Enhanced exonuclease–Cas9 systems promote multiple nucleotide deletions with higher efficiency and broader targeting scope in plants

ABSTRACT

CRISPR–Cas9 is a widely used platform for plant genome editing, but its outcomes are typically dominated by small insertions and deletions (indels). Such limited mutation profiles restrict its utility in functional studies of non‐coding RNAs and regulatory elements, such as microRNAs (miRNAs), untranslated regions (UTRs), and promoter sequences, where larger sequence disruptions are often required. Here, we developed enhanced exonuclease–Cas9 platforms, termed multiple nucleotide deletion Cas9 (MND–Cas9) systems, for efficient generation of large deletions in rice. By screening four exonucleases (RecJ, T5, TREX2, and SbcB), we established MND–Cas9v1 systems based on TREX2 or SbcB that produced substantially larger deletions without reducing editing efficiency. Further optimization with an inserted DNA‐binding domain (DBD) between Cas9 and exonuclease yielded MND–Cas9v2, which simultaneously enhanced efficiency and deletion size. To expand PAM compatibility, we introduced PAM‐relaxed Cas9–NG and SpG variants, generating MND–Cas9–NG/SpGv2 systems with broader targeting scope and superior performance compared to their parental nucleases. Finally, we demonstrated the utility of these systems in two applications: MND–Cas9v2 efficiently knocked out the miRNA gene OsMIR530, producing larger seeds, and generated extended deletions in the 3′UTR of OsGhd2, which upregulated its expression and increased grain size. These results demonstrate that MND–Cas9 systems enable high‐efficiency generation of extended deletions and facilitate functional analyses of non‐coding RNAs and regulatory sequences. Overall, this work establishes a versatile and expandable exonuclease–Cas9 platform that substantially broadens the mutational spectrum and application potential of CRISPR‐Cas9 for plant genome engineering.

Cas9 preferentially generates small indels. By tethering an exonuclease and a DNA‐binding domain (DBD) to Cas9, the Exo‐DBD‐Cas9 system efficiently promotes multiple nucleotide deletions. These expanded deletions enable effective disruption of miRNA‐mediated regulation and relieve translational repression through targeted editing of 3′ untranslated regions.

INTRODUCTION

Clustered regularly interspaced short palindromic repeats (CRISPRs) and their associated nuclease Cas9 have been developed into a versatile and widely adopted platform for genome editing in diverse organisms, including plants (Sander and Joung, 2014; Tang and Zhang, 2023; Tuncel et al., 2025). Cas9 is directed by a single guide RNA (sgRNA) to complementary DNA sequences and requires a protospacer adjacent motif (PAM) for target recognition and cleavage (Jinek et al., 2012). The broad utility of this system has greatly accelerated both fundamental research and crop improvement.

While conventional Cas9 has been extensively used for generating loss‐of‐function alleles, the mutational outcomes are typically restricted to short insertions or deletions (indels), mostly in the range of 1–2 bp (Tang et al., 2016; Tang et al., 2018; He et al., 2024a). Such small indels, when present in coding regions, are sufficient for gene knockouts but often fall short for dissecting non‐coding sequences. For instance, point mutations or minimal indels are usually insufficient to block microRNA (miRNA) maturation or disrupt cis‐regulatory motifs embedded in promoter regions (Zhou et al., 2017, 2023; Liu et al., 2025; Zebell et al., 2025). To enlarge Cas9‐induced deletions, several strategies have been explored. Expression of TREX2 with Cas9 has been reported to promote extended deletions in multiple plant species (Cermak et al., 2017; Weiss et al., 2020; Ordon et al., 2023; Capdeville et al., 2024; He et al., 2024a). Similarly, fusion of exonucleases such as T5 or SbcB to Cas9 was shown to enhance deletion length in rice and zebrafish, respectively (Clements et al., 2017; Zhang et al., 2020a). However, previous studies mainly focused on improving editing efficiency, without systematically characterizing how exonuclease fusion alters Cas9‐induced mutation patterns. Moreover, the potential applications of these expanded deletion profiles in editing regulatory elements, promoters, or miRNA loci have rarely been explored in plants.

Another key factor shaping Cas9 utility is PAM compatibility. The widely used Streptococcus pyogenes Cas9 (SpCas9) requires a canonical NGG PAM (Jiang et al., 2013), restricting access to many genomic sites. Engineered variants with altered PAM recognition, such as SpCas9‐VQR, –EQR, and –VRER (Hu et al., 2016), iSpyMacCas9 (Chatterjee et al., 2020; Sretenovic et al., 2021), Cas9‐NG (Nishimasu et al., 2018; Zhong et al., 2019; Wu et al., 2022; Fan et al., 2024), SpG (Walton et al., 2020; Li et al., 2021), and SpRY (Walton et al., 2020; Ren et al., 2021a, 2021c; Wu et al., 2022; Chen et al., 2024), have broadened the target range. Among these, Cas9‐NG and SpG are capable of recognizing relaxed NGN PAM, offering greatly increased targeting scope in plant genomes. However, trade‐offs between expanded PAM compatibility, editing efficiency, and off‐target effects remain an important consideration (Tang et al., 2018; Walton et al., 2020; Wu et al., 2022; He et al., 2024b). It remains very important to improve the editing efficiency of such Cas9 variants so that they can be broadly adopted for genome editing in plants.

