TnpB, a compact RNA-guided nuclease ancestral to Cas12, is optimized for plant genome editing. A 99-nt enhanced RNA and T5 exonuclease fusion generate TnpBe5, boosting rice editing 2.5-fold. Coupling with a single-strands annealing (SSA)-responsive HYYG surrogate (TnpBmax) enriches edited cells, achieving up to 81.5% efficiency and high homozygosity across rice and tomato.

Adenine and cytosine base editing using dCas-SF01 and the 35S-CmYLCV-U6 composite promoter successfully introduced targeted base substitutions at multiple loci in rice, with average editing efficiency of 33.3%. Using the protospacer adjacent motif (PAM)-relaxed SF01-IKRR variant enabled base editing using 5'-NTN PAMs in rice.

Transient gene expression in mesophyll protoplasts is a valuable approach for investigating gene function, plant physiological processes, and molecular mechanisms. Rubber dandelion (Taraxacum kok-saghyz, TKS) is an ideal model for studying rubber biosynthesis and serves as a promising source of natural rubber and inulin. However, developing efficient protoplast-based systems for TKS remains challenging. In this study, we established a robust method for isolating mesophyll protoplasts from TKS by optimizing enzymatic conditions for cell wall digestion. We subjected the protoplasts to PEG/calcium-mediated transfection and evaluated promoter activities, expressed and detected target proteins, confirmed the subcellular localization of the target proteins, and examined transcription factor-DNA interactions under physiological conditions. We also developed a rapid assessment strategy for genome-editing tools in TKS protoplasts using multiple reporter systems. We evaluated these optimized tools in a tissue-culture-free hairy root transformation system, establishing a dual-platform toolkit for functional genomics in TKS. This work provides an efficient approach for TKS protoplast preparation, facilitating studies of gene function and advancing biotechnological research in this rubber-producing crop.

Prime editing (PE) is limited by low efficiency in dicot plants. Here we develop an optimized PE system for soybean, GmPEplus, by deleting the RNase H domain, introducing a V223A substitution within the reverse transcriptase domain, inserting a viral nucleocapsid protein between Cas9 and reverse transcriptase, and co-expressing a dominant-negative engineered allele of the endogenous GmMLH1. GmPEplus achieves editing efficiencies of up to 81.3% in stable transgenic lines. Subsequently, we show that nicking the non-edited strand using an additional sgRNA via the tRNA processing system enhances editing efficiency, and that optimizing its expression with an independent AtU6 cassette boosts efficiency by 13.1-fold. Importantly, a Csy4-mediated multiplex PE system (CMMPE) is established for simultaneous editing of 2-12 genes in soybean hairy roots and up to 3 genes in stable transgenic lines. GmPEplus and CMMPE offer powerful, versatile tools for precise, multiplex and heritable genome editing in soybean breeding.

A newly developed rice guanine base editor (OsGTBE) achieves targeted and efficient G-to-T editing (C-to-A in the opposite strand) in rice. Using OsGTBE to edit endogenous herbicide-resistant loci generated several novel alleles conferring herbicide resistance, highlighting its utility in creating valuable germplasm and enhancing genetic diversity..

Succulents, valued for their drought tolerance and ornamental appeal, are important in the floriculture market. However, only a handful of succulent species can be genetically transformed, making it difficult to improve these plants through genetic engineering. In this study, we adapted the recently developed cut-dip-budding (CDB) gene delivery system to transform three previously recalcitrant succulent varieties - the dicotyledonous Kalanchoe blossfeldiana and Crassula arborescens and the monocotyledonous Sansevieria trifasciata. Capitalizing on the robust ability of cut leaves to regenerate shoots, these plants were successfully transformed by directly infecting cut leaf segments with the Agrobacterium rhizogenes strain K599. The transformation efficiencies were approximately 74%, 5% and 3.9%-7.8%, respectively, for K. blossfeldiana and C. arborescens and S. trifasciata. Using this modified CDB method to deliver the CRISPR/Cas9 construct, gene editing efficiency in K. blossfeldiana at the PDS locus was approximately 70%. Our findings suggest that succulents with shoot regeneration ability from cut leaves can be genetically transformed using the CDB method, thus opening up an avenue for genetic engineering of these plants.

