Circular RNAs in hepatocellular carcinoma: a systematic qualitative review of regulatory mechanisms, therapeutic resistance, and clinical translation

Abstract

Background

Hepatocellular carcinoma (HCC) is characterized by high mortality, frequent recurrence, and limited long-term benefit from systemic therapy and radiotherapy. Circular RNAs (circRNAs), a class of covalently closed non-coding RNAs with high molecular stability and tissue-specific expression, have emerged as important regulators of HCC biology and treatment response.

Methods

We searched PubMed from database inception to May 2, 2026 and manually screened the reference lists of eligible studies and relevant reviews for studies investigating circRNA-related molecular mechanisms, tumor progression, therapeutic resistance, radiotherapy response, or translational applications in HCC. A total of 694 records were identified in PubMed, 694 records were screened after duplicate removal, and 42 studies were included in the qualitative synthesis. Original studies involving human tissues, clinical cohorts, cell lines, or animal models were eligible. Title/abstract screening, full-text assessment, and data extraction were performed independently by two reviewers, with disagreements resolved by consensus or senior-author adjudication. Given substantial heterogeneity in study design, model systems, molecular endpoints, and reported outcomes, findings were synthesized qualitatively rather than by meta-analysis, and a structured qualitative evidence appraisal was used to weight mechanistic and translational conclusions.

Results

The included evidence indicates that circRNAs participate in HCC progression through miRNA sponging, RNA-binding protein interactions, transcriptional regulation, epigenetic modulation, and, in selected contexts, translational activity. Distinct oncogenic and tumor-suppressive circRNAs converge on pathways related to proliferation, epithelial-mesenchymal transition, metastasis, ferroptosis, stemness, and DNA-damage repair. CircRNAs were also associated with sorafenib and lenvatinib response, radioresistance, and radiosensitivity. Emerging translational applications include tissue and liquid-biopsy biomarkers, especially exosomal circRNAs, as well as junction-targeted oligonucleotide strategies.

Conclusions

Current evidence supports circRNAs as mechanistically relevant regulators and promising translational candidates in HCC. However, progress toward clinical application is constrained by etiologic heterogeneity, limited standardization, incomplete specificity validation for therapeutic targeting, and the lack of multicenter prospective validation studies.

Background

Hepatocellular carcinoma (HCC) is the dominant histological subtype of primary liver cancer and remains a major cause of cancer-related mortality worldwide. Global cancer statistics indicate that HCC accounts for approximately 75%–85% of primary liver cancer cases; notably, incidence and mortality are particularly high in China, which contributes to more than half of the global disease burden [1, 2]. Despite advances in multimodal management, including surgical resection, locoregional therapies, molecular targeted agents, immunotherapy, and radiotherapy, the long-term outcomes remain unsatisfactory. Late-stage diagnosis, frequent recurrence, and the rapid emergence of therapeutic resistance continue to limit durable clinical benefit. Recent oncology biomarker research has also linked tumor progression and immune-microenvironment modulation to translational biomarker development [3]. Recent studies have further emphasized the rapidly evolving therapeutic landscape of HCC, including immunotherapy rechallenge and surgical prognostic factors, the importance of molecular drivers of tumor progression, and the growing clinical relevance of etiology-specific disease contexts such as NAFLD-related HCC [4–9].

Circular RNAs (circRNAs) are endogenous RNA molecules generated by back-splicing that form covalently closed loops, a configuration associated with high molecular stability and cell- and tissue-specific expression patterns [10–12]. Compared with most linear RNAs, circRNAs are less susceptible to exonuclease-mediated degradation, can be enriched in extracellular vesicles, and contain back-splice junctions that enable more specific detection or junction-targeted inhibition. These features provide practical advantages for biomarker development and therapeutic targeting in HCC. In HCC, accumulating evidence links dysregulated circRNAs to malignant phenotypes and treatment response, supporting their potential roles as mechanistic regulators as well as candidate biomarkers or therapeutic targets [10–12]. However, prior reviews have often emphasized descriptive cataloguing of circRNA functions, whereas less attention has been given to the robustness of mechanistic evidence, therapy- and radiotherapy-resistance contexts, etiology-specific heterogeneity, and the methodological barriers that determine clinical translatability. In this context, integrating circRNA biology with broader advances in HCC progression, therapeutic vulnerability, and etiologic heterogeneity may help better position circRNA-based biomarkers and interventions within current translational research frameworks [4, 7–9].

In this systematic review, we synthesize current evidence on circRNA research in HCC, with emphasis on biogenesis, regulatory mechanisms, tumor progression, therapeutic resistance, radiotherapy response, and translational relevance. We also evaluate how the available literature supports biomarker development and circRNA-directed therapeutic strategies, and we highlight the principal methodological and clinical barriers that must be addressed before routine implementation.

Methods

Information sources and search strategy

This systematic review was conducted to summarize current evidence on the roles of circular RNAs (circRNAs) in hepatocellular carcinoma (HCC), with a focus on molecular mechanisms, tumor progression, therapeutic resistance, radiotherapy response, and translational applications. We searched PubMed from database inception to May 2, 2026 using the concepts “circular RNA” OR “circRNA” combined with “hepatocellular carcinoma” OR “liver cancer”. Reference lists of eligible articles and relevant review papers were also screened manually to identify additional potentially eligible studies. Manual reference-list screening was performed as a supplementary check and did not identify additional eligible studies beyond those captured by the PubMed search.

