Assessing and engineering the IscB-ωRNA system for programmed genome editing

Abstract

ωRNA-guided endonuclease IscB, the evolutionary ancestor of Cas9, is an attractive system for in vivo genome editing due to its compact size and mechanistic resemblance to Cas9. However, wildtype IscB/ωRNA systems show limited activity in human cells. Here we report enhanced OgeuIscB, which, with eight amino-acid substitutions, displayed a 4-fold increase in in vitro DNA-binding affinity and a 30.4-fold improvement in insertion/deletion (indel) formation efficiency in human cells. Paired with structure-guided ωRNA engineering, the enhanced OgeuIscB/ωRNA systems efficiently edited the human genome across 26 target sites, attaining up to 87.3% indel and 62.2% base-editing frequencies. Both wildtype and engineered OgeuIscB/ωRNA showed moderate fidelity in editing the human genome, with off-target profiles revealing key determinants of target selection including an NARR target-adjacent motif (TAM) and the TAM-proximal 14 nucleotides in the R-loop. Collectively, our engineered OgeuIscB/ωRNA systems are programmable, potent, and sufficiently specific for human genome editing.

Introduction

IscB is an RNA-guided endonuclease residing in a distinct family of IS200/IS605 transposons^1–3. Despite being 1/3 the size of Cas9, IscB contains all the essential domains found in Cas9, including the HNH nuclease domain, the split RuvC nuclease domain, and in particular, the Cas9-specific arginine-rich bridge helix (Fig. 1a). IscB further encodes an N-terminal PLMP motif-containing domain (PLMP), a C-terminal target-associated motif (TAM) interaction domain (TID, shared with some Cas9 proteins), and a few other degenerate structural motifs. The α-helical recognition lobe (REC; ~600 residues) commonly found in Cas9 is largely absent in IscB. In place of the CRISPR RNA (crRNA) and tracrRNA heteroduplex in the Cas9 ribonucleoprotein (RNP) is a single noncoding RNA, named OMEGA RNA (ωRNA), the structured portion of which serves the equivalent function of the REC domain in Cas9^4–6. Structural studies further revealed strong mechanistic similarity between IscB and Cas9 in target DNA recognition, R-loop formation, and conformational dynamics in the HNH nuclease^4–6.

Figure 1 |. Engineering OgeuIscB for increased genome-editing efficiency.

a, Domain architectures of SpCas9 and OgeuIscB. PLMP, HNH, RuvC, P1D (P1 interaction domain), and TID domains are denoted. Protein lengths are drawn to scale. aa: amino acid. b, Multiple sequence alignment of OgeuIscB and seven IscB orthologs. Representative regions are shown, with candidates for mutagenesis highlighted in red boxes. Full sequence alignment is provided in Supplementary Fig. 1. c, Workflow to determine the cellular activity of OgeuIscB/ωRNA and its derivatives. d, Indel levels at VEGFA_site 1 generated by wildtype OgeuIscB and OgeuIscB variants bearing single-point mutations. Mutations included in each OgeuIscB variant are listed in Supplementary Table 2. Fold changes in indel frequencies are plotted in Supplementary Fig. 2a. wt_nt: wildtype OgeuIscB and non-targeting ωRNA. Three independent replicates were carried out in HEK293T cells (mean ± s.d.). Cells were harvested 48 h post-transfection. P values were determined by two-tailed Student’s t-test. * P ≤ 0.05; ** P ≤ 0.01; *** P ≤ 0.001; **** P ≤ 0.0001. e, Projection of all evaluated mutations onto the cryo-EM structure of the OgeuIscB/ωRNA/DNA complex. PDB entry: 7UTN⁴. Non-target strand DNA: red; target strand DNA: blue; guide region: orange; ωRNA scaffold: grey; protein: light brown; beneficial mutations: magenta spheres; other mutations: gray spheres. The dashed red line delineates a plausible path for the non-target strand to enter the RuvC active site upon locking of the R-loop by the HNH endonuclease domain. f, Surface charge distributions of wildtype OgeuIscB and an OgeuIscB variant with beneficial mutations projected. The structures adopt the same orientation as in e.

While retaining the versatility of conventional Cas9 systems, the compact size renders the IscB/ωRNA system highly attractive for in vivo genome-editing applications⁷. Identified in the human gut metagenome, OgeuIscB/ωRNA has been reported to generate double-stranded breaks (DSBs) in mammalian cells². However, the use of OgeuIscB for programmed genome editing is currently hindered by its overall low editing activity and unstable performance across different target sites.

Practical applications of IscB/ωRNA as genome-editing agents also require careful assessment of its editing fidelity. The OgeuIscB/ωRNA complex has been shown to specify a 16-nt region in target DNA through R-loop formation², which is 4-nt shorter than that by Streptococcus pyogenes Cas9 (SpCas9). This may potentially lead to a higher incidence of offtarget cleavage. Nevertheless, structural studies showed that at least the first 14 base pairs of the ωRNA/target DNA heteroduplex are extensively recognized by OgeuIscB upon R-loop formation (PDB: 8CSZ). In the case of SpCas9, trimming the spacer from 20-nt to 17-nt increased the mismatch sensitivity and reduced off-target DNA cleavage⁸. Whether the specificity of IscB/ωRNA is sufficient for mammalian genome editing has not been systematically evaluated.

In this study, we assessed and engineered the OgeuIscB/ωRNA system for programmed genome editing. We developed enhanced OgeuIscB variants with boosted DNA cleavage activity both in vitro and in human cells. The enhanced activity of engineered OgeuIscB/ωRNA further translates to more robust base editing, an advanced genome-editing technology that directly alters the nucleobase identity without requiring a DSB^9–11. We assessed the fidelity of the OgeuIscB/ωRNA system using genome-wide, unbiased identification of DSBs enabled by sequencing (GUIDE-seq)¹². We detected off-target cleavage in the human genome by the wildtype complex and find that this moderate fidelity is inherited by the engineered OgeuIscB/ωRNA systems. Collectively, we demonstrate that transposon-associated RNA-guided endonuclease IscB, despite being much smaller and more evolutionarily ancient than Cas9, can be engineered into a robust genome-editing agent.

Results

Rational approaches to identify beneficial mutations in IscB

An RNA-guided nuclease may function poorly in eukaryotic cells for many reasons. The nuclease may be a poor DNA binder or cleaver inherently; have trouble accessing target DNA wrapped inside nucleosomes; or have problems with expression or RNP assembly. Our first line of probing using an electrophoretic mobility shift assay (EMSA) revealed that OgeuIscB/ωRNA binds DNA very poorly at concentrations lower than 125 nM (detailed below in the in vitro characterization section). This stands in sharp contrast to the 10–30 nM dissociation constant (K_d) reported for RNA-guided Cas nucleases^13–15.

We hypothesized that introduction of positively charged residues such as lysine (K) and arginine (R) may increase the affinity of OgeuIscB for nucleic acids, which should in turn improve its DNA cleavage activity. Similar strategies have been successfully applied to increase the activity and targeting ranges of SpCas9^{16, 17}, CjCas9¹⁸, AsCas12a¹⁹, BhCas12b²⁰, Cas12i2²¹, UnCas12f1²², AsCas12f1²³, OsCas12f1²⁴, RhCas12f1²⁴, and an eukaryotic transposon-associated endonuclease, Fanzor²⁵. We chose our target residues based on the multiple sequence alignment of IscB orthologs (Supplementary Fig. 1 and Supplementary Table 1), rather than focusing solely on residues near the ωRNA and target DNA, as revealed by the cryo-EM structures^{4, 5}. Aligned positions that harbor positively charged residues in IscB orthologs, but neutral or negatively charged residues in OgeuIscB, were selected for mutagenesis (Fig. 1b). We chose this strategy over a structure-guided approach because the available structures may not have captured all relevant conformational states. This strategy also helps to avoid testing protein variants with major defects in folding or complex formation, as the introduced substitutions are found naturally among IscB orthologs.

