Control of GVHD by regulatory T cells depends on TNF produced by T cells and TNFR2 expressed by regulatory T cells

Acute graft-versus-host disease (GVHD) is one of the major causes of morbidity and mortality following allogeneic hematopoietic stem cell transplantation (allo-HCT). In murine models, cell therapy using CD4⁺CD25^highFoxp3⁺ regulatory T cells (Tregs) was shown to efficiently prevent GVHD^1-3  without hampering immune reconstitution⁴  or the graft-versus-leukemia (GVL) effect.^5,6  The efficacy and reproducibility of this therapy in preclinical models over the past 15 years by different groups in different mouse models using murine or human polyclonal or antigen-specific Treg cells led to the development of clinical trials with promising preliminary results.^7-9  The mechanisms enabling this action rely on the ability of Tregs to directly act on conventional T cells (Tconvs).^4-6,10  Here, we hypothesize that as Tregs are capable of modulating Tconv response, likewise Tconvs also modulate Treg action, resulting in an effective feedback mechanism.

Tumor necrosis factor (TNF) constitutes a likely candidate for mediating the effect of Tconvs on Tregs for several reasons: (1) This cytokine is abundantly produced during the cytokine storm following allo-HCT.¹¹  (2) In the setting of autoimmunity, TNF was previously shown to boost proliferation and suppressive activity of Tregs by signaling through its type 2 receptor (TNFR2).^12-15  This was observed in mice and humans, both in healthy individuals and type 1 diabetes patients.^16-18  (3) Tregs express higher levels of TNFR2 compared with Tconvs, and Tregs expressing the highest level of TNFR2 are most suppressive in mice and humans.^12,13,15 

The role of TNF/TNFR2 interaction in the control of GVHD by Tregs has not been studied. Addressing this question may help us to better understand the physiopathology of the disease. It is also relevant in regard to anti-TNF treatment that is sometimes administered to patients with severe GVHD. The activation of TNF receptor type 1 (TNFR1) could be associated with increased inflammation and tissue damage in allo-HCT¹⁹  and blocking this pathway could ameliorate GVHD by reducing cell death of gastrointestinal cells. Thus, one may hypothesize that anti-TNF treatment would have a beneficial effect by blocking TNFR1 triggering but a detrimental effect by blocking TNFR2 triggering on Tregs, the latter phenomenon being assessed in this work. Here, we explored the hypothesis that in the context of Treg-based therapy of GVHD, Tconvs regulate Tregs via TNFR2 signaling. Using 3 different experimental approaches to prevent TNF/TNFR2 interaction, we showed that the suppression of GVHD obtained by transfer of Tregs strongly depended on TNFR2 expressed by Tregs and TNF produced by Tconvs.

Wild-type (WT) C57BL/6 (B6 H-2^b) and B6C3HF1 (H-2^kxb) mice were purchased from Harlan Laboratories (Gannat, France) and Charles River Laboratories (Saint-Germain-Nuelles, France). TNF^−/− and TNFRs1b^−/− (TNFR2^−/−) mice were purchased from the Jackson Laboratory (Bar Harbor, ME). All mice were on a C57BL/6 background. Mice were housed under specific pathogen-free conditions. All experimental protocols were approved by the local ethics committee (authorization number 11/12/12-5B) and are in compliance with European Union guidelines.

Tregs were prepared as previously described.²⁰  Briefly, spleens and lymph nodes from C57BL/6 female mice were collected and mechanically dilacerated. Cell suspension was stained with biotin-coupled anti-CD25 monoclonal antibody (mAb) (7D4; BD Biosciences, San Diego, CA), followed by anti-biotin microbeads (Miltenyi Biotec, Paris, France), and CD25⁺ cells were positively selected through magnetic large selection column (Miltenyi Biotec). Selected cells were stained with the following mAb’s: CD4–fluorescein isothiocyanate (FITC; eBioscience, San Diego, CA), CD62L–phycoerythrin (PE; eBioscience), CD25-biotin (BD Biosciences), and streptavidin-PE-Cy5 (eBioscience). CD4⁺ CD25^high CD62L^high cells were then sorted using a MoFlo Legacy (Beckman Coulter, Villepinte, France), with a purity of 99%. For HY-Treg preparation, cells were cultured for 3 to 4 weeks in the presence of recombinant murine interleukin-2 (IL-2) (10 ng/mL; PeproTech, Neuilly-sur-Seine, France) and weekly stimulated with CD8⁺ dendritic cells previously loaded with the HY peptide (10 µg/mL, N-15-S, NY; PolyPeptide, Strasbourg, France) in the presence of granulocyte-macrophage colony-stimulating factor (20 ng/mL; PeproTech). CD8⁺ dendritic cells were isolated from splenocytes of C57BL/6 mice, as previously described.²⁰  For recipient-specific (rs)–Treg preparation, cells were cultured for 3 to 4 weeks in the presence of recombinant murine IL-2 and weekly stimulated with irradiated total splenocytes from C3H female mice, as previously described.^6,10 

