Developing high-efficiency base editors by combining optimized synergistic core components with new types of nuclear localization signal peptide

Feipeng Wang,Chengwei Zhang,Wen Xu,Shuang Yuan,Jinling Song,Lu Li,Jiuran Zhao*,Jinxiao Yang*

Beijing Key Laboratory of Maize DNA Fingerprinting and Molecular Breeding, Beijing Academy of Agriculture & Forestry Sciences, Beijing 100097,China

ABSTRACT The clustered regularly interspaced short palindromic repeats (CRISPR)-CRISPR-associated protein(Cas)system has been widely used for genome editing.In this system,the cytosine base editor (CBE) and adenine base editor (ABE) allow generating precise and irreversible base mutations in a programmable manner and have been used in many different types of cells and organisms. However, their applications are limited by low editing efficiency at certain genomic target sites or at specific target cytosine(C)or adenine(A)residues.Using a strategy of combining optimized synergistic core components, we developed a new multiplex super-assembled ABE (sABE) in rice that showed higher base-editing efficiency than previously developed ABEs.We also designed a new type of nuclear localization signal(NLS)comprising a FLAG epitope tag with four copies of a codon-optimized NLS(F4NLSr2)to generate another ABE named F4NLS-sABE. This new NLS increased editing efficiency or edited additional A at several target sites.A new multiplex super-assembled CBE(sCBE)and F4NLSr2 involved F4NLS-sCBE were also created using the same strategy. F4NLS-sCBE was proven to be much more efficient than sCBE in rice.These optimized base editors will serve as powerful genome-editing tools for basic research or molecular breeding in rice and will provide a reference for the development of superior editing tools for other plants or animals.

1.Introduction

The clustered regularly interspaced short palindromic repeats(CRISPR)-CRISPR-associated protein(Cas)system has been used for genome editing in many different types of cells and organisms[1-4].Base editing is a newly developed genome-editing technology that permits the programmable generation of precise and irreversible base mutations without inducing double-strand breaks or requiring exogenous repair template DNA[5-7].Thus,base editing has been widely used to generate point mutations for basic research, gene therapy, or molecular breeding [8-17]. There are two base editors(BEs):the cytosine base editor(CBE),which converts target cytosineguanine(C-G)to thymine-adenine(T-A)base pairs,and the adenine base editor(ABE),which converts A-T to G-C.The most commonly used deaminase in CBEs in animals and plants is the rat cytidine deaminase rAPOBEC1 or Petromyzon marinus cytidine deaminase 1(PmCDA1) [5,6,8,11], while that in ABEs is a wild-type adenine deaminase ecTadA fused with its variant ecTadA* (ecTadA &ecTadA*) [7,18,19]. Both CBEs and ABEs comprise several essential components including sgRNA, Cas9 nickase (Cas9n), a deaminase,and nuclear localization signal (NLS) peptides. CBEs,but not ABEs,also require a uracil DNA glycosylase inhibitor(UGI).

Although ABEs and CBEs represent a useful system for base editing,they still show low editing efficiency at certain genomic target sites or at specific target C or A residues. To improve editing efficiency, the core components have been optimized for use in animals and plants. For example, the scaffold of sgRNA was modified to obtain enhanced sgRNA (esgRNA) and the Cas9n DNA sequence was extensively codon-optimized[18,20-23]. More effective cytidine deaminases have also been developed and employed,such as the hyperactive human AID,Anc689 APOBEC, human APOBEC3A, evoAPOBEC1, and evoCDA1, while a single ecTadA* in ABE was shown [24-28] to be more effective than the conventional ecTadA&ecTadA*.For the NLS,both a codon-optimized bipartite NLS(bpNLS)and an additional FLAG epitope tag at the N terminus of NLS (FNLS)improved base-editing efficiency[23,25,29].Additional copies of the UGI increased the editing frequency of CBEs[30].

