Calibrating T cell responsiveness through interactions with self

Abstract

During an immune response, T cells face one of the most consequential decisions of their lifespan upon recognition of a ligand they have not previously encountered: whether to exit the naive basal state, undergo clonal expansion and acquire effector functions. This process is often portrayed as a binary switch, in which naive cells from a highly diverse repertoire transition from an ‘off’ state to an ‘on’ state. However, this digital view overlooks the crucial prior information that T cells integrate through T cell receptor (TCR) interactions with self-peptide–MHC (self-pMHC). During thymic selection, immature T cells encounter a unique self-pMHC ligandome that shapes their development. After maturation, naive T cells continue to engage self-ligands as they patrol secondary lymphoid organs. Here we review evidence that these encounters with self-peptides are not only essential for T cell survival but also have lasting consequences that dynamically tune T cell function when called into action. The naïve off state, therefore, is neither fixed nor functionally neutral. We argue that a deeper understanding of an individual’s self-peptide repertoire is crucial for deciphering TCR self and non-self discrimination and for effectively harnessing T cell responses to foreign antigens.

Introduction

A conceptually attractive yet overly simplistic model to explain how T cells discriminate self from non-self holds that the pool of circulating naive T cells is essentially devoid of cells bearing antigen receptors that recognize self, largely as a result of negative selection [G] in the thymus¹. In this view, most cognate peptides encountered by a naive T cell in the periphery are therefore foreign, providing a reductive self–non-self framework for how the immune system mounts robust responses against an array of pathogens while maintaining a peaceful coexistence with self-antigens. However, this framework fails to capture key features of T cell biology. First, clonal deletion is incomplete: autoreactive T cells are readily detected in healthy individuals and are restrained by peripheral tolerance [G] mechanisms; this preserves T cell receptor (TCR) diversity to maintain immune coverage^2–8. Second, and central to this review, the simplified self–non-self dichotomy conceals what we now know to be a striking paradox: although the TCR repertoire has been pruned of overtly autoreactive cells, all T cells require a degree of self-reactivity for survival, during both positive selection [G] in the thymus and during peripheral homeostasis. This raises fundamental questions. If self-reactivity is an inherent feature of all naive T cells, what underlies its evolutionary and functional necessity? How does the immune system nonetheless discriminate self from non-self?

Increasingly, self-recognition is understood not just as a transient developmental checkpoint but as a determinative programme that is initiated in the thymus and sustained in the periphery to “train” T cells how to interpret and respond to foreign antigens. One provocative hypothesis posits that moderate TCR cross-reactivity and exposure to a well-chosen subset of self-peptides promotes a form of learning from encountered self-peptides to bias the repertoire against responding to self while preserving reactivity to foreign antigens⁹. Complementing this view, recent work challenges the classical role ascribed to positive selection as being primarily in place to enforce MHC restriction [G]. Instead, data suggest that positive selection imprints functional properties based on early self-recognition events^10–14. Crucially, T cell engagement with self-peptide presented by MHC (self-pMHC) does not end at thymic exit; ongoing interactions in the periphery sustain T cell survival and maintain a poised, ready state that enables rapid activation upon antigen encounter. These insights converge on a compelling possibility: that self–non-self discrimination, and the T cell response more broadly, are not governed solely by elimination of dangerous clones, but by a more complex relationship with the self-peptides encountered.

Despite its importance throughout the life of a T cell — from ensuring the selection of a functional and diverse T cell repertoire in the thymus to the survival and trafficking dynamics of naive T cells in the periphery¹⁵ — the identity and characteristics of self-peptides that meaningfully shape T cell fate remain largely undefined “dark matter”. Although self-peptidome studies have characterized several hundred to a few thousand MHC-bound peptides at a time, such analyses typically still require very large numbers of starting material in excess of 10⁸ cells^16–19. Given that antigen presenting cells (APCs) express on average 10⁵ MHC class I (MHCI) or class II (MHCII) molecules^20,21, it remains unclear whether the detectable self immunopeptidome [G] represents only the tip of the iceberg and, crucially, which of these self-peptides are actually engaged by individual TCRs in vivo. Thus, much like dark matter in astrophysics whose existence is deduced from its effects rather than directly measured, the sub-threshold interactions [G] of conventional ab T cells via their unique TCR with self-peptide ligands (those that do not lead to overt activation and, hence, autoimmunity) are mostly inferred from the imprints they leave, shaping cell fate and behaviour without being directly detected. The effects of sub-threshold self-peptide interactions are revealed through indirect proxies and surrogate readouts (Box 1) given that it remains technically challenging to detect the transient, low affinity interactions that T cells make with self-pMHC. Here we discuss emerging evidence that these elusive self-peptide interactions shape the T cell response more broadly than previously appreciated, prompting a conceptual shift from viewing self-recognition as a liability to recognizing it as a fundamental determinant of T cell fate, responsiveness and self–non-self discrimination.

Box 1. Surrogate markers of T cell reactivity for self-peptide–MHC.

Thymocytes and naive T cells expressing an ab T cell receptor (TCR) require interactions with self-peptides presented by MHC (self-pMHC) for their development and maintenance. In the periphery, self-pMHC interactions promote what is referred to as tonic or sub-threshold TCR signalling with an intensity that generally scales with the self-reactivity of their TCR. Readouts of TCR signalling likely integrate both the strength and frequency of TCR engagement. For example, T cells experiencing stronger, and perhaps more frequent, tonic signals show increased TCR z-chain phosphorylation and recruitment of the tyrosine kinase ZAP70 without full T cell activation^131,132. Indeed, although each interaction of a T cell with a cognate self-pMHC likely results in a transient signal, multiple signals experienced serially are “remembered” and can impact cell fate^70,133. Downstream tonic TCR signalling resembles canonical agonist-induced signal transduction, requiring the adaptor protein LAT, and results in calcium flux, translocation of NFAT to the nucleus, phosphorylation of ERK, and activation of mTOR pathways^95,134,135. Although tonic TCR signals are below the threshold for activation and do not induce clonal expansion or effector differentiation, the integrated strength of T cell interactions with self-peptide can nonetheless influence gene expression in naive T cells^71,104. Therefore, differentially expressed genes can serve as surrogate markers of self-reactivity, which circumvents the challenge of directly detecting TCR signals (e.g., tyrosine phosphorylation) in viable cells^{13,71,95,136,137} (see figure).

