [144, 145], thereby validating this receptor as a major therapeutic target in oncology. In HCC, exosomal circCCAR1 is transferred into CD8⁺ T cells, where it stabilizes the PD-1 protein, thereby causing CD8⁺ T-cell exhaustion and driving resistance to anti-PD1 immunotherapy [114]. Similarly, circRNAs can upregulate other emerging checkpoints, such as TIM-3, an inhibitory checkpoint that is frequently co-expressed with PD-1 on terminally exhausted T cells and functionally exhausted NK cells [146]. Ligation of TIM-3 by its ligands delivers inhibitory signals that suppress effector functions, including IFN-γ and TNF-α secretion. In HCC, exosomal circUHRF1 enters NK cells and acts as a miR-449c-5p sponge to upregulate TIM-3, which specifically suppresses NK cell-derived IFN-γ and TNF-α secretion and drives resistance to anti-PD-1 therapy [147].

Fig. 7.

Regulation of effector cell exhaustion by circRNAs. A Tumor-derived exosomal circRNAs enter effector cells and upregulate the expression of PD-1 and TIM-3. B Exosomal circRNAs act as protein scaffolds to silence the transcription of cytotoxic molecules, such as IFN-γ and Granzyme B. C CircRNAs indirectly drive effector cell exhaustion by modulating the homeostasis of extracellular metabolites

CircRNAs can drive exhaustion by epigenetically silencing genes critical for effector function (Fig. 7B). For instance, exosomal circ_0006896 enters CD8⁺ T cells and acts as a protein scaffold for the histone deacetylase HDAC1. This interaction reduces histone H3 acetylation at the LEF1 promoter, a pivotal transcription factor that governs CD8⁺ T cell effector differentiation and cytotoxic function, and silences the transcription of critical cytotoxic molecules such as IFN-γ and Granzyme B [148]. High expression of circ_0006896 in both acute myeloid leukemia (AML) cells and patient-derived exosomes correlates with poor prognosis and relapse in AML patients.

CircRNAs can also drive exhaustion by remodeling the extracellular metabolite milieu, indirectly suppressing effector cell function (Fig. 7C). This can be achieved either through exosomal circRNAs that manipulate the immunosuppressive adenosine pathway or through circRNA-encoded peptides that deplete essential nutrients. For example, exosomal circTMEM181 enters macrophages and sponges miR-488-3p to upregulate CD39, an ectonucleotidase that coordinates with CD73 on tumor cells, elevating extracellular adenosine (ADO) levels, which suppress effector CD8⁺ T cell function and drive anti-PD-1 resistance [149]. HCC-derived exosomal circTMEM181 is significantly elevated in HCC patients who respond poorly to anti-PD1 therapy and in those with poor postoperative prognosis. In a distinct but converging mechanism, the tumor-intrinsic circRNA-encoded peptide circPETH-147aa impairs antitumor immunity by stabilizing SLC43A2 mRNA in HCC cells. This depletes environmental essential amino acids, specifically methionine and leucine, thereby starving CD8 + T cells and pushing them into an exhausted and dysfunctional state [89].

Regulation of circRNAs in modulating immunosuppressive cells

Major immunosuppressive populations in the TME include regulatory T cells (Tregs), pro-tumor N2-polarized tumor-associated neutrophils (TANs), and polymorphonuclear MDSCs (PMN-MDSCs) [150–152]. Tumor-derived exosomal circRNAs can potently induce and reprogram these populations, thereby establishing an immunosuppressive milieu that restrains antitumor immunity and limits the efficacy of immunotherapy.

Tregs are a specialized subset of CD4⁺ T cells characterized by expression of the master transcription factor FOXP3. Their differentiation and expansion are critically driven by TGF-β/Smad signaling [153]. Within the TME, tumor-derived exosomal circRNAs can hijack this signaling axis to aberrantly expand Tregs and enforce immune escape. In HCC, for instance, exosomal circGSE1 derived from HCC cells is transferred to CD4⁺ T cells, where it acts as a molecular sponge for miR-324-5p, thereby upregulating TGFBR1 and downstream Smad3 signaling cascade. This promotes the differentiation and expansion of Tregs [154]. These expanded Tregs potently suppress antitumor CD8⁺ T cell function, thereby facilitating HCC immune escape and disease progression.

