Background

Single-nucleotide substitutions, gene expression changes, or removal of deleterious genes are molecular basis of many important agronomic traits in plants [1]. Stacking traits or changing several key factors of regulatory pathways would greatly advance crop breeding [1]. Diversity and simplicity of CRISPR-Cas systems provide powerful molecular toolboxes [2,3,4,5,6,7,8,9,10]. Several strategies have been employed to implement multiplex applications in bacteria, yeast, and mammalian cells [11,12,13,14,15,16]. The most commonly used multiplex strategies for orthogonal genome manipulation include several orthogonal CRISPR systems forming the multi-functional CRISPR system, such as a dual-functional method using SpCas9 variants for adenine base editor (ABE) and SaCas9 for cytosine base editor (CBE) [17] or a tri-functional method using LbCpf1 variant for CRISPRa, SpCas9 variant for CRISPRi, and SaCas9 variant for deletion [15]. However, these strategies require delivering multiple Cas proteins simultaneously, and each Cas protein needs its own PAM recognition [15, 17]. On the other hand, various RNA aptamers were incorporated into CRISPR RNA scaffolds, and these aptamers could recruit their binding proteins to Cas9-targeted sites [14]. This strategy has been used to recruit gene activation or repression effectors to target different genomic sites to perform dual function systems [14], and to recruit sets of fluorescent proteins to label multiplexed genomic site [18]. Nonetheless, tools for performing multi-functional editing are still very limited, especially in plants [1].

Compared with Cas9 and dCas9, nCas9 has not been exploited to its full potential for multiplex genome engineering. To generate multiple genetic modifications of plant genomes, we envisioned a multiplex genome editing system that would achieve simultaneous wide-editing induced by a single system (SWISS) based on nCas9 nuclease (Fig. 1a). The system contains sgRNA scaffolds, with different RNA aptamers recruiting cognate binding proteins (BPs), fused with cytidine or adenosine deaminases, and could carry out cytidine base editing and adenine base editing at different target sites simultaneously. By using another pair of sgRNAs to this dual-function system, SWISS could introduce a double-strand break (DSB) at a third target site, obtaining a tri-functional genome editing at multiple sites (Fig. 1a).

Fig. 1
figure 1

Multiple RNA scaffolds and binding protein orthologs mediate efficient C-to-T conversion. a The CRISPR scaffold RNA-programmed multiplex genome editing system based on nCas9 nuclease. Abbreviation: BP, binding protein. b Architecture of the pOsU3-esgRNA-2×MS2 construct with two MS2 hairpins at the 3′-end of the esgRNA. Abbreviation: SUP4 Term, transcription terminator for the S. cerevisiae SUP4 tRNA gene. c Architectures of PBEc1-c5. Abbreviations: XTEN, 16-aa linker; NLS, nuclear localization signal; CaMV, cauliflower mosaic virus; Term, terminator. d Comparison of C-to-T conversion using a BFP-to-GFP reporter system by PBE and the five PBEcs in rice protoplasts (n = 3). Values and error bars indicate means ± s.e.m. of three independent experiments. e Architectures of PBEc6-c8. Abbreviations: XTEN, a 16-aa linker; NLS, nuclear localization signal; CaMV, cauliflower mosaic virus; Term, terminator. f Schematic of the scRNAs with MS2, PP7, boxB, or com RNA hairpins in the tetraloop and stem loop2 or the 3′-end of the sgRNA and esgRNA. g Comparison of C-to-T conversion using the BFP-to-GFP reporter system induced by various scRNAs and their cognate PBEcs in rice protoplasts (n = 3). The f6 aptamer hairpin binds MCP specifically. Two PP7 hairpin variants were adopted. Values and error bars indicate means ± s.e.m. of three independent experiments. h Comparison of C-to-T editing frequencies of rice endogenous genes induced by four scRNAs and their cognate PBEcs (n = 3). An untreated protoplast sample served as control. Values and error bars indicate means ± s.e.m. of three independent experiments.

