TCR ligand density and affinity determine peripheral induction of Foxp3 in vivo

To explore the impact of TCR/pMHC affinity on peripheral Foxp3 induction we used 5C.C7 T cells, which recognize a peptide from moth cytochrome c (MCC; 88–103) in the context of the MHCII molecule I-E^k (Davis et al., 1998). The 5C.C7 TCR is ideal for addressing this question because, in addition to the natural ligand MCC, a wide variety of related peptide ligands have been characterized for their TCR binding properties and in vitro activating potencies (Reay et al., 1994; Rabinowitz et al., 1996; Wilson et al., 1999; Li et al., 2004; Krogsgaard et al., 2005). In addition to MCC, we assessed Foxp3 induction by the weak agonist peptide 102S (Reay et al., 1994; Rabinowitz et al., 1996) and the superagonist K5 (Krogsgaard et al., 2003, 2005; Li et al., 2004), both of which contain mutations in TCR contact residues that result in a shorter or longer half-life of 5C.C7 TCR–pMHC interactions, respectively (Corse et al., 2010; Huppa et al., 2010). All of the peptides used in our study bind to MHC with comparable affinity (Krogsgaard et al., 2003), which is essential to separate the influences of TCR ligand potency versus density of pMHC ligand upon T cell responses. To confirm the rank order of the peptides in our experimental setting, we stimulated 5C.C7 T cells with irradiated splenocytes and the indicated titration of peptide (Fig. 1 A). As expected, at equivalent doses of peptide 102S induced less 5C.C7 proliferation and K5 more proliferation, relative to the natural ligand MCC. This difference in in vitro potency is apparent in the higher and lower EC₅₀ values for 102S and K5, respectively (Fig. 1 A).

To assess the ability of the various ligands to induce Foxp3 in vitro, 5C.C7 T cells were stimulated in the presence of exogenous IL-2 and a titration of peptide. Interestingly, to reach peak in vitro Foxp3 induction for each ligand, the peptide concentration had to be adjusted to compensate for affinity. Relative to MCC, a higher concentration of 102S and a lower concentration of K5 was required (Fig. 1 B). For all three peptides, the optimal concentration for in vitro Foxp3 induction is below the concentration of peptide required for expansion of the Foxp3-negative compartment within the same cultures (Fig. 1 C). These results are consistent with a recently published study, comparing in vitro Foxp3 induction and proliferation in response to two peptides differing in their EC₅₀ values (Turner et al., 2009), and with in vivo data indicating that Foxp3 induction is favored by conditions suboptimal for proliferation of responding cells (Kretschmer et al., 2005).

The addition of exogenous TGF-β, a factor known to facilitate TCR-dependent Foxp3 induction (Chen et al., 2003; Selvaraj and Geiger, 2007), widened the range of TCR stimulation over which Foxp3 induction occurred in vitro (Fig. S1). Furthermore, the superagonist K5 resulted in the most Foxp3⁺ cells, peaking at concentrations of peptide higher than favorable for Foxp3 induction in the absence of exogenous TGF-β. Thus, in terms of Foxp3 induction, TGF-β reduces the sensitivity to strong stimulation through the TCR. This is consistent with the requirement of TGF-β and other negative regulators for Foxp3 induction in systems relying on relatively strong TCR stimulation (Chen et al., 2003; Zheng et al., 2006; Selvaraj and Geiger, 2007). Our in vitro data suggest that the combination of ligand density and affinity determine an optimal range of TCR stimulation conducive for expression of Foxp3. Sensitivity to variations in these parameters is decreased in the presence of exogenous TGF-β.

