Recent Progress and Uses of Plant Gene Editing Tools

The first-generation cytosine base editing system (CBE) was successfully developed by David R. Liu's group in 2016 based on the CRISPR/Cas9 platform, consisting of a fusion protein of dCas9 and a cytidine deaminase together with an sgRNA. Four different cytidine deaminases—hAID, hAPOBEC3G, rAPOBEC1, and pmCDA1—were evaluated for their deamination activity on single-stranded DNA, and rAPOBEC1 (rAC1) from rat showed the highest activity. This enzyme was fused via an amino acid linker to the N-terminus of dCas9 to generate CBE1, but its C-to-T editing efficiency in HEK293T cells was only 0.8–7.7% [10]. The editing process of this system (Figure 1A) involves three key steps: sgRNA directs the fusion protein to the target DNA sequence, the cytidine deaminase catalyzes the deamination of cytosine (C) to uracil (U), and subsequently U is recognized as thymine (T) during DNA replication or repair, while the complementary guanine (G) is replaced by adenine (A), thereby achieving precise C/G-to-T/A base substitution. To improve efficiency, successive optimization of CBE was carried out. Komor et al. [10] first developed CBE2 by fusing a uracil DNA glycosylase inhibitor (UGI) to the C-terminus of CBE1, which blocked uracil excision repair, increased efficiency threefold to 20%, and reduced unintended indels. They then replaced dCas9 with nCas9, which retains single-strand nicking activity, generating CBE3 and further improving editing efficiency by 2–6 fold, up to 37%. Subsequently, researchers developed CBE4 by adding a second UGI to the C-terminus of CBE3 and extending the linker sequences between the deaminase, nCas9, and UGI. This new editor achieved 15–90% efficiency, representing a 50% improvement over CBE3 [11]. Later studies demonstrated that nuclear localization signals (NLSs) also influenced base editing performance. Koblan et al. [12] tested six NLS configurations and found that adding a bipartite NLS (bpNLS) to both the N- and C-termini of CBE4 yielded the best performance. With further codon optimization of the entire construct, including APOBEC1 and nCas9 domains, they created CBE4max, which enhanced editing efficiency across various cell types by 1.8–9 fold, reaching up to 89%. Particularly at low dosage or at difficult-to-edit sites, CBE4max showed remarkable advantages, for example, achieving 77% efficiency at the SCN9a splice acceptor site in mouse neuroblastoma cells, a 5.5-fold increase compared to CBE4 (14%). Whether optimizing codons or nuclear localization signals (NLS), the ultimate goal is to enhance editing efficiency by increasing the expression level of nCas proteins in the nucleus. Given the critical role of cytidine deaminases in base editing, extensive efforts have been made to identify or engineer novel deaminases with improved properties. Gehrke et al. (2018) [13] replaced rAPOBEC1 with human APOBEC3A (A3A) and introduced an N57G mutation to create eA3A, which conferred higher precision by favoring TC motifs, though with an efficiency of 20–50% similar to CBE3. Thuronyi et al. (2019) [14] employed the bacteriophage-assisted continuous evolution (BE-PACE) system to evolve three deaminases: APOBEC1, CDA1, and FERNY (the latter reconstructed through ancestral sequence reconstruction from node 656 of the APOBEC family phylogenetic tree, which is 29% smaller than APOBEC1). This process yielded evoAPOBEC1 (H122L+D124N), evoCDA1 (A123V), and evoFERNY (H102P+D104N). These variants were subsequently incorporated into the CBE4max editor. EvoAPOBEC1-BE4max exhibited markedly enhanced editing efficiency at GC-rich sites, increasing from 2.3% to 58% at the HEK3 locus. EvoFERNY-BE4max showed even higher activity at GC-rich sites than evoAPOBEC1, achieving 70% efficiency at the HEK3 locus. In contrast, evoCDA1-BE4max significantly improved editing at difficult-to-target loci, such as the TMC1 hearing-loss model, reaching 24% efficiency, which represents a 2.6-fold increase over CDA1-BE4max. Interestingly, Chen et al. [15] engineered an adenine deaminase TadA-8e by introducing an N46L mutation to abolish its adenine deaminase activity and confer cytidine deamination capability, while Neugebauer et al. [16] further evolved TadA-8e using phage-assisted approaches to generate TadA-CD variants with strong activity toward deoxycytidine. These engineered TadA-8e variants were used to develop new CBEs (Td-CBE or TadCBE). In HEK293T cells, Td-CBEmax (E27R) achieved C-to-T editing efficiencies of 57.7–94.9% at 13 endogenous loci, compared with 44.9–93.2% for BE4max [15], while TadCBE averaged 51–60% efficiency across nine sites, comparable to BE4max (47%) and evoAPOBEC1-BE4max (55%), but superior to evoFERNY-BE4max (41%) [16]. Notably, TadA-derived CBEs not only maintained or exceeded the editing activity of conventional CBEs but also offered smaller size, lower indel frequencies, and dramatically reduced Cas-independent DNA and RNA off-target editing.