Cytosine and adenosine base editors (CBE and ABE) have been widely used in plants, greatly accelerating gene function research and crop breeding. Current base editors can achieve efficient A-to-G and C-to-T/G/A editing. However, efficient and heritable A-to-Y (A-to-T/C) editing remains to be developed in plants. In this study, a series of A-to-K base editor (AKBE) systems were constructed for monocot and dicot plants. Furthermore, nSpCas9 was replaced with the PAM-less Cas9 variant (nSpRY) to expand the target range of the AKBEs. Analysis of 228 T0 rice plants and 121 T0 tomato plants edited using AKBEs at 18 endogenous loci revealed that, in addition to highly efficient A-to-G substitution (41.0% on average), the plant AKBEs can achieve A-to-T conversion with efficiencies of up to 25.9 and 10.5% in rice and tomato, respectively. Moreover, the rice-optimized AKBE generates A-to-C conversion in rice, with an average efficiency of 1.8%, revealing the significant value of plant-optimized AKBE in creating genetic diversity. Although most of the A-to-T and A-to-C edits were chimeric, desired editing types could be transmitted to the T1 offspring, similar to the edits generated by the traditional ABE8e. Besides, using AKBEs to target tyrosine (Y, TAT) or cysteine (C, TGT) achieved the introduction of an early stop codon (TAG/TAA/TGA) of target genes, demonstrating its potential use in gene disruption.

