Development of hRad51–Cas9 nickase fusions that mediate HDR without double-stranded breaks
Development of hRad51-Cas9 nickase fusions that mediate HDR without double-stranded breaks
Holly A. Rees 0 1 2
Wei-Hsi Yeh 0 1 2
David R. Liu
0 Merkin Institute of Transformative Technologies in Healthcare, Broad Institute of Harvard and MIT , Cambridge, MA 02142 , USA
1 Howard Hughes Medical Institute, Harvard University , Cambridge, MA 02142 , USA
2 Department of Chemistry and Chemical Biology, Harvard University , Cambridge, MA 02138 , USA
In mammalian cells, double-stranded DNA breaks (DSBs) are preferentially repaired through end-joining processes that generally lead to mixtures of insertions and deletions (indels) or other rearrangements at the cleavage site. In the presence of homologous DNA, homologydirected repair (HDR) can generate specific mutations, albeit typically with modest efficiency and a low ratio of HDR products:indels. Here, we develop hRad51 mutants fused to Cas9 (D10A) nickase (RDN) that mediate HDR while minimizing indels. We use RDN to install disease-associated point mutations in HEK293T cells with comparable or better efficiency than Cas9 nuclease and a 2.7-to-53-fold higher ratio of desired HDR product:undesired byproducts. Across five different human cell types, RDN variants generally result in higher HDR:indel ratios and lower off-target activity than Cas9 nuclease, although HDR efficiencies remain strongly site- and cell type-dependent. RDN variants provide precision editing options in cell types amenable to HDR, especially when byproducts of DSBs must be minimized.
W disruption by generating insertions and deletions
idely used genome editing strategies include gene
(indels) at a targeted locus following a
doublestranded DNA break (DSB)1, homology-directed repair (HDR)
following a targeted DSB2, and base editing, which enables the
precise installation of transition point mutations (C to T, G to A,
A to G, or T to C) without creating DSBs3?5. Among these three
strategies, HDR offers access to the broadest possible range of
changes to genomic DNA in mammalian cells (Fig. 1a)6. The use
of single-stranded DNA oligonucleotides containing
PAMblocking mutations as donor templates can improve HDR
outcomes by preventing re-cutting of the target site after
successful HDR (Fig. 1a)7. Nevertheless, because HDR is usually
initiated by a DSB, HDR is accompanied by undesired cellular
side-effects including high levels of indel formation7,8, DNA
translocations9, large deletions10, and p53 activation11,12.
We sought to improve ratios of desired:undesired HDR
products by exploring the initiation of HDR from a DNA nick rather
than a DSB. In contrast to DSBs, DNA nicks generally do not
induce undesired genome modification13?15, a principle exploited
by base editors to minimize editing byproducts3,5,16. Mutating
catalytic residues in programmable nucleases can result in
C and HDR
HEK site 2
programmable nickases that cleave only one of the two strands of
DNA at the target locus17?20. Although single nicks can lead to
more favorable HDR:indel ratios than double-stranded DNA
breaks19,21,22, nicks usually lead to much lower frequencies of
genome editing when compared to DSBs (typically 5?20-fold)23,
making nickases substantially less useful than nucleases as
genome editing tools17,20,22,24.
In this study we achieve DSB-free HDR with minimal
byproducts and reduced off-target editing by fusing hRad51 variants to
a programmable nickase to generate hRad51?Cas9 (D10A)
nickase fusions (RDN variants). We chose hRad51 due to its
known involvement in the repair of nicked DNA17,22. RDN is
capable of stimulating HDR at a DNA nick, resulting in a much
higher ratio of HDR product:indel formation in human cells (up
to 53-fold at the eight genomic loci tested here), substantially
lower off-target editing. A known mutant of hRad51 that cannot
bind BRCA225,26 can be used in RDN to further increase the
HDR:indel ratio. A second known hRad51 mutant that cannot
self-associate25,26 increases overall HDR efficiency while slightly
lowering HDR:indel ratios. RDN-mediated HDR is a one-step
procedure that does not require inclusion of PAM-blocking
mutations7 and can use readily synthesized 100-mer single
stranded DNA (ssDNA) oligonucleotides as donor templates.
Although RDN remains limited by its dependence on cellular
DNA repair processes underlying HDR, RDN may be useful for
applications that require precise genome edits not accessible to
base editing while minimizing undesired consequences of DSBs.
Indels caused by single Cas9 nickases. Cas9 contains two
independent nuclease domains, either of which can be disabled to
generate a nickase that selectively cleaves either the guide
RNApaired strand (Cas9(D10A) nickase) or the opposite strand (Cas9
(H840A) nickase) (Fig. 1b)27. We used high-throughput DNA
sequencing (HTS) to systematically compare the editing
outcomes of Cas9, Cas9(D10A), or Cas9(H840A) nickases at eight
genomic loci in three human cell lines.
While both nickases resulted in substantially fewer indels than
intact Cas9, nick-induced indel formation was highly
stranddependent and locus-dependent (Fig. 1c). The Cas9(D10A) and
Cas9(H840A) nickases displayed different relative activities when
paired with different sgRNAs; for example, at HEK site 2 the Cas9
(H840A) nickase generated 24 ? 5% indels (? values represent
standard deviations for three biological replicates) and the Cas9
(D10A) nickase generated only 1.1 ? 0.2% indels, while at HEK
site 3 Cas9(H840A) nickase resulted in only 0.73 ? 0.38% indels
but Cas9(D10A) nickase treatment generated 7.9 ? 1.4% indels
(Fig. 1c). One of the eight sgRNAs we tested, targeted to the
SERPA1 locus, did not lead to detectable indels when combined
with either nickase despite robust indel formation when
combined with Cas9 (Fig. 1c). A similar pattern of indel
formation at nicked sites was observed in HeLa and U2OS cells
(Supplementary Fig. 1a, b), and with the ABEmax base editor,
which contains a Cas9(D10A) nickase, although other base
editors resulted in reduced indel frequencies compared to their
component nickase domains alone (Supplementary Fig. 6 and
Supplementary Note 1). Observed indel frequencies did not
correlate with the presence of microhomology as predicted using
inDelphi28 (Supplementary Fig. 2b). These results suggest that the
cellular response to single nick generation is site-dependent and
unpredictable by microhomology, though in general leads to
substantially lower indel formation than the cellular response
We hypothesized that the site dependence of nickase-induced
indels could be explained if the induced nicks were converted to
DSBs by a separate cellular process such as DNA replication.
When a replication fork encounters a nick, it becomes a DSB29.
To test this possibility, we analyzed two sgRNAs (211 and 210)
that target DNA either 28 bp upstream (sgRNA 210) or 18 bp
downstream (sgRNA 211) of HEK site 2, a particularly
asymmetric locus that results in high levels of Cas9(H840A)
nickase-mediated indels but low levels of Cas9(D10A)
nickaseinduced indels. While Cas9(H840A) nickase and the HEK site
2 sgRNA resulted in high indel levels (24 ? 5%), nicking the same
strand slightly upstream or downstream of HEK site 2 resulted in
17-fold lower indel formation (Fig. 1d). These observations
indicate that the high indel frequency generated by Cas9(H840A)
nickase when paired with the HEK site 2 sgRNA is strongly
dependent on the exact site being targeted. These data suggest
that the cellular response to nicks is highly sgRNA-dependent.
