Reprogrammed Foxp3⁺ regulatory T cells provide essential help to support cross-presentation and CD8⁺ T cell priming in naive mice

SUMMARY

Foxp3⁺ regulatory T (Treg) cells can undergo reprogramming into a phenotype expressing proinflammatory cytokines. However, the biologic significance of this conversion remains unclear. We show that large numbers of Treg cells undergo rapid reprogramming into activated T-helper cells following vaccination with antigen plus Toll-like receptor 9 (TLR-9) ligand. Helper activity from converted Treg cells proved essential during initial priming of CD8⁺ T cells to a new cross-presented antigen. Help from Treg cells was dependent on CD40L, and (unlike help from conventional non-Treg CD4⁺ cells) did not require pre-activation or prior exposure to antigen. In hosts with established tumors, Treg cell reprogramming was suppressed by tumor-induced indoleamine 2,3-dioxygenase (IDO), and vaccination failed due to lack of help. Treg reprogramming, vaccine efficacy and anti-tumor CD8⁺ T cell responses were restored by pharmacologic inhibition of IDO. Reprogrammed Treg cells can thus participate as previously unrecognized drivers of certain early CD8⁺ T cell responses.

INTRODUCTION

Foxp3⁺ regulatory T (Treg) cells are a unique cell population. They are a critical component of normal self-tolerance, yet, paradoxically, in some settings the Treg cell lineage is required in order to support the early phase of normal immune responses (Lund et al., 2008). It has been unclear how Treg cells could both suppress and promote immune responses in different contexts (Zhou et al., 2009a).

Recently, it has been observed that Treg cells retain an unexpected degree of phenotypic plasticity. Under certain conditions, Treg cells may lose their suppressor phenotype and become “reprogrammed” into T-helper-like cells (Sharma et al., 2009; Yang et al., 2008; Zhou et al., 2009b). Phenotypically, these former Treg cells resemble proinflammatory effector cells (Duarte et al., 2009; Williams and Rudensky, 2007), and large numbers of such cells may be found in mice dying of infection, in chronic autoimmune disorders, or in mice rejecting tissue allografts (Oldenhove et al., 2009; Vokaer et al., 2010; Zhou et al., 2009b). However, these are all highly abnormal conditions, and it has been unclear whether reprogrammed Treg cells play a physiologic role in normal, protective immune responses.

We have shown that Treg cell reprogramming can be regulated in vitro by the enzyme indoleamine 2,3-dioxygenase (IDO). IDO is an innate immunoregulatory mechanism that participates in tolerance and immunosuppression in pregnancy, mucosal tolerance and other settings (Mellor and Munn, 2008). Hosts with established tumors can show markedly elevated levels of IDO in tumor-draining lymph nodes (Munn et al., 2004), and tumor-induced IDO can directly activate Foxp3⁺ Treg cells for enhanced suppressor activity (Sharma et al., 2007). Conversely, if IDO is blocked (e.g., by the pharmacologic IDO-inhibitor 1-methyl-tryptophan) then the Treg cells in tumor-bearing hosts become unstable, and can be driven by inflammation to undergo reprogramming into helper-like T cells, expressing IL-17 and other proinflammatory cytokines (Sharma et al., 2009). However, it has been unclear whether these phenotypically-reprogrammed Treg cells play any functional role in anti-tumor immunity.

In the current study, we now show that cells of the Foxp3⁺ lineage can participate as an integral part of the CD4⁺ T-helper system. In certain settings, reprogrammed Treg cells were found to play an indispensable helper role, allowing innate inflammation to drive the early (priming) phase of CD8⁺ T cell response to new antigen. Further, we show that in mice with established tumors, one key reason for the failure of therapeutic immunization can be the inhibition of normal Treg cell reprogramming by tumor-induced IDO.

RESULTS

Treg cells undergo reprogramming in vaccine-draining lymph nodes

Treg cell reprogramming was studied using vaccination with a whole-protein antigen (chicken ovalbumin, OVA) which must be processed by DCs and cross-presented on MHC class I to CD8⁺ T cells. In this cross-presentation model, the CD8⁺ T cell response is heavily dependent on CD4⁺ help to “license” the DCs (Bennett et al., 1998). Vaccinations were performed in C57BL/6 mice bearing a Foxp3-GFP fusion protein targeted to the Foxp3 locus (Fontenot et al., 2005). We have previously shown that Treg cells from these mice display detectable GFP fluorescence for at least 4 days after reprogramming (Sharma et al., 2009). Foxp3^GFP mice received adoptive transfer of OVA-specific OT-I cells (CD8⁺, recognizing the SIINFEKL peptide of OVA), followed by immunization with whole OVA protein plus the TLR9-ligand CpG-1826, emulsified in incomplete Freund’s adjuvant (IFA).

Figure 1A shows analysis of CD4⁺ cells in vaccine-draining lymph nodes (LNs) following immunization. Treg cells and conventional (non-Treg) CD4⁺ cells were distinguished based on Foxp3-GFP expression. Treg cells are known to respond rapidly to proinflammatory signals (O'Gorman et al., 2009); based on these reports we analyzed activation-induced phosphorylation of STAT5. As early as 6 hrs after immunization, large numbers of Treg cells (GFP⁺) expressed phosphorylated STAT5, whereas non-Treg CD4⁺ cells (GFP⁻) did not respond until 24–48 hrs after immunization. Similar results were seen with the early activation marker CD69 (lower panels). Thus, in this model, Treg cells were the first and most rapid responders to vaccine-induced activation.

Figure 1. Foxp3⁺ Treg cells undergo phenotypic reprogramming following vaccination.

(A) Foxp3^GFP mice received adoptive transfer of OT-I cells and immunization with OVA+CpG+IFA vaccine. At different time points, draining LNs were stained for phospho-STAT5 and surface CD69; graphs show the percentage of gated CD4⁺ Treg cells (GFP⁺) and non-Treg cells (GFP⁻) expressing the antigens at each time-point.

(B) Foxp3^GFP mice received OT-I cells and were immunized as above, with or without CpG in the vaccine. Draining LN cells were harvested 4 days after vaccination, incubated for 4 hrs with low-dose PMA+ionomycin in the presence of brefeldin A, then stained for intracellular cytokines.

(C) Foxp3^GFP mice received OT-I cells and immunization as above, and draining LNs were stained for CD40L expression at different times (staining was directly ex vivo, without PMA treatment). Graph shows pooled data from 3–5 experiments at each time-point (bars show SD).

