DNA methylation can alter CRISPR/Cas9 editing frequency and DNA repair outcome in a target‐specific manner

Summary

• The impact of epigenetic modifications on the efficacy of CRISPR/Cas9‐mediated double‐stranded DNA breaks and subsequent DNA repair is poorly understood, especially in plants.

• In this study, we investigated the effect of the level of cytosine methylation on the outcome of CRISPR/Cas9‐induced mutations at multiple Cas9 target sites in Nicotiana benthamiana leaf cells using next‐generation sequencing.

• We found that high levels of promoter methylation, but not gene‐body methylation, decreased the frequency of Cas9‐mediated mutations. DNA methylation also influenced the ratio of insertions and deletions and potentially the type of Cas9 cleavage in a target‐specific manner. In addition, we detected an over‐representation of deletion events governed by a single 5′‐terminal nucleotide at Cas9‐induced DNA breaks.

• Our findings suggest that DNA methylation can indirectly impair Cas9 activity and subsequent DNA repair, probably through changes in the local chromatin structure. In addition to the well described Cas9‐induced blunt‐end double‐stranded DNA breaks, we provide evidence for Cas9‐mediated staggered DNA cuts in plant cells. Both types of cut may direct microhomology‐mediated DNA repair by a novel, as yet undescribed, mechanism.

Short abstract

See also the Commentary on this article by Raffan et al., 235: 2146–2148.

Introduction

CRISPR/Cas9 has been extensively used in gene editing in eukaryotes, and many of the mechanistic details of Cas9‐induced double‐stranded breaks (DSBs) have been well described. Cas9‐mediated cleavage mostly occurs as a blunt DSB three nucleotide upstream of the protospacer adjacent motif (PAM) (Jinek et al., ²⁰¹²; Josephs et al., ²⁰¹⁵). However, animal and yeast studies indicate that the Cas9 can also induce staggered DSBs with single 5′ nucleotide overhangs (Li et al., ²⁰¹⁵; Lemos et al., ²⁰¹⁸). Double‐stranded breaks can be repaired by several endogenous DNA repair mechanisms. In eukaryotes, the predominant mechanism in differentiated cells is classical nonhomologous end joining (c‐NHEJ), in which blunt DNA ends are directly re‐ligated. Alternatively, short overhangs, resulting for example from CRISPR/Cas9 staggered cuts, can be filled in and then ligated forming short insertions (IN) (Beying et al., ²⁰²¹). Double‐stranded breaks can also be repaired through homology‐directed repair. When the repair relies on extensive nucleotide sequence similarity/homology (such as 20–100 bp) at 3′ DNA ends it comprises two main pathways: single‐strand annealing and homologous recombination (Roth et al., ²⁰¹²; Bothmer et al., ²⁰¹⁷; Beying et al., ²⁰²¹). The remaining dominant pathway is microhomology‐mediated end joining (MMEJ), also known as alternative end joining (Alt‐EJ), which relies on annealing of short complementary regions referred to as microhomologies present on both 3′ DNA ends close to the break, which can result in short deletions (DEL) (Chiruvella et al., ²⁰¹³; Beying et al., ²⁰²¹; Vu et al., ²⁰²¹).

Successful CRISPR/Cas9‐induced mutagenesis is conditioned by the selection of an appropriate target sequence. Several bioinformatics tools have been developed to aid the selection of optimal sequences to be targeted by Cas9. Many of these algorithms provide predicted mutagenic efficiencies (on‐target scores) and, if a whole‐genome sequence is available, some programs will also provide predictions of off‐target sequences (Liu et al., ²⁰¹⁷; Cui et al., ²⁰¹⁸). While these algorithms facilitate the design of gRNAs, the predicted on‐target efficiencies can often be inaccurate (Naim et al., ²⁰²⁰), indicating that our understanding of the principles underlying Cas9 binding and cleavage of target DNA and subsequent repair of DSB in vivo is far from complete. A possible reason for the discrepancies between the in silico predictions and experimental results could be that most of the gRNA design tools do not consider the epigenetic status of the target locus. It is likely that the chromatin structure of a locus (in addition to the underlying base sequence) will be an important determinant for the efficacy of Cas9 binding, cleavage and subsequent repair. Indeed, there have been reports that the frequency of Cas9‐induced mutagenesis is lower for heterochromatin compared with euchromatin (Chen et al., ²⁰¹⁶; Daer et al., ^{2017, 2018}). However, other reports have suggested that Cas9‐induced mutation frequencies are comparable for heterochromatin and euchromatin (Yu et al., ²⁰¹³; Feng et al., ²⁰¹⁶; Kallimasioti‐Pazi et al., ²⁰¹⁸). This indicates that epigenetic factors influencing the CRISPR/Cas outcomes are more complex than just considering the broad heterochromatin/euchromatin dichotomy (Schep et al., ²⁰²¹). For instance, it has been shown that the epigenetic availability of the PAM sequence is crucial to the action of Cas9, such that PAMs associated with nucleosomes cannot be located by Cas9 (Hinz et al., ²⁰¹⁵; Horlbeck et al., ²⁰¹⁶; Yarrington et al., ²⁰¹⁸). However another epigenetic effect that could influence the ability of Cas9 to find and cleave DNA is the methylation status of the target locus. While DNA methylation at the PAM was found to have no effect on Cas9 activity in human cell lines (Hsu et al., ²⁰¹³; Fujita et al., ²⁰¹⁶), this phenomenon is yet to be tested in other organisms.

Here we present a detailed study on Nicotiana benthamiana in which we use Tobacco Rattle Virus (TRV) as a molecular switch to change the chromatin state of a reporter gene from an actively transcribed to a transcriptionally silenced state. Unlike previously used systems, our unique approach enabled us to interrogate different chromatin states of the same locus with the same set of CRISPR/Cas9 genome editing reagents and systematically describe the effect of chromatin state on the frequency and type of mutations induced at various Cas9 targets in a huge set of independently edited cells.

Materials and Methods

Generation of CRISPR/Cas9 gene‐editing constructs

Cas9 target sites were identified by Geneious v.11.1.5 (http://www.geneious.com) (Kearse et al., ²⁰¹²) and potential off‐targets were analysed by Sol Genomics Network Blast (https://www.solgenomics.net/tools/blast) (Fernandez‐Pozo et al., ²⁰¹⁵) and CRISPR RGEN Cas‐OFFinder (http://www.rgenome.net/cas‐offinder) (Bae et al., ²⁰¹⁴). The corresponding oligonucleotides with BbsI compatible overhangs (Supporting Information Table S1) were annealed into a dsDNA duplex, phosphorylated with T4 Polynucleotide Kinase (New England BioLabs, Hitchin, UK) and subsequently cloned into BbsI‐digested entry vector pEn_Chimera (Fauser et al., ²⁰¹⁴). The resulting gRNA expression cassettes (pEn_Chimera_x; x = ut, ub, pt, pb, mt, mb, dt, db, utg, mbg, dtg; Table S2) were verified by Sanger sequencing. The BASTA resistance cassette (CaMV35S::bar::nos) was deleted from the original destination binary vector pDe_CAS9 (Fauser et al., ²⁰¹⁴) by HindIII restriction enzyme digestion and re‐ligation to avoid interference with the 35S sequence. The gRNA expression cassettes were then recombined into the modified destination vector (pDe_CAS9_no_basta) by Gateway cloning (LR reaction) according to the manufacturer’s instruction (Life Technologies, Carlsbad, CA, USA), which yielded the pDE_CAS9_no_basta_x binary vectors (Table S2) for plant transformation.