Building on these advances, we systematically engineered and optimized exonuclease–Cas9 architectures for efficient multi‐nucleotide deletions (MNDs) in rice. We first generated MND–Cas9v1 by fusing exonucleases (e.g., Trex2 or SbcB) to Cas9, which significantly increased deletion size without compromising editing efficiency. To further enhance editing outcomes, we introduced DNA‐binding domains to tether the exonuclease to Cas9, creating MND–Cas9v2 (Trex2–DBD–Cas9 and SbcB–DBD‐Cas9). We further expanded the target range of the MND–Cas9v2 system by incorporating SpG and Cas9–NG variants and showed that MND–SpG and MND–Cas9–NG conferred higher editing efficiency with enlarged deletions. As proof‐of‐concept demonstrations, we applied MND–Cas9v2 to generate much larger deletions than wild‐type Cas9 in a microRNA gene (OsmiR530) and the 3′UTR regulatory region of OsGhd2. In both cases, efficient knockout resulted in enlarged grain size. Together, the established MND–Cas9v2 system is a versatile platform that expands both the mutational spectrum and the targeting scope of CRISPR‐Cas9 editing in plants.

RESULTS

Establishment of exonuclease–Cas9 fusion systems for multi‐nucleotide deletion in rice

Cas9 has been reported to show a relatively slow dissociation rate from DNA double‐strand break (DSB) substrates, preferentially releasing the 3′ end of the non‐target strand (Richardson et al., 2016). We reasoned that this unique biochemical property could be exploited by coupling Cas9 with exonucleases that process single‐stranded DNA. Specifically, we hypothesized that fusion of exonucleases to Cas9 would resect the liberated 3′ end, thereby generating staggered overhangs and biasing repair through the non‐homologous end joining (NHEJ) pathway toward multiple nucleotide deletions (MNDs) rather than simple indels (Figure 1A).

Figure 1.

Development of the MND–Cas9 system in rice

(A) Schematic illustration of the working principles of the MND–Cas9 genome editing system. (B) Editing efficiencies of different systems in rice protoplasts at 6 target sites. (C) Deletion size distributions generated by different systems in rice protoplasts at 6 target sites. (D) Deletion position profiles generated by different systems in rice protoplasts at 6 target sites. Each dot represents a biological replicate. Data are presented as mean values ± SD. Data were analyzed using a two‐tailed unpaired t‐test. ns, P > 0.05; **P < 0.01. Solid line, median; dashed line, quartiles.

To systematically evaluate this strategy, we selected four exonucleases with distinct origins and catalytic features: RecJ (E. coli 5′ → 3′ exonuclease), T5 exonuclease (bacteriophage‐derived, 5′ → 3′ activity), SbcB (E. coli exonuclease I, 3′ → 5′ activity), and Trex2 (human 3′ → 5′ exonuclease). Each nuclease was fused to either the N‐ or C‐terminus of SpCas9, generating a panel of eight candidate MND‐Cas9 systems (Figure S1A). Rice protoplasts were used as a rapid and robust assay platform, and six endogenous sites were chosen to represent diverse sequence contexts. Editing outcomes were quantified by next‐generation sequencing (NGS) of PCR amplicons, ensuring single‐nucleotide resolution of deletion events. Initial comparisons revealed that N‐terminal fusions consistently outperformed C‐terminal fusions in terms of editing efficiency, suggesting that spatial positioning of the nuclease domain is critical for functional coupling with Cas9 (Figure S1B). Consequently, all subsequent experiments focused on N‐terminal fusion constructs. Among these, T5‐Cas9 showed a significant enhancement in the overall editing efficiency relative to wild‐type Cas9, whereas SbcB–Cas9, Trex2–Cas9, and RecJ–Cas9 maintained comparable efficiencies (Figure 1B).

Beyond overall efficiency, detailed profiling of editing signatures highlighted clear differences between these systems. Wild‐type Cas9 and RecJ‐Cas9 primarily generated 1–2 bp deletions (Figure S2A, E), consistent with canonical Cas9 editing patterns. T5‐Cas9 modestly increased the deletion size range to 1–3 bp (Figure S2D). In sharp contrast, SbcB–Cas9 and Trex2–Cas9 substantially shifted the indel spectrum toward larger deletions, with a pronounced peak at 6–12 bp (Figures 1C, S2B, C). Analysis of editing profiles further demonstrated that Cas9‐generated deletions clustered within a narrow 4–5 bp position upstream of the PAM, while SbcB–Cas9 and Trex2–Cas9 expanded the deletion range to 4–15 bp, indicating an extended editing range (Figure 1D).

