RevIewS

Central CD4+ T cell tolerance: deletion versus regulatory T cell differentiation

Ludger Klein1*, Ellen A. Robey2 and Chyi-Song Hsieh3

Abstract | The diversion of MHC class II-restricted thymocytes into the regulatory T (Treg) cell lineage is driven by intrathymic encounter of agonist self-antigens in a similar manner to the clonal deletion of thymocytes. Somewhat paradoxically , it thus seems that the expression of an autoreactive T cell receptor is a shared characteristic of T cells that are subject to

clonal deletion and T cells that are diverted into the Treg cell lineage. Here, we discuss how thymocyte-intrinsic and thymocyte-extrinsic determinants may specify the choice between these two fundamentally different T cell fates.

Recessive tolerance This Review is dedicated to the memories of Bruno decisions of thymocytes, reductionist ‘signal strength’ Any mode of tolerance that Kyewski and Harald von Boehmer, who both made models do not explain phenomena such as the intra- depends on clonal deletion or 13,14 seminal contributions to our understanding of central clonal competition between Treg cell precursors , the anergy induction. The term tolerance and who passed away in 2018. overlapping TCR repertoires of conventional T cells and recessive refers to the fact that 15,16 in order to be effective, Treg cells or the concomitant clonal deletion that occurs essentially all self-reactive A series of landmark studies in the late 1980s confirmed in essentially all TCR-transgenic systems where neo-an- 17–20 T cells need to be physically the longstanding hypothesis that tolerance to self tigen expression drives efficient Treg cell diversion . eliminated or functionally invol­ves the clonal deletion in the thymus of T cells All of these observations argue against the existence of inactivated. with potentially dangerous, self-reactive T cell receptors sharply demarcated affinity and/or avidity thresholds 1,2 Dominant tolerance (TCRs) . Around the same time, the transplantation of that separate clonal deletion and Treg cell diversion. In Any mode of tolerance that thymic epithelium was reported to confer a mode of tol- this Review, we discuss how the cell fate choice between depends on the suppressive erance to donor-type tissues that could not be explained clonal deletion and diversion into the Treg cell lineage action of regulatory cells and by recessive tolerance mechanisms such as clonal dele- may be influenced by distinct modalities of intrathymic can therefore be transferred from tolerant to naive tion, as it was transferrable to naive animals through self-antigen expression and presentation, spatiotemporal individuals. the injection of lymphocytes and hence was referred aspects of thymocyte differentiation and contextual input to as dominant tolerance3–5. Several independent lines of that accompanies the agonist TCR signal. investigation subsequently illustrated the importance of dominant tolerance by showing that the elimination of Topology of self-antigen expression particular subsets of CD4+ T cells, all later shown to be the Self-antigens can be broadly categorized into those with a equivalent of forkhead box protein P3 (FOXP3)+ regula- ubiquitous expression pattern and those with an expression 1Institute for Immunology, Biomedical Center tory T (Treg) cells, unleashed a latent autoimmune poten- pattern that is confined to one or a few anatomical sites, 6–8 Munich (BMC), Ludwig tial in the normal T cell repertoire . It is believed that known as tissue-restricted antigens (TRAs). It is now widely

Maximilians Universität, approximately 80% of the Treg cell repertoire originates recognized that TRAs are also likely to be visible to the nas- Planegg-Martinsried, 9,10 from the thymus (known as thymus-derived Treg cells) , cent T cell repertoire owing to the phenomenon of promis- Germany. + where FOXP3 Treg cell differentiation can be instructed by cuous gene expression in medullary thymic epithelial cells 2 Division of Immunology and encounter of agonist self-antigen11,12. Thus, the recogni- (mTECs) (Box 1). Although it is undisputed that promiscu- Pathogenesis, Department of central tolerance Molecular and Cell Biology, tion of self during thymocyte differentiation can not only ous gene expression extends the scope of University of California generate holes in the T cell repertoire by clonal deletion to a much broader spectrum of self-antigens than previ- Berkeley, Berkeley, CA, USA. but also specify the diversion of cells into a self-reactive ously appreciated, insights are only beginning to emerge 3Department of Internal T cell lineage that lacks autoimmune potential and instead regarding whether and how distinct patterns of antigen Medicine, Division of confers tolerance through immune regulation. expression may specify the mechanism of central toler- Rheumatology, Washington The conceptually most simple models to explain these ance induction for CD4+ T cells and how this affects the University School of Medicine, St. Louis, MO, USA. markedly different outcomes of self-antigen recognition composition of the polyclonal TCR repertoire. MHC class II tetramer *e-mail: ludger.klein@ in the thymus are based on the affinity and/or avidity Two recent studies used technology + med.lmu.de of the underlying TCR–peptide–MHC (TCR–pMHC) to trace the fate of small cohorts of polyclonal CD4 T cells https://doi.org/10.1038/ interactions (Fig. 1). However, although there is little when their cognate antigens were expressed as neo-self-an- s41577-018-0083-6 doubt that TCR signals crucially impact the cell fate tigens under the control of various promoters to achieve

Nature Reviews | Immunology Reviews

a Affinity-based model b Avidity-based model Thymus Increasing TCR affinity (fixed Increasing cognate pMHC density cognate pMHC density) (fixed TCR affinity)

Thymocyte

Escape Treg cell induction Clonal deletion Escape and Treg cell induction Clonal deletion only Treg cell induction and clonal deletion

Low TCR Intermediate High TCR Low pMHC Intermediate High pMHC affinity TCR affinity affinity density pMHC density density

None None

Conventional Treg cells T cells Combined thymic output of different Thymic output for a given TCR affinity TCRs for a given pMHC density at different pMHC densities

Fig. 1 | Signal strength models of clonal deletion versus regulatory T cell differentiation. The term affinity describes the strength of a single interaction between a T cell receptor (TCR) and its peptide–MHC (pMHC) ligand, whereas the term avidity is proportional to the multiplicity of such interactions. a | According to an affinity-based model of thymocyte development, at a given fixed density of cognate pMHC in the thymus, T cells bearing the highest affinity TCRs are removed from the repertoire by clonal deletion, whereas T cells bearing TCRs of intermediate affinity are diverted into the

regulatory T (Treg) cell repertoire. T cells bearing TCRs of low affinity escape into the periphery as conventional T cells. Although this prediction has so far not been addressed in a polyclonal setting, the affinity-based model is supported by a

study showing that at a fixed level of neo-self-antigen expression, the efficiency of Treg cell selection was directly correlated with the reactivity of different TCRs119. b | An avidity-based model posits that in addition to the TCR affinity, the density of the pMHC ligand is also crucial. At a given fixed TCR affinity, increasing or decreasing the density of antigen presented on

thymic antigen-presenting cells will bias the cell fate towards either clonal deletion or Treg cell differentiation, respectively. Several experimental observations are consistent with this model. First, TCR-transgenic models have shown a clear correlation between antigen expression levels and the ensuing cell fate20,120. Second, the titrated administration of agonist peptide to TCR-transgenic fetal thymic organ cultures or through intravenous injection into adult mice revealed a positive

correlation between agonist concentration and the number of Treg cells; above a certain threshold, further increasing the peptide dose resulted in thymocyte deletion121,122. Third, at fixed levels of antigen expression, attenuating cognate antigen presentation through the knockdown of MHC class II expression resulted in decreased clonal deletion and increased 55 emergence of Treg cells . Finally , low-affinity or high-affinity peptide ligands for the same TCR can instruct Treg cell differentiation with equal efficiency provided that the density of the low-affinity ligand is correspondingly higher123.

21,22 −/− either a ubiquitous or a TRA-like expression pattern . revealed holes in the Treg cell repertoire of Aire mice, Ubiquitous antigen expression led to a large-scale reduc- and an intact Aire gene was crucial for the capacity of tion in the number of antigen-specific CD4+ T cells, which several TCRs of unknown specificity from naturally + was evident already in the nascent CD4 single-positive occurring Treg cells to promote intrathymic clonal diver- 23,24 (SP) compartment, indicating efficient clonal deletion. By sion into the Treg cell lineage . In further support of

contrast, under circumstances of TRA-like antigen expres- TRAs shaping the composition of the Treg cell repertoire, + +/+ sion, antigen-specific CD4 T cells were only mildly or not adoptively transferred neonatal Treg cells from Aire at all affected in terms of their absolute numbers yet har- mice, but not Aire−/− mice, prevented the autoimmune + 25 Thymus-derived Treg cells boured a substantial fraction of FOXP3 Treg cells. On this manifestations seen in AIRE-deficient mice . Cells that enter the regulatory basis, it was suggested that the mechanism of CD4+ T cell Although these findings establish that intrathymic T (Treg) cell lineage as a consequence of cell fate tolerance is determined by the antigen expression pattern, antigen expression in a TRA-like manner can specify specification during T cell whereby ubiquitous expression favours deletional toler- Treg cell diversion, they do not rule out that promiscuous maturation in the thymus. By ance, whereas TRA-like modalities of antigen expression gene expression also contributes to clonal deletion. Indeed, contrast, peripherally derived promote diversion into the Treg cell lineage. observations in several TCR-transgenic systems have T cells arise through reg Studies in mice lacking the autoimmune regulator protein linked promiscuous gene expression and AIRE expres- conversion of conventional −/− + 26–29 forkhead box protein P3 (AIRE) (Aire mice), which have decreased expression of sion to deletional CD4 T cell tolerance . Similarly, + (FOXP3)− CD4+ cells in many TRAs, are consistent with a role of TRAs in Treg cell tetramer-based studies showed that polyclonal CD4 secondary lymphoid organs. generation. For example, large-scale TCR sequencing T cells specific for an eye-restricted TRA were deleted