Neutrophils serve as a crucial component of the innate immune system and can infiltrate the TME as TANs [155]. TANs exhibit functional plasticity and, analogous to the M1/M2 classification of macrophages, can be classified into N1 and N2 subtypes [156]. The N1 phenotype exhibits cytotoxic and antitumor functions, whereas the N2 phenotype supports tumor growth and progression [157]. A key driver of N2 polarization is TGF-β, which reprograms neutrophils toward a pro-tumorigenic phenotype. In CRC, exosomal circPACRGL is significantly upregulated and serves as a sponge for miR-142-3p and miR-506-3p, thereby facilitating TGF-β1 expression. This, in turn, drives the differentiation of neutrophils from the N1 to the N2 phenotype, promoting a pro-metastatic TME and CRC progression [158].

PMN-MDSCs are a heterogeneous population of pathologically activated immature myeloid cells possessing potent immunosuppressive capabilities [159]. Within the TME, infiltrating PMN-MDSCs contribute significantly to cancer progression and diminished responses to immunotherapy, with their abundance being strongly linked to unfavorable clinical outcomes [160]. PMN-MDSCs exert their immunosuppressive functions primarily through the elevation of ROS and arginase-1 (ARG-1), upregulation of fatty acid transporter protein 2 (FATP2)-driven lipid metabolism, and suppression of receptor-interacting protein kinase 3 (RIPK3)-mediated necroptotic signaling [161, 162]. In bladder cancer, tumor-derived exosomal circRNA_0013936 is transferred to PMN-MDSCs and enhances their immunosuppressive activity by sponging miR-320a and miR-301b to upregulate FATP2 and downregulate RIPK3, respectively, thereby inhibiting CD8⁺ T cells and driving immune evasion and tumor progression [163].

Clinical potential of CircRNAs in cancer

CircRNAs possess good stability and are secreted into body fluids, positioning them as promising liquid biopsy biomarkers. Furthermore, circRNA exhibits low innate immunogenicity and programmable translational capacity, establishing them as a therapeutic platform for cancer immunotherapy [164].

CircRNAs in cancer diagnosis and prognosis

Liquid biopsy, the analysis of tumor-derived materials in bodily fluids such as blood, urine and sputum, enables real-time, non-invasive monitoring in precision oncology. Owing to their covalently closed loop structure, circRNAs exhibit exceptional stability in body fluids [165, 166], underpinning the clinical translational value of exosomal circRNAs as diagnostic and prognostic biomarkers, with evidence ranging from foundational research to clinical validation.

Blood biomarkers

Blood biomarkers offer accessibility, reproducibility, and the capacity for longitudinal monitoring [167]. Circulating circRNAs have demonstrated robust diagnostic and prognostic performance across diverse malignancies. For diagnosis, the circRNA F-circEA, generated from the EML4-ALK fusion gene in NSCLC, is specifically detected in the plasma of EML4-ALK-positive patients but is absent in those without the fusion gene [168], underscoring its promise as a non-invasive liquid biopsy biomarker for identifying candidates for ALK-targeted therapy.

Furthermore, Zheng and colleagues conducted a comprehensive clinical investigation of exosomal circLPAR1 in CRC using a cohort of 112 CRC patients [169]. Its expression levels were significantly decreased during CRC development and recovered post-surgery, demonstrating a dynamic correlation with tumor burden. When combined with conventional clinical biomarkers, such as carcinoembryonic antigen (CEA) and carbohydrate antigen 19 − 9 (CA19-9), the diagnostic panel achieved an area under the curve (AUC) of 0.875, substantially outperforming these markers individually. This rigorous validation underscores the potential of exosomal circRNAs, such as circLPAR1, as reliable non-invasive tools for cancer detection.

CircRNAs have also emerged as promising prognostic indicators. In chemotherapy, exosomal hsa_circ_0004771 is overexpressed in CRC serum and further elevated in patients resistant to 5-fluorouracil (5-FU), supporting its role as a predictive biomarker for 5-FU resistance and poor prognosis [170]. In the context of immunotherapy, exosomal circZNF451 is markedly higher in lung adenocarcinoma patients with progressive disease (PD) than those with partial remission (PR) after PD-1 blockade therapy, indicating its potential to predict PD-1 blockade sensitivity [171]. Similarly, in HCC, circUHRF1 is secreted via exosomes, and its elevated levels correlate with poor prognosis, NK cell dysfunction, and resistance to anti-PD-1 therapy, positioning it as a liquid-biopsy-based biomarker for identifying patients unlikely to benefit from immune checkpoint blockade [147].

