Discovery of Therapeutic Antibodies Targeting Complex Multi-Spanning Membrane Proteins

Abstract

Complex integral membrane proteins, which are embedded in the cell surface lipid bilayer by multiple transmembrane spanning polypeptides, encompass families of proteins that are important target classes for drug discovery. These protein families include G protein-coupled receptors, ion channels, transporters, enzymes, and adhesion molecules. The high specificity of monoclonal antibodies and the ability to engineer their properties offers a significant opportunity to selectively bind these target proteins, allowing direct modulation of pharmacology or enabling other mechanisms of action such as cell killing. Isolation of antibodies that bind these types of membrane proteins and exhibit the desired pharmacological function has, however, remained challenging due to technical issues in preparing membrane protein antigens suitable for enabling and driving antibody drug discovery strategies. In this article, we review progress and emerging themes in defining discovery strategies for a generation of antibodies that target these complex membrane protein antigens. We also comment on how this field may develop with the emerging implementation of computational techniques, artificial intelligence, and machine learning.

Key Points

Complex multi-spanning membrane protein targets offer significant opportunities for therapeutic biologics.

Discovery of therapeutic antibodies targeting complex membrane proteins is technically challenging, due largely to the properties of complex integral membrane proteins and the complexity of producing these targets as antigens.

Significant progress is being made in this area by employing new methods to address these technical challenges and new and varied strategies for antibody discovery.

Introduction

Monoclonal antibodies (mAbs) are an established and successful class of therapeutics that exhibit clinical efficacy in the treatment of disease in the areas of oncology, immune disorders, infection, and metabolic, haematological, and neurological diseases [1]. The growth in approved antibody therapeutics is illustrated by a reported 162 antibodies approved by at least one regulatory agency (as reported June 2022), with 16 antibody therapeutics gaining first approval in 2023 [2, 3]. The growth of mAbs and related biologics as effective therapeutics is based on the variety of mechanisms of action achievable, specificity to their target epitopes, and high efficacy, leading to antibodies being generally well tolerated, with a lower risk of unanticipated safety issues than other therapeutics [4]. The effectiveness of antibodies with specific mechanisms of action can be accessed by tailoring discovery strategies and molecular engineering of antibodies to target desired epitopes or to introduce new properties. Antibodies can bind to and neutralise target proteins to block function (e.g. cytokines) and inhibit cellular signalling pathways [4]. Antibodies may bind cell surface receptors and inhibit the action of native ligands or directly agonise signalling pathways [4]. Antibodies can bind cell surface protein targets and engage Fcγ receptors (FcγR), via their Fc domains, to drive antibody effector functions such as antibody-dependent cellular cytotoxicity (ADCC) and antibody-dependent cellular phagocytosis (ADCP) to cause cell killing and depletion [4]. Cell killing can also be achieved by conjugating antibodies to a toxic chemical payload or radionuclide to give rise to an antibody–drug conjugate (ADC) or radionuclide–antibody conjugate (RAC), respectively, or by engineering Fc domains to improve affinity for FcγR to enhance effector function [5, 6]. In addition to the mechanisms of action of individual mAbs, they also have a long duration of action when administered, due to their long plasma half-life (~ 11–30 days), which can be extended by Fc engineering to enhance binding to neonatal Fc receptors (FcRn) to exploit the FcRn-mediated recycling pathway [7, 8]. mAbs have been particularly effective in oncology (where antibody effector functions or ADC toxins eliminate tumour cells) and immunological diseases via the inhibition of inflammatory signalling pathways or elimination of cells driving disease [1]. Further opportunities exist following development of novel formats such as bispecific antibodies, in which two binding modes are present in a single molecule. This can involve binding to two target receptors on a cell to inhibit signalling of both receptors, or binding to a tumour-associated antigen and a target protein expressed on an effector cell that allows recruitment of an effector function (e.g. CD3 receptor on T cells) [9].

Antibodies developed to date for clinical use have mainly targeted soluble ligands (e.g. TNFα, interleukin (IL)-5, IL-6, IL-17, vascular endothelial growth factor (VEGF)-A, thymic stromal lymphopoietin (TSLP)), cell surface receptors (e.g. epidermal growth factor receptor (EGFR), programmed cell death protein (PD)-1, PD-L1, IL-2R, human epidermal growth factor receptor (HER)2), and viral proteins (e.g. respiratory syncytial virus (RSV) F-protein, SARS-CoV-2 spike protein) [2, 10-12]. In the case of cell surface transmembrane receptors, these are often type I or II membrane proteins with large extracellular ligand-binding domains with a single transmembrane helix. There is, however, significant interest in discovering antibodies targeting complex polytopic integral membrane proteins where they are identified as important potential therapeutic targets [13, 14]. These complex membrane proteins (CMPs) are characterised by multiple hydrophobic transmembrane spanning domains and include G protein-coupled receptors (GPCRs), ion channels, transporters, tight junction proteins (e.g. claudins), and enzymes that are members of large protein families [15-17]. Taking GPCRs as an example, analysis of the human genome predicts approximately 1000 genes and the existence of approximately 350 non-olfactory GPCRs [18]. Targeting GPCRs with small molecules is a proven strategy, as illustrated by the clinical success of targeting GPCRs. A recent review showed that small-molecule drugs targeting GPCRs represent the largest class of marketed drugs, with around 34% of Food and Drug Administration (FDA)-approved drugs targeting GPCRs [19]. Small-molecule drugs can, however, face challenges in progress to the clinic, due to toxicity and off-target effects.

These target classes present an opportunity to exploit the excellent specificity of antibodies and other features such as their long serum half-life, ability to engage effector function(s) to drive ADCC and ADCP, and modulation of properties such as affinity and Fc effector functions by protein engineering [20]. GPCRs, ion channels, and transporters typically exhibit multiple conformational states and can bind a wide variety of small-molecule, peptide, and protein ligands (including toxins) and, in the case of ion channels, respond to changes in membrane potential. Antibodies that bind these targets may be selected to act by a variety of mechanisms: antibodies may act as antagonists or agonists (particularly where GPCRs and ion channels interact with native ligands and toxins) or by binding to conformational states of the proteins that have a desired pharmacological effect. mAbs may also be used to exploit mechanisms of action that involve cellular depletion or elimination by binding to specific cells expressing a target protein and recruitment of effector functions mediated by customised antibody Fc domains [21]. This potential is reflected in a growing pre-clinical and clinical pipeline of mAbs and alternative protein scaffolds targeting CMPs such as GPCRs and claudins [14-16, 22]. To date, however, there are few approved and marketed biologics addressing these target classes. Taking GPCRs as an example, there are only three antibodies in clinical use in 2024. Mogamulizumab, an afucosylated anti-C-C chemokine receptor (CCR)4 mAb, received initial approval for clinical use in Japan in 2012, followed by approvals in the United States (US) and Europe [23-25]. This humanised anti-CCR4 antibody causes depletion of CCR4-expressing cells in patients with T cell lymphomas by enhanced natural killer cell activation enabled by reduced Fc fucosylation. Erenumab was approved in the US as a first-in-class mAb that targets the calcitonin gene-related peptide receptor (CGRP-R) complexed to receptor activity-modifying protein 1 (RAMP1) and is being used clinically as a therapeutic for migraine [26]. Most recently, talquetamab received approval in 2023 for the treatment of multiple myeloma. Talquetamab is a bispecific T cell engager that targets GPRC5D, a GPCR, and CD3 [27]. Finally, there are several mAbs approved that target the multi-spanning membrane protein CD20 (e.g. rituximab, ocrelizumab, ofatumumab) and are used to deplete tumour cells [28]. These approvals demonstrate the potential of developing clinically effective mAbs against these target classes. However, there are no approved antibody therapeutics to date that target ion channels, reflecting the technical challenges of isolating effective and developable mAbs against these target classes. Technical issues arise due to challenges in expressing, purifying, and stabilising multi-spanning membrane proteins to produce pure, conformationally relevant protein antigen and due to immunological tolerance arising from the high sequence conservation of some proteins across species. That said, our knowledge of CMPs is accelerating due to the development and application of new technologies for preparing membrane proteins and increasing knowledge of protein structure driven by advances in structural techniques such as cryogenic electron microscopy (cryo-EM), which has been particularly effective in solving structures for membrane proteins and requires less protein than crystallographic methods [29, 30]. Advances in computational techniques such as the application of AlphaFold2, RoseTTAFold2, and ESMFold are also driving improved methods for modelling integral membrane protein structures [31]. In this review, we explore the technical challenges in developing mAbs, and related protein scaffolds, that bind CMPs, and focus on the technologies and options available to facilitate membrane protein antigen preparation and strategies for antibody discovery to isolate antibodies that bind and allow modulation of the function of these targets.

Generation of Complex Membrane Protein Antigens: A Key Challenge for Therapeutic Antibody Discovery

Therapeutic antibody discovery requires the application of established and emerging antibody discovery platforms. These can broadly be separated into two approaches: immunisation of laboratory animals with a target antigen, exploiting the natural immune response to generate antibodies, and in vitro technologies such as phage, yeast, and mammalian display. Identification of antibodies to multi-spanning membrane proteins presents challenges due to technical issues in expression, purification, and formulation of integral membrane proteins to use as antigens. In some cases, extracellular N- or C-terminal domains are present, which may be expressed and purified as functional, correctly folded domains. In other cases, extracellular loops of the membrane proteins, which can vary in length, can provide limited potential epitopes (see Figure 1). Heterologous expression of the full-length membrane protein often requires empirical screening in multiple expression systems to obtain a high expression level (see for example, McCusker et al. [32] and Kaipa et al. [33]). Options for preparation of membrane protein antigens for antibody discovery will vary with the choice of discovery platform employed (as this may dictate formats of antigen that can be used) and the very diverse structures observed for integral membrane protein families, which can influence the choice of heterologous expression methods (see Fig. 1). For example, some membrane proteins have large extracellular domains (ECDs) that can be targeted, whilst others have small surface-exposed loops that provide limited potential epitope exposure and are challenging to target. Structures of antibody–membrane protein complexes are, however, revealing how antibodies (and related protein-binding scaffolds) bind these proteins (see examples in Fig. 1). In the following sections, we review methods available for isolation of membrane protein antigens in formats suitable for antibody discovery. We also review strategies that can be employed to isolate functional antibodies.

