Articles

Broad specificity profiling of TALENs results in

engineered nucleases with improved DNA-cleavage
specificity

npg

2014 Nature America, Inc. All rights reserved.

John P Guilinger1,2, Vikram Pattanayak1,2, Deepak Reyon3,4, Shengdar Q Tsai3,4, Jeffry D Sander3,4,
J Keith Joung3,4 & David R Liu1,2
Although transcription activatorlike effector nucleases
(TALENs) can be designed to cleave chosen DNA sequences,
TALENs have activity against related off-target sequences.
To better understand TALEN specificity, we profiled 30 unique
TALENs with different target sites, array length and domain
sequences for their abilities to cleave any of 1012 potential
off-target DNA sequences using in vitro selection and highthroughput sequencing. Computational analysis of the selection
results predicted 76 off-target substrates in the human genome,
16 of which were accessible and modified by TALENs in human
cells. The results suggest that (i) TALE repeats bind DNA
relatively independently; (ii) longer TALENs are more tolerant
of mismatches yet are more specific in a genomic context; and
(iii) excessive DNA-binding energy can lead to reduced TALEN
specificity in cells. Based on these findings, we engineered a
TALEN variant that exhibits equal on-target cleavage activity but
tenfold lower average off-target activity in human cells.

The ability to engineer site-specific changes in genomes is a

powerful capability with important research and therapeutic implications. TALENs are fusions of the FokI restriction
endonuclease cleavage domain with a DNA-binding TALE repeat
array (Fig. 1a). These arrays consist of multiple 34-amino acid
TALE repeats, each of which uses a repeat variable diresidue
(RVD), the amino acids at positions 12 and 13, to recognize
each of the four DNA nucleotides1,2. Thus, one can construct a
TALE repeat to bind virtually any DNA sequence. TALENs can
be engineered to be active only as heterodimers using obligate
heterodimeric FokI variants3,4. In this configuration, two distinct TALEN monomers each bind one target half-site and cleave
the DNA spacer sequence between the two half-sites. In cells,
TALEN-induced double-strand breaks can result in targeted gene
knockout through nonhomologous end joining5 or in precise targeted alteration of genomic sequence through homology-directed
repair with an exogenous DNA template6,7. TALENs have been

used to manipulate genomes in a variety of organisms6,811 and

cell lines5,7,12,13.
Although TALENs do not cause widespread genomic off-target
modification1417, cleavage at off-target sites can result in unintended mutations at genomic loci. Whereas recent studies have
identified closely related off-target sites containing two or fewer
mismatches in zebrafish18 and in human cell lines13, more distantly related off-target sites are of particular interest because
one would expect a typical 36-bp target site to be approximately
eight or more mutations away from any sequence in the human
genome. In previous studies, two distant genomic off-target sites
have been verified to be cleaved in human cell lines from 19
potential off-target sites predicted using systematic evolution of
ligands by exponential enrichment (SELEX)7, an in vitro method
to identify binding sites of DNA-binding domains in isolation.
Only three off-target sites have been identified using an integrasedeficient lentiviral vectorbased approach19,20 to capture
off-target double-strand break sites in cells. The limited number
of off-target TALEN sites identified in previous studies suggests
that further research is needed both to better understand the
extent of TALEN-induced genomic off-target mutations and to
improve TALEN specificity to minimize these unwanted effects.
Principles that determine specificities of TALEN proteins
remain poorly characterized. Although SELEX experiments
and a high-throughput study of TALE activator specificity have
described the DNA-binding specificities of monomeric TALE
proteins5,7,9 and a single TALE activator21, respectively, the DNAcleavage specificities of active, dimeric nucleases can differ from
the specificities of their component monomeric DNA-binding
domains22. For example, zinc-finger nucleases, which are different
engineered dimeric nucleases, demonstrate compensation effects
between monomers22. Cellular methods to study off-target genomic
modification such as whole-genome sequencing or integrasedeficient lentiviral vectorbased capture can be complicated by
DNA accessibility, which varies from site to site and between cell

1Department of Chemistry and Chemical Biology, Harvard University, Cambridge, Massachusetts, USA. 2Howard Hughes Medical Institute, Harvard University,

Cambridge, Massachusetts, USA. 3Molecular Pathology Unit, Center for Cancer Research, and Center for Computational and Integrative Biology, Massachusetts General
Hospital, Charlestown, Massachusetts, USA. 4Department of Pathology, Harvard Medical School, Boston, Massachusetts, USA. Correspondence should be addressed to
D.R.L. (drliu@fas.harvard.edu).

Received 25 September 2013; accepted 11 January 2014; published online 16 february 2014; doi:10.1038/nmeth.2845

nature methods | VOL.11 NO.4 | APRIL 2014 | 429

Articles
a

RVD
RVD code

N-terminal
domain

TALE repeats

FokI cleavage domain

5
3

C-terminal domain
Canonical
Q3
Q7
28-aa

DNA library of
~1012 target sites

TALEN
expression
plasmid
In vitrocoupled
transcription and
translation (2 h)

L S R

L S R

L
L

S R

L S R

L S R

S R

L S R

L S R

S R

L S R

L S R

(i) Digestion with TALEN

(ii) Selection (2 d)
S

Blunting of the overhangs

Ligation of adaptor 1

L S R

L S R

L S R

L S R

L S R

R
S

Digestion with TALEN

Rolling-circle
amplification
(overnight)
L

L S R

2014 Nature America, Inc. All rights reserved.

L S R

Circularization
(overnight)
L

npg

L S R

PCR with two primers:

(i) adaptor 2constant sequence
5
(ii) adaptor 1

L S R

(i) High-throughput
sequencing (overnight)
(ii) Data analysis (1 d)

S R

S R

L S R

Amplicons containing cleaved library members

Gel purification
S

Specificity profile

S R

L S R

Figure 1 | TALEN architecture and selection scheme. (a) Schematic of a TALEN. The 12th and 13th amino acids (the RVD) of each TALE repeat recognize
a specific DNA base pair. Two different TALENs bind their corresponding half-sites, which allows FokI dimerization and DNA cleavage. The C-terminal
domain variants used in this study are shown at the bottom. (b) Outline of the selection. Single-stranded library of DNA oligonucleotides contain
partially randomized left half-site (L), spacer (S), right half-site (R) and constant region (thick black line). Double arrows represent concatameric repeats
of a DNA target site variants. (c) Selection of TALEN-cleaved library members by adaptor ligation and high-throughput DNA sequencing.

