3 Selection Strategies II: Antibody Phage Display

3.1 Introduction

The production of polyclonal antibodies by immunisation of animals is a method established for more than a century. The first antibody serum was directed against Diphterie and produced in horses (von Behring and Kitasato, 1890). Hybridoma technology was the next milestone, allowing the production of monoclonal antibodies by fusion of an immortal myeloma cell with an antibody producing spleen cell (Köhler and Milstein 1975). However, hybridoma technology has some limitations, the possible instability of the aneuploid cell lines, most of all its inability to produce human antibodies and to provide antibodies against toxic or highly conserved antigens (Winter and Milstein 1991).

To overcome the limitations of hybridoma technology, antibodies or antibody fragments can be generated by recombinant means (Fig. 1). The most common used antibody fragments are the Fragment antigen binding (Fab) and the single chain Fragment variable (scFv). The Fab fragment consists of the fd fragment of the heavy chain and the light chain linked by a disulphide bond. The variable region of the the heavy chain (VH) and the variable region of the light chain (VL) are connected by a short peptide linker in the scFv. A major breakthrough in the field of antibody engineering was the generation of antibody fragments as recombinant proteins in the periplasmatic space of E. coli(Better et al. 1988, Huston et al. 1988, Skerra and Plückthun 1988). To circumvent the instability of hybridoma cell lines, the genes encoding VHand VLof a monoclonal antibody can be cloned into an E. coliexpression vector in order to produce antibody fragments in the periplasmatic space of E. coliwhich preserve the binding specificity of the parental hybridoma antibody (Toleikis et al. 2004).

The production of mouse derived monoclonal antibody fragments in E. colidid not remove the major barrier for the broad application of antibodies in therapy as repeated administration of mouse derived antibodies causes a human anti-mouse antibody (HAMA) response (Courtenay-Luck et al. 1986). This problem can be overcome by two approaches: By humanisation of mouse antibodies (Studnicka

et al. 1994) or by employing repertoires of human antibody genes. The second approach was achieved in two ways. First, human antibody gene repertoires were inserted into the genomes of IgG-knockout-mice, allowing to generate hybridoma cell lines which produce human immunoglobulins (Jakobovits 1995, Lonberg und Huszar 1995, Fishwild et al.1996). However, this method still requires immunisation and has limitations in respect of toxic and conserved antigens.

These restrictions do not apply for the more rational second approach: the complete in vitrogeneration of specific antibodies from human antibody gene repertoires. There, despite of the constant suggestion of novel methods like bacterial surface display (Fuchs et al. 1991, for review see Jostock and Dübel 2005), ribosomal display (Hanes and Plückthun 1997), puromycin display (Roberts and Szostak 1997) or yeast surface display (Boder and Wittrup et al. 1997), phage display has become the most widely used selection method (Table 3.1, chapter “An exciting start- and a long trek” Fig. 1.3), which is based on the groundbreaking work of Smith (1985). The genotype and phenotype of a polypeptide were linked by fusing short gene fragments to the minor coat protein III gene of the filamentous bacteriophage M13. This resulted in the expression of this fusion protein on the surface of phage, allowing affinity purification of the gene of interest by the polypeptide binding. The first antibody gene repertoires in phage were generated and screened by using the lytic phage Lambda (Huse et al.1989, Persson 1991) with limited success. Consequently, antibody fragments were presented on the surface of M13, fused to pIII (McCafferty 1990, Barbas et al.1991, Breitling et al.1991, Clackson et al.1991, Hogenboom et al.1991, Marks et al.1991). By uncoupling antibody gene replication and expression from the phage life cycle by locating them on a separate plasmid (phagemid), genetic stability, propagation and screening of antibody libraries was greatly facilitated (Barbas et al.1991, Breitling et al.1991, Hoogenboom 1991, Marks et al.1991). To date, “single-pot” (see below) antibody libraries with a theoretical diversity of up to 1011independent clones were assembled (Sblattero and Bradbury 2000) to serve as a molecular repertoire for phage display selections.

