Development of an ultra-efficient prime editing system in tomato

Abstract

Prime editing (PE) enables precise genome modifications without donor DNA or double-strand breaks, but its application in dicot plants has faced challenges due to low efficiency, locus dependence, and poor heritability. Here, we develop an ultra-efficient prime editing (UtPE) system for dicots by integrating evolved PE6 variants (PE6c and PE6ec), an altered pegRNA (aepegRNA), an RNA chaperone, and a geminiviral replicon. UtPE significantly improves editing performance in tomatoes, with UtPEv1 excelling in simple edits (unstructured RTTs) and UtPEv3 effective for complex targets (structured RTTs or multiple nucleotide changes). Compared to a PE2max-based tool, UtPE increases desired editing efficiency by 3.39 to 8.89-fold, enables editing at previous inaccessible sites, achieves an average of 16.0% desired editing efficiency in calli, and produces high-frequency desired edits in up to 87.5% of T0 plants. Multiplexed editing at up to three loci and stable T1 inheritance are also achieved, resulting in traits such as jointless pedicels and glyphosate resistance, while minimizing off-target effects.

Low efficiency, locus-dependence, and poor heritability affect the application of prime editing to dicot plants. Here, the authors solve these problems by developing an ultra-efficient prime editing (UtPE) system for dicots by integrating evolved PE6 variants, an altered pegRNA, an RNA chaperone, and a geminiviral replicon.

Introduction

Prime editing (PE) technology¹ enables the precise installation of desired edits at genomic sites in monocots with high efficiency^2–20. PE applications in dicot plants faced more challenges than in monocots, as various attempts resulted in very low efficiency, and no heritable PE event was observed^21–25. Recently, we developed a dicot PE system that also enabled efficient and heritable PE in dicots, such as tomato and Arabidopsis²⁶. Around the same time, the PE tool was also applied to engineering source-sink relations in tomato²⁷, Arabidopsis^28,29, and tobacco²⁹. However, the PE tools have still mainly been locus-dependent and demanded substantial improvement for routine applications for dicot breeding. In our previous work, only six out of eleven sites in tomato exhibited moderate to high desired PE efficiency, and heritable PE events were not achieved at some targets, such as SlHKT1;2, SlEPSPS1, SlCENH3(sub), and SlDMR6²⁶. Therefore, we sought to develop a more efficient and locus-independent PE system for PE-inactive sites, thereby enabling the routine application of PE tools in dicots.

Recently, Liu’s group developed PE6 variants based on the evolution of the PEmax tool and a Tf1 reverse transcriptase (RT), which significantly improved PE efficiency in mammalian cells³⁰. The PE6c and PE6d variants have been shown to be more efficient than the PEmax system in rice using T-DNA with Agrobacterium-mediated transformation (AMT)^31,32. However, none of the PE6 variants have been adapted in dicots. Moreover, there were various PE6 variants such as PE6e, PE6f, PE6g, and combinations of the evolved nCas9 and Tf1 (PE6ec, PE6fc, and PE6gc)³⁰, which showed different PE characteristics and capacities in human cells, but those variants have not been evaluated in plants.

In our previous work, we harnessed the PEmax protein, engineered it with an altered engineered PE guide RNA (aepegRNA) system^26,33, and enhanced it with a nucleocapsid (NC) RNA chaperone¹⁴. Together, these components delivered by a geminiviral replicon vector broke the PE efficiency barrier in tomato and Arabidopsis. We hypothesize that a hyper-efficient and locus-independent PE system should inherit all the best and updated features of the current PE components.

In this work, we integrate and evaluate the most efficient and potential PE6 variants, such as PE6c and PE6ec, with or without the NC chaperone, within our recently developed dicot PE system (Fig. 1a), at multiple tomato targets. Ultimately, we develop an ultra-efficient PE (UtPE) system to improve desired PE efficiency. The UtPE system works very efficiently at all the tested PE-inactive sites and may thus enable locus-independent PE in tomato and beyond.

Fig. 1. System design and initial assessment of the UtPE system in tomato.

a The UtPE system design. The UtPE variants (1, 2, and 3) were expressed using a CaMV promoter and EURb7 terminator and guided by the aepegRNA produced by the pU6cm system. The PE protein and aepegRNA expression cassettes were delivered into plant cells via a super Agrobacterium strain-mediated transformation and amplified by the geminiviral replicon. Created in BioRender https://biorender.com/6kvi9l3. b All the PE6 variants, their derivatives, and engineered variants with the NC peptide that were tested in this study. The bottom panel details the characteristics of the aepegRNA. Created in BioRender https://biorender.com/9kse9o7. Comparison of the PE efficiency of PE6c, PE6d, and their NC fusion variants with PE2max (c) as well as between PE6c and PE2max-NC (d). The pegRNA was designed to insert the 5’-AAATTGTGA-3’ sequence into the 5’ UTR region of SlCAB13. In c, d tomato cotyledon explants were transformed using the PE tools. After the co-cultivation stage, the explants were incubated under the T34 incubation regime (as detailed in the “Methods” section). Callus samples were collected at 16 days post-transformation and analyzed using targeted deep sequencing. The experiments were conducted in triplicate (n = 3). The “imprecise edits” is the sum of “modified prime edited portion” and “prime edited portion with scaffold insertion”. Statistical analysis was performed using GraphPad Prism 9 with a two-sided, uncorrected Fisher’s LSD test. All data points are displayed in the plots, and the error bars indicate ± SEM. Source data are provided as a Source Data file.

