Enhancing Circular RNA Translation Efficiency Through Dual Internal Ribosome Entry Sites

TL;DR

Researchers improved the translation efficiency of circular RNA by using two IRES elements from the EMCV family, enhancing its potential for drug delivery.

Contribution

A novel dual-IRES strategy using EMCV-derived IRES elements to boost circRNA translation efficiency is introduced.

Findings

Translation efficiency is significantly improved with dual EMCV-derived IRES elements.

EMCV IRESs at both 5′ and 3′ ends of the CDS work cooperatively to enhance expression.

The strategy is compatible with multiple coding sequences.

Abstract

Circular RNAs are stable mRNA molecules that do not require expensive chemical modifications, but their application is limited by low translational efficiency because they lack a 5′ cap and rely on IRESs for translation. In this study, we designed a dual-IRES strategy to enhance circRNA translation and tested this approach in 293T cells. Our results show that translation efficiency is significantly improved only when both the 5′ and 3′ IRESs are derived from the EMCV family. This suggests that EMCV IRESs possess structural features that allow effective cooperation in a dual-IRES configuration.

Circular RNA (circRNA) has emerged as a promising vector for drug delivery because, unlike linear mRNA, it does not require costly chemical modifications and offers greater stability and sustained expression in cells. Lacking the canonical 5′ cap structure, circRNA relies primarily on internal ribosome entry sites (IRES) to initiate translation, but IRES-mediated initiation is less efficient than cap-dependent translation. To overcome this limitation, we devised a dual-IRES strategy that introduces a second IRES to drive translation of the coding sequence (CDS). By testing several IRES elements known for high translational activity, this study shows that IRESs derived from the EMCV (Encephalomyocarditis virus) family can enhance expression when placed at the 3′ of the CDS, in coordination with the 5′ EMCV-derived IRES. The optimal dual-IRES combinations identified in this study display compatibility with two different coding sequences, offering a useful strategy to enhance circRNA translation.

Figures

Full text

1. Introduction

During the COVID-19 pandemic, mRNA vaccines demonstrated their potential in epidemic control, propelling them to the forefront of molecular medicine [1]. RNA delivery systems have shown intrinsic safety, high adaptability, and broad applications, permitting direct protein synthesis without integration into the host genome [2]. Owing to this favorable safety profile and versatility, mRNA platforms have become powerful tools for vaccines, protein-replacement therapies, and gene-regulation strategies [3,4,5]. Nevertheless, canonical linear mRNAs are susceptible to exonuclease-mediated degradation, resulting in relatively short intracellular half-lives [6]. In contrast, covalently closed circRNAs lack free ends susceptible to exonuclease attack, thereby exhibiting enhanced stability and prolonged persistence in cells [7,8]. Consequently, circRNAs enable sustained protein expression and may reduce dosing frequency. In addition, circRNA production avoids 5′ cap analogs, modified nucleotides such as pseudouridine (Ψ), and a defined poly(A) tail [9]. This further reduces manufacturing complexity and cost. Collectively, these attributes position circRNAs as an emerging and highly promising modality within the expanding landscape of RNA therapeutics.

Although the head-to-tail configuration of circular RNAs shields them from exonuclease degradation [10], it also eliminates the 5′ cap required for efficient cap-dependent translation. To circumvent this limitation, IRES elements have been incorporated to drive cap-independent translation of circRNAs. IRES elements are structured RNA motifs that recruit ribosomes and initiate translation [11], and recent screens have identified several IRES elements with notably high translational activity [12]. However, unlike canonical cap-dependent initiation, IRES-mediated translation often bypasses certain initiation factors [13], which typically reduces its efficiency to approximately one-fifth to one-half of that achieved by cap-dependent mechanisms [14]. This lower efficiency remains a major obstacle to the therapeutic use of circular RNAs. Therefore, improving IRES-mediated translation in circRNA remains a major challenge for realizing its therapeutic potential.

