Convergences and Divergences of Thymus- and Peripherally Derived Regulatory T Cells in Cancer

The expansion of regulatory T cells (Treg) is a common event characterizing the vast majority of human and experimental tumors and it is now well established that Treg represent a crucial hurdle for a successful immunotherapy. Treg are currently classified, according to their origin, into thymus-derived Treg (tTreg) or peripherally induced Treg (pTreg) cells. Controversy exists over the prevalent mechanism accounting for Treg expansion in tumors, since both tTreg proliferation and de novo pTreg differentiation may occur. Since tTreg and pTreg are believed as preferentially self-specific or broadly directed to non-self and tumor-specific antigens, respectively, the balance between tTreg and pTreg accumulation may impact on the repertoire of antigen specificities recognized by Treg in tumors. The prevalence of tTreg or pTreg may also affect the outcome of immunotherapies based on tumor-antigen vaccination or Treg depletion. The mechanisms dictating pTreg induction or tTreg expansion/stability are a matter of intense investigation and the most recent results depict a complex landscape. Indeed, selected Treg subsets may display peculiar characteristics in terms of stability, suppressive function, and cytokine production, depending on microenvironmental signals. These features may be differentially distributed between pTreg and tTreg and may significantly affect the possibility of manipulating Treg in cancer therapy. We propose here that innovative immunotherapeutic strategies may be directed at diverting unstable/uncommitted Treg, mostly enriched in the pTreg pool, into tumor-specific effectors, while preserving systemic immune tolerance ensured by self-specific tTreg.


INTRODUCTION
Tumor onset is a very complex process, in which cells of both innate and adaptive immune system play crucial roles in inhibiting or fostering tumor development. The awareness that the immune system could act as an extrinsic tumor suppressor or as a tumorsculpting player resulted in the cancer immunoediting theory, which described the interaction between tumor and host as consisting of three different phases: elimination, equilibrium, and escape (1). During the last phase of this process, transformed cells acquire the ability to subvert the control exerted by immune cells thus originating the clinically evident pathology. The escape is due to different mechanisms, including reduced immunogenicity (low expression level of MHC class I and loss of antigen expression), acquired resistance to the cytotoxic functions of immune cells, and accumulation in the tumor microenvironment of immunosuppressive mediators, like regulatory T cells (T reg ) (1). The first marker to be identified as distinguishing T reg from the other CD4 + T lymphocytes was CD25 (2) and depletion of CD25-positive cells unveiled anti-tumor immunity in experimental models (3). Few years later, the transcription factor Forkhead Box P3 (Foxp3) was indicated as the master regulator of T reg (4,5). In support of the crucial roles played by Foxp3 in T reg fate determination and immune homeostasis, Foxp3 mutations have been recognized as responsible for human Immune Dysfunction, Polyendocrinopathy, Enteropathy, X-linked (IPEX) syndrome (6,7), and for the phenotype of scurfy mutant mice (8), both characterized by fatal autoimmune lymphoproliferation linked to severe defects in T reg development/functions. However, very recent data have demonstrated that the complete development of the T reg lineage requires the concomitant, Foxp3-independent, establishment of a T reg -specific pattern of DNA hypomethylation (9).
According to recently proposed recommendations (10), T reg are classified into two principal subsets based on their developmental origin: thymus-derived T reg (tT reg ) develop in the thymus, while peripherally induced T reg (pT reg ) develop in vivo in the periphery from conventional T cells (T conv ), through a process called "conversion" (11). T reg can also be induced in vitro (and are called iT reg ) under TGF-β and/or retinoic acid exposure, but their complete commitment into fully differentiated T reg is still under debate (12). In physiological conditions, the pool of T reg , encompassing both tT reg and pT reg , which represents about the 5-10% of the circulating CD4 + T lymphocytes, assures peripheral selftolerance and prevents exacerbated immune responses (7,8). A huge amount of data now demonstrates that tumor onset and progression perturb T reg homeostasis and lead to increased T reg /T conv and T reg /CD8 ratios both in peripheral blood and in the tumor microenvironment (13). The accumulation of T reg at tumor sites may be due to the concomitant or the preferential occurrence of distinct events, such as the recruitment of T reg from periphery, the proliferation of pre-existing T reg in the tumor microenvironment, and the de novo conversion of tumor-infiltrating CD4 + lymphocytes (TIL) into pT reg (14,15). Despite controversy existing over the dominant suppression mechanism, and despite the incomplete understanding of the biological meaning of T reg accumulation in cancer, it is well accepted that T reg are crucial players in tumor biology and that the modulation of their function is an indispensable requisite for the development of successful anti-tumor immune-therapies.

MECHANISMS OF T reg SUPPRESSION IN TUMORS
It was recently demonstrated that T reg infiltrating different tissues have a specific gene signature (16), thus T reg may use peculiar suppression mechanisms in response to microenvironmental stimuli. This specialization may represent the basis for designing immune interventions targeting specific T reg functions in a given tissue while sparing systemic immune homeostasis. Even though a tumor T reg -specific gene signature has not been delineated yet, some mechanisms have been described to contribute to T reg suppression in tumors, which can be clustered in three main types: surface molecules, enzymatic activities, and cytokines (Figure 1).
Both human and mouse T reg constitutively express on their surface cytotoxic T lymphocyte-associated antigen 4 (CTLA-4), a coinhibitory member of the CD28/B7 family, endowed with strong immune-regulatory properties (17). The relevance of CTLA-4 in regulating T reg function was demonstrated in several settings, including autoimmune diseases and different tumor types (18). A comparative gene expression profile between T reg and T conv revealed that T reg specifically up-modulate lymphocyte activation gene 3 (LAG-3) (19), a homolog of CD4, that binds to MHC class II on antigen-presenting cells (APCs). LAG-3 is upregulated in tumor-infiltrating T reg and experiments with anti-LAG-3 mAb demonstrated that functional LAG-3 is required for maximal T reg suppressive function (20,21). T reg -DC interaction is also mediated by the transmembrane protein neuropilin-1 (Nrp-1), expressed on T reg membrane, which ensures the stability of T reg -DC interaction and allows T reg to efficaciously suppress DC (22). A study conducted on patients with early-stage cervical tumor showed that T reg infiltrating the tumor-draining lymph node of patients with metastasis have a higher expression of several immune-modulatory molecules, including Nrp-1 (23). The receptor activator of nuclear factor-kB ligand (RANKL), member of the tumor necrosis factor family, was found to be highly expressed on T reg isolated from tumor-bearing hosts, and substantial evidence indicates that RANKL expressed by T reg is involved in the onset of metastasis in both mammary (24) and prostate tumors (25).
Regulatory T cell suppression may be mediated by enzymatic activities, such as CD39/CD73 (26,27), granzyme B, and perforin (28). CD39 and CD73 are two ecto-enzymes that dephosphorylate extracellular ATP and generate pericellular adenosine, which in turn exerts a strong pro-tumorigenic role modulating the function of numerous tumor-infiltrating immune cells. CD73-deficient mice develop a stronger anti-tumor immune response compared to CD73-sufficient mice (29). T reg are also able to control the (1) Among the surface molecules expressed by T reg , CTLA-4, LAG-3, Nrp-1, and RANKL have a well-demonstrated role in promoting tumor progression, mainly modulating DC activation and function. In particular CTLA-4 and LAG-3, binding to CD80/CD86 (B7-1/2), and MHCII respectively, significantly impair DC capacity to activate T conv . In addition CTLA-4 promotes IDO expression and the production of the pro-apoptotic metabolite kynurenine. Nrp-1 instead stabilizes T reg -DC contact, allowing T reg to adequately suppress DC. Although the course of action of RANKL is not yet well defined, its expression is associated to tumor metastatization. (2) The two ecto-enzymes CD39 and CD73 generate from ATP pericellular adenosine, which is endowed with strong tolerogenic effects. Also cAMP, similarly to adenosine, interferes with T conv activation and survival. Granzyme and perforin induce the apoptosis of target cells by cytolysis. (3) T reg secrete several immune-modulatory cytokines, which could directly modulate T conv functions (TGF-β, IL-10, and IL-35), or indirectly promote the establishment of pro-tumorigenic microenvironment (VEGF).
proliferation and function of different immune cells via the Perforin pathway (30). In mouse models of melanoma, lymphoma, and acute myeloid leukemia it has been demonstrated that tumorinfiltrating T reg , but not naïve T reg , secrete high amounts of both perforin and granzyme B, which in turn induce NK and CD8 + T cell death (28).
Immunosuppressive cytokines, like TGF-β and IL-10, are critical players in T reg biology, being involved in both their differentiation and suppressive potential, especially in tumors. T reg -derived TGF-β was found relevant in suppression of anti-tumor T cell response in both mouse (31) and human (32,33) tumors. IL-10 is a well-known immunosuppressive mediator, and several pieces of evidence highlight the relevance of T reg -derived IL-10 in controlling inflammation at the mucosal interfaces such as gut and lung (34,35). Despite these data, little information is available Frontiers in Immunology | Immunological Tolerance about the roles of T reg -derived IL-10 in tumor microenvironments. We have recently demonstrated that tumor-associated T reg secrete high amounts of IL-10, which in turn impairs DC migration to the draining lymph nodes and the mounting of a specific anti-tumor immune response. This phenotype could be reverted by stimulating the receptor OX40 on the surface of intratumoral T reg . Indeed, OX40-triggered T reg showed reduced secretion of IL-10 as a consequence of the down-modulation of the interferon regulatory factor 1 (IRF1), a transcription factor active in the IL-10 promoter (36). Another cytokine recently described as critical for the full T reg suppressive function is IL-35, formed by Epstein-Barr-virus-induced gene 3 (Ebi3) and IL-12α (p35) (37). In two different mouse transplantable tumor models (melanoma and colon carcinoma), it was observed that T reg secrete abundant IL-35, thus promoting the differentiation of induced IL-35-secreting T reg (37). It is well known that tumor growth is associated with a consistent process of new angiogenesis in response to hypoxia. A circuit involving tumor hypoxia, T reg recruitment, and angiogenesis has been recently discovered (38). In the hypoxic tumor microenvironment, the chemokine axis CCL28-CCR10 plays a determinant role in the recruitment of T reg , which secrete huge amounts of vascular endothelial growth factor (VEGF), further stimulating the new angiogenesis process and the establishment of a tolerogenic microenvironment (38).