(D) F1(Foxp3-GFP-cre × ROSA26-YFP) mice received OT-I cells and immunization as above, and draining LNs were analyzed on day 4. GFP and YFP are both seen in the FL1 channel.

(E) Foxp3^GFP mice received CFSE-labeled OT-I cells (Thy1.1 congenic), followed by OVA vaccine containing IFA plus graded amounts of CpG. CD4+ cells were analyzed for surface CD40L; OT-I cells (gated Thy1.1⁺) were analyzed for CFSE dye-dilution and intracellular granzyme B.

In each panel, data are representative of 3–5 experiments (17 experiments for panel B). See also Supplemental Figure S1.

We next asked whether activated Treg cells showed evidence of phenotypic plasticity (reprogramming) following vaccination. One defining characteristic of such plasticity is that reprogrammed Treg cells acquire the ability to express proinflammatory cytokines (IL-2, IL-17, TNFα) when re-stimulated in vitro with PMA plus ionomycin (Sharma et al., 2009; Yang et al., 2008). Expression of these cytokines implies a major alteration in the Treg cell phenotype, because they would normally be profoundly suppressed in the Foxp3⁺ lineage (Williams and Rudensky, 2007). For all in vitro re-stimulation studies we used a low concentration of PMA+ionomycin (as described in Supplemental Methods online), in order to minimize the nonspecific induction of cytokine expression due to the in vitro stimulation alone. Under these conditions, Figure 1B shows that prior to vaccination, resting Treg cells from Foxp3^GFP mice produced no IL-2 or TNFα when challenged with PMA (left-hand panels). However, after vaccination (right-hand panels) many Treg cells had acquired the ability to produce IL-2 and TNFα, and large numbers also co-expressed IL-17, suggesting acquisition of a polyfunctional “helper-like” phenotype.

The expression of proinflammatory cytokines suggested that Foxp3 was no longer fully suppressing these genes. In some reports, Treg cells undergoing reprogramming in vitro may lose Foxp3 expression, although in other cases the reprogrammed Tregs can maintain their expression of Foxp3 (Yang et al., 2008). In our system, Treg cells that underwent vaccine-induced reprogramming in vivo continued to express the Foxp3-GFP reporter protein throughout the experiment (>4 days), as shown in Figure 1A. As a technical caveat, however, Supplemental Figure S1A (available online) shows that, following reprogramming, the Foxp3 protein became difficult to detect by the usual intracellular staining methods, suggesting that Foxp3 might become sequestered or otherwise altered in its subcellular location (although it remained readily detectable by GFP fluorescence and immunoblot, Figure S1B). Thus, for our purposes, the Foxp3-GFP reporter mice provided the most reliable marker to identify reprogrammed Treg cells.

Reprogrammed Treg cells upregulate cell-surface CD40-ligand

We next examined the in vivo induction of cell-surface CD40-ligand expression on Treg cells. CD40L is an important functional mediator of T cell help (Bennett et al., 1998; Schoenberger et al., 1998), and it could be measured directly ex vivo, without requiring in vitro stimulation. Figure 1C shows that CD40L was upregulated on a subset of Treg cells beginning at approximately 15 hrs after vaccination. Of note, under our experimental conditions, the only cells expressing CD40L at these early time-points derived exclusively from the Treg cell population (GFP⁺). Somewhat fewer Treg cells expressed CD40L than could be induced to express cytokines with PMA, but up to 25% of all Treg cells upregulated CD40L. Additional studies, not shown, confirmed that CD40L was expressed specifically on the subset of reprogrammed Treg cells that also co-expressed proinflammatory cytokines (polyfunctional phenotype). Although it is known that some Treg cells may express inducible or intracellular CD40L under various conditions (Taylor et al., 2001), in our model the constitutive expression of surface CD40L appeared an informative marker of the reprogrammed phenotype.

Results from the Foxp3^GFP gene targeted mice were confirmed in a second reporter strain, bearing a Bac-transgenic GFP-cre-recombinase fusion protein under the Foxp3 promoter (Zhou et al., 2009b). When this mouse is crossed with a ROSA26-YFP (floxed-stop) reporter strain, the F1 offspring have >95% of Foxp3⁺ cells irreversibly marked by YFP (Zhou et al., 2009b). Figure 1D shows that immunization of these mice likewise caused phenotypic reprogramming of Treg cells (cytokine expression) and constitutive upregulation of surface CD40L. (In the analysis in Figure 1D, both the GFP and YFP signals are combined in the green channel, such that all cells that currently or previously expressed Foxp3 are marked, see Supplemental Figure S1C and S1D.)

Taken together these results demonstrated that Treg cells responded rapidly to vaccination (more rapidly than the conventional, non-Treg CD4⁺ cells). Following vaccination, many Treg cells acquired the ability to express proinflammatory cytokines and CD40L, suggestive of a helper T cell phenotype.

Treg cell reprogramming is driven by CpG-induced inflammatory signals

Figure 1B (above) showed that successful reprogramming of Treg cells required the inclusion of CpG in the vaccine. CpG is known to inhibit the suppressor activity of Treg cells in vivo via a signaling pathway requiring MyD88, the adapter protein downstream of TLR9 (Pasare and Medzhitov, 2003). Consistent with this, Figure S1E shows that Treg cells adoptively transferred into hosts lacking MyD88 were unable to undergo normal reprogramming in response to CpG vaccine. IL-6 is a key proinflammatory cytokine downstream of MyD88 (Pasare and Medzhitov, 2003), which is known to drive Treg cell reprogramming in vitro (Yang et al., 2008). Figure S1E shows that Treg cells transferred into hosts lacking IL-6 also failed to undergo normal reprogramming in response to vaccination. Finally, Figure 1E shows that the response to CpG was fully titratable, with increasing doses of CpG producing progressively more reprogramming of Treg cells (measured in these experiments by CD40L upregulation).

Thus, the reprogramming observed in our system appeared consistent with other examples of reprogramming reported in the literature, and was driven by the innate immune system (TLR-ligand, MyD88 and IL-6). Moreover, the degree of Treg cell reprogramming closely tracked the ability of the vaccination to stimulate CD8⁺ T cell responses (i.e., proliferation of OT-I cells, as shown in Figure 1E). This concordance prompted us to next ask whether reprogrammed Treg cells were providing functional helper activity for the CD8⁺ T cells.

