An Fc-free EGFR-specific 4-1BB-agonistic Trimerbody Displays Broad Antitumor Activity in Humanized Murine Cancer Models without Toxicity

Modulating immune responses using mAbs is one of the most promising approaches for cancer immunotherapy (1). Probably most well known is the mAb-mediated blockade of the programmed cell death protein 1 (PD-1) inhibitory pathway, which prevents PD-1–mediated immunosuppressive signaling in T cells and can restore effector functions to anergic tumor-infiltrating T cells (2). PD-1/PD-ligand 1 (PD-L1) axis blockade has shown long-term durable responses in a wide range of cancers, but their efficacy is limited to 10% to 30% of patients (3). Another immunotherapeutic approach involves the stimulation of costimulatory receptors, such as 4-1BB, with agonistic mAbs (4). 4-1BB, also known as CD137, is a member of the TNF receptor (TNFR) superfamily which can be induced on a variety of leukocyte subsets. 4-1BB is a type I single-pass transmembrane receptor with four extracellular cysteine-rich domains (CRD) and an intracellular signaling domain (5). On T cells, 4-1BB is expressed following activation through the T-cell receptor (TCR). Binding of its natural ligand [4-1BB-ligand (4-1BBL), TNFSF9] or agonistic mAbs enhances T-cell proliferation and effector functions (6–8), prevents T-cell exhaustion (8), protects from programmed cell death (9, 10), and promotes memory cell differentiation, which may support persistence of tumor-specific T cells (11). Anti-4-1BB–agonistic mAbs have been explored in preclinical cancer models and shown to promote rejection of a range of poorly immunogenic tumors (12–14). However, off-tumor toxicity have been the major impediment to the clinical development of full-length anti-human 4-1BB (anti-hu4-1BB)-agonistic IgGs, and several studies suggest that the toxicity is mainly dependent on Fc–FcγR interactions (15–17). The anti-hu4-1BB human IgG₄ urelumab (BMS-663513) caused dose-dependent liver toxicity and was implicated in two deaths (18, 19). Subsequent studies revealed that lower doses reduced liver toxicity, but at the cost of efficacy (19). The anti-hu4-1BB human IgG₂ utomilumab (PF-05082566) has an improved safety profile relative to urelumab, but is also a less potent 4-1BB agonist (20).

New strategies are being actively sought to avoid the off-tumor toxicities associated with Fc–FcγR interactions while retaining the antitumor activity associated with 4-1BB costimulation. These approaches aim to confine 4-1BB costimulation to the tumor microenvironment and draining lymph nodes. We have recently described Fc-free tumor-specific trimerbodies targeting a tumor-associated antigen (TAA), such as EGFR (15) or carcinoembryonic antigen (CEA; ref. 21), and murine 4-1BB in an agonistic manner. Both trimerbodies were potent costimulators in vitro and the EGFR-targeted 4-1BB-agonistic trimerbody showed enhanced tumor penetration and powerful antitumor activity in immunocompetent mice, while alleviating the systemic cytokine production and T-cell–mediated liver toxicities that are associated with IgG-based 4-1BB agonists (15). More recently, we showed in a liver-specific human EGFR-transgenic immunocompetent mouse that systemic administration of anti-4-1BB–agonistic IgGs resulted in nonspecific immune stimulation and hepatotoxicity, whereas in mice treated with the Fc-free EGFR-specific 4-1BB-agonistic trimerbody no such immune-related adverse effects were observed (22).

Here, we generated and characterized a humanized EGFR-targeted 4-1BB-agonistic trimerbody (4-1BB^N/CEGFR), consisting of three anti-hu4-1BB single-chain antibody fragments (scFv) and three anti-human EGFR (huEGFR) single-domain antibodies (V_HH)_. The humanized 4-1BB^N/CEGFR is structurally similar to that of the mouse trimerbody (15), costimulates human T cells in vitro in the presence of huEGFR, and delayed the progression of an EGFR⁺ human colorectal cancer and triple-negative breast cancer (TNBC) cell line–derived xenografts (CLDX) and a patient-derived xenograft (PDX) of EGFR⁺ non–small cell lung cancer (NSCLC), as monotherapy in immune-reconstituted mice. Furthermore, the combination of 4-1BB^N/CEGFR with the ICB atezolizumab significantly improved the antitumor immune response, with a near-complete inhibition of tumor growth in humanized mice bearing aggressive EGFR⁺ PD-L1⁺ human TNBC CLDX.

NOD.Cg-Prkdc^SCIDIL2rg^tm1Wjl/SzJ (NSG) female mice were supplied by Charles River Lboratories, Hsd:athymic Nude-Foxn1^nu female mice were supplied by Envigo RMS SPAIN S.L., and 129S4-Rag2tm1.1Flv Il2rgtm1.1Flv/J (Rag2^−/− IL2Rγ null) female mice were bred in the animal facility of CIMA. Animals were maintained under specific-pathogen-free condition with daily cycles of 12 hours light/12 hours darkness, and sterilized water and food were available ad libitum. All animal procedures conformed to European Union Directive 86/609/EEC and Recommendation 2007/526/EC, enforced in Spanish law under RD 1201/2005. Animal protocols were approved by the respective Ethics Committee of Animal Experimentation of the participant institutions (IDIPHISA, imas12, CIEMAT and CIMA); they were performed in strict adherence to the guidelines stated in the International Guiding Principles for Biomedical Research Involving Animals, established by the Council for International Organizations of Medical Sciences. The experimental study protocols were additionally approved by local government (PROEX 094/15, 108/15, 076/19, and 166/19).