The search strategy was designed to maximize sensitivity for mechanistic and translational circRNA studies in HCC. The PubMed syntax combined circRNA-related terms with HCC/liver-cancer terms: (“circular RNA” OR “circular RNAs” OR circRNA OR circRNAs) AND (“hepatocellular carcinoma” OR HCC OR “liver cancer” OR “liver neoplasm*”). Both MeSH and Title/Abstract terms were used. The complete reproducible PubMed search string is provided in Supplementary Table S1. No meta-analysis-oriented search filters were applied because the review aimed to capture a broad range of mechanistic, preclinical, translational, and clinically oriented original studies. In total, 694 records were identified in PubMed.

Eligibility criteria and study selection

We included original studies investigating circRNA-associated molecular mechanisms, biological functions, biomarker potential, therapeutic resistance, or radiotherapy response in HCC. Eligible studies could involve human HCC tissues, adjacent liver tissues, blood-derived samples, serum, plasma, exosomes, clinical cohorts, HCC cell lines, or animal models. Studies were considered eligible if they provided mechanistic, functional, diagnostic, prognostic, or translational information relevant to circRNAs in HCC.

We excluded review articles, editorials, commentaries, conference abstracts lacking sufficient primary data, studies not focused on HCC, and reports without meaningful mechanistic or translational relevance to the aims of this review. Where overlapping datasets were suspected, the most informative and methodologically complete report was preferentially retained.

After removal of 0 duplicate records, 694 records underwent title and abstract screening. Of these, 560 were excluded during initial screening. A total of 134 reports were sought for retrieval, and 127 full-text articles were assessed for eligibility after 7 reports could not be retrieved. Eighty-five full-text articles were excluded, most commonly because they were reviews/editorials/comments (n = 24), not focused on HCC (n = 18), lacked mechanistic circRNA data (n = 21), lacked therapeutic-resistance, radiotherapy, or translational relevance (n = 16), or contained overlapping or insufficiently detailed datasets (n = 6). Ultimately, 42 studies were included in the qualitative synthesis.

Study selection was performed through sequential screening of titles/abstracts followed by full-text assessment. Two reviewers independently screened titles/abstracts and full texts against the eligibility criteria. Disagreements were first resolved by discussion and, when consensus could not be reached, by consultation with the senior author. Data extraction was also performed independently by two reviewers using a standardized extraction framework, followed by cross-checking for consistency. Because this review synthesized highly heterogeneous mechanistic and translational literature, studies were grouped narratively into the following synthesis domains: circRNA biology and functional modes of action, circRNAs regulating HCC progression, circRNAs associated with therapeutic resistance, circRNAs associated with radiotherapy response, and circRNAs with potential diagnostic, prognostic, or therapeutic translational relevance.

Data items, data extraction, and synthesis

For each eligible study, we extracted the circRNA investigated, study model or sample type, mechanistic class, principal downstream targets or signaling pathways, and the reported biological or translational effect. Additional variables of interest included whether the evidence related to tumor progression, epithelial-mesenchymal transition, metastasis, ferroptosis, stemness, DNA-damage repair, targeted therapy response, radiotherapy response, biomarker potential, exosome-mediated signaling, or etiology-related context such as HBV-associated disease.

Given the substantial heterogeneity across included studies in model systems, experimental platforms, endpoints, and outcome reporting, we did not perform a quantitative meta-analysis. Because most included studies were mechanistic laboratory studies, translational analyses, or small retrospective clinical studies, conventional pooled GRADE certainty assessment was not appropriate. Instead, we used a structured qualitative evidence appraisal to avoid assigning equal narrative weight to studies with substantially different validation depth. The appraisal considered study model, number and relevance of cell lines, animal-model support, patient-sample validation, rescue-experiment design, endogenous interaction validation, expression-dose plausibility, statistical reporting, and clinical/translational support. The resulting validation level is reported in the summary tables and was used to temper the strength of mechanistic and translational conclusions.

Results were summarized narratively and presented using structured thematic subsections, summary tables, and a schematic figure. Heterogeneity was considered primarily in relation to etiologic background, experimental context, therapeutic setting, and analyte type (e.g., tissue circRNA, circulating circRNA, or exosomal circRNA), and these sources of variability were addressed qualitatively in the Discussion. A review protocol was not prospectively registered; this limitation and its implications for interpretability are explicitly addressed in the Limitations section. A completed PRISMA 2020 checklist accompanies the submission, the study-selection flow diagram is presented in Fig. 1, and the complete PubMed search strategy is provided in Supplementary Table S1. Artificial intelligence (AI) tools were used solely for English language editing and formatting support and did not influence study selection, evidence interpretation, or the scientific conclusions of the review.

Fig. 1.

PRISMA 2020 flow diagram of study selection

Evidence appraisal and reporting transparency

Because the evidence base is dominated by mechanistic studies rather than randomized or directly comparable clinical studies, we performed a structured qualitative appraisal (Additional file 3) rather than a pooled risk-of-bias or GRADE-based quantitative certainty assessment. For preclinical animal experiments, appraisal items were adapted from commonly used animal-study domains, including model description, allocation/randomization where reported, blinding where reported, completeness of outcome reporting, and statistical appropriateness. For cell-based mechanistic studies, the appraisal considered the number of independent cell lines, use of gain- and loss-of-function approaches, positive/negative controls, biological replication, rescue experiments, concordance with endogenous expression levels, and whether the proposed mechanism was validated beyond prediction and reporter assays. For patient-derived studies, sample type, cohort size, clinicopathologic annotation, independent validation, and diagnostic/prognostic modeling were considered. This framework was used to classify conclusions as patient-supported or multi-model findings, animal-supported findings, cell-line-supported findings, or exploratory mechanistic hypotheses.

