Tumor Tolerance–Promoting Function of Regulatory T Cells Is Optimized by CD28, but Strictly Dependent on Calcineurin

CD4⁺ Foxp3⁺ regulatory T cells (Treg) play an essential role in maintaining immune homeostasis and regulating antipathogen responses, but their physiological functions also restrict immune responses against malignant tumors and thus promote tumor progression (1). Therapeutic depletion of Treg enhances immunological tumor control in preclinical models, but sustained and effective depletion in cancer patients is predictably limited by the ensuing systemic autoimmunity. More selective approaches are needed to target specifically those Treg that are engaged in the antitumor response. This will not only require Treg-selective drug targeting approaches, but also an understanding of the pathways that regulate the maintenance and functions of different populations of Treg at various states of activation and in different tissues.

Treg are composed of subpopulations with different migration patterns (2, 3). CD44^low “resting” or “central” Treg (cTreg) express the homing receptors CD62L and CCR7 and recirculate through secondary lymphoid organs (SLOs). In contrast, CD44^hi “activated” or “effector” Treg (eTreg) lack CD62L and CCR7, but proliferate more rapidly and express chemokine receptors that endow them with the ability to enter nonlymphoid and inflamed tissues. They also express elevated levels of proteins required for their maintenance and suppressive function, such as ICOS and CTLA-4. Treg that newly emerge from thymic development have a cTreg phenotype and can give rise to eTreg, whereas eTreg are not able to revert to cTreg (3, 4). cTreg to eTreg conversion occurs during immune homeostasis, depends on cognate Ag recognition, and is enhanced by inflammatory signals (3). During antitumor responses, self Ag-specific Treg rapidly expand in tumor-draining lymph nodes (tdLNs) in an MHC II–dependent manner (5). Treg found in tumor tissue phenotypically resemble eTreg and are highly suppressive (6).

We have previously investigated Treg responses to tumor growth by tracking TCR-transgenic clonal Treg populations and found that TCR-driven activation of cTreg in tdLNs is a prerequisite for their subsequent migration to tumors (7). We also observed that upon entry into tumor tissue, only those eTreg that locally re-encounter their cognate Ag are able to regulate the function of tumor-infiltrating CTL and prevent tumor rejection (7). Therefore, Treg depend on TCR signals both at immune induction sites for Treg activation and at immune effector sites for their suppressive activity. In further support of a role of TCR signals in regulating cTreg to eTreg differentiation, genetic deletion of TCRs in mature, polyclonal Treg has a greater impact on the maintenance and gene expression pattern of eTreg than cTreg (8, 9). These observations support a scenario whereby TCR signals, potentially in concert with effector T cell–derived IL-2 (2, 3, 5, 10–12), induce cTreg to differentiate into eTreg and sustain their suppressive activity (2–4, 13). In addition, CD28 costimulation, which both complements and amplifies TCR signals and promotes cell survival, is not only essential for thymic Treg development, but also important for optimal Treg function, because mice with targeted deletion of CD28 in mature Treg slowly develop autoimmunity (14–16).

Although the general importance of TCR signals for Treg function is therefore established, the role of various signaling pathways activated by the TCR and costimulatory receptors for their ability to promote immune tolerance is less clear. Differences in the signaling response of Treg and conventional T cells to TCR stimulation have been noted (17–19), but whether they differ in their dependence on specific pathways for their functions in vivo, and which pathways regulate which functions of mature Treg, is unknown.

A major TCR-dependent pathway essential for effector T cell differentiation and function is Ca²⁺- and calcineurin-dependent NFAT activation. Classic studies on calcineurin inhibitor therapy showed that calcineurin supports Treg suppressive activity in vivo, but were unable to separate a direct inhibitory effect on Treg from indirect effects, for example, through blockade of IL-2 production by conventional T cells or APCs (20). Meanwhile, reports of residual suppressive activity in Treg lacking one or two of the three T cell–expressed NFAT family proteins (NFAT1, 2, and 4) put the importance of this pathway into question (21–23). In addition, other studies suggested that NFAT activity in Treg is at least partly TCR-independent and constitutive (24, 25).

In this study we have used genetic deletion of the regulatory B subunit of calcineurin (CnB) or of CD28 specifically in mature Treg to investigate their role in the activation and differentiation of eTreg that promote tumor tolerance. We found that without CnB, Treg are unable to activate NFAT1 and 2. Although their suppressive activity is only marginally reduced in vitro, and their numbers at least in peripheral lymph nodes (LNs) are comparable to wild-type (WT) mice, CnB-deficient Treg fail to maintain immunological tolerance, to differentiate into tumor-homing eTreg, and to suppress immune responses against aggressive tumors, which as a result are controlled without additional intervention. CD28, on the other hand, in contrast to its essential role on the activation of conventional T cells, is not critical for the formation, but augments the fitness of tumor-homing eTreg. Accordingly, CD28 deletion in Treg does not abrogate, but considerably reduces their ability to promote tumor growth. Our findings identify an essential role for calcineurin to activate NFAT proteins in Treg and induce TCR-dependent expression of genes that characterize the transition from cTreg to tumor-homing eTreg. CD28, on the other hand, enables tumor tolerance through a fitness-optimizing role in eTreg.

CnB^fl/fl (26), Foxp3^YFP-Cre (27), Foxp3^CreERT2 (28), Foxp3^DTR (29), TCRα^−/−, C57BL/6, and BALB/c mice were from Jackson Laboratories. CD28^fl/fl mice and pgk-HA × TCR-HA double-transgenic mice as a source of hemagglutinin (HA)-specific Treg were previously described (14, 30). All animals were housed, bred, enrolled in authorized experiments, and euthanized according to the guidelines of the Institutional Animal Care and Use Committee at Massachusetts General Hospital. In particular, animals were euthanized when mandated by a reduced state of health, with guidance from veterinary staff. The age at euthanasia was recorded in Kaplan­–Meier plots.

Tumor growth was longitudinally measured following s.c. injection into the flank of mice of 3–5 × 10⁵ MCA-205, 10⁶ MC38, or 10⁶ CT26-HA tumor cells.

To delete floxed CnB or CD28 genes in CnB^fl/fl × Foxp3^CreERT2 or CD28^fl/fl × Foxp3^CreERT2 mice, 1 mg tamoxifen (Sigma-Aldrich) dissolved in 10% ethanol/90% olive oil was injected i.p. for five consecutive days.

