Carcinoembryonic Antigen (CEA)-Specific 4-1BB-Costimulation Induced by CEA-Targeted 4-1BB-Agonistic Trimerbodies

4-1BB (CD137) is an inducible costimulatory receptor that promotes expansion and survival of activated T cells; and IgG-based 4-1BB-agonistic monoclonal antibodies exhibited potent antitumor activity in clinical trials. However, the clinical development of those antibodies is restricted by major off-tumor toxicities associated with FcγR interactions. We have recently generated an EGFR-targeted 4-1BB-agonistic trimerbody that demonstrated strong antitumor activity and did not induce systemic inflammatory cytokine secretion and hepatotoxicity associated with first-generation 4-1BB agonists. Here, we generate a bispecific 4-1BB-agonistic trimerbody targeting the carcinoembryonic antigen (CEA) that is highly expressed in cancers of diverse origins. The CEA-targeted anti-4-1BB-agonistic trimerbody consists of three 4-1BB-specific single-chain fragment variable antibodies and three anti-CEA single-domain antibodies positioned around a murine collagen XVIII-derived homotrimerization domain. The trimerbody was produced as a homogenous, non-aggregating, soluble protein purifiable by standard affinity chromatographic methods. The purified trimerbody was found to be trimeric in solution, very efficient at recognizing 4-1BB and CEA, and potently costimulating T cells in vitro in the presence of CEA. Therefore, trimerbody-based tumor-targeted 4-1BB costimulation is a broadly applicable and clinically feasible approach to enhance the costimulatory environment of disseminated tumor lesions.


INTRODUCTION
In the last decade, immunostimulatory monoclonal antibodies (mAbs) have risen to prominence as a clinically viable approach to induce a tumor-specific immune response in cancer patients (1). The antagonistic mAbs used in immune checkpoint blockade are able to block T cell-inhibitory signaling from receptors such as cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) and programmed cell death-1 (PD-1), and have been successfully used in the treatment of several types of cancers (1). Another type of cancer immunotherapy uses agonistic mAbs to achieve the same effect, but from the opposite direction (2).The agonism of T cell costimulatory receptors can enhance T cell effector functions, proliferation, and survival, which has been demonstrated using several receptors from the tumor necrosis factor (TNF) receptor superfamily (TNFRSF). One such TNFRSF member, 4-1BB (CD137, TNFRSF9) (3), forms the basis for the immunostimulatory approach used in this article.
In addition to activated T cells, 4-1BB is expressed on dendritic cells, NK cells, B cells, and endothelial cells. It is only expressed and present on T cells for a short interval occurring after TCR engagement, and has a single identified ligand, 4-1BBL, found on antigen presenting cells (4). The engagement of 4-1BB by its ligand or agonistic mAbs provides substantial boosts to the T cell response (4), which prompted the incorporation of 4-1BB intracellular signaling domain into TCR-like chimeric antigen receptors (CARs) and ultimately greatly improved their functionality (5). Melero et al. (6) were the first to report that anti-4-1BBagonistic mAbs induce regression of established tumors in mice. The long-lasting tumor-specific immune response was primarily mediated by CD8+ T cells (6). Two anti-human 4-1BBagonistic mAbs have been generated and are undergoing clinical evaluation. Urelumab (BMS-663513) is a non-ligand-blocking, fully human IgG 4 mAb with an engineered hinge region to improve stability. A monotherapy urelumab trial was associated with serious hepatic toxicity that resulted in two deaths, leading to its interruption (7). Utomilumab (PF-05082566) is a ligandblocking, fully human IgG 2 mAb with a better safety profile than urelumab, but with only modest antitumor activity (3).
Therefore, and considering that toxicity has been the major impediment for IgG-based anti-4-1BB-agonistic mAbs, and that different studies suggest that organ toxicities are dependent on FcγR interactions, new costimulatory strategies that are not based on full-length antibodies, are urgently needed. Targeted strategies confining the costimulatory activity to the tumor site appears as attractive approaches to improve the implementation of immunomodulatory molecules (8,9). We have recently generated a tumortargeted 4-1BB-agonistic trimerbody, consisting of three anti-4-1BB single-chain fragment variable (scFv) antibodies, and three anti-EGFR single-domain (V HH ) antibodies positioned in an hexagonal conformation around a murine collagen XVIII-derived homotrimerization (TIE XVIII ) domain (10). The Fc-free trimerbody exhibited potent antitumor activity, and did not induce systemic cytokine secretion and hepatotoxicity associated with first-generation 4-1BB-agonistic mAbs (10).
EGFR is widely expressed in normal, non-neoplastic tissues, and its use as a target antigen can lead to severe on-target, offtumor immunotoxicity. For this reason, we have investigated the targeting of carcinoembryonic antigen (CEA) by a bispecific 4-1BB-agonistic trimerbody. CEA is a heavily glycosylated protein that belongs to the CEA-related cell adhesion molecule (CEACAM) family of the immunoglobulin gene superfamily, and is highly expressed in cancers of diverse origins, especially carcinomas (11).
The CEA-targeted 4-1BB-agonistic trimerbody was produced as a homogenous, non-aggregating, soluble protein purifiable by standard affinity chromatographic methods. The purified trimerbody was found to be trimeric in solution, very efficient at recognizing 4-1BB and CEA, and potently costimulated T cells in vitro in the presence of CEA.