Urine biomarkers

Urine-based liquid biopsy is particularly advantageous for the diagnosis and prognosis of urological malignancies, as urine is in direct contact with the urothelium and the prostate, where tumor-derived circRNAs are readily detectable and increasingly recognized for their diagnostic potential [172]. In bladder urothelial carcinoma, circPRMT5 is significantly elevated in both plasma and urine exosomes, and its upregulation correlates with tumor metastasis. Across two independent cohorts (240 and 257 patients), high circPRMT5 levels independently predicted advanced clinical stage, lymph node metastasis, and shorter disease-free survival, establishing it as a non-invasive prognostic tool capable of informing surgical decision-making prior to intervention [173].

For prostate cancer, a urine extracellular vesicle circRNA classifier (comprising circPDLIM5, circSCAF8, circPLXDC2, circSCAMP1, and circCCNT2) was developed and validated across three cohorts (263, 497, and 505 patients) for detecting high-grade prostate cancer, consistently outperforming two standard-of-care risk calculators (PCPT-RC 2.0 and ERSPC-RC). Combining the classifier with the risk calculators further improved predictive accuracy. This assay requires no pre-collection digital rectal examination, integrates easily into routine workflows, and offers a repeatable, non-invasive strategy to reduce unnecessary biopsies while improving detection of clinically significant disease [174].

Sputum biomarkers

Sputum, given its direct contact with the respiratory tract and proximity to lung tumors, represents another attractive biofluid for non-invasive detection. Bai and colleagues identified sputum circ_0006949 as a diagnostic biomarker for NSCLC, showing significant upregulation in the sputum of patients with stage I-II NSCLC. High expression of this circRNA correlates with poor prognosis and outperforms conventional plasma markers such as CEA and Cytokeratin 19 fragment (CYFRA 21 − 1), demonstrating its potential for early detection [175]. Further investigation into the prognostic role of sputum circRNAs and the integration of plasma, urine, and sputum circRNA-based assays could establish a comprehensive multi-fluid biopsy strategy for precision oncology.

CircRNAs in cancer immunotherapy

Current therapeutic strategies utilizing synthetic circRNAs in cancer immunotherapy primarily include cancer vaccination, localized intratumoral immunomodulation, and adoptive cell therapy. These approaches exploit the coding capacity of circRNAs and require efficient delivery systems, such as lipid nanoparticles (LNPs), which protect circRNA from degradation and ensures durable expression in target cells [176].

Cancer vaccination

CircRNA vaccines encoding tumor antigens can elicit potent antitumor immune responses. For instance, a synthetic circRNA encoding fused HPV16 E7/E6 antigens and encapsulated with lipid nanoparticles (LNP-circRNA^E7E6) has been developed to treat HPV16-related solid tumors [177]. In mice bearing TC-1 tumors, a model of HPV-positive cancer, administration of LNP-circRNA^E7E6 elicits complete regression of established tumors and induces robust, long-term immunological memory that protects cured animals against tumor rechallenge. Furthermore, the circRNA cancer vaccine TI-0093, the clinical version of LNP-circRNA^E7E6, received Investigational New Drug (IND) clearance from China’s National Medical Products Administration (NMPA, CXSL2500461). The first patient was dosed in a multicenter Phase I trial in November 2025, marking a pivotal advance toward establishing circRNA vaccines as a new pillar of cancer immunotherapy.

The route of vaccine administration critically affects both efficacy and safety. Conventional routes, such as intramuscular and intravenous injections, may have limitations in inducing mucosal immunity at tumor barriers and carry risks of systemic off-target accumulation and inflammation [178]. Intranasal delivery of LNP-circRNA vaccines leverages the respiratory mucosa to induce robust local T cell immunity and durable memory while reducing systemic exposure. This approach also synergizes with chimeric antigen receptor T (CAR-T) cell therapy, offering a compelling strategy for pulmonary malignancies [179].

Localized intratumoral immunomodulation

Localized intratumoral delivery represents another immunotherapy strategy, which aims to convert immunologically “cold” tumors into “hot” ones (Fig. 8). As a representative example, Xu and colleagues show that a single intratumoral injection of LNPs carrying IL-12-encoding circRNA is sufficient to induce robust antitumor immunity and marked tumor regression in a Lewis lung carcinoma (LLC) model [180]. This treatment substantially increases CD45⁺ leukocyte and CD8⁺ T cell infiltration into the tumor parenchyma. Similarly, researchers develop circILNb, a synthetic circRNA co-expressing IL-15 and anti-PD-L1 nanobody. Intratumoral injection of LNP-encapsulated circILNb achieves sustained local co-expression of both payloads and effective tumor control [181].

