Phage Display Technology as a Powerful Platform for Antibody Drug Discovery

Antibody drugs with a high affinity and specificity are effective and safe for intractable diseases, such as cancers and autoimmune diseases. Furthermore, they have played a central role in drug discovery, currently accounting for eight of the top 20 pharmaceutical products worldwide by sales. Forty years ago, clinical trials on antibody drugs that were thought to be a magic bullet failed, partly due to the immunogenicity of monoclonal antibodies produced in mice. The recent breakthrough in antibody drugs is largely because of the contribution of phage display technology. Here, we reviewed the importance of phage display technology as a powerful platform for antibody drug discovery from various perspectives, such as the development of human monoclonal antibodies, affinity enhancement of monoclonal antibodies, and the identification of therapeutic targets for antibody drugs.


Introduction
In 1985, Smith et al. demonstrated phage display technology using artificial peptide sequences on the N-terminus of a bacteriophage surface protein [1]. Phage display technology allows the construction of libraries in which various peptides and proteins are displayed on the phages, and then the most suitable clone is selected from the library by in vitro panning. Therefore, the clones with an affinity for the target of interest or those with the ability to migrate to the target tissue are enriched from the library. Therefore, the development of phage display technology provides optimal sequences to target peptides or proteins, unlike conventional alanine scanning and other methods, and aids understanding of their molecular evolution [2][3][4][5][6][7]. Although various kinds of molecular display technologies such as ribosome display [8,9], and yeast display technologies [10,11] have been proposed, phage display technology is frequently employed because of a high diversity of molecules that can be displayed and ease of handling [12,13].
In particular, phage display technology has become a powerful platform for drug discovery in life science because it is easy to produce antibodies in vitro. Winter et al. displayed a single-chain variable fragment (scFv) antibody that consists of the variable heavy chain (VH) and the variable light chain (VL) joined together by a flexible peptide linker in 1990 [14]. Subsequently, several antibody formats, such as scFv, fragment antigenbinding (Fab), and variable fragment (VHH) derived from heavy chain antibodies of Camelidae, have been reported to be displayed on the phages ( Figure 1). Therefore, the phage display technology allows the construction of various phage antibody libraries such as naïve [15,16], immunized [17,18] and synthetic phage antibody [19,20] and contributes to the current development of antibody drugs. More than 70 phage-derived monoclonal antibodies have entered clinical studies, and 14 of them have been approved for use till May 2020. Here, we review the importance of phage display technology as a powerful platform for drug discovery, from target discovery to antibody drug development.
Three frequently used antibody formats (scFv, Fab and VHH) are shown.

Development of Antibody Drugs Using Phage Display Technology
In 1975, Köhler and Milsterin established a method for generating monoclonal antibodies using hybridomas [21]. Since this discovery, monoclonal antibodies with high specificity and affinity to target molecules were expected to be used as magic bullets for various clinical applications [22]. However, clinical trials have largely failed because mouse monoclonal antibodies recognized as heterologous proteins in the human body are highly immunogenic, and they showed reduced efficacy because of human anti-mouse antibodies, so-called HAMAs [23]. This problem has been overcome by antibody engineering. For example, human chimeric antibodies were produced by replacing the Fc sequence of mouse antibodies with that of human antibodies to reduce immunogenicity [24,25]. Moreover, humanized antibodies were produced by replacing their protein sequences with the complementarity-determining region (CDR) of mouse antibodies, which is important for binding to the antigen. Using the same technology, rituximab (Rituxan ® ), infliximab (Remicade ® ), and trastuzumab (Herceptin ® ) were approved in the United States between 1997 and 1998 [24,25]. Nowadays, the production of human antibodies has been established using phage human antibody library [24][25][26] and human antibody-producing transgenic mice [24,25,27].
Taken together, phage display technology has made a significant contribution to antibody engineering. Therefore, we summarized the fundamental mechanisms of antibody engineering and the contributions of phage display technology.