Reprogrammed Treg cells provide help for cross-presentation

To assess the functional activity of converted Treg cells, we used host mice with a targeted disruption of the TCRα-chain (Tcra^−/− mice), which lack all endogenous CD4⁺ or CD8⁺ T cells. (These mice have B cells, NK cells and γ δ T cells, and so are not globally lymphopenic, but selectively lack αβ T cells). Tcra^{−/ −} hosts received adoptive transfer of either sorted Foxp3GFP Treg cells, sorted non-Treg cells (CD4⁺GFP⁻ cells from the same mice), or no CD4⁺ cells. Mice were rested, then received CFSE-labeled OT-I cells and were immunized with OVA+CpG+IFA vaccine, and draining LNs analyzed 4 days later. Fig. 2A shows that Tcra^{−/ −} mice receiving no CD4⁺ cells, or receiving only the non-Treg fraction of CD4⁺ cells, supported little proliferation of OT-I cells, and no upregulation of the differentiation marker granzyme B. (Also, in many experiments, the OT-I cells that attempted to activate in the absence of CD4 help showed a reduction in CD8 expression, as visible in the figure.) In contrast, the OT-I cells in mice receiving the Treg cell fraction showed a robust proliferative response to vaccination and extensive upregulation of granzyme B. Thus, in our model, it appeared that virtually all of the available helper activity by transferred CD4⁺ cells was being contributed by the Treg cell fraction, not by the conventional non-Treg CD4⁺ cells.

Figure 2. Reprogrammed Treg cells are functionally required to support early CD8⁺ T cell response to a cross-presented vaccine antigen.

(A–C) T cell-deficient Tcra^{−/ −} host mice received adoptive transfer of sorted Treg cells or CD⁴⁺ non-Treg cells as indicated. Mice were rested for 1–7 days, then received CFSE-labeled OT-I cells and immunization with OVA+CpG+IFA. Responses were analyzed in draining LNs after 4 days.

(A) Foxp3^GFP Treg cells or non-Treg cells (CD4⁺GFP⁻) were transferred; controls received no CD4⁺ cells. Comparison of OT-I cell responses in draining LN.

(B) Treg cells (CD4⁺CD25⁺) were sorted from WT B6 or Cd40lg^{−/ −} mice and transferred into Tcra^{−/ −} hosts (controls received no CD⁴⁺ cells), followed by adoptive transfer of OT-I cells and vaccination with OVA+CpG+IFA. Draining LNs were analyzed for DCs (CD11c⁺) and OT-I cells (CD8⁺).

(C) Tcra^{−/ −} mice received Foxp3^GFP Treg cells or no Treg cells. A third group received no Treg cells but were treated with anti-CD40 (clone FGK45, 250 ug i.p. on the day of vaccination and 100 ug i.p. 2 days later). Plots show CFSE-labels OT-I cells in vaccine-draining LNs on day 4.

(D) CD40L-deficient host model: Cd40lg^{−/ −} mice received adoptive transfer of CD40L-sufficient Foxp3^GFP Treg cells (2–4 × 10⁵ CD4⁺GFP⁺) or non-Treg cells (1 × 10⁶ CD4⁺GFP⁻), both congenically marked with Thy1.1. All mice received CFSE-labeled OT-I cells and immunization as above, and response of OT-I cells and transferred CD4⁺ cells analyzed on day 4.

In each panel, data are representative of 3–6 experiments. See also Supplemental Figure S2.

This potent helper activity contained within the Treg cell lineage was not an artifact of Treg cells derived from the Foxp3^GFP gene-targeted strain, because sorted CD4⁺CD25⁺ Treg cells from wild-type B6 mice produced identical results, and sorted Treg cells from F1(Foxp3-GFP-cre × R26-YFP) donors (as used in Figure 1D) were also excellent helper cells in the Tcra^{−/ −} model (data not shown).

CD40L on reprogrammed Treg cells drives DC activation and CD8⁺ T cell proliferation

CD40L is a key molecular mechanism by which helper T cells activate (“license”) DCs to cross-present antigens to CD8⁺ T cells (Bennett et al., 1998; Schoenberger et al., 1998). We asked whether the surface CD40L expressed by reprogrammed Treg cells allowed them to activate host DCs. DC activation was assessed in these experiments as upregulation of CD80 and CD86. In Figure 2B, Tcra^{−/ −} mice received Treg cells either from WT donors or from Cd40lg^{−/ −} donors (control mice received no Treg cells); then all mice received OT-I cells and OVA+CpG+IFA vaccine. In Tcra^{−/ −} mice receiving no Treg cells, the DCs in vaccine-draining LNs were unable to upregulate CD80 and CD86; and the OT-I cells did not divide or express granzyme B (shown in the paired dot-plots below). In contrast, in hosts receiving WT Treg cells the DCs expressed high levels of CD80 and CD86 after vaccination, and OT-I cells divided robustly and upregulated granzyme B. However, mice reconstituted with Treg cells lacking CD40L showed no upregulation of CD80 and CD86 on DCs, and no proliferation of OT-I cells. Supplemental Figure S2A shows that upregulation of costimulatory molecules on the CD8α⁺ subset of DCs (which are the specific APCs critical for cross-presentation to CD8⁺ T cells) was dependent on help from reprogrammed Treg cells.

Thus, adoptive-transfer studies unexpectedly revealed that virtually all of the spontaneous helper activity of resting CD4⁺ T cells in our system was contained in the Treg cell fraction, not in the conventional, non-Treg CD4⁺ cells. In our model, in which a new antigen must be cross-presented for the first time to naive CD8⁺ T cells, it proved to be the Treg population that rapidly underwent reprogramming, upregulated CD40L and licensed DCs for cross-presentation.

