The T cell costimulatory molecule CD28 plays an important role in the thymic generation of Foxp3+ regulatory T cells (Tregs) essential for the maintenance of self-tolerance. In this study, we show that a cell-intrinsic signal from CD28 is involved in the generation of cytokine-responsive Foxp3− precursors using studies of mixed bone marrow chimeras as well as TCR-specific generation of Foxp3+ cells using intrathymic transfer of TCR-transgenic thymocytes expressing a natural Treg TCR. Contrary to a previous report, the analysis of CD28 mutant knockin mice revealed that this cell-intrinsic signal is only partially dependent on the Lck-binding PYAP motif. Surprisingly, even though the absence of CD28 resulted in a 6-fold decrease in thymic Tregs, the TCR repertoires of CD28-deficient and sufficient cells were largely overlapping. Thus, these data suggest that CD28 does not operate by markedly enlarging the repertoire of TCRs available for Treg development, but rather by improving the efficiency of Treg development of thymocytes expressing natural Treg TCRs.
Foxp3+ CD4+ regulatory T cells (Tregs) are essential for the maintenance of self-tolerance, as mice that are defective in the development or function of these cells develop spontaneous autoimmune disease (1, 2). The signals that lead to the development of uncommitted thymocytes to Tregs have been the subject of substantial research. The seminal discovery that the expression of cognate Ag in the thymus could drive Treg development in TCR-transgenic mice led to the hypothesis that Tregs develop due to interactions with self-Ags at an avidity window between negative and positive selection (3–5). This was further supported by studies demonstrating that the Treg and non-Treg TCR repertoires mostly differed (6, 7). Recently, the use of Treg TCR-transgenic mice revealed that TCR-specific natural Treg development is often restricted by a small developmental niche (8, 9). Thus, TCR-derived signals are important for thymic Treg differentiation.
Although TCR activation is essential for the selection of thymocytes into the Treg subset, additional signals are also important. In particular, costimulation by CD28 is required for efficient Treg development, as mice deficient in CD28, or its ligands CD80/CD86, have a dramatic reduction of thymic and peripheral Treg numbers (10–13). Although one function of CD28 is augmentation of IL-2 secretion, a potentially cell-extrinsic mechanism, the presence of normal thymocytes in mixed bone marrow chimeras was unable to rescue Treg differentiation in CD28 knockout (KO) cells (11). These data suggest that CD28 primarily regulates Treg development via a cell-intrinsic mechanism. Consistent with this observation, the frequency of Foxp3+ cells in hyperactive Stat5 (Stat5CA) transgenic CD28 KO mice was markedly lower than that in Stat5CA-transgenic mice (14, 15), suggesting that enhanced cytokine signaling can only partially complement a deficiency in CD28 for thymic Treg development.
Recently, we and others proposed that thymic Treg development can be divided into at least two discrete steps (15–18). Consistent with studies demonstrating that CD25 and glucocorticoid-induced TNFR can be upregulated in a Foxp3-independent manner during thymic Treg development (19–21), we observed that the CD25hiGITRhiFoxp3− CD4+CD8− subset is enriched in Treg precursors (16). Characterization of these cells suggested a model for thymic Treg development in which TCR-derived signals lead to the development of a cytokine-responsive Treg precursor, which then responds to signals from IL-2 or IL-15 for the induction of Foxp3. Consistent with this model, Stat5CA diminished the requirement for TCR specificity in Treg generation (15). Visualizing the development of Tregs after intrathymic injection of TCR-transgenic cells into cognate Ag-expressing hosts was also consistent with this model (17). Taken together, these data suggest that thymic Treg development is a multistep process involving signals from TCR, cytokines, and other receptors, including CD28.
Although CD28 is important in thymic Treg development, it is unclear whether it acts before or after the generation of Foxp3− Treg precursors. Note that we use “Treg precursors” in a different manner than in a recent study examining the role of CD28 and Lck in stabilizing Foxp3 mRNA, which used that term to refer to thymic Foxp3+ cells as precursors to peripheral Tregs (22). CD28 may regulate Treg development by promoting cell-extrinsic IL-2 production and/or generating cell-intrinsic signals. Costimulation might facilitate the generation of Tregs by increasing the aggregate TCR signal, thereby recruiting thymocytes with lower TCR avidity to self-Ags into the Treg subset. Conversely, CD28 could provide a signal that increases the efficiency of Treg development of thymocytes expressing the natural repertoire of Treg TCRs. Thus, the process by which CD28 promotes Treg development is currently unknown.