To dissect the effects of antigen density and affinity on Foxp3 induction in vivo, we opted for a single intravenous dose of peptide, which has previously been shown to tolerize a monoclonal population of antigen-specific T cells (Thorstenson and Khoruts, 2001). Low doses of MCC peptide were able to induce Foxp3 expression in adoptively transferred 5C.C7 T cells (Fig. 2 A). Over a 100-fold titration of peptide, we found 0.1 µg MCC to be optimal for peripheral Foxp3 induction, both in terms of the percentage of 5C.C7 expressing Foxp3 and the number of 5C.C7 Foxp3⁺ cells (Fig. 2 B). This dose was below the threshold of activation required for complete CD44 up-regulation of the 5C.C7 Foxp3⁻ cells (Fig. 2 C) and resulted in only partial CFSE dilution (Fig. 2 D). This suggests that in our system, induction of Foxp3 expression is coincident with a weaker TCR signal (in terms of a lower peptide dose). A time course after injection of the optimal dose showed that although Foxp3 can be detected in some 5C.C7 T cells as early as 2.5 d after immunization, peak accumulation of Foxp3⁺ 5C.C7 T cells occurs between 4 and 10 d after immunization (Fig. 2 D). 5C.C7 T cells induced to express Foxp3 under our experimental conditions express high levels of CD25 and CTLA-4 and display phenotypic characteristics common to T reg cells (Fig. S2). Consistent with previous studies (Kretschmer et al., 2005; Josefowicz et al., 2009), these Foxp3⁺ 5C.C7 T cells had undergone only a few cell divisions, with the majority remaining undivided at early time points (Fig. 2 D, days 2.5 and 4; and Fig. S2 D).

We addressed the influence of TCR/pMHC affinity on in vivo Foxp3 induction by comparing varying doses of the weak agonist 102S and the superagonist K5 with MCC 8 d after peptide injection (Fig. 3). All three peptides, including the weak agonist 102S, are able to stimulate canonical in vivo T cell responses, including proliferation, cytokine production, and memory, upon subcutaneous immunization with LPS (Corse et al., 2010). In response to intravenous peptide injection, we found K5 to induce comparable numbers of Foxp3⁺ 5C.C7 T cells as MCC at the same low doses of peptide (Fig. 3 B). Given results that low levels of TCR stimulation favor Foxp3 induction (Kretschmer et al., 2005; Haxhinasto et al., 2008; Sauer et al., 2008; Josefowicz et al., 2009), we were surprised to find that the weak agonist 102S appeared diminished in its ability to induce expression of Foxp3. Although intravenous administration of 102S peptide did result in some Foxp3 expression, the percentage of 5C.C7 expressing Foxp3 was fourfold less than that observed with the optimal dose of MCC 8 d after peptide injection (Fig. 3, A and C). Furthermore, normalization to remove the effect of the Foxp3⁻ 5C.C7 T cells from the calculation revealed a drastically smaller number Foxp3⁺ 5C.C7 in the 102S-treated mice, compared with both high-affinity ligands (Fig. 3 B). We detected even less Foxp3 induction upon further increase of the dose to 100 µg (Fig. 3, A and B). This indicates that optimal Foxp3 induction by the 102S ligand cannot be accomplished by further increasing the concentration of peptide. This could not be attributed to decreased in vivo stability of 102S compared with MCC, as both peptides appeared to be cleared at a comparable rate (Fig. S3).

To better understand the diminished ability of 102S to induce Foxp3 in vivo, we examined proliferation of Foxp3⁺ and Foxp3⁻ 5C.C7 T cells in response to the various peptides and doses. In contrast to stimulation that favored Foxp3 induction (0.1 µg MCC; Fig. 2D), 10 µg 102S resulted in proliferation by the majority of 5C.C7 T cells (Fig. 4 A, left). We hypothesized that decreasing the level of cumulative TCR stimulation in conditions where the majority of 102S-stimulated cells were dividing would favor Foxp3 induction. We sought to do this by manipulating the duration of TCR stimulation. It has previously been demonstrated using two-photon imaging that T cell–APC interactions can be disrupted in vivo by injection of an MHCII antibody after peptide injection (Celli et al., 2007). Using this approach, we interfered with TCR–pMHC interactions at various times after peptide administration by injecting a monoclonal antibody to the MHCII molecule I-E^k (Fig. 4).