types23, or by DNA repair and integration pathways after cleavage

that could obscure the determination of intrinsic TALEN specificity. Purely cellular studies are also limited to the stochastic
handful of off-target sites in a given genome that are similar to the
target sequence and thus cannot be used to evaluate the ability of
TALENs to cleave a very large number of off-target sites necessary
for a broad and in-depth study of TALEN specificity.
Using a previously described in vitro selection method22,24, we
interrogated TALENs for their abilities to each cleave 1012 potential off-target DNA substrates related to their intended target
sequences. The resulting data are, to our knowledge, the first comprehensive profiles of TALEN cleavage specificities. The selection
results suggest a model in which excess nonspecific DNA-binding
energy gives rise to greater off-target cleavage relative to on-target
cleavage. Based on this model, we engineered TALENs with a
modified architecture and substantially improved specificities of
DNA cleavage in vitro. In human cells, these modified TALENs
exhibited 24-fold to more than 120-fold greater specificity for the
most readily cleaved off-target sites than currently used TALEN
constructs. Our results demonstrate four key findings: (i) TALENs
430 | VOL.11 NO.4 | APRIL 2014 | nature methods

are highly specific for their intended target base pair at 103 of
the 104 positions profiled, with specificity increasing near the
N-terminal TALEN end of each TALE repeat array (corresponding to the 5 end of the bound DNA); (ii) longer TALENs are
more specific in a genomic context, whereas shorter TALENs
have higher specificity per nucleotide; (iii) TALE repeats each
bind their respective base pairs relatively independently; and (iv)
excess DNA-binding affinity leads to increased TALEN activity
against off-target sites and therefore decreased specificity.
RESULTS
Specificity profiling of CCR5- and ATM-targeted TALENs
We profiled the specificities of 30 unique heterodimeric TALEN
pairs (hereafter referred to as TALENs) harboring different
C-terminal, N-terminal and FokI domain variants, and targeted to half-sites of various lengths. The number of base pairs
recognized by each half-site that we list in this paper includes
the 5 thymine (T) recognized by the N-terminal domain. Most
of the TALENs we tested were obligate heterodimers with FokI
Q105E,I118L in one TALEN and FokI E109K,I157K for the

Articles
a
25
20

35

CCR5A preselection
library
CCR5A TALEN
digestion

15
10

30
Sequences (%)

30
Sequences (%)

b
35

25
20

L18 CCR5A TALEN

R18 CCR5A TALEN

ATM preselection
library
ATM TALEN
digestion

15
10

L18 ATM TALEN

R18 ATM TALEN

0
0
10
5
Total number of mutations in half-sites

0
5
10
Total number of mutations in half-sites

2014 Nature America, Inc. All rights reserved.

other (EL/KK); we also used a more active heterodimeric variant with FokI Q105E,I118L,N115D in one TALEN and FokI
E109K,I157K,H156R in the other (ELD/KKR), and homodimeric
FokI nuclease domains, as specified below3,25.
We designed TALENs as previously reported12 to target one
of three distinct sequences, CCR5A, CCR5B or ATM, in two
different human genes, CCR5 and ATM (Supplementary Fig. 1).
We determined the specificity profiles using a previously described
in vitro selection method22,24. Briefly, we digested preselection
libraries of > 1012 DNA sequences, each theoretically containing at
least 10 copies of all possible DNA sequences with six or fewer mutations relative to the on-target sequence, with 3 nM to 40 nM of an
in vitrotranslated TALEN (Online Methods, Supplementary
Table 1, Supplementary Fig. 2 and Supplementary Results).
Cleaved library members harbored a free 5 monophosphate that
enabled them to be captured by adaptor ligation (Fig. 1b,c). We
isolated DNA fragments of length corresponding to 1.5 target
sites (an intact target site and a repeated half-site up to the
point of TALEN-induced DNA cleavage) by gel purification.
High-throughput sequencing and computational analysis of
TALEN-treated or control samples that survived this selection
revealed the abundance of all TALEN-cleaved sequences as
well as the abundance of the corresponding sequences before
selection (Supplementary Notes). In the control sample, all
members of the preselection library were cleaved by a restriction
endonuclease at a constant sequence to enable their capture by
adaptor ligation and isolation by gel purification. We calculated
the enrichment value for each library member that survived selection by dividing the abundance of its sequence after selection by
that before selection.
For all TALEN variants and under all tested conditions, the
DNA that survived the selection contained significantly fewer
mean mutations in the targeted half-sites than were present in
the preselection libraries (P < 107, Fishers exact test; Fig. 2a,b
and Supplementary Tables 2 and 3). For all selections, the
on-target sequences were enriched 8-fold to 640-fold with

e
Cleavage efficiency of discrete
sequence (relative to on-target)

Figure 2 | In vitro selection results. (a,b) Sequences surviving selection (TALEN digestion) compared
to preselection libraries for CCR5A TALENs (a) and ATM TALENs (b) with EL/KK FokI domains as a
function of the number of mutations in both half-sites (left and right half-sites combined excluding
the spacer). Each selection was performed once with more than 28,900 sequences analyzed per
selection. (c,d) Specificity scores for the CCR5A TALENs (c) and ATM TALENs (d) at all positions in
the target half-sites plus a single flanking position. A score of 0 indicates no specificity. Boxed data
represent the intended target base. For the right half-site, the R18 TALENs, data for the sense strand
are shown. (e) Correspondence between discrete in vitro TALEN cleavage efficiency (cleaved DNA as a
fraction of total DNA) for mutant sequences normalized to on-target sequence cleavage (value of 1)
versus their enrichment values in the selection normalized to the enrichment values of the on-target
sequence (value of 1) listed in Supplementary Figure 3. The Pearsons r = 0.90 between normalized
cleavage efficiency and normalized enrichment value.

1.0
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1.0

Right half-site

Left half-site

npg

Specificity
score

1.2
1.0
0.8
0.6
0.4
0.2
0
0

0.2
0.4
0.6
0.8
1.0
Enrichment value (relative to on-target)

1.2

a mean enrichment value of 110-fold (Supplementary Table 4).

To validate our selection results in vitro, we assayed the ability
of the CCR5B TALENs targeting 13-bp left and right halfsites (L13 + R13) to cleave each of 16 diverse off-target
substrates (Supplementary Fig. 3). The efficiencies correlated
well (r = 0.90) with the observed enrichment values from the
selection (Fig. 2e).
To quantify specificities of DNA cleavage at each position in the
TALEN target sites for all four possible base pairs, we calculated
specificity scores as the differences between preselection and
postselection base-pair frequencies, normalized to the maximum
possible change of the preselection frequency from complete
specificity (defined as 1.0) to complete antispecificity (defined
as 1.0). For all TALENs tested, the targeted base pair at every
position in both half-sites was preferred, with the sole exception
of the base pair closest to the spacer for some ATM TALENs at
the right half-site (Fig. 2c,d and Supplementary Figs. 49). The
5 T recognized by the N-terminal domain was highly specified,
and the 3 DNA end (targeted by the C-terminal TALEN end)
generally tolerated more mutations than the 5 DNA end; both of
these observations are consistent with previous reports26,27. All
12 of the positions targeted by the Asn-Asn (NN) RVDs in the
ATM and CCR5A TALENs were enriched for guanine, confirming
previous reports5,7,26,28 that the NN RVD specifies guanine.
TALEN off-target cleavage in cells
For TALENs that target 36 base pairs (bp), potential off-target
sites in the human genome are expected on average to contain
approximately eight or more mutations relative to the on-target
site (Supplementary Table 5), more mutations than theoretically
are covered in the in vitro selection. Therefore, we used a machine
learningbased classifier algorithm29 trained on the tens of
thousands of off-target sites revealed by the in vitro selection
to identify rare TALEN candidate off-target sites in the human
genome (Supplementary Results). Using this algorithm, we identified the 36 best-scoring heterodimeric candidate off-target sites
nature methods | VOL.11 NO.4 | APRIL 2014 | 431

