Dynamics of Immune Repertoires: Exploration and Translation

B cells are an evolutionary system, undergoing rapid somatic hypermutation and antigen-driven selection as part of the adaptive immune response. B cell lineage trees inferred from B cell receptor sequencing data represent the history of mutations in a lineage, and can also link multiple forms of B cell diversity. For example, B cell lineage trees sampled from multiple tissues from the same subject could represent B cell migration between tissues. Trees sampled at different timepoints in the same subject represent clonal persistence over time. Here, we introduce new phylogenetic methods that use lineage trees to understand patterns of B cell migration, differentiation, and evolution over time. We demonstrate how this framework has been used to understand the role of B cell migration and differentiation in multiple immune conditions. Further, we show how longitudinally sampled data can be used to test for ongoing evolution in B cell lineages. Using large, publicly available AIRR-seq datasets, we demonstrate how different infections, vaccinations, and other conditions differ significantly in their ability to stimulate evolution over time in B cells. Some conditions, such as HIV infection, stimulate high strong signatures of B cell evolution, while others such as influenza vaccination produce a more compartmentalized response detectable only with direct germinal center sequencing. These methods are widely applicable and implemented in the R package dowser, available at https://bitbucket.org/kleinstein/dowser.

Immune repertoires provide a unique fingerprint reflecting the immune history of individuals, with potential applications in precision medicine. Can this information be used to identify a person uniquely? If it really is a personalised medical record, can it inform us about the outcomes of a COVID-19 infection? I will explore how learning about randomness can help us read this information we all carry within us.

Clonal expansion is a core aspect of T cell immunity. However, little is known with respect to the relationship between replicative history and the formation of distinct CD8+ memory T cell subgroups. To address this issue, we developed a genetic-tracing approach, termed the DivisionRecorder, that reports the extent of past proliferation of cell pools in vivo. Using this system to genetically ‘record’ the replicative history of different CD8+ T cell populations throughout a pathogen-specific immune response, we demonstrate that the central memory T (TCM) cell pool is marked by a higher number of prior divisions than the effector memory T cell pool, owing to the combination of strong proliferative activity during the acute immune response and selective proliferative activity after pathogen clearance. These data provide the first evidence that a stem-cell-like memory T cell pool that reconstitutes the CD8+ T cell effector pool upon reinfection is marked by prior quiescence.

Adaptive immunity is driven by specific binding of hyper-variable receptors to diverse molecular targets. The many-to-many mapping between receptors and their epitope targets determines how the immune system senses threats, but an understanding of its global structure has remained elusive. In my talk I will introduce an analytical framework to quantify structure in recent high-throughput measurements of such maps. I will discuss how a combination of ideas from population genetics, molecular biophysics, and machine learning is enabling us decipher the probabilistic rules of the degenerate molecular binding code of the immune system.

T cells play an important role in adaptive immunity. An enormous clonal diversity of T-cells with a different specificity, encoded by the T-cell receptor (TCR), protect the body against infection. Most TCR$\beta$ chains are generated from a V-, D-, and J-segment during recombination in the thymus. Although complete absence of the D-segment is not easily detectable from sequencing data, we find convincing evidence for a substantial proportion of TCR$\beta$ rearrangements lacking a D-segment. Additionally, sequences without a D-segment are more likely to be abundant within individuals and/or shared between individuals. Our analysis indicates that such sequences are preferentially generated during fetal development and persist within the elderly. Summarizing, TCR$\beta$ rearrangements without a D-segment are not uncommon, and tend to allow for TCR$\beta$ chains with a high abundance in the naive repertoire.

The T cell receptor (TCR) diversity necessary for the recognition of a broad antigen spectrum is determined by the interaction between T cell receptors (TCRs) and cognate ligands presented by major histocompatibility complex (MHC) molecules during T cell selection in the thymus. Evolution might have favored an optimal diversity in the copy number-variable MHC, defined by a trade-off between the benefits of presenting a larger number of foreign antigens and the disadvantage of a limited mature T cell repertoire following negative thymic selection, due to presentation of a larger number of self-antigens. However, our understanding of how the initial T cell repertoire is shaped is still very limited. Three-spined sticklebacks have a completely functional adaptive immune system and exhibit a natural level of diversity at the MHC. The small size of this wild fish allows an easier estimate of the systemic TCR diversity for each individual, ideal in eco-immunological studies. We have developed a cDNA-based 5’RACE protocol with unique molecular identifiers and a stickleback TCRß gene reference library. By characterizing the systemic TCRß repertoire diversity among male and female naive lab bred individuals belonging to different families and harboring copy number-variable MHC genotypes ranging from low to high diversity, we have directly tested the association between gender, MHC, and naive TCR diversity. Our results contribute to the understanding of the T cell selection process and the balance between the recognition of pathogenic vs self antigens during the evolution of the vertebrate adaptive immune system.

