Agonist antibody discovery: experimental, computational, and rational engineering approaches

Abstract

Agonist antibodies that activate cellular signaling have emerged as promising therapeutics for treating myriad pathologies. Unfortunately, the discovery of rare antibodies with the desired agonist functions is a major bottleneck during drug development. Nevertheless, there has been important recent progress in discovering and optimizing agonist antibodies against a variety of therapeutic targets that are activated by diverse signaling mechanisms. Herein, we review emerging high-throughput experimental and computational methods for agonist antibody discovery as well as rational molecular engineering methods for optimizing their agonist activity.

Introduction

Given that numerous pathologies are characterized by inadequate signal transduction,^1–3 strategies to restore cellular signaling to normal physiological levels or to controllably augment cellular signaling have the potential for broad therapeutic applications. Although development of natural ligands toward this goal represents a logical strategy that has achieved some success,^4–8 numerous natural ligands are not viable as therapeutics because of concerns around stability, developability, and pleiotropic activity.^9,10

Antibodies have emerged as a promising therapeutic strategy for receptor agonism.¹¹ These biomolecules often have several favorable attributes, including high affinity, specificity, stability, developability, long half-life, and potential for multiple effector functions. Thus, the race has begun to develop antibodies with agonist function, with various antibodies in clinical trials for immune cell activation. However, efficient discovery of these antibodies, which varies greatly in the level of difficulty based on the target endogenous signaling mechanisms, remains a key biotechnological hurdle.¹¹

To date, many clinically approved therapeutic antibodies inhibit cellular signaling.^12,13 The clinical translation of these antagonistic molecules provides some insight toward the development of agonist antibodies. However, lessons learned indicate that many design principles for the discovery and optimization of antagonists simply do not hold true for agonists, which illustrates the unique complexities inherent to these molecules and their efficient discovery.

Herein, we comprehensively review recent developments in the high-throughput experimental and computational discovery and optimization of agonist antibodies and other biologics (Figure 1). We address experimental discovery methodologies that use high-throughput approaches for selecting lead antibody candidates based on affinity or biological activity, and computational approaches and rational engineering approaches for generating and optimizing agonist antibodies.

Figure 1.

Overview of approaches for discovering and optimizing agonist antibodies. Affinity-based selections involve first identifying antigen-specific antibodies and subsequently screening them for agonist activity. Activity-based selections directly screen for activity during the antibody discovery process. Computational design approaches involve either predicting lead antibodies with agonist activity or antibody mutants with increased agonist activity. Rational molecular engineering methods involve using different antibody formats, valences, and Fc engineered variants to increase agonist activity.

Principles for affinity-based selection of agonist antibodies

Conventional affinity-based selection, namely in vivo (e.g., immunization and hybridoma technology) and in vitro (e.g., phage display and yeast surface display) approaches, have been used for decades for the discovery and optimization of high-affinity antibodies. These methods have displayed variable and difficult-to-predict levels of success in generating agonist antibodies, which is expected because antibodies that activate cellular receptors represent, at most, a subset of molecules that bind a given target receptor.

Therefore, a key challenge using these technologies is to improve their ability to reliably identify agonist antibodies. In particular, there has been significant progress in adapting, focusing, and biasing conventional affinity-based selection methods to discover agonist antibodies that activate cellular signaling through four specific mechanisms of action: (i) induction of receptor clustering; (ii) stabilization of ligand–receptor interactions; (iii) natural ligand mimicry via binding at a receptor active site; and (iv) allosteric binding.

Affinity-based selection: agonists that activate via receptor clustering

Endogenous receptor signaling could provide valuable insights toward activating receptors with antibodies and other biologics. Productive receptor–receptor interactions that lead to signal transduction can involve two receptors (either homo- or heterodimers) or three or more receptors (clusters). Receptors that signal naturally via ligand-induced homodimerization (e.g., various tyrosine kinase receptors) are theoretically highly amenable to agonism via IgG antibodies because of equivalent valencies.^14,15 Bivalent antibodies have the potential for simultaneous, multivalent target engagement of two receptor molecules (one per Fab arm). Several agonist antibodies have been discovered that activate receptors via dimerization.^16–18

In theory, receptors that require clustering for signal transduction represent more challenging targets, but also hold high value in regard to therapeutic significance. Such receptors (e.g., TNFα family) are widely studied for applications in activating immune cells against cancer. The TNFα family of receptors, including OX40, CD137, and CD40, are observed to signal via homotrimeric clustering and further network superclustering, which is typically induced by ligands that are likewise assembled as trimers.^11,19 Although bivalent IgG antibodies alone are typically suboptimal in the context of inducing assembly of three or more receptors, external crosslinking does enable increased activity.^11,20,21 Importantly, this has relevance in vivo given the presence of FcγR-expressing cells (e.g., effector cells) that contribute to Fc-mediated crosslinking of antibodies and can endow agonist activity.

Agonist antibodies that act through receptor clustering have been routinely generated through conventional affinity-based selections.^21–24 However, there is significant interest in focusing discovery efforts against target epitopes that are most likely to yield potent agonist antibodies and establishing principles to better understand the phenomenon of signal transduction. In one notable recent study, investigators sought to systematically identify agonist antibodies against the TNFα family receptor OX40, and characterize the potential relationship between agonist activity and antibody epitope.²¹ A panel of mouse antibodies was discovered through immunization with mouse OX40 (mOX40) extracellular domain (ECD) and hybridoma technology, and separately through panning a synthetic phage library against mOX40. The investigators characterized the antibody target epitopes based on binding to the four different extracellular cysteine-rich domains (CRDs) of OX40: CRD1 (membrane distal), CRD2 and CRD3 (both involved with ligand binding), and CRD4 (membrane proximal; Figure 2a). Next, the authors used epitope binning to identify a panel of four representative antibodies that each bind to a unique OX40 CRD (Figure 2a). The authors used this antibody panel to compare agonist activity in several in vitro and in vivo models, and identified that lead antibodies targeting CRD2 and CRD4 demonstrated consistently superior in vivo activity as judged in two biological models [B8R peptide immunization model (Figure 2b) and CT26 anti-tumor efficacy model (Figure 2c)]. Altogether, these results suggest that interaction at the ligand-binding site is not requisite for OX40 agonism.

Figure 2.

Antibodies targeted against specific OX40 receptor domains induce potent agonism. (a) Graphic illustration of differential agonist activity and ligand-blocking properties as a function of the target cysteine-rich domain (CRD). (b) Antibodies against CRD2 and CRD4 strongly enhance CD8+ T effector cell activation in a B8R peptide immunization model. (c) Antibodies targeting CRD2 and CRD4 reduce cancer progression rate in vivo in BALB/c mice. Adapted from ²¹.