Development of Human Monoclonal Antibodies Using Phage Display Technology
As mentioned above, fewer mouse sequences in the monoclonal antibodies are essential to reduce their immunogenicity in the human body. Therefore, there is an urgent need to create monoclonal antibodies with all-human sequences. In this respect, two approaches to generate such monoclonal antibodies using phage display technology have been established.
The first approach is the use of a phage display-derived human antibody library, in which the antibody repertoire in the human body is displayed in vitro on the phages. For example, Cambridge Antibody Technology (CAT) and Dyax constructed phage displayderived human naïve scFv and Fab antibody libraries by extracting antibody genes from human spleen, peripheral blood lymphocytes, tonsils, and fetal liver [15,16,28,29]. A number of antibodies that are currently approved or in ongoing clinical studies have been isolated from these libraries, such as belimumab against B-lymphocyte stimulator (BLyS) isolated from CAT's library (Benlysta ® ) [30] and avelumab against programmed death-1 ligand-1 (PD-L1) isolated from Dyax's library (Bavencio ® ) [31].
The other approach is humanization of mouse monoclonal antibody using the guided selection method [32,33]. Guided selection is based on chain shuffling of variable genes using phage display technology. As an example, in the guided selection procedure using the Fab format, the first step is to clone the variable regions of a mouse antibody to a phagemid vector containing human antibody constant domains, resulting in a chimeric Fab. The next step is to construct a human VL shuffled library by replacing the mouse VL with human VL repertoires and then selecting binders after panning against a target antigen, which results in the selection of human VL paired with the mouse VH. The third step is to construct a human VH shuffled library by replacing the mouse VH with human VH repertoires and then selecting binders after panning against a target antigen, which results in the selection of complete-human Fab clones. Finally, the epitope specificity and affinities of selected human Fabs are confirmed using the appropriate assays. For example, adalimumab against tumor necrosis factor-alpha (TNFα; Humira ® , clone: D2E7), the first fully human antibody-blockbuster drug, was created from mouse anti-human TNFα (clone: Mab32) as a template using the guided selection method [34].
Therefore, phage display technology contributes to the development of antibody drugs by overcoming immunogenicity as a bottleneck for the magic bullet. The approved antibody drugs derived from phage display technology are shown in Table 1.

In Vitro Affinity Maturation of Monoclonal Antibodies Using Phage Display Technology
Antibodies are effective and safe drugs that can achieve complete remission for intractable diseases, such as cancers and autoimmune diseases [57,58] because they have a high affinity and specificity for their antigens. However, many antibodies, isolated from naïve phage, antibody libraries with an antibody repertoire for any antigen have low affinity for the antigens because somatic hypermutations did not occur. Therefore, in order to use antibodies with low affinity for the antigens as research tools for therapeutics and diagnostics, it is necessary to enhance the affinity for the antigens. Phage display technology is commonly used to improve the affinity for antigens, which is important for high antibody efficacy. A number of antibody drugs such as moxetumomab pasudotox against CD22 (Lumoxiti ® ) [52] and ranibizumab against vascular endothelial growth factor A (VEGFA) (Lucentis ® ) [35] have been manufactured by in vitro affinity maturation using phage display technology. In this approach, gene libraries are constructed with mutations in the antigen-binding regions such as VL and VH, the libraries are displayed on the phages, and the antibodies with high affinity for the antigens are then isolated by biopanning. In order to construct gene libraries with mutations, there are two major approaches: (1) Random mutagenesis and (2) site-specific mutagenesis.
In random mutagenesis, mutagenesis is induced by anticancer agents [59], radiation [59], or by gene recombination with chain shuffling [14,60,61]. Among these techniques, error-prone polymerase chain reaction (error-prone PCR) is often employed [62]. This approach leverages the natural error rate of a low-fidelity DNA polymerase. In this approach, the affinity of monoclonal antibodies is enhanced using the following steps. Mutations in the CDRs of the ideal monoclonal antibody clone for affinity maturation are randomly introduced by error-prone PCR. From these mutation libraries, the clones that strongly bind to the antigen are enriched by biopanning. By repeating these steps, antibody clones with a higher affinity than the template antibody can be generated. The method of constructing mutation libraries by error-prone PCR and selecting the antibody clone with high affinity by biopanning has been applied to various antibodies, and their affinities have been improved. In particular, using the error-prone DNA shuffling method combined with DNA shuffling, the affinity of anti-fluorescein scFv was improved with dissociation constant (KD) of up to 50 fM and slower dissociation kinetics (half-time > 5 days) than those for the streptavidin-biotin complex [63]. As random mutagenesis using methods such as error-prone PCR is not site-specific and randomly causes mutations throughout the sequence, the introduction of logically inconceivable mutations can enhance the interaction and increase the stability of the structure of the antibody. Conversely, there are concerns that the three-dimensional structure of the antibody contact region may be disrupted by introducing random mutations.
Against this background, site-specific mutagenesis can be used to theoretically estimate the sites to be mutated from the structural data of the antibody [64] or the hotspot sequence of the somatic hypermutation [65,66], and then mutate them into a sequence encoding 20 different amino acids using PCR with mutagenic primers [67,68]. Thus, the antibody with improved affinity can be efficiently prepared without disrupting the structure of the antibody, if the sites in which mutations are introduced are appropriate. For example, Chowdhury et al. constructed a library of randomly mutated hotspot sequences of somatic hypermutations, such as A/G-G-C/T-A/T and AGT in the VL CDR3 of antimesothelin scFv, and enriched high-affinity antibodies for mesothelin by biopanning. A series of procedures resulted in the successful isolation of the higher-affinity scFvs with KD of several hundred pM [65].
Therefore, using the phage display technology, high-affinity antibodies can be efficiently isolated from various types of mutagenesis libraries, which have their own advantages and disadvantages and complement each other (Table 2).
Taken together, monoclonal antibodies with high affinity for antigens and low immunogenicity have been produced using phage display technology, according to the flow chart in Figure 2. Furthermore, as this technology can be applied to a variety of proteins as well as antibody drugs, it is a versatile and powerful platform for drug discovery. Table 2. Advantages and disadvantages of random mutagenesis and site-specific mutagenesis for constructing gene libraries with mutations.