Fig. 1.

Examples of the structural diversity of complex multi-spanning membrane proteins and the impact on antibody isolation challenges. The structures of complex integral membrane proteins selected illustrate the diverse topologies and variety of potential surface extracellular epitopes accessible to antibodies. Structures were generated in PyMOL (Schrӧdinger) using the following Protein Databank (PDB) entries: P2X4, 4DW1 [34]; apelin receptor, 6KNM [35]; glucagon-like peptide 1(GLP-1) receptor, 6LN2 [36]; eel sodium channel in complex with beta subunit, 5XSY [37]; claudin-4 complexed with enterotoxin, 8U4V [38]; potassium channel Kv1.3, 7SSZ [39]. GPCR structures (apelin receptor and GLP-1 receptor), claudin-4, and Kv1.3 were selected to show complexes with sdAb, nanobody, and Fab bound as labelled in the figure (chains coloured pale cyan). Note that four nanobodies are bound to Kv1.3, one per monomer of the tetrameric ion channel. Amino acid residues exposed on the extracellular surface of the membrane proteins that could potentially be epitopes are shown in blue, and hydrophobic transmembrane and intracellular regions are shown in pale green. The beta subunit and enterotoxin are both depicted as light orange. The multi-pass membrane protein residues were coloured according to transmembrane and extracellular regions as defined in their respective UniProt entries for the membrane proteins. The plasma membrane is represented by the grey lines. Key challenges to antibody discovery, which will be target protein dependent, are highlighted in the box. GPCR G protein-coupled receptor, sdAb single-domain antibody

Heterologous Expression Systems for Complex Membrane Protein Production

A critical step for an antibody discovery campaign is the generation of high-quality, functionally validated target protein that can be used as an antigen and a source of the target protein in a format suitable for screening antibodies for relevant binding and functional properties. For therapeutic antibody discovery, a key initial step involves bioinformatic sequence analysis of the target protein to determine the most relevant common human variant target sequence for use in drug discovery. This step would also evaluate whether the target has sequence variants that may be important (e.g. splice variants) and options for generation of the target based on the available sequence and structural information considered. Bioinformatics servers and databases such as UniProt, ENSEMBL, Protein databank (PDB), PsiPred, PredMP, and Orientation of Proteins in Membranes (OPM) may be used to facilitate this analysis [40-46].

Synthetic peptides derived from the target protein sequence have been successfully used in antibody discovery and should be considered a valid approach on a case-by-case basis (reviewed in Lee et al. [47]). We have focused this review on strategies to produce intact, natively folded target proteins. Membrane proteins are typically natively expressed at low levels, and their hydrophobic properties make protein production difficult. To provide CMP in the milligram quantities required for drug discovery, recombinant protein expression requires overexpression of the target proteins in a suitable host. Heterologous overexpression of CMPs is challenging and requires consideration of options for expression vector construct design, choice of host cells, optimisation of expression conditions, and development of methods to solubilise the protein from the cell membrane, purify the protein, and stabilise the protein outside of its native lipid environment. This typically requires significant effort to investigate options and optimise methods. Recent in-depth reviews explore the strategies employed to express and purify integral membrane proteins [48-50], and Table 1 highlights commonly used key host systems for heterologous expression and their advantages and challenges. Furthermore, cells overexpressing the target protein of interest can themselves be used as a source of antigen for both in vivo and in vitro antibody discovery and can be directly used as screening tools to identify antibodies that bind target protein in its native membrane context. Cells expressing the target protein are also used to screen antibody function using a variety of read outs (e.g. fluorescence-based cell binding, reporter assays, patch-clamp electrophysiology, radioactive ligand-binding or radioactive ion flux assays) as reviewed by Colley et al. [51].

Table 1.

Commonly used expression systems for recombinant protein production. This table provides an overview of the expression hosts commonly used in both academic and pharma/biotechnology settings

Expression host	Host types	Example uses	Advantages	Challenges
Bacteria	Escherichia coli	Expression of ECDs	Rapid growth, simple growth conditions, cost effective	Can result in inclusion bodies; very limited post-translational modifications
Yeast	Saccharomyces cerevisiae, Pichia pastoris	Suitable for soluble and membrane protein expression	Fast growing, relatively simple media, can control expression	Difficult cell disruption; high mannose glycosylation; hyperglycosylation
Insect cells	Sf9, Sf21, Hi5	Expression of complex proteins; expression of multiple subunits/domains	Effective for complex proteins; correct folding and some post-translational modification	Time-consuming methods compared to E.coli; higher cost for culture; insect cell glycosylation pattern different to mammalian pattern
Mammalian	HEK, CHO	Production of therapeutic proteins; excellent for complex proteins	Transient or stable options for expression; human-like glycosylation and post-translational modifications; scalable	Longer production time compared to some hosts; expensive media and complex culture conditions

CHO Chinese hamster ovary, ECD extracellular domain, HEK human embryonic kidney

CMPs derived from eukaryotes are typically not well suited for heterologous expression in bacterial hosts such as Escherichia coli (E.coli). This can be due to the absence of native post-translational modification (PTM) such as glycosylation, differences in the bacterial lipid composition, and sub-optimal protein folding due to the reducing cell environment and lack of appropriate chaperones [52]. Despite limitations, examples of successful bacterial expression of eukaryotic proteins exist, and a recent approach using directed evolution in E. coli to derive higher levels of GPCR expression and stability has been successful [53-55].

Insect cell expression of CMPs using baculovirus has been valuable, particularly for GPCRs [56]. Insect cell expression supports high recombinant protein expression levels, large-scale culture, and PTMs such as glycosylation, albeit with simpler oligomannosidic glycosylation. Recently, a plasmid-based, baculovirus-free protein expression method was described that reduces overall timelines, as virus production is not necessary. The expression vector is delivered using polyethyleimine (PEI), and the insect cells retain their secretory and membrane protein transport pathways, normal cell growth, and viability [57]. This direct insect cell expression method was used to successfully express SARS-CoV-2 antigen. Whilst insect cells are a well-established system, it is important to assess protein function and quality, as high-level expression does not always equate to functional, correctly folded protein. This has been demonstrated experimentally for the GPCR, human angiotensin II type 1 receptor, and the serotonin transporter [58] and for the human Ether-a-go-go-related gene (hERG) ion channel, where comparison of different expression hosts identified human embryonic kidney (HEK) expression as the best host for functional protein purification [59].

Expression in yeast has also been a successful strategy for membrane protein expression and allows scalable production of biomass to high levels [60]. It is also possible to perform directed evolution in yeast to isolate proteins with improved levels of expression and stability, and interestingly, the increased expression observed in yeast is maintained when evolved constructs are transferred to insect cells for expression, with up to a 26-fold increased functional protein expression reported [61].

For therapeutic antibody discovery, expression of human target protein antigens in mammalian cells typically involves use of HEK or Chinese hamster ovary (CHO) cells as the most prevalent hosts. Membrane protein expression can be achieved by transient transfection of cells using a transfection reagent (e.g. PEI, lipofectamine) and expression vector or by stable expression following genomic integration of the transgene [62, 63]. Each method has advantages and challenges, and both typically require evaluation of several expression constructs to identify optimal expression levels. Several commercial host cells for transient transfection are available, with HEK EXPI293 and EXPICHO being widely adopted in pharma and biotech companies [64]. These platform technologies show demonstrably higher protein expression levels than HEK293 and CHO-S host cells but are relatively costly to deploy. Generation of stable cell lines that overexpress CMPs presents an alternative to transient expression methods. We routinely use recombinant lentivirus to generate stable cell pools that can be used to isolate clonal cell lines expressing the gene of interest at a high level (reviewed by Elegheert et al. [65]). The plasma membrane may accumulate large numbers of target protein molecules, and we have observed 50,000 to 3 million target proteins per cell with expression levels being target protein dependent (as measured using Bangs Laboratories calibration beads—unpublished observation). We have also used the Jump-In™ cell system (Thermo Fisher), which allows targeted integration of genes into cells with an engineered specific genomic loci to overexpress CMPs (e.g. formyl peptide receptor (fMLP) receptor) [66]. One issue observed upon high-level, constitutive expression of membrane proteins is cellular toxicity. The most successful strategy for managing toxicity has been to provide control of gene expression by use of an inducible expression system. Expression is regulated by insertion of a chemically inducible promoter upstream of the target gene. Options include, but are not limited to, tetracycline or doxycycline (e.g. T-Rex and Tet-On 3G). Andréll et al. (2016) demonstrated use of tetracycline-inducible cell lines, with fluorescence-activated cell sorting (FACS) used to isolate stable cell lines, with improved levels of membrane protein overproduction monitored by fluorescent protein (FP) tagging of the target membrane proteins [67]. Transposase-based systems (e.g. PiggyBac) may also be used to generate inducible cell lines for membrane protein expression (68). Quantitation of surface protein expression can be achieved using calibration beads (e.g. Bangs Laboratories) and flow cytometry if a suitable antibody recognising the target protein is available [69]. Alternatively, fluorescently labelled ligands or radioactive ligands may be used if available for the target of interest [70, 71]. Cell surface proteomics can also be used to quantify target CMPs if no tool reagents are available [72].

The recombinant cells expressing the target protein may be used directly as an immunogen for antibody discovery and screening. They may also be used to derive purified protein or be exploited to generate a wide range of CMP antigen formats, which are described in Sect. 4.

