CD137 (4-1BB)-Based Cancer Immunotherapy on Its 25th Anniversary

The group of Byoung Kwon genetically searched for molecules expressed by activated but not resting T lymphocytes (1). A cDNA encoding a transmembrane protein was named 4-1BB (2). Cloning of the mouse and human homologs permitted the generation of monoclonal antibodies (mAb) that eventually led to the discovery of the protein (3) and its designation as CD137, becoming a new member of the tumor necrosis factor receptor (TNFR) gene family, termed TNFRSF9. The antibodies were functionally able to enhance proliferation and cytokine production while mitigating cell death in T-lymphocyte cultures (3, 4). The group of Robert Mittler, based at Bristol Myers Squibb, was able to show that the artificial in vivo costimulation of CD8+ T cells with agonist anti-CD137 mAb was able to enhance CD8+ T-cell responses while attenuating humoral responses through a CD4 T-cell–dependent mechanism (5).

Twenty-five years ago, we reported that an infusion of anti-CD137 agonist mAbs was able to eradicate a number of established transplantable mouse tumors including those derived from the Ag104 and P815 cell lines (6). Antitumor effects in every case required the presence in the mouse of CD8+ T cells, while the requirements for natural killer (NK) and CD4 T cells were tumor model–dependent (6–8).

In the meantime, scientists at the Immunex company were able to clone a TNF family natural ligand for CD137, termed CD137L (4-1BBL, TNFSF9), which also exerted costimulatory effects on T lymphocytes as a trimeric molecule (9). In this regard, we showed in mouse models that murine tumor cell transfectants expressing CD137L enhanced CD8 antitumor immune responses toward themselves and against untransfected concomitant counterparts in a manner that was enhanced if the fellow costimulation ligand CD80 was cotransfected to stimulate CD28 (10). Such results were improved if instead of the CD137L moiety, anti-CD137 mAb single-chain variable fragments (scFv) encompassing a transmembrane domain were transfected to transplantable tumor cell lines (11). Importantly, such effects could be recapitulated against established tumors by intratumoral injections of an adenovirus encoding such engineered surface scFv antibodies (12) or 4-1BBL (13, 14). These experiments were an application of the previously described costimulation with adenovirally transferred 4-1BBL of antiviral CD8+ T-cell responses (15).

These discoveries were coincidental in time with reports that CTLA-4 blockade with antagonist antibodies was able to eradicate a particular transplantable colon cancer (16). In this regard, to the best of our knowledge, the therapeutic use of CD137 mAb opened up the field of agonist immunomodulatory mAbs (17). Figure 1 shows a timeline of the key milestones regarding these and subsequent preclinical and clinical developments.

Of utmost importance is the fact that CD137 expression is not constitutive on T cells but is induced following T-cell receptor (TCR)/CD3 activation on antigen-primed T lymphocytes. It can also be expressed by other activated leukocytes including activated NK cells, where it is stimulatory (8, 18), B cells (19), and other myeloid leukocytes including dendritic cells (20–22), where its functions remain elusive for the most part. As a result of being inducible, it can be used as a marker to identify and select T cells recently involved in antigen recognition (23, 24). Of note, costimulation by CD28 (10) and CD137 itself (25) notably enhances the levels of CD137 expression, as does reportedly hypoxia (26). Following massive CD137 cross-linking, internalization of the moiety occurs to endosomal compartments from where it may keep signaling (27). Consequences of internalization of CD137 are underexplored and might be important to optimize stimulation schedules, because chronic stimulation may desensitize the pathway, therefore favoring intermittent exposure to achieve maximal effects.

CD137 belongs to the TNFR subfamily that is devoid of cytoplasmic death domains (28) and therefore does not directly engage with the caspase activation machinery. The molecule is composed of four extracellular cysteine-rich domains (CRD), a transmembrane region, and a 42-aa-long cytoplasmic domain bereft of any known intrinsic enzymatic activity. As in the case of all other TNFR members, the activity of CD137 relies on the interaction with members of a family of adapter proteins termed TRAFs (TNF receptor–associated factors; ref. 29). The first pieces of evidence for TRAF2 interaction with CD137 were based on a two-hybrid methodology (30) and coimmunoprecipitation (31). CD137 interaction with TRAF2 and TRAF1, as well as TRAF3 and TRAF5, has been reported (29–33). Indeed, using high-resolution proteomics, we have recently demonstrated the presence of those four adapters in CD137 immunoprecipitates. In addition to these, cIAP1 and cIAP2 and the LUBAC members HOIP and HOIL-1L were coprecipitated (J. Glez-Vaz; submitted for publication).

