Human OX40 tunes the function of regulatory T cells in tumor and nontumor areas of hepatitis C virus–infected liver tissue


  • Potential conflict of interest: Prof. Antonelli received grants from Siemens and Vin.

  • This work was supported by the following grants obtained by V.B.: Associazione Italiana per la Ricerca sul Cancro (AIRC; progetto “Investigator Grant” [IG]-2010/13 no. 10756); European Union grants (IMECS no. 201169, FP7-Health-2007-A, and SPHYNX no. 261365, FP7-Health-2010); Ministero della Sanità (Ricerca finalizzata [RFPS-2006-3-337923 and RFPS-2007-1-636647] and Istituto Superiore di Sanità [Progetto AIDS-2008]); Ministero dell'Istruzione, dell'Università e della Ricerca (MIUR; Programmi di ricerca di interesse nazionale [PRIN]-2008/10 no. 7245/1; [PRIN]-2011/13 no. 2010LC747T-004; Ateneo Sapienza [2009-C26A09PELN, 2010-C26A1029ZS, 2011-C26A11BYWP, and 2012-C26A12JL55]; and Fondo per gli investimenti di ricerca di base [FIRB]-2011/13 no. RBAP10TPXK); Fondazione Cariplo (progetti no. 13535 and 3603 2010/12); FISM (Fondazione Italiana Sclerosi Multipla onlus) grant no. 2011/R/4; Fondazione Italiana per la Ricerca sull'Artrite (FIRA 2010); and Istituto Italiano di Tecnologia (IIT; A2 project 2013). This work was also supported by grants obtained by S.P. from Associazione Italiana Ricerca sul Cancro (MFAG 8726) and from Ministero dell'Istruzione, dell'Università e della Ricerca (FIRB-Futuro in ricerca RBFR12I3UB_002).

  • See Editorial on Page 1461


Regulatory T cells (Tregs) can be considered as a mixed population of distinct subsets, endowed with a diverse extent and quality of adaptation to microenvironmental signals. Here, we uncovered an opposite distribution of Treg expansion, phenotype, and plasticity in different microenvironments in the same organ (liver) derived from patients with chronic hepatitis C: On the one side, cirrhotic and tumor fragments were moderately and highly infiltrated by Tregs, respectively, expressing OX40 and a T-bethighIFN-γ “T-helper (Th)1-suppressing” phenotype; on the other side, noncirrhotic liver specimens contained low frequencies of Tregs that expressed low levels of OX40 and highly produced interferon-gamma (IFN-γ; T-bet+IFN-γ+), thus becoming “Th1-like” cells. OX40-expressing and Th1-suppressing Tregs were enriched in the Helios-positive subset, carrying highly demethylated Treg cell-specific demethylated region that configures committed Tregs stably expressing forkhead box protein 3. OX40 ligand, mostly expressed by M2-like monocytes and macrophages, boosted OX40+ Treg proliferation and antagonized the differentiation of Th1-like Tregs. However, this signal is counteracted in noncirrhotic liver tissue (showing various levels of inflammation) by high availability of interleukin-12 and IFN-γ, ultimately leading to complete, full Th1-like Treg differentiation. Conclusion: Our data demonstrate that Tregs can finely adapt, or even subvert, their classical inhibitory machinery in distinct microenvironments within the same organ. (Hepatology 2014;60:1494–1507)








alcoholic fatty liver disease


carboxyfluorescein succinimidyl ester


chronic hepatitis C


dendritic cells


flow cytometry


forkhead box protein 3


hepatocellular carcinoma


hepatitis C virus


healthy donor










mean fluorescence intensity


mononuclear cells






nontumoral liver


peripheral blood


programmed death-1






conventional T cell


T helper


tumor necrosis factor alpha


regulatory T cell


Treg cell-specific demethylated region



Regulatory T cells (Tregs) represent crucial gatekeepers for the maintenance of immune homeostasis and the protection from uncontrolled immune responses.[1] The Treg lineage is determined by the transcription factor, forkhead box protein 3 (FOXP3), and by epigenetic, FOXP3-independent events.[2] In particular, Treg stability is associated with a demethylated form of the Treg cell-specific demethylated region (TSDR) in the Foxp3 locus.[3]

Among the several molecules suggested to regulate Treg expansion and/or suppressive function, programmed death-1 (PD-1) and OX40 have recently received particular attention. PD-1, considered as an inhibitory receptor delivering negative signals to conventional T cells (Tconvs), sustains development, maintenance, and suppressive function of peripherally induced Tregs upon engagement with its own ligand (PD-L1).[4] On the other side, PD-1 tempers expansion and function of already established Tregs.[5, 6] Contrary to PD-1, OX40 is considered as a costimulatory molecule conveying prosurvival signals to Tconvs.[7] OX40 is constitutively expressed by murine Tregs and inhibits Treg suppressive function,[8-10] a property that has been exploited to induce antitumor response in mouse models[8, 11, 12] and even in cancer patients.[13] However, in some contexts, OX40 signal can foster Treg fitness and proliferation.[14-17] OX40 is expressed by human Tregs or Tconvs not constitutively, but only upon appropriate activating signals7: Whether this inducible form of OX40 delivers prosurvival or inhibitory signals in human Tregs is not completely clear.

