Entering the Age of Specificity in T Cell Immunity

1. Introduction

Adaptive immunity is defined by antigen‐specific recognition and recall [1]. For conventional αβ T cells, the target epitope is a peptide (p) bound in the groove of a major histocompatibility complex molecule (MHC). The combined discoveries of MHC restriction, epitope‐associated peptides, and the T cell receptor led to sustained efforts to identify pMHC epitopes in experimental and human disease studies, including infections and tumors [2]. In mice, these efforts could be fairly comprehensive because of the inbred nature of laboratory strains and recurrent use of common MHC haplotypes (e.g., H‐2b or H‐2d haplotypes) for a small number of MHC genes (K^b or D^b).

For humans, epitope mapping is substantially more complicated because of the diversity of the human leukocyte antigen (HLA) locus. The three Class I loci A, B, and C and the three Class II loci DR, DQ, and DP together generate at least 12 maternally and paternally derived molecules per individual, and hundreds to thousands of alleles exist in the population [3]. Dominant epitopes are therefore generally identified only in highly prevalent HLA contexts, such as HLA‐A*02:01 [4]. Comprehensive epitope mapping has been performed for a few highly relevant infections but is not the norm [5]. Mapping individual epitopes is tedious work with limited evident short‐term reward. As a result, major epitope mapping efforts have been largely restricted to extraordinary events, such as the emergence of SARS‐CoV‐2, or to the generation of new reagents for specific populations.

Tracking epitope‐specific responses has generally relied on two types of experimental approaches. The first, and most common in the literature, is stimulation‐dependent phenotypic readouts, including ELISPOT, intracellular cytokine staining, and activation‐induced marker assays [6, 7]. These approaches are robust and relatively low cost, and prior knowledge of a subject's HLA is not required. Stimulation usually occurs with pools of peptides, though minimal peptide epitopes or whole inactivated pathogens can also be used. When pools or pathogens are used, the resulting measurement is a bulk estimate of total response magnitude, with variable information about the state of responding cells. Higher resolution is obtained with flow cytometry‐based methods, but these still typically resolve states rather than specificities (Figure 1).

At the other end of the spectrum are MHC multimer‐based methods, usually tetramers of peptide–MHC molecules linked by streptavidin–biotin binding to fluorescent probes [8]. Since their introduction in the late 1990s, tetramers have been the mainstay of epitope‐specific detection, offering high sensitivity and specificity. Their drawbacks include a requirement for HLA knowledge in human studies and relatively high cost. Because they only capture information on epitopes that are directly assayed and because cost and other limitations restrict the number of epitopes tested, multimer‐based studies can only provide focused measurements of selected assayed specificities rather than global estimates of response magnitude.

Stimulation‐based methods overcome the cost and HLA limitations of multimer staining but have unknown and potentially variable levels of both sensitivity and specificity. Stimulation can overestimate responses by driving activation in bystander populations and underestimate responses by failing to trigger a particular effector phenotype under the chosen conditions. In addition, when examining cells from ongoing infections or tumors, T cells may already be activated, making it difficult to attribute observed changes to experimental stimulation.

Collectively, the complexity of human epitope distributions resulting from HLA diversity and the bulk resolution of stimulation‐based methods or selectively targeted multimer approaches mean that comprehensive specificity profiling of human T cell responses against most antigenic targets is rare. These constraints, along with the cost and limited throughput of epitope profiling assays, in combination contribute to the absence of robust epitope‐aware correlates of protection analogous to antibody titers for major infectious diseases. Formal tests of the hypothesis that individual epitope‐specific responses determine variation in disease manifestations across human populations have been sparse.