The TNFα family co-stimulatory receptors have significant structural similarity and the question emerges as to whether the relationship between epitope (CRD) and agonist activity is generalizable across this receptor family. A similar investigation evaluated the relationship between agonist activity and epitope (CRD-binding) for the related TNFα family receptor CD40.²⁵ Based on the characterization of a panel of nine CD40 antibodies, including several clinical-stage antibodies, the results indicated that antibodies targeting the membrane distal CRD1 had superior agonist activity and such antibodies did not compete with the natural ligand CD40L for receptor binding. Thus, despite structural similarity among the TNFα receptor family, the relationship between epitope and agonist activity might be receptor specific.^21,25,26 Future investigations will be important to evaluate the generality of these results by testing larger panels of domain-specific agonist antibodies against various TNFα receptors.

Although the examples discussed thus far illustrate the discovery and optimization of agonist antibodies that induce homotypic clustering, agonist antibodies have also been described that are capable of agonism via heterotypic receptor clustering.²⁷ By panning a phage-displayed antibody-like protein library against the TNFα family receptors death receptor 4 (DR4) and death receptor 5 (DR5), the investigators discovered a cross-reactive DR4 and DR5 molecular therapeutic (surrobody). In both in vitro and in vivo models, this newly discovered surrobody demonstrated greater potency compared with single-specificity mAbs against either DR4 or DR5, as well as the combination of these mAbs. Mechanistically, the cross-reactive surrobody might facilitate enhanced receptor clustering via the formation of both homotypic and heterotypic signaling complexes. Overall, this work demonstrates the potential of hetero-oligomeric clustering for enhanced agonist activity, and further demonstrates that such clustering can be achieved without conventional bispecific antibodies.

Affinity-based selection: agonists that activate by stabilizing ligand--receptor interactions

Another general approach for activating cellular receptors using antibodies is stabilizing ligand–receptor interactions, which often results in augmented natural ligand-mediated signal transduction. Notably, this strategy is inherently different from traditional agonists that directly induce receptor signaling, because it is reliant on local endogenous levels of ligand for signal activation. This ‘limitation’ might have therapeutic benefit given that it might circumvent possible toxicity concerns associated with the systemic delivery of ligands or other agonists and might be especially attractive if direct agonist strategies are not suitable.

For the focused discovery of such biologics, affinity-based selection often involves positive selection for binding to ligand–receptor complexes and negative selection against binding to individual components (e.g., monomeric ligand or receptor). This strategy requires high purity of the target ligand–receptor complex, which can be complicated by the low stability of such complexes. To mitigate these stability concerns, engineered ligands with ultra-high affinity have been used in place of natural ligands.^28,29 Alternatively, molecular tethering approaches have also been used.³⁰

In one notable example, antibody fragments (camelid V_HH nanobodies) against the IL-6:gp80 (ligand–receptor) complex were discovered that selectively engage a ‘junctional epitope’ at the interface of the ligand–receptor complex.³⁰ These antibodies were discovered by immunizing camels with IL-6 fused to gp80 via a flexible peptide linker. Whereas most of the identified nanobodies recognized either IL-6 or gp80 alone, a few rare clones selectively recognized the fusion protein. The investigators solved the crystal structure of one of these rare clones in complex with IL-6:gp80 and found that this nanobody binds a newly formed epitope that almost evenly spans the junction of IL-6 and domain I of gp80. This nanobody exhibited agonist activity, increasing the intensity and duration of IL-6-mediated gp80 signal transduction.

Interestingly, similar discovery campaigns have yielded antibodies that act through unique modes of binding (i.e., other than direct engagement of ligand–receptor complex interfaces). In one notable example, investigators identified a rare antibody that stabilizes the ternary complex of IL-4 (ligand) and IL-4Rα and γ_c (two receptor subunits).^28,29 This antibody fragment was observed to interact at an epitope involving both receptor subunits, specifically at the membrane proximal ‘stem’ interface, and indirectly stabilized ligand binding. In another example, antibody fragments were identified that have interesting specificities, including allosteric, conformationally selective antibodies that bound IFNAR2 (receptor) only upon interferon (IFN, ligand) binding and others that exhibited prebinding to IFNAR2 and likewise stabilized IFN binding.³¹ This study yielded antibody fragments that increased IFN signaling potency by ~100 fold.

Affinity-based selection: agonists that activate by mimicking natural ligands

Although many agonist antibodies are described as capable of replacing ligand function, here ligand mimics are reported to both recapitulate ligand-induced receptor signal activation and bind target receptors in a manner that is competitive with the natural ligand (i.e., the antibody epitope overlaps with the ligand-binding site). Notably, general strategies have been used that directly bias discovery toward ligand-mimic antibodies. One such innovative approach is anti-idiotypic antibody discovery, in which antibodies are selected based on affinity for the variable regions of other antibodies (Figure 3a).

Figure 3.

Discovery of anti-idiotypic antibodies with agonist activity against the prolactin (PRL) receptor. (a) Graphic illustration of anti-idiotypic antibody discovery. The antibodies are first generated against the natural ligand PRL, and then serve as an antigen source for a second immunization. The resulting antibodies display agonist activity against the PRL receptor via ligand mimicry (b). The agonist mAb B-7 competes with the natural ligand (PRL) for binding. (c) The monoclonal antibody (mAb) B7 activates cellular proliferation in ligand-responsive cells. Adapted from ³².

The discovery of agonist, anti-idiotypic antibodies typically involves two immunizations in series in two animal species, and requires a known agonist molecule as an initial immunogen and scaffold for molecular discovery.32,33 Resulting polyclonal antibodies from the initial immunization are isolated, processed, and purified to remove the Fc region, and then F(ab’)2 regions are used as the immunogen for the second immunization. The resulting anti-idiotypic antibodies are isolated and screened for agonist properties.

This general approach has been used in several proof-of-concept studies for the discovery of agonist antibodies.^32,33 For example, investigators identified an antibody mimic of a growth hormone that activates the JAK2/STAT5 pathway and stimulates insulin-like growth factor secretion by porcine hepatocytes.³³ In another example, investigators discovered anti-idiotypic antibodies that mimic the binding and agonist activity of the hormone ovine prolactin (PRL) (Figure 3).³² One such antibody was further characterized, and found to compete with the ligand PRL for target receptor engagement (Figure 3b), and induce cellular responses, similar to that of PRL, albeit to a lesser extent (Figure 3c). Anti-idiotypic antibody discovery has been primarily demonstrated to replace natural peptide agonists, but could be an attractive strategy for the discovery of ligand-mimic antibodies to replace a variety of natural ligands.^32,33 However, research is warranted to establish the generality and typical efficiency of this approach.

