briefings in functional genomics and proteomics-2002-carmen-189-203

Sara Carmen

is a senior scientist and

Lutz Jermutus

is the head of the technology

development team within the

display technology group at

CAT (Cambridge Antibody

Technology). The team is

dedicated to exploring

innovative solutions in library

design and selection strategies,

as well as phage and ribosome

display.

Keywords: phage display,antibody, library, valency,selection

Sara Carmen,

Display Technology Development,

Cambridge Antibody Technology,

The Science Park,

Melbourn,

Cambridgeshire,

SG8 6JJ, UK

Tel: +44(0)1763 269 284

Fax: +44 (0)1763 263 413

E-mail: sara.carmen@

cambridgeantibody.com

Concepts in antibodyphage displaySara Carmen and Lutz JermutusDate received (in revised form): 15th April 2002

AbstractThis paper introduces the reader to antibody phage display and its use in combinatorial

biochemistry. The focus is on overviewing phage display formats, library design and selection

technology, which are the prerequisites for the successful isolation of specific antibody

fragments against a diverse set of target antigens.

Abbreviations

CDR, complementarity determining region N1, first N-terminal domain of g3p

VH, variable heavy chain N2, second N-terminal domain of g3p

VL, variable light chain CT, C-terminal domain of g3p

scFv, single chain variable fragment IG, intergenic region

g3p, gene 3 protein RT, reverse transcription

INTRODUCTION TOPHAGE DISPLAYAnimal immunisation followed by

hybridoma technology has been used to

generate monoclonal antibodies against a

variety of antigens. Over the past ten

years, advances in molecular biology have

allowed for the use of Escherichia coli to

produce recombinant antibodies. By

restricting the size to either a Fab, a Fv or

a linker-stabilised single chain Fv (scFv)

(Figure 1) such antibody fragments can

not only be expressed in bacterial cells but

also displayed by fusion to phage coat

proteins.1

The phage display concept was first

introduced for short peptide fragments in

1985.2 Fragments of the EcoRI

endonuclease, displayed as a polypeptide

fusion (phenotype) to the gene 3 protein

(g3p), were encoded on the DNA

molecule (genotype) encapsulated within

the phage particle. The linkage of

genotype to phenotype is the fundamental

aspect of phage display. Subsequently,

phage display of functional antibody

fragments was shown3 when the VH and

VL fragments of the anti-lysozyme

antibody (D1.3) were introduced, with a

linker, into a phage vector at the N-

terminus of g3p. Since then, large scFv,

Fab and peptide repertoires have been

generated using a variety of phage display

formats.

One of the major advantages of phage

display technology of antibody fragments

compared with standard hybridoma

technology is that the generation of

specific scFv/Fab fragments to a particular

antigen can be performed within a couple

of weeks. The starting point is usually an

antibody library, of either naive or

immune origin, comprising a population

of, ideally, 109 –1011 clones. After usually

two to, maximally, three rounds of

selection, the population is enriched for a

high percentage of antibody fragments

specific for the target antigen. The display

of human antibody fragments is of

particular interest since any therapeutic

derived from these antibody fragments is

believed to have a minimal risk of an

immune response in patients.

Both scFv and Fab formats have been

used successfully in antibody libraries

displayed on phage. Fab fragments are

usually more stable than Fv fragments and

have less potential to dimerise than scFvs.4

In addition to the VH and VL segments,

they also possess the constant regions (CH

and CL) of the heavy and light chains.

They are reputed to be displayed at lower

& HENRY STEWART PUBLICATIONS 1473-9550. B R I E F I N G S I N F U N C T I O N A L G E N O M I C S A N D P R O T E O M I C S . VOL 1. NO 2. 189–203. JULY 2002 1 8 9

by guest on January 15, 2011bfg.oxfordjournals.org

Dow

nloaded from

http://bfg.oxfordjournals.org/

frequency on the surface of phage, thus

alleviating the avidity effects associated

with some well-expressed scFvs.

Synthetic and naive Fab libraries have

been used successfully for the generation

of antibody fragments to a variety of

antigens.5–8 Expression of the scFv

fragment has a less toxic effect on the cell

than the larger Fab molecules.4 This

results in a better yield and diversity in

scFv libraries. The remainder of this paper

will, therefore, focus primarily on scFv

libraries since they are generally the more

popular choice.9

Expression in E. coli ensures that

sufficient quantities of scFv, for screening

and characterisation, can be produced

with relative ease. Many secreted

eukaryotic proteins such as antibodies

require disulphide bonds for stability, and

the oxidising environment of the E. coli

periplasm, where filamentous phage

assembles, provides the appropriate

conditions for antibody folding. Human

antibodies against human proteins can be

isolated from diverse human antibody

libraries. Moreover, antibodies to toxic

molecules such as doxorubicin10 can be

obtained, a task difficult with

immunisation/hybridoma techniques.

M13 phage biologyM13 is a filamentous bacteriophage. The

native particle is a thin, cylindrical shape,

usually 900 nm long and 6–7 nm in

diameter.11,12 It contains a single-stranded

DNA genome (6,407 base pairs in length)

which encodes 11 genes,13 five of which

are coat proteins. The major coat protein

is the gene 8 protein (g8p) which is

present in almost 2,700 copies and

responsible for encapsulating the phage

DNA (Figure 2). The distal end of the

phage particle is capped by five copies

each of g7p and g9p. At the proximal end,

four to five copies each of g6p and g3p

are present. M13 bacteriophage infects

only male bacteria, ie those E. coli cells

that bear the F-plasmid which encodes

the F-pilus. Infection is mediated by the

interaction between g3p of the phage and

the F-pilus. Filamentous phage have the

characteristic that once they have infected

their host cell they do not lyse the cell,

but instead are able to replicate and are

released from the cell membrane while

the host cell continues to grow and

divide, in contrast to the lytic phages T4

and T7.