3.2 The Phage Display System

Due to its robustness and straightforwardness, phage display has been the selection method most widely used in the past decade. Display systems employing insertion of antibody genes into the phage genome have been developed for phage T7 (Danner and Balesco 2001), phage Lambda (Huse et al. 1989, Mullinax et al. 1990, Kang et al. 1991a) and the Ff class (genus inovirus) of the filamentous
phages f1, fd and M13 (McCafferty 1990). Being well established for peptide display, the phage T7 is not well suited for antibody phage display because it is assembled in the reducing enviroment of the cytoplasm, thus leaving most antibodies unfolded (Danner and Balesco 2001). In contrast, the oxidizing milieu of the bacterial periplasm allows antibody fragments to be folded and assembled properly (Skerra and Plückthun 1988). The Ff class non-lytic bacteriophages are assembled in this cell compartment and allow the production of phages without killing the host cell (Fig. 1). This is a major advantage compared to the lytic phages Lambda (Huse et al. 1989). In addition, filamentous phages allow the production of soluble proteins by introducing an amber stop codon between the antibody gene and gene III when using phagemid vectors. In an E. colisupE suppressor strain, the fusion proteins will be produced, whereas soluble antibodies are made in a non suppressor strain (Marks et al. 1992a, Griffiths et al. 1994), but expression in suppressor strains is also possible (Kirsch et al.2005). Therefore, the members of the Ff class are the phages of choice for antibody phage display.

To achieve surface display, five of the M13 coat proteins (Fig. 1) have been used in fusion to foreign proteins, protein fragments or peptides. In the commonly used system the antibody fragment is coupled to the N-terminus or second domain of the minor coatprotein pIII (Barbas et al. 1991, Breitling et al. 1991, Hoogenboom et al. 1991). The function of the 3–5 copies of pIII, in particular
their N-terminal domain, is to provide interaction of the phage with the F pili expressed on the surface of E. coli(Crisman and Smith 1984). The major coat protein pVIII has been considered as an alternative fusion partner, with only very few success reports in the past decade (Kang et al. 1991b). pVIII fusions are obviously more useful for the display of short peptides (Cwirla et al. 1990, Felici et
al. 1991). Fusions to pVI have also been tried, but not yet with antibody fragments (Jespers et al. 1995). pVII and pIX were used in combination, by fusing the VLdomain to pIX and the VHdomain to pVII, allowing the presentation of a Fv fragment on the phage surface. Thus this format offers the potential for heterodimeric display (Gao et al. 1999). However, the fusion with pIII remains the most
widely used system for antibody phage display and is still the only system of practical relevance.

Two different systems have been developed for the expression of the antibody::pIII fusion proteins. First, the fusion gene can be inserted directly into the phage genome substituting the wildtype (wt) pIII (McCafferty et al. 1990). Second, the fusion gene encoding the antibody fusion protein can be provided on a separate plasmid with an autonomous replication signal, a promoter, a resistance marker and a phage morphogenetic signal, allowing this “phagemid” to be packaged into assembled phage particles. A helperphage, usually M13KO7, is necessary for the production of the antibody phage to complement the phage genes not encoded on the plasmid. Due to its mutated origin, the M13KO7 helperphage genome is not efficiently packaged during antibody phage assembly when compared to the phagemid (Vieira and Messing 1987).

