A combined approach of structure‐based virtual screening and NMR to interrupt the PD‐1/PD‐L1 axis: Biphenyl‐benzimidazole containing compounds as novel PD‐L1 inhibitors

Immunotherapy has emerged as a game‐changing approach for cancer treatment. Although monoclonal antibodies (mAbs) targeting the programmed cell death protein 1/programmed cell death protein 1 ligand 1 (PD‐1/PD‐L1) axis have entered the market revolutionizing the treatment landscape of many cancer types, small molecules, although presenting several advantages including the possibility of oral administration and/or reduced costs, struggled to enter in clinical trials, suffering of water insolubility and/or inadequate potency compared with mAbs. Thus, the search for novel scaffolds for both the design of effective small molecules and possible synergistic strategies is an ongoing field of interest. In an attempt to find novel chemotypes, a virtual screening approach was employed, resulting in the identification of new chemical entities with a certain binding capability, the most versatile of which was the benzimidazole‐containing compound 10. Through rational design, a small library of its derivatives was synthesized and evaluated. The homogeneous time‐resolved fluorescence (HTRF) assay revealed that compound 17 shows the most potent inhibitory activity (IC50) in the submicromolar range and notably, differently from the major part of PD‐L1 inhibitors, exhibits satisfactory water solubility properties. These findings highlight the potential of benzimidazole‐based compounds as novel promising candidates for PD‐L1 inhibition.

2022, Grant/Award Number: 2022YWZWB2; POIR, Grant/Award Number: 04.04.00-00-420F/17-00; European Union library of its derivatives was synthesized and evaluated.The homogeneous timeresolved fluorescence (HTRF) assay revealed that compound 17 shows the most potent inhibitory activity (IC 50 ) in the submicromolar range and notably, differently from the major part of PD-L1 inhibitors, exhibits satisfactory water solubility properties.These findings highlight the potential of benzimidazole-based compounds as novel promising candidates for PD-L1 inhibition.

| INTRODUCTION
Cancer-targeted therapy is booming, and novel drugs are constantly emerging.Among the most interesting anticancer approachesemerged so far is immunotherapy.It started with ipilimumab, an anti-CTLA-4 (cytotoxic T-lymphocyte protein 4) monoclonal antibody (mAb), approved in 2011 for the treatment of melanoma, and able to reactivate the immune system against cancer cells. [1]CTLA-4 together with programmed cell death protein 1/programmed cell death protein 1 ligand 1 (PD-1/PD-L1), indoleamine 2,3-dioxygenase (IDO), T-cell immunoglobulin and mucin domain-containing protein 3 (TIM-3), and lymphocyte-activation gene 3 (LAG-3) are the immune checkpoints receptor that have garnered the most attention so far.Specifically, PD-1 binds two natural ligands: PD-L1 and PD-L2 which are both transmembrane proteins belonging to the immunoglobulin superfamily.In cancer cells, overexpression of PD-L1 leads to the progression of T cells into an exhausted state and decreased tumor cell apoptosis.[4][5][6] Currently, all clinically approved anti-PD-1/PD-L1 therapeutics are highly selective mAbs, such as nivolumab, pembrolizumab, cemiplimab, and dostarlimab targeting PD-1, or atezolizumab, avelumab, and durvalumab against PD-L1. [7,8]Once they entered the market, those mAbs have revolutionized the treatment landscape of many cancer types, especially in the metastatic setting.To date, more than 1000 clinical trials have evaluated the antitumor property of anti-PD-1/ anti-PD-L1 antibodies, and the list of approvals is constantly expanding toward more and more cancer types.However, to reduce the costs and the side effects of mAbs, while improving the patient's compliance (e.g., oral administration), the finding and developing of small molecules as PD-1/PD-L1 inhibitors has started.In fact, lowmolecular-weight ligands are less expensive and have even the advantages of higher tissue permeability, and more controllable pharmacokinetics, providing, at least in principle, a promising alternative strategy to mAbs.2][13][14] Among all, the most promising results have been shown in 2021 for ARB-272572, [15] a C2-symmetrical structure with significative activity on mice and with evixapodlin, [16] MAX-10181 [17] and INCB086550, [18] being the most advanced one (clinical phase II).
Nonetheless, they still lack the level of in vivo activity displayed by mAbs and most of them suffer from insolubility problems and/or toxicological issues, so the research of more soluble small molecules and possible synergistic strategies must go on.Interestingly, a new wave of anti-PD-L1-based therapies encompasses a combination of mAbs and anti-PD-L1 small molecules. [19]Herein, in the attempt to find novel chemotypes, with potentially improved water solubility (a known limit of most PD-L1 inhibitors), a virtual screening (VS) of the ZINC12 database (DB) and of an in-house DB [20] has been performed and the most promising virtually-identified hits (11 compounds) were screened by performing one-dimensional (1D) nucleic magnetic resonance (NMR) experiments with PD-L1 protein.As a result, five molecules displayed a certain binding capability to PD-L1.Among these, we selectively focused on EML258 (10), due to our knowledge in benzimidazole medicinal chemistry [21][22][23] and to its property as a privileged scaffold.In fact, this hetero-aromatic bicyclic ring is metabolically stable, synthetically accessible, and offers a high degree of structural diversity.[26][27][28] Thus, starting from 10 we synthesized a small library of derivatives.Among the novel compounds, the homogeneous timeresolved fluorescence (HTRF) binding assay revealed that compound 17 is the most potent one in interrupting the PD-1/PD-L1 complex (IC 50 in the submicromolar range).Notably, differently from the major part of PD-L1 inhibitors, it exhibits unexpected satisfactory solubility properties.Through two-dimensional (2D) NMR, we unambiguously identified the binding site where 17 is hosted, which turned out to be almost superimposable to that of 1.A combined approach of NMR and molecular docking calculations revealed the interaction mode within PD-L1 paving the way for further optimization in binding affinities.
Altogether, these findings highlight the potential of our new benzimidazole-based compounds as novel, promising candidates for effective PD-L1 inhibition.