Biochemically, the cell-intrinsic effect is primarily mediated by the C-terminal PYAP motif, as expression of a mutant CD28 transgene in which this motif is mutated failed to restore Treg development in CD28-deficient mice (11). However, previous studies of the structural basis for CD28 have been plagued by inconsistencies and conflicting results, in part due to variable levels of expression and the use of heterologous promoter systems (23). As CD28 expression varies throughout thymic development and as a consequence of T cell activation, faithful recapitulation of expression levels is essential to accurately determine the role played by CD28 in the development of Tregs (24).
To address these issues, we used targeted knockin mice expressing mutations of CD28 to definitively address the role of specific CD28 signaling motifs in Treg development. We found that the Lck-binding PYAP motif only partially accounts for CD28 signals, whereas the PI3K-binding Y170 motif is dispensable. We also determined that cell-intrinsic CD28 signals are required for the efficient generation of Foxp3− Treg precursors. Unexpectedly, analysis of the Treg TCR repertoire suggests that CD28 acts mostly by increasing the efficiency of TCR-instructed Treg selection rather than by dramatically expanding the TCR repertoire capable of Treg development.
Materials and Methods
C57BL/6.SJL (CD45.1) mice were purchased from the National Cancer Institute (Frederick, MD). CD28 KO mice (25) were generated by C. Thompson (University of Pennsylvania, Philadelphia, PA). Homozygous CD28-AYAA and -Y170F mice were previously described (23, 26). Foxp3gfp reporter mice (27) were provided by A. Rudensky (Memorial Sloan-Kettering, New York, NY). Mice were housed in a specific pathogen-free facility at Washington University and used under protocols approved by the Institutional Animal Care and Use Committee.
Fluorescently conjugated mAbs were purchased from eBioscience (San Diego, CA), Biolegend (San Diego, CA), and BD Biosciences (San Jose, CA). Human IL-2 (hIL-2) was obtained from the National Cancer Institute Biological Resources Branch (Frederick, MD) preclinical repository. Murine IL-7 was purchased from PeproTech (Rocky Hill, NJ).
CD25hiFoxp3− (5 × 103) or Foxp3− (1 × 105) CD4 single-positive (SP) cells were FACS purified and cultured with hIL-2 (50 U/ml) in the presence of murine IL-7 (5 ng/ml). Foxp3gfp expression was analyzed by flow cytometry at 24 h.
Bone marrow chimeras
A total of 3–5 × 106 red cell-lysed bone marrow cells were i.v. transferred into lethally irradiated hosts (10.5 Gy). Mice were analyzed 6–8 wk after reconstitution.
G113 TCR-transgenic thymocytes largely devoid of Tregs were mixed with wild-type (WT) Thy1.1+ cells to allow assessment of injection efficiency, labeled with DDAO-SE (Invitrogen, Carlsbad, CA), and intrathymically injected (8). Induction of Foxp3 in the donor cells was assessed by flow cytometry of the entire thymus.
TCR repertoire analysis
TRAV14 TCRα sequences from thymic Foxp3+ and Foxp3− Vβ6+ CD4SP thymocytes isolated from 6–10-wk-old mice were obtained as previously described (7). As before, the CDR3 amino acid sequence was used as a unique identifier for individual TCRs. Three of the WT thymus CD4SP data sets have been published (16).
Two-tailed Student t test was used for calculation of statistical significance unless otherwise stated. Estimation of repertoire diversity by the abundance coverage estimator was obtained using EstimateS (version 7.5; R.K. Colwell, http://viceroy.eeb.uconn.edu/estimates).
The CD28 Lck-binding PYAP motif is involved in thymic Treg development
A previous report that used transgenic expression of various mutated CD28 constructs in a CD28-deficient background suggested that the Lck-binding PYAP motif is important for the function of CD28 in Treg development (11). However, transgenic expression may not faithfully recapitulate the endogenous expression pattern of CD28 as it is developmentally regulated (Supplemental Fig. 1A) (24). Because studies of gene dosage suggested that the level of CD28 affects its function (23), we assessed the role of the PYAP motif using CD28-AYAA knockin mice, which have targeted P187A and P190A mutations in the cytoplasmic tail (23). In addition, we analyzed the CD28-Y170F mutant, as PI3K-induced Akt can negatively regulate Treg development (26, 28). As expected, the mutant knockin mice express CD28 at comparable levels to that of WT control (Supplemental Fig. 1B).
