Systematic QM/MM investigation of factors that affect the cytochrome P450-catalyzed hydrogen abstraction of camphor

Systematic QM/MM Investigation of Factors that Affectthe Cytochrome P450-Catalyzed Hydrogen Abstraction of

Camphor

AHMET ALTUN,1 SASON SHAIK,2 WALTER THIEL1

1Max-Planck-Institut fur Kohlenforschung, Kaiser-Wilhelm-Platz 1, D-45470 Mulheim an derRuhr, Germany

2Department of Organic Chemistry and the Lise-Meitner-Minerva Center for ComputationalQuantum Chemistry, The Hebrew University of Jerusalem, 91904 Jerusalem, Israel

Received 12 December 2005; Accepted 10 January 2006DOI 10.1002/jcc.20398

Published online in Wiley InterScience (www.interscience.wiley.com).

Abstract: The hydrogen abstraction reaction of camphor in cytochrome P450cam has been investigated in the nativeenzyme environment by combined quantum mechanical/molecular mechanical (QM/MM) calculations and in the gasphase by density functional calculations. This work has been motivated by contradictory published QM/MM results. Inan attempt to pinpoint the origin of these discrepancies, we have systematically studied the factors that may affect thecomputed barriers, including the QM/MM setup, the optimization procedures, and the choice of QM region, basis set,and protonation states. It is found that the ChemShell and QSite programs used in the published QM/MM calculationsyield similar results at given geometries, and that the discrepancies mainly arise from two technical issues (optimizationprotocols and initial system preparation) that need to be well controlled in QM/MM work. In the course of thesesystematic investigations, new mechanistic insights have been gained. The crystallographic water 903 placed near theoxo atom of Compound I lowers the hydrogen abstraction barrier by ca. 4 kcal/mol, and thus acts as a catalyst for thisreaction. Spin density may appear at the A-propionate side chain of the heme if the carboxylate group is not properlyscreened, which might be expected to happen during protein dynamics, but not in static equilibrium situations. Thereis no clear correlation between the computed A-propionate spin density and the hydrogen abstraction barrier, and hence,no support for a previously proposed side-chain mediated transition state stabilization mechanism. Standard QM/MMoptimizations yield an A-propionate environment close to the X-ray structure only for protonated Asp297, and not fordeprotonated Asp297, but the computed barriers are similar in both cases. An X-ray like A-propionate environment canalso be obtained when deprotonated Asp297 is included in the QM region and His355 is singly protonated, but thisCompound II-type species with a closed-shell porphyrin ring has a higher hydrogen abstraction barrier and should thusnot be mechanistically relevant.

© 2006 Wiley Periodicals, Inc. J Comput Chem 27: 1324–1337, 2006

Key words: QM/MM calculations; cytochrome P450; H-abstraction reaction; camphor; compound I

Introduction

The cytochrome P450 enzymes constitute a ubiquitous hemopro-tein superfamily in bioorganisms. They catalyze a vast amount ofstereospecific and regioselective COH bond hydroxylation pro-cesses into several organic substrates, which are of vital impor-tance for regulating drug metabolism, detoxification and biosyn-thesis.1–3 The principal active species of P450 enzymes is assumedto be an elusive high-valence oxoiron–cationic porphyrin species,known as Compound I (Cpd I).4 Many experimental observations,such as stereochemical scrambling5 and intrinsic kinetic isotopeeffect measurements,6 as well as theoretical studies1,2 support a

two-state rebound mechanism of COH hydroxylation (see Scheme1).7 The rate-determining step of the rebound process is the initialH-abstraction reaction of the substrate RH, in which the reactant

Correspondence to: W. Thiel; e-mail: [email protected]

Contract/grant sponsor: the German Federal Ministry of Education andResearch (BMBF) within the framework of the German-Israeli ProjectCooperation (DIP) (to S. Shaik)

Contract/grant sponsor: the Volkswagenstiftung (to W. Thiel)

This article contains Supplementary Material available athttp://www.interscience.wiley.com/jpages/0192-8651/suppmat

© 2006 Wiley Periodicals, Inc.

complex RC is connected to a hydroxo intermediate radical speciesHYD by a transition structure TSH. In the subsequent COO bondformation step, a ferric–alcohol complex PC is formed via atransition structure TSR, which has a significant barrier on thequartet surface, but a negligible one on the doublet surface8–10 sothat the corresponding transition state TSR has rarely been locatedfor the doublet.8

P450cam from the bacterium Pseudomonas putida hydroxylatescamphor at the C5 position producing only the exo alcohol.11