Costimulatory signaling via 4-1BB is largely contingent upon cross-linking by ligand at least to trimerize or form larger order of magnitude lattice-shaped structures (29). Following cross-linking, K63 polyubiquitination reactions take place (27) that originally were thought to involve the E3 RING domain of TRAF2. However, this catalytic domain reportedly is structurally unable to mediate ubiquitination reactions (34), and other E3 ubiquitin ligases must be involved including cIAPs, which have been shown to avidly associate with TRAF2 (35) also in the 4-1BB signalosome (J. Glez-Vaz; submitted for publication). Of note, a dominant-negative cIAP transgene in mice abrogated T-cell costimulation by 4-1BB mAbs (36), and we now have evidence that cIAPs are required for downstream signaling and for the antitumor effects (J. Glez-Vaz; submitted for publication). McPherson and colleagues (37) had previously shown that knockdown of cIAP2 in cIAP1 knockout T cells abrogated 4-1BB–dependent NF-kB activation, indicating that either cIAP1 or cIAP2 was required for NF-kB1 activation. These authors also showed that 4-1BB signaling causes degradation of TRAF3 leading to activation of the NF-kB2 pathway that was delayed with respect to activation of the classical NF-kB pathway.

The system seems to be highly controlled by deubiquitinases such as A20 and CYLD that constitutively associate with the 4-1BB signalosome complex in which various substrates are ubiquitinated including TRAF2 (38).

The biochemical function of TRAF1, which is prominently induced by NF-κB activation, remains elusive, but reportedly together with TRAF2, it recruits cIAPs with more affinity to the signaling complex (35). TRAF1 has also been functionally linked to T-cell apoptosis inhibition through Bim downregulation (39). Once early ubiquitination takes place, docking sites appear for TAB1/2–TAK1, and subsequently the IKK complex and various MAPKs become activated. K63 polyubiquitination of IKKγ (NEMO) is also probably a key event in signaling via NF-κB1 (40). In any case, TRAF1 seems to enhance classical NF-kB1 activation. Of note, TRAF1 itself is a strong transcriptional target of NF-kB, suggesting an interesting positive feedback mechanism in this signaling route as a result of the inducibility of TRAF1 expression (41).

Remarkably, CD137 also activates noncanonical NF-κB through elusive mechanisms that reportedly involve TRAF1 (37) in a yet poorly understood fashion. In this regard, the presence of TRAF3 in the 4-1BB signalosome suggests that active 4-1BB could sequester TRAF3, thereby preventing the targeting of NIK for proteasomal degradation through K48 polyubiquitination. Signaling via the alternative NF-κB pathway (NF-κB2) may be a differential feature of CD137 as compared with other costimulatory TNFR family members (42). This is important because most of these signaling mechanisms are shared by other T cell–expressed TNFR family members, and it remains to be seen what the essential qualitative or quantitative differences among them are.

There are soluble CD137 (sCD137) forms dependent on variants splicing out the transmembrane domain (43) or on protein shedding by metalloprotease cleavage (44). Soluble forms resulting from T-cell activation can competitively block CD137L to tone down costimulatory functions as a negative feedback mechanism (45). The expression of the CD137 ligand seems to be restricted to activated dendritic cells, B cells, and macrophages (46, 47). Reverse signaling via CD137L has been studied and seems to be important at regulating myeloid leukocyte biology (48). CD137 molecular physiology and signaling are summarized in Fig. 2.

Almost since its discovery, CD137 was found to be involved in fostering a number of functions in T cells. Once ligated, it increases T-cell proliferation, enhances cytokine production (3), reduces lymphocyte cell death (39), enhances memory differentiation (49), and may reverse anergy/exhaustion (50). CD137 ligation triggers transcriptomic changes that underlie these phenotypic observations with additional layers of complexity because CD137-induced cytokines may potentially contribute to these phenotypic changes in an autocrine/paracrine manner. In addition to transcriptional effects, experimental evidence indicated that CD137 elicits epigenetic chromatin reprogramming that controls DNA methylation (51) and histone-3 methylation/acetylation changes (52). CD137 costimulation potently reprograms metabolism fostering mitochondrial mass and function (53, 54). The costimulation via CD137 is not restricted to TCRαβ T lymphocytes because it is also functionally expressed by TCRγδ counterparts (55, 56).

