Insights into the mechanism and inhibition of fatty acid amide hydrolase from quantum mechanics/molecular mechanics (QM/MM) modelling

Enzyme Mechanisms: Fast Reaction and Computational Approaches 363 Insights into the mechanism and inhibition of fatty acid amide hydrolase from quantum mechanics/molecular mechanics (QM/MM) modelling Alessio Lodola*, Marco Mor*, Jitnapa Sirirak and Adrian J. Mulholland 1 *Dipartimento Farmaceutico, Università degli Studi di Parma, viale G.P. Usberti 27/A Campus Universitario, I-43100 Parma, Italy, and Centre for Computational Chemistry, School of Chemistry, University of Bristol, Bristol BS8 1TS, U.K. Abstract FAAH (fatty acid amide hydrolase) is a promising target for the treatment of several central nervous system and peripheral disorders. Combined QM/MM (quantum mechanics/molecular mechanics) calculations have elucidated the role of its unusual catalytic triad in the hydrolysis of oleamide and oleoylmethyl ester substrates, and have identified the productive inhibitor-binding orientation for the carbamoylating compound URB524. These are potentially crucial insights for designing new covalent inhibitors of this drug target. Introduction Computational modelling [e.g. with combined QM/MM (quantum mechanics/molecular mechanics) methods] is increasingly making practical contributions to our understanding of enzyme-catalysed reaction mechanisms [1]. QM (quantum chemical electronic structure) methods allow the modelling of chemical reactions, and are potentially highly accurate for small molecules. However, electronic structure calculations require large computational resources, which significantly limits the size of the system that can be treated, making calculations on whole enzymes (composed of thousands of atoms) prohibitive. Calculations on enzymecatalysed reactions are possible with QM/MM methods: they link QM and MM representations of different parts of a system [2]. The combination of these approaches contains the fundamentals necessary to describe the PESs (potential energy surfaces) relevant to enzymatic chemistry, at least to a first approximation [3]. In the QM/MM framework, the reactive region of the enzyme active site is described by a QM technique (e.g. semi-empirical, density functional theory or ab initio molecular orbital methods); the QM region should be of sufficient size to include crucial electronic structure effects in the reaction [4,5]. Semi-empirical molecular orbital techniques such as AM1 and PM3, although not suitable for all systems, allow large QM regions to be treated but often overestimate reaction barriers [6]. Reliable calculations have increasingly been made possible by the development of QM/MM methods based on density functional theory and Key words: fatty acid amide hydrolase (FAAH), inhibitor design, quantum mechanics/molecular mechanics (QM/MM), reaction mechanism, tetrahedral intermediate, transition state. Abbreviations used: FAAH, fatty acid amide hydrolase; OA, oleamide; OME, oleoylmethyl ester; PES, potential energy surface; QM/MM, quantum mechanics/molecular mechanics; TI, tetrahedral intermediate; TS, transition state. 1 To whom correspondence should be addressed (email Adrian.Mulholland@bristol.ac.uk). on ab initio electron correlation approaches such as the MP2 perturbation method and coupled-cluster theory [7,8]. Many important drug targets are enzymes. Ligand design could significantly benefit from detailed, molecularlevel knowledge of their reaction mechanisms. QM/MM modelling has the potential to show in atomic detail how TSs (transition states) and intermediates are stabilized within enzymes. Increasingly, mechanistic simulations will provide important insights, of the type described here, for medicinal chemists and will thus help in the design of new inhibitors and the development of predictive models of drug metabolism [9]. FAAH (fatty acid amide hydrolase): a case study for QM/MM reaction modelling FAAH is