Protein structure and computer tools

The tridimensional structure of the dimeric form of the extracellular domain of PD-L1 was retrieved from the Protein Data Bank (PDB, https://www.rcsb.org/) under the PDB code 5J89 [32]. The GOLD 5.3 software (Cambridge Crystallographic Data Centre, Cambridge, UK) was used to perform molecular docking analysis. Prior to the docking operations, the structure of each ligand was optimized using a classical Monte Carlo (MC) conformational searching procedure via the BOSS (Biochemical and Organic Simulation System) v4.9 software [33]. Molecular graphics and analysis were performed using Discovery Studio Visualizer, Biovia 2020 (Dassault Systèmes BIOVIA Discovery Studio Visualizer 2020, San Diego, Dassault Systèmes, 2020).

Compounds studied as potential PD-L1 binders

The docking analysis was performed in two steps: (i) with BQR and the reference products BMS-202 and FLB; and (ii) with BQR analogs and derivatives. Different series of BQR analogs have been designed and tested as anti-leukemic agents [34, 35]. Non-carboxylic acid derivatives of BQR have revealed an improved cellular potency compared to the parent product [36]. We selected four compounds in this series to evaluate their PD-L1 binding capacity: the fluorinated analog BQR-4, amide compound BQR-8, and the N-methoxy and N-ethoxy acetamide derivatives BQR-13 and BQR-14 (Figure 2). These four compounds have been identified as potent inhibitors of DHODH and cytotoxic agents, notably BQR-13 which inhibited DHODH more potently than the parent compound BQR and showed a higher cytotoxic potential [36].

Display full size

Figure 2. Structures of the studied BQR analogs. BQR: brequinar

In addition, three hybrid compounds were evaluated as potential PD-1 binders: B1, B2, and B3 (Figure 2). These molecules incorporate a BQR moiety coupled to a triphenylphosphine (TPP) to target mitochondria. They differ by the length of the (CH₂)_n alkyl linker between the BQR and TPP units, with n = 2, 4, or 5 for compounds B2, B1, and B3, respectively. Compound B2 has been shown to target DHODH most efficiently, triggering the formation of reactive oxygen species (ROS), promoting mitochondrial lipid peroxidation, and inducing ferroptosis in cultured cancer cells [27]. This hybrid compound showed no apparent adverse effects in mice and most importantly, it revealed an unexpected capacity to downregulate PD-L1 in cancer cells. The anticancer performance of B2 was clearly superior to that of the parent compound BRQ in a melanoma B16F10 murine model [27].

Molecular docking procedure

The docking procedure includes the following steps:

• (1)

  An MC conformational search of the ligands using the BOSS software, freely available to academic users. The best starting geometries (minimum-energy conformers) were defined for each compound. Within BOSS, MC simulations were performed in the constant-temperature and constant-pressure ensemble (NPT), as described [37].

• (2)

  Determination of the free energy of hydration for ligand structure. The molecular mechanics/generalized Born surface area (MM/GBSA) procedure was used to evaluate the free energies of hydration (ΔG), in relation to aqueous solubility [33]. MM/GBSA does not provide information about the number, position, and free energy of water molecules but the contribution of water molecules in the binding site is estimated in the calculation of Gibb’s free energy of binding (ΔG) values [38]. MC search and computation of ΔG were performed within BOSS using the xMCGB script as previously described [33, 39]. The best ligand structure was then considered for the docking procedure.

• (3)

  Definition of the protein-ligand sites of interaction. Drug-binding sites were searched using the CASTp 3.0 software for active site prediction. With the PDB structure 5J89, the flexible amino acids are Tyr56, Met115, Asp122, Tyr123, and Lys124 (monomer A), and Tyr56, Gln66, Met115, Asp122, and Tyr123 (monomer B). The side chains of these amino acids were considered fully flexible during docking. The side-chain protonation state of ionizable residues (Asp, Glu, His, Tyr, Lys) and the tautomeric forms of histidine (His) are controlled. In general, up to 100 poses considered energetically reasonable are selected during the search for the correct binding mode of the ligand. The decision to select a trial pose is based on ranked poses, using the PLP (Piecewise Linear Potential) fitness scoring function. The PLP fitness value is a pairwise additive scoring function incorporated into GOLD [40]. The same procedure was used to establish molecular models for all compounds.