The CD28 KO, AYAA, and Y170F strains were bred to the Foxp3gfp reporter for analysis of Treg development (16, 27). Overall CD4 and CD8 thymic development appeared normal in both CD28 mutant mice (Fig. 1A, Table I). Flow cytometric analysis of the CD4SP subset revealed that the percentage of Foxp3+ cells was normal in Y170F mice (3.1 ± 0.5%), but decreased 2-fold in AYAA mice (1.4 ± 0.3%; p < 0.01) (Fig. 1B, 1C). However, contrary to the previous report showing that the mutation of the PYAP motif completely abolishes the function of CD28 for Treg development (11), we found that the frequency of Tregs was significantly higher in AYAA compared with CD28 KO mice (1.4 ± 0.2% versus 0.6 ± 0.1%; p < 0.01). These results imply that in addition to the PYAP motif, other sequences in the tail of CD28 contribute to thymic Treg development.
A role for CD28 in the generation of cytokine-responsive Treg precursors
Previously, we proposed that thymic Treg development can be divided into TCR-dependent and -independent phases (16). CD28 could act coordinately with TCR in the first phase or facilitate production of IL-2 for the latter step. To address this, we examined CD28 KO and mutant mice for the frequency of CD25hiFoxp3− CD4SP cells, which are enriched in Treg precursors (16). Flow cytometric analysis showed that the CD25hiFoxp3− CD4SP subset was decreased in CD28 KO mice (2.1 ± 0.3% versus 2.8 ± 0.4% in WT; p < 0.01), whereas the frequency of the subset in Y170F and AYAA mice was comparable to that in WT mice (Fig. 1B, 1C). Consistent with the decreased number of CD25hiFoxp3− cells present in CD28 KO mice, there was also a reduction in the CD122hiCD25hiFoxp3− subset (Fig. 1D), which expresses the trimeric IL-2 receptor and contains the highest frequency of Treg precursors in normal mice (16). However, the cells that complete Treg development and express Foxp3 in CD28 KO mice express WT levels of CD122 and CD25 (Fig. 1D, Supplemental Fig. 2). In contrast to CD28 KO mice, the percentage of CD122hi cells in the CD25hiFoxp3− subset in AYAA mice was comparable to that of WT mice (Fig. 1D). Also, we did not detect any difference in the level of Foxp3gfp reporter expression in the KO and mutant mice (Supplemental Fig. 2), which is not consistent with a recent report associating the PYAP motif with stability of Foxp3 mRNA in thymic Tregs (22). Altogether, these data demonstrate that CD28 is involved in the upregulation of IL-2R chains via signaling motifs other than Y170 or PYAP.
Although the CD25hiFoxp3− cell subset is enriched in Treg precursors in normal mice, it was possible that CD28 could have differential effects on CD25 upregulation versus Treg development, thus affecting the enumeration of Treg precursors in CD28 KO mice. For example, CD28 could be more important for Treg precursor generation than CD25 upregulation. Conversely, the decreased percentage of CD25hi cells in the CD28 KO may reflect the dependence of CD25 expression on CD28 (25) rather than reflecting the percentage of Treg precursors. We therefore decided not to use CD25 to subset CD4SP cells and instead functionally assessed the frequency of cytokine-responsive Treg precursors in the entire Foxp3− subset after IL-2 stimulation in vitro. We found that the percentage of cytokine-responsive Treg precursors in CD28 KO mice was decreased by half comparing to WT control (Fig. 2A). Mutation of the PYAP motif also reduced the precursor population by one-fourth. By contrast, the Y170F mutation had no effect on the Treg precursor (Fig. 2A). These results suggest that the PYAP motif only partially accounts for the involvement of CD28 in the generation of Treg precursors and does so via a mechanism independent of IL-2R upregulation.