Precise sequence insertion or replacement in plants is technically challenging but is of great importance in crop breeding because many agronomic traits are affected by DNA fragment variations. Although prime editing (PE) has been continuously optimized to improve its activity in plants (Jiang et al., 2022; Li et al., 2022a,b; Zong et al., 2022), it is still inefficient for targeted insertion or replacement of longer sequences. Similar strategies, twinPE (Anzalone et al., 2022) and GRAND editing (Wang et al., 2022), in which a pair of PE guide RNAs (pegRNAs) are partially complementary to each other in their reverse transcriptase template (RTT) but are not homologous to the genomic sequences, were recently developed to facilitate longer sequence insertion (Figure 1a). HPPD-inhibitor herbicides such as β-triketones are effective in controlling resistant weeds that have emerged. The HIS1 gene in rice confers broad-spectrum resistance to triketone herbicides, whereas a dysfunctional his1 allele with a 28-bp fragment deletion was found in triketone-sensitive Indica varieties (Maeda et al., 2019). A genetic survey for 631 Indica varieties commonly used in rice breeding revealed that the 28-bp deletion is widely distributed, including 50.7% 3-line restorers, 40.7% 2-line restorers and 18.1% conventional varieties (Lv et al., 2021), which causes a huge risk for applying HPPD-inhibitor herbicides in Indica rice cultivating area. S1035 is an elite conventional Indica cultivar that sensitive to triketone due to the 28-bp deletion at HIS1. PE and GRAND editing strategies were tested to targeted insert the 28-bp fragment. Different from the design of PE (Figure S1), GRAND editing uses a pair of pegRNAs to delete the 18-bp genomic sequence between the two nicks and to insert a 46-bp designed sequence, including the to-be-inserted 28-bp and the to-be-replaced 18-bp sequences in which synonymous mutations were introduced to reduce the homology between RTT and genomic sequences (Figure 1b). It was reported that the sequence complementarity within the RTTs significantly affects the insertion efficiency (Wang et al., 2022), 10-, 18- or 26-bp sequence complementarity was designed in our test (Figure 1b). PE yielded 1.46% precise insertion events after protoplast transfection, indicating its low efficiency for DNA fragment insertion. GRAND editing with 10-bp complementary RTT sequences (RTT-10) achieved higher efficiency (9.88%) than that of RTT-18 (3.76%) or RTT-26 (0.59%) (Figure 1c). We then further evaluated GRAND editing during stable transformation. The transgenic lines were directly treated by 60 μm mesotrione, a widely used β-triketone herbicide. Nine (11.5%) resistant lines were generated from RTT-10 transformation, while only one was obtained each from RTT-18 and RTT-26 transformations (Figure 1d; Figures S2 and S3). Then RTT-10 construct was used to edit MingHui86, an elite 3-line restorer with the 28-bp deletion at HIS1. 13 (15.5%) of the 84 transgenic lines were recovered mesotrione-resistance (Figure 1d and Figure S4). T-DNA free, homozygous offsprings were identified from S1035 edited lines in the T1 generation (Figure 1e and Figure S5, Table S1). The expression level of the repaired HIS1 gene was comparable with that of wild type (WT; Figure 1f), but the mesotrione-resistance of the edited plants was similar with that of XiuShui134, a functional HIS1 gene-containing Japonica rice variety (Figure 1g). These results indicated that GRAND editing can be used to rescue other defective varieties for quickly solving the risk of applying HPPD-inhibitor herbicides in Indica rice cultivating area. PE and GRAND editing strategies were also tested for introducing the glyphosate-resistant T173IP177S mutations (C518T + C529T) into the OsEPSPS gene (Figure 1h; Figure S6). Protoplast test showed the efficiency of GRAND editing was 1.86 fold of that of PE (9.98% vs. 5.37%; Figure S7). Only 3 (2.0%) heterozygous plants from PE contained desired TIPS mutations, other mutants were either chimeras or that the C518T and C529T substitutions occurred separately (Figures S6 and S8). In contrast, GRAND editing generated 5 (3.4%) homozygous and 18 (12.2%) heterozygous edited lines (Figure 1i; Figure S8). Since the plants carrying homozygous TIPS mutations were seriously affected in their growth or even died after transplanting to soil (Figure S9), the heterozygous TIPS mutants were tested by glyphosate and showed no symptoms of damage but the WT plants withered (Figure S10). Their offsprings in the T1 generation inherited the glyphosate-resistant trait (Figure 1j,k; Table S1). To further evaluate the activity of GRAND editing, we designed to replace a 28-bp fragment (containing three amino acid substitutions) in the TVHYNP domain of the OsSLR1 gene (Figure 1l). We obtained 4 (6.9%) homozygous, 19 (32.8%) heterozygous plants from 58 transgenic lines (Figure 1m). As expected, both homozygous and heterozygous mutants displayed a dwarf phenotype (Figure 1n,o). During preparation and review of our manuscript, Li et al. (2023) reported the use of GRAND editing in knock-in of protein tags. In combination with site-specific recombinases, large DNA fragments of up to 11.1 kb were targeted inserted in the rice genome (Sun et al., 2023). Collectively, ours and these results demonstrated the technical advancement for insertion or replacement of DNA fragments in the plant genome, which is of great importance in genetic research and crop breeding. This work was supported by Bellagen Biotechnology Co. Ltd., Nanfan special project, CAAS (ZDXM04, ZDXM23014), National Key R&D Program of China (2021YFA1300404 to J.-K.Z.), National Natural Science Foundation of China (32188102 to J.-K.Z., 32271524 and 31901046 to M.W.), the Youth Innovation Promotion Association, CAS (2020272) to M.W., the Key R&D Program of Ningxia (2021BEB04075) and Ningxia Natural Science Foundation (2022AAC03007) to X.L. The authors declare no competing interests. J.-K.Z. and M.W. conceived of and designed the research. X.L., Y.W., H.W., Y.H., Y.S., Z.L., M.L., C.W., Y.D. L.X. and J.Z. conducted the experiments and analysed the data. J.-K.Z. and M.W. wrote the manuscript. Figures S1-S10 Supplementary Figures Tables S1-S2 Supplementary Tables Data S1 Materials and methods. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.

Summary The CRISPR/Cas type V‐I is a family of programmable nuclease systems that prefers a T‐rich protospacer adjacent motif (PAM) and is guided by a short crRNA. In this study, the genome‐editing application of Cas12i3, a type V‐I family endonuclease, was characterized in rice. We developed a CRIPSR/Cas12i3‐based Multiplex direct repeats (DR)‐spacer Array Genome Editing (iMAGE) system that was efficient in editing various genes in rice. Interestingly, iMAGE produced chromosomal structural variations with a higher frequency than CRISPR/Cas9. In addition, we developed base editors using deactivated Cas12i3 and generated herbicide‐resistant rice plants using the base editors. These CRIPSR/Cas12i3‐based genome editing systems will facilitate precision molecular breeding in plants.