DOUBLE EFFECTS OF T reg ON PROGNOSIS
Since their discovery, T reg were considered one of the main obstacles for tumor clearance, according to their tolerogenic properties and their accumulation along tumor progression. In this view, several anti-tumor immune-therapies focus on T reg depletion or inhibition, in order to "contrasuppress" T reg inhibitory functions and to block the conversion of non-regulatory cells (non-T reg ) into regulatory cells (15). Reduced T conv /T reg ratio was observed in patients with pancreatic tumor (39, 40), breast cancer (39), ovarian cancer (41), Hodgkin lymphoma (42), and melanoma (43). Increased T reg frequency is generally associated to advanced tumor stage and poor prognosis, as recently demonstrated in a study on ovarian cancer (44). In the ascites of patients with advanced tumor, the percentage of T reg was increased compared to the ascites of patients with early-stage tumor. Same results were obtained with the mouse WF-3 transplantable ovarian tumor model, showing augmented percentages of T reg in both spleen and tumorassociated cells of mice with advanced tumors, compared to naïve or mice with early lesions. In addition, the treatment of tumorbearing mice with the T reg -depleting mAb PC61 (αCD25) reduced tumor growth and prolonged mice survival (44). Similarly, it has been demonstrated that T reg number inversely correlated with the therapy outcome in melanoma patients treated with nonmyeloablative chemotherapy, in combination or not with total body irradiation, followed by adoptive T cell transfer (45). Responder patients had a lower frequency of T reg in peripheral blood compared to non-responder patients (45). A study conducted on patients affected by invasive ductal carcinoma showed a positive correlation among T reg , Th17, and tumor aggressiveness. These data imply that T reg and Th17 cells may concomitantly expand during tumor progression, with T reg mainly suppressing protective effector T cell proliferation while sparing Th17 proangiogenic activities, fostering cooperatively tumor progression, and the metastatic process (46).
Nevertheless, recent data, in particular tumor systems, point out that T reg may exert a protective role for the host (13, 47,48). The connection between tumor and inflammation is a well-assessed process (49), but now it is clearly emerging that the type of tumorassociated inflammation imprints the behavior of T reg , becoming detrimental or beneficial for the host. Type-1 inflammation, characterized by high concentration of IFN-γ and IL-12 and fully active cytotoxic cells, represents efficient anti-tumor immunity (49). In this setting, the inhibitory properties of T reg may promote tumor escape and aggressiveness (47). On the contrary, immune responses dominated by cytokines like TNF-α, IL-1β, IL-6, IL-23, and IL-17 act as pro-tumoral mediators (47). In this environment T reg may suppress a pro-tumoral inflammatory status, thus playing a protective role for the host (47).
These unexpected anti-tumoral T reg properties were observed in patients with colorectal cancer (CRC) (50)(51)(52). In these patients, with different tumor stages, a better prognosis and an increased overall survival were associated with higher infiltration of FOXP3 + T cells compared to patients with a poor tumor outcome. These data suggested the hypothesis that FOXP3 + T cells could be considered as an independent prognostic factor for CRC. Following a strong activation, both conventional CD4 + (53) and CD8 + (54) lymphocytes up-regulate Foxp3 expression in colonic tissue. These observations indicated that the CRC prognosis positively correlated with non-regulatory FOXP3 + cells rather than to T reg . However in vitro suppression assays demonstrated that FOXP3 + cells, isolated from CRC tissues, were endowed with suppressive functions, confirming their nature as regulatory cells (55). In a recent study conducted on 65 patients with different stages of CRC, FOXP3 expression was systematically evaluated in both tumorinfiltrating lymphocytes and neoplastic cells, and was correlated to tumor progression and clinical-pathological features (56). From this study the notion emerged that high FOXP3 expression in tumor cells correlated with poor tumor outcome, compared to tumors poorly expressing FOXP3; on the contrary, no correlation was observed between CRC prognosis and FOXP3 expression by T cells (56).
A protective role of T reg was also found in head and neck squamous cell carcinoma (HNSCC) (57). Univariate and multivariate analysis demonstrated that the locoregional control of the tumor was positively associated with CD4 + FOXP3 + regulatory cell infiltration (57). However, also for this type of neoplasia, there are some discordant data regarding the role of T reg in tumor progression. Indeed, another study showed that T reg frequency and suppressive function were higher in the peripheral blood of tumor-bearing patients than in healthy volunteers (58).
The discrepancies observed in these studies may be due to the number of patients included, different strategies of analysis and non-homogeneity of tumor samples (stage, metastasis, etiology). Certainly, to properly define the role of T reg in tumor outcome, the new studies should take into account the tumor stage and the related inflammatory features, depending on the anatomical localization. In general, those tumors arising from chronic inflammation, almost at their initial stage, can benefit from the suppressive properties of T reg . In fact, during the inflammatory www.frontiersin.org process, T reg highly accumulate in the site of inflammation such to prevent exacerbated immune responses and tissue damage, which are the prelude to neoplastic transformation. On the contrary, in the presence of an established tumor, T reg may reduce anti-tumor immunity thus favoring tumor escape. A more specific definition of T reg contribution in tumor development and progression is desirable for the design of new and more effective immunotherapies, allowing the discrimination among tumors that will benefit or not from T reg depletion/inhibition.