The high degree of sgRNA-dependence associated with
nickinduced indels may explain previously conflicting reports of the
relative inactivity of the H840A nickase in human cells30?32.
HDR stimulated by single Cas9 nickases. The use of HDR for
precision genome editing in mammalian cells is limited by low
efficiency in many cell types (T cells being a notable exception33),
and the excess of indels and other undesired cellular outcomes
that result from DSB formation. Previous work with Cas9
nickases18,20,22,24,32, homing endonucleases converted to
nickases17, and zinc finger nickases19 demonstrates that nicks can
induce low levels of HDR when combined with a donor DNA
We wondered whether the observed variability among
nickinduced indel formation also applies to nick-induced HDR. To
assess this possibility, we designed 100-mer single-stranded DNA
oligonucleotide (ssODN) templates for each of eight genomic loci
and co-delivered them with Cas9 nuclease, Cas9 nickases, and
catalytically dead Cas9 (dCas9). For three loci (HBB, SERPA1,
and LDLR), the ssODN encoded a single human pathogenic SNP
located in the protospacer. For the remaining five loci, the donor
templates were designed to incorporate an SNP within the
protospacer as well as a PAM-altering SNP, as described in the
CORRECT method for HDR donor template design7. We
lipofected a plasmid encoding Cas9, Cas9 nickase, or dead
Cas9, a plasmid expressing the indicated sgRNA, and the
corresponding ssODN donor template into HEK293T cells. Four
days post-lipofection, genomic DNA was purified and analyzed
by high throughput sequencing (HTS). We used Crispresso234,35
to filter out reads containing indels from our alignment prior to
assessing HDR efficiency to ensure that reads containing both
indels and HDR did not contribute to tabulated HDR efficiencies.
DNA strands that contained both indels and HDR events were
counted as indels during tabulations (see Methods section).
At seven of eight sites, we detected HDR with one or both Cas9
nickases (Fig. 1e). Linear regression analysis identified a weak
positive correlation (R2 = 0.57, p = 0.031 for the Cas9(D10A)
nickase, R2 = 0.51, p = 0.045 for the H840A nickase) between
indel formation and HDR frequencies with nickases, but no
significant correlation with Cas9 nuclease (R2 = 0.08 p = 0.475)
(Supplementary Fig. 2a). Although the absolute frequencies of
HDR were 2.0-fold to 2.5-fold higher with Cas9 nuclease than
with either Cas9 nickase (average across eight sites of 10% HDR
product for Cas9, 5.0% for Cas9(H840A), and 4.0% for Cas9
(D10A)), the HDR:indel ratio was 9.1-fold to 9.6-fold higher
when using a nickase than Cas9 nuclease (the average HDR:indel
ratio was 0.23 for Cas9, 2.1 for H840A, and 2.2 for Cas9(D10A))
(Fig. 1f). Importantly, we did not detect HDR above a frequency
of 0.2% when dCas9 was paired with the same sgRNAs and donor
templates (Fig. 1e), indicating that observed HDR frequencies are
strongly dependent on Cas9 nicking, and are not artifacts of the
donor template acting as a primer during the PCR reaction prior
to HTS, a source of artificially high apparent HDR frequencies
(Supplementary Fig. 7). To ensure that the donor templates did
not participate in the PCR reactions used a size-selective DNA
purification step (see Methods section and Supplementary Fig. 7).
These experiments establish that nick-induced HDR results in
improved HDR:indel ratios compared to DSB-mediated HDR.
However, the unpredictable nature of whether a nickase will be
able to mediate HDR at a particular locus, as well as generally low
efficiency, limits the utility of simple nickase-mediated HDR.
Modulating HDR by manipulating cellular repair proteins. To
address these limitations, we sought to better understand the
cellular proteins involved in catalyzing nick-induced HDR. To
date, several studies have manipulated cellular DNA repair
processes to favor HDR over NHEJ17,22,24,36?38. Previous efforts have
identified key cellular DNA repair modulators that can be
inhibited (such as p53 binding protein 1 (53BP1))36,37 or
overexpressed (such as Rad5236) to improve HDR:indel ratios in
response to a targeted DSB. Knockdown of cellular hRad51, or
inhibition of hRad51 by overexpression of the dominant negative
mutant hRad51(K133R), increases both indel and HDR
frequencies at targeted nicks17,22,24. Guided by these observations,
we chose to manipulate DNA repair modulators and study the
resulting effects on DSB and nick-induced HDR.
We overexpressed either human hRad51 or hRad51(K133R) in
conjunction with Cas9 or the Cas9(D10A) nickase (Fig. 2a).
Overexpression of hRad51 led to a significant (p < 0.05; Student?s
two-tailed t-test, Supplementary Table 4) decrease in HDR
frequency at two of eight tested loci for Cas9(D10A)
nickmediated HDR (Fig. 2b) and at five of eight loci for
DSBmediated HDR (Fig. 2d). Conversely, overexpression of hRad51
(K133R), which inhibits cellular hRad51 activity, led to an
increase in the efficiency of nick-induced HDR, but not
DSBinduced HDR (Fig. 2b, d). Finally, HDR:indel ratios remained
largely unchanged by overexpression of hRad51 or hRad51
(K133R). Together, these data demonstrate that hRad51
inhibition increased both HDR and indel frequencies at nick sites, but
not at Cas9-induced DSBs (Fig. 2c, e). Intriguingly,
overexpression of hRad51(K133R) led to low but detectable levels of
HDR at the previously refractory SERPA1 site (Fig. 2b).
To test the potential effect of p53 binding protein 1 (53BP1) on
nick-induced HDR, we overexpressed i53, a protein inhibitor of
53BP137. 53BP1 directs DSBs towards NHEJ-mediated repair by
preventing end resection, a key event on the HDR pathway39.
Overexpression of i53 with Cas9 led to a significant (defined as p
< 0.05, Student?s two-tailed t-test) increase in the absolute
frequency of HDR at four of eight tested loci and an improvement
in the HDR:indel ratio at six of eight loci compared to Cas9 alone
(Fig. 2d, e, Supplementary Tables 4 and 5). No such HDR
improvements from i53 overexpression were observed, however,
when Cas9(D10A) nickase was used instead of Cas9 (Fig. 2b, c),
indicating that 53BP1 is unlikely to be a key modulator of
nickmediated HDR. Overexpression of Rad52, an interaction partner
of hRad51, did not increase the efficiency of HDR arising from
nicks or DSBs, but significantly improved the HDR:indel ratio at
four of eight loci when HDR was stimulated by a nick (Fig. 2b?e).
Together, these findings suggest that global inhibition of cellular
hRad51, but not inhibition of 53BP1 or elevating Rad52 levels,
can increase the frequency of HDR in response to a DNA nick.
Development of Cas9(D10A)nickase fusions that promote
HDR. Based on the above findings, we generated fusion
constructs between the Cas9(D10A) nickase or the Cas9(H840A)
nickase and hRad51(K133R) to mediate local inhibition of
hRad51 at the target site. We anticipated that such fusion
constructs might be more effective and less perturbative than global
inhibition of hRad51, which causes chromosomal instability40,
and that the hRad51 fusion partner would serve to modulate
repair of the DNA nick. Under normal circumstances, cellular
hRad51 binds to exposed genomic ssDNA after end-resection at
the nick, leading to perfect, non-mutagenic repair of the nick13,22.