To create mixed bone marrow chimeras, 6 × 10⁶ bone marrow cells from Foxp3^DTR mice were injected together with 12 × 10⁶ bone marrow cells from CnB ΔTreg, CD28 ΔTreg, or Foxp3^YFP-Cre mice into TCRα^−/− hosts irradiated with 400 rad. After 8 wk, healthy mice were implanted with tumors and received diphtheria toxin (Calbiochem) i.p. at 25 μg/kg on the first day, followed by daily injections of 5 μg/kg.

Single cell suspensions from spleen and LNs of CnB ΔTreg^het, CD28 ΔTreg^het, or Foxp3^YFP-Cre/wt mice were enriched for Treg through immunomagnetic selection of CD4⁺ CD25⁺ cells (Miltenyi). YFP⁺ CD44^low CD62L^hi cTreg and YFP⁺ CD44^hi CD62L^neg eTreg were further purified by FACS. Varying numbers of cTreg or eTreg were added to 2.5 × 10⁴ T cell–depleted splenocytes as APCs and 1 × 10⁴ CD4⁺ CD25⁻ T cells labeled with 5 μM CellTrace Violet (CTV) as responder cells, and stimulated with 250 ng/ml of anti-CD3 mAb (clone 145-2c11; BioLegend). Responder cell proliferation (CTV dilution) and absolute numbers of remaining Treg was read out after 72 h. To quantify proliferation, we extrapolated the number of progenitor cells giving rise to each CTV peak by dividing the number of cells in each peak by the number of cells that would have originated from a single precursor (e.g., four cells in the peak corresponding to two divisions would have derived from one progenitor). To compute the T cell fold increase, we then divided the total number of recorded Treg by the extrapolated total number of progenitors. Percentage of suppression was scaled from 0 (proliferation of responders in absence of Treg) to 100 (complete absence of proliferation).

To follow cTreg proliferation, an identical assay was set up, but Treg, not responder T cells, were labeled with CTV.

Pooled splenocytes and LN cells were prestained with anti-CD25 BV421 mAbs (PC61; BioLegend) and stimulated at 37°C using plate-immobilized anti-CD3 (10 μg/ml, clone 145-2c11) and soluble anti-CD28 (10 μg/ml, clone 37.51) Abs. At different time points, aliquots were quickly transferred into a 2-fold excess of 4% paraformaldehyde (PFA), yielding a final PFA concentration of 2.6%, incubated at room temperature for 10 min, and washed in PBS/0.5% BSA. Cells were permeabilized by dropwise addition of 1 ml ice-cold methanol and incubation on ice for 20 min, then carefully washed twice in PBS/0.5% BSA. Staining for CD4 (100 ng/ml, GK1.5; BioLegend), YFP (Alexa Fluor 488–conjugated anti-GFP/YFP polyclonal rabbit IgG; Life Technologies), Foxp3 (FJK-16s; eBioscience), pERK Thr202/Tyr204 (197G2; Cell Signaling), pS6 Ser235/236 (D57.2.2E; Cell Signaling), and IκB (L35A5; Cell Signaling) was performed at room temperature for 1 h, followed by careful washing. For quantification, we calculated the percentage of marker-positive (pERK) or -negative (IκB) cells. For pS6, we calculated the product of percentage of positive cells and mean fluorescence intensity (MFI) of positive cells (referred to as MFI*) (30) to account for both the frequency of signaling cells and their content of phosphoprotein. To permit comparisons across experiments, we normalized each parameter in individual experiments by assigning a value of 100 to the maximal recorded response.

Spleen and LN cells from control, CnB ΔTreg, or CD28 ΔTreg mice before onset of overt autoimmunity were kept on ice immediately upon harvest, during enrichment of Treg using the Miltenyi mouse Treg selection kit, and dead cell staining with Zombie Red (BioLegend). Some samples were stimulated with ionomycin (1 μg/ml; Sigma) for 10 min at 37°C, whereas others were kept on ice to preserve their in situ NFAT activation state. All samples were allowed to adhere to poly-l-lysine–coated cover glasses for 5 min at 37°C, then fixed in 4% PFA for 10 min and stained with anti-NFAT1 (D43B1; Cell Signaling), anti-NFAT2 (7A6; BioLegend), and anti-Foxp3 (FJK-16s; eBioscience) Abs. Primary Ab binding was revealed using Alexa Fluor 633–conjugated anti-rabbit IgG, Alexa Fluor 647–conjugated anti-mouse IgG Fab (Jackson Immunoresearch), and Alexa Fluor 488–conjugated anti-rat IgG (Invitrogen), respectively. Slides were mounted using ProLong antifade medium (Thermo Fisher) and analyzed on a Zeiss LSM510 confocal microscope. Nuclear localization of NFAT proteins was quantified as the signaling index of individual cells (30), which has a value of 0 for full cytoplasmic localization and 1 for complete nuclear localization.

Splenocytes and LN cells from CnB ΔTreg^het, CD28 ΔTreg^het, or Foxp3^YFP-Cre/wt mice were assayed immediately after isolation or after 6 h of culture at 37°C, to reveal Treg susceptibility to apoptosis, as described (3). Viability of CD4⁺ YFP⁺ CD62L^hi CD44^low cTreg was read out by Annexin V and 7-AAD staining, performed as per the manufacturer’s instructions (BioLegend).

Tumors, ear skin, lungs, liver, and colon were finely minced and processed with tissue-specific protocols as follows: tumor tissue was digested in RPMI 1640 5% FCS containing 1.5 mg/ml collagenase IV and 50 U/ml DNase I for 30 min; skin was digested in DMEM 2% FCS 10 mM HEPES containing 0.5 mg/ml hyaluronidase, 1.5 mg/ml collagenase IV, and 50 U/ml DNase I for 1 h; lungs were digested in DMEM 2% FCS 10 mM HEPES containing 0.5 mg/ml hyaluronidase, 1.5 mg/ml collagenase IV, and 50 U/ml DNase I for 20 min; and liver was digested in RPMI 1640 5% FCS supplemented with 1 mg/ml collagenase IV for 30 min. To isolate lymphocytes from the colon lamina propria, intraepithelial lymphocytes were eliminated by two 20 min washes in RPMI 1640 5% FCS 1.5 mM EDTA 1 mM DTT at 37°C. After a brief wash in PBS to remove EDTA, the tissue fragments were incubated in RPMI 1640 5% FCS containing 1 mg/ml collagenase IV for 45 min. All digestions were performed at 37°C under agitation, residual aggregates were mechanically disrupted, and cell suspensions were filtered before immunophenotyping.