Eight- to 12-week-old recipient B6C3HF1 female mice received a 10 Gy irradiation followed by retro-orbital infusion of 10 × 106 bone marrow (BM) cells + 2 × 106 CD3⁺ T cells, with or without HY-Treg cells in a 1:1 ratio (ie, 2 × 106 HY-Treg cells). BM and T-cell suspensions were prepared using leg bones and splenocytes, respectively, as previously described. All infused cells were isolated from female C57BL/6 mice (semi-allogeneic model). As recipient and donor mice were females, HY-Treg cells were activated in vivo by repeated retro-orbital infusions of 100 μg of the HY peptide (at D0, D1, D3, and D6), as previously described.²⁰  For rs-Treg experiments, mice were transferred with rs-Treg cells in a 1:1 ratio.

Anti-TNFR2 (TR75-54.7) mAb was purchased from Bio X Cell (West Lebanon, NH). Recipient mice were treated with 3 intraperitoneal injections of 500 μg of the antibody on days 0, 2, and 4.

GVHD clinical score was calculated 2 to 3 times per week. Each of the 5 following parameters was scored 0 (if absent) or 1 (if present): weight loss >10% of initial weight, hunching posture, skin lesions, dull fur, and diarrhea. Dead mice received a global score of 5. Mice were euthanized in case of weight loss >30% of initial weight or maximal clinical grade (ie, 5/5).

Livers, lungs, skin, and small and large bowel samples were preserved in Bouin's fixative and embedded in paraffin. For these organs, 5-µm-thick sections were stained with hematoxylin and eosin for histological examination as previously described.^6,10  Briefly, 1 pathologist analyzed slides in a blinded fashion to assess the intensity of GVHD. GVHD lesions in each sample were scored according to a semiquantitative scoring system described by Hill et al with minor modifications.²¹ 

Two weeks after transplantation (ie, at D13, D0 being the date of transplantation and Treg cell injection), recipient mice were euthanized and their spleens collected. Because of the low proportion of Treg cells among splenocytes and the low overall spleen cellularity at D13, cell suspensions obtained for each spleen were enriched in CD4⁺ and CD8⁺ cells after labeling with anti-CD4 and anti-CD8 microbeads (Miltenyi Biotec) and positive magnetic selection through large selection columns (Miltenyi Biotec). Selected cells were then stained with the following mAb’s: CD4-FITC, CD4-allophycocyanin and CD4-Vioblue (Miltenyi Biotec), Foxp3-PE-Cy5 and Foxp3-V450 (eBioscience), CD25-PE-Cy7 (eBioscience), CD62L-PE (eBioscience), inducible T-cell costimulator (ICOS)–PE (eBioscience), cytotoxic T lymphocyte associated protein 4 (CTLA4)–biotin (followed by streptavidin-PE-Cy7; eBioscience), interferon γ (IFNγ)–PE (Miltenyi Biotec), TNFα-FITC (Miltenyi Biotec), and CD8α-FITC (eBioscience). Intracellular Foxp3 staining was performed according to the manufacturer’s instructions, using the Foxp3 staining buffer set from eBioscience. For intracellular cytokine staining, cells were restimulated with 1 μg/mL PMA (Sigma Aldrich, Saint Quentin Fallavier, France) and 0.5 μg/mL Ionomicyn (Sigma Aldrich) for 5 h, in the presence of GolgiPlug (1 μL/mL; BD Biosciences). Events were acquired on a FACSCanto II flow cytometer (BD Biosciences) and analyzed using FlowJo software vX.0.7 (FlowJo LLC, Ashland, OR).

