During late stages of thymic development, T cells must chose between different fates, dictated by their TCR specificity. Typically, this is thought of as a choice between three alternatives (being positive selection for useful T cells vs negative selection or neglect for harmful or useless T cells). However, there is growing evidence for a fourth alternative, in which T cells are positively selected by agonist ligands, which would normally be expected to induce T cell deletion. In this review, we will discuss where and when agonist selection is induced and whether this should be considered as a novel form of thymic selection or as an alternative differentiation state for Ag-exposed T cells.
Developing T cells face a series of checkpoints designed to determine their fitness for the lineage. The final selection step, however, does not simply test whether the right proteins are expressed at the right time, it monitors the “suitability” of the T cell for survival and utility in that particular individual. This scrutiny, which occurs as cells expressing an αβTCR reach the CD4+CD8+ double-positive (DP)3 stage, is usually discussed in the context of three potential fates: A DP thymocyte can undergo positive selection and become a mature T cell, undergo negative selection in which it is actively instructed to die, or suffer neglect, being the default passive cell death pathway. Which path a T cell follows depends on its Ag receptor specificity. The standard model is that T cells expressing no TCR, or a TCR with negligible affinity for self-MHC molecules, are doomed to die via neglect. A “medium” affinity interaction with self-MHC identifies T cells that may be useful, having a receptor that can work with the particular MHC alleles expressed by that individual, but not be overtly autoaggressive. T cells bearing a TCR with high affinity for self-peptide/MHC ligands (sufficient to activate a mature T cell, and hence classed as an agonist ligand) are thought to be eliminated through negative selection. Any T cell fortunate enough to survive this process via positive selection would then be induced to commit to a lineage, becoming either CD8 or CD4 single positive (SP).
Negative selection itself is typically pictured to occur via clonal deletion at the DP stage. However, this appealingly simply story has long been insufficient to explain all the data. Thus, some models for negative selection argue that deletion occurs at the DP stage, while others suggest deletion happens at the SP stage. Other models show that autoreactive DP thymocytes may not be eliminated immediately, but have an opportunity to edit their TCRs to avoid the death signal. Clonal anergy rather than deletion has also been proposed as a product of negative selection.
In all of these models, the autoreactive population is not imagined to undergo positive selection, rather it is dealt with through alternative negative selection mechanisms. Also, self-reactive T cells in these models are functionally impaired, either because they have lost their autoreactive TCR (editing), have become unable to respond (anergy), or are dead (being the ultimate inactivation).
There has been growing evidence, however, for positive selection of T cells by agonist ligand, in which cells with overtly self-reactive TCRs are directed into a mature T cell lineage. For simplicity and consistency, this type of agonist-mediated positive selection will be referred to as “agonist selection” for the remainder of this review. These cells display unusual phenotypic and functional properties, which appear to allow them to regulate the immune response. Several TCRαβ T cell populations have been proposed to be generated via agonist selection; including NK-T cells, the CD8αα subset of intestinal intraepithelial lymphocytes (IEL), and CD4+CD25+ regulatory T cells (Treg) (Fig. 1⇓). These individual cell populations have been the topic of recent excellent reviews, to which we would refer the reader (1, 2, 3, 4, 5). In addition, certain TCRγδ T cells may also be selected by agonist interactions, and recent data suggest this may occur in the thymus (6). However, the focus of this review will be on how agonist selected TCRαβ T cells are generated, with a special focus on whether agonist selection represents a distinct developmental lineage chosen during T cell maturation (the “fourth way” of the title), or whether these cells represent a differentiation state that can be reached by both developing and mature T cells at multiple stages (Fig. 2⇓).
Where are agonist selected T cells produced?
This question has been one of the most contentious in the field. Starting with the dogma that a DP thymocytes encounter with agonist self-peptides results in clonal deletion has led to the prediction that agonist-selected cells must be produced extrathymically, and some reports appeared to support this model. However, in more recent years there is increasing evidence that agonist-selected cells can, and perhaps typically do, arise in the thymus.
