Research ArticleCancer

Tumor-targeted 4-1BB agonists for combination with T cell bispecific antibodies as off-the-shelf therapy

See allHide authors and affiliations

Science Translational Medicine  12 Jun 2019:
Vol. 11, Issue 496, eaav5989
DOI: 10.1126/scitranslmed.aav5989

For the love of 4-1BB

Antibodies activating the 4-1BB costimulatory molecule did not succeed as cancer immunotherapy due to adverse events in patients. Claus et al. have now taken a protein engineering approach to make a next-generation 4-1BB agonist. They generated constructs with a trimeric 4-1BBL portion that engages 4-1BB on T cells without harmful Fc receptor cross-linking. The constructs also include tumor antigen–targeting portions and can be used in conjunction with other immunotherapies. These proteins successfully stimulated T cells in explanted samples from patients with cancer and improved survival in multiple mouse models. This approach reveals a potentially better way to stimulate 4-1BB for cancer immunotherapy.


Endogenous costimulatory molecules on T cells such as 4-1BB (CD137) can be leveraged for cancer immunotherapy. Systemic administration of agonistic anti–4-1BB antibodies, although effective preclinically, has not advanced to phase 3 trials because they have been hampered by both dependency on Fcγ receptor–mediated hyperclustering and hepatotoxicity. To overcome these issues, we engineered proteins simultaneously targeting 4-1BB and a tumor stroma or tumor antigen: FAP–4-1BBL (RG7826) and CD19–4-1BBL. In the presence of a T cell receptor signal, they provide potent T cell costimulation strictly dependent on tumor antigen–mediated hyperclustering without systemic activation by FcγR binding. We could show targeting of FAP–4-1BBL to FAP-expressing tumor stroma and lymph nodes in a colorectal cancer–bearing rhesus monkey. Combination of FAP–4-1BBL with tumor antigen–targeted T cell bispecific (TCB) molecules in human tumor samples led to increased IFN-γ and granzyme B secretion. Further, combination of FAP– or CD19–4-1BBL with CEA-TCB (RG7802) or CD20-TCB (RG6026), respectively, resulted in tumor remission in mouse models, accompanied by intratumoral accumulation of activated effector CD8+ T cells. FAP– and CD19–4-1BBL thus represent an off-the-shelf combination immunotherapy without requiring genetic modification of effector cells for the treatment of solid and hematological malignancies.


Cancer immunotherapy has changed the cancer treatment paradigm, but there remains a substantial need for improvement, particularly for patients with tumors lacking cytotoxic T cell infiltrates at baseline (13). Costimulation of activated T cells (4), natural killer (NK) cells, and other immune cells (5) via the tumor necrosis factor receptor superfamily (TNFRSF) member 4-1BB (CD137) has emerged as a promising approach for cancer immunotherapy (3, 6, 7). 4-1BB costimulation of T cells enhances proliferation, cytotoxicity, T helper cell (TH1) polarization and cytokine secretion (4, 810), metabolic fitness (7), modification of DNA methylation (11), and T cell memory formation (7), and counteracts exhaustion (9) and activation-induced cell death (10).

Today, two approaches relying on 4-1BB agonism have entered clinical trials: (i) agonistic anti-human 4-1BB antibodies (3, 6, 12, 13) and (ii) second-/third-generation 4-1BB/CD3ζ chimeric antigen receptor (CAR) T cells (7, 14, 15). The agonistic anti-human 4-1BB human immunoglobulin G4 (IgG4) antibody (anti–hu4-1BB huIgG4) urelumab (BMS-663513) caused dose-dependent hepatitis in patients (3, 6, 12, 13), likely due to 4-1BB cross-linking via FcγRIIb-expressing liver-resident cells such as hepatic myeloid and sinusoidal endothelial cells (16, 17). Subsequent studies revealed that, when urelumab was administered safely at a reduced dose of 0.1 mg/kg, it only mediated limited efficacy (3, 13). The anti–hu4-1BB huIgG2 utomilumab (PF-05082566) displays a better safety profile but lower agonistic potency (3). Thus, despite a decade-long effort, agonistic anti–hu4-1BB antibodies have not progressed beyond early stage clinical trials. Second-/third-generation CAR T cell treatments successfully use 4-1BB costimulation, and although approved and very effective in relapsed/refractory patients with B cell malignancies, they require patient-individualized and expert-dependent infrastructure and logistics. In addition, CAR T cells have not yet been successful for treatment of solid tumors (14). Recent publications have described bispecific tumor-targeted 4-1BB agonists (18, 19), which overcome liver toxicity but display a fast clearance from circulation (18).

To overcome these limitations, we developed tumor antigen (TA)–targeted 4-1BB ligand fusion proteins (TA–4-1BBL) for systemic administration. TA–4-1BBL effectively costimulates T cells for improved tumor cell killing in both solid and hematological cancers while avoiding liver toxicity. Here, we describe two versions of such an agonist: one targeting CD19, expressed on normal and malignant B cells, and another targeting fibroblast activation protein (FAP), expressed on tumor stroma and, to a lower degree, on lymphoid fibroreticular cells (20, 21). As the Fc region of the engineered agonists contains mutations abrogating cross-linked by Fcγ receptors (FcγRs) (22), TA–4-1BBL molecules can only induce 4-1BB activation when cross-linked via TA-expressing cells, representing a replacement of FcγR-mediated cross-linking, while maintaining favorable pharmacokinetics.


Design of a tumor-targeted 4-1BB agonist, which is independent of Fc-mediated hyperclustering for activity

As agonistic anti–hu4-1BB antibodies have failed so far in the clinic, our goal was to achieve potent 4-1BB stimulation at the tumor site without systemic FcγR cross-linking for a broad range of cancers. Initial experiments using human FcγR-expressing Chinese hamster ovary (CHO) cells or protein A beads confirmed that anti–hu4-1BB antibodies rely on Fc-mediated cross-linking to induce 4-1BB activation (fig. S1, A to C). To overcome this dependency, we generated TA–4-1BBL characterized by the following features (Fig. 1A): (i) a trimeric split 4-1BBL that binds 4-1BB, (ii) a monovalent Fab fragment that binds to CD19 or FAP, and (iii) a heterodimeric Fc region containing P329G L234A/L235A (PGLALA) mutations (22), which inhibit FcγR binding but retain FcRn binding for IgG-like pharmacokinetics (Fig. 1B), whereas bispecific tumor-targeted 4-1BB agonists lacking an IgG-like pharmacokinetic are cleared from circulation within 50 hours (18). Having tested different stoichiometries, we confirmed the strongest cross-linking-dependent activity of TA–4-1BBL at a ratio of 3:1 between the 4-1BB receptor and TA-binding domains (fig. S1D) (19). A recently described tumor-targeted 4-1BB agonist with a 2:2 ratio (23) was inferior to TA–4-1BBL (fig. S1, E and F).

