Adaptive Immune Recognition in Infectious Disease and Cancer

The Immune Epitope Database (IEDB) and Cancer Epitope Database and Analysis Resource (CEDAR) are twin databases that as part of their activities in cataloguing B and T cell epitopes, have compiled T- and B cell receptor (TCR and BCR) sequence data from over 3,800 papers, spanning more than 9,000 epitopes collectively. In this talk I will outline the IEDB’s efforts to accurately annotate receptor sequences, as well as the demonstrated utility of this data - from use as training data for machine learning algorithms, to providing valuable insights in repertoire sequencing data.

I'll describe our work aimed at developing predictive models of T cell receptor specificity, including our predictions in the IMMREP2025 TCR specificity challenge and our attempts to design TCRs and antibodies that are specific for peptide-MHC complexes.

Immune repertoires need to stay diverse in order to prepare our immune system against the breadth of pathogens and their escape mutants. I will describe a model based on biological knowledge in which responding effector T cells suppress in each other based on their affinity for the antigen. I will show how this results in equalizing the frequencies of different T cell clones regardless of their affinity. I will describe experiments in mice that confirm those predictions. This equalization allows for maintaining diversity in the response pool, which we show experimentally results in a higher coverage of possible escape mutants by the immune system.

The adaptive immune system consists of highly diverse B- and T-cell receptors, which can recognize a multitude of diverse pathogens. Immune recognition relies on molecular interactions between immune receptors and pathogens, which in turn is determined by the complementarity of their 3D structures and amino acid compositions, i.e., their shapes. Immune shape space has been previously introduced as an abstraction of molecular recognition in the immune system. However, the relationships between immune receptor sequence, protein structure, and specificity are very difficult to quantify in practice. I will introduce physically motivated machine learning approaches to learn representations of protein micro-environments in general, and of immune receptors, in particular. Despite no direct training on immune-antigen data, the implicit physical reasoning in our model enables us to make accurate predictions for antigen-induced T-cell activities across diverse viral epitopes and cancer neo-antigens. Leveraging our T-cell receptor recognition model, we develop a computational protocol for de-novo design of immunogenic peptides. We use our generative framework to quantify the diversity of the peptide recognition landscape for various TCR-MHC complexes, offering key insights into T-cell specificity in both humans and mice. Our approach provides a platform for immunogenic antigen design, opening new computational paths for T-cell vaccine development against viruses and cancer.

The $\alpha$ $\beta$ T cell receptor TCR is encoded by two diverse protein chains which together determine the receptor antigen specificity. A key unsolved challenge is to determine an ambiguous map from TCR sequence to specificity, and from peptide/MHC (pMHC) back to TCR sequence. In this presentation we will review progress in using SCEPTR, a large language based transformer model to map TCRs of the same or similar specificity into nearby regions of high-dimensional transformed space. We will apply this framework to define metaclones, sets of TCRs from different individuals with shared pMHC specificity, exploring applications in both infectious diseases and cancer. Finally, we will present TcRictionary a novel graph-based database which captures the many-to-many nature of TCR ontology and allows complex querying from multiple perspectives.

The diversity of antigen receptors created by somatic recombination is a hallmark of adaptive immunity. Yet, it is becoming increasingly clear that this diversity is substantially reshaped by convergent selection pressures. In this talk, I will present some of our recent progress on two fundamental questions in the field. First, what do convergence patterns reveal about the molecular determinants of T cell specificity? Second, how can we best exploit convergence signals for the discovery of disease-associated T cell receptor metaclones?

Advances in next-generation sequencing (NGS), combined with machine learning approaches, have positioned T-cell receptor (TCR) repertoire analysis as a promising strategy for the discovery of novel biomarkers and therapeutic targets. This approach is particularly relevant, and challenging, in the context of autoimmune diseases (ADs), where immune heterogeneity and disease complexity remain major obstacles. To address this critical gap, we performed high-throughput TCR repertoire sequencing across multiple autoimmune conditions. We sorted CD4⁺ T effector cells (Teff), CD4⁺ regulatory T cells (Treg), and CD8⁺ T cells from peripheral blood samples of 300 individuals diagnosed with type 1 diabetes (T1D), rheumatoid arthritis (RA), osteoarthritis (OA), spondyloarthritis (SpA), and systemic lupus erythematosus (SLE), as well as from healthy volunteers (HV). TCR repertoires were analyzed using an innovative machine learning framework based on sparse least squares discriminant analysis (sPLS-DA). In addition, we applied this analytical strategy to longitudinal samples from SLE patients treated with low-dose IL-2, enabling us to explore dynamic immune changes in response to therapy. In this presentation, I will discuss our findings and highlight both the opportunities and the persistent challenges in leveraging TCR repertoire profiling for biomarker discovery in autoimmune diseases.

