Directed evolution of sulfotransferases and paraoxonases by ancestral libraries

Article abstract of DOI:10.1016/j.jmb.2011.06.037

Large libraries of randomly mutated genes are applied in directed evolution experiments in order to obtain sufficient variability. These libraries, however, contain mostly inactive variants, and the very low frequency of improved variants can only be isol

Full text of DOI:10.1016/j.jmb.2011.06.037

doi:10.1016/j.jmb.2011.06.037
J. Mol. Biol. (2011) 411, 837–853
Contents lists available at www.sciencedirect.com
Journal of Molecular Biology
journal homepage: http://ees.elsevier.com.jmb
Directed Evolution of Sulfotransferases and Paraoxonases
by Ancestral Libraries
Uria Alcolombri, Mikael Elias⁎ and Dan S. Tawfik⁎
Department of Biological Chemistry, Weizmann Institute of Science, Rehovot 76100, Israel
Received 7 April 2011;
received in revised form
Large libraries of randomly mutated genes are applied in directed evolution
experiments in order to obtain sufficient variability. These libraries,
however, contain mostly inactive variants, and the very low frequency of
improved variants can only be isolated by high-throughput screening.
Small but efficient libraries comprise an attractive alternative. Here, we
describe the application of ancestral libraries—libraries based on mutations
predicted by phylogenetic analysis and ancestral inference. We designed
and constructed such libraries using serum paraoxonases and cytosolic
sulfotransferases (SULTs) as model enzymes. Both of these enzyme families
exhibit a range of activities in drug metabolism and detoxification of
xenobiotics. The ancestral serum paraoxonase and SULT libraries were
screened by low-throughput means, including HPLC, using substrates
and/or reactions with which all family members exhibit low activity. The
libraries showed a remarkably high frequency of highly polymorphic and
functionally diverse variants. Screening of as few as 300 variants enabled
the isolation of mutants with up to 50-fold higher activity than the starting
point enzyme. Structural and kinetic characterizations of an evolved SULT
variant show how few ancestral mutations reshaped the active site and
modulated the enzyme's specificity. Ancestral libraries therefore comprise a
means of focusing diversity to positions and mutations that readily trigger
changes in substrate and/or reaction specificity, thereby facilitating the
isolation of new enzyme variants for a variety of different substrates and
reactions by medium-throughput or even low-throughput screens.
1
4 June 2011;
accepted 20 June 2011
Available online
2
4 June 2011
Edited by J. Karn
Keywords:
serum paraoxonase (PON);
cytosolic sulfotransferases;
SULT;
directed evolution;
protein phylogeny
©
2011 Elsevier Ltd. All rights reserved.
*
Corresponding authors. E-mail addresses:
Mikael.Elias@weizmann.ac.il; tawfik@weizmann.ac.il.
Abbreviations used: SULT, cytosolic sulfotransferase;
REAP, Reconstructed Evolutionary Adaptive Path; PON,
serum paraoxonase; TBBL, 5-thiobuyl-γ-butyrolactone;
TBL, γ-thiobutyrolactone; OP, organophosphate; CMP,
-cyano-4-methyl-7-coumaryl-cyclohexyl-ester; DEPCYC,
diethylphosphoryl-3-cyano-4-methylcoumarin; 3NPA,
-nitrophenyl acetate; 4AAP, 4-acetoxy-acetophenone;
IMP, O-isopropyl-O-(p-nitrophenyl) methyl phosphonate;
PAPS, 3-phosphoadenosine 5-phosphosulfate; pNP,
-nitrophenol; 3CyC, 3-cyanoumbelliferone; BPA,
bisphenol A; L-thyroxine, 3,5,3′,5′-tetraiodothyronine;
Introduction
Directed evolution is traditionally performed with
large libraries created by random mutagenesis.
Although large libraries are essential for many tasks,
small but effective libraries are of crucial importance,
especially when complex or difficult-to-assay activi-
3
1–7
3
ties are needed.
In random libraries, deleterious
mutations are most common, with a frequency of
8
≥0.33, whereas beneficial mutations occur at fre-
−
3 9,10
4
quencies around 10
.
Thus, as mutations accu-
mulate, most of the library genes become ‘nonviable,’
9
PDB, Protein Data Bank; PEG, polyethylene glycol; PAP,
-phosphoadenosine 5-phosphate; HPLC, high pressure
liquid chromatography.
as they do not encode folded functional proteins.
1
1
6
3
Methods such as family shuffling, SCHEMA,
2
iterative saturation mutagenesis, simultaneous
0022-2836/$ - see front matter © 2011 Elsevier Ltd. All rights reserved.

838
Directed Evolution with Ancestral Libraries
3
,12–14
multiple-site saturation mutagenesis,
consensus
Results
15
libraries, and Reconstructed Evolutionary Adaptive
Path (REAP) were developed with the aim of
minimizing the explored sequence space while max-
imizing functional variability.
4
Designing ancestral PON libraries
Here, we describe ancestral libraries that, similarly
to other methods such as family shuffling, aim at
exploring new combinations of substitutions that
have already been explored by nature and proven
functional. However, the mutational diversities
explored here are borrowed from the evolutionary
past and not from the present. The ancestral libraries
described here are composed of active-site sub-
stitutions that were predicted to have existed at
various nodes and branches of the evolutionary
trajectories of a given enzyme family. These ances-
tral forms are long gone, but their sequences can be
The library design strategy is schematically
described in Fig. 1.
Our first test case involved PONs, a family that
includes the mammalian families PON1, PON2, and
2
1
PON3 sharing 55–85% amino acid identity. PONs
are calcium-dependent hydrolases that catalyze the
hydrolysis of a broad range of substrates such as
esters, lactones, and phosphoesters. However, their
primary substrates are lactones, and the remaining
2
2
activities are promiscuous. The different families
differ in their lactonase specificity (PON1 and
PON3's best substrates are lipophilic lactones,
whereas PON2's best substrate is a homoserine
quorum-sensing lactone) and in their promiscuous
1
6
reconstructed from the phylogenetic tree. The
libraries explored all residues that are located close
to, or within, the enzyme's active site and that differ
between the starting point enzyme and the ancestor
2
2–24
activities.
Sequences of PON family members
were collected. To increase the fidelity of prediction,
we removed sequences with partial lengths and
sequences in which PON's characteristic sequence
motifs were missing. The alignment was fine-tuned
manually to improve its reliability at gap positions
(the alignment is provided as Supplementary Fig. 1).
The alignment was used to create a phylogenetic
tree, and a bootstrapping analysis was performed to
evaluate it (Supplementary Fig. 2). The tree was
checked manually to assess its validity (clustering of
known PON family members, compliance with the
tree of life, etc.). It should be noted that most of these
steps (removing sequences that might be irrelevant,
refining the alignment, and assessing the tree's
quality) are manual and family specific. A standard
methodology cannot be described, and the only
general rule is to run several iterative rounds of
alignment and tree making until the tree has
converged to its most convincing format.
(
or ancestors) along the trajectories that gave rise to
this enzyme. Some of these substitutions are unique,
but many of these are seen in certain contemporary
family members. However, the libraries included
only predicted ancestral substitutions, and not the
entire diversity seen in existing family members.
Random combinations of these ancestral mutations
were generated by spiking synthetic oligonucleo-
tides at the background of a starting point gene that
1
7
encodes an existing family member. Two exam-
ples are described: mammalian serum paraoxonase
(
(
PON) 1 and human cytosolic sulfotransferase
SULT) 1A1. The PON1 case provided a basis for
comparison, as we have described the directed
evolution of this enzyme by random and targeted
mutageneses using high-throughput or even ultra-
high-throughput screens,
libraries and medium-throughput screens.
1
8,19
as well as neutral drift
1,20
In
contrast, SULT1A1 comprises a new case; to our
knowledge, only one directed evolution experiment
has been described with this enzyme or with any
other SULT (Amir Aharoni et al., submitted for
publication).
Once the tree had been finalized, ancestral pre-
dictions were obtained for all nodes of the tree, and
the most probable ancestral sequences were deter-
mined as described in Materials and Methods. To
maximize the potential for changes in enzymatic
specificity, we considered for library diversification
all positions residing within 12 Å of the catalytic
calcium that resides at the bottom of PON1's active
site and all other positions of the active-site wall.
This list included 94 positions, or 26% of the
enzyme's total length.
The first library was based on the ancestral node
N8. This ancestor combines all three mammalian
PON families (PON1, PON2, and PON3) and
chicken PON, whose specificity is unknown. N8
comprises the deepest ancestor in this phylogeny
(i.e., most ancient) for which the reliability of
prediction is still high. Later, the same process was
applied on a deeper node, N6—the ancestor for all
vertebrate PONs for which the prediction's fidelity is
Here, we provide a generic protocol for the
preparation of ancestral libraries, with the aim of
generating ‘first-generation’ mutants that exhibit
improved activity with the desired target sub-
strate/reaction. Such variants could then be
subjected to standard directed evolution tech-
niques to obtain higher activity. We show that
ancestral libraries comprise a relatively small array
of variants that are, on average, more active and
evolvable than variants obtained by random
mutagenesis. Structural and kinetic characteriza-
tions of one of the evolved SULT1A1 mutants
were also performed, thus indicating how few
ancestral mutations can modulate an enzyme's
active site.