To address whether CD28 facilitated the generation of Treg precursors via a cell-intrinsic mechanism, we analyzed bone marrow chimeras generated with CD45.2 CD28 KO or AYAA bone marrow cells mixed in a 1:1 ratio with CD45.1 WT bone marrow. This approach has been used to provide a more normal cytokine environment during thymic Treg development (11). Consistent with previous results using CD25 as a Treg marker (11), we found that the generation of thymic Tregs was not restored in CD28 KO cells despite the presence of WT thymocytes (Fig. 2B, 2C). However, we observed only a partial defect in Treg frequency in AYAA mutant cells (Fig. 2B, 2C) rather than a complete defect. We also analyzed the generation of Treg precursors by purifying CD45.2 Foxp3− CD4SP donor cells and culturing them with IL-2 in vitro. We found that the presence of WT thymocytes made essentially no difference in the generation of cytokine-responsive AYAA or CD28 KO Treg precursors (Fig. 2A, 2D), even though Treg development of WT thymocytes appeared normal in the chimeras (Supplemental Fig. 3). These data therefore suggest that CD28 delivers a cell-intrinsic signal involved in the generation of Treg precursors, as it cannot be compensated by the presence of normal thymocytes.
CD28 ligands are required for efficient de novo Treg development
The previous set of experiments examined Treg development during the steady state. To evaluate de novo Treg generation, we used TCR-transgenic line expressing the naturally arising Treg TCR G113 (8). As the Treg developmental niche for this TCR is quite small, G113 αβ-TCR–transgenic thymocytes contain a very low frequency of Foxp3+ cells. Intrathymic transfer of the G113 thymocyte population can therefore be used to study Treg development (8). Three days after intrathymic transfer into WT recipients, G113 Foxp3+ cells can be readily found (8.3 ± 4.2%; Fig. 3). However, the lack of CD28 ligands CD80/CD86 in the recipients inhibited the development of Foxp3+ cells 16-fold, with only a small frequency of Foxp3+ cells observed (0.5 ± 0.4%, p < 0.01; Fig. 3). This appeared to be related to cell-intrinsic CD28-derived signals rather than an environmental effect, as Treg development was not inhibited and perhaps slightly enhanced posttransfer of G113 thymocytes into CD28 KO recipients. Thus, these data suggest that CD28 interaction with B7 is required for the efficient de novo generation of Ag-specific Tregs.
CD28 signaling does not markedly increase the number of TCRs capable of facilitating Treg development
The above analysis of monoclonal T cell populations demonstrated that CD28-derived signals were important for G113-TCR–mediated Treg development. However, it remained possible that other TCRs would exhibit variable dependencies on CD28 signals for Treg development. One hypothesis is that CD28 signaling directly contributes to the overall TCR signaling strength, thereby greatly expanding the repertoire of TCRs that undergo Treg development, resulting in the increased frequency of Tregs in WT versus CD28 KO mice. An alternative, non-mutually exclusive possibility is that TCR alone dictates development into Foxp3+ cells and that CD28 provides a parallel signal to improve the efficiency of this process. The former hypothesis would therefore predict that the Treg TCR repertoire in CD28 KO and WT mice would differ greatly, whereas the latter would show strong similarity in the TCR repertoires. Although there are some contested data suggesting that CD28 facilitates negative selection in TCR-transgenic models (11, 29), it remains largely unknown whether CD28 alters the TCR repertoire.
Because the TCR repertoire of fully polyclonal mice is too diverse for experimental analysis at the individual TCR level, we analyzed the thymic TRAV14 (Vα2) TCR repertoires of CD28 KO and WT mice in a fixed TCRβ model (7). Similar to the fully polyclonal setting, TCliβ TCR-transgenic Foxp3gfp Tcra+/− (TCliβ-tg) mice on a CD28 KO background have decreased frequencies of thymic CD4SP Foxp3+ cells (Supplemental Fig. 4A). Furthermore, we did not observe skewing of Vα2 usage in CD28 KO and WT TCliβ-tg mice (Supplemental Fig. 4B). Thus, the TRAV14 (Vα2) subset in TCliβ-tg mice appeared suitable for evaluating the role of CD28 on the thymic Treg TCR repertoire.