Full size image

Results

Engineered constructs and scRNAs for efficient C-to-T conversion

We previously developed a plant cytosine base editor [19], PBE, consisting of cytidine deaminase APOBEC1 [20], nCas9 (D10A), and uracil DNA glycosylase inhibitor (UGI). In the present work, we firstly generated a CRISPR RNA scaffold (scRNA) construct with two MS2 hairpins at the 3′-end of the esgRNA (esgRNA-2×MS2), which mediates efficient activation in human cells [14] and is driven by the OsU3 promoter (Fig. 1b; Additional file 2: Sequences S1 and S2). To generate MS2-recruited PBE constructs (PBEcs), we used a T2A “self-cleaving” peptide to express nCas9 and MCP (MS2 coat protein)-deaminase fusion modules simultaneously; the nCas9 (D10A) was fused with or without APOBEC1 or UGI as the RNA-programmed module, while MCP fused with APOBEC1 or UGI or both was the recruited module, generating PBEc1-c5. These PBEcs were codon-optimized for crop plants and driven by the Ubi-1 promoter of maize (Fig. 1c; Additional file 2: Sequences S3).

We tested PBEc1-c5 with the esgRNA-2×MS2 scaffold in our former BFP-to-GFP reporter system using rice protoplasts [19], in which GFP fluorescence requires the codon CAC (His66) to be converted to TAC (Tyr66). PBE together with a conventional sgRNA construct was used as control. The average proportion of GFP+ rice cells ranged from 0.7 to 10.8%, with the MCP-APOBEC1-UGI (PBEc4)-recruited module the most efficient and its frequency being about 2.9-fold higher than obtained with PBE (Fig. 1d; Additional file 1: Figure S1), while the yield with MCP-UGI-APOBEC1 (PBEc5) was only about 1.2-fold higher than that with PBE. C-to-T activity using MCP-APOBEC1 as recruited module (PBEc1 and PBEc2) was comparable to that of PBE. However, when MCP-UGI was the recruited module (PBEc3), C-to-T activity declined dramatically (Fig. 1d), though MCP bound to its RNA aptamer hairpin as a dimer [21]. Therefore, we chose the structure of PBEc4 for further development.

To develop a platform for multiplex recruitment and create several scRNAs for efficient C-to-T conversion, we replaced the MCP in PBEc4 with PCP, N22p, and Com [14, 18], generating PBEc6-c8, which recognize the well-characterized viral RNA hairpins PP7, boxB, and com, respectively (Fig. 1e; Additional file 2: Sequences S3). For scRNAs, besides one, two, or three hairpins of these RNA aptamers (MS2, PP7, boxB, or com) were introduced to the 3′-end of the sgRNA or esgRNA [22, 23], RNA hairpins on the tetraloop and stem loop2 of sgRNA or esgRNA and a quad-hairpin scRNA (sgRNA4.0) were generated for comparison [18, 24] (Fig. 1f; Additional file 1: Figure S2; Additional file 2: Sequences S2).

We then compared the activity of these scRNAs and cognate PBEcs using the BFP-to-GFP reporter system in rice protoplasts. To our surprise, we found that all the scRNAs with RNA hairpins on the tetraloop and stem loop2, including the MS2, PP7, and boxB hairpins, yielded very low frequencies of GFP+ signals, ranging from 0.1 to 0.4% (Fig. 1g; Additional file 1: Figure S3). On the other hand, all the scRNAs bearing two or three RNA aptamer hairpins at their 3′-ends, and esgRNA-1×com incorporating one 3′-end hairpin, yielded robust frequencies of GFP+ signals, ranging from 1.8 to 8.8% (Fig. 1g; Additional file 1: Figure S3). Of them, PBEc4 combined with esgRNA-2×MS2 (7.5%), esgRNA-3×MS2 (8.0%), and sgRNA4.0 (8.8%), and PBEc8 combined with esgRNA-2×com (6.9%), generated even higher frequencies of GFP+ signals than the combinations of the PBE and sgRNA (1.7%) or esgRNA (6.0%) (Fig. 1g; Additional file 1: Figure S3). The different outcomes for the two scRNA conformations could be due to the fact that we used a double-stranded linker between hairpin repeats to improve the conformational stability of the 3′-ends of the multi-hairpin scRNAs [14].