In the system using direct insertion into the phage genome, every pIII protein on a phage is fused to an antibody fragment. This is of particular advantage in the first round of panning, where the desired binder is diluted in millions to billions of phages with unwanted specifi city. The oligovalency of these phages improves the chances of a specific binder to be enriched due to the improvement of apparent affinity by the avidity effect. This advantage, however, has to be weighted to a number of disadvantages. The transformation efficiency of phagemids is two to three orders of magnitude better than the efficiency of phage vectors, thus facilitating the generation of large libraries. Second, the additional protein domains fused to pIII may reduce the function of pIII during reinfection. In a phagemid system, the vast majority of the pIII assembled into phage are wt proteins, thus providing normal pilus interaction. This may explain why only two “single-pot” antibody libraries (Griffiths et al. 1994, O’Connel et al. 2002) were made using a phage vector. In contrast, in phagemid systems both replication and expression of foreign fusion proteins are independent from the phage genome. There is no selection pressure, as propagation of the phagemid occurs in the absence of helperphage. The fusion proteins can be produced in adjustable quantities and allowing to use the amber/suppressor system for switching to soluble expression of antibody fragments without a pIII domain. Finally, despite usually not derived from highest copy plasmids, the dsDNA of phagemids is more easy to handle than phage DNA, facilitating cloning and analysis. Therefore, most pIII display systems use the phagemid approach. There is, however, a disadvantage which originates from the two independent sources for the pIII during phage packaging. During assembly, the wt pIII of the phage are inserted into the phage particles with much higher rate than the pIII fusion protein. As a result, the vast majority of resulting phage particles carry no antibody fragments at all. The few antibody phages in these mixtures are mainly monovalent, with phage carrying two or more antibody fragments being extremely rare. This allows to select for antibodies with a high monovalent affinity, since avidity effects decreasing the dissociation rate from the panning antigen can be avoided. In the first panning
round, however, when a few binders have to be fished out of a huge excess of unwanted antibody phages, the fact that only a few percent of the phages carry antibodies hampers the efficiency of the system (Barbas et al. 1991, Breitling et al. 1991, Hoogenboom et al. 1991, Garrard et al. 1991, Lowman et al. 1991, O’Connel et al. 2002). This problem can be overcome by using a newly developed helperphage “Hyperphage”. Hyperphage does not have a functional pIII gene and therefore the phagemid encoded pIII antibody fusion is the sole source of pIII in phage assembly offering multivalent display for phagemid vectors. This method improves antibody phage display by two orders of magnitude and vastly improves panning efficiency (Rondot et al. 2001).

Multivalent display can be also achieved by the integration of two amber stop codons into the gIII gene of the helper phage genome, offering the production of a functional helper phage “Ex-phage” in an E. colisuppressor strain. In the associated phagemid pIGT3, the antibody::pIII fusion is made without an amber stop codon and the antibody phage is produced in an E. colinon suppressor strain
(Baek et al. 2002). However, the necessary deletion of the amber stop codon in the phagemid makes it imperative to subclone the antibody gene or to use a protease to produce soluble antibodies in contrast to the Hyperphage system where the amber/suppressor system can be used for switching to soluble expression of antibody fragments without a pIII domain.

A niche application of phage display is the selective infective phage (SIP) technology. Here, antibody fragments are fused to the N-terminal domain of pIII by cloning into the phage genome, therefore every pIII carries an antibody and the deletion of the pIII N-terminal region made the phage non-infective. In turn, the antigen is fused to the C-terminal end of seperately produced soluble pIII Nterminal domain. The functional, F pili binding pIII is reconstituted when the antibody phage binds to the antigen, allowing only the correct antibody phage to infect E. coliand to be propagated (Spada and Plückthun 1997). However, due to the fast kinetics of pIII/pilin interactions and very low concentrations of the three reaction partners if not coexpressed in the same cell, this method is not applicable for the convenient panning of larger libraries

3.3 Selection and Evaluation of Binders

The novel procedure for isolating antibody fragments by their binding activity in vitrowas called “panning”, referring to the gold washers tool (Parmley and Smith 1988). The antigen is immobilized to a solid surface, such as nitrocellulose (Hawlisch et al.2001), magnetic beads (Moghaddam 2003), a column matrix (Breitling et al.1991) or, most widely used, plastic surfaces as polystyrol tubes (Hust et al.2002) or 96 well microtitre plates (Barbas et al.1991). The antibody phages are incubated with the surface-bound antigen, followed by thorough washing to remove the vast excess of non-binding antibody phages. The bound antibody phages can subsequently be eluted and reamplified by infection of E. coli. This amplifi cation allows detection of a single molecular interaction during panning, as a single antibody phage, by its resistance marker, can give rise to a bacterial colony after elution. This illustrates the tremendous sensitivity of the method. The selection cycle can be repeated by infection of the phagemid bearing E. colicolonies from the former panning round with a helperphage to produce new antibody phages, which can be used for further rounds of panning until a significant enrichment is achieved. The number of antigen specific  antibody clones should increase with every panning round. Usually 2–6 panning rounds are necessary to select specifically binding antibody fragments (Fig. 3.2). For an overview of available stratagies and protocols, refer to McCafferty et al.(1996) and Kontermann and Dübel (2001). High throughput methods using microtitreplates and robotics can facilitate and enhance the panning procedure (for review see Konthur et al. 2005).

