FS118, a Bispecific Antibody Targeting LAG-3 and PD-L1, Enhances T-Cell Activation Resulting in Potent Antitumor Activity

Immunotherapy has significantly improved the clinical outcome of patients with cancer. Targeting of immune checkpoint inhibitors such as programmed cell death protein 1 (PD-1) or its ligand (PD-L1) shifts immune cells within the tumor microenvironment (TME) from an exhausted or immunosuppressive state to a proinflammatory state favoring antitumor response. Unfortunately, only a limited proportion of patients benefit from these therapies due to the development of resistance mechanisms within the TME.

PD-1, present on the surface of immune cells, has a role in downregulating immune responses and promoting self-tolerance by suppressing T-cell inflammatory activity. This prevents autoimmune diseases, but it can also prevent the immune system from killing cancer cells. PD-L1 is expressed by a broad range of both nonhematopoietic and immune cells. It has also been detected in a variety of solid tumors (1, 2) on both tumor cells and host immune cells. A wide range of tumors upregulate PD-L1 in response to proinflammatory cytokines such as IFNγ (3, 4). Engagement of PD-L1 with PD-1 on activated tumor-infiltrating lymphocytes (TILs) can deliver inhibitory signals to protect the tumor from immune destruction (5, 6).

Preclinical (6–10) and clinical data (11–13) have demonstrated that blockade of PD-1/PD-L1 interaction can restore functional T-cell responses in several clinical settings, providing clinical benefit to patients with advanced cancer. Immune checkpoint blockade (ICB) has been integrated in treatment regimen of patients with advanced, relapsed/refractory and, more recently, in patients with earlier stages of disease (14–19).

Currently, several PD-L1 antibodies are in development or approved for various cancer indications. Although these agents provide encouraging responses, only a small percentage of patients respond to these treatments with many showing innate and adaptive resistance to therapy (20, 21). Emerging data suggest that a mechanism of adaptive resistance to PD-1 or PD-L1 therapy is the upregulation of additional checkpoint molecules such as lymphocyte activation gene-3 (LAG-3; refs. 22–27).

LAG-3 is expressed on the surface of B cells, natural killer cells, plasmacytoid dendritic cells, γδ T cells, and activated T cells where it also functions as a negative regulator of T-cell function (28–30). Expression of LAG-3 on exhausted T cells or regulatory T cells (Tregs) in patients' tumors may also be a key factor responsible for resistance to therapies targeting PD-1 or PD-L1 (31, 32). LAG-3 is upregulated on both human and mouse TILs and several studies blocking LAG-3 have shown enhanced antitumor T-cell responses (33–36). Several LAG-3–targeting molecules are in clinical development, but while these are well-tolerated, only moderate responses as monotherapies have been reported (37–39).

The combination of LAG-3 and PD-1 mAbs has demonstrated synergistic improvement of the antitumor response in murine tumor models compared with blocking either one alone (36, 40). Clinical activity of the combination has also been established with a LAG-3–blocking mAb (relatlimab) plus PD-1 mAb (nivolumab) combination (22) and a number of LAG-3–targeting approaches are being investigated in clinical trials. In a small cohort of heavily pretreated melanoma patients, the majority of whom were PD-1/PD-L1 relapsed or refractory, LAG-3 expression enriched for responsiveness to the combination of relatlimab plus nivolumab, with an approximately threefold increase in overall response rate where TIL LAG-3 expression was ≥1%.

FS118 was developed as a human tetravalent bispecific antibody (mAb²) that is a full-length human IgG₁ PD-L1 mAb with a distinct LAG-3–binding capability introduced into the fragment crystallizable (Fc) region (named Fcab: Fc with antigen binding; ref. 41). It is anticipated that such a bispecific antibody could provide the benefit of ICB combination approaches with additional efficacy and safety benefits through more efficient recruitment and activation of tumor-specific T cells. To minimize antibody-dependent cell-mediated cytotoxicity, additional mutations [leucine-to-alanine amino acid substitutions (LALA) at positions 1.3 and 1.2 (IMGT)] were engineered into the CH2 domain to decrease Fcγ receptor binding (42). Further to that, FS118 has been through a standard manufacturing development process and the expression titer of the FS118 clonal manufacturing cell line is within the industry standards for mammalian expression systems. Fcabs are designed to retain key structural and functional properties of the wild-type Fc fragments, retain FcRn binding, and can be purified by protein A resins (43). The activity of FS118 was characterized with in vitro pharmacologic studies using cell lines and primary human immune cells. To further understand the in vivo mechanism of action of FS118, a surrogate mouse bispecific antibody (mLAG-3/PD-L1 mAb²) was created (41) and tested in vitro and in vivo. Both the human FS118 and mouse surrogate mLAG-3/PD-L1 mAb² demonstrated simultaneous binding and neutralization of human or mouse LAG-3 and PD-L1 inhibitory signaling that were comparable with the combination of LAG-3 and PD-L1 single antibody treatment. In vivo mouse studies using two different murine genetic backgrounds demonstrate significant tumor growth inhibition (TGI). Moreover, the mLAG-3/PD-L1 mAb² exhibited a novel mechanism of blocking immune-mediated suppression in vivo compared with the administration of a single antibody.