Thymic CD4SP Foxp3+ and Foxp3− cells were purified by flow cytometry, and a total of five and six data sets were obtained from WT and KO TCliβ-tg mice, respectively (Supplemental Table I). Initial analysis showed that CD28 deficiency resulted in a decreased TCR diversity in the Foxp3+ subset, which may contribute to the decreased number of Tregs in the CD28 KO (Supplemental Table I). Consistent with the similarity in Foxp3− CD4SP numbers between CD28 KO and WT mice, virtually all of the top 10 Foxp3− TCRs in the CD28 KO data set could be found in the WT data set (Supplemental Fig. 5), suggesting that CD28 has no major impact on the Foxp3− TCR repertoire. As expected, the top 10 TCRs of Foxp3− and Foxp3+ subsets are largely distinct in both CD28KO and WT mice, similar to previously published results (Fig. 4A, 4B) (7).
We next analyzed the effect of CD28 deficiency on the Treg TCR repertoires. The Treg TCR repertoire in CD28 KO mice appeared to more oligoclonal, as the top 10 TCRs comprised a greater portion of the total TCRs sequenced. Also, the individual frequencies of the top 10 Foxp3+ TCRs often differed between the CD28KO and WT data sets (Fig. 4C, 4D, Supplemental Fig. 6), suggesting that CD28 signaling has variable effects on Treg development depending on the particular TCR specificity. Despite these observed differences between the CD28 KO and WT Treg TCR repertoires, almost all of the top 10 Treg TCRs in one data set could be found in the other. Thus, it appeared unlikely that CD28 costimulation enlarges the Treg TCR repertoire sufficiently to account for a 6-fold increase in Treg numbers (Fig. 1, Supplemental Fig. 4).
To obtain greater perspective regarding the similarity between the CD28 KO and WT mice, the frequencies of all TCRs greater than an average frequency of 0.25% in the KO or WT data sets were plotted in two dimensions (Fig. 5). This threshold frequency was chosen as 0.23% represents the minimal population frequency of a TCR that would be found in the data set at least once with 90% confidence, as calculated using the binomial distribution with a sample size of 1000. These data revealed that most of these Treg TCRs were present in both CD28 KO and WT mice. Remarkably, the overall pattern of TCR usage appeared similar regardless of whether Treg or Foxp3− TCRs were plotted. Although the overall pattern was similar between Treg and Foxp3− TCRs, we did observe 10 Treg TCRs that were found only in the CD28 KO and 4 only in the WT data sets (Fig. 5). By contrast, only one Foxp3− TCR was found exclusively in either the CD28 KO or WT data set. Because the use of the average TCR frequency between CD28 KO and WT data sets could favor overlapping TCRs, we also verified these observations using a threshold of 0.5% in either the CD28 KO or WT data sets (Supplemental Fig. 7). Although infrequent TCRs cannot be easily analyzed via this approach, one interpretation of these results is that these Treg TCRs found in only the CD28 KO or WT data sets represent a small repertoire shift such that a few TCRs undergo negative selection in the presence of CD28 (points on the left edge) and are thus not found in the WT data set and that a few others cannot facilitate Treg selection without CD28 costimulation (points on the bottom edge).
Taken together with statistical estimates of TCR diversity (Supplemental Table I), these data suggest that the increased diversity of the Treg TCR repertoire from CD28 costimulation is primarily due to the inclusion of infrequent TCRs. Although these infrequent TCRs can substantially increase the diversity of the Treg TCR repertoire, CD28-mediated recruitment of cells expressing these TCRs cannot, however, account for the 6-fold greater Treg number found in CD28 WT versus KO mice. In fact, it appears that some of these TCRs are inefficient at inducing Treg development, thereby requiring costimulation by CD28 (Supplemental Fig. 8). Thus, although CD28 increases the pool of TCRs available for Treg selection, this does not appear to be the major mechanism by which CD28 facilitates the generation of the thymic Treg population.
In this report, we examined the role of CD28 in thymic Treg development. Our results demonstrate that the number of Treg precursors is substantially decreased in the absence of cell-intrinsic signals provided by CD28 (Fig. 2). In contrast to previous reports, these CD28-derived signals are only partially dependent on the PYAP motif, as AYAA mutant knockin mice have intermediate frequencies of both Treg precursors and mature Tregs in comparison with CD28 KO mice (Figs. 1, 2). Finally, we infer that CD28 signaling primarily improves the efficiency of Treg development of thymocytes with Treg-selecting TCRs rather than markedly increases the repertoire of TCRs available for Treg development.