In this study,we developed a multiplex super-assembled ABE system (sABE) by combining optimized synergistic core components, including esgRNA, a plant codon-optimized Cas9n, rice codon-optimized ecTadA&ecTadA*, and bpNLS.We found this new system was more efficient than previously developed ABEs.We further optimized sABE by using a new FLAG with 4*NLS(F4NLS) to generate F4NLS-sABE, which increased editing efficiency or edited additional A at several target sites. We also developed a multiplex super-assembled CBE system (sCBE) for rice with a rice codon-optimized rAPOBEC1 and double copies of UGI(2*UGI)as well as the same components of the sABE system(except for ecTadA & ecTadA*). The optimized F4NLS-sCBE also displayed greatly improved editing efficiency.

2. Materials and methods

2.1. Plasmid construction

The nucleotide sequences of key components are described in the supplemental files. sABE was constructed based on our previous reported vector SpCas9n-pBE-basic [20]. The fragment 3 in SpCas9n-pBE-basic was replaced by the codonoptimized bpNLSr&edTadAr&ecTadA*r&Cas9np&bpNLSr&TNos fusion sequence with SnaB I and Avr II, synthesized by GenScript Corp.(Nanjing,Jiangsu,China).The OsU3 promoter was then substituted by the OsU6a promoter with Kpn I and BamH I to generate vector sABE-basic. Three synthesized fragments: FNLSr1& edTadAr& ecTadA*r& Cas9np& NLSr1&TNos, F4*NLSr2& edTadAr& ecTadA*r& Cas9np& 4*NLSr2&TNos, and edTadAr& ecTadA*r& Cas9np& 4*NLSr2& TNos,were used to replace bpNLSr&edTadAr&ecTadA*r&Cas9np&bpNLSr& TNos in sABE-basic to generate the vectors FNLSsABE-basic, F4NLS-sABE-basic, and 4NLS-sABE-basic, respectively. Following Ma et al. [31], targets were added to each basic vector to generate sABE, FNLS-sABE, F4NLS-sABE, and 4NLS-sABE for rice transformation. These four ABEs all included the target sites T1 to T5. Two plasmids were constructed for each ABE. The target sites T1, T2, and T3 were placed under the control of respectively the OsU6a,OsU3, and OsU6c promoters in one plasmid; and the target sites T4 and T5 were placed under the control of the OsU6a and OsU6c promoters in the other plasmid. For the CBE systems,the codon-optimized bpNLSr&rAPOBECr&Cas9np&2*UGIr&bpNLSr&T35s fusion sequence with SnaB I and Avr II was synthesized by GenScript and then used to replace fragment 3 in SpCas9n-pBE-basic to generate vector sCBEbasic. Three fragments composed of synthesized FNLSr1&rAPOBECr,a PCR fragment of Cas9np,and synthesized 2*UGIr&NLSr1&T35s were used to replace fragment 3 in SpCas9n-pBEbasic, leading to vector FNLS-sCBE-basic. Similarly, synthesized F4*NLSr2& rAPOBECr, a PCR fragment of Cas9npand synthesized 2*UGIr& 4*NLSr2 & T35s were combined to substitute fragment 3 to obtain vector F4NLS-sCBE-basic.Target sites T6,T7,T8,and T9 were cloned before the esgRNAs in each basic vector using Bsa I following Xie et al. [32] and finally three vectors, sCBE, FNLS-sCBE, and F4NLS-sCBE were constructed for rice transformation. Target site sequences and primers used are listed separately in Tables S1 and S2.

2.2. Rice transformation

The Agrobacterium tumefaciens strain EHA105 was transformed with the binary vectors using the freeze/thaw method.Embryogenic calli induced from mature seeds of rice variety Nipponbare (Oryza sativa L. japonica. cv. Nipponbare) were used for Agrobacterium-mediated rice transformation as previously [33] described. Three days after Agrobacterium infection, calli were cultured on selection medium (containing 50 μg mL−1hygromycin)for four weeks to obtain hygromycinresistant calli.Resistant calli were transferred to regeneration medium (not containing hygromycin) for shoot induction for one month. The shoots were transferred to rooting medium for root induction when they were 4-5 cm long. After two weeks,T0plants were harvested.

2.3. DNA extraction and identification of transgenic T0 plants

T0plants were subjected to genomic DNA extraction using a DNA-quick Plant System kit (TIANGEN Biotech, Beijing,China). The target locus was amplified by PCR with Cas9-specific primers (Table S2) and samples with an 853-bp nucleic acid fragment in agarose gel electrophoresis were identified as transgenic T0plants.

