Main

Programmable RNA-guided endonucleases, including CRISPR-Cas9, are driving advances in genome editing for both fundamental research and biotechnology. The ability to genetically modify plant genomes has allowed for the creation of rationally designed phenotypes. However, efficient delivery of genome editing reagents to plants remains a major challenge. The most common strategy is to encode RNA-guided genome editors (for example, CRISPR-Cas enzymes) within transgenes and use tissue culture and plant transformation approaches to make transgenic plants, after which genetic crosses are required to remove the transgenic material but retain the edits1,2,3. However, current plant transformation methods are limited to specific plant species and genotypes, often require considerable time, resources and technical expertise, and can cause unintended changes to the genome and epigenome1.

An approach to circumvent these limitations is to use plant viral vectors to deliver genome editing reagents such as meganucleases or zinc finger nucleases (ZFNs) for targeted mutagenesis4,5. While the use of meganucleases and ZFNs for viral-mediated plant genome editing was a notable advance, the ability to encode an easily programmable RNA-guided CRISPR system would be highly advantageous. As such, several viral vectors have been engineered to encode guide RNAs (gRNAs) for delivery to transgenic plants already expressing Cas9, resulting in somatic and germline editing and transmission of edits to the next generation6,7,8,9. Because plants have evolved mechanisms to restrict viral infection of meristem and germ cells, most viruses are rarely sexually transmitted10. However, transient invasion of meristem cells by viral RNAs encoding gRNAs can allow these cells to be edited and for these edits to be seed transmissible6,7,8,9. While these approaches represent important advances, they still require the use of nucleases that can be challenging to engineer (such as meganucleases and ZFNs), or transgenic plants expressing the CRISPR-Cas endonuclease protein.

A strategy to avoid the need for transgenic plant materials has been the use of viral vectors with large cargo capacities, capable of expressing entire RNA-guided editing systems (for example, Cas9 and the gRNA). This approach has been met with some success; however, it still requires plant regeneration steps because these viruses do not cause germline editing and heritability of the edits11,12,13,14. On the other hand, encoding entire CRISPR systems in viruses that are capable of germline transmission has been challenging because of their limited cargo capacity6,7,8,9,15.

To overcome this cargo size limit, we explored the potential of TnpB, a class of ultracompact RNA-guided endonucleases (~400 amino acids)16,17,18, to be encoded in a plant RNA viral vector. As ancestors of Cas enzymes, TnpBs similarly utilize a programmable RNA guide, called an omega RNA (ωRNA), to be directed to any target site and induce genome edits. Previously, TnpBs ISDra2, ISYmu1 and ISAam1 were shown to be capable of targeted genome editing in mammalian cells, and ISDra2 and ISYmu1 in monocot rice plant cells16,19,20,21. Here we tested the ISDra2, ISYmu1 and ISAam1 TnpBs for genome editing in the dicot plant, Arabidopsis. Given the single cargo site in the TRV vector that is typically used, we sought to express both the TnpB protein and its guide RNA within the same mRNA transcript under a single promoter, similar to their natural expression arrangement16,17,18.

To test the activities of TnpB and its gRNA encoded in a single transcript, we first expressed these three TnpBs and assessed their RNA-guided plasmid interference activities in bacteria. We co-expressed the TnpB and gRNA from the same promoter as a single transcript, maintaining their natural sequences without codon optimization. We compared two configurations of the 3’-guide region: one extended continuously without a terminator to mimic the natural TnpB condition, and another capped by the hepatitis delta virus (HDV) ribozyme, as previously used in bacteria16 (Extended Data Fig. 1). Our results showed that without the HDV ribozyme, only ISDra2 demonstrated plasmid interference activity whereas with the HDV ribozyme, all three TnpBs exhibited robust activity at both 26 °C and 37 °C (Fig. 1a and Extended Data Fig. 2). These findings revealed that single transcript expression cassettes with an HDV ribozyme sequence at the 3’ end are capable of cleaving plasmid DNA in bacteria.

Fig. 1: Expression of TnpB and guide RNA in a single transcript for plant genome editing.
figure 1