Here, we discuss T cell specificity in the strict sense: the recognition, by specific TCRs, of specific peptides presented by specific MHC molecules. We distinguish this from the specificity of T cells with undefined TCRs toward whole proteins or even entire pathogens, as is frequently measured using stimulation‐based methods. The key distinction is that when both the paired TCR and pMHC sequences are known, the interaction can be reproduced in vitro without any additional information, and reasonable predictions can even be made for closely related, previously untested TCR or pMHC sequences. In contrast, the results of ELISpot or intracellular cytokine staining stimulation experiments cannot be reproduced without access to the original T cell samples.

In this review, I explore the hypothesis that a substantial fraction of the variation in human susceptibility to infectious diseases and responsiveness to tumor immunotherapies reflects differences in epitope‐level specificity and the T cell responses those epitopes elicit. Although this idea is broadly accepted in principle, it has rarely been tested systematically, for the methodological and conceptual reasons outlined above. I first review efforts to link epitope‐specific responses to protection or risk, highlighting cases in which the identity of the targeted epitope, rather than the overall magnitude of the T cell response, appears to be the critical determinant of outcome. I then consider unconventional T cell populations, which illustrate how developmentally programmed and semi‐invariant repertoires complicate simple, state‐based views of specificity. Next, I discuss the prevailing functionalist framework in which T cell “states” are treated as central explanatory units and examine how this framework has both advanced and constrained our understanding of protective and pathogenic responses. Finally, I argue that recent advances in TCR repertoire profiling and epitope‐mapping technologies now make it feasible to test specificity‐driven hypotheses directly, by quantifying the distribution of epitope‐specific responses in an unbiased and scalable way. I close by outlining the experimental, statistical, and conceptual steps needed to incorporate specificity as a primary variable in studies of human infection and cancer.

2. Evidence for Specificity as a Determinant of Immune Efficacy

3. Beyond Conventional αβ T Cells

Thus far, I have focused on conventional αβ T cells, whose peptide–MHC targets are comparatively well characterized and which comprise the majority of antigen‐specific responses examined in human studies. Unconventional T cells, including γδ T cells, MAIT cells, NKT cells, Tregs, and a growing set of regulatory CD8 populations such as KIR+CD8s, ILTCKs, “hobit” cells, and CD8 suppressors, are usually treated as separate objects of study [77, 78, 79]. I consider them here not because they form a single biological class, but because they expose two limitations of a purely state‐based account of specificity. First, unconventional and regulatory populations can intermingle with conventional antigen‐specific T cells in tissues and tumors and may be difficult to distinguish in antigen‐agnostic phenotypic or single‐cell datasets, leading to misassignment of cellular states and functions. Second, several such populations acquire distinctive functional programs during thymic development or early tissue adaptation, providing particularly clear examples of how receptor features, developmental context, and specificity can jointly constrain subsequent cell fate. Understanding these extreme cases may help define the more subtle ways in which TCR specificity influences differentiation within conventional αβ T cell responses.

γδ T cells are the most distinct unconventional population, as they utilize separate TCR loci and gene families [80]. They are therefore readily distinguished from conventional αβ T cells in sequencing‐based studies, but can be misidentified in flow cytometry analyses without proper comprehensive staining. γδ T cells vary widely across species. Mice have large γδ T cell populations marked by degenerate γδ TCR rearrangements associated with distinct tissue locations (e.g., Vγ5+ T cells in the skin), whereas human γδ repertoires have more diverse CDR3‐sequences overall, but show variation in the extent of diversity based on V‐region usage (i.e., more CDR3 variation in Vδ1 TCRs compared to Vδ2 TCRs). Known γδ ligands include nonclassical MHC molecules [81], certain lipids presented by CD1 and CD1 itself [82, 83, 84], and conformational states of butyrophilin proteins [85, 86], but the ligand landscape remains incomplete. For many γδ subsets, specificity, tissue localization, activation potential, and effector mechanisms are tightly correlated, suggesting that their functional roles are heavily programmed by TCR specificity and developmental context. In some cases, γδ effector states mirror those of conventional T cells; for example, lung Vγ6 cells in mice share IL‐17‐dominated phenotypes with Th17 cells [87]. In others, peripheral activation is clearly TCR‐dependent but the relevant ligands are unknown. Human Vγ9Vδ2 cells are among the best understood. Phosphoantigen metabolites associated with evolutionarily important pathogens such as Mycobacterium tuberculosis and Plasmodium spp bind to the intracellular surface of butyrophilin molecules and induce conformational changes that trigger broad Vγ9Vδ2 activation [85]. Although distinct from classical peptide–MHC recognition, this represents a highly structured specificity tightly linked to functional outcome. Importantly, γδ T cell ligands and the cellular contexts that generate them are not necessarily unique to γδ T cells. Defining these ligands and the molecular states they report may also expose underappreciated targets or modes of recognition for αβ T cells, particularly at the boundary between conventional peptide–MHC recognition and responses to stress‐induced, non‐peptidic, or conformationally altered self.