Fig. 1 Design of a bispecific tumor-targeted 4-1BB ligand (TA–4-1BBL), which features IgG-like pharmacokinetics.

(A) TA–4-1BBL consists of an Fc, devoid of FcγR binding (huIgG1PGLALA) (22), a trimeric hu4-1BBL, and a Fab, binding to FAP or CD19. Heterodimerization and correct assembly are achieved by knob into hole (kih) mutation in the Fc, CH1-CL domain crossover, and mutations in CH1(EE) and CL(RK) (49, 50). Class averages based on negative stain transmission electron microscopy of the whole molecule and x-ray crystallography (for data collection and refinement statistics, see table S1) of the trimeric 4-1BBL (missing CH1/CL parts in the structure are shown in dashed lines) were used to further characterize TA–4-1BBL. (B) Pharmacokinetic profile of TA–4-1BBL after a single dose injected intravenously, 12 mice per group. After therapy, three mice per group were bled at indicated time points. Concentration of drug in serum was measured by enzyme-linked immunosorbent assay (ELISA). Shown is mean (SD) of n = 3 mice per time point and group. FAP–4-1BBL and CD19–4-1BBL were tested once in two different experiments and combined in this graph. (C) Simultaneous binding by surface plasmon resonance to both targets, 4-1BB and FAP (left) or 4-1BB and CD19 (right). Results were confirmed in two independent experiments.

We demonstrated in vivo that liver inflammation induced by 4-1BB antibodies can be prevented by impeding FcγR interactions. Treatment with anti-mouse 4-1BB (anti–mu4-1BB) antibodies led to accumulation of 4-1BB+ proliferating CD8+ T cells and CD68+ monocytes/macrophages in the liver of C57BL/6J mice, as previously described (24, 25). Introduction of D265A P329G mutations (DAPG) in the Fc region abrogating FcγR binding (26) did not affect pharmacokinetics (fig. S2A) but prevented immune cell infiltration and hepatocellular tissue damage (fig. S2, B to E).

Structural analysis by negative-stain transmission electron microscopy showed an IgG-like structure (Fig. 1A) with a compact trimeric 4-1BBL in different conformations, indicating structural flexibility (fig. S3A). To confirm that the trimeric organization of human 4-1BBL (27, 28) is also a feature of our fusion protein, we determined the crystal structure of the split 4-1BBL–CH1/CL fragment at 2.69 Å (table S1). TA–4-1BBL contains one 4-1BBL trimer, closely resembling the trimer of the described 4-1BBL (27, 28) and of other TNFSF members (fig. S3B).

TA–4-1BBL bound to recombinant human and cynomolgus/rhesus monkey 4-1BB but was not cross-reactive to mu4-1BB (fig. S3C). The monovalent affinities for hu4-1BB were in the three-digit nanomolar range (table S2), and half-maximal effective concentration (EC50) binding values on activated human T cells were in the low nanomolar range (fig. S4, A to C). No binding to resting huPBMCs was observed (fig. S4D). The Fab monovalent affinities for recombinant FAP or CD19 were in the picomolar range (table S2) with EC50 binding values on TA-expressing cells in the low nanomolar range (fig. S4, E and F). FAP– and CD19–4-1BBL were shown to simultaneously bind hu4-1BB and huFAP or huCD19, respectively (Fig. 1C). Relatively high affinity for the TA was selected to enable effective tumor targeting. In a rhesus monkey bearing a natural colorectal adenocarcinoma (CRC) with FAP+ tumor-associated fibroblasts in the tumor and fibroblastic reticular cells in the lymph nodes (LNs) (Fig. 2A), we tested tumor-targeting of FAP–4-1BBL. A high FAP-mediated uptake was confirmed for 89Zr-FAP–4-1BBL by immunoPET/CT imaging with increasing target-to-background uptake ratios between 48 and 120 hours after drug administration (Fig. 2, B and C, and table S3).

Fig. 2 FAP–4-1BBL targets to FAP+tumor tissue and LN in a CRC-bearing rhesus monkey.

(A) Immunohistochemistry staining of FAP in the peritumoral LN and in tumor tissue after necropsy (n = 1). (B) Maximum intensity projection of a CRC-bearing rhesus monkey 120 hours after intravenous injection of FAP–4-1BBL (0.5 mg/kg) mixed with tracer amounts of 89Zr-labeled FAP–4-1BBL (165 MBq). Tissue accumulation of radioactive tracer is described on a minimum-maximum standard uptake value (SUV) scale. (C) SUV of indicated organs determined by positron emission tomography after 48 and 120 hours. Organs of FAP expression, perfusion, and excretion (including organs of antibody metabolism) are indicated. Targeting and retention of FAP–4-1BBL are defined as an increase of SUV from 48 to 120 hours.

FAP–4-1BBL increases T cell activation, proliferation, and CD8+ EM T cell formation

The functionality of FAP–4-1BBL was assessed in different in vitro experiments. In a hu4-1BB–expressing reporter cell line, FAP–4-1BBL activated the nuclear factor κB (NFκB) pathway only in the presence of FAP-expressing cells, whereas minimal activation occurred in the absence of FAP or using untargeted control DP47–4-1BBL (Fig. 3A, DP47 is a germline control). Known anti–hu4-1BB antibodies were either not active (MOR-7480.1 huIgG2) or not able to induce activation to the same extent (20H4.9 huIgG4) owing to the lack of proper cross-linking (no addition of FcγR-expressing cells as in fig. S1B).

Fig. 3 FAP–4-1BBL induces in vitro T cell activation in the presence of signal 1.

