Preclinical Development of the Anti-LAG-3 Antibody REGN3767: Characterization and Activity in Combination with the Anti-PD-1 Antibody Cemiplimab in Human PD-1xLAG-3–Knockin Mice Free

Therapeutic mAbs targeting inhibitory receptors such as cytotoxic T-lymphocyte–associated protein 4 (CTLA-4) or programmed cell death 1 (PD-1) show impressive clinical activity with an acceptable benefit risk ratio in several tumor types (1–4). However, sustained responses are only achieved in a minority of patients, suggesting that combination approaches may be required to overcome tumor immune escape mechanisms (5). Antibodies blocking the interaction of the inhibitory receptor lymphocyte-activation gene 3 (LAG-3) with its ligand, MHC class II, may invigorate the immune response to cancer, especially in combination with antibodies blocking the PD-1/PD-L1 axis (reviewed in refs. 6–8).

The inhibitory receptor LAG-3, a CD4-like molecule, is expressed on activated CD4⁺ and CD8⁺ T cells, a subset of regulatory T cells (Treg), natural killer (NK) cells, B cells, and plasmacytoid dendritic cells (DC; refs. 8–11). LAG-3 suppresses T-cell activation, proliferation, and homeostasis (11, 12) and has been reported to play a role in Treg-suppressive function (13). LAG-3 binds to MHC class II, which is expressed on DCs, macrophages, B cells, and epithelial cancer cells, and serves to present peptides to CD4⁺ T cells (14–16). LAG-3 has been shown to be expressed on dysfunctional T cells in chronic viral infections (17, 18) and during tumor progression in patients with cancer (19, 20). CD8⁺ T cells coexpressing PD-1 and LAG-3 display severely impaired effector functions in a murine model of self-antigen tolerance (21), and antigen-specific T cells in human ovarian cancer are negatively regulated by PD-1 and LAG-3 (20). Multiple reports demonstrate that a large fraction of PD-1–expressing CD8⁺ and CD4⁺ tumor-infiltrating lymphocytes (TIL) coexpress LAG-3, supporting the dominance of PD-1 in modulating antitumor T-cell responses, as well as a direct role of LAG-3 in further suppressing the activity of a subset of PD-1–expressing T cells (18, 20–24). Mice deficient for both PD-1 and LAG-3 show spontaneous autoimmunity and lethality, not seen in either single knockout (22, 25). In addition, dual blockade of PD-1 and LAG-3 with antibodies synergistically inhibits tumor growth in preclinical tumor models (22, 23, 26). Thus, it stands to reason that antagonistic anti-LAG-3 antibodies may improve efficacy in combination with anti-PD-1 therapies in the clinic (7, 8, 27–30). Currently, there are several LAG-3–targeting treatments in various phases of clinical development (31–34). In particular, combination therapy of anti-LAG-3 (BMS-986016) plus anti-PD-1 (nivolumab) has shown promising clinical efficacy in patients with melanoma (33, 34).

We have previously generated cemiplimab, a human anti-PD-1 antibody blocking PD-1/PD-L1–mediated T-cell inhibition (35). Here we describe the generation of REGN3767, a fully human high-affinity anti-LAG-3 antibody, using VelocImmune mice containing human immunoglobulin gene segments (36, 37). In vitro, REGN3767 binds LAG-3 with high affinity and blocks LAG-3/MHC class II–driven T-cell inhibition. In vivo, REGN3767 enhances the antitumor activity of the anti-human PD-1 antibody cemiplimab against syngeneic colorectal carcinomas in mice humanized for the extracellular domains of PD-1 and LAG-3. The preclinical results presented here supported the initiation of clinical trials using REGN3767 in monotherapy or in combination with cemiplimab in patients with solid tumors.