Together, these results support our hypothesis that coupling 3′ → 5′ exonucleases to Cas9 can generate large deletions (Figure 1A). Hence, we established SbcB–Cas9 and Trex2–Cas9 fusions as a first‐generation multiple nucleotide deletion system (MND–Cas9v1). These tools not only maintain the efficiency of canonical Cas9 but also introduce a distinct editing signature characterized by expanded deletion sizes and broader editing ranges in rice cells. This laid the foundation for further engineering of advanced MND–Cas9 variants.

Exo–DBD–Cas9 fusions establish second‐generation MND tools with enhanced efficiency and deletion profiles

Previous studies on base editors have demonstrated that fusing a single‐stranded DNA‐binding domain (DBD) between Cas9 and deaminases significantly enhances editing efficiency, largely by stabilizing the displaced non‐target strand and extending the time window for enzymatic activity (Zhang et al., 2020b; Tan et al., 2022; Zheng et al., 2024). Inspired by this strategy, we hypothesized that linking exonucleases to Cas9 through a DBD could similarly stabilize the liberated non‐template strand, thereby providing exonucleases with prolonged access and improving both editing efficiency and deletion size (Figure 2A).

Figure 2.

Optimization of the MND–Cas9 system in rice

(A) Schematic illustration of the Exo–DBD–Cas9 genome editing system. (B) Vector design of Exo–DBD–Cas9. (C) Editing efficiencies of different systems in rice protoplasts at six target sites. (D) Deletion size distributions generated by different systems in rice protoplasts at 6 target sites. (E) Deletion position profiles generated by different systems in rice protoplasts at 6 target sites. Each dot represents a biological replicate. Data are presented as mean values ± SD. Data were analyzed using a two‐tailed unpaired t‐test. **P < 0.01; ***P < 0.001.

To test this concept, we focused on SbcB and Trex2, two 3′ → 5′ exonucleases that had shown the greatest potential for enlarging deletion size and broadening deletion range in the first‐generation systems. We constructed two new architectures: SbcB–DBD–Cas9 and Trex2–DBD–Cas9 (Figure 2B) and generated multiplex constructs targeting six endogenous rice sites (Figure S3A). To rigorously assess the contribution of the DBD module, we selected the same set of target sites that were previously used for the first‐generation multiple nucleotide deletion system (MND–Cas9v1), enabling direct comparison across Cas9, Exo–Cas9, and Exo–DBD–Cas9 architectures under identical sequence contexts. Following transformation into rice protoplasts, editing efficiencies were quantified by NGS of PCR amplicons. Both SbcB–DBD–Cas9 and Trex2–DBD–Cas9 displayed significantly higher editing efficiencies compared with wild‐type Cas9 (Figures 2C, S3B). Analysis of editing profiles revealed that SbcB–DBD–Cas9 produced deletions predominantly ranging from 6 to 14 bp, while Trex2–DBD–Cas9 generated deletions between 5 and 11 bp, both markedly larger than those produced by Cas9 (Figures 2D, S4A, B). Furthermore, analysis of editing profiles indicated that SbcB–DBD–Cas9 and Trex2–DBD–Cas9 expanded the deletion position to 5–18 bp and 5–15 bp upstream of the PAM, respectively, representing a substantial extension of the deletion range relative to Cas9 (Figure 2E).

Collectively, these results demonstrate that the incorporation of a DBD into exonuclease–Cas9 fusions markedly improves genome editing outcomes by enhancing both editing efficiency and multi‐nucleotide deletion capacity. In contrast to the MND–Cas9v1 system, which only expanded deletion sizes, the Exo–DBD–Cas9 architecture achieves coordinated improvements in both editing efficiency and deletion size. The resulting SbcB–DBD–Cas9 and Trex2–DBD–Cas9 represent second‐generation multiple nucleotide deletion Cas9 systems (MND–Cas9v2) with superior performance in rice cells.

Expansion of MND–Cas9 systems via PAM‐flexible Cas9 variants

Although the first‐ and second‐generation MND–Cas9 systems showed markedly improved deletion efficiencies and expanded editing ranges, their activity remained constrained by the NGG PAM requirement of SpCas9. To further broaden the targeting scope, we explored PAM‐flexible Cas9 variants, Cas9–NG and SpG, which recognize NGN PAM and therefore theoretically quadruple the accessible target space in the rice genome (Figure 3A). We fused SbcB–DBD or Trex2–DBD to the N‐terminus of Cas9–NG and SpG, producing two additional MND–Cas9v2 systems (Figure 3B). For each system, we tested eight endogenous rice loci using two multiplex constructs per system, with each construct containing four sgRNAs targeting distinct loci (Figure S5A). Following protoplast transformation, editing outcomes were assessed by NGS. For SpG, the addition of SbcB–DBD enhanced editing efficiency at NGA, NGC, and NCC PAM sites, while Trex2–DBD consistently improved editing efficiency across all four NGN PAM types (Figures 3C, S5B). In the case of Cas9–NG, SbcB–DBD fusion yielded editing efficiencies comparable to that of Cas9–NG, whereas Trex2–DBD conferred robust enhancement at all tested PAM variants (Figures 3C, S5B).