Human embryonic kidney (HEK) 293 cell lines (Thermo Fisher Scientific) ectopically expressing human or mouse PD-L1 or LAG-3 were generated (HEK293-hLAG-3, HEK293-mLAG-3, HEK293-hPD-L1, and HEK293-mPD-L1). In addition, lentivirus constructs were created by cloning human or mouse LAG-3 and PD-L1 into lentiviral vectors [Lenti-X expression vector (pLVX); Clontech, 631253] and viral particles were generated using the Lenti-X HTX Packaging System (Clontech, 632180). The chicken ovalbumin (OVA aa 323–339)-specific DO11.10 murine T-cell hybridoma cell line expressing endogenous PD-1 (Jewish National Health) or the LK35.2 B cell line were transduced with the LAG-3 or PD-L1 lentiviral particles, respectively, to ectopically express the receptors on the cell surface: DO11.10-mLAG-3, DO11.10-hLAG-3, LK35.2-mPD-L1, and LK35.2-hPD-L1. The cell lines were also transduced with empty particles (pLVX): DO11.10(pLVX) and LK35.2(pLVX). Raji cells were purchased from ATCC. All cell lines were tested for Mycoplasma contamination internally every 4 weeks using a R&D MycoProbe Mycoplasma Detection Kit, and were authenticated by IDEXX BioAnalytics providing a short tandem repeat profile, IMPACT PCR profile, and testing for interspecies contamination.

All antibodies, except the commercial rat mouse IgG LAG-3 antibody clone C9B7W (BioXCell), were fully human IgG₁ antibodies and contained the LALA (amino acid) mutation in the Fc region.

FS118 was constructed by combining a human LAG-3 Fcab (41) with the fragment antigen-binding (Fab) domains of an IgG₁ human PD-L1 (hPD-L1) mAb (“hPD-L1 mAb”). During the drug discovery process, the optimal pairing of Fab and Fcab was selected using the DO11.10 T-cell activation system described below. A mouse LAG-3/PD-L1 mAb² was generated for use as a FS118 surrogate in murine tumor models because FS118 has no functional activity in the mouse. The mLAG-3/PD-L1 mAb² was constructed by combining a mouse LAG-3 Fcab with the Fab domains of an IgG₁ mouse cross-reactive PD-L1 mAb (G1-AA/S1, sequence from clone YW243.55.S1, Genentech patent publication US20100203056A1; “mPD-L1 mAb”). To characterize the properties of the human and surrogate Fcabs in the mAb² format, a “mock” LAG-3 mAb² (LAG-3/mock mAb²) was generated and contained the human or the mouse LAG-3 Fcab and a Fab region, which can bind to a nonrelevant antigen, the FITC (clone 4420; refs. 41, 44). The human and mouse LAG-3/mock mAb² are called “hLAG-3/mock mAb²” and “mLAG-3/mock mAb²,” respectively. Other comparator antibodies included G1-AA/S1 (“mPD-L1 mAb”), the human IgG₁ isotype control (anti-FITC, G1-AA/4420), and rat monoclonal mouse LAG-3 antibody [clone C9B7W, BioXCell; “mLAG-3 mAb (Rat IgG)”]. The sequence of the Fab domain of the rat mouse LAG-3 mAb was used to generate a chimeric mouse LAG-3 antibody with a human IgG₁ Fc backbone (“mLAG-3 mAb”).

Biacore T200 or Biacore 3000 Instruments (GE Healthcare) were used to assess the binding affinity and kinetics of FS118 or mLAG-3/PD-L1 mAb² to recombinant human or mouse LAG-3 or PD-L1, respectively, by surface plasmon resonance (SPR).

For the human protein studies, FS118 or control human PD-L1 mAb were captured on a sensor chip by immobilized protein L (Thermo Fisher Scientific), and various concentrations of human LAG-3-Fc (R&D Systems) or human PD-L1-Fc (R&D Systems) were flowed across the chip. Simultaneous binding of FS118 to human LAG-3 and PD-L1 was also assessed by SPR using LAG-3 and PD-L1 Fc fusion proteins. Human PD-L1-Fc was directly immobilized on the carboxymethyl dextran-5 (CM5) sensor chip surface, and FS118 or the comparator human PD-L1 mAb, was captured by bound PD-L1-Fc. Then the LAG-3-Fc was flowed across the bound antibody.

