EXPERT improves prime editing to make larger DNA changes more efficiently

Both cis nicks and upstream binding are essential for the differential editing capacity of the EXPERT compared to canonical PEs. We first confirmed the role of the two cis nicks in EXPERT, by comparing it with two variants, EXPERT-a and EXPERT-b, which each generate only one nick using either the ups-sgRNA or the ext-pegRNA, respectively (Supplementary Fig. 2a). These variants were then used to edit the HEK4_2 locus and introduce a 40-bp sequence replacement in HEK293T cells (Supplementary Fig. 2b). As expected, EXPERT (with two cis nicks) achieved 6.1% efficiency, in contrast to the 2.84% and below 0.1% efficiencies achieved by EXPERT-a and EXPERT-b, respectively. Mechanistically, we speculate that this is because EXPERT generates two cis nicks that enhance the detachment of the original single-stranded DNA fragment from the genome, thereby promoting subsequent processes (Supplementary Fig. 2c). These results confirmed the importance of the two cis nicks for the EXPERT.

We next aimed to confirm the role of the upstream binding. We constructed several PE2 variants (Supplementary Fig. 2a). (i) PE2: it generates one nick by using the ups-sgRNA as its pegRNA. It does not have the upstream binding. (ii) PE2-a: it generates one nick by using the ups-sgRNA as its pegRNA. It does not have the upstream binding. We included a truncated ext-pegRNA in PE2-a although it does not create a 2^nd nick. (iii) PE2-b: it generates two cis nicks, one using the ups-sgRNA, and another by using the ext-pegRNA. PE2-b also does not have the upstream binding. We compared these three PE2 variants with EXPERT, which has the upstream binding, to edit the same HEK4_2 locus for introducing a 40-bp sequence replacement (Supplementary Fig. 2b). All these PE2 variants lacking the upstream binding design, regardless of the design to generate one nick or two cis nicks, achieved low efficient edits, ranging from 0.05% to 0.37%, consistent with the knowledge that current PEs are inefficient for large fragment edits. Remarkably, the efficiency achieved by EXPERT is 6.1%, 122.1-fold higher than that by PE2. These results confirmed the essential role of the upstream binding for the EXPERT.

One consideration with the generation of two nicks in EXPERT is whether this will increase the unintended indel rates at the on-target site. Our results in the above experiments suggest that the presence of two cis nicks does not increase the likelihood of indel events in comparison to the one nick system PE2. The indel rate at the HEK4_2 site for EXPERT was 0.28%, comparable to that by PE2 (with 0.2% indels) (Supplementary Fig. 2b).

In summary, by introducing an extra upstream guide RNA (ups-sgRNA) to create an additional cis nick and an extended pegRNA (ext-pegRNA) that contains an upstream binding sequence, we construct a PE tool EXPERT.

Canonical PEs are unable to edit the upstream region of the pegRNA nick. The EXPERT, thanks to the introduction of two cis nicks and the use of ext-pegRNA, in theory, should be able to edit that unreachable region by canonical PEs.

To validate this hypothesis, we performed two edits at the VEGFA_1 site, both located in the upstream region of the ext-pegRNA nick: (i) replacing a 37-bp sequence (VEGFA_1 -37to-1 replace 37 bp); (ii) deleting a 37-bp sequence (VEGFA_1 -37to-1 del 37 bp) (Fig. 1b). The VEGFA_1 site was chosen because it provides multiple NGG sequences on the same strand, thereby facilitating the design of ext-pegRNA and ups-sgRNA. For the first edit, a high replacement efficiency at 33.7% with low indel rate (0.52%) was achieved. Similarly, for the second edit, the precise deletion rate was high at 18.8% with a low indel rate at 0.8%. We also attempted a third edit at this site: (iii) VEGFA_1 -37to-1 replace extended loxp and +1 TtoC, to test whether EXPERT allows for simultaneous editing on both sides of the ext-pegRNA nick. The results showed that the efficiency of this simultaneous editing on both sides of the ext-pegRNA nick has reached 10.7%, with a low indel rate at 0.38% (Fig. 1c).

