Tolerance induction to alloantigens remains a major challenge in transplant immunology. Progress in the last decade of our understanding of T-cell activation has led to the development of new immunotherapeutic strategies to replace conventional immunosuppression which inhibits the immune system in a nonspecific way. In particular, positive and negative costimulatory molecules of the CD28 family have been consistently demonstrated to be critical for the development of productive immune responses as well as the establishment and maintenance of peripheral tolerance. However, recent discoveries of novel costimulatory interactions confer a novel dimension to the immunoregulatory interactions within the B7:CD28 family and compels a revised view within a “quintet” of costimulatory molecules: CD28/B7/CTLA-4/PD-L1/ICOSL. Complexity introduced in this more detailed costimulatory pathway has important implications in therapeutic interventions against human immunological diseases and, especially, highlight the fundamental differences in selectively targeting CD28 molecules instead of B7 counterparts. In this review, we discuss these differences and emphasize different CD28-specific immunomodulating strategies evaluated in experimental models of transplantation and autoimmune diseases.
Although significant medical advances in these last decades have led to an important reduction in acute rejection rates and improvements in short-term allograft survival, these improvements have not resolved the problem of late morbidity due to immunological and nonimmunological causes (e.g. chronic rejection, calcineurin inhibitors (CNI) toxicity, noncompliance). Immunomodulation or the induction of regulatory mechanisms might complement these advances and improve long-term outcomes by replacing CNI and installing immune tolerance.
As T lymphocytes orchestrate immune responses after allotransplantion and autoimmunity, the aim of most immunosuppressive treatments is to prevent T-cell activation. T-cell activation is triggered by specific antigen recognition and reinforced by the engagement of costimulatory molecules that regulate T-cell differentiation into either pathogenic effector cells or anti-inflammatory regulatory cells. Although CD28 remains the most important positive costimulatory molecule for effector T cells (Teff) and the therapeutic potential of CD80/86 (B7) blockade has reached the market with Abatacept (Orencia®, Bristol-Myers Squibb, New York, NY, USA) in autoimmune disease and Belatacept (Nulogix®, Bristol-Myers Squibb) in kidney transplantation, the field of costimulation has greatly evolved since CD28-B7 and CTLA4-B7 interactions have been described, with heightened complexity on both sides of the equation.
The CD28/B7/CTLA-4/PD-L1/ICOSL Quintet of Costimulation
The CD28/B7/CTLA-4 pathway is classically depicted as a triad of costimulatory molecules where, constitutively expressed, CD28 binds B7 molecules to provide a costimulatory signal that is important for sustaining T-lymphocyte activation, proliferation and a pro-inflammatory response. CTLA-4, the other CD80/86 ligand which is inducible after activation, is a coinhibitory molecule which delivers antiproliferative signals to T cells, triggers indoleamine 2,3-dioxygenase (IDO) production in antigen-presenting cells (APCs) and is essential for the suppressive function of regulatory T cells (Tregs) and the induction of tolerance. However, recent significant discoveries have hugely complicated and revised this simple dichotomous vision. First, the group of Gordon Freeman revealed a specific bidirectional inhibitory interaction between CD80 and PD-L1 in mice (1) and humans (2). This unsuspected interaction is functionally significant with an affinity intermediate to that of CD80/CD28 and CD80/CTLA-4 and results in diminished expression of cell-surface activation markers, decreased T-cell proliferation and reduced cytokine production. Last year, another group revealed that ICOSL, so far known to interact only with ICOS, was also a ligand for CD28 and CTLA-4 in humans (3). ICOSL/CD28 interaction appeared essential for the costimulation of human T cells’ primary responses to allogeneic antigens and memory recall responses. However, whether ICOSL/CTLA-4 interaction in addition results in coinhibitory signals remains to be explored. Because this novel ICOSL reactivity has not been found in mice, this discovery also warrants careful re-evaluation of knowledge and translational interpretation of experimental models. Consequently, these newly described interactions confer a novel dimension to the immunoregulatory interactions within the B7:CD28 family and compels a revised view of this “quintet” CD28/B7/CTLA-4/PD-L1/ICOSL of costimulatory molecules (Figure 1), which have important implications in therapeutic interventions against human diseases.