Although somewhat of a surprise, CD137 or CD137L deficiency in mice does not drive a severe immunodeficient phenotype because such mice are mostly normal with only relatively mild defects in the setting of more severe model viral infections such as influenza or lymphocytic choriomeningitis virus (57–59) for which CTL responses are critical to clear the infections. One of the reasons for these mild phenotypes is probably related to the overlapping function of other costimulatory TNFR molecules such as OX40, CD27, and GITR. In humans, a few cases suffering homozygous loss-of-function mutations of CD137 have been associated with severe Epstein–Barr virus infections including lymphoproliferative syndromes (60–62).

It comes as a surprise that the same agonist mAbs that were able to elicit tumor regressions also improve the outcomes of autoimmune disease models in mice driven by autoreactive CD4 T cells such as experimental autoimmune encephalomyelitis (63), collagen-induced arthritis (64, 65), or lupus (66). The reasons for this are far from clear but were traced to the fostering of activation-induced cell death of such pathogenic CD4⁺ T cells (63, 65). In contrast, agonist anti-CD137 mAbs were shown to exacerbate CD8-mediated diabetes in NOD mice (67, 68) and caused discrete levels of T-cell infiltrates in the liver causing mild periportal hepatitis in the treated mice (69, 70). Of note, the pathogenesis in these models is mainly mediated by CD8⁺ T cells.

CD4⁺ Foxp3⁺ regulatory T cells (Treg) also acquire CD137 expression that is very prominent on the Tregs present in the tumor microenvironment (TME; ref. 71). Conspicuously, the result of CD137 ligation on Tregs seems not to increase or be detrimental for their immunosuppressive functions but apparently may contribute to their expansion (72, 73). Interestingly, in mouse models, CD137 can be used as a target to deplete Tregs from the TME with properly engineered mAbs (74).

Although costimulation of T-cell responses by CD137 mAbs in general requires antigen–TCR engagement, the effect of CD137 mAb on memory T cells can be antigen-independent as shown by proliferation and activation of memory T cells in mice without exposure to a specific antigen (75). Because memory T cells accumulate in the liver and bone marrow, it is possible that this antigen-independent CD8⁺ T-cell activation may be partially responsible for CD137 mAb–mediated side effects in theses organs.

The inhibition of T-cell apoptosis is perhaps the most prominent and relevant effect of CD137 costimulation. This has been ascribed to the NF-κB activation of Bcl-xL and Bfl-1 (76), as well as to downregulation of Bim (39). The inducible nature of the expression of 4-1BB and 4-1BBL mediating this prominent antiapoptosis role prompted T.H. Watts to propose 4-1BB signaling as a signal-4 in T-cell activation to distinguish it from constitutively operative CD28 costimulation (77). It was also important to establish that 4-1BB costimulation is actually independent from CD28, as shown in CD28 knockout mice (78).

NK cells are also able to prominently express CD137 on their plasma membrane upon activation (8). Other investigators have reported that CD137 ligation on NK cells leads to costimulation of cytokine secretion without increasing cytolytic activity at the effector phase including antibody-dependent cellular cytotoxicity (79) but strengthening their cytolytic machinery and making them resistant to TGFβ inhibition (80).

CD137 putative functional expression on other cell lineages has been documented. On endothelial cells in the tumor microvasculature in which it is expressed in response to hypoxia, its ligation promotes T-cell extravasation via ICAM-1 and VCAM upregulation (81). Expression and ligation on adipocytes mediate subtle metabolic consequences (82) and a role for CD137–CD137L interactions has been reported in the progression of atherosclerosis (83).

From the early discoveries of the antitumor effects exerted by CD137 agonistic mAbs, multiple attempts to enhance efficacy upon combination with other modalities of cancer therapy have been made (84, 85). For instance, chemotherapy can enhance efficacy against transplantable tumor models (86–88). This could result from the immunomodulatory effects of chemotherapy or because of the generation of more tumor antigen cross-presentation from dying tumor cells (89). Notably, cross-priming by cDC1 of CD8⁺ T cells is needed for anti–4-1BB mAbs to work (90, 91). Radiotherapy also synergizes with 4-1BB, possibly through similar mechanisms (92–95). Even surgery in neoadjuvant 4-1BB treatment settings synergized to prevent metastatic relapses (96).