Compared with the other agonist-selected populations, the site for NK-T development has long been proposed to involve the thymus. Even here, however, the developmental steps are quite distinct from mainstream T cells. When speaking of NK-T cells, we will mostly refer to CD1d-restricted NK-T population bearing the invariant Vα14 rearrangement (also called iVα14 NK-T cells). The first surprise in NK-T development is their absolute requirement to encounter CD1d on bone marrow-derived cells in the thymus, most likely the immature thymocytes themselves (2). While recent articles have shown that CD1d is expressed on cortical epithelial cells, this is not sufficient for selection of NK-T cells, even when overexpressed (7, 8). It is believed that CD1d is associated with (undefined) endogenous glycolipid(s) that constitute the agonist ligand driving agonist selection. The first population of T cells expressing receptors specific for the CD1d/α-GalCer (used to identify iVα14) cells appears in the thymus as CD8− (either CD4+8− or CD4−8−) mature T cells. These cells initially lack CD44 and NK1.1, but acquire both markers further in development (9, 10). The thymus is required for NK-T development, and this population appears to derive from conventional DP precursors, as shown by transfer experiments (10). Although deriving from DP thymocytes, the selected population of NK-T cells (at least in mice) lacks expression of CD8 (being either CD4+8− or CD4−8−), and forced expression of CD8 appears to abort NK-T maturation (11). This fits with a model in which agonist selection on CD1d crosses the line into negative selection with persistent CD8 expression.
The CD8αα T cells that are present in the gut represent another agonist-selected population. This population expresses forbidden TCR Vβs in strains of mice where endogenous superantigens led to deletion of conventional T cells (12), but these cells have been more extensively studied using TCR transgenic models for agonist selection. CD8αα IEL are especially intriguing because, unlike NK-T cells, they are suggested to be an alternative fate for “conventional” T cells (1). Thus, in several TCR transgenic model systems, where exposure to agonist self-ligand led to deletion of conventional T cells, CD8αα T cells in the gut were expanded (13, 14, 15, 16, 17). Accumulating data suggests that CD8αα T cells arise in the thymus, rather than in an extrathymic pathway as was previously proposed (16, 18, 19). In fact, CD8αα T cells arise in fetal thymic organ cultures when a high-affinity agonist is present (20, 21, 22, 23). It is not currently clear where the agonist ligand needs to be expressed to drive thymic CD8αα IEL selection, although cortical epithelium is evidently not sufficient (24). However, it is still unclear whether CD8αα T cells derive from double-negative (DN) or DP progenitors. On the side of the DN is data from models where CD8αα IEL develop in animals that have very few DP progenitors (e.g., in the H-Y TCR transgenic male mouse). Using the same model, it was found that expression of CD8β, which is required for positive selection of most class I-restricted T cells, is dispensable for agonist-selected IEL (16). Also, CD8αα IEL express the FcεR1γ chain in place of (or in addition to) the TCRζ chain in the CD3 complex, a feature in common with DN thymocytes (1, 25). Finally, the gene expression profile of CD8αα IEL is similar to γδ IEL, which develop from DN progenitors, and is quite distinct from the CD8αβ IEL (26). In contrast, Eberl and Littman recently showed that CD8αα IEL express a reporter gene that is activated only after β selection, suggesting that this population derives from at least a stage IV DN progenitor (19). In addition, CD8αα and DN T cells can be generated directly from TCR transgenic DP progenitors by interaction with agonist-bearing stromal cells, at least in reaggregate cultures (27, 28). As these cells develop, they lose their expression of CD4 and CD8β (regardless of the class specificity of the TCR) and, in some systems, commit to expression of CD8αα (28). H-Y TCR transgenic T cells driven in this way express receptors typically found on NK cells and acquire rapid functional reactivity, suggesting that their role might be like “innate immune cells” (28). However, because normal T cells also express these molecules and properties after antigenic stimulation (29, 30, 31, 32), it is not clear that this is necessarily a developmental program per se (see below). Lastly, it should be noted that the CD8αα T cells generated in TCR transgenic models may not be the most appropriate method to study CD8αα IEL, given that TCRαβ transgenes are often expressed at the DN stage rather than the typical DP stage. Indeed, preliminary data using transgenic mice that express the H-Y-specific TCR at the DP stage (rather than at the DN stage, which is characteristic of conventional TCR transgenic mice) shows that exposure to agonist ligands (in the form of male Ag) fails to produce a CD8αα IEL population (T.A.B. and K.A.H., unpublished observations). Using such models may allow a better definition of when the window for agonist selection is open during T cell maturation. Additional studies on the repertoire and function of the CD8αα IEL that exist in normal mice is also needed, and in the future it may be useful to create transgenic models using the specific TCR that are normally expressed in this population to understand their developmental and functional properties.