Following the development of the cytosine base editing system (CBE), the team led by David R. Liu successfully established the adenine base editing system (ABE) in 2017, with the key innovation being the replacement of the cytosine deaminase in CBE with an engineered adenine deaminase. By fusing a mutated TadA* (TadA A106V and D108N) with nCas9, the researchers constructed the first-generation adenine base editor, ABE1.2, which achieved an editing efficiency of only 3.2 ± 0.88% [17]. The working principle of ABE is similar to that of CBE (Figure 1B): sgRNA directs the nCas9–deaminase fusion protein to a specific DNA site, where the adenine deaminase catalyzes the deamination of adenine (A) to produce inosine (I). During DNA replication, I is recognized as guanine (G), while the complementary thymine (T) is corrected to cytosine (C), thereby achieving a targeted A/T-to-G/C base pair conversion. Importantly, naturally occurring adenosine and adenine deaminases such as E. coli TadA and ecTadA, human ADAR2, and mouse ADA act exclusively on various forms of RNA. Through directed evolution, the research team re-engineered TadA and successfully endowed it with DNA editing activity, representing a critical breakthrough in the realization of ABE technology [17]. To further enhance editing efficiency, Gaudelli and colleagues [17] subjected wild-type TadA to artificial evolution using error-prone PCR, DNA shuffling, and other strategies to generate random amino acid substitutions, and after seven rounds of optimization, developed a series of ABE systems ranging from ABE1.2 to ABE7.10. ABE7.10 was designed by fusing nCas9 with a heterodimer of wild-type TadA and an evolved TadA7.10, which increased the editing efficiency from 3% to 58%. Koblan et al. [12] replaced the SV40 NLS of ABE7.10 with a bipartite nuclear localization signal (bis-bpNLS) and optimized codon usage with a GenScript strategy, generating ABEmax, which further improved efficiency, achieving 46% editing at the HBG promoter site—seven times higher than ABE7.10. Gaudelli's team [18] next evolved ABE7.10 using an adenine deaminase variant library to create the ABE8 series (40 variants classified into ABE8.x-m, monomers composed solely of the evolved TadA* without wild-type TadA, about 500 bp shorter in sequence than dimeric versions, and ABE8.x-d, heterodimers fusing wild-type TadA with evolved TadA*). Among these, ABE8.20-m showed optimal performance, reaching 60% editing efficiency in CD34 cells and as high as 98–99% in primary human T cells, thereby greatly expanding editing efficiency and applicability. Meanwhile, Richter et al. [19] applied phage-assisted non-continuous and continuous evolution (PANCE and PACE) to evolve the deaminase component of ABE7.10, producing ABE8e, which exhibited deamination activity 590 times higher than ABE7.10, with editing efficiencies in mammalian cells increasing from 1.7–20% to 18–86%.

