The evolution and practical breeding of crops depend on genetic variations. Conventionally, these variations result from natural mutations or physical and chemical mutagenesis, both of which occur randomly and lack direction. To accelerate crop improvement, targeted mutagenesis methods are highly desired. Clustered regularly interspaced palindromic repeat (CRISPR)‐Cas9 systems have been engineered for genome‐targeted mutagenesis in eukaryotic organisms, including plants (Chen et al. , 2019). The CRISPR‐Cas9 system produces DNA double‐strand breaks (DSBs) by site‐specific cleavage, which typically results in small indels. To efficiently induce point mutations, CRISPR‐mediated base editors (BEs) have been further developed to generate targeted nucleotide substitution within editing windows. Combined with an sgRNA library, the BE platform has the potential to generate a large number of point mutations to screen pivotal amino acids (AAs) and to drive directed evolution of proteins in vivo (see review in Packer and Liu, 2015).

Acetyl‐CoA carboxylase (ACCase) catalyses the first step of fatty acid biosynthesis. Loss‐of‐function mutations in ACCase are lethal or lead to serious developmental arrest in plants (Baud et al. , 2004). In field production, ACCase is the target of a major group of commercial herbicides. All known ACCase‐inhibiting herbicide‐resistant mutations occur in the carboxyltransferase (CT) domain of ACCase, which directly interacts with the herbicide (Jang et al. , 2013). In rice, the OsACC gene (LOC_Os05g22940 ) has 35 exons and encodes a 2327‐AA protein, in which amino acids that determine herbicide resistance are largely unclear. To exploit dominant mutations endowing resistance in OsACC, 141 sgRNAs were designed in the 1653‐bp coding sequence of the CT domain located in the 34th exon for the base‐editing screen (Figure 1a). To conduct C∙G to T∙A substitutions, the eBE3 and eCDA systems, which previously had been optimized by stacking multiple copies of uracil DNA glycosylase inhibitor (UGI) (Qin et al. , 2019), were used to construct the domain‐specific libraries separately. To achieve efficient A∙T‐to‐G∙C substitution, the previously reported plant adenine BE (ABE) was optimized by fusing triple copies of the nuclear localization sequence to develop an enhanced ABE (eABE) for constructing a library (Li et al. , 2019; Li et al. , 2018). For each sgRNA, an oligo pair of forward and reverse 20‐bp guide sequences flanked with a 4‐bp ligation adapter was synthesized and annealed following the instructions of the pHUN411 vector (Xing et al. , 2014). Then, the annealed oligos were pooled and ligated into the BE vectors to replace the spectinomycin resistance gene between two Bsa I sites. Approximately 1500 kanamycin‐positive and spectinomycin‐negative clones were pooled in each library. To evaluate the library quality, 96 clones of each vector were randomly selected and Sanger sequenced. As indicated in Figure 1b, more than 93.8% of clones carried correct sgRNA, indicating the high accuracy of the plasmid library. At least 64 unique sgRNAs were identified from 96 clones, and most sgRNAs were detected no more than twice. These results indicate the good evenness of sgRNA distribution of the libraries. Then, the plasmids were transformed into the Agrobacterium EHA105 strain, and ~3000 clones were pooled per library. A random sequencing assay indicated that the sgRNAs in Agrobacterium had similar accuracy and uniformity to those in E. coli (Figure 1b).