| Virtual screening
][31][32] Despite small differences in accessory points, all models present a central aromatic scaffold, a biphenyl moiety, and a basic amine group (see Supporting Information S1: Figure S1).The relative distances between the aromatic moieties, in the known ligands, can be slightly different.Generally, most of the active compounds known so far possess a two-atoms linker between the biphenyl and the central scaffold (as in compounds 1, 2, 4), [9,10,12,14] which can be even part of a cycle (see compound 3 in Figure 1), [11] less frequently, a bit longer linker is present (see compound 5 in Figure 1). [13]Lately, we published a five-points pharmacophore model for extended/symmetrical biphenyl ligands. [29]As the aim of the present study was to find novel small PD-L1 binders, we used a simplified version of our model, containing three aromatic points (one for the central scaffold and two for the biphenyl moiety) and a positively charged point to mimic the basic amine function (Supporting Information S1: Figure S1).The subset "Leads Now" of the ZINC12 DB (about 2 million compounds) together with an in-house library of ~4000 small molecules [20] was screened with the aid of such a pharmacophoric model.Only those structures matching all the four points were retained for subsequent docking-based VS.Thus, almost 900 hits were docked into the dimeric structure of PD-L1 (PDB code: 5J89) [30] by means of Glide. [33]The receptor-based screening returned ~180 hits that were all visually inspected.The final choice was made based on the superposition with 1, on their chemical heterogeneity, and on the synthetic feasibility.Thus, 11 hits (6-16), seven from our in-house DB (LOR1, 11752305, CDM-55, EML432, EML258, EML746, 1332287), and four from the ZINC DB (Z1686815855, Z295636908, Z109836016, Z110038292; name or ZINC codes and chemical structure visible in Table 1) were selected for NMR 1D screening.