Camphor hydroxylation by Cpd I of P450cam was recently inves-tigated by independent quantum mechanical/molecular mechanical(QM/MM) calculations in the native enzyme environment by thegroups of Shaik/Thiel (S/T)8 and Guallar/Friesner (G/F).12,13 S/Tstudied the complete rebound mechanism for four different snap-shots of a molecular dynamics (MD),8,14 while G/F addressed onlythe initial hydrogen abstraction step on the quartet surface startingfrom an X-ray-derived structure (pdb code: 1DZ9).15 The S/Tsystem comprises the enzyme and a surrounding water layer of 16Å (total of 24,394 atoms, 16,956 solvent atoms, net charge of�10e). The G/F system has 7448 atoms and no net charge becausethe polar surface residues (Lys, Arg, Glu, and Asp) that do notform salt bridges were neutralized to account for screening bywater solvent, which was not treated explicitly. During QM/MMgeometry optimizations S/T kept the outer region of their systemfixed and optimized only the inner region,8,14 while G/F con-strained the positions of the neutralized polar surface residues. Theheme side chains were incorporated into the QM region by G/F,but not by S/T in their standard setup. The QM/MM boundary wastreated by frozen orbitals (G/F) and link atoms with charge shiftshift model (S/T). The calculations were carried out using theQSite16 (G/F) and ChemShell17 (S/T) programs. (ChemShell is amodular QM/MM program developed in the European QUASIproject under the coordination of P. Sherwood. See: http://www.cse.clrc.ac.uk/qcg/chemshell).

Internally consistent results were found by S/T8 in the unre-stricted open-shell B3LYP (UB3LYP)/CHARMM hybrid studyfor all snapshots considered, TSH and HYD being ca. 21 (22) and15 (11) kcal/mol above RC on the quartet surface at the R1/B1(R2/B2) level (see below for definitions). By contrast, TSH andHYD were only 11.7 and 4.5 kcal/mol above RC in the restrictedopen-shell B3LYP (ROB3LYP)/OPLS-AA study of G/F at theR2s/B�g* level (see below for definitions). The inclusion of thecrystallographic water 903 (w903) placed near the oxo ligand ofCpd I into this QM region did not change the ROB3LYP/OPLSrelative energies.12,13 According to ROB3LYP/OPLS results, thelone pair orbitals of carboxylate oxygen in the A-propionate (A-prop) heme side chain are not fully occupied, yielding excess spinpolarization on O2A atom of A-prop, which decreases from 0.20ein RC to 0.10e and 0.12e in TSH and HYD. This spin densitydecrease is accompanied by a concomitant increase of negativecharge on O2A, which has been proposed to provide a differentialstabilization of TSH and HYD through favorable electrostaticinteractions with the positively charged hydrogen atom onArg299.12,13 To check the effect of the side chains, S/T performedextensive single-point UB3LYP (gas phase) and UB3LYP/CHARMM (protein environment) calculations employing largeQM regions that extend into the propionate side chains of hemeand beyond; in one of these systems the geometry was optimizedat the UB3LYP/CHARMM level.8 These calculations do not showa barrier lowering effect by the side chains or any spin density atthe propionates as long as these negatively charged groups arescreened by the neighboring Arg112 and Arg299 residues. Like-wise, propionate side chains do not carry spin density in recentQM/MM studies on Cpd I of human isoforms of P45018 and ofhorseradish peroxidase (HRP) when large basis sets are used;19 theformer study18 also treats Cpd I of P450cam, but without includingthe heme side chains in the QM region.

As outlined above, the published QM/MM investigations oncamphor hydroxylation by P450cam have followed the same gen-eral strategy, but differed in system setup and QM/MM method-ology, and there are significant discrepancies between the pub-lished results. It is obviously important to identify the source ofthese discrepancies, which calls for a systematic study of thefactors that affect the barrier heights and spin densities in theQM/MM calculations. A brief account of this work is givenelsewhere.20 Here, we report the detailed results for the initialhydrogen abstraction by Cpd I of P450cam on the quartet surface.We focus on the lowest energy electromers, RC(IV), TSH(III), andHYD(III), which have previously been studied with conflictingresults.8,12,13

Computational Details

Setup of the System

All systems used in this study include only the monomer corre-sponding to the protein chain A and solvent chain Z of thecrystallographic dimer in pdb file 1DZ9.15 After adding missinghydrogen atoms, G/F employed this X-ray structure as startingpoint for their QM/MM calculations. To allow for direct compar-isons, we adopt an analogous structure (labeled snapX) for our

Scheme 1.

H-abstraction Reaction by P450cam 1325

Journal of Computational Chemistry DOI 10.1002/jcc

present QM/MM calculations. Going back to the S/T setup proce-dure (see ref. 14 for technical details), we start from the X-raystructure (1DZ9), add missing hydrogen atoms as well as an outerwater layer, and perform a CHARMM21 minimization to removeclose contacts keeping the coordinates of heme-Cys357 unit andouter 8 Å solvent layer fixed.