Reprogrammed Treg cells rescue the helper defect in Cd40lg^{−/ −} hosts

Tcra^{−/ −} mice are partially lymphopenic, which might influence Treg cell reprogramming (Duarte et al., 2009). Therefore, we confirmed these findings in a second, non-lymphopenic model. Mice with a targeted disruption of the CD40-ligand gene (Cd40lg^{−/ −} mice) have a profound defect in antigen cross-presentation to CD8⁺ T cells due to a lack of functional CD4 help (which requires CD40L) (Bennett et al., 1998). We asked whether CD40L-sufficient Treg cells (and only the Treg cells) could rescue this selective T-helper cell defect in Cd40lg^{−/ −} mice. Treg cells and non-Treg CD4⁺ cells were sorted from CD40L-sufficient Foxp3^GFP donors, and adoptively transferred into Cd40lg^{−/ −} recipients. All mice then received CFSE-labeled OT-I responder cells and were immunized with OVA+CpG+IFA vaccine, as in the Tcra^{−/ −} model. Figure 2D shows that the transferred Treg cells underwent reprogramming following vaccination, upregulated CD40L, and supplied potent help for OT-I activation. In contrast, the non-Treg fraction of CD4⁺ cells did not upregulate CD40L or proinflammatory cytokines following transfer, and failed to provide help for OT-I cells. (Likewise, none of the endogenous Treg cells in Cd40lg^{−/ −} mice underwent detectable reprogramming, which is consistent with the known defects in Treg cell function in these mice.) The helper effect of CD40L-sufficient Treg cells was fully titratable (shown in Supplemental Figure S2B). Thus, these findings in the non-lymphopenic Cd40lg^{−/ −} model confirmed all of the essential findings from the Tcra^{−/ −} model.

Conventional (non-Treg) CD4⁺ cells can deliver help if they are pre-activated

The lack of helper activity in the conventional (non-Treg) CD4⁺ cells was somewhat surprising, because these are the classical “T-helper” cells. However, the naive CD4⁺ repertoire contains an extremely low frequency of clones specific for a new antigen (Moon et al., 2007). We hypothesized that the failure of conventional CD4⁺ cells to provide help in our priming model was due to the low number of OVA-specific clones. (In contrast, the Treg cells – with their high frequency of self-reactive TCRs – would have no such limitation.) To determine whether increasing the clonal frequency would allow cognate help from conventional (non-Treg) CD4⁺ cells, we pre-immunized donor mice with OVA vaccine and boosted them multiple times, to create OVA-specific memory cells. Supplemental Figure S2C shows that under these conditions the conventional, non-Treg CD4⁺ cells now became able deliver help to CD8⁺ T cells (whereas the same population from naive, un-primed mice could not). Similarly, if conventional (non-Treg) CD4⁺ cells were isolated from TCR-transgenic OT-II mice, which have a high clonal frequency of OVA-specific CD4⁺ cells, these likewise showed good helper activity (Supplemental Figure S2D). Thus, conventional (non-Treg) CD4⁺ cells could deliver help, as expected, as long as they were pre-immunized with cognate antigen or were present at high clonal frequency. However, in a normal “resting” CD4⁺ repertoire, the non-Treg CD4⁺ cells showed little spontaneous helper activity for a new antigen. Indeed, even when non-Treg cells were selectively enriched for cells with an “antigen-experienced” (CD62L^lo) phenotype, non-Treg cells from (naïve) resting hosts still failed to provide any spontaneous helper activity (Supplemental Figure S2E). The non-Treg CD4⁺ population only became able to deliver spontaneous help if they were first pre-activated in vivo by systemic administration of CpG for 7–10 days (Supplemental Figure S2E). Thus, whereas both Treg cells and conventional CD4⁺ non-Treg cells were able to mediate help, the unique feature of reprogrammed Treg cells was that they could provide immediate, spontaneous help for new antigens, without requiring prior pre-activation.

Established B16F10 tumors induce progressive unresponsiveness to vaccine

We next evaluated the response to vaccination in mice with established tumors. This is a very different setting from naive mice, because established tumors actively inhibit T cell responses against tumor-associated antigens. We used the B16F10 melanoma model because B16 tumors are known to suppress anti-tumor immune responses (Quezada et al., 2008), and the B16F10 subline is particularly immunosuppressive (Shields et al., 2010).

Figure 3 demonstrates the progressive loss of T cell response to vaccination during B16F10 tumor growth. Mice bearing tumors of different durations received TCR-transgenic pmel-1 cells (CD8⁺, recognizing the shared self-tumor antigen gp100) (Overwijk et al., 2003), then were vaccinated with an altered peptide ligand (human gp100) in CpG + IFA (Overwijk et al., 2003). In mice without tumors, this vaccine produced robust proliferation of pmel-1 cells; however, in mice with day 3 tumors, response to vaccine was markedly reduced, and by day 7 of tumor growth the response to vaccination was essentially lost. The kinetics of suppression depended on the specific tumor cell line and the size of the tumor inoculum, but similar unresponsiveness was seen with E.G7 lymphoma (not shown), and our findings are consistent with reports by others (Oizumi et al., 2008).

Figure 3. Progressive unresponsiveness to vaccination induced by established tumors.

C57BL/6 mice were implanted with 1 × 10⁶ B16F10 tumor cells (control mice received no tumor). After 3–7 days of tumor growth, mice received adoptive transfer of CFSE-labeled resting pmel-1 cells (sorted CD8⁺) and gp100+CpG+IFA vaccine. Four days later, tumor-draining LNs were analyzed (or vaccine-draining LN in mice without tumors). Representative of a total of 9 experiments on day 7, and 3 experiments each on other days.

Tumor-induced IDO blocks Treg cell reprogramming

We have previously shown that B16 tumors upregulate host expression of the immunosuppressive enzyme IDO (Munn et al., 2004). In vitro, IDO can stabilize the suppressive (regulatory) phenotype of Treg cells, and inhibit their reprogramming into helper-like cells (Sharma et al., 2009). Therefore, we asked whether elevated IDO expression in hosts with established B16F10 tumors prevented Treg cells from undergoing normal reprogramming. Figure 4A (middle column of dot-plots) shows that mice with established tumors displayed markedly impaired reprogramming of Treg cells in following vaccination (seen as reduced IL-17 and CD40L expression, and complete absence of IL2 and TNFα expression). In contrast, when IDO was blocked with the IDO-inhibitor drug 1-methyl-D-tryptophan (1MT), then the same vaccination regimen cause extensive Treg cell reprogramming (right-hand dot-plots). (The figure shows data from tumor-draining LNs; similar results were also seen in LNs draining the vaccination site.)

Figure 4. Treg cell reprogramming in tumor-bearing hosts is antagonized by IDO, and restored by pharmacologic IDO-inhibitor.

