T-Cell Receptor Repertoire Sequencing in the Era of Cancer Immunotherapy

Our immune system is composed of the innate and adaptive immune systems; the innate immune system protects against general threats and conserved pathogenic sequences while the adaptive immune system targets and retains memory of specific pathogens (1). T cells are critical mediators of adaptive immunity, specifically driving cellular immunity as opposed to B cells that govern humoral immunity. Functionally, T cells kill infected cells or release cytokines to recruit other immune cells after recognizing foreign antigens presented on infected cells (2). Immature T cells migrate from hemopoietic tissues to the thymus where they mature into naïve T cells that express a unique and functional T-cell receptor (TCR) that grants them the ability to recognize specific antigens (3). The thymus is also where T cells differentiate from an initial CD4/CD8 double-positive state into either a CD4⁺ or CD8⁺ lineage. CD4⁺ T cells execute helper functions while CD8⁺ T cells are cytotoxic (4). Naïve T cells become antigen-experienced T cells after encountering their cognate antigen presented on major histocompatibility complex (MHC) molecules through their TCRs.

TCRs are highly diverse heterodimeric surface receptors that enable T cells to provide protection against a variety of different pathogens. A functional TCR is comprised of two paired protein chains, with either alpha (α) and beta (β) chains or gamma (γ) and delta (δ) chains (Fig. 1A). The majority of T cells express αβ TCRs that recognize antigens presented on MHC proteins (3). Only 1%–5% of T cells express γδ TCRs, which are not MHC restricted and participate in innate immune responses. It is unclear which exact ligands γδ T cells bind (5). For these reasons, αβ TCRs are typically the focus of TCR sequencing and subsequent clinical applications.

Regardless of type, the chains of a TCR contain an N-terminus variable region capable of antigen recognition and a C-terminus constant region. The variable region is assembled from variable (V), diversity (D), and joining (J) gene segments through an ordered process called V(D)J recombination, in which one allele of each gene segment is randomly recombined with other gene segment alleles to form a functional antigen recognition region (Fig. 1B; refs. 6, 7). TCRα and γ chains are comprised of V and J gene segments, while TCRβ and δ chains also include the D gene segment, offering more diversity in structure. The overall combinatorial diversity of the gene segments is accompanied by junctional diversity, which is driven by the random addition or deletion of nucleotides at junctions between gene segments. Both combinational and junctional diversity grant T cells a substantial range of antigen specificities, potentially accounting for 10¹⁵ to 10²⁰ possible TCR chains (8). Importantly, immature T cells commit to either an αβ or γδ lineage during V(D)J recombination, where the commitment to αβ is facilitated by the successful recombination of the β-chain and commitment to γδ is facilitated by successful recombination of both the γ and δ chain (9).

The variable domains for each of the TCR chains have three complementarity-determining regions (CDR): CDR1, CDR2, and CDR3. CDR1 and CDR2, which are encoded by the V gene segment, primarily facilitate the interaction between the TCR and MHC through contact with the conserved α-helices of MHC. CDR3 is encoded by the junction of V and J or D and J gene segments, resulting in high variability. This region is responsible for binding peptide antigens presented in the binding groove of MHC (8, 10). Because of its direct interaction with antigens and inherent hypervariability, CDR3 offers a wealth of knowledge about TCR specificity and is consequently a commonly used target region for TCR sequencing.

TCRs are of no consequence in the absence of their critical interaction partners, MHC molecules. MHC class I proteins are expressed by all nucleated cells and are recognized by cytotoxic CD8⁺ T cells while MHC class II proteins are expressed by professional antigen-presenting cells (APC) such as dendritic cells, macrophages, and B cells for recognition by CD4⁺ T cells. Antigen presentation by both classes of MHC proteins is an essential process in adaptive immunity (11, 12). These MHC molecules present processed peptides at the cell's surface for recognition by T cells through their specific TCRs (Fig. 1C). Antigen-processing pathways differ between both MHC classes; however, all pathways lead to the expression of a fully assembled peptide-MHC (pMHC) complex at the cell surface for TCR recognition (13). Structurally, the binding surface of both classes of MHC proteins is comprised of two domains; this surface originates from a single α-chain for MHC class I and from an α-chain and β-chain heterodimer for MHC class II (12). Accordingly, this diversity of MHC sequence and structure must be accounted for by any predictive software seeking to identify high-confidence MHC epitopes. Methods for predicting MHC epitopes are further discussed in the Computational Tools for Predicting Antigen Specificity section.

TCR specificity serves as the foundation for using TCR sequencing to understand T-cell biology and dynamics, as well as to study T cells in disease contexts and informing the development of therapeutics. Although identical TCRs may tend to recognize the same antigens, cross-reactivity has been observed, in which a single TCR has the ability to recognize multiple pMHC pairs (14–16). Because cross-reactivity of TCRs has the potential to complicate the characterization and surveillance of T-cell populations, it must not be overlooked in analysis of TCR sequencing and validation of antigen specificity.