(A) HeLa-hu4-1BB-NFκB-luc reporter cells were cocultured with or without huFAP-expressing cells, 4-1BB agonists, or control proteins. Luciferase activity was measured as units of released light (URLs) (n = technical duplicates). Significance was calculated, comparing the area under the curve (AUC) by unpaired one-way analysis of variance (ANOVA) with Tukey’s multiple comparison test. EC50 values for FAP–4-1BBL are indicated as mean (SD). (B) Carboxyfluorescein diacetate succinimidyl ester (CFSE)–labeled human peripheral blood mononuclear cells (huPBMCs) were cocultured with irradiated NIH/3T3-huFAP fibroblasts in the presence or absence of agonistic 2 nM CD3 huIgG1 and titrated 4-1BB agonists. Proliferation (CFSE dilution), 4-1BB, and CD25 expression of CD4+ or CD8+ T cells after 5 days [mean (SD), n = technical triplicates] are shown. Significance was calculated as in (A). (C) CFSE-labeled huPBMCs were cocultured with irradiated FAP-expressing cells (NIH/3T3-huFAP or U87MG), 2 nM CD3 huIgG1 alone, or in combination with DP47–4-1BBL or FAP–4-1BBL. CD4+ and CD8+ T cells were characterized for their expression of CD45RO, CD45RA, and CD62L by flow cytometry. Total counts of 4-1BB–expressing T cells divided in CM, EM, naive, and Teff/TEMRA subpopulations are shown [mean (SD), n = technical triplicates]. Unpaired one-way ANOVA with Tukey’s multiple comparison test was used. (D) CFSE-labeled HLA-A*02:01/NLV peptide–specific EM CD8+ T cells were cocultured with NLV peptide pulsed (1 or 10 nM) or unpulsed HLA-A*02:01+ FAP+ MV-3 melanoma cells and 4-1BB agonists or controls. After 24 hours, brefeldin A and monesin were added for 4 hours and cells were analyzed for 4-1BB and IFN-γ expression by flow cytometry [mean (SD), n = technical triplicates]. Significance was calculated as in (A). (E and F) CFSE-labeled huPBMCs of five different donors were cocultured with irradiated NIH/3T3-huFAP and carcinoembryonic antigen (CEA)+ PD-L1+ gastric tumor cells (MKN45-huPD-L1), FAP–4-1BBL, carcinoembryonic antigen-targeted T cell bispecific antibody (CEA-TCB) or atezolizumab alone or in combination [atezolizumab shown only in (F)]. After 4 days, proliferation (CFSE dilution), CD25, and 4-1BB expression of CD8+ T cells [mean (SD), n = technical triplicates] were analyzed by flow cytometry. IFN-γ in supernatant was measured by multiplex (n = technical duplicates). Significance was analyzed as in (C). Each shown experiment represents results of at least three independent and similar experiments. For huPBMC-based assays, at least three different donors were tested, but one representative donor is shown. n.s., not significant.

We then used huPBMCs from healthy donors to activate human T cells with FAP–4-1BBL and agonistic anti-human CD3 huIgG1 providing signal 1. Only in the presence of CD3 stimulation did FAP–4-1BBL induce an increase in CD8+ and CD4+ T cell proliferation (Fig. 3B), leading to increased total CD8+ and CD4+ T cell count after 4 days (fig. S5A). An up-regulation of activation and costimulatory markers such as 4-1BB (detected with not-4-1BBL–competing anti–hu4-1BB antibody clone 4B4-1; fig. S5B), CD25, OX40, ICOS, and CD69 was observed (Fig. 3B and fig. S5C). In contrast, DP47–4-1BBL or anti–hu4-1BB antibodies did not activate T cells similarly (Fig. 3B). FAP–4-1BBL–mediated costimulation could be induced in the presence of cells with different FAP expression (Fig. 3C and fig. S5, A, C, and D), leading to an increase in the total number of 4-1BB+ central memory (CM) and effector memory (EM) T cells (Fig. 3C) and frequency of EM T cells (fig. S5D).

We next investigated whether FAP–4-1BBL can enhance antigen-specific EM CD8+ T cell function. Blood-sorted and in vitro (on feeder cells) expanded HLA-A*02:01/NLV peptide–specific EM CD8+ T cells were stimulated by FAP+ HLA-A*02:01+ MV-3 melanoma cells pulsed with NLV peptide. FAP–4-1BBL increased 4-1BB and interferon-γ (IFN-γ) expression only in the presence of NLV peptide (Fig. 3D).

Subsequently, huPBMCs were treated with different combinations of FAP–4-1BBL, carcinoembryonic antigen-targeted T cell bispecific antibody (CEA-TCB) (fig. S5E) (2931), and PD-L1–blocking antibody (atezolizumab) for 4 days in the presence of PD-L1+ CEA+ MKN45 human gastric cancer cells, as well as FAP+ fibroblasts. Combination of FAP–4-1BBL and CEA-TCB increased proliferation and 4-1BB and CD25 expression on CD8+ T cells (Fig. 3E), and the addition of atezolizumab further increased IFN-γ secretion (Fig. 3F). These data show that FAP–4-1BBL in the presence of signal 1 (CD3ε or T cell receptor stimulation) induces strong, FAP-dependent T cell activation and proliferation with subsequent expansion of the memory T cell pool.

FAP–4-1BBL costimulates T cells in ex vivo patient-derived tumor tissue

To assess whether FAP–4-1BBL can improve the functionality of tumor-infiltrating T cells, tumor digests of epithelial ovarian cancer (EOC) samples positive for FAP (Fig. 4A) were stimulated with agonistic anti–huCD3 huIgG1 and FAP–4-1BBL. Unlike huPBMCs, we did not observe a statistically significant increase of 4-1BB expression and proliferation (Ki67) of CD8+ T cells (Fig. 4B). As tumor-infiltrating T cells mainly display a memory phenotype, we decided to focus on cytokine secretion. Stimulation of the remaining digest of EOC patient 4 with anti-huCD3 huIgG1 and FAP–4-1BBL led to significant (P < 0.0001) IFN-γ secretion (Fig. 4C). Digests of melanoma lesions were analyzed for FAP expression and frequency of CD3+ T cells (Fig. 4D). Stimulation with agonistic anti-huCD3 huIgG1 and FAP–4-1BBL led to an increase of several inflammatory cytokines in the supernatant in all three patient samples (Fig. 4E). IFN-γ and the granzyme B were mainly affected; however, the patterns of cytokine modulation were strongly patient dependent.