Under precise microenvironment-driven signals and in pathological processes dominated by a certain T-helper (Th) subset (Th1, Th2, Th17, and TFH), Tregs (or Treg subsets) can express the corresponding T-helper-associated transcription factor(s), thus acquiring specialized suppressive functions selectively directed against that T-helper subset.[18] Unstable Tregs, functionally and/or molecularly “deprogrammed” into cytokine-producing (i.e., “Th1-like” or “Th17-like”) Tregs, likely contribute to, rather than suppress, inflammatory responses.[19]

Whether Treg heterogeneity and plasticity are modulated in diverse pathologic liver microenvironments, and whether expression or engagement of costimulatory and inhibitory receptors is involved in Treg adaptation to microenvironmental cues, is mostly unknown. Here, we show that, within a certain organ (i.e., the human liver from CHC patients [pts]), inflammatory, cirrhosis, or tumor compartments were characterized by the expansion of distinct Treg subsets, expressing different levels of OX40, Helios (a transcription factor previously associated with Treg molecular and functional stability[20]) and TSDR methylation, and displaying divergent specialization or reprogramming in response to cytokines and surface signals provided by the different microenvironments.

Patients and Methods

Patients, Samples, and Processing

29 CHC patients with or without hepatocellular carcinoma (HCC) were studied (Supporting Table 1): 23 with underlying cirrhosis (c), 6 without cirrhosis (noncirrhotic, nc). Peripheral blood mononuclear cells (PBMCs) and liver specimens were obtained from patients undergoing surgery or liver transplantation at Istituto Nazionale dei Tumori “Regina Elena” or “Sapienza” Università di Roma - Policlinico Umberto I. Human studies have been performed in accordance to the ethical guidelines of the 1975 Declaration of Helsinki and were approved by each Institutional Ethical Committee. Informed consent was obtained from all patients. Methods for HCV-RNA viral load assessment, PBMC isolation and liver fragment processing are detailed in the Supporting Information.

Flow Cytometry

A complete list of antibodies can be found in the Supporting Information. Cells were pretreated with human Fc-Receptor Binding Inhibitor (eBioscience) and incubated 20 minutes at 4°C with Abs for surface antigens, except anti-OX40L and IL-12R-b2 PE (30 minutes at RT). Intracellular staining for FOXP3 was performed using the anti-FOXP3 mAb and FOXP3/Transcription Factor Staining Buffer Set according to manufacturer's instructions (eBioscience). Before IFN-γ and T-bet staining, cells were stimulated 4 hours with Cell Stimulation Cocktail (plus protein transport inhibitors, eBioscience). For mono/MF cytokine production, mononuclear cells were stimulated 18 hours with LPS (400 ng/ml, Difco) alone or plus IFN-γ (50 ng/ml, eBioscience), in the presence of Protein Transport Inhibitor Cocktail (eBioscience), and intracellular staining for TNF-α and IL-10 was performed in gated CD14+ cells using Cytofix/Cytoperm and Perm/Wash buffers according to manufacturer's instructions (BD Bioscience). Data were acquired on LSR Fortessa (Becton Dickinson) and analyzed with FlowJo software (Tree Star Inc, version 8.8.7).

In Vitro Assays of Treg Proliferation, Th1-Like Polarization or Suppression

Tregs were enriched using the CD4+CD25+ Regulatory T Cell Isolation Kit (Miltenyi Biotec). CFSE labeling was performed by incubation for 15 minutes at 37°C with 10 μM CFSE. In some experiments, Tregs were preactivated for 18 hours in U-bottomed 96-well plate with equal numbers of T cell-stimulating beads (Treg Suppression Inspector, Miltenyi Biotec) plus IL-2 (100 IU/ml, Roche) and TNF-α (50 ng/ml, RnDSystems). OX40 was stimulated with rOX40L (20 ng/ml, containing a polyHistidine tag, RnDSystems) cross-linked with a secondary tag-specific mAb (10 mg/ml, AD1.1.10, mouse IgG1, RnDSystems). Details of T-cell culture and cytokine stimulation, and of Th1-like Treg polarization ex vivo, can be found in the Supporting Information.

TSDR Demethylation Analysis

Helioshigh and Helioslow subsets in act-Treg or non-Treg gates were sorted using a FACSAria (Becton Dickinson), and real-time PCR using methylation- and demethylation-specific primers was performed on bisulfite-treated genomic DNA, as detailed in the Supporting Information.


Immunohistochemical analyses were performed on formalin-fixed paraffin embedded HCC tissue samples (n = 8) obtained from the archives of the Human Pathology Unit of the Department of Health Science, University of Palermo. Antibodies and methods are described in the Supporting Information.

Statistical Analysis

Statistical analysis was performed using Prism software (version 4, GraphPad). Unpaired Student t test, 2-tailed, was used to analyze in vitro data, while Mann-Whitney test, 2-tailed, or Wilcoxon matched pairs test, 2-tailed, was applied to compare groups of ex vivo samples. Correlations have been calculated using the nonparametric Spearman's correlation test, 2-tailed. Every in vitro assay was performed in triplicates or quadruplicates when possible. In all graphs, bars show means ± SEM. In all tests, P < 0.05 was considered statistically significant.


Tregs Are Differentially Distributed in Distinct Liver Microenvironments in Line With Variable OX40 Expression

First, we quantified the percentage of FOXP3+CD127low Tregs,[21] within the CD4+ T-cell pool, in mononuclear cells (MNCs) isolated from nontumor liver (NT-LIV) or hepatocellular carcinoma (HCC; tumor [TUM]) fragments, derived from the same CHC livers, carrying underlying cirrhosis (c) or not (nc; Supporting Table 1), as well as from peripheral blood (PB) of the same pts or healthy donors (HDs), as a control. Confirming previous studies,[22, 23] Tregs were significantly expanded at the TUM site compared to the other districts (Fig. 1). Interestingly, Tregs were significantly enriched in NT-LIV and in PB of c, compared to nc, pts (Fig. 1). Such cirrhosis-related Treg expansion was peculiar to CHC, because an unaltered or even decreased Treg frequency was detected in NT-LIVc specimens obtained from pts with hepatitis B or alcoholic fatty liver disease (AFLD) pts (data not shown).