We evaluated 36 variants of OgeuIscB, each bearing a single-point mutation, for insertion/deletion (indel) formation at VEGFA_site 1 in human embryonic stem (HEK) 293T cells (Fig. 1c and Supplementary Table 2). Nuclear localization signals (NLS) were appended to both termini of OgeuIscB to facilitate nuclear import. Wildtype OgeuIscB generated 2.0 ± 0.2% (mean ± s.d.) indels at VEGFA_site 1 in our hands (Fig. 1d). This editing level is consistent with previous studies^{2, 5}. Among the 36 single-point mutations, nine increased DNA cleavage in comparison to wildtype OgeuIscB; twelve significantly reduced the indel level; and the rest were neutral (Fig. 1d). We followed up with eight beneficial single-point mutations, M102R, F137K, V159K, N281R, Q324R, Y327K, H368R, and L393K, which increased the indel frequency by 1.6–2.6-fold (Supplementary Fig. 2a).

Although these eight hotspot mutations were not chosen using a structure-guided approach, their beneficial effects can be rationalized to a good extent based on their locations in the OgeuIscB/ωRNA/R-loop structure (PDBs: 8CSZ, 8CTL)⁴. For example, M102 resides in the bridge helix (residues 85–122) that contacts the guide region of the ωRNA at positions 9 to 16 together with a cluster of arginine residues (Fig. 1e)^{4, 5}. Removal of a single arginine from the bridge helix abolished DNA cleavage in DtIsrB (R104A), a nickase homologous to OgeuIscB⁶. We reason that M102R further strengthens the bridge helix/ωRNA interactions and helps with guide RNA display, which in turn promotes R-loop formation. F137 and V159 are both constituents of the REC linker that connects the bridge helix to RuvC II. F137K and V159K add two positive charges to stabilize the mid-region of the RNA/DNA heteroduplex (Fig. 1f). N281 belongs to the HNH endonuclease domain that undergoes conformational changes to lock and stabilize the RNA/DNA heteroduplex; N281R may stabilize the locked R-loop state, which in turn promotes target-strand cleavage by HNH. Q324, Y327, and H368 reside on the surface of the RuvC domain. They are close to the path through which the non-target strand (NTS) DNA enters the RuvC active site. Q324R, Y327K, and H368R likely promote the recruitment and cleavage of NTS by RuvC. L393K may improve the DNA-binding affinity of OgeuIscB by creating a favorable contact with the backbone of TAM-containing DNA. We note that structural studies have not captured the pre-R-loop formation state, nor the partial R-loop state, in which these residues may have different modes of DNA contact. Alternatively, some of the identified mutations may promote DNA binding and cleavage through mechanisms beyond supplementing electrostatic interactions.

Combining beneficial mutations led to enhanced OgeuIscB

Next, we investigated whether OgeuIscB activity could be further improved by combining beneficial mutations. OgeuIscB variants carrying double mutations were generally more active than their single-mutation counterparts (Fig. 2a, Supplementary Fig. 2b, and Supplementary Table 2), suggesting that the beneficial effects are additive. Indeed, activity continued to build as we combined three, four, five, and eight mutations. The best variant OgeuIscB-v6.1, combining all eight mutations (M102R/F137K/V159K/N281R/Q324R/Y327K/H368R/L393K), boosted the indel level at VEGFA_site 1 to 18.6 ± 2.2%, a 9.2-fold increase over the wildtype protein activity. Importantly, the same pattern was recapitulated when we programmed the combinatorial OgeuIscB variants to target EMX1_site 1, albeit to a lesser extent (18.9 ± 0.8% indels with OgeuIscB-v6.1, compared to 6.7 ± 0.3% by wildtype OgeuIscB; Fig. 2b and Supplementary Fig. 2c). These findings suggest the activity improvement to OgeuIscB is independent of target sequences. We hereby rename OgeuIscB-v6.1 as enhanced OgeuIscB, or enOgeuIscB.

Figure 2 |. Characterization of enOgeuIscB in human cells and in vitro.

a, Indel levels at VEGFA_site 1 generated by OgeuIscB variants bearing 2–8 mutations. OgeuIscB-v2, 3, 4, 5, and 6 denote OgeuIscB variants carrying 2, 3, 4, 5–7, and 8 single-point mutations, respectively. b, Indel levels at EMX1_site 1 generated by selected OgeuIscB variants. c, Metaplot showing insertion and deletion patterns generated by wildtype OgeuIscB, OgeuIscB-v5.1, and enOgeuIscB. d, Electrophoretic mobility shift assay of target DNA with OgeuIscB/ωRNA and enOgeuIscB/ωRNA. The experiment was repeated independently twice and the representative result is shown. For a and b, three independent replicates were carried out in HEK293T cells (mean ± s.d.). Cells were harvested 48 h post-transfection. Fold changes in indel frequencies are plotted in Supplementary Fig. 2b and 2c. wt_nt: wildtype OgeuIscB and non-targeting ωRNA.

Deep sequencing analysis²⁶ revealed that all OgeuIscB variants caused predominantly small deletions rather than insertions at the target site (Supplementary Fig. 3), which is consistent with a previous study². The indel pattern formed by enOgeuIscB was similar to that by wildtype OgeuIscB (Fig. 2c), indicating that the mutations did not alter the cleavage pattern of the RNA-guided nuclease.

To investigate whether our engineering efforts improved the DNA binding or cleavage activity of OgeuIscB/ωRNA, we purified wildtype and enhanced OgeuIscB in complex with the VEGFA_site 1-targeting ωRNA from E. coli for biochemical analysis. In EMSA experiments, enOgeuIscB/ωRNA bound the DNA target at concentrations as low as 31 nM, corresponding to a 4-fold improvement over the wildtype complex (Fig. 2d). A similar 4-fold improvement was recapitulated in the concentration-dependent cleavage assay, where DNA cleavage was apparent in the presence of 31 nM of enOgeuIscB/ωRNA, as opposed to 125 nM of the wildtype complex (Extended Data Fig. 1a). These results suggest the mutations improved the affinity of OgeuIscB for the DNA target. In contrast, little activity difference was found between wildtype and enhanced OgeuIscB in the time-course cleavage assay at a higher RNP concentration (500 nM; Extended Data Fig. 1b), suggesting that the mutations did not alter the rate-limiting steps in RNA-guided DNA cleavage. Collectively, we conclude that our identified mutations enabled enOgeuIscB to locate the DNA target more efficiently, a feature that may prove particularly beneficial within the complex context of the human genome.

Structure-guided ωRNA engineering

Guide RNA engineering has been shown to enhance the genome-editing performance of several Cas nucleases^{22, 23, 27–29}. The 222-nt ωRNA folds into a tertiary structure significantly more sophisticated than its counterpart in the CRISPR-Cas9 system (Extended Data Fig. 2a)^{4, 5}. It also makes more extensive interactions with its protein partner, OgeuIscB. Given that RNA folding and assembly may represent a unique challenge to IscB/ωRNA, we explored ways to engineer the ωRNA as an orthogonal strategy to further improve editing efficiency.

We removed 20 nucleotides 3’ to P5 as they do not contribute to the assembly of OgeuIscB/ωRNA^{4, 5}; P5 was repurposed as a transcription terminator for human pol III by appending eight consecutive Us to its 3’ end. We further truncated a significant portion of P1, P4, and P5 helices in ωRNA because these distal regions are not contacted by OgeuIscB, hence their removal will unlikely disrupt the RNA-guided DNA cleavage^{4, 5}. Specifically, the poorly defined distal region of P1 (U31-U45) was replaced with a GAAA tetraloop; the distal region of P4 (G147-U156) was replaced with either a sephadex-binding aptamer D8³⁰ (ωRNA-v1) or a GAAA tetraloop (ωRNA-v2); and the P5 triloop (U191A192A193) was optimized to a more stable GAAA tetraloop. The OgeuIscB/ωRNA structures further define three A•C mismatches in P1, P2, and J2. Lacking obvious functions in the OgeuIscB/ωRNA structure, we consider these mismatches “weak spots” as they may lead to alternative RNA structure formation, especially in the context of truncated ωRNAs. We therefore introduced three point mutations, C24U, A88G, and C138U, to reinstate perfect Watson-Crick base pairing in P1, P2, and J2, respectively (Fig. 3a, Extended Data Fig. 2b, and Supplementary Table 3).