(A) Foxp3^GFP host mice were implanted with B16F10 tumors. On day 7 of tumor growth, mice received adoptive transfer of resting pmel-1 cells (sorted CD8⁺), with or without gp100+CpG+IFA vaccine. Groups were treated either with continuous 1MT in drinking water (beginning on day 6, one day prior to vaccination) or with vehicle control, as indicated. Four days after vaccination, tumor-draining LNs were harvested, treated with low-dose PMA+ionomycin, and stained as in Figure 1. The lower set of dot-plots show the gated GFP⁺ (Treg cell) population.

(B) One day prior to tumor implantation, Treg cells (CD4⁺CD25⁺) were enriched from GCN2-deficient Eif2ak4^{−/ −} mice or from WT B6 controls, and transferred into congenic Thy1.1⁺ hosts. B16F10 tumors were then implanted, and 7 days later mice received pmel-1 cells and vaccine as in the previous panel. (None of the mice were treated with 1MT). After 4 days, the transferred Treg cells were analyzed in tumor-draining LNs.

(C) Prior to tumor implantation, T cell-deficient Tcra^{−/ −} hosts received a mixture of CD⁴⁺ cells, comprising Thy1.2⁺ Treg cells from Foxp3^GFP donors plus congenic Thy1.1⁺ non-Treg cells (CD4⁺GFP⁻Thy1.1⁺). All mice then received B16F10 tumors, and 7 days later were treated with pmel-1 cells and vaccine, with or without oral 1MT as indicated. Four days after vaccination, tumor-draining LNs were analyzed for transferred Treg cell and non-Treg populations, gated separately on Thy1 allotype.

Each panel is representative of 3–6 experiments. See also Supplemental Figure S3.

In other studies, reprogramming was seen with both a representative Class B and Class C oligonucleotide (Supplemental Figure S3), and we have previously shown that extensive reprogramming occurs following lentiviral vaccines as well (Sharma et al., 2009); thus vaccine-induced Treg cell reprogramming was a generalized phenomenon seen in multiple vaccine systems.

The oral 1MT used in Figure 4A was a systemic inhibitor of IDO. To more specifically test the hypothesis IDO directly affected Treg cells, we asked whether the effect of IDO was lost when Treg cells lacked the GCN2 kinase signaling pathway. GCN2 (encoded by the Eif2ak4 gene) senses amino acid deprivation, and an intact GCN2 pathway is required in target T cells in order for IDO to exert many of its effects (Munn et al., 2005; Sharma et al., 2007). Treg cells were enriched from mice with a targeted disruption of the GCN2 gene (Eif2ak4^{−/ −} mice) or WT B6 controls (both on the Thy1.2 background) and transferred into congenic Thy1.1⁺ hosts (Figure 4B). Mice then received tumors and were vaccinated 7 days later, but were not treated with 1MT. Under these conditions (IDO active), there was little detectable reprogramming of the transferred WT Treg cell cohort, as expected; however, the Treg cells from GCN2-deficient Eif2ak4^{−/ −} mice were resistant to the effects of IDO, and underwent normal reprogramming following vaccination, despite the presence of tumor, and without the need for 1MT.

Taken together, these data indicated that established tumors progressively created a milieu in tumor-bearing hosts in which the normal, vaccine-induced reprogramming of Treg cells was suppressed, and could no longer occur. This inhibition of reprogramming was mediated by tumor-induced IDO, as shown both by a pharmacologic IDO-inhibitor, and by using IDO-resistant Eif2ak4^{−/ −} Treg cells.

Mature pre-existing Tregs undergo vaccine-induced reprogramming

It was formally possible that the cytokine-expressing Foxp3-GFP⁺ cells seen in Figure 4A, might actually arise from naive CD4⁺ cells that merely upregulated Foxp3 de novo (rather than deriving from mature Treg cells that had converted). To definitively establish whether the cells that we observed were deriving from mature, pre-existing Foxp3⁺ Treg cells, we co-transferred marked populations of Treg cells and non-Treg CD4⁺ cells into the same Tcra^{−/ −} hosts (Figure 4C). The Treg cell subset was sorted from Foxp3^GFP donors (Thy1.2⁺) and the non-Treg cell subset (CD4⁺GFP⁻) from congenic Foxp3^GFP-Thy1.1 donors. Cells were mixed in the original ratio (1:5) and transferred into Tcra^{−/ −} hosts; then all mice were inoculated with B16F10 tumors, and 7 days later received pmel-1 cells and gp100 vaccination, with or without 1MT. Figure 4C shows that after vaccination, all of the CD4⁺ cells in tumor-draining LNs expressing proinflammatory cytokines and CD40L derived exclusively from the original Treg cell population; whereas the non-Treg cell population remained quiescent, and contributed none of these activated helper cells.

Thus, in our model of resting CD4⁺ cells acutely exposed to anti-tumor vaccination in a tumor-bearing host, it was the highly-responsive Treg cell population that accounted for virtually all of the vaccine-induced helper-cell response, during the initial 4-day window on which we were focused. Of note, however, the Treg cells were not able to contribute this inflammatory response unless tumor-induced IDO was first blocked by 1MT administration.

Reprogrammed Treg cells support CD8⁺ T cell response to anti-tumor vaccine

We next asked whether blocking IDO also restored the CD8⁺ T cell responses that were progressively lost in hosts with established tumors (see Figure 3). Figure 5A shows that when tumor-bearing mice were treated with 1MT at the time of vaccination, then pmel-1 cells in tumor-draining LNs once again became able to divide in response to vaccination, and could upregulate granzyme B and the chemokine receptor CXCR3 (a functionally important marker of CD8⁺ T cell maturation, because it is required for homing to sites of inflammation). Within the tumor itself, activated, proliferating pmel-1 cells were found in mice treated with 1MT during vaccination, but not in mice without 1MT. As a proxy for functional anti-tumor effect, tumors were dissected at the end of the experiment (day 11) and their size measured, as shown in the bar graph. This was not a measure of long-term tumor regression, because all tumors eventually grew back following a single dose of vaccine; however, the rapid and significant reduction in tumor size compared to controls served as useful experimental confirmation that the proliferating, granzyme B-expressing pmel-1 cells were associated with biologically-relevant anti-tumor activity. In other confirmatory experiments (data not shown) similar effects of 1MT on vaccination were seen in the E.G7 tumor system as well.

Figure 5. Inhibition of IDO enhances efficacy of anti-tumor vaccination by restoring help from reprogrammed Treg cells.