Plant material

Wild‐type N. benthamiana plants, N. benthamiana 16c line expressing the green fluorescent protein (GFP) reporter gene under the constitutive CaMV 35S promoter (Ruiz et al., ¹⁹⁹⁸) and its transcriptionally silenced progeny (S1: the virus‐free progeny of TRV‐35S‐infected 16c; Fig. 1) (Fei et al., ²⁰²¹) were grown in SANYO/Panasonic growth chambers with 16 h : 8 h, light : dark photoperiods at 21°C under 150 μmol m⁻² s⁻¹ light.

Fig. 1.

Virus‐induced epigenetic switch system. (a) Schematic diagram of the GFP reporter gene under the control of the constitutive CaMV promoter P35S. (b) Schematic diagram of Tobacco rattle viruses (TRV): wild‐type virus (TRV‐WT) and its derivatives harbouring a 120‐bp segment from P35S (TRV‐35S) and GFP (TRV‐GFP). (c) Scheme of the four plant conditions used for the analysis of epigenetic changes at the 35S promoter. Noninfected and TRV‐WT‐infected plants contain transcriptionally active euchromatin, while the TRV‐35S‐infected plants and their noninfected progeny (S1) have induced chromatin silencing at the 35S promoter. (d) Scheme of GFP expression for the different plant conditions. 16c plants express GFP; TRV‐WT plants have partially, nonspecifically reduced GFP expression due to ongoing virus infection; TRV‐35S‐infected plants exhibit reduced GFP expression due to TRV‐35S‐mediated formation of sRNAs that trigger dense cytosine methylation in the target 120‐bp region of P35S; S1 plants have a slightly reduced GFP expression due to inherited DNA methylation at the 35S promoter. (e) Fluorescence of GFP in representative plants of the different plant conditions, imaged under UV light. Pictures of plants from all biological replicates can be found in Supporting Information Fig. S1. (f) Quantification of GFP mRNA and TRV RNA by RT‐qPCR for the different plant conditions. **, * and ns represent significant difference for P < 0.01, P < 0.05 and not significant, respectively. Individual pairs were analysed by two‐tailed t‐test with α = 0.05. Error bars represent standard deviations. (g) Average of methylated cytosines in the 120‐bp target region in individual epigenotypes and sequence contexts from three independent biological replicates. (h) DNA cytosine methylation of P35S in detail for the four plant conditions. A top schematic diagram of P35S shows the position of the TRV‐35S target and positions of CRISPR/Cas9 target sites (colour‐coded CRISPR/Cas9 target sites ut, ub, pt, pb, mt, mb, dt, db are explained in Fig. 2). Only cytosines that are present in the leading strand are shown. nd, not done; the bottom bar is the sequence context of individual cytosines.

Viruses and virus sap

Wild‐type TRV, recombinant TRV‐35S and TRV‐GFP, and virus sap collection from N. benthamiana plants have been described in Fei et al. (2021).

Transient expression of gene‐editing reagents in planta

Agrobacterium tumefaciens AGL1 strain was transformed with pDe_Cas9_no_basta_x vectors (Table S2) by electroporation and selected on solidified LB plates using 100 µg ml⁻¹ spectinomycin and 50 µg ml⁻¹ rifampicin. For each construct, 200 µl of overnight starter culture was added to 10 ml LB supplemented with the above antibiotics and incubated for 24 h at 28°C and 11 g . The bacterium culture was then centrifuged at 3000  g for 10 min, and the pellet was resuspended in infiltration medium (IM, 10 mM MES (pH 5.6), 300 µM acetosyringone, 10 mM MgCl₂) and incubated for 3–4 h at room temperature. The optical density was adjusted to 1.5 and 0.15 OD₆₀₀ by adding a fresh IM buffer. An Agrobacterium suspension was infiltrated with a 1‐ml syringe into the abaxial side of three leaves (4^th–6^th youngest leaves) of 5‐wk‐old N. benthamiana. For agro‐infiltrating virus‐infected plants, the first and the second leaf of 3‐wk‐old N. benthamiana were rub‐inoculated with the corresponding virus sap as previously described (Fei et al., ²⁰²¹) and the plants were kept under standard conditions (described in ‘Plant material’ in the Materials and Methods section) for another 2 wk before agro‐infiltration. All experiments were independently repeated three to five times.

Imaging GFP fluorescence

GFP expression was regularly monitored under UV light using a handheld mercury UV lamp (UVP Blak‐Ray™ B‐100AP High‐Intensity UV Inspection Lamp, Upland, CA, USA). Photographs were taken using a Canon G16 camera as described previously (Fei et al., ²⁰²¹).

Sample collection

A leaf disc (8 mm in diameter) was cut out from each infiltrated leaf (4^th, 5^th and 6^th) and combined into a single biological sample 5 d post infiltration (dpi). Three samples were collected from each plant, frozen in liquid nitrogen and stored at −80°C until further use.

Mutation analysis

DNA was isolated using the DNeasy Plant Mini Kit (Qiagen). A T7 endonuclease assay was carried out as previously described (Pyott et al., ²⁰¹⁶). For assessing the frequency of gene editing, the Cas9 targeted loci were amplified with the corresponding primers (Table S1), and the resulting DNA was cleaned up using SAP‐ExoI digestion (NEB). Sanger sequencing was performed using the BigDye Terminator v.3.1 Cycle Sequencing Kit (ThermoFisher Scientific, Waltham, MA, USA) and completed on an ABI 3730XL instrument by Edinburgh Genomics (Edinburgh, UK). The efficacy of gene editing was calculated using Tide (https://tide.nki.nl/) (Brinkman et al., ²⁰¹⁴). For creating a next‐generation sequencing (NGS) library, the target loci were amplified with the corresponding Illumina compatible primers (Table S1), which were subsequently barcoded using the Nextera XT Index Kit v.2 (Illumina, San Diego, CA, USA). The NGS library was sequenced on an Illumina MiSeq instrument. The data were analysed using the CRISPResso2 tool (Clement et al., ²⁰¹⁹) followed by additional analysis using in‐house Python scripts (release 3.7.5) and MS Excel 365 (Microsoft, Redmond, WA, USA).

RNA analysis

RNA was isolated using the RNeasy Plant Mini Kit (Qiagen). After Turbo DNase treatment (Ambion), cDNAs were made from 0.5 μg RNA using random hexamers primers and Superscript III (Life Technologies). Quantitative RT‐PCR was performed using the SYBR Green I Master Mix on a LightCycler480 (Roche) using gene‐specific primers (Table S1). Data were converted with Lc480 software (v.2014.1), and C_t values were calculated using LinRegPCR (v.2012.0) (Ramakers et al., ²⁰⁰³), both software are available at https://www.medischebiologie.nl/files/. The analysis was carried out in MS Excel 365 (Microsoft). Three technical replicates were performed for each biological replicate.

DNA methylation analysis

Genomic DNA was converted using the EZ DNA Methylation‐Gold Kit (Zymo Research, Irvine, CA, USA). Target loci were amplified using bisulfite PCR primers (Table S1) and EpiMark^® Hot Start Taq DNA Polymerase (NEB). The resulting PCR products were ligated into the pGEM^®‐T Vector (Promega), which were then used to transform Escherichia coli (DH5α) cells. In total, 8–10 clones from each sample were sequenced using BigDye 3.1, as described previously. DNA methylation patterns were analysed using Geneious v.11.1.5 (http://www.geneious.com) (Kearse et al., ²⁰¹²).