Regulatory T cells present a particularly complicated case for tracking specificity [88]. Indeed, one could argue that Tregs should not be classified as “unconventional” based on TCR recognition, as Treg populations are routinely observed within conventional epitope‐specific responses, especially in mouse models where high‐avidity tetramers and well‐characterized antigens are available. Many of these populations are induced Tregs that arise during inflammatory responses [89]. Natural thymic Tregs, by contrast, are thought to emerge as a distinct pool of relatively self‐reactive cells, leading to the hypothesis that they have a repertoire skewed toward self antigens and partially distinct from conventional effector populations. Multiple studies have examined the TCR repertoires of thymic and peripheral Tregs, reaching varying conclusions about how distinct they are from conventional CD4 T cells. Some analyses suggest largely separate repertoires with limited overlap, consistent with a self‐reactive agonist‐selected Treg pool [90, 91]. Others find substantial sharing of TCR sequences between Tregs and conventional cells, particularly when induced Tregs are included [92, 93]. Methodological limitations complicate these comparisons. Paired‐chain sequencing throughput is often modest, while deep sequencing of single chains can overestimate overlap or fail to capture clonotype‐level structure. A robust algorithm for identifying bona fide natural Tregs from TCR sequence alone has not yet emerged, and the degree to which Treg fate is encoded by the TCR versus extrinsic signals remains an open question.

Beyond this repertoire ambiguity, the antigens recognized by thymic Tregs remain incompletely defined. Many are not readily detected with conventional foreign peptide–MHC multimers, and their activation kinetics can precede those of conventional effector cells. For example, lung Tregs responding to murine influenza show activation as early as day three and appear to derive from resident populations present before infection [94]. Their rapid activation suggests that these cells are unlikely to be specific for influenza peptides and may instead target self or tissue antigens or use noncanonical docking modes. Functionally and phenotypically, however, they are difficult to distinguish from Tregs that share conventional TCRs. For a review centered on epitope‐level specificity, Tregs therefore illustrate two related problems: protective regulatory activity may depend on poorly mapped self, tissue, or non‐canonical ligands rather than pathogen‐derived epitopes; and identical or overlapping TCRs can support regulatory or conventional fates depending on developmental and tissue context. Repertoire sequence alone is consequently insufficient to infer either functional identity or functional specificity.

The most recently described, and least well defined, unconventional populations include various KIR‐positive CD8 T cells, ILTCKs, and related CD8 suppressor populations [95, 96, 97]. These were identified in gene expression and phenotyping studies and appear to perform nontraditional activities, including suppression and tissue organization. From a TCR repertoire perspective, they are polyclonal and do not contain obvious convergent CDR3 motifs. A related population, characterized by expression of ZNF683 and sometimes referred to as “hobit” cells, has been reported to have biased amino acid usage in relatively long CDR3 regions, including enrichment of hydrophobic residues, suggesting that while these cells do not cluster in the same way as conventional antigen‐specific clones, they may still have selective receptor features [98]. Collectively, these cells may be particularly confounding in tumor and tissue studies because they can share surface phenotypes with effector or effector memory populations, including CD45RA re‐expression, and may be misclassified as conventional memory cells. Additional studies are needed to resolve the function, ontogeny, and specificity of these cells, including possible recognition of equally unconventional antigens. An important open question is whether these CD8 regulatory populations, like thymic Tregs, are selected for characteristic self‐ or tissue‐associated reactivities and whether their suppressive programs are linked to identifiable TCR features, developmental cues, or both. Resolving these questions could provide a useful comparative framework for understanding how receptor specificity contributes to regulatory fate across T cell lineages.