Figure 1.

Tregs are significantly enriched in tumor and nontumor cirrhotic tissues. Treg frequency was estimated by FCM as the percentage of FOXP3+CD127low cells among CD4 T cells, in PB of HD, or PB, NT-LIV or TUM fragments obtained from CHC pts, c, or nc, listed in Supporting Table 1. Representative FCM data (left) and data overview (right) are shown. *P < 0.05; **P < 0.01; ***P < 0.005; by Mann-Whitney's test, two-tailed.

Importantly, the frequency of OX40+ Tregs was significantly higher (compared to PB) in NT-LIVc and (at higher levels) TUM, but not in NT-LIVnc, irrespective of PD-1 expression (Fig. 2A). OX40 was also expressed in a low fraction of Tconvs (CD4+FOXP3) from TUM and NT-LIVc, albeit at a greatly lower degree than in Tregs from the same districts (Fig. 2A). In addition, OX40+ Tregs also expressed OX40 at a higher level (in terms of mean fluorescence intensity [MFI]) than the Tconv counterpart (Supporting Fig. 1). Instead, the frequency of PD-1+OX40 cells was similar between Tconvs and Tregs in each district; in both Tconvs and Tregs, the PD-1+OX40 subset was significantly enriched in the liver, compared to PB, but irrespective of whether infiltrating NT-LIVc, NT-LIVnc, or TUM specimens (Fig. 2A). Notably, total Treg frequency was directly correlated with the percentage of OX40+ Tregs in TUM and tended to be inversely associated with the percentage of PD-1+OX40 Tregs (Fig. 2B), suggesting OX40 as a receptor associated to, or even functionally involved in, Treg expansion in cirrhosis and cancer, and PD-1 as a possible negative regulator of Treg expansion, as previously demonstrated in chronically hepatitis C virus (HCV)-infected settings.[6]

Figure 2.

OX40 is highly expressed by Tregs in tumor and nontumor cirrhotic tissues. (A) Frequencies of PD-1OX40+, PD-1+OX40+, and PD-1+OX40 cells were evaluated by FCM in CD4+FOXP3 Tconvs or CD4+FOXP3+CD127low Tregs, in different samples (PB, NT-LIV, and TUM) obtained from CHC pts. Data overview (left) or representative FCM data (right) are shown. *P < 0.05; **P < 0.01; ***P < 0.005; by Mann-Whitney's test, two-tailed. (B) Spearman's correlation (r) between frequency of Tregs and frequency of OX40+ (PD-1+/–) or PD-1+OX40 Tregs in different liver specimens from CHC pts. *P < 0.05; ns, not significant.

Th1-Suppressing and Th1-Like Tregs Show Phenotypic, Functional, and Spatial Dichotomy in Liver

Many data demonstrate that, in contexts characterized by a Th1-oriented inflammation, Tregs become specialized Th1-suppressing cells (gaining T-bet expression) or even become Th1-like cells (acquiring the competence for interferon-gamma [IFN-γ] production), thus inhibiting or contributing to inflammation, respectively.[24] Within both Tconvs and Tregs, two subpopulations could be clearly detected in the different liver districts (upon short stimulation with a phorbol myristate acetate/ionomycin-containing reagent): (1) T-bet+IFN-γ+ cells, representing cells fully differentiated into Th1 Tconvs or Th1-like Tregs, and (2) T-bethighIFN-γ cells, representing cells undergoing an incomplete Th1 differentiation, that, in the case of Tregs, have been identified as the specialized Th1-suppressing cells[24, 25] (Fig. 3A). However, the T-bet+IFN-γ+ Treg or Tconv (Th1-like or Th1, respectively) subsets were maximally expanded in NT-LIVnc and less represented in NT-LIVc and TUM (Fig. 3B, upper plots). Conversely, the T-bethighIFN-γ subpopulation of Tregs (Th1-suppressing), but not of Tconvs, was expanded in NT-LIVc and (to a significant extent) in TUM (Fig. 3B, lower plots). Th1-like (T-bet+IFN-γ+), but not Th1-suppressing (T-bethighIFN-γ), Treg frequencies positively correlated with the frequency of Th1 Tconvs in PB, NT-LIVc, and TUM (Fig. 3C), indicating that a Th1-oriented microenvironment may favor the polarization of Th1-like Tregs. Importantly, OX40 was differentially expressed in the two subsets, being the OX40+ Treg percentage significantly higher in the Th1-suppressing T-bethighIFN-γ subset, compared to T-bet+IFN-γ+ Th1-like counterpart, in each sample (Fig. 3D), suggesting that OX40 may be preferentially associated with an activated and specialized suppressive function of Tregs in vivo.

Figure 3.