These results demonstrate that EXPERT has the unprecedented ability to edit upstream region of the pegRNA nick, and simultaneously edit both sides regions. All types of edits have low indel rates, which preliminarily supports the hypothesis that the presence of two cis nicks does not elevate the occurrence of indel events.

We then assessed the impact of DCN on editing efficiency by EXPERT at two endogenous loci: VEGFA_1 and HEK4_1. The results showed that the editing rates peaked when the DCNs were between 32 and 40 nt (Fig. 2b). At the VEGFA_1 locus, the highest editing rate of 33.7% with an indel rate of 0.52% was achieved when the DCN was 37 nt long. At the HEK4_1 locus, the highest editing rate of 11.9% with an indel rate of 2.23% was achieved when the DCN was 32 nt long. Interestingly, at this site (HEK4_1), a slightly shorter DCN of 24 nt almost completely abolished the intended edits. Similar results were observed in the 293T-reporter cells: when a 23 nt DCN was used, the editing rate was only 0.24%, whereas, with a 38 nt, the editing rate peaked at 7.44% (Fig. 2a). We speculate that these observations suggest that a DCN of 24 nt or shorter may cause steric hindrance between two adjacent nCas9 proteins. Based on this, DCNs of 24 nt or shorter should be avoided in EXPERT.

These data demonstrate that the distance between two nicks affects the editing efficiency of EXPERT. Our results suggest using a DCN between 32 and 40 nt as the starting design.

When the editing region is away from either of the nick sites (i.e., the ext-pegRNA nick and the ups-sgRNA nick), there is a possibility for the genomic 3′ or 5′ Flap of the original single-stranded DNA to hybridize with the complementary ext-pegRNA, which could potentially hinder the synthesis by the reverse transcriptase and subsequently affect the production of the newly synthesized DNA strand (Fig. 2c and Supplementary Fig. 4a). In support of this, EXPERT exhibited notably low editing activity when the editing regions were away (≥14 nt) from the ext-pegRNA nick, both at the VEGFA_1 site (edit: VEGFA_1 -34to-18 del 17 bp, 0.3%) and the HEK4_1 site (edit: HEK4_1 -32to-15 replace 6*His, 0.54%) (Fig. 2e, mismatch number = 0). Similarly, when the editing region was away (≥14 nt) from the ups-sgRNA nick, EXPERT also showed low editing activity at the HEK4_1 site (edit: HEK4_1 -18to-1 replace 6*His, 0.28%) (Supplementary Fig. 4c, mismatch number = 0). Likely due to the same reason, EXPERT also demonstrated poor performance in creating small edits in the upstream region of the ext-pegRNA nick (Supplementary Fig. 5a).

We thus hypothesized that introducing mismatches in the regions corresponding to DNA 3′ Flap complementarity on the ext-pegRNA will disrupt the hybridization of DNA Flap and ext-pegRNA, therefore improving the editing efficiency of EXPERT.

We first tested if this strategy improves the EXPERT’s editing efficiency to the editing regions that are away from the ext-pegRNA nick. We introduced mismatches starting from the nucleotide on the ext-pegRNA corresponding to the first nucleotide of the complementary DNA 3′ Flap, and tested the effects of the number of mismatches from 0 to 17 (Fig. 2d and Supplementary Fig. 6). The results demonstrated that when mismatches were introduced, as few as only one, the editing efficiency could be improved, for example at the HEK4_1 site from 0.54% to 3.93%. The extent of improvement was positively correlated with the number of mismatches introduced. The highest editing efficiency was achieved with 10 or 11 mismatches at both tested sites: 24.81% at the VEGFA_1 site and 14.43% at HEK4_1 site (Fig. 2e). Analysis of product purity (the ratio of intended edits: indels) revealed that introducing mismatches enhanced product purity at both the VEGFA_1 and HEK4_1 sites (Fig. 2f).

