Repair of Cas9-induced DSBs involves c-NHEJ to varying degrees at different targets
If persistent binding and residence of Cas9-sgRNA at its target after DNA cleavage influence DSB repair pathway choice, the involvement of c-NHEJ in repair of Cas9-induced DSBs would change between targets with different binding affinities and residence duration for Cas9-sgRNA. Inactivation of c-NHEJ would thus lead to varying degrees of mutagenic NHEJ (m-NHEJ) stimulation at these sites. To test this hypothesis, we used Cas9-sgRNA to induce site-specific DSBs in an NHEJ reporter integrated in the genome of mouse embryonic stem cells (mESC) as done before19, and analyzed the extent of the c-NHEJ involvement in repair of Cas9-induced DSBs at different sites (Fig. 1a). In this NHEJ reporter, only m-NHEJ can lead to GFP+ cells, and the frequency of GFP+ cells thus represents the relative efficiency of m-NHEJ as part of both c-NHEJ and a-NHEJ (Fig. 1a). We found that neither DNA-PKcs inhibition by NU7441 nor XRCC4 deletion changed the frequencies of Cas9-induced m-NHEJ at the two sites targeted by the sgRNA gEJW3 or gEJW7, while the inactivation of c-NHEJ inhibited Cas9-induced m-NHEJ at the four sites targeted by gEJC5, gEJW4, gEJW5 and gEJW6 to different extents (Fig. 1b). This indicates that the participation of c-NHEJ varies in repair of Cas9-induced DSBs at different targets.
Surprisingly, no stimulation of m-NHEJ was observed in repair of Cas9-induced DSBs by inactivation of c-NHEJ at some targets. To determine whether this was unique to repair of Cas9-induced DSBs, we used the same NHEJ reporter cells but with the first I-SceI recognition site being destroyed by CRISPR/Cas9 genome editing to imitate Cas9-induced single DNA breakage in the reporter so that GFP+ cells can only be generated by m-NHEJ in repair of I-SceI-induced DSBs (Fig. S1a). Inhibition of c-NHEJ with NU7441 increases I-SceI-induced m-NHEJ by more than 2-fold (Fig. S1b,c). We then wondered why inactivation of c-NHEJ stimulates I-SceI-induced m-NHEJ, but not Cas9-induced m-NHEJ. Given that the repair of Cas9-induced DSBs is mostly accurate5,6, it is possible that Cas9-sgRNA in the cells is abundant and could recleave target DNA until the targeted sites are mutated by NHEJ, thus accumulating m-NHEJ events (Fig. 1c). Inactivation of c-NHEJ would then reduce m-NHEJ events (Fig. 1b). By reducing the amount of Cas9 or sgRNA transfected, we predicted that the target recleavage would be reduced and Cas9-induced m-NHEJ would be stimulated by DNA-PKcs inhibition (Fig. 1c). We found that overall Cas9-induced m-NHEJ was reduced with a low amount of Cas9-gEJW6 as expected (Fig. S1d). While NU7441 suppressed m-NHEJ induced by a high amount of Cas9 and gEJW6 typically at 0.25µg each, the inhibitor started to stimulate m-NHEJ when the amount of Cas9 and gEJW6 was both reduced to 0.001µg (Fig. S1d,e). In contrast, DNA-PKcs inhibition did not alter m-NHEJ induced by Cas9 and gEJW7 at an amount ranging from 0.25µg to 0.0001µg, indicating no involvement of c-NHEJ in repair of Cas9-induced DSB at the gEJW7 target (Fig. S1f,g). At the two sites targeted by gEJC5 or gEJW5 with the transfection amount of Cas9-sgRNA at 0.001µg, DNA-PKcs inhibition and XRCC4 deletion still inhibited Cas9-induced m-NHEJ but this inhibition was reduced (Fig. 1d). However, this c-NHEJ inactivation abolished or even reversed the suppression of Cas9-induced m-NHEJ to stimulation at the targets with gEJW4 and gEJW6 and remained to exert no effect on Cas9-induced m-NHEJ at the two sites with gEJW3 or gEJW7 (Fig. 1d). This further confirms that the involvement of c-NHEJ varies at different targets in repair of Cas9-induced DSBs.
