<scp>AI</scp> ‐designed <scp>OpenCRISPR</scp> ‐1 performs robust knockout, base editing, and prime editing in rice

Plant OpenCRISPR-1 (POC1) mediates knockout and adenine base editing in rice. RNA-guided programmable nucleases enable high-precision genome engineering for applications in agriculture, medicine, and biotechnology. Genome editing in biological organisms relies on a limited set of naturally occurring enzymes. Widely used nucleases such as Cas9 (Jinek et al., 2012) and Cas12a (Zetsche et al., 2015), as well as emerging effectors including TnpB (Karvelis et al., 2021; Karmakar et al., 2024a) and IscB (Altae-Tran et al., 2021), are derived from bacterial systems and often exhibit constraints in non-native eukaryotic contexts (Pacesa et al., 2024). Artificial intelligence (AI)-driven protein designing is now transforming nuclease engineering by leveraging large language models trained on the vast natural diversity of protein sequences (Nijkamp et al., 2023). Recently, OpenCRISPR-1 (OC1), an AI-generated RNA-guided nuclease comprising 1380 amino acids and differing from the prototypical SpCas9 by 403 amino acid substitutions, was shown to support efficient genome editing in human cells (Ruffolo et al., 2025). OC1 was released as an open-source nuclease and has been proposed to be freely available for licensing for academic research and commercial development (Ruffolo et al., 2025); consequently, it has the potential to substantially enhance global access to genome-editing technologies and represents an attractive alternative to the widely used SpCas9. However, its compatibility with diverse tracrRNA architectures, amenability to the development of advanced editing platforms such as cytosine and adenine base editors and prime editors, and functionality in plant systems have not yet been established. Here, we develop and validate a suite of OC1-based genome-editing tools for plants, taking rice as a model, that support highly efficient gene knockout, base editing, and prime editing, demonstrating the versatility of AI-designed nucleases in plant systems. For simplicity, we hereafter refer to this platform as plant-OpenCRISPR-1 (POC1). This study highlights the translational potential of AI-designed nucleases for plant genome engineering. We implemented a standardized workflow in which POC1-based plasmid constructs and their SpCas9-based counterparts were transfected into rice protoplasts (Panda et al., 2024), followed by genomic DNA isolation 72 h post-transfection and deep amplicon sequencing to quantify editing outcomes (Fig. 1a1). Protoplast transfection efficiencies consistently exceeded 80%, as assessed by reporter expression (Fig. 1a2). To establish a platform for targeted double-strand break (DSB) induction in plant genomes, we codon-optimized the original OC1 sequence (Ruffolo et al., 2025) for rice using the GenSmart™ Codon Optimization algorithm (GenScript, 2020; Supporting Information Figs S1–S3), and cloned it into the pRGEB32 backbone (Xie et al., 2015), replacing the SpCas9 coding region to generate the POC1 nuclease expression vector (Figs 1b1,b2, S1–S3). For an initial assessment, two guide RNAs targeting the OsCYP75B4 gene, spaced 573 bp apart, were cloned into a polycistronic tRNA–gRNA (PTG) cassette (Xie et al., 2015) and expressed under the OsU3 promoter in the POC1 vector (Figs S4–S6). Protoplast transfection followed by PCR and gel analysis revealed a c. 570-bp deletion (Fig. 1b3), indicating that POC1 induced DSBs at both target sites, resulting in excision of the intervening region. The smaller PCR amplicon was cloned, and the targeted deletion was confirmed by Sanger sequencing (Notes S1.1). We next evaluated POC1-mediated editing at the OsSWEET11, OsSWEET14 and OsZ3 loci in rice protoplasts. Deep amplicon sequencing revealed robust indel formation at all three targets, with editing efficiencies ranging from 10.0 to 16.8% (Fig. 1b4–b6; Notes S1.2–S1.7). POC1 exhibited editing efficiencies comparable to those of SpCas9 across all tested loci, with no statistically significant differences observed (SWEET11, P = 0.538; SWEET14, P = 0.261; Z3, P = 0.505). In most SpCas9-mediated genome-editing applications, target recognition is guided by a chimeric single-guide RNA (sgRNA) generated by fusing the crRNA and tracrRNA via an artificial linker (Jinek et