Figure 3 |. Structure-guided ωRNA engineering.

a, Schematic of ωRNA variants. b, Indel levels at VEGFA_site 1 generated by ωRNA variants paired with wildtype and engineered OgeuIscB. Sequences of ωRNA variants are provided in Supplementary Table 3. All ωRNA variants were evaluated in HEK293T cells (mean ± s.d.; n = 3 independent experiments). Cells were harvested 48 h post-transfection. Fold changes in indel frequencies are plotted in Extended Data Fig. 2c.

When directed to VEGFA_site 1, ωRNA-v1 and ωRNA-v2 led to a global increase in indel levels across all tested OgeuIscB variants: wildtype OgeuIscB, OgeuIscB-v1.3, OgeuIscB-v3.3, OgeuIscB-v5.1, and enOgeuIscB (Fig. 3b and Extended Data Fig. 2c). ωRNA-v2 was particularly potent, generating 1.2–4.3-fold more indels than the wildtype ωRNA. These results confirmed that the peripheral regions of P1 and P4, along with extra nucleotides succeeding P5, are dispensable in the OgeuIscB/ωRNA complex.

Encouraged by the orthogonal boost from ωRNA engineering, we constructed 14 more ωRNA variants to explore additional ways to improve the editing activity (Fig. 3). Protecting the guide region by appending 5’-to-3’ exoribonuclease-resistant RNA (xrRNA)³¹ (ωRNA-v1 vs. v3 and ωRNA-v2 vs. v5) or Csy4 RNA³² (ωRNA-v2 vs. v7 and ωRNA-v2 vs. v8) to the 5’ end of the ωRNA did not improve indel formation. Removal of an internal nucleotide flip-out in P5 (A201) lowered the indel levels (ωRNA-v4 vs. v5 and ωRNA-v6 vs. v7), whereas reintroduction of extra sequences following P5 had no discernible effect (ωRNA-v2 vs. v9). P5 interacts with the PLMP domain of OgeuIscB, deletion of which only mildly reduced the DNA cleavage activity of the OgeuIscB/ωRNA RNP in vitro⁴. However, OgeuIscB-ΔPLMP failed to generate appreciable levels of indels in human cells (Extended Data Fig. 2d), suggesting that the interactions between P5 and PLMP, albeit dispensable for target DNA cleavage, are likely essential for RNP assembly or stability. Consistently, deletion of the PLMP domain in DtIsrB was shown detrimental to its nickase activity⁶. Collectively, both the accessibility of the 5’ end and the integrity of P5 in ωRNA promote indel formation by OgeuIscB/ωRNA in human cells.

In ωRNA-v10–12, we further truncated ωRNA-v2 at P1, P2, and P4. While activity was not affected by removing two more nucleotides from P1, a 4-nt truncation in P2 or P4 deletion abolished indel formation (Fig. 3b and Extended Data Fig. 2c). The apical loop of P3 and residues trailing J1 form a pseudoknot integral to the ωRNA tertiary structure. Disruption of the pseudoknot inactivates OgeuIscB/ωRNA⁵. As a result, we deemed alteration to P3 not an option. Thus, ωRNA-v2 represents a minimized ωRNA sensitive to additional stem-loop truncations. While further ωRNA consolidation is conceivable, such effort demands editing of long-range interactions and possible rearrangement of the core structure of the ωRNA.

Lastly, we evaluated the contribution of the three mismatch-correcting mutations (C24U, A88G, and C138U, in P1, P2, and J2, respectively) by reverting them individually and collectively in ωRNA-v2. Reverting A88G in P2 (ωRNA-v14) or all three mutations (ωRNA-v16) markedly reduced the indel levels, whereas reverting C24U in P1 (ωRNA-v13) or C138U in J2 (ωRNA-v15) had minimal impact (Fig. 3b and Extended Data Fig. 2c). These results suggest that A88G was the main contributor to activity improvement in the truncated ωRNA.

An observation worth noting is that all engineered ωRNA variants had lower expression levels than that of the wildtype ωRNA except for ωRNA-v13 and ωRNA-v16 (Extended Data Fig. 2e). However, ωRNAs co-expressed with enOgeuIscB consistently showed higher abundances than those paired with wildtype OgeuIscB. This enhanced protection may arise from more efficient assembly between enOgeuIscB and ωRNA, although this hypothesis requires additional evidence to substantiate. Collectively, despite its complex tertiary structure and multifaceted function, we successfully engineered the ωRNA with shortened sequences (186 nt vs. 222 nt for the wildtype ωRNA) and increased potency. We proceeded with ωRNA-v2 and ωRNA-v13.

Evaluating the indel formation efficiency of enOgeuIscB/ωRNA

We evaluated the genome-editing performance of OgeuIscB/ωRNA across 26 target sites in HEK293T cells (Supplementary Table 4). All target sites feature 3’-NWRRNA (W = A or T; R = A or G), a TAM determined by an in vitro plasmid cleavage assay². Wildtype OgeuIscB/ωRNA generated 1.6 ± 2.2% indels across the 26 target sites (median: 0.6%; Fig. 4a, Fig. 4b). These results are consistent with a previous study showing that although OgeuIscB/ωRNA can generate DSBs in human cells, its activity is generally low and fluctuates across target sites². In contrast, enOgeuIscB/ωRNA produced 18.6 ± 12.2% indels across the 26 target sites (median: 17.9%), corresponding to an average 30.4-fold increase in activity. Specifically, enOgeuIscB edited the 12 sites rejected by the wildtype protein (<0.5% indels; 0.2 ± 0.1%) to 10.6 ± 6.3%. The enhanced activity, coupled with broader access to target sites, underscores the success of our protein engineering efforts.

Figure 4 |. Genome editing facilitated by engineered OgeuIscB/ωRNA.

a, Indel frequencies mediated by wildtype OgeuIscB and enOgeuIscB paired with the wildtype ωRNA in HEK293T cells. b, Violin plot showing the distribution of indel levels achieved by OgeuIscB/ωRNA and enOgeuIsc/ωRNA across 26 target sites. c, Indel levels achieved by HA tag-bearing and HA tag-free enOgeuIsc and enIscB³³ paired with the wildtype ωRNA across 26 target sites. d, Indel frequencies generated by enOgeuIscB/ωRNA-v2 and enIscB/ωRNA* at 6 target sites. e, Indel frequencies generated by enOgeuIscB/ωRNA-v2 and enOgeuIscB/ωRNA-v13 encoded in a single plasmid at six target sites in HEK293T cells. Editing outcomes in A549 and HeLa cells are provided in Extended Data Fig. 6. f, A:T-to-G:C editing at VEGFA_site 1 generated by individual fusions of TadA8e^V106W and enOgeuIscB (D60A) paired with the wildtype ωRNA, ωRNA-v2, and ωRNA-v13. Heatmaps are plotted for the entire 16-nt protospacer using editing rates averaged from three independent experiments. g, A:T-to-G:C editing achieved by wildtype OgeuIscB/ωRNA, enOgeuIscB/ωRNA-v2, enOgeuIscB/ωRNA-v13, and enIscB/ωRNA* in the dual TadA8e^V106W architecture across six target sites. The best edited A sites are plotted for individual target sites. Editing events occurred in the entire protospacers are plotted in Extended Data Fig. 7. All genome-editing experiments shown in the figure were carried out in HEK293T cells (mean ± s.d.; n = 3 independent experiments). Cells were harvested 72 h post-transfection. For both b and c, the width of the violin represents the density of data points at each value. The thick centerline and the plus sign within the violin indicate the median and mean, respectively, while the dashed lines represent the first and third quartiles. The violin plot is truncated at the minimum and maximum values. n = 26 target sites.