www.nature.com/nri Reviews

30 Tissue-restricted antigens in an AIRE-dependent manner . Expression of green Remarkably, despite relatively low levels of TCR expres- (TRAs). Self-antigens that are fluorescent protein (GFP) as a TRA-like neo-self-anti- sion on the surface of DP thymocytes, they are exqui- expressed by only one or a few gen under the control of the insulin promoter resulted sitely sensitive to antigen. In fact, whereas the estimated peripheral tissues. The term in a significant decrease in the number of polyclonal minimum number of pMHC complexes required TRA is an operational definition + based on available expression GFP-specific CD4 T cells coincident with the emergence to activate mature peripheral T cells is believed to be + catalogues, according to which of FOXP3 cells among the remainder of the GFP-specific approximately 100, as few as one or two agonist pMHC TRAs are expressed in less cells, which suggests that clonal deletion and diversion complexes seem to be sufficient to elicit clonal deletion than 5 tissues of 60 tested. 22 34,35 into the Treg cell lineage occurred in parallel . Of note, the at the DP thymocyte stage . Importantly, this implies intensity of tetramer staining on these T cells was lower that at the DP stage (and at least for high-affinity TCR– Central tolerance reg Any mechanism of tolerance than that on their naive counterparts in the absence of pMHC combinations), there is almost no possibility of that shapes the T cell receptor GFP, which suggests that T cell clones expressing TCRs a lower avidity interaction as a consequence of lower composition and functionality with high affinity for GFP may have been deleted. pMHC density that might favour Treg cell diversion. of the T cell repertoire during In sum, the propensity of ubiquitous and TRA-like Indeed, although it is well established that corti- selection in the thymus. modes of antigen expression to favour clonal deletion and cal antigen encounter by DP thymocytes can result in 36,37 MHC class II tetramer Treg cell development, respectively, may relate to quanti- clonal deletion , its contribution to Treg cell differ- A staining reagent used to tative differences in the amount of self-antigen, which entiation remains contentious. For example, FOXP3- detect antigen-specific CD4+ is in keeping with an avidity-based model. However, expressing cells are almost exclusively confined to the T cells. It consists of four as we discuss in the following sections, we also need to thymic medulla, a compartment that contains hardly recombinantly expressed biotinylated MHC class II consider the following factors: the different spatial dis- any DP cells. In addition, the delayed appearance of + molecules (each carrying a tribution of antigen expression and consequently the FOXP3 cells in ontogeny or in bone marrow chimaeras covalently attached peptide developmental stage at which these antigens are first seen suggests that entry into the Treg cell lineage requires thy- epitope of interest) that by a thymocyte; the likelihood and frequency of serial mocyte maturation beyond the DP stage and/or involves are tetramerized via a 38,39 fluorochrome-labelled antigen encounters; the antigen-presenting cell (APC) medulla-associated signals . Such correlations entail streptavidin core. types that display the respective self-antigen; and, related the caveat that the timing and intrathymic positioning to this, the co-signals such as cytokines or co-stimulatory of FOXP3 expression may be poor indicators of when Autoimmune regulator ­molecules that accompany the TCR signal. and where the underlying TCR stimulus occurred (as protein the transit time through the DP compartment may be (AIRE). AIRE was discovered through genetic mutation in T cell developmental stage shorter than the temporal delay between a TCR signal patients with the monogenically Intrathymic T cell development is characterized by a being received and FOXP3 expression being induced), inherited disease autoimmune highly ordered migration of thymocytes between distinct but it is clear that antigen encounter at the DP stage is polyendocrine syndrome 1. It is areas of the thymus31,32. Ubiquitously expressed antigens not a prerequisite for T cell differentiation. Instead, a transcriptional regulator that reg controls the expression of are expected to be visible to developing thymocytes in diversion into the Treg cell lineage can efficiently be initi- tissue-restricted antigens in the cortex as soon as these cells express a functionally ated at the SP stage, as shown by intrathymic transfer of medullary thymic epithelial cells rearranged TCR at the CD4+CD8+ double-positive (DP) antigen-naive CD4+ SP cells into an antigen-expressing (mTECs) but may also have stage. By contrast, owing to their confined expression in thymus40. These experiments also showed that the pro- similar or distinct functions in mTECs, TRAs are expected to be encountered only after pensity of CD4+ SP cells to enter the T cell lineage cells other than mTECs. reg CD4+ or CD8+ lineage-committed SP T cells have relo- inversely correlated with their progressive maturation cated to the medulla33. This raises the question of how — that is, more mature cells were less likely to become

the developmental stage at which thymocytes receive an Treg cells (and instead underwent proliferative expansion). agonist TCR signal influences the mode of tolerance. Although this is reminiscent of a previously recognized

Box 1 | Promiscuous gene expression in the thymus tightly controlled expression of genetic programmes is a hallmark of tissue identity. However, in violation of this rule, medullary thymic epithelial cells (mteCs) express a large number of mrNas encoding self-antigens whose expression is otherwise under stringent developmental, spatial and temporal control, known as tissue-restricted antigens (tras)106,107. the expression of more than 3,000 tra transcripts is controlled by the autoimmune regulator protein (aire). aire lacks features of classical transcription factors and instead seems to facilitate ectopic transcription indirectly through binding to inactive histone marks and promoting the subsequent recruitment of factors associated with chromatin opening, release of stalled rNa polymerase and transcript maturation108,109. AIRE is essential for T cell tolerance, as is evident from the autoimmune manifestations that occur in aire-deficient humans and mice. Promiscuous gene expression of tras is not fully abolished in the absence of aire110, and recent data suggest that the transcription factor Fez family zinc-finger protein 2 (FeZF2) might be a second, independent regulator of promiscuous gene expression111. in support of this, FeZF2-dependent tra expression is not affected in aire-deficient mteCs, and FeZF2 deficiency in thymic epithelium elicits a spectrum of autoimmune manifestations that is partly distinct from that in aire-deficient mice. at the population level, mteCs express the highest number of genes (>85% of all protein-coding transcripts) observed in any cell type112 and thereby provide a comprehensive representation of peripheral self-antigens. However, the expression of a given tra is, in many cases, confined to only 1–3% of all mteCs33,107. For example, pioneering early studies showed that an adult mouse thymus contains a few hundred cells that express pancreas-specific tras113, whereas the total number of mteCs is in the order of 100,000. recent comparisons of gene expression profiles of mteCs at the single-cell level confirmed that there is a substantial degree of cell-to-cell variation in the transcriptomes of mteCs, which further substantiates the idea that promiscuous gene expression entails a stochastic component33,112,114,115.

Nature Reviews | Immunology Reviews

Thymus Periphery expressed and hence early encountered antigens may preferentially operate through clonal deletion. By con- Cortex Medulla trast, a TRA-like pattern of antigen expression in the medulla would be permissive (but not per se instruc- Clonal Treg cell Proliferation and deletion induction effector function tive) for Treg cell differentiation because these antigens are likely to be seen only once the capacity to undergo

Treg cell differentiation has been established. Whether antigen recognition at this late stage induces clonal dele-

tion or Treg cell differentiation might then be determined ? by asynchronous loss of the susceptibility to assume either cell fate or by the modalities of TCR signalling in conjunction with other parameters,as discussed below. Intrinsically programmed response programmed Intrinsically to agonist TCR stimulation Serial antigen encounters

+ + + The mosaic-like pattern of expression of TRAs, such

+ CD8 that only a few cells in the medulla express a given + T cells self-antigen (Box 1), poses obvious limitations on the CD4 Peripheral Mature CD4 CD4 likelihood that a given TRA is seen at all by its cognate DP thymocytesImmatureSP thymocytes CD4 SP thymocytes T cells. In addition, it is conceivable that the number of Progressive T cell maturation APCs that present a given antigen may directly affect the ensuing mode of CD4+ T cell tolerance. In this regard, it Fig. 2 | How T cell-intrinsic developmental windows of opportunity may influence the mode of CD4+ T cell is of particular relevance that most models of thymocyte tolerance. Immature thymocytes have an inherent selection, particularly those based upon affinity and/or predisposition to undergo apoptotic cell death upon avidity, assume that cell fate decisions in the thymus are agonist antigen encounter, whereas fully mature T cells based on a single thymocyte–APC interaction. However, respond to antigenic stimulation by proliferation and the this may not, or at least may not always, be the case. acquisition of effector function. The transition from a Real-time imaging of thymocyte motility in thymic deletion-sensitive to a deletion-resistant state seems to slices showed that in the absence of cognate antigen, coincide with progressive thymocyte maturation at the medullary thymocytes display rapid motion in a ‘random single-positive (SP) stage. Thus, cortical double-positive walk’ pattern and scan their environment for cognate (DP) thymocytes and medullary SP thymocytes with an hi pMHC ligands through engaging in 5–8 short-lived con- immature (HSA ) phenotype are susceptible to clonal 42,43 deletion, whereas more mature (HSAlow) cells are not41. tacts with APCs per hour . Administration of cognate It remains poorly understood when and where during peptide to such thymus slices resulted in the migratory + thymocyte development the capacity to undergo arrest of specific CD8 SP T cells within minutes, fol- 44 regulatory T (Treg) cell differentiation is established, but it is lowed by their apoptotic cell death within a few hours . conceivable that concomitant with gradually losing their Thus, clonal deletion resulting from abundantly availa- susceptibility to deletion, medullary CD4+ SP thymocytes ble antigen may indeed involve a single, yet sustained, enter a transient phase of inclination towards the Treg cell thymocyte–APC interaction. By contrast, when cog- lineage. Consistent with this, efficient Treg cell nate antigen was expressed endogenously in a TRA-like differentiation was observed when the developmental expression pattern in a model of steady-state negative window for clonal deletion was experimentally bypassed selection of CD8+ T cells, there was no evidence for the by intrathymic injection of semi-mature, antigen-naive + T cell receptor (TCR)-transgenic CD4+ SP cells into a migratory arrest of T cells. Instead, CD8 SP cells contin- microenvironment in which clonal deletion was the ually migrated, albeit with decreased velocity, indicating prevalent tolerance mechanism in the steady state40. that they had experienced one or several previous anti- gen encounters42. Whereas these experimental systems focused on TCR-transgenic CD8+ SP T cells, a similar similar transition from deletion susceptibility to deletion pattern of confined migration was observed for a frac- resistance at the CD4+ SP cell stage41, it is tempting to tion of polyclonal MHC class I-restricted and MHC speculate that the developmentally programmed predis- class II-restricted medullary thymocytes42. Thus, it is + position to undergo Treg cell differentiation is maintained possible that for CD4 SP thymocytes also, particular for longer during CD4+ SP cell maturation than is the modalities of self-antigen presentation do not deliver susceptibility to undergo clonal deletion. an immediate and persistent stop signal. In such a On the basis of these combined results, it is conceiva- multiple hit scenario, a mosaic-like pattern of antigen ble that thymocytes transit through cell-intrinsically reg- expression is more likely to result in transient TCR trig- ulated developmental windows of opportunity in which gering, whereas ubiquitously expressed and presented antigen encounter is differentially interpreted (Fig. 2). self-antigens would provide continual TCR stimulation. Whereas susceptibility to clonal deletion is established Consistent with the idea that discontinuous TCR

