Advances in Phage Display—A Perspective

Phage Antibody Libraries from Immunized Humans, Chickens, and Other Animals

In many phage antibody projects, antibody domains are obtained from animals that are immune to one or more antigens of interest, either because they have been artificially immunized with a vaccine or other immunogen or because they have been naturally infected by a pathogen. For example, Jing Yi Lai and Theam Soon Lim of Universiti Sains Malaysia have reviewed projects in which the donors are humans who have been infected by viruses, bacteria, or parasites (Lai and Lim 2020).

Junho Chung of Seoul National University and his colleagues introduce the chicken immune system (see Introduction: Generation of Chicken Antibody Libraries and Selection of Antigen Binders [Yang et al. 2025c]) and detail the construction and use of a chicken phage antibody library displaying scFv domains against a chosen antigen (see Protocol: Chicken Immunization Followed by RNA Extraction and cDNA Synthesis for Antibody Library Preparation [Yang et al. 2025a], Protocol: Generation of a Phage Display Chicken Single-Chain Variable Fragment Library [Yang et al. 2025b], Protocol: Preparation of VCSM13 Helper Phage for Display Library Reamplification and Bio-Panning [Yang et al. 2025d], and Protocol: Selection of Antigen Binders from a Chicken Single-Chain Variable Fragment Library [Yang et al. 2025e]). The project that the students carry out in the Phage Display Course largely follows these articles, using a pComb3 family phagemid vector developed at Scripps (Andris-Widhopf et al. 2000; Cary et al. 2000).

The chicken antibody system differs from the human and mouse systems in several respects that make library generation simpler, and therefore particularly appropriate for the Phage Display Course (see Introduction: Generation of Chicken Antibody Libraries and Selection of Antigen Binders [Yang et al. 2025c]). There are only two immunoglobulin loci in chickens, one for the heavy (H) chain and one for the λ light chain; there is no κ light chain. There is only a single functional germline V_H gene at the H locus, followed by four diversity (D) segments and a single joining (J) segment. Diversity is generated by VDJ joining and multiple somatic gene conversion events with more than 80 upstream V_H pseudogenes. There is also only a single functional V_λ gene at the λ locus, followed by a single J segment. Diversity is generated by VJ joining and multiple somatic gene conversion events with 25 upstream V_λ pseudogenes. At both loci, there is no sequence variation near the end of assembled V genes in antibody-producing B cells, and very limited diversity near the beginning. The entire repertoire of V_H genes can be amplified with only three forward primers combined with a single reverse primer, and the same is true for the V_λ gene repertoire (see Protocol: Generation of a Phage Display Chicken Single-Chain Variable Fragment Library [Yang et al. 2025b]). Most of the repertoire can be captured with a single forward primer for V_λ genes and a single forward primer for V_H genes, as is done in the Phage Display Course. This simplicity is in sharp contrast to the profusion of primers required to capture the repertoire of V_H, V_λ, and V_κ genes in human B cells (see Protocol: Generation of Antibody Libraries for Phage Display: Human Fab Format [Peng and Rader 2024b]). Constructing an scFv library from chickens is therefore relatively straightforward, making it a practical exercise in the Phage Display Course.

Chicken scFv libraries are constructed from laying hens that have been hyperimmunized with the antigen of interest (Andris-Widhopf et al. 2000; Cary et al. 2000; see also Protocol: Generation of a Phage Display Chicken Single-Chain Variable Fragment Library [Yang et al. 2025b]). The progress of the hens’ immune response to the antigen is assayed by titering immunoglobulin in blood or in egg yolks, each of which contains up to 100 mg of IgY (the chicken homolog of mammalian IgG). When the titer is adequate, the hens are euthanized, and RNA is extracted from lymphoid organs, such as spleen and bone marrow. The RNA is reverse-transcribed to create cDNA molecules that include V_H and V_λ coding sequences. The cDNAs in turn serve as templates for PCR amplification of the hens’ entire V_H and V_λ repertoires, using the simple primer sets described in the previous paragraph. The primers are designed to create a flexible linker between the V_H and V_λ amino acid sequences so that they fold into a functional scFv domain, and to install suitable flanking restriction sites so that the scFv construct can be inserted into the phagemid vector in the correct reading frame, as illustrated for SfiI cloning sites in Figure 7. Vector DNAs bearing the scFv inserts are transfected into E. coli host cells; when the transfected cells are infected with helper phages, they release virions bearing scFv domains on their recombinant pIII subunits. These virions are the input to the first round of affinity selection.

Chicken phage antibody libraries can be smaller than the all-purpose libraries discussed in the next subsection. All-purpose libraries, either synthetic or representing large numbers of individual donors, are intended to include antibody domains that bind almost any chosen antigen of interest. Hundreds of millions to hundreds of billions of independent phage clones are required to meet this goal. A large portion of the library must be used to ensure that individual clones are represented by a sufficient number of virions. A chicken phage antibody library, in contrast, represents the antibody repertoire of chickens that have been hyperimmunized with the antigen of interest, so that a substantial number of their circulating antibodies are specific for the immunogen. A few million independent clones usually suffice to include high-affinity antibodies specific for the antigen.