We next assayed ωRNA-v2 and ωRNA-v13 in combination with either wildtype OgeuIscB or enOgeuIscB across eight genomic loci. At VEGFA_site 1, VEGFA_site 2, EMX1_site 2, and TTLL11_site 3, ωRNA-v2 and ωRNA-v13 boosted indel levels by 1.3–4.1- and 1.3–3.4-fold, respectively, compared to the wildtype ωRNA (Extended Data Fig. 3a). However, editing levels with ωRNA-v2 and ωRNA-v13 at CXCR4_site 1 and DYNC1H1_site 1 were comparable to wildtype ωRNA levels and decreased at EMX1_site 1 and DNMT1_site 3. Similar trends persisted when testing enOgeuIscB/ωRNA-v2 across 26 genomic loci—while the average indel level increased compared to enOgeuIscB/ωRNA (20.2 ± 17.7% vs. 18.6 ± 12.2%), not all genomic sites benefited from the engineered ωRNA scaffold (Extended Data Fig. 3b and 3c). These results suggest that the sequence in the guide region may influence ωRNA folding, which could in turn affect RNP assembly. Consistent with this speculation, the extra protective effect observed with enOgeuIscB for VEGFA_site 1-targeting ωRNA-v2 and ωRNA-v13 diminished when these ωRNAs were targeting EMX1_site 1 (Extended Data Fig. 3d). Additional engineering effort is required to further improve the robustness of ωRNA folding and reduce its activity fluctuation.

Our OgeuIscB constructs carry three consecutive human influenza hemagglutinin (HA) epitopes at the C terminus, each bearing two aspartic acid residues. These negative charges may repel nucleic acids. Indeed, eliminating the 3× HA tag improved indel formation efficiency at all target sties, regardless of the protein identity—1.7–6.5-, 1.3–3.7-, and 1.0–2.3-fold more indels were observed with wildtype OgeuIscB, OgeuIscB-v5.1, and enOgeuIscB, respectively (Extended Data Fig. 3a and 4a–c). The activity improvement persisted when transitioning from the wildtype ωRNA to ωRNA-v2 and ωRNA-v13. Consistent with our interpretation that the negative charges on the 3× HA tag reduced indel formation, we found that positively charged tags slightly increased the activity of wildtype OgeuIscB/ωRNA (Supplementary Fig. 4). However, the same tags on enOgeuIscB decreased the indel frequency, likely due to an oversaturation effect from an excess amount of positively charged residues. These results support our hypothesis that electrostatic interactions between OgeuIscB, ωRNA, and target DNA have a major impact on the DNA targeting and cleavage activity of the complex.

We further compared the performance of our enOgeuIscB with a recently published engineered OgeuIscB variant enIscB³³. This variant was previously evaluated after sorting cells for IscB expression, whereas our indel frequency measurement did not include this pre-selection step. To enable an equal-footing comparison, we cloned the two OgeuIscB variants into the same vectors, and compared their indel-generating capacities without pre-selection. When paired with the wildtype ωRNA, the two variants showed nicely correlated editing levels across 26 target sites. In the HA-tagged format, our enOgeuIscB outperformed enIscB by 1.47-fold (Fig. 4c and Extended Data Fig. 5a). Upon HA tag removal, the two variants generated comparable levels of indels (Fig. 4c and Extended Data Fig. 5b). When paired with their corresponding best-performing ωRNA, the indel frequencies increased steadily (Fig. 4d). Collectively, enOgeuIscB demonstrated similar indel-forming capacity and better compatibility with the HA tag compared to the reported enIscB.

Considering the uptake challenges associated with co-administration of OgeuIscB- and ωRNA-encoding plasmids, we cloned OgeuIscB and ωRNA into a single plasmid and reassessed their performance. We observed 64.5–87.3% indels with enOgeuIscB/ωRNA-v2 and enOgeuIscB/ωRNA-v13 across six genomic loci, without imposing any means of selection (Fig. 4e). These indel levels approach the practical limit for gene editing in HEK293T cells via plasmid transfection.

Since IscB-mediated genome editing has not been reported for cell lines beyond HEK293, we carried parallel experiments in A549 and HeLa cells, and attained indel levels up to 15.8% and 23.7%, respectively (Extended Data Fig. 6). Note that these two cell lines are known to be more refractory to lipid-mediated plasmid delivery, which could have caused the reduced indel levels. Editing potency broadly translated from HEK293T to A549 and HeLa cells across different target sites. One exception was TTLL11_site 3 in HeLa cells, which hosts a single-nucleotide variation 3 nt into the TAM-proximal R-loop region. enOgeuIscB/ωRNA rejected this mismatch in the TAM-proximal RNA-guided region, as indel formation was reduced to the background level. Collectively, enOgeuIscB/ωRNA variants show robust indel formation activity across multiple sites in human cells.

Base editing mediated by enOgeuIscB/ωRNA

Compared to existing type-V mini-CRISPR proteins that process DNA with a single RuvC domain^34–37, IscB hosts two separate nuclease domains, each dedicated to the cleavage of one DNA strand. This makes IscB a more versatile platform to support CRISPR 2.0 applications, such as base editing^9–11. Herein, we evaluated TadA8e-fused³⁸ RuvC-inactivated (D60A) OgeuIscB nickase for A:T-to-G:C editing in HEK293T cells. A V106W mutation was installed to improve the editing fidelity of TadA8e^{38, 39}. While single TadA8e^V106W fusions at either the N or C terminus of enOgeuIscB showed moderate activity (15.3–28.5% editing in the presence of the wildtype ωRNA, ωRNA-v2, or ωRNA-v13, considering the best edited A at VEGFA_site 1), the dual TadA8e^V106W fusion (TadA8e^V106W-enOgeuIscB-TadA8e^V106W) boosted the editing level to 45.6–57.2% (a 1.9–3.5-fold increase compared to single TadA8e^V106W fusions, Fig. 4f and Extended Data Fig. 7a). Consistent with a previous report³³, IscB-derived base editors cover a broad editing window; editing was observed across A₁ to A₁₁ in the R-loop, with peak activity at A₆ (Fig. 4f).

We next evaluated the activity of adenine base editors derived from wildtype OgeuIscB, our enOgeuIscB, and the reported enIscB in the same fusion architecture (TadA8e^V106W-OgeuIscB-TadA8e^V106W) across six genomic loci. As expected, both engineered OgeuIscB variants outperformed the wildtype enzyme (36.1 ± 21.1% editing for enOgeuIscB/ωRNA-v13 and 35.2 ± 24.3% editing for enIscB/ωRNA* versus 20.3 ± 21.2% editing for wildtype OgeuIscB/ωRNA; Fig. 4g and Extended Data Fig. 7b), while enOgeuIscB/ωRNA-v13 generated base editing at levels comparable to enIscB/ωRNA*. Collectively, these findings suggest that although our engineering efforts were devoted to increase cleavage-induced indel formation, the improvements directly translate to base editing. These observations further support the notion that the bottleneck in IscB-mediated DNA cleavage is at the target searching rather than the DNA cleavage step, and that our mutation combination specifically improved this activity (Extended Data Fig. 1a and 1b).