For miRNA-sponging or ceRNA claims, evidence was considered stronger only when bioinformatic prediction and luciferase assays were complemented by endogenous circRNA/miRNA interaction assays, such as RNA pull-down, Ago2-RIP/RISC-related evidence, expression concordance, and physiologically plausible rescue experiments. Claims based only on overexpression systems, reporter assays, or inverse expression correlations were interpreted cautiously as proposed axes rather than definitive proof of functional miRNA sequestration.

For therapy-resistance studies, special attention was given to model type. Stepwise drug-selected cell lines were considered useful for hypothesis generation but insufficient by themselves to establish clinically acquired resistance. Evidence was considered more translationally persuasive when supported by patient-derived samples, longitudinal treatment-response data, organoid or xenograft models, or in vivo resistance-validation experiments.

Results

Study selection

The literature screening process is illustrated in Fig. 1. A total of 694 records were identified in PubMed. After removal of 0 duplicate records, 694 records underwent title and abstract screening, and 560 records were excluded. Of the 134 reports sought for retrieval, 7 could not be retrieved; 127 full-text articles were assessed for eligibility and 85 were excluded with reasons. Ultimately, 42 studies were included in the qualitative synthesis.

Overview of circRNA biology relevant to HCC

CircRNAs are produced through a non-canonical back-splicing mechanism in which a downstream splice donor site is covalently linked to an upstream splice acceptor site [13]. Based on genomic origin, circRNAs are commonly classified as exonic circRNAs, intronic circRNAs, exon-intron circRNAs, or intergenic circRNAs [13, 14]. Exonic circRNAs constitute the majority and are mainly localized in the cytoplasm, whereas intronic and exon-intron circRNAs are often enriched in the nucleus. The circular configuration confers marked stability and permits accumulation in cells and body fluids, a property that underpins much of their biomarker appeal in HCC. Together, these core biological features provide the framework for interpreting the mechanistic and translational studies summarized below.

Functional modes of circRNA action

Building on their biogenesis and cellular localization, circRNAs regulate gene expression through multiple, not mutually exclusive mechanisms. The most extensively studied function is their role as competing endogenous RNAs (ceRNAs) that sequester microRNAs and thereby relieve repression of downstream target genes. CircRNAs can also interact directly with RNA-binding proteins (RBPs), functioning as molecular scaffolds or decoys that influence protein localization and activity. In addition, circRNAs may regulate transcription of their parental genes or, in certain contexts, be translated into functional peptides, reflecting the broad functional repertoire of circRNAs in HCC and other solid tumors [13–15]. Through these diverse mechanisms, circRNAs participate in complex regulatory networks involved in cancer development and progression.

circRNAs regulating HCC progression

Oncogenic circRNAs promoting HCC progression

Accumulating evidence indicates that numerous circRNAs function as oncogenic drivers in HCC, facilitating malignant progression through distinct molecular pathways [16]. Many oncogenic circRNAs exert their effects primarily through miRNA sponging.

Circ_0007386 promotes HCC cell proliferation, migration, invasion, and EMT by sponging miR-507, thereby derepressing cyclin T2 (CCNT2) and accelerating cell-cycle progression [17]. Circ_PIAS1 enhances tumor aggressiveness by sequestering miR-455-3p, leading to up-regulation of nuclear protein 1 (NUPR1) and suppression of ferroptosis [18]. Circ_RanGAP1 promotes tumor growth and metastasis through activation of the miR-27b-3p/NRAS/ERK signaling axis [19].

Circ_TOLLIP, whose expression is regulated by eukaryotic initiation factor 4 A-3 (EIF4A3), facilitates HCC invasion and metastasis by sponging miR-516a-5p and up-regulating PBX3, thereby activating EMT-related pathways [20]. These studies collectively highlight the central role of miRNA-mediated regulation in circRNA-driven oncogenic processes.

Notably, some oncogenic circRNAs function through miRNA-independent mechanisms. Circ_NUP54 interacts with the RBP human antigen R (HuR), promotes its cytoplasmic translocation, stabilizes BIRC3 mRNA, and activates NF-κB signaling, thereby enhancing HCC cell survival and growth [21]. Circ_0000518 exhibits dual regulatory functions: as a ceRNA, it sponges miR-326 to regulate ITGA5 expression, and as an RBP-binding molecule, it binds HuR, stabilizing ITGA5 mRNA, promoting malignant phenotypes and metabolic reprogramming via the Warburg effect [22] (Table 1).

Table 1.

Oncogenic circRNAs in HCC

circRNA	Mechanism	Model / validation level	Effect	Ref.
circ_0007386	Proposed miR-507/CCNT2 axis	HCC cell lines and tissues; functional assays with ceRNA validation; clinical expression association	Promotes proliferation, migration, invasion, and EMT	[17]
circ_PIAS1	Proposed miR-455-3p/NUPR1/FTH1 axis; ferroptosis regulation	Cell lines, animal models, and patient tissues; multi-level functional/mechanistic validation	Enhances tumor aggressiveness and suppresses ferroptosis	[18]
circ_RanGAP1	Proposed miR-27b-3p/NRAS/ERK axis	Cell lines, animal models, and clinical samples; functional and in vivo validation	Promotes tumor growth and metastasis	[19]
circ_TOLLIP	EIF4A3-regulated circRNA; proposed miR-516a-5p/PBX3/EMT pathway	Cell lines, tissues, and animal models; functional, mechanistic, and in vivo validation	Facilitates invasion and metastasis; activates EMT	[20]
circ_NUP54	HuR interaction; BIRC3 mRNA stabilization; NF-kB signaling	Cell lines, tissues, and animal models; RBP-based mechanistic validation	Promotes survival and growth	[21]
circ_0000518	Proposed miR-326/ITGA5 axis; HuR-mediated ITGA5 stabilization	Cell lines and tissues; combined ceRNA and RBP-binding validation	Promotes malignant phenotypes and metabolic reprogramming	[22]