Several titratable residues in G/F and S/T structures havedifferent protonation states. S/T8,14 assigned histidines 17, 176,347, and 391 {21, 62, 337, and 361} to be protonated at the � (Hse){� (Hsd)} nitrogen only, while the other histidines are fully pro-tonated at both N (Hsp); Glu366 is protonated. In the unsolvatedneutral G/F sytem, the following residues are neutralized: Arg 79,90, 109, 130, 161, 212, 277 and 365; Lys 126, 197, 214, 239, 313,314, 344, 372, 392, and 412; Glu 20, 47, 76, 84, 91, 94, 107, 128,133, 152, 156, 171, 172, 195, 198, 209, 269, 279, 306, and 329;Asp 27, 77, 97, 104, 125, 153, 173, 188, 202, 304, 339, and 407;histidines 347, 352, and 391 {21, 62, 80, 176, 270, 337, 355, and361} are protonated at the � {�} nitrogen only, and the otherhistidines are fully protonated at both N. In both definitions, theremaining titratable residues are all charged, and one K� ion ispresent.

His355, which forms an H-bond with the D-propionate (D-prop) side chain of heme (see Fig. 1), is the only residue around theactive center that has been assigned a different protonation state byS/T (doubly protonated) and G/F (singly protonated; H-bond withD-prop is present). Previous Poisson–Boltzman electrostatic con-tinuum model calculations22 on the pentacoordinated ferricP450cam complex suggest a doubly protonated His355, while pKa

calculations on the pdb structure 1DZ9 with the recently developedempirical PROPKA method23 indicate that both protonation statesare possible for His355 (pKa 6.46). Asp297 is situated near theA-prop side chain of heme; both S/T and G/F assumed it to bedeprotonated, although it should be protonated according to pre-vious22 and present pKa calculations (PROPKA, 8.44). ProtonatedAsp297 proved to be necessary for obtaining geometries close tothe X-ray structure in a recent QM/MM study on resting state of

P450cam.24 Computed PROPKA pKa values of all titratable resi-dues are given in Supplementary Material. (PROPKA is a newlydeveloped efficient empirical method for the prediction of pKavalues of all ionizable residues in a protein. The method providespKa values with an overall RMSD of 0.79 from experiment. Theoverall accuracy is comparable to current state-of-the-art proteinpKa prediction methods, such as those based on Poisson–Boltz-mann electrostatics, but seems to make better predictions for Asp,Glu, and Cys residues with highly shifted pKa values. For detailssee: http://ghemical.chem.uiowa.edu.)

QM/MM Procedure

The general QM/MM methodology and protocol adopted in thisstudy were described extensively elsewhere8,14 and summarizedbriefly in the Introduction. In the following, we only address someaspects relevant to the present study.

The QM/MM calculations were performed with the ChemShellpackage17 using the TURBOMOLE program25 for unrestrictedopen-shell density functional theory (DFT) calculations with theB3LYP functional26 and the DL-POLY program27 for MM calcu-lations with the CHARMM22 force field.21 The HDLC opti-mizer28 in ChemShell was employed for geometry optimizations.A rational function optimizer with the Powell update (P-RFO) foran explicit Hessian28 was used in transition-state (TS) optimiza-tions. Full TS optimizations were only performed for the small QMregions R1 and R2 (definitions see below). For the other QMregions, the TS was taken to be the highest point on the potentialenergy profile along the OOH5exo reaction coordinate, which wasscanned with an increment of 0.01 Å. An electronic embeddingscheme29 was applied to include the polarizing effect of theenzyme environment on the QM region. Hydrogen link atoms30

were used at the QM/MM boundary along with the charge shiftmodel.31 No cutoffs were introduced for the nonbonding MM andQM/MM interactions.

To probe the effect of different QM/MM partitioning schemes,we employed a number of QM regions (see Fig. 2), which com-prised the following sets of atoms: (1) R1 (51 QM atoms): Iron–oxo–porphyrin (without heme side chains), C4HC5H2C6H2 part ofcamphor which corresponds to propane in gas-phase calculations,and sulfur atom of coordinating Cys357. (2) R1� (67 QM atoms):R1 region with full camphor. (3) R2 (84 QM atoms): Iron–oxo–porphyrin (without heme side chains), full camphor, Cys357, COgroup of Leu356, and NHOC�H unit of Leu358. The proximalligand in R2 will be called extended Cys. (4) Rm (70 QM atoms):R1� with methyl mercaptide (OSCH3) representation of proximalligand. (5) The p{s}[w] label after the name of a QM regionindicates the inclusion of propionate heme side chains {all hemeside chains} [w903] into the corresponding region, for example,R1p (67 QM atoms), R1 with propionate side chains of heme; R2s(120 QM atoms), R2 with all heme side chains; R2sw (123 QMatoms), R2s with w903, etc. (6) R2sE (170 QM atoms): R2s withw903, CH3NHC(NH2)2

� representing Arg299 and Arg112,CH3COOH representing Asp297, and methyl substituted imida-zole representing His355 (see Fig. 1). The effects of the protona-tion states of Asp297 and His355 in R2sE were also investigated.When included in the QM region, Asp297 and His355 wererepresented as follows: CH3COOH (protonated Asp297),

Figure 1. Active site of P450cam in Cpd I (pdb code: 1DZ9), abbre-viated as R2sE.

