FS222, a CD137/PD-L1 Tetravalent Bispecific Antibody, Exhibits Low Toxicity and Antitumor Activity in Colorectal Cancer Models

Immunomodulatory mAbs are a promising approach for patients with cancer. Immune checkpoint inhibitors targeting programmed cell death (PD)-1, PD-L1, and cytotoxic T-lymphocyte–associated protein 4 are the most advanced immunotherapy agents for oncology. However, only a subset of patients benefit from long-term survival, and there remains an unmet clinical need (1). Although bispecific T-cell engagers, such as blinatumomab (BLINCYTO) targeting CD3 and CD19, are the most advanced next-generation immuno-oncology modalities, their use is limited to hematologic malignancies and further limited by acute safety concerns (2). We believe agonist antibodies against specific costimulatory receptors from the tumor necrosis factor receptor superfamily may represent the next stage in solid cancer treatment.

CD137 (4-1BB) is a costimulatory molecule and widely known to be upregulated on CD8⁺ T cells following activation (3). CD137 can also be expressed on activated CD4⁺ helper T cells, B cells, regulatory T cells (Treg), natural killer (NK) cells, natural killer T cells, and dendritic cells (DC; ref. 4). Engagement of CD137 by its ligand CD137L results in receptor trimer formation, and subsequent clustering leads to CD137 signaling cascade activation. This provides a survival signal to T cells, thereby sustaining effective T-cell activation and generation of immunologic memory. The primary functional role of CD137 in enhancing T-cell cytotoxicity was first described in 1997 (5), and soon thereafter CD137 mAbs were proposed as anticancer therapeutics.

Clinical development of CD137 mAbs has been hampered by dose-limiting high-grade liver inflammation associated with CD137 agonist antibody treatment. Urelumab (Bristol-Myers Squibb, BMS-663513), a human IgG4 isotype antibody, was the first CD137 mAb to enter clinical trials, but these were halted after significant, on target, dose-dependent liver toxicity was observed (6–8). This outcome was not predicted because urelumab failed to preclinically identify liver inflammation due to its low affinity for the cynomolgus monkey target molecule (9). More recently, clinical trials of urelumab in the treatment of solid cancers were recommended; however, urelumab dosing in these trials had to be limited and efficacy results were disappointing with no objective response reported in the 64 patients with solid tumors treated with monotherapy (6).

No dose-limiting toxicity has been observed with CD137 mAb utomilumab (PF-05082566, Pfizer), a human IgG2 isotype antibody, in dose-escalation phase I clinical trials dosing up to 10 mg/kg in phase I clinical trials of advanced cancer (6, 8). However, the overall objective response rate with this antibody was only 3.8% in patients with solid tumors, potentially indicating that utomilumab has a weaker potency and clinical efficacy than urelumab, while having a more favorable safety profile (6, 8). Trials of utomilumab in combination with radiotherapy or chemotherapy, as well as in combination with other antibody therapies, are ongoing with early results showing no dose-limiting toxicities for doses up to 5 mg/kg and a 26% patient response rate for the combination of utomilumab and pembrolizumab (10).

PD-1 and its ligands PD-L1 (CD274, B7-H1) and PD-L2 (B7-DC) deliver inhibitory signals that regulate the balance between T-cell activation, tolerance, and immunopathology. Consequently, PD-L1 expression by cells can mediate protection against cytotoxic T lymphocyte killing. Cancer, as a chronic and proinflammatory disease, subverts this immune-protective pathway through upregulation of PD-L1 expression to evade the host immune response. PD-L1 expression has been shown in a wide variety of solid tumors (11), and clinical trials have shown the benefit of targeting PD-L1 in patients leading to the approval of three PD-L1–targeting mAbs to date: atezolizumab (MPDL3280A, Tecentriq, Hoffmann-La Roche, Genentech), a humanized IgG1 antibody; avelumab (MSB0010718C, Bavencio, Merck KGaA, Pfizer), a fully human IgG1 antibody; and durvalumab (MEDI4736, Imfinzi, AstraZeneca) a fully human IgG1 antibody. The PD-L1/PD-1 immune checkpoint is also being targeted by three approved PD-1 mAbs, namely pembrolizumab (Keytruda, Merck), nivolumab (Opdivo, BMS), and cemiplimab (Libtayo, Regeneron Pharmaceuticals).

In mouse models resistant to single-agent treatment with either CD137 agonists or PD-1/L1 blockade, significant synergistic effects have been observed when antibodies targeting both pathways are combined. The mechanistic basis for this synergy, even in poorly immunogenic mouse tumor models, is that tumor-infiltrating lymphocytes coexpress PD-1 and CD137, and following combination treatment CD8⁺ T cells can now effectively respond to tumor-associated neoantigens (12). The mechanistic basis for CD137 agonist antibodies alone can be two-fold. Firstly, their ability to induce effector cell function can result in antitumor activity in some preclinical models. Secondly, their alternative function to deplete Tregs has been described as the most effective mechanism of action for CD137 mAbs (13). However, an alternative approach is to direct potent CD137 agonist activity to tumor-reactive T cells while not relying on Treg depletion. Avoiding Treg depletion can be achieved by removing FcγR binding by L234A and L235A (LALA) mutation while retaining agonist activity via an alternative mechanism of cross-linking as described in this article.

As only a fraction of patients respond to monotherapies that block the PD-1/PD-L1 pathway (14), and CD137-targeting agonistic molecules have yet to demonstrate significant responses in patients with cancer without toxicity (7, 8), we believe there remains a need to develop treatments that combine PD-L1 blockade and provoke strong CD137 agonism in safe and efficacious therapies that do not rely on a combination approach. An alternative to combining CD137 and PD-L1 monotherapies is the development of a bispecific antibody that encompasses the two modalities. It is anticipated that such a bispecific mAb could deliver superior antitumor efficacy over combining monotherapies. There are existing preclinical approaches combining CD137 mAb activity with PD-L1 mAb activity into bispecific therapies. These can be subdivided into two broad range categories, non–IgG- and IgG-like molecules, both of which can be further divided by their binding valency for each target.