2.4. Mutant identification

Several transgenic T0plants were used to identify C-to-T or Ato-G conversions. Target loci were amplified with specific primers and the PCR products were purified using an EasyPure PCR Purification Kit (TransGen Biotech). The PCR products were subjected to Sanger sequencing (Tsingke Biological Technology,Beijing,China)to identify mutations.Base editing efficiency in T0plants was defined as the percentage of mutants with any target C-to-T or A-to-G substitution among all the transgenic samples. Editing efficiency of single C-to-T conversion was defined as the percentage of mutants with Cto-T substitution at a specific single position among all the transgenic samples. The primers used are listed in Table S2.

3. Results and discussion

3.1.Development of a multiplex super-assembled ABE system(sABE) in rice

All of the core components of an ABE, including sgRNA, NLS,Cas9n, and adenine deaminase, affect its editing efficiency.Previous studies[18,20-22,25,29]have shown that esgRNA and codon-optimized bpNLS increase the base-editing efficiency of ABEs and CBEs in rice. Because the Cas9n DNA sequence optimized with favored human codons showed an improved base mutation frequency in human cells [23], we speculated that the editing efficiency of the adenine deaminase in the rice system might be increased by introduction of favored rice codons.Accordingly,to improve the base editing efficiency of ABE, we combined esgRNA, bpNLS, a plant-favored codonoptimized Cas9n and a rice-favored adenine deaminase to generate the super-assembled ABE for rice.

We developed a codon-optimized bpNLS [25] for rice,designated as bpNLSr. For Cas9n, we chose the previously reported high-efficiency plant codon-optimized Cas9n(Cas9np),which has a higher GC content in the 5′terminal region like most genes in Gramineae genomes [31]. For the adenine deaminases ecTadA and ecTadA*,we used the same design as previously reported [34]. These were separately codonoptimized for rice and were designated as ecTadArand ecTadA*r, respectively. These codon-optimized adenine deaminases showed a higher rice-favored codon usage frequency than other previously used [18,29,35,36] adenine deaminases (Fig. S1). Finally, we combined esgRNA with bpNLSr, Cas9np, ecTadAr, and ecTadA*rto generate sABE(Fig.1).For greater efficiency,we designed sABE as a multiplex base-editing system in which at least three targets could be edited simultaneously(Fig.1).

3.2. Improved base-editing efficiency of sABE

To directly compare the performance of sABE with that of other ABEs, the previously reported T1, T2, T3, T4, and T5 target sites were selected from the genes OsNRT1.1B,OsSPL14,OsWRKY45, OsSLR1, and OsACC, respectively [18,19,36] (Table S1). OsNRT1.1B encodes a nitrogen transporter [37], OsSPL14 encodes a protein involved in plant architecture [38],OsWRKY45 encodes a critical regulator of the salicylic acid signaling pathway [39], OsSLR1 encodes a DELLA protein that serves as a repressor in the gibberellic acid signaling pathway[40], and OsACC encodes an acetyl-coenzyme A carboxylase that is essential for lipid biosynthesis[41].

Fig. 1 - Diagram of the sABE, FNLS-sABE, F4NLS-sABE, and 4NLS-sABE base editors used in this study.

