Small molecules from antibody pharmacophores (SMAbPs) as a hit identification workflow for immune checkpoints

Abstract

Small-molecule modulators of immune checkpoints are poised to revolutionize cancer immunotherapy. However, efficient strategies for hit identification are lacking. We introduce small molecules from antibody pharmacophores (SMAbPs), a workflow leveraging cocrystal structures of checkpoints with antibodies to create pharmacophore maps for virtual screening. Applying SMAbPs to five immune checkpoints yielded hits with submicromolar potency in both cell-free and cellular assays. Notably, SMAbPs identified the most potent T cell immunoglobulin and mucin-domain containing-3 and V-domain immunoglobulin suppressor of T cell activation (VISTA) inhibitors reported to date and first-in-class modulators of B and T lymphocyte attenuator, 4-IBB, and CD27. Targeting inhibitory and costimulatory checkpoints with hits identified through SMAbPs demonstrated remarkable in vivo antitumor activity, exemplified by MG-V-53 (VISTA inhibitor) and MG-C-30 (CD27 agonist), which significantly reduced tumor volumes in MC38 and EG7-OVA mouse models, respectively.

SMAbPs is a workflow for identifying potent small molecule modulators of immune checkpoints.

INTRODUCTION

Immune checkpoint inhibitors (ICIs) have revolutionized the landscape of cancer therapy, providing a means to achieve sustained remission across a variety of cancer types by reinvigorating the host immune system to combat tumor cells (1–4). These inhibitors function by modulating the activity of immune cells through pathways that either inhibit [such as programmed cell death protein 1 (PD-1)] or stimulate [such as inducible T cell costimulator (ICOS)] immune responses (5, 6). The clinical success of ICIs, exemplified by the remarkable outcomes in melanoma, non–small cell lung cancer, and renal cell carcinoma, has marked a remarkable advancement in oncology (1–4). However, despite these successes, a substantial proportion of patients with cancer exhibit either primary or acquired resistance to ICIs, necessitating the exploration of alternative therapeutic targets and strategies (7–10). One of the key mechanisms behind this resistance is the up-regulation of alternative immune checkpoints, such as V-domain immunoglobulin suppressor of T cell activation (VISTA) and T cell immunoglobulin and mucin-domain containing-3 (TIM-3), which can bypass the inhibitory effects of current ICIs (11). This resistance highlights the urgent need for the identification and targeting of alternative immune checkpoints in combination therapies to enhance therapeutic efficacy.

Currently, all US Food and Drug Administration–approved ICIs are monoclonal antibodies (mAbs). While mAbs have demonstrated remarkable efficacy as cancer therapies, they pose several limitations. These include suboptimal tumor penetration due to their large size, high production costs, and the potential for immunogenicity, which can lead to adverse reactions (12). Moreover, the fixed dosing regimens required for mAbs limit their flexibility in clinical use (12). In contrast, small molecules present several advantages that could potentially address these limitations. Small molecules are generally orally bioavailable, allowing for easier administration and improved patient compliance (13, 14). Small molecules can also penetrate tumors more effectively due to their smaller size. They can also be designed for flexible dosing regimens, which can be tailored to individual patient needs. These properties suggest that small molecules could improve the safety, efficacy, and accessibility of cancer immunotherapy. In this context, our laboratory has pioneered the discovery of first-in-class small-molecule ICIs for TIM-3, VISTA, and ICOS (15–19).

One of the primary challenges for the development of small-molecule ICIs lies in the nature of immune checkpoint proteins, which often feature flat and dynamic binding surfaces that are not easily targeted by small molecules (13). Traditional drug discovery approaches, including both random and virtual screening methods, have revealed limited efficacy in the discovery of potent small-molecule ICIs (20–22). Here, we introduce small molecules from antibody pharmacophores (SMAbPs), a pharmacophore-based screening approach for multiple checkpoints. SMAbPs use cocrystal structures of immune checkpoints with mAbs to generate pharmacophore maps, which guide virtual screening of the ZINC database to identify potential hits (Fig. 1). Hits are validated using biorthogonal assays and cellular screening (Fig. 1).

Fig. 1. Overview of the SMAbP workflow.

(A) Identify key interacting residues of mAb with target immune checkpoint from cocrystal structure, (B) pharmacophore-based virtual screening, and (C) hit validation. RFU, relative fluorescence intensity.

The SMAbP approach involves several key steps. First, we use PocketQuery (23), a computational tool that identifies key interacting residues within the protein-protein interfaces of mAb-bound checkpoint proteins. PocketQuery analyzes these interfaces to calculate properties such as binding affinity and druggability scores, which are essential for the rational design of small-molecule inhibitors (23). Using the pharmacophore maps generated from these analyses, we perform virtual screening of the ZINC database, which includes millions of purchasable compounds. This extensive screening process aims to identify small molecules that can mimic the binding interactions of mAbs with their target checkpoints. Following virtual screening, identified hits undergo a rigorous validation process. This includes biorthogonal assays to confirm binding affinity and specificity, as well as cellular assays to assess the functional impact of the small molecules on immune checkpoint activity. By integrating the high–binding affinity characteristics of mAbs with the vast chemical diversity available in the ZINC database, the SMAbP strategy offers a powerful and efficient method for the discovery of potent small-molecule ICIs.

Our research represents a remarkable advancement in the field of cancer immunotherapy, offering a promising alternative to mAb-based treatments. The development of small-molecule ICIs has the potential to overcome the current limitations associated with mAbs, providing a more versatile and cost-effective approach to targeting immune checkpoints. By addressing both the primary and acquired resistance mechanisms that limit the efficacy of current ICIs, our work paves the way for more effective combination therapies and improved outcomes for patients with cancer.