Off-target effects of OgeuIscB/ωRNA systems

We employed GUIDE-seq to assess the genome-wide specificity of OgeuIscB/ωRNA¹². Compared to widely used in vitro off-target profiling methods such as Digenome-seq⁴⁰, CIRCLE-seq⁴¹, and ONE-seq⁴², GUIDE-seq reports all DSB events in live cells—on-target cleavage, off-target cleavage, or natural DSBs—through the integration of a phosphorylated double-stranded oligonucleotide (dsODN). As dsODN integration favors blunt-end over sticky-end DSBs, we first investigated whether GUIDE-seq is sensitive enough to detect IscB-mediated cleavage, which generates 5’-end protruding sticky-ends. We co-administered the dsODN and the OgeuIscB- and ωRNA-encoding plasmids into HEK293T cells via transient transfection. OgeuIscB/ωRNA, enOgeuIscB/ωRNA-v2, and enOgeuIscB/ωRNA-v13 produced indels at frequencies of 4.4–19.3%, 26.2–47.6%, and 25.8–38.6%, respectively, at VEGFA_site 1, VEGFA_site 2, CXCR4_site 1, and DYNC1H1_site 1 (Fig. 5a); indel rates were lower due to co-delivery of the dsODN. Importantly, we detected integration rates of 0.3–2.2% across the four target sites. Although lower than the values obtained for the blunt-end cutter SpCas9, these integration rates are sufficient for GUIDE-seq profiling. Similar integration rates have been reported from GUIDE-seq analyses of other Cas nucleases producing sticky ends, such as AsCas12a, LbCas12a, and AsCas12f1^{23, 43}.

Figure 5 |. Genome-wide specificity of wildtype and engineered OgeuIscB/ωRNA.

a, On-target indel and dsODN integration frequencies in GUIDE-seq samples. b, Numbers of off-target sites captured by GUIDE-seq. Hits with up to seven mismatches to the 22-nt bait are retained. The TAM sequence is specified as NWRRNA. c, Top-ranked off-target editing loci for VEGFA_site 1-, VEGFA_site 2-, CXCR4_site 1-, and DYNC1H1_site 1-targeting OgeuIscB/ωRNA reported by GUIDE-seq. Uncropped GUIDE-seq results for OgeuIscB/ωRNA, enOgeuIscB/ωRNA-v2, and enOgeuIscB/ωRNA-v13 targeting VEGFA_site 1, VEGFA_site 2, CXCR4_site 1, and DYNC1H1_site 1 are provided in Supplementary Fig. 5–8. Mismatch positions are highlighted in colors. Numbers of unique integration events detected by GUIDE-seq are provided to the right of the corresponding sequences. Black squares indicate the on-target loci. All GUIDE-seq experiments were carried out in HEK293T cells.

The TAM preference of OgeuIscB was previously defined as NWRRNA in biochemical assays². Given the short spacer (16 nt) and degenerate TAM (6 nt), we allowed up to 7 mismatches within the 22-nt bait in GUIDE-seq analysis (Fig. 5b, Fig. 5c, and Supplementary Fig. 5–8) and detected 148 potential off-target sites for VEGFA_site 1-targeting OgeuIscB/ωRNA. The on-target site ranked second among all hits, supporting the programmable nature of OgeuIscB/ωRNA. GUIDE-seq also captured another site in the human genome identical to VEGFA_site 1 with a slightly altered TAM (GGAGTG) (Supplementary Fig. 5).

VEGFA_site 2-targeting OgeuIscB/ωRNA generated dsODN integration at 43 genomic loci, with the on-target site ranking second. The top-ranked hit carries an A1G mismatch and a perfectly complied TAM (CAGACA; Fig. 5c and Supplementary Fig. 6). When targeting CXCR4_site 1 and DYNC1H1_site 1, we detected 38 and 8 potential off-target sites, respectively, for OgeuIscB/ωRNA (Fig. 5c and Supplementary Fig. 7–8). Similar to our observations for VEGFA_site 1- and VEGFA_site 2-targeting OgeuIscB/ωRNA, on-target sites consistently ranked among the top hits. The number of potential off-target sites increased from OgeuIscB/ωRNA to enOgeuIscB/ωRNA-v2 and enOgeuIscB/ωRNA-v13 (Fig. 5b). The concurrent trend in on-target and off-target activity points to similar specificity between the engineered and wildtype OgeuIscB/ωRNA systems.

To verify the off-target sites identified by GUIDE-seq, we measured indel levels at a total of 16 off-target sites (Extended Data Fig. 8). As anticipated, the indel frequencies increased from wildtype OgeuIscB/ωRNA to the engineered systems, indicating a potential tradeoff between potency and specificity. Nevertheless, the on-target sites remain the top edited sites in all cases. Collectively, our GUIDE-seq results corroborated well with targeted deep sequencing results.

Mismatch and TAM tolerance inferred from GUIDE-seq

Next, we inspected mismatch tolerance and TAM preferences of the OgeuIscB/ωRNA system in its off-target profiles. We considered two factors that may dictate whether an off-target site is captured by GUIDE-seq—the incomplete representation of all possible off-target sequence combinations in the human genome and the preferences of OgeuIscB/ωRNA among the available off-target sites. To this end, we extracted potential off-target sites from the human genome by Cas-OFFinder⁴⁴ (Fig. 6a). We consider hits of ≤2 mismatches with the 16-nt spacer as a surrogate for “all available off-target sites in the human genome” (Supplementary Table 5). Meanwhile, we reanalyzed GUIDE-seq data, allowing up to 2 mismatches to the spacer without specifying a TAM. The resulting sites serve as a surrogate for “the subset of off-target sites recognized by OgeuIscB/ωRNA”. Mismatch tolerance of OgeuIscB/ωRNA at individual protospacer positions is defined as the ratio of mismatches detected experimentally by GUIDE-seq to those identified in silico by Cas-OFFinder.

Figure 6 |. Mismatch tolerance and TAM preferences of OgeuIscB/ωRNA.

a, Sequence logos representing off-target sites bearing up to two mismatches with VEGFA_site 1, VEGFA_site 2, CXCR4_site 1, and DYNC1H1_site 1 identified in silico by Cas-OFFinder (top) and captured expeirmentally by GUIDE-seq (bottom). b, Mismatch tolerance at individual protospacer positions plotted as ratios of mismatches detected by GUIDE-seq and those identified by Cas-OFFinder. c, Sequence logos representing TAMs detected at off-target sites of up to two mismatches identified in silico by Cas-OFFinder (top) and expeirmentally by GUIDE-seq (bottom). d, Close-up view of TAM interactions in the OgeuIscB/ωRNA/DNA complex. PDB entry: 7UTN⁴. Similar plots for enOgeuIscB/ωRNA-v2 and enOgeuIscB/ωRNA-v13 are provided in Extended Data Fig. 9 and 10. Plots considering numbers of unique integration events in GUIDE-seq are available in Supplementary Fig. 10.

Mismatches were mostly rejected at positions 3–16 by wildtype OgeuIscB for all four spacers (Fig. 6a, Fig. 6b)⁴⁵. Weaker stringency for the first two nucleotides bordering the TAM-distal side can be rationalized by the cryo-EM structure as this region of the ωRNA/target strand (TS) heteroduplex is not contacted by OgeuIscB, whereas the rest of the ωRNA/TS heteroduplex is sandwiched between the body of OgeuIscB and the latched HNH nuclease domain—mismatches are indirectly sensed from distortions in the heteroduplex backbone⁴. Similar PAM-distal (1–3) mismatch tolerance has been reported for SpCas9^46–48, and is likely a shared feature among IscB and its derived RNA-guided endonucleases.

Increased mismatch tolerance was observed at position 16 (first nucleotide 5’ to the TAM) for VEGFA_site 2-, and to a lesser extent, VEGFA_site 1- and CXCR4_site 1-targeting OgeuIscB/ωRNA (Fig. 6b). This tolerance appears to be negatively correlated with the base-pairing strength, as all three of these mismatch sites contain an A or T. In contrast, DYNC1H1_site 1 containing a C at position 16 did not tolerate any mismatches therein. Given that the TS is rotated 180° above position 16, and stacking is suboptimal at this base pair; it is plausible that a weaker nucleotide (A/T) in the guide may tolerate mismatches through sheared base-pair formation. Detailed structural biology study is required to understand the cause of this mismatch tolerance.