A variety of immunotherapy strategies potently synergize with 4-1BB stimulation to enhance antitumor immunity. Among the most prominently efficacious in mouse tumor models is perhaps the combination with anti–PD-1/PD-L1 (97–99) and local delivery of IL12 (14, 100–102). Other combinations have also been shown to be synergistic including the use of an anti–CTLA-4 mAb (103) and the codepletion of Tregs (104). As a mechanism of mutual synergy, it is worth mentioning that CD137 costimulation fosters IFNγ secretion from T cells, and this cytokine in turn is arguably the main inducer of PD-L1 expression in the TME (105). Experimental cancer vaccines have also been shown to synergize with anti-CD137 mAbs (106–110). Other immunostimulatory mAbs may also act in synergy with agonist anti-CD137 mAb, as has been observed when targeting LAG3 (50), OX40 (110, 111), and CD40 (112). Reported synergies of CD137 agonists in preclinical and clinical research are summarized in Fig. 3.

All things considered, arguably the most efficacious synergy of CD137 costimulation is probably established with adoptive T-cell therapies (113). Evidence for synergy includes TCR-transgenic T cells (111) and ex vivo cultures of tumor-infiltrating T lymphocytes (114). With regard to tumor-infiltrating lymphocytes (TIL), anti-CD137 mAb also helps at the preinfusion stages to select those T cells with tumor reactivity and to enhance the yield of the cultures (115, 116). In this regard, CD137 costimulation has been successfully exploited to expand NK-cell cultures for infusion purposes in clinical trials (117, 118). This is consistent with observations in which the presence of 4-1BBL in artificial antigen-presenting cells had a dramatic effect to induce T-cell expansion ex vivo in culture (119).

Adoptive T-cell therapy using the gene transfer of chimeric antigen receptors (CAR) against CD19 and BCMA has revolutionized the treatment of B-cell malignances (120). The group of Carl June introduced the 4-1BB signaling domain as part of the intracellular sequence of CARs, and showed that in CD8⁺ T cells it leads to longer persistence upon infusion as compared with those encompassing the CD28 signaling domain (121). Alternatively, the group of Michel Sadelain showed that 4-1BBL + CD80 gene transfer to the T cells cotransferred with CD28ζ CARs also gained much efficacy against prostate cancer and lymphoma xenografted to mice (122).

In the clinic, at least three 4-1BB containing CARs have received approval for commercialization in Western countries due to unprecedented efficacy against refractory cases of acute lymphoblastic leukemia, B-cell lymphomas, and multiple myeloma (120, 123). Reportedly, in CD4 T cells, the costimulation domain of ICOS surpasses the efficacy of CD137 (124), but in CD8⁺ T cells, CD137 seems more effective and perhaps leads to less severe cytokine release syndrome in the clinic as compared with CD28-containing CARs (125, 126). Recent efficacy evidence has been reported for allogeneic NK cells transduced with CARs following expansion in culture driven by cytokines and CD137 ligation. Such CARs expressed in NK cells also encompass the CD137 cytoplasmic domain for costimulation in addition to CD3ζ (127).

Chimeric membrane molecules other than conventional CARs can be envisaged using synthetic biology to convey CD137 costimulation. Among these, an extracellular FAS/intracellular 4-1BB chimera has shown remarkable potential in mice (128), both because of conferring 4-1BB costimulation and being a dominant-negative decoy ligand for FASL (129).

In summary, ways to confer CD137 costimulation are powerful allies of adoptive T-cell therapy. This could be especially important as adoptive T-cell therapy faces the formidable challenge of providing efficacy against solid malignances. In that regard, CD137 could be crucial due to its ability to prevent or reverse anergy (50, 130) in addition to promoting survival and longer persistence. Enhanced mitochondrial/metabolic fitness (131) and the NF-κB2 pathway (132) are probably key differential superiority elements in favor of CD137. Perhaps as a consequence of these mechanisms, CD137 tonic signaling from the CARs seems to be beneficial rather than detrimental for CAR T performance and has been associated with a unique transcriptional profile (130, 133). These reasons are probably behind the superiority of 4-1BB–based CARs over CD28 ones in terms of persistence and safety.