Directed Evolution with Ancestral Libraries
839
Fig. 1. A schematic representation of the ancestral library methodology. (a) Orthologues and paralogues of the target
enzyme are identified, and their sequences are aligned. (b) A phylogenetic tree is generated. The ancestral nodes are
assigned, and their sequences are predicted. The reference node (or nodes) for the library design (highlighted in red)
typically encompasses sequences of several paralogous families that exhibit ≥50% sequence identity. (c) The diversified
library positions are determined based on the enzyme's three-dimensional structure. These positions (yellow) include all
positions of the active-site wall and any other position within 12 Å of the enzyme's catalytic center (e.g., the catalytic metal
or a key catalytic residue). (d) The sequence of the chosen ancestral node (N8) is compared to that of the starting enzyme
(
WT). In most positions, the same amino acid is present (e.g., position 102). At positions where the sequence diverged,
ancestral substitutions are identified (e.g., A54X, etc.). (e) The ancestral substitutions are incorporated by mutagenesis
oligonucleotides that are combinatorially incorporated into the target enzyme's gene by assembly PCR. (f) The resulting
gene library is cloned, transformed into E. coli, and screened to identify variants with the desired specificity.
much lower. Of the 94 positions considered, 21
positions differed between the starting gene and the
N8 ancestor, and 27 positions differed with respect
to the N6 ancestor. In the N8 library, all the ancestral
substitutions could be identified in at least one
contemporary family member. In contrast, in the N6
library, 9 out of 27 ancestral substitutions do not
appear in any mammalian PON sequence (Table 1
and Fig. 2; Supplementary Table 1).
express in E. coli in a soluble and functional form).
The recombinant variant also exhibits higher stabil-
2
5
ity than human PON1 and may thus tolerate a
2
6
larger number and variety of mutations.
The N8-PON library
The N8-PON library was constructed by assembly
PCR using DNase-I-generated fragments of the
rePON1 gene and an equimolar mixture of 21
mutagenesis oligonucleotides, each encoding one
ancestral substitution (Supplementary Table 2). The
library was cloned into an expression vector. Sequenc-
ing of randomly picked clones indicated 4±2 ancestral
substitutions per gene plus 0.1±0.3 mutations that
were randomly incorporated during the PCR.
As our starting gene, we choose rePON1-G3C9
(
dubbed here rePON1), an Escherichia-coli-expressed
PON1 variant whose sequence is ∼95% identical
with rabbit PON1 and has enzymatic specificity that
is essentially identical with rabbit and human
2
0
PON1. We chose the engineered rePON1 rather
than a wild-type PON1 because we needed a
bacterially expressed protein for the library screens
To explore the viability and diversity of the N8
ancestor library, we screened a small number of
(human PON1 and other mammalian PON1s do not

840
Directed Evolution with Ancestral Libraries
Table 1. Diversified positions in the PON ancestral libraries
Position rePON1 (library starting point)
PON1
PON2
PON3
N8 mammalian+chick ancestor N6 vertebrate ancestor
5
5
6
7
7
7
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
3
3
5
L
E
S
I/L
D/E
I
I
D
I
D
S
L
K
A
H
Q
F
I
a
6
D
T
7
S/A/T
I/L/V
I/K/M
Q/N/D/E
D/L/H
S/A/V/Y
P/F/Q
Y/T
S/V
L
K/H
A
H/E
Q/F
P/F
F/I
Y/F
L
S/T/N
M/L
P/V
A
H/N
M
F/S/Y
L/V/F
F/L/T
L/F
I/M/L
F/I/V
S/C/N
S
T
L
P
4
I
5
M
D
G
S
P
Y
S
a
8
S
36
37
89
90
93
94
96
97
21
22
37
39
40
41
44
71
72
82
91
93
94
29
32
H
a
H
G
I
a
I
F
a
S/Q
W/L
V/I/M/L M/T
M
W
M
H
D
F
L
M
Y
Y
S
V
D
I
L
N
L
S
T
L
F
L
V
Y/H
D/Y
F/S/Y
I/V
E/D
L/I/V
L/M
K/T/R
L/I
S/F
V/T
I/L
F/Y/S
Y
Q/H
T/S
Y
Y/D
S
V/I
D
I
L
N/E
L
S
F
Y
F
I
V/I
D
V
E
L
L
K
I
D
V/A
S/T/A/F
N
L/I
L/F
N
I
L
a
S
V
I
Y
Y
Q
T
S/T
A
E
a
V/T
L
I
L
L
V/Y
D/Y
Q
N/I/M
Y
M
a
Y
Q
S
H
a
Q
T/S
I
S
S
Listed are all positions that differ between the starting point (rePON1) and the N8 or N6 ancestors, and the amino acids that occupy these
positions in the three mammalian PON families. Positions that differ between N8 and N6 are marked in boldface.
a
Unique ancestral substitutions (i.e., substitutions to amino acids that do not appear in paralogous families).
Fig. 2. Locations of the diversified library positions. (a) Backbone representation of rePON1. Marked in color are all
positions considered for diversification: the active-site wall and any position within 12 Å of the catalytic calcium (top
green sphere). The diversified positions in the N8 and N6 ancestral library positions are shown in blue. The remaining
positions (yellow) are those in which the ancestral sequence was identical with that of rePON1. (b) Backbone
representation of human SULT1A1. The diversified positions in the ancestral library positions are shown in blue, and
those that do not differ between 1A1 and the ancestor are shown in yellow. The substrate (p-nitrophenol) and the sulfate
donor's product (PAP) are shown in green.

Directed Evolution with Ancestral Libraries
841
randomly picked variants (276 clones) for six
different substrates. These substrates represent
three different hydrolytic activities: lactones [5-
thiobuyl-γ-butyrolactone (TBBL) and γ-thiobutyro-
lactone (TBL)], organophosphates (OPs) [methyl
phosphonic acid 3-cyano-4-methyl-7-coumaryl-
cyclohexyl-ester (CMP) and diethylphosphoryl-3-
cyano-4-methylcoumarin (DEPCYC)], and aryl es-
ters [3-nitrophenyl acetate (3NPA) and 4-acetoxy-
acetophenone (4AAP)] (Supplementary Fig. 3). A
control library that contained, on average, 3.2±2.7
random mutations per gene was generated by error-
prone PCR and tested for comparison (Fig. 3).
For the six substrates tested, ≥50% of the N8
library variants exhibited detectable activity (≥0.2-
fold activity relative to the starting point rePON1),
with each of the six substrates tested (Table 2).
Moreover, 3.6–22.4% of the screened variants
exhibited ≥2-fold higher activity than the wild-
type-like rePON1 with one or more of the tested
substrates. Furthermore, 1.3–10.7% variants exhib-
ited ≥5-fold higher activity compared to the wild-
type-like rePON1.
chromogenic lactone TBBL, a large portion of the
library variants (50.8%) exhibited activities that are
essentially identical with those of the wild-type-like
rePON1. In addition, none of the tested N8 library
variants exhibited significantly improved TBBLase
activity (N2-fold higher than rePON1), and only few
improved variants were found in the N6 library
(1.4%). This stands in contrast to 3.6–22.4% of
significantly improved variants observed with all
other substrates. Furthermore, the highest improve-
ment in TBBL seen among 552 screened variants
taken from both the N8 and N6 libraries was 3-fold,
relative to up to 52-fold improvements with the
other substrates. This trend relates to the fact that
TBBL was designed to mimic lipophilic γ-lactones
and thus reflects the native activity of PON1 and, to
a large degree, of all PONs. The rarely observed
improvements in TBBLase activity suggest that N8,
and even the more the ancient ancestor N6,
exhibited TBBLase activity levels similar to those
of contemporary PON1.
Library variants exhibiting the highest changes in
activity were sequenced and recloned to validate
their activity. These variants were screened with two
additional organophosphate substrates: O-isopro-
pyl-O-(p-nitrophenyl) methyl phosphonate (IMP)
and paraoxon. The most active clones that emerged
from this screen can be divided into three major
groups (Table 3): the first group (variants 2G1, 1B10,
1D2, 3D7, 1E11, and 2F6) showed higher activity
than wild-type PON1 towards organophosphate
2
7
Differences in lysate activities stem from differ-
ences in specific activities and expression levels.
Expression levels did vary between library mem-
bers, but to a much lower degree than specific
activities (see the text below). That most differences
can be ascribed to differences in specific activities is
also manifested in different activities varying to
different degrees. For lactonase activity with the
Fig. 3. Activity changes in the ancestral libraries compared to the random mutagenesis library. The libraries were
screened with six different substrates: (1) TBBL, (2) TBL, (3) CMP, (4) DEPCYC, (5) 3NPA, and (6) 4AAP. The color bar
represents activity compared to the starting point rePON1. The black background denotes activity levels that are lower
than those of rePON1. The red-to-white scale denotes equal or higher activity relative to rePON1, whereby white denotes
≥
10-fold higher activity.

