The full-length SaCas9 was cloned from the SaCas9-gRNA vector reported in Qin et al. (2019). Fragments P1, P2, P3, and P4, fused with intein^N or intein^C, were synthesized by Shenggong Bioengineering (Shanghai). The full-length SaCas9 or fused fragments were cloned into the pCXUN backbone for stable rice transformation or into RNAγ1 or RNAγ2 vectors for VIGE in sheepgrass. The RNAγ1 and RNAγ2 vectors were constructed following the method described by Cheuk and Houde (2018). All subcloning procedures were carried out using the pEASY-Uni Seamless Cloning and Assembly Kit (TransGen Biotech). The plasmid maps, including full-length SaCas9, split-v1, split-v2, and split-v3, are shown in Supplementary Figure S1. The gRNAs targeting OsNYC1, OsNYC4, and OsPDS were synthesized using overlapping PCR. Taking OsNYC1 as an example, PCR amplification was performed using the SaCas9-gRNA template with the primer pairs U3pmeF/OsNYC1-T1F and OsNYC1-T1F/U3pmeR, producing two PCR products. These products were mixed in a 1:1 ratio and used as the template for a subsequent amplification with U3pmeF/U3pmeR. The resulting PCR product was then cloned into the Pme I restriction site of vectors containing either full-length or split SaCas9. For the gRNAs targeting the LcHRC, LcGW2, and LcTB1 genes, taking LcHRC as an example, the gRNA scaffold was obtained by amplifying the SaCas9-gRNA template using the primers γ1-LcHRC-T1F/γ1-gRNAR. This scaffold was then seamlessly cloned into the Apa I restriction site of the RNAγ1 backbone using the Assembly Kit. All primers used in this study are provided in Supplementary Table S1.

The stable transformation of rice was conducted following previously published protocols (Hiei et al., 1994; Saha et al., 2006). In brief, the vectors described above were transformed into the Agrobacterium tumefaciens strain EHA105 by electroporation. Dehulled seeds of the japonica rice (Oryza sativa L.) variety Zhonghua 11 were surface sterilized and cultured on the N6D solid medium to induce callus formation. Rice calli were pre-cultured and inoculated with the A. tumefaciens strain EHA105 carrying the recombinant vector. After 3 days of co-cultivation, the calli were washed thoroughly with sterilized water to remove residual bacteria and transferred to the N6-S medium for selection under antibiotic pressure for 2 weeks. Resistant calli were then transferred to the RE-III medium for a 2-week cultivation period and subsequently sub-cultured onto the fresh RE-III medium every 2 weeks to promote regeneration. Regenerated plants were successfully obtained, with a minimum of 18 independent transgenic lines generated for each vector.

The Western blot and immunoprecipitation experimental methods were primarily conducted following previously published protocols (Saha et al., 2006). In brief, the protein extract from wild-type and transgenic rice plants was separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to a polyvinylidene difluoride (PVDF) membrane (Merck Millipore, IPVH00010). For Western blot and immunoprecipitation, SaCas9 was assayed using the mouse anti-SaCas9 monoclonal antibody [6H4] (EpiGentek, A-9001). The loading control was probed using the anti-actin antibody (Abmart, M20009), with a goat anti-mouse antibody (Abcam, ab6789) utilized as the secondary antibody.

The constructed plasmids were transformed into A. tumefaciens EHA105 strains for agroinfiltration of wild-type or overexpressed RNAα and RNAβ Nicotiana benthamiana leaves. Equal volumes of Agrobacterium strains harboring individual plasmids were mixed to a final OD₆₀₀ of 0.3 and infiltrated into leaves of 3- to 4-week-old N. benthamiana plants, as described previously (Yuan et al., 2011). Virus-infected leaves were harvested after 7 days postinfection (dpi) and ground in 10 mM phosphate buffer each containing 0.5% of celite 545 (Roth) on ice. The virus was used to inoculate the fully emerged third leaves of 2- to 3-week-old sheepgrass using the finger-rub method. Each treatment included a minimum of four replicates.

The mutation types of the transgenic rice plant were assayed by Sanger sequencing of the PCR product and analyzed by CRISPR-GE DSDecodeM software (http://skl.scau.edu.cn/.). For next-generation sequencing (NGS), mutations were analyzed by the online tool high-throughput tracking of mutations 2.0 (Hi-TOM 2.0) (Sun T. et al., 2024). In brief, the target region was amplified from genomic DNA using site-specific primers containing barcode sequences. The resulting amplicons were submitted for next-generation sequencing at the State Key Laboratory of Rice Biology and Breeding, utilizing the Hi-TOM platform (China National Rice Research Institute, Chinese Academy of Agricultural Sciences, Hangzhou).

Statistical analysis was performed using GraphPad Prism version 8 for Windows (https://www.graphpad.com/), while the figures were further processed with Adobe Illustrator 2020 and Adobe Photoshop CC 2019 (https://www.adobe.com/).