Dead cells were stained using the fixable viability dye Zombie Red according to the manufacturer’s instructions. Chemokine receptors CCR4 (clone 2G12), CXCR3 (CXCR3-173), and CCR5 (HM-CCR5) were stained for 1 h at 37°C in cell culture medium. Staining for CD4 (GK1.5), CD25 (PC61), CD44 (IM7), CD45 (30F11), CD62L (MEL14), and ICOS (C398.4A) was performed at 4°C for 20 min. All these Abs were from BioLegend. To detect intracytoplasmic and intranuclear Ags, cells were fixed and permeabilized using the Mouse Regulatory T cell Staining Kit from eBioscience, and stained for Ki67 (16A8; BioLegend), CTLA-4 (UC10-4B9; BioLegend), GFP/YFP (Alexa Fluor 488–conjugated polyclonal rabbit IgG; Life Technologies), and Foxp3 (FJK-16s; eBioscience).

For some experiments, as indicated, blood-borne were distinguished from tissue-resident Treg by an i.v. injection of 3 μg CD45.2-PE mAb (clone 104) 3 min before euthanasia.

DNA of Foxp3^YFP-Cre or CnB ΔTreg mice was extracted either from tail tissue or from CD4⁺ Foxp3⁺ Treg sorted to >99% purity from LNs using the Arcturus PicoPure DNA extraction kit from Applied Biosystems. A fragment of the CnB locus containing the CnB open reading frame (and both flanking loxP sites in CnB ΔTreg) was amplified by PCR using the primers: 5′-CAATGCAGTCCGCTGTAGTTC-3′ and 5′-AGCCTCCACATACACAGATAC-3′.

For data passing the Kolmogorov–Smirnov normality test, experimental groups were compared through Student t test. Otherwise, Mann–Whitney U test was used.

A fraction of Treg in LNs continuously recognizes their cognate tissue-derived Ags (4, 31, 32), which is thought to enhance their suppressive function (8, 9) and to maintain immune homeostasis by extinguishing incipient autoimmune effector responses (33, 34). We sought to examine in this context the role of TCR- and calcineurin-dependent NFAT activation on the one hand, and of CD28-dependent complementation or amplification of TCR signals on the other hand. We therefore crossed Foxp3^YFP-Cre knock-in mice (27) with animals in which both alleles either of the regulatory B subunit of calcineurin (CnB) (26) or of CD28 (14) genes were flanked by loxP sites to create Foxp3^YFP-Cre × CnB^fl/fl (henceforth “CnB ΔTreg”) or Foxp3^YFP-Cre × CD28^fl/fl (“CD28 ΔTreg”) mice. Because Foxp3 expression is initiated during the final steps of thymic Treg development (35), CnB or CD28 are efficiently deleted in mature Treg in these animals (Supplemental Fig. 1A) (14). CnB ΔTreg mice were born at Mendelian ratios and did not show macroscopic abnormalities until their fourth week of life, when they started to develop a hunched posture, scaly skin, crusted ears, and alopecia, and stopped thriving, whereas CD28 ΔTreg mice remained indistinguishable from littermates (Fig. 1A, 1B). CnB ΔTreg mice, but not CD28 ΔTreg mice, required euthanasia at a median age of 32 d because of a rapidly progressive wasting disease (Fig. 1C). In contrast to their normal appearance, CD28 ΔTreg mice already showed moderate splenomegaly and lymphadenopathy at 4 wk of age, but these signs were more prominent in CnB ΔTreg mice (Fig. 1D). Concomitantly, expansion of conventional CD8⁺ and CD4⁺ T cells with a CD44^hi CD62L^neg effector phenotype was already detectable at 2 wk of age in CnB ΔTreg mice, and more pronounced in these animals at 4 wk, when it became first detectable in CD28 ΔTreg mice (Fig. 1E, Supplemental Fig. 1B). Histological examination of nonlymphoid tissues of CnB ΔTreg mice revealed severe inflammatory infiltrates and associated pathological changes in the lungs, skin, and liver (Fig. 1F). Heart, kidney, adrenal glands, eye, and brain lacked infiltrates, whereas stomach and small and large intestines were only marginally affected (data not shown). By contrast, lungs of CD28 ΔTreg mice showed only focal inflammatory cell infiltrates and the majority of lung tissue remained unaffected. Skin and liver were marginally infiltrated (Fig. 1F), and all other organs appeared normal at this age (data not shown). Thus, whereas CD28 ΔTreg mice largely maintain immune homeostasis, and only later in life develop a skin-focused, nonlethal autoimmune disease (14, 15), CnB ΔTreg mice rapidly develop a lymphoproliferative inflammatory syndrome that, with regard to kinetics and organ distribution, closely resembles the scurfy phenotype in animals that either express a nonfunctional Foxp3 protein (36), in which Foxp3 is deleted (37), or in which Treg are ablated starting at birth (29).

Preceding and at the onset of overt disease in CnB ΔTreg mice at 4 wk of age, the frequency of Treg in LNs was only moderately reduced in CnB ΔTreg and CD28 ΔTreg mice, if at all (Fig. 2A), indicating that a Treg functional defect, not a defect in maintenance, causes lymphoproliferative disease in these animals. In contrast to LNs, we observed a clear and progressive reduction in Treg frequencies in spleens of both strains. This may result in part from preferential accumulation of the expanding populations of CD44^hi CD62L^neg conventional effector and effector memory T cells in spleens, including the splenic red pulp, as opposed to LNs (38), causing a relative decrease in Treg frequencies mostly in spleens. In addition, if splenic accumulation of conventional effector and effector memory T cells is in part based on their homing preferences, not local activation, they likely will produce less IL-2 to enable a concomitant local expansion of Treg to maintain their density (33). Thus, failed immune homeostasis in CnB ΔTreg and CD28 ΔTreg mice cannot be explained by Treg absence. However, when we examined the Treg compartment in SLOs for the proportion of CD44^low CD62L^hi cTreg and CD44^hi CD62L^neg eTreg, we found that eTreg were underrepresented in LNs and spleen of CnB ΔTreg mice, whereas the ratios of cTreg and eTreg were normal in CD28 ΔTreg mice (Fig. 2B). These observations suggested a role for calcineurin activity in the TCR-dependent formation of eTreg, which may contribute to failed immune homeostasis in CnB ΔTreg mice despite the presence of Foxp3⁺ Treg.