Figure 3.

Expansion of PAM compatibility using Cas9–NG and SpG variants

(A) Schematic illustration of Cas9, Cas9–NG, and SpG targeting the rice genome. (B) Vector designs of different genome editing tools. (C) Editing efficiencies of different systems in rice protoplasts. (D) Deletion size distributions generated by different systems in rice protoplasts at eight target sites. (E) Deletion position profiles of the SpG editing system in rice protoplasts at eight target sites. (F) Deletion position profiles of the Cas9–NG editing system in rice protoplasts at eight target sites. Each dot represents a biological replicate. Data are presented as mean values ± SD. Data were analyzed using a two‐tailed unpaired t‐test. *P < 0.05; **P < 0.01; and ***P < 0.001.

Editing profile analysis further highlighted the advantages of the Exo–DBD fusions. Both SbcB–DBD–SpG and Trex2–DBD–SpG shifted the deletion size distribution toward larger events compared with SpG alone (Figures 3D, S6A–C), a trend that was also observed for Cas9–NG fusions (Figures 3D, S7A–C). Analysis of editing profiles revealed that SpG‐mediated deletions were largely confined to a narrow 4–7 bp range upstream of the PAM. By contrast, SbcB–DBD–SpG expanded the deletion range to 5–18 bp, while Trex2–DBD–SpG extended it to 5–15 bp (Figure 3E). Similarly, Cas9–NG fusions showed broadened deletion range of 5–15 bp for SbcB–DBD–Cas9‐NG and 5–14 bp for Trex2–DBD–Cas9–NG, compared with 4–7 bp for Cas9–NG (Figure 3F).

Together, these data establish that Exo–DBD–Cas9–NG/SpG fusions effectively overcome PAM restrictions while simultaneously enhancing editing efficiency and expanding deletion profiles. These systems substantially broaden the versatility of CRISPR‐Cas9‐based genome editing, enabling efficient multi‐nucleotide deletion across a wider array of genomic loci in rice.

Functional application of MND–Cas9 for targeted deletion of OsmiR530

MicroRNAs (miRNAs) are critical regulators of gene expression at the post‐transcriptional level, often exerting strong control over developmental and physiological processes in plants (Yu et al., 2025). Previous studies have shown that conventional Cas9‐induced mutations predominantly occur around 3 bp upstream of the PAM, typically resulting in 1–2 bp indels. Such mutations rarely coincide with the critical stem or loop regions essential for miRNA processing, making it difficult to effectively disrupt miRNA maturation and function. (Zhou et al., 2017; He et al., 2024a; Liu et al., 2025; Zheng et al., 2025). By contrast, our MND–Cas9 systems typically generate multiple nucleotide deletions, which are well suited to abolish miRNA formation by disrupting critical structural elements. To explore the functional utility of MND–Cas9 for knocking out miRNA genes in stably transformed rice, we selected OsmiR530 as a test case, given its reported role in negatively regulating OsPL3, a gene influencing grain size (Sun et al., 2020). Based on prior findings, we hypothesized that deletion of OsmiR530 would relieve repression of OsPL3, thereby leading to increased grain size (Figure 4A).

Figure 4.

Generation of OsMIR530 mutants using the MND–Cas9 system

(A) Schematic illustration showing that OsMIR530 knockout upregulates OsPL3 expression, resulting in enlarged grains. (B) Predicted secondary structures of pre‐OsMIR530 mutants. (C) Genotype of MND–Cas9v2‐mediated genome editing at the OsmiR530 site in rice T0. Red background indicates the OsmiR530‐3p mature sequence. (D) Grain size of different OsMIR530 mutants compared with WT, Bars = 1 cm. (E) Expression levels of OsPL3 in WT and OsMIR530 mutants. (F) Representative plants of different OsMIR530 mutants and the wild type (WT). Bars = 30 cm. (G) Panicle number per plant of different OsMIR530 mutants and WT. (H) Plant height of different OsMIR530 mutants and WT. (I) Grain length of different OsMIR530 mutants and WT. (J) Grain width of different OsMIR530 mutants and WT. Each dot represents a biological replicate. Data are presented as mean values ± SD. Data were analyzed using a two‐tailed unpaired t‐test. **P < 0.01; ***P < 0.001.

To test this hypothesis, we constructed an editing vector based on the efficient MND–Cas9v2 system, targeting the OsmiR530‐3p region. By screening T0 lines resulting from Agrobacterium‐mediated transformation, we successfully obtained three independent biallelic mutants, designated osmir530‐m01, osmir530‐m02, and osmir530‐m03 (Figure 4C). These deletion sizes of these mutant alleles ranged from 7 bp to 19 bp (Figure 4C). The predicted secondary structures of the mutant alleles, analyzed using the Vienna RNA WebSuite (Gruber et al., 2008), revealed striking alterations relative to the wild‐type OsmiR530 precursor (Figure 4B). Hence, these structural disruptions are expected to abolish or severely impair mature miRNA biogenesis.