(A) C57BL/6 mice with day 7 established B16F10 tumors received CFSE-labeled pmel-1 cells and vaccination with gp100+CpG+IFA, with or without 1MT in drinking water as shown. Four days later, gated pmel-1 cells (Thy1.1⁺) were analyzed in tumor-draining LN and disaggregated tumor. Representative of 6 experiments with tumors and 11 experiments with tumor-draining LNs. Tumor size (product of orthogonal diameters) is shown after dissection at necropsy (day 11). * p<.01 by t-test (n=16 tumors in 8 separate experiments)

(B) Tcra^{−/ −} hosts received pre-adoptive transfer of Foxp3^GFP Treg cells, or CD4⁺GFP⁻ non-Treg cells, or no CD4⁺ cells, as indicated. Mice were then implanted with B16F10 tumors, and on day 7 received pmel-1 cells and gp100+CpG+IFA vaccine, with or without 1MT as shown. Four days later, pmel-1 cells were analyzed in tumor-draining LN, and tumor size measured at necropsy. Pooled data from 13 experiments (number of tumors indicated in the bar graph, * p<.05 by ANOVA).

(C) Treg cells (CD4⁺CD25⁺) were sorted from GCN2-deficient Eif2ak4^{−/ −} mice or WT control donors and 2 × 10⁵ cells transferred into WT (C57BL/6) hosts. All mice were then implanted with B16F10 tumors, and on day 7 received pmel-1 cells, gp100+CpG+IFA vaccination and 1MT as shown. CFSE-labeled pmel-1 cells were measured in TDLNs on day 11. Representative of 3 experiments; * p<.05 by ANOVA). See also Supplemental Figure S4.

We next asked whether the beneficial effect of 1MT on CD8⁺ T cell responses was specifically due to its ability to restore Treg reprogramming. The pmel-1 system was informative in this regard, because pmel-1 cells are known to require CD4⁺ T cell help in order to achieve optimal anti-tumor efficacy (Antony et al., 2005). In Figure 5B, Tcra^{−/ −} host mice were pre-loaded with different CD4⁺ populations (either sorted Foxp3^GFP Treg cells, non-Treg CD4⁺ cells, or no CD4⁺ cells). All mice were implanted with B16F10 tumors, and on day 7 received pmel-1 cells and gp100 vaccine, with or without 1MT. In the absence of all CD4⁺ cells, pmel-1 cells showed poor response to vaccination, despite 1MT administration. Adoptive transfer of non-Treg CD4⁺ cells provided no detectable help for pmel-1 cells, even with 1MT. Only the Treg cell fraction supported proliferation of pmel-1 cells, granzyme B upregulation, and anti-tumor effect in the presence of 1MT (third dot-plot). The beneficial effects of Treg cells was lost if 1MT was omitted (last dot-plot), consistent with the fact that Treg cells could not undergo normal reprogramming when IDO was active (as shown in Figure 4A, above). Thus, the beneficial effects of 1MT on anti-tumor vaccination appeared heavily dependent on its ability to restore Treg cell reprogramming, and the helper activity they provided.

To further test this hypothesis, we asked whether Treg cells that could undergo reprogramming even in the presence of IDO (i.e., GCN2-deficient Treg cells from Eif2ak4^{−/ −} mice, see Figure 4B, above) would be able to bypass the need for 1MT treatment. C57BL/6 host mice received a cohort of either Eif2ak4^{−/ −} Treg cells or WT control Treg cells, then tumor were implanted and established tumors treated on day 7 with vaccination and pmel-1 cells. Figure 5C shows that the presence of Eif2ak4^{−/ −} Treg cells provided spontaneous help for pmel-1 T cells, even in the face of established tumors, without the need for 1MT treatment. The help provided spontaneously by GCN2-deficient Treg cells (without 1MT) was comparable to wild-type Treg cells treated with 1MT. (There was also a further benefit of adding 1MT to the Eif2ak4^{−/ −} Treg cells, which was expected since the pmel-1 cells and all host cells were still susceptible to IDO).

Taken together, the preceding data suggested that a critical immunologic defect imposed by tumor-induced IDO was the suppression of normal Treg cell reprogramming. This implied that Treg cells might actually play a beneficial role in tumor immunotherapy (if they could be reprogrammed); yet this seemed counterintuitive given that Treg cell depletion has been used to enhance anti-tumor immune responses. (Although in this regard it is not clear that Treg cell depletion offers benefit once tumors become established (Quezada et al., 2008)). To resolve this, we rigorously depleted Treg cells from mice with established tumors, using a diphtheria-toxin receptor driven by the Foxp3 promoter, as shown in Supplemental Figure S4. This allowed inducible depletion of >95% of Treg cells. The effect of this rigorous Treg cell ablation was a modest enhancement of anti-tumor efficacy and pmel-1 responses in the absence of 1MT, but with a complete loss of the beneficial effects of 1MT (which, as we showed above, was dependent on Treg cell conversion). Thus, complete ablation of Treg cells did indeed confer some benefit, but it came at the cost of losing the helper effect of reprogrammed Treg cells conferred by 1MT.

Helper activity of reprogrammed Treg cells for anti-tumor responses requires CD40L

We knew from Figure 2B that reprogrammed Treg cells could activate DCs in a CD40L-dependent fashion. We therefore asked whether CD40L was also required for DC activation when tumor-bearing mice were treated with vaccine+1MT. In resting control mice (without tumors), DCs expressed low basal levels of CD80 and CD86 (Figure 6A, first set of histograms). In mice with established tumors, DC activation was almost completely suppressed following vaccination (second set of histograms). However, if mice were treated with 1MT at the time of vaccination, then DCs expressed high amounts of CD80 and CD86 (third set of histograms). The ability of 1MT to restore expression of CD80 and CD86 was entirely lost in host mice lacking CD40L (fourth set of histograms), suggesting that CD40L was an important downstream mediator of DC activation by 1MT+vaccination.

Figure 6. Helper activity of reprogrammed Treg cells in tumor-bearing hosts is mediated via CD40L.

(A) C57BL/6 or Cd40lg^{−/ −} hosts were implanted with B16F10 tumors, then on day 7 received pmel-1 cells and gp100+CpG+IFA vaccine, with or without oral 1MT as indicated. Four days later, DCs (CD11c⁺) were analyzed in tumor-draining LNs. Control mice received no tumor and no treatment.