In particular, we believe there is a widespread semantic confusion with the use of the “CD28 blockade” terminology when referring to bioreagents targeting B7 molecules. In fact, outcomes and consequences of targeting B7 instead of CD28 molecules are fundamentally different (for review, Ref. 4). Although antagonists of B7 and of CD28 both prevent the key B7/CD28 costimulatory interaction, B7 antagonists do inhibit CTLA-4 signals crucial to the function of Tregs and tolerance, the bidirectional CD80/PD-L1 coinhibitory interaction, but do not prevent the recently described CD28/ICOSL costimulatory interaction. In contrast, selectively targeting CD28 allows for the blockade of all costimulatory while preserving crucial coinhibitory interactions (Figure 1).
Antibodies directed against CD28 can be divided into two classes based on their target epitope and their stimulatory activity on T lymphocytes: first, “superagonistic anti-CD28 antibodies” refer to a nonphysiological engagement of CD28 by its basolateral domain resulting in a polyclonal activation of T lymphocytes even in the absence of TCR stimulation. Second, “conventional anti-CD28 antibodies” cross-link CD28 by any epitope lying outside the basolateral domain and costimulate T cells only in synergy with a TCR stimulation. Monovalent fragments from conventional anti-CD28 antibodies cannot costimulate T-cell activation even with TCR signals, presumably because they cannot multimerize CD28. In this review, we summarize works from both our lab and other labs, which have evaluated specific CD28 therapies in experimental models of transplantation and autoimmune diseases (Table 1).
Table 1. CD28-specific antibodies in experimental models
Does not induce human T-cell activation or cytokine release in vivo
Prevents GVHD by CTLA-4 dependent mechanisms
Superagonist Anti-CD28 Antibodies
Superagonist anti-CD28 antibodies present a unique capacity of inducing TCR-independent T-lymphocytes activation, proliferation and cytokine release (5). They were characterized to bind exclusively to the laterally exposed C’’D loop of CD28 and therefore do not antagonize CD28/B7 interaction (6), Figure 2. Because they were described to expand more efficiently natural regulatory T cells (Tregs) over Teff (7–11), several studies evaluated their potential to induce or restore immune tolerance in experimental models of allotransplantation or autoimmune diseases. In fact, they elicit two qualitatively distinct waves of T-cell activation: a first phase of Teff polyclonal activation, followed by a second phase of Treg expansion dependent on paracrine IL-2 secretion from stimulated Teff (12,13). This superagonist-induced Treg expansion was sufficient in rodents to prevent autoimmune diseases in numerous experimental models (13–20) and to protect from acute GVHD (21). In solid organ transplantation, superagonist anti-CD28 antibodies delayed acute rejection of heart (10) and liver (22) allograft, induced long-term survival of liver allograft in synergy with blood donor-specific transfusion (DST) and was sufficient in monotherapy to induce donor-specific tolerance in kidney allotransplantation (9). In rodents, lymphopenia induced at high doses was well tolerated and did not induce important pro-inflammatory cytokines release (12,15). In rodents, it was shown that Treg suppressed the inflammatory responses since, after Tregs elimination, CD28 superagonist led to significant pro-inflammatory cytokines release (13).
In preclinical studies in monkeys, which received doses of up to 500 times higher than humans, no toxicity was observed after injection of the humanized superagonist anti-CD28 antibody TGN1412 (23). Monkeys showed only a weak pro-inflammatory cytokine induction after injection and an anticipated expansion and activation of peripheral T lymphocytes. Affinity of this CD28 superagonist toward human and monkey CD28 receptors was found comparable (24). Furthermore, Fcγ receptor sequences demonstrated a high degree of similarity between these species and IgG4 binding to human and monkey Fcγ receptors was virtually the same (25).