Phenotypic evaluation of the three osmir530 mutants showed no significant differences from the wild type in plant height or tiller number (Figure 4F, G, H). However, all three mutants displayed significantly increased grain length and width compared with the wild type (Figure 4D, I, J). Consistent with the expected regulatory relationship, qRT‐PCR analysis revealed that the expression of OsPL3 was significantly elevated in all mutant lines (Figure 4E), in agreement with earlier reports linking OsmiR530 to the repression of OsPL3 (Sun et al., 2020). By successfully abolishing OsmiR530 function and uncovering its phenotypic consequence on grain size, we establish MND–Cas9 as a powerful tool for probing and manipulating miRNA‐mediated regulatory pathways in plants.

Disruption of the OsGhd2 3′UTR SMITE element relieves translational repression

SMITE (small miniature inverted‐repeat transposable element) sequences embedded within 3′ untranslated regions (3′UTRs) have been reported to function as translational repressors in rice (Naito et al., 2009; Wei et al., 2014; Shen et al., 2017). For example, the SMITE element in OsGhd2 3′UTR forms a stable stem–loop structure that restricts protein accumulation (Shen et al., 2017; Figure 5A). To investigate the functional contribution of this element, we designed two distinct genome editing strategies using the MND–Cas9v2 system. In the first strategy, paired sgRNAs were introduced upstream and downstream of the SMITE sequence to generate a complete deletion of the element. In the second strategy, a single sgRNA was designed to target the stem–loop structure within the SMITE sequence, aiming to disrupt its secondary structure.

Figure 5.

Generation of OsGhd2 3′UTR‐edited lines using the MND–Cas9 system

(A) Schematic illustration showing that OsGhd2 3′UTR editing enhances its translation and influences grain size. (B) Genotype of MND–Cas9v2‐mediated genome editing at the OsGhd2 site in rice T0. Red background marks the mutation generated by the dual‐sgRNA strategy (OsGhd2‐m01). Blue background marks the mutations generated by the single‐sgRNA strategy (OsGhd2‐m02 and OsGhd2‐m03). (C) Grain size of different OsGhd2 3′UTR mutants compared with WT, Bar = 1 cm. (D) Expression levels of OsGhd2 in WT and OsGhd2 3′UTR mutants. (E) Representative plants of different OsGhd2 3′UTR mutants and the wild type (WT). Bar = 30 cm. (F) Panicle number per plant of different OsGhd2 3′UTR mutants and WT (G) Plant height of different OsGhd2 3′UTR mutants and WT. (H) Grain length of different OsGhd2 3′UTR mutants and WT. (I) Grain width of different OsGhd2 3′UTR mutants and WT. (J) Schematic representation of the dual‐fluorescence reporter construct showing the GFP reporter module containing simulated 3′UTR mutations. The internal control expression cassette is omitted for clarity. (K) Fluorescence observation of rice protoplasts transformed with the dual‐fluorescence reporter system. (L) ZsGreen/mCherry ratio in different 3′UTR mutants. Each dot represents a biological replicate. Data are presented as mean values ± SD. Data were analyzed using a two‐tailed unpaired t‐test. *P < 0.05.

Through stable rice transformation, we obtained three independent mutants. OsGhd2‐m01 carried a full deletion of the SMITE element generated by the dual‐sgRNA strategy, whereas OsGhd2‐m02 and OsGhd2‐m03 harbored partial deletions of the stem–loop structure produced by the single‐sgRNA strategy (Figure 5B). Phenotypic assessment revealed no significant differences in plant height or tiller number compared with the wild type (Figure 5E–G). However, all three mutants showed significantly increased grain length and width (Figure 5C, H, I), suggesting that SMITE disruption enhances grain traits.

We next examined OsGhd2 expression at the transcript level by qRT‐PCR. Consistent with earlier studies, transcript abundance was not significantly altered in the three mutants relative to the wild type (Figure 5D), supporting the notion that the SMITE element primarily regulates translation rather than transcription. To further validate this conclusion, we established a dual‐fluorescence reporter assay (Zhang et al., 2024; He et al., 2024b) incorporating the OsGhd2 promoter and terminator sequences to mimic the native transcriptional context. The 5′UTR of OsGhd2 was fused upstream of ZsGreen, while the 3′UTR was replaced with either the wild‐type OsGhd2 3′UTR or the mutant 3′UTRs (m01, m02, and m03) (Figure 5J). In parallel, an mCherry expression cassette was co‐expressed within the same construct to serve as an internal normalization control. These reporter constructs were transiently delivered into rice protoplasts, and the ZsGreen/mCherry fluorescence ratio was quantified to assess translational output. The results showed that ZsGreen expression normalized to mCherry was significantly higher when driven by the mutant 3′UTRs compared with the wild‐type 3′UTR (Figure 5K, L), indicating that deletion of the SMITE element indeed alleviates translational repression.