Prism (GraphPad Software) was used for statistical analysis. Kaplan-Meier survival curves were compared using log-rank test. For analysis of GVHD clinical grading curves, area under curve (AUC) was calculated for each mouse, and then Student t test or 1-way analysis of variance (ANOVA) with post hoc analysis was performed depending on the number of comparatives. For cytometry analysis, we have normalized the mean fluorescence intensity (MFI) values with T-cell + Treg-cell control group. Then we used unpaired, 2-tailed Student t tests for generation of P values.

To assess the role of TNF on GVHD protection by Treg administration, we first used our recently described model in which the disease was prevented by transfer in female recipients of donor Tregs specific for the exogenous (ie, nondonor, nonrecipient) HY antigen at time of HCT, followed by their in vivo reactivation by HY peptide immunization.²⁰  In a semi-allogeneic condition C57BL/6 → [B6xC3H]F1 of BM transplantation, GVHD protection at 1:1 Treg/Tconv ratio strictly depends on HY immunization. We evaluated the role of TNFR2 in Treg-mediated GVHD protection using a blocking anti-TNFR2 mAb (Figure 1A-B). Mice transferred with Tconvs developed severe GVHD that was prevented by the cotransfer of HY-Tregs. The beneficial effect of Treg administration was fully abolished in mice that were treated with the blocking anti-TNFR2 mAb. These latter mice displayed high clinical GVHD scores and decreased survival, as compared with HY-Treg-treated control mice. In order to assess whether a higher Treg/Tconv ratio could overcome the disadvantage because of TNFR2 blockade, we reproduced the same experiment doubling the number of HY-Treg infused in recipient mice (2:1 Treg/Tconv ratio). Even with this increased number of therapeutic Tregs, blocking TNFR2 fully abolished the Treg-dependent GVHD protection (supplemental Figure 1, available on the Blood Web site).

We then used TNFR2-deficient HY-specific Tregs, obtained from TNFR2–knockout (KO) mice, to confirm that the Treg control of GVHD by TNFR2 was mediated by TNFR2 expression by Tregs. Whereas TNFR2-sufficient control Tregs fully protected from GVHD, TNFR2-deficient Tregs completely failed to prevent the disease (Figure 2). Survival of the mice and clinical scores of GHVD were identical in mice receiving donor Tconvs alone and mice receiving donor Tconvs and TNFR2-deficient Tregs.

We then assess the role of TNFR2 in Treg-mediated protection in another transplant setting. Tregs naturally present in the donor T-cell inoculum were present in sufficient number to attenuate GVHD because their depletion accelerated the disease.^1-3  Here, we observed that blocking TNFR2 led to a similar GVHD aggravation. Indeed, in mice grafted with BM cells and whole T cells containing Tregs at physiological level, administration of the blocking anti-TNFR2 mAb induced an accelerated GVHD (Figure 1C-D). Histological analyses performed at day 14 confirmed the exacerbated GVHD in the liver (P < .001) and a trend to an increase of GVHD for the small intestine and the colon of anti-TNFR2–treated mice (supplemental Figure 2). Surprisingly, we observed a reduced GVHD in the skin of grafted animals after anti-TNFR2 mAb treatment. This could be explained by the distinct biological feature of cutaneous Tregs vs Tregs of other tissues, including a specific pattern of homing receptors as recently reviewed,²²  which may modify their sensitivity to TNF/TNFR2 signaling.

The number of splenocytes collected at day 14 importantly varied between the 2 groups of mice. Whereas spleens of mice grafted with T cells contain 65.4 × 10⁶ ± 2.2 cells, this number fell sharply to 12.3 × 10⁶ ± 9.6 in mice treated with anti-TNFR2 mAb, probably reflecting an accelerated GVHD. We next evaluated the effect of blocking TNFR2 on cytokine production in the spleen of mice developing GVHD after transfer of WT donor T cells. Mice treated with the anti-TNFR2 mAb had an increase in IFNγ and TNFα production in both CD4 and CD8 donor T cells (Figure 3).

In order to reinforce the robustness of our observations, we used a third transplant setting consisting of infusing Tregs that were rendered specific for recipient-type alloantigen (namely, rs-Treg) instead of HY-Treg to prevent GVHD. Whereas GVHD was prevented by rs-Treg administration, this protective effect was fully abolished when using anti-TNFR2 mAb (Figure 4).