Results

Integrating the PE6 protein variants into the dicot PE system

All the PE6 variants, their derivatives, and engineered PE proteins with the C-terminal fusion of a nucleocapsid (NC) RNA chaperone peptide tested in this study were co-expressed with an aepegRNA cassette and delivered by  a geminiviral replicon system (Fig. 1a and b, and Supplementary Fig. 1). The aepegRNA scaffold contains a stabilized small hairpin³⁴ and the optimized ‘flip and extension’ (F + E) tetraloop³⁵, and it is protected by a trimmed evopreQ (tevopreQ1) loop³³ at its 3’ end (Fig. 1b). The transcription of aepegRNA was driven by a U6 composite promoter (pU6cm), a polII-polIII fusion promoter³⁶, and a double terminator (EURb7). The EuRb7 terminator is an enhanced terminator that contains a genomic insulator (Rb7), which strongly enhanced gene expression (>160-fold increase compared to a nopaline synthase (NOS) terminator) and was further enhanced when combined with a geminiviral replicon³⁷. The double terminator was also used to regulate the transcription of the PE6 protein variants together with a long CaMV promoter (Fig. 1a and b, and Supplementary Fig. 1). The PE system, when delivered by an EHA105-based super Agrobacterium version 2 (EHA105spv2)-mediated transformation of the geminiviral replicon, plays a critical role in boosting pegRNA transcription and PE protein production²⁶ and ultimately constitutes an ultra PE (UtPE) system (Fig. 1a).

UtPEv1 outperformed PE6d and the previous dicot PE system in tomato

Recently, PE6c and PE6d variants were adopted in rice with various performances for base substitutions³¹ and in-locus protein tagging³². Therefore, we initially integrated the PE6c and PE6d protein variants (Fig. 1b) to precisely insert nine nucleotides (5’-AAATTGTGA-3’) at the SlCAB13 site (Supplementary Table 1) that exhibited moderate desired PE efficiency in our recent study²⁶. Analyzing callus explants by targeted deep sequencing, we found that the PE6c variant outperformed the PE2max and PE6d variants (Fig. 1c). The highest average desired PE efficiency reached 11.74 ± 1.74% using PE6c compared to 2.95 ± 0.57% for PE2max, representing a 3.98-fold enhancement (Fig. 1c and Supplementary Table 2). PE6c slightly increased the desired/imprecise edit ratios and did not increase indel mutations compared to PE2max (Fig. 1c and Supplementary Table 2), similar to the observation in human cells³⁰. Meanwhile, the PE6d variant was just slightly better than PE2max with a 1.29-fold improvement.

The processivity of RT enzymes may be enhanced by an RNA chaperone that facilitates the destabilization of stem loops formed by a complexed RTT¹⁴. Indeed, we previously employed the NC RNA chaperone to enhance PE activity by fusing it to the C-terminus of PE2max in tomato and Arabidopsis²⁶. Unexpectedly, when we added NC to the C-terminal of PE6c, the PE efficiency was only comparable to that of PE6c alone (Fig. 1c and Supplementary Table 2). In contrast, adding NC at the C-terminal of PE6d significantly improved its performance (Fig. 1c and Supplementary Table 2). The PE6c variant also outperformed PE2max-NC with 1.35-fold higher desired PE efficiency, reaching up to 16.03 ± 1.34% at the SlCAB13 site (Fig. 1d and Supplementary Table 3). The system using the PE6c and PE6c-NC variants is designated as UtPEv1 and UtPEv2, respectively (Fig. 1a).

To further validate the UtPEv1 activity, we employed UtPEv1 to install the desired edits at six additional sites (Supplementary Tables 1 and 4) that had previously shown nil to moderate PE activity²⁶. In the callus stage, we found that UtPEv1 outperformed the previous PE2max-based dicot PE system at all tested sites, with a 3.39- to 8.89-fold increase (Fig. 2a and Supplementary Table 4). Interestingly, though the PE2max-based dicot PE system did not work at the SlDMR6 site, UtPEv1 exhibited 8.09 ± 0.25% desired PE efficiency, albeit with high imprecise PE (mostly scaffold inserted) and indel mutation rates (Supplementary Table 4). The desired/imprecise edit ratio of UtPEv1 was only reduced at inactive sites, such as SlCENH3(del) and SlHKT1;2, compared to PE2max. At moderate to highly active sites, UtPEv1 even slightly improved the desired/imprecise edit ratio while the indel mutation rate was comparable (Supplementary Table 4). Among the sites, the desired PE efficiency reached an average of 14.86 ± 0.47%. Moreover, except for the SlHKT1;2 site, all the others resulted in more than 2% of the desired PE efficiency at the callus stage (Fig. 2a), a practical level for obtaining PE events carrying the desired edits at high frequencies.

Fig. 2. Development of the UtPE tool for locus-independent prime editing in tomato.

a Utilizing UtPEv1 for precise installation of desired edits at six additional sites, compared to PE2max. b Enhancement of desired PE efficiency at the SlHKT1;2 site using UtPEv1, UtPEv2, PE6ec, PE6fc, PE6gc, and their NC fusion variants. c Comparison of the desired PE efficiency of UtPEv1, UtPEv2, and UtPEv3 at SlOR, SlCENH3(sub), and SlCAB13 sites. d The desired/imprecise edit ratios of UtPEv1, UtPEv2, and UtPEv3 at the four tested sites. e Potential off-target analysis of UtPEv1 in comparison with PE2max. The callus explants subjected to the analysis were the same in (a). f Potential off-target analysis of UtPEv1, UtPEv2, and UtPEv3 in comparison with PE2max. The callus explants subjected to the analysis were the same in (c, d). In (a–c), tomato cotyledon explants were transformed using the PE tools. After the co-cultivation stage, the explants were incubated under the T34 incubation regime (as detailed in the “Methods” section). Callus samples were collected at 16 days post-transformation and analyzed using targeted deep sequencing. The experiments were conducted in triplicate (n = 3). The “imprecise edits” is the sum of “modified prime edited portion” and “prime edited portion with scaffold insertion”. In (d), the desire/imprecise edit ratios were calculated by dividing the desired PE efficiency by the imprecise efficiency of the PE tools. Statistical analysis was performed using GraphPad Prism 9 with a two-sided, uncorrected Fisher’s LSD test. All data points are displayed in the plots, and the error bars indicate ±SEM. Source data are provided as a Source Data file.

