Fluorescence activated cell sorting as a general ultra-high-throughput screening method for directed evolution of glycosyltransferases

Article abstract of DOI:10.1021/ja104167y

Glycosyltransferases (GTs) offer very attractive approaches to the synthesis of complex oligosaccharides. However, the limited number of available GTs, together with their instability and strict substrate specificity, have severely hampered the broad application of these enzymes. Previous attempts to broaden the range of substrate scope and to increase the activity of GTs via protein engineering have met with limited success, partially because of the lack of effective high-throughput screening methods. Recently, we reported an ultra-high-throughput screening method for sialyltransferases based on fluorescence-activated cell sorting (Aharoni et al. Nat. Methods 2006, 3, 609-614). Here, we considerably improve this method via the introduction of a two-color screening protocol to minimize the probability of false positive mutants and demonstrate its generality through directed evolution of a neutral sugar transferase, β-1,3-galactosyltransferase CgtB. A variant with broader substrate tolerance than the wild-type enzyme and 300-fold higher activity was identified rapidly from a library of >10⁷ CgtB mutants. Importantly, the variant effected much more efficient synthesis of G _M1a and asialo G_M1 oligosaccharides, the building blocks of important therapeutic glycosphingolipids, than did the parent enzyme. This work not only establishes a new methodology for the directed evolution of galactosyltransferases, but also suggests a powerful strategy for the screening of almost all GT activities, thereby facilitating the engineering of glycosyltransferases.

Full text of DOI:10.1021/ja104167y

Published on Web 07/12/2010
Fluorescence Activated Cell Sorting as a General
Ultra-High-Throughput Screening Method for Directed
Evolution of Glycosyltransferases
Guangyu Yang,†,§,‡ Jamie R. Rich,† Michel Gilbert,# Warren W. Wakarchuk,#
Yan Feng,§,‡ and Stephen G. Withers*,†
Centre for High-throughput Biology (CHiBi) and Department of Chemistry, UniVersity of British
Columbia, 2036 Main Mall, VancouVer, British Columbia V6T 1Z1, Canada, School of Life
Sciences and Biotechnology, Shanghai Jiao Tong UniVersity, 800 Dongchuan Road, Shanghai
200240, P. R. China, Key Laboratory for Molecular Enzymology and Engineering of Ministry of
Education, Jilin UniVersity, 2519 Jiefang Road, Changchun 130023, P. R. China, and National
Research Council of Canada, Institute for Biological Sciences, 100 Sussex DriVe, Ottawa,
Ontario K1A 0R6, Canada
Received May 14, 2010; E-mail: withers@chem.ubc.ca
Abstract: Glycosyltransferases (GTs) offer very attractive approaches to the synthesis of complex
oligosaccharides. However, the limited number of available GTs, together with their instability and strict
substrate specificity, have severely hampered the broad application of these enzymes. Previous attempts
to broaden the range of substrate scope and to increase the activity of GTs via protein engineering have
met with limited success, partially because of the lack of effective high-throughput screening methods.
Recently, we reported an ultra-high-throughput screening method for sialyltransferases based on
fluorescence-activated cell sorting (Aharoni et al. Nat. Methods 2006, 3, 609-614). Here, we considerably
improve this method via the introduction of a two-color screening protocol to minimize the probability of
false positive mutants and demonstrate its generality through directed evolution of a neutral sugar
transferase, ꢀ-1,3-galactosyltransferase CgtB. A variant with broader substrate tolerance than the wild-
type enzyme and 300-fold higher activity was identified rapidly from a library of >107 CgtB mutants.
Importantly, the variant effected much more efficient synthesis of GM1a and asialo GM1 oligosaccharides,
the building blocks of important therapeutic glycosphingolipids, than did the parent enzyme. This work not
only establishes a new methodology for the directed evolution of galactosyltransferases, but also suggests
a powerful strategy for the screening of almost all GT activities, thereby facilitating the engineering of
glycosyltransferases.
Introduction
addition, enzymes isolated directly from nature are not always
suitable for synthetic processes that employ high concentrations
The chemical synthesis of complex oligosaccharides is often
an extremely challenging process because of the need for
selective, labor-intensive protection/deprotection steps.1,2 En-
zymatic synthesis by glycosyltransferases (GTs), on the other
hand, offers significant advantages over purely chemical meth-
ods due to the exquisite control conferred over the stereo- and
regiochemistry of the reaction products.3-5 However, broad
application of GTs is hampered by the limited availability of
enzymes and by their often strict substrate specificities. In
of substrate and possibly also organic solvents. Therefore, access
to GTs tailored for the efficient synthesis of specific oligosac-
charides and glycoconjugates would be of great interest.
However, while the field of protein engineering has undergone
explosive growth over the past 10 years, there have been
relatively few success stories with respect to the directed
evolution of GTs compared with other important enzyme
families such as lipases, proteases, and glycosidases. This is
mainly due to the fact that typical screens for GTs are based on
microtiter plate assays and thus require significant amounts of
costly reagents, and have limited throughput.6-9 Indeed, most
previous directed evolution studies involving GTs screened no
more than 1000 colonies. Screening of larger libraries would
allow the exploration of a much larger protein sequence space,
† University of British Columbia.
§ Shanghai Jiao Tong University.
‡ Jilin University.
# National Research Council of Canada, Institute for Biological Sciences.
(1) Wu, D.; Fujio, M.; Wong, C. H. Bioorg. Med. Chem. 2008, 16, 1073–
1083.
(2) Hanson, S.; Best, M.; Bryan, M. C.; Wong, C. H. Trends. Biochem.
Sci. 2004, 29, 656–663.
