Efficient circular RNA synthesis for potent rolling circle translation

The TRIC synthesizes circRNAs efficiently

Group I introns can also act as trans-ribozymes^20,21 (Fig. 1c). Therefore, once the target sequence is connected to the 3′ end of the ribozyme (Fig. 1d), the splicing reaction produces circRNAs. This trans-ribozyme-based circularization (TRIC) approach keeps the intact ribozyme upstream of target sequences and thus allows efficient local folding of the entire ribozyme.

To test the TRIC method, we selected the Anabaena transfer (t)RNA^Leu group I intron as it is short (249 nt) and highly active (Extended Data Fig. 1)^6,22. The Anabaena intron is situated in the tRNA anticodon arm (ACA) and utilizes its internal guide sequence (IGS) to form P1 and P10 interactions with flanking exons¹³ (Fig. 2a and Extended Data Fig. 1). To retain P1 and P10, part of the tRNA^Leu was preserved in TRIC-V0 for circularizing a 141 nt 3×Flag sequence (Fig. 2b). We analysed the in vitro transcription (IVT) sample using a 12% urea–polyacrylamide gel and identified the 403 nt full-length precursor (FL), the 257 nt spliced linear intron, and two unknown species (I and II; Fig. 2c, left). In a subsequent 6% urea–polyacrylamide gel, the electroeluted RNAs I and II both exhibited a minor species that either matched the linear intron or the nicked 3×Flag, suggesting that I is the circular intron and II is the circular 3×Flag (circ3×Flag; Fig. 2c, right). The circularity and identity were further confirmed by reverse transcription followed by PCR (RT–PCR) and Sanger sequencing (Fig. 2d,e). Due to the apparent presence of circ3×Flag, we concluded that the TRIC-V0 construct circularizes the 3×Flag sequence efficiently during IVT (Extended Data Fig. 2a,b).

a, A scheme of the Anabaena tRNA^Leu group I intron. The Anabaena intron divides the tRNA^Leu into a 34 nt left half (L34) and a 51 nt right half (R51). b, TRIC-V0 retains L15 and R30 of the tRNA sequence. Insert: the two-step reaction of the V0. c, Left: cotranscriptionally circularized sample of V0-3×Flag analysed on a 12% urea–polyacrylamide gel. Right: cotranscriptionally circularized sample of V0-3×Flag and electroeluted linear intron, and band I and II loaded on a 6% urea–polyacrylamide gel. The RNA ladder is labelled on the left. The black triangle, blue line and red circle indicate the FL, nicked circRNA and circRNA, respectively. This applies to all the following figures. d, RT–PCR using cotranscriptionally circularized sample of V0-3×Flag, and RNAs band I and II as templates. The positions of forward (For) and reverse (Rev) primers are indicated in b. e, Sanger sequencing of the RT–PCR product from band II RNA. f, Cotranscriptionally circularized samples of V1 variants analysed by 6% urea–polyacrylamide gel. The electroeluted band II was used as a reference. Inserts: diagrams of V1.0–2 showing introduction of the EGS and the internal loop. g, Five GOIs spanning from 1,638 nt (EGFP) to 8,706 nt (Factor 8) were assembled into V1.0 and PIE constructs. All resulting circRNAs contained a CVB3 IRES and a CDS of EGFP, Fluc, Spike-EGFP, Cas9-EGFP or Factor 8-EGFP. For each GOI, circRNAs generated by the V1.0 and the PIE have identical sequence. h, Cotranscriptionally circularized samples of V1.0 and PIE for long GOIs were analysed on a 0.8% native agarose gel operated at 25 W constant power. Cotranscriptional splicing products are indicated with vertical blue lines or blue arrows. Additional post-transcriptional circularization was achieved by adding 2 mM GTP and heating at 55 °C for 20 min. i, FL samples of V1.0 for long GOIs were circularized for various durations (0, 10, 20 or 40 min) and then analysed in a 0.8% native agarose gel.