The earliest approaches for characterizing the T-cell repertoire relied on Sanger sequencing. Because this approach is only able to capture small alterations in the genome, such as substitutions and short insertions/deletions, it is not robust enough to account for the immense diversity of TCRs (3). The advent of high-throughput sequencing (HTS) platforms has allowed for the rapid and comprehensive sequencing of genomic DNA (gDNA) or RNA, fueling major advances in the understanding of the T-cell repertoire (17). HTS methods, such as next-generation sequencing (NGS; ref. 12), are available for all TCR chains, but most kits target TCRαβ chains, as αβ T cells make up the majority of the T-cell population (8). While there are a multitude of different NGS platforms offered by companies like Illumina, Life Technologies, and Oxford Nanopore Technologies, the general workflow for using HTS in research remains consistent (18–20). The process begins with library preparation, followed by NGS, then computational analysis and deconvolution of the sequencing data, and finally, experimental validation (19). For TCR sequencing, each of these steps involves specialized techniques or programs designed to accommodate the complexity of TCRs.

An important first step of NGS library preparation is targeted enrichment of transcripts to be sequenced (21). Multiplex PCR and rapid amplification of cDNA ends (5′ RACE) are two commonly used methods for targeted amplification during T-cell repertoire library preparation. Both methods commonly aim to amplify the CDR3 region of the TCR, which offers the most information on antigen specificity. Multiplex PCR is widely used and it involves the amplification of either gDNA or RNA of the CDR3 region using primers for J gene segment alleles or the TCR constant region in conjunction with a mixture of primers for known V gene segment alleles. Inherently, multiplex PCR is limited by the set of primers available; because of this, novel V-alleles cannot be accurately characterized. Furthermore, primer bias within multiplex PCR risks uneven amplification of alleles, leading to inaccurate relative TCR sequence frequencies (8). Dual barcoding and unique molecular identifier (UMI) modifications to standard PCR protocols facilitate the removal of artifacts introduced during PCR amplification and support accurate downstream sequence analysis (21, 22).

Alternatively, 5′ RACE utilizes RNA and employs only one primer pair that targets the constant region of a TCR chain and the 5′ mRNA end (23). This eliminates the bias presented by multiplex PCR and allows for the capture of all TCRs present in a given sample. However, 5′ RACE may not be completely efficient in recapitulating entire recombination sequences, as it has been demonstrated to exclude V gene segments in some contexts; this necessitates the need to optimize such techniques further (8, 23). Optimizations to reduce the risk for primer bias during cDNA amplification include template switching modifications which add universal oligonucleotide sequences to 3′ ends of cDNA so that universal 3′ primers can be used during PCR amplification to amplify strands in a sequence-independent manner (24). The addition of unique barcodes to 5′ ends of cDNA further monitors for amplification errors and biases (25, 26).

With regards to sequencing methods, either bulk sequencing or single-cell sequencing can be used to analyze T-cell repertoires. Bulk sequencing, which sequences the aggregation of unpaired TCR chains within a sample, is commonly used to analyze large-scale TCR diversity and compare populations across patient cohorts (8). Historically, bulk analysis methods have focused on sequencing the TCRβ chain (or δ in γδ T cells) due to the presence of the additional D gene segment and greater combinatorial diversity compared with the α-chain. Moreover, T cells participate in “allelic exclusion,” in which only one functional β-chain is present in a T cell whereas multiple α-chains may be expressed (27). However, a major limitation of only sequencing the β-chain is that it does not provide information about αβ-chain pairings or in vivo biological function of the sequenced T cell (3, 8).

Conversely, single-cell sequencing focuses on individual immune cells. Advances in single-cell isolation techniques like fluorescence-activated cell sorting and microfluidic platforms have laid the groundwork for the success of single-cell TCR sequencing (28). While bulk sequencing offers valuable information regarding TCR diversity, it lacks the resolution of full TCRαβ chain pairing of single-cell analysis. Single-cell sequencing offers a more targeted and specific analysis of the T-cell repertoire, providing important information on paired α- and β-chain sequences; however, this method is limited by the number of cells that can be processed from a given sample and may not be capable of detecting rarer clonotypes. Overall, by analyzing gene expression in individual cells, single-cell analysis can provide a deeper understanding of intercellular genetic diversity and in vivo biological function (28).

Several companies offer TCR sequencing platforms, including Adaptive Biotechnologies (ImmunoSEQ), Beijing Genomics Institute (IR-SEQ), iRepertoire, Inc (RepSeq), Clonotech Takara Bio USA, Inc. (SMARTer Human TCR a/b Profiling Kit), CD Genomics (TCR-Seq), and Thermo Fisher Scientific (Oncomine TCR Beta-LR Assay), among others (8). Notably, 10X Genomics offers a single-cell profiling method equipped with feature barcoding technology to provide extensive information regarding gene and protein expression, paired TCR sequences, and antigen specificity (Table 1; ref. 29).

Moreover, several research groups have developed sequencing methods to address current experimental and technological limitations. For example, Howie and colleagues developed and validated pairSEQ, which assesses the diversity of TCR sequences in a sample and accurately pairs TCRα and TCRβ sequences without the need for single-cell methodologies. This platform is currently available through Adaptive Biotechnologies (30). Also of note, recently developed single-cell RNA sequencing technologies, including TCR functional landscape estimation supervised with scRNA-seq analysis (tessa) and clonotype neighbor graph analysis (CoNGA), can capture both TCR and gene expression data, mapping the functional relevance of specific TCRs (31, 32).