Representative oncogenic circRNAs in HCC, their principal regulatory mechanisms, model or sample sources, validation level, and reported biological effects

Tumor-suppressive circRNAs in HCC

In contrast to oncogenic circRNAs, several circRNAs exert tumor-suppressive effects in HCC by restraining malignant behaviors such as proliferation, invasion, and metastasis [11]. Circ_CSPP1 suppresses HCC progression through the circ_CSPP1/miR-493-5p/HMGB1 axis, and loss of circ_CSPP1 significantly accelerates malignant progression [23]. Circ_ELMOD3 inhibits tumor growth by sponging miR-6864-5p, stabilizing the E3 ubiquitin ligase TRIM13, and activating the p53 signaling pathway [24].

Circ_GPR137B suppresses tumorigenesis and metastasis through a positive feedback loop involving miR-4739 and the m6A demethylase FTO, linking circRNA regulation to epigenetic modification [25]. Beyond miRNA-dependent mechanisms, circ_ATF6 acts as a molecular scaffold to promote calreticulin degradation and suppress Wnt/β-catenin signaling [26]. Circ_PSD3 inhibits vascular invasion and metastasis by sequestering histone deacetylase 1 (HDAC1) and modulating the urokinase-type plasminogen activator system [27].

More recently, circ_VAMP3 has been shown to inhibit HCC proliferation by driving CAPRIN1 phase separation and suppressing c-Myc translation [28]. In HBV-related HCC, viral protein HBx induces m6A-dependent degradation of the tumor-suppressive circRNA cFAM210A, revealing a virus-specific layer of circRNA regulation [29].

Recent studies have further expanded the landscape of circRNA-mediated regulation in HCC, underscoring their functional diversity and context dependence (Table 2).

Table 2.

Tumor-suppressive circRNAs in HCC

circRNA	Mechanism	Model / validation level	Effect	Ref.
circ_CSPP1	Proposed miR-493-5p/HMGB1 axis	Cell lines and tissues; functional assays with ceRNA validation	Suppresses HCC progression	[23]
circ_ELMOD3	miR-6864-5p sponging; TRIM13 stabilization; p53 activation	Cell lines, animal models, and tissues; functional and in vivo validation	Inhibits tumor growth; activates p53	[24]
circ_GPR137B	miR-4739/FTO feedback loop; m6A-related regulation	Cell lines, animal models, and tissues; multi-level mechanistic validation	Suppresses tumorigenesis and metastasis	[25]
circ_ATF6	Scaffold function; calreticulin degradation; Wnt/beta-catenin inhibition	Cell lines, tissues, and animal models; miRNA-independent mechanistic validation	Suppresses Wnt/beta-catenin signaling	[26]
circ_PSD3	HDAC1 sequestration; uPA system modulation	Cell lines, tissues, and animal models; RBP/protein-interaction validation	Inhibits vascular invasion and metastasis	[27]
circ_VAMP3	CAPRIN1 phase separation; suppression of c-Myc translation	Cell lines, tissues, and animal models; validation beyond ceRNA model	Inhibits HCC proliferation	[28]
circ_cFAM210A	HBx-induced m6A-dependent degradation; YBX1-related regulation	HBV-related HCC models and tissues; etiology-specific mechanistic validation	Suppresses HBV-related HCC progression	[29]

Representative tumor-suppressive circRNAs in HCC, their principal regulatory mechanisms, model or sample sources, validation level, and reported biological effects

Mechanistic convergence and regulatory networks

Although individual circRNAs regulate distinct molecular axes, many converge on common oncogenic pathways, including PI3K/AKT, ERK/MAPK, NF-κB, Wnt/β-catenin, EMT, and metabolic signaling networks. While miRNA sponging remains the most frequently reported mechanism, increasing evidence indicates that miRNA-independent functions, such as RBP interaction, epigenetic regulation, and phase separation, play equally important roles in circRNA-mediated regulation in HCC [13, 16]. Representative oncogenic and tumor-suppressive circRNA networks discussed in this review are summarized schematically in Fig. 2.

Fig. 2.