To evaluate the effectiveness of esgRNA-2×MS2, esgRNA-3×MS2, sgRNA4.0, and esgRNA-2×com in converting C-to-T in endogenous rice genes, we expressed the sgRNAs in rice protoplasts using these four scRNAs and co-transfected PBEc4 or PBEc8. PBE with conventional sgRNA and esgRNA constructs served as controls. The base editing efficiencies at C3 to C9 of these five tested target sites (OsACC-T1, OsDEP1-T1, OsDEP1-T2, OsEV, and OsOD) were enhanced using esgRNA-2×MS2 (average 18.0%), esgRNA-3×MS2 (average 15.0%), and esgRNA-2×com (average 11.1%) compared with the conventional sgRNA (average 4.8%), esgRNA (8.0%), and sgRNA4.0 (average 4.7%) (Fig. 1h). The results with esgRNA-2×MS2, esgRNA-3×MS2, and esgRNA-2×com were 2.3- to 3.8-fold superior to those with the conventional sgRNA (Additional file 1: Figure S4) and had the same primary C-to-T base editing window (Fig. 1h; Additional file 1: Figure S5). Moreover, the C-to-T base editing efficiencies of narrow window APOBEC1 variants (YE1, EE, and YEE) can also be improved 1.4- to 1.8-fold in central positions (C5 for OsEV, C6 for OsOD) by esgRNA-2×MS2 with PBEc4 architecture than sgRNA with PBE architecture (Additional file 1: Figure S6; Additional file 2: Sequences S3).

Thus, we have shown that incorporating different RNA aptamers into sgRNA provides an effective approach to multiplex recruitment of RNA-programmed nCas9 (D10A) in plants. In addition, PBEc4 combined with esgRNA-2×MS2 or esgRNA-3×MS2 and PBEc8 combined with esgRNA-2×com can be chosen as candidates for develo** multiplex genome editing systems.

Engineering constructs for RNA scaffolds mediated A-to-G conversion

We previously created the plant adenine base editor [25], PABE-7, composed of a laboratory-evolved deoxyadenosine deaminase dimer ecTadA-ecTadA7.10 [26], nCas9 (D10A), and three copies of the SV40 NLS at the C terminus. To repurpose PABE-7 into RNA aptamer-recruiting architecture using nCas9 (D10A) platform, we generated PABE constructs (PABEc1-c3) for recruiting esgRNA-2×MS2 by optimizing the linker length and location between MCP and adenosine deaminase (Fig. 2a; Additional file 2: Sequences S3). The mGFP-to-GFP reporter system was used to test A-to-G conversion activity in rice protoplasts [25]; in this case, an A-to-G conversion on the non-coding strand converts TAG to CAG (Gln69) on the coding strand. In contrast to the increased C-to-T editing efficiency obtained by using PBEc4, the A-to-G editing efficiency of PABEc1-c3 was lower (1.7–8.0%) than that of PABE-7 (14.4%). Of which, PABEc3 showed higher A-to-G editing efficiency (Fig. 2b; Additional file 1: Figure S7). Therefore, we chose the PABEc3 architecture for further multiplex development and tried to improve its activity using other RNA aptamers.

Fig. 2
figure 2

Optimization of plant adenine base editor constructs using multiple RNA scaffolds and binding protein orthologs. a Architectures of PABEc1-c3. Abbreviations: ecTadA7.10, evolved Escherichia coli TadA; aa, amino acid; XTEN, a 16 aa linker; NLS, nuclear localization signal; CaMV, cauliflower mosaic virus; Term, terminator. b Comparison of A-to-G conversion using the mGFP-to-GFP reporter system induced by PABE and the three PABEcs in rice protoplasts (n = 3). Values and error bars indicate means ± s.e.m. of three independent experiments. c Architectures of PABEc4-c6. Abbreviations: ecTadA7.10, evolved Escherichia coli TadA; aa, amino acid; XTEN, a 16 aa linker; NLS, nuclear localization signal; CaMV, cauliflower mosaic virus; Term, terminator. d Comparison of A-to-G conversion using the mGFP-to-GFP reporter system induced by various scRNAs and their cognate PABEcs in rice protoplasts (n = 3). Values and error bars indicate means ± s.e.m. of three independent experiments. e Comparison of the A-to-G editing frequencies of endogenous rice genes induced by five scRNAs and their cognate PABEcs (n = 3). An untreated protoplast sample served as control. Values and error bars indicate means ± s.e.m. of three independent experiments

