Copyright by Wei-Cheng Lu 2013

140
Copyright by Wei-Cheng Lu 2013

Transcript of Copyright by Wei-Cheng Lu 2013

Copyright

by

Wei-Cheng Lu

2013

The Dissertation Committee for Wei-Cheng Lu Certifies that this is the approved

version of the following dissertation:

EVOLVED ENZYMES FOR CANCER THERAPEUTICS AND

ORTHOGONAL SYSTEMS

Committee:

Andrew D. Ellington, Supervisor

George Georgiou

Walter Fast

Christian Whitman

Hal Alper

EVOLVED ENZYMES FOR CANCER THERAPEUTICS AND

ORTHOGONAL SYSTEMS

by

Wei-Cheng Lu, B.S.; M.S.

Dissertation

Presented to the Faculty of the Graduate School of

The University of Texas at Austin

in Partial Fulfillment

of the Requirements

for the Degree of

Doctor of Philosophy

The University of Texas at Austin

August 2013

Dedication

To my mom and dad, and my brother and sister

For your love and support.

Acknowledgement

I would like to first thank my advisor, Andy Ellington. He gave me a lot of time

and freedom to try and learn in his lab. You need to be very independent to survive in his

lab and I survive.

I would also like to thank my committee members, especially George Georgiou

for giving me an opportunity to join the CGL group. Working in the therapeutics field is

my favorite. I really enjoy in this work. Also, I am grateful to Dr. Fast, Dr. Whitman

and Dr. Alper for giving me lots of feedback to my dissertation and work.

I would like to thank members in the Ellington lab. Thank Matt to guide me in

the beginning and help me a lot in BirA work. I would like to thanks Christien, Supriya,

Amrita. You guys give me warm hugs all the time. Thank Adam, Jared, Tony and

Bingling. You made this lab fun and it is important to have fun in the lab. Thank

Michelle, Arti, Paulina and Jorge. You help me not only on experiments but also on

many other works. Thank Randy, Alex and Xi. You are not just postdocs but also role

models. You teach us how to fight with Andy and expose his weakness to us.

Thank Kaiyin, Reilin, Kengming, Yating and Li. You are my family at Austin.

Finally, I would like to thank my parents. Thank for your sacrifice and support.

In the past several years, I felt upset in most of time and even wanted to give up in

the middle of this journey. Thank all of you to encourage and support me from the

beginning to the end.

vi

Evolved Enzymes for Cancer Therapeutics and Orthogonal System

Wei-Cheng Lu, Ph.D.

The University of Texas at Austin, 2013

Supervisor: Andrew D. Ellington

Directed evolution has been explored for a long time. Various ideas, methods,

have been shown to be feasible and successful in the enzyme field. We were interested in

evolving enzymes for applications. Therefore, we evolved human cystathionine gamma-

lyase (hCGL) and E. coli biotin ligase for therapeutic and biotechnology applications.

Wild-type human cystathionine gamma-lyase does not have any methionine-

degrading activity, unlike the high methionine-degrading abilities of bacterial methionine

gamma-lyase (MGL) found in Pseudomonas putida. The ability to engineer hCGL to

breakdown methionine can be a potential cancer treatment by targeting the methionine-

dependent cancer cells. However, the methionine-degrading activity of previously

engineered hCGL has only shown 1% activity compared to MGL, too low to be useful in

practical cancer therapeutics. By using a combination of protein design and phylogenetic

analysis, we further evolved hCGL to achieve a higher methionine-degrading activity,

with one variant displaying as much as 7% activity compared to bacterial MGL, making

it a more likely candidate in cancer treatment.

In addition, it has been shown that new orthogonal pairs of biotin protein ligase

and biotin have many biotechnology applications. Therefore, we have developed

selection scheme for directing the evolution of E. coli biotin protein ligase (BPL, gene:

vii

BirA) via in vitro compartmentalization, and have altered the substrate specificity of BPL

towards the utilization of the biotin analogue desthiobiotin. Following just 6 rounds of

selection and amplification several variants that demonstrated higher activity with

desthiobiotin were identified. The best variants from Round 6, BirA6-40 and BirA6-47,

showed 17-fold and 10-fold higher activity, respectively, their abilities to use

desthiobiotin as a substrate. Further characterization of BirA6-40 and the single

substitution variant BirAM157T revealed that they had 2- to 3-fold higher kcat values for

desthiobiotin, and 3- to 4-fold higher KM values. The kcat/KM values for these enzymes

were around 0.7-fold that of BirAwt. It is interesting the selections did not lower the KM

for desthiobiotin and actually led to a less efficient enzyme. This is an example of how

“you get what you select for”. Because peptide:DNA conjugates were distributed such

that there was on average one template or less per emulsion compartment there was

selection only for the catalytic rate (kcat) of desthiobiotinylation and not for turnover.

Given these conditions, it might be anticipated that the peptide substrate, rather than

desthiobiotin, should be bound better by the winning variants, and in fact BirA6-40 showed

a reduced KM value for BAP.

viii

Table of Contents

List of Tables ......................................................................................................... xi

List of Figures ....................................................................................................... xii

List of Figures ....................................................................................................... xii

Chapter 1: Directed enzyme evolution: In vitro selection of protein for therapeutic

and biotechnology applications.......................................................................1

In vitro selection of proteins via emulsion compartments ..............................1

Directed evolution of proteins in emulsions ..........................................4

Optimization of binding................................................................6

Optimization of catalysis ..............................................................7

Technical issues in emulsion selections: the oil:aqueous phases.9

Technical issues in emulsion selections: in vitro transcription and

translation...........................................................................10

Technical issues in emulsion selections: template recovery and

amplification ......................................................................12

The future of emulsion selections: double-bagging............................13

The future of emulsion selections: selection for fecundity........15

The future of emulsion selections: synthetic circuits.................17

Advantages and disadvantages ............................................................17

Therapeutic target: human cystathionine gamma-lyase................................18

Cancer and amino acid-dependence.....................................................18

Methionine-dependent cell activities ...................................................20

Pre-clinical trials and Clinical trials.....................................................21

Biotechnology target: E.coli biotin protein ligase ........................................23

Streptavidin:biotin................................................................................23

Applications of streptavidin:biotin orthogonal pair.............................23

Biotinylation reaction...........................................................................24

Function of E.coli biotin protein ligase................................................24

References.....................................................................................................27

ix

Chapter 2: Engineering human cystathionine gamma-lyase to degrade methionine for

cancer treatment ............................................................................................34

Introduction...................................................................................................34

Methionine gamma-lyase.....................................................................34

Cystathionine gamma-lyase .................................................................35

Dr. Stone and Olga Paley’s work.........................................................36

Results...........................................................................................................40

Phylogenetic analysis of MGLs and CGLs:.........................................40

Pilot experiment (test 11 selected positions from phylogenetic analysis)

.....................................................................................................41

Combinatorial library (11 positions from phylogenetic analysis) .......47

Comprehensive phylogenetic analysis of CGLs and MGLs................51

Substrate promiscuity...........................................................................57

Phylogenetic analysis on positions 59, 119 and 339 of hCGL ............59

Discussions ...................................................................................................62

Advantages and disadvantages ............................................................62

Other targets.........................................................................................65

Amino acid-depletion enzymes...................................................65

Antibody-directed enzyme prodrug therapy (ADEPT)...............65

Materials and methods ..................................................................................67

Site-directed mutagenesis ....................................................................67

Combinatorial library construction ......................................................67

96-well plate screening ........................................................................68

Kinetic analysis using MBTH..............................................................69

References.....................................................................................................70

Chapter 3: Directed evolution of the substrate specificity of E.coli biotin ligase 72

Introduction...................................................................................................72

Results and discussions.................................................................................75

Development of a scheme for BirA directed evolution .......................75

Library construction.............................................................................78

Selection for DTB utilization...............................................................80

x

Comparing DTB-utilizing variants ......................................................82

Kinetic characterizations of biotin ligase variants ...............................85

Evolutionary paths ...............................................................................88

Materials and Methods..................................................................................95

Library preparation ..............................................................................95

Cross-linking BAP with DNA .............................................................95

In vitro compartmentalization selection...............................................96

Gel-shift assay......................................................................................96

BPL protein purification ......................................................................97

Pyrophosphate detection ......................................................................97

References.....................................................................................................99

Chapter 4: Demonstration of cooperation and co-evolution of synthetic operon102

Introduction.................................................................................................102

Results and Conclusions .............................................................................105

Construction scheme..........................................................................105

Selection scheme................................................................................106

Operons are functional .......................................................................107

Mock selection ...................................................................................109

The selection of the representative pool (4 operons) .........................110

Non-specific DNA binding of anti-His antibody agarose beads .......111

References...................................................................................................119

Bibliography ........................................................................................................120

Vita…...................................................................................................................126

xi

List of Tables

Table 2-1: Kinetic analysis of Pseudomonas putida methionine gamma-lyase,

human cystathionine gamma-lyase and engineered human

cystathionine gamma-lyase (hCGL-NLV). ...................................38

Table 2-2: List of kinetic values for 22 hCGL variants in methionine

breakdown .......................................................................................45

Table 2-4: List of kinetic values of hCGL variants in methionine breakdown

...........................................................................................................56

Table 2-5: Substrate specificities of hCGL-NLV and hCGL-NLV+RGS+LM.

...........................................................................................................58

Table 2-6: Summary of engineered hCGL variants in methionine breakdown.

...........................................................................................................61

Table 3-1: List of substitutions of isolated biotin ligase variants from Round 6

pool ...................................................................................................82

Table 3-2: Substitution distribution of selected BirA pool (Round 6) by Next-

Generation sequencing ...................................................................91

xii

List of Figures

Figure 1-1: Display technologies for the directed evolution of proteins. .........3

Figure 1-2: In vitro compartmentalized selection for HaeIII methylase variants

.............................................................................................................5

Figure 1-3: In vitro compartmentalized selection for functional streptavidin

variants...............................................................................................8

Figure 1-4: Selections in double emulsions. ......................................................15

Figure 2-1: Enzyme reaction of methionine gamma-lyase. .............................35

Figure 2-2: Enzyme reaction of cystathionine gamma-lyase. .........................36

Figure 2-3: Structure comparison of human cystathionine gamma-lyase (PDB:

3COG) and Trichomoas vaginalis (PDB: 1E5E) with inhibitor

propargylglycine (PAG) .................................................................37

Figure 2-4: Evaluation of PEGylated hCGL-NLV in athymic mice bearing

LAN-1 xenografts............................................................................39

Figure 2-5: Phylogenetic analysis of cystathionine gamma-lyases and methionine

gamma-lyases...................................................................................41

Figure 2-6: Phylogenetic analysis of the carboxyl terminals of 5 eukaryotic

cystathionine gamma-lyases and 6 bacterial methionine gamma-

lyases.................................................................................................43

Figure 2-7: Kinetic analysis of engineered hCGL variants in methionine

breakdown .......................................................................................46

Figure 2-8: 96-well plate screening process ......................................................48

Figure 2-9: Combinatorial library screening ...................................................49

xiii

Figure 2-10:Kinetic analysis of hCGL-NLV and hCGL-NLV+RGS in

methionine breakdown ...................................................................50

Table 2-3: List of kinetic values of 20 hCGL variants in methionine

breakdown. 20 hCGL variants were purified for detailed kinetic

analysis .............................................................................................53

Figure 2-11:Kinetic analysis of hCGL-NLV+RGS+LM in methionine

breakdown .......................................................................................54

Figure 2-12: Methionine and similar molecules. ...............................................58

Figure 2-13: Phylogenetic analysis at positions 59, 119 and 339 .....................60

Figure 2-14:Kinetic analysis of hCGL-NLV+RGS+LM and hCGL-

IAV+RGS+LM in methionine breakdown. ..................................61

Figure 3-1: Biotinylation reaction .....................................................................73

Figure 3-2: Selection scheme of directed evolution of BPL using IVC ..........76

Figure 3-3: Mock selection. Before the real selection, a mock selection was

performed to validate the selection scheme ..................................78

Figure 3-4: Two libraries construction. ............................................................80

Figure 3-5: The selection process of BPL library 1 and BPL library 2. .........81

Figure 3-6: Activity assay of isolated variants (gel-shift assay). .....................84

Figure 3-7: Kinetic characterization of BirAwt, BirA6-40 and BirAM157T86

Figure 3-8: Activity assay of variants with single reintroduced substitutions (gel-

shift assay)........................................................................................89

Figure 3-9: Mapping Met-157 onto the structure of E.coli biotin ligase with

biotin. E.coli biotin ligase is shown here with biotin (1HXD) ...93

Figure 4-1: Structure of Lac operon................................................................103

Figure 4-2: Synthetic operons ..........................................................................104

xiv

Figure 4-3: A Representative synthetic operon pool ......................................106

Figure 4-4: Selection scheme of synthetic operons using IVC .......................107

Figure 4-5: Protein expression of synthetic operons (western blot) .............109

Figure 4-6: Mock selection. ...............................................................................110

Figure 4-7: One round of real selection ...........................................................111

Figure 4-8: Dissection of the selection ..............................................................114

Figure 4-9: One round of selection with new wash buffer ............................115

Figure 4-10:Different capture approach..........................................................116

1

Chapter 1: Directed enzyme evolution: In vitro selection of protein for

therapeutic and biotechnology applications

IN VITRO SELECTION OF PROTEINS VIA EMULSION COMPARTMENTS

The directed evolution of proteins has been used to generate enzymes with a wide

variety of kinetic and physical properties (Griffiths and Tawfik 2000; Bloom and Arnold

2009; Romero and Arnold 2009). In the past, directed evolution of proteins has typically

been carried out in the context of cells. This is both because the production of proteins

requires the complex and difficult-to-maintain translation apparatus, and because the

cellular membrane provides a convenient means of separating genotypes and phenotypes

from one another. However, such in vivo selections have a number of disadvantages.

Manipulating live organisms is time- and labor-consuming, and it is very difficult to

control selection conditions and stringencies within a cell. Quite often these experiments

yield survival of the organism by an otherwise unanticipated route, rather than selection

for a particular protein property. In addition, it is difficult to work outside the well-

known boundary conditions of living systems; for example, selecting for proteins that

operate at extremes of pH or that utilize unnatural amino acids. Also, it has been found

that overexpression of up to 51% of endogenous E.coli ORFs can cause severe growth

defects, while 77% of these ORFs cause growth inhibition (Kitagawa, Ara et al. 2005).

These results strongly argue that in vivo selection experiments for many different

functions might disfavor truly active protein variants. Finally, the population sizes that

can be examined in cells are inherently limited by the gross inefficiencies of

transformation or transfection. In order to overcome these difficulties in vitro systems for

directed evolution were developed.

2

In vitro selections use much the same principles for molecular evolution as in vivo

selections. A pool of heritable information is generated, parsed via function, and

selectively amplified. There are various ways to achieve these three steps with proteins

in vitro, including ribosome display, mRNA display and CIS display (Lipovsek and

Pluckthun 2004; Ullman, Frigotto et al. 2011) (Figure 1-1). For example, ribosome

display involves the in vitro translation of mRNAs that lack a stop codon and that

therefore couple the nascent peptide indirectly to the mRNA via the stalled ribosome.

Ribosome display was first demonstrated by Jozef Hanes and Andreas Plückthun and

used for selection of a single-chain antibody fragment (scFv) that bound hemagglutinin.

In five rounds of purely in vitro selection, the anti-hemagglutinin scFv was enriched from

the starting population in which it was diluted by an anti-β-lactam scFv by 108-fold

(Hanes and Pluckthun 1997; Zahnd, Amstutz et al. 2007). mRNA display is similar to

ribosome display but relies on the formation of a covalent link between a puromycin-

containing mRNA and the nascent translated peptide. In a model selection, a mRNA

fusion with Myc peptide epitope could be enriched 20-40 fold from the mixed population

by immunoprecipitation (Roberts and Szostak 1997). The Szostak group also used this

system for the de novo identification of ATP-binding motifs from a completely random-

sequence library. They found that one functional protein could be recovered from a

sequence space that spanned 1011 different proteins (Keefe and Szostak 2001; Lipovsek

and Pluckthun 2004). In contrast to mRNA display, CIS display relies on a linkage (non-

covalent or covalent) between the template DNA and the nascent translated peptide

instead of linkage of the mRNA to the nascent peptide. The McGregor group has used

RepA, a DNA replication initiator, to link the DNA template and nascent peptide

(Odegrip, Coomber et al. 2004). CIS display can also be engineered to provide a covalent

3

linkage to the DNA template via the replication initiator protein, P2A (Reiersen, Lobersli

et al. 2005).

Figure 1-1: Display technologies for the directed evolution of proteins. Ribosome

display, mRNA display and CIS display provide different means of linking

genotype with phenotype. For ribosome display, mRNA sequences are

engineered without stop codons and the ribosome stalls, leaving the mRNA

and the nascent peptide linked to one another via the ribosome. In mRNA

display, puromycin is covalently conjugated to mRNA, and couples itself to

a newly translated peptide, providing a direct covalent linkage between

mRNA and peptide. For CIS diplay, the RepA (or other DNA binding)

protein binds a specified sequence on a DNA template, such as the ori

sequence. Peptides fused to RepA can bind to their templates, a non-

covalent linkage similar to ribosome display.

The general advantages of these methods relative to in vivo selection are larger

library sizes and ease of manipulation. However, it can be difficult to fully control the

selection environment, especially given that many of the components necessary for

translation or enzymatic activity are diffusible. This is why it has proven useful to try to

both control expression and restrict diffusion within emulsions. Water-in-oil emulsions

can contain 1010

compartments and are stable at 25o C for 24 hours with no obvious

change in the distribution of compartment sizes (Tawfik and Griffiths 1998). This

4

technology has enabled the revolution in NextGen DNA sequencing (Margulies, Egholm

et al. 2005), but has also proved to be of great utility for the directed evolution of

proteins.

Directed evolution of proteins in emulsions

Directed evolution via in vitro compartmentalization (IVC) was first

demonstrated in a seminal paper by Dan Tawfik and Andrew Griffiths (Tawfik and

Griffiths 1998). First, these authors showed that functional dihydrofolate reductase

(DHFR) and HaeIII methyltransferase could be produced via in vitro transcription and

translation in compartments and still retain over 60% activity compared to proteins

produced in nonemulsified reactions. Functional HaeIII methyltransferases could then

individually feedback on the survival of their own templates. In a population of folA

(encoded DHFR) mixed with M.HaeIII (encoded methyltransferase HaeIII), methylated

M.HaeIII genes were protected from HaeIII endonuclease digestion and could be

enriched from a pool by a factor of 107-fold after only 2 rounds of selection (Figure 1-2).

For all selections, some manner of genotype-phenotype linkage is necessary.

There are a surprising number of ways to link the genotype and phenotype during IVC

selections. For example, noncovalent linkage by the co-scalled STABLE system (Doi

and Yanagawa 1999), zinc finger protein binding to DNA templates (Sepp and Choo

2005) and covalent linkage by a HaeIII methyltransferase fusion protein (Bertschinger

and Neri 2004) and SNAP-tag (Stein, Sielaff et al. 2007) have been used for the selection

of binding proteins. We will examine a number of these systems below. For enzymes,

fluorescent substrates and FACS have been used to confine genotype and phenotype

either on the surface of beads or in a double emulsion (Bernath, Hai et al. 2004; Aharoni,

5

Amitai et al. 2005; Mastrobattista, Taly et al. 2005). Again, we will see examples of this

with enzymes as diverse as restriction endonucleases, methyltransferases, polymerases,

and phosphotriesterases.

Figure 1-2: In vitro compartmentalized selection for HaeIII methylase variants

(Tawfik and Griffiths 1998). Tawfik and Griffiths first demonstrated that

functional M.HaeIII could be produced in emulsified compartments. In a

model selection, templates for M.HaeIII and a different protein (DHFR,

encoded by the folA gene) were mixed at several different ratios (1:104,

1:105, 1:10

6, and 1:10

7 ). Only functional M.HaeIII (not functional DHFR)

methylated the restriction/methylation sites at the 3’-termini of associated

templates and thereby protected these templates from subsequent HaeIII

restriction enzyme digestion. After only two round selections, the M.HaeIII

gene could be enriched from the pool that started with a 1:107 ratio.

6

Optimization of binding

The methods that Tawfik and Griffiths developed can be generalized to changing

the substrate specificities of a variety of binding proteins. For example, the Ghadessy

group used an anti-p53 antibody to isolate p53 variants that bound to a double-stranded

DNA containing a so-called low response element. When introduced into cells, these

variants could also better mediate transactivation of genes controlled by low response

elements (Fen, Coomber et al. 2007). Similarly, Yu Chen et al. demonstrated that they

could carry out selections with a fusion between a RNA-binding protein and a DNA-

binding zinc finger. The RNA-binding protein could be evolved to recognize an

immobilized target RNA, while the DNA-binding zinc finger bound to the template

DNA; in essence the dual protein served as a bridge between the RNA target and the

DNA template. In this way, these authors demonstrated that the method should be able to

change the recognition specificity of RNA binding protein (Chen, Mandic et al. 2008).

Finally, Levy and co-workers carried out an emulsion-based selection to alter the binding

specificity of streptavidin. The gene encoding streptavidin was modified with the biotin

analogue desthiobiotin, and streptavidin variants that were able also able to bind their

own DNA templates were further isolated via binding of an anti-His antibody to a His tag

on the streptavidin (Figure 1-3). Interestingly, selected variants still bound biotin quite

well, but had greatly reduced off-rates and longer dissociation half-times for

desthiobiotin. Nonetheless, the specificity differences were great enough that the wild-

type and mutant proteins could be used in differential labeling schemes (Levy and

Ellington 2008).

7

Optimization of catalysis

The specificities of enzymes can also be altered by directed evolution in

emulsions. The Griffiths group has used in vitro compartmentalization and directed

evolution to change the recognition specificity of the M.HaeIII methyltransferase, and

identified enzyme variants that could methylate AGCC rather than the canonical

sequence GGCC. Moreover, the mutant methylase shows higher catalytic activity with

AGCC compared to wild-type M.HaeIII with the canonical sequence (Cohen, Tawfik et

al. 2004). Similarly, Nobuhide et al. have shown that it is feasible to use IVC to select

for restriction endonuclease activity. These authors introduced a FokI restriction site into

the template for the gene, and recovered cleaved DNA templates by ligation to to dUTP-

biotin (Doi, Kumadaki et al. 2004).

Enzymes other than those used as molecular biology reagents can also be

optimized. Griffths and Tawfik demonstrated that IVC and FACS could be combined for

the directed evolution of phosphotriesterases (Griffiths and Tawfik 2003). Genotype and

phenotype were linked by immobilizing both genes and translated proteins on

microbeads. The microbeads were then re-emulsified with enzymatic substrates. Finally,

the products were also captured on the microbeads and further detected via an anti-

product antibody binding and FACS. It should be noted that this method selects not only

for catalysis, but for multiple turnover catalysis, and the kcat of the phosphotriesterase was

improved by 63-fold. While library size is limited relative to other emulsion methods (

FACS only selects 107 clones per hour) this combined system can nonetheless be adapted

to the evolution of catalysts that are not readily engineered by other in vitro and in vivo

methods.

8

Figure 1-3: In vitro compartmentalized selection for functional streptavidin

variants. Biotinylated DNA templates encoding streptavidin variants were

compartmentalized in water-in-oil emulsions containing in vitro

transcription and translation reagents. Functional streptavidin variants

translated in compartments could bind to their encoding DNA templates.

After breaking the emulsion, all streptavidin variants were isolated via an

encoded His tag and the genes for the functional streptavidin variants were

recovered by PCR.

Additionally, IVC has been used not only for evolving proteins but also for

evolving RNA. Levy and co-workers developed an IVC selection for multiple turnover

RNA catalysts (Levy, Griswold et al. 2005). In this selection, microbeads that

immobilized a gene pool and one half of a RNA ligase substrate were emulsified with the

9

components of an in vitro transcription reaction and a fluorescent, second half of a RNA

ligase substrate. Transcribed, active ribozyme variants could then link the fluorescent

substrate to the microbeads containing their own templates. The most active ligase

variants were sorted following breaking of the emulsion by FACS, and the templates

were amplified in vitro and re-immobilized for additional rounds of selection and

amplification.