early at the DP thymocyte stage, the capacity for Treg cell stimulation is favourable or even essential for Treg cell Ectopic transcription differentiation may be acquired only in close temporal differentiation, FOXP3 induction upon in vitro stimula- Expression of a tissue-specific + + mRNA in a cell type where the proximity to CD4 lineage commitment and the reloca- tion of CD4 SP T cells with CD3-specific antibodies was + respective gene product does tion of CD4 SP cells to the medulla. This offers an expla- most efficient when the TCR stimulus was prematurely not have any obvious function. nation as to why CD4+ T cell tolerance to ubiquitously terminated or when, subsequent to an initial phase of

www.nature.com/nri Reviews

TCR stimulation, phosphoinositide 3-kinase (PI3K) or Thymic antigen-presenting cells its downstream signalling modules — protein kinase B Developing thymocytes may encounter self-antigens on (also known as AKT1) or mechanistic target of rapamycin various types of APCs33, including classical dendritic (mTOR) — were inhibited45. Along these lines, mathe- cell (cDC) subsets50,51, plasmacytoid DCs (pDCs)52, matical models have suggested a signal integration model B cells53,54, cortical thymic epithelial cells and mTECs55. It of thymocyte selection that provides a conceptual frame- is widely accepted that all these thymic APC subtypes can work to understand how the TCR stimulation history may induce T cell tolerance. However, their relative contribu- determine the outcome of thymic selection based upon a tions to shaping the polyclonal T cell repertoire remain synthesis of both sustained and transient TCR signalling poorly understood with regard to both the spectrum of levels46. This model posits that clonal deletion of thymo- self-antigens that they display and whether self-antigen cytes may result from a strong, isolated signalling event, encounter on distinct APCs preferentially or even

whereas Treg cell differentiation may ensue from repeated, selectively results in clonal deletion or Treg cell induction. weak TCR stimulation, and it is clear how ubiquitous and Mouse models lacking cDCs or subsets of mTECs, high-level expression of self-antigens versus mosaic-like or with absent or aberrant MHC class II expression on and low-level expression of self-antigens may favour subclasses of thymic APCs, have been used to address (Fig. 3) 12 clonal deletion versus Treg cell differentiation . these questions (reviewed previously ). Irrespective of In sum, a growing body of observations indicates that whether antigen presentation by DCs or mTECs was clonal deletion can result from a single hit of antigen that inhibited, these models showed an increase in the num- promotes immediate arrest of the thymocyte on the APC ber of CD4+ SP T cells, which is thought to reflect the surface, which may in turn amplify the intensity and/or non-redundant contribution of the respective APCs to duration of TCR signalling43,47–49. However, it remains clonal deletion of MHC class II-restricted T cells55–57. By

to be seen whether, by contrast, Treg cell differentiation contrast, the overall size of the polyclonal Treg cell com- involves, or even requires, multiple antigen encounters. partment was not markedly affected upon interfering

a b Thymic APC Thymic APC expressing high expressing low level of pMHC level of pMHC

TCR TCR Apoptotic thymocyte

Antigen- Antigen- specific specific thymocyte thymocyte Treg cell

Threshold for clonal deletion TCR signal intensity TCR signal intensity

Time Time of thymocyte of thymocyte Migration velocity Migration velocity Migration Time Time Fig. 3 | Model for thymocyte integration of T cell receptor signalling strength and number of antigen encounters. a | Antigens that are expressed ubiquitously and presented at high levels — high levels of peptide–MHC (pMHC) shown in dark red — on thymic antigen-presenting cells (APCs) may favour clonal deletion of specific thymocytes through the induction of an isolated strong T cell receptor (TCR) signal. Immediate arrest of strongly TCR-signalled thymocytes on the APC surface may prolong the duration of TCR signalling and thereby establish a feedforward loop that amplifies the aggregate signal input. b | A mosaic pattern of self-antigen expression and presentation at comparably low levels (as indicated by the lighter red shading of thymic APCs) — as is typical for tissue-restricted antigens — may allow for serial, yet temporally separated, sub-threshold antigen encounters during continual migration of thymocytes through the 46 thymic medulla. This mode of antigen recognition was suggested to favour regulatory T (Treg) cell differentiation .

Nature Reviews | Immunology Reviews

Peptide–MHC ligandome with the capacity of individual APC subsets to con- cell fate determinants may be provided in the form of + The repertoire of peptides that tribute to central tolerance induction for CD4 T cells. co-stimulatory molecules or soluble factors, and they are bound by MHC molecules However, sequencing of TCRs on Treg cells in mice lack- represent potential determinants of deletion versus on a given cell type. ing DCs or with decreased MHC class II expression on diversion into the Treg cell lineage. Common cytokine receptor mTECs clearly pointed to a non-redundant contribution γ-chain family cytokines of these APCs to shaping the Treg cell repertoire at the IL-2. Together with the TCR stimulus, IL-2 has a cru- 24 58 (γc cytokines). The level of individual antigen specificities . Thus, in both cial role in promoting thymic Treg cell development . heterodimeric or mouse models, distinct sets of TCRs were absent from The relationship between TCR and cytokine signals heterotrimeric receptors for at the T cell repertoire and instead were present in the has been framed in terms of a two-step model, whereby least six cytokines (IL-2, IL-4, reg IL-7, IL-9, IL-15 and IL-21) repertoire of conventional T cells. strong TCR signals (step one) lead to the upregulation of share a common γ-chain In sum, there is no evidence to suggest that a sin- expression of CD25 (also known as IL-2Rα), which con-

(known as γc or CD132). The gle thymic APC subset is specialized and required fers high-affinity IL-2 binding (step two) and, in turn, receptors for IL-2 and IL-15 59,60 + − for either clonal deletion or Treg cell differentiation. promotes FOXP3 upregulation . CD25 FOXP3 cells also share the β-subunit (CD122) but contain distinct Nonetheless, it is becoming increasingly clear that are present at a frequency of 2–3% in the steady-state + α-subunits (CD25 and CD215, thymic APC subtypes, owing to differences in the way CD4 SP T cell compartment. A substantial fraction of + + respectively). Both α-subunits that they sample and process self-antigens, differ in these cells gives rise to mature CD25 FOXP3 Treg cells confer high-affinity cytokine their peptide–MHC ligandome (reviewed previously33) upon in vivo transfer into an MHC class II-deficient thy- binding but are not directly (Box 2). Irrespective of the ensuing mode of tolerance, mus or upon in vitro culture with IL-2, which indicates involved in signal transduction. the presentation of partly unique arrays of self-peptides that a fraction of CD25+FOXP3− CD4+ SP thymocytes by different APC subsets is expected to translate into a are at an intermediate developmental stage, where TCR non-redundant role for these subsets in central tolerance signals are dispensable and IL-2 is sufficient for the com- 60 to autoantigens that are not ubiquitously available. pletion of Treg cell differentiation . Further support for a model in which CD25 expression precedes the induction

Co-stimulation and soluble factors of FOXP3 in the majority of developing Treg cells comes In addition to having partly non-overlapping peptide– from molecular timing studies in vivo61. However, there

MHC ligandomes, thymic APC subsets may also differ are also indications that some thymic Treg cells differen- in their capacity to provide non-antigenic signals that tiate via a pathway in which the upregulation of FOXP3 act on developing thymocytes simultaneously or in close expression precedes CD25 induction62,63. Moreover, temporal proximity to TCR stimulation. Such extrinsic other common cytokine receptor γ-chain family cytokines

(γc cytokines), in particular IL-15, can have a redundant 62,64–66 Box 2 | The peptide–mHC class II ligandomes of thymic antigen-presenting cells role in promoting Treg cell development . thymic antigen-presenting cell (aPC) subsets are expected to display partly Levels of IL-2 in the thymus are likely to be a lim- 65,67 non-overlapping peptide–MHC class ii ligandomes owing to differences in their origin iting factor for Treg cell differentiation , which and the way in which they capture and process antigens (reviewed previously33). raises questions about the cellular sources of IL-2. In Cortical thymic epithelial cells (cteCs) express unusual proteases whose proteolytic a thymic tissue slice model that recapitulates many