a key enzyme in endocannabinoid metabolism [10], and is a promising target for treatment of nervous system disorders such as anxiety, pain and depression [11]. FAAH is a member of the amidase signature family endowed with alys 142 Ser 217 Ser 241 catalytic triad [12]. FAAH shows a remarkable ability to hydrolyse amides and esters at similar rates, by a mechanism in which acylation is rate limiting [13]. Considerable experimental effort [14] has been invested into elucidating the detailed mechanism of hydrolysis by FAAH. Lys 142 appears to serve as a key acid and base in distinct steps of the catalytic cycle. As a base, Lys 142 activates the Ser 241 nucleophile for attack on the substrate amide carbonyl. As an acid, Lys 142 readily protonates the substrate-leaving group leading to its expulsion (Figure 1). The impact of Lys 142 on Ser 241 nucleophile strength and leaving group protonation occurs indirectly, via the bridging Ser 217 of the triad, which acts as a proton shuttle [15]. The three-dimensional structure of FAAH has been determined by X-ray crystallography [16], prompting computational investigations on the interactions of FAAH Biochem. Soc. Trans. (2009) 37, 363 367; doi:10.1042/bst0370363

364 Biochemical Society Transactions (2009) Volume 37, part 2 Figure 1 Proposed acylation mechanism for FAAH [17] YisNH 2 for OA (the label N am is used for its leaving group nitrogen in the text) and CH 3 OforOME(O es is used as a label for its leaving group oxygen in the text). Figure 2 PM3-CHARMM22 QM/MM potential energy profiles for the hydrolysis of OA (black) [17,21] and OME (grey) in FAAH A G are relevant configurations along the calculated pathways: A is the substrate complex; C is the TI; E is the protonated TI; G is the product; B, D and F are the (approximate) TS structures for the different steps. Y is NH 2 for OA and CH 3 OforOME. with substrates and with inhibitors. We have modelled the mechanism of the first step of the acylation reaction of the substrate OA (oleamide) [17]. An adiabatic mapping approach was applied to scan the PES for the reaction at the PM3-CHARMM22 QM/MM level. Density functional theory calculations (with the B3LYP functional) were also performed, for a better description of the reaction energetics [18]. In our model, the first reaction step of the acylation consists of three events (Figure 1). A proton is abstracted from Ser 241 and, via the bridging residue cis-ser 217, is transferred to the general base Lys 142. The activated Ser 241 attacks the carbonyl group of the OA substrate, leading to the formation of the TI (tetrahedral intermediate). To model these events, two reaction co-ordinates were applied, defined in terms of interatomic distances: R x defined as [d(o 1, H 1 ) d(o 2,H 1 ) d(o 1, C)] including proton abstraction from Ser 241 by Ser 217 and the nucleophilic attack by Ser 241 ; R y, defined as [(do 2,H 2 ] d[(n, H 2 )], describes the proton transfer between Ser 217 and Lys 142. R x and R y were increased respectively in steps of 0.2 and 0.1 Å (1 Å = 0.1 nm) with harmonic restraints of 5000 kcal mol 1 Å 2 (1 kcal = 4.184 kj) as described in [17]. The PM3-CHARMM22 PES indicated that the deprotonation of the Ser 241 nucleophile is the rate-limiting step of the reaction, with a barrier of 36 kcal mol 1 at this level of theory (Figure 2). As the subsequent nucleophilic attack takes place with no barrier, TI formation and Ser 241 deprotonation are concerted events. The TI (configuration C, Figure 2) was calculated to be less stable than the Michaelis complex (substrate complex, A) by 26 kcal mol 1, indicating its transient character. The TI is significantly stabilized by the enzyme; however, comparison of the PM3-CHARMM22 PES and the in vacuo PM3 PES showed that the highest level of enzyme stabilization along the pathway was reached at TI due to crucial hydrogen bonds involving the charged oxygen of the OA and NH groups of the oxyanion hole residues.