Full size image

An extended guide sequence (EGS) and the internal loop between EGS and IGS are key for enhancing the efficiency of trans-ribozymes^23,24. Therefore, we next constructed TRIC-V1 by introducing a 20 or 23 bp EGS and internal loops to TRIC-V0 (Fig. 2f, right and Extended Data Fig. 2d). All tested V1 variants circularized the 3×Flag efficiently (Fig. 2f, left). Since V1.0 yielded the highest circRNA to intron ratio, we chose this construct for subsequent optimizations (Extended Data Fig. 2a,c).

A new protocol enables synthesis of circRNAs >8,000 nt

Next, we wanted to evaluate the performance of V1.0 in circularizing long GOIs and compare it with the PIE (Fig. 2g). Gel analysis suggested that V1.0 has a higher cotranscriptional splicing efficiency than the PIE for the Coxsackievirus B3 (CVB3)-contained enhanced green fluorescent protein (EGFP), Firefly luciferase (Fluc), spike protein of SARS-CoV-2 (Spike) and Cas9 (Fig. 2h). Twenty minutes of post-transcriptional circularization at 55 °C reduced the amount of FL for all constructs. However, due to severe nicking and degradation, we were unable to identify circRNA bands.

Mg²⁺ ions and elevated temperatures are the primary causes of RNA nicking⁶. Since both are essential for IVT and circularization, we reasoned that suppressing cotranscriptional circularization and accelerating post-transcriptional circularization are rational ways to reduce nicking of the circRNA product. We titrated Mg²⁺ in the IVT reactions and found that cotranscriptional circularization was suppressed using ≤16 mM Mg²⁺ at 24 mM NTP concentration²⁵ (Extended Data Fig. 2e). Notably, IVT yield is only significantly reduced when the Mg²⁺ concentration is lower than 12 mM (Extended Data Fig. 2f). Using the generated FL for post-transcriptional circularization, V1.0 circularized all GOIs efficiently, including the 8,706 nt Factor 8 (Fig. 2i and Extended Data Fig. 2g–j) (see below for circRNA band assignment). Prolonged circularization decreased the FL, but also increased circRNA nicking. Moreover, a direct comparison between V1.0 and PIE at 20 min of circularization revealed comparable efficiencies (ratio of converted FL) (Extended Data Fig. 2k). Notably, the absence of circular Spike, Cas9 and Factor 8 in the widely used co-post-transcriptional circularization protocol (Fig. 2h) suggests that the FL post-transcriptional circularization protocol is essential for synthesis of circRNAs >5,000 nt.

Native agarose gels can separate circRNAs from their linear counterparts

It is assumed that native agarose gels cannot resolve circRNA and its linear form^4,11. However, we consistently observed two distinct bands, which were both resistant to RNase R exonuclease treatment, at the rough position of circEGFP in a 0.8% agarose gel (Fig. 3a,b). When the gel-purified lower band was linearized by RNase H, it migrated similarly to the upper band, indicating that the lower and the upper bands are circular and nicked EGFP, respectively (Fig. 3c). This suggests that circEGFP moves faster than nicked EGFP in native agarose gels.

a, Cotranscriptionally circularized sample of the V1.0 EGFP analysed on a native agarose gel. The FL EGFP is 1,963 nt and the circEGFP is 1,638 nt. Two bands were observed around the circEGFP position. b, FL and circularized (circ.) EGFP were treated with RNase R (R) and analysed on a native agarose gel. c, RNA of the lower band in the gel from b was purified, RNase R digested and then linearized with RNase H (H) in the presence of a 34 nt DNA primer and analysed on a native agarose gel. d–i, FL and circularized RNAs analysed on 0.8% (d), 1.5% (e) and 3.0% (f) native agarose gels or 1.5% urea–agarose gels with 6 M (g), 4 M (h) or 2 M (i) urea. P2A-EGFP has no CVB3 IRES. Electrophoresis of urea–agarose gels was operated in a vertical system with specified settings.

