A CRISPR tool designed for efficient editing of multiple genes at once

All animal experiments were approved and supervised by the Animal Welfare Committee of China Agricultural University (Approval ID: AW72205202-1-01).

Mammalian codon-optimized Cas12i.3 with nuclear localization signal, linker, Triplex, 20-target array, 30-target array, and T5 exonuclease (T5E) were synthesized by the GeneWiz Co., Ltd. and incorporated into a mammalian expression vector under the CBh promoter by Gibson Assembly. crRNA oligos, hammerhead (HH), and hepatitis delta virus (HDV) ribozyme oligos were synthesized by the GeneWiz Co., Ltd., annealed, and subsequently cloned under the U6 promoter within the previously mentioned mammalian expression vector. The sequences mentioned above are listed in. Supplementary Tables S1–S6

HEK293T cells, NIH-3T3 cells, sheep fibroblasts, and tdTomato sheep fibroblasts were cultured at 37°C with 5% CO₂ with high-glucose Dulbecco’s Modified Eagle’s Medium (DMEM, Gibco), 10% (v/v) fetal bovine serum (FBS, Gibco), 1% (v/v) penicillin–streptomycin (100×, Gibco). For transient transfection of HEK293T cells, the cells were seeded in 12-well plates and transfected at ∼70% confluency. 1.5 μg of vectors were transfected into HEK293T cells using Lipofectamine 3000 (Thermo Fisher Scientific) following the manufacturer’s protocol. For transient transfection of NIH-3T3 cells, sheep fibroblasts, and tdTomato sheep fibroblasts, cells were seeded at 70%–80% confluency in 10-cm dishes and then 2 × 10⁶ cells were electroporated with 7 μg of plasmid using Nucleofector II/2b device (Lonza) with program A-033.

A small sample of ear tissue was harvested and disinfected in 75% ethanol for 1 min, washed three times with phosphate-buffered saline, and then minced into ∼1 mm³ pieces with sterile scissors. The tissue was then placed in 200 μl of FBS in a 10-cm dish and incubated upside down at 37°C with 5% CO₂ for 1 h. After incubation, complete culture medium [DMEM (Gibco), 10% (v/v) FBS (Gibco), and 1% (v/v) penicillin–streptomycin (Gibco)] was gently added to avoid displacing the tissue. Fibroblasts began migrating from the explants after approximately one week. At ∼90% confluency, cells were trypsinized and collected for subsequent transfection.

To enable sensitive and efficient detection of editing efficiency across optimization steps, a tdTomato expression cassette was inserted into the single-copy ZFY gene to establish a stably transfected reporter cell line. Using CRISPR–Cas9 combined with homology-mediated end joining (HMEJ), the first intron of the ZFY gene in sheep fibroblasts was targeted to insert the CBh-tdTomato-SV40 poly(A) cassette. This reporter cell line enabled the assessment of CRISPR–Cas-mediated genome editing activity by measuring the ratio of quenched tdTomato fluorescence or weakly fluorescent cells following crRNA targeting.

A fluorescence reporter system combined with fluorescence-activated cell sorting (FACS) analysis was employed to evaluate editing activity under various conditions, including crRNA scaffold optimization, T5E fusion, codon optimization, and the screening of high-efficiency mutation combinations. Cells were dissociated with 0.25% trypsin-EDTA (Gibco) and suspended in FBS-containing DMEM. To assess Cas12i.3 nuclease activity using the fluorescence reporter system, ∼10 000 single cells were recorded with a BD LSRFortessa 48 h post-transfection. Editing efficiency was determined by measuring the ratio of quenched tdTomato fluorescence and weakly fluorescent cells among EGFP-positive cells.

T7 endonuclease I (T7E1) assays were performed to evaluate editing activity under various conditions, including crRNA scaffold optimization, T5 exonuclease fusion, codon optimization, and single-point mutant screening, as well as for comparisons between IOCas12i and Cas12i.3 or SpCas9. The polymerase chain reaction (PCR) products generated with the PrimeSTAR Max DNA Polymerase were hybridized in NEBuffer 2 (New England Biolabs) on a T100 thermal cycler (Bio-Rad). The primers for amplifying target loci in the T7E1 assay are provided in Supplementary Tables S2, S3, and S6. The annealed samples were then digested with T7E1 (New England Biolabs) for 15 min and separated by electrophoresis on a 2% agarose gel stained with GelStain (TransGen Biotech). The mutation frequency, represented as the percentage of insertions and deletions (indels), was calculated with the formula: Indel (%) = 100 × [1 − (1 − fraction cleaved)^1/2].