Computational analyses are performed on TCR sequencing data acquired from NGS to obtain useful information about the properties of the T-cell repertoire. Common metrics used in the analysis of the T-cell repertoire include T-cell density, diversity, and clonality (Fig. 2A). T-cell density refers to the total number of T cells that are found within an area. In the context of cancer, density serves as an estimate of T-cell infiltration into a tumor (33). T-cell diversity accounts for both TCR sequence “richness” and “evenness,” which reflect the number of unique TCR sequences and the distribution of TCR sequences, respectively (33, 34). Specific diversity measures include Hill numbers that refer to the number of effective species in a sample and Rényi entropy values which quantify the randomness of the given system. These measures serve as the basis for diversity-related indices that aim to assess clonal dominance, such as Shannon entropy, the Gini–Simpson index, and Pielou's coefficient (34). Clonality integrates density and diversity metrics and assesses the degree of clonal expansion of different T cells within a sample. Shannon clonality and Simpson clonality are clonality functions that are often included in TCR repertoire analysis, and they are derived from Shannon entropy and Gini–Simpson index diversity metrics, respectively. Recommended by Adaptive Biotechnologies, Simpson clonality is less sensitive to differences in sample size.

In addition to metrics that report on the properties of a single sample, there are indices that aid in the comparison of T-cell repertoires between different samples. Two similarity indices that are routinely utilized are the Jaccard index and the Morisita overlap index (MOI). The Jaccard index and related indices like the Sorensen index are solely based on the presence or absence of specific TCR sequences across multiple samples. These methods are useful in comparing heterogeneous repertoires but are unable to compare relative frequencies of particular TCR sequences between samples. In contrast, MOI accounts for relative frequencies as well as presence of specific TCR sequences, offering a more comprehensive comparison of T-cell repertoires from different samples (Fig. 2B; ref. 34). TCR metrics and overlap data can be presented in numerous ways for publication (Fig. 3).

Research groups like our own have developed more advanced approaches to characterize the T-cell repertoire. A limitation in the calculations of standard diversity or clonality metrics is that a group of highly similar TCR sequences within a given sample has the same weight as a group of dissimilar TCR sequences. Numeric embedding of TCR sequences allows for the fine-grained capture of TCR similarities, accounting for a continuous scale of TCR similarity when characterizing the T-cell repertoire (31, 35). In addition, computational tools like GLIPH, iSMART, and TCRdist that cluster TCRs based on sequence similarity can help researchers understand the biological relevance of TCR sequences rather than counting total TCR clonotypes (36–39). Additional analysis pipelines, such as Immunarch and Scirpy, facilitate large-scale analysis and visualization of TCR sequencing data (40, 41). Gene usage analysis reports allelic frequency of V, D, and J alleles and can be used to highlight intrinsic allelic bias within samples as well as extrinsic bias in response to disease (42, 43). Clone tracking and differential abundance analyses quickly identify expanded and contracted T-cell populations across longitudinal samples and enable researchers to both monitor and predict immunologic responses to disease and therapy (44–46).

Because of the expanded mapping of TCR repertoires through HTS, several groups have curated databases to provide a centralized location for TCR-related information (Table 2). ImMunoGeneTics (IMGT) is an information system that was created in 1989 to compile and standardize immunogenetic data. As the first and largest branch of IMGT, IMGT/LIGM-DB offers a collection of known antibody and TCR sequences from humans and other vertebrate species (47, 48). With regards to T-cell antigens, the Immune Epitope Database (IEDB) contains a collection of experimentally isolated antigens from a variety of contexts, including infectious agents, allergens, cancer, and self-antigens (49). The Adaptive Immune Receptor Repertoire (AIRR) Community established the AIRR Data Commons (ADC), an open-source repository for immune receptor sequencing data from individual external databases such as iRepertoire and VDJ Server (50, 51). With over 5.4 billion sequence annotations, the ADC represents the largest database of immune cell receptor sequences and an incredible resource for researchers (51). McPAS-TCR assembles TCR sequences associated with specific pathologies, like infections, cancer, and autoimmune diseases, as well as their corresponding antigens (52). VDJdb is a database that uniquely links TCR sequences with their pMHC ligands, contributing to a more comprehensive characterization of TCR interactions. In contrast to McPAS-TCR which solely focuses on pathologic conditions, VDJdb does not limit the biological contexts in which the collected TCR-pMHC pairs occur (53). In addition, 10X Genomics offers a dataset of over 15,000 distinct paired TCR sequences with specificity for at least one pMHC within the panel used in their single-cell immune profiling platform. Curated by the same group that created VDJdb, TCRdb leverages datasets from 10X Genomics single-cell immune profiling technology and contains more than 270 million TCR sequences from across clinical conditions, tissue, and cell types. The large-scale nature of TCRdb enabled by a new pipeline contributes to its novelty, as previous databases were manually curated and only contained around 100,000 TCR sequences (54). Such databases serve as the foundation for many computational tools that analyze important aspects of the TCR repertoire and immune response.

