Dear Editor,

Genome editing techniques have been rapidly developing in recent decades1. Among them, site-specific cleavage of genomic loci in various organisms by homing endonucleases (HEases)2, Zinc finger nucleases (ZFNs)3, transcription activator-like effector nucleases (TALENs)4, and most recently the CRISPR (clustered regularly interspersed short palindromic repeats)/Cas9 system5, has been utilized widely not only in laboratories but also for translational studies. The central issue of genome editing is how to achieve specific and robust recognition of particular genomic sequences. In the case of HEases, ZFNs, and TALENs, this is achieved by specific intermolecular interactions between nucleotides and protein motifs, while for CRISPR/Cas9, the specificity is due to Watson-Crick base pairing between CRISPR RNA (crRNA) and its recognition sequences. The crRNA targets a 20-bp complementary target DNA sequence, which is flanked by a proto-spacer adjacent motif (PAM). Recent crystal structure studies and single-molecule DNA curtain experiments suggested that PAM site is essential for the initiation of Cas9 binding while the seed sequence corresponding to 3′ end of the crRNA complementary recognition sequence which is also directly adjacent to PAM is critical for subsequent Cas9 binding, R-loop formation, and activation of nuclease activities in Cas96,7,8. While extensive efforts have been focused on optimizing the efficiency of targeting and cleavage by the CRISPR/Cas9 system in various organisms, relatively few studies investigated the mistargeting, so called off-targeting, activities9,10,11,12. In those studies, limited numbers of potential target DNA sequences with point or combined mismatches in comparison with the authentic targeting sites were tested for in vitro and in vivo cleavage activities. While these studies suggest the off-targeting activities of Cas9 on those examined sites are limited, for a comprehensive view an unbiased genome-wide analysis is required. Here, by a robust unbiased approach, we demonstrated that CRISPR/Cas9 had crRNA-specific off-target binding activities in human genome. However, most of those binding off-targets could not be efficiently cleaved both in vivo and in vitro, suggesting that the off-target cleavage activity of CRISPR/Cas9 in human genome is very limited.

To unbiasedly determine the off-targets of CRISPR/Cas9 in vivo, we hypothesized that the Cas9/crRNA complex must first bind significantly to those off-targets, which could be revealed by chromatin immunoprecipitation and high throughput genome sequencing (ChIP-seq). We tagged Venus protein to either N- or C-terminal of humanized Cas9 (hCas9) protein (Figure 1A), which was subsequently co-transfected together with different single guide RNAs (sgRNAs) into human HEK-293T cells. Cas9-Venus C-terminal fusion protein (Cas9-CV) showed similar sgRNA-dependent in vivo cleavage activity on its genomic target as untagged wild type protein, while N-terminal Venus-Cas9 (NV-Cas9) fusion showed no activity (Supplementary information, Figure S1A). In addition, Cas9-CV also showed similar cleavage activity on a previously identified emx1 off-target (emx1-OT4)9,11 but did not cleave the control locus (Supplementary information, Figure S1B-S1D). Mutations of both HNH nuclease and RuvC catalytic domains (DM-Cas9-CV) abolished the cleavage activity (Supplementary information, Figure S1A, lane5). We performed chromatin immunoprecipitation using high affinity nanobody for the Venus protein (GBP)13. Cas9-CV was significantly enriched in the emx1 locus but not control egfa-t1 locus in an sgRNA-dependent manner, while DM-Cas9-CV showed a greatly enhanced binding in comparison with Cas9-CV (Figure 1B and Supplementary information, Figure S1E). Importantly, emx1-OT4 could also be significantly enriched by the ChIP approach (Figure 1C and Supplementary information, S1F-S1I). Therefore, we used DM-Cas9-CV in all subsequent ChIP experiments.

Figure 1
figure 1

GFP nanobody-based ChIP-seq analysis to unbiasedly identify genome-wide off-targets of CRISPR/Cas9 in human genome. (A) Schematic view of Cas9 constructs. Venus was fused to wild type hCas9 or hCas9 double mutant (D10A and H840A). (B) The sgRNA-specific in vivo binding of Cas9-CV and DM-Cas9-CV to endogenous emx1 and egfa-t1 loci. 293T cells were transfected with sgRNA for the emx1 locus or without sgRNA as control. The ChIP signals are divided by the no-sgRNA signals. (C) ChIP enrichment on emx1, previously identified emx1 off-target (emx1-OT4), control efga-t1, as well as other genomic loci containing sequences with 2-3 nucleotide mismatches in comparison to the original emx1 target sequence, was examined in 293T cells transfected with sgRNA for emx1. (D) ChIP signals (normalized read counts) around target and off-target sites in ChIP-seq libraries. (E) ChIP-seq peaks identified in sgEMX1 libraries. (F) Consensus binding motif analysis of peaks in sgEMX1 libraries. Sequence motifs were identified within 101 base pairs around peak summits using MEME. (G) Confirmation of ChIP-seq-identified peaks in sgEMX1 libraries by quantitative PCR. All 12 overlapped peaks as well as randomly selected peaks from the rest of individual sgEMX1 libraries were checked. (H) In vitro cleavage assay of identified binding off-target sites. Error bars indicate the standard deviation (STDEV). Student's t-test was performed (*P< 0.05, ***P< 0.0005). All experiments were replicated at least twice. See also Supplementary information, Figure S1 and Table S1.