This study aimed to enhance translation from circular RNAs by introducing a second IRES downstream of the CDS and upstream of the 3′ untranslated region (UTR) to support the 5′ IRES. Several commonly used viral IRES elements were selected, including EMCV (Type II) and Type I IRESs derived from or closely related to enteroviruses, such as EV-A71, HRV, CVB3 [15], and the engineered iEV-B107 variant, which shares sequence and functional characteristics with enterovirus IRESs. These IRES elements were first evaluated for their ability to mediate EGFP expression in circular RNA constructs. We then generated pairwise combinations of these IRESs to screen for dual-IRES configurations capable of boosting protein output. Finally, we assessed the general applicability of the identified configuration across different protein-coding sequences, resulting in markedly higher expression levels compared with previously reported high-efficiency IRES elements. Taken together, these findings provide a framework for the rational design of dual-IRES circRNAs with enhanced translational efficiency.

2. Materials and Methods

2.1. Sample Preparation

2.1.1. Molecular Cloning

The plasmids used in this study are listed in Table 1. The plasmid backbone used in this study was pUC-SPK, a commercial backbone from Sangon Biotech (Shanghai, China) derived from pUC-57 and modified by Sangon internally. The IRES-containing plasmids were custom-ordered from Sangon Biotech, and all subsequent recombinant plasmids used in this study were constructed based on this pUC-SPK backbone. Plasmids were assembled using the Multi Seamless Cloning Kit (Sangon Biotech, Shanghai, China) according to the manufacturer’s instructions. All PCR reactions were performed using FidCycle High-Fidelity DNA Polymerase (Sangon Biotech, Shanghai, China), and PCR products were purified using Hieff NGS^®^ DNA Selection Beads (Yeasen, Shanghai, China). All recombinant plasmids were verified by Sanger sequencing.

2.1.2. Generation of dsDNA Templates

A sequence-verified bacterial clone was cultured in LB medium supplemented with kanamycin for 16 h. Plasmid DNA was extracted using the E.Z.N.A.^®^ Endo-Free Plasmid DNA Midi Kit (Omega Bio-Tek, Norcross, GA, USA; Cat. No. D6904-03). The plasmid was linearized with XhoI (Thermo Fisher Scientific, Waltham, MA, USA), followed by precipitation with 0.7 volumes of isopropanol. The mixture was incubated at −20 °C and then centrifuged at 12,000 rpm for 15 min at 4 °C. The DNA pellet was washed twice with ice-cold 75% ethanol, air-dried, and dissolved in nuclease-free water. The concentration of the DNA template was determined using a NanoDrop spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA), and its purity was assessed by agarose gel electrophoresis.

2.1.3. RNA Preparation

In vitro transcription (IVT). IVT was performed in a 20 µL reaction mixture containing 0.5 µg of linearized dsDNA template, 40 mM Tris–HCl (pH 8.0), 35 mM MgCl_2_, 17 mM DTT, 1.5 mM spermidine, 8.75 mM NTPs, 0.05 mU/µL pyrophosphatase (PPase), 1 U/µL RNase inhibitor (RI), and 10 U/µL T7 RNA polymerase (all from Sangon Biotech, Shanghai, China). The reaction buffer was prepared in-house according to standard IVT formulations and filtered through a 0.22 µm membrane before use. Reactions were incubated at 37 °C for 3 h with a heated lid.

RNA purification. Following transcription, 80 µL of nuclease-free water was added to each reaction, and RNA was precipitated with 7.5 M LiCl at −20 °C for 30 min. The RNA was collected by centrifugation at 12,000 rpm for 15 min at 4 °C, washed twice with 1 mL of ice-cold 75% ethanol (centrifuged at 12,000 rpm for 5 min each at 4 °C), air-dried, and dissolved in nuclease-free water.

Circularization. For circular RNA generation, we employed the PIE strategy, which utilizes a self-splicing group I intron rearranged to flank the RNA sequence of interest, allowing efficient intramolecular ligation and formation of circular RNA in vitro. To enhance circularization efficiency, RNA was denatured at 70 °C for 5 min and immediately placed on ice for at least 2 min. Circularization was carried out at 55 °C for 20 min in a buffer containing 10 mM MgCl_2_, 50 mM Tris–HCl (pH 7.5), 1 mM DTT, and 2 mM GTP. The resulting RNA was purified using the RNeasy Plus Kit (Qiagen, Hilden, Germany; Cat. No. 74104).

RNase R digestion. To remove residual linear RNA, purified RNA was treated with RNase R (Beyotime Biotechnology, Shanghai, China) at a ratio of 1 U enzyme per µg RNA and incubated at 37 °C for 15 min. The reaction was terminated by heating at 70 °C for 10 min to inactivate the enzyme, and the RNA was further purified using the RNeasy Plus Kit for subsequent transfection experiments.