Technical issues in emulsion selections: the oil:aqueous phases

In vitro compartments are typically made of two phases, an aqueous phase and an

oil-surfactants phase. The aqueous phase contains genotype and what are reagents for

expression of phenotype expression. In the case of proteins, this will typically be

transcription and translation, but for ribozymes only transcription is required. The

function of oil-surfactant phase is to produce artificial membranes separating each

aqueous compartment and confining each genotype and phenotype. By stirring the

aqueous phase with the oil-surfactant phase, typically upwards of 1010

compartments can

be generated in a 1 mL emulsion.

The composition of the oil-surfactant phase is a critical issue, in part because the

hydrophobic phase or molecules therein can inhibit complex biochemical processes such

as translation, and in part because the interface can denature newly produced proteins

(Miller, Bernath et al. 2006). In order to stabilize emulsions to heat thermal cycling,

0.05% Triton X-100 or ABIL EM 90 can be added; the latter can prevent compartment

leakage over 35 thermal cycles (Williams, Peisajovich et al. 2006). In addition, unlike

other surfactants ABIL EM 90 does not lead to the rapid inhibition of rabbit reticulocyte

lysates (Ghadessy and Holliger 2004). Because of the idiosyncracies associated with

10

systems and proteins in emulsions, it is typically wise to test different oil-surfactant

phases with different aqueous phases (lysates, below) prior to carrying out a directed

evolution experiment (Davidson, Dlugosz et al. 2009). For example, an expressed

protein may have acceptable yields, but may not be necessarily active due to inhibition by

components of the emulsion reaction (e.g., surfactants).

Technical issues in emulsion selections: in vitro transcription and translation

There are a variety of options for how to produce proteins in emulsion. An E.coli

lysate is generally used as the aqueous phase for most of IVTT-IVC systems. However,

E.coli lysates cannot support post-translational modifications on eukaryotic proteins

while at least rabbit reticulocyte lysates appear to be able to. Moreover, it is apparent that

different proteins translate more or less well in different lysates, perhaps due to the

different surfactants that are available, as alluded to above (Tawfik and Griffiths 1998;

Griffiths and Tawfik 2003; Yonezawa, Doi et al. 2003; Ghadessy and Holliger 2004;

Chen, Mandic et al. 2008). Therefore, wheat germ or rabbit reticulocyte lysates have also

been adapted to emulsion methods.

In our experience, there are a wide variety of variables that impact the efficiency

of expression in these different lysates, and the complexity of the systems makes it

difficult to predict in advance which lysate or conditions may work best for which

emulsion selection. For example, template DNA purification is important for translation

yield in lysates. In our hands, phenol-extracted DNA seems to be better than DNA

purified via spin columns which is in turn better than DNA purified on gels (Davidson,

Dlugosz et al. 2009). A recent selection model for oxygen resistant [FeFe] hydrogenase

overcame expression inefficiencies by using emulsion PCR to increase the amount of

11

template DNA on beads, ultimately generating larger amounts of protein in each

compartment (Stapleton and Swartz 2010).

One commonality between good expression systems seems to be the absence or

inhibition of nucleases. Nucleases in lysates can quickly degrade DNA construct

(genotype) and thereby reduce both the phenotype expression and the recovery of a

functional genotype. In our experience, 1 ng of 2.3kb DNA could not be recovered from

E. coli lysate after 2 hours incubation at 30o C. Adding a non-specific carrier such as

salmon sperm DNA may help, as will making modifications to the DNA template, such

as biotinylation or the use of phosphorothiolated nucleotides (Takei, Kadomatsu et al.

2002). Also, it has been found that Gam protein, a bacteriophage λ protein, can inhibit

exonuclease RecBCD in lysates (Sitaraman, Esposito et al. 2004). A PCR-derived linear

template was protected by Gam against purified RecBCD for up to 4 hours, leading to an

increase in protein production. However, since there are other many other nucleases in

E.coli lysates (such as ExoIII, VIII, and EndoIV) it may be wise into the future to use a

fully defined, recombinant translation system (PURE) to avoid DNA degradation

(Shimizu, Kanamori et al. 2005). In order to enhance RNA production from those

mRNAs that are made, changes in translation initiation can be introduced. For example,

the EMCV IRES has been shown to increase protein expression in rabbit reticulocyte

lysates (Bochkov and Palmenberg 2006). In at least one instance, a construct with an

EMCV IRES increased luciferase expression 4- to 5-fold in rabbit reticulocyte lysates

relative to constructs with Kozak sequences. For similar experiments in wheat germ

extracts, the TMV IRES can be used (Yonezawa, Doi et al. 2003).

Once a protein is produced, there is no guarantee that it will be active, since the

emulsion interface is quite different than a phospholipid interface. In many instances,

proteins likely denature at the interface, even in the presence of emulsifying agents.

12

Once again, it may be necessary to screen emulsifiers to find those that are most useful

for functional expression of a given protein. Adding a bulk protein such as BSA or

negative charged surfactants such as sodium deoxycholate (Tawfik and Griffiths 1998)

can also prevent translated proteins from being trapped at the oil-water interface

(Ghadessy and Holliger 2004). The use of chaperones may also prove helpful in

avoiding interface-mediated denaturation. Parent et al. have found that GroEL- and

GroES-overexpression in lysates can rescue destabilized, temperature-sensitive

bacteriophage P22 coat protein assembly (Parent, Ranaghan et al. 2004). Similarly,

Tokuriki and Tawfik have demonstrated that overexpression of GroEL/GroES helps

enzymes to accumulate mutations that would otherwise be destabilizing, which in turn

allows more pathways to be followed during the evolution of new substrate specificities

(Parent, Ranaghan et al. 2004; Tokuriki and Tawfik 2009). Along these lines, the Tawfik

group has shown that a small library that is first allowed to accumulate neutral mutations

is a better starting point for the evolution of serum paraoxonases (PON1) with different

substrate specificities than a library centered on the wild-type enzyme (Amitai, Gupta et

al. 2007; Gupta and Tawfik 2008). This is likely because mutations that are nominally

'neutral' can actually improve the stability of the wild-type enzyme to denaturation. The

use of such libraries would be a clear advantage in emulsion selections, as well.

Technical issues in emulsion selections: template recovery and amplification

Following the actual selection, it’s important to be able to efficiently recover

functional variants from the pool. Diethyl ether is typically used to break the emulsions

(Miller, Bernath et al. 2006), although other organic solvents such as chloroform and

hexane can also be used (Davidson, Dlugosz et al. 2009). In our experience, there are

13

two potential choke points for recovery. First, it is increasingly difficult to recover longer

genes, especially if you are recovering mRNA rather than DNA. This is likely related to

the aforementioned problem of nucleases in lysates. In addition, there is a related issue.

With low recovery, the possibilities for accumulating additional mutations or

amplification artifacts are proportionately greater, because larger numbers of

amplification cycles are required to generate material for additional rounds of emulsion-

based selection. This can actually lead to a situation not unlike the evolutionary

conundrum known as Muller's Ratchet, in which the mutation rate is sufficiently high and

the recovery rate is sufficiently low that the functionality of the population actually

decreases during the course of a selection, irrespective of stringency.

A secondary issue with respect to the recovery and amplification of mRNAs is

that secondary structure may impede reverse transcription. We have attempted to

circumvent this problem by creating synthetic genes whose mRNA transcripts will have

greatly reduced structure-forming potential.

The future of emulsion selections: double-bagging

It has proven possible to emulsify an emulsion, creating an aqueous suspension of

oil drops that in turn surround aqueous contents. First, an aqueous phase is mixed with

an oil-surfactant phase containing, for example, cholesterol, Span 60, and decane to

generate the water-in-oil emulsion (Figure 1-4). Subsequently, this first emulsion

reaction is further mixed with PBS containing 0.5% Tween 20 and the

aqueous:oil:aqueous compartments are extruded through a 8 µm pore-size membrane

(Miller, Bernath et al. 2006). Double emulsions are particularly useful for looking at

otherwise diffusible products or at individual cells (Bernath, Hai et al. 2004). For

14

example, the Griffiths group has evolved the classic E. coli Ebg enzyme to have greater

β-galactosidase activity (Mastrobattista, Taly et al. 2005). Enzyme variants were

translated in individual compartments, as with single emulsions, but the fluorescent

substrate of the reaction, fluorescein di-β-D-galactopyranoside, was then kept with an

individual template and enzyme variant by a second emulsion. The enzyme variants that

produced the greatest amount of fluorescent product (fluorescein) were separated by

FACS. Eight selected variants showed 300-fold higher kcat/KM values relative to the

unevolved Ebg enzyme.

Bacteria can also be ensconced within double emulsions. This is useful because

in vitro lysates are often inefficient at protein production, and because folding within the

bacteria may prevent the denaturation problems cited above. The Tawfik group has used

this technique to screen individual bacteria that express variants of serum paraoxonase

(PON1). One great advantage of using bacteria as production vehicles is that they can

generate from 104-10

5 enzyme molecules per bacterium per compartment, whereas in

vitro transcription and translation yield only about 10-102

molecules per compartment.

Turnover of a fluorescent substrate and FACS resulted in the identification of enzymes

that were 100-fold improved in thiolactonase activity. Serum paraoxonase can also

hydrolyze organophosphates at a low rate, and in a subsequent experiments variants were

identified that had 105-fold improved activity (kcat/KM) against a coumarin derivative of

Sp-cyclosarinhigher (Gupta, Goldsmith et al. 2011).

15

Figure 1-4: Selections in double emulsions (Bernath, Hai et al. 2004; Miller, Bernath et

al. 2006). A double emulsion further encapsulates water-in-oil emulsions

within an additional aqueous phase. The internal aqueous phase still

contains the in vitro transcription and translation reagents necessary for

protein expression. Surfactants such as Span 60 or Tween 20 provide a

boundary between the internal aqueous phase and the outer oil phase. These

surfactants also help to stabilize encapsulation in a second aqueous phase,

creating an environment akin to a lipid-coated cell. Those ribozyme or

protein templates that generate the most fluorescent products within the first

aqueous layer can be sieved from the multitude of cell-like compartments by

FACS.

The future of emulsion selections: selection for fecundity

While turnover of substrates is a useful way to select for the activity of many

enzymes, self-amplification can be used to select for polymerase function. The Holliger

group has developed a system they term “compartmentalized self-replication” (CSR) for

polymerase selections. They first express Taq DNA polymerase in cells, and then

compartmentalize the cells in emulsions with amplification reagents. Thermocycling is

also carried out directly in emulsions, with the first denaturation step lysing the bacteria.

16

Those polymerase variants that are most active will better amplify their own genes

(Ghadessy, Ong et al. 2001). DNA polymerases with higher thermostabilities (the

isolated Taq clone has 11-fold longer half-life at 97.5oC than the wild type Taq enzyme)

and that are more resistant to the DNA polymerase inhibitor, heparin (130-fold better

than wild type) have been evolved. The same strategy has also been used to expand the

substrate specificity of Taq DNA polymerase. One mutant has been isolated that can

actually use both dNTP and NTP equally well, while another has been shown to use 2’-

substituted nucleotides (Ong, Loakes et al. 2006). The Holliger group has now worked

on evolving polymerases to further expand substrate utilization to primers with a

hydrophobic base analog (Loakes, Gallego et al. 2009) and to the nucleotides 5-

nitroindole, 5-nitroindole-3-carboxamide, FITC-12-dATP(Ghadessy, Ramsay et al.

2004), biotin-16-dUTP (Ghadessy, Ramsay et al. 2004), Cy3-dCTP, and Cy5-dCTP

(Ramsay, Jemth et al. 2010).

Additionally, it is worth mentioning selective gene amplification (SGA) (Kelly

and Griffiths 2007), a method that further mimics natural selection. As an example,

active or inactive phosphotriesterase genes (OPD) and a substrate were immobilized on

streptavidin-coated microbeads. The two different microbead populations (active and

inactive) were emulsified with an in vitro transcription and translation reaction.

Following expression and breaking of the emulsion, conversion of the substrate to a

product was detected by the addition of anti-product antibodies coupled to primers

necessary for the amplification of the original gene. When the microbeads were re-

emulsified with a PCR reaction mixture those genes that originally yielded active

phosphotriesterase variants were preferentially amplified, and could participate in

subsequent rounds of selection and amplification. This method might be adapted to a

wider variety of reactions, depending on the availability of good anti-product antibodies.

17

The future of emulsion selections: synthetic circuits

While individual proteins and enzymes can be expressed in emulsions, it would

be interesting and useful to increase the complexity of selection schemes by having

multiple steps in a synthetic circuit feedback on survival. For example, the Holliger

group demonstrated that CSR could be adapted to select not only polymerases, but also

the nucleoside diphosphate kinase (NDK). In this selection scheme, functional NDK

molecules provide enough dNTPs for DNA polymerase to function and ultimately

amplify the DNA templates encoding functional NDKs. We are similarly working on

demonstration of protein cooperation in a synthetic operon. The desthiobiotin-utilizing

streptavidin variant (Levy and Ellington 2008) that was previously selected has been

combined with a selected biotin ligase (BirA) variant that can also utilize desthiobiotin.

By adding a peptide substrate to the DNA template encoding the operon, it may be

possible to simultaneously select for desthiobiotin addition and capture.

Advantages and disadvantages

Emulsion-based selections have several advantages over competing methods.

They allow selection for multiple turnover reactions, unlike ribosome display, mRNA-

peptide display and CIS display, which primarily select for binding. By avoiding the

need for cells, toxic substrates or unnatural substrates can be used in selections.

Emulsion selections also provide much greater control over the selection process. For

example, by iteratively lowering substrate concentrations, variants with improved KM

values can be selected; by reducing reaction times or adding competitive substrates,

variants with improved kcat values can be isolated. One of the greatest advantages of

emulsion methods is that larger library sizes can be sieved, but this advantage remains

largely unrealized because of the many disadvantages that attend the method.

18

As we have already seen, some of the primary problems with emulsion selections

have to do with the robustness of protein production. Beyond this consideration, there is

considerable variability between emulsion reactions, both from batch to batch and even

within batches (not all emulsion bubbles are of equal volume). In current methods, stir

bars (Yonezawa, Doi et al. 2003), homogenizers (Agresti, Kelly et al. 2005), or

microcapillary devices (Utada, Lorenceau et al. 2005) are used to produce emulsions.

While the first two methods are generally more variable, higher stirring speed or

homogenization may impair enzyme activity (Miller, Bernath et al. 2006). Fortunately,

the general drive for emulsion PCR has led to the development of devices for sorting

emulsions. For example, instruments made by the company RainDance are able to

generate consistent, stable emulsion droplets at a rate of 107 droplets/hour, and the

droplets can potentially not only be sorted, but also merged. Into the future such devices

should help to further standardize selection conditions, and may enable the step-wise

addition of substrates and other small molecules (Kintses, van Vliet et al. 2010).

THERAPEUTIC TARGET: HUMAN CYSTATHIONINE GAMMA-LYASE

Cancer and amino acid-dependence

Cancer has become a leading cause of human death and around one million

people have out of a variety cancers in the United of States. In the current state, cancers

are tough diseases to treat as is it is difficult to preferentially kill cancer cells without

hurting normal cells in the treatment. Conventional chemotherapy is toxic to cancer cells

but also has adverse effects on normal cells. Therefore, cancer cell-specific treatments

have been pursued as a better method.

19

In cancer biology research, it has been found that cancer cells exhibit different

metabolism compared to normal cells (Kroemer and Pouyssegur 2008). It has been

shown that some types of cancer are dependent on particular amino acids because they

are not able to synthesize these amino acids themselves. They must acquire them from

extracellular sources to maintain cell activities such as protein synthesis. Thus this

resource dependent might be a target for preferential killing of cancer cells.

In cancer research, it has been discovered that several amino acids are important

to maintain cancer cell growth such as asparagine, tyrosine, phenylalanine, glutamine,

leucine and methionine. Without these amino acids in culture media, some cancer cell

lines could not survive and stop growing in vitro. Hepatocellular carcinoma (HCCs),

pancreatic tumor and melanomas require exogenous L-arginine (Gong, Zolzer et al. 2000;

Ensor, Holtsberg et al. 2002; Bowles, Kim et al. 2008; Hernandez, Morrow et al. 2010).

Childhood acute lymphoblastic leukemia (ALL), ovarian carcinomas and non-Hodgkin

lymphomas need L-asparagine from extracellular sources (Cooney and Handschumacher

1970; Lorenzi, Llamas et al. 2008; Cantor, Panayiotou et al. 2012). Some cell lines of

neuroblastomas and glioblastomas and breast, lung, colon, kidney carcinomas have been

found to be methionine-dependent (Hoffman 1985; Breillout, Antoine et al. 1990;

Poirson-Bichat, Goncalves et al. 2000; Kokkinakis, Hoffman et al. 2001).

We are interested in methionine-dependent cancer cells for two major reasons. In

the past research, methionine-depletion has been shown to be a promising way to inhibit

tumor growth in vitro and in tumor-bearing animals. (Breillout, Hadida et al. 1987; Guo,

Lishko et al. 1993; Guo, Herrera et al. 1993; Poirson-Bichat, Gonfalone et al. 1997).

Moreover, synergistic effect of methionine-depleting total parenteral nutrition (lacking

methionine and cysteine) with 5-fluorouracil (5-FU) has been reported in tumor-bearing

rats and in clinical trails with gastrointestinal tract cancers (Goseki, Yamazaki et al.

20

1995). Therefore, methionine depletion in serum became our goal to pursue for cancer

treatment.

It is not clear why some cancer cells are sensitive to methionine deprivation. In

past studies, it has been reported that some methionine-dependent cancer cells associate

with a loss of function of methionine synthase, methylthioadenosine phosphorylase

(MTAP) or methylenetetrahydrofolate reductase (MTHFR). These enzymes are

important in the pathway of methionine and homecysteine metabolism (Cellarier,

Durando et al. 2003). Cancer cells would not be able to synthesize methionine

themselves with the loss of function of these enzymes. Therefore, acquiring methionine

from the environment would be the major source for methionine-dependent cancer cells.

Generally, there are two ways to reduce the serum methionine levels: nutritional

deprivation and enzymatic deprivation. The first approach is to provide a methionine-

free diet (Methionine-, Homocysteine-) to lower methionine absorbed in the body. The

second approach is to implement methionine gamma-lyase to degrade absorbed

methionine. By depleting methionine from the environment and therefore, methionine-

dependent cancer cells would not acquire enough methionine could not survive in low

concentration methionine condition.

Methionine-dependent cell activities

Methionine is involved in four major cell activities. First, methionine is one of 20

essential amino acids for all protein synthesis in cells. Without methionine, the synthesis

of all required proteins would be interrupted. Second, methionine is important for

production of polyamines (spermine and spermidine) which are associated with nuclear

and cell division activities (Thomas and Thomas 2001). Third, methionine is involved in

21

the transsulfuration pathway which generates glutathione. Glutathione is an important

antioxidant and can protect cell from oxidative stress (Anderson 1998). Fourth,

methionine provides the methyl group for methylation of DNA and other molecules.

Without methionine, all above cell activities would be interrupted and cancer cells would

be arrested at S-G2 phase of cell cycle (Pavillard et al., 2006) .

Pre-clinical trials and Clinical trials:

It has been reported that some pre-clinical trails have shown efficacy of

methionine deprivation on tumor cells (Breillout, Hadida et al. 1987; Guo, Lishko et al.

1993; Guo, Herrera et al. 1993; Poirson-Bichat, Gonfalone et al. 1997). Purified

methionine gamma-lyase from Pseudomonas putida was used to treat nude mice with

rodent or human tumors. There was no apparent toxicity of the treatment and the growth

of tumors were greatly slowed (Tan, Xu et al. 1996). Furthermore, recombinant

Psuedomonas putida methionine gamma-lase has been used on primates to test the

immunogenicity. However, it has been shown that re-challenge with recombinant

enzyme resulted in anaphylactic shock and death of one primate (Yang, Wang et al.

2004). This result indicated the antigenicity of bacterial enzymes would be a major

hurdle for practical treatment of patients. Recently, PEGylated methionine gamma-lyase

has been generated. PEGlyated therapeutic enzymes have been shown to have greater

resistance to degradation by proteases, longer half-life in vivo, lower clearance and less

immunogenicity (Veronese 2001; Harris and Chess 2003; Tan, Xu et al. 2010).

PEGylated methionine gamma-lyase is one option to reduce antigenicity in practical

treatment.

22

Besides pre-clinical trails, several clinical trails have been reported with positive

results. Methionine-depleting total parenteral nutrition (lacking methionine and cysteine)

displayed synergistic effects with 5-fluorouracil (5-FU) on patients with gastrointestinal

tract cancers (Goseki, Yamazaki et al. 1995). Cystemustine treatment combined one-day

methionine-free diet in patients with metastatic melanoma or recurrent glioma was well

tolerated in toxicity and nutritional status (phase II clinical trail) (Thivat, Farges et al.

2009). Moreover, purified endotoxin-free methionine gamma-lyase from Pseudomonas

putida has been used to test side effects on patients with advanced breast cancers. The

approach significantly lowered methionine concentration in serum (~0.1 uM) and did not

cause any signs of side effects (pilot phase I clinical trail) (Tan, Zavala et al. 1996; Tan,

Zavala et al. 1997).

Taken together, methionine-depletion is a promising approach to specifically

target methionine-dependent cancer cells. Besides methionine-depleting total parenteral

nutrition, an active enzyme with methionine-degrading activity is a powerful option for

this purpose. The hurdle is there is no human enzyme showing methionine-degrading

activity that would be non-immunogenic. Therefore the question is “How could we avoid

immunogenicity using a bacterial methionine gamma-lyase?” or “How do we engineer a

human enzyme to acquire methionine-degrading activity?” Interestingly, recent work has

shown an engineered human cystathionine gamma-lyase is able to degrade methionine

(Stone, Paley et al. 2012) and can be a candidate for further improvement. A further

engineering of human cystathionine gamma-lyase to degrade methionine is described in

Chapter 2.

23

BIOTECHNOLOGY TARGET: E.COLI BIOTIN PROTEIN LIGASE

Streptavidin:biotin

The interaction between biotin and streptavidin is the strongest non-covalent

binding in the nature with a Kd of ~10-15

M. Since the interaction of biotin:streptavidin

pair is so tight and specific, it has been widely applied for protein labeling or purification.

However, the single streptavidin:biotin pair is limited in application. Thus people keep

attempting to expand orthogonal pairs to streptavidin:biotin.

Applications of streptavidin:biotin orthogonal pair

Dr. Levy (an former member of The Ellington lab) implemented an IVC system to

evolve the streptavidin for altering specificity to biotin analog, desthiobiotin (Levy and

Ellington 2008). In his selection, he isolated the variants which show similar binding

affinity to desthiobiotin (~10-13

M) and have around 50 times slower off rate compared to

wild type streptavidin. He has demonstrated that two orthorgonal pairs, streptavidin

variant (SAR7-6):desthiobiotin and wild-type streptavidin:biotin could be used for

different fluorescence labeling on slides. Based on this idea, we questioned whether

there are any interesting biotin relevant targets. Thus, we thought E.coli biotin protein

ligase (BPL) might be an interesting target and it might have potential applications to

generate BPL variants in use of desthiobiotin as a new orthogonal pair. We implemented

in vitro compartmentalization (IVC) as a system to evolve E.coli biotin ligase in use of

desthiobiotin. Detailed work is described in Chapter 3.

