The latest X-ray structure of human MAGL in complex with the piperazinyl-pyrrolidine 3l reversible inhibitor (Figure 1, PDB code: 5ZUN) [16] was used as a reference in the present study. The X-ray complex shows that the co-crystallized inhibitor forms key H-bond and hydrophobic interactions within MAGL binding site (Figure 1). The piperazine-carbonyl-thiazole portion of the inhibitor nicely occupies the bottom of the receptor pocket, while the biphenyl-pyrrolidin-2-one moiety resides at the entrance of the hydrophobic tunnel. The pyrrolidine carbonyl of the co-crystallized inhibitor establishes H-bond interactions with the backbone nitrogen of A51 and M123, within the oxyanion hole of MAGL catalytic site, that are considered as fundamental for MAGL inhibitory activity. Furthermore, the carbonyl group linked to the thiazole ring of the ligand forms an H-bond with R57; notably, the same carbonyl group also shows an interaction with a structural water molecule that forms an H-bond network with E53 and H272. These water-bridged H-bonds between the co-crystallized inhibitor and the two residues are supposed to contribute to the strong binding affinity for the target. A significant number of hydrophobic interactions were also noted between the co-crystallized ligand and the enzyme binding site. The phenyl ring of the inhibitor is encircled by the sidechains of A51, I179, L213 and L241, establishing strong van der Waals contacts with these residues. The terminal o-chlorophenyl ring of the inhibitor forms hydrophobic contacts with I179 and L205, while the electron-rich chlorine was found to be in proximity to I179, L205 and L213, thus forming additional van der Waals interactions. Finally, the thiazole ring of the inhibitor forms a face-to-face π-π stacking with Y194 that contributes to the ligand inhibitory activity.

After analyzing the main protein-ligand interactions in the enzyme binding site, as a first step of our study, the X-ray structure of MAGL in complex with its piperazinyl-pyrrolidine inhibitor was subjected to a preliminary MD simulation of 20 ns in order to evaluate the stability of the above-mentioned H-bond interactions, with particular attention to the H-bond network mediated by the structural water molecule. In fact, the H-bond interaction between the thiazole carbonyl and the structural water was believed to be of remarkable importance, due to the presence of this water molecule in various MAGL X-ray structures [19]. Therefore, we envisioned that targeting the structural water mediating the H-bond network with E53 and H272 could provide a novel strategy for the development of potent and reversible MAGL inhibitors. The MD analysis highlighted a remarkable stability of the ligand disposition throughout the simulation, possessing an average RMSD of 0.4 Å with respect to the starting coordinates and thus well maintaining the hydrophobic and van der Waals interactions identified in the X-ray complex. Similarly, the structural water was found to maintain its disposition, with an average RMSD of 0.5 Å during the MD (Figure S1). The H-bond interactions of the ligand with A51 and M123 were maintained for more than 85% of the MD, whereas the interactions with R57 and the structural water molecule were observed for about 70% of the MD.

The MD simulation thus confirmed the stability of all the interactions observed in the reference ligand-protein X-ray structure, which were subsequently exploited to develop a pharmacophore model. A receptor-based pharmacophore model was constructed using Ligandscout program [20], considering both ligand-protein interactions and the excluded volume of the binding pocket. A total of eight pharmacophore features were considered, including five mandatory and three optional features (Figure 2). The four H-bond acceptor features corresponding to the interactions of both ligand’s carbonyl groups with A51, M123, R57 and the structural water (Figure 1) were set as mandatory. A hydrophobic feature representing the phenyl ring of the ligand involved in van der Waals contacts with A51, I179, L213 and L241 residues, was considered as the fifth mandatory feature. The three optional features of the pharmacophore included three additional hydrophobic features representing: (a) the chlorine of the co-crystallized inhibitor forming van der Waals contacts with I179, L205 and L213, (b) the terminal phenyl ring showing hydrophobic contacts with I179 and L205, (c) the ligand thiazole ring exhibiting face-to-face π-π stacking with Y194 (Figure 2).

