Enhancing gene editing specificity by attenuating DNA cleavage kinetics


Engineered nucleases have gained broad appeal for their ability to mediate highly efficient genome editing. However the specificity of these reagents remains a concern, especially for therapeutic applications, given the potential mutagenic consequences of off-target cleavage. Here we have developed an approach for improving the specificity of zinc finger nucleases (ZFNs) that engineers the FokI catalytic domain with the aim of slowing cleavage, which should selectively reduce activity at low-affinity off-target sites. For three ZFN pairs, we engineered single-residue substitutions in the FokI domain that preserved full on-target activity but showed a reduction in off-target indels of up to 3,000-fold. By combining this approach with substitutions that reduced the affinity of zinc fingers, we developed ZFNs specific for the TRAC locus that mediated 98% knockout in T cells with no detectable off-target activity at an assay background of ~0.01%. We anticipate that this approach, and the FokI variants we report, will enable routine generation of nucleases for gene editing with no detectable off-target activity.


Engineered nucleases have transformed the practice of genome editing by enabling methods that are general, robust and highly efficient1. However the specificity of these reagents remains a concern2,3,4. This issue is especially important for therapeutic applications, as even a single unintended genome alteration can potentially drive oncogenesis5,6. The challenge of ensuring adequate specificity is compounded when very high levels of on-target modification are required, as driving desired outcomes toward saturation can require a disproportionate increase in nuclease dose. This consideration impacts diverse applications for which utility scales with editing efficiency, including in vivo and ex vivo therapies7,8,9,10, where the need for high levels of biallelic or multiplexed editing can require near complete modification of each target.

A number of strategies have been described for enhancing DNA cleavage specificity. The details of these approaches have varied with the features and limitations of the nucleases to which they have been applied. For all-protein systems (for example, transcription activator-like effector nucleases (TALENs)11, ZFNs12, and meganucleases13) a direct approach for improving specificity has been to reengineer the base-sensing interface of individual reagents14,15,16,17. Other more general strategies have included breaking dimerization symmetry to suppress formation of unintended nuclease species18 as well as removing nonspecific DNA contacts19. For CRISPR systems, which feature shorter targets, a recurrent focus has been extension of the recognition event required for cleavage, to remedy limitations with unique addressing in complex genomes20,21,22,23. Opportunities for engineering the base-sensing interface have been comparatively limited, given the dominant role of Watson–Crick interactions in target recognition, although recent studies have begun to examine the interface with the protospacer adjacent motif (PAM)24,25,26 and synthetic guides27,28. The specificity of Cas9 has also been improved via removal of nonspecific contacts2,29,30, albeit at the cost of reduced activity for resulting variants31,32,33,34. Recent studies have begun to explore selection systems for identifying Cas9 variants with improved on-target preference3,34,35. Each of these approaches has yielded reagents with improved specificity relative to a parental framework. However, to date, no engineered nuclease has demonstrated complete modification, in an application-relevant cell type, that is uniquely specific when assessed via unbiased, high-sensitivity methods.

A shared feature of these strategies for enhancing specificity has been their emphasis on initial target recognition for interpretation and rationale. The potential for improving cleavage preference by engineering downstream steps, and in particular by optimization of reaction kinetics, has remained comparatively less explored. This would appear to present an important untapped dimension for improving performance, given the minimal catalytic requirements for successful genome editing, which can require just one cleavage event per cell. Although recent studies of Streptococcus pyogenes Cas9 have begun to explore cleavage dynamics36,37,38 and structural intermediates39,40,41, efforts to capitalize on these insights have thus far been modest and have not demonstrated unique specificity in highly modified cells2. Moreover, no such studies have been reported for TALENs or ZFNs, whose longer binding sites enable more mismatches and a greater energetic penalty for interaction with non-targeted sequences, to which kinetic optimization might be added to yield highly specific cleavage.

In this study, we show that ZFN specificity may be substantially improved by re-engineering the FokI cleavage domain. Screening single-residue variants of a previously published ZFN dimer42, we identify diverse substitutions that fully preserve on-target activity while globally suppressing off-target cleavage up to 1,000-fold. Using a highly sensitive indel assay that we have developed, we demonstrate a 3,000-fold increase in on- to off-target preference via replacement of a single residue that contacts the catalytic active site. We also characterize a substitution within the zinc finger motif that eliminates a conserved nonspecific DNA contact and provides a complementary means for improving ZFN specificity. By combining these approaches, we develop ZFNs that disrupt the TRAC locus (encoding the constant portion of the T-cell antigen receptor alpha chain) in human T cells with an efficiency of >98% and no detectable off-target activity.


Engineering the FokI nuclease domain as a strategy for enhancing ZFN specificity

In pursuing this work, a key motivation was our observation that ZFNs can exhibit a high degree of binding specificity that may not be recapitulated in cellular cleavage studies (Supplementary Fig. 1a,b). Such behavior could be due to catalysis occurring too quickly to allow discrimination between on- and off-target sequences. As both a test of this model and a potential means for improving ZFN performance, we sought to identify mutations that might slow down cleavage to the point where the kinetic constant for this step would be slow, relative to dissociation from off-target sites. Viewed in the context of the standard Michaelis–Menten framework (Supplementary Fig. 1c), our goal was to reduce k2 to a value that was much less than k−1 for all off-target sites43. To accomplish this, we focused engineering efforts on the FokI cleavage domain, given its direct role in catalysis. This domain also exhibits very low affinities for dimerization44 and for binding DNA45, which suggests a minimal contribution to the overall stability of the ZFN–DNA complex.

Our first step was to perform a screen of nuclease variants bearing single-residue substitutions for the ability to retain on-target activity while showing reduced cleavage at known off-target sites. Our test system consisted of a ZFN pair targeted to the human AAVS1 safe harbor42 (Fig. 1a) expressed in K562 cells and monitored for activity at the intended locus and four known off-target sites (OT1, OT2, OT3 and OT4). In choosing residue substitutions for this initial scan, we adopted a strategy of reducing positive charge that involved replacement of each basic residue with serine, as well as each amide residue with its carboxylate side chain counterpart. This provided a distribution of variants that avoided gross locational bias while testing an initial hypothesis that global charge might influence FokI domain localization in the context of the DNA-bound ZFN and thus alter cleavage rate. A scan of these mutations (40 total; Fig. 1b) in the context of the left ZFN revealed 7 where the mutated nuclease showed a clear increase in on-target preference (3- to 80-fold) and good retention of activity (62–124%) relative to the parent (Fig. 1c and Supplementary Table 1). Screening results with both the right and left ZFN were broadly concordant (Supplementary Table 1). In a structural model of the FokI cleavage complex, these residues (R416, R422, N476, Q481, K525, N527 and Q531) are all predicted to lie within 10 Å of the DNA interface, which represents a significant enrichment of hits from this proximity class (Fisher’s exact test, P < 0.001; Supplementary Table 1). These residues also cluster near the active site (Fig. 1d).