To dissect the TAM preferences of OgeuIscB/ωRNA in human cell editing experiments, we expanded our analysis to include the 6-nt sequence downstream of the target site (Fig. 6c). While Cas-OFFinder hits showed minimal enrichment at these six positions, an NARRNN motif emerged in GUIDE-seq-identified off-target sites. Similar TAM preferences were also observed using the default GUIDE-seq analysis pipeline with a pre-specified NWRRNA TAM (Supplementary Fig. 9). The RR requirement at TAM-3 and 4 is stringent; few pyrimidines were found at off-target sites. This RR preference is specified by hydrogen bonds between N7s of A3 and G4 and the main-chain amide groups of G460 and R461, as shown in the cryo-EM structure (Fig. 6d); pyrimidines could not accommodate the same contacts⁴. TAM-2 showed a preference for A in GUIDE-seq. This position is specified from the minor groove side by PID linker residues His397 towards O2 of T2 in the NTS, and K380 towards N3 of A2 in the TS (Fig. 6d)^{4, 5}. G:C pairs in either orientation would be rejected due to steric clashes with the N2 of guanosine. Here our data further suggests that only one A:T orientation is preferred.

A weak preference for adenosine at TAM-6 was observed in the in vitro plasmid cleavage assay, which is attributed to a favorable hydrophobic contact with the methyl group of T6 in the NTS⁴. Here our GUIDE-seq data does not support a strong TAM-6 specification in human cells. Overall, our findings suggest that the TAM contacts in cellular settings are weighted slightly differently from those in vitro. This will have implications in target selection practice.

Lastly, we analyzed off-target sites impacted by enOgeuIscB/ωRNA-v2 and enOgeuIscB/ωRNA-v13 and obtained similar results (Extended Data Fig. 9 and 10), suggesting that the engineered derivatives retain the specificity and TAM preferences of wildtype OgeuIscB/ωRNA. The correlation between the number of unique dsODN integration events and indel propensity also prompted us to reassess mismatch tolerance and TAM preferences of the OgeuIscB/ωRNA systems, weighting off-target sites based on their GUIDE-seq read counts; the results remain consistent (Supplementary Fig. 10). We note that three of the four chosen target sites are A:T rich. Whether the observed genome-wide specificity translates fully to more GC-rich target sites awaits future investigation. Collectively, the OgeuIscB/ωRNA system specifies an “NARR” TAM when searching for targets in human cells, favors complementarity at positions 3–16, and tolerates mismatches at positions 1–2. Information here will guide the design of future editing experiments; sites with weak TAMs or many competing off-target sites can be computationally avoided.

Discussion

In this study, we explored ways to improve the potential of the OgeuIscB/ωRNA system for programmed genome editing. Using rational protein engineering, we increased the in vitro DNA binding affinity of OgeuIscB by 4-fold and the ex vivo indel formation activity by 30.4-fold. It is worth noting that although our coevolutionary approach⁴⁹ is efficient at avoiding deleterious mutations, it may miss potential mutations in regions of extremely high or low sequence homology. Utilizing exhaustive arginine-scanning mutagenesis, a recent study identified enIscB, which carries four indel-boosting mutations—E84R, H368R, S386R and S456R (Supplementary Fig. 11)³³. There the mutations were denoted E85R, H369R, S387R and S457R due to the inclusion of an extra M at the N terminus. In addition to the shared H368R, the two studies independently identified S386R (enIscB) and L393K (our enOgeuIscB), two mutations closely positioned at the protein/DNA interface (Supplementary Fig. 11). While the two independent solutions lead to a similar level of activity improvement, some target sites are edited more efficiently by one variant and vice versa. This suggests neither variant has reached the plateau in robustness. Considering that the two variants share only one mutual mutation, future effort to combine mutations may yield an even more robust IscB variant.

Since large RNAs have a tendency to either adopt multiple conformations or be kinetically trapped at misfolded states, optimization of ωRNA represents an additional challenge for the IscB/ωRNA system. Our best RNP combination, enOgeuIscB/ωRNA-v2, delivers up to 87.3% indels in HEK293T cells. Nevertheless, its editing efficiency still fluctuates among target sites, suggesting that the folding of the ωRNA core structure is not robust enough. Moreover, despite the substantial improvement in activity, enOgeuIscB/ωRNA remains a weak DNA binder in vitro in comparison to SpCas9. As such, we believe we have not reached the full potential of the OgeuIscB/ωRNA system.

Prokaryotes rely on CRISPR-Cas systems to survive the constant bombardment by phages and other mobile genetic elements. There is a high evolutionary demand for the CRISPR-Cas effectors to be both efficient and accurate. The ωRNA-guided nucleases in transposable elements are not held to the same standard—they only need to mediate occasional site-specific homing or transposition^{3, 50}. The moderate activity and fidelity of the IscB/ωRNA system is consistent with its natural function. Numerous studies have demonstrated that transposase activity can be readily improved through gain-of-function selection^51–54; our progress so far reinforces this view. We envision that the IscB/ωRNA system may be further improved for genome-editing applications through rational engineering or directed evolution.

With its small stature and controllable strand scission activity, IscB/ωRNA is a particularly attractive system for CRISPR2.0 applications, such as base editing^9–11 and prime editing⁵⁵. We and others have demonstrated the compatibility of TadA8e³⁸ and APOBEC3A⁵⁶ with IscB; however, the base-editing window remained too wide to deliver controllable clinical outcomes³³. We envision that this problem can be solved by coupling IscB with effectors of intrinsic target specification properties, such as context-specific deaminases for base editing and pegRNAs for prime editing. We are optimistic that our enhanced OgeuIscB/ωRNA systems serve as a steppingstone towards a robust, compact, and programmable genome-editing tool.

Online Methods

Plasmid construction

Oligonucleotides were ordered from Integrated DNA Technologies (IDT). Plasmids encoding wildtype OgeuIscB and the ωRNA scaffold were gifts from Feng Zhang (Addgene plasmids #176540 and #176541)². ABE8e (TadA8e^V106W) was a gift from David Liu (Addgene plasmid # 138495)³⁸. For cloning of plasmids encoding OgeuIscB variants, polymerase chain reactions (PCR) were carried out using Phusion U Hot Start DNA Polymerase (Fisher Scientific, F555L). Plasmids were constructed by USER assembly (New England Biolabs, M5505L) following the manufacture’s protocol. Plasmids encoding ωRNA were constructed by Golden Gate assembly. Specifically, DNA fragments were amplified using Q5 High-Fidelity DNA Polymerase (New England Biolabs, M0491L). The fragments were digested by Type IIS restriction enzyme BsmBI-v2 (New England Biolabs, R0739L) and assembled by T4 DNA ligase (M1804) per manufacture’s protocol. DH5-alpha F’I^q competent cells (New England Biolabs, C2992I) were used for cloning and plasmid propagation. The sequence of all plasmids was confirmed by Sanger sequencing (Azenta Life Science) or Nanopore sequencing (Plasmidsaurus).

For mammalian genome editing, OgeuIscB variants are expressed under a cytomegalovirus (CMV) promotor; ωRNAs are expressed with a U6 promoter. In adenine base editors containing a single TadA8e^V106W domain, a 32 amino acid linker (linker 1) was used to connect OgeuIscB and TadA8e^V106W. For dual TadA8e^V106W editors, linker 1 was used to connect the N terminus of OgeuIscB to one TadA8e^V106W monomer and a 34 amino acid linker (linker 2) was used to connect the C terminus of OgeuIscB to the second TadA8e^V106W monomer. All OgeuIscB variants and the derived adenine base editor fusions were flanked by two NLS peptides. The sequences of linkers and NLS peptides are provided in Supplementary Table 6. Key plasmids generated by this work will be deposited to Addgene.

Cell culture and transfection

A549, HeLa, and HEK293T cells were cultured in Dulbecco’s Modified Eagle’s Medium with high glucose, sodium pyruvate, and L-Glutamine (DMEM, Sigma-Aldrich, D5796) supplemented with 10% fetal bovine serum (Corning, 10–013-CV). Cells were maintained in 5% CO₂ at 37 °C and passaged regularly upon reaching 80% confluence. Cells were seeded 24 h prior to transfection to ensure ~50% confluence at the time of transfection. Transfection was performed with jetOPTIMUS DNA transfection reagent (Polyplus, 101000006) or Lipofectamine 3000 transfection reagent (Invitrogen, L3000015) following the manufacturers’ instructions.