Of the more than 370 000 species of higher plants in nature, fewer than 0.1% can be genetically modified due to limitations of the current gene delivery systems. Even for those that can be genetically modified, the modification involves a tedious and costly tissue culture process. Here, we describe an extremely simple cut-dip-budding (CDB) delivery system, which uses Agrobacterium rhizogene to inoculate explants, generating transformed roots that produce transformed buds due to root suckering. We have successfully used CDB to achieve the heritable transformation of plant species in multiple plant families, including two herbaceous plants (Taraxacum kok-saghyz and Coronilla varia), a tuberous root plant (sweet potato), and three woody plant species (Ailanthus altissima, Aralia elata, and Clerodendrum chinense). These plants have previously been difficult or impossible to transform, but the CDB method enabled efficient transformation or gene editing in them using a very simple explant dipping protocol, under non-sterile conditions and without the need for tissue culture. Our work suggests that large numbers of plants could be amenable to genetic modifications using the CDB method.

Improved cytosine and adenine base editors and an efficient dual editor were applied in targeted evolution of ACETYL COA CARBOXYLASE in rice, resulting in the generation of dozens of herbicide-resistant mutations, at least three of which, W2125L, W2125Q and C2186H, have not been reported previously.

A new deaminase, TadA8e, was recently evolved in the laboratory. TadA8e catalyzes DNA deamination over 1,000 times faster than ABE7.10.We developed a high-efficiency adenine base editor, rABE8e(rice ABE8e), combining monomeric TadA8e, bis-bpNLS and codon optimization. rABE8e had substantially increased editing efficiencies at NG-protospacer adjacent motif(PAM) and NGG-PAM target sequences compared with ABEmax. For most targets,rABE8e exhibited nearly 100% editing efficiency and high homozygous substitution rates in the specific editing window, especially at Positions A5 and A6. The ability to rapidly generate plant materials with homozygous base substitutions will benefit gene function research and precision molecular breeding.