al., 2012). In addition to this canonical short tracrRNA (tracrRNA-S), Streptococcus pyogenes encodes a longer tracrRNA variant (tracrRNA-L) that folds into a natural single-guide RNA and directs Cas9 to repress its own promoter, thereby preventing autoimmunity (Workman et al., 2021). We recently showed that tracrRNA-L and a truncated derivative (ΔtracrRNA-L) can efficiently support genome editing in eukaryotic systems (Karmakar et al., 2024b). Here, we generated POC1 expression vectors incorporating tracrRNA-L and ΔtracrRNA-L modules for guide RNA expression to assess DSB induction and indel formation (Figs 1c1,c2, S7–S9). We compared POC1 and canonical SpCas9 systems using tracrRNA-S, tracrRNA-L, and ΔtracrRNA-L at the OsPDS, OsSPL14, and OsSWEET14 loci in rice protoplasts. SpCas9 supported efficient indel formation at all three loci with tracrRNA-S and ΔtracrRNA-L, whereas tracrRNA-L alone consistently showed minimal activity (Fig. 1c3–c5; Notes S1.8–S1.28). When using the canonical tracrRNA-S, POC1-mediated editing efficiencies were significantly lower at the OsPDS and OsSPL14 loci, but significantly higher at the OsSWEET14 locus, compared with SpCas9 (Fig. 1c3–c5; Notes S1.8–S1.28). POC1 exhibited indel generation efficiencies comparable to those of SpCas9 across all three tested loci when tracrRNA-L was used, with no statistically significant differences observed. By contrast, using ΔtracrRNA-L resulted in significantly reduced POC1 activity relative to SpCas9. Next, we evaluated the compatibility of POC1 for base editor development. Base editing enables precise installation of targeted single-nucleotide substitutions without introducing genomic DSB (Molla et al., 2020, 2021). Adenine base editors (ABE) mediate A·T-to-G·C conversions (Gaudelli et al., 2017), whereas cytosine base editors (CBE) install C·G-to-T·A substitutions (Komor et al., 2016). Modern ABEs are typically generated by fusing a Cas9 (D10A) nickase to the laboratory-evolved adenosine deaminase TadA8e (Richter et al., 2020). Following this strategy, we engineered D10A nickase variants of both Cas9 and POC1 and fused TadA8e to their N termini via a flexible linker to generate Cas9-ABE8e and POC1-ABE8e, respectively (Figs 1d1,d2, S10, S11). POC1-ABE8e efficiently mediated A·T-to-G·C substitutions at both tested loci, OsZ3 and OsWSL5. Notably, POC1-ABE8e exhibited significantly higher editing efficiency at the OsWSL5 locus (15%) compared with Cas9-ABE8e (6.6%). By contrast, editing at the OsZ3 locus was significantly reduced relative to Cas9-ABE8e (17.8% vs 23%) (Fig. 1d3; Notes S1.29–S1.32). Cas9-ABE typically exhibits an optimal editing window spanning positions 4–8 within the protospacer (Molla & Yang, 2019; Molla et al., 2020). POC1-ABE8e appears to exhibit a broader editing window, as indicated by c. 30-fold increase in editing efficiency at an adenine located at protospacer position 2 of the Wsl5 site relative to Cas9-ABE8e (Fig. 1d4,d5; Notes S1.31, S1.32). Since this observation of an expanded editing window is based on a single target, it warrants validation across additional loci. To generate POC1-CBE and Cas9-CBE, we fused a human APOBEC3A variant (A3A-Y130F) and a uracil glycosylase inhibitor to the N and C termini of POC1(D10A) and Cas9(D10A), respectively (Figs 2a1,a2, S12, S13). The cytidine deaminase A3A-Y130F has previously been shown to confer high C-to-T editing efficiency in plant CBE architectures (Li et al., 2021). Across all three target sites tested, OsPDS, OsGS3, and OsCGRS55, POC1-CBE induced significantly higher C·G-to-T·A substitutions than Cas9-CBE (Fig. 2a3; Notes S1.33–S1.38). Across these targets, POC1-CBE exhibited an activity window spanning protospacer positions 2–15 (Fig. 2a4; Notes S1.33–S1.38). Although negligible, POC1-ABE generated similar or reduced Indels byproduct, while POC1-CBE yielded a similar Indels profile except for OsGS3 (Notes S1.39). Collectively, these results demonstrate that POC1 is compatible with both adenine and cytosine base editor platforms and supports highly efficient base editing. Prime editing enables precise installation of all types of base substitutions as well as predefined insertions and deletions at target loci without the need for a donor template or the induction of DSB (Anzalone et al., 2019). To construct prime editors (PEs), a