The Bristol Myers Squibb company led the initial development of a fully human IgG4 clinical-grade agonist anti-CD137 mAb (eventually termed urelumab; ref. 134). The mAb was safe in non-human primates but as was noted later, the affinity of urelumab for macaque CD137 was considerably lower as compared with the human protein. A conventional 3 + 3 dose-escalation phase I clinical trial as monotherapy was conducted in which dose-limiting toxicities were not encountered. Evidence of objective activity in cases of melanoma was observed, as well as evidence for pharmacodynamic activity, although transient elevations of transaminases were reported in some of these dose-scalation patients (135).

Phase II clinical trials were undertaken using every 3 week doses of 5 mg/kg, and it was observed that around 20% of patients developed grade ≥ 3 abnormalities in liver function tests, with most of them experiencing milder elevations in circulating transaminases (135). Indeed, two fatalities occurred. These serious drug-related adverse events, although apparently idiosyncratic and limited to a subset of patients, put the urelumab program on hold.

Liver toxicity could be modeled in mice, in which polyclonal CD8⁺ T-cell accumulations in the liver parenchyma were found following repeated treatment with agonist anti-mouse CD137 (69, 136). Fc receptors (137) and proinflammatory cytokines such as IFNγ and IL27 were also shown to be critical, perhaps involving myeloid cells in the liver (138). The exact mechanism of liver toxicity remains elusive, although baseline expression of 4-1BB has been substantiated on memory T cells in the bone marrow and in the liver (139). Intriguingly, liver inflammation was abrogated in GITR knockout mice (139). It is tempting to speculate that once the pathway is stimulated, it results in more 4-1BB expression on activated CD8⁺ T cells as a vicious circle.

In a parallel effort, Pfizer developed utomilumab, a fully human anti-CD137 mAb of the IgG2 isotype (140). In this case, the agonist activity on the receptor is moderate and is fully cross-reactive with macaque CD137. Following uneventful toxicology studies, dose-escalation clinical trials were performed with excellent tolerability and no dose-limiting toxicities (141). However, clinical activity as a single agent was modest, although at least two cases of Merkel carcinoma of the skin responded to treatment (141). A recent report confirmed marginal clinical activity of utomilumab for immune checkpoint inhibitor–experienced melanoma and non–small cell lung cancer patients with evidence for frequent induction (46.3%) of antidrug antibodies (142).

Of important note, urelumab is an IgG4 that keeps avidity for FcR-I, whereas utomilumab is an IgG2 with limited binding to Fc receptors (143). Conceivably, the agonist activity of these antibodies in vivo is contingent on Fc receptor–mediated cross-linking. However, in the case of urelumab, some level of costimulation is observed even without an apparent need for cross-linking.

Urelumab reentered clinical trials to document the safe doses that could be tolerable in terms of liver side effects, and a phase I study declared 0.1 mg/kg, 0.3 mg/kg, and a flat dose of 8 mg as safe (144). Skin reactions and neutropenia were also mitigated by dose reduction. However, at these doses, objective responses were restricted to 13% of 31 heavily pretreated non-Hodgkin B-lymphoma cases.

As mentioned above, preclinical research in mouse models demonstrated that combinations of CD137 agonists and PD-(L)1 checkpoint inhibitors were synergistic. In light of such results, the combinations of urelumab + nivolumab and utomilumab + pembrolizumab have been tested in pilot single-arm trials. In the case of urelumab, patients with melanoma responded to the combination [overall response rate (ORR) = 39%], but evidence of the contribution of low-dose urelumab in these patients with checkpoint inhibitor–naïve melanoma was not conclusive (145). In the case of utomilumab + pembrolizumab, several interesting objective responses were documented, including cases of complete response in one patient with anaplastic thyroid cancer and in another with small cell lung cancer (146). In a recent report, utomilumab in combination with rituximab demonstrated promising clinical activity (ORR = 21.2%) in patients with relapsed/refractory follicular lymphoma and other CD20⁺ non-Hodgkin lymphomas (147).

In a very interesting combination approach, the group of Elizabeth Jaffee has tested urelumab (8 mg doses) in combination with nivolumab and a GVAX vaccine in a neoadjuvant + adjuvant randomized clinical trial for resectable pancreatic cancer. Urelumab was found to enhance tumor CD8⁺ T-cell infiltration in the resected tumor tissue, with promising clinical outcomes of relapse-free survival (148).