CD25+CD4+ T cells (Treg) are another population selected for by high-affinity agonist self-ligands. Recent data from the Rudensky laboratory showed that retroviral expression of TCR cloned from CD25+CD4 Treg cells, but not from CD25−CD4 naive T cells, led to rapid expansion in lymphopenic hosts, directly demonstrating the “self-reactivity” of normal Treg (33). As for CD8αα IEL, conventional TCR transgenics, particularly the influenza-reactive 6.5 (also called TS1) strain, has been the predominant tool used to study thymic development of Treg cells. Treg can clearly be generated either during thymic development (34, 35, 36), or through certain forms of Ag encounter in the periphery (37, 38). For CD4+CD25+ Treg populations selected in the thymus, Ag expressed on radio-resistant cells including thymic epithelium appears to be most efficient (34, 36, 39), and although Treg selection on bone marrow-derived cells is observed, this seems to give rise preferentially to CD4+CD25− cells (34). This raises an important point that there are different classes of Treg cells, characterized by different phenotypes (CD4+CD25+, CD4+CD25−, and CD8 populations) and functional mechanisms. While we are focusing on CD4+CD25+ Treg populations here, the selection of other regulatory T cell pools may have distinct developmental requirements.
In addition to development in the thymus, CD4+CD25+ Treg cells can be generated from mature naive CD4 T cells, both in vitro and in vivo (37, 38, 40). These cells appear indistinguishable from thymically selected Treg populations, including their gene expression profile (H. von Boehmer, personal communication), with the important implication that commitment to the Treg pool is not necessarily a developmental lineage, but can represent a differentiation state (Fig. 2⇑). Interestingly, the repertoire of TCR specificities expressed by CD25+CD4 T reg in normal mice was largely different from that of the naive CD25−CD4+ T cells, but did show some overlap, suggesting that both modes of development (developmental and differentiative) could operate in vivo (33).
Can “conventional” T cells be selected with agonists?
The capacity of normal CD4 or CD8αβ T cells to undergo positive selection on agonist ligands has been addressed by many groups. In some model systems, agonist ligands delivered at low doses appeared to induce positive selection of CD8 T cells, supporting an avidity model of thymic selection (in which the same ligand could drive positive and negative selection depending on dose or density) (41, 42). However, like the populations described above, these cells appeared to be altered in their development. With increasing agonist strength used for selection, there appeared to be decreased expression of CD8 overall, and CD8β in particular. Ultimately, these agonist-selected class I-restricted cells start resembling CD8αα populations (20, 21, 28). Perhaps more important, agonist-selected cells showed functional impairment, in that they could not respond to the original ligand (20, 21, 22, 43, 44, 45, 46). Hence, these studies fit well with the studies discussed above that imply that T cells selected in the presence of agonists are pushed into a phenotypically and functionally altered state.
However, there are some reports arguing that conventional T cells can be selected by agonist ligands. Mintern and colleagues used peptide/class I MHC tetramers expressing agonist ligands to drive thymic selection in fetal thymic organ culture (47). Their data suggested that conventional T cells, with normal CD8αβ expression and functional reactivity, could be produced in this manner. Germain and colleagues used two-step reaggregate cultures, the first step of which involved DP thymocytes cultured with agonist-bearing dendritic cells, and observed selection of CD4 T cells that appeared to be functional (48). In these cases, the exposure to agonist ligand was artificial (not presented by a cell) or interrupted, and hence it is unclear if this type of selection would occur in nature. Nevertheless, such data leave open the possibility that agonist-selected populations might sometimes appear within the conventional T cell pool.