The DNA sequences of antibody and T cell receptor gene rearrangements can be used to trace clonally-related lymphocytes through different tissues, in different individuals and over time. In this presentation, I will describe methods for inferring clonal relatedness from adaptive immune repertoire profiling data. I will describe networks of clones in different tissues of human organ donors and show how features of gene rearrangements yield insights into tissue compartmentalization of immune responses. I will also discuss some of our recent data on antibody responses to SARS-CoV-2 vaccines, highlighting how clonal lineage analysis can be used to understand how antibody responses evolve over time. Finally, I will describe some of our recent efforts to study shared repertoire features between individuals.

In the coming years, most translation of immunological basic science is likely to involve an artificial intelligence or machine learning (AI/ML) component. Based upon considerable recent experience in translating AI-driven medical projects from scientific insight all the way to real-world clinical products, I will talk about a translation framework which I have developed. The framework is used in at least one of the top-3 pharma companies in the world. And I have used it extensively in supporting medical AI translation projects. Examples, where appropriate, will be taken from a recent project on donor matching.

The adaptive immune receptor repertoire (AIRR) is a snapshot of past and ongoing immune events, such as infections, vaccinations, allergies and (autoimmune) diseases. This information is remembered as a mixture of unknown event-specific immune signals that are distributed across immune receptors. It is assumed that each immune signal is a set of immune-event specific motifs and each immune event may be associated with multiple motifs. Linking these immune signals to immune status and classification of immune repertoires based on immune signals is one of the main goals of immunodiagnostics. Machine learning (ML) provides various approaches to detect complex signals in high-dimensional data. Earlier, ML and statistical approaches have been successfully applied to classify CMV, cancer and multiple sclerosis immune repertoires. However, there is still no systematic machine learning approach for the specific problem of immune repertoire classification. Here we present our plans for comprehensive benchmarking of machine learning methods applied to AIRR data. We plan to investigate a range of machine learning and deep learning algorithms to learn, classify and generate immune-event specific motifs on simulated AIRR datasets. We will focus on the understanding of how machine learning approaches learn and reconstruct an immune signal from AIRR-seq data. This research will make the first step toward establishing clear guidelines on how to apply machine learning to AIRR-seq data.

B cells are the central actors in the adaptive system and encode highly diverse and mutable pathogen-engaging receptors. They can counter a multitude of pathogens by directly neutralizing invaders or by storing memory to respond to reinfections efficiently. Upon infection, activated B cells seed germinal centers (GCs) where they hypermutate and are selected for enhanced affinity to pathogens. On longer timescales, memory B cells from previous GC reactions can seed new GCs during reinfection and mutate further; however, the extent of their role in response to reinfections is unclear. Because B cells evolve only in GCs, standard dynamical models with continuous accumulation of mutations fail to describe this interrupted evolution. We introduce a stochastic telegraph process to model the B-cell lifecycle by capturing the entry and exit of B cells from GCs and constraining the accumulation of mutations to the GC residents only. We use this model to reconstruct time-resolved evolutionary histories of B cells from a longitudinal dataset of immune repertoires from individuals with HIV and obtain posteriors of the rates across repertoires and patients. Our results elucidate the lifecycle of B cells and clarify the role of memory B cells in secondary responses on a repertoire-wide scale.

We consider the maintenance of a population of ``product'' cells from progenitor cells via one or more intermediate compartments. If there is only one intermediate compartment, a large ratio of product cells to progenitors can only be achieved at the cost of the product cell population being dominated by large families of cells descended from individual progenitors, and large average number of divisions separating product cells from progenitors. These undesirable features can be avoided if there are multiple intermediate compartments. A sequence of compartments is, in fact, an efficient way to maintain a product cell population from a progenitor population, avoiding excessive clonality and minimising the number of rounds of division ``en route''.