We then tested if this strategy improves the EXPERT’s editing efficiency to the editing regions that are away from the ups-sgRNA nick. Following a similar strategy, we introduced mismatches starting from the nucleotide on the ext-pegRNA corresponding to the first nucleotide of the complementary DNA 5′ Flap, with the number of mismatches varying from 0 to 14 as a way to improve EXPERT’s editing efficiency to the editing regions that are away from the ups-sgRNA nick (Supplementary Fig.and Supplementary Fig.). The results again demonstrated a positive correlation between the editing rates and the number of mismatches. When three mismatches were introduced, the editing efficiency reached ~10%. When 11 mismatches were introduced, the efficiency increased to 12.4%, and under these conditions, the highest product purity was achieved (Supplementary Fig.). 4 6 4c

We also tested an alternative strategy for introducing mismatches, specifically by introducing mismatches at intervals of every 3 and 5 nucleotides, respectively (Supplementary Fig. 7). The results showed that the editing efficiency of ext-pegRNA with such mismatches closely resembles that observed with ext-pegRNA containing 11 consecutive mismatches. These results suggest an alternative mismatch design for ext-pegRNA to achieve effective editing with EXPERT. Moreover, EXPERT exhibits differential unintended indel rates at the VEGFA_1 and HEK4_1 loci, possibly attributable to the sequence characteristics of these sites or the editing sequence. It is important to note that incorporating appropriate mismatches can enhance editing efficiency and reduce indel rates (Fig. 2e and Supplementary Fig. 4c).

Lastly, we evaluated EXPERT for introducing small edits (small insertions, small deletions and base substitutions) in the upstream region of the ext-pegRNA nick, with or without the inclusion of mismatches (Supplementary Fig. 5). The results demonstrated that EXPERT can perform all these types of small edits in the upstream region of the ext-pegRNA nick (Supplementary Fig. 5). Again, introducing mismatches significantly improved editing efficiency, in line with our speculation that the introduced mismatches reduce the homology between the ext-pegRNA and the genomic DNA 3′ Flap, thereby preventing their hybridization (Fig. 2c). Nevertheless, it is noted that, despite the lower unintended on-target indel rates, the efficiency and product purity of the small edits achieved by EXPERT are generally lower than those achieved by PE3 (Supplementary Fig. 5b).

All these results demonstrate that the editing efficiency of EXPERT can be improved by introducing mismatches.

PBS plays an important role in initiating reverse transcription and ensuring efficient PE efficiency^1,24. To assess whether the length of the PBS affects the editing efficiency of EXPERT, experiments were conducted at two loci: the VEGFA_1 and PRNP loci, with PBS lengths ranging from 8 to 17 nt (Fig. 2g). The results showed that PBS lengths ranging from 9 to 16 nt could lead to efficient editing outcomes: 11.64% to 15.41% at the VEGFA_1 locus, and 2.42% to 5.53% at the PRNP locus. From a practical point, we would recommend using PBS length of 12 nt because at both loci this parameter led to the highest editing efficiency, and avoid using PBS shorter than 9 nt as apparently this drastically reduced the editing rate in one tested locus.

Subsequently, we conducted an investigation to determine the optimal length of HS. Analysis of the editing efficiency across different HS lengths revealed that effective editing could be achieved with HS lengths ranging from 12 to 22 nt (Supplementary Fig.). When the HS length is shorter than 16 nt, only one locus exhibits an editing efficiency greater than 6% (1 out of 12 or 8%). In contrast, when the HS length is 16 nt or longer, six loci show an editing efficiency exceeding 6% (6 out of 7 or 86%) (Supplementary Fig.). Therefore, we recommend an HS length of 16 nt or longer as the starting point in EXPERT design. 8 8

These results indicate that EXPERT is capable of inserting large DNA fragments (up to 100 bp) at the upstream region of the ext-pegRNA nick, but the efficiency remains to be improved. To address this, we included an extra gRNA, positioned between the ups-sgRNA and ext-pegRNA, which we refer to as the Helper gRNA (Fig. 3c). We reason that this Helper gRNA would direct the nCas9-RT to create a third cis nick, and recruit an additional nCas9-RT enzyme between the original two nicks, thereby promoting the detachment of the original DNA strand and increasing the overall reverse transcription.