Is agonist selection a form of negative selection, positive selection, or Ag reactivity?
While the appearance of all these populations in the thymus is correlated to their exposure to agonist ligands, there is an unresolved question as to whether these cells are instructed to develop by MHC/self-Ag recognition at a particular stage of development (which would be a form of positive selection), or whether they arise as a mechanism of avoiding or compensating from agonist stimulation (this would be a form of negative selection).
Situations in which conventional positive selection is impeded can still allow for selection of CD8αα IEL populations. For example, development of HY TCR transgenic cells in a female mouse requires expression of CD8β, yet selection of CD8αα IEL in the male does not (16). This has been proposed as an argument against agonist selection following positive selection. However, interpretation of these experiments is complicated by the fact that the selecting ligand is very different in the two environments, and hence “agonist positive selection” may not require normal CD8 engagement. Indeed, data discussed above in which T cells were selected with agonist ligands presented by peptide/MHC tetramers argued that CD8 engagement of the ligand was not needed (47). Furthermore, CD8α-deficient T cells can be positively selected into the CD8 lineage with weak agonist ligands (49, 50). Development of NK-T cells has also been shown to have unusual signaling requirements, compared with conventional T cells, in that NK-T development is highly dependent on Fyn, NIK, RelB, Ets-1, and Ikaros, while it does not require normal activity of the Erk pathway (2, 51, 52, 53). In contrast, NK-T development (like CD8αα IEL development (16)) seems to require the α-CPM motif of the TCRα-chain, which is characteristically required for positive but not negative selection (54). As in other aspects of agonist-selected cells then, the signaling requirements for development of these cells show features of both negative and positive selection.
A related issue regarding agonist-selected cells is whether their unusual phenotype allows them to survive agonist encounter, or whether agonist encounter drives their phenotype. In particular, these two alternatives speak to the issue of total signal strength dictated by the phenotype of the progenitor vs TCR affinity in the development of agonist-selected cells. One model would be that a small subset of developing T cells stochastically begins differentiating into an “agonist-selected” lineage, and these cells selectively survive when the agonist is expressed. For example, this subset of cells might be partially impaired in the quality or quantity of the signal they receive from an agonist ligand, allowing them to evade negative selection. This argument has been applied to the expression of CD8αα rather than CD8αβ on agonist-selected IEL populations, because CD8αα is a “weaker” coreceptor than CD8αβ. However, recent data showing a remarkably efficient production of agonist-selected CD8αα T cells (beginning with nonselected DP thymocytes) makes this less likely and suggests active instruction of T cells to adopt the phenotype(s) of such agonist-selected cells (28). Likewise, the expression of CD44 and NK1.1 on NK-T cells appears following their selective enrichment in the thymus upon agonist ligand engagement (9, 10). Nevertheless, while TCR affinity might initially dictate the developmental program of the T cell, changes in T cell expression of coreceptors, costimulators, or signaling molecules may still be key for emergence of the agonist-selected population.
A distinct question, though, is whether agonist selected cells are actually representative of a form of differentiated effector/memory cell. In this model (Fig. 2⇑), the T cells may have begun (or even completed) maturation before they begin responding to the agonist ligand. If agonist is expressed throughout the thymus, there may be a minimal amount of time in which T cells can mature before they begin an overt response to the same Ag. Hence, there may not be a clear preactivated intermediate detectable in the thymus. This model is not mutually exclusive with the idea of agonist selection as a developmental program—it may be that cells begin development under agonist instruction and then participate in a (limited and atypical) response against the agonist ligand. Again, the acquisition of NK1.1 on α-GalCer/CD1d-reactive NK-T cells is detected well after they initially appear as a selected population (9, 10). Hence the phenotype and function of agonist-selected cells may be dependent on repeated encounters with Ag. Rather than a developmental switch then, agonist selection may be more analogous to a chronic (albeit unusual) immune response.
In what state are agonist-selected T cells maintained?