RESULTS

Small-molecule TIM-3 inhibitors

The optimum criteria for pharmacophore maps that result in a hit rate of ~10 to 15% using SMAbPs are (i) maps that display four points of mAb/checkpoint interaction; (ii) maps that are derived from more than two amino acid residues from the hotspot of mAb/checkpoint interaction; (iii) maps that feature at least one hydrophobic or aromatic interaction; (iv) the presence of at least two hydrogen bond donors/acceptors; and (v) a cluster of amino acids that have a druggability score of more than 0.85 [PocketQuery scoring system (23)]. In our proof-of-concept study, we performed pharmacophore-based screening using at least four pharmacophore maps for each checkpoint. Subsequently, we procured ~15 to 20 top hits for each pharmacophore map based on root mean square deviation (RMSD) values, GlideScores from Glide docking, and commercial availability.

We recently reported a first-in-class small-molecule TIM-3 inhibitor based on screening using pharmacophore maps generated by a machine learning approach, PyRod (15) The top hit from PyRod-based screening (compound A; fig. S1) displayed an estimated dissociation rate constant (K_D) value of 11.54 μM for TIM-3 binding (15). We applied the SMAbP workflow to a cocrystal structure of TIM-3 and anti-TIM-3 mAb, M6903 [Protein Data Bank (PDB) ID: 6TXZ] (24), using five pharmacophore maps (table S1) and identified a focused chemical library of 88 potential small-molecule TIM-3 inhibitors (tables S2 to S6). M6903 has been reported to block the interaction of TIM-3 with Phosphatidylserine (PtdSer), Carcinoembryonic antigen-related cell adhesion molecule 1 (CEACAM1), and Galectin-9 (Gal-9) by direct competition with the ligand binding sites on TIM-3 (24). Screening the 88 compounds for TIM-3 binding using microscale thermophoresis (MST) validated 10 hits with K_D values in the range of 0.26 to 1.71 μM (table S7). Our top hit from this screening (MG-T-19; Fig. 2A) revealed an MST K_D value of 0.26 μM [95% confidence interval (CI), 0.21 to 0.32 μM]. In comparison to our TIM-3 inhibitor (15) and recently reported small-molecule TIM-3 inhibitor (25), MG-T-19 is the most potent TIM-3 binder reported to date. As shown in Fig. 2B, the identification of MG-T-19 was an outcome of virtual screening using a pharmacophore map derived from the interaction of Tyr³⁴, Try⁹³, and Val⁹⁷ residues from M6903 with TIM-3. Notably, MG-T-19 demonstrated significant inhibition of TIM-3 interactions with PtdSer, CEACAM1, and Gal-9 at 2.5 μM (figs. S2 to S4). MG-T-19 increased tumor necrosis factor–α (TNF-α) and interferon-γ (IFN-γ) production of peripheral blood mononuclear cells (PBMCs) from healthy donors (figs. S5 and S6). Furthermore, we validated the therapeutic potential of MG-T-19 by demonstrating its capability to augment the capacity of PBMCs to impede the proliferation of Kasumi-1 cells (acute myelogenous leukemia cell line) in a coculture assay (Fig. 2C).

Fig. 2. Discovery and evaluation of MG-T-19 as a TIM-3 inhibitor.

(A) Chemical structure of MG-T-19. (B) The overlay of the pharmacophore map derived from PDB ID: 6TXZ over amino acid residues of M6903 (top) and MG-T-19 (bottom). Green spheres represent hydrophobic interactions. Yellow arrows represent hydrogen bonding interactions. (C) Proliferation of Kasumi-1 cells as a single culture and coculture with PBMCs in the absence (control) and presence of M6903 (100 nM) and MG-T-19 (2.5 μM). Proliferation was analyzed by H3 thymidine uptake, and results were given as counts per minute (CPM). Error bars represent SD (n = 5). **P < 0.01 and ***P < 0.001.

Small-molecule VISTA inhibitors

Our pioneering work in the realm of small-molecule VISTA inhibitors (16) has sparked research endeavors by others to find VISTA-targeted small molecules (26, 27). We applied the SMAbP workflow to a cocrystal structure of VISTA and a blocking anti-VISTA mAb that has been reported to block VISTA binding to T cells (PDB ID: 6MVL) (28). Using five pharmacophore maps (table S8) derived from different clusters of amino acid residues of the VISTA mAb interacting with VISTA, we identified and procured a small library of 100 potential VISTA hits (tables S9 to S13). Biophysical screening validated 12 hits as VISTA binders with K_D values in the range of 0.13 to 7.52 μM (table S14). Our top hit (MG-V-53; Fig. 3A) had an MST K_D value of 0.13 μM (95% CI, 0.07 to 0.19 μM). Orthogonal screening using surface plasmon resonance (SPR) validated the submicromolar VISTA binding affinity of MG-V-53 (Fig. 3B) with a K_D value of 106 nM (95% CI, 76.18 to 147.9 nM). To the best of our knowledge, MG-V-53 is the most potent small-molecule VISTA binder reported to date. To confirm that MG-V-53 mimics the interaction of VISTA mAb from PDB ID: 6MVL with VISTA based on the pharmacophore map used, we used our library of single-point VISTA mutants we previously reported (16) for the evaluation of small-molecule VISTA inhibitors. The identification of MG-V-53 was based on a pharmacophore map (table S8, cluster V3) that involves the interaction with Phe⁹⁴, Arg⁸⁶, and Arg¹⁵⁹ residues in VISTA. We validated that compound MG-V-53 exhibits a remarkable reduction in the VISTA binding affinity (based on MST analysis) to these single-point VISTA mutants (Phe⁹⁴Ala, Arg⁸⁶Ala, and Arg¹⁵⁹Ala) in comparison to wild-type VISTA (fig. S8). We observed the same trend for the VISTA mAb from the VISTA/mAb cocrystal structure (PDB ID: 6MVL) (fig. S8). These findings confirm that compounds from our screening workflow mimic the pharmacophores and exhibit similar binding sites to the mAbs on the surface of the immune checkpoints.