EVIDENCE FOR pT reg OR tT reg ACCUMULATION IN TUMORS DISTINGUISHING FEATURES OF pT reg AND tT reg
Many efforts have been recently addressed to the identification of phenotypic, molecular, and functional features distinguishing tT reg and pT reg , besides their site of origin (11). Some markers have been proposed to distinguish pT reg and tT reg , even though with some limitations: the initial enthusiasm for the suggestion of Helios as able to identify tT reg (59) has been soon moldered by the observation of Helios expression in pT reg (60); the recent finding of the Nrp-1 as a marker of tT reg (61,62) has application limited to murine cells, being not expressed on human T reg (63). Several attempts have been conducted to identify genetic (64)(65)(66) and/or epigenetic signatures distinguishing pT reg and tT reg . The T reg -specific demethylated region (TSDR) is involved in the stable commitment of the T reg lineage, and controversy still remains on whether iT reg or pT reg can efficiently demethylate this region and become fully committed T reg (66)(67)(68)(69). Despite this growing amount of information, distinguishing the relative contribution of pT reg and tT reg to immune suppression in physiological and pathological conditions remains hard. However, some pieces of evidence have accumulated in the last years that speak in favor of tT reg or pT reg prevalence or concomitance in tumors.

EVIDENCE FOR tT reg ACCUMULATION IN CANCER
One of the first attempts to distinguish between pT reg conversion and tT reg expansion in cancer was pursued by Bui and colleagues who adoptively transferred CD4 + CD25 + cells, mixed at 1:10 ratio with CD25-depleted Thy1.1-congenic splenocytes, into immunodeficient mice bearing a progressive sarcoma (70). The analysis performed 10 days after tumor injection showed that the vast majority (around 80%) of tumor-infiltrating CD4 + CD25 + cells derived from expansion/recruitment of the transferred T reg , rather than from conversion of non-T reg . This and other reports, appeared in the "pre-Foxp3" era, were biased by the idea that CD25 was the most stringent T reg marker and that CD25-depleted cells represented a suitable precursor population to efficiently detect de novo generation of pT reg . However, subsequent studies have demonstrated that the CD25 + subset of Foxp3 − T conv is enriched in pT reg precursors (69,71), thus the extent of pT reg differentiation from CD25-depleted cells represents probably an underestimation of the actual contribution of pT reg induction in the tumor context. Other authors have shown that tT reg may dominate pT reg not only quantitatively but also qualitatively, in terms of suppressive function: indeed, IDO + plasmacytoid dendritic cells, derived from mouse tumor-draining lymph nodes, were capable to induce Foxp3 + pT reg at very high levels but were unable to activate the suppressive function of these cells to an extent comparable to tT reg (72). Many studies have clearly shown T reg proliferation (in terms of de novo DNA synthesis and/or cell division) in tumor-bearing mice or cancer patients, thus indirectly supporting the idea that expansion of pre-existing tT reg might prevail over pT reg differentiation in building the tumor-associated T reg pool. For instance, T reg have been shown to incorporate high levels of BrdU in tumor-draining lymph nodes and at cancer sites in several experimental models (73,74). A study conducted in patients with multiple myeloma showed that the TREC content in naive cells was significantly lower in T reg (identified as CD4 + CD25 high cells) than CD4 + CD25 − or CD25 low cells, suggesting that the T reg pool mainly derived from peripheral expansion rather than recent thymic emigration (75). However, the observation of high T reg proliferation at tumor sites cannot be considered as an unequivocal proof of tT reg prevalence over pT reg , since both subsets could be endowed with the same proliferative potential in vivo. Indeed, several pieces of evidence indicate that conversion and proliferation may represent uncoupled and independent events (see pTreg Development and tTreg Expansion as Independent Processes).

EVIDENCE FOR pT reg INDUCTION IN CANCER
Some studies support the idea that pT reg conversion actually occurs in tumor-bearing hosts at higher efficiency than in physiological conditions, even if controversy still exists on whether pT reg may prevail numerically over tT reg at the tumor site. We have in the past demonstrated that thymectomized and CD25-depleted mice, subsequently transplanted with carcinoma cells, showed a significantly higher T reg recovery in tumor-draining than in contralateral lymph nodes, suggesting that in tumor-bearing mice the T reg pool might be replenished mostly by newly derived pT reg than by proliferation of residual T reg . To prove this possibility, CD25-depleted CD4-purified T cells were transferred into immunocompetent, Thy1.1-congenic, CT26 tumor-bearing mice. In this setting, we could show that the transferred cells acquired CD25 and Foxp3 at significantly higher levels in draining lymph nodes, compared to contralateral lymph nodes of tumor-bearing mice, or to the lymph nodes of tumor-free mice (76). This result clearly showed that tumor progression actively promoted the conversion of nonregulatory precursors into pT reg . Some tumor-derived molecular signals were found to be involved in tumor-associated conversion. For instance, in different mouse models, tumor cells have been shown to induce in vitro T reg conversion through TGF-β, and TGFβ neutralization abrogated T reg accumulation at the tumor site (77). Human leukemic cells converted in vitro non-T reg into T reg through the tumor cell-restricted IDO activity, and IDO blockade prevented pT reg induction in vivo in a leukemia mouse model (78). A confirmation of extensive pT reg infiltration in murine tumors has been recently obtained thanks to the recent discovery of Nrp-1 as a tT reg -restricted marker (61,62). The analysis of Nrp-1 expression has indeed revealed that Nrp-1-negative bona fide pT reg cells were significantly enriched at tumor site compared to spleen, ranging around 40-90% of total tumor-infiltrating T reg depending on the tumor type (61). These Nrp-1-negative cells also presented a gene signature (Helios low , SWAP-70 low , and Dapl1 high ) compatible with the pT reg identity (61). Unfortunately, human T reg do not express Nrp-1 (63), thus this marker cannot be used to estimate the relative contribution of tT reg or pT reg in human cancers.

DEVELOPMENTAL AND FUNCTIONAL RELATIONS BETWEEN pT reg AND tT reg IN CANCER pT reg DEVELOPMENT AND tT reg EXPANSION AS INDEPENDENT PROCESSES
Many attempts have been made to understand whether tT reg accumulation and pT reg development are mutually exclusive or rather cooperative in establishing immune suppression. The evidence that tT reg may "educate" T conv to convert into T reg through the secretion of cytokines, such as TGF-β and IL-10 (79), may support the latter possibility. This event would generate a cascade of suppressive function transmitted from T reg to bystander cells, establishing a loop of immunosuppression, reminiscent of a phenomenon called as "infectious tolerance" (80). Zhou and coworkers have addressed this issue in the tumor setting, and have demonstrated that tumor-antigen-specific pT reg could indeed arise from T reg -depleted cells (adoptively transferred in mice carrying the cognate antigen-expressing tumor), but that the extent of pT reg induction was not affected by the concomitant presence of tT reg , either exogenous (adoptively co-transferred) or endogenous (preexisting in the host) (81,82). This result indicated that tT reg and pT reg accumulate in tumors in a reciprocally independent fashion and that "infectious tolerance" may play minor roles in shaping the tumor-associated T reg pool.
A comprehensive scenario of T reg accumulation in tumors should take into account, beside de novo conversion, the active proliferation of not only tT reg but also pT reg . Proliferation plays opposite roles in the differentiation of T conv into pT reg versus the expansion of already differentiated pT reg . Regarding the former aspect, we have demonstrated that T conv proliferation was not required for their conversion into pT reg , since CD25 + Foxp3 + cells could develop in tumor-bearing mice from CD25-depleted cells treated with an anti-proliferative agent (76). A seminal study by Kretschmer and colleagues showed that T conv proliferation was not only dispensable but also detrimental to conversion: indeed, low levels of T cell proliferation, in conditions of suboptimal antigen presentation, lack of costimulation, and IL-2 paucity, favored TGF-β-mediated pT reg induction, thus suggesting that an inverse relationship might exist between T conv proliferation and conversion into pT reg (83). However, once developed, pT reg promptly proliferated in response to experimental antigens (83) and, more importantly, in response to tumor antigens (81,82). Experiments performed in CNS1-mutated mice, which are genetically unable to generate pT reg , have shown that pT reg and tT reg may occupy distinct "niches": indeed, the efficiency of pT reg differentiation from T conv was higher when those T conv were co-transferred, in lymphopenic recipients, with a CNS1-deficient (non-containing pT reg ) compared to a CNS1-sufficient (containing pT reg ) counterpart, suggesting that not only the tT reg pool, but also the pT reg niche, may be controlled by autonomous homeostatic mechanisms (84).