To predict TCR antigen specificity, one must first identify antigens which can be presented to T cells via MHC molecules. As such, numerous tools have been developed to address this critical challenge. Polymorphisms in MHC molecules affect characteristics of the peptide-binding groove and restricts the types of epitopes which can be presented (12). MHC haplotypes are commonly used as inputs into MHC prediction tools, such as NetMHC and the IEDB T-cell epitope prediction tool, to predict the likelihood of known proteins being presented by MHC molecules (55–57). Using the unique amino acid sequence of the peptide-binding groove in MHC molecules and inherent anchor positions important for peptide binding, these prediction algorithms predict the binding affinity for peptides and likelihood of presentation (58, 59). However, such algorithms only predict the presentation of a given peptide sequence and cannot predict what peptides are likely generated by cells through proteasomal degradation (60). Proteasomal processing prediction algorithms, such as Proteasome Cleavage Prediction Server, SpliceMet, and those offered by IEDB, can be used in conjunction with MHC prediction tools to more faithfully predict antigen presentation by MHC molecules and to identify potential sets of personalized tumor antigens using somatic mutations as inputs (61–63). Ultimately, although these tools are unable to predict T-cell reactivity, they do predict the stability of the formed peptide-bound MHC (pMHC), which contributes to its half-life and likelihood of being seen by any given TCR.

In addition to facilitating the characterization of T-cell repertoires by specific metrics, TCR sequencing provides information for the determination of T-cell antigen specificity. Identifying specific antigens targeted by different TCRs holds value in regards to quantifying antigen specificity groups within an individual, predicting disease outcomes, and isolating reactive T cells for therapeutic purposes. There is particular interest in the field of cancer immunology to identify and target neoantigens, peptides that arise from somatic mutations unique to cancer cells (64). With the recent improvements in HTS, many groups including our own have developed machine learning–based and algorithm-based tools aimed at elucidating antigen specificity for a variety of purposes (Table 3).

In 2017, Glanville and colleagues developed GLIPH (grouping of lymphocyte interactions by paratope hotspots) to automatically cluster TCR sequences into different groups of shared specificity. GLIPH integrates global and local TCR sequence similarity of CDR3 sequences to identify sets of TCRs that potentially recognize the same pMHC ligands. Clustering by global similarity refers to grouping α and β CDR3 sequences differing by up to one amino acid and clustering by local similarity refers to grouping sequences with shared enriched CDR3 motifs (36). More recently, the same group introduced GLIPH2, which efficiently processes and clusters millions of TCR sequences; this serves as an improvement to the original algorithm which lost accuracy when greater than 10,000 TCR sequences were inputted (37). Yet, both versions of GLIPH require TCR information from established databases and thus cannot discern new epitopes or TCR-pMHC pairs.

Zhang and colleagues also developed a method in 2020 to cluster TCRs based on antigen specificity, named iSMART (immuno-Similarity Measurement by Aligning Receptors of T cells). The researchers coupled iSMART with their TRUST algorithm (Tcr repertoire utilities for solid tissue) that detects CDR3 hypervariable regions from bulk RNA-sequencing data (65). Within the study, TRUST and iSMART were utilized to profile tumor-associated T cells, using multi-omic data from The Cancer Genome Atlas. Ultimately, iSMART aided in the identification of novel cancer neoantigens (38).

While GLIPH and iSMART indirectly facilitate antigen discovery through clustering TCR sequences based on shared specificity, other algorithms aim to directly predict TCR and antigen pairs. Springer and colleagues developed a deep learning–based method termed ERGO (pEptide tcR matchinG predictiOn) to predict TCR-peptide binding using existing large-scale TCR-peptide datasets as reference. ERGO is based on long short-term memory networks and solely considers the β-chain CDR3 sequence in determining whether a specific TCR will bind to different peptides (66). Because antigen specificity is not solely governed by the β-chain CDR3, ERGO-II was developed to include the α-chain CDR3 sequence, MHC typing, and V and J gene segments (67).

Similar to ERGO, TCRMatch aims to identify the specific antigens targeted by different TCRs. Using the β-chain CDR3 sequence as an input, TCRMatch employs a comprehensive k-mer matching algorithm to both identify TCRs with matches in IEDB and provide the specificity of the matches. The use of IEDB eliminates the need to perform preliminary experiments to train algorithms and streamlines the overall process; however, the range of antigen identification is limited to the dataset availabilities. To offset this limitation, TCRMatch can be used in conjunction with other tools like GLIPH2 (68).

NetTCR, designed by Jurtz and colleagues, predicted the interactions between TCRs and peptides specifically presented by HLA-A*02:01, a common human MHC class I allele. Like TCRMatch and the first version of ERGO, NetTCR used the β-chain CDR3 sequence; however, it was based on convolutional neural networks (CNN), which are uniquely suited to reconcile unaligned peptides and TCR sequences of different lengths (bioRxiv 433706). NetTCR-2.0 retains the advantage of accounting for differing TCR sequence lengths while gaining the improved processing ability for paired TCRαβ sequence data. In this study, the authors demonstrated that models using α-chain CDR3 or β-chain CDR3 data alone showed variability in results and decreased performance (69).

We developed pMHC–TCR binding prediction network (pMTnet) to predict specific TCR binding interactions with MHC class I-bound T-cell neoantigens. pMTnet uses the β-chain CDR3 sequence as an input and relies on a novel branch of deep learning called “transfer learning,” which utilizes related TCR and pMHC data that lack pairing information, as well as a training program that allows for the differentiation between binding versus nonbinding TCRs. pMTnet achieved an AUROC (area under the ROC curve) of greater than 0.83 on multiple cohorts. This accuracy extends to TCRs, epitopes, and HLAs that had not been observed before. Using pMTnet as a knowledge discovery tool, we validated the high immunogenicity of human endogenous retrovirus E (HERV-E) in kidney cancers and showed HERV-E is more immunogenic than most tumor neoantigens. With pMTnet, it is possible to develop prognostic and predictive tools for patients with cancer, using TCR and tumor antigen data (35).