To compare the editing efficiency of optimized constructs (EOCas12i–Combo1 and EOCas12i–Combo2) with Cas12i.3, IOCas12i, CasSF01, SpCas9, and LbCas12a, targeted deep sequencing was performed. Genomic regions of interest were amplified by PrimeSTAR Max DNA Polymerase (Takara). Primers for PCR amplification of target genomic sites are listed in Supplementary Tables S4 and S5. The amplified products were purified using TaKaRa MiniBEST DNA Fragment Purification Kit Ver.4.0 (Takara) and sequenced by a DNBSEQ-T7 platform with 150-bp paired-end reads (Benagen Co., Ltd). Sequencing reads were analyzed by CRISPResso2 [15] for the quantification of editing efficiency.

To evaluate the multiplex editing capability of the EOCas12i platform, whole-genome sequencing (WGS) was performed at 100× coverage using the DNBSEQ-T7 platform. Raw reads were initially quality-filtered and trimmed to ensure a Q score of 30 or higher. The BWA-MEM aligner was used to map the reads to the GRCh38.p14 reference genome, and the resulting BAM files were sorted with Samtools and deduplicated using GATK. The final BAM files were then used to analyze editing efficiencies at each target site with CRISPRessoWGS [15], applying the parameters -wc -4 -w 6.

Cells were lysed in radioimmunoprecipitation assay (RIPA) buffer (Aidlab) supplemented with protease inhibitor cocktail. The supernatants of the lysates were quantified using a BCA Protein Assay Kit (Aidlab) and adjusted to a uniform concentration using RIPA buffer. After heating at 95°C for 5 min, equal volumes of cell lysate were run on 6% polyacrylamide gels and transferred to polyvinylidene difluoride (PVDF) membranes (Millipore). The membrane was then blocked in 5% nonfat milk in Tris-buffered saline with Tween 20 (TBST) buffer and incubated overnight with the primary antibody [anti-DYKDDDDK tag (Beyotime) and anti-Tubulin (Beyotime)] at 4°C. After three washes with TBST, the membranes were incubated with the secondary anti-mouse antibody (ZSGB-BIO) at room temperature for 1 h. The protein band was visualized using chemiluminescent substrates (Invitrogen, WP20005), according to the manufacturer’s instructions.

The structural model of Cas12i.3 was constructed using the SWISS-MODEL server (https://swissmodel.expasy.org↗), based on the structure of Cas12i2. The associated crRNA scaffold was similarly adapted from that of Cas12i2, substituting the sequence with validated ZFX and EMX1 target sites [16]. Cas12i.3 was then paired with crRNAs targeting ZFX or EMX1, and PyMOL (Version 2.3.2) was used to identify Cas12i.3 residues within 2 Å of the crRNAs targeting ZFX or EMX1. In parallel, Cas12i.3–crRNA complexes for ZFX or EMX1 were analyzed with protein–ligand interaction profiler (PLIP, https://plip-tool.biotec.tu-dresden.de↗) to identify interacting residues. Structural alignment with Cas12i2 further revealed three evolutionarily conserved residues in Cas12i.3—K181, Q432, and T453—which correspond to positions previously shown to enhance editing efficiency when mutated in Cas12i2 [17]. These conserved residues were therefore also included in the mutational screen. All candidate residues identified from spatial proximity, interaction profiling, and conserved functional sites were pooled for mutagenesis. Among these, non-positively charged residues were mutated to arginine to generate Cas12i.3 variants with potentially enhanced editing activity.

To generate linear DNA templates for in vitro transcription (IVT), IOCas12i and EOCas12i coding region with T7 promoter was amplified by PCR using Q5 High-Fidelity DNA Polymerase (New England Biolabs) and primers IVT-mRNA-F and -R (IVT-mRNA-F: 5′-TAATACGACTCACTATAGGGGCCACCATGGACTATAAGGACC-3′ and IVT-mRNA-R: 5′-TCCCCAGCATGCCTGCTATTCT-3′). The PCR products were purified using the TaKaRa MiniBEST DNA Fragment Purification Kit Ver.4.0 (Takara) and eluted in nuclease-free water. IVT reactions were performed using mMESSAGE mMACHINETM T7 ULTRA Kit (Ambion), following the manufacturer’s instructions. The resulting mRNA was eluted in nuclease-free water and stored at −80°C until use.