activity differs from the antigen processing that occurs in other aPCs. these aspects of steady-state thymic Treg cell development, cteC-specific proteases are not at all or only weakly expressed in other cell types, and Il2−/− DCs were significantly less efficient in their abil- they have been implicated in the generation of unique MHC class i-bound and MHC ity to induce agonist-driven Treg cell development than class ii-bound peptides for positive selection. Medullary teCs (mteCs), like cteCs, are Il2+/+ DCs, which suggests that APCs may simultane- inefficient in loading extracellular self-antigens onto MHC class ii molecules. instead, ously provide both IL-2 and the TCR ligand required both cteCs and mteCs seem to preferentially use non-canonical MHC class ii loading for T cell differentiation68. By contrast, a recent study pathways for endogenous antigens116. in the case of mteCs, a peptide–MHC class ii reg ligandome predominantly composed of endogenously derived substrates is expected using T cell-specific, B cell-specific or DC-specific con- to include tissue-restricted antigens (tras). variability in the expression of tras ditional Il2 knockout showed that IL-2 derived from between individual mteCs is likely to result in a corresponding degree of variability also thymocytes themselves is necessary and sufficient for in their peptide–MHC class ii ligandomes. thymic Treg cell development, in particular in the absence Classical dendritic cells (cDCs) in the thymus segregate into at least two subsets50. of IL-15 (ref.69). Importantly, this study also used mixed Migratory CD8−sirPα+ cDCs migrate into the thymus from the circulation, and their chimaeras to show that in Il2−/− recipients of an equal peptide–MHC class ii ligandomes may contain epitopes derived from self-antigens that mixture of wild-type and T cell-specific Il2-knockout have been captured in the periphery117. resident CD8+sirPα− cDCs arise through bone marrow, IL-2-deficient and IL-2-sufficient thymo- intrathymic differentiation. similar to their migratory counterparts, they efficiently sample cytes underwent T cell differentiation with identical extracellular sources of self-antigen. although both types of thymic cDC may thereby reg efficacy, which indicates that autocrine IL-2 is dispen- capture and present mteC-derived antigens, resident cDCs may be particularly efficient in doing so118, as supported by T cell receptor repertoire studies24. approximately one-third of sable for Treg cell differentiation and that IL-2 is instead thymic DCs are plasmacytoid DCs (pDCs); pDCs are of extrathymic origin and are capable provided by neighbouring T cells. The exact identity of inducing central tolerance to self-antigens that have been captured at peripheral sites52. of the bystander T lineage cells that produce IL-2 to thymic B cells may originate from intrathymic B lymphopoiesis, but they can also be support Treg cell differentiation remains to be estab- of extrathymic origin. unlike mainstream peripheral B cells, thymic B cells have lished, but CD4+ SP cells differentiating into the con- prominent aPC features, such as high levels of expression of MHC class ii molecules and ventional CD4+ T cell lineage are obvious candidates. CD80. Curiously, they also express the autoimmune regulator protein (aire), but the Intriguingly, SP thymocytes acquire the competence for functional importance of this finding remains to be established53,54. it has been IL-2 production upon TCR stimulation only gradually speculated that the self-peptide–MHC class ii ligandome of thymic B cells may mirror at a late phase of maturation70. It is therefore conceiv­ that of germinal centre B cells and/or contain self-antigens that have been captured via able that the availability of IL-2 to drive the second step autoreactive B cell receptors. of Treg cell differentiation is restricted to the immediate

www.nature.com/nri Reviews

a b

FOXP3+ Rare APC with mature Conventional cognate pMHC Treg cell CD4+ T cell

IL-2 trans- presentation IL-2 Rare APC consumption with cognate MHC class II pMHC CD25 IL-2 TCR IL-2Rγ Free IL-2 gradient CD25 ? IL-2Rβ Apoptosis Treg cell Treg cell progenitor progenitor

FOXP3+ mature T cell Conventional reg CD4+ T cell c First incoming precursor Second incoming precursor

Thymocyte #3 #2 #1 #3 #2 #1 #3 #2 #1 precursor

CD25–FOXP3– CD25–FOXP3– CD25–FOXP3– T cell progenitor T cell progenitor reg Treg cell reg progenitor Lack of Step one: Step one: TCR signal TCR signal TCR signal

CD25+FOXP3– Rare APC + – Rare APC with cognate Rare APC CD25 FOXP3 with cognate Treg cell with cognate T cell pMHC reg pMHC intermediate pMHC intermediate Step two: IL-2 signal Sequestration Conventional of pMHC CD4+ T cell IL-2 Lack of IL-2 gradient consumption IL-2 signal

CD25+FOXP3+ + + + + mature Treg cell CD25 FOXP3 CD25 FOXP3 Apoptosis mature Treg cell mature Treg cell Competition for antigen Competition for IL-2

Fig. 4 | model of limited developmental niches for antigen-specific thymocyte progenitors to undergo step two of Treg cell differentiation. regulatory T cells. a | A thymic niche for regulatory T (Treg) cell development IL-2-deprived progenitors may either revert to the conventional T cell is defined by a rare antigen-presenting cell (APC) that presents a particular, lineage or, perhaps more likely , undergo deletion. c | A schematic of possibly tissue-restricted antigen-derived, agonist self-peptide and by an intraclonal competition for antigen and IL-2 between Treg cell progenitors IL-2 source in the immediate vicinity. IL-2 produced by a relatively mature, of identical specificity is shown. The first thymocyte progenitor (light yellow) + conventional self-reactive CD4 single-positive thymocyte (possibly an that enters the thymic niche undergoes complete maturation into a Treg cell escapee from clonal deletion) remains associated with the niche as a result (left panel). Once it is mature, this cell may sequester cognate peptide–MHC of limited diffusion of IL-2 and/or the capture and trans-presentation of IL-2 (pMHC) complexes and deplete local IL-2, thereby interfering with step one by CD25 expressed on the APC. The Treg cell progenitor receives both a (middle panel) and/or step two (right panel), respectively , of Treg cell strong T cell receptor (TCR) signal and IL-2, allowing its development into differentiation of subsequently incoming progenitors of the same + a mature forkhead box protein P3 (FOXP3) Treg cell. b | A schematic of TCR specificity. The alternative fate of outcompeted precursors could be escape specificity-independent feedback inhibition of Treg cell differentiation is from central tolerance induction to become conventional T cells (more likely shown. Mature Treg cells in the thymus, including both recently differentiated in the case of competition for antigen) or deletion through apoptosis (more cells and re-immigrants from the peripheral circulation, require and likely in the case of competition for IL-2). Both types of intraclonal consume IL-2 as a survival factor and thereby deplete IL-2 from the thymic competition would require, or at least be increased by , prolonged retention niche. IL-2 depletion limits the ability of newly TCR-primed (step one) of recently matured Treg cells in the selecting niche.

vicinity of a few mature, pre-emigration, conventional cells at the corticomedullary junction71. Interestingly, a SP T cells that produce IL-2 upon encountering rare ago- fraction of both thymic cDCs and mTECs expresses the nist pMHC ligands (Fig. 4a). Consistent with this idea, high-affinity CD25 chain of the IL-2 receptor but not in situ analyses showed that IL-2 protein is confined to the signalling chain CD122 (ref.69), which suggests that ‘haloes’ surrounding a small number of IL-2-producing these APCs may capture T cell-derived IL-2 and restrict its

Nature Reviews | Immunology Reviews

+ − Trans-presentation free diffusion. Such local retention of IL-2 by neighbour- or not an antigen-primed CD25 FOXP3 thymocyte + + Cytokine trans-presentation is ing APCs may not only reinforce the spatial constraints precursor gives rise to a mature CD25 FOXP3 Treg cell, best established for IL-15, in IL-2 availability but also allow for trans-presentation72 the fate of those CD25+FOXP3− thymocytes that lose the whereby signalling in a of surface-bound IL-2 to developing T cells. competition for IL-2 remains to be established; it may responder cell results from reg recognition of complexes of The amount of IL-2 available to support thymic Treg cell depend on the developmental stage at which they IL-15 with the high-affinity generation is not only limited by the number of IL-2- encounter antigen, the type of APC and the affinity of the IL-15Rα subunit (CD125) on the producing cells but also controlled by a negative feed- TCR. On the one hand, it is possible that IL-2-deprived surface of a trans-presenting back loop in which mature Treg cells restrain the de novo autoreactive thymocytes may revert to a conventional cell type, which may itself differentiation of T cells (Fig. 4b). Using thymic tissue CD4+ T cell fate and leave the thymus as escapees from produce IL-15. It is controversial reg whether the IL-2Rα subunit slices, it was shown that existing thymic Treg cells within central tolerance induction. On the other hand, given the (CD25) can bind IL-2 the thymic slice environment limited cytokine availabil- dual function of IL-2 as a factor for both differentiation independently of other receptor 68 ity and inhibited the development of new Treg cells . In a and survival, it seems more likely that these cells die. subunits and thereby elicit IL-2 further intriguing twist, T cells can migrate back to the Consistent with that idea, TCR-transgenic presumptive signalling in trans. reg thymus from the periphery and so may place additional Treg cells that were incapable of signalling via CD25 were 73 76 CD28–B7 signalling axis pressure on the system by also competing for IL-2 (ref. ). lost from the T cell repertoire , which suggests that for

CD28 is the prototypical Curiously, the recirculation of peripheral Treg cells requires cells that are poised to enter the Treg cell lineage, IL-2 is co-stimulatory receptor AIRE, which seems to reflect a role of AIRE in chemo­ indeed a crucial determinant of life or death. Surprisingly expressed on the surface of kine production by mTECs rather than in promoting TRA little is known with regard to the possibility that classi- T cells. It can bind to CD80 74 (also known as B7.1) or CD86 expression . It remains to be established whether recircu- cal clonal deletion may likewise involve intermediate (also known as B7.2) on the lation of Treg cells to the thymus is developmentally con- cell stages that are responsive to γc cytokines, but it is surface of antigen-presenting trolled, involves environmental triggers and/or selectively noteworthy that IL-7 can negate negative selection in cells. applies only to certain antigen specificities. particular models of clonal deletion77,78. Competition for IL-2 per se is an antigen-independent process, but it also has implications for another enig- CD28. Cd28−/− or Cd80−/−Cd86−/− mice fail to gener- 79 matic aspect of thymic Treg cell differentiation — namely, ate a Treg cell pool of normal size . The small number + −/− the intraclonal competition between Treg cell precursors of mature FOXP3 Treg cells in the thymus of Cd28 of shared TCR specificity. Models in which the fate of mice is reminiscent of the phenotype of Il2−/− mice. a thymocyte is solely specified by its TCR predict that However, although co-stimulation has been impli-

transgenic expression of a Treg cell-derived TCR would cated in cytokine production, the Treg cell deficiency in −/− −/− −/− lead to increased Treg cell development. However, initial Cd28 or Cd80 Cd86 mice is not caused by depri- analyses of transgenic mice with TCRs cloned from nat- vation of exogenous cytokines but primarily reflects a