UtPEv3 improved desired PE efficiency at the sites with more complex RTTs

The desired edit to be installed at the SlHKT1;2 site is complex with three desired nucleotide changes (+10 A to G; +19 A to G; and +21 T to C), including synonymous mutations (Supplementary Table 1), forming a predicted stem loop within its RTT strand (Supplementary Fig. 2) and may require a more robust PE variant for installing it. The PE6c and PE6d were developed based on the combinations of nCas9 from PEmax and evolved Tf1 RT and MMLV RT, respectively³⁰. While PE6c and PE6d were well characterized in human cells and recently adapted to monocots^31,32, the evolved nCas9 configurations of PE6e-g, which showed potential PE improvement in human cells, have not yet been tested elsewhere. In this experiment, we engineered and tested PE6ec (combined evolved nCas9 of PE6e and evolved Tf1 RT of PE6c), PE6fc (combined evolved nCas9 of PE6f and evolved Tf1 RT of PE6c), and PE6gc (combined evolved nCas9 of PE6g and evolved Tf1 RT of PE6c) (Fig. 1b) for improving PE efficiency at the SlHKT1;2 site. However, none of the combinations could enhance the activity of the PE tools as UtPEv1 still resulted in the highest desired PE efficiency at the target site (callus stage) (Fig. 2b). Except for the PE6ec variant, all the other combinations of the evolved nCas9s and Tf1 RT resulted in a dramatic reduction of PE performance compared to UtPEv1 (Fig. 2b). Simultaneously, we fused the RNA chaperone NC to the C-terminal of PE6ec, PE6fc, and PE6gc variants (Fig. 1b) and assessed its impacts on their activity. The addition of NC could enhance the desired efficiency of PE6ec, resulting in a 1.37-fold increase over UtPEv1. Similarly, UtPEv2 was slightly better (1.26-fold increase) than UtPEv1 (Fig. 2a and Supplementary Table 5). The NC chaperone also led to more imprecise edits as the desired/imprecise edit ratios reduced in the cases of UtPEv2 and PE6ec-NC (Supplementary Table 5). Interestingly, the PE6ec-NC variant showed the highest desired PE efficiency, 1.76 ± 0.23%, at this site. It is designated as UtPEv3 (Fig. 1a).

Since the UtPEv1, UtPEv2, and UtPEv3 exhibited comparable desired PE efficiency at the SlHKT1;2 site, we assessed their activities at three more targets (SlOR, SlCENH3 (sub), and SlCAB13) and found that the desired PE efficiency was almost comparable among the UtPE tools at the callus stage (Fig. 2c). The desired PE efficiency was highest with UtPEv1 at SlOR (9.81 ± 0.76%) and SlCAB13 (14.68 ± 1.14%) and with UtPEv2 at SlCENH3(sub) (1.94 ± 0.53%) site (Fig. 2c and Supplementary Table 6). Moreover, the UtPE variants exhibited similar desired/imprecise edit ratios (Fig. 2d) and indel mutation rates (Supplementary Tables 5 and 6) compared to the PE2max-based dicot PE system at the four tested sites.

The UtPE system shows low gRNA-dependent off-target activity

Since the activity of the UtPE variants is significantly higher than that of PE2max, we wonder if they could induce higher pegRNA-dependent off-target activity. We employed the Cas-OFFinder³⁸ for identifying potential off-target sites of the gRNAs in tomato (Supplementary Data 1). Assessing two potential off-target sites per pegRNA revealed no off-target traces of PE editing and only background levels of indel mutation frequency from the UtPE and PE2max systems (Fig. 2e and f), as also observed earlier²⁶.

The tRNA-HDV-based UtPE outperformed the Csy4-aided UtPE

The Csy4 protein (also known as Cas6f) is the key enzyme responsible for crRNA production in CRISPR subtype I-F³⁹. Csy4 bound with equal affinity to both its pre-crRNA substrate and Csy4 stem-containing 3’ end of the processed crRNA³⁹, raising the possibility of 3’ end protection by Csy4 protein^39,40 (Fig. 3a). The Csy4-processing system was employed to support pegRNA and sgRNA production in human cells^41,42, wheat⁴³, rice⁴⁰, and tomato²⁷. We compared the Csy4-aided aepegRNA processing system with the tRNA-HDV-based aepegRNA processing (Fig. 1a) UtPE tool by replacing the tRNA-HDV sequences with the Csy4 recognition sequence (Csy4RS) and co-expressed Csy4 protein together with the PE tool (Fig. 3b). The Csy4-UtPEv1-containing construct has a slightly higher size (~2.3 kb) than the UtPEv1-delivering replicon thanks to the addition of the expression cassette of Csy4 protein (Fig. 3b). Unexpectedly, our data revealed that the Csy4-assissted UtPEv1 system showed significantly lower PE efficiency (0.4-0.86-fold lower) compared to the UtPEv1 tool at all three tested sites (Fig. 3c and Supplementary Table 7).