Commercially available antibodies used in the experiments are listed in Supplementary Table S1. Recombinantly produced antibodies are listed in Supplementary Table S2. HEK293 (CRL-1573), MDA-MB-231 (HTB-26), A431 (CRL-1555), NIH/3T3 (CRL-1658), and CHO-K1 (CCL-61) cells were obtained from the ATCC and cultured in DMEM (Lonza) supplemented with 2 mmol/L l-glutamine, 10% (vol/vol) heat-inactivated FCS (Merck Life Science), and antibiotics (100 units/mL penicillin, 100 mg/mL streptomycin; all from Life Technologies) referred as to DMEM complete medium (DCM), at 37°C in 5% CO₂ humidity. NIH/3T3 cells expressing huEGFR (3T3^huEGFR; ref. 23) were kindly provided by A. Villalobo (IIBm). The hu4-1BB–expressing HEK293 cell line (HEK239^hu4–1BB) was generated by transfection with the expression vector pCMV3-Flag-TNFRSF9 (SinoBiological) and selected in DCM with 500 μg/mL G418 (Life Technologies). CHO-K1 Cells expressing human FcγRIIb (CD32) were from Promega (No. JA2251). All cell lines were used within 3 months of thawing and checked for Mycoplasma using PCR every month using the Mycoplasma Plus TM Primer Set (Biotools B&M Labs).

To generate the SAP3.28 scFv-based N-terminal trimerbody, the DNA fragments encoding the FLAG-strep II-SAP3.28^HL (V_H-linker-V_L) scFv was synthesized by Geneart AG and subcloned as HindIII/NotI into the expression vector pCR3.1-MFE23^N (24) resulting in pCR3.1-FLAG-strepII-SAP3.28^HL-N-myc/His. The C-terminal myc/His tag-sequence was removed by PCR from the plasmids with Fw-CMV and Stop-XbaI-Rev primers (Supplementary Table S3). The Flag-strep II-SAP3.28^HL scFv gene was subcloned as HindIII/NotI into a vector containing the human collagen XVIII-derived homotrimerization (TIE^XVIII) domain and the anti-human EGFR single-domain antibody (V_HH; EGa1; ref. 25), resulting in the bispecific trimerbody-expressing vector pCR3.1-FLAG-strepII-SAP3.28^HL-N18/C18EGa1. All the sequences were verified using primers FwCMV and RvBGH (Supplementary Table S3).

4-1BB–dependent activation of activated NFκB assay was performed on thaw-and-use GloResponseNFkB-luc2/4-1BB Jurkat cells (Promega, No. JA2351) according to the manufacturer's instructions (Supplementary Materials and Methods).

Human PBMCs or isolated T cells (Supplementary Materials and Methods; 1.5 × 10⁵ cells/well) were plated in triplicate in flat bottom 96-well plates, in RPMI supplemented with 10% FCS and 50 μmol/L β-mercaptoethanol (Life Technologies) and cocultured with 45 Gy irradiated target cells (3T3 or 3T3^hEGFR) at an effector/target ratio of 5:1. The anti-hu4-1BB agonists antibodies and controls were added at 10-fold serial dilutions in the presence of anti-huCD3 (OKT3) mAb at 0.05 μg/mL. After 72 hours, cell-free supernatants were analyzed by ELISA for cytokine secretion. Irradiated EGFR⁺PD-L1⁻ cells (3T3^huEGFR) or EGFR⁺PD-L1⁺ cells (MDA-MB-231; 3 × 10⁴ cells/well) were seeded with huPBMCs (1.5 × 10⁵ cells/well), activated with anti-huCD3 at 0.05 μg/mL, in the presence of anti-PD-L1 (atezolizumab) alone (10 μg/mL) or combined with 4-1BB^N/CEGFR (1 μg/mL). Cell-free supernatants were measured for IFNγ after 72 hours by ELISA (Diaclone, No. 851560005).

HT29 cells (1 × 10⁶) were implanted subcutaneously into the dorsal space of 6-week-old Rag2^−/− IL2Rγ null female mice, followed by the intraperitoneal infusion of freshly huPBMCs (1 × 10⁷ cells/mouse). Tumor growth was monitored by caliper measurements three times a week, and when tumors reached approximately 0.4 cm in diameter, mice were randomized to receive treatment (n = 7–8/group). Measurements were conducted in a random order by the investigator who was blinded to the treatment assignment. Mice were treated every 3 days with five intraperitoneal injections of CEA^N or 4-1BB^N/CEGFR trimerbodies (4 mg/kg) or every week with three intraperitoneal injections of 4-1BB IgG (4 mg/kg). MDA-MB-231 cells (2 × 10⁶) were resuspended in PBS and mixed with matrigel (30%). Cells were implanted subcutaneously on right dorsal flank of 6-week-old NSG female mice, followed by an intraperitoneal injection of freshly isolated huPBMCs (1 × 10⁷cells/mouse). Tumor growth was monitored by caliper measurements three times a week. Tumor-bearing mice (0.2 cm diameter) were randomly divided into four groups (n = 5–6/group) and the investigator was blinded for treatment allocation. Mice were treated every three days with five intraperitoneal injections of 4-1BB^N/CEGFR trimerbodies (4 mg/kg), or every week with three intraperitoneal injections of PD-L1 IgG (4 mg/kg), alone or in combination. Mice weights were measured twice a week to monitor toxicity. Mice were euthanized at any sign of distress and/or due to 10% to 15% of weight loss.