The structure and function of g3p have

been well studied. The protein is

responsible for phage infection and for

release of the phage particle following

assembly.14,15 It has two N-terminal

domains (N1 and N2) and a C-terminal

domain (CT, Figure 3). N1, which has

been shown to be essential for

infectivity,16 interacts with the TolA

protein of the TolQRA complex located

across the periplasmic space between the

inner and outer E. coli membranes. The

primary interaction of phage with an

E. coli cell is mediated by N2 binding the

F-pilus and it is this association that is

thought to bring N1 into close proximity

with TolA. It is not yet understood how

TolA contributes to phage infection;

however, it is known that the F-pilus

retracts and that the M13 genome is

injected into the cytoplasm. CT is

required for the termination of phage

assembly in the periplasm and release of

the phage from the cell membrane. Once

the cells have been infected and phage

protein production commences, the cells

are no longer able to be infected. The

presence of only very small amounts of

the g3p of f1 filamentous phage has been

shown to be associated with a resistance of

the cell to infection from filamentous

The oxidising periplasmof E. coli favoursdisulphide bondformation

Filamentous phageassemblies in the E. Coliperiplasm

Figure 1: Antibody structure and derivedfragments. (A) IgG, 160 kilodaltons (kD); (B)Fab fragment, 50 kD; (C) Fv fragment, 30 kD;(D) scFv with linker, 30 kDa

A. B. C. D.

1 9 0 & HENRY STEWART PUBLICATIONS 1473-9550. B R I E F I N G S I N F U N C T I O N A L G E N O M I C S A N D P R O T E O M I C S . VOL 1. NO 2. 189–203. JULY 2002

Carmen and Jermutus


Dow

nloaded from


phage. Its presence in the E. coli

membrane disrupts the membrane

integrity, causing a number of effects

including defective F-pili.17

The five coat proteins have all been

used as fusion proteins for phage

display.18,19 Genomic or cDNA libraries

are favoured as C-terminal fusions to g8p

since expression constraints due to

frameshifts and stop codons are avoided.20

C-terminal polypeptide fusions to g3p

have also been demonstrated.21 Since g3p

is the most popular fusion protein, this

paper will concentrate on display of

antibody fragments on g3p in a phagemid

system.

PHAGE DISPLAY FORMATSEarly phage display formats involved the

fusion of peptides to the N-terminus of

g3p or g8p in the M13 phage vector. The

polypeptide or antibody fragment was

displayed in a multivalent format, since all

copies of the coat protein are translated as

fusion proteins,22,23 although it was not

All coat proteins can beused for phage display

Figure 3: The modular structure of g3p. The N1 domain interacts withthe TolA protein in the E. coli membrane. The N2 domain binds the F-pilus and the CT is necessary for termination of phage assembly. Theglycine-rich linkers, G1 and G2, are thought to confer flexibility to thedomains, facilitating the infection process. Antibody fragments can bedisplayed at the N-terminus of g3p. Alternatively, they can be displayed asan N-terminal fusion to CT

Figure 2: The structure of M13 phage particle. The phage particle is of cylindrical shape (900 nm long) and only 6–7 nm indiameter. Non-structural proteins are unshaded. All coat proteins are shaded according to their corresponding gene in thegenome. The intergenic regions are marked by IG. G8p is the major coat protein and is present in about 2,700 copies. Theminor coat proteins g3p, g6p, g7p and g9p are present in approximately five copies each

III

VIII

VII

IV

IX

X

I

XI

II VIG

6.4 kb

ori

IG

VI


Concepts in antibody phage display


Dow

nloaded from


possible to display larger polypeptides or

proteins without affecting the function of

g8p. Six residue insertions are usually

tolerated, but, when the insert is increased

to eight residues, only 40 per cent of

phage are infectious and, at 16 residues,

this number drops to less than one per

cent.24 Larger fusion proteins to g3p are

tolerated, but, in a multivalent display

format, such fusions may cause a decrease

in infectivity2 by sterically hindering the

interaction of N2 and the F-pilus. These

problems are overcome by the use of

phagemids.

Phagemids are plasmids

(�4.6 kilobases) which encode a signal

sequence, the phage coat protein and an

antibiotic resistance marker. The antibody

fragment/polypeptide is cloned upstream

of the g3p/g8p coat protein sequence and

expression is controlled by the use of a

promoter such as lacZ. The relatively

small size of these vectors means that they

have higher transformation efficiencies

than phage vectors,25 hence facilitating

the construction of large repertoires or

libraries of peptide or antibody fragments.

The incorporation of an amber stop

codon between the displayed protein and

the phage coat protein permits fusion

protein expression in suppressor strains of

E. coli such as TG1.26 Non-suppressor

strains, such as HB2151,27 will not

incorporate a glutamine at the amber

codon, thereby resulting in production of

only the antibody/polypeptide moiety.

A phagemid cannot produce infective

phage particles alone. A helper phage such

as M13KO7 or VCSM13 is required. The

helper phage provides the genes which

are essential for phage replication and

assembly, including a wild-type copy of

the coat protein used for display. Cells

already containing the phagemid vector

are superinfected with the helper phage.

Glucose in the growth media represses the

lacZ promoter, preventing expression of

g3p-fusion, which would inhibit

superinfection. Once the helper phage

genome is incorporated into the cell, the

glucose is removed and phage production

commences. The M13KO7 genome

possesses a modified intergenic region

(IG),28 causing it to be replicated and

packaged less efficiently than the

phagemid which carries the wild-type

M13 IG. This ensures that the genotype

(phagemid) and phenotype (g3p–scFv

fusion) are linked in a single (phage)

molecule, which is the key feature of

phage display (Figure 4). The antibody

fragment can also be fused to a truncated

Phagemid vectorsfacilitate construction oflarge libraries

Figure 4: Sequence of events depicting phage infection, assembly and secretion using a phagemid system. (1) Phagedisplaying scFv infects cell. Single-stranded phagemid DNA is injected into cytoplasm. (2) Helper phage superinfectproviding genomic information encoding remaining phage proteins for phage replication and assembly. (3) ScFv–g3p fusion,from the phagemid vector, and all other structural proteins (including wild-type g3p) from the helper phage vector, aredirected to the periplasm. (4) Phage assembly occurs in the E. coli periplasm. Phage containing DNA encoding scFvsequence (genotype) and displaying scFv (phenotype) are secreted from cell membrane


Carmen and Jermutus


Dow

nloaded from


CT domain in a phagemid system since

there will always be copies of wild-type

g3p from the helper phage for infection.

Selectively infective phage (SIP)

technology (reviewed in Jung et al.29)

exploits the modular nature of g3p and

can be used for identifying protein– ligand

interactions. The ligand is coupled to

N1–N2 either by the use of an expression

vector or, if it is sufficiently small (eg an

organic molecule), it can be chemically

coupled. The receptor/scFv is fused to

the CT domain of g3p as part of the

phage particle. It is the interaction

between the ligand and the receptor

which restores the domains of g3p,

thereby rendering the phage infectious

(Figure 5). The main difference between

this technology and standard phage

display is that the selection and infection

process are coupled. Since interaction

leads to infection, there is no need for

elution.