Full size image

To see whether this is a general trend, we analysed circRNAs spanning lengths from 813 to 5,757 nt on 0.8%, 1.5% and 3% native agarose gels (Fig. 3d–f). In the 0.8% gel, the P2A-EGFP (no CVB3, 813 nt) and the Fluc (2,601 nt) behaved similarly to the EGFP (Fig. 3d). For the Spike, two largely overlapping bands appeared below the FL. However, a new band appeared above the Cas9-FL. This slow-moving band remained in the 1.5% gel and was also observed for the Spike, suggesting it is circRNA (Fig. 3e). The separation for the other circRNAs is reduced in the 1.5% gel, indicating that increasing agarose concentration more significantly decreases the mobility of circRNAs than that of linear RNAs. This was further confirmed using a 3% gel, where all circRNAs ran slower than their linear counterparts (Fig. 3f). Circularity of these slow-moving bands in the 3% gel was confirmed by RNase R digestion (Extended Data Fig. 3a). These results demonstrate that sufficient separation between circRNA and its linear counterpart is achievable in native agarose gels (see also Extended Data Fig. 4).

The urea–agarose gel system provides excellent separation between circRNAs and their linear equivalents

Next, we tested denaturing agarose gels for separation of circRNAs and their linear forms. In formaldehyde agarose gels, we observed separation only for Spike and Cas9 in a 1.5% gel (Extended Data Fig. 3b,c). Due to its toxicity, formaldehyde can be substituted with urea²⁶. Notably, in a 6 M–1.5% urea–agarose gel, all tested circRNAs moved drastically slower than their linear counterparts (Fig. 3g). Lowering the urea or agarose concentration reduced the separation (Fig. 3h,i and Extended Data Fig. 3d,e). In a 6 M–4% gel, even the 141 nt circ3×Flag ran slower than its nicked form (Extended Data Fig. 3f). We also observed that extending the running time or increasing the running power enhanced the separation (Extended Data Fig. 3g–j). Notably, electrophoresis of these gels only takes <30 min, making it a rapid method for circRNA identification (Extended Data Fig. 4).

The TRIC-V2 enables RNA circularization without unwanted sequences

An ideal TRIC construct should have minimal sequence requirements while remaining efficient (Fig. 4a). However, both V1.0 and PIE rely on native splicing sequences that will remain part of the final circRNAs. To address this limitation, we reduced the tRNA sequence to retain only P1 and P10 (V1.30; Extended Data Fig. 5). However, this markedly reduced the circularization efficiency (Fig. 4b). Restoring the R30 (V1.33) recovered the efficiency to V1.0 level while restoring the L15 (V1.32) or introducing a 3′ exon–EGS interaction (V1.31) did not. Further optimizations of the L/R lengths (V1.34–1.39) showed that the 17 nt ACA (V1.39) is the minimal sequence for achieving the V1.0 efficiency²⁵.

a, The concept of the TRIC-V2, which has minimal sequence requirements yet remains high efficiency. eIGS, extended IGS. b, The L15/R30 of the tRNA sequence in V1.0 were systematically engineered. For V1.30–39 (V1.3X), the tRNA sequences were gradually reduced. L8/R9 in V1.39 is the ACA of the tRNA. For V2.0–2.2 (V2.X), the sequences of the anticodon stem were reversed (re) (V2.0) or extended (15 bp, V2.1), or the sequences of P1/P10 were reversed (V2.2). Cotranscriptionally circularized samples were analysed on a 6% urea–polyacrylamide gel. c, An effective eACA should consist of a 7 nt loop with a U at the third position and a stem ≥5 bp. The circularization site can be placed either in a UTR or CDS. d, Circularization sites of the T2A-EGFP, T2A-Nluc and circZnf609 were placed in CDS with stem lengths of 5, 7 and 11 bp, respectively. V2 circZnf609 is identical to the natural circZnf609 except for nine non-sense mutations in CDS. FL of each RNA was circularized for 20 min and analysed on a 4 M–1.5% urea–agarose gel.