Cas9(H840A) nickase is typically fused to a Moloney murine leukemia virus reverse transcriptase (M-MLV-RT) (Anzalone et al., 2019). Following this strategy, we generated a corresponding POC1(H850A) nickase and fused M-MLV-RT to its C terminus to generate POC1-PE (Figs 2b1,b2, S14, S15). Following the PE3max strategy, we designed engineered pegRNAs (epegRNAs) and nicking guide RNAs to install three distinct edits: a C·G-to-G·C substitution in OsPEPC, a C·G-to-A·T substitution in OsALS, and a G·C-to-C·G substitution at the OsACC locus. Compared with Cas9-PE3max, POC1-PE exhibited similar editing efficiency at the OsPEPC target, with no statistically significant difference observed (Fig. 2b3; Notes S1.40, S1.41). Contrastingly, at the OsALS and OsACC loci, POC1-PE achieved precise editing efficiencies of 4.0% and 6.18%, respectively (Fig. 2b3; Notes S1.42–S1.45), which were significantly lower than those obtained with Cas9-PE3max. Interestingly, POC1-PE exhibited comparable ratios of desired to undesired edits at the OsPEPC and OsALS loci, whereas a significantly improved ratio was observed at the OsACC locus (Fig. 2b4). Finally, we evaluated the performance of POC1 for editing an endogenous target gene in regenerated rice plants. The OsCYP57B4 locus was targeted using the POC1 vector harboring the PTG cassette constructed in the preliminary experiment (Figs S5, S6). Rice plants were regenerated following Agrobacterium-mediated transformation of calli (Molla and Yang, 2020). Among the hygromycin-resistant plants obtained (47 nos. for pRGEB32-CYP75B4 and 40 nos. for POC1-CYP75B4) (Fig. 2c1), Sanger sequencing of the target locus in 25 independent lines per construct revealed mutation frequencies of 36% (9 of 25 plants) for POC1 and 48% (12 of 25 plants) for pRGEB32 (Fig. 2c2,c3). There was no statistically significant difference (P = 0.39) in mutation frequency between pRGEB32-CYP75B4 and pKb-POC1-CYP75B4 under the tested conditions. We recovered biallelically edited plants harboring a 573-bp deletion that encompassed the intervening sequence between the two induced DSBs (Fig. 2c2; Notes S1.46). These results demonstrate that POC1 supports efficient genome editing in regenerated rice plants. In summary, we developed POC1-mediated versatile genome-editing platforms and demonstrated efficient NHEJ-driven indel formation, adenine and cytosine base editing, and prime editing in rice. Our results establish that the AI-designed nuclease OpenCRISPR-1 (Ruffolo et al., 2025), following codon optimization, is compatible with multiple genome-editing modalities and achieves performance comparable to that of the naturally occurring SpCas9 across diverse target loci in rice. Although the OC1 exhibited significantly reduced off-target editing compared to SpCas9 in human cells (Ruffolo et al., 2025), its off-target activity in plant systems remains to be evaluated. Similarly, the performance of OC1 should be assessed across a broader range of monocot and dicot species to support its wider adoption as an alternative to SpCas9. The expanding repertoire of AI-designed effectors (Thomson et al., 2025), including nucleases and deaminases, is poised to unlock new opportunities in eukaryotic genome engineering by overcoming intrinsic limitations of naturally evolved systems. This work was supported by the Indian Council of Agricultural Research (ICAR), Department of Agricultural Research and Education, Government of India, through the Plan Scheme Incentivizing Research in Agriculture. Funding from the Ignite Life Science Foundation, Bangalore (Projects/AgSci/22-23/01), to KAM is gratefully acknowledged. Additional support was provided by the Department of Biotechnology (DBT), Government of India, through the NSF–DBT TRTech–PGR program (Project No. IC-12048(12)/4/2024-MED-DBT) and the DIGNITY project (BT/PR53972/PBN/18/25/2024) to KAM. PD acknowledges the fellowship from the Department of Biotechnology (DBT), Government of India-JRF program. RS and CG appreciate the fellowship received from Ignite Life Science Foundation, Bangalore, India. The pRGEB32 and pG3H-PE3max vectors used in this study were gifts from Yinong Yang and Qijun Chen, respectively. The authors declare that GPT 5.2 was used to assist with text proofreading. The authors have subsequently reviewed, edited and approved all the generated contents and