Fig. 8.

Intratumoral delivery of circRNA-encoded immunomodulators. Intratumoral injection of circRNAs encoding immunostimulatory molecules (e.g., IL-12/IL-15) enables sustained local protein expression, converting immunologically “cold” tumors into a “hot” phenotype while minimizing systemic toxicity, which highlights the paradigm shift from using circRNAs merely as targets to harnessing them as active therapeutic agents

Adoptive cell therapy

Conventional adoptive cell therapies, particularly CAR-T cell therapy, face challenges for broader clinical application in cancer immunotherapy [182]. CircRNA-based in vivo CAR platforms are emerging as a promising alternative. A recent study reports a comprehensive in vivo panCAR platform based on circRNA-encoded CARs (circRNA^CAR) delivered by immunocyte-tropic LNPs, which generates CAR-T, CAR-NK, and CAR-macrophage in vivo [183]. In murine tumor models, this multi-effector response significantly inhibits tumor growth and reshapes the TME from an immunosuppressive to an immune-active state. Notably, combining in vivo panCAR therapy with circRNA-based vaccines encoding the corresponding HER2 antigen produces synergistically enhanced antitumor immunity across multiple mouse models.

Conclusions and future perspectives

CircRNAs have emerged as novel transcriptional and post-transcriptional regulators that link tumor biology to antitumor immunity by acting as both intracellular signaling modulators and intercellular messengers.

Current circRNA expression profiles derive largely from bulk RNA sequencing, which obscures the cellular heterogeneity of the TME. Consequently, information on circRNA distribution across different cell types remain lacking. The development of sensitive single-cell circRNA sequencing and spatial transcriptomics offers a promising avenue to resolve the spatial distribution of circRNAs.

The clinical application of circRNA-based immunotherapeutics is primarily hindered by inefficient circularization and suboptimal translational output. Therefore, advancing circRNAs into precision cancer immunotherapies necessitates the development of novel circularization strategies and optimized translation-enhancing elements to achieve robust and sustained antigen expression.

Abbreviations

CAF

Cancer-associated fibroblast

CAR-T

Chimeric antigen receptor T

CDK

Cyclin-dependent kinase

ceRNA

Competitive endogenous RNA

circRNA

Circular RNA

CRC

Colorectal cancer

DC

Dendritic cell

DDR

DNA damage repair

EMT

Epithelial-mesenchymal transition

ESCC

Esophageal squamous cell carcinoma

FASN

Fatty acid synthase

GC

Gastric cancer

H2AX

H2A.X variant histone

HCC

Hepatocellular carcinoma

HIF-1α

Hypoxia-inducible factor 1α

HLA

Human leukocyte antigen

IFN

Interferon

LLPS

Liquid-liquid phase separation

LNP

Lipid nanoparticle

m⁶A

N⁶-methyladenosine

MDSC

Myeloid-derived suppressor cell

MHC

Major histocompatibility complex

NK cell

Natural killer cell

NSCLC

Non-small cell lung cancer

PD-1

Programmed cell death protein 1

PMN-MDSC

Polymorphonuclear MDSC

PKM2

Pyruvate kinase M2

Pol II

RNA polymerase II

pre-mRNA

Precursor messenger RNA

pre-tRNA

Precursor transfer RNA

RBP

RNA-binding protein

RCD

Regulated cell death

RIG-I

Retinoic acid-inducible gene I

ROS

Reactive oxygen species

TAM

Tumor-associated macrophage

TAN

Tumor-associated neutrophil

TCA

Tricarboxylic acid

TCR

T cell receptor

TME

Tumor microenvironment

TNBC

Triple-negative breast cancer

Treg

Regulatory T cell

VEGF

Vascular endothelial growth factor

Authors’ contributions

Z.Y.J. contributed to the literature searching and original draft writing. Y.P. and X.J.Y. supervised the work and revised the manuscript. All authors approved the final version.

Funding

This work was supported by the National Natural Science Foundation of China (82273168) and the 1.3.5 Project for Disciplines of Excellence, West China Hospital of Sichuan University (ZYGD23027).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.