(6) Park, S. H.; Park, H. Y.; Sohng, J. K.; Lee, H. C.; Liou, K.; Yoon,
Y. J.; Kim, B. G. Biotechnol. Bioeng. 2009, 102, 988–994.
(7) Persson, M.; Palcic, M. M. Anal. Biochem. 2008, 378, 1–7.
(8) Williams, G. J.; Goff, R. D.; Zhang, C. S.; Thorson, J. S. Chem. Biol.
2008, 15, 393–401.
(3) Macmillan, D.; Daines, A. M. Curr. Med. Chem. 2003, 10, 2733–
2773.
(4) Lairson, L. L.; Henrissat, B.; Davies, G. J.; Withers, S. G. Annu. ReV.
Biochem. 2008, 77, 521–555.
(5) Weijers, C. A.; Franssen, M. C.; Visser, G. M. Biotechnol AdV. 2008,
26, 436–456.
(9) Williams, G. J.; Zhang, C.; Thorson, J. S. Nat. Chem. Biol. 2007, 3,
657–662.
9
10570 J. AM. CHEM. SOC. 2010, 132, 10570–10577
10.1021/ja104167y  2010 American Chemical Society

FACS as a Screening Method for Directed Evolution of GTs
A R T I C L E S
thereby improving the chances of identifying useful mutations.10,11
The development of high-throughput, low-cost screens for the
interrogation of large GT libraries is therefore of great interest.
Fluorescence-activated cell sorting (FACS) is a specialized
application of flow cytometry.12,13 In a FACS instrument,
multiple fluorescent parameters (size, structure, and fluorescent
signals) of individual sample particles (cells, microbeads, or
emulsions) can be analyzed by a focused laser beam at rates of
approximately 107/h. A charge can be applied to the particles
of interest, which are deflected into a collection tube by a
charged plate. Originally developed for the analysis of
protein-ligand interactions, FACS has recently emerged as a
powerful tool for screening enzyme libraries due to its unique
sensitivityanditsabilitytoanalyzeasmanyas108 mutants/day.14-16
The major challenge in FACS screening is how to maintain the
link between genotype and phenotype because most enzyme
substrates do not remain associated with the cells. Several
different FACS screening approaches have been developed,
largely based on the attachment of products to cell surfaces or
entrapment of the product inside the cells, and these have been
applied to the directed evolution of proteases,17-20 peroxidases,21
esterases,22 glutathione transferases,23-25 and nucleoside ki-
nases26 in order to improve their expression level, stability,
ligand binding, activity, or substrate specificity. We have
recently reported a novel FACS-based, high-throughput screen-
ing method, which makes use of differences in physical
properties between the substrate and product, for the directed
evolution of the sialytransferase CstII from Campylobacter
jejuni.27 Libraries of mutant sialyltransferase (ST) genes were
expressed in Escherichia coli cells that are naturally permeable
to an exogenous glycosyl acceptor (Bodipy-lactose, 1) (Figure
1). Active STs catalyzed sialylation of 1, entrapping the charged
fluorescent product within the cell and thereby co-localizing the
Figure 1. Fluorescently labeled lactose (1) and differentially labeled
N-acetyl-D-galactosamine (2 and 3) derivatives used in screens for glyco-
syltransferase activity.
genotype and phenotype (Figure 2a). Using FACS, the library
was screened at a rate of more than 107 events/h and an ST
variant with up to 400-fold improved activity for 1 was thus
identified. However, improvements were specific to 1, as
enhanced activity had arisen from the evolution of a binding
site for the substrate-appended fluorophore, as demonstrated
crystallographically.27 An approach that avoids the coincident
evolution of dye binding sites is therefore necessary.
Here, we present a method to evolve improved GT activity
through simultaneous use of two selection substrates bearing
the same sugar but chemically distinct fluorophores, thereby
minimizing the probability of selecting for a dye-binding site
(Figure 2b). Significantly, we have applied this strategy to a
ꢀ-1,3-galactosyltransferase (CgtB), which transfers the neutral
sugar galactose, thereby demonstrating that cellular entrapment
is not restricted to charged products and highlighting the
generality of this ultra-high-throughput screen for GTs. Screen-
ing of a library of >107 CgtB mutants against fluorescently
labeled derivatives of N-acetylgalactosamine (2 and 3) identified
a range of improved mutants, with up to 300-fold enhanced
catalytic activities and improved ability to synthesize GM1a and
asialo GM1 oligosaccharides, the building blocks of important
therapeutic glycosphingolipids.
(10) Aharoni, A.; Amitai, G.; Bernath, K.; Magdassi, S.; Tawfik, D. S.
Chem. Biol. 2005, 12, 1281–1289.
(11) Dietrich, J. A.; McKee, A. E.; Keasling, J. D. Annu. ReV. Biochem.
2010, 79, 563–590.
(12) Givan, A. L. Flow Cytometry: First Principles; Wiley-Liss: NY, 1992.
(13) Shapiro, H. M. Practical Flow Cytometry, 4th ed.; Wiley-Liss: NY,
2004.
(14) Becker, S.; Schmoldt, H. U.; Adams, T. M.; Wilhelm, S.; Kolmar, H.
Curr. Opin. Biotechnol. 2004, 15, 323–329.
(15) Farinas, E. T. Comb. Chem. High Throughput Screening 2006, 9, 321–
328.
Experimental Section
(16) Yang, G.; Withers, S. G. ChemBioChem 2009, 10, 2704–2715.
(17) Olsen, M. J.; Stephens, D.; Griffiths, D.; Daugherty, P.; Georgiou,
G.; Iverson, B. L. Nat. Biotechnol. 2000, 18, 1071–1074.