Viral evolution is often framed through the lens of immune escape, where mutations reduce recognition by antibodies or T cells. Yet mutations can also generate novel HLA binders, leading to immune exposure. To systematically assess the extent of HLA escape and exposure events across viral genomes, we developed a novel computational framework and used it to analyze the mutational landscapes of influenza A/H1N1, SARS-CoV-2, and HIV-1. We found that escape and exposure frequently co-occur, with major antigenic shifts marked by early increases in escape, followed by compensatory rises in exposure. Sites of antibody escape were disproportionately enriched for HLA exposure mutations, suggesting that humoral and cellular pressures are coupled. Experimental testing of 80 peptide-mutant pairs confirmed 19 escape and 14 exposure events, with several of them altering functional T-cell recognition. Importantly, allele-specific escape/exposure ratios correlated with severe COVID-19 outcomes and tracked global SARS-CoV-2 mortality. These findings challenge the escape-centric paradigm of viral evolution and suggest that exposure mutations may play a critical role in shaping immune responses across a range of viruses. Understanding this dynamic interplay between escape and exposure has broad implications for vaccine design, antiviral strategies, and predicting viral adaptation.

Unlike jawless fish, the adaptive immune system of jawed vertebrates originated in a singular series of events involving insertions of a transposon into a gene encoding immunoglobulin-like domains. Since no descendants of the earliest jawed vertebrates survive to the present day, the key events and processes must be inferred from living animals. Available evidence from comparative genomics and immunology suggest that the antigen-specific receptors diversified by RAG genes were part of the primordial MHC, and that gene co-evolution dependent on genetic recombination (as well as genome-wide duplication events) were fundamental processes that co-opted disparate genes to shape the adaptive immune system thereafter. These processes have continued with accident being as important as selection, including different lineages of jawed vertebrates losing key genes and then recovering, non-immune genes being co-opted for novel functions, and genomic organisation leading to different strategies of immune function. For example, MHC is organised differently in typical mammals like humans and mice, galliform birds like chickens, and passerine birds like sparrows and zebra finch, with significant functional consequences. Among those consequences is the relative importance of a spectrum of MHC alleles from promiscuous generalists to fastidious specialists, which illustrates the strategic tradeoffs in the immune response.

If MHC class I (MHC-I) peptide presentation obeyed simple mass-action rules, the immunopeptidome would mirror the translatome, since all (non-exported) proteins are eventually degraded in the cell that synthesize them. Yet CD8⁺ T cells routinely detect peptides derived from extremely low-abundance tumor or viral gene products—very often early after viral infection. This mismatch implies that the MHC I antigen processing machinery is physically organized to prioritize certain antigens. In this talk I will describe evidence supporting the idea that antigen processing, like protein synthesis itself, is “channeled” to increase the efficiency of recognizing viral and aberrant cell peptides.

Specificity is widely considered the defining hallmark of adaptive immunity. For decades, immunologists have focused on the most specific responses: the immunodominant clones that expand massively, the TCR with the highest avidity for a given peptide, the T cells that bind most strongly to experimental tools such as tetramers. These highly specific responses unquestionably exist - they are experimentally convenient and biologically spectacular. But do they truly represent how the immune system operates in real life? Several observations challenge this view. Individuals with extremely restricted TCR repertoires—sometimes comprising only a few thousand TCRs—can nevertheless mount effective immune responses against diverse pathogens. Likewise, the well-documented phenomenon of heterologous immunity reveals that immune responses to one pathogen can profoundly influence responses to structurally unrelated pathogens. Yet such observations have often been treated as curiosities rather than fundamental properties of T-cell immunity. Recent advances in high-resolution TCR repertoire analysis now allow us to revisit the assumption that extreme specificity is the central feature of T-cell biology. Emerging data instead suggest that a substantial fraction of T cells carry polyspecific TCRs capable of recognizing multiple unrelated viral peptides. These findings suggest that the immune system may not operate primarily/exclusively through rare, highly specific clones. Rather, it may rely heavily on populations of broadly cross-reactive/polyspecific-T cells, capable of responding to diverse antigens with varying avidities. In this view, the fundamental property of the T-cell response may not be exquisite specificity but rather functional degeneracy and fuzziness.