Full size image

Since structures, rather than sequences, are crucial for group I intron activity²⁷, we next generated various V2 constructs that do not rely on native splicing sequences (Extended Data Fig. 5l–n) with altered anticodon stems (V2.0 and V2.1) or reversed P1 and P10, while keeping the U for the G•U base pair, essential for the splicing reaction (V2.2). All constructs circularized efficiently (Fig. 4b), thus, V2 minimally requires an extended ACA (eACA) comprising a 7 nt loop with a U at the third position and a ≥5 bp stem for efficient circularization (Fig. 4c and Extended Data Fig. 6a). We then tested circularization of three longer protein coding circRNAs. Multiple eACAs were found in each coding sequence (CDS) and all GOIs were circularized efficiently (Fig. 4d). Due to the small size and sequence requirements of this eACA motif, suitable sites can be easily identified or, in case of protein CDS, engineered by making use of codon redundancy. Alternatively, short (<5 nt) insertions can be made in non-coding regions (Fig. 4c).

The TRIC-V2 is faster than the PIE

The circularization efficiency was highest for the natural circZnf609²⁸, which exhibits the longest stem (11 bp) among the three constructs (Fig. 4d), suggesting that longer stems lead to higher efficiency. To further investigate this, we generated V2 constructs with stem lengths of 15 and 25 bp. As expected, V2 (stem of 15 or 25 bp) outperformed V1.0 (stem of 5 bp) in circularizing EGFP (Fig. 5a). Notably, further extending the EGS to 40 nt did not obviously enhance the efficiency.

a, Comparison between V1.0 and V2 in circularizing EGFP. EGSs were either 20 or 40 nt. The eACA stem for V2 constructs was either 15 or 25 bp. FL of V1.0 and V2 were circularized for 4 min and analysed on a 2% native agarose gel. There was no separation between circRNA and its linear form in 2% gels. b–d, The comparison between V2 (stem of 25 bp) and PIE for circularizing EGFP (b), Spike (c) and Cas9 (d). FL were circularized for 1–4 min. e, Michaelis–Menten analysis of the V2 (stem of 25 bp) and the PIE for circularizing EGFP. Calculations were based on the reaction processes observed in native agarose gels (Extended Data Fig. 6b–q). f, FL of V2 (stem of 25 bp) EGFP and PIE EGFP were circularized for either 3 and 8 min or 8 min and analysed on a native gel (Extended Data Fig. 6r). The intensities of the remaining FL, circRNA and nicked RNA were measured in ImageJ. Efficiency was calculated by dividing the intensity of spliced FL by the total FL, yield was determined by the intensity of circRNAs divided by total FL and nicking was calculated as the ratio between nicked RNAs and the sum of circRNA and nicked RNA. The red lines indicate the reported values of yield and nicking for a circEGFP⁶. g,h, The comparison between V2 using either the Ana group I intron or the Tetra group I intron and the Tetra-STS and the Tetra-Rzy constructs: CVB3 (coxsackievirus B3)-EGFP was cloned to the best Tetra-STS (AU-rich no. 16) and Tetra-Rzy (CVB3 IRES-GFP) constructs, and FL were circularized for the designated time and analysed on native agarose gels (g) or urea–agarose gels (h). The data in e and f are mean ± s.d. for three independent replicates tested using unpaired two-tailed t-tests.

Source data

Full size image

The superior efficiency of V2 over V1.0 suggests that V2 would probably outperform PIE as well. To compare V2 and PIE, we circularized EGFP, Spike and Cas9-FL for 1–4 min (Fig. 5b–d). V2 (stem of 25 bp) outperformed PIE in efficiency for all tested candidates. For instance, in 1 min, V2 (stem of 25 bp) converted over half of the FL into circEGFP, while PIE only converted a small portion (Fig. 5b). The higher efficiency is reflected by the construct kinetics: using the Michaelis–Menten analysis²², we found that V_max of the V2 (stem of 25 bp) is 3.7 times higher than the PIE V_max in the EGFP circularization (Fig. 5e and Extended Data Fig. 6b–q). V2 (stem of 25 bp) showed higher efficiency compared with V2 (stem of 15 bp), suggesting potential space for further efficiency improvement. Additionally, PIE generated a substantial number of concatenations, which were mostly absent from V2 (Fig. 5b).