Articles
predicted off-target genomic sites

Site
OnCCR5A
OffC-5
OffC-15
OffC-16
OffC-28
OffC-36
OffC-38
OffC-49
OffC-69
OffC-76

npg

2014 Nature America, Inc. All rights reserved.

Site
OnATM
OffA-1
OffA-11
OffA-13
OffA-16
OffA-17
OffA-23
OffA-35

No TALEN (%)
<0.006
<0.006
<0.020
<0.006
<0.009
<0.006
<0.006
<0.006
<0.010
<0.006
No TALEN (%)
0.007
<0.006
<0.006
<0.006
<0.006
<0.051
0.018
<0.006

CCR5A EL/KK CCR5A ELD/KKR

CCR5A
FokI (%)
FokI (%)
Homo FokI (%)
9.8
0.53
<0.014
<0.006
0.014
<0.006
ND
ND
ND
ND
ATM EL/KK
FokI (%)
6.8
<0.006
<0.006
0.008
<0.006
<0.14
<0.006
<0.006

28
2.3
0.23
0.031
0.16
0.15
ND
ND
ND
ND

47
2.3
0.043
<0.006
0.056
0.028
0.067
0.110
0.089
0.149

ATM ELD/KKR
ATM
FokI (%)
Homo FokI (%)
16
0.026
0.036
0.025
<0.006
<0.17
0.29
<0.006

18
0.077
0.39
<0.006
0.057
0.94
0.23
0.070

For cells treated with either no TALEN or CCR5A TALENs containing heterodimeric EL/KK,
heterodimeric ELD/KKR or the homodimeric (Homo) FokI cleavage domain variants, cellular
modification rates are shown as the percentage of observed insertions or deletions (indels)
consistent with TALEN cleavage relative to the total number of sequences for on-target (On)
and predicted off-target sites (Off). ND, no data were collected. Same as above for ATM
TALENs. Sample sizes and P values are given in Supplementary Tables 7 and 8.

for the ATM TALENs and 48 of the best-scoring candidate off-target

sites for the CCR5A TALENs (Supplementary Table 6). These
sites differed from the on-target sequence at 714 positions.
We amplified 76 of 84 predicted off-target sites for CCR5A and
ATM TALENs from genomic DNA purified from U2OS human
cells expressing either CCR5A or ATM TALENs12. We considered
sequences containing insertions or deletions of three or more base
pairs in the DNA spacer of the potential genomic off-target sites
and present in significantly greater numbers (P < 0.005 Fishers
exact test) in the TALEN-treated samples versus the untreated control sample to be TALEN-induced modifications. Consistent with
a previous report3, CCR5A or ATM TALENs containing ELD/KKR
and homodimeric FokI domains demonstrated increased on-target
activity compared to EL/KK FokI domains. Of the 45 amplified
CCR5A off-target sites, we identified nine off-target sites with
TALEN-induced modifications; likewise, of the 31 amplified ATM
off-target sites, we observed seven off-target sites with TALENinduced modifications (Table 1 and Supplementary Tables 7
and 8). Therefore, 16 of 76 total assayed off-target candidates
were accessible and modified by TALENs in cells. The inspection of modified on-target and off-target sites yielded a prevalence of deletions ranging from three to dozens of base pairs
(Supplementary Fig. 10), consistent with previously described
characteristics of TALEN-induced genomic modification30. We
also found that our approach outperformed TALENoffer31, a
Figure 3 | In vitro specificity as a function of TALEN length. Enrichment
value of on-target (zero mutation) and off-target sequences containing
16 mutations for CCR5B TALENs of varying TALE repeat array lengths with
EL/KK FokI domains. Each selection was performed once with more than
34,900 sequences analyzed per selection.
432 | VOL.11 NO.4 | APRIL 2014 | nature methods

purely computational approach for the identification of TALENinduced off-target substrates in cells (Supplementary Tables 7
and 8, Supplementary Fig. 11, and Supplementary Results).
Repeat-binding independence and effects of TALEN length
The extensive number of quantitatively characterized off-target
substrates in the selection data enabled us to address several
key questions about TALEN specificity. First, we assessed
whether mutations at one position in the target sequence
affect the ability of TALEN repeats to productively bind other
positions. We found that TALE repeats bound their respective
DNA base pairs independently, beyond a slightly increased
tolerance for adjacent mismatches (Supplementary Fig. 12 and
Supplementary Results).
The independent binding of TALE repeats simplistically
predicts that TALEN specificity per base pair is independent of
target-site length. To experimentally characterize the relationship between TALE array length and off-target cleavage, we constructed TALENs targeting 10 bp, 13 bp and 16 bp (including
the 5 T) for both the left (L10, L13 and L16) and right (R10,
R13 and R16) half-sites. We subjected TALENs representing all
nine possible combinations of left and right CCR5B TALENs
to in vitro selection. The results revealed that shorter TALENs
had greater specificity per targeted base pair than longer
TALENs (Supplementary Table 2). For example, sequences
cleaved by the L10 + R10 TALEN contained a mean of 0.032
mutations per recognized base pair, whereas those cleaved by
the L16 + R16 TALEN contained a mean of 0.067 mutations per
recognized base pair.
For selections with the longest CCR5B TALENs targeting 16
bp plus 16 bp or CCR5A and ATM TALENs targeting 18 bp plus
18 bp, the mean selection enrichment values did not follow a simple
exponential decrease as function of mutation number (Fig. 3 and
Supplementary Table 9). It is possible these TALENs had greater
affinity than is required to substantially bind and cleave the target
site (referred to hereafter as excess DNA-binding energy). Thus,
we hypothesize that excess DNA-binding energy from the larger
number of TALE repeats in longer TALENs reduces specificity by