In most cases, the first step in the evaluation process of potential binders is an ELISA with polyclonal phage preparations from each panning round on coated target antigen and a control protein, e.g. BSA. The next step is the production of soluble monoclonal antibody fragments – from the panningrounds showing a significant enrichment of specific binders in polyclonal phage ELISA – in microtitreplates, followed by ELISA on coated antigen and in parallel on control protein. Soluble Fab fragments can be detected by their constant domains, wheras soluble scFvs can be detected by their engineered tags, e.g. his- or cmyc-tag. The clones producing specific antibody fragments will be further analysed by sequencing. Here, specific binders in duplicate can be rejected. A subcloning into E. coliexpression vectors, like pOPE101 (Schmiedl et al. 2000) offers high scale production of antibody fragments for further analysis, e.g. analysis by flow cytometry (Schirrmann and Pecher 2005). Another important feature of an antibody is its affinity, which is analysed by surface plasmon resonance (BIAcore) (Lauer et al.2005). After analysis of specificity and affinity the selected antibody fragment can be subcloned into other formats like IgG or scfv::Fc-fusion in order to achieve avidity and immunological effector functions (Jostock et al.2004).

3.4 Phage Display Vectors

A large number of different phage display vectors have been constructed. Table 3.2 lists a selection of phage display vectors, without pretending to be complete. Some of them have not been used for the construction of a library up to now, but have been included since they offer ideas and alternatives, e.g. a system which allows to control the success of antibody gene cloning by green fluorescent protein (GFP) expression (Pascke et al. 2001)

A variety of different promoters have been employed for the expression of antibody fragments on the surface of phages. Widely used is the lac Z promoter (lacZ) derived from the lactose operon (Jacob and Monod 1961). The gIII promoter (gIII) from the bacteriophage M13 (Smith 1985), the tetracycline promoter (1x teto/p) (Zahn et al. 1999) and the phoA promotor of the E. colialkaline phosphatase (Garrard et al. 1991) were also successfully used. It seems that very strong promoters, e.g. the synthetic promoter PAI/04/03 (Bujard et al. 1987), are rather a disadvantage (Dübel personnel communication). To our knowledge, a systematic comparison of the different promotors has not been done.

The targeting of the antibody fragments to the periplasmatic space of E. colirequires the use of signal peptides. The pelB leader of the pectate lyase gene of Erwinia caratovora (Lei et al. 1987) is commonly used. The gIII leader (Smith 1985), the phoA leader of the E. colialkaline phosphatase and the ompA leader of E. coliouter membrane protein OmpA have also been used, being common to
many protein expression vectors (Skerra et al. 1993, Skerra and Schmidt 1999). Further examples are the heat-stable enterotoxin II (stII) signal sequence (Garrard et al. 1991) and the bacterial chloramphenicol acetyltransferase (cat) leader (McCafferty et al. 1994)

Due to the inability of E. colito assemble complete IgG, with one exception (Simmons et al. 2002), smaller antibody fragments are used for phage display. In particular, Fabs and scFvs have been shown to be the antibody formats of choice. As aforementioned, in Fabs, the fd fragment and light chain are connected by a disulphide bond. In scFvs, the VHand VLare connected by a 15–25 amino acid linker (Bird et al. 1988, Bird and Walker 1991, Huston et al. 1988). Soluble scFvs tend to form dimers, in particular when the peptide linker is reduced to three to twelve amino acid residues. Diabodies or tetrabodies are produced if the linker between VHand VLis reduced to a few amino acids (Kortt et al. 1997, Arndt et al. 1998, Le Gall et al. 1999). The dimerization agravates the determination of the affinity, due to the possible avidity effect of the antibody complex (Marks et al. 1992b). Furthermore, some scFvs show a reduced affinity up to one order of magnitude compared to the corresponding Fabs (Bird and Walker 1991). ScFvs with a higher affinity than the corresponding Fabs were rarely found (Iliades et al. 1998). Small antibody fragments like Fv and scFv can easily be produced in E. coli. The yield of functional Fvs expressed in E. coliis higher than the yield of the corresponding Fabs, due to a lower folding rate of the Fabs (Plückthun 1990, Plückthun 1991). In one example, the stability in long-term storage was much higher for Fabs than for scFvs. After 6 month the functionality of scFvs stored at 4ºC was reduced by 50 %, Fabs, however, showed no significant loss of functionality after one year (Kramer et al. 2002). The overall yield of Fvs expressed in E. colivary from 0.5 to 10 mg/l culture compared to 2 to 5 mg/l culture for Fabs (Ward 1993), but very high yields of 1–2 mg/l of soluble and functional F(ab′)2 have been reported (Carter et al.1992). Therefore, the choice of the antibody format, scFv or Fab, depends on the desired application.