DIVISION OF LABOR BETWEEN tT reg AND pT reg IN CANCER
Both tT reg and pT reg have been generally recognized as immune suppressive cells in a variety of in vivo and in vitro experimental settings (12). However, whether the two subsets are endowed with peculiar activities remains unclear and is a matter of intense investigation.
Gene expression profiling revealed that the pT reg and tT reg signatures were mostly overlapping but also presented some differentially expressed genes, encoding for proteins involved in T reg suppressive function, suggesting that pT reg may preferentially exploit different molecules and related mechanisms to exert suppression (64)(65)(66). The Nrp-1 itself is not only a marker discriminating murine tT reg from pT reg , but also plays a role in T reg suppression: since this molecule prolongs T reg interactions with immature dendritic cells, tT reg may benefit from this pathway in preferentially modulating dendritic cell and cognate T cell activation (22). Many data suggest that pT reg may be specialized suppressors of immune responses at interfaces with external environments, such as airways, gut, and maternal-fetal interface (64,(84)(85)(86)(87). Of note, a peculiar T reg suppressive molecule, IL-10, plays crucial roles at environmental interfaces, therefore pT reg may perform their specialized activity through IL-10 secretion (34,88). IL-10 is critically involved in the establishment of tumor-associated immune suppression, and we have clearly demonstrated IL-10 production by around 40% of tumor-infiltrating T reg in murine transplanted tumors (36). It would be interesting to understand whether the fraction of IL-10-producing T reg is enriched in pT reg , rather than tT reg , in tumors. One study has directly addressed the issue of induced T reg functional specialization in tumors, by generating in vitro tumor-specific iT reg and co-culturing these cells, or tT reg as control, with NK cells: these authors found that iT reg and tT reg equally suppressed IL-2-induced NK activation, but only iT reg were endowed with the surprising ability not to suppress, but to enhance, NK cytotoxicity induced by tumor target cell contact (89). This observation may speak in favor of differential roles played by tT reg and pT reg in cancer, with the former more involved in preventing target cell-independent, and possibly selfdirected, unwanted immune responses, and the latter concurrently enhancing tumor-specific immunity.
This division of labor may result in the progressive shaping of the immune response toward an effective anti-tumor immunity with minimal side effects. Such dichotomy is also reminiscent of the double role played by T reg in different cancers, according to the hypothesis that high T reg frequency is associated to poor or good prognosis in non-inflammatory or inflammatory cancer onset, respectively (13, 47). In the former case, i.e., noninflammatory cancers in which protective type-1 responses are suppressed by high T reg infiltration, T reg may mainly recognize tumor-associated self-antigens, and mostly include tT reg ; conversely, in the case of inflammatory cancers, related to chronic low-dose type-17 cytokines, which are typical of mucosal tissues, high numbers of pT reg may suppress pro-tumoral inflammation through IL-10, relevantly produced by pT reg at those sites. We are tempting to speculate that tT reg may dominate in suppressing anti-tumor type-1 responses, while pT reg may prevail in shaping pro-tumor type-2 and type-17 inflammatory responses. Notably, the prototypical example of an inflammation-related tumor in which T reg accumulation associates to good prognosis is CRC (50), developing in the gastrointestinal mucosa, in which immune tolerance is under the control of pT reg (84).

ANTIGEN SPECIFICITY OF tT reg VERSUS pT reg IN CANCER
Antigen recognition may play a crucial role in dictating whether tT reg or pT reg will prevail in the tumor context. Controversy still exists on the antigen specificity of these two populations. On the one side, tT reg are generally believed to recognize self-antigens encountered during thymic selection (90). On the other side, pT reg , deriving from T conv , are thought to display the same TCR repertoire of T conv and thus to mainly recognize foreign antigens. Indeed, only a small overlap exists between TCR repertoires of pT reg and tT reg (66), and pT reg are believed to recognize nonself-antigens such as commensal microbiota, allergens, and fetal alloantigens (84,87).
Tumor cells can express a variety of antigens that can be broadly classified into: (i) self-antigens physiologically expressed as in the tissue of origin, (ii) self-antigens aberrantly expressed, in terms of expression level, developmental stage, or histotype (called tumorassociated antigens or TAAs), and (iii) neo-antigens uniquely expressed by tumor cells, mostly derived from oncogenic mutations (named tumor-specific antigens or TSAs). Based on the above considerations, self-antigens and TAA should be recognized by tT reg , while TSA would induce and activate pT reg . However, a complex picture arises from studies analyzing the TCR specificity of tumor-associated T reg .

T reg can recognize TAA and TSA in tumors
In different tumor models, TCR-transgenic T reg have been shown to promptly proliferate in response to the cognate antigen specifically expressed by tumor cells, suggesting that T reg can undergo tumor-antigen-driven activation and expansion (74,81,82,91). Antigen presentation in the tumor context may favor T reg expansion: in a mouse model of spontaneous prostate cancer, an efficient T reg induction/expansion occurred only when TCR-transgenic, antigen-specific CD4 T cells encountered the cognate antigen expressed in the context of prostate cancer cells, rather than nontransformed cells or viral vector-infected cells (91). In this model, TAA-specific T reg were recognized as pT reg induced in vivo in a TGF-β-independent fashion.
This evidence of TAA-responding T reg has been confirmed in human tumors. CD4 clones derived from cancer patients resulted to be regulatory cells and to recognize peptides derived from TAAs, such as LAGE1 (92), ARTC1 (93), TRAG-3, NY-ESO-1 (94-96), Melan-A (97), survivin, TRP1, and gp100 (94) in melanoma patients, and WT-1 in leukemia patients (98). By using MHCII/peptide tetramer technology, other authors failed to detect T reg specific for NY-ESO-1 in the peripheral blood of ovarian cancer patients (99). Bonertz et al. developed an in vitro system to screen the suppressive function of T reg in response to single peptides and, with this approach, could detect T reg specific for several TAA in the peripheral blood of colon carcinoma patients but not in healthy donors; notably, T reg depletion in vitro unveiled TAA-specific T conv responses (100).
The possibility that tumor-associated T reg may recognize not only TAA but also TSA is demonstrated by the observation that T reg specific for exogenous viral antigens, acting as TSA, may arise in virus-related cancers. T reg clones specific for human papilloma virus (HPV), and suppressing the cognate antigen-directed T cell response, have been obtained from tumor-draining lymph nodes and tumor biopsies of cervical cancer patients (101). T reg clones specific for antigens of the Epstein-Barr virus (EBV), associated to several hematological and solid malignancies, can be generated from the peripheral blood of healthy donors (102).

T reg can recognize self-antigens in tumors
Several data in mouse models confirm that T reg responding to selfantigens can play a role in suppressing the anti-tumor responses. Immunization with serologically defined auto-antigens was found to enhance tumor growth in different mouse models, and this event was dependent on the expansion of T reg responding to those self-antigens (103). This study confirmed that self-antigensspecific T reg could suppress anti-tumor immunity, but did not explore the T reg response to self-antigens expressed by tumor cells themselves during tumor progression. This issue has been instead addressed in an experimental model in which a foreign antigen, artificially expressed in transgenic mice under tissue-specific promoter, was seen (peripherally and/or thymically) by the immune system as a self-antigen and elicited the generation of a pool of memory T reg specific for that antigen (74). If those mice were injected with the cognate antigen-bearing tumor, the memory T reg pool specific for that self-antigen was hugely expanded in tumors and tumor-draining lymph nodes, confirming that self-specific T reg can respond to self-antigens expressed by tumor cells (74). A seminal paper has recently reported the immunoscope analysis of T reg infiltrating spontaneous prostate tumors in a mouse transgenic model, and described the clonal enrichment of a single T reg specificity that was directed not to a unique TSA but to a self-antigen expressed also by normal prostate cells (104). The development of T reg specific for peripheral tissue-restricted selfantigens occurred in the thymus under the control of the Aire molecule, which allows the expression of peripheral antigens in thymic epithelial cells (104). These findings clearly demonstrate that T reg can recognize self-antigens in cancer and suggest that maintaining self-antigen expression may help transformed cells to overcome the immune surveillance through self-specific T reg expansion.

