Massively parallel, high-throughput measurements of protein function are essential for deep learning approaches promising to revolutionize antibody engineering. In this talk, I will describe our latest technologies enabling near-comprehensive exploration of functional sequence space. I will show the mapping of antibody escape mutants for the SARS-CoV-2 Spike RBD. Then, I will describe our antibody yeast display Fab platform which allows tracking of simultaneous heavy and light chain mutations.

Antibody repertoire established from previous exposures can persist in circulation and exert a major influence on the nature of subsequent responses. However, determining the relative contributions from pre-existing and newly-elicited antibodies can be difficult. We combine B cell repertoire sequencing with liquid-chromatography tandem mass-spectrometry proteomics to identify and quantify individual antibody clonotypes across multiple time points for in-depth analysis of the immunological memory and longevity of antibody responses in circulation.

There has been a surge of interest in the potential of adaptive immune receptor repertoires to serve as a source of diagnostic and prognostic biomarkers. In this talk, I will discuss machine learning approaches for discovering repertoire sequence patterns that distinguish individuals with a shared clinical phenotype or outcome from control groups. Examples will include B cell receptor patterns associated with multiple sclerosis and T cell receptor patterns associated with cancer and cancer outcomes. Additionally, I will discuss the challenges of translating these approaches to the clinic.  

Antibody discovery relies on screening tools that allow for efficient prioritization and selection of candidates for downstream characterization. Typically, hundreds to thousands of antibodies must be screened, expressed, and tested to identify neutralizing antibody candidates for further characterization. To overcome these obstacles, we developed LIBRA-seq, the high-throughput antibody discovery platform that maps paired heavy/light chain antibody sequences to antigen specificity using next-generation sequencing. Furthermore, we have incorporated new features to LIBRA-seq, including ligand blocking to evaluate antibody functionality at the screening step. Overall, LIBRA-seq presents a general platform with applications to virtually any area targeting the identification of antibody candidates.

In silico design is emerging as an alternative way to generate nanobodies, facilitating the discovery of nanobodies that target epitopes with pre-determined biological functions. However, binding affinity of the designed nanobodies is hardly optimal, limiting their practical application. We sought to introduce in vitro selection from a synthetic library to optimize the computer-designed nanobodies. The same pipeline is repurposable to identify novel epitope-specific nanobodies with high binding affinity. With further assistance from computational tools, the selected nanobodies are predicted to have optimal stability and solubility.

Antibody discovery technologies have made rapid progress against simple targets like soluble ectodomains, but discovery remains difficult against proteins like diverse/broad viral families, disordered proteins, and membrane proteins, delaying new antibody drug development against difficult targets. Here we will share case studies of effective library-scale approaches to engineer new antibodies against the disordered malaria circumsporozoite protein (CSP), broad HIV-1 viral strains, and against membrane protein targets.

In this talk, we will discuss a new process for the single cell characterization of antigen-specific B cells from humanely sampled peripheral blood of living mice. We leverage this protocol to screen the sequence diversity of the immune repertoire during immunization and used the information to steer the immune response toward our design goal.

Characterizing immune response during disease or profiling genetically modified immune cells for cancer treatment has enabled new challenges in quantifying functional immune response. Here, I describe two examples using the dynamic single-cell technology platforms we developed to address these problems: (1) the use of integrated dynamic and transcriptional single-cell profiling to identify clinical response biomarkers for T-cell therapy; (2) the nature of the cytotoxic T-cell responses elicited upon SARS-CoV-2 infection.

Epstein-Barr virus (EBV) infects over 80% of the population and periodically reactivates causingantibody response to EBV contains epitopes that mimic host proteins and epitope response differences could be important to understanding disease. We developed an EBV epitope peptide microarray to resolve epitope level differences in IgG and IgM responses in a COVID +/- cohort.

We have developed a comprehensive process of NGS-based analysis of phage panning to follow the enrichment of antibody sequences along the different panning rounds. To obtain information on cognate chain pairs we used long-read PacBio sequencing method and developed an NGS analysis pipeline to analyze the enrichment of paired VH/VL sequences of antibodies. Our results indicated a 125% increase in identifying novel antibody sequences over traditional panning method.

Integral membrane proteins are the targets of many therapeutics, including antibodies. To accelerate the process of antibody discovery in an epitope-specific manner we have devised a novel approach to generate single particle cryoEM reconstructions of heterogeneous polyclonal antibody-antigen complexes at high resolution directly from immune sera. By combining these data with B cell receptor repertoire analysis we circumvent the need to isolate individual B cells and generate monoclonal antibodies for further characterization.

Polyreactive antibodies compromise screening pipelines and are generally intractable for clinical development. We designed a set of experiments using a synthetic camelid antibody fragment (‘nanobody’) library to enable machine learning models to accurately assess polyreactivity from protein sequence (AUC > 0.8). Our models provide quantitative scoring metrics that predict the effect of amino acid substitutions on polyreactivity. We experimentally tested our models’ performance on three independent nanobody scaffolds, where over 90% of predicted substitutions successfully reduced polyreactivity. Importantly, the model allowed us to diminish the polyreactivity of an angiotensin II Type I receptor antagonist nanobody, without compromising its functional properties.

With over 100 approved antibody-based therapeutics, the format is a well-established starting point for drug discovery. Despite this success, lead antibodies may suffer from undesired molecular properties. Thus, we present the Therapeutic Antibody Developability Analysis (TA-DA). A tool built by testing hundreds of sequence- and structure-based descriptors at differentiating clinical antibodies from non-natively paired human repertoire antibodies.