Indel-generating activity of OgeuIscB variants

OgeuIscB variant screening and ωRNA engineering (related to Fig. 1–3, Extended Data Fig. 2 and 3d, Supplementary Fig. 2 and 3) were carried out in 96-well plates with jetOPTIMUS DNA transfection reagent. Specifically, 120 ng of the OgeuIscB plasmid and 80 ng of the ωRNA plasmid were delivered with 0.2 μL of jetOPTIMUS DNA transfection reagent in 12.5 μL of jetOPTIMUS buffer. Cells were incubated for 48 h post-transfection before harvest. The same protocol was used for genome editing across 26 target sites in HEK293T cells (related to Fig. 4a–d, Extended Data Fig. 3a–3c, 4 and 5, Supplementary Fig. 4) with a minor modification in workflow that cells were incubated for 72 h instead of 48 h post-transfection.

Lipofectamine 3000 transfection reagent was used to deliver enOgeuIscB and engineered ωRNA as a single plasmid (related to Fig. 4e and Extended Data Fig. 6). Specifically, 200 ng of plasmid DNA and 0.2 μL of P3000 reagent were mixed in 10 μL of Opti-MEM reduced serum medium (Thermo Fisher Scientific, 31985–062). Meanwhile, 0.3 μL of Lipofectamine 3000 was diluted in 10 μL of Opti-MEM and added to the above solution. The mixture was incubated at room temperature for 15 min before adding to cells seeded in 96-well plates. Cells were harvested 72 h post-transfection. The same transfection protocol was used for A549, HeLa, and HEK293T cells.

Adenine base editing by engineered OgeuIscB/ωRNA

Adenine base editing experiments were carried out in 96-well plates. Two separate plasmids that encode ωRNAs and base editors (fusion proteins of OgeuIscB nickase and TadA8e^V106W) were delivered into HEK293T cells using Lipofectamine 3000 transfection reagent (related to Fig. 4f–g and Extended Data Fig. 7). Briefly, 200 ng of base editor plasmid and 100 ng of ωRNA plasmid were mixed with 0.2 μL of P3000 reagent in 10 μL of Opti-MEM reduced serum medium. To the solution was added 0.3 μL of Lipofectamine 3000 reagent pre-diluted in 10 μL of Opti-MEM. The mixture was incubated at room temperature for 15 min before adding to cells seeded in 96-well plates. Cells were harvested 72 h post-transfection.

Amplicon deep sequencing

Cells were washed with PBS and lysed in 50 μL of lysis buffer (10 mM Tris-HCl, 0.05% SDS, 25 μg/mL proteinase K; pH 8.0). The lysate was incubated at 37 °C for 15 min, 55 °C for 60 min, 85 °C for 15 min, and 95 °C for 10 min. 1 μL of the cell lysate was subjected to two rounds of PCR to add Illumina adaptors and barcodes. The sequences of primers used for library preparation are listed in Supplementary Table 7. The pooled library underwent gel purification and 150 bp paired-end sequencing on an Illumina MiSeq instrument. Indel frequencies were analyzed by CRISPResso2 using the batch mode²⁶. For adenine base editing, MiSeq reads were aligned to reference sequences with BWA-MEM⁵⁷ and sorted using Samtools 1.13⁵⁸. A:T-to-G:C editing rates at given positions were calculated as number-of-G-reads/number-of-total-reads.

Quantification of ωRNA expression

Cells were transfected with jetOPTIMUS DNA transfection reagent as described above and incubated for 48 h. Total RNA was extracted by Direct-zol RNA Microprep Kits (Zymo Research, R2053-A) with on-column DNA digestion per the manufacture’s instruction. RNA was reverse transcribed using GoScript reverse transcriptase (Promega, A501D) with target-specific primers (Supplementary Table 7). The relative levels of ωRNAs were determined by quantitative PCR normalized to GAPDH.

In vitro DNA binding and cleavage

OgeuIscB/ωRNA complexes were expressed and purified following a published protocol⁴. Briefly, T7 Express Competent E. coli cells (New England Biolabs) were transformed with OgeuIscB- and ωRNA-encoding plasmids. The transformed E. coli, supplemented with 0.75 mg/ml L-cysteine in the culture, were grown until the OD 600 nm reached 0.8. Protein expression was induced by adding isopropyl-β-D-thiogalactopyranoside (IPTG) to a final concentration of 0.5 mM. Bacteria were grown at 16 °C for an additional 16 h prior to harvest. The OgeuIscB/ωRNA complexes were purified following a protocol reported previously⁴. Protein concentrations were quantified using BioTek Cytation 5 (Agilent).

Target DNA was prepared by PCR. Forward primer, /Cy5/ CTGTCTCTTATACACATCT; reverse primer, /FAM/ CAAGCAGAAGACGGCATAC; template, CTGTCTCTTATACACATCTGAGGGCCCCACGAATTCTTCTTCAGACTTCTAGTCTCGTTCACTCTTTTTGATGATGCTTCGTAACTCGTGGGAAGTCGAGGTCAGCCCGTCGTAGTATGCCGTCTTCTGCTTG. The fluorophore labeled PCR products were purified with a GeneJet gel extraction kit (Thermo Scientific).

For time-course DNA cleavage, 5 nM of target DNA was incubated with 500 nM of OgeuIscB/ωRNA in a cleavage buffer (100 mM NaCl, 50 mM HEPES, 2 mM DTT, 5 mM MgCl₂, 10 μM ZnCl₂; pH 7.5) at 37°C. The reaction was stopped with 20 mM of EDTA at designated time points followed by additional incubation at 95 °C for 5 min. The sample was then treated with 0.8 U proteinase K at 37 °C for 15 min. To each sample, an equal volume of 100% formamide was added, and the samples were heated to 95 °C for 10 min before being analyzed by 12% urea-polyacrylamide gel electrophoresis (PAGE). The fluorescent signal was captured using ChemiDoc (Bio-Rad). For protein dose-dependent cleavage, the same protocol was followed with a fixed 30 min incubation period. For EMSA, 5 nM of target DNA was incubated with different concentrations of OgeuIscB/ωRNA at 37 °C for 30 min. The mixture was then directly analyzed by a 1.5% Tris-borate (TB) gel. The fluorescent signals were captured using a Typhoon scanner.

GUIDE-seq

OgeuIscB and ωRNA were delivered as a single plasmid into cells pre-seeded in a 24-well plate. To each well, 800 ng of plasmid DNA and 4 pmol of dsODN were co-delivered with 0.8 μL of jetOPTIMUS DNA transfection reagent in 50 μL of jetOPTIMUS buffer. Cells were incubated for 72 h followed by genomic DNA extraction with the Quick-DNA Miniprep kit (Zymo Research, D3025). Amplicon deep sequencing was performed as described above to measure indel and dsODN integration frequencies.

GUIDE-seq libraries were prepared with 500 ng of genomic DNA following a reported protocol¹² and sequenced on an Illumina Nextseq 2000 platform using paired-end sequencing (R1 40 cycle, I1 8 cycle, I2 16 cycle, R2 70 cycle). Sequencing data were mapped to human genome reference (GrCh37) and processed using the open-source guideseq software (https://github.com/tsailabSJ/guideseq)⁵⁹. For genome-wide specificity analysis, the PAM was set to NWRRNA with a maximum mismatch number of 7. For mismatch tolerance analysis, the PAM was set to NNNNNN with a maximum mismatch number of 2 in the protospacer.

In silico off-target prediction

To evaluate mismatch tolerance of the OgeuIscB/ωRNA systems, Cas-OFFinder⁴⁴ was employed to predict potential off-target sites. Genomic loci carrying up to 2 mismatches to the on-target sequence were extracted. We did not specify a PAM for in silico off-target prediction. Sequence logos were plotted using WebLogo 3⁶⁰.