All targets could be edited by previously developed ABEs(Table S3).However,the new sABE showed much higher base editing efficiencies at all tested targets in T0plants,except for one target that was edited with comparable efficiency(Tables 1, S4 and S5).At the target site T1, the base-editing efficiency of sABE (75%) was 1.3-fold that of the previously developed ABE(59.1%)(Tables 1,S3).Adenine residues at positions 4,6,8,and 12 in the protospacer region(counting from the 5′ end of the target sequence)were efficiently edited by sABE(Table 1),whereas the previously developed ABE could not edit the A residue at position 12,nor could it produce the mutation types of A4A6 > G4G6, A8A12 > G8G12, and A4A6A8 > G4G6G8 in T0transgenic lines (Table S3, Fig. 2). For target T2, sABE showed highly improved base-editing efficiency (70.8%), 2.7-fold that of the previously developed ABE (26.1%) (Tables 1, S3).Although both sABE and the previously developed ABE could edit A residues at positions 5, 7, and 10, the latter could not produce the mutation type A5A7A10>G5G7G10(Tables 1,S3,Fig. 3-A, B). The base-editing efficiency of sABE at target site T3 was 75% in T0plants (Table 1, Figs. S2-A, S3) and 80% in calli (data not shown),28% higher than that of the previously developed ABE (62.3%) in calli (Table S3). Interestingly, unlike at the T1 and T2 target sites,sABE edited only the A residue at position 7, whereas the previously developed ABE edited A residues at positions 4 and 7(Table S3).Among all five targets,T4 was the only one with the same base-editing frequency(12.5%) for both sABE and the previously developed ABE(Tables 1, S3, Figs. S2-B, S4). As the editing window of ABE ranged from positions 4 to 7 [7] and there was only one editable A residue at position 6 in this range,only A6>G6 was produced by both ABEs(Tables 1,S3).The editing efficiency of the previously developed ABE was 20.6% at T5, but that of sABE was 3.2-fold higher (66.7%) (Tables 1 and S3). The point mutation C2186R in OsACC endows rice with resistance to the herbicide haloxyfop [18]. C2186R corresponds to A7 > G7 in target T5.Given that the targeted A7>G7 mutation frequency of sABE (66.7%) was much higher than that of the previously developed ABE (1.3%) (Tables 1, S3, Fig. 3-C, D), sABE could be used for efficient generation of haloxyfop-resistant rice lines in future.

3.3. Optimization of sABE using different NLS

The base editor sABE showed higher base-editing efficiency than did a previously developed ABE at most tested sites.However, further optimization of sABE was desirable to improve its base-editing efficiency, especially at sites with low editing efficiency and sites where fewer A residues were edited by sABE. Given that various types, numbers, and locations of the NLS have been shown to influence the editing efficiency of ABE and CBE [18,23,25,29], we focused on optimizing sABE by changing the bpNLSr.

We first selected the previously developed NLS with a FLAG epitope tag at the N terminus(FNLS)[23].Although this NLS resulted in increased editing efficiency of CBE in human and mouse cells, it has never been used to increase baseediting frequency in rice. The FNLS was codon-optimized for rice and then designated as FNLSr1. FNLSr1was then used to replace bpNLSrin sABE, generating FNLS-sABE (Fig. 1). The editing efficiency at the target sites T1 to T5 was tested with FNLS-sABE (Tables S6 and S7, Figs. S5-S9). Unexpectedly, in rice T0plants, FNLS-sABE displayed 2.1-fold and 1.2-fold higher editing efficiencies than sABE only at sites T4 and T5,while keeping comparable efficiency at target T1 (77.3% vs.75.0%) (Table 1). The editing efficiencies of FNLS-sABE at targets T2 (54.5%) and T3 (63.6%) were lower than those of sABE (70.8% for T2 and 75.0% for T3) (Table 1). Statistical results suggested that FNLSr1did not show better base editing efficiency improvement than bpNLSr(Fig.S10).FNLSr1did not perform as well as in mouse and human cells, where FNLS greatly increased C-to-T base editing [23]. This may be because the FNLSr1we used was not a suitable codonoptimized type for rice, because of the difference in genome environments in plant and animal cells, or because theperformance of FNLS differed between CBE and ABE. Whatever the reason, the results showed that not all NLS that perform well in animals will perform well in rice, and they should be thoroughly tested in each experimental system.

Table 1-Summary of base editing efficiencies at several sites for sABE,FNLS-sABE,F4NLS-sABE and 4NLS-sABE.