Th1-like and -suppressing Tregs accumulate in noncirrhotic liver or in cirrhotic/tumor liver, respectively. (A) Representative intracellular staining of IFN-γ versus T-bet ex vivo in gated Tregs and Tconvs from PB or NT-LIVnc of a CHC pt. (B) Frequency of T-bet+IFN-γ+ (Th1-like) or T-bethighIFN-γ (Th1-suppressing) in gated Tregs (left plots), and analogous subsets in gated Tconvs (right plots), from different specimens. *P < 0.05 by Mann-Whitney's test, two-tailed. (C) Spearman's correlation (r) between frequencies of T-bet+IFN-γ+ (Th1-like), or of T-bethighIFN-γ (Th1-suppressing) Tregs, and frequency of T-bet+IFN-γ+ (Th1) Tconvs in different specimens from CHC pts. *P < 0.05; **P < 0.01; ns, not significant; na, not available. (D) Percentage of OX40+ cells in gated T-bet+IFN-γ+ (Th1-like) versus T-bethighIFN-γ (Th1-suppressing) Tregs in each TUM sample from CHC patients. *P < 0.05 by Wilcoxon's matched pairs test, two-tailed.

To ascertain the role of OX40 in human Treg-suppressive function, we set up an in vitro experimental assay by using Tregs enriched from PB of HDs (because of the paucity of liver-derived MNCs that did not allow the availability of high numbers of purified Treg required for these experiments) and preactivated in vitro for 18 hours with T-cell-stimulating beads, interleukin (IL)-2, and tumor necrosis factor alpha (TNF-α) to induce OX40 up-regulation.[26] Such preactivated Tregs were more suppressive than fresh Tregs in a standard in vitro suppression assay, performed by coculturing Tregs with scaled dilutions of carboxyfluorescein succinimidyl ester (CFSE)-labeled Tconvs in the presence of irradiated peripheral blood mononuclear cells and anti-CD3 (Supporting Fig. 2A). Strikingly, highly purified OX40+ Tregs, isolated from preactivated Tregs, were significantly more suppressive than the OX40 counterpart (Supporting Fig. 2B). To test whether OX40 engagement by an agonistic molecule inhibited the suppressive function of human Tregs, as previously observed in mice,[9, 10, 12] we performed experiments in which preactivated OX40-expressing Tregs, obtained as described above, were incubated for 6 hours with cross-linked rOX40L and then added to the suppression assay. In this setting, we could not detect any change in Treg-suppressive function in vitro after OX40 engagement (Supporting Fig. 2C). Altogether, these data suggest that OX40 signal neither improves nor diminishes the suppressive ability of human Tregs on a per cell basis. Rather, OX40 may mark highly suppressive Tregs in humans.

OX40 Up-Regulation and Th1-Oriented Plasticity Occur in Distinct Treg Subpopulations

To investigate whether Th1-like (mostly OX40) and Th1-suppressing (mostly OX40+) subsets belonged to distinct Treg subpopulations, we performed an extensive multiparameter flow cytometric (FCM) analysis aimed at identifying a marker, or a set of markers, cosegregating with the two phenotypes, in Tregs freshly isolated from the different liver compartments. First, we subdivided Tregs into CD45RA+FOXP3low resting, CD45RAFOXP3high activated (act-) Tregs, or CD45RAFOXP3low nonsuppressive (non-) Tregs.[27] We noticed that the subset of act-Tregs accumulated in TUM tissue in both nc and c pts, but was significantly higher in NT-LIV from c, compared to nc, pts (Fig. 4A). Both non-Tregs and act-Tregs, as well as the Tconv counterpart as a control, were further subdivided into two populations according to the high or low expression of Helios.[20] This analysis revealed that Helios expression appeared to be mutually exclusive with respect to IFN-γ production (Fig. 4B), recalling what was observed for OX40 expression (see Fig. 3D). A relevant proportion of Helioshigh Tregs could be detected in the non-Treg gate, but Helioshigh Tregs were maximally enriched in the act-Treg gate (Supporting Fig. 3A). When cells were stratified into Helioshigh and Helioslow subsets, a clear cosegregation of multiple markers appeared. Indeed, within all the populations studied, the Helioshigh cells were enriched in OX40+ (PD-1+ or –), in T-bethighIFN-γ and in proliferating (Ki67+) CD39+ cells (Fig. 4C). By contrast, the Helioslow subset was enriched in PD1+OX40, T-bet+IFN-γ+, CD39, and Ki67 cells (Fig. 4C). Some quantitative differences were observed among act-Tregs, non-Tregs, and Tconvs: (1) the Helioshigh act-Tregs contained the highest proportions of OX40+, T-bethighIFN-γ, and CD39+Ki67+ cells and the lowest frequencies of PD1+OX40 cells, as compared with the Helioshigh counterparts in non-Tregs and Tconvs, and (2) the Helioshigh subgroup of both act-Tregs and non-Tregs contained the lowest fractions of T-bet+IFN-γ+, as compared with the Helioshigh Tconvs (Fig. 4C). When we compared samples from nc and c pts, we could observe that Helioshigh/low subsets showed similar frequency (Supporting Fig. 3A) and similar phenotype (in terms of OX40, PD-1, CD39, and Ki67 expression), but different plasticity: Indeed, Helioslow act-Tregs contained a higher Th1-like Treg frequency in nc versus c pts (Supporting Fig. 3B), suggesting that both quantitative (different act-Treg frequency; see Fig. 4A) and qualitative modulations might contribute to shaping the Treg pool in distinct microenvironments. To demonstrate an association between Helios expression and Treg commitment, we analyzed, in sorted Helioshigh or Helioslow act-Tregs or non-Tregs, the frequency of cells carrying a demethylated TSDR, critically controlling the specification of Treg lineage.[3] We found that Helioshigh Tregs contained a significantly higher proportion of cells with demethylated TSDR, compared to Helioshigh Tregs, in both non-Tregs and act-Tregs subpopulations (Fig. 4D). We could not verify a higher TSDR demethylation in liver-infiltrating Helioshigh Tregs, because of the insufficient amounts of recovered cells required for this analysis. However, we could observe, in the bulk Treg pool obtained from NT-LIVc of a CHC pt, high levels of TSDR demethylation, in line with high Helioshigh frequency (Fig. 4D, right). In summary, our data demonstrate that OX40+ Th1-suppressing Tregs are mostly contained in Helioshigh, TSDR-demethylated, and act-Tregs, accumulating in cirrhosis and cancer, tempting us to hypothesize that OX40 may play a functional role in shaping the pool of activated, suppressive, and protumoral Tregs.