a, Barplots of interference assay testing the single transcript expression TnpB vectors for cleavage in E. coli. Data are from experiments performed at 26 °C (top) and at 37 °C (bottom). Bars indicate absence (black) or presence (purple) of a PAM on the target plasmid. The Y axis is a log10 scale of the normalized c.f.u.s ml−1. The X axis displays the three TnpBs tested using the single expression transcript design without or with an HDV ribozyme. The s.e.m. was calculated for each experiment, with 3 replicates per experiment. b, Schematic of the single expression transcript TnpB-ωRNA plasmid design used for plant genome editing. The green arrow symbolizes the AtUBQ10 promoter; the dark grey boxes indicate the 2×-FLAG, SV40 NLS and HDV ribozyme sequences; the light grey boxes indicate the TnpB-ωRNA and guide sequences; the red box symbolizes the rbcS-E9 terminator; the black arrow indicates the orientation of the TnpB-ωRNA expression cassette. c, Barplot displaying the average editing efficiencies (±s.e.m.) for protoplast experiments using ISDra2, ISYmu1 and ISAam1 TnpBs. Each dot represents the average editing efficiency (percent indel reads) of a gRNA from Extended Data Fig. 3a, with number of samples indicated at the top of the plot. d,e, ISYmu1 somatic editing in T1 transgenic plants for ISYmu1 gRNA2 (d) and ISYmu1 gRNA12 (e). The genotypes are plotted along the X axis and the editing efficiencies (percent indel reads) (±s.e.m.) are plotted on the Y axis. Each dot indicates a single T1 transgenic plant. The room and HS treatments stand for room temperature and heat-shock plant growth conditions, respectively.

Full size image

To test the single expression cassette for targeted genome editing in Arabidopsis, we used the AtUBQ10 promoter to drive expression of the TnpB-ωRNA and a gRNA targeting the PHYTOENE DESATURASE3 (AtPDS3) gene region, followed by the HDV ribozyme and rbcS-E9 terminator (Fig. 1b). We tested 20 ISDra2 sites, 10 ISYmu1 sites and 7 ISAam1 sites for editing capabilities in Arabidopsis protoplast cells (Supplementary Table 1)22. ISDra2 and ISYmu1 demonstrated active editing ranging from 0–4.8% and 0.1–4.2%, respectively, as measured by next-generation amplicon sequencing (amp-seq) (Extended Data Fig. 3a). ISAam1 was much less active, with editing efficiency ranging 0–0.3% (Extended Data Fig. 3a). On average, we observed editing efficiencies of 1% for ISDra2, 2.1% for ISYmu1 and 0.1% for ISAam1 (Fig. 1c). In line with previous reports, the DNA repair profiles consisted of deletion-dominant repair outcomes for all three TnpBs (Extended Data Fig. 3b)16,19,20. These data demonstrate that ISDra2, ISYmu1 and ISAam1 are all capable of targeted genome editing in Arabidopsis plant cells using the single transcript expression design.

To evaluate TnpB-mediated editing in transgenic plants we selected ISYmu1, as it demonstrated the highest average editing efficiency in Arabidopsis protoplast cells and was shown to exhibit no off-target editing in rice19. Two gRNAs with the most active editing were selected, each targeting a unique genomic context. gRNA2 targeted the coding region of AtPDS3, whereas gRNA12 targeted the promoter region directly upstream of the AtPDS3 gene. Transgenic plants were created via standard floral dip transformation utilizing the same plasmids as for the protoplast experiments23. To test for sensitivity to temperature, transgenic plants expressing ISYmu1 were either grown at room temperature or subjected to a heat-shock treatment. We tested editing in wild-type (WT) plants, as well as in the rna dependent rna polymerase 6 (rdr6) mutant which is known to have reduced transgene silencing24. Analysis using amp-seq revealed an average editing efficiency of 1.6% and 2.5% for gRNA2 in WT and rdr6, respectively (Fig. 1d). Analysis of gRNA12 revealed greater editing than gRNA2, averaging 44.9% editing in WT and 75.5% in rdr6 (Fig. 1e). Comparison of editing efficiency in the plants grown at room temperature with those that received the heat-shock treatment revealed a preference for increased temperature for both target sites in the WT background, demonstrating 6.3-fold and 1.4-fold increases in editing for gRNA2 and gRNA12, respectively (Fig. 1d,e). In rdr6, we observed a 13-fold increase in editing for gRNA2, but little change in editing for gRNA12 (Fig. 1d,e). The editing outcomes from transgenic T1 plants expressing ISYmu1 consisted of chimaeric, deletion-dominant, DNA repair profiles (Extended Data Fig. 4). These data demonstrate that ISYmu1, encoded as a transgene, is capable of performing efficient genome editing in Arabidopsis plants, and that heat treatment and the rdr6 silencing mutant can be used to increase editing efficiency.

Encouraged by the ISYmu1 activity in transgenic Arabidopsis plants, we next tested ISYmu1 for TRV-mediated genome editing. TRV is a bipartite RNA virus composed of TRV1 and TRV2 (Fig. 2a). Previous work has shown that the TRV2 RNA can be engineered by inserting a cargo expression cassette downstream of the pea early browning virus promoter (pPEBV) (Fig. 2a)25,26. To test ISYmu1 for genome editing capabilities via TRV delivery to Arabidopsis, we engineered two TRV2 cargo architectures. In TRV2 Architecture_A, the tRNAIleu was directly downstream of the TnpB and gRNA sequences (Fig. 2a). In TRV Architecture_B, we included an HDV ribozyme sequence between the guide and tRNAIleu sequence (Fig. 2a). We included tRNAIleu in both designs as it was previously shown to promote systemic TRV movement and transmission of edited alleles to the next generation25,26.