Schematic overview of oncogenic and tumor-suppressive circular RNAs in HCC. The figure summarizes representative oncogenic circRNAs that promote HCC proliferation, EMT, invasion/metastasis, and metabolic reprogramming, as well as tumor-suppressive circRNAs that restrain these malignant phenotypes. Key regulatory modes include ceRNA/miRNA sponging, RNA-binding protein (RBP)-mediated post-transcriptional regulation, and transcriptional/epigenetic control (e.g., m6A-related regulation), collectively shaping downstream oncogenic pathways in HCC

Across the included mechanistic studies, miRNA sponging was the most frequently reported mode of action. However, the evidentiary depth varied widely: many studies relied on bioinformatic prediction and luciferase assays, whereas fewer incorporated endogenous pull-down, RISC-occupancy, stoichiometric analysis, or multi-model validation. Accordingly, these findings should be read as mechanistic hypotheses supported at varying levels of robustness rather than uniform proof of ceRNA function.

circRNAs associated with therapeutic resistance in HCC

Among currently approved targeted agents for HCC, circRNA-related evidence has predominantly focused on sorafenib and, more recently, lenvatinib. These two agents are therefore highlighted in this section, whereas circRNA-mediated mechanisms underlying resistance to other targeted therapies remain largely unexplored.

circRNAs involved in sorafenib resistance

Resistance to systemic therapy, particularly sorafenib, remains a major clinical challenge in advanced HCC. Several circRNAs have been implicated in mediating resistance to targeted agents. Circ_DCBLD2 promotes sorafenib resistance by sponging miR-345-5p and stabilizing TOP2A mRNA, thereby attenuating apoptosis [30]. Circ_RBM23 enhances resistance through modulation of the miR-338-3p/RAB1B axis [31]. Circ_001241 regulates sorafenib resistance via the miR-21-5p/TIMP3 pathway [32]. Circ_TTC13 contributes to resistance by inhibiting ferroptosis through the miR-513a-5p/SLC7A11 axis [33].

Conversely, certain circRNAs enhance sensitivity to targeted therapy. Circ_DDX17 increases sorafenib responsiveness by sponging miR-21-5p, restoring PTEN expression, and suppressing PI3K/AKT signaling, ultimately promoting apoptosis and inhibiting tumor growth [34].

circRNAs associated with lenvatinib response

Lenvatinib is a first-line multikinase inhibitor widely used for advanced HCC, yet the emergence of drug resistance substantially limits its long-term therapeutic benefit. Recent studies indicate that circRNAs contribute to lenvatinib response through multiple regulatory mechanisms.

Exosomal circ_0007132 has been identified as a clinically relevant mediator of lenvatinib resistance. This circRNA is upregulated in serum exosomes from HCC patients with progressive disease after lenvatinib treatment and promotes resistance by stabilizing the RNA-binding protein NONO, thereby enhancing ZEB1 expression and malignant progression [35]. CircRNA-mTOR has also been implicated in lenvatinib resistance by promoting cancer stemness. Elevated circRNA-mTOR expression is associated with poor prognosis in HCC and reduced sensitivity to lenvatinib, primarily through activation of stemness-related signaling pathways [36].

In contrast, certain circRNAs appear to enhance lenvatinib sensitivity. Circ_PIK3C3 is downregulated in lenvatinib-resistant HCC and acts as a competing endogenous RNA to suppress oncogenic signaling, thereby attenuating tumor progression and partially reversing drug resistance [37]. Similarly, circ_CCNY has been shown to enhance lenvatinib efficacy by inhibiting MAPK signaling and modulating the tumor immune microenvironment, including suppression of immune evasion [38]. Collectively, these findings suggest that circRNAs participate in regulating lenvatinib response in HCC through diverse mechanisms, including exosome-mediated communication, regulation of cancer stemness, and modulation of oncogenic and immune-related pathways. Although the current evidence base remains limited compared with that for sorafenib, these findings highlight circRNAs as emerging determinants of lenvatinib sensitivity and resistance that warrant further validation in large clinical cohorts (Table 3).

Table 3.

circRNAs and TKI response

circRNA / therapy	Mechanism	Model type / translational support	Effect	Ref.
circ_DCBLD2 / sorafenib	Proposed miR-345-5p/TOP2A axis	Sorafenib-resistant HCC cell lines; mainly preclinical evidence	Promotes sorafenib resistance	[30]
circ_RBM23 / sorafenib	Proposed miR-338-3p/RAB1B axis	Sorafenib-resistant HCC cell lines; mainly preclinical evidence	Enhances sorafenib resistance	[31]
circ_001241 / sorafenib	Proposed miR-21-5p/TIMP3 pathway	Sorafenib-resistant HCC cell models; mainly preclinical evidence	Regulates sorafenib resistance	[32]
circ_TTC13 / sorafenib	Proposed miR-513a-5p/SLC7A11 axis; ferroptosis inhibition	Sorafenib-resistant cell lines and experimental models; ferroptosis-related preclinical validation	Contributes to sorafenib resistance	[33]
circ_DDX17 / sorafenib	Proposed miR-21-5p/PTEN/PI3K-AKT axis	HCC cell lines and animal models; preclinical evidence with in vivo support	Enhances sorafenib sensitivity	[34]
circ_0007132 / lenvatinib	Exosomal circRNA stabilizes NONO; enhances ZEB1/EMT signaling	Patient serum exosomes plus HCC experimental models; patient-derived exosomal evidence	Promotes lenvatinib resistance	[35]
circRNA-mTOR / lenvatinib	PSIP1/c-Myc axis; stemness regulation	HCC cell lines and experimental models; clinical/prognostic association	Promotes stemness and lenvatinib resistance	[36]
circ_PIK3C3 / lenvatinib	Proposed miR-452-5p/SOX15/Wnt/beta-catenin axis	Lenvatinib-resistant HCC models; mainly preclinical evidence	Enhances lenvatinib sensitivity	[37]
circ_CCNY / lenvatinib	Scaffold for SMURF1-mediated HSP60 degradation; immune evasion modulation	HCC experimental models; preclinical immune-related mechanism	Enhances lenvatinib efficacy	[38]

Representative circRNAs associated with sorafenib and lenvatinib response in HCC, including resistance- and sensitivity-related mechanisms, model type, translational support, and reported effects

circRNAs associated with radiotherapy response in HCC

Radiotherapy is increasingly used as a treatment option for patients with HCC who are not eligible for surgical resection or locoregional ablation, and is often discussed within multidisciplinary treatment strategies for advanced-stage disease, particularly in the context of evolving systemic therapy paradigms [39]. However, the clinical efficacy of radiotherapy varies substantially among patients, largely due to intrinsic tumor heterogeneity and differences in molecular characteristics, as exemplified by aggressive subtypes such as macrotrabecular-massive HCC [40].