To generate anti-human LAG-3 antibodies, VelocImmune mice, carrying genes encoding human immunoglobulin heavy and kappa light chain variable regions (36, 37), were immunized with recombinant human LAG-3–mFc protein (Regeneron), containing the extracellular domain of LAG-3 (amino acids 1–450) and the Fc portion of mouse IgG2a. Hybridomas were generated from splenocytes and supernatants were screened for binding to HEK293 cells transfected with human LAG-3. REGN3767 was engineered as a human IgG4 containing a serine to proline substitution (S228P) in the hinge region to minimize half-antibody formation (38). To reduce binding to Fcγ receptors, REGN3767 also contains three amino acid substitutions (P₂₃₆VA₂₃₈) derived from the lower hinge region of IgG2 replacing four amino acids (E₂₃₆FLG₂₃₉) in the corresponding lower hinge region of IgG4 (39). The amino acid sequences of the heavy and light chains of REGN3767 (GenBank accession numbers: MN200290 and MN200291) are shown in Supplementary Fig. S1. The hIgG4 (S228P) anti-human PD-1 antibody cemiplimab (REGN2810), that binds to both human and monkey PD-1 proteins and blocks PD-1 interactions with PD-L1 and PD-L2, was described previously (35).

Binding kinetics of REGN3767 mAb to LAG-3 proteins at 25°C and pH 7.4 was measured in Surface Plasmon Resonance-Biacore studies by first capturing REGN3767 on a CM5 (Biacore) surface chip precoated with anti-human kappa goat polyclonal F(ab′)2 Ab (GE Life Sciences) and then injecting various concentrations of LAG-3 proteins over the surface. Following capture, hLAG-3.mmH (myc-myc-polyhistidine tag), rLAG-3.mmH, or mLAG-3.mmH at concentrations ranging from 50 to 0.78 nmol/L, or hLAG-3.hFc protein ranging from 25 to 0.39 nmol/L were individually injected over the REGN3767-captured surfaces. The kinetic parameters were obtained by globally fitting the data to a 1:1 binding model using Biacore T200 Evaluation.

The ability of REGN3767 to block the binding of hLAG-3 to human MHC II–positive (Raji) B cells or murine MHC II–positive (A20) B cells was assessed using a cell–cell adherence assay. Fluorescently labeled Raji or A20 cells were examined for adherence to HEK293/hLAG-3 cells in the presence or absence of REGN3767. The parental HEK293 cell line has no detectable expression of human LAG-3 as determined by flow cytometry. To confirm expression of MHC II, Raji cells were stained with anti-human HLA-DR (clone Tu36, BD Biosciences) and A20 cells were stained with anti-mouse I-A/I-E (clone M5/114.15.2, BioLegend). Raji or A20 cells were labeled with Calcein AM (Life Technologies). Human Fc block (BD Pharmingen) was added to prelabeled Raji cells and mouse Fc block (BD Pharmingen) was added to A20 cells at a final concentration of 10 μg/mL. HEK293 parental and HEK293/hLAG-3 cells were added at 1.2 × 10⁴ cells/100 μL into 96-well plates and cultured overnight. REGN3767 or the isotype matched control antibody REGN2759 (Regeneron) were added at increasing concentrations for 1 hour, followed by incubation with 1.2 × 10⁵ labeled Raji or A20 cells for 1 hour. Nonadherent cells were removed by washing. Relative fluorescence units of adherent labeled Raji or A20 cells were measured at an excitation/emission wavelength of 485 nm/535 nm on a VICTOR X5 plate reader.

VelociGene technology was used to generate human PD-1xLAG-3 knockin mice as described previously (40). Briefly, a Lag3-targeting vector was engineered that replaced 1750 bp of the extracellular portion of the mouse Lag3 gene (containing exons 2–4) with the corresponding 1741 bp region of the human gene (exons 2–4). Dual humanized mouse embryonic stem cells (ESC) were created by electroporation of the Lag3-targeting vector into C57BL/6N mouse ESCs that contained a humanized Pdcd1 gene encoding the extracellular portion of human PD-1 and the transmembrane and intracellular portion of mouse Pdcd1 (35). Correct gene targeting in ESC clones was identified by a loss of allele assay (41). Dual humanized ESC clones were used to implant female mice to generate a litter of pups containing both humanized genes (i.e., Lag3 and Pdcd1).

