Basis for discrimination by engineered CRISPR/Cas9 enzymes

CRISPR/Cas9 is a programmable genome editing tool that has been widely used for biological applications. While engineered Cas9s have been reported to increase discrimination against off-target cleavage compared with wild type Streptococcus pyogenes (SpCas9) in vivo, the mechanism for enhanced specificity has not been extensively characterized. To understand the basis for improved discrimination against off-target DNA containing important mismatches at the distal end of the guide RNA, we performed kinetic analyses on the high-fidelity (Cas9-HF1) and hyper-accurate (HypaCas9) engineered Cas9 variants. While DNA unwinding is the rate-limiting step for on-target cleavage by SpCas9, we show that chemistry is seriously impaired by more than 100-fold for the high-fidelity variants. The high-fidelity variants improve discrimination by slowing the rate of chemistry without increasing the rate of DNA rewinding—the kinetic partitioning favors release rather than cleavage of a bound off-target substrate because chemistry is slow. Further improvement in discrimination may require engineering increased rates of dissociation of off-target DNA.

Because these mismatches (Fig. S1) have been shown to play an important role in single molecule experiments of both DNA unwinding and conformational dynamics within the enzyme [17][18][19][20][21] , we chose to study the effects of this off-target using comprehensive kinetic analyses.
Before beginning our kinetic analyses, we determined the enzyme active site concentration for each of our Cas9 variants 22,23 . Measuring the amount of product formed in a titration of enzyme with increasing concentration of DNA, revealed active site concentrations of 31 nM, 26 nM, 23 nM for SpCas9, HypaCas9, and Cas9-HF1, respectively, for enzyme samples with a 100 nM nominal concentration based on absorbance at 280 nm as described (Fig. S2). It is important to note that the concentration of DNA required to saturate the signal is equal to the concentration of product formed, which eliminates concerns that some of the enzyme might bind DNA but not react. All subsequent experiments were set up using the concentration of active enzyme determined in the active site titration.
To compare the kinetics of on-or off-target DNA substrates of SpCas9 with the engineered variants, we first examined the time course of target strand (HNH) cleavage for each enzyme ( Fig.   1 and S3). Comparison of the cleavage rates of on-and off-target substrates by wild-type SpCas9 shows that the 3 bp PAM-distal mismatch slows the enzyme 13-fold (from 1 s -1 to 0.076 s -1 ). Both high-fidelity Cas9 variants dramatically decrease the rate of cleavage of on-target DNA substrates 21-to 35-fold compared to SpCas9 (0.028 s -1 for HypaCas9 and 0.047 s -1 for Cas9-HF1 vs 1 s -1 for SpCas9). Moreover, HypaCas9 and Cas9-HF1 further reduce the rates of off-target DNA cleavage 8-to 290-fold (rates of 0.0033 s -1 and 0.00016 s -1 , respectively) relative to their respective rates with on-target DNA.
Since our previous work identified R-loop formation as rate limiting for on-target cleavage and others subsequently suggested that R-loop formation rates dictate enzyme specificity for SpCas9 and Cas9-HF1 16 , we tested whether HypaCas9 would employ a similar mechanism. To directly measure the rates of R-loop formation for all enzymes, we used a stopped-flow assay by measuring fluorescence of tC o , a fluorescent tricyclic cytosine analog that is quenched by base stacking in dsDNA so that opening of the duplex results in a large increase in fluorescence. In the presence of Mg 2+ , SpCas9, HypaCas9, and Cas9-HF1 unwind the on-target DNA substrate at nearly identical rates (~2 s -1 ) (Fig. 2). Surprisingly, the rate of R-loop formation of off-target DNA substrates for all Cas9 variants was also largely unchanged (between 0.85 s -1 and 2.59 s -1 ).
Therefore, DNA unwinding is not rate-limiting and is not correlated with rates of cleavage for the high-fidelity variants.
Since enzyme specificity is a function of all steps leading up to the first largely irreversible step, steps other than the observed rate of R-loop formation must be considered. To estimate the intrinsic cleavage rate, SpCas9 was mixed with off-target DNA to form the SpCas9.DNA complex in the absence of Mg 2+ to allow binding and conformational changes to come to equilibrium. We then initiated the chemical reaction by the addition of 10 mM Mg 2+ . The rates of HNH and RuvC cleavage were measured to be 0.12 s -1 and 0.14 s -1 , respectively ( Figure S5c and S5d).
Intriguingly, these results show that the rate-limiting step in the enzyme pathway of SpCas9 offtarget cleavage is the chemistry step since the rate of R-loop formation we measured was 0.85 s -1 (Fig. 2b). In contrast, our previous, identical experiments showed that R-loop formation was ratelimiting with wild-type SpCas9 and on-target DNA, so discrimination is based, in part, by a change in rate-limiting step.
We repeated these experiments with HypaCas9 and Cas9-HF1 with on-or off-target DNA.