Evidence increasingly supports a role for circRNAs in shaping radiotherapy response in HCC by coordinating DNA damage repair, pro-survival signaling, and malignant phenotypes. Circ_LARP1B promotes radioresistance through the miR-578/IGF1R axis, enhancing tumor proliferation and invasion [41]. Circ_ROBO1 reduces radiosensitivity by up-regulating RAD21 via miR-136-5p sponging [42]. Circ_EYA3 stabilizes DTX3L mRNA through interaction with IGF2BP2, thereby attenuating radiation-induced DNA damage [43].

In contrast, circ_0071662 is downregulated in radioresistant HCC and correlates with poor prognosis, suggesting its potential value as a biomarker of radiosensitivity [44]. Circ_NOP14 enhances radiosensitivity by inhibiting Ku70-dependent DNA damage repair, sensitizing HCC cells to both internal and external irradiation [45] (Table 4).

Table 4.

circRNAs and radiotherapy response

circRNA	Mechanism	Model / validation level	Effect	Ref.
circ_LARP1B	Proposed miR-578/IGF1R axis	Radioresistant HCC cell models; mainly preclinical functional validation	Promotes radioresistance; enhances proliferation and invasion	[41]
circ_ROBO1	Proposed miR-136-5p/RAD21 axis	HCC cell models and radiosensitivity assays; preclinical mechanistic validation	Reduces radiosensitivity	[42]
circ_EYA3	IGF2BP2 interaction; DTX3L mRNA stabilization	HCC cell models and radiation-response assays; RBP-based preclinical validation	Attenuates radiation-induced DNA damage	[43]
circ_0071662	Downregulation in radioresistant HCC	Clinical HCC samples; diagnostic/prognostic analysis; biomarker-oriented evidence	Potential biomarker of radiosensitivity	[44]
circ_NOP14	Inhibition of Ku70-dependent DNA repair	HCC cell models and irradiation models; preclinical mechanistic validation	Enhances radiosensitivity	[45]

Representative circRNAs implicated in radioresistance or radiosensitivity in HCC, their proposed mechanisms, model or sample sources, validation level, and reported effects

Discussion

circRNAs as diagnostic and prognostic biomarkers

Owing to their covalently closed circular structure, circRNAs are highly stable and relatively resistant to exonuclease-mediated degradation, which supports their detection in tumor tissues as well as in circulating biospecimens [10–13]. Across the studies included in this review, dysregulated circRNA expression was frequently associated with tumor progression, invasiveness, recurrence-related features, and survival outcomes in HCC, suggesting potential diagnostic and prognostic utility [10–13]. More broadly, recent biomarker studies in oncology and HCC, including work on tumor-progression and immune-microenvironment markers as well as capsular lymphatic vessel density in early-stage HCC, highlight the importance of demonstrating decision-relevant incremental prognostic value before candidate biomarkers are translated into clinical practice [3, 46]. In tissue-based settings, some circRNAs may help distinguish HCC from adjacent non-tumorous liver tissue, whereas in blood-based settings, circulating and exosomal circRNAs have emerged as promising candidates for minimally invasive biomarker development [47, 48].

However, translational enthusiasm should be interpreted in the context of current clinical practice. HCC surveillance in at-risk populations still relies mainly on ultrasound with or without serum alpha-fetoprotein (AFP), and any circRNA-based biomarker strategy must demonstrate clinically meaningful incremental value beyond these established tools rather than merely statistical association [49]. In practical terms, this means future biomarker studies should compare circRNA panels head-to-head against ultrasound ± AFP, prespecify clinically actionable decision thresholds, and evaluate performance across disease stages and etiologic subgroups [49].

Exosomal circRNAs are of particular interest because extracellular vesicles may protect RNA cargo from degradation, enrich tumor-associated molecular signals, and reflect biologically relevant intercellular communication associated with progression and treatment resistance [47, 48]. Nonetheless, it remains premature to conclude that exosomal circRNAs are categorically superior to non-vesicular circulating circRNAs. The optimal analyte may differ according to the intended clinical application, assay workflow, and patient population [47, 48]. At present, most available studies remain retrospective and involve relatively small cohorts, limiting generalizability [47, 48]. Therefore, large multicenter prospective validation studies with harmonized pre-analytical and analytical procedures are essential before circRNA-based biomarkers can be integrated into routine HCC care [47–49].

Therapeutic strategies targeting circRNAs

Beyond their biomarker potential, circRNAs are increasingly recognized as mechanistically actionable nodes within HCC progression, drug resistance, and radiotherapy response networks [16]. Nevertheless, the most realistic near-term translational role of circRNA-targeted therapy is likely to be as an adjunctive or sensitization strategy rather than as a standalone treatment modality.