Repertoire analysis as an estimation of pT reg /tT reg balance
The direct comparison between the repertoires of tumorassociated T reg and T conv may help understanding the processes underlying T reg enrichment in cancer. Some authors have reported that the analysis of TCR diversity (performed with the immunoscope technology) showed that T reg infiltrating murine transplanted tumors displayed a biased TCR repertoire toward "public" CDR3 sequences (i.e., shared by different mice), suggesting T reg intra-tumor clonal expansion driven by the recognition of dominant antigens (105). Also tumor-infiltrating activated T conv showed a biased TCR repertoire, but it was distinct from the T reg spectrum, and the minimal overlap between the two subsets was mainly confined to "private" specificities (105). By using a similar approach, others have reported distinct and not overlapping TCR repertoires of T reg and T conv infiltrating prostate tumors in a genetically engineered mouse model of spontaneous prostate carcinogenesis (104). Lack of overlap between T conv and T reg repertoires was also found in tumors and tumor-draining lymph nodes in a mouse model of chemical carcinogenesis (106). Overall, the lack of overlap between T reg and T conv has been interpreted in many cases as the result of negligible pT reg conversion at the tumor site; however, pT reg and tT reg may share more specificities than expected, since tT reg -associated antigens may preferentially drive T conv fate decision toward the conversion into pT reg rather than toward the conventional activation; moreover, already established pT reg may then undergo intra-tumor clonal expansion together with tT reg in response to the same antigens. Therefore, the overall overlap between T conv and T reg specificities may not accurately estimate the extent of pT reg induction in tumors. Indeed, in one study performed in advanced melanoma patients, TAA-specific TCRs, expressed by tumor T reg clones, could be detected in both T reg and T conv populations, demonstrating that TAA-specific T reg may be comprised of pT reg derived from the conversion of T conv (95).
Indirect data support the notion that TAA-specific T reg may contain pT reg . TAA-specific T reg clones, obtained from patients with advanced melanoma, suppressed in vitro the cognate antigenspecific T cell response, but produced high levels of Th1 and/or Th2 cytokines (95), and showed low FOXP3 expression and TSDR demethylation, indicating that these cells may represent an incompletely uncommitted T reg population, which more likely belongs to the pT reg than to the tT reg pool (95).
A recent study has directly evaluated the consequences of pT reg and tT reg antigen specificities in tumor-bearing hosts. Indeed, Schreiber et al. have shown that, if purified polyclonal tT reg and T conv , differentially labeled with green or red fluorescence, were cotransferred in CD4-null mice, the tT reg progeny exceeded the newly T conv -derived pT reg population in tumor-draining lymph nodes as well as in the spleen; conversely, when transgenic, tumor-antigenspecific, tT reg and T conv were injected, tT reg and pT reg reached comparable frequencies in tumor-draining lymph nodes (107). These results suggest that tT reg and pT reg are mostly specific for self-or tumor-antigens respectively, and that the balance between pT reg and tT reg may be fine-tuned by the relative prevalence of TSAs versus self-antigens expressed by tumor cells.

HETEROGENEITY AND PLASTICITY OF tT reg AND pT reg T reg HETEROGENEITY IN CANCER: RELATIONS WITH THE pT reg /tT reg DICHOTOMY
During the latest years, it has become increasingly clear that T reg , meant as Foxp3-positive cells, are not a homogeneous lineage, but rather represent a mixture of subpopulations. Indeed, beside the tT reg /pT reg distinction based on their developmental origin, diverse T reg subsets can be identified endowed with peculiar features in terms of suppression, proliferation, and stability, even though not properly classifiable as developmentally distinct lineages (Figure 2). Tumor microenvironmental signals may differentially affect these subsets, thus shaping T reg heterogeneity to the advantage of tumor progression.

T reg stability and epigenetic commitment in cancer
Foxp3 inherent stability, rather than Foxp3 expression in a given moment and tissue, is intimately linked to an actual commitment to the T reg lineage and therefore to the maintenance of immune suppression. Pioneer studies have demonstrated that Foxp3 stability is strictly related to an epigenetic imprinting of FIGURE 2 | Functional dynamics of tT reg and pT reg in cancer. This picture summarizes development, heterogeneity, plasticity, antigen specificity, and function of pT reg and tT reg in cancer. Activated T reg , which are epigenetically committed and mostly self-and TAA-specific, can transiently lose Foxp3 without methylating TSDR thus becoming latent T reg ; in some conditions, they can acquire T-bet expression thus becoming specialized suppressors, detrimental to the anti-tumor type-1 response. Activated T conv , mostly foreign (TSA) antigen-specific, can promiscuously express Foxp3 without demethylating TSDR. However, a fraction (CD25 + , or CD39 + ) of activated T conv can convert into pT reg , progressively moving from an uncommitted to a committed stage. Through IL-10, committed pT reg can suppress pro-tumoral inflammatory and type-17 responses, thus exerting beneficial roles for the host in some cancer types. In some contexts, uncommitted pT reg (and possibly activated T conv ) can move back to exT reg stage, acquiring the ability to produce inflammatory cytokines. Therefore, in some tumors such as colon cancer, Th17-like T reg may foster type-17 inflammation thus supporting tumor growth; in other tumor contexts, Th1-like T reg can favor type-1 responses that rather block tumor growth. Green, cells specific for self-antigens and TAA; light blue, cells specific for foreign antigens including TSA. Yellow dash, demethylated TSDR; blue dash, methylated TSDR. Red "F" in yellow circles, stable Foxp3; yellow "F" in empty circles, unstable Foxp3. Dashed arrows, unclear events. Orange rounded arrows, proliferation in the tT reg or the pT reg homeostatic niche. Light green frames, conditions in which T reg are beneficial to the host; light orange frames, conditions in which T reg are detrimental to the host.
CpG demethylation in the TSDR region of the Foxp3 locus (67,86,108). TSDR demethylation was then recognized as the mechanism featuring the distinction between committed (demethylated) and uncommitted (methylated) T reg , irrespective of Foxp3 expression (9). Controversy exists on whether pT reg show complete or partial TSDR demethylation and can then be considered as committed T reg . Many studies show that iT reg have a partially or completely methylated TSDR (9, 67-69), while pT reg have been described as TSDR-demethylated (68), TSDR-methylated (66), or as a mixed population of stable and unstable cells, characterized by CD25 high or low expression respectively (69).
Few data exist on the extent of TSDR demethylation in tumorassociated T reg . The frequency of TSDR-demethylated cells is higher in peripheral and intratumoral leukocytes of lung, colon, prostate, or breast cancer patients, in relation to a higher T reg www.frontiersin.org frequency as determined by flow cytometry or immunohistochemistry (109). Of note, the extent of TSDR demethylation in CRC patients was only slightly higher than in healthy volunteers, in contrast to the significantly increased T reg frequency in these samples shown by previous studies (110,111). This discrepancy may be explained with the peculiar nature of this inflammation-related and mucosal tissue-located cancer, in which the inflammatory response may specifically involve pT reg , possibly containing more uncommitted (TSDR-methylated) cells than tT reg .
The evaluation of TSDR demethylation has been used as a reliable analytical tool for the estimation of committed T reg in some tumor conditions and therapies. An increased frequency of epigenetically committed (TSDR-demethylated) T reg has been determined in tumor-infiltrating cells of ovarian, colorectal, and bronchial cancers compared to non-tumoral tissue counterparts (112). TSDR demethylation was decreased in the peripheral blood of metastatic renal carcinoma patients receiving tumor vaccination (113), and increased in patients treated with dendritic cell vaccination and cytokine therapy (114).

