Profiling disease-selective drug targets: from proteomics to ligandomics

Abstract

Despite advancements in omics technologies, including proteomics and transcriptomics, identification of therapeutic targets remains challenging. Ligandomics recently emerged as a unique technology of functional proteomics for global profiling of cell-binding protein ligands. When applied to diseased versus healthy vasculatures, comparative ligandomics systematically maps novel disease-restricted ligands that allow selective targeting of pathological but not physiological pathways, providing high efficacy with intrinsic safety. In this review, we discuss the potential of cellular ligands as therapeutic targets and summarize the development of ligandomics. We further compare the advantages and limitations of different omics technologies for drug target discovery and discuss target selection criteria to improve drug R&D success rates.

Teaser:

Comparative ligandomics is a new platform to identify high-quality drug targets with wide therapeutic windows by global profiling of disease-selective cell-binding protein ligands.

Introduction

The advent of proteomics and transcriptomics has been instrumental in dissecting disease molecular mechanisms and was widely anticipated to improve drug discovery success rates and reduce pharmaceutical R&D costs. However, these technologies provided only limited success in drug development, and drug attrition rates and R&D costs remain high.^1,2 Despite the generation of extensive data by omics technologies with many prospective druggable targets, the anticipated boom in the approval rate of new drug entities by the US Food and Drug Administration (FDA) has not been realized.^1,3 Drug target discovery remains a daunting task.

With central roles in cell–cell communication (CCC), extracellular ligands and cell surface receptors regulate diverse physiological and pathological processes and have high intrinsic therapeutic potential. Given that protein ligand–receptor interactions (LRIs) can be amplified and ramified through multiple layers of intracellular signaling cascades, pharmacological interventions directed at early cell surface signaling events might be therapeutically more effective compared with interventions at locations further downstream. Indeed, of all FDA-approved drugs, plasma membrane receptors, including G-protein-coupled receptors, represent the most popular targets.⁴ However, only a small number of cellular ligands, such as insulin, vascular endothelial growth factor (VEGF), and programmed death-ligand 1 (PD-L1), have been successfully exploited as therapeutic agents or drug targets.^5,6 Unlike receptors that can be readily identified based on their transmembrane domains and cell surface expression,⁷ secreted ligands with equivalent potential as drug targets are more elusive and are traditionally discovered on a case-by-case basis with inherent technical challenges. Not all secreted proteins are cell-binding ligands. Ligandomics recently emerged as a new technology to address these challenges by globally profiling cell-binding protein ligands.^8,9

In this review, we compare ligandomics with functional proteomics, RNA-sequencing (RNA-seq) and single-cell RNA-seq (scRNA-seq) and discuss the advantages and limitations in terms of drug target discovery. We focus specifically on target validation and criteria to identify high-quality drug targets for developing novel therapies with high efficacy, wide safety margins, and potentially improved success rates. Of note, the term ‘ligand’ in this review refers to extracellular protein ligands binding to cell surface receptors but not small-molecule (SM) ligands and nonprotein ligands that bind to receptors in the cytoplasm, nucleus, or other subcellular compartments.

Functional proteomics

Proteomics is the large-scale study of protein expression and function. Expression proteomics aims to quantify protein expression in cells or tissues under defined environments. Although comparative expression proteomics of diseased versus healthy cells can reveal disease-associated proteins, the approach cannot interrogate protein–protein interactions (PPIs) and requires additional mechanistic studies that are often time-consuming and labor-intensive to investigate disease-associated proteins as drug targets (Table 1). By contrast, functional proteomics is the large-scale study of PPIs and is directly relevant to signaling regulation and drug target discovery. We summarize different technologies associated with functional proteomics below.

Table 1.