1326 Altun, Shaik, and Thiel • Vol. 27, No. 12 • Journal of Computational Chemistry


CH3COO� (deprotonated Asp297), methyl substituted imidazole(protonated His355) and methyl substituted imidazolate (deproto-nated His355).

Several basis sets were examined to check for consistency. Inthe case of the B1, B2, B2�, and Bg sets, the iron atom is describedby a small-core effective core potential (ECP) together with theassociated double-� quality LACVP basis.32 The label “*” afterthese abbreviations means that the LACVP* basis is used insteadof LACVP at iron, both having the same definition except for thecontraction coefficients in one shell. If W appears at the end of anabbreviation, the iron atom is described by Wachters all-electronbasis set augmented with diffuse d and a set of f polarizationfunctions in the contraction [8s7p4d1f],33 which has given con-verged results for the pentacoordinated ferric and ferrous P450cam

complexes.34 The other atoms are described as follows. (1) B1:6-31G35 on all other atoms. (2) B2: 6-31�G*35 on the six atomscoordinated to iron, the C5 and O atoms of camphor, the carbox-ylate (OCOO�) atoms of the propionates, and O atom of w903;6-31��G**35 on the H5exo atom of camphor and H atoms ofw903; 6-31G on the remaining atoms. (3) B2�: 6-31�G**35 onH5exo; otherwise as in B2. (3) Bg: 6-31G*35 on all atoms coordi-nated to iron, the C4, C5, and C6 atoms of camphor, the carboxylate

unit of the propionates, and the heavy atoms of proximal ligand;6-31G on all other atoms. (4) B�g: 6-31G* on all other atoms.

Throughout this study, the QM level employed in a particularQM/MM calculation is denoted in the form QM region/basis set,for example, R1/B1, R2s/B2, etc. The standard QM/MM optimi-zations involved all residues (including crystallographic and sol-vent waters) that have atoms within a distance of 2.4 Å around anyatom of the reactive center (defined to consist of the heme, coor-dinating Cys357 and camphor). Larger optimized regions (up to 8Å) were examined to assess the internal consistency of the results.

Results

Basis Set Dependence in the Gas Phase

The sensitivity of UB3LYP results with regard to basis set exten-sion has been studied in ref. 8 only for the H-abstraction barrier ofthe model system [FeO(SH)(NH3)4]� � CH4. As an additionalcheck, the barrier (TSH–RC gap) and endothermicity (HYD–RCgap) for the H-abstraction reaction were computed by optimizingthe R1 regions of RC, TSH, and HYD in the gas phase at the

Figure 2. Definition of QM regions.

TABLE 1. Basis Set Dependence of the UB3LYP Barrier/Endothermicity (kcal/mol) for the Isolated R1Model Optimized at UB3LYP/B1 Level.a

B1 B1W Bg Bg* BgW B2 B2� B2W

No. of basis functions 329 363 374 374 408 396 395 430Barrier/endothermicity 19.2/13.5 19.7/11.8 21.7/14.5 22.2/15.7 20.6/10.5 18.9/10.4 18.9/10.4 17.9/6.9

aThe lowest energy hydroxo intermediate has an Fe(IV) center in the gas phase. Thus, the computed endothermicityrefers to HYD(IV). The O—H5exo distance of TSH was kept at 1.20 Å during the R1/B1 optimization.



UB3LYP/B1 level, followed by single-point UB3LYP calculationswith different basis sets (see Table 1).

LACVP and LACVP* basis sets at iron give essentially thesame results (Bg vs. Bg*). The use of one set of diffuse functionsat H5exo seems appropriate (B2 vs. B2�). Compared to the preferredB2W basis set, the largest deviations are obtained with Bg-basedbasis sets, indicating that the inclusion of polarization functionswithout diffuse functions on the coordinated atoms gives unbal-anced results. Going from B1 to B2, the barrier/endothermicity islowered by ca. 0.3/3.1 kcal/mol. The use of Wachters all-electronbasis set at iron instead of LACVP does not have a significanteffect on the barrier, but affects the endothermicity by 2–4 kcal/mol. In an overall assessment, there is no obvious correlationbetween barrier and endothermicity; the endothermicity, with val-ues in the 7–16 kcal/mol range, is more sensitive to basis setimprovements than the barrier, which varies in the 18–22 kcal/molrange. To limit the computational effort for larger QM regions, theB1 and B2 basis sets will mostly be employed despite theirtendency to overestimate the endothermicity. Bg and Bg* will alsobe used to compare QM/MM results from ChemShell17 andQSite.16

Compatibility of ChemShell and QSite QM/MM Results

R2s/B'g* and R2sw/B�g* ROB3LYP/OPLS calculations yield abarrier/endothermicity of 11.7/4.5 kcal/mol and spin densities of0.20e/0.10e/0.12e on O2A atom of the A-prop heme side chain forRC/TSH/HYD.12,13 To investigate the compatibility of ChemShell(UB3LYP/CHARMM, link atom at QM/MM boundary withcharge shift model) and QSite (ROB3LYP/OPLS, frozen density atQM/MM boundary) results at given geometries, we performedsingle-point UB3LYP/CHARMM calculations on G/F geometrieswith various selections of QM regions and basis sets (see Table 2).