24

Biotinylation reaction:

Biotin is an essential cofactor for biotin dependent enzymes to transfer a carboxyl

group in several metabolic pathways. In order to attach biotin to biotin dependent

enzymes such as acetyl-CoA carboxylase, biotin protein ligase is required for this

biotinylation reaction. In this biotinylation reaction, there are two steps. First, BPL

generates an intermediate, biotinoyl-5’-AMP by sequentially recruiting biotin and ATP to

their binding sites. Subsequently, BPL transfers the biotin moiety of biotinoyl-5’-AMP to

the specific lysine of biotin carboxyl carrier protein (BCCP) and then forms an amide

linkage between the carboxyl group of biotin and the ɛ-amino group of lysine. (Chapman-

Smith and Cronan 1999; Chapman-Smith and Cronan 1999)

Function of E.coli biotin protein ligase

E.coli BPL is a 35.5 kDa protein and its structure with bound biotin has been

solved (1HXD) (Weaver, Kwon et al. 2001). It is known that there are three domains, an

N-terminal DNA binding domain, a central catalytic domain, and a C-terminal domain

with unknown function. The N-terminal DNA binding domain has been shown to have a

characteristic helix-turn-helix structure which is associated with repression of biotin

synthesis. E.coli BPL is a bifunctional protein which is not only a biotin ligase but also a

transcription repressor of biotin biosynthesis. In the repression process, biotinoyl-5’-

AMP is a co-repressor which can trigger E.coli BPL to form dimer and bind the operator

to suppress the transcription of the biotin operon. The central catalytic domain contains

binding sites for biotin, ATP and the BCCP as these molecules are required substrates for

biotinylation reaction. Moreover, it has been shown that there are three different

conformation states of E.coli BPL in thermodynamic studies. The first conformation

25

change is caused by biotin binding and helps to recruit ATP binding. After generating

biotinoyl-5’-AMP, it goes through another structural alteration which allows BCCP

binding for biotinylation, or forming a dimer as a repressor (Chapman-Smith and Cronan

1999). Although the function of the C-terminal domain is not clear, it has been found

that some mutations within this domain impaired enzymatic activity. It is thought that

the C-terminal domain is required for catalytic activity and facilitates interaction of ATP

and the BCCP (Chapman-Smith, Mulhern et al. 2001). Regarding protein substrates for

E.coli BPL, it has been shown that there is a conserved peptide sequence (MKM) among

different identified protein substrates and the middle lysine is the specific residue for

formation of the amide bond with biotin. In addition, a 14 amino acid peptide, biotin

acceptor protein (BAP) has been isolated from a peptide library (Schatz 1993). This

small peptide tag has been widely used in protein labeling or purification.

Besides applying streptavidin:biotin orthogonal pair for labeling, Ting’s group has

shown that biotin ligases could be good enzymes to perform site-specific labeling on

proteins with biotin analogs. Labeling proteins with biotin analogs would help provide

an additional functional group to further expand the chemical reaction availability on

proteins (Slavoff, Chen et al. 2008). They assayed biotin ligases from several species and

found desthiobiotin azid and cis-propargyl biotin could be labeled on a short peptide by

P. horikoshii BPL and yeast BPL respectively. In addition, they evolved a new peptide

which could only be specifically recognized by yeast BPL and form a new orthogonal

pair, yeast BAP:yeast BPL. (Chen, Choi et al. 2007). These two orthogonal pairs, yeast

BAP:yeast BPL and E.coli BAP:E.coli BPL orthogonal pairs have been shown to label

different fluorescences on surface. Furthermore, they evolved E.coli lipoic acid ligase (a

homolog of E.coli biotin protein ligase) which has similar ligation reaction to BPL.

26

Engineered E.coli lipoic acid ligase is capable of recognizing 7-hydroxycoumarin and can

be used as a site-specific fluorophore ligase (Uttamapinant, White et al. 2010).

27

REFERENCES

Agresti, J.J., Kelly, B.T., Jaschke, A., and Griffiths, A.D. (2005). Selection of ribozymes

that catalyse multiple-turnover Diels-Alder cycloadditions by using in vitro

compartmentalization. Proc Natl Acad Sci U S A 102, 16170-16175.

Aharoni, A., Amitai, G., Bernath, K., Magdassi, S., and Tawfik, D.S. (2005). High-

throughput screening of enzyme libraries: thiolactonases evolved by fluorescence-

activated sorting of single cells in emulsion compartments. Chem Biol 12, 1281-

1289.

Amitai, G., Gupta, R.D., and Tawfik, D.S. (2007). Latent evolutionary potentials under

the neutral mutational drift of an enzyme. HFSP J 1, 67-78.

Anderson, M.E. (1998). Glutathione: an overview of biosynthesis and modulation. Chem

Biol Interact 111-112, 1-14.

Bernath, K., Hai, M., Mastrobattista, E., Griffiths, A.D., Magdassi, S., and Tawfik, D.S.

(2004). In vitro compartmentalization by double emulsions: sorting and gene

enrichment by fluorescence activated cell sorting. Anal Biochem 325, 151-157.

Bertschinger, J., and Neri, D. (2004). Covalent DNA display as a novel tool for directed

evolution of proteins in vitro. Protein Eng Des Sel 17, 699-707.

Bloom, J.D., and Arnold, F.H. (2009). In the light of directed evolution: pathways of

adaptive protein evolution. Proc Natl Acad Sci U S A 106 Suppl 1, 9995-10000.

Bochkov, Y.A., and Palmenberg, A.C. (2006). Translational efficiency of EMCV IRES

in bicistronic vectors is dependent upon IRES sequence and gene location.

Biotechniques 41, 283-284, 286, 288 passim.

Bowles, T.L., Kim, R., Galante, J., Parsons, C.M., Virudachalam, S., Kung, H.J., and

Bold, R.J. (2008). Pancreatic cancer cell lines deficient in argininosuccinate

synthetase are sensitive to arginine deprivation by arginine deiminase. Int J

Cancer 123, 1950-1955.

Breillout, F., Antoine, E., and Poupon, M.F. (1990). Methionine dependency of

malignant tumors: a possible approach for therapy. J Natl Cancer Inst 82, 1628-

1632.

Breillout, F., Hadida, F., Echinard-Garin, P., Lascaux, V., and Poupon, M.F. (1987).

Decreased rat rhabdomyosarcoma pulmonary metastases in response to a low

methionine diet. Anticancer Res 7, 861-867.

Cantor, J.R., Panayiotou, V., Agnello, G., Georgiou, G., and Stone, E.M. (2012).

Engineering reduced-immunogenicity enzymes for amino acid depletion therapy

in cancer. Methods Enzymol 502, 291-319.

28

Cellarier, E., Durando, X., Vasson, M.P., Farges, M.C., Demiden, A., Maurizis, J.C.,

Madelmont, J.C., and Chollet, P. (2003). Methionine dependency and cancer

treatment. Cancer Treat Rev 29, 489-499.

Chapman-Smith, A., and Cronan, J.E., Jr. (1999a). The enzymatic biotinylation of

proteins: a post-translational modification of exceptional specificity. Trends

Biochem Sci 24, 359-363.

Chapman-Smith, A., and Cronan, J.E., Jr. (1999b). Molecular biology of biotin

attachment to proteins. J Nutr 129, 477S-484S.

Chapman-Smith, A., Mulhern, T.D., Whelan, F., Cronan, J.E., Jr., and Wallace, J.C.

(2001). The C-terminal domain of biotin protein ligase from E. coli is required for

catalytic activity. Protein Sci 10, 2608-2617.

Chen, I., Choi, Y.A., and Ting, A.Y. (2007). Phage display evolution of a peptide

substrate for yeast biotin ligase and application to two-color quantum dot labeling

of cell surface proteins. J Am Chem Soc 129, 6619-6625.

Chen, Y., Mandic, J., and Varani, G. (2008). Cell-free selection of RNA-binding proteins

using in vitro compartmentalization. Nucleic Acids Res 36, e128.

Cohen, H.M., Tawfik, D.S., and Griffiths, A.D. (2004). Altering the sequence specificity

of HaeIII methyltransferase by directed evolution using in vitro

compartmentalization. Protein Eng Des Sel 17, 3-11.

Cooney, D.A., and Handschumacher, R.E. (1970). L-asparaginase and L-asparagine

metabolism. Annu Rev Pharmacol 10, 421-440.

Davidson, E.A., Dlugosz, P.J., Levy, M., and Ellington, A.D. (2009). Directed evolution

of proteins in vitro using compartmentalization in emulsions. Curr Protoc Mol

Biol Chapter 24, Unit 24 26.

Doi, N., Kumadaki, S., Oishi, Y., Matsumura, N., and Yanagawa, H. (2004). In vitro

selection of restriction endonucleases by in vitro compartmentalization. Nucleic

Acids Res 32, e95.

Doi, N., and Yanagawa, H. (1999). STABLE: protein-DNA fusion system for screening

of combinatorial protein libraries in vitro. FEBS Lett 457, 227-230.

Ensor, C.M., Holtsberg, F.W., Bomalaski, J.S., and Clark, M.A. (2002). Pegylated

arginine deiminase (ADI-SS PEG20,000 mw) inhibits human melanomas and

hepatocellular carcinomas in vitro and in vivo. Cancer Res 62, 5443-5450.

Fen, C.X., Coomber, D.W., Lane, D.P., and Ghadessy, F.J. (2007). Directed evolution of

p53 variants with altered DNA-binding specificities by in vitro

compartmentalization. J Mol Biol 371, 1238-1248.

29

Ghadessy, F.J., and Holliger, P. (2004). A novel emulsion mixture for in vitro

compartmentalization of transcription and translation in the rabbit reticulocyte

system. Protein Eng Des Sel 17, 201-204.

Ghadessy, F.J., Ong, J.L., and Holliger, P. (2001). Directed evolution of polymerase

function by compartmentalized self-replication. Proc Natl Acad Sci U S A 98,

4552-4557.

Ghadessy, F.J., Ramsay, N., Boudsocq, F., Loakes, D., Brown, A., Iwai, S., Vaisman, A.,

Woodgate, R., and Holliger, P. (2004). Generic expansion of the substrate

spectrum of a DNA polymerase by directed evolution. Nat Biotechnol 22, 755-

759.

Gong, H., Zolzer, F., von Recklinghausen, G., Havers, W., and Schweigerer, L. (2000).

Arginine deiminase inhibits proliferation of human leukemia cells more potently

than asparaginase by inducing cell cycle arrest and apoptosis. Leukemia 14, 826-

829.

Goseki, N., Yamazaki, S., Shimojyu, K., Kando, F., Maruyama, M., Endo, M., Koike,

M., and Takahashi, H. (1995). Synergistic effect of methionine-depleting total

parenteral nutrition with 5-fluorouracil on human gastric cancer: a randomized,

prospective clinical trial. Jpn J Cancer Res 86, 484-489.

Griffiths, A.D., and Tawfik, D.S. (2000). Man-made enzymes--from design to in vitro

compartmentalisation. Curr Opin Biotechnol 11, 338-353.

Griffiths, A.D., and Tawfik, D.S. (2003). Directed evolution of an extremely fast

phosphotriesterase by in vitro compartmentalization. EMBO J 22, 24-35.

Guo, H., Lishko, V.K., Herrera, H., Groce, A., Kubota, T., and Hoffman, R.M. (1993a).

Therapeutic tumor-specific cell cycle block induced by methionine starvation in

vivo. Cancer Res 53, 5676-5679.

Guo, H.Y., Herrera, H., Groce, A., and Hoffman, R.M. (1993b). Expression of the

biochemical defect of methionine dependence in fresh patient tumors in primary

histoculture. Cancer Res 53, 2479-2483.

Gupta, R.D., Goldsmith, M., Ashani, Y., Simo, Y., Mullokandov, G., Bar, H., Ben-David,

M., Leader, H., Margalit, R., Silman, I., et al. (2011). Directed evolution of

hydrolases for prevention of G-type nerve agent intoxication. Nat Chem Biol 7,

120-125.

Gupta, R.D., and Tawfik, D.S. (2008). Directed enzyme evolution via small and effective

neutral drift libraries. Nat Methods 5, 939-942.

Hanes, J., and Pluckthun, A. (1997). In vitro selection and evolution of functional

proteins by using ribosome display. Proc Natl Acad Sci U S A 94, 4937-4942.

Harris, J.M., and Chess, R.B. (2003). Effect of pegylation on pharmaceuticals. Nat Rev

Drug Discov 2, 214-221.

30

Hernandez, C.P., Morrow, K., Lopez-Barcons, L.A., Zabaleta, J., Sierra, R., Velasco, C.,

Cole, J., and Rodriguez, P.C. (2010). Pegylated arginase I: a potential therapeutic

approach in T-ALL. Blood 115, 5214-5221.

Hoffman, R.M. (1985). Altered methionine metabolism and transmethylation in cancer.

Anticancer Res 5, 1-30.

Keefe, A.D., and Szostak, J.W. (2001). Functional proteins from a random-sequence

library. Nature 410, 715-718.

Kelly, B.T., and Griffiths, A.D. (2007). Selective gene amplification. Protein Eng Des Sel

20, 577-581.

Kintses, B., van Vliet, L.D., Devenish, S.R., and Hollfelder, F. (2010). Microfluidic

droplets: new integrated workflows for biological experiments. Curr Opin Chem

Biol 14, 548-555.

Kitagawa, M., Ara, T., Arifuzzaman, M., Ioka-Nakamichi, T., Inamoto, E., Toyonaga, H.,

and Mori, H. (2005). Complete set of ORF clones of Escherichia coli ASKA

library (a complete set of E. coli K-12 ORF archive): unique resources for

biological research. DNA Res 12, 291-299.

Kokkinakis, D.M., Hoffman, R.M., Frenkel, E.P., Wick, J.B., Han, Q., Xu, M., Tan, Y.,

and Schold, S.C. (2001). Synergy between methionine stress and chemotherapy in

the treatment of brain tumor xenografts in athymic mice. Cancer Res 61, 4017-

4023.

Kroemer, G., and Pouyssegur, J. (2008). Tumor cell metabolism: cancer's Achilles' heel.

Cancer Cell 13, 472-482.

Levy, M., and Ellington, A.D. (2008). Directed evolution of streptavidin variants using in

vitro compartmentalization. Chem Biol 15, 979-989.

Levy, M., Griswold, K.E., and Ellington, A.D. (2005). Direct selection of trans-acting

ligase ribozymes by in vitro compartmentalization. RNA 11, 1555-1562.

Lipovsek, D., and Pluckthun, A. (2004). In-vitro protein evolution by ribosome display

and mRNA display. J Immunol Methods 290, 51-67.

Loakes, D., Gallego, J., Pinheiro, V.B., Kool, E.T., and Holliger, P. (2009). Evolving a

polymerase for hydrophobic base analogues. J Am Chem Soc 131, 14827-14837.

Lorenzi, P.L., Llamas, J., Gunsior, M., Ozbun, L., Reinhold, W.C., Varma, S., Ji, H.,

Kim, H., Hutchinson, A.A., Kohn, E.C., et al. (2008). Asparagine synthetase is a

predictive biomarker of L-asparaginase activity in ovarian cancer cell lines. Mol

Cancer Ther 7, 3123-3128.

Margulies, M., Egholm, M., Altman, W.E., Attiya, S., Bader, J.S., Bemben, L.A., Berka,

J., Braverman, M.S., Chen, Y.J., Chen, Z., et al. (2005). Genome sequencing in

microfabricated high-density picolitre reactors. Nature 437, 376-380.

31

Mastrobattista, E., Taly, V., Chanudet, E., Treacy, P., Kelly, B.T., and Griffiths, A.D.

(2005). High-throughput screening of enzyme libraries: in vitro evolution of a

beta-galactosidase by fluorescence-activated sorting of double emulsions. Chem

Biol 12, 1291-1300.

Miller, O.J., Bernath, K., Agresti, J.J., Amitai, G., Kelly, B.T., Mastrobattista, E., Taly,

V., Magdassi, S., Tawfik, D.S., and Griffiths, A.D. (2006). Directed evolution by

in vitro compartmentalization. Nat Methods 3, 561-570.

Odegrip, R., Coomber, D., Eldridge, B., Hederer, R., Kuhlman, P.A., Ullman, C.,

FitzGerald, K., and McGregor, D. (2004). CIS display: In vitro selection of

peptides from libraries of protein-DNA complexes. Proc Natl Acad Sci U S A

101, 2806-2810.

Ong, J.L., Loakes, D., Jaroslawski, S., Too, K., and Holliger, P. (2006). Directed

evolution of DNA polymerase, RNA polymerase and reverse transcriptase activity

in a single polypeptide. J Mol Biol 361, 537-550.

Parent, K.N., Ranaghan, M.J., and Teschke, C.M. (2004). A second-site suppressor of a

folding defect functions via interactions with a chaperone network to improve

folding and assembly in vivo. Mol Microbiol 54, 1036-1050.

Pavillard, V., Nicolaou, A., Double, J.A., and Phillips, R.M. (2006). Methionine

dependence of tumours: a biochemical strategy for optimizing paclitaxel

chemosensitivity in vitro. Biochem Pharmacol 71, 772-778.

Poirson-Bichat, F., Goncalves, R.A., Miccoli, L., Dutrillaux, B., and Poupon, M.F.

(2000). Methionine depletion enhances the antitumoral efficacy of cytotoxic

agents in drug-resistant human tumor xenografts. Clin Cancer Res 6, 643-653.

Poirson-Bichat, F., Gonfalone, G., Bras-Goncalves, R.A., Dutrillaux, B., and Poupon,

M.F. (1997). Growth of methionine-dependent human prostate cancer (PC-3) is

inhibited by ethionine combined with methionine starvation. Br J Cancer 75,

1605-1612.

Ramsay, N., Jemth, A.S., Brown, A., Crampton, N., Dear, P., and Holliger, P. (2010).

CyDNA: synthesis and replication of highly Cy-dye substituted DNA by an

evolved polymerase. J Am Chem Soc 132, 5096-5104.

Reiersen, H., Lobersli, I., Loset, G.A., Hvattum, E., Simonsen, B., Stacy, J.E., McGregor,

D., Fitzgerald, K., Welschof, M., Brekke, O.H., et al. (2005). Covalent antibody

display--an in vitro antibody-DNA library selection system. Nucleic Acids Res

33, e10.

Roberts, R.W., and Szostak, J.W. (1997). RNA-peptide fusions for the in vitro selection

of peptides and proteins. Proc Natl Acad Sci U S A 94, 12297-12302.

Romero, P.A., and Arnold, F.H. (2009). Exploring protein fitness landscapes by directed

evolution. Nat Rev Mol Cell Biol 10, 866-876.

32

Schatz, P.J. (1993). Use of peptide libraries to map the substrate specificity of a peptide-

modifying enzyme: a 13 residue consensus peptide specifies biotinylation in

Escherichia coli. Biotechnology (N Y) 11, 1138-1143.

Sepp, A., and Choo, Y. (2005). Cell-free selection of zinc finger DNA-binding proteins

using in vitro compartmentalization. J Mol Biol 354, 212-219.

Shimizu, Y., Kanamori, T., and Ueda, T. (2005). Protein synthesis by pure translation

systems. Methods 36, 299-304.

Sitaraman, K., Esposito, D., Klarmann, G., Le Grice, S.F., Hartley, J.L., and Chatterjee,

D.K. (2004). A novel cell-free protein synthesis system. J Biotechnol 110, 257-

263.

Slavoff, S.A., Chen, I., Choi, Y.A., and Ting, A.Y. (2008). Expanding the substrate

tolerance of biotin ligase through exploration of enzymes from diverse species. J

Am Chem Soc 130, 1160-1162.

Stapleton, J.A., and Swartz, J.R. (2010). Development of an in vitro

compartmentalization screen for high-throughput directed evolution of [FeFe]

hydrogenases. PLoS One 5, e15275.

Stein, V., Sielaff, I., Johnsson, K., and Hollfelder, F. (2007). A covalent chemical

genotype-phenotype linkage for in vitro protein evolution. Chembiochem 8, 2191-

2194.

Stone, E., Paley, O., Hu, J., Ekerdt, B., Cheung, N.K., and Georgiou, G. (2012). De novo

engineering of a human cystathionine-gamma-lyase for systemic (L)-Methionine

depletion cancer therapy. ACS Chem Biol 7, 1822-1829.

Takei, Y., Kadomatsu, K., Itoh, H., Sato, W., Nakazawa, K., Kubota, S., and Muramatsu,

T. (2002). 5'-,3'-inverted thymidine-modified antisense oligodeoxynucleotide

targeting midkine. Its design and application for cancer therapy. J Biol Chem 277,

23800-23806.

Tan, Y., Xu, M., Guo, H., Sun, X., Kubota, T., and Hoffman, R.M. (1996a). Anticancer

efficacy of methioninase in vivo. Anticancer Res 16, 3931-3936.

Tan, Y., Xu, M., and Hoffman, R.M. (2010). Broad selective efficacy of recombinant

methioninase and polyethylene glycol-modified recombinant methioninase on

cancer cells In Vitro. Anticancer Res 30, 1041-1046.

Tan, Y., Zavala, J., Sr., Han, Q., Xu, M., Sun, X., Tan, X., Magana, R., Geller, J., and

Hoffman, R.M. (1997). Recombinant methioninase infusion reduces the

biochemical endpoint of serum methionine with minimal toxicity in high-stage

cancer patients. Anticancer Res 17, 3857-3860.

Tan, Y., Zavala, J., Sr., Xu, M., Zavala, J., Jr., and Hoffman, R.M. (1996b). Serum

methionine depletion without side effects by methioninase in metastatic breast

cancer patients. Anticancer Res 16, 3937-3942.

33

Tawfik, D.S., and Griffiths, A.D. (1998). Man-made cell-like compartments for

molecular evolution. Nat Biotechnol 16, 652-656.

Thivat, E., Farges, M.C., Bacin, F., D'Incan, M., Mouret-Reynier, M.A., Cellarier, E.,

Madelmont, J.C., Vasson, M.P., Chollet, P., and Durando, X. (2009). Phase II trial

of the association of a methionine-free diet with cystemustine therapy in

melanoma and glioma. Anticancer Res 29, 5235-5240.

Thomas, T., and Thomas, T.J. (2001). Polyamines in cell growth and cell death:

molecular mechanisms and therapeutic applications. Cell Mol Life Sci 58, 244-

258.

Tokuriki, N., and Tawfik, D.S. (2009). Chaperonin overexpression promotes genetic

variation and enzyme evolution. Nature 459, 668-673.

Ullman, C.G., Frigotto, L., and Cooley, R.N. (2011). In vitro methods for peptide display

and their applications. Brief Funct Genomics 10, 125-134.

Utada, A.S., Lorenceau, E., Link, D.R., Kaplan, P.D., Stone, H.A., and Weitz, D.A.

(2005). Monodisperse double emulsions generated from a microcapillary device.

Science 308, 537-541.

Uttamapinant, C., White, K.A., Baruah, H., Thompson, S., Fernandez-Suarez, M.,

Puthenveetil, S., and Ting, A.Y. (2010). A fluorophore ligase for site-specific

protein labeling inside living cells. Proc Natl Acad Sci U S A 107, 10914-10919.

Veronese, F.M. (2001). Peptide and protein PEGylation: a review of problems and

solutions. Biomaterials 22, 405-417.

Weaver, L.H., Kwon, K., Beckett, D., and Matthews, B.W. (2001). Corepressor-induced

organization and assembly of the biotin repressor: a model for allosteric activation

of a transcriptional regulator. Proc Natl Acad Sci U S A 98, 6045-6050.

Williams, R., Peisajovich, S.G., Miller, O.J., Magdassi, S., Tawfik, D.S., and Griffiths,

A.D. (2006). Amplification of complex gene libraries by emulsion PCR. Nat

Methods 3, 545-550.

Yang, Z., Wang, J., Yoshioka, T., Li, B., Lu, Q., Li, S., Sun, X., Tan, Y., Yagi, S.,

Frenkel, E.P., et al. (2004). Pharmacokinetics, methionine depletion, and

antigenicity of recombinant methioninase in primates. Clin Cancer Res 10, 2131-

2138.

Yonezawa, M., Doi, N., Kawahashi, Y., Higashinakagawa, T., and Yanagawa, H. (2003).

DNA display for in vitro selection of diverse peptide libraries. Nucleic Acids Res

31, e118.

Zahnd, C., Amstutz, P., and Pluckthun, A. (2007). Ribosome display: selecting and

evolving proteins in vitro that specifically bind to a target. Nat Methods 4, 269-

279.