Next, we used our previously reported codon-optimized NLS for rice [34] and obtained FNLSr2. Considering that the number and locations of NLS affect the base-editing efficiency of ABE, and that four copies of N-terminal SV40-NLS resulted in 10-fold more efficient editing in knockout mouse cells[42],we combined F4*NLSr2and 4*NLSr2with sABE to generate F4NLS-sABE (Fig. 1). Given that in a previous study [18], the presence of three NLS at the C-terminus of Cas9n maximized the editing efficiency of ABE, we also constructed 4NLS-sABE with only one copy of 4*NLSr2at the C terminus of Cas9np(Fig.1). At target T4, F4NLS-sABE increased the base editing efficiency up to 38.1%, 3-fold that of sABE (12.5%) and 1.5-fold that of FNLS-sABE (26.1%) (Tables 1, S8, Fig. S11).Compared with sABE at T1 and T5 (75.0% and 66.7%efficiencies, respectively), F4NLS-sABE showed comparable base-editing efficiencies (83.3% and 61.9%), but did not edit any other A residues (Tables 1, S8 and S9, Figs. S12 and S13).Although F4NLS-sABE edited the targets T2 and T3 slightly less effectively than did sABE,the A residues at positions 12 at T2 and 4 at T3 were efficiently mutated by F4NLS-sABE but not by sABE (Tables 1, S9, Figs. S14 and S15). Thus, although F4NLS-sABE did not perform as well as we expected, it outperformed the previously developed ABEs and showed increased editing efficiency at many targets, especially those with low editing efficiency by sABE, or edited additional A.The base-editing efficiency of 4NLS-sABE was poor, and it edited a narrower range of A residues than did F4NLS-sABE and sABE (Tables 1, S10 and S11, Figs. S16-S20). This result was also unexpected,given that the version with three NLS at the C terminus of Cas9n performed best among the seven tested NLS combinations [18]. Given that two copies of three NLS, NLS with the FLAG epitope tag, and bpNLS were not tested in Li's study [18], it is unknown whether these would perform better.

Fig.2-Targeted editing of rice genome at the T1 target site by sABE.(A)Schematic view of the T1 target site in OsNRT1.1B.The target sequence is highlighted in red.(B)Sequence chromatograms of targeted base editing at the T1 target site edited by sABE.Arrows point to positions with an edited base.

The recently developed SurroGate systems [34], show greatly improved base-editing efficiencies of multiplex ABE or CBE in rice and the editing efficiency of F4NLS-sABE could be further optimized by use of these systems. Moreover, a simplified ABE system containing only ecTadA* instead of ecTadA & ecTadA* showed much higher editing efficiency at the T1 to T5 targets in rice[28].We speculate that the editing efficiency of F4NLS-sABE could be further improved by use of only ecTadA*and that the combination of SurroGate and only ecTadA* might perform best for optimizing F4NLS-sABE. In addition, considering the off-target effects of adenine base editors that usually occur at the RNA but not the DNA level[43-46], an F148A mutation to both ecTadA and ecTadA*,which was used to ensure a complete absence of RNA offtarget effects while maintaining efficient DNA on-target activity in human cells [47], could be tested in F4NLS-sABE for more precise editing.

3.4. Higher editing efficiency of optimized F4NLS-sCBE in rice

Considering the good performance of sABE and F4NLS-sABE,we hypothesized that similar architectures would also work well in multiplex CBE in rice. Because FNLS significantly increased C-to-T base editing in animals [23], we also tested FNLS in the CBE system, selecting the commonly used rAPOBEC1-based CBE system with UGI. We first codonoptimized rAPOBEC1 and UGI with rice-favored codons,generating rAPOBEC1rand UGIr(Figs.S21 and S22).rAPOBEC1r,Cas9np, and introduced two copies of UGIr(2*UGIr) together with bpNLSr,or FNLSr1and NLSr1or F4*NLSr2and 4*NLSr2into a tRNA-sgRNA system to generate three multiplex CBE systems:sCBE,FNLS-sCBE, and F4NLS-sCBE(Fig.4-A).

Fig.3-Targeted editing of rice genome at the T2 and T5 target sites by sABE.(A)Schematic view of the T2 target site in OsSPL14.(B)Sequence chromatograms of targeted base editing at the T2 target site edited by sABE.(C)Schematic view of the T5 target site in OsAcc.(D)Sequence chromatograms of targeted base editing at the T5 target site edited by sABE.The target sequence is highlighted in red (A,C).Arrows point to positions with an edited base(B,D).T-to-C conversions in the opposite strand are shown(B,D).