Fig. 4 FAP–4-1BBL in combination with CD3 activation induces cytokine release from FAP-expressing human cancer samples.

(A) Single-cell suspensions of EOC tumor lesions were tested for FAP expression. (B) Single-cell suspensions of EOC patients’ tumor tissue were cultured unstimulated, with agonistic CD3 huIgG1 alone, or in combination with FAP–4-1BBL. Shown are the frequencies of 4-1BB+ and Ki67+ CD8+ T cells determined by flow cytometry. Significance was calculated using paired one-way ANOVA with Tukey’s multiple comparison test. (C) Tumor suspension of patient 4 was incubated in medium, with CD3 huIgG1 or FAP–4-1BBL alone or in combination. Supernatant was analyzed for cytokines by multiplex. Technical duplicates are shown as individual curves. Significance was calculated, comparing the AUC using unpaired one-way ANOVA with Tukey’s multiple comparison test. (D) Melanoma digests from three patients were tested for frequency of FAP- and CD3-expressing cells by flow cytometry. (E) FAP+ tumor digests were cultured unstimulated or with CD3 huIgG1, FAP–4-1BBL, or DP47–4-1BBL alone or in combination. Secreted cytokines were analyzed using cytometric bead array (CBA). Depending on tumor size sample, n = 2 (patient 5 and 6) or n = 1 (patient 7) technical replicate per condition was measured. Significance was calculated using unpaired two-way ANOVA with Tukey’s multiple comparison test. (F) Colon adenocarcinoma, lung adenocarcinoma, lung adenosquamous carcinoma, or melanoma tumor lesions were analyzed for FAP, CEA, or tyrosinase-related protein (TYRP1) expression. (G) Fresh tumor tissue from lesions shown in (F) were placed in bioreactors (each symbol represents one bioreactor), supplied with 8 ml of medium/reactor, and for patient 8, with autologous huPBMCs. Bioreactors were incubated with medium alone (untreated) or stimulatory molecules as indicated. Supernatant was analyzed for IFN-γ using CBA. Significance was analyzed using unpaired one-way ANOVA multiple comparison, uncorrected Fisher’s LSD test. Because of the limited available patient material, all performed assays are shown.

To test whether natural expression of FAP (Fig. 4F) in nondisrupted human tumor tissues is sufficient to provide functional cross-linking, we combined FAP–4-1BBL with CEA-TCB or a melanoma-specific TYRP1-TCB (fig. S5E). Tyrosinase-related protein (TYRP1) is involved in melanin synthesis and found on the surface of many melanoma cells (32). In the first experiment, a CEA+ and FAP+ colon adenocarcinoma tumor sample (patient 8) was incubated in a perfused system enabling three-dimensional tissue culture (bioreactor) (33). As this tumor displayed low T cell infiltration, autologous huPBMCs were added. Stimulation with FAP–4-1BBL and CEA-TCB led to a strong increase of IFN-γ production (Fig. 4G). We further tested tumor samples from lung adenocarcinoma (patient 9), lung adenosquamous carcinoma (patient 10), and melanoma (patient 11) without addition of autologous huPBMCs (Fig. 4G). As the melanoma tumor was CEA but TYRP1+ (Fig. 4F), we used the TYRP1-TCB to deliver CD3 stimulation. Combination of TCBs with FAP–4-1BBL strongly increased IFN-γ production by tumor-infiltrating lymphocytes (Fig. 4G). This shows that FAP–4-1BBL can be effectively combined with different TCBs and that expression and distribution of FAP in human tumor tissues are sufficient to provide target-mediated costimulation.

Combination of CEA-TCB and FAP–4-1BBL induces CD8+ T cell infiltration and tumor regression

CEA-TCB treatment induces 4-1BB up-regulation in tumor-infiltrating CD8+ T cells (fig. S6A). Therefore, we combined FAP–4-1BBL with CEA-TCB in human stem cell (HSC)–engrafted NSG mice bearing subcutaneous MKN45 xenograft tumors cografted with mouse NIH/3T3 fibroblasts to mimic FAP expression comparable to human patient samples (Fig. 5A). The combination of FAP–4-1BBL and CEA-TCB inhibited tumor growth (Fig. 5, B and C), whereas the respective monotherapies or the combination of CEA-TCB with DP47–4-1BBL did not. The improved tumor growth control was accompanied by an increase of CD8+ T cells in the tumor as confirmed by flow cytometry (Fig. 5D) and immunohistochemistry staining (Fig. 5, E and F). The ratio of CD8+ to CD4+ T cells was increased, but there was no significant shift in the CD8+/Treg ratio (Fig 5D). No immune cell accumulation in the liver occurred; however, the low efficiency of myeloid cell engraftment of the HSC-NSG mice model (34) may not completely reflect a fully immunocompetent setting. Of note, in HSC-NSG, mice bone marrow necrosis occurred after treatment with FAP–4-1BBL (Fig. 5G), whereas no such effects could be observed in HSC-NSG mice treated with CD19–4-1BBL or in fully immunocompetent C57BL/6J mice receiving surrogate muFAP-4-1BB therapy or rhesus or cynomolgus monkeys receiving FAP–4-1BBL therapy. This observation is indicative of a model-related bone marrow effect, occurring only in HSC-NSG mice most probably because of bone marrow remodeling and therefore muFAP up-regulation after HSC transplantation.

Fig. 5 Combination of FAP–4-1BBL and CEA-TCB in vivo decreases tumor growth and increases intratumoral CD8+ T cell accumulation.