Strikingly, injection of anti-MHCII between 6 to 16 h after peptide administration resulted in enhanced Foxp3 induction by the weak agonist 102S (Fig. 4, A and B), which was assayed 6 d after peptide injection. This was true both in terms of the frequency and number of Foxp3⁺ 5C.C7 T cells (Fig. 4, B and C). Disrupting TCR–pMHC interactions during the hours in which formation and persistence of stable T cell–APC contacts dictate subsequent effector function (Hugues et al., 2004; Miller et al., 2004; Celli et al., 2007; but not 36 h after peptide injection; Fig. 4, A and B), lowered the level of cumulative stimulation, as assessed by CFSE dilution, and changed the response of the 5C.C7 T cells. This resulted in an increased frequency of Foxp3-expressing cells, whereas proliferation of the entire 5C.C7 population was reduced (Fig. 4, D and E). In contrast, administration of anti-MHCII subsequent to stimulation with the optimal dose of MCC results in fewer Foxp3⁺ 5C.C7 T cells (Fig. S4), suggesting that the reduced level of TCR stimulation under these conditions falls below the range conducive for Foxp3 induction.

Slightly reduced proliferation and increased Foxp3 expression upon blockade was seen in mice injected with 10 µg MCC, with the effect being less drastic than for 102S (Fig. 4, D and E). This may be explained by the observation that 102S-stimulated cells express the proliferation antigen ki67 with a temporal delay, compared with MCC at the same dose of peptide, indicating that the cells responding to stimulation with the lower affinity ligand may require more time to enter the cell cycle (Fig. S5 A). At 10 µg, the majority of MCC-stimulated cells are ki67⁺ by 24 h, whereas ki67 expression is not detected in most 102S-stimulated cells until after 48 h (Fig. S5 A). Thus, disruption of TCR–pMHC interactions after 102S injection resulted in an increased portion of 5C.C7 T cells below the threshold of stimulation required for ki67 expression, whereas cells stimulated with 10 µg MCC are all ki67⁺ by 48 h, regardless of antibody treatment (Fig. S5 B). This is consistent with the modest effect of MHC blockade on the proliferative capacity and Foxp3 expression of 5C.C7 T cells responding to MCC (Fig. 4, D and E) but does not exclude the possibility that there are doses of MCC at which Foxp3 induction could be more efficiently enhanced by MHC blockade.

Considering that disruption of TCR–pMHC interactions in vivo leads to reduced proliferation coincident with Foxp3 induction, we examined whether there was a dose of 102S that would stimulate cell division, which is comparable to low doses of MCC. In this more refined in vivo titration, we also included the peptide 102N, another weak agonist that induces comparable or slightly less in vitro IL-2 production and conjugate formation than 102S (Egen and Allison, 2002). A titration of MCC, 102S, and 102N in vivo showed that dose could compensate for pMHC potency in stimulating comparable amounts of proliferation. There was a dose of each peptide that resulted in a similar intermediate level of CFSE dilution 5 d after peptide injection (0.3 µg MCC, 3 µg 102S, and 10 µg 102N; Fig. 5, A and B [boxed data points]; and Fig. S6 A, full titration). This readout suggests that all three peptides are capable of achieving a similar level of TCR stimulation. Consistent with this, 2 d after stimulation with these doses of MCC and 102S we detected equivalent levels of ki67 expression and Akt phosphorylation, which represents the degree of activity of Akt kinase and mTORC2 complex (Fig. 5 E). This comparable amount of Akt phosphorylation is of particular interest considering recent studies reporting that the Akt–mTor pathway, which is essential for T cell activation and proliferation, may antagonize peripheral Foxp3 induction (Haxhinasto et al., 2008; Sauer et al., 2008). The level of phospho-Akt correlated with ki67 expression, regardless of stimulating peptide (Fig. 5 F).

As expected, the dose for each peptide that triggered partial division of the responding 5C.C7 population was also the peak of Foxp3 induction, based on the percentage of Foxp3⁺ 5C.C7 T cells (0.3 µg MCC, 3 µg 102S, and 10 µg 102N; Fig. 5, B and C). Although the peak of Foxp3 induction by 102S peptide in this more refined titration (3 µg) produced slightly more Foxp3 induction than 10 µg, both doses still resulted in diminished numbers of Foxp3⁺ 5C.C7 T cells compared with MCC. The numbers of 102S-induced Foxp3⁺ 5C.C7 T cells were lower at day 8 (Fig. 3) than at day 5 (Fig. 5 D). Thus, to gain more insight into the fate of cells stimulated by these strong and weak agonists, we quantitated the number of Foxp3⁺ and Foxp3⁻ 5C.C7 T cells over time (Fig. 6 and Fig. S7).