The glycosylase base editor (GBE) was first developed by Zhao et al. in 2020 [20], overcoming the limitation of cytosine base editors (CBEs) and adenine base editors (ABEs), which could only mediate transitions between pyrimidines or between purines, and enabling pyrimidine-to-purine conversions. GBE1.0 achieved editing efficiencies ranging from 5.3% to 53.0% in mammalian cells. The core components of GBE include a cytidine deaminase, uracil-DNA glycosylase (UNG), nCas9, and sgRNA. Guided by the sgRNA, the fusion protein binds to the target DNA, where the cytidine deaminase converts cytosine (C) into uracil (U). UNG subsequently removes U, generating an apurinic/apyrimidinic (AP) site. The formation of this AP site, together with a nick introduced by nCas9 on the non-edited strand, triggers DNA repair and replication, thereby favoring the insertion of guanine (G) at the AP site and ultimately achieving precise C/G-to-G/C base pair conversion (Figure 1C). In the same year, Kurt and colleagues [21] developed the CGBE1 and miniCGBE1 editors for C/G-to-G/C transversions. CGBE1 consists of an RNA-guided Cas9 nickase, an Escherichia coli–derived uracil DNA N-glycosylase (eUNG), and a rat APOBEC1 cytidine deaminase variant (R33A), achieving an average editing efficiency of 14.4% across 25 sites. MiniCGBE1, generated by removing the eUNG domain, showed a slightly reduced average efficiency of 13% but significantly lowered the frequency of insertions and deletions (indels). To further enhance editing efficiency, Sun et al. [22] subsequently developed GBE2.0 (APOBEC(R33A)-nCas9-Rad51-Ung1), which incorporated several key enhancements. These included replacing human Ung with the more active yeast Ung1, introducing the APOBEC1 R33A variant, and fusing the DNA repair protein Rad51. The single-stranded DNA-binding domain (ssDBD) of Rad51 plays a critical role in DNA repair and enhances the deaminase's affinity for ssDNA, thereby boosting the editor's efficiency. This optimized version achieved an average editing efficiency of 30.88% across 17 chromosomal loci, representing a twofold increase compared with GBE1.0 (15.54%). Dong et al. [23] adopted an activation-based strategy by fusing VP64 to GBE to generate VP64-GBE, which significantly improved performance, yielding an average efficiency of 24.96% across 18 sites in HEK293T cells, compared with only 11.91% for the original GBE. Further incorporation of the SunTag system produced SunTag-GBEs, which not only enhanced efficiency and product purity but also expanded the effective editing window to the C6–C8 positions, raising the editing efficiency at the hard-to-edit locus PPP1R12C-5 to 35%. Chen et al. [24] developed another class of CGBEs by harnessing the base excision repair (BER) pathway. Based on CBE3 and CBE4, they removed UGI and fused core BER proteins including XRCC1, DNA ligase III (LIG3), and the 8-kDa domain of DNA polymerase β (PB), constructing more than 30 editors. Among them, rAPOBEC-nCas9-rXRCC1 performed best, achieving an average efficiency of 15.4% for C-to-G editing in human cells, with a maximum of 37%. Interestingly, by introducing an N46L mutation into TadA-8e to abolish its adenine deaminase activity, researchers generated a TadA-8e derivative with cytidine deaminase activity. The resulting Td-CGBE enabled highly efficient and precise C/G-to-G/C editing, reaching up to 72.8% efficiency in HEK293T cells [15].