(B) Tcra^{−/ −} mice received pre-transfer of a mixed CD⁴⁺ population comprising 1 × 10⁶ non-Treg cells (CD4⁺GFP⁻ cells from Foxp3^GFP mice) plus 2 × 10⁵ Treg cells (enriched as CD4⁺CD25⁺) from either C57BL/6 or Cd40lg^{−/ −} mice. All mice were implanted with B16F10 tumors, and on day 7 received pmel-1 cells and gp100+CpG+IFA vaccine, with or without 1MT as shown. Four days later, pmel-1 cells were analyzed in tumor-draining LNs, and tumor size measured at necropsy (* p<.05 by ANOVA; n per group shown in the figure).

(C) Tcra^{−/ −} mice received pre-transfer of Treg cells (CD⁴⁺CD25⁺) from either WT or Cd40lg^{−/ −} mice, as shown. All mice were implanted with B16F10 tumors, and on day 7 received pmel-1 cells and gp100+CpG+IFA vaccine, with or without 1MT as indicated. One group also received activating anti-CD40 (clone FGK45, as in Figure 2C).

In each panel, data are representative of 3–5 experiments. See also Supplemental Figure S5.

We next asked whether the relevant site of this critical CD40L expression was specifically on reprogrammed Treg cells. Tcra^{−/ −} hosts were pre-loaded with a mixture of Treg cells plus non-Treg CD4⁺ cells (Figure 6B). In half of the mice the Treg cells were taken from Cd40lg^{−/ −} donors, while the other half received CD40L-sufficient (WT) Treg cells. In all mice, the non-Treg cell fraction (GFP⁻) was taken from Foxp3^GFP mice with intact CD40L. Thus, following reconstitution, all host cells and all non-Treg CD4⁺ T cells had intact CD40L, and the groups differed only in whether the Treg cell population could express CD40L. All mice then received pmel-1 cells and gp100 vaccination, with or without 1MT as shown. Figure 6B shows that only those mice receiving CD40L-sufficient Treg cells – but not mice receiving Cd40lg^{−/ −} Treg cells–were able to support full CD8⁺ T cell responses in the presence of 1MT (proliferation, granzyme B expression and anti-tumor activity). Thus, CD40L was required for CD4 helper activity in this model, and the relevant source of CD40L was specifically derived only from the Treg cell population. (In these studies, the CD4⁺CD25⁺ sorted Treg cell preparation contained a small number of contaminating non-Treg cells; however, this was irrelevant because all mice already received a large excess of CD40L-competent non-Treg cells, with no effect.)

Figure 6C shows that the defect in helper activity of Cd40lg^{−/ −} Treg cells could be substantially rescued by treating mice with a cross-linking antibody against CD40. The antibody was not as effective as authentic CD40L, but it could restore both proliferation and granzyme B expression, thus supporting a direct mechanistic contribution of the CD40 pathway.

Finally, Supplemental Figure S5 confirms the preceding results in a second model, asking whether CD40L expression specifically on Treg cells could restore the defective anti-tumor helper activity seen in Cd40lg^{−/ −} hosts. In Cd40lg^{−/ −} hosts with established tumors, adoptively-transferred wild-type Treg cells underwent successful vaccine-induced reprogramming when host mice were treated with 1MT, and this restored the ability of the CD40L-deficient hosts to support anti-tumor CD8⁺ T cell responses. Only the Treg fraction (but not the non-Treg CD4⁺ fraction) was able to restore defective helper activity in Cd40lg^{−/ −} hosts with tumors; and Treg cell reprogramming could occur only if IDO was blocked with 1MT.

Thus, taken together, our data suggested that one key contribution of reprogrammed Treg cells to anti-tumor vaccination was the provision of CD40L-mediated signals. If tumor-induced IDO was active, then Treg cells were prevented from undergoing vaccine-induced reprogramming and could not provide this critical signal. However, if IDO was blocked by 1MT then Treg cells could reprogram, CD40L was expressed, and response to anti-tumor vaccination was enhanced.

DISCUSSION

Using a standard vaccine model, we show that in certain settings reprogrammed Treg cells can be required in order to support the initial (priming) response of CD8⁺ T cells to a cross-presented antigen. Conventional (non-Treg) CD4⁺ cells could supply similar help if they were already pre-activated or pre-immunized. But in a resting mouse exposed for the first time to a new antigen, it was the reprogrammed Treg cells that provided virtually all of the helper activity to support the initial cross-presentation to CD8⁺ cells.

In our system, Treg cells became activated within hours after vaccination, and reprogramming was driven by adjuvant-induced inflammatory signals (CpG, MyD88, IL-6). Traditionally, these early events following vaccination have been considered strictly the province of the innate immune system. Our findings do not conflict with this model, but rather identify reprogrammed Treg cells as an integral downstream component of this “innate” inflammatory cascade. We show that Treg cells, via their ability to rapidly upregulate CD40L, can act as a mechanistic intermediate linking vaccine-induced inflammation with CD40-mediated licensing of DCs.

Our experiments were performed using a CpG-based vaccine, but we hypothesize that the phenomenon may be broadly relevant to any context in which a new antigen is encountered without a large, pre-existing cohort of antigen-specific conventional helper cells (i.e., any priming exposure to a new antigen). Under these conditions, we hypothesize that Treg cells constitute a critically important pool of “pre-formed” helper cells. This role for Treg cells addresses a long-standing paradox in adaptive immunity (Cohn, 2009): namely, how do helper-dependent CD8⁺ immune responses ever get started? The precursor frequency of naive CD4⁺ cells against a new antigen is extremely low (Moon et al., 2007), and clonal expansion takes many days. Yet the initial presentation of “helper-dependent” antigens to CD8⁺ T cells implicitly assumes that CD4⁺ cells are available from the start in order to license the DCs (Bennett et al., 1997; Schoenberger et al., 1998). Where do these early helper cells come from? We now show that a large fraction of Treg cells are constitutively ready to undergo rapid conversion into helper cells in response to innate inflammation, and that, under certain conditions, help from these cells can be indispensable to support initial CD8⁺ T cell responses to a new antigen. Thus, while classical non-Treg CD4⁺ cells are sufficient to provide help under most circumstances, in the specialized case of a new antigen encountered by a naive host, Treg cells can play an unexpectedly critical role as helper cells. This may help explain, in part, the otherwise paradoxical observation that mice in which the Foxp3 lineage is inducibly ablated become unable to mount a protective early immune response to a new viral infection (Lund et al., 2008).