For this study, the previously amplified lung PDX TP103 was selected according to its histologic type, genetic background (EGFR and TP53 mutated), and huEGFR cell surface expression (26). Tumors were cut into ≈50-mm³ pieces, and implanted subcutaneously through a tiny incision into the dorsal space of anesthetized 6-week-old NSG female mice. Tumor growth was monitored by caliper measurements every 3 to 4 days, and when tumors reached approximately 0.5 cm in diameter, mice were randomized into groups (n = 6–7/group) with similar mean tumor sizes and SDs, and freshly isolated huPBMCs (1 × 10⁷ cells/mouse) from healthy donors were intraperitoneally infused. Mice were treated every 3 days with five intraperitoneal injections of 4-1BB^N/CEGFR (4 mg/kg). Mice weights were measured once a week to monitor toxicity. Mice were euthanized when the weight loss was ≥10% to 15%, when tumor size reached a diameter of 1.0 cm any dimension, when tumors ulcerated, or at any sign of mouse distress.

Statistical analysis was performed using GraphPad Prism Software version 6.0. In general, the in vitro experiments were done in triplicates and values are presented as mean ± SD from one of at least three separate experiments. Significant differences (P value) were discriminated by applying a two-tailed, unpaired Student t test assuming a normal distribution. P values are indicated in the corresponding figures for each experiment. EC₅₀ were calculated using a nonlinear regression curve (log agonist vs. normalized response-variable response). Mean tumor volume are presented for each group using a scatter plot as mean ± SD. To assess the differences between treatment groups, P values were determined by one-way ANOVA adjusted by the Bonferroni correction for multiple comparison tests.

Anti-hu4-1BB trimerbodies were generated using scFv-encoding genes derived from the anti-hu4-1BB-agonistic SAP3.28 mAb (Supplementary Fig. S1A), which binds to hu4-1BB CRD-1 (27). The SAP3.28 IgG (hereafter referred to as 4-1BB IgG) is a chimeric molecule displaying a humanized V_L domain and a partially humanized V_H domain that preserves the murine FR3 region to retain antigen binding, and the Fc region of murine IgG₁ (27). Like urelumab, which recognizes the N-terminus of CRD-1 (28), 4-1BB IgG does not block the hu4-1BB receptor/hu4-1BBL interaction (Supplementary Fig. S2A–S2C). Furthermore, we showed that the epitopes of 4-1BB IgG and urelumab do not overlap (Supplementary Fig. S2D and S2E).We designed a SAP3.28 scFv-based anti-hu4-1BB N-terminal trimerbody (4-1BB^N) by fusing the SAP3.28 scFv to the human collagen XVIII-derived homotrimerization (TIE^XVIII) domain by a flexible linker (Supplementary Fig. S1B and S1C), and a bispecific trimerbody by fusing the anti-EGFR EGa1 V_HH antibody (25) to the C-terminus of the 4-1BB^N to generate the construct called 4-1BB^N/CEGFR (Fig. 1A). Both trimerbodies were purified from conditioned medium from stably transfected HEK293 cells by Strep-Tactin affinity chromatography, with proteins yields (3.5 and 4.5 mg/L, respectively) that were > 95% pure (Supplementary Fig. S3A). Mass spectrometry (using MALDI-TOF, not shown) confirmed the absence of the signal sequences in the purified antibodies. SEC-MALS experiments on both 4-1BB^N and 4-1BB^N/CEGFR yielded major peaks with molar masses of 111 and 160 kDa, respectively (Supplementary Fig. S3B and S3C), which are consistent with trimeric molecules. Minor peaks at smaller volumes with molar masses of 217 and 340 kDa indicate the presence of dimers of trimers, as previously observed for other trimerbodies (15). Circular dicroism measurements show predominant b-sheet structures and cooperative thermal denaturations (T_m≈ 60°C; Supplementary Fig. S3D and S3E). Small angle X-ray scattering (SAXS) was used to study the three-dimensional structure of both trimerbodies. The 4-1BB^N trimerbody shows a flat distribution, with a well-defined TIE^XVIII core in the center and the scFvs partially extended on the same plane, like the spokes on a wheel (Supplementary Figs. S4 and S5; Supplementary Table S4) The 4-1BB^N/CEGFR trimerbody maintains the same planar configuration of 4-1BB^N with its additional small-sized EGFR V_HH domains interspersed between the 4-1BB scFvs to resemble a six-bladed ninja star (Fig. 1B; Supplementary Fig. S5; Supplementary Table S4).