Analogue markers

Two established surrogate analogue markers of naive T cell self-reactivity in mouse models are CD5 and Nur77^12,46,102, with evidence suggesting both CD5 and Nur77 levels also correlate with self-reactivity in human T cells^136,138. Surface levels of CD5 increase during thymic positive selection, and CD5 remains expressed on all peripheral T cells to varying levels⁴⁶. Notably, the mean fluorescence intensity of CD5 is greater on naive CD4⁺ T cells than on CD8⁺ T cells. CD5 surface expression levels scale with stronger TCR signalling, as confirmed by increased TCR z-chain phosphorylation in CD5^hi versus CD5^low cells^12,14. The cytoplasmic domain of CD5 contains multiple tyrosine phosphorylation sites that may modulate TCR signalling, so variable expression of CD5 on T cells may itself act as a rheostat for regulating TCR signal strength¹³⁹. By contrast, Nur77, encoded by Nr4a1 and part of the NR4A family of nuclear receptors, is an intracellular immediate response gene that is rapidly expressed upon TCR stimulation¹⁴⁰. Independent transgenic mouse strains have been generated where a fluorescent reporter is expressed under the transcriptional control of the Nr4a1 promoter (referred to as Nur77-GFP) and where fluorescence intensity tracks with self-reactivity in naive T cells, facilitating functional studies of T cells across a spectrum of intrinsic self-reactivity^102,141. GFP expressed in this model is uncoupled from endogenous Nur77 and has a longer half-life; as a result, reporter fluorescence accumulates in proportion to Nr4a1 transcription¹⁰². Sorting polyclonal naive T cells into subpopulations of high, medium or low self-reactivity based on Nur77-GFP or CD5 expression has shown key differences in transcriptional cell state and biases in T cell function (Box 3). Due to the nature of both markers, relative expression differences in CD5 or Nur77-GFP reflect previously experienced TCR signalling and are not necessarily reflective of real-time TCR signal intensity. Therefore, short-term temporal variations (minutes) in signalling may be difficult to resolve with either marker, and differences in their kinetics might potentially result in divergence in downstream readouts depending on whether naive T cells are sorted based on Nur77-GFP or CD5 expression. Both CD5 and Nur77-GFP expression will, however, decrease over the longer term (hours) when T cells are removed from tonic self-pMHC signals^71,76,102. Tools such as TCR transgenic models offer one useful perspective on the role of TCR specificity and tonic TCR signal strength, with data showing that both CD5 and Nur77-GFP expression vary in intensity between monoclonal TCR transgenic populations. While the variation in self-reactivity as read out by surrogate markers is narrower in monoclonal than in polyclonal T cell populations, variations in self-reactivity exist within cell populations that express the same TCR and can be used to parse functional biases^12,100. Underscoring this point, Nur77-GFP^low and Nur77-GFP^hi TCR transgenic cells exhibit differences in two-dimensional affinity [G] for cognate pMHC^96,99, but it remains an open question as to why this is the case.

Digital markers

Alternative surrogate markers used to parse naive T cell functional diversity are expressed in a digital (on/off) fashion, of which Ly6C has been best studied so far, although notably only in mouse T cells as it does not have a human orthologue. In naive mouse T cells, Ly6C expression in a polyclonal population is bimodal. Among naive CD4⁺ T cells, the Ly6C⁺ subset exhibits lower self-reactivity than the Ly6C^– population⁹⁵. In line with this, withdrawal from self-pMHC leads to the upregulation of Ly6C expression on CD4⁺ T cells¹⁴². By contrast, Ly6C expression among naive CD8⁺ T cells is observed in mice with increased microbial colonization — a result of greater type I interferon levels and increased tonic TCR signalling — and marks cells with a greater basal activation state, increased CD5 expression and enhanced effector function^143,144.

Setting the tone: positive selection in the thymus

While negative selection is mediated by a variety of different thymic APCs, including dendritic cells (DCs), medullary thymic epithelial cells (mTECs) and B cells, a single stromal cell type, cortical thymic epithelial cells (cTECs), is absolutely required for the positive selection of conventional ab T cells^1,22. Evidence suggests that this specialized role of cTECs is enabled, at least in part, by non-canonical modes of self-antigen processing that likely endow cTECs with a unique cTEC self-immunopeptidome, distinct from the peptides generated by other thymic and peripheral APCs (Fig. 1a). This specialized peptide repertoire may play a role not only in the binary checkpoint of thymocyte survival and lineage commitment, but also in an analogue tuning process that actively and persistently shapes the functionality of naive T cells.

Figure 1. Generation of a distinct repertoire of self-peptides presented by cortical thymic epithelial cells for positive selection of T cells in the thymus.

(a) Wild-type cortical epithelial cells (cTECs), in addition to the standard repertoire of self-peptides generated via the housekeeping proteasome or cathepsin S proteases in lysosomes, express specialized molecular machinery that leads to the production of a cTEC-specific self-peptidome. In the MHC class I (MHCI) pathway, cTECs use thymoproteasomes that contain the b5t catalytic subunit, while in the MHC class II (MHCII) pathway, cTECs use the specialized protease cathepsin L and thymus-specific serine protease (TSSP) to process lysosomal substrates. (b,c) Deletion of either cathepsin L (b), or of the b5t thymoproteasome (c), leads to the presentation of only the standard self-peptides by cTECs by MHCI or MHCII, respectively, which has a substantial impact on the self-reactivity and CD5 expression levels of the selected T cell repertoire. Cathepsin L deficiency in cTECs leads to the loss (dotted lines) of monoclonal populations at the low end of the self-reactivity spectrum, and remaining T cells receive reduced selecting signal strength. b5t deficiency in cTECs leads to the loss of T cell clones at the lower end of the self-reactivity spectrum (dotted lines) but above this shifted positive selection threshold, CD8⁺ T cell self-reactivity is unaffected.

Unique pathways of self-peptide generation in cTECs

In the MHCI pathway, cTECs use so-called thymoproteasomes containing the β5t catalytic subunit (encoded by Psmb11) to generate MHCI-bound peptides, whereas mTECs and haematopoietic APCs process cytoplasmic substrates with the standard β5 subunit or immunoproteasomes incorporating the inducible β5i subunit²³. In the MHCII pathway, cTECs preferentially use the protease cathepsin L (encoded by Ctsl) to process lysosomal substrates, in contrast to other APCs that mostly rely on cathepsin S²⁴. Another unusual protease implicated in the generation of MHCII-bound peptides in cTECs is thymus-specific serine protease (TSSP, encoded by Prss16), which is not expressed by other APC subsets^25,26 (Fig. 1a). Furthermore, unlike haematopoietic APCs that mainly acquire MHCII ligands from endocytosed extracellular proteins, cTECs predominantly generate pMHCII ligands by shuttling endogenous cytoplasmic substrates into lysosomes via autophagy^27,28.