Mechanistically, we hypothesize that local production of IL-6 is a key signal that instructs Treg cells when to reprogram for helper function. Earlier studies implicated IL-6 as a mechanism by which TLR-ligands could “turn off” the suppressive activity of Treg cells (Pasare and Medzhitov, 2003). IL-6 is known to drive Treg cell reprogramming in vitro (Yang et al., 2008), and we found that Il6^{−/ −} mice failed to undergo normal reprogramming in our system. We have also shown that IL-6 production by DCs can be directly suppressed by IDO in an autocrine or paracrine fashion, via IDO-induced activation of the GCN2 pathways in the DCs (Sharma et al., 2009). Now we demonstrate that IDO can also act directly on the Treg cells themselves via activation of their own endogenous GCN2 pathway, to inhibit reprogramming. We have previously shown that GCN2 is an important pathway by which IDO activates Treg cells for enhanced suppression (Sharma et al., 2007); and others have shown that GCN2 suppresses Th17 cell lineage differentiation (Sundrud et al., 2009). Thus, IDO acts to inhibit Treg cell reprogramming both indirectly, by suppressing IL-6 production in DCs, and directly, via the GCN2 pathway in Treg cells.

In mice with established B16F10 tumors, the net effect of the IDO-GCN2 pathway was to dominantly inhibit Treg cell reprogramming after vaccination. For these studies, we chose the B16F10 subline because it is highly immunosuppressive (Shields et al., 2010), and it is also a potent inducer of IDO (Sharma et al., 2007). In this model, IDO suppressed the normal vaccine-induced Treg cell conversion, which proved to be a major mechanism contributing to the inability of vaccination to prime anti-tumor T cells. This inhibition of Treg cell reprogramming was not an absolute defect, since it could be bypassed by providing cognate tumor-specific CD4⁺ help (e.g., TCR-transgenic OT-II cells), and it is certainly not the only defect in tumor-bearing hosts. Nevertheless, IDO-induced suppression of Treg cell reprogramming proved to be a critical and previously unappreciated checkpoint in our tumor system. Moreover, it could be targeted with a clinically-relevant drug (1-methyl-D-tryptophan, 1MT) which is currently in Phase I clinical trials. Granted, not all tumors may be so heavily dependent on IDO as our B16F10 model, and we examined only a single, short-term priming vaccination (in which conventional helper memory cells were not yet available); nonetheless our model serves to make the mechanistic point that, in at least some settings, it may actually be more beneficial to reprogram Treg cells in situ than to rigorously deplete them.

Additional studies will be required to elucidate the antigen specificity of the reprogrammed Treg cells. In general, Foxp3⁺ Treg cells are thought to include many TCRs that recognize self peptides (Fisson et al., 2003) We hypothesize that this recognition of ubiquitously-expressed self antigens allows a large pool of Treg cells to interact spontaneously with APCs during vaccination, thus allowing delivery of help without requiring cognate specificity for the new (vaccine) antigen.

Taken together, our findings suggest a previously unappreciated helper role for cells in the Foxp3⁺ Treg cell lineage. The help provided by converted Treg cells was not in itself different from that provided by conventional CD4⁺ cells; rather, Treg cells were distinguished by their unique ability to deliver help immediately and spontaneously, without needing prior priming or pre-activation. It has long been recognized that Treg cells are unusual in that many Treg cells appear to circulate in a constitutively activated state (Fisson et al., 2003). Resting Treg cells respond much more rapidly to inflammatory signals than do conventional CD4⁺ cells (O'Gorman et al., 2009). We propose that this rapid-response, “hair-trigger” capability may in fact be the fundamental, distinguishing attribute of the Foxp3⁺ lineage. We hypothesize that Foxp3-lineage cells represent a pool of constitutively-primed “first-responder” cells: capable of rapid suppression where that is appropriate; but also able (when directed by the innate immune system) to supply potent helper activity during the early phases of the immune response, at a time when conventional antigen-specific T-helper cells are not yet available.

EXPERIMENTAL PROCEDURES

Reagents and mice

Details of reagents and mouse strains are given in Supplemental online materials. Animal studies were approved by the Institutional Animal Care and Use Committee of the Medical College of Georgia.

Tumors

B16F10 cell line was obtained from ATCC. Tumor implantation was performed as described (Sharma et al., 2007), using 1 × 10⁶ cells in order to ensure rapid tumor growth and immune suppression. Tumor area was measured at necropsy on day 11 as the product of orthogonal diameters following dissection.

Vaccines

CpG-1826 was synthesized as described in Supplemental Materials. Human gp100_25-33 (KVPRNQDWL) was synthesized by Southern Biotechnology (Overwijk et al., 2003). Whole OVA protein was obtained from Sigma (catalog #A-5503). Vaccines were prepared by emulsifying 100 ug of OVA protein, or 25 ug gp100 peptide, with 50 ug CpG-1826 in incomplete Freund’s adjuvant (Sigma F-5506) and administered in the footpad.

Adoptive transfer

For CD8⁺ T cell transfers, OT-I or pmel-1 spleen cells were enriched for CD8⁺ cells by magnetic bead isolation (Miltenyi Biotech), labeled with CFSE as described (Munn et al., 2005), and 2 × 10⁶ cells injected i.v. All CD4⁺ transfers were sorted from spleens of donor mice by MoFlo cell-sorter with doublet discrimination (>95% post-sort purity). Treg cells (usually 2 × 10⁵ cells unless otherwise indicated) or non-Treg cells (CD4⁺GFP⁻, 1 × 10⁶ cells) were injected i.v. For Cd40lg^{−/ −} and Eif2ak4^{−/ −} donors, Treg cells were enriched by sorting for CD4⁺CD25⁺ cells (>90% Foxp3 cells by intracellular staining). All experiments using the non-Treg cell fraction were always sorted based on Foxp3^GFP fluorescence, to rigorously exclude any Treg cells.

FACS staining

Details of staining and the low-dose PMA activation protocol are given online. LNs were disaggregated by passage through 40 um mesh. Tumors were disaggregated as described (Sharma et al., 2009).

Statistical analysis

Multiple treatment groups were compared by ANOVA with Tukey’s HSD test.