Biolayer interferometry (BLI) was used to measure the association and dissociation kinetics of 4-1BB^N and 4-1BB^N/CEGFR binding to hu4-1BB, and of 4-1BB^N/CEGFR and the anti-EGFR ATTACK antibody (29) binding to huEGFR (Fig. 1C). The bispecific ATTACK antibody is an evolution of the tandem trimerbody format (30) which combines three EGFR-binding V_HH antibodies with a single CD3-binding scFv (29). All interactions were of high affinity (with low picomolar K_D values), indicating functional trivalence of the trimerbodies toward the antigens displayed on a biosensor surface (Supplementary Table S5). The kinetics of huEGFR binding by these trivalent antibodies is consistent with previous studies (15, 29). In a complementary experiment, 4-1BB^N and the 4-1BB^N/CEGFR were first loaded onto hu4-1BB immobilized on the surface of biosensors, which were then transferred into buffer containing huEGFR. 4-1BB^N/CEGFR, but not 4-1BB^N, was able to bind soluble huEGFR while remaining bound to the immobilized hu4-1BB, further confirming its bivalence and its capability to bind both antigens simultaneously (Fig. 1D). Furthermore, 4-1BB^N/CEGFR bound to mouse (mo-), cynomolgus (cy-) and huEGFR, as well as to cy4-1BB and hu4-1BB, but to a much lower extent to mo4-1BB (Supplementary Fig. S6A and S6B). Their ability to detect hu4-1BB and huEGFR in a cellular context was analyzed by flow cytometry. The 4-1BB^N/CEGFR trimerbody bound to wild-type HEK293 (EGFR⁺) cells, to HEK293 cells transfected to express hu4-1BB on their cell surface (HEK293^hu4-1BB), and to mouse 3T3 cells expressing huEGFR (3T3^huEGFR) but not to wild-type 3T3 cells (Supplementary Fig. S7). In contrast, the 4-1BB IgG only bound HEK293^hu4-1BB cells. To further assess the multivalent binding of 4-1BB^N/CEGFR, we studied its capacity to inhibit proliferation and EGFR phosphorylation in A431 cells (25). Both 4-1BB^N/CEGFR and cetuximab, an EGF-competitive inhibitor (31), but neither the anti-human CD20 rituximab nor the parental 4-1BB IgG, inhibited A431 proliferation, in a dose-dependent manner (P = 0.003 and P = 0.0005, respectively, for the higher doses of both antibodies, vs. equimolar doses of control antibodies), as well as EGFR phosphorylation (Supplementary Fig. S8A and S8B).

The agonist activities of the three SAP3.28-derived antibodies and urelumab were assessed using NFκB-luc2/4-1BB Jurkat cells (Jurkat^NFκB) that constitutively express hu4-1BB on the cell surface and a luciferase reporter driven by a NFκB response element. Jurkat^NFκB reporter cells were cocultured with target cells stably expressing either huFcγRIIb (CHO^huFcγRIIb) or huEGFR (3T3^huEGFR), as well as nontransfected CHO or 3T3 cells as negative controls; the expression of cell surface huFcγRIIb and huEGFR were demonstrated by flow cytometry (Fig. 2A and B). Titrations of bivalent (4-1BB IgG or urelumab), or trivalent (4-1BB^N or 4-1BB^N/CEGFR) anti-hu4-1BB antibodies were then added to the cocultured cells. In the absence of Fc- or EGFR-mediated antibody cross-linking at the target cell surface (i.e., in cocultures with nontransfected CHO or 3T3 cells), 4-1BB IgG showed little to no induction over untreated Jurkat^NFκB cells at all tested concentrations, both anti-hu4-1BB trimerbodies showed an approximately 10-fold induction, and urelumab showed an approximately 20-fold induction (Fig. 2C and D). In the presence of FcγRIIb-mediated cross-linking (i.e., using CHO^huFcγRIIb as target cells), 4-1BB IgG induced a NFκB dose-dependent activation with a 26-fold induction (P = 0.0008) and urelumab's induction was further increased to 40-fold (P = 0.003; Fig. 2C). Neither trimerbody showed a FcγRIIb-mediated increase in induction (Fig. 2C). The trimerbody-mediated 4-1BB signaling was significantly strengthened when target cells expressed huEGFR (P = 0.0008), leading to a 40-fold increase of NFκB luciferase reporter activity (Fig. 2D). Induction by 4-1BB IgG, urelumab, and 4-1BB^N was not affected by huEGFR expression (Fig. 2D). The negative control antibodies moIgG1, huIgG4, and CEA^N, a trimerbody recognizing CEA, showed no activation (Supplementary Fig. S9A and S9B).We then used huPBMCs or T cells from healthy donors to investigate the effect of the anti-hu4-1BB antibodies on IFNγ secretion when cocultured with irradiated 3T3 or 3T3^huEGFRcells, both with and without a suboptimal dose of anti-huCD3 mAb. The 4-1BB^N/CEGFR trimerbody had a dose-dependent activating effect on IFNγ secretion only when huPBMCs or T cells were cocultured with EGFR⁺ cells; no induction was observed with EGFR⁻ cells (Fig. 2E and F). Under these conditions, the effect of 4-1BB IgG and CEA^N was minimal and independent of EGFR expression (Fig. 2E; Supplementary Fig. S10). These data show that 4-1BB^N/CEGFR induces strong, EGFR-dependent T-cell costimulation and IFNγ secretion that requires initial signaling through the TCR/CD3 complex (signal 1). Subsequently, huPBMCs were cocultured with irradiated EGFR⁺PD-L1⁻ (3T3^huEGFR) or EGFR⁺PD-L1⁺ (MDA-MB-231) cells (Fig. 2G) in the presence of 4-1BB^N/CEGFR and the PD-L1–blocking antibody atezolizumab. When combined with a suboptimal dose of anti-huCD3 mAb, the 4-1BB^N/CEGFR trimerbody significantly enhanced IFNγ secretion (P = 0.0007 3T3^huEGFR cells; P = 0.0002 MDA-MB-231 cells; Fig. 2H). The addition of atezolizumab significantly increased IFNγ levels when huPBMCs were cocultured with MDA-MB-231 cells in the presence of 4-1BB^N/CEGFR (P = 0.02; Fig. 2H).