Fig. 3. Assessment of MG-V-53 as a small-molecule VISTA inhibitor.

(A) Chemical structure of MG-V-53. (B) Dose-response curve of MG-V-53 binding to VISTA using SPR. Error bars represent SD (n = 3). (C) Tumor volume in the MC38 mouse model of vehicle control, MG-V-53 (20 mg/kg, po), anti–PD-L1 mAb, and combination treatment groups of C57BL/6 mice. Error bars represent SEM (n = 8). **P < 0.01 and ***P < 0.001.

Using our previously reported screening platforms for small-molecule VISTA inhibitors (16), MG-V-53 had a median inhibitory concentration (IC₅₀) value of 121 nM (95% CI, 112.5 to 130.4 nM) in the VISTA–V-Set and Immunoglobulin domain containing 3 (VSIG3) fluorescence resonance energy transfer (FRET) assay (fig. S9) and blocked VISTA-VISTA interaction (fig. S10) between Chinese hamster ovary (CHOK1) VISTA cells and Nano-Glo interleukin-2 (IL-2)/Jurkat VISTA cells in a bioluminescent cell–based T cell activation assay. Notably, we validated the cellular target engagement of VISTA by MG-V-53 using a reported protocol for cellular thermal shift assay (29). As shown in fig. S11, MG-V-53 resulted in an average stabilization of 4.51° ± 0.93°C in the cell lysate of VISTA-expressing Raji cells. Using four ovarian cancer cell lines (OVKATE, SKOV3, COV504, and A2780) and four endometrial cancer cell lines (RL952, HEC1A, AN3CA, and Ishikawa) with reported varying VISTA gene expression (16), we demonstrated that the introduction of MG-V-53 to cocultured T cells with cancer cell lines (figs. S12 and S13) selectively reinstated T cell proliferation in the presence of high VISTA-expressing cell lines (OVKATE, SKOV3, COV504, RL952, and HEC1A), while having minimal effect on cancer cell lines exhibiting low VISTA expression (A2780, AN3CA, and Ishikawa). Moreover, both IFN-γ and TNF-α were found to be up-regulated in the presence of MG-V-53 in the supernatants of cocultures of purified human T cells with OVKATE and RL952 cell lines displaying high VISTA expression (figs. S14 and S15).

Upon demonstrating favorable pharmacokinetic properties of MG-V-53 (table S15) and the binding affinity of MG-V-53 to mouse VISTA (MST K_D = 146 nM; 95% CI, 121 to 175 nM), we assessed the ability of MG-V-53 to potentiate the antitumor activity of anti–Programmed Cell Death Ligand 1 (PD-L1) mAb using the MC38 colon cancer model in C57BL/6 mice. As shown in Fig. 3C, cotreatment with MG-V-53 [20 mg/kg, per os (po)] for 21 days enhanced the tumor inhibition rate of anti–PD-L1 mAb to 74.5% (in comparison to 45.2% for MG-V-53 and 58.6% for PD-L1 mAb groups). As shown in figs. S17 and S18, the combination of MG-V-53 and anti–PD-L1 mAb led to increased levels of both IFN-γ and granzyme B based on the analysis of the excised tumors on day 21.

First-in-class small-molecule B and T lymphocyte attenuator inhibitors

We applied our workflow to a crystal structure (PDB ID: 8F6O) (30) of B and T lymphocyte attenuator (BTLA) in complex with h22B3 (anti-BTLA mAb) aiming to identify first-in-class small-molecule BTLA inhibitors. The interaction between the BTLA on T cells with herpes virus entry mediator (HVEM) on cancer cells results in the inhibition of T cell activation (31). The anti-BTLA mAb (22B3) used in our study (PDB ID: 8F6O) has been reported to mimic HVEM binding to BTLA (30). Dual inhibition of BTLA and PD-1 has resulted in synergistic effects in various preclinical models of immuno-oncology (31–33). On the basis of the outcome of five pharmacophore maps (table S16), we purchased a potential 100 BTLA hits (tables S17 to S21) and validated 11 true hits (table S22). Our top hit (MG-B-28; Fig. 4A) binds BTLA with an MST K_D value of 531 nM (95% CI, 457 to 617 nM; Fig. 4B) and inhibits BTLA-HVEM interaction in a competitive enzyme-linked immunosorbent assay (ELISA) with an IC₅₀ value of 906 nM (95% CI, 841 to 952 nM; fig. S19). In a cell-based assay, MG-B-28 promotes T cell activation in a dose-dependent manner upon blocking BTLA-HVEM interaction upon coculturing BTLA/Jurkat_NFAT (nuclear factor of activated T cells) cells with HVEM/Chinese hamster ovary (CHO)–T cell receptor (TCR) cells (Fig. 4C). These findings render MG-B-28 a promising candidate for future screening in preclinical models as first-in-class small-molecule inhibitors of BTLA-HVEM.