High-quality validated antibodies recognising CMP can be difficult to source commercially, making detection of target protein overexpression with commercial tool antibodies challenging. Alternative strategies to detect and optimise protein expression are often required. Epitope tags can be engineered in surface-exposed N-terminal or extracellular loops of the protein, allowing detection with anti-tag-specific antibodies. Caution is required when inserting tags, to avoid perturbing protein folding and function or the potential to disrupt potentially important epitopes (see Maue for review [73]). Thermo Fisher’s Tag-On-Demand platform reduces the risk of impacting protein function by using amber suppression to enable expression of C-terminally tagged and native membrane proteins from the same expression construct. In our hands, amber suppression was rare and required isolation of a stable transfected cell clone containing the tRNA synthetase and aminoacyl tRNA with the desired characteristics; however, having achieved this, expression of the tag could be controlled by exogenously added non-natural amino acid to allow enrichment of high-expressing membrane protein cell lines [74].

FPs such as green FP (GFP), typically fused to the C-terminus of the membrane protein, provide another strategy to detect expression of membrane proteins. FP fusions allow screening for membrane protein expression using fluorescence from the FP as a surrogate for expression of the membrane protein–FP fusion. Although GFP is the most widely used FP, many alternatives are available, with different emission spectra. Care should be exercised when selecting an FP; FPs without the monomerising A206K mutation are prone to dimerise and accumulate in the endoplasmic reticulum [75], and addition of FPs can increase protein multimerisation, even when ‘monomeric’ FPs are used. It has been demonstrated in one study that the use of fusion partners mEmerald, mVenus, mCerulean, and mCherry as FPs can impact target membrane protein expression levels, with mVenus systematically giving higher expression levels [76]. FP fusion tags offer significant options in enabling studies of membrane protein expression, including sorting of single target protein engineered cell clones via FACS, allowing high-expressing stable clones to be isolated. FP fusions can also be used for fluorescence microscopy or imaging, enabling verification of cell surface expression in cases where antibody detection is not possible. FP fusions allow for monitoring of detergent solubilisation and quality of purified protein using the technique fluorescence size exclusion chromatography (FSEC) [77]. Finally, it is possible to use FPs as an indicator of expression without fusing to the final expressed target protein by using a 2A viral sequence separating the target protein and FP. 2A sequences act as a ‘self-cleaving’ peptide that actually operates via a non-proteolytic, ribosome skipping mechanism allowing expression of both target protein and FP from a bicistronic construct in the same cell [78].

Looking to the future, much of what has been described here involves screening a relatively small number of parameters to identify an expression system (protein sequence, vector, host) producing sufficient membrane protein to support an antibody discovery campaign. To put this in context, although yields of protein can vary widely depending on the expression system, typically 0.05–2 mg of protein can be obtained from a litre of mammalian host culture. An antibody campaign may require 5–10 mg protein. To facilitate a more complete exploration of expression parameters, high-throughput processes and automation may allow a more efficient means to identify successful expression constructs/host combinations and rapidly exclude those that are unsuitable. This type of approach has been described by Birch and Quigley and is addressed in Sect. 4.3 [79, 80].

Complex Membrane Protein Antigen Formats

Choice of antigen format for antibody discovery is dependent on compatible solubilisation and purification methods for the target of choice and the downstream antibody discovery platform used. The formats used range from simple soluble regions of the protein—such as soluble ECDs or peptides – to CMPs purified in detergent or lipidic environments, to complex membranous environments and whole cells. In this review, we focus on the more complex antigen formats and solubilisation methods shown in Figure 2, including detergent solubilisation, discoidal CMPs solubilised by membrane-scaffold proteins or peptides or by polymer solubilisation, reconstitution in liposomes, and production of virus-like particles (VLPs). These CMP formats more accurately recapitulate the native environment and, therefore, have often been more successful in leading to isolation of functional antibodies to CMPs, whereas simple antigens, such as peptides, have failed.

Fig. 2.

Schematic of extraction of membrane proteins expressed on the cell surface isolated by detergents, MSPs/NSPs, proteoliposomes, polymers, and budding of VLPs from the cell membrane. The table describes the need for detergent extraction, solubilisation, purification methods, and the pros and cons of each method for the solubilisation and purification of CMPs used in biologics discovery. Figure created with BioRender.com. AIEC anion exchange chromatography, CMP complex membrane protein, MSP membrane scaffold protein, NSP nanodisc scaffold peptide, SEC size exclusion chromatography, UV ultraviolet, VLP virus-like particle, N/A not applicable.

Virus-Like Particles

VLPs are becoming a go-to antigen display technology for antibody discovery campaigns. VLPs are non-replicative and non-infectious particles that bud from the host cells when a viral group-specific antigen (gag) polyprotein is expressed (Fig. 2). VLPs are similar to immature virions and consist of host cell membrane proteins and the lipid bilayer, generally 100–200 nm in diameter [81]. VLPs are generated from transient transfection of host cells, such as HEK293 cells, with a plasmid encoding the target CMP or from stable cell lines. For discovery of antibodies against epitopes close to the membrane surface or in conformations of membrane protein only available in lipid environments, VLPs as an antigen display technology are becoming indispensable. Integral Molecular are a leader in the field for producing VLPs using murine leukaemia virus (MLV) gag polyprotein. The MLV-VLPs combined with immunisation of chickens has led to successful campaigns to produce state-specific mAbs against open and closed conformations of glucose transporter 4 (GLUT4) [82] and highly specific claudin-6 (CLDN6) mAbs without cross-reactivity to CLDN9, which has previously hampered clinical trials [83]. These two examples are explained in detail in Sect. 5.1. In antibody discovery campaigns, there are several benefits to using VLPs as an antigen display technology compared to other formats. The VLPs can provide a native-like lipid environment with high concentrations of target membrane protein on the surface, with fewer host cell membrane proteins present compared to a stable cell line. The VLPs are stable, capable of prolonged storage at 4 °C and at −80 °C, and are less prone to aggregation and degradation compared to other antigen formats. VLPs are highly immunogenic and often do not need to be conjugated to a toxin adjuvant to elicit a strong immune response, due to the viral gag protein acting as the adjuvant (see Zepeda-Cervantes et al. for review of VLPs and immune responses) [84]. Furthermore, VLPs can be produced on a cell background that is matched to the immunisation host, such as mouse immunisation with murine colorectal carcinoma cell line (CT26), to effectively make the host VLP immunologically ‘silent’ and lead to more antibodies produced against the target antigen. In comparison to VLPs, recombinant ECDs and discoidal membrane protein extracts can have a higher percentage purity and yield of the target membrane protein, which is needed for some applications. To obtain highly pure VLPs, they can be purified from contaminating components such as broken membrane, exosomes, micro vesicles, and apoptotic vesicles, which have a similar size [85].

Alternative Virus-like Particles

Most recently, an alternative strategy to use virus particles has emerged. Smith et al. described methods by which membrane proteins can be displayed on vaccinia virus [86]. Vaccinia virus requires special containment, but a variant exists (vaccinia Ankara) that has a better safety profile for laboratory containment. For type I membrane proteins, ECD can be displayed by fusing the C-terminal to a viral transmembrane protein A56R, which is expressed on the surface of vaccinia extracellular enveloped virus (EEV). To express more complex integral membrane proteins such as GPCRs, a different fusion strategy is adopted in which the C-terminus of the GPCR is fused, via a Gly-Ser linker, to the N-terminus of a viral protein F13L, which associates with the inner membrane surface of the EEV via palmitoylation. This was used to demonstrate expression of the GPCRs CXCR4 and Frizzled 4 and 7. These virus particles can be used to isolate antibodies by phage panning on the virus expressing the target antigen. This system is potentially cleaner than VLP methods, as the vaccinia virus is more selective in which proteins are incorporated into the membrane [86].

Virus-like Particle Purification

There are several methods, described below, to isolate VLPs from contaminants. As most viruses are negatively charged at neutral pH, due to their acidic to neutral pI [87], this makes them amenable to anion exchange chromatography (AIEC) as a purification method [88, 89]. Heparin affinity purification leads to high purity of the VLPs, due to differences in affinity to the contaminating extracellular vesicles [89-91]. Multimodal resins, such as Capto™ Core 700 and 400, combine size exclusion chromatography (SEC) with a hydrophobic and positively charged bead core that allows small impurities to bind within the bead while large molecules/VLPs flow around the beads and elute earlier [91]. SEC can also be used to isolate VLPs from membrane fragments and aggregates and is often the final ‘polishing’ step (Lorenzo et al. [89]). Combinations of AIEC or heparin affinity with Capto™ Core columns can lead to high yields and high purity of VLPs. For methods for quality assessment of VLPs, see references [92-95]. As mentioned previously, good tool antibodies for the characterisation of targets in cell lines is also a requirement for the validation of targets produced in VLPs, which are not always available for CMPs.

Antigen Fusions: Tagging, Modifying, and Increasing Antigen Surface Density

To circumvent issues with target validation due to lack of tool antibodies, the target CMP can be modified with tags that can also be useful in downstream assays, such as screening phage libraries and in binding assays (Fig. 3). Fusion proteins are commonly used to stabilise proteins during expression and for purification of proteins. In this section, fusion proteins will be discussed in terms of their use in VLPs, but in some cases, they can be applied to recombinant cell lines and soluble proteins. Fusion proteins can be formed to incorporate tags such as an FP, mentioned in Sect. 3, which are useful in microscopy, fluorescence correlation spectroscopy, and flow virometry for quantification of the VLP number, size, and antigen concentration on the VLP surface [96, 97]. Addition of externally displayed tags, such as His- or FLAG-tags, can be beneficial if there are limited tool antibodies available or if the tool antibodies are limited to the intracellular portion of the protein. These tags can also be useful to determine whether the membrane protein is displayed in the correct orientation within the membrane. Other beneficial tags include the AviTag, which allows for biotinylation of the antigens on the VLPs for capture and use in phage display campaigns, screening assays, and binding assays [98]. If the N- and C-termini of the antigen is luminal, then the antigen can be fused to a transmembrane domain that acts as a ‘snorkel’ and permits addition of surface-exposed tags (Fig. 3c). Transmembrane snorkels already tested by others include transmembrane domains from glycoprotein 41 (gp41) [99], mouse beta-type platelet-derived growth factor receptor (PDGFR-ß) [100, 101], potassium voltage-gated channel subfamily E member 1 (KCNE1) [102], truncated neuraminidase, and hemagglutinin [103]. In VLPs, the snorkel transmembrane domain can be fused to the C-terminus of the gag protein or another host cell membrane protein for external display of a tag on ‘empty’ VLPs containing no target antigen [99]. Soluble antigens can also be displayed on VLPs by fusing the soluble antigen to snorkel-tagged gag proteins or other membrane anchors. Additionally, VLPs can be engineered to increase the antigen density at the surface by fusing the antigen directly or indirectly, by a tag-capture system, to the gag protein. The number of gag proteins per VLP is ~ 2500, thereby increasing the number of antigens incorporated into the VLPs [104]. However, careful quality assessment of the VLP and the structure of the CMP is required to ensure overloading of the VLP does not lead to distortion of the CMP or VLP. Tag-capture systems include SpyTag-SpyCatcher [105], cTag-cClamp [106], GFP-nanobodies, and ALFA-nanobodies [107]. Variable domain sequences from the heavy chain (VHH) (also known as a Nanobody^®, a registered trademark of Ablynx) are 15-kDa proteins that have high specificity to their target. The cTag-cClamp and nanobody systems can also be used for protein purification when coupled to beads. The structure of SloI, a Ca²⁺-activated K⁺ channel, was solved using SloI tagged with ALFA- and GFP-tags in membrane-derived vesicles and purified by nanobody technology. The SpyTag-SpyCatcher system can be utilised in multiple ways, as a method to increase antigen density in VLPs by tagging the antigen-SpyTag to gag-SpyCatcher; to tag adjuvants (such as diphtheria toxin) to the antigen; or to externally tag an antigen or antigen fragment onto the VLP [108]. The cTag-cClamp and nanobody systems can also be used for protein purification.