Comparison of protein-related omics technologies for drug target discovery

Feature	RNA-seq/scRNA-seq and expression proteomics	Functional proteomics	Ligandomics
Profiling capacity	Gene expression at mRNA or protein level	PPIs	Cell-binding protein ligands
Functional interaction	No	Yes	Yes
Protein subcellular location	All types of protein	Intracellular proteins	Extracellular cell-binding protein ligands
Quantification capacity	Yes: quantifies gene and protein expression	No	Yes: quantifies ligand-binding activity
Throughput capacity	Global mapping of gene and protein expression	Global mapping of bait-binding proteins	Global mapping of cell-binding ligands
Technology maturity	Well developed	Well developed	New/under development
Quantitative criteria for drug target selection	Differential expression in diseased versus healthy cells or tissues	Not available	Disease selectivity or ligand-binding activity ratio of diseased versus healthy cells
Target validation	qRT-PCR, immunohistochemistry, western blot, etc.	Individually verify PPIs by co-immunoprecipitation, protein pull-down assay, etc.	Binding assay with clonal phages, FIHC, functional assays with purified ligands, therapy assay with neutralizing pAbs
Major advantages	Highly efficient for global mapping Technologies are well developed Quantitative criteria for target selection Convenient to validate gene or protein expression	Maps functional activity rather than protein expression. Highly efficient mapping of bait-binding proteins TPP maps PPIs without AP/TAP	Global mapping of cell-wide binding ligands Binding activity quantification Disease selectivity as the quantitative criterion for target selection Multiple validation assays
Major limitations	Profiles gene/protein expression, but not function Target selection criterion is too inclusive and lacks stringency Functional validation is labor-intensive, time-consuming, and technically challenging	Maps PPIs one bait at a time by AP/TAP-MS Inability to compare diseased versus healthy cells No binding activity quantification No quantitative criteria for drug target selection Functional validation of identified targets is technically difficult	Immature technology; not yet widely used Cannot simultaneously identify ligand-binding receptors Relies on phage display-based technology with inherent limitations of protein folding and PTMs

Mass spectrometry-based functional proteomics

Mass spectrometry (MS)-based functional proteomics is one of the most efficient approaches to globally map PPIs. Briefly, bait proteins with an affinity tag are expressed in cultured cells where they complex directly or indirectly with other proteins and form interactomes that are subsequently affinity purified (AP) and identified by MS.¹⁰ Human interactomes for 338 bait proteins have been successfully mapped by AP-MS.¹¹ A major challenge of AP-MS is spurious protein purification, which generates a high degree of false positives. Tandem affinity purification (TAP), in which two different tags are attached to the same bait for sequential affinity purifications to eliminate nonspecific binding proteins, was developed to solve this problem.¹²

AP/TAP-MS has the following advantages: (i) efficient mapping of entire protein interactomes for hundreds of bait proteins to generate comprehensive PPI networks; (ii) identification of proteins with both direct or indirect binding to bait; and (iii) interactome formation in native intracellular environments, rather than cell lysates. However, the same advantages also limit the application to intracellular interactomes, rather than including LRIs on the cell surface (Table 1). Other drawbacks include mapping interactomes with one bait at a time and applicability restricted to cultured cells but not animal models.¹³ With the possible exception of cancer cell lines, it is often difficult to isolate sufficient quantities of diseased cells for large-scale AP/TAP-MS profiling ex vivo.¹⁴ Disease phenotypes can be lost in isolated primary cells during in vitro culture.¹⁵ As a result, AP/TAP-MS most efficiently maps physiological PPIs but suffers an inherently compromised capability to identify disease-associated therapeutic targets (Table 1).

To circumvent the requirement of chemical modifications or affinity tags, advanced proteomics methods, such as thermal proteome profiling (TPP), solvent-induced protein precipitation (SIP), and limited proteolysis (LiP), have been developed to profile intracellular protein interactions with SM ligands, drugs, or metabolites.^16–18 Advanced TPP was further developed to map intracellular PPIs without AP/TAP or to quantify the binding affinity of SM ligand–target interactions.^19,20

