Structural insights reveal interplay between LAG-3 homodimerization, ligand binding, and function

Abstract

Lymphocyte activation gene-3 (LAG-3) is an inhibitory receptor expressed on activated T cells and an emerging immunotherapy target. Domain 1 (D1) of LAG-3, which has been purported to directly interact with major histocompatibility complex class II (MHCII) and fibrinogen-like protein 1 (FGL1), has been the major focus for the development of therapeutic antibodies that inhibit LAG-3 receptor-ligand interactions and restore T cell function. Here, we present a high-resolution structure of glycosylated mouse LAG-3 ectodomain, identifying that cis-homodimerization, mediated through a network of hydrophobic residues within domain 2 (D2), is critically required for LAG-3 function. Additionally, we found a previously unidentified key protein-glycan interaction in the dimer interface that affects the spatial orientation of the neighboring D1 domain. Mutation of LAG-3 D2 residues reduced dimer formation, dramatically abolished LAG-3 binding to both MHCII and FGL1 ligands, and consequentially inhibited the role of LAG-3 in suppressing T cell responses. Intriguingly, we showed that antibodies directed against D1, D2, and D3 domains are all capable of blocking LAG-3 dimer formation and MHCII and FGL-1 ligand binding, suggesting a potential allosteric model of LAG-3 function tightly regulated by dimerization. Furthermore, our work reveals unique epitopes, in addition to D1, that can be targeted for immunotherapy of cancer and other human diseases.

Keywords: LAG-3; cancer immunotherapy; dimerization; immune checkpoint; structural biology.

Fig. 1.

Structure of the glycosylated LAG-3 dimer. (A) Negative stain EM 3D reconstruction map model of LAG-3 dimer and selected reference-free 2D class averages. Four ECDs (D1 to D4) were labeled on the corresponding positions in both LAG-3 molecules of the dimer map model. (B) Surface representation of domains 1 to 3 in the dimeric LAG-3 structure. Glycans are shown in green. (C) SDS-PAGE of mouse LAG-3 ECD expressed and purified from Expi293F cells, Expi293F GnTI- cells, and Expi293F GnTI- cells, followed by EndoHf treatment. The predicted molecular weights of the LAG-3 ECD and EndoHf are 46.2 kDa and 70 kDa, respectively. The shift in molecular weight between glycosylated and deglycosylated mouse LAG-3 represents approximately 10 kDa of N-linked glycans. Glycosylated protein from Expi293F cells was used for crystal structure determination. (D–F) Cartoon structures of the partially deglycosylated mouse LAG-3 domains 1 and 2 (PDB 7TZE), the glycosylated mouse LAG-3 domains 1 and 2 (PDB 8DGG; this paper), and the partially deglycosylated human LAG-3 domains 1 and 2 (PDB 7TZG). Glycans are shown in green. The glycan-protein interaction at the Top of the D2 dimer interface brings two monomers together and increases the buried surface area from 440 Ų in the partially deglycosylated mouse LAG-3 structure to 511 Ų in the glycosylated mouse LAG-3 structure.

Fig. 2.

Glycosylation and homodimerization at the D2 dimer interface. (A) Contact residues in the D2 dimer interface are shown. R192 and the negatively charged glycan on N184 can be visualized in the Top view. (B) The side view shows an extensive network of hydrophobic interacting residues. (C) Graphical representation of flow-FRET experiments using two-chain, two-plasmid system with Expi293F cells. (D) Flow-FRET data from Expi293F cells comparing CFP-YFP fused by a linker to double transfections of mLAG-3^CFP/hCD80^YFP, hCD80^CFP/hCD80^YFP (known homodimer), mLAG-3^CFP/mLAG-3^YFP, and hLAG-3^CFP/hLAG-3^YFP. (E) Flow-FRET data showing the effect of the N184Q and W180D mutations on LAG-3 dimerization. Reduction in %FRET⁺ cells was calculated relative to WT. The data in (D) and (E) are presented as the mean ± SEM and are representative of at least two independent experiments. All statistics were determined by Student’s t test with P values noted in the figure.

Fig. 3.