Random mutagenesis
The introduction of logically inconceivable mutation can enhance the interaction of antigen-antibody and increase the stability of the structure of the antibody.
The three-dimensional structure of the antibody contact region may be disrupted by introducing random mutations.

Site-specific mutagenesis
The antibody with higher affinity could be efficiently prepared without disrupting the structure of the antibody.
The introduction of logically inconceivable mutations cannot be established. Antibody genes (VL and VH) are amplified from the immune tissues, such as in peripheral blood lymphocytes, spleen, and tonsils of animals such as human and mouse. Then, their genes are cloned into phagemid vectors. Therefore, the antibody repertoires of each animal can be displayed in vitro on the phages. This library has the theoretical advantage of producing monoclonal antibodies against a variety of antigens. Conversely, it has the disadvantage of generally isolating monoclonal antibodies with low affinity for the antigens.

・Immunized library
The experimental animals, such as mouse, rat, and rabbit, are immunized with the target antigen. Alternatively, animals, including human that have been exposed to the target antigen, are selected. Antibody genes are amplified from the immune tissues of the animal, and then, their genes are cloned into phagemid vectors. Therefore, the antibody repertoires, mainly against the target antigen, can be displayed in vitro on the phages. This library has the advantage of efficiently isolating monoclonal antibodies with high affinity against the target antigen. Conversely, it has the disadvantage of being less versatile because it is specific against the target antigen.

・Synthetic library
The amino acids and length of the antigen-binding region, especially CDR3 of VH, of a template antibody with characteristics such as high frequency or high stability. Therefore, the library offers advantages, such as isolation of monoclonal antibodies with high expression in the host cells, excellent stability and solubility, and ease of genetic engineering. Conversely, it has the disadvantage that the sequences of the isolated antibodies could be partially artificial.  The characteristics of three frequently used phage antibody libraries are mentioned here. For other processes, refer to the respective sections in the text.