Fig. 1: Identification of FokI substitutions that improve ZFN on-target cleavage preference.

a, Sketch of the host ZFN pair used for these studies bound to its target site within the human AAVS1 safe harbor locus42. ZFN-L and ZFN-R denote the left and right monomers, respectively, while ELD and KKR indicate the corresponding heterodimer FokI variants18,60. Note that ZFN-R uses a longer linker between its second and third fingers (counting from the N terminus), which allows skipping of a single base as indicated55. b, Sequence of the FokI cleavage domain used in these studies with residues numbered as in ref. 61. The 40 positions queried via residue substitutions are shown in red. Underlines indicate the 7 residues (R416, R422, N476, Q481, K525, N527 and Q531) that resulted in retention of >60% parental activity with more than threefold improved specificity upon substitution. Active site residues are colored blue, and residues that differ between the ELD (Q486E, I499L and N496D) and KKR (E490K, I538K and H537R) heterodimer variants are colored green. c, Summary of screening results when the indicated substitutions were introduced into the left ZFN of the AAVS1 dimer and nucleases were then screened for activity at the intended target and four known off-target sites. The blue data points (left axis) indicate specificity as gauged by the ratio of activity at the on-target site (AAVS1) to the cumulative activity at the four characterized off-target sites. Gray bars (right axis) indicate on-target indel frequencies. Data are presented as the mean ± s.d. from four biological replicates. The specificity for the parent ZFN dimer (leftmost data point) is highlighted by a black arrow. Specificity values for the seven variants with the most promising behavior are bounded by a red box. Screening results with the right ZFN were broadly concordant (Supplementary Table 1). d, Structural relationship between seven residues of interest (shown in red) and the active site residues (shown in blue) in a molecular model of a ZFN dimer bound to DNA18. e, Activity and specificity of selected variants identified in FokI substitution screens of the AAVS1 ZFN dimer. Each variant was tested as a dimer in which both ZFN-L and ZFN-R bore the indicated substitution. Each indel frequency value represents the average from three biological replicates. The standard deviations for all samples were below 2%; standard deviations were below 0.1% for samples with mean values below 1%. The second column provides the indel frequency measured at the intended target, while the third column indicates the aggregate indel frequency measured across four previously known off-target sites. The fourth column lists the ratio of on- to off-target indels (that is, second column/third column); this ratio is not applicable (NA) for the GFP sample as there is no detectable on-target activity. To highlight relative signal intensities, table values are embedded in heat maps. Note that each variant was characterized in a larger mRNA titration series from which data on a single dose yielding levels of on-target modification comparable to, or greater than, the parental level were extracted for presentation here. The samples shown used the following mRNA amounts for transfection: 800 ng (parent, R416D, R416S, R418E and R422H); 400 ng (N476G, I479Q, Q481E, K525A, K525S and GFP); 200 ng (R416E, R416N, Q481A and K525T); and 100 ng (Q481D). For the full titration series and full off-target dataset, see Supplementary Table 9.

Building off these results, we next performed a more intensive scan of residues predicted to lie within 10 Å of the DNA interface, with substitutions biased toward residues also observed in FokI homologs to minimize the fraction of inactive variants (Supplementary Table 2). This study broadly confirmed the results of the first scan and identified substitutions at two additional positions that improved on-target cleavage preference by factors of 10- to 40-fold (S418D and I479T; Supplementary Table 3). Next, these two positions, plus the seven identified in the first scan, were comprehensively varied to each alternative amino acid. Screens of this test set identified diverse substitutions that improved on-target cleavage preference by a factor of up to 178 (Supplementary Tables 48). Critically, these studies were performed under conditions of modest activity for the parent ZFNs (<70% on-target indels) that moreover showed a clear dose response, ruling out the possibility that these substitutions served merely to weaken the affinity of a saturated system. Finally, 14 of the most encouraging variants were directly compared with the parent dimer for activity and specificity via replicate titration analysis (Supplementary Table 9). This study revealed 12 variants that retained full on-target activity, with off-target modification reduced by a factor of at least 100. For the 2 most effective ones (I479Q and Q481A), off-target cleavage was reduced by over 1,000-fold (Fig. 1e).

A single-residue substitution globally suppresses off-target cleavage

A key feature of our approach for improving ZFN specificity is its potential to yield global effects. To assess this issue, we submitted the two most specific variants (I479Q and Q481A) to genome-wide specificity analysis using an oligonucleotide capture method similar to GUIDE-seq46. Nucleases were delivered at high doses (>83% on-target indels) to improve the likelihood of detecting off-target signal. The results of this analysis suggested comprehensive suppression of off-target activity, as gauged by the substantial reduction in the number of candidate cleavage sites identified (725 for the parent dimer versus 10 and 12 for I479Q and Q481A, respectively) as well as the much larger fraction of capture events associated with the intended target (13% for the parent versus 98% for each variant; Fig. 2a). This was confirmed via follow-up indel analysis, which showed no significant activity by either variant at the respective candidate off-target loci (Fig. 2b). As a further test of specificity, we assessed all dimers for cleavage activity at the 100 most highly ranked loci of the parent ZFN capture list. For the Q481A dimer, this study revealed global and complete suppression of cleavage at all off-target sites (Fig. 2c), down to a median assay background signal of <0.01% (Supplementary Table 10). The I479Q dimer exhibited broadly similar behavior, albeit with <0.1% indels detected at a subset of sites (Fig. 2c). Notably, across this entire study we observed no instance of an off-target locus that was either uniquely or more highly modified by a variant ZFN than by the parent, confirming that these FokI mutations globally reduce off-target cleavage.

Fig. 2: Global suppression of off-target cleavage by the I479Q and Q481A FokI domain variants.

a, Summary of oligonucleotide duplex capture results with the parent and variant dimers. Each capture experiment was performed in biological quadruplicate. The proportion of on-target capture events is indicated by the largest blue wedge in each chart (13% for the parent and 98% for each variant). Remaining wedges represent capture signal seen at candidate off-target sites. Capture experiments were performed under conditions that yielded on-target indel levels of 74.6%, 84.5% and 83.6% for the parent, I479Q and Q481A dimers, respectively. b, Table view of oligonucleotide duplex capture results for the I479Q and Q481A variant ZFNs. The first and second columns provide locus coordinates, the third column lists total unique integrations from four biological replicates and the fourth and fifth columns contain indel levels observed in a follow-up study in K562 cells. Each indel value represents a single measurement. Loci are ranked in order of decreasing number of capture events, with the intended target at the top. The sixth column contains P values for the corresponding ZFN–GFP comparisons (see Supplementary Note 2 for details of the statistical test), revealing significant modification of only the intended target. c, Plot of the indel frequencies measured at the intended target and the 99 most active off-target loci for the parent AAVS1 ZFN pair expressed in human K562 cells (gray bars). Each bar represents a single measurement. These loci were also assessed for indel frequency in cells treated with the I479Q and Q481A variant ZFNs (orange and red bars, respectively). Indel values that are significant relative to a negative control are plotted as three-dimensional bars; non-significant indel values are plotted as flat rectangles. For assay values used to generate this plot, see Supplementary Table 10. d, Left, log-scale plot of indel frequencies at OT1 in cells treated with parent ZFNs, the Q481A dimer or a negative control, as measured using the oversampling-based indel assay. The OT1 values for the parent are the average of 12 technical replicates while the OT1 values for Q481A and the GFP negative control are the average of 48 technical replicates performed to sample sufficient genomes to achieve the desired sensitivity; values for individual data points are given in Supplementary Fig. 3. Error bars represent the 95% confidence intervals. Right, indel frequencies from single on-target measurements of the same samples.