A variation of this technology —

termed Cys display — has recently been

presented.30 Here, the g3p and the

antibody fragment to be displayed are

both expressed separately in the cell,

although both with a terminal Cys. In the

oxidising periplasm, g3p and scFv can

interact, allowing for the formation of a

g3p–S–S–scFv fusion which can later be

eluted by reducing agents such as DTT.

No selection results have so far been

reported with this technology, but there is

the possibility of unpaired cysteine

residues interfering with disulphide bond

formation in the scFv, resulting in

misfolding and reduced yields.31

Competition from g3p of the helper

Figure 5: Formats for scFv display and vectors used. (A) Wild-type phage displaying multiple scFvs at the N-terminus ofthe N1 domain. (B) and (C) scFv displayed at the N-terminus of N1 or CT, respectively. Both these formats incorporatecopies of wild-type g3p. (D) SIP technology. scFv is displayed at the end of CT domain of wild-type g3p in the phage vector.Infectivity is dependent on the scFv interacting with a ligand attached to the N-terminal domains. (E) Hyperphage. Thehelper phage genome lacks a g3p so all copies of g3p are derived from the phagemid vector system, resulting in multivalentphage




Dow

nloaded from


phage and proteolytic degradation of the

g3p-fusion usually result in phage

populations where as little as ten per cent

of phage display a fusion protein.32 This is

an important consideration when trying

to represent large repertoires of antibody

fragments, since at least ten times more

phage are required to reach the theoretical

diversity. Use of a gene 3-less helper

phage helps to restore fusion protein

display levels.33 Hyperphage is a modified

helper phage which displays the g3p

phenotype but which contains a phage

vector that lacks the gene 3 sequence.34

Production of such hyperphage particles

necessitates exogenous g3p by culturing

in the E. coli strain DH5Æ/pIII. These

cells possess a lacZ-regulated gene 3

sequence on their chromosome. The

resulting infective hyperphage particles

are superinfected into cells containing a

phagemid, and, during phage production,

the only g3p produced is the scFv-fusion

from the phagemid vector. This results in

multivalent display, a phenomenon also

resulting from the use of phage vectors.

The potential advantage of this system is

the improved chance of finding a clone of

interest from large populations, due to a

lower background from phage displaying

only the helper phage g3p. Alternatively,

the g3p could be supplied from cells

containing a plasmid encoding the gene 3

sequence regulated by the phage shock

promoter (psp).35 This operon is induced

following infection by filamentous phage,

resulting in production of g3p which can

complement a helper phage with a

deleted gene 3 sequence.

CONSTRUCTION OFPHAGE ANTIBODYLIBRARIESNaive antibody librariesThe genomic information coding for

antibody variable domains is usually

derived from B cells of either ‘naıve’

(non– immunised) or immunised

donors.36 An antibody repertoire from

immunisation is generally restricted to

generating antibodies against the antigen

of the original immunogenic response,

whereas naıve libraries have the advantage

that they can theoretically be used for an

unlimited range of antigens. Construction

of these libraries involves relatively

straightforward molecular biology

techniques such as RT of mRNA,

followed by polymerase chain reaction

(PCR) with germline-specific primers to

amplify the VH and VL gene segments

from the cDNA template, and restriction-

based cloning to incorporate the

rearranged antibody segments into an

appropriate phagemid display vector.

Finally, the vectors are transformed into

E. coli cells to generate the antibody

repertoire. The first naıve scFv phagemid

library was constructed using peripheral

blood lymphocytes (PBLs) and was

predicted to have a diversity of .107.

Antibody fragments isolated from this

library demonstrated micromolar

affinities.37 About five years later, a scFv

library of .1010 recombinants, derived

from tonsil and PBLs, was described.10

This library allowed for the selection of

scFvs with affinities in the subnanomolar

range. These results support theoretical

predictions38 that the size of the library is

thought to be proportional to the

affinities of the isolated antibodies. The

limiting factor for generating large library

sizes is the transformation step; even if this

is optimised, library sizes in the range of

108 –1011 can be accomplished only after

numerous electroporations. Recently,

protocol improvements, primarily relating

to DNA purification and choice of strain,

have been described that allow the

generation of library sizes of 1010 in a

single electroporation step.39

Synthetic antibody librariesConstruction of synthetic libraries

involves rearranging VH and VL gene

segments in vitro and introducing artificial

complementarity determining region

(CDRs) of varying loop lengths using

PCR and randomised oligonucleotide

primers. One of the earliest synthetic

repertoires used a diverse repertoire of

human VH gene segments paired with a

constant Vº3 (DPL16) light chain

Most phage do notdisplay a fusion protein

Library sizes havecontinued to increase

Naıve libraries containnatural CDRs

Synthetic librariescontain artificial CDRs


Carmen and Jermutus


Dow

nloaded from


sequence. A randomised VH CDR3

ranging from four to 12 residues was

introduced40 which resembles the natural

loop length diversity more closely than an

earlier synthetic repertoire which had

loop lengths of either five or eight

residues.41 A library of .108

recombinants was generated from which

antibodies to a variety of human antigens,

as well as a diverse collection of non-

human proteins, were isolated.

Most antibody fragments usually

require an oxidising environment for

folding, although it is possible to isolate

antibodies that are able to fold into a

functional and stable antibody domain

structure in a reducing environment.

These antibodies are relatively rare and

might be restricted to certain germlines.

They may possess an above average

thermodynamic stability and should be

particularly useful for intracellular

applications. A previously isolated scFv

(F8), shown to be functionally expressed

in both bacterial and plant cytoplasm, was

used as a scaffold framework.42 Specific

residues in the CDR3s of the heavy and

light chain were randomised and a library

(diversity of 5 3 107) was generated.

Antibodies of submicromolar affinity,

which can be functionally expressed in

the cytoplasm, have been successfully

isolated to a variety of targets, such as

proteins from plant viruses, as well as

more common protein targets such as

lysozyme.

Some of the more recently constructed

synthetic repertoires have opted to restrict

the number of frameworks used. The

HuCAL library43 is a fully synthetic

antibody library which has yielded

antibodies with nanomolar (nM) affinities

to a number of antigens. The framework

diversity is limited to seven heavy and

seven light chain consensus sequences.