The data discussed above argue that agonist-selected cells are phenotypically and functionally different from naive “normal” T cells, but distinct from each other: Treg cells express CD25 and are CD45RBlow; NK-T cells are NK1.1+ and CD44high, and CD8αα IEL T cells are also elevated for several NK lineage markers including CD94 (although typically not NK1.1). However, in these respects they exhibit markers that overlap with activated conventional T cells. Up-regulation of CD25 and CD44 and loss of CD45RB is typical of effector/memory cells, while NK1.1, CD94 and CD8αα have all been observed on stimulated T cells (31, 32, 55). This again raises the question of whether agonist-selected cells are actually representative of a distinct, committed lineage or are cells undergoing a form of chronic stimulation. This difference is relevant, because the “developmental lineage” model predicts that the DP thymocyte is able to interpret TCR signals in four separate ways.
Are agonist-selected cells maintained by chronic stimulation? Analysis of mice expressing β2-microglobulin (β2m) in the thymus alone suggested that CD8αα IEL populations did not persist if the Ag was not present in the gut. However, the interpretation of these data are complicated by the proposal that CD8αα IEL require engagement of both their TCR and CD8αα (with thymic leukemia Ag molecules expressed on gut epithelium) to develop and survive (1, 56). Because thymic leukemia Ag expression is also dependent on β2m, the role of sustained TCR engagement has not been fully addressed in this case. The capacity of NK-T cells to persist in the absence of their ligand, CD1d, has been shown in lymphopenic settings (57), but not in normal hosts. For Treg populations, initial data from in vitro assays indicated these cells were anergic. This idea must be revised in light of data suggesting CD4+CD25+ Treg cells proliferate vigorously to cognate peptide/MHC ligands in vivo (58). It appears a subset of Treg cells are nondividing and can be maintained long term in the absence of specific agonist, while maintaining expression of activation markers (59, 60). Such data argue against an absolute need for chronic agonist stimulation in Treg maintenance. Indeed, recently activated Treg populations may have a shorter life span (60). Hence it appears that continuous Ag may not be required for maintenance of the Treg phenotype, but might influence survival of these cells.
In summary, several subsets of TCRαβ T cells can be induced by agonist ligands during their development. However, current studies are still unclear about whether selection of these cells is confined to a particular stage in thymic development or whether agonist selection is better visualized as a differentiation state that can be reached by cells starting at various different points in their ontogeny. At present, production of NK-T cell development fits well in the developmental lineage model, but the situation is less clear for CD8αα IEL. Treg cells can clearly be induced by both thymic and extrathymic pathways, although it is not clear at which stage(s) of thymic development they can become committed.
Although studies using TCR transgenic models have greatly facilitated the analysis of agonist-selected subsets, interpretation of these may be complicated due to various nonphysiological aspects of such mice (including precocious TCRα-chain expression and vastly biased frequencies of specific cells). Refinement of these models and examination of polyclonal populations may help define the normal stages at which agonist selection or direction can operate. In particular, analysis of CD8αα IEL and Treg populations will be greatly helped by characterizing the TCRs expressed by polyclonal T cells that develop into these pools to determine the nature of their self reactivity and selection. However, although NK-T, CD8αα IEL, and Treg populations can all be induced through agonist interactions, it is not at all clear that they share a common developmental pathway. Just as functional and phenotypic properties distinguish these subsets, their differentiation might also follow distinct rules. With further study on the developmental requirements of these cells, and analysis of the gene expression profiles that agonist selection induces, it will be interesting to see whether common features emerge in these subsets. Or whether there are as many ways for T cells to deal with agonists as there are to skin a cat.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
↵1 This work was supported by the National Institutes of Health Grants AI39560 (to K.A.H.) and AI38903 (to S.C.J.).
↵2 Address correspondence and reprint requests to Dr. Stephen C. Jameson, Center for Immunology, Mayo Mail Code 334, 420 Delaware Street SE, Minneapolis, MN 55455. E-mail address:
↵3 Abbreviations used in this paper: DP, double positive; SP, single positive; IEL, intraepithelial lymphocyte; DN, double negative; Treg, regulatory T cell; β2m, β2-microglobulin.
- Received July 30, 2004.
- Accepted September 30, 2004.
- Copyright © 2004 by The American Association of Immunologists