The role of cTEC-specific self-peptides in T cell repertoire generation has been explored by deleting key components involved in peptide presentation. Deletion of TSSP or constituents of the autophagic machinery, including ATG5, VPS34 or LAMP2, which are presumed to impact the MHCII ligandome of cTECs, does not alter the overall size of the CD4⁺ T cell compartment, but impairs the selection of specific transgenic TCRs and alters the composition of the polyclonal repertoire^28–33. More profound defects are observed upon genetic ablation of β5t or cathepsin L, leading to a marked reduction in the thymic CD8 or CD4 single-positive (SP) compartments, respectively^25–27. Importantly, overall levels of MHCI in β5t-deficient or MHCII in cathepsin L-deficient cTECs are not diminished, and indirect evidence indicates that a substantial degree of diversity within their respective pMHC ligandomes is preserved in both cases, likely due to compensatory activity by other proteases^34,35. Thus, mere ligand deprivation is an unlikely explanation for the reduced T cell output observed in these models. Furthermore, excessive negative selection, resulting from an high degree of similarity between the pMHC ligandomes of cTECs and those of negatively selecting thymic APCs when cTEC-specific proteolytic pathways are eliminated — as proposed by the altered ligand hypothesis [G] — can, at most, explain only a minor portion of the contraction observed in the thymic CD4 or CD8 SP compartments^34–38. Although other scenarios, and even peptide-independent functions of β5t or cathepsin L³⁹, cannot be formally excluded, current evidence converges on the view that efficient positive selection of both CD8⁺ and CD4⁺ T cells crucially depends on the presentation of specialized pMHC ligands by cTECs. Corroborating this idea, biochemical studies involving mass spectrometry and peptide sequencing indicate that cTEC-specific proteolytic pathways can indeed generate distinct MHCI-bound or MHCII-bound peptides^40–42. However, the existence of truly unique pMHC ligands on cTECs in vivo remains largely inferred. Even with improved methods for analyzing MHC-bound peptides from scarce cells, only a handful of relatively abundant cTEC peptides has been characterized, and key aspects of the molecular composition and biophysical characteristics of the cTEC-specific immunopeptidome remain unresolved.

Peptide “quality” imprints naive T cell reactivity during positive selection

Studies using TCR transgenic models and large-scale repertoire analyses have shown that the numeric contraction of the CD8⁺ and CD4⁺ T cell compartments resulting from β5t or cathepsin L deficiency, respectively, reflects a broad, clonotype-selective bottleneck in progression through the positive selection checkpoint^27,34,35,43. Given the presumed peptide specificity of positive selection⁴⁴, this may be an expected consequence of an altered pMHC ligandome. A second, more surprising, observation is that the CD8⁺ or CD4⁺ T cell clones that do manage to pass selection despite the absence of β5t or cathepsin L, respectively — that is, clones that appear superficially β5t- or cathepsin L-independent — emerge qualitatively reprogrammed.

Cathepsin L expression in cTECs shapes the CD4⁺ T cell repertoire by modulating the strength of positive selection signals across the full spectrum of self-pMHCII affinities (Fig. 1b). Supporting the notion that positively selecting signals in mice lacking cathepsin L in TECs (Ctsl^ΔTEC) are globally attenuated, polyclonal CD4⁺ T cells have reduced expression of CD5³⁵. This is not due to preferential loss of clones that are typically CD5^hi in wild-type mice; in fact, repertoire sequencing revealed that TCR loss in Ctsl^ΔTEC mice was more pronounced among clones that are typically CD5^low. Thus, clones selected through relatively low affinity interactions in a wild-type setting may fail to receive sufficient signals and die by neglect in Ctsl^ΔTEC mice. By contrast, clones positively selected with relatively high affinity interactions in a wild-type setting remain, potentially owing to compensatory sensitization of proximal TCR signalling pathways. Such compensation may involve reduced expression of CD5 itself and extends to other negative regulators, including CD6, PD-1 and BTLA, thereby enabling sufficiently strong signalling to persist in a cathepsin L-deficient thymic environment. Consequently, the CD4⁺ T cell repertoire selected in the absence of cathepsin L becomes skewed towards clones bearing TCRs that would normally occupy the CD5^hi spectrum (Fig. 1b). However, these clones now masquerade as artificial CD5^low cells with impaired functional competence. Transgenic expression of cathepsin L-independent TCRs showed that these clones, despite efficiently completing positive selection in absence of cathepsin L, are functionally mis-programmed in terms of their response to antigen. Moreover, when selected in the absence of cathepsin L, they also display a reduced capacity for homeostatic proliferation following tonic MHCII encounter and have a survival disadvantage compared to T cells bearing the identical TCR selected in wild-type mice³⁵. These functional differences in vivo are consistent with lower CD5 expression in the absence of cathepsin L, which is particularly notable because CD5 is not only a proxy for cumulative TCR signal strength during selection but also serves as a rheostat of TCR signalling sensitivity^45,46 (Box 1). Decreased CD5 levels may impair in vivo survival, as CD5 levels are thought to calibrate NF-κB signalling⁴⁷, which in turn establishes proliferation competence and enhances responsiveness to homeostatic cytokines such as IL-7^48–50.

Phenotypic and functional differences have also been observed between CD8⁺ T cells selected in the presence or absence of β5t. Polyclonal CD8 SP thymocytes from β5t-deficient mice exhibited elevated CD5 and Nur77 levels, suggesting that β5t-containing thymoproteasomes promote weaker selecting TCR–pMHC interactions^51,52. Supporting this notion, biochemical assays showed that thymoproteasomes generate peptides with distinct sequence preferences near the central MHC anchor positions, thereby shaping the MHCI-restricted immunopeptidome in a way that may favour presentation of lower-affinity TCR ligands⁴⁰. An inverse correlation between CD5 levels on three TCR transgenic CD8⁺ T cell clones and their dependency on β5t was reported; the weaker the selecting signal for a given TCR, the more reliant it was on β5t^34,52 (Fig. 1c). In line with the preferential selection of T cells with high reactivity to self-pMHC, polyclonal CD8⁺ T cells from thymoproteasome-deficient mice have been found to exhibit increased homeostatic cell division in lymphoreplete or lymphopenic conditions compared with their wild-type counterparts^34,52. Further suggesting qualitative differences emerging in the absence of the thymoproteasome, the β5t-independent CD8⁺ T cell clone OT-I was reported to be hyporesponsive to cognate antigen when selected in the absence of β5t^52,53.

It is tempting to speculate that the different outcomes of removing a subset of thymic self-peptides through cathepsin L or β5t deficiency in TECs reflects fundamental differences in the modalities and kinetics of TCR signalling in MHCI versus MHCII restricted thymocytes⁵⁴, potentially contributing to the long-standing yet enigmatic observation that polyclonal naive CD4⁺ T cells express overall higher and more heterogeneous levels of CD5 and Nur77 than their CD8⁺ T cell counterparts⁵⁵. Nevertheless, for both CD4⁺ and CD8⁺ T cells, these findings reinforce the emerging view that, rather than merely ensuring the production of a sufficient number of self-MHC-restricted circulating T cells, positive selection imprints distinct programmes of activation thresholds, homeostatic responsiveness and effector potential, providing an intriguing link to the hypothesis that positive selection exhibits a preference for TCRs that confer superior responsiveness to foreign antigens^12,56.