Zong et al. (2017) [25] constructed a plant base editor (nCas9-PBE) by combining rat APOBEC1 cytidine deaminase, nCas9, and the uracil glycosylase inhibitor (UGI), then they performed codon optimization for cereal plants and cloning under the maize ubiquitin-1 (Ubi-1) promoter. In rice, wheat, and maize protoplasts, the system achieved C-to-T editing efficiencies of 5.8%, 6.8%, and 4.0%, respectively. At the OsSPL14 locus in rice, the base C-to-T editing efficiency reached 7.07%, while at the TaLOX2 locus in wheat, multiplexed C-to-T editing reached 12.48%. Building on this system, Zong et al. (2018) [26] replaced rat APOBEC1 with codon-optimized human APOBEC3A for cereals, generating A3A-PBE. This editor reached C-to-T editing efficiencies of 24.5% in rice protoplasts and up to 82.9% in regenerated rice plants at the OsCDC48 locus. In wheat protoplasts, four target sites showed an average efficiency of 13.1%, and in regenerated plants, the TaALS locus reached 22.5%. Notably, the editing window of A3A-PBE expanded to 17 nucleotides (positions 1–17), compared with only 7 nucleotides (positions 3–9) for the original PBE. Zeng et al. (2020) [27] codon-optimized hA3A, evoAPOBEC1, evoCDA1, and evoFERNY for rice and paired them with rice codon-optimized Cas9n-NG [28], which recognizes NG PAM sites, to construct four PhieCBEs: PevoAC1-NG, PevoFERNY-NG, PevoCDA1-NG, and PhA3A-max-NG. Among them, PevoFERNY-NG demonstrated the most robust overall editing characteristics, achieving C-to-T efficiencies as high as 86.3% in rice. Li et al. (2018) [29], based on the ABE7.10 architecture (ecTadA-ecTadA* fused to nCas9-D10A), constructed seven plant ABE variants (PABE-1 to PABE-7) by altering the position of the deaminase and modifying the number and placement of nuclear localization signals (NLSs). PABE-7 (deaminase at the N-terminus with three C-terminal NLSs) performed best, reaching 32.8% A-to-G efficiency in rice and wheat protoplasts, 15.8–59.1% in regenerated rice plants, but only 0.4% and 1.1% in regenerated wheat plants.Yan et al. (2018) [30] optimized codons of wild-type E. coli TadA and its mutant TadA*7.10 and fused them to nCas9 to construct rBE14, which enabled A-to-G editing in rice genes OsMPK6, OsWRKY13, OsSERK2, and OsWRKY45, with efficiency peaking at 62.26% in OsWRKY45. Wei et al. (2021) [31] constructed rABE8e using codon-optimized ABE8e for plants, achieving 85.7% and 82.1% A-to-G editing efficiencies at EPSPS and ALS loci in rice, with some Wx gene targets reaching 100%. Yan et al. (2021) [32] introduced V82S/Q154R mutations into TadA8e to generate TadA9, and constructed rBE46b and rBE48b using TadA9 and TadA8e, respectively. In rice, rBE46b achieved A-to-G efficiencies of 56.25%, 86.96%, and 97.92% at OsMPK6, OsMPK13, and Tms9-1, while rBE48b efficiencies were 58.33%, 18.75%, and 14.29%. Interestingly, TadA9 monomers displayed comparable or even superior activity within the editing window.Tan et al. (2022) [33] developed hyper ABE8e (hyABE8e) by inserting the single-stranded DNA-binding domain (DBD) from RADIATION SENSITIVE 51 (Rad51) [34] between TadA8e and nCas9 and performing codon optimization for rice. This system reached an average A-to-G editing efficiency of 78.5% across 29 rice loci, with four sites (TS4, TS12, TS14, TS16) achieving 100% efficiency. Sretenovic et al. (2021) [35] adapted CGBE for plants by using maize codon-optimized Cas9 (zCas9) and rice codon-optimized components, producing three versions: pYPQ265K (UNG-rAPOBEC1(R33A)-zCas9(D10A)), pYPQ265N1 (rAPOBEC1-zCas9(D10A)-UNG), and pYPQ265O1 (rAPOBEC1-zCas9(D10A)-rXRCC1). In rice and tomato protoplasts, pYPQ265K showed the highest efficiency, achieving C-to-G editing rates of 0.25–1.75% and 0.3–0.7%, respectively. In rice T₀ plants, only pYPQ265O1 generated C-to-G edits at the OsALS-sgRNA32 locus, with an