MC38.Ova cells were described previously (35) and authenticated by short tandem repeat profiling in 2016 (IDEXX BioResearch). MC38.Ova cells (5 × 10⁵) were injected subcutaneously in the hind flank of 8–10 weeks old female human PD-1xLAG-3 knockin mice. Tumors were measured semiweekly using a caliper and reported as mm³ (length × width²/2). In a prophylactic tumor model, antibodies were administered intraperitoneally in 200 μL starting on day 3 after tumor implantation, and then twice a week for 2 weeks. In an established tumor model, mice were randomized on day 10–11 after implantation when tumors reached 100 mm³. Mice were treated with antibodies on the randomization day and then twice a week for 2 weeks. Mice were euthanized when the tumor volumes reached 2,000 mm³ or when tumors ulcerated. The protocol was approved by the Regeneron Pharmaceuticals Institutional Animal Care and Use Committee.

Tumors were harvested and processed using GentleMacs Cell Disruptors (Miltenyi) to generate single-cell suspensions. Lymphocytes were enriched by Percoll (GE Healthcare) gradient centrifugation, stained with fluorescently labeled antibodies, and analyzed on a FACS Canto Flow Cytometer (BD Biosciences). Ex vivo TILs stimulation was performed with 1 μg/mL Ova_257-264 MHC I peptide and 5 μg/mL Ova_323-339 MHC II peptide (InvivoGen) in the presence of Brefeldin A (1 μg/mL) for 4 hours at 37°C. Draining lymph nodes (dLN) were processed into single-cell suspension by mechanical dissociation. Cells were stimulated with phorbol 12-myristate 13-acetate (PMA; 20 ng/mL) and ionomycin (500 ng/mL) for 4 hours in the presence of Brefeldin A (1 μg/mL), and then processed with the Cytofix/Cytoperm Kit (BD Biosciences) for intracellular cytokine staining. The following antibody clones were used: mCD45 (30-F11, BioLegend), mCD4 (GK1.5, BioLegend), mCD8a (53-6.7, BioLegend), mIFNγ (XMG1.2, eBioscience), mTNFα (MP6-XT22, BioLegend), hPD-1 (EH12.2H7, BioLegend), hLAG-3 (3DS223H, eBioscience), mouse PD-1 (J43, eBioscience), and mouse LAG-3 (C9B7W, BioLegend).

Spleens were harvested from tumor-bearing mice and manually dissociated into single-cell suspensions. Spleen homogenates were clarified by centrifugation and the protein content was quantified using a BCA Assay (Pierce). A total of 50 μg of supernatant from a spleen sample and 25 μL of serum were used in duplicates to measure cytokine concentrations by the V-PLEX Proinflammatory Panel 1 mouse kit according to the manufacturer's instructions (Meso Scale Discovery).

Female cynomolgus monkeys (5 animals/dose group) received a single intravenous infusion (1, 5, or 15 mg/kg) or subcutaneous injection (1 or 15 mg/kg) of REGN3767. Blood samples were collected for measurement of functional REGN3767 concentrations and antidrug antibody (ADA) in serum at predose and postdose at various times for up to 56 days. Concentrations of functional REGN3767 in serum were determined using ELISA. This procedure employed microtiter plates coated with a recombinant protein containing the extracellular domain of human LAG-3 and utilized REGN3767 as the reference standard. REGN3767 captured on the plate was detected using a biotinylated mouse anti-human IgG4 mAb REGN1298 (Regeneron), followed by NeutrAvidin conjugated with horseradish peroxidase.

Clone REGN3767 was selected from a panel of human antibodies generated by immunization of VelocImmune mice transgenic for human variable regions (36, 37). REGN3767 bound specifically to human LAG-3, with a K_D of 3.22 nmol/L and 0.13 nmol/L for monomeric and dimeric recombinant proteins, respectively. No cross-reactivity to mouse and rat LAG-3 was detected, consistent with low amino acid sequence identity between the extracellular regions of human and mouse (70%) or rat (67%) LAG-3. REGN3767 bound HEK293 cells engineered to overexpress human or cynomolgus monkey LAG-3 with a similar EC₅₀ value of 1.1 nmol/L and 0.8 nmol/L, respectively (Fig. 1A). Flow cytometry confirmed REGN3767 binding to LAG-3 on activated primary human and cynomolgus monkey CD4⁺ and CD8⁺ T cells (Supplementary Fig. S2A and S2B). REGN3767 bound activated CD8⁺ T cells, expressing either low or high level of early activation marker CD69, with comparable EC₅₀ values of 0.65–1.3 nmol/L and 0.7–1.0 nmol/L for human and cynomolgus monkey donors, respectively (Supplementary Fig. S2C and S2D). The staining intensity was lower on activated CD4⁺ T cells compared with CD8⁺ T cells in both species, suggesting lower LAG-3 expression; this observation is in line with published studies (42, 43). In summary, REGN3767 exhibited similar EC₅₀ values for binding to human and cynomolgus monkey LAG-3 overexpressed on HEK293 cells or expressed endogenously on activated primary cells.

