Generation of precise and accurate engineered circRNAs using enzymatic ligation Authors: Amrita Singh, Adela Dujsikova, Noah Mueller, Y Grace Chen Published: 2026 ABSTRACT Abstractmessenger RNA-based therapeutics have revolutionized the treatment and prevention of infectious, neurological, and cancer diseases. However, their linear topology makes them susceptible to rapid degradation in vivo, which limits their therapeutic efficacy. Engineered circular RNAs (circRNAs) due to their closed ends and high stability are emerging as a promising alternative to linear RNA therapies. Engineered circRNAs are also increasingly used to mimic naturally occurring circRNAs in functional studies. Both applications, however, depend on production of precise circRNAs with homogenous sequences to enable accurate interpretation of biological outcomes. To address this, we developed and optimized methods for generating precise circRNAs. We employed enzymatic ligation of linear RNAs rather than autocatalytic splicing to produce circRNAs to minimize extraneous nucleotides remaining from the ribozymes. We carefully designed the DNA transcription template to maintain sequence and structural integrity. A permuted DNA template leveraging three internal guanosines (Gs) was synthesized and amplified using a reverse primer containing two 2′-O-methyl groups. This approach optimally produced the linear precursor RNA with correct 5′ and 3′ ends. After testing multiple workflows, we found that GMP-primed in vitro transcription, T4 RNA ligase 2-mediated circularization, and urea–polyacrylamide gel electrophoresis (PAGE) gel extraction produced the highest fidelity circRNAs. Introduction The COVID-19 pandemic revealed the power of messenger RNA (mRNA)-based therapeutics with billions of doses of mRNA vaccines distributed globally. The concept of using mRNAs to produce proteins was first proposed in the 1960s [1], but translating engineered mRNAs into viable therapeutics required decades of innovation to overcome substantial biological challenges, including instability, innate immune activation, inefficient translation, and limitations to in vitro synthesis [2]. Engineered circular RNAs (circRNAs), which lack free ends, have emerged as attractive candidates for novel therapeutics [3]. Their covalently closed structure confers resistance to exonucleases, substantially enhancing stability relative to linear RNAs. Moreover, incorporation of internal ribosome entry sites (IRES), some as small as a single modified nucleotide can drive efficient translation of the encoded open reading frame (ORF) [4–6]. In functional studies of endogenous circRNAs, engineered circRNAs also present a distinct advantage over plasmid-based overexpression systems, which often produce undesired linear RNA artifacts through trans-splicing [7] or incomplete circularization [8]. By contrast, synthetic circRNAs can be precisely designed and purified to avoid such contaminants. They can also serve as powerful tools for modeling viral infections, as certain viruses generate circRNAs that are encapsulated within virions and delivered into host cells prior to viral transcription [9]. Altogether, these advances highlight the therapeutic and experimental potential of circRNAs and emphasize the need for efficient and precise methods to generate them with defined sequences and structures. Engineered circRNAs are primarily generated through two approaches: autocatalytic splicing or enzymatic ligation of linear RNA precursors. Autocatalytic splicing strategies commonly employ group I ribozymes derived from the T4 phage td gene, Anabaena pre-tRNA, or Tetrahymena introns, which are positioned to flank the exon targeted for circularization [10]. During in vitro transcription, the resulting linear RNAs form conformations that facilitate intramolecular transesterification reactions, producing circRNA molecules. However, this process frequently results in circRNAs that retain portions of the flanking intronic sequences introducing hundreds of nucleotides from the introns into the resulting circRNA, which can form double-stranded regions that trigger innate immune responses [11]. Thus, generating circRNAs with precisely defined sequences is critical for being an accurate template for protein translation in therapeutics and reliably assessing their biological functions. Enzymatic ligation represents an alternative approach that minimizes the retention of extraneous sequences [10]. Therefore, to enable the synthesis of circRNAs suitable for both therapeutic applications and biological studies, our objective was to develop a strategy that yields circRNAs with precisely defined sequences. We employed an enzymatic ligation–based approach to maintain complete control over circRNA sequence composition. During optimization, we observed that conventional enzymatic ligation protocols often lead to the incorporation of unintended nucleotides resulting from in vitro transcription into the circularized product. To overcome this challenge, we developed approaches to avoid inclusion of these extraneous sequences and systematically evaluated the circularization efficiency and accuracy of DNA and RNA ligases. We optimized each stage of the circRNA synthesis workflow including sequence design, in vitro transcription, generation of 5′-monophosphorylated RNA precursors, ligation, and purification to maximize precision and yield. Sequence integrity was verified at multiple steps and final circRNA preparations were purified by urea–polyacrylamide gel electrophoresis (PAGE) gel extraction to eliminate unligated linear species and concatenated by-products. Materials and methods RNA structural modeling (ViennaRNA RNAfold) RNA secondary structures were predicted using RNAfold (ViennaRNA package v2.7) at 