T reg functional dynamics in cancer
The idea of Foxp3 as the master transcription factor of T reg lineage has been challenged by the observation that some T reg features are Foxp3-independent, and that Foxp3 plays T reg -unrelated functions (115). This is especially true for human FOXP3-positive cells, since human activated T conv can transiently express this transcription factor that acts as an intrinsic T cell regulator (116). The concomitant analysis of CD45RA and FOXP3 in human T reg in both physiological and pathological contexts allowed delineating a classification into three subsets: CD45RA + FOXP3 low resting T reg (rT reg ), CD45RA − FOXP3 low non-T reg , and CD45RA − FOXP3 high (CD45RO + ) activated T reg (aT reg ), endowed with different potentials of proliferation, suppression, stability, and plasticity (117). Whether each subset mainly contains tT reg or pT reg is unclear. While rT reg were recognized as CD31 + recent thymic emigrants, thus belonging to the tT reg pool, aT reg can be considered as a mixed population of activated tT reg (derived from rT reg ) and pT reg (derived from non-T reg or T conv ). The CD45RA − FOXP3 low non-T reg subset may represent a mixture of activated T conv (promiscuously and unstably expressing FOXP3), latent T reg (transiently downregulating FOXP3), and recently converted pT reg (117).
The three human T reg subsets can be differentially expanded in distinct pathologies. In conditions characterized by exacerbated immune responses, such as autoimmune diseases, rT reg and non-T reg are expanded; conversely, in diseases associated to immune unresponsiveness, such as sarcoidosis, the aT reg subset is instead enriched in the peripheral blood (117). The tumor context, which conceivably belongs to the latter category, should be characterized by aT reg expansion. In line with this hypothesis, CD45RO + FOXP3 high aT reg were found significantly expanded in the peripheral blood, and much more at the tumor site, in patients with malignant melanoma (118). Also the non-T reg and the rT reg fractions were increased, but only in the peripheral blood, in cancer patients compared to healthy controls, and both subsets positively correlated with tumor progression (118). The non-T reg pool produced some IFN-γ and its frequency returned to normal levels after tumor removal, thus probably representing aberrantly activated T conv , or T reg with attenuated FOXP3 activity (118). A much deeper knowledge on T reg dynamics in cancer is needed to better understand the role played by each specialized component in suppressing anti-tumor type-1, or pro-tumor inflammatory, responses.

T reg subsets specified by functional/developmental markers
Several surface or intracellular markers have been suggested to identify T reg subsets endowed with peculiar abilities other than suppressive functions. A portion of human T reg with an effector/memory-like phenotype (26,119,120) expresses CD39, which has been proposed as a target to enrich human suppressive T reg (119). CD39 was found overrepresented in peripheral and tumor-infiltrating T reg from HNSCC, and was further increased in patients with advanced-stage disease or after radiochemotherapy (120,121). CD39 is also expressed by a T conv subset, which produces lower levels of pro-inflammatory cytokines than the bulk T conv population, and is more prone to convert, at least in vitro, into T reg (120). Both CD39 + T reg and CD39 + T conv were enriched in peripheral blood, and further increased at tumor site, in HNSCC patients, and a positive correlation existed between frequencies of these two populations (120). Therefore, these data suggest that tumor-associated CD39 + T conv may represent a reservoir of CD39 + T reg precursors. As a consequence, it could be suggested that CD39 + T reg may include both tT reg and pT reg , and that both T reg subsets can exploit the CD39-mediated suppressive mechanisms of ATP degradation and adenosine generation.
Not only the functional arms of suppression, but also the activation requirements may differ in tT reg and pT reg : for instance, TNFR2 expression is needed to activate tT reg but not iT reg suppressive ability in experimental colitis (122). Of note, TNFR2-positive T reg have been found enriched in murine tumors, in association with a higher suppressive ability, ex vivo, in the standard suppression assay (123). In a mouse model of metastatic melanoma, TNF-α caused enhanced tumor progression through the TNFR2mediated T reg expansion at the site of metastasis (124). These data suggest that TNFR2 expression may label tumor-infiltrating T reg of thymic origin, and that TNF-α at the tumors site may preferentially expand and activate tT reg . Supporting the idea that tT reg may represent more stable cells, TNFR2 was found to be involved in the maintenance of Foxp3 stability in mouse models of inflammation (125). Also in human peripheral blood, CD25 and TNFR2 co-expression identifies cells highly expressing FOXP3, showing an effector/memory phenotype and strong suppression, ex vivo (126). The TNF-α/TNFR2 pathway may amplify T reg activation also through the induction of a NF-kB-driven transcriptional program enriched for other members of TNF superfamily, such as 4-1BB, FAS, and OX40 (127).
The early idea that Helios could differentiate tT reg from pT reg (59,128) prompted the use of this marker in delineating tT reg accumulation in cancer. The vast majority of tumor-infiltrating T reg were found to express Helios in a mouse model of glioblastoma (129), in glioblastoma multiforme patients (129), and renal cell carcinoma patients (130). However, the value of Helios as univocal marker of tT reg has been questioned by other studies that showed Helios also expressed in pT reg (60,131), and in association to T reg suppression (128,131) and commitment. Indeed, Helios − FOXP3 + cells freshly isolated from healthy donors or autoimmune disease patients showed decreased TSDR demethylation compared to Helios + FOXP3 + (132,133), and also displayed a higher plasticity in terms of cytokine production (133). In a murine transplanted tumor model, tumor-infiltrating T reg were enriched in Helios + cells, representing the subset with the highest proliferative potential (as shown by Ki67 staining) (131). In summary, the well-recognized enrichment of Helios + T reg in several human and mouse tumors may be attributed, rather than to preferential attraction and expansion of tT reg , to the tumor-driven local activation and/or commitment of both tT reg and pT reg .

SPECIALIZATION AND PLASTICITY OF tT reg /pT reg IN CANCER
It is now well established that T reg (or better, their specific subsets) adapt their molecular programs to optimize their in vivo suppressive function in distinct inflammatory milieus, which may be alternatively dominated by Th1, Th2, Th17, or T FH responses. Strikingly, these T reg specialized programs are orchestrated by the same transcription factors that drive the polarization of the targeted T-helper subset: therefore, T-bet, IRF4, Stat3, and Bcl6 expression are respectively and selectively required for the T reg specialized suppression of Th1 (134, 135), Th2 (136), Th17 (137), and T FH (138,139) responses. Indeed, by acquiring master T-helper genes, T reg may gain the expression of chemokine receptors driving the recruitment of specialized T reg into inflamed tissues. However, in some contexts, T reg (or, again, some T reg subsets) can express not only T-helper-related transcription factors and migratory molecules, but also cytokines such as IFN-γ or IL-17, thus turning from specialized suppressors into so-called Th1-like or Th17-like T reg that may rather contribute to inflammation (140). Some data, mostly from mouse experimental models, suggest that such T reg plasticity is not a lineage reprograming of committed T reg , which appear instead quite stable; rather, Th1-like or Th17-like T reg may derive from uncommitted cells expanded in inflammatory conditions (69,141). However, other studies have shown that in both mouse and human pathologies T reg can produce relevant amounts of type-1 and type-17 cytokines even though preserving Foxp3 expression (142-146).