Statistical analysis

Unpaired Student’s t-tests (two tailed) were used to compare samples (GraphPad Prism 9.0). n.s. P > 0.05; * P ≤ 0.05; ** P ≤ 0.01; *** P ≤ 0.001; **** P ≤ 0.0001. Investigators were not blinded to experimental conditions or outcome assessments.

Extended Data

Extended Data Fig. 1 |. In vitro DNA cleavage activity of wildtype OgeuIscB and enOgeuIscB.

a. Concentration-dependent in vitro DNA cleavage by wildtype OgeuIscB and enOgeuIscB. Target DNA was supplied at a concentration of 5 nM. b. Time-course in vitro DNA cleavage by wildtype OgeuIscB and enOgeuIscB. Target DNA: 5 nM; (en)OgeuIscB/ωRNA: 500 nM. All experiments were repeated independently three times with similar results.

Extended Data Fig. 2 |. Structure-guided ωRNA engineering.

a. Cryo-EM structure of the OgeuIscB RNP bound to target DNA illustrating the locations of the truncations in ωRNA (PDB entry: 7UTN⁴). Non-target strand DNA: red; target strand DNA: blue; guide RNA: orange; ωRNA scaffold: grey. Rectangle boxes indicate regions subject to truncation. b. Schematic of ωRNA variants. Changes in ωRNA-v2 and ωRNA-v13 are indicated in red. c. Fold changes of indel levels at VEGFA_site 1 generated by ωRNA variants paired with wildtype and engineered OgeuIscB proteins. Sequences of ωRNA variants are provided in Supplementary Table 3. Raw indel frequencies are plotted in Figure 3b. d. Indel levels at VEGFA_site 1 generated by wildtype ωRNA paired with wildtype OgeuIscB and OgeuIscB-ΔPLMP. e. Relative levels of ωRNAs when co-expressed with wildtype OgeuIscB and enOgeuIscB. RNA levels were normalized to that of GAPDH. All experiments were carried out in HEK293T cells, with indel frequencies and RNA abundances quantified 48 h post-transfection (mean ± s.d.; n = 3 independent experiments).

Extended Data Fig. 3 |. Genome editing facilitated by combining engineered OgeuIscB and ωRNA variants.

a, Indel frequencies generated by HA-tagged wildtype OgeuIscB, OgeuIscB-v5.1, and enOgeuIscB paired with the wildtype ωRNA, ωRNA-v2, and ωRNA-v13 at eight target sites. b. Indels generated by enOgeuIscB paired with wildtype and ωRNA-v2 at 26 genomic sites. c, Violin plot showing the distribution of indel levels achieved by enOgeuIscB/ωRNA and enOgeuIsc/ωRNA-v2 across 26 target sites. The width of the violin represents the density of data points at each value. The thick centerline and the plus sign within the violin indicate the median and the mean, respectively, while the dashed lines represent the first and third quartiles. The violin plot is truncated at the minimum and maximum values. d, Relative levels of EMX1_site 1-targeting ωRNA when co-expressed with wildtype OgeuIscB and enOgeuIscB. RNA levels were normalized to that of GAPDH. All experiments were carried out in HEK293T cells (mean ± s.d.; n = 3 independent experiments), with indel frequencies quantified 72 h post-transfection (a-c) and RNA levels quantified 48 h post-transfection (d). Statistical analysis was done using a two-tailed Student′s t-test. n.s. P > 0.05; *** P = 0.0002.

Extended Data Fig. 4 |. The HA tag negatively impacts the editing activity of OgeuIscB/ωRNA.

a. Design of plasmid constructs used for activity evaluation. b. Indel frequencies generated by HA tag-free wildtype OgeuIscB, OgeuIscB-v5.1, and enOgeuIscB paired with the wildtype ωRNA, ωRNA-v2, and ωRNA-v13 at eight target sites. c. Box plot of indel frequencies achieved by HA tag-bearing and HA tag-free wildtype OgeuIscB, OgeuIscB-v5.1, and enOgeuIscB across eight target sites. The whiskers extend to minima and maxima. Lower and upper hinges denote first and third quartiles. The central line and plus sign represent the median and mean, respectively. n = 8 target sites. All experiments were carried out in HEK293T cells (mean ± s.d.; n = 3 independent experiments). Cells were harvested 72 h post-transfection.

Extended Data Fig. 5 |. Side-by-side comparison of the indel formation activity of enOgeuIscB and enIscB.

Indel frequencies mediated by HA tag-bearing (a) and HA tag-free (b) enOgeuIscB and enIscB paired with the wildtype ωRNA. All experiments were carried out in HEK293T cells (mean ± s.d.; n = 3 independent experiments). Cells were harvested 72 h post-transfection.

Extended Data Fig. 6 |. Indel frequencies generated by enOgeuIscB/ωRNA-v2 and enOgeuIscB/ωRNA-v13 at six target sites in A549 (a) and HeLa (b) cells.

enOgeuIscB and the engineered ωRNA were encoded in a single plasmid to facilitate delivery. Three independent replicates were carried out (mean ± s.d.). Cells were harvested 72 h post-transfection.

Extended Data Fig. 7 |. Adenine base editing by engineered OgeuIscB/ωRNA.

a. A:T-to-G:C editing at VEGFA_site 1 by three versions of TadA8e^V106W and enOgeuIscB (D60A) fusions, programmed with the wildtype ωRNA, ωRNA-v2, or ωRNA-v13. b. A:T-to-G:C editing achieved by wildtype OgeuIscB/ωRNA, enOgeuIscB/ωRNA-v2, enOgeuIscB/ωRNA-v13, and enIscB/ωRNA*, in the dual TadA8e^V106W fusion scheme, across six target sites. The editing efficiency at each adenosine residue across the entire protospacer region was plotted. Three independent replicates were carried out in HEK293T cells (mean ± s.d.). Cells were harvested 72 h post-transfection.

Extended Data Fig. 8 |. Off-target (OT) editing at GUIDE-seq-nominated OT sites.

Indel frequencies generated by OgeuIscB/ωRNA, enOgeuIscB/ωRNA-v2, and enOgeuIscB/ωRNA-v13 in GUIDE-seq samples (a) and cells independently transfected in the absence of GUIDE-seq dsODN (b). All experiments shown in the figure were carried out in HEK293T cells. Cells were harvested 72 h post-transfection (a, n = 1; b, mean ± s.d., n = 3 independent experiments).

Extended Data Fig. 9 |. Mismatch tolerance of enOgeuIscB/ωRNA-v2 and enOgeuIscB/ωRNA-v13.

a. Sequence logos representing off-target sites captured expeirmentally by GUIDE-seq for enOgeuIscB/ωRNA-v2 and enOgeuIscB/ωRNA-v13 targeting VEGFA_site 1, VEGFA_site 2, CXCR4_site 1, and DYNC1H1_site 1. Off-target sites bearing up to two mismatches with the 16-nt spacer are included in the analysis. b, Mismatch tolerance at individual positions plotted as ratios of mismatches detected by GUIDE-seq and those identified by Cas-OFFinder. Similar analyses for wildtype OgeuIscB/ωRNA are provided in Figures 6a and 6b.

Extended Data Fig. 10 |. Sequence logos representing putative TAMs at off-target sites captured by GUIDE-seq.

Off-target sites bearing up to two mismatches with the 16-nt spacer are included in the analysis. Similar analyses for wildtype OgeuIscB/ωRNA are provided in Figure 6c.

Data availability

The cryo-EM structure of the OgeuIscB/ωRNA/DNA complex were available at PDB (entry: 7UTN, https://www.rcsb.org/). The human reference genome GRCh37 is available at EBI (https://ftp.ebi.ac.uk/). Deep sequencing data generated by this study have been deposited to the NCBI Sequence Read Archive under accession number PRJNA983934.