Fig. 3.

Schematic showing different tagging systems for use in VLPs. a Fluorescent proteins (green beta-barrel) can be fused to the luminal (or extracellular) side of the antigen to aid detection of the VLPs. b Extracellular tags (orange rectangle) can be used for detection and for capture. c Transmembrane domains, ‘snorkels’ (pale pink transmembrane domain), can be added to membrane proteins lacking an external terminus for addition of tags, fluorescent proteins, or tag-capture systems. d Fusion of the antigen to the gag protein can increase antigen density in VLPs. e Addition of tag-capture systems, such as Tag-Clamp, can be used to increase antigen density or immunogenicity. Figure created with BioRender.com. gag group-specific antigen, VLP virus-like particle

Extraction Methods and Formulation of Antigens

Other antigen formats that have increased in popularity due to advances in cryo-EM driving the field of structural biology (reviewed by Harrison et al. [109]) are the CMPs in detergents and nanodiscs. The biggest bottleneck to isolation and purification of CMPs in these formats is the solubilisation of the CMP. Three main approaches have been employed to solubilise membrane proteins for purification: detergents, protein-based scaffolds, and polymer-based scaffolds (discussed further below and summarised in Fig. 2).

Detergents

Detergents such as n-dodecyl-β-D-maltoside (DDM), n-dodecyl-phosphocholine (Fos-12), and laurylmaltose neopentylglycol (LMNG) are the most efficient compounds at solubilising membrane proteins. The company Cube Biotech has a solubilisation database for membrane proteins expressed in HEK293 cells, comparing the solubilisation and liquid chromatography-mass spectrometry (LC/MS) identification of proteins in a range of detergents, derivatives of the polymers poly(styrene-co-maleic acid) (SMA), poly(diisobutylene-alt-maleic acid) (DIBMA), poly (acrylic acid-co-styrene [AASTY]), and amphipols. The detergents had a higher solubilisation capability compared to the other reagents tested (https://cube-biotech.com/solubilization-database). Screening of detergents is required with each new CMP tested, as each protein will solubilise to different extents dependent on the detergent used [80]. While being efficient at solubilisation, the drawback to using detergents is the loss of the lipid bilayer. The membrane protein is, therefore, in a non-native environment, which can distort the native structure and induce aggregation, degradation, and precipitation. This distortion can be slightly mitigated by addition of lipids during detergent extraction, and purification steps to promote protein stabilisation as the environment changes. In particular, cholesteryl hemisuccinate (CHS) greatly improves protein stability, but a range of lipids can be screened to find a lipid that optimally stabilises the CMP of interest [110]. Detergent-solubilised CMPs can then be affinity purified by conventional methods (discussed in Section 4.2.5). For antibody discovery, use of detergents can pose issues, with the potential to mask epitopes, particularly in proteins with small ECDs [111]. However, DDM detergent micelles containing the ion channel glycine receptor α3 (GlyRα3) were successfully used for immunisation in mice and produced six mAbs, which is described further in Sect. 5.2 [112] (Table 2).

Table 2.

Monoclonal antibody isolation strategies for multi-spanning membrane proteins – selected examples

Target	Target class	Antigen format	Antibody isolation method	Antibody function/reference	Number of hits	Functional hits
Endothelin A receptor	GPCR	Overexpressing Cells; N-terminal ECD	Immunisation/mouse	Inhibitor[188]	613	Multiple; 1 mAb selected
Apelin receptor	GPCR	Thermostabilised receptor in nanodiscs	Immunisation/camel; single-domain antibody phage library	Inhibitor[35]	186	106 binders identified; 1 antagonist converted to agonist by structure-based design
CGRP receptor	GPCR	Purified protein: CGRP receptor ECD with RAMP1	Immunisation/Xenomouse	Competitive inhibitor[175]	Not reported	Inhibitor antibody
GPRC5D	GPCR	Plasmid DNA Boost with RBL cells overexpressing GPCR5D	Immunisation/mice	GPCR5D-binding antibody[206]	Not reported	1 antibody reformatted to bispecific (dual GPCR5D, CD3 binder)
ASIC1a	Acid-sensing ion channel	ASIC1a protein in nanodisc	Phage library selections	Inhibitor[207]	6 described	1 identified
P2X4	Ligand-gated ion channel	P2X4 purified protein	Phage library selections	Inhibitor/potentiator[208]	33	6 inhibitors; 8 potentiators
Kv1.3	Voltage-gated ion channel	Plasmid DNA; purified Kv1.3 reconstituted in liposomes	Immunisation/chicken; immunisation/llama	Inhibitor[209]	50 from chicken; 19 from llama	10 identified [9 from chicken, 1 from llama]
GLUT4	Glucose transporter	GLUT4 VLP	Immunisation/chicken; immune phage display library	Inhibitor[82]	29	2 state-specific mAbs
ASCT2	Glutamine transporter	Rat hepatoma cells expressing human ASCT2	Immunisation/rat	Inhibitor[210]	Not disclosed	1 antagonist
Glycine receptor	Ligand-gated ion channel	Purified protein in detergent	Immunisation/mouse	Activator[112]; inhibitor	6	3 activators; 3 inhibitors
Claudin-6	Tight junction protein	DNA/VLP	Immunisation/chicken and scFv library build	Claudin-6-binding antibody[83]	1891 unique scFv	5 specific binders
STEAP2	Metalloreductase	Ad293 cells expressing hybrid STEAP3-2 engineered protein	Immunisation; Del-1 humanised mouse; immunisation STEAP2 knockout mouse	Binding antibodies[146]	2	2 mAb described

Examples of antibody isolation strategies generating antibodies and single-domain antibodies binding multi-spanning membrane proteins representing several protein families

CGRP calcitonin gene-related peptide, ECD extracellular domain, GLUT4 glucose transporter 4, GPCR G protein-coupled receptor, mAb monoclonal antibody, STEAP six-transmembrane epithelial antigen of prostate, ASIC1a  acid-sensing ion channel 1a, ASCT2 alanine serine cysteine transporter 2, scFv single-chain variable fragment, RAMP receptor activity-modifying protein, RBL rat basophilic leukaemia cells, VLP virus-like particle

Proteoliposomes

Purified detergent-extracted CMPs can be reconstituted into liposomes to create proteoliposomes. Proteoliposomes provide a more native-like environment to study CMPs than detergents, with the CMP inserting into a spherical lipid bilayer of synthetic or animal-derived lipids. The proteoliposome size ranges from 15 nm to 500 nm or larger dependent on the formation method and lipids used [113]. Lipids play an important role in modulation and activation of many membrane proteins, which can be studied by controlled selection of lipid and ratios of lipids in the absence of other CMPs [114, 115].

The challenges of working with proteoliposomes include successful reconstitution and correct orientation of the CMPs within the liposome, as the proteins insert randomly. Quantification of CMP orientation is usually required, but the use of tags and soluble components on the extraluminal side of the membrane protein can increase insertion in the correct orientation and/or selection of liposomes with correctly orientated proteins [107, 116]. While proteoliposomes offer a more native-like environment, they are not able to recapitulate the asymmetry of lipid bilayers or the complexity of native-lipid mixes or lipid rafts. Yet, proteoliposomes have been successfully used in antibody discovery campaigns.

A truncated GPCR, M₂ muscarinic acetylcholine receptor, in phosphatidylcholine liposomes was used in immunisations to produce a single-chain variable fragment (scFv) phage library. The GPCR was also incorporated into liposomes containing biotinylated lipids, which allow capture onto streptavidin beads for screening in phage display. An scFv specifically binding the receptor was identified and reformatted for further evaluation [117].

Protein-Based Nanodiscs

Membrane scaffold proteins (MSPs) are amphipathic proteins derived from apolipoprotein A-1 (ApoA1), a key protein involved in transport of fat in high-density lipoprotein particles. MSP self-assembles in the presence of lipids to form a belt of protein around membrane proteins, capturing lipids associated to the membrane proteins in the process. Solubility of the extracted membrane proteins is maintained, with the hydrophilic regions of the MSP interacting with the environment and the hydrophobic region interacting with the lipids (Fig. 2). For extraction of membrane proteins by MSPs, the membrane proteins are first solubilised in detergent then reconstituted into synthetic lipids and/or pre-defined lipids to form a nanodisc of ~ 10 nm in diameter prior to purification to isolate the target CMP [118]. The latest advances in the engineering of MSPs uses the SpyCatcher-SpyTag technology to circularise the MSPs in vivo, leading to a tenfold higher yield in protein extraction. Dependent on the circularised MSP used, nanodiscs can be homogeneously formed in size ranges from 11 to 100 nm [119]. Nanodiscs formed from MSP 1D1 encapsulating the GPCR human apelin receptor, which was stabilised with additional mutations, were successfully used in camel immunisation to produce camelid heavy-chain antibodies, described further in Sect. 5.2 [35] (Table 2).