Fig. 4. Biophysical and cellular evaluation of MG-B-28.

(A) Chemical structure of MG-B-28. (B) Dose-response curve of MG-B-28 binding to BTLA using MST. (C) Luminescence signal from the BTLA/NFAT luciferase reporter assay in the presence of anti-BTLA mAb (100 nM) and various concentrations of MG-B-28. Error bars represent SD (n = 5). **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Targeting stimulatory checkpoints with small molecules

Seeking to confirm the effectiveness of SMAbPs for both negative and positive immune checkpoints, we commenced screening endeavors to find first-in-class small-molecule binders of 4-IBB and CD27, using four pharmacophore maps for each checkpoint (tables S23 and S29). 4-IBB is a costimulatory checkpoint expressed on the surface of T cells that plays a key role in potentiating cytotoxic CD8⁺ T cells (34). In multiple cancer models, 4-IBB agonistic mAbs can induce remarkable tumor suppression as a result of the potentiated T cell response (35). CD27, an emerging T cell costimulatory checkpoint, has been approached with agnostic anti-CD27 mAbs showing encouraging therapeutic results when combined with anti–PD-1 mAbs in both preclinical and clinical investigations (36–38). Upon screening a library of 71 compounds (tables S24 to S27), on the basis of a cocrystal structure of urelumab (an agonist 4-IBB mAb) with 4-IBB (PDB ID: 6MHR) (39), we identified eight true 4-IBB hits with an MST K_D value of 420 nM (95% CI, 310 to 540 nM) for the top hit, MG-I-9 (table S28). On the basis of CD27/MK5890 (an activating anti-CD27 mAb) interaction (PDB ID: 8DS5) (37), we screened 80 compounds (tables S30 to S33) and identified 12 hits (table S34).

Using 4-IBB and CD27 reporter cell assays, we identified three hits (fig. S21) and five hits (fig. S22) with the ability to augment 4-IBB and CD27 signaling in vitro, respectively. The top hits (MG-I-62 and MG-C-30) had median effective concentration values of 1.28 μM (95% CI, 1.16 to 1.43 μM) and 0.84 μM (0.74 to 1.19 μM) in the cell-based reporter assays, respectively (figs. S23 and S25). Using PBMCs from healthy donors, we validated the ability of MG-C-30 (Fig. 5A) to activate natural killer cells (Fig. 5B), along with potentiating the secretion of TNF-α and IFN-γ (figs. S27 and S28). Similarly to anti-CD27 mAb, MG-C-30 delayed tumor growth in EG7-OVA tumor–bearing mice (Fig. 5C).

Fig. 5. Assessment of the therapeutic potential of small-molecule agonists of stimulatory checkpoints.

(A) Chemical structure of MG-C-30. (B) Activation of natural killer cells from PBMCs of healthy donors as revealed by the frequency of CD69⁺ cells in the presence of anti-hCD27 mAb (250 nM) and multiple concentrations of MG-C-30. Error bars represent SD (n = 5). (C) Tumor volume in the EG7-OVA mouse model of vehicle control, MG-C-30 (25 or 50 mg/kg, po), and anti-mCD27 mAbs groups of C57BL/6 mice. Error bars represent SEM (n = 8). ns denotes nonsignificant; *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001 relative to control.

DISCUSSION

In this study, we establish a workflow for the identification of hits targeting immune checkpoints. This workflow, SMAbPs, is based on using cocrystal structures of checkpoints with mAbs in building pharmacophore maps for virtual screening. The success of SMAbPs lies in its innovative use of antibody pharmacophores to guide small-molecule discovery. This approach addresses the key challenge of targeting flat and often undruggable protein surfaces by focusing on well-defined interaction hotspots. The integration of PocketQuery enhances the precision of pharmacophore maps, ensuring that selected residues are optimal for small-molecule binding. In addition, the use of the ZINC database for virtual screening streamlines the identification of commercially available compounds, accelerating the hit-to-lead process. We observed a ~10 to 15% hit rate (table S36) upon biophysical screening (MST) of the focused chemical libraries (comprising ~70 to 100 compounds for each immune checkpoint). The selection of compounds for the focused chemical libraries was based on pharmacophore maps that feature four points of mAb/checkpoint interaction, the involvement of more than two amino acid residues from the mAb in the interaction, a combination of at least one hydrophobic (or aromatic) interaction and two hydrogen bonding interactions, and having a druggability score of more than 0.85 in the PocketQuery scoring system. To select the best maps for our screening workflow, we ranked them according to their higher druggability scores from PocketQuery. In our proof-of-concept study, we performed pharmacophore-based screening using at least four pharmacophore maps for each checkpoint. Subsequently, we procured ~15 to 20 top hits for each pharmacophore map based on RMSD values, GlideScores from Glide docking, and commercial availability.

By leveraging the high specificity and affinity of mAb interactions, pharmacophore-based screening using SMAbPs overcomes the traditional challenges associated with targeting flat protein-protein interaction surfaces. In this study, we showcase the success of SMAbPs with three inhibitory immune checkpoints (TIM-3, VISTA, and BTLA) and two stimulatory checkpoints (4-IBB and CD27). One of the standout results from this study is the identification of MG-T-19, a TIM-3 inhibitor with unprecedented potency. TIM-3 is an inhibitory checkpoint that, when blocked, can reinvigorate T cell responses against tumors. The discovery of MG-T-19 underscores the potential of SMAbPs to identify high-affinity small-molecule inhibitors for immune checkpoints. Similarly, the identification of MG-V-53 as a VISTA inhibitor with in vivo antitumor activity highlights the versatility of SMAbPs. VISTA is another inhibitory checkpoint implicated in tumor immune evasion.