Target Discovery Using Phage Display Technology
By developing and using the technologies described in the previous section, antibody drugs currently account for eight of the top 20 pharmaceutical products worldwide by sales (Table 3). Furthermore, many monoclonal antibodies are in ongoing clinical trials [69,70]. On the other hand, as the number of targets for these antibody drug candidates is small, many monoclonal antibody clones have been developed for the same promising targets. For example, among the antibodies approved or clinically tested in Japan, four clones including nivolumab (Opdivo ® ) [71,72], pembrolizumab (Keytruda ® ) [73,74], spartalizumab [75,76] and cemiplimab [77,78] have been developed for programmed death 1 (PD-1), and three clones including avelumab (Bavencio ® ) [79,80], atezolizumab (Tecentriq ® ) [81,82] and durvalumab (Imfinzi ® ) [83,84] have been developed for PD-L1. It has recently been noted that the targets of antibody drugs are dwindling. Therefore, it is essential to identify novel targets for the future development of antibody drugs. In this section, we discuss examples of utilizing phage display technology for the discovery of novel therapeutic targets.

Search for Antigens that React with Autoantibodies Using Phage Display Technology
Autoimmune diseases are intractable diseases in which the immune system fails to function properly and attacks its own body tissues. The causes of autoimmune disease, such as rheumatoid arthritis (RA) [85,86], systemic lupus erythematosus (SLE) [87,88] and multiple sclerosis (MS) [89,90] remain unknown. Although, the recent development of biologics, mainly antibody drugs, has partially improved the outcomes of diseases, such as RA [91][92][93][94], the prognosis for these diseases remains poor. Therefore, understanding the molecular pathogenesis of autoimmune diseases and searching for new therapeutic targets is essential for the development of new therapies.
Antibodies to self-antigens produced in autoimmune diseases can be used as biomarkers. For example, in vasculitis, the number of antibodies to myeloperoxidase and proteinase 3 in neutrophils was clinically tested [95,96]. In SLE, the number of antibodies to doublestranded DNA and small nuclear ribonucleoprotein in the nucleus has also been clinically tested [97,98]. Therefore, the search for antigens that bind to autoantibodies in autoimmune diseases is a promising approach to elucidate molecular pathogenesis and target identification.
In this respect, self-antigens specific for autoimmune diseases can be efficiently explored using phage display technology. To achieve this, a complementary DNA (cDNA) phage display library is first constructed based on mRNA extracted from biological tissues. Next, the phage clones that bind to the antibodies collected from blood and other sources from patients with autoimmune diseases are enriched by biopanning using this library. Finally, by analyzing the sequences of cDNA in the eluted phages, self-antigens are identified using high-throughput technologies. For example, in the case of MS, self-antigens binding to IgG in cerebrospinal fluid and serum of patients have been explored by constructing cDNA libraries from the human brain. As a result, 14 different self-antigens have been identified, including DEAD-box helicase 24 [99]. Moreover, similar approaches have also been applied to autoimmune diseases such as RA [100] and SLE [101].
Besides autoimmune diseases, cancer antigens have also been identified from autoantibodies in cancer patients. For example, in head and neck cancers, the cancer-specific antigens displayed on the phages were enriched by subtraction biopanning. Validation of 21 clones demonstrated L23 to be a novel cancer antigen, which is highly expressed in head and neck cancers compared that in normal keratinocytes. Knockdown of L23 inhibited proliferation, invasion, and cell survival, whereas its overexpression showed opposite effects [102]. Furthermore, similar approaches have also been applied to various types of cancers, such as breast [103] and prostate cancers [104], and paraneoplastic neurological syndrome [105].