(18) Varadarajan, N.; Gam, J.; Olsen, M. J.; Georgiou, G.; Iverson, B. L.
Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 6855–6860.
(19) Varadarajan, N.; Rodriguez, S.; Hwang, B. Y.; Georgiou, G.; Iverson,
B. L. Nat. Chem. Biol. 2008, 4, 290–294.
Generation of Random-Mutagenesis Library. DNA encoding
amino acids 1-272 of CgtB was subjected to error-prone PCR using
the Gene Morph II Mutagenesis kit (Stratagene) with 0.03 ng or
0.06 ng of pCW-CgtB∆C30-MalE as template, following the
manufacturer’s protocol. The primers used were CgtB-F, 5′-GAA-
AGGGAGCTCACATATGTTTAAAATTTCAATCATCTTACC-
AAC-3′ and CgtB-R, 5′-GAAGGTCGGAATTCCGGTTTTATTT-
TATATATTTGAATATATAGC-3′. The PCR product was di-
gested and ligated to pCW-MalE-C vector using NdeI and EcoRI
restriction sites and the ligation mixture was electroporated to E.
cloni 10G elite cells (Lucigen) according to the manufacturer’s
protocol. The transformed cells were grown overnight at 37 °C in
LB medium supplemented with ampicillin (100 mg/mL) and the
library plasmid DNA was extracted. Several individual clones
from each library were sequenced and shown to have an average
mutational frequency of ∼2 and ∼5 mutations per gene in the
0.03 and 0.06 ng template libraries, respectively. The plasmid
DNA from the two libraries was mixed in a 1:1 ratio and was
transformed into E. coli JM107* for screening.28 These were the
(20) Varadarajan, N.; Georgiou, G.; Iverson, B. L. Angew. Chem., Int. Ed.
2008, 47, 7861–7863.
(21) Lipovsek, D.; Antipov, E.; Armstrong, K. A.; Olsen, M. J.; Klibanov,
A. M.; Tidor, B.; Wittrup, K. D. Chem. Biol. 2007, 14, 1176–1185.
(22) Becker, S.; Ho¨benreich, H.; Vogel, A.; Knorr, J.; Wilhelm, S.; Rosenau,
F.; Jaeger, K. E.; Reetz, M. T.; Kolmar, H. Angew. Chem., Int. Ed.
2008, 47, 5085–5088.
(23) Kawarasaki, Y.; Griswold, K. E.; Stevenson, J. D.; Selzer, T.;
Benkovic, S. J.; Iverson, B. L.; Georgiou, G. Nucleic. Acids. Res. 2003,
31, e126.
(24) Griswold, K. E.; Kawarasaki, Y.; Ghoneim, N.; Benkovic, S. J.;
Iverson, B. L.; Georgiou, G. Proc. Natl. Acad. Sci. U.S.A. 2005, 102,
10082–10087.
(25) Griswold, K. E.; Aiyappan, N. S.; Iverson, B. L.; Georgiou, G. J.
Mol. Biol. 2006, 364, 400–410.
(26) Liu, L.; Li, Y.; Liotta, D.; Lutz, S. Nucleic Acids Res. 2009, 37, 4472–
4481.
(27) Aharoni, A.; Thieme, K.; Chiu, C. P.; Buchini, S.; Lairson, L. L.;
Chen, H.; Strynadka, N. C.; Wakarchuk, W. W.; Withers, S. G. Nat.
Methods. 2006, 3, 609–614.
(28) Antoine, T.; Heyraud, A.; Bosso, C.; Samain, E. Angew. Chem., Int.
Ed. 2005, 44, 1350–1352.
9
J. AM. CHEM. SOC. VOL. 132, NO. 30, 2010 10571

A R T I C L E S
Yang et al.
Figure 2. Schematic view of the fluorescent product entrapment method employed to screen mutant GT libraries. Fluorescently labeled acceptors such as
1, 2, or 3 are transported into the cell by a transmembrane sugar transport protein (permease). After an incubation period during which the acceptor may be
modified enzymatically, unreacted acceptor is removed in a washing step. E. coli cells containing catalytically active GTs retain the fluorescent product
inside the cell as it is no longer a substrate for the permease. Catalytically active enzymes can be identified and isolated by fluorescence activated cell
sorting: (a) in the original method, a single fluorescent substrate was used to screen the library of mutant sialyltransferases;27 (b) in this improved method,
two selection substrates were employed simultaneously in order to avoid the evolution of dye binding sites. E, glycosyltransferase; P, permease.
same engineered cells used for the original FACS screening work
in which both the lacZ ꢀ-galactosidase gene and the sialic acid
aldolase gene (nanA) have been knocked out. Knockout of the
nanA gene was not essential for this work, nor was it deleterious.
Generation of DNA Shuffling Library. To construct the
shuffling library, the plasmid DNA isolated from the third round
of FACS enrichment was amplified by PCR with CgtB-F and
CgtB-R primers. The amplified genes were purified and mixed with
the wild-type CgtB genes in a 50:1 ratio. The gene mixture was
digested using Benzonase (Novagen) and those DNA fragments
smaller than 100 bp were collected and reassembled using the
following PCR profile: 95 °C for 1 min; 45 cycles of 94 °C for
30s, 50 °C for 30 s, 72 °C for 1 min; and 72 °C for 5 min. Analysis
of the reassembly product by agarose gel electrophoresis revealed
a smear encompassing the ∼800 bp target length. A portion of the
assembly reaction mixture was used as template for PCR amplifica-
tion with CgtB-F and CgtB-R primers. The resulting genes were
purified, digested, and ligated into pCW-MalE-C vector and
transformed into E. coli JM107* competent cells for screening.