More than 20 years ago, a few hens were hyperimmunized with the dye fluorescein conjugated to bovine serum albumin (FL-BSA), as described previously (Andris-Widhopf et al. 2000). The hens made not only antibodies that reacted with BSA, but also antibodies that reacted with the fluorescein dye. Small molecules like fluorescein that provoke an antibody response when conjugated to a protein antigen are called haptens; the proteins they are conjugated to are called carriers. RNA was extracted from the hens’ spleens and bone marrows and frozen in aliquots for long-term storage. Aliquots have served as the starting point for the Phage Display Course's student projects ever since. The availability of that RNA stock, as well as the small size of the library required to represent the repertoire of anti-fluorescein scFv domains, have made it practical for the students to complete their project in the 2 weeks available.

More generally, hyperimmunizing a few hens with a chosen antigen of interest, euthanizing them and harvesting their spleens and bone marrow, and constructing a small phage antibody library and affinity-selecting antibodies against the antigen or hapten using easily obtained supplies and standard protocols—that can be an attractive alternative to constructing a large all-purpose library (next subsection), if such a resource is unavailable or too costly to buy or construct.

All-Purpose Phage Antibody Libraries: Naive and Synthetic

All-purpose libraries are designed to represent such a diversity of independent clones that they are likely to include high-affinity antibody domains specific for almost any antigen of interest. There are two major types (Almagro et al. 2019): natural naive libraries, meant to express the entire natural antibody repertoire of many (sometimes as many as 200) donors (human or nonhuman) who have not been purposely immunized with an antigen of interest, and synthetic libraries, in which randomized synthetic complementarity-determining regions (CDRs) are installed in the backbone of an animal's (usually a human's) natural antibody variable regions.

A 2017 review summarizes anti-pathogen antibodies that have been affinity-selected from dozens of natural naive libraries assembled from RNA pooled from many healthy human donors (Chan et al. 2017). A series of articles by Haiyong Peng (University of Florida) and Rader details construction of naive human libraries in Fab format (see Protocol: Generation of Antibody Libraries for Phage Display: Chimeric Rabbit/Human Fab Format [Peng and Rader 2024a], Protocol: Generation of Antibody Libraries for Phage Display: Human Fab Format [Peng and Rader 2024b], Protocol: Generation of Antibody Libraries for Phage Display: Library Reamplification [Peng and Rader 2024c], Protocol: Generation of Antibody Libraries for Phage Display: Preparation of Electrocompetent E. coli [Peng and Rader 2024d], Protocol: Generation of Antibody Libraries for Phage Display: Preparation of Helper Phage [Peng and Rader 2024e], Protocol: Phage Display Selection of Antibody Libraries: Panning Procedures [Peng and Rader 2024f], and Protocol: Phage Display Selection of Antibody Libraries: Screening of Selected Binders [Peng and Rader 2024g]).

Like natural naive libraries, synthetic libraries do not entail immunizing live animals or humans. However, they also have a critical advantage: Antibodies against self-antigens are not censored by immunological tolerance. Tolerance might block discovery of high-affinity human antibody domains against human antigens, for instance. Human anti-human antibodies are an increasingly important class of pharmaceuticals.

Three articles by Sidhu, Veggiani, Maryna Gorelik (University of Waterloo), and Shane Miersch (University of Waterloo) review the “minimalist” approach to synthetic human libraries that the Sidhu laboratory has been pursuing for more than two decades (see Overview: Structural Survey of Antigen Recognition by Synthetic Human Antibodies [Gorelik et al. 2025], Introduction: Beyond Natural Immune Repertoires: Synthetic Antibodies [Veggiani and Sidhu 2024a], and Protocol: Generation and Selection of Synthetic Human Antibody Libraries via Phage Display [Veggiani and Sidhu 2024b]). The overall goal is to develop a practical way to construct libraries from which promising therapeutic antibodies for any chosen antigen can be affinity-selected. They focus on two desirable physical properties of such antibodies: high affinity for the antigen and high “developability,” meaning possession of properties such as solubility and stability that facilitate manufacture and favorable pharmacokinetic behavior. The defining feature of the minimalist approach is using an already existing therapeutic antibody with favorable developability as the sole scaffold for mutagenesis, thus greatly simplifying library construction and hopefully increasing the chance that antibody domains selected from the library will share the favorable developability of the chosen scaffold antibody. The scaffold for many minimalist projects has been the anti-HER2 antibody trastuzumab. Minimalism's key assumption is that high affinity can be achieved without exploiting the entire natural diversity of scaffolds in natural antibodies. This assumption has been largely vindicated in the sequel.

Minimalist libraries are constructed by introducing random mutations in strategically chosen codons in the scaffold antibody's six CDRs. One of the three articles mentioned above provides a general protocol for constructing such a library (see Protocol: Generation and Selection of Synthetic Human Antibody Libraries via Phage Display [Veggiani and Sidhu 2024b]):

• A “stop template” is cloned in a phagemid vector. The stop template encodes the scaffold sequence (e.g., trastuzumab), but with a stop codon at each position chosen for mutagenesis.

• The resulting vector is transfected into a mutant E. coli host that incorporates deoxy-U (dU) in place of many T nucleotides. The transfected cells are infected with helper phages and incubated in medium that selects for both phagemid and helper. Virions are prepared from the culture supernatant (there are no phagemid-encoded pIII molecules on these virions).