Fig. 3. Csy4-aided and multiplexed UtPE.

a Expression cassette of Csy4-altered epegRNA and Csy4-processed product. The tRNA and HDV sequence of the altered pegRNA was replaced by the Csy4 recognition sequences (Csy4RS). After transcription, the Csy4 protein processed the Csy4RS-altered epegRNA-Csy4RS into altered epegRNA-Csy4RS with the Csy4 protein stably bound to the Csy4RS stem at the 3’ terminal of the processed RNA molecule. Created in BioRender https://biorender.com/3amip9l. b Construct maps of the Csy4-aided UtPE and multiplexed UtPE. The Csy4-aided UtPEv1 was evaluated in comparison with the UtPEv1 at the SlOR, SlCAB13, and SlPRD sites. The Csy4-aided UtPEv1 was included with the Csy4-altered epegRNA and Csy4 expression cassette (Supplementary Data 2). The multiplexed constructs were designed for simultaneous editing at two (UtPEv1-MP1) and three (UtPEv1-MP2 and UtPEv3-MP2) targets. The figure was created using Microsoft PowerPoint 365. c Desired PE efficiencies of the UtPEv1 and Csy4-aided UtPEv1 tools assessed by targeted deep sequencing. d Desired PE efficiencies of the UtPEv1 and the multiplexed tools assessed by targeted deep sequencing. In (c, d), tomato cotyledon explants were transformed using the PE tools. After the co-cultivation stage, the explants were incubated under the T34 incubation regime (as detailed in the Methods section). Callus samples were collected at 16 days post-transformation and analyzed using targeted deep sequencing. The experiments were conducted in triplicate (n = 3). Statistical analysis was performed using GraphPad Prism 9 with a two-sided, uncorrected Fisher’s LSD test. All data points are displayed in the plots, and the error bars indicate ±SEM. Source data are provided as a Source Data file.

The UtPE tool enabled efficient multiplexing PE in tomato

Efficient and simultaneous editing of multiple targets at high precision is highly desirable for engineering traits that are linked to various genes/alleles⁴⁴. Our previous version of the dicot PE tool was shown to be feasible for multiplexed PE, with slightly reduced PE efficiencies compared to single-PE tools²⁶. We tested whether the UtPE tool could also enable multiplexed PE. UtPEv1 and UtPEv3 were employed to simultaneously install desired edits at two (SlOR and SlCAB13) and three (SlOR, SlCAB13, and SlPRD) targets (Fig. 3b). Via NGS, we found that the UtPEv1 and UtPEv3 tool showed comparable desired PE efficiencies with the single UtPE construct and UtPEv3 was slightly better than UtPEv1 in multiplexed PE with three targets (Fig. 3d and Supplementary Table 7).

The UtPE system efficiently produced the desired PE events at all the tested sites

Screening T0 transformant revealed potential events carrying desired PE edits from all the targets with the UtPE tools, with 5/8 sites resulting in events carrying homozygous desired edits (Table 1, Supplementary Figs. 3 and 4). The desired PE efficiency of the UtPE tools at the plant stage varied between 2.9 and 87.5%, depending on the targets, and is much higher compared with PE2max and PE2max-NC (Table 1). Interestingly, UtPEv3 exhibited the highest desired PE efficiency levels at the sites with longer and multiple edit-containing RTTs such as SlCAB13 (87.5%) and SlHKT1;2 (6.5%), while UtPEv1 and UtPEv2 resulted in more events without containing scaffold-inserted alleles for the sites with simpler RTTs like SlOR (33.3%) and SlCENH3(sub) (5.9%), respectively (Supplementary Figs. 3 and 4, and Table 1). We could then validate most of the representative PE events by targeted deep sequencing (Supplementary Table 8). At the SlHKT1;2 target, while no event was obtained with UtPEv2 and two potential events with UtPEv3 (Supplementary Fig. 3 and Table 1), only potential events generated by UtPEv3 could be confirmed to carry a high desired PE frequency (Supplementary Table 8). This result may be due to the low desired edit frequency (less than 15%) of both the potential events obtained with UtPEv1, whereas those obtained with UtPEv3 showed a ~ 40% desired edit frequency in the ICE Synthego screening.

Table 1.

Desired PE efficiency of the UtPE tools assessed at the T0 plant stage

No.	Target	PE variant	Total no. of analyzed plantsa	No. of plants with desired PE				No. of plants contained scaffold-inserted allelesc	No. of plants contained indel	No. of plants contained WT allele	No. of desired PE events	Desired PE efficiency (%)d
				Hob	He	Chi	Total
1	SlOR	PE2max	33	0	6	0	6	1	0	27	5	15.2
2		UtPEv1	30	5	7	5	17	7	1	13	10	33.3
3		UtPEv2	20	1	4	1	5	2	2	14	3	20.0
4		UtPEv3	30	3	8	5	16	8	8	12	8	26.7
5	SlCAB13e	PE2max	30	0	2	2	4	-	4	22	4	13.3
6		PE2max-NC	25	1	7	1	9	-	4	13	9	36.0
7		UtPEv1	33	1	13	4	18	-	12	11	18	54.5
8		UtPEv2	26	2	9	2	13	-	12	7	13	50.0
9		UtPEv3	32	8	16	4	28	-	8	2	28	87.5
10	SlCENH3 (del)	PE2max	30	0	0	0	0	-	0	30	0	0.0
11		UtPEv1	23	0	1	1	2	-	0	21	2	8.7
12	SlCENH3 (sub)	PE2max	33	0	0	0	0	0	0	33	0	0.0
13		UtPEv1	34	0	0	1	1	0	0	33	1	2.9
14		UtPEv2	34	0	0	2	2	0	0	32	2	5.9
15		UtPEv3	32	0	0	3	3	2	0	29	1	3.1
16	SlHKT1;2	PE2max	31	0	0	1	1	0	0	30	1	3.2
17		UtPEv1	33	0	0	2	2	-	0	31	2	6.1
18		UtPEv2	31	0	0	0	0	-	1	31	0	0.0
19		UtPEv3	31	0	0	2	2	0	2	27	2	6.5
20	SlEPSPS1	PE2max	34	0	6	1	7	1	0	27	6	17.6
21		UtPEv1	37	1	5	4	10	0	0	28	10	24.3
22	SlMBP21	PE2max	37	0	0	2	2	1	0	35	1	2.7
23		UtPEv1	27	6	1	10	17	0	0	10	17	63.0
24	SlDMR6	PE2max	30	0	0	0	0	0	0	30	0	0.0
25		UtPEv1	34	1	1	1	3	0	5	27	3	8.8
26	SlPRD	PE2max	20	0	1	2	3	-	0	17	3	15.0
27		UtPEv1	22	0	8	1	9	-	0	13	9	40.9

^aTotal no. of analyzed plants’ might be smaller than the sum of ‘No. of plants with desired PE’, ‘No. of plants contained indel’, and ‘No. of plants contained WT allele’. An analyzed plant may contain both desired PE edits and imprecise edits or indels.