Ligand-mimic antibodies have also been discovered via more conventional campaigns. In one report, a notable anti-PRL receptor antibody was observed to have unique biological activity compared with the natural ligand PRL.³⁴ Interestingly, this antibody induced selective, ligand-mimic activity via activation of AKT, ERK1/2, STAT1, and STAT3, but did not induce ligand-like signaling via the STAT5 pathway. Further characterization revealed that this antibody partially competes with PRL, suggesting that it engages an epitope that partially overlaps with the orthosteric site. More generally, partial orthosteric binding could represent a general strategy for antibodies to selectively mimic desirable aspects of ligand-induced signaling.

Affinity-based selection: agonists that activate via allostery

Allosteric antibodies bind to a receptor at a site other than the ligand-binding site. These antibodies are more likely to avoid conserved epitopes associated with pleiotropic binding and, therefore, are generally attractive. Allosteric agonist antibodies have been discovered via conventional ³⁵ as well as mechanism-focused strategies.^36–38 One general, mechanism-focused approach involves panning molecular libraries against a target receptor in the presence of saturating concentrations of natural ligand, thereby blocking orthosteric binding sites and biasing discovery toward molecules that bind allosterically.

For example, investigators screened for binders to insulin receptor (INSR) in the presence of saturating levels of insulin, which led to the discovery of an allosteric agonist against the insulin receptor.³⁶ The newly discovered antibody activated INSR to ~20% of the level induced by insulin, and did not inhibit insulin-mediated receptor activation.^36,37 Notably, this antibody showed only partial downstream signaling relative to insulin, selectively activating INSR through the Akt pathway, but not the Erk pathway. Overall, this antibody has the insulin-like capacity to regulate glucose levels in diabetes (Akt pathway) without recapitulating potentially deleterious, insulin-like induction of cell proliferation (Erk pathway).^37,39

Principles of activity-based selection of agonist antibodies

Although affinity-based selection represents a proven agonist discovery method, inherent limitations exist. For example, low-throughput evaluation of the biological activity of myriad lead candidates can be a laborious, inefficient and rate-limiting step. The high-throughput screening of antibody libraries directly for biological activity offers an innovative approach for use in place of, or in series with, affinity-based selection, and might show utility for efficient agonist discovery where conventional approaches have failed.

As with traditional discovery approaches, function-based screening requires a link between the sequence encoding a molecular therapeutic (genotype) and biological activity (phenotype). This genotype–phenotype linkage has been accomplished such that either a single cell has an antibody gene and reports on activity (autocrine system) or two cells are contained within close proximity in which one cell has an antibody gene and another cell reports on activity (paracrine-like system).

Activity-based selection: autocrine-based agonist discovery

Notably, direct function-based screening of libraries using mammalian cell reporter systems has been used in several proof-of-concept agonist campaigns, illustrating a diversity of discovery systems and biomedical applications. For example, discovery has successfully been used against specific cell surface receptor targets^9,40–43 as well as in target agonistic approaches^41,44–47. Antibody libraries have been displayed on the cell surface,^{9,40,43,45,46} secreted,^41,42,46 or expressed intracellularly⁴⁷. Selection strategies have been based on reporter cell gene expression (e.g., fluorescent protein linked to transcription factor activation)^9,40 as well as phenotypic responses (e.g., cellular proliferation, migration, and prevention of cell death)^{19,41,42,45,47}. Both mono-^{9,40,43,45–47} and bispecific^41,42 antibodies have been identified using this approach. Although their discovery has been accomplished primarily using in vitro systems,^{9,40–43,46,47} in vivo systems⁴⁵ have also been used. Conversely, these general function-based screening approaches have also been applied successfully for the selection of antagonist antibodies ⁴⁰.

A typical workflow of autocrine, function-based screening suing surface-displayed antibody variants involves multiple steps. First, genes encoding a library of antibodies (e.g., antibody genes fused to a single transmembrane domain via a flexible peptide linker) are cloned into a lentiviral transfer cassette. Next, lentiviral particles are prepared and used to stably integrate antibody genes into mammalian reporter cells, typically such that each cell has a single antibody gene. Upon lentiviral transduction, reporter cells constitutively express surface-displayed antibodies, allowing for interaction of the antibodies with the target receptor and, in productive cases, this event leads to downstream signal activation. Clones that activate reporter cells can be isolated by a selectable phenotype. Genomic DNA is harvested from isolated cells and used as the template DNA for amplification of lead antibody genes. The recovered genes can be evaluated for sequence identity directly or probed by next-generation sequencing. Finally, lead candidates are generated in a soluble format for additional vetting of biological activity and biophysical properties.

In autocrine systems using surface-displayed antibody libraries, antibodies are constrained in close proximity to target receptors, thereby presenting a relatively high effective concentration of lead antibody on the cell surface compared with concentrations typically used for the screening of soluble leads. As a result, these screens have reduced stringency for antibody affinity, and might be beneficial toward promoting the identification of clones with rare and desirable biological properties that might otherwise be lost during affinity-based screening. This might be beneficial, especially for target agnostic screening applications. However, when the target receptor is known, affinity-based screening can serve a complementary role in evolving antibodies that have both desirable activity and affinity in soluble formats.

For example, a recent study showcased the selection of agonist antibodies against the tyrosine kinase receptor TrkB using an autocrine reporter system in which target receptors are activated by membrane-tethered antibodies.⁹ The goal of this study was to discover antibodies that replace the cognate ligand, brain-derived neurotrophic factor (BDNF). For molecular discovery, the investigators used a two-step process, namely affinity-based selection followed by activity-based selection. First, a phage scFv library (>10¹⁰) was enriched for binding to the human TrkB ectodomain. Binding clones from the affinity-based screen were subcloned into a lentiviral transfer plasmid for mammalian surface display, transduced into TrkB model cell lines, and subjected to high-throughput, activity-based screening. Encouragingly, a top identified antibody in soluble format showed comparable TrkB agonist activity in vitro relative to the natural ligand. This example demonstrates the successful use of affinity-based pre-enrichment to focus discovery on agonist antibodies with drug-like affinity.