While this genome-wide analysis highlighted the capacity of single-residue substitutions to comprehensively reduce off-target cleavage, it was not able to measure the full magnitude of this effect for the Q481A variant given its lack of significant indel signals for any off-target sites. To address this, we developed a more sensitive protocol for sequence-based indel analysis that involves oversampling each input allele by a factor of at least ten and running multiple technical replicates. Oversampling enables identification and removal of rare technical artifacts by virtue of their individual frequencies lying below the threshold expected from a multiply sampled input mutation. Running technical replicates enables common background artifacts to be more accurately subtracted via the use of a negative control. A titration study demonstrated good linearity over a 1,000-fold range of input indel levels (<2.5-fold variance from predicted indel signals; Supplementary Fig. 2). We then applied this protocol to the quantification of indel levels at OT1 in cells treated with the Q481A variant, using 1.44 million input haploid genomes. This allowed discernment of a significant indel signal at a level of 0.0024%, as compared to 5.53% in cells treated with the parent ZFNs (Fig. 2d and Supplementary Fig. 3). Factoring in the higher on-target signal in variant-treated cells (85.5% versus 66.3%), this result indicates that the single Q481A substitution improved discrimination against off-target cleavage by ~3,000-fold. Moreover, the ratio of on-target to OT1 signal exhibited by the Q481A variant in this study (85.5%/0.0024% = 35,600) approximates the binding preference seen in our original biochemical analyses of the parent (20,500-fold; Supplementary Fig. 1a), resolving the discrepancy that initially led to these studies.

Although the key observations to this point were consistent with our conceptual framework for these studies, they did not rule out other possible mechanisms for improved specificity such as an increase in on-target binding preference or a global reduction in ZFN–DNA affinity. To further constrain the mechanism, we investigated the properties of the Q481A variant in three related biochemical studies. In the first, a competitive ELISA was used to assess the parent and variant ZFNs for relative affinity for the on-target site and OT1. This revealed a <1.3-fold change in binding preference (Supplementary Fig. 4). In the second study, off-rates were gauged for each ZFN in complex with its target sequence as a proxy for comparing dissociation constants under the assumption of identical association rates47. Our study revealed dissociation rates for parent and variant ZFNs that were highly similar (<10% variance in half-life), providing a strong indication that the observed effects were not due to global weakening of affinity (Supplementary Fig. 5). Finally, we directly compared the cleavage kinetics of ZFNs bearing either the parental FokI domain or the Q481A variant and observed a >20-fold reduction in apparent k2 for the variant (Supplementary Fig. 6 and Supplementary Note 1). Together, these results provide further support for catalytic slowing as a mechanism of specificity improvement for our variants.

Finally, we sought to gain insight into the generality of FokI domain diversification as a means for reducing off-target cleavage. To accomplish this, key variants were tested in the context of a second previously published ZFN pair targeted to the gene encoding human PD-1 (PDCD1)48. Eleven variants were identified that when incorporated into both ZFNs of the pair yielded nucleases that retained full on-target activity (Supplementary Table 11), with on- to off-target ratios improved by up to 230-fold (Supplementary Table 12). This study demonstrates that our strategy for improving specificity via modification of the FokI domain could be applied to other ZFNs.

Improving specificity via removal of a zinc finger phosphate contact

Although experiments to this point had established engineering of the FokI domain as a simple and general means for improving specificity, a consideration of our conceptual framework (Supplementary Fig. 1c) suggested that this approach might not work for ZFNs for which the off-rate (k−1) is especially slow. Given the finite cellular half-life of ZFNs, it seemed possible that designs with especially slow off-rates might vacate both off- and on- target sites with similar kinetics via degradation processes, thus limiting the utility of a reduced k2 for enhancing specificity. A second consideration was that ZFNs exhibiting too much nonspecific binding energy might be sequestered by bulk genomic DNA, reducing availability for on-target binding and cleavage49. Given this, we sought to identify alternative substitutions that might reduce nonspecific DNA affinity50 and that could be combined with FokI variants for improving performance. As the focus for these studies, we investigated the arginine at position ‘−5’ of the zinc finger framework (Fig. 3a; numbering as described in ref. 51). This residue contacts the DNA backbone in each finger of the natural zinc finger protein (ZFP) from which most ZFNs are derived51, and alanine substitutions at this position have been explored for reducing the activity of engineered zinc finger transcription factors52. This position is also highly conserved as a basic residue within the family of DNA-binding ZFPs, suggesting functional relevance.

Fig. 3: Improving ZFN cleavage specificity via removal of a nonspecific DNA contact in the zinc finger repeat.

a, Diagram showing the location of arginine (−5), which makes a nonspecific contact to a phosphate. The structure shown is finger two of the zinc finger protein Zif268 (ref. 51). b, Sketch of the host ZFN pair for these studies, which targets a sequence within the human BCL11A gene at hg38:chr2: 60495250–60495290 (−). c, Effect of increasing the number of R(−5)Q-substituted fingers in the context of the five-finger protein BCL11A ZFN-R. The activities at the intended target (gray bars) and the cumulative activity at two previously known off-target sites (red bars) for the indicated BCL11A ZFN-R zinc finger variant and parent BCL11A ZFN-L co-expressed in human K562 cells are plotted. Each value is the average of three biological replicates, where error bars represent the standard deviation and dots represent values from individual replicates. For ZFN design information and plotted values, see Supplementary Tables 13 and 14.

Our studies proceeded as follows. First, starting with a ZFN dimer targeted to the erythroid enhancer of BCL11A (Fig. 3b; see also refs. 53,54), derivatives were generated for both the left and right ZFNs bearing varying numbers of glutamine substitutions of the arginine of interest. These derivatives spanned all substitution levels, up to full replacement of every finger with its R(−5)Q counterpart. Next, these ZFNs were screened for activity at the intended target and a previously characterized off-target locus. The results revealed divergent behavior for the two ZFNs. For the left ZFN, we observed a response that was consistent with a progressive reduction in nonspecific binding energy across the substitution series. This ZFN exhibited lower levels of both on-target and off-target cleavage in variants bearing a higher fraction of glutamine-substituted fingers, but with little evidence of increased specificity beyond what could be achieved by simply reducing dose (Supplementary Table 13 and Supplementary Fig. 7). In contrast, the right ZFN showed enhanced specificity with increasing fractions of glutamine-substituted fingers, with the fully converted derivative exhibiting a >100-fold reduction in off-target activity with no loss of on-target cleavage (Supplementary Fig. 7). This was confirmed via retesting this substitution series in triplicate and with analysis of an additional off-target site (Fig. 3c and Supplementary Table 14).

Finally, variants of the BCL11A right ZFN were characterized biochemically for dissociation rate from the on-target sequence. This study revealed a progressive rise in off-rate with increasing fraction of R(−5)Q substitutions. The increase in off-rate with each additional substitution was remarkably consistent, ranging from 1.6- to 3.4-fold, with fully substituted ZFNs dissociating 50-fold faster than the unmodified parents (Supplementary Fig. 8). These studies support the replacement of framework residues for improving cleavage specificity for at least a subset of ZFNs. Consistent with the possibility that the FokI mutations and zinc finger backbone mutations function through different mechanisms, an initial test of combining the two approaches demonstrated that R(−5)Q substitutions could be combined with FokI variants to further improve specificity (Supplementary Table 15). This test was performed in human CD34+ cells, demonstrating that both approaches could improve ZFN performance in therapeutically relevant contexts.

Complete knockout of a therapeutic target with no detectable off-target effects

Although our studies to this point had established the capacity of FokI and zinc finger substitutions to substantially improve ZFN specificity, they had not assessed performance at the limit of near-complete on-target cleavage. This regime is relevant to many editing applications, including in vivo and ex vivo therapies (for example, see refs. 7,8,9,10), where the need for high levels of biallelic or multiplexed editing can require highly efficient modification of each target locus. Conditions with a very high rate of modification also present a much more rigorous test of specificity, as the higher nuclease levels required to drive a desired editing event towards saturation can also yield very large increases in modification at off-target sites. This was illustrated by the parent ZFN dimer for our BCL11A studies, for which there was a >50-fold increase in cleavage of OT1 as the on-target indel levels increased from 59% to 85% (Supplementary Table 13).