We performed ChIP-seq analysis in HEK-293T cells co-transfected which DM-Cas9-CV and no sgRNA or sgRNAs targeting either the emx1 or efga-t1 locus. Biological repeats were performed to reduce potential noises in the assay. In pooled ChIP-seq libraries, the original targeting sites of emx1, egfa-t1, as well as the known off-target of emx1, emx1-OT4, showed significant specific sgRNA-dependent enrichment (Figure 1D). To achieve more stringent identification of the off-targets, during MACS peak calling, we set the threshold as FDR < 0.5%. In ChIP-seq libraries generated from 293T cells without transfected sgRNA, no peak was identified, while in libraries generated from cells transfected with egfa-t1 sgRNA, only the original target site was identified in biological repeats (Supplementary information, Table S1A). For emx1 sgRNA, 50 and 63 peaks were identified in each biological repeat, and 12 overlapped peaks were finally obtained (Figure 1E). Interestingly, most of the 50 (39/50) and 63 (42/63) peaks contain conserved motifs which correlate well with PAM and its 5′ 10-12 bp seed region, while all 12 peaks that appeared in both biological repeats contain such conserved motifs (Figure 1F and Supplementary information, Figure S1J-S1L and Table S1C-S1E). We further confirmed these identified peaks by quantitative PCR. Most of them showed significant sgRNA-dependent specific enrichment, with some showing comparable enrichment as the original emx1 locus (Figure 1G and Supplementary information, Figure S1M). Finally, we checked whether these sites corresponding to peaks could indeed be cut by Cas9. Surprisingly, in both in vitro and in vivo cleavage assays, most of these binding off-targets could not be significantly cleaved while the emx1 original site and its known off-target (emx1-OT4) were almost completely cleaved by Cas9/sgRNAemx1 (Figure 1H and Supplementary information, Figure S1I, S1N). Only two binding off-targets, OT2-1 and OT2-4, reproducibly showed weak cleavage (Figure 1H and Supplementary information, Figure S1N). These results suggest that substrate binding could be uncoupled from the cleavage step in the CRISPR/Cas9 system.

One of the major concerns about genome editing is the potential off-target effect of editing enzymes which may lead to unexpected genomic instabilities such as mutations and chromosomal translocations. By an unbiased genome-wide ChIP-seq approach, we analyzed binding off-targets of CRISPR/Cas9 in human genome. Surprisingly, while Cas9 could bind to various genomic sequences containing PAM and conserved seed sequences in an sgRNA-specific manner, its cleavage off-targets are very limited in comparison with other genome-editing enzymes, such as HEases, ZFNs, and TALENs. This might be largely due to additional involvement of the target sequence annealing step in activating the cleavage activities of CRISPR/Cas9 complex on its targets. On the other hand, the sgRNA-specific off-target binding activities may significantly affect other recently developed approaches which combine the nucleotide sequence binding specificity of CRISPR/Cas9 with other non-cleavage associated functions such as transcription regulation14 and fluorescent labeling15.

For the sgRNA targeting emx1, there are many more genomic loci which contain the PAM and conserved seed (10+3 base pairs) region in the human genome (Figure 1F and data not shown). It could be speculated that binding of Cas9/sgRNAemx1 to those loci might be blocked by multiple factors including cell type- and/or development-specific local chromatin structure and modifications. Our preliminary results from HeLaS3 cells (Supplementary information, Figure S1P, Table S1B and S1F) have identified 19 potential binding off-targets for sgRNA emx1, and most of them did not overlap with the binding off-targets identified in HEK293T cells. Nevertheless, most of those HeLa cell-specific off-targets also contain similar conserved seed and PAM regions (Supplementary information, Figure S1Q-S1R). This suggests that off-targets might be cell type dependent and determined by various complicated factors in addition to primary DNA sequences. In addition, we hypothesize that different sgRNAs might have greatly variable levels of off-target binding activities which might correlate with the kinetics of target DNA duplex disruption, formation of DNA-RNA heteroduplex, and R-loop expansion. The contribution of nucleotide sequence and composition of both the seed region and its 5′ surrounding region needs further detailed studies. Our unbiased approach provides a valuable tool to further investigate the molecular mechanism of CRISPR/Cas9 and to optimize its in vivo applications.


During the preparation of this letter, two Nat Biotechnol papers appeared online ( and with similar conclusions.