urally occurring Treg cells of unknown specificity showed T cell-intrinsic function of the CD28–B7 signalling axis. that such monoclonal systems have only very small For example, the failure of Cd28−/− thymocytes to effi- + 13,14 numbers of FOXP3 cells . Efficient Treg cell develop- ciently enter the Treg cell lineage is not rescued by the ment was restored upon reducing the frequency of cells presence of bystander Cd28+/+ cells80. 13,14 bearing the respective Treg cell-derived TCR , which Co-stimulation through CD28 is believed to func-

indicates that Treg cell differentiation involves a limiting tion simultaneously to the TCR signal at an early stage 40,81,82 niche that is saturated at very low clonal frequencies. of Treg cell differentiation , and one possibility is Competition for antigen is likely to be a rate-limiting that its role is to increase the aggregate magnitude of the

determinant that restrains Treg cell differentiation at step underlying agonist stimulus. In the context of models

one. This is supported by the observation that a grad- of Treg cell differentiation versus clonal deletion centred ual reduction of the frequencies of T cell precursors on signal strength, one may therefore hypothesize that resulted in a proportional increase in the level of TCR for cells that engage in TCR–pMHC interactions of rel- 75 stimulation in developing Treg cells . The mosaic-like atively low affinity or avidity, co-stimulation may favour

expression pattern of many self-antigens in the thymus Treg cell generation over the differentiation of conven- suggests that the thymus contains plenty of different tional T cells. By contrast, for TCR–pMHC interactions

niches tailored to promote the development of Treg cells of relatively high affinity and/or avidity, co-stimulation

with distinct self-antigen specificities. Considering the may tip the balance between Treg cell differentiation evidence for linked display of agonist pMHC complexes and clonal deletion in favour of the latter. According

and trans-presented IL-2 by the same APC, it is conceiv- to such a scenario, the smaller Treg cell compartment of

able that within these niches, Treg cell precursors may CD28-deficient mice would be expected to contain a compete for access to IL-2 as well as antigen. It is tempt- substantially altered TCR repertoire, which on the one

ing to speculate that mature Treg cells may preferentially hand is depleted of cells that have escaped into the deplete sources of IL-2 associated with the same APC conventional T cell repertoire and on the other hand that selected them originally. This would primarily limit harbours TCR clones that otherwise would have under-

the development of Treg cells with related specificities, gone deletion. However, the TCR composition of the −/− which potentially explains why low-frequency Treg cell residual Treg cell pool of Cd28 mice is surprisingly similar precursors develop more efficiently than high-frequency to that of wild-type mice, which suggests that CD28 (Fig. 4c) Treg cell precursors . does not function primarily to amplify the TCR signal How do these findings relate to clonal deletion? strength but to provide a qualitatively distinct signal

Whereas competition for a developmental niche that is that somehow supports the formation of a Treg cell com- defined by the availability of IL-2 may influence whether partment of normal size81. The possibility that CD28

www.nature.com/nri Reviews

drives the proliferative expansion of a small precursor Transforming growth factor-β. Mice lacking transform- 83 pool of Treg cells has been excluded , and so, it seems that ing growth factor-β1 (TGFβ1) or with T cell-specific

CD28 functions in thymic Treg cell generation through deletion of TGFβ receptors develop a fatal inflammatory a true lineage-specifying function and/or through a disorder, which highlights the crucial role of TGFβ in pro-survival function. In keeping with the latter function, T cell tolerance95. Owing to the pleiotropic functions of and in further evidence against a role for co-stimulation in TGFβ in the immune system, it has been challenging to signal amplification, CD28 deficiency in a TCR-transgenic conclusively dissect the role of TGFβ during thymocyte

model of agonist-driven Treg cell differentiation resulted in differentiation and central tolerance induction. Thus,

the physical loss of presumptive Treg cells rather than their although it is widely accepted that TGFβ supports the escape as conventional FOXP3− T cells83. These findings in vitro conversion of naive T cells into induced FOXP3+ 96 are consistent with CD28 signalling preventing the death Treg cells and is crucial for the in vivo conversion of naive

of TCR-activated Treg cell precursors. They suggest that T cells into peripherally derived Treg cells in physiological whether or not CD28-mediated co-stimulation accompa- as well as therapeutic settings97,98, an analogous role for

nies the underlying agonistic TCR signal can be a binary TGFβ in thymic Treg cell differentiation has remained con-

switch that specifies diversion into the Treg cell lineage or tentious. Initial analyses of TGFβ-deficient mice showed clonal deletion, respectively. Intriguingly, observations that these animals have more, rather than fewer, thymic

in other cellular contexts indicate that CD28 can nega- Treg cells when analysed at approximately 2 weeks of tively regulate the expression of the pro-apoptotic protein age99,100. However, it was later found that mice lacking the BCL-2-interac­ting mediator of cell death (BIM; also TGFβ receptor on T cells are essentially devoid of thymic 84 known as BCL2L11) , which is an essential mediator Treg cells when analysed at an earlier time point and that 85 of classical clonal deletion . However, BIM deficiency the apparently normal numbers of thymic Treg cells at −/− does not rescue the thymic Treg cell defect in Cd28 a later stage were secondary to systemic inflammation 86 mice , which indicates that CD28 signalling in Treg cell and increased levels of IL-2, which, in turn, expand what 101 precursors either suppresses other death pathways or, would otherwise be a very small Treg cell population . in addition to suppressing BIM, concomitantly has a These data clearly show that TGFβ is required for the for-

lineage-specifying function. mation of a thymic Treg cell pool of normal size; however, The role of CD28-mediated co-stimulation in clas- different underlying mechanisms have been proposed. sical clonal deletion has been controversial. In vitro On the one hand, it has been suggested that the experiments showed that CD28 signalling can increase function of TGFβ in the thymus is as a survival factor. the death of TCR-stimulated thymocytes77,87. By con- Apoptosis was massively increased among FOXP3+ trast, numerous in vivo studies, involving either thymocytes lacking the TGFβ receptor86. Of note, cell superantigen-reactive T cells or TCR-transgenic sys- death in the context of classical clonal deletion mediated tems of negative selection, have failed to support a cru- by CD3-specific antibody or endogenously expressed cial requirement for CD28 in this process88–92. A more neo-antigen was also increased in the absence of TGFβ recent investigation revisited the interpretation that signalling, which indicates that the anti-apoptotic co-stimulation via CD28 is dispensable for deletional function of TGFβ is not restricted to cells that have

tolerance in vivo and provided an unexpected new twist. entered the Treg cell lineage but applies more generally In this study, superantigen-reactive T cells were absent to all thymocytes that receive an agonist TCR stimu- from the mature thymocyte compartment irrespective of lus86,102. Ablation of the pro-apoptotic molecule BIM

whether or not co-stimulation was available, but in the restored the thymic Treg cell pool to normal size in the absence of CD28, this was not a consequence of their absence of TGFβ signalling, which suggests that TGFβ clonal deletion but resulted from their re-direction into is required to inhibit BIM-dependent apoptosis during − − + a CD4 CD8 TCRαβ compartment that may contain the thymic Treg cell differentiation but is not essential for the precursors of CD8αα+ intraepithelial lymphocytes93. Thus, induction of FOXP3 expression. In line with this, a con- in the case of superantigen-mediated clonal deletion of served cis-element in the Foxp3 gene containing a TGFβ thymocytes, CD28 seems to relay a signal that favours response element is crucial for the induction of peripher-

cell death over another agonist-driven cell fate. ally derived Treg cells in gut-associated lymphoid tissues 98,103 Taking the results together, it seems that CD28- but is dispensable for thymic Treg cell differentiation . mediated co-stimulation is crucially required for both On the other hand, some findings support the idea + CD8αα intraepithelial clonal deletion, at least as far as superantigen-mediated that TGFβ not only keeps T cells alive but also specifies lymphocytes reg Intraepithelial lymphocytes are T cell death is concerned, and the efficient formation of entry into the Treg cell lineage. For example, the thymic enriched in αβ T cells that a Treg cell compartment of normal size. Somewhat par- Treg cell compartment is not decreased in size when the express CD8 homodimers. αα adoxically, during Treg cell differentiation, CD28 seems TGFβ receptor is inactivated late in thymic Treg cell dif- These cells are thought to have to counteract cell death, whereas in the context of clonal ferentiation using a Foxp3–Cre system104. This indicates a crucial function in immune regulation at mucosal deletion, it seems to promote cell death. This is reminis- that TGFβ signalling is dispensable for the survival of interfaces, in particular in the cent of earlier observations in T cell hybridomas that mature thymic Treg cells, although it remains possible intestine. There is evidence showed, depending on the signalling context, that CD28- that it exerts a crucial pro-survival function at an earlier that CD8αα+ intraepithelial mediated co-stimulation can induce either pro-apoptotic stage of thymic Treg cell differentiation. Upon intrathymic lymphocytes emerge from or anti-apoptotic genes94, and it will be interesting to injection of polyclonal CD4−CD8− thymocytes, 2–3% of agonist selection in the thymus, + + but their exact developmental dissect whether distinct molecular effectors down- these cells give rise to FOXP3 CD4 SP cells, and this is origin remains poorly stream of CD28 operate in Treg cell differentiation versus believed to mimic steady-state Treg cell differentiation. understood. clonal deletion. TGFβ receptor-deficient thymocyte precursors failed

Nature Reviews | Immunology Reviews

+ to give rise to FOXP3 Treg cells in this setting, which in TCR signalling intensity are translated into a binary indicates that TGFβ signalling has a direct cell-intrinsic cell fate decision. For example, almost nothing is known role in the initiation of Foxp3 transcription104. As a note regarding how TCR signalling events in the context of

of caution, however, it is not known whether the small step one of the two-step model of Treg cell differentia- fraction of input cells that would normally assume a tion may qualitatively or quantitatively differ from a

Treg cell fate indeed failed to initiate Foxp3 transcription TCR stimulus that results in clonal deletion. Similarly, or instead failed to survive. the molecular basis of differential integration of TCR In conclusion, further study is required to unravel the signals and non-antigenic signals (co-stimulation and exact roles of TGFβ during thymic selection, although cytokines) at subsequent stages of thymocyte develop- it is reasonable to assume that a general pro-survival ment is only poorly understood, and it will be inter- function of TGFβ for TCR-signalled thymocytes and esting to unravel how stage-specific sensitization or

a function as a bona fide differentiation factor for Treg cells de-sensitization of the TCR signalling machinery are not mutually exclusive. Irrespective of its exact func- and/or transcriptional or epigenetic control may poise tion, the net effect of TGFβ in the thymus seems to be to thymocytes for particular fate decisions.