Site-Directed Mutagenesis. Site-directed mutagenesis was
performed using the Quikchange methodology (Stratagene) fol-
lowing the manufacturer’s protocol. Mutations were confirmed by
DNA sequencing. The primers used were
cells were spun down and the excess acceptor sugars were removed
by washing the cells with LB media and phosphate-buffered saline
(PBS, pH 7.4). The sample was placed on ice and taken to the
FACS for further analysis and sorting.
The cells were diluted with PBS as needed to obtain a flow
cytometric event rate of ∼4000/s in the FACSAria flow cytometer
(Becton-Dickinson) using PBS as sheath fluid. The threshold for
event detection was set to forward and side scattering. The average
sort rate was ∼4000 events/s, using a 70 µm nozzle, exciting argon
ion (488 nm), and 405 nm lasers, and measuring emissions passing
the 530 ( 20 nm (FITC) band-pass filter for the bodipy emission,
and the 450 nm (violet 1) filter for the coumarin emission. Cells
were sorted into Eppendorf tubes containing 0.5-1 mL of LB
medium. Pools of sorted positives were plated on LB agar plates
supplemented with ampicillin (100 mg/mL). Colonies were allowed
to grow overnight at 37 °C and the cells were removed from the
agar plates for the next round of FACS enrichment. FACS data
were processed using the Flowjo Software (Tree Star).
Protein Expression and Purification. Expression and purifica-
tion of wild-type CgtB and its mutants were carried out as
previously described.29 Briefly, the E. coli cells containing CgtB
variants were inoculated in 2YT medium supplemented with
ampicillin (100 mg/mL) at 37 °C overnight. The cells were
transferred into fresh medium and grown at 37 °C. When A600
reached 0.6-1.0, IPTG was added to a final concentration of 1
mM, and the cultures were transferred to 20 °C and grown
overnight. The cells were harvested and suspended in 10% (w/v)
Buffer A (which contains 20 mM HEPES, pH 7.0, 200 mM NaCl,
5 mM ꢀ-mercaptoethanol, 1 mM EDTA, and 10% glycerol). After
ultrasonic cell disintegration, the cell suspension was centrifuged
at 12 000 rpm for 30 min. The supernatant was diluted as needed
in Buffer A and applied to a 20 mL column of amylose resin (New
England Biolabs). After sample application, the column was washed
with 2 column volumes to elute unbound proteins. The bound
protein was eluted by washing the column with buffer B (Buffer
A and 20 mM maltose). The fractions containing the target protein
were collected and dialyzed overnight against 100 vol of 50 mM
sodium acetate, pH 6.0, and 20% glycerol at 4 °C. The purity of
all of the variants of CgtB was greater than 90% (see the Supporting
Information). The protein concentration was determined according
to the Bradford method, and bovine serum albumin (BSA) was used
as the standard.
CgtB_K151E_F: AATTATTGCAGAGAAAAATTTATATTG-
GACTATGTGG,
CgtB_K151E_R:
CAATATAAATTTTTCTCTG-
CAATAATTTTTTTTAC,
CgtB_L227G_F: GAATGAAGTGCGTGTTAAAAATAATAT-
TCAAGAGTTGC,
CgtB_L227G_R: GAATATTATTTTTAACACGCACTTCAT-
TCTTAGTATTAC,
CgtB_E234D_F: AAATAATATTCAAGATTTGCAGTTGGTTT-
TAAACTA
TTTAAGG,CgtB_E234D_R:GTTTAAAACCAACTGCAAATCT-
TGAATATTATTTTTAACAAGCAC.
Flow Cytometric Screening. Flow cytometric screening of CgtB
for glycosyltransferase activity was carried out essentially as
described before,27 except that two fluorescent substrates were used
simultaneously. Briefly, plasmid DNA (pCW-CgtB∆C30-MalE)
encoding the CgtB libraries was transformed into JM107* com-
petent cells and used to directly inoculate (without plating on agar)
LB media supplemented with ampicillin (100 mg/mL) and grown
overnight at 37 °C. The cells were then diluted 1/50 in M9 mineral
cultured media and grown at 37 °C with vigorous shaking. When
the absorbance at 600 nm (A600) reached 0.5-0.7, IPTG was added
to a final concentration of 1 mM, and the cultures were transferred
to 20 °C and grown overnight. Cells were spun down (1 mL) and
resuspended in 50 µL of M9 media supplemented with 0.2-0.5
mM concentrations of 2 and 3. Following 10-30 min of incubation,
Kinetic Analysis. Kinetic analysis of the CgtB variants was
performed at 37 °C in 20 mM HEPES buffer, pH 7.5, 10 mM
MgCl2, 10 mM MnCl2, 50 mM KCl, 0.25 mM NADH, 0.7 mM
(29) Bernatchez, S.; Gilbert, M.; Blanchard, M. C.; Karwaski, M. F.; Li,
J.; DeFrees, S.; Wakarchuk, W. W. Glycobiology 2007, 17, 1333–
1343.
