Cyclohexylcarbamic acid 3'- or 4'-substituted biphenyl-3-yl esters as fatty acid amide hydrolase inhibitors: synthesis, quantitative structure-activity relationships, and molecular modeling studies.

Fatty acid amide hydrolase (FAAH) is a promising target for modulating endocannabinoid and fatty acid ethanolamide signaling, which may have important therapeutic potential. We recently described a new class of O-arylcarbamate inhibitors of FAAH, including the cyclohexylcarbamic acid biphenyl-3-yl ester URB524 (half-maximal inhibitory concentration, IC(50) = 63 nM), which have significant anxiolytic-like properties in rats. In the present study, by introducing a selected group of substituents at the meta and para positions of the distal phenyl ring of URB524, we have characterized structure-activity profiles for this series of compounds and shown that introduction of small polar groups in the meta position greatly improves inhibitory potency. Most potent in the series was the m-carbamoyl derivative URB597 (4i, IC(50) = 4.6 nM). Furthermore, quantitative structure-activity relationship (QSAR) analysis of an extended set of meta-substituted derivatives revealed a negative correlation between potency and lipophilicity and suggested that small-sized substituents may undertake polar interactions with the binding pocket of the enzyme. Docking studies and molecular dynamics simulations, using the crystal structure of FAAH, indicated that the O-biphenyl scaffold of the carbamate inhibitors can be accommodated within a lipophilic region of the substrate-binding site, where their folded shape mimics the initial 10-12 carbon atoms of the arachidonyl moiety of anandamide (a natural FAAH substrate) and methyl arachidonyl fluorophosphonate (a nonselective FAAH inhibitor). Moreover, substituents at the meta position of the distal phenyl ring can form hydrogen bonds with atoms located on the polar section of a narrow channel pointing toward the membrane-associated side of the enzyme. The structure-activity characterization reported here should help optimize the pharmacodynamic and pharmacokinetic properties of this class of compounds.


Introduction
Fatty acid amide hydrolases (FAAHs) 1,2 are intracellular enzymes responsible for the hydrolysis of endogenous fatty acid ethanolamides (FAEs), 3,4 a reaction that, along with transport into cells, [5][6][7] terminates the biological effects of these lipid mediators. At least two distinct classes of cellular receptors are thought to mediate such effects. The polyunsaturated FAE anandamide 8 (arachidonoylethanolamide, 1, Figure 1) activates cannabinoid receptors, G protein-coupled receptors found in brain and immune cells that may play essential roles in the intrinsic regulation of pain, anxiety, and memory. 1 On the other hand, the monounsaturated FAE oleoylethanolamide 9,10 (OEA, 18:1∆ 9 ) activates the peroxisome proliferator-activated receptor-R (PPAR-R), a nuclear receptor involved in the control of satiety, energy metabolism, and inflammation. The saturated FAE palmitoylethanolamide 11,12 (PEA, 16:0) also is biologically active, but its mechanism of action is still undefined.