(A) A representative example of FAP expression in the tumor at the start of therapy (200 mm3). (B and C) Shown are two of three independent experiments. CEA+ human gastric cancer cells MKN45 and NIH/3T3-huFAP fibroblasts were coinjected subcutaneously into human stem cell–engrafted NSG (HSC-NSG) mice. At an average tumor size of 200 mm3, mice were treated with vehicle, CEA-TCB, FAP–4-1BBL, or DP47–4-1BBL alone or in combination as indicated (n = 10 per group). Shown is mean (SEM). Statistical significance of tumor volumes at the end point was calculated using unpaired one-way ANOVA with Tukey’s multiple comparison test (B) or unpaired, two-tailed Student’s t test (C). (D) Digested tumor tissues (day 52) were analyzed by flow cytometry gating on living human CD4+ or CD8+ T cells. Regulatory T (Treg) cells were defined as FoxP3+ CD25+ CD4+ T cells. Each symbol represents one individual mouse; bars indicate mean (SD). Significance was calculated using unpaired one-way ANOVA with Tukey’s multiple comparison test. (E) Representative immunohistochemistry staining for CD3+ and CD8+ cells is shown. Tissue sections were scanned, and whole scans were analyzed by Definiens. Tumors were taken from the experiment shown in (C) at termination. (F) CD3+ and CD8+ T cells count per square millimeter of immunohistochemistry analyzed by Definiens for the whole section. Each symbol represents one individual mouse; shown is mean (SD). Significance was calculated as in (D). (G) Ten HSC-NSG mice per group were treated weekly with vehicle, FAP–4-1BBL, or DP47–4-1BBL intravenously. Three days after first (day 3) and second injection (day 10), bone marrow from femur (n = 5 per group) was analyzed for cell counts by flow cytometry. Tumor experiments were repeated in three independent experiments; bone marrow evaluation experiment was assessed once.

As hu4-1BBL is not mouse cross-reactive and the analogous mu4-1BBL fusion protein could not be prepared, we generated a bispecific mouse surrogate antibody (muFAP-4-1BB) for mechanistic studies in fully immunocompetent mice. The surrogate contained two 4-1BB–binding sites and one FAP-binding site (fig. S6B) and a muIgG1-Fc containing DAPG mutations to abrogate FcγR binding while retaining antibody-like half-life (26). As the mu4-1BBL naturally forms a dimer (35) and not a trimer as in humans (27, 28), we considered muFAP-4-1BB a suitable mouse surrogate. We tested the efficacy of muFAP-4-1BB in combination with mouse surrogates for CEA-TCB (muCEA-TCB; fig. S5D) or atezolizumab (muPD-L1) in huCEA transgenic C57BL/6J mice (36), bearing huCEA-expressing MC38 tumors (MC38-huCEA). Although this tumor is infiltrated by endogenous muFAP+ mouse fibroblasts, it displays lower FAP expression at baseline as compared to human tumors (fig. S6C). Treatment with muFAP-4-1BB alone did not reduce tumor growth but supported tumor remission when combined with muCEA-TCB or muPD-L1 (fig. S6, D to F). As expected, no immune cell accumulation in the liver was observed (fig. S6G). Therefore, FAP–4-1BBL is a promising combination partner for cancer immunotherapies, including checkpoint inhibitors and TCBs.

Cross-linked CD19–4-1BBL costimulates T cells in vitro

Last, we applied the same concept for CD19+ B cell lymphoma. CD19–4-1BBL was not internalized after binding to CD19+ B cells (fig. S7A). The functional properties were confirmed by 4-1BB+ reporter cells incubated with a variety of CD19+ tumor cell lines (Fig. 6A). Combined with a CD3/CD28 bead or PHA-L stimulus, CD19–4-1BBL increased IFN-γ production by huPBMCs, whereas untargeted DP47–4-1BBL did not (Fig. 6B). These data showed that, similarly to FAP–4-1BBL, CD19–4-1BBL is functionally active.

Fig. 6 CD19–4-1BBL induces in vitro T cell activation and localizes with CD20-TCB in the immunological synapse.

(A) HeLa-hu4-1BB-NFκB-luc reporter cells were cocultured in the absence or presence of huCD19-expressing cell lines and CD19–4-1BBL or DP47–4-1BBL. Shown is the luciferase activity after 6 hours as URLs (n = technical duplicates). Significance was calculated comparing AUC by unpaired two-tailed Student’s t test. EC50 values are indicated as mean (SD). (B) huPBMCs (n = technical duplicates) were incubated with 1.2 × 106 CD3/CD28 beads/ml (top) or PHA-L (1 μg/ml) (bottom) and CD19–4-1BBL or DP47–4-1BBL, whereby CD19–4-1BBL is cross-linked via B cells. IFN-γ secretion was analyzed by ELISA after 2 days. Significance was calculated as in (A), and EC50 values are indicated as mean (SD). (C) Human CD3+ T cells isolated from buffy coat (n = technical duplicates) were cocultured with WSU-DLCL2 cells and CD20-TCB. Expression of 4-1BB and CD69 on CD8+ T cells was determined by flow cytometry. (D) WSU-DLCL2–bearing HSC-NSG mice (n = 4) were treated weekly intravenously with vehicle or CD20-TCB. Tumors were isolated 2 days after second injection, and tumor-infiltrating CD8+ T cells were analyzed by flow cytometry for 4-1BB expression. Each symbol indicates one individual mouse. Shown is mean (SD); significance was calculated by unpaired two-tailed Student’s t test. (E) CD3+ T cells were cocultured with WSU-DLCL2 cells, CD19–4-1BBL, and CD20-TCB as indicated. After 18 hours, supernatants were analyzed by ELISA for IFN-γ secretion. Shown is the mean (SD) of n = technical duplicates, indicated with symbols. Significance was calculated using an unpaired one-way ANOVA with Tukey’s multiple comparison test. (F) Localization of CD19–4-1BBL (magenta), WSU-DLCL2 cells (blue), and activated 4-1BB–expressing CD8+ T cells (green) by confocal fluorescence microscopy. (G) The localization of CD19–4-1BBL, WSU-DLCL2 cells, and CD20-TCB–activated 4-1BB–expressing CD8+ T cells was monitored for up to 90 frames. In the graph, the intensity of AF647-labeled CD19–4-1BBL in the surface contact area between 4-1BB+ CD8+ T cells and CD19+ WSU-DLCL2 cells is shown. Experiments shown in (A) to (D) were repeated at least three times in independent experiments. If primary cells were used, then at least three different donors were tested. Experiments shown in (E) to (G) were performed only once.