• Single-stranded plus-strand DNA is prepared from the virions; that DNA has numerous dU nucleotides in place of T nucleotides.

• Synthetic minus-strand mutagenic oligonucleotides, one for each of the six CDRs, are hybridized to the plus-strand virion DNA. These oligonucleotides have randomized sense codons opposite each of the stop codons in the plus strand. The minus strand is then closed by incubation with DNA polymerase and ligase. This reaction creates double-stranded heteroduplex DNA whose plus strand has numerous dU nucleotides and stop codons at the sites chosen for mutagenesis, and whose minus strand has no dU nucleotides and has randomized sense codons at the sites chosen for mutagenesis. Many (but not all) of the heteroduplexes will be covalently closed circles. This DNA is stored frozen in aliquots for use in preparing fresh libraries, as described in the next bullet.

• To prepare a new library, an aliquot of the heteroduplex DNA is electroporated into electrocompetent E. coli prepared from helper-infected cells. The electroporated cells are cultured in medium that selects for both helper and phagemid genomes. The cells do not incorporate dU nucleotides into DNA, and have an intact dU repair system that cuts the plus strand into small segments. The great majority of replicating DNAs, and of the DNAs in the secreted virions, are therefore derived from covalently closed circular minus strands, in which the stop codons have been replaced with randomized sense codons. Phages displaying antibody domains that bind a chosen antigen can then be affinity-selected from the resulting library.

By now, many human antibodies that have been affinity-selected from minimalist libraries are available for comparison to natural antibodies (see Overview: Structural Survey of Antigen Recognition by Synthetic Human Antibodies [Gorelik et al. 2025]). These authors compared 50 antibodies affinity-selected from minimalist libraries with 50 natural antibodies from the same V_H and V_κ subgroups. High-resolution structures of the antibody–antigen interaction were available for all 100 antibodies. By several broad criteria, the structures of the CDRs in the two types of antibody and the interaction of the CDRs with the antigen were strikingly similar. Figure 8 shows an example of the authors’ analysis. All CDR amino acids defined as “core” in the figure are represented in at least 60% of the structures analyzed. To facilitate graphical comparison, those core amino acids are mapped onto the backbone of the trastuzumab Fab (even though the backbone conformation differs significantly from one antibody to another). The size of each colored circle in the figure is proportional to the percent of structures in which that core amino acid participates significantly in the CDR–antigen interface; amino acids for which this percent was <20 are omitted. The striking similarity of the graphs for synthetic and natural antibodies implies that synthetic and natural CDRs interact with antigen in very similar ways. The authors argue that their results suggest considerable commonality of CDR structure, regardless of the backbone in which they reside, and that high-affinity antibodies can be selected from libraries in which only one of the natural antibodies’ myriad scaffolds is represented.

View larger version:

• In this window
• In a new window

Figure 8.

Graphical representation of the interaction between “core” complementarity-determining region (CDR) amino acids and antigen in synthetic and natural antibodies. The definition of core amino acids and their representation in the figure are explained in the text. Figure reprinted from Overview: Structural Survey of Antigen Recognition by Synthetic Human Antibodies (Gorelik et al. 2025) with permission from Cold Spring Harbor Laboratory Press.

A series of four articles by Mary Ann Pohl of Ailux Biologics and Juan Almagro of GlobalBio, Inc., describe an elaboration of the minimalist concept that they call semisynthetic libraries (see Protocol: Semisynthetic Phage Display Library Construction: Design and Synthesis of Diversified Single-Chain Variable Fragments and Generation of Primary Libraries [Almagro and Pohl 2024a], Protocol: Semisynthetic Phage Display Library Construction: Generation of Filtered Libraries [Almagro and Pohl 2024b], Protocol: Semisynthetic Phage Display Library Construction: Generation of Single-Chain Variable Fragment Secondary Libraries [Almagro and Pohl 2024c], and Introduction: Considerations for Using Phage Display Technology in Therapeutic Antibody Drug Discovery [Pohl and Almagro 2024]).

Such libraries are created in three stages:

• Construction of four primary libraries: Each of the four primary scFv libraries is constructed on one of four human V_κ scaffolds in the N-terminal position, connected by a flexible linker to a common human V_H scaffold in the C-terminal position. The V_H scaffold's V gene segment is followed by a mixture of 90 combinations of human diversity (D) and joining (J) gene segments. The sequence from the beginning of CDR3 to the end of FR4 (corresponding to the last codon of the V gene segment through the D and J gene segments) is a temporary placeholder that will be replaced in construction of the secondary libraries (last bullet). Hopefully, the very limited variation in the placeholder's sequence and length is sufficient for the purposes of “filtration” (next bullet). Meanwhile, random variation is introduced at strategically chosen codons in all three CDRs of the V_κ scaffolds and CDR1 and CDR2 of the V_H scaffold. The four primary libraries are assembled by chemical DNA synthesis and stored as frozen aliquots.