We found that intravenous injection with either MCC or 102S resulted in a dose-dependent expansion and sharp contraction, just after the peak of the response (Fig. S7, A and B). As a control, we immunized with the same dose of peptide subcutaneously, at the base of tail with LPS, which generates a more typical effector response and does not result in such drastic cell loss (Fig. S7 A). The contraction of 5C.C7 T cells responding to intravenous peptide injection is consistent with studies reporting deletion of T cells stimulated by intravenous peptide as a mechanism of tolerance (Kearney et al., 1994; Liblau et al., 1996). When comparing the doses of MCC, 102S, and 102N that result in a comparable level of proliferation, we observed a similar decline in number of 5C.C7 T cells in both the LNs and the spleen (Fig. S7, C and D). This apparent deletion was coincident with the presence of a fraction of 5C.C7 T cells positive for active caspases, which was not present in recipients that did not receive peptide injection and is consistent with cell death in response to tolerizing antigen administration (Fig. S7 E).

The two weak agonist peptides assessed in our system are able to induce a substantial frequency of Foxp3⁺ 5C.C7 T cells if the right dose of peptide is administered. At early time points, all three peptides induce a comparable percentage of Foxp3⁺ 5C.C7 (Fig. 6, days 2.5 and 5). However, Foxp3⁺ 5C.C7 induced by the low-affinity ligands disappear quickly after the peak of the response, within the same time frame as the Foxp3⁻ compartment, whereas a low dose of MCC results in increased frequencies and numbers of Foxp3⁺ 5C.C7 throughout the course of the experiment (Fig. 6). Considering the loss of the Foxp3⁻ 5C.C7 T cells in all of these conditions (Fig. 6 A), it is the persistence of the Foxp3⁺ 5C.C7 population after low-dose MCC stimulation that is specific, rather than the deletion of either 5C.C7 compartment after stimulation with weak ligand. We do not attribute the difference in Foxp3⁺ 5C.C7 persistence to an inability of 5C.C7 T cells responding to 102S peptide to produce IL-2 (Fig. S8).

Our measurements of the degree of proliferation and initial Foxp3 induction suggest that increasing the dose of 102S results in the same quantity of stimulation as received by MCC-stimulated cells. However, time course experiments revealed that a low dose of the strong agonist peptide, but not a high dose of weak agonist, was able to induce a persistent population of Foxp3⁺ T cells. Thus, there are two phases that must be considered. In the first phase, antigen dose can compensate for potency to achieve the optimal quantity of TCR stimulation for initial Foxp3 induction, whereas in the second phase, high doses of peptide are detrimental to the persistence of the stimulated 5C.C7 T cells, including the induced Foxp3⁺ T cells in the case of the weak agonist peptide. This highlights limitations of comparing in vivo and in vitro systems, the latter of which do not address the issue of persistence.

Foxp3⁺ T reg cell suppression of self-reactive effector T cells is critical for the prevention of autoimmunity (Sakaguchi, 2004). Peripheral induction of Foxp3 in mature CD4⁺ T cells may be an important mechanism of maintaining this self-tolerance. The partial overlap of TCR repertoires of induced T reg cells and thymic T reg cells indicates that peripheral Foxp3 induction could potentially offer a second chance for self-reactive cells that escape thymic selection to enter the T reg cell lineage (Lathrop et al., 2008). Alternatively, cells could convert into T reg cells in response to non–self-antigens presented in tolerizing conditions (Curotto de Lafaille and Lafaille, 2009). Understanding the conditions optimal for peripheral Foxp3 induction in vivo could provide insight into how this process may be manipulated in a therapeutic setting.

Several studies altering antigen dose and degree of Akt and/or mTor activity have shown that efficient Foxp3 induction occurs in conditions suboptimal for activation (Kretschmer et al., 2005; Haxhinasto et al., 2008; Sauer et al., 2008), suggesting that a weaker TCR stimulation favors peripheral conversion. It was less clear how defined TCR/pMHC binding parameters, such as affinity and off rate, influence this process. In this study, we address the cumulative influence of antigen density and the strength of TCR–pMHC interactions on the peripheral induction of Foxp3 in vivo.