Owing to the emerging physiological functions of the FAEs, small-molecule inhibitors that selectively target intracellular FAAH activity may not only be useful as research tools but also serve as prototypes for the development of novel therapeutic agents. [13][14][15][16][17][18] Therefore, starting from the assumption that carbamic acid esters may act as site-directed inhibitors of FAAH, we have developed a series of O-aryl-N-alkylcarbamic acid esters, which are highly potent at inhibiting FAAH activity both in vitro and in vivo, but do not significantly interact with several other serine hydrolases (e.g. acetylcholinesterase and monoglyceride lipase) or with cannabinoid receptors. In rats, these compounds display profound anxiolytic-like properties, which may be attributed to their ability to elevate brain anandamide levels. 19 An initial structure-activity relationship (SAR) investigation into the steric requirements for the lipophilic O-aryl moiety of these compounds suggested that nonlinearly shaped structures may be associated with greater inhibitory potencies. 20 This idea is supported by two findings. First, the curved shape of the most successful FAAH inhibitors resembles the folded conformation assumed by fatty acids in complex with binding proteins, as well as one of the conformations predicted for anandamide binding to the CB 1 cannabinoid receptor. 21,22 Second, and more importantly, the crystal structure of FAAH inhibited by methyl arachidonyl fluorophosphonate (MAPF) revealed that the arachidonyl chain of this covalent inhibitor assumes a folded conformation in the substrate-binding site of the enzyme. 23 Previous three-dimensional quantitative SAR (3D-QSAR) analyses of the alkylcarbamic acid aryl esters 20 indicated that space occupancy of a region corresponding to the meta position of an O-phenyl ring is positively correlated with FAAH inhibition, which is in turn suggestive of a favorable interaction of the inhibitor with the enzyme binding site. The most potent compound in this series was the biphenyl-3-yl derivative URB524 (2, Figure 1), which inhibited FAAH activity in rat brain membranes with a half-maximal concentration (IC 50 ) value of 63 nM. These experiments did not attempt, however, to identify the lipophilic and electronic, but only the steric, requirements of the binding site. Therefore, in the present study we have taken URB524 as the starting point for a systematic exploration of the effect of phenyl substitution on FAAH inhibition, introducing at the meta and para positions of the distal phenyl ring of URB524 a set of substituents with balanced variation in their lipophilic and electronic properties. The distal ring has been selected for two reasons. First, our previous 3D-QSAR analyses suggested that this moiety is critical to achieve significant inhibitory potency. Second, substitutions on the distal ring are in principle devoid of direct resonance effects upon the carbamate group, and are expected therefore to yield compounds the potencies of which should correlate directly with noncovalent binding interactions.
Our experimental design comprised two steps. We first tested the sensitivity of the meta and para positions of the distal phenyl ring to the lipophilic and electronic properties of a small set of moderate-size substituents. Next, after the more responsive position of the phenyl ring was identified, we expanded the series of substituents while maintaining significant and independent variation among the variables describing their lipophilic, electronic, and steric properties, as required to investigate QSARs by multiple regression analysis (MRA). 24,25 Chemistry Cyclohexylcarbamic acid aryl esters 4a-c,e-i,l,k, m-w and 4x,y were obtained by addition of cyclohexylisocyanate to phenylphenols 3a-c,e,g-i,k,m-w (Scheme 1), or 3x,y (Scheme 3), respectively. Removal of the protective benzyloxycarbonyl from 4x,y afforded 4j,d (Scheme 3). 4z was prepared by hydrogenation of 4t.
The synthesis of 3p is reported in Scheme 4: the propenyl derivative 11, obtained by a Wittig reaction from aldehyde 9 was reduced to give 8p, which was eventually hydrolyzed to furnish the desired biphenol.

Results and Discussion
We measured FAAH activity in rat brain membranes, using [ 3 H]anandamide as a substrate. The IC 50 values for compounds 4a-z are reported in Table 1. At 30-100 µM, the compounds (i) had no effect on acetylcholinesterase activity (electric eel) or butyrylcholinesterase activity (horse serum); (ii) did not displace the binding of [ 3 H]WIN-55212-2 (10 nM) to rat CB 1 cannabinoid receptors or human recombinant CB 2 receptors; and (iii) did not affect [ 3 H]anandamide (100 nM) transport in human astrocytoma cells in culture (data not shown).
We first examined a small set of URB524 analogues (4a-l) the substituents of which (methyl, trifluoromethyl, amino, and carbamoyl) represent the four combinations of positive and negative levels for the π and σ (σ m or σ p ) descriptors; small (fluoro) and bulky (cyclohexylcarbamoyloxy or cyclohexylureidocarbonyl) substituents were also included in this explorative set. The results suggested that substitutions in the meta position of the distal phenyl ring yield the most potent inhibitors. Thus, the 3′-methyl (4h) and 3′-amino (4j) derivatives were as potent as the parent compound (URB524), while the 3′-carbamoyl derivative 4i (URB597) was an order of magnitude more potent than URB524. By contrast, with the only exception of the 4′-fluoro derivative 4e, all parasubstituted compounds were less active than URB524. In contrast with the general SAR trend, according to which meta-substituted compounds were more active than the corresponding para isomers, 3′-fluoro derivative 4k, the meta isomer of 4e, was less active than URB524. The requirements of the group matching the area of the receptor near the 4′-position seem more strict than those of the 3′-group, as hinted by the comparison of IC 50 ratio of compounds 4b,4a and 4h,4g, respectively. Given the results of this first set of derivatives, further modifications at the 4′-position were discharged. Moreover, compounds 4f and 4l demonstrated that the introduction of more extended fragments is detrimental to activity.