Other technologies of functional proteomics

Several other technologies have been developed to discover PPIs. The yeast two-hybrid system (Y2H) is a molecular technology that has been successfully used to identify many binding proteins, as previously reviewed.²¹ The limitations of Y2H are time-consuming and labor-intensive screening procedures with one bait at a time, detection of PPIs in yeast but not mammalian cells, and high rates of false positives because of overexpression of both bait and prey proteins in the same yeast cells.²¹

Protein microarray is another technology to detect binding proteins using surface-immobilized arrays of antibodies, proteins, or peptides.²² However, technical issues, such as protein purification, impurity, and stability, present practical challenges for the preparation of thousands of different antibodies or bait proteins. Additionally, inflexibility of the bait proteins on microarray plates, limited detection sensitivity, and interactome formation in cell lysates but not the native cytoplasm also hinder the widespread application of the technique.

Protein proximity labeling (PPL) covalently fuses an enzyme, such as biotin ligase, to a bait protein, which is expressed in cells and covalently tags both direct and indirect binding proteins with a substrate moiety in the vicinity of the enzyme.²³ Tagged proteins are affinity purified and sequenced by MS. Advantages of PPL include the labeling of proteins in live cells without cell disruption, thereby minimizing false positives and preserving transient interactions. Limitations are the high background noise resulting from nonspecific labeling of bystander proteins, low efficiency with one bait protein at a time, and applicability limited to cultured cells.

Transcriptomics

Transcriptomics globally profiles cellular RNA transcripts under defined conditions using DNA microarrays²⁴ and/or RNA-seq.²⁵ Given the drawbacks associated with microarrays, including limited detection sensitivity without signal amplification by PCR and reliance on predesigned DNA probes with little flexibility, RNA-seq has emerged as the approach of choice for transcriptomics profiling.²⁵

RNA-seq and scRNA seq

For RNA-seq, RNAs are converted to cDNA by reverse transcription, amplified by PCR, and sequenced by next-generation sequencing (NGS).²⁵ Sequencing data are aligned to databases to reveal the identity and abundance of mapped RNAs, including mRNA.

Compared with bulk RNA-seq, scRNA-seq has two major technical differences: (i) single cells are compartmentalized by sorting into individual droplets or microwells; and (ii) oligo(dT) primers for reverse transcription have nucleotide barcodes to index individual cells.²⁶ As a result, cDNAs of individual cells from different compartments can be pooled, sequenced by NGS, and indexed back to their cell identities. scRNA-seq is useful to classify cell subtypes or heterogeneity based on global mRNA profiles, whereas bulk RNA-seq lacks the sensitivity to detect changes in mRNA profiles in specific cell subpopulations. Therefore, the main advantage of scRNA-seq is its ability to profile transcriptomes regardless of heterogeneity in a cell population.²⁶ However, low capture efficiency for less abundant mRNA remains a challenge for scRNA-seq.²⁶

Computational transcriptomics and limitations

Although RNA-seq/scRNA-seq can be applied to diseased versus healthy cells or tissues to map disease-associated gene expression, these techniques are the equivalent of expression proteomics, rather than functional proteomics, to map PPIs (Table 1). An emerging trend of RNA-seq/scRNA-seq is to computationally infer PPIs and LRIs, such as transcriptional²⁷ and CCC networks,²⁸ based on curated databases of known PPIs and LRIs. For CCC inference, all known cellular ligands are extracted from scRNA-seq data to predict ligand-expressing cells, and all known receptors are extrapolated from the data to deduce ligand-responding cells. Intercellular crosstalk is inferred by matching LRIs through the databases. A major caveat of these computational approaches is the inability to predict functional PPIs and LRIs solely based on gene expression data with reliability concerns, as discussed below.

Poor correlation between mRNA and protein expression levels

Genome-wide correlation between expression levels of mRNA and protein is vital for RNA-seq data analysis to predict PPI/CCC. However, several large-scale quantitative analyses show that concordance between mRNA and protein expression ranges only from 17% to 50%.^29–32 In general, the correlation is higher in cells under a ‘steady state’ than those under a disease or evolving state.^33,34 The poor correlation is largely attributed to diverse mechanisms of post-transcriptional, translational, and post-translational regulations.