Human LAG-3 binds both human and mouse MHC II (15, 16), so the ability of REGN3767 to block these interactions was assessed using a cell–cell adherence assay. A similar assay was previously used to demonstrate MHC II interactions with CD4 (14) and to identify the domains of LAG-3 responsible for binding to MHC II (16). REGN3767 blocked the binding of MHC II–positive human Raji B cells and murine A20 B cells, respectively, to HEK293/hLAG-3 cells with EC₅₀ values of 4.2 and 7.1 nmol/L, respectively (Fig. 1B). Binding of HEK293/hLAG-3 to Raji cells or A20 cells was reversed when anti-MHC class II antibody (anti-human HLA-DR or anti-mouse HLA I-A/I-E, respectively) was added to the system (Supplementary Fig. S3), suggesting that MHC class II plays a major role in this process. Neither Raji nor A20 cells displayed significant adherence to parental HEK293 cells, demonstrating that LAG-3 expression on HEK293 is required for adherence.

The potential of REGN3767 in combination with cemiplimab to rescue T-cell activity was evaluated in in vitro functional assays. In an engineered T-cell/antigen-presenting cell (APC) luciferase–based bioassay (Supplementary Fig. S4; Supplementary Materials and Methods), activation of engineered JRT3.T3.5 T cells is driven by engagement of a TCR (Ob2F3) with its cognate peptide MBP_85-99 presented on MHC II by engineered HEK293 cells that serve as APCs (44). JRT3.T3.5 T cells were additionally engineered to express chimeric human LAG-3 and human PD-1. Chimeric human LAG-3 was constructed of the human LAG-3 extracellular domain and the cytoplasmic domain of the human inhibitory receptor CD300a, which contains immunoreceptor tyrosine-based inhibition motifs. APC express human MHC II with or without exogenously expressed human PD-L1. In both T-cell/APC formats, REGN3767 was titrated in the presence of 30 nmol/L of cemiplimab. In the absence of PD-1/PD-L1 signaling, REGN3767 restored T-cell activity with an EC₅₀ value of 4.7 nmol/L; the activity was unaffected by the presence of cemiplimab, as expected. When the bioassay included the PD-1/PD-L1 signaling axis, REGN3767 increased T-cell activation when combined with cemiplimab.

In an additional in vitro functional assay of allogeneic T cells: peripheral blood mononuclear cell (PBMC) mixed lymphocyte reaction (MLR), cemiplimab or REGN3767, used as single agents augmented IFNγ release by the alloreactive T cells (Fig. 1C). Combination of cemiplimab and REGN3767 further enhanced T-cell reactivity in the presence of a T-cell receptor stimulus. To show that REGN3767 alone, or in combination with cemiplimab, does not induce nonspecific T-cell activation, we performed proliferation and cytokine release assay in PBMC ex vivo. The addition of cemiplimab and/or REGN3767 had no effect on PBMC proliferation and did not induce significant production of IFNγ, TNFα, IL2, and IL10 from freshly isolated or preactivated PBMC. Finally, we tested the ability of REGN3767 to induce antibody-dependent cellular cytotoxicity (ADCC) and complement-dependent cytotoxicity (CDC) in vitro. REGN3767 at concentrations ranging from 95 fmol/L to 100 nmol/L did not mediate ADCC or CDC activity against Jurkat/hLAG-3 target cells in the presence of human NK cells or human serum complement, respectively. Taken together, these data suggest that REGN3767 and cemiplimab treatments could synergistically enhance effector T-cell responses without nonspecific T-cell activation.