Moreover, several computational tools have been developed to help understand V(D)J recombination and subsequent selection processes. IGoR (Inference and Generation of Repertoires) is a software tool that takes immune sequence reads of cDNA or gDNA as input and then quantifies the statistics of receptor generation (70). IGoR assumes that random choice of germline segments and indels contribute to the recombination of receptor sequences and that there is a constant error probability throughout the sequence when dealing with the TCR. IGoR can be used to process V(D)J recombination–related tasks, such as inferring the recombination statistics, analyzing specific recombination scenarios and their likelihoods, and generating synthetic sequences (70). Recently, IGoR has contributed significantly to T-cell research, facilitating the distinction of T-cell lineages, comparing the efficacy of TCR profiling methods, and characterizing diversity of TCR repertoires (71–73).

OLGA (Optimized Likelihood estimate of immunoGlobulin Amino-acid sequences) is a computational tool developed to estimate the probability of generating TCR and B-cell receptor (BCR) sequences or motifs with dynamic programming (74). OLGA can be directly applied to human and mouse TCRβ loci, as well as human TCRα and IGH loci. Other species and loci can also be analyzed by learning recombination models using IGoR. Studies in many fields have benefited from OLGA, including vaccine development and computational immunology (75–78).

During development, T cells undergo the selection process based on their binding properties; Sethna and colleagues developed a computational method, called SONIA, to quantify the selection pressure of CDR3 amino acid sequence (79). SONIA outputs the selection pressure factors for features of the input TCR CDR3β sequence. These selection factors can be used to indicate how likely a sequence is to pass the selection process. The postselection probability of this sequence can be calculated using this selection factor.

Numerous other methods have also been developed to tackle the issue of TCR antigen-binding prediction, including DeepTCR, TCRGP, and TITAN (Tcr epITope bimodal Attention Networks). DeepTCR uses CNNs to parse through TCR sequencing data; it integrates α-chain CDR3 or β-chain CDR3 sequences with V/D/J gene segment usage to classify and cluster antigen-specific TCRs (80). TCRGP is a computational tool using Gaussian process, which has the ability to generate a robust prediction from small datasets, to predict the binding of TCR and epitopes (81). TITAN is a bimodal neural network method developed to predict the binding of TCR and epitope. It uses k-nearest neighbor algorithm as the baseline and encodes both TCR and epitope sequences for the generalization purpose (82).

Overall, these algorithm-based tools for analyzing TCR sequencing data aim to help researchers better understand antigen specificity, offering a wealth of information regarding the complexity of the immune response. In their respective manners, these tools have all contributed to developing new ways to characterize, monitor, and treat a variety of human diseases, including infectious disease, autoimmune diseases, and cancer.

Because of TCR-antigen promiscuity, functional assays must be performed to validate TCR-pMHC pairing predictions made by algorithm-based TCR sequencing analysis tools (83). Common experimental techniques for such validation include ELISA, enzyme-linked immune absorbent spot (ELIspot), multimer staining, and activation marker expression. Both ELISA and ELIspot assess T-cell activation through cytokine release, which occurs when T cells bind to their cognate pMHC and is routinely used as an indication of activated T cells. ELISA assesses the total concentration of cytokines released, while ELIspot detects individual cytokine-secreting clones (84, 85). On the other hand, multimer sorting is used to assess the binding of a potential TCR-pMHC pair. Multimers, like tetramers and dextramers, can be engineered to present target pMHC for target T cells. Because multimers are labeled with fluorophores, multimer stains allow for the quantification and isolation of T cells by flow cytometry (86).

Beyond these commonly used techniques, the upregulation of both stimulatory and inhibitory proteins during T-cell activation can be monitored for functional validation. 4-1BB, OX40, and CD103 have been found to enhance T-cell expansion and survival while PD-1, TOX, and CD39 have been shown to dampen T-cell responses; all of these markers have been found to be enriched on antigen-specific T cells in different contexts (87–92). Yet, the utility of these markers is limited, as their discovery stems from evaluating only a small minority of antigen-specific T-cell responses and they fail to distinguish between antitumor and bystander T cells. Several novel functional assays were developed to address such shortcomings while also enabling the processing of more TCRs and antigens. Yeast antigen display libraries, which involve the expression of human MHC on the surface of yeast, facilitate the rapid identification of peptide ligands and T-cell recognition events (93). Another assay was developed to track T-cell trogocytosis, a phenomenon where T cells share their lipid membrane and membrane-associated proteins while conjugated with their target APCs (94). Signaling and antigen-presenting bifunctional receptors on APCs have been developed to induce TCR-like signaling after binding of TCR with its cognate antigen (95). Finally, tandem minigenes offer the ability to assess several potential TCR-antigen pairs by designing multiple antigen sequences into one construct (96).