restrain clonal deletion while supporting Treg cell devel- With regard to a multiple hit model of thymocyte opment. Consequently, TGFβ is expected to adjust the selection, improved intravital methodologies to visualize balance between these two modes of central tolerance, antigen recognition events simultaneously with the ensu- which raises questions regarding the sources of TGFβ in ing cell fate choice will be necessary to dissect whether the thymic microenvironment and whether the availabil- sequential and intermittent encounters with agonist

ity of TGFβ is subject to temporal and spatial regulation. ligands in vivo are a prerequisite for Treg cell differen- The total amount of active TGFβ in the thymus grad- tiation or simply reflect the pre-emigration behaviour 104 ually increases over the neonatal period , showing a of Treg cells subsequent to and independent of the actual strong correlation with the delayed emergence of the first lineage-instructing TCR stimulus. Of note, it remains to 38 Treg cells at approximately day 3 after birth . Intriguingly, be seen how a model of Treg cell generation that is based the extent of thymocyte apoptosis seems to directly con- upon serial, short-lived encounters with antigen can be trol the supply of intrathymic TGFβ, whereby contact reconciled with a limiting niche model that presumes

with dying thymocytes promotes TGFβ production prolonged retention of Treg cells and competition for by thymic macrophages and DCs (and possibly also antigen and IL-2 in the instructing microenvironment. TECs)104; this suggests some form of cross-regulation Finally, a deeper understanding of the determinants of 105 between clonal deletion and Treg cell differentiation . clonal deletion versus Treg cell diversion in the thymus is complicated by the fact that both cell fates might be achieved Perspectives via pathways that are more heterogeneous than commonly Recent data have revealed intriguing correlations between assumed. On the one hand, there are hints that classical ubiquitous antigen expression in the thymus and clonal clonal deletion differs in its requirements for co-stimulation deletion of thymocytes and between TRA-like mosaic and/or apoptotic effector pathways depending on the antigen expression in the thymus and diversion of thymo­ developmental stage of thymocytes41. On the other hand,

cytes into the Treg cell lineage. We have discussed how Treg cell development may occur via different pathways that these findings may relate to thymocyte-intrinsic and are characterized by differences in the order of expression thymocyte-extrinsic determinants that specify the fate of of CD25 and FOXP3. Importantly, irrespective of whether + − autoreactive MHC class II-restricted thymocytes. As a con- Treg cell differentiation proceeds via CD25 FOXP3 or − + sequence, the thymus-derived Treg cell repertoire should CD25 FOXP3 intermediates, both pathways of Treg cell have a strong bias towards the recognition of peripheral differentiation seem to involve apoptosis-prone tran- antigens while being largely devoid of TCR specificities sitional stages at which cells compete for cytokines for widely expressed antigens. This remains to be tested. as survival factors. Thus, a substantial fraction of

None of the cell fate determinants outlined here has TCR-signalled Treg cell intermediates may die because emerged as a single switch between clonal deletion and of cytokine deprivation, which raises the possibility that

clonal diversion, and a coherent model that incorporates blunted Treg cell differentiation represents a significant, if all potential parameters is lacking. The strength of the not major, pathway of clonal deletion at the CD4+ SP stage. underlying TCR stimulus is likely to be a prime determi- Published online xx xx xxxx nant, but it remains unclear how incremental variations

1. Kappler, J. W., Roehm, N. & Marrack, P. T cell 5. Salaun, J. et al. Thymic epithelium tolerizes activated T cells expressing IL-2 receptor alpha-chains tolerance by clonal elimination in the thymus. Cell 49, for histocompatibility antigens. Science 247, (CD25). Breakdown of a single mechanism of 273–280 (1987). 1471–1474 (1990). self-tolerance causes various autoimmune diseases. 2. Kisielow, P., Bluthmann, H., Staerz, U. D., Steinmetz, M. 6. Fowell, D. & Mason, D. Evidence that the T cell J. Immunol. 155, 1151–1164 (1995). & von Boehmer, H. Tolerance in T cell-receptor repertoire of normal rats contains cells with the 9. Thornton, A. M. et al. Expression of Helios, an Ikaros transgenic mice involves deletion of nonmature potential to cause diabetes. Characterization of transcription factor family member, differentiates CD4+8+ thymocytes. Nature 333, 742–746 (1988). the CD4+ T cell subset that inhibits this autoimmune thymic-derived from peripherally induced Foxp3+ T 3. Modigliani, Y. et al. Lymphocytes selected in potential. J. Exp. Med. 177, 627–636 (1993). regulatory cells. J. Immunol. 184, 3433–3441 (2010). allogeneic thymic epithelium mediate dominant 7. Powrie, F., Leach, M. W., Mauze, S., Caddle, L. B. & 10. Weiss, J. M. et al. Neuropilin 1 is expressed on tolerance toward tissue grafts of the thymic epithelium Coffman, R. L. Phenotypically distinct subsets of CD4+ thymus-derived natural regulatory T cells, but not haplotype. Proc. Natl Acad. Sci. USA 92, 7555–7559 T cells induce or protect from chronic intestinal mucosa-generated induced Foxp3+ T reg cells. (1995). inflammation in C. B-17 scid mice. Int. Immunol. 5, J. Exp. Med. 209, 1723–1742 (2012). 4. Ohki, H., Martin, C., Corbel, C., Coltey, M. & 1461–1471 (1993). 11. Hsieh, C. S., Lee, H. M. & Lio, C. W. Selection of Le Douarin, N. M. Tolerance induced by thymic epithelial 8. Sakaguchi, S., Sakaguchi, N., Asano, M., Itoh, M. regulatory T cells in the thymus. Nat. Rev. Immunol. grafts in birds. Science 237, 1032–1035 (1987). & Toda, M. Immunologic self-tolerance maintained by 12, 157–167 (2012).

www.nature.com/nri Reviews

12. Klein, L. & Jovanovic, K. Regulatory T cell lineage 36. Daley, S. R., Hu, D. Y. & Goodnow, C. C. Helios marks 59. Burchill, M. A. et al. Linked T cell receptor and cytokine commitment in the thymus. Semin. Immunol. 23, strongly autoreactive CD4+ T cells in two major waves signaling govern the development of the regulatory 401–409 (2011). of thymic deletion distinguished by induction of PD-1 T cell repertoire. Immunity 28, 112–121 (2008). 13. Bautista, J. L. et al. Intraclonal competition limits the or NF-kappaB. J. Exp. Med. 210, 269–285 (2013). 60. Lio, C. W. & Hsieh, C. S. A two-step process for thymic fate determination of regulatory T cells in the thymus. This study reveals the extent, stage and molecular regulatory T cell development. Immunity 28, 100–111 Nat. Immunol. 10, 610–617 (2009). nature of two distinct waves of clonal deletion in (2008). 14. Leung, M. W., Shen, S. & Lafaille, J. J. TCR-dependent the normal thymus. 61. Bending, D. et al. A timer for analyzing temporally differentiation of thymic Foxp3+ cells is limited to small 37. Stritesky, G. L. et al. Murine thymic selection dynamic changes in transcription during differentiation clonal sizes. J. Exp. Med. 206, 2121–2130 (2009). quantified using a unique method to capture deleted in vivo. J. Cell. Biol. 217, 2931–2950 (2018). 15. Hsieh, C. S., Zheng, Y., Liang, Y., Fontenot, J. D. T cells. Proc. Natl Acad. Sci. USA 110, 4679–4684 62. Marshall, D., Sinclair, C., Tung, S. & Seddon, B. & Rudensky, A. Y. An intersection between the (2013). Differential requirement for IL-2 and IL-15 during self-reactive regulatory and nonregulatory T cell This paper uses an elegant method to assess the bifurcated development of thymic regulatory T cells. receptor repertoires. Nat. Immunol. 7, 401–410 quantitative impact of clonal deletion at different J. Immunol. 193, 5525–5533 (2014). (2006). stages of thymocyte differentiation. 63. Tai, X. et al. Foxp3 transcription factor is proapoptotic 16. Pacholczyk, R., Ignatowicz, H., Kraj, P. & Ignatowicz, L. 38. Fontenot, J. D., Dooley, J. L., Farr, A. G. & and lethal to developing regulatory T cells unless Origin and T cell receptor diversity of Foxp3+ CD4+ Rudensky, A. Y. Developmental regulation of Foxp3 counterbalanced by cytokine survival signals. Immunity CD25+ T cells. Immunity 25, 249–259 (2006). expression during ontogeny. J. Exp. Med. 202, 38, 1116–1128 (2013). 17. Apostolou, I., Sarukhan, A., Klein, L. & von Boehmer, H. 901–906 (2005). 64. Bayer, A. L., Lee, J. Y., de la Barrera, A., Surh, C. D. & Origin of regulatory T cells with known specificity for 39. Lee, H. M. & Hsieh, C. S. Rare development of Foxp3+ Malek, T. R. A function for IL-7R for CD4+CD25+Foxp3+ antigen. Nat. Immunol. 3, 756–763 (2002). thymocytes in the CD4+CD8+ subset. J. Immunol. 183, T regulatory cells. J. Immunol. 181, 225–234 (2008). 18. Jordan, M. S. et al. Thymic selection of CD4+ CD25+ 2261–2266 (2009). 65. Vang, K. B. et al. IL-2, -7, and -15, but not thymic regulatory T cells induced by an agonist self-peptide. 40. Wirnsberger, G., Mair, F. & Klein, L. Regulatory T cell stromal lymphopoeitin, redundantly govern CD4+Foxp3+ Nat. Immunol. 2, 301–306 (2001). differentiation of thymocytes does not require a regulatory T cell development. J. Immunol. 181, 19. Klein, L., Khazaie, K. & von Boehmer, H. In vivo dedicated antigen-presenting cell but is under 3285–3290 (2008). dynamics of antigen-specific regulatory T cells not T cell-intrinsic developmental control. Proc. Natl 66. Yao, Z. et al. Nonredundant roles for Stat5a/b in directly predicted from behavior in vitro. Proc. Natl Acad. Acad. Sci. USA 106, 10278–10283 (2009). regulating Foxp3. Blood 109, 4368–4375 (2007). Sci. USA 100, 8886–8891 (2003). 41. Kishimoto, H. & Sprent, J. Negative selection in the 67. Bayer, A. L., Yu, A., Adeegbe, D. & Malek, T. R. Essential 20. Picca, C. C. et al. Thymocyte deletion can bias Treg thymus includes semimature T cells. J. Exp. Med. 185, role for interleukin-2 for CD4+CD25+ T regulatory cell formation toward low-abundance self-peptide. Eur. J. 263–271 (1997). development during the neonatal period. J. Exp. Med. Immunol. 39, 3301–3306 (2009). 42. Le Borgne, M. et al. The impact of negative selection 201, 769–777 (2005). 21. Legoux, F. P. et al. CD4+ T cell tolerance to on thymocyte migration in the medulla. Nat. Immunol. 68. Weist, B. M., Kurd, N., Boussier, J., Chan, S. W. tissue-restricted self antigens is mediated by 10, 823–830 (2009). & Robey, E. A. Thymic regulatory T cell niche size is antigen-specific regulatory T cells rather than deletion. 43. Ueda, Y. et al. Mst1 regulates integrin-dependent dictated by limiting IL-2 from antigen-bearing dendritic Immunity 43, 896–908 (2015). thymocyte trafficking and antigen recognition in the cells and feedback competition. Nat. Immunol. 16, 22. Malhotra, D. et al. Tolerance is established in polyclonal thymus. Nat. Commun. 3, 1098 (2012). 635–641 (2015). CD4+ T cells by distinct mechanisms, according to 44. Dzhagalov, I. L., Chen, K. G., Herzmark, P. & Robey, E. A. 69. Owen, D. L. et al. Identification of cellular sources of self-peptide expression patterns. Nat. Immunol. 17, Elimination of self-reactive T cells in the thymus: IL-2 needed for regulatory T cell development and 187–195 (2016). a timeline for negative selection. PLOS Biol. 11, homeostasis. J. Immunol. 200, 3926–3933 (2018). References 21 and 22 indicate that in the polyclonal e1001566 (2013). 70. Boursalian, T. E., Golob, J., Soper, D. M., Cooper, C. J. repertoire, a TRA-like pattern of antigen expression 45. Sauer, S. et al. T cell receptor signaling controls Foxp3 & Fink, P. J. Continued maturation of thymic emigrants