Combinatorial pairing of independently recombined TCRα and TCRβ chains is central to diversifying the T cell receptor (TCR) repertoire. TCRs recognizing the same epitopes frequently feature similar sequences with dominant motifs in one or both chains. However, the extent to which a single chain determines epitope recognition remains unclear. Here, we systematically test the constraints of TCR chain coupling by enforced pairing of a specific chain with hundreds of thousands of diverse partner chains. We show that while nearly all chains can stably pair, only a small – and highly variable – fraction of these combinations retain the desired epitope binding. The most promiscuous chains preserved specificity with approximately 10% of the partner repertoire, whereas the most selective maintained binding with fewer than 1 in 1000 partners. This approach expanded the known repertoire space for 10 epitopes, generating >70,000 novel epitope-specific αβTCR sequences from just a few unexposed individuals. Using the resulting greatly expanded datasets, we train classifiers distinguishing binders from highly similar non-binder TCRs with high accuracy and develop structural models of recognition requirements. Our approach highlights the inherent limitations of single-chain sequencing data and provides a new experimental and analytical framework for refining inference of TCR-pMHC specificity.

Estimating production and loss rates of cell populations is essential but difficult. The current state-of-the-art method to estimate these rates involves mathematical modelling of deuterium labelling experiments. Current models typically assume that the labelling in the precursors of the population of interest (POI) is proportional to the deuterium enrichment in body water/glucose. This assumption is not always true and it is known that this can have a significant effect on the rates estimated from labelling experiments. Here we quantify the effect that different turnover (replacement) rates of the precursors can have on the estimated proliferation and loss rates of a POI by explicitly modelling the dynamics of the precursors. We first confirm earlier results that the labelling curve of the POI only reflects its own turnover rate if either the turnover rate of the precursors is sufficiently fast, or the contribution from the precursors is sufficiently small. Next, we describe three realistic scenarios with a slowly turning over precursor population, and show how this changes the interpretation of the different parameter estimates. Our analyses underpin that uniquely identifying the turnover rate of a POI requires measurements (or knowledge) on the turnover of its immediate precursors. DOI: 10.1371/journal.pcbi.1013052

Mathematical models of germinal centers have had a great impact on understanding how high affinity and diverse antibodies evolve in response to challenging pathogens. For example, it was predicted that B cells undergo iterative steps of somatic hypermutation and selection with respect to the antigen in question. It was also predicted that T follicular helper cells provide the limiting step in B cell selection. Further, that the product of germinal centers, the antibodies, impact on the germinal center, thus suggesting a negative feedback loop. All these prediction were meanwhile confirmed. According to textbook knowledge, class switch recombination happens in the course of germinal center reactions. In a combined experimental and simulation approach, it will be shown that this belief is wrong and that class switch mostly happens before the reaction. Further, we will provide theoretical evidence for the idea that the frequency of somatic hypermutation is regulated in an affinity-dependent way rather than being the same for every B cell division. Indeed, experiments now confirmed that this is really the case and we provided a mechanism how this was realized by nature. The results provide a success story of the iterative cycle between theoretical prediction and experimental validation that can serve as a motivation for the development of digital twins of tissues and organs.

Adaptive immune responses to chronic viral infections have long been studied. Inflationary CD8 T cell responses to human cytomegalovirus (CMV) are a paradigm of such a chronic response. In my talk, I will address the dynamics of the recently discovered chronic NK cell response to CMV. Analysing neutral somatic variants that accumulate steadily, we infer the clonal dynamics of CMV-responsive NK cell clones. We find large expanded NK cell clones that, remarkably, show strong signatures of positive selection. I will discuss mechanistic implications for the origin and maintenance of such clones. This research is in collaboration with Chiara Romagnani and Timo Rückert at Charité Berlin.