While the previously described assays provide valuable information, they are ultimately either limited in their sensitivity and scope or biased by the focus on a specific TCR. These limitations make assessing the repertoire of functional tumor-associated T cells particularly difficult. As a result, Danilova and colleagues developed MANAFEST (Mutation-Associated NeoAntigen Functional Expansion of Specific T cells), which uses TCR sequencing of peptide-stimulated cultures and a bioinformatics platform to both analyze antigen-specific clonal expansion and characterize the expanded TCR clonotypes in a high-throughput and sensitive manner. This assay can drive the development of personalized cancer immunotherapies, as well as the identification of treatment response biomarkers (97). Importantly, adaptations of MANAFEST can be used to detect T-cell responses to specific viral, bacterial, and self-antigens. The same group developed ViraFEST, which has the ability to characterize T-cell responses against different viral antigens including human immunodeficiency virus epitopes (98).

With major advances in TCR sequencing, T-cell antigen specificity prediction, and functional validation of T-cell antigen specificity, there is motivation to develop strategies to streamline the whole process. In 2015, Klinger and colleagues developed MIRA (Multiplexed Identification of T-cell Receptor Antigen specificity), which combines deep TCR sequencing and several immune assays to simultaneously identify and validate large numbers of antigen-specific T cells (99). Separately, Zhang and colleagues presented tetramer-associated T-cell receptor sequencing (TetTCR-seq) as a means to connect TCRs to their specific antigen, in which functional TCR-pMHC binding is determined using a library of DNA-barcoded antigen tetramers (100). TetTCR-seqHD combines TetTCR-seq with the new ability to characterize gene and protein expression to offer a more informed characterization of T-cell activation. These methods are effective in profiling large amounts of single T cells with precision, while also uniquely identifying T-cell cross-reactivity (101). In addition, Arnaud and colleagues developed NeoScreen as a pipeline for the sensitive identification of rare tumor neoantigens and cognate TCRs expressed by tumor-infiltrating lymphocytes (TIL). This method is based on the early exposure of TILs isolated from tumors to antigens of choice which are loaded onto CD40-activated B cells; TCR and neoantigen predictions are then validated against autologous tumors (102).

Advances in TCR sequencing and TCR repertoire analysis have significantly aided in the characterization of tumors (Fig. 4). Notably, TCR sequencing has promoted the description of immune tumor heterogeneity, which also relates to and cooperates with the genomic heterogeneity of tumor cells in tumorigenesis. Performing TCR sequencing for non–small cell lung cancer (NSCLC), our group has demonstrated significant intratumor differences in T-cell density and clonality between distinct tumor regions and correlation between genomic and immune intratumor heterogeneity, indicating that TCR intratumor heterogeneity is potentially driven by specific antigens in different tumor regions (46). More recently, Joshi and colleagues also characterized the heterogeneity of the NSCLC TCR repertoire and provided more evidence that TCR intratumor heterogeneity reflects the mutational landscape across different tumor regions (103). Significant TCR intratumor heterogeneity was also observed in NSCLC metastases, as well as other cancers like melanoma and renal cell carcinoma (104–106). In the context of limited-stage small cell lung cancer (LS-SCLC), whole-exome and TCR sequencing revealed a homogeneous mutational landscape along with a heterogeneous TCR repertoire; this sheds light on the unique nature of LS-SCLC as having high copy-number alteration levels and a particularly immunosuppressive tumor microenvironment (107). In addition to characterizing intratumor heterogeneity, TCR sequencing can be utilized to observe intertumor heterogeneity. Specifically, our group utilized TCR sequencing to observe heterogeneous T-cell profiles across synchronous metastases from lung cancer and melanoma that correspond to heterogeneous mutational and methylation landscapes (104, 105). Through these studies, TCR sequencing of multiple synchronous tumors enables the assessment of differential tumor growth in treatment-naïve patients, in addition to lesion-specific therapeutic responses to targeted therapy and immune checkpoint blockade between metastases within the same patient (105). Overall, TCR intratumor and intertumor heterogeneity have important implications in elucidating mechanisms of cancer immunity and predicting therapeutic responses to immunotherapy.

In addition to informing on heterogeneity, TCR sequencing offers a way to track tumor evolution through immune monitoring. Combining gene expression profiling and TCR sequencing, Dejima and colleagues showed that T cell–mediated antitumor immunity decreases as lung cancers progress from preneoplastic lesions to invasive lung adenocarcinomas. Specifically, as lung cancers become more malignant, immune-activation pathways are downregulated while immunosuppressive pathways are upregulated and T-cell clonality and T-cell tumor infiltration decreases (108). Cui and colleagues reported that the diversity of the TCR repertoire in the peripheral blood gradually decreased during tumor progression from cervical intraepithelial neoplasia to established cervical cancer and that specific clonotypes with similar CDR3 motifs were present between patients (109). The identification of immunologic trends during tumor evolution can help in the classification of tumor grades and aid in the optimization of immunotherapies based on cancer stage.

On the basis of recent observations that cancer-associated TCRs share similar biochemical profiles which distinguish them from TCRs in normal tissues, Beshnova and colleagues developed a machine learning method called DeepCAT (Deep CNN Model for Cancer-Associated TCRs) to predict whether cancer-associated TCRs are present in a patient's peripheral blood. DeepCAT demonstrated high prediction accuracy when distinguishing between patients with cancer and healthy individuals, with an AUC greater 0.95 for several types of early-stage cancers, including breast cancer, ovarian cancer, and melanoma. Overall, DeepCAT represents a potential method for noninvasive, early cancer detection (110).