Our studies focused on a ZFN pair that cleaves the TRAC locus (Fig. 4a), which is a knockout target for engineering tumor-targeted T cells. Although this pair had exhibited no detectable off-target activity when assayed at an on-target level of 81% indels55, when delivered at higher levels (~95% on-target indels) it manifested substantial activity at a candidate off-target site (Supplementary Table 16). We sought to further characterize and eliminate the off-target activity of this ZFN as follows. First, we generated seven variants of each ZFN, which differed from the parents in having a discrete FokI domain substitution and/or three R(−5)Q-substituted fingers (Fig. 4b). We then assessed the full matrix of 64 resultant dimers for activity and specificity in K562 cells and identified 13 that exhibited highly efficient modification of TRAC (≥94% indels) with complete abrogation of the off-target activity (Supplementary Table 16). These were then submitted to unbiased genome-wide oligonucleotide duplex capture analysis under conditions of very high on-target modification (88–96% indels), and the results were used to choose a subset of pairs for specificity assessment in human T cells (Fig. 4c and Supplementary Fig. 9). For each tested pair, this involved indel analysis at candidate off-target loci identified via capture analysis (Supplementary Table 17) as well as at verified off-target sites for the parent dimer (Supplementary Table 18). This identified two pairs (pairs 14 and 60) with global and complete suppression of cleavage at all off-target sites, down to a median assay background signal of 0.01%, with on-target indel levels of 98.2% and 97.0%, respectively (Fig. 4d). In a parallel assessment of on-target activity, flow cytometry analysis revealed a loss of TCR cell-surface expression in 98.5% and 97.3% of ZFN-treated T cells, respectively (Fig. 4e), confirming essentially complete modification of the intended locus with no detectable off-target cleavage.

Fig. 4: Development of ZFNs for highly efficient modification of TRAC in T cells with no detectable off-target effects.

a, Sketch of the parent ZFN dimer for this study, which targets a site in the human TRAC locus. b, Summary of the variant ZFNs generated for these studies. The top entry in each table corresponds to the indicated parent ZFN, while remaining entries describe variants bearing either the indicated FokI type and/or a panel of three R(−5)Q substitutions in the zinc finger DNA-binding domain. Substituted fingers were the first, third and fifth for ZFN-L and the first, third and fourth for ZFN-R. c, Summary of oligonucleotide duplex capture results with the parent and the indicated variant dimers. Capture experiments were performed in biological quadruplicate; the average indel frequency at the intended target is shown. The proportion of on-target capture events is indicated by the largest blue wedge in each chart. Remaining wedges represent capture signal seen at candidate off-target sites. For a summary of the results for all ZFN pairs tested, see Supplementary Fig. 9. d, Plot of indel frequencies measured at the intended target (TRAC) and at the 41 off-target loci yielding detectable modification by the parent ZFN pair in T cell studies (gray bars) and at the same loci in cells treated with the indicated variant pairs (colored bars). Values are plotted as in Fig. 2c with each bar representing a single measurement. For assay values used to generate this plot, see Supplementary Table 18. e, Flow cytometry data for pooled samples treated with ZFN variant pairs #14 and 60 and control T cells (empty trace), confirming the essentially complete knockout of TCR expression achieved by these ZFN pairs under experimental conditions that yielded no detectable off-target activity in human T cells. This experiment was performed on one aliquot of the same preparation of cells treated with pairs 14 and 60 that were processed for indel analysis in d. Red lines indicate the gates used to quantify the loss of CD3.


Although cleavage specificity has been a long-standing concern18, efforts to improve the target preference of gene editing nucleases have mostly focused on one aspect of the enzymatic process—target engagement. In this work, we have sought to address this issue in the context of the ZFN architecture by engineering the FokI domain, which mediates catalysis. Via large-scale screening, we have identified single residue substitutions that globally suppress off-target activity without affecting on-target affinity or binding preference. This behavior is exemplified by the Q481A variant, which suppresses modification by the AAVS1 ZFNs of all 94 quantifiable off-target sites to below the limits of detection (median assay background of <0.01%), with >1,000-fold reduction at the strongest off-target sites.

We have rationalized these studies within the context of the standard framework of enzyme function (Supplementary Fig. 1c). In particular, we interpreted the initial conundrum—discordant binding and cleavage preferences—as potentially being due to an imbalance between k2 and k−1. A supporting consideration was the chimeric nature of the ZFN structure12, which employs the FokI cleavage domain outside of its original evolutionary context. Given the faster dissociation rates reported for the parent FokI enzyme56 relative to five- or six-finger ZFNs57, it seemed possible that the FokI domain might cleave too quickly to fully discriminate on-target versus off-target sites in its ectopic context. Viewed in this light, our strategy may be seen as being designed to reduce k2 into a range that enables cleavage specificity to reflect binding preference, by allowing sufficient time before catalysis for dissociation from suboptimal targets (Supplementary Fig. 10). Consistent with this framework, Q481A (one of the most impactful substitutions) eliminates a contact to the active site residue D467 (Supplementary Fig. 11) and yields a >20-fold reduction in catalytic rate (Supplementary Note 1). Moreover, the increased cellular specificity of the variant nuclease resolves the discordance between the binding and cleavage preference of the AAVS1 ZFNs for the on-target site versus OT1, resulting in values that differ by less than a factor of two.

While a global reduction in catalytic rate provides a simple and intuitive framework for these studies, we note that this model neither uniquely nor completely explains all observations, and that different substitutions may enhance specificity via pathways that are mechanistically distinct. For example, an alternative possible consequence of the substitutions we have identified may be to enhance functional sensitivity to minor variations in docking configuration (as might occur upon binding a noncognate target). Such behavior would also reduce catalytic rate, albeit differentially for distinct sites. Another possible outcome could be to impart a sequence preference to the FokI domain itself, although this seems less likely given both the global effects of the substitutions and their locations (away from the floor of the major groove). It is also possible that some FokI substitutions act, at least in part, to reduce ZFN affinity for nonspecific DNA sequences, which could lower bulk sequestration by the genome49,50 and provide an explanation for the apparently increased activity manifested by a subset of variants. Distinguishing among these and other possibilities, which will help to further improve specificity, will be investigated in future studies.

A corollary of our proposed mechanistic framework is that it should preclude identification of a single optimal substitution for all applications, as each nuclease pair will present distinct dissociation rates for its on-target and off-target sites. The four most specific AAVS1 variants (S418E, I479Q, Q481A and Q481E) exemplify this point, enabling the best performance for the AAVS1 dimer (Fig. 1e and Supplementary Table 9) but yielding suboptimal activity in the context of the PDCD1 ZFNs (Supplementary Table 11). Accordingly, we anticipate that practical application of this strategy will involve screening of a small panel of variants (for example, those in Fig. 1e) to identify the best performers for any given ZFN dimer. As demonstrated with the BCL11A and TRAC reagents, this may involve each ZFN monomer bearing a different FokI substitution type.