2.2. Cell Culture and Transfection

HEK293T cells were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China, accession number SCSP-502) and cultured in DMEM supplemented with 10% fetal bovine serum (FBS) at 37 °C in a humidified atmosphere containing 5% CO_2_. For routine passaging, cells were dissociated using 0.25% trypsin and neutralized with complete growth medium.

For transfection, cells were seeded into 24-well or 96-well plates at densities of 2.5 × 10^4^ or 4.0 × 10^3^ cells per well, respectively. In 24-well plates, 0.75 µL of Lipofectamine 3000 (Thermo Fisher Scientific, Waltham, MA, USA; Cat. No. L3000015) per well was used; in 96-well plates, 0.126 µL per well was applied. To minimize pipetting errors associated with very small volumes, the Lipofectamine solution was prepared as a master mix prior to combining with RNA samples. Typically, 300 ng of EGFP circRNA was transfected per well in 24-well plates, and 0.015 ng of nanoLuc circRNA per well in 96-well plates. RNA and transfection reagent were gently mixed and added dropwise to each well. Transfected cells were incubated for 48 h without changing the culture medium.

2.3. In Vitro NanoLuc Assay

At 48 h post-transfection with nanoLuc circRNA, the 96-well plates were equilibrated at room temperature for 15 min. Nano-Glo^®^ Luciferase Assay Reagent (Promega, Madison, WI, USA; Cat. No. N1110) was prepared according to the manufacturer’s instructions, and 100 µL was added to each well. The plates were gently shaken in the dark for 3 min to allow complete cell lysis and signal stabilization. Subsequently, 40 µL of the resulting lysate from each well was transferred to a white 96-well microtiter plate (Sangon Biotech, Shanghai, China). To minimize signal crosstalk, wells were arranged with empty wells in between. Bioluminescence was measured using a multimode microplate reader (Thermo Fisher Scientific, Waltham, MA, USA).

2.4. Microscope Imaging

For imaging, HEK293T cells were seeded in 24-well plates at a density of 4.5 × 10^3^ cells per well. Fluorescence images were acquired at 24 and 48 h post-transfection using a Leica fluorescence microscope (Leica Microsystems, Wetzlar, Germany) equipped with appropriate excitation light sources and filter sets and a 10× objective. EGFP fluorescence was detected using an excitation wavelength of 488 nm and an emission wavelength of 509 nm. Bright-field and EGFP fluorescence images were merged to visualize protein expression levels.

2.5. Quantification of EGFP Expression

EGFP expression levels were quantified using the Abcam GFP Quantification Kit (Abcam, Cambridge, UK; Cat. No. ab235672) according to the manufacturer’s instructions. Briefly, GFP standard protein was dissolved in 100 µL of GFP Assay Buffer to obtain a 1 µg/µL stock solution, which was further diluted to generate a 10 ng/µL working solution. Standard curves were generated using serial dilutions corresponding to 0, 40, 80, 120, 160, and 200 ng EGFP per well. Fluorescence signals were background-subtracted using assay buffer alone (0 ng/well) and used to construct the standard curve.

For sample preparation, culture medium was gently aspirated, and cells were washed twice with PBS. Assay buffer was added (120 µL per well for 24-well plates), and plates were incubated on ice for 10 min to ensure complete cell lysis. Lysates were collected, centrifuged at 12,000 rpm for 5 min at 4 °C, and 50 µL of the supernatant was transferred to a black, opaque 96-well plate. Fluorescence was measured at Ex/Em = 488/510 nm using a microplate reader. EGFP amounts were multiplied by a factor of 2.4 to account for the ratio between the total lysate volume and the measured volume, yielding total EGFP expression per well.