Nanodisc scaffold peptides (NSPs) are derived from the MSP ApoA1 protein but are engineered into peptides and linkers. These ‘peptidiscs’ have a short amphipathic bi-helical peptide derived from two repeats of a sequence in ApoA1. The most recent peptidiscs have superior solubility and improved stabilisation of the membrane protein. They have been used to isolate a variety of membrane proteins in ~ 12-nm nanodiscs for functional studies, including mechanically activated ion channels [120], and to study the interactions between an adenosine triphosphate (ATP) binding cassette (ABC) transporter, BtuCD, and a soluble binding partner, BtuF, using mass photometry [121].

Saposins are another family of lipid-binding proteins that, like MSPs, have been utilised to make nanodiscs, by surrounding the membrane protein and lipids in a belt of saposin A protein to create saposin-lipoprotein (Salipro) particles. This can be achieved with initial detergent solubilisation and reconstitution of the membrane protein with lipids into particles [122], or by a more direct extraction with minimal detergent [123]. Both MSP and Salipro technologies add complexity into membrane protein quality control (QC), as they interfere with ultraviolet (UV) spectroscopy, absorbing at 280 nm. Use of MSP/NSP nanodiscs as a CMP antigen format in immunisations for antibody discovery will yield mAbs that also bind to MSP/NSP, which should be taken into consideration when choosing antigen formats.

Polymer-Based Nanodiscs

Poly (styrene-co-maleic acid) (SMA) polymers are an alternative to detergent extraction and circumvent the use of scaffold proteins. The SMA-based polymer is amphipathic and, like the MSPs, forms a belt around the membrane protein and lipid. SMA allows direct extraction of protein without the need for detergents, to create SMA lipid particles (SMALPs). The SMALP network (https://www.smalp.net/) provides protocols and tips for purification of membrane proteins with polymers, along with references and upcoming conferences. The presence of native lipids is important to maintain protein stability, function, and the conformational flexibility of membrane proteins. For instance, cholesterol is important for the modulation of many GPCRs, and several lipids such as phosphatidylglycerol and phosphatidylethanolamine are allosteric regulators. The importance of lipids is further highlighted in a review of GPCRs and SMA-like copolymers [124]. Isolation of a yeast tetraspanin, CD81, in SMA2000, by nickel affinity and SEC, allowed for conformational mobility of the protein, and two distinct conformations could be identified by antibody binding [125]. SMAs have also been successfully used in the isolation of a low-yielding GPCR, human β₂-adrenergic receptor, through optimisation of purification conditions [126]. SMAs are currently the best-characterised polymers and have been used to solubilise a range of membrane proteins for structural studies, immunisations, and antibody-binding assays [127]. The second generation of SMA-like copolymers have been developed to improve their properties with sidechain modifications. These include poly(styrene-co-(N-(3-N′,N′-dimethylaminopropyl)maleimide)) (SMI), which lacks the maleic acid of SMA and has improved tolerance to divalent cations and pH sensitivity [128] [129]. Replacing the maleic acid with acrylic acid (AASTY polymers) forms a more homogeneously sized nanodisc [130]. A maleic acid derivative, poly (diisobutylene-alt-maleic acid) (DIBMA) allows for larger particles to be produced and used with divalent cations. The replacement of the styrene moiety with an aliphatic diisobutylene means absorption can be measured at 280 nm without interference, as with SMA and AASTY [124]. A comprehensive review of polymers can be found in Orekhov et al. [131].

Poly-acrylic acid (PAA) with modified side chains makes up amphipathic polymers called amphipols, which like MSPs require the CMP to first be solubilised with detergent before reconstitution into the amphipols. Amphipols, such as A8-35 and its derivatives, have mostly been used for solubilisation of CMP used in structural biology with cryo-EM and mass spectrometry. Most recently, cyclo-alkane modified and aromatic modified amphipols have been developed that can extract the CMPs directly from the cell membrane [132, 133]. Amphipols have the added benefit that they do not absorb at 280 nm and are more tolerant to divalent ions than most SMAs.

Amphipathic protein scaffolds and charged polymers, such as many SMA derivatives, can influence protein–protein interactions, modify protein functions, or, if they are of opposing charge to the membrane protein, become ineffective at solubilising/reconstituting. Pentyl-functionalised inulin polymers are non-ionic and have been shown to solubilise membrane protein efficiently [134]. They have been successfully used to reconstitute a functional redox complex into nanodiscs [135]. Sulfo-SMA and Sulfo-DIBMA are also neutrally charged, but still water-soluble, allowing for native protein–protein interactions to be observed [136].

When CMPs are extracted, their native environment becomes perturbed, which affects some proteins more than others. Although lipids are present around the membrane protein to create a native-like environment, the lipid packing becomes distorted, which can affect mobility, conformation, and function, e.g. conductance of ion channels [137, 138]. The choice of nanodisc and size can also influence the structure of the CMP. Using the Erwinia ligand-gated ion channel (ELIC) as an example, differences in the cryo-EM structure of ELIC was dependent on the nanodisc formed when comparing saposin, SMA, and two circularised MSPs. Molecular dynamics (MD) simulations indicate that larger nanodisc sizes may better replicate native environments and structures, allowing conformational flexibility [139]. Experimentally, the photoactivation and full conformational changes of a rhodopsin, bRho, could only be observed when extracted with the polymer DIBMA, which forms larger particles when compared to SMA and SMI. The styrene of SMA and SMI may have also interfered with the rhodopsin [140]. While conformational flexibility is needed for studying CMP activity, the trapping of CMP in one conformation may be beneficial for some antibody discovery campaigns and also for structural analysis, as shown with an ABC transporter extracted with SMA2000P [141].

Purification Methods

All methods described in Sects. 4.3.1–4 to solubilise membrane proteins subsequently require the purification of the target CMP from other proteins for downstream applications. Purification of extracted CMPs is most frequently performed utilising affinity-tagged purification. The most popular tag to use is the His-tag, followed by the Strep-tag and FLAG-tag. Cobalt-based resin can lead to higher purity of protein when used instead of nickel resin, due to higher specificity of the cobalt ions to the His-tag, although this can be protein and isolation method specific. The use of Step-tags can yield higher purity than His-tag purification [33]; an added benefit is that in antibody library screens Strep-tags are not often used. Furthermore, the third-generation Twin-Strep-tag^® has improved affinity, increasing capture of low-yielding proteins with high specificity. Combining a Twin-Strep-tagged membrane protein with Strep-Tactin XT-based purification can yield highly pure CMPs [142]. Strep-Tactin modified silica can also be used to capture Twin-Strep-tagged membrane proteins maintained within the membrane for use in screening assays for ligands or antibodies [142]. SEC can then be performed to isolate monomeric CMP from aggregates. The use of fluorescence-detection SEC (FSEC) of an FP-tagged CMP can be highly advantageous for identifying the CMP and assessing yields, peak profiles and elution times [77]. Antibody-capture systems can lead to CMP capture and elution with 99% purity in one step. The company Cube Biotech use the Rho1D4 system, where a nine-residue tag is added to the C-terminus, which leads to high purity of low-yielding membrane proteins. Nanobody agarose resins and magnetic particles, so called Nano-Traps, are also available for capture of a range of tags, including, but not limited to, ALFA, GFP, Myc, Spot, and V5 [107, 143].

In antibody discovery, the choice of antigen display technology and understanding the limitations of each method is dependent on the mode of action of the antibody needed and how the epitopes of the target are displayed. To find the optimal conditions for protein expression, solubilisation, and purification, it is important to screen protein constructs, expression conditions, detergent/polymer/MSP/NSP for protein solubilisation, stability, and buffer conditions. QC of CMPs is more challenging than for a recombinant ECD; for QC methods, see references [122, 144] and the SMALP network.

Automation and High-Throughput Screening

Two of the most challenging parts of membrane protein purification are gaining a high yield of protein and its extraction in a soluble form for downstream work. The use of automation and high-throughput methods are hugely valuable for screening and identifying optimal conditions at all stages of protein solubilisation and purification. The Quigley Lab have clear protocols for high-throughput cloning, solubilisation, buffer screening, and small-scale purification of proteins from mammalian cells [80]. Often CMPs have a low yield; therefore, screening of different constructs with tags in different locations can identify those giving the greatest quantity of the target membrane protein. Similarly, point mutations can increase stability [145], or grafting of ECDs from difficult to express membrane proteins onto a family member that is easier to express, as shown for six-transmembrane epithelial antigen of prostate-2 (STEAP2) and STEAP3, can provide a format for epitope display in yields that can be used in antibody discovery [146]. Screening of detergents or polymers for solubilisation and small-scale purification in a 12- to 96-well format can rapidly identify the best conditions for scale up.

Discovery Strategies to Identify Therapeutic Antibody Leads Targeting Complex Membrane Protein Targets

Therapeutic antibody isolation strategies for CMP targets ipso facto focus on recognition of accessible extracellular epitopes. As mentioned, this presents significant challenges, particularly for CMPs that expose small extracellular loops and, therefore, limited potential accessible epitope(s). A further challenge is that for some preparations of CMPs, antibodies may be isolated that target intracellular epitopes preferentially due to there being more exposed potential epitope area. This situation requires antibody discovery strategies that will favour isolation of rare extracellular epitope binders.