Here, we describe a fully human, tetravalent, IgG bispecific antibody (mAb², FS222) comprising a PD-L1–specific mAb with 5 amino acid insertions and 7 amino acid substitutions in the CH3 region of the Fc domain to create two binding sites forming an Fc fragment antigen-binding (Fcab) for CD137 (15). FS222 blocked PD-L1 and activated CD137⁺ tumor-reactive T cells in a PD-L1–dependent manner. It demonstrated similar potency in primates, and preliminary toxicity studies in this species showed significant pharmacodynamic (PD) responses and a lack of toxicity. A surrogate mouse cross-reactive CD137/PD-L1 mAb² with homologous mechanisms of action to FS222 was observed to provide a substantial survival benefit in multiple mouse tumor models with no toxicity and showed potent in vivo PD changes related to antitumor immune responses.

The CD137/PD-L1 mAb² molecule named FS222 consisting of an IgG1 molecule comprising the human CD137 Fcab was prepared by substituting part of the CH3 domain comprising the AB, CD, and EF loops (15), for the corresponding region of the CH3 domain of a PD-L1 mAb (E12v2). Fcab generation has been previously described (16). FS222 incorporates a LALA mutation (leucine to alanine at positions 234 and 235 according to Eu numbering) in the CH2 domain (AA) to reduce Fcγ receptor binding (17, 18). FS222 was expressed transiently using HEK293 6E (National Research Council Canada, Canada) cells. Supernatants were purified on MabSelect SuRe LX Protein-A prepacked columns using ÄKTAxpress instrument (both GE Healthcare Life Sciences). IgG protein content was quantified by BioLayer Interferometry (BLI) using the Octet QK^e System (FortéBio) platform with Protein A quantitation biosensors (FortéBio; 18-5013). FS222 was purified by Protein A affinity chromatography using mAb SelectSure columns.

After purification, size exclusion–high-performance liquid chromatography (SE-HPLC) was performed on an Agilent 1100 series HPLC Value System (Agilent Technologies, Inc.), fitted with a TSKgel SUPERSW3000 HPLC 4.6 mm ID × 30 cm column (Tosoh Bioscience, LLC) using 20 mmol/L sodium phosphate and 200 mmol/L sodium chloride, pH 6.8, as a mobile phase. Quantification of percentage monomer was performed using ChemStation software (Agilent Technologies, Inc.). Capillary electrophoresis–sodium dodecyl sulfate (CE-SDS) analysis was performed on a 2100 Bioanalyzer Capillary Electrophoresis System (Agilent Technologies, Inc.), according to manufacturer's instructions. For reducing conditions, dithiothreitol (DTT) was added and samples were denatured at 70°C for 5 minutes.

His-tagged human PD-L1 antigen was coated on to a CM5 chip to approximately 1,100 RU and was used to immobilize FS222 when injected at 100 nmol/L which resulted in approximately 300 RU of FS222 being captured. Fc-tagged human CD137 antigen was then injected at a single concentration (100 nmol/L), using the Biacore T200 (GE Healthcare Life Sciences), to observe dual binding.

DO11.10 T cells overexpressing human CD137 were cultured in DMEM containing 10% heat-inactivated FBS, 1 mmol/L sodium pyruvate, and 50 μg/mL puromycin. HEK cells overexpressing human PD-L1 were cultured in DMEM containing 10% FBS, 100 μg/mL hygromycin B, 15 μg/mL blasticidin, and 1 μg/mL doxycycline. Cells were resuspended in 40 μL of FS222, CD137/Ctrl(HelD1.3) mAb², CD137(MOR7480.1) mAb, Ctrl(4420) mAb, or PD-L1(E12v2) mAb (Supplementary Table S1) titrations prepared in DPBS and then washed and resuspended in a secondary human IgG detection antibody (A-21445, Thermo Fisher Scientific) which had been diluted in DPBS. Cells were then resuspended in the viability dye 7-AAD (A1310, Thermo Fisher Scientific) and examined with either a BD FACSCanto II or BD LSRFortessa II (BD Biosciences) before being analyzed using FlowJo V10 (TreeStar, Inc.).

Peripheral blood mononuclear cells (PBMC) were isolated from leukocyte cones, obtained from platelet donors, by Ficoll (GE17-1440-02, Sigma-Aldrich) gradient separation. Pan T cells were isolated from the PBMCs present in the eluent using the Pan T Cell Isolation kit II according to the manufacturer's instructions (130-096-535, Miltenyi Biotec). Pan T cells were incubated overnight at 37°C and 5% CO₂ in RPMI supplemented with 10% heat-inactivated FBS, 1 mmol/L 2-mercaptoethanol, penicillin (100 U/mL)/streptomycin (100 U/mL), and 1 μmol/L sodium pyruvate. Dynabeads Human T-Activator CD3/CD28 beads (11132D, Thermo Fisher Scientific) were used to activate T cells and upregulate CD137 and PD-L1 surface expression. Beads were washed from the T cells using a DynaMag-15 Magnet (12301, Thermo Fisher Scientific) following the manufacturer's instructions.

Stimulated pan human primary T-cell suspensions were resuspended in 40 μL FS222, CD137/Ctrl(HelD1.3) mAb², PD-L1 mAb, CD137(MOR7480.1) mAb, or Ctrl(4420) mAb (Supplementary Table S1) titrations prepared in DPBS and treated with AF647 goat anti-human IgG (H + L; 1:500; A-21445, Thermo Fisher Scientific), anti-hCD4 FITC (1:200; 550628, BD Biosciences), and anti-hCD8 eF450 (1:200; 48-0087-42, Thermo Fisher Scientific) prepared in DPBS. 7-AAD (A1310, Thermo Fisher Scientific) was used as a viability dye, and samples were then examined with a BD FACSCanto II before being analyzed using FlowJo V10 Prism software.