102
Enrichment value
(post selection abundance pre selection abundance)

Table 1 | Cellular modification induced by TALENs at on-target and

101

100

L16 + R16
L16 + R13

101

L13 + R16
L16 + R10
L13 + R13
L10 + R16

102

L13 + R10
L10 + R13
L10 + R10
103

1
2
3
4
5
6
Total number of mutations in left and right half-sites

Articles
80
70
60
50
40
30
20
10
0

CCR5A TALENs

On-target site
enrichment value

On-target site
enrichment value

Canonical

Q3

Q7

28-aa

600

ATM TALENs

DNA

500

Left half-site

d
Right half-site

DNA

Left half-site

Right half-site

400
300
200
100
0

Canonical

Q3

Q7

1.0

1.0

npg

2014 Nature America, Inc. All rights reserved.

enabling the cleavage of sequences with more mutations, without a corresponding increase in the cleavage of sequences with
fewer mutations, because the latter are already nearly completely
cleaved. Indeed, the in vitro cleavage efficiencies of discrete DNA
sequences for these longer TALENs were independent of the presence of a small number of mutations in the target site (Fig. 4cf),
suggesting there was nearly complete binding and cleavage of
sequences containing few mutations. Likewise, higher TALEN
concentrations also resulted in decreased enrichment values of
sequences with few mutations and increased enrichment values
of sequences with many mutations (Supplementary Table 4).
These results together support a model in which excessive TALEN
binding arising from either long TALE arrays or high TALEN
concentrations decreases the observed TALEN DNA-cleavage
specificity for each recognized base pair. Despite the fact that
TALENs designed to cleave longer target sites are less specific per
base pair, this model predicts that such TALENs have higher overall specificity than those that target shorter sites, when considering the number of potential off-target sites in the human genome
(Supplementary Fig. 13 and Supplementary Results).
Engineering TALENs with improved specificity
The findings above suggest that TALEN specificity could be
improved by reducing non-specific DNA binding energy to
only support efficient on-target cleavage. We hypothesized that
reducing the cationic charge of the canonical 63-amino-acid
(aa) TALE C-terminal domain, which contains ten cationic residues4,5,7,9,10,12, would decrease nonspecific DNA binding32 and
improve the specificity of TALENs.
We constructed two variants in which we changed three (K788Q,
R792Q and R801Q; named Q3) or seven (K777Q, K778Q, K788Q,
R789Q, R792Q, R793Q and R801Q; named Q7) cationic arginine
or lysine residues in the canonical 63-aa C-terminal domain to
glutamine (Fig. 1a). We performed in vitro selections on CCR5A
and ATM TALENs containing the canonical C-terminal domain,
the engineered Q3 domain and the engineered Q7 domain as
well as a previously reported 28-aatruncated C-terminal
domain5 with a theoretical net charge (1) identical to that

Cleavage efficiency

Cleavage efficiency

Canonical
Canonical
Figure 4 | In vitro selection specificity and discrete
0.9
0.9
Q7
Q7
0.8
0.8
cleavage efficiencies of TALENs containing canonical or
0.7
0.7
engineered C-terminal domains. (a,b) On-target enrichment
0.6
0.6
values for selections of CCR5A TALENs (a) and ATM TALENs (b)
0.5
0.5
0.4
0.4
containing indicated domains with EL/KK FokI cleavage
0.3
0.3
domains (Supplementary Table 4a,b). 28-aa indicates
0.2
0.2
28-aatruncated variant. Each selection was performed
0.1
0.1
0
0
once with more than 4,622 sequences analyzed per
OnA A1 A2 A3 A4 A5 A6 A7 A8
OnC C1 C2 C3 C4 C5 C6 C7 C8
selection. (c) CCR5A on-target sequence (OnC) and
double-mutant sequences with mutations highlighted in red. (d) ATM on-target sequence (OnA), single-mutant sequences and double-mutant sequences
with mutations highlighted in red. (e,f) Discrete in vitro cleavage efficiency of DNA sequences listed in c with CCR5A TALENs (e) and of DNA sequences
listed in d with ATM TALENs (f) containing indicated domains with EL/KK FokI domains. Error bars, s.d. from three biological replicates. Average is shown
for C4 from two replicates. See Supplementary Results for P values.

of the Q7 C-terminal domain. The enrichment values of the

on-target sequence in the CCR5A and ATM selections increased
substantially as the net charge of the C-terminal domain decreased
(Fig. 4a,b). For example, the enrichment values of the on-target
sequences in ATM selections were 510, 50 and 20 for the Q7, Q3
and canonical 63-aa C-terminal domain variants, respectively.
Similarly, substituting one, two or three cationic residues in the
TALEN N terminus with glutamine also increased cleavage specificity (Supplementary Table 4, Supplementary Fig. 14 and
Supplementary Results). Consistent with the selection results,
TALENs containing Q7 C-terminal domains showed about fourfold or greater specificity of DNA cleavage in vitro for 11 of the 16
CCR5A and ATM off-target sites containing one or two mutations
(Fig. 4cf and Supplementary Results).
Improved specificity of engineered TALENs in human cells
To determine whether the increased specificity of the engineered
TALENs observed in vitro also occurs in human cells, we mea
sured TALEN-induced modification rates of the on-target and
top 36 predicted off-target sites for CCR5A and ATM TALENs
containing all six possible combinations of the canonical 63-aa,
Q3 or Q7 C-terminal domains and the EL/KK or ELD/KKR FokI
domains (12 TALENs total). We did not analyze TALENs containing a 28-aa C-terminal domain in these experiments because both
the ATM and CCR5A on-target sites have DNA spacer lengths of
18 bp, which is outside the 28-aa C-terminal domains preferred
DNA spacer length range (Supplementary Figs. 15 and 16, and
Supplementary Results). For both FokI variants, the TALENs
with Q3 C-terminal domains demonstrated substantial on-target
activities ranging from 8% to 24% modification, comparable to the
activity of TALENs with the canonical 63-aa C-terminal domains.
TALENs with canonical 63-aa or Q3 C-terminal domains and the
ELD/KKR FokI domain were both fivefold to ninefold more active
in modifying the CCR5A and ATM on-target site in cells than the
corresponding TALENs with the Q7 C-terminal domain (Fig. 5
and Supplementary Table 7).
Compared to the canonical 63-aa C-terminal domains, TALENs
with Q3 C-terminal domains demonstrated a mean increase in
nature methods | VOL.11 NO.4 | APRIL 2014 | 433