Based on sequence analysis of human

variable antibody domain genes, the

authors determined that the majority of

canonical structures are represented by

these consensus sequences. For each

antibody germline family, scFv sequences

were optimised for expression by

avoiding rare codons in E. coli. All six

CDRs are flanked by restriction sites to

facilitate affinity maturation.

In vivo recombined antibodylibrariesTo get around the earlier limits imposed

by the transformation step, some libraries

have been constructed using in vivo

recombination or combinatorial infection

to generate highly diverse repertoires of

either synthetic or naive antibody

fragments. A synthetic Fab library was

constructed using two vectors,5 each with

a different antibiotic selectable marker.

One was a plasmid derived from pUC19

and the second was a phage vector

(fdDOG), and the heavy and light chain

antibody sequences were cloned into

these vectors, respectively. The vectors

were then incorporated into the same cell

by a simple phage rescue strategy and cells

were subsequently co-infected by the

chloramphenicol-resistant bacteriophage

P1. P1 provides the Cre recombinase

which recognises the loxP sites flanking

the heavy and light chain antibody

sequences and recombines them within

the cell. The library was predicted to

contain 6.4 3 1010 recombinants, the

recombinants being detected on the basis

of their resistance to all three antibiotics.

Recently, a single vector system also

using Cre recombinase has been

described.44 A phagemid vector

containing both VH and VL segments was

transfected into cells and the phage

population produced from this primary

library was then allowed to infect bacteria

at a high multiplicity of infection (ratio of

phage to cells), so ensuring that more than

one vector is incorporated per cell. The

VH and VL segments were then exchanged

between vectors generating multiple new

VH/VL combinations. Using this method,

a naive scFv library with a diversity of

3 3 1011 was created. This library is

potentially more stable than those

previously constructed using in vivo

recombination because of the use of the

smaller phagemid vector.

Antibodies that formfunctional proteins in areducing environmentare rare

Large library sizes canalternatively be createdin vivo recombination




Dow

nloaded from


Quality control of antibodylibrariesAs larger libraries are possible with

improved technology, the task of assessing

diversity remains difficult. BstNI

fingerprinting is a quick way of checking

for diversity since the scFv sequence can

be amplified directly from colonies by

PCR. The product is then digested using

the BstNI enzyme and different sequences

will give different band patterns on an

agarose gel. Sequencing is a preferable

method since it also provides information

on clones that have frameshifts or stop

codons and demonstrates additional

diversity which may not be obtained from

a BstNI restriction digest; however, the

sample size required to assess library sizes

above 109 is well beyond most laboratory

sequencing capabilities. Furthermore,

information attained by sequence

diversity may not equate to the functional

diversity of the library. A clone can

contain a complete and in-frame VH –VL

sequence, but it may not produce

functional protein because the antibody

fragment may be aggregated, misfolded or

toxic to the cell. Aggregated antibody

may result in ‘bald’ phage — ie phage that

display only g3p from the helper phage

vector. As the vast majority of phage are

either monovalent or ‘bald’, large

populations of phage, much greater than

the library size, are required for efficient

representation of the library diversity. The

best measure to assess library quality is

how well the library performs in

selections. Populations of medium to high

affinity (nM range) antibodies should be

isolated directly from a quality library for

a highly diverse panel of antigens.

Ways to improve display levels, other

than those mentioned earlier, may include

co-expression of scFv fragments with the

E. coli Skp and/or FkpA periplasmic

proteins. These chaperones have been

shown to improve both soluble and

displayed scFv levels.45–47 Alternatively,

clones that display g3p from the helper

phage, due to absence of

g3-fusion protein, can be effectively

removed using a modified helper phage.

A protease cleavage site introduced into

the gene 3 sequence of the helper phage

genome between N2 and CT resulted in

the KM13 helper phage,48 which was

then used to rescue a phagemid library.

Clones that expressed misfolded scFvs

and, as a consequence, packaged only

helper phage g3p, were rendered non-

infectious by incubation with trypsin.

Trypsin can also be used effectively as an

elution reagent49,50 which also removes

‘bald’ clones via their protease cleavage

site. It can also be used sequentially

following, for example, competitive

elution with the original antigen.51 Using

this method, a dramatic reduction in

background from non-specific phage was

observed compared with standard elution

methods. This can probably be explained

by effective removal of all frameshifted, or

poorly expressing, clones from the

population, when previously they had

been seen to have a growth advantage.52,53

SELECTIONTECHNOLOGIESWhile a lot of work has concentrated on

the generation of larger and more rational

or designed libraries, other researchers

have concentrated their efforts on the

selection process itself. The selection

protocol can be divided up into five main

steps: (i) coating of antigen; (ii) block; (iii)

incubation of phage with antigen; (iv)

washing to remove non-specific phage;

and (v) elution (Figure 6A–D).

Surfaces can be coated with antigen in

a variety of ways, the most common

being direct adsorption to a plastic surface

where it is non-covalently associated via

electrostatic and Van-der-Waals

interactions. If it is a short peptide or a

small organic molecule, it is more

commonly conjugated to a carrier protein

such as BSA (bovine serum albumin) or

KLH (keyhole limpet haemocyanin)

before immobilisation onto the plastic

surface, as it will not remain bound on its

own or, if bound, will be sterically

difficult to access by the phage library.

