Six-transmembrane epithelial antigen of prostate-2 (STEAP2) is a complex membrane protein that is highly expressed on prostate cancer cells with limited distal normal tissue expression. High species homology and small extracellular domains make STEAP2 a very challenging protein to target. I will share reflections on the multifaceted discovery campaigns that enabled the isolation of STEAP2-specific antibodies for the development of an armored STEAP2 chimeric antigen receptor T cell therapy.

GPCRs play critical roles in nearly all aspects of cancer progression, including growth and survival, invasion and metastasis, and drug resistance. However, signaling inhibition can be complicated by high affinity/sensitivity to ligand, constitutive activity/lack of known ligand, and the challenge of developing selective high-affinity antagonists. Thus, we propose an induced-proximity/bispecific antibody-based approach to selectively deplete GPCRs for sustained signaling inhibition, by co-opting cancer cell-surface recycling receptors.

Our group has generated antibody-drug conjugates (ADCs) directed against G protein-coupled receptor (GPCR) targets highly expressed in colorectal and other cancer types. ADCs are one of the fastest growing classes of anticancer drugs and consist of highly specific monoclonal antibodies linked to potent cytotoxic payloads. These GPCR-targeted ADCs were generally well-tolerated and show promising efficacy in preclinical models of colorectal cancer, yet tumors were not completely eliminated. Currently, we are evaluating ADCs in combination with FDA-approved therapies to enhance efficacy and overcome resistance and relapse.

T cell engagers utilizing CD3 are an increasingly validated class of multispecific antibodies, but are inherently complicated to generate. Reducing engineering complexity through a CD3-specific heavy chain–only antibody (HCAb) would be beneficial. Using a yeast-based platform, we demonstrate the discovery and engineering of a panel of high-affinity, CD3-specific, HCAbs from immunized camelids, and validate their use in functional bispecific molecules.

The 3rd generation bioSens-All platform combines BRET-based biosensors that are highly adaptable to the needs of discovery projects for small molecules, peptides, and antibodies. The platform has been successfully used internally to identify biased small molecule negative allosteric modulators for protease-activated receptor 2 (PAR2). The platform revealed different mechanisms-of-action of our lead compound when benchmarked against other antagonists of PAR2. In addition, the platform was used to develop assays for high-throughput screening for the challenging adhesion GPCR, ADGRE5. These examples will demonstrate how the bioSens-All platform was used to advance projects from discovery to preclinical candidate nomination.

Discovery of biotherapeutics against challenging targets such as integral membrane proteins, membrane protein complexes, and heavily glycosylated surface proteins using display technologies remains a challenge. We have utilized therapeutic-ready phage- and yeast-display platforms expressing a diversity of formats to pan against both cells and virus-like particles. Using these novel reagents and protocols, we have managed to discover biotherapeutics to traditionally display "unfriendly" targets.

Structural analysis of complex membrane proteins has provided valuable insights, leading to the improved design of powerful biologics. Cryo-electron microscopy has revolutionized structural characterization of biological macromolecules, enabling high-resolution analysis of drug targets previously inaccessible to structural interrogation. With the advent of cryo-EM, significant strides made through structure-based drug design approaches have augmented the specificity and efficacy of therapeutic interventions. In this presentation, we will explore the obstacles encountered by a multi-pass transmembrane protein, illustrating how structural biology can help overcome these challenges and navigate remediation of liabilities though protein engineering efforts.

Here we describe a homogenous assay for the detection of membrane protein antibodies using nanodisc-based platform. We demonstrate the feasibility of this assay by detecting HLA and ion channel antibodies. Using HLA-nanodisc complexes, we have successfully performed donor specific HLA antibody screening using serum obtained from kidney transplantation recipients.

InterAx combines real-time cellular assays and mathematical pathway modelling to define novel pharmacodynamic parameters. These parameters quantify signaling and kinetic biases, allowing to (1) determine the optimal signaling profile for therapeutic efficacy and safety and (2) apply machine learning methods to efficiently select and optimize drugs. We will show how we use this approach to discover diabetes and obesity drugs that minimize nausea without compromising effectiveness.

De novo–designed miniproteins represent a groundbreaking new modality, offering unprecedented flexibility in engineering biologics with an array of desirable features tailored for specific applications. We developed a novel GLP1R-agonist miniprotein designed de novo to exhibit improved biased signaling with a strong emphasis on G protein-coupled signaling, while significantly reducing ß-arrestin signaling.

Molecules with bioactivity for GPCRs represent a subset of the vast space of small drug-like molecules. We compare machine learning models, including dilated graph convolutional networks, that conduct binary classification to quickly identify active GPCR ligands. The models are trained and validated using 600,000+ active, inactive, and decoy compounds. The best-performing model, GPCRLigNet, was a feedforward dense neural network mapping from Morgan fingerprints to activity. Incorporation of GPCRLigNet into virtual screening is demonstrated for class A and class B GPCRs. This work provides proof of principle that GPCRLigNet can effectively hone the chemical search space towards GPCR ligands.

We developed GPCR-BERT, a transformer-based model, to understand the sequential design of G protein-coupled receptors (GPCRs). By fine-tuning the model with prediction tasks on conserved motifs, we elucidated relationships between binding pocket residues and motifs. Attention weights and hidden states were analyzed to determine amino acid contributions. Embedding analysis over 3D structures revealed higher-order interactions within receptor conformations. GPCR-BERT demonstrates the potential of language models in therapeutic design.