Human primary CD8⁺ T cells were isolated from PBMCs obtained from leucocyte depletion cones using the CD8⁺ T-cell isolation kit II (130-096-495, Miltenyi Biotec Ltd.) according to the manufacturer's instructions. For cell-based cross-linking, HEK293 cells overexpressing hPD-L1 (HEK.hPD-L1) or HEK wild-type cells or a mixture of the two populations were plated on to CD3 mAb–coated (8 μg/mL, Clone UCHT1, R&D Systems, MAB100-SP) 96-well flat-bottom plates in 100 μL T-cell culture medium [RPMI medium with 10% FBS, 1X penicillin–streptomycin, 1 mmol/L sodium pyruvate, 10 mmol/L Hepes (Sigma-Aldrich, H0887), and 50 μmol/L 2-mercaptoethanol (Gibco, M6250)]. CD8⁺ T cells were added. Cells were treated with a titration of FS222, CD137(20H4.9) mAb, or Ctrl(4420) mAb (Supplementary Table S1). Supernatants were assayed with human IL2 ELISA Ready-SET-Go! kit (88-7025-88, Fisher Scientific) following the manufacturer's instructions. Plates were read at 450 nm using the plate reader with the Gen5 Software. The concentration of human IL2 (hIL2) was plotted versus the log concentration of antibody, and the resulting curves were fitted using the log (agonist) versus response equation in GraphPad Prism.

Human primary CD4⁺ T cells were isolated from leukocyte cones using a Human CD4⁺ T Cell Isolation Kit (130-096-533, Miltenyi Biotec Ltd.) according to the manufacturer's instructions. Dynabeads Human T-Activator CD3/CD28 (11131D, Thermo Fisher Scientific) were used in the presence of 50 IU/mL recombinant human IL2 (PeproTech, 200-02) with 3:1 bead to cell ratio to expand cells for 7 days. Dynabeads were removed, and CD4⁺ T cells were rested overnight with reduced 10 IU/mL recombinant human IL2.

Monocytes were isolated from human PBMCs using a Human Pan Monocyte Isolation Kit (130-096-537, Miltenyi Biotec Ltd.) following the manufacturer's instructions. Monocytes were differentiated to iDCs using Human Mo-DC Differentiation Medium (130-094-812, Miltenyi Biotec Ltd.) following the manufacturer's instructions.

Expanded T cells were cultured in AIM V Medium (12055091, Thermo Fisher Scientific) and incubated overnight. Titrations of FS222, PD-L1(E12v2) mAb and CD137(20H4.9) mAb, PD-L1(E12v2) mAb, CD137/Ctrl(HelD1.3) mAb² and PD-L1(E12v2) mAb, CD137(20H4.9) mAb, CD137/Ctrl(HelD1.3) mAb², or Ctrl(4420) mAb (Supplementary Table S1) were used to treat a 1:10 mix of iDC cells and expanded CD4⁺ T cells in AIM V Medium for 5 days. Supernatants were analyzed for IFNγ using Human IFN gamma ELISA Ready-SET-Go! Kit (88-7316-86, Thermo Fisher Scientific). Plates were read at 450 nm using the plate reader with the Gen5 Software. The concentration of human IFNγ was plotted versus the log concentration of antibody, and the resulting curves were fitted using the log (agonist) versus response equation in GraphPad Prism.

CD8⁺ T-cell activation was achieved by antigen stimulation of genetically modified OT-1 T cells, isolated from C57BL/6 OT-1 mice (003831, The Jackson Laboratory) having a T-cell receptor specific for ovalbumin peptide 257–264, and was determined by the release of IFNγ. OT-1 T cells were incubated with B16-F10 melanoma cells, which had previously been cultured in the presence of 20 ng/mL IFNγ (AF-315-05-100UG, PeproTech) to induce PD-L1 expression, and that were then pulsed with 500 nmol/L SIINFEKL peptide for 1 hour at 37°C, to drive T-cell activation. The efficacy of surrogate FS222 was subsequently assessed by ELISA for secreted mIFNγ (88-7314-88, Thermo Fisher Scientific) after 3 days. This assay was also carried out utilizing MC38.OVA cells that express ovalbumin, in an identical protocol, except for peptide pulsing which was not necessary.

The CT26.WT colon carcinoma cell line (ATCC) was initially expanded, stored, and then prescreened by IDEXX Bioresearch for pathogens using the IMPACT I protocol and shown to be pathogen-free. BALB/c female mice (Charles River) aged 8 to 10 weeks and weighing 18 to 22 g each received 1 × 10⁵ CT26.WT cells injected subcutaneously in the left flank in 100 μL DMEM serum-free culture medium.

The MC38 colon carcinoma cell line (ATCC) was initially expanded, stored, and then prescreened by IDEXX Bioresearch for pathogens using the IMPACT I protocol and shown to be pathogen free. C57BL/6 female mice (The Jackson Laboratory) aged 9 to 10 weeks and weighing 18 to 24 g each received 1 × 10⁶ MC38 cells injected subcutaneously in the right flank in 100 μL DMEM serum-free culture medium.

The surrogate FS222 and control antibodies were injected intraperitoneally into mice at appropriate μg per mouse in DPBS, 100 mmol/L arginine, and 0.05% Tween 80 on days 7, 9, and 11 following tumor inoculation. Tumor volume measurements were taken with callipers to determine the longest axis and the shortest axis of the tumor, and the following formula was used to calculate the tumor volume:

Where L = longest axis; S = shortest axis.

A single-dose PD study was run in the same CT26.WT syngeneic tumor model as described above. This study comprised three dosing groups, receiving either control antibody or surrogate FS222 at one of two doses. Samples from tumor tissue and blood were analyzed over 8 time points (2, 6, 24, 48, 72, 96, 120, and 192 hours). Each dosing cohort had 64 mice (8 mice per timepoint). Each animal received 1 × 10⁵ CT26.WT cells injected subcutaneously in the left flank in 100 μL DMEM. Eleven days following tumor cell inoculation, each mouse received the test sample via a 100 μL i.v. injection.

Tumor tissue and blood were tested for drug-bound–positive T cells, T-cell proliferation, and free PD-L1. Blood (100 μL) was collected into EDTA-coated capillaries by tail vein bleeding and was lysed twice in red blood cell lysis buffer (Thermo Fisher Scientific, 00-4333-57) according to manufacturer's instructions. Tumor tissue was collected by dissection and was disaggregated to single-cell suspension by standard mechanical and enzymatic methods. Red blood cells were lysed in red blood cell lysis buffer according to the manufacturer's instructions.