For crRNA synthesis, crRNA targeting DNMT1 locus with T7 promoter was amplified by PCR using the same polymerase but with primer IVT-DNMT1-F and -R (IVT-DNMT1-F: 5′-TAATACGACTCACTATAGGGAGAGAATGTGCGCATAGTCGCAC-3′ and IVT-DNMT1-R: 5′-AGGTCATCTACAAAGCCCCTTCTGTGCGACTATGCGCACA-3′). The PCR products were purified using the TaKaRa MiniBEST DNA Fragment Purification Kit Ver.4.0 (Takara) and eluted in nuclease-free water. IVT reactions for crRNA were performed using the MEGAshortscript^™ T7 kit (Ambion), following the manufacturer’s instructions. The resulting crRNA was eluted in nuclease-free water and stored at −80°C until use.

Seven- to eight-week-old female mice were injected with 5 IU of pregnant mare serum gonadotropin, followed 48 h later by 5 IU of human chorionic gonadotropin. Hormone-treated females were then placed in a mating with fertile stud males and embryos were collected the following day.

For crRNA and mRNA delivery, mixed crRNA and mRNAs were microinjected into zygotes from female C57 mice. After microinjections, embryos were either immediately transferred into pseudopregnant females or cultured in KSOM until they reached the blastocyst stage (E4.5). To generate post-implantation embryos or live pups, microinjected embryos were surgically transferred into the oviducts of pseudopregnant females immediately after microinjection. Ear clips were collected from 2-week-old mice, and genomic DNA was extracted using HiPure Tissue DNA Micro Kit (Magen Biotech). PCR was conducted on extracted DNA with PrimeSTAR Max DNA Polymerase (Takara).

Individual blastocyst stage embryos were incubated in 10 μl of embryo lysis buffer at 55°C for 30 min. PCR was performed using a 1 μl of sample with PrimeSTAR Max DNA Polymerase (Takara). The PCR products from the edited embryos or mice (ear clip samples) were subjected to 12% polyacrylamide gel electrophoresis, stained with GelStain (TransGen Biotech) for 30 min, and visualized using a UV transilluminator (Azure Biosystems). The primers used are listed in the. Supplementary Table S2

Forty-eight hours post-transfection, EGFP-positive cells (20 000–40 000 cells) were collected and genomic DNA was extracted using HiPure Tissue DNA Micro Kit (Magen Biotech). Potential off-target sites were predicted by Cas-OFFinder (https://www.rgenome.net/cas-offinder/↗, with mismatch numbers ≤4 and DNA bulge size ≤2) and amplified by PCR with PrimeSTAR Max DNA Polymerase (Takara). Off-target site information and primers were listed in Supplementary Table S4. The amplicons were purified using TaKaRa MiniBEST DNA Fragment Purification Kit Ver.4.0 (Takara) and sequenced by a DNBSEQ-T7 platform with 150-bp paired-end reads (Benagen Co., Ltd). Sequencing reads were analyzed by CRISPResso2 for the quantification of editing efficiency.

GUIDE-seq was employed to assess the on-target and off-target frequencies of EOCas12i–Combo1, EOCas12i–Combo2, and SpCas9 targeting ADRB2 and CXCR4, as previously described [18, 19]. GUIDE-seq and data analysis were completed by GeneRulor Co., Ltd. Briefly, 5 μl of dsODN (100 μM) and 10 μg of plasmids expressing nucleases and guide RNAs (gRNAs) were co-transfected into HEK293T cells. After 72 h, DNA was extracted and subjected to dsODN–PCR for cleavage verification. GUIDE-seq libraries were generated by DNA fragmentation, Y-adapter ligation, two rounds of PCR, and sequencing on the MGISEQ-2000RS (PE150, paired-end) with customized settings for 16-bp UMIs. The resulting data were demultiplexed using in-house Python scripts and analyzed using GUIDEseq v1.1 [20].

Cas12i.3 is a single crRNA-guided endonuclease with small protein size and simple PAM requirements, which recently discovered CRISPR–Cas system from metagenomes (Fig. 1A and Supplementary Table S1). We first tested genome-editing of this naturally occurring Cas12i.3 in mammalian cells. A construct expressing Cas12i.3 and a ZFX-targeting crRNA was generated (Supplementary Fig. S1A). SpCas9 targeting the same ZFX site served as a reference. The indel frequency of Cas12i.3 was only 8.6% in sheep fibroblasts, whereas that of SpCas9 reached 15.2%, nearly double the efficiency of Cas12i.3 (Supplementary Fig. S1B–D and Supplementary Table S2).