We selected four target sites, T6 to T9, from OsWaxy,OsCDC48, and OsSNB (Table S1) and combined them into one vector. In transgenic T0plants, we found that although sCBE functioned as a super-assembled system,there were still sites that could not be edited (Tables 2, S12). Of 32 T0plants, none showed C-to-T editing at T6 or T8 by sCBE and we observed very low editing efficiency at T7 and medium editing efficiency at T9 (Table 2, Figs. S23 and S24). However, F4NLSsCBE enabled efficient editing at T6 and showed 8-fold and 1.5-fold higher editing efficiency than sCBE at targets T7 and T9, respectively (Tables 2, S13, Figs. 4-B, C, S2-C, D, S25 and S26). F4NLS-sCBE showed increased editing efficiency for all the edited single C residues (Table S14, Fig. 4-D). FNLS-sCBE also performed better than sCBE (Tables 2, S15, Figs. S27 and S28),which is consistent with those obtained in animals[23].

The editing efficiency of F4NLS-sCBE was thus superior to those of sCBE and FNLS-sCBE. The finding that F4NLS-sCBE performed better than FNLS-sCBE suggests that FNLSr2is more suitable than FNLSr1for CBE,or that additional copies of NLS can increase editing efficiency in the CBE system. In contrast, F4NLS-sABE did not show increased editing efficiency at all tested targets.This may be because of differences between the ABE and CBE systems, so that FNLSr2may be more suitable for CBE than for ABE,or additional copies of NLS might work well in CBE but not in ABE.To further improve the editing efficiency of F4NLS-sCBE,further studies should focus on using more effective cytidine deaminases, such as PmCDA1, which showed better base editing performance than rAPOBEC1 [48,49], other evolved deaminases(evoAPOBEC1, evoCDA1, Anc689 APOBEC) [25], and the Surro-Gate system together or separately.The third-generation base editor(BE3)-type CBEs induces genome-wide off-target mutations not only at the DNA but also at the RNA level in human and rice cells [43-46]. To avoid off-target effects at the RNA level, F4NLS-sCBE could be further optimized by use of engineered cytosine deaminases, such as rAPOBEC1 with W90Y + R126E point mutations and human APOBEC3A with R128A orY130F mutations [47]. CBEs' off-target effects at the DNA level have remained a challenge that invites further study.

Fig.4-Cytosine base editing by multiplex super-assembled CBEs in rice.(A)Diagram of the sCBE,FNLS-sCBE and F4NLS-sABE base editors used in this study.(B)Schematic view of the T6 target site in OsWaxy.The target sequence is highlighted in red.(C)Sequence chromatograms of Line 22 at the OsWaxy target site edited by F4NLS-sCBE.The arrow points to the position with an edited base.G-to-A conversions in the opposite strand are shown.(D)Frequencies of single C-to-T conversions at the target site of T6,T7,T9 edited by sCBE and F4NLS-sCBE.

Table 2-Base editing efficiencies of sCBE,FNLS-sCBE,and F4NLS-sCBE at several sites.

4. Conclusions

We used a strategy of combining many synergistic core components to develop new multiplex super-assembled ABE and CBE systems in rice, sABE and sCBE. The base-editing efficiencies of sABE were higher than those of previously developed ABEs.On this basis,we also designed a new type of NLS, F4NLSr2, to generate F4NLS-sABE, which enhanced editing efficiency or edited additional target A at several target sites. We further applied F4NLSr2to develop a more efficient multiplex super-assembled CBE system in rice,F4NLS-sCBE. These improved base editors will serve as powerful tools for basic research or molecular breeding in rice,and provide a reference for the development of improved genome-editing tools for other plants and animals.

Declaration of competing interest

The authors have submitted patent applications based on the results reported in this paper.

Acknowledgments

This work was supported by the Beijing Scholars Program[BSP041]. We thank Jennifer Smith, PhD, from Liwen Bianji,Edanz Group China (www.liwenbianji.cn/ac), for editing the English text of a draft of this manuscript.

Author contributions

Jinxiao Yang and Jiuran Zhao designed the experiments.Feipeng Wang constructed all vectors. Jinling Song and Lu Li performed rice transformation.Shuang Yuan performed plant DNA extraction and target site amplification. Feipeng Wang,Chengwei Zhang, and Wen Xu analyzed the results. Jinxiao Yang and Jiuran Zhao supervised the project. Jinxiao Yang,Jiuran Zhao, Chengwei Zhang, and Feipeng Wang wrote the manuscript.

Appendix A.Supplementary data

Supplementary data for this article can be found online at https://doi.org/10.1016/j.cj.2020.01.003.