Articles

npg

2014 Nature America, Inc. All rights reserved.

on-target:off-target activity ratio of more than 12-fold and more

than ninefold for CCR5A and ATM sites, respectively, with the
ELD/KKR FokI domain (Fig. 5, and Supplementary Tables 7 and
10a,b). These mean improvements can only be expressed as lower
limits owing to the absence or near-absence of observed cleavage
events by the engineered TALENs for many off-target sequences.
For the ATM TALENs containing Q7 C-terminal domains, the
cleavage efficiency of both the on-target and off-target sites was
so low that we could not determine their specificity (Fig. 5 and
Supplementary Tables 7 and 10a,b). For the most abundantly
cleaved off-target site (CCR5A off-target site 5), the Q3 C-terminal
domain was 24-fold more specific, and the Q7 C-terminal domain
was >120-fold more specific (Fig. 5), than the canonical 63-aa
C-terminal domain. To determine whether the increased specificity of the engineered TALENs observed for CCR5A and ATM
TALENs applies more generally, we constructed three additional
TALENs, targeting sequences in the PMS2, SDHD and HDAC1
genes12, using the canonical 63-aa, Q3 or Q7 C-terminal domains
and ELD/KKR FokI domains. Of the 64 TALENs reported previously12, TALENs targeting these three genes had target sequences
with closely homologous genomic off-target sites containing
one to five mutations. For each of these TALENs, we measured
modification rates for genomic on-target and off-target sites.
PMS2, SDHD and HDAC1 TALENs with Q3 C-terminal domains
demonstrated on-target activities ranging from 6% to 28% modification, comparable to the activity of TALENs with the canonical
63-aa C-terminal domains (Supplementary Table 10c and 11).
Although the PMS2, SDHD and HDAC1 TALENs with Q3
C-terminal domains had similar on-target activity levels as
canonical TALENs, they demonstrated a fivefold to sevenfold
increase in on-target:off-target activity ratio. For the PMS2
TALENs, the Q7 C-terminal domains demonstrated a 53-fold
and 64-fold increase in on-target:off-target activity ratio in cells,
although as observed above, the Q7 TALENs were less active
on the target site than TALENs containing the canonical or Q3
C-terminal domains (Supplementary Tables 10c and 11).
DISCUSSION
The 16 confirmed TALEN off-target sites containing 812 mutations identified from the 76 predicted sites assayed in this study
represent more bona fide genomic off-target sites in the human
genome than have been revealed collectively to date by other
methods. These 16 sites were modified with 0.032.3% efficiency
in human cells, which demonstrated that TALENs can have appreciable off-target activities in human cells even at sites that are
eight or more mutations away from the on-target sequence. Site
accessibility in cells, influenced by histone proteins, transcription factors and DNA modification23, likely accounts for at least
some of the difference between our in vitro, computational and
cell-based results. We compared our method with other methods
434 | VOL.11 NO.4 | APRIL 2014 | nature methods

On-target site
Off-target site 5
On:off-target activity:

19

176 >8,760

12 284 >1,450

19 635 >576

100.00
Cellular modification
efficiency (%)

Figure 5 | Specificity of engineered TALENs in human cells. The cellular

modification efficiency of canonical and engineered TALENs expressed as a
percentage of insertions and/or deletions consistent with TALEN-induced
modification out of total sequences is shown for the on-target CCR5A site
and for CCR5A off-target site 5. On:off target activity, defined as the ratio
of on-target to off-target modification, is shown above each pair of bars.
Cellular modification experiments were performed once with more than
7,190 sequences analyzed per condition.

10.00
1.00
0.10

0.01
C-terminal domain:
Can. Q3 Q7
FokI domain:
Homo

Can. Q3 Q7
ELD/KKR

Can. Q3 Q7
EL/KK

for characterizing TALEN specificity and identifying genomic

TALEN off-target sites in Supplementary Discussion.
The observed decrease in specificity for TALENs with more
TALE repeats or more cationic residues in the C-terminal domain
or N terminus is consistent with a model in which excess TALEN
binding affinity leads to increased promiscuity. This excess DNAbinding energy model may explain reports that NN RVDs bind
either A or G (refs. 2,27,33). Those studies used TALE arrays of
more than 14 RVDs, which may have created a scenario in which
excess DNA-binding energy permits a suboptimal NN RVD interaction with A compared to G. We observed NN RVDs can discriminate between A and G, consistent with reports using shorter
TALE arrays of 13 RVDs28. Excess DNA-binding energy could
also explain the previously reported promiscuity at the 5 terminal
T of TALENs with longer C-terminal domains34 and is consistent with observations of higher TALEN protein concentrations
inducing more off-target cleavage9. Although decreasing TALEN
protein expression in theory could reduce off-target cleavage,
TALE arrays are reported with on-target DNA binding affinities
as high as 2.8 nM (dissociation constant, Kd)26, sufficient to
theoretically saturate target sites even when expressed at modest,
mid-nanomolar concentrations in a cell. The difficulty of improving the specificity of such TALENs by lowering their expression,
coupled with the need to maintain sufficient TALEN concentrations to effect desired levels of on-target cleavage, highlight the
value of engineering TALENs with higher intrinsic specificity.
Our findings suggest that mutant C-terminal domains with
reduced nonspecific DNA binding may be used to alter the
DNA-binding affinity of TALENs such that on-target sequences
are cleaved efficiently but with minimal excess DNA-binding
energy, which results in better discrimination between on-target
and off-target sites. As TALENs targeting up to 46 bp have
been shown to be active in cells14, it may be possible to further
improve specificity by engineering TALENs with a combination of
variant N-terminal and C-terminal domains that impart reduced
nonspecific DNA binding, a greater number of TALE repeats to
contribute additional on-target DNA binding and lower-affinity
RVDs such as the NK RVD to recognize G27,28. It is tempting to
speculate that the strategy of substituting residues that contribute
to nonspecific DNA binding to improve DNA specificity may also
apply to other proteins used for genome engineering, including
Cas9 and zinc-finger nucleases.
Our findings and the resulting improved TALENs would have
been difficult to generate using purely cellular off-target cleavage
methods. The ability of our profiling method to reveal the broad,
unobscured DNA-cleavage specificity of TALENs in the absence

Articles
of cellular complications enabled the elucidation of the inherent DNA-cleavage specificity of TALENs. The small number of
genomic sequences that are closely related to a target sequence
also intrinsically limits studies of cellular off-target cleavage. In
contrast, we evaluated each active, dimeric TALEN in this study
for its ability to cleave any of 1012 close variants of its on-target
sequence, a library size several orders of magnitude greater than
the number of different sequences in a mammalian genome. This
dense coverage of off-target sequence space enabled the elucidation of detailed relationships between DNA-cleavage specificity
and target base pair position, TALE repeat length, TALEN concentration, mismatch location and TALEN domain composition.
These results collectively reveal principles for characterizing and
improving TALENs with greater specificity that may enable a
wider range of genome-engineering applications.