MATERIALS AND METHODS
Mice C57BL/6 female mice were purchased from Envigo Crs (Barcelona, Spain), and were maintained in controlled environment: temperature (21 ± 1 • C), humidity (50 ± 5%), and 12 h light/dark cycles. Manipulation was performed under aseptic conditions in a laminar flow hood, when was necessary, and sterilized water and food were available ad libitum.

Expression and Purification of Recombinant Antibodies
HEK293 cells were transfected with the appropriate expression vectors using Lipofectamine 2,000 (Life Technologies) according to the manufacturer's protocol and selected in DCM with 500 µg/ml G418 to generate stable cell lines. Conditioned media from transiently and stably transfected cells were analyzed by ELISA, western blotting and FACS. Recombinant antibodies were purified from conditioned media with HisTrap HP columns (GE Healthcare, Uppsala, Sweden).

Western Blotting
Samples were separated under reducing conditions on 12% Tris-glycine gels and transferred to a nitrocellulose membrane using iBlot 2 Dry Blotting System (Life Technologies). The membrane was blocked with Odyssey Blocking Buffer (LI-COR Biosciences, Lincoln, NE, USA) and probed with anti-c-myc mAb (clone 9E10, Abcam, Cambridge, UK), followed by incubation with DyLight800-goat anti-mouse (GAM) IgG (Rockland Immunochemicals, Gilbertsville, PA, USA).Visualization of protein bands was performed with the Odyssey R system (LI-COR Biosciences).

ELISA
The ability of trimerbodies to bind purified antigens was studied by ELISA as described (10,14). Briefly, Maxisorp (NUNC Brand Products, Roskilde, Denmark) plates were coated (2 µg/well) with recombinant human CEACAM5/CD66e (hCEA; Sino Biological, Beijing, P.R.China) or recombinant mouse 4-1BB:human Fc chimera (m4-1BB; R&D Systems, Minneapolis, MN, USA). After washing and blocking with 300 µl 5% BSA in PBS, 100 µl of neat conditioned media from transfected HEK293 cells or purified protein solution (1 µg/ml), were added and incubated for 1 h at room temperature. After three washes, 100 µl of HRP-conjugated goat anti-mouse IgG (Sigma-Aldrich) were added for 1 h at room temperature, after which the plate was washed and developed. To demonstrate the simultaneous binding of the two antibody domains present in the bispecific trimerbody a dual ELISA was performed. Briefly, Maxisorp plates were coated with hCEA and after washing and blocking, 100 µl of neat supernatant from transfected HEK293 cells, were added and incubated for 1 h at room temperature. After washing, 100 µl of m4-1BB (2 µg/ml) were added followed by 100 µl of HRP-conjugated goat anti-human-Fc IgG (Sigma-Aldrich) and incubated for 1 h at room temperature. The plate was washed and developed.