favours Treg cell induction over deletion, which reveals expression via PI3K, Akt, and mTOR. Proc. Natl Acad. in the periphery. Nat. Immunol. 5, 418–425 (2004). a correlation between self-antigen expression Sci. USA 105, 7797–7802 (2008). 71. Yang-Snyder, J. A. & Rothenberg, E. V. Spontaneous patterns and the mode of CD4+ T cell tolerance. This study uses in vitro approaches to suggest that expression of interleukin-2 in vivo in specific tissues of 23. Malchow, S. et al. Aire enforces immune tolerance by temporal aspects of TCR signalling might be crucial young mice. Dev. Immunol. 5, 223–245 (1998).

directing autoreactive T cells into the regulatory T cell for FOXP3 induction during Treg cell differentiation. 72. Wuest, S. C. et al. A role for interleukin-2 lineage. Immunity 44, 1102–1113 (2016). 46. Khailaie, S., Robert, P. A., Toker, A., Huehn, J. & trans-presentation in dendritic cell-mediated T cell 24. Perry, J. S. et al. Distinct contributions of Aire and Meyer-Hermann, M. A signal integration model of activation in humans, as revealed by daclizumab antigen-presenting-cell subsets to the generation of thymic selection and natural regulatory T cell therapy. Nat. Med. 17, 604–609 (2011). self-tolerance in the thymus. Immunity 41, 414–426 commitment. J. Immunol. 193, 5983–5996 (2014). 73. Thiault, N. et al. Peripheral regulatory T lymphocytes (2014). This is a bioinformatic study that introduces a TCR recirculating to the thymus suppress the development of 25. Yang, S., Fujikado, N., Kolodin, D., Benoist, C. & signal integration model of thymic selection, their precursors. Nat. Immunol. 16, 628–634 (2015). Mathis, D. Regulatory T cells generated early in life describing how thymocyte cell fates might be The results from this paper introduce the concept of

play a distinct role in maintaining self-tolerance. based upon the TCR stimulation history. feedback inhibition of intrathymic Treg cell differentia­

Science 348, 589–594 (2015). 47. Au-Yeung, B. B. et al. Quantitative and temporal tion by recirculating Treg cells from the periphery. 26. Anderson, M. S. et al. The cellular mechanism of Aire requirements revealed for Zap70 catalytic activity during 74. Cowan, J. E. et al. Aire controls the recirculation of control of T cell tolerance. Immunity 23, 227–239 T cell development. Nat. Immunol. 15, 687–694 murine Foxp3+ regulatory T cells back to the thymus. (2005). (2014). Eur. J. Immunol. 48, 844–854 (2018). 27. DeVoss, J. et al. Spontaneous autoimmunity prevented 48. Daniels, M. A. et al. Thymic selection threshold 75. Moran, A. E. et al. T cell receptor signal strength in by thymic expression of a single self-antigen. J. Exp. defined by compartmentalization of Ras/MAPK Treg and iNKT cell development demonstrated by a Med. 203, 2727–2735 (2006). signalling. Nature 444, 724–729 (2006). novel fluorescent reporter mouse. J. Exp. Med. 208, 28. Klein, L., Klein, T., Ruther, U. & Kyewski, B. CD4 T cell 49. Mariathasan, S. et al. Duration and strength of 1279–1289 (2011). tolerance to human C-reactive protein, an inducible extracellular signal-regulated kinase signals are 76. D’Cruz, L. M. & Klein, L. Development and function of serum protein, is mediated by medullary thymic altered during positive versus negative thymocyte agonist-induced CD25+Foxp3+ regulatory T cells in the epithelium. J. Exp. Med. 188, 5–16 (1998). selection. J. Immunol. 167, 4966–4973 (2001). absence of interleukin 2 signaling. Nat. Immunol. 6, 29. Liston, A., Lesage, S., Wilson, J., Peltonen, L. & 50. Wu, L. & Shortman, K. Heterogeneity of thymic 1152–1159 (2005). Goodnow, C. C. Aire regulates negative selection of dendritic cells. Semin. Immunol. 17, 304–312 77. Kishimoto, H. & Sprent, J. Several different cell surface organ-specific T cells. Nat. Immunol. 4, 350–354 (2005). molecules control negative selection of medullary (2003). 51. Hu, Z. et al. CCR7 modulates the generation of thymic thymocytes. J. Exp. Med. 190, 65–73 (1999). 30. Taniguchi, R. T. et al. Detection of an autoreactive regulatory T cells by altering the composition of 78. Rubin, R. L. & Hermanson, T. M. Plasticity in the T cell population within the polyclonal repertoire that the thymic dendritic cell compartment. Cell Rep. 21, positive selection of T cells: affinity of the selecting undergoes distinct autoimmune regulator (Aire)- 168–180 (2017). antigen and IL-7 affect T cell responsiveness. mediated selection. Proc. Natl Acad. Sci. USA 109, 52. Hadeiba, H. et al. Plasmacytoid dendritic cells transport Int. Immunol. 17, 959–971 (2005). 7847–7852 (2012). peripheral antigens to the thymus to promote central 79. Salomon, B. et al. B7/CD28 costimulation is 31. Kurd, N. & Robey, E. A. T cell selection in the thymus: tolerance. Immunity 36, 438–450 (2012). essential for the homeostasis of the CD4+CD25+ a spatial and temporal perspective. Immunol. Rev. 53. Yamano, T. et al. Thymic B cells are licensed to present immunoregulatory T cells that control autoimmune 271, 114–126 (2016). self antigens for central T cell tolerance induction. diabetes. Immunity 12, 431–440 (2000). 32. Petrie, H. T. & Zuniga-Pflucker, J. C. Zoned out: functional Immunity 42, 1048–1061 (2015). 80. Tai, X., Cowan, M., Feigenbaum, L. & Singer, A. CD28 mapping of stromal signaling microenvironments in the 54. Yamano, T., Steinert, M. & Klein, L. Thymic B cells costimulation of developing thymocytes induces Foxp3 thymus. Annu. Rev. Immunol. 25, 649–679 (2007). and central T cell tolerance. Front. Immunol. 6, 376 expression and regulatory T cell differentiation 33. Klein, L., Kyewski, B., Allen, P. M. & Hogquist, K. A. (2015). independently of interleukin 2. Nat. Immunol. 6, Positive and negative selection of the T cell repertoire: 55. Hinterberger, M. et al. Autonomous role of medullary 152–162 (2005). what thymocytes see (and don’t see). Nat. Rev. thymic epithelial cells in central CD4+ T cell tolerance. 81. Lio, C. W., Dodson, L. F., Deppong, C. M., Hsieh, C. S. Immunol. 14, 377–391 (2014). Nat. Immunol. 11, 512–519 (2010). & Green, J. M. CD28 facilitates the generation of 34. Ebert, P. J., Ehrlich, L. I. & Davis, M. M. Low ligand 56. Ohnmacht, C. et al. Constitutive ablation of dendritic Foxp3− cytokine responsive regulatory T cell requirement for deletion and lack of synapses in cells breaks self-tolerance of CD4 T cells and results in precursors. J. Immunol. 184, 6007–6013 (2010). positive selection enforce the gauntlet of thymic T cell spontaneous fatal autoimmunity. J. Exp. Med. 206, 82. Vang, K. B. et al. Cutting edge: CD28 and maturation. Immunity 29, 734–745 (2008). 549–559 (2009). c-Rel-dependent pathways initiate regulatory T cell 35. Peterson, D. A., DiPaolo, R. J., Kanagawa, O. & 57. van Meerwijk, J. P. et al. Quantitative impact of thymic development. J. Immunol. 184, 4074–4077 (2010). Unanue, E. R. Cutting edge: negative selection of clonal deletion on the T cell repertoire. J. Exp. Med. 83. Hinterberger, M., Wirnsberger, G. & Klein, L. B7/CD28 immature thymocytes by a few peptide-MHC 185, 377–383 (1997). in central tolerance: costimulation promotes maturation complexes: differential sensitivity of immature and 58. Malek, T. R. The biology of interleukin-2. Annu. Rev. of regulatory T cell precursors and prevents their clonal mature T cells. J. Immunol. 162, 3117–3120 (1999). Immunol. 26, 453–479 (2008). deletion. Front. Immunol. 2, 30 (2011).