^bHomozygous (Ho): 95–100%; Heterozygous (He): 45-94%; Chimeric (Chi): 10–44%; WT: <10%.

^cHomozygous events of the SlOR target containing 100% desired edits with partial (<50%) scaffold insertion were counted as events with desired edits.

^dDesired PE efficiency (%) = 100*(No. of desired PE events/total number of analyzed plants).

^eScaffold insertion was not found in homozygous events. “-” means the scaffold insertion cannot be determined with the chromatograms.

Subsequently, in our screening for transgenes, we identified two out of sixteen representative T0 events that are likely free of replicon and show very low T-DNA amplification (Supplementary Fig. 5). These data indicate that the developed UtPE system is hyper-efficient and could enable locus-independent PE in tomatoes. We recommend using UtPEv1 for targets with simple RTTs and UtPEv3 for sites with long and complex RTTs.

The desired PE alleles installed by the UtPE system were stably inherited and caused phenotypic changes

It is essential to verify that the desired PE edits installed by the UtPE system are heritable and produce the desired phenotypic changes. We, therefore, assessed the inheritance of the desired PE edits in the next generation of the PE events generated by the UtPE tools. We identified the desired PE alleles in T1 plants from representative T0 events, in both homozygous and heterozygous forms (Supplementary Table 9). We found that plants with chimeric scaffold-inserted alleles or indels did not pass these alleles to the next generation, indicating they were likely temporary changes. In contrast, T0 plants with a mix of scaffold-inserted or indel alleles did pass these alleles to the T1 generation, following Mendel’s laws of inheritance. Among the homozygous T1 plants, we observed a relatively high frequency of plants lacking T-DNA and replicon amplification (18 out of 42 analyzed plants) (Supplementary Fig. 6). The T1 plants carrying homozygous KO allele pairs of SlMBP21 exhibited the jointless phenotype (Fig. 4a), as reported earlier⁴⁵. In the case of plants carrying the homozygous SlEPSPS1 (TIPS) allele, 5 mM glyphosate resistance was observed (Fig. 4b). However, it appears that the TIPS protein variant has weaker catalytic activity in the shikimate pathway⁴⁶, leading to growth retardation in the plants. Further enhancement of SlEPSPS1 (TIPS) expression by promoter engineering may result in glyphosate tolerance and normal growth as observed in overexpressing EPSPS1 (TIPS) driven by a strong promoter in rice⁴⁷ and cassava⁴⁸. Assessing the haploid induction efficiency of the SlCENH3(del) and SlCENH3(sub) alleles involved crossing T1 plants that carried homozygous forms of these alleles with hairless/trichomeless plants (Supplementary Fig. 7a). This approach revealed some potential haploid F1 progenies that exhibited reduced hair on their stems (Supplementary Fig. 7b). However, upon sequencing the SlCENH3 gene, it became evident that all potential F1 progenies were either chimeric or heterozygous (Supplementary Fig. 7c and Supplementary Table 10). This finding indicates that haploid induction in these progenies was either unsuccessful or incomplete.

Fig. 4. Phenotypic assessment of T1 PE plants carrying homozygous desired edits.

a The jointless phenotype appeared in T1 plants with homozygous knockout (KO) alleles of SlMBP21, generated using the UtPEv1-based tool. The red arrows denote the pedicel abscission zones. b Glyphosate tolerance conferred by T1 plants that carry homozygous pairs of the SlEPSPS1(TIPS) alleles. Eight-leaf stage tomato seedlings, carrying homozygous SlEPSPS1 (TIPS) and wild type (WT), were sprayed twice at five-minute intervals with a 5 mM glyphosate isopropylamine (Farmhannong, Seoul, Korea) solution containing 0.02% SilwettL-77. The sprayed plants were photographed one and two weeks after spraying. The experiment was performed twice with consistent outcomes.

Discussion

In this study, we developed and validated an ultra-efficient PE (UtPE) system that addresses key limitations of existing PE tools in dicots^{21,23,24,26–29}. The UtPE system significantly enhances PE efficiency in dicot plants, addressing common challenges like low editing rates, locus dependence, and limited heritability. By combining evolved PE6 protein variants (PE6c and PE6ec) with a stabilized aepegRNA scaffold, the NC RNA chaperone, and a geminiviral replicon delivery system, UtPE integrates several enhancements into a single platform (Fig. 1a and b). This study adapts and evaluates PE6 variants in dicots, achieving improved editing efficiency of up to 8.89-fold at multiple loci, including previous unresponsive targets such as SlHKT1;2, SlCENH3, and SlDMR6 (Fig. 2, Supplementary Fig. 3 and Table 1). In addition, the UtPE tool enables efficient multiplexed PE with three tested sites (Fig. 3d). Furthermore, the system enabled the recovery of heritable edits in up to 87.5% of T0 transformants, a level that has not been achieved before in tomato and other dicots^22,26–29. More importantly, the specific genotypic changes were linked to observable phenotypic traits in homozygous T1 plants (Fig. 4). These findings indicate that UtPE may be a valuable platform for enhancing the application of PE in improving dicot crops.