To directly compare the function of calcineurin- and CD28-deficient cTreg as well as eTreg, we examined their in vitro suppressive activity separately. In order to avoid the potentially confounding effects of overt or subclinical systemic inflammation in CnB ΔTreg and CD28 ΔTreg mice on the function of Treg, we generated female Foxp3^YFP-Cre/wt × CnB^fl/fl (“CnB ΔTreg^het”) and Foxp3^YFP-Cre/wt × CD28^fl/fl (“CD28 ΔTreg^het”) animals. Because the Foxp3 locus is located on the X chromosome, random X inactivation will predictably lead to expression of YFP-Cre and deletion of CnB or CD28 in only half of all Treg (YFP⁺), whereas the other half (YFP⁻) does not delete these genes and maintains immune homeostasis (Fig. 2C). These mice had normal lifespans (Supplemental Fig. 2A) and remained healthy with no signs of conventional effector T cell expansion (Supplemental Fig. 2B) or lymphocytic tissue infiltration as observed in CnB ΔTreg and CD28 ΔTreg mice (Supplemental Fig. 2C). However, we noted that consistently <50% of Treg expressed YFP-Cre in lymphoid tissues even in the absence of any floxed allele (Supplemental Fig. 2D), potentially reflecting toxic effects of the Cre recombinase on cryptic loxP sites in the mouse genome (39). To control for this possibly confounding effect, we used YFP-Cre–expressing Treg from Foxp3^YFP-Cre/wt mice as controls (“WT Treg”).

When we cocultured purified YFP⁺ cTreg from CnB ΔTreg^het, CD28 ΔTreg^het, or Foxp3^YFP-Cre/wt mice for 3 d with CD4⁺ responder T cells and T cell–depleted splenocytes in the presence of anti-CD3 Abs, we found that CnB-deficient cTreg suppressed responder cell proliferation at almost normal levels, and defective suppression only became evident at low cTreg:responder ratios (Fig. 2D, 2E). Of note, reduced suppressive activity of CnB-deficient Treg correlated with lower Treg numbers retrieved at the end of the 3-d assay (Fig. 2F). In contrast, CD28-deficient cTreg functioned normally and accumulated similarly to controls (Fig. 2E, 2F). Lower numbers of CnB-deficient Treg did not appear to result from enhanced apoptosis (Fig. 2G), but mostly from a pronounced proliferation defect (Fig. 2H). This suggested that poor proliferation and accumulation of CnB-deficient Treg may primarily account for their reduced in vitro–suppressive activity. Therefore, to determine the suppressive capacity of CnB-deficient and CD28-deficient cTreg on a per cell basis, we quantified their suppression of responder cell proliferation as a function of their numbers at the end of the assay. By this measure, both CnB-deficient and CD28-deficient cTreg showed near-normal suppressive activity, as they allowed only ∼20 and ∼10% more responder T cell proliferation than control cells, respectively (Fig. 2I). eTreg were more potent suppressors than cTreg on a per cell basis, as expected, and their function was not detectably impaired in absence of CD28 (Fig. 2I). CnB-deficient eTreg, because of their paucity in CnB ΔTreg^het mice, could not be isolated in sufficient numbers for functional testing. Together, these observations reveal no or only minor defects in cTreg lacking CnB or CD28. Because prior studies correlated similarly mild in vitro–suppressive defects in Treg with no or only mild autoimmune phenotypes (40, 41), we hypothesized that failed immune homeostasis, and especially the scurfy phenotype in CnB ΔTreg mice, may result mostly from a reduced capacity of CnB- or CD28-deficient eTreg to migrate to and persist in nonlymphoid tissues.

In contrast to cTreg, which primarily express the chemokine receptors CCR7 and CXCR4, whose ligands are expressed in lymphoid tissues, eTreg are responsive to chemokines that guide their migration to nonlymphoid tissues (2, 11, 42, 43). For instance, genetic deletion of CCR4 in Treg strongly impairs their capacity to migrate into skin and lung, and leads to severe skin and pulmonary inflammation (13). This suggests that immune homeostasis not only requires Treg activity in SLOs, where immune responses are induced, but also in nonlymphoid tissues. We therefore hypothesized that impaired eTreg differentiation in absence of CnB or CD28 may account for reduced accumulation in nonlymphoid tissues, causing the widespread tissue inflammation described above. To test this hypothesis, we examined female CnB ΔTreg^het and CD28 ΔTreg^het mice, where tissue distribution of YFP⁺ Treg lacking either CnB or CD28 can be studied in a noninflammatory context (see Fig. 2C, Supplemental Fig. 2). Here, the frequency of YFP⁺ Treg was decreased in LNs of CnB ΔTreg^het relative to control animals (Fig. 3A), suggesting that the normal Treg numbers in LNs of CnB ΔTreg mice are sustained in part by inflammation. Moreover, specifically the proportion of CD44^hi eTreg was diminished in these animals (Fig. 3B, Supplemental Fig. 2E). To investigate eTreg formation in more detail, we measured the expression of the Treg activation markers ICOS and CTLA-4 and of the cell cycle entry marker Ki67. Basal expression of ICOS and CTLA-4 in cTreg was unchanged. However, their elevated expression in WT eTreg was diminished by nearly 10-fold in the absence of CnB, and ∼2-fold in the absence of CD28. Ki67 was equally reduced in both cases (Fig. 3C). Furthermore, CXCR3, CCR4, and CCR5, chemokine receptors that are expressed in a fraction of CD44^hi eTreg and confer tropism for nonlymphoid tissues, were undetectable in Treg lacking CnB, but only reduced in those lacking CD28 (Fig. 3C), suggesting a critical role for CnB-driven NFAT activity in inducing expression of eTreg-associated chemokine genes. To examine this further, we parsed a published NFAT1 chromatin immunoprecipitation sequencing data set obtained from CD8⁺ T cells (44) to determine the distance from the transcription start sites (TSSs) of chemokine receptor genes to the closest NFAT1 binding sites. In comparison with all other genes in this data set, or a collection of control genes implicated in stem cell function, TSS-NFAT1 binding site distances for chemokine receptor genes were very short, with 80% being <10 kb (compared with ∼40% of the stem cell genes and 36% for all genes), and 56% being <2 kb (compared with 25 and 20% for stem cell–related and all genes) (Supplemental Fig. 3A). Specifically, CCR4, CXCR3, and CCR5 had NFAT1 binding sites within 72, 1763, and 4709 bp of their TSSs (Supplemental Fig. 3). This suggests that the CnB/NFAT axis directly regulates chemokine receptor gene expression in T cells, and that CnB-deficient Treg fail to induce CXCR3, CCR4, and CCR5 at least in part because of impaired transactivation of these genes by NFAT proteins. Treg that lacked costimulation via CD28 also expressed these chemokine receptors less frequently, but those that did, expressed them at similar levels as control Treg (Fig. 3C).