Th17-like T reg in cancer
Regulatory T cells may shift to a Th17-like phenotype in inflamed microenvironments dominated by type-17 cytokines, thus favoring, rather than suppressing, pro-tumoral mechanisms of chronic inflammation. According to this idea, human T reg have been found to spontaneously secrete IL-17 in the intestine of patients carrying inflammatory bowel disease (145,147) and colon carcinoma (147). In epithelial ovarian cancer, a malignancy associated to chronic inflammation, T conv were found to secrete high levels of IL-17 (and other cytokines) when cultured ex vivo with IL-2; under similar conditions, tumor-infiltrating T reg were prone to FOXP3 downregulation, attenuation of suppressive function, and prompt IL-17 production (148). In human lung adenocarcinoma, FOXP3 message amounts correlated with Th17-related transcripts enriched at the tumor site, where IL-17 antagonized the development of the anti-tumor, T-bet-dependent, Th1 response (149). Myeloid antigen-presenting cells and cytokines such as IL-2, TGF-β, IL-1, IL-23, and IL-6 may initiate T reg polarization into Th17-like cells in these tumor contexts (147)(148)(149). In a mouse model of hereditary colon polyposis, as well as in human colon cancer, the Th17-like T reg co-expressed the Th17-related transcription factor ROR-γt, and fostered tumor progression, also through the promotion of mast cell local expansion (150,151). This study clearly demonstrated that microenvironmental signals could direct T reg plasticity toward pro-inflammatory and pro-tumoral activities.
One group has demonstrated that Th17-like T reg can also arise in experimental tumors as an outcome of vaccination strategies (152). In this study, vaccination with antigen plus TLR-9 ligand induced T reg reprograming into polyfunctional T-helper-like cells, producing a wide array of cytokines including IL-2, TNFα, and IL-17, and expressing cell-surface CD40L, thus providing efficient T cell help for tumor-antigen cross-presentation and development of anti-tumor response (152). The IDO immunosuppressive enzymatic activity was responsible for preventing this anti-tumor T reg polarization, which was instead enhanced using an IDO blocker (152).
Little data exist on the precursors of Th17-like T reg in cancer. In the peripheral blood of healthy subjects, the CD45RA − FOXP3 low non-T reg subset was found enriched in Th17-related transcripts and in cells actively secreting IL-17, even at higher levels than naïve or memory T conv , a data suggesting that this population contains Th17 or Th17-like precursors (117). It would be interesting to understand whether the Th17 potential resides, within the non-T reg gate, in activated T conv , in latent T reg , and/or in recently induced pT reg , possibly co-expressing FOXP3 and RORγt and thus pre-committed to the Th17 lineage.

Th1-like T reg in cancer
Pioneer studies from Koch and colleagues demonstrated that, following exposure to IFN-γ in Th1-dominated microenvironments, a subset of T reg can up-regulate the Th1-related transcription factor T-bet, which drives T reg expansion, migration (CXCR3mediated), and function specifically during type-1 inflammation (134). Further experiments have shed light on the developmental requirements and the alternative fates of murine T-bet + T reg : following IFN-γ stimulation, T reg could gain T-bet expression but failed to fully polarize into IFN-γ-producing Th1-like T reg , due to an impaired T reg susceptibility to IL-12. Indeed, IL-12 receptor β2, which is inducible in T conv in an IFN-γ-and T-bet-dependent fashion, is epigenetically inaccessible in T reg (135). Only longlasting IFN-γ preconditioning could unlock IL-12 responsiveness, thus allowing the complete T reg polarization into Th1-like cells (135). Presumably, in contexts characterized by chronic IFN-γ and IL-12 abundance, such as autoimmune, inflammatory, and viral diseases, T reg will be oriented to a full reprograming into Th1-like cells. Supporting this idea, IFN-γ-producing T reg have been reported in mouse models of graft-versus-host disease (153), viral (154) or parasite (142) infection, in human multiple sclerosis (144), and diabetes (146,155). In one of these systems, IFN-γproducing T reg were recognized to be specific for a foreign (viral) antigen (154). Whether such Th1-like T reg can be yet considered as classical regulatory cells is still debated. One study has shown that in vitro polarized Th1-like T reg were less suppressive than conventional T reg in the standard in vitro suppression assay, and that suppression could be partially rescued with concomitant IL-10 www.frontiersin.org and IFN-γ neutralization (144). Another study has proven that IFN-γ-producing human iT reg were equally functional as natural T reg in suppressing both proliferation and cytokine production of responder T cells (156). In a mouse model of graft-versus-host disease, IFN-γ produced by stable (TSDR-demethylated) T reg was shown to be even required for T reg protective effect (153), suggesting that IFN-γ-releasing T reg can display in vivo unexpected functions depending on the context.
Conversely, it could be envisaged that, in the tumor context, the low levels of IFN-γ derived from T conv , NK, and CD8 cells, and the paucity of IL-12 production by tumor-associated APCs, may concur to induce a pool of T reg expressing T-bet but not secreting IFN-γ, thus specialized in suppressing antitumor type-1 immunity. In line with this possibility, TAA-specific T conv , but not TAA-specific T reg , produced IFN-γ in patients with epithelial ovarian cancer (99). In both healthy subjects (117) and malignant melanoma patients (118), IFN-γ-producing cells were enriched within the circulating CD45RA − FOXP3 low (CD45RO + ) non-T reg subset, mostly including activated T conv and/or uncommitted T reg . Conversely, in ovarian cancer, tumor-infiltrating T reg were enriched in CXCR3 + cells, highly expressing T-bet but poorly producing IFN-γ, and strongly suppressing Th1 response ex vivo (157). Tumor-associated CXCR3 + T reg were mostly Heliospositive, and T-bet + T reg could be generated in vitro by culturing CD45RA + CCR7 + rT reg (mostly containing tT reg ) under Th1-polarizing conditions (157), suggesting their derivation from committed tT reg . This finding was in accordance to Koch et al. who showed that T-bet + T reg derived from T-bet − T reg , rather than from activated T conv (134). These data support the idea that tT reg , rather than pT reg , may contain the precursors for Th1-specialized suppressors, thus playing critical roles in suppressing protective responses in tumors whose high T reg frequency correlates with poor prognosis (13).
Some therapeutic interventions can force tumor-associated T reg toward a fully differentiated Th1-like phenotype. For instance, circulating T reg from melanoma patients showed significantly higher IFN-γ secretion following a protocol of tumor peptide vaccination plus IL-2 and cyclophosphamide, in line with enhanced serum IL-12 (158). On the whole, these data suggest that, especially in the human system, the transition from T-bet + T reg , specialized Th1 suppressors, into T-bet + IFN-γ + T reg , Th1-like plastic cells, may not only depend on the availability and the responsiveness to exogenous stimuli, but may differentially occur in distinct T reg precursors: on the one side, tT reg , enriched in committed and self-specific cells, may be forced to arrest to the specialization (T-bet + ) endpoint; on the other side, pT reg , containing less committed and foreign antigens-specific cells, may be more prone to the complete reprograming into pro-inflammatory (T-bet + IFN-γ + ) cells. Future studies will elucidate the mechanisms by which different growing tumors may favor the expansion of protumoral specialized Th1 suppressors or the induction of Th1-like plastic T reg .