Cells were stained with Fixable Viability Dye eFluor 780 (65-0865-14, Thermo Fisher Scientific) following the manufacturer's instructions. Cells were stained with an antibody staining panel (Supplementary Table S4, all but Ki67 and FoxP3 antibodies) in the presence of CD16/CD32 mAb Fc block (1:50, 14-0161-86, Thermo Fisher Scientific) and then fixed and permeabilized with the eBioscience Foxp3 staining Kit (00-5523-00, Thermo Fisher Scientific) according to the manufacturer's instructions. Cells were stained with Ki67 and Foxp3 antibodies in the presence of Fc block and then examined in a BD Fortessa flow cytometer. Data were analyzed with FlowJo, Excel, and GraphPad Prism.

A preliminary dose-range finding study was conducted to evaluate the pharmacokinetic (PK)/PD response to and tolerability of FS222 in cynomolgus monkeys. The study was performed using Mauritian cynomolgus macaques at Charles River Laboratories in line with Institutional Animal Care and Use Committee guidelines and in accordance with the “Guide for the Care and Use of Laboratory Animals” (1996) by the Institute of Laboratory Animals Research Commission on Life Sciences (National Research Council, Washington, DC).

Briefly, FS222 was administered to cynomolgus monkeys (1/sex/group) via intravenous infusion at 3 mg/kg as a single dose on day 1 or at 0.1, 1, 10, or 30 mg/kg as repeat doses on days 1, 8, 15, and 22. For the 3 mg/kg dose group, serial serum samples were collected for PK assessment on day 1 (predose, 0.083, 0.5, 2, 6, and 12 hours postdose) and days 2, 3, 4, 6, 8, 11, 15, 22, 29, 36, and 43. For the remaining groups, PK serum samples were collected on day 1 (predose, 0.083, 0.5, 2, 6, and 12 hours postdose) and days 2, 3, 7, 8 (predose and 0.083 hours postdose); day 15 (predose, 0.083, 0.5, 2, 6, and 12 hours postdose); days 16, 17, 21, and 22 (predose and 0.083 hours postdose); and day 25. Serum levels of FS222 were measured using a qualified Gyros-based immunoassay developed in-house to specifically detect free drug (human biotinylated PD-L1 was used as a capture reagent and human Alexa Fluor labeled CD137 as a detection reagent).

For the evaluation of tolerability, standard toxicology parameters such as body weight, food consumption, clinical observations, hematology, and blood chemistry were evaluated over the duration of the study. The study was terminated 25 days after administration of first dose in repeat dose animals and 43 days after administration of single dose (PK group of animals).

For evaluation of the PD response to FS222, immunophenotyping of peripheral blood was performed to assess peripheral lymphocyte populations (monocytes, T cells, B cells, and NK cells) as well as the induction of proliferation and activation of central memory and effector memory CD4⁺ and CD8⁺ T-cell subpopulations. Serial blood samples (on EDTA) were collected prior to first dosing of FS222 (predose) and on the indicated study days over the course of the study; stained with antibodies against CD45, CD3, CD4, CD8, CD16, CD28, CD25, FoxP3, CD95, CD69, and Ki67 (Supplementary Table S3); and analyzed using a FACSCanto II flow cytometer (BD Biosciences) and FACSDiva (BD Biosciences), Excel, and GraphPad Prism software.

FS222 was created by incorporating into a proprietary PD-L1 mAb, an engineered IgG1 Fc region termed Fcab (Fc-region with antigen binding), where high-affinity antigen binding for CD137 was introduced in the C-terminal region (Fig. 1A). Binding to FcγR was removed by the introduction of L234A and L235A (LALA) mutations (Supplementary Fig. S1A), while binding to human FcRn was maintained (Supplementary Fig. S1B). The mAb² is tetravalent, with two binding sites for PD-L1 (one in each Fab region) and two binding sites for CD137 (one in each CH3, due to the homodimeric nature of the Fc region), and maintains the IgG1 structure.

The simultaneous binding of FS222 to human CD137 and human PD-L1 was tested via SPR and then individually to cells expressing either human CD137 or human PD-L1.

FS222 simultaneously bound to both targets as observed by SPR (Fig. 1B) with affinities of 0.66 nmol/L to dimeric human CD137 and 0.19 nmol/L to monomeric human PD-L1 (Supplementary Fig. S1C–S1E). The cell-binding assays demonstrated that FS222 bound human CD137 to an equivalent level as the component CD137 Fcab (CD137/Ctrl(HelD1.3) mAb²) alone (EC₅₀ 6.2 nmol/L; Fig. 1C). FS222 also bound human PD-L1 to an equivalent level as the component Fab (PD-L1(E12v2) mAb) alone (EC₅₀ 3.7 nmol/L; Fig. 1D). PD-L1(E12v2) mAb was found to be just as effective as the PD-L1(S70) mAb (YW243.55.S70) at blocking PD-L1 binding to PD-1 as investigated in a DO11.10 human PD-L1 T-cell activation assay (data not shown). As shown in Fig. 1E, FS222 bound to activated CD4⁺ and CD8⁺ primary T cells with an EC₅₀ of 0.8 and 0.9 nmol/L, respectively. This was equivalent to the PD-L1(E12v2) mAb–binding characteristics but not equivalent to the CD137/Ctrl(HelD1.3) mAb², showing that this cell binding was driven primarily through PD-L1 rather than CD137. The positive control CD137 mAb, CD137(MOR7480.1) mAb, bound to activated human primary T cells as expected, whereas the CD137/Ctrl(HelD1.3) mAb² had minimal binding to the same cells. Therefore, FS222 bound to PD-L1 with high affinity and, through design, bound CD137 with high avidity, which are two features of the molecule critical for the cross-linking–dependent activity as described below.

Subsequent investigations were aimed to dissect the nature of FS222-mediated CD137 agonism. To demonstrate that FS222 can be cross-linked to mediate human CD137 signaling only in the presence of cells expressing human PD-L1, human primary CD8⁺ T cells stimulated by plate-bound CD3 mAb were cocultured with wild-type HEK 293 cells, HEK 293 cells engineered to overexpress human PD-L1, or mixtures of the two cell lines in different proportions. This allowed the investigation of varying ratios of human PD-L1–expressing cells to wild-type cells to model the likely heterogeneity of expression present within different human tumors.