Bacterial-derived CRISPR/Cas systems are versatile platforms to engineer site-specific gene editing tools. Compared to the canonical Cas9-mediated DNA cleavage systems, which induce a high-proportion of frame-shift mutations, the recently developed base editing (BE) tools allow more precise and predictable base substitutions within a CRISPR/Cas9-defined editing window. Initially, such tools made use of engineered cytosine deaminases or evolved adenine deaminases to catalyse base deamination when fused to a Cas9 nickase (nCas9) (Rees and Liu, 2018). The deamination of cytosines and adenines results in the conversion of C·G and A·T pairing to T·A and G·C respectively. Since this method facilitates precise gene editing in a repair template-independent manner, it was soon applied to a wide range of species, including plants (Mao et al., 2019). However, the narrow target range and low editing activity still limit the application of BEs to gene function studies and crop breeding. As a solution, engineered SpCas9 nickase variants with relaxed PAM preferences, such as Cas9-NG and SpRY, were used to produce newer versions of BE tools (Nishimasu et al., 2018; Walton et al., 2020). Although this strategy helped to expand the target range of BE systems, efficiency losses were usually observed (Zhong et al., 2019). To overcome this limitation, two previously developed rice-optimized BE systems (based on nCas9-NG fused to Anc689BE4max and ABEmax respectively) were further adapted for use in Arabidopsis (Koblan et al., 2018; Wang et al., 2019). The four recombinant base editors, CBE/ABEmax-nCas9 and CBE/ABEmax-nCas9NG, were expressed using an Arabidopsis-optimized vector under the control of the RPS5a promoter and NOS terminator. Previous studies in Arabidopsis have shown that CRISPR/Cas9 gene editing efficiencies can be significantly improved by suppressing the plant RNA silencing pathways (Mao et al., 2018). To minimize the silencing effects, we fused a commonly used silencing suppressor (p19) to the above BE systems, generating the CBE/ABEmax-nCas9-p19 and CBE/ABEmax-nCas9NG-p19 systems (Figure 1a). We first investigated the effect of co-expressing p19 in the two CBEmax systems by targeting a site in the Arabidopsis Acetolactate Synthase (ALS) gene with a canonical NGG PAM. The chosen target, AtALS-599, contains three successive cytosines (C5, C6 and C7) within the editing window (Figure 1b). Transgenic lines were produced by the Agrobacterium-mediated floral dipping method. At least 20 10-day-old seedlings were bulked as a single biological replicate for DNA extraction and SNP detection using next-generation sequencing (NGS). For each construct, three biological replicates were analysed. Compared to the nCas9 control, the editing efficiencies of nCas9NG at the three cytosines were decreased by about eight-folds. However, co-expression of p19 substantially increased the editing efficiencies of the nCas9NG system to the nCas9 levels. Interestingly, co-expression of p19 only had a marginal enhancing effect on the activity of the canonical nCas9-based BE system (Figure 1b). To study whether the beneficial effect of p19 extends to other targets as well as different BE systems, a second site in the ALS gene was targeted for editing. The AtALS-1700 site contains two cytosines (C6, C7) and two adenines (A4, A5) located within the editing window (Figure 1c,d). Analysis of editing efficiency in transgenic Arabidopsis lines again showed that the non-canonical nCas9NG systems were 5- to 10-fold less efficient than the wild-type nCas9 systems for both cytosine and adenine base editing. Nevertheless, as observed above, the lower nCas9NG efficiencies can be restored to nCas9 levels by p19 co-expression (Figure 1c,d). To evaluate the efficiency of the relaxed PAM BE systems at targets with NG PAMs, four overlapping targets within the Arabidopsis AtTIC236 gene (AT2G25660) were selected for editing with the CBEmax-nCas9NG system using four different sgRNAs (sgR1 to sgR4) (Figure 1e). Based on the NGS data of bulked T1 plants, none of the cytosines within the editing window exhibited an editing frequency over 10% on average from the three replicates (Figure 1h,i). Co-expression of p19 improved efficiencies overall, although the effect was heavily dependent on the guide RNA and the nucleotide position within the editing window. In some cases, spectacular efficiency improvements were observed, with the editing frequency of C14 increasing from 2.1% to 11.9% and from 8.7% to 41.4% by using the guides sgR1 and sgR2 respectively. In other cases, such as for C6 and C8, editing frequencies were increased by 15–20 folds, but the absolute values were still very low (<10%) (Figure 1h,i). Although p19 had a positive effect on BE efficiencies, it also induced strong developmental phenotypes when highly expressed in plants. Counter selection for T-DNA free BE plants in subsequent generations is necessary for removing this side effect (Mao et al., 2018). We queried whether the efficiency of non-canonical BE systems could be improved using an alternative strategy, without causing severe developmental phenotypes. We targeted the same AtTIC236 gene sites in the Arabidopsis dicer-like (DCL) 2/3/4 triple mutant background using the CBEmax-nCas9NG system (Henderson et al., 2006). We observed a good correlation between the results in the dcl234 mutants and those observed using the p19 co-expression strategy. Nucleotides showing efficiency improvements by p19 co-expression in WT plants also showed improvements in the dcl234 mutant without p19 co-expression. Overall, the use of dcl234 mutant was more efficient than p19 co-expression (Figure 1h,i). Our results suggest that the efficiencies of non-canonical BE systems are constrained at the post-transcriptional level. Targeting four sites within the AtCRE1 gene (AT2G01830) with NG PAMs using the CBE/ABEmax systems supported our observations above (Figure 1f,g). In this case, BE efficiencies were extremely low with only one cytosine and one adenine showing editing frequencies above 1% (C14 and A13) for a single sgRNA (sgR1) (Figure 1j–m). As observed above, suppression of the plant RNA-silencing pathways by either co-expression of p19 or the use of dcl234 mutant increased BE efficiencies in some instances, but not always. The practical threshold of 10% was never reached by co-expression of p19, but it was achieved on one occasion (C14 with the sgR1 guide) using the CBEMax-nCas9NG system in the dcl234 mutant and almost reached for one adenine nucleotide (A13 with the sgR1 guide) using the ABEMax-nCas9NG system in the dcl234 mutant (Figure 1j–m). To determine whether the relaxed PAM systems can generate heritable BE events, we selected two dcl234 T1 populations transformed with CBEmax-nCas9NG construct for further analysis. The C/T editing frequency of the TIC236 C14 site using sgR1 and sgR2 was evaluated by Hi-Tom sequencing. Forty-eight individual T1 seedlings were sequenced for each population. Four of the T1 lines for sgR1 and 23 lines for sgR2 exhibited editing efficiencies above 50% in the sampled leaves (Figure 1o). In the T2 generation of two of them (sgR1-#6 and sgR2-#48), heritable C/T conversion events were validated by Sanger sequencing with three different editing outcomes observed, suggesting stable inheritance of BE events (Figure 1n). Our results suggest that the efficiency loss observed in non-canonical BE systems could be due to suboptimal expression of the engineered base editors. Increasing expression levels using strong promoters can overcome this limitation as shown in a previous study, demonstrating that the use of the AtRPS5a promoter enables ABE7.10-nCas9-mediated base editing in Arabidopsis (Kim, 2018). An alternative strategy is to reduce RNA degradation using silencing suppressors as shown in this work where we demonstrated that relaxed PAM base editing can be achieved in Arabidopsis seedlings with acceptable efficiencies. We thank the bioinformatics facility of PSC for NGS data analysis. This work was supported by the Strategic Priority Research Program, CAS (Grant No. XDA26040104), the National Natural Science Foundation of China (Grant No. 31870350) and the Youth Innovation Promotion Association, CAS (Grant No.2017004, No.2020272). The authors declare no competing financial interests. Y.M. and M.W. designed the experiments; Y.M., Y.Z, Q.W. and B.H. performed the experiments; Q.Z. analysed the data; Y.M. wrote the manuscript and J.B. and J.-K.Z. edited the manuscript; J.-K.Z. supervised the project.