Antigen can be biotinylated for selections

in solution. The phage–antibody–antigen

Library quality can beassessed by itsperformance inselections

Co-expression ofchaperones canimprove display levels

Small molecules arefirst conjugated to acarrier molecule


Carmen and Jermutus


Dow

nloaded from


complexes are subsequently captured

using streptavidin-coated paramagnetic

beads. Selections on cell surfaces have also

been successfully demonstrated,54–59 as

well as on protein bands from

nitrocellulose membranes following two-

dimensional (2D) electrophoresis.50,60,61

Antigen purity is important since

contaminants presented to a naıve

antibody library are very likely to generate

antibodies to proteins other than the

target antigen. The proportion of protein

that is immobilised while retaining its

native structure and how often each

epitope is presented and accessible are

further considerations. While antigen is

usually coated onto solid phase by passive

adsorption, this process may cause

denaturation of the molecules.62,63 This,

in turn, can sometimes favour non-

specific phage or antibody binding,64

since the denatured form will present

exposed hydrophobic residues, causing a

‘sticky’ surface. Factors such as the

chemistry of antigen immobilisation, the

blocking agent used and the density of

immobilised antigen can influence how

well the antibody fragments can access a

given epitope (Figure 7).64

The main purpose of the blocking

agent is to coat any exposed areas of the

solid surface that are not already coated

with an antigen molecule. During the

blocking phase, antigen that is reversibly

bound may well desorp.64 It is essential to

rinse thoroughly after blocking since

residual blocking agent containing

desorped antigen may compete for

Pure antigen in nativeconformation improvesphage output specificity

Reversibly boundantigen may reduce thenumber of specificantibody hits

Figure 6: The phage display/selection cycle. (A) Antigen is immobilised onto plastic. (B) Library/outputs are incubatedwith the antigen surface. (C) Unbound phage are removed and the surface is washed. (D) Phage are eluted. (E) E. coli cellsare infected with eluted phage. (F) Cells are plated onto selective agar plates. (G) Selection output is amplified. (H) Helperphage superinfect cells containing phagemid vector. (I) Phage replication and assembly provides population of enrichedphage for next round of selection




Dow

nloaded from


specific antibody fragments after addition

of the library. Zhuang et al.64 recommend

the use of streptavidin-coated beads and a

biotinylated antigen since, unlike passive

absorption, this process does not favour

denaturation of the antigen and, perhaps

more importantly, immobilisation/

coating is nearly covalent due to the tight

interaction of biotin and streptavidin.

Incubation with the library can be

performed at various temperatures

ranging from 48C to 378C. Some

protocols shake/agitate the selection

tubes/plates to improve the chances of

specific phage encountering antigen.

Phage displaying scFv that are not able to

bind the antigen will either remain in

solution or may bind non-specifically.

The washing step is thought to be crucial

for the removal of non-specific phage.

Many repeated quick rinses may favour

the removal of non-specific and low

affinity clones since their fast off-rates will

cause the majority of them to be in

solution when the wash buffer is

discarded. Repeated washes are thought

to have a dilution effect as more and more

non-specific clones are removed.

There are many elution reagents to

choose from, ranging from those which

are pH dependent (triethylamine,

glycine/HCl) to those with enzymatic

mechanisms (trypsin and chymotrypsin)

or competitive elution with soluble

antigen. The nature of the antigen can

influence how well specifically bound

phage is eluted from the antigen. For

instance, a protein that has an overall

positive charge may favour elution of

specific phage at a low pH — such as

glycine/HCl at pH 2.2.

It is crucial to try to maximise the

sequence diversity accessed from the total

repertoire at round one since it is this

subset that is used for further selection

rounds and its diversity will influence the

success of selection. At round one, the

number of clones that will not bind the

antigen far exceeds the number that do.

Limiting antigen concentration or lack of

exposure of certain epitopes may result in

the loss of specific clones, and potent

clones present at low level may be missed.

During subsequent selection rounds,

growth advantages may result in the

over-representation of some clones thus,

adopting more than one selection strategy

increases the chance of obtaining diverse

round one output populations.

Selections comparing immunotube and

multipin surfaces on ten to 11 antigens65

resulted in the isolation of 124 different

antibodies in total, each strategy

producing very different panels of

Quality of round oneoutput is critical toselection success

Small changes inselection protocol caninfluence outputdiversity

Figure 7: Accessibilityof epitope. The amountof accessible epitope canbe significantly lowerthan the antigenconcentration. Antigen isrepresented by shadedcircles, with the epitopebeing represented inblack. White circlesrepresent the blockingagent. The plastic surfaceis represented by a thinblack line. Adapted fromZhuang et al.

64


Carmen and Jermutus


Dow

nloaded from


antibodies with very few clones common

to both strategies. Small variations such as

the surface area and the concentration of

antigen and phage were sufficient to cause

these differences in output diversity.

Three alternative selection strategies were

adopted (B. M. Edwards et al.,

unpublished data) to isolate antibodies

from a 1011 library to BLyS

(B lymphocyte stimulator). The antigen

was (i) immobilised on immunotubes; (ii)

biotinylated and coupled to streptavidin-

coated plates; or (iii) biotinylated and used

in soluble selection. The selection resulted

in 2,730 scFvs which specifically

recognised the BLyS antigen; of these,

1,287 were sequence unique and most

were only isolated once. By using diverse

selection strategies, the authors maximised

sequence diversity. Cell surface selections

can be problematic since a number of

antigens in addition to the target protein

may be displayed on the cell surface.

Edwards et al.59 used seven different

selection strategies for cell surface

selections on human adipose tissue. Out

of 109 unique antibodies, shown

specifically to bind adipocyte plasma

membrane, 87 were isolated from just one

of the strategies. These few examples

support the theory that the choice of

selection methods may influence output

numbers and diversity. Selection

methodologies should be tailored to fit

each antigen class.

More recently developed selection

techniques include the use of antigen

prepared by western blotting on

nitrocellulose membranes as a target for

selections.50,60,61 2D gel electrophoresis

was used for the production of protein

profiles from patients with different

disease states. Isolation of phage antibodies

to proteins on these blots is likely to

prove extremely useful in the further

characterisation of these diseases. Antigen

can also be displayed as a fusion to a

Lpp–OmpA’ hybrid on the surface of the

E. coli cells.66 This allows a novel selection

approach called ‘delayed infectivity

panning’ (DIP). The antigen on the

bacterial cell surface is exposed to a

population of phage displaying antibody

fragments. Non-specific infection is

prevented by growing the cells at 168C,

which prevents formation of the F-pilus.

Once non-specific phage have been

removed, the temperature is raised to

378C, which facilitates F-pilus formation

and, thereby, permits infection.

FUTUREAs phage display technology moves into

its second decade, a number of phage

display-derived antibodies are now in

advanced clinical trials. Library sizes are

getting larger and library design is more

tailored to the needs and later application

of antibody fragments. Much more is

understood about antibody structure and

important contact residues involved in

antibody–antigen interactions.67 The

expression of eukaryotic proteins in E. coli

has improved with the use of frameworks

that do not have detrimental effects on

the host cell. Production of secondary

libraries for affinity maturation is greatly

facilitated by restriction sites flanking

whole V gene segments, permitting heavy

and light chain shuffling,68 or flanking

specific CDRs, allowing hypervariable

regions to be shuffled within the

frameworks using randomised CDR

cassettes.43 Furthermore, automation of

phage display enables high throughput

selection of specific antibodies and renders

the process further compatible with

hybridoma technology.69

Alternative display technologies, such

as ribosome display, have been optimised

(for review, see Amstutz et al.70). Libraries

with a diversity of 1014 are possible using

this cell-free system. It may be possible to

isolate slightly different populations of

scFv fragments since constraints on

diversity from expression in E. coli are

avoided with this in vitro technology and

affinity maturation can occur

simultaneously with selections. A new

‘display-less’ technology has been shown

to isolate antibodies with subnanomolar

affinities; termed ‘periplasmic expression

and cytometric screening’ (PECS), this

technology directs soluble scFv to the

Selections on cellsurface are morecomplex




Dow

nloaded from


periplasmic space.71 Fluorescently labelled

ligand (up to 10 kilodaltons) is then

incubated with the cell population.