Enzyme Mechanisms: Fast Reaction and Computational Approaches 365 Figure 3 Structures of important points along the QM/MM reaction pathway for the acylation reaction with OA in FAAH ([17], see also Figure 2) Left-hand panel: TI (C in Figure 2). Middle panel: protonated TI (E in Figure 2). Right-hand panel: acylenzyme (G in Figure 2). Energy correction at the higher B3LYP/6-31+G(d) level gave a barrier of 18 kcal mol 1, close to the experimentally deduced activation barrier for the overall reaction of OA of 16 kcal mol 1 [13]. The effect of conformational fluctuations on catalysis is a key issue of debate in enzymology [19]. QM/MM calculations based on energy minimization alone, e.g. adiabatic mapping along a reaction co-ordinate (as opposed to simulations incorporating the effects of thermal fluctuations, e.g. molecular dynamics or Monte Carlo simulations [20]), may hide potential pitfalls due to conformational transitions at the protein active site. We have investigated the relation between conformational fluctuations and reactivity for the first step of the acylation reaction between FAAH and OA. PESs were calculated at the PM3-CHARMM22 and B3LYP/6-31+G(d)//PM3-CHARMM22 levels by using several conformations (four) of the enzyme substrate complex. The results showed that geometrical fluctuations of the active site can significantly affect the overall energetic barrier and that important conformational changes occur in the active site in the reaction [21]. Reaction of OA apparently proceeds via a minor conformation of the enzyme, in which the TS is better stabilized. We have also modelled the second step of the FAAH OA acylation reaction at the PM3-CHARMM22 QM/MM level, employing the same type of adiabatic mapping approach. Formation of the acylenzyme from the TI requires several proton transfer processes involving both Lys 142 and Ser 217. Once the leaving group nitrogen is protonated, the breakdown of the amide bond can be completed. Various reaction co-ordinates were tested in QM/MM calculations of PESs for this reaction step, and only one was found to give a smooth, satisfactory shape. This surface was calculated starting from the TI, splitting the modelling of the reaction into two events: (i) protonation of the leaving group and (ii) breaking of C N bond. For step (i), two reaction coordinates were employed, namely R r defined as the difference between atomic distances [d(o 2, H 1 ) d(n am, H 1 )] moving the proton H 1 from Ser 217 to N am of OA, and R s defined as [d(n, H 2 ) d(o 2,H 2 )] moving proton H 2 from Lys 142 back to Ser 217. Both reaction co-ordinates were restrained to a sequence of values with harmonic restraints of 5000 kcal mol 1 Å 2, and increased in steps of 0.1 Å; geometry optimization was performed at each point to achieve an energy gradient of 0.01 kcal mol 1 Å 1. The PM3-CHAR- MM22 PES showed a mechanism where the protonation of the basic nitrogen of OA (N am )isconcertedwiththe transfer of proton H 2 from Lys 142 to Ser 217. However, visual inspection of the structures along the modelled pathway showed a rearrangement of the TI due to the pyramidal inversion of the N am. This event is crucial, because the lone pair on N am needs to be correctly oriented towards the oxygen of Ser 217 to accept the H 1 proton (Figure 3). All these events lead to configuration E (Figure 2) overcoming a barrier of 40 kcal mol 1 (relative to the substrate complex), 3 kcal mol 1 higher than that calculated for the deprotonation of Ser 241. The protonated TI is calculated to be more stable than the TI itself: the energy of E is 12 kcal mol 1 with respect to the substrate complex. Finally, the expulsion of the leaving group (e.g. neutral ammonia in the reaction of OA) was modelled by increasing (in steps of 0.1 Å) the reaction co-ordinate R z,[d(c, N am )]. This reaction led to the formation of the acylenzyme G overcoming a barrier of 16 kcal mol 1 (relative to the substrate complex). The full energy profile for the acylation reaction indicates that formation of the TI is not the rate-limiting step, because the highest overall barrier was found for the protonation of the leaving group. However, the small