37°C with Turner 2004 thermodynamic parameters. Linear RNAs were folded with the –noPS option, and circRNAs with the –circ –noPS option. Input sequences were formatted as single-line RNA strings by converting DNA bases (T) to uracil (U) and uppercasing. Minimum free-energy structures were obtained in dot–bracket notation. Design of permuted DNA templates for transcription Internal GGG motifs were randomly selected as transcription start sites to generate permuted DNA templates initiating from within the sequences. The two DNA halves were amplified separately using Phusion polymerase (NEB #M0530L) followed by an overlap extension polymerase chain reaction (PCR) with a forward primer containing a 5′ T7 promoter overhang. The amplified products were cloned into pcDNA3.1 using Gibson assembly (NEB #E2611L). Plasmid sequences were confirmed by whole plasmid sequencing (Plasmidsaurus). Unmodified DNA templates were amplified from this plasmid using Phusion polymerase and standard primers. Modified DNA templates were amplified using Phusion polymerase with a standard forward primer and a reverse primer containing two 2′-O-methyl RNA residues at their 3′ terminus. Both the DNA templates were extracted from agarose using gel DNA recovery kit (Zymo #D4008). Primer sequences are listed in Supplementary file 1. in vitro transcription in vitro transcription reactions of 20 µL included 100 ng of DNA templates, 7.5 mM of each of the NTPs, 1× of T7 reaction buffer, and 2 µL of T7 RNA polymerase enzyme at 37°C for either 2 or 16 h (NEB #E2040S). The addition of 2 units of Turbo DNase (Invitrogen #AM2238) per transcription reaction at 37°C for 15 min degraded DNA templates. Reactions were purified using RNA cleanup columns (NEB #T2050L). RNA concentrations were measured using Nanodrop One. TapeStation analysis RNA samples were analyzed by Agilent HS RNA Screen Tape on an Agilent 4200 TapeStation System. 2 µL of RNA samples (typically 10 ng/µL) were mixed with 1 µL of High Sensitivity RNA dye. Samples were vortexed vigorously for 30 s followed by heating at 72°C for 3 min and immediately transferred on ice for at least 3 min before loading on the Screen Tape. Generation of monophosphorylated RNA 5′-monophosphorylated RNAs were generated using three different approaches. One approach involved dephosphorylation of 1 pmol of 5′-triphosphate transcripts using 5 units of quick calf intestinal phosphatase (NEB #M0525L) at 37°C for 15 min, followed by purification and addition of monophosphate to transcripts using 10 units of T4 polynucleotide kinase (NEB #M0201L) at 37°C for 30 min. In the second approach, pyrophosphates were removed by using 10 units of pyrophosphatase RppH (NEB #M0356S) on 2 µg of in vitro transcribed RNAs at 37°C for 30 min. The third approach used 5× or 10× excess guanosine monophosphate (GMP) to GTP in the transcription reactions at 37°C for 2 or 16 h. in vitro transcription reactions of 20 µL each included 100 ng of DNA templates, 6.25 mM GMP, 1.25 mM GTP, 7.5 mM each of ATP, CTP, and UTP, 1× of T7 reaction buffer, and 2 µL of T7 RNA polymerase enzyme. The concentration of MgCl2 in the 1× T7 reaction buffer remained unchanged between transcription reactions with and without GMP. The addition of 2 units of Turbo DNase (Invitrogen #AM2238) per reaction at 37°C for 15 min degraded the DNA templates. The final reactions from all the approaches were purified using RNA cleanup columns (NEB #T2030L). XRN1 assay One microgram of monophosphorylated RNA mixture was treated with 1 unit of XRN1 enzyme (NEB #M0338S) at 37°C for 60 min. Triphosphorylated RNAs were used as negative controls. The reactions were purified using 1.8× reaction volume of RNAClean XP beads (Beckman Coulter #A63987). XRN1 degrades 5′-monophosphorylated RNAs, therefore percent yields of monophosphorylation were calculated according to the RNAs that were degraded. Poly(A) tailing Five micrograms of RNA were poly(A)-tailed using 5 units of Escherichia coli poly(A) polymerase (NEB #M0276L) for 30 min at 37°C. The polymerase units and ATP concentration were adjusted proportionally to the amount of RNA in the reaction. Reactions were purified using RNA cleanup columns (NEB #T2030L). 3′ end sequencing of transcribed RNA Poly(A) tails were added to the 3′ ends of RNAs transcribed from unmodified and modified DNA templates. The 3′ ends were reverse transcribed with Superscript III first strand synthesis system (Invitrogen #18 080 051) using oligo dT primers. The products were amplified using a forward primer close to the 3′ end (eGFP 3′ end check_F sequence is listed in Supplementary file 1) and oligo dT. PCR products were cloned into pcDNA3.1 using Gibson assembly (NEB #E2611L) and were subjected to Sanger sequencing. RNase R treatment RNAs were treated with RNase R (NEB #M0100S) for 30 min at 37°C. A total of 20 units of RNase R were used for 10 µg of RNA; the units were adjusted proportionally to the amount of RNA in a reaction. Lucigen’s RNase R (#RNR07250) was also tested at 20 units for 10 µg of RNA. Reactions were purified using RNA cleanup columns (NEB #T2030L). Enzymatic ligation to generate circRNAs 5′-monophosphorylated RNAs were circularized with ligases. For a 100 µL splint reaction, 2 µL of 10 µM of linear RNA were used. DNA or RNA splint was added at 1:3 RNA to splint ratio and annealed in a oligonucleotide annealing buffer (50 mM, Tris pH 7.4, 100 mM NaCl) by heating the RNA and splint mixture to 90°C for 3 min followed by slowly cooling to 25°C at a rate of 0.1°C per s. Subsequently, linear precursor RNAs were ligated using one of the following enzymes: DNA ligase, RNA ligase 1, or RNA ligase 2. 