Temporal and quantitative aspects of self-ligand interactions during positive selection

On the basis of constrained peptide thymic selection models, it has been proposed that positive selection may not require an extensive diversity of peptides to obtain a normal number of naive T cells, but it is not clear how large the set of presented self-pMHC complexes in the thymus needs to be^56,57. In TCR transgenic models, a single defined peptide can drive positive selection and is sufficient to shape CD5 expression, but in other instances multiple distinct peptides were shown to support selection of the same TCR (with a mixture of peptides possibly being more effective)^58–65. Notably, although positive selection is associated with serial TCR signalling events that are experienced over a period of days⁶⁶, it is still unknown whether thymocytes repeatedly engage the same self-pMHC or instead integrate signals from different self-peptides over the course of positive selection. Addressing this question is complicated by two major limitations: first, the thymic self-peptidome, particularly within the cortex, remains undefined, and second, only a small number of endogenous positively selecting ligands have been identified for individual TCRs^59–61. Nevertheless, several observations suggest that more than one selecting ligand may operate for at least a subset of TCRs. For some MHCI-restricted TCRs, positive selection may involve multiple temporally distinct signalling phases, consistent with the existence of “two waves” of TCR signalling⁶⁵. Indeed, DP thymocytes expressing certain TCR transgenes have been shown to migrate into the medulla while continuing to receive TCR signals, implying that signalling can persist beyond the cortical stage⁶⁶. More recently, it has been proposed that thymocytes exhibiting sustained or recurrent signalling in the medulla may preferentially express β5t-independent TCRs⁶⁷.

Importantly, positive selection signalling outcomes depend on more than the TCR–pMHC interface alone, as coreceptor engagement with MHC is integral to how “self-reactive” a given clonotype is perceived. During coreceptor scanning, an antigen-engaged TCR must encounter an LCK-associated CD4 or CD8 molecule, such that the frequency of LCK-loaded coreceptors effectively sets dwell-time thresholds for productive signalling and tunes selection thresholds to self-ligands⁶⁸. Consistent with this, modulation of the MHCI–CD8 interaction can reorder the agonist hierarchy among peptides for the same TCR, demonstrating that coreceptor engagement can reshape functional sensitivity independently of peptide sequence⁶⁹. Despite this apparent complexity, constraining the temporal window during which these transient self-pMHC signals can occur, or reducing TCR signal strength through inhibition of ZAP70, markedly diminishes the efficiency of positive selection⁷⁰. Importantly, however, thymocytes that successfully undergo positive selection display comparable levels of Nur77-GFP reporter expression regardless of such perturbations, indicating that a defined signalling threshold must be reached to complete selection⁷⁰. Regardless of the threshold set, post-selection thymocytes exhibit a broad spectrum of CD5 and Nur77-GFP expression, and transcriptional and epigenetic profiling suggests that the gene expression programmes characteristic of peripheral naive T cells may already be established during thymic development^71,72.

Maintaining the pitch: peripheral calibration of naive T cells by self-pMHC

Once selected in the thymus, naive T cells enter the circulation and begin a lifelong surveillance programme in which they continuously scan self-pMHC complexes displayed in secondary lymphoid organs (Fig. 2a). Within these tissues, self-peptide is presented not only by professional APCs, including DCs, but also by a variety of non-haematopoietic stromal cells, where each cell type contributes differently to the spatial and qualitative landscape of self-peptide presentation^73,74. In secondary lymphoid organs, MHCI-restricted self-peptides are largely generated by the constitutive proteasome, although professional APCs such as DCs also express immunoproteasomes even at steady state, with broader induction occurring under inflammatory conditions⁷⁵. Self-peptide interactions support lymph node retention and trafficking. It is estimated that naive CD4⁺ T cells scan around 150–300 DCs during each transit through a lymph node, and in the absence of MHCII, lymph node transit time is reduced from 12 to 7.5 hours⁷⁶. Whether the frequency of self-pMHC contacts, beyond the strength of individual interactions, affects surrogate self-reactivity marker expression (Box 1) or gene expression programmes in T cells remains an open question.

Figure 2. Transcriptional heterogeneity among naive T cells is established during thymic selection and maintained by peripheral interactions with self.

(a) Upon expression of an ab T cell receptor (TCR) at the double positive (DP) stage of T cell development, thymocytes integrate a broad spectrum of sub-threshold signals through the TCR over a range of interaction frequencies, leading to their positive selection into single positive (SP) cells. Mature SP thymocytes subsequently leave the thymus and enter the peripheral naive T cell repertoire. As they circulate through secondary lymphoid organs (SLOs) but not while in the blood, naive T cells experience ongoing self peptide–MHC (self-pMHC) interactions, which maintain their transcriptional cell state and provide survival signals. Removal of self-pMHC tonic signals leads to transcriptional changes among naive T cells and also reveals differences between T cells that have high or low self-reactivity, which are likely epigenetically imprinted. (b) Naive T cells are heterogeneous at the transcriptional level, and at least some of this diversity is a result of differences in tonic sub-threshold self-pMHC signal strength (see also Box 1 and 3). Here we illustrate a model suggesting that this transcriptional heterogeneity impacts both how “poised” a naive T cell is to respond to cognate antigen stimulation (dimension-2) while at the same time being enriched for genes involved in negative regulation of T cell activation (dimension-1), perhaps to ensure that they do not become overtly activated by self-ligands.

Importantly, the loss of self-peptide signalling, as demonstrated experimentally through the depletion of TCR or MHC expression or modulation of proximal TCR signalling, leads to a gradual erosion in naive T cell numbers with a half-life of 20–40 days^77–81. In lymphoreplete environments, self-peptides prevent attrition, whereas under lymphopenic conditions, increased access to self-pMHC can trigger IL-7-dependent homeostatic proliferation⁸². Ultimately, sub-threshold engagements with self-peptide support not only the survival of individual clones but also sustain a diverse repertoire through intraclonal competition for self-pMHC^83,84. Importantly, beyond these roles in homeostasis and trafficking, the nature and intensity of self-pMHC recognition by naive T cells dynamically calibrates their signalling machinery, establishing a set point that shapes both the threshold and quality of future responses to foreign antigen. As in the thymus, self-peptides in the periphery thus serve not merely as neutral ligands but as active participants in the tuning of signal thresholds and in shaping the structure of the TCR repertoire.