Flow Cytometry
The ability of trimerbodies to bind to cell surface 4-1BB was studied by flow cytometry as described previously (10). Briefly, HEK293 or HEK293 4−1BB cells were incubated with supernatants or purified trimerbodies (5 µg/ml) and anti-c-myc mAb for 30 min. After washing, the cells were treated with appropriate dilutions of PE-conjugated goat F(ab') 2 anti-mouse IgG (Jackson Immuno Research, Newmarket, UK). The samples were analyzed with a SonySH800 (Sony, Tokyo, Japan). The mouse anti-human HLA-ABC mAb (clone W6/32, Abcam) was used as a control in the flow cytometry studies.

Circular Dichroism (CD)
Circular dichroism measurements were performed with a Jasco J-810 spectropolarimeter (JASCO). The spectrum was recorded on a protein sample at 2.8 µM in PBS using a 0.2 cm path length quartz cuvette at 25 • C. Thermal denaturation curve from 7 to 95 • C was recorded on the same protein sample and cuvette by increasing temperature at a rate of 1 • C/min and measuring the change in ellipticity at 218 nm.

Size Exclusion Chromatography-Multiangle Light Scattering (SEC-MALS)
Static light scattering experiments were performed at room temperature using a Superdex 200 Increase 10/300 GL column (GE Healthcare) attached in-line to a DAWN-HELEOS light scattering detector and an Optilab rEX differential refractive index detector (Wyatt Technology Corporation, Santa Barbara, CA, USA). The column was equilibrated with running buffer (PBS, 0.1 µm filtered) and the SEC-MALS system was calibrated with a sample of BSA at 1 g/L in the same buffer. Then a 100 µL sample of the antibody at 0.3 g/L in PBS was injected into the column at a flow rate of 0.5 mL/min. Data acquisition and analysis were performed using ASTRA software (Wyatt Technology Corporation). Based on numerous measurements on BSA samples at 1 g/L under the same or similar conditions we estimate that the experimental error in the molar mass is around 5%.

Kinetics and Binding Studies Using Biolayer Interferometry
The ability of 1D8 N/C CEA.1 to bind biosensor-immobilized m4-1BB and hCEA in solution at the same time was investigated using biolayer interferometry (BLI) on an Octet RED96 system (Fortebio, Menlo Park, CA, USA). AR2G biosensors (Fortebio) were activated using s-NHS and EDC, and then coated with 10 µg/mL m4-1BB in 10 mM sodium acetate at pH 6 for 20 min, followed by quenching with ethanolamine. After 10 min of equilibration in kinetics buffer (PBS with 0.1% BSA and 0.05% Tween20), the coated biosensors were incubated in kinetics buffer containing 4 nM of 1D8 N18 , 1D8 N/C CEA.1, or only kinetics buffer for 1 h. The biosensors were then transferred into a solution of 50 nM hCEA in kinetics buffer or kinetics buffer for 2 h, followed by a final incubation in kinetics buffer for 1 h. Raw sensor data was exported and smoothed by Savitky-Golay filtering. Similarly, the kinetics of the interaction between m4-1BB and the trimerbody were investigated by using m4-1BB-coated biosensors, incubating with 2 or 4 nM of 1D8 N/C CEA.1 in kinetics buffer for 1 h, and then monitoring dissociation in kinetics buffer for 3 h. The data were fit to a 1:1 binding model using the Octet Analysis software.

Statistical Analysis
All the experiments were done in triplicates and the statistical analysis was performed using Prism software (GraphPad Software, San Diego, CA, USA). Significant differences (P value) were discriminated by applying a two-tailed, unpaired Student's t test assuming a normal distribution with * P < 0.05, * * P < 0.01, * * * P < 0.001. Values are presented as mean ± SD.

Structural Characterization of the CEA-Targeted 4-1BB-Agonistic Trimerbody
The 1D8 N/C CEA.1 was produced in stably transfected HEK293 cells and purified from conditioned medium by immobilized metal affinity chromatography, which yielded proteins with a high degree of purity as determined by reducing SDS-PAGE ( Figure 3A). Mass spectrometry indicates a molecular weight of 50.4 kDa, which is the expected weight for the protomer   after signal sequence processing (data not shown). The SEC-MALS chromatogram ( Figure 3B) shows a major peak eluting at 12.6 mL with a molar mass of 143 kDa at the center of the peak, which indicates the formation of trimers in solution (with a calculated mass of 151.2 kDa). The chromatography also shows a minor proportion of aggregated material eluting at 8.7 mL, the exclusion volume of the column. The circular dichroism spectrum, with a minimum at 218 nm is consistent with a predominantly β-sheet secondary structure, which is stable up to ∼50 • C (Figures 3C,D). The thermal denaturation is irreversible as a large pellet was observed in the cuvette.

