Systematic identification of CRISPR off-target effects by CROss-seq

Graphical abstract

Dear Editor,

The CRISPR-mediated genome editing tools, including nucleases, base editors (ABE/CBE), transposases/recombinases, and prime editor (PE), have been extensively applied in basic and clinical researches, although the off-target effect remains a major concern (Anzalone et al., 2020). Recently, various methods have been developed to assess the specificity and accuracy of different tools (Zhang et al., 2021), yet each method is designed for limited editing systems, and none of them can simultaneously detect off-target sites in vivo and in vitro. A versatile method for profiling genome-wide off-target effects of various tools remains lacking.

A common feature of CRISPR-mediated genome editors is their dependence on RNA-guided site-specific DNA recognition, where an R-loop structure is formed and the opposite single-stranded DNA (ssDNA) is exposed (Anzalone et al., 2020). The recently reported KAS-seq (kethoxal-assisted single-stranded DNA sequencing) provides a unique and facile approach to map ssDNA in situ (Weng et al., 2020; Wu et al., 2020). We thus speculated that the binding events of CRISPR-mediated genome editors could be revealed via detecting the ssDNA component of R-loop. To test this, we carried out CRISPR-mediated genome editing in cells or isolated genomic DNA. The N₃-kethoxal labeled ssDNA region allows subsequent biotinylation through azido group-associated click reaction and can be enriched for deep sequencing (Fig. 1A). Meanwhile, an optimized labeling condition was developed for in vitro CRISPR-Cas targeting assay (Fig. S1). We termed this method CROss-seq (CRISPR Off-targeting ssDNA sequencing) and speculated that it could detect targeting specificity of multiple CRISPR-mediated genome editors and facilitate cross-validation of gRNA dependent off-target effects in vivo and in vitro (Fig. 1A).

First, we designed sgRNA for two target sites (HBG1 and HBG2) (Liang et al., 2019) and conducted SpCas9 cleavage assay in vitro. To determine the optimal reaction condition, we used a gradient of N₃-kethoxal working concentrations. Samples treated in parallel but without sgRNA were applied as negative controls. We found highly enriched signals at target sites in editing samples compared to the control sample (Fig. 1B). In addition, the enriched fragments showed a clear cleavage signal precisely at the putative editing site, confirming the effective identification of SpCas9 targeting (Fig. 1B).

Next, we defined a CROSS score that calculates the number of reads with cleavage signal at a given locus under normalized sequencing depth, and applied this score to evaluate the reliability (Fig. S2) (Kim et al., 2015). The optimized labeling condition yielded the largest number of putative off-target sites (Fig. S3A and S3B; Table S1). Consensus sequence analysis of these sites reflected the expected spacer sequence of the sgRNA, with better alignment in the PAM-proximal region (Fig. S3C). We then compared the candidate sites to the off-target sites reported by Digenome-seq using the same sgRNA (Liang et al., 2019). Among the 45 reported sites, 18 of them could be confirmed by CROss-seq (Fig. S3D). Notably, the Digenome-seq cleavage scores of the overlapped sites were significantly higher than the others (Fig. S3E). We tested other two sgRNAs (HEK4 and FANCF) and compared to previous methods, showing the reproducibility of CROss-seq (Fig. S3F and S3G). To further validate the accuracy, we performed targeted deep sequencing on the selected top off-targets as well as two on-target sites according to CROSS scores (Fig. 1C). Indels were observed at 13 out of the 30 sites, with frequencies ranging from 0.28% to 20% (Fig. 1D). The use of another sgRNA targeting HEK2 showed similar results (Fig. S4; Table S2).

We next applied CROss-seq in vivo. After 12- or 24-h Cas9 transfection, cells were treated with N₃-kethoxal for 45 min. A negative control without sgRNA supplement was conducted in order to eliminate the background signals derived from endogenous ssDNA regions. The 24-h group showed strong CROss-seq signal intensity at the HBG1 and HBG2 target sites (Fig. S5A). Interestingly, reads around the targeting sites displayed both indels and truncations, suggesting that N₃-kethoxal can effectively label the target region before and after the Cas9-induced DNA repair (Fig. S5A). The candidate off-target sites were then ranked and filtered by their CROSS scores (Fig. S2). Besides on-target sites, we identified three off-target sites from the 12-h group and 36 off-target sites from the 24-h group (Fig. S5B and S5C; Table S3), suggesting that the off-target effects positively correlated to the editing intensity in a time-dependent manner. Consensus sequence analysis of off-target sites recapitulated the designed spacer sequence, supporting the accuracy of CROss-seq in detecting sgRNA-dependent off-targets (Fig. S5D). CROss-seq using cells treated with sgRNA targeting another gene HEK2 showed comparable results (Fig. S6; Table S4).