A first plausible route is the inhibition of oncogenic circRNAs that drive resistance to systemic therapy or radiotherapy. In principle, junction-spanning antisense oligonucleotides (ASOs) or siRNAs directed against the back-splice junction may offer greater specificity than approaches targeting shared linear transcript sequences, because they exploit the unique circular junction [16]. However, such specificity cannot be assumed and requires rigorous experimental confirmation. Candidate therapeutics should demonstrate selective target engagement, minimal interference with host linear transcripts, and reproducible phenotypic rescue in relevant HCC models. From a clinical perspective, the most meaningful endpoint may be re-sensitization to established therapies—such as restoration of apoptosis, ferroptosis, or radiation-induced vulnerability—rather than complete tumor suppression by circRNA targeting alone [30, 33, 34, 45].

A second major consideration is delivery. Many RNA-based therapeutic concepts fail at the level of in vivo delivery, but HCC presents a potentially favorable translational setting because liver-directed uptake and locoregional treatment approaches are already part of routine clinical oncology practice [39, 47, 50]. This suggests that successful circRNA-targeting strategies should be developed with delivery feasibility in mind from the outset, including liver-targeted formulations or locoregional administration strategies compatible with interventional or radiation oncology workflows [39, 47, 50]. In this context, demonstration of tumor exposure, on-target engagement, and acceptable immune and off-target safety profiles should be treated as core development milestones rather than secondary considerations [50, 51].

A third emerging opportunity lies in intercepting exosome-mediated resistance signaling. For circRNAs implicated in resistance phenotypes transmitted through exosomal communication, it may be feasible to target the circRNA cargo itself or disrupt its downstream stabilization axis in recipient cells [35, 47]. This concept is particularly attractive because it links a clinically accessible biospecimen compartment, such as serum exosomes, to a longitudinally monitorable therapeutic mechanism [35, 47]. Even so, such approaches remain early-stage and will require careful validation of biological specificity, pharmacodynamic readouts, and compatibility with combination treatment paradigms commonly used in advanced HCC [35, 51].

Overall, therapeutic development in this field should move away from broad claims that circRNAs are inherently “druggable” and instead focus on a smaller number of mechanistically credible candidates supported by robust junction specificity, feasible delivery, and clinically relevant combination rationale [16, 50, 51].

Importantly, the key translational limitation is not only delivery but also external validity. Most resistance studies used stepwise drug-selected cell lines that approximate adaptive tolerance under simplified conditions; they do not fully reproduce acquired resistance in patients, where pharmacokinetic pressure, stromal remodeling, immune effects, and intratumoral heterogeneity all shape response. Patient-derived organoids, patient-derived xenografts, and longitudinal clinical cohorts should therefore be prioritized to test whether candidate circRNAs track clinically acquired resistance and time-to-progression.

Current challenges in clinical translation

Despite rapid progress in circRNA research, several major barriers continue to limit clinical translation in HCC. One central challenge is delivery. Similar to other RNA-based therapeutic platforms, circRNA-targeting approaches face persistent problems related to off-target effects, immune activation, limited in vivo persistence, and insufficient tissue specificity [50, 51]. For this reason, delivery should be considered a gatekeeping translational criterion, with priority given to candidates that can be paired with liver-feasible delivery systems and that demonstrate target engagement in vivo before broader mechanistic or therapeutic claims are advanced [47, 50].

A second challenge is the highly context-dependent nature of circRNA biology. The functional role of a circRNA may differ according to molecular subtype, disease stage, etiologic background, treatment exposure, or tumor microenvironment [13, 16]. This complexity reduces the likelihood that a single circRNA signature will function as a universal biomarker or therapeutic target across all HCC settings. It also raises concern that modulation of a given circRNA could have variable or unintended consequences depending on host-gene context and regulatory network interactions [13, 16]. Future studies should therefore anchor candidate circRNAs to clearly defined clinical contexts and validate them in stratified cohorts rather than presenting them as broadly generalizable markers.

A third major limitation is methodological heterogeneity. Most available studies remain preclinical, single-center, retrospective, or exploratory in design. Detection platforms, normalization strategies, sample types, processing conditions, and reporting methods are often insufficiently standardized, making it difficult to compare results across studies or judge reproducibility [47, 48]. For biomarker development, harmonization of pre-analytical variables, assay performance characteristics, and reporting standards will be essential. Without such standardization, promising circRNA signatures may continue to generate associative signals without progressing toward clinically deployable tests [47, 48].

Etiologic heterogeneity presents an additional translational obstacle of particular importance in HCC. HBV-related, HCV-related, and NAFLD-related HCC differ in inflammatory milieu, fibrosis background, genomic context, and prior treatment exposure, all of which may influence circRNA expression and biomarker performance [8, 49]. As a result, etiology-specific circRNA models may prove more clinically useful than a single universal panel. Prospective multicenter studies should therefore prespecify stratified analyses, calibration strategies, or subgroup-specific performance assessment across major etiologic categories [8, 49].

Mechanistic caution is also warranted for the large fraction of studies that infer ceRNA activity from bioinformatic prediction and reporter assays. In many cases, the evidence chain stops short of demonstrating endogenous binding, RISC engagement, stoichiometric sufficiency, or competition with the native miRNA pool. The present review therefore treats miRNA-sponging interpretations as provisional unless supported by higher-order validation in tissue, animal, or patient-derived models. Future studies should report whether endogenous circRNA abundance is sufficient to sequester biologically meaningful fractions of the relevant miRNA pool and should distinguish genuine functional sponging from bioinformatics-inferred correlational associations.