2000 units of T4 DNA ligase (NEB #M0202M) were used for ligation at 37°C for 1 h. RNA ligase 2 (NEB #M0239L) reactions were either set up with DNA splint or RNA splint. Reaction conditions remained the same for both the splints: 15 units of enzyme with 10% polyethylene gylcol (PEG), 100 mM MgCl2, and RNA ligase 2 buffer, and reactions were incubated at 25°C for 1 h. For RNA ligase 1 (NEB #M0204S) reactions, 25 units of the enzyme were added with 10% PEG, 1× RNA ligase buffer and incubated at 37°C for 1 h. When DNA splints were included, the addition of 2 units of Turbo DNase (Invitrogen #AM2238) for 15 min at 37°C degraded the DNA immediately after the end of ligation incubation. All ligation reactions were purified using RNA clean up columns (NEB #T2050L). Subsequently, ligated RNAs were poly(A)-tailed (NEB #M0276L) and treated with RNase R (NEB #M0100S) as per aforementioned conditions. After column purification, the RNAs were quantified using Nanodrop One. For gel purification instead of poly(A) tailing and RNase R, reactions were run on a 3% urea–PAGE gel following ligation. Urea–PAGE gel analysis The 3% urea–PAGE gel was polymerized by adding 3 mL of 30% bis acrylamide (Bio-Rad), 1× Tris–borate ethylenediaminetetraacetic acid (EDTA; TBE), and nuclease free water up to 10 mL. The mixture was heated for 10–20 s, followed by cooling on ice until reaching room temperature. Then, 10% ammonium persulphate and 10 µL of TEMED were added and thoroughly mixed to initiate free radical formation to polymerize the gel. The gel was poured into the plates and allowed to solidify for 15 min, followed by a pre-run at 120 V for 15 min in 1× TBE. Samples were prepared with a 2× gel loading buffer, heated at 70°C for 5 min, and immediately placed on ice. Wells were properly flushed to remove residual urea and the samples were carefully loaded into the wells. The gel was run at 120 V for 2 h in 4°C followed by staining using 1× SYBR™ Green I nucleic acid gel stain (Invitrogen) in 1× TBE buffer solution for 15 min with gentle rocking. 5 min of destaining using 1× TBE buffer was conducted. The gel was visualized in the EtBr channel using Bio-Rad’s Gel Doc Go Imaging System. Urea–PAGE gel purification (crush-and-soak method) Desired bands were sliced from the urea–PAGE gel. Gel slices were transferred to a 1.5 mL Eppendorf tube and weighed, frozen on dry ice, and crushed with a 1 mL pipette tip. Then, the volume of 100 mM NaCl, 10 mM Na-MOPS pH 6.0, and 3 mM EDTA buffer was added according to 3× the weight of the gel slices. This mixture was rotated for 2–3 h at 4°C and passed through a 3 mL syringe with a 0.22 µm filter to remove gel pieces. To precipitate the RNA, 2–2.5× of 100% ethanol, 1/5th volume of 3 M sodium acetate pH 5.2, and 0.5 µL of glycoblue (Invitrogen) were added and kept at −80°C for at least 1 h. Tubes were centrifuged at 15,000 rpm for 10 min. The supernatant was discarded and the blue pellets were washed twice with 70% chilled ethanol. Pellets were dissolved in an acidic buffer containing 6 mM Na-MES buffer, pH 6.0, to minimize the hydrolysis of circRNAs to linear RNA. The volume of the buffer was chosen according to pellet size and downstream experiments. RNA concentration was measured using Nanodrop One. 50 nanograms of purified samples were run on a 3% urea–PAGE gel to analyze the integrity of the extracted circRNAs. EX E-gel purification Ligated RNAs were run for 20 min on a 2% agarose EX E-gel (Invitrogen #G401002) to separate circRNAs from the linear RNAs of the same size. CircRNA bands were excised, weighed, and either extracted using the crush-and-soak method or by gel RNA recovery kit (Zymo #R1011). Recovered RNAs were quantified and 50 ng of purified samples were run on a 3% urea–PAGE gel to check the RNA integrity. Reverse transcription-polymerase chain reaction (RT-PCR) Reverse transcription for ligation junction accuracy used the iScript kit (Bio-Rad #1708 891) containing random hexamers. Junctions were amplified using primers flanking around 200 bases of the ligation junction such that only the circRNAs are amplified (ligation primer sequences are listed in Supplementary file 1). PCR products of desired size were gel purified using the gel DNA recovery kit (Zymo #D4008) followed by Sanger sequencing. RNase H based circularity confirmation assay RNAs were annealed to a single-stranded DNA (ssDNA) probe (eGFP probe sequence listed in Supplementary file 1) in a buffer (0.2 M NaCl, 0.1 M Tris–HCl) by heating the mixture to 90°C for 3 min followed by slowly cooling to 25°C at a rate of 0.1°C per s. The probe-bound RNA was immediately treated with 2.5 units of RNase H (NEB #M0297L) for 30 min at 37°C. DNA probes were removed using 2 units of DNase I (NEB #M0303L) for 30 min at 37°C. Reactions were purified using 1.8× RNA clean XP beads (Beckman Coulter #A63987). The recovered RNA was analyzed using TapeStation. Results Optimization of in vitro transcription templates for precise circRNA synthesis As a model to optimize circRNA synthesis conditions, we used a 1322-nucleotide (nt) eGFP-encephalomyocarditis virus IRES construct. Precursor linear RNAs were generated through T7 RNA polymerase–dependent in vitro transcription (Fig 1A). Because T7 RNA polymerase initiates transcription most efficiently from a sequence containing three consecutive guanosines (GGG) [12], the use of a canonical T7 promoter positioned upstream of the eGFP–IRES sequence would introduce three additional, unwanted nucleotides into the final circRNA. These sequences could lead to frameshifting of the open reading frame, impacting protein translation. RNAfold analysis comparing circRNAs with and without these extra Gs also revealed that the additional nucleotides altered the predicted secondary structure (Fig. 1B), suggesting that even minor sequence additions could affect RNA folding and function. Design of transcription template suitable