The discovery of modulators for BTLA, 4-1BB, and CD27 further showcases the broad applicability of SMAbPs. Each of these checkpoints plays a distinct role in immune regulation, and their modulation can have remarkable therapeutic benefits. For example, BTLA inhibition can enhance T cell responses, while 4-1BB and CD27 agonists can boost immune cell activation and proliferation. The small molecules identified in this study, such as MG-C-30, not only exhibit strong binding affinities but also show promising in vivo activity, indicating their potential for therapeutic development.

In vivo studies with MG-V-53 and MG-C-30 further validate the therapeutic potential of these small molecules. MG-V-53 showed significant tumor volume reduction in the MC38 mouse model, while MG-C-30 demonstrated similar efficacy in the EG7-OVA mouse model. Note that while mAbs typically mediate receptor cross-linking to trigger signaling, small molecules are unlikely to operate through this mechanism. We hypothesize that small molecules likely stabilize a specific conformational state of the immune checkpoint that mimics the active state typically achieved through cross-linking. This conformational change promotes signaling similarly to agonistic antibodies but without the need for cross-linking. We anticipate that future research by us and others will investigate this hypothesis further.

In conclusion, the SMAbP workflow represents a remarkable advancement in the field of cancer immunotherapy. By combining the specificity of mAb interactions with the efficiency of virtual screening, SMAbPs offer a powerful tool for the discovery of small-molecule modulators of immune checkpoints. Note that the top compound for each immune checkpoint in this study (MG-T-19, MG-V-53, MG-B-28, MG-I-9, and MG-C-30) had a GlideScore of −8.41 kcal/mol or lower in the Glide docking experiment. Therefore, using this score as a cutoff parameter in future applications of this screening workflow can potentially maximize the number of potent hit compounds. This approach has the potential to expand the repertoire of available cancer therapies, improving patient outcomes and providing avenues for research and development. These results highlight the potential of small-molecule immune checkpoint modulators to achieve remarkable antitumor responses, offering a promising alternative to mAb-based therapies. Future studies will focus on optimizing the top hit compounds for clinical use, exploring their synergistic effects in combination therapies, and expanding the SMAbP approach to additional immune targets.

MATERIALS AND METHODS

Pharmacophore-based screening and docking studies

The three-dimensional (3D) pharmacophore models used in virtual screening based on the immune checkpoint/mAb interaction were established from available cocrystal structures reported, specifically, a cocrystal structure of TIM-3 and anti-TIM-3 mAb, M6903 (PDB ID: 6TXZ) (24), a cocrystal structure of VISTA and anti-VISTA mAb (PDB ID: 6MVL) (28), a cocrystal structure of h22B3 (anti-BTLA mAb) and BTLA(PDB ID: 8F6O) (30), a cocrystal structure of urelumab with 4-IBB (PDB ID: 6MHR) (39), and a cocrystal structure of anti-CD27 mAb with CD27 (PDB ID: 8DS5) (37). The identification of hotspots of immune checkpoint/mAb interaction, defining clusters interacting residues, generation of the pharmacophore maps, and subsequent pharmacophore-based virtual screening of the ZINC library was done using the web service PocketQuery (http://pocketquery.csb.pitt.edu/), as previously reported (23). PocketQuery relies on computing several energetic, structural, and evolutionary properties for each residue of the immune checkpoint/mAb interaction (23), specifically, ΔGFC, an estimate of the change of free energy (in kilocalories per mole) for a residue upon complexation; ΔΔGR, an estimate of the change in free energy of an alanine mutation; ΔSASA, the change in solvent-accessible surface area (SASA) of a residue; ΔSASA%, the relative SASA as computed by naccess; Cons, a conservation score computed using Scorecons; and Rate, an evolutionary rate computed using Rate4Site (23).

We used the PocketQuery search interface to select the chains corresponding to the mAb in the cocrystal structure. After ranking the clusters by PocketQuery based on druggability score (23), we selected the top 4 to 5 clusters for each immune checkpoint based on the following criteria: (i) clusters that display four points of mAb/checkpoint interaction; (ii) clusters that are derived from more than two amino acid residues from the hotspot of mAb/checkpoint interaction; (iii) clusters that feature at least one hydrophobic or aromatic interaction; (iv) the presence of at least two hydrogen bond donors/acceptors; and (v) a cluster of amino acids that have a druggability score of more than 0.85 [PocketQuery scoring (23)]. As previously reported (23), the pharmacophore maps associated with the identified cluster/immune checkpoints interactions were directly exported from PocketQuery into a pharmacophore-based virtual screening workflow, ZINCPharmer (http://zincpharmer.csb.pitt.edu/). ZINCPharmer screens for compounds using a 3D pharmacophore, the spatial arrangement of the critical features of the interaction (e.g., hydrophobic interactions or hydrogen bonding). The size of the ZINCPharmer library is ~100 to 200 million conformations of 10 to 20 million compounds associated with commercial sources to enable purchasing the compounds immediately for experimental validation. We procured ~15 to 20 top hits for each pharmacophore map based on (i) RMSD values, (ii) GlideScores from Glide docking, and (iii) commercial availability. All the identified clusters, druggability scores, generated pharmacophore maps, overlay of the clusters on the pharmacophore maps, RMSD values, preliminary hits from each cluster, overlay of the hits over the pharmacophore maps, GlideScores (in kilocalories per mole) from Glide docking, and validated hits for each immune checkpoint are reported in the Supplementary Materials.