• Construction of four filtered libraries: The “filtered” libraries consist of members of the primary libraries whose scFv domains are resistant to heat shock and are therefore presumably well folded into a stable native three-dimensional structure. Construction begins by PCR-amplifying an aliquot of each of the four primary libraries, splicing the amplified DNA into the SfiI cloning sites of a pComb3-like phagemid vector, and propagating the library by electroporating the spliced DNA into an E. coli host and infecting with a helper phage. Partially purified phagemid virions are prepared from the four culture supernatants. The virions are then heat-shocked and cooled to allow renaturation of the displayed scFvs (the virions themselves are impervious to the heat shock). Superparamagnetic beads coated with protein L are used to affinity-purify virions from each of the four heat-shocked virion preparations, discarding virions that are not captured on the beads. Protein L binds FR1 of all four V_κ scaffolds in well-folded scFvs. The affinity-purified virions are therefore enriched for virions displaying particularly stable scFvs. Virions from each of the four libraries are released from the beads and propagated separately by infecting E. coli host cells. Aliquots of phagemid-bearing cells are frozen to serve as input to construction of the secondary libraries (next bullet).

• Construction of four secondary libraries: An aliquot of frozen phagemid-bearing cells from each of the four filtered libraries is thawed and cultured in medium that selects for the phagemid (the cells are not infected with helper phages, and the medium does not select for helpers). Phagemid DNA is extracted from the four cultures and partially purified. A secondary library is prepared from each of the four phagemid DNAs by replacing the 90 placeholder CDR3-FR4 sequences at the end of the V_H regions with a mixture of hundreds of millions of natural CDR3-FR4 sequences prepared from 200 adult human donors. Replacement is mediated by a highly conserved nucleotide sequence at the end of human V_H FR3. The conserved sequence serves as a PCR priming site for fusing the natural and filtered library sequences. Each of the four fused DNAs is spliced into the SfiI cloning sites of the pComb3-like phagemid vector, which is electroporated into E. coli host cells. The phagemid-bearing cells are infected with helper phages and cultured. Phagemid virions prepared from the culture supernatants comprise the four secondary libraries.

The resulting ensemble of four semisynthetic libraries is called “ALTHEA Gold Plus Library.” It is a successor to the “ALTHEA Gold Library,” an ensemble of two semisynthetic libraries, representing two, rather than four, human V_κ scaffolds (Valadon et al. 2019); for the ALTHEA Gold Library, filtration was accomplished with protein A rather than protein L.

High-affinity, conformationally stable scFvs have been affinity-selected from both the ALTHEA Gold Library (Valadon et al. 2019; Pedraza-Escalona et al. 2021; Dao et al. 2022a,b) and the ALTHEA Gold Plus Library (Guzmán-Bringas et al. 2023; Mata-Cruz et al. 2025). Mata-Cruz and her colleagues, for example, sought human scFvs that bind human CD36, a multifunctional membrane protein that is implicated in numerous diseases and is the target of drug development projects (Guerrero-Rodríguez et al. 2022). After three rounds of affinity selection using the extracellular domain of human CD36 as selector, they analyzed phagemids from 90 randomly selected clones. Fifty-seven of the phagemids, representing four distinct scFvs, bound strongly and specifically to the CD36 selector. All but 17 of the 90 phagemids, including all 57 that bound strongly to the selector, bound strongly to protein L, indicating that their displayed scFvs were predominantly in a natural immunoglobulin conformation. One of the four distinct scFvs, called D11, was shown to mimic the anti-metastatic effects of a mouse monoclonal antibody that binds both mouse and human CD36 (Pascual et al. 2017). These results commend D11 as a possible pharmaceutical lead. Whether or not that promise is realized, the results demonstrate the effectiveness of the Althea Gold Plus Library as a source of well-folded, high-affinity scFvs for a chosen selector.

Affinity Maturation

In the course of a natural antibody response, the affinity of the antibodies for the antigen increases progressively as a result of somatic mutation of the V regions in the responding B cells, along with clonal competition and increasingly stringent selection for high affinity by the antigen. An artificial analog of natural affinity maturation can be imposed on phage antibody clones that emerge from affinity selection with the antigen of interest (next section). DNA sequences from the selected clones are mutagenized at strategic positions and used to create a library of mutagenized phages. The mutagenized library is subjected to increasingly stringent affinity selection, in the hope of creating phage antibody clones with even higher affinity (Tiller et al. 2017).

Yield Versus Stringency

Theoretically, the yield of a phage clone from a round of affinity selection is the ratio of the number of virions of that clone in the input to that round to the number of virions of that clone in the unamplified output of that round. It is not generally possible to measure yield directly, but it is possible to estimate the relative yields of different phage clones bioinformatically, as explained in the final subsection of this section.

There are two types of yield:

• Specific yield is dependent on the identity of the displayed peptide. The obvious reason for high specific yield is that the clone in question displays a peptide with high affinity for the selector—exactly the desired property in an affinity selection project. Such specific clones are selector-specific, as are their yields. However, some clones owe their relatively high specific yields to something other than high affinity of their displayed peptides for the selector. One example is a clone whose displayed peptide binds the substrate, rather than the selector. Such clones are examples of selector-unrelated phages, or SUPs, which are the subject of the next subsection.