An unexpected challenge of this work was our initial inability to fully quantify the specificity of the most effective FokI variants, owing to off-target indels occurring too rarely to measure. To address this, we developed a new strategy for discriminating rare assay artifacts from bona fide indels that can allow signals below 0.001% indels to be detected, which represents an improvement of at least 100-fold relative to the conventional version of the assay4,58. This allowed detection of indels induced by the Q481A variant at the OT1 off-target site. The key to this approach—oversampling of input genomes—may be implemented using the same methodologies and sequencing platforms already employed by numerous laboratories. Given this, we anticipate that this approach will prove generally useful for quantifying indels under conditions requiring especially high sensitivity, for example, the derisking of cancer-relevant loci in a therapeutic context.

In addition to varying the FokI sequence, our studies have established disruption of a nonspecific zinc finger–phosphate contact as a complementary approach for improving ZFN specificity. A key feature of this modification is that it offers bona fide tunability: increasing the fraction of R(−5)Q-substituted fingers yields a monotonic progression of on-target and off-target performance (Supplementary Fig. 7 and Supplementary Table 13) as well as dissociation rates (Supplementary Fig. 8). Notably, when applied to our BCL11A ZFNs, this substitution yielded divergent behavior between the left and right monomers. While the latter demonstrated improved specificity with an increasing fraction of substituted fingers, the former showed coordinated reductions in on-target and off-target cleavage that were indistinguishable from a lowering of the dose. These results are consistent with a model in which the ZFNs exhibit mismatched on-target affinities, with the right ZFN saturated in our cellular studies. They point to the potential utility of these substitutions for optimizing specificity via balancing of the dose responsiveness of a ZFN dimer pair. We also note that these mutations should be broadly useful for improving the cellular specificity of ZFP fusion proteins beyond ZFNs, for example, designed ZFP transcription factors59.

Given the potency and mechanistic complementarity of our FokI domain and zinc finger substitutions, it seemed possible that they might be combined to yield ZFNs that could achieve complete knockout of a target locus with no detectable off-target cleavage. Such extremely active and specific behavior is of particular interest for therapeutic applications, where efficacy often requires highly efficient modification. Accordingly, our final studies focused on development of the TRAC ZFNs, which yielded >98% on-target indels with no detectable off-target activity (down to a median assay background level of 0.01%), in a therapeutically relevant context.

Finally, we note that although these studies were performed in the context of ZFNs, the FokI domain has been broadly applied to impart cleavage properties onto diverse DNA-binding domains including TALEs11 and CRISPR–Cas9 (refs. 21,22), and we anticipate that our FokI variants will prove broadly useful for improving specificity in these systems. More generally, having incompatible kinetics when fusing targeting and catalytic domains from different contexts is likely to be a common problem across the full scope of protein engineering activities. As such, we envision that the principles demonstrated in this study will find broad utility in optimizing the performance of other chimeric systems.


ZFN constructs

The sequence of each parent ZFN construct is given in Supplementary Table 19. FokI variants for the AAVS1 ZFNs were generated using the QuikChange XL kit (Agilent) according to the manufacturer’s instructions and using the starting AAVS1 ZFN constructs as a template. Primer sequences used for the QuikChange procedure can be found in Supplementary Table 20. FokI variants for the PDCD1 constructs were generated by standard subcloning with BamHI and XhoI to combine FokI variants from the AAVS1 FokI variant constructs and the zinc finger domains of the starting PDCD1 ZFN constructs. The ZFP backbone and FokI variants for the BCL11A and TRAC ZFNs were generated by reassembling zinc finger arrays from modules as previously described55 and cloning the resulting zinc finger array into a vector containing the desired FokI variant. All constructs were verified by Sanger sequencing with two sequencing primers to yield overlapping sequence reads of the entire ZFN coding sequence.

mRNA production

mRNA was transcribed from a PCR template using the 5× mMessage mMachine T7 ULTRA kit (Invitrogen) according to the manufacturer’s instructions. The PCR template was generated using AccuPrime Pfx DNA Polymerase (Invitrogen). Primers were used at a final concentration of 0.4 µM with the following thermocycling conditions: initial melt of 95 °C for 3 min; 30 cycles of 95 °C for 30 s, 63.3 °C for 30 s and 68 °C for 2 min; and a final extension at 68 °C for 3 min. PCR products were purified using the QIAquick PCR purification kit (Qiagen). mRNA was purified using the RNeasy kit (Qiagen). mRNA concentration was determined using the Quant-iT RNA Assay kit (Invitrogen). Primer sequences for the two primers (N80pt and R560) can be found in Supplementary Table 20.

Gene editing in human K562 cells

K562 (ATCC, CCL243) cells were obtained from ATCC and were maintained in RPMI1640 with 10% FBS and 1× penicillin–streptomycin–glutamine (PSG) (Gibco, 10378-016) at 37 °C with 5% CO2. Various doses of mRNA encoding green fluorescent protein (GFP) or paired ZFN were electroporated into K562 cells using the SF cell line 96-well Nucleofector kit (Lonza, V4SC-2960) following the manufacturer’s instructions. In brief, cells were washed twice with 1× PBS (divalent cation-free) and resuspended at 2 × 105 cells per 15 µl of supplemented SF cell line 96-well Nucleofector solution. For each transfection, 15 µl of the cell suspension was mixed with 5 µl of mRNA and transferred to the Lonza Nucleocuvette plate and electroporated using the protocol for K562 cells (Nucleofector program 96-FF-120) on an Amaxa Nucleofector 96-well Shuttle System (Lonza). Electroporated cells were incubated at room temperature for 10 min and then transferred to 150 µl of prewarmed complete medium in a 96-well tissue culture plate. Cells were incubated for 72 h and then harvested for indel quantification.

Gene editing in human CD34+ hematopoietic stem and progenitor cells

CD34+ stem and progenitor cells were purified from mobilized peripheral blood leukapheresis products obtained from AllCells. Platelet depletion was performed, CD34+ cells were enriched using the Miltenyi Biotec CliniMACS Plus instrument (Miltenyi Biotec) and aliquots were frozen.

For each transfection, aliquots were thawed in a 37 °C water bath, added to prewarmed X-Vivo medium with PSG and then centrifuged at 300g for 15 min. Then, cells were resuspended in C-Vivo medium with 1× CC100 and PSG at a density of 1 × 106 cells ml−1 and incubated at 37°C and 5% CO2 for 48 h. On the day of transfection, cells were centrifuged at 300g for 15 min and resuspended gently in BTXpress electroporation solution to obtain 2 × 105 cells in a 100-µl volume for each sample. Cell suspension was mixed with the appropriate amount of mRNA and electroporated with a BTX 96-well plate instrument using the following pulse conditions: mode = LV; voltage = 250 V; and P.Length = 4 ms. Cells were transferred to a 48-well plate with 400 µl of warmed X-Vivo medium with CC100 and PSG per well, incubated for 72 h at 37 °C and 5% CO2 and then collected. Genomic DNA was extracted using Quickextract and characterized using the MiSeq indel assay.

Gene editing in human T cells

Negatively-selected, cryopreserved peripheral blood CD3+ Pan T cells (AllCells, PB009-1F) were revived and activated for 72 h at 1 × 106 cells per milliliter in RPMI1640 with 10% human serum (Valley Biomedical, HS1017), Dynabeads human T-activator CD3/CD28 (cells to beads ratio of 1:3; Life Technologies, 11131D) and IL-2 (100 IU ml−1; Thermo Fisher, CTP0023) without antibiotics. Various doses of paired ZFN mRNA were transfected into 0.2 × 106 activated T cells using the BTX transfection system (mode = LV; interval = 100 ms; polarity = unipolar) in 100 µl of BTXpress high performance electroporation solution (BTX, 45-0805) in BTX 96-well disposable electroporation plate (2-mm gap, 125 µl, one plate; BTX, 45-0450-M). Electroporated cells were transferred to 100 µl of complete medium in a 96-well tissue culture plate and incubated at 30 °C and 5% CO2 for 24 h. Cells were then moved to 37 °C with 5% CO2 and incubated for 24 h. Cells were mixed with a pipette and incubated at 37 °C and 5% CO2 for another 48 h. Cells were then collected for flow cytometry and indel quantification.