Next, we compared EGFP circularization efficiency, yield and nicking between EGFP V2 (stem of 25 bp) and PIE constructs (Fig. 5f and Extended Data Fig. 5r). In 3 min, V2 (stem of 25 bp) achieved slightly higher efficiency and yield than PIE in 8 min, with significantly less nicking. Extending the circularization time to 8 min improved V2 (stem of 25 bp) to 97.8% efficiency and 74.8% yield (90.3% of the yield limit), but also increased nicking. Notably, using the FL post-transcriptional circularization protocol, PIE improved yield and nicking from previously reported values of ~50% and ~20% to 65.5% and 9.3%, respectively⁶. Further yield enhancement from 65.5% (PIE) to 74.8% (V2) should be attributed to the higher circularization rate of V2.

Recent studies have described Tetrahymena thermophila (Tetra) group I intron-derived constructs, the Tetra-STS²⁹ and the Tetra-Rzy³⁰, for circRNA synthesis. We compared these with V2 by cloning CVB3-EGFP into Tetra-STS (AU-rich no. 16) and Tetra-Rzy (CVB3 IRES-GFP) constructs. As shown in Fig. 5g,h, V2 outperforms both constructs. Additionally, the Tetra-V2 also produced circCVB3-EGFP efficiently, demonstrating that optimizations from the Ana intron can be effectively applied to other group I introns.

Immunogenicity of circRNAs is low

The immunogenicity of circRNAs remains debatable^{8,11,16,31,32}. Recent research suggested that PIE-derived residual bacterial sequences cause circRNA immunogenicity¹⁶. Given that V2-derived circRNAs lack bacterial sequences, they might be less immunogenic than PIE circRNAs. To test this, we designed V2 CVB3-EGFP and CVB3-Nano luciferase (Nluc) constructs, where the eACA is either formed by the CDS (EGFP, stem of 5 bp) or derived from the 25ES7b stem of human 28S ribosomal RNA (rRNA) (Nluc, stem of 24 bp). CircRNAs were purified by three consecutive high-performance liquid chromatography (HPLC) runs followed by RNase R digestion⁸ (HR purification; Fig. 6a). Subsequently, we transfected these circRNAs and controls into A549 (human lung carcinoma) cells and monitored expression of RIG-I (5′-triphosphate sensor) and several cytokines³³. As expected, poly(I:C) (double-stranded RNA mimic) and unmodified mRNAs substantially increased RIG-I and cytokine expression compared with the lipofectamine-only control (Fig. 6b). CircRNAs exhibited lower immunogenicity compared with unmodified mRNAs, but still induced notable RIG-I and cytokine expression. We consistently observed reduced immune responses with V2 circNluc compared with PIE circNluc, but not for V2 circEGFP. Therefore, the immunogenicity comparison between V2- and PIE-derived circRNAs remains inconclusive.

a, CircRNAs were purified using HR, with circularized RNA purified by three consecutive HPLC runs and a final RNase R digestion. Late circRNA fractions from each injection were collected and 50 ng of each purified circRNA was analysed on a 6 M–1.5% urea–agarose gel. b, For circRNAs purified using HR, 100 ng of each purified circRNAs, poly(I:C) (double-stranded RNA mimic) and both modified and unmodified linear (Lin.) mRNAs (capped and poly-A tailed) were transfected into 10,000 A549 (human lung carcinoma) cells. The mRNA levels of RIG-I and several cytokines were monitored by RT–qPCR 24 h later. The y axes represent fold changes compared with the mock. c, CircRNAs were purified using PRH purification, with circularized RNA treated with alkaline phosphatase and RNase R before HPLC injection. Late circRNA fractions from each injection were collected and 50 ng of each purified circRNA was analysed on a 6 M–1.5% urea–agarose gel. d, CircRNAs purified using PRH were transfected and and monitored as in b. Insert: CCL5 expression of mock, lipofectamine only (lipo.) and circRNAs. e, Group I introns function as TER to remove a middle (bridge) sequence from their target sequences. f, The TERIC produces circRNAs. g, Circularization of EGFP using TERIC were conducted in 10 µl volumes with 0.1, 0.5 or 2 mM GTP. In protocol A, in vitro transcribed TERs and pGs were initially mixed at a 5:1 ratio, refolded, supplemented with splicing buffer and heated at 55 °C for 20 min. Subsequently, splicing reactions were halted using 2 μl of 100 mM EDTA and analysed on a 0.8% native agarose gel. h, FL and 20 min circularized circZnf609 analysed on a native agarose gel. FL were synthesized either with or without 100% m⁶A or N¹Ψ modification (Modi). TER was unmodified. i, CircRNAs were purified using either HR or PRH purification, as in a and c. j, CircRNAs purified using either HR or PRH purification were transfected and and monitored as in b. IFN-α, interferon-α; IFN-β, interferon-β; TNF-α, tumour-necrosis factor-α; IL-6, interleukin-6; CCL5, chemokine (C-C motif) ligand 5; AP, alkaline phosphatase; R, RNase R. Data in b, d and j are mean ± s.d. for three biological replicates. *P < 0.05, **P < 0.01 and ***P < 0.001, unpaired two-tailed t-test.