There is a strong rationale to combine CD19–4-1BBL with CD20-TCB (fig. S5D) (37, 38) as CD20-TCB induced in vitro (Fig. 6C) and in B cell lymphoma–bearing HSC-NSG mice (Fig. 6D) 4-1BB up-regulation on CD8+ T cells. Addition of CD19–4-1BBL and CD20-TCB to a coculture of WSU-DLCL2 and T cells induced an increase in INF-γ secretion (Fig. 6E). To elucidate the mode of action of this combination therapy, time-lapse confocal microscopy was applied to monitor immunological synapse formation on activated T cells. Dynamic localization of CD19–4-1BBL during CD20-TCB–mediated T cell and tumor cell cross-linking could be observed mainly at sites of interaction of immune and tumor cells in a CD20-TCB dose-dependent manner (Fig. 6, F and G). CD20-TCB cross-linking initiated the interaction, followed by redistribution of CD19–4-1BBL to the immunological synapse, possibly mediated by lipid rafts attracted by binding to CD20 (39). These data indicate that simultaneous targeting of CD19–4-1BBL and CD20-TCB-occured, leading to a synapse formation between B cells and T cells.

Combination of CD20-TCB and CD19–4-1BBL eradicates tumors in vivo

The combination of CD19–4-1BBL with CD20-TCB was studied in HSC-NSG mice bearing aggressive human lymphoma WSU-DLCL2. A suboptimal dose of CD20-TCB showed moderate tumor growth inhibition on day 20, whereas combination with CD19–4-1BBL resulted in an earlier onset of tumor reduction and complete eradication of tumors without signs of toxicity (Fig. 7A). This was accompanied by marked infiltration of T cells (Fig. 7, B and C) and increase of the intratumoral CD8+/CD4+ T cell ratio (Fig. 7B). Immunohistochemistry revealed enhanced granzyme B secretion and CD3+ T cell infiltration, whereas CD20 and CD19 expression declined. As tumors were rejected at day 46, we assume that this is a sign of tumor eradication and not of antigen down-regulation, e.g., tumor escape (Fig. 7C).

Fig. 7 Combination of CD19–4-1BBL and CD20-TCB in vivo inhibits tumor growth and increases intratumoral CD8+ T cell accumulation and activation.

(A) Human WSU-DLCL2 cells were injected subcutaneously in HSC-NSG mice. At 250 mm3 tumor size, mice were treated with vehicle or CD20-TCB or CD19–4-1BBL alone or in combination. Shown is the mean (SEM) of the number of mice per group as indicated. Statistical significance of tumor volumes at the end point was calculated using unpaired, two-tailed Student’s t test. (B) Digested tumor single-cell suspensions from animals euthanized at day 20 (n = 4 per group indicated by symbols) were analyzed by flow cytometry. Shown is mean (SD), and significance was calculated using unpaired one-way ANOVA with Tukey’s multiple comparison test. (C) Representative immunohistochemistry staining of tumor tissue from animals euthanized at day 20 (n = 4 per group) for huCD3+, granzyme B (GnzB)+, huCD20+, and huCD19+ cells is shown for each group. Tissue sections were scanned, and whole scans were analyzed by Definiens. The shown experiment represents results of three independent experiments.

We evaluated this additive effect with Nalm-6 B cell lymphoma displaying a much lower CD20 expression than WSU-DLCL2 (fig. S7B). Again, only the combination of CD20-TCB and CD19–4-1BBL resulted in improved tumor control (fig. S7C) correlating with an accumulation of T cells in the tumor (fig. S7, D and E).


In this study, we focused on the effect of TA–4-1BBL on T cell activation. Other immune cells express 4-1BB (e.g., NK cells, myeloid cells, and neutrophils) and play a role in tumor immunology but were not a focus of this study. The role of 4-1BBL reverse signaling blockage on myeloid cells (40) as well as on long-lasting tumor-specific memory formation of T cells was also not explored.

Several preclinical and clinical studies have highlighted the importance of 4-1BB costimulation to maintain long-term functionality of T cells by anti-apoptotic and metabolic adaptation, TH1 polarization, increase in memory formation, and cell survival (4, 7, 9). Second/third-generation CAR T cells, including the U.S. Food and Drug Administration–approved tisagenlecleucel (14, 15), exploit agonism of 4-1BB signaling to enhance T cell activity and persistence. However, in contrast to CAR T cell therapies, agonistic anti–hu4-1BB antibodies have not advanced beyond early clinical trials. Urelumab is limited by hepatotoxicity, and utomilumab has shown lower efficacy (3). Here, we propose that these limitations come from the dependence of such agonists on FcγR-mediated cross-linking in vivo to induce sufficient 4-1BB signaling and describe an alternative approach based on a molecular format for systemically delivered 4-1BB agonists.

Human TNFRSF members form complexes with their natural ligands based on three receptor units binding to a single homotrimeric ligand, but typically two or more such complexes need to cluster together on the plasma membrane to initiate sufficient signaling (41). Therefore, bivalent binding molecules such as IgGs cannot induce substantial TNFRSF member agonism, unless receptors or antibody-receptor complexes are further hyperclustered. In vivo, this can be mediated by binding FcγRs via the Fc domain, particularly the inhibitory FcγRIIB (17, 42), which has a comparable cellular expression pattern between mice and humans (42, 43). An unwanted side effect of these anti-TNFRSF antibodies is the interaction with FcγRIIB in liver, where it is expressed by sinusoidal endothelial cells, macrophages, and dendritic cells, leading to major adverse events as shown for CD95 or DR5 (16, 17). This is in line with the described T cell and macrophage liver accumulation induced by anti–mu4-1BB rat IgG2a antibodies (24, 25) cross-linked via muFcγRIIB (44). In patients with cancer, urelumab showed dose-limiting grade 4 hepatitis (12), leading to subsequent dose reduction (13). Here, we demonstrate that the in vivo activity of agonistic anti–mu4-1BB antibodies is affected by FcγR interactions and that mutating the Fc region to abolish such interactions is effective in preventing hepatic inflammation. Beyond side effects, anti-tumor efficacy may also be affected by intratumoral FcγRIIB expression on tumor-associated myeloid cells and macrophages (42, 43). Intratumoral FcγRIIB density is variable across patients and can be much lower than observed in typical mouse tumor models.