npg

2014 Nature America, Inc. All rights reserved.

Methods
Methods and any associated references are available in the online
version of the paper.
Accession codes. Sequence Read Archive: SRP035232.
Note: Any Supplementary Information and Source Data files are available in the
online version of the paper.
Acknowledgments
J.P.G., V.P. and D.R.L. were supported by Defense Advanced Research Projects
Agency HR0011-11-2-0003 and N66001-12-C-4207, US National Institutes of
Health (NIH) NIGMS R01 GM095501 (D.R.L.), and the Howard Hughes Medical
Institute (HHMI). D.R.L. was supported as a HHMI Investigator. V.P. was
supported by award T32GM007753 from US National Institute of General
Medical Sciences. D.R., S.Q.T., J.D.S. and J.K.J. were supported by a NIH Director
Pioneer Award (DP1 GM105378). J.K.J. was supported by the Jim and Ann
Orr Massachusetts General Hospital Research Scholar Award. We thank M.L.
Maeder for preforming transfections and isolating genomic DNA, and C. Khayter
and M. Goodwin for technical assistance.
AUTHOR CONTRIBUTIONS
J.P.G., V.P., D.R., J.D.S. and S.Q.T. performed the experiments, designed
the research, analyzed the data and wrote the manuscript. J.K.J. and D.R.L.
designed the research, analyzed the data and wrote the manuscript.
COMPETING FINANCIAL INTERESTS
The authors declare competing financial interests: details are available in the
online version of the paper.
Reprints and permissions information is available online at http://www.nature.
com/reprints/index.html.
1. Moscou, M.J. & Bogdanove, A.J. A simple cipher governs DNA recognition
by TAL effectors. Science 326, 1501 (2009).
2. Boch, J. et al. Breaking the code of DNA binding specificity of TAL-type
III effectors. Science 326, 15091512 (2009).
3. Doyon, Y. et al. Enhancing zinc-finger-nuclease activity with improved
obligate heterodimeric architectures. Nat. Methods 8, 7479 (2011).
4. Cade, L. et al. Highly efficient generation of heritable zebrafish gene
mutations using homo- and heterodimeric TALENs. Nucleic Acids Res.
40, 80018010 (2012).
5. Miller, J.C. et al. A TALE nuclease architecture for efficient genome
editing. Nat. Biotechnol. 29, 143148 (2011).
6. Bedell, V.M. et al. In vivo genome editing using a high-efficiency TALEN
system. Nature 491, 114118 (2012).
7. Hockemeyer, D. et al. Genetic engineering of human pluripotent cells
using TALE nucleases. Nat. Biotechnol. 29, 731734 (2011).

8. Cermak, T. et al. Efficient design and assembly of custom TALEN and

other TAL effector-based constructs for DNA targeting. Nucleic Acids Res.
39, e82 (2011).
9. Tesson, L. et al. Knockout rats generated by embryo microinjection of
TALENs. Nat. Biotechnol. 29, 695696 (2011).
10. Moore, F.E. et al. Improved somatic mutagenesis in zebrafish using
transcription activator-like effector nucleases (TALENs). PLoS One 7,
e37877 (2012).
11. Wood, A.J. et al. Targeted genome editing across species using ZFNs and
TALENs. Science 333, 307 (2011).
12. Reyon, D. et al. FLASH assembly of TALENs for high-throughput genome
editing. Nat. Biotechnol. 30, 460465 (2012).
13. Mussolino, C. et al. A novel TALE nuclease scaffold enables high genome
editing activity in combination with low toxicity. Nucleic Acids Res. 39,
92839293 (2011).
14. Li, T. et al. Modularly assembled designer TAL effector nucleases for
targeted gene knockout and gene replacement in eukaryotes. Nucleic Acids
Res. 39, 63156325 (2011).
15. Ding, Q. et al. A TALEN genome-editing system for generating human stem
cell-based disease models. Cell Stem Cell 12, 238251 (2013).
16. Lei, Y. et al. Efficient targeted gene disruption in Xenopus embryos
using engineered transcription activator-like effector nucleases (TALENs).
Proc. Natl. Acad. Sci. USA 109, 1748417489 (2012).
17. Kim, Y. et al. A library of TAL effector nucleases spanning the human
genome. Nat. Biotechnol. 31, 251258 (2013).
18. Dahlem, T.J. et al. Simple methods for generating and detecting
locus-specific mutations induced with TALENs in the zebrafish genome.
PLoS Genet. 8, e1002861 (2012).
19. Gabriel, R. et al. An unbiased genome-wide analysis of zinc-finger
nuclease specificity. Nat. Biotechnol. 29, 816823 (2011).
20. Osborn, M.J. et al. TALEN-based gene correction for epidermolysis bullosa.
Mol. Ther. 21, 11511159 (2013).
21. Mali, P. et al. CAS9 transcriptional activators for target specificity
screening and paired nickases for cooperative genome engineering.
Nat. Biotechnol. 31, 833838 (2013).
22. 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, 765770 (2011).
23. Maeder, M.L. et al. Rapid open-source engineering of customized
zinc-finger nucleases for highly efficient gene modification. Mol. Cell
31, 294301 (2008).
24. Pattanayak, V. et al. High-throughput profiling of off-target DNA cleavage
reveals RNA-programmed Cas9 nuclease specificity. Nat. Biotechnol.
31, 839843 (2013).
25. Miller, J.C. et al. An improved zinc-finger nuclease architecture for highly
specific genome editing. Nat. Biotechnol. 25, 778785 (2007).
26. Meckler, J.F. et al. Quantitative analysis of TALE-DNA interactions suggests
polarity effects. Nucleic Acids Res. 41, 41184128 (2013).
27. Christian, M.L. et al. Targeting G with TAL effectors: a comparison
of activities of TALENs constructed with NN and NK repeat variable
di-residues. PLoS One 7, e45383 (2012).
28. Cong, L., Zhou, R., Kuo, Y.C., Cunniff, M. & Zhang, F. Comprehensive
interrogation of natural TALE DNA-binding modules and transcriptional
repressor domains. Nat. Commun. 3, 968 (2012).
29. Sander, J.D. et al. In silico abstraction of zinc finger nuclease cleavage
profiles reveals an expanded landscape of off-target sites. Nucleic Acids
Res. 41, e181 (2013).
30. Kim, Y., Kweon, J. & Kim, J.S. TALENs and ZFNs are associated with
different mutation signatures. Nat. Methods 10, 185 (2013).
31. Grau, J., Boch, J. & Posch, S. TALENoffer: genome-wide TALEN off-target
prediction. Bioinformatics 29, 29312932 (2013).
32. McNaughton, B.R., Cronican, J.J., Thompson, D.B. & Liu, D.R.
Mammalian cell penetration, siRNA transfection, and DNA transfection
by supercharged proteins. Proc. Natl. Acad. Sci. USA 106, 61116116
(2009).
33. Streubel, J., Blucher, C., Landgraf, A. & Boch, J. TAL effector
RVD specificities and efficiencies. Nat. Biotechnol. 30, 593595
(2012).
34. Sun, N., Liang, J., Abil, Z. & Zhao, H. Optimized TAL effector nucleases
(TALENs) for use in treatment of sickle cell disease. Mol. Biosyst. 8,
12551263 (2012).