Full size image

We proceeded to replace the C terminal MCP in PABEc3 with PCP, N22p, and Com, generating PABEc4-c6 (Fig. 2c; Additional file 2: Sequences S3). These cognate scRNAs were tested with PABEc3-c6 using the mGFP-to-GFP reporter system in rice protoplasts. To our surprise, the A-to-G activity of PABEc5 combined with esgRNA-2×boxB (26.5%) was increased and was comparable to that of PABE-7 combined with esgRNA (25.6%) (Fig. 2d; Additional file 1: Figure S8). The highest A-to-G activities in other RNA aptamer groups were PABEc3 combined with esgRNA-MS2+f6 (18.1%), PABEc4 combined with esgRNA-1×PP7-1 (21.0%), and PABEc6 combined with esgRNA-2×com (22.5%); however, they were still lower than that of PABE-7 combined with esgRNA (Fig. 2d; Additional file 1: Figure S8). As observed with the PBEcs using the scRNAs of RNA hairpins in the tetraloop and stem loop2 conformation, the A-to-G editing activities of PABEcs with these scRNAs also induced much lower GFP+ signals, ranging from 1.1 to 7.2% (Fig. 2d; Additional file 1: Figure S8).

We chose esgRNA-2×MS2, esgRNA-MS2+f6, esgRNA-1×PP7-1, esgRNA-2×boxB, and esgRNA-2×com to evaluate the effectiveness of PABEcs in converting A-to-G in endogenous rice genes. Six appropriate sgRNAs were inserted into these scRNAs, and they were co-transfected with cognate PABEcs into rice protoplasts (Additional file 1: Table S1). Of these combinations, PABEc5 combined with esgRNA-2×boxB had the highest A-to-G base editing efficiency (average 4.7%) in A4 to A8, which was lower than PABE-7 combined with esgRNA (average 7.6%), but was comparable to that of PABE-2 combined with sgRNA (average 4.8%), a similar construct used in human cells [26] (Fig. 2e; Additional file 1: Figure S9). Therefore, we selected PABEc5 combined with esgRNA-2×boxB as the scRNA for ABE in the multiplex genome editing system.

Multiplex genome editing with Cas9 nickase and RNA scaffolds

The successful development of scRNA-mediated CBE or ABE in rice protoplasts paved the way to multiplexed orthogonal CBE and ABE editing on different targets using the nCas9 (D10A) platform. To fully harness the properties of nCas9 (D10A)-mediated multiplex editing, we first set out to express three sgRNAs in a dual-function system designated as SWISS version 1.1 (SWISSv1.1) based on PBEc4, esgRNA-2×MS2, and paired sgRNAs [27], which should perform simultaneous cytosine base editing and generate paired-nCas9 mediated DSB (Fig. 3a). Toward develo** such a platform, we designed two sets of sgRNAs (Additional file 1: Table S2) and assembled multiple sgRNAs in the same vector under the OsU3 or TaU6 promoter (Additional file 1: Figure S10; Additional file 2: Sequences S4). C-to-T efficiencies ranged from 0.3 to 31.3% at C3 to C9, while indels efficiency ranged from 1.7 to 2.5% (Fig. 3a). Encouraged by the activity of SWISSv1.1, we used another dual-function strategy for adenine base editing and simultaneous DSB production based on PABEc5, esgRNA-2×boxB, and paired sgRNAs, designated as SWISS version 1.2 (SWISSv1.2) (Fig. 3b). A-to-G frequencies in the two tested groups reached 2.9%, and indels efficiency reached 2.5% (Fig. 3b). Moreover, with both SWISSv1.1 and SWISSv1.2, more than 79% of the indels reads were deletions induced by the paired nCas9 (D10A) (Additional file 1: Figure S11). These findings establish that scRNA-mediated CBE and ABE can induce multiple sgRNAs to perform base editing and indels dual-function, which shows that the paired nCas9 (D10A) provides an alternative way to induce indels when using PBEc4 and PABEc5.