34

Chapter 2: Engineering human cystathionine gamma-lyase to degrade

methionine for cancer treatment

INTRODUCTION

Methionine gamma-lyase

Methionine gamma-lyase (EC 4.4.1.11, also called methioninase or MGL) is a

pyridoxal 5’-phosphate (PLP) dependent enzyme and belongs to the gamma-lyase family

which can break a carbon-sulfur bond at the gamma carbon. The substrate for this

enzyme is L-methionine which is degraded through alpha, gamma-elimination to produce

alpha-ketobutyrate, methane thiol and NH3 as products (Figure 2-1). In nature, it is found

in bacteria, plants, fungi and protozoa but not in mammals. It is known that bacteria and

protozoa use MGL to degrade methionine for ATP generation (Inoue, Inagaki et al. 1997;

Ragsdale 2003; Sato and Nozaki 2009). Importantly, MGL has shown promise for use as

an anti-cancer drug. Depletion of methionine in mice with xenografted tumors inhibited

tumor growth (Tan, Sun et al. 1999; Poirson-Bichat, Goncalves et al. 2000; Kokkinakis,

Hoffman et al. 2001) and several clinical trials reported that a methionine-free diet shown

an improved outcome of cancer treatment (Goseki, Yamazaki et al. 1995; Durando,

Thivat et al. 2008; Thivat, Farges et al. 2009). However, bacterial MGL would be

expected to induce strong immune responses in mammals. In fact, it has been shown to

cause anaphylactic shock and death in primates when bacterial MGL was re-challenged

(Yang, Wang et al. 2004). Therefore, it would be better to use a humanized MGL or a

human enzyme which is able to break down methionine for methionine depletion in

cancer treatment.

35

Figure 2-1: Enzyme reaction of methionine gamma-lyase. Methionine gamma-lyase

is a PLP-dependent enzyme and breaks down L-methionine to methanethiol,

NH3 and 2-oxobutanoate (alpha ketobutyrate).

Cystathionine gamma-lyase

Based on this idea, a paralog of MGL, human cystathionine gamma-lyase

(EC 4.4.1.1, also called cystathionase or hCGL) was chosen by Dr. Stone and Olga Paley

(Georgiou Lab, UT Austin) for engineering to acquire methionine-breakdown activity

(Stone, Paley et al. 2012). hCGL shares around 60% amino acid similarity with

Pseudomonas putida MGL and also belongs to the gamma-lyase family. hCGL is PLP-

dependent enzyme that can break the carbon-sulfur bond of cystathionine to produce

cysteine, NH3 and alpha-ketobutyrate (Figure 2-2). The catalytic mechanism of MGL

and hCGL is the same. Moreover, the substrate, cystathionine is similar and larger than

methionine. Thus hCGL was chosen to be engineered in use of methionine as substrates.

The physiological function of hCGL is involved in a transsulfuration pathway which

converts methionine to cysteine (Rao, Drake et al. 1990; Cavuoto and Fenech 2012) and

aside from diet, it is the only way for humans to synthesize cysteine.

36

Figure 2-2: Enzyme reaction of cystathionine gamma-lyase. Cystathionine gamma-

lyase is a PLP-dependent enzyme and break L-cystathionine to cysteine,

NH3 and 2-oxobutanoate by alpha-gamma elimination.

Dr. Stone and Olga Paley’s work (Stone, Paley et al. 2012):

In order to engineer hCGL to break down methionine, Everett Stone and Olga

Paley did a structural comparison of hCGL (PDB: 3COG) (Sun, Collins et al. 2009) and

Trichomonas vaginalis MGL (tMGL, PDB: 1E5E) and found the side chains of Glu-59,

Arg-119 and Glu-339 are important for the cysteine portion of cystathionione (Figure 2-

3). However, the corresponding positions of tMGL are more hydrophobic amino acids,

Ile-55, Ala-116 and Val-337 (Stone, Paley et al. 2012). Based on this observation, they

hypothesized these positions are important for substrate specificity. Thus, they

performed a combinatorial pairwise saturation mutagenesis on Glu-59, Arg-119 and Glu-

339 of hCGL. First, they carried out saturation mutagenesis on Arg-119 and Glu-339 and

obtained a variant with R119L and E339V showing activity in breakdown of methionine.

And then, they used this variant as a template and further optimized the positions of 59

and 119 (fix the E339V) by another round of saturation mutagenesis. Finally, they

obtained a variant, hCGL-NLV (E59N, R119L and E339V) which displayed around 1%

activity in the breakdown of methionine compared to Pseudomonas putida MGL

(pMGL). The kcat and KM of hCGL-NLV are 7.9 s-1

(kcat of pMGL: 20 s-1

) and 14 mM

(KM of pMGL: 0.34 mM) respectively (Table 2-1). Moreover, the engineered hCGL-

37

NLV is more stable than pMGL in pooled human serum at 37oC. The half-lives of hCG-

NLV and pMGL in pooled human serum are t1/2=78 and t1/2=1.9 hours respectively.

Additionally, the PEGylated hCGL-NLV displayed longer t1/2 (30 hours) than

unPEGylated hCGL-NLV (t1/2=1.8 hours) in the blood of injected mice. Importantly,

mice with neuroblastoma xenografts displayed nearly complete cessation of tumor

growth in PEGylated hCGL-NLV treatment (Figure 2-4) (Stone, Paley et al. 2012). In

order to enhance the efficacy for depletion of methionine in practical cancer therapy, a

more active hCGL variant in break of methionine will be needed in the future.

Figure 2-3: Structure comparison of human cystathionine gamma-lyase (PDB:

3COG) and Trichomoas vaginalis (PDB: 1E5E) with inhibitor

propargylglycine (PAG) (Stone, Paley et al. 2012).

38

Table 2-1: Kinetic analysis of Pseudomonas putida methionine gamma-lyase,

human cystathionine gamma-lyase and engineered human cystathionine

gamma-lyase (hCGL-NLV) (Stone, Paley et al. 2012).

39

Figure 2-4: Evaluation of PEGylated hCGL-NLV in athymic mice bearing LAN-1

xenografts (Stone, Paley et al. 2012). (�) control group which maintained

normal diet (N=10). (�) animals maintained on methionine(-)

homecysteine(-)choline(-) diet (N=10). (�) animals treated with 100 U

PEG-hCGL-NLV in combination with methionine(-) homecysteine(-)

choline(-) mouse feed (N=10). Treatment days are designated by (▲).

Tumor growth rate was expressed as mean SEM (standard error of the

mean) for each group. *, p < 0.01 for the PEG-hCGL-NLV treated group

relative to the two untreated groups.

40

RESULTS

Phylogenetic analysis of MGLs and CGLs:

Besides random mutagenesis, structural comparison, and DNA shuffling,

phylogenetic analysis is an established method of enzyme engineering. The idea is that

enzymes in the same family derive from the same ancestral protein and share similar

protein scaffolds. Evolving to different substrate specificities, these enzymes display

unique conservations at specific amino acid positions and can be further categorized into

different sub-families. For example, both methionine gamma-lyases (MGLs) and

cystathionine gamma-lyases (CGLs) are in the same gamma-lyase family and can be

further categorized into two sub-families showing unique conservations.

MGLs and CGLs are paralogs and both belong to the gamma-lyase family of

proteins. They use the same catalytic mechanism to break the carbon-sulfur bond on the

gamma-carbon of their substrates. The idea of phylogenetic analysis is that “the

differences of conservation in sequence alignment of paralogs may result in the

differences of substrate specificity” (Cole and Gaucher 2011). Theoretically, there are

four situations in the phylogenetic analysis of MGLs and CGLs (Figure 2-5, hypothetical

situations). For Situation 1, the amino acid (S: serine, no difference between CGLs and

MGLs) should be responsible for the properties of gamma-lyase and may be very

important to the catalytic mechanism of gamma lyase. For Situation 2 (CGLs: E, MGLs:

V), 3 (CGLs: S) and 4 (MGLs: V), the conserved amino acids should associate with the

individual sub-family properties. For example, in Situation 3, S (serine) may be

important for substrate specificity of CGLs and V (valine) may be important to the

substrate specificity of MGLs. Substituting S (CGLs: S) with V (MGLs: V), we may be

able to shift the substrate specificity from CGL to MGL (CGLs and MGLs). Since our

goal is to engineer hCGL to acquire the activity of MGLs, we should keep the amino acid

41

(S) as in Situation 1. In addition, the S (CGLs: S) should be substituted with V (MGLs:

V) in the situation 2. In Situation 3, we should use any amino acids of MGLs (A, C, D,

E, T, M, P or Y) to replace S of CGLs and use V to replace amino acid of the CGLs (D,

E, T, M or A) in the Situation 4.

Figure 2-5: Phylogenetic analysis of cystathionine gamma-lyases and methionine

gamma-lyases. (Left) Theoretically, paralogs are from the same precursor

protein. In evolution of different paralogs, they may share the same scaffold

and have different conservations for substrate specificity. (Right) There are

five possible situations for comparison of paralogs.

Pilot experiment (test 11 selected positions from phylogenetic analysis)

The goal is to engineer hCGL to break down methionine. However there is no

strong evidence that phylogenetic analysis will work in this project. Therefore, we

performed a pilot experiment to test the idea.

42

From the solved structure (PDB: 2NMP), hCGL is a tetramer protein with the

active site composed of the C-terminal domain of one monomer and the partial N-

terminal domain of an adjacent monomer. Since the active site is important for substrate

specificity thus we focused on the phylogenetic analysis of the C-terminal domain.

Several eukaryotic cystathionine gamma-lyases (eCGLs: human, cattle, rat, mouse,

zebra) were chosen to compare with several bacterial methionine gamma-lyases (bMGLs:

Ruegeria, Pseudomonas, Ferrimonas, Porphyromonas, Shewanella and Bacillus).

Geneious (sequence analysis software) was used for sequence alignment (Figure 2-6).

K268R, F270C, G273A, C307M, T311G, V314I, G343D, F344A, L350H, I 353S and

V359Y were chosen for site-directed mutagenesis on wild-type hCGL and hCGL-NLV

backbone. These chosen substitutions matched with Situation 2 (Figure 2-5).

43

Figure 2-6: Phylogenetic analysis of the carboxyl terminals of 5 eukaryotic

cystathionine gamma-lyases and 6 bacterial methionine gamma-lyases.

The 11 red arrows indicated conservations chosen as candidates for shifting

substrate specificity.

44

Initially, 11 hCGL mutants (11 aforementioned substitutions) on the wild-type

hCGL backbone were individually screened on a 96-well plate (Figure 2-8, Material and

Method). Unfortunately, none of the 11 mutants showed better activity compared to

hCGL-NLV. However, 2 of the 11 hCGL mutants (11 aforementioned substitutions) on

the hCGL-NLV backbone showed around 2-fold better activity in the breakdown of

methionine compared to hCGL-NLV. These 11 mutants (on hCGL-NLV backbone) were

further purified by Ni-NTA chromatography and characterized to determine the kcat and

KM of methionine breakdown. Interestingly, hCGL-NLV+G and hCGL-NLV+S (T311G,

I353S on hCGL-NLV) both showed around 2-fold better kcat/KM (1.1 and 0.96 s-1

mM-1

)

compared hCGL-NLV (0.56 s-1 mM-1) (Table 2-2). The kcat of hCGL-NLV+G, hCGL-

NLV+S, and hCGL-NLV were 3.1, 4.4, and 7.9 s-1

respectively and KM of them were 2.8,

4.6, and 14 mM respectively (Figure 2-7). Apparently, the KM of hCGL-NLV+G and

hCGL-NLV+S were much improved (3 to 5-fold better) in the breakdown of methionine

compared to hCGL-NLV. Mapping T311 and I353 on the solved structure of wild-type

hCGL (PDB: 2NMP), only I353 is around the active site (similar to N59, L119 and L339)

and T311 is further away from the active site. This indicates that I353S may cooperate

with N59, L119, and L339 to improve the KM in the breakdown of methionine. It is

difficult to explain how T311G affects the KM of the hCGL-NLV+G without solved

structure. hCGL-NLV+R, hCGL-NLV+C and hCGL-NLV+I (K268R, F270C or V314I

on hCGL-NLV backbone) showed similar kcat/KM in the breakdown of methionine

compared to hCGL-NLV.

Furthermore, T311G and I353S were combined on hCGL-NLV backbone to test

for cooperativity. Kinetic assays showed hCGL-NLV+GS (T311G and I353S on hCGL-

NLV) displayed 2.2-fold better kcat/KM value compared to hCGL-NLV.

45

Table 2-2: List of kinetic relative values for 22 hCGL variants in methionine

breakdown. The aforementioned 11 positions were substituted on wild-

type hCGL or hCGL-NLV. Each variant was tested for its ability to break

down methionine compared to hCGL-NLV.

46

0

0.5

1

1.5

2

2.5

3

3.5

4

-5 0 5 10 15 20 25

Activity assay

hCGL-NLV+RhCGL-NLV+GhCGL-NLV+S

1/s

[Methionine] mM

y = (M0*m1)/(M0 + m2)

ErrorValue

0.135473.8756kcat

0.592756.7953Km

NA0.99453R

y = (M0*m1)/(M0 + m2)

ErrorValue

0.0327423.0793kcat

0.100622.7715Km

NA0.9989R

y = (M0*m1)/(M0 + m2)

ErrorValue

0.0733394.3739kcat

0.221144.6378Km

NA0.99825R

Figure 2-7: Kinetic analysis of engineered hCGL variants in methionine

breakdown. Three engineered hCGL variants were purified to determine

the kcat and KM in breakdown of methionine. Interestingly, hCGL-NLV+G

and hCGL-NLV+S displayed two-fold better kcat/KM value compared to

hCGL-NLV.

47

Combinatorial library (11 positions from phylogenetic analysis)

Based on the pilot experiment, it is very possible some of these chosen candidates

(11 aforementioned substitutions) would benefit each other and need to be present at the

same time for cooperation. Therefore a small combinatorial library was generated based

on the 11 aforementioned candidates. Since we did not have a high-throughput screening

method we were not able to screen a library generated from saturation mutagenesis at all

11 positions (2011

variants). Thus, we designed each position to have two amino acid

options (211

=2048 variants) in the combinatorial library to probe either an amino acid

from wild-type hCGL or an amino acid from the phylogenetic analysis of CGLs and

MGLs. We expected the selected amino acids from the phylogenetic analysis of MGLs

and CGLs to be more potent than other amino acids.

This library, “11p-NLV-I” (mutagenesis at 11 positions, on hCGL-NLV

backbone, library I) was constructed by overlapping PCR and verified by sequencing 20

clones. Only 45% (9 clones) contained full-length coding sequences with the desired

substitutions. Others were truncated (premature stop codons) or variants with insertions

or deletions. Theoretically, there are 2048 variants in this library and we screened forty

96-well plates which contained approximately 4000 clones to attempt full coverage.

(Figure 2-8 and 2-9).

Initially, 34 clones displayed higher activity compared to hCGL-NLV in the

screening. The top 10 of these active clones were chosen for further protein purification

and characterization. However, only one variant, P2A11 (plate 2- row A- column 11)

displayed 2.5-fold higher kcat/KM compared to hCGL-NLV in the breakdown of

methionine. The kcat and KM value of P2A11 are 4.5 s-1

and 3.5 mM respectively (Figure

2-10). Sequencing P2A11 showed it contained K268R, T311G and I353S on hCGL-

NLV backbone. Mapping K268 to the solved structure of hCGL (PDB: 2NMP) it is

48

away from the active site and found near the interface of monomers (hCGL is a tetramer)

(Sun, Collins et al. 2009). Perhaps this mutation allows the monomers to juxtapose in a

way that facilitate a more compliant active site. Still, it is difficult to exactly know how it

affects methionine-breakdown activity without a solved structure.

Figure 2-8: 96-well plate screening process. First, a combinatorial library was

transformed to BL21(DE3) competent cells. Individual colonies were used

to inoculate 96-well plates for small scale culture. Second, cells were then

induced for protein expression by IPTG for 2 hours. Induced cells were

lysed by mild detergent and centrifuged to separate soluble protein from cell

debris. The soluble fraction was incubated with methionine for 10 hours.

Finally, the reaction was mixed with MBTH to determine methionine-

degraded products by Absorbance320. Positive variants were further isolated

and purified for detailed kinetic analysis.

49

Figure 2-9: Combinatorial library screening. Green blocks indicate hCG-NLV and

orange blocks indicate hCGL-NLV+G. Pink blocks indicate background.

Yellow and white blocks represent different hCGL variants. Red numbers

are highlighted hCGL variants which may have higher activity in the

breakdown of methionine. In the library screening, we first isolated 34

candidates and then focus on 10 candidates. These 10 candidates were

isolated and further purified for kinetic analysis in detail. Finally, only one

P2A11 (hCGL-NLV+RGS) displayed higher activity in the breakdown of

methionine compared to hCGL-NLV+GS.

50

-1

0

1

2

3

4

5

-10 0 10 20 30 40 50

Activity assay

hCGL-NLVhCGL-NLV+RGS

1/s

[Methionine] mM

y = (m1*M0)/( m2 + M0)

ErrorValue

0.164036.245m1

0.9462815.243m2

NA0.99727R

y = (m1*M0)/( m2 + M0)

ErrorValue

0.091254.5221m1

0.260643.5395m2

NA0.99412R

Figure 2-10:Kinetic analysis of hCGL-NLV and hCGL-NLV+RGS in methionine

breakdown. The kinetic analysis was carried out at 37oC. The kcat/KM of

hCGL-NLV+RGS is approximately 3-fold better than hCGL-NLV.

51

The results of the two experiments indicate the phylogenetic analysis works for

rationally designing proteins. We obtained hCGL-NLV+RGS (K268R, T311G and

I353S) which displayed 3.2-fold better kcat/KM compared to hCGL-NLV in breakdown of

methionine. In addition, it was noted that only one variant was better than hCGL-NLV in

10 isolated clones from screening. Since protein expression could not be normalized by

this screening approach it is possible the other 9 isolated clones expressed more enzymes

and produced false positive signals.

Comprehensive phylogenetic analysis of CGLs and MGLs

Since the previous experiments showed positive results, a comprehensive

phylogenetic analysis was performed to compare full-length eukaryotic CGLs (eCGLs),

bacterial MGLs (bMGLs) and bacterial CGLs (bCGLs). 16 eukaryotic CGLs, 8 bacterial

CGLs and 18 bacterial MGLs were selected for phylogenetic analysis and 67 amino acid

positions were identified for potential mutations. These 67 amino acid candidates not

only included conservations of Situation 2 but also Situation 3 and 4 in phylogenetic

analysis (Figure 2-5). In order to reduce the potential library to a manageable size,

twenty candidates around the active site were chosen for site-directed mutagenesis:

Q49F, P52A, G53E, H55G, S63L, L91M, A92G, T94S, D112T, G116C, W155Y,

T160A, S218G, M222A, Q240K, N241D, P247L, Y317E, E349Q and A357S. These

substitutions were generated individually on the hCGL-NLV+RGS backbone. In order to

avoid the false positive effect in 96-well plate screening, these 20 variants were directly

purified by Ni-NTA column (20 ml bacterial culture) without pre-screening. Since the

initial slope of kinetic assays indicate the approximate kcat/KM values, these variants were

tested at several low methionine concentrations to determine approximate kcat/KM.

52

Variants with G116C, T160A, or P247L substitutions had no enzymatic activity. Variants

with D112T or Q240K substitution expressed poorly and did not produce enough

enzymes for testing. Additionally, variants with Q49F, P52A, G53E, H55G, A92G,

T94S, W155Y, S218G, M222A, Q240K, N241D, E349Q or A357S substitution

displayed similar or worse kcat/KM values compared to hCGL-NLV+RGS. Variants

containing S63L, L91M or Y317E substitution showed improvement compared to

hCGL-NLV+RGS in breakdown of methionine. These three variants were further

analyzed by detailed kinetic analysis to determine kcat and KM values in breakdown of

methionine (Table 2-3). The kcat of hCGL-NLV+RGS+L and hCGL-NLV+RGS+M

were 6.8 and 5.8 s-1, respectively compared to hCGL-NLV+RGS, (4.6 s-1). The KM of

these two variants are 3.3 and 2.8 mM, respectively which are similar to hCGL-

NLV+RGS (2.4 mM). Both variants had slightly better kcat value and similar KM value

compared to hCGL-NLV+RGS in breakdown of methionine and their kcat/KM values

improved only slightly.

53

Table 2-3: List of kinetic values of 20 hCGL variants in methionine breakdown.

20 hCGL variants were purified for detailed kinetic analysis. Three of

them, hCGL-NLV-RGS with S63L, L91M or Y317E substitution showed

around 2.7-fold better kcat/KM compared to hCGL-NLV.

The strategy for this work was to identify individual beneficial mutations by

phylogenetic analysis and activity assays and then further combine the mutations to test

synergy. For instance, the beneficial mutations (T311G and I353S) isolated in the pilot

experiment, worked together to make hCGL-NLV+GS. Therefore, S63L and L91M

were combined to test if they were able to synergistically improve the ability to

54

methionine breakdown. Surprisingly, hCGL-NLV+RGS+LM did not improve the KM

but significantly changed the kcat value to 2-fold higher compared to hCGL-NLV+RGS in

the breakdown of methionine (Figure 2-11). This variant displayed 5.4-fold better kcat/KM

value compared to the initial hCGL variant, hCGL-NLV. In addition, we further

combined Y317E with hCGL-NLV+RGS+LM. This variant did not show better activity

in the breakdown of methionine. Mapping S63 and L91 on the solved structure of wild-

type hCGL (PDB: 2NMP), these two positions are around active site, indicating S63L

and L91M may work with other active site substitutions such as E59N, R119L, E339V,

T311S.

-2

0

2

4

6

8

-5 0 5 10 15 20

Activity assayhCGL-NLV+RGS+LM

1/s

[Methionine] mM

y = (M0*m1)/(M0+ m2 )

ErrorValue

0.240269.0738m1

0.22692.9115m2

NA0.99451R

Figure 2-11:Kinetic analysis of hCGL-NLV+RGS+LM in methionine breakdown. Since the hCGL-NLV+RGS+L and hCGL-NLV+RGS+M variants showed

higher activity in breakdown of methionine we further combined S63L and

L91M onto hCGL-NLV+RGS. This variant displayed around 5-fold better

kcat/KM value in the breakdown of methionine compared to hCGL-NLV.

55

Compacting all kcat and KM values of isolated variants (Table 2-4), the KM

significantly reduced from 14 mM (hCGL-NLV) to around 3 mM (hCGL-NLV+RGS and

hCGL-NLV+RGS+LM) in use of methionine as substrates. Concurrently, the kcat of

variants dropped from 7.9 s-1 (hCGL-NLV) to 4.6 s-1 (hCGL-NLV+RGS) and then

improved to 9.1 s-1

(hCGL-NLV+RGS+LM). Overall, we further engineered hCGL-

NLV via phylogenetic analysis to yield a better variant, hCGL-NLV+RGS+LM which

displayed around 5% methionine breakdown activity of pMGL (3.1 s-1 mM-1 vs. 59 s-1

mM-1

).

56

Table 2-4: List of kinetic values of hCGL variants in methionine breakdown.

57

Substrate promiscuity

Besides testing the ability to break down methionine, it is also important to test if

hCGL-NLV+RGS+LM is active on other substrates. Is it a methionine-specific enzyme

or a promiscuous enzyme? We tested hCGL-NLV and hCGL-NLV+RGS+LM with L-

methionine, L-cystathionine, L-cysteine and DL-homecysteine (Figure 2-12). Since

hCGL activity is sensitive to changes in salt concentrations these two enzymes were

further purified by size-exclusion chromatography (HiLoad 16/600 Superdex 200 prep

grade, GE Healthcare Life Science) after Ni-NTA column purification. Purified hCGL-

NLV+RGS+LM showed not only 5-fold better kcat/KM value in breakdown of methionine

but also 2-fold better kcat/KM value in use of cysteine compared to hCGL-NLV (7.2 s-

1mM

-1 vs. 2.6 s

-1mM

-1 ) (Table 2-5). Moreover, it displayed 2-fold worse kcat/KM in use

of its original substrate, L-cystathionine, compared to hCGL-NLV (2.1 s-1

mM-1

vs. 1.1

s-1

mM-1

). In use of DL-homocysteine, there was no significant difference between

hCGL-NLV and hCGL-NLV+RGS+LM. In general, hCGL-NLV+RGS+LM seems to be

more like pMGL compared to hCGL-NLV in this engineering process. Additionally,

since L-cysteine and DL-homocysteine are smaller molecules and more similar to L-

methionine than L-cystathionine, it is not surprising that engineered hCGL was active in

use of L-cysteine and DL-homocysteine as in use of methionine (Figure 2-12).