Flow cytometry

Between half a million and a million T cells were collected by centrifugation and washed twice with divalent-cation-free 1× PBS at room temperature. Cell pellets were then resuspended in 100 µl of 1× PBS containing 1% BSA, 5 µl of APC mouse anti-human CD3 antibody (BD Biosciences, 340440) and 1 µl of fixable viability dye eFluor 780 (diluted between 1 and 20 times; Thermo Fisher, 65-0865-14). This suspension was incubated for 30 min at room temperature in the dark. After the incubation, cells were washed with divalent-cation-free 1× PBS twice at room temperature and then resuspended in 200 µl of 1× PBS. Flow cytometry analysis was performed using an Attune NxT acoustic focusing cytometer (Thermo Fisher), and data were analyzed using FlowJo (v.10.4, FlowJo). See Supplementary Note 2 for details on the gating strategy.

Oligonucleotide capture assay

Our oligonucleotide capture assay was adapted from the GUIDE-seq protocol of Tsai et al.46, with modifications designed to enhance robustness and sensitivity for analysis of ZFNs. A key adaptation involved the addition of randomized 4-base 5′ overhangs to the donor duplex to provide a closer match with the ends resulting from ZFN cleavage. This yielded a threefold improvement in capture efficiency relative to the blunt ends used by Tsai et al. (Supplementary Fig. 12, compare 0 bp overhang with 5′ overhang of 4 bp), that was preserved even when the duplex and cleavage target exhibited substantial overhang length mismatch (Supplementary Fig. 12, compare 5′ overhangs of 1–5 bp; see also Supplementary Figs. 1315). This provides assurance that our assay efficiently recovers all ZFN cleavage events including those yielding alternative overhang lengths (for example, 3 or 5 base pairs; see refs. 50,62). We also assessed the overhangs yielded by FokI variants of the AAVS1 and TRAC ZFNs by analyzing indel data and observed no evidence for altered cleavage patterns (Supplementary Figs. 14 and 15). This provided assurance that the capture assay would be equally effective at identifying candidate off-target sites for parental ZFNs as well as their FokI variant counterparts.

The donor duplex sequence was as follows: 5′-P-N*N*NNGTTTAATTGAGTTGTCATATGTTAATAACGGT*A*T-3′ and 5′-P-N*N*NNATACCGTTATTAACATATGACAACTCAATTAA*A*C-3′ where P represents 5′ phosphorylation and an asterisk indicates a phosphorothioate linkage. Oligonucleotides were ordered from IDT using the following codes: /5Phos/(25252525)*(25252525)*(25252525)(25252525)GTTTAATTGAGTTGTCATATGTTAATAACGGT*A*T and /5Phos/(25252525)*(25252525)*(25252525)(25252525)ATACCGTTATTAACATATGACAACTCAATTAA*A*C.

The protocol was otherwise essentially as previously described except for the following points: (i) the experiment was performed in biological quadruplicate; (ii) only a single set of duplex-specific primers were used, with sequences obtained in the minus direction; (iii) sequencing data were processed without the use of filters for the intended target; (iv) individual reads were required to map to a region of hg38 that was unambiguous, non-repetitive (fewer than three loci) and >10 kb from the intended target; and (v) candidate targets were defined by clusters of unique integrations lying within 100 base pairs of each other that also contained at least fivefold more events than the same locus in untreated control cells.

Standard indel analysis

PCR primers for loci of interest were designed using Primer3 with the following optimal conditions: amplicon size of 200 nucleotides; a melting temperature of 60 °C; primer length of 20 nucleotides; and GC content of 50%. Adaptors were added for a second PCR reaction to add the Illumina library sequences (forward primer: ACACGACGCTCTTCCGATCT; reverse primer: GACGTGTGCTCTTCCGAT). See Supplementary Table 20 for a list of primer sequences. Regions of interest were amplified in 25 µl using 100 ng of genomic DNA with AccuPrime HiFi (Invitrogen). Primers were used at a final concentration of 0.1 µM with the following thermocycling conditions: initial melt of 95 °C for 5 min; 35 cycles of 95 °C for 30 s, 55 °C for 30 s and 68 °C for 40 s; and a final extension at 68 °C for 10 min. PCR products were diluted 1:20 in water. Two microliters of diluted PCR product was used in a 20-µl PCR reaction to add the Illumina library sequences with Phusion High-Fidelity PCR MasterMix with HF Buffer (NEB). Primers were used at a final concentration of 0.5 µM with the following conditions: initial melt of 98 °C for 30 s; 12 cycles of 98 °C for 10 s, 60 °C for 30 s and 72 °C for 40 s; and a final extension at 72 °C for 10 min. A second PCR reaction was then performed to add sample specific sequence barcodes. These barcode primers are essentially as previously described14. PCR libraries were purified using the QIAquick PCR purification kit (Qiagen). Samples were quantified with the Qubit dsDNA HS Assay kit (Invitrogen) and diluted to 2 nM. The libraries were then run according to the manufacturer’s instructions on either an Illumina MiSeq using a standard 300-cycle kit or an Illumina NextSeq 500 using a mid-output 300-cycle kit. See Supplementary Note 2 for details on how the data were processed.

NextSeq high-sensitivity oversampling indel analysis

PCR conditions and primer sequences were similar to those in the standard indel analysis except that genomic DNA was purified and quantified using the Qubit dsDNA assay kit and 100 ng of this purified DNA was used per PCR reaction. Multiple technical replicates (each with 100 ng of genomic DNA) were performed for each genomic DNA sample. Lower cluster densities (ideally fewer than 120,000 clusters per mm2) and higher amounts of PhiX control DNA (30–50%) were also required to obtain high-quality data, especially when only a single amplicon was used for an entire NextSeq run. Sequence reads per technical replicate were also much higher than in the standard indel assay, ideally >300,000 reads per technical replicate. Either a 300-cycle mid-output kit or a 300-cycle high-output kit was used depending on the number of technical replicates loaded on a given sequencing run. See Supplementary Note 2 and Supplementary Table 21 for details on how the data were processed.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability

Illumina sequencing data underlying all key experiments have been deposited in the NCBI Sequence Read Archive under accession code PRJNA540312.

Code availability

Custom computer scripts used to perform the standard indel analysis and increased-sensitivity indel analysis can be found in Supplementary Note 2. Custom computer scripts used to automate more standard portions of the data analysis pipeline are available upon request.


  1. 1.

    Urnov, F. D. et al. Genome editing with engineered zinc finger nucleases. Nat. Rev. Genet. 11, 636–646 (2010).

    CAS  Article  Google Scholar 

  2. 2.

    Chen, J. S. et al. Enhanced proofreading governs CRISPR–Cas9 targeting accuracy. Nature 550, 407–410 (2017).

    CAS  Article  Google Scholar 

  3. 3.

    Hu, J. H. et al. Evolved Cas9 variants with broad PAM compatibility and high DNA specificity. Nature 556, 57–63 (2018).

    CAS  Article  Google Scholar 

  4. 4.

    Tsai, S. Q. & Joung, J. K. Defining and improving the genome-wide specificities of CRISPR–Cas9 nucleases. Nat. Rev. Genet. 17, 300–312 (2016).