difference between the calculated barriers (36 compared with 40 kcal mol 1 ) suggests that the acylation depends on both reaction steps, with no step being clearly rate limiting. The general picture of the mechanism of OA hydrolysis in FAAH is similar to that found in similar computational work by Tubert-Brohman et al. [22], who performed free energy perturbation simulations for the reaction at the PDDG/PM3 QM/MM level. It is encouraging that independent simulations by different groups with different QM/MM techniques arrive at the same qualitative mechanistic conclusions, and strengthen these conclusions. A small difference between the results is that Tubert-Brohman et al. [22] found the highest

366 Biochemical Society Transactions (2009) Volume 37, part 2 barrier for TI formation (rather then collapse). This difference may be related to the different calculated relative energies of the TI, as the energy cost for the protonation and expulsion of the leaving group (E-F-G pathway in Figure 2) is comparable with both methods ( 15 kcal mol 1 relative to the energy of the TI). In our calculations, the TI is less stable than the substrate complex by 26 kcal mol 1, whereas with the QM/MM/FEP (free energy perturbation) method the TI is predicted to be more stable than the substrate complex by 0.4 1.9 kcal mol 1 [22]. This may be due to the more extensive conformational sampling performed by Tubert-Brohman et al. As noted above, potential energy barriers (as reported here) are dependent on the conformation of the enzyme used for modelling; another important consideration is that the PM3 method overestimates barriers for the reaction. The potential energy profiles described here (e.g. Figure 2) should not be directly compared with experimental values for these reasons. They are, however, useful, for comparison of alternative substrates and conformations. To investigate the experimental observation that FAAH cleaves amides somewhat faster than esters [13], we have also modelled the acylation mechanism for OME (oleoylmethyl ester). The FAAH OME system was prepared by applying the same procedure described for OA [17]. The same PM3- CHARMM22 adiabatic mapping approach was used to explore PESs of the reaction. The PES for the first step of the acylation was explored by applying the same reaction co-ordinates (R x and R y ) and the same harmonic restraints as for OA. Deprotonation of Ser 241 was found to have a barrier of 44 kcal mol 1, relative to the substrate complex A (Figure 2). This event is in concert with the nucleophilic attack and leads to the TI. The subsequent breakdown of the TI was divided into two events, similar to the modelling of the OA reaction: (i) protonation of the alcohol-leaving group and (ii) breaking of the C O bond. For step (i), two reaction co-ordinates were employed, namely R r defined as [d(o 2,H 1 ) d(o es H 1 )] moving the proton H 1 from Ser 217 to O es of OME, and R s defined as [d(n, H 2 ) d(o 2,H 2 )] moving proton H 2 from Lys 142 back to Ser 217. R r and R s were increased in steps of 0.1 Å with harmonic restraints of 5000 kcal mol 1 Å 2. For step (ii) (expulsion of methanol), the reaction co-ordinate R z {the distance [(C, O es )]} was increased in steps of 0.1 Å. The highest energy point along the path was the TS for protonation of the leaving group oxygen (O es ). This process led to E (Figure 2) overcoming a barrier of 47 kcal mol 1 relative to the substrate complex A. The subsequent expulsion of methanol happened spontaneously, as no effective barrier divides E from the acylenzyme. For OME, as for OA, the reaction rate is predicted to depend to a similar extent on TI formation and collapse. Tubert- Brohman et al. [22] have also compared the reactions of these two substrates, but only the first step of acylation for OME. Figure 2 summarizes the energetics (at the PM3- CHARMM22 level) for acylation of OA and OME. The overall calculated barrier (relative to A) for the amide cleavage is 40 kcal mol 1, 7 kcal mol 1 less than the barrier for the ester. These results are in qualitative agreement with kinetic investigations on OA and OME substrates: the experimental barrier for the