FS222 showed maximum activity, measured by human IL2 release, from activated human primary CD8⁺ T cells, when 100% of HEK 293 cells expressed PD-L1 (Fig. 2A). The maximum IL2 release (E_max) reduced in proportion to the reduction in the percentage of cells expressing human PD-L1 present; however, the EC₅₀ value remained broadly the same at 0.05 nmol/L (Fig. 2A).

The activity of FS222 was tested in a mixed lymphocyte reaction (MLR) which utilizes human primary CD4⁺ T cells and immature monocyte-derived DCs (iDC) expressing endogenous levels of both targets. The PD-L1–specific antibody [PD-L1(E12v2) mAb] showed potent activity in the MLR assay (EC₅₀ 0.08 nmol/L). The human CD137 mAb, CD137(20H4.9) mAb, even when cross-linked with hCH2 mAb, did not elicit activity. This suggested that CD137 signaling alone is ineffective in this assay. However, a combination of PD-L1(E12v2) and CD137(20H4.9) mAbs cross-linked with hCH2 mAb showed potent activity (EC₅₀ 0.14 nmol/L) with a higher maximum IFNγ release compared with PD-L1(E12v2) mAb alone, which indicated a synergistic effect of the two mAbs (Fig. 2B). FS222 showed similarly potent activity to the combination of the two separate monospecific antibodies with an EC₅₀ of 0.07 nmol/L and comparable maximum IFNγ release level (Fig. 2B). FS222 was also tested against each component part of the mAb² that makes up FS222, the CD137/Ctrl(HelD1.3) mAb², and the PD-L1(E12v2) mAb, alone and in combination. CD137(HelD1.3) mAb² showed no activity in this assay, indicating that the CD137 Fcab component had an inability to activate T cells in an MLR. There was no additional effect on activation by treating with a CD137/Ctrl(HelD1.3) mAb² plus PD-L1(E12v2) mAb combination above that already seen with PD-L1(E12v2) mAb alone. FS222 showed a similar potency to the combination of its component parts CD137/Ctrl(HelD1.3) mAb² and PD-L1(E12v2) mAb (EC₅₀ of 0.07 and 0.06 nmol/L, respectively) with superior efficacy as indicated by higher IFNγ production (E_max; Fig. 2B).

FS222 did not bind to mouse CD137 (data not shown) so it was not possible to evaluate FS222 in mouse syngeneic tumor model systems in vivo. Therefore, a mouse surrogate of FS222 was created using an Fcab targeting mouse CD137 selected using yeast display. The Fcab against mouse CD137 was selected based on affinity measurements and cross-link–dependent activation of CD137 and was tested in similar mouse systems to those used to determine the function of FS222 in human systems. Surrogate FS222 was tested for cell binding similar to FS222 and was found to have an EC₅₀ of 2.7 and 2.6 nmol/L to cells engineered to overexpress mouse CD137 and mouse PD-L1, respectively.

The functional activity of surrogate FS222 was determined in a primary assay where CD137 and PD-L1 are endogenously expressed on activated T cells and B16-F10 tumor cells, respectively. B16-F10 mouse melanoma cells that had previously been pulsed with OVA peptide (SIINFEKL) were cocultured with CD8⁺ antigen–specific OT-1 T cells. CD8⁺ T-cell activation (IFNγ release) was increased after treatment with either mCD137(Lob12.3) mAb, PD-L1(S70) mAb, or a combination approach. However, the greatest potency was seen upon treatment with the surrogate FS222 (EC₅₀ 0.003 nmol/L; Fig. 3A). The same result was achieved using a similar assay described above but utilizing MC38 cells expressing ovalbumin as the source of both OVA peptide and PD-L1 (Supplementary Fig. S2A). As with FS222, no activation was detected in the absence of PD-L1 cross-linking which indicated a similar mode of action between the surrogate and human-specific molecules.

To evaluate the antitumor effects of surrogate FS222 in vivo, CT26.WT or MC38 tumor cell lines were injected subcutaneously into the flank of BALB/c and C57BL/6 mice, respectively. In the CT26.WT tumor model, mice were injected intraperitoneally on days 7, 9, and 11 after tumor implantation with approximately 10 mg/kg surrogate FS222, PD-L1(S70) mAb, mCD137(Lob12.3) mAb, or Ctrl(HelD.13) isotype control mAb. Surrogate FS222 was shown to substantially reduce tumor growth (Fig. 3B), and 42% of mice remained tumor free up to 68 days after treatment (Fig. 3C and Table 1). In contrast, treatment with PD-L1(S70) mAb or mCD137(Lob12.3) mAb failed to show significant survival advantage (Fig. 3C and Table 1; Supplementary Tables S8 and S9).

In the MC38 tumor model, surrogate FS222 was able to eradicate all tumors at a lower dose of 1 mg/kg. Mice were injected intraperitoneally 7, 9, and 11 days after tumor cell inoculation with a dose of approximately 1 mg/kg of either surrogate FS222, PD-L1(S70) mAb, mCD137(Lob12.3) mAb, a combination of both, or Ctrl(4420) isotype control mAb. Surrogate FS222–treated mice showed full tumor regression in all mice which remained tumor free until day 49 when the study was ended. In contrast, single and combined treatment of PD-L1(S70) mAb and mCD137(Lob12.3) mAb resulted in durable tumor regression in only a fraction (4/12 or less) of animals (Fig. 3D and E, Table 1; Supplementary Tables S10 and S11).

In the syngeneic CT26.WT tumor model, we assessed surrogate FS222 dose-dependency in vivo and showed dose-dependent survival benefit between doses of approximately 0.1 mg/kg and approximately 10 mg/kg. To evaluate dose-dependent efficacy, surrogate FS222 dose levels equivalent to approximately 0.1, 0.3, 1, and 10 mg/kg, and Ctrl(4420) mAb isotype control dosed at 10 mg/kg, were administered intraperitoneally using the same study design described above. Surrogate FS222 showed an antitumor efficacy from 0.3 mg/kg and durable tumor regression in 21% of treated animals at 1 mg/kg and 40% at 10 mg/kg (Supplementary Fig. S2B). Using the log-rank (Mantel–Cox) test, a significant survival dose dependency was shown for surrogate FS222 at 0.3 mg/kg up to 1 mg/kg compared with Ctrl(4420) mAb treatment, but no significant benefit raising from 1 mg/kg to 10 mg/kg surrogate FS222 despite the median survival extending from 29 to 39 days, respectively (Fig. 4 and Table 2).