We tested this design in the 293T-reporter cell line. The addition of the Helper gRNA to the system resulted in a 3-fold improvement for a 71-bp replacement edit and a 1.7-fold improvement for a 96-bp replacement edit (Fig. 3d). This was further confirmed at two endogenous sites: (i) an 88-bp replacement at the EMX1 site; and (ii) a 75-bp replacement at the VEGFA_1 site (Fig. 3e). The replacement efficiency at the EMX1 locus increased from 0.21% to 3.1%, a 15-fold improvement. Likewise, at the VEGFA_1 locus, the efficiency increased 6.1-fold from 1.13% to 6.86%. In all cases, the indel rates were at low levels ranging from 0.15% to 0.2%, further indicating that introducing nicks on the same DNA strand does not elevate the risk of generating DSBs.

These findings demonstrate that including a Helper gRNA in the EXPERT improves its capacity for large fragment replacement editing without elevating the indel rate.

The use of the ups-sgRNA in the EXPERT raises the question of whether it introduces higher indel rates. To systematically evaluate this, we compared EXPERT with the representative single-pegRNA system PE2 and the two-pegRNA system twinPE.

Next, we evaluated the gRNA (ext-pegRNA or ups-sgRNA)-independent and gRNA-dependent off-target (OT) effects of EXPERT using whole-genome sequencing (WGS). The EXPERT-expressing plasmids carrying three types of targeting RNAs were used: (i) with non-target gRNAs as the vehicle control; (ii) targeting the EMX1 locus (EMX1 -17 loxp + 6 * His + AscI ins); and (iii) targeting the VEGFA_1 locus (VEGFA_1 -34to-3 del 32 bp). The plasmids were transfected into HEK293T cells, respectively. The genomic DNA of transfected cells of these groups, as well as those of the non-transfected wild-type cells, were subjected to WGS.

We first determined whether EXPERT induces gRNA-independent OT mutations genome-wide. The analysis showed that there was no significant difference in the level of genome-wide base substitutions and indels among the four groups of cells, indicating that EXPERT does not induce whole-genome gRNA-independent OT effects (Fig. 4e, f).

Next, we evaluated the gRNA-dependent OT effects of EXPERT. We determined the on-target editing efficiencies of EXPERT with amplicon sequencing and assessed the frequencies of indels in potential OT sites by WGS. The results showed that the frequencies of indels for all predicted potential OT sites are below 0.1% (Supplementary Fig.). 12

These results demonstrate that both gRNA-independent and gRNA-dependent OT effects by EXPERT are minimal.

In theory, the use of ups-gRNA and ext-pegRNA is compatible with all PE systems as long as they are based on nicking the target DNA.

We further compared EXPERTmax with later versions of PEmax (PE3max, PE4max, and PE5max). Comparing to PE2max, PE3max uses an additional nicking sgRNA, PE4max uses an additional MLH1dn component, whereas PE5max utilizes both additional components (i.e., nicking sgRNA and MLH1dn).

Without including the additional nicking sgRNA, EXPERTmax had an average 1.7-fold higher editing efficiency than PE3max (Supplementary Fig. 14). Inclusion of the additional nicking sgRNA slightly but not dramatically further improved the EXPERTmax’s editing efficiency, resulting in an average 1.9-fold higher than those by PE3max. EXPERTmax + MLH1dn outperformed PE4max by an average of 3.6-fold, and EXPERTmax + MLH1dn + additional nicking sgRNA resulted in an average 2.4-fold increase in editing efficiency compared to PE5max (Fig. 5a and Supplementary Fig. 13). Furthermore, product purity results demonstrated that EXPERTmax consistently achieved higher purity compared to its corresponding PEmax systems (Supplementary Fig. 15).

These results demonstrate that the EXPERT strategy is compatible and readily adaptable to other PE systems.

We next assessed EXPERTmax in different cell types in different species.

First, we used EXPERTmax to edit four different human cell lines: the lymphoblast cell line K562, the leukemia T lymphocyte cell line Jurkat, the cervical carcinoma-derived cell line Hela, and the human fetal lung fibroblast HFL1, respectively.