for circularization. (A) Efficient in vitro transcription with T7 RNA polymerase requires minimal T7 promoter and GGG at the transcription start site. Linear precursor RNA is generated by addition of the polymerase and nucleotides to the DNA template at 37°C for 2–16 h. (B) RNAfold modeled structures of circRNAs generated from unpermuted transcription template with extra GGG were different from circRNAs with no extra Gs, where transcription was initiated from an internal GGG site (permuted transcription template). (C) Schematic representation of the permutation strategy, showing that 5′ and 3′ ends were different between linear RNAs from unpermuted and permuted transcription templates. Subsequent circularization produced contiguous circRNAs that were ligated at different junctions. (D) Treatment with XRN1 which degrades 5′-monophosphorylated RNAs showed that addition of excess GMP to GTP in in vitro transcription reactions resulted in higher amounts of 5′-monophosphorylated linear precursors than other methods. Representative data are from a mean of n = 3 technical replicates with standard error of the mean (SEM). (E) Tapestation analysis of transcribed RNA showed that in vitro transcription reactions incubated at 37°C for 2 h had fewer shorter contaminating RNAs compared to incubations for 16 h. (F) in vitro transcription with only GTP had higher yields at 16 h than 2 h. Addition of 5× or 10× GMP reduced the yields for 16 h reactions which was comparable to the yields at 2 h. 10× GMP at 2 h had the lowest yields among all of the conditions. Representative data are from a mean of n = 3 technical replicates. (G) Efficiency of 5′-monophosphosphate transcripts was comparable between 5× GMP and 10× GMP in vitro transcription reactions. Representative data are from a mean of n = 3 technical replicates with SEM. To preserve structural integrity, we redesigned the DNA template leveraging GGGs found within the sequence. Because the resulting circRNA is sequence–contiguous, transcription of its linear precursor can in principle begin at any point along the RNA sequence without changing the final circular product. We therefore identified 15 naturally occurring GGG sites within the eGFP–IRES sequence and selected GGG located at position 375 as a new transcription initiation site generating a permuted DNA template (Fig. 1C). Thus, a permuted DNA template with internal GGG as the transcription start site was chosen for use. Generation of 5′–monophosphorylated precursors for enzymatic ligation Ligation requires 5′-monophosphorylated RNA, but in vitro transcription reactions produce RNAs with 5′-triphosphates. Hence, we optimized the generation of linear precursors with appropriate 5′ ends. To achieve this, we evaluated three approaches for producing monophosphorylated transcripts (Supplementary Fig. S1) [10]. In the first method, in vitro transcribed 5′-triphosphate RNAs were dephosphorylated to 5′-hydroxyls using alkaline phosphatase and subsequently monophosphorylated by T4 polynucleotide kinase. The second method employed RNA 5′-pyrophosphohydrolase (RppH), which directly converts 5′-triphosphates to 5′-monophosphates. The third approach leveraged the idea that since the first nucleotide in the transcript is always a G, co–transcriptional incorporation of GMP, by substituting a portion of GTP with excess GMP during the in vitro transcription reaction, would result in transcripts commencing with 5′-GMP [13]. To determine the efficiencies of each method, we treated the resulting RNA product with XRN1, an enzyme that selectively degrades 5′- monophosphorylated RNAs, and quantified the fraction of degraded transcripts (Fig. 1D). The co–transcriptional GMP method proved most effective, yielding 51% monophosphorylation with 10× GMP and 48% with 5× GMP, compared with 37% for the phosphatase-kinase approach and 32% for the RppH method (Fig. 1D). Based on these results, supplementing the in vitro transcription reaction with GMP was selected as the preferred strategy for generating 5′-monophosphorylated precursors suitable for circularization. Improvement of linear RNA transcription for generation of precise circRNAs Having optimized the transcription start site and established a reliable strategy for generating 5′-monophosphorylated precursors, we next focused on ensuring that the in vitro transcription reactions produced linear RNAs of precise length and integrity. Prolonged in vitro transcription can yield a mixture of full-length and truncated transcripts, which would result in undesired and heterogeneous circular products after ligation. TapeStation analysis revealed that long in vitro transcription reactions (16 h) produced truncated transcripts alongside full–length products, whereas shorter times (2 h) resulted in cleaner preparations with minimal small transcripts (Fig. 1E). RNA yields were higher for a typical in vitro transcription reaction with triphosphate nucleotides for a 16 h incubation compared to a reaction with excess GMP and lower GTP; however, inclusion of GMP during synthesis equalized yields between 16 and 2 h reactions (Fig. 1F). We also compared monophosphorylation efficiencies under optimized short-transcription conditions. Using the XRN1 assay as before, 2 h in vitro transcription reactions with 5× or 10× GMP generated transcripts with 59% and 54% 5′-monophosphorylation, respectively (Fig. 1G). Based on this balance of accuracy, efficiency, and yield, we selected a 2 h transcription reaction time using 5× GMP and the permuted DNA template with internal GGG as our standard method for generating linear precursors for circularization. Minimizing 3′-end heterogeneity of in vitro transcribed RNAs While optimizing transcript initiation and phosphorylation improved the 5′ accuracy of linear