Figure 4.

Helioshigh subset is enriched in OX40+ and Th1-suppressing Tregs. (A) Representative plots of CD45RA versus FOXP3 staining in gated Tregs in different specimens from CHC pts, showing subsets of CD45RA+FOXP3low resting Tregs, CD45RAFOXP3high act-Tregs, or CD45RAFOXP3low non-Tregs (left). Percentage of each Treg subset in gated CD4+ T cells in the indicated samples (right). *P < 0.05; **P < 0.01; by Mann-Whitney's test, two-tailed, calculated for act-Treg. (B) Representative plots of Helios versus IFN-γ staining in gated Tconvs, non-Tregs, or act-Tregs from a cirrhotic liver specimen. (C) Representative plots of PD-1 versus OX40 (upper plots), IFN-γ versus T-bet (middle plots), and Ki67 versus CD39 (lower plots) in gated Helioshigh or Helioslow subsets of Tconvs, non-Tregs, or act-Tregs from a cirrhotic liver specimen. (D) TSDR demethylation levels in sorted Helioshigh or Helioslow subsets from act-Tregs or non-Tregs obtained from PB of 2 independent HDs (left) or in purified Tregs or Tconvs from PB or NT-LIVc of a CHC pt (right).

Liver-Derived, M2-Like, Macrophages Express OX40L and Induce Treg Proliferation

To verify whether an actual OX40 engagement by its ligand (OX40L) might explain the accumulation and expansion of OX40+ Tregs in NT-LIVc and much more in TUM tissue in vivo, we first identified the cell type mostly accounting for OX40L expression. Among the various cell populations examined, that is, CD8 and CD4 T cells, B cells, myeloid, plasmacytoid dendritic cells (DCs), slanDC (a strongly inflammatory DC subset involved in response to pathogens[28]), and basophils, monocytes/macrophages (mono/MF) represented the most abundant cell type that expressed OX40L at the highest levels in all samples examined in CHC pts and HDs (not shown). Interestingly, OX40L expression was particularly prominent in the so-called classical CD14highCD16low, as compared with nonclassical CD14lowCD16high, mono/MF subsets[29] in both PB and liver (Fig. 5A). Furthermore, it was maximally up-regulated on a fraction of classical CD14highCD16low liver macrophages that expressed (in contrast to the nonclassical ones) high levels of CD163 and CD206, two markers related with an alternative M2 program of monocyte differentiation (M2-like mono/MF30; Fig. 5A). These liver-infiltrating mono/MF showed a basal TNF-α and IL-10 production ex vivo that was not boosted by lipopolysaccharide (LPS) or LPS+IFN-γ stimulation, in contrast to their peripheral counterparts (Fig. 5B), a phenotype confirming their M2-like identity.[30-32] In line with FCM data, in situ immunohistochemical (IHC) analysis showed that macrophages with spindle or stellate morphology expressing the M2 marker, CD163, abundantly infiltrated the HCC parenchyma and showed direct spatial interaction with FOXP3-expressing Tregs. Double-marker IHC stainings for CD163 and OX40 and for OX40 and FOXP3 confirmed the expression of OX40 on Treg cells contacting CD163+ macrophages (Fig. 5C). Despite the fraction of CD206+ mono/MF was expanded in liver samples, compared to PB (and significantly in NT-LIVc), OX40L expression by mono/MF was comparable in PB, NT-LIV, and TUM, suggesting that tissue-derived signals unlikely regulate OX40L expression on mono/MF (Fig. 5D). Rather, virus-dependent factors may affect OX40L levels: Indeed, OX40L expression directly correlated with plasmatic viral load in all samples examined (Fig. 5E).[33] Suggesting OX40L as a signal playing peculiar functions in CHC, we found a trend toward a lower OX40L expression level in mono/MF from PB, NT-LIV, or TUM tissues from a small cohort of pts with non-CHC-related cirrhosis, even if these differences lacked statistical significance (Supporting Tables 2 and 3; Fig. 5D). The positive correlation between the percentage of OX40+ Tregs and Treg frequency (see Fig. 2B) suggested the idea that OX40/OX40L interaction may lead to the promotion of Treg proliferation. To explore this possibility, fresh OX40L-expressing monocytes or a cross-linked recombinant (r) form of OX40L (as a control) were tested for their capacity to promote the proliferation of OX40+ Tregs (that were previously preactivated in vitro, as described above), in response to T-cell-stimulating beads, through the OX40/OX40L axis, in the presence or absence of a blocking anti-OX40L antibody (Ab). We found that rOX40L significantly enhanced Treg proliferation (Fig. 5F) and also boosted the expression level (in terms of MFI) of its own receptor, OX40, both in the presence and absence of TNF-α, suggesting the existence of a positive feedback in the OX40/OX40L axis (Fig. 5G). More important, OX40L-expressing monocytes hugely increased the proliferation of Tregs in coculture assays in a fashion that was partially, but significantly, dependent on OX40L/OX40 interaction (Fig. 5F). Overall, these data suggest that the OX40L, shown to enhance proliferation of OX40+ Tregs, may be provided by M2-like mono/MF in vivo. However, this signal unlikely determines the differential distribution of OX40+ Tregs at distinct sites in vivo, being that OX40L was expressed with no significant differences in NT-LIVnc, NT-LIVc, or TUM samples.