We also used a single-copy GFP reporter gene and two natural genome loci to further assess variable involvement of c-NHEJ in repair of Cas9-induced DSBs (Fig. S2a). Among 7 sgRNAs designed for GFP gene knock-out (KO), all converted over 60% GFP+ cells into GFP− cells (Fig. S2b). DNA-PKcs inhibition by NU7441 had no significant effect on GFP KO at all 7 sites (Fig. S2b). To limit recleavage by Cas9, we transfected mESC with much lower amount of Cas9-sgRNA at 0.001µg. While limited recleavage lowered CRISPR/Cas9-mediated GFP KO editing, DNA-PKcs inhibition slightly increased GFP KO at 1 site targeted by gGC7, but made little change on GFP KO at the other 6 sites (Fig. S2c). Using targeted PCR amplicon deep sequencing, we additionally measured Cas9-induced m-NHEJ at two natural genome loci Cola1 and Rosa26 in mESC. We found that DNA-PKcs inhibition reduced the editing efficiency at the sites targeted by Cola1 gC2 and Rosa26 gR3, stimulated at the sites by Cola1 gC3, and had minimal effect at the rest of the sites including gC1 and gC4 for Cola1 and gR1, gR2 and gR4 for Rosa26 (Fig. S2d,e). These results together indicated variable involvement of c-NHEJ in CRISPR/Cas9 genome editing at different sites or even no involvement of c-NHEJ at some sites.
Inactivation of c-NHEJ induces varying stimulation of Cas9-induced HDR among targets
Inactivation of c-NHEJ also channels DSBs for repair from c-NHEJ towards HDR with a certain probability in each repair round, causing the increased use of HDR regardless of DNA recleavage by Cas9. Thus, using a single-copy HDR reporter in mESC20, we analyzed the impact of c-NHEJ inactivation on HDR induced by Cas9-sgRNA targeting at 5 different sites surrounding the I-SceI site in the reporter (Fig. 1e). Like I-SceI-induced HDR, Cas9-induced HDR was increased by NU7441 at the sites targeted by gHRC1, gHRC2 and gHRC3 and the extent of this stimulation was different among these three targets (Fig. 1f). Surprisingly, DNA-PKcs inhibition did not elevate HDR induced by Cas9-gHRC4 and Cas9-gHRC5 (Fig. 1f). We also used CRISPR/Cas9 gene editing to generate isogenic wild-type, DNA-PKcs−/− and Ku80−/− mESC clones containing the HDR reporter (Fig. S3a,b). Using one of these clones, along with isogenic XRCC4+/+ and XRCC4−/− HDR reporter mESC previously established21, we found that deletion of DNA-PKcs, Ku80 or XRCC4 significantly enhanced HDR induced by gHRC1, gHRC2 or gHRC3 in complex with SpCas9, as well as HDR induced by I-SceI (Figs. 1g and S3c). The extents of the HDR stimulation varied among these three different targets (Fig. 1g). However, deletion of DNA-PKcs or Ku80, like DNA-PKcs inhibition, caused little increase in HDR induced by Cas9-gHRC4 and Cas9-gHRC5 whereas XRCC4 deletion had modest stimulation (Fig. 1g). This suggests that inactivation of c-NHEJ stimulates Cas9-induced HDR to varying extents among targets due to different involvement of c-NHEJ at these targets. At the targets where Cas9-induced HDR is not stimulated by either inhibition of DNA-PKcs or deletion of DNA-PKcs or Ku80, it is possible that repair of Cas9-induced DSBs by NHEJ does not involve DNA-PKcs or Ku70/Ku80.
Reduced target-binding affinity of Cas9-sgRNA shifts repair of Cas9-induced DSBs towards c-NHEJ
To further determine whether c-NHEJ repair of Cas9-induced DSBs is influenced by target-binding affinity or post-cleavage target residency of Cas9-sgRNA, we mutated either sgRNA or SpCas9 for two sites targeted by gEJC5 and gEJW7 in the NHEJ reporter to reduce the target-binding affinity of the Cas9-sgRNA complex. Consistent with previous observation that weakened Cas9-sgRNA target binding generally reduces the efficiency of genome editing22–25, Cas9-induced m-NHEJ was less efficient with mismatched or truncated sgRNA variants (i.e. C2A, T15A and 16nt for gEJC5, and A1T, A4C and T15A for gEJW7) and with SpCas9 variants eSpCas9 and SpCas9-HF1, both of which were engineered to have lower binding affinity and higher specificity to target DNA (Fig. 2a). As in Fig. 1b, XRCC4 deletion and DNA-PKcs inhibition reduced Cas9-induced m-NHEJ at the site targeted with gEJC5, again suggesting significant DNA recleavage by Cas9 (Fig. 2a). In contrast, the gEJC5 variants (C2A, T15A and 16nt), eSpCas9 and SpCas9-HF1 alleviated or even reversed this reduction (Fig. 2a). This suggests that in repair of Cas9-induced DSBs, the weaker the Cas9-sgRNA target binding is, the less Cas9-induced DNA recleavage occurs and the more preferentially c-NHEJ is engaged. At the site targeted by gEJW7, neither DNA-PKcs inhibition nor XRCC4 deletion had effect on Cas9-induced m-NHEJ as shown in Fig. 1b (Fig. 2a), indicating no engagement of c-NHEJ at this site. However, gEJW7 mismatch variants (i.e. A1T, A4C and T15A), eSpCas9 and SpCas9-HF1 allowed significant stimulation of Cas9-induced m-NHEJ up to 3.5-fold by DNA-PKcs inhibition or XRCC4 deletion (Fig. 2a), suggesting that weakened target binding of Cas9-sgRNA would induce strong involvement of c-NHEJ.