| NMR binding assay
The binding capabilities of the selected compounds (11) were evaluated by 1 H NMR spectroscopy.Small molecule inhibitors were F I G U R E 1 Chemical structures of the most representative programmed cell death protein 1 ligand 1 (PD-L1) inhibitors along with their IC 50 values.
T A B L E 1 Names or ZINC codes and chemical structures of the 11 compounds selected for NMR evaluation.tested against PD-1 binding single domain of PD-L1 (amino acids 18-134).We showed that only five of the compounds (7, 9, 10, 13, the spectra as in the case of BMS-like compounds. [34]This indicates that the protein does not undergo oligomerization under the influence of the tested compounds.The behavior of these inhibitors was similar to compound STD4, a weak binder that did not induce oligomerization of PD-L1. [35]

| Hit optimization
In NMR studies, 5 out of 11 hits were demonstrated to weakly bind PD-L1.Among them, we decided to focus on the benzimidazolecontaining compound 10.In fact, we envisaged that the benzimidazole, a privileged chemotype in medicinal chemistry, metabolically stable and synthetically easily accessible, on which we have synthetic expertise, [22,23] could work as the central aromatic scaffold.Thus, the binding mode of 10 and its superimposition with the crystallographic structure of 1 complexed with dimeric PD-L1, guided hit optimization (Figure 3).proper ionic interaction with the A Asp122 and thus may represent another point of modification.Herein, according to these observations, a small series of analogs of 10 have been synthesized (17-21, Table 2).In the novel derivatives, the 2-methylbiphenyl moiety, an essential fragment for interaction with PD-L1, was introduced at position 2 of the benzimidazole central core while the pyridinesulfonamide group was replaced with a small range of diverse polar side chains, linear or within a cycle, that were demonstrated to increase the potency of 1. [9] As a negative control, compound 22, featuring a 1,1'-biphenyl-4-yl substituent in place of the pivotal twisted biphenyl moiety, was prepared.

As shown in
Compound 10 was prepared as depicted in Scheme 1, according to our previously reported procedures. [21]Briefly, 4-amino-3-nitrobenzenesulfonyl chloride 23 [22] was reacted with 2-aminopyridine, using pyridine as a solvent at 0°C, to give the corresponding benzenesulfonamide 24.

| In vitro anti-PD-1/PD-L1 activity
The newly synthesized compounds were tested for their ability to inhibit PD-1/PD-L1 interaction using the well-established HTRF assay. [14,34,36,37]ta presented in Table 2 (Supporting Information S1: Figure S3) intensity changes are highlighted in blue (Figure 4).In particular, upon the addition of the ligand, the cross-peaks of the free protein in the 2D 1 H- 15 N HSQC spectrum decrease in the 55-59 and 118-124 amino acid ranges.These regions include both residues known to directly interact with small molecule inhibitors (such as Tyr56), and part of the hydrophobic residues typically involved in a cage surrounding them (Ile 54 , Ser 117 ).Interestingly, amino acids are not likely to interact with the ligand but are known to stabilize the PD-L1 homodimer by establishing interchain contacts (Asp 58 , Arg 113 ) at the interface, [29] showing reduced intensity.These NMR results suggest that 17 has a binding surface of PD-L1 almost overlapping with that of 1. [14,29] With the aim to rationalize the newly provided structure-activity relationships data (Table 2), molecular docking simulations of 17 with PD-L1 were carried out (Figure 5, left panel) and a comparison with 1 was provided (Figure 5, right panel).The protein structure cocrystallized with 1 (PDB Code: 5J89) [30] was employed for our simulations.A substantial pose convergence was predicted from our calculations, enforcing the reliability of the design strategy.In particular, 17 makes a T-shape π-π stacking interaction with A Tyr56 (biphenyl moiety) and a π-π stacking interaction with B Tyr 56 (benzimidazole ring).The critical role of Tyr56 due to its involvement Together, these data suggest the promising role of benzimidazole in the development of PD-L1 ligands.Indeed, we have shown that this aromatic system can properly orient the interacting side chains while forming itself important contacts with target proteins.As confirmed by systematic reports on the pharmacological profile of benzimidazole, fine-tuning the substituents on this scaffold can lead to a large effect on potency, activity, and selectivity.Thus 17 has the chance to be further improved in its IC 50 .
T A B L E 2 Chemical structure of compounds 17-22 and inhibitory activity against PD-1/PD-L1 interaction.Solubility has a crucial role not only in the late phase of drug development but also in the hit identification stage. [38]Insoluble compounds are likely to give assay artifacts and solubility issues could be related to false positive or false negative results.For this reason, before testing the new derivatives, we determined the solubility profile of the compounds in aqueous solutions with 0.2% DMSO by performing nephelometric measurements.We tested three concentrations (10, 50, and 100 µM) within a time frame (0, 24, and 48 h) to properly plan also the scheduled experiments in a cellular contest.
As shown in Figure 6, all compounds displayed good solubility at all the concentrations tested, with the only exception of compound 19 which is not soluble at 100 μM.
Also, the physicochemical characteristics of a drug molecule are pivotal factors that significantly impact its pharmacokinetic profile.
To check the drug-likeliness properties of our best compound 17, the in-silico absorption, distribution, metabolism, excretion (ADME) calculation was done using Qikprop utility (QikProp, Schrödinger, LLC, 2023).The pharmacokinetic profile was checked and the druggability of the molecule was assessed together with that of the reference compound 1.
The results obtained are shown in Table 3