Taken together, these unconventional populations reinforce the central argument of this review. They show that receptor sequence, ligand class, developmental factors, and tissue context can be inseparable determinants of T cell function, while also illustrating how easily state‐based or antigen‐agnostic analyses can conflate biologically distinct populations. A specificity‐centered framework must therefore accommodate not only peptide–MHC‐defined αβ T cell responses, but also the receptor and ligand architectures that generate non‐canonical or regulatory programs.

4. Specificity, Noise, and the Limits of Current Methods

The examples above make a qualitative case that specificity matters. Particular epitopes in HIV, HCV, influenza virus, and tumors are associated with protection, risk, or detrimental outcomes, and unconventional lineages illustrate that some TCRs are developmentally hard wired into distinct functional programs. Yet, despite these case studies, specificity rarely emerges as a central explanatory variable in human immunology. Instead, we tend to describe functional states and their relative magnitudes, such as Tex, Tpex, Tfh‐like, or Trm populations, and to treat these functional states as if they were the primary objects of selection [99].

One reason is that our standard tools inject substantial noise and bias into specificity measurements. Stimulation‐based assays, including ELISPOT, ICS, and activation‐induced marker assays, rely on particular effector readouts and can overestimate responses by activating bystander or cross‐reactive cells and underestimate responses by missing specificities that do not produce the measured cytokines under the chosen conditions. Multimer‐based approaches only sample predefined epitopes, require prior HLA knowledge, and can miss low‐affinity or structurally constrained interactions. They are highly specific for what they test but blind to everything else. Layered on top of HLA diversity, private antigens, especially in cancer, and under‐characterized ligands for unconventional cells, these constraints mean that much of the specificity space is simply unobserved.

The result is that specificity often functions as an unmeasured confounder. When we cluster cells by transcriptomic state or surface phenotype and correlate those clusters with outcomes, we usually lack information on which epitopes those cells recognize, what their cross‐reactivity profiles are, or how their target antigens are distributed across cells and tissues. Some fraction of between‐individual noise in such analyses may reflect genuine differences in epitope targeting and clonotype composition that our assays do not capture. Conversely, our functional classifications and subset names, including Tex, Tpex, Trm, and “stem‐like” memory cells, combine multiple biological attributes, such as lineage, differentiation state, localization, and antigen exposure, into single labels that are not explicitly tied to specificity, making it difficult to compare across studies or to integrate epitope‐level information when it is available. Some newer methods, including the “reverse phenotyping” approach, merge specificity and state assessment into a single analytical workflow [100].

In the remainder of this review I take two complementary approaches to this problem. First, I argue that our nomenclature and conceptual frameworks for T cell subsets should more explicitly accommodate specificity, whether defined by epitope, restricting element, or TCR motif, in parallel with functional state. Second, I outline how recent advances in TCR repertoire profiling, reverse epitope discovery (identifying epitopes starting from TCRs with certain features, such as biochemical motifs, sharing across infected subjects, or clonal expansion after antigen exposure), and high‐throughput pMHC libraries now make it realistic to measure specificity at scale and how study designs and analytical strategies can be adapted to treat specificity as a primary variable rather than latent noise.