hydrolysis of OA is lower by 0.6 kcal mol 1 than that for OME [13]. The calculations suggest that this unusual preference for the amide depends both on a more efficient stabilization of the TS for the deprotonation of Ser 241 and on higher basicity of the nitrogen N am compared with that of oxygen O es in the TI. We have also used this QM/MM modelling approach to investigate the reaction between FAAH and the carbamic inhibitor URB524 [23]. Carbamates and urea compounds inactivate FAAH by carbamoylating the Ser 241 nucleophile [24,25]. URB524 and its derivatives can be docked in two possible orientations within the FAAH catalytic site (called orientations I and II), both of which place the carbamic group close to Ser 241. Traditional computational tools employed in drug discovery (such as docking and scoring) failed to discriminate clearly between these two binding orientations [26,27]. We have studied the carbamoylation reaction of this inhibitor in FAAH, employing the PM3-CHARMM22 potential, with B3LYP/6-31G+(d) energy corrections. PESs were calculated for each binding orientation, and TS structures and intermediates were identified along the reaction profiles. The calculations clearly showed that reaction was significantly energetically preferred in orientation II, identifying II as the productive binding mode [23]. These findings are consistent with the recent development of potent new FAAH inhibitors, and with linear interaction energy calculations [28,29] for a series of 22 new FAAH carbamoylating compounds [30]. Altogether, our results indicate that QM/MM modelling of reaction mechanisms of covalent inhibitors in enzyme targets has the potential to provide detailed information for drug design in cases where docking approaches alone may be inadequate [31]. Acknowledgements We thank our co-authors and collaborators in relevant work reviewed here. Funding A.J.M. is supported by an Engineering and Physical Sciences Research Council Leadership Fellowship [grant number EP/G007705/1]. References 1 Cavalli, A., Carloni, P. and Recanatini, M. (2006) Target-related applications of first principles quantum chemical methods in drug design. Chem. Rev. 106, 3497 3519 2 Field, M.J., Bash, P.A. and Karplus, M. (1990) A combined quantum mechanical and molecular mechanical potential for molecular dynamics simulations. J. Comput. Chem. 11, 700 733 3 Warshel, A. (2003) Computer simulations of enzyme catalysis: methods, progress, and insights. Annu. Rev. Biophys. Biomol. Struct. 32, 425 443 4 Friesner, R.A. and Guallar, V. (2005) Ab initio quantum chemical and mixed quantum mechanics/molecular mechanics (QM/MM) methods for studying enzymatic catalysis. Annu. Rev. Phys. Chem. 56, 389 427 5 Senn, H.S. and Thiel, W. (2007) QM/MM studies of enzymes. Curr. Opin. Chem. Biol. 11, 182 187

Enzyme Mechanisms: Fast Reaction and Computational Approaches 367 6 Ridder, L. and Mulholland, A.J. (2003) Modeling biotransformation reactions by combined quantum mechanical/molecular mechanical approaches: from structure to activity. Curr. Top. Med. Chem. 3, 1241 1256 7 Claeyssens, F., Harvey, J.N., Manby, F.R., Mata, R.A., Mulholland, A.J., Ranaghan, K.E., Schütz, M., Thiel, S., Thiel, W. and Werner, H.-J. (2006) High accuracy computation of reaction barriers in enzymes. Angew. Chem. Int. Ed. Engl. 45, 6856 6859 8 Mulholland, A.J. (2007) Chemical accuracy in QM/MM calculations on enzyme-catalysed reactions. Chem. Cent. J. 1, 1 19 9 Mulholland, A.J. (2005) Modelling enzyme reaction mechanisms, specificity and catalysis. Drug Discov. Today 10, 1393 1402 10 Piomelli, D. (2003) The molecular logic of endocannabinoid signalling. Nat. Rev. Neurosci. 4, 873 884 11 Piomelli, D., Tarzia, G., Duranti, A., Tontini, A., Mor, M., Compton, T.R., Dasse, O., Monaghan, E.P., Parrott, J.A. and Putman, D. (2006) Pharmacological profile of the selective FAAH inhibitor KDS-4103 (URB597). CNS Drug Rev. 12, 21 38 12 Labar, G. and Michaux, C. (2007) Fatty acid amide hydrolase: from characterization to therapeutics. Chem. Biodiversity 4, 1882 1902 13 McKinney, M.K. and Cravatt, B.F. (2003) Evidence for distinct roles in catalysis for residues of the serine-serine-lysine catalytic triad of fatty acid amide hydrolase. J. Biol. Chem. 278, 37393 37399 14 McKinney, M.K. and Cravatt, B.F. (2005) Structure and function of fatty acid amide hydrolase. Annu. Rev. Biochem. 