Mechanistic models are widely used to describe and explain biological processes over time. However, they typically rely on a limited number of observable compartments and sparse longitudinal data. As a result, these models are often either too simple to capture complex biological phenomena or they face identifiability issues, particularly when considering interindividual variability in the form of nonlinear mixed-effects models based on systems of differential equations. In parallel, with ongoing technological advances, longitudinal high-throughput data (e.g., -omics, including transcriptomics and proteomics data) are increasingly available in various contexts and could bring valuable information into mechanistic models to better capture underlying biological processes. However, integrating such high-dimensional data to inform the dynamics remains a major challenge, both mathematically and for broader interpretation in public health applications. In this talk, I will present two complementary approaches for integrating large- to high-dimensional biomarkers into mechanistic models. The first approach, called lasso-SAMBA, addresses robust covariate selection in ODE-based non-linear mixed-effect models. It extends the original SAMBA algorithm (which is an iterative model-building algorithm that fastly and sequentially identifies relevant covariates on parameters while estimating model using the SAEM algorithm) by replacing stepwise inclusion with Lasso regression combined with stability selection, ensuring a more reliable identification of relevant covariates while preserving the monotonic decrease of the information criterion. The second approach uses observed -omics data to infer the dynamics of unobserved immune compartments. It relies on an iterative algorithm that alternates between a regularization step, which identifies the most informative biomarkers through penalized likelihood derivatives, and a mechanistic inference step, where population parameters are estimated using the SAEM algorithm in Monolix. This framework enables the selection of biomarkers whose temporal patterns best reflect the latent compartments in our model. Together, these methods provide powerful tools for integrating and selecting high-dimensional biological data in mechanistic modeling. They will be illustrated on exemples of immune dynamics after vaccination for Varicella-Zoster virus and immune-viral dynamics after SARS-CoV-2 infection.

We investigate the development of immunological memory, starting out from single antigen-specific lymphocytes in vivo and in vitro. Thereby, we break down adaptive immune responses into their most fundamental building blocks. In the past, this experimental approach of "single-cell fate-mapping" has shown us that even physiological immune responses to infection do not harness the full protective potential available in every antigen-specific lymphocyte. We want to make this full potential accessible through tailored vaccination strategies and optimised immunotherapeutic approaches. Central to all these endeavours is our goal to identify 'stem-like' T and NK cells that stand at the origin of all differentiation processes underlying lasting immunological memory in the context of acute and chronic infection, malignancy and autoimmune disease. By identifying key factors that regulate these stem-like lymphocytes, we hope to open up new avenues for effective vaccination and immunotherapy.

CD8+ T cells can recognize with their T cell receptor (TCR) epitope targets presented by MHC class I molecules with high specificity and sensitivity. Adoptive T cell therapies with primary T cells, like tumor-infiltrating lymphocytes (TILs) or donor-lymphocyte infusion (DLI) or virus-specific T cells (VSTs) have demonstrated in individual cases outstanding clinical efficacy, including curative responses where other currently available therapies fail (including solid cancer). Gaining a better understanding on how CD8+ T cells can exert highly effective protective immunity and to engineer this protective immunity for adoptive T cell therapy of patients that cannot endogenously generate or maintain such T cell responses, is a major challenge in the field. In patients and pre-clinical mouse models, we have demonstrated over the past years that effective antigen-specific T cell immune responses are usually characterized by the initial recruitment of highly polyclonal populations from the naïve compartment, consisting of a repertoire of different T cell receptors (TCRs) with varying affinities/avidities for cognate pMHC ligands. By implementing and developing advanced technologies (like single cell TCR sequencing/identification, traceable TCR retrogenic T cells, Crispr-Cas9-mediated orthotopic TCR replacement, structural and functional TCR avidity measurement, high-throughput TCR reporter cell screening), we could define correlates for quality of protection and evolutionary TCR repertoire changes during chronic antigen-exposure. Thereby, variable cross-reactivity patterns within polyclonal TCR repertoires can provide protection towards epitope escape. Cross-reactivity maps for all possible single amino-acid exchange variants can be determined cost- and time-efficiently by machine-learning guided support (P-TEAM) and can provide important information for selection of TCRs for clinical applications. Since mouse and human TCR germline components (V-D-J) are basically identical in their amino acid sequence, we decided to generate saturating ultra-deep TCR libraries (> 1000 individual pairs) for 4 different nominating epitopes presented by two different murine MHC class-I alleles from more that 650 donors. The depth of this - to our knowledge currently unique - dataset allows to estimate the total space of TCR solutions for defined epitope specificities within physiological peripheral TCR repertoires, as well as to determine fundamental features of functional pMHC recognition that apply for murine as well as human TCRs.