nature methods | VOL.11 NO.4 | APRIL 2014 | 435

npg

2014 Nature America, Inc. All rights reserved.

ONLINE METHODS
Oligonucleotides, PCR and DNA purification. All oligonucleotides were purchased from Integrated DNA Technologies
(IDT). Oligonucleotide sequences are listed in Supplementary
Notes. PCR was performed with 0.4 l of 2 U/l Phusion Hot
Start II DNA polymerase (Thermo-Fisher) in 50 l with 1
HF buffer, 0.2 mM dNTP mix (0.2 mM dATP, 0.2 mM dCTP,
0.2 mM dGTP and 0.2 mM dTTP) (NEB), 0.5 M to 1 M of
each primer and a program of 98 C, 1 min; and 35 cycles of
(98 C, 15 s; 62 C, 15 s; 72 C, 1 min) unless otherwise noted.
Many DNA reactions were purified with a QIAquick PCR
Purification Kit (Qiagen), referred to below as Q-column
purification, or MinElute PCR Purification Kit (Qiagen) referred
to below as M-column purification.
TALEN construction. The canonical TALEN plasmids were
constructed by the fast ligation-based automatable solid-phase
high-throughput (FLASH) method12 with each TALEN targeting
1018 bp. Sequences encoding proteins with substitutions in
the N termini were cloned by PCR with Q5 Hot Start Master
Mix (NEB) (98 C, 22 s; 62 C, 15 s; 72 C, 7 min)) using phosphorylated TAL-N1fwd (for N1), phosphorylated TAL-N2fwd
(for N2), or phosphorylated TAL-N3fwd (for N3) and phos
phorylated TAL-Nrev as primers. 1 l DpnI (NEB) was added, and the
reaction was incubated at 37 C for 30 min and then subjected
to M-column purification. ~25 ng of eluted DNA was blunt-end
ligated intramolecularly in 10 l 2 Quick Ligase buffer, 1 l of
Quick Ligase (NEB) in a total volume of 20 l at room temperature
(~21 C) for 15 min. 1 l of this ligation reaction was transformed
into Top10 chemically competent cells (Invitrogen). Sequences
encoding proteins with C-terminal domain substitutions were
cloned by PCR using TAL-Cifwd and TAL-Cirev primers, and
then Q-columnpurified. ~1 ng of this eluted DNA was used as
the template for PCR with TAL-Cifwd and either TAL-Q3 (for Q3)
or TAL-Q7 (for Q7) for primers and then Q-columnpurified.
~1 ng of this eluted DNA was used as the template for PCR with
TAL-Cifwd and TAL-Ciirev for primers, and then Q-column
purified. ~1 g of this DNA fragment was digested with HpaI
and BamHI in 1 NEBuffer 4 and cloned22 into ~2 g of desired
TALEN plasmid pre-digested with HpaI and BamHI. TALENs
containing the N1, N2 and N3 N-terminal variant domains
and TALENs containing the canonical, Q3 and Q7 C-terminal
domains are available from Addgene (5143851449). Protein
sequences are listed in Supplementary Notes.
In vitro TALEN expression. TALEN proteins, all containing a
3 Flag tag, were expressed by in vitro transcription-translation.
800 ng of TALEN-encoding plasmid or no plasmid (empty lysate
control) was added to an in vitro transcription-translation reaction using the TNT Quick Coupled Transcription-Translation
System, T7 Variant (Promega) in a final volume of 20 l at 30 C
for 1.5 h. Western blots were used to visualize protein using
1 l of anti-Flag M2 monoclonal antibody (Sigma-Aldrich, SKU
F3165). TALEN concentrations were calculated by comparison to
standard curve of 1 ng to 16 ng N-terminally Flag-tagged bacterial
alkaline phosphatase (Sigma-Aldrich).
In vitro selection for DNA cleavage. Preselection libraries
were prepared with 10 pmol of oligo libraries containing
nature methods

partially randomized target half-site sequences (CCR5A, ATM or

CCR5B) and fully randomized 10-bp to 24-bp spacer sequences
(Supplementary Notes). Oligonucleotide libraries were separately circularized by incubation with 100 units of CircLigase II
ssDNA ligase (Epicentre) in 1 CircLigase II Reaction buffer
(33 mM Tris-acetate, 66 mM potassium acetate and 0.5 mM
dithiothreitol, pH 7.5) supplemented with 2.5 mM MnCl2 in
20 l total for 16 h at 60 C then incubated at 80 C for 10 min.
2.5 l of each circularization reaction was used as a substrate
for rolling-circle amplification at 30 C for 16 h in a 50-l
reaction using the Illustra TempliPhi 100 Amplification Kit
(GE Healthcare). The resulting concatemerized libraries were
quantified with Quant-iT PicoGreen dsDNA Kit (Invitrogen),
and libraries with different spacer lengths were combined in an
equimolar ratio.
For selections on the CCR5B sequence libraries, 500 ng of
preselection library was digested for 2 h at 37 C in 1 NEBuffer 3
with in vitro transcribed/translated TALEN plus empty lysate
(30 l total). For all CCR5B TALENs, concentrations of in vitro
transcribed-translated TALENs were quantified by western blot
(during the blot, TALENs were stored for 16 h at 4 C) and then
TALEN was added to 40 nM final concentration per monomer. For
selections on CCR5A and ATM sequence libraries, the combined
preselection library was further purified in a 300,000 molecular
weight cutoff spin column (Sartorius) with three 500-l washes
in 1 NEBuffer 3. 125 ng of preselection library was digested for
30 min at 37 C in 1 NEBuffer 3 with a total 24 l of fresh in vitro
transcribed-translated TALENs and empty lysate. For all CCR5A
and ATM TALENs, 6 l of in vitro transcription/translation
left TALEN and 6 l of right TALEN were used, corresponding
to a final concentration in a cleavage reaction ranging from
14 nM to 18 nM for CCR5A TALENs or from 10.5 nM to 13.5 nM
for ATM TALENs. These TALEN concentrations were quantified
by western blot performed in parallel with digestion.
For all selections, the TALEN-digested library was incubated
with 1 l of 100 g/l RNase A (Qiagen) for 2 min and then
Q-columnpurified. 50 l of purified DNA was incubated with
3 l of 10 mM dNTP mix (10 mM dATP, 10 mM dCTP, 10 mM
dGTP and 10 mM dTTP) (NEB), 6 l of 10 NEBuffer 2 and
1 l of 5 U/l Klenow Fragment DNA polymerase (NEB) for
30 min at room temperature and Q-columnpurified. 50 l of
the eluted DNA was ligated with 2 pmol of heated and cooled
#1 adaptors containing barcodes corresponding to each sample
(selections with different TALEN concentrations or constructs;
Supplementary Notes). Ligation was performed in 1 T4 DNA
ligase buffer (50 mM Tris-HCl, 10 mM MgCl 2, 1 mM ATP and
10 mM DTT, pH 7.5) with 1 l of 400 U/l T4 DNA ligase (NEB)
in 60 l total volume for 16 h at room temperature, and then
Q-columnpurified.
6 l of the eluted DNA was amplified by PCR in 150 l total
reaction volume (divided into three 50-l reactions) for 14 to 22
cycles using the #2A adaptor primers (Supplementary Notes).
The PCR products were purified by Q column. Each DNA sample
was quantified with Quant-iT PicoGreen dsDNA Kit (Invitrogen)
and then pooled into an equimolar mixture. 500 ng of pooled
DNA was run a 5% TBE 18-well Criterion PAGE gel (Bio-Rad) for
30 min at 200 V and DNAs of length ~230 bp (corresponding to
1.5 target site repeats plus adaptor sequences) were isolated and
purified by Q column. ~2 ng of eluted DNA was amplified by PCR
doi:10.1038/nmeth.2845