5. Specificity and Nomenclature

A recent T cell nomenclature consensus statement addresses two important problems with current practice [101]. It asks authors to state explicitly how they define T cell subsets in the methods sections of primary research papers and proposes a modular scheme that separates lineage, function, migration, differentiation state, and antigen status into distinct descriptors. A modular framework that decouples these axes is a sensible response and is likely to improve transparency and comparability of state‐based analyses.

However, specificity remains largely implicit in this framework. Antigen appears only as a contextual flag, for example indicating whether cognate antigen is thought to persist or has been cleared, rather than as an explicit description of what the TCR recognizes. The modular schema allows one to denote, for instance, a CD8 T cell that is exhausted, resident, and antigen‐experienced, yet the specificity information remains in surrounding text, figures, or methods, not in the naming scheme itself. When examples in the consensus article refer to antigen‐specific cells, they generally assume that specificity has already been established by techniques such as tetramer staining or recent vaccination history, but they do not introduce any formal field for epitope identity, restricting element, or TCR‐defined motif. In practical terms, two studies can describe the same cell subset (e.g., CD8 Tex cells) responding to completely different antigens, and there is no systematic way to determine whether they are discussing the same epitopes or even the same broad antigenic targets.

For infections and cancer, this is a nontrivial omission. The case studies above show that epitope and clonotype specificity can determine whether T cell responses are protective, neutral, or detrimental and can influence exhaustion phenotype, tissue localization, and memory potential. Yet our dominant naming conventions encode function and migration in considerable detail while often omitting the variable most likely to explain inter‐individual differences in outcome, namely what the TCR actually recognizes. State labels are useful shorthand for differentiation and localization, but without explicit specificity information they collapse responses directed at very different antigens into single conceptual bins.

One way to address this gap, without discarding the modular framework for describing T cell states, is to treat specificity as a coequal axis that is reported in a standardized way whenever it is known. In the same spirit that current guidelines urge authors to define subset criteria in the methods, one can argue that when antigen specificity has been established, for example by tetramers, TSCAN and SABR‐like pMHC libraries [102, 103], or reverse epitope discovery [104], the epitope or epitope class should be stated explicitly for each major subset. For an infection, this might mean calling out CD8 Tex cells specific for an HLA‐B27 restricted NS5B epitope in HCV or lung CD8 Trm cells specific for an influenza NP₃₆₆ epitope. In tumors, it might involve specifying whether a given Tex or Trm population is directed against a defined neoantigen, a shared oncogenic driver, or a viral antigen in virus‐associated cancers. This information does not have to be forced into the nomenclature string, which is already arguably too baroque, but it should be present in a systematic and easily accessible format, for example under a “Specificity and TCR” subheading in the methods and in a summary table linking each named subset to its known specificities.

In many human studies, especially in cancer and autoimmunity, exact epitopes are not known or are too numerous to name individually. Even in these settings, it is possible to capture aspects of specificity by reporting TCR‐defined clusters. Algorithms such as TCRdist [105, 106], GLIPH [107, 108], ALICE [109], and CoNGA [98] can identify recurring motifs or clusters of TCRs that likely share specificity, even when the cognate epitope has not been mapped. In such cases, authors could at least associate major T cell subsets with dominant TCR clusters, indicating, for example, that a CD8 Tex population in a tumor is enriched for a particular motif‐defined cluster or that a Trm population in an infection is dominated by a small number of public clonotypes. As noted earlier, efforts to find both protective and pathogenic repertoire correlates in tuberculosis identified distinct TCR subgroups for each [39]. Ideally a reporting standard would require minimal information to express and confirm the specificity of any TCR in vitro: TRAV, TRBV, TRAJ, TRBJ, and CDR3αβ.