However, the liver safety issues and lack of evidence of consistent clinical monotherapy activity deprioritized development of urelumab and utomilumab in such a manner that currently there are no active clinical trials with these agents. Nevertheless, several important conclusions can be drawn from these early-phase clinical experiences: (i) These agonist antibodies are pharmacodynamically active and can achieve certain clinical activity in monotherapy; (ii) liver inflammation is a key limiting factor to be overcome for this class of agonist agents; (iii) other less serious safety issues may appear in terms of leukopenia, as previously reported in mice (135, 149), and skin toxicity that are probably on-target; (iv) use of CD137 agonists and PD-(L)1 blockade is most likely feasible in combination regimens with no evidence for serious accumulative toxicity, and there is a rationale for therapeutic synergy of these combined regimens.

Biotechnology for therapeutic mAbs and recombinant protein constructs has advanced a great deal over the last 25 years. Several strategies have been followed to construct agents with multiple specificities and activities built into a single molecule (150). The main objective here is to attain meaningful agonist activity on CD137 while making it tolerable for the patient when given either as monotherapy or in combination with other treatments. As previously discussed, the main feature to exert agonistic activity on CD137 is to induce cross-linking by multimeric ligands or by means of attaching the ligand to a solid phase such as a juxtaposed plasma membrane. Hence, designing and making constructs that would only or selectively cause cross-linking in the TME and/or in tumor-draining lymph nodes is the goal of this strategy. Recent evidence supports the superiority of cross-linking CD137 from the plasma membrane of a cell that is presenting cognate antigen or is engaging CD3–TCR by other means such as a T-cell engager bispecific antibody (42). The superiority of the so-called “in cis” costimulation over “in trans” costimulation, when TCR–CD3 engagement occurs from different cells, speaks to some degree of receptor cross-talk between CD3 and CD137. The molecular underpinnings of such cross-talk are being clarified. Having said this, even if less molecularly active, intense trans costimulation might be as efficacious for cancer treatment.

Three main strategies are being followed: (i) bispecific antibodies that bind to tumor selectively expressed surface antigens; (ii) bispecific antibodies that bind to immune receptors or checkpoint ligands expressed in the TME; and (iii) conditional antibodies that become active only in the TME. Figure 4 summarizes the different alternatives that are being developed, and several of these agents have already entered early-phase clinical development. Ongoing and recently completed clinical trials with this new wave of CD137 agonists are included in Table 1.

The formats of the multispecific compounds are variable and are represented in Fig. 4 for each agent. Conventional antibody chemistry still seems to be the most popular way, involving variations of antibody engineering both on constant regions and on the avidity of antigen-binding sites. However, other chemistries and constructs are used including nanobodies and calin- and collagen-based proteins. Both antibodies and trimeric forms of the natural ligand are under development. Details regarding these important aspects may account for different pharmacodynamics and tissue penetration features whose technical description is beyond this review (151).

This strategy is being pursued with several targets such as HER2 (152), EGFR (153, 154), 5T4 (42), PSMA (155), Glypican-3, Nectin-4 (156), Mesothelin (157), and B7-H3 (158), which are preferentially overexpressed by tumor cells. The principle of selective cross-linking and T-cell costimulation works with all these constructs in T-cell cocultures with the proper target-expressing cell lines. For hematologic malignancies, agents targeting CD137 agonists have been redirected to CD30 (159) and to CD19 (160). In these cases, the activity of the compounds is contingent on the expression of the different molecules on malignant cells. This feature is frequently disease-specific and variable even within a given malignancy. The key models to preclinically test these compounds are CD34- or peripheral blood mononuclear cell–humanized mice (161, 162) and human CD137 knock-in mice (163).

In the clinic, the only available clinical results are with HER2 × 4-1BB bispecific antibody (PRS-343; refs. 152, 164). The treatment as monotherapy is remarkably safe, signs of CD8⁺ T-cell functional stimulation have been found, and preliminary clinical activity has been reported (ORR = 40% at doses over 8 mg/kg) in HER2 cancer patients with trastuzumab-refractory disease (164).

Fibroblast-activated protein (FAP) is a transmembrane protein selectively and intensely expressed on tumor-associated fibroblasts in many solid malignancies. Targeting compounds to FAP could permit selective homing to tumors in mice and humans. Cross-linking of 4-1BB at this level provides significant costimulation to T cells and shows efficacy in mouse models (165). Two compounds are being developed in clinical trials based on these concepts: a FAP-binding antibody × trimeric CD137L (165, 166) developed by Roche and a more classic bispecific FAP–CD137 antibody that is being developed by Boehringer Ingelheim. In these cases, costimulation is provided in trans with respect to tumor antigens, but at least with T-cell engagers, these agents provide meaningful costimulation in vivo (165, 167).