For the expression of Fabs in E. colitwo polypeptide chains have to be assembled. In the monocistronic systems, e.g. pComb3, the antibody genes are under control of two promoters and each has its own leader peptide (Barbas et al. 1991), whereas in plasmids like pCES1 with a bicistronic Fab operon, both chains are under control of a single promoter, leading to a mRNA with two ribosomal binding sites (De Haard et al. 1999). The bicistronic system is more sufficient for the expression of Fabs (Kirsch et al. 2005).

Two variants of the antibody::pIII-fusion have been made. Either full size pIII or truncated version of pIII were used. The truncated version was made by deleting the pIII N-terminal domain. This domain mediated the interaction with the F pili of E. coli. Infection is provided by wt pIII, as only a small percentage of phage in phagemid-based systems are carrying an antibody. These truncated vectors are therefore not compatible with the use of Hyperphage or Ex-phage, as the fullsize pIII is necessary for infection (Rondot et al. 2001, Baek et al. 2002). Some phagemids, e.g. pSEX81 (Welschof et al. 1997) allow the elution of antibody phages during panning by protease digestion instead of pH shift. This is possible due to a protease cleavage site between pIII and the antibody fragment. Therefore complete recovery of specifically antigen bound antibody phages is possible, even in case of very strong antigen binding.

Most of the described phagemids have an amber stop codon between the antibody gene and gIII. This allows the production of soluble antibody fragments after transformation of the phagemid to a non suppressor bacterial strain like HB2151 (Griffiths et al. 1994). For phagemids like pComb, it is necessary to delete the gIII by digestion and religate the vector before tranformation into E. coli (Barbas et al. 1991). In the case of phagmids like pSEX81, the selected antibody genes have to be subcloned into a separate E. coliexpression vector like the pOPE series (Breitling et al. 1991, Dübel et al. 1993, Schmiedl et al. 2000).

3.5 Phage Display Libraries

Various types of phage display libraries have been constructed. Immune libraries are generated by amplification of V genes isolated from IgG secreting plasma cells of immunised donors (Clackson et al. 1991). From immune libraries antibody fragments with monovalent dissociations constants in the nM range can be isolated. Immune libraries are typically created and used in medical research to select an antibody fragment against one particular antigen, e.g. an infectious pathogen, and therefore would not be the source of choice for the selection of a large number of different specifi cities. Naive, semi-synthetic and synthetic libraries have been subsumed as “single-pot” libraries, as they are designed to isolate antibody fragments binding to every possible antigen. A correlation is seen between the size of the repertoire and the affinities of the isolated antibodies. Antibody fragments with a µM affinity have been isolated from a “single-pot” library consisting of approximately 107clones, whereas antibody fragments with nM affinities were obtained from a library consisting of 109independent clones (Hoogenboom 1997). It is evident that the chance to isolate an antibody with a high affinity for a particular antigen increases almost linearly to the size of the library. According to the source of antibody genes, “single pot” libraries (Table 3.3) can be naive libraries, semisynthetic libraries or fully synthetic libraries. Naive libraries are constructed from rearranged V genes from B cells (IgM) of non-immunized donors. An example for this library type is the naive human Fab library constructed by de Haardt et al.(1999), yielding antibodies with affinities up to 2.7 ×10−9
M. Semi-synthetic libraries are derived from unrearranged V genes from pre B cells (germline cells) or from one antibody framework with genetically randomized complementary determining region (CDR) 3 regions, as described by Pini et al.(1998). The antibody fragments obtained from this library
show affinities between 10−8M  and  10−9M, with one scFv having a dissociation constant of 5 ×10−11M. A combination of naive and synthetic repertoire was used by Hoet et al. (2005). They combined light chains from autoimmune patientswith a fd fragment containing synthetic CDR1 and CDR2 in the human VH3-23 framework and naive, origined from autoimmune patients, CDR3 regions. The fully synthetic libraries have a human framework with randomly integrated CDR cassettes (Hayashi et al. 1994). Antibody fragments selected from fully synthetic libraries exhibit affinities between 10−6M and 10−11M (Knappik et al. 2000). All library types – immune, naive, synthetic and their intermediates – are useful sources for the selection of antibodies for diagnostic and therapeutic purposes.