Specific binding of the ligand with the

scFv in the periplasmic space results in an

increased fluorescence which allows

positive cells to be separated from the

remaining population using fluorescence-

activated cell sorting (FACS).

While technologies will further

improve to allow for faster, more

automated selection of higher-affinity

binders in fewer selection cycles, the

complexity of antigens will gradually

increase. This will provide further

opportunities for the continued

development of creative approaches to

library design and selection strategies, as

exemplified by the advent of the first

in vivo selection in humans.72

References

1. Griffiths, A. D. and Duncan, A. R. (1998),‘Strategies for selection of antibodies by phagedisplay’, Curr. Opin. Biotechnol., Vol. 9, pp.102–108.

2. Smith, G. P. (1985), ‘Filamentous fusionphage: novel expression vectors that displaycloned antigens on the virion surface’, Science,Vol. 228, pp. 1315–1317.

3. McCafferty, J, Griffiths, A. D., Winter, G. andChiswell, D. J. (1990), ‘Phage antibodies:filamentous phage displaying antibody variabledomains’, Nature, Vol. 348, pp. 552–554.

4. Arndt, K. M., Muller, K. M. and Pluckthun,A. (2001), ‘Helix-stabilized Fv (hsFv) antibodyfragments: substituting the constant domains ofa Fab fragment for a heterodimeric coiled�coildomain’, J. Mol. Biol., Vol. 312, pp. 221–228.

5. Griffiths, A. D., Williams, S. C., Hartley, O.et al. (1994), ‘Isolation of high affinity humanantibodies directly from large syntheticrepertoires’, EMBO J., Vol. 13, pp. 3245–3260.

6. de Haard H. J., van Neer, N., Reurs, A. et al.(1999), ‘A large non-immunized human Fabfragment phage library that permits rapidisolation and kinetic analysis of high affinityantibodies’, J. Biol. Chem., Vol. 274, pp.18218–18230.

7. Parhami, S. B., Viswanathan, M. andMargolies, M. N. (2002), ‘Selection of highaffinity p-azophenyarsonate Fabs from heavy-chain CDR2 insertion libraries’, J. Immunol.Meth., Vol. 259, pp. 43–53.

8. de Carvalho, N. C., Williamson, R. A.,

Parren, P. W. H. I., Lundkvist, A., Burton,D. R. and Bjorling, E. (2002), ‘Neutralisinghuman Fabs fragments against measles virusrecovered by phage display’, J. Virol., Vol. 76,pp. 251–258.

9. Holt, L. J., Enever, C., de Wildt, R. M. T.and Tomlinson, I. M. (2000), ‘The use ofrecombinant antibodies in proteomics’, Curr.Opin. Biotechnol., Vol. 11, pp. 445–449.

10. Vaughan, T. J., Williams, A. J., Pritchard, K.et al. (1996), ‘Human antibodies with sub-nanomolar affinities isolated from a large non-immunised phage display library’, Nat. Biotech.,Vol. 14, pp. 309–314.

11. Makowski, L. and Russel, M. (1997),‘Structure and assembly of filamentousbacteriophages’, in Kay, B., Winter, G. andMcCafferty, J., Eds., ‘Structural Biology ofViruses’, Oxford University Press, New York;pp. 352–380.

12. Webster, R. E. (1996), ‘Biology of thefilamentous bacteriophage’, in ‘Phage Display ofPeptides and Proteins’, Academic Press Inc., SanDiego, CA., pp. 1–20.

13. Sambrook, J. and Russell, D. W. Eds. (2001),‘Molecular Cloning: A Laboratory Manual’(3rd edn), Cold Spring Harbour LaboratoryPress, New York, NY.

14. Holliger, P. and Riechmann, L. (1997),‘A conserved infection pathway forfilamentous bacteriophages is suggested by thestructure of the membrane penetration domainof the minor coat protein g3p from phage fd’,Curr. Biol., Vol. 5, pp. 265–275.

15. Stengele, I., Bross, P., Garces, X., Giray, J. andRasched, I. (1990), ‘Dissection of functionaldomains in phage fd adsorption protein –discrimination between attachment andpenetration sites’, J. Mol. Biol., Vol. 212,pp. 143–149.

16. Riechmann, L. and Holliger, P. (1997), ‘TheC-terminal domain of Tol A is the coreceptorfor filamentous phage infection of E.coli’, Cell,Vol. 90. pp. 351–360.

17. Boeke, J. D., Model, P. and Zinder, N. D.(1982), ‘Effects of bacteriophage f1 gene IIIprotein on the host cell membrane’, Mol. Gen.Genet., Vol. 186, pp. 185–192.

18. Gao, C., Mao, S., Lo, C. L., Wirsching, P.,Lerner, R. A. and Janda, K.D. (1999), ‘Makingartificial antibodies: a format for phage displayof combinatorial heterodimeric arrays’, Proc.Natl. Acad. Sci. USA, Vol. 96, pp. 6025–6030.

19. Jespers, L. S., Messens, J. H., De Keyser, A.et al. (1995), ‘Surface expression and ligand-based selection of cDNAs fused to filamentousphage gene VI’, BioTechnology, Vol. 13, pp.378–382.

20. Fuh, G., Pisabarro, T. M., Li, Y., Quan, C.,Lasky, L. A. and Sidhu, S.S. (2000), ‘Analysisof PDZ domain-ligand interactions using


Carmen and Jermutus


Dow

nloaded from


carboxyl-terminal phage display’, J. Biol.Chem., Vol. 275, pp. 21486–21491.

21. Fuh, G. and Sidhu, S. (2000), ‘Efficient phagedisplay of polypeptides fused to the carboxy-terminus of the M13 gene-3 minor coatprotein’, FEBS Lett., Vol. 480, pp. 231–234.