To overcome these problems, we developed TA–4-1BBL. In contrast to conventional antibodies, the activity of TA–4-1BBL is independent of FcγR cross-linking, and 4-1BB signaling is only triggered upon binding of TA–4-1BBL to the tumor stroma target FAP (trans) or the tumor target CD19 (cis) (fig. S8). FcRn binding and IgG-like pharmacokinetic properties of TA–4-1BBL are not affected by the introduction of PGLALA mutations in the Fc region. This allows the molecule to have high systemic exposure, a requirement for sufficient tumor accumulation of macromolecules (45). In addition, we confirmed that the trimeric 4-1BBL, cross-linking three 4-1BB receptors with one tumor target, gave maximal potency for 4-1BB agonism (19). Other valency ratios were less active, indicating that an optimal ratio between 4-1BB agonist and tumor-targeting domain exists.

In patients with preexisting or induced T cell–inflamed tumors, TA–4-1BBL may be combined with immune checkpoint inhibitors to improve tumor cell killing. In the T cell–inflamed, syngeneic MC38-huCEA mouse tumor model, muFAP-4-1BB induced tumor regression when combined with an anti–PD-L1 antibody. This supports TA–4-1BBL application in cancers responsive to checkpoint inhibition. Especially, as 4-1BB has been reported as a marker for antigen-specific effector T cells in the tumor (46).

However, most patients with cancer do not have sufficient endogenous anti-cancer T cell intratumoral infiltration. Here, synthetic T cell redirection using either TCBs or individualized CAR T cell therapies is a potential treatment approach. In the case of hematological B cell tumors, both TCBs and CAR T cell therapies have shown promising efficacy and some are already approved (15, 47, 48). Our group has developed CD20-TCB (RG6026) (37, 38), which is currently being tested in phase 1 clinical trials (NCT03075696). Here, we show that the combination of CD20-TCB with CD19–4-1BBL enhances intratumoral T cell activity and treatment efficacy in preclinical models. For the treatment of non–T cell–inflamed solid tumors overexpressing huCEA, we developed CEA-TCB (RG7802) (2931), which has recently reported promising efficacy in very hard to treat, third-line-plus patients with colorectal cancer who are microsatellite stable (31). In such solid tumors, FAP is frequently expressed, which supports a future combination of CEA-TCB and FAP–4-1BBL. We believe that clinical development of the described TA–4-1BBL antibody fusion proteins in combination with TCBs provides a potent, safe, and convenient “off-the-self” alternative to CAR T cell therapies given the ease of administration, pharmacokinetic properties of administrating an antibody fusion protein, and, in the case of CEA-TCB and FAP–4-1BBL, improved tumor specificity by targeting two orthogonal targets on tumor cells and tumor stroma.


Study design

This study was designed to characterize TA–4-1BBLs developed to costimulate T cells at the tumor site. For in vitro assays in general, at least three independent experiments were performed, and in the case of primary cells, at least three donors were tested if not indicated differently. All studies were performed in technical duplicates or triplicates as indicated. Buffy coats were obtained from Zurich blood donation center in accordance with the Declaration of Helsinki. Donors signed a written informed consent before sample collection. In vivo, we use three different humanized mouse tumor models and one syngeneic tumor mouse model (muFAP-4-1BB surrogate). The Cantonal Veterinary Office, Zurich, Switzerland, approved the protocols (P2011/128, ZH193-2014, and ZH223/17) in accordance with the Swiss Animal Protection Law. Tumor-bearing mice were randomized 1 day before treatment by tumor size into groups of 10 mice per treatment. Repeats of experiments are indicated in the figure legends. An imaging study was conducted with one male CRC-bearing rhesus monkey (21 years, 8.5 kg). The procedure was approved by the Wake Forest Institutional Animal Care and Use Committee, in compliance with the U.S. Animal Welfare Act, the Guide for the Care and Use of Laboratory Animals, the Office of Laboratory Animal Welfare, and public health service regulations. Wake Forest is accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care, International. For ex vivo patient material, all patients signed a written informed consent before sample collection. Ethical approvals were given as follows: EOC samples were collected at the University Hospital Basel and approved by the local governmental ethical commission Nordwestschweiz EK321/10; other tumor samples were collected at the University Hospital Zurich or Hirslanden Hospital, Zurich, and approved by the local governmental ethical commission Kanton Zurich. None of the experiments were blinded. Primary data are reported in data file S1.

Statistical analysis

GraphPad Prism 7.04 and JMP version 12 were used for statistical analysis. For titration curves, AUC was calculated and used for statistical comparison. EC50 values were determined using nonlinear regression curve, fit variable slope (four parameters), and least-squares fit. Data are shown as mean with SD (n = 3 or more) or individual curves (n < 3). Technical repeats or single patients or mice are indicated by single symbols if possible. The statistical tests used are indicated in the figure legends for each experiment. Tumor growth curves are shown as mean with SEM. P values are indicated by asterisks as *P < 0.05, **P < 0.01, ***P < 0.0001, and ****P < 0.00001.


Materials and Methods

Fig. S1. TA–4-1BBL features a distinct functionality compared to other 4-1BB agonists.

Fig. S2. Anti–4-1BB antibodies cross-linked via FcγR induce immune cell accumulation in the liver.

Fig. S3. TA–4-1BBL is a flexible molecule with a trimeric 4-1BBL structure.

Fig. S4. TA–4-1BBL binds specifically to cell-expressed targets.

Fig. S5. In vitro FAP–4-1BBL costimulation improves T cell activation and memory phenotype differentiation.

Fig S6. muFAP-4-1BB works in combination with CEA-TCB or anti–PD-L1 in syngeneic mice.

Fig S7. CD19–4-1BBL and CD20-TCB combination induces tumor regression also in CD20low-expressing B cell lymphoma.

Fig S8. TA–4-1BBL works in cis- and trans-presentation in the presence of signal 1.

Table S1. X-ray data collection and refinement statistics for human 4-1BBL–CH1/CL.

Table S2. Surface plasmon resonance affinity constants of TA–4-1BBLs.

Table S3. FAP–4-1BBL tumor targeting and tissue biodistribution in a CRC-bearing rhesus monkey.

Table S4. Monomer content of FAP–4-1BBL in production batches.

Data file S1. Primary data.