The adaptive immune system in vertebrates has evolved to recognize non-self antigens, such as proteins expressed by infectious agents and mutated cancer cells. T cells play an important role in antigen recognition by expressing a diverse repertoire of antigen-specific receptors, which bind epitopes to mount targeted immune responses. Recent advances in high-throughput sequencing have enabled the routine generation of T-cell receptor (TCR) repertoire data. Identifying the specific epitopes targeted by different TCRs in these data would be valuable. To accomplish that, we took advantage of the ever-increasing number of TCRs with known epitope specificity curated in the Immune Epitope Database (IEDB) since 2004. We compared seven metrics of sequence similarity to determine their power to predict if two TCRs have the same epitope specificity. We found that a comprehensive k-mer matching approach produced the best results, which we have implemented into TCRMatch, an openly accessible tool (http://tools.iedb.org/tcrmatch/) that takes TCR β-chain CDR3 sequences as an input, identifies TCRs with a match in the IEDB, and reports the specificity of each match. We anticipate that this tool will provide new insights into T cell responses captured in receptor repertoire and single-cell sequencing experiments and will facilitate the development of new strategies for monitoring and treatment of infectious, allergic, and autoimmune diseases, as well as cancer.

Advances in our understanding of the germline immunoglobulin (IG) gene repertoires of humans and mice will be reviewed. Hypotheses arising from repertoire studies over 35 years will also be described. For both humans and mice, advances in our knowledge of germline genes have recently delivered surprises. Important advances in our knowledge of human IG genes suggest that there may be less diversity than has been assumed, for many common haplotype blocks have now been identified. Advances in our understanding of inbred mouse genes show a diversity in the IGH locus that is not explained by the reported breeding histories of the strains. An hypothesis will be outlined suggesting that the chaotic breeding histories of some inbred mice may explain their susceptibility to autoimmune diseases. It will be suggested that this susceptibility is normally controlled by the co-evolution of heavy and light chain genes. A related hypothesis will be outlined suggesting that the co-evolution of heavy and light chain genes may make the offspring of individuals from divergent populations susceptible to autoreactivity. Via this and related mechanisms, if populations diverge sufficiently, antibody genes may act to promote reproductive isolation. IG genes may therefore be critical to the process of speciation. Repertoire studies can and must also contribute to the elucidation of the coordinated functioning of antibodies, which differ significantly between humans and mice. It remains unclear how useful effector function develops despite the frequent production by an individual of antibodies of diverse isotypes with often opposing activities.

The SARS-COV-2 pandemic has provided a rare opportunity to study the earliest adaptive immune responses to infection in humans. We performed a longitudinal analysis of the peripheral blood T cell receptor repertoire in a cohort of 39 healthcare workers who became infected by SARS-COV-2 during the first wave of the pandemic in London. We observed a set of TCRs which expanded and then contracted in the first four weeks of the study. The wave of expansion was very fast, peaking around the time of the first PCR+ test. This set of expanding TCRs was highly enriched for TCRs which had been independently annotated with SARS-COV-2 specificity in functional assays. The epitopes recognised by these annotated early expanding T cells showed little sequence homology to circulating seasonal human coronaviruses. The epitopes were, however, highly conserved between all the major strains of SARS-Cov-2. The expanding TCRs were enriched for public (shared) high-abundance TCRs present in healthy pre-pandemic repertoires, and formed part of a cross-individual network of clusters of TCRs with shared sequence motifs. We hypothesise that a set of high abundance TCRs in the repertoire may allow the immune system to respond rapidly and contribute to containing the magnitude of the infection, before the appearance of neutralising antibody.

Most human genes code for more than one transcript. Different ratio of transcripts of the same gene can be found in different cell types or states, indicating differential use of transcription start site or differential splicing. Such differential transcript usage events (DTUs) enable an additional layer of regulation and protein diversity. Except for PTPRC and CIITA, there are very few reported cases of DTU in the immune system. In order to rigorously map DTUs between human immune cell types, we studied two publicly available RNA sequencing datasets that profiled multiple human immune cell types. We identified 129 DTUs between five human naïve immune cell types that appeared in both datasets. The patterns of the DTUs are mostly cell type or lineage specific, and correlate with the expression of splicing factors and transcription factors, suggesting a model for regulation of DTUs in the naïve human immune system. Of the several stimulations studied, only sepsis affected the splicing of more than a few genes, and only in innate immune cells. As alternative first exons compared larger change in use between cell types compared to alternative splicing events, and transcription factors expression patterns are richer compared to splicing factor expression patterns, splicing regulation seems to play a minor role in immune differentiation and an even smaller role in immune activation.