This study unveils three groundbreaking contributions to PE in dicot plants: (1) it marks the pioneering systematic evaluation of PE6-derived editors (PE6c, PE6d, PE6ec, PE6fc, and PE6gc) and their fusion with the NC RNA chaperone in a dicot context; (2) it reveals the dynamic interplay of component interactions, demonstrating that NC fusion can enhance or hinder activity based on the PE6 backbone, while the co-expression of Csy4 offers insights into optimizing high-expression environments; and (3) it inspires practical tool selection, guiding the choice between UtPEv1 and UtPEv3 based on the complexity of the targeted region. These discoveries empower us to deepen our understanding and expand the potential for PE in dicots. It is important to note that UtPE is not a one-size-fits-all solution. Instead, it serves as an adaptable toolkit that allows for the fine-tuning of PE performance based on the characteristics of specific target loci. Among the UtPE variants, UtPEv1 and UtPEv3 differ in their PE protein configurations and target specificity. UtPEv1, based on PE6c, performed best at loci with simple RTTs, offering high efficiency and lower imprecise editing. The performance was further enhanced by UtPEv3, which combined evolved nCas9 (PE6e) with Tf1 RT and an NC RNA chaperone, particularly benefiting targets with complex RTT structures, such as SlHKT1;2 and SlCAB13. The more robust activity of UtPEv3 in dealing with complex RTT strands also led to higher scaffold-incorporated alleles besides desired edits of unstructured RTTs in the cases of SlOR and SlCENH3(sub) (Supplementary Fig. 3 and Table 1). While both are effective, UtPEv3 is better suited for editing sites with structured RTTs or multiple nucleotide changes. These increases in editing efficiency did not result in detectable off-target activity, underscoring the precision of the UtPE system. The Csy4-aided UtPEv1 tool showed reduced PE efficiency across all tested targets (Fig. 3c), likely due to Csy4 overaccumulation, which causes cytotoxic effects, as previously observed in human⁴¹ and plant systems⁴⁹. Because the UtPE tool uses a geminiviral replicon that enhances expression of pegRNA and PE components²⁶, it may also elevate Csy4 levels, leading to cell death and poor explant growth compared to the non-Csy4 UtPE system.

It is essential to recognize that the identification of the replicon in non-transgenic plants provides valuable insights into plant regeneration. Research indicates that the geminiviral replicon can persist as an episome during the early stages of regeneration, even after the loss of integrated T-DNA^50,51. This provides a compelling explanation for why we occasionally observe replicon signals in T-DNA–negative plants (Supplementary Figs. 5 and 6). Moreover, crucial findings reveal that T-DNA integration is not a prerequisite for achieving genome-edited plants⁵¹. Consistent with the findings of Cermak et al.⁵¹, T-DNA serves primarily as a delivery vehicle for viral replicons, implying that its integration into the genome is unnecessary. The persistence of replicons in T0 and T1 plants in our study likely reflects differences in construct design compared to earlier HDR systems. Unlike the replicon described by Cermak et al.⁵¹, the UtPE replicon lacks the opposing promoters that can trigger strong RNAi-mediated silencing, a factor known to accelerate the loss of replicons^26,50. Given that the coat protein is deleted, the replicon is non-infectious; thus, its detection suggests high episomal stability rather than active viral infection or genomic integration. Additional investigation may be necessary to fully elucidate the mechanism of replicon persistence in T-DNA–free plants.

The system’s capability to create stable, heritable edits was validated by recovering T1 plants that carried the desired genetic modifications and displayed expected phenotypic traits, such as jointless development (SlMBP21 knockout) and glyphosate resistance (SlEPSPS1 (TIPS)). The latter example underscores a potential trade-off between introducing new traits and maintaining plant fitness, which could be addressed through further optimization of gene expression levels. We generated the SlCENH3(del) and SlCENH3(sub) alleles; however, our preliminary crossing experiments did not produce true haploid progenies. The chimeric or heterozygous F1 individuals suggest that haploid induction was incomplete or inefficient in tomatoes. Similar findings in Arabidopsis show that only specific alterations in the CENH3 histone fold domain yield stable haploid inducers⁵². This likely indicates that our tomato CENH3 variants retain some centromere function, hindering complete genome elimination during fertilization. Further optimization, such as domain-specific replacements or conditional knockdowns, may be needed for effective haploid induction in solanaceous crops. Nonetheless, UtPE’s ability to introduce these precise modifications shows promise for studying centromere biology and developing haploid-inducing genotypes in dicots. Furthermore, the next frontier for plant PE involves engineering cis-regulatory elements, in-locus protein tagging, precise sequence insertion, and the creation of gain-of-function alleles to precisely modulate trait expression without altering gene structure, thereby supporting gene function studies. The modular UtPE platform enables targeted tuning of promoters, enhancers, and protein-coding regions, expanding PE’s impact from basic gene studies to precision crop design. Taken together, the UtPE platform represents a significant advancement in enhancing the efficiency and reliability of PE in dicot plants. Its strong performance across various loci indicates potential for wider applications in trait development. Future research should focus on evaluating its versatility among different species and exploring integration with tissue culture-free delivery methods to maximize its impact on precision plant breeding.