The 4-1BB^N/CEGFR trimerbody retained close to 100% of its initial binding activity after 4 days in human serum at 37°C (Supplementary Fig. S11A and S11B). Chelation with p-SCN-Bn-Deferoxamine (Df) of the 4-1BB^N/CEGFR trimerbody did not alter its SDS-PAGE migration pattern nor compromise its binding activity (Supplementary Fig. S12A and S12B). After radiolabeling, the RCY (radiolabeling yield) and RQP (radiochemical purity) of purified [⁸⁹Zr]Zr-Df-4-1BB^N/CEGFR were 40% and 95%, respectively. The AIC values were 10.97 and −22.66 for one and two compartment of [⁸⁹Zr]Zr-Df-4-1BB^N/CEGFR, respectively; thus, the disposition of the 4-1BB^N/CEGFR trimerbody was better explained through a bicompartmental model (Supplementary Table S6). After intravenous administration, the elimination of [⁸⁹Zr]Zr-Df-4-1BB^N/CEGFR was biphasic, with a half-time of 7.3 hours for the rapid distribution phase and 66.8 hours for the slow distribution phase (Fig. 3A). The volume of distribution at steady state was 66.5 mL (2.63 L/kg) and the plasma clearance 0.97 mL/hour (37.6 mL/kg/hour). As the blood-to-plasma ratio was 0.62, the blood clearance value obtained was very low (0.062 L/kg/hour) compared with the cardiac output (21.7 L/kg/hour in mouse), which is generally desirable for developing a drug with a low dosage regimen (32).

We tested the 4-1BB^N/CEGFR trimerbody for antitumor activity in huPBMC-driven humanized immunoavatar mouse models. Rag2^−/− IL2Rγ^null mice were intraperitoneally injected with huPBMCs and then human HT-29 colorectal cancer cells were subcutaneously inoculated (Fig. 3B). Transferred human T cells become activated and develop pathogenic xeno-reactivity, a process called xenograft-versus-host disease (xGVHD; ref. 33), which is a valuable model for testing immunomodulatory strategies, where the engrafted human T cells are amenable for modulation by therapeutic agents (34–36). When tumors reached approximately 0.4 cm in diameter, mice were treated with five trimerbody (CEA^N or 4-1BB^N/CEGFR) intraperitoneal injections at 3/4-day intervals, or three weekly equimolar doses of 4-1BB IgG, as depicted in Fig. 3B. The dose and treatment schedule was designed in a similar way to what was conducted with the anti-mo4-1BB agonists in an immunocompetent model of colorectal cancer (15). The 4-1BB^N/CEGFR-treated group showed a significantly slower tumor growth compared with the untreated group (P = 0.01), and the CEA^N-treated groups (P = 0.004; Fig. 3C). Notably, the humanized 4-1BB^N/CEGFR trimerbody provided antitumor activity in vivo comparable with the 4-1BB IgG (Fig. 3C).

We next sought to determine whether the antitumor effect would also occur in an EGFR⁺ NSCLC PDX-bearing huPBMC-driven humanized NSG mice model (TP103; Fig. 3D and E). As shown in Fig. 3F, the 4-1BB^N/CEGFR-treated mice showed a reduced tumor growth compared with the control group. The improved tumor growth control was accompanied of significant changes in the tumor-infiltrating lymphocyte (TIL) infiltration pattern. In both groups, a diffuse infiltration of CD3⁺ T lymphocytes surrounding and involving tumor cell nests was detected (Supplementary Fig. S13). In the PBS-treated mice, there was a prevalence of CD4⁺ T cells with a CD4/CD8 ratio of 2.8 (Fig. 3G–I). In the 4-1BB^N/CEGFR-treated group, a significant increase in the number of CD8⁺ T cells (P = 0.04) was observed, accompanied by a reduction in the number of Foxp3⁺ cells (P = 0.01; Fig. 3G–I; Supplementary Fig. S13).