Together, these results demonstrate that both complete and partial deletions of the OsGhd2 3′UTR SMITE element using Trex2–DBD–Cas9 relieve translational inhibition and enhance grain size. This highlights the potential of our MND–Cas9v2 systems to dissect and manipulate non‐coding regulatory sequences such as 3′UTRs for agronomic trait improvement in plants.

DISCUSSION

The CRISPR‐Cas9 system has transformed plant functional genomics and crop improvement by enabling efficient and precise genome modifications (Molla et al., 2021; Wang and Doudna, 2023; Li et al., 2024; Gilbertson et al., 2025). While standard Cas9 nucleases are effective at introducing small indels, they generally show limited capacity for generating large deletions, which restricts their application to functional interrogation of non‐coding regulatory elements such as untranslated regions (UTRs), promoters, and non‐coding RNAs. Expanding the editing profiles of Cas9 to encompass efficient multi‐nucleotide deletions would therefore broaden its utility for plant genome engineering.

In this study, we systematically engineered Cas9‐based systems to enhance large‐deletion outcomes and broaden targeting scope. Among our first‐generation MND–Cas9v1 systems, fusion of 3′–5′ exonucleases TREX2 or SbcB outperformed fusions based on 5′–3′ exonucleases, conferring comparable editing efficiency to Cas9 but much larger deletions. This result is consistent with the model that supports their mechanism of action (Figure 1A). These systems were further upgraded to MND–Cas9v2, which features tethering the exonucleases to Cas9 via a DNA‐binding domain (DBD). To expand targeting scope, we also incorporated Cas9–NG and SpG into this MND–Cas9v2 version. MND–Cas9v2 systems not only enhanced the size of deletions but also significantly improved editing efficiency relative to Cas9, demonstrating that spatial anchoring of the exonuclease can promote more robust and extensive DNA resection. Using MND–Cas9v2, we achieved efficient knockout of a rice miRNA locus (OsMIR530) and the 3′UTR of OsGhd2, highlighting the system's potential for dissecting non‐coding regulatory elements whose functional validation often requires the removal of relatively large DNA fragments.

Compared to CRISPR‐Cas9, CRISPR‐Cas12a generates much larger deletions (Tang et al., 2017; Zhong et al., 2018; Zhang et al., 2021; Liu et al., 2025), demonstrating its promising applications in editing miRNA genes and promoters (Zhou et al., 2022, 2023; Tang and Zhang, 2023; Zheng et al., 2025). However, CRISPR‐Cas12a has two main constraints that limit its wide adoption to many crops. First, Cas12a nucleases are more temperature‐sensitive than the widely used SpCas9 (Malzahn et al., 2019), despite recent improvements with engineered variants such as ttLbCas12a (Schindele and Puchta, 2020), ttLbCas12a Ultra (Xin et al., 2024), and LbCas12a‐RRV or RVQ (Zhang et al., 2023). Unless highly efficient Cas12a variants are used, CRISPR‐Cas12a‐mediated genome editing may face lower efficiency than CRISPR–SpCas9 in many crops. Second, the canonical PAM of Cas12a nucleases are TTTV, which is already more stringent than the NGG PAM of SpCas9. SpCas9 variants such as Cas9–NG and SpG further expand the targeting scope, making them ideal tools to pinpoint the target sites, such as specific non‐coding sequences. In this regard, our engineered MND–Cas9v2 system allows targeting at NGG PAM sites (with wild‐type Cas9) as well as NG PAM sites (with Cas9–NG and SpG), greatly broadening its applications with expanded targeting scope. Interestingly, we also observed that the performance of exonuclease–DBD fusions varied depending on the Cas9 backbone. Specifically, SbcB‐DBD enhanced editing efficiency of Cas9–NG at multiple target sites but showed little or no improvement when fused to SpG at the same target sites. Since Cas9–NG and SpG may differ slightly in PAM recognition and in the structural configuration of the DNA ends that they generate (Walton et al., 2020; Li et al., 2021), such subtle mechanistic differences could affect how exonucleases engage and resect DNA ends. Therefore, our results highlight that the effects of exonuclease fusion are not universally transferable across Cas9 variants, emphasizing the need to empirically test diverse nuclease‐exonuclease combinations to identify the most effective architectures for each context. Collectively, our MND–Cas9v2 platform provides a compelling solution to efficiently achieve targeted large deletions in plants. Although exonuclease fusion to Cas9 could, in principle, increase the likelihood of larger deletions, our current analyses did not detect extensive deletion events in either rice protoplasts or stable transformants. The largest deletions observed were 42 bp in protoplast assays and 28 bp in stable lines, suggesting the deletion sizes are still largely confined by the efficient NHEJ pathway. Hence, our results do not indicate an increased risk of extensive deletions associated with exonuclease–Cas9 fusions in plants. In summary, we developed a series of multi‐nucleotide deletion‐optimized Cas9 (MND–Cas9) systems through stepwise engineering. Our results establish MND–Cas9v2 systems as a versatile platform that complements existing Cas9‐based editors by specifically enabling robust large‐fragment deletions while retaining high editing efficiency and broadened targeting capability. This system expands the functional reach of CRISPR‐Cas9 in plants and lays the foundation for future applications in regulatory genomics, synthetic biology, and precision crop improvement.