Immunotherapeutic approaches that use engineering principles to generate tailored immune responses, as e.g. equipping immune cells with a sentinel receptor against cancer-specific targets (e.g. CAR-T cells), have increased our ability to treat various diseases. However, current approaches still tend to fail in numerous patients and several hard-to-treat diseases, mainly due to the inability of understanding immune processes within tissues. With tumor microenvironments, as well as tissue-specificities influencing the efficacy of immune responses, revealing the interplay between immune cell motility, proliferation, differentiation and function is essential to improve current therapeutic strategies and treatment approaches. Here, we show how the combination of multi-modal experimental and clinical data with methods from mathematical modelling, image analysis and machine learning can be used to determine the spatio-temporal dynamics of immune responses within and across tissues. By combining individual cell-based mathematical models and cell population approaches with experimental and clinical data our analyses indicate the importance of tissue structure and immune cell subset dynamics for the control of viral infections and the effect of CAR T cell responses against malignant diseases. These analyses could help to provide insights on key regulators of host-pathogen interactions and immune cell dynamics in order to support the design of novel therapeutic strategies that rely on the modulation of adaptive immune responses.

In the secondary lymphoid tissues, human immunodeficiency virus (HIV) can replicate both in the follicular and the extrafollicular compartments. Yet, virus is concentrated in the follicular compartment in the absence of antiretroviral therapy, in part due to the lack of cytotoxic T lymphocyte (CTL)-mediated activity there. CTL home to the extrafollicular compartment, where they can suppress virus load to relatively low levels. We use mathematical models to show that this compartmentalization can explain seemingly counterintuitive observations. First, it can explain the observed constancy of the viral decline slope during antiviral therapy irrespective of the presence of CTL in SIV-infected macaques, under the assumption that CTL-mediated lysis significantly contributes to virus suppression. Second, it can account for the relatively long times it takes for CTL escape mutants to emerge during chronic infection even if CTL-mediated lysis is responsible for virus suppression. Third, the compartmental structure has important implications for the evolution of viral mutants in general, influencing mutant fixation probabilities and fixation times. The talk will discuss these results.

Tumor evolution reflects a dynamic interplay between somatic mutation and immune surveillance. The ratio of nonsynonymous to synonymous mutations within the immunopeptidome—termed immune dN/dS—provides a quantitative measure of immune selection acting on neoantigenic mutations. By adapting evolutionary models originally developed for molecular evolution, immune dN/dS distinguishes between tumors that are immune edited, where antigenic mutations are depleted by negative selection, and immune escaped, where antigenicity persists through immune evasion mechanisms. Analysis of large-scale cancer cohorts demonstrates that immune-edited tumors exhibit reduced antigenicity and high CD8⁺ T cell infiltration, whereas immune-escaped tumors retain antigenic diversity and are more responsive to immune checkpoint blockade. As a dynamic biomarker, immune dN/dS integrates tumor genomics and immune ecology, offering a mechanistic framework to stratify patients by evolutionary immune pressure and to refine predictions of immunotherapy response.

The network of interactions comprising CD28, CTLA4 and PD-1 and their ligands CD80, CD86, and PD-L1 have profound implications for T cell activation. Despite their critical role, factors that determine their receptor occupancy when all three ligands are present remain incompletely understood. Using a supported lipid bilayer to replicate the antigen-presenting cell membrane facilitated a quantitative analysis of CD80, CD86, and PD-L1 interactions. Our observations reveal that PD-1 blockade operates through a previously unrecognized competitive mechanism beyond conventional checkpoint inhibition: liberated PD-L1 molecules redistribute to form CD80-PD-L1 heterodimers that strengthen CD28-CD80 interactions and outcompete CD86 for CD28 binding. This preferential binding enhances T cell activation through superior CD80 co-stimulation. Furthermore, the redistribution by PD-1 blockade reduces CD80 homodimer availability, limiting CTLA4 binding to CD80 and enhancing CD28 signaling. These findings provide vital insights into the competitive principles that allow B7-CD28 family receptors to regulate T cell activation through network-level effects.