• Nonspecific background yield is independent of the identity of the displayed peptide. Some virions, whatever peptide they display, end up in the output just by chance, because no physical separation process is perfect. Nonspecific background yield is very low in affinity selection, but it is not zero.

Stringency is the degree to which a round of affinity selection favors phage clones whose displayed peptides bind the immobilized selector with high affinity over clones whose displayed peptides bind the immobilized selector with low affinity.

High yield and high stringency are both desirable characteristics of a round of affinity selection, but they are mutually antagonistic: High stringency almost always entails low yield and vice versa. High yield is mainly promoted by a high surface density of selectors on the substrate, while high stringency is promoted by a low surface density of selectors on the substrate. In the Phage Display Course, for example, stringency of selection for scFvs that bind the fluorescein (FL) hapten selector could be adjusted by immobilizing mixtures of FL-conjugated ovalbumin (FL-Ova) and unconjugated Ova in different proportions on the substrate surface. The smaller the proportion of FL-Ova, the higher the stringency.

High yield is imperative in the first round of affinity selection, when the input is a large library with billions of clones, even though that entails low stringency. Each individual clone in such a library is represented by a limited number of genetically identical virions—10 or 100, say. That includes the most desired clones: those displaying peptides that bind the selector with high affinity. If stringency is so high that not even a single virion from such a clone is represented in the unamplified output of the first round of affinity selection, that clone cannot magically reappear in subsequent rounds.

Stringency can be greatly increased in the second and subsequent rounds. That is because even when yield in the first round is as high as practical (i.e., stringency is as low as practical), the overall yield is still quite low. The number of clones represented in the output is almost never more than 0.01% of the number of clones represented in the starting library, and usually is much lower. That means that each clone can be represented by at least 10,000 times more identical virions in the input to the second round than in the input to the first round, assuming that the total number of virions in the inputs to the first and second rounds are the same.

Selector-Unrelated Phages, or SUPs

SUPs are unwanted phage clones that increase in prevalence with successive rounds of affinity selection for reasons other than affinity of their displayed peptides for the selector (Thomas et al. 2010). They can be classified into two types.

• Some SUPs reflect high specific yield during the capture or release stages of affinity selection (the capture, release, and amplification stages of affinity selection are described in the first paragraph of this section). For instance, the displayed peptide might bind some component of the affinity selection apparatus other than the immobilized selector—the substrate itself, for example.

• Other SUPs reflect overreplication during the amplification stage in each round of affinity selection. This unwanted overreplication is specific for the displayed peptide, but it is seldom caused by the displayed peptide. Much more often, it is the result of a property that is determined by genetic alterations that are linked to the coding sequence for the displayed peptide in the same phagemid chromosome.

I argue that with a few exceptions, measures to eliminate SUPs from a library have limited success (see Overview: Principles of Affinity Selection [Smith 2024]). A more promising general approach today is to identify SUPs bioinformatically, with the aid of high-throughput sequencing (HTS) of the clones that emerge from an affinity selection project. HTS is already increasingly favored as the “primary readout” in affinity selection projects; that is, the first stage in analysis of the selected clones. The role of HTS in identifying phage clones of interest, while eliminating SUPs from consideration, is the subject of the next subsection. That subsection discusses HTS in the context of affinity selection from phage antibody libraries, but its application to other types of libraries and selectors should be apparent.

High-Throughput Sequencing (HTS) as Primary Readout in Affinity Selection Projects

A series of articles by Brandon DeKosky and his colleagues at the Massachusetts Institute of Technology introduces HTS (see Introduction: Beyond Single Clones: High-Throughput Sequencing in Antibody Discovery [Fahad et al. 2025b]) and details its use in antibody engineering (see Protocol: Antibody Data Analysis from Diverse Immune Libraries [Fahad et al. 2025a], Protocol: Clonal Lineage and Gene Diversity Analysis of Paired Antibody Heavy and Light Chains [Fahad et al. 2025c], and Protocol: Clonal Variant Analysis of Antibody Engineering Libraries [Fahad et al. 2025d]).

HTS can deliver partial sequences of the antibody domains displayed by millions of virions in the output of a round of affinity selection. That is usually enough for a thorough census of the selected antibody domains in the outputs of the first and second rounds of affinity selection, giving the number of times each clone occurs; that number is referred to as the clone's reads. A clone's reads divided by the sum of the reads for all clones in the output is a measure of the clone's prevalence. The amplified output of the first round of affinity selection is the input to the second round. Therefore, the ratio of a clone's prevalence in the amplified output of the second round of affinity selection to its prevalence in the amplified output of the first round is a measure of the clone's yield in the second round of affinity selection relative to the yields of the other clones in that round. If, following the advice in the first subsection above, stringency is greatly increased in the second round, clones whose displayed antibody domains bind the selector with high affinity are strongly favored over clones displaying low-affinity antibody domains. This profile of the clones’ relative yields in the second round may be enough to choose the most promising clones for further investigation. Promising clones can be recovered from the second-round output by PCR with clone-specific primers.

Not all clones with high second-round yields necessarily display antibody domains with high affinity for the selector. Some of them might be SUPs, as explained in the previous subsection. Unwanted SUPs can be distinguished from desirable high-affinity clones if affinity selections with multiple selectors are carried out in parallel from a common library. The parallel selections should be as identical as possible, apart from the identity of the selectors. SUPs reveal themselves because their yield is high in all (or almost all) the parallel selections, regardless of the selector.