Notably, direct function-based screening might afford inherent advantages for molecular discovery, and show utility even when affinity-based selection alone fails. An important recent investigation sought to discover antagonist and agonist antibodies against the human apelin receptor (APJ).⁴⁰ This receptor is a member of the class A G-protein-coupled receptor (GPCR) family, which are difficult to raise antibodies against because of a limited, exposed extracellular region. For discovery, an existing nanobody library was used that was generated via immunizing camels with APJ nanodiscs and subcloning the isolated repertoire into a phagemid vector (Figure 4a). This library was previously used in an affinity-based selection campaign for functional antibodies, in which several antagonist antibodies were identified. However, no agonist antibodies were found, raising the question as to whether agonist antibodies exist within this immune repertoire.

Figure 4.

Autocrine-based functional screening for agonist nanobodies against the G-protein-coupled receptor (GPCR) apelin receptor. (a) Antibody libraries were generated by first immunizing a camel with apelin receptor (APJ) nanodiscs. The resulting immune repertoire was cloned into a phage display vector and the library was pre-enriched against APJ-binding clones using phage display. Next, the sublibrary was cloned into a lentiviral transfer plasmid for mammalian cell display via GPI-anchoring. (b) Cytograms for the parental GPI-anchored library and after the third round of activity-based sorting. A high ratio of product/substrate indicates agonist activity in the APJ B-arrestin reporter cells. (c) After three rounds of sorting, reporter cells bearing nanobody genes were sorted as single cells per well and evaluated for agonist activity. (d) The top-performing agonist nanobody (JN300) demonstrated agonist activity in the soluble format, as shown for PathHunter B-arrestin assay. Adapted from ⁴⁰.

Next, the investigators sought to discover functional antibodies using this same immune repertoire-derived library by suing phage display (one round to enrich for initial binders) followed by high-throughput, function-based screening in mammalian reporter cells (Figure 4a). For antagonist discovery, selection for lack of signal transduction was conducted in the presence of cognate natural ligand apelin at an 80% maximal dose, whereas for agonist discovery, selection for signal transduction was conducted in the absence of natural ligand. Notably, cells were subjected to three rounds of activity-based sorting with culturing between rounds as shown for agonist selection in Figure 4b. Single cells were then sorted (one per well), expanded, and evaluated for antagonist or agonist function (Figure 4c). Top lead sequences were recovered by RT-PCR, sequenced, and expressed as soluble single-domain antibodies (sdAbs). In the antagonist discovery campaign, ten lead clones were evaluated as soluble antibodies, seven of which had IC₅₀ values <200 nM. For agonist discovery, ten leads showed >50% signal activation relative to cognate ligand when evaluated in surface-displayed format. One agonist lead (JN300) was further characterized in soluble format and was observed to both bind APJ-expressing cells and agonize the APJ receptor with an EC₅₀ of 80–90 nM (Figure 4d).

Overall, this investigation contributes several innovations that warrant further discussion. First, the mammalian surface display method used is unique from previously reported approaches in that antibody library members were fused to the domain Decay Accelerating Factor, which allows for anchoring to glycosylphosphatidylinositol on lipid rafts, which is notable given that GPCRs are also commonly associated with lipid rafts.⁴⁰ Second, the successful round-over-round activity-based screening (without subcloning) is attractive for more rapid function-based antibody discovery. Although the authors do report concerns, such as increased false positives because of paracrine-like activation of cells, the investigators suggest engineered, inducible antibody display as a potential solution. Third, this investigation provides a general blueprint for the discovery of antagonist antibodies, which is notable considering that antagonist antibodies, similar to their agonist counterparts, cam represent rare sequences within a given repertoire that are challenging to select solely based on affinity. Fourth, the investigators demonstrate the use of well-defined controls for system validation (i.e., surface-displayed agonist and antagonist biologics). Overall, these principles can be broadly useful for function-based screening applications.

Activity-based screening has also been used for the discovery of bispecific agonist antibodies.^41,42 In one notable study, investigators sought to identify agonist antibodies against a known target, erythropoietin receptor (EpoR), that replace the activity of the natural ligand erythropoietin (EPO).⁴² After prescreening a naïve human antibody library for binding to EpoR and subcloning into a secreted scFv-Fc format, the investigators infected growth advantage-based reporter cells overexpressing EpoR with lentiviruses containing antibody genes at a multiplicity of infection of two (each cell should express, on average, two scFv library members). Several lead colonies were recovered from diffusion-limiting methylcellulose agar, and each colony had between one and four different antibody variants (Figure 5a). Interestingly, infecting multiple scFv-Fc antibody variants per cell led to a phenomenon referred to as intracellular combinatorial libraries, in which both monospecific and bispecific antibodies were secreted. By extension, biological activity can result from single antibodies, antibody combinations, or bispecific antibodies, thereby greatly increasing library diversity.

Figure 5.

Autocrine-based functional screening for agonist antibodies against the erythropoietin (EPO) receptor (EpoR). (a) A lentiviral antibody library is transduced into TF-1 cells in which cell proliferation is dependent on EpoR activation. Rare antibodies and combinations thereof that induce EpoR signal activation result in colony formation. Activity can theoretically result from single antibodies, antibody combinations, bispecific antibodies, or combinations thereof. (b) In vitro analysis reveals that the most-active molecular species is a bispecific antibody V-1/V-2, whereas monoclonal antibodies V1 and V2 do not show agonist activity either alone or in combination. (c) The mechanism of action appears to be that the bispecific antibody V-1/V-2 forces EpoR dimers into a conformation similar to that induced by the natural ligand (EPO). Adapted from ⁴².

In this study, single antibodies exhibited only modest in vitro bioactivity, with the top performing agonist reaching ~60% of the activity of cognate natural ligand EPO. The investigators then examined ~50 combinations of antibody genes (each gene pair identified within the same colony) for in vitro bioactivity and observed that most mixtures, containing both single and bispecific antibodies, demonstrated improved activity. One particular gene combination that showed synergy (V-1/V-2) and comparable bioactivity to the natural ligand EPO was further evaluated. To determine the molecular species responsible for the observed bioactivity, the investigators produced all possible molecular constructs (Figure 5b). Interestingly, monospecific antibodies and monospecific antibody mixtures did not exhibit agonist activity, whereas the V-1/V-2 bispecific antibody showed agonist activity comparable to that of EPO (Figure 5b). Further competition analysis showed that V-2 is competitive with EPO binding, whereas V-1 binds EpoR at a nonoverlapping epitope, offering potential mechanistic insight into the observed activity (Figure 5c). Overall, this work demonstrates the successful high-throughput discovery of a high potency biepitopic antibody; further investigation is warranted toward evaluating this general approach in applications in which biepitopic antibodies are expected to be advantageous, such as against target receptors that mechanistically signal via higher order receptor clustering. Furthermore, this strategy might be broadly useful for the high-throughput identification and evaluation of bispecific antibodies and synergistic antibody combinations.