High-Throughput Validation of Therapeutic Target Candidates Using Phage Display Technology
In current drug discovery methods, it is important to identify not only therapeutic targets but also biomarkers for understanding pathological conditions, including approaches such as companion diagnostics. An approach to identify these molecules involves comprehensive omics studies, such as genomics [106,107] and transcriptomics [107,108] have been attracting attention. In particular, proteomics is a large-scale study in which proteins, end products of the central dogma, are comprehensively analyzed, and proteomics plays a central role in post-genome research [109,110]. An approach to identify therapeutic targets and biomarkers involves comparing proteins expressed in cells or tissues of healthy and diseased cases, called disease proteomics, is commonly used [111,112]. This research area has made significant progress because even small numbers of differentially expressed proteins can be efficiently identified by improving the performance of mass spectrometry [113,114]. Therefore, the remaining issue for this research area is the high-throughput validation of these disease-related proteins.
In this respect, monoclonal antibodies with high a affinity and specificity for the antigen proteins have been commonly used [115][116][117]. Techniques using monoclonal antibodies, such as enzyme-linked immunosorbent assay, western blotting, fluorescent imaging, and tissue microarray (TMA) staining are extremely useful for examining the function and distribution of proteins [118,119]. Usually, monoclonal antibodies are generated using hybridomas; however, this approach is laborious and time-consuming, and it requires a large number of recombinant antigens. Furthermore, protein production using this approach often requires gene engineering for heterologous expression, which takes time for optimization. Therefore, it is impractical to produce monoclonal antibodies against the multiple candidate proteins identified when using a proteomics approach for protein selection. To address this issue, an approach called antibody proteomics technology has been developed, which uses a phage antibody library and TMA analysis to rapidly and comprehensively isolate monoclonal antibodies against candidate proteins for the identification of potential biomarkers and therapeutic targets [120]. In this section, we describe the development and evaluation of this novel technology.
First, in order to produce monoclonal antibodies against a large number of diseaserelated proteins efficiently, we focused on a naïve phage antibody library that can produce monoclonal antibodies against various types of antigens in vitro. Thus, we previously constructed an improved naïve phage antibody library by designing a PCR primer set that allowed for comprehensive amplification of the VH and VL genes, so that rapid isolation of monoclonal antibodies against multiple target proteins can be performed in vitro [121,122]. Subsequently, we attempted to construct a monoclonal antibody preparation method using the proteins (in the order of ng) that can be recovered directly from two-dimensional differential in-gel electrophoresis (2D-DIGE) gels, which are commonly used in proteomics studies, without the preparation of recombinant protein as an antigen. Monoclonal antibodies can be isolated even from small amounts of antigen (only 0.5 ng in the case of kinase insert domain receptor protein) using a nitrocellulose membrane with excellent protein adsorption capacity as an immobilized carrier for antigen proteins, which is called nitrocellulose membrane panning method [120]. These processes allowed us to establish a method for rapid antibody preparation in vitro for many candidates identified using 2D-DIGE analysis. Finally, in order to efficiently validate the candidates in a large number of clinical samples, we focused on TMA using clinical tissue sections from multiple cases mounted on a single glass slide. Immunohistochemical staining of TMA with monoclonal antibodies against the antigen of interest allowed us to elucidate the expression profiles of a large number of cases in a single analysis. In addition, we could also analyze the correlation of expression profiles with clinical information, including age, sex, medical history, and medications. By optimizing the conditions, we established a method for immunohistochemical staining of TMA using monoclonal antibodies displayed on the phage, without the preparation of recombinant monoclonal antibodies [120]. By combining these techniques, we could perform the entire process from proteome analysis to identification of therapeutic targets and biomarker proteins in approximately one month.
The antibody proteomics technology comprises four stages: (1) Search for diseaserelated proteins by proteomics-based analysis using 2D-DIGE, (2) identification of the candidate proteins using mass spectrometry analysis, (3) isolation of monoclonal antibodies against the candidate proteins using a phage antibody library, and (4) validation of the candidate proteins using TMA analysis ( Figure 3).
Hence, this technology accelerates the identification of proteins that are potentially useful as biomarkers or therapeutic targets. Moreover, it could be an alternative to the conventional approach in which prioritized proteins are validated one by one among the identified proteins, and it can become a fundamental system for drug discovery.
To evaluate the practicality of this technology, we applied it to the search for therapeutic targets for breast cancer [120,123]. We identified a novel therapeutic target that is specifically expressed in the testes of normal tissues and refractory cases of breast cancer (triple-negative breast cancer) called Eph receptor A10 (EphA10) [124][125][126]. EphA10 expression was also shown to be correlated with breast cancer stage and lymph node metastasis, suggesting that it is a promising target for breast cancer therapy [127]. In order to demonstrate the proof of concept that inhibition of EphA10 specifically expressed in breast cancers induces an anti-tumor effect, we generated a neutralizing monoclonal antibody against EphA10 and then analyzed the therapeutic effect in a xenograft mouse model. Analysis of tumor volumes showed that anti-EphA10 neutralizing monoclonal antibody inhibited tumor growth in a concentration-dependent manner [123,128]. In addition, EphA10 was highly expressed not only in breast cancer but also in prostate cancer [129], suggesting that EphA10 could be a promising target for various kinds of tumors. (1) Candidate proteins ar tected using two-dimensional differential in-gel electrophoresis (2D-DIGE) and (2) identified mass spectrometry analysis (MSA). Simultaneously, (3) monoclonal antibodies against all th teins identified by 2D-DIGE are produced using a phage antibody library. Finally, (4) the pro are validated as biomarkers and therapeutic targets using tissue microarray (TMA). Therefo using this technology, the candidate proteins can be comprehensively validated, and the mo useful proteins can be selected.