The three base‐editing libraries were stably transformed into an elite rice variety (Oryza sativa L. japonica . cv. Feigeng2020). To screen herbicide‐resistant events, an aryloxyphenoxypropionate (APP) group ACCase‐inhibited chemical, haloxyfop‐R‐methyl, was added to the regeneration medium. Approximately 5000 hygromycin‐positive calli of each library were screened under the herbicide pressure of two concentrations (1 and 2 μM), both of which completely inhibited the growth of control transformants (Figure 1c). After 4–8 weeks of selection, 15 independent shoots were recovered from the calli transformed with the eABE library. In addition, 10 independent events of eABE calli that exhibited herbicide resistance but failed in regenerating plants were also selected. Genomic DNA was isolated from resistant plants and calli to identify sgRNA by PCR sequencing. We found that 23 out of 25 resistance events, including 14 out of 15 plants and nine out of 10 calli, were transformed with sgRNA‐127. In the sgRNA‐127‐targeted genomic region, all these events carried monoallelic or biallelic T₇‐to‐C conversion, leading to a C2186R mutation in the resistant cells (Figure 1d). sgRNA‐128 was observed in one resistant callus, which carried monoallelic T₁₃‐to‐C conversion in its spacer region, leading to the same C2186R mutation. sgRNA‐10 was detected in one regenerated plant, which resulted in a monoallelic A₁₆‐to‐G conversion and led to the I1879V mutation in OsACC. The remaining region of the CT domain in the resistant events was further screened by Sanger sequencing, and no additional mutation was found. Resistant plants and calli were also screened from the eBE3 library. Two plants and one callus, all of which harboured sgRNA‐121, were selected by haloxyfop‐R‐methyl. A monoallelic G₁‐to‐C substitution was detected in all three resistance events, resulting in a W2125S mutation. No resistant event was detected from the eCDA library.

A previous study showed that the ABE‐induced C2186R substitution in OsACC conferred herbicide resistance to rice in medium (Li et al. , 2018). Here, two additional single mutations, I1879V and W2125S, were produced by CRISPR‐mediated base screening to confer APP herbicide resistance in rice. Notably, the I1879V and C2186R mutations in rice correspond to the I1781V and C2088R resistance‐endowing mutations in weeds. However, the W2125S mutation is different from the conserved Cys substitution on the corresponding 2027 Trp (W2125C) in weeds, providing a new resistance allele. After transfer to greenhouse, serious growth retardation and complete sterility were observed in W2125S and homozygous C2186R plants. Furthermore, no homozygous mutant was detected in the progeny of heterozygous C2186R plants, suggesting that the homozygous C2186R mutation might be lethal to rice embryos. The growth and reproduction retardation suggest the fitness cost of these mutations. In the T₂ generation, T‐DNA‐free homozygous I1879V plants and heterozygous C2186R plants were isolated and sprayed with a commercial APP herbicide Gallant to confirm resistance and evaluate the potential usage of mutations in fields. Wild‐type (WT) plants died after 7 days, while all mutants demonstrated significant resistance (Figure 1e). Compared with WT, no obvious differences were detected in T₁ homozygous I1879V plants in traits of height, flag leaf length and setting rate (Figure 1f). Several reports also indicated that the I1781V mutation of ACCase had negligible fitness costs or even fitness advantages in various weeds (Jang et al. , 2013). Together with our observations, the I1879V mutation may be an ideal OsACC allele to acquire herbicide resistance when breeding commercial rice varieties.

Efficient nucleotide conversion can be achieved in the editing window of different BEs, such as positions 4‐8, 3‐9 and 2‐4 of eABE, eBE3 and eCDA, respectively (Li et al. , 2018; Shimatani et al. , 2017; Zong et al. , 2017). Accordingly, in the 1653‐bp targeted sequence, the 611‐nt (including 167 editable As and 162 editable Ts), 788‐nt (including 165 editable Cs and 192 editable Gs) and 393‐nt (including 84 editable Cs and 92 editable Gs) regions are covered by the 141 sgRNAs of the eABE, eBE3 and eCDA libraries, respectively. In grass weeds, substitutions at 7 different amino acids of the CT domain were implicated in naturally occurring resistance (Jang et al. , 2013). Among them, four mutations can be generated by A∙T//G∙C conversion in rice, whereas only the nucleotide corresponding to C2186R is located inside the window of the library. Although non‐typical substitutions, such as the G‐to‐C conversion for W2125S and the mutation at position 16 of sgRNA for I1879V, were revealed, the efficiency of these unconventional substitutions should be limited using the current tools. We believe that optimized BEs with improved editing efficiency, broadened PAM recognition, expanded editing windows and more flexible conversion types would greatly facilitate in planta CRISPR‐Cas9‐mediated base‐editing screens in the near future.