58

Figure 2-12: Methionine and similar molecules.

Table 2-5: Substrate specificities of hCGL-NLV and hCGL-NLV+RGS+LM.

59

Phylogenetic analysis on positions 59, 119 and 339 of hCGL

We further engineered hCGL-NLV based on phylogenetic analysis and found

hCGL-NLV+RGS+LM did improve the activity in breakdown of methionine. We then

questioned if substitutions of E59N, R119L and E339V also matched the phylogenetic

analysis. We revisited positions 59, 119 and 339 which were chosen by structural

analysis and screened from a saturation mutagenized library. Interestingly, we found that

positions 59, 119 and 339 could be potential candidates based on the phylogenetic

analysis (Figure 2-13). The phylogenetic analysis implied isoleucine might be a better

option than asparagine at position 59 and alanine might be more beneficial than leucine at

position 119. Therefore, three variants (hCGL-INV+RGS+LM, hCGL-NAV+RGS+LM

and hCGL-IAV+RGS+LM) were generated to test the activity in methionine breakdown

and surprisingly, hCGL-IAV+RGS+LM showed higher activity in methionine

breakdown compared to hCGL-NLV+RGS+LM. The kcat/KM of hCGL-IAV+RGS+LM

was 5.4 s-1

mM-1

which was 1.5-fold better than the kcat/KM of hCGL-NLV+RGS+LM.

Both hCGL-ILV+RGS+LM and hCGL-NAV+RGS+LM were also better than hCGL-

NLV+RGS+LM though hCGL-IAV+RGS+LM was the best one (Figure 2-14 and Table

2-6).

60

Figure 2-13: Phylogenetic analysis at positions 59, 119 and 339. 16 eukaryotic CGLs,

7 bacterial CGLs and 16 bacterial MGLs were aligned by Geneious

software.

61

2

4

6

8

10

12

-2 0 2 4 6 8 10

Activity assay

hCGL-NLV+RGS+LMhCGL-IAV+RGS+LM

1/s

[Methionine] mM

y = m3+(M0*m1)/(M0 + m2)

ErrorValue

0.154997.8718kcat

0.147562.1658Km

0.0732583.3901m3

NA0.99797R

y = m3+(M0*m1)/(M0 + m2)

ErrorValue

0.210789.7936kcat

0.142021.833Km

0.113333.2275m3

NA0.99718R

Figure 2-14:Kinetic analysis of hCGL-NLV+RGS+LM and hCGL-IAV+RGS+LM

in methionine breakdown.

Table 2-6: Summary of engineered hCGL variants in methionine breakdown.

62

DISCUSSIONS

In past research, Dr. Stone has shown engineered hCGL-NLV displayed 1%

activity in breakdown of methionine compared to pMGL. Moreover, the hCGL variant

was able to inhibit tumor growth of atymic mice with LAN-1 xenografts (Stone, Paley et

al. 2012). In this work, we significantly improved the activity of hCGL variant from 1%

to 7% in breakdown of methionine compared to pMGL. We lowered the KM ten-fold

(from 12.2 mM to 1.8 mM) and restored the kcat (from 9.2 s-1

, 4.5 s-1

to 9.8 s-1

) in

methionine breakdown in the engineering process. Although we only have shown the

improved activity in methionine breakdown in in vitro assays, we are working with others

to perform in vivo analysis for the best hCGL variant (hCGL-IAV+RGS+LM). It will be

very interesting to characterize this variant for stability in pooled serum, IC50, PK and PD

in mice and eventually primates.

Advantages and disadvantages:

Phylogenetic analysis (sequence alignment of paralogs) is a classical method for

defining positions and amino acids which may be important to paralogs in the same

protein family. This simple strategy is based on the idea that “the important amino acids

and positions would be conserved in the evolution”. Since paralogs in the same protein

family are from the same ancient protein, phylogenetic analysis is useful for finding the

“conserved differences” (important positions and amino acids) which may cause the

major different properties among paralogs such as substrate specificity, pH optimum,

temperature optimum, etc. Therefore, knowing these “conserved differences” would be

useful for engineering the properties among paralogs.

This approach is different from directed evolution which starts from a randomly

mutagenized library and then follows several rounds of high-throughput screening. The

63

common directed evolution approach is a short-term evolution and usually finds limited

useful substitutions (positions and amino acids). In order to evolve more functional

substitutions, you will need iterative rounds of mutagenesis and selection. Moreover, you

may need to perform additional saturation mutagenesis to optimize the selected

substitutions. The information (conserved differences) from phylogenetic analysis are

optimized as the result of long-term evolution by nature. By replacing the conserved

differences among related paralogs it is possible to shift certain properties among

paralogs. Based on this idea, we compared bacterial MGLs and eukaryotic CGLs and

found the conserved differences which might define substrate specificity among MGLs

and CGLs. In fact, we did change the substrate specificity of hCGL by implementing this

approach. Moreover, there are still many potential candidates (positions and amino

acids) to tests.

There are several advantages in using of this approach. First, it is easy to perform

phylogenetic analysis compared to structural analysis and there are more sequences than

solved structures in the database. Second, the information (positions and amino acids)

from phylogenetic analysis is more than the information from structural analysis.

Enzyme reaction is dynamic and there are several steps involved in such as substrate

binding, catalysis, conformational change and product release. However, structural

analysis only focuses on the one state-specific interaction (the solved structure is usually

at a specific state such as apo-enzyme or enzyme with substrate analog) and most

interactions are found around the active site. Since phylogenetic analysis shows the

“conserved differences” to specific properties (such as substrate specificities), they may

be involved in substrate binding, catalysis, conformational change or product release.

Moreover, some of the interaction may not be around the active site, such as K268R (at

interface of monomers) and T311G. These key candidates far away from the active site

64

are not easy to identify by structural analysis. Third, phylogenetic analysis could help to

generate a smaller and higher yield library, and identify mutations which may work

synergistically. In this hCGL engineering work, 38 site-directed mutagenized hCGL

mutants and one combinatorial library (2048 variants) were screened for activity in use of

methionine. Surprisingly, 9 out of 38 site-directed mutagenized hCGL mutants are more

active compared to their backbone protein in the breakdown of methionine. In contrast,

only one positive hCGL variant was selected from the combinatorial library.

It seems we have discovered beneficial substitutions from phylogenetic analysis

which have higher possibility to work synergistically such as S63L and L91M. hCGL-

NLV+RGS+LM (5% activity of pMGL) is more active than hCGL-NLV+RGS+L or

hCGL-NLV+RGS+M (both are around 3% activity of pMGL). Another example is

T311G and I353S. Wild-type hCGL with the T311G or I353S substitutions individually

did not show improvement in the breakdown of methionine but hCGL-NLV (E59N,

R119L and E339V) with T311G or I353S substitution displayed synergistic work in the

breakdown of methionine compared to hCGL-NLV.

This approach also has a disadvantage. In order to perform phylogenetic analysis,

you need two groups for comparison. If there is no bacterial MGLs or eukaryotic CGLs

in nature, we would not be able to compare MGLs and CGLs, preventing us finding

conserved difference to shift substrate specificity. The only solution for this situation

would be to perform structural analysis or performing selection or screening from a

randomized pool.

It is very different among phylogenetic analysis, structural analysis and common

directed evolution from a randomized pool. In phylogenetic analysis, you know all

beneficial positions and amino acids which have already been optimized by nature. It is

65

like you have done many rounds of selection and just need to perform recombination to

discover the best combination.

Other targets

Amino acid-depletion enzymes

Besides engineering hCGL, phylogenetic analysis can be applied to similar work

associated with amino acid-depletion enzymes for human cancer treatment such as

asparagine, tyrosine, phenylalanine, glutamine, leucine and cysteine. In addition,

engineered hCGL-NLV+RGS+LM also displays ability in degrading cysteine and can be

used to target cysteine-dependent cancer cells.

Antibody-directed enzyme prodrug therapy (ADEPT)

It has been reported that methionine gamma-lyase could be applied to ADEPT

with selenomethionine for cancer treatment. ADEPT is designed for cancer treatment

and is composed of two parts, non-toxic prodrug and antibody fused enzyme which can

convert the non-toxic prodrug to toxic drug (Xu and McLeod 2001; Zawilska,

Wojcieszak et al. 2013). The idea is to localize the therapeutic enzyme onto cancer cells

with cancer marker targeting antibodies and then systemically administer the prodrug to

be converted in high concentration at tumor site. It has been found that methylselenol

(product of selenomethionine by MGL) inhibited cell growth and arrested cells at G1 and

G2 phase. Subsequently, arrested cells would undergo apoptosis (Zeng, Wu et al. 2009;

Zeng, Briske-Anderson et al. 2012). In order to achieve this goal, the therapeutic enzyme

has to be absent or in low concentration in normal cells. In addition, the enzyme should

66

not induce an immune response, hence it is better to use a human enzyme. Therefore,

engineered hCGL variant with MGL activity can be a better choice than bacterial MGL in

this approach. In addition, the concept of ADEPT is different from amino acid-

deprivation treatments for cancer, and can be used to target different cancer cells with

specific surface markers rather than amino acid-dependent cancer cells.

67

MATERIALS AND METHODS

Site-directed mutagenesis:

We used QuickChange Site-Directed Mutagenesis kit (Agilent Technology) to

generate single site-mutated hCGL variants. All used oligonucleotides were designed

based on the requirement of the kit protocol and ordered from Integrated DNA

Technologies. Mutant strand synthesis reaction (Thermal cycling): 50 ul reaction

included 250 ng oligonucleotide pair (2 X 125 ng), 50-100 ng template plasmid (hCGL

was cloned into pET28a vector), 10 nmole dNTP, 10-fold concentrated Pfu turbo buffer

(Agilent Technology) and 2.5 U Pfu turbo DNA polymerase (Agilent Technology).

Cycling Parameter: one cycle of 95oC for 30 seconds and 16 cycles of 95

oC for 30

seconds, 55oC for 1minutes, 68

oC for 2 minutes. After thermal cycling, reaction would

be treated with 20 unit of Dpn I (New England BioLabs) at 37oC for 1 hour. Finally, 4 ul

of reaction was used for heat-shock transformation on BL21(DE3) competent cells.

Combinatorial library construction:

11 positions were picked from phylogenetic analysis for shifting substrate

specificity and overlapping PCR was used to generate combinatorial library. We used

“DNAWorks” software to design oligonucleotides for combinatorial library. All

oligonucleotides ordered from Integrated DNA Technologies were further purified by

PAGE gel for higher purity. 16 oligonucleotides were designed to cover 336 basepair of

C-terminal of hCGL. Thermal cycling condition for fragment assembly: 50 ul reaction

included 0.1 uM oligonucleotides (final concentration), 20 nmole dNTP, KOD hot start

DNA polymerase buffer (EMD4Biosciences), 1 unit of KOD hot start DNA polymerase

(EMD4Biosciences). The program for thermal cycling: one cycle of 94oC for 2min and

68

24 cycles of 94oC for 30 seconds, 60oC for 30 seconds, 72oC for 1 minutes. 2.5 ul of

assembled reaction would be used for further amplification and amplified product would

be gel extracted for cloning into pET28a vector.

96-well plate screening (Stone, Paley et al. 2012):

Single colonies of E.coli BL21(DE3), containing plasmids encoding either hCGL-

NLV or variant hCGL enzymes were picked into 96-well culture plates containing 75 ul

of TB media/well and 50 ug/ml kanamycine. Cells were grown at 37oC on a plate shaker

to an OD600 of ~0.8-1, cooled to 25oC, whereupon an additional 75 ul of media containing

50 ug/ml kanamycine, and 2 mM IPTG were added and incubation with shaking was

continued for 2 hours at 25oC. Subsequently, 100 ul of culture/well were transferred to a

fresh 96 well plate (assay plate). The assay plates were centrifuged (10 minutes X 3500

g) to pellet the cells, the media was removed, and the cells were chemically lysed by

addition of 50 ul/well of B-PER protein extraction reagent (Pierce) and mixing for 5

minutes on a plate shaker. Then 20 ul of 5 mM L-methionine at pH7.3 was added to the

lysates and incubated at 37oC for approximately 12 hours. The alpha-keto acid reaction

product is then derivatized by addition of 146 ul of MBTH solution to 54 ul of reaction

and heated for 40 minutes at 50oC. The absorbance at 320 nm was determined

spectrophotometrically using a microtiter plate reader. Variants exhibiting high

absorbance values and indicative of production can thus be identified and selected for

further characterization.

69

Kinetic analysis using MBTH (Stone, Paley et al. 2012):

Kinetic analyses were performed as described elsewhere (5) to measure

ketobutyrate production by reaction with 3-methyl-2-benzothiazolone hydrazone

hydrochloride (MBTH) to generate a chromophore with a λmax of ~320 nm. Eppendorf

tubes (1.5 ml) containing 220 µL of substrate in a 100 mM sodium phosphate buffer, 1

mM EDTA (pH 7.3), and 10 uM PLP were incubated at 37°C in a heat block. Reactions

were started by adding 20.3 µL of enzyme solution and quenched with 26.7 µL of 50%

trichloroacetic acid after a set length of time (without shaking). Reactions and blanks

were then mixed with 733 µL of MBTH solution (2.2 ml: 1.6 ml of 1M acetate buffer pH

5.0 and 0.6 ml of 0.1% MBTH in same) and incubated at 50°C for 40 minutes. After

cooling, the samples were transferred to cuvettes and the A320 nm was determined. The

assay was shown to be linear between 0 - 320 µM α-ketobutyrate with a lower detection

limit of 15 µM. One unit of MGL activity was defined as the amount of enzyme that

produced 1 µmol of α-ketobutyrate per minute at infinite concentration of L-methionine.

We also followed enzyme reactions in a continuous assay by detection of product

methane-thiol with 5,5'-Dithiobis(2-nitrobenzoic acid) (DTNB) (2) using 96-well plates

and a microtiter plate reader set at 405 nm.

70

REFERENCES

Cavuoto, P., and Fenech, M.F. (2012). A review of methionine dependency and the role

of methionine restriction in cancer growth control and life-span extension. Cancer

Treat Rev 38, 726-736.

Cole, M.F., and Gaucher, E.A. (2011). Exploiting models of molecular evolution to

efficiently direct protein engineering. J Mol Evol 72, 193-203.

Durando, X., Thivat, E., Farges, M.C., Cellarier, E., D'Incan, M., Demidem, A., Vasson,

M.P., Barthomeuf, C., and Chollet, P. (2008). Optimal methionine-free diet

duration for nitrourea treatment: a Phase I clinical trial. Nutr Cancer 60, 23-30.

Goseki, N., Yamazaki, S., Shimojyu, K., Kando, F., Maruyama, M., Endo, M., Koike,

M., and Takahashi, H. (1995). Synergistic effect of methionine-depleting total

parenteral nutrition with 5-fluorouracil on human gastric cancer: a randomized,

prospective clinical trial. Jpn J Cancer Res 86, 484-489.

Inoue, H., Inagaki, K., Eriguchi, S.I., Tamura, T., Esaki, N., Soda, K., and Tanaka, H.

(1997). Molecular characterization of the mde operon involved in L-methionine

catabolism of Pseudomonas putida. J Bacteriol 179, 3956-3962.

Kokkinakis, D.M., Hoffman, R.M., Frenkel, E.P., Wick, J.B., Han, Q., Xu, M., Tan, Y.,

and Schold, S.C. (2001). Synergy between methionine stress and chemotherapy in

the treatment of brain tumor xenografts in athymic mice. Cancer Res 61, 4017-

4023.

Poirson-Bichat, F., Goncalves, R.A., Miccoli, L., Dutrillaux, B., and Poupon, M.F.

(2000). Methionine depletion enhances the antitumoral efficacy of cytotoxic

agents in drug-resistant human tumor xenografts. Clin Cancer Res 6, 643-653.

Ragsdale, S.W. (2003). Pyruvate ferredoxin oxidoreductase and its radical intermediate.

Chem Rev 103, 2333-2346.

Rao, A.M., Drake, M.R., and Stipanuk, M.H. (1990). Role of the transsulfuration

pathway and of gamma-cystathionase activity in the formation of cysteine and

sulfate from methionine in rat hepatocytes. J Nutr 120, 837-845.

Sato, D., and Nozaki, T. (2009). Methionine gamma-lyase: the unique reaction

mechanism, physiological roles, and therapeutic applications against infectious

diseases and cancers. IUBMB Life 61, 1019-1028.

Stone, E., Paley, O., Hu, J., Ekerdt, B., Cheung, N.K., and Georgiou, G. (2012). De novo

engineering of a human cystathionine-gamma-lyase for systemic (L)-Methionine

depletion cancer therapy. ACS Chem Biol 7, 1822-1829.

Sun, Q., Collins, R., Huang, S., Holmberg-Schiavone, L., Anand, G.S., Tan, C.H., van-

den-Berg, S., Deng, L.W., Moore, P.K., Karlberg, T., et al. (2009). Structural

71

basis for the inhibition mechanism of human cystathionine gamma-lyase, an

enzyme responsible for the production of H(2)S. J Biol Chem 284, 3076-3085.

Tan, Y., Sun, X., Xu, M., Tan, X., Sasson, A., Rashidi, B., Han, Q., Wang, X., An, Z.,

Sun, F.X., et al. (1999). Efficacy of recombinant methioninase in combination

with cisplatin on human colon tumors in nude mice. Clin Cancer Res 5, 2157-

2163.

Thivat, E., Farges, M.C., Bacin, F., D'Incan, M., Mouret-Reynier, M.A., Cellarier, E.,

Madelmont, J.C., Vasson, M.P., Chollet, P., and Durando, X. (2009). Phase II trial

of the association of a methionine-free diet with cystemustine therapy in

melanoma and glioma. Anticancer Res 29, 5235-5240.

Xu, G., and McLeod, H.L. (2001). Strategies for enzyme/prodrug cancer therapy. Clin

Cancer Res 7, 3314-3324.

Yang, Z., Wang, J., Yoshioka, T., Li, B., Lu, Q., Li, S., Sun, X., Tan, Y., Yagi, S.,

Frenkel, E.P., et al. (2004). Pharmacokinetics, methionine depletion, and

antigenicity of recombinant methioninase in primates. Clin Cancer Res 10, 2131-

2138.

Zawilska, J.B., Wojcieszak, J., and Olejniczak, A.B. (2013). Prodrugs: a challenge for the

drug development. Pharmacol Rep 65, 1-14.

Zeng, H., Briske-Anderson, M., Wu, M., and Moyer, M.P. (2012). Methylselenol, a

selenium metabolite, plays common and different roles in cancerous colon

HCT116 cell and noncancerous NCM460 colon cell proliferation. Nutr Cancer 64,

128-135.

Zeng, H., Wu, M., and Botnen, J.H. (2009). Methylselenol, a selenium metabolite,

induces cell cycle arrest in G1 phase and apoptosis via the extracellular-regulated

kinase 1/2 pathway and other cancer signaling genes. J Nutr 139, 1613-1618.

72

Chapter 3: Directed evolution of the substrate specificity of E.coli

biotin ligase

INTRODUCTION

Biotin is an essential cofactor and is covalently conjugated to biotin-dependent

enzymes such as acetyl-CoA carboxylase by biotin protein ligase (BPL). The E.coli BPL

first generates an intermediate, biotinoyl-5’-AMP, and subsequently transfers the biotin

moiety to a specific lysine of in biotin carboxyl carrier protein (BCCP) (Chapman-Smith

and Cronan 1999; Chapman-Smith and Cronan 1999) (Figure 3-1). E.coli BPL is a 35.5

kDa protein and its structure with bound biotin has been solved (1HXD) (Weaver, Kwon

et al. 2001). The central catalytic domain contains binding sites for biotin, ATP, and

BCCP, and undergoes a series of conformational changes in the multistep reaction

(Chapman-Smith and Cronan 1999). Protein substrates for E.coli BPL contain a

conserved sequence (MKM) that directs lysine modification. Alternatively, a 14 amino

acid peptide (biotin acceptor peptide, BAP) isolated from a peptide library (Schatz 1993)

has been widely used as a tag for protein biotinylation and subsequent purification.

By expanding the substrate specificity of biotin ligase, it may prove possible to

generate a wider array of protein-labeling reagents. For example, Ting’s group purified

biotin ligases from several species (human, S.cerevisiae, B. subtilis, P. acnes, L.

mesenteroides, T. cruzi, M. jannaschii and P. horikoshii), tested these BPLs with a

variety of biotin analogs (desthiobiotin azid, cis-propargyl biotin, trans-propargyl biotin,

iminobiotin, diaminobiotin, nitrobenzoxadiazole r-amino butyric acid, iodouracil valeric

acid and thiouracil valeric acid), and found that P. horikoshii BPL and yeast BPL could

label peptides with desthiobiotin azid and cis-propargyl biotin (Slavoff, Chen et al. 2008).

73

The range of proteins that can be biotinylated has also been expanded by using a

promiscuous E.coli biotin ligase variant that releases the reactive intermediate, biotin-5’-

AMP, and subsequently labels any adjacent lysine residue (Choi-Rhee, Schulman et al.

2004; Cronan 2005). Fusion versions of this variant have been used to label and identify

interaction partners in protein complexes (Cronan 2005; Chen, Choi et al. 2007).

BPL should be an excellent target for directed evolution in vitro, in part because

the strength and utility of the biotin:streptavidin linkage. Phage display has previously

been used to evolve a short peptide that was specifically recognized by yeast BPL. The

yeast BPL and peptide substrate proved to be orthogonal to the E. coli BPL and its

substrate, allowing two color site-specific protein labeling on cell surfaces with

fluorescent streptavidin conjugates (Chen, Choi et al. 2007).

Figure 3-1: Biotinylation reaction. The E.coli BPL first generates an intermediate,

biotinoyl-5’-AMP, and subsequently transfers the biotin moiety to a specific

lysine of in biotin carboxyl carrier protein (BCCP).

74

However, phage display can be limited by library size and skewing due to the

impact of different proteins on phage fitness. For example, a fusion protein to the phage

tail spike (pIII) may impact the functionality of the pIII signal peptide and thereby reduce

infectivity. To overcome these limitations, directed evolution has been carried out in

water-in-oil emulsions. By transcribing and translating individual DNA templates within

individual emulsion bubbles very large libraries can be screened or selected for function

(1010 variants / milliliter; (Tawfik and Griffiths 1998; Griffiths and Tawfik 2003;

Bernath, Hai et al. 2004; Cohen, Tawfik et al. 2004; Miller, Bernath et al. 2006). For

example, Griffths and Tawfik have demonstrated that IVC and FACS could be combined

to select phosphotriesterases from libraries of over 3.4 X 107 variants that have a 63-fold

higher kcat value for the degradation of paraoxon, a potent acetylcholinesterase-inhibiting

insecticide (Griffiths and Tawfik 2003; Bernath, Hai et al. 2004; Aharoni, Amitai et al.

2005; Mastrobattista, Taly et al. 2005). It should also prove possible to use similar

methods for the directed evolution of the substrate specificity of E. coli biotin ligase,

especially as we had previously used in vitro compartmentalization to evolve a biotin-

binding protein, streptavidin, with altered affinities for the biotin analog, desthiobiotin

(Levy and Ellington 2008).

75

RESULTS AND DISCUSSIONS

Development of a scheme for BirA directed evolution

Since biotin ligase is a critical enzyme in many organisms, alteration of its

substrate specificity would likely lead to toxic effects, which would contraindicate an in

vivo selection scheme. In vitro selection via phage-display methods is possible but is not

compatible with the eventual use of enzymes in trans, rather than in cis, and does not

afford the same flexibility in reaction conditions as would selections in the context of an

emulsion (Fernandez-Gacio, Uguen et al. 2003).

Therefore, we have pursued the development of a scheme for the directed

evolution of BirA in emulsions (Figure 3-2). Genes for BirA are conjugated to a peptide

tag that is a substrate for the enzyme, and are distributed into emulsion vesicles such that

there is less than one template / vesicle. The BirA genes are transcribed and translated,

and functional variants can label the peptide tags associated with their own genes. Upon

breaking the emulsion, biotinylated genes are captured with streptavidin-coated magnetic

beads, similar to a scheme we previously developed for the directed evolution of

streptavidin function (Levy and Ellington 2008). The captured genes are amplified and

redistributed to emulsion vesicles. Multiple cycles of selection and amplification are

anticipated to sieve the most functional BirA variants from a library.