Peripheral self-peptide availability and naive T cell responsiveness

T cells that are acutely deprived of self-pMHC rapidly lose responsiveness to activation with cognate antigen. Within minutes, they exhibit TCR ζ-chain dephosphorylation, disruption of lipid raft localization and blunted responses to previously effective doses of foreign pMHC^85,86. This change in T cell readiness is apparent even when comparing cells from circulation (where they are transiently away from self-pMHC signals) to cells from secondary lymphoid organs^76,86 (Fig. 2a). These findings support a model in which continuous low-level self-recognition maintains a signalling-competent state through recruitment of key kinases such as LCK and ZAP70 as well as regulation of inhibitory phosphatases like SHP1⁸⁶. Over the longer-term (a day or more), removal from tonic TCR signals eventually leads to changes in transcriptional state as well, revealing the extent to which variation in gene expression between T cells with differences in self-reactivity is actively modulated by ongoing self-pMHC interactions versus stably imprinted in the thymus⁷¹ (Fig. 2a). Thus, self-pMHC interactions in the periphery not only preserve the size and diversity of the naive T cell pool but also ensure that its members remain functionally poised to respond rapidly and effectively to pathogen challenge.

Experimental ablation of sub-threshold TCR signals has been used to determine the impact of complete loss of interactions with the self-peptidome on naive T cell function, but more subtle physiological changes in the self-immunopeptidome might also affect naive T cell responsiveness. For instance, infection can cause a shift in peptide processing machinery and gene expression, changing the peripheral self-peptidome^87,88. Infection also alters T cell scanning by downregulating chemokines important for DC positioning, ultimately dampening T cell priming to an unrelated pathogen⁸⁹. Furthermore, it is tempting to speculate that the spatial reorganization of the lymph node during infection may disrupt the ability of T cells to engage in sub-threshold homeostatic interactions with self-pMHC, impairing their responsiveness to subsequent infection. Similarly, secondary lymphoid organs show evidence of fibrosis, structural remodeling, and altered chemokine and cytokine niches with age, impairing naive T cell motility as well as T cell access to stromal and antigen-presenting niches where low-affinity self-pMHC interactions normally occur, potentially contributing to naive T cell hyporesponsiveness^90–94. Decoding how shifting self-peptide landscapes and lymphoid architecture recalibrate naive T cell sensitivity may unlock new strategies to preserve immune vigor during infection and with age.

Heterogeneity in self-reactivity

Collectively, polyclonal naive T cells experience a continuum of sub-threshold TCR signal strength as inferred from their broad distribution of CD5 and Nur77 (Box 1), whereas individually, each cell experiences a narrower range based on their degree of self-reactivity (Box 1 and 2). Where each T cell sits on the continuum may influence their set point or reactivity to foreign antigen, as well as their effector function (Box 3). Transcriptomic profiling of naive CD4⁺ T cells, sorted based on CD5, Nur77-GFP and Ly6C expression, suggests that highly self-reactive cells exhibit gene signatures that share features with bona fide activated T cells^71,95,96, potentially rendering them more “poised” to respond. Some of these features, including accessibility at regulatory genes such as Eomes, Ikzf2 and Id2, appear to be imprinted during thymic development and persist independently of ongoing self-pMHC engagement⁷¹ (Fig. 2a). At the same time, highly self-reactive cells also show upregulation of negative regulators of TCR signalling (such as Ctla4, Pdcd1, Lag3 and Cd200) alongside epigenetic remodeling at exhaustion-associated loci like Tox and Havcr2 (which encodes TIM3)^96,97. Markers such as FolR4 and CD73, associated with T cell hyporesponsiveness, are also elevated⁹⁸. Indeed, the loss of the negative regulator E3 ligase CBL-B reverses the diminished responsiveness of Nur77-GFP^hi cells⁹⁹, suggesting that highly self-reactive cells are tuned in their responsiveness depending on the expression of other molecules impacting TCR signalling. Collectively, these findings support a model in which strong self-reactivity initiates both activation-prone and tolerance-associated gene programmes within the same cell, raising the possibility that naive T cells on the higher end of the self-reactivity spectrum are simultaneously poised and restrained (Fig. 2b).

Box 2. T cell receptor determinants of peptide–MHC binding strength.

Cell-extrinsic processes, such as the stochastic nature of the interactions with specific self-peptides encountered during thymic development and the surveillance of secondary lymphoid organs, can lead to variations in sub-threshold T cell receptor (TCR) signals among individual naive T cells. Indeed, even two T cells with identical TCRs may differ in their sub-threshold TCR signal intensity and functional potential (Box 1 and 3). However, the structural and biophysical features of the unique TCR sequence expressed by a T cell plays a key deterministic role in its self-reactivity, although what the sequence features are and how the a and b chains jointly contribute is only beginning to be defined.

TCR sequence features of self-reactivity

Insights from TCR repertoire sequencing, in vitro binding studies and structural modelling have revealed factors that influence whether a given clonotype will occupy the high or low end of the self-reactivity spectrum. TCR sequence features arise from both germline-encoded V and J gene segments and the somatically generated complementarity-determining region 3 (CDR3) created during V(D)J recombination. The amino acid composition of the CDR3 plays a crucial role: hydrophobic, positively charged and cysteine residues within the CDR3, particularly in the β chain, are enriched in T cells exhibiting higher self-reactivity, likely enhancing peptide contacts and stabilizing interactions with self-peptide–MHC (self-pMHC)^145–149. CDR3 length is also associated with self-reactivity, with shorter CDR3 loops generally detected in cells with increased self-reactivity and cross-reactivity^145,150,151. Lastly, V and J gene usage contribute to self-reactivity in conventional mouse CD4⁺ T cells by providing germline-encoded contacts through CDR1 and CDR2 loops; certain V-J combinations are preferentially detected in mouse and human CD8⁺ T cell populations that receive strong TCR signals during development (e.g., CD8aa intestinal epithelial progenitors and regulatory T cells), and this has been suggested to be a consequence of early TCR selecting signals after TCRa recombination¹⁴⁸. With sufficiently large datasets of TCR sequences, biases can be detected in the T cell repertoire as a consequence of thymic positive selection when comparing TCRs present pre- and post-selection^145,152.

Relationship between self-pMHC and foreign-pMHC reactivity

Expression levels of surrogate self-reactivity markers such as CD5 and Nur77-GFP vary substantially across different MHC class I- and MHC class II-restricted TCR transgenic populations, supporting the idea that TCR specificity and self-reactivity are linked. In polyclonal CD4⁺ T cells, higher self-reactivity generally correlates with stronger binding to cognate foreign pMHC, suggesting that the most self-responsive clones are also the most sensitive to foreign ligands¹². Notably, however, studies of two Listeria monocytogenes-specific TCRs have shown that antigen receptors with identical specificity and similar affinities for target pMHC can nonetheless display markedly different CD5 levels¹⁵³. These observations indicate that, under certain circumstances (e.g., monoclonal TCR transgenic settings) foreign ligand affinity and self-reactivity can be experimentally uncoupled.