作物的进化和实际育种取决于遗传变异。传统上，这些变异是由自然突变或物理和化学诱变产生的，两者均随机发生且缺乏方向。为了加速作物改良，非常需要有针对性的诱变方法。聚簇的规则间隔的回文重复序列（CRISPR）-Cas9系统经过精心设计，可在包括植物在内的真核生物中进行基因组定向诱变（Chen等，2019）。CRISPR-Cas9系统通过位点特异性切割产生DNA双链断裂（DSB），通常导致小插入缺失。为了有效地诱导点突变，CRISPR介导的碱基编辑器（BEs）得以进一步开发，以在编辑窗口内产生靶向核苷酸取代。结合sgRNA文库，BE平台具有产生大量点突变以筛选关键氨基酸（AA）并驱动体内蛋白质定向进化的潜力（请参阅Packer和Liu的综述，2015年）。

乙酰辅酶A羧化酶（ACCase）催化脂肪酸生物合成的第一步。ACCase的功能丧失突变是致死性的或导致植物严重发育停滞（Baud等，2004）。在田间生产中，ACCase是一大类商业除草剂的目标。所有已知的抑制ACCase的除草剂抗性突变都发生在ACCase的羧基转移酶（CT）域中，该域与除草剂直接相互作用（Jang等，2013）。在水稻中，OsACC基因（LOC_Os05g22940）具有35个外显子，编码2327-AA蛋白，其中决定除草剂抗性的氨基酸尚不清楚。为了利用赋予OsACC抗性的显性突变，在第34外显子的CT域的1653-bp编码序列中设计了141个sgRNA，用于碱基编辑筛选（图1a）。为了进行C∙G到T∙A的取代，eBE3和eCDA系统之前已通过堆叠多份尿嘧啶DNA糖基化酶抑制剂（UGI）进行了优化（Qin等人，2019），分别用于构造特定于域的库。为了实现有效的A∙T到G∙C取代，先前报道的植物腺嘌呤BE（ABE）通过融合三倍核定位序列以开发增强型ABE（eABE）来构建文库进行了优化（Li等。，2019 ;李等人，2018）。对于每个sgRNA，按照pHUN411载体的指导合成并退火一个寡核苷酸对，该寡核苷酸对是一个正向和反向20 bp指导序列，两侧是一个4 bp的连接衔接子，并进行了退火（Xing等，2014）。然后，将退火的寡核苷酸汇集并连接到BE载体中，以替换两个Bsa之间的壮观霉素抗性基因。我的网站。每个文库中约有1500个卡那霉素阳性和壮观霉素阴性克隆。为了评估文库质量，随机选择每种载体的96个克隆，并对Sanger进行测序。如图1b所示，超过93.8％的克隆带有正确的sgRNA，表明质粒文库的准确性很高。从96个克隆中至少鉴定出64个独特的sgRNA，并且大多数sgRNA的检测不超过两次。这些结果表明文库中sgRNA分布的良好均匀性。然后，将质粒转化到农杆菌EHA105菌株中，每个文库汇集约3000个克隆。随机测序分析表明，农杆菌中的sgRNA与大肠杆菌中的sgRNA具有相似的准确性和均一性 （图1b）。