The largest B2 basis gives similar barriers as the smallest B1basis, whereas endothermicities are ca. 4 kcal/mol lower with B2

(see R1, R2, and R2s). Compared to the B2 basis set, Bg-basedbasis sets (Bg, Bg* and B�g*) overestimate the barrier/endother-micity by ca. 2/4 kcal/mol, consistent with the gas-phase results(see Table 1). LACVP and LACVP* at iron give almost the sameresults (Bg vs. Bg*). When 6-31G* is assigned for all atoms(B�g*) rather than only coordinating atoms, the Bg* results do notimprove, indicating that the basis set on the atoms outside thecoordination sphere is not critical.

Partial inclusion of camphor in a QM region underestimates therelative energies by ca. 2 kcal/mol in the enzyme (R1 vs. R1�). TheQM regions R1� and R2 as well as R1�p and R2s yield very similarresults, indicating that the smallest sulfhydryl (OSH�) represen-tation of the proximal ligand is appropriate for estimating therelative energies in QM/MM calculations. The inclusion of the sidechains in the QM region adds only 1–2 kcal/mol to relativeenergies (see R2/B1 vs. R2s/B1 and R2/B2 vs. R2s/B2), whichjustifies the cost-effective MM treatment of heme side chains inQM/MM energy calculations with a 2 kcal/mol target accuracy.

In agreement with previous results,12,13 the inclusion of w903into QM region R2s to yield R2sw does not change relativeenergies much (by less than 1 kcal/mol, see B�g* results). Thepublished12,13 B�g* barrier (11.7 kcal/mol) is reproduced by bothB1 and B2 basis sets (12.3 kcal/mol) for the analogous QM region(R2s). The published12,13 R2s/B�g* endothermicity (4.5 kcal/mol)is within 2 kcal/mol of the present R2s results (Table 2) and thusin reasonable agreement in view of technical differences (QM/MMmethodology, force field selection, chosen DFT approach, etc.)and the strong basis set dependence of the endothermicity (seeTables 1 and 2). In summary, with appropriate selections of QMregion and basis set, the QM/MM implementations in QSite andChemShell yield relative energies that agree within 2 kcal/mol ata given geometry.

The published12,13 B�g* spin density of 0.20e/0.10e/0.12e forRC/TSH/HYD is well reproduced with the B1 basis set and Bg-based basis sets (see Table 2). Thus, as long as the geometries areidentical, QSite and ChemShell give consistent results not only forthe relative energies but also for spin densities with analogousbasis sets. When the B1 and Bg basis sets are extended to B2, thespin density on the carboxylate unit of A-prop is reduced by ca.0.1e, and becomes quite small for RC (0.07e for R2s, 0.04e forR1�p) and almost vanishes for TSH and HYD (0.01e or less).Hence, the computed spin density on A-prop carboxylate dimin-ishes upon basis set improvement, as also noted in our previousQM/MM study on HRP Cpd I.19 The B1 and B2 barriers aresimilar in spite of a large A-prop carboxylate spin density differ-ence of 0.1e between B1 and B2, indicating that there is nocorrelation between the H-abstraction barrier and A-prop carbox-ylate spin density. The chosen size of the proximal ligand has nomarked effect on A-prop spin densities and relative energies inenzyme (see R1�p/B2 and R2s/B2).

To account for screening by solvent water molecules, G/Fneutralized a number of Arg, Lys, Glu, and Asp residues, whichwould normally be charged (see pKa values in SupplementaryMaterial). To check the effect of these nonstandard protonationstate assignments, we performed some R1/B1 single-point calcu-lations after adding H atoms to these residues in the G/F structures.When (a) only Arg and Lys, and (b) all Arg, Lys, Glu and Aspresidues are protonated, the barrier/endothermicity becomes (a)

TABLE 2. Single-Point UB3LYP/CHARMM Relative Energies (kcal/mol) and Mulliken Spin Densities � (e) on Carboxylate Unit of theA-Propionate Heme Side Chain for RC/TSH/HYD of G/F Structures.a

Level Energy (kcal/mol) �(A-prop COO�)