Statistical analysis

Statistical analyses were performed using the Ncss 9 and Spss 28 statistical software. Significance level α was always set to 0.05. All variables were checked for outliers, and assessed for normality using the Shapiro–Wilk test and Levene’s test for equal variance. A one‐tailed or two‐tailed two‐sample independent t‐test was used in which the two independent groups were compared, based on expectations from the groups being compared. To determine whether the independent variables had an interaction, two‐way analysis of variance (ANOVA) tests were performed, followed by Tukey’s post hoc test if possible.

Results

Virus‐induced gene silencing can be used as an ‘epigenetic switch’ to alter the cytosine methylation status of a target locus

To test how chromatin state affects the efficacy of CRISPR/Cas9 mutagenesis in planta, we developed a system that allowed us to change the chromatin state of a single locus and subsequently analyse CRISPR/Cas9‐induced mutations at that locus. In this system, we made use of a recombinant RNA virus to induce DNA methylation by triggering the plant’s RNA‐dependent DNA methylation (RdDM) pathway at the target locus. The RdDM pathway can be triggered by some viruses when short interfering RNAs (siRNAs) are produced by the plant’s antiviral RNA silencing response. Certain classes of these virus‐derived siRNAs can direct plant‐encoded DNA methyl transferases to methylate cytosine residues of DNA that bear sequence complementarity to the siRNA. Plant viruses can be turned into virus‐induced gene silencing (VIGS) vectors if sequences present in the plant’s genome are cloned into the viral genome. Infection with such viruses will result in the production of siRNAs that are complementary to the region of the plant’s genome that was inserted into the virus. If a promoter sequence is inserted into the VIGS vector, the siRNA‐mediated RdDM can lead to cytosine methylation of that promoter and consequently result in transcriptional gene silencing (TGS) (Jones et al., ²⁰⁰¹; Přibylová et al., ²⁰¹⁹).

For our experimental system, we utilised a previously described and well characterised transgenic N. benthamiana line (named 16c) that contained a single‐copy genomic insertion of the GFP reporter gene under the constitutive 35S promoter (Fig. 1a) (Ruiz et al., ¹⁹⁹⁸). In the 16c line, the 35S::GFP cassette is highly expressed, implying that the transgenic cassette is integrated into an epigenetically active region of the plant’s genome. In a previous study (Fei et al., ²⁰²¹), we adapted a TRV VIGS vector (Jones et al., ²⁰⁰¹) to induce RdDM of the 35S promoter by targeting a short segment (120 bp) upstream of the transcription start site and therefore transcriptionally silence the expression of GFP in 16c plants. We named this vector TRV‐35S (Fig. 1b). To control for possible effects of the virus infection itself (unrelated to the VIGS‐induced RdDM of the target) we used a TRV vector without the 120‐bp section of the 35S promoter inserted and named this vector TRV‐WT (Fig. 1b). Furthermore, we collected seeds from the TRV‐35S‐infected plants to allow us to evaluate the effects of cytosine methylation at the target locus in plants containing no viral RNA. The basis of this is that TRV transmission to progeny seedlings was extremely low (< 1%) (Jones et al., ²⁰⁰¹) but cytosine methylation that is in a symmetrical sequence context (i.e. CG or CHG, where H represents A, C or T) can be maintained through both mitotic and meiotic cell divisions and exhibits transgenerational inheritance (Jones et al., ²⁰⁰¹). We use the term S1 to refer to these plants grown from the seeds collected from TRV‐35S‐infected plants as they are the first‐generation progeny of the transcriptionally silenced plants (Fig. 1c). Therefore, our system allows for the comparison of two different epigenotypes (high/low cytosine methylation) at the 35S promoter in the context of virus‐infected and nonvirus‐infected tissue (Fig. 1d).

First, we tested the efficacy of our TRV‐based ‘epigenetic switch’. When imaged under UV light, 16c plants that had been rub‐inoculated with TRV‐35S glowed red due to the autofluorescence of chlorophyll and the green glow of GFP was completely absent. By contrast, for both the noninfected and TRV‐WT‐infected 16c plants a green GFP signal was discernible and was particularly visible in the upper leaves and petioles. A weak GFP signal could be seen in the S1 plants, indicative of only a partial silencing of GFP (Figs 1e, S1). This silencing of the GFP expression was confirmed in three randomly selected independent biological replicates by RT‐qPCR. TRV‐WT‐infected plants had a significantly higher level of GFP mRNA compared with TRV‐35S‐infected plants (P = 0.006, two‐tailed t‐test), which had negligible levels of GFP mRNA. The noninfected 16c plants had more than twice the level of GFP mRNA than S1 plants (P = 0.011, two‐tailed t‐test), but there was still a weak expression of GFP mRNA in the S1 plants (Fig. 1f). Moreover, we did not observe any difference in the abundance of TRV RNA between the TRV‐35S and TRV‐WT infections (P = 0.441, two‐tailed t‐test).

To confirm that the basis of the observed GFP silencing was a result of VIGS‐induced RdDM, we performed bisulphite sequencing. DNA samples taken from TRV‐WT‐infected plants showed almost no methylation in the 120‐bp target region, and weak methylation was observed in the upstream promoter region, mainly in the CG context. By contrast, DNA samples from TRV‐35S‐infected plants showed strong methylation in the target region with a weak spreading of methylation towards 3′ and 5′ ends of the 35S promoter, independent of the C sequence context. In the upstream part of the promoter, we observed increased methylation in the symmetrical CG and CHG context, mainly in the region where even TRV‐WT‐infected plants showed some CG methylation. As expected, noninfected plants showed a similar methylation pattern as TRV‐WT‐infected plants: no methylation in the target region and limited methylation in the upstream region of the promoter. S1 plants inherited an increased level of mostly symmetric CG and CHG methylation within the target region of the 35S promoter, but only at some C positions (Fig. 1g,h). Taken together, our data show that infection with TRV‐35S can switch the 35S::GFP transgene of 16c plants from an epigenetically active to an epigenetically inactive (silenced) state due to virus‐induced methylation of the 35S promoter and that this silenced epigenotype is partially inherited to the virus‐free progeny.

The efficacy of CRISPR/Cas9 editing is inversely correlated with the level of cytosine methylation for targets within a promoter region

Having demonstrated that we could alter the cytosine methylation status of a promoter by VIGS, we sought to use this as a tool to determine how DNA methylation at a given locus might influence the frequency of CRISPR/Cas9‐induced mutations (Fig. 2a). To this end, we designed CRISPR gRNAs that would direct Cas9 to the region of the 35S promoter that we had targeted for RdDM using TRV‐35S (Fig. 2b). Additionally, we designed gRNAs to direct Cas9 to a region of the 35S promoter that was upstream of the TRV‐35S targeted region but still showed an increase in VIGS‐induced cytosine methylation (Figs 1h, 2b). Within the TRV‐35S targeted region of the promoter, we designed gRNAs to direct Cas9 to three positions which we referred to as proximal, middle and distal (based on their position relative to the start of the promoter). For each of the four regions of the 35S promoter (namely: upstream, proximal, middle and distal) we designed pairs of gRNAs that would anneal to each of the two complementary strands of the target DNA, which for convenience we referred to as the top and bottom strands. As far as possible we designed these pairs of gRNAs so that they would overlap, although we could not obtain a 100% overlap of the gRNAs directed to the top and bottom DNA strands due to the constraints of PAM availability. For ease of reference, we named each of the gRNAs using acronyms to describe both their position in the 35S promoter (upstream/proximal/middle/distal) and their cognate DNA strand (top/bottom). Therefore, we named the eight gRNAs ut, ub, pt, pb, mt, mb, dt, db (Fig. 2b).