In order to search for statistical relationships between physicochemical properties and inhibitor potency, we inserted at the meta position twelve additional substituents (4m-z, Table 1), which adequately span the space of lipophilic, steric, and electronic properties. Certain substituents were also selected for their topological similarity with the carbamoyl group (4r) or with portions of it (4s, 4v, or 4z). The IC 50 values for 4m-z indicate that hydrophilic (4q-w) groups have a favorable impact on pharmacological activity. On the contrary, the introduction of large lipophilic groups (4mp) is detrimental to activity. Several compounds in this set are more active than URB524, although none of them is better than the m-carbamoyl derivative URB597 (4i).
The nineteen (including hydrogen) biphenyl substituents listed in Table 1 constitute a set with a large variation in both lipophilicity (almost 4 π units) and steric bulk (35 MR units). Furthermore, π and molar refractivity (MR) values are practically uncorrelated to the electronic effects (r with σ m of -0.19 and -0.16, respectively), though some correlation is present between lipophilic (π) and steric (MR) descriptors (r ) 0.63), due to the known difficulty in obtaining big hydrophilic substituents.
Multiple regression analysis (MRA) employing eight common physicochemical descriptors (π, σ m , F, R, MR,   Table 2) and π 2 did not yield a statistically significant model. However, a simple plot of pIC 50 vs lipophilicity (not shown) indicated a clear relationship, albeit masked by the presence of an outlier, the aminomethyl derivative 4z. This can be attributed to the basicity of the compound, which is expected to be largely protonated at neutral pH. When 4z was omitted from the regression set, the relationship between lipophilicity and potency gave the regression equation 1.
The negative correlation with lipophilicity was unexpected in that we had previously hypothesized that the biphenyl moiety may mimic the fatty acyl chain of a FAE bound to the hydrophobic binding site of FAAH. Yet, no MRA model including up to five variables turned out to be statistically better than, or equivalent to, that represented in eq 1. Only the inclusion of an indicator variable, set to one for substituents able to give hydrogen bonds (HB), and to zero in other cases, allowed the detection of an alternative model (eq 2) having comparable descriptive (r 2 ) and predictive (q 2 ) power. The pIC 50 values calculated by eq 2 are reported in Table 3, together with independent variables employed in the QSAR models.
Although this model is more complex than that represented by eq 1, it could provide a possible interpretation for the positive effect of substituent hydrophilicity within a supposedly lipophilic binding pocket: eq 2 relates this behavior to the formation of hydrogen bonds between meta substituents and polar amino acid residues of the enzyme. This kind of interaction is strictly dependent on distances and angles between the atoms involved, which may explain the fact that, in the case of para-substituted compounds, no positive effect for the polar carbamoyl (4c) and amino (4d) substituents was observed. The last compounds of the series (4rw), prepared to validate the temporary QSAR models built with the data previously available, confirmed the indications reported above, even if none of them resulted more potent than the carbamoyl derivative 4i.
The resolution of the crystal structure of FAAH covalently bound to methyl arachidonyl phosphonate (MAP) 23 allowed us to rationalize our QSAR models by performing docking, and molecular dynamics (MD) simulations, experiments. Bracey et al. observed that the binding site for MAP lies in a channel that spans the entire length of the enzyme, from the membranebound surface (Figure 2A, bottom) to the cytosol at the opposite side (Figure 2A, top). The surface of this channel, colored according to lipophilicity, is represented in Figure 2. The channel has a complex topography: it comprises a central hydrophilic area around the catalytic residue Ser241, which is surrounded in turn by extended lipophilic surfaces. Proceeding toward the membrane, the duct bifurcates into a lipophilic bulge (lower part of Figure 2A, back), in which the terminal atoms of the arachidonyl chain of MAP are located, and a narrow tunnel with a hydrophilic ridge (lower part of Figure 2A, front). The polar head of the FAEs might temporarily link to this ridge en route to the catalytic site.