If such TCR clusters were routinely named and deposited, future studies could recognize shared specificities and reuse these annotations, much as B cell epitope names are reused across vaccines and cohorts. At present, each group tends to rediscover and rename its own motif definitions, and cross‐study synthesis requires substantial bespoke effort to determine whether apparently distinct clusters represent shared specificities. A simple convention in which major TCR clusters are given stable identifiers, linked to their associated epitopes when known and referenced in text and tables alongside functional state and migration descriptors would make specificity information discoverable. It would also facilitate precisely the kind of variance‐partitioning analyses discussed earlier, in which one asks how much outcome variance is explained by specificity, magnitude, state, or host factors. Indeed, we are no longer writing papers solely for other scientists to read. Rather, we are producing datasets that can be analyzed and reanalyzed by multiple groups. For gene expression analyses, this is now a relatively mature process, with standard data formats and repositories. No comparable enforced, systematic reporting process or format exists for repertoire data though the Adaptive Immune Receptor Repertoire (AIRR) Community has proposed standards that have been adopted by some journals and societies [110]. Without such steps, we are likely to continue refining our vocabulary for T cell states and migratory patterns while leaving specificity, the parameter most likely to explain a substantial share of outcome variance in infection and cancer, largely unannotated.

6. Looking Toward the Future of TCR Specificity Profiling

The findings discussed here show that there are strong signals of specificity as an independent correlate of risk and protection, in both human association studies and animal models. These studies have largely relied on either coarse‐grained measures of specificity, including broad HLA associations, whole‐protein stimulation assays, and peptide pools, or highly specific epitope associations restricted to a single HLA or animal model. Collectively, they point to the importance of specificity, but as a field we have lacked methods for systematically testing the role of specificity with both precision, meaning epitope‐level resolution, and breadth, meaning independence from a few selected HLA types or model systems. Despite its centrality to our concept of adaptive immunity, we have never formally tested the contribution of specificity at scale in infections or tumors.

The emergence of technical and computational approaches for TCR repertoire analysis provides a new model for how to formally assess specificity without compromising either breadth or precision. Several advances have been critical. High‐throughput sequencing platforms first made acquisition of repertoire data tractable. Initial single‐chain 5′ RACE‐based approaches [111, 112] and DNA‐based multiplex primer methods generated large datasets [113] for developing analytical methods. The development of one‐cell‐per‐well paired‐chain sequencing allowed capture of complete TCRαβ pairs and formal confirmation of specificity associations by cloning and re‐expression, typically on the order of tens to hundreds of cells [114, 115, 116]. The next major advance, single‐cell RNA sequencing [117] with integrated TCR capture, scaled these workflows by two to three orders of magnitude, making it routine to obtain tens of thousands of paired‐chain TCRs from a sample.

Even so, ten to twenty thousand cells from a human sample is a drastic undersampling of true repertoire diversity, often returning only low hundreds of unique clonotypes because of the heavy‐tailed distribution of TCR frequencies [118, 119, 120]. A small number of extremely large clones occupy a significant proportion of the repertoire and are disproportionately sampled in random draws from blood or tissue. Pathogen‐specific clones, even at the peak of clonal expansion, can range from 10⁻¹ to 10⁻⁵ of total T cells in peripheral blood [121]. For unbiased specificity tracking, methods that can obtain paired‐chain sequences from millions of cells are therefore essential. Recent approaches, including combinatorial barcoding schemes (TIRTL‐Seq [122], PAIR‐Seq [123]), REFLEX [124], SPLiT‐seq [125], and related protocols, can now routinely generate repertoires from millions of input cells. This brings the sampling depth required to resolve low‐frequency epitope‐specific clonotypes into reach.

In parallel, antigen‐agnostic methods for identifying specific responses within repertoires have been developed. A key insight was that TCRs targeting the same epitope are often similar in a defined sequence space, as quantified using distance metrics such as TCRdist and clustering approaches like GLIPH and related methods [105, 107]. Subsequent refinements including ALICE [109], GLIPH2 [108], and tcrdist3 [106] have defined metaclonotypes or specificity groups, clusters of TCRs linked by sequence motifs and enriched for shared antigen recognition, informed by recombination probabilities learned from deep single‐chain sequencing [126]. These metaclonotypes can be detected in bulk repertoire data even when the exact epitope is unknown and provide a way to project millions of TCRs into a smaller number of putative specificity groups.