This non-mutagenic repair process is inhibited by the dominant
negative hRad51(K133R) mutant, which forms mixed filaments
with wild-type hRad51 that can perform a DNA homology
search, but cannot hydrolyze ATP to initiate DNA strand
invasion41, even when low levels of the mutant protein are present42.
We began by optimizing the parameters for transfection by
performing a titration of plasmid and donor template quantities
and by measuring HDR and indel efficiencies at two loci with
both the Cas9(D10A) nickase and the hRad51(K133R)?Cas9
(D10A) fusion (Supplementary Figs. 3a?d). To our surprise, a
small quantity of ssODN (50 ng) was sufficient for efficient HDR,
and increasing the ssODN amount to 400 ng reduced HDR
efficiency. Fusion of hRad51(K133R) to the N-terminus of the
Cas9(D10A) nickase increased HDR efficiency in HEK293T cells
by an average of 2.4-fold without altering the favorable HDR:
indel ratio observed with the Cas9(D10A) nickase alone (Fig. 3b).
We refer to this fusion construct, hRad51(K133R)?Cas9(D10A)
nickase, as RDN(K133R). Compared to RDN(K133R), moving
the position of hRad51(K133R) to the C terminus of the Cas9
(D10A) nickase did not significantly alter HDR frequencies
(Fig. 3b), nor did fusing an additional monomer of hRad51
(K133R) to the N-terminus of Cas9(D10A) (Fig. 3b). Fusion of
one hRad51(K133R) monomer to the N-terminus and one to the
C-terminus, however, reduced both HDR and indel formation,
possibly due to the association of multiple fusion proteins into an
extended multimer (Fig. 3b). Consistent with the data showing
that inhibition or overexpression of hRad51 does not have a
substantial effect on DSB-mediated HDR, fusion between Cas9
and hRad51(K133R) led to a slight reduction to average HDR
frequency at the loci tested (Fig. 3b; Supplementary Figs. 4a, b).
Fusion between hRad51(K133R) and the Cas9(H840A) nickase
also did not improve HDR frequency or HDR:indel ratios
(Supplementary Figs 5e and 5f). The nickase strand preference of
HDR enhancement upon hRad51(K133R) fusion may arise from
the position of the nick introduced by Cas9(H840A) in the R-loop
of displaced genomic DNA, compared with the position of the
nick from Cas9(D10A) in the DNA:RNA duplex (Fig. 1b).
Surprisingly, fusion of wild-type hRad51 to Cas9(D10A),
hereafter referred to as RDN, also resulted in increased HDR
efficiency (Fig. 3c), even though overexpression of hRad51 in
trans with the Cas9(D10A) nickase lead to slightly decreased
HDR efficiency (Fig. 2b). These results suggest that increased
HDR frequency mediated by RDN results from a mechanism
distinct from global inhibition of hRad51. Together, these data
demonstrate that localizing hRad51 to a targeted DNA nick
through the RDN fusion increases nick-mediated HDR efficiency
without inhibition of strand invasion mediated by cellular
Next, we sought to understand if the HDR frequency
enhancement associated with RDN and RDN(K133R) arises
from simple steric occlusion of DNA repair proteins from
accessing the nick, or whether the affinity of hRad51 for
singlestranded DNA leads to localization of the single-stranded DNA
donor to the nick. To illuminate these possibilities, we created
fusions between the Cas9(D10A) nickase and RecA or
bacteriophage T4-derived single-stranded binding protein (SSB). RecA
is a bacterial homolog of hRad51 that catalyzes strand invasion
between homologous strands of DNA. Neither RecA?Cas9
or Cas9(D10A) i53, hRad51,
SNP without PAM
(HBB, SERPA1, LDLR)
SNP with PAM
D10A + hRad51
D10A + hRad51-K133R
D10A + i53
D10A + hRad52
* * *
TSH ithw 5
TSH ithwH10 **
(D10A) nor SSB?Cas9(D10A) resulted in HDR enhancement
(Fig. 3c). Furthermore, incorporation of three additional hRad51
mutants (R310A, R235E and G151D) into RDN to generate RDN
(R310A), RDN(R235E) and RDN(G151D) all displayed HDR
enhancement frequencies indistinguishable from that of RDN
and RDN(K133R) (Fig. 3c, and Supplementary Figs. 5i, j), in spite
of their differing catalytic and DNA-binding characteristics
(Fig. 3a)43?45. Taken together, these observations reveal that
neither the fusion orientation of hRad51 relative to Cas9(D10A)
nor the strand invasion and strand exchange activities of hRad51
are critical for the ability of RDN to mediate HDR.
Donor template optimization. When possible, including a
PAMaltering mutation together with the target mutation in a donor
template is an effective approach to improve HDR efficiency7,46
by preventing re-cutting and subsequent modification of the
desired HDR product. HDR efficiencies are highly dependent on
the distance between the DNA cleavage site and the mutation that
is being incorporated7,46. The above experiments used donor
templates that contain PAM-blocking mutations at five of the
eight loci tested (sgRNA 1, sgRNA 2, HEK site 2, HEK site 3, and
HEK site 4), and donor templates that lacked PAM-blocking
mutations due to unavailability of a silent PAM-blocking
mutation in addition to the target point mutation at the remaining
three sites (LDLR, HBB, and SERPA1). Since indels are generated
much less efficiently with nick-induced HDR compared to
DSBinduced HDR (Fig. 1e), we sought to test whether PAM-blocking
mutations are necessary for nick-induced HDR and to define the
region between the PAM and target mutation that can support
We designed a series of eight ssODN templates targeting the
HEK site 3 locus, each containing a SNP located in a different
position within the protospacer from position 7 to 25, counting the
PAM as positions 21?23. Two sets of donor templates were used.
The first set of ssODNs incorporated a PAM mutation (replacing
the TGG PAM with TTT) alongside the target mutation, while the
second set only encoded each target mutation. As expected, we
observed an increase in the frequency of Cas9-mediated HDR when
the PAM-blocking template was used compared to the
non-PAMblocking template (Fig. 4a). By contrast, incorporating a PAM
mutation into the donor ssODN did not lead to increased HDR
frequency for nick-induced HDR, mediated either by Cas9(D10A)
or RDN(K133R), as long as the target mutation is located within the
sgRNA protospacer sequence (Fig. 4a).