After deletion of the MAP fragment from the enzyme and refinement of the protein model (see Experimental Section), we performed molecular docking with URB524 and URB597 (4i). The results of these analyses showed that the biphenyl moiety of URB524 may readily replace the arachidonyl chain of MAP, with the meta position of the distal phenyl ring pointing exactly toward the hydrophilic ridge ( Figure 2B). Indeed, a superposition placing the first two double bonds of the arachidonyl chain of MAP, in the conformation of its complex with FAAH, onto the biphenyl moiety of URB524 highlights the steric similarity of the two fragments and supports our initial hypothesis ( Figure 3).
Docking of URB597, the most active compound of the present series, suggests that its carbamoyl group may form two distinct hydrogen bonds: one as an acceptor, with the hydroxyl group of Thr488, and the other as a donor, with the main chain carbonyl group of Leu192 ( Figure 4). Docking of other meta-substituted structures, with the common biphenyl scaffold held in the same lipophilic region as in Figure 4, showed that all polar groups can form hydrogen bonds with the enzyme. However, the hydrogen bond of the carbamoyl group of URB597 is the strongest, possibly because its bidentate structure is absent in the acetyl and hydroxymethyl derivatives 4s and 4v. Accordingly, URB597 is more potent than the latter compounds (Table 1). This structure-based analysis, evidencing a role for specific hydrogen bonding interactions of our inhibitors with FAAH, may account for the empirical correlation described by eq 2 and provide an explanation for the negative correlation with lipophilicity (eq 1). The low tolerance to para substitution could be explained by the observation that, in the docking mode described above, the para position resulted so close to the side chains of Phe381, Phe432, and Leu404 that only a hydrogen or fluorine atom may avoid clashes with their aryl or alkyl groups (data not shown).
The fact that URB597 is the most potent inhibitor of our series prompted us to investigate in greater detail its interaction with FAAH using theoretical methods. Docking analysis revealed that a binding mode alternative to the one described above might be possible for this compound. In fact, a 180°rotation of the molecule (i) brings the N-cyclohexyl ring in a portion of the space occupied by the arachidonyl chain of MAP; and (ii) allows the biphenyl scaffold to lie in the funnel pointing to the cytosolic outlet, which is part of the FAAH channel. In this orientation it is still possible for the carbamoyl group of URB597 to form two hydrogen bonds, one with the backbone carbonyl of Leu380 and the other with the side-chain NH 2 of Gln273 (data not shown).
To test which of the two orientations is more stable we performed MD runs, maintaining the tertiary structure of FAAH observed in its crystallized complex with MAP and allowing movement of amino acid residues located at a maximum distance of 8 Å from the docked carbamate inhibitor. Starting from the orientation of Figure 4, URB597 gave a stable oscillation around a conformation very similar to its initial one, with an average distance between the nucleophilic oxygen of Ser241 and the electrophilic carbamate carbon of 3.00 ( 0.15 Å (mean ( SD). On the other hand, when MD runs were started from the alternative orientation, the ligand position was much less stable, the average distance between the two centers being of 4.10 ( 0.65 Å. Moreover, the m-carbamoyl group of URB597 left its original position and was able to undergo alternative polar interactions only at the cost of a worse adaptation of the carbamate fragment to the catalytic site. These results suggest that the binding mode depicted in Figure  4 may favor the nucleophilic attack of Ser241 after substrate recognition.
The stabilizing effect of the 3′-carbamoyl group in this binding orientation was further confirmed by comparing the MD trajectory of URB597 with that of its parent  compound URB524. The first structure showed reduced fluctuations, with respect to the initial conformation, indicating that the additional interaction provided by the carbamoyl group contributes to limit the movements of the inhibitor within the active site; this increases the probability of a reaction between Ser241 and the carbamate group, and could be related to the observed improvement of potency conferred by polar substituents in the 3′-position.