Figure 5.

OX40L is expressed by M2-like monocytes in the liver. (A) FCM analysis of CD206, CD163, and OX40L expression in mono/MF, gated as CD14+CD16high/low cells, from a representative of 7 CHC pts analyzed. fmo, fluorescence-minus-one negative control. Similar results have been obtained from TUM samples. (B) Intracellular staining for TNF-α and IL-10 in gated mono/MF from 1 representative of 13 patients analyzed. Similar results have been obtained from TUM samples. (C) IHC analysis of CD163, OX40, and FOXP3 in a representative HCC sample, showing close proximity between CD163+ M2-like mono/MF and OX40+ Tregs. (D) Frequency of CD206+ cells (left plot) and expression level of OX40L (right plot), shown as rMFI (representing the ratio between OX40L median fluorescence intensity in monocytes and in lymphocytes in each sample), in gated mono/MF from different specimens. Bars represent mean ± standard error of the mean. (E) Spearman's correlation (r) between serum viral load (HCV RNA, IU/mL) and OX40L expression in CD14+CD16high/low mono/MF, calculated in each compartment (PB, liver, or tumor) of CHC pts. *P < 0.05; **P < 0.01. (F) Tregs purified from PB of HDs were CFSE labeled and cultured for 6 days with equal numbers of T-cell-stimulating beads and IL-2; where indicated, after 18 hours, we added cross-linked recombinant OX40L (rOX40L), or autologous monocytes expressing OX40L, with or without anti-OX40L blocking monoclonal Ab (αOX40L). Treg proliferation in gated FOXP3+ cells at day 6 is shown. Data shown are from a representative of seven experiments performed with individual donors; each condition has been tested in triplicates or quadruplicates. *P < 0.05; **P < 0.01; ***P < 0.005; by unpaired Student t test, two-tailed. (G) Tregs purified from PB of HDs were CFSE labeled and cultured for 6 days with equal numbers of T-cell-stimulating beads and IL-2, with or without TNF-α and cross-linked recombinant OX40L (added after 18 hours). OX40 expression in gated FOXP3+ cells at day 6 is shown. Data shown are from a representative of three experiments performed with individual donors.

IL-12 Local Availability, Rather Than IL-12 Responsiveness, Shifts the Balance Between Th1-Like and Th1-Suppressing Tregs

Then, we asked whether the NT-LIVnc microenvironment provided selective signals promoting Th1-like polarization and counteracting the OX40/OX40L pathway in those districts.

The amounts of IL-12 and IFN-α, but not of other cytokines, such as IL-27, IL-1β, IL-6, and TNF-α, spontaneously released per gram of tissue in 6 hours from unprocessed tissue fragments, were significantly reduced in both NT-LIVc and TUM, as compared with NT-LIVnc specimens (Fig. 6A), indicating that local availability of these cytokines may orientate the outcome of Treg polarization in different microenvironments.

Figure 6.

IL-12 favors, whereas OX40 antagonizes, Th1-like Treg polarization. (A) Amounts of different cytokines released in 6 hours in tissue-conditioned medium from NT-LIVnc (n = 2), NT-LIVc (n = 5), and TUM (n = 4) fragments. Only detectable cytokines, among all the tested molecules, are depicted. Means ± standard error of the mean are shown. *P < 0.05; **P < 0.01, by unpaired Student t test, two-tailed, calculated for each cytokine with respect to NT-LIVnc. (B) Representative plots showing IFN-γ versus IL-12R-β2 in gated Tregs or Tconvs from PB and NT-LIVc, from 1 of 3 tested patients. (C) Intracellular staining of IFN-γ and T-bet in gated Tregs or Tconvs from NT-LIVc cultured with T-cell-stimulating beads and treated or not ex vivo for 3 days with IL-12. Data shown are from a representative experiment of three individual replicates performed with NT-LIVc or TUM specimens showing similar results. (D) Tregs purified from PB of HDs were preconditioned 6 days with IFN-γ (pre-IFN-γ), cultured for a further 3 days with IL-12, and stained for IFN-γ, T-bet, and IL-12R-β2. Data shown are from a representative of three independent experiments performed with individual donors; each condition has been tested in triplicates. *P < 0.05; **P < 0.01; ***P < 0.005; by unpaired Student t test, two-tailed. (E) Representative plots showing IFN-γ versus T-bet expression in gated Helioshigh or Helioslow Tregs polarized in vitro under Th1 conditions (pre-IFN-γ>IL-12). (F) Representative plots showing OX40 versus IFN-γ, T-bet, or IL-12R-β2 expression, in gated Tregs ex vivo from NT-LIVc or TUM of CHC pts (upper plots), or in Tregs polarized in vitro into Th1-like cells (pre-IFN-γ>IL-12; lower plots). (G) Frequency of T-bet+IFN-γ+ (Th1-like) Tregs, polarized in vitro (pre-IFN-γ>IL-12), under exposure to TNF-α, and/or cross-linked rOX40L. Data shown are from a representative of two independent experiments performed with individual donors; each condition has been tested in triplicates. *P < 0.05; **P < 0.01; by unpaired Student t test, two-tailed. (H) OX40 expression in gated Tregs at day 6 of the in vitro polarization setting, after culture with IFN-γ.