Fig. 2: Somatic and heritable editing in Arabidopsis using TRV to deliver ISYmu1 TnpB and guide RNA targeting AtPDS3.
figure 2

a, Schematic of the TRV1 and TRV2 plasmids. Green arrows indicate the RNA-dependent RNA polymerase (RDRP) and pPEBV promoters for TRV1 and TRV2, respectively; the grey boxes in TRV1 and TRV2 indicate the native TRV components; the red Cargo box in TRV2 indicates the location of either Architecture_A or Architecture_B; below TRV2 are schematics of the components, Architecture_A or Architecture_B, cloned into the TRV2 Cargo slot. b, Representative picture of a plant displaying white sectors in leaves (yellow arrows) ~3 weeks after TRV delivery. c,d, Barplot displaying the somatic editing efficiencies (percent indel reads) (Y axis) for ISYmu1 gRNA2 in WT and ku70 genetic backgrounds (c) and for ISYmu1 gRNA12 in WT (d). The TRV2 cargo architectures are plotted along the X axis with either room or HS treatment. Each dot represents an individual plant that underwent agroflood TRV delivery. The s.e.m. was calculated for each experiment. e,f, DNA indel repair profile for an individual WT plant that underwent delivery of TRV Cargo Architecture_B with ISYmu1 gRNA2 (e) or with ISYmu1 gRNA12 (f) under the heat-shock treatment. The top five most common indel types are listed on the left. The read counts for each indel are listed on the right. The PAM is identified by the red box, and the target site is outlined by the black box, in the Reference sequence. The total read number and editing efficiency are listed below each indel profile. g, Representative image of albino and green progeny seedlings from a WT plant showing 54.54% somatic editing using the TRV2 Architecture_B design with gRNA2 that underwent heat-shock treatment. h, Sanger sequencing trace file screenshot from one of the albino plants in Fig. 3a. Top: sequence of the wild-type reverse complement. Middle: the ISYmu1 gRNA2 target and PAM (grey box). Bottom: the ab1 trace file displaying a homozygous 4 bp deletion. i, Table summarizing the transmission of edited alleles from two individual plants that underwent agroflood delivery using ISYmu1 gRNA12. The ‘Progeny screened’ column indicates the number of seedlings genotyped; the ‘Biallelic edits (%)’ column indicates the number of seedlings containing biallelic edits; and the ‘Monoallelic edits (%)’ column indicates the number of plants harbouring monoallelic edits.

Full size image

First, we evaluated TRV-mediated editing potential with gRNA2 using both TRV2 Architecture_A and Architecture_B. gRNA2 was selected because it targets the AtPDS3 coding sequence, enabling easy phenotypic screening for editing due to white photobleaching of cells containing biallelic mutations25,26. We delivered TRV vectors to both WT and the ku70 genetic mutant. Ku70 plays a role in the non-homologous end joining (NHEJ) double strand break repair pathway27. ISYmu1-mediated editing efficiency should be greater in the ku70 genotype if double-stranded breaks generated by ISYmu1 are repaired through NHEJ. Each TRV2 plasmid was co-delivered with the TRV1 plasmid to Arabidopsis plants using the agroflood method26. White speckles were observed on some of the leaves at ~3 weeks post agroflooding, suggesting that sectors of cells contained biallelic mutations in the target AtPDS3 gene (Fig. 2b). Amp-seq analysis revealed an average of 0.1% and 0% editing efficiency in leaf tissue of WT and ku70 plants agroflooded with TRV2 Architecture_A and grown under room temperature, respectively (Fig. 2c). For the heat-shock-treated plants, we observed an average editing efficiency of 0.4% in WT and 0.7% in ku70 plants agroflooded with TRV2 Architecture_A (Fig. 2c). Using TRV2 Architecture_B, we observed an average of 0.6% and 2% editing in WT and ku70, respectively, for the room-temperature-grown plants. For the plants that received TRV2 Architecture_B and a heat shock, we observed an average editing efficiency of 3.3% in WT and 8.9% in ku70 (Fig. 2c). These results show that Architecture_B, containing the HDV ribozyme, generated higher editing than Architecture_A, and that the ku70 mutant can enhance editing efficiency.

Next, we tested gRNA12 utilizing the TRV2 Architecture_B, since this architecture demonstrated the highest levels of editing for gRNA2. Using the same agroflood TRV delivery method to WT plants, we observed an average of 8.51% and 4.27% editing efficiency in room-temperature and heat-treatment growth conditions, respectively (Fig. 2d). Further, 6/57 plants displayed editing greater than 40%, with 4 plants showing greater than 75% editing when room-temperature treatment was used (Fig. 2d). Again, analysis of the repair outcomes showed deletion-dominant profiles for ISYmu1 gRNA2 and gRNA12 (Fig. 2e,f).