Impaired eTreg formation predicts reduced Treg accumulation in nonlymphoid tissues. Indeed, in all nonlymphoid tissues studied, including those that are most severely inflamed in CnB ΔTreg mice, we noted that CnB-deficient YFP⁺ Treg were essentially absent (Fig. 3D). In contrast, the reduction of Treg lacking CD28 was milder and more variable, and most pronounced in skin, the tissue that is most affected by the inflammatory disease that CD28 ΔTreg mice develop later in life (14, 15). These findings suggest a scenario whereby TCR-driven CnB activity is critical and CD28 costimulation optimizes the formation of eTreg whose migration to and survival in nonlymphoid tissues is required for immune homeostasis.

The effects of CnB- or CD28-deletion on the ability of Treg to populate nonlymphoid sites may also restrict their migration and persistence in tumor tissue, and thereby limit their ability to establish local tumor tolerance. To test this hypothesis, and also to examine the role of CnB in enabling the formation of eTreg and their trafficking into a newly forming immune effector site, we implanted female CnB ΔTreg^het mice with fast-growing MCA-205 mesenchymal tumors. Eleven days after tumor implantation the frequency of YFP⁺ Treg in tdLNs of control and CnB ΔTreg^het mice was comparable (Fig. 4A). However, CnB-deficient Treg were absent from tumor tissue, suggesting that without calcineurin activity, Treg cannot acquire the functions that enable them to enter and survive in tumor tissue. Similar to Treg in which TCRs are deleted (8, 9), we found a small but consistent reduction in Foxp3 expression in CnB-deficient Treg in tdLNs (Fig. 4B), which possibly reflects the reported role of CnB-dependent NFAT activity on CNS2 to stabilize Foxp3 expression (45). We also noted that although essentially all CnB-sufficient Treg in tumor tissue and more than 50% in tdLNs were CD44^hi CD62L^neg eTreg, <20% of CnB-deficient Treg in tdLNs showed this phenotype, and those that did had only partially downregulated CD62L (Fig. 4C). Hence, lack of CnB limits, but does not abolish, CD44 up- and CD62L downregulation, cardinal features of T cell activation.

Because migration of Treg into tumor tissue likely requires appropriate trafficking receptors, we analyzed expression of chemokine receptors CXCR3, CCR4, and CCR5, which have previously been implicated in Treg accumulation in tumor tissue (46–48). All three receptors were expressed by varying fractions of WT CD44^hi eTreg in tdLNs (Fig. 4D), and the majority of Treg in tumor tissue expressed these receptors, demonstrating that their expression correlated with the ability of Treg to accumulate in tumors. In contrast, CnB-deficient Treg completely failed to express any of these chemokine receptors (Fig. 4D). Hence, failure to induce expression of appropriate trafficking molecules is one likely reason for their inability to enter tumor tissue.

In addition to trafficking receptors, eTreg express elevated levels of CTLA-4, required for their suppressive function (49), and ICOS, which has been suggested to promote their survival (3). Strikingly, the rare CD44^hi CnB-deficient Treg in tdLNs completely failed to induce expression of ICOS and CTLA-4 above baseline levels found in cTreg, and their proliferation was strongly impaired (Fig. 4E). In contrast, all tumor-infiltrating WT Treg expressed high levels of ICOS, CTLA, and the cell cycle entry marker Ki67, suggesting that these features are either correlated with or a prerequisite for accumulation in tumor tissue.

These observations suggest, but do not formally prove that CD44^low cTreg convert into CD44^hi eTreg during tumor challenge upon TCR-dependent activation in tdLNs. To more stringently test this model, we seeded mice with clonal populations of TCR-transgenic cTreg specific for the model Ag influenza HA (7, 30) and tracked their proliferation and concomitant differentiation into eTreg, as well as their accumulation in tumors expressing HA. We found that cTreg upregulated CCR4, CCR5, and CXCR3, whereas downregulating CCR7 (Supplemental Fig. 4A), and also upregulated CTLA-4 and ICOS (Supplemental Fig. 4B) during Ag-driven clonal expansion in tdLN. Similarly to polyclonal Treg, only cells that had undergone extensive proliferation and expressed the most CTLA-4 and ICOS accumulated in tumor tissue. Thus, TCR signaling and, more specifically, CnB activation play a cell-intrinsic and nonredundant role in driving cTreg to eTreg differentiation in tdLNs, and their subsequent accumulation within tumors.

In conventional T cells, calcineurin is essential for TCR-mediated activation and nuclear accumulation of the T cell–expressed NFAT proteins NFAT1, NFAT2, and NFAT4. However, recent reports have questioned the generality of this mechanism by describing a pathway of NFAT activation through IL-7 in thymocytes (50), as well as constitutive, TCR- and calcineurin-independent NFAT activity in Treg (21–25). To examine the mechanism of NFAT activation in Treg in SLOs, we purified Treg under conditions optimized to preserve their in situ NFAT activation state and examined subcellular localization of NFAT1 and NFAT2. Both proteins were nuclear in a sizeable fraction of WT Treg. In contrast, NFAT1 and NFAT2 proteins were cytoplasmic in essentially all CnB-deficient Treg (Fig. 5A, 5B). Treatment with the Ca²⁺ ionophore ionomycin triggered further nuclear translocation of NFAT1 and 2 in the majority of WT Treg, but was completely ineffective in CnB-deficient Treg (Fig. 5A, 5B). Similar observations were made under stimulation of Treg with Con A, which crosslinks the TCR and various costimulatory receptors (data not shown). Hence, a fraction of Treg in lymphoid tissues receives signals that activate NFAT1 and NFAT2 in the steady state, but this activity, as well as additional TCR and Ca²⁺-induced activity, is fully dependent on calcineurin.