We previously measured the frequency of HDR at HEK site 3
using a donor template with a PAM-blocking mutation (replacing
the TGG PAM with TCC, Supplementary Tables 1 and 2)
using Cas9 (Fig. 1e), Cas9(D10A) (Fig. 1e), or RDN(K133R)
BRCA2 Self- ssDNA
binding association binding
Mean HDR:indel ratio
Mean HDR:indel ratio
Preventing BRCA2 binding to site Stochastic indel
C Desired product
Preventing selfassociation of hRad51 monomers
10 15 20 25
Position in protospacer
Cas9, no PAM mutation
Cas9(D10A), no PAM mutation
RDN(K133R), no PAM mutation
Cas9, with PAM mutation
Cas9(D10A), with PAM mutation
RDN(K133R), with PAM mutation
Fig. 4 Characterization of positional dependence and off-target editing of
nick-mediated HDR. a HDR frequencies measured by high-throughput
sequencing in unsorted HEK293T cells using ssODNs with point mutations
distributed along the sgRNA protospacer sequence of the HEK 3 sgRNA
site. In previous figures, an oligonucleotide with a different PAM-blocking
mutation at HEK Site 3 was used to measure an SNP incorporated at
position 12 in the protospacer. b Indel frequencies at off-target genomic loci
in cells treated with Cas9 nuclease, Cas9(D10A) nickase, or Cas9(D10A)
fusions with hRad51 or the indicated mutants thereof. Dead Cas9 (dCas9)
treated cells were included as a negative control. All data are shown as
individual data points and mean ? s.d. for n = 3 independent biological
replicates, performed on different days. Source data are provided in the
Source Data file
(Supplementary Fig. 5c). The HDR frequencies from Cas9 and
RDN(K133R) were very similar when these different
oligonucleotides were used. For example, Cas9 yielded 4.7 ? 0.5% HDR with a
TTT-blocking mutation, and 5.7 ? 0.9% with a TCC-blocking
mutation. However, the mean value for Cas9(D10A) increased
from 2.6 ? 1.0% with the TCC PAM blocking mutation to 7.9 ?
3.2% with the TTT PAM blocking mutation, an unexpected result
that suggests some ssODN dependence for
Unlike DSB-induced HDR, in which HDR efficiency steeply
declines as the distance between the DSB and the incorporated
mutation increases7,46 (Fig. 4a), we observed comparable HDR
efficiencies when RDN(K133R) was paired with different donor
templates that introduced mutations from position 7 to 18 in the
protospacer (Fig. 4a). This greater apparent independence of
HDR efficiency from the location of the mutation to be installed
relative to the protospacer suggests that RDN may offer more
flexibility with regards to guide RNA choice than Cas9
We also tested donor template oligonucleotides that were
oriented in the same sense as the sgRNA (forward template,
which was used for all other experiments in this study) and in the
opposite sense (reverse template). We did not observe any
significant differences (Student?s two-tailed t-test) in the resulting
HDR efficiencies mediated by Cas9(D10A), Cas9, Cas9(H840A),
or RDN(K133R) (Supplementary Fig. 4c), indicating that ssODN
orientation is not a substantial determinant of HDR efficiencies
under the conditions tested.
RDN with additional hRad51 mutants. Although the
development of RDN as a tool to mediate HDR led to consistently
improved HDR:indel ratios, the overall frequency of
RDNmediated HDR is similar to that of Cas9-mediated HDR (Fig. 3c).
In an attempt to improve overall HDR efficiency further while
maintaining favorable HDR:indel ratios, we assessed four
additional mutants of hRad51 in RDN constructs.
In addition to their role in catalyzing DNA strand invasion,
hRad51 monomers directly bind to BRCA247?49, or to other
hRad51 monomers25,50. Mutants of hRad51 that have lost either
or both of these capabilities have been engineered25,26 (Fig. 3a).
We installed these mutations into the RDN context and assayed
HDR and indel outcomes of the resulting constructs to assess
whether these binding interactions influence editing outcomes
(Fig. 3d?f). The results revealed that using hRad51 mutants
incapable of self-association, but which maintain BRCA2 binding,
increased HDR efficiency in HEK293T cells at the eight tested
sites to an average of 14% (F86E mutant, RDN(F86E)) or 15%
(A89E mutant, RDN(A89E)), compared to 10% for RDN. Both of
these mutants were associated with a modest reduction in HDR:
indel ratio, from an average of 1.9 for RDN to 0.93 for RDN
(F86E) or 0.98 for RDN(A89E).
In contrast, removing the BRCA2-binding ability of hRad51
using the double mutant (RDN(S208E A209D)) only slightly
improved HDR efficiency relative to RDN (to an average of 12%),
but substantially improved the HDR:indel ratio (to 3.3),
suggesting that abolishing recruitment of BRCA2 to the nick
promotes more favorable HDR:indel partitioning. We should
note that even with these improvements, the efficiency of
nickinduced HDR remains more sgRNA-dependent than the
efficiency of DSB-induced HDR. For example, pairing original
or mutant RDN constructs with sgRNA SERPA1 leads to modest
(<3%) HDR frequencies compared with Cas9 (11.1 ? 0.6%).
We tested a final hRad51 A190L A192L double mutant that
lacks both BRCA2-binding and hRad51 self-association ability.
RDN(A190L A192L) mediated HDR with an average efficiency of
14% and an HDR:indel ratio of 1.6, offering intermediate levels of
HDR efficiency and HDR:indel ratio compared to the above RDN
These analyses inform potential mechanisms by which RDN
can mediate efficient HDR with favorable HDR:indel ratios. The
data are consistent with a model in which self-association of
hRad51 is important to maintain a high HDR:indel ratio but also
limits HDR efficiency by promoting perfect repair of the DNA
nick. In contrast, recruitment of BRCA2 to the nick site reduces
the rate of perfect repair of the nick (Fig. 3g). For applications
that benefit most from maintaining the highest possible HDR
efficiency, RDN(A89E) is the most useful, whereas applications
that require maximizing the HDR:indel ratio will benefit from use
of the RDN (S208E A209D) variant.
Off-target modification by RDN variants. Cas9 nuclease51 and
Cas9-derived proteins such as base editors3,5,52 can induce
offtarget editing in an sgRNA-dependent fashion. We characterized
off-target editing at known off-target sites associated with three
well-studied sgRNAs51: HEK site 2, HEK site 3, and HEK site 4,
which is a notoriously promiscuous sgRNA53,54. Although the
homology required between the target genomic locus and the
ssODN prevents significant off-target HDR products from being
generated by Cas9 combined with a ssODN, indel formation from
Cas9 nuclease activity at off-target sites under these conditions is
common. Off-target indel formation was measurable (>0.1%)
with Cas9 treatment at all tested Cas9 off-target sites, and
offtarget indel formation ranged in efficiency from 0.12 to 98%
(Fig. 4b). In contrast, Cas9(D10A) nickase and RDN edited only
two of the 12 off-target loci (>0.1% indel formation) (Fig. 4b).
The more efficient RDN(A89E) edited four of 12 off-target sites at
a frequency >0.1%, all of which are associated with the
promiscuous HEK site 4 guide RNA. These results indicate that
RDN-mediated HDR offers substantially lower off-target DNA
modification than nuclease-based HDR, and that this trend even
applies to RDN(A89E), which typically results in higher on-target
HDR frequencies than Cas9.
HDR in other human cell types. HEK293 and HEK293T cells are
known to be particularly amenable to ssODN-mediated HDR55.
Indeed, some other commonly used immortalized cell lines
including HeLa and U2OS are thought to be completely
refractory to ssODN-mediated HDR55. We compared RDN- and
Cas9mediated HDR outcomes in other immortalized cell lines and in
primary human cells, including HeLa cells, U2OS cells, human
induced pluoripotent stem (hiPS) cells and K562 cells.