For the mouse protein studies, recombinant mouse LAG-3-Fc (R&D Systems) or mouse PD-L1-Fc (R&D Systems) were directly immobilized onto a CM5 sensor chip followed by an injection of serially diluted mLAG-3/PD-L1 mAb² to generate binding curves and equilibrium K_D values. Finally, to demonstrate mLAG-3/PD-L1 mAb² could simultaneously bind both mouse LAG-3 and PD-L1, mPD-L1-Fc was covalently linked to the surface of a CM5 sensor chip. The mLAG-3/PD-L1 mAb² flowed across at a fixed concentration, and then mLAG-3-Fc was applied.

An IL2 release assay utilizing both the DO11.10 T cells and LK35.2 B cell lines was set up to assess the function of mAb² and control antibodies. DO11.10 T cells transduced either with human, mouse LAG-3 construct, or an empty vector, and LK35.2 B cells transduced either with human, mouse PD-L1, or empty vector construct were used in the assay. Briefly, serial dilutions of mAb² or control antibody were prepared in media (DMEM, 10% FBS, and 1 mmol/L sodium pyruvate). Diluted mAb² or control antibodies were added to a mixture of DO11.10 T cells (2 × 10⁴ cells) and LK35.2 B cells (antigen-presenting cells, 2 × 10⁴ cells) in presence of OVA peptide (H-ISQAVHAAHAEINEAGR-OH, Pepscan; 1 μmol/L) and incubated for 24 hours. Supernatants were collected and assayed with a mouse IL2 ELISA Kit (eBioscience or R&D Systems) following the manufacturer's instructions. Plates were read on a plate reader using Gen5 Software (BioTek).

For the Staphylococcal enterotoxin B (SEB) assay, human peripheral blood mononuclear cells (PBMCs) were isolated from leukocyte cones (anonymized samples from consenting donors at NHS Blood and Transplant Service) by layering over Ficoll-Paque Plus per the manufacturer's instructions (GE Healthcare). CD4⁺ T cells and CD14⁺ monocytes were isolated by magnetic selection (Miltenyi Isolation Kits, Miltenyi Biotec). CD4⁺ T cells were activated with αCD3/CD28 Beads (Thermo Fisher Scientific) for 8 days prior to freezing. Autologous monocytes were cultured and differentiated into immature dendritic cells (iDCs) for 7 days and then frozen. For coculture and stimulation, cells were thawed and then cocultured (1 × 10⁵ T cells plus 1 × 10⁴ iDC) for 4 days with SEB (Sigma) at a single concentration (0.1 ng/mL) in the presence of various concentration of test antibodies. Culture supernatants were harvested and analyzed for IFNγ release by ELISA (Human IFN-gamma Quantikine ELISA Kit, R&D Systems). Assays were performed in triplicate.

Three mouse colon carcinoma cell lines, MC38 (NCI), MC38.OVA (an engineered MC38 cell line expressing ovalbumin), and CT26 (ATCC) were used. Tumor model studies used 8- to 10-week-old female C57BL/6 mice (Jackson Laboratory) or 8- to 11-week-old female Balb/c mice (Charles River Laboratories). Experiments were conducted under a United Kingdom Home Office Project License and approved by an Animal Welfare and Ethical Review Board in accordance with the United Kingdom Animal (Scientific Procedures) Act 1986 and with EU Directive EU 86/609. For TGI studies, mice were inoculated subcutaneously with 1 × 10⁶ MC38 cells in C57BL/6 mice or 1 × 10⁵ CT26 cells in Balb/c mice. When tumors were palpable, 6 days postinoculation (CT26 tumor–bearing mice) or 8 days postinoculation (MC38 tumor–bearing mice), respectively, the animals received an intraperitoneal injection of the test antibodies every 3 days for a total of three administrations. Tumor volume (mm³) was measured throughout the study using calipers. The following formula was used to calculate tumor volume: (length × width²)/2. In the study where combination therapy was investigated, animals receiving a single test antibody treatment were coinjected with a control antibody [10 mg/kg rat (HRP-1, BioXCell) or human IgG₁ (G1-AA/4420 isotype control)] to ensure that mice received the same amount of total antibody as the mice treated with combination of single antibodies.

For the tumor immune-profiling study, C57BL/6 mice were injected subcutaneously with 1.5 × 10⁶ MC38.OVA or MC38 cells. When the tumor reached a size of approximately 65 mm³, the animals were treated with a single intraperitoneal dose of the test antibodies. Mice were sacrificed at different time points postdosing, then tumors and spleens were collected, processed, and analyzed by flow cytometry. Serum was isolated from peripheral blood to assess the level of soluble LAG-3 (sLAG-3) and PD-L1 proteins.

C57BL/6 mice were injected subcutaneously with 1 × 10⁶ MC38 tumor cells (NIH). Two days after inoculation, the first dose of 10 mg/kg CD8-depleting antibody (clone 2.43, IgG2b, 2BScientific) was given intraperitoneally. Depleting antibody (10 mg/kg) was given 1 day prior to dosing with 10 mg/kg mLAG-3/PD-L1 mAb² or 10 mg/kg IgG₁ isotype control (G1-AA/4420) for a total of three administrations, and a last time at day 11, 1 day after the last dose with mLAG-3/PD-L1 mAb². Tumor volume (mm³) was measured throughout the study using calipers. The following formula was used to calculate tumor volume: (length × width²)/2.