Epitope Discovery

“Epitope discovery”—identification of peptides or protein domains that are recognized by antibodies—has been an important application of phage display technology from its earliest days (Cortese et al. 1994). Antibodies—either monoclonal or polyclonal—are used to affinity-select antibody-binding peptides from two types of phage display library: random peptide libraries (RPLs) and natural peptide libraries (NPLs). RPLs display synthetic peptides with randomized sequences; NPLs display the entirety or fragments of natural proteins.

Two articles by Gregg Silverman (see Introduction: Insights from the Study of B-Cell Epitopes of a Microbial Pathogen by Phage Display [Silverman 2025a] and Protocol: Cloning and Selection from Antigen Fragment Libraries for Epitope Identification [Silverman 2025b]), summarizing recent work by his group at New York University (Hernandez et al. 2020a,b), exemplify epitope discovery from NPLs. His laboratory has developed a suite of techniques for rapid identification of short peptide epitopes, for potential incorporation into modular vaccines against infectious diseases. To this end, genes for target pathogen proteins are randomly fragmented into small DNA pieces, which are cloned into a pComb-derived vector called pComb-Opti8. This vector displays the cloned peptides on subunits of the major coat protein pVIII through a flexible linker, rather than on subunits of minor coat protein pIII (as illustrated in Fig. 4). The resulting NPLs include DNA inserts in all possible reading frames in both orientations, but there is sufficient redundancy for many thousands of in-frame inserts, spanning the entire protein-coding sequence, to be represented. Monoclonal or polyclonal antibodies of interest are then used to affinity-select phage clones displaying protein fragments that bind those antibodies. Individual affinity-selected clones can be sequenced to identify the selected epitope-containing peptides. Alternatively, and increasingly, outputs of affinity selection are analyzed en masse by HTS, as outlined in the final subsection of the previous section.

The investigators’ “proof-of-principle” project targets the leucocidins, which are major virulence factors in invasive infection by Staphylococcus aureus, including by methicillin-resistant S. aureus (MRSA). The leucocidins, reviewed by Spaan et al. (2017), are eight-member β-barrel rings of alternating type S and type F subunits. Such rings kill cells in infected hosts by creating pores in the cell membrane, thus breaching the barrier between inside and outside. There are multiple leucocidins, with different type S and type F pairs. In most cases, the two types of subunits are secreted separately. The type S subunit binds a receptor on a target cell; then, additional type S subunits and type F subunits assemble into the β-barrel pore at the receptor site. Different leucocidins target different cell types, depending on the receptors they bind. Receptors have been identified for the major leucocidins, and X-ray crystallographic structures are available for many leucocidins and individual type S and type F subunits.

Two illustrative articles from Silverman's group report construction of gene fragment libraries from two type S subunits in the pComb-Opti8 vector, and affinity selection of phage-borne peptides from each library, using as selectors a mouse monoclonal antibody elicited by immunization with the respective unfragmented parent type S subunit (Hernandez et al. 2020a,b). Clone-by-clone sequencing (Hernandez et al. 2020b) or HTS analysis (Hernandez et al. 2020a) of the affinity-selected clones from each library identified a minimal epitope recognized by the selecting monoclonal antibody. Both epitopes were also recognized by antibodies from patients recovering from invasive S. aureus infection. The availability of the X-ray crystallographic structure of the type S subunits allows the minimal epitopes to be located in the native folded proteins. One epitope corresponds to a solvent-exposed β-turn at the tip of a β-hairpin (Hernandez et al. 2020a); the other corresponds to an entire solvent-exposed β-hairpin (Hernandez et al. 2020b).

Remarkably, when mice were immunized with these small peptides conjugated to keyhole limpet hemocyanin, they produced antibodies that bind not only the immunizing peptides, but also the intact parent proteins from which the peptides derive. This ability to elicit antibodies that react with the target pathogen protein in its native state has been called immunogenic fitness (Matthews et al. 2002); it is presumably a necessary condition for almost any effective vaccine component. It is not clear how general strong immunogenic fitness will be for short peptide epitopes affinity-selected from fragment libraries. For some parent proteins, many or most epitopes may be discontinuous, composed of amino acids from distant parts of the linear amino acid sequence. In any case, the logic of these investigators’ strategy does not depend on the size distribution of the protein fragments. A gene could be randomly fragmented into much larger pieces, some of which happen to encode peptides that fold into natural subdomains that are present on the intact protein. Discontinuous epitopes on such subdomains are likely to be shared by the pathogen or pathogenic protein itself, and therefore have good prospects of strong immunogenic fitness.

Display of Nonantibody Antigen-Binding Domains

Antibody V regions are folded β-sheets supporting three loops with diverse amino acid sequences at one end. The β-sheet is conformationally stable, serving as a secure “scaffold” that keeps the loops in place despite their diversity. It is the diversity of the loops that accounts for the antibodies’ ability to bind specifically to an enormous diversity of antigens. The loops that mediate binding by an antibody are called complementarity-determining regions (CDRs), because it is their sequence diversity that confers an almost unlimited diversity of complementarity (i.e., antigen-binding specificity) on antibodies.