Dimerization has a critical role in LAG-3 function. 3A9 cells were stably transduced with LAG-3 variants and tested for their ability to bind (A) non-cognate pMHCII tetramer and (B) FGL1 dimer. (C) LAG-3 expressing 3A9 cells co-cultured with 293T cells expressing cognate peptide-MHCII to evaluate LAG-3 function. LAG-3-expressing DO11.10 cells were also tested for their ability to bind (D) non-cognate pMHCII tetramer and (E) FGL1 dimer. (F) LAG-3-expressing DO11.10 cells were co-cultured with LK35.2 B cells pulsed with cognate peptide to evaluate LAG-3 function. Empty vector transduced T cells are represented by (−). The data are presented as the mean ± SEM and are representative of at least two independent experiments. All statistics were determined by Student’s t test with P values noted in the figure.

Fig. 4.

C9B7W binds to D2 dimerization interface, disrupts LAG-3 dimerization, and blocks ligand binding. (A) Fine epitope mapping of C9B7W by yeast surface display reveals a binding epitope at the dimer interface of LAG-3. LAG-3 residues important for C9B7W binding (W180, N209, F214) are displayed as red spheres on the LAG-3 dimer. (B) Negative stain EM showing selected 2D class averages and the 3D map model of the C9B7W Fab with mouse LAG-3. (C) Flow-FRET data showing the effect of C9B7W Fab on LAG-3 dimerization. Reduction in %FRET⁺ cells was calculated relative to the isotype. (D) LAG-3 CAR assay to evaluate LAG-3 ligand engagement in the presence of C9B7W. Jurkat cells stably expressing a NFκB-GFP reporter and a LAG-3 CAR were co-cultured with 293T cells stably expressing pMHCII or transmembrane-fused FGL1. (E) IL-2 secretion to measure LAG-3-dependent T cell inhibition in the presence or absence of C9B7W. DO11.10 cells expressing LAG-3 were co-cultured with LK35.2 cells pulsed with cognate peptide. Empty vector transduced T cells are represented by (−). The data in (C) and (D) are presented as the mean ± SEM and are representative of at least two independent experiments. All statistics were determined by Student’s t test with P values noted in the figure.

Fig. 5.

M8-4-6 antibody binds to D1 tip, disrupts LAG-3 homodimerization, and potently blocks ligand binding. (A) Fine epitope mapping of D1 antibody (M8-4-6) by yeast surface display reveals a binding epitope at the tip of LAG-3 D1. LAG-3 residues important for M8-4-6 binding (T146, R148, N151, R152) are displayed as green spheres on the LAG-3 dimer. (B) LAG-3 CAR assay to evaluate LAG-3 ligand engagement in the presence of M8-4-6. Jurkat cells stably expressing a NFκB-GFP reporter and a LAG-3 CAR were co-cultured with 293T cells stably expressing pMHCII or transmembrane-fused FGL1. (C) IL-2 secretion to measure LAG-3-dependent T cell inhibition in the presence or absence of M8-4-6. DO11.10 cells expressing LAG-3 were co-cultured with LK35.2 cells pulsed with cognate peptide. Empty vector transduced T cells are represented by (−). (D) Flow-FRET data showing the effect of M8-4-6 Fab on LAG-3 dimerization. Reduction in %FRET⁺ cells was calculated relative to the isotype. (E) Negative stain EM showing selected 2D class averages and the 3D map model of the M8-4-6 Fab with mouse LAG-3. The data in (B) and (C) are presented as the mean ± SEM and are representative of at least two independent experiments. All statistics were determined by Student’s t test with P values noted in the figure.

Fig. 6.

410C9 antibody disrupts LAG-3 dimerization and blocks ligand binding. (A) Negative stain EM showing selected 2D class averages and the 3D map model of the dimer-disrupting, 410C9 Fab with mouse LAG-3. (B) Flow-FRET data showing the effect of 410C9 Fab on LAG-3 dimerization. Reduction in %FRET⁺ cells was calculated relative to the isotype. (C) LAG-3 CAR assay to evaluate LAG-3 ligand engagement in the presence of 410C9. Jurkat cells stably expressing a NFκB-GFP reporter and a LAG-3 CAR were co-cultured with 293T cells stably expressing pMHCII or transmembrane-fused FGL1. (D) IL-2 secretion to measure LAG-3-dependent T cell inhibition in the presence or absence of 410C9. DO11.10 cells expressing LAG-3 were co-cultured with LK35.2 cells pulsed with cognate peptide. Empty vector transduced T cells are represented by (−). The data in (B) and (C) are presented as the mean ± SEM and are representative of at least two independent experiments. All statistics were determined by Student’s t test with P values noted in the figure.