3. Results

3.1. Construction and Preparation of Circular RNAs with Single or Dual IRES Elements

We constructed two pUC-SPK vectors carrying either a single or dual IRES-driven protein expression module. Each module is embedded within the Anabaena Permuted Intron–Exon (PIE) system and placed under the control of a T7 promoter (Figure 1A). Following linearization with the XhoI restriction enzyme, the resulting linear templates are approximately 6 kb in length, consistent with the expected size (Figure 1B). In the agarose gel, lane 1 shows the uncut single IRES plasmid, while lane 2 shows the same plasmid after XhoI digestion. Lanes 3–4 and 5–6 show two different dual-IRES plasmids before and after linearization. These linearized templates were then used for IVT, yielding RNA bands predominantly around 2 kb in size. The RNA transcribed from the dual IRES templates (lanes 2–3) migrates more slowly than that from the single IRES template (lane 1), consistent with their increased sequence length (Figure 1C). Capillary electrophoresis analysis was performed to assess the efficiency of circularization (Figure 1D). Following circularization, the precursor RNA peaks are eliminated, indicating effective conversion of linear transcripts to circular RNA. Concurrently, new peaks appear at positions preceding the former precursor peaks, representing the circularized RNA products.

In addition, two closely spaced short RNA species of approximately 100 nt were observed after circularization. These species represent excised introns released during the circularization process and are indicated by arrows and shown at higher magnification in the inset. Based on the relative peak areas, the circularization efficiency of the single-IRES construct was estimated to be approximately 85%. Circularization efficiency of a dual-IRES construct was comparable (Supplementary Figure S1). Subsequent RNase R treatment removes residual linear RNA species, including intron fragments and precursor RNA, resulting in highly purified circular RNA products (Figure 1E). The resulting circular RNA samples were then used for transfection to evaluate their translational efficiency. Fluorescence microscopy and merged images indicate that the EV-A71 (Enterovirus A71) IRES produces the strongest EGFP signal, followed by CVB3 (Coxsackievirus B3), iEV-B107 (Enterovirus B107), and HRV (Human Rhinovirus). In contrast, among the EMCV-derived IRES variants, EMCV-6A mediates weak but detectable EGFP expression, whereas EMCV-ΔATG11 and EMCV-7A produce minimal to undetectable signals, confirming differential translational activities among the tested circRNA constructs (Figure 1F).

3.2. Screening of Synergistically Functional Dual-IRES Combinations

To assess whether combining two IRES elements can enhance circular RNA expression compared with single-IRES constructs, we generated a matrix of dual-IRES circular RNAs by pairing various known IRES elements. Contrary to our initial expectation, the six dual-IRES constructs (EV-A71+EV-A71, iEV-B107+HRV, HRV+iEV-B107, EV-A71+EMCV-ΔATG11, EMCV-ΔATG11+EV-A71, and EMCV-ΔATG11+CVB3; bars 15–20) showed lower EGFP expression than the single-IRES circular RNAs (EMCV-ΔATG11, EMCV-6A, EMCV-7A, iEV-B107, HRV, CVB3, and EV-A71; bars 1–7) (Figure 2D). Among these combinations, EV-A71+EMCV-ΔATG11 exhibited the highest EGFP expression, followed by iEV-B107+HRV, HRV+iEV-B107, and dual EV-A71. In contrast, EMCV-ΔATG11+EV-A71 and EMCV-ΔATG11+CVB3 showed minimal fluorescence (Figure 2A, bars 15–20 in Figure 2D).

Our quantitative analysis revealed that, when the 5′ IRES is fixed, the choice of the 3′ IRES influences expression. When EV-A71 occupies the 5′ position, the EV-A71+EMCV-ΔATG11 construct showed approximately 2.7-fold higher expression than the dual EV-A71 construct (bars 18, 15 in Figure 2D), indicating that EMCV-ΔATG11 outperforms EV-A71 as a 3′ IRES in dual-IRES configurations.

Likewise, upon swapping the IRES (5′ IRES to 3′ IRES and vice versa), the EV-A71+EMCV-ΔATG11 construct showed approximately 4.2-fold higher expression than EMCV-ΔATG11+EV-A71, while iEV-B107+HRV showed approximately 1.5-fold higher expression than HRV+iEV-B107 (bars 18, 19, 16, and 17 in Figure 2D).