There are more than 170 post-transcriptional modifications of RNA, of which N6-methyladenosine (m⁶A) is the most prevalent, affecting over 7000 human transcripts.³⁵ m⁶A levels are regulated by RNA methyltransferases and demethylases and recognized by a group of RNA-binding proteins that regulate RNA stability, translation, splicing, and export.³⁶ Therefore, RNA posttranscriptional modifications have an important role in the discordance between mRNA and protein expression levels.

mRNA isoform switching can also contribute to the poor reliability for predicting PPIs based on the mRNA level. For example, the Bcl-x isoform, which promotes apoptosis and tumor suppression in normal cells, can switch to a different isoform that enhances the survival of cancer cells.³⁷ The relatively short read length of NGS for RNA-seq is a major limitation for accurate mapping of isoform switching.³⁸

Protein trafficking regulation can adversely affect the reliability of computational prediction. For example, the immune inhibitory molecule CTLA-4 is constitutively expressed in T cells, but its functional activity is tightly regulated by restricted protein trafficking to the cell surface and rapid internalization.³⁹ Protein translocation to different subcellular compartments, including nuclear translocation of transcriptional factors (e.g., HIF1), is a well-known mechanism that changes PPIs without altering gene expression.⁴⁰ Additionally, regulation of protein turnover rates by ubiquitination⁴⁰ can contribute to the discordance between mRNA and protein abundance.

Taken together, the above mechanisms lead to the poor correlation between the mRNA and protein expression levels, thereby reducing the reliability of inferring PPIs and LRIs by computational analyses exclusively based on RNA-seq/scRNA-seq data.

Signaling regulation of PPIs without altering mRNA expression

Not all PPIs are regulated by gene expression. Many signaling pathways are well known to be regulated only by protein modifications, such as phosphorylation, dephosphorylation, prolyl hydroxylation, acetylation, and methylation.⁴¹ These signaling regulations alter LRIs or PPIs without changes in gene expression,³³ thereby reducing the reliability of RNA-seq-based computational analyses.

Promiscuous LRIs

PPIs and LRIs are renowned for promiscuity,⁴² such as the TGFβ ligand–receptor family.⁴³ Specific TGFβ ligands can bind to multiple TGFβ receptors, and TGFβ receptors cross-react with different TGFβ ligands. Promiscuous LRIs and PPIs with variable binding affinities and availability can regulate both physiological and pathological processes, but are not reliably inferred by computational analysis of scRNA-seq data.

Coreceptor regulation

Coreceptors are distinct receptors that form heteroreceptor complexes. Examples are CCR5/CD4/CXCR4 oligomers, which regulate HIV1 entry.⁴⁴ Additionally, coreceptors include plasma membrane proteins with minimal cytosolic domain (e.g., neuropilin 1)⁴⁵ or GPI-anchored membrane proteins without transmembrane domains (e.g., human erythrocyte acetylcholinesterase),⁴⁶ many of which are yet to be identified. The computational inference of CCC networks is complicated by coreceptor-dependent LRIs.

Orphan ligands and receptors

Despite multiple PPI databases,⁴⁷ there are many unknown PPIs, including the so-called ‘orphan’ ligands/receptors with unidentified binding partners. Well-known examples include orphan G-protein-coupled receptors,⁴⁸ and the less-known secretogranin-related ligand Scg3.⁸ Computational inference of LRIs for orphan ligands and receptors based on scRNA-seq is technically impossible.

Ligandomics

Ligandomics was developed based on open reading frame phage displays (OPDs) to globally map cell-binding protein ligands. The rational design of ligandomics and its relevance to drug target discovery are discussed below.