Upon comparing ligands of varying affinities across an in vitro peptide titration, we found that there was a limited range of TCR stimulation over which Foxp3 induction occurred. Relative to the natural ligand MCC, peak Foxp3 induction by the superagonist K5 occurred at lower concentrations of peptide, whereas the peak induction by the weak agonist 102S occurred at higher concentrations. All three ligands were able to induce a comparable magnitude of in vitro conversion of 5C.C7 T cells, so long as peptide concentration was altered to compensate for a higher or lower affinity ligand. For each peptide, the TCR stimulation range favoring conversion was below the concentration of peptide required for expansion of the nonconverted cells. The addition of exogenous TGF-β widened the range of TCR stimulation over which conversion occurred, which allowed for Foxp3 induction at concentrations of peptide that were otherwise too high. This is consistent with previously published reports demonstrating the requirement for TGF-β and other negative regulators of T cell activation in scenarios where T cells are receiving high levels of TCR stimulation (Chen et al., 2003; Zheng et al., 2006; Selvaraj and Geiger, 2007).

When MCC, K5, and 102S were titrated in vivo, we did not observe the same simple relationship between ligand affinity and optimal Foxp3 induction dose seen in vitro. The superagonist peptide K5, which peaked at concentrations of peptide 100-fold lower than MCC in vitro, was comparable to MCC in vivo, resulting in efficient conversion of 5C.C7 T cells at the same low dose. Strikingly, stimulation with the weak agonist 102S resulted in drastically reduced numbers of Foxp3⁺ 5C.C7 T cells 8 d after peptide injection, compared with either of the strong ligands. This was surprising given the studies reporting that weaker TCR stimulation favors Foxp3 induction (Kretschmer et al., 2005; Haxhinasto et al., 2008; Sauer et al., 2008). Thus, we went on to modulate multiple parameters influencing the level of TCR stimulation received by 102S-stimulated cells to optimize Foxp3 induction, and subsequently assessed the frequencies and numbers of Foxp3⁺ 5C.C7 T cells induced by the strong and weak agonists over time.

Upon assessing proliferation of cells responding to 10 µg 102S, the dose where we had observed a detectable frequency of Foxp3⁺ 5C.C7, it appeared that induction of Foxp3 by 5C.C7 T cells responding to 10 µg 102S was accompanied by proliferation more robust than that in conditions that are optimal for Foxp3 induction (i.e., 0.1 µg MCC peptide). The idea that such a strong stimulation would be detrimental to Foxp3 induction is consistent with studies demonstrating that Akt activity can antagonize expression of Foxp3 (Haxhinasto et al., 2008; Sauer et al., 2008). In addition, proliferation may oppose Foxp3 induction by cell cycle–dependent recruitment of maintenance DNA methyltransferase, which results in a silenced state of the Foxp3 locus (Josefowicz et al., 2009).

Considering these findings, we sought to decrease the cumulative TCR stimulation in response to this high density of 102S pMHC complexes by injecting antibodies to MHCII 6–16 h after injection of peptide. During this time period, T cell–APC contacts form and persist, and antibodies to MHCII have been shown to result in disruption of T cell–APC interactions (Hugues et al., 2004; Miller et al., 2004; Celli et al., 2007). By blocking TCR–pMHC interactions, we were able to both decrease the frequency of 102S-stimulated cells entering cell cycle and greatly lower their overall proliferative capacity, which correlated with increased initial induction of Foxp3. These results highlight the influence of pMHC potency, density, and duration of interactions with TCR on the overall quantity of TCR stimulation that a cell receives and, thus, its decision to divide or express Foxp3.

Together, our data are in agreement with the idea that weak TCR stimulation that is suboptimal for proliferation favors initial Foxp3 induction. Indeed, a refined titration of peptides in vivo revealed that both MCC and the two weak ligands 102S and 102N could produce this favorable quantity of TCR stimulation if the dose was adjusted to compensate for ligand strength. At these doses, which span nearly two orders of magnitude between MCC and 102N, stimulation by all three peptides resulted in equivalent Foxp3 induction, based on the frequency of Foxp3⁺ 5C.C7 at early time points. In conditions favoring optimal Foxp3 induction, there was no difference in the ability of strong and weak ligands to trigger Akt phosphorylation, which simply correlated with ki67 expression, regardless of stimulating peptide.