Box 3. Imprinting of functional biases by sub-threshold interactions with self-peptide–MHC.

In addition to influences from the cytokine milieu and lymphoid organ trafficking history, self-peptide reactivity contributes to transcriptional heterogeneity among naive T cells^{71,143,154–157}. This transcriptional heterogeneity may underlie some of the biases in effector differentiation that have been described, whereby markers of self-reactivity (Box 1) have been used to subdivide the naive T cell population and compare cell fate decisions that occur after encounter with foreign peptide–MHC (pMHC). These processes are not fully understood yet, with experimental conclusions in some instances that are in apparent contradiction, perhaps as a function of how self-reactivity is measured. However, taken as a whole, it appears that the functional heterogeneity at the population-level, driven by variations in self-reactivity at the individual cell-level, can impact cell fate decisions during effector differentiation — perhaps contributing to a robust and diverse functional response.

Regulatory T cells

One consistent observation that has been made using each of the three major surrogate markers of self-reactivity relates to regulatory T (T_reg) cell differentiation. Strikingly, CD5^hi, Ly6C^– and Nur77-GFP^hi naive CD4⁺ T cells all have a higher propensity than their lower self-pMHC affinity counterparts to express FOXP3 when stimulated under conditions that promote T_reg cell differentiation by naive CD4⁺ T cells^104,142,158. An interpretation of these data is that the most self-reactive naive CD4⁺ T cells may have escaped pathways leading to thymic central tolerance and thymic T_reg cell differentiation yet have an opportunity to contribute to tolerance through peripheral T_reg cell differentiation.

T follicular helper cells

CD5^hi cells and Nur77-GFP^hiLy6C^– naive CD4⁺ T cells with relatively high levels of self-reactivity express higher levels of the transcription factor Bcl6 and the receptors Cxcr5 and Pdcd1, along with lower expression of Prdm1 (BLIMP1)^71,96, suggestive of a gene expression profile that shares features with differentiated T follicular helper (T_FH) cells. Consistent with this, CD5^hi naive CD4⁺ T cells generated a higher percentage of CXCR5⁺PD1⁺ T_FH cells, compared with CD5^low cells, in response to infection with lymphocytic choriomeningitis virus (LCMV)⁷¹. This observation also fits with a model of early divergence between T_FH cells and non-T_FH cells being a function of TCR signal strength and IL-2 production, whereby strong TCR signals generate IL-2 producers poised for T_FH cell differentiation, and the IL-2 signals received by cells making weaker TCR interactions reinforces their non-T_FH cell differentiation¹⁵⁹. Indeed, greater IL-2 production is a hallmark of CD5^hi naive CD4⁺ T cells¹⁴. Recent work has also shown that the naive CD4⁺ T cell repertoire contains autoreactive T_FH-like cells that are CD5^hi and express PD1, BCL6 and EOMES, and which expand upon the ablation of T_reg cells¹⁶⁰. In contrast, Nur77-GFP^hi naive CD4⁺ T cells were shown to generate fewer T_FH cells than Nur77-GFP^low cells during LCMV infection, although this may have been affected by the sorted cells being transferred into lymphopenic recipients in which TCR signal strength is altered as a function of a lack of competitor T cells^84,161.

T_H1 and T_H2 cells

When CD5^low and CD5^hi naive CD4⁺ T cells from mice and humans are stimulated in vitro under T_H1 cell-polarizing conditions, CD5^hi cells generate fewer IFNg-expressing cells and less IFNg per cell, suggesting a bias against robust T_H1 cell responses by the most self-reactive cells^100,135,136, an affect that was recapitulated in vivo in mice¹⁰⁰. In line with this, co-cultures of congenically distinct CD5^low and CD5^hi naive mouse CD4⁺ T cells points to an IL-4 producing T_H2 cell bias in the CD5^hi subset, an effect that is attributed to differences in tonic mTORC1 signalling among CD5^low and CD5^hi naive CD4⁺ T cells¹³⁵. In contrast, recent single-cell RNA-sequencing studies identified a CD5^hi cluster of naive CD4⁺ T cells that undertakes a T_H1 cell-associated trajectory early after antigen challenge in vivo¹⁵⁵. Together, these findings point to potential biases that the strength of tonic TCR signalling can introduce into the differentiation decisions of naive CD4⁺ T cells.

IL-17-producing T cells

Relatively less is known about the effects of sub-threshold interactions with self-peptide on T_H17 cell responses. However, IL-17-producing CD8⁺ Tc17 cells have been analyzed in the context of self-reactivity. CD5^low naive CD8⁺ T cells have a higher propensity to differentiate into IL-17 producing Tc17 cells in vitro and pathogenic IFNg⁺IL-17⁺ effector cells in a transfer model of colitis¹⁶².

Memory T cells

Several studies suggest that self-reactivity can influence the recruitment of naive T cells into the memory pool. In mouse and human samples, memory CD4⁺ T cells express higher levels of CD5 than their naive T cell counterparts^12,136. In adoptive transfer experiments, CD5^hi polyclonal naive CD4⁺ T cells are preferentially recruited into the memory population following infection in mice¹². On the other hand, it has been suggested that CD5 decreases with memory differentiation on human CD8⁺ T cells¹⁶³, and murine LCMV-specific P14 TCR transgenic T cells with relatively lower levels of CD5 generated fewer CD127⁻KLRG1⁺ short-lived effector phenotype cells and more memory precursor CD127⁺KLRG1^– cells than the CD5^hi population¹⁴⁴. These studies support the possibility that a path from naive to memory can be influenced by the degree of self-reactivity exhibited by an individual cell.

Although the level of self-reactivity is broadly associated with increased functional potential, experimental findings reveal a more nuanced and sometimes contradictory relationship between self-peptide recognition and T cell responsiveness. Self-reactivity can act as a tuning mechanism, enhancing the responsiveness of naive T cells to foreign antigen. CD5^hi cells show greater ERK phosphorylation and IL-2 production after stimulation and exhibit a competitive advantage over CD5^low cells in response to infection^12,13,100. Consistent with this idea, recent thymic emigrants, which exhibit increased sensitivity to TCR engagement with self-pMHC (and greater CD5 expression), demonstrate enhanced TCR signal transduction and expand more robustly in response to low-affinity antigens compared to their mature counterparts^101,102. In contrast, other studies suggest that heightened self-reactivity induces functional restraint. T cells with low self-reactivity (e.g., CD5^low or Nur77-GFP^low Ly6C⁺) show stronger calcium flux, cytokine responses upon stimulation, and preferentially expand in competitive settings^99,103,104. Together, these findings underscore the complexity of sub-threshold interactions with self-peptides, which may both prime and restrain T cells depending on developmental context, signal intensity and environmental cues. Future work is needed to resolve when strong self-reactivity promotes responsiveness versus tolerance. Reported differences in how self-reactivity shapes T cell function may reflect differences in the marker of self-reactivity used or a distinction between population-level trends and the behaviour of individual TCRs. While markers of self-reactivity generally correlate with functional potential, this relationship reflects an average across the repertoire rather than a strict rule at the level of individual clones, which may help to explain divergent findings in TCR transgenic models in both the context of responsiveness and effector function.