precursors, the 3′ ends of T7–transcribed RNAs can also exhibit heterogeneity due to the addition of non-templated nucleotides by T7 RNA polymerase [14, 15]. The number and identity of these nucleotides can vary depending on the terminal sequence of the DNA template, introducing unwanted sequence variability into the resulting circRNAs. To assess this variability in our eGFP–IRES transcripts, we analyzed the 3′-ends by adding poly(A) tails and performing reverse transcription with oligo-dT followed by PCR amplification and Sanger sequencing. We found that the 3′ ends of the linear precursor had extraneous nucleotides (23.5% of C, 17.6% of T, 17.6% of TC, 11.8% of G, and 5.9% of CC) (Fig. 2A). Such heterogeneity would produce circRNAs with variable ligation junctions, compromising sequence and protein fidelity. We could not reliably determine whether T7 RNA polymerase added additional adenosines since the 3′ end sequencing was dependent on the addition of poly(A) tails. Improvement of in vitro transcription conditions for synthesizing precise precursor linear RNAs. (A) Sanger sequencing of 3′ ends of the linear RNA transcribed from unmodified DNA templates showed heterogeneity. (B) Schematic of amplification of DNA templates using reverse primer with two 2′-O-methyl RNA nucleosides at 3′ end to synthesize transcripts with homogeneous ends. (C) Sanger sequencing revealed homogenous ends of RNAs transcribed from modified DNA templates. (D) RNAfold modeled structures showed that addition of dinucleotides CC and CU at the 3′ends of linear precursor drastically impact the structure of the downstream circRNAs while the addition of G/T/C had no influence. To overcome the issue of extra nucleotides, we sought to improve the homogeneity of the transcription termination site. The templates for in vitro transcription were amplified using a modified reverse primer containing two consecutive 2′–O–methyl RNA residues at the 3′-terminus, which blocks T7 RNA polymerase from adding non-templated nucleotides after run–off transcription [16]. Use of this modified template significantly improved 3′ end precision, as confirmed by RT-PCR and Sanger sequencing of the in vitro transcription products (Fig. 2B and C). Structural modeling further demonstrated that incorporation of dinucleotides CC and TC significantly influenced a portion of the circRNA structure while addition of extra nucleotides G/C/T/A did not impact the configuration (Fig. 2D and Supplementary Fig. S2). This strongly highlights the importance of a precise 3′ end of the linear transcript to ensure uniform circRNA sequences and structures to produce only the desired RNA or protein. Our final optimized in vitro transcription protocol employed a permuted DNA template starting with a previously internal GGG to avoid extraneous Gs incorporated into the final circRNA and amplified using a reverse primer containing two 2′–O–methyl residues to ensure accurate 3′ ends. The template was used in a 2 h transcription reaction including a 1:5 ratio of GTP:GMP to achieve high 5′-monophosphorylation. This combination consistently produced high–quality linear RNAs suitable for precise enzymatic circularization. Evaluation of ligases for efficiency and accurate circulation To identify the most reliable enzymatic ligation conditions, we tested three ligases: T4 DNA ligase, RNA ligase 1, and RNA ligase 2. Each enzyme has distinct substrate preferences for forming a 3′–5′ phosphodiester bond between a 5′-monophosphate donor and a 3′-hydroxyl acceptor. DNA ligase requires a local region of double strandedness [17] whereas RNA ligase 2 tolerates both single- and double-stranded substrates but ligates double stranded region more efficiently [18]. To facilitate ligation by these enzymes, we introduced a splint: a short nucleic acid complementary to the 5′ and 3′ ends of the linear RNA to bring them together and promote circularization. For DNA ligase, which requires a DNA splint, we designed a 40–nt long ssDNA with 20-nt complementarity to either end of the linear RNA. RNA ligase 2, which functions in RNA:RNA or RNA:DNA duplexes, was tested using the same DNA splint or an RNA splint. The 42-nt RNA splint included two unpaired Gs at the 5′-end resulting from the GG required to initiate T7 RNA polymerase-based transcription of the RNA splint. DNA ligase 1 requires perfect complementarity of the splint to RNA for ligation, while RNA ligase 2 can accommodate a few mismatches with the splint around the ligation site. RNA ligase 1 primarily acts on single–stranded RNA substrates and exhibits relatively low reaction specificity, which can lead to mis–ligation events depending on RNA length and structure of the RNA ends [19]. Some studies have improved ligation efficiency and accuracy by using RNA ligase 1 together with imperfect DNA or RNA splints that leave a few nucleotides unpaired at each end [19, 20]. In our study, we used RNA ligase 1 without a splint. To evaluate circularization, 5′–monophosphorylated linear RNAs generated from both unmodified and modified DNA templates were subjected to ligation with each of the three enzymes. Because ligation reactions also produce unwanted products such as unligated linear RNA and linear or circular concatemers, we implemented an RNase R–based strategy to eliminate linear RNA contaminants. RNase R selectively digests linear RNAs with accessible 3′-hydroxyl ends. Its efficiency can be enhanced by polyadenylating the RNA substrates, which results in more accessible 3′ ends for RNase R degradation. Thus, ligation products were first poly(A)–tailed using poly(A) polymerase and then treated with RNase R (Fig. 3B). Circularization was assessed using a 3% denaturing urea–PAGE, where circRNAs can be distinguished from linear RNAs because they