The identification that CD28 is involved in the generation of Treg precursors is perhaps not surprising, as CD28 signaling is thought to be concomitant with TCR engagement. However, it remains difficult to determine whether CD28 is only involved in the generation of Treg precursors or is also needed at latter stages of Treg development. First, the reduction of Tregs in CD28 KO mice is greater than that observed for cytokine-responsive Treg precursors (Figs. 1, 2). One potential explanation for this is that the in vitro assay, although useful for identifying cells that can potentially become Foxp3+, overrepresents the frequency of cells that become Foxp3+ after intrathymic transfer (16). It is therefore possible that CD28 KO Treg precursors progress less efficiently to Tregs in vivo than their WT counterparts, which could be explained by decreased levels of IL-2 in CD28 KO mice. However, we were unable to discern a clear requirement for environmental cytokines using mixed CD28 KO and WT bone marrow chimeras. In fact, we found that although the frequency of cytokine-responsive Treg precursors in these chimeras remained constant (Fig. 2), the percentage of CD28 KO Tregs was actually one-third lower in the presence of WT cells than in CD28 KO mice (Figs. 1, 2). This observation was not previously reported (11), perhaps related to the use of CD25 rather than Foxp3gfp as a Treg marker. One interpretation of these results is that CD28, in addition to facilitating the generation of Treg precursors themselves, also improves the competitive fitness of Tregs or their precursors for cytokines or other factors during the generation of the Treg population.
Multiple signaling molecules can bind to CD28, including kinases PI3K and Lck that bind to YMNM and PYAP motifs, respectively. Our data show that the Lck-binding motif is only partially involved in the generation of Treg precursors, whereas the PI3K-binding motif is dispensable (Fig. 1). This result suggests that prolonged or enhanced signaling by CD28-associated Lck is important for the development of Treg precursors. These data are consistent with a previous study that shows that the Lck-binding motif is important for the thymic Treg generation and also T cell activation in an autoimmunity model caused by CTLA-4 deficiency (11, 30). However, the decrease in Treg number is less than previously reported. One potential explanation is differences in CD28 expression between the two systems (23). Because the mutation of the PYAP motif only partially recapitulates the defect in Treg development observed in CD28 KO mice, this implies that additional signaling motifs are involved.
Thymic Treg selection is commonly attributed to TCRs with intermediate affinity to self-Ags in an avidity window between positive and negative selection (18). As CD28 signals are thought to be coincident with TCR-derived signals (31), we hypothesized that the loss of CD28 would decrease the overall TCR signal, allowing only the highest affinity TCRs the ability to induce Treg development. The analysis of the Foxp3+ TCR repertoire of CD28 KO showed a lower diversity compared with that of WT. This may, in part, be due to the 6-fold decrease in the number of Tregs in CD28 KO mice. Furthermore, individual TCRs exhibited variable dependence on CD28 signaling, with some TCRs found at higher or lower frequency in CD28 KO compared with WT mice. The variable importance of CD28 may reflect the level of CD80/CD86 among the various APCs capable of mediating Treg selection, such as cortical thymic epithelial cells, medullary thymic epithelial cells, and bone marrow-derived dendritic cells. However, we did not observe a large number of Treg TCR sequences unique to the CD28 WT subset, suggesting that CD28 signaling itself does not dramatically increase the pool of TCRs available for Treg selection and thereby cannot account for the increased Treg number in CD28 WT mice relative to CD28 KO mice (Fig. 5). Although the molecular mechanism of CD28 in Treg development remains to be clearly defined, these data imply that Lck and other signals do not primarily enhance the TCR signals relevant for crossing the avidity threshold for Treg selection, but rather provide additional parallel signals that facilitate Treg development.
We thank N. Santacruz and J. Hunn for expert technical assistance.
Disclosures The authors have no financial conflicts of interest.
This work was supported by the Arthritis Foundation (to C.-S.H.), the Burroughs Wellcome Fund (to C.-S.H.), the National Institutes of Health (to C.-S.H. and J.M.G.), and the Tertiary Education Services Office, Macau Special Administrative Region (C.-W.J.L.).
The online version of this article contains supplemental material.
Abbreviations used in this paper:
- human IL-2
- regulatory T cell
- Received January 11, 2010.
- Accepted March 23, 2010.
- Copyright © 2010 by The American Association of Immunologists, Inc.