For the docking studies, refinement of crude PDB structure of receptor was performed [PDB ID: 6TXZ (24) for TIM-3, PDB ID: 6MVL (28) for VISTA, PDB ID: 8F6O (30) for BTLA, PDB ID: 6MHR (39) for 4-IBB, and PDB ID: 8DS5 (37) for CD27]. Polar hydrogens were added, Kollman charges were assigned, and atomic solvation parameters were added. The optimized receptors were then used for docking simulation of the compounds into the rigid receptors. The central coordinates of the box were calculated as the average coordinates of each heavy atom in the cluster. The size of the box was determined by the size of the cluster in the PDB complex. The 2D structures of the compounds were built and then converted into 3D using vLife MDS 3.0 software. The 3D structures were energetically minimized up to the RMS gradient of 0.01 using CHARM22 force field. Glide docking was performed using Maestro 13.4 (Schrödinger). GlideScore, an empirical scoring function that approximates ligand binding free energy, was used to evaluate the binding.

MST assay for TIM-3, VISTA, BTLA, 4-IBB, and CD27

For MST screening, we used Monolith NT.115 instrument from NanoTemper to assess the compound/immune checkpoint interaction. We procured His-tagged human immune checkpoint (TIM-3, VISTA, BTLA, 4-IBB, or CD27) from SinoBiological and labeled it with His-tag Labeling Kit RED-tris-NTA 2nd Generation from NanoTemper (catalog no. MO-L018). The immune checkpoint was dissolved in phosphate-buffered saline (PBS) buffer (pH 7.4) with 0.1% bovine serum albumin and 0.05% Tween 20. We kept the concentration of the fluorescently labeled immune checkpoint constant at (50 nM for TIM-3, 25 nM for VISTA, 20 nM for BTLA, 50 nM 4-IBB, or 25 nM CD27). A volume of 5 μl of the corresponding samples was filled in MST capillaries. Subsequently, we incubated the samples within the capillaries for 20 min at room temperature before the measurements. We detected changes in thermophoretic properties as a change in fluorescence intensity upon incubation of various concentrations of the tested compounds with fluorescently labeled immune checkpoints. For the determination of K_D values, we performed dose-dependent studies using multiple concentrations of each tested compound. MST data are represented as normalized changes in fluorescence (F_norm) upon ligand binding. F_norm is the ratio of fluorescence measured before and during thermophoresis. The assay was performed in triplicates in three independent runs.

TIM-3/Gal-9 inhibition

Inhibition of the TIM-3/Gal-9 interaction was assayed using competitive ELISA, as previously reported (24). We coated ELISA plates with recombinant human Gal-9 protein (2 μg/ml; from R&D Systems), followed by the addition of biotinylated TIM-3 (prepared by biotinylating recombinant human TIM-3–Fc chimera protein from R&D Systems) and various concentrations of the tested compound. Biotinylated TIM-3 was detected using streptavidin tag peroxidase conjugated antibody and a substrate of horseradish peroxidase (HRP) [3,3′,5,5′-tetramethylbenzidine (TMB)]. Subsequently, we measured the optical density at 450 nm using a plate reader. The assay was performed in three independent runs (n = 5 each).

TIM-3/CEACAM1 inhibition

We used a previously reported ELISA-based platform to examine the inhibition of the interaction between TIM-3 and CEACAM1 (24). In this assay, various concentrations of the tested compound were added to ELISA plates precoated with 0.5 μg of recombinant humanTIM-3–immunoglobulin G1–Fc in 1× tris-buffered saline (Thermo Fisher Scientific). After incubation at room temperature (1 hour), we added CEACAM1-polyHis (ARCO, CE1-H5220) and incubated the plates for an additional 1 hour with shaking. Anti–6×His-HRP (BioLegend) and TMB were used for signal detection, as described above. The assay was performed in three independent runs (n = 5 each).

TIM-3/PtdSer inhibition

To evaluate the ability of the tested compound to block the interaction of TIM-3 with PtdSer, we performed a blocking assay using Jurkat cells expressing PtdSer. Pretreatment of Jurkat cells with Staurosporine to induce apoptosis results in the expression of PtdSer on the cell surface (24). Jurkat cells (1 × 10⁷ cells) were treated with Staurosporine (2 μg/ml) for 18 hours to induce apoptosis. Untreated Jurkat cells were used as a negative control. We incubated apoptotic Jurkat cells with Alexa Fluor 647 (AF647)–labeled TIM-3 preincubated with various concentrations of tested compound. The binding of AF647-labeled TIM-3 to Jurkat cells was measured by flow cytometry as previously described on the basis of the assessment of the median fluorescence intensity of AF647 (24). The assay was performed in three independent runs (n = 5 each).

Assessment of cytokines production by PBMCs isolated from healthy donors

Human Pan-T Cell Isolation Kit (Miltenyi Biotec) was used to isolate human T cells from PBMCs from healthy donors cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. PBMCs were pretreated with M6903 (100 nM), MG-T-19 (2.5 μM), or vehicle control (n = 5) for 48 hours and stimulated with phorbol 12-myristate 13-acetate and ionomycin for 4 hours. Culture supernatants were harvested and the levels of TNF-α and IFN-γ were determined by ELISA (QIAGEN, Valencia, CA, USA).