Beyond function-based screening for antibody discovery directed against specific targets, several interesting examples highlight targeted agnostic discovery.^41,44–46 In one innovative demonstration of this approach, investigators sought to discover antibodies capable of replacing various Yamanaka transcription factors (Sox2, c-Myc, and Oct3/4 and Lkf4).⁴⁶ Expression of these key factors in trans (e.g., gene delivery via lentiviral transduction) represents a standard method for cellular reprogramming to generate induced pluripotent stem cells (iPSCs), and antibodies capable of replacing these factors are significant for advancing iPSC-reprogramming capabilities in vitro and in vivo. In this study, the investigators identified an antibody to replace Sox2 and found that it binds and antagonizes brain acid soluble protein 1 (Basp1), a protein that was not previously reported to have a role in cellular reprogramming. In another noteworthy study, investigators directly identified antibodies that promote cellular migration to the brain for applications in regenerative medicine and the treatment of neurodegenerative disease.⁴⁵ Overall, these examples and others highlight function-based screening for a broad range of biomedical applications.

Activity-based selection: paracrine-based agonist discovery

Paracrine-like systems combine two different cell types, in which one cell type produces an antibody or biologic that acts on another cell type to induce signal transduction. Although genotype and phenotype are contained within two separate cells, genotype–phenotype linkage can be retained typically via diffusion-limiting environments. To date, paracrine-like systems have been devised that either combine cells of the same species or cells of unique species. Although developing such systems incurs the inherent challenge and complexity of co-culturing multiple cell types together, these systems also afford potential advantages associated with each cell type. For example, organisms such as phage, bacteria, and yeast are commonly used for in vitro surface display technology, and are attractive because of the capacity for large library size, ease of affinity-based selection, and facile iterative sorting for enrichment of lead clones. Mammalian cellular systems are amenable to function-based, high-throughput screening, and are attractive for the expression and presentation of target antigens in a biologically relevant context.

In one notable example, functional antibodies were selected using the simple approach of fluorescence-based microscopy and cell isolation via micropipettors, which was commensurate in throughput to fluorescence-activated cell sorting (FACS).⁴⁸ In this study, the investigators identified agonist antibodies against DR4 and DR5 by co-encapsulating primary B cells (derived from chickens immunized with DR4 and DR5) and reporter cells in low melting-point agarose-based microdroplets (~100 m diameter). Cells with functional antibodies were isolated from large numbers of primary B cells based on fluorescence patterns that report on antigen binding and apoptosis response.

Co-culture systems combining multiple species have also been reported. For example, investigators sought to combine the traditional approach of phage display with function-based screening by developing a paracrine-like agonist selection system in which phage-producing Escherichia coli were co-encapsulated with mammalian reporter cells in microdroplet ecosystems.⁴⁹ To establish proof of concept, several studies were first conducted in the absence of microdroplet ecosystems. The investigators co-cultured E. coli-producing phages that displayed an anti-TrkB antibody with TrkB reporter cells and demonstrated a significant increase in cellular activation compared with treatment with E. coli-producing random phage. Furthermore, the investigators demonstrated that the majority of mammalian cells were viable after 24 h of co-culture with E. coli. Next, using the picoliter-sized droplet system, the investigators observed that bacteria produced sufficient phage to induce reporter cell activation. Further investigation might be warranted to develop this similar system for FACS-based screening by using microdroplets that are stable in the aqueous phase.

Activity-based selection: emerging technologies

Several investigations establish innovative yeast–mammalian co-culture systems that could be further adapted for high-throughput activity-based antibody screening.^50–52 For example, using an affinity-based selection approach, investigators cultured a yeast-displayed library with viable mammalian cells overexpressing the target receptor of interest, namely human acid-sensing ion channel receptor 1a.⁵⁰ Resulting yeast–mammalian cell complexes were screened for antigen binding by FACS, which allowed for the maturation of an antagonist antibody with enhanced affinity and potency relative to the parent molecule. Another potentially untapped opportunity for function-based discovery involves yeast reporter cells. Interestingly, yeast cells have been engineered to report on receptor activation (analogous to mammalian reporter cells) via expression of reporter genes linked to signaling pathways relevant to human disease.^53,54

Computational tools for agonist antibody discovery and engineering

Myriad computational approaches have been used to improve the antibody discovery and engineering process. One task that is well suited for computation is mutational scanning of the antibody–antigen interface, which can be applied to identify key antibody residues that are important for binding interactions. Many software suites can perform this task, including Rosetta,^55,56 FoldX,^57,58 and SAAMBE-3D.⁵⁹ Rosetta and FoldX use built-in energy functions to compute the energies of protein assemblies before and after mutations to determine the energetic effects of such mutations. With these tools, several positions can be mutated simultaneously. By contrast, SAAMBE-3D is a machine-learning algorithm that can quickly calculate the change in binding energy for a single amino acid substitution. SAAMBE-3D was trained on the SKEMPI 2.0 database, which comprises paired mutations and experimentally determined G of binding. As mutational data sets continue to grow, the utility of machine learning should likewise advance as a powerful and accurate method for such applications. Each of these approaches traditionally relies on crystal structures of the antibody–antigen complex, which might be unknown and difficult to produce. However, in some of these cases, a crystal structure exists of the antibody and antigen individually, but not in complex with each other. Several docking protocols, including HDOCK,⁶⁰ ZDOCK,⁶¹ and RosettaDock,⁶² can compute the antibody–antigen complex from the individual structures. In other cases, only the antigen structure is known and, in this scenario, a homology model of the antibody can be created using computational tools, such as Rosetta⁶³ and FoldX,⁶⁴ which can then be used in docking protocols. In this case, it is helpful to have experimental measurements to determine which residues are involved in antigen binding to assess whether the homology models and docking procedures accurately reflect the binding interactions.