9
10572 J. AM. CHEM. SOC. VOL. 132, NO. 30, 2010

FACS as a Screening Method for Directed Evolution of GTs
A R T I C L E S
Scheme 1. (a) Partial Structure of Lipooligosaccharide Outer Core from C. jejuni O:19 Strain OH4384.33 (b) Reaction Catalyzed by CgtB
Glycosyltransferases from C. jejuni. (c) Structure of Representative Ganglioside (Ganglioside GM1a with Galꢀ1,3GalNAcꢀ Subunit
)
Highlighted in Red. (d) Minimal Core 1 and Core 2 Glycoprotein O-Glycan Structures with Galꢀ1,3GalNAcR Subunit Highlighted in Red
PEP, and 0.1% (w/v) BSA. Release of UDP was coupled to the
oxidation of NADH (λ ) 340 nm, ε ) 6.22 mM-1 cm-1) using the
enzymes pyruvate kinase (PK) and lactate dehydrogenase (LDH).30
Background hydrolysis of UDP-Gal in the absence of acceptor sugar
was measured for the different acceptors and subtracted from the
enzyme transfer rate. Absorbance measurements were obtained
using a UV-2550 spectrophotometer (Shimadzu). Grafit 4.0 (Er-
ithacus software) was used to calculate kinetic parameters and
associated errors by direct fit of the initial rate to the appropriate
equations.
common to both the core 1 and 2 O-glycans as well as the
ganglioside glycolipids (Scheme 1c,d).29 Previous improvements
in the C. jejuni OH4384 CgtB variant involving a C-terminal
deletion and fusion to maltose binding protein (CgtB∆C30-
MalE) provided a starting point for directed evolution in this
study.29
FACS-based screening for CgtB activity in E. coli (Figure
1b) required that (i) fluorescent acceptor substrates reach the
cytoplasm, (ii) glycosylation of the acceptor result in product
entrapment, and (iii) unreacted acceptor be washed out of the
cell. To this end, we synthesized bodipy- and coumarin-labeled
fluorescent GalNAc acceptors 2 and 3, with the expectation that
they would be freely transported into and out of E. coli. The
expectation then was that upon galactosylation of 2 and 3 by
plasmid-encoded CgtB, the product disaccharides would be
entrapped within the cell, allowing unreacted acceptors to be
washed away without loss of fluorescent product (Figure 2b).
In this scenario the fluorescent intensity of the cells should
correlate with CgtB activity.
Specific activities of the CgtB variants using asialo GM2-FCHASE
as acceptor were determined as described previously.29
Synthesis. See the Supporting Information.
Results and Discussion
The CgtB galactosyltransferases (CAZy family GT2)31 are a
group of related enzymes from different strains of C. jejuni that
catalyze the biosynthesis of the Galꢀ1,3GalNAc linkage within
their respective structurally distinct lipooligosaccharides (Scheme
1a,b)32 To our knowledge, these are the only glycosyltrans-
ferases identified to date that utilize both R- and ꢀ-configured
N-acetylgalactosamine as an acceptor, thus, suggesting simul-
taneous promise for synthesis of the Galꢀ1,3GalNAc subunit
To test the feasibility of this strategy, E. coli cells expressing,
respectively, native CgtB, its improved variant CgtB∆C30-
MalE, and empty pCW-MalE-C plasmid were incubated with
both fluorescent GalNAc acceptors 2 and 3 at concentrations
that optimize the dynamic range accessible while minimizing
selection for low Km values. An E. coli strain JM107*, which
(30) Gosselin, S.; Alhussaini, M.; Streiff, M. B.; Takabayashi, K.; Palcic,
M. M. Anal. Biochem. 1994, 220, 92–97.
(31) Cantarel, B. L.; Coutinho, P. M.; Rancurel, C.; Bernard, T.; Lombard,
V.; Henrissat, B. Nucleic Acids Res. 2009, 37, D233–238.
(32) Gilbert, M.; Brisson, J. R.; Karwaski, M. F.; Michniewicz, J.;
Cunningham, A. M.; Wu, Y. Y.; Young, N. M.; Wakarchuk, W. W.
J. Biol. Chem. 2000, 275, 3896–3906.
(33) Aspinall, G. O.; McDonald, A. G.; Pang, H.; Kurjanczyk, L. A.;
Penner, J. L. Biochemistry. 1994, 33, 241–249.
9
J. AM. CHEM. SOC. VOL. 132, NO. 30, 2010 10573

A R T I C L E S
Yang et al.
Figure 3. (a) Visualization of fluorescence entrapment within E. coli cells
expressing native CgtB, the CgtB∆C30-MalE variant, or empty plasmid
(control). (b) Flow cytometry profile of whole cell fluorescence in cells
expressing empty plasmid (left, negative control) and CgtB∆C30-MalE
(right) after 15 min incubation with bodipy-GalNAc (2, 0.25 mM) and
coumarin-GalNAc (3, 0.25 mM), followed by a washing step.
lacks ꢀ-galactosidase (lacZ), was used to minimize the in ViVo
degradation of substrate and product.28 After extensive washing,
the fluorescence intensities of the cells were visualized under
ultraviolet light. As shown in Figure 3a, the cells expressing
empty plasmid were only weakly fluorescent. However, cells
expressing CgtB∆C30-MalE showed a strong fluorescence
signal, while an intermediate level of fluorescence intensity was
seen for the cells expressing native CgtB (Figure 3a). The flow
cytometry profile showed that CgtB∆C30-MalE cells have both
substantially higher green and blue fluorescence corresponding
to entrapment of both bodipy and coumarin disaccharide
conjugates, respectively (Figure 3b). The fluorescence signals
are stable for more than 2 h at 4 °C, providing an ample window
for the FACS analysis and sorting.
Figure 4. (a) FACS analysis of the error-prone PCR library of CgtB∆C30-
MalE and enrichment of the E. coli population with significantly higher
fluorescence entrapment after three rounds of screening. The shaded area
(pink) defines the gate employed to sort cells in the top 0.3%-1.0% of
both green (bodipy) and blue (coumarin) fluorescence intensity. A highly
fluorescent population was evolved after 3 rounds of enrichment. (b).
Fluorescence intensity (under UV irradiation) of the 26 mutants E1-E26
randomly selected from the enriched population obtained after three rounds
of sorting the EP-PCR library. WT: the parent enzyme, CgtB∆C30-MalE.