Fig. 2.

The impact of epigenotype on CRISPR/Cas9‐induced mutation frequency. (a) Summary of the experimental design. Here, 3‐wk‐old 16c plants were infected with the Tobacco Rattle Virus (TRV) constructs. Noninfected plants were used as controls. At 2 wk later, the newly emerging leaves were infiltrated with Agrobacterium tumefaciens carrying the corresponding CRISPR/Cas9 constructs. Samples were collected 5 d post infiltration (dpi) and were subsequently analysed for the efficacy of CRISPR/Cas9‐mediated gene editing. (b) Schematic diagram of the CRISPR/Cas9 target sites at the P35S::GFP::NosT locus. gRNAs were named according to the position of the protospacer adjacent motif (PAM) sequence (circled). t, top; b, bottom; u, upstream; p, proximal; m, middle; d, distal. The region of P35S targeted for induced cytosine methylation is indicated with a black line. (c) CRISPR/Cas9‐induced mutation frequencies in plants infected with TRV‐WT or TRV‐35S (left panel), and in noninfected 16c and TRV‐35S‐noninfected progeny (S1) plants (right panel). Individual points (o, x, □) represent next‐generation sequencing (NGS) data (OD₆₀₀ = 0.15) from three independent biological replicates. Lower range, box, upper range and a horizontal line represents the first quartile, second and third quartile, fourth quartile and median, respective. Two‐way ANOVA was used to examine the effect of the target position and epigenotype on mutation frequency and showed a statistically significant interaction between the target position and the epigenotypes (F(7, 32) = 3.017, P = 0.015) in infected plants. By contrast, the interaction was not significant for noninfected plants (F(7, 32) = 0.960, P = 0.477). For both groups, infected and noninfected, the simple and main effect of the target site was significant (P < 0.001). To examine the effect of the epigenotype in the individual target site, a one‐tailed t‐test was used, as individual target sites had only two independent variables. * and ns indicate significant difference for P < 0.05 and not significant, respectively.

To test the in planta activity of these gRNAs, we cloned them individually into the pDe‐Cas9 binary vector under the control of the Arabidopsis thaliana U6 promoter. This binary vector also contained a Cas9 expression cassette driven by the constitutive Ubiquitin4–2 promoter from Petroselinum crispum (PcUbi4‐2) (Fauser et al., ²⁰¹⁴). Syringe infiltration of A. tumefaciens cultures containing these gRNA and Cas9 expression vectors was used to transiently transform the leaves of 16c plants with these CRISPR/Cas9 constructs. To optimise the conditions for A. tumefaciens‐mediated delivery of these constructs, we performed each infiltration at two different densities of bacterial suspension (namely OD₆₀₀ = 1.5 and OD₆₀₀ = 0.15). DNA was extracted from the infiltrated leaves and a T7 assay was performed to semiquantitatively assess the mutagenic capacity of each of the gRNA‐containing constructs. Encouragingly, the target amplicons digested with T7 endonuclease produced the expected cleavage products for all of the gRNA constructs, indicating that all of the gRNAs are capable of directing Cas9 to the respective target loci to induce mutations using both the high and the low bacterium densities (Fig. S2A).

Next, we used our VIGS system to induce methylation at the 35S promoter to determine any effects of this induced methylation on the mutagenesis frequency of each of the gRNAs. Here, 16c plants were infected with either TRV‐35S or TRV‐WT and were incubated for 2 wk to allow the infection to become systemic and initiate RdDM. Newly formed leaves of these systemically infected plants were then infiltrated with the different gRNA‐containing CRISPR/Cas9 constructs (Fig. 2a). In parallel, noninfected plants and S1 plants were infiltrated with the same CRISPR/Cas9 constructs to analyse the active/silenced epigenotypes in the absence of TRV infection. DNA was then extracted from the infiltrated leaves to determine the rate of mutagenesis under each of the different conditions. To obtain more detailed and quantitative data than would be possible using a T7 assay, we analysed the samples by Sanger sequencing. First, we plotted the mutation frequencies for all eight gRNAs infiltrated at either OD₆₀₀ = 1.5 or OD₆₀₀ = 0.15 for each of the four different plant variants, from this point forwards referred to as epigenotypes (TRV‐WT‐infected, TRV‐35S‐infected, noninfected and S1). In general, regardless of the epigenotype, higher bacterial densities resulted in higher mutation frequencies (P < 0.001, two‐tailed t‐test). However, differences in mutation rate between the epigenotypes could be seen mainly with the low bacterial density and we assumed that they were partially masked when the plants were infiltrated with the higher bacterial density (Fig. S2B). Interestingly, TRV‐WT‐infected plants exhibited equally high mutation frequencies for both bacterial densities (P = 0.575, two‐tailed t‐test) and the mutation frequencies for the TRV‐WT‐infected plants were higher than in any of the other plant variants, regardless of the bacterial density used for delivering the CRISPR/Cas9 components. When comparing noninfected and TRV‐WT‐infected 16c plants, which were infiltrated at OD₆₀₀ = 0.15, there was a significant difference in mutation frequency (P = 0.001, two‐tailed t‐test). This suggests that the TRV infection somehow predisposes the plant to a lower fidelity of DNA repair, which we speculate could arise from the plant’s stress response to the virus infection or even interference of TRV RNA with the DNA repair pathways. This result highlighted the importance of our experimental controls to accurately interpret our data. Due to the observed effect of TRV infection on mutation frequency, it is only meaningful to make pairwise comparisons between the TRV‐WT‐infected and TRV‐35S‐infected plants or between the noninfected and S1 plants when assessing the effect of chromatin state.

Based on the global analyses outlined above, we chose to analyse the mutation frequencies for each gRNA individually for the four different plant variants using the lower bacterial density (OD₆₀₀ = 0.15). To accurately discern the effects of promoter methylation on the efficacy of CRISPR/Cas9‐induced gene editing for each locus, we analysed the mutation frequency of the corresponding amplicons using TIDE‐coupled Sanger sequencing (Fig. S2C) and subsequently by Illumina NGS (Fig. 2c) using at least three independent biological replicates. In agreement with previous findings (Sentmanat et al., ²⁰¹⁸), there was a strong correlation between the Sanger and NGS datasets for all epigenotypes and targets, although mutation frequencies determined by NGS were, on average, 1.14 times higher than the estimates determined by Sanger sequencing (Fig. S3). Comparing the TRV‐WT‐infected and TRV‐35S‐infected plants, the mutation frequency was clearly higher for the promoter with active chromatin compared with the silenced, methylated promoter and this trend was consistent across the majority of targets, with pb, mt, dt and db targets showing statistically significant differences (P < 0.05, one‐tailed t‐test; Figs 2c, S2C, left panels). Interestingly, the difference in mutation frequency became significant for ut when the number of biological repeats was increased from three (NGS) to five (Sanger; Fig. S2C).