Choice of ligandomics platforms

OPDs can appropriately display cellular proteins, including ligands, in correct reading frames, as summarized in previous reviews.^9,49 By contrast, conventional phage display is mainly used to display antibody (Ab) libraries with predictable reading frames or random short peptide libraries with minimal concern for reading frames, but not cellular proteins with unpredictable reading frames.⁵⁰ Ligandomics was developed by combining OPD and NGS in the context of cell-binding selection.⁸

Instead of other protein display techniques, phage display was chosen as a platform to develop ligandomics for the reasons described in Box 1. Given that phage display has been widely used for in vivo selection to screen for vessel-binding clones, possible phage leakage from blood vessels is an important consideration for phage vector selection. Healthy and diseased vessels are permeable to ~60 kDa (66-kDa albumin has a radius of 3.5 nm) and ~2000 kDa proteins, respectively.^51,52 Healthy fenestrated vessels, such as capillaries in the kidney, intestine, pancreas, endocrine glands, and choroids, have a pore size limit of 6–15 nm.⁵³

Box 1. Different display technologies as platforms for ligandomics.

Diverse protein display technologies have been reported, including ribosome display,⁶² mRNA display,⁶³ baculovirus display,⁶⁴ retrovirus display,⁶⁵ bacterial display,⁶⁶ yeast display,⁶⁷ mammalian cell display,⁶⁸ and phage display.⁵⁰ These display technologies have distinct advantages and limitations for ligandomics profiling.

Ribosomal and mRNA display require in vitro translated proteins to be attached to ribosomes and coding mRNAs, respectively, and are designed to screen protein binders (i.e., biopanning with immobilized protein bait), but not cell binders.^62,63 This is because of the difficulty in separating and recovering encoding mRNAs tethered to the displayed proteins from bait cells after binding selection. Additionally, bait cells can release RNase to adversely affect the stability of the tethered mRNAs, and these cell-free display systems lack posttranslational modifications (PTMs).

Baculovirus display and retrovirus display have the advantage of expressing displayed proteins in eukaryotic cells with appropriate protein folding and PTMs. However, retrovirus display is not suitable for identifying cell-binding proteins because retrovirus can nonspecifically infect mammalian bait cells.⁶⁵ Baculovirus has the same issue, even though it replicates only in insect cells.⁶⁹

Cell surface display techniques, including bacterial display, yeast display, and mammalian cell display, are designed to select protein binders but not cell binders.^66–68 This is because heterogeneous proteins on the surface of bait cells represent a major challenge for target selection by cell surface display. Additionally, steric hindrance of the interaction of bait cells with multiple target cells is another issue that prevents efficient selection of protein-displaying target cells. Furthermore, lipopolysaccharides on the bacterial surface can exert endotoxin toxicity on mammalian bait cells.

Compared with all the above protein display technologies, phage display is the most appropriate for ligandomics, as demonstrated in recent studies.^8,9,57,58 The limitations of phage display include possible misfolding of mammalian proteins and lack of PTMs in bacteria. Both nonlytic filamentous phage and lytic T7 phage have been developed as display platforms. As discussed in the main text, filamentous phage display has been widely used to display Abs and random peptides, whereas T7 phage has been applied to display cellular proteins for ligandomics to identify cell-binding protein ligands. These two phage display systems are yet to be thoroughly compared to define their advantages and limitations for ligandomics.

There are two major phage display systems: nonlytic filamentous phages and lytic T7 phages. Filamentous phages with a diameter of ~6.6 nm and length of 880–4000 nm can leak out of diseased vessels,⁵⁴ whereas icosahedral T7 phage with a diameter of 55 nm, equivalent to ~90 000 kDa, is impermeable even under disease conditions.⁸ Indeed, recent studies confirmed undetectable leakage of circulating T7 phage from neonatal retinal vasculature and fenestrated choriocapillaris with or without pathological angiogenesis,^55–57 supporting T7 phage as the preferred vector for in vivo ligand binding. Additionally, filamentous phage, but not T7 phage, requires secretion of displayed proteins into the bacterial periplasmic space, thereby introducing possible bias for mammalian protein display. To this end, T7 phage is preferred for ligandomics to globally profile cellular ligands,^8,58 whereas filamentous phage has been widely used to display Abs and random peptides with minimal issues for compatibility with bacterial secretion systems.

