Recent progress in the analysis of αβ T cell and B cell receptor repertoires

Abstract

T cell receptors (TCRs) and B cell receptors (BCRs) are vertebrate evolution’s best answer to the threat of microbial pathogens that can evolve much faster than ourselves. These antigen receptors are generated during T cell or B cell development by combinatorial rearrangement of germline genome V, D and J gene segments, and with junctional residues capable of enormous diversity. For decades the complexity of these receptor repertoires has limited their analysis, but advances in DNA sequencing technology and an array of complementary tools have now made their study much more tractable, filling a major gap in our ability to understand immunology as a system. Here, we summarize the recent approaches and discoveries that are enabling these advances, with some suggestions as to what may lie ahead.

The αβ TCR repertoire

Ever since T cells were recognized as having diverse specificities in the 1970s, there have been efforts to understand their response repertoire, since that has much to tell us about their capabilities. But due to the limits of technology, these efforts could only estimate the overall diversity based on simple assumptions about the recombining units [1] or calculating the naïve frequency of a particular antigen specificity [2].

The modern era of repertoire studies began with the use of next generation DNA sequencing technologies by Robins and others to sequence hundreds of thousands of TCRs in a single experiment from populations of T cells in the blood [3,4]. These data confirmed the extreme complexity of TCRβ sequences, estimating approximately 5 × 10¹¹ amino acid sequences but that only 3 × 10⁶ of these possibilities were present in the average individual [3]. Subsequent work has delineated pathogen-specific repertoires [5]. A number of interesting results have come from these early TCR sequencing studies, such as the common origins of memory resident versus peripheral CD8⁺ T cells [6], the reduced TCR repertoire in older individuals [7] and the discovery that T cell clonal expansion and functional attributes are very broad at first in the context of a pathogen response, but then narrow to predominantly one cell type as the disease progresses [8].

But taking full advantage of the power of T cell repertoire analysis requires obtaining both chains of the TCR heterodimer from single cells, and Han et al. [9] developed the first relatively high throughput method, where hundreds of TCR pairs could be obtained in one set of experiments and this could be quickly built up to thousands. An alternative was also developed by Howie et al. [10] using a dilution strategy. These pairs of TCRs could then be transfected into reporter T cells and challenged with candidate antigens [11^••] or expressed in a soluble form and used to screen yeast libraries that Garcia et al. have used to display up to one billion variant peptides bound to a given MHC molecule [12,13^•]. This later approach is more arduous but is necessary when it is not clear what antigens to test. Other general screening methods have also been developed [14^•], for more focused searches, such as surface molecules [14^•] or cDNAs from particular cell types [15^•].

It is important to recognize that there are two distinct types of T cell repertoire to consider. One is the TCR sequence repertoire, which is vast and consists of all possible sequences of TCR α and β, and their potential pairings. But the other, often more biologically relevant, repertoire is the ligand repertoire, which is ~1000× smaller. For the former repertoire, if every TCR α could pair with every TCR β, which is unlikely, one could imagine an upper limit for αβ TCR diversity of 10¹⁵, far more than the number of T cells in a human being (2 × 10¹¹) or a mouse (3 × 10⁸). This explains why even monozygotic twins have very few identical TCR sequences, even though they have the same HLA alleles and are often recognizing the same antigens.

But what about the ligand repertoire? Here the best numbers come from studies using peptide-MHC tetramers [16] or higher order multimers. Using the advanced enrichment techniques used by Jenkins [17], where frequencies of approximately one in one hundred thousand and one in one million are seen in humans [18,19], and are only modestly less rare in mice [17], indicating an intrinsic limitation, possibly positive selection in the thymus. Multiple TCR motifs can be seen binding to the same peptide-MHC tetramer (an average of five noted by Glanville et al. [11^••]). It seems that the ligand repertoire for a given MHC allele might easily be in the low millions. This is a much more manageable figure for most purposes, although given the thousands of MHC alleles in human beings, it still covers a lot of ground! But how to convert the relatively easy to obtain sequence data on TCRs-thousands to millions-into the much more biologically relevant ligand information typically desired? Here two bioinformatic approaches have been developed that can accomplish this task. One is the program GLIPH (Grouping of Lymphocyte Interactions by Paratope Hotspots, [11^••]) which used tetramer gathered T cell sequence data to identify short 3–4 aa motifs and other parameters as key determinants of specificity, presaged in part by the kmer work of Chain et al. [20] and the extensive Tdist analysis of Dash et al. [21^••]. These programs take large raw TCR sequence datasets, alpha or beta, or both, and use them to produce clusters of TCRs which share the same peptide-MHC specificity, and also to predict the relevant MHC allele, where the input population has sufficient diversity.