Since EMCV-ΔATG11 performs better as a 3′ IRES in our dual-IRES constructs, we generated circular RNAs containing two EMCV-ΔATG11 IRES elements to evaluate whether a dual-EMCV-ΔATG11 construct could further enhance expression. Fluorescence imaging showed that the dual-EMCV-ΔATG11 construct produces higher signal intensity than the single EMCV-ΔATG11 construct (Figure 2B). Consistent with these observations, protein quantification revealed approx. 2.8-fold higher expression for the dual-EMCV-ΔATG11 construct than for the single EMCV-ΔATG11 construct, based on the average of paired measurements (bars 1 and 8 in Figure 2D). Although the signal remained weaker than that of the HRV construct (bar 5), dual-EMCV-ΔATG11 circular RNAs clearly showed increased protein expression relative to the single EMCV-ΔATG11 construct.

Comparison among EMCV-ΔA TG11, EMCV-6A, and EMCV-7A shows that the IRES variant EMCV-ΔATG11 has truncations at both ends relative to EMCV-6A and EMCV-7A, while the core region is identical among all three variants. EMCV-7A contains an extra A at the A-bulge (Figure 2E) compared to EMCV-ΔATG11 and EMCV-6A. Although the three EMCV variants share a highly conserved core sequence, they exhibit distinct levels of translational output.

Consistent with the pattern of the single-IRES EMCV variants (Figure 2D, bars 1–3), combinations with EMCV-6A at the 5′ position gave superior expression among the EMCV-derived dual-IRES constructs. As the most efficient IRES within the EMCV family, all EMCV-6A–leading combinations (EMCV-6A+EMCV-ΔATG11, EMCV-6A+EMCV-6A, and EMCV-6A+EMCV-7A) resulted in higher expression levels than combinations led by EMCV-ΔATG11 or EMCV-7A (Figure 2B). All EMCV-6A—leading combinations exhibited expression levels comparable to single CVB3 IRES and consistently higher than those of HRV or iEV-B107 IRES constructs. In terms of relative increase, all EMCV-6A—leading combinations reached expression levels approximately 2-fold higher than that of the single EMCV-6A construct. Similarly, the EMCV-7A+EMCV-7A construct reached an expression level about 2-fold higher than that of single EMCV-7A (bars 3, 11, 12, and 13; bars 2, 9, and 10 in Figure 2D).

To further optimize the dual-EMCV construct, domain I of the 3′ EMCV IRES was deleted (Figure 2E). Fluorescence microscopy indicated that this deletion does not negatively affect expression (Figure 2B). In 24-well plate cultures, this construct produced an average protein yield of over 1 µg per well, with an expression level approximately 3.7-fold higher than that of the single EMCV-6A IRES and comparable to the highest-expressing EV-A71 IRES (bars 3, 14 in Figure 2D). To examine whether the dual-IRES configuration affects the functional stability of circRNAs, we performed a time-course analysis of EGFP expression comparing EV-A71 single-IRES and EMCV-6A+EMCV-ΔDI dual-IRES constructs. Comparable expression persistence was observed up to 96 h post-transfection, indicating that the dual-IRES design does not compromise circRNA stability (Figure S2).

3.3. The Optimized Dual IRES Configuration Achieves Maximal Nanoluc Efficiency

To assess whether the optimized dual-EMCV IRES configuration can accommodate other coding sequences, we evaluated its performance using a nanoLuc reporter assay. A series of optimized dual-IRES circular RNA constructs, primarily consisting of EMCV-6A—leading combinations (Figure 3, bars 8–11; EMCV-6A+EMCV-7A, EMCV-6A+EMCV-ΔATG11, EMCV-6A+EMCV-6A, and EMCV-6A+EMCV-ΔDI), together with a dual EMCV-ΔATG11 construct (bar 7; EMCV-ΔATG11+EMCV-ΔATG11), were compared with single-IRES circular RNAs (bars 1–6; EMCV-ΔATG11, EMCV-6A, HRV, EV-A71, CVB3, and iEV-B107), with all fold-over-mock values summarized in Table 2. Upon replacing EGFP with NanoLuc, the translational performance of single-IRES—driven circular RNAs differed from that observed with EGFP. Specifically, iEV-B107, rather than EV-A71, exhibited the highest translational activity, followed by CVB3, EMCV-6A, EV-A71, and HRV (Figure 3). Despite this shift, dual-EMCV configurations consistently maintained strong translational performance. RNAfold predictions suggested that potential base-pairing may exist between the IRES and the CDS, and that the extent of such base-pairing can vary depending on the combination of IRES and CDS (Supplementary Figure S3). Similarly to the results obtained with EGFP constructs, the dual EMCV-ΔATG11 construct exhibited lower expression than high-performance single-IRES elements; however, it reached an expression level approximately ninefold higher than that of the single EMCV-ΔATG11 construct (bars 7 and 1 in Figure 3).