In conclusion, our findings indicate that the biphenyl scaffold in O-arylcarbamate inhibitors of FAAH may mimic the initial fatty acyl segment of anandamide and other FAEs, as they interact with the enzyme's active site. The results further suggest that polar substituents in the meta position of the distal phenyl group can form hydrogen bonds with hydrophilic amino acid residues within the FAAH channel. MD simulations supported this hypothesis, showing that such hydrogen bonding may increase the time spent by the electrophilic carbamate carbon near the catalytic residue Ser241, which should in turn favor the catalytic process. Since carbamate inhibitors of FAAH activity exert potent in vivo effects of therapeutic relevance, 29 including anxiolyticlike actions, a detailed knowledge of their SAR profile should be useful for further developments in this field.

Experimental Section
(a) Chemicals, Materials, and Methods. All reagents were purchased from Sigma-Aldrich in the highest quality commercially available. Solvents were RP grade unless otherwise indicated. Purification of the crude material obtained from the reactions, affording the desired products, was carried out by flash column chromatography on silica gel (Kieselgel 60, 0.040-0.063 mm, Merck). TLC analyses were performed on precoated silica gel on aluminum sheets (Kieselgel 60 F 254, Merck). Melting points were determined on a Bü chi SMP-510 capillary melting point apparatus and are uncorrected. The structures of the unknown compounds were unambiguously assessed by MS, 1 H NMR, IR, and elemental analysis. EI-MS spectra (70 eV) were recorded with a Fisons Trio 1000 spectrometer. Only molecular ions (M + ) and base peaks are given. 1 H NMR spectra were recorded on a Bruker AC 200 spectrometer. Chemical shifts (δ scale) are reported in parts per million (ppm) relative to the central peak of the solvent. IR spectra were obtained on a Shimadzu FT-8300 or a Nicolet Atavar 360 FT spectometer. Absorbances are reported in ν (cm -1 ). Elemental analyses were performed on a Carlo Erba analyzer. All products had satisfactory (within (0.4 of theoretical values) C, H, N analysis results.
(c) QSAR. MRA calculations were performed with an Excel (Microsoft Co., version 97) spreadsheet, employing the builtin statistical functions and automated macro procedures. Substituent properties were parametrized by the variables π, π 2 σm, F, R, MR, L, B1, B5. When available, their values were taken from the van de Waterbeemd data set; 47 for the benzyl, hydroxyethyl, and aminomethyl substituents they were taken from the Hansch collection; 48 π and MR values for the cyclohexylcarbamoyloxy group (4l) were estimated by the Chem Prop module in Chem Draw, 49 and those of the electronic variables from similar alkylaminocarbonyloxy groups in the Hansch collection. MRA models were calculated for all the possible combinations of maximum five variables. Standard deviation of the errors in prediction (SDEP) and the relative predictivity parameter, q 2 , were calculated by cross-validation, omitting one compound at a time from the set, according to the leave-one-out technique (LOO). 50 (d) Molecular Modeling. Molecular models were built, refined, and analyzed by Sybyl version 6.8. 51 Energy calculations were performed employing the Merck molecular force field (MMFF94s for geometry optimization and MMFF94 for molecular dynamics), 52 implemented in Sybyl, with the dielectric set constant to 1. FAAH structure coordinates were taken from those of the covalent adduct with a MAP chain, reported in the Protein Data Bank 53 (PDB code: 1MT5). The first structure refinement (addition of missing side chains and hydrogens) was done by the Biopolymer module, and it was followed by a visual inspection of histidine tautomerism, which was modified to maximize the number of possible hydrogen bonds. The geometry of added atoms was then relaxed by energy minimization to a gradient of 0.1 kcal/(mol‚Å), the MAP atoms removed, and the hydrogen atoms at catalytic site residues were reassigned to get a hydrogen bond between Ser241 and Ser217, and one between Ser217 and the NH2 group of Lys142. The inhibitors were then fitted into the enzyme channel, optimizing their position and conformation first by the Dock_minimize procedure, then by energy minimization of the complex, allowing movements of the residues at maximum 8 Å from the inhibitor. Molecular dynamics simulations (step size of 1 fs) started with seven 3000-step heating cycles, gradually raising T from 0 to 310 K and continued with 200000 fs of simulation; after 100000 fs of equilibration, a snapshot of the trajectory was saved every 500 fs for subsequent analysis. The surface of the enzyme channel shown in Figures 2 and 4 was built by the MOLCAD module in Sybyl.