We compared the toxicity profile in huPBMC-driven humanized NSG mice treated with 4-1BB IgG or 4-1BB^N/CEGFR trimerbody (6 mg/kg) once a week for 3 weeks and euthanized 1 week later. The histologic study of the livers revealed that 4-1BB IgG treatment exacerbated xGVHD. Details of the liver infiltration in a representative mouse of each group of treatment are depicted in Fig. 3J, showing extensive perivascular mononuclear cell infiltration in the group treated with the IgG-based 4-1BB agonist. We then studied the concentrations of human IFNγ in serum samples collected at sacrifice. 4-1BB IgG treatment significantly increase IFNγ levels over 4-1BB^N/CEGFR treatment (P = 0.001), where the levels were comparable with PBS-treated animals (Fig. 3K).

The therapeutic potential of combining 4-1BB^N/CEGFR with the PD-L1 blocker atezolizumab was investigated in huPBMC-driven humanized NSG mice bearing human EGFR⁺PD-L1⁺ MDA-MB-231 (Fig. 2G) TNBC xenografts (Fig. 4A). Atezolizumab monotherapy was able to reduce tumor growth by approximately 60%, while 4-1BB^N/CEGFR monotherapy showed an approximately 90% tumor growth reduction (Fig. 4B). The combination of atezolizumab plus 4-1BB^N/CEGFR resulted in an additional decrease in tumor growth (Fig. 4B). In the PBS-treated group, large nests of neoplastic pleomorphic cells with intense cytokeratin (CK) expression with dense lymphocyte infiltration (Fig. 4C and D) were observed. Importantly, the percentage of CK⁺ cells was significantly lower in the 4-1BB^N/CEGFR monotherapy group (P = 0.04) and in the combination therapy group (P = 0.0002) than in atezolizumab monotherapy group (Fig. 4C). With combination therapy, the percentage of CK⁺ cells was at most 30% in 5 of 6 mice and in one mouse, TNBC cells were completely eradicated (Fig. 4E). This reduction in tumor burden was associated with a significantly increased proportion of CD8⁺ T cells in the 4-1BB^N/CEGFR-treated groups (P = 0.03 and P = 0.04; Fig. 4D and E).

Immune checkpoint receptors, both coinhibitory and costimulatory, are membrane molecules expressed by immune cells that regulate the activation and effector functions of T cells (37). These regulatory receptors can be manipulated by the exogenous administration of antibodies to enhance preexisting antitumor immunity (38). The blockade of inhibitory checkpoints, such as CTLA-4 and PD-1/PD-L1, with antagonistic mAbs have shown remarkable efficacy in several types of cancer, with manageable toxicity profiles; however, their overall response rate remains around 30% (39). Agonistic antibodies targeting costimulatory checkpoints, such as 4-1BB, OX40, CD40, GITR, and ICOS, are able to bias T cells toward an effector outcome and overcome anergy-inducing immunosuppressive signaling in the tumor microenvironment, thus providing strong rationale to be combined with ICB (38). However, despite significant interest and effort, no such antibody has yet received regulatory approval. Among them, anti-4-1BB mAbs have shown robust antitumor activity in preclinical models (12). However, the clinical development of full-length IgG anti-hu4-1BB-agonistic mAbs is facing serious challenges due to low efficacy (utomilumab) or severe hepatotoxicity (urelumab; ref. 18). We recently generated Fc-free tumor-targeted murine 4-1BB-agonistic trimerbodies that induced effective antitumor immunity without liver toxicity in immunocompetent mice (15).

Here, we characterize a Fc-free bispecific humanized trimerbody which binds to human 4-1BB and EGFR. This 4-1BB^N/CEGFR trimerbody is efficiently produced and structural studies showed that the trimerbody primarily forms the intended trimeric structure, with protein folding and configuration nearly identical to that of the murine trimerbody (15). The antibody domains are positioned around the human collagen XVIII homotrimerization domain in a hexagonal configuration, and the binding studies provided quantitative evidence for multivalent interactions with both human 4-1BB and EGFR. These results demonstrate the robustness of the trimerbody scaffold to generate functional multivalent and multispecific molecules with a predictable and well-defined structure.

The binding of three TNFRSF receptors to a single trimeric ligand nucleates receptor clustering to induce signaling, but multiple complexes are required for signaling to reach effective levels (40). Anti-4-1BB–agonistic mAbs can be classified as either strong or weak agonists. A strong agonist (e.g., urelumab) can induce signaling activation without FcγR-mediated cross-linking, while a weak agonistic (e.g., utomilumab) requires FcγR-mediated cross-linking to meaningfully induce 4-1BB signaling (41). Here, we demonstrate in a hu4-1BB–reporting cell line that a bivalent (IgG) anti-hu4-1BB antibody derived from the SAP3.28 antibody (27) is dependent on the presence of FcγRIIb to induce 4-1BB signaling and can therefore be classified as weak agonists. However, SAP3.28-derived trimerbodies induce partial 4-1BB signaling even without additional cross-linking, which emphasizes the relevance of trimerbody valence and stoichiometry in the context of agonizing a multimerizing receptor. Importantly, in the presence of EGFR-expressing cells, the trimerbody-4-1BB complexes are further cross-linked, resulting in increased agonistic activity that significantly exceeds that achieved by a 4-1BB IgG cross-linked by FcγRIIb-expressing cells, and was similar to that observed with FcγRIIb-cross-linked urelumab.