In HEK293T cells, RDN(A89E) offers the highest HDR
frequency (Fig. 3e) and RDN(S208E A209D) offered the highest
HDR:indel ratio (Fig. 3f) of all the constructs tested, so we tested
these two constructs in the wider range of cell types. For this
comparison, we used oligonucleotides designed without PAM
mutations to maximize the generality of the results and due to our
conclusions that nick-mediated HDR does not benefit from PAM
blocking mutations (Fig. 4a). Unless otherwise specified, we
report results from unsorted cells as percentages of the entire cell
population, not as percentages of edited or modified cells, which
would greatly increase apparent editing efficiencies.
RDN (containing wild-type hRad51) led to substantially
reduced HDR frequencies when compared to Cas9 in all
nonHEK293T cell types tested. For example, in K562 cells the average
reduction in efficiency was from a mean of 16% with Cas9 to
3.8% with RDN (Fig. 5a). The mean HDR:indel ratio, however,
was improved 87-fold in K562 cells and 3-fold in HeLa cells
(Fig. 5b, d). RDN(S208E A209D) demonstrated slightly improved
HDR:indel ratios when compared to RDN, but the overall
efficiency of HDR remained low compared to that achieved by
Cas9 (Fig. 5).
When RDN(A89E) was used, however, the average HDR
efficiency was substantially improved, with mean HDR
frequencies of 8.3% in K562, 1.3% in U2OS, and 1.8% in HeLa cells
(Fig. 5a, c, e). HDR efficiencies in these three non-HEK293T cell
types were on average 2.1-fold lower than those following Cas9
treatment. RDN(A89E) was associated with a 15-fold
improvement in HDR:indel ratio in K562 cells and 7-fold in HeLa cells
compared to Cas9 treatment. This improvement was not
observed in U2OS cells, which exhibited a slight reduction in
HDR:indel ratio when RDN(A89E) was used (Fig. 5f). In hiPS
cells, only one of the three tested loci was amenable to RDN
(A89E)-mediated HDR, demonstrating that this approach may be
more site-dependent in hiPS cells than in immortalized cell lines.
To test if this limitation was due to poor expression of RDN
(A89E) in hiPS cells, we generated Cas9 and RDN(A89E)
constructs tagged with P2A GFP to enable isolation of
Cas9expressing or RDN(A89E)-expressing cells. With
Cas9?P2A?GFP, isolating GFP-positive cells resulted in 1.8%
average HDR efficiencies in hiPS cells with an average HDR:indel
ratio of 0.03 (Fig. 5i, j). Among GFP-positive cells expressing
RDN(A89E)?P2A?GFP, average HDR efficiencies were 1.0%,
with an average HDR:indel ratio of 46 (Fig. 5i, j), reflecting a
modest decrease in HDR efficiency but a >1000-fold
improvement in HDR:indel ratio. (Fig. 5i, j, Supplementary Fig. 8).
Among GFP-positive cells isolated with the
RDN(A89E)-P2AGFP construct, average indel frequency was 1.6% and the vast
majority showed no target site modification. This observation
suggests that the majority of nicks induced by RDN(A89E)
construct are perfectly repaired in hiPS cells; in contrast,
GFPpositive cells containing Cas9?P2A?GFP contained an average of
These data together reveal that RDN(A89E) mediates more
efficient HDR than Cas9 nuclease in HeLa and HEK293T cells,
maintains similar levels of HDR efficiency in K562 cells, and
offers improved HDR:indel ratios in HeLa, HEK293T, K562,
and hiPS cells. Neither an efficiency nor a product purity
advantage from any tested RDN variant was observed in U2OS
cells, possibly as a result of unusual regulation of DNA repair in
U2OS cells55,56. This variability is likely due to the reliance of
RDN on cellular repair processes that are highly cell
The method developed in this study enables precise and
specific changes to be made to genomic DNA through
homologydirected repair, without generating a double stranded DNA
break. Use of the fusion construct hRad51?Cas9(D10A)
(RDN) or variants of this construct in which hRad51 has been
replaced by hRad51 mutants, can address some of the
challenges associated with using HDR to make precise changes to
genomic DNA in certain human cell types.
The HDR:indel ratio generated by RDN is generally
improved compared to that which can be achieved using a DSB.
This improvement in the purity of editing outcomes is
particularly important for genome editing applications in which
gene knockout resulting from indel formation opposes desired
biological outcomes, or in which mixtures of many different
edited genotypes?the typical cellular response to DSBs?is
undesired. The RDN(S208E A209D) construct is particularly
useful under such circumstances since it offers ~3.2-fold more
HDR product than indels (Fig. 3d). In addition, the efficiency of
HDR mediated by RDN and RDN(A89E) is higher than that of
Cas9 in some (but not all) cell types (Figs. 3e, 5a), although
HDR efficiency remains modest, likely limited by dependence
on cellular DNA repair processes. RDN and its variants also
offer substantially higher DNA specificity (lower off-target
indel formation) compared to Cas9 nucleases combined with
the same sgRNAs, even when applied to a notoriously
promiscuous guide RNA with many known off-target loci (Fig. 4b).
RDN with wild-type hRad51 offers the greatest degree of DNA
specificity among the mutants tested, but this difference was
only notable at the promiscuous HEK Site 4, as were not able to
detect off-target editing at frequencies above 0.2% at any other
tested loci following use of RDN, RDN(A89E) or RDN(S208E
A209D) (Fig. 4b). Finally, since RDN variants cannot directly
generate DSBs, we anticipate that the likelihood of inducing
translocations, large deletions, or p53 activation will be greatly
reduced compared to nuclease-based genome editing methods.
Additional studies using are needed to fully characterize the
scope of cellular responses to targeted nicks compared to
We anticipate that RDN(A89E) or RDN(S208E A209D) will
be useful for applications in which efficiency or cleanliness of
genome editing are critical. Recent work whereby saturation
genome editing was performed to investigate variants of
unknown significance in BRCA157 highlight the utility of a
tool with the ability to generate mutations with single
nucleotide resolution. Nuclease-mediated approaches to
saturation editing can only be performed on essential genes
HEK site 3
HEK site 3
HEK site 3
LDLR HEK site 3
eqS hHD 5
LDLR HEK site 3
LDLR HEK site 3
because of the requirement that cells in which indels are
induced must be excluded from the analysis. The favorable
HDR:indel ratio and HDR efficiency offered by RDN
may permit mutagenesis with nucleotide-level resolution on
non-essential genes. Finally, we hope that the principles
illuminated in this work will be enabling for researchers seeking
to develop new methods for studying and manipulating
cellular DNA damage and repair.
Plasmid cloning. All mammalian cell expression plasmids were constructed by
USER cloning from gBlock gene fragments (Integrated DNA Technologies) with
USER junctions sized between 14 and 20 nucleotides58. Phusion U Green Multiplex
PCR Master Mix (ThermoFisher) was used for amplification of DNA. sgRNA
plasmids were constructed by blunt end ligation of a linear PCR product generated
by encoding the 20-nt variable protospacer sequence onto the 5? end of an
amplification primer and treating the resulting piece to KLD Enzyme Mix (New
England Biolabs) according to the manufacturers? instruction. Mach1 chemically
competent E. coli (ThermoFisher) cells were used.