将三个基础编辑库稳定地转化为优良水稻品种（Oryza sativa L. japonica）。简历。Feigeng2020）。为了筛选抗除草剂事件，向再生培养基中添加了芳氧基苯氧基丙酸酯（APP）基团被ACCase抑制的化学物质haloxyfop-R-methyl。在两个浓度（分别为1和2μM）的除草剂压力下，每个文库中大约有5,000潮霉素阳性愈伤组织被筛选，两者都完全抑制了对照转化子的生长（图1c）。选择4-8周后，从用eABE文库转化的愈伤组织中回收了15个独立的芽。此外，还选择了10个独立的eABE愈伤组织事件，这些事件表现出对除草剂的抗性但在再生植物中失败。从抗性植物和愈伤组织中分离基因组DNA，以通过PCR测序鉴定sgRNA。我们发现25种抗药性事件中有23种，包括15种植物中的14种和10种愈伤组织中的9种，用sgRNA-127转化。在sgRNA-127靶向的基因组区域中，所有这些事件均携带单等位基因或双等位基因T₇到C的转换，导致耐药细胞中发生C2186R突变（图1d）。在一个抗性愈伤组织中观察到sgRNA-128，该愈伤组织在其间隔区进行单等位基因T ₁₃到C的转化，导致相同的C2186R突变。在一个再生植物中检测到sgRNA-10，导致单等位基因A ₁₆转化为G，并导致OsACC中的I1879V突变。通过Sanger测序进一步筛选了抗性事件中CT结构域的其余区域，未发现其他突变。还从eBE3文库中筛选了抗性植物和愈伤组织。haloxyfop-R-methyl选择了两株植物和一个愈伤组织，它们都带有sgRNA-121。单等位基因G ₁在所有三个耐药事件中均检测到C到C取代，导致W2125S突变。从eCDA库中未检测到耐药事件。

先前的研究表明，OsACC中ABE诱导的C2186R取代赋予了介质对水稻的除草剂抗性（Li等，2018）。在这里，通过CRISPR介导的碱基筛选产生了两个额外的单突变I1879V和W2125S，以赋予水稻APP除草剂抗性。值得注意的是，水稻中的I1879V和C2186R突变对应于杂草中的I1781V和C2088R赋予耐性的突变。但是，W2125S突变不同于杂草中相应的2027 Trp（W2125C）上保守的Cys取代，提供了新的抗性等位基因。转移到温室后，在W2125S和纯合C2186R植物中观察到严重的生长迟缓和完全不育。此外，在杂合C2186R植物的子代中未检测到纯合突变体，这表明纯合C2186R突变可能对水稻胚致死。生长和繁殖迟缓表明这些突变的适应性代价。在T_分离第2代，不含T‐DNA的纯合I1879V植物和杂合C2186R植物，并用市售APP除草剂Gallant进行喷雾，以确认其抗性并评估田间突变的潜在用途。野生型（WT）植物在7天后死亡，而所有突变体均表现出明显的抗性（图1e）。与野生型相比，在T ₁纯合I1879V植物中，在株高，剑叶长度和结实率方面无明显差异（图1f）。一些报道还表明，ACCase的I1781V突变在各种杂草中的健身成本甚至可以忽略不计（Jang等人，2013）。结合我们的观察，I1879V突变可能是育种商品水稻品种时获得除草剂抗性的理想OsACC等位基因。

可以在不同BE的编辑窗口中实现有效的核苷酸转换，例如分别位于eABE，eBE3和eCDA的4-8、3-9和2-4位（Li等人，2018年; Shimatani等人，2017年） ; Zong等，2017）。因此，在1653 bp的靶向序列中，611-nt（包括167个可编辑的As和162个可编辑的Ts），788-nt（包括165个可编辑的Cs和192个可编辑的Gs）和393-nt（包括84个可编辑的Cs和92个可编辑的Ts）。 Gs）区分别被eABE，eBE3和eCDA文库的141个sgRNA覆盖。在禾本科杂草中，CT结构域的7个不同氨基酸处的取代与自然产生的抗性有关（Jang等人。，2013年）。其中，水稻中的A∙T // G∙C转化可产生四个突变，而仅对应于C2186R的核苷酸位于文库窗口内。尽管发现了非典型的替代，例如W2125S的G到C转换和I1879V的sgRNA的16位突变，但仍应使用当前工具来限制这些非常规替代的效率。我们认为优化的BE可以提高编辑效率，扩展PAM识别能力，扩大编辑窗口并提供更灵活的转换类型，这将在不久的将来极大地促进Planta CRISPR-Cas9介导的基础编辑屏幕的发展。