Fig. 3
figure 3

CRISPR RNA scaffold-programmed simultaneous multiplex genome editing based on an nCas9 (D10A) platform in rice protoplasts. a Simultaneous CBE and DSB formation induced by PBEc4 with esgRNA-2×MS2, and paired sgRNAs. Left, schematic of the SWISSv1.1 strategy. Right, the two sets of sgRNAs tested (n = 3). A CBE target, esgRNA-2×MS2, and paired sgRNAs for creating DSB were assembled in the same vector. b Simultaneous ABE and DSB induced by PABEc5 with esgRNA-2×boxB, and paired sgRNAs. Left, schematic of the SWISSv1.2 strategy. Right, the two sets of sgRNAs tested (n = 3). One ABE target with esgRNA-2×boxB and paired sgRNAs for creating DSB were assembled in the same vector. c Simultaneous CBE and ABE induced by SWISSv2. Left, schematic of the SWISSv2 strategy. Right, the two sets of sgRNAs tested (n = 3). One CBE target with esgRNA-2×MS2 and one ABE target with esgRNA-2×boxB were assembled in the same vector. d Simultaneous CBE, ABE, and DSB induced by SWISSv3. Top, schematic of the SWISSv3 strategy. Bottom, the two sets of sgRNAs tested (n = 3). One CBE target with esgRNA-2×MS2, one ABE target with esgRNA-2×boxB, and paired sgRNAs for DSB were assembled in the same vector. e The scope of SWISSv2 and SWISSv3 multiplex genome editing strategies could be expanded by an nCas9-NG PAM variant. The two sets of sgRNAs for SWISSv2 and one set of sgRNA for SWISSv3 tested (n = 3). Multiple sgRNAs were assembled in the same vector. a–e An untreated protoplast sample served as control. Values and error bars indicate means ± s.e.m. of three independent experiments. f–i Numbers of total indels and SNVs (f), total indels (g), total SNVs (h), and total C-to-T SNVs (i) identified in the PBE and SWISS plants. Both of the pH-SWISSv2/v3 and pH-PBE binary constructs were transformed without the sgRNA cassette. Each dot represents the number of indels or SNVs from an individual plant. Horizontal lines and error bars indicate mean number of mutations ± SD (n = 6). P values were calculated by the two-tailed Mann-Whitney U test

Full size image

To determine whether scRNA-mediated base editing can promote dual-function CBE and ABE on different target sites simultaneously when designated as SWISS version 2 (SWISSv2) (Fig. 3c), we co-expressed nCas9 (D10A), MCP-APOBEC1-UGI, and ecTadA-ecTadA7.10-N22p using T2A under one Ubi-1 promoter (Additional file 1: Figure S12a; Additional file 2: Sequences S5). The esgRNA-2×MS2 for cytosine base editing was installed under the TaU6 promoter, and the esgRNA-2×boxB for adenine base editing was controlled by the OsU3 promoter (Additional file 1: Figure S12b). Two groups of targets were tested in rice protoplasts (Additional file 1: Table S2). Amplicon deep sequencing showed that SWISSv2 induced efficient CBE and ABE dual-function on two different targets at both groups. The C-to-T efficiency ranged from 1.8 to 13.2% in C3 to C9, and the A-to-G efficiency ranged from 0.5 to 4.3% in A4 to A8 (Fig. 3c).

Encouraged by the results above, we then introduced paired sgRNAs into SWISSv2 designated as SWISS version 3 (SWISSv3) (Fig. 3d) and tested two groups of target sites (Additional file 1: Figure S12c and Table S2). As expected, we observed that SWISSv3 acted as a programmable CBE, ABE, and DSB tri-functional editing system performing C-to-T (0.4–26.7%) and A-to-G (0.5–2.6%) editing on primary editing window of two targets and simultaneously creating indels (2.1–2.2%) at the other target (Fig. 3d). We also compared the editing efficiency of the SWISS system (SWISSv2 and SWISSv3) with PBE, PBEc4, PABE-2, PABEc5, and paired nCas9 (Additional file 1: Figure S13 and Table S2). The results showed that the base editing efficiencies of SWISSv2 and SWISSv3 (average C-to-T 15.3%; average A-to-G 2.0%) were lower than the original PBE (average C-to-T 15.5%) and PABE-2 (average A-to-G 5.1%) base editors, but comparable to the PBEc4 (average C-to-T 15.0%) and PABEc5 (average A-to-G 2.3%) (Additional file 1: Figure S13a, b). The indel efficiency of SWISSv3 (2.1%) was similar to that of paired nCas9 (D10A) (2.1%) (Additional file 1: Figure S13b). These data support that SWISS is a reliable multi-functional genome editing tool.