• (4)

  Docking procedure with GOLD. A typical docking process considers 100 energetically reasonable poses (according to the ChemPLP scoring function) to search for the correct ligand binding mode. The decision to maintain a trial pose is based on ranked poses, using the PLP fitness scoring function. Six poses are kept usually. The empirical potential energy of the interaction ΔE for the ranked complexes was evaluated using the expression ΔE (interaction) = E (complex) − [E (protein) + E (ligand)]. Calculations of the final energy are performed on the basis of the SPASIBA force field. SPASIBA (integrated into CHARMM) has been specifically developed to provide refined empirical molecular mechanics force field parameters [41]. SPASIBA reproduces vibrational frequencies (with a higher accuracy than molecular mechanics potentials), and potential energy distributions of normal modes. It has been shown to be excellent at reproducing crystal-phase infrared data. The parameters are derived from vibrational wavenumbers obtained in the infrared and Raman spectra of a large series of compounds including organic molecules, amino acids, saccharides, nucleic acids, and lipids [42].

The same procedure was used to establish molecular models for all compounds. We have previously employed the same computational approach to identify PD-L1 binding small molecules, including both natural products and synthetic compounds [31, 43, 44].

Interaction of BQR with PD-L1 dimer

BQR was docked into the cavity at the interface of two adjacent PD-L1 monomers, delimited by four tyrosine residues (two Y123 + Y56 on each side). This cavity is known to accommodate small molecules [45]. The modeling analysis indicated that BQR can form complexes with PD-L1 in a manner comparable to that observed with BMS-202. The molecule inserts its biphenyl unit at the junction of the two PD-L1 monomers, in a relatively elongated cavity. The fluoro-biphenyl portion projects deeply into the cavity in a hydrophobic region, not accessible to the solvent whereas the methylquinoline moiety is at the surface of the protein groove, as represented in Figure 3. The elongated structure of the drug is well adapted to the size of the protein cavity.

Display full size

Figure 3. Molecular model of BQR bound to the PD-L1 dimer. Views (a) and (b) show the drug located at the interface of the protein dimer. The two protein units (in cyan and green) sandwich the small molecule ligand. The inset shows the deep insertion of the drug in the binding site. (c) A close-up view of BQR bound to the PD-L1 interface. The SAS around the drug binding zone is represented with the indicated color code. (d) Binding map contacts for BQR bound to PD-L1 (color code indicated). (e) Superimposed models of BMS-202 and BQR. BQR: brequinar; PD-L1: programmed death-ligand 1; SAS: solvent-accessible surface

BQR can form a stable complex with the PD-L1 dimer but the calculated empirical potential energy (ΔE = –59.00 kcal/mol) is significantly superior (less negative) to that calculated with the reference product BMS-202 (ΔE = –82.03 kcal/mol) under identical conditions. It is inferior to that calculated with FLB (ΔE = –47.44 kcal/mol). However, the free energy of hydration is much less favorable (ΔG = –15.02 kcal/mol for FLB vs. –5.59 kcal/mol for BQR). A number of molecular interactions, including H-bonds, π-stacking interactions, and van der Waals contacts, stabilize the drug-protein dimer complex (not shown) but the extent of binding of BQR to PD-L1 is relatively modest. Overall, the data suggest that BQR is not a robust PD-L1 binder.

Interaction of BQR derivatives with PD-L1 dimer

We compared the capacity of compounds BQR-4, BQR-8, BQR-13, and BQR-14 to form complexes with PD-L1 using the same methodology as for BQR. From the calculated ΔE/ΔG values, some structure-binding relationships can be determined (Table 1).

The replacement of the acid function of BQR with an amide (BQR-8) showed no improvement in the PD-L1 binding capacity. In contrast, the incorporation of a trifluoroethanol substituent in place of the acid group provided a derivative (BQR-4) capable of forming much more stable complexes with PD-L1 than the parent compound. The observation is interesting because this compound BQR-4 has been shown to inhibit DHODH much less efficiently than BQR (IC₅₀ = 0.48 and 808 nM for BQR and BQR-4, respectively) while maintaining a significant cytotoxic potential [36]. BQR-4 shows a high capacity to interact with PD-L1, even superior to that of BMS-202, and this interaction may participate in compound bioactivity. Both BQR-4 and BQR-14 can form stable complexes with the PD-L1 dimer (Figure 4). Their biphenyl unit inserts at the surface of the protein dimer, while the quinoline unit remains accessible to the solvent. Different molecular contacts, notably a π-sulfur interaction with residue Met115 and π-π stacking interaction with Tyr56, stabilize the drug-protein complexes [Figure 5 with notable contacts shown in pink (π-π stacking) and in orange (π-sulfur interaction)]. Other molecular contacts, in particular, van der Waals interactions and key ligand-protein H-bonds (in green) contribute to stabilizing the molecular edifices. The multiplicity of the contacts contributes the stabilizing the drug-protein complex. The trifluoroethanol substituent of BQR-4 provides a better anchorage to the protein than the N-methoxyacetamide (BQR-13) or N-ethoxyacetamide (BQR-14) unit.