Once methods are established to produce validated formulations of CMP antigen, a variety of approaches can be considered to generate antibodies. There are two main approaches to generate antibodies targeting CMPs. These typically rely on immunisation of laboratory animals, harnessing the natural immune response, or in vitro display methods, such as phage or yeast display [147-149]. Following immunisations in mice, hybridoma technology is the best-known methodology for mAb discovery and involves immortalisation of antibody-secreting cells by fusion with myeloma cells. Screening can then be performed on immortal clone culture supernatants to identify antibodies of interest [150]. In recent years, there has been an increased focus on directly screening for antigen-specific B cells using microfluidic platforms (for review see Schardt et al. [151]). These methods can use closed systems in which single B cells are encapsulated in water-in-oil droplets for screening or open systems such as the Beacon (Berkeley Lights) technology, in which cells are captured in a microfluidic chamber or ‘pen’ [152]. In the case of phage display methods, display libraries may be generated from lymphoid tissue of immunised hosts or by generation of designed, synthetic libraries [148, 153-156]. Most functional antibodies to date against CMP targets have been generated directly by immunisation strategies or by generation of immune display libraries from isolated lymphoid tissue of immunised hosts. In this review, we will consider immunisation and phage display strategies and how these can be tailored according to the antigen generation strategies adopted to drive antibody discovery campaigns. We will also consider some emerging themes in how biotech and pharma companies are approaching this challenge.

Table 2 provides selected examples from recent publications highlighting choice of antigen format, antibody isolation strategy, and outcomes. A common host used for antibody generation is the mouse, which likely is a consequence of ready availability, ease of handling, and well-established protocols. Antibody generation in mice also exploits the natural immune system and mechanisms (including clonal selection and somatic hypermutation), which produce mAbs with high affinity and a good developability profile [157, 158]. Advances in this field have included the development of transgenic humanised animals, which obviates the requirement for humanisation of mouse antibodies. Examples of transgenic mice include Xenomouse^® (Abgenix), HuMAb^® mouse (Medarex), and ATX-Gx™ (Alloy Therapeutics) [12, 159, 160]. OmniAb also market several transgenic options, with OmniRat, OmniMouse, OmniChicken, and OmniTaur, with the latter platform expressing cow-derived antibodies that have ultralong complementarity determining region 3 (CDR3) lengths [161-163]. This latter point is linked to an emerging theme of using alternative hosts for antibody discovery. Due to immune tolerance, immunisations can result in a poor immune response to antigens that are highly homologous to the host’s own proteins. A recent review showed that mounting an effective immune response in mice may be limited due to their evolutionary closeness to humans [164]. The authors’ analysis showed that for human drug targets, 99% have a murine orthologue, with half having ≥ 90% sequence identity and a quarter ≥ 95% identity. A further observation in therapeutic mAbs is the frequency of antibodies having short heavy-chain CDR3 lengths [9–13 residues]. This leads to antibodies that tend to bind flat, featureless epitopes rather than accessing pockets in proteins. These observations illustrate a trend towards using alternative hosts for immunisation. Thus, chickens are being used, as they are evolutionarily more distant to mammals, having diverged 310 million years ago. The impact of this should be enhanced immune response to human targets, as only 15% of human drug targets are highly conserved (> 90%) in chickens. Camelids, which include llamas and alpacas, also provide an established platform that is built on the prevalence of a camelid antibody structure of a small heavy-chain antibody that can be formatted as a VHH or nanobody [165]. These nanobodies are small (~ 15 kDa) and structurally are composed of antiparallel ß-strands supporting three CDRs that constitute the binding surface, the paratope. Nanobodies are easily converted to phage display formats for drug discovery, including generation of entirely synthetic single-domain antibody (sdAb) libraries such as the Sybody library [166]. The benefit of generating synthetic libraries is the ability to vary the length of the CDR3, giving rise to different paratope interaction surfaces: concave, convex, and long protruding loop. These nanobodies can then bind varied epitopes, including, for example, using longer heavy-chain CDR3 loops, which can target ligand-binding pockets on membrane proteins more readily. The unique properties of these nanobodies requires careful consideration of their potential therapeutic use—whilst the small size favours humanisation and tissue penetration, it can also be associated with rapid clearance via the kidney and poor pharmacokinetics. This can be addressed by making use of the modular nature of antibody domains and engineering into multivalent formats (e.g. bispecifics) or by conjugation to Fc domains to improve half-life.

Immunisation strategies used to generate lead antibodies that bind to CMPs fall into three broad categories when considering options to produce target proteins as immunogens: (1) engineered recombinant cell lines overexpressing the target protein, (2) pure (or enriched) protein antigen derived from a recombinant heterologous expression system, and (3) genetic immunisation using DNA, RNA, or engineered virus encoding the target protein. These immunisation methods may, and often are, used in combination to ensure isolation of a diverse panel of antibodies and are explored in more detail in the following sections [167].

Antibody Discovery Strategies Using Engineered Recombinant Cell Lines and Complex Membrane Formats

Recombinant engineered cell lines designed to overexpress a target protein may be used directly as a source of membrane protein in its native conformation to use as an immunogen. These cell lines may also be used as a source of target protein for purification and reconstitution methods or provide a source for target protein enrichment via generation of VLPs. Cells expressing target protein are also critical for use as a screening tool to screen for antibodies that bind to the target CMP in a native membrane environment using assay platforms such as the iQue flow cytometer or Mirrorball (high-throughput fluorometric microvolume assay technology), using the host cell to eliminate any non-CMP target binders. As stated in previous sections, the benefit of using CHO and HEK cells as overexpressing cell lines for immunisation is that they present the target protein to the immune system in its native conformation, maximising the prospect of generating functional binders to the target. Since neither HEK or CHO cells are derived from a commonly used animal hybridoma host, the host cell proteome of engineered HEK or CHO cells, in addition to the expressed target CMP antigen, can potentially be recognised as non-self by the host animal immune system. Whilst this may have some benefit in that host cell proteins could produce an adjuvant effect to stimulate an immune response to the target antigen, the host cell proteins will also give rise to an animal immune response. This will impact the breadth and depth of antibody generation to the target CMP. It is, therefore, important to screen engineered cell lines to ensure high-level expression of the recombinant target CMP. This can be very challenging for GPCRs, ion channels, and transporters and, in our experience, is highly target dependent. We typically aim for expression levels of 5 × 10⁵–10⁶ target proteins per cell. However, we have estimated that with these levels of target protein expression and using an immunisation protocol with 10⁷ cells per animal, the amount of target protein introduced as immunogen is a tenth of what is typically immunised with a purified target protein.

Given immunisation with recombinant HEK or CHO cells overexpressing the target antigen is likely to lead to an immune response to host cell proteins, we can consider engineering cell lines using host cells closely matched genetically to the host animal. In the case of using mice as the host animal, this approach can employ isogenic mouse cell lines such as XS63 or CT26, both derived from BALB/c mice [168]. In using isogenic host cells to express the target CMP antigen, the host cell is theoretically immunologically ‘silent’, focusing the immune response on the human CMP target and minimising immune response to the host cell proteome [168, 169]. An alternative strategy to drive an immune response to the CMP target is to immunise mice in which the mouse homologue of the human target CMP is removed from its genome (i.e. transgenic knockout [KO] mice) [170]. In a further development to improve generation of antibodies to CMP targets, a recent study described use of cells engineered to express both the target antigen and cell membrane-bound cytokines as adjuvants. Huang et al. described the principle of ‘cell adjuvants’ by expressing membrane-bound mouse(m)IL-2, mIL-18, or mouse granulocyte macrophage colony stimulating factor (mGM-CSF) on syngeneic BALB/3T3 cells, which were effective in stimulating splenocyte proliferation in vitro. These cells were subsequently transiently transfected with a type I transmembrane protein, ecotropic viral integration site 2B (EVI2B), and used to immunise BALB/C mice. The 3T3 cells co-expressing GM-CSF and EVI2B stimulated the highest anti-EVI2B response measured by mouse serum cell-based enzyme-linked immunosorbent assay (ELISA) using xenogeneic 293 cells expressing EVI2B. This approach was successfully extended to a GPCR, CXCR2 [171].

A further example of use of recombinant engineered cells to drive antibody isolation was described by Zanvit et al. [146]. In this study, the prostate cancer antigen, STEAP2, was the target for an antibody discovery campaign. To generate STEAP2-specific antibodies, a cell immunisation hybridoma campaign was used. Interestingly, STEAP2 could not be cell surface expressed in non-prostate cells such as Ad293. To circumvent this issue, a cell line was made in which STEAP2 extracellular loops were grafted onto the backbone of the STEAP3 protein that was found to localise at the cell surface. This created a STEAP3-2 chimeric protein, and this also included a C-terminal snorkel tag comprising a PDGFR-ß transmembrane domain with a triple FLAG-tag to allow cell surface detection using anti-FLAG antibodies. Humanised Del-1 mice were immunised with Ad293 STEAP3-2 cells to generate an immune response. A B cell enrichment screening approach was used in which STEAP2-specific B cell clones were enriched through a deselection step using LNCaP STEAP2 KO cells. This allowed non-binding clones to be taken forward for hybridoma fusion. Hybridoma clones were identified by primary screening on Ad293 and Ad293 STEAP3-2 cells and LNCaP/LNCaP STEAP2 KO cells. A secondary screening step was also performed to eliminate binders to STEAP1, 3, and 4 family members, allowing identification of a STEAP2-binding antibody.

Recombinant cell lines expressing integral membrane proteins can also be used in a phage display setting. Kelil et al. have described a process they term CellectSeq, which combines phage selections (using a synthetic Fab phage library) on cells expressing target protein with a next-generation sequencing approach [156]. In this example, the authors targeted a tetraspanin (CD151) and a heterodimeric receptor (integrin α11ß1). For CD151, HEK293T cells were engineered to overexpress CD151, and a control cell line was made by depleting CD151 with a short hair-pin RNA. These cells were subjected to multiple rounds of phage selection against CD151-positive and CD151-negative cells, with the aim of producing pools enriched with clones specific for CD151 and a negative pool enriched for non-specific clones. Next-generation sequencing combined with a scoring algorithm was then used to identify candidate clones enriched by this process and allowed isolation of four highly specific antibodies.