Preparation of plasmids for mammalian cell transfection. To obtain endotoxin
free plasmids for transfection, 45 mL of Mach1 cells expressing freshly-transformed
plasmid were pelleted by centrifugation (6000?g, 10 min, 4 ?C) and purified using
ZymoPURE II Plasmid Midi Prep Kits (Zymo Research), according to the
manufacturer?s instructions with the inclusion of the optional step of passing the
plasmid across the EndoZero Spin column (Zymo Research). Plasmid yield was
quantified using a Nanodrop and by electrophoresis on a 1% agarose Tris/Borate/
EDTA gel supplemented with ethidium bromide.
Mammalian cell culture. All cells were cultured and maintained at 37 ?C with 5%
CO2. Antibiotics were not used for cell culture. HEK293T cells (ATCC CRL-3216)
and HeLa cells (ATCC CCL-2) were cultured in Dulbecco?s modified Eagle?s
medium (DMEM) plus GlutaMax (ThermoFisher) supplemented with 10% (v/v)
fetal bovine serum (FBS). K562 cells (ATCC CCL-243) were cultured in Roswell
Park Memorial Institute (RPMI) 1640 Medium plus GlutaMax (ThermoFisher)
supplemented with 10% (v/v) FBS. U2OS cells were cultured in MyCoy?s 5A
Medium plus GlutaMax (ThermoFisher) supplemented with 10 % (v/v) FBS.
hiPS cells (human episomal iPS cell line; A18945; ThermoFisher) were cultured
in Essential 8 Flex Medium (ThermoFisher) supplemented with RevitaCell after
passaging (ThermoFisher) according to the manufacturer?s directions. Versene
(Thermo Fisher) was used for cell passaging and dissociation. Prior to
nucleofection, cells were harvested with Accutase (ThermoFisher).
For data shown in Fig. 5i, j, nuclease expression plasmids were constructed
whereby the Cas-enzyme construct (Cas9 or RDN(A89E)) was proceeded by
P2AGFP to enable isolation of transfected cells. iPS cells were flow sorted at the MIT
FACS core 3?5 days after nucleofection and genomic DNA was isolated directly
Mammalian cell lipofection and genomic DNA isolation. HEK293T cells were
seeded on 48-well poly-D-lysine coated plates (Corning) 16?20 h before lipofection.
Lipofection was performed at a cell density of 65%. Unless otherwise stated, cells
were transfected with 231 ng of nuclease-editor or base-editor expression plasmid
DNA, 69 ng of sgRNA expression plasmid DNA, 50 ng (1.51 pmol) 100-nt ssODN
(PAGE-purified; Integrated DNA Technologies) and 1.4 ?L Lipofectamine 2000
(ThermoFisher) per well. For experiments where global inhibition or
overexpression of a cellular HDR-component was performed 100 ng of the appropriate
plasmid was included. Cells were harvested 4 days post-transfection and genomic
DNA isolation and purification was performed with Agincort DNAdvance Kit
(Beckman Coulter), according to the manufacturer?s protocol. Size-selective DNA
purification was necessary to prevent contamination of gDNA with donor ssODN
HDR templates. For analysis of indel formation in Supplementary Fig. 1, HeLa and
U2OS cells were transfected according to the above protocol except they were
transfected at a density of 80% with 1.4 ?L Lipofectamie 3000 and 1 ?L of P3000
(ThermoFisher) per well.
Nucleofection of mammalian cells. For data generated in Fig. 5, nucleofection of
K562, HeLa and U2OS cells was performed. For these three cell types, 350 ng
nuclease-expression plasmid, 150 ng sgRNA-expression plasmid and 200 pmol
(6.6 ?g) 100-nt ssODN (PAGE-purified; Integrated DNA Technologies) was
nucleofected in a final volume of 20 ?L per sample in a 16-well Nucleocuvette
strip (Lonza). K562 cells were nucleofected using the SF Cell Line
4DNucleofector X Kit (Lonza) with 5 ? 105 cells per sample (program FF-120),
according to the manufacturer?s protocol. U2OS cells were nucleofected using
the SE Cell Line 4D-Nucleofector X Kit (Lonza) with 3?4 ? 105 cells per sample
(program DN-100), according to the manufacturer?s protocol. HeLa cells were
nucleofected using the SE Cell Line 4D-Nucleofector X Kit (Lonza) with 2 ? 105
cells per sample (program CN-114), according to the manufacturer?s protocol.
Cells were harvested 48 hours after nucleofection; genomic DNA was purified
using the Agincort DNAdvance Kit (Beckman Coulter), according to the
hiPS cells were nucleofected with 400 ng nuclease-expression plasmid, 400 ng
sgRNA-expression plasmid and 200 pmol (6.6 ?g) 100-nt ssODN (PAGE-purified;
Integrated DNA Technologies) in a final volume of 20 ?L per sample in a 16-well
Nucleocuvette strip (Lonza) using the CB-150 program in the P3 Primary Cell
4DNucleofector X Kit (Lonza) with 0.75?1.5 ? 106 cells per sample.
Preparation of genomic DNA for high throughput sequencing. Sites of interest
were amplified using the primers listed (Supplementary Table 3). Amplification
primers for the first PCR reaction (PCR1) were designed with primer2 and had 5?.
extensions to enable amplification with an Illumina barcoding primer in a second
PCR reaction (PCR2). Phusion U Green Multiplex PCR Master Mix
(ThermoFisher) was used for both PCR1 and PCR2. For PCR1, each reaction contained 0.5
?M of the appropriate forward and reverse primer (Supplementary Table 3) and
30?100 ng of genomic DNA was as a template. Cycling conditions were 98 ?C for 1
min 30 s, then 30 cycles of (98 ?C for 10 s, 61 ?C for 15 s, and 72 ?C for 15 s)
followed by a final extension of 1 min at 72 ?C per 30 ?L reaction. PCR1 products
were verified on a 2% agarose gel Tris/Borate/EDTA gel supplemented with
ethidium bromide. For PCR2, 1 ?L of unpurified PCR1 plus 0.5 ?M of each of a unique
forward and reverse barcoding primer pair were added to each sample for a final
volume of 30 ?L. Cycling conditions were 98 ?C for 1 min 30 s, then 7 cycles of (98 ?
C for 10 s, 61 ?C for 15 s, and 72 ?C for 15 s) followed by a final extension of 1 min
at 72 ?C. PCR2 products were purified by gel electrophoresis on a 2% agarose gel
using the QIAquick Gel Extraction Kit (Qiagen). Purified product was passed over
a second Minelute column (Qiagen) for a further round of purification before
quantification with QBit ssDNA HS Assay Kit (ThermoFisher) and sequenced
using an Illumina MiSeq with 230?270-bp single end reads according to the
Analysis of HTS data. Demultiplexing of pooled sequencing reads was
performed using the MiSeq Reporter software (Illumina). Crispresso-v235 was used
to perform alignments between sequenced amplicons and reference amplicons.