However, the phase I clinical trial of the TGN1412 CD28 superagonist induced a rapid and massive cytokine storm that caused severe and life-threatening adverse effects (26). After this accident, a flood of articles tried to elucidate why such toxicity had not been observed in preclinical species. A study showed that TGN1412 induces calcium responses in human naive and memory CD4+ T lymphocytes but not in monkey T lymphocytes (27), which could be explained by a lack of CD28 expression by CD4 effector memory cells in macaques (28). Whereas the TGN1412-binding region is perfectly conserved between human and monkey, monkey CD28 sequences diverge from human CD28 in three transmembrane residues (24), which could alter CD28 association with molecular partners, rendering TGN1412-mediated signaling weaker in monkeys. Moreover, expression at high level of inhibitory sialic acid-recognizing Ig-superfamily lectins (Siglec, CD33) on lymphocytes was lost during human evolution, in particular Siglec-5 (29,30). In consequence, human T and B cells are more reactive than chimpanzee T cells to a wide variety of stimuli (anti-TCR Abs, L-phytohemagglutin, Staphylococcus aureus superantigen, CD28 superagonist and MLRs). This suggests therefore that TGN1412-induced cytokine storms could be due to an overall over-reactivity of human lymphocytes in comparison to monkeys.
Conventional Anti-CD28 Antibodies
Agonistic bivalent antibodies
The degree of transmission of the costimulatory signal by CD28 is directly correlated to its multimerization (Figure 2). CD28 being a homodimeric protein, like antibodies, the action of CD28 antibodies in their IgG form usually results in CD28 cross-linking and T-cell costimulation in synergy with TCR signals (31). This is only partially dependent on Fc engagement by Fc receptors since ``silenced'' divalent CD28 antibodies with a mutated Fc domain that fail to interact with Fc receptors still costimulate T cells (32). However, when divalent anti-CD28 antibodies also prevent CD28/B7 interaction, they become able in vivo to inhibit autoimmune diseases in experimental models of EAU (33) and type 1 diabetes when administrated during the neonatal period (34). Such antibodies demonstrated efficacy also in preventing GVHD, by depleting alloantigen-activated T cells via an IFN-γ dependent apoptosis mechanism (35,36) or increasing Tregs frequency in inflammatory settings (37). Nevertheless, divalent anti-CD28 antibodies presenting CD28/B7 inhibitory capacities also displayed agonist properties in vivo and induced an unregulated migration of memory T cells to extra-lymphoid tissue independently of TCR-derived signals and homing receptors (38).
Modulating bivalent antibodies
The mouse anti-rat CD28 monoclonal antibody JJ319 was one of the most studied anti-CD28 antibodies since, even if it costimulates T cells in vitro, it induces a down modulation of CD28 on the surface of T lymphocyte in vivo and therefore acts as a functional antagonist of the CD28/B7 interaction (39–41). As a consequence, JJ319 demonstrated efficacy in the prevention of type 1 diabetes (17) and acute GVHD (41) where it induced apoptosis of alloreactive T cells after few cycles of division. Results were most impressive in kidney allotransplantation, where a short-term treatment of JJ319 was sufficient to prevent acute and chronic rejection as well as induce reproducible donor-specific tolerance (42,43). In heart and liver allotransplantation, even if JJ319 delayed acute rejection, association with CNI, donor-specific blood transfusion or CD40-Ig treatments was required to achieve tolerance induction (22,39,40,44). Furthermore, almost all studies in transplant settings described induction/expansion of T and non-T regulatory cells such as CD4+ CD25+ Foxp3+ (42), CD4+ CD45RC– Foxp3– T lymphocytes (22), CD3– IDO+ cells (44) and CD3– B7+ INOS+ myeloid-derived suppressor cells (45).
Fc-silent bivalent antibodies
To avoid agonist properties of divalent anti-CD28 antibodies, Fc-mutated anti-CD28 antibodies were generated to prevent the cross-linking of CD28 through Fc/FcγR interactions (Figure 2). In a rat cardiac allograft model, a divalent anti-CD28 with Fc-silenced function and antagonist properties in vitro, dose dependently prolonged graft survival, synergized with a suboptimal treatment of CNI and suppressed alloantigen-initiated proximal TCR signaling events (46). However, a humanized version of Fc-silent anti-CD28 antibody was described to generate agonistic signals in vitro and promoted T-cell proliferation and cytokine release when stimulated with monocytes or endothelial cells expressing no or low levels of B7 molecules (32). In contrast, when T cells were stimulated with CD86-transfected monocytes, the same antibody inhibited both proliferation and cytokine induction, probably due to the engagement of CD86 with CTLA-4 on responding T cells. In vivo, administration of this antibody in humanized SCID mouse models reduced T-cell-mediated skin allograft rejection (32), as well as epidermis thinning and lymphocyte infiltrates of human psoriasis plaques transplanted to SCID mice (47).