Taken together, the field now needs fewer descriptive reports of newly identified circRNAs and more translationally disciplined studies that address reproducibility, specificity, delivery feasibility, and decision-level clinical utility [16, 47–51]. Multicenter prospective protocols with prespecified thresholds, quality-controlled workflows, and clinically relevant endpoints are likely to provide the most efficient path from retrospective association to meaningful application [47–49].

Limitations of this review

Several limitations should be acknowledged. First, the protocol was not prospectively registered, which may increase the possibility of post hoc methodological decisions. To mitigate this limitation, the revised manuscript provides complete reproducible search strings, a PRISMA 2020 checklist, an in-text flow diagram, explicit eligibility criteria, and a transparent account of the qualitative synthesis process. Nevertheless, non-registration should be considered when interpreting the certainty of the review conclusions.

Second, the database search was limited to PubMed and supplemented by manual screening of reference lists from eligible studies and relevant reviews. Although this strategy captured a focused body of PubMed-indexed circRNA-HCC literature, relevant studies indexed only in Web of Science, Scopus, Embase, or other bibliographic databases may have been missed.

Third, the evidence base was too heterogeneous for quantitative meta-analysis or conventional pooled GRADE assessment. The included studies differed substantially in sample type, experimental model, molecular endpoint, circRNA-detection platform, therapeutic context, and clinical annotation. Therefore, the review used a structured qualitative evidence appraisal to distinguish exploratory cell-line findings from animal-supported, patient-supported, or multi-model validated findings. This approach improves interpretability but cannot fully substitute for standardized risk-of-bias assessment across future primary studies.

Fourth, many included mechanistic studies relied on overexpression systems and proposed miRNA-sponging mechanisms without demonstrating endogenous RISC engagement or stoichiometric feasibility. Similarly, most drug-resistance studies were based on stepwise-selected resistant cell lines, which do not fully reproduce clinically acquired resistance shaped by pharmacokinetic pressure, tumor microenvironmental remodeling, immune contexture, and intratumoral heterogeneity. These limitations reinforce the need for patient-derived organoids, patient-derived xenografts, longitudinal clinical cohorts, and prespecified biomarker-performance analyses before circRNA-based applications can be translated into clinical decision-making.

Conclusions

CircRNAs have emerged as key regulators of HCC progression, therapeutic resistance, and radiosensitivity through diverse mechanisms that include miRNA sequestration, RNA-binding protein interaction, and epigenetic control. The available evidence synthesized in this review supports their relevance across malignant phenotypes, targeted-therapy response, and DNA-damage repair–associated radiosensitivity. However, the translational literature remains dominated by preclinical studies and small retrospective cohorts, which limits immediate clinical generalizability.

Future progress will depend less on cataloging additional circRNAs and more on advancing a smaller set of mechanistically credible candidates through translational proof points. In particular, the field should prioritize whether circRNA-based models improve real clinical decisions, such as surveillance or treatment selection, and whether circRNA-directed interventions can achieve tumor-relevant exposure with acceptable safety. Mechanistic studies should treat endogenous interaction evidence, physiologically plausible rescue design, and stoichiometric feasibility of the ceRNA mechanism as minimum validation criteria rather than optional confirmatory experiments. Among the currently described candidates, circ_0007132, circ_0071662, circ_DCBLD2, and circ_NOP14 appear especially relevant for follow-up because they connect to patient-supported or clinically actionable phenotypes (lenvatinib resistance, radiotherapy response, sorafenib resistance, and radiosensitivity). Because circRNA expression may vary across HBV-, HCV-, and NAFLD-related HCC, prospective validation should determine whether etiology-specific models outperform pooled signatures.

Ultimately, circRNA research in HCC will have clinical impact only if it changes management—either by improving risk-stratified surveillance beyond ultrasound ± AFP or by enabling add-on interventions that measurably resensitize tumors to TKIs or radiotherapy. Demonstration of decision impact, etiologic robustness, reproducible assay performance, and feasible delivery/safety in multicenter prospective settings will be essential before circRNA programs can move into guideline-facing evaluation and later-phase development.

Abbreviations

AFP

Alpha-fetoprotein

ASO

Antisense oligonucleotide

ceRNA

Competing endogenous RNA

circRNA

Circular RNA

EMT

Epithelial-mesenchymal transition

GRADE

Grading of Recommendations Assessment, Development and Evaluation

HBV

Hepatitis B virus

HCC

Hepatocellular carcinoma

HCV

Hepatitis C virus

LLM

Large language model

m6A

N6-methyladenosine

NAFLD

Non-alcoholic fatty liver disease

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

RBP

RNA-binding protein

RISC

RNA-induced silencing complex

siRNA

Small interfering RNA

TKI

Tyrosine kinase inhibitor

Authors’ contributions

Beining ZHANG conceptualized the review and drafted the manuscript. Ninggang ZHENG conducted the literature search and assisted in manuscript writing and revision. Jiangye WANG and Guoliang SUN extracted and synthesized evidence and prepared the tables/figure. Cheng HUANG supervised the study, critically revised the manuscript, and approved the final version. All authors read and approved the final manuscript.

Funding

This work was supported by the Gansu Provincial People’s Hospital In-Hospital Fund (No.23GSSYA-17) and the Natural Science Foundation of Gansu Province (No.26JRRA760). The funders had no role in the design of the review, data collection, analysis, interpretation, manuscript preparation, or decision to submit the manuscript for publication.

Data availability

All data generated or analysed during this study are included in this published article and its supplementary information files.

Declarations

Ethics approval and consent to participate

Not applicable. This systematic review did not involve new studies of human participants, human data, human tissue, or animals performed by the authors.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.