The early-phase CD137L–FAP developed by Roche (RG7826 or RO7122290) has shown an excellent safety profile in monotherapy or when combined with PD-L1 blockade using atezolizumab. It is pharmacodynamically active, as it induces peripheral blood T-cell activation and proliferation as well as accumulation of activated CD8⁺ T cells in the TME (168, 169). There are signs of modest clinical activity in monotherapy and more promising evidence in combination with atezolizumab, but because responding patients were not previously refractory to anti–PD-(L)1 therapy, further evidence for clinical activity is needed. An ongoing clinical trial is testing the combination of a T-cell engager cotargeting CEACAM5 (CEA) and CD3 (cibisatamab) together with RG7826 based on interesting preclinical data in an attempt to provide both signal 1 and signal 2, thereby functionally mimicking CARs (165, 167).

Given the preclinical synergy of CD137 agonists with PD-(L)1 blockade, various PD-L1 × CD137 combinations are being developed. The idea is to use PD-L1 both to target and agonistically cross-link CD137 in the TME while simultaneously blocking PD-L1 binding to PD-1 (170–173). The mouse surrogate of this idea shows evidence for immunologic and therapeutic activities (174–176), albeit the optimal doses for blockade and agonist activity are likely to be different. Several formats for such bispecific antibodies have been generated (Fig. 4).

In the clinic, results are presently available only for a molecule developed by Genmab (GEN1046). This agent has shown clear clinical signs of T-cell activation in patients. Regarding safety, around 20% of patients developed controllable abnormalities in liver function tests. Importantly, there is evidence for clinical activity in monotherapy against tumors previously having been determined to be refractory to anti–PD-(L)1 therapy, as well as clear signs of pharmacodynamic activity (170).

For hematologic malignancies, a trimeric CD137L binding to CD19 has been developed by Roche (RG6076 or RO7227166) to treat refractory B-cell malignancies. Preliminary reported evidence speaks of safety and frequent objective responses upon combination with a CD20 T-cell engager (177) in patients pretreated with obinutuzumab (an afucosylated anti-CD20 mAb) to deplete endogenous B cells and cytoreduce as much as possible the number of remaining malignant cells. A bispecific antibody binding to CD30 and CD137 is also being developed for hematologic malignancies, chiefly including cases of refractory Hodgkin lymphoma (159).

Interestingly, bispecific constructs can also exploit awakening the function of two immune receptors involved in triggering antitumor immunity. In this regard, it is worth considering the efforts using a CD40 × CD137 bispecific (GEN1042; ref. 178) already in the clinic (179). This bispecific antibody seems to mediate CD40 activation/maturation of dendritic cells while selectively cross-linking CD137 on T cells in contact with such dendritic cells in a bidirectional fashion (Fig. 4). This concept of cross-linking from dendritic cells that are cross-presenting tumor antigens is of much interest and presumably can also be achieved by PD-L1 × CD137 bispecifics because there is expression of PD-L1 on dendritic cells (C. Luri-Rey; manuscript in preparation).

To attain selective activity in the TME and tumor-draining lymph nodes, probody technology has been applied. This consists of masked antibodies in which the antigen-binding site is not functional because of a masking cognate peptide. Such a peptide is tethered to the light chain of the antibody by a protease-sensitive linker that is cleaved by proteases selectively present in tumors. This strategy in mouse models with an anti-CD137 probody increases liver safety and synergizes with other treatments including PD-1 blockade and adoptive T-cell therapy (180).

Benefiting from the fact that the interstitial fluid of tumors is very rich in ATP concentrations, a conditional antibody binding and cross-linking CD137 only in ATP-rich media has been generated and has shown remarkable safety and efficacy in human CD137 knock-in mouse models (181, 182). Such a compound is progressing toward clinical development.