Methods

System design and plasmid cloning

The PE2max and PE2max-NC variants were cloned from a plasmid of our published work²⁶ (Addgene plasmid #227693). The PE6c and PE6d variants were cloned according to the published work of Doman and coworkers³⁰ using Addgene plasmids (#207853 and #207854, respectively). The PE6ec, PE6fc, and PE6gc were designed and cloned by combining nCas9 from PE6e (Addgene plasmid #207855), PE6f (Addgene plasmid #207856), and PE6g (Addgene plasmid #207857), respectively, with the evolved Tf1 RT (evoTf1: P70T, G72V, S87G, M102I, K106R, K118R, I128V, L158Q, F269L, A363V, K413E, S492N, K118R, S188K, I260L, S297Q, R288Q) of the PE6c plasmid (see Supplementary Data 2). The RNA chaperone NC (from pH-ePPE, Addgene plasmid #183097) was fused to the C-terminal of PE6 variants to generate PE6c-NC, PE6d-NC, PE6ec-NC, PE6fc-NC, and PE6gc-NC (Fig. 1a and b and Supplementary Data 2). The PE protein expression cassettes were driven by the CaMV 35S promoter (Addgene plasmid #50267) and EURb7 terminator (Fig. 1a and Supplementary Data 2)²⁶. All the pegRNAs were previously designed and cloned, with an aepegRNA configuration that was used with the pU6cm by Vu and coworkers²⁶ (Fig. 1b and Supplementary Fig. 1 and Supplementary Table 1). The Csy4-aepegRNAs were cloned by replacing the tRNA(gly) and HDV sequences of the aepegRNAs with Csy4RS sequences (Fig. 3a and Supplementary Data 2). The Csy4 expression cassette was driven by the CaMV 35S promoter (Addgene plasmid #50267) and NOS terminator (Addgene plasmid #50339) (Fig. 3b and Supplementary Data 2). All the PE constructs were cloned into a geminiviral replicon vector system (Addgene plasmid #227697)^26,50 for plant transformation with a selection marker (NptII, Addgene #51144) (Supplementary Fig. 1). All cloning work was performed using the Golden Gate assembly method⁵³. All bioparts (Supplementary Data 2) were incorporated into the Moclo level 0 plasmids and assembled into the PE protein combinations and binary vectors, as shown in Fig. 1a and b, and Supplementary Fig. 1.

Tomato transformation

The PE tools were transformed into tomato cotyledon slices by Agrobacterium (super Agrobacterium version 2 strain (EHA105spv2))-mediated transformation protocol reported earlier by Vu and coworkers²⁶. The tomato Money Maker (MM) variety was used in this study. In detail, cotyledons that were 7 days old were excised for transformation. Agrobacterium containing the PE constructs was cultured, harvested, and resuspended in a modified ABM-MS medium. This medium consisted of ABM (composed of 2.21 g/L K₂HPO₄, 1.13 g/L KH₂PO₄, 1.5 g/L NH₄NO₃, 0.15 g/L KCl, 0.31 g/L MgSO₄, 0.015 g/L CaCl₂, 0.003 g/L FeSO₄, 3.9 g/L MES, and 20 g/L glucose at pH 5.5) mixed in a 1:1 ratio with 1/4 MSB5 (which includes MS basal salts with Gamborg B5 vitamins from Duchefa Biochemie B.V, Haarlem, Netherlands, Cat. M0231, along with 5 g/L sucrose, pH 5.2). The optical density (OD_600nm) of the Agrobacterium suspension was adjusted to 0.8, and 200 µM acetosyringone (Sigma-Aldrich Inc., Darmstadt, Germany, Cat. D134406) was added. The Agrobacterium cells were then activated at 28 °C for 1 h in a shaker set to 180 rpm. After this activation, the Agrobacterium cells were mixed with the cotyledon explants and incubated at room temperature for 20 min. The transformed explants were then placed on co-cultivation medium, which consisted of semisolid ABM-MS containing 1 mg/L zeatin trans-isomer (Duchefa Biochemie B.V, Haarlem, Netherlands, Cat. Z0917), 1 mM putrescine dihydrochloride (Sigma-Aldrich Inc., Darmstadt, Germany, Cat. P5780), 200 µM acetosyringone, and 7.5 g/L agar. They were incubated at 25 °C for 2 days before being washed and transferred to a selection medium. This selection medium contained full MSB5, 1.0 mg/L zeatin trans-isomer, 0.1 mg/L 3-Indoleacetic acid (Sigma-Aldrich Inc., Darmstadt, Germany, Cat. I2886), 1 mM putrescine dihydrochloride, 30 g/L glucose, 300 mg/L timentin, and 80 mg/L kanamycin, with a pH of 5.8. Explants were then incubated at 34 °C for 5 days and subsequently maintained at 28 °C for an additional 5 days before being shifted to 25 °C for the remaining stages (T34 incubation regime). These conditions were previously shown to result in the maximal activity of the PE2max tool by Vu et al.²⁶. Regenerated shoots were rooted in rooting medium ((haft MSB5, 1 mg/L Indole-3-butyric acid (Sigma-Aldrich Inc., Darmstadt, Germany, Cat. 57310), and 30 g/L glucose, 300 mg/L timentin, and 80 mg/L kanamycin, pH 5.8), and then transferred to soil.

Targeted deep sequencing

To assess PE efficiency in the callus stage, samples were collected at 16 days post-transformation (dpt) and analyzed via targeted deep sequencing. Targeted deep sequencing was performed as previously reported²⁶. Briefly, genomic DNA was extracted from cotyledon explants or plant leaves using the GENTi™ Advanced Genomic DNA Extraction Kit and the GeneAll® GENTi™ 32 Advanced automated nucleic acid extraction system (Jinol Biotechnology Co., Ltd., Seoul, Korea) using the manufacturer’s protocol. Targeted deep sequencing was performed using the Illumina MiniSeq platform (KAIST Bio-Core Center). Libraries were prepared via three rounds of PCR, with sample indexing in the third round using manufacturer-supplied primers (see Supplementary Data 3). FASTQ files were analyzed using CRISPResso2⁵⁴.