Across all tested constructs, EMCV-6A—leading dual-IRES configurations outperformed all single-IRES designs. Among these, the EMCV-6A+EMCV-ΔDI configuration reached an expression level approximately 2.1-fold higher than that of the iEV-B107 single-IRES construct (bars 11, 6 in Figure 3). Consistent with the pattern observed in the EGFP reporter system, all EMCV-6A–leading dual-IRES combinations reached expression levels approximately twofold higher than that of the single EMCV-6A construct (bars 8–11 and bar 2 in Figure 3).

Together, these results demonstrate that the optimized dual-EMCV IRES configuration substantially enhances translational efficiency of circular RNAs and remains effective in the context of different coding sequences, outperforming several commonly used high-expression single-IRES elements.

4. Discussion

Despite the growing interest in circular RNAs as alternatives to linear transcripts, improving their translational efficiency remains a key unresolved challenge. To address this limitation, we evaluated a dual-IRES strategy aimed at enhancing circRNA translation. Our findings indicate that only a subset of IRES combinations function productively in a dual configuration, and compatible pairings support higher protein output than single-IRES constructs. Consistent with previous reports [16], high-activity IRESs do not necessarily cooperate when combined, and certain pairings can even impair translation efficiency. In our system, combinations involving EMCV-family IRESs placed at both the 5′ and 3′ ends of the CDS showed reproducibly enhanced translation, suggesting that the EMCV family may possess structural or functional features that are more compatible with cooperative initiation in a dual-IRES context. This indicates that enhancing circRNA translation is not simply a matter of increasing initiation sites, but instead requires careful matching of the 5′ and 3′ IRES elements.

To further explore the mechanistic basis of dual-IRES compatibility, we considered how intrinsic differences between Type I and Type II IRES elements may influence their performance in a dual configuration. Type I IRESs, such as those derived from HRV, EV-A71, and CVB3 [18], rely on multiple internal ribosome entry site trans-acting factors (ITAFs), including polypyrimidine tract-binding protein (PTB) and poly(rC)-binding protein 2 (PCBP2), to achieve the conformational states required for efficient initiation [19,20,21]. In contrast, the EMCV Type II IRES exhibits much lower ITAF dependence and can recruit ribosomes through a more direct interaction with the translational machinery [22,23]. These distinctions suggest that combining highly active IRESs does not necessarily yield improved translation, potentially due to incompatibilities in structural requirements and cofactor dependencies. When two ITAF-dependent IRESs flank the coding sequence, their concurrent demand for the same cofactors may generate competitive pressure. This competition could be amplified by the fact that ITAFs are present at abundances hundreds to thousands of times lower than ribosomes in mammalian cells [24,25,26], which may hinder proper folding of one or both IRES elements and reduce overall initiation efficiency. In comparison, EMCV-derived IRESs, which rely less on cellular ITAFs, may be more tolerant in dual configurations. Moreover, the 3′ EMCV IRES could potentially increase the local concentration of ribosomes, thereby reducing the time required for ribosome recruitment by the 5′ IRES. This mechanistic model provides a plausible explanation for the relative compatibility of EMCV-based pairings and aligns with established distinctions between Type I and Type II IRES function, although further studies are needed to validate these hypotheses.