Considering only the initial induction of Foxp3, our data varying ligand dose, potency, and duration of TCR–pMHC interactions in vivo would suggest that perhaps it is simply the quantity of TCR stimulation that is important. This would be consistent with a model in which all of these factors contribute to a cumulative signal, which determines T cell responses (Rosette et al., 2001; Rachmilewitz and Lanzavecchia, 2002). Our system suggests that the range of TCR stimulation favoring this process is very narrow, as the dose response for all three peptides is very steep. Based on our in vitro data, where the addition of TGF-β allowed for Foxp3 induction in response to higher concentrations of peptide, we would expect that a particular cytokine environment could also widen the range of TCR stimulation over which Foxp3 induction could occur in vivo. In addition, our results suggest that it may be conceivable to expand the range of antigenic conditions over which Foxp3 induction can occur in a therapeutic setting through manipulation of the duration of TCR–pMHC interactions. Disrupting T cell–APC interactions allowed us to decrease the cumulative TCR stimulation, resulting in Foxp3 induction when ligand density would otherwise be unfavorable. In doing so, it may be possible to alter the frequency of antigen-specific T reg cell, which could skew a potentially pathogenic T cell response toward tolerance.

Despite finding doses of 102S and 102N that yielded similar proliferation and initial Foxp3 induction compared with optimal doses of MCC, these weak agonists still generated diminished numbers of Foxp3⁺ 5C.C7 T cells at longer times after peptide injection. Time course experiments revealed that at the doses of each peptide yielding optimal initial Foxp3 induction, injection of weak agonist peptides resulted in similar deletion of both the Foxp3 negative and positive populations, whereas stimulation with MCC resulted in specific persistence of the induced Foxp3⁺ population of 5C.C7 T cells. Thus, there are two phases influencing the numbers of Foxp3⁺ 5C.C7 observed after in vivo stimulation. During initial induction of Foxp3, antigen dose could compensate for potency to achieve the optimal quantity of TCR stimulation, whereas in the second phase, maintenance versus deletion of this population appeared to be determined by unique stimulation criteria. This highlights the importance of studying in vivo systems, as persistence is not addressed in short term in vitro assays. Low-potency ligands are efficient at achieving the weak TCR signals favoring Foxp3 induction both in vitro and in vivo. However, our data demonstrate that these interactions do not necessarily induce a persistent population of Foxp3⁺ T cells in vivo.

Our findings demonstrate that ligand potency and density are noninterchangeable factors. Increasing the dose of low-potency ligand appeared to achieve a similar level of cumulative TCR stimulation (comparable proliferation, Akt phosphorylation, and Foxp3 induction at early time points) but failed to generate persistent Foxp3⁺ 5C.C7, compared with the strong agonist MCC. Therefore it is not simply a matter of getting the same quantity of TCR stimulation; there are qualitative differences that must be considered. This difference in cell fate suggests that there may be mechanisms allowing T cells to discriminate between signals comings from a few strong TCR–pMHC interactions, versus many complexes with weaker interactions. One can imagine the triggering of survival signals specifically downstream of a strong TCR–pMHC interaction, as opposed to proapoptotic signals which are induced by the triggering of many separate TCRs. There is precedent for cell-intrinsic mechanisms of ligand discrimination via negative signaling feedback loops, which is proposed to be important in the discrimination of self- and foreign ligands (Stefanová et al., 2003). In addition, it has been suggested that the density of antigen may influence TCR signal inhibition, based on an in vivo model comparing tolerance in response to high and low levels of self-antigen (Singh and Schwartz, 2003).

