The binary vector backbone PC-35S was first modified to contain a CsU6-tRNA-gRNA scaffold cassette (Huang et al., 2020) with two AarI sites for the gRNA insertion. The vector also contains a unique XbaI site downstream of the 35S promoter and a unique EcoRI site right upstream of the HSP terminator. Dicot plant codon-optimized Cas9 gene from the pXH1 vector (Huang et al., 2020) was first mutated at D10 to A to make Cas9^D10A nickase or nCas9. The evolved TadA8e (Richter et al., 2020) was PCR amplified using the ABE8e plasmid (Addgene) as template and primers ABE8-F1/R1; nCas9 was PCR amplified using primers ABE8-F2/R2 and pXH1 as template. The modified PC-35S was digested with EcoRI + XbaI. Ligation of TadA8e, nCas9, and EcoRI/XbaI-digested vector was performed using the in-fusion cloning method (Takara Bio) to make vector PC-ABE8e. The CmYLCV promoter (Huang et al., 2022), which confers high gene editing efficiency in citrus, was PCR amplified with primers CmY-F2/CmY-R2. The 35S promoter in PC-ABE8e was replaced with the CmYLCV promoter to make the final vector PC-CmYLCV-ABE8e. Primers LOBBE-F1/LOBBE-R1 (for gRNA GTT​TAT​ATA​GAG​AAA​GGA​AA) were annealed and cloned into AarI-digested PC-CmYLCV-ABE8e. All constructs were verified with Sanger sequencing.

The PC-ABE8e vector described above was digested with SbfI + BspEI to remove TadA8e. The CmYLCV promoter (Huang et al., 2022) was PCR amplified using primers CmY-F1/R3. The fragment A3A-RAD51DBD (Supplementary Information S1) was synthesized by Integrated DNA Technologies, Inc. (Coralville, IA, United States). Ligation of the CmYLCV promoter, A3A-RAD51DBD, and SbfI/BspEI-digested vector was performed using the in-fusion cloning method (Takara Bio) to make an intermediate vector CmYLCV-A3A-RAD51-nCas9. CmYLCV-A3A-RAD51-nCas9 was further digested with EcoRI. The UGI was PCR amplified using primers UGI-F1/R1 and cloned into EcoRI site of the CmYLCV-A3A-RAD51-nCas9 vector via the in-fusion cloning method (Takara Bio) to construct the final CBE vector. Two gRNAs for two different alleles of the citrus ALS gene were designed. Primers ALS-F/ALS-R were used to amplify gRNA scaffold-tRNA unit using the plasmid pXH1 (Huang et al., 2020) as template. The amplicon was digested with BsaI and cloned into the AarI-digested CBE to make the CBE-2xALS construct. All constructs were verified by Sanger sequencing.

The constructs were transformed into Agrobacterium strain EHA105. Agrobacterium-mediated transformation of citrus epicotyl was performed as described previously (Jia et al., 2019a; Huang et al., 2020; Huang et al., 2022). Shoots of GFP positive, TATA-edited citrus were micro-grated onto Carrizo rootstock seedlings. Survived plants were transplanted in soil after establishment in a glasshouse. For the selection of herbicide chlorsulfuron resistant citrus, citrus epicotyl segments were cultured on kanamycin-containing selection media (100 mg/L) for 1 week under dark at 30°C. After 1 week, the citrus epicotyl segments were transferred to chlorsulfuron (Fisher Scientific, Catalog No.50-255-082) containing media (150 nM) without kanamycin under light at room temperature. Every 3 weeks, the citrus epicotyl segments were transferred to new chlorsulfuron-containing media to select chlorsulfuron-resistant shoots. After three rounds of subculture with chlorsulfuron selection, chlorsulfuron-resistant shoots were visible on the media.

We performed genotyping of citrus transformants as described previously (Huang et al., 2022). Citrus genomic DNA was extracted using the CTAB (cetyltrimethylammonium bromide) method. Detection of editing in the target genes was performed via amplifying the target regions (primers in Supplementary Table S1) with high fidelity DNA polymerase Q5 (New England Biolabs, Ipswich, MA, United States), followed by cloning of PCR products and sequencing. Primers LOBpro-F1/LOBpro-R1 were used for the LOB1 promoter genotyping. Primers CsALSgt-F2/CsALSgt-R2 were used for ALS genotyping. Primers Cas9gt-F1/Cas9gt-R1 were used for Cas9 genotyping. Primers CsALSgt-F1/CsALSgt-R1 were used for the PCR detection of ALS.

Xanthomonas citri subsp. citri (Xcc) wild type strain 306 and dLOB2 containing Xcc pthA4:Tn5 (Hu et al., 2014; Zhang J. et al., 2017; Teper et al., 2020; Huang et al., 2022) were suspended in 20 mM MgCl2 at 10⁸ CFU/ml. The bacterial suspensions were syringe-infiltrated into fully expanded young leaves of wild type grapefruit, sweet orange Hamlin plants, or edited lines. Three different leaves from each genotype were included for the Xcc inoculation assays. Inoculated plants were kept in a temperature-controlled (28°C) glasshouse with high humidity. Pictures were taken 8 days post inoculation for disease resistance evaluation.