Single-cell suspensions of splenocytes were obtained following mechanical dissociation through a 70-μm nylon cell strainer and treatment with red blood cell lysis (RBC Lysis Buffer, Miltenyi Biotec). Mouse tumor tissues were mechanically dissociated with the GentleMACS Mouse Tumor Dissociation Kit (Miltenyi Biotec) following the manufacturer's instructions. Cell suspensions were washed with RPMI (Gibco) and filtered through 70-μm cell strainer (BD Falcon).

Cells were then stained with a viability Dye Live/Dead e780 (eBioscience) or fixable Near-IR Dead Cell Stain (Life Technologies) and then blocked with anti-CD16/32 (eBioscience) before staining with fluorescence-conjugated antibodies in flow cytometry–staining buffer. For surface staining the following antibodies were used: CD45-BV510 and CD45-Alexa 700 (clone 30-F11) from BioLegend and eBioscience, respectively; CD8a-BUV395 and CD8a-BUV737 (clone 53-6.7) from BioLegend and BD Bioscience, respectively; CD4-BV786 and CD4-BUV395 (clone RM4-5) from eBioscience and BD Bioscience, respectively; CD3e-BUV737 and CD3e-Alexa 488 (clone 145-2C11) from BD Bioscience and eBioscience, respectively; LAG-3-PE (clone C9B7W) from BD Biosciences; and PD-L1-Alexa488 [clone G1-AA/S1 conjugated with Alexa488 using the Lightning Link Rapid Alexa488 Labeling Kit (Innova Bioscience)]. Intracellular staining was performed using the FoxP3/Transcription Factor Staining Buffer Set (eBioscience). The following antibodies were used for intracellular staining: Foxp3-APC and FoxP3-PerCP-Cy5.5 (clone FJK-16s) from eBioscience and Invitrogen, respectively, and Ki-67-PE-Cy7 (clone SolA15) from eBioscience. Data were acquired on a LSRFortessa (BD Biosciences) and analysis performed using FlowJo (Tree Star, v10).

Meso scale discovery (MSD) technology platform was used to measure total sLAG-3 from the serum of tumor-bearing mice. Briefly, biotinylated goat polyclonal antibodies against-mouse LAG-3 (R&D Systems) were captured on MSD streptavidin plates and washed three times with PBS/0.05% Tween. Serum samples and standards were added to the plates and incubated. Following washes, sulfo-tagged detection goat polyclonal anti-mouse LAG-3 antibody was added and incubated. The plates were then washed and read on the MSD instrument. Total mouse soluble PD-L1 concentrations in the mouse serum samples were determined using the mouse PD-L1/B7-H1 PicoKine ELISA Kit (Boster) following the manufacturer's instructions.

Statistical analyses were performed with GraphPad Prism 5 software. Nonparametric testing was performed and means ± SEM or means ± SD are shown on all graphs.

STATA/IC 15.1 software was used to implement a mixed model for analysis of the tumor growth rate. Statistical significance was shown pairwise for growth rates over the full time of study using the mixed-model analysis comparing all groups. A separate model was fitted to each pair of treatments of interest. The model was

A and B are the intercept and slope, respectively; they are different for each mouse, and include a fixed effect for the group and a random effect for the animal:

T is a dummy variable representing the treatment group with value 0 in one group and 1 in the other. The random effects are distributed with a normal distribution:

where σ_A and σ_B are the SDs of the interanimal variability in the intercept and slope, respectively. The intra-animal variability is also normally distributed with SD σ:

For each pair of treatments, the model above was fitted to the data. The model parameters were reported, together with their 95% confidence intervals. For A₁ and B₁, the (two-sided) P value for a difference from zero was calculated; a P value below 0.05 is statistically significant evidence for a difference between the treatment groups.

FS118, hLAG-3/PD-L1 mAb², bound to recombinant human LAG-3-Fc with a K_D value of 0.056 nmol/L (Table 1). The binding affinity of FS118 for human PD-L1-Fc (K_D = 1 nmol/L) was similar to that of the hPD-L1 mAb composing the Fab part of the mAb² (K_D = 1.2 nmol/L), indicating that introduction of a LAG-3–binding site into the CH3 domain of the antibody did not affect PD-L1 binding. Concurrent dual-target binding (LAG-3 and PD-L1) by FS118 was also confirmed by SPR where FS118 binding to PD-L1-Fc did not interfere with the bispecific molecule binding to LAG-3-Fc (Fig. 1). FS118 specifically bound to PD-L1 and LAG-3 because binding to closely related proteins, PD-L2 and CD4, was not observed (Supplementary Fig. S1).