To test whether intein-mediated protein splicing is an optional approach to reduce the size of SaCas9 transcription unit in plants and its potential use in VIGE, we analyzed the sequence of SaCas9, and results showed that there are 105 potential sites that can be split by intein derived from NpuDnaE (Supplementary Figure S2). In this study, the SaCas9 protein was first split into a 533 aa N-terminal fragment (P1) and a 520 aa C-terminal fragment (P2), with the split site located at HNH endonuclease domains that cleave the DNA strands complementary to the guide RNA (split-v1), and these two fragments were fused with N-terminal and C-terminal of NpuDnaE intein, respectively (Figure 1 and Supplementary note S1). The fused coding sequence was driven by maize ubiquitin and the double 35 S promoter, respectively (Figure 1B and Supplementary note S1). The full-length SaCas9 and split SaCas9 without fusion with intein (split-v2) were designed as control groups for comparison (Figure 1B and Supplementary note S1). Subsequently, three gRNAs driven by an OsU3 promoter targeted to the OsPDS, OsNYC1, and OsNYC4 genes were constructed and cloned into the vector containing the full-length or split SaCas9 expression cassette, respectively (Figure 1B and Supplementary note S1). Sequence analysis of the transgenic rice plants revealed that the editing efficiency of split-v1 is comparable to that of full-length SaCas9 across all three target genes. Specifically, the editing efficiencies of split-v1 in OsNYC1, OsNYC4, and OsPDS were 72.2%, 96.1%, and 81%, respectively, while the corresponding efficiencies for full-length SaCas9 were 75%, 92%, and 80%. For the OsNYC4 and OsPDS targets, the editing efficiency of split-v1 was slightly higher than that of full-length SaCas9; however, the proportion of homologous and biallelic mutations was lower, with split-v1 achieving 69.2% and 47.7% for OsNYC4 and OsPDS, respectively, compared to 76% and 65% for full-length SaCas9 (Figures 1D, E). However, split-v2 showed no editing activity (Figrure 1D). These findings are in line with the Western blot and immunoprecipitation results: the full-length SaCas9 was detected in split-v1 transgenic plants but not in those with split-v2. Possibly due to antibody characteristics or the lower abundance of the target protein, full-length SaCas9 could not be detected via Western blot, but it can be detected using the immunoprecipitation method (Figure 2A).

To explore the possibility of splitting SaCas9 into three parts, we divided P2 into P3 and P4 and fused them with intein^N or intein^C (split-v3), respectively (Figures 1A, B). The results demonstrated a significant decrease in efficiency, with only the vector targeted to OsNYC4 showing an editing efficiency of 23.3%, and among these edited plants, no homozygous/biallelic lines were found (Figure 1D). The lower efficiency observed in split-v3 may be attributed to the misassembly of P1, P3, and P4, which could potentially be improved using different inteins.

The phenotype of randomly selected Ospds, Osnyc1, and Osnyc4 mutants confirmed the sequencing results, with photobleaching observed in the Ospds mutant (Figure 2B) and a stay-green phenotype in dark-induced leaf senescence of Osnyc1 and Osnyc4 mutants (Figure 2C).

To test whether split SaCas9 can be utilized in VIGE, we modified the barley stripe mosaic virus (BSMV) vector as previously described (Cheuk and Houde, 2018). The modified vectors have the four-component BSMV system and have been demonstrated to allow overexpression of cDNAs of up to 2,100 nucleotides. In this study, we utilized RNAγ1 to express a gRNA targeting the LcHRC gene of sheepgrass (Leymus chinensis). Specifically, we constructed three vector combinations: in the first combination (Comb1), RNAγ1 expressed P1-intein^N and intein^C-P2; in the second combination (Comb2), RNAγ1 expressed P1-intein^N, while RNAγ2 expressed intein^C-P2; and in the third combination (Comb3), RNAγ2 expressed both P1-intein^N and intein^C-P2 (Figure 3). These constructs were then introduced into N. benthamiana plants for further analysis. After 7 days, the homogenate containing BSMV was harvested and used to infect sheepgrass (Figures 3A, B). After 14 days, we sequenced the LcHRC gene of the infected leaves and new leaves using next-generation sequencing. The results show that no editing events were detected in the sheepgrass. We hypothesize that the lack of genome editing activity may be attributed to the excessive amount of Agrobacterium components simultaneously injected into N. benthamiana. To test this, we separately introduced three vector combinations into N. benthamiana plants overexpressing RNAα and RNAβ and subsequently used the resulting virus to infect sheepgrass leaves. Next-generation sequencing revealed detectable genome editing activity in the Comb3-infected leaves, with an average editing efficiency reaching 30.95% (Figures 3B–D). In contrast, genome editing activity was undetectable in the leaves infected with Comb1 and Comb2, as well as in the control group, where RNAγ1 was used to express the gRNA and RNAγ2 was used to express the full-length SaCas9. To evaluate the efficiency of this approach and its applicability as a high-throughput gRNA screening platform in sheepgrass, additional target sites were designed for validation. These included LcHRC-T1 targeting the LcHRC gene; LcGW2-A-T1, LcGW2-B-T1, LcGW2-A-T2, and LcGW2-B-T2 targeting the LcGW2 gene; and LcTB1-A-T1 and LcTB1-B-T1 targeting the LcTB1 gene. All target sites exhibited detectable editing activity, with genome editing efficiencies ranging from 10.40% to 37.03%. Surprisingly, genome editing activity was not detectable in the new leaves of plants that had not been infected with the virus. This may be due to the size of the split SaCas9 still having some impact on the pathogenicity or mobility of the virus. Overcoming this limitation can be achieved by 1) optimizing viral vectors to enhance the ability of viruses to express exogenous genes and 2) improving the gene editing efficiency of reported smaller nucleases like CasΦ (Pausch et al., 2020; Liu et al., 2022; Li et al., 2023; Wang et al., 2023; Sun Y. et al., 2024; Zhao et al., 2024), Cas12f1 (Bigelyte et al., 2021; Wu et al., 2021; Kim et al., 2022), Casλ (Al-Shayeb et al., 2022), and TnpB (Karvelis et al., 2021) in plants and employing inteins to facilitate splicing and achieve smaller transcription units.