We tested the antitumor activity of REGN3767 as a single agent or in combination with Regeneron's clinical anti-PD-1 antibody cemiplimab in a mouse MC38.Ova colon carcinoma model. Because cemiplimab and REGN3767 do not cross-react with mouse PD-1 and LAG-3, respectively, we genetically engineered dual human PD-1xLAG-3 knockin mice to express the extracellular domains of human PD-1 and human LAG-3 as chimeric proteins containing the corresponding mouse transmembrane and cytoplasmic domains from the respective mouse loci (Supplementary Fig. S5A). As shown in Supplementary Fig. S5B, PD-1xLAG-3 dual knockin mice faithfully express human PD-1 and LAG-3 proteins on activated T cells. The absence of staining with anti-mouse LAG-3 and anti-mouse PD-1 antibodies confirmed that full-length mouse PD-1 and LAG-3 were deleted in PD-1xLAG-3 dual knockin mice. Of note, as human PD-1 binds both human and mouse PD-L1 with similar affinity, and human LAG-3 binds both human and mouse MHC II (15, 16, 45), humanization of PD-L1 and MHC II are not required. Homozygous human PD-1xLAG-3 dual knockin mice displayed a normal life span and no overt signs of autoimmunity for at least 1 year, whereas PD-1xLAG-3-knockout mice develop strain and tissue-specific autoimmunity (22, 25).

In MC38.Ova tumors in PD-1xLAG-3–knockin mice, PD-1 and LAG-3 coexpression was largely limited to CD4⁺ and CD8⁺ TILs, while splenic T cells predominantly expressed PD-1 but not LAG-3 (Supplementary Fig. S5C), consistent with previous reports (22). Approximately 47% of CD8⁺ and 50% of CD4⁺ TILs cells were double positive, supporting the hypothesis that these two receptors could directly cooperate in T-cell suppression.

In a prophylactic MC38.Ova tumor model, monotherapy with cemiplimab at 10 mg/kg was moderately efficacious, while REGN3767 dosed at 25 mg/kg had a minimal effect as a monotherapy (Fig. 2A and C). However, combination treatment prolonged survival (P < 0.0001) compared with isotype antibody control treatment at the end of the study on day 46. Five of 12 mice were tumor free in the combination treatment group, whereas only 1 of 12 and 2 of 12 mice were tumor free in REGN3767 and cemiplimab monotherapy groups, respectively. In a therapeutic model, treatment of 100 mm³ MC38.Ova tumors with cemiplimab at 10 mg/kg was partially efficacious, while there was no single-agent activity of REGN3767 at 25 mg/kg (Fig. 2B and D). Combination therapy showed a significant reduction of tumor growth on day 22 relative to isotype control (P < 0.05) and to REGN3767 (P < 0.01) groups.

To determine the relative contribution of cemiplimab and REGN3767 to the combination effect, we tested regimens, in which the dose of one antibody was constant and the other was titrated (Supplementary Table S1). When the anti-PD-1 antibody, cemiplimab, was administered at 10 mg/kg (five injections within 2 weeks), addition of REGN3767 at increasing doses (5 mg/kg or 25 mg/kg) on the same schedule enhanced efficacy. Similarly, when REGN3767 was dosed at 25 mg/kg, lowering cemiplimab dose from 10 to 1 mg/kg reduced the antitumor activity. Thus, the coblockade of PD-1 and LAG-3 is superior to either monotherapy and the efficacy is more dependent on anti-PD-1 activity.