To evaluate the accuracy of CROss-seq in vivo, we selected the top off-target sites according to the CROSS scores and performed targeted deep sequencing. NGS data revealed that 10/11 (91%) of these sites showed an off-target cleavage signal in cellulo (Fig. 1E), and seven of them harbored obvious indels with frequency ranging from 1% to 40% (Fig. 1F). Compared to other cell-based and cell-free methods, CROss-seq showed the highest validation rate of off-targets (Fig. S5E).

In vivo and in vitro assays for off-target detection usually show widespread discrepancies (Kim and Kim 2018). CROss-seq offered the opportunity to evaluate the off-target effects in vitro and in vivo simultaneously. Through a comparative analysis, we found that in vitro CROss-seq assay revealed more off-target sites (Fig. S5F). To further evaluate their consistency, we found a strong correlation (R = 0.85) between CROSS score and indel level for the off-target sites detected in vivo. However, no such correlation was observed for in vitro assay (R = −0.11) (Fig. 1G). These results highlighted the importance of cross-validation of off-target effects through multiple strategies.

Compact chromatin structure hinders the nuclease access and thus results in fewer off-target sites (Kim and Kim 2018). We therefore compared the off-target sites to the DNase I hypersensitivity sites (DHSs), and found that sites identified by CROss-seq in vivo assay were more frequently overlapped with the DHSs (Fig. S5G). Moreover, the average DHSs signal intensity of these sites was significantly higher (Fig. 1H). These results collectively indicated the influences of chromatin accessibility in CRISPR-Cas9 genome editing, which again underlined the need for a cross-validation method applied simultaneously in vitro and in vivo.

The performance of CROss-seq in the SpCas9 system prompted us to apply it in base editor systems (Komor et al., 2016). After BE3 editing, the CROss-seq in vitro assay revealed an evident peak on the expected targeting locus compared to negative control (Fig. S7A). We also observed a clear truncation signature towards the cleavage site, reflecting the nickase activity of nCas9 in BE3 editing (Fig. S7A). Then we used the CROSS score for off-target sites detection and discovered 12 putative sites with a consensus sequence resembling the target sequence (Fig. S7B and S7C; Table S5). Notably, CROss-seq detected more off-target sites than Digenome-seq did (Kim et al., 2017), and only the on-target site was commonly detected by both methods (Fig. S7D). We then evaluated the off-target effects of the in vivo BE4max system (Koblan et al., 2018). We therefore adjusted the CROSS-scoring to assess the editing effects by calculating the ratio of reads with C-to-T or A-to-G conversion within the editing window (Fig. S8). As expected, we observed apparent CROss-seq peaks and C-to-T conversion at the HEK2 target site (Fig. 2A; Table S6). Meanwhile, we did not detect any off-target site, indicating a minor sgRNA-dependent off-target effect of BE4max in vivo.

Next, we applied CROss-seq to examine the off-target effects of the adenine base editor (ABE) (Gaudelli et al., 2017; Richter et al., 2020). Similar to CBE, the in vitro assay captured abundant reads distributed near the HEK2 target site, showing a truncation signature resulting from nCas9 (Fig. S7E). We identified 25 off-target sites according to the CROSS score (Fig. S7F and S7G; Table S7). CROss-seq reads from the in vivo assay were also enriched near the target site, most of which contained an A-to-G editing signature (Fig. 2B). Meanwhile, six off-target sites were identified by the scoring program (Fig. 2C; Table S8). Consensus sequence analysis of the captured off-target sites agreed with the expected target site, and the PAM-proximal region was aligned with more accordance (Fig. 2D). Among these six off-target sites, five of them displayed A-to-G transition signal (Fig 2E and 2F; Fig. S9), indicating high reliability of CROss-seq in base editor systems.