migrate slower than their linear counterparts of identical length and sequence [21]. Circularization efficiencies of DNA ligase and RNA ligases 1 and 2 were calculated as the ratio of RNA remaining after ligation, poly(A) tailing, and RNase R with respect to the amount of input linear RNA (Fig. 3C). Distinct ligation efficiencies were observed among the enzymes and between unmodified and modified DNA templates (Fig. 3C). We found that use of modified DNA templates improved ligation yields, especially for DNA ligase 1 and RNA ligase 2, both of which depend on accurate base pairing of the splint at the junction. DNA ligase produced the lowest yields (1.2% with unmodified and 3.2% with modified template) while RNA ligase 1, known for its higher base-pairing permissiveness, achieved moderate efficiencies of 5.4% and 5.9% showing little dependence on template modification. RNA ligase 2 exhibited the highest circularization efficiency of 17% when linear RNAs from modified templates were ligated with an RNA splint, compared to 8% for RNAs from unmodified templates under the same conditions. When a DNA splint was used, RNA ligase 2 reactions produced yields of 8.6% and 10.4% for unmodified and modified templates, respectively. Following the successful formation of circRNAs, urea–PAGE analysis showed multiple bands corresponding to unligated and partially ligated RNAs that persisted even after RNase R digestion. Treatment with RNase R from a different manufacturer performed similarly and failed to remove these contaminants (Supplementary Fig. S3). We also note that contaminating bands were most prominent for reactions from modified DNA templates that were ligated using RNA ligase 2 with RNA splint reactions, likely contributing to the high 17% yield observed post poly(A) and RNase R treatments (Fig. 3C). Subsequent analysis of the ligation junctions by RT-PCR and Sanger sequencing confirmed that circRNAs derived from modified linear precursors were accurate. Junctions formed from unmodified templates were accurate for DNA ligase 1 and RNA ligase 2 but showed errors with RNA ligase 1, consistent with its lower substrate specificity (Fig. 3D). Based on both yield and sequence accuracy, RNA ligase 2 proved most efficient for generating precise circRNAs from modified templates; however, contaminating RNAs still persisted in these samples. Evaluation of circularization efficiency and accuracy by different ligases. (A) Schematic depicting requirements and features of DNA ligase, RNA ligase 1, and RNA ligase 2. (B) Workflow of circRNA generation using enzymatic ligation and RNase R based purification which can be improved by addition of poly(A) tails to linear RNAs. (C) 3% urea–PAGE showed that all ligases were able to circularize 5′-monophosphate RNAs. Boxed bands depict the circRNAs that run slower than their linear counterparts. Contaminating RNAs of lower and higher size than circular or linear RNA were also observed suggesting that poly(A) tailing and RNase R treatments were insufficient to degrade them. Efficiency of ligation was calculated as percentage yields of RNAs remaining after all treatments divided by input RNA for each ligation reaction. CircRNAs derived from modified transcription templates (mod) had higher efficiencies particularly for DNA ligase and RNA ligase 2 than those derived from unmodified templates (unmod). RNA ligase 2 had the highest circularization efficiency, especially with circRNAs derived from mod templates in presence of an RNA splint. Representative data are from a mean of n = 3 technical replicates with SEM; (*) P ≤.05 (unpaired t-test). (D) Sanger sequencing of ligation junctions showed accurate sequences with circRNAs derived from modified transcription templates using all ligases. CircRNAs made with RNA Ligase 1 had errors with the linear RNAs from unmodified templates which were corrected with the use of modified templates. Purification and validation of circRNAs As observed in our earlier experiments, RNase R treatment alone was insufficient to completely remove unligated or partially ligated RNA species from the circRNA samples. To obtain homogeneous preparations, we therefore performed gel purification to isolate pure circRNAs and eliminate remaining contaminants. Circular bands were more intense in the urea-PAGE gel electrophoresis following ligation compared to those observed after poly(A) tailing and RNase R treatment (Supplementary Fig. S4), suggesting a reduction of circRNAs during subsequent processing steps. To maximize recovery and purity, we conducted gel extractions immediately after ligation instead of subsequent to poly(A) tailing and RNase R treatment. We tested different gel extraction methods on circRNAs generated using RNA ligase 2 and a DNA splint. The desired bands were excised and recovered using an optimized crush–and–soak urea–PAGE extraction protocol. Ligated circRNAs were also separated from unligated linear RNAs of identical length and sequence using EX E-gel electrophoresis (Supplementary Fig. S5) [22]. The slower migrating band above the linear RNA band consistent with circRNA was extracted either through the same crush-and-soak approach or using a commercial column–based kit. As controls, linear RNAs were also purified using all three methods: urea–PAGE extraction, EX E-gel + crush-and-soak extraction protocol, and EX E-gel + column purification. To evaluate RNA integrity and purity after extraction, samples were analyzed on a 3% urea–PAGE gel (Fig. 4A). Intact linear transcripts were recovered efficiently by all three methods. For circRNAs, however, lower–migrating bands appeared in all extraction methods but were particularly strong in samples purified using the EX-E-gel + column kits suggesting nicking occured during extraction. The