Using the integrated single-copy GFP gene (Fig. S2a), we also analyzed whether weakened Cas9-sgRNA target binding promoted c-NHEJ, which is largely accurate in repair of Cas9-induced DSBs, and suppressed GFP KO. As expected, Cas9-mediated GFP KO at the gGW5 and gGC15 sites was less efficient with mismatched sgRNA mutants (i.e. C1T, T15A and C16G for gGW5, and A2C, A15T and G16C for gGC15) or SpCas9 variants eSpCas9 and xCas9 (Fig. S4a,b). Although DNA-PKcs inhibition had no effect on GFP KO mediated by wild-type SpCas9-sgRNA at either site (Fig. S2bc), it stimulated GFP KO mediated by these Cas9-sgRNA variants (Fig. S4a,b). Using endogenous genomic loci, we also found that the editing efficiency with the mismatch variants of Cola1 gC4 (i.e. C1T and G16C) and Rosa26 gR4 (i.e. A16T) was reduced due to weaker target binding, but elevated by DNA-PKcs inhibition (Fig. 2b,c), although DNA-PKcs inhibition had minimal effect on the editing efficiency at the sites targeted by Cola1 gC4 and Rosa26 gR4 (Fig. S2d,e). This indicates that reduced Cas9-sgRNA target binding promotes c-NHEJ. Taken together, these results suggest that the variable target-binding affinity of Cas9-sgRNA may determine the extent of c-NHEJ involvement in repair of Cas9-induced DSBs.
Reduced Cas9-sgRNA target-binding affinity enhances stimulatory effect of c-NHEJ inactivation on Cas9-induced HDR
Consistent with previous studies21,26, inactivation of c-NHEJ stimulates HDR induced by CRISPR nucleases as well as I-SceI (Figs. 1f,g and S3c). We expected that this stimulatory effect would be further enhanced if HDR were induced by Cas9-sgRNA variants with reduced target-binding affinity because reducing Cas9-sgRNA target binding promotes c-NHEJ. Therefore, we compared HDR induced by mutated Cas9-sgRNA between cells proficient and deficient in c-NHEJ. Due to reduced efficiency of DNA cutting, Cas9-induced HDR was generally less efficient with mismatched or truncated sgRNA variants (i.e. G1C, G2C and 17nt for gHRC4, and A1T, C2A and 17nt for gHRC2) and SpCas9 variants eSpCas9, SpCas9-HF1 and xCas9 (Fig. 2d,e). At the site targeted by gHRC4, where Cas9-induced HDR was not affected by either DNA-PKcs inhibition, DNA-PKcs deletion or Ku80 deletion, and modestly stimulated by deletion of XRCC4, Cas9-induced HDR with the sgRNA variants or the SpCas9 variants was significantly elevated by inactivating c-NHEJ (Fig. 2d). At the site targeted by gHRC2, where Cas9-induced HDR was increased by DNA-PKcs inhibition or deletion of DNA-PKcs, Ku80 or XRCC4, respectively, stimulation of Cas9-induced HDR by inactivation of c-NHEJ was further enhanced with the Cas9-sgRNA variants (Fig. 2e). This confirms that reducing target-binding affinity of Cas9-sgRNA promotes c-NHEJ, providing the basis for the enhanced stimulatory effect of c-NHEJ inactivation on Cas9-induced HDR.