These results define intrinsic HNH cleavage rates of 0.035 s -1 and 0.0023 s -1 for HypaCas9 with on-and off-target substrates, respectively ( Figure S5g and S5k). Intrinsic cleavage rates of Cas9-HF1 for on-and off-target substrates were measured as 0.038 s -1 and 0.00014 s -1 , respectively ( Figure S5o and S5q). These intrinsic cleavage rates are somewhat faster than those measured with the simultaneous addition of DNA and Mg 2+ , indicating that some step other than R-loop formation but preceding chemistry may slow the net rate. Nonetheless, these results show that the intrinsic cleavage rates for on-target DNA are reduced ~100-fold for both HypaCas9 and Cas9-HF1 relative to SpCas9. For off-target DNA the intrinsic cleavage rates are reduced 50-or 860-fold for HypaCas9 or Cas9-HF1, respectively, relative to SpCas9.
Discrimination is not defined solely by the relative rates of DNA cleavage. Rather, because R-loop formation is fast, discrimination is a function of the kinetic partitioning between the rates of DNA release versus cleavage. In order to quantify the kinetic partitioning, we incubated enzyme and labeled DNA in the absence of Mg 2+ , which allows for R-loop formation, but prevents catalysis 15 and then added Mg 2+ and an excess of unlabeled DNA to serve as a trap. Comparison between parallel experiments performed in the presence and absence of the DNA trap provides an estimate for the fractional kinetic partitioning for dissociation versus cleavage of bound DNA.
Once SpCas9 was bound to on-target DNA, it was cleaved rapidly, and the DNA trap had little effect (Fig. 3a). In contrast, 33% of the off-target DNA disassociated from the enzyme, while ~67% of the DNA was committed to cleavage ( Fig. 3b and S4a). These results show that SpCas9 discriminates against the PAM-distal mismatched DNA by decreasing the rate of cleavage, increasing the fraction of DNA that is released rather than cleaved. However, the effect is small because the dissociation rate is so slow.
Next, we examined the kinetic partitioning for HypaCas9 and Cas9-HF1 bound to on-target DNA ( Figure 3C, E, S4B, and S4D). HypaCas9 and Cas9-HF1 show ~75% and ~92% of the ontarget DNA was cleaved in the presence of the trap. Because the intrinsic cleavage rate for ontarget DNA by HypaCas9 (0.035 s -1 ) and Cas9-HF1 (0.038 s -1 ) is 100-fold slower than with SpCas9 (4.3 s -1 ), the major effect of the two variants is to drastically slow the rate of cleavage.
This gives time for the DNA to dissociate before it is cleaved, but the dissociation rate is too slow to have a significant impact.
The increased kinetic partitioning to favor dissociation was further enhanced when HypaCas9 and Cas9-HF1 react with off-target DNA because the cleavage rates were further reduced to 0.0023 s -1 and 0.00014 s -1 , respectively (Figure 3d, f, S4c, and S4e). These rates are 50-to 860-fold slower, respectively, compared to SpCas9 on an off-target substrate. Accordingly, only ~24% and ~10% of the bound off-target DNA was committed to going forward for cleavage by HypaCas9 and Cas9-HF1, respectively, in the presence of trap DNA. Taken together, these results show that the engineered high-fidelity variants acquired improved specificity against the PAM-distal mismatched DNA through a markedly decreased rate of cleavage, which alters kinetic partitioning to favor release rather than cleavage of the bound substrate. Calculation of the apparent dissociation rates (Equation 4) show that the high-fidelity variants do not increase the rate of DNA release (Table 1). Rather, the increased discrimination is entirely attributed to decreases in the rate of cleavage.
To understand enzyme specificity, rate constants must be interpreted in the context of all kinetically relevant steps as illustrated in a free energy profile (calculated using Equation 5).
Because we have direct measurements of the rate constants for each relevant step (Scheme 1), we can construct a bona fide free energy profile ( Fig. 4 and S7-9). The free energy profiles comparing SpCas9, HypaCas9, and Cas9-HF1 show a change in rate-limiting and specificity determining steps. Enzyme specificity is defined by kcat/Km and is given by the highest overall barrier relative to the starting state, while the maximum rate, kcat, is defined by the highest local barrier relative to the preceding state. Because the rates of DNA binding do not change significantly with different substrates and enzymes, specificity is governed by the kinetic partitioning of the Cas9 R-loop (EDH) state to either go forward resulting in irreversible cleavage versus release via re-annealing of the DNA and ejection from the enzyme. The higher barriers for cleavage increase the kinetic probability for dissociation of the DNA.