76

Figure 3-2: Selection scheme of directed evolution of BPL using IVC. There are

three major steps in this selection. First, linear constructs (library) need to

covalently conjugate with biotin acceptor peptide (BAP). Second, the

conjugates and E.coli lysates are mixed with oil-surfactant to generate

individual compartments. Functional variants which can use desthiobiotin as

substrates would attach desthiobiotin to BAP. Finally, the functional

variants could be separated from other nonfunctional variants by

streptavidin-coated magnetic beads and recovered by PCR amplification.

In order to make this scheme work, each step had to be optimized, including

conjugation of peptide tag with the gene template, in vitro transcription and translation,

breaking of the emulsion, and recovery of the DNA template. Each of these

optimizations was undertaken separately, and the results combined into a workable

protocol. First, we use a chemical cross-linking reagent, sulfo-SMCC

(sulfosuccinimidyl-4-(N-maleimidomethyl)cyclohexane-1-carboxylate, from Thermo

Scientific) to conjugate the peptide tags with DNA templates, and have optimized the

77

efficiency to 35% to 50%. A number of different in vitro transcription and translation

(IVTT) options were explored, and ultimately we settled on an E.coli lysate (RTS E.coli

HY kit, from Roche). Salmon testes DNA was added to reduce DNA degradation caused

by nucleases in the lysate (Miller, Bernath et al. 2006). The recovery of biotinylated

DNA templates from the lysate was around 80% in the absence of emulsification. The

efficiency of recovery was diminished following emulsification, but optimization of yield

was partially recovered from 30% by breaking the emulsion with hexane to 60% by

breaking the emulsion with ether. The buffer used for capturing the desthiobiotin-

conjugated templates was also a variable, and surprisingly PBS was found to give better

recovery than a suggested high salt buffer. Finally, the choice of DNA polymerase for

template amplification was very important. KOD DNA polymerase was found to have

better processivity on the full-length BirA template than did Pfu and Taq DNA

polymerase, as has previously been observed (Takagi, Nishioka et al. 1997).

In order to demonstrate the viability of this scheme, we initially selected

functional BirA variants relative to a truncated, inactive enzyme (Figure 3-3). The

truncated variant was recovered from emulsions 50-fold less well than the wild-type

gene, suggesting that there might be substantive purification on a per cycle basis in a real

selection. We therefore mixed small amounts of wild-type BirA with the truncated gene,

and selected for ligation function. As shown in (Figure 3-3), when the wild-type gene

was present at only 1 part in ten, it could be quantitatively recovered.

78

Figure 3-3: Mock selection. Before the real selection, a mock selection was

performed to validate the selection scheme. We prepared four different

pools, full-length BirA pool (positive control pool), truncated BirA pool

(negative control pool) and two mixed pools. For two mixed pools, full-

length wild type BirA (functional) was mixed with truncated BirA (non-

functional) in 1:9 and 1:999 ratios. After one round of selection in emulsion,

recovery of the full-length BirA control pool was defined as 100% while

only 2% of the truncated BirA control pool was recovered. Additionally, the

1:9 ratio pool had 11% recovery. Theoretically, we should recover 10%

DNA from the 1:9 mixed pool because there are 10% of 1:9 mixed pool

DNA is full-length BirA which has enzyme activity in the selection. This

verified that the selection scheme could distinguish functional BirA from

non-functional BirA and recover functional BirA from mixed pool.

Library construction

We wanted to show the utility of our emulsion selection by identifying BirA

variants that could efficiently and orthogonally use desthiobiotin in place of biotin. Two

different libraries were constructed, one that targeted residues that were thought to

interact with biotin, and one that contained random mutations introduced by error-prone

79

PCR. For both pools, we added a C-terminal GGGS link and His6 tag at the C-terminal

end for protein detection and purification.

The structure of E.coli biotin ligase-biotin complex has been solved (Wilson,

Shewchuk et al. 1992), and Tyr-132, Gly-204 and Gly-206 of E.coli are close to the

thiophene ring of biotin. Additionally, Gly-117, Lys-183, Gly-186 and Leu-188 of E.coli

biotin ligase are close to the junction of pentanoic acid and thiophene ring in biotin. In

desthiobiotin, there is no thiophene ring and hexanoic acid directly links to uredio ring.

This difference may affect the binding of E.coli biotin ligase to desthiobiotin (Wu and

Wong 2004). Thus we generated a BirA template that was completely randomized at

these 7 positions by substituting codons with NNK (K=G or T), which spans the 20

amino acids contains only 3% stop codons relative to NNN (Braunagel and Little 1997).

Of course, residues other than those abutting the thiophene ring may prove

important for catalysis with desthiobiotin, and we therefore also generated a pool using

traditional error-prone PCR methods (Fromant, Blanquet et al. 1995). The substitution

rate of the randomized pool was around 2.4% / position, or ca. 24 mutations / gene

(Figure 3-4).

80

Figure 3-4: Two libraries construction. There are two libraries were generated. One is

generated by error-prone PCR for random mutagenesis and the substitution

rate is 2.4%. The substitutions would randomly distribute over the entire

coding sequence. The other library which was generated by overlapping

PCR with designed oligonucleotides (containing 7 positions with NNK

codon).

Selection for DTB utilization

The two libraries were initially selected individually for their ability to conjugate

desthiobiotin to the peptide-gene chimera. A high concentration of DTB (200

micromolar) was utilized to help ensure efficient labeling by enzymatic variants. After

several rounds, the concentration of DTB was dropped to 50 micromolar, and the two

pools were combined for the final three rounds of selection. We did not include a

counterselection with biotin, given that such a counterselection previously failed to yield

81

orthogonal streptavidin variants that could utilize desthiobiotin in preference to biotin

(Levy and Ellington 2008). Therefore, the current selection was primarily to expand and

/ or alter the substrate preference of the E. coli BirA (Slavoff, Chen et al. 2008;

Uttamapinant, White et al. 2010).

The pools were cloned after four (just prior to mixing the pools) and six rounds of

selection. Interestingly, only variants from the PCR-mutagenized pool survived the

selection, which may indicate that only relatively subtle structural alterations to the

biotin-binding pocket remain active with desthiobiotin.

Figure 3-5: The selection process of BPL library 1 and BPL library 2.

82

The coding region of BirA is 963 nucleotides and can be divided into three

domains composed of 321 amino acids. We sequenced 19 full-length clones from the

Round 6 pool (Table 3-1). Alignment revealed a predominant substitution, M157T (15

clones). Another 2 clones had a valine substitution at this position. There were other

positions where statistically significant substitutions occurred, including A129 (V or T)

(36%, 7 clones), and I224 (V or T) (21%, 4 clones). Residues I40, F51, G74, A82, K163

and T225 had less frequent but still significant substitutions (15%, 3 substitutions).

Table 3-1: List of substitutions of isolated biotin ligase variants from Round 6

pool. Mutations of 10 active biotin ligase variants from Round 6 pool are

listed in the table. Position 157 is the most well-conserved substitution and

shows in all isolated biotin ligase variants. In addition, position 129, 224

and 225 are the other three with conserved substitutions.

Comparing DTB-utilizing variants

A gel-shift assay was used to characterize the activity of biotin ligase variants

with desthiobiotin as a substrate. Radiolabeled (35

S-methionine labeled) BirA protein

83

variants and a peptide-tagged chloramphenicol acetyltransferase were synthesized

separately in lysates. The proteins were then co-incubated in the presence of desthiobiotin

for 8 minutes. After reaction, recombinant streptavidin was added and following gel

electrophoresis a shift in the radioactive CAT band was indicative of BirA activity. The

activity of variants was initially assessed by determining the ratio of shifted signal

(complex amount) to the amount of enzyme present (enzyme amount).

Ten clones showed higher activity than wild-type with desthiobiotin compared to

BirAwt. These 10 most improved BirA variants all contained similar substitutions: M157

to T or V (100%), A129 to T or V (40%), I224 to T or V (40%) and T225 to A (30%)

(Table 3-1). The variants BirA6-22, BirA6-47, and BirA6-40 were best able to utilize

desthiobiotin, and were 8-fold, 10-fold, and 17-fold better than the wild-type in our assay,

respectively (Figure 3-6).

Since the pool was selected for desthiobiotin utilization it was unclear whether it

would retain its ability to utilize biotin as a substrate. We therefore assayed BirA6-22,

BirA6-47, and BirA6-40 using biotin and found that their activities were 0.06-fold, 0.19-

fold, and 0.91-fold that of the wild-type BirA, which would imply 130-fold, 54-fold, and

19.2-fold improvements in the use of desthiobiotin relative to biotin compared with

BirAwt (Figure 3-6). The ability to best utilize desthiobiotin correlated well with the

retention of the ability to utilize biotin, while the greatest orthogonality occurred in

enzymes that had disproportionately lost the ability to utilize biotin. This observation is

consistent with the notion we and others have previously put forward, that enzymes that

are undergoing changes in substrate specificity proceed through a more 'generalist'

intermediate (Khersonsky and Tawfik 2010). That said, when we assayed BirAwt, BirA6-

22, BirA6-22(V106A), BirA6-40, and BirA6-47 with iminobiotin as a substrate none of them

showed easily measured activity. Thus, the remodeling of the active site appeared to

84

proceed along a path that moved gradually from specificity for biotin to specificity for

desthiobiotin.

Figure 3-6: Activity assay of isolated variants (gel-shift assay). This is a radiolabeled

gel-shift assay. All proteins were labeled with 35

S-methionine and then

exposed to phosphor screen. Different BirA variants and CAT-BAP (peptide

substrate) were synthesized with in vitro transcription and translation

reagents containing 35

S-methionine and then to test the activity of BirA

variants in use of biotin or desthiobiotin. Higher activity variants would

modify more CAT-BAP (peptide substrate) and then generate more shifted

signal (complex of modified CAT-BAP and recombinant streptavidin). 4

variants with higher activity in use of desthiobiotin were chosen for further

comparison. The variants showed differing activities compared to BirAwt.

Increased “fold” activity was compared in 20 uM biotin: BirA4-77 (0.8-fold),

BirA6-22 (0.06-fold), BirA6-40 (0.91-fold), BirA6-47 (0.19-fold) and in 50 uM

desthiobiotin: BirA4-77 (10-fold), BirA6-22 (8-fold), BirA6-40 (17-fold),

BirA6-47 (10-fold). These data imply 12-fold, 130-fold, 19-fold and 54-fold

activity improvement of variants over wild-type protein in the presence of

desthiobiotin relative to biotin.

85

The positions of the most frequent substitutions were mapped onto the biotin-

BirA structure (1HXD) (Weaver, Kwon et al. 2001). The substituted position A129 (to T

or V) is near to the biotin-binding loop (amino acids 110 to 128), while the substituted

positions I224 (to V or T) and T225 (to A) are in a different loop (amino 212 to 233) that

is known to change its conformation from disordered to ordered upon biotin-binding

(Weaver, Kwon et al. 2001). Interestingly, the most well-conserved substitution, T157

(V or T) is far from the biotin binding site.

Kinetic characterizations of biotin ligase variants

More detailed kinetic characterizations were carried out with the best

desthiobiotin-utilizing variants. Our assay was based on the release of pyrophosphate in

the first step of the ligase reaction. We initially measured the kcat and KM of BirAwt with

ATP, and found that the measured KM value (0.22 mM) was very similar to the value that

was previously published (0.2 mM). Similarly, the measured kcat, 0.48 s-1

, was close to

similar to the range of values that have appeared (0.17 to 0.44 s-1

) (Chapman-Smith,

Mulhern et al. 2001).

We proceeded to use this method with desthiobiotin. However, since

desthiobiotin is not readily soluble in water it was initially dissolved in dimethyl

sulfoxide (DMSO) and then diluted with water. While it has been shown that DMSO can

impair wild-type biotin ligase activity, we characterized all of the BirA variants using the

a concentration of DMSO (2.4%) that has been shown to leave wild-type enzyme activity

largely intact (90%) (Nah 2009). Surprisingly, the kcat of BirA6-40 and BirAM157T were

found to be 0.73 s-1

and 0.53 s-1

, respectively; 2-fold faster than the kcat of BirAwt with

desthiobiotin (0.29 s-1

). However, the KM of BirA6-40 and BirAM157T for desthiobiotin

86

were 6.7 mM and 4.3 mM, values that are 2- to 3-fold worse than the KM of BirAwt (0.17

mM) (Figure 3-7).

Figure 3-7: Kinetic characterization of BirAwt, BirA6-40and BirAM157T. Purified

BirAwt, BirA6-40 and BirAM157T were characterized by kinetic analysis via

detection of pyrophosphate generated in the biotin ligase reaction. BirA6-40

and BirAM157T show much lower kcat/KM compared to BirAwt in use of

biotin. However, the KM values for BirA6-40 and BirAM157T are too high to

define. Additionally, BirA6-40 and BirAM157T show 0.6-fold and 0.7-fold

kcat/KM in the presence of desthiobiotin as compared to BirAwt. Moreover,

BirA6-40 and BirAM157T have around 2-fold higher kcat value with

desthiobiotin as compared to BirAwt.

87

Selected variants were also characterized for their ability to use biotin. The very

high KM values of both BirA6-40 and BirAM157T for biotin prevented measurement of

individual kinetic constants, but the ratio kcat/KM could be determined. This value was

similar for both biotin ligase variants, around 0.01 s-1 mM-1. This value is roughly 5

orders of magnitude less than that of the wild-type BirA (106 s

-1 M

-1) (Figure 3-7).

The pyrophosphate assay proved limiting for attempting to characterize the

kinetics of the second substrate, the peptide BAP. However, if we consider the gel-shift

assays (Figure 3-6 and 3-8) and kinetic assays (Figure 3-7) for BirAwt and BirA6-40 with

50 uM desthiobiotin, it might be expected that BirAwt should be faster than BirA6-40 with

desthiobiotin, since BirAwt has a larger kcat/KM value. In fact, the result of the gel-shift

assay shows a stronger signal for BirA6-40 (this was consistent across two independent

gel-shift assays, Figure 3-6 and Figure 3-8). Taken together, these results imply that

BirA6-40 has a smaller KM for BAP than does the wild-type enzyme.

The observed changes in the kinetic parameters seem to mirror the particular

features of the emulsion-based selection method. The fact that the amount of

desthiobiotin provided was essentially saturating throughout the selection led to 2-fold

improvements in kcat for desthiobiotin, and may have also allowed an increase in the KM

for desthiobiotin. Conversely, the low peptide concentration in each compartment (1

peptide / compartment, or ca. 0.3 nM) led to selective pressure for improved binding of

the peptide. While it is possible that variants exist that show greatly improved kcat and

KM values for each substrate, such variants were not seen and are apparently rare relative

to variants where only the kinetic parameter actually under selection (kcat) is improved.

In general the stepwise optimization of individual kinetic parameters would seem to be

more productive than attempts to find rare variants that are simultaneously optimized

over a range of kinetic and biophysical parameters.

88

Evolutionary paths

In order to determine how the various prevalent mutations contributed to the

change in substrate specificity, we created variants containing individual amino acid

substitutions based on the three clones with the highest activities. Some 12 single

substitutions were introduced into the wild-type protein: D3G, I11V, Q41R, P85S,

A106V, A129V, M157T, I224T and T225A, K163E, E215D and Q255R, and the

activities of these variants were again assessed using both desthiobiotin and biotin as

substrates (Figure 3-8).

89

Figure 3-8: Activity assay of variants with single reintroduced substitutions (gel-

shift assay). This is a radiolabeled gel-shift assay. All proteins were

labeled with 35

S-methionine and then exposed to phosphor screen. Different

site-directed mutagenized BirA variants and CAT-BAP (peptide substrate)

were synthesized with in vitro transcription and translation reagents

containing 35

S-methionine and then to test the activity of BirA variants in

use of biotin or desthiobiotin. Higher activity variants would modify more

CAT-BAP (peptide substrate) and then generate more shifted signal

(complex of modified CAT-BAP and recombinant streptavidin). 12

substitutions from isolated biotin ligase variants were reintroduced into wild

type BirA and assayed the activities in use of desthiobiotin. In this assay,

variants showed a high level of improvement in activity over the wild type.

BirAA129V (8.7-fold), BirAM157T (16.3-fold), BirAE215D (9.0-fold) and

BirAI224T (7.8-fold) showed the greatest increase in activity over the wild

type in the presence of desthiobiotin.

90

The single, most highly observed mutation (95.7% of the Round 6 pool, Next-Gen

sequencing, Table 3-2), BirAM157T, on its own shows much higher activity with

desthiobiotin (a kcat of 0.53 s-1

and a KM of 4.3 mM with desthiobiotin, compared with a

kcat of 0.73 s-1 and a KM of 6.7 mM for the best isolated BirA variant, BirA6-40) (Figure 3-

7). While the M157T substitution does not directly contact biotin it is close to highly

conserved residues such as Tyr-132, Gly-204, Ala-205 and Gly-206 which in turn are

close to the thiophene ring of biotin (Wu and Wong 2004). Also Met-157 is in proximity

to Gly-186 and Ile-187 that are near the junction of the pentanoic acid and thiophene ring

of biotin (Figure 3-9, left). All of the substitutions at this position are smaller than

methionine (T (59% of the total), V (30%), I (4%) and A (1%)). Interestingly, it has been

shown that Pyrococcus horikoshii biotin ligase can use desthiobiotin-azide (DTB-Az)

(Slavoff, Chen et al. 2008), and the position corresponding to Met-157 in this enzyme is

valine. However, this may not be significant as other biotin ligases from

Bacillus, Leuconostoc, and Methanococcus also have valine at this position and these

were found not to be able to use DTB-Az as a substrate.

91

Table 3-2: Substitution distribution of selected BirA pool (Round 6) by Next-

Generation sequencing. Selected BirA pool (Round 6) was analyzed by

454 Next-Generation sequencing. 2389 reads were organized by frequency.

We hypothesize that by reducing the size of Met-157 desthiobiotin may be bound

more loosely and can therefore find new poses within the enzyme active site that improve

one or more steps of catalysis. A similar result was observed during the directed

evolution of streptavidin to utilize desthiobiotin (Levy and Ellington 2008; Magalhaes,

Czekster et al. 2011). The binding pocket of a streptavidin variant that could bind

desthiobiotin was wider than was the case for the wild-type streptavidin, and the extra

92

space gave the open thiophene ring more rotational freedom. Thus, although the M157T

does not directly contact to desthiobiotin reducing the size of Met-157 might mean fewer

structural constraints on immediately adjacent residues such as Tyr-132, Gly-204, Ala-

205, Gly-206, Gly-186, and Ile-187, and thus possibly also more rotational freedom for

binding desthiobiotin (Figure 3-9, right).

The next most highly observed substitutions BirAA129V and BirAI224T also show

higher activity with desthiobiotin than the wild-type E.coli biotin ligase, as does

BirAE215D (Figure 3-6). That said, there are clearly multiple paths to high activity clones,

as the best variant, BirA6-40, contains only 2 of the 4 highly observed substitutions, while

the next most active variants contain 3 of the 4 highly observed substitutions.

In contrast, BirAA106V has lost activity with both biotin and desthiobiotin. This

substitution was introduced based on the sequence of BirA6-22. It seems likely that

A106V is a deleterious mutation that randomly arose, and has been compensated for by

the other mutations in BirA6-22. To test this hypothesis, we 'repaired' the A106V

substitution in BirA6-22 and found there was no significant difference compared to BirA6-

22 in the use of biotin or desthiobiotin. While none of the other 9 active variants

originally screened contained substitutions at positions 106, deep sequencing revealed

that 7% of selected variants contained substitutions at this position. Thus, A106V may be

a potentiating or compensatory mutation that improves the fitness variants that proceed

along a specific mutational path. Deep sequencing provided sufficient data to search for

nearby covariations, and we discovered that A106V demonstrated significant covariance

with A129V (p<0.0001), a mutation that appeared 22.1% in Round 6 pool (Rhee, Liu et

al. 2007).

93

Figure 3-9: Mapping Met-157 onto the structure of E.coli biotin ligase with biotin.

E.coli biotin ligase is shown here with biotin (1HXD). Tyr-132 (blue

spheres), Gly-204, Ala-205, Gly-206 (pink spheres), Ile-187, Gly-186

(yellow spheres), Met-157 (left, assorted color spheres) and M157T (right,

assorted color spheres, mutagenized by PyMOL) are labeled on the 1HXD.

Tyr-132, Gly-204, Ala-205, Gly-206, Gly-186 and Ile-187 are close to

biotin’s thiophene ring or to the junction of the pentanoic acid and thiophene

ring in biotin. M157T substitution does not directly interact with biotin but

may affect aforementioned conserved amino acids (blue, pink and yellow

spheres). Substitution of the methionine with threonine or another smaller

amino acid (valine, isoleucine or alanine based Next-Gen sequencing result)

may provide more space for desthiobiotin at binding pocket of enzyme.

The identification of a deleterious substitution in an otherwise active variant is

consistent with the fact that while 10 of the 19 clones from Round 6 had improved

activities, 8 had virtually no activity with desthiobiotin. While this could indicate that the

selection stringency was not sufficiently high, this hypothesis is contradicted by the fact

94

that we were able to recover clones with improved activities. Therefore, it seems more

likely that this result indicates that there is a large mutational burden that occurs as a

result of the amplification process.

95

MATERIALS AND METHODS:

Library preparation:

To begin selection, two biotin protein ligase pools were created. The first was

generated by error-prone PCR for the random mutagenesis of E. coli BirA per Fromant et

al. (Fromant, Blanquet et al. 1995). Buffer conditions were: 10 mM Tris, pH 8.7, 50 mM

KCl, 5 µg/ml BSA, 0.5 µM of each of the two primers, 0.12 mM dATP, 0.10 mM dCTP,

0.55 mM dGTP, 3.85 mM dTTP, 0.5 mM MnCl2, 4.8 mM Cl2, 20 pg of template DNA

(966 bp), and 2 units of Taq DNA polymerase (New England Biolabs). Amplification

was carried out over 25 cycles and amplified product was subsequently gel-extracted then

further mutagenized for another 25 cycles. The second library was generated by

saturation mutagenesis at 7 specific positions. After mutagenesis, the assembled

fragments were overlapped with N-terminal and C-terminal fragments of the E. coli BirA

gene in order to generate a full coding region. Finally, the randomized coding regions

were overlapped with commercial fragments (Roche) containing T7 promoters and

terminators to generate the full linear constructs.

Cross-linking BAP with DNA:

Sulfo-SMCC (Pierce) was used to cross-link the BAP (23 amino acids,

CGGGSGGGSGLNDIFEAQKIEWH, from Bio-Synthesis) and DNA linear constructs

(1.3 kb). First, DNA library was amplified with an oligonucleotide primer containing a

5’-amino group. After this, 2 µg of amplified library DNA was mixed with 10 mM, 500

µl sulfo-SMCC at room temperature for 1 hr. In the next step, activated DNA library was

purified by silicon-member-based DNA purification spin column (Qiagen) and then

96

incubated with BAP overnight at room temperature. Finally, the BAP-library conjugates

were purified by silicon-membrane-based DNA purification spin column again.

In vitro compartmentalization selection (Miller, Bernath et al. 2006; Levy and

Ellington 2008):

The DNA library was added into RTS E. coli lysate (Roche) with 0.5% sodium

deoxycholate. Then it was emulsified with 500 µl of an oil-surfactant mixture containing

mineral oil, 4.5% Span 80, 0.5% Tween 80, and 0.1% Triton X-100 by stirring for 4 min

on ice. Subsequently, the emulsion reactions were incubated at 30 °C for 2 hr. To stop the

reactions, the emulsions were incubated at 90 °C for 15 min. The aqueous phase was

extracted the first time with 900 µl of water-saturated ether in the presence of 500 µl

PBS. It was further extracted two more times with 900 µl water-saturated ether. Excess

ether was removed by vacuum centrifugation for 10 min at room temperature. Free

desthiobiotin in the extracted reaction was removed by silicon-membrane-based DNA

purification (Qiagen). Purified DNA was incubated with streptavidin-coated magnetic

beads for capture. After three PBS washes, captured DNA was recovered by PCR

amplification.