To expand the utility of this compact protein, we aimed to systematically optimize CRISPR–Cas12i.3. We first developed a tdTomato fluorescent reporter cell line based on sheep fibroblasts for evaluating editing efficiency during optimization (Supplementary Fig. S2). In this system, crRNA-guided cleavage within the tdTomato coding sequence results in frameshift mutations that either lead to nonfunctional proteins or introduce premature stop codons resulting in truncation, as well as in-frame deletions that remove essential residues—both of which lead to fluorescence quenching. The reporter cell line was generated by integrating tdTomato into the endogenous single-copy ZFY locus, enabling sensitive assessment of genome-editing efficiency by measuring the ratio of quenched fluorescence and weakly fluorescent cells following transfection with Cas12i.3 and tdTomato-targeting crRNA (Fig. 1B).

In the native CRISPR system, full-length Direct-Repeat crRNA (DRf-crRNA) is initially expressed and then processed by Cas12a into truncated Direct-Repeat crRNA (DRt-crRNA) [21]. In Cas12a, DRf-crRNA is more efficient than DRt-crRNA [10]. Additionally, extending the 5′ end of crRNA in Cas12a improves the editing efficiency [22]. To examine the impact of DRf-crRNA and DRt-crRNA on the nuclease activity of Cas12i.3, we transfected the tdTomato fluorescent reporter cells with the plasmids expressing Cas12i.3 and either DRf-crRNA or DRt-crRNA targeting tdTomato. The quenched fluorescence ratio of DRt-crRNA was 1.8-fold that of DRf-crRNA, with a similar trend observed for weakly fluorescent cells (Supplementary Fig. S3A, C, and E, and Supplementary Tables S1 and S2). T7E1 assay at the endogenous ZFX locus showed that DRt-crRNA achieved 1.6-fold the efficiency of DRf-crRNA in sheep fibroblasts (Supplementary Fig. S3H, and Supplementary Tables S1 and S2). These data indicate that DRt-crRNA is more effective than DRf-crRNA in the context of Cas12.3.

RNA polymerase III (Pol III) promoters preferentially initiate transcription with a guanine (G), but DRt-crRNA lacks a 5′ G, necessitating an extra nucleotide that may affect its activity. Moreover, Pol III-transcribed crRNAs often carry variable 3′ U-tails, which can influence editing efficiency [23]. To express precise DRt-crRNA efficiently, we inserted self-cleaving ribozymes—the HH and HDV ribozymes—at the 5′ and/or 3′ ends of the crRNA. Three crRNA expression plasmids were tested (HH–crRNA, crRNA–HDV, and HH–crRNA–HDV), all driven by the U6 promoter. For clarity, these are referred to as HH, HDV, and HH–HDV, respectively. Flow cytometry revealed that all three constructs exhibited significantly lower ratios of quenched fluorescence and weakly fluorescent cells compared to the DRt-crRNA control (Supplementary Fig. S3B, D, and F, and Supplementary Tables S1 and S2). Similar results were observed across the groups at the sheep endogenous ZFX locus (Supplementary Fig. S3I, and Supplementary Tables S1 and S2). These results indicate that HH and/or HDV ribozyme did not enhance CRISPR–Cas12i.3 editing efficiency.

Mismatched base pairs in the crRNA stem–loop impair Cas12a activity, while A:U to G:C substitutions has been shown to improve editng efficiency by stabilizing crRNA [10, 24]. To explore whether similar stem–loop engineering could enhance Cas12i.3 function, eight stem–loop modifications were engineered by replacing A:U pairs with G:C pairs and were named DRf-1 to DRf-8 (Fig. 1C and D, and Supplementary Table S1). Among them, DRf-4 exhibited the highest activity, with a quenched fluorescence ratio of 32.2%, which was 1.8-fold that of wild-type DRf-crRNA (17.7%; DRf-WT; Fig. 1E, and Supplementary Tables S1 and S2). Additionally, DRf-4 significantly increased the ratio of weakly fluorescent cells from 7.9% in DRf-WT to 10.7% (Supplementary Fig. S3G, and Supplementary Tables S1 and S2). At the sheep endogenous ZFX locus, DRf-4 similarly demonstrated superior editing efficiency (Supplementary Fig. S3J, and Supplementary Tables S1 and S2). These results establish DRf-4 as a structurally optimized crRNA that significantly improves Cas12i.3-mediated genome editing.