Single molecule FRET measurements have suggested that the DNA rewinding is the major determinant of improved discrimination. In the absence of a correlation with chemistry steps, fluorescence signals are difficult to interpret unambiguously. Therefore, we tested the FRETpaired DNA substrates to determine what the FRET signal was measuring relative to the rates of cleavage. Cy3 and Cy5 were labeled on position -6 nt of the target strand and -16 nt on the nontarget strand, respectively 16 . First, we examined the time course of target strand (HNH) cleavage of the Cy3/Cy5 labeled DNA for each enzyme ( Figure S10). With wild-type Cas9, the reaction follows a single exponential with a markedly reduced rate (0.013 s -1 vs 1 s -1 for γ-32 P-labeled substrates), which represents a ~100-fold decrease in the rate of chemistry for SpCas9. The engineered HypaCas9 and Cas9-HF1 exhibited rates of 0.005 s -1 and 0.0064 s -1 , respectively, which are 5.6-fold and 7.3-fold slower than γ-32 P-or 6-FAM-labeled substrates measured under identical conditions. Interference by the Cy3/Cy5 labels masks the full impact of the high-fidelity variants.
Next, we measured DNA unwinding of FRET paired labeled on-and off-target DNA in the presence of Mg 2+ by stopped-flow methods. We observed the expected decrease in FRET due to increase in distance between donor and acceptor pairs with R-loop formation. However, the rate of R-loop formation of on-target DNA substrate for all Cas9 variants was significantly reduced by the Cy3/Cy5 label, from 1.69 s -1 to 0.018 s -1 . The FRET-pair label slows DNA unwinding rates by ~938-fold compared to the signal derived from tC o -or 2AP-labeled substrate ( Figure S11). Note that our control experiments demonstrated that tC o -or 2AP-labeled substrates did not affect rates of cleavage. Taken together, labeling of the DNA with bulky Cy3 and Cy5 labels dramatically impacted the enzyme, making these results difficult to interpret with respect to enzyme discrimination on native DNA.
In this study, we provide a comprehensive understanding of enhanced specificity of highfidelity Cas9 variants. HypaCas9 and Cas9-HF1 are seriously impaired in terms of enzyme efficiency of DNA cleavage. For each variant, the cleavage rate was 100-fold slower for on-target cleavage as compared to SpCas9. HypaCas9 and Cas9-HF1 gain discrimination mainly through slowing down chemistry, which shifts kinetic partitioning to favor release rather than cleavage of the bound DNA. We propose that Cas9 uses an induced-fit mechanism analogous to DNA polymerases, where a conformational change after initial substrate binding is an important determinant of enzyme specificity. For SpCas9, the conformational change is rate-limiting and determines specificity because R-loop formation is largely irreversible and is followed by fast chemistry 15 . The higher-fidelity variants have extraordinarily slow chemistry, allowing for release of the substrate from the enzyme before the irreversible cleavage reaction even though the dissociation rate is slow. DNA polymerases have evolved to dramatically increase the rate of dissociation of mismatched nucleotides in addition to decreasing the rate of catalysis 24,25 . Engineered Cas9 enzymes achieve improved discrimination against mismatches at the distal end of the guide sequence only by decreasing the rate of catalysis. This moderate improvement in discrimination may be sufficient in the context of the full recognition sequence to disfavor offtarget cleavage in vivo. However, further improvements to enable gene therapy may require engineered enzymes that increase rates of off-target DNA dissociation without requiring such drastic reductions in efficiency of on-target cleavage.
Cleavage probability = k chem k off + k chem
In vitro sgRNA transcription and refolding.
In vitro sgRNA transcription and refolding were performed as described previously 15 . The primers used for the templates of sgRNA transcription are listed (Table S2). Equimolar concentrations of complementary oligos were mixed in 10 mM Tris-HCl, pH 8.0, 50 mM NaCl, 1 mM EDTA and heated to 95℃ for 5 minutes, then slowly cooled to room temperature in about 60 minutes. The sgRNA was in vitro transcribed using the HiScribe Qiuk T7 RNA synthesis kit (New England Biolab) following the manufactory protocol. The transcribed sgRNA was further purified using a PureLink column (Thermo Scientific) and refolded by heating to 95℃ and then slowly cooled to room temperature in 10 mM Tris-HCl, pH 8.0, 50 mM NaCl, 1 mM EDTA.
DNA duplex formation and probe labelling. duplex used for HNH cleavage assays was prepared by -32 P or 6-FAM labeling the target strand before annealing with cold non-target strand at a 1:1.15 molar ratio.

Stopped-flow kinetic assay.
Stopped flow experiment was performed as previously described. 15 Briefly, cas9-gRNA complex Global analysis of kinetic data.
The kinetic data defining CRISPER-Cas9 cleavage were globally fit to the models shown in Scheme 1 by KinTek Explorer software (KinTek Corporation. Austin, TX) to obtain rate constants ( Figures S7-S9). FitSpace confidence contour analysis was performed to define the lower and upper limits for each kinetic parameter.

Active-site titration assay
To measure the active-site concentration of WTCas9 in off-target cleavage, a fixed concentration of enzyme (100 nM, estimated from absorbance at 280 nm) of WTCas9.gRNA was allowed to react with various concentrations of off-target DNA (5' end labeled on the target strand) in the presence of 10 mM Mg 2+ . According to a previous studyt 9 and our results, the intrinsic cleavage