74, 411 432 15 Ahn, K., McKinney, M.K. and Cravatt, B.F. (2008) Enzymatic pathways that regulate endocannabinoid signaling in the nervous system. Chem. Rev. 108, 1687 1707 16 Bracey, M.H., Hanson, M.A., Masuda, K.R., Stevens, R.C. and Cravatt, B. F. (2002) Structural adaptation in a membrane enzyme that terminates endocannabinoid signaling. Science 298, 1793 1796 17 Lodola, A., Mor, M., Hermann, J.C., Tarzia, G., Piomelli, D. and Mulholland, A.J. (2005) QM/MM modelling of oleamide hydrolysis in fatty acid amide hydrolase (FAAH) reveals a new mechanism of nucleophile activation. Chem. Commun., 4399 4401 18 Mulholland, A.J. (2008) Computational enzymology: modelling the mechanisms of biological catalysts. Biochem. Soc. Trans. 36, 22 26 19 Klähn, M., Braun-Sand, S., Rosta, E. and Warshel, A. (2005) On possible pitfalls in ab initio quantum mechanics/molecular mechanics minimization approaches for studies of enzymatic reactions. J. Phys. Chem. B 109, 15645 15650 20 van der Kamp, M.W., Shaw, K.E., Woods, C.J. and Mulholland, A.J. (2008) Biomolecular simulation and modelling: status, progress and prospects. J. R. Soc. Interface 5, S173 S190 21 Lodola, A., Mor, M., Zurek, J., Tarzia, G., Piomelli, D., Harvey, J.N. and Mulholland, A.J. (2007) Conformational effects in enzyme catalysis: reaction via a high energy conformation in fatty acid amide Hydrolase. Biophys. J. 92, L20 L22 22 Tubert-Brohman, I., Acevedo, O. and Jorgensen, W.L. (2006) Elucidation of hydrolysis mechanisms for fatty acid amide hydrolase and its Lys142Ala variant via QM/MM simulations. J. Am. Chem. Soc. 128, 16904 16913 23 Lodola, A., Mor, M., Rivara, S., Christov, C., Tarzia, G., Piomelli, D. and Mulholland, A.J. (2008) Identification of productive inhibitor binding orientation in fatty acid amide hydrolase (FAAH) by QM/MM mechanistic modelling. Chem. Commun., 214 216 24 Alexander, J.P. and Cravatt, B.F. (2005) Mechanism of carbamate inactivation of FAAH: implications for the design of covalent inhibitors and in vivo functional probes for enzymes. Chem. Biol. 12, 1179 1187 25 Ahn, K., Johnson, D.S., Fitzgerald, L.R., Liimatta, M., Arendse, A., Stevenson, T., Lund, E.T., Nugent, R.A., Nomanbhoy, T.K., Alexander, J.P. and Cravatt, B.F. (2007) Novel mechanistic class of fatty acid amide hydrolase inhibitors with remarkable selectivity. Biochemistry 46, 13019 13030 26 Mor, M., Rivara, S., Lodola, A., Plazzi, P.V., Tarzia, G., Duranti, A., Tontini, A., Piersanti, G., Kathuria, S. and Piomelli, D. (2008) Cyclohexylcarbamic acid 3 -or4 -substituted biphenyl-3-yl esters as fatty acid amide hydrolase inhibitors: synthesis, quantitative structure-activity relationships, and molecular modeling studies. J. Med. Chem. 47, 4998 5008 27 Tarzia, G., Duranti, A., Gatti, G., Piersanti, G., Tontini, A., Rivara, S., Lodola, A., Plazzi, P.V., Mor, M., Kathuria, S. and Piomelli, D. (2006) Synthesis and structure activity relationships of FAAH inhibitors: cyclohexylcarbamic acid biphenyl esters with chemical modulation at the proximal phenyl ring. ChemMedChem 1, 130 139 28 Carlson, H.A. and Jorgensen, W.L. (1995) An extended linear response method for determining free energies of hydration. J. Phys. Chem. 99, 10667 10673 29 Zhou, R., Friesner, R.A., Ghosh, A., Rizzo, R.C., Jorgensen, W.L. and Levy, R.M. (2001) New linear interaction method for binding affinity calculations using a continuum solvent model. J. Phys. Chem. B 105, 10388 10397 30 Mor, M., Lodola, A., Rivara, S., Vacondio, F., Duranti, A., Tontini, A., Sanchini, S., Piersanti, G., Clapper, J.R., King, A.R. et al. (2008) Synthesis and quantitative structure activity relationship of fatty acid amide hydrolase inhibitors: modulation at the n-portion of biphenyl-3-yl alkylcarbamates. J. Med. Chem. 51, 3487 3498 31 Lodola, A., Woods, C.J. and Mulholland, A.J. (2008) Applications and advances of QM/MM methods in computational enzymology. Annu. Rep. Comput. Chem. 4, 155 169 Received 17 November 2008 doi:10.1042/bst0370363