The pharmacophore model was screened against a combination of four commercial databases (i.e., Enamine, ChemBridge, Pharmeks and Vitas-M) that comprise around 4 million compounds. The pharmacophore screening step allowed to select a large set of diverse compounds showing the desirable protein-ligand interactions in the MAGL binding site, which were ranked according to the number of satisfied features. In total, 276,150 compounds matched at least the 5 mandatory features of the pharmacophore, while 5707 molecules showed to satisfy all 8 features of the model (Table 1). Given the large number of pharmacophore hits identified, only those matching at least seven features were selected for the subsequent stage of the vs. workflow.

The next step was the validation of the docking procedures, which involved the assessment of various programs to identify the methods able to reproduce the binding orientation of the X-ray ligand. The co-crystallized inhibitor 3l was thus subjected to self-docking studies using 13 docking methods. The root mean-square deviation (RMSD) between the position of the co-crystallized ligand predicted by docking and its experimental disposition was calculated. The comprehensive results of preliminary self-docking studies on 5ZUN X-ray structure are shown in Table 2: 9 out of 13 docking methods showed good results, with RMSD values below 2.0 Å.

The docking solutions were also visually checked to confirm that the fundamental ligand-protein interactions with A51, R57, M123 and the structural water molecule within MAGL binding pocket were maintained. After this analysis, only 6 docking methods (i.e., Dock6, Glamdock, Glide SP and Gold with ASP, PLP and Goldscore fitness functions) were selected for the next steps, since only the binding poses produced with these procedures showed all fundamental ligand-protein interactions. The 19,314 compounds retrieved from the pharmacophore screening were eventually docked in the X-ray structure of MAGL using the six selected docking methods. The top ranked docking solutions produced for each ligand by each docking method were subjected to a pose filtering analysis in search for compounds maintaining the interactions represented by the pharmacophore model. In particular, only compounds forming the H-bonds with A51, R57, M123 and the structural water molecule and respecting at least two out of the four hydrophobic features, where retained. By intersecting the six different subsets of compounds obtained by filtering the results of the six different docking methods, we identified 113 compounds showing the desired interactions in the binding poses predicted by all methods. Finally, these ligands were subjected to a pose consensus analysis in which the six docking results obtained for each compound were clustered together in search for common binding modes [21,22]. Only the 40 compounds for which a full pose consensus among all six docking procedures was observed (i.e., compounds for which all predicted binding modes had reciprocal RMSDs below 2.0 Å) were selected and considered as potential MAGL inhibitors. The selected compounds were then subjected to 20 ns of MD simulations with the same parameters and conditions as used in the preliminary MD study of the reference X-ray complex. The Gold Goldscore pose of each ligand was used as the starting ligand disposition within MAGL binding site, since Goldscore procedure showed the best RMSD value in the preliminary self-docking analysis. The trajectories of the 40 protein-ligand complexes were analyzed in terms of H-bond stability and ligand RMSD with respect to the starting coordinates. The compounds that showed an average RMSD value lower than 2.0 Å and the maintenance of the H-bonds with R57, the structural water molecule and at least A51 or M123 for more than 70% of the simulation were retained. Out of the 40 analyzed compounds, seven ligands were ultimately selected by the MD filter and thus subjected to enzyme inhibition assays in order to test their inhibitory activity against MAGL (see Materials and methods for details). As shown in Table 3, four compounds out of the seven tested showed a MAGL inhibition activity; considering the high similarity between VS2 and VS4, the vs. study results in a hit rate of 50% with the two best MAGL inhibitors (i.e., VS1 and VS2) with an IC₅₀ of 34.7 µM and 48.9 µM. A similarity search for compounds VS1 and VS2 against the already published hydrolase inhibitors revealed that there are no hydrolase inhibitors with a similarity score greater than 80 (score = 100 means that the two compounds are identical).