Monovalent CD28 antagonist
Several years ago, we described that monovalent fragments (i.e. Fab or scFv) from a conventional anti-CD28 antibody which prevented CD28/B7 interaction could be used as a true antagonist of CD28 since it inhibited T-cells activation, proliferation and cytokines secretion without cross-linking the CD28 receptor (31). They could, furthermore, induce T-cell anergy and prevent cytotoxic T-lymphocyte generation in human mixed lymphocytes reactions (48). In vitro, specific CD28 antagonist monovalent antibodies prevented T-cell activation not only by preventing CD28 engagement with its ligands, but also by promoting CTLA-4 engagement (49) (Figure 2). In fact, CTLA-4 could inhibit the TCR-induced stop signal that otherwise allows T-cell immobilization (50). Therefore, in the presence of a monovalent CD28 antagonist antibody, human T lymphocytes stay motile and cannot form stable immunological synapses with APCs (49). In contrast, B7-mediated blockade (anti-B7 and CTLA4-Ig), which also prevents CD28-mediated costimulation, cannot inhibit cell arrest on APC and immunological synapses formation. This is of particular interest, because shorter T cell arrest times when encountering APC has been associated with tolerant states in vivo by biphotonic studies in rodents (51) whereas the formation of stable synapses and T-cell immobilization was associated with immune responses. Moreover, Treg suppression activity, which is now unanimously demonstrated to be CTLA-4-dependent but not CD28-dependent (52), is not affected (and is even reinforced in inflammatory settings) with selective CD28 targeting whereas B7-mediated blockade switches it off (49).
In vivo, monovalent CD28 antagonist prevented EAE in rodents both during initial Ag priming and after the onset of clinical signs of EAE (53), whereas B7-mediated reagents were not able to block ongoing disease, presumably by interfering with regulatory B7/CTLA-4 interaction. In heart allotransplantation in mice, monovalent CD28 antagonist delayed acute rejection and inhibited chronic rejection when combined with CNI or anti-CD40L antibody (54). Graft acceptance was associated with decreased alloantibody production, increased proportion of early graft infiltration by Tregs and increased expression of regulatory dendritic cell genes. Interestingly, coadministration of a blocking anti-CTLA-4 antibody led to prompt rejection in all animals and inhibited expression of Foxp3, PD-1 and IDO in the graft. In nonhuman primates, we recently reported that a monovalent CD28 antagonist delayed acute rejection when given as monotherapy and synergized with CNI to prevent acute and chronic allograft rejection, as well as alloantibody production, in kidney and heart transplant models (49). We observed acquisition of posttransplant donor-specific hyporesponsiveness that could be related to an increase of Tregs in the periphery and their accumulation in the allograft where molecular signatures of immune regulation (e.g. HO-1, IDO, TGF-β) were observed. Finally, we recently developed FR104, a novel humanized pegylated anti-CD28 Fab’ antibody fragment. This novel antagonist, which prevents human T-lymphocyte proliferation and activation, is devoid of agonist properties on human T cells in vitro, even in the presence of anti-CD3 antibodies or when cross-linked with secondary antibodies (Poirier et al., in press). To further investigate the immunological safety profile of CD28 compound on human T lymphocytes, we established a “trans vivo” immunotoxicity model in which human PBMC were challenged in vivo in NOD/SCID mice. After infusion into a lymphopenic host, human T lymphocytes acquired an activated/memory phenotype, which makes the model relevant to evaluate in vivo immunotoxicity of human CD28 antibodies because it was identified that effector memory T cells, rather than naïve T cells, were the predominant source of proinflammatory cytokine release under TGN1412 stimulation (28). In this model, injection of a commercial superagonist anti-CD28 antibody resulted in a rapid and significant increase of human proinflammatory cytokines release as well as increased the frequency of CD25 and CD69-positive T cells. In contrast, FR104 injection did not result in any cytokine release as compared to the control situation and even reduced release of IFNγ, probably as a result of the blockade of xenoreactivity, and did not elicit any activation marker on human T lymphocytes (Poirier et al., in press). Furthermore, the binding epitope on CD28 of our monovalent antagonist FR104 is composed of a conformational epitope formed by the CDEF loops of CD28 (49), different from the C″D epitope defined as mandatory binding site to obtain the “superagonist” effect of CD28 antibodies (6). For all these arguments, our feeling is that the monovalent anti-CD28 FR104 is a true antagonist of CD28, devoid of agonist activities on human T cells in vitro and in vivo. However, there is no doubt that only the first clinical evaluation will reveal whether this CD28 antagonist is safe in humans. Finally, with time, these humanized mice suffered from acute and severe xenogenic GVHD due to the activation of human T cells. Administration of FR104 completely prevented GVHD in a CTLA4-dependent manner, because the protection was abrogated by the administration of a CTLA-4 antagonist (Poirier et al., in press).