The ability of FS118 to bind to the native form of cell surface–expressed proteins was assessed by flow cytometry. The results demonstrated strong binding of FS118 to human LAG-3 (EC₅₀ = 3.2 nmol/L) and human PD-L1 (EC₅₀ = 3.2 nmol/L) expressed on the surface of HEK293 cells (Table 1; Supplementary Fig. S2). Binding of FS118 to LAG-3 and PD-L1 in a concentration-dependent manner was also confirmed on CD3/CD28-activated human CD4⁺ T cells by flow cytometry (Supplementary Fig. S2).

Moreover, the ability of FS118 to block binding of LAG-3 and PD-L1 to their respective ligands was measured. FS118 inhibited binding of LAG-3 to Raji cells endogenously expressing MHC class II (MHCII). In the same way, FS118 also inhibited PD-L1 binding to PD-1 or CD80 (Supplementary Fig. S3).

FS118 was tested in a LAG-3/PD-L1–dependent T-cell assay where OVA_323–339-specific mouse DO11.10 T cells expressing human LAG-3, or nonexpressing control cells were incubated with LK35.2 B cells (antigen-presenting cells) expressing, or not, human PD-L1 in presence of OVA peptide and isotype control or test antibodies. The addition of FS118 to the cultured cells reversed T-cell inhibition mediated by LAG-3 or PD-L1 (Fig. 2A and B; Supplementary Table S1). FS118 demonstrated greater activity than the combination of its component parts hLAG-3/mock mAb² plus hPD-L1 mAb (EC₅₀ = 0.61 nmol/L and 2.07 nmol/L, respectively), when both human LAG-3 and human PD-L1 were overexpressed (Fig. 2C). When only human LAG-3 or human PD-L1 was overexpressed, FS118 showed similar activity compared with the single or the combination component antibodies.

FS118 and single-component antibodies were also tested in a primary SEB-activated human T-cell assay. The magnitude of the effect of test antibodies in the SEB T-cell assay was donor dependent (Fig. 2D). However, FS118 and the comparators, hPD-L1 mAb and hLAG-3/mock mAb² plus hPD-L1 mAb, increased IFNγ levels above that of the isotype control in all donor samples. FS118 demonstrated similar or, in some donors, enhanced T-cell activation compared with the single-agent combination. The potency values (EC₅₀) of FS118 were on average, 0.125 nmol/L. FS118 was also able to enhance CD8⁺ T-cell cytokine production and cytotoxicity in an antigen recall assay, as measured by intracellular IFNγ and CD107a staining (Supplementary Fig. S4).

Overall, these data demonstrate that FS118 can reverse PD-L1- and LAG-3–mediated inhibition of T-cell activation and effector function.

As FS118 showed no functionality in a mouse LAG-3 and PD-L1 assay system (data not shown), a murine surrogate mAb² (mLAG-3/PD-L1 mAb²) was generated to evaluate the potential for a mLAG-3/PD-L1 mAb² to mediate an antitumor response in mouse tumor models. The binding and function of mLAG-3/PD-L1 mAb² in mouse systems was compared with that of FS118 in human systems (Table 1; Supplementary Fig. S5).

In a SPR assay, mLAG-3/PD-L1 mAb² bound to immobilized mLAG-3-Fc fusion protein with a K_D of 0.98 nmol/L and to mPD-L1-Fc with a K_D of 0.09 nmol/L. Simultaneous binding of LAG-3 and PD-L1 was also demonstrated by SPR (Supplementary Fig. S5A).

In addition, mLAG-3/PD-L1 mAb² bound to mouse LAG-3 or mouse PD-L1 overexpressed on HEK293 cells in a concentration-dependent manner (EC₅₀ values of 2.3 and 12 nmol/L, respectively; Table 1; Supplementary Fig. S5B).

The functional activity of mLAG-3/PD-L1 mAb² was evaluated in the DO11.10 T-cell activation system. When both mouse LAG-3 and mouse PD-L1 are expressed in the same assay format as described in Fig. 2A–C, mLAG-3/PD-L1 mAb² activated T cells in a dose-dependent manner comparable with the combination of the mouse LAG-3 (mLAG-3/mock mAb²) and PD-L1 mAb (Supplementary Fig. S5C). The single-agent controls (mouse LAG-3 or mouse PD-L1 mAb alone) did not increase T-cell activation when both mouse targets were overexpressed. (Supplementary Fig. S5C).

While the binding affinity of mLAG-3/PD-L1 mAb² for recombinant mouse LAG-3 is approximately 10-fold lower than FS118 affinity for human LAG-3 and binds to mouse PD-L1 with approximately 10-fold higher affinity than FS118 for human PD-L1, functional activity of mLAG-3/PD-L1 mAb² and FS118 against each single target (LAG-3 or PD-L1) within their corresponding species (mouse or human) was found to be similar (Table 1). Although the functional activity of FS118 (EC₅₀ = 0.61 nmol/L) was greater than mLAG-3/PD-L1 mAb² (EC₅₀ = 5.8 nmol/L) when both LAG-3 and PD-L1 were expressed, this difference was within 10-fold and therefore mLAG-3/PD-L1 mAb² was considered a suitable surrogate of FS118 for mouse studies.