c-NHEJ deficiency increases off-target activity of CRISPR/Cas9
As mismatches in base pairing between sgRNA and off-target sites reduce the binding affinity of Cas9-sgRNA to off-target sites and lower the occurrence of DNA recleavage at these sites, it is anticipated that c-NHEJ would be engaged proportionally more at off-target sites than at on-target sites. Thus, inactivation of c-NHEJ would also increase the use of m-NHEJ at off-target sites, exacerbating off-target effects in CRISPR/Cas9 genome editing. To test this hypothesis, we analyzed the effects of DNA-PKcs inhibition and XRCC4 deletion on off-target activities of Cas9 at 7 potential off-target sites for gPnpla3 and 6 potential off-target sites for gMertk. We found that both NU7441 and XRCC4 deletion slightly reduced on-target editing by Cas9-gPnpla3 and Cas9-gMertk, indicating significant on-target DNA recleavage. In contrast, off-target editing efficiencies were not reduced by either DNA-PKcs inhibition or XRCC4 deletion, but even increased at some of these off-target sites, e.g. OT5 and OT7 for gPnpla3 and OT1 for gMertk (Fig. 3a,b). This suggests that inactivation of c-NHEJ exerts more stimulatory effect on off-target editing than on on-target editing, aggravating off-target effect in CRISPR/Cas9 genome editing.
Chemical inhibition and genetic inactivation of c-NHEJ are often used to increase the efficiency of Cas9-induced HDR-mediated gene knock-in or replacement27–31. Given that NU7441 stimulated Cas9-induced HDR in the HDR reporter at the targets by gHRC1 and gHRC2 (Fig. 1f), we also performed off-target analysis through amplicon deep sequencing of 6 potential off-target sites for Cas9-gHRC1 and Cas9-gHRC2 respectively. The on-target editing efficiencies of Cas9-gHRC1 and Cas9-gHRC2 were slightly lowered by DNA-PKcs inhibition, again indicating significant on-target DNA recleavage. Unlike on-target editing, the editing efficiencies at the 6 off-target sites were not reduced by NU7441. Instead, relative off-target effects represented by the ratio of the off-target activity to the on-target activity were elevated (Fig. 3c,d). We also found that relative off-target effects were significantly enhanced at the off-target sites of gHRC1 and gHRC2 in XRCC4−/− cells (Fig. 3c,d). This again suggests that inactivation of c-NHEJ could exacerbate off-target effects in CRISPR/Cas9 genome editing.
Local transcription does not affect involvement of c-NHEJ in repair of Cas9-induced DSBs
After DNA cleavage by Cas9-sgRNA, some ends can be exposed by spontaneous dissociation of Cas9-sgRNA from the cleaved target DNA and engage c-NHEJ, and the others may require local transcription machinery or DNA replication fork to dislodge the target-bound Cas9-sgRNA complex. If the dislodgement of Cas9-sgRNA by transcription underlies the disengagement of c-NHEJ, we reasoned that the gene silencing activity (i.e. the transcription-blocking capability) of catalytically dead Cas9 (dCas9)-sgRNA at a target would be negatively correlated with the extent of c-NHEJ participation in repair of Cas9-induced DSBs at the same site. Thus, using the integrated single-copy GFP in mESC, we examined any potential correlation between the gene silencing activities of dSpCas9 at various sites and c-NHEJ involvement in SpCas9-induced GFP KO editing at the same sites. While dSpCas9-sgRNA exhibited variable gene silencing activities at many of these sites (Fig. 4a), the effect of DNA-PKcs inhibition on GFP KO varied from no effect to 4-fold stimulation among targets (Figs. 4b and S4c). No apparent bias towards either template strand of transcription or non-template strand was detected in both transcriptional silencing and DNA-PKcs involvement (Fig. 4c). Importantly, no correlation was observed between the gene silencing activities by dSpCas9-sgRNA and the DNA-PKcs participation in SpCas9-induced m-NHEJ (Fig. 4d; P = 0.78).
To further determine the effect of the collision between local transcription and Cas9-sgRNA on the engagement of c-NHEJ for Cas9-induced m-NHEJ, we used catalytically dead Staphylococcus aureus Cas9 (dSaCas9)-sgRNA to block the translocating RNA polymerase (RNAP), preventing its collision with downstream site-specific DSBs induced by SpCas9 (Fig. 4e). Among 6 sgRNAs tested for transcriptional blockage, only gSaGW1 and gSaGW2, in complex with dSaCas9, mediated efficient gene silencing, indicating a strong capability of blocking RNAP (Fig. 4e). The GFP KO frequency induced by SpCas9-gGC4 and SpCas9-gGW5 at a transfection amount of 0.25µg or 0.001µg was not altered by DNA-PKcs inhibition, and this non-effect was not changed by co-transfection with either dSaCas9-gSaGW1 or dSaCas9-gSaGW2 (Fig. 4f). This suggests that transcription blockage by dSaCas9-sgRNA would not affect the extent of c-NHEJ engagement in repair of SpCas9-induced DSBs and further confirms that a collision with local transcription do not control c-NHEJ engagement in repair of Cas9-induced DSBs.