Thus, using 2 different approaches (anti-TNFR2 mAb and TNFR2-deficient Tregs) and different types of Tregs (Tregs of the T-cell inoculum, therapeutic HY-Tregs, or rs-Tregs), we demonstrate that the control of GVHD by Tregs is TNFR2 dependent.

TNF can be produced by multiple cell types after allo-HCT. Because donor T cells are key actors of GVHD and one of the major TNF producers,¹¹  we evaluated whether GVHD prevention by Tregs would be maintained when using TNFα-deficient donor T cells collected from TNFα-deficient mice. We first compared GVHD in mice transferred with TNFα-deficient and TNF-sufficient whole donor T cells in the absence of additional therapeutic Tregs. Both groups developed a similar clinical GVHD (Figure 5C-D). However, histological and biological analyses performed at day 14 showed that mice receiving TNFα-deficient T cells developed a more severe GVHD compared with mice transferred with TNF-sufficient T cells. Histological scores in the liver and small intestine were higher in the former mice compared with the latter mice (supplemental Figure 3). Also, numbers of splenocytes were reduced in mice transferred with TNFα-deficient T cells compared with TNFα-sufficient T cells. This was associated with an increase in IFNγ production in both CD4 and CD8 donor T cells in mice receiving TNFα-deficient T cells (Figure 6).

We then assessed whether GVHD protection by therapeutic HY-specific Tregs was maintained when donor T cells were TNF deficient. Although the cotransfer of HY-specific Tregs protected from GVHD in mice receiving TNF-sufficient T cells, this beneficial Treg effect was entirely lost with TNFα-deficient donor T cells (Figure 5A-B). Actually, we even observed that mice grafted with TNFα-deficient donor T cells supplemented with Tregs experienced an even greater degree of GVHD (trends in an accelerated mortality) compared with mice receiving only TNFα-deficient donor T cells. The loss of GVHD protection by Tregs when TNF-deficient T cells were transferred was confirmed when therapeutic rs-Tregs were injected in place of HY-Tregs (Figure 4). Thus, TNFα produced specifically by donor T cells is required for the Treg suppressive effect in GVHD.

To analyze by what mechanism the control of GVHD by Tregs depends on TNF/TNFR2 interaction, we measured the proportion and activation markers of Tregs in the spleen collected at day 13 in mice grafted with HY-Tregs and either WT T cells treated with anti-TNFR2 mAb or TNF-deficient T cells. First, the expression level of Foxp3 was significantly reduced when TNF/TNFR2 interaction was inhibited in both settings, whereas Treg proportions remained unchanged (Figure 7A-B). This lower Foxp3 expression among whole Tregs was characterized by a reduced proportion of Foxp3^high-expressing cells and an increased proportion of Foxp3^int-expressing cells in both experimental models (supplemental Figure 4). A likely explanation would be that, in the absence of TNFR2 signaling in Tregs, Foxp3 would be downmodulated, suggesting that TNF stabilized Foxp3 expression in Tregs. We further analyzed the expression of CD25, the α chain of the IL-2 receptor constitutively expressed at the Treg cell surface membrane. We observed a dramatic decrease of the percentage of CD25⁺ cells and CD25 expression level among Tregs when anti-TNFR2 mAb was administered to grafted mice. Because in experimental allo-HCT CD25 expression is upregulated by IL-2,²³  these results suggest that TNFα increased IL-2 responsiveness of Tregs in this context. Finally, we evaluated the expression of ICOS and CTLA4, which are important molecules in Treg biology. In both models of inhibition of TNF/TNFR2 interaction, ICOS and CTLA-4 expressions were reduced compared with controls, with a more pronounced effect in mice treated with the anti-TNFR2 mAb than mice grafted with TNFα-deficient T cells (Figure 7), probably reflecting the more complete abrogation of TNF signaling in the presence of the mAb.