R1/B1 0.0/8.9/4.7 —R1/Bg 0.0/10.7/3.9 —R1/Bg* 0.0/10.8/4.1 —R1/B�g* 0.0/10.8/4.0 —R1/B2 0.0/7.7/0.3 —R1�/B1 0.0/11.4/6.2 —R2/B1 0.0/11.0/5.9 —R2/B2 0.0/10.1/1.6 —R2s/B1 0.0/12.3/6.2 0.17/0.09/0.12R2s/Bg 0.0/14.1/6.2 0.20/0.12/0.11R2s/Bg* 0.0/14.2/6.4 0.20/0.12/0.11R2s/B�g* 0.0/15.9/6.8 0.22/0.12/0.14R2sw/B�g* 0.0/15.0/6.3 0.22/0.12/0.15R2s/B2 0.0/12.3/2.4 0.07/0.01/0.01R1�p/B2 0.0/12.8/3.0 0.04/0.00/0.01

aNo spin density on D-propionate side chain of heme.



8.1/3.8 kcal/mol and (b) 10.0/3.4 kcal/mol. Compared with theR1/B1 result at the original G/F geometry (8.9/4.2 kcal/mol), theprotonation of these residues obviously has no significant effect onthe computed energetics.

QM/MM Geometry Optimizations on G/F Structures

We performed several R1/B1 UB3LYP/CHARMM optimizationson the G/F structures to check how the single-point R1/B1 barrier/endothermicity of 8.9/4.7 kcal/mol at the original G/F geometry(see Table 2) changes upon optimization. When individualQM/MM energy minimizations are done for each of the G/Fgeometries (fixed OOH5exo distance for TSH) by relaxing the QMregion and all residues and waters that have atoms within 2.4 (4.0)[8.0] Å around the reactive center, the computed barrier/endother-micity becomes 11.7/7.6 (11.8/8.0) [9.5/8.3] kcal/mol. Thus, rel-ative energies are shifted by typically 3 kcal/mol when each of theoriginal G/F geometries is reoptimized by UB3LYP/CHARMMcalculations.

In the next step, we started from the UB3LYP/CHARMM-optimized RC (see above), moved the migrating H5exo atom man-ually to positions typically occupied in TSH and HYD, respec-tively, and optimized the resulting geometries at the R1/B1 level:this led to a computed barrier/endothermicity of 17.4/10.4 kcal/mol (using an optimized region of 2.4 Å). Analogous R1/B1results for the barrier/endothermicity were found with this proce-dure when the initial structure is derived from the UB3LYP/CHARMM-optimized TSH (16.2/9.9 kcal/mol) or HYD (17.1/10.3kcal/mol) rather than from RC. As most of the protein atoms arefixed during the present UB3LYP/CHARMM optimizations, buthave previously been minimized using the OPLS force field, it maybe more appropriate to relax the G/F structures by CHARMM first.When CHARMM minimizations with fixed coordinates of thereactive center are followed by full UB3LYP/CHARMM optimi-zations for TSH, which is then relaxed to find RC and HYD, thebarrier/endothermicity is computed to be 15.8/10.3 kcal/mol atR1/B1 and 15.6/6.1 kcal/mol at R1/B2, which agrees with thecorresponding results without initial MM relaxation (R1/B1: 16.2/9.9 kcal/mol, see above). Fixing water molecules during theCHARMM minimizations as in the G/F study12,13 does not haveany major effect on the barrier/endothermicity (R1/B1: 15.2/9.5kcal/mol).

The optimizations described in the preceding paragraph thusyield barriers and endothermicities that are consistently higherthan those obtained from single-point calculations at the G/Fgeometries, by ca. 7 and 5 kcal/mol, respectively. What is theorigin of these differences? In an attempt to answer this question,we focus on the barrier and compare the geometries for RC andTS

Hobtained from G/F (R1/B1 barrier of 8.9 kcal/mol) and from

the R1/B1 optimizations described above, starting from RC (labelA, barrier of 17.4 kcal/mol) and from TSH (label B, barrier of 16.2kcal/mol). At first sight, these geometries (G/F vs. A vs. B) seemto be quite similar, especially with regard to the reactive center andthe position of the migrating H5exo atom. Previous experienceindicates that it is essential to have consistent hydrogen bondingnetworks in reactant and transition state, and that barriers may beaffected by changes in hydrogen bonding during the reaction.Close to the reactive center, there are two important hydrogen

bonds (see below), which show analogous changes in all three setsof geometries (G/F, A, and B) when going from RC to TSH: theO � � � H distance decreases by 0.06–0.07 Å in the hydrogen bondbetween w903 and the oxygen atom of FeO, while it increases by0.05–0.07 Å between camphor and Tyr96 due to the motion ofcamphor towards the heme. In the case of the reoptimizedUB3LYP/CHARMM geometries (A and B) these are the onlychanges in hydrogen bond distances (RC vs TSH) that exceed 0.05Å. By contrast, there are 14 other such changes in the G/Fgeometries, both in the vicinity of the reactive center and far away.In 11 of these cases, the hydrogen bond is shorter (stronger) in TSH