Maintenance of cell wall integrity is of great importance not only for plant growth and development, but also for the adaptation of plants to adverse environments. However, how the cell wall integrity is modulated under salt stress is still poorly understood. Here, we report that a nuclear-localized Agenet domain-containing protein SWO1 (SWOLLEN 1) is required for the maintenance of cell wall integrity in Arabidopsis under salt stress. Mutation in SWO1 gene results in swollen root tips, disordered root cell morphology, and root elongation inhibition under salt stress. The swo1 mutant accumulates less cellulose and pectin but more lignin under high salinity. RNA-seq and ChIP-seq assays reveal that SWO1 binds to the promoter of several cell wall-related genes and regulates their expression under saline conditions. Further study indicates that SWO1 interacts with importin ɑ IMPA1 and IMPA2, which are required for the import of nuclear-localized proteins. The impa1 impa2 double mutant also exhibits root growth inhibition under salt stress and mutations of these two genes aggravate the salt-hypersensitive phenotype of the swo1 mutant. Taken together, our data suggest that SWO1 functions together with importin ɑ to regulate the expression of cell wall-related genes, which enables plants to maintain cell wall integrity under high salinity.

CRISPR–Cas9 methods have been applied to generate random insertions and deletions, large deletions, targeted insertions or replacements of short sequences, and precise base changes in plants 1 – 7 . However, versatile methods for targeted insertion or replacement of long sequences and genes, which are needed for functional genomics studies and trait improvement in crops, are few and largely depend on the use of selection markers 8 – 11 . Building on methods developed in mammalian cells 12 , we used chemically modified donor DNA and CRISPR–Cas9 to insert sequences of up to 2,049 base pairs (bp), including enhancers and promoters, into the rice genome at an efficiency of 25%. We also report a method for gene replacement that relies on homology-directed repair, chemically modified donor DNA and the presence of tandem repeats at target sites, achieving replacement with up to 130-bp sequences at 6.1% efficiency.