Indels were quantified in a 10-bp window surrounding the expected cut site for
each sgRNA. For quantification of HDR, we discarded reads that contained
indels from the alignment to the reference sequence using ?discard-indel-reads?
filter. This approach ensured that we did not erroneously count reads that
contained both an SNP incorporated through HDR and an indel as an HDR
event, as has been previously described7. The resulting alignment contained only
reads that do not contain an indel within the 10-bp window around the sgRNA
cleavage site. Separately from the alignment matrix, the output of Crispresso-v2
reported the percentage of reads that had been excluded from the alignment
because they contained an indel (%cells with indel). For each target point
mutation that was incorporated via HDR, the alignment alone could be used to
determine the % of non-indel containing cells (% indel-free cells with target
mutation) that had successfully incorporated the target mutation. In order to
assess the % of all cells that had the target mutation, the following correction was
% Cells with target mutation ? % indelfree cells with target mutation ?
For calculation of HDR:indel ratio, the % cells with indel-free HDR at the
indicated sequence was divided by the % cells with an indel in the 10-bp window
surrounding the cleavage site. For experiments with HEK293T cells, where
robust (>1%) HDR and indel percentages were detectable for many conditions,
HDR:indel ratios were not calculated if HDR frequency was less than 1% for a
particular sample, to avoid reporting artificially high HDR:indel ratios that could
accompany very low frequency events. For the data shown in Fig. 5, HDR and
indel frequencies were measured in cell types less able than HEK293T cells to
support HDR. For these instances, an HDR:indel ratio was not reported if the
HDR frequency was <0.1% for the same reason. For calculations in the text in
which averages across sites were made, if an HDR:indel ratio was not calculated
due to a low HDR rate, then the HDR:indel ratio was set to zero when calculating
the mean to avoid artificially inflating HDR:indel ratios.
% cells with indel
Plasmids encoding the constructs used in this study are available on Addgene, accession
numbers can be found in Supplementary Table 6. The accession numbers have been
listed in Supplementary Table 6. The source data underlying Figs. 1c?f, 2b?e, 3b?f, 4a?b,
and 5a-e and Supplementary Figs. 1a?b, 3a?d, 4, 5a?j, 6a?b, and 7a?b are provided as a
Source Data file. High-throughput DNA sequencing data has been deposited in the NCBI
Sequence Read Archive with BioProject accession number PRJNA515942 (SRP180368).
H.A.R. and D.R.L. designed the study; H.A.R. generated reagents; H.A.R and W-H.Y.
performed experiments. D.R.L. supervised the research. All authors wrote the manuscript
and assisted with proofing and revisions.
Supplementary Information accompanies this paper at
Competing interests: D.R.L. is a consultant and co-founder of Editas Medicine, Beam
Therapeutics, and Pairwise Plants, companies that use genome editing. H.A.R. and D.R.L
have filed a patent application on RDN-mediated genome editing. H.A.R and D.R.L are
co-inventors on a provisional patent application filed by the Broad Institute covering the
use of hRad51 and its variants fused to nickases for enhanced HDR. The remaining
author declares no competing interests.
Reprints and permission information is available online at http://npg.nature.com/
Journal peer review information: Nature Communications thanks the anonymous
reviewers for their contribution to the peer review of this work.
Publisher?s note: Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the article?s Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not included in the
article?s Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this license, visit http://creativecommons.org/
1. Rouet , P. , Smih , F. & Jasin , M. Introduction of double-strand breaks into the genome of mouse cells by expression of a rare-cutting endonuclease . Mol. Cell Biol . 14 , 8096 - 8106 ( 1994 ).
2. Doudna , J. A. & Charpentier , E. Genome editing. The new frontier of genome engineering with CRISPR-Cas9 . Science 346 , 1258096 ( 2014 ).
3. Gaudelli , N. M. et al. Programmable base editing of A*T to G*C in genomic DNA without DNA cleavage . Nature 551 , 464 - 471 ( 2017 ).
4. Komor , A. C. , Badran , A. H. & Liu , D. R. CRISPR-based technologies for the manipulation of eukaryotic genomes . Cell 168 , 20 - 36 ( 2017 ).
5. Komor , A. C. , Kim , Y. B. , Packer , M. S. , Zuris , J. A. & Liu , D. R. Programmable editing of a target base in genomic DNA without double-stranded DNA cleavage . Nature 533 , 420 - 424 ( 2016 ).
6. Rouet , P. , Smih , F. & Jasin , M. Expression of a site-specific endonuclease stimulates homologous recombination in mammalian cells . Proc. Natl Acad. Sci. USA 91 , 6064 - 6068 ( 1994 ).
7. Paquet , D. et al. Efficient introduction of specific homozygous and heterozygous mutations using CRISPR/Cas9 . Nature 533 , 125 - 129 ( 2016 ).
8. Chapman , J. R. , Taylor, M. R. & Boulton , S. J. Playing the end game: DNA double-strand break repair pathway choice . Mol. Cell 47 , 497 - 510 ( 2012 ).
9. Richardson , C. & Jasin , M. Frequent chromosomal translocations induced by DNA double-strand breaks . Nature 405 , 697 - 700 ( 2000 ).
10. Kosicki , M. , Tomberg , K. & Bradley , A. Repair of double-strand breaks induced by CRISPR-Cas9 leads to large deletions and complex rearrangements . Nat. Biotechnol . 36 , 765 - 771 ( 2018 ).
11. Haapaniemi , E. , Botla , S. , Persson , J. , Schmierer , B. & Taipale , J. CRISPR-Cas9 genome editing induces a p53-mediated DNA damage response . Nat. Med . 24 , 927 - 930 ( 2018 ).
12. Ihry , R. J. et al. p53 inhibits CRISPR-Cas9 engineering in human pluripotent stem cells . Nat. Med . 24 , 939 - 946 ( 2018 ).
13. Maizels , N. & Davis , L. Initiation of homologous recombination at DNA nicks . Nucleic Acids Res . 46 , 6962 - 6973 ( 2018 ).
14. Caldecott , K. W. Single-strand break repair and genetic disease . Nat. Rev. Genet . 9 , 619 - 631 ( 2008 ).
15. Lindahl , T. Instability and decay of the primary structure of DNA . Nature 362 , 709 - 715 ( 1993 ).
16. Rees , H. A. & Liu , D. R. Base editing: precision chemistry on the genome and transcriptome of living cells . Nat. Rev. Genet ., https://doi.org/10.1038/s41576- 018-0059- 1 ( 2018 ).
17. Davis , L. & Maizels , N. Homology-directed repair of DNA nicks via pathways distinct from canonical double-strand break repair . Proc. Natl Acad. Sci. USA 111 , E924 - E932 ( 2014 ).
18. Davis , L. , Zhang , Y. & Maizels , N. Assaying repair at DNA nicks . Methods Enzymol . 601 , 71 - 89 ( 2018 ).
19. Ramirez , C. L. et al. Engineered zinc finger nickases induce homology-directed repair with reduced mutagenic effects . Nucleic Acids Res . 40 , 5560 - 5568 ( 2012 ).
20. Kan , Y. , Ruis , B. , Takasugi , T. & Hendrickson , E. A. Mechanisms of precise genome editing using oligonucleotide donors . Genome Res . 27 , 1099 - 1111 ( 2017 ).