MATERIALS AND METHODS

Plant materials

The Japonica cultivar Nipponbare was used in this study. The protoplast transformation materials are prepared by growing sterilized rice seeds in 1/2MS culture medium for 10–12 d at 28°C in the dark, resulting in rice seedlings suitable for protoplast transformation. For stable transformation, sterilized rice seeds are placed in callus induction medium (N6‐D) as described previously (Ren et al., 2021b), at 32°C with 24 h of light exposure. They are cultured for 7–10 d to obtain calli suitable for stable transformation of rice.

Construction of the vectors

The rice codon‐optimized NLS‐Trex2‐GGGGSx3_linker fragment was synthesized by Genscript (Nanjing, China) and assembled with the Zea mays ubiquitin 1 (ZmUbi1) promoter, NLS–Cas9–NLS, and the Arabidopsis thaliana heat shock protein (AtHSP) terminator using the Golden Gate assembly with BsaI‐HFv2 (New England Biolabs), generating the Trex2–Cas9 entry vector (pZR055). The sgRNA expression cassette driven by the OsUbi1 promoter and processed by tRNAs (pTX1290: pOsUbi1‐tRNA‐BsaI site‐lacZα‐BsaI site‐gRNA scaffold‐tRNA‐tpinII) was used to produce the sgRNA entry vector. The Trex2‐Cas9 entry vector, sgRNA entry vector, and pMOD_C0000a were assembled into the T‐DNA backbone pTRANS_210d (Addgene #91109) to generate the Trex2–Cas9 backbone (pGEL1277) using Golden Gate assembly as described previously (Zhang et al., 2013; Zhou et al., 2021; Zhong et al., 2023; Tang et al., 2024). Backbones of T5–Cas9, RecJ–Cas9, and SbcB–Cas9 (pGEL1278) were constructed in the same way.

For the Trex2–DBD–Cas9 construct, the rice codon‐optimized DBD fragment (Genscript, Nanjing, China) was assembled with ZmUbi1, NLS‐Trex2‐GGGGSx3_linker, NLS‐Cas9‐NLS, and the AtHSP terminator to obtain the Trex2–DBD–Cas9 entry vector (pZR170). The final Trex2–DBD–Cas9 backbone (pGEL1279) was generated by assembling the entry vector, the sgRNA entry vector, and pMOD_C0000a into pTRANS_210d. The SbcB–DBD–Cas9 backbone (pGEL1280) was constructed similarly. For the Trex2–DBD–SpG construct, NLS‐Trex2‐GGGGSx3_linker was assembled with ZmUbi1, DBD, NLS‐Cas9‐SpG‐NLS, and the AtHSP terminator to obtain the entry vector pZR324. The final Trex2–DBD–SpG backbone (pGEL1281) was assembled with the sgRNA entry vector and pMOD_C0000a into pTRANS_210d. The SbcB–DBD–SpG backbone (pGEL1282) was generated in the same way. For the Trex2–DBD–Cas9–NG construct, NLS‐Trex2‐GGGGSx3_linker was assembled with ZmUbi1, DBD, NLS‐Cas9‐NG‐NLS, and the AtHSP terminator to generate the entry vector pZR325. The final Trex2–DBD–Cas9–NG backbone (pGEL1283) was assembled with the sgRNA entry vector and pMOD_C0000a into pTRANS_210d. The SbcB–DBD–Cas9–NG backbone (pGEL1285) was constructed similarly.

Sanger sequencing was used to confirm the integrity of all vectors. Table S1 provides a comprehensive list of the target sites utilized in this study. Protein sequences of Trex2, T5, RecJ, SbcB, and DBD are shown in Figure S8. All plasmids constructed in this study are in the process of being submitted to Addgene and will be available to the research community upon publication.

Rice protoplast transformation

Seedlings of the Japonica rice cultivar Nipponbare were grown on 1/2MS solid medium for 10–12 d in the dark at 28°C. Protoplasts were isolated and transformed following previously established protocols, with minor modifications (Tang et al., 2017; Liu et al., 2022). In brief, healthy leaves were cut into 0.5–1.0 mm strips and incubated in enzyme solution. After 30 min of vacuum infiltration, the samples were gently agitated in the dark at 25°C (60–80 rpm) for 6 h. The digested mixture was filtered through a 40 μm nylon mesh, and protoplasts were washed twice with W5 buffer, examined under a microscope, and adjusted to a final density of 2 × 10⁶ cells/mL.