CD28/CTLA-4 and Tregs
Literature showed that selective CD28 blockade posttransplantation resulted in Treg accumulation in allografts in rodents and primates and was associated with tolerance. In contrast, CTLA4-Ig administration in humans decreases Tregs’ frequency in patients with rheumatoid arthritis (55) and in kidney transplant recipients (56), which could explain higher rejection rate in patient receiving high dose of Belatacept in clinical phase III (57). In rodents, a recent study reported that nTregs of thymus origin (Helios+) were decreased in periphery and allografts after CTLA4-Ig administration, leading to accelerated allograft rejection in a MHC class-II mismatch model, in which long-term allograft survival is dependent on Tregs (58). In fact, CD28 signaling has been demonstrated to be crucial for thymic generation, peripheral homeostasis and survival of nTregs (59,60). Therefore, the remaining question is how could selective CD28 blockade promote Tregs after transplantation? Recent data shed light on the question: Treg subsets could present different costimulation requirement. Highly suppressive ICOS+ Tregs appear independent of CD28 for expansion in mice and men whereas CD28 engagement would be needed for maintenance of ICOS-Tregs (61,62). Therefore, it is possible that selective CD28 blockade could indeed impact some but not all Tregs. In addition, B7/CD28 interaction has opposing roles in nTregs versus iTregs, because strong CD28 costimulation suppresses generation of iTregs (63–65). It is then conceivable that, conversely, blocking selectively CD28 costimulation could promote generation of iTregs, which is also promoted by signals mediated by CTLA-4 (66–71). Finally, the suppressive function of Tregs which is CTLA-4 dependent (72,73) but CD28 independent (52,72,74), would not be damped by selective CD28 blockade.
Costimulation blockade with CTLA4Ig has demonstrated its potential to replace CNI in kidney transplantation and control autoimmunity, demonstrating the central role of the B7/CD28/CTLA4 triad in controlling T-cell reactivity (75,76). However, because Treg function needs co-inhibitory signals transmitted to CTLA-4 by B7 molecules, inducing immune tolerance to allo- or autoantigens would require another strategy. A growing number of studies have suggested that antagonizing CD28 with appropriate antibodies can indeed result in a blockade of T-cell costimulatory signals while allowing for correct function of coinhibitory signals and Treg cells. In animal models, this translated into induction of Treg cells and tolerance. Nevertheless, some important questions remain pending: (1) Is peripheral immune tolerance as observed in laboratory animals desirable in man? In other words, what is the real importance of inducing Tregs in the physiopathology of transplant rejection/acceptance and autoimmunity in man? (2) What would be the effect of long-term CD28 blockade on natural Treg homeostasis and consequences of such effect? (3) Although our knowledge of T-cell physiology has increased, it is clear that new costimulatory interactions are still being uncovered which means a thorough comprehension of the consequences of costimulation blockade cannot yet be claimed. In addition, the recent discovery of significant differences between mice and humans in the function of costimulatory and co-inhibitory molecules as well as between humans and other primates complicates preclinical extrapolations and suggests that a clinical evaluation only will reveal whether the CD28-targeted strategy holds its promises.
Work on CD28 costimulation blockade at INSERM UMR-S 1064 and at Effimune is funded by the “Tolerance Restoration In Autoimmune Disease” FP7 Health 2011 Programme #EC-GA n°281493 of the European Commission, by the French Agence Nationale de la Recharche ``TOLESTIM'' grant # ANR- 09-BIOT-013-02 and by the Progreffe Fundation (Nantes, France).
The authors of this manuscript have conflicts of interest to disclose as described by the American Journal of Transplantation. B.V. and N.P. are shareholders and employees in Effimune, a company developing CD28 antagonists.