migration pattern of these bands matched that of linear RNA, suggesting that the harsher chemical conditions used in commercial kits may contribute to circRNA hydrolysis. We chose to purify circRNAs from urea–PAGE using the crush–and–soak protocol because it resulted in intact circRNAs. Using this method, we recovered 8.5% circRNAs that were generated from modified DNA templates using RNA ligase 2 with DNA splint, which was a slight decrease from the 10.4% observed after poly(A) tailing and RNase R treatment. CircRNAs ligated with RNA ligase 2 using an RNA splint were also extracted from urea–PAGE to assess post-purification yields (Supplementary Fig. S6). Following urea–PAGE purification, the yields of circRNAs derived from modified DNA templates was 7.1%, a sharp decline from the 17% circularization observed after poly(A) tailing and RNase R treatment confirming that the original higher yield was due to contaminating RNAs. As urea–PAGE purification of circRNAs ligated with RNA ligase 2 using a DNA splint produced slightly higher yields than those ligated with RNA splint, we used RNA ligase 2 with a DNA splint for further experiments. Purification of circRNAs and extending the RNA ligase 2 (RL2)-dependent circularization method to other RNAs. (A) CircRNAs synthesized with RNA ligase 2 using DNA splint were purified using three different approaches: from 3% urea–PAGE using crush and soak method, or from EX E-gels either using the crush and soak method or using column-based kit. As a control, linear RNAs were also extracted using the same methods. CircRNAs extracted from 3% urea–PAGE or EX E-gel using a crush and soak method had more intact circRNAs with less nicking compared to those extracted from EX E-gel using column-based kits. Linear RNAs on the other hand remained intact with each of the approaches. (B) Schematic of RNase-H based circularity confirmation assay that uses a short ssDNA probe which cleaves intact circRNAs into a single linear band, while nicked circRNAs or linear RNAs are cut into two shorter bands. (C) RNase-H based assay confirmed circularity of EGFP-IRES circRNAs. Linear RNAs were cleaved into two shorter bands of expected sizes while circRNAs derived from modified DNA templates were linearized to the size of full-length linear precursor. (D, E) 5′-monophosphate linear precursors of human immunodeficiency virus (HIV) and mCherry were ligated using RNA ligase 2 and respective DNA splints. For circHIV, urea–PAGE purification of circRNAs derived from modified templates had the highest yields with the least contaminating RNAs. Yields of mCherry circRNAs were much higher with polyA + RNase R approach on RNAs from modified template ligated using RNA ligase 2, however urea–PAGE showed higher and lower sized undesired RNAs. Representative data are from a mean of n = 3 technical replicates with SEM. (F) Sanger sequencing confirmed accuracy of circRNAs. Clean chromatograms were observed for ligation junctions of both HlV and mCherry circRNAs derived from modified DNA templates purified either through poly(A) tailing and RNase R treatment or from urea–PAGE purification. To confirm that the purified products were indeed circRNAs, we performed an RNase H–based cleavage assay. Since RNase H specifically cleaves RNA within RNA–DNA hybrids, treatment of a linear RNA substrate in the presence of a complementary DNA oligonucleotide would yield two shorter products, whereas cleavage of a circRNA would result in linearization to produce a single product (Fig. 4B). A 20–nt DNA probe complementary to a region within the RNA was used. As expected, linear RNAs generated from modified DNA templates were efficiently cleaved into two fragments of ∼320-nt and ∼1002-nt (Fig. 4C). By contrast, circRNAs derived from modified templates produced a single band corresponding to the linearized circRNA, confirming their circularity (Fig. 4C). Taken together, these results demonstrate that precise circRNAs, free of extraneous sequences and linear contaminants are synthesized through our optimized approach with yields of 8.5%. This method integrates (i) a permuted DNA template with internal Gs and a modified reverse primer containing two 2′–O–methyl residues for accurate 5′ and 3′ ends, (ii) RNA ligase 2 mediated ligation, and (iii) careful urea–PAGE–based purification. The resulting circRNAs are structurally uniform and suitable for downstream functional and therapeutic investigations. Application of the optimized method to additional RNAs To test the versatility of our approach, we applied it to two additional circRNAs of different lengths and sequences: a 655–nt HIV–derived RNA and a 766–nt mCherry RNA, both shorter than the 1322–nt eGFP–IRES construct. Permuted DNA templates were designed to use internal GGG for T7 polymerase initiation. Linear RNAs from modified DNA templates were synthesized in 2 h in vitro transcription reactions containing 5× GMP, yielding 5′-monophosphorylated precursors that were ligated using RNA ligase 2 and a 40–nt DNA splint. For both HIV and mCherry transcripts, we compared urea–PAGE purification with the poly(A) tailing and RNase R treatment approach (Fig. 4D and E). Efficiencies were calculated as the proportion of RNA recovered after ligation, poly(A) tailing, and RNase R treatment relative to the amount of input linear RNA, or alternatively, as the fraction of RNA recovered after urea–PAGE purification relative to the linear RNA input. For the circHIV construct, urea–PAGE–purified samples contained fewer unligated contaminants compared to those subjected to RNase R treatment (Fig. 4D). The overall yields were 11.5% for urea–PAGE–purified circRNAs and 10% for RNase R–treated samples derived from modified templates (Fig. 4D). Similar trends were observed for the mCherry