842
Directed Evolution with Ancestral Libraries
Table 2. Percentage of active variants in the N8 and N6 ancestral libraries and in the random mutagenesis library
TBBL TBL
N8 N8
a
a
Relative activity
N6
RM
Relative activity
N6
RM
c
Actb0.2
43.5±4.8
55.1±4.1
1.4±0.7
49.2±6.2
50.8±6.2
0.0±0.0
60
40
0
Actb0.2
2NAct≥0.2
Act≥2
11.2±6.5
80.4±7.6
8.4±1.3
22.1±12.7
73.6±11.7
4.3±1.1
16
84
0
b
2
NAct≥0.2
Act≥2
CMP
N8
DEPCYC
N8
a
a
Relative activity
N6
RM
Relative activity
N6
RM
b
Actb0.2
56.2±3.5
38.8±2.3
5.0±1.3
31.2±5.3
46.4±0.7
22.4±5.1
65
34
1
Actb0.1
51.1±6.6
46.4±6.4
2.5±0.7
22.1±17.2
72.8±15.8
5.1±2.7
58
42
0
b
2
NAct≥0.2
2NAct≥0.1
Act≥2
Act≥2
3
NPA
4AAP
a
b
a
c
Relative activity
N6
N8
RM
Relative activity
N6
N8
RM
b
Actb0.2
60.2±4.6
36.2±5.1
3.6±2.3
54.0±35.1
42.4±32.8
3.6±2.5
61
39
0
Actb0.2
2NAct≥0.2
Act≥2
70.3±2.5
14.0±3.0
7.7±2.5
58.4±32.0
33.7±25.7
7.9±6.4
68
32
0
b
2
NActb0.25
Act≥2
RM=random mutagenesis library.
Noted for each substrate is the percentage of variants that show a certain activity level relative to the wild-type-like starting point
rePON1.
a
Act=activity relative to rePON1.
b
The limit for the lowest detectable activity was defined as b0.1-fold relative to the starting point rePON1, or up to b0.25-fold,
depending on the rate of background hydrolysis and the assay's variability for the tested substrate.
c
For the N6 and N8 libraries, three plates were screened, with 92 library clones per plate. The error ranges reflect the deviations in the
percentage of active clones between these three plates. Only one plate was screened from the random mutagenesis library, and error
ranges could not be provided. However, for most substrates, the differences between the ancestral N6 and N8 libraries and the random
mutagenesis library are well beyond the error range. The exceptions are the aryl esters 3NPA and 4AAP, which exhibit high background
rates and low PON1-catalyzed rates, and therefore show relatively high error ranges.
substrates. Variants in this group contained the
mutation F222S, which has also been identified from
previous screens for an increased organophosphate
contained at least one other mutation (or sometimes
few mutations) that is likely to mediate a higher
organophosphate hydrolase activity, such as Y190F
1
,19
1
hydrolysis activity of PON1.
These variants
and Y293F. The second group (variants 2D9, 2F10,
Table 3. Selected variants isolated from the N8-PON library
Variant
CMP
IMP
DEPCYC PARA
TBL
TBBL
3NPA
4AAP
Substitutions
2G1
52.4±12.2
7±1.9 6.8±1.5
7.1±1.8 1.2±2.3 1.4±0.3 3.8±1.0 12.5±3.1
L55I, I74L, Y190F, F222S,
a
F264L, Y293F, T332S
1
1
3
B10
D2
D7
46.6±8.6 22.7±3.5 6.5±1.0 11.8±2.1 1.1±0.3 1.3±0.2 1.1±0.4 2.6±0.3
29.9±7.4 21.9±4.4 2.1±0.4 2.5±0.6 1±0.5 1.1±0.2 1.6±0.3 4.9±1.1
25.2±2.3 11.6±1.1 7±0.6
D78A, F222S, T→C (309)
E56D, I74L, F222S, I237V
L55I, S193F, F222S,
V282T, Y293F, T332S
L55I, S137Q, F222S, I237V,
K244N, G→C(732)
11±0.8 1.1±0.5 0.4±0.0 1.2±0.2 1.8±0.2
3.2±0.3 1.4±0.8 0.7±0.1 1±0.2 1.3±0.2
1
E11
F6
20.7±3.4 14.3±2.0 2.2±0.4
16.7±4.5 11.7±3.0 1.8±0.5
2
2.3±0.7 0.9±0.2 0.6±0.2 0.8±0.1 1.3±0.2 L55I, S137Q, Y190I, F222S, C→T
993)
(
a
1
2
3
1
2
2
B6
4.7±1.4
4.4±1.0
3.7±1.5
3.6±1.0
1.4±0.2
1±0.2
5.5±1.6 6.8±1.9 10.1±3.2 6.2±2.5 1.3±0.2 5.8±1.5 4.8±1.2 M75K,Y190I, L259S, Y293F, T332S
1.9±0.2 6±1.1 7.4±1.0 6±1.0 1.2±0.1 4.1±0.6 3.2±0.6 L55I, P189F, S193F, Y293F, T332S
D4
E8
A2
D9
F10
4.1±1.6 5.1±1.9
6.2±2.2 9.8±3.7
0.4±0.1 1.1±0.1
0.2±0.1 0.9±0.2
5.5±2.1 5.4±2.0 1.4±0.3 4.4±1.4 5.4±1.8
6±2.2 1.2±0.4 0.9±0.3 0.7±0.2 3.9±1.2
M75K, S137Q, I237V, T332S
E56D, F222S, I291L, T332S
2.4±0.3 22.8±1.9 1.7±0.1 0.8±0.3 9.9±1.2 M75K, S137Q, I237V, I291L, T332S
2±0.4 12±2.2 0.9±0.2 0.6±0.2 3.8±0.9
L55I, P189F, I237V,
I291L, T332S, T→C (180)
P189F, I237V, I291L, T332S
L55I, I74L, V282T, I291L
—
3
1
A8
E6
0.9±0.2
0.8±0.3
1±0.3
0.2±0.0 0.8±0.2
0.1±0.1 0.2±0.2
1.5±0.3 5.5±1.0 0.6±0.1 0.7±0.1 2.9±0.3
0.1±0.2 2.8±1.0 1.3±0.1 0.5±0.1 3.5±0.3
rePON1-G3C9
1±0.1
1±0.2
1±0.1
1±0.7 1±0.0 1±0.1
1±0.3
Listed are the activity levels with various substrates relative to the wild-type-like rePON1. Standard deviations were obtained by
measuring the activities of four individually grown colonies of the same variant.
a
Random mutations incorporated by PCR amplifications of the library.

Directed Evolution with Ancestral Libraries
843
3
A8, and 1E6) was improved mainly with the
rePON1). The improvements observed were up to
20-fold (Table 2).
thiolactone TBL and the aryl ester 4AAP, and also
exhibited a decrease in organophosphate hydrolase
activity, especially with IMP. Variants belonging to
this group all contained the mutation I291L that has
been previously seen to affect aryl esterase and
N6 library variants that exhibited the largest
changes in specificity were recloned and sequenced
(Table 4). A large number of N6 variants improved
with different substrates. Most notable were large
improvements towards aryl esters, in particular
with 4AAP (e.g., 20-fold higher activity of variant
2F10 compared to rePON1), that were not observed
in the N8 library. The improvements with organo-
phosphates were, however, much lower than
observed with N8 library variants (≤5-fold as
opposed to ≤52-fold). These differences probably
relate to mutations like F222S that appear in the N8
library but not in the N6 library, and mutations such
as P189G that appear only in N6. Notably, the latter
is also a mutation that appears in the ancestral node
N6, but not in any of the family members that
diverged from it (Table 1). Thus, exploring more
than one node can further broaden the diversity of
ancestral libraries.
1
8,28
lactonase activities.
Other mutations found in
various members of this group, such as I74L and
2
9
S193F, are also likely to mediate new specificities.
The third group (variants 1B6, 2D4, 3E8, and 1A2)
was improved with all different substrates tested to
a similar degree, suggesting higher protein stability
and solubility. Variants of this group all contained
the T332S mutation, which contributed to their
higher TBLase activity and is likely to mediate their
1
8
higher expression levels.
The N6-PON library
N6, a node that is deeper and more challenging to
predict, was also examined in order to test the impact
of the particular choice of node. This library includes
2
7 substituted positions compared to 21 positions in
Ancestral versus random libraries
the N8 library (Table 1). The N6 library also included
substitutions that do not appear in any of the
mammalian PON family members (9 out of 27). As
with the N8 library, we analyzed the activity of 276
randomly picked library clones and compared them
to the random mutagenesis library (Fig. 3). The N6
library exhibited 3.6–8.4% of improved variants,
depending on the substrate examined (≥2-fold
higher activity compared to the starting point
In total, by screening 276 variants from each
library, we identified 75 improved variants (≥2-fold
higher activity with at least one of the six substrates)
with the N8 library and 37 improved variants in the
N6 library. Thus, on average, 1 out of 5 library
variants showed improved activity with at least one
substrate. In comparison, only 2 mildly improved
variants were identified by screening 92 variants in
Table 4. Selected variants isolated from the N6-PON ancestral library and the random mutagenesis library
Variant
TBBL
TBL
CMP
DEPCYC
3NPA
4AAP
Substitutions
a
2
3
2
3
2
1
2
3
F10
G10
E3
3±0.2
1.6±0.1
2.2±0.7
1.9±0.2
1.9±0.2
1.9±0.3
1.7±0.1
0.5±0.0
12±1.9
1.8±0.2
1.1±0.1
3.1±0.2
4±0.9
8.8±2.7
5±0.5
1.1±0.1
3.6±0.5
3.6±0.2
1.9±0.9
2.5±0.2
1.9±0.2
2.5±0.5
1.4±0.4
0.7±0.02
1.7±0.2
1±0.02
0.3±0.1
2.3±0.1
1.5±0.1
3.7±1.0
2.8±0.2
0.3±0.01
6.3±0.6
4.3±0.4
2.2±0.6
2.3±0.2
2.2±0.2
2.7±0.6
2±0.6
20.1±3.0
13.5±0.4
9.1±3.6
7.9±1.2
6.3±0.8
4.9±1.0
4.4±0.4
4.2±0.1
S67T, D136H, P189G, T332S
I74L, Y190I, E239D
b
R32H, I74L, M196V, L241F, Y293M
D1
Y293M, T332S
a
a
A9
S272E, V282I, Y293M, T332S
a
G11
G11
B4
L55I, P189G, T332S
S67T, Y190I, W194L, I237V, T332S
a
a
a
3.6±1.3
E56T, I74L, S137H, P189G,
a
M196V, E239D, S272E
a
a
3
1
1
3
1
A5
F12
C10
B8
2.1±0.2
1.9±0.2
1±0.3
0.5±0.1
1.5±0.1
4.9±0.8
2.2±0.2
2.6±0.9
1.2±0.1
1.9±0.1
0.4±0.04
1.4±0.1
0.4±0.1
4±0.6
0.6±0.1
1.7±0.2
0.3±0.1
2.5±0.6
2±0.2
2.2±2.8
1.1±0.1
4±1.4
2.5±0.7
1.1±0.2
2.9±1.1
1.9±0.1
1.7±0.5
1.7±0.4
1.6±0.1
L55I, S137H, L241F, V282I, I291L
D136H, L241F
a
a
a
S137H, S193M, I291L, Q329I
a
S67T, M75P, S137H, E239D
E1
5.3±0.2
L55I, S67T, M75P, D136H, Y190I,
a
W194L, L241F, V282I
a
b
a
1
C8
B4
0.5±0.2
0.5±0.1
0.03±0.03
1.1±0.1
1±0.1
1.4±0.6
2.9±2.8
0.9±0.1
1±0.1
1±0.4
3.6±1.1
1.3±0.1
1.4±0.0
1±0.1
1.1±0.4
1.2±0.3
0.5±0.2
0.9±0.0
1±0.1
4.6±3.6
1.2±0.2
0.5±0.1
1±0.1
1.5±0.2
0.5±0.3
0.6±0.3
0.5±0.03
1±0.3
E56T, S193M, H197R, V282I, T332S
a a
2
M75P, S137H, V282I
b
RM-A3
RM-A12
rePON1-G3C9
H115Q
b
F24I
1±0.1
1±0.1
—
RM=random mutagenesis library.
Listed are the activity levels with various substrates relative to the wild-type-like rePON1. Standard deviations were obtained by
measuring the activities of four individually grown colonies of the same variant.
a
Unique ancestral substitutions to amino acids that do not appear in any of the exiting mammalian PONs (see Table 1).
Random mutations incorporated by PCR amplifications of the library.
b