Conclusions and Prospects
In this review, we discussed the significant contribution of phage display techn for the development of antibody drugs, which play a central role in drug discovery various perspectives.
In addition to the ones mentioned above, bispecific antibodies that recogniz antigens in a single antibody format are currently being commercialized [132][133][134]. display technology is also used as an essential technology for the developm bispecific antibodies [135][136][137]. Furthermore, this technology is used for the develop of antibody-mimicking peptides, microantibodies, as potential next-generation bio [138,139]. Therefore, it is expected that phage display technology will continue t powerful platform to lead innovative drug discovery in the future [140].    (2) identified by mass spectrometry analysis (MSA). Simultaneously, (3) monoclonal antibodies against all the proteins identified by 2D-DIGE are produced using a phage antibody library. Finally, (4) the proteins are validated as biomarkers and therapeutic targets using tissue microarray (TMA). Therefore, using this technology, the candidate proteins can be comprehensively validated, and the most useful proteins can be selected.
We also applied this technology to the search of biomarkers of metastasis in lung cancers, an important prognostic factor in cancers, and found that the expression levels of oxysterol-binding proteins, such as oxysterol-binding protein-related protein 5 (OSBPL5) and calumenin (CALU), were significantly higher in positive cases of lymph node metastasis than in negative cases [130]. Interestingly, 80 % of co-expression cases with both proteins were positive for lymph node metastasis, whereas approximately 50-60 % of cases with expression of at least one of the two proteins were positive for lymph node metastasis. Furthermore, invasion assays showed that the lung cancer cell lines in which these genes were overexpressed or knocked down significantly enhanced or inhibited the invasion capacity, suggesting that both proteins had the ability to promote lung cancer cell invasion. Therefore, OSBPL5 and CALU play a role in lymph node metastasis by enhancing the invasiveness of lung cancer cells.
Finally, we applied the technology to the search for biomarkers for companion diagnostics. While, cisplatin has been commonly used as a major anticancer agent against malignant mesothelioma, it has many adverse side effects. In a previous study, we identified that the malignant mesothelioma cell lines in which the Annexin A4 gene was overexpressed or knocked down showed significantly inhibition or enhancement of the cytotoxicity of cisplatin, demonstrating that Annexin A4 is involved in cisplatin resistance [131].
In this manner, antibody proteomics technology has enabled selection of promising proteins from disease-related proteins based on scientific evidence. Consequently, it is expected that antibody proteomics technology will aid identification of a large number of therapeutic targets and biomarker proteins for the development of novel diagnostic and therapeutic agents.

Conclusions and Prospects
In this review, we discussed the significant contribution of phage display technology for the development of antibody drugs, which play a central role in drug discovery from various perspectives.
In addition to the ones mentioned above, bispecific antibodies that recognize two antigens in a single antibody format are currently being commercialized [132][133][134]. Phage display technology is also used as an essential technology for the development of bispecific antibodies [135][136][137]. Furthermore, this technology is used for the development of antibodymimicking peptides, microantibodies, as potential next-generation biologics [138,139]. Therefore, it is expected that phage display technology will continue to be a powerful platform to lead innovative drug discovery in the future [140].