Proliferation of Kasumi-1 cells

Kasumi-1 cells (from American Type Culture Collection; 50,000 cells) were grown for 6 days in RPMI 1640 medium with 10% inactivated pooled human serum at 37°C in a humidified atmosphere containing 5% CO₂. The cells were grown as a single culture and a coculture with PBMCs in the absence and presence (n = 5) of M6903 (100 nM) and MG-T-19 (2.5 μM). To assess the proliferation of Kasumi-1 cells, the cells were labeled for 16 hours with 37 MBq of H3 thymidine per well. After harvesting the cells, the radioactivity was quantified by liquid scintillation counting, and results were obtained as counts per minute (CPM) values of H3 thymidine uptake.

VISTA FRET assay

We performed the VISTA FRET assay as we previously reported (16). Human recombinant VISTA extracellular domain (ECD; 10× His tag), catalog no. 13482-H08H, was obtained from Sino Biological and labeled with mAb anti–6× His Tb cryptate from Cisbio. VISTA antibody (MAB71263-SP) was obtained from Novus Biologicals and labeled using the Mix-n-Stain CF 647 Antibody Labeling Kit (Sigma-Aldrich). The assay plates were medium-binding white 384-well plates (Greiner, #784075). The assay buffer was PBS (pH 7.0). time-resolved (TR)–FRET measurements were done on Tecan Infinite M1000 PRO (top read, z-position height: 25,500 μm; donor read-620: excitation: 340 to 20 nm, emission: 620 to 5 nm, and gain value: 255; acceptor read-665: excitation: 340 to 20 nm, emission: 665 to 5 nm, and gain value: 255; both with 50 flashes per well; 200-μs integration time, 100 Hz, and 50-μs lag time). VISTA Tb cryptate (25 μM) and CF647-labeled VISTA antibody (1.6 μM) stock solutions were prepared in PBS (pH7.0). VISTA Tb cryptate and CF647-labeled VISTA antibody (assay mixture) were immediately mixed before the assay to a final concentration of His-tagged VISTA (10 nM), mAb anti–6× His Tb cryptate gold (1 nM), and CF647-labeled VISTA antibody (10 nM). Stock solutions of the tested compound in dimethyl sulfoxide (DMSO; 50 nl) were applied to 384-well plates. The assay mixture (10 μl) was added to the plated compounds and incubated for 3 hours at room temperature. Measurements were performed as described above, and TR-FRET signals were calculated as a ratio as follows: (intensity of 665 nm)/(intensity of 615 nm) × 10,000. The IC₅₀ values were calculated by plotting compound concentrations versus TR-FRET signals. The dose-response curves were analyzed by nonlinear regression using GraphPad Prism (GraphPad Software Inc., La Jolla, CA, USA). The dose-response screening was performed in triplicates in three independent runs. Error bars represent SD.

VISTA SPR protocol

The validation of the VISTA binding affinity of MG-V-53 was performed on a Biacore T200 (GE Healthcare Biosciences, Sweden) using Series S Sensor Chip NTA (GE Healthcare Biosciences, Sweden, catalog no. 28-9949-51). Human recombinant VISTA protein (His tag), catalog no. 13482-H08H, obtained from Sino Biological was captured in PBS (pH 7.0) on sensor chip using NTA reagent kit (GE Healthcare Biosciences, Sweden, catalog no. 28-9950-43).

Serial dilutions of MG-V-53 were prepared while maintaining final DMSO concentration to 5% in all tested samples. The tested dilutions of MG-V-53 in PBS (pH 7.0) containing 5% DMSO and 0.05% Tween 20 were allowed to flow through both ligand-captured flow cells and reference flow cells at the same rate (30 μl/min) and contact time (120 s). Solvent correction was included to avoid the impact of DMSO on surface plasmon effect during binding analysis. Extra wash of the flow system using 50% DMSO in PBS buffer was then performed following each run. Equilibrium K_D value was calculated using Biacore T200 Evaluation software following the 1:1 Langmuir binding model with global fit parameters for solvent-corrected sensograms.

Coculture cellular assays for VISTA inhibitor (MG-V-53)

SKOV3 ovarian cancer, RL952 endometrial cancer, HEC1A endometrial cancer, AN3CA endometrial cancer, and Jurkat T cell lines were obtained from the American Type Culture Collection. COV504 ovarian cancer and A2780 ovarian cancer cell lines were obtained from the European Collection of Authenticated Cell Cultures. OVKATE ovarian cancer and Ishikawa endometrial cancer cell lines were obtained from W. Fantl laboratory, Stanford University. Cells were cultured in RPMI 1640 medium with 10% FBS, penicillin (100 U/ml), and streptomycin (100 μg/ml). All cells were grown at 37°C in a humidified atmosphere containing 5% CO₂. Briefly, 3 × 10⁴ Jurkat T cells were cultured alone or cocultured with 1.5 × 10⁵ of eight cancer cell lines used in the assay in 24-well plates. After incubation with MG-V-53 (5 μM) or VISTA antibody (MAB71263-SP; 1 μM) for 24 hours, Jurkat T cells were harvested from the supernatant. Presto Blue Viability Assay (Thermo Fisher Scientific, Waltham, MA) was then performed. The assay was performed in triplicates in three independent runs. Error bars represent SD.