844
Directed Evolution with Ancestral Libraries
1
8
the random mutagenesis library (Table 4). If we
consider ≥5-fold at the threshold for improved
activity, the average is 1 out of 10 for the N6 and N8
ancestral libraries, and nil for the random library.
The outcome with ancestral libraries is also
favorable if we consider the “viability” of a library
high-throughput means.
However, we also
searched for new variants in a more challenging
system where the screen is much more laborious. To
this end, we explored the enzyme family of SULTs.
The SULT family comprises enzymes that use 3-
phosphoadenosine 5-phosphosulfate (PAPS) as a
donor molecule to transfer a sulfate group to
different acceptor substrates. We used two known
colorimetric substrates, 4-nitrophenol (pNP) and 3-
cyanoumbelliferone (3CyC), but the actual sub-
strates of interest, bisphenol A (BPA) and 3,5,3′,5′-
tetraiodothyronine (L-thyroxine), could only be
screened by low-throughput assays such as high
pressure liquid chromatography (HPLC) (Supple-
mentary Fig. 5). We chose the latter two substrates
because all characterized SULTs sulfonate them
—
meaning the frequency of folded and functional
enzyme variants. The latter is reflected in the ability
to hydrolyze the lactone substrate TBBL: 60% of
“
dead” variants (lactonase activity within back-
ground rates, i.e., ≤0.2 relative to rePON1) were
observed in the random library compared to an
average of 43.4% in the N8 and N6 libraries (Table 2).
Expression levels of library variants
3
2–34
The expression levels of library variants were also
examined (Supplementary Fig. 4). The expression
levels in all three libraries (N8, N6, and random
mutagenesis library variants) were, on average,
about half relative to rePON1 and more widely
distributed as expected. The expression levels in
library variants varied from no expression up to 1.7-
fold higher expression than the starting point
rePON1. As expected, most inactive clones (as
determined by activity with TBBL) showed no
expression or very low expression. Since the ob-
served increases in activity in improved variants
were much higher than 1.7-fold (Tables 3 and 4) and
since the improvements in most cases were sub-
strate-specific (except for variants belonging to the
third group of the N8 library: 1B6, 2D4, 3E8, and 1A2;
see the text above), it is clear that most improvements
stem from changes in specific activities.
with low rates
and because of the potential
interest in their detoxification.
Design of the ancestral SULT library
The design protocol for the SULT library followed
that of the PON libraries. Candidate residues for
mutagenesis included all residues within 12 Å of the
reaction center of SULT1A1 [the center was defined as
the hydroxyl group of the p-nitrophenyl substrate;
3
5
Protein Data Bank (PDB) ID: 1LS6], including the
binding pocket cleft wall and 4 Å around the binding
pocket of the donor molecule PAPS (Fig. 2b;
Supplementary Table 3). From these positions (99 in
total), we choose the ones that differ between wild-
type human SULT1A1 (our starting point gene) and
the ancestor N8. The latter comprises the ancestor of
the “1s” SULT family, which includes the 1A, 1B, 1C,
1
D, and 1E families (the alignment is provided as
Ancestral SULT library
Supplementary Fig. 6, and the tree is provided as
Supplementary Fig. 7). We chose N8 because the
prediction for the deeper node (N7) was far less
accurate (bootstrapping value of 44 out of 100) and
also because the inclusion of more SULT families
created too many gaps in the alignment. The ancestral
The substrates used with the PON library exper-
iments are all colorimetric and/or fluorogenic,
making it relatively easy to screen for new activities
and specificities by high-throughput or even ultra-
Table 5. Positions that differ between human SULT1A1 and its N8 ancestor
Residue
SULT1A1
SULT1A
SULT1B
SULT1C
SULT1D
SULT1E
SULT-N8 ancestor
1
1
2
4
5
7
7
8
8
1
1
1
1
2
2
2
0
P
P
A
S
V
F
M
A
I
Y
H
E
S
V
Q
F
P
P
ANT
SN
VM
FY
LIMAV
AICVD
VIMACE
YH
RHKN
DENH
CS
ILV
TPCQSER
VIFL
KRQ
NPDE
NGSD
VATP
IVL
TVSY
AVNESR
LVA
PQRSL
EKAPTSQNG
NTIWM
ACSV
R
E
EHQ
S
VI
Y
KR
I
FI
YH
Q
KE
AS
IM
DK
DE
DTSKE
YF-VE
QYFN
VA
Q
E
N
A
I
Y
N
I
L
H
R
D
C
I
1
7
4
4
TIM
LVI
6
FQYHL
DHEVQKL
LRIF
PQLISK
SQYH
RK
DNAH
LVSCA
LVIDW
TKSNAVG
IYVL
F
N
7
6
TKNS
VLM
IFLH
LR
DHNP
P
LVM
DE
VIMS
9
LQIAKV
DH
43
44
51
68
43
45
47
LQ
LAHFY
AS
LVI
−
ST
S
I
VLMI

Directed Evolution with Ancestral Libraries
845
Table 6. Selected variants isolated from the SULT-N8 library
a
b
b
b
b
Variants
pNP
3CyC
BPA
l-Thyroxine
Mutations
c
d
a10
b2
2.41±0.05
2.51±0.17
3.66±0.11
4.24±0.49
2.45±0.2
6.97±0.9
19.3±4.4
1.03±0.3
9.4±0.4
0.48±0.1
2.09±0.2
1.11±0.1
0.5±0.1
2.76±0.1
10.49±1.8
0.18±0.0
S44A, M77N, V79M, A86I, E151D, F247I, M260I
d
M77N, A101S, Y143H, E151D, S168C, V243I, F247I
b9
P10Q, M77N, E151D, S168C, V243I, F247I
d
d
d
d6
P11E, Q56E, L67V, F76Y, A101S, Y143H,
c d c
M145I, S168C, H213R, F247I, H250Y
e
d
d
d
d
d
d
1
1
1
A1-1C5
0.13±0.01
0.27±0.02
0.02±0.01
1±0.02
2.53±0.3
2.07±0.5
0.2±0.3
1±0.01
1.29±0.08
2.04±0.1
0.21±0.1
1±0.1
1.58±0.0
1.18±0.0
0.33±0.00
1±0.1
Q56E, L67V, A101S, H213R, F247I, M260I
d d d d d d
e
A1-1E9
L67V, A101S, Q177K, V211L, F222K, F247I
f
E1
1E1
1A1
Human 1A1
a
Listed are activities as fold change relative to human SULT1A1, as measured with purified enzyme samples. Standard deviations
were obtained from three independent measurements.
b
Assay conditions for each substrate tested (enzyme concentration, substrate concentration, and PAPS concentration, respectively): (1)
[
E]
0
=2.86 μM, 50 μM pNP, and 0.5 mM PAPS; (2) [E]
0 0
=1.43 μM, 1 μM 3CyC, and 100 μM PAPS; (3) [E] =2.86 μM, 100 μM BPA, and
5
00 μM PAPS; (4) [E]
0
=2.86 μM, 25 μM L-thyroxine, and 100 μM PAPS.
c
Random mutations incorporated by PCR amplifications of the library.
Stabilizing mutations.
d
e
Variants 1A1-1C5 and 1A1-1E9 were obtained by introducing consensus, stabilizing mutations, and selecting for higher expression
levels in E. coli and higher thermal stability. They also contained the F247I active-site mutation (Amir Aharoni, submitted for publication)
that was also included in the ancestral substitutions and may account for the changes in substrate specificity observed in variants 1A1-
1
C5 and 1A1-1E9. The magnitude of changes in the ancestral library variants, however, is much larger, and this mutation on its own is
therefore incapable of inducing the observed changes.
f
The activity observed with human SULT1E1 is provided for reference.
substitutions included in the library are listed in Table
. As can be seen, all the introduced substitutions
enzymes. Active variants exhibiting a detectable
5
level of sulfotransferase activity (92 in total, ∼50% of
the variants) were collected, regrown, and tested
with colorimetric pNP. The ratio of activities with
3CyC to activities with pNP served as a preliminary
indication for changes in substrate specificity. Four
selected variants that exhibited the highest de-
viations relative to human 1A1 were subsequently
collected, recloned, overexpressed, and purified.
The purified enzyme variants were tested with
3CyC and pNP, and with the target substrates BPA
and L-thyroxine, by monitoring the exchange of
PAPS into 3-phosphoadenosine 5-phosphate (PAP)
appear in at least one SULT member. To increase the
mutational tolerance of the library variants, we
included stabilizing consensus mutations in the
3
0,36
library.
These mutations have been found to
stabilize human 1A1 and to increase its levels of
functional expression in E. coli (Amir Aharoni et al.,
submitted for publication). In total, we introduced by
oligonucleotide spiking and shuffling 16 ancestral
mutations and 8 stabilizing mutations at an average
of 5.1±2.4 mutations per gene.
3
2
Screening the SULT-N8 library
by HPLC.
These four variants exhibited activity improve-
ments compared to wild-type SULT1A1 for each of
the substrates tested: pNP, ≤4.3-fold; 3CyC, ≤19.3-
fold; BPA, ≤9.4-fold; and L-thyroxine, ≤10.5-fold
We started with 184 randomly chosen library
variants that were tested for quenching the fluores-
cence of 3CyC as an indicator for folded active
Fig. 4. Kinetic characterization of the SULT1A1-b9 mutant compared to wild-type SULT1A1. Michaelis–Menten curve
for wild-type SULT1A1 (full circles) and its b9 (empty circles), with pNP (a) (left; [E] =2.86 μM) and 3CyC (b) (right;
0
[
E]
0
=1.43 μM) fitted to the substrate inhibition model (GraphPad Prism 5). Error bars were derived from two independent
measurements.