than in RC. The most striking example is the hydrogen bondbetween w62 and the O2A atom of the A-propionate side chain ofthe heme where the O � � � H distance decreases from 2.16 Å (RC)to 1.84 Å (TSH) in the G/F geometries while it remains almostunchanged in the UB3LYP/CHARMM geometries (A: 1.71 vs.1.69 Å; B: 1.74 vs. 1.72 Å); it should be noted that the reductionfrom 2.16 Å (G/F) to 1.71 Å (A) in RC occurs upon UB3LYP/CHARMM reoptimization of RC and is thus expected to contrib-ute significantly to the observed increase of the barrier by ca. 3kcal/mol in the corresponding calculations (see first paragraph ofthis section). For the sake of brevity, we do not discuss the other10 cases encountered in the G/F geometries where the hydrogenbond is shortened by more than 0.05 Å when going from RC toTSH (7 of them near the surface); it seems plausible, however, thatthese changes will combine to produce an overall (artificial) tran-sition state stabilization and hence a lowering of the barrier.

The procedure adopted for the UB3LYP/CHARMM optimiza-tions (see above) ensures that the stationary points (RC, TSH, andHYD) are connected by a continuous path on the potential energysurface (PES) and that the protein environment remains consistentduring the reaction. We have confirmed that RC and TSH shareidentical networks of hydrogen bonds X � � � HOY (X � N or O,Y � N or O, X � � � H distance between 1.2 and 2.5 Å, X � � � HOYangle greater than 150°) in the UB3LYP/CHARMM geometries:there are 926 such hydrogen bonds overall with A. To a largeextent, this is also true for the G/F geometries (929/927 suchhydrogen bonds for RC/TSH). The main distinction between thethree sets of geometries (G/F, A, and B) is thus not the integrity ofthe overall hydrogen bonding network, but the strength of variousindividual hydrogen bonds. Judging from the geometries (A andB), the UB3LYP/CHARMM optimizations generally yield hydro-gen bonds of similar strength in RC and TSH. By contrast, the G/Fgeometries feature a number of hydrogen bonds (not directlyrelated to the reaction, see above) that appear to be of differentstrength in RC and TSH. In our experience, this may be due toincomplete optimization and/or convergence to unconnected localminima, which are common dangers in the work on complexmolecules. On the basis of the available evidence, we believe thatthe G/F geometries12,13 suffer from such problems. When the G/Fgeometries are reoptimized along pathways with a consistentprotein environment, the computed barrier/endothermicity in-creases by ca. 7/5 kcal/mol, thus accounting for half of the dis-crepancies in the published QM/MM results.8,12,13

To probe the effects of the external solvent layer that is presentonly in the S/T setup, a water layer of 16 Å thickness wasconstructed around the enzyme segment of the original unsolvatedneutral TSH structure of G/F as described previously.14,34 The



resulting structure was optimized by CHARMM keeping the co-ordinates of the reactive center and the outer 8 Å solvent layerfixed. Following the same procedure as before (see above), anR1/B1 QM/MM optimization was then performed for the TSstructure with fixed OOH5exo distance, and subsequent energyminimization starting from the optimized TS geometry led to RCand HYD. The resulting barrier/endothermicity of 14.9/9.6 kcal/mol is close to the unsolvated case (see above). Thus, the solventlayer has no significant influence on the relative energies com-puted.

QM and QM/MM Results on SnapX

To assess the effect of the chosen initial geometry (MD snapshotsvs. X-ray coordinates) on the energetics, the initially preparedsnapX starting structure (see Computational Details) was opti-mized at the R1/B1 UB3LYP/CHARMM level and then subjectedto an R1/B1 QM/MM-PES scan along the OOH5exo distance. Theoptimized R1/B1 stationary points were refined at the R1/B2 level.Transition structures were located at both R1/B1 and R1/B2 levelsfor the systems with protonated and deprotonated Asp297 (seeTable 3). The other entries in Table 3 correspond to single-pointresults at R1/B2 geometries.20

The computed R1/B1 (16.7/8.3 kcal/mol) and R1/B2 (16.3/4.1kcal/mol) energies for snapX are very similar to those obtained byrescanning the OOH5exo distance of the G/F structure (see above,R1/B1: 16/10 kcal/mol, R1/B2: 16/6 kcal/mol). However, they areca. 5 kcal/mol lower than those obtained previously for the snap-shots from an MD simulation.8,14

The results reported in the following sections refer to the snapXsystem.