IL-12 plays crucial roles in the development of conventional Th1 cells and also in the polarization of Th1-like Tregs,[24, 25] as well as expression of IL-12R-β2 (the inducible subunit of IL-12 receptor) in dictating the complete polarization of murine Th1-like Tregs.[24] Therefore, we tested whether defective IL-12 susceptibility could be responsible for the reduced frequency of Th1-like Treg in cirrhosis and cancer. An impaired IL-12R responsiveness of Tregs from these districts was excluded for two main reasons: First, IL-12R-β2 was expressed in vivo by liver-infiltrating Tregs, especially the IFN-γ-producing Tregs, irrespective of their tissue microenvironment origin (Fig. 6B and Supporting Fig. 4); second, a short-time ex vivo culture of cirrhosis- or tumor-infiltrating cells in the presence of exogenous IL-12 increased IFN-γ secretion not only by Tconvs, but also by Tregs (Fig. 6C). To dissect the specific functions operated by IL-12 in Treg plasticity, we set up a two-stage in vitro system of Th1-like Treg polarization, adding IFN-γ for the first 6 days and then IL-12 for a further 3 days to freshly isolated Tregs, according to a protocol developed by others using murine Tregs.[24] Preconditioning with IFN-γ increased both the frequency of T-bethighIFN-γ Tregs and the expression of IL-12R-β2, events that were further amplified by IL-12 addition; however, as with murine Tregs,[24] and in line with ex vivo data shown above, IL-12 addition resulted in a significant polarization of human Th1-like Tregs (Fig. 6D). Of note, IFN-α (which was also significantly less released in both NT-LIVc and TUM, compared to NT-LIVnc) was not able to induce Th1-like Treg polarization in the same in vitro assay (Supporting Fig. 5). In line with data ex vivo (Fig. 4C), also in this polarizing culture in vitro the Helioslow Treg subset was enriched in T-bet+IFN-γ+ cells, compared to the Helioshigh counterpart (Fig. 6E). In summary, these data show that the preferential polarization of Th1-like Tregs in NT-LIVnc (compared to NT-LIVc or TUM) contexts may be especially oriented by IL-12, more abundant in chronically inflamed liver tissues that are not affected by cirrhosis or cancer. We cannot exclude that other cytokines abundant in NT-LIVnc, such as IFN-α and IL-27, and the modulation of their receptors, may possess overlapping, or even contrasting, functions in Th1-like Treg polarization.

To explore the idea that preferential accumulation of Th1-like (mostly OX40) and Th1-suppressing (mostly OX40+) Tregs in NT-LIVnc and NT-LIVc/TUM districts, respectively, was a result of different availability of Th1-polarizing cytokines (e.g., IL-12), we performed experiments to determine whether OX40 and Th1 polarization showed phenotypical and/or functional antagonism. OX40 expression appeared to be mutually exclusive with respect to IFN-γ and IL-12R-β2, but not T-bet, expression, in both liver-derived Tregs ex vivo and Th1-like Tregs polarized in vitro, indicating that OX40 may be associated with a blocked transition from Th1-suppressing into Th1-like cells, reasonably at the step of IL-12R-β2 up-regulation[24] (Fig. 6F). To test the possibility that OX40 up-regulation and Th1-oriented plasticity might represent alternative processes that are functionally regulated, we analyzed the consequences of OX40 engagement, by cross-linked rOX40L, on Th1-like Treg polarization assay in vitro. In this context, we also tested the role of TNF-α, based on the above-described effects of TNF-α on Treg expansion and OX40 induction.[26, 34] We could observe that IFN-γ acquisition was significantly decreased after OX40 engagement and nearly abolished upon TNF-α exposure, demonstrating that OX40 (as well as TNF-α) prevented Treg deprogramming into Th1-like cells (Fig. 6G). Vice versa, we observed that IFN-γ exposure significantly reduced OX40 expression level (in terms of MFI) on the Treg surface (Fig. 6H), suggesting that a functional reciprocal regulation existed between OX40 and Th1 signals, which may contribute to fix the divergence between Th1-like and -suppressing pathways.


This study represents the first comprehensive analysis of Treg phenotypic and functional complexity within diverse human tissue microenvironments, characterized by chronic inflammation, fibrosis, or oncogenic transformation, within the same organ (i.e., the liver). We found that cirrhosis and tumor microenvironments favor the expansion of Tregs, especially act-Tregs, marked by a phenotypic and molecular signature of Treg stability and specialized Th1 suppression (T-bethighIFN-γ, OX40+, Helioshigh, CD39+, demethylated TSDR, and high suppressive function); on the contrary, chronic HCV-associated liver inflammation in the absence of cirrhosis provides a series of proinflammatory signals that rather promoted the preferential diversion of Tregs into Th1-like cells (T-bet+IFN-γ+, OX40, Helioslow, CD39, methylated TSDR, and low suppressive function). In addition, the finding that the percentage of OX40-expressing Tregs is positively correlated with Treg frequency and proliferation ex vivo supports the idea that OX40 can provide signals contributing to Treg survival and expansion in cirrhotic or tumor microenvironments in humans. Taken together, these data suggest that the cirrhotic microenvironment favors recruitment and expansion of committed Tregs, thus establishing a state of immune tolerance, ultimately contributing to the evolution of cirrhosis into cancer. By contrast, recruitment and expansion of unstable Th1-like Tregs in the inflammatory liver microenvironment would contribute to amplify inflammation, rather than suppression.