T cells can detect as few as a single antigenic peptide-MHC (pMHC) complex among thousands of endogenous pMHCs displayed on the surface of antigen-presenting or target cells. This remarkable sensitivity is striking given the relatively low affinity and in principle degenerate nature of TCR-pMHC interactions, and despite more than three decades of intense research, its mechanistic basis remains incompletely understood. We attribute this gap stems largely from the inherent complexities of membrane biophysics, the transient nature of the biochemical and structural changes that accompany T-cell antigen recognition, and the limitations of conventional biochemical approaches, which fail to capture the unique context of the immunological synapse. To meet these challenges, we have developed non-invasive live-cell imaging modalities that yield quantitative, molecular-level insights with millisecond temporal resolution below the diffraction limit of visible light. I will present our experimental strategies and discuss some of our key findings on how T-cell antigen receptors engage their antigens within the immunological synapse. I will further go over more recent results in the contexts of tumor recognition and autoimmunity, reflecting our broader goal of translating mechanistic insights into tangible clinical benefits. I will finally present work in progress and planned activities focused on antigen discovery in the context of osteosarcoma and also a more quantitative understanding of T-cell multi-recognition arguably a cornerstone of adaptive immunity.

CD4+ Foxp3+ regulatory T cells (Treg) play critical functions in maintaining tolerance to self antigens, regulating immune responses to pathogens and in the pathophysiology of anti-tumoural immunity. Treg ontogeny is complex since they are generated during T cell thymic development (thymic Treg), but are also induced from mature conventional T cells (inducible Treg) when activated by foreign antigen with appropriate additional cues. How these distinct ontogenic pathways contribute to the maintenance and function of the circulating mature Treg compartments throughout the life course, and during development of malignant disease remains unclear. Here, we address this using two distinct, but complementary in vivo fate mapping models in mice; i) temporal fate mapping in busulfan chimeric mice to track de novo generation of Treg subsets throughout life and ii) genetic fate mapping using an inducible Foxp3-CreERT fate reporter mice. Data were subsequently analysed using computational models to define precise lineage topologies. We find that naive and effector/memory (EM) Treg pools exhibit distinct dynamics but are both continuously replenished by de novo generation throughout life. We show that naive Treg and not conventional T cells, are the predominant precursors of circulating EM Treg in adults. EM Treg numbers are normal in germ free mice and are not influenced by host commensal load, revealing a central role for self recognition for their development. To investigate the ontogeny of Treg development in malignant disease, we used the same fate reporter systems to characterise the Treg infiltrate of three different model tumours. In all three cases, we found that Treg derived from pre-existing, EM Treg. Together, these results reveal a predominantly linear pathway of Treg development from thymic origin to EM Treg, driven by self antigen recognition, and that are specifically activated and recruited to tumours in malignant disease.

When cells exchange information via receptor-ligand interactions localized in specific membrane regions of the interacting cells, the local cell morphology may impose constraints on the early signaling events. An important example is T cell activation, where T cells probe the surface of adjacent antigen presenting cells with dendrite-like protrusions. The small number of proteins involved in the initial stage of T cell receptor activation raises additional questions about the potential impact of stochastic fluctuations on the ensuing signaling response. In my talk, I will describe a simulation method that avoids frequently encountered approximations related to the structure of the membrane surfaces between the interacting cells and the stochastic motion of T cell receptor and peptide-MHC ligand complexes. It simulates grid-free motion of single particles on adjacent curved surfaces and allows one to perform highly resolving analyses of how the interplay between kinetic segregation and proofreading determines the kinetics of TCR signal initiation. This work was supported by the Division of Intramural Research of NIAID, NIH.

There are approximately 400000000000 naive CD4 T cells in your body, about the same as the number of stars in our galaxy. On the other hand, the number of cells of one TCR clonotype is a small integer that increases or decreases by one cell at a time, when cells divide or die. New clonotypes are released from the thymus and compete with other clonotypes in the periphery for specific and non-specific resources. Mean clonal sizes can therefore be calculated from mean clonal lifetimes. For example, if the ratio of thymic production to peripheral division is four percent, then the number of distinct T-cell clonotypes in the human body is about nine percent of the total number of (naive CD4) T cells. In mice, most TCR clonotypes may consist of just one or two T cells. TCRs recognize peptides (or epitopes), typically 8­14 amino acids long, bound to MHC molecules on antigen-presenting cells. There cannot only exist a single TCR which recognizes a given peptide because the possible number of peptides is far greater than the number of T cells in a mouse or in one person. Therefore, individual TCRs must recognize multiple peptides if a mammal's T cell repertoire is to be capable of providing coverage against the majority of new pathogens a host might encounter in its lifetime. Patterns of recognition of epitopes by T cell clonotypes (a set of cells sharing the same T cell receptor) are represented as edges on a bipartite network. We introduce a circular space of epitopes, so that T cell cross-reactivity is a quantitative measure of the overlap between clonotypes that recognize similar (that is, close in epitope space) epitopes.