Structure-guided agonist discovery

These computational methods are increasingly being used in concert with experimentally determined structural information to design agonist antibodies and other biologics.^10,65,66 In one notable example, an antagonistic sdAb discovered via immunization was converted into an agonist through rational mutation, guided by structural data.⁶⁵ In this study, the investigators sought to develop an agonist sdAb against the class A GPCR APJ, which could be used for the treatment of chronic heart failure. Toward this goal, the authors immunized camels with nanodiscs containing thermally and conformationally stabilized APJ and used phage display of the immune library to isolate APJ binding sdAbs. Next, resulting antibodies were screened for antagonist and agonist properties. The authors identified 186 unique sdAbs that bound to APJ with picomolar to nanomolar range affinities. Of these, 106 were potent antagonists and the remaining 80 were neutral binders. Unfortunately, no agonist antibodies were discovered.

Toward the unfulfilled goal of agonist discovery, the investigators next sought to develop a rational design method to convert one potent antagonistic sdAb (JN241) identified from the initial campaign into an agonist.⁶⁵ JN241 had the longest CDR3 of all the antibodies identified. Moreover, JN241 engaged APJ at an epitope overlapping with the ligand-binding site. The investigators solved the crystal structure of the antibody–APJ complex, which was used to identify key interactions between APJ and JN241. Several APJ and JN241 mutants were created with alanine substituted at these crucial positions. The binding of JN241 and mutant APJ was tested as well as that of APJ and mutant JN241. Importantly, the investigators found that alanine mutations in CDR3 that were located in the ligand-binding pocket did not disrupt the binding interaction. Thus, mutations in this region could be used to convert this sdAb into an agonist.

Based on previous studies involving a potent anti-APJ peptide agonist AMG3054, the authors noted that the antagonist sdAb JN241 lacked hydrophobic interactions with two residues of APJ that were shown to be important for activating the receptor.⁶⁵ They hypothesized that either substituting in aromatic residues (tyrosine, phenylalanine, or tryptophan) near the tip of CDR3 or inserting these residues could result in the formation of essential hydrophobic interactions that would transform the antagonist into an agonist (Figure 6a). Of the three insertion mutants, tyrosine insertion resulted in agonism with EC₅₀ values between 36 and 47 nM (Figure 6b,c). Importantly, this example highlights how antagonist molecules that conversely inhibit biological signaling hold value for agonist discovery when used in the proper context, offering a new path for discovery.

Figure 6.

Conversion of an antagonistic nanobody into an agonist nanobody by rational design. (a) Insertion of a tyrosine into the tip of CDR3 of the wild-type nanobody results in new contacts that also occur between the natural ligand and receptor. In vitro analysis reveals that this insertion converts the wild-type antagonist into a full agonist, as the agonist displays poor receptor inhibition (b) and strong receptor activation (c). Adapted from [55].

Computational approaches can also be broadly useful for tuning the specificity and safety profile of biologics. In one example, investigators computationally designed IL-2 cytokine-mimetic proteins with improved safety profiles compared with the natural cytokine by reducing polyspecificity.¹⁰ By extension, computational approaches might also have value for reducing the toxicity of agonist therapeutics that are limited clinically.

Agonist Fc engineering

In addition to optimizing antibody variable domains for affinity and biological activity, Fc engineering is another important avenue for improving agonist therapeutics. The Fc region of antibodies, comprising the C_H2 and C_H3 domains, has the potential to alter in vivo cellular signaling as a result of Fc-mediated crosslinking, engineered Fc–Fc interactions, and isotype specific conformations.

One engineering strategy reported to enhance agonist activity is controlling Fc interactions with Fc receptors (Fc Rs). There are six human Fc Rs (i.e., Fc RI, Fc RIIA, Fc RIIB, Fc RIIC, Fc RIIIA, and Fc RIIIB), and each has a unique affinity and specificity for different human IgG subclasses of antibody. There are several important studies that highlight the influence of Fc–Fc R interactions on agonist activity, such as by introducing Fc mutations to enhance the affinity to Fc RIIB while reducing or eliminating affinity for other Fc Rs.67–74 In one study, investigators introduced mutations in the CH2 domain of a CD40 agonist IgG1 antibody to find Fc variants with improved Fc RIIB binding.⁶⁹ The investigators discovered several mutations that increased binding affinity to Fc RIIB while reducing affinity to Fc RIIA. Interestingly, one such Fc mutant displayed a 96-fold increase in binding to Fc RIIB, which led to a 25-fold increase for in vitro agonist activity compared to wild type. Furthermore, in vivo antitumor studies showed that the Fc mutant significantly improved the antitumor response relative to control antibodies. Similarly, agonist antibodies against other immune receptors, such as CD137, have also been reported with improved agonist activity resulting from enhanced affinity to Fc RIIB receptors ⁶⁷. One potential mechanistic explanation is the divergent internalization pathways associated with these receptors, because Fc RIIA is typically degraded upon Fc binding whereas Fc RIIB is typically recycled,^75,76 thereby enabling a higher degree of clustering. Collectively, these results demonstrate the importance of optimizing Fc–receptor binding affinity and selectivity as an approach to improve agonist antibody activity.

Although engineering Fc–Fc R interactions is an important approach for optimizing agonist antibodies, this approach relies on optimal expression of Fc Rs on effector cells, which has been reported to vary significantly among Fc R-expressing cells and is challenging to predict in vivo ^77–79. One innovative strategy to obviate this limitation is to crosslink antibodies in an Fc R-independent manner. Engineering Fc–Fc interactions has emerged as an innovative strategy toward this goal. In one important study, the investigators demonstrated that Fc mutations, T437R and K248E, facilitated hexamerization of antibody Fc regions only when bound to OX40, thereby promoting clustering of antibody-bound receptors.⁶⁸ Crystal structures revealed that these mutations promoted stabilizing interactions between Fc regions when in close proximity. Using an in vitro reporter assay, the Fc mutants showed a 30% improvement in Fc R-independent agonist activity compared with the natural ligand.

Antibody isotype has also been observed to influence agonist activity as a result of molecular conformation and geometry. One study evaluated the impact of IgG subclass on agonist activity of a CD40 antibody.⁷¹ The authors identified that IgG2 isotype antibodies induced significantly improved T cell activation in Fc RIIB-knockout mice compared with IgG1, and induced agonist activity in an Fc R-independent manner. Mechanistically, the authors established that the C_H1 and hinge regions have a significant role in the observed improved activity. For example, the investigators observed that the h2B isoform of IgG2 more potently elicited cellular signaling compared with other IgG2 isoforms. This isoform involves rearrangement of two of the hinge disulfide bonds to form new disulfide bonds with C_L and C_H1, allowing the antibody to adopt a more compact conformation, such that Fab arms are located in close proximity to the hinge region. This compact conformation enables close packing of target receptors, and might be attractive in the context of signal transduction via receptor-mediated clustering. Similarly, another report also demonstrated the unique ability of the h2B isoform of an IgG2 antibody to show improved biological activity against the immune receptor CD200R.⁷² Overall, these studies add to the burgeoning body of evidence supporting Fc engineering and optimization to improve agonist activity.