To test for transmission of edited alleles to the next generation, we first screened the progeny of a WT plant showing 54.54% somatic editing using the TRV2 Architecture_B design with gRNA2 that underwent heat-shock treatment. In total, 2,318 seeds were sown on ½ MS plates containing 3% sucrose. After 10 days, 68 albino seedlings were observed, suggesting biallelic mutations in the PDS3 gene (Fig. 2g). To confirm that AtPDS3 was mutated, we performed Sanger sequencing on the two white seedlings shown in Fig. 2g, which revealed both plants to be homozygous for a 4-bp frame-shift deletion (Fig. 2h). To further characterize transmission of edited alleles, amp-seq on 209 seedlings (41 albino and 168 green) showed that all of the albino seedlings contained biallelic mutations, with the majority of mutations being the 4-bp deletion observed in Fig. 2h (Supplementary Table 2). Of the 168 green seedlings, 8 were heterozygous (4-bp deletion in WT) (Supplementary Table 2).

Next, we characterized transmission of edited alleles from two individual lines, plant 54 (80.5% somatic editing) and plant 69 (77.1% somatic editing), that underwent agroflood using gRNA12 TRV2 Architecture_B with the room-temperature condition. As expected, we did not observe any albino seedlings, probably because this target site is located upstream of the AtPDS3 transcription start site. Using Sanger sequencing, we analysed the genotypes of 148 and 75 progeny seedlings from plants 54 and 69, respectively. Sanger sequencing analysis of the progeny from plant 54 revealed 27 (18%) biallelic and 25 (17%) monoallelic edited plants (Fig. 2i, Supplementary Table 2 and Extended Data Fig. 5a)28. For plant 69, we observed higher transmission of edited alleles, totalling 32 (43%) biallelic and 15 (20%) monoallelic edited plants (Fig. 2i, Supplementary Table 2 and Extended Data Fig. 5b)28. These data demonstrate the heritability of edits generated via TRV delivery of ISYmu1 at two distinct target sites.

To test the applicability of this approach to another locus, we designed six ISYmu1 gRNAs targeting the AtCHLl1 gene (AT4G18480) (Supplementary Table 1). AtCHLl1 was chosen due to the obvious yellow phenotypic readout that AtCHLl1 homozygous mutant plants display29. The agroflood method was used to individually deliver TRV Architecture_B vectors targeting each of the six AtCHLl1 sites. Plants were exposed to either room temperature or heat-shock growth conditions as previously described. About 2 weeks post agroflood, we observed yellow sectors on some of the plants infected with TRV ISYmu1 gRNA4, gRNA6 and gRNA9 (Fig. 3a). To quantify somatic editing, tissue samples from plants infected with TRV targeting gRNA4, gRNA6 and gRNA9 were collected for amp-seq analysis. We observed an average of 8.3%, 2.9% and 1% somatic editing for gRNA4, gRNA6 and gRNA9 for the room-temperature condition, respectively (Fig. 3b). For the plants that underwent the heat shock, we detected an average editing frequency of 18.4%, 1.2% and 0% for gRNA4, gRNA6 and gRNA9, respectively (Fig. 3b). Further, 4/47 (8.5%) and 4/12 (33.3%) plants infected with ISYmu1 gRNA4 displayed somatic editing greater than 40% for room-temperature and heat-shock samples, respectively (Fig. 3b).

Fig. 3: Somatic and heritable editing in Arabidopsis using TRV to deliver ISYmu1 TnpB and guide RNA targeting AtCHLl1.
figure 3

a, Representative pictures of plants displaying yellow sectors ~2 weeks after TRV delivery. The gRNA and somatic editing efficiency is indicated in the upper left corner of each picture. b, Barplot displaying the somatic editing efficiencies (±s.e.m.) for ISYmu1 gRNA4, gRNA6 and gRNA9 in WT. The gRNA target site is plotted along the X axis. The Y axis indicates the editing efficiencies (percent indel reads). Each dot represents an individual plant that underwent agroflood TRV delivery. c, Representative image of yellow and green progeny seedlings from a WT plant showing 67.4% somatic editing using the TRV2 Architecture_B design with gRNA4 that underwent heat-shock treatment. d, Table summarizing the transmission of edited alleles from four and one individual plants that underwent agroflood delivery using ISYmu1 gRNA4 and gRNA6, respectively. e, Representative Sanger sequencing trace file screenshots from a yellow plant harbouring an edit at gRNA4 (top) or gRNA6 (bottom). For each panel: top, wild-type sequence; middle, the ISYmu1 gRNA target and PAM; bottom, the ab1 trace file displaying a homozygous deletion.