Given that Cas12i.3 generates DNA cleavage products with long 5′ overhangs prone to accurate repair without causing indels, we investigated whether a 5′ exonuclease could enhance its editing efficiency by removing these overhangs (Fig. 2A). T5E has previously been shown to boost editing efficiency for Cas9 and Cas12a [25, 26]. We fused T5E to either the N- or C-terminus of the Cas12i.3 protein using four distinct linkers: (GGGGS)₂, (GGGGS)₃, SGGSGGSGGS, and XTEN, designated as linker 1, linker 2, linker 3, and linker 4, respectively. The term “N-linker 1″ refers to the fusion of T5E to the N-terminus of Cas12i.3 using linker 1, with similar meanings for “N-linker 2,” “N-linker 3,” and so on (Fig. 2B and Supplementary Table S1). Flow cytometry revealed that C-linker 3 achieved the highest quenched fluorescence ratio (46.1%), which was 2.3-fold that of the control (20.0%) (Fig. 2C and Supplementary Table S2). Furthermore, C-linker 3 significantly increased the proportion of weakly fluorescent cells within the EGFP-positive population compared to the control (8.4% versus 5.0%; Supplementary Fig. S4A and Supplementary Table S2). Consistent with these findings, T7E1 assays on the endogenous ZFX locus in sheep fibroblasts demonstrated that both C-linker 1 and C-linker 3 yielded an editing efficiency of 15.9%, a 35.9% improvement over the control (11.7%; Supplementary Fig. S4B and Supplementary Table S2). Taken together, these results indicate that the fusion of T5E to the C-terminus of Cas12i.3 via linker 3 (amino acid sequence: SGGSGGSGGS) most effectively enhances Cas12i.3-mediated genome editing.

Codon optimization of Cas12i.3 was further explored as a strategy to enhance editing efficiency by improving protein expression through alignment with the host’s codon usage. We evaluated four mammalian-optimized codon schemes (Codon OptimWiz, GeneOptimizer, Jcat, and General Biol) alongside an unoptimized sequence. All optimized schemes showed increased codon adaptation index (CAI) values for both humans and sheep, with Codon OptimWiz and Jcat group achieving near-maximal CAI values (Supplementary Fig. S4C and Supplementary Table S1). However, Jcat had >70% GC content in certain regions, potentially affecting expression (Supplementary Fig. S4D). Fluorescence microscopy, flow cytometry, and western blotting confirmed Codon OptimWiz as the top performer in Cas12i.3 expression (Fig. 2D and E, and Supplementary Fig. S4E).

To further investigate the four codon optimizations on Cas12i.3 editing efficiency, expression vectors with different codon optimizations were constructed. Quenched fluorescence ratio for the Unoptimized, Codon OptimWiz, GeneOptimizer, Jcat, and General Biol groups were 21.1%, 30.1%, 28.0%, 25.2%, and 21.3%, respectively, with Codon OptimWiz group achieving the highest rate at 1.4-fold that of the Unoptimized group (Fig. 2F and Supplementary Table S2). Similarly, a higher proportion of weakly fluorescent cells was observed in the Codon OptimWiz group compared to the Unoptimized group (Supplementary Fig. S4F and Supplementary Table S2). For the endogenous ZFX locus, Codon OptimWiz, GeneOptimizer, and Jcat groups also showed higher editing efficiencies, at 15.6%, 15.8%, and 14.5%, respectively, while Unoptimized and General Biol groups showed lower efficiencies of 13.2% and 8.2%, respectively (Supplementary Fig. S4G and Supplementary Table S2). Together, these data demonstrate that Codon OptimWiz optimizations enhance both Cas12i.3 expression and editing efficiency.