Nature Reviews | Immunology Reviews

84. Murray, M. E. et al. CD28-mediated pro-survival 99. Li, M. O., Sanjabi, S. & Flavell, R. A. Transforming 115. Meredith, M., Zemmour, D., Mathis, D. & Benoist, C. signaling induces chemotherapeutic resistance in growth factor-beta controls development, Aire controls gene expression in the thymic epithelium multiple myeloma. Blood 123, 3770–3779 (2014). homeostasis, and tolerance of T cells by regulatory with ordered stochasticity. Nat. Immunol. 16, 85. Bouillet, P. et al. BH3-only Bcl-2 family member Bim is T cell-dependent and -independent mechanisms. 942–949 (2015). required for apoptosis of autoreactive thymocytes. Immunity 25, 455–471 (2006). 116. Nedjic, J., Aichinger, M., Mizushima, N. & Klein, L. Nature 415, 922–926 (2002). 100. Marie, J. C., Letterio, J. J., Gavin, M. & Rudensky, A. Y. Macroautophagy, endogenous MHC II loading and 86. Ouyang, W., Beckett, O., Ma, Q. & Li, M. O. TGF-β1 maintains suppressor function and Foxp3 T cell selection: the benefits of breaking the rules. Transforming growth factor-beta signaling curbs expression in CD4+CD25+ regulatory T cells. J. Exp. Curr. Opin. Immunol. 21, 92–97 (2009). thymic negative selection promoting regulatory T cell Med. 201, 1061–1067 (2005). 117. Bonasio, R. et al. Clonal deletion of thymocytes by development. Immunity 32, 642–653 (2010). 101. Liu, Y. et al. A critical function for TGF-β signaling in circulating dendritic cells homing to the thymus. 87. Punt, J. A., Havran, W., Abe, R., Sarin, A. & Singer, A. the development of natural CD4+CD25+Foxp3+ Nat. Immunol. 7, 1092–1100 (2006). T cell receptor (TCR)-induced death of immature regulatory T cells. Nat. Immunol. 9, 632–640 118. Perry, J. S. A. et al. CD36 mediates cell-surface CD4+CD8+ thymocytes by two distinct mechanisms (2008). antigens to promote thymic development of the differing in their requirement for CD28 costimulation: 102. Chen, W. et al. Requirement for transforming growth regulatory T cell receptor repertoire and allo-tolerance. implications for negative selection in the thymus. factor β1 in controlling T cell apoptosis. J. Exp. Med. Immunity 48, 923–936 (2018). J. Exp. Med. 186, 1911–1922 (1997). 194, 439–453 (2001). 119. Lee, H. M., Bautista, J. L., Scott-Browne, J., Mohan, J. F. 88. Dautigny, N., Le Campion, A. & Lucas, B. Timing 103. Schlenner, S. M., Weigmann, B., Ruan, Q., Chen, Y. & & Hsieh, C. S. A broad range of self-reactivity drives and casting for actors of thymic negative selection. von Boehmer, H. Smad3 binding to the foxp3 enhancer thymic regulatory T cell selection to limit responses to J. Immunol. 162, 1294–1302 (1999). is dispensable for the development of regulatory T cells self. Immunity 37, 475–486 (2012). 89. Jones, L. A., Izon, D. J., Nieland, J. D., Linsley, P. S. with the exception of the gut. J. Exp. Med. 209, 120. Simons, D. M. et al. How specificity for self-peptides & Kruisbeek, A. M. CD28–B7 interactions are not 1529–1535 (2012). shapes the development and function of regulatory required for intrathymic clonal deletion. Int. Immunol. 104. Konkel, J. E., Jin, W., Abbatiello, B., Grainger, J. R. T cells. J. Leukoc. Biol. 88, 1099–1107 (2010). 5, 503–512 (1993). & Chen, W. Thymocyte apoptosis drives the 121. Atibalentja, D. F., Murphy, K. M. & Unanue, E. R. 90. Page, D. M., Kane, L. P., Allison, J. P. & Hedrick, S. M. intrathymic generation of regulatory T cells. Proc. Natl Functional redundancy between thymic CD8α+ and Two signals are required for negative selection of Acad. Sci. USA 111, E465–E473 (2014). Sirpα+ conventional dendritic cells in presentation of CD4+CD8+ thymocytes. J. Immunol. 151, 1868–1880 105. Chen, W. & Konkel, J. E. Development of thymic Foxp3+ blood-derived lysozyme by MHC class II proteins. (1993). regulatory T cells: TGF-β matters. Eur. J. Immunol. 45, J. Immunol. 186, 1421–1431 (2011). 91. Tan, R., Teh, S. J., Ledbetter, J. A., Linsley, P. S. & 958–965 (2015). 122. Feuerer, M. et al. Enhanced thymic selection of Teh, H. S. B7 costimulates proliferation of CD4–8+ 106. Anderson, M. S. et al. Projection of an immunological FoxP3+ regulatory T cells in the NOD mouse model T lymphocytes but is not required for the deletion of self shadow within the thymus by the Aire protein. of autoimmune diabetes. Proc. Natl Acad. Sci. USA immature CD4+8+ thymocytes. J. Immunol. 149, Science 298, 1395–1401 (2002). 104, 18181–18186 (2007). 3217–3224 (1992). 107. Derbinski, J., Schulte, A., Kyewski, B. & Klein, L. 123. Relland, L. M. et al. Affinity-based selection of regulatory 92. Walunas, T. L., Sperling, A. I., Khattri, R., Thompson, C. B. Promiscuous gene expression in medullary thymic T cells occurs independent of agonist-mediated induction & Bluestone, J. A. CD28 expression is not essential for epithelial cells mirrors the peripheral self. Nat. of Foxp3 expression. J. Immunol. 182, 1341–1350 positive and negative selection of thymocytes or Immunol. 2, 1032–1039 (2001). (2009). peripheral T cell tolerance. J. Immunol. 156, 108. Mathis, D. & Benoist, C. A decade of AIRE. Nat. Rev. 1006–1013 (1996). Immunol. 7, 645–650 (2007). Acknowledgements 93. Pobezinsky, L. A. et al. Clonal deletion and the fate of 109. Peterson, P., Org, T. & Rebane, A. Transcriptional L.K. receives support from the European Research Council autoreactive thymocytes that survive negative regulation by AIRE: molecular mechanisms of (ERC-2016-ADG 742290 - TOLERANCE FOOTPRINT) and the selection. Nat. Immunol. 13, 569–578 (2012). central tolerance. Nat. Rev. Immunol. 8, 948–957 Deutsche Forschungsgemeinschaft (CRC 1054 ‘Control and This study reconciles conflicting data on the role of (2008). plasticity of cell fate decisions in the immune system’). E.A.R. is CD28 in clonal deletion. 110. Derbinski, J. et al. Promiscuous gene expression in supported by grants from the US National Institutes of Health 94. Collette, Y., Benziane, A., Razanajaona, D. & Olive, D. thymic epithelial cells is regulated at multiple levels. (NIH AI064227 and AI065537). C.-S.H. is supported by the US Distinct regulation of T cell death by CD28 J. Exp. Med. 202, 33–45 (2005). National Institutes of Health (NIH AI079187 and AI131349). depending on both its aggregation and T cell 111. Takaba, H. et al. Fezf2 orchestrates a thymic program receptor triggering: a role for Fas–FasL. Blood 92, of self-antigen expression for immune tolerance. Cell Author contributions 1350–1363 (1998). 163, 975–987 (2015). The authors contributed equally to all aspects of the article. 95. Li, M. O. & Flavell, R. A. TGF-beta: a master of all 112. Sansom, S. N. et al. Population and single-cell genomics T cell trades. Cell 134, 392–404 (2008). reveal the Aire dependency, relief from Polycomb Competing interests 96. Chen, W. et al. Conversion of peripheral CD4+CD25− silencing, and distribution of self-antigen expression in The authors declare no competing interests. naive T cells to CD4+CD25+ regulatory T cells by TGF-β thymic epithelia. Genome. Res. 24, 1918–1931 (2014). induction of transcription factor Foxp3. J. Exp. Med. 113. Smith, K. M., Olson, D. C., Hirose, R. & Hanahan, D. Publisher’s note 198, 1875–1886 (2003). Pancreatic gene expression in rare cells of thymic Springer Nature remains neutral with regard to jurisdictional 97. Kretschmer, K. et al. Inducing and expanding regulatory medulla: evidence for functional contribution to T cell claims in published maps and institutional affiliations. T cell populations by foreign antigen. Nat. Immunol. 6, tolerance. Int. Immunol. 9, 1355–1365 (1997). 1219–1227 (2005). 114. Brennecke, P. et al. Single-cell transcriptome analysis Reviewer information 98. Zheng, Y. et al. Role of conserved non-coding DNA reveals coordinated ectopic gene-expression patterns Nature Reviews Immunology thanks T. Malek, P. Marrack and elements in the Foxp3 gene in regulatory T cell fate. in medullary thymic epithelial cells. Nat. Immunol. 16, other anonymous reviewer(s) for their contribution to the Nature 463, 808–812 (2010). 933–941 (2015). peer review of this work.

www.nature.com/nri