From ligandomics to comparative ligandomics for drug target discovery

The basic ligandomics profiling procedure includes: (i) incubation of OPD cDNA libraries with cells of interest to select cell-binding ligand clones; (ii) thorough washing of cells to remove unbound clones; (iii) recovery and amplification of cell-bound clones; (iv) reselection of cell-binding clones until all relevant cell surface receptors are fully occupied by cognate ligand clones; (v) NGS sequencing of all enriched clones to identify cell-binding ligands and quantify their respective copy numbers (i.e., relative binding activities);⁹ and (vi) data analysis.⁸

Ligandomics was initially applied to globally map phagocytosis ligands by incubating OPD libraries with phagocytes at 4°C, followed by phagocytosis at 37°C for a limited time. After removal of nonphagocytosed clones from the cell surface, phagocytosed clones are released by cell lysis for subsequent amplification, reselection, and NGS analysis. Ligandomics has been successfully applied to globally identify phagocytosis ligands in microglia and retinal pigment epithelial cells.⁵⁸

Subsequently, ligandomics was extended to profile endothelial ligands with simultaneous binding activity quantification in diabetic and healthy mice.⁸ Profiling and comparative analyses of cellular ligands for healthy versus diseased cells or vessels are termed ‘comparative ligandomics’. For drug target discovery, comparative ligandomics for diabetic versus healthy retinal vessels systematically identified ‘disease-high’ ligands with increased binding to diabetic vessels, ‘disease-low’ ligands with decreased binding, and ‘disease-unchanged’ ligands. Scg3 was discovered as a disease-high or disease-selective endothelial ligand by comparative ligandomics using two different vascular disease models: diabetic retinopathy (DR) and choroidal neovascularization (CNV).^8,57 After extensive target validation, Scg3-neutralizing monoclonal Abs and related humanized (h)Abs were generated with high efficacy to ameliorate pathological angiogenesis or vascular leakage in mouse models of CNV, retinopathy of prematurity (ROP), and DR.^8,55–57

Criteria and reliability of drug target selection

A challenge for drug R&D involves the identification of high-quality drug targets, and different omics technologies have distinct criteria and reliability for target selection. In comparative transcriptomics or proteomics, drug targets are selected based on their differential expression, instead of their functional activity, in diseased versus healthy conditions (Table 1). A dilemma is that such a lenient and inclusive criterion identifies too many targets and that subsequent large-scale functional validation consumes too much time and too many resources. Likewise, functional proteomics without binding quantification of PPIs is not designated to globally map disease-associated targets.

A unique advantage of ligandomics is to directly quantify ligand-binding activity and calculate the binding activity ratio for diseased versus healthy cells or disease selectivity for all mapped ligands (Table 1).^8,57 A case in point involves Scg3 and VEGF, ligands that were identified by ligandomics with disease-specific or indiscriminate binding activity, respectively.^8,55–57 Retinopathy of prematurity (ROP) is a retinal neovascular disease with concurrent physiological and pathological angiogenesis in preterm infants, in which healthy developing retinal vessels are highly susceptible to nonspecific angiogenic inhibition.⁵ In the mouse model of ROP, disease-selective anti-Scg3 hAbs stringently inhibit pathological but not physiological angiogenesis with a wide therapeutic window and no detectable adverse side effects.⁵⁵ By contrast, nondisease-selective VEGF inhibitors at therapeutic doses indiscriminately suppress both types of angiogenesis and confer adverse effects by blocking normal vessel development. These studies indicate the dramatic influence of disease selectivity on the therapeutic windows and confirm such selectivity as an important criterion for ligandomics in selecting optimal targets for improved drug R&D success.

Drug target validation

A major bottleneck of functional proteomics is the lack of throughput validation of identified intracellular PPIs. Protein targets identified by RNA-seq can be conveniently verified for their expression but not function (Table 1). In fact, it is technically difficult to functionally validate intracellular proteins identified by functional proteomics and RNA-seq in specific diseased versus healthy cells on a large scale.