Here, we describe in the context of alloreactive immune responses a novel feedback mechanism by which Tconvs boost Treg activation. Indeed, we clearly demonstrate that the protection of GVHD by Tregs in allo-HCT strictly depends on TNF produced by effector T cells and TNFR2 expressed by Tregs. We believe our findings are solid because they rely on 3 different experimental approaches (blocking TNFR2 mAb, TNFR2-deficient Tregs, and TNFα-deficient Tconvs) and using 3 types of protective Tregs (polyclonal Tregs present at physiological level in the T-cell inoculum, therapeutic HY-Tregs, and therapeutic rs-Tregs). If a similar phenomenon exists in human, TNF-TNFR2 interaction would be critical in the control of GVHD by Tregs in allo-HCT performed either in routine or in clinical trials when therapeutic Tregs are injected.

These findings are in accordance with recent reports showing that, in the context of autoimmune diabetes, effector T-cell activation was able to boost Treg cell expansion, favoring disease control, a phenomenon partly dependent on TNF.^14-16,24  Our results extend, in an allo-HCT setting, the observation that Tconvs exert a powerful boost of Treg activity. We also demonstrate for the first time that the sole defect of TNF production by donor T cells was sufficient to completely abolish the Treg suppressive effect in GVHD. Our findings expand our understanding of the diverse and complex effects of inflammation on Treg biology. Indeed, it was shown that some inflammatory factors, such as IL-6 or IL-4, promoted Foxp3 downmodulation and Treg instability.^25,26  To the contrary, our study suggests that another inflammatory factor, TNF, maintained a high level of Foxp3 in Tregs in an inflamed environment. This is consistent with previous data in the context of autoimmune diseases in mice¹⁶  and with our recent findings that TNF increased Foxp3 expression in human Tregs in vitro.²⁷ 

These results shed light on the conflicting data regarding the effect of anti-TNF drugs. These agents are widely used in the treatment of various autoimmune diseases. However, they are not efficient in all patients, and there are numerous reports describing paradoxical exacerbation of the initial disease or the occurrence of new autoimmune syndromes.^28-31  Similarly, a number of clinical studies failed to demonstrated efficacy for anti-TNF treatment in preventing or treating GVHD.^32,33  Our results support that the potential benefit of anti-TNF drugs in GVHD patients may also be counteracted by the inhibition of Treg function, thus promoting a vigorous alloreactive T-cell response.

Finally, we believe that our results pave the way for a novel immune checkpoint therapy to modulate alloreactivity after allo-HCT. Indeed, the TNF/TNFR2 signaling pathway represents a potential target for modulating alloreactivity. We previously observed that ex vivo Treg depletion from donor lymphocyte infusions could improve the GVL effect.^34,35  TNF inhibition could also potentiate the GVL effect of donor lymphocyte infusions by blocking Treg effect with a much more simple and direct procedure. On the other hand, the alloreactivity could be strongly reduced to control GVHD by using TNFR2 agonist molecules in order to expand Tregs ex vivo.¹⁷  Indeed, it was recently observed that serum of recipient animals during acute GVHD induced Treg activation and enhanced their suppressive function through a TNF-dependent mechanism without inhibiting the GVL effect.³⁶  These 2 aspects remain to be confirmed both in experimental GVHD and in xeno-GVHD induced by human cells. This represents a potential therapeutic strategy to tailor the intensity of the alloreactive response to the particular need over the course of the patient disease.

The authors are indebted to Dan Avi Landau for critical reading of the manuscript.

M.L. was supported by a “poste d'accueil Inserm,” S.N. by the “région Ile de France, DIM Biothérapie,” and the département hospitalo-universitaire “Virus Immunité Cancer.” This work was supported by Association Française contre les Myopathies, Agence de la Biomédecine, Association de Recherche contre les hémopathies malignes, Fondation pour la recherche médicale (Victor et Erminia Mescle price) et Fondation Ramsay Générale de Santé (Cellular Therapy and Regenerative Medicine price).

Contribution: M.L., S.N., B.L.S., and J.L.C. conceived the study and wrote the manuscript; M.L., S.N., C. Pilon, C.D., Y.B., and F.C. performed experiments; and M.L., S.N., C. Pilon, A.T., G.H.M., C. Pouchy, F.C., S.M., B.L.S., and J.L.C. analyzed the data.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: José L. Cohen, Hôpital Henri Mondor, 53 Av du Mal de Lattre de Tassigny, 94010 Créteil, France; e-mail: jose.cohen@inserm.fr.