construct, where urea–PAGE purification yielded purer circRNAs than RNase R treatment. However, unlike HIV circRNA, RNase R–treated mCherry samples displayed additional high–molecular–weight RNA bands migrating above the main circRNA species, likely corresponding to linear and circular concatemers (Fig. 4E). The yields for mCherry circRNAs were 13.6% after urea–PAGE purification and 42% after RNase R treatment of ligated RNAs from modified templates (Fig. 4E). Although RNase R treatment resulted in higher apparent yields for mCherry samples, this increase likely reflected the presence of co–purifying contaminants. Therefore, despite lower recovery, urea–PAGE purification provided a more reliable approach for obtaining high–purity circRNA preparations. To validate the accuracy of the circRNA products, ligation junctions were amplified and analyzed by Sanger sequencing. HIV circRNAs and mCherry circRNAs synthesized from modified DNA templates exhibited accurate ligation junctions. Together, these results demonstrate that our optimized design and synthesis workflow combining permuted templates with 2′–O–methyl primers, co–transcriptional GMP incorporation, RNA ligase 2 ligation, and urea–PAGE purification enables the efficient generation of precise, accurate, and structurally homogeneous circRNAs across multiple sequences and lengths. Discussion An RNA’s primary sequence determines its translation, structure, stability, and interactions with RNA–binding proteins. Thus, maintaining sequence precision is essential for regulating or modeling biological function. Depending on the position of these unwanted extraneous nucleotides, translation frameshifting could occur, compromising protein production and function. Here, we present a robust enzymatic ligation approach for generating precisely defined circRNAs by careful design of the transcription template to achieve unprecedented RNA sequence fidelity. By using internal guanosines as transcription start sites and introducing two 2′–O–methyl residues at the 3′ end of the reverse primer, we maintained the desired sequence during each step of the workflow and improved ligation accuracy. The use of RNA ligase 2 yielded the highest circularization efficiencies with accurate junction sequences. We found that urea–PAGE purification consistently produced circRNAs with fewer linear contaminants compared to RNase R treatment. Moreover, this workflow was applied across different transcript sizes and sequences (eGFP–IRES, HIV, and mCherry), with yields that varied with each RNA and downstream purification methods (Supplementary Fig. S7). For all of the constructs, circRNAs derived from modified DNA templates had higher yields (Supplementary Fig. S7). This improvement showed that we could enhance efficiency of conventional enzymatic ligation by increasing the availability of accurately defined 3′ ends in linear RNA precursors generated from modified DNA templates. Autocatalytic splicing systems traditionally retain long sequences from the introns into the circRNAs that affect RNA function. We note that autocatalytic splicing based on group I and group II introns have been developed to minimize the retention of intronic sequences by concealing these retained sequences (scars) into the IRES or coding regions of the circRNAs [23–25]. Cis-splicing methods that negate the need for spacer sequences have also emerged [26, 27]. These strategies, referred to as “scarless” systems, typically exploit genetic code degeneracy to preserve protein sequence, structure, and function; however, while doing so, they alter the underlying RNA sequence. Conventional enzymatic ligation methods that use linear RNA precursors with extraneous sequences that arise from the T7 RNA polymerase’s requirement for 5′ guanosines or from the 3′ end heterogeneity also significantly affect the RNA sequence. Unlike scarless autocatalytic splicing methods or conventional enzymatic ligation, our approach requires no alteration of the RNA sequence, preserving the native transcript and thereby enabling direct assessment of RNA sequence and structure, its translation efficiency, stability, and fidelity of the translated protein. Autocatalytic scarless systems require careful customization for every new RNA sequence, in contrast, our enzymatic ligation approach relies only on naturally occurring internal guanosine motifs, which are present in nearly all transcripts, allowing easy adaptation to new targets. For additional versatility and adaptability, T7 RNA polymerase can initiate at GG motifs, albeit at lower efficiency than at GGG sites; thus, for shorter constructs, permuted templates starting with two Gs could be evaluated. Since transcriptional efficiency is impacted by primary sequence, selection of optimal GGG or GG initiation sites could include consideration of the +4 to +8 nucleotides [28]. Moreover, permuted templates should be designed so that GGG or GG initiation sites and the 3′ termini are positioned within minimally structured regions, to facilitate efficient ligation. Other RNA polymerases, such as SP6 which has a different motif preference like GAA, could also be used according to the sequence of the RNA that needs to be circularized [29]. Future analysis to compare the stability, translational efficiency, and immunogenicity of circRNAs generated by different purification and synthesis methods will be valuable. In conclusion, our optimized enzymatic ligation approach, based on readily available enzymes and modified primers, offers distinct advantages in universality, design simplicity, and sequence fidelity. This strategy provides a practical and flexible platform for generating precise and high–purity circRNAs suitable for diverse biological and therapeutic applications. Supplementary Material