are taking full responsibility for the contents of the publication. None declared. KAM and MJB conceived and designed the study. KAM supervised the entire project. KAM, PD and RS planned the experiments. PD and RS performed experiments and collected all data. DP performed the protoplast transfection and prepared the POC1-based alternative tracr vectors. CG prepared the pKb-POC1 mother vector and assisted in cloning. PD, RS, DP, SPA and SP performed data analysis. SPA and SP constructed the control vectors for ABE and PE experiments. PD, RS, DP and SP analyzed the mutant plant lines. KAM, PD and RS prepared the manuscript. RS and PD prepared figures. KAM and MJB revised and finalized the manuscript. PD and RS contributed equally to this work. All data supporting the findings of this study are given in the main text and its Supporting Information (Figs S1–S15; Methods S1–S8; Notes S1, S2; Tables S1–S6). Amplicon sequencing data were submitted to the Gene Bank (NCBI Bioproject ID-PRJNA1401667). The vectors developed in this study will be made available through Addgene under the following plasmid IDs: #255129 (pKb-POC1), #255130 (pKb-POC1-ABE8e), #255131 (pKb-POC1-CBE), and #255132 (pKb-POC1-PE). All sequences are available in the supporting information file as Notes S2. Fig. S1 A diagrammatic flowchart showing the process of constructing pKb-POC1 vector. Fig. S2 Plasmid map showing different components of the pKb-POC1 mother vector, used for knockout experiments in rice. Fig. S3 Nucleotide sequence alignment between original OpenCRISPR-1 (Ruffolo et al., 2025) and rice codon-optimized POC1 nuclease (this study). Fig. S4 A diagrammatic flowchart showing the process of constructing pKb-POC1-CYP75B vector. Fig. S5 Schematic diagram showing the architecture of OsCYP75B4 gene comprising two exons, intervened by one intron. Fig. S6 Plasmid map showing different components of the pKb-POC1-CYP75B4, used for multiplexed knockout of OsCYP75B4 gene in rice. Fig. S7 A diagrammatic flowchart showing the process of constructing pKb-POC1-alternative tracr vector. Fig. S8 Plasmid map showing different components of the pKb-POC1-tracrL mother vector, used for alternative tracrRNA experiments in rice. Fig. S9 Plasmid map showing different components of the pKb-POC1-ΔtracrL mother vector, used for alternative tracrRNA experiments in rice. Fig. S10 A diagrammatic flowchart showing the process of constructing pKb-POC1-ABE8e vector. Fig. S11 Plasmid map showing different components of the pKb-POC1-ABE mother vector, used for Adenine Base Editing experiments in rice. Fig. S12 A diagrammatic flowchart showing the process of constructing pKb-POC1-CBE vector. Fig. S13 Plasmid map showing different components of the pKb-POC1-CBE mother vector, used for cytosine base editing experiments in rice. Fig. S14 A diagrammatic flowchart showing the process of constructing pKb-POC1-PE vector. Fig. S15 Plasmid map showing different components of the pKb-POC1 PE mother vector, used for prime editing experiments in rice. Methods S1 Vector construction for POC1-mediated knockout in rice. Methods S2 Vector construction for POC1-based adenine base editor: pKb-POC1-ABE8e. Methods S3 Vector construction for POC1 based cytosine base editor: pKb-POC1-CBE. Methods S4 Vector construction for POC1 based primed editor: pKb-POC1-PE. Methods S5 Protoplast isolation, transfection of constructs. Methods S6 Genomic DNA isolation, amplification of the targets and deep sequencing. Methods S7 Stable rice transformation. Methods S8 Statistical analysis. Notes S1 Sanger sequencing and Illumina deep sequencing snapshot obtained from transfected protoplast population and CRISP-Id analysis of regenerated T0 plant for cyp75b4 knockout using POC1. Table S1 Different vectors, polynucleotides, polypeptide sequences and their sequence IDs. Table S2 Target genes with locus ID and guide sequence. Table S3 List of primers used for cloning different guides. Table S4 List of primers used for screening of mutant plants. Table S5 List of primers used for deep amplicon sequencing (first round). Table S6 List of primers used for deep amplicon sequencing (second round). Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article. The New Phytologist Foundation remains neutral with regard to jurisdictional claims in maps and in any institutional affiliations.