To test whether mLAG-3/PD-L1 mAb² has antitumor activity, tumor-bearing mice were administered with mLAG-3/PD-L1 mAb² or IgG₁ control. mLAG-3/PD-L1 mAb² had significant antitumor activity in both the C57BL/6 MC38 and Balb/c CT26 tumor models (Fig. 3A and B, respectively). In all studies performed, mLAG-3/PD-L1 mAb² was well-tolerated at all dose levels tested with no signs of weight loss or toxicity.

A dose–response experiment was performed in the MC38 tumor model to understand the optimal dose level for further investigation. Significant differences in mean tumor volume were observed between the IgG₁ control and mLAG-3/PD-L1 mAb²–treated MC38-bearing mice at the 3, 10, or 20 mg/kg dose levels but not at 1 mg/kg (Fig. 3A; Supplementary Fig. S6A). A dose of 3 mg/kg was determined to be the minimal dose level required to drive this effect. Subsequent TGI studies used 10 mg/kg as an optimal dose level to ensure maximal antitumor effects.

Antitumor activity was observed with treatment of mLAG-3/PD-L1 mAb² and compared with mouse LAG-3 (clone C9B7W) and PD-L1 (clone G1-AA/S1) mAb given alone or in combination. mLAG-3/PD-L1 mAb² eliminated the tumor in 6 of 8 mice and slowed tumor growth in the remaining 2 mice (Fig. 3C; Supplementary Fig. S6B). The combination of the test antibodies did not induce any complete tumor regressions, although it did slow tumor growth (Supplementary Fig. S6B). Although the administration of the individual mouse LAG-3 antibody or mouse PD-L1 antibody did slow the tumor growth, it did not suppress it to the extent of either the combination or the mAb². Therefore, in the MC38 model, mLAG-3/PD-L1 mAb² induced an antitumor immune response as potent as the combination of PD-L1 and LAG-3 mAbs. The antitumor effect of the mAb² was confirmed to be CD8⁺ T-cell mediated as the mAb² failed to control tumor growth when CD8⁺ T cells were depleted (Supplementary Fig. S7A).

Studies were carried out to investigate the pharmacodynamic effects of a single dose of the mouse LAG-3/PD-L1 mAb² on tumor immune cell infiltrates and on splenic T cells in a mouse tumor model.

Administration of mLAG-3/PD-L1 mAb², single agents alone or in combination resulted in no or only marginal change in the proportion of CD4⁺ or CD8⁺ T cells in TILs (Fig. 4A and B) and splenic cells (Supplementary Fig. S8A). While there was no significant change in the frequencies of Ki-67–positive (proliferating) CD4⁺ or CD8⁺ T cells in the tumor (Supplementary Fig. S8B), treatment with mLAG-3/PD-L1 mAb², the combination of single antibodies or the mouse PD-L1 mAb increased the percentage of proliferating CD4⁺ T cells in the spleen at 96 hours postdosing compared with control IgG₁ (Supplementary Fig. S8C). In a separate experiment, this effect was confirmed to be independent of CD8⁺ T cells (Supplementary Fig. S7B). PD-L1 and LAG-3 antibodies were used to determine the expression of PD-L1 and LAG-3 on tumor-infiltrating T cells. Characterization by flow cytometry demonstrated that the mouse PD-L1 flow cytometry antibody competed with mLAG-3/PD-L1 mAb², and therefore measures free PD-L1 surface expression. The mouse LAG-3 antibody (clone C9B7W) used to detect cell surface LAG-3 was noncompetitive with mLAG-3/PD-L1 mAb² and therefore detected total LAG-3 (free and bound) cell surface receptor. Treatment with mLAG-3/PD-L1 mAb² resulted in a transient decrease over several days in the percentage of LAG-3⁺ T cells in tumor and spleen, while the combination or the mouse PD-L1 mAb induced a transient increase of the frequency of LAG-3⁺ T cells (Fig. 4C–E; Supplementary Fig. S9A). In the mLAG-3/PD-L1 mAb² treatment group, LAG-3 expression on CD4⁺ and Treg TILS was reduced at 24 hours and persisted for 96 hours (Fig. 4C and E; Supplementary Fig. S9A). LAG-3 expression on CD8⁺ TILs was also reduced 8 hours posttreatment with mAb² and remained low for 96 hours (Fig. 4D; Supplementary Fig. S9A). The decrease in LAG-3 expression could not be attributed to a loss of T-cell populations because overall proportion of T cells did not alter following treatment (Fig. 4A and B), nor could the decrease in LAG-3 expression be attributed to internalization of the mAb² as this did not show significant internalization over a 3-hour period in an in vitro assay (Supplementary Fig. S10).

In contrast, treatment with mouse PD-L1 mAb alone appeared to increase expression of LAG-3 on CD4⁺ and CD8⁺ T cells in both the tumor and spleens (Fig. 4C–E; Supplementary Fig. S9A).