Striking the right chord: cross-reactivity and self–non-self discrimination

Evidence from both the thymus and the periphery shows that the classic binary model, in which T cells remain off until encountering cognate foreign antigen and then switch on, is incorrect. Interactions with self-pMHC are not avoided; instead, they are essential for T cells to function properly. If T cells react to both self and foreign peptides, even with different affinities, this begs the question: just how different are self and foreign antigens really? It is still unclear whether this is solely a function of the context in which pMHC is presented (i.e., together with damage-associated or pathogen-associated molecules) or whether the TCR sequence plays a deterministic role independent of antigen presentation context. The latter seems likely, given that even during infection or injury, T cells generally do not become activated by the self-pMHC they recognize. Importantly, our understanding of self–non-self discrimination has been limited by a lack of data on TCR sequences paired with the self and non-self peptides they recognize. Indeed, a better definition of whether — and how — self and non-self peptides might differ from or resemble each other may enable us to determine to what extent there are self or non-self peptide intrinsic features that are key to TCR-based discrimination. While it is clear that any given TCR is able to bind many pMHCs, mapping this cross-reactivity remains a formidable challenge: there are, for instance, 20⁹ (512 billion) possible 9-mer amino acid sequences, even if not all of these are presented by MHC.

T cell cross-reactivity

Cross-reactivity is a fundamental and well-studied property of TCRs that is central to explaining how the same TCR can engage both self and foreign peptides across a wide range of affinities. However, a fundamental challenge remains to identify which pMHCs a given TCR can recognize¹⁰⁵. Current information on TCR–pMHC interactions comes from screens that either explore “neighbourhoods” of a given index peptide that is mutated at different positions (summarized in ref.¹⁰⁶), or where a larger set of the peptide landscape is generated against which T cells are queried^107–109. Such screens have shown what might be expected from the structural flexibility of both the TCR and the pMHC: while the peptides recognized may share specific structural similarities, they do not always appear “related” when compared by standard amino acid sequence similarity metrics¹⁰⁷. Thus, a simple model in which each TCR has a single “optimal” peptide that it binds most strongly, with variants of this peptide binding progressively more weakly as their sequences become more “distant” from the optimal peptide, is unlikely to be correct (Fig. 3a). Instead, the landscape of peptides as viewed through the TCR-recognition lens, is likely to be “rugged” rather than “smooth”, meaning any given TCR has multiple “optimal” peptides it is able to recognize with a similar binding strength^108–111 (Fig. 3a). This polyspecificity of TCRs has, however, been challenging to fully define, given the vastness of the possible peptide space. Taken together, experiments identifying peptides recognized by a given TCR have shown that many peptide sequences for which a TCR is cross-reactive can cluster in groups of peptides that differ by only a few amino acids^106,107. However, even a single, seemingly innocuous point mutation in a T cell epitope (e.g., substituting a leucine for an isoleucine) can sometimes completely abrogate TCR recognition^106,112. In other cases, the same TCR can recognize two different peptides that have no single amino acid in common¹⁰⁷.

Figure 3. Models of T cell receptor (TCR) cross-reactivity and the peptide landscape recognized by individual TCRs.

(a) Illustration of different models to explain the relationship between a TCR and the set of peptides it recognizes. In a smooth peptide landscape, there is one optimal peptide (the one with the greatest binding strength) with related peptide variants binding with reduced strength as a function of the distance to the optimal peptide. In a rugged peptide landscape, TCRs bind multiple peptides optimally, some of which may not be related in amino acid sequence or other biochemical parameters of the peptide. (b) It remains unknown whether there are decipherable features that distinguish self from non-self peptides that would separate them in shape space. It is possible that the identification of peptide distance metrics would uncover previously unappreciated differences between self and non-self peptides such that they form separate clusters or some cluster substructure. Alternatively, there may be no way to distinguish self from non-self by peptide sequence even when large enough experimental datasets of both self and non-self peptides are generated for a given MHC allele.

Recent years have seen enormous efforts to generate big datasets of TCR–epitope interactions, largely geared towards training artificial intelligence (AI) models to predict which TCRs can recognize a given pMHC and, vice versa, which pMHCs can be recognized by a given TCR¹⁰⁵. Such datasets are also required to define meaningful, structure-aware “distance functions” between peptides that would allow us to predict the likely extent of cross-reactive T cell immunity between a given peptide pair. Currently, simple sequence-base metrics like Hamming distance or edit distance — possibly enriched with information about physiochemical amino acid properties — are often used for this purpose^113–116. However, such methods implicitly assume that all peptides recognized by the same TCR are closely related to each other, which, as argued above, is improbable; the real “peptide distance function” is likely too complex to be well represented by simple sequence-based methods, especially given that the TCR can engage with pMHC complexes in different geometrical arrangements and that the pMHC complex itself is conformationally adaptable^117,118. Large-scale sequence datasets, perhaps augmented by experimental and predicted structures or other biochemical features of sequences of amino acids, may lead to the successful training of AI-based peptide distance metrics. These new metrics could lead to unexpected insights into whether the TCR is able to distinguish self from non-self based on peptide sequence alone or whether this is imposed by other signals.

Self–non-self peptide dictionaries

While data on foreign peptides paired with the TCRs that recognize them remains sparse across MHC alleles^105,119, our knowledge of the self-peptide “dictionary” on which T cells are trained — and later encounter in the periphery — is even more limited. The self-immunopeptidome is highly individual in nature, shaped by the MHC alleles expressed (of which there are thousands in the human population¹²⁰) and polymorphisms in proteins encoded in our genomes, as well as being distinct between cell types as a function of gene expression patterns and the peptide processing machinery available in a given cell¹²¹. Moreover, only a limited number of self-peptide ligands have been identified for specific TCRs that are important for T cell development and naive T cell homeostasis^59–61. Thus, given the incomplete characterization of the self-peptide landscape, theoretical and computational models have become essential tools for conceptualizing how T cells achieve self–non-self discrimination.