Several stable nonantibody scaffolds have been explored for construction of “phage nonantibody libraries” (as I call them here) representing a great diversity of binding sites. From such libraries, specific, high-affinity ligands for a great diversity of selectors can be affinity-selected. I call the selectors “antigens” here to highlight the parallelism between nonantibody binding domains and canonical antibodies. Such nonantibody antigen-binding domains, like antibody V regions, are composed of a structurally stable scaffold holding in place CDRs with highly diverse sequences and antigen-binding specificities.

Akiko Koide and Shohei Koide of New York University Grossman School of Medicine summarize the construction and use of monobodies, whose scaffolds are fibronectin type III (FN3) domains (see Overview: Use of Phage Display and Other Molecular Display Methods for the Development of Monobodies [Koide and Koide 2024]). The FN3 motif is widespread in nature: Humans produce 4104 of them in 673 proteins, a few of which have served as scaffolds for monobodies. Like antibody V regions, FN3 domains are folded β-sheets, but with seven rather than nine β-strands. As in the case of antibody V regions, there are three loops at one end of the folded sheet, connecting pairs of anti-parallel β-strands. The amino acid sequences in these loops can vary greatly without weakening the folded β-sheet structure; that sequence diversity gives monobodies the ability to bind specifically to a great diversity of antigens. Unlike an antibody V region, a monobody is not paired with another monobody in a stable superstructure comparable to a Fab. They are expressed as stand-alone antigen-binding domains, like the VHH domains in some camelid antibodies (Muyldermans 2013).

The investigators set out to display monobodies fused to the C-terminal domain of filamentous phage pIII protein, using a phagemid vector in the pComb3 family (see “Phagemid Vectors for Phage Antibody Display”), and have explored other display formats as well. Efficient display required changing pComb3's ompA signal peptide to a signal peptide dependent on the signal recognition particle (SRP), so that the protein was secreted through the inner membrane cotranslationally, rather than post-translationally. They argue that this change was necessitated by one of the FN3 domain's chief attractions: its extraordinary noncovalent stability, which is not reinforced by an intradomain disulfide bond. Rapid formation of the folded structure in the cytoplasm would arguably interfere with post-translational secretion.

Stefan Ståhl and his colleagues at KTH Royal Institute of Technology in Stockholm have authored articles introducing affibodies, another type of nonantibody antigen-binding domain (see Protocol: Selection of Affibody Molecules Using Escherichia coli Display [Dahlsson Leitao et al. 2024], Protocol: Selection of Affibody Molecules Using Phage Display [Hjelm et al. 2024], Protocol: Selection of Affibody Molecules Using Staphylococcal Display [Löfblom et al. 2024], Protocol: Cloning of Affibody Libraries for Display Methods [Ståhl et al. 2024a], and Introduction: Engineering of Affibody Molecules [Ståhl et al. 2024b]). Here, the scaffold is a three-helix bundle, and the CDRs are not loops protruding from one end of the bundle, but rather 13 amino acids in helices 1 and 2 that protrude from one side of the bundle and are not required for its structural integrity. The investigators explain how to display affibodies on a pComb3 family phagemid vector, on the surface of E. coli bacteria, and on the surface of Staphylococcus carnosus bacteria. With only 58 amino acids, affibodies are smaller than any other antigen-binding domain, yet they can have subnanomolar and even subpicomolar affinities for their target antigens. These features commend them for biomedical applications requiring rapid and long-lasting binding to antigens in human subjects, along with rapid clearance of unbound molecules. For example, affibody ABY-25 has 76 pm affinity for the breast cancer marker HER2 (Feldwisch et al. 2010). When conjugated to the chelator DOTA and labeled with the positron-emitting radionuclide ⁶⁸Ga, whose half-life is only 68 min, it allows metastatic lesions to be imaged at high resolution in HER2-positive breast cancer patients by positron emission tomography (Velikyan et al. 2019; Alhuseinalkhudhur et al. 2020).

Display of Specialized Protein Domains

Gregory Martyn of the University of Toronto and Veggiani describe libraries of modified Src homology 2 (SH2) domains, which bind phosphorylated tyrosine (pTyr) residues on specific receptor tyrosine kinases or other proteins (see Overview: Phage-Displayed SH2 Domain Libraries: From Ultrasensitive Tyrosine Phosphoproteome Probes to Translational Research [Martyn and Veggiani 2024]). The human genome includes 122 of these ∼100-amino-acid domains, present on intracellular signal transduction proteins. When an SH2 domain binds pTyr, the protein that it is part of is triggered to set in train a signal transduction cascade. Being able to exploit this specificity to detect or modify the behavior of specific regulatory pTyr residues could have many potential biomedical applications. However, although SH2 domains generally bind specifically to their cognate pTyr residues, their affinity is generally too weak for most applications. The article describes procedures for randomizing strategic codons in an SH2-coding sequence, using phage display to affinity-select variants with higher affinity or altered specificity.

