CRISPR/Cas9-mediated homology-directed repair by ssODNs in zebrafish induces complex mutational patterns resulting from genomic integration of repair-template fragments

Abstract

Targeted genome editing by CRISPR/Cas9 is extremely well fitted to generate gene disruptions, although precise sequence replacement by CRISPR/Cas9-mediated homology-directed repair (HDR) suffers from low efficiency, impeding its use for high-throughput knock-in disease modeling. In this study, we used next-generation sequencing (NGS) analysis to determine the efficiency and reliability of CRISPR/Cas9-mediated HDR using several types of single-stranded oligodeoxynucleotide (ssODN) repair templates for the introduction of disease-relevant point mutations in the zebrafish genome. Our results suggest that HDR rates are strongly determined by repair-template composition, with the most influential factor being homology-arm length. However, we found that repair using ssODNs does not only lead to precise sequence replacement but also induces integration of repair-template fragments at the Cas9 cut site. We observed that error-free repair occurs at a relatively constant rate of 1-4% when using different repair templates, which was sufficient for transmission of point mutations to the F1 generation. On the other hand, erroneous repair mainly accounts for the variability in repair rate between the different repair templates. To further improve error-free HDR rates, elucidating the mechanism behind this erroneous repair is essential. We show that the error-prone nature of ssODN-mediated repair, believed to act via synthesis-dependent strand annealing (SDSA), is most likely due to DNA synthesis errors. In conclusion, caution is warranted when using ssODNs for the generation of knock-in models or for therapeutic applications. We recommend the application of in-depth NGS analysis to examine both the efficiency and error-free nature of HDR events.This article has an associated First Person interview with the first author of the paper.

Keywords: CRISPR/Cas9; HDR; Homology-directed repair; Next-generation sequencing; Zebrafish.

Fig. 1.

Impact of repair-template homology-arm length, strand complementarity and symmetry on HDR efficiency. (A) Illustration of an sgRNA target site for the zebrafish smad6a gene, with protospacer sequence GGGTACAGGCGGCCCACAC. Following Cas9 recruitment and sgRNA binding, the DNA is cleaved 3 bp upstream of the PAM sequence (NGG, displayed in bold), which is visualized by the dashed red line and indicated as the ‘Cas9 cut site’. We designed ten repair templates, 60, 120 or 180 nucleotides (nt) in length, either corresponding to the sgRNA target strand (‘target’ – T) or to the complementary strand (‘non-target’ – NT). The repair templates were either symmetrically (S) or asymmetrically (A) positioned around the Cas9 cut site. The position of the different repair templates relative to the Cas9 cut site is depicted in the lower panel. The sequence of the 60 nt repair templates for smad6a is shown in the upper panel, and the nucleotide sequences of all other repair templates are listed in Table S8. Each repair template contains several synonymous nucleotide changes, depicted in red, relative to the reference sequence, including replacement of a guanine nucleotide in the PAM, whenever possible. (B) For each target site (smad6a, tprkb, pls3 or slc2a10) and each repair-template type (NT 60 S, NT 120 S, NT 180 S, NT 120 A left, NT 120 A right, T 60 S, T 120 S, T 180 S, T 120 A left, T 120 A right), average total HDR efficiencies resulting from five independent experiments were plotted. Calculated repair rates represent the number of sequencing reads containing the intended base-pair substitution closest to the Cas9 cut site. The HDR rates were split up into two categories: ‘perfect repair %’, representing the percentage of NGS reads containing at least the base-pair change closest to the Cas9 cut site (plain bars), and ‘erroneous repair %’, representing the reads containing erroneous integration events of repair-template fragments (dashed bars). NT, non-target; T, target; 60-120-180, total repair-template length; S, symmetrical; A, asymmetrical. Error bars represent the s.e.m. for five independent biological replicates each consisting of a pooled sample of 20 embryos. Repair rates depicted in this graph are listed in Table S1. Statistical tests performed: one-way ANOVA with blocking (60 nt vs 120 nt, 60 nt vs 180 nt, 120 nt vs 180 nt, target vs non-target) for symmetrical templates, non-parametric Kruskal–Wallis test, followed by pairwise comparison with Dunn–Bonferroni correction, for asymmetrical templates; *P<0.05 and **P<0.01.

Fig. 2.

Simplified schematic representation of erroneous repair-template integration events at the smad6a zebrafish gene. For both the ‘NT 120 S’ and ‘T 120 S’ repair templates, three examples of NGS reads are schematized to clarify the erroneous repair patterns that were encountered in this study. For each example, the black line with arrows represents the reference sequence, and the two black dashed lines indicate the approximate location of repair-template insertions. Inserted fragments are depicted as rectangles, colored in blue or yellow, in accordance with the used repair template (indicated in the scheme as ‘non-target’ or ‘target’ repair template on top of the three example sequences). Gray rectangles depict ‘random’ sequences not corresponding to the repair-template sequence. The color gradient clarifies to which part of the repair template the insertional fragment corresponds and defines the orientation of the fragment. This is additionally illustrated by arrows overlaying the rectangles. A ‘c’ in the rectangle means that it corresponds with the repair template's complementary sequence. In ‘NT 120 S’ example sequence 3, the integrated repair template fragments could not be assigned to a certain location in the reference sequence, due to the limited length of NGS sequence reads (250 bp), which is depicted by two question marks ('?'). This scheme is a simplified representation. A more detailed scheme of these examples, using the NGS read sequence as reference point, is depicted in Fig. S2.

Fig. 3.

Impact of the position of the substituted nucleotide in the repair template on repair rates. (A) Repair-template composition with specified locations of the base pair alterations, shown for an ‘NT 60 S’ repair template. Each repair template included in this study contains several synonymous nucleotide changes relative to the reference sequence. Whenever possible, a guanine nucleotide of the PAM was replaced, in order to avoid undesired cleavage of the repair template by Cas9. In addition, five nucleotide changes were included in the region surrounding the Cas9 cut site (designated as point zero, visualized with a gray dashed line). All nucleotide changes are depicted in red. The position of the PAM sequence is marked with a full gray line. (B) The HDR efficiency (perfect repair only) for each base-pair substitution in each repair template is plotted as a function of the relative distance to the Cas9 cut site. Each plotted data point shows average values of five independent experiments, each analyzing 20 embryos. To improve clarity, error bars were omitted. NT, non-target; T, target; 60-120-180, total repair-template length; S, symmetrical; A, asymmetrical. Repair rates depicted in this graph are listed in Table S3.

Fig. 4.

Influence of chemical compound administration through injection on HDR efficiency. Injection mixes containing the NT 120 S repair template were complemented with chemical compounds that either inhibit specific components of the NHEJ pathway, including SCR7, NU7441 and KU0060648, or that were shown to stimulate the HDR pathway, including RS1 and L755507. Five independent experiments were carried out and average total HDR rates are shown, split into two categories: perfect repair % (plain bars) and erroneous repair % (dashed bars). Error bars represent the s.e.m. for five independent biological replicates each consisting of a pooled sample of 20 embryos. Repair rates depicted in this graph are listed in Table S4. Statistical tests performed: independent samples t-test for normal distributed groups and the non-parametric independent samples Mann–Whitney U-test for non-normal distributed groups.

Fig. 5.

Determination of germline transmission of precise base-pair substitutions introduced by CRISPR/Cas9-mediated HDR. For two of the four target sites included in this study (the slc2a10 and smad6a zebrafish genes), embryos injected with injection mixes containing the NT 120 S repair template were grown until adulthood. For each target site, eight adult fish (labeled founder fish # 1, # 2, … # 8) were screened for the presence of the precise base-pair substitutions located closest to the Cas9 cut site in their germ cells. Therefore, DNA was extracted from collected eggs or sperm, and subjected to NGS analysis. HDR rates are shown, split into two categories: perfect repair % (plain bars) and erroneous repair % (dashed bars). Repair rates depicted in this graph are listed in Table S6.

Fig. 6.

Mechanisms of DNA double-strand break repair. (A) A schematic representation of different described mechanisms of DSB-induced HDR is shown, categorized by repair template type: double-stranded (gray) or single-stranded (‘target’ – yellow, ‘non-target’ – blue). DSB repair by HDR is always initiated by DNA end resection, resulting in 3′ single-stranded DNA tails, which participate in strand invasion into homologous sequences. DSB repair with double-stranded repair templates can be carried out by three different mechanisms: double-strand break repair (DSBR or Holliday junction resolution), Holliday junction dissolution, or synthesis-dependent strand annealing (SDSA) (Allers and Lichten, 2001; Szostak et al., 1983). DSB repair using single-stranded templates, sometimes referred to as SSTR, acts via SDSA (Davis and Maizels, 2016; Kan et al., 2017; Paix et al., 2017). (B) Homology-arm length influences the efficiency of SDSA. The left panel depicts the composition of all included repair templates in this study, where the four templates that perform significantly worse in comparison to the others in terms of total repair rates (see Fig. 1B) are highlighted (1-4). The location of the DSB is marked with a red dashed line. These four templates all contain a short (30 nt) homology arm at the side of the repair template that does not anneal to one of 3′ overhangs (marked with ‘II’ sign), but participates in the resolution step, as also depicted in the right panel, where each of these inefficient templates is fitted to the SDSA model for DSB repair with single-stranded templates. The red dots refer to the DSB location, with left and right homology arms marked with red arrows and length indications. (C) Mechanisms of template switching in SDSA leading to erroneous repair. Following SDSA initiation, dissociation of unstable intermediates can lead to template switching. Four possible mechanisms for template switching during SDSA are shown. These mechanisms are based on microhomology (binding marked in blue) and can result in erroneous replication (red dashed arrows).