efficiency of 4.8%, though all versions produced C-to-T byproducts. Tian et al. (2022) [36] improved UNG codon optimization and used highly active deaminases to construct OsCGBE03, which achieved only 1.69% efficiency in rice protoplasts but reached an average of 21.3% C-to-G efficiency across five genes (OsIPA1, OsbZIP5, OsSIR1, OsALS1, NRT1.18) in rice plants. Fan et al. (2024) [37] tested TadA-8e-derived cytosine base editors (TadCBEa, TadCBEd, TadCBEd V106W) in rice and tomato protoplasts, achieving high-efficiency, high-purity, and low-off-target C-to-T editing. In rice protoplasts, efficiency peaked at 55.4%, with extremely low indel rates, making these editors superior to traditional CBEs for applications requiring minimal off-target effects and byproducts. Most recently, Jiang et al. [38] replaced human UNG (hUNG) in CGBEs with cod UNG (coUNG), a uracil DNA glycosylase derived from cold-adapted codfish Gadus morhua, thereby improving C-to-G editing efficiency by 1.71- to 2.54-fold. In rice T₀ plants, efficiencies ranged from 33.33% to 68.75%. They further integrated TadA-8e-derived cytidine deaminase components (TadA-CDs) by fusing coUNG with TadA-CDc and SpCas9 nickase, generating CDc-CGBEco. This editor achieved homozygous C-to-G editing without byproducts in up to 52.08% of transgenic lines, while heterozygous editing reached 83.33%, with no significant off-target effects observed.

With the ongoing optimization of plant base editors in terms of efficiency, specificity, and editing window, their applications have rapidly expanded from fundamental research to practical crop improvement. A series of highly efficient and precise editing systems have established a solid technical foundation for the targeted development of novel germplasm in major crops, demonstrating promising potential in rice, wheat, and a variety of other plant species.

As one of the world's most important staple crops, rice was the first system in which base editing technology was tested and successfully applied. As shown in Table 1, base editing was applied to the relevant rice genes to confer important agronomic traits, including herbicide resistance and improved nitrogen use efficiency. In 2017, Li et al. [39] used BE3 to target three sites within the OsSBEIIb gene, which encodes starch branching enzyme IIb, and the OsPDS gene, which encodes phytoene desaturase. Their work demonstrated that base editing achieved much higher efficiency in targeted base correction compared with HDR-mediated gene replacement, while also showing that base-editing vectors were easier to construct than HDR plasmids (Table 1). In the same year, Lu and Zhu [40] employed a base editor lacking the UGI domain to edit the NRT1.1B gene, which encodes a nitrogen transporter, and the SLR1 gene, which encodes a DELLA protein where amino acid substitutions in or near its TVHYNP motif reduce plant height. They achieved C-to-T/G editing efficiencies of 2.7% and 13.3%, respectively (Table 1). In 2018, Li et al. [29] introduced point mutations into the acetyl-CoA carboxylase (ACC) gene of rice using base editing, thereby generating herbicide-resistant rice lines (Table 1). In 2020, Kuang et al. [41] employed both CBE and ABE to generate nearly saturated mutations of the endogenous herbicide target gene OsALS1, uncovering novel herbicide-resistance loci and creating new resistant germplasm (Table 1). In the same year, Liu et al. [42] performed precise base editing of the α-tubulin homologous gene OsTubA2, producing rice germplasm resistant to the herbicides trifluralin and dinitroaniline (Table 1). In 2022, Tian et al. [36] used the newly developed OsCGBE03 base editor to target genes such as OsIPA1, OsbZIP5, OsSIR1, and OsALS1, successfully conferring either high nitrogen-use efficiency or herbicide resistance to rice plants (Table 1).