CT26.WT tumor–bearing mice were treated with a single intravenous dose of surrogate FS222 (∼1 mg/kg and ∼10 mg/kg) 11 days after subcutaneous inoculation of CT26.WT. Tumor tissue and blood were tested for T-cell–bound surrogate FS222, T-cell proliferation, and PD-L1 receptor occupancy over time (between 2 and 192 hours). Total CD137 receptor expression was also assessed.

A high percentage of peripheral and tumor-resident T cells showed bound surrogate FS222 as early as 2 hours after intravenous administration (Fig. 5A). There was a dose-dependent correlation in the longevity of binding, with surrogate FS222 no longer detected after 96 hours on T cells isolated from mice administered with 1 mg/kg, whereas surrogate FS222 was still detected between 120 and 192 hours after administration of 10 mg/kg.

Ki67 expression was used as a marker for T-cell proliferation on CD4⁺ and CD8⁺ T cells. T cells isolated from tumor tissue exhibited a higher frequency of Ki67⁺ T cells as expected in an inflammatory tumor microenvironment. At both dose levels, surrogate FS222 resulted in increases in the frequency of Ki67⁺ peripheral blood T cells when compared with Ctrl(4420) mAb isotype control, indicating a PD response (Fig. 5B). The effect appeared stronger for CD8⁺ T cells which is in line with CD8⁺ T cells expressing higher CD137 levels than CD4⁺ T cells.

A proportion of cells isolated from the blood and tissue of mice dosed with Ctrl(4420) mAb isotype control were saturated with 100 nmol/L surrogate FS222 ex vivo which acted as control for 100% PD-L1 receptor engagement and was confirmed by fully blocking binding with a competing mouse PD-L1 mAb (clone 10F.9G2). Cells isolated from tumor tissue and the blood of mice treated with surrogate FS222 at the 10 mg/kg dose showed near-complete PD-L1 blockade for 8 days, as shown in Fig. 5C represented by 100% PD-L1 receptor occupancy. At 1 mg/kg, surrogate FS222 achieved near-complete PD-L1 receptor occupancy for approximately 72 hours on peripheral T cells. PD-L1 receptor engagement on T cells present in blood showed an accelerated decrease compared with T cells present in the tumor tissue which retained a greater PD-L1 receptor occupancy with greater longevity.

Serum cytokines were analyzed by multiplex electrochemiluminescent immunoassay (Meso Scale Discovery, MSD) to assess cytokine production after surrogate FS222 treatment in the same model. Surrogate FS222, when dosed at 10 mg/kg, resulted in increased serum proinflammatory cytokines IFNγ, TNFα, and IL6. The immunosuppressive cytokine IL10 likewise shows increase in serum after dosing, presumably to counter the proinflammatory response. This effect was dose-dependent, and serum cytokine levels remained similar in mice treated with 1 mg/kg surrogate FS222 compared with Ctrl(4420) mAb–treated mice (Fig. 5D). Antitumor activity observed via tumor growth inhibition, tracked by measuring excised tumors at indicated timepoints, still showed highly significant tumor growth inhibition in mice treated with a single dose of 1 mg/kg surrogate FS222 (Supplementary Fig. S3A). This indicated localized antitumor cytotoxic activity without systemic exposure to inflammatory cytokines.

In 2008, clinical trials investigating urelumab in solid tumors were halted due to severe treatment-related immune events which manifested in the liver as severe hepatotoxicity resulting in patient deaths (7). More recently, urelumab has been administered at far reduced dose levels to mitigate this toxicity. Preclinical mechanistic work was undertaken in mice wherein animals dosed with CD137 agonist mAbs showed similar hepatotoxicity. These studies showed a requirement for T cells and CD137 expression in the resultant hepatotoxicity (9, 19). Therefore, these animal models have some translational relevance for predicting the risk of hepatotoxicity in the clinic in human patients following administration of other CD137 agonists such as FS222. Mice from our CT26.WT syngeneic tumor studies showed no overt signs of toxicity following repeated dosing with surrogate FS222 and maintained normal bodyweight throughout. To determine whether immune activation and antitumor activity observed as a result of treatment with 1 mg/kg surrogate FS222 correlated with hepatotoxicity, liver samples were taken at necropsy for histologic assessment. Surrogate FS222–treated and control mice were necropsied 4, 7, and 14 days after the last administration whereby liver samples were excised and examined.

Each liver section was scored for pathology, and the frequencies of mice showing zero, minimal, slight, and moderate effects within each group are shown in Supplementary Table S2 (0 = zero, 1 = minimal, 2 = slight, 3 = moderate). Surrogate FS222–treated animals showed minimal liver pathology (Supplementary Fig. S3B and Supplementary Table S2). Specifically, the livers showed minimal to slight hepatocellular necrosis with mixed lymphocyte infiltrate in the parenchyma, minimal to slight mixed inflammatory cells in periportal tracts, no degenerative hepatocytes, and minimal to slight increased mitoses (Supplementary Table S2). These findings are not deemed to represent adverse hepatotoxicity, as observed with other examples of CD137 agonist mAbs.