Source data

Full size image

RNase R might enrich structured linear RNAs that are immunogenic¹⁷. We therefore moved the RNase R digestion before HPLC (RH purification; Extended Data Fig. 7a). Subsequent tests indicated comparable immunogenicity of circRNAs generated by the HR and RH purifications (Extended Data Fig. 7b). However, the V2 circNluc elicited heightened RIG-I expression, suggesting 5′-triphosphate contamination. Thus, we further added an alkaline phosphatase treatment before RNase R (PRH purification; Fig. 7c and Extended Data Fig. 7c). The phosphatase-treated unmodified RNAs indeed exhibited a significantly reduced immunogenicity^11,16. Remarkably, all PRH-purified circRNAs failed to induce RIG-I or cytokine expression compared with the lipofectamine only (Fig. 6d). An exception was noted for V2 circEGFP, which induced minimal yet significant CCL5 expression (Fig. 6d, insert). This might be due to FL contamination resulting from the low circularization efficiency (stem of 5 bp). In conclusion, these results support the notion that circRNAs have low immunogenicity^8,32.

a, Equimolar amounts of N¹Ψ-modified linear Nluc, V2 circNluc and PIE circNluc were transfected into 10,000 A549 cells. Nluc expression was measured over 7 days. b, The cumulative expression of Nluc from N¹Ψ-modified linear Nluc, V2 circNluc and PIE circNluc. c, When circRNAs contain only a CDS but no stop codons, ribosomes can translate such circRNA indefinitely and produce a polyprotein, a process known as the RCT. d, When a 2A sequence is present, such as T2A or P2A, the polyprotein will be converted to protein monomers. e, The secondary structure of the 373 nt CSFV IRES. Multiple stop codons are present in all possible frames: 4, 10 and 8 in frames 1–3, respectively. Frame 1 was cleaned of stop codons by three mutations (U52A, A171G and A191G), and one deletion (d349U) to align with the CDS. f, Equimolar amounts of circOR4F17-Nluc-RCT, circCSFV-Nluc, circCSFV-Nluc-RCT, V2 circNluc and N¹Ψ-modified linear Nluc were transfected into HEK 293F cells and 24 h later, Nluc expression was monitored. The data in a, b and f are mean ± s.d. for three (a and b) or four (f) biological replicates. ***P < 0.001, unpaired two-tailed t-test.

Source data

Full size image

Synthesis and immunogenicity of modified circRNAs

The significant residual immune response to phosphatase-treated linear mRNAs highlights the potential immunogenicity of nicked circRNAs (Fig. 6d), which could be diminished by RNA modifications^34,35,36. However, the PIE, and probably the TRIC as well, are unable to synthesize fully modified circRNAs since modifications could disrupt the ribozyme activity⁸. Since group I introns can also function as trans-excision ribozymes (TER)^37,38,39 (Figs. 1e and 6e), relocating the 3′ and 5′ parts of the bridge to 5′ and 3′ ends of a GOI, respectively, would allow circularization of the GOI in trans (Figs. 1f and 6f and Extended Data Fig. 8a). With this trans-excision ribozyme-based circularization (TERIC), modified GOIs can be circularized by unmodified TERs.