In K562 cells, the editing efficiency of EXPERTmax was 9.6–21.6 times higher than that of PE2max. The average improvement was 15.4-fold (Fig. 5b and Supplementary Fig. 16a). For instance, the efficiency of inserting a Flag tag at the EXM1 site (EMX1 -17 Flag ins) increased from 2.07% to 25.27%. In Jurkat cells, the editing efficiency of EXPERTmax was 2.9–46 times higher than that of PE2max. The average improvement was 17.4-fold (Fig. 5c and Supplementary Fig. 16b). The efficiency of inserting a Flag tag at the EXM1 site (EMX1 -17 Flag ins) increased from 0.07% to 3.22%, representing a remarkable 46-fold increase. Similarly in Hela cells, the efficiency by EXPERTmax was 1.5 to 8.4-fold higher than that by PE2max, with an average improvement of 4-fold (Fig. 5d and Supplementary Fig. 16c). In addition, the overall product purity of EXPERTmax was higher than that of PE2max in these three cell lines (Supplementary Fig. 17). In HFL1 cells, the efficiency of EXPERTmax was 1.5 to 25.8-fold higher than that of PE2max, while the unintended indel rates remained either lower than or comparable to those of PE2max (Supplementary Fig. 18). These findings demonstrate that the EXPERT can be utilized in different human cell types.

We then tested EXPERTmax in a mouse cell line N2a and in pig fetal fibroblast (PFF) cells to assess its applicability in non-human species. In N2a cells, the intended edits are “-34to-1 replace loxp” at the Ifnar1 locus, and “-32to-1 replace Flag+AscI” at the Tgfb1 locus. In pig PFF cells, the intended edits are “-17 Flag ins” at the ANTXR1 locus, and “-41to-1 replace anti-JEV mutation” at the CALR locus. The results demonstrated that EXPERTmax significantly enhanced editing efficiency in both mouse and pig cells. Compared to PE2max, the editing efficiency of EXPERTmax was 3.9–4.3 times higher in N2a cells (Fig. 5e) and 3.4–39.7 times higher in PFF cells (Fig. 5f). Of note, in PPF cells at the susANTXR1 site, the editing efficiency increased from 0.23% to 9.13%, representing a 39.7-fold increase. Consistent with findings in HEK293T cells, the unintended indel rates did not significantly increase compared to PE2max for most edits. As a result, the overall product purity achieved by EXPERTmax was also higher than that by PE2max (Fig. 5e, f).

These findings suggest that EXPERTmax can be used in different mammalian species. Notably, in pig species, there have been no previous reports of using PE systems for large fragment edits (e.g., replacement of 41 bp), as attempted here. EXPERTmax achieved an extraordinary efficiency of 23.4%.

We then employed EXPERTmax to generate disease-relevant mutations in human cells as a proof-of-concept demonstration for its potential use in human biomedical research. Cystic fibrosis (CF) is a fatal inherited disease caused by mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) gene, which is inherited in an autosomal recessive manner^25–27. To date, over 1900 mutations have been identified, with more than 300 known to cause the disease²⁸. Traditional gene-editing techniques often have limited capabilities, leading to suboptimal therapeutic outcomes. To demonstrate EXPERT’s applicability in CF gene editing, here we worked to generate a large fragment mutant sequence (36 bp) in exon 4 that carries multiple known CFTR mutations (Fig. 5g). This region cannot be edited by canonical PEs using the initially selected pegRNA because it is located at the upstream region of this pegRNA. However, it can be achieved using EXPERT by using this pegRNA spacer. We compared the editing efficiencies among (i) PE2max (using the ups-sgRNA as the pegRNA); (ii) PE3max (using the ups-sgRNA as the pegRNA); (iii) EXPERTmax (using the initially selected pegRNA as the ext-pegRNA); and EXPERTmax (using the initially selected pegRNA as the ext-pegRNA) + an additional nicking sgRNA (as used in PE3max) (Fig. 5g). The results demonstrated that the utilization of EXPERTmax (9.18%) significantly enhanced the editing efficiency in comparison to PE2max (2.33%) and PE3max (3.66%). The use of “EXPERTmax + nicking sgRNA” further enhanced the efficiency over that by EXPERTmax along, reaching an average editing efficiency of 11.8%, 3.2 times higher than that by PE3max and 5.1 times higher than that by the PE2max. Moreover, EXPERTmax exhibited remarkably low indel rates and showed higher product purity compared to PEs (Fig. 5h). These findings suggest the potential value of EXPERTmax in translational research of human diseases such as CF.