Full size image

Next, we screened the progeny from plants infected with TRV targeting gRNA4 and gRNA6 to quantify transmission of edited alleles. Seedlings were grown on ½ MS plates containing 3% sucrose, and after 10 days we observed yellow seedlings, consistent with the biallelic mutation genotype of this gene (Fig. 3c)29. In total, we observed 3.5–8.5% yellow progeny from plants infected with TRV ISYmu1 gRNA4 (Fig. 3d). For the progeny of a plant infected with gRNA6, we observed fewer yellow seedlings, totalling 2/145 (1.4%) (Fig. 3d). Sanger sequencing revealed that all of the yellow plants harboured biallelic mutations at the gRNA4 or gRNA6 target site (Fig. 3e, Extended Data Fig. 5c and Supplementary Table 2). Next, Sanger sequencing revealed that 9/143 (6.3%) and 1/137 (0.7%) green seedlings from plants 2154_12-26_HS (gRNA4) and 2156_12-65 (gRNA6), respectively, contained monoallelic edits (Fig. 3d and Supplementary Table 2). These data indicate that TRV-mediated editing with ISYmu1 is capable of generating targeted somatic mutations at three target sites of the AtCHLl1 gene, and that edited alleles can be transmitted to the next generation.

It has been demonstrated that TRV is not transmitted to the next generation following agroflood inoculation of plants25,26. To confirm that the TRV was not present in the progeny of a TRV-infected plant, RT–PCR was performed on 5 albino plants harbouring homozygous 4-bp deletions at AtPDS3. Consistent with the literature, TRV was not detected in any of the albino plants (Extended Data Fig. 6)25,26. These data indicate that TRV-mediated biallelic edits using ISYmu1 are heritable and virus-free.

To evaluate off-target editing, we surveyed 3 individual albino plants harbouring biallelic mutations generated by ISYmu1 TRV2 Architecture_B gRNA2. Whole-genome sequencing was performed to generate an average of 770× coverage, with greater than 99% of the genome covered by mapped reads (Supplementary Table 3). In all 3 samples, we confirmed the targeted mutations in the AtPDS3 gene, as previously identified using amp-seq. In addition, we found a large number of variant differences compared with the Col-0 reference genome both in the control and the edited plants (Supplementary Table 4), suggesting that most of the variants detected are due to spontaneous mutations present in our lab strain of Arabidopsis. To screen for variants potentially caused by ISYmu1 off-target editing, all variants in the edited plants were filtered with variants already present in the control background. Variants with coverage lower than 30-fold were also filtered out. The remaining variants were checked manually for any false positive variant calling. In the 3 albino plants we sequenced, only 5, 5 and 4 variants were detected, and these variants are all outside the predicted potential off-target sites based on sequence similarity to the AtPDS3 gRNA2 sequence (Extended Data Figs. 7–10, and Supplementary Tables 4 and 5)30. In line with ISYmu1 off-target analysis reported in rice and human cells19,20, these data further demonstrate the high target-site specificity of ISYmu1.

A long-term goal of plant scientists has been the development of fast and easy means of editing plant genomes without the need for tissue culture and transgenesis. Very recently, low levels of tissue-culture-free heritable gene editing was demonstrated by delivering Cas9 and the gRNA to Nicotiana benthamiana using the tobacco ringspot virus (TRSV)31. They improved heritability by co-delivering the Cas9-gRNA TRSV with an rdr6 virus-induced gene silencing knockdown sequence on the apple latent spherical virus (ALSV)31. Here we developed a streamlined and easy-to-use approach utilizing the ultracompact site-specific TnpB genome editor, ISYmu1, together with tobacco rattle virus, for heritable plant genome editing. These results should accelerate high-throughput genome editing for both basic and applied research. We anticipate this approach to be applicable to other novel TnpBs, various viral vectors and a number of plant species for genome editing. Recent work has uncovered many TnpB systems from diverse microbial sources, including enzymes with unique protospacer adjacent motif (PAM) sequence specificities32, which can increase the range of target DNA sequences that could be edited using this approach. The TRV virus used in this study has a broad host range of over 400 species, including many solanaceous plants such as tomato, ornamental plants and other crops33. In addition, plant viruses with similar cargo capacities, such as potato virus X and barley stripe mosaic virus, are likely to be amenable to this approach since it has been demonstrated that they are capable of viral-mediated heritable gene editing by delivering the gRNA to a Cas9-expressing transgenic plant34. Further, because this approach can create sectors of tissue harbouring somatic biallelic edits, it may also serve as a tool to enable the study of genes that cause embryonic lethality or severe pleiotropic effects as homozygous mutants. Finally, in addition to being an important tool for crop biotechnology, viral delivery of TnpBs could enable high-throughput CRISPR screens in model plant species such as Arabidopsis, further unlocking their potential for genetic discovery.