At the two ends of a conceptual spectrum lie what we might term the “far non-self” and the “near non-self” hypotheses. In one model, self and non-self are sufficiently distinct for T cells to distinguish them after “learning” to identify specific patterns characteristic of self-peptides (Fig. 3b). This far non-self model can be understood by analogy to language classification: a monolingual child who speaks English (self) has no difficulty identifying French words as non-English (non-self) because of differences in the frequency, distribution and order of letters in English compared to French words (the orthographic regularities of a language)¹²². Importantly, children accurately classify many words (but not all — some words may appear in both languages or seem English although they are actually French) as native or foreign, even if they know only a fraction of the English dictionary by a process called generalization. Similarly, computer algorithms trained on large enough data sets can robustly discriminate between languages based on just a few letters of text, recognizing that letter combinations such as “the” or “ght” are likely to be English¹²³. In the far non-self model, the prediction would be that immunogenicity increases with distance from self, with the most immunogenic peptides being the most different from self. In contrast, the near non-self model postulates that self and non-self are indistinguishable based on amino acid sequence, structure or other parameters relevant to TCR binding (Fig. 3b). Using the language analogy, this might correspond to comparing real English words to pseudo-English ones, which are not real words but follow the same general orthographic rules (words like “guidation” or “anachrone”). In this model, selecting T cells that recognize self-peptides weakly may predispose them to react strongly to altered versions of the same strings. The most immunogenic peptides would therefore be the ones that are just far enough away from self to escape tolerance mechanisms, whereas peptides that are very far away would be less immunogenic (this has also been referred to as the “shell model”)¹²⁴.

One approach to determine whether there are features that distinguish self-peptides from non-self-peptides has been to computationally process the human proteome into 9-mers (for MHCI presentation) and compare these with 9-mers generated from common pathogens, aiming to identify features that might distinguish self and non-self. Such analyses have suggested that there are no major differences in amino acid usage, nor are there any characteristic short amino acid combinations found in only self or non-self peptides^9,124. Thus, entirely proteomic-based comparisons seem to favour the near non-self hypothesis as every non-self peptide is only a few amino acid substitutions away from a self peptide¹²⁴, rendering it unlikely that self and non-self peptides are clearly separated from each other in distinct regions of the peptide “shape space” (Fig. 3b). However, these in silico peptide comparisons do not necessarily consider which peptides are processed for MHC loading across cell types. These computational approaches also do not yet consider structural parameters relevant to TCR binding, and we are not yet able to predict peptide recognition based on antigen receptor sequence alone¹⁰⁵.

How could reliable self–non-self discrimination work if self and non-self are not that different? It is conceivable that in comparing data of actual self-peptides eluted from MHC presented in the thymus (rather than computationally generated ones), we would find that they differ in a systematic, previously unappreciated way from pathogen-derived peptides presented by MHC during infection, perhaps not to the degree that self and non-self cluster entirely separately, but with some detectable differences (Fig. 3b). Large enough datasets of presented foreign and self-peptides coupled with machine learning (ML) approaches could, for instance, reveal novel parameters by which some separation between self and non-self can be achieved — along the lines of recent ML-based methods to compress high-dimensional cytokine dynamics of activated T cells into a low-dimensional latent space to separate T cells with the same TCR by their affinity for the antigen they were activated by¹²⁵. Indeed, simulations have suggested that non-random presentation of self-peptides in the thymus during negative selection could improve self–non-self discrimination⁹. Whether a unique cTEC-specific immunopeptidome might similarly provide an optimized self-peptide training set for self–non-self generalization during positive selection has not yet been considered.

Summary and future perspectives

The concept of self-pMHC recognition by T cells as a continuous, instructive process is reframing fundamental aspects of T cell biology. Understanding why thymocytes integrate self-pMHC signals during positive selection to calibrate TCR signal intensity and establish functional biases in naive T cells will be important to unraveling the logic of T cell repertoire formation. One conceptual possibility, akin to bet-hedging strategies in evolutionary biology, is that ensuring a high degree of repertoire heterogeneity and a spectrum of functionally distinct ground states represent a way to balance responsiveness and safety in an unpredictable microbial world. Mapping the repertoire of self-peptides presented across cells is fundamental to decoding how T cells develop, respond to foreign antigens and maintain tolerance. Illuminating the self-peptide dark matter has only recently become feasible, driven by advances in low-input sample preparation from primary cells, increased sensitivity of mass spectrometry instrumentation, and improvements in data analysis software^126–128. This will enable characterization of the features of the peptides that shape T cell fate, clarifying whether the peptides that promote naive T cell survival and responsiveness in the periphery are indeed those encountered during thymic selection, a long-standing assumption that may not hold if these two self-peptidomes are non-overlapping. Indeed, it remains unclear whether the specialized proteasomes expressed in the thymus, which co-evolved with the machinery responsible for antigen receptor diversity^129,130, produce a peptidome that selects a T cell repertoire better able to distinguish self from non-self. Establishing reference self-peptidomes across MHC alleles will also provide a framework for assessing whether self and non-self peptides can be distinguished within a parameter space and whether the immunogenicity of a peptide can be predicted as a function of its similarity or difference to the self-peptide or self-peptides recognized by a given T cell. Knowledge of the self-peptidome coupled with large-scale screens and AI-based analysis tools may ultimately make it possible to link a TCR sequence to both the self-peptide(s) and the cognate foreign antigen(s) it recognizes. Together, resolving these open questions will not only advance basic immunology but also inform our understanding of disease susceptibility, as well as strategies for personalized, “self-aware” vaccine design.

Glossary Box

Altered ligand hypothesis (also known as the peptide-switch hypothesis)

A model proposing that T cells need to be selected on different peptides in the thymic cortex and medulla to ensure that not all positively selected T cells are subsequently negatively selected

Immunopeptidome

The set of peptides derived from self-proteins that are presented by the MHC alleles present in an individual

MHC restriction

The principle that a T cell can only recognize and respond to peptides when they are presented by specific MHC alleles present in an individual

Negative selection

The process by which immature thymocytes that bind self-pMHC with high affinity are removed from the T cell repertoire via apoptosis

Peripheral tolerance

Mechanisms outside of the thymus that eliminate or render overtly self-reactive mature T cells less responsive to TCR stimuli through various mechanisms, such as suppression by Treg cells or induction of a transcriptional program that causes T cell anergy

Positive selection

Low to moderate affinity interactions (below the negative selection threshold) between immature CD4+CD8+ double positive thymocytes and self-pMHC that promote survival and differentiation

Sub-threshold interactions

TCR-pMHC interactions that are of insufficient affinity or duration to trigger an activation program

Two-dimensional affinity

An effective measure of binding when the TCR and pMHC complexes are anchored on opposing cell membranes. Because diffusion and orientation are more restricted in a membrane-anchored context, 2-dimensional affinity can differ from 3-dimensional affinity measured using soluble proteins