Although many studies have demonstrated successful applications of base editing in rice, reports in wheat remain comparatively scarce. In 2019, Zhang et al. [43] applied ABE to edit the TaALS-P174 gene, which encodes acetolactate synthase, generating herbicide-resistant wheat. Double mutants of TaALS-P174 and TaALS-G631 showed slightly greater resistance to pyroxsulam than single mutants, while ALS mutations in TaALS-P174 conferred strong resistance to this herbicide. Using CBE, they also introduced a substitution at position 174 of the TaALS gene, converting proline to serine, alanine, or phenylalanine, thereby conferring resistance to sulfonylurea, imidazolinone, and aryloxyphenoxypropionate herbicides (Table 1). In 2022, Han et al. [44] developed a highly efficient wheat ABE (WhieABE) based on the hyperactive adenine deaminase TadA8e. WhieABE was used to edit wheat tubulin alleles located on chromosomes 1A, 1B, 1D, 4A, and 4D, introducing A-to-G edits that caused Met-to-Thr substitutions in wheat tubulin and enhanced resistance to dinitroaniline herbicides (Table 1).

Beyond the major cereal crops, base editing technology has also been applied in other fields and horticultural crops. In 2019, Veillet et al. [45] used a CBE system to edit the StGBSS1 gene in tetraploid potato, introducing base substitutions within the KTGGL catalytic domain, successfully applying base editing to a polyploid crop and creating potato mutants with impaired amylose biosynthesis across four alleles (Table 1). In 2018, Kang et al. [46] applied an ABE system to edit the FT and PDS3 genes in Arabidopsis and Brassica napus protoplasts, generating either single amino acid substitutions in FT or aberrant splicing in PDS3 transcripts, resulting in transgenic plants with delayed flowering or albino phenotypes (Table 1). In 2020, Wu et al. [47] applied CBE to edit BnALS1 and BnALS3 genes in rapeseed, introducing C-to-T substitutions at codon P197, thereby creating transgene-free homozygous rapeseed mutants resistant to bensulfuron (Table 1). In the same year, Qin et al. [48] developed a novel cotton base-editing system, GhBE3, based on CBE, and edited the GhCLA and GhPEBP genes, achieving efficiencies of 26.67% and 57.78%, respectively (Table 1). In 2022, Wang et al. [49] applied ABE for the first time in allotetraploid cotton, achieving A-to-G editing efficiencies up to 64% at genomic target sites. They further developed a new editor, GhABE7.10dCpf1 (7.10TadA+dCpf1), which recognizes T-rich PAMs and successfully edited the GhPEBP gene to introduce amino acid substitutions that generated a compact plant architecture, serving as a prototype for mechanized cotton harvesting (Table 1). In 2020, Li et al. [50] used CBE to edit two non-allelic acetolactate synthase genes in maize, ZmALS1 and ZmALS2, with in vivo C-to-T efficiencies up to 13.9%, successfully generating transgene-free homozygous ZmALS1 mutants or double mutants conferring resistance to chlorsulfuron (Table 1). More recently, Fu et al. (2025) [51] tested optimized cytosine and adenine base editors (evoAPOBEC1, evoFERNY, evoCDA1, TadA8.20, TadA8e) in maize, editing ZmACC1 and ZmACC2 to generate herbicide-resistant materials. Among them, ZmACC2 single mutants showed greater resistance than ZmACC1 single mutants, while homozygous double mutants appeared to be lethal (Table 1). In 2017, Shimatani et al. [52] applied a codon-optimized PmCDA1 system in Arabidopsis to edit endogenous DELLA and ETR1 genes in tomato, obtaining homozygous DELLA mutants with reduced serrated leaflet formation (Table 1). In 2019, Veillet et al. [53] used a CBE system to edit SlALS in tomato and StALS in potato, successfully generating chlorsulfuron-resistant plants, with tomato editing efficiencies reaching as high as 71% (Table 1). In 2018, Tian et al. [54] applied a modified BE3 system in watermelon to edit the ALS gene, successfully creating herbicide-resistant germplasm resources (Table 1).