21. Mali , P. et al. RNA-guided human genome engineering via Cas9 . Science 339 , 823 - 826 ( 2013 ).
22. Davis , L. & Maizels , N. Two distinct pathways support gene correction by single-stranded donors at DNA nicks . Cell Rep . 17 , 1872 - 1881 ( 2016 ).
23. Cong , L. et al. Multiplex genome engineering using CRISPR/Cas systems . Science 339 , 819 - 823 ( 2013 ).
24. Bothmer , A. et al. Characterization of the interplay between DNA repair and CRISPR/Cas9-induced DNA lesions at an endogenous locus . Nat. Commun . 8 , 13905 ( 2017 ).
25. Pellegrini , L. et al. Insights into DNA recombination from the structure of a RAD51-BRCA2 complex . Nature 420 , 287 - 293 ( 2002 ).
26. Yu , D. S. et al. Dynamic control of Rad51 recombinase by self-association and interaction with BRCA2 . Mol. Cell 12 , 1029 - 1041 ( 2003 ).
27. Jinek , M. et al. A programmable dual-RNA-guided DNA endonuclease in adaptive bacterial immunity . Science 337 , 816 - 821 ( 2012 ).
28. Shen , M. W. et al. Predictable and precise template-free CRISPR editing of pathogenic variants . Nature 563 , 646 - 651 ( 2018 ).
29. Payen , C. , Koszul , R. , Dujon , B. & Fischer , G. Segmental duplications arise from Pol32-dependent repair of broken forks through two alternative replication-based mechanisms . PLoS Genet . 4 , e1000175 ( 2008 ).
30. Shen , B. et al. Efficient genome modification by CRISPR-Cas9 nickase with minimal off-target effects . Nat. Methods 11 , 399 - 402 ( 2014 ).
31. Ran , F. A. et al. Double nicking by RNA-guided CRISPR Cas9 for enhanced genome editing specificity . Cell 154 , 1380 - 1389 ( 2013 ).
32. Miyaoka , Y. et al. Systematic quantification of HDR and NHEJ reveals effects of locus, nuclease, and cell type on genome-editing . Sci. Rep . 6 , 23549 ( 2016 ).
33. Roth , T. L. et al. Reprogramming human T cell function and specificity with non-viral genome targeting . Nature 559 , 405 - 409 ( 2018 ).
34. Pinello , L. et al. Analyzing CRISPR genome-editing experiments with CRISPResso . Nat. Biotechnol . 34 , 695 - 697 ( 2016 ).
35. Clement , K. et al. CRISPResso2 provides accurate and rapid genome editing sequence analysis . Nat. Biotechnol . 37 , 224 - 226 ( 2019 ).
36. Paulsen , B. S. et al. Ectopic expression of RAD52 and dn53BP1 improves homology-directed repair during CRISPR-Cas9 genome editing . Nat. Biomed. Eng . 1 , 878 - 888 ( 2017 ).
37. Canny , M. D. et al. Inhibition of 53BP1 favors homology-dependent DNA repair and increases CRISPR-Cas9 genome-editing efficiency . Nat. Biotechnol . 36 , 95 - 102 ( 2018 ).
38. Lin , S. , Staahl , B. T. , Alla , R. K. & Doudna , J. A. Enhanced homology-directed human genome engineering by controlled timing of CRISPR/Cas9 delivery . Elife 3 , e04766 ( 2014 ).
39. San Filippo, J., Sung , P. & Klein , H. Mechanism of eukaryotic homologous recombination . Annu. Rev. Biochem . 77 , 229 - 257 ( 2008 ).
40. Schlacher , K. et al. Double-strand break repair-independent role for BRCA2 in blocking stalled replication fork degradation by MRE11 . Cell 145 , 529 - 542 ( 2011 ).
41. Stark , J. M. et al. ATP hydrolysis by mammalian RAD51 has a key role during homology-directed DNA repair . J. Biol. Chem . 277 , 20185 - 20194 ( 2002 ).
42. Kim , T. M. et al. RAD51 mutants cause replication defects and chromosomal instability . Mol. Cell Biol . 32 , 3663 - 3680 ( 2012 ).
43. Prasad , T. K. , Yeykal , C. C. & Greene , E. C. Visualizing the assembly of human Rad51 filaments on double-stranded DNA . J. Mol. Biol . 363 , 713 - 728 ( 2006 ).
44. Mason , J. et al. Non-enzymatic roles of human RAD51 at stalled replication forks . BioRxiv . https://doi.org/10.1101/359380 ( 2018 ).
45. Marsden , C. G. et al. The tumor-associated variant RAD51 G151D induces a hyper-recombination phenotype . PLoS Genet 12 , e1006208 ( 2016 ).
46. Kwart , D. , Paquet , D. , Teo , S. & Tessier-Lavigne , M. Precise and efficient scarless genome editing in stem cells using CORRECT . Nat. Protoc . 12 , 329 - 354 ( 2017 ).
47. Yang , H. et al. BRCA2 function in DNA binding and recombination from a BRCA2-DSS1-ssDNA structure . Science 297 , 1837 - 1848 ( 2002 ).
48. Yang , H. , Li , Q. , Fan , J. , Holloman , W. K. & Pavletich , N. P. The BRCA2 homologue Brh2 nucleates RAD51 filament formation at a dsDNA-ssDNA junction . Nature 433 , 653 - 657 ( 2005 ).
49. Liu , J. , Doty , T. , Gibson , B. & Heyer , W. D. Human BRCA2 protein promotes RAD51 filament formation on RPA-covered single-stranded DNA . Nat. Struct. Mol. Biol . 17 , 1260 - 1262 ( 2010 ).
50. Ma , C. J. , Kwon , Y. , Sung , P. & Greene , E. C. Human RAD52 interactions with replication protein A and the RAD51 presynaptic complex . J. Biol. Chem . 292 , 11702 - 11713 ( 2017 ).
51. 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 ).
52. Rees , H. A. et al. Improving the DNA specificity and applicability of base editing through protein engineering and protein delivery . Nat. Commun . 8 , 15790 ( 2017 ).
53. 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 ).
54. Hu , J. H. et al. Evolved Cas9 variants with broad PAM compatibility and high DNA specificity . Nature 556 , 57 - 63 ( 2018 ).
55. Richardson , C. D. et al. CRISPR-Cas9 genome editing in human cells occurs via the Fanconi anemia pathway . Nat. Genet . 50 , 1132 - 1139 ( 2018 ).
56. Tay , Y. , Tan , S. M. , Karreth , F. A. , Lieberman , J. & Pandolfi , P. P. Characterization of dual PTEN and p53-targeting microRNAs identifies microRNA-638/Dnm2 as a two-hit oncogenic locus . Cell Rep . 8 , 714 - 722 ( 2014 ).
57. Findlay , G. M. et al. Accurate classification of BRCA1 variants with saturation genome editing . Nature 562 , 217 - 222 ( 2018 ).
58. Badran , A. H. et al. Continuous evolution of Bacillus thuringiensis toxins overcomes insect resistance . Nature 533 , 58 - 63 ( 2016 ).
This work was supported by U.S. NIH U01 AI142756, RM1 HG009490, R01 EB022376, and R35 GM118062, DARPA HR0011-17-2-0049, and HHMI . We thank A. Hamidi , J.
? The Author(s) 2019