References (5162)

References and Notes

Acknowledgments: We acknowledge the technical and scientific support of teams at Roche Innovation Centers Zurich, Munich, and Basel. We thank J. T. Regula (Roche Innovation Center Munich) for his contribution to optimize the fusion protein design; J. Kim from Genentech for providing muPD-L1; M. Canamero, N. Rieder, and H. Sade (Roche Innovation Center Munich) for the transfer of the FAP immunohistochemistry assay; and M. Schätz, M. Galm, T. Schlothauer, J. Fischer, J. Endl, and S. Seeber (Roche Innovation Center Munich) for providing huFcγR-expressing CHO cells. We also thank M. Levesque, J. Martinez-Gomez (University Zürich, Dermatology, Switzerland), and D. Tschopp and colleagues (Hirslanden Hospital, Zurich, Switzerland) for providing primary tumor samples. Funding: C.C. was initially supported by a Roche postdoctoral fellowship. A.G.B.-R. was financially supported by NIH grant no. P40 OD012217 and the University of Puerto Rico. This work was supported, in part, by the shared resources of the Wake Forest Clinical and Translational Science Institute NIH/NCAT grant UL1TR001420 and the shared resources of the Wake Forest Baptist Comprehensive Cancer Center’s NIH/NCI Cancer Center Support grant P30CA012197-39. A.Z. was supported by grants from the Swiss National Science Foundation (320030_162575). Author contributions: C.C. performed, designed, and supervised cell-based in vitro experiments with TA–4-1BBL; discussed and interpreted results; and wrote the manuscript. C.F. designed molecules, performed and supervised biochemical experiments of TA–4-1BBL, discussed and interpreted results, and wrote the manuscript. W.X. supervised the research and designed cell-based in vitro experiments with CD19–4-1BBL, discussed and interpreted results, and wrote the manuscript. J.S. and S.L. designed and coordinated the in vivo mouse tumor models and ex vivo analysis by flow cytometry of mouse tissues and pharmacokinetic studies, discussed and interpreted results, and wrote the manuscript. F.U. performed and designed ex vivo human experiments with EOC patient samples, discussed and interpreted results, and wrote the manuscript. S.H., R.S., T. Hüsser, and S.D. performed and designed ex vivo human tumor sample assays and discussed and interpreted the results. R.A. and J.C. developed and performed cell-based in vitro experiments. E.M. and P.B. contributed to the design of the molecules, and R.J.H. and T. Hofer developed FAP and CD19 antibodies, respectively. C.J. was responsible for protein crystallization. J.B., C.J., P.R., H.S., and M.L. designed, coordinated, performed, and analyzed the x-ray and NS-EM experiments. M.P. and S. Chen performed, designed, and analyzed confocal microscope assays. P.L.B.M., C. Küttel, and V.N. performed and analyzed immunohistochemistry staining. M.C.B. and A.F.-G. performed setup and NCS-Bz-Df-FAP–4-1BBL labeling. R.G. performed biochemical experiments. A.O. and C.P. performed cloning and purification of molecules. J.M.C., G.D., C.T.W., and K.K.S.S. planned, organized, supervised, and performed the noninvasive immunoPET/CT imaging study in the CRC-bearing rhesus monkey. D.L.C. performed and analyzed immunohistochemistry of rhesus monkey tissues. A.G.B.-R. provided the CRC-bearing rhesus monkey. M.H. and M.C. planned, organized, supervised, and analyzed the radioactive labeling of FAP–4-1BBL and noninvasive immunoPET/CT imaging. A.M.G., F.C., W.D., and P.N.M. planned and supervised dose finding, pharmacokinetics, and toxicity studies. V.L. supervised, discussed, and interpreted results of the initial FAP–4-1BBL project. M.A. and S.G.-R. performed initial experiments, designed assays, and discussed and interpreted the results. T.v.H., S.T., and M.M. developed up- and downstream processing and analytics. S. Colombetti coordinated and supervised in vivo mouse experiments and discussed and interpreted the results. T.F., M.B., and V.T. discussed and interpreted results. V.H.-S. as medical surgeon provided EOC samples. A.Z. designed and supervised ex vivo human EOC experiments with patient samples, discussed and interpreted results, and wrote the manuscript. C. Klein and P.U. initiated and oversaw research, discussed and interpreted results, and wrote the manuscript. P.U. initially proposed the concept. All authors reviewed and approved the final manuscript. Competing interests: All authors were Roche employees at the time of study conduction, except the authors affiliated with the University of Basel, Wake Forest School of Medicine, and Caribbean Primate Research Center, whose work was funded by Hoffmann–La Roche. Roche authors declare ownership of Roche stock (options). M.C. received consulting/advisor fees from Pfizer Women’s Health Advisory Board (2014/15). A.Z. received consulting/advisor fees from Bristol-Myers Squibb, Merck Sharp & Dohme, Hoffmann–La Roche, NBE Therapeutics, Secarna, ACM Pharma, and Hookipa. M.C. received honoraria from the American College of Veterinary Pathologists (2018), the Society for Toxicologic Pathology (2019), the National Toxicology Program (2018), and the University of Arkansas for Medical Sciences (2018). A.Z. received honoraria from Bristol-Myers Squibb, Merck Sharp & Dohme, Hoffmann–La Roche, NBE Therapeutics, Secarna, ACM Pharma, and Hookipa. M.C., H.S., and A.Z. maintain noncommercial research agreements with Hoffmann–La Roche. A.Z. maintains further noncommercial research agreements with NBE Therapeutics, Secarna, ACM Pharma, Hookipa, Crescendo, and Beyondsprings. M.C. was invited as speaker to Hoffmann–La Roche and received partial travel reimbursement from Hoffmann–La Roche (March 2017). M.C. and H.S. received research grants and funds from Hoffmann–La Roche. The Postdoctoral position of F.U. was partially financed by research grants from the Swiss National Science Foundation and research funds from Hoffmann–La Roche (hold by A.Z.). M.A., P.B., C.C., C.F., S.G.-R., C. Klein, E.M., S.L., V.L., and P.U. are inventors on patent applications (WO2016075278, WO2018114754, and WO2018114748) held/submitted by F. Hoffmann La Roche AG that cover TA–4-1BBLs and their combination therapy. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. All materials are available from Roche Glycart AG (Roche Innovation Center Zurich) under a material transfer agreement (MTA). Coordinates for human 4-1BBL are deposited with RCSB under the Protein Data Bank (PDB) ID 6FIB.
View Abstract

Navigate This Article