To further enhance efficiency, we integrated optimized crRNA design, Cas12i.3 codon optimization, and exonuclease fusion to develop IOCas12i system (Fig. 3A). Interestingly, a C-to-A mutation at the third codon of amino acid 273 (S273R) in IOCas12i occurred by chance during vector construction. Given IOCas12i’s superior editing efficiency, this mutation was retained. Compared with Cas12i.3, IOCas12i exhibited a quenched fluorescence ratio 2.0-fold that of Cas12i.3, with a similar trend observed for weakly fluorescent cells (Fig. 3B, and Supplementary Fig. S5A and Supplementary Table S2). Editing efficiency at the endogenous ZFX locus increased from 11.2% with Cas12i.3 to 26.8% with the optimized IOCas12i, as measured by T7E1 assay (Supplementary Fig. S5B and Supplementary Table S2). To evaluate the broad applicability of IOCas12i across mammalian species, we next assessed its editing performance in human HEK293T and mouse NIH-3T3 cells. In HEK293T cells, IOCas12i achieved 2.3–2.7-fold the editing efficiencies of Cas12i.3 at EMX1 site 1, EMX1 site 2, and FANCF (Fig. 3C, and Supplementary Fig. S5C–E and Supplementary Table S3). In NIH-3T3 cells, IOCas12i demonstrated 2.7–3.9-fold the editing efficiencies of Cas12i.3 at DNMT1 site 1, DNMT1 site 2, EMX1, and FANCF (Fig. 3D, and Supplementary Fig. S5F–I and Supplementary Table S3). Across all tested sites, Cas12i.3 and IOCas12i showed average editing efficiencies of 11.2% and 27.8%, respectively, in HEK293T cells, and 7.3% and 24.9% in NIH-3T3 cells. We further compared the editing activity of IOCas12i and SpCas9 at EMX1 site 1, EMX1 site 2, and FANCF in HEK293T cells using identical spacer sequences (5′-TTN PAM for IOCas12i and NGG-3′ PAM for SpCas9). At EMX1 site 1, SpCas9 was 1.1 times as efficient as IOCas12i. In contrast, at EMX1 site 2 and FANCF, IOCas12i achieved 1.1 and 1.4 times the efficiency of SpCas9, respectively (Fig. 3E, and Supplementary Fig. S5J–L and Supplementary Table S3). Importantly, both Cas12i.3 and IOCas12i exhibited minimal off-target activity, whereas SpCas9 showed potential off-target effects at a predicted site (Fig. 3F and G, and Supplementary Table S4). These findings demonstrate that IOCas12i, through the integration of optimized crRNA, codon-optimized Cas12i.3, and exonuclease fusion, substantially enhances genome-editing efficiency without increasing off-target effects.

To evaluate IOCas12i’s impact on deletion length, PCR amplicons surrounding the gRNA target sites were subjected to next-generation sequencing (NGS) analysis. The results revealed that IOCas12i primarily produces deletions of 11–40 bp (74.8%), whereas Cas12i.3 and SpCas9 predominantly generated shorter deletions of 0–10 bp (71.9% and 81.7%, respectively), as shown in Fig. 3H, and Supplementary Fig. S6 and Supplementary Table S4. Overall, these findings indicate that IOCas12i induces longer deletions compared to Cas12i.3 and SpCas9.

To assess the applicability of IOCas12i for embryo editing and gene-edited animal generation, we employed a triplex structure to place the crRNA targeting the mouse DNMT1 gene downstream of the IOCas12i coding sequence. In vitro-transcribed mRNA was microinjected into single-cell mouse embryos, which were subsequently cultured to the blastocyst stage and lysed for editing efficiency analysis via polyacrylamide gel electrophoresis (PAGE). Among 16 embryos analyzed, two displayed additional shorter bands. TA cloning and sequencing revealed 9- and 18-bp deletions in these embryos, indicating that 2 out of 16 embryos (12.5%) carried detectable edits (Fig. 3I, and Supplementary Fig. S7A and Supplementary Table S2). Subsequently, 40 embryos were transferred into two recipient females, resulting in eight pups. One pup exhibited marked growth retardation and appeared noticeably smaller than its littermates. PAGE and sequencing revealed an 18-bp deletion at the DNMT1 locus in this individual, representing an editing efficiency of 12.5% (1/8) among the offspring (Fig. 3J, Supplementary Fig. S7B and C; Supplementary Table S2). These findings demonstrate the efficacy of IOCas12i for both embryo editing and gene-edited animal generation.