When the study was conducted on activated huPBMCs or isolated T cells, the 4-1BB^N/CEGFR trimerbody did not increase IFNγ secretion above the basal levels, at any of the concentrations analyzed, in the absence of EGFR-specific cross-linking. These results demonstrate that, in contrast to the results from the Jurkat cell-based hu4-1BB reporter assay, under near physiologic conditions using activated primary T cells the trimerbody-mediated 4-1BB clustering does not provide effective 4-1BB costimulation without additional EGFR-mediated cross-linking. This aspect is particularly relevant, as it shows that effective costimulation is not induced, despite saturating binding of the trimerbody to 4-1BB. This has important implications with regard to off-tumor safety issues, as these results indicate that the trimerbody is not capable of inducing 4-1BB costimulation in TAA-negative tissues.

The humanized EGFR-targeted trimeric 4-1BB-agonistic trimerbody exhibits improved serum stability and a circulatory half-life of nearly 3 days. We hypothesize that strategies aiming to reinvigorate preexisting tumor-specific exhausted T cells (42) could benefit from an intermittent boosting strategy, to reduce systemic exposure and potential toxicity. This may be especially important when using a relatively ubiquitous TAA, such as EGFR, for targeted 4-1BB costimulation. The relatively short half-life combined with the TAA-targeted approach would allow the selective accumulation of the 4-1BB^N/CEGFR trimerbody in the tumor area and in tumor cell–infiltrated lymph nodes. In fact, it has been shown that EGFR expression is related with lymph node involvement and tumor grade in colorectal cancer. Also, lymph node–involved colorectal cancers showed higher scores of EGFR staining than control groups (43).

We have recently shown that treatment of immunocompetent transgenic mice expressing huEGFR in the liver (ΔEGFR-tg) (44) with IgG-based anti-mouse 4-1BB agonist resulted in nonspecific immune stimulation and hepatotoxicity (22). In contrast, none of these features were observed in ΔEGFR-tg mice treated with the Fc-free EGFR-specific anti-mouse 4-1BB-agonistic 1D8^N/CEGa1 trimerbody (22), despite the fact that the anti-EGFR EGa1 V_HH recognize huEGFR and moEGFR. These results further validate the safety profile of Fc-free trimerbodies in systemic cancer immunotherapy protocols. Here, we demonstrated that treatment of huPBMC-driven immunoavatar mice (36) with the anti-4-1BB–agonistic IgG resulted in enhanced activation of adoptively transferred human T cells and exacerbation of hepatic xGVHD. In contrast, treatment with the 4-1BB^N/CEGFR trimerbody induced human IFNγ serum levels and liver infiltration similar to that observed in PBS-treated animals.

Despite the limitations of current mouse models for the study of human tumors, such as the development of xGVHD, and insufficient engraftment of some human immune subsets (45), we demonstrated that in huPBMC humanized immunoavatar mouse models of human colorectal cancer and TNBC, 4-1BB^N/CEGFR monotherapy provided significant antitumor activity. This effect was confirmed in a humanized PDX model of human NSCLC where treatment with 4-1BB^N/CEGFR showed a significant reduction in tumor growth, which was associated with a significant increase in the percentage of CD8⁺ TIL and a substantial improvement of the CD8⁺ T cell/Treg (regulatory T cell) ratio, from 7.5 in the control-treated mice to 50 in the 4-1BB^N/CEGFR-treated mice. EGFR can be an effective target for the development of a broadly applicable tumor-specific 4-1BB–mediated immunotherapy. In most solid tumors (including lung, colorectal, prostate, pancreatic, head and neck, liver, renal, urothelial, and endometrial cancers, along with glioblastoma) more than 50% of patients display a moderate to strong expression of EGFR (46). Furthermore, in humanized mice bearing aggressive EGFR⁺ TNBC CLDX expressing high levels of PD-L1, the combination of 4-1BB^N/CEGFR with atezolizumab further improved antitumor activity, resulting in 1 of 6 mice undergoing complete regression.

The primary antitumor mechanism of anti-4-1BB IgG₁ mAbs is the activation of CD8⁺ T cells after coengagement of the inhibitory FcγRIIb receptor (47). However, FcγRIIb interactions in the liver are responsible for the liver toxicity of anti-4-1BB IgG_2a mAbs (48–50), which primarily act by depleting Tregs through interactions with activating FcγR receptors (47). However, these activating FcγR interactions and the subsequent FcγR-mediated depletion of 4-1BB⁺cells may compromise antitumor immunity (41). These findings outline a very intricate scenario in which the number of CD8⁺ TILs, Tregs, and FcγR⁺ cells, along with the relative abundances of activating and inhibitory FcγRs, are likely to determine patient outcomes and must be considered before deciding whether full-length IgG anti-4-1BB agonists are indicated (51). An additional consideration is the competition between administered therapeutic mAbs and endogenous Igs for FcγRs, which will affect the Fc receptor binding of anti-4-1BB agonists and thereby affect their mechanism of action (51). Recently, several IgG-based tumor-targeted 4-1BB agonists have been engineered with effector-silent Fc regions that retain FcRn-driven half-life extension while reducing binding to FcγRs (52, 53), indicating that it can be desirable for a therapeutic antibody to avoid these interactions. However, these mutations may affect mAb stability, and introduce potentially immunogenic sites (54). The trimerbody described in this study has no Fc region and thereby ensures completely FcγR-independent 4-1BB clustering and cross-linking, avoiding residual binding activities shown in “silenced” Fc regions (55), and may be more easily systematized and applied in a clinical setting.