22. Scott, J. K. and Smith, G. P. (1990), ‘Searchingfor peptide ligands with an epitope library’,Science, Vol. 249, pp. 386–390.

23. Greenwood, J., Willis, A. E. and Perham,R. N. (1991), ‘Multiple display of foreignpeptides on a filamentous bacteriophage’,J. Mol. Biol., Vol. 220, pp. 821–827.

24. Iannolo, G., Minenkova, O., Petruzzelli, R.and Cesareni, G. (1995), ‘Modifyingfilamentous phage capsid: limits in the size ofthe major capsid protein’, J. Mol. Biol., Vol.248, pp. 835–844.

25. Hoogenboom, H., Griffiths, A. D., Johnson,K. S., Chiswell, D. J., Hudson, P. and Winter,G. (1991), ‘Multi-subunit proteins on thesurface of filamentous phage: methodologiesfor displaying antibody (Fab) heavy and lightchains’, Nucleic Acid Res., Vol. 19, pp. 4133–4137.

26. Gibson, T. J. (1984), ‘Studies on the Epstein-Barr Virus genome’, PhD thesis. Cambridge,UK: Cambridge University.

27. Carter, P., Bedouelle, H. and Winter, G.(1985), ‘Improved oligonucleotide sitedirected mutagenesis using M13 vectors’,Nucleic Acid Res., Vol. 13, pp. 4431–4443.

28. Viera, J. and Messing, J. (1987), ‘Production ofsingle-stranded plasmid DNA’, Meth.Enzymol., Vol. 153, pp. 3–11.

29. Jung, S., Arndt, K. M., Muller, K. M. andPluckthun, A. (1999), ‘Selectively infectivephage (SIP) technology: scope and limitations’,J. Immunol. Meth., Vol. 231, pp. 93–104.

30. Proceedings of the 11th Annual InternationalConference on Antibody Engineering,December 2000 (San Diego, CA, USA).

31. Waibel, R., Alberto, R., Willuda, J. et al.(1999), ‘Stable one–step technetium-99mlabeling of his-tagged recombinant proteinswith a novel Tc(I)-carbonyl complex’,Nat. Biotechnol., Vol. 17, pp. 897–901.

32. McCafferty, J. (1996), ‘Phage display: factorsaffecting panning efficiency’, in ‘Phage Displayof Peptides and Proteins’, Academic Press, Inc.,San Diego, CA., pp. 261–276.

33. Jackson, R. H., McCafferty, J., Johnson, K. S.et al. (1992), ‘Selection of variants of antibodiesand other protein molecules using phagedisplay on the surface of bacteriophage fd’, inRees, D.C., Sternberg, M. J. E. and Wetzel,R., Eds., ‘Protein Engineering: A PracticalApproach’, IRL Press, Oxford, pp. 277–301.

34. Rondot, S., Koch, J., Breitling, F. and Dubel,S. (2001), ‘A helper phage to improve single-

chain antibody presentation in phage display’,Nat. Biotechnol., Vol. 19, pp. 75–78.

35. Rakonjac J., Jovanovic, G. and Model, P.(1997), ‘Filamentous phage infection-mediated gene expression: construction andpropagation of the gIII deletion mutanthelper phage R408d3’, Gene, Vol. 198, pp.99–103.

36. Clackson T., Hoogenboom, H. R., Griffiths,A. D. and Winter, G. (1991), ‘Makingantibody fragments using phage displaylibraries’, Nature, Vol. 352, pp. 624–628.

37. Marks, J. D., Hoogenboom, H. R., Bonnet,T. P., McCafferty, J., Griffiths, A. D. andWinter, G. (1991), ‘By-passing immunisation’,J. Mol. Biol., Vol. 222, pp. 581–597.

38. Perelson, A. S. and Oster, G. F. (1979),‘Theoretical studies of clonal selection:minimal antibody repertoire size and reliabilityof self-non-self discrimination,’ J. Theor. Biol.,Vol. 81, pp. 645–670.

39. Sidhu, S. S., Lowman, H. B., Cunningham,B. C. and Wells, J. A. (2000), ‘Phage displayfor selection of novel binding peptides’, Meth.Enzymol., Vol. 328, pp. 333–363.

40. Nissim, A., Hoogenboom, H. R., Tomlinson,I. M. et al. (1994), ‘Antibody fragments from a‘single pot’ phage display library asimmunochemical reagents’, EMBO J., Vol. 13,pp. 692–698.

41. Hoogenboom, H. R. and Winter, G. (1992),‘By-passing immunisation, human antibodiesfrom synthetic repertoires of germline VH genesegments rearranged in vitro’, J. Mol. Biol.,Vol. 227, pp. 381–388.

42. Desiderio, A., Franconi, R., Lopez, M. et al.(2001), ‘A semi-synthetic repertoire ofintrinsically stable antibody fragments derivedfrom a single-framework scaffold’, J. Mol. Biol.,Vol. 310, pp. 603–615.

43. Knappik, A., Ge, L., Honegger, A. et al.(2000), ‘Fully synthetic human combinatoriallibraries (HuCAL) based on modular consensusframeworks and CDRs randomised withtrinucleotides’, J. Mol. Biol., Vol. 296,pp. 57–86.

44. Sblattero, D. and Bradbury, A. (2000),‘Exploiting recombination in single bacteria tomake large phage antibody libraries’, Nat.Biotechnol., Vol. 18, pp. 75–80.

45. Bothmann, H. and Pluckthun, A. (1998),‘Selection for a periplasmic factor improvingphage display and functional periplasmicexpression’, Nat. Biotech., Vol. 16, pp. 376–380.

46. Bothmann, H. and Pluckthun, A. (2000), ‘Theperiplasmic Escherichia coli peptidylprolylcis,trans isomerase FkpA. I. Increasedfunctional expression of antibody fragmentswith and without cis-prolines’, J. Biol. Chem.,Vol. 275, pp. 17100–17105.




Dow

nloaded from


47. Hayhurst, A. and Harris, W. J. (1999),‘Escherichia coli skp chaperone coexpressionimproves solubility and phage display of scFvantibody fragments’, Prot. Expr. Purif., Vol. 15,pp. 336–343.

48. Kristensen, P. and Winter, G. (1998),‘Proteolytic selection for protein folding usingfilamentous bacteriophages’, Fold. Design,Vol. 3, pp. 321–328.