Functional Characterization of the CEA-Targeted 4-1BB-Agonistic Trimerbody
The binding kinetics of the bispecific 1D8 N/C CEA.1 was investigated by biolayer interferometry (BLI) using biosensors coated with m4-1BB ( Figure 4A). As shown in Figure 4A, the 1D8 N/C CEA.1 showed saturating binding to m4-1BB at low nanomolar concentrations and an extremely slow dissociation (with an interaction half-life >40 h, as <5% dissociation was observed over 3 h), which is consistent with the binding behavior previously reported for m4-1BB-binding trimerbodies (10). We also used BLI to demonstrate the simultaneous binding of 1D8 N/C CEA.1 to both m4-1BB and hCEA. 1D8 N/C CEA.1 and 1D8 N18 were both able to associate with sensor-immobilized m4-1BB, and 1D8 N/C CEA.1 showed a subsequent binding to hCEA in solution (Figures 4B,C). Importantly, this indicates that the binding to either antigen does not sterically inhibit binding to the other antigen, which would prevent 1D8 N/C CEA.1 from participating in the crosslinking of T cells and tumor cells.
To further assess the multivalence and multispecificity of the 1D8 N/C CEA.1 trimerbody, we performed cell adhesion assays to plastic immobilized hCEA or laminin (Lm111), as positive control. As shown in Figure 5B, HEK293 m4−1BB cells adhered to wells coated with hCEA after incubation with 1D8 N/C CEA.1. Moreover, 1D8 N/C CEA.1 was as efficient as Lm111 in promoting the adhesion of 4-1BB-expressing cells (Figure 5C). The adhesion of HEK293 m4−1BB cells was specific since no adhesion of wild-type HEK293 cells to BSA-and hCEA-coated wells was observed ( Figure 5A). Furthermore, wells coated with BSA-and hCEA, preincubated with 1D8 N18 did not support any significant adhesion of 4-1BB-positive and 4-1BB-negative cells (Figures 5A,B).
To determine whether 1D8 N/C CEA.1 retained the baseline costimulatory capacity observed for 1D8 N18 (10), and whether this was improved by the interaction to hCEA. Mouse CD8a + T cells were stimulated with plate-bound anti-CD3 mAb and the costimulatory antibodies 1D8 IgG, 1D8 N18 and 1D8 N/C CEA.1 in solution, in the presence or absence of immobilized hCEA. A rat IgG 2a and the CEA-specific MFE-23 scFv-based N-terminal trimerbody MFE-23 N18 (10) were used as controls. In the absence of hCEA the 1D8 N/C CEA.1 showed a costimulatory effect similar to 1D8 N18 , but in the presence of immobilized hCEA T cell proliferation was significantly increased (P = 0.003), whereas IFNγ secretion showed a trend to increase, without statistical significance (Figures 5D,E). In the absence of concomitant anti-CD3 mAb-stimulation, the 1D8 N/C CEA.1 did not costimulate T cells, even in the presence of immobilized hCEA (Supplementary Figure 1).