The pegRNA-expressing cassette and puromycin-expressing sequence were amplified from pU6-pegRNA-GG-acceptor (Addgene: 132777) and pKLV2-U6gRNA5(BbsI)-PGKpuro2AZsG-W (Addgene: 67975), respectively. These components were then cloned into the pCMV-PE2 plasmid (Addgene: 132775) to generate the PE2-U6-puro vector. The Csy4 expression cassette was synthesized (GenScript) and then cloned into the PE2-U6-puro plasmid to generate the PE2-U6-puro-Csy4 vector. The PGK-Puro-T2A-ZsGreen sequence were amplified from pKLV2-U6gRNA5(BbsI)-PGKpuro2AZsG-W (Addgene: 67975) and then were cloned into pCMV-PEmax (Addgene: 174820) and pCMV-PEmax-P2A-MLH1dn (Addgene: 174828) to generate the pCMV-PEmax-puro-ZsGreen vector and the pCMV-PE4max-puro-ZsGreen vector, respectively. PCR was performed using the KAPA HiFi PCR Kit (Roche). Gel extraction of the amplified DNA fragments was performed using FastPure Gel DNA Extraction Mini Kit (Vazyme). The ClonExpress II One Step Cloning Kit (Vazyme) was used to construct PE2-U6-puro, PE2-U6-puro-Csy4, pCMV-PEmax-puro-ZsGreen, and pCMV-PE4max-puro-ZsGreen vectors.

Three versions of EXPERT vectors were utilized in this study. Schematic diagrams depicting the structure of these vectors are presented in Supplementary Fig. 19. Of note, both the ext-pegRNA and ups-sgRNA are expressed in the same one vector in all versions. The main difference in plasmid vectors between PE and EXPERT is that the EXPERT vector includes an additional guide RNA expression module. Overall, EXPERT v1 version was used in Figs. 1b and 2; EXPERT v3 version was used in Figs. 1c, 2g, 3, 4 and 5. All ext-pegRNA, ups-sgRNA, and pegRNA sequences were synthesized and cloned into PE2-U6-puro (digested with AarI), PE2-U6-puro-Csy4 (digested with AarI), pCMV-PEmax-puro-ZsGreen (digested with NruI), and pCMV-PE4max-puro-ZsGreen (digested with NruI) by Tsingke Biological Technology or GenScript to perform our experiments. The detailed information for the ext-pegRNA, ups-sgRNA, and pegRNA sequences used in this study is provided in Supplementary Data 1. The detailed sequences of all constructs are listed in Supplementary Data 2.

A series of twinPE pegRNA pairs for these different loci were evaluated, and shown in Supplementary Fig. 11. Specifically, for each locus, we tested various combinations of pegRNA pairs, with the PBS of each pegRNA designed using the online tool pegFinder³⁵. The detailed sequence information for these pegRNAs is provided in Supplementary Data 1. The best-performing twinPE pegRNA pairs, with a precise editing efficiency of at least 2% and the highest product purity, for each locus were selected and used for comparison with EXPERT.

HEK293T (ATCC, CRL-11268) and 293T-reporter cells were maintained in DMEM (Gibco) + 10% FBS (CELLiGENT) + 1% penicillin-streptomycin (Gibco); K562 (ATCC, CRL-3344) and Jurkat (ATCC, TIB-152) cells were maintained in RPMI Medium 1640 (Gibco) + 10% FBS (CELLiGENT) + 1% penicillin-streptomycin (Gibco); Hela (Procell, CL-0101), HFL1 (Procell, CL-0106) and N2a (Procell, CL-0168) cells were kindly provided by Procell Life Science & Technology Co., Ltd. Hela (Procell) and N2a (Procell) cells were maintained in MEM (Procell) + 10% FBS (CELLiGENT) + 1% penicillin-streptomycin (Gibco). HFL1 (Procell) cells were maintained in Ham’s F-12K (Procell) + 10% FBS (CELLiGENT) + 1% penicillin-streptomycin (Gibco); PFF (isolated by our laboratory from a 35-day-old pig embryo) cells were maintained in DMEM (Gibco) + 10% FBS (Gibco) + 1% penicillin-streptomycin (Gibco). All these cells were cultured at 37 °C with 5% CO₂ and regularly tested to exclude mycoplasma contamination.