In the last decade, VLPs have emerged as a valuable tool for presenting CMP antigens for antibody discovery [82, 172]. Detailed information on VLP production has been described in Sect. 4.2. A recent example of the use of VLPs described isolation of highly specific antibodies against CLDN6, a potential oncotherapeutic target due to its differential expression on cancer cells. This presented a significant challenge, as CLDN6 has only three extracellular amino acid differences to CLDN9, which is widely expressed in tissues. The strategy used to isolate rare antibodies specifically targeting CLDN6 was to use DNA immunisation of chickens with a plasmid encoding CLDN6, followed by boosting with CLDN6 VLP. B cells were then isolated and used to generate an scFv phage display library. Phage panning was then carried out using CLDN6 VLP and CLDN9 VLP to allow for positive selection or deselection, respectively. Selected binders could then be reformatted as IgG. This approach was successful in isolating rare anti-CLDN6 antibodies that had picomolar affinity, and showed minimal cross-reactivity to CLDN9 and other CLDN family members [83].

Antibody Discovery by Immunisation or Display Methods Using Purified or Enriched Recombinant Protein

Many options can be considered for antibody discovery using purified or reconstituted CMP antigens. In designing an antibody discovery campaign, the target membrane protein topology, structure, interacting partners, and potential epitopes are important considerations for antigen design. The final strategy adopted will depend on the feasibility of expressing the target protein in a chosen format. Synthetic peptides designed to mimic extracellular loop(s) may be used either as linear peptide(s) or as a conformationally constrained peptide (reviewed in Lee et al. [47]). This can be successful, but effective target epitopes are often discontinuous (conformational) [173], and in this review, we focus on recombinant protein generation strategies. The choice of strategy to express and purify a CMP target is directed by knowledge of the topology and structure of the protein and options available for protein engineering. If the CMP target has a large ECD, which can fold to give a stable structure, it is possible to express a recombinant version of the ECD and purify as a soluble protein that can be used as an antigen. This approach is exemplified by class B GPCRs that have structured N-terminal domains which bind peptide hormone ligands. An example of this is the isolation of antibodies, by phage display, to the N-terminal ECD of glucose-dependent insulinotropic polypeptide receptor (GIPr) [174]. In another example, the complex of the ECD of the CGRP-R and the ECD of the RAMP1 was used to generate anti-CGRP-R antibodies following immunisation of the Xenomouse^® [175]. Where the CMP targets have less well-defined ECDs, the proteins may be expressed in a heterologous expression system and purified.

Two recent examples of antibody generation using purified proteins can be used to exemplify strategies used for immunisations. In the first, Simard et al. isolated antibodies targeting the ligand-gated glycine receptor [112]. Pentameric GlyRα3 was expressed in insect cells by use of recombinant baculovirus, and the receptor was purified from insect cell membranes by detergent extraction and affinity purification. The purified protein in detergent micelles was used to immunise wild-type mice, and hybridoma supernatants were screened by ELISA using Ni-NTA plates to capture receptor. To focus on antibodies targeting conformational epitopes, hybridomas showing binding to 8 M urea denatured proteins were eliminated, and binding to correctly folded GlyRα3 was confirmed by FSEC using a GlyRα3-GFP fusion. A further step was taken to eliminate binders to linear epitopes by excluding hybridomas that bound to denatured GlyRα3 in sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) Western blots. This process allowed identification of six mAbs that modulated ion channel function in cells stably expressing GlyR.

In a second example, a thermally and conformationally stabilised human apelin receptor (a class A GPCR) was expressed in insect cells, affinity purified following detergent extraction, and reconstituted into nanodiscs [35]. Stabilisation was achieved by the introduction of three amino acid mutations into the apelin receptor. sdAbs were identified by immunising a camel with apelin receptor nanodiscs and screening for immune response using apelin receptor nanodisc ELISA. A camel immune sdAb phage library was generated and used to isolate sdAbs by phage panning using biotinylated apelin receptor liposomes. Positive clones were shown to be specific by eliminating clones binding to an irrelevant GPCR liposome preparation. Interestingly, a high-affinity sdAb (JN241) identified as an antagonist was used to solve a co-crystal structure of the apelin receptor–JN241 complex, and this structure was used to guide design of an sdAb in which mutation of the CDR3 (insertion of tyrosine) converted the antagonist to a full agonist of the receptor.

Phage display methods can be readily used to isolate antibodies to ECDs of membrane proteins using standard protocols developed for soluble proteins with very little modification. Examples of this were presented earlier in this section. Phage selections on full-length, purified membrane protein antigens displayed in either detergents, proteoliposomes, or nanodiscs provide an alternative to ECD-derived antigens for phage selections and for cases where the membrane protein has limited accessible extracellular loops. Unlike phage selections on peptides and ECDs, formats such as nanodiscs and detergent micelles expose both extracellular and intracellular protein surfaces for phage binding; for therapeutic applications, antibody isolation strategies must focus on the extracellular exposed protein surface/domains. Phage selection and antibody screening methods must, therefore, be tailored to select antibodies binding to these extracellular epitopes.

Genetic Immunisation

Plasmid DNA Immunisation

Genetic immunisation discovery strategies, typically based on DNA immunisation, use expression plasmids encoding the target gene of interest. These DNA expression constructs are delivered to the selected host species by an intradermal route. These delivery routes include intradermal injections (including Dermajet) [176, 177], gene gun (Bio-Rad) delivery [165, 178, 179], where DNA-coated colloidal gold particles are fired into cells in the dermis using compressed helium, or electroporation [180-184]. Electroporation requires injection of a DNA vector encoding the target gene into skin or muscle and application of microneedle electrodes at the injection site. A high-voltage electrical pulse is applied that creates nanopores in cell membranes and uptake of the DNA vector. Each delivery method aims to introduce the DNA vector encoding the target gene into cells at the delivery site to drive target CMP synthesis to stimulate an immune response. This methodology avoids issues relating to use of foreign cell lines expressing the target protein, as the antigen is expressed in a host cell that is regarded as self by the immune system. In vivo expression host cells should also allow presentation of the target protein in a native conformation in the host cell membrane and allow PTM. In practice, however, the low and transient expression level combined with a limited immune response to DNA-based immunisations can restrict success. Expression vectors can be modified to optimise expression by evaluating different promoters with the aim of altering promoter strength or using promoters specific for a host tissue (e.g. muscle, skin). An example of this is a study carried out Vij et al. [211], who compared the activities of human elongation factor-1α (EF1α), spleen focus forming virus (SFFV), ubiquitin C (UbC) and CAG promoters in vitro  (using HaCat keratinocyte cells and HEK293 cells) and in vivo using luciferase as a reporter. Interestingly, the CAG promoter gave the best expression profile in both cells and in vivo following gene gun delivery. A further observation was that antigen expression level declines rapidly in vivo by approximately tenfold each day post-delivery. This may explain how, in contrast to cell-, protein-, and peptide-based immunisation protocols, the immune response to DNA immunisation typically gives lower serum titres and is slow to develop, and a prolonged immunisation campaign may be required, with significantly fewer lead antibodies isolated (reviewed in Saade and Petrovsky [185]). In practice, DNA immunisations may be combined with alternative antigens (e.g. cells expressing target protein), and one published study showed a combined DNA/cell immunisation strategy, although giving the lowest antibody titre produced the highest number of antigen-specific clones [167].

Several approaches have been described that use alternative delivery methods or different ways of encoding the target protein. In one example, Hazen et al. [186] described a generation of antibodies to multi-drug-resistant protein 4 (MRP4), a 12 transmembrane spanning transporter. Antibody discovery campaigns using phage display and hybridoma technology and sources of MRP4 antigen (including detergent-extracted MRP4, and HEK and LNCaP cells expressing MRP4) had failed to identify anti-MRP4 antibodies. An alternative strategy was followed using an expression plasmid encoding MRP4, which was co-administered with plasmids expressing the immune modulators foetal liver tyrosine kinase 3 ligand (Flt3L) and GM-CSF. The plasmids were administered by hydrodynamic tail vein (HTV) injection, a technique that enables delivery of a large volume of DNA and subsequent high-level expression in liver; antibodies binding to extracellular loop epitopes were obtained. Only DNA immunisations resulted in antibodies that bound specifically to cell surface-expressed MRP4. Furthermore, co-administration of the immune modulators improved reproducibility of the serum response to MRP4. However, there was only a marginal increase in the number of hybridomas expressing extracellular MRP4-binding antibodies. This example also included use of different promoters (CAG and CMV promoters), and the CAG promoter increased the overall efficiency of DNA immunisation compared to the CMV promoter, allowing a reduction in the frequency of immunisations whilst maintaining a high immune response to the target protein. This group used the same technique to generate antibodies to the prostaglandin D₂ receptor, CRTH2 (described in a patent application [187]). Takatsuka et al. [184] also employed an immunisation regime delivering CCX-CKR and GroEL expression plasmids (employing E.coli GroEL as an adjuvant) using in vivo electroporation immunisations, followed by a final boost by HTV injection to obtain anti-CCX-CKR antibodies.

Messenger RNA Immunisation

Recent developments in messenger RNA (mRNA) technologies, driven by their importance in the context of the coronavirus disease 2019 (COVID-19) pandemic, are offering another potential avenue for therapeutic antibody discovery. The ability to generate mRNA by cell-free in vitro transcription provides an opportunity to drive expression of a target protein in cells or in vivo by using mRNA encoding the protein and could have utility for CMP antigen delivery [188]. Production of mRNA for protein expression is somewhat more challenging than DNA-based technologies, as mRNA is more susceptible to degradation by nucleases, is inherently less stable, and can directly stimulate inflammatory responses and in vivo delivery can be inefficient [189]. Advances have been made in mRNA stabilisation and delivery methods, with formulation in lipid nanoparticles (LNPs) becoming a common method for in vivo delivery. A recent example demonstrates the potential of this technology. Hsu et al. used mRNA-LNP encoding the receptor-binding domain (RBD) of SARS-CoV-2 spike protein to immunise mice and compared this with immunisation of RBD protein [190]. The mRNA construct encoding the RBD protein was shown to produce RBD protein by transduction of HEK293T cells. Having confirmed protein expression in vitro, immunisations of mice with mRNA-LNP and RBD protein were performed in parallel. Evaluation of serum titres of antibodies showed that mRNA-LNP induced higher titres than RBD protein immunisations. The mice were then used for hybridoma production, and six mAbs were identified from the mRNA-LNP immunised mice that showed specific binding to SARS-CoV-2 RBD. This study demonstrated the potential value of mRNA technology to produce mAbs to an extracellular protein domain, which could be extended to more CMPs or at least provide an option for consideration in designing therapeutic antibody campaigns.