Engineering agonist valency and specificity

Rational molecular engineering to optimize valency and specificity has emerged as another promising route to augment agonist activity and improve safety, and is especially relevant for antibodies that act via receptor clustering. A recent study evaluated the potential of both increased valency and targeting multiple epitopes for enhancing Fc-crosslinking independent OX40 agonism.²⁰ The investigators systematically evaluated the activity of bivalent and tetravalent antibodies that incorporate either one type of antibody-binding domain (monoepitopic) or two different antibody-binding domains specific for nonoverlapping epitopes (biepitopic). The tetravalent antibodies were constructed in a variety of molecular formats, including a dual variable domain format (DVD). For monoepitopic constructs, tetravalent antibodies showed improved Fc-crosslinking-independent bioactivity compared with bivalent controls. Moreover, tetravalent biepitopic variants showed superior activity in vitro in T cell models relative to all constructs in the panel (Figure 7c). Importantly, tetravalent DVD antibodies exhibited similar pharmacokinetic profiles relative to IgG controls in vivo in a mouse model. Furthermore, tetravalent biepitopic antibodies showed superior pharmacodynamic profiles in a standard model to observe T cell-dependent immune response (Figure 7d). Interestingly, tetravalent biepitopic antibodies achieved maximum agonism even when lacking affinity for FcγRs in this in vivo model.

Figure 7.

Tetravalent biepitopic targeting of OX40 improves FcγR-independent receptor agonism. (a) OX40:Fab1:Fab2 ternary complex [adapted from Protein Data Bank (PDB): 60GX] illustrates Fab1 and Fab2 binding to unique, nonoverlapping epitopes on the OX40 receptor. (b) The tetravalent, biepitopic antibody is proposed to engage two OX40 molecules at unique epitopes and promote daisy-chain-like, higher order receptor clustering. (c) In vitro activation of CD4+ T cells without FcγR crosslinking or CD28 co-stimulation demonstrates superior agonism by tetravalent, biepitopic antibodies. (d) In vivo activation of CD4+ effector memory T cells in a KLH-immunization in human OX40 knock-in mice reveals superior agonism by tetravalent, biepitopic molecules independent of FcγR-mediated clustering. Adapted from ²⁰.

To better understand the phenomenon by which tetravalent biepitopic antibodies agonize OX40 receptors, the investigators solved the crystal structure of the ternary complex of OX40:Fab1:Fab2 (Figure 7a).²⁰ These structural data suggest that Fab1 and Fab2 are incapable of binding a single OX40 receptor molecule when used in a tetravalent, biepitopic DVD format. The geometric constraints of this format bias binding toward two OX40 receptors, one receptor by Fab arm 1 and one receptor by Fab arm 2, thereby favoring the assembly of higher order OX40 receptor complexes (Figure 7b). Moreover, the investigators used size-exclusion chromatography to assess the degree of receptor crosslinking by incubating various molecular constructs in either the absence or presence of OX40 receptor extracellular domain. Interestingly, co-incubation of biepitopic tetravalent antibodies with OX40 ECD resulted in the largest antibody–receptor complexes. Receptor complexation was also improved for tetravalent monoepitopic antibodies relative to bivalent monoepitopic antibodies, albeit to a lesser extent. Mechanistically, biepitopic binding can in theory afford extensive daisy-chain-like receptor crosslinking (Figure 7b) as long as receptor engagement is biased toward binding of multiple OX40 receptors.

Biepitopic targeting and higher order valency were recently applied in another notable example, further highlighting the utility of this emerging rational approach.⁸⁰ Investigators explored a DR5 agonist treatment comprising an equimolar mixture of two DR5 antibodies that bind nonoverlapping epitopes. Each antibody contained an Fc mutation observed to induce IgG hexamerization upon binding to target DR5 on the cell surface. Agonism was observed to be independent of FcγR-mediated antibody crosslinking. Furthermore, biepitopic treatment was demonstrated to induce superior agonist response relative to monoepitopic control treatments with the same Fc mutations in both in vitro and in vivo models. Further analysis revealed that heterohexamer assembly induced complete DR5 agonism, as opposed to mixtures of homohexamer molecular species. Based on promising in vitro and in vivo preclinical results in multiple cancer models, the treatment is in an active clinical trial (GEN1029) against solid malignant tumors.

Bispecific engineering has also been applied to improve drug safety and efficacy.^81–85 A recent investigation improved the safety profile of an antifibroblast growth factor receptor 1 agonist antibody by focusing receptor agonism to target adipose tissue via co-targeting of β-Klotho.⁸¹ Likewise, another study showed reduced hepatotoxicity of a CD137 agonist via bispecific engineering to bind epidermal growth factor receptor for tumor localization.^82,83 One interesting study reported a tetravalent, bispecific antibody that targets both CD137 and OX40.⁸⁶ The bispecific tetravalent molecule showed reduced liver T cell infiltration relative to CD137 mAb control. Notably, the antibody also exhibited FcyR-independent T cell activation, despite having only two binding domains against each target per antibody molecule. This activation was dependent on co-target engagement. Further investigation is warranted to elucidate the underlying mechanism of T cell activation for this tetravalent bispecific molecule.

In addition to rational engineering approaches, natural antibody formats also endow valency beyond two binding sites and might be attractive for receptor clustering-mediated receptor agonism. For example, IgM antibodies (deca- to dodecavalent) are produced naturally by B cells during initial antigen response. Although IgM isotype antibodies are less explored for clinical applications compared with IgG antibodies, IgM might have utility in a variety of biomedical applications in which strong avidity effects are desired.⁸⁷ One recent investigation reports favorably on the development of IgM agonist antibodies for the treatment of ocular disease via molecular targeting of the receptor tyrosine kinase Tie2 and DR4.