When we compared the mutation frequencies for gRNAs targeted to either the top or bottom strand for a given locus, we found no global correlation between the orientation (top or bottom) of the target and the mutation frequency (data not shown). As the impact of cytosine methylation on the mutation frequency in the promoter region was significant, when comparing the nonmethylated TRV‐WT with heavily methylated TRV‐35S plants (P = 0.009, two‐tailed t‐test; Figs S5, S6), we were interested to also assess the impact of cytosine methylation of transcribed DNA on the efficacy of CRISPR/Cas9‐mediated gene editing. To this end, we induced RdDM at the coding region of GFP with a recombinant TRV VIGS vector (TRV‐GFP) and subsequently analysed the frequency of Cas9‐induced mutations by amplicon sequencing (Notes S1; Fig. S4). We observed no statistically significant difference in mutation frequency between the TRV‐GFP and TRV‐WT (control) samples for individual target sites (Fig. S4) and no statistically significant effect of the cytosine methylation on the mutation frequency (P = 0.640, two‐tailed t‐test; Figs S5, S6), implying that the TRV‐GFP‐induced cytosine methylation has little or no effect on the frequency of CRISPR/Cas9‐induced mutations for three GFP targets including dtg (Figs 2b top panel, S4G,H, S5, S6). Taken together, our data showed that the frequency of CRISPR/Cas9‐induced mutations at the promoter DNA was generally inversely correlated with its level of cytosine methylation, but this phenomenon did not apply to coding regions of DNA, suggesting that the effect of DNA methylation on the efficacy of gene editing was rather indirect (see the second paragraph of Discussion section).

The mutational outcome of CRISPR/Cas9 editing is affected by DNA methylation

Next, we were interested to see if DNA cytosine methylation could affect not only the frequency of CRISPR/Cas9‐induced mutations but also the mutational outcome. To this end, we used the output of CRIPSResso2 software (Clement et al., ²⁰¹⁹, p. 20) to analyse our NGS data to resolve the specific features of insertion (IN), deletion (DEL) and substitution (SUB) events within our dataset. SUB events occurred along the whole analysed DNA regions without any specific accumulations near the predicted Cas9 cut sites, including negative controls that had been infiltrated with a CRISPR/Cas9 cassette with no gRNA, and accounted for c. 2% of reads. All these data indicated that SUB events can be considered as sequencing or PCR errors, therefore they were omitted from subsequent analyses.

Detailed analysis of total reads of IN and DEL events in individual target sites showed that each target site had a specific pattern of IN : DEL ratios. At some target sites IN dominated over DEL and vice versa (Figs 3, S7A). Interestingly, for three sites, the IN : DEL ratio was obviously shifted in connection with RdDM, either in favour of insertions (mt and db site) or deletions (dtg site; Fig. S7B). Moreover, we found that RdDM had a significant effect on the frequency of insertions and deletions at several individual target sites (pb and dt sites: Fig. 3a, left panel; and pb, mt, dt, db and dtg sites: Fig. 3b, left panel).

Fig. 3.

The impact of epigenotype on CRISPR/Cas9‐induced mutations. (a) Frequencies of insertion and (b) deletion events at each target site and epigenotype. Individual points (o, x, □) represent next‐generation sequencing data (OD₆₀₀ = 0.15) from three independent biological replicates. Lower range, box, upper range and a horizontal line represents the first quartile, second and third quartile, fourth quartile and median, respective. Two‐way ANOVA was used to examine the effect of the target position and epigenotype on the frequency of insertions and deletions and showed statistically significant interaction between the target position and the epigenotypes (F(8, 36) = 5.097, P < 0.001 and F(8, 36) = 7.347, P < 0.001, respective) in infected plants. The interaction was not significant for noninfected plants nor for insertions or deletions (F(8, 36) = 1.094, P = 0.390 or F(8, 36) = 1.087, P = 0.394). For both groups, infected and noninfected, the simple and main effect of the target site was significant (P < 0.001). To examine the effect of the epigenotype in the individual target site a one‐tailed t‐test was used, as individual target sites had only two independent variables. * and ns indicate significant difference for P < 0.05 and not significant, respectively. Terms TRV‐WT, TRV‐35S, noninfected and S1 are explained in Fig. 1; colour‐coded CRISPR/Cas9 target sites ut, ub, pt, pb, mt, mb, dt, db are explained in Fig. 2 and dtg in Supporting Information Fig. S4.

When we analysed IN mutations, c. 98% of them represented single‐nucleotide insertions (Fig. S8) which were located just next to the predicted Cas9 cut site. We observed a strong bias within the inserted nucleotides towards the 4^th nucleotide upstream of the PAM at most target sites, which is consistent with a previous report from a nonplant model that Cas9 can produce either blunt ends or staggered ends with single‐nucleotide (the 4^th nucleotide from PAM) 5′ overhangs (Figs 4, S9) (Li et al., ²⁰¹⁵). The only exceptions were pb and mb, for which nontemplated T and A nucleotides were the most prevalent single‐nucleotide insertions over the templated Gs (Figs 4, S9). Interestingly, the identity of the inserted nucleotide also appeared to be affected by the epigenotype to varying degrees for different target loci, with the biggest effect seen for the site targeted by the dt gRNA (Figs 4, S9).

Fig. 4.

Detailed analysis of single‐nucleotide insertion events detected next to the Cas9 cut site. 1‐bp insertion events were split into four categories, one each for every nucleotide (A, C, G, T). The top bar shows which nucleotide was in the original sequence on the 4^th position upstream from the protospacer adjacent motif (PAM) sequence. Each data point represents the mean of next‐generation sequencing data (OD₆₀₀ = 0.15) from three independent biological replicates. Terms TRV‐WT, TRV‐35S, noninfected and S1 are explained in Fig. 1; colour‐coded CRISPR/Cas9 target sites ut, ub, pt, pb, mt, mb, dt, db are explained in Fig. 2 and dtg in Supporting Information Fig. S4.

We then evaluated DEL events based on the length and position of deletions relative to the predicted CRISPR/Cas9 cut site and the PAM. We found that only a small fraction of reads (3%) contained mutations outside a 20‐nt window around the Cas9‐induced DSB (Group 5; Fig. 5a). Close inspection of DEL events within this 20‐nt window resulted in the identification of four groups. DEL events in group 1 spanned the cut site in both a PAM‐proximal and PAM‐distal direction. Group 2 contained DEL events that were exclusively in the PAM‐distal region adjacent to the Cas9 cut site. Group 3 contained DEL events that could be mostly mapped on the PAM distal end where terminal nucleotides can be mapped to either side of the Cas9 cut site. Finally, group 4 contained DEL events that were exclusively in the PAM‐proximal region adjacent to the Cas9 cut site (Fig. 5a,b). We found the distribution of mutations within each of these four groups to be variable for the different gRNAs with no noticeable effect of the epigenotype. Groups 1, 2, 3 and 4 represented 1%, 61%, 27% and 7% of the total reads, respectively, indicating that most deletions occurred in the PAM‐distal region adjacent to the Cas9 cut site (Figs 5c, S10). Intriguingly, we found that in many cases, especially in group 2, the deletion was insulated from the predicted Cas9 cleavage site by at least one additional nucleotide, which was almost exclusively aligned to the 4^th nucleotide upstream of the PAM. This observation is consistent with Cas9‐induced staggered ends with single‐nucleotide 5′ overhang (see Fig. 4), which mostly did not vary between epigenotypes (Figs 5d, S11).