Local replication abolishes c-NHEJ engagement at Cas9-induced DSBs
To address whether a replication fork could locally dissociate the Cas9-sgRNA complex from cleaved target DNA, disfavoring c-NHEJ, we transfected HEK293 cells with a plasmid containing an SV40 origin-ATG-GFP-P2A-FLuc NHEJ reporter cassette, together with expression plasmids for SV40 large T antigen (LT), I-SceI or the SpCas9-gEJW10 complex, and the Renilla luciferase (RLuc) gene as internal control. The expression of SV40 LT drives bidirectional DNA replication via the SV40 origin, and the expression of I-SceI or SpCas9-gEJW10 induces a site-specific DSB between the “Koz-ATG” and the “ATG-GFP-P2A-FLuc” (Fig. 5a). Repair of I-SceI- or Cas9-induced DSBs by m-NHEJ can proportionally reframe the originally out-of-frame firefly luciferase (FLuc) gene in the NHEJ reporter plasmids to in-frame in the cells and induce synthesis of active firefly luciferase. The frequency of this NHEJ repair can thus be measured as a relative ratio of FLuc to RLuc by luminescence assays. Treatment with NU7441 reduced I-SceI-induced NHEJ in this assay, but the level of this reduction was similar with or without the expression of SV40 LT (Fig. 5a), suggesting little effect of SV40 LT on I-SceI-induced NHEJ. However, while Cas9-induced NHEJ was also suppressed by NU7441, DNA replication initiated by SV40 LT significantly attenuated this repressive effect (Fig. 5a). This suggests that local DNA replication driven by SV40 LT might inhibit the involvement of c-NHEJ in repair of Cas9-induced DSBs.
We also wondered whether a collision with local DNA replication would prevent potential competition between c-NHEJ and HDR in repair of Cas9-induced DSBs by blocking c-NHEJ engagement, thus removing the stimulatory effect of DNA-PKcs inhibition on Cas9-induced HDR. Using U2OS cells containing an integrated single-copy HDR reporter (Fig. 5b), in which an SV40 origin is located between TrGFP and I-SceI-GFP, we analyzed the effect of DNA-PKcs inhibition on HDR induced by I-SceI and SpCas9. DNA-PKcs inhibition stimulated HDR induced by SpCas9 in complex with gHRC1, gHRC2, gHRC3, gHRC4 and gHRC5 to different degrees, as well as by I-SceI (Figs. 5b and S5a), indicating variable but detectable engagement of the competing c-NHEJ pathway in repair of these I-SceI- or Cas9-induced DSBs. After expression of SV40 LT, both I-SceI- and Cas9-induced HDR were repressed in a dose-dependent manner (Figs. 5b and S5b,c). DNA-PKcs inhibition stimulated I-SceI- or Cas9-induced HDR, and the expression of SV40 LT attenuated this stimulation or even abolished it at a transfection amount of 0.16 µg (1/5 of total DNA transfected) (Figs. 5b and S5b,c). This suggests that local DNA replication driven by SV40 LT could collide with both I-SceI and SpCas9-sgRNA after DNA cleavage to dislodge I-SceI and Cas9-sgRNA from its cleaved target and restrict the engagement of c-NHEJ in repair of exposed DSBs.
By restricting c-NHEJ due to a collision with replication fork, the DSB repair pathway choice would be biased toward HDR. To test this possibility, we used the HDR reporter to measure the bias between HDR and NHEJ in repair of the same DSB induced by SpCas9-sgRNA that tightly bound with its target and by SpCas9-sgRNA variants with weakened target-binding affinities. In the HDR reporter, repair of the same Cas9-induced DSBs around the I-SceI site of I-SceI-GFP by HDR generates the “WT GFP”, whereas NHEJ generates “mutant GFP” due to disruption of the I-SceI site (Fig. 5c). We can separate these two repair outcomes in mESC by nested PCR and evaluate the HDR bias (i.e. the ratio of HDR to total edited) by deep sequence analysis. After HDR and NHEJ induced by SpCas9-gHRC4 in mESC, we found the HDR bias was nearly 3-fold lower with gHRC4 variants (G1C and 17nt) than with gHRC4, indicating a reduced HDR preference when the SpCas9-sgRNA binding affinity to its target is weakened (Fig. 5d). At the site targeted by gHRC2, where the HDR stimulation by DNA-PKcs inhibition was nearly abolished by the expression of SV40 LT in U2OS cells (Fig. 5b), SV40 LT expression increased the HDR bias, indicating a repair pathway shift from NHEJ to HDR (Fig. 5d). Therefore, for Cas9-sgRNA tightly bound to its cleaved target in cells, it is likely that its release from cleaved DNA requires local DNA replication, thus promoting HDR over NHEJ in repair of exposed DSBs.