The T cell antigen receptor (TCR) sequence determines the specific peptide(s) a given T cell recognizes and the interaction between the TCR and peptide­-MHC plays a key role in determining T cell fate, from the thymus during development to the differentiation following activation during an immune response. Yet, whether there are TCR sequence features which predispose a T cell towards certain functions or fates remains poorly understood. Here, we used an unbiased machine learning (ML) approach to characterize murine CDR3β sequence characteristics that impact how strongly naive CD4+ T cells recognize self peptides, as measured by their expression levels of CD5. While most CDR3β sequences can lead to both strong and weak self-­reactivity, our ML model identified subsets of CDR3β sequences that more consistently predisposed a T cell towards a particular fate. Our analysis determined that non­templated nucleotide additions in the TCRβ chain tend to weaken self­reactivity, reducing the likelihood of positive selection in the thymus. Further, self­ reactivity is impacted by V and J chain usage and biases in amino acids found in specific positions. We confirmed these findings in a separate validation data set, tested specific self­-reactivity predictions using TCR retrogenic experiments of TCRs not in our dataset, and showed that our ML analysis tool generates new insights when applied to previously published datasets. Taken together, our results highlight how the CDR3β sequence gives rise to a trade­off between repertoire diversity and recognition strength.

Adaptive immune receptor repertoire (AIRR) data are complex and carry disease and infection relevant information in the form of sequence-based immune signals. For an AIRR containing an immune signal, the fraction of AIR sequences bearing this signal may be very low. One of the major unresolved challenges in AIR diagnostics and therapeutics discovery, is to predictively isolate the immune signal from an AIRR and then to use this signal to engineer novel sequences. This is a machine learning (ML) problem that bridges repertoire and sequence level classification. Existing ML methods rely on engineered features or on a predefined knowledge about repertoire generation. To address this problem, We developed AIRRTM, an end-to-end generative model based on an encoder-decoder architecture and topic modeling (TM). AIRRTM reaches stable performance in identification and generation of disease specific sequences using repertoire labels but not sequence labels.

Adoptive cell therapy (ACT) using tumor-infiltrating lymphocytes (TILs) has demonstrated efficacy in certain cancer patients yet there is a need to discern the multifactorial mechanisms driving clinical response to develop next-generation TIL products and to identify predictive biomarkers to better select patients for ACT therapy. Of interest, the dynamics and profiles of TIL clonotypes that populate infusion products and then persist in blood or infiltrate tumors after ACT remain poorly understood. Here, we analyzed 13 metastatic melanoma patients enrolled in a TIL-ACT trial (NCT03475134) comprising 6, 3 and 4 patients with partial responses (PR), stable disease (SD) or progressive disease (PD), respectively (best overall response rate at 3 months by RECIST v1.1). Using bulk and single-cell (sc) T-cell receptor (TCR)-sequencing (seq), we tracked TIL clonotypes in tumors, in ACT products (ACTP) and in post-ACT blood and tumor samples. By mapping these TCRs to scRNA-Seq data, we identified transcriptomic signatures of TILs a) expanding in vitro, 2) persisting in blood post-ACT and/or 3) infiltrating tumors post-ACT. Moreover, patient-derived tumor cell lines allowed the interrogation of tumor-specificity. A library of 141 tumor-reactive clones and 187 non-tumor-reactive ones allowed the investigation of their dynamics and profiles. TCR repertoires overlaps between baseline tumors and ACTP were variable, yet significantly higher in responders. Conversely, progressors ACTP minimally overlapped with tumors but significantly overlapped with blood clonotypes (i.e. blood obtained at the time of the surgery). Also, in responders, higher fractions of ACTP TILs persisted in blood and infiltrated tumors post-ACT. Furthermore, tumor-specific clonotypes were more abundant in responders tumors and preferentially expanded in vitro and persisted in vivo. Associating TILs dynamics to molecular profiles of cells, we found similarity in genes and transcription factors associated to tumor-specificity with in vitro expansion and in vivo persistence and infiltration. Finally, a gene signature of tumor-reactivity massively overlapped with the signatures of TILs dynamics as well as the signature of clinical responses. These data indicate that the gene signature predictive of clinical responses is associated to the global TILs dynamic (in vitro and in vivo) and TILs tumor-specificity.