In ligandomics, four assays were developed to independently validate identified ligands, including in vivo ligand binding (IVLB), functional immunohistochemistry (FIHC), and ‘function-first’ and ‘therapy-first’ analyses (Table 2 and Figure 1). All four assays are applicable to in vitro and in vivo settings.

Table 2.

Comparison of four types of validation assay for ligandomics

Feature	Ligand binding assay	FIHC	Function-first assay	Therapy-first assay
Assay method	Binding assay using clonal phage displaying individual ligands (Ligand-Phages)	Combines Ligand-Phage binding with immunohistochemistry	Uses purified protein ligands to modulate functional activity	Uses commercially available ligand-specific pAbs to inhibit disease activity
Readout	Quantify cell-bound Ligand-Phage number using plaque assay	Visualize cell-bound Ligand-Phage by immunostaining	Quantify cellular functions altered by ligands	Quantify disease activity inhibited by pAbs
Binding activity quantification	Yes	No (Qualitative)	Yes	Yes
Disease selectivity quantification	Yes	No	Yes	No
Throughput assay	Yes	Yes	No	No
Advantages	No protein purification and labeling required Requires only Ligand-Phages	No protein purification and labeling required Signals of bound Ligand-Phage are amplified ~400 times by tagging FLAG to phage	Characterizes functional role before mechanism research	Validates therapeutic targets before drug development
Limitations	Not all ligands displayed on phage have binding activity Need to pretest Ligand-Phage binding activity Binding domain identified by ligandomics should be included	Not all ligands displayed on phage have binding activity Need to pretest Ligand-Phage binding activity Binding domain identified by ligandomics should be included	Depends on availability of purified ligands and animal models/assay systems Gene expression can be used to replace purified ligands	Depends on availability of ligand-specific pAbs and disease models/assay systems

Figure 1.

Comparative ligandomics to discover disease-restricted targets for novel therapies. Open-reading frame phage display (OPD) cDNA libraries are amplified and purified, followed by intravenous (i.v.) injection into diseased and control animal models. After three rounds of in vivo binding selection to diseased or healthy vessels, cDNA inserts of enriched vessel-bound phages are sequenced by next-generation sequencing (NGS) with simultaneous binding activity quantification. Comparative ligandomics data analysis for diseased versus healthy vessels systematically identify disease-selective ligands, which are independently validated by four different assays: in vivo ligand binding, functional immunohistochemistry (FIHC), and function-first and therapy-first analyses. Disease selectivity confers the quality of identified targets to develop novel disease-targeted therapeutics.

IVLB and FIHC with throughput capacity can be used to validate therapeutic targets and independently confirm disease selectivity. For example, clonal phage displaying Scg3 (Scg3-Phage) and VEGF (VEGF-Phage) were used for in vivo ligand-binding assay to quantify the binding activity of Scg3-Phage or VEGF-Phage and independently compare their disease selectivity. Scg3 selectively bound to diseased but not healthy vessels, whereas VEGF bound indiscriminately.^55–57 Furthermore, FIHC was used to visualize the binding sites and patterns (e.g., diffused versus clustered signals) of Scg3-Phage and VEGF-Phage on diseased versus healthy vessels.^55–57

Traditional in vivo ligand-binding quantification and imaging assays using ¹²⁵I-labeled VEGF are unreliable, because VEGF dimers of 45 kDa freely leak out from the vasculature and cannot be fully removed by intracardial perfusion.⁵⁹ Therefore, to our knowledge, IVLB and FIHC are the only methods currently available to reliably quantify and visualize in vivo binding of endothelial ligands.^55–57