Reverse Transcription-Quantitative (RT-qPCR) was performed essentially as described previously (Huang et al., 2022). Leaves of wild-type and a representative grapefruit TATA-edited plant with and without Xcc inoculation were sampled at 48 h post inoculation. The citrus house-keeping gene GAPDH was used as an endogenous control. The primers QLOB1-F1/QLOB1/R1 and GAPDH-F1/R1 for qPCR are listed in Supplementary Table S1.

To analyze potential off-targets, we analyzed the putative off-targets using a web-based software (http://crispr.hzau.edu.cn/cgi-bin/CRISPR2/CRISPR). Genomic DNA from transgenic lines was used as template, and the primers listed in Supplementary Table S1 were used to amplify the fragments spanning the off-targets. Finally, the PCR products were subjected to Sanger sequencing.

To test if precise gene editing works in citrus, we first tested adenine base editors (ABE), ABE8e which mediates A⋅T-to-G⋅C base changes (Richter et al., 2020). The citrus optimized ABE8e vector was constructed (Figure 1A). TadA8e (Richter et al., 2020) was N-terminally fused with Cas9^D10A nickase. There is a nuclear localization signal (NLS) at each end of TadA8e- Cas9^D10A to increase its nucleus transportation. The CsU6-tRNA-gRNA-scaffold unit for multiplex editing in this ABE system was described previously (Huang et al., 2020). We sought to edit the TATA box (Figures 1B,D) located upstream of the EBE of the promoter of the citrus canker susceptibility (S) gene LOB1 (locus ID: Cs7g27640, C. sinensis v2.0 genome) with ABE8e. A nearby NGG PAM site enables the TATA box within the editing window of ABE8e. The EBE region of the LOB1 promoter in citrus is responsible for binding by the TAL effector PthA4 of Xcc (Hu et al., 2014) to activate its expression. The TATA box of the LOB1 promoter overlaps with the EBE region (Figures 1B,D). Hu et al. (2014) previously showed that mutation of TATA box abolishes LOB1 induction by PthA4 in the transient assay. We first investigated if the ABE construct targeting the TATA box can edit the target as expected through Xcc-facilitated agroinfiltration of citrus leaves (Jia and Wang, 2014). The transient assay (Figure 1C) showed that the construct edited the TATA box as anticipated (Figures 1D,E). The ABE mutated TATA to CACA (from TATA to TGTG for the complementary strand). Test on another gene CsTub (Cs1g21050) through transient assay also demonstrated precise editing (Figures 1F,G).

Biallelic editing of CsLOB1 coding region results in Xcc resistance while maintaining normal plant development and growth (data not shown). We expected that editing of the TATA box of CsLOB1 would not affect plant development and growth either. Agrobacterium-mediated stable transformation of grapefruit (Citrus paradise) and sweet orange (Citrus sinensis) Hamlin epicotyl tissues was conducted to edit the TATA box in the CsLOB1 promoter of both varieties. Two transgenic grapefruit plants and one transgenic sweet orange were obtained. Genotyping showed that in the grapefruit line #2 and Hamlin sweet orange mutant the TATA box was successfully edited (Figures 2A,B). The TATA box in the LOB1 promoter in grapefruit line #2 and Hamlin sweet orange mutant was 100% edited into CACA. The purity of editing was confirmed through direct sequencing of PCR products and colony sequencing of cloned PCR products. In grapefruit line #1, the first T in TATA box was 100% mutated to C, while 56% of the second T in the TATA box was mutated into C (9 clones out of 16). Another T to C editing was observed immediately upstream of TATA box in the grapefruit line #1 (31.2%, 5 clones out of 16). The bystander editing or proximal base editing has been observed in other studies too (Molla et al., 2020). Using the ABE system, we achieved high rate (66.7%) of biallelic/homozygous editing of the TATA box in the LOB1 promoter. After micro-grafting, one transgenic grapefruit plant and one transgenic sweet orange plant survived. Inoculation of the TATA-edited plants with Xcc demonstrated that the TATA-edited plants were resistant to Xcc (Figure 2C). On the contrary, all plants were susceptible to Xcc pthA4:Tn5 dLOB2, which carries a designer TAL dLOB2 for LOB2 induction to cause canker disease in citrus plants. LOB2, a LOB1 homolog, can cause canker symptoms when artificially induced, such as in the presence of designer TAL dLOB2 (Zhang F. et al., 2017; Teper et al., 2020; Huang et al., 2022). Xcc pthA4:Tn5 dLOB2 was included in the inoculation assay as a control. The onset of canker disease by Xcc pthA4:Tn5 dLOB2 on the same leaves excluded the possibility that lack of canker disease on the TATA-edited mutant plants is due to leaf age. To explore the outcome of TATA box editing on LOB1 induction by Xcc, we performed RT-qPCR for both wild type and TATA-edited grapefruit plant. The results showed that at 48 h after Xcc inoculation, LOB1 was induced about 64 folds in wild type citrus, while LOB1 was only induced 6.8 folds in TATA-edited citrus (Figure 2D). This result provides evidence that editing of TATA box of LOB1 dramatically compromises its inducibility by Xcc. The basal expression level of LOB1 is not significantly different between WT and TATA-edited citrus. These results demonstrated that base editor ABE8e can efficiently and precisely edit the citrus genome (both grapefruit and sweet orange). These results also showed a promising strategy by editing the EBE-associated TATA box of S genes to breed Xanthomonas-resistant crops through base editor ABE8e.