    CAS  Article  Google Scholar 

  5. 5.

    Hacein-Bey-Abina, S. et al. LMO2-associated clonal T cell proliferation in two patients after gene therapy for SCID-X1. Science 302, 415–419 (2003).

    CAS  Article  Google Scholar 

  6. 6.

    Donsante, A. et al. AAV vector integration sites in mouse hepatocellular carcinoma. Science 317, 477 (2007).

    CAS  Article  Google Scholar 

  7. 7.

    Ran, F. A. et al. In vivo genome editing using Staphylococcus aureus Cas9. Nature 520, 186–191 (2015).

    CAS  Article  Google Scholar 

  8. 8.

    Gao, X. et al. Treatment of autosomal dominant hearing loss by in vivo delivery of genome editing agents. Nature 553, 217–221 (2018).

    CAS  Article  Google Scholar 

  9. 9.

    Huang, C. H., Lee, K. C. & Doudna, J. A. Applications of CRISPR–Cas enzymes in cancer therapeutics and detection. Trends Cancer 4, 499–512 (2018).

    CAS  Article  Google Scholar 

  10. 10.

    DiCarlo, J. E., Mahajan, V. B. & Tsang, S. H. Gene therapy and genome surgery in the retina. J. Clin. Invest. 128, 2177–2188 (2018).

    Article  Google Scholar 

  11. 11.

    Miller, J. C. et al. A TALE nuclease architecture for efficient genome editing. Nat. Biotechnol. 29, 143–148 (2011).

    CAS  Article  Google Scholar 

  12. 12.

    Kim, Y. G., Cha, J. & Chandrasegaran, S. Hybrid restriction enzymes: zinc finger fusions to Fok I cleavage domain. Proc. Natl Acad. Sci. USA 93, 1156–1160 (1996).

    CAS  Article  Google Scholar 

  13. 13.

    Grizot, S. et al. Efficient targeting of a SCID gene by an engineered single-chain homing endonuclease. Nucleic Acids Res. 37, 5405–5419 (2009).

    CAS  Article  Google Scholar 

  14. 14.

    Miller, J. C. et al. Improved specificity of TALE-based genome editing using an expanded RVD repertoire. Nat. Methods 12, 465–471 (2015).

    CAS  Article  Google Scholar 

  15. 15.

    Rebar, E. J. & Pabo, C. O. Zinc finger phage: affinity selection of fingers with new DNA-binding specificities. Science 263, 671–673 (1994).

    CAS  Article  Google Scholar 

  16. 16.

    Jarjour, J. et al. High-resolution profiling of homing endonuclease binding and catalytic specificity using yeast surface display. Nucleic Acids Res. 37, 6871–6880 (2009).

    CAS  Article  Google Scholar 

  17. 17.

    Oakes, B. L. et al. Multi-reporter selection for the design of active and more specific zinc-finger nucleases for genome editing. Nat. Commun. 7, 10194 (2016).

    CAS  Article  Google Scholar 

  18. 18.

    Miller, J. C. et al. An improved zinc-finger nuclease architecture for highly specific genome editing. Nat. Biotechnol. 25, 778–785 (2007).

    CAS  Article  Google Scholar 

  19. 19.

    Guilinger, J. P. et al. Broad specificity profiling of TALENs results in engineered nucleases with improved DNA-cleavage specificity. Nat. Methods 11, 429–435 (2014).

    CAS  Article  Google Scholar 

  20. 20.

    Ran, F. A. et al. Double nicking by RNA-guided CRISPR Cas9 for enhanced genome editing specificity. Cell 154, 1380–1389 (2013).

    CAS  Article  Google Scholar 

  21. 21.

    Tsai, S. Q. et al. Dimeric CRISPR RNA-guided FokI nucleases for highly specific genome editing. Nat. Biotechnol. 32, 569–576 (2014).

    CAS  Article  Google Scholar 

  22. 22.

    Guilinger, J. P., Thompson, D. B. & Liu, D. R. Fusion of catalytically inactive Cas9 to FokI nuclease improves the specificity of genome modification. Nat. Biotechnol. 32, 577–582 (2014).

    CAS  Article  Google Scholar 

  23. 23.

    Bolukbasi, M. F. et al. DNA-binding-domain fusions enhance the targeting range and precision of Cas9. Nat. Methods 12, 1150–1156 (2015).

    CAS  Article  Google Scholar 

  24. 24.

    Kleinstiver, B. P. et al. Engineered CRISPR–Cas9 nucleases with altered PAM specificities. Nature 523, 481–485 (2015).

    Article  Google Scholar 

  25. 25.

    Kleinstiver, B. P. et al. Broadening the targeting range of Staphylococcus aureus CRISPR–Cas9 by modifying PAM recognition. Nat. Biotechnol. 33, 1293–1298 (2015).

    CAS  Article  Google Scholar 

  26. 26.

    Gao, L. et al. Engineered Cpf1 variants with altered PAM specificities. Nat. Biotechnol. 35, 789–792 (2017).

    CAS  Article  Google Scholar 

  27. 27.

    Yin, H. et al. Partial DNA-guided Cas9 enables genome editing with reduced off-target activity. Nat. Chem. Biol. 14, 311–316 (2018).

    CAS  Article  Google Scholar 

  28. 28.

    Ryan, D. E. et al. Improving CRISPR–Cas specificity with chemical modifications in single-guide RNAs. Nucleic Acids Res. 46, 792–803 (2018).

    CAS  Article  Google Scholar 

  29. 29.

    Kleinstiver, B. P. et al. High-fidelity CRISPR–Cas9 nucleases with no detectable genome-wide off-target effects. Nature 529, 490–495 (2016).

    CAS  Article  Google Scholar 

  30. 30.

    Slaymaker, I. M. et al. Rationally engineered Cas9 nucleases with improved specificity. Science 351, 84–88 (2016).

    CAS  Article  Google Scholar 

  31. 31.

    Kulcsár, P. I. et al. Crossing enhanced and high fidelity SpCas9 nucleases to optimize specificity and cleavage. Genome Biol. 18, 190 (2017).

    Article  Google Scholar 

  32. 32.

    DeWitt, M. A. et al. Selection-free genome editing of the sickle mutation in human adult hematopoietic stem/progenitor cells. Sci. Transl. Med. 8, 360ra134 (2016).

    Article  Google Scholar 

  33. 33.

    Zhang, D. et al. Perfectly matched 20-nucleotide guide RNA sequences enable robust genome editing using high-fidelity SpCas9 nucleases. Genome Biol. 18, 191 (2017).

    Article  Google Scholar 

  34. 34.

    Vakulskas, C. A. et al. A high-fidelity Cas9 mutant delivered as a ribonucleoprotein complex enables efficient gene editing in human hematopoietic stem and progenitor cells. Nat. Med. 24, 1216–1224 (2018).

    CAS  Article  Google Scholar 

  35. 35.

    Casini, A. et al. A highly specific SpCas9 variant is identified by in vivo screening in yeast. Nat. Biotechnol. 36, 265–271 (2018).

    CAS  Article  Google Scholar 

  36. 36.

    Sternberg, S. H., LaFrance, B., Kaplan, M. & Doudna, J. A. Conformational control of DNA target cleavage by CRISPR–Cas9. Nature 527, 110–113 (2015).

    CAS  Article  Google Scholar 

  37. 37.

    Dagdas, Y. S., Chen, J. S., Sternberg, S. H., Doudna, J. A. & Yildiz, A. A conformational checkpoint between DNA binding and cleavage by CRISPR–Cas9. Sci. Adv. 3, eaao0027 (2017).