Function-first and therapy-first assays are designed to confirm functional activity and therapeutic potential of identified ligands before extensive mechanistic characterization and pharmaceutical development. The assay protocols are determined by the cell types and disease models used by ligandomics to identify each specific ligand, as described in detail elsewhere.⁴⁹ The assays take advantage of circulating extracellular protein ligands or their inhibitors, such as commercially available neutralizing polyclonal (p)Abs, to conveniently target only receptor-expressing cells, whereas validation of intracellular protein targets identified by other omics technologies is more technically challenging (see above). Indeed, these two nonthroughput assays conveniently confirmed the proangiogenic and vascular leakage activity of Scg3 and therapeutic activity of commercially available anti-Scg3 pAbs before Scg3-neutralizing mAbs were generated for drug development.^8,49

Advantages and limitations of different omics technologies

RNA-seq and expression proteomics are well-developed technologies for efficient mapping of gene and protein expression but with suboptimal target selection criteria (Table 1). Functional proteomics is not designed for drug target discovery and lacks a reliable criterion for target selection.

The ability to quantify ligand-binding activity and disease selectivity and independently validate targets represents the major advantages of ligandomics (Table 1). Conventional ligands, including angiogenic factors, were traditionally identified by case-by-case approaches based on the functional activity on healthy vessels and subsequent extrapolation to disease pathogenesis. Such angiogenic factors lack disease selectivity, and their therapeutic neutralization can pose serious side effects by simultaneously blocking parallel physiological function because of on-target side effects. Disease-selective Scg3 was first identified by molecular cloning in 1990, but was not recognized as a cellular ligand until comparative ligandomics profiling in 2017.⁸ This highlights the technical hurdles inherent in identifying disease-selective ligands by conventional case-by-case approaches without prior knowledge of disease associations. Thus, we predict the presence of other disease-selective ligands, which are supported by chemokine receptor CCR3, which is specifically expressed on CNV endothelial cells, in humans with wet age-related macular degeneration.⁶⁰ In addition to Scg3, several disease-selective angiogenic factors identified by comparative ligandomics are under independent validation as drug targets.

Limitations of ligandomics include possible misfolding of mammalian proteins displayed on the phage surface and lack of posttranslational modifications (PTMs), such as glycosylation (Table 1). However, many functionally active protein ligands expressed and purified from bacteria are commercially available, implying that such problems are relatively minimal at the proteome scale.⁸ For example, glycosylation of VEGF is necessary for its secretion but not its receptor binding.⁶¹ Ligandomics could be applied to investigate the long-lasting enigma of how many mammalian proteins can be appropriately displayed on the phage surface with functional activity at the proteome scale.

Ligandomics can identify cell-binding ligands but not their cognate receptors (Table 1). However, identified ligands can be used as molecular probes to identify their receptors. Although ligandomics can be applicable to isolated healthy and diseased cells, such applications to identify drug targets with ex vivo models have yet to be demonstrated. In another limitation, ligandomics, similar to functional proteomics and RNA-seq, is incapable of identifying nonprotein ligands, such as glucocorticoids and prostaglandins.

Concluding remarks

Most omics technologies for drug target discovery focus on high-throughput mapping of druggable targets with suboptimal target selection criteria and limited capacity to independently validate and assess the quality of identified drug targets. Ligandomics represents a unique functional proteomics platform that is distinguished from other omics technologies with the following important advantages: (i) global mapping of cell-binding ligands with high intrinsic therapeutic potential; (ii) profiling of protein function, instead of expression; and (iii) disease selectivity as a quantitative criterion for drug target selection. In ligandomics, disease selectivity as reflected by binding data is the highest priority for drug target selection, predictive of minimal on-target toxicity and wide therapeutic windows. Disease selectivity, functional activity, pathogenic role, and therapeutic potential of all identified ligands can be independently validated by four integrated assays. The discovery of Scg3 as a high-fidelity disease-selective therapeutic target highlights the validity and utility of comparative ligandomics for drug target discovery with high quality and reliability.

Highlights.

• Cellular ligands are among the most valuable drug targets

• They are traditionally identified on a case-by-case basis with technical challenges

• Ligandomics systematically maps cellular ligands and disease-restricted ligands

• Four validation assays to evaluate the quality of identified drug targets

• Scg3 was discovered as a disease-restricted target for anti-angiogenic therapy