Screening T0 transformants and T1 plants carrying PE alleles

T0 transformants that survived the hardening step and T1 offspring plants from T0 events were screened for PE alleles. Three leaf fragments were collected from three different compound leaves of each plant and combined to isolate gDNA using the GENTi™ Advanced Genomic DNA Extraction Kit and the GeneAll® GENTi™ 32 Advanced automated nucleic acid extraction system (Jinol Biotechnology Co., Ltd., Seoul, Korea) using the manufacturer’s protocol. PCR reactions were used to amplify DNA sequences using primers (Supplementary Data 3) that flank the targeted sites, and the PCR products were sequenced by the Sanger method. The resulting sequencing chromatograms were analyzed using the Synthego ICE Analysis tool (version 3)⁵⁵ to screen for events carrying the desired PE alleles. Some representative PE events were subsequently validated by targeted deep sequencing. In Table 1, The number of desired PE events was calculated by subtracting the number of plants containing scaffold-inserted alleles from the total number of plants with the desired PE.

‘Total number of analyzed plants’ might be smaller than the sum of ‘Number of plants with desired PE’ ‘Number of plants contained indel’, and ‘Number of plants contained WT allele’, since an analyzed plant may contain both desired PE edits and imprecise edits or indels. The desired PE efficiency was calculated using the following equation:

Desired PE efficiency (%) = 100 × (number of desired PE events/total number of analyzed plants) (1).

Assessment of T-DNA and replicon presence within PE events

The replicon cassette is part of the T-DNA but can still replicate briefly after the T-DNA is removed or modified^50,51. This allows us to detect replicon DNA even if the original T-DNA sequences are missing. To address this, we created a dual genotyping method to distinguish between stable T-DNA integration and the persistence of episomal replicons. The presence of T-DNA and replicon within the PE events was assessed by PCR²⁶. In general, PCR reactions using primer pairs specified for the right border (RB) of the T-DNA and the circularized forms of the replicon (Supplementary Data 3) were conducted with SlGAPDH (Solyc05g014470) as an internal control. PCR products were separated on a 1% agarose gel, and the absence of T-DNA and replicon was determined by the lack of bands corresponding to their expected sizes without having any other unspecific bands. Each experiment was performed twice and yielded similar results.

Potential off-target analysis

The potential off-target analysis was conducted similarly to a previous report by Vu et al.²⁶. In brief, the gRNA sequences of the pegRNAs (see Supplementary Table 1) were analyzed using the Cas-OFFinder³⁸ web-based application to identify potential off-target sites in the tomato (Solanum lycopersicum) genome database. This search was conducted using the assembly variant SL2.40, accessed in March 2024 at https://solgenomics.net/. Only off-target sites that had fewer than four mismatches to the gRNA sequences were considered in this analysis. Two potential off-target sites per target gRNA/pegRNA were selected for the assessment (Supplementary Data 1). The 16-dpt cotyledon explants that were transformed with pegRNAs having identified potential off-target sites were subjected to analysis by targeted deep sequencing using primers listed in Supplementary Data 3. Targeted deep sequencing was performed using the Illumina MiniSeq platform (KAIST Bio-Core Center), and FASTQ files were analyzed using the CRISPResso2 program.

Herbicide resistance assay

Eight-leaf stage tomato seedlings carrying homozygous SlEPSPS1 (TIPS) and WT were sprayed two times (a five-minute interval) with 5 mM glyphosate isopropylamine (Farmhannong, Seoul, Korea) solution containing SilwettL-77 (0.02%). Sprayed plants were photographed after one and two weeks post-spraying.

Assessment of haploid induction by the SlCENH3(del) and SlCENH3(sub) alleles

Due to the lack of flow cytometry and karyotyping facilities for ploidy assessment, we employed an alternative method to assess the haploid induction caused by the tomato SlCENH3(del) and SlCENH3(sub) alleles. T1 plants carrying a homozygous form of the desired edits (upper stem containing hairs/trichomes) were cross-pollinated with plants carrying a homozygous knockout form (a recessive allele) of the HAIRS ABSENT gene⁵⁶, which are hairless. The resulting first filial generation (F1) plants were screened for the presence/absence of hairs and sequenced to identify the SlCENH3 allelic forms. Normal fertilization results in plants carrying heterozygous alleles (SlCENH3/SlCENH3(del) or SlCENH3/SlCENH3(sub)) and hairs, while male haploid plants induced by SlCENH3 mutations should possess the wild type (WT) CENH3 allele with a hairless phenotype (Supplementary Fig. 7a).

Data analysis and presentation

All the experiments were carried out at least in triplicate. Editing data, as well as any applicable statistical analyses and scatter plots, was subsequently processed using Microsoft Excel 365 and the GraphPad Prism 9.0 software. Multiple comparisons were executed utilizing the two-sided, uncorrected Fisher’s LSD test. Figures 1a and b and 3a, as well as Supplementary Fig. 7a were prepared using Biorender (biorender.com) under a paid subscription.

Reporting summary

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

Author contributions

T.V.V. and J.Y.K. conceived and designed the research. T.V.V., N.T.N., J.K., and T.H.N. conducted experiments. T.V.V., N.T.N., and J.Y.K. analyzed data. T.V.V. and N.T.N. wrote the manuscript. T.V.V. and J.Y.K. finalized the manuscript. All authors read and approved the manuscript.

Peer review

Peer review information

Nature Communications thanks Sangsu Bae, Qijun Chen, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The authors declare that all the data supporting the findings of this study are available in the article and its Supplementary Information files. The datasets from the high-throughput sequencing experiments generated in this study have been deposited in the National Center for Biotechnology Information (NCBI) database under accession code PRJNA1308257. The tomato (Solanum lycopersicum) genome database (Tomato variants SL2.40; https://solgenomics.net/) was accessed in March 2024. Source data are provided with this paper.

Competing interests

All the authors are inventors one of a Korean patent application (Application number 1020240167196) entitled “Method for increasing prime editing efficiency in plants”, which describes the UtPE system reported in the study. J.Y.K. is the founder and CEO of Nulla Bio Inc.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-67874-3.