Similarly, an increase of free mouse PD-L1 was observed when mice were treated with mLAG-3/mock alone. Interestingly, it was observed that when mouse PD-L1 antibody was administered, LAG-3 expression increased and when mouse LAG-3 mAb treatment was given, free PD-L1 appeared to increase (Fig. 4F and G; Supplementary Fig. S9B).

Treatment with the combination of mLAG-3/mock plus mouse PD-L1 mAb appeared to induce an overall increase in total surface LAG-3 (Fig. 4C and D). Total LAG-3 surface level was more difficult to ascertain because all the commercially available PD-L1 mAbs tested competed with mAb² for PD-L1 binding, although it appeared that combination treatment might increase available PD-L1 in this setting.

Taken together, these data show that treatment with single dose of mLAG-3/PD-L1 mAb² prevented the compensatory upregulation of the LAG-3 receptor on T cells in TILs observed with the combination treatment.

LAG-3 is known to undergo a rapid cleavage event from the cell surface of activated T cells by the transmembrane metalloproteases ADAM 10 and ADAM 17 (45). Therefore, total sLAG-3 (free and bound to antibody) was measured in the serum of treated mice. A single 10 mg/kg treatment with mLAG-3/PD-L1 mAb² or mLAG-3/mock plus PD-L1 significantly increased sLAG-3 in the serum up to fourfold over negative controls. This increase persisted up to 96 hours following dosing (Fig. 5A). Similarly, the levels of total soluble PD-L1 following treatment with mLAG-3/PD-L1 mAb² or mLAG-3/mock plus PD-L1 increased compared with isotype control–treated mice, although this was not apparent until 48 hours posttreatment (Fig. 5B). These results suggest that the mLAG-3/PD-L1 mAb² reduced LAG-3 cell surface expression by accelerating its shedding from the cell surface of T cells.

Despite significant advances made by PD-1:PD-L1 blockade therapy, a large proportion of patients do not derive clinical benefit from these therapies. An underlying resistance mechanism is the upregulation of checkpoint inhibitors such as LAG-3 (22, 23, 27). FS118 is a tetravalent bispecific antibody that was selected on the basis of its ability to overcome the immunosuppressive signaling mediated by both LAG-3 and PD-L1 in patients with cancer. Here, we have reported characterization of the activity of FS118 and unveiled a novel mechanism of action utilizing a surrogate mouse bispecific antibody.

We have demonstrated that FS118 simultaneously bound to LAG-3 and PD-L1, blocked LAG-3:MHCII, PD-L1:PD-1, and PD-L1:CD80 interactions and as a result reversed T-cell inhibition. In a human primary T-cell activation assay, FS118 reversed immune suppression to enhance T-cell activation and was more potent than the combination of individual components of FS118 in some of PBMC donors tested. FS118 enhanced cytokine production by both CD4⁺ (in a SEB assay) and CD8⁺ T cells (in an antigen recall assay), two central players of the antitumor response. These results demonstrated that the LAG-3–targeting Fcab and PD-L1 Fab regions of FS118 can work in synergy to overcome LAG-3- and PD-L1–mediated inhibition of T-cell activation. The stronger T-cell response observed in PBMC donors treated with the bispecific over the combination may be explained by the avidity effects of the molecule created by the dual cis-binding to a single cell or trans-binding between two cells resulting in long-lasting blocking of immunosuppressive signals.

A mouse surrogate of FS118 (mLAG-3/PD-L1 mAb²) was generated to characterize functional activity in vitro and in vivo. mLAG-3/PD-L1 mAb² demonstrated comparable binding properties and in vitro functional activity against mouse LAG-3 and PD-L1 when compared with FS118. The antitumor efficacy of the surrogate mAb² was evaluated in MC38 and CT26 syngeneic mouse tumor models. The results showed the potential to elicit CD8⁺ T-cell–dependent antitumor immune response that was greater than mouse PD-L1 or mouse LAG-3 antibody monotherapy and was at least as effective as the combination of mouse LAG-3 plus mouse PD-L1 mAbs. These data demonstrate that the mLAG-3/PD-L1 mAb² can provide dual blockade of LAG-3 and PD-L1 in vivo and enhance the antitumor immune response.