Although is probable that cell-intrinsic mechanisms could be in place to aid in discrimination between high-potency versus high-density ligands, it is also possible that cell-extrinsic factors play a role in this process. This seems especially important considering that the maturation state of APCs is known to play an important role in T cell fate under tolerizing conditions. Low levels of costimulation favor Foxp3 induction (Benson et al., 2007), anergy, and deletion (Hawiger et al., 2001). Both dose and potency of peptide ligand have been reported to influence the kinetics and type of effector T cell–APC contact formation (Bousso and Robey, 2003; Skokos et al., 2007; Henrickson et al., 2008). Thus, it is possible that low doses of MCC and high doses of 102S achieve comparable cumulative TCR stimulation, but in the context of different types of T cell–APC interactions. For example, very low concentrations of peptide may only result in short interactions, which, given a potent ligand, may be sufficient to provide the minimal stimulation required for Foxp3 induction. In this scenario, more stable T cell–APC interactions resulting from higher concentrations of peptide could play a role in the subsequent deletion of these cells. Transient TCR–pMHC interactions upon low-dose injection of peptide could explain the similarity in levels of Foxp3 induction by MCC and K5 we observed, as the increased potency of the high-affinity superagonist may be minimized in the absence of more stable conjugate formation.

We propose that a low density of a high-affinity TCR ligand is optimal to achieve the quantity and quality of stimulation required to induce a persistent population of Foxp3⁺ T cells in vivo. In fact, considering that the Foxp3⁻ 5C.C7 T cells are deleted in these same conditions, and that injection of high doses of either MCC or 102S results in comparable loss of 5C.C7 T cells, specificity is most apparent in the survival of the Foxp3⁺ 5C.C7 in response to small amounts of strong ligand. Deletion and Foxp3 induction are two proposed mechanisms of peripheral tolerance. However, how the strength and density of TCR ligand steers a self-reactive T cell between these fates is understudied. TCR repertoire analysis has suggested that T reg cells may recognize self-antigens with low avidity (Hsieh et al., 2004). We suggest that low concentrations of specific self-pMHC complexes could contribute to peripheral Foxp3 induction in response to self-antigens, even if the TCR–pMHC interaction is of a high affinity. In the case of more abundantly expressed self-pMHC complexes, our data suggest that deletion of the cognate T cells is a more likely mechanism of tolerance, regardless of the strength of the ligand and extent of Foxp3 induction. It is interesting to speculate that there could be detrimental effects of maintaining T reg cells that recognize ubiquitous self-ligands, such as bystander suppression (Masteller et al., 2005; Tang and Bluestone, 2008).

In summary, initial in vivo Foxp3 induction can be achieved by optimizing the quantity of cumulative TCR stimulation, which can be thought of as a sum of the factors of ligand potency, density, and duration of TCR–pMHC interactions. However, TCR ligand potency and density are clearly noninterchangeable in determining the persistence of the cells induced to express Foxp3. These findings provide important insight into how affinity and prevalence of antigen may determine the route to peripheral tolerance.

5C.C7 TCR transgenic RAG2^−/− mice (Taconic) were bred to B10.A CD45.1 (provided by W. Paul via the National Institute of Allergy and Infectious Diseases contract facility at Taconic) to generate 5C.C7 RAG2^−/− CD45.1 mice, which were used as donor mice in adoptive transfer experiments. Male B10.A mice (Taconic) were used as adoptive transfer recipients at between 6 and 9 wk of age. All mice were maintained in microisolator cages, and treated in accordance with the regulations of the National Institutes of Health and the American Association of Laboratory Animal Care. Experiments in this study were approved by the Memorial Sloan-Kettering Cancer Center Institutional Animal Care and Use Committee.

Cells were cultured in a 37°C humidified chamber with 5% CO₂ in complete RPMI1640 (supplemented with 10% FCS, 2 µM glutamine, 100 U/ml penicillin and streptomycin, and 2 µM 2-mercaptoethanol). Single cell suspensions were prepared from LNs harvested from 5C.C7 RAG2^−/− mice (routinely >90% Vαll⁺Vβ3⁺ by flow cytometry) and stimulated in duplicate or triplicate, as indicated, in 96-well round bottom plates. In T cell proliferation assays, 3 × 10⁴ 5C.C7 cells were stimulated with 2 × 10⁵ irradiated splenocytes and the indicated peptides for 60 h. 1 µCi/well ³H-methyl-thymidine was used to monitor T cell proliferation. Cells were harvested onto glass-fiber filters using a Tomtec harvester, and filters were counted using a MicroBeta scintillation counter (PerkinElmer). For in vitro T reg cell conversion assays, 10⁵ 5C.C7 T cells were cultured with 4 × 10⁵ irradiated splenocytes that had been depleted of T cells using CD90.2 microbeads (Miltenyi Biotec), the indicated peptide, and 100 U/ml of recombinant human IL-2 (PeproTech), with or without human TGF-β1 (PeproTech). Cells were cultured for 90–96 h before cells were stained for flow cytometry analysis. In some cultures the absolute number of cells was determined using an automated cell counter (Guava Technologies Inc.).