For transformation, 30 μg of plasmid DNA (1 μg/μL, prepared using the Qiagen Midiprep kit) in 30 μL was mixed with 200 μL of protoplast suspension and 230 μL of 40% PEG solution. Following a 30‐min incubation in the dark, the reaction was stopped by adding 1 mL of W5 buffer. Protoplasts were collected by low‐speed centrifugation and transferred into a 12‐well plate for incubation in the dark at 32°C for 48 h.

Rice stable transformation

Following our previous work (Zhou et al., 2019; Zheng et al., 2023), the Japonica cultivar Nipponbare was used for stable Agrobacterium‐mediated transformation. Rice seeds were dehulled, surface‐sterilized, and germinated on solid N6‐D medium. Calli derived from precultured seedlings were inoculated with Agrobacterium tumefaciens strain EHA105 harboring the recombinant expression vector. After a 3‐d co‐cultivation period on co‐culture medium, calli were washed with sterile water and transferred to N6‐S medium for 2 weeks of selection. Subsequently, proliferating calli were moved to RE‐III medium and cultured for an additional 2 weeks. Resistant calli were subcultured onto fresh RE‐III medium every 2 weeks until plant regeneration was achieved.

Detection and quantification of genome editing

The efficiency of genome editing in rice protoplasts was evaluated using NGS of PCR amplicons. Forty‐eight h after protoplast transformation, genomic DNA was extracted using the CTAB method (Stewart and Via, 1993). Target‐specific primers, each containing a unique 6‐base barcode at the 5′ end to allow sample identification, were synthesized (Zhong et al., 2019). Successful amplification was confirmed by gel electrophoresis. The amplicons of the editing regions were amplified by 2x Rapid Taq Master Mix (Vazyme, Nanjing, China). Amplicons were sent to Novogene (China, Tianjin) for deep sequencing by the Novaseq. 6000 platform, which produced 150 bp paired‐end reads. The editing frequency was analyzed by CRISPRMatch (You et al., 2018).

For plants generated through stable transformation, genomic DNA from T0 individuals was similarly extracted using the CTAB method. Target loci were amplified and subjected to Sanger sequencing. The resulting sequences were decoded using DSDecode (Xie et al., 2017) to determine the genotype of each T0 plant. Mutation types were classified by comparison with the reference genome.

RNA extraction and qRT‐PCR

Real‐time qPCR was conducted using the ChamQ Universal SYBR qPCR Master Mix (Vazyme, Nanjing, China) following the manufacturer's instructions, with OsActin mRNA serving as an internal control. The relative levels of gene expression were calculated using the 2^−ΔΔCt method. Three biological replicates (three independent mutant leaves) were examined to ensure reproducibility. The experiment was performed three times independently and similar results were obtained. All primers used in this study are listed in Table S2.

Data analysis

The data were analyzed with GraphPad Prism 9.0 software, and the figures were made using Adobe Photoshop and Adobe Illustrator software.

CONFLICTS OF INTEREST

The authors declare no conflicts of interest.

AUTHOR CONTRIBUTIONS

Y.Z. proposed the project and designed the experiments. R.Z. and X.T. generated all the constructs. R.Z. performed rice protoplast transformation. R.Z. and X.T. analyzed the mutation frequencies in protoplasts. R.Z., Q.R., and W.W. conducted rice stable transformation. R.Z. and W.W. analyzed the mutation frequencies in rice stable lines. R.Z. and Y.H. performed data analysis and generated the figures. R.Z. Y.Q., and Y.Z. analyzed the data and wrote the paper with input from other authors. All authors read and approved the final version of the manuscript.

Supporting information

Additional Supporting Information may be found online in the supporting information tab for this article: http://onlinelibrary.wiley.com/doi/10.1111/jipb.70155/suppinfo

Figure S1. Multiplex genome editing with Cas9 and exo–Cas9 in rice protoplasts

Figure S2. Deletion profiles induced by Cas9 and exonuclease fusions in rice protoplasts

Figure S3. Multiplex genome editing with Exo–DBD–Cas9 in rice protoplasts

Figure S4. Deletion profiles of Exo–DBD–Cas9 variants in rice protoplasts

Figure S5. Multiplex genome editing with Exo–DBD–‐SpG/Cas9–NG in rice protoplasts

Figure S6. Deletion profiles induced by SpG and Exo–DBD–SpG variants in rice protoplasts

Figure S7. Deletion profiles induced by Cas9–NG and Exo–DBD–Cas9–NG variants in rice protoplasts

Figure S8. Protein sequences of exonucleases and DNA‐binding domain (DBD)

Table S1. Target sites used in this study

Table S2. Oligos and sequence used in this study

Biographies

Zhang, R ., Tang, X ., He, Y ., Wang, W ., Ren, Q ., Qi, Y ., and Zhang, Y . (2026). Enhanced exonuclease–Cas9 systems promote multiple nucleotide deletions with higher efficiency and broader targeting scope in plants. J. Integr. Plant Biol. 68: 3316–3328.