Gel-shift assay:

35S-methionine-labeled BPL and CAT-BAP (peptide substrate) were generated

separately from Roche HY100 E. coli lysate after 30 °C incubation for 3 hr.

Subsequently, BPL and CAT-BAP were mixed together with 4 mM ATP; these solutions

were further mixed with either 20 µM biotin or 50 µM desthiobiotin for 8 min. Reactions

were then terminated by heating the mixtures for 10 min at 95 °C. In order to

97

differentiate the modified CAT-BAP (biotinylated or desthiobiotinylated), 3.8 µmol

recombinant streptavidin was added into the reactions to bind modified CAT-BAP. A

complex of streptavidin and modified CAT-BAP showed a shifted signal on a 4-20%

gradient SDS-PAGE gel. The signals of all proteins were be detected by phosphor

screen.

BPL protein purification (Naganathan and Beckett 2007):

The BirA variants were first subcloned into a pRS.1 plasmid containing a tac

promoter and then transformed into JM109 competent cells for protein expression. Cells

grew in Terrific Broth medium (Sigma-Aldrich) and protein expression was induced by 1

mM IPTG for 5 hr after OD600 reached 0.6-0.9. Soluble protein was separated from

debris by 40,000 g centrifugation for 30 min and further purified using a Ni-NTA column

(Qiagen). The purified BPL was then run through FPLC with HiLoad 16/600 Superdex

75 pg column (GE) for higher purity and buffer exchange. Storage buffer for purified

BPL was composed of 50 mM Tris-HCl, pH 7.5, and 200 mM KCl.

Pyrophosphate detection:

An EnzChek pyrophosphate assay kit (Life Technologies) was used for the

detection of generated pyrophosphate from the biotinylation reaction; steps were

modified as necessary to fit the experiment. In the first step, the biotinylation (or

desthiobiotinylation) reactions (50 mM bicine buffer, pH 8.3, 10 mM Magnesium

oxaloacetate, 2 mM ATP, 70 µM BAP, and 0.5-1 µM enzyme in 50 µl volume) were

carried out at 30 °C (with different concentrations of biotin or desthiobiotin).

Subsequently, the reactions were inactivated at 90 °C for 4 min at different time points.

98

Finally, these terminated reactions were mixed with 50 µl pyrophosphate reagents and

incubated at 22 °C for 1 hr. 95 µl of the final reaction were transferred to a new 96-well

plate and measured for absorbance at 360 nm. Readouts were taken at various time points

to determine the rate of reaction at different concentrations of biotin or desthiobiotin.

99

REFERENCES

Aharoni, A., Amitai, G., Bernath, K., Magdassi, S., and Tawfik, D.S. (2005). High-

throughput screening of enzyme libraries: thiolactonases evolved by fluorescence-

activated sorting of single cells in emulsion compartments. Chem Biol 12, 1281-

1289.

Bernath, K., Hai, M., Mastrobattista, E., Griffiths, A.D., Magdassi, S., and Tawfik, D.S.

(2004). In vitro compartmentalization by double emulsions: sorting and gene

enrichment by fluorescence activated cell sorting. Anal Biochem 325, 151-157.

Braunagel, M., and Little, M. (1997). Construction of a semisynthetic antibody library

using trinucleotide oligos. Nucleic Acids Res 25, 4690-4691.

Chapman-Smith, A., and Cronan, J.E., Jr. (1999a). The enzymatic biotinylation of

proteins: a post-translational modification of exceptional specificity. Trends

Biochem Sci 24, 359-363.

Chapman-Smith, A., and Cronan, J.E., Jr. (1999b). Molecular biology of biotin

attachment to proteins. J Nutr 129, 477S-484S.

Chapman-Smith, A., Mulhern, T.D., Whelan, F., Cronan, J.E., Jr., and Wallace, J.C.

(2001). The C-terminal domain of biotin protein ligase from E. coli is required for

catalytic activity. Protein Sci 10, 2608-2617.

Chen, I., Choi, Y.A., and Ting, A.Y. (2007). Phage display evolution of a peptide

substrate for yeast biotin ligase and application to two-color quantum dot labeling

of cell surface proteins. J Am Chem Soc 129, 6619-6625.

Choi-Rhee, E., Schulman, H., and Cronan, J.E. (2004). Promiscuous protein biotinylation

by Escherichia coli biotin protein ligase. Protein Sci 13, 3043-3050.

Cronan, J.E. (2005). Targeted and proximity-dependent promiscuous protein

biotinylation by a mutant Escherichia coli biotin protein ligase. J Nutr Biochem

16, 416-418.

Fernandez-Gacio, A., Uguen, M., and Fastrez, J. (2003). Phage display as a tool for the

directed evolution of enzymes. Trends Biotechnol 21, 408-414.

Fromant, M., Blanquet, S., and Plateau, P. (1995). Direct random mutagenesis of gene-

sized DNA fragments using polymerase chain reaction. Anal Biochem 224, 347-

353.

Griffiths, A.D., and Tawfik, D.S. (2003). Directed evolution of an extremely fast

phosphotriesterase by in vitro compartmentalization. EMBO J 22, 24-35.

Khersonsky, O., and Tawfik, D.S. (2010). Enzyme promiscuity: a mechanistic and

evolutionary perspective. Annu Rev Biochem 79, 471-505.

100

Levy, M., and Ellington, A.D. (2008). Directed evolution of streptavidin variants using in

vitro compartmentalization. Chem Biol 15, 979-989.

Magalhaes, M.L., Czekster, C.M., Guan, R., Malashkevich, V.N., Almo, S.C., and Levy,

M. (2011). Evolved streptavidin mutants reveal key role of loop residue in high-

affinity binding. Protein Sci 20, 1145-1154.

Mastrobattista, E., Taly, V., Chanudet, E., Treacy, P., Kelly, B.T., and Griffiths, A.D.

(2005). High-throughput screening of enzyme libraries: in vitro evolution of a

beta-galactosidase by fluorescence-activated sorting of double emulsions. Chem

Biol 12, 1291-1300.

Miller, O.J., Bernath, K., Agresti, J.J., Amitai, G., Kelly, B.T., Mastrobattista, E., Taly,

V., Magdassi, S., Tawfik, D.S., and Griffiths, A.D. (2006). Directed evolution by

in vitro compartmentalization. Nat Methods 3, 561-570.

Naganathan, S., and Beckett, D. (2007). Nucleation of an allosteric response via ligand-

induced loop folding. J Mol Biol 373, 96-111.

Nah, B.N.L. (2009). High-throughput assays for biotin protein ligase: A novel antibiotic

target. Master's thesis, 1-145.

Rhee, S.Y., Liu, T.F., Holmes, S.P., and Shafer, R.W. (2007). HIV-1 subtype B protease

and reverse transcriptase amino acid covariation. PLoS Comput Biol 3, e87.

Schatz, P.J. (1993). Use of peptide libraries to map the substrate specificity of a peptide-

modifying enzyme: a 13 residue consensus peptide specifies biotinylation in

Escherichia coli. Biotechnology (N Y) 11, 1138-1143.

Slavoff, S.A., Chen, I., Choi, Y.A., and Ting, A.Y. (2008). Expanding the substrate

tolerance of biotin ligase through exploration of enzymes from diverse species. J

Am Chem Soc 130, 1160-1162.

Takagi, M., Nishioka, M., Kakihara, H., Kitabayashi, M., Inoue, H., Kawakami, B., Oka,

M., and Imanaka, T. (1997). Characterization of DNA polymerase from

Pyrococcus sp. strain KOD1 and its application to PCR. Appl Environ Microbiol

63, 4504-4510.

Uttamapinant, C., White, K.A., Baruah, H., Thompson, S., Fernandez-Suarez, M.,

Puthenveetil, S., and Ting, A.Y. (2010). A fluorophore ligase for site-specific

protein labeling inside living cells. Proc Natl Acad Sci U S A 107, 10914-10919.

Weaver, L.H., Kwon, K., Beckett, D., and Matthews, B.W. (2001). Corepressor-induced

organization and assembly of the biotin repressor: a model for allosteric activation

of a transcriptional regulator. Proc Natl Acad Sci U S A 98, 6045-6050.

Wilson, K.P., Shewchuk, L.M., Brennan, R.G., Otsuka, A.J., and Matthews, B.W. (1992).

Escherichia coli biotin holoenzyme synthetase/bio repressor crystal structure

delineates the biotin- and DNA-binding domains. Proc Natl Acad Sci U S A 89,

9257-9261.

101

Wu, S.C., and Wong, S.L. (2004). Development of an enzymatic method for site-specific

incorporation of desthiobiotin to recombinant proteins in vitro. Anal Biochem

331, 340-348.

102

Chapter 4: demonstration of cooperation and co-evolution of synthetic

operon

INTRODUCTION

Circuits and pathways in living organisms are complex, as they are finely tuned

and co-optimized throughout the course of evolution. An operon is a simple form of a

genetic circuit which contains a cluster of genes under the control of a single promoter.

For example, the canonical lac operon contains lacY (beta-lactose permease), lacZ (beta-

galactosidase), and lacA (thiogalactoside acetyltransferase) driven by a single promoter

(Figure 4-1). The lac operon is required for the lactose catabolism and is tightly

controlled to response to the environment stimulus by cooperation of lacY and lacZ. The

beta-lactose permease transfers lactose into cells and then the beta-galactosidase digests

lactose to galactose and glucose (Juers, Matthews et al. 2012). Therefore, bacteria use

simple sugars as carbon source. Although the role of thiogalactoside acetyltransferase is

controversial, it has been hypothesized to play a role in detoxification (Roderick 2005).

Tight regulation of lac operon is based on the availability of lactose. Without

lactose present, the lac I (repressor) is constitutively expressed and binds the operator (O)

which is just downstream of the lac operon promoter (P). The repressor blocks the RNA

polymerase from binding the promoter and transcribing lacY, lacZ and lacA in order to

avoid unnecessary enzyme production (Wilson, Zhan et al. 2007). With lactose present,

the metabolite, allolactose, binds to the repressor which changes its conformation to

allow release from the operator. Once the operator is unbound, RNA polymerase can

bind to the promoter and transcribe lacY, lacZ and lacA to conduct lactose metabolism.

The long evolutionary history of the lac operon has allowed for fine tuning of

individual genes, as well as, the interplay between genes in the operon. The creation of a

synthetic operon with analogous complexity and evolved cooperativity would enable

103

better understanding of the process of natural evolution. Through the course of an in vitro

evolutionary experiment, similar evolutionary development can be observed as evolution

orchestrates the co-optimization of parts (Davidson, Meyer et al. 2012).

Figure 4-1: Structure of Lac operon. The lac operon is composed of three structure

genes, lacY, lacZ and lacA, which encode three different enzymes for

lactose metabolism. Besides three structure genes, it has one promoter (P),

operator (O) and lacI (repressor). The lacI is constitutively expressed to

block the transcription of structure genes by binding operator and prevent

the wasting of enzymes production without lactose present as a carbon

source in the environment. These regulators tightly control the cooperation

of these three structures for lactose metabolism in bacteria.

Biotin ligase and streptavidin are functionally relevant proteins and have been

evolved by our lab (Chapman-Smith and Cronan 1999; Levy and Ellington 2008;

Magalhaes, Czekster et al. 2011) (Figure 4-2). It would be interesting to combine a biotin

ligase (BirA) with streptavidin (SA) under the same T7 promoter to demonstrate

cooperation and co-evolution of synthetic operons. The idea is to demonstrate the

synthetic operon in use of desthiobiotin.

104

Figure 4-2: Synthetic operons. Levy has evolved streptavidin to acquire better off-rate

to desthiobiotin and we have evolved E.coli biotin ligase with higher

activity in use of desthiobiotin at low peptide concentration. Combining a

streptavidin and a biotin ligase under T7 promoter, we generated synthetic

operons which should cooperate in use of desthiobiotin.

105

RESULTS AND CONCLUSIONS

Construction scheme:

In order to demonstrate cooperation and co-evolution of synthetic operons, two

genes (streptavidin and biotin ligase) with two versions (wild-type version optimized for

biotin and evolved version optimized for desthiobiotin) were chosen to generate four

different synthetic operons (Figure 4-3) (Chapman-Smith and Cronan 1999; Levy and

Ellington 2008; Nah 2009; Magalhaes, Czekster et al. 2011). The constructs: [SAwt-

BirAwt], [SAwt-BirA6-40], [SA7-6-BirAwt] and [SA7-6-BirA6-40] were placed under the same

T7 promoter for transcription. Each of these operons also had two separate ribosomal

binding sites for translation and all conjugated with biotin acceptor peptide (BAP) at two

ends (5’ of DNA). BAP is a peptide substrate for biotin ligase (Schatz 1993). These

four different synthetic operons (as a representative pool) were selected in use of

desthiobiotin.

106

Figure 4-3: A Representative synthetic operon pool. Two different streptavidin, wild-

type streptavidin (SAwt), streptavidin variant (SA7-6) and two different biotin

ligases, wild-type BirA (BirAwt) and BirA variant (BirA6-40) were used to

generate four synthetic operons as a representative pool. They were [SAwt-

BirAwt], [SAwt-BirA6-40], [SA7-6-BirAwt] and [SA7-6-BirA6-40]. Each of

synthetic operon has unique restriction site in the middle of construct and

can be differentiated from each other by restriction enzyme digestion.

Selection scheme:

A selection scheme was designed to test these 4 synthetic operons (Figure 4-4).

In order to “survive” in this selection scheme, two genes in the same operon have to

cooperate together in emulsions. First, a representative pool (four operons) is conjugated

to BAP which is a substrate for biotin ligase (BAP:operon conjugates). Then they are

emulsified with E.coli lysates to generate compartments which is less than one conjugate

/ compartment and the conjugates are transcribed and translated in emulsions. A

synthesized biotin ligase recognizes desthiobiotin and then attaches it on BAP of

conjugate. A synthesized streptavidin binds the modified conjugate. After breaking the

emulsions, anti-His antibody agarose beads capture the complex (His-tagged streptavidin

and desthiobiotinylated BAP:operon conjugates). Finally, selected operon recovers by

PCR amplification and analyzes by restriction enzymes. Different synthetic operons

survive in different selection conditions. We would expect that [SAwt-BirAwt] survives

with biotin present and [SA7-6-BirA6-40] survives with desthiobiotin present in the

selection. The selection platform is water-in-oil emulsions and is similar to a scheme we

implemented in Chapter 3 (Figure 4-4) (Griffiths and Tawfik 2003; Miller, Bernath et al.

2006; Levy and Ellington 2008; Davidson, Dlugosz et al. 2009).

107

Figure 4-4: Selection scheme of synthetic operons using IVC. There are several steps

in this selection. First, a representative operon pool (four synthetic operons)

needs to covalently conjugate with BAP (BAP:operon conjugates). The

conjugates and E.coli lysates are then mixed with oil-surfactant to generate

individual compartments. The functional conjugates use desthiobiotin as

substrates and then attach desthiobiotin to BAP. Finally, the functional

conjugates are separated from conjugates by anti-His antibodies agarose

beads and recovered by PCR amplification.

Operons are functional:

Each operon contained one streptavidin gene, one biotin ligase gene, one T7

promoter, one T7 terminator, two separate ribosomal binding sites, and one unique

restriction site (Figure 4-3). The length of each operon was 2.1 kb. Before the selection,

four synthetic operons were tested for expression and activity in E.coli lysates (Figure 4-

108

5, western blot). Four operons were separately incubated with E.coli lysates at 30oC for 1

hour and then divided into two sets for gel electrophoresis and then western blot. One set

of samples was boiled and the other set of samples was not boiled before gel

electrophoresis (Figure 4-5). Lane1, 2, 3, and 4 are [SAwt-BirAwt], [SAwt-BirA6-40], [SA7-

6-BirAwt] and [SA7-6-BirA6-40], respectively and lane5, 6, 7, and 8 are aforementioned

operons with additional CAT-BAP (Chloramphenicol AcetyltTansferase with Biotin

Acceptor Peptide) DNA as substrates for biotin ligase. The western blot was stained

with alkaline phosphatase conjugated anti-His antibodies (aHis Ab-AP) (left part of

Figure 4-5) to detect synthesized His-tagged biotin ligases and streptavidin and then

further stained with alkaline phosphatase conjugated streptavidin (SA-AP) to detect

biotinylated CAT-BAP (right part of Figure 4-5). The expression of biotin ligases and

streptavidin of four operons were good (left part of Figure 4-5, lane1-4). Moreover, since

the protein samples were not boiled before gel electrophoresis the complex of

biotinylated CAT-BAP and streptavidin was not disrupted. That said, the complex

(shifted signal) was detected by aHis Ab-AP (left part of Figure4-5, lane 5-8). The

biotinylated CAT-BAP (not the complex) was differentiated from the shifted complex by

SA-AP (right part of Figure 4-5, lane 5-8). Taken together, these 4 synthetic operons

expressed and functioned well in the in vitro transcription and translation reaction.

109

Figure 4-5: Protein expression of synthetic operons (western blot). We tested 4

synthetic operons if they were all able to express functional protein in E.coli

lysates. (Left: stained by aHis Ab-AP) Lane1 to lane4 were [SAwt-BirAwt],

[SAwt-BirA6-40], [SA7-6-BirAwt] and [SA7-6-BirA6-40], respectively and all of

them expressed well. Lane5 to lane8 were aforementioned operons with

additional CAT-BAP (substrate of biotin ligase) and displayed different

pattern compared to lane1 to lane4. It indicated CAT-BAP was biotinylated

by biotin ligases and formed shifted complex with streptavidin. (Right:

stained with SA-AP) In addition, CAT-BAP (not complex) was biotinylated

by biotin ligase and was stained by SA-AP.

Mock selection:

In order to demonstrate the feasibility of selection scheme in emulsions, a mock

selection was performed. We generated three pools to test and they were three ratios

(0:1, 1:9 and 1:99) of mixing “non-functional” operon (no BAP-tagged) and “functional”

operon (BAP-tagged) (Figure 4-6). Without BAP-tagged, operon was not recovered in

the selection. Additionally, an Xho I restriction site was introduced into non-functional

operon. Thus it was able to differentiate from functional operon after selection. One

round of selection was performed with 20 uM of what present in the emulsions and we

110

found functional operon in 1:9 pool was enriched (Figure 4-6). This result proved the

selection worked. Although we did not find significant enrichment of functional operon

in the 1:99 pool it probably needed more rounds of selection to achieve it.

Figure 4-6: Mock selection. Non-functional and functional operons were mixed in three

different ratios (0:1, 1:9, and 1:99) for the mock selection. Non-functional

operons were not BAP-tagged and had an Xho restriction site. Thus it was

able to differentiate from functional operons in Xho I restriction enzyme

reaction. After one round of selection, we found functional operons in the

1:9 pool were enriched and proved the feasibility of selection scheme.

The selection of the representative pool (4 operons)

After mock selection, the real selection was implemented in emulsions with 50

uM desthiobiotin present. Four BAP:operon conjugates were added into in vitro

transcription and translation reagents (RTS E.coli HY kit, Roche) and then emulsified to

generate small vesicles with one conjugate / compartment. After one round of selection,

recovered operons were further digested with 4 restriction enzymes (Cla I, Sac I, Sal I or

Xho I) individually for differentiation (Figure 4-7). Comparing the signal intensity of

digested fragments, we were able to find which operon was enriched in the selection.

111

However, there was no enrichment in anyone of 4 operons in the selection (Figure 4-7)

and the result was consistent among several repeats.

Figure 4-7: One round of selection. Four operon conjugates were mixed in the same

amount and competed in the selection with 50 uM desthiobiotin present in

emulsions. After one round of selection, we expected to recover [SA7-6-

BirA6-40] more than the other three. However, there was no difference

among all synthetic operons.

Non-specific DNA binding of anti-His antibody agarose beads

As previously demonstrated, the BirA6-40 is a better enzyme in use of

desthiobiotin compared to BirAwt with 50 uM desthiobiotin present in the selection

(Chapter 3). In addition, the koff of SA7-6 is 50-fold better than SAwt. That indicates we

should have more SA7-6 binding on desthiobiotinylated operons than SAwt after washes

(Levy and Ellington 2008). In order to figure out why the selection did not work, we

dissected the selection into two steps. First, [SAwt-BirA6-40] and [SA7-6-BirA6-40] should

modify themselves more than [SAwt-BirAwt] and [SA7-6-BirAwt] with 50 uM desthiobiotin

present in the selection. Second, after pulling out by anti-His antibody agarose beads and

washes, SA7-6 should bind tighter to modified BAP:operon conjugates than SAwt in

washes. Taken together, we should see [SAwt-BirA6-40] and [SA7-6-BirA6-40] modify

112

themselves more than [SAwt-BirAwt] and [SA7-6-BirAwt] in the first step of selection.

Subsequently, we should see more [SA7-6-BirA6-40] than [SAwt-BirA6-40] after washes.

In order to check the first step, we did PCR cleanup after the selection and then

used streptavidin magnetic beads to capture functional conjugates. Captured conjugates

were then amplified by PCR. Thus, we avoided the interference of synthesized

streptavidin and anti-His antibody agarose beads in the selection. It was clear that we had

lower signals in background than recovered functional conjugates (Figure 4-8 B, lane1

and lane2). Moreover, we used four restriction enzymes to differentiate which operon

used desthiobiotin better than other operons (Figure 4-8 D). The result showed [SAwt-

BirA6-40] and [SA7-6-BirA6-40] used desthiobiotin better than [SAwt-BirAwt] and [SA7-6-

BirAwt]. Therefore, the first step of the selection was not a problem.

Besides using streptavidin magnetic beads, we also performed a selection using

anti-His antibody agarose beads to pull out functional conjugates. However, when we

performed PCR amplification to recover operons from the selection we found high

background signal (untagged operon) compared to BAP:operon conjugates (Figure 4-8 C,

lane5 and lane6). It was consistent across experiments. Since untagged operon should

not be recovered from the selection it was very possible that anti-His antibody agarose

beads had high non-specific DNA binding and covered real signal.

113

114

Figure 4-8: Dissection of the selection. (A) The selection had two steps. First,

synthesized biotin ligase modified BAP of conjugates with desthiobiotin and

then synthesized streptavidin bound to modified conjugates in the

compartment. Second, anti-His antibodies agarose beads were used to pull

out the complex (modified conjugates with streptavidin). And then we

recovered operon by PCR amplification. BirA6-40 has better activity in use

of desthiobiotin and SA7-6 has better off rate in binding to desthiobiotin

compared to their corresponding wild-type protein. Therefore, we should

pull out [SA7-6-BirA6-40] after one round of selection with 50 uM

desthiobiotin present. However, we did not have more [SA7-6-BirA6-40] back

in the selection. (B and C) After breaking emulsions we used streptavidin

magnetic beads or anti-His antibodies agarose beads to pull out modified

conjugates. Then we recovered conjugates with PCR amplification.

Streptavidin magnetic beads avoided the interference of synthesized

streptavidin and anti-His antibody agarose beads in the selection. (B,

streptavidin magnetic beads) The signal of recovered functional operons in

PCR amplification (lane 1) was lower than background (lane2). (C, anti-His

Ab agarose beads) However, the background signal was stronger than

recovered functional conjugates. (D) Further analysis of the recovered

conjugates (B, lane1) using restriction enzymes, we found we had more

[SAwt-BirA6-40] and [SA7-6-BirA6-40] back than the other two at the first step

of selection. It indicated modification of conjugates was not a problem.

The problem should be the non-specific DNA binding to anti-His Ab

agarose beads.

Therefore, we tried different wash buffers and capture approaches to avoid non-

specific DNA binding. The original wash buffer was TBS with Tween 20 (TBST) which

was used in streptavidin selection by Levy (100 mM Tris-HCl [pH7.4], 150 mM NaCl,

0.1% Tween 20) (Levy and Ellington 2008). Therefore, we tested wash buffers with

different salt concentration (150 mM, 300 mM, 600 mM and 1 M NaCl) and there was no

significant improvement in non-specific DNA binding of anti-His Ab agarose beads.