49. Engberg, J., Andersen, P. S., Nielsen, L. K.,Dziegiel, M., Johansen, L. K. and Albrechtsen,B. (1996), ‘Phage-display libraries of murineand human antibody Fab fragments’, Mol.Biotechnol., Vol. 6, pp. 287–310.

50. Ravn, P., Kjaer, S., Jensen, K. H. et al. (2000),‘Identification of phage antibodies toward theWerner protein by selection on western blots’,Electrophoresis, Vol. 21, pp. 509–516.

51. Goletz, S., Christensen, P. A., Kristensen, P.,Blohm, D., Tomlinson, I., Winter, G. andKarsten, U. (2002), ‘Selection of largediversities of antiidiotypic antibody fragmentsby phage display’, J. Mol. Biol., Vol. 315,pp. 1087–1097.

52. de Bruin, R., Spelt, K., Mol, J., Koes, R. andQuattrocchio, F. (1999), ‘Selection of highaffinity phage antibodies from phage displaylibraries’, Nat. Biotechnol., Vol. 17, pp. 397–399.

53. Carcamo, J., Ravera, M. W., Brissette, R.et al. (1998), ‘Unexpected frameshifts fromgene to expressed protein in a phage displayedpeptide library’, Proc. Natl. Acad. Sci. USA,Vol. 95, pp. 11146–11151.

54. Marks, J. D., Ouwehand, W. H., Bye, J. M.et al. (1993), ‘Human antibody fragmentsspecific for human blood group antigens froma phage display library’, Bio/Technology, Vol.11, pp. 1145–1149.

55. De Kruif, J., Terstappen, L., Boel, E. andLogtenberg, T. (1995), ‘Rapid selection of cellsubpopulation-specific human monoclonalantibodies from a synthetic phage antibodylibrary’, Proc. Natl. Acad. Sci. USA, Vol. 92,pp. 3938–3942.

56. Pereira, S., Maruyama, H., Siegel, D. et al.(1997), ‘A model system for detection andisolation of a tumor cell surface antigen usingantibody phage display’, J. Immunol. Meth.,Vol. 203, pp. 11–24.

57. Palmer, D. B., George, A. J. T. and Ritter,M. A. (1997), ‘Selection of antibodies to cellsurface determinants on mouse thymicepithelial cells using a phage display library’,Immunology, Vol. 91, pp. 473–478.

58. Osbourn, J. K., Earnshaw, J. C., Johnson,K. S., Parmentier, M., Timmermans, V. andMcCafferty, J. (1998), ‘Directed selection ofMIP-1 alpha neutralizing CCR5 antibodiesfrom a phage display human antibody library’,Nat. Biotechnol., Vol. 16, pp. 778–781.

59. Edwards, B. M., Main, S. H., Cantone, K. L.,Smith, S. D., Warford, A. and Vaughan, T. J.(2000), ‘Isolation and tissue profiles of a largepanel of phage antibodies to the humanadipocyte cell surface’, J. Immunol. Meth.,Vol. 245, pp. 67–78.

60. Liu, B. and Marks, J. D. (2000), ‘Applyingphage antibodies to proteomics: Selectingsingle chain Fv antibodies to antigens blottedon nitrocellulose’, Anal. Biochem., Vol. 286,119–128.

61. Liu, B., Huang, L., Sihlbom, C., Burlingame,A. and Marks, J. D. (2002), ‘Towardsproteome-wide production of monoclonalantibody by phage display’, J. Mol. Biol.,Vol. 315, pp. 1063–1073.

62. Kontermann, R. E. (2001), ‘Immunotubeselections’, in Kontermann, R. and Dubel, S.,Eds., ‘Antibody Engineering’, SpringerLaboratory Manuals, Berlin, pp.137–148.

63. Friguet, B., Djavadi-Ohaniance, L. andGoldberg, M. E. (1984), ‘Some monoclonalantibodies raised with a native protein bindpreferentially to the denatured antigen’, Mol.Immunol., Vol. 21, pp. 673–677.

64. Zhuang, G., Katakura, Y., Furuta, T.,Omasa, T., Kishimoto, M. and Suga, K.(2001), ‘A kinetic model for a biopanningprocess considering antigen desorption andeffective antigen concentration on a solidphase’, J. Biosci. Bioeng., Vol. 91, pp. 474–481.

65. Lou, J., Marzari, R., Verzillo, V. et al. (2001),‘Antibodies in haystacks: how selectionstrategy influences the outcome of selectionfrom molecular diversity libraries’, J. Immunol.Meth., Vol. 253, pp. 233–242.

66. Benhar, I., Azriel R., Nahary, L. et al. (2000),‘Highly efficient selection of phage antibodiesmediated by display of antigen as Lpp–OmpA’fusions on live bacteria’, J. Mol. Biol., Vol. 301,pp. 893–904.

67. MacCallum, R. M., Martin, A. C. andThornton, J. M. (1996), ‘Antibody–antigeninteractions: contact analysis and binding sitetopography’, J. Mol. Biol., Vol. 265, pp. 732–745.

68. Marks, J. D., Griffiths, A. D., Malmqvist, M.,Clackson, T. P., Bye, J. M. and Winter, G.(1992), ‘By-passing immunisation: buildinghigh affinity human antibodies by chainshuffling’, Bio/Technology, Vol. 10, pp. 779–783.

69. Krebs, B., Rauchenberger, R., Reiffert, S.et al., (2001), ‘High-throughput generationand engineering of recombinant humanantibodies’, J. Immunol. Meth., Vol. 254,pp. 67–84.

70. Amstutz, P., Forrer, P., Zahnd, C. andPluckthun, A. (2001), ‘In vitro displaytechnologies: novel developments and


Carmen and Jermutus


Dow

nloaded from


applications’, Curr. Opin. Biotechnol., Vol. 12,pp. 400–405.

71. Chen G., Hayhurst, A., Thomas, J. G.,Harvey, B. R., Iverson., B. L. andGeorgiou, G. (2001), ‘Isolation ofhigh-affinity ligand-binding proteins byperiplasmic expression with cytometric

screening (PECS)’, Nat. Biotechnol., Vol. 19,pp. 537–542.

72. Arap, W., Koloninm, M.G., Trepel, M. et al.(2002), ‘Steps toward mapping the humanvasculature by phage display’, Nat. Med.,Vol. 8, pp. 121–127.




Dow

nloaded from


briefings in functional genomics and proteomics-2002-carmen-189-203

Documents