It is becoming increasingly clear that the costimulatory activity of CD137 is of great interest to enhance the effects of tumor immunotherapy, and the maximal effect is very likely to be found deploying the new agents in synergistic combination therapy schemes. The obvious combination with anti–PD-(L)1 therapy is ongoing in the clinic, but combinatorial approaches with adoptive T-cell therapy, cytokines, and chemotherapy should be clinically explored. Even in the case of PD-L1 × CD137 bispecific constructs, the combination with an additional antibody blocking PD-1 therapy is clearly worth considering given the fact that the optimal doses for CD137 stimulation might be lower than those for complete PD-L1 blockade. We have to bear in mind that to permit cross-linking of the two targets by a bispecific construct, optimal concentrations must be found, because an excess of the compound leads to saturation of each target with no opportunity for cross-linking (17). This is consistent with clinical findings in the sense that bimodal dose–response curves are observed for immune pharmacodynamic effects and efficacy.

Much of the success will depend on the discovery and validation of predictive biomarkers that are to be thoroughly investigated in clinical pretreatment and on-treatment samples. From the pharmacodynamic point of view, early evidence of T-cell activation is likely to be predictive of disease outcomes. Recent evidence indicates that CD137, either soluble or membrane bound, provides an accurate reflection of functional CD137 ligation (25). In peripheral blood, concentrations of IFNγ and CXCL10 as well as proliferating Ki-67⁺ CD8⁺ T cells are useful pharmacodynamic biomarkers in addition to T-cell infiltration and activation in on-treatment tumor biopsies (168, 170). Quantitatively significant expression of the targets for the bispecific compounds in the TME is likely to be a crucial condition to attain meaningful therapeutic effects.

The most relevant scientific questions in the field are: (i) to precisely decipher the molecular and cellular requirements to attain antitumor efficacy; (ii) to understand the role of 4-1BB costimulation in T-cell memory, dysfunction, and exhaustion; (iii) to understand and overcome CD137-related liver inflammation; (iv) to preclinically and clinically define the most synergistic partner treatments for combinations; (v) to extend the benefit of CD137-based immunotherapy to nonmalignant diseases; and (vi) to design and implement late-phase clinical trials to document benefit over standard of care.

Overall, CD137-based approaches hold much hope for cancer immunotherapy beyond their successful application in CAR T cells. We certainly expect that consistent efficacy in patients will be reached soon. Clinical science and pharmaceutical development are often unpredictable, but the magnitude of the current academic and industrial effort in CD137-based immunotherapy and the preliminary reports on the new wave of CD137 agonist compounds speak to a bright future, which will be harvesting the fruits of 25 years of continuous translational research efforts.

I. Melero reports grants and personal fees from Bristol Myers Squibb, Genmab, and Roche and personal fees from Alligator, F-Star, Merus, Pieris, and Boston Therapeutics during the conduct of the study, as well as grants and personal fees from Bristol Myers Squibb, Roche, AstraZeneca, Genmab, and PharmaMar and personal fees from F-Star, Gossamer, Alligator, Hotspot, Biolinerx, Bright Peak, Third Rock, Boston Therapeutics, Pieris, Servier, and Pierre Fabre outside the submitted work. M.F. Sanmamed reports grants from Bristol Myers Squibb and Roche during the conduct of the study, as well as grants and personal fees from Roche and Bristol Myers Squibb, and personal fees from Numab outside the submitted work. No disclosures were reported by the other authors.

I. Melero has been supported by grants PID2020-112892RB funded by MICIN/AEI/10.13039/501100011033 and SAF2017-83267-C2-1-R funded by MICIN/AEI/10.13039/501100011033 and by FEDER “Una manera de hacer Europa,” the Fundación de la Asociación Española Contra el Cáncer (AECC; HR21–00083), the Fundación La Caixa and Fundación BBVA, the Instituto de Salud Carlos III (PI20/00002 and PI19/01128), cofinanced by the Fondos FEDER “A way to make Europe” and Joint Translational Call for Proposals 2015 (JTC 2015), TRANSCAN456 2 (code TRS-2016-00000371), and the Gobierno de Navarra Proyecto LINTERNA (reference 0011-1411-2020-000075). M.F. Sanmamed has received a Miguel Servet I (MS17/00196) contract from the Instituto de Salud Carlos III cofinanced by the Fondo Social Europeo “Investing in your future” and a grant from Instituto de Salud Carlos III, Fondo de Investigacion Sanitaria (PI19/00668). J. Wang has received a sponsored research grant from RootPath Genomics in the past 12 months and is funded by NIH grants R21AI163924 and R37CA273333. L. Chen has received sponsored research grants from NextCure, DynamiCure, and Normunity in the past 12 months and is funded by NIH grants CA196530 and CA016359 and a United Technologies Corporation endowment in cancer research.