Also, we tried longer wash time, larger volume, more washes and there were no

significant improvement. But blocking anti-His Ab agarose beads with salmon testes

DNA (Sigma) offered minor improvements. The new wash buffer was 100 mM Tri-HCl,

[pH7.4], 150 mM NaCl, 0.1% Tween 20 and washed resin 7 times with 1.2 ml buffer.

115

A selection was performed with 50 uM desthiobiotin present and used pre-

blocked anti-His Ab agarose beads (by salmon testes DNA) for capture. Also, new wash

buffer was applied for washes (100 mM Tri-HCl, [pH7.4], 150 mM NaCl, 0.1% Tween

20 and washed resin 7 times with 1.2 ml buffer). The non-specific binding was reduced

(Figure 4-9, lane 4 and lane 5). We further analyzed the selection (Figure 4-9, lane 5) by

restriction enzymes but still did not see any significant enrichment in any operons. It

indicated that the non-specific DNA binding was still too high compared to enriched

signal.

Figure 4-9: One round of selection with new wash buffer. (top figure) After one

round of real selection it was clear that we lowered the background with

compared to recovered operons (lane4 and lane5). (bottom figure)

However, we still did not see any difference among 4 operons after one

round of real selection.

116

Besides the aforementioned tests, we also tested two beads to avoid non-specific

DNA binding. They were anti-His Ab resin (R&D systems) and cobalt-immobilized

metal affinity chromatography (IMAC) magnetic beads (Invitrogen). Three of them

(including original anti-His Ab from Sigma) were tested (but not in emulsions and also

had no free biotin in the reaction) for non-specific binding to beads (Figure 4-10). The

result showed the original beads were better than others in recovery. By comparing the

signal of untagged operon (lane 1) and biotin-tagged operon (lane 2), the anti-His Ab

agarose beads may be the best one so far.

Also, it should be noted the recovery of biotin tagged operon in this test was

better than in the selection because all operons were biotin-tagged. Furthermore, there

was no free biotin in the reaction. Therefore, there was no competition to binding sites of

synthesized streptavidin and it improved the recovery of biotin-tagged operon in this test.

Figure 4-10:Different capture approach. We suspected the problem was non-specific

DNA binding to agarose beads so we compared different anti-His antibodies

(Sigma or R&D) and Cobalt-IMAC magnetic beads to capture complex

(streptavidin with modified operon). However, the original anti-His

antibodies was the best one compared to others.

117

In fact, besides the nonspecific binding to beads, there were several factors that

also affected the selection. First, the operon (2.1 kb) is longer than the linear construct

(0.7 kb) used in Levy’s work and it has been found the short DNA has less non-specific

binding to beads compared to 2.1 kb linear construct. Second, free desthiobiotin in the

selection also affects the recovery of modified operon because it competes binding sites

of streptavidin. Higher recovery would help to differentiate from background (non-

specific binding). Third, the expression of both biotin ligase and streptavidin in an

emulsion vesicle is not as high as only expression of biotin ligase or streptavidin in an

emulsion vesicle. It would reduce the modification of BAP-tagged operon (by biotin

ligase) and recovery of modified operon (by streptavidin).

Taken together, non-specific binding to beads and the competition effect of free

desthiobiotin are likely the two major factors. Since we would not be able to reduce the

size of operons the best way is to find a better capture approach for the modified operon.

Optimize the use of Co-IMAC magnetic beads would be an option because we already

knew magnetic beads cause much lower background than agarose beads. In the

experiment of “evolving of biotin ligase for desthiobiotin”, the linear construct was 1.3

kb and did not show significant non-specific binding to magnetic beads. Thus the Co-

IMAC magnetic beads might a potential candidate. In addition, the free desthiobiotin in

the selection is 50 uM which is around 1.5 x 105 molecules of free desthiobiotin in an

emulsion vesicle. Although we don’t know the concentration of synthesized streptavidin

in an emulsion vesicle it would be better to reduce free desthiobiotin for the competition

of binding sites. However, reducing desthiobiotin concentration in the reaction may also

affect the activity of biotin ligase. Therefore, these two factors influence the selection

conversely and would be difficult to tune. Although they affect the selection in opposite

ways, the concentration of desthiobiotin in modification reaction is not a limiting step but

118

the concentration of BAP (one BAP-tagged operon in one compartment: 0.3 nM).

Therefore, we would be able to reduce the concentration of free desthiobiotin without

influence the modification reaction and help to recover more modified operons from the

selection.

119

REFERENCES

Chapman-Smith, A., and Cronan, J.E., Jr. (1999). The enzymatic biotinylation of

proteins: a post-translational modification of exceptional specificity. Trends

Biochem Sci 24, 359-363.

Davidson, E.A., Dlugosz, P.J., Levy, M., and Ellington, A.D. (2009). Directed evolution

of proteins in vitro using compartmentalization in emulsions. Curr Protoc Mol

Biol Chapter 24, Unit 24 26.

Davidson, E.A., Meyer, A.J., Ellefson, J.W., Levy, M., and Ellington, A.D. (2012). An in

vitro autogene. ACS Synth Biol 1, 190-196.

Griffiths, A.D., and Tawfik, D.S. (2003). Directed evolution of an extremely fast

phosphotriesterase by in vitro compartmentalization. EMBO J 22, 24-35.

Juers, D.H., Matthews, B.W., and Huber, R.E. (2012). LacZ beta-galactosidase: structure

and function of an enzyme of historical and molecular biological importance.

Protein Sci 21, 1792-1807.

Levy, M., and Ellington, A.D. (2008). Directed evolution of streptavidin variants using in

vitro compartmentalization. Chem Biol 15, 979-989.

Magalhaes, M.L., Czekster, C.M., Guan, R., Malashkevich, V.N., Almo, S.C., and Levy,

M. (2011). Evolved streptavidin mutants reveal key role of loop residue in high-

affinity binding. Protein Sci 20, 1145-1154.

Miller, O.J., Bernath, K., Agresti, J.J., Amitai, G., Kelly, B.T., Mastrobattista, E., Taly,

V., Magdassi, S., Tawfik, D.S., and Griffiths, A.D. (2006). Directed evolution by

in vitro compartmentalization. Nat Methods 3, 561-570.

Nah, B.N.L. (2009). High-throughput assays for biotin protein ligase: A novel antibiotic

target. Master's Thesis, 1-145.

Roderick, S.L. (2005). The lac operon galactoside acetyltransferase. C R Biol 328, 568-

575.

Schatz, P.J. (1993). Use of peptide libraries to map the substrate specificity of a peptide-

modifying enzyme: a 13 residue consensus peptide specifies biotinylation in

Escherichia coli. Biotechnology (N Y) 11, 1138-1143.

Wilson, C.J., Zhan, H., Swint-Kruse, L., and Matthews, K.S. (2007). The lactose

repressor system: paradigms for regulation, allosteric behavior and protein

folding. Cell Mol Life Sci 64, 3-16.

120

Bibliography

Agresti, J.J., Kelly, B.T., Jaschke, A., and Griffiths, A.D. (2005). Selection of ribozymes

that catalyse multiple-turnover Diels-Alder cycloadditions by using in vitro

compartmentalization. Proc Natl Acad Sci U S A 102, 16170-16175.

Amitai, G., Gupta, R.D., and Tawfik, D.S. (2007). Latent evolutionary potentials under

the neutral mutational drift of an enzyme. HFSP J 1, 67-78.

Anderson, M.E. (1998). Glutathione: an overview of biosynthesis and modulation. Chem

Biol Interact 111-112, 1-14.

Bernath, K., Hai, M., Mastrobattista, E., Griffiths, A.D., Magdassi, S., and Tawfik, D.S.

(2004). In vitro compartmentalization by double emulsions: sorting and gene

enrichment by fluorescence activated cell sorting. Anal Biochem 325, 151-157.

Bertschinger, J., and Neri, D. (2004). Covalent DNA display as a novel tool for directed

evolution of proteins in vitro. Protein Eng Des Sel 17, 699-707.

Bochkov, Y.A., and Palmenberg, A.C. (2006). Translational efficiency of EMCV IRES

in bicistronic vectors is dependent upon IRES sequence and gene location.

Biotechniques 41, 283-284, 286, 288 passim.

Braunagel, M., and Little, M. (1997). Construction of a semisynthetic antibody library

using trinucleotide oligos. Nucleic Acids Res 25, 4690-4691.

Cellarier, E., Durando, X., Vasson, M.P., Farges, M.C., Demiden, A., Maurizis, J.C.,

Madelmont, J.C., and Chollet, P. (2003). Methionine dependency and cancer

treatment. Cancer Treat Rev 29, 489-499.

Chapman-Smith, A., and Cronan, J.E., Jr. (1999). The enzymatic biotinylation of

proteins: a post-translational modification of exceptional specificity. Trends

Biochem Sci 24, 359-363.

Chapman-Smith, A., Mulhern, T.D., Whelan, F., Cronan, J.E., Jr., and Wallace, J.C.

(2001). The C-terminal domain of biotin protein ligase from E. coli is required for

catalytic activity. Protein Sci 10, 2608-2617.

Chen, I., Choi, Y.A., and Ting, A.Y. (2007). Phage display evolution of a peptide

substrate for yeast biotin ligase and application to two-color quantum dot labeling

of cell surface proteins. J Am Chem Soc 129, 6619-6625.

Chen, Y., Mandic, J., and Varani, G. (2008). Cell-free selection of RNA-binding proteins

using in vitro compartmentalization. Nucleic Acids Res 36, e128.

Choi-Rhee, E., Schulman, H., and Cronan, J.E. (2004). Promiscuous protein biotinylation

by Escherichia coli biotin protein ligase. Protein Sci 13, 3043-3050.

121

Cronan, J.E. (2005). Targeted and proximity-dependent promiscuous protein

biotinylation by a mutant Escherichia coli biotin protein ligase. J Nutr Biochem

16, 416-418.

Davidson, E.A., Dlugosz, P.J., Levy, M., and Ellington, A.D. (2009). Directed evolution

of proteins in vitro using compartmentalization in emulsions. Curr Protoc Mol

Biol Chapter 24, Unit 24 26.

Davidson, E.A., Meyer, A.J., Ellefson, J.W., Levy, M., and Ellington, A.D. (2012). An in

vitro autogene. ACS Synth Biol 1, 190-196.

Doi, N., Kumadaki, S., Oishi, Y., Matsumura, N., and Yanagawa, H. (2004). In vitro

selection of restriction endonucleases by in vitro compartmentalization. Nucleic

Acids Res 32, e95.

Doi, N., and Yanagawa, H. (1999). STABLE: protein-DNA fusion system for screening

of combinatorial protein libraries in vitro. FEBS Lett 457, 227-230.

Fen, C.X., Coomber, D.W., Lane, D.P., and Ghadessy, F.J. (2007). Directed evolution of

p53 variants with altered DNA-binding specificities by in vitro

compartmentalization. J Mol Biol 371, 1238-1248.

Fernandez-Gacio, A., Uguen, M., and Fastrez, J. (2003). Phage display as a tool for the

directed evolution of enzymes. Trends Biotechnol 21, 408-414.

Fromant, M., Blanquet, S., and Plateau, P. (1995). Direct random mutagenesis of gene-

sized DNA fragments using polymerase chain reaction. Anal Biochem 224, 347-

353.

Ghadessy, F.J., and Holliger, P. (2004). A novel emulsion mixture for in vitro

compartmentalization of transcription and translation in the rabbit reticulocyte

system. Protein Eng Des Sel 17, 201-204.

Ghadessy, F.J., Ong, J.L., and Holliger, P. (2001). Directed evolution of polymerase

function by compartmentalized self-replication. Proc Natl Acad Sci U S A 98,

4552-4557.

Ghadessy, F.J., Ramsay, N., Boudsocq, F., Loakes, D., Brown, A., Iwai, S., Vaisman, A.,

Woodgate, R., and Holliger, P. (2004). Generic expansion of the substrate

spectrum of a DNA polymerase by directed evolution. Nat Biotechnol 22, 755-

759.

Griffiths, A.D., and Tawfik, D.S. (2003). Directed evolution of an extremely fast

phosphotriesterase by in vitro compartmentalization. EMBO J 22, 24-35.

Gupta, R.D., Goldsmith, M., Ashani, Y., Simo, Y., Mullokandov, G., Bar, H., Ben-David,

M., Leader, H., Margalit, R., Silman, I., et al. (2011). Directed evolution of

hydrolases for prevention of G-type nerve agent intoxication. Nat Chem Biol 7,

120-125.

122

Gupta, R.D., and Tawfik, D.S. (2008). Directed enzyme evolution via small and effective

neutral drift libraries. Nat Methods 5, 939-942.

Hanes, J., and Pluckthun, A. (1997). In vitro selection and evolution of functional

proteins by using ribosome display. Proc Natl Acad Sci U S A 94, 4937-4942.

Juers, D.H., Matthews, B.W., and Huber, R.E. (2012). LacZ beta-galactosidase: structure

and function of an enzyme of historical and molecular biological importance.

Protein Sci 21, 1792-1807.

Kelly, B.T., and Griffiths, A.D. (2007). Selective gene amplification. Protein Eng Des Sel

20, 577-581.

Khersonsky, O., and Tawfik, D.S. (2010). Enzyme promiscuity: a mechanistic and

evolutionary perspective. Annu Rev Biochem 79, 471-505.

Kintses, B., van Vliet, L.D., Devenish, S.R., and Hollfelder, F. (2010). Microfluidic

droplets: new integrated workflows for biological experiments. Curr Opin Chem

Biol 14, 548-555.

Kitagawa, M., Ara, T., Arifuzzaman, M., Ioka-Nakamichi, T., Inamoto, E., Toyonaga, H.,

and Mori, H. (2005). Complete set of ORF clones of Escherichia coli ASKA

library (a complete set of E. coli K-12 ORF archive): unique resources for

biological research. DNA Res 12, 291-299.

Kroemer, G., and Pouyssegur, J. (2008). Tumor cell metabolism: cancer's Achilles' heel.

Cancer Cell 13, 472-482.

Levy, M., and Ellington, A.D. (2008). Directed evolution of streptavidin variants using in

vitro compartmentalization. Chem Biol 15, 979-989.

Levy, M., Griswold, K.E., and Ellington, A.D. (2005). Direct selection of trans-acting

ligase ribozymes by in vitro compartmentalization. RNA 11, 1555-1562.

Lipovsek, D., and Pluckthun, A. (2004). In-vitro protein evolution by ribosome display

and mRNA display. J Immunol Methods 290, 51-67.

Loakes, D., Gallego, J., Pinheiro, V.B., Kool, E.T., and Holliger, P. (2009). Evolving a

polymerase for hydrophobic base analogues. J Am Chem Soc 131, 14827-14837.

Magalhaes, M.L., Czekster, C.M., Guan, R., Malashkevich, V.N., Almo, S.C., and Levy,

M. (2011). Evolved streptavidin mutants reveal key role of loop residue in high-

affinity binding. Protein Sci 20, 1145-1154.

Margulies, M., Egholm, M., Altman, W.E., Attiya, S., Bader, J.S., Bemben, L.A., Berka,

J., Braverman, M.S., Chen, Y.J., Chen, Z., et al. (2005). Genome sequencing in

microfabricated high-density picolitre reactors. Nature 437, 376-380.

Mastrobattista, E., Taly, V., Chanudet, E., Treacy, P., Kelly, B.T., and Griffiths, A.D.

(2005). High-throughput screening of enzyme libraries: in vitro evolution of a

123

beta-galactosidase by fluorescence-activated sorting of double emulsions. Chem

Biol 12, 1291-1300.

Miller, O.J., Bernath, K., Agresti, J.J., Amitai, G., Kelly, B.T., Mastrobattista, E., Taly,

V., Magdassi, S., Tawfik, D.S., and Griffiths, A.D. (2006). Directed evolution by

in vitro compartmentalization. Nat Methods 3, 561-570.

Naganathan, S., and Beckett, D. (2007). Nucleation of an allosteric response via ligand-

induced loop folding. J Mol Biol 373, 96-111.

Nah, B.N.L. (2009). High-throughput assays for biotin protein ligase: A novel antibiotic

target. Master's Thesis, 1-145.

Ong, J.L., Loakes, D., Jaroslawski, S., Too, K., and Holliger, P. (2006). Directed

evolution of DNA polymerase, RNA polymerase and reverse transcriptase activity

in a single polypeptide. J Mol Biol 361, 537-550.

Parent, K.N., Ranaghan, M.J., and Teschke, C.M. (2004). A second-site suppressor of a

folding defect functions via interactions with a chaperone network to improve

folding and assembly in vivo. Mol Microbiol 54, 1036-1050.

Pavillard, V., Nicolaou, A., Double, J.A., and Phillips, R.M. (2006). Methionine

dependence of tumours: a biochemical strategy for optimizing paclitaxel

chemosensitivity in vitro. Biochem Pharmacol 71, 772-778.

Ramsay, N., Jemth, A.S., Brown, A., Crampton, N., Dear, P., and Holliger, P. (2010).

CyDNA: synthesis and replication of highly Cy-dye substituted DNA by an

evolved polymerase. J Am Chem Soc 132, 5096-5104.

Reiersen, H., Lobersli, I., Loset, G.A., Hvattum, E., Simonsen, B., Stacy, J.E., McGregor,

D., Fitzgerald, K., Welschof, M., Brekke, O.H., et al. (2005). Covalent antibody

display--an in vitro antibody-DNA library selection system. Nucleic Acids Res

33, e10.

Rhee, S.Y., Liu, T.F., Holmes, S.P., and Shafer, R.W. (2007). HIV-1 subtype B protease

and reverse transcriptase amino acid covariation. PLoS Comput Biol 3, e87.

Roberts, R.W., and Szostak, J.W. (1997). RNA-peptide fusions for the in vitro selection

of peptides and proteins. Proc Natl Acad Sci U S A 94, 12297-12302.

Roderick, S.L. (2005). The lac operon galactoside acetyltransferase. C R Biol 328, 568-

575.

Schatz, P.J. (1993). Use of peptide libraries to map the substrate specificity of a peptide-

modifying enzyme: a 13 residue consensus peptide specifies biotinylation in

Escherichia coli. Biotechnology (N Y) 11, 1138-1143.

Sepp, A., and Choo, Y. (2005). Cell-free selection of zinc finger DNA-binding proteins

using in vitro compartmentalization. J Mol Biol 354, 212-219.

124

Shimizu, Y., Kanamori, T., and Ueda, T. (2005). Protein synthesis by pure translation

systems. Methods 36, 299-304.

Sitaraman, K., Esposito, D., Klarmann, G., Le Grice, S.F., Hartley, J.L., and Chatterjee,

D.K. (2004). A novel cell-free protein synthesis system. J Biotechnol 110, 257-

263.

Slavoff, S.A., Chen, I., Choi, Y.A., and Ting, A.Y. (2008). Expanding the substrate

tolerance of biotin ligase through exploration of enzymes from diverse species. J

Am Chem Soc 130, 1160-1162.

Stapleton, J.A., and Swartz, J.R. (2010). Development of an in vitro

compartmentalization screen for high-throughput directed evolution of [FeFe]

hydrogenases. PLoS One 5, e15275.

Stein, V., Sielaff, I., Johnsson, K., and Hollfelder, F. (2007). A covalent chemical

genotype-phenotype linkage for in vitro protein evolution. Chembiochem 8, 2191-

2194.

Stone, E., Paley, O., Hu, J., Ekerdt, B., Cheung, N.K., and Georgiou, G. (2012). De novo

engineering of a human cystathionine-gamma-lyase for systemic (L)-Methionine

depletion cancer therapy. ACS Chem Biol 7, 1822-1829.

Takagi, M., Nishioka, M., Kakihara, H., Kitabayashi, M., Inoue, H., Kawakami, B., Oka,

M., and Imanaka, T. (1997). Characterization of DNA polymerase from

Pyrococcus sp. strain KOD1 and its application to PCR. Appl Environ Microbiol

63, 4504-4510.

Takei, Y., Kadomatsu, K., Itoh, H., Sato, W., Nakazawa, K., Kubota, S., and Muramatsu,

T. (2002). 5'-,3'-inverted thymidine-modified antisense oligodeoxynucleotide

targeting midkine. Its design and application for cancer therapy. J Biol Chem 277,

23800-23806.

Tawfik, D.S., and Griffiths, A.D. (1998). Man-made cell-like compartments for

molecular evolution. Nat Biotechnol 16, 652-656.

Thivat, E., Farges, M.C., Bacin, F., D'Incan, M., Mouret-Reynier, M.A., Cellarier, E.,

Madelmont, J.C., Vasson, M.P., Chollet, P., and Durando, X. (2009). Phase II trial

of the association of a methionine-free diet with cystemustine therapy in

melanoma and glioma. Anticancer Res 29, 5235-5240.

Thomas, T., and Thomas, T.J. (2001). Polyamines in cell growth and cell death:

molecular mechanisms and therapeutic applications. Cell Mol Life Sci 58, 244-

258.

Tokuriki, N., and Tawfik, D.S. (2009). Chaperonin overexpression promotes genetic

variation and enzyme evolution. Nature 459, 668-673.

Ullman, C.G., Frigotto, L., and Cooley, R.N. (2011). In vitro methods for peptide display

and their applications. Brief Funct Genomics 10, 125-134.

125

Utada, A.S., Lorenceau, E., Link, D.R., Kaplan, P.D., Stone, H.A., and Weitz, D.A.

(2005). Monodisperse double emulsions generated from a microcapillary device.

Science 308, 537-541.

Uttamapinant, C., White, K.A., Baruah, H., Thompson, S., Fernandez-Suarez, M.,

Puthenveetil, S., and Ting, A.Y. (2010). A fluorophore ligase for site-specific

protein labeling inside living cells. Proc Natl Acad Sci U S A 107, 10914-10919.

Williams, R., Peisajovich, S.G., Miller, O.J., Magdassi, S., Tawfik, D.S., and Griffiths,

A.D. (2006). Amplification of complex gene libraries by emulsion PCR. Nat

Methods 3, 545-550.

Wilson, C.J., Zhan, H., Swint-Kruse, L., and Matthews, K.S. (2007). The lactose

repressor system: paradigms for regulation, allosteric behavior and protein

folding. Cell Mol Life Sci 64, 3-16.

Yonezawa, M., Doi, N., Kawahashi, Y., Higashinakagawa, T., and Yanagawa, H. (2003).

DNA display for in vitro selection of diverse peptide libraries. Nucleic Acids Res

31, e118.

Zahnd, C., Amstutz, P., and Pluckthun, A. (2007). Ribosome display: selecting and

evolving proteins in vitro that specifically bind to a target. Nat Methods 4, 269-

279.

126

Vita

Wei-Cheng Lu was born in Taipei, Taiwan. He entered the Tzu-Chi University

and received his B.S. degree in Laboratory Medicine and Biotechnology. After

mandatory military service he then started work for Institute of Biomedical Sciences,

Academia Sinica at Taiwan. He entered the Cell and Molecular Biology graduate

program at the University of Texas at Austin, under the supervision of Dr. Andrew D.

Ellington. His publications include:

Lu, W. C., Liu, Y. N., Kang, B. B., Chen, J. H. Trans-activation of heparanase promoter

by ETS transcription factor. Oncogene 22: 919-923, 2003

Ru, H. Y., Chen, R. L., Lu, W. C., Chen, J. H. hBUB1 defects in leukemia and

lymphoma cells. Oncogene 21: 4673-4679, 2002

KC Chen, CH Wu, CY Chang, WC Lu, Q Tseng, ZM Prijovich, W Schechinger, YC

Liaw, YL Leu and SR Roffler. Directed evolution of a lysosomal enzyme with

enhanced activity at neutral pH by mammalian cell-surface display. Chem Biol.

15:1277-86, 2008

CP Chen, YT Hsieh, ZM Prijovich, HY Chuang, KC Chen, WC Lu, Q Tseng, YL Leu,

TL Cheng, SR Roffler. ECSTASY, an adjustable membrane-tethered/soluble

protein expression system for the directed evolution of mammalian proteins

Protein Eng Des Sel., 25:367-375, 2012

Lu WC, Ellington AD. In vitro selection of proteins via emulsion compartments.

Methods. 15;60(1):75-80, 2013

Email: [email protected]

This dissertation was typed by the author.