    Article  Google Scholar 

  38. 38.

    Raper, A. T., Stephenson, A. A. & Suo, Z. Functional insights revealed by the kinetic mechanism of CRISPR/Cas9. J. Am. Chem. Soc. 140, 2971–2984 (2018).

    CAS  Article  Google Scholar 

  39. 39.

    Singh, D., Sternberg, S. H., Fei, J., Doudna, J. A. & Ha, T. Real-time observation of DNA recognition and rejection by the RNA-guided endonuclease Cas9. Nat. Commun. 7, 12778 (2016).

    CAS  Article  Google Scholar 

  40. 40.

    Jiang, F. et al. Structures of a CRISPR–Cas9 R-loop complex primed for DNA cleavage. Science 351, 867–871 (2016).

    CAS  Article  Google Scholar 

  41. 41.

    Jiang, F. & Doudna, J. A. CRISPR–Cas9 structures and mechanisms. Annu. Rev. Biophys. 46, 505–529 (2017).

    CAS  Article  Google Scholar 

  42. 42.

    De Ravin, S. S. et al. Targeted gene addition in human CD34+ hematopoietic cells for correction of X-linked chronic granulomatous disease. Nat. Biotechnol. 34, 424–429 (2016).

    Article  Google Scholar 

  43. 43.

    Bisaria, N., Jarmoskaite, I. & Herschlag, D. Lessons from enzyme kinetics reveal specificity principles for RNA-guided nucleases in RNA interference and CRISPR-based genome editing. Cell Syst. 4, 21–29 (2017).

    CAS  Article  Google Scholar 

  44. 44.

    Kaczorowski, T., Skowron, P. & Podhajska, A. J. Purification and characterization of the FokI restriction endonuclease. Gene 80, 209–216 (1989).

    CAS  Article  Google Scholar 

  45. 45.

    Li, L., Wu, L. P., Clarke, R. & Chandrasegaran, S. C-terminal deletion mutants of the FokI restriction endonuclease. Gene 133, 79–84 (1993).

    CAS  Article  Google Scholar 

  46. 46.

    Tsai, S. Q. et al. GUIDE-seq enables genome-wide profiling of off-target cleavage by CRISPR–Cas nucleases. Nat. Biotechnol. 33, 187–197 (2015).

    CAS  Article  Google Scholar 

  47. 47.

    Kim, J. G., Takeda, Y., Matthews, B. W. & Anderson, W. F. Kinetic studies on Cro repressor–operator DNA interaction. J. Mol. Biol. 196, 149–158 (1987).

    CAS  Article  Google Scholar 

  48. 48.

    Beane, J. D. et al. Clinical scale zinc finger nuclease-mediated gene editing of PD-1 in tumor infiltrating lymphocytes for the treatment of metastatic melanoma. Mol. Ther. 23, 1380–1390 (2015).

    CAS  Article  Google Scholar 

  49. 49.

    Jen-Jacobson, L. Protein–DNA recognition complexes: conservation of structure and binding energy in the transition state. Biopolymers 44, 153–180 (1997).

    CAS  Article  Google Scholar 

  50. 50.

    Pattanayak, V., Ramirez, C. L., Joung, J. K. & Liu, D. R. Revealing off-target cleavage specificities of zinc-finger nucleases by in vitro selection. Nat. Methods 8, 765–770 (2011).

    CAS  Article  Google Scholar 

  51. 51.

    Pavletich, N. P. & Pabo, C. O. Zinc finger–DNA recognition: crystal structure of a Zif268–DNA complex at 2.1 Å. Science 252, 809–817 (1991).

    CAS  Article  Google Scholar 

  52. 52.

    Khalil, A. S. et al. A synthetic biology framework for programming eukaryotic transcription functions. Cell 150, 647–658 (2012).

    CAS  Article  Google Scholar 

  53. 53.

    Bauer, D. E. et al. An erythroid enhancer of BCL11A subject to genetic variation determines fetal hemoglobin level. Science 342, 253–257 (2013).

    CAS  Article  Google Scholar 

  54. 54.

    Vierstra, J. et al. Functional footprinting of regulatory DNA. Nat. Methods 12, 927–930 (2015).

    CAS  Article  Google Scholar 

  55. 55.

    Paschon, D. E. et al. Diversifying the structure of zinc finger nucleases for high precision editing. Nat. Commun. 10, 1133 (2019).

    Article  Google Scholar 

  56. 56.

    Pernstich, C. & Halford, S. E. Illuminating the reaction pathway of the FokI restriction endonuclease by fluorescence resonance energy transfer. Nucleic Acids Res. 40, 1203–1213 (2012).

    CAS  Article  Google Scholar 

  57. 57.

    Kim, J. S. & Pabo, C. O. Getting a handhold on DNA: design of poly-zinc finger proteins with femtomolar dissociation constants. Proc. Natl Acad. Sci. USA 95, 2812–2817 (1998).

    CAS  Article  Google Scholar 

  58. 58.

    Akcakaya, P. et al. In vivo CRISPR editing with no detectable genome-wide off-target mutations. Nature 561, 416–419 (2018).

    CAS  Article  Google Scholar 

  59. 59.

    Liu, P. et al. Regulation of an endogenous locus using a panel of designed zinc finger proteins targeted to accessible chromatin regions. Activation of vascular endothelial growth factor A. J. Biol. Chem. 276, 11323–11334 (2001).

    CAS  Article  Google Scholar 

  60. 60.

    Doyon, Y. et al. Enhancing zinc-finger-nuclease activity with improved obligate heterodimeric architectures. Nat. Methods 8, 74–79 (2011).

    CAS  Article  Google Scholar 

  61. 61.

    Wah, D. A., Hirsch, J. A., Dorner, L. F., Schildkraut, I. & Aggarwal, A. K. Structure of the multimodular endonuclease FokI bound to DNA. Nature 388, 97–100 (1997).

    CAS  Article  Google Scholar 

  62. 62.

    Orlando, S. et al. Zinc-finger nuclease-driven targeted integration into mammalian genomes using donors with limited chromosomal homology. Nucleic Acids Res. 38, e152 (2010).

    Article  Google Scholar 

Download references


We thank M. Lal for performing preliminary experiments.

Author information




J.C.M. and E.J.R. designed experiments, supervised experiments, analyzed data and wrote the paper. F.F., D.P.P. and P.L. designed experiments, performed experiments and analyzed data. A.R., D.E.P. and L.Z. designed experiments, supervised experiments and analyzed data. G.L. designed experiments. D.A.S. and Y.R.B. wrote custom computer code. S.J.H. supervised experiments. C.B.P., D.F.X., H.W.R., N.A.S., S.C.L., T.W. and Y.Z. performed experiments.

Corresponding author

Correspondence to Edward J. Rebar.

Ethics declarations

Competing interests

All authors are full-time employees of Sangamo Therapeutics.

Additional information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary Information

Supplementary Figs. 1–15, Supplementary Tables 1–20 and Supplementary Notes 1 and 2

Reporting Summary

Supplementary Table 21

Detailed information for high-sensitivity OT1 indel assay results for K562 cells treated with Q481A AAVS1 ZFNs or GFP

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Miller, J.C., Patil, D.P., Xia, D.F. et al. Enhancing gene editing specificity by attenuating DNA cleavage kinetics. Nat Biotechnol 37, 945–952 (2019). https://doi.org/10.1038/s41587-019-0186-z

Download citation

Further reading


Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing