Uncovering the dynamics of precise repair at CRISPR/Cas9-induced double-strand breaks

Abstract

CRISPR/Cas9 is widely used for precise mutagenesis through targeted DNA double-strand breaks (DSBs) induction followed by error-prone repair. A better understanding of this process requires measuring the rates of cutting, error-prone, and precise repair, which have remained elusive so far. Here, we present a molecular and computational toolkit for multiplexed quantification of DSB intermediates and repair products by single-molecule sequencing. Using this approach, we characterize the dynamics of DSB induction, processing and repair at endogenous loci along a 72 h time-course in tomato protoplasts. Combining this data with kinetic modeling reveals that indel accumulation is determined by the combined effect of the rates of DSB induction processing of broken ends, and precise versus error repair. In this study, 64-88% of the molecules were cleaved in the three targets analyzed, while indels ranged between 15-41%. Precise repair accounts for most of the gap between cleavage and error repair, representing up to 70% of all repair events. Altogether, this system exposes flux in the DSB repair process, decoupling induction and repair dynamics, and suggesting an essential role of high-fidelity repair in limiting the efficiency of CRISPR-mediated mutagenesis.

Fig. 1. UMI-DSBseq quantitative single-molecule sequencing of DSBs and repair products at three targets in tomato.

A Collection of time-course: mesophyll cell protoplasts are isolated from 2 to 3 week-old seedlings of M82 Solanum lycopersicum. Duplicate samples are prepared with 200,000 protoplasts for each of the 7 time-points along 72 h. CRISPR RNPs are introduced by PEG-mediated transformation. Samples are frozen at 0, 6, 12, 24, 36, 48, and 72 h after RNP introduction and DNA is extracted. B UMI-DSBseq target design: a primer specific to the target sequence (blue arrow), is coupled with a restriction enzyme site flanking the sgRNA target sequence to create an available end on intact molecules (WT or Indel) for ligation of adaptors. C UMI-DSBseq library preparation: DNA extraction from time-course collection, containing WT (1), unrepaired DSBs (2), and intact molecules containing indels (3), is restricted in vitro with the restriction enzyme identified flanking the target cut site (dashed oval). Following end-repair by fill-in and A-addition, Y-shaped adaptors composed of P7 Illumina flow-cell sequences and containing i7 indexes (orange) and 9 bp unique molecular identifiers (UMIs) in purple, are ligated to the unrepaired DSBs and restricted ends. Target-specific amplification by ligation-mediated PCR follows, with one primer identical to the adaptor sequence and containing the P7 Illumina tail (orange arrow) and one primer specific to the target sequence (blue arrow) with the P5 Illumina tail (red). This results in the amplification of a single end of the DSB between the SpCas9 cut site and the primer. The red X represents the non-captured end of the DSB.

Fig. 2. Patterns of error-prone repair along 72 h time-course.

A–F Percent of molecules identified as unrepaired DSBs (red) and NHEJ-mediated indel (blue) out of total consensus sequences along 72 h, A–C experimental and D–F control time-courses for (A, D) Psy1, (B, E) CRTISO, and (C, F) PhyB2 with dots representing percent of molecules from each replicate and a line representing the mean of the duplicates (see related: Fig. S2).

Fig. 3. Positions of DSB.

A–C Proportion of total DSBs (normalized to total molecules obtained in the in vivo time course) plotted by position along the target sequence with the expected cut site between -3 and −4 bp upstream of the PAM site, (A) Psy1, (B) CRTISO and (C) PhyB2. Bars indicating the proportion of DSBs captured at different positions over all time points of the time courses (n = 14) are shown with different colours based on the position, as shown in the legend. The positions of DSB for individual time points of the time courses are shown in Fig.S3. D-E Proportion of DSBs induced in vitro (normalized to total DSBs) plotted by the position of the captured end at (D) Psy1, (E) CRTISO, and (F) PhyB2; (n = 4).

Fig. 4. A 3-state model of DSB induction and repair.

A schematic and equations for a 3-state model of DSB induction K_cut in red, precise repair, P_repair in green and Error-prone repair E_repair, in blue. B–D Predicted fit of the model (lines) at Psy1 (B), CRTISO (C), and PhyB2 (D) for Intact molecules (green), DSB (red), and indels (blue). Confidence intervals are shown as shadings and calculated from 100 bootstraps of the data. Observed data are represented as dots. E–G Rate constants estimated at Psy1 (E), CRTISO (F), and PhyB2 (G) in terms of number of events per hour per molecule. The smoothed distribution of the estimates obtained through the bootstrap procedure is shown as a violin plot. Mean estimate is shown as a black point. Grey points represent 100 instances of stratified bootstrap. (see related Table 1, Fig. S2, Table S1).

Fig. 5. Characterizing putative intermediates of DSB repair.

A Schematic of categories of DSB types: Direct DSB (−3 bp from the PAM site), Processed DSBs (putative repair intermediates) classified as Guide, PAM, and Extended DSBs. B–D Percent of DSB types along the time-course of repair, for Psy1 (B), CRTISO (with perfect DSBs including both -3 and −4 bp) (C) and PhyB2 (D). Direct DSBs are shown in blue, guide-side DSBs in yellow, PAM-side DSBs in green and extended DSBs with extensions that do not match the target sequence in red.

Fig. 6. A 4-state model of DSB induction and repair, including processed ends.

A schematic and equations for a 4-state model of DSB induction, K_cut, ends processing, K_processing, repair from directly induced DSBs, namely precise repair, P_direct and error repair E_direct, and from putative repair intermediates ‘processed’ DSB, including precise repair P_processed and error repair E _processed. B–D Violin plots of rate constants estimated at Psy1 (B), CRTISO (C), and PhyB2 (D) in terms of proportion per hour (as in Fig.4E–G). E–G Schematic representation of the dynamic flow estimated in terms of percent of total molecules following 72 h, at CRTISO when (E) the −4 position DSB is considered as a directly induced (DSB) or (F) as a ‘processed’ DSB, (G) Psy1, and (H) PhyB2. Grey arrows indicate rates estimated as 0, dashed arrows represent confidence intervals overlapping 0. CIs indicated in grey brackets and calculated from 100 iterations of the bootstrap. See also Table S4, Fig. S12.

Fig. 7. High resolution time courses over 24 h.

A–C Fit of the 3-state model to the data and respective estimates of the rate constants (D, E) for Psy1 (A, D), CRTISO (B, E), PhyB2 (C, F). Mean trajectory and rate estimates are shown as continuous lines (A–C) and black dots (D–F), while the distribution of 100 bootstraps is shown in lighter colours.

Fig. 8. UMI-DSBseq coupled with likelihood-based modeling reveals variables in the processes of DSB induction and repair.

A Decoupling of induction and repair revealing dynamics characterized by relatively efficient cutting and prominent precise repair. B Model of DSB repair in the absence of precise repair was rejected by AIC relative likelihood. Models of DSB repair in the presence of precise repair can explain the rate of indels formation at various loci (3-state is shown here for simplicity, but the same conclusion applies to the 4-state model). The thickness of the arrows indicates the efficiency, i.e., thicker arrows have a higher proportion of molecules.