In a similar liver pharmacology mouse study, which included dosing CT26.WT tumor–bearing mice with 10 mg/kg of a CD137 targeting mAb clone 3H3 known to induce liver toxicity (20), both surrogate FS222–treated (also at 10 mg/kg) and CD137(3H3) mAb–treated animals showed CD3⁺ T-cell liver infiltration from day 13 onward to similarly high levels above control mice (Fig. 5E). However, activated CD8⁺ T cells remained at significantly higher levels with greater longevity in CD137(3H3) mAb–treated animals compared with animals treated with surrogate FS222 (Fig. 5F). The CD8⁺ T-cell response for both CD137(3H3) mAb and surrogate FS222 peaks at approximately 95% positive for Ki67 across days 8 and 13 after first dose. However, for surrogate FS222, this returned to baseline by day 16, whereas for CD137(3H3) mAb, this did not happen by the end of the study at day 28 after first dose (Fig. 5F). This indicated a difference in the mode of action of these two CD137 targeting agents; CD8⁺ T cells activated by surrogate FS222 proliferated less so and for a shorter period compared with CD8⁺ T cells activated by CD137(3H3) mAb. CD137(3H3) mAb has been shown to lead to hepatic degeneration previously (20). CD4⁺ FoxP3⁺ Tregs were present at higher levels in surrogate FS222–treated livers compared with CD137(3H3) mAb–treated livers at day 8 and day 13 after first dose (Supplementary Fig. S3C). Approximately 30% of CD4⁺ T cells were positive for FoxP3 after surrogate FS222 treatment, whereas the level after CD137(3H3) mAb remained nearer baseline at 10% (Supplementary Fig. S3C). This indicated a potentially more immunosuppressive environment which could dampen the damaging effect of activated CD8⁺ T-cell accumulation, shown to otherwise lead to hepatocyte death (21).

Due to the potential preclinical limitations of surrogate molecules and mouse models for predicting CD137-induced liver toxicity, we ran a non-GLP PK/PD toxicity study of single and repeat dosing of FS222 in cynomolgus monkeys.

FS222 was shown to be fully cross-reactive in cell binding assays and primary immune cell functional assays using PBMCs from human or cynomolgus blood (Supplementary Fig. S4A–S4D). Therefore, the PK behavior of FS222 was characterized in cynomolgus monkeys after intravenous administration of FS222 in a non-GLP dose-range finding study (Fig. 6A). FS222 displayed a dose proportional increase in C_max and AUC (0–168 hours; Supplementary Table S5) and linear plasma clearance (at doses ≥ 1 mg/kg; Fig. 6A). FS222 had a mean terminal half-life of approximately 148 hours, which is in line with human antibodies in monkeys targeting PD-L1 (atezolizumab BLA #761034 Pharmacology Review). In general, FS222 PK followed a linear dose response (at dose levels ≥ 1 mg/kg) and clearance rates (CLp), and the volumes of distribution were similar between animals (Supplementary Tables S6 and S7). FS222 was generally well tolerated up to 30 mg/kg dosed weekly as determined by clinical chemistry and histopathology results (Table 3).

As shown in Fig. 6B, the levels of serum soluble PD-L1 (sPD-L1) were quantified as a measure of direct target engagement and indicative of downstream cell activation (22). Increased serum sPD-L1 levels were observed in all animals on day 1, with an apparent peak at 168 hours after end of infusion, following which the levels declined in line with the decline in the systemic levels of FS222. Repeat administration of FS222 resulted in prolonged increase in serum sPD-L1 in animals that were shown to have no or low levels of antidrug antibodies. Consistent with the findings of the study to assess the PD response of the surrogate FS222 in a syngeneic mouse tumor model described previously, a drug-related increase in cell proliferation and activation was also observed in NK cells (Fig. 6C) and CD4⁺ and CD8⁺ central memory T cells (Fig. 6D and E). In many animals, Ki67 expression reached plateau at day 11, remained high at day 15, and deceased progressively to reach baseline expression between days 18 and 22 with a maximum response being observed between 3 and 10 mg/kg. A moderate but transient increase in the relative percentage and absolute counts of CD4⁺ FoxP3⁺ Tregs was also seen (Supplementary Fig. S4E).

FS222, a CD137/PD-L1 tetravalent bispecific antibody, exhibited potent in vitro CD137-mediated T-cell activation upon engagement of PD-L1. No cross-reactivity was observed to mouse CD137; therefore, a mouse surrogate molecule was developed. Surrogate FS222 outperformed CD137 and PD-L1 monospecific mAbs as monotherapies or in combination in multiple syngeneic mouse tumor models.

No liver pharmacology or toxicity, previously reported with other CD137 agonist mAbs, was observed with FS222 or the mouse surrogate. Contrasting observations of liver toxicity in the clinic, with CD137 mAbs urelumab and utomilumab, suggest that targeting of CD137 is not an intrinsically toxic pathway for therapy, but the way it is targeted is crucial. Urelumab is a potent fully human IgG4 antibody but causes dose-dependent and on-target liver toxicity, whereas utomilumab demonstrates no dose-limiting toxicity but weaker potency on a human IgG2 backbone. FS222, although a human IgG1, had no Fc-mediated effector function, and its potent CD137 activity was dependent upon PD-L1 expression. This resulted in a highly active molecule as seen in vitro and in vivo in multiple syngeneic tumor models, with no liver toxicity. Furthermore, the results from our preliminary toxicology study indicated that FS222, which is cross-reactive with cynomolgus monkeys and has the same in vitro potency in this species and human, had potent in vivo pharmacologic activity in the cynomolgus monkey and is well tolerated up to 30 mg/kg.

Despite being able to bind cell-expressed human CD137, the Fcab component of FS222 was unable to cluster and activate CD137 in the absence of PD-L1–mediated cross-linking, a significant safety feature of the molecule. Coupled with directing CD137 activity to areas of PD-L1 expression, for example tumor microenvironments, FS222 is designed to overcome the adverse side effects associated with CD137 agonists currently in the clinic. This is further strengthened by reducing FcγR binding and allows FS222 to not rely on FcγR-expressing cells to provide the cross-linking necessary for CD137 clustering in current mAb therapies. The combination approach adopted for in vitro experiments of using two separate monospecific antibodies relied not only on releasing the PD-1/PD-L1 blockade via one antibody, but also on maximum cross-linking of CD137(20H4.9) mAb by a hCH2-specific mAb. For this combination in the clinic, the natural cross-linking mechanism would be via FcγR cross-linking via the Fc region of IgG4-based urelumab (20H4.9). Not only are FcγR-expressing cells found throughout the body, therefore bringing another significant safety concern for aberrant IgG cross-linking and CD137 agonism, they are also varied in prevalence with diverse FcγR expression levels, making them an unreliable source of cross-linking–dependent activation within a tumor (23). Therefore, FS222 mitigated high systemic toxicity and variable antitumor activity by not relying on FcγR cross-linking for potent site-specific activity. FS222 also did not rely on Fc-mediated cell killing as a mechanism of action on account of significantly reduced FcγR binding. This resulted in highly potent activity which did not come at the loss of important CD137- or PD-L1–expressing immune cells that can then potentiate the cell-mediated tumor killing action.