| CONCLUSIONS
In conclusion, this study showcases a comprehensive approach combining ligand-based and receptor-based VS with NMR spectroscopy for the discovery of small molecules targeting the PD-1/ PD-L1 pathway.By employing this strategy, the benzimidazole hit 10 was identified as a promising candidate, possessing a privileged scaffold, and exhibiting a certain (low) binding affinity to PD-L1.
Subsequently, a rational lead optimization approach was undertaken, leading to the synthesis of analog 17.This derivative demonstrated enhanced potency and solubility compared to its precursor.The successful first round of optimization about benzimidazole-based small molecules represents a significant step forward in the development of novel, soluble PD-L1 inhibitors.
Based on our NMR-based binding mode of 17, a second round of synthesis may be envisaged to increase the potency of our derivatives which would allow a full exploration of in-cell efficacy and potential synergistic strategies with monoclonal antibodies, or other drugs might be investigated.Ultimately, the double VS coupled with NMR furnish, besides 10, novel scaffolds, available to the scientific community, on which to work.and Lipinski filters to select compounds provided with better druglike properties.The obtained ZINC molecules together with those coming from our in-house DB (other 4000 molecules), [20] were prepared using LigPrep [39] with Epik. [40]Conformational sampling was performed on all DB molecules using the ConfGen search algorithm. [41]ing Phase, [42] the DB was indexed with the automatic creation of pharmacophoric sites for each conformer to allow rapid DB alignment and screening.

| Pharmacophore-based VS
Based on our expertise in pharmacophore building, [43,44] a simplified version of our docking-based pharmacophore model [29] was obtained with Phase (see Supporting Information S1: Figure S1).The distance

| Receptor-based VS
As for the protein, among the human PD-L1 X-ray structures available at the time of the experiments, we chose the PD-L1 dimer structure in complex with the molecule BMS-202, with the PDB code 5J89. [30]The PDB structure was prepared by employing the graphical interface of the Schrödinger's molecular modeling platform, Maestro [45] v. 12.7.156.In particular, the protein was prepared with the Protein Preparation Wizard, [46,47] included in Maestro.Hydrogen atoms were added and missing side chains were filled in by using the Prime [48] module, while crystallographic water molecules were deleted.The N-terminal and C-terminal residues were capped with the acetyl (ACE) and N-methyl amide (NME) groups, respectively.To properly describe the protonation state of the protein residues and to also describe the hydrogen bonding networks correctly at neutral pH, the protonation states were assigned evaluating their pK a with the Propka [49] program included in Maestro.An inspection of the histidines microenvironment was F I G U R E 6 Solubility determination (nephelometry).The solubility of the compounds was determined using the instrument Nepheloskan Ascent ® (Labsystems).The experiments were conducted at room temperature in a 96-well plate with a final volume of 300 μL.Each compound was tested in triplicate at the concentrations of 10, 50, and 100 µM in phosphate-buffered saline (PBS) with 0.2% DMSO.The measurements were performed at three different times (0, 24, 48 h) from the preparation of the samples.Data obtained were compared with control (PBS with 0.2% DMSO) and the ratio sample/control was determined for each compound.The compounds are considered soluble if the ratio is ≤3.
T A B L E 3 Calculated pharmacokinetic properties of compound 17 and the reference compound 1. performed, given the labile equilibrium at neutral pH of this residue, and none of it was considered to be protonated.Finally, a relaxation procedure was performed by running a restrained minimization only on the initially added hydrogen atoms according to the OPLS2005 [50] force field.A receptor grid of 30 Å × 30 Å × 30 Å was computed for the PD-L1 structure around the centroids of the cocrystallized ligand BMS-202.During the grid calculation, the cocrystallized ligand BMS-202 was deleted from the threedimensional (3D) structure.Docking simulations were performed using the Glide SP [51][52][53] software included in Maestro using default parameters.Sampling of nitrogen atom inversions (when not belonging to cycles) and of different ring conformations were allowed, while nonplanar amide conformations were penalized.In the simulations, the receptor was kept fixed, while the ligand is treated as flexible.Docking converged to a well-defined binding mode; indeed, an almost complete overlap of the ligand was possible for the best 20 predicted poses.