Methods

Plasmids used in this study

Plasmids used for bacterial assay were generated as follows. The single expression cassette containing TnpB and ωRNA sequences were synthesized as geneblocks from Integrated DNA Technologies (IDT) and were golden-gate cloned using BsmbI restriction enzyme (E1602S) into a vector (chloramphenicol resistance) under a single tetracycline-inducible promoter (TetR/pTet) to make the TnpB-ωRNA plasmid. Target sites with various PAM sequences and target sites were golden-gate cloned with BbsI restriction enzyme (R3539S) into a vector (ampicillin/carbenicillin resistance).

Plasmids were generated for protoplast and floral dip experiments in a two-step cloning strategy. In step one, the ISDra2, ISYmu1 and ISAam1 protein coding sequences and their ωRNAs were synthesized as geneblocks by IDT. Then, starting with the pC1300_pUB10_pcoCASphi_E9t_MCS_version2 vector35, we used NEBuilder HiFi DNA Assembly (catE2621) and PCR to assemble the TnpB-ωRNA geneblocks into plant expression vectors with a toxic ccdB insert flanked by PaqCI sites immediately downstream of the ωRNA scaffold and preceding an HDV ribozyme sequence. The HiFi reactions were then transformed into One Shot ccdB Survival 2 T1R Competent Cells (A10460) to obtain the pMK003 (ISDra2), pMK025 (ISYmu1) and pMK024 (ISAam1) intermediate vectors for facile guide sequence cloning (Supplementary Table 6). In step two, guide sequences were synthesized as individual top and bottom strands with 4 base pair overhangs from IDT, phosphorylated and annealed, and then used for golden-gate assembly using the NEB PaqCI (R0745) enzyme (Supplementary Table 7). When transformed into NEB 10-beta competent E. coli (C3019), vectors that still contained the ccdB gene would kill the cells, leaving behind only the transformants possessing successfully assembled TnpB plant expression vectors harbouring a guide RNA sequence.

TRV vectors targeting AtPDS3 were created with the pDK3888 TRV2 plasmid as a base vector26. NEBuilder HiFi DNA Assembly (E2621) was used to clone the ISYmu1 gRNA2 Architecture_A, ISYmu1 gRNA2 Architecture_B and ISYmu1 gRNA12 Architecture_B into the TRV2 cargo slot. First, pDK3888 was digested using NEB ZraI (R0659), NEB PmlI (R0532) and NEB Quick CIP (M0525) overnight, and purified using Qiagen QiaQuick purification column (28104). Next, three PCR reactions were performed to amplify the fragments needed for NEBuilder HiFi DNA Assembly, followed by purification using a Qiagen QiaQuick purification column (28104) (Supplementary Table 8). Then, the digested and purified pDK3888 plasmid and purified PCR fragments were used to assemble the final TRV2 plasmid using NEBuilder HiFi DNA Assembly according to manufacturer protocol. Finally, the NEBuilder HiFi DNA Assembly reaction was transformed into NEB 10-beta competent E. coli (C3019). TRV2 vectors targeting AtCHLl1 were created using golden-gate assembly36. Two oligos corresponding to the target site were phosphorylated and annealed. Then, the annealed double-stranded DNA was used in a PaqCI (R0745S) golden-gate reaction with the pMK435 ccdb intermediate vector (Supplementary Table 8). The golden-gate reaction was then transformed into NEB 10-beta competent E. coli (C3019). Correct plasmids were confirmed using Primordium whole-plasmid sequencing. Plasmids and their descriptions can be found in Supplementary Table 6.

Bacterial interference assay

For the bacterial interference assay, we co-transfected 100 ng of the TnpB-ωRNA plasmid and 100 ng of the target plasmid to 33 µl of NEB 10-beta electrocompetent E. coli cells (C3020K). Specifically, the target plasmid contains a target site either flanking the canonical PAM (TTGAT for ISYmu1 and ISDra2 and TTTAA for ISAam1) or flanking a non-canonical PAM (GGGGG). The cells were recovered in 1 ml of NEB 10-Beta Stable/Outgrowth media (B9035S) for 1 h. Following recovery, a series of 5-fold dilutions of the recovery culture were prepared. Each dilution (5 μl) was spot plated onto LB-agar plates containing double antibiotics (34 μg ml−1 chloramphenicol, 100 μg ml−1 carbenicillin and 2 nM anhydrotetracycline) and onto control plates with a single antibiotic (34 μg ml−1 chloramphenicol and 2 nM anhydrotetracycline). If no colonies were visible on the serial dilution plates, 400 μl of the 1 ml recovery culture was plated entirely on the double antibiotic plate to enhance detection sensitivity. Plates were left overnight at either 26 °C or 37 °C, and colony-forming units (c.f.u.s) were counted on all plates the next morning. The normalized c.f.u.s were calculated by taking the ratio of c.f.u.s on the double antibiotic plates to the c.f.u.s on the single antibiotic plates. The normalized c.f.u.s in the canonical PAM conditions were compared to those in the non-canonical PAM conditions. Experiments were performed in triplicate.