FS222 showed potent activity in human primary T-cell assays, but only when PD-L1–expressing cells were present. There was also no activity when the CD137 Fcab was paired with an irrelevant, non–PD-L1 binding Fab domain, HelD1.3. Surrogate FS222 had comparable in vitro activity to FS222, therefore justifying its use in syngeneic mouse tumor models. We believe the tumor control activity shown by surrogate FS222 addresses one of the hurdles of treating a PD-L1–insensitive tumor, or one that has become refractory to PD-L1 therapy. It does this by harnessing PD-L1 target expression in an alternative way to exert direct cytotoxic T-cell activation through CD137 engagement and clustering. The highly immunogenic MC38 tumor model demonstrated insensitivity to PD-L1 mAb treatment and Fc-disabled CD137 mAb treatment as monotherapies at the dose levels employed in this in vivo study. However, despite also being Fc-disabled, treatment with surrogate FS222 resulted in complete tumor eradication and 100% animal survival. In this model, PD-L1 expression, presumably in the tumor microenvironment, provided a setting in which surrogate FS222 can exert superior activity and efficacy to either monotherapy. This is clearly through PD-L1–dependent cross-linking of FS222 and CD137 receptor clustering on T cells resulting ultimately in enhanced tumor-specific CD8⁺ T-cell cytotoxic activity. Surrogate FS222 and therefore FS222 could be bridging a PD-L1–expressing tumor cell and tumor-infiltrating T cell, localizing T-cell cytotoxic activation to the tumor cell/T-cell interface. The same is true for the significant activity and superiority over monotherapy of surrogate FS222 in the less immunogenic CT26.WT model (24), which in our hands is also insensitive to PD-L1 mAb treatment.

The PD changes after treatment with surrogate FS222 in a CT26.WT tumor–bearing mouse model were investigated after administration of a single high (10 mg/kg) and a single low (1 mg/kg) dose. Strikingly, T cells with bound drug were present in the tumor at 2 hours after administration for both dose levels. These T cells had prolonged PD-L1 occupancy specifically in the tumor, the longevity, but not magnitude, of which was correlated with dose. By the end of the study (8 days after drug administration), intratumoral PD-L1 occupancy was still approximately 80% on T cells at 10 mg/kg, whereas peripheral PD-L1 occupancy on T cells had decreased substantially. This highlights the potential of surrogate FS222 to locate to the tumor microenvironment, in preference to remaining in the periphery. Evidence of cytokine production as a consequence of surrogate FS222 treatment was also observed in the serum of mice treated with 10 mg/kg, whereas this was less pronounced with the lower dose level perhaps indicating efficacy without systemic cytokine exposure.

Given the relevance of preclinical studies in mice for risk assessment of severe hepatotoxicity in human patients treated with CD137 agonist agents, the lack of hepatotoxicity in mice in these studies indicates that a mAb² agonizing CD137 via PD-L1–mediated cross-linking has a significantly reduced risk of inducing hepatotoxicity in human patients treated at therapeutic doses. FS222 had a mean terminal half-life of approximately 6 days in cynomolgus monkeys, in line with PD-L1–targeting antibodies such as atezolizumab and followed a linear dose response at dose levels ≥ 1 mg/kg. FS222 was generally well tolerated up to 30 mg/kg dosed weekly. With comparable potency between the cynomolgus and human immune systems, we believe these findings will translate successfully to a human setting. Both the surrogate molecule in mouse and FS222 in cynomolgus monkeys caused a drug-related increase in T-cell proliferation and activation as measured by Ki67 expression. This would indicate that the PD responses seen in our mouse models translate to the effect of FS222 on cynomolgus monkey T cells.

In summary, we have developed FS222, a CD137/PD-L1 tetravalent bispecific antibody with a novel mode of action, and potentially improved therapeutic index for the treatment of human cancer. FS222 did not cause evident toxicity in cynomolgus monkeys upon repeated dosing which we believe further encourages clinical development targeted at tumors where a significant unmet medical need exists in immunotherapy. Checkpoint inhibitors are failing or only providing modest clinical benefit in many tumor settings, and for many of those, we feel there is a mechanistic rationale for improvement in clinical outcomes with FS222.

All authors are current or former employees of F-star Therapeutics Ltd.

Conception and design: M.A. Lakins, A. Koers, J. Munoz-Olaya, S. Batey, D. Gliddon, M. Tuna, N. Brewis

Development of methodology: M.A. Lakins, A. Koers, R. Giambalvo, M. Tuna

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.A. Lakins, A. Koers, R. Giambalvo, R. Hughes, E. Goodman, F. Wollerton

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.A. Lakins, A. Koers, R. Giambalvo, J. Munoz-Olaya, R. Hughes, E. Goodman, S. Marshall, F. Wollerton, S. Batey, D. Gliddon

Writing, review, and/or revision of the manuscript: M.A. Lakins, J. Munoz-Olaya, R. Hughes, E. Goodman, S. Marshall, F. Wollerton, S. Batey, D. Gliddon, M. Tuna, N. Brewis

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.A. Lakins, S. Marshall

Study supervision: M.A. Lakins, M. Tuna, N. Brewis

Other (Coordinated cynomolgus monkey DRF study and associated data analysis and interpretation): S. Marshall

The authors would like to thank the F-star Protein Sciences, in vivo, Assay Development and Drug Discovery team; Cristian Gradinaru for statistical analyses; Jacqueline Doody for scientific contributions; Alison McGhee for critical review; Babraham BSU staff members for animal husbandry and technical assistance; Dr. Sarah Taplin for pathology assessment; and Dr. Sarah Burl and Natalie Allen for article editing and review.

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.