3.6 Generation of Phage Display Libraries

Various methods have been employed to clone the genetic diversity of antibody repertoires. After the isolation of mRNA from the desired cell type and the preparation of cDNA, the construction of immune libraries is usually done by a two step cloning or assembly PCR (see below). Naive libraries are constructed by two or three cloning steps. In the two step cloning strategy, the amplified repertoire of light chain genes is cloned into the phage display vector first, as the heavy chain contributes more to diversity, due to its highly variable CDRH3. In the second step the heavy chain gene repertoire is cloned into the phagemids containing the light chain gene repertoire (Johansen et al. 1995, Welschof et al. 1997, Little et al. 1999). In the three step cloning strategy, separate heavy and light chain libraries are engineered. The VHgene repertoire has then to be excised and cloned into the phage display vector containing the repertoire of VLgenes (De Haardt et al. 1999). Another common method used for the cloning of naive (McCafferty et al. 1994, Vaughan et al. 1996), immune (Clackson et al. 1991) or hybridoma (Krebber et al. 1997) scFv phage display libraries is the assembly PCR. The VHand VLgenes are amplified seperately and connected by a subsequent PCR, before the scFv encoding gene fragments are cloned into the vector. The assembly PCR is usually combined with a randomization of the CDR3 regions, leading to semi-synthetic libraries. To achieve this, oligonucleotide primers encoding various CDR3 and J gene segments were used for the amplication of the V gene segments of human germlines (Akamatsu et al. 1993). The CDRH3 is a major source of sequence variety (Shirai et al. 1999). Hoogenboom and Winter (1992) and Nissim et al. (1994) used degenerated CDRH3 oligonucleotide primers to produce a semi-synthetic heavy chain repertoire derived from human V gene germline segments and combined this repertoire with an anti-BSA light chain. In some cases a framework of a well known antibody was used as scaffold for the integration of randomly created CDRH3 and CDRL3 (Barbas et al. 1992, Desiderio et al. 2001). Jirholt et al.(1998) and Söderlind et al.(2000) amplified all CDR regions derived from B cells before shuffling them into one antibody framework in an assembly PCR reaction. An example for an entirely synthetic library, Knappik et al.(2000) utilized seven different VHand VL germline master frameworks combined with six synthetically created CDR cassettes. The construction of large naive and semi-synthetic libraries (Hoet et al. 2005, Løset et al. 2005, Little et al. 1999, Sheets et al. 1998, Vaughan et al. 1996) requires significant effort to tunnel the genetic diversity through the bottleneck of E. colitransformation, e.g. 600 transformations were necessary for the generation of a 3.5 ×1010
phage library (Hoet et al. 2005)

To move the diversity potentiating step of random VH/VLcombination behind the bottleneck of transformation, the Cre-lox or lamda phage recombination system has been employed (Waterhouse et al. 1993, Griffiths et al. 1994, Geoffrey et al. 1994). However, libraries with more than 1010independent clones have now been accomplished by conventional transformation, rendering most of these complicated methods unnecessary in particular as they may result in decreased genetic stability. A remarkable exception is the use of a genomically integrated CRE recombinase gene (Sblattero and Bradbury 2000) which is expected to solve the instability issue and allows the generation of libraries with complexities above the limit achievable by conventional cloning.

In summary, antibodies with nanomolar affinities can be selected from either type of library, naive or synthetic. If the assembly by cloning or PCR and preservation of molecular complexity is carefully controlled at every step of its construction, libraries of more than 1010independent clones can be generated.

Antibody phage display is delivering and will deliver high affinity human antibodies and antibody fragments for research as well as for diagnostic and therapeutic applications in the future.