Base editors, presently including cytidine base editors (CBEs) and adenine base editors (ABEs), enable precise base alterations in the genome without inducing DNA double-stranded breaks (DSBs). Base editors are valuable tools for precision plant molecular breeding since many agronomic traits are controlled by variations in one or few DNA bases. The early developed CBE and ABE systems, consisting of the rat cytidine deaminase APOBEC1 (rAPOBEC1) or activation-induced cytidine deaminase (AID) PmCDA1, and the evolved tRNA adenine deaminase TadA, respectively, have been applied to many plant species. To improve the base editing efficiency, more effective cytidine deaminases such as the human APOBEC3A have been tested (Zong et al., 2018). On the other hand, for expanding the base editing scope in plants, several SpCas9 and SaCas9 variants such as VQR-Cas9, VRER-Cas9 and SaKKH-Cas9 that recognize PAMs other than the canonical NGG motif were introduced into the CBE and ABE toolbox (Hua et al., 2018a; Qin et al., 2018). However, relative to the widely used CRISPR/Cas gene editing technologies for inducing DSBs and subsequent repair-caused mutations, the efficiency of base editing is still low. In addition, base editors reported thus far are constrained by recognition of only a few kinds of PAM sequences. We have previously reported the initial adoption of CBEs and ABEs in rice (Hua et al., 2018b; Lu and Zhu, 2017). In our CBE system, we fused rAPOBEC1 to the N-terminus of SpCas9 nickase (Cas9n, D10A) using the unstructured 16-residue peptide XTEN as linker. A traditional nuclear localization signal, SV40 NLS peptide, was added to the C-terminus of the Cas9n. Two agronomically important genes of rice, NRT1.1B and SLR1, were selected for editing by this CBE system. However, the base substitution efficiencies were low, with only 2.7% for NRT1.1B and 13.3% for SLR1, respectively (Lu and Zhu, 2017). In our ABE system, we synthesized wild-type ecTadA and its mutant form ecTadA*7.10 and linked them together using a 32-amino acid (aa) linker; the resulting recombinant protein was fused to the N-terminus of the SpCas9 or SaCas9 nickase with the same linker. Testing at different targets showed that the base substitution efficiencies ranged from 5% to 60%, with most of the target sites having efficiencies lower than 30% as reported by other groups (Hua et al., 2018b). Recently, Koblan et al. (2018) found that the expression levels of base editors are major bottlenecks for base editing efficiency. They improved BE4 and ABE7.10 base editors by adopting bipartite nuclear localization signals (bpNLS), optimizing codon usage and ancestral reconstruction of the deaminase component. The resulting BE4max, AncBE4max and ABEmax editors showed increased editing efficiencies in a variety of settings, especially under suboptimal conditions or at sites previously edited with low efficiencies (Koblan et al., 2018). To improve the base editing efficiency in plants, we directly adopted the above GenScript codon-optimized nucleotide sequences of bpNLS-Anc689 APOBEC-32 aa Linker and bpNLS-adenine deaminase of ABE7.10-32 aa Linker into our previous CBE and ABE editors, resulting in Anc689BE4max and ABEmax, respectively (Figure 1a,b). To directly compare the performance of Anc689BE4max with our previous CBE, the NRT1.1B and SLR1 were selected for editing using the previously tested sgRNA. As shown in Figure 1c, 72.4% of the transgenic rice lines harboured the target C to T replacement at NRT1.1B target site, and 76.2% of these lines (55.2% of total transgenic lines) are homozygous (Figure 1d). Most of the regenerated plantlets transformed with Anc689BE4max-sgRNASLR1 displayed an obvious dwarf phenotype (Figure 1e,f). Genotyping and sequencing results showed that 82.8% of the transgenic lines converted C to T at their target site, and 72.5% of the transgenic lines were homozygous or biallelic C to T substitutions (Figure 1g,h). The sequencing results are also consistent with the phenotype of each plantlet (Figure 1f,i). Compared with the efficiency of 2.7% for NRT1.1B and 13.3% for SLR1 from our previous CBE (Lu and Zhu, 2017), the Anc689BE4max showed much higher base editing efficiencies. We also noticed that the deamination window ranged from the 4th to 15th target bases, but the substitutions were concentrated at the 4th to 7th bases, which is similar to our previously reported CBE (Lu and Zhu, 2017). To further evaluate the efficiency of