(c) FACS histogram analysis of the green and blue fluorescence of
CgtB∆C30-MalE (pink), its randomized library (green), and the S42 mutant
(blue) in E. coli cells.
While the mechanism of transport and entrapment of the
fluorescent substrate/product has not been clearly established,
several pieces of evidence point to mediation by lactose
permease. No fluorescence entrapment was obtained when using
E. coli Tuner strain (Novagen) in which the lactose permease
gene (lacY) is knocked out, suggesting that the fluorescent
acceptor could not get into the cells without the lacY gene (see
the Supporting Information for details). In addition, when regular
E. coli JM107* cells (which contain the lacY gene) were
incubated with the fluorescent acceptors together with high
concentrations of lactose, transport of the fluorescent acceptor
into the cells was inhibited (see the Supporting Information).
Since we have independently demonstrated that lactose does
not appreciably inhibit CgtB-catalyzed galactosyl transfer to
Bodipy-GalNAc, this result suggests that the lactose competi-
tively inhibits entry of Bodipy-GalNAc via the lactose permease,
but that the galactosylated product does not get out. Conse-
quently, this screening system can be used with GTs that use
galactose derivatives as acceptors. Indeed, in addition to the
ꢀ-1,3-galactosyltransferase CgtB reported in this work, we have
shown that this E. coli-based screening system can be readily
applied to sialyltransferases, R-1,4-galactosyltransferases, and
an R-1,3-galactosyltransferase (unpublished work). Importantly,
it should be possible to also use other sugar transport proteins
to further extend the application of this strategy.
Having established a valid reporter of galactosyltransferase
activity, DNA encoding amino acids 1-272 of CgtB was
subjected to error-prone PCR (EP-PCR) and cloned into the
pCW-MalE-C vector to generate a library of CgtB∆C30-MalE
variants with ∼2-5 mutations per gene. The library was
transformed into E. coli JM107* to generate ∼2 × 107 colonies.
Three iterative rounds of sorting were performed by FACS to
enrich those cells in the top 0.3-1.0% of both green and blue
fluorescence intensity (Figure 4a). After each round of sorting,
the “positive” cells were collected into growth medium and
plated on agar prior to the next round of enrichment. Twenty
9
10574 J. AM. CHEM. SOC. VOL. 132, NO. 30, 2010

FACS as a Screening Method for Directed Evolution of GTs
A R T I C L E S
Table 1. Specific Activities of CgtB∆C30-MalE and Selected Mutants Using Benzyl Glycosides of N-Acetylgalactosamine as Acceptors
specific activity (mU/mg)
enzyme
mutations
5 mM GalNAc-R-OBna
5 mM GalNAc-ꢀ-OBna
CgtB∆C30-MalE
E8b
NONE
E234D
178
796
1110
593
2510
301
1205
590
18
104
111
51
274
55
E14b
N26K, C40S, K50R, K151E, L227G
K130I, L70I
N26K, L68I, K151E, L227G, E234D
K151E
K151E, E234D
L227G
E20b
S42c
HS1d
HS2d
143
65
HS3d
a Errors in the data were less than 8%. b Error-prone PCR library. c DNA shuffling library. d Hot spot mutants constructed using site directed
mutagenesis.
Figure 5. Evolution pathway of ꢀ-1,3-galactosyltransferase CgtB from C. jejuni OH4384. In an earlier report, CgtB∆C30-MalE was constructed and
showed a much higher activity than the native CgtB.29 In our current work, the CgtB∆C30-MalE gene was randomized in stage I using the Gene Morph II
Mutagenesis kit, ∼2 × 107 colonies were screened, and several improved mutants were obtained. In stage II, the positive mutants from stage I were shuffled
together with the parental CgtB∆C30-MalE gene, 5 × 106 colonies were screened, and the best mutant in this study, S42, was isolated. In stage III, specific
mutants were constructed by site-directed mutagenesis and their effects on the activity improvement were determined and analyzed. The specific activities
were measured using GalNAc-R-OBn as acceptor.
six clones were randomly selected after the third round of sorting
and about 40% of them showed much higher activities than the
wild type CgtB∆C30-MalE, as indicated by the fluorescent
intensity of the cells under UV irradiation in the presence of 2
and 3 (Figure 4b). Indeed, the specific activities of these mutants
for transfer of galactose to simple glycosides of N-acetylgalac-
tosamine showed significant improvements over that of the
parent enzyme CgtB∆C30-MalE (Vide infra). Eleven of the
CgtB genes carrying apparent beneficial mutations were se-
quenced and five of these were found to be of identical
composition (equivalent to the mutant E14 in Table 1, see the
Supporting Information). This finding suggests that these
mutants have the highest activities in the entire EP-PCR library
and have therefore been substantially enriched during the FACS
screening.
In an effort to improve upon the already significant gains in
glycosyltransferase activity obtained via the EP-PCR approach,
CgtB genes on plasmids from the enriched population (Figure
4a) were backcrossed with the parental CgtB gene by DNA
shuffling, giving rise to a second library consisting of ∼5 ×
106 colonies. After three rounds of FACS as above, a total of
48 positive clones were assessed for improvements in activity
relative to CgtB∆C30-MalE, as suggested by the fluorescence
intensity of the cells under UV irradiation in the presence of 2
and 3. From within this group, the CgtB genes from eight clones
with improved fluorescence retention evident to the naked eye
were sequenced. Among these proteins, the most active enzyme
S42 (Figure 4c), contained five amino acid mutations. As
previously observed after FACS sorting of the EP-PCR library,
the enriched pool contained multiple enzymes bearing the same
mutations. In this case, three of the eight sequenced CgtB genes
were identical to the most active mutant S42, further substantiat-
ing the capacity of this screening method to effectively enrich
the most active colony in a large population. Inspection of the
mutations observed in positive clones from the EP-PCR library
and the shuffling library identified three “hot spots”, Lys151Glu
(11 times), Leu227Gly (9 times), and Glu234Asp (8 times), sites
at which mutations appear particularly beneficial to activity.