Virus Immunisation

Viruses have been exploited as a strategy to deliver antigen encoding genes to cells for antibody generation. Adenovirus vectors have been explored in the context of vaccinations, due to ease of genome manipulation, the ability to accommodate large target gene fragments, high-titre virus production, and high transduction efficiency. Use of adenovirus as an immunisation strategy in biologics discovery has been reported by Amgen for two CMP targets: the iron transporter ferroportin [181, 191, 192] and transient receptor potential ankyrin 1 (TRPA1) ion channel [193]. In the studies reported, C57Bl/6 mice received a priming injection of recombinant adenovirus containing a CMV promoter driving expression of the target gene of interest fused to the Pan DR-binding epitope (PADRE) [194]. PADRE is a 13-amino-acid peptide that activates CD4+ T cells and initiates an innate immune response resulting in an adjuvant effect. The mice were boosted three times by electroporation with plasmid DNA encoding the target protein–PADRE fusion. Finally, mice were boosted 5 days prior to hybridoma fusions with a membrane preparation derived from cells expressing the ferroportin–PADRE fusion protein. In the case of ferroportin, two mice were used to generate 4000 hybridomas, and hybridoma supernatants were screened for ferroportin binding on ferroportin-expressing 293T cells. Thirty-seven ferroportin-binding antibodies were identified, with one functionally active. The authors used an additional strategy to generate Ferroportin antibodies from two strains of transgenic mice (Xenomouse™ IgG2κλ and IgG4κλ). In this case, mice were immunised with HEK293 cells transiently expressing ferroportin or membrane preparations from cells expressing ferroportin via subcutaneous or intraperitoneal injection followed by additional antigen boosts until antibody responses to ferroportin were observed. This hybridoma approach led to screening of 7600 hybridoma supernatants and identification of 200 ferroportin-binding antibodies, with 11 identified as functionally active (ferroportin patent application, Amgen [191]). This case study highlights an important feature of antibody discovery campaigns against CMP targets. Multiple complementary strategies are often necessary to ensure a successful outcome. It also highlights that extensive screening is required to identify the rare antibodies that target desired functional epitopes; in this example, 0.025–0.14% of the supernatants screened yielded antibodies with the desired function.

The three immunisation strategies described above (overexpressing cell lines, recombinant proteins, and genetic immunisation using DNA, RNA, or viruses) have been used in isolation to successfully generate functional binders to CMPs. However, it is more common to observe different antigen sources being used in combinations and employed in parallel to maximise diversity of antibodies isolated and increase opportunities to isolate antibodies with desired biophysical properties and diverse epitopes [176, 177, 179]. This may include combining human and a species variant antigen to drive selection of species cross-reactive antibodies. A large panel of binders provides the best opportunity to enable the most appropriate epitope to be targeted for function, as well as identifying antibodies with optimal characteristics in terms of ‘developability’ (e.g. solubility, stability, manufacturability, formulation) or species cross-reactivity to allow for translational studies and successful candidate drug development.

The Impact of Structural Biology, Artificial Intelligence, and Machine Learning on Antibody Discovery for Complex Membrane Protein Targets

Development of mAbs and next-generation antibody modalities such as bispecific antibodies, engagers, and ADCs requires a highly complex development process. This process aims to identify a clinical candidate molecule in which multiple properties of the biologic drug are optimised (e.g. affinity, biophysical properties, safety). As we have highlighted in this review, certain classes of protein targets provide additional challenges to generate a suitable antigen format for antibody generation and screening, which have been addressed by several experimental and technical advances.

In recent years, there have been advances in structural biology and computational tools/techniques that look poised to have a dramatic impact on antibody discovery. In the structural biology area, advances in cryo-EM have increased the number of protein structures available, particularly of membrane proteins, which is advancing our knowledge. Artificial intelligence has the potential to impact many aspects of biologics discovery – driven by advances in computer hardware (e.g. graphics processing unit (GPUs) and central processing unit (CPUs)), new software algorithms, and cloud computing coupled to enhanced focus on gathering larger volumes of data, with machine learning (ML) and deep learning (DL) making a significant impact. The development of AlphaFold and RoseTTAFold has enhanced our ability to predict 3D-protein structures directly from amino acid sequences [195, 196]. Further developments in this field make de novo protein design a realistic vision. In recent years, the emergence of databases capturing sequence (e.g. Abysis) and antibody structure (e.g. SAbDab) provides opportunities for application of ML [197]. Initial applications focus on epitope and paratope identification, with developments emerging allowing application of DL to use a sequence-based approach to optimise an existing antibody candidate [198]. In another example, Hie et al. used general protein language models to evolve human antibodies without providing information on target antigen, binding specificity, or structure [199]. This allowed improvements in binding affinities for four highly matured, clinically relevant antibodies up to sevenfold, and for three unmatured antibodies, up to 160-fold. The promise here is that application of DL methods may allow de novo design of antibodies against a target antigen structure in a ‘zero-shot’ manner where all antibody designs are generated with no follow-up optimisation [200]. These advances are having a significant impact on mAb and sdAb formats and, in time, may be extended to more complex, multi-specific formats.

ML is also having an impact in allowing optimisation of membrane proteins that could be extended to drug discovery. For example, Muk et al. developed computational ML method(s) that combined sequence, structure, and dynamics-based molecular features to predict thermostabilising mutations of GPCRs [201]. In another approach, expression and functionality of channel rhodopsins were improved [202]. This type of approach could enhance our ability to generate membrane proteins with better properties for drug discovery applications. In a more recent example, Goverde et al. have opened up the exciting prospect of being able to design soluble analogues of CMPs, including claudins and GPCRs [203]. In this study, de novo protein design based on a DL pipeline was used to design complex protein folds and soluble analogues of integral membrane proteins. This used protein design based on the inversion of the AlphaFold 2 network coupled to sequence design using ProteinMPNN [204, 205]. In a tour de force of protein design, examples of claudin and GPCR proteins were designed and expressed in E.coli, followed by purification as soluble proteins. It also proved possible to solve the crystal structures of some of the designed molecules, confirming the designed fold as well as preserving native functional motifs. Whilst these proteins were not used for drug discovery, the authors noted that further developments in this area could be transformative in creating soluble analogues of membrane proteins that have been somewhat recalcitrant to antibody drug discovery. Taking these emerging themes, the prospect for wholly in silico-designed antibodies targeting membrane protein epitopes has a genuine chance to deliver.

Conclusions

The continuing growth of approved mAb therapeutics and related biologics to treat a diverse range of diseases illustrates the advances made in isolation, development, and clinical approval of these biologics. Complex integral membrane protein targets, which include GPCRs, ion channels, transporters, and adhesion molecules such as claudins, are attractive targets for biologics discovery, and significant efforts are being directed to discover antibodies and related scaffolds targeting these molecules. Antibodies that bind to the selected targets may directly modulate protein function or be re-configured into formats such as ADCs, bispecifics, or chimeric antigen receptor (CAR) T cell therapeutics, which enable alternative modes of therapeutic intervention. These complex, multi-spanning integral membrane protein targets present technical challenges for therapeutic antibody discovery for several reasons. High-level expression of these proteins in heterologous systems and subsequent extraction from the membranes as enriched or purified high-quality, conformationally stable protein antigens require significant investment of resources and diverse, often empirical strategies. Strategies for antigen generation to enable antibody isolation are influenced by the structural diversity of these proteins. In the case of proteins with large ECDs with defined structure, recombinant expression of the domains offers a relatively ‘simple’ route to generation of a suitable high-quality antigen. For such cases, this may offer a clear strategy for antibody isolation via immunisation or antibody display methods. It is often the case, however, that multiple antigen formats must be explored to ensure a stable, correctly folded protein is generated for use as an antigen. The selected, optimised antigen formats may then be used to drive antibody discovery by appropriate application of strategies that employ hybridoma, B cell platform(s), display technologies, or a combination of these techniques to ensure a diverse panel of antibodies is isolated. This field is quite dynamic, with new strategies emerging that are employing new antigen formats, antigen delivery methods, alternative hosts (e.g. transgenic mice, chickens, and camelids), conformationally constrained or engineered antigens, and a variety of microfluidic discovery platforms. In this review, we have outlined developments in the options available for antibody discovery, and which are shaping antibody generation strategies. It is our view, however, that there remains a lack of a clear consensus on discovery approaches, and often many options are explored to generate a diverse panel of antibodies binding to selected targets of interest. Generation of high-quality CMP formats can often be the rate-limiting step in a biologics drug discovery campaign. The continued growth in publications and congress reports describing identification of antibodies in pre-clinical research programmes along with increased numbers of antibodies and related biologics entering clinical studies suggest we are finding solutions to the challenges implicit in targeting complex integral membrane proteins. Developments in the structural understanding of complex membrane targets coupled to the acceleration of computational and ML technologies in understanding proteins and impacting antibody discovery hold significant promise for breakthroughs in this field of research. The pace of research in computational approaches suggests that design of antibodies entirely in silico, perhaps initially with simple frameworks such as sdAb, is a realistic ambition. The combination of enhancements to antibody discovery from both experimental and computational methods provides a reason to believe that we will see further approvals for antibodies against these complex membrane targets and potential clinical benefits to patients.

Declarations

Conflict of Interest

Amberley Stephens (AS) and Trevor Wilkinson (TW) are full-time employees of AstraZeneca, a company generating therapeutic antibodies for the treatment of a variety of diseases.

Funding

No sources of funding were used to support the writing of this review.

Ethics Approval

Not applicable.

Consent to Participate

Not applicable.

Consent for Publication

Not applicable.

Availability of Data and Materials

Not applicable.

Code Availability

Not applicable.

Author Contributions

AS and TW both contributed to the design and writing of this article.