846
Directed Evolution with Ancestral Libraries
Table 7. Kinetic characterization of the SULT1A1-b9 mutant compared to wild-type SULT1A1
pNP
3CyC
Wild-type 1A1
Kinetic parameters
Fitted parameters
Wild-type 1A1
Mutant b9
Mutant b9
cat (min−
(μM)
(μM)
max (μM/min)
Activity units (nmol/min/mg)
1
)
5.36±1
7±2
60.±14
9.25±0.5
92.5±5
17.2±2.4
24±7
283±100
30.5±1.1
305±11
0.45±0.03
6.2±0.9
—
0.55±0.1
11±2
2.94±0.3
0.72±0.2
38±12
3.5±0.02
70±0.4
k
K
m
K
i
Apparent parameters
V
The kinetic parameters are derived from the plots in Fig. 4.
(
Table 6). These activities are also higher than the
mutations is partly due to the parallel incorporation
of compensatory consensus mutations (1.8±0.67 per
variant, on average).
activities exhibited by 1E1 and any other SULT
3
2
family, as indicated by published data. The
mutations that appeared in these variants included
P10Q, S44A, M77N, A86I, S168C, V243I, and F247I
Kinetic and structural characterizations of
SULT1A1-b9
(
Table 6). Several of these mutations have been
reported to be involved in substrate specificity (e.g.,
3
7
32,38
A86I and F247I
).
The kinetic parameters of the selected variant b9
with both pNP and 3CyC were determined and
compared to those of human SULT1A1 (Fig. 4 and
Table 7). SULTs, particularly human 1A1, are known
to exhibit substrate inhibition. The data were
As is the case with PON ancestral libraries, the
active variants carried a high number of mutations,
up to six ancestral mutations in variant b9 or b2
(
Table 6). This high number of accommodated
Table 8. Data collection and refinement statistics
Human SULT1A1 variant b9+PAP+pNP
Human SULT1A1 variant b9+PAP+3CyC
Data collection
PDB accession number
Number of images
3QVU
125
3QVV
135
Oscillation step (°)
1
1
Wavelength (Å)
1.54178
2.5
1.04628
2.35
P2 2 2
1 1 1
Resolution (Å)
Space group
P2
1
1
2 2
1
Unit cell parameters a, b, c (Å)
Number of observed reflections (last bin)
Number of unique reflections (last bin)
66.86, 71.54, 122.77
101,999 (11,010)
20,787 (2277)
98.7 (99.8)
71.79, 71.91, 122.59
145,889 (24,812)
26,761 (4512)
98.4 (99.5)
Completeness (last bin) (%)
a
R
merge (last bin) (%)
12.6 (45.5)
7.1 (52.7)
b
R
measured (last bin) (%)
I/σ(I) (last bin)
14.0 (50.9)
7.8 (58.0)
10.62 (3.22)
2.60–2.50
13.36 (3.06)
2.50–2.35
Last-resolution shell
Redundancy (last bin)
4.91 (4.84)
5.45 (5.49)
Refinement statistics
Resolution range (Å)
48.85–2.50
19,746
24.7/20.8
46.93–2.35
25,385
27.6/23.0
Number of reflections used in refinement
c
Rfree/Rwork
RMSD from ideal
Bond length (Å)
Bond angle (°)
0.003
0.587
0.005
0.947
Ramachandran plot
Preferred regions (%)
Allowed regions (%)
Outliers (%)
97.1
2.6
94.2
5.1
0.3
0.7
nh
a
P
P P
Rsym = Rmerge =
j=h − Ih;i j =
Ih;i.
h
h
i
qffiffiffiffiffiffiffiffiffiffi n
n
h
n
P
P
h
P P
P
h
b
c
nh
1
nh
Rmeas
=
j=h − Ih;i j =
Ih;i with =h =
Ih;i.
nh − 1
h
i
h
i
i
R
work =∑‖F
0
−|F
c
‖=∑|F
0
|, where F
0
denotes the observed structure factor amplitude and F
c
denotes the structure factor amplitude
calculated from the model. Rfree is as for Rwork but calculated with 5% of randomly chosen reflections omitted from the refinement.

Directed Evolution with Ancestral Libraries
847
accordingly fitted to the simplest model of substrate
carried the P10Q, M77N, E151D, S168C, V243I, and
F247I mutations at the background of wild-type
human 1A1. We obtained structures in complex
with pNP and PAP at 2.5 Å resolution (PDB ID:
3QVU), and those in complex with 3CyC and PAP at
2.35 Å resolution (PDB ID: 3QVV; Table 8).
3
9
inhibition. The turnover number (kcat) of the b9
mutant was higher than that of wild-type 1A1, ∼3-
fold higher for pNP, and 6.5-fold for 3CyC. The Km
value for pNP increased (3-fold), resulting in kcat/Km
values similar to those of wild type. In contrast, the
Km value for 3CyC decreased 8.5-fold such that the
3
5
The structures of wild-type human SULT1A1
kcat/K of variant b3 with CyC increased by N50-fold
and its b9 mutant are highly similar (RMSD of
∼0.33 Å for 287 carbon α atoms). The binding of
the acceptor pNP is also similar to that in the
complexes of wild type in b9. However, significant
changes could be observed in active-site volume
and configuration. The acceptor's site of the b9
variant is considerably enlarged compared to wild-
type SULT1A1, primarily due to the F247I muta-
tion (Fig. 5a and b). This residue has also been
identified as a key position for accommodating the
binding of acceptor molecules (Amir Aharoni et al.,
m
relative to wild type. Substrate inhibition by pNP
also decreased (the apparent K for wild-type 1A1 is
i
6
3
0 μM compared to 283 μM for the b9 mutant). With
CyC as substrate, no substrate inhibition was
observed for the wild type, whereas substrate
inhibition could be seen for the b9 mutant
(
K =37.7 μM). However, the K /K ratio was
i i m
relatively high (N50 as opposed to ∼10 with pNP).
To unravel the effect of the ancestral mutations,
we solved the crystal structure of the b9 mutant that
Fig. 5. Structural changes in the active site of the evolved b9 variant. (a) Superposition of the acceptor's binding pocket
of the b9 variant (blue) onto that of the human 1A1 structure (PDB ID: 1LS6; gray mesh). The acceptor pockets of both wild
m
type and b9 can readily accommodate pNP, although the wild type's pocket shows a better fit (and a 3-fold lower K for
pNP). (b) The wild type's pocket is too small to accommodate the larger substrate 3CyC (green), whereas in b9, the F247I
mutation (sticks) enlarged this pocket to accommodate 3CyC and even larger substrates. (c) The b9 mutant structure in
complex with PAP and pNP (blue) superposed onto the wild-type SULT1A1 structure (gray). The incorporated ancestral
mutations created a hydrophobic cluster of three isoleucines (I89, I243, and I247) that seem to have induced the closure of
active-site loops by up to 1.5 Å. (d) The reshaping of the secondary acceptor's tunnel. The secondary tunnel of b9 (cyan) is
displaced relative to wild type and is also narrower such that the inhibitory pNP molecule seen in human 1A1 (white
sticks) cannot be accommodated.