npg

2014 Nature America, Inc. All rights reserved.

for 58 cycles with #2B adaptor primers (Supplementary Notes)

and purified by M column.
10 l of eluted DNA was purified using 12 l of AMPure XP
beads (Agencourt) and quantified with an Illumina/Universal
Library Quantification Kit (Kapa Biosystems). DNA was prepared
for high-throughput DNA sequencing according to Illumina
instructions and sequenced using a MiSeq DNA Sequencer
(Illumina) using a 12 pM final solution and 156-bp pairedend reads. To prepare the preselection library for sequencing,
the preselection library was digested with 14 l of appropriate
restriction enzyme (CCR5A, Tsp45I; ATM, Acc65I; CCR5B, AvaI
(NEB)) for 1 h at 37 C then ligated as described above with
2 pmol of heated and cooled #1 library adaptors. Preselection
library DNA was prepared as described above using #2A library
adaptor primers and #2B library adaptor primers in place of
#2A adaptor primers and #2B adaptor primers, respectively
(Supplementary Notes). The resulting preselection library DNA
was sequenced together with the TALEN-digested samples.
Discrete in vitro TALEN cleavage assays. Discrete DNA substrates for TALEN digestion were constructed by combining
pairs of oligonucleotides as specified in Supplementary Notes
with restriction cloning22 into pUC19 (NEB). Corresponding
cloned plasmids were amplified by PCR (59 C annealing for
15 s) for 24 cycles with pUC19Ofwd and pUC19Orev primers (Supplementary Notes) and Q-columnpurified. 50 ng of
amplified DNAs were digested in 1 NEBuffer 3 with 3 l each of
in vitrotranscribed-translated TALEN left and right monomers
(corresponding to a ~16 nM to ~12 nM final TALEN concentration), and 6 l of empty lysate in a total reaction volume of 120 l.
The digestion reaction was incubated for 30 min at 37 C, then
incubated with 1 l of 100 g/l RNase A (Qiagen) for 2 min and
purified by M column. The entire 10 l of eluted DNA with glycerol added to 15% was analyzed on a 5% TBE 18-well Criterion
PAGE gel (Bio-Rad) for 45 min at 200 V, then stained with
1 SYBR Gold (Invitrogen) for 10 min. Bands were visualized
and quantified on an AlphaImager HP (Alpha Innotech).
Cellular TALEN cleavage assays. TALENs were cloned into
mammalian expression vectors12, and the resulting TALEN vectors transfected into U2OS-EGFP cells, a clonal U2OS human
cell line with an integrated construct that constitutively expresses
an EGFP-PEST fusion protein, as previously described 12.
Genomic DNA was isolated after 2 d as previously described 12.
For each assay, 50 ng of isolated genomic DNA was amplified by PCR (98 C, 15 s 67.5 C, 15 s; 72 C, 22s) for 35 cycles
with pairs of primers with or without 4% DMSO as specified
in Supplementary Notes. Two PCR reactions were performed

doi:10.1038/nmeth.2845

for OffC-5 to improve the limit of detection. The relative dsDNA

content of the PCR reaction for each genomic site was quantified with Quant-iT PicoGreen dsDNA Kit (Invitrogen) and
then pooled into an equimolar mixture, keeping no-TALEN and
all TALEN-treated samples separate. DNA corresponding to
150350 bp was purified by PAGE as described above.
44 l of eluted DNA was incubated with 5 l of 1 T4 DNA
ligase buffer and 1 l of 10 U/l polynucleotide kinase (NEB) for
30 min at 37 C and Q-columnpurified. 43 l of eluted DNA was
incubated with 1 l of 10 mM dATP (NEB), 5 l of 10 NEBuffer 2,
and 1 l of 5 U/l DNA Klenow fragment (3 5 exo; NEB)
for 30 min at 37 C and purified by M column. 10 l of eluted DNA
was ligated as above with 10 pmol of heated and cooled genomic
(G) adaptors (Supplementary Notes), and purified by Q column.
8 l of eluted DNA was amplified by PCR for 68 cycles with G-B
primers containing barcodes corresponding to each sample. Each
sample DNA was quantified with Quant-iT PicoGreen dsDNA
Kit (Invitrogen) and then pooled into an equimolar mixture. The
combined DNA was subjected to high-throughput sequencing
using MiSeq as described above.
Data analysis. Illumina sequencing reads were filtered and parsed
with scripts written in Unix Bash as outlined in Supplementary
Notes. DNA sequences are available upon request. Source code
is available as Supplementary Software. Specificity scores were
calculated as previously described22. Sample sizes for sequencing
experiments were maximized (within practical experimental considerations) to ensure greatest power to detect effects. Statistical
analysis of the distribution of number of mutations in various
TALEN selections in Supplementary Table 2 was performed
as previously described22. Statistical analysis of TALEN modified genomic sites in Supplementary Tables 7, 8 and 10 was
performed as previously described29 with multiple comparison
correction using the Benjamini-Hochberg method35,36.
To determine extrapolated mean enrichment curves mutation
enrichment value as function of mutation number were fit to
an exponential function, a eb, with R2 reported using the
nonlinear least-squares method. The a, b and R2 values and
the mutation range for these fits are reported in Supplementary
Table 12. The exponential decrease, b, was used to extrapolate
all mean enrichment values beyond five mutations to determine
the extrapolated mean enrichment.
35. Benjamini, Y. & Hochberg, Y. Controlling the false discovery ratea
practical and powerful approach to multiple testing. J. Royal Stat. Soc. B
57, 289300 (1995).
36. Noble, W.S. How does multiple testing correction work? Nat. Biotechnol.
27, 11351137 (2009).

nature methods