To further enhance genome-editing activity of IOCas12i, we applied a protein engineering strategy targeting its DNA–crRNA binding pocket. We hypothesized that certain amino acid substitutions with this region might modulate the interaction between the Cas protein and DNA or crRNA. To test this, we introduced a total of 42 individual amino acid substitutions at three distinct positions within the binding interface and constructed a plasmid library co-expressing each IOCas12i variant with EGFP. Following transfection of HEK293T cells and FACS sorting of EGFP-positive populations, genomic DNA was extracted and editing activity was assessed using T7E1 assays (Fig. 4A and Supplementary Fig. S8A). Several mutations were found to significantly improve editing efficiency in mammalian cells, including T235R, D267R, S477R, G478R, D551R, L662R, and T850R. In the next stage, we combined these mutations with a set of previously reported beneficial substitutions, including S7R, N168R, D233R, L332R, T505R, S599R, and D851R (Fig. 4B–D and Supplementary Fig. S8B–G). To systematically evaluate the combinatorial effect of these mutations, we grouped them based on their spatial proximity into three regions—designated Region 1, Region 2, and Region 3. Within each region, representative combinations were constructed and evaluated using a tdTomato-targeting assay in fluorescent reporter sheep fibroblasts (Fig. 4E–I and Supplementary Fig. S9). In Region 1, among the six tested single and combined variants, L332R and D851R individually conferred the greatest enhancement in editing efficiency, each reaching 1.1-fold that of the IOCas12i (Fig. 4F and Supplementary Fig. S9A). In Region 2, due to the higher density of beneficial candidates, we initially evaluated various combinations of N168R, D233R, T235R, and D267R. Among the eight tested variants, N168R and N168R/D267R showed the most notable improvement. Further gains were achieved by incorporating S7R, T505R, and S599R into these backbones, with N168R/S599R and S7R/N168R/T505R emerging as top performers (Fig. 4G and H, and Supplementary Fig. S9B and C). In Region 3, of 11 variants tested, G478R and G478R/D551R demonstrated the most pronounced enhancement in editing activity (Fig. 4I and Supplementary Fig. S9D).

Having identified the most effective amino acid substitutions in each of the three defined structural regions of IOCas12i, we next sought to integrate these region-specific mutations to generate enhanced variants. Because IOCas12i itself incorporates multiple layers of optimization—including engineered crRNA architecture, codon usage adaptation, and exonuclease fusion—the resulting variants inherently combine these features with the newly introduced amino acid substitutions.

Based on prior data, N168R/S599R and S7R/N168R/T505R represented the top-performing combinations in Region 2. Building upon these backbones, we further introduced high-efficiency substitutions from Regions 1 and 3 to generate two series of variants: combinations 1–1 to 1–8 (derived from N168R/S599R) and combinations 2–1 to 2–8 (from S7R/N168R/T505R). Among the N168R/S599R-derived combinations, combinations 1–5 (N168R/L332R/G478R/S599R) exhibited the highest editing efficiency, while combinations 2–6 (S7R/N168R/L332R/G478R/T505R/D551R) showed the best performance in the S7R/N168R/T505R background. The top-performing constructs were designated as EOCas12i–Combo1 and EOCas12i–Combo2, respectively (Fig. 5A–C and Supplementary Fig. S10).

To evaluate the genome-editing capabilities of the optimized variants EOCas12i–Combo1 and EOCas12i–Combo2, we performed deep-targeted sequencing at eight representative endogenous loci—ADRB2, CHRM4, FANCF, CXCR4, AR, CD2, GRIN2B, and HBB—in HEK293T cells. On average, EOCas12i–Combo1 and EOCas12i–Combo2 generated indels at these eight sites with frequencies of 64.1% and 66.6%, respectively. Compared with wild-type Cas12i.3, the two engineered variants exhibited editing efficiencies ranging from 2.5- to 22.8-fold and 3.0- to 60.0-fold those of Cas12i.3 (Fig. 5D and E, and Supplementary Table S5). Relative to IOCas12i, their editing efficiency ranged from 1.2- to 3.1-fold (Combo1) and 1.1- to 8.1-fold (Combo2) those of IOCas12i across the tested sites (Fig. 5F and Supplementary Table S5). Notably, at the CHRM4, FANCF, CD2, and HBB loci, both EOCas12i variants surpassed SpCas9 in editing efficiency, while showing comparable or slightly reduced activity at ADRB2, CXCR4, and AR, with the exception of GRIN2B. When benchmarked against LbCas12a, both EOCas12i–Combo1 and EOCas12i–Combo2 demonstrated similar editing efficiencies at all loci except GRIN2B (Fig. 5G and Supplementary Table S5). A similar trend was observed when compared with CasSF01, with both variants showing comparable editing efficiencies at seven of the eight tested loci (Supplementary Fig. S11). Collectively, these results highlight the strong and competitive editing performance of EOCas12i–Combo1 and EOCas12i–Combo2.