| Docking calculations
As for compound 17: the molecule was built and prepared through the LigPrep [39] module of Maestro, employing the OPLS2005 force field.Epik [40,54] was used to evaluate the ligands' pKa at neutral pH and so to properly describe its protonation state.Then, the obtained ligand was optimized at the molecular mechanics level through the MacroModel [55] program included in the Schrödinger suite of programs.For the docking procedure, the grid described above was employed and the simulations performed with Glide SP [49][50][51] as well.All the presented figures were obtained using ChimeraX [56] and assembled by the means of Gimp (2.10.22 revision 3).

| Protein expression and purification
The proteins were expressed and purified as described previously by us [29] using the drop-wise dilution method of refolding.Briefly, protein was expressed in Escherichia coli BL21 (DE3) strain as inclusion bodies which were collected by centrifugation, washed, and dissolved using the 6 M guanidine hydrochloride buffer.The

| NMR Spectroscopy
For NMR measurements, the buffer was exchanged by gel filtration to PBS pH 7.4 (PD-L1).Samples of 0.23 mM concentration of PD-L1 were measured in 3 mm NMR tubes and 10% (v/v) of D2O were added to the samples to provide the lock signal.NMR experiments were performed at 300 K on a Bruker 600 MHz Avance III spectrometer equipped with a cryogenically cooled protonoptimized TCI probe.The interaction of the compounds with PD-L1 was evaluated by monitoring the perturbations in chemical shifts of NMR resonances in the 1 H-15 N 2D SOFAST HMQC [57] upon titration with the compound.NMR spectra were processed with Topspin 3.2 (Bruker) (see the Supporting Information for the resulting spectra).TFA) were used with a flow rate of 20 mL/min.A general gradient of 0-3 min at 5% B, 3-14 min increasing from 5% to 35% B, and 14-18 min increasing from 35% to 90% B was used, followed by a 90% B flush for another 2 min.Small variations in this purification method were made as needed to achieve ideal separation for each compound. 1H spectra were recorded at 400 MHz on a Bruker Ascend 400 spectrometer while 13   provided the title compound (25 mg, 35%) as a white solid. 1

and 14 )
bind to PD-L1 at the elevated concentration (molar ratio above 1:5 protein:inhibitor), as the NMR signals splitting were observed (Figure2; Supporting Information S1: FigureS2).To be sure that in the NMR spectra weak interactions of the inhibitor with the PD-L1 protein were observed, a blank test was also performed using only dimethylsulfoxide (DMSO, in the equivalent volume as the sample used for NMR titration).The results clearly confirmed that the observed changes were due to interactions between the chemical probe and the protein.As the concentration of inhibitor in the sample increases, signals splitting are observed at −0.05 and −0.35 ppm, indicating protein-inhibitor interactions.Despite the aromatic groups in the compound, no broadening of the NMR signals is observed in