Peptides were synthesized and HPLC purified (≥95%) by Biosynthesis, Inc.

Flow cytometry was done on an LSRII (BD). Antibodies to surface markers were obtained from BD, eBioscience, or BioLegend. Foxp3 was detected using antibodies (eBioscience) and fixation/permeabilization reagents. For detection of phosph-Akt (Ser473), we used an Alexa Fluor 647–conjugated antibody (Cell Signaling Technology; DE9), after immediate fixation with 2% paraformaldehyde and subsequent permeabilization with cold methanol. CTLA-4 and ki67 antibodies (BD) were detected after fixation and permeabilization as required for Foxp3 or phospho-Akt analysis, depending on the experiment.

Between 5 × 10⁵ and 1 × 10⁶ 5C.C7 RAG2^−/− CD45.1 T cells were transferred into B10.A recipients by tail vein injection. Mice were subsequently immunized intravenously the same or next day with 0.03–100 µg of peptide diluted in PBS. When disruption of T cell–APC interactions was desired, 500 µg of monoclonal antibody to the mouse MHC class II molecule I-E^k (clone 14–4-4S; American Type Culture Collection; produced by Memorial Sloan Kettering Cancer Center [MSKCC] Monoclonal Antibody Core Facility [MACF]) or isotype control antibody (clone 20–8-4s, recognizing H-2Kb, also produced by the MSKCC MACF) was injected into the tail vein at the indicated time point after peptide injection. In some experiments cells were CFSE labeled before transfer. To enrich for 5C.C7 T cells before flow cytometry analysis, CD4⁺ T cells were purified from LN suspensions by Dynal negative selection (Invitrogen).

Flow data were analyzed using FlowJo (version 8.8.6; Tree Star, Inc.) and further processed using either Prism 5.0 (GraphPad Software, Inc.) or Excel (Microsoft). Proliferative capacity was calculated from CFSE profiles as previously described (Gudmundsdottir et al., 1999).

Fig. S1 shows the influence of exogenous TGF-β on in vitro Foxp3 induction by peptides of varying potency. Fig. S2 shows characteristics of Foxp3⁺ 5C.C7 T cells that are common to T reg cells. Fig. S3 demonstrates comparable stability of 102S and MCC peptides in vivo. Fig. S4 shows the impact of MHCII blockade on Foxp3 induction by an optimal low dose of MCC. Fig. S5 contains analysis of ki67 expression after in vivo stimulation with varying doses of MCC and 102S, with and without disruption of TCR–pMHC interactions. Fig. S6 includes additional peptide doses from titration and timecourse experiments shown in Figs. 5 and 6. Fig. S7 characterizes loss of 5C.C7 T cells after injection of intravenous peptide, including dose dependence and expression of active caspases. Fig. S8 shows ex vivo IL-2 production by 5C.C7 T cells, harvested 6 d after in vivo peptide stimulation.

We thank A. Trumble-Koncelik and J. Geddes for technical assistance and M. Huse, S.Z. Josefowicz, P.A. Savage, S.A. Quezada, Y. Zheng, R. Niec, T. Pentcheva-Hoang, M.A. Sepulveda, and W. Walkowicz for helpful discussion and critical reading of the manuscript.

R.A. Gottschalk is a predoctoral fellow of the Cancer Research Institute. E. Corse was supported by a postdoctoral fellowship from the Irvington Institute of Immunological Research. J.P. Allison is an investigator of the Howard Hughes Medical Institute and is supported by grants from the National Institutes of Health.

The authors have no conflicting financial interests.