The natural ligand of OX40 (OX40L) can be expressed by a variety of cell types, including macrophages.[7] Here, we show, to our knowledge, for the first time, that M2-like mono/MF[30] express high OX40L levels and expand highly suppressive OX40+ Tregs, directly by the OX40/OX40L interaction. These mono/MF spontaneously produced TNF-α and IL-10, a phenotype resembling IL-10-producing, M2-like, tumor-associated macrophages[30] and reminiscent of chronic, low-dose TNF-α exposure that plays critical roles in tumor-promoting inflammation.[31] However, such “alternatively activated” macrophages appear equally distributed in NT-LIVnc and NT-LIVc/TUM districts, suggesting that other signals likely drive the differential accumulation of Th1-like versus Th1-suppressing Tregs in distinct contexts. Contrary to OX40L, the expression of its receptor, OX40, in Tregs is indeed differently modulated in distinct microenvironments, being expressed at higher frequency in cirrhosis and tumor. Not only OX40L itself, but also other signals may be involved in OX40+ Treg accumulation at those sites. Among others, we have underscored a negative role for IFN-γ in OX40 up-regulation in vitro, an observation suggesting that, in NT-LIVnc, OX40+ Treg frequency could be maintained at lower levels by the higher IFN-γ amounts produced by Th1 cells in that context, which are sustained by IL-12. Conversely, in cirrhosis and cancer, OX40+ Tregs are allowed to accumulate and interact with OX40L+ mono/MF. In turn, OX40L may stabilize their phenotype, further increasing their proliferation and OX40 expression and counteracting Th1-like plasticity. Therefore, the dichotomy in distribution of Th1-suppressing and Th1-like Tregs in different districts may be, at least partly, attributable to the interplay between the OX40/OX40L axis and type-1 cytokines, especially to IL-12 that is available in NT-LIVnc, but not in NT-LIVc/TUM districts (probably depending on both quality and quantity of the inflammatory response in each tissue), and can achieve a significant polarization of human Th1-like Tregs in vitro. This observation recommends the possible usage of compounds stimulating IL-12 production in tumors to favor the generation of antitumor effector cells, rather than immunosuppressive committed Tregs.[35] In tumor/pretumor contexts, the paucity of type-1 cytokines may not only prevent an extensive Th1-like Treg polarization, but also allow OX40 expression (which is inhibited by IFN-γ), and the OX40/OX40L interaction with OX40L+ mono/MF that can contribute to the expansion of OX40+ Tregs only in those districts. In turn, OX40L signal can help in amplifying and fixing this Treg phenotype by fostering Treg proliferation, inhibiting Treg plasticity, and up-regulating its own receptor, OX40 (Supporting Fig. 6).

In line with data by others,[33] OX40L expression by mono/MF directly correlated with HCV viral load. Because some data suggest high viremia as a risk factor for HCC development or recurrence,[36] and because Treg intratumoral accumulation is associated with unfavorable HCC prognosis,[37] we are tempted to speculate that high HCV load may support tumor progression also through OX40L-mediated local Treg expansion and the resulting immune suppression. This hypothesis can account for the observation that both OX40+ Treg expansion and high mono/MF-OX40L expression in cirrhosis or tumor was peculiar of CHC, but not of hepatitis B or AFLD, pts. Further studies are in progress to verify whether Treg expansion in tumor and cirrhosis is dependent on the synergistic effects by OX40 and antigen-specific (HCV-related? Tumor-associated?) signals.

Our results support the notion that Helios may label truly committed Tregs with stable FOXP3 expression and demethylated TSDR, which are less susceptible to diversion into Th1-like cells, but are more prone to up-regulate OX40, to proliferate and become specialized Th1 suppressors. In particular, whereas Helios appears to mark committed Tregs as developmentally determined, OX40 may be viewed as an activation marker, expressed preferentially by committed Tregs infiltrating tumor or pretumor microenvironments (i.e., HCC and cirrhosis). Therefore, the predominant effect of OX40 triggering in vivo would be the expansion of highly suppressive OX40+ Tregs, which, in turn, would favor tumor progression, an issue that should be carefully taken into account when designing OX40-targeted anticancer immunotherapies.[13] In this context, targeting, rather than triggering, the OX40 receptor may represent a suitable strategy to block Tregs in cancer immunotherapy in humans (at least in HCC), by eliminating or paralyzing tumor-associated OX40+ Tregs, comprising specialized Th1-suppressing and committed (Helioshigh) cells, while sparing systemic immune homeostasis, maintained by peripheral Helios+ Tregs not expressing OX40. In conclusion, our data point out that human Tregs phenotypically and functionally adapt to microenvironmental signals in liver tissues affected by peculiar pathological conditions, and suggest that a global comprehension of Treg heterogeneity and plasticity is mandatory to well-designed therapies for chronic inflammatory diseases and cancer.


The authors thank Marco Cassatella and Federica Calzetti (Università di Verona) for providing anti-M-DC8 antibody, Maria Cristina Gagliardi (Istituto Superiore di Sanità, Rome) for providing anti-CD206 antibody, Carla Guarnotta (Università di Palermo) for technical assistance in immunohistochemical stainings, and Stefania Morrone (“Sapienza” Università di Roma, Rome) for help with cell sorting. The authors also acknowledge Massimo Locati and Alberto Mantovani for their helpful discussion.