IMPLICATIONS FOR CANCER IMMUNOTHERAPY
The initial enthusiasm on the use of therapeutic cancer vaccines has been soon disappointed by the observation of a low response rate in many trials (159). After the discovery of T reg as potent immune suppressive cells hampering the establishment of antitumor immunity, it soon became clear that anti-tumor vaccination might fail to elicit an effective immune response and to achieve successful tumor eradication, because of the immune suppressive barrier created by T reg . In addition, since T reg may recognize TAA and TSA at higher frequency than T conv , tumor-antigen-based vaccines may expand/induce T reg rather than effector cells, thus inhibiting rather than boosting the anti-tumor response. Indeed, Zhou et al. first demonstrated that TCR-transgenic CD4 T cells specific for a TAA, adoptively transferred into mice bearing TAA-expressing tumor cells, proliferated extensively after administration of a therapeutic tumor vaccine (in the form of a recombinant vaccinia virus encoding the antigen), but tumor-antigen-experienced cells were mostly regulatory cells, ex vivo suppressive, and anergic to subsequent stimulation (81).
In cancer patients receiving tumor-antigen vaccination, the expansion of antigen-specific T reg has been documented. Circulating NY-ESO-1-specific T reg spontaneously develop in late-stage melanoma patients and are expanded following immunization with NY-ESO-1 protein supplemented with adjuvants (96). Therapeutic vaccination with an HPV synthetic long peptide vaccine, administered to patients with HPV-positive cervical carcinoma, induced both CD8 and CD4 T cell immunity, but also enhanced the HPV-specific T reg pool (160). The pool of vaccine-specific T reg may derive not only from the expansion of pre-existing tumor-antigen-specific clones, but also from de novo generation of vaccine-specific pT reg . This is suggested by results obtained vaccinating melanoma patients with an HLA-DP4-restricted MAGE-A3 peptide: in this setting, a subset of vaccine-specific T reg becomes detectable only after vaccination (161). Vaccine-elicited T reg showed some degree of heterogeneity: out of five CD25 + regulatory clones isolated from vaccinated patients, four expressed high FOXP3 mRNA levels, produced TGF-β, and showed demethylated TSDR; one clone expressed less FOXP3, had methylated TSDR and produced some Th2 cytokines (161). These data suggest that antigen-specific T reg , induced in the periphery following antigen exposure and thus recognizable as pT reg , can contain both committed and uncommitted cells.
The concomitant and detrimental T reg expansion in antitumor vaccination can be avoided by using CD8 T cell-targeted approaches. A melanoma vaccination protocol based on an MHCI-restricted Melan-A peptide significantly decreased the frequency of Melan-A-specific T reg , in association with an improved and more diverse Th1 response (97).
Some attempts have also been made to combine active immunotherapy with T reg depletion or functional blockade. Several studies showed that depletion of CD25 + cells in vivo in cancer patients could enhance the tumor-specific T cell responses induced by cancer vaccines (15). However, CD25-directed strategies may fail to achieve sustained results, since activated effector cells may be concomitantly eliminated and pT reg may replenish the T reg pool after depletion (15). Interestingly, a recent study has demonstrated that different T reg -depleting agents, either CD25-targeted (IL-2/diphtheria toxin fusion protein, or anti-CD25 antibody) or not (low-dose cyclophosphamide), failed to consistently eliminate more than 50% of committed T reg , as identified by TSDR demethylation (162).

Frontiers in Immunology | Immunological Tolerance
Therefore, alternative strategies are needed to counteract the "hard core" of immune suppression that is represented by epigenetically committed T reg . We have proposed in the past that T reg functional inactivation, rather than depletion, may represent a successful strategy to prevent massive pT reg induction and concomitantly block T reg suppression (15). This idea may be corroborated by the observation that markers associated to T reg suppressive functions, and therapeutically targetable, may show enriched or restricted expression in epigenetically committed T reg . For instance, GITR stimulation has been shown to attenuate T reg suppression and favor the rejection of experimental tumors (163). A recent study has demonstrated that GITR engagement in vivo led to the downregulation of Foxp3 expression in intratumoral T reg (164). Of note, GITR + T reg were found enriched in Helios + cells, representing highly committed T reg (131), thus GITR targeting may preferentially block the strongest suppressors among the T reg pool. A similar possibility could be envisaged for therapeutic strategies aimed at TNF-α/TNFR2 blockade, since this axis may be mainly involved in the activation of more committed and stable cells (122)(123)(124)(125)(126). Committed T reg may also be targeted by virtue of their high proliferative potential: indeed, high proliferation rates, in terms of Ki67 positivity, were detected in healthy subjects within the aT reg subset, enriched in stable and committed T reg (117), and also in murine tumor-infiltrating Helios + T reg (131). Therefore, treatments based on the depletion of proliferating cells, such as low-dose cyclophosphamide, may efficiently target committed T reg .
An innovative way to improve immunotherapy would be to reprogram tumor-associated T reg into fully armed effector cells, which would then become "exT reg ." Different from other vaccinebased approaches, T reg reprograming is expected to trigger antitumor response very rapidly, since T reg are already located at the tumor site and already tumor-antigen-experienced, thus not requiring a de novo T cell priming. Therefore, exT reg may function in an "innate-like" manner, promptly providing co-stimulatory and pro-inflammatory signals when adequately modulated, before a novel adaptive anti-tumor response develops (140). An example of this approach has been proposed by Sharma et al. who demonstrated that reprograming of mature pre-existing tumorassociated T reg into CD40L-expressing helper effector cells was needed to achieve tumor regression in a model of immunotherapy combining antigen vaccination, TLR-9 stimulation, and IDO blockade (152).
The above-discussed data overall indicate that tT reg and pT reg may not be equally susceptible to functional reprograming, but this dichotomy may turn into a benefit for the efficacy and safety of the evoked response. Indeed, on the one hand, tT reg , predominantly self-specific, highly committed, and hard to be reprogrammed into T-helper-like cells, would be preserved, thus ensuring immune tolerance to self-antigens and maintaining systemic immune homeostasis. On the other hand, pT reg , mainly representing tumorspecific and uncommitted cells, may be more easily converted into exT reg , thus mounting an immediate helper and/or effector response in a mostly tumor-antigen-specific fashion.
Reprograming into exT reg may be achieved by immunotherapies aimed at subverting the immune suppression mechanisms established by innate cells in tumor microenvironments. For instance, in the above-reported model of tumor vaccination, CD40L upregulation by T reg following TLR-9 stimulation was strictly dependent on host-derived MyD88 and IL-6 signals (152). In melanoma patients, tumor peptide antigen vaccination combined with low-dose cyclophosphamide and low-dose IL-2 evoked Th1-like T reg accumulation, in line with a less tolerogenic microenvironment and with enhanced IL-12 availability (158). Of note, in this system, depletion of proliferating (conceivably committed and thymus-derived) T reg by means of cyclophosphamide allowed the functional reshuffling of innate cells that in turn unveiled the emergence of Th1-like exT reg .
However, it is arguable that microenvironmental rearrangements would better accomplish full T reg reprograming with the concomitant direct modulation of T reg activities, aimed especially at enhancing T reg susceptibility to external signals. For instance, expression of IL-12 receptor, which is epigenetically regulated in T reg (135), could be artificially boosted by pharmacological approaches. Also, targeting with monoclonal antibodies some receptors expressed on T reg surface and correlated with T reg stability (such as TNFR2 and GITR) could result in enhancing T reg propensity to reprograming. In line with this idea, treatment of murine melanomas with a GITR agonistic antibody resulted in the accumulation of exT reg at the tumor site (164). Suppressor of cytokine signaling (SOCS) 1 and 2, which maintain Foxp3 stability and prevent T reg polarization into effector cells (165,166), may be pharmacologically inhibited to unlock T reg responsiveness to pro-inflammatory microenvironmental cytokines.

CONCLUSION
Even though discrimination between pT reg and tT reg by simple surface phenotyping is not yet possible many pieces of evidence indicate that both subsets contribute to the T reg pool conditioning the tumor microenvironment. Nevertheless, the development/expansion of pT reg and tT reg are independent processes, possibly resulting from disparate antigens and signals, and their activities seem characterized by very peculiar features in terms of specificity, stability, and specialization. On the one side, tT reg may expand at tumor site in response to self-antigens expressed by tumor cells, mostly include committed (TSDR-demethylated) Helios-and TNFR2-expressing cells, and contain the precursors of specialized T-bet + Th1-suppressing cells, thus representing not only the guardians of systemic immune homeostasis but also the "hard core" of tumor immune escape. On the other side, pT reg may mostly develop following local encounter with TAA or TSA antigens, possibly represent a mixed population of committed (TSDRdemethylated) and uncommitted (TSDR-methylated) cells, and are more prone to be reprogramed into Th1-like or Th17-like effector cells. We envisage that future successful immunotherapies may not only target committed T reg but also favor "recycling" uncommitted T reg into prompt anti-tumor effectors.