While spontaneous dissociation of Cas9-sgRNA from cleaved DNA results in a conventional two-ended DSB, DNA replication that releases Cas9-sgRNA from its cleaved target may generate a three-ended DSB, with the leading strand likely forming a blunt end on one sister chromatid and the lagging strand a 3’-overhanging end with long ssDNA on the other sister chromatid (Fig. 5e). These two ends each can rejoin with the other blunt end of the DSB, or have a potential to directly ligate with each other, generating a palindromic chromosome from sister chromatid fusion (SCF) and potentially promoting chromatid breakage-fusion-bridge (BFB) cycles32–34 (Fig. 5e). Because neither DNA-PKcs nor Ku80 is engaged at Cas9-induced DSBs at the gHRC4 target site for repair in the HDR reporter in mESC (Fig. 1f,g), it is likely that Cas9-gHRC4 at this site may collide with a replication fork after DNA cleavage, generating a three-ended DSB and allowing subsequent fusion of two sister chromatids and production of a palindromic chromosome. Because the product contains the palindromic DNA sequence surrounding the junctions, a single primer could in theory be annealed to both the leading strand template and the newly synthesized lagging strand in the repair product for PCR amplification. However, no PCR products were detected from repair of Cas9-induced DSBs at the gHRC4 target site in the HDR reporter in mESC and U2OS cells with a single primer e.g. TF1, TF2 or TF3 (data not shown), likely due to the interference in PCR amplification by palindromic DNA sequences35. Given potentially asymmetric sequence deletion at the junction of SCF, we thus paired a distal primer to the break (TF2 or TF3) with the most proximal primer TF1 to minimize the length of palindromic DNA sequence in PCR amplification of repair products induced by SpCas9-gHRC4 in the HDR reporter and detected PCR bands over 250bp in mESC (Figs. 5f and S6a). In U2OS cells, these PCR bands were detected only after expression of SV40 LT (Figs. 5f and S6a). This is consistent with the observations that DNA-PKcs inhibition stimulates HDR induced by Cas9-gHRC4 in U2OS cells, but neither in mESC nor in U2OS cells highly expressing SV40 LT.
To further confirm that the PCR bands for these repair products were indeed fusions of sister chromatids via end ligation of Cas9-induced DSBs, we first cloned PCR products into a plasmid for Sanger sequencing. Among 40 clones for PCR bands with TF1 and TF2, 17 were from mESC and the rest from U2OS cells. Among 31 clones for PCR bands with TF1 and TF3, 29 were from mESC and the rest from U2OS cells. Sanger sequencing revealed only two sequence variations in each PCR band: DL251R6 and DL268R1 for the PCR band with TF1 and TF2 and DL231R5 and DL386R45 for the PCR band with TF1 and TF3 (Figs. 5g and S6b). They all contained some GFP sequences inverted around the break site but no palindromic GFP sequences, indicating that SCF may occur but palindromic sequences may be lost during repair or may not be amplified by PCR (Table S1). In addition, the deletion length in each sequence was distinctly asymmetric surrounding the break point, long at 231bp, 251bp, 268bp or 386bp at one direction and short at 1bp, 5bp, 6bp or 45bp at the other direction (Figs. 5g and S6b). It is likely that the collision between DNA replication and Cas9-sgRNA could generate long ssDNA at the lagging strand end and little or no ssDNA overhang at the leading strand end. Long ssDNA could be easily degraded, generating long deletion. PCR targeted amplicon sequencing also confirmed inverted GFP sequences with no palindromic fragments around the sites of Cas9-induced DSBs, but with more junction sequence variations (Fig. S7a,b). Taken together, these results indicate that Cas9-sgRNA could be released from its cleaved DNA target upon encountering local DNA replication, generating three-ended DSBs and subsequently inverted duplication via end-joining of sister chromatids.