Real-time observation of DNA recognition and rejection by the RNA-guided endonuclease Cas9

Binding specificity of Cas9–guide RNA complexes to DNA is important for genome-engineering applications; however, how mismatches influence target recognition/rejection kinetics is not well understood. Here we used single-molecule FRET to probe real-time interactions between Cas9–RNA and DNA targets. The bimolecular association rate is only weakly dependent on sequence; however, the dissociation rate greatly increases from <0.006 s−1 to >2 s−1 upon introduction of mismatches proximal to protospacer-adjacent motif (PAM), demonstrating that mismatches encountered early during heteroduplex formation induce rapid rejection of off-target DNA. In contrast, PAM-distal mismatches up to 11 base pairs in length, which prevent DNA cleavage, still allow formation of a stable complex (dissociation rate <0.006 s−1), suggesting that extremely slow rejection could sequester Cas9–RNA, increasing the Cas9 expression level necessary for genome-editing, thereby aggravating off-target effects. We also observed at least two different bound FRET states that may represent distinct steps in target search and proofreading.

1) The manuscript cites very few references despite rapidly growing number of studies that investigate Cas9 binding. As already stated above authors should compare in much more detail their achievements with respect to previous studies in CRISPR-field and to what extent the authors go beyond. The manuscript will certainly benefit from more comprehensive comparison that could also include R-loop formation in other CRISPR systems.
2) The authors report second bound state. How clearly defined is that state? It appears to me that authors only allow 3 states when obtaining their idealized FRET trajectories (e.g. Suppl. Fig 4b).
Could it in reality be that second state actually consists of more states? Second state is very broadly distributed in transition density plots.
3) Please state more clearly what interpretation of precipitously decreased life time for PAMproximal mismatches is. I suspect that in these cases R-loops that reach PAM-distal end are not forming at all, such that only target scanning occurs? Or do authors suspect destabilized R-loops due to presence of the PAM proximal mismatches? 4) Figure 3 is very busy and labels within some of figures become very small (e.g. Fig. 3C). I suggest to subdivide it into two figures, e.g. by splitting of model (Fig. 3e).

Reviewer #2 (Remarks to the Author):
Singh et al use single-molecule FRET analysis to probe real-time interactions between RNAprogrammed Cas9 and its DNA targets. The key finding is that the dissociation rate of Cas9 from DNA is highly dependent on mismatches between the guide RNA and the DNA sequence that are proximal to the PAM sequence. Mismatches distal to the PAM sequence having little to no effect on Cas9 binding to DNA. The authors also detect two different bound FRET states, which may represent distinct steps in target searching and/or proof reading.
Although, the concept of PAM, seed sequence and off-rate controlled binding have been presented before this study provides novel insight into the effects on mismatches between the target and the guide RNA, at the single-molecule level (at a resolution not before seen). Also novel is the Detection of two bound FRET states, which provides evidence for a two-step mechanism of Cas9 DNA binding -the first step being PAM surveillance and the second involving RNA-DNA heteroduplex formation upon PAM recognition (which is also supported by recent structural data).
The experiments were expertly performed and interpreted. The data is of high quality and supports the conclusions presented. I recommend publication of this work as is.
We appreciated the constructive criticisms of the reviewers and have addressed their concerns by adding more text, explanations, references and figures. We hope that the revised manuscript will comply with the referees' remarks.

Point by point response, Reviewer #1:
1) The manuscript cites very few references despite rapidly growing number of studies that investigate Cas9 binding. As already stated above authors should compare in much more detail their achievements with respect to previous studies in CRISPR-field and to what extent the authors go beyond. The manuscript will certainly benefit from more comprehensive comparison that could also include

R-loop formation in other CRISPR systems.
The introduction of the article has been reworked with the addition of a new paragraph that references a majority of previous studies that have been conducted to investigate off-target effects of Cas9-RNA and related CRISPR-Cas design tools.
Also included is a more detailed comparison with a single-molecule 1 study that investigated the Cas9-RNA induced RNA-DNA heteroduplex formation via magnetic tweezers, and found that 11 PAM-proximal matches are sufficient for stable RNA-DNA heteroduplex formation with StCas9 (the Cas9 ortholog from Streptococcus thermophilus). In our current study using SpCas9 (from Streptococcus pyogenes, simply referred to as Cas9), we found 9-10 PAM-proximal matches to be sufficient for ultra-stable Cas9-RNA binding. The stability of RNA guided CRISPR enzymes and DNA targets depends on energetic contributions of the RNA-DNA heteroduplex and interactions between the DNA target and amino-acid residues of the CRISPR-Cas enzymes, which may explain subtle mechanistic differences between the various Cas9 orthologs.
2) The authors report second bound state. How clearly defined is that state? It appears to me that authors only allow 3 states when obtaining their idealized FRET trajectories (e.g. Suppl. Fig 4b). Could it in reality be that second state actually consists of more states? Second state is very broadly distributed in transition density plots.
The reviewer is correct in suggesting that the second bound state may consist of more  This is a good point. In response, we have modified a sentence in the main text as following.
"The amplitude-weighted lifetime of the bound state (τ avg ) decreased precipitously even with just 2 bp PAM-proximal mismatches, likely because the R-loop failed to extend beyond the mismatches." Figure 3 is very busy and labels within some of figures become very small (e.g. Fig. 3C). I suggest to subdivide it into two figures, e.g. by splitting of model (Fig.   3e).

4)
We thank the reviewer for this helpful suggestion. Figure 3 has now been divided into two separate main text figures, i.e. Figure 3 with all the lifetime data and Figure 4 with the kinetic model of Cas9-RNA targeting.

Point by point response, Reviewer #2:
5) The experiments were expertly performed and interpreted. The data is of high quality and supports the conclusions presented. I recommend publication of this work as is.
We are pleased with the reviewer's positive feedback.