Next, we tested if cytosine base editors (CBE), which can mediate C⋅G-to-T⋅A base changes, work in citrus. A citrus optimized CBE vector was constructed (Figure 3A). We chooseAPOBEC3A (A3A) deaminase for citrus CBE considering its wide deamination window and high editing efficiency (Zong et al., 2018; Zhang et al., 2020). An inhibitor of uracil DNA glycosylase (UGI) was fused immediately downstream of nCas9 to increase efficacy of C to T editing. A single-stranded DNA-binding domain from Rad51 protein (RAD51-DBD) was inserted between the deaminases and nCas9, which was reported to substantially increase activity and an expand editing window of CBE (Zhang et al., 2020). Tan et al. (2022) recently reported that RAD51-DBD can also dramatically increase editing efficiency of ABE. This CBE system contains a CsU6-tRNA-gRNA-scaffold unit for multiplex editing. The final CBE construct also carries a GFP expression cassette for direct visualization of transgene integration. We first determined whether the citrus acetolactate synthase (ALS) gene (locus ID: Cs7g22130, C. sinensis v2.0 genome) can be edited with this CBE. Most citrus varieties, owing to their hybrid nature, have two different alleles of the ALS gene responsible for herbicide resistance. We designed two gRNAs targeting both ALS alleles. The transient expression result showed that the CBE construct edited both alleles of the ALS gene as expected (Figures 3B,C).

Acetolactate synthase (ALS) is an enzyme required for the biosynthesis of multiple branched-chain amino acids, such as valine, leucine, and isoleucine. Chlorsulfuron is a known ALS inhibitor that kills plants, thus being used as a herbicide. A single amino acid mutation in ALS genes in various plants confers resistance to chlorsulfuron (Kuang et al., 2020; Malabarba et al., 2021). For example, when the amino acid P in the “QVPRRMI” amino stretch of ALS protein is mutated to S or F (Figure 3B, Figure 4C), the plant becomes resistant to chlorsulfuron. We first tested the sensitivity of citrus to chlorsulfuron. The growth of citrus Carrizo citrange, a hybrid of Citrus sinensis “Washington” sweet orange X Poncirus trifoliata, was completely inhibited by chlorsulfuron (Figure 4A). Agrobacterium-mediated stable transformation of Carrizo epicotyl tissues was performed with chlorsulfuron selection in the culture media after 1 week of culture. Three chlorsulfuron-resistant plants grew on the chlorsulfuron-containing medium while most of the transformed epicotyl tissues did not grow (Figure 4B). One of these three plants showed green fluorescence, indicating transgene integration. Genotyping of the other two chlorsulfuron-resistant plants showed that both alleles of the ALS gene were edited in the two edited plants (Figures 4C,D). Only C-to-T substitutions, but no other by-product substitutions, were observed in the edited plants. The editing rendered the “QVPRRMI” amino stretch of ALS protein to “QVSRRMI” in one allele, and to “QVFWRMI” in another allele. Both mutant plants have same genotypes in ALS locus. Intriguingly, these two plants did not show green fluorescence although the construct used for transformation contains a GFP expression cassette (Figure 3D). To further confirm the absence of T-DNA integration, we performed PCR amplification of the Cas9 gene in the mutant plants and the Cas9 gene was undetectable in the herbicide-resistant citrus plants (Figure 4E). This indicates that the ALS-edited plants are transgene-free, which likely resulted from the transient expression of the CBE construct, representing the first transgene-free gene-edited citrus using the CRISPR technology.

To investigate whether base editors can introduce off-target mutations, we amplified and sequenced top 12 potential off-target sites in both Grapefruit and sweet orange Hamlin TATA-edited mutant lines. The top potential off-targets all carry 4 or more mismatches. Sequencing results showed that no edits were detected at potential off-target sites (Supplementary Table S2). For ALS-edited plants, only 1 potential off-target exists with four or fewer mismatches within the protospacers. Sequencing results showed that no edits were detected at the potential off-target site.