Fig. 5.

Detailed analysis of deletion events detected next to the Cas9 cut site. (a) Schematic of mutation categories: (1) deletions on both DNA ends, (2) only on the protospacer adjacent motif (PAM) distal end, (3) mostly on the PAM distal end where terminal nucleotides can be mapped to either side of Cas9 cut DNA, (4) only on proximal DNA end, and (5) not directly adjacent to the cut site. (b) Schematic of mutations at nucleotide resolution for categories 1–4 using representative sequences from the ut next‐generation sequencing (NGS) data as an example. (c) Type and frequency of deletion mutations in CRISPR/Cas9 cut sites. (d) Detailed analysis of group 2 and its first adjacent position next to the cut site on the PAM distal DNA end. Position can contain plain deletion or a mutation with a remaining single nucleotide (A, C, G or T) adjacent to the predicted double‐stranded break. The top bar represents which nucleotide was in the original sequence on the 4^th position upstream from the PAM sequence. The top panel shows a detailed example of ‘ut’ gRNA target for a TRV‐WT‐infected sample. Each value represents the mean of NGS data (OD₆₀₀ = 0.15) from three independent biological replicates. Terms TRV‐WT, TRV‐35S, noninfected and S1 are explained in Fig. 1; colour‐coded CRISPR/Cas9 target sites ut, ub, pt, pb, mt, mb, dt, db are explained in Fig. 2 and dtg in Supporting Information Fig. S1.

DNA ends with single‐stranded overhangs in nondividing cells are prone to be repaired by the MMEJ pathway if an identical sequence of 2–25 nucleotides is present at both 3′ DNA ends near the cut site (Allen et al., ²⁰¹⁹; Beying et al., ²⁰²¹). DNA repair by this MMEJ mechanism results in a deletion (in the simplest scenario that does not involve insertions) that spans one of the two matching microhomology regions and the sequence in between them. Consequently, the preserved copy of the microhomology region can be mapped to either side of the deletion, and this mutational signature can be used to infer likely cases of MMEJ‐directed repair. In our dataset, we observed increased occurrence of such deletions only for ub and dt targets, in which trinucleotides present at the terminus of the PAM strand were repeated on the PAM‐distal DNA end near the Cas9 cut site (Fig. 6a). This indicated that the MMEJ repair pathway did not play a significant role in our experimental system, which is consistent with its predominant activity in dividing cells. However, the surprisingly high occurrence of 1‐nt microhomologies in our dataset (Fig. 6a) implied that a significant number of the deletions were formed by a single‐nucleotide‐dependent microhomology repair distinct from the classical MMEJ pathway. Moreover, when we analysed the 1‐nt microhomologies in more detail for individual gRNA targets, we noted that they nearly always aligned to the nucleotide that was either 3^rd or 4^th nt upstream of the PAM (Figs 6b, S12). This again was consistent with our conclusion that Cas9 could cut either in front of the 3^rd or at the 4^th nt upstream of the PAM on the PAM‐containing strand. When we considered all deletion mutations in our dataset, we observed an enrichment of deletion sizes corresponding to the distance between the predicted Cas9 blunt/staggered cleavage site and the position of a single‐nucleotide microhomology at the PAM‐distal end of DSB (Figs 6c, S13). Together with the fact that the majority of DELs occurred on PAM‐distal DNA strands (Figs 5c, S10, group 2) our data suggested that the nonrandom mutations were produced by a previously unidentified 5′ DNA single‐nucleotide microhomology‐mediated DNA repair. This highlighted the importance of considering the potential for triggering single‐nucleotide microhomology‐mediated DNA repair when designing gRNAs.

Fig. 6.

Microhomology‐mediated DNA repair between the protospacer adjacent motif (PAM)‐proximal (PAM strand) and distal (non‐PAM strand) DNA ends can explain increased frequency in deletion patterns. (a) Analysis of the size of microhomologies surrounding deletion reads. Red lines represent the probability of such microhomologies happening by chance, for each size class. As the largest detected microhomology was 5 nucleotides (nt) long, the probability for no microhomology (indicated on chart as 0) was calculated as subtracting the sum of the calculated probabilities of the different microhomology sizes from 100%. For normalisation, the sum of all deletion reads in individual samples was normalised to 100% and further divided into individual categories according to the microhomology size. The horizontal black line within the box, the ‘x’ and the diamond symbol represent the median, the mean and the outliers, respectively. (b) Normalised frequency of deletion reads surrounded by no (0 nt) or 1 nt microhomology in individual target sites and epigenotypes. The colour code indicates the occurrence of individual nucleotides. Sequences above each gRNA target represent the target sequence with highlighted and colour‐coded 3^rd and 4^th nucleotide upstream from PAM.  Individual biological replicates can be found in Supporting Information Fig. S12. (c) Enrichment of deletion sizes that match the distance between the nucleotide adjacent to the cut positions on the PAM strand and the corresponding nucleotide position upstream from the cut site on the non‐PAM‐distal DNA strand in ‘db’ gRNA target. The example above the chart represents the enrichment of 5 and 9 nt deletions, which correspond to the 3^rd and 4^th nucleotide microhomologies, respectively. Deletion patterns for the remaining target sites can be found in Fig. S13. Each data point represents the mean of next‐generation sequencing (NGS) data (OD₆₀₀ = 0.15) from three independent biological replicates. Terms TRV‐WT, TRV‐35S, noninfected and S1 are explained in Fig. 1; colour‐coded CRISPR/Cas9 target sites ut, ub, pt, pb, mt, mb, dt, db are explained in Fig. 2 and dtg in Fig. S4.

Discussion

The primary aim of this study was to investigate the effects of an epigenetic modification, namely DNA cytosine methylation, on the efficacy of generating targeted mutations using CRISPR/Cas9. Previous studies have suggested that the frequency of CRISPR/Cas9‐induced mutations may be affected by the epigenetic status of the target DNA, using genomewide analyses to look for associations between the frequency of mutations and the epigenetic status (euchromatin vs heterochromatin) of the target, for multiple endogenous loci (Chen et al., ²⁰¹⁶; Daer et al., ^{2017, 2018}). To our knowledge, no study has to date compared the effect of DNA methylation on CRISPR/Cas9‐induced mutations by comparing the mutation frequency at the same locus containing either a high or low cytosine methylation content. In this work, we developed an ‘epigenetic switch’ that enabled us to control the methylation status of the genomic loci, therefore allowing us to compare the efficacy of CRISPR/Cas9‐induced mutations for a given locus containing either a high or low level of cytosine methylation.