Prime editor (PE) holds great promise in genome editing for its high specificity and expanded editing scope (Anzalone et al., 2019). Although nDigenome-seq (Kim et al., 2020) has been developed to assess the off-target effects of PE in a cell-free assay, the off-target effects of PE have not been tested in mammalian cells. To assess the off-target effects of PE in vivo, we transfected HEK293T cells with plasmids encoding PE2 and pegRNA targeting HEK3, and performed CROss-seq after 24 h. CROss-seq sequencing reads were evidently enriched near the target site, most of which contained a CTT insertion signature at the +1 position adjacent to the pegRNA-induced nick (Figs. 2G and S10A). Strikingly, only the on-target site could pass the CROSS-scoring cutoff (Figs. S10B, S10C and S11; Table S9). A recent study reported evident off-target effects of PE2 editing system by using nDigenome-seq (Kim et al., 2020). We therefore applied CROss-seq to investigate the editing with the pegRNA targeting the same locus HEK4. We identified the designed on-target site and one off-target site across the genome (Fig. 2H–J; Table S10). Of note, recent reports also indicated that PEs rarely induce off-target (Jin et al., 2021). These results together demonstrated the minimal off-target effects of PE.

In summary, we developed CROss-seq, a versatile, efficient, and cost-effective method for genome-wide off-target profiling in multiple genome editing systems. We applied CROss-seq both in vivo and in vitro to evaluate the off-target effects of Cas9 nuclease, base editors, and prime editor, while this method should also be adapted to other currently available or even potentially future CRISPR-mediated genome editors. Our results showed that the in vivo assay was more specific while in vitro assay was more sensitive, highlighting the necessity of coupling in vivo and in vitro assays to achieve a comprehensive and precise assessment of the off-target effects. CROss-seq also has its limitations. For example, the guanine density may influence the N₃-kethoxal labeling efficiency, although the vast majority of human or mouse genome contains at least one G in the putative target regions (Fig. S12). To reduce the background noise, sample treated without sgRNA can be used as a negative control, which means CROss-seq can only identify gRNA-dependent editing events. We also anticipate applying CROss-seq in tissues and clinical samples with specific optimization. In conclusion, CROss-seq contributes to a deeper understanding of CRISPR-mediated tools in broader application scenarios.

References

[1] Anzalone AV, Koblan LW, Liu DR. Genome editing with CRISPR–Cas nucleases, base editors, transposases and prime editors. Nat Biotechnol 2020;38:824–844.

[2] Anzalone AV, Randolph PB, Davis JR et al. Search-and-replace genome editing without double-strand breaks or donor DNA. Nature 2019;576:149–157.

[3] Gaudelli NM, Komor AC, Rees HA et al. Programmable base editing of A·T to G·C in genomic DNA without DNA cleavage. Nature 2017;551:464–471.

[4] Jin S, Lin Q, Luo Y et al. Genome-wide specificity of prime editors in plants. Nat Biotechnol 2021;39:1292–1299.

[5] Kim D, Bae S, Park J et al. Digenome-seq: genome-wide profiling of CRISPR-Cas9 off-target effects in human cells. Nat Methods 2015;12:237–243, 1 p following 243.

[6] Kim D, Kim J. DIG-seq: a genome-wide CRISPR off-target profiling method using chromatin DNA. Genome Res 2018;28:1894–1900.

[7] Kim D, Lim K, Kim S et al. Genome-wide target specificities of CRISPR RNA-guided programmable deaminases. Nat Biotechnol 2017;35:475–480.

[8] Kim DY, Moon SB, Ko J et al. Unbiased investigation of specificities of prime editing systems in human cells. Nucleic Acids Res 2020;48:10576–10589.

[9] Koblan LW, Doman JL, Wilson C et al. Improving cytidine and adenine base editors by expression optimization and ancestral reconstruction. Nat Biotechnol 2018;36:843–846.

[10] Komor AC, Kim YB, Packer MS et al. Programmable editing of a target base in genomic DNA without double-stranded DNA cleavage. Nature 2016;533:420–424.

[11] Liang P, Xie X, Zhi S et al. Genome-wide profiling of adenine base editor specificity by EndoV-seq. Nat Commun 2019;10:1–9.

[12] Richter MF, Zhao KT, Eton E et al. Phage-assisted evolution of an adenine base editor with improved Cas domain compatibility and activity. Nat Biotechnol 2020;38:883–891.

[13] Weng X, Gong J, Chen Y et al. Keth-seq for transcriptome-wide RNA structure mapping. Nat Chem Biol 2020;16:489–492.

[14] Wu T, Lyu R, You Q et al. Kethoxal-assisted single-stranded DNA sequencing captures global transcription dynamics and enhancer activity in situ. Nat Methods 2020;17:515–523.

[15] Zhang H, Li T, Sun Y et al. Perfecting targeting in CRISPR. Annu Rev Genet 2021;55:453–477.

RIGHTS & PERMISSIONS

©The Author(s) 2022. Published by Oxford University Press on behalf of Higher Education Press.