The next step toward comprehensive specificity determination was the recognition that longitudinal sequencing [121, 127], or comparative cohort studies [128, 129, 130], could be used to capture condition‐associated TCRs. Sampling an individual at baseline and during the acute response, or at the acute peak and in convalescence, yields sets of clonotypes that expand and contract over time and can be inferred to be pathogen‐ or vaccine‐specific responses. Alternatively, identifying TCRs present only in exposed groups but absent in well‐matched controls can generate a list of potential condition‐associated TCRs. Coupled with clustering approaches, these dynamic clonotypes can be recovered from multiple individuals with the same exposure, creating maps of metaclonotypes in which each motif corresponds, in principle, to an epitope‐specific response (Figure 2). This strategy has been applied extensively in SARS‐CoV‐2 cohorts, where large amounts of public repertoire data and HLA annotation have enabled robust inference of condition‐associated TCR clusters and their likely restricting elements [106, 130].

Even at this stage, a form of specificity association is possible. Metaclonotype maps provide insight into the distribution and magnitude of epitope‐class‐specific responses across individuals, and one could test for associations between these motif‐level responses and clinical outcomes. However, the field is accustomed to thinking in terms of individual epitopes, and moving from sequence motifs to mapped peptide–MHC epitopes remains important. The first step in this process, called “reverse epitope discovery,” is assigning HLA restriction to each metaclonotype cluster [104, 131]. In large cohorts with HLA annotation, this can be done by testing for enrichment of particular alleles among individuals who carry the cluster, and even in cohorts without formal typing, HLA alleles can be inferred from repertoire data using established tools, especially in adults whose memory repertoires reflect shared common exposures [132, 133, 134].

Once HLA is assigned, the more difficult problem is peptide mapping. Two bottlenecks must be overcome to routinely link clusters to specific peptides. The first is the capacity to synthesize large numbers of TCRs for cloning and expression, enabling high‐throughput functional mapping. Initially, this was facilitated by DNA synthesis services that could produce expression‐ready TCR constructs at moderate cost. Recent advances, including streamlined cloning protocols such as TCRAFT [135], have reduced the cost and labor required to generate TCRs to a level where testing hundreds or thousands of receptors per study is feasible [136]. The second bottleneck is antigen discovery. For constrained immunogens, such as Spike‐only vaccines, de‐orphanization can be accomplished with peptide scans or combinatorial peptide pools. For larger antigens, combinatorial pooling and matrix designs can substantially increase efficiency [137, 138].

For extremely large pathogens or complex autoimmune targets, whole‐proteome peptide synthesis and screening, for example across all herpesvirus proteins or all self antigens, remains impractical. To address this, several library‐scale screening platforms have been developed. These include systems that utilize large yeast display libraries with soluble TCRs for panning [139], couple TCRs or antigens to lentiviral barcodes [140, 141, 142], or display libraries of candidate antigens on antigen‐presenting cells, followed by tagging recognized cells for co‐sorting and sequencing [102, 103, 143, 144, 145]. These methods have already been used to de‐orphan clinically important TCRs in autoimmune diseases such as Sjögren's syndrome and axial spondyloarthritis [28]. As efficiency improves and adoption widens, these platforms will allow construction of increasingly complete metaclonotype maps that link TCR sequence clusters to defined specificities across diverse HLA backgrounds.

The end result of these advances will be TCR codices, mapped TCR motifs that collectively cover the epitope space of a pathogen or target antigen across a representative distribution of HLA alleles. Once such codices exist, longitudinal sequencing will no longer be strictly required to assign specificity. Instead, bulk or single‐cell repertoire data can be directly annotated by matching TCRs to motif libraries, yielding epitope‐level specificity calls at scale. As additional pathogens and antigens are curated, these codices will mature, becoming more complete as new targets and variants are added. Direct assessment of specificity will be possible from a single repertoire sequencing reaction at a sensitivity that exceeds functional assays and even multimer‐based methods, with little or no HLA bias. Emergence of new epitopes through mutation and loss of previously mapped responses through escape or antigen loss will be detectable as changes in motif‐level frequencies. It is difficult to see any other near‐term method that offers comparable tractability and comprehensiveness for specificity mapping.

