Optimizing base editors for improved efficiency and expanded editing scope in rice

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.