Next, we determined the impact of anti-PD-1 and anti-LAG-3 treatment on T cells in MC38.Ova tumors. Upon treatment of mice with 100 mm³ MC38.Ova tumors with a combination of cemiplimab and REGN3767 the average tumor size was significantly smaller than in the isotype control or anti-LAG-3–treated groups (Fig. 3A), and combination treatment, but not monotherapy, resulted in significant increase of tumor-infiltrating CD4⁺ T cells (Fig. 3B). To examine antigen-specific T cells in tumors, TILs were restimulated ex vivo with OVA MHC I and MHC II peptides followed by intracellular staining for IFNγ and TNFα. Neither mono- nor combination therapy had a significant impact on the frequency of IFNγ- or TNFα-producing CD8⁺ TILs, however, the frequency of IFNγ- and TNFα-producing CD4⁺ TILs was significantly higher in cemiplimab and dual-antibody–treated groups as compared with the control treatment (Fig. 3C). As T-cell priming occurs in tumor dLN, we examined antigen-specific and IFNγ-expressing T cells in dLNs. There was no expansion of OVA-specific CD8⁺ T cells in dLN as a result of single or dual anti-PD-1 and anti-LAG-3 treatment. However, we observed a significant increase in IFNγ-producing CD4⁺ and CD8⁺ T cell in dLNs in the dual anti-PD-1 and anti-LAG-3 treatment group (Fig. 3D). In contrast to tumors, there were no changes in TNFα production by T cells in response to single or dual antibody therapy in dLNs. Taken together, these data suggest that cemiplimab and REGN3767 combination immunotherapy reduces tumor growth by increasing the proportion of effector T cells both in the tumor and in dLNs. In addition, we detected increased concentrations of IFNγ, TNFα, and IL10 in serum and spleen lysates of PD-1xLAG-3–knockin mice treated with both REGN3767 and cemiplimab compared with treatment with either antibody alone (Fig. 4).

To explore the underlying molecular mechanisms of MC38.Ova tumors response to anti-PD-1 and anti-LAG-3 treatment, we conducted RNA sequencing of MC38.Ova tumors in PD-1xLAG-3 mice following treatment with one or two doses of cemiplimab, REGN3767, or their combination. Mice were randomized on day 10 after MC38.Ova cells implantation when tumors reached 100 mm³. Treatment with two doses of REGN3767, cemiplimab, or their combination (on randomization day 10 and on day 14) significantly reduced tumor volumes compared with isotype control treatment, while single-dose treatment on day 10 did not show antitumor activity (Fig. 5A). Both REGN3767 and cemiplimab promoted robust transcriptional changes in tumors, and combined PD-1 and LAG-3 blockade resulted in enhanced immune activation signatures. The combination treatment resulted in significant upregulation of 1,616 and 2,256 genes after one or two doses, respectively (fold changes vs. control group >1.5; Fig. 5B). While the number of upregulated genes with REGN3767 treatment kept increasing (131 genes after one dose vs. 731 genes after two doses), the number of cemiplimab-induced genes decreased (696 upregulated genes after one dose vs. 161 genes after two doses), indicating different kinetics of response to these monotherapies. Transcriptional gene expression profiles indicative of different TILs subpopulations in the combination group showed the enrichment of immune cell signatures, including neutrophils, macrophages, NK, and T cells; all signatures were more pronounced after two doses (Fig. 5C). The combination therapy also enhanced immune responses promoted by either antibody alone, including genes associated with T cells' activation and effector function. In the T cells signature, a prominent increase in the expression of Cd40lg receptor, which is involved in costimulatory T-cell receptor signaling, was observed in a combination treatment (Fig. 5C). Blocking PD-1 and LAG-3 in combination therapy also resulted in upregulation of inhibitory PD-L1 and VISTA ligands that can suppress T-cell responses, suggesting a potential mechanism of adaptive immune escape (Fig. 5D). Cytotoxic Gzma, Gzmb, and Prf1 as well as selected T-cell immunoregulatory and Th1 signaling genes were upregulated in the combination treatment group, consistent with our finding that combination treatment enhanced the cytotoxic function of intratumoral T cells after tumor antigen restimulation (Fig. 3C).

Twenty-five (25) female cynomolgus monkeys (5 animals/dose group) received a single intravenous infusion (1, 5, or 15 mg/kg) or subcutaneous injection (1 or 15 mg/kg) of REGN3767 and animals were followed over the 8-week study period. Single dose pharmacokinetic of functional REGN3767 showed linear kinetics with a half-life of 10–12 days (Fig. 6). The immunogenicity of REGN3767 was high, with 5 of 15 and 5 of 10 animals showing positive ADA responses following intravenous and subcutaneous administrations, respectively, but there were no observable adverse effects. Following intravenous infusion of REGN3767, the pharmacokinetic parameters were estimated using noncompartmental analysis (Supplementary Table S2). The mean C_max increased in an approximately dose-proportional manner across the tested dose levels. Following intravenous infusion of REGN3767, the mean AUCinf increased in a dose-proportional manner as demonstrated by similar AUCinf/dose values. Consistent with this finding, the mean clearance values were similar for the three dose groups, ranging from 4.11 to 4.47 mL/day/kg. The elimination phase t_1/2 values for the three dose groups were also similar, ranging from 10.8 to 11.5 days.