TCR sequencing is necessary to improve upon current T cell–based therapies. Adoptive T-cell transfer therapy previously relied on the autologous infusion of expanded tumor-infiltrating T cells; however, many patients do not respond to this therapy due to the low frequencies of neoantigen-specific T cells within the tumor-infiltrating group (111). Advances in T-cell therapies including genetically modified T cells expressing novel TCRs have been developed to elicit tumor-specific immune responses (112). TCR sequencing plays an important role in increasing the efficacy of T-cell therapies through the identification of optimal target antigens and tumor antigen-specific TCRs for these novel T-cell therapies (111, 113). Isolated neoantigen-specific T cells can undergo either bulk or single-cell TCR sequencing to identify dominant TCR clonotypes within a tumor (111).

In addition to driving the specificity of T-cell therapies, TCR sequencing can also be used as a tool to evaluate and track T-cell clonotypes over time after administration of immunotherapies. For example, deep sequencing on the variable region of the TCRβ chain was used to track the in vivo frequency of TCRs reactive against mutant KRAS G12D in a patient with metastatic colorectal cancer who experienced tumor regression after adoptive cellular therapy. Interestingly, the most dominant clonotype upon infusion was not detected after some time while less dominant clonotypes persisted (114). Furthermore, sequencing of the TCRβ chain revealed that clonal diversity of chimeric antigen receptor (CAR)-T cells was highest in the blood of patients with acute lymphoblastic leukemia and non–Hodgkin lymphoma during infusion and declined over time; it was also found that individual T-cell clonotypes exhibited distinct clonal kinetics and had variable impacts on the CAR-T cell pool after infusion (115).

Analysis of the TCR repertoire also allows researchers to elucidate underlying mechanisms of action for immunotherapy, as well as its effects on immune cells. Sipuleucel-T is an autologous cellular therapy and a type of cancer vaccine where APCs are activated against prostatic acid phosphatase antigen, which is widely expressed on prostate cancers (116). Sheik and colleagues employed TCR sequencing to characterize the mechanism of action for sipuleucel-T. They showed that the treatment facilitated the recruitment of T cells into the prostate rather than promoting expansion of T-cell clones, as diversity of circulated T cells decreased while diversity of tumor-infiltrating T cells increased (117). In another study, Wieland and colleagues utilized TCR sequencing to examine the composition of the peripheral blood and tumor-infiltrating CD8⁺ T cells that were activated upon anti–PD-1 therapy in a patient with melanoma. They found an oligoclonal repertoire of CD8⁺ T cells in peripheral blood and showed that the majority of dominant peripheral blood clones did not expand following treatment, while expanding clones were found within the resected tumor (118).

Aside from informing the development of immunotherapies, TCR sequencing and derived TCR repertoire metrics have utility in predicting patient outcomes. TCR diversity represents a prognostic biomarker in patients with melanoma, in which patients with greater T-cell evenness and richness in their peripheral blood, as well as in lymph node metastases, had longer progression-free and overall survival (119). Increased diversity in the baseline TCR repertoire of tumor-infiltrating T cells was found to be prognostic in various cancers, including breast cancer, melanoma, lung cancer, and kidney cancer (120). In a study focused on high-grade serous ovarian carcinoma, high T-cell clonality combined with either genomic instability signatures, which includes homologous recombination deficiency and copy-number variation, or T-cell infiltration into the tumor resulted in high prognostic value (121). Furthermore, patients with expanding T cells exhibiting higher affinity to truncal neoantigens, which are derived from mutations with higher variant allele frequency, were more likely to demonstrate better prognosis and response to immunotherapies in lung cancer and melanoma (35).

The TCR repertoire can also be used to predict patient responses to different cancer immunotherapies. Both increased richness and evenness of TCR diversity in the peripheral blood of patients with advanced melanoma prior to treatment with ipilimumab, an inhibitor of CTLA-4, corresponded to greater clinical benefit upon treatment (122). From a study on advanced NSCLC, patients who exhibited higher TCR richness in their peripheral blood during treatment had the greatest clinical benefit from anti-PD-1 immunotherapy, as well as prolonged progression-free and overall survival. In the same study, patients with greater Jaccard similarity index values between pretreatment and various treatment timepoints showed improved progression-free survival, highlighting the importance of immune monitoring during treatment (123). Both increased diversity of the TCR repertoire at baseline and increased clonality during treatment related to significant responses to nivolumab, an inhibitor of PD-1, in patients with classical Hodgkin lymphoma, as well as to autologous Epstein–Barr virus (EBV)-expanded CTLs immunotherapy in patients with EBV-positive nasopharyngeal carcinoma (124, 125). In addition, greater TCR clonality of pretreatment tumor-infiltrating T cells was predictive for the efficacy of anti-PD-1 immunotherapy in patients with metastatic melanoma (120).

Although cancer immunotherapies have promoted significant antitumor immune responses, they also have the potential to induce inflammation-related toxicities. In addition to serving as prognostic and predictive biomarkers for patient outcome and immunotherapy responses, TCR repertoire metrics can also offer information on potential immunotherapy-related toxicities. Clonality was assessed in the context of immune-related adverse events (irAE) after ipilimumab anti-CTLA-4 treatment in patients with prostate cancer; the expansion of greater than 55 CD8⁺ T-cell clones in the peripheral blood preceded the development of severe irAEs (126). A separate study associated CD4⁺ memory T-cell density and diversity in the peripheral blood of patients with melanoma with the development of severe irAE upon anti-PD-1 monotherapy and anti-PD-1 and anti-CTLA-4 combination therapy (127). TCR repertoire metrics can correlate with both positive and negative clinical outcomes, and it is important to leverage the knowledge gained from TCR sequencing to develop treatments and strategies that promote favorable outcomes while minimizing potential toxicities (Fig. 4).