The validation of these results indicates that one can use these programs to interrogate any collection of TCR sequences and reduce at least a fraction of the sequences into meaningful specificity groups. These groups in turn can be queried for whether they are important in any given biological or clinical situation. For example, one can ask whether a cohort of individuals who respond successfully to a particular vaccine differs significantly in their T cell repertoire response from those who do not. Given the same distribution of MHC alleles in the cohort, one might find that the diversity of the responders TCR repertoire is significantly greater, or that specific TCRs are expressed in one group or the other. In either case, one does not have to know what the antigen or MHC is to see what a critical trend might be. This is critical, since as the data sets get larger and larger, you can be dealing with thousands of specificity groups, with no hope of finding ligands for more than a fraction.

Progress should also be noted in the use of peptide MHC multimers to analyze T cell responses to many different ligands at once. Here the development of combinatorial labeling strategies, first using fluorescent probes [22,23] and more recently with metal labels and mass cytometry [24,25] or with oligonucleotide barcodes [26] enables hundreds of different T cell specificities to be analyzed simultaneously, together with detailed phenotyping.

In TCR analyses, there has been frequent speculation about ‘public’ versus ‘private’ TCRs, where public refers to TCR identities between individuals, either at the DNA or protein sequence levels. A recent paper by DeWitt et al. [27], analyzes 80 million TCR sequences from over 660 people, and finds that ~14% are ‘public’. While some of these might represent TCRs that have been repurposed for important stereotypical interactions, as the γδ TCRs in the dendritic epithelial cells in the skin of mice [28], most are probably just the result of chance and less complex rearrangements, as shown by Ethanati et al. [29^••], although they do contain many pathogen related sequences [21^••].

For a more extensive review of TCR repertoire work than space allows here, please see Bradley and Thomas [30^•].

BCR and antibody analysis

High-throughput DNA sequencing (HTS) methods now enable the analysis of millions of Ig gene rearrangements in a single experiment. Early efforts were applied to the analysis of malignant B cell populations, zebrafish antibodies, healthy donor Ig repertoires and phage display libraries [31–34] and then a wide range of autoimmune, infectious disease, immunodeficiency, and vaccination conditions. Scaled-up sequencing of physically joined Ig heavy and light chain cDNA from single cells followed, enabling synthesis and characterization of native monoclonal antibodies [35,36]. More recently, BCRs have been analyzed from single cell RNASeq data [37], or enriched Ig-containing cDNA libraries, analogous to published TCR approaches [38]. A new addition to this toolbox is DNA barcoding of reagent monoclonal antibodies for protein phenotyping on single cells via DNA sequencing of the antibody tags [39,40]. Similar methods could label antigens and enable highly multiplexed analysis of B cell antigen specificity and BCR sequences. New analysis algorithms have been developed to address the wealth of new data, with tradeoffs between their ability to handle very large datasets, their sophistication of modeling the structure of Ig germline genes and rearrangements, and their applicability to single-cell data [37,41–45]. Equally important are new initiatives formulating data standards, data structures and mechanisms to improve quality assurance and the ability to share and reanalyze Ig sequence data [46–48]. In parallel with the genetic tools, improved protein fragment analysis using mass spectrometry has enabled comparison of serum antibody proteins and BCR gene sequences in an individual, to characterize dominant clones in influenza vaccine responses [49,50].

The published literature using these new methods includes many proof-of-concept papers featuring relatively few subjects. Looking ahead, studies of larger cohorts will be needed to characterize the variation associated with population groups, age, sex, and health status. Analysis of tissues other than blood will also expand understanding of human immune responses, following the lead of rhesus macaque studies using serial fine-needle aspiration of lymph nodes to study germinal center B cell reactions after vaccination [51].