Our results showed that the level of cytosine methylation can affect the efficacy of CRISPR/Cas9 mutagenesis in a target site‐specific manner (Figs 2c, S2C). The RdDM pathway, which we used to trigger local chromatin modifications, is known to result in transcriptional silencing if a promoter sequence is targeted (Jones et al., ²⁰⁰¹). By contrast, cytosine methylation introduced to the transcribed region of a gene does not prevent transcription and in some cases has even been reported to positively correlate with high level transcription (To et al., ²⁰¹⁵). As we did not observe any change in the mutagenesis frequency in the transcribed GFP coding region, which was heavily methylated in TRV‐GFP‐infected plants (Fig. S4), we concluded that the induced DNA methylation affects CRISPR/Cas9 mutagenesis indirectly. In plants, cytosine methylation in the promoter region can attract SUVH proteins and therefore induce histone H3K9 dimethylation (Johnson et al., ²⁰⁰⁷), which is associated with repressive chromatin. In addition, methyl‐CpG binding proteins can attract histone deacetylases (Zemach & Grafi, 2003), which increased positive charge density on the histone tails, allowing the histones to wrap DNA more tightly. Methylated DNA can also prevent the incorporation of the H2A histone variant H2A.Z, which has been shown to affect the transcription of nearby genes (Coleman‐Derr & Zilberman, 2012). These and other associated chromatin changes could make DNA less accessible for the CRISPR/Cas9 machinery in the targeted P35S promoter. Indeed, in mouse cells, Heterochromatin protein 1 (HP1)‐associated stable heterochromatin has been shown to reduce the ability of the CRISPR/Cas9 complex to scan for potential target sites (Knight et al., ²⁰¹⁵), and a very recent study using proliferating human cells also demonstrated that chromatin state influences CRISPR/Cas9 mutagenesis (Schep et al., ²⁰²¹). As DNA methylation can affect nucleosome positioning in the promoter region (Kelly et al., ²⁰¹²; Zhong et al., ²⁰²¹), we could also envision that changes in the nucleosome occupancy of CRISPR/Cas9‐targeted loci would influence the efficacy of genome editing. Indeed, it has been shown (both in vitro and in yeast cells) that recognition of PAM sequences by CRISPR/Cas9 complexes is impaired when the PAM is situated at the nucleosome core (Hinz et al., ²⁰¹⁵; Yarrington et al., ²⁰¹⁸).

In contrast with our observation for the P35S promoter region, ongoing transcription of the GFP gene can be associated with repeated chromatin remodelling (Farnung et al., ²⁰²¹) therefore preventing methylation‐induced changes in the chromatin structure. This would be in line with previous in vitro studies in human cell cultures, for which target site methylation had no impact on CRISPR/Cas9 activity (Hsu et al., ²⁰¹³; Fujita et al., ²⁰¹⁶). Intriguingly, the higher mutagenesis rate that we found within the GFP coding region in TRV‐35S‐infected plants compared with TRV‐GFP‐infected plants (Fig. S4H) suggested that active transcription may negatively interfere with CRISPR/Cas9 binding and/or cleavage, which was in agreement with previous findings (Daer et al., ²⁰¹⁷). Alternatively, DSBs might be more accurately repaired (resulting in lower rates of mutagenesis) by the RNA‐templated DSB repair mechanism, in which RNA polymerase II transcripts can serve as the DNA repair template (Keskin et al., ²⁰¹⁴). Further experiments are required to determine which of the above mechanisms may underpin the effect of DNA methylation on CRISPR/Cas9 activity in plants.

In addition to observing that induced cytosine methylation could affect the rate of CRISPR/Cas9 mutagenesis, we noted that several other factors within our experimental system also had a significant effect (Fig. 7a). Virus infection generally increased the frequency of mutagenesis (Figs 2c, S2C, left vs right panels). We speculate that this effect was likely to be due to the stress caused by the infection, which might have caused a higher error rate in the cells’ DNA repair mechanisms. Indeed, a previous report has linked heat stress to higher rates of CRISPR/Cas9 mutagenesis (LeBlanc et al., ²⁰¹⁸). Additionally, we observed that the density of Agrobacterium used for infiltrating plants with the CRISPR/Cas9 components had a significant effect on the rate of mutagenesis. The higher bacterial density of OD₆₀₀ = 1.5 resulted in consistently higher rates of mutagenesis compared with the lower bacterial density of OD₆₀₀ = 0.15 (Fig. S2B). From this, we concluded that the level of expression of CRISPR/Cas9 components or the number of cells successfully transformed within the infiltrated patch was likely to be a limiting factor for the mutagenesis frequency, over this range of OD₆₀₀ values. Testing a wider range of Agrobacterium densities would be required to determine the optimum density of Agrobacterium suspension to achieve maximal mutagenesis frequencies. Importantly, our results highlight the need to carefully determine the most appropriate Agrobacterium density to use for studies aimed at understanding specific mechanisms of CRISPR/Cas9 editing. In our case, we chose to use a lower bacterial density (OD₆₀₀ = 0.15) for the subsequent analyses, because the effect of cytosine methylation on CRISPR/Cas9‐induced mutagenesis frequency was partially masked when a higher bacterial density was used (OD₆₀₀ = 1.5; Fig. S2B). This might be due to chromatin/histone changes that are related to quickly repeating cycles of Cas9 cleavage.

Fig. 7.

Overview of CRISPR/Cas9 mutagenesis outcomes. (a) A table summarising the main conclusions of our study. Checkmarks and crosses indicate statistically significant and nonsignificant effects, respectively. Mixed marks designate statistical significance for some, but not all target sites. (b) Proposed model for how DNA tension may influence the outcome of Cas9‐induced double‐stranded breaks (DSBs) to produce either a blunt or a staggered DSB and 5′ single‐nucleotide mediated DNA repair. On relaxed DNA, Cas9 forms a blunt DSB three nucleotides upstream from protospacer adjacent motif (PAM) due to the RuvC and HNH domains of Cas9 cleaving between the 3^rd and 4^th nucleotides upstream from the PAM position on the PAM strand and non‐PAM strand, respectively (left panel). DNA under tension could stretch the PAM strand, causing the RuvC domain of Cas9 to cleave between the 4^th and 5^th nucleotides upstream of the PAM. This, together with HNH‐mediated cleavage between the 3^rd and 4^th nucleotides upstream of the PAM position on the non‐PAM strand would result in a staggered DSB with a 1‐nt 5′ overhang on the PAM strand (right panel). After cleavage by Cas9, the PAM‐distal DNA strand is probably released first because most of the deletions happened on the PAM‐distal DNA (see Fig. 5). Moreover, the nucleotide adjacent to the cut on the PAM‐containing proximal strand serves as a single‐nucleotide searching for complementarity to allow specific 5′ microhomology‐mediated DNA repair (bottom panels).

As well as observing that extrinsic factors (virus infection and bacterial density) significantly and consistently affected the rates of mutagenesis for different target loci, we noted a strong intrinsic effect of the sequence itself for the different target loci (Fig. 7a). In other words, some target loci appear to be more prone to mutagenesis than others. Previous studies in human and yeast cells have come to similar conclusions that the efficacy of CRISPR/Cas9 mutagenesis is dependent on the sequence of the target locus (van Overbeek et al., ²⁰¹⁶; Lemos et al., ²⁰¹⁸). It is worth bearing in mind that the introduction of a mutation at a target locus relies not only on Cas9 recognition and subsequent DSB formation but also on the erroneous repair of that DSB. The sequence of a target locus could potentially affect all three of these steps (recognition, cleavage, and erroneous repair) that underlie mutagenesis. Parsing these effects remains a significant challenge to understanding how the underlying sequence of a target locus can predetermine the efficacy of CRISPR/Cas9 mutagenesis. Interestingly, even when we targeted the very same locus but on complementary DNA strands, we observed differences in the rates of mutagenesis. This effect was most visible for the mt/mb positions in all epigenotypes. The mutation frequency induced by the mt gRNA in TRV‐WT‐infected plants was approximately four times higher than that of the mb gRNA, which targeted the same locus on the complementary strand (Figs 2c, S2C).