Plant materials and growth conditions

For protoplast preparation, Arabidopsis Columbia ecotype (Col-0) seeds were suspended in a 0.1% agarose solution and kept at 4 °C in the dark for 3 days to stratify. Following stratification, seeds were planted on Jiffy pucks and grown under a 12-h/12-h light/dark photoperiod with low-light condition at 20 °C for 3–4 weeks22.

For the creation of transgenic plants, the Arabidopsis Col-0 ecotype was used. The ku70 (SALK_123114) genotype was obtained from Feng Zhang lab at the University of Minnesota. The rdr6 genotype was created using CRISPR-Cas9, resulting in a 616-bp deletion in the gene body of rdr6. Floral dip transformation was performed according to the protocol as previously outlined using the Agl0 Agrobacterium strain23. Transgenic T1 plants were screened using ½ MS plates with 40 μg ml−1 hygromycin B under a 16-h/8-h light/dark cycle at 23 °C. After 1 week, transgenic seedlings that passed selection were transferred to soil and moved to a greenhouse (23 °C) for the rest of their life cycle.

For agroflood experiments, sterilized seeds were sown on ½ MS agar plates and stratified for 5 days. After 5 days, the seeds were moved to a growth room and grown under a 16-h/8-h light/dark cycle at 23 °C for 8–10 days. The seedlings were then used for TRV delivery.

A subset of transgenic T1 plants and plants that underwent agroflood were subjected to a heat-shock treatment modified from ref. 37. Seedlings that passed selection or underwent agroflood were then transplanted to soil and grown in a greenhouse (23 °C) for 1 week. After 1 week, plants that did not receive a heat-shock treatment continued to grow in the greenhouse (23 °C); however, plants that underwent heat-shock treatment were exposed to 8 h (9:00–17:00) of heat exposure at 37 °C every day for 5 days, followed by 2 days of recovery at a greenhouse (23 °C). This heat-shock regime lasted for 2 weeks.

Protoplast isolation and transfection

Arabidopsis mesophyll protoplast isolation was performed as previously described22. Plasmid transfections into Arabidopsis protoplasts were performed using 20 µg of plasmid following ref. 35. The concentrations of plasmids were determined using a nanodrop spectrophotometer. Plasmids were added to the bottom of each transfection tube, and the volume of the plasmids was supplemented with water to reach 20 µl. Protoplasts (200 µl) were added, followed by 220 µl of fresh and sterile polyethylene glycol (PEG)-CaCl2 solution. The samples were mixed by gently tapping the tubes and incubated at room temperature for 10 min. After 10 min, 880 µl of W5 solution was added and mixed with the protoplasts by inverting the tube two to three times to stop the transfection. Next, protoplasts were collected by centrifuging the tubes at 100 relative centrifugal force (RCF) for 2 min and resuspended in 1 ml of WI solution. The protoplast cells were then plated in 6-well plates precoated with 5% calf serum. Protoplast cells in the 6-well plates were incubated at 26 °C for 48 h. During the 48-h incubation, the protoplast cells were subjected to a 37 °C heat-shock treatment for 2 h at 16 h post transfection. At 48 h post transfection, protoplasts were collected for genomic DNA extraction.

TRV delivery to Arabidopsis seedlings

TRV delivery was performed as previously described26. TRV1 and TRV2 vectors were first introduced into the GV3101 Agrobacterium strain. The Agrobacterium harbouring TRV vectors were then grown in 200 ml of lysogeny broth (LB) with antibiotics for 18 h at 28 °C. Agrobacterium cultures were centrifuged for 20 min at 3,500 × g. The LB was discarded and the Agrobacterium cells were resuspended in 200 ml of sterile water. The resuspended Agrobacterium was centrifuged for 10 min at 2,109 × g. The supernatant was discarded and the pellet was resuspended in sterile agro-infiltration buffer containing 10 mM MgCl2, 10 mM 2-(N-morpholino) ethanesulfonic acid and 250 µM acetosyringone to otical density (OD)600 = 1.5. The Agrobacterium cells were then incubated at 23 °C for 3 h with slow shaking. After 3 h, the Agrobacterium harbouring TRV1 and TRV2 were mixed in a 1:1 ratio, and 15 ml of this 1:1 mixture of TRV was delivered to seedlings at 8–10 days old. After 4 days of agroflood co-culture, seedlings were transplanted to soil.

Screening the progeny of TRV-infected plants for edits

Seeds were harvested from the TRV-infected plants ~12 weeks after TRV delivery. The seeds were sown on ½ MS plates supplemented with 3% sucrose and stored at 4 °C in the dark for 5 days to stratify. After 5 days, the seeds were moved to a growth room and grown under a 16-h/8-h light/dark cycle at 23 °C for 10–12 days. Next, a subset of plants was sampled for genotyping. A single piece of leaf tissue was sampled, and DNA was extracted using Invitrogen Platinum Direct PCR Universal Master Mix (A44647500) according to manufacturer instructions. The DNA was then used for amp-seq or Sanger sequencing using primers listed in Supplementary Table 9.