BCR repertoire formation

Human Ig germline loci (heavy, kappa, and lambda on chromosomes 14, 2, and 22, respectively) are poorly characterized in most populations. These repetitive loci defy conventional mapping or assembly, and conceal large-scale insertions, deletions, and complex rearrangements. Recent sequencing of functionally haploid hydatidiform mole cell lines has begun to address these deficiencies [52,53]. Unsurprisingly, few genome-wide association studies (GWAS) have examined linkage between Ig locus variation and immunological diseases or phenotypes. A recent study of rheumatic heart disease due to Streptococcus pyogenes infection in Oceanic populations showed the potential for such associations, finding disease-susceptibility in subjects with the *02 allele of heavy chain gene segment IGHV4–61 [54^•]. Prior work showed that IGHV1–69 alleles with phenylalanine at position 54 preceded by a hydrophobic residue at position 53 in CDR-H2, are preferentially used in broadly neutralizing antibodies that bind the influenza hemagglutinin stalk [53]. Thorough characterization of Ig loci in diverse populations will enable the search for other disease or therapeutic response associations.

It is currently unclear how diverse the Ig repertoire needs to be to provide competent humoral immunity over a human lifespan. An estimate of 10⁷–10⁸ IgH clonotypes and 10¹⁶–10¹⁸ potential heavy-light chain pairs in healthy donor blood B cells was recently published, based on HTS from multiple biological replicates per individual, as previously described for TCRβ measurements [55,56]. Non-circulating B cells in secondary lymphoid organs, the gastrointestinal tract and other sites could further increase these estimates. Key questions include whether the Ig repertoire diversity decreases in old age, as reported for the TCRβ repertoire, and whether decreased Ig diversity is associated with poorer vaccine responses [56].

Selection

Naïve B cell repertoires in human infants encounter an onslaught of antigens from microbiota, foods, vaccinations, and pathogens, triggering clonal expansion, isotype switching, somatic mutation, and differentiation to memory, plasma cell and other fates. We recently used IgH HTS to analyze B cell maturation in the Stanford STORK birth cohort of infants and young children with small-volume blood samples from years 1, 2, and 3 of life [57^•]. Correlations with environmental or pathogen exposures, and clinical symptoms showed that infants with disrupted skin barriers due to eczema had increased IgE somatic mutation (SHM), while upper respiratory infection was associated with increased IgM and IgD SHM. Antigen exposure contributes to the development, or prevention, of peanut allergy [58]; increased SHM in IgE-expressing B cells could reflect sensitization occurring via the disrupted skin barrier. Further analysis of antigen-specific B cells will be needed to clarify other childhood immunological events, such as the effects of the types of initial influenza virus infections on responses in adulthood [59].

Despite the diversity of the Ig repertoire, HTS has identified highly similar sequences in people exposed to the same antigens, such as Dengue virus [60], influenza [61,62], Haemophilus influenzae type b, tetanus [63], and HIV [64]. Strikingly, patients infected with Plasmodium falciparum exhibit a novel kind of convergent antibody, with portions of the LAIR1 gene (from chromosome 19) inserted between IGHV and IGHD-J gene segments, or into the switch region downstream from IGHJ gene segments [65^••]. LAIR1 insertions confer binding to the RIFIN surface antigens of the malaria parasite. Although an individual’s humoral responses consist mostly of private clones, convergent clonotypes could be diagnostically or prognostically useful by revealing prior antigen exposures. Ongoing work will add to reference databases of convergent Ig sequences, and test any clinical associations.

Conclusion

These new tools for characterizing TCR and BCR repertoires and functions are already populating the literature and databases such as VDJB at an exponential rate, increasing our systems-level understanding of immunological diseases, vaccinations and other interventions. These data will continue to expand our ability to discern broad trends and common patterns in these types of immune responses, as well as potential interactions, such as pathogen-specific TCR motifs in autoimmune repertoires, as already noted by Bradley et al. [26].