Figure 3 , 2
10 adapts itself at the hydrophobic surface with the benzene ring of the benzimidazole moiety well superimposed with the central pyridine ring of 1. Notably, the carboxyl moiety of 10 overlaps with the terminal phenyl ring of 1, thus residing in the hydrophobic region formed by A Met115, A Ala121, B Ala121, and B Tyr123, and undoubtfully represents the first point of modification in the process of hit optimization.On the other side of the molecule, the basic pyridine ring is too far away to form a T A B L E 1 (Continued) Aliphatic part of 1 H NMR spectra for the titration of hPD-L1 with 10.Blue: the reference programmed cell death protein 1 ligand 1 (PD-L1); red: PD-L1/10 in the molar ratio 1:1, green: 1:5, purple: 1:10, and orange: only with DMSO, respectively.
supported our design strategy.Indeed, the novel derivatives 17-21, exhibited low micromolar IC 50 .Compound 17, featuring an ammino ethanolic chain, turned out to be the best-performing compound of the series having an IC 50 submicromolar.As expected, 22, is completely inactive and was indeed synthetized as a negative control.These results indicate that benzimidazole-based compounds could be recognized as valuable novel hits for PD-1/PD-L1 inhibition.2.5 | NMR and computational studies of the 17/ PD-L1 complex2D 1 H-15 N HSQC NMR experiments were performed to obtain information about the binding region of the complex PD-L1/17.To this aim, the variation in signal intensity for the protein (PD-L1, amino acids 18-134) in 2D 1 H− 15 N HSQC spectra was monitored in the presence of 50 μM of 17 (protein:ligand ratio equal to 1:2 considering PD-L1 as dimer).A decrease in signal intensities of the protein is diagnostic of a ligand-induced perturbation.The residues exhibiting the largest The putative binding mode of 10 (golden sticks) into the programmed cell death protein 1 ligand 1 (PD-L1) structure overlapped with the crystal structure of 1 (magenta sticks, PDB Code: 5J89).The protein is shown as a light green (Chain A) and light blue (Chain B) cartoon with interacting residues as sticks.Polar hydrogens only are shown for clarity purposes.in specific interactions with 17 through both chains and enlightened by our docking simulations, is in excellent agreement with NMR results showing that its 2D 1 H-15 N HSQC intensity is importantly reduced.Moreover, our computational study unveils that 17 is surrounded by the hydrophobic residues A,B Ile54, A,B Met115, A,B Ala121 and by A,B Ser117 belonging to one of the two mostly perturbed regions found in the NMR experiment.Regarding the polar side chain, it is differently oriented with respect to that of 1 being involved in interactions with A Asp122 (even supported by our NMR data) through the amine group, and with A Lys124 through the hydroxyl group.Our new combined NMR and computational results suggest that other chemical modifications to the polar side chain of 17 are needed to enforce the interactions with the PD-L1 binding surface.
matching tolerance was set to 2.0 Å as a balance between stringent and loose-fitting matching alignment.Screening molecules were required to match four out of four sites.Only those compounds possessing all four F I G U R E 4 Graphical representation of the per-residue intensity changes for programmed cell death protein 1 ligand 1 (PD-L1) in the presence of 17 (50 μM).The residues exhibiting the highest decreases in signal intensities have been colored in blue.F I G U R E 5 Schematic representation of the lowest energy pose of the 17/programmed cell death protein 1 ligand 1 (PD-L1) complex from molecular docking simulations alone (left panel) and overlapped with the crystal structure of 1/hPDL1 (right panel, PDB Code: 5J89).17and 1 are represented as orange and magenta sticks, respectively.Residues included in the binding site are highlighted in green (Chain A) and azure (Chain B) sticks, residues whose two-dimensional (2D) 1 H-15 N HSQC signal is reduced in intensity and belonging to the binding site are highlighted in blue sticks, while the rest of the protein is shown as light green (Chain A) and light blue (Chain B) cartoon.Polar hydrogens only are shown for clarity purposes.pharmacophoric sites were considered for subsequent receptorbased VS.
solubilized inclusion bodies being added in four portions to the refolding buffer: 0.1 M Tris pH 8.0, 1 M L-Arg hydrochloride, 2 mM ethylenediaminetetraacetic acid (EDTA, 0.25 mM oxidized glutathione, and 0.25 mM reduced glutathione.Refolded PD-L1 was dialyzed three times over 48-72 h against buffer containing 10 mM Tris pH 8.0 and 20 mM NaCl.Finally, the protein was concentrated and loaded to a size exclusion chromatography column HiLoad 26/600 Superdex 75 (GE Healthcare) preequilibrated with PBS pH 7.4 buffer and concentrated to obtain samples for NMR measurements.