In addition to NFAT proteins, calcineurin might help to activate other signaling activities, such as the NF-κB pathway via the CARMA1/Bcl10/MALT1 complex (51). To investigate these functions of calcineurin in Treg, we conducted phosphoprotein analysis by flow cytometry on YFP⁺ Treg from CnB ΔTreg^het and Foxp3^YFP-Cre/wt mice. Upon TCR stimulation of CnB-deficient cells, we observed no defect in Ras/ERK activation, no defect in mTORC1 activity as reflected by S6 phosphorylation, and slightly enhanced, but not diminished, NF-κB activity, as reflected by degradation of IκB (Fig. 5C, 5D). Similar results were obtained using suboptimal doses of anti-CD3 stimulation, ruling out a masking effect of excessive TCR stimulation at high anti-CD3 doses, as well as with pharmacological stimulation by PMA and ionomycin (data not shown). Hence, activation of NFAT1 and 2 is fully dependent on calcineurin activity, and NFAT activation is the primary effector pathway in Treg, suggesting that the failure of CnB-deficient Treg to form eTreg and maintain immune homeostasis mainly results from lack of NFAT activity.

In light of the critical need for TCR-driven NFAT activation via calcineurin for induction of the eTreg trafficking program and elevated suppressive function, we asked whether costimulation via CD28 had an augmenting and qualitatively similar or a qualitatively distinct role in this process. CD28 is critical for the formation of conventional effector T cells and has been described as a general amplifier of TCR-induced signaling pathways, including the activation of NFAT proteins (52, 53), but is also critical for triggering distinct activities, including maximal NF-κB activation (54, 55). To first determine if CD28 synergizes with calcineurin in NFAT activation in Treg, we examined YFP⁺ Treg from SLOs of CD28 ΔTreg^het mice. In situ activation of NFAT1 and 2 and their ionomycin-triggered nuclear translocation were normal in CD28-deficient Treg (Fig. 6A, data not shown). In contrast, and consistent with previous analyses on conventional T cells (54, 55), we found CD28 to be critical for TCR-dependent NF-κB activation in mature Treg (Fig. 6B). Therefore, in Treg, CD28 costimulation regulates TCR-dependent signaling activities distinct from those that depend on calcineurin (Fig. 5D).

To determine the role of CD28 in the formation of eTreg that can infiltrate tumor tissue, we compared their distribution in CD28 ΔTreg^het and Foxp3^YFP-Cre/wt control mice implanted with MCA-205 tumors. In contrast to the normal frequency of eTreg in LNs of CD28 ΔTreg or CD28 ΔTreg^het mice (Figs. 2B, 3B), we observed a 50% decrease in CD28-deficient eTreg in tdLNs of CD28 ΔTreg^het mice (Fig. 6C). This difference may result from the fact that in CD28 ΔTreg^het mice, CD28-deficient Treg have to compete with CD28-sufficient Treg for potentially limiting resources, for example, ICOS ligands (3), which restrict their survival as eTreg. However, in contrast to CnB-deficient Treg, CD28-deficient eTreg migrated to tumor tissue (Fig. 6C), albeit at a frequency of <10% of normal (data not shown). Although fewer CD28-deficient Treg expressed CXCR3, CCR4, and CCR5 in tdLNs, those that did, expressed these receptors at the same levels as WT eTreg (Fig. 6D). Similar observations were made for CTLA-4 and ICOS: CD28-deficient eTreg expressing these proteins were less abundant, but expressed them and proliferated at the same levels as WT eTreg (Fig. 6E).

In line with a selection of Treg responsive to inflammatory chemokines for tumor entry, the majority of CD28-deficient Treg in tumor tissue expressed CXCR3, CCR4, and CCR5 (Fig. 6F). CCR5 expression was slightly reduced in tumor-infiltrating Treg lacking CD28, suggesting that this receptor is less important for entry or local persistence than CXCR3, which was normal, or CCR4, which was slightly elevated (Fig. 6F).

Interestingly, in contrast to tdLN, eTreg expression of ICOS and, more variably, CTLA-4 was reduced, indicating that CD28 costimulation is necessary to sustain optimal Treg function at effector sites (Fig. 6G). Thus, although CnB-deficient Treg are unable to acquire to any degree the state of elevated functional activity and trafficking capabilities of eTreg, CD28-deficient Treg remain capable of both and migrate to tumor tissue. Here they accumulate at reduced frequency and in a partially diminished state of activation.

Global ablation of Treg in mouse models enhances immunological control of tumors in both prophylactic and therapeutic settings. We wanted to explore the effect specifically of preventing or reducing eTreg formation on tumor tolerance. Although meaningful tumor growth studies in CnB ΔTreg mice are not possible because of their short life span and severe inflammatory disease, we first examined the importance of CD28 costimulation for the ability of Treg to control antitumor immunity. When we implanted MCA-205 tumors into homozygous CD28 ΔTreg mice, we observed a significant deceleration of tumor growth compared with control animals, indicating enhanced antitumor immunity (Fig. 7A). Interestingly, the accumulation of CD28-deficient Treg in tumor tissue was less impaired (2–3-fold) in this setting than in CD28 ΔTreg^het mice (15–20-fold), where CD28-deficient and CD28-sufficient Treg potentially compete for resources (Fig. 7B, 7C). When we implanted the less aggressive colon carcinoma line MC38 into CD28 ΔTreg mice, tumor control was even more pronounced and two out of four animals rejected their tumors (Fig. 7D), suggesting that costimulation of Treg via CD28 is critical for tumor tolerance.