848
Directed Evolution with Ancestral Libraries
submitted for publication). This enlarged cavity
facilitates the binding of 3CyC, a bigger substrate
than pNP, and accounts for the observed improve-
ments in activity for 3CyC and with the other
large substrates, L-thyroxine and BPA. Within the
resolution limits, the structures of the b9 mutant in
complex with 3CyC and pNP, including the
reacting hydroxyl groups that coalign, appear
essentially identical (Supplementary Figs. 8 and 9).
We also observed changes with respect to the
secondary tunnel; this cavity, previously seen in
human SULT1A1, binds a second pNP molecule in a
nonreactive mode and is likely to be responsible for
the observed substrate inhibition. The alteration of
this tunnel was mediated by two mutations (V243I
and F247I). It appears that F247I, by being in van der
Waals contact with I89, stabilized a new rotamer of
I89. The resulting hydrophobic cluster (V243I, F247I,
and I89) provokes the closing of the active loops and
the narrowing of the secondary tunnel (Fig. 5c and
d). In accordance, substrate inhibition in b9 is ∼3-
fold weaker than that in wild-type 1A1 (Fig. 4 and
Table 7). Other mutations, such as M77N and S168C,
are in close vicinity to the active site and might also
contribute to the observed change in specificity.
However, their role does not seem to be reflected in
the crystal structures.
Most of the incorporated mutations can be found in
various family members, in particular in the case of
the SULT library (Table 5). Only in certain libraries
(e.g., the N6-PON library) unique substitutions not
found in any known family member comprise about a
third of the library (Table 1). Thus, a similar outcome
could, in principle, be achieved through shuffling of
various family members. However, the diversity of
substitutions that appear in these positions within
contemporary family members is vast (e.g., Table 5)
and could not be explored within reasonable library
sizes. The choice of ancestral substitution limits
library sizes while maintaining sufficient functional
diversity. Furthermore, family shuffling is limited to
closely related sequences (≥70% amino acid
3
8
7
identity). Moreover, nearly all paralogues (i.e.,
members of families exhibiting different functional
specificities) show sequence identity lower than 70%.
Indeed, family shuffling would not be applicable for
the SULT1 and PON families explored here, as their
most diverse members show only 42% and 55%
amino acid identities, respectively. Methods that
overcome the identity barrier have been described
7
(e.g., by shuffling fragments at predefined positions).
An alternative approach is to have a library based on
amino acids that appear in structurally equivalent
active-site positions. For example, by screening
∼
60,000 library variants derived from a Pseudomonas
fluorescens esterase, variants that enantioselectively
Discussion
hydrolyze 3-phenylbutyric acid esters (up to 240-fold
faster than wild type) were identified. However, the
3
We describe the construction of ancestral libraries
as a tool for facilitating directed enzyme evolution.
We exemplified this approach with mammalian
PONs and SULTs. Both these enzyme families are of
potential interest in drug metabolism and detoxifi-
cation of xenobiotics. Although screened by low-
throughput means, the ancestral PON and SULT
libraries show a remarkably high frequency of
polymorphic and functionally diverse variants.
Screening of as few as 200–300 variants enabled
the isolation of dozens of active variants with a
range of different specificities and higher activities,
including those with substrates with which all
known family members exhibit low activity.
Ancestral libraries comprise a means of focusing
diversity to positions that can readily promote
changes in substrate specificity. The number of
explored positions is relatively high (in the range
of 20 positions or more). However, mutational
diversity is highly focused on a given ancestral
node or on a few nodes that could be explored in
separate libraries (as here, with the N6-PON and
N8-PON libraries) or combined into one library.
This limited diversity is nonetheless wide enough
to promote large functional variability and thereby
enables the isolation of ‘first-generation’ variants
as starting points for further directed evolution by
standard methods.
choice of diversified library positions was largely
dictated by early reports that explored various
relevant positions of this enzyme. For many enzymes,
however, knowledge of specificity-altering positions
is limited or nonexistent. Furthermore, as indicated in
Tables 1 and 5, the variability within all family
members is huge; thus, family shuffling and the
related methods may result in too large and insuffi-
ciently viable diversity. In the ancestral libraries, the
choice of which mutations from existing family
members would actually be explored is given by the
ancestral sequence. In addition, a significant number
of mutations and mutational compositions that are
not seen in extant members are also explored.
The high viability of the ancestral libraries is
particularly notable in light of the high frequency of
mutations. The number of mutations in active clones
from the ancestral libraries (N4 ancestral substitutions,
on average, in the N8-PON and N6-PON variants,
and N7 in the SULT variants) is much larger than in
active variants isolated from random libraries (a
single mutation in the random library screened
here). This despite the average mutational rate in the
unselected libraries being very similar. Indeed, the
likelihood of finding a combination of four or more
beneficial and/or neutral mutations in a randomly
1
,40
mutated library is very low.
In contrast, the
ancestral libraries can easily accommodate high

Directed Evolution with Ancestral Libraries
849
mutational loads. In particular, it is clear that the
accumulation of several active-site mutations, while
maintaining configurational stability and enzymatic
activity, is normally a laborious and lengthy process
that demands several consecutive rounds of random
mutagenesis and screening of a large number of
library variants. In the ancestral libraries described
here, variants with up to 11 mutations within and
near the active site were readily identified by
screening as few as 200 variants.
catalytic activity. Although the general outlines for
library design by REAP have been partially
4
5
provided, a more detailed and well-established
protocol might be needed for a wider implementation.
In addition, although REAP resembles our approach in
the use of ancestral inference, these approaches differ
in several important ways. Including positions that
play a role in mediating a specific trait that appears in
one branch (e.g., the nucleoside triphosphate promis-
cuity of viral polymerases) can help focus library
diversity. However, this approach is limited to a
specific functional property that existed in some form
or another along the evolutionary history of the
targeted protein family. We included the entire
diversity found in various ancestral nodes throughout
phylogeny. Our approach does not require knowledge
on the specificity of the different members of the
targeted protein family. Instead, it explores a large
fraction of the phylogenetic sequence space by
combinatorial libraries and thus obtains activities and
specificities that did not exist in any of the ancestors, let
alone in contemporary members of the targeted
family.
Other approaches that enable the incorporation of a
large number of mutations while maintaining a
relatively high fraction of viable variants are consen-
sus and neutral drift libraries. Consensus libraries are
composed of targeted mutations that bring the
sequence of the target gene closer to the family
consensus. The consensus sequence may overlap the
sequence of the family ancestor (but not vice versa) or
may differ. In certain positions, the ancestor and
consensus sequences overlap (e.g., position 55; Table 1,
first line). However, in many positions, the ancestral
sequence does not match the consensus or differs
altogether from the existing diversity (e.g., position 56
in the N6-PON ancestor; Table 1, second line).
Furthermore, consensus mutations usually residing
away from the active site exhibit mostly stabilizing
In summary, the ancestral library approach
described here complements other approaches to
4
6
targeted mutagenesis such as family shuffling,
41
47
4
1
effects. Indeed, the main aim of consensus libraries is
to obtain variants with higher stability, and not to alter
SCHEMA, REAP, neutral drift, and simulta-
neous multiple-site saturation mutagenesis. It has,
3
14
the enzyme's specificity. Consensus ancestor muta-
tions were also included in our SULT library, and the
recombinant PON1 variant used as starting point for
however, its unique features and provides a new
and effective way of performing protein engineer-
ing. We also describe a range of new PON and SULT
variants that comprise valuable starting points for
engineering, detoxifying, and decontaminating en-
zymes for organophosphates and other environ-
mental hazards such as bisphenol A, or for potential
therapeutic uses (e.g., prevention of organophos-
phate toxicity, treatment of hyperthyroidism).
42
the PON libraries carries several such mutations.
Neutral drift libraries are another related method for
directed evolution. Comparison of ancestral PON
libraries to previously generated neutral libraries of
1
rePON1 reveals a similar frequency of improved
variants and a similar magnitude of improvements (in
particular for the high mutational load neutral drift
library; Supplementary Fig. 10). The generation of
neutral drift libraries is more laborious but requires no
prior knowledge of the enzyme's structure or phylog-
eny. Thus, these approaches (ancestral and neutral
drift) may apply to different target enzymes.
Materials and Methods
Ancestral reconstruction
4
As this work progressed, Benner and Gaucher et al.
Sequences of PON family members were collected from
the National Center for Biotechnology Information non-
redundant protein sequence database (nr) using protein
alignment BLAST (blastp) with its default settings. Only
fully sequenced genes and those that contain all PONs'
also described the application of ancestral inference to
enzyme engineering. Their method called REAP was
applied to engineer polymerase variants that accept
nucleoside triphosphates with 3′-ONH for DNA
2
4
29
sequencing. REAP was implemented to identify
active site and other essential residues were included.
patterns of change in the evolutionary history of
polymerases indicated by deviations from simple
Redundant sequences exhibiting ≥90% identity to other
4
8
sequences were removed using Cd-hit, and the align-
4
9
31,43,44
ment was created with PRANK using the F+ option and
50
Markov models.
Because viral polymerases are
the LG substitution matrix. The alignment was fine-
tuned manually to improve its reliability at gap positions
likely to accept modified nucleotides, amino acid
replacements that might have been responsible for the
functional divergence between viral and nonviral
polymerases were identified. The analysis thereby
assisted the identification of candidate amino acid
exchanges that could give rise to polymerase variants
that accept modified nucleosides without the loss of
(
Supplementary Fig. 1) and then used to create a
51
phylogenetic tree using the Phyml program. Bootstrap-
ping analysis was performed to evaluate the tree.
Ancestral predictions for the various nodes were obtained
5
2
with FastML, and the most probable ancestral sequence
was used for library design. The SULT ancestral

