An optimized thymine base editing toolkit with various editing windows enables targeted T-to-G base conversions in rice

Abstract

Base editors offer potential for site saturation mutagenesis, yet deaminase-based editors are constrained to adenine and cytosine targets (Li et al., 2023a). Recently, glycosylase-based base editors (gBEs), which fuse engineered glycosylases with SpCas9 nickase (SpCas9n, D10A) to excise specific guanine or thymine bases, achieve base conversions through DNA repair over abasic sites (He et al., 2024; Tong et al., 2023; Ye et al., 2024). While glycosylase-based guanine base editors (gGBEs) show efficient guanine conversion in plants (Liu et al., 2024; Tian et al., 2024), thymine base editors (TBEs) remain unexplored (Figure S1).

Previous studies identified that the Y147A mutation in human uracil DNA glycosylase (hUNG) produces a thymine DNA glycosylase variant (hTDG). Highly active variants, TDG-EK (He et al., 2024) and TDG3 (Ye et al., 2024), were engineered using protein-language-assisted design and directed evolution, respectively, to enhance thymine editing. Cas9-embedding strategy further enhances base editing efficiency in mammalian cells (Figure 1a; He et al., 2024). To engineer efficient TBE tools for plants, we inserted three plant-codon-optimized TDG variants (hTDG, TDG-EK and TDG3; Figure 1b) into SpCas9n at various positions (I1029-G1030, F1046-I1063 and P1249-E1250) with a GGGGS linker.

Initially, we assessed the thymine editing efficiency of three TDG variants (hTDG, TDG-EK and TDG3) in rice protoplasts using a dual-luciferase reporter system, with hUNG as a control. The system utilized ZmUBI-driven firefly luciferase (FLuc) as a reference and 2x35S-driven NanoLuc (NLuc) as the reporter. A nonsense mutation was introduced at Gly69 (GGA > TGA), which was converted to a read-through codon by successful thymine editing, restoring NLuc translation (Figure S2). Three sgRNAs were designed to edit specific thymine positions (T9, T10 and T11; Figure 1c). Transient assays revealed that TDG3 outperformed hTDG by 2.2- to 20.8-fold and TDG-EK by 1.3- to 4.5-fold, with CE₁₀₄₆TDG3 showing the highest activity at T9 (Figure 1d, Figure S3).

To evaluate editing efficiencies of Cas9-embbeded TBEs (CE-TBEs) in transgenic rice plants, we tested six endogenous targets (SLR1-g1, SLR1-g2, SLR1-g3, ALS1-g1, EPSPS-g1, TB1-g1; Table S1). A total of 1681 T₀ plants edited by nine CE-TBEs were regenerated and analysed via Hi-TOM sequencing (Table S2). Plants with chimerism >10% were considered valid edits (Figure S4) and confirmed by subcloning sequencing (Figure S5). Genotyping revealed that the hTDG variant induced only 3.8% and 1.5% indels at SLR1-g2 and ALS1-g1, respectively (Figure 1e).

For the TDG-EK variant, the CE₁₀₂₉TDG-EK and CE₁₂₄₉TDG-EK constructs showed no detectable thymine editing activity, while CE₁₀₄₆TDG-EK achieved T-to-G transversion at T11 of SLR1-g3 and T3 of TB1-g1 with 3.6% (1/28) and 2.1% (1/48), respectively (Figure 1f). Consistent with protoplast results, the TDG3 variant significantly enhanced glycosylase activity in rice, increasing thymine editing and indel efficiencies (Figure 1e,f). CE₁₀₄₆TDG3 outperformed CE₁₀₂₉TDG3 and CE₁₂₄₉TDG3, with average efficiencies of 29.1% for indels and 16.4% for thymine editing (Figure 1g). The highest thymine editing efficiency was 38.2% (13/34) at SLR1-g3 (Figure 1f).

The CE-TDG3 constructs predominantly yielded T-to-G editing in rice T₀ plants, except for CE₁₀₄₆TDG3, which induced 5.9% (2/34) T-to-A editing at the SLR1-g3 target. No T-to-C editing was detected. Notably, different embedding variants exhibited distinct activity windows. CE₁₀₂₉TDG3 showed a wide editable range (T-6 to T17; with the PAM located at 21–23; Figure 1h), while CE₁₀₄₆TDG3 displayed a narrower editing window (T-1 to T11; Figure 1i). The CE₁₂₄₉TDG3 led to a backward-shifted editing window (mainly T9–T14; Figure 1j).

At the SLR3-g3 target, plants with a high proportion of amino acid substitutions or in-frame deletions at the TVHYNP motif exhibit a semi-dwarf phenotype (Figure 1k,l, Figure S6). Notably, T-to-G conversions were observed within protospacers at all tested targets except ALS1-g1, where CE₁₀₂₉TDG3, CE₁₀₄₆TDG3 and CE₁₂₄₉TDG3 induced high proportions of thymine mutations outside the protospacer or on the targeted strand (Figure 1m,n, Table S2). Although most edits were chimeric, likely due to sustained CE-TDG3 activity, T-DNA-free T-to-G mutants were obtained in the T₁ generation (Figure 1o, Table S4).

In this study, we compared a series of Cas9-embedded glycosylase constructs and developed three efficient TBEs: CE₁₀₂₉TDG3, CE₁₀₄₆TDG3 and CE₁₂₄₉TDG3. Genotyping of 1681 T₀ plants demonstrated that CE-TDG3 constructs enabled targeted T-to-G editing within distinct activity windows, achieving up to 38.2% efficiency in rice. Unlike editing in mammalian cells (predominantly T-to-C/G; He et al., 2024) and Escherichia coli (mainly T-to-A; Ye et al., 2024), our rice-optimized TBE (CE-TDG3) primarily induced T-to-G editing, with no significant off-target activity (Table S5). Improving TDG activity would further enhance editing efficiency and purity (Tong et al., 2024). Combining CE-TDG3 with ABE and AKBE (Li et al., 2023b; Tan et al., 2022; Wu et al., 2023) allows the conversion of T:A base-pair to any desired base-pair (Figure 1p), broadening possibilities for generating elite germplasm.

Y.T. and X.L. designed the research; X.L., Y.T. and R.S. performed the experiments; Y.T. and X.L analysed the data; Y.T., Y.P. and K.T. wrote and revised the manuscript.

The authors declare no competing interests.