According to the original and widely accepted model, Cas9 nuclease produces blunt‐ended DSBs (Jinek et al., ²⁰¹²; Ahmad et al., ²⁰²⁰). More recent studies on animal and yeast models showed that, in vivo, the Cas9 endonuclease can also form staggered DSBs with single‐nucleotide overhangs (Li et al., ²⁰¹⁵; Lemos et al., ²⁰¹⁸). These results were also supported by molecular dynamics simulations (Zuo & Liu, 2016). Our analyses of insertion and deletion events clearly showed that a high proportion of these mutations can be most simply explained as repairs of staggered DSBs with single‐nucleotide 5′ overhangs that are first filled in and then ligated presumably by c‐NHEJ (Beying et al., ²⁰²¹), which is the most prevalent DNA repair pathway in differentiated nonproliferating cells such as the leaf tissue in our study. We reviewed data from previous studies describing CRISPR/Cas9 mutations in several other plant species (including Arabidopsis, maize and cotton) and found that staggered‐end formation can also explain their most frequently inserted nucleotides (Li et al., ^{2013, 2019}; Nekrasov et al., ²⁰¹³; Lee et al., ²⁰¹⁹). These observations suggested that Cas9 can form both blunt and staggered ends in planta. Interestingly, our data also revealed that the frequency of blunt and staggered‐end cuts is predetermined by the CRISPR/Cas9 target site (Figs 4, 5, S9, S11), and also partially influenced by the epigenotype (e.g. pt, dt and db). To our knowledge, this is the first description of such an effect in any model system. To explain this phenomenon, we speculate that the process of Cas9 unwinding the DNA duplex at a target locus is sensitive to the physical tension of DNA in the locus, such that relaxed DNA duplexes are more easily unwound than DNA under tension. The tensioned conformation of DNA could therefore affect the position of the RuvC nuclease domain of Cas9 relative to the nontargeted strand, causing a shift of the cut by one nucleotide further from the PAM and resulting in a staggered DSB (Fig. 7b). Therefore, the relaxed DNA might favour the formation of blunt DSBs, whereas DNA under tension may favour the formation of a staggered DSB. In agreement with our model, Cas9 in vitro assays with DNA in a relaxed conformation produced blunt cuts (Sternberg et al., ²⁰¹⁴), whereas when we analysed data from in vivo studies, for which DNA is more likely to be under tension at certain loci, we inferred that Cas9 frequently produces staggered DSBs (Li et al., ^{2013, 2019}; Nekrasov et al., ²⁰¹³; Lee et al., ²⁰¹⁹). DNA tension could be affected by the local chromatin environment, mainly by DNA‐interacting proteins such as histones, transcription factors and polymerases. Their binding to DNA can be affected (or even directed) by cytosine methylation, which could result in a highly localised and strand‐specific increase or decrease in DNA tension.

In our system, we also observed that the epigenetic status of gRNA target sites could influence the frequency of CRISPR/Cas9‐induced deletions (Fig. 3b). Interestingly, the vast majority of deletion events happened on the PAM‐distal DNA end (Figs 5c, S10, group 2). We also found similar deletion patterns in other plant studies (Li et al., ^{2013, 2019}; Nekrasov et al., ²⁰¹³; Kumar et al., ²⁰¹⁸; Lee et al., ²⁰¹⁹; Fellenberg et al., ²⁰²⁰), but not in animal model systems (van Overbeek et al., ²⁰¹⁶). Based on these findings, we assumed that in plants the PAM‐distal cleavage product is released first from the CRISPR/Cas9 complex, and not just the nontargeted PAM‐distal DNA strand, as suggested by an animal in vitro study (Richardson et al., ²⁰¹⁶). In such a release mechanism, the PAM‐distal DNA end is more exposed to exonucleases, leading to deletions occurring predominantly at this side of the cut, whereas the PAM‐proximal DNA end remains protected and mostly unchanged (Fig. 7b). Moreover, our analysis suggests that a significant proportion of such deletions may be directed by a 5′ single nucleotide at the PAM‐proximal end of the Cas9‐induced DSB (Figs 7b, S13), which we also observed in other plant data sets (data not shown). Intriguingly, the well characterised microhomology‐dependent DNA repair pathway (MMEJ/alt‐EJ) is associated with 3′ DNA overhangs and proliferating cells (Sfeir & Symington, 2015). Therefore, it is plausible that the 5′ single‐nucleotide microhomology‐mediated DNA repair that we described in this study may be governed by NHEJ or novel genetic factors. However, more data including genetic evidence is required to identify the underpinning molecular mechanism and consequently assign the corresponding DNA repair pathway.

Taken together, our findings (Fig. 7a) have demonstrated that DNA methylation can influence the mutation frequency and mutational outcomes in the context of CRISPR/Cas9 genome editing, probably indirectly through changes in chromatin state. Our findings including the Cas9 staggered‐end‐mediated insertions and deletions and 5′ DNA single‐nucleotide mediated DNA repair (Fig. 7b) may contribute to better target selection and gRNA designs for diverse future applications in plants and beyond.

Competing interests

None declared.

Author contributions

AM performed conceptualisation; AM and AP performed methodology; AP performed software; AP, AM, LF and AB investigated the research; AP, LF and AM wrote original draft; DEP and AB reviewed and edited; AP visualised; and AM supervised the study.

Supporting information

Fig. S1 GFP fluorescence phenotypes for individual epigenotypes before agroinfiltration.

Fig. S2 Analysis of mutagenesis frequency at CRISPR/Cas9 cut sites.

Fig. S3 Comparison of the overall frequency of mutations induced by CRISPR/Cas9 in P35S.

Fig. S4 The impact of gene‐body methylation on CRISPR/Cas9‐induced mutations.

Fig. S5 Analysis of DNA (cytosine) methylation at CRISPR/Cas9 gRNA target sites in individual epigenotypes.

Fig. S6 Correlation analysis between the mutation frequencies and the percentage of methylated cytosines in the target locus in the compared epigenotypes.

Fig. S7 Comparison of insertion (IN) and deletion (DEL) events in individual CRISPR/Cas9 target sites.

Fig. S8 Insertion events analysis in P35S and GFP coding region.

Fig. S9 Detailed analysis of single‑nucleotide insertion events detected next to the Cas9 cut site in individual epigenotypes.

Fig. S10 Detailed analysis of the type and frequency of deletion mutations in individual epigenotypes.

Fig. S11 Detailed analysis of group 2 and its first adjacent position next to the cut site on the PAM‐distal DNA end in individual epigenotypes.

Fig. S12 Detailed analysis of the normalised frequency of deletion reads surrounded by no (0 nt) or 1 nucleotide microhomology in individual target sites and epigenotypes.

Fig. S13 The pattern of the deletion sizes in individual target sites.

Notes S1 DNA cytosine methylation within coding sequences does not alter the frequency of CRISPR/Cas9‐induced mutations.

Table S1 List of primers used in this study.

Table S2 List of vectors used in this study.

Please note: Wiley Blackwell are not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

Data availability

The NGS data have been deposited in the ArrayExpress database at EMBL‐EBI (www.ebi.ac.uk/arrayexpress) under accession number E‐MTAB‐10256.