Here, we show that the combination of a humanized tumor-specific Fc-free 4-1BB-agonistic trimerbody with an ICB mAb resulted in a greater therapeutic index compared with either monotherapy. These results demonstrate the benefits of combination therapies using both costimulatory and ICB immunotherapy strategies, and the suitability of the trimerbody platform for enacting costimulatory strategies with high efficacy while avoiding adverse reactions mediated by the Fc region.

M. Compte reports current employment at Leadartis. A. Erce-Llamazares reports current employment at LeadArtis. E.M. Garrido-Martin reports other from PharmaMar and personal fees from Bristol Myers Squibb and Pfizer outside the submitted work. C. Domínguez-Alonso reports grants from the Spanish Minister of Science and Innovation during the conduct of the study. M. Zonca reports grants from Spanish Ministry of Science, Innovation and Universities during the conduct of the study, and reports formal employment at Leadartis. I. Melero reports grants and personal fees from BMS, Roche, AstraZeneca, Bioncotech, Alligator, Pharmamar, and Genmab, and personal fees from F-Star, Numab, Gossamer, EMD, Amunix, and MSD outside the submitted work. L. Paz-Ares reports personal fees from BMS, AstraZeneca, MSD, Lilly, Roche, Blueprint, Bayer, Mirati, Angem, Jansen, Sanofi, and Pharmamar outside the submitted work. L. Sanz reports grants from Carlos III Health Institute during the conduct of the study and is a co-founder of Leadartis. L. Alvarez-Vallina reports grants from the Spanish Ministry of Science, Innovation and Universities, the CRIS Cancer Foundation, and the Spanish Association Against Cancer during the conduct of the study; in addition, L. Alvarez-Vallina has a patent for WO/2019/234187 licensed to Leadartis SL, and is a co-founder of Leadartis SL. No disclosures were reported by the other authors.

M. Compte: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. S.L. Harwood: Formal analysis, investigation, methodology, writing–review and editing. A. Erce-Llamazares: Validation, investigation. A. Tapia-Galisteo: Investigation, methodology. E. Romero: Investigation, methodology. I. Ferrer: Investigation, visualization, methodology. E.M. Garrido-Martin: Investigation, visualization, methodology. A.B. Enguita: Investigation, visualization, methodology. M.C. Ochoa: Investigation, visualization, methodology. B. Blanco: Validation, investigation, visualization, methodology. M. Oteo: Investigation, visualization, methodology. N. Merino: Investigation. D. Nehme-Álvarez: Investigation, visualization. O. Hangiu: Investigation. C. Domínguez-Alonso: Investigation. M. Zonca: Investigation, methodology. A. Ramírez-Fernández: Investigation. F.J. Blanco: Supervision, funding acquisition, investigation, methodology, writing–original draft. M.A. Morcillo: Conceptualization, supervision, funding acquisition, investigation, methodology, writing–original draft. I.G. Muñoz: Resources, supervision, funding acquisition, validation, investigation, methodology, writing–original draft. I. Melero: Conceptualization, resources, supervision, validation, investigation, writing–original draft. J.L. Rodriguez-Peralto: Resources, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–review and editing. L. Paz-Ares: Resources, supervision, funding acquisition, validation, investigation, visualization, writing–review and editing. L. Sanz: Conceptualization, supervision, funding acquisition, investigation, methodology, writing–review and editing. L. Alvarez-Vallina: Conceptualization, resources, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, writing–review and editing.

We thank M. Glennie and A. Villalobos for reagents, and Diamond Light Source (United Kingdom) for provision of synchrotron radiation at beamline B21. This work was supported by grants from the European Union [IACT Project (602262), H2020-iNEXT (1676)]; the Spanish Ministry of Science, Innovation and Universities and the Spanish Ministry of Economy and Competitiveness (SAF2017-89437-P, CTQ2017-83810-R, RTC-2016-5118-1, RTC-2017-5944-1), partially supported by the European Regional Development Fund; the Carlos III Health Institute (PI16/00357), co-founded by the Plan Nacional de Investigación and the European Union; the CRIS Cancer Foundation (FCRIS-IFI-2018); and the Spanish Association Against Cancer (AECC, 19084). C. Domínguez-Alonso was supported by a predoctoral fellowship from the Spanish Ministry of Science, Innovation and Universities (PRE2018-083445). M. Zonca was supported by the Torres Quevedo Program from the Spanish Ministry of Economy and Competitiveness, co-founded by the European Social Fund (PTQ-16-08340).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.