Promega VISTA bioluminescent cell–based T cell activation assay for MG-V-53

The assay is based on coculturing Nano-Glo IL-2–NLP/Jurkat VISTA cells and CHOK1 VISTA cells. VISTA-VISTA interaction that results from coculturing stimulates IL-2 signaling, which is detected by a luminescence signal. Compounds with VISTA binding affinity are identified by their ability to impede VISTA-VISTA interaction and consequently result in reduced IL-2 signaling. The assay kit including Nano-Glo IL-2–NLP/Jurkat VISTA cells and CHOK1 VISTA cells was obtained from Promega, WI. The assay was then performed according to the provider’s protocol. Briefly, Jurkat cells (5 × 10⁴ cells) and 3 × 10⁴ cells of CHOK1 cells (CHOK1 parental or CHOK1 VISTA cells) were cocultured overnight in RPMI 1640 medium with 10% FBS. Antihuman CD3 antibody (1000 ng/ml) was then added, and the cocultures were incubated for 6 hours at 37°C in a humidified atmosphere containing 5% CO₂. Promega Nano-Glo reagent was then reconstituted and used at the end of the incubation time according to the manufacturer’s instructions. Each data point from this assay represents the average of three independent measurements. Error bars represent SD.

Assessment of cytokines production by ELISA for VISTA inhibitor (MG-V-53)

Human Pan-T Cell Isolation Kit (Miltenyi Biotec) was used to isolate human T cells from PBMCs cultured in RPMI 1640 medium supplemented with 10% FBS and 1% penicillin/streptomycin. Human IFN-γ and TNF-α ELISA kits (QIAGEN, Valencia, CA, USA) were used to analyze cytokine production in coincubation experiments with OVKATE and RL952 cells cultured in RPMI 1640 medium with 10% FBS. The assay was performed in triplicates in three independent runs. Error bars represent SD.

MC38 mouse model for MG-V-53

C57BL/6 mice (6 weeks old, female; Charles River Laboratories) were used in this study. Mice were housed and handled according to the guidelines approved by the Institutional Animal Care and Use Committee (IACUC) of our institution (protocol 28894). MC38 cells (5 × 10⁵ per mouse) were subcutaneously inoculated into the right flank of C57BL/6 mice. When the average tumor volume reached ~100 mm³, the mice were randomly assigned to four groups of eight mice each: control group, 20 mg/kg (po) of MG-V-53 treatment group, PD-L1 mAb group (200 μg per mouse; Bio X Cell, catalog no. BE0101), and combination treatment group (20 mg/kg of MG-V-53 + PD-L1 mAb). MG-V-53 was administered by oral gavage once a day. PD-L1 mAb was administered by intraperitoneal injection every 4 days. Tumor size and body weight were measured every other day. All mice were euthanized after 21 days of treatment, and the tumors were harvested for further analysis. IFN-γ and granzyme B levels were determined using ELISA. The tumor growth inhibition rate (TGI) was calculated using the following formula: TGI (%) = [1 − (V_t /V_V)] × 100, where V_t represents the tumor volumes of the treatment group and V_V represents the tumor volumes of the vehicle control.

BTLA-HVEM cell–free ELISA

We procured the BTLA:HVEM (biotinylated) inhibitor screening assay kit (catalog no. 72008) from BPS Bioscience and performed the assay using the manufacturer’s recommended protocol. The assay was performed in triplicates in a single run. Error bars represent SD.

BTLA/NFAT luciferase reporter assay

We purchased the BTLA/NFAT - Luciferase Reporter - Jurkat Recombinant Cell Line (catalog no. 79476) and HVEM/CHO-TCR activator stable cell line (catalog no. 79551) from BPS Bioscience and performed the assay using the manufacturer’s recommended protocol. The assay was performed in a single run (n = 5). Error bars represent SD.

4-IBB Promega assay

We purchased the 4-IBB bioassay from Promega (catalog no. JA2351) and performed the assay using the manufacturer’s recommended protocol. The assay was performed in a single run (n = 5). Error bars represent SD.

CD27 reporter assay

We procured the CD27/NF-kB Reporter-Jurkat Recombinant Cell Line (catalog no. 79509) from BPS Bioscience and performed the assay using the manufacturer’s recommended protocol. The assay was performed in a single run (n = 5). Error bars represent SD.

EG7-OVA tumor mouse model for MG-C-30

C57BL/6 mice (8 weeks old, female; Charles River Laboratories) were used in this study. Mice were housed and handled according to the guidelines approved by the IACUC of our institution (protocol 28975). The mice were subcutaneously inoculated with 5 × 10⁵ EG7-OVA cells. When the average tumor volume reached ~50 mm³, the mice were randomly assigned to four groups of eight mice each: control group, 25 mg/kg (po) of MG-C-30 treatment group, 50 mg/kg (po) of MG-C-30 treatment group, and anti-mCD27 mAb group (10 mg/kg; Bio X Cell, BE0348). MG-C-30 was administered by oral gavage once a day, and anti-mCD27 mAb was administered by intraperitoneal injection twice a week for 3 weeks. Tumor size and body weight were measured every other day. The tumor volume was determined according to following the formula: V = length × width (2) × 0.5 (in cubic millimeters). Mice were euthanized when the average tumor volume reached the ethical end point (∼2000 mm³). The TGI was calculated using the following formula: TGI (%) = [1 − (V_t/V_V)] × 100, where V_t represents the tumor volumes of the treatment group and V_V represents the tumor volumes of the vehicle control.

Statistical analysis

All the experiments described in the current study were performed at least three times. The data are plotted and analyzed using GraphPad Prism software (9.5.0). Plotting data were normalized with respect to proper experimental controls as mentioned in the corresponding figure legends.

Supplementary Materials

This PDF file includes:

Figs. S1 to S28

Tables S1 to S36