In a cynomolgus monkey toxicology study, 5 animals/sex/group were given 0, 2, 10, or 50 mg/kg/week REGN3767 by i.v. infusion for 4 consecutive weeks followed by an 8-week recovery period. REGN3767 was well-tolerated at all dose levels. There were no unscheduled deaths during the study and no drug-related clinical signs evident. There were no REGN3767-related effects on body weights, body temperatures, clinical pathology, cardiovascular, or neurological endpoints, with no substantial sex differences noted. In addition, there were no REGN3767-related changes in lymphocyte subpopulations in PBMC, organ weights, or findings from the gross and microscopic examination of tissues from primary necropsy (day 30) or after the 8-week recovery period (day 85). The no observable adverse effect level for this study was considered to be 50 mg/kg, the highest dose level evaluated.

Blockade of inhibitory pathways on activated T cells with mAbs against the immune checkpoint receptors PD-1 and CTLA-4 has shown remarkable antitumor activity in diverse human cancers. Furthermore, therapeutic cotargeting of PD-1 and CTLA-4 showed enhanced clinical activity compared with single agents. These successes have led to the exploration of additional inhibitory signaling pathways on immune cells. LAG-3 receptor is another inhibitory checkpoint on T cells; its potential role in cancer is supported by extensive preclinical studies (8, 22, 23). Notably, analysis of mouse and human tumors suggests that coexpression of LAG-3 and PD-1 is associated with T-cell dysfunction (21, 22), indicating that inducible LAG-3 expression on activated T cells could be a potential mechanism of therapeutic resistance to PD-1 checkpoint blockade (46). These findings have prompted efforts to explore using anti-LAG-3–blocking antibodies, either alone or in combination with anti-PD-1, in cancers (32–34).

To target LAG-3, we generated a human mAb REGN3767 that binds to human and monkey LAG-3 and inhibits the interaction with its ligand, MHC class II. REGN3767 is engineered as human IgG4 isotype and contains one mutation in the hinge region to minimize half-antibody formation and three additional point mutations in Fc region to reduce the potential for inducing antibody-dependent cytotoxicity against LAG-3–expressing cells. We confirmed that REGN3767 does not elicit antibody Fc-mediated ADCC or CDC activity. In cell-based assays, REGN3767 rescued T-cell activation as a single agent and enhanced responses to a clinical anti-PD-1 antibody, cemiplimab, demonstrating that LAG-3 delivers an independent inhibitory signal to activated T cells during PD-1 blockade. In an engineered T-cell/APC bioassay, the combination of REGN3767 and cemiplimab was able to overcome the inhibitory effects of MHC II/LAG-3 and PD-L1/PD-1 signaling. In a second assay, REGN3767 was able to enhance the response to cemiplimab of primary CD4⁺T cells in allogeneic T cell:PBMC MLR assay.

Consistent with in vitro findings, REGN3767 therapy reduced tumor growth in a subset of human PD-1XLAG-3–knockin mice, and improved antitumor efficacy of cemiplimab, thus providing preclinical proof of concept for dual PD-1 and LAG-3 blockade in patients with cancer.

The potential mechanism underlying the additive effect of REGN3767 and cemiplimab was consistent with T-cell activation both in tumors and the periphery of MC38.Ova-bearing mice. Combination treatment was associated with increased intratumoral CD4⁺ and CD8⁺ T cells producing IFNγ, TNFα, as well as elevated IFNγ, TNFα, and IL10 levels in blood and spleen. Although IL10 is normally associated with reduced antitumor activity (47), some reports describe a role of IL10 in expanding tumor CD8⁺ T cells (48).