Based on our experimental observations, several principles emerge that govern the performance of dual-IRES circRNA systems. First, the activity of the 5′ IRES sets the upper limit of overall translational output. Combinations beginning with a high-activity IRES, such as EV-A71+EMCV-ΔATG11 and iEV-B107+HRV, maintain higher expression levels than constructs initiated by weaker IRESs, even when 3′ IRESs exert inhibitory effects. This trend is also observed in EMCV-family constructs, where 5′ EMCV-6A consistently drives higher translation than 5′ EMCV-ΔATG11 or EMCV-7A (Figure 2). Second, placing an EMCV-family IRES at the 3′ site improves compatibility and preserves translation efficiency. For example, for dual-IRES constructs containing EV-A71 at the 5′ position, placing EMCV-ΔATG11 at the 3′ position (Figure 2D, bar 18) results in a smaller reduction in expression than placing EV-A71 at the 3′ position (Figure 2D, bar 15). Moreover, the strong performance of dual EMCV-ΔATG11 constructs further supports that EMCV elements tolerate dual-IRES configurations better than ITAF-dependent Type I elements. This observation aligns with the notion that EMCV IRESs rely minimally on ITAFs, introducing less competition with the 5′ Type I IRES and thereby ensuring proper function of both IRES elements. Third, the apparent ranking of individual IRESs can shift when the coding sequence is exchanged, indicating that IRES activity is influenced by local sequence context. This context dependence was evident in our observation that the relative strengths of EV-A71 and iEV-B107 changed when the CDS was replaced. We propose that these changes may be due to the unique head-to-tail topology of circular RNAs. Based on the observations described above, the closed conformation of circular RNAs could allow non-specific base-pairing between the IRES and the CDS, which may interfere with IRES function and generate CDS-dependent preferences [27]. These context-dependent effects may explain why the relative performance of certain IRESs differs between the two reporter genes. Consistent with these predictions, RNA secondary structure analyses suggest that potential base-pairing interactions between the IRES and the CDS can vary depending on the reporter sequence (Supplementary Figure S3).

Among the three EMCV variants examined, EMCV-6A and EMCV-7A differ at an A bulge, with EMCV-7A containing an additional nucleotide. Previous studies indicated that this extra nucleotide increases PTB dependence, which can destabilize the IRES conformation and reduce translation efficiency [28]. The EMCV-ΔATG11 variant lacks part of domain I as well as the 11th AUG, which is considered the authentic initiation codon [29,30]. These structural differences account for the lower activity of these variants compared with EMCV-6A. In combinations led by EMCV-6A, the identity of the 3′ EMCV, whether it is EMCV-6A, EMCV-7A, or EMCV-ΔATG11, does not significantly affect the overall translation efficiency. We speculate that this is due to the inherent directionality of IRES elements and coding sequences, whereby only the 5′ IRES can initiate translation. Consequently, the translational output is primarily determined by the 5′ IRES, while the 3′ EMCV primarily serves a supportive role, facilitating ribosome recruitment for the 5′ EMCV-6A. This mechanism explains why all EMCV-6A—leading combinations exhibit expression levels roughly twice that of the single EMCV-6A IRES, and why dual EMCV-6A constructs do not outperform EMCV-6A+EMCV-ΔATG11 or EMCV-6A+EMCV-7A. Given the limited impact of the intrinsic translational strength of the 3′ EMCV on the overall output of dual-EMCV constructs, and supported by previous studies demonstrating that deletion of domain I does not abolish EMCV IRES activity [31,32], we removed domain I from the 3′ EMCV in our dual-IRES design. This modification preserves the ability of the 3′ EMCV to support translation while simplifying the IRES design in the dual-EMCV configuration.

In summary, our study highlights a dual-IRES design strategy for circular RNAs in which a second IRES element is placed at the 3′ end of the CDS to enhance translational efficiency. The dual-EMCV configuration in particular provides a reproducible and substantial improvement in protein expression and demonstrates compatibility with different coding sequences, suggesting that this approach may serve as a potentially useful strategy for circRNA optimization. Nonetheless, several limitations remain. The generalizability of these findings for coding sequences beyond the two analyzed here, diverse cell types, and in vivo systems awaits investigation. In addition, the precise mechanistic basis of IRES–IRES interactions in circular RNAs requires further investigation to fully exploit this strategy. Looking forward, this dual-IRES design may serve as a foundation for optimizing circRNA translation and guiding rational selection of IRES combinations in related experimental contexts.

5. Conclusions

In this study, we developed and systematically evaluated a dual-IRES—driven circular RNA design strategy and successfully identified IRES combinations capable of functioning cooperatively. Through comparative analyses using reporter gene expression, we demonstrate that appropriately paired dual-IRES configurations can significantly enhance protein translation compared with conventional single-IRES designs. Importantly, the optimized dual-IRES constructs exhibited consistent performance in the context of two different coding sequences, indicating good compatibility and robustness. Together, these findings establish dual-IRES engineering as an effective approach to improving translational efficiency of circular RNAs and provide a practical framework for the rational design of high-expression circRNA platforms.