Next-generation amplicon sequencing

DNA was extracted from protoplast samples with Qiagen DNeasy plant mini kit (Qiagen, 69106). Tissue was collected from transgenic plants by sampling and pooling leaf tissue from 3 random leaves on a single plant 3 weeks after being transplanted to soil. For the plants that underwent agroflood, leaf tissue was sampled by collecting and pooling tissue from 3 random (however, if white or yellow sectors were visible, they were sampled) leaves on a single plant distal to the TRV delivery site 3 weeks after being transplanted to soil. Once tissue samples were collected, they were frozen at −80 °C overnight. The samples were then ground and DNA was extracted using the Invitrogen Platinum Direct PCR Universal Master Mix (A44647500) according to manufacturer instructions. For the progenies of plants that underwent agroflood, a single leaf tissue was sampled and DNA was extracted using Invitrogen Platinum Direct PCR Universal Master Mix (A44647500) according to manufacturer instructions. The DNA was then used for next-generation amplicon sequencing.

Following ref. 35, editing efficiency was characterized using single-end next-generation sequencing on the Illumina NovaSeqX platform. Libraries were prepared via a 2-step PCR amplification method. In the first round of amplification, each target site was amplified using primers flanking the target site (Supplementary Table 9). After 25 cycles of amplification, the reactions were cleaned using 1.0× Ampure XP bead purification (Beckman Coulter, A63881). Next, each sample went through 12 additional cycles of amplification using Illumina indexing primers. The samples were cleaned using 0.7× Ampure XP bead purification. Samples were checked for purity on a 2% agarose gel, quantified using a nanodrop spectrophotometer, normalized and pooled.

Next-generation amplicon sequencing analysis

Amplicon sequencing analysis was performed following ref. 35. Single-end reads were used for analysis. Reads were adapter trimmed using Trim Galore default settings. Remaining reads were mapped to the target genome region using the BWA aligner (v.0.7.17, BWA-MEM algorithm). Sorted and indexed bam files were used as input files for further analysis using the CrispRvariants R package (v.1.14.0). Each mutation pattern with corresponding read counts was exported using the CrispRvariants R package. After assessing all control samples, a criterion to classify reads as edited was established: only reads with a ≥3-bp deletion or insertion (indel) of the same pattern (indels of same size starting at the same location) with ≥10 read counts from a sample were counted as edited reads. Single nucleotide variants were also filtered out.

Off-target analysis

Off-target analysis was performed as previously described35. DNA from single Arabidopsis seedlings was extracted with the Qiagen DNeasy plant mini kit and sheared to 300-bp size with a Covaris sonicator. Library preparation was performed with a Tecan Ovation Ultralow V2 DNA-seq kit. For variant calling, WGS reads were aligned to the TAIR10 reference genome using BWA mem (v.0.7.17)38 with default parameters. GATK (4.2.0.0)39 MarkDuplicatesSpark was used to remove PCR duplicate reads. Then GATK HaplotypeCaller was used to call raw variants. Raw single nucleotide polymorphisms (SNPs) were filtered with QD < 2.0, FS > 60.0, MQ < 40.0 and SOR > 4.0. Raw InDels were filtered with QD < 2.0, FS > 200.0 and SOR > 10.0 and used for base quality score recalibration. The recalibrated bam was further applied to GATK and Strelka (v.2.9.2) SNPs/InDel calling. Only SNPs/InDels called by both GATK and Strelka were used for further filtering. The intersection of SNPs/InDel called by GATK with Strelka (v.2.9.2)40 was obtained using BedTools (v.2.26.0)41. SNPs/InDel were filtered with wild-type background using BedTools (v.2.26.0). Variants with depth coverage lower than 30 were filtered.

RT–PCR

Total RNA from TRV-infected progeny plants was extracted using Zymo Research Direct-zol RNA MiniPrep kit (R2052). Total RNA was converted to cDNA using the Invitrogen SuperScript IV VILO Master Mix (11766050). The RT–PCR control was performed using primers targeting the AtIPP2 gene (Supplementary Table 10). PCR was performed to check for the presence/absence of the TRV vector using SP9238 and SP9239 (Supplementary Table 10)26. PCR was performed with New England Biolabs Q5 High-Fidelity 2× Master Mix (M0492L) according to manufacturer instructions, using 2 µl of cDNA in a 25-µl reaction. PCR conditions included a 98 °C initial denaturation step for 30 s, 35×(98 °C, 10 s; 55 °C, 20 s; 72 °C, 10 s) and 72 °C for 2 min. PCR amplicons (10 µl) were analysed using 2% agarose gel electrophoresis.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.