Although CD28 ΔTreg mice appear healthy, their latent lymphoproliferative disease may still confound tumor growth studies. To address this caveat, and also to enable comparative studies of animals with CD28- and CnB-deficient Treg, we devised two different strategies either to acutely delete CnB or CD28 in Treg, or to remove CnB- or CD28-sufficient Treg. First, we crossed CnB^fl/fl and CD28^fl/fl to Foxp3^CreERT2 mice (28) to create CnB ΔTreg^ind and CD28 ΔTreg^ind mice, in which Cre activity is induced and CnB or CD28 genes are deleted in Treg only upon tamoxifen treatment. When we implanted MC38 carcinomas into these animals and initiated tamoxifen treatment 4 d later, both strains exerted partial control over tumor growth, but this control was more pronounced in CnB ΔTreg^ind mice (Fig. 7E). However, this approach is likely limited by incomplete deletion of floxed genes (8). Therefore, we devised a second approach and created mixed bone marrow chimeras from either CnB ΔTreg or CD28 ΔTreg together with Foxp3^DTR bone marrow cells. In these animals, Foxp3^DTR Treg maintain immune tolerance, but can be effectively ablated through diphtheria toxin treatment, acutely creating functional CnB ΔTreg and CD28 ΔTreg mice. When we implanted MC38 colon carcinomas in these animals and initiated diphtheria toxin treatment 4 d later, Foxp3^DTR Treg were efficiently deleted (data not shown). As a result, both strains exerted partial control over tumor growth, but control was again more pronounced in mice in which Treg lacked CnB (Fig. 7F). Thus, in two different experimental settings, tumor tolerance was strongly impaired when Treg lacked expression of CnB or CD28. This effect was more pronounced upon deletion of CnB in Treg, correlating with their complete block in eTreg differentiation, whereas loss of CD28 still permits a reduced number of eTreg to enter and function in tumor tissue, explaining the more partial loss of tumor tolerance.

In this study, we have defined the respective roles of calcineurin-dependent NFAT activation and CD28 costimulation in the formation and maintenance of effector Treg endowed with the ability to migrate to and accumulate in nonlymphoid tissues, as well as in malignant tumors. We show that cognate Ag encounters in tdLNs activate resting cTreg and enable their clonal expansion and differentiation into eTreg that upregulate markers of elevated suppressive function and chemokine receptors enabling their migration to tumor tissue. In the absence of calcineurin activity, this differentiation process does not occur, and Treg fail to populate tumor and nonlymphoid tissues. Despite residual suppressive activity of CnB-deficient cTreg in vitro, this defect is associated on the one hand with superior immune control of aggressive tumors, but on the other hand with failed immune homeostasis, reflected by an early-onset, rampant lymphoproliferative inflammatory disease replicating the scurfy phenotype. In contrast, eTreg formation still occurs in absence of CD28, but at a diminished rate. Accordingly, a milder form of inflammatory disease occurs only later in life, as previously reported (14–16), accompanied by more variably reduced eTreg accumulation in nonlymphoid tissues. Nevertheless, immune control of aggressive tumors is strongly enhanced.

Our observations consolidate and extend a model in which, under homeostatic conditions, a fraction of Treg in SLOs is continuously exposed to self-antigen–derived TCR ligands, some of which are likely tissue-specific and trigger cTreg differentiation into eTreg capable of entering and persisting in nonlymphoid tissues (3–5, 13, 31, 32). In addition, some nonlymphoid tissues are seeded with specialized Treg populations soon after birth (56). Based on the complete absence of CnB-deficient Treg in all tested nonlymphoid tissues including the colonic lamina propria, which normally harbors thymically as well as peripherally induced Treg (57), we can conclude that both initial neonatal seeding as well as continued seeding during adult life depend on calcineurin activity subsequent to induction of Foxp3 expression. The requirement for CD28, on the other hand, is much more variable, and very pronounced for instance in skin, but less stringent in lung or colon.

In contrast to normal tissues, tumors rapidly develop de novo in the adult animal. The absence of CnB-deficient Treg at these sites and failed tumor tolerance demonstrate that calcineurin is a key effector of TCR signals that license Treg to populate immune effector sites and regulate ongoing immune responses.

Dendritic cells are known to regulate Treg numbers in an MHC II–dependent manner (58), suggesting a role for direct self-antigen–dependent interactions. However, we found that cTreg numbers in blood and lymphoid tissues and their in vitro–suppressive function are largely preserved in the absence of calcineurin, suggesting that either TCR signals or specifically NFAT activation are not important for maintenance of this subset. This agrees with findings that dendritic cells regulate Treg in SLOs indirectly by activating effector T cells that in turn provide IL-2 to sustain cTreg in a TCR-independent manner (10, 33, 59).

However, cTreg activity does not seem to suffice to maintain systemic immune homeostasis, and the severe inflammatory disease in mice lacking CnB in Treg further underscores the importance of the formation of eTreg with the capacity to traffic to nonlymphoid tissues. Prior studies have shown that deletion of the TCR in mature Treg leads to decreased expression of Foxp3 and of genes important for the suppressive function of eTreg (8). Our observations indicate that calcineurin, and by extension, NFAT proteins, are key drivers of CTLA-4 and ICOS expression. Although induction of Foxp3 expression during Treg development and its maintenance in mature Treg primarily depends on NF-κB proteins, especially c-Rel (60–62), NFAT binding to the CNS2 enhancer element of Foxp3 is important to sustain optimal Foxp3 expression in eTreg (45). NFAT proteins may therefore increase the expression of genes important for Treg-suppressive activity directly through binding on their promoters or regulatory elements, as shown for CTLA-4 and ICOS (44, 63), but also indirectly by optimizing expression of Foxp3. Both of these mechanisms, along with the reported cooperation of NFAT proteins with Foxp3 (64, 65) may thus contribute to the calcineurin-dependent, elevated suppressive function of eTreg triggered by TCR signaling.

In addition to genes closely associated with Treg function, calcineurin-driven NFAT activity also supports Treg proliferation and is essential for the expression of all receptors for nonlymphoid chemokines we have examined. Restricting calcineurin function will therefore not only prevent expansion and optimal suppressive function, but also trafficking of Treg to tumors and to nonlymphoid tissues.

Little is known about the role of TCR-dependent signaling pathways, and in particular of calcineurin or NFAT activity, in the direct regulation of chemokine receptor expression. Prior reports have implicated NFAT proteins in the regulation of CX3CR1 in human PBMCs (66), CCR2 in neurons (67), but also of CXCR5 in follicular Treg, a subset of eTreg (68). Our survey of published chromatin immunoprecipitation sequencing data on NFAT1 binding sites in CD8 T cells (44) revealed that these are frequently located in close proximity to the TSSs of chemokine receptors, especially of the Treg activation–dependent receptors CXCR3, CCR4, and CCR5 studied here. This suggests that in addition to other factors such as T-bet, which has been shown to induce CXCR3 expression in Treg (69), TCR- and calcineurin-dependent NFAT activity is not only critical for expression of these receptors, but also directly drives their expression.