#	Structure	IC50 (µM)
VS1		34.7 ± 2.4
VS2		48.9 ± 4.2
VS3		79.4 ± 6.3
VS4		138 ± 11
VS5		>200
VS6		>200
VS7		>200
17b		0.90 ± 0.08
3l		0.042 ± 0.001

Figure 3 shows the minimized average structure of compound VS1 in complex with MAGL obtained by the last 10 ns of MD simulation. The sulphonamide moiety of VS1 displays multiple H-bond interactions in the oxyanion hole of MAGL binding site. The sulfonamide oxygen of the ligand forms an H-bond with the backbone nitrogen of A51 observed for 82% of the simulation. Although the interaction with the backbone nitrogen of M123 was lost during the MD, the sulphonamide nitrogen of the ligand establishes an additional stable H-bond with the backbone oxygen of A51 maintained for 97% of the MD, which effectively anchors the ligand core to the protein binding site. The carbonyl oxygen of the ligand pyrimidinone moiety forms an H-bond interaction with R57 maintained for 98% of the simulation and an H-bond network with the structural water molecule, E53 and H272 observed for 94% of the MD. Moreover, the pyrimidinone nitrogen forms an additional direct H-bond interaction with E53 maintained throughout the whole simulation. Finally, the o-chloro-p-fluorophenyl ring of VS1 shows van der Waals contacts with L205, L213, L241 and A51, while a face-to-face π-π stacking is observed between the pyrimidinone ring of VS1 and Y194. These hydrophobic interactions, which well mimic those identified in the reference X-ray structure, most likely contribute to the binding affinity of the ligand for MAGL.

Besides VS1, the hit compound VS2 showed a promising MAGL inhibitory activity with an IC₅₀ below 50 µM and also VS3 and VS4 demonstrated MAGL inhibition with IC₅₀ values of 79.4 µM and 138 µM, respectively (Table 3). As shown in Figure 4, the carbonyl group of VS2 core ring establishes both H-bonds with A51 and M123, maintained for 95% and 74%, respectively, during the simulation. The furan carbonyl of VS2 is involved in H-bond interactions with R57 and the structural water molecule observed for more than 80% of the simulation, whereas the water-bridged interactions with E53 and H272 were maintained for about 70% of the MD. The m-chlorophenyl ring of the inhibitor forms van der Waals contacts with L213, L241 residues while the furan ring shows a π-π stacking with Y194 (Figure 4). The possible explanation behind the lower IC₅₀ value of VS1 as compared to VS2, could be ascribed to the presence of a direct H-bond between VS1 and the carboxyl group of E53, which is instead absent in the binding mode predicted for VS2. Compounds VS3 and VS4 show a predicted pattern of ligand-protein interactions like that described for VS2. However, both VS3 and VS4 show only a single H-bond with the residues of the oxyanion hole (Figure S2), which can probably explain their lower inhibitory activity with respect to VS2.

In order to verify whether the most promising compounds could interact with cysteine residues of MAGL enzyme, the activity of the two most potent inhibitors VS1 and VS2 was also tested in presence of the thiol-containing agent 1,4-dithio-D,L-threitol (DTT). As shown in Figure 5, the IC₅₀ value of the two compounds was only very slightly, but not significantly, influenced by the presence of DTT, shifting for VS1 from 35.4 µM when assayed in the absence of DTT to 34.1 µM when assayed in the presence of 10 µM DTT and for VS2 from 47.5 µM to 49.5 µM, thus excluding any significant interaction of the two ligands with MAGL cysteine residues. Furthermore, with the aim of establishing whether the mechanism of inhibition was reversible or irreversible, the effects of preincubation on the inhibitory activity of compounds VS1 and VS2 were evaluated. In this assay, the compounds were preincubated with the enzyme for 0, 30 and 60 min before adding the substrate to start the enzymatic reaction. An irreversible inhibitor should show a higher potency after longer incubation times, whereas a reversible inhibitor should display a constant inhibition potency that is independent from the incubation time. As shown in Figure 5, this assay confirmed the reversible character of both compounds, as they did not show any significant increase in inhibitory potency at longer incubation times. Taken together, these experimental studies, aimed at better characterizing the mode of action of compounds VS1 and VS2, further supported the reliability of the binding mode predicted for the two ligands.