To test this idea, the IGS and 5′ half of the P9.0 were retained in TERs to create P1/P10 and P9.0 interactions between TERs and GOIs^38,39. Additionally, the key elements for the TRIC were also preserved (Extended Data Figs. 1 and 8a). Initial tests showed that the TER(1,226) circularized unmodified EGFP efficiently (Fig. 6g and Extended Data Fig. 8b,c). The circularization was confirmed and further optimized (Extended Data Fig. 8d–h). Neither PIE nor TRIC or TERIC produced 100% N⁶-methyladenosine (m⁶A) or N¹-methylpseudouridine (N¹Ψ) modified circRNAs (Extended Data Fig. 9). We realized that these modifications could weaken the AU-rich IGS of TER(1,226) and thus additionally analysed circZnf609, whose IGS is CCGCC. As expected, TERIC synthesized 100% m⁶A and N¹Ψ modified circZnf609 efficiently, but not PIE or TRIC (Fig. 6h). Of note, precursors containing N¹Ψ were circularized more efficiently than m⁶A-modified ones. In summary, TERIC enables efficient synthesis of modified circRNAs.

Next, we tested immunogenicity of the circZnf609 with or without modifications (Fig. 6i,j). Consistently, HR-purified, but not PRH-purified, unmodified circRNAs exhibited significant immunogenicity, regardless of the source (V2 or PIE). However, the modified circZnf609 did not show significant advantages over PRH-purified unmodified circRNAs⁸. It is possible that longer transfection times or higher doses are required to observe potential differences. In contrast, m⁶A-modified circZnf609 induced significant RIG-I expression.

TRIC circNluc produces more proteins than PIE circNluc

To study the translation efficiency of TRIC circRNAs, we investigated protein expression from circRNAs containing a CVB3 IRES. In A549 cells, Nluc expression from N¹Ψ-modified mRNA peaked at day 1 and then declined rapidly, whereas expression from V2 circNluc peaked at day 2 and remained significantly higher than day 1 up to day 3 (Fig. 7a). Similar patterns were seen in HEK 293F cells (Extended Data Fig. 10a,b). Meanwhile, PIE circNluc expressed similarly to V2 circNluc on day 1, but showed no increase afterwards. Consequently, V2 circNluc led to higher expression after day 2 (Fig. 7a,b). On day 7, V2 circNluc expression remained at almost half the level of that from N¹Ψ-modified mRNA on day 1.

Translation efficiency of RCT is increased by over 7,000-fold

Unlike single-shot translation, RCT primarily produces a polyprotein. This can be converted into protein monomers provided with 2A skipping sequences⁴⁰ (Fig. 7c,d). Instead of using circRNAs with only a CDS, which are inefficient in translation initiation^9,10,40, we selected a potent short IRES (OR4F17)⁴⁰ for the RCT construct. As expected, the circOR4F17-Nluc-RCT produced a significant number of proteins. However, the translation was still orders of magnitude less efficient compared with N¹Ψ-modified mRNA (Fig. 7f).

We then asked whether potent viral IRESs could be used for RCT. However, viral IRESs are typically long and highly structured, which could block RCT either by in-frame stop codons or by causing ribosome stalling. Nevertheless, we tested the 373 nt CSFV IRES^6,40,41 (Fig. 7e). As expected, multiple stop codons were identified in each frame. We therefore engineered frame 1 to eliminate stop codons and align with the CDS. Notably, the RCT using CSFV IRES showed over a tenfold increase in Nluc expression compared with its single-shot translation and was over 7,000-fold more efficient than the RCT using the OR4F17 IRES (Fig. 7f and Extended Data Fig. 10c).

• SEO Powered Content & PR Distribution. Get Amplified Today.
• PlatoData.Network Vertical Generative Ai. Empower Yourself. Access Here.
• PlatoAiStream. Web3 Intelligence. Knowledge Amplified. Access Here.
• PlatoESG. Carbon, CleanTech, Energy, Environment, Solar, Waste Management. Access Here.
• PlatoHealth. Biotech and Clinical Trials Intelligence. Access Here.
• Source: https://www.nature.com/articles/s41551-024-01306-3