Compared with traditional genome editing technologies such as CRISPR/Cas9, base editing technology demonstrates significant advantages, including high target specificity, precise and efficient editing, and a low frequency of off-target effects [55]. In recent years, with the continuous optimization of research systems, this technology has achieved remarkable progress, but it still has certain limitations. For example, in practical applications, it is restricted by multiple factors such as editing efficiency, PAM recognition range, editing window width, types of base conversion, and off-target effects. Systematic optimization of these factors remains the key to advancing the application of base editing technology in crop breeding.

As the third generation of genome editing technology, base editing (BE) integrates deaminases with the CRISPR/Cas system to achieve targeted modification of DNA or RNA bases. With advantages such as high precision, efficiency, and the ability to operate without introducing double-strand breaks (DSBs), BE has shown great promise in crop genetic improvement. By precisely modifying herbicide-resistance genes (e.g., OsALS), disease/stress-resistance genes, and quality-related genes, researchers have successfully generated novel germplasm with enhanced herbicide resistance, stronger stress tolerance, and improved quality. Although prime editing (PE) has advanced rapidly and allows for arbitrary base conversions and more complex genome rewriting, it is important to note that regulatory policies in countries such as the United States, Japan, Argentina, and China generally classify gene-edited plants lacking exogenous DNA—where only point mutations or small fragment deletions occur—as non-transgenic, or simplify their safety evaluation process [95]. This means that base editing, which only introduces point mutations, faces fewer regulatory restrictions and is more readily accepted and applied. In contrast, PE applications involving large-fragment insertions remain subject to stricter transgenic-related regulations. Therefore, base editing still holds significant practical value in plants. Its future development may focus on the following directions: (1) Optimization of editor performance: This involves further engineering of deaminases and Cas proteins through protein engineering and directed evolution to enhance efficiency, expand the editing window, and minimize bystander editing. (2) Expansion of PAM compatibility: This involves the development of Cas variants with relaxed PAM recognition to enlarge the editable genomic scope. (3) Application and diversification of editors: This involves promoting the application of existing editors such as GBE and AKBE, while developing novel base editors to enrich the spectrum of achievable base conversions. (4) Exploration of novel editing platforms: This involves establishing highly efficient Cas12-based base editing systems and other innovative approaches. (5) Integration with Artificial Intelligence (AI): In recent years, the field of AI has advanced rapidly, demonstrating revolutionary potential in areas such as medicine and agriculture. Deep integration of AI with base editing technology can accelerate its development toward greater speed and precision. For instance, Wei et al. recently employed the AI model ProMEP to perform a series of protein engineering efforts on Cas9, successfully designing a high-performance and highly versatile Cas9 variant (Cas9-AI-8.3). This variant enhances the efficiency of multiple mainstream base editors by an average of 2- to 3-fold [96]. First, AI enables precise navigation and target prediction by deeply mining multi-omics data to accurately identify key single-nucleotide variations associated with crop quality, yield, and disease resistance. Second, AI-driven protein design platforms such as AlphaFold [97] and RFdiffusion [98] can rapidly and accurately resolve the three-dimensional structures of base editors. These platforms facilitate de novo design or optimization of components such as deaminases, Cas proteins, and linkers, thereby improving the performance of base editors. Finally, AI models trained on vast amounts of on-target and off-target editing data can learn to identify genomic sequence features that may lead to unintended edits. This allows for the prediction of off-target risks before the deployment of base editors, enhancing their safety profile. In summary, base editing technology is steadily maturing, with broad application prospects in crop breeding. It is expected to provide powerful technological support for global food security and sustainable agricultural development.