As noted, bulk single chain sequencing remains the dominant mode of TCR repertoire profiling, due to its relatively low cost, technical advantages, and commercial availability. Its value for specificity inference, however, depends heavily on information external to the sequence itself. Once a pathogen‐specific TCR codex has been developed, convergent enrichment of multiple single‐chain motifs associated with that pathogen can provide probabilistic evidence that a categorized specific response has occurred. Confidence increases when these annotations are concordant with HLA genotype, longitudinal clonal dynamics, exposure or serologic history, tissue context, or independent functional measurements [133, 146]. These inferences should nevertheless be distinguished from direct specificity assignment. On their own, an isolated α or β chain, even one containing a strong, well‐characterized motif, will not maintain specificity when paired with the vast majority of randomly drawn partner chains [147, 148]. Bulk single‐chain data can therefore support association, clustering, and hypothesis generation, but generally cannot establish the cognate pMHC specificity of an individual clonotype. Paired αβ sequence information remains the preferred basis for definitive receptor‐level specificity assignment and for functional confirmation by cloning and re‐expression.

Finally, it is important to acknowledge parallel efforts to develop algorithms for TCR specificity decoding from sequence alone, with the most promising approaches leveraging structural modeling and deep learning [149, 150]. These tools will provide a powerful complement to the empirical strategies described above, improving resolution within metaclonotype maps, predicting specificity for rare or private clonotypes, and increasing the efficiency of the de‐orphanization process. Together, empirical codices and predictive models make it realistic, for the first time, to treat specificity as a measured variable at scale rather than an inferred property of a few selected epitopes.

7. Conclusion–the Age of Specificity

The extent to which specificity contributes to human immune variation has never been formally quantified. That long‐standing unknown is now within reach. Advances in TCR profiling methods have matured to the point that we are on the cusp of routine, comprehensive specificity profiling in laboratory experiments, diagnostic assessment, and therapeutic response monitoring. Initial efforts to build infection‐specific codices will demonstrate the power and efficiency of these approaches for identifying epitope‐level correlates of risk and protection and for improving rational vaccine design that targets specific responses rather than generic magnitudes.

The low cost and high throughput of new TCR repertoire profiling methods permit the development of routine clinical platforms that could be used for vaccine response monitoring, disease diagnostics, and predicting future susceptibility to infections. As structure‐aware and data‐driven methods for TCR specificity prediction improve, they will further unlock these approaches for therapeutic development and clinical applications, including TCR‐engineered therapies and specificity‐aware checkpoint strategies. Most importantly, treating specificity as an explicit, resolved variable will allow us, for the first time, to ask and answer quantitative questions about how much of human variation in infection, vaccination, and cancer therapy outcomes is driven by which epitopes are seen, not just by how many T cells respond or what state they occupy.

Funding

This work was supported by R01 AI136514, the St. Jude Center for Influenza Research and Response (SJCEIRR) contract 75N93021C00016, and the Center for Influenza Vaccine Research in High Risk Populations (CIVR‐HRP) contract 75N93019C00052.

Conflicts of Interest

P.G.T. has served on the Scientific Advisory Boards of Immunoscape, Shannon Bio, Osivax, and Intrcon, has received research support and personal fees from Elevate Bio, and consulted for 10× Genomics, Illumina, Pfizer, Cytoagents, Sanofi, Merck, and JNJ. P.G.T. has patents related to TCR amplification, cloning, individual TCR sequences, and/or applications thereof.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analysed during the current study.