To release the requirement of PAMs and expand the editing scope of SWISSv2 and SWISSv3, we replaced the nCas9 (D10A) with nCas9-NG (D10A) PAM variant (VRVRFRR) [28], generating the NG version (Additional file 1: Figure S12a; Additional file 2: Sequences S5). Three groups of targets were tested, including two for SWISSv2 and one for SWISSv3 (Additional file 1: Table S2). The C-to-T (0.1–5.9%) and A-to-G (0.2–1.6%) editing on two targets were efficient in SWISSv2 of Cas9-NG PAM variant (Fig. 3e). Similarly, the C-to-T (up to 5.9%), A-to-G (up to 0.3%), and indels (0.6%) editing on three targets were also observed in SWISSv3 of Cas9-NG PAM variant (Fig. 3e). Taken together, SWISSv2 and SWISSv3 provide alternative tools for gene stacking and genetic modification in plants.

Simultaneous CBE, ABE, and indels formation in rice plants

To test the potential of SWISSv3 in rice plants, we used multiple sgRNAs targeting OsALS [29], OsACC [30], and OsBADH2 [31] and assembled into the binary vector (Additional file 1: Figure S14a). The editing targets in regenerated plants were examined by T7EI assay and confirmed by Sanger sequencing (Additional file 1: Figure S14b, c). Efficient C conversion (25.5%), A conversion (16.4%), and indels formation (52.7%) were evident in 55 regenerated rice seedlings (Table 1). Totally, 10 plants (18.1%) were involved in simultaneous dual-function editing on different targets, including 1 plant (1.8%) containing simultaneous C edits and A edits, 7 plants (12.7%) containing simultaneous C edits and indels, and 2 plants (3.6%) containing simultaneous A edits and indels (Table 2). Importantly, 4 plants (7.3%) contained simultaneous C edits, A edits, and indels at separate targets, and SWISSv3 also produced individual editing events with C edits (3.6%), A edits (3.6%), and indels (29.1%) at the three targets (Table 2).

Table 1 Mutation frequencies induced by SWISSv3 in T0 rice plants
Full size table
Table 2 Multiplex genome editing using SWISSv3 in T0 rice plants
Full size table

We also evaluated the potential for off-target effects; we searched the genomic sequence for all target sites that contained sequences with up to a 3-nt mismatch and sequenced these sites in the triple mutants. We found no off-target mutations in any of the triple mutants (Additional file 1: Table S3). Previous studies showed that APOBEC1-based CBE induced Cas9-independent genome-wide mutations in rice and mouse [32, 33]. To examine specificity of the SWISS systems, twelve transgenic rice plants expressing the SWISSv2/v3 or PBE without sgRNA were analyzed by whole genome sequencing at an average depth of 60× with high quality (Additional file 1: Table S4). We filtered out background mutations using ten wild-type plants. The results showed that there was no significant difference in the average number of total indels and single-nucleotide variants (SNVs) between the SWISS and PBE groups (Fig. 3f-i). Thus, the genome-wide Cas9-independent off-target effects of the SWISS system was comparable to that of PBE [32].

Taken together, our findings demonstrate that SWISSv3 acts as a tri-functional synthetic programmable genome editing system with CRISPR RNA scaffolds in plants. This ability will facilitate molecular design breeding in crops.

Discussion

Multiplex genome editing using multiple sgRNAs could be exploited in two ways; one is performing the same type of editing events on different targets [3, 34] and the other is performing a range of different editing events on different targets [11,12,13, 15]. Although Cas9 or Cas12a has been engineered as dual-functional genome editing system using a truncated sgRNA (or crRNA) and another full-length sgRNA (or crRNA), this strategy was restricted in editing gene regulation and cleave on different targets simultaneously [11,12,13]. Other multi-functional genome editing could also be achieved with CRISPR-Cas orthologs by recognizing different target sites [15, 17], but large cargo-capacity vector or co-delivery multiple vectors are needed and specific PAM sequences are required. In our SWISS system, we used a single Cas9 nickase (D10A) and different scRNAs for multiplex editing; the reduced vector size and the use of an NG PAM Cas9 variant would further expand the targeting scope of the SWISS systems.

In the studies described above, we used the RNA polymerase III promoters OsU3 and TaU6 to express multiple sgRNAs [35, 50] and TIDE [51].

Off-target analysis

Potential off-target sites were predicted using the online tool Cas-OFFinder [52]. Sites containing up to 3-nt mismatches were examined. The whole-genome sequencing assay and genome-wide Cas9-independent off-target analysis were conducted as reported [32].

Statistical analysis

GraphPad Prism version 7.0 was used for all data analysis. All numerical values are presented as means ± s.e.m. Statistical comparison adjustments were performed using two-tailed Mann-Whitney U test.