Anc689BE4max, we designed an sgRNA (ALS-sg1) for modifying the acetolactate synthase gene (ALS) in rice. It is known that a mutated form of ALS, ALSS627N (G1880 to A in Nipponbare DNA sequence), confers tolerance to imidazolinone herbicides (Piao et al., 2018) (Figure 1j). Similar to the results from the base editing of NRT1.1B and SLR1, 71.4% of the transgenic lines contained C to T substitution at their target site in ALS, and most of them were homozygous or biallelic (Figure 1k,l). Our result also shows that although the target C ranging from 4th to 10th of the protospacer could be replaced by T, the substitution preferentially occurred within the window from the 5th to 7th base. Only two of the edited lines contained the intended G1880 to A conversion, since this target base is located outside of the 'hot spot' of the deamination window (Figure 1k). Recently, Nishimasu et al. (2018) reported that a rationally engineered SpCas9 variant, SpCas9-NG, containing the R1335A/L1111R/D1135V/G1218R/E1219F/A1322R/T1337R seven amino acid alteration, can recognize relaxed NG PAMs in human cells. During the preparation of our manuscript, Endo et al. (2018) reported that the nickase of this variant (D10A) fused to cytidine deaminases such as PmCDA1 and rAPOBEC1 could mediate C to T conversion at sites bearing NG PAMs in rice calli, but the nSpCas9NG-APOBEC1 base editor showed a low activity at most of the tested target sites. To expand the access range of base editors, we adopted the SpCas9NG nickase into our Anc689BE4max and ABEmax to replace the SpCas9 nickase, resulting in Anc689BE4max-nCas9NG and ABEmax-nCas9NG system, respectively (Figure 1a,b). Testing of the Anc689BE4max-nCas9NG system at SLR1 using the same sgRNA with GGG PAM showed an editing efficiency of 40.9%, lower than that of Anc689BE4max-nCas9 (Figure 1s). To facilitate the G1880 to A substitution in ALS, we designed another sgRNA, ALS-sg2, harbouring the AGC PAM, for editing by the Anc689BE4max-nCas9NG system (Figure 1j). 57.1% of the transgenic lines showed the intended G1880 to A replacement, although the combined base substitution efficiency was lower than that of Anc689BE4max-nCas9 with ALS-sg1 (Figure 1m,s). The mutants homozygous for the G1880 to A substitution were tolerant to imidazolinone herbicide, whereas wild-type plants were not (Figure 1n,o). Taking advantage of the nCas9NG-derived base editor that can recognize relaxed PAMs, we further applied it to modify the 5-enolpyruvylshikimate-3-phosphate synthase (EPSPS) gene. Previous work showed that a single base transition of C317-T within OsEPSPS (C530 in Nipponbare genome), changing proline-106 to leucine (P106L), led to resistance to another herbicide, glyphosate, in the EPSPS-deficient Escherichia coli strain AB2829 (Zhou et al., 2006). Here, we designed an sgRNA harbouring the TGA PAM (EPSPS-sg1) for editing by our Anc689BE4max-nCas9NG system, since there are no suitable NGG PAMs near the protospacer (Figure 1p). We successfully obtained five plantlets containing the targeted C530 to T replacement from 29 transgenic lines (Figure 1p,q,s). The glyphosate tolerance will be tested in the T1 generation since there was no homozygous mutant in the T0 plants. To evaluate the activity of our ABEmax editor, we designed a third sgRNA (ALS-sg3) to edit the ALS gene by ABEmax and our previous ABE side by side. The results showed that ABEmax doubled the editing efficiency of ABE (Figure 1r,t). We further tested ABEmax at the ALS-sg1 target site, and the results showed that 48.3% of the transgenic lines harboured A to G substitution (Figure 1t). The general editing efficiencies of ABEmax seem lower than those of Anc689BE4max (Figure 1s,t). We further evaluated the ABEmax-nCas9NG system with non-canonical NGG PAMs. Testing at the EPSPS-sg2 target site harbouring the AGT PAM showed an editing efficiency of 41.2%. However, testing at another target site (ALS-sg4) bearing the CGT PAM showed the editing efficiency lower than 10% (Figure 1t). In summary, our upgraded base editors not only show substantially increased editing efficiencies, but also have expanded editing scopes compared to previously reported CBEs and ABEs. These improved base editors are more powerful tools for molecular breeding of crops, although more plant species and more target sites need to be tested in the future. This work was supported by the Chinese Academy of Sciences. The authors declare no conflicts of interest with respect to this work.