T cells express three NFAT proteins, NFAT1, 2, and 4 (NFATc2, c1, and c3). Compound deletion of either NFAT1 and 2 or 1 and 4 does not abrogate the suppressive function of thymically derived Treg, indicating that expression of either NFAT 2 or 4 by itself suffices to sustain Treg activity. This has led to the conclusion that Treg are less dependent on NFAT than conventional T cells (22, 23), but the absolute requirement for NFAT expression has not been tested. In fact, deletion of all three T cell expression NFAT proteins was necessary to disable NFAT activity in CD8 T cells (44), and it is likely that the same is required to disable NFAT activity in Treg. Our findings demonstrate residual cTreg function, but a complete block in eTreg differentiation in absence of any calcineurin-dependent NFAT activity. Interestingly, Treg that lack CnB can still upregulate CD44 and at least partially downregulate CD62L expression, commonly used markers of eTreg differentiation (3). This shows that some features of T cell activation are preserved in the absence of CnB, whereas other critical aspects of eTreg function are disrupted.

Although calcineurin and NFAT activation were essential for eTreg formation, CD28 appeared to play only an optimizing role. CD28 is essential in Treg development (70), and only conditional deletion in mature Foxp3⁺ Treg has recently started to reveal the distinct role of CD28 costimulation in their survival and proliferation (14, 71). Extending these prior studies we found that eTreg lacking CD28 were less numerous in tumor-draining LNs and tumor tissue and less numerous or absent in nonlymphoid tissues. This could reflect reduced induction of eTreg differentiation in absence of CD28-mediated amplification of TCR-driven signaling pathways. Because Treg express TCRs with a range of peptide–MHC affinities, only those Treg with the highest TCR affinity may be activated in the absence of costimulation. This in turn would lead to a narrowing of the eTreg repertoire, which has been shown to increase the propensity for autoimmune disease (72). Alternatively, Treg may be activated at normal efficiency, but their proliferation or the survival of proliferating cells is reduced as a result of impaired induction of CD28-dependent survival factors (73) or metabolic adaptation (74). Impaired survival may be compounded by reduced expression of ICOS, which promotes eTreg survival in IL-2–poor environments, such as in tumor tissue (3). The fact that the frequency of CD28-deficient Treg was more severely decreased in tumors and in nonlymphoid tissues than in SLOs, and that CD28-deficient Treg were more compromised in tumors when they were in competition with CD28-sufficient Treg, argues for a reduction in fitness.

One of the central functions of CD28 costimulation in synergy with TCR signals is to facilitate robust NF-κB activation via the PDK1/PKCθ/CARMA1 pathway (55). Interestingly, the phenotype of mild lymphoproliferative disease and impaired tumor tolerance in CD28 ΔTreg mice resembles that in mice with Treg-specific conditional deletion of the NF-κB protein c-Rel (62), suggesting that impaired c-Rel activation may account in large part for the defects in CD28-deficient Treg. c-Rel is partially redundant with the NF-κB protein p65/RelA, and deletion of both proteins in Treg is required to produce the scurfy phenotype (62). RelA appears to be more effectively activated in Treg by members of the TNF receptor superfamily, such as GITR, than by the TCR (75), probably explaining why CD28-deficient Treg do not exhibit more severe defects, despite the virtually complete block in NF-κB activation we observed upon TCR activation in vitro.

Do CD28 and calcineurin signals synergize or promote eTreg function independently? PKCθ and Akt, which are both regulated by CD28 signaling, can prolong NFAT activation through inhibition of GSK3β, which phosphorylates NFAT and thereby accelerates its inactivation (76). However, we did not observe any defect in NFAT nuclear translocation in CD28-deficient Treg, suggesting that at least in Treg, CD28 was instead most important for the activity of calcineurin-independent signaling pathways, such as NF-κB activation via PKCθ. Conversely, lack of calcineurin did not attenuate CD28-regulated activities, such as ERK, mTORC1, and NF-κB signaling. In this regard one should also note that, although individual deletion of the NF-κB proteins c-Rel and RelA in Treg produced less severe disease, their compound deletion produced a scurfy phenotype similar to the one we observe in CnB ΔTreg mice. However, the entire Treg compartment appears to be destabilized in the animals, resulting in expression of effector T cell genes (62), in contrast to the selective block in cTreg to eTreg differentiation and the failure to accumulate in nonlymphoid tissues in the absence of CnB. Thus, calcineurin on the one hand and CD28 and NF-κB proteins on the other hand likely support eTreg formation and function through different mechanisms in Treg.

Understanding the roles of various TCR- and costimulation-dependent signaling pathways in regulating the formation and function of eTreg has the potential to inform strategies for their therapeutic manipulation to antagonize their tumor tolerance–supporting function without systemic disruption of immunological self-tolerance in cancer patients. Our findings demonstrate that an absolute block in eTreg formation in the absence of CnB results in immune control of highly aggressive tumors. This is reminiscent of the long-standing observation that calcineurin inhibitors prevent the formation of long-term immune tolerance to transplanted allogeneic solid organs (77). Although prior efforts to therapeutically target a toxin to Treg via CD25 have not been successful in cancer patients (78), likely at least in part because of insufficient selectivity of CD25 expression on Treg, other cell surface proteins, such as the chemokine receptors CCR4 and CCR8, have recently been described to be more selectively expressed on tumor-reactive eTreg (42, 43, 79). It may be possible to use these receptors for Treg-specific delivery of calcineurin inhibitors to Treg to curb tumor growth.

In contrast to calcineurin, CD28 costimulation seems to continually optimize the fitness of eTreg in nonlymphoid tissue, given their more dramatic decline at these sites compared with SLOs, and selectively antagonizing CD28 in Treg will likely directly affect already formed eTreg, including those in tumor tissue. Several studies have already shown how disruption in Treg of individual CD28-driven pathways, such as PI3K signaling (80), Akt-dependent Foxo1 inactivation (6), and, more recently, c-Rel (81), can promote tumor control. The capability to antagonize these pathways or CD28 itself selectively in Treg may dictate in part which target will have the greatest utility in cancer immunotherapy in the future.

We thank Jason McCarthy, Esteban Carrizosa, Mauro Di Pilato, Jasper Pruessmann, and Shariq Usmani and the Massachusetts General Hospital Department of Pathology Flow and Image Cytometry Research Core for helpful discussions and technical support.

The authors have no financial conflicts of interest.