DISCUSSION
Enhancing 4-1BB costimulation by systemic administration of agonistic mAbs is a viable therapeutic strategy, hampered by off-tumor toxicities associated with FcγR interactions (7,10,15). Arguably, reducing the toxicity of immune potentiating antibodies without compromising their antitumor activity is paramount to fully exploiting their clinical potential (16). We have previously shown that an EGFR-targeted 4-1BB-agonistic trimerbody is efficiently targeted to EGFRexpressing tumors in vivo and exhibits antitumor activity similar to IgG-based 4-1BB-agonistic mAbs, with no evidence of toxicity (10).
Given the broad expression profile of EGFR, a CEAtargeted 4-1BB-agonistic trimerbody was generated by fusing the anti-human CEA single-domain antibody CEA.1 (13) to the C-terminus of the 4-1BB-agonistic scFv-based trimerbody 1D8 N5 (10) through a 17-residue linker. The trimerbody 1D8 N/C CEA.1 was expressed by gene-modified human cells as soluble protein and was very efficient at recognizing antigens. Biolayer interferometry analysis revealed that the 1D8 N/C CEA.1 trimerbody was capable of binding concurrently to 4-1BB and CEA, indicating that each domain independently binds to its cognate antigen. SEC-MALS showed that the 1D8 N/C CEA.1 primarily formed the expected trimeric structure. These data suggest that despite the different configuration, with shorter linkers on both ends of the collagen XVIII homotrimerization domain, the overall structure and biophysical properties of the molecule were not modified, suggesting that the 1D8 N/C CEA.1 may have a structure similar to the one observed for the 1D8 N/C EGa1, with the six binding domains in an extended hexagonal conformation around the homotrimerization domain (10). These results also show that the trimerbody format has an enormous structural plasticity and could generate molecules fully adapted to different pathological contexts.
CEA is a non-internalizing antigen (17), which may be advantageous compared to other antigens that are rapidly internalized after antibody binding (18), since the CEA targeted 4-1BB agonistic trimerbody may be available to bind to T cell 4-1BB for a prolonged period after binding to tumor cell CEA. Furthermore, in normal gastrointestinal tract the CEA is expressed at the apical surface of glandular epithelial cells, where it is poorly accessible to systemically administered antibodies due to tight junctions (11,19), which might reduce potential on-target/off tumor toxicity.
Importantly, 1D8 N/C CEA.1 recognized cell surface expressed 4-1BB as efficiently as the anti-4-1BB mAb 1D8, and it was very specific at inducing adhesion of 4-1BB-expressing cells on plastic-bound CEA, but not on plastic-bound BSA. Appropriate stimulation by the members of the TNFRSF involves oligomerization of homotrimer receptors into higher-order complexes at the immunological synapse (3), and accordingly, the potency of the CEA-targeted 1D8 N/C CEA.1 trimerbody was superior to first-generation anti-4-1BB mAbs. The CEA-specific 4-1BB-agonistic trimerbody demonstrated better costimulation of CD8a+ T cells in the presence of immobilized hCEA. These results support the 1D8 N/C CEA.1 trimerbody-induced formation of clustered 4-1BB signaling complexes at the contact interface between T cells and the CEAexpressing cells.
Trimerbodies correspond to a new class of therapeutic agents that offer important advantages over canonical mAbs to manipulate the immune system for therapeutic purposes (20)(21)(22)(23), including reduced toxicity (10). Tumor-targeted 4-1BB-agonistic trimerbodies offer additional advantages compared to conventional IgG-based antibodies in terms of biodistribution, tumor localization, and therapeutic index. In fact tumor-specific trimerbodies have been employed for molecular imaging of cancer with modalities such as nuclear (24) and optical imaging (21,25). We consider that radiolabeled trimerbodies will have potential as theranostic tools in cancer therapy, to improve personalized treatment based on the molecular characteristics of cancer cells and in the intratumoral distribution of immune cell populations, such as 4-1BB positive T cells. Additional advantages of tumor-targeted trimerbodies including the simultaneous targeting of different receptors to exert synergistic or additive effects for the promotion of protective antitumor immunity. For example, treatment with anti-4-1BB and anti-PD-L1 mAbs (26). Yet the complexity of generating and access to multiple clinical-grade mAbs, often manufactured by different companies, essentially precludes their combined exploration in clinical trials.
In summary, our results confirm the value of tumor-targeted 4-1BB agonistic trimerbodies as an alternative to canonical immunomodulatory antibodies in order to realize the therapeutic potential of 4-1BB costimulation.

DATA AVAILABILITY
The datasets generated for this study are available on request to the corresponding author.

ETHICS STATEMENT
Animal protocols were approved by the Ethic Committee of Animal Experimentation of the Instituto Investigación Sanitaria Puerta de Hierro-Segovia de Arana (Hospital Universitario Puerta de Hierro Majadahonda, Madrid, Spain). Procedures were additionally approved by the Animal Welfare Division of the Environmental Affairs Council of the Government of Madrid (66/14).