850
Directed Evolution with Ancestral Libraries
predictions were performed as described above, with
some modifications. Firstly, a 95% identity threshold was
applied. Using a threshold of 90% resulted in the loss of all
but two members of the 4A1 family, as well as the loss of
family members with known structures such as human
The rates were corrected for background nonenzymatic
rates and compared to the activity of the human-like
rePON1-G3C9 starting point. From the N6-PON library,
276 randomly picked clones were similarly screened. For
the SULT library screen, we used 10 μl of crude lysate in
a total reaction volume of 100 μl for screening with the
1
A2 and 1A3. Secondly, because the structures of most
SULT family members are available (as opposed to only
PON1), the alignment was structure-based using T-Coffee
0
colorimetric pNP (50 μM; [PAPS] =0.5 mM, absorption
measured at 405 nm) or with the fluorogenic 3CyC
53
Expresso. The alignment after the manual refinement of
the gap positions (Supplementary Fig. 6) was used for
constructing the SULT tree (Supplementary Fig. 7) and
ancestral predictions, as described above.
(1 μM; [PAPS] =100 μM, excitation at 408 nm and
0
emission at 450 nm). The ratio of activity between these
two substrates relative to the ratio exhibited by the
stabilized human SULT1A1 variants (1A1-1C5 and 1A1-
1
E9) was used to identify changes in specificity. Four
variants showing the highest changes in specificity were
purified and assayed further.
Library preparation
We applied the ISOR (Incorporating Synthetic Oligo-
nucleotides via Gene Reassembly) protocol for generat-
ing DNA libraries of random combinations of specific
Purification of enzyme variants and enzymatic assays
17
mutations. The starting point for the PON library was
a recombinant PON1 variant, rePON1-G3C9, that
expresses function in E. coli, exhibits ∼94% identity to
rabbit PON1, and exhibits the same enzymatic specific-
The SULT variants were purified by Ni-NTA. The
purified enzymes were assayed with pNP and 3CyC as
described above, and by HPLC essentially as described
3
2
previously.
Briefly, the 100-μl reactions contained
2
0
ity as rabbit and human PON1. For each ancestral
2.9 μM purified enzyme and BPA (100 μM) and PAPS
(0.5 mM), or L-thyroxine (25 μM) and PAPS (100 μM).
The reactions were stopped at different time points
(5 min, 10 min, and 20 min) by adding acetonitrile to
30% vol/vol (final concentration) and immediately
freezing in liquid nitrogen. Samples were diluted 3-
fold with buffer A (see the text below), centrifuged to
remove the protein precipitate, and injected into HPLC.
HPLC was performed with AKTA-Basic using an ion-
exchange column (Sepax-SAX-NP5) and a gradient of
buffer A (10 mM Tris, pH 7) and buffer B (buffer A plus
0.25 M NaCl)—from 0% buffer A to 50% buffer B in
10 min—and at a flow rate of 1 ml/min. The conversion
of PAPS into PAP was monitored at 259 nm (retention
times of 9.5 min and 7.4 min, respectively). The
concentration of the produced PAP was plotted against
time, and the initial rates were extrapolated. Kinetic
assays with pNP and 3CyC were performed using a
microtiter plate reader and 0.1-ml reactions in 10 mM
phosphate buffer (pH 7.0) while monitoring optical
density or florescence, respectively. The data were fitted
to the equation: v =k [E] [S] /(K +[S] (1+[S] /K )),
mutation, a mutagenesis oligonucleotide ∼30 bp long
(
∼15 bp from each side of the mutation) was
synthesized (Supplementary Tables 2 and 4). Typically,
pmol of an equimolar mixture of all mutagenesis
2
oligonucleotides and 50 ng of DNase I (Takara)-
generated fragments of the starting gene (rePON1-
G3C9; GenBank AY499193) were assembled as de-
1
7
scribed previously, followed by a nested PCR with
external primers. The control library of random muta-
tions was generated by error-prone PCR (GeneMorph®
II Random Mutagenesis Kit; Stratagene, Agilent Tech-
nologies, USA) of rePON1-G3C9, and the mutation load
was tuned to a similar average of ∼4 mutations per
gene. The various PON libraries (N6, N8, and Mu
libraries) were separately ligated into a modified pET-32
expression plasmid that introduces a His-tag at PON1's
19
C-terminus. The ligated plasmids were transformed
into E. coli BL21-DE3 cells. The transformed cells were
plated on agar, and single colonies were isolated and
grown on 96-deep-well plates for protein expression
1
and activity screen. The SULT library was similarly
0
cat
0
0
m
0
0
i
generated. As a starting point, we used two stabilized
human SULT1A1 variants, dubbed hSULT1A1-1C5 and
hSULT1A1-1E9, each carrying several consensus muta-
tions (listed in Table 6). These were kindly provided by
Amir Aharoni (submitted for publication; Department
of Life Sciences, Ben-Gurion University of the Negev,
Israel). Although their enzymatic specificities deviate
from wild-type human 1A1, these deviations are much
smaller than those seen in the ancestral library variants
whereby K is the apparent dissociation constant for
i
39
substrate binding in the inhibitory mode.
Crystallization
Freshly purified SULT1A1-b9 was concentrated to
33 mg/ml. One microliter of the donor's product PAP
(100 mM in phosphate buffer, pH 7.0) and 1 μl of the
acceptor pNP (200 mM in 100% dimethyl sulfoxide)
were added to 30 μl of protein solution. Crystallization
was performed using the hanging-drop vapor diffusion
method. Equal volumes (1 μl) of protein and reservoir
solutions were mixed, and the resulting drop was
allowed to reach equilibrium with a 500-μl reservoir
solution made of 18–22% (wt/vol) polyethylene glycol
(PEG) 3350 and 50 mM Tris–HCl buffer (pH 8). Crystals
appeared after 2 days at 293 K. For exchange of the
acceptor molecule, crystals were rinsed for 5 min in a
drop of a solution composed of the mother liquor
(
Table 6).
Library screens
From the N8-PON library, 276 randomly picked
clones were grown on 96-well plates, and crude bacterial
lysates were screened with the target substrates (Sup-
1
plementary Fig. 3), as described previously. The lysate
volume taken with each substrate varied from 5 μ to
2
0 μl such that initial rates could be reliably measured.

Directed Evolution with Ancestral Libraries
851
solution, 25% PEG 600, and 5 mM 3CyC, and then
soaked for 1 h in a fresh drop of the same solution.
References
1
2
3
4
. Gupta, R. D. & Tawfik, D. S. (2008). Directed enzyme
evolution via small and effective neutral drift libraries.
Nat. Methods, 5, 939–942.
. Reetz, M. T. & Carballeira, J. D. (2007). Iterative
saturation mutagenesis (ISM) for rapid directed evolu-
tion of functional enzymes. Nat. Protoc. 2, 891–903.
. Helge, J. & Bornscheuer, U. T. (2010). Natural
diversity to guide focused directed evolution. Chem-
BioChem, 11, 1861–1866.
. Chen, F., Gaucher, E. A., Leal, N. A., Hutter, D.,
Havemann, S. A., Govindarajan, S. et al. (2010).
Reconstructed Evolutionary Adaptive Paths give
polymerases accepting reversible terminators for
sequencing and SNP detection. Proc. Natl Acad. Sci.
USA, 107, 1948–1953.
Data collection and structure refinement
Crystals were transferred into a cryoprotectant solution
containing the reservoir solution and 25% PEG 600 for
1
min, and then flash-cooled in liquid nitrogen. X-ray
diffraction data for the crystal of mutant b9 in complex with
pNP and PAP (2.5 Å resolution) were collected at 100 K on a
Rigaku R-AXIS IV++ imaging plate area detector mounted
on a Rigaku RU-H3R generator with CuKα radiation
focused by Osmic confocal mirrors. A second data set
collected using a mutant b9 crystal soaked with 3CyC and
PAP at 2.35 Å resolution was collected at 100 K with
synchrotron radiation at ID23-1 beamline (European
Synchrotron Radiation Facility, Grenoble, France) using an
ADSC Quantum Q315r detector. The data were integrated
5
6
. Lutz, S. (2010). Beyond directed evolution—semi-
rational protein engineering and design. Curr. Opin.
Biotechnol. 21, 734–743.
. Carbone, M. N. & Arnold, F. H. (2007). Engineering by
homologous recombination: exploring sequence and
function within a conserved fold. Curr. Opin. Struct.
Biol. 17, 454–459.
5
4
and scaled using XDS and XSCALE. Both crystals
belonged to the same space group yet exhibited significantly
different unit cell dimensions (Table 8). Molecular replace-
55
ment was performed using MOLREP, with the structure
of human SULT1A1 (PDB ID: 2D06) used as starting model.
5
5
Manual model improvement was performed using Coot.
Refinement of the models was performed using
7
8
9
. Sieber, V., Martinez, C. A. & Arnold, F. H. (2001).
Libraries of hybrid proteins from distantly related
sequences. Nat. Biotechnol. 19, 456–460.
. Wickens, M. & Cox, M. M. (2009). Critical reviews in
biochemistry and molecular biology. Introduction.
Crit. Rev. Biochem. Mol. Biol. 44, 2.
. Bershtein, S. & Tawfik, D. S. (2008). Ohno's model
revisited: measuring the frequency of potentially
adaptive mutations under various mutational drifts.
Mol. Biol. Evol. 25, 2311–2318.
55
REFMAC5. TLS refinement was applied to the 3CyC-
containing structure. Figures were prepared with PyMOL
(
DeLano Scientific LLC), and figures displaying cavities
were produced with Caver 2.1.1†.
Accession numbers
Coordinates and structure factors of the b9 mutant, in
complex with PAP and pNP and in complex with PAP and
1
1
0. Soskine, M. & Tawfik, D. S. (2010). Mutational effects
and the evolution of new protein functions. Nat. Rev.
Genet. 11, 572–582.
1. Raillard, S., Krebber, A., Chen, Y., Ness, J. E.,
Bermudez, E., Trinidad, R. et al. (2001). Novel enzyme
activities and functional plasticity revealed by recom-
bining highly homologous enzymes. Chem. Biol. 8,
3
CyC, have been deposited in the PDB under accession
numbers 3QVU and 3QVV, respectively.
Supplementary materials related to this article can be
found online at doi:10.1016/j.jmb.2011.06.037
8
91–898.
1
2. Amitai, G., Gupta, R. D. & Tawfik, D. S. (2007). Latent
evolutionary potentials under the neutral mutational
drift of an enzyme. HFSP J. 1, 67–78.
Acknowledgements
1
3. Bloom, J. D., Romero, P. A., Lu, Z. & Arnold, F. H.
(2007). Neutral genetic drift can alter promiscuous
We are very grateful to Hagit Bar and Tal Pupko
for their assistance in performing phylogenetic
analysis, to Amir Aharoni and Dotan Amar for
providing the stabilized SULT1A1 variants and for
assisting in SULT enzymatic assays, and to Daniel
Tal for assisting in protein production and HPLC
assays. Financial support by the National Institutes
of Health (W81XWH-07-2-0020) and the Defense
Threat Reduction Agency (HDTRA 1-07-C-0024) is
gratefully acknowledged. M.E. is a fellow sup-
ported by the IEF Marie Curie Program (grant
protein functions, potentially aiding functional evo-
lution. Biol. Direct, 2, 17.
4. Jochens, H., Aerts, D. & Bornscheuer, U. T. (2010).
Thermostabilization of an esterase by alignment-
guided focussed directed evolution. Protein Eng. Des.
Sel. 23, 903–909.
5. Flores, H. & Ellington, A. D. (2005). A modified
consensus approach to mutagenesis inverts the cofac-
tor specificity of Bacillus stearothermophilus lactate
dehydrogenase. Protein Eng. Des. Sel. 18, 369–377.
6. Thornton, J. W. (2004). Resurrecting ancient genes:
experimental analysis of extinct molecules. Nat. Rev.
Genet. 5, 366–375.
1
1
1
1
2
52836). D.S.T. is the Nella and Leon Benoziyo
Professor of Biochemistry.
7. Herman, A. & Tawfik, D. S. (2007). Incorporating
Synthetic Oligonucleotides via Gene Reassembly
(
ISOR): a versatile tool for generating targeted
†
http://www.caver.cz
libraries. Protein Eng. Des. Sel. 20, 219–226.