To investigate effects of each hot-spot mutation on catalytic
activity, the individual mutants K151E, L227G, and K151E/
E234D were constructed by site-directed mutagenesis and the
kinetic parameters of each were determined along with those
for the wild-type CgtB and some of the most active variants
from the EP-PCR and DNA shuffling libraries, which included
the E234D single mutant. The enzymes were expressed and
purified to greater than 90% homogeneity as described in the
Supporting Information. Specific activities of these mutant
enzymes with UDP galactopyranose donor and the R- and
ꢀ-linked benzyl glycosides of GalNAc (GalNAc-OBn) as
acceptors were determined and the data obtained are presented
in Table 1. In accordance with their fluorescence profiles, all
of the variants were significantly more active than the parent
CgtB∆C30-MalE. Mutant S42 showed the highest activity, an
increase of 14-fold over the parent, irrespective of the anomeric
configuration of the acceptor. Insight into the origin of this rate
increase is obtained from data on the three individual hot-spot
mutations (K151E, L227G, E234D). Activities of these three
mutants were 1.7-, 3.3-, and 4.5-fold higher than wild-type when
using RGalNAc-OBn acceptor, and 3.0-, 3.6-, and 5.8-fold
higher for ꢀGalNAc-OBn. Indeed, similar activity enhancements
were seen for each substrate with each mutant, clearly showing
that we have eliminated false positive mutants arising from the
9
J. AM. CHEM. SOC. VOL. 132, NO. 30, 2010 10575

A R T I C L E S
Yang et al.
Table 2. Kinetic Parameters for the Transfer of Galactose from
UDP-Galactopyranose to GM2 Oligosaccharide by Selected CgtB
Mutants
effects of the hot spot mutations appear to be independent of
each other, thus non-interacting (Figure 5). On the other hand,
the contributions of the N26K and L68I mutations to the rate
enhancement seen in S42 seem to be very small, if not
negligible, since K151E, L227G, and E234D already account
for most of the activity increase.
substrate
UDPGala,d
GM2 oligosaccharide,d
KM (mM) kcat/KM
c
c
enzyme
kcat (min-1
)
KM (mM) kcat/KM kcat (min-1
)
To assess the effects of these mutations on glycosylation of
a physiologically relevant substrate, kinetic parameters for the
transfer of galactose to the GM2 oligosaccharide were determined
(see Supporting Information for structures of acceptor and
product), and these are shown in Table 2. The specificity
constant (kcat/KM) of the double mutant K151E/E234D for GM2
is greater than that of either single mutant K151E or E234D
but lower than that of the S42 mutant, which contains all three
“hotspots” (K151E, E234D, L227G), indicating that the effects
of these mutations are essentially additive for acceptor kinetics.
The influence of the K151E mutation is exerted via an
improvement in kcat, while L227G and E234D affect activity
via decreased KM values. Indeed the KM value of S42 for GM2
is some 80-fold lower than that of the parent enzyme. With the
tetrasaccharide acceptor, a 16-fold improvement in kcat/KM was
observed for S42 over the parent enzyme.
CgtB∆C30-MalE
S42
K151E
E234D
L227G
54
34
150
65
55
170
0.22
0.03
0.24
0.07
0.11
0.08
250
1160
630
900
490
460
91
1080
120
140
410
23
0.3
26
1.4
2.3
2.8
20
320
42
88
61
K151E/E234D
2200
150
a Kinetic measurements at 5 mM GM2 oligosaccharide and variable
UDP-Gal concentration. b Kinetic measurements at 10 mM UDP-Gal
c
and variable GM2 oligosaccharide concentrations. Measured in min-1
mM-1
.
d The errors in the kcat and KM data are ∼5%, and the errors in
the kcat/KM data are less than 10%.
Scheme 2. Galactosylation of 4 Catalyzed by Native Enzyme or
Mutants, Including E14 and S42, Yields the Same Product
To confirm that the observed rate enhancements are of
practical use, we sought to establish that the mutants generated
the same reaction products as the parent enzyme. The galacto-
sylation of p-nitrophenyl 2-acetamido-ꢀ-D-galactopyranoside 4
was compared using CgtB∆C30-MalE, and mutants E8, E14,
E20, and S42 (Scheme 2). Analysis of the reactions by thin-
layer chromatography provided initial confirmation that the same
products were formed in each case. This was then confirmed
for the products formed by CgtB∆C30-MalE, E14, and S42
generation of dye binding sites. Our strategy of simultaneous
screening against differentially labeled substrates therefore
appears to have been successful. It is also of interest that the
1
through their isolation and analysis by H NMR spectroscopy.
Scheme 3. Some Useful Transformations Catalyzed by CgtB Galactosyltransferase Mutant S42
9
10576 J. AM. CHEM. SOC. VOL. 132, NO. 30, 2010

FACS as a Screening Method for Directed Evolution of GTs
A R T I C L E S
Table 3. The Specific Activities (mU/mg) of CgtB∆C30-MalE and Mutant S42 with Different Donor and Acceptor Substrates
acceptor
donor
CgtB∆C30-MalE spec. act.
S42 spec. act
enhancement (fold)
Asialo GM2-FCHASEa
GalNAc-R-OBn
UDP-galactose
UDP-glucose
100
<0.13
2800
38.0
28
>298
a For structures of FCHASE oligosaccharides, see the Supporting Information.