The role of REGN3767 in modulating T-cell responses was further supported by robust transcriptional changes consistent with T-cell expansion and activation in tumors in response to REGN3767 therapy. Combination therapy elicited additional immune-related gene changes, not seen with either antibody alone. Surprisingly, in addition to T-cell expansion and activation, REGN3767 therapy engaged other types of tumor-associated leukocytes, including macrophages, neutrophils, and NK cells, suggesting that the myeloid compartment may contribute to the therapeutic response following anti-LAG-3 therapy. The physiologic role of LAG-3 on immune cell types other than T cells, including plasmacytoid DCs, NK cells, and B cells is poorly understood (8, 10), and may be related to additional LAG-3 ligands reported recently. It has been suggested that LSECtin, a member of the DC-SIGN family, and Galectin-3, both expressed in many tumors, are ligands that can also regulate LAG-3–expressing CD8⁺ T cell and NK cells in tumor microenvironment (49, 50). Fibrinogen-like protein FGL1, a liver-secreted protein, is another LAG-3 functional ligand independent of MHC II (51). At this time, it is not clear how REGN3767 affects LAG-3 interaction with additional putative ligands. Detailed studies as to when and where these putative ligands are expressed and how they interact with LAG-3 will further facilitate unravelling the mechanisms of LAG-3 targeting.

Neither REGN3767 nor cemiplimab, as monotherapy or in combination caused nonspecific lymphocyte activation in an ex vivo cytokine release assay. Cynomolgus monkey toxicology studies were performed separately for REGN3767 and cemiplimab and there were no potential adverse effects related to lymphocyte-binding patterns of these antibodies.

Our results expand on the existing preclinical data that combination blockade of PD-1 and LAG-3 signaling has potential for increased therapeutic benefit in cancer treatment. REGN3767 in combination with cemiplimab is currently being investigated clinically in multiple tumor types.

E. Burova has ownership interest (including patents, stock, and stock options) in Regeneron and is co-inventor on a pending patent application relating to the subject matter of the manuscript. A. Hermann has ownership interest (including patents and stock) in Regeneron. E. Ullman has ownership interest (including patents and stock) in Regeneron. G. Halasz has ownership interest (including patents) in Regeneron Pharmaceuticals stock options. T. Potocky has ownership interest (including patents and stock) in Regeneron. O. Allbritton has ownership interest (including stock) in Regeneron. W. Poueymirou is an executive director animal production and vivarium operations (paid consulting) at Regeneron Pharmaceuticals. J. Martin has ownership interest (including patents and stock) in Regeneron. W.C. Olson is vice president (paid consulting) at Regeneron Pharmaceuticals. A. Murphy has ownership interest (including patents, stock, and stock options) in Regeneron. E. Ioffe has ownership interest (including patents) in Regeneron, and is co-inventor on a pending patent application relating to the subject matter of the manuscript. M. Mohrs has ownership interest (including patents, stock, and stock options) in Regeneron and co-inventor on a pending patent application relating to the subject matter of the manuscript. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E. Burova, A. Hermann, J. Dai, E. Ullman, T. Potocky, W.C. Olson, A. Murphy, E. Ioffe, G. Thurston, M. Mohrs

Development of methodology: E. Burova, A. Hermann, J. Dai, E. Ullman, T. Potocky, M. Liu, J. Pei, A. Rafique, A. Murphy, E. Ioffe

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Burova, A. Hermann, J. Dai, E. Ullman, T. Potocky, M. Liu, O. Allbritton, A. Woodruff, J. Pei, W. Poueymirou

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Burova, A. Hermann, J. Dai, E. Ullman, G. Halasz, T. Potocky, S. Hong, M. Liu, O. Allbritton, A. Woodruff, J. Pei, A. Rafique, A. Murphy, G. Thurston

Writing, review, and/or revision of the manuscript: E. Burova, S. Hong, M. Liu, A. Rafique, J. Martin, W.C. Olson, E. Ioffe, G. Thurston, M. Mohrs

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. Burova, A. Hermann, J. Dai, T. Potocky, J. Pei, D. MacDonald

Study supervision: E. Burova, A. Hermann, J. Dai, J. Martin, W.C. Olson, A. Murphy, E. Ioffe, M. Mohrs

The authors thank all Regeneron employees who contributed to the generation and characterization of REGN3767 including R. Leidich, A. Badithe, P. Kruger, C.-J. Siao, A. Mujica, H. Polites, G. Kroog, I. Lowy, and N. Papadopoulos. This work is funded by Regeneron Pharmaceuticals, Inc.

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.