For transfection assays, HEK293T, Hela, and N2a cells were seeded into 6-well plates and transfected (~80% confluent) with 2.5 µg plasmid using JetPRIME (PolyPlus) according to the manufacturer’s instructions. K562, Jurkat, HFL1, and PFF cells were transfected using an electroporation instrument (Celetrix). Briefly, a total of 1 × 10⁶ cells were transfected with 2.5 µg plasmid using a 20 µL electroporation cuvette where the cells were shocked with 520 V.

All data in this study were collected through sequencing of transfection-positive cells. K562, Jurkat, and HFL1 cells were enriched with positive cells using flow cytometry sorting. Specifically, At 96 h post-transfection, GFP-positive cells (transfection-positive cells) were FACS-enriched (BD Aria^TM III) for subsequent experiments. HEK293T, Hela, N2a, and PFF cells were enriched by adding puromycin (InvivoGen) to the medium at the final concentration of 2 µg/mL, 24 h post-transfection. At 96 h post-transfection, transfection-positive cells were collected and lysed with freshly prepared lysis buffer and incubated at 65 °C for 40 min, followed by incubation at 95 °C for 15 min.

Flow cytometry analysis or sorting was performed 96 h post-transfection. 293T-reporter cells were collected after PBS washing and trypsin digestion and resuspended in PBS for flow cytometry analysis (Beckman CytoFLEX). Data were analyzed by CytExpert software. For Flow cytometry sorting, K562, Jurkat, and HFL1 cells were collected after PBS washing, resuspended in PBS, and sorted using a BD Aria™ III flow cytometer, followed by enrichment of GFP-positive cells.

The resulting cell lysis solution was subjected to PCR amplification of the target region using specific primers (Supplementary Data 3) with barcodes. Equal amounts of PCR products were pooled, purified, and commercially sequenced (Annoroad) using the NovaSeq platform. The obtained amplicon sequencing reads (FASTQ files) were then demultiplexed using bcl2fastq (Illumina) and analyzed by aligning the amplicon reads to a reference sequence by CRISPResso2³⁶. To quantify the frequency of precise editing and indels, CRISPResso2 was run in the HDR mode. The precise editing efficiencies were calculated as the ratio of the number of desired reads to the total number of reads, while indel rates were calculated as the ratio of the number of indel-containing reads to the total number of reads.

Genomic DNA was extracted from transfected and wild-type cells using the TIANapm Genomic DNA kit (TIANGEN) according to the manufacturer’s instructions. WGS was performed by using the MGI DNBSEQ-T7 platform (Annoroad). Raw data was processed by SOAPnuke with default parameters to ensure the quality of data used in further analysis³⁷. The obtained clean reads were mapped to the reference genome GRCH38.p12 from the National Center for Biotechnology Information (NCBI, https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_000001405.38/↗) by Burrows-Wheeler aligner³⁸. SAMtools were performed to sort reads and remove low-quality reads³⁹. Duplicate reads from PCR were detected and removed by MarkDuplicates module of Genome Analysis Toolkit (GATK)⁴⁰. The HaplotypeCaller module of GATK was performed for base substitutions and indels calling. The obtained variants were filtered by VariantFiltration module of GATK. The filtering settings were as follows: Base Substitution: QD < 2.0, ReadPosRankSum < –8.0, FS > 60.0, QUAL < 30.0, DP < 4.0, MQ < 40.0, MappingQualityRankSum < −12.5 and INDEL: QD < 2.0, ReadPosRankSum < −20.0, FS > 200.0, QUAL < 30.0, DP < 4.0. The number of genome-wide base substitutions and indels are listed in Supplementary Data 4. Potential gRNA-dependent off-target sites were predicted by Cas-OFFinder⁴¹, allowing up to five mismatches. Off-target sites identified (Supplementary Data 4) by WGS shown in Supplementary Fig. 12 were randomly selected from genome-wide base substitutions.

At least three biological replicates were conducted per experiment to ensure the reliability of the data. Data are presented as mean ± s.d. Statistical analyses were performed with GraphPad Prism (version 9). To determine the statistical significance of a few results, the two-tailed Student’s t-test was employed.

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