Treatment with mLAG-3/PD-L1 mAb² was shown to reduce LAG-3 expression on the cell surface of CD4⁺ and CD8⁺ T cells in TILs and splenic T cells. This effect was not observed with PD-L1 blockade alone and therefore suggests that LAG-3 downregulation was due to the LAG-3–targeting moiety in the mAb² format. This contrasts with treatment with mouse LAG-3 antibody that increased surface expression of PD-L1 and treatment with mouse PD-L1 antibody that caused an increase in surface LAG-3. The idea that single targeting of a checkpoint inhibitor receptor could result in the upregulation of an additional checkpoint receptor has been previously reported in murine tumor models (24, 26). This observation could allude to the numbers of patients who become refractory to current monospecific immune therapy as a reflexive immune compensatory response (22, 23). Here, following dosing with mLAG-3/PD-L1 mAb², total LAG-3 was rapidly reduced from the surface of the T cells in tumor and spleen limiting the impact of the compensatory upregulation of LAG-3 on T-cell activation and enhancing the antitumor response. The cross-talk between CD4⁺ and CD8⁺ T cells for suppression of tumor growth is a hallmark feature of the immune system. While the antitumor effect in MC38 tumor mouse model was driven by cytotoxic CD8⁺ T cells, mLAG-3/PD-L1 mAb² increased proliferation of peripheral CD4⁺ cells. This could reflect the activity of CD4⁺ T cells to provide helper signals to tumor -specific CD8⁺ T cells and subsequently augment antitumor immunity.

Although the mechanism regulating LAG-3 expression is not fully understood, one could hypothesize that mLAG-3/PD-L1 mAb² altered the cellular trafficking and/or the shedding of LAG-3 from the cell surface, potentially via the metalloproteases ADAM10 and ADAM17 (45). In unstimulated T cells, LAG-3 exists as an intracellular pool that is localized to the surface upon stimulation (46). Cross-linking of LAG-3 through cis- or trans-binding of the mAb² could potentially regulate LAG-3 surface expression by altering the signaling pathway that controls its translocation from early/recycling endosomal compartments to the cell surface with a putative impact on PD-1 expression and function as well. Indeed, a study by Huang and colleagues suggested that LAG-3 and PD-1 shared intracellular localization and trafficking pathway, and that LAG-3 may enable the rapid translocation of PD-1 from early/recycling endosomal compartments to the immunological synapse (40).

Our data revealed that the action of the mLAG-3/PD-L1 mAb² on LAG-3 cell surface expression correlated with an increased level of sLAG-3 in the serum of the treated animals suggesting that the mLAG-3/PD-L1 mAb² accelerated the shedding of mouse LAG-3 from the cell surface into the blood. Similarly, the combination treatment increased sLAG-3 into the blood following treatment. However, if this increase of sLAG-3 resulted from the shedding of cell surface LAG-3, it did not reduce drastically or prevent the increased cell surface LAG-3 expression observed upon combination therapy. Here, sLAG-3 is hypothesized to be a potential surrogate indicator of T-cell activation in response to LAG-3/PD-L1 mAb² treatment. sLAG-3, through its binding to MHCII, has been reported to stimulate antigen-presenting cells such as macrophages and dendritic cells to activate T-cell responses and enhance tumor-specific cytotoxic T cells (47, 48). The relevance of this mechanism in human cancers is supported by evidence that the presence of sLAG-3 positively correlates with improved survival in patients with breast cancer and gastric cancer (42, 49). In addition to PD-L1 and LAG-3 inhibition, FS118 may promote the recognition and elimination of the tumor by bridging the immune effector cells and the cancer cells favoring a directed and local response.

In conclusion, these preclinical results demonstrate that LAG-3/PD-L1 mAb² have the potential to drive a potent antitumor response and unveil a differentiated mechanism of action compared with the combination therapy. These data support the evaluation of FS118 in patients with cancer.

D. Gliddon is an employee/paid consultant for and holds ownership interest (including patents) in F-star Biotechnology. S. Batey is an employee/paid consultant for F-star Biotechnology. L. Young holds ownership interest (including patents) in F-star. N. Brewis is an employee/paid consultant for, reports receiving commercial research grants from and reports receiving other commercial research support from, and holds ownership interest (including patents) in F. star. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Kraman, D. Gliddon, M.M. Wydro, J. Winnewisser, M. Tuna, M. Morrow, N. Brewis

Development of methodology: M. Kraman, N.L. Allen, K. Kmiecik, C. Seal, S. Batey

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Kraman, N.L. Allen, K. Kmiecik, A. Koers, M.M. Wydro, J. Winnewisser

Analysis and interpretation of data (ie, statistical analysis, biostatistics, computational analysis): M. Kraman, M. Faroudi, N.L. Allen, K. Kmiecik, C. Seal, A. Koers, L. Young, J. Doody, M. Morrow, N. Brewis

Writing, review, and/or revision of the manuscript: M. Kraman, M. Faroudi, N.L. Allen, D. Gliddon, A. Koers, S. Batey, J. Winnewisser, M. Tuna, J. Doody, M. Morrow, N. Brewis

Administrative, technical, or material support (ie, reporting or organizing data, constructing databases): M. Kraman

Study supervision: M. Kraman, D. Gliddon, M. Tuna, J. Doody

Other (antibody generation): M.M. Wydro

The authors would like to thank the F-star protein sciences team; F-star in vivo team; F-star drug discovery team; Cristian Gradinaru for statistical analyses; Alison McGhee for critical review; Babraham BSU staff members for animal husbandry and technical assistance; and Dr. Sarah Burl 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.