Catalytic Role of Water 903

In the previous S/T study,8,14 w903 (see Fig. 1) situated near theoxo ligand in the X-ray structure (1DZ9) moved to the cavityaround camphor during the dynamics runs of the setup, while it isplaced close to the oxo ligand in snapX and in the G/F study.12,13

To see whether this is the source of 5 kcal/mol energy differencebetween the results on published S/T snapshots8,14 and snapX, weremoved w903 from snapX (deprotonated Asp297) and reopti-mized the structures, which led to an R1/B1 barrier/endothermicityof 20.8/13.9 kcal/mol. These values are similar to the publishedR1/B1 values of S/T (20.6/14.9 kcal/mol).8 Hence, the presence of

w903 lowers the barrier/endothermicity by ca. 4/6 kcal/mol be-cause of its interactions with the oxo ligand.

R1w/B2 calculations on the system with protonated Asp297yield a barrier/endothermicity of 17.2/6.4 kcal/mol, which matchesthe R1/B2 result of 16.7/6.1 kcal/mol quite well. Thus, inclusion ofw903 into the QM region does not have any notable effect on theenergetics in agreement with previous results.12,13 This suggeststhat the stabilization due to w903 has mainly electrostatic origin.

R1�/B1 (20.7/15.9 kcal/mol) and R1�w/B1 (17.7/10.6 kcal/mol)single-point gas-phase UB3LYP calculations on the structuresextracted from R1/B2-optimized UB3LYP/CHARMM geometriesof snapX with protonated Asp297 (see below for the effect ofAsp297 protonation state) yield differential barrier/endothermicitylowerings of 3.0/5.2 kcal/mol by w903, similar to the proteinQM/MM values of ca. 4/6 kcal/mol. Thus, gas-phase energies canbe used to analyze the source of stabilization by w903 in enzyme.Interaction energies between w903 and R1� in the gas phase arecomputed to be �8.9/�11.9/�14.2 kcal/mol for RC/TSH/HYD atthe B3LYP/B1 level without counterpoise corrections. Estimatingthe classical electrostatic interaction energies between each atomof w903 and the oxo ligand as well as the migrating H5exo atomwith the use of Mulliken charges yields similar values of �8.6/�10.9/�14.5 kcal/mol for RC/TSH/HYD. The stabilization byw903 thus comes from favorable electrostatic interactions in H-bonds that are stronger in TSH and HYD than in RC as a result ofincreasing negative charge at the oxo ligand with decreasingOOH5exo distance. w903 is almost neutral overall and should thusexhibit relatively small long-range electrostatic interactions.

Protonation State of Asp297 and His355

The H-abstraction barrier is rather insensitive to the protonationstate of Asp297, whereas the endothermicity increases by typically2 kcal/mol when Asp297 is protonated (see Table 3). The A-propenvironment is quite different in systems with protonated anddeprotonated Asp297 (Asp297 in the MM region, selected dis-tances see Table 4). In the case of deprotonated Asp297, the planeof the carboxylate group of A-prop rotates by ca. 45° and Arg299is aligned such that it screens Asp297 rather than the O2A atom ofA-prop. This type of A-prop environment is not present in any ofthe available P450cam crystal structures. In the snapX system withprotonated Asp297, the A-prop environment is compatible with theX-ray structure. It has been shown previously for the restingstate24 and the pentacoordinated complexes34 of P450cam that all

TABLE 3. R1/B1 and R1/B2 UB3LYP/CHARMM Barrier/Endothermicity (kcal/mol) on SnapX withProtonated and Deprotonated Asp297, and Single-Point Results on R1/B2-Optimized Structuresfor Larger QM Regions.

Asp297 R1/B1 R1/B2 R1p/B2 R1�/B2 R1�p/B2 R2/B2 R2s/B1 R2s/Bg* R2s/B2

Deprotonateda 16.7/8.3 16.3/4.1 16.9/4.4 19.0/4.6 19.9/4.9 18.8/4.6 19.0/8.8 22.3/8.6 19.7/5.0Protonated 16.9/10.1 16.7/6.1 16.9/6.1 19.3/6.8 19.7/6.9 18.7/6.5 18.1/9.7 21.0/9.8 19.1/6.4

aCarboxylate unit of A-prop carries spin density of 0.08e/0.06e/0.06e (R2s/B1) and of 0.13e/0.08e/0.11e (R2s/Bg*) forRC/TSH/HYD in the system with deprotonated Asp297. In all other cases, there is no spin density on propionates.



interactions around propionate side chains of heme remain intactduring MD simulations with protonated Asp297.

The local A-prop environment tunes the spin density distribu-tion in the heme side chains. No A-prop spin density is found whenArg299 screens the O2A atom of A-prop (see first column of Table4) as in the X-ray structure. This is the case for snapX withprotonated Asp297, but also for the UB3LYP/CHARMM-opti-mized G/F and snap40 structures with deprotonated Asp297. Theoriginal G/F geometries12,13 with deprotonated Asp297 show somespin density on A-prop (see Table 3), which vanishes uponUB3LYP/CHARMM reoptimization due to better screening ofO2A by