852
Directed Evolution with Ancestral Libraries
1
1
2
8. Aharoni, A., Amitai, G., Bernath, K., Magdassi, S. &
Tawfik, D. S. (2005). High-throughput screening of
enzyme libraries: thiolactonases evolved by fluores-
cence-activated sorting of single cells in emulsion
compartments. Chem. Biol. 12, 1281–1289.
sulfotransferases. Drug Metab. Pharmacokinet. 17,
221–228.
34. Ozawa, S., Shimizu, M., Katoh, T., Miyajima, A., Ohno,
Y., Matsumoto, Y. et al. (1999). Sulfating-activity and
stability of cDNA-expressed allozymes of human
phenol sulfotransferase, ST1A3⁎1 ((213)Arg) and
ST1A3⁎2 ((213)His), both of which exist in Japanese
as well as Caucasians. J. Biochem. 126, 271–277.
35. Gamage, N. U., Duggleby, R. G., Barnett, A. C.,
Tresillian, M., Latham, C. F., Liyou, N. E. et al. (2003).
Structure of a human carcinogen-converting enzyme,
SULT1A1. Structural and kinetic implications of
substrate inhibition. J. Biol. Chem. 278, 7655–7662.
36. Lehmann, M. & Wyss, M. (2001). Engineering proteins
for thermostability: the use of sequence alignments
versus rational design and directed evolution. Curr.
Opin. Biotechnol. 12, 371–375.
37. Liu, M. C., Suiko, M. & Sakakibara, Y. (2000).
Mutational analysis of the substrate binding/catalytic
domains of human M form and P form phenol
sulfotransferases. J. Biol. Chem. 275, 13460–13464.
38. Barnett, A. C., Tsvetanov, S., Gamage, N., Martin, J. L.,
Duggleby, R. G. & McManus, M. E. (2004). Active site
mutations and substrate inhibition in human sulfo-
transferase 1A1 and 1A3. J. Biol. Chem. 279,
18799–18805.
39. Copeland, R. A. (2000). Enzymes, 2nd edit. Wiley, New
York, NY.
40. Bershtein, S., Segal, M., Bekerman, R., Tokuriki, N. &
Tawfik, D. S. (2006). Robustness–epistasis link shapes
the fitness landscape of a randomly drifting protein.
Nature, 444, 929–932.
41. Lehmann, M., Kostrewa, D., Wyss, M., Brugger, R.,
D'Arcy, A., Pasamontes, L. & van Loon, A. P. (2000).
From DNA sequence to improved functionality: using
protein sequence comparisons to rapidly design a
thermostable consensus phytase. Protein Eng. 13,
49–57.
42. Khersonsky, O., Rosenblat, M., Toker, L., Yacobson,
S., Hugenmatter, A., Silman, I. et al. (2009). Directed
evolution of serum paraoxonase PON3 by family
shuffling and ancestor/consensus mutagenesis, and
its biochemical characterization. Biochemistry, 48,
6644–6654.
43. Gaucher, E. A. (2007). In Ancestral Sequence Recon-
struction (Liberles, D. A., ed.), Oxford University
Press, New York, NY.
44. Benner, S. A. (2003). Interpretive proteomics—finding
biological meaning in genome and proteome data-
bases. Adv. Enzyme Regul. 43, 271–359.
45. Cole, M. F. & Gaucher, E. A. (2010). Exploiting models
of molecular evolution to efficiently direct protein
engineering. J. Mol. Evol. 72, 193–203.
46. Stemmer, W. P. (1994). DNA shuffling by random
fragmentation and reassembly: in vitro recombination
for molecular evolution. Proc. Natl Acad. Sci. USA, 91,
10747–10751.
47. Meyer, M. M., Silberg, J. J., Voigt, C. A., Endelman, J. B.,
Mayo, S. L., Wang, Z. G. & Arnold, F. H. (2003). Library
analysis of SCHEMA-guided protein recombination.
Protein Sci. 12, 1686–1693.
48. Li, W. & Godzik, A. (2006). Cd-hit: a fast program for
clustering and comparing large sets of protein or
nucleotide sequences. Bioinformatics, 22, 1658–1659.
9. Gupta, R. D., Goldsmith, M., Ashani, Y., Simo, Y.,
Mullokandov, G., Bar, H. et al. (2011). Directed
evolution of hydrolases for prevention of G-type
nerve agent intoxication. Nat. Chem. Biol. 7,
120–125.
0. Aharoni, A., Gaidukov, L., Yagur, S., Toker, L.,
Silman, I. & Tawfik, D. S. (2004). Directed evolution
of mammalian paraoxonases PON1 and PON3 for
bacterial expression and catalytic specialization. Proc.
Natl Acad. Sci. USA, 101, 482–487.
2
1. Draganov, D. I. & La Du, B. N. (2004). Pharmacoge-
netics of paraoxonases: a brief review. Naunyn-
Schmiedeberg's Arch. Pharmacol. 369, 78–88.
2. Khersonsky, O. & Tawfik, D. S. (2005). Structure–
reactivity studies of serum paraoxonase PON1 sug-
gest that its native activity is lactonase. Biochemistry,
2
44, 6371–6382.
2
3. Draganov, D. I., Teiber, J. F., Speelman, A., Osawa, Y.,
Sunahara, R. & La Du, B. N. (2005). Human
paraoxonases (PON1, PON2, and PON3) are lacto-
nases with overlapping and distinct substrate speci-
ficities. J. Lipid Res. 46, 1239–1247.
2
4. Teiber, J. F. & Draganov, D. I. (2011). High-performance
liquid chromatography analysis of N-acyl homoserine
lactone hydrolysis by paraoxonases. Methods Mol. Biol.
692, 291–298.
2
5. Gaidukov, L., Bar, D., Yacobson, S., Naftali, E.,
Kaufman, O., Tabakman, R. et al. (2009). In vivo
administration of BL-3050: highly stable engineered
PON1–HDL complexes. BMC Clin. Pharmacol. 9, 18.
6. Bloom, J. D., Labthavikul, S. T., Otey, C. R. & Arnold,
F. H. (2006). Protein stability promotes evolvability.
Proc. Natl Acad. Sci. USA, 103, 5869–5874.
2
2
2
7. Khersonsky, O. & Tawfik, D. S. (2006). Chromogenic
and fluorogenic assays for the lactonase activity of
serum paraoxonases. ChemBioChem, 7, 49–53.
8. Aharoni, A., Gaidukov, L., Khersonsky, O., McQ
Gould, S., Roodveldt, C. & Tawfik, D. S. (2005). The
‘
evolvability’ of promiscuous protein functions. Nat.
Genet. 37, 73–76.
2
9. Harel, M., Aharoni, A., Gaidukov, L., Brumshtein, B.,
Khersonsky, O., Meged, R. et al. (2004). Structure and
evolution of the serum paraoxonase family of detox-
ifying and anti-atherosclerotic enzymes. Nat. Struct.
Mol. Biol. 11, 412–419.
0. Bershtein, S., Goldin, K. & Tawfik, D. S. (2008). Intense
neutral drifts yield robust and evolvable consensus
proteins. J. Mol. Biol. 379, 1029–1044.
1. Gaucher, E. A., Thomson, J. M., Burgan, M. F. &
Benner, S. A. (2003). Inferring the palaeoenvironment
of ancient bacteria on the basis of resurrected proteins.
Nature, 425, 285–288.
2. Allali-Hassani, A., Pan, P. W., Dombrovski, L.,
Najmanovich, R., Tempel, W., Dong, A. et al. (2007).
Structural and chemical profiling of the human
cytosolic sulfotransferases. PLoS Biol. 5, e97.
3. Nishiyama, T., Ogura, K., Nakano, H., Kaku, T.,
Takahashi, E., Ohkubo, Y. et al. (2002). Sulfation
of environmental estrogens by cytosolic human
3
3
3
3

Directed Evolution with Ancestral Libraries
853
4
5
5
5
9. Loytynoja, A. & Goldman, N. (2008). Phylogeny-aware
gap placement prevents errors in sequence alignment
and evolutionary analysis. Science, 320, 1632–1635.
0. Le, S. Q. & Gascuel, O. (2008). An improved general
amino acid replacement matrix. Mol. Biol. Evol. 25,
53. Armougom, F., Moretti, S., Poirot, O., Audic, S.,
Dumas, P., Schaeli, B. et al. (2006). Expresso: automatic
incorporation of structural information in multiple
sequence alignments using 3D-Coffee. Nucleic Acids
Res. 34, W604–W608.
54. Kabsch, W. (1993). Automatic processing of rotation
diffraction data from crystals of initially unknown
symmetry land cell constants. J. Appl. Crystallogr. 26,
795–800.
1307–1320.
1. Guindon, S. & Gascuel, O. (2003). A simple, fast, and
accurate algorithm to estimate large phylogenies by
maximum likelihood. Syst. Biol. 52, 696–704.
2. Pupko, T., Pe'er, I., Shamir, R. & Graur, D. (2000).
A fast algorithm for joint reconstruction of ancestral
amino acid sequences. Mol. Biol. Evol. 17, 890–896.
55. Collaborative Computational Project Number 4. (1994).
The CCP4 Suite: programs for protein crystallography.
Acta Crystallogr., Sect. D.: Biol. Crystallogr. 50, 760–763.