Developability

The translation of agonist antibodies to the clinic is challenging, presenting both general antibody developability hurdles and also those unique to agonists. Biophysical properties, manufacturability, safety, and efficacy are all important considerations for predicting and determining the likelihood that an agonist lead candidate will progress from discovery to the clinic. Several approaches have been used for early-stage antibody developability assessment, including experimental techniques to determine antibody self-interactions, cross-interactions, expression titer, and stability, and computational methods to predict biophysical properties based on antibody sequences and structures.^88–95 Agonist antibodies also require unique considerations for dosing and administration to demonstrate in vivo safety and efficacy, and avoid adverse effects associated with on-target and off-target binding. Previous investigations pursuing natural ligands and ligand–Fc fusion proteins for therapeutic applications in the clinic could offer insights in this regard.^96–100 Natural ligand-induced cellular signal transduction is tightly regulated and often short-lived, and natural ligands themselves generally have low stability and are degraded rapidly.¹⁰¹ Successfully applying antibodies as agonist therapeutics will require careful optimization of dosing and administration to effectively control signal transduction both temporally and spatially.

Although engineered bispecific and multivalent antibodies have emerged as promising formats for next-generation agonists with increased functionality and improved safety and efficacy, these complex formats also demand specialized developability considerations. For manufacturing, it is important to efficiently produce antibodies at high yield and purity using standard biotechnology platforms.⁸⁸ Initial efforts at producing bispecific antibodies were frustrated by the formation of products that are difficult to separate and heterogeneous because of heavy chain and light chain mispairing. Importantly, these heavy chain¹⁰² and light chain^103,104 pairing problems have largely been addressed, allowing for the facile production of pure, homogenous bispecific antibody products of interest. For engineered multivalent antibodies,^105–107 several molecular formats are also available with excellent manufacturability, including for tetravalent^108,109 and hexavalent formats.⁸⁰ Beyond manufacturing considerations, a multi-tiered approach could improve early-stage developability assessment by first evaluating developability on the basis of each constituent antibody domain, and then assessing developability of the engineered molecule.⁸⁸ Encouragingly, several bispecific antibodies have received US Food and Drug Administration (FDA) approval, including blinotumomab (Amgen), emicizumab (Genentech), and amivantamab (Jannsen).^106,107,110 This list notably includes bispecific antibodies with agonist function, such as the FDA-approved bispecific T cell engager (blinotumomab), which simultaneously engages CD19 on malignant B cells and binds and activates CD3 on T cells. Overall, significant groundwork has been established to facilitate the development of agonist therapeutics. Continued advancements toward the early and accurate assessment of lead candidate developability will be crucial for improving the reliable clinical translation of agonist biologics, and avoiding costly failures during clinical trials.^88,89

Future directions

Several first-generation agonist antibodies encountered challenges in clinical development, including toxicity, off-target effects, and lack of efficacy, ultimately thwarting their clinical use.^11,26 Considering the future of agonists in the clinic, a new wave of agonist therapeutics could include molecules that bypass the typically IgG format to improve activity and safety, especially if inherent developability challenges because of molecular complexity can be addressed.

Toward future agonist discovery, strategies that enable high-throughput, function-based selection of soluble format agonist antibodies that are amenable to screening in a variety of antibody molecular formats (of varying valency and specificity) would be attractive. In this regard, new hybrid platforms aimed at leveraging combined strengths and hedging against weaknesses represent one promising future direction to improve agonist discovery. Improving the general reliability of agonist discovery would also represent a tremendous advancement, and strategies to effectively derisk discovery are needed in future campaigns. One key risk factor is antibody molecular format, including the screening format (e.g., surface-displayed antibody) versus drug format (e.g., soluble IgG). A second key risk factor for future consideration is the antigen (experimental versus physiological). More specifically, lead molecules screened for binding to simplified versions of antigens might lack or show impaired binding to the antigen in a physiological context.^22,111

Another important future direction is to establish robust, mechanism-focused discovery strategies and principles, which could include function-based screening on the basis of receptor class (e.g., for ion-channel linked, G-protein-coupled, enzyme-linked receptors, or subsets thereof) or similarly on the basis of endogenous receptor signaling mechanism (e.g., dimerization, higher order oligomerization, and conformational changes). Mechanism-focused discovery strategies hold promise for developing agonists against the numerous target receptors that have not yet been successfully targeted.

Concluding remarks

Affinity-based, activity-based, and structure-guided approaches have enabled the discovery of numerous agonist antibodies and biologics. Molecular engineering has also served a crucial role in optimizing agonist function, remedying the inherent limitations of bivalent IgG monoclonal antibodies to adequately recapitulate the activity of many natural ligands. Moreover, endogenous ligand–receptor interactions have provided a valuable blueprint toward receptor signal transduction and warrant thorough consideration to inform agonist discovery. Overall, agonist antibodies hold tremendous potential to treat myriad pathological conditions via the activation of cellular processes. However, continued research, development, and innovation are needed before these biomolecules can achieve broad clinical use. The race is on to establish general strategies to efficiently and reliably identify rare agonist antibodies with drug-like safety and efficacy.

Highlights.

• Agonist antibodies have emerged as promising therapeutics for receptor agonism

• Discovery methods have been adapted based on the endogenous signaling mechanisms

• High-throughput screening for agonist function enables improved discovery efficiency

• Computational methods are being used to design and optimize agonist antibodies

• Molecular engineering has remedied the limited activity and safety of some IgGs

Biography

John S. Schardt earned his BSc from Lehigh University, and his PhD from the University of Maryland in bioengineering. His thesis focused on multivalent engineering of molecular therapeutics for improved cancer therapy. He studied jointly as a fellow with the NIH National Cancer Institute on this research. His postdoctoral research at the University of Michigan focuses on the discovery and engineering of immune co-stimulatory agonist antibodies, and is supported by a NIGMS F32 award. More broadly, his research interests include biologics discovery, function-based high-throughput screening, and molecular engineering.

Harkamal S. Jhajj received his BSc in chemistry from the University of Michigan-Dearborn in 2013. Before entering graduate school, he worked as a research technician at Massachusetts General Hospital, Harvard Medical School, where his work focused on the role of inflammation in vascular diseases. He is currently a PhD student in the Biomedical Engineering Department at the University of Michigan. Specifically, his research work focuses on discovering novel agonist antibodies to modulate the immune system. Broadly, his research interests include drug discovery, protein engineering, and immunology.

Peter M. Tessier received his PhD in chemical engineering from the University of Delaware and performed his postdoctoral studies at the Whitehead Institute for Biomedical Research at MIT. He is the Albert M. Mattocks Professor in the Departments of Chemical Engineering, Pharmaceutical Sciences and Biomedical Engineering, and a member of the Biointerfaces Institute at the University of Michigan. His research focuses on designing, optimizing, characterizing, and formulating therapeutic antibodies, a class of protein therapeutics that hold great potential for detecting and treating human disorders ranging from cancer to neurodegenerative diseases.