The utility of TCR sequencing extends beyond applications to cancer. Because T cells and their TCRs play an integral role in the mammalian allogeneic response by recognizing and eliminating incompatible tissues, researchers utilize TCR sequencing to characterize responses to organ transplants (128). Specifically, the TCR repertoire can offer information on the risk of rejection through T cell–mediated rejection and graft-versus-host disease and has been evaluated in the context of kidney, liver, and hematopoietic stem cell transplantations (129–132). In addition to transplants, TCR sequencing has potential use in the biological characterization, disease stratification, and monitoring of autoimmune diseases. For example, from analyzing TCR repertoires, researchers have gained insight into the spatial T-cell dynamics in multiple sclerosis, identified TCR clonotypes specific to celiac disease, and defined both clonotype expansion and TCR chain pairing in type I diabetes (133–135). T cells are also instrumental in the immune response to pathogens. As a result, TCR sequencing is valuable for studying the dynamics of adaptive immunity, surveilling the progress of infection, and developing novel immunotherapies for a number of different viral infections, including SARS-CoV-2 (136, 137). In other health conditions, like ischemic heart disease and acute coronary syndrome, TCR repertoire analysis can be performed for the purpose of risk stratification and diagnosis (138, 139). Overall, TCR sequencing and analysis of the TCR repertoire represents a significant tool in an array of biological contexts beyond the scope of this review (Fig. 4).

Recent technological advances have expanded the scope of possibilities pertaining to analysis of the T-cell repertoire, streamlining the identification and development of novel biomarkers and therapeutics. Single-cell transcriptomics equipped with feature barcoding has been implemented in conjunction with TCR sequencing to obtain paired TCRαβ chain sequences while also providing information about cell surface marker expression and antigen specificity, allowing scientists to effectively link antigen specificity and TCR sequence to phenotype for the first time (29, 140). Importantly, interactions between T cells and their surrounding cells have the ability to influence the function and transcriptome of T cells and require the implementation of spatial transcriptomic technologies like 10X Genomics Visium Spatial Gene Expression platform and Advanced Cell Diagnostic's RNAscope assay, which allows the design of custom probes which could be used to identify specific CDR3β sequences and T-cell clonotypes in tissues (141–143). These methodologies can also be utilized to assess the localization and tumor infiltration of antigen-specific T cells and clonotypes of CAR and TCR engineered T cells. With an increasing number of studies now evaluating the importance of cell distribution, distances between cell types and their impact on clinical outcomes, these advances are poised to contribute to the next generation of discoveries in the TCR sequencing landscape (144, 145). Moreover, use of novel protein folding prediction tools such as Google's DeepMind and AlphaFold are likely to enhance our ability to understand pMHC and TCR interactions going forward and help leverage this knowledge to improve upon immunotherapies and therapeutic responses (146).

TCR sequencing is an invaluable tool for the characterization of a wide range of immune responses, informing on the nature of TCR repertoires and aiding in the identification of TCR-antigen pairs. With specific regards to cancer, TCR sequencing can contribute to the basic characterization of tumor heterogeneity and tumor evolution. From a clinical standpoint, TCR sequencing serves both as the foundation of T cell–based immunotherapies and a tool to assess the efficacy of various types of immunotherapies. Furthermore, TCR repertoire metrics derived from sequencing data analysis can be significant predictors of patient outcomes, anticancer treatment responses, and the risk of immune-related toxicities in many different cancer types. NGS methodologies and TCR data analysis tools are continually improving, exhibiting greater processing capabilities and expanding research and clinical applications.

J.V. Heymach reports other support from AstraZeneca, EMD Serono, Boehringer-Ingelheim, Catalyst, Genetech, GlaxoSmithKline, Hengrui Therapeutics, Eli Lilly, Spectrum, Sanofi, Takeda, Mirati Therapeutics, Bristol Myers Squibb, BrightPath Biotherapeutics, Janssen Global Services, Pneuma Respiratory, RefleXion, and Chugai Pharmaceuticals outside the submitted work. J. Zhang reports personal fees from AstraZeneca, Geneplus, Hengrui, Innovent, Merck, and Roche; grants and personal fees from Johnson and Johnson and Novartis; and grants from OrigMed outside the submitted work. A. Reuben reports personal fees from Adaptive Biotechnologies outside the submitted work. No disclosures were reported by the other authors.

A. Reuben was supported by the Exon 20 Group, Rexanna's Foundation for Fighting Lung Cancer, the Waun Ki Hong Lung Cancer Research Fund, MD Anderson's Lung Cancer Moon Shot, the Petrin Fund, the University Cancer Foundation via the Institutional Research Grant program at the University of Texas MD Anderson Cancer Center, the Happy Lungs Project, RETpositive/LUNGevity (FP00015320), and the Cancer Prevention & Research Institute of Texas (RP210137).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.