Display full size

Figure 4. Molecular model of compounds BQR-4 (top) and BQR-14 (bottom) bound to the PD-L1 dimer. (a, d) The drug is located at the interface of the two protein units (in cyan and green). (b, e) A detailed view of the drug inserted in the binding site. (c, f) The hydrophobicity surface around the drug binding zone is represented with the indicated color code. BQR: brequinar; PD-L1: programmed death-ligand 1

Display full size

Figure 5. Interaction of BQR analogues with PD-L1. Binding map contacts for (a) BQR-4 and (b) BQR-14 bound to PD-L1 (color code indicated). BQR: brequinar; PD-L1: programmed death-ligand 1

The case of compound BQR-14 is interesting because this compound is a 35-fold weaker inhibitor of DHODH than compound BQR-13 (IC₅₀ = 0.12 and 4.2 nM for BQR-13 and BQR-14, respectively) but its cytotoxic potency remains very high [36]. Apparently, its cytotoxicity is not only dependent on DHODH inhibition. Our data suggest that the BQR scaffold can be exploited to design novel PD-L1 inhibitors.

Interaction of BQR hybrids with PD-L1 dimer

The PD-L1 binding capacity of hybrid compounds B1–B3 was compared to that of BQR. Interestingly, the docking analysis indicated that B2 can form stable complexes with PD-L1, with a PD-L1 binding capacity superior to that of analogs B1 and B3. The calculated empirical potential energy (ΔE = –104.10 kcal/mol) is favorable and the free energy of hydration (ΔG = –13.30 kcal/mol) is improved for B2 compared to B1 and B3 (Table 1). A molecular model of B2 bound to the PD-L1 dimer is shown in Figure 6. In this case, the quinoline moiety is inserted into the interfacial protein cavity maintaining the TPP unit essentially outside the cavity. Most of the molecular contacts are provided by the BQR portion of the hybrid (Figure 6d). Nevertheless, the TPP moiety participates in the stability of the drug-protein complex, through interaction with residue Tyr56. The elongated drug covers the entire protein groove, from Tyr56 on the PD-L1 unit-A to Tyr56 on the facing PD-L1 unit-B. There are over 30 drug-protein contacts ensuring good complex stability. The hybrid B2 is apparently a better PD-L1 binder than the derivative BQR-4. These observations may contribute to the reported capacity of these BQR-TPP hybrid compounds, notably the lead B2, to induce a downregulation of PD-L1 in cancer cells.

Display full size

Figure 6. Molecular model of compound B2 bound to the PD-L1 dimer. (a, b) The drug located at the interface of the two protein units (in cyan and green). (c) The SAS around the binding site with the indicated color code. (d) Binding map contacts for B2 bound to PD-L1 (color code as in Figure 5). PD-L1: programmed death-ligand 1; SAS: solvent accessible surface

Altogether, the molecular docking analysis indicated a modest capacity of BQR to interact with PD-L1 but the extent of binding can be reinforced significantly through modulation of the drug scaffold. The replacement of the acid function on the methylquinoline unit of BQR with a trifluoroethanol substituent afforded a very interesting molecule. This compound BQR-4 is a weak DHODH inhibitor but it maintains a good capacity to inhibit cell proliferation. PD-L1 blockade may contribute partially to the bioactivity of compounds like BQR-4 and BQR-14. Similarly, the docking analysis suggested that the PD-L1 binding capacity can be enhanced upon coupling of the BQR unit with a mitochondria-targeted TPP unit. Compound B2 displays a remarkable capacity to interact with PD-L1, covering the long groove between the two PD-L1 units. The study underlines the possibility of designing novel PD-L1 binders based on the BQR scaffold.