Similarly, the pentasaccharide product formed upon galactosy-
lation of GM2 oligosaccharide with CgtB mutant S42 had an
identical 1H NMR spectrum to that obtained from an authentic
sample of GM1a oligosaccharide.
or the other CgtB variants found in previous work.29 Thus, use
of S42, in conjunction with improved GalNAc transferases,
would make this route practical and economically viable.35,36
Conclusion
The synthetic potential of the best mutant identified, S42,
was demonstrated by synthesis of GM1a phenyl glycoside (7,
GM1OPh) from its precursor tetrasaccharide GM2OPh 6 as shown
in Scheme 3a, and further explored by testing its donor sugar
specificity using acceptor 4. In keeping with the specificity of
wild-type CgtB, neither UDP-GlcNAc nor UDP-GalNAc served
as donors. However, surprisingly UDP-Glc served as a substrate
for CgtB S42. Specific activity was enhanced at least 300-fold
over the parent enzyme, the activity of which was only barely
detectable with UDP-Glc donor (Scheme 3b, Table 3). This
tolerance of S42 toward structural modifications at C-4 of the
galactose donor suggests its potential in the chemoenzymatic
synthesis of C-4 modified variants of ganglioside GM1 or
O-glycans containing the T-antigen structural subunit. Moreover,
S42 should be an excellent starting point for the evolution of
novel glycosyltransferases with new substrate profiles.
Another important class of synthetic targets is that of the
asialo GM1a-type oligosaccharides such as 10. While these
compounds can be obtained by sialidase-catalyzed degradation
of the more complex GM1a oligosaccharides (Scheme 1c), this
reaction is typically only performed on an analytical scale.34
Further, in addition to the fact that such a degradative approach
is inherently inefficient, the commercial ganglioside preparations
that are needed are isolated from natural sources such as bovine
brain and canine blood, and are, thus, expensive and potential
vectors for cross-species contamination by prions or other
infectious agents, and, therefore, are unattractive.
The use of ultra-high-throughput screening methods such as
FACS sorting allows the interrogation of mutant libraries of a
size that would be inaccessible by plate-based assays on either
cost or time considerations. This study generalizes our previous
approach,27 both by demonstrating its application to neutral
sugar transferases and by showing that simultaneous use of
chemically and spectroscopically distinct fluorescent tags on the
same sugar acceptor removes the selective bias toward evolution
of dye binding sites. It is probable that, by the use of acceptors
with both different sugar moieties and different fluorescent dyes,
one could create positive and negative selection pressures for
the isolation of highly active GT mutants with tailored alterations
in substrate specificity. The merit of this methodology lies not
only in its thousand times greater throughput than previous
microtiter assays, but also in its use of E. coli cells as
microreactors. In such a small system, the volumes of reagent
needed for screening are millions-fold less than that required
for robotic screening, resulting in huge savings in cost and
availability of reagents.37
This work paves new avenues for the engineering of almost
all GTs. Given the available information on sequence, structure,
and function of sugar transport proteins (http://www.membran-
etransport.org/; http://www.tcdb.org/),38 it may be possible to
develop a repertoire of engineered E. coli strains that can be
used as a tool box for the directed evolution of many GTs. This
technique will greatly increase the rate at which novel engi-
neered GTs are identified, thereby contributing to the continued
development of glycoscience.
Since asialo GM2-type oligosaccharides (such as 9) are
available from lactosides via a single GalNAc transferase-
catalyzed step, transfer of a Gal residue to asialo GM2-type
oligosaccharides by Campylobacter CgtB represents a viable
approach toward the assembly of asialo GM1a structures. Mutant
S42 transfers galactose to the asialo GM2 oligosaccharide 9 30-
fold faster than do either the parental enzyme CgtB∆C30-MalE,
Acknowledgment. We thank Amir Aharoni for helpful discus-
sions, Hongming Chen for providing activated bodipy dye, and
Marie-France Karwaski for technical help. We acknowledge the
China Scholarship Council for a fellowship to G.Y., the Michael
Smith Foundation for Health Research for a fellowship to J.R.R.,
the Canada Research Chairs Program for a Tier 1 chair to S.G.W.
and the Natural Sciences and Engineering Research Council of
Canada and the Canadian Institutes of Health Research for grant
support.
(34) Iwamori, M.; Kaido, T.; Iwamori, Y.; Ohta, Y.; Tsukamoto, K.; Kozaki,
S. J. Biochem. 2005, 138, 327–334.
(35) Gilbert, M.; Karwaski, M. F.; Bernatchez, S.; Young, N. M.; Taboada,
E.; Michniewicz, J.; Cunningham, A. M.; Wakarchuk, W. W. J. Biol.
Chem. 2002, 277, 327–337.
Supporting Information Available: Structures, syntheses,
sequencing results, purification of mutants, and spectra as
mentioned in text. This material is available free of charge via
the Internet at http://pubs.acs.org.
(36) Yamashiro, S.; Haraguchi, M.; Furukawa, K.; Takamiya, K.; Yama-
moto, A.; Nagata, Y.; Lloyd, K. O.; Shiku, H.; Furukawa, K. J. Biol.
Chem. 1995, 270, 6149–6155.
(37) Agresti, J. J.; Antipov, E.; Abate, A. R.; Ahn, K.; Rowat, A. C.; Baret,
J. C.; Marquez, M.; Klibanov, A. M.; Griffiths, A. D.; Weitz, D. A.
Proc. Natl. Acad. Sci. U.S.A. 2010, 107, 4004–4009.
(38) Saier, M. H. Mol. Microbiol. 2000, 35, 699–710.
JA104167Y
9
J. AM. CHEM. SOC. VOL. 132, NO. 30, 2010 10577