The immunoglobulin gene loci (IG), which encode all antibody genes, are amongst the most polymorphic in the human genome. Genetic diversity in these regions include high levels of allelic variation of multiple genes, as well as whole gene deletion and duplications, resulting in copy number variation (CNV), which impacts antibody responses to pathogens. Using germline antibody sequencing and inference analysis of repertoire sequences we have uncovered a high level of genetic variability in IGHV, IGKV, IGLV and IGHC genes in South African CAPRISA donors. Genetic variation within the IGHV genes includes single nucleotide polymorphisms (SNV) in the V-region, leader and intronic regions. We also observed mismatched leader regions, for example within the 19 individuals that had IGHV1-18*01, all had a L-PART1 that matched IGHV1-18*04 rather than IGHV1-18*01. In addition to allelic diversity, we have observed changes in CNV with >5% of our participants having duplications of IGHV1-3, IGHV4-4, and IGHV2-26. Genetic diversity within the light chains remains largely unexplored compared to heavy chains. In a small study using expressed kappa and lambda antibody repertoires of 13 individuals we have identified over 20 novel light chain alleles, indicating that the vast genetic diversity we observed in the IGH locus extends to the other loci as well. Lastly, we have observed >10 novel alleles within IGHG3, IGHG1, IGHA1 and IGHA2 constant region genes. Some of the novel IGHC alleles involve amino acid changes associated with increased Fc effector function and/or HIV-1 neutralization. Having a greater understanding of germline antibody genetic diversity and its impact on immune responses to HIV-1 allows us to engineer better antibodies for prevention and treatment as well as improve on germline targeting HIV-1 vaccination strategies.

Cell-to-cell communication networks have critical roles in coordinating diverse organismal processes, such as tissue development or immune cell response. In particular, cytokine-driven differentiation of Th cells into the well-known Th1, Th2, Tfh subtypes, contain multi-layered regulatory circuits in terms of interaction by diffusible ligands. Compared with intracellular signal transduction networks, the function and engineering principles of cell-to-cell communication networks are far less understood. Major complications include: cells are themselves regulated by complex intracellular signaling networks; individual cells are heterogeneous; and importantly, cells are subject to proliferation and cell death, and often show exponential growth at the onset of an immune response. In this talk, I will outline the key challenges on the way to data-driven models of immune-cell interactions, and describe our strategy to tackle those challenges with an integrated framework of experimental, statistical and analytical tools. Using this framework, we developed a fully annotated model of Th cell dynamics in LCMV infection in mice. Overall, we found that data-driven modeling of immune cell interaction dynamics can open new perspectives on key decision-making processes at the onset of an immune response.

A diverse B cell receptor/antibody repertoire is required for effective and safe immune responses. Positive and negative selection forces act on this repertoire during an immune response and therefore effects of infection, vaccination and dysregulation due to disease can be seen on the repertoire. It is also important to distinguish between different types of B cells in these analyses since there are many different functions of B cells and their repertoire may be affected differently depending on the stimulus. I will present some of our data showing the effects of vaccination, age, and infection on B cell populations.

We developed a computational tool, IgDiscover, which allows rapid and precise individualized genotyping of V, D, and J alleles from expressed, rearranged B or T cell receptor sequences. Application of IgDiscover to a study of immunoglobulin heavy chain (IGH) germline gene diversity in 80 individuals from 4 different human sub-populations: African, East Asian, South Asian, and European, revealed frequent copy number variation and numerous novel alleles. Many of these were present at widely differing frequencies in the different sub-populations, providing a baseline for studies of antibody responses in humans. We also applied personalized IGH genotyping and haplotype analysis to a cohort of 14 SARS-CoV-2 convalescent health care workers and demonstrated an exceptional diversity in the immunoglobulin heavy chain V (IGHV) gene and allele content between the individuals. We selected one study participant with an interesting IGHV genotype for the isolation of neutralizing monoclonal antibodies against the SARS-CoV-2 spike, which allowed us to address the functional impact of allelic variation. Application of IgDiscover and inferred haplotype analysis to define IGHV gene and allele content in a family group provided insight into how the high variation in the IGH locus may arise. Overall, our results underscore the importance of personalized genotyping of adaptive immune receptor genes in studies involving AIRR data.

The B cell can serve as a controllable model system to illuminate how cellular programming and complex signal integration operates to alter multiple fates simultaneously. When experimentally derived principles and rules are captured into suitable probabilistic models, population dynamics can be recreated that includes the automatic generation of enormous heterogeneity in cell phenotype. Such models may be useful for simulations of naive B cell activation, for germinal center dynamics and for insights into how a complex system can evolve from responses 're-purposed' from organismal development programs.