To assess EOCas12i’s impact on deletion length, we performed deep sequencing analysis on PCR amplicons encompassing gRNA target sites in HEK293T cells. Both the IOCas12i and EOCas12i systems induced a higher proportion of deletions ranging from 11 to 40 bp at ADRB2, CHRM4, FANCF, and CXCR4, whereas deletions by Cas12i.3, SpCas9, and LbCas12a were predominantly restricted to 10 bp or smaller, especially for SpCas9 (Fig. 6A, and Supplementary S12 and Supplementary Table S5). These findings indicate that EOCas12i shares a similar deletion pattern with IOCas12i, favoring the generation of larger deletions.

To investigate the specificities of EOCas12i, we analyzed the genome-wide off-target effects of EOCas12i–Combo1, EOCas12i–Combo2, and SpCas9 at ADRB2 and CXCR4 loci using GUIDE-seq in HEK293T cells. At the ADRB2 locus, the number of off-target sites was similar for all three systems. In contrast, at the CXCR4 locus, SpCas9 exhibited more off-target sites than both EOCas12i–Combo1 and EOCas12i–Combo2 (Fig. 6B). These results suggest that EOCas12i–Combo1 and EOCas12i–Combo2 may offer superior specificity compared to SpCas9.

To further evaluate the application of the EOCas12i–Combo2 in generating gene-edited animals, in vitro-transcribed mRNA and crRNA targeting DNMT1 were microinjected into collected single-cell mouse embryos. A total of 65 two-cell embryos were transferred into the oviducts of three recipient females, resulting in ten pups. PAGE analysis of the genomic DNA from these pups revealed that four (40%) exhibited additional shorter bands alongside the wild-type, indicative of indel mutations. Subsequent TA cloning and sequencing confirmed deletions at the DNMT1 locus (Fig. 6C and Supplementary Table S2). These results validate the effectiveness of the EOCas12i–Combo2 system in generating gene-edited animals.

We evaluated the multiplexed genome editing capabilities of the EOCas12i platform using two CRISPR arrays: a 20-target array (12 loci: ACTB, GAPDH, LMNA, AR, ADRB2, CCR4, CCR10, CD2, CHRM4, CXCR4, HBB, and IL1RN) and a 30-target array (17 loci: ACTB, GAPDH, LMNA, AR, ADRB2, CCR4, CCR10, CD2, CHRM4, CXCR4, HBB, IL1RN, DNMT1, EMX1, FANCF, GRIN2B, and VEGFA). These arrays were positioned downstream of the EOCas12i coding sequence in a triplex structure (Fig. 7A and Supplementary Table S6). WGS followed by CRISPRessoWGS analysis was used to assess editing efficiency of EOCas12i–Combo1 and EOCas12i–Combo2 in HEK293T cells for both arrays. The results revealed consistently high editing efficiency at all target loci, regardless of whether a 20-target or 30-target array was used, or whether the vector was EOCas12i–Combo1 or EOCas12i–Combo2 (Fig. 7B). Specifically, T7E1 assays were conducted at positions 3, 5, 7, 8, 9, and 12 for the 20-target array, and at positions 3, 5, 7, 8, 9, 12, 21, 24, 28, and 30 for the 30-target array. Efficient cleavage was observed at all loci, except at position 5 of the 20-target array in EOCas12i–Combo1. Notably, the crRNA at position 30 of the 30-target array achieved editing efficiencies of 31.9% and 41.0% for EOCas12i–Combo1 and EOCas12i–Combo2, respectively (Fig. 7C and Supplementary Fig. S13). Statistical analysis revealed that EOCas12i–Combo1 generally exhibited higher average editing efficiency compared to EOCas12i–Combo2 for both the 20- and 30-target arrays (Fig. 7D). When the first 20 targets of the 30-target array were compared to the 20-target array, no significant difference in editing efficiency was observed, suggesting that the number of targets in the latter positions of the array does not substantially affect editing efficiency at the earlier loci (data not shown). Despite the multiplexed genome editing involving up to 30 targets, both vectors primarily generated long indels (>10 bp), similar to those observed in single-gene editing (Fig. 7E). These results confirm that both EOCas12i–Combo1 and EOCas12i–Combo2 are effective for multiplexed genome editing.

The raw sequencing data from targeted deep sequencing, whole-genome sequencing (WGS), and GUIDE-seq generated in our study have been deposited in the NCBI Sequence Read Archive (SRA) under the BioProject PRJNA1247221.