An article by Chen T. Liang, Olivia Roscow, and Wei Zhang summarizes affinity selection of high-affinity ligands for chosen selectors from libraries of ubiquitin (Ub) variants (see Overview: Generation and Characterization of Engineered Ubiquitin Variants to Modulate the Ubiquitin Signaling Cascade [Liang et al. 2024]). Ub is a highly conserved 76-amino-acid protein present in virtually all eukaryotic cells. The article emphasizes how detailed knowledge of the structure and properties of Ub can guide affinity selection projects. Ub might well have been included in the previous subsection, as yet another attractive scaffold for “phage nonantibody libraries,” as I call them there. Following the pioneering work of the Sidhu laboratory (Ernst et al. 2013), however, almost all proteins chosen as selectors for Ub variants already interact specifically with Ub (Tang et al. 2023).

Ub is central to a sprawling network of cellular functions.

• Ubiquitination—enzymatic covalent conjugation of Ub to target proteins, which are called “substrates”—is at the heart of the Ub system. It is carried out by an enzyme cascade composed of three types of enzymes acting in succession: E1 activating enzymes, followed by E2 conjugating enzymes, followed by E3 ligating enzymes. The human proteome includes two E1s (Barghout and Schimmer 2021), about 40 E2s (Stewart et al. 2016), and well over 600 E3s (Yang et al. 2021). It is the E3 ligating enzymes that are responsible for conjugating Ub to specific substrate proteins, creating conjugates with single Ubs, or branched or unbranched chains of Ubs. The combinations of substrates that can be ubiquitinated by the E3 ligating enzymes have only begun to be mapped (Suiter et al. 2025).

• Countering ubiquitination are about 100 deubiquitinating enzymes with different specificities (Snyder and Silva 2021).

• Ubiquitinated proteins are recognized by a plethora of Ub-binding effector proteins, triggering myriad physiological responses. Prominent among the effector proteins is the proteasome, which degrades the ubiquitinated proteins it recognizes (Ciechanover and Stanhill 2014; Grice and Nathan 2016); that indeed was the context in which Ub was first discovered. However, there are an unknown number of other effector proteins, mediating a largely unknown array of responses. Recognition of ubiquitinated proteins by effector proteins is primarily mediated by numerous small Ub-binding domains (UBDs), falling into 10 or 11 classes (Hurley et al. 2006). Effector proteins typically have multiple UBDs. Using Ub variants to probe specific effector functions is a major goal of this arena of investigation.

Affinity selection of Ub variants that bind specific elements of the Ub network is a promising new way to interrogate a great diversity of cellular functions. Binding of Ub-binding proteins to wild-type Ub is specific but weak; it seldom interferes with affinity selection of high-affinity ligands for specific Ub-binding proteins from libraries of Ub variants.

A project by Veggiani and his colleagues (Veggiani et al. 2022) exemplifies application of the principles outlined (see Overview: Generation and Characterization of Engineered Ubiquitin Variants to Modulate the Ubiquitin Signaling Cascade [Liang et al. 2024]). They prepared a panel of 60 different selectors, representing the Ub-interacting motif (UIM) class of UBDs (last bullet in the previous paragraph) from human proteins. Such UIMs bind weakly, but specifically, to wild-type Ub, with dissociation equilibrium constants in the range K_D = 100–1000 μm. The selectors were used to affinity-select ligands from a library of Ub variants displayed on phages. The library was designed in light of the contact amino acids in the known structure of a Ub variant in complex with a UIM from yeast. Altogether 23 Ub amino acids were chosen for “soft” randomization, such that at each randomized position, ∼50% of the codons would encode the wild-type amino acids, while the other 50% would encode other amino acids. The resulting library contained 17 billion different Ub variants. A total of 138 distinct Ub variants, affinity-selected by 42 of the 60 UIM selectors, bound their selectors well above background levels. Further analysis of the most selective variant for each of the 42 selectors revealed high affinity and specificity for their cognate selectors:

• Phage ELISA showed that 26% of the 42 variants recognized only their cognate selectors, and 52% recognized three or fewer selectors.

• Competitive phage ELISA showed that 57% of the variants gave IC₅₀ values <100 nm with their cognate selectors, while wild-type Ub gave IC₅₀ values between 100 and 500 μm.

These results indicate that many UIMs, involved in a wide variety of physiological functions, can be specifically addressed with affinity-selected Ub variants. It seems likely that this approach will apply to UBDs other than UIMs as well. Design of a Ub variant library for each class of UBD can be guided by available structures of Ub in complex with members of that UBD class. A vast array of cellular effector functions would thus be opened up for detailed investigation.

The UIM project seems tailor-made for using HTS, rather than clone-by-clone analysis, as the first step in analyzing the output of affinity selections, as recommended above in “High-Throughput Sequencing (HTS) as Primary Readout in Affinity Selection Projects.” Thus, two rounds of affinity selection with all 60 UIM selectors would be carried out in parallel, stringency being greatly increased in the second round. HTS of the 120 first- and second-round outputs would allow clones of interest, with high specific affinity for their cognate UIM selectors, to be distinguished from the expected selector-unrelated phages (SUPs): phages that bind the carrier or other elements of the selection apparatus, phages that overreplicate during amplification, and phages that bind off-target UIMs. This line of experimentation would arguably be a streamlined way to choose a small number of clones for detailed individual investigation.