B and T cells perform functions critical to human health and they develop, differentiate, and expand in spatially distinct sites across the body. Both B and T cells express clonal heritable antigen receptors that confer exquisite molecular (i.e., antigen) specificity. Antigen receptors can be defined by sequencing, but these methods require tissue dissociation, which loses the anatomical location, and the surrounding functionally relevant environmental cues. Linking specific clonal sequences to their molecular and cellular surroundings, i.e., ‘clonal niche’, could help us understand and harness B and T cell activity. A technological bottleneck has been to capture the location of antigen receptor sequences, and by extension B and T cell clonal responses, directly within tissues. To address this, we recently developed a spatial transcriptomics-based approach (Spatial VDJ) and associated computational pipelines to reconstruct B and T cell clonality in human tissues.Using this technology, we spatially resolve B and T cell receptors within immune and tumor tissues, as well as B cell clonal evolution within germinal centers. Combined, Spatial VDJ links B and T cell clonal responses to their microenvironment with applications to various immune-related pathologies, including infections, cancer and autoimmune diseases.

Early in the Covid-19 pandemic, neutralising antibodies were being developed as therapeutics, highlighted by their rapid approval by the FDA. Today, most of these antibody drugs have since lost their efficacy due to viral escape. The pandemic showcased and continues to showcase a coevolutionary race between the human immune system and SARS-CoV-2, during which the immune system generates neutralising antibodies targeting the SARS-CoV-2 spike protein’s receptor binding domain (RBD), crucial for host cell invasion, while the virus evolves to evade antibody recognition. In the presented manuscript, we establish a synthetic coevolution system combining high-throughput screening of antibody and RBD variant libraries with protein mutagenesis, surface display, and deep sequencing. Additionally, to significantly extend our interrogation of sequence space, we train a protein language machine learning model that predicts antibody escape to RBD variants and demonstrate its capability of generalising to a larger mutational space and mutations at positions unseen during training. Through explainable AI techniques, we probe the model and identify biologically meaningful coevolution trends.

Analysis of the cell populations and antigen receptor repertoires of human B cells in responses to vaccines or infections, or in pathological immunological contexts such as food allergy, has been limited by the difficulty and expense of gathering antigen-specific data. We have been developing and applying highly multiplexed panels of DNA-tagged antigens from SARS-CoV-2 variants, influenza viruses, and food allergen proteins, among other antigens, to sort specific B cells for single cell transcriptome, BCR heavy and light chain and antigen tag sequencing experiments. These data enable determination of the specificity and antigen variant binding breadth of thousands of B cells in a single experiment, and correlation with cellular phenotypes and receptor sequences, opening a new scale of investigation into the B cell mechanisms affecting serum antibody responses and B cell memory.

Most human T cell analyses are performed from peripheral blood, although 98% of T cells reside in other tissues. In this study, we single-cell sequenced 5.7 million αβ T cells from autologous blood and tonsils of 10 donors. Using this unparalleled multimodal dataset, we sought to answer critical questions in T cell receptor biology previously unanswerable by smaller-scale experiments. We identified distinct clonal expansions and distributions, with TCM subsets being more clonal in tonsils, while TEM and MAIT cells being predominantly clonal in blood. Clonal overlap analysis revealed surprisingly low sharing across tissues (1-7%), which increased with age. Additionally, expanded tonsillar T cells were often not detected in blood, highlighting the distinct and compartmentalized nature of T cell distributions. T cells with identical TCR sequences demonstrated phenotypic overlap at broad annotation levels (e.g., CD4/CD8, naive/memory), though this concordance decreased with more granular phenotypes. Critically, T cell profiles in blood did not reliably recapitulate the differentiation state and functional role of identical clones in tissue, indicating substantial differentiation plasticity and tissue-specific adaptations. Analysis of antigen specific CD8 T cells revealed location as a primary determinant of their frequency, phenotype, and immunodominance. Finally, the tissue repertoire recalibrates our current TCR diversity estimates, and we provide a refined estimate of whole-body repertoire. Given the tissue-restricted nature of T cell phenotypes, functions, differentiation potential, and clonality across this unprecedented dataset, we conclude that tissue T cell analysis is crucial for accurate TCR repertoire analysis and effectively monitoring changes after perturbing therapies.

The T cell receptor (TCR) repertoire is an extraordinarily diverse collection of TCRs essential for maintaining the body’s homeostasis and response to threats. In this study, we compiled an extensive dataset of more than 4200 bulk TCR repertoire samples, encompassing 221,176,713 sequences, alongside 6,159,652 single-cell TCR sequences from over 400 samples. From this dataset, we then selected a representative subset of 5 million bulk sequences and 4.2 million single-cell sequences to train two specialized Transformer-based language models for bulk (CVC) and single-cell (scCVC) TCR repertoires, respectively. We show that these models successfully capture TCR core qualities, such as sharing, gene composition, and single-cell properties. These qualities are emergent in the encoded TCR latent space and enable classification into TCR-based qualities such as public sequences. These models demonstrate the potential of Transformer-based language models in TCR downstream applications.

The adaptive immune system relies on many types of B and T cells, whose functions are reflected in the distinct molecular features of their receptor sequences. Here, we introduce an inference framework, soNNia, which integrates interpretable knowledge-based models of immune receptor generation with flexible and powerful deep-learning approaches to characterize sequence determinants of receptor function. Using soNNia, we characterize sequence-specific selection associated with receptors harvested from different cell types and tissues. On a dataset of human TCR repertoires from thymocyte subsets, we characterize the variability between individuals generated during the TCR V(D)J recombination process and how it is shaped through all stages of T-cell maturation and differentiation.

LIBRA-seq (LInking B-cell Receptor to Antigen specificity through sequencing) enables high-throughput antibody discovery through simultaneous screening of B cells against a theoretically unlimited number of antigens at a time. Using LIBRA-seq, we have identified antibodies with unique phenotypic properties against HIV-1, influenza, coronavirus, and other pathogens of biomedical significance. Large-scale LIBRA-seq analysis of human antibody repertoires is revealing novel insights into the fundamental rules of antibody-antigen interactions. Overall, the LIBRA-seq technology offers unmatched capabilities for high- throughput discovery of novel antibody therapeutics and for assessment of vaccine efficacy.

To become specialized binders, antibodies undergo a process called affinity maturation to maximize their binding affinity. Despite this process, some antibodies retain low-affinity binding to diverse epitopes in a phenomenon called polyreactivity. Here we seek to understand the molecular basis of this polyreactivity in antibodies. Our results highlight that polyreactive antigen-binding fragments (Fabs) bind their targets with low affinities, comparable to T cell receptor recognition of autologous classical major histocompatibility complex. Extensive mutagenic studies find no singular amino acid residue or biochemical property responsible for polyreactive interaction, suggesting that polyreactive antibodies use multiple strategies for engagement. Finally, our crystal structures and all-atom molecular dynamics simulations of polyreactive Fabs show increased rigidity compared to their monoreactive relatives, forming a neutral and accessible platform for diverse antigens to bind. Together, these data support a cooperative strategy of rigid neutrality in establishing the polyreactive status of an antibody molecule.

Current methods in Adaptive Immune Receptor Repertoire sequencing (AIRR-seq) resolve variable region sequences of antibody transcripts with minimal detailed resolution of the constant region (IGHC) therefore hindering characterization of the extent of IGHC diversity, which is known to impact downstream effector functions. Dr. Smith will present a recently developed immunogenomics tool that resolves near complete antibody heavy chain (IGH) genotypes and repertoires, near Full-Length Antibody Heavy Chain Repertoires (FLAIRR-seq). We further will present the application and potential of FLAIRR-seq, alongside it’s partner genotyping tool, IGenotyper, to identify the impact of immunogenomic variation in vaccine response, autoimmunity, and immunotherapy responses.

Unraveling factors that shape the antibody (Ab) repertoire is crucial for understanding adaptive immunity and disease risk. Our study, utilizing long-read genomic and adaptive immune receptor repertoire (AIRR) sequencing, highlighted the significant influence of genetic variation within the immunoglobulin heavy chain locus (IGH) on IgM and IgG Ab repertoires in human peripheral blood mononuclear cells (PBMCs). Extending these findings, we conducted deep AIRR sequencing in six individuals to explore the Ab repertoire across B cell development stages—pro-B, pre-B, immature, and naive. We also sequenced the IGH, IG kappa, and IG lambda loci to create personalized germline allele databases and identify SNPs, indels, and structural variants affecting the Ab repertoire. We found high gene usage correlations across B cell stages and replicated genetic variants in pro-B cells linked to inter-individual variation in the peripheral Ab repertoire, highlighting strong genetic effects on V(D)J recombination. Our analysis revealed associations between IGH variants and specific heavy and light chain gene usage profiles, suggesting trans-effects of IGH locus polymorphisms on light chain repertoire likely due to constraints on chain pairing. Variants correlated with both heavy and light chain usage also significantly influence the CDR3 region variation, including rs74091765, which affects gene usage for IGHV3-64, 13 IGL, and 6 IGK genes, as well as the CDR3 amino acid composition across IGH, IGK, and IGL. Our study demonstrates that genetic variation in the IG loci significantly contributes to Ab repertoire diversity, emphasizing the relevance of IG loci genomics in understanding Ab responses and human disease.

Evolution is the powerful force driving both the real-time emergence of pathogen resistance to drugs and immunity, as well as the diversity of natural forms and functions that have emerged over longer timescales. Modern evolutionary models, especially those that leverage advances in machine learning, can improve our ability to design new proteins in the laboratory. This talk will cover how models of protein sequences and structures can learn evolutionary rules that help guide the artificial evolution of human antibodies. First, we will cover how algorithms known as protein language models can guide the affinity maturation of antibodies against diverse viral antigens using sequence information alone and without requiring any task-specific data. Next, we will cover how multimodal language models, which also take into account information about protein structure, can further improve the ability for unsupervised models to guide antibody evolution, which we use to improve the neutralization potency of clinical antibodies against viral escape variants.

High throughput sequencing of B cell receptors (BCRs) is increasingly applied to study the immense diversity of antibodies. Learning biologically meaningful embeddings of BCR sequences is beneficial for predictive modeling. Several embedding methods have been developed for BCRs, but no direct performance benchmarking exists. Moreover, the impact of the input sequence length and paired-chain information on the prediction remains to be explored. We evaluated the performance of multiple embedding models to predict BCR sequence properties and receptor specificity. Despite the differences in model architectures, most embeddings effectively capture BCR sequence properties and specificity. BCR-specific embeddings slightly outperform general protein language models in predicting specificity. In addition, incorporating full-length heavy chains and paired light chain sequences improve the prediction performance of all embeddings. This study provides insights into the properties of BCR embeddings to improve downstream prediction applications for antibody analysis and discovery.

T cells targeting self-proteins are important mediators in autoimmune diseases such as type 1 diabetes (T1D). T cells express unique cell-surface receptors (TCRs) which recognize peptides presented by major histocompatibility molecules. Here, we deep-sequenced the TCR beta chain (TCRβ) repertoires from longitudinal peripheral blood DNA samples at four time points beginning early in life from children that progressed to clinical T1D (n=29) and age/sex/HLA-matched pancreatic islet autoantibody negative controls (n=25). From 53 million TCRβ sequences, we show that the TCRβ repertoire is extraordinarily diverse early in life and narrows with age independent of disease. We demonstrate the ability to identify and track shared antigen-specific TCRβ sequences, those responding to viral peptides and separately islet proteins. Public islet antigen-specific TCRβ sequences had different patterns of accumulation based upon self-antigen specificity in the preclinical T1D cases (e.g., those responding to insulin, glutamic acid decarboxylase, or zinc transporter 8). As an independent validation, we sequenced and analyzed TCRβ repertoires from a cohort of newly diagnosed T1D patients (n=143), identifying the same islet-antigen reactive TCRs. Furthermore, 73 public islet-antigen TCRβ sequences were present in higher frequencies and numbers in T1D samples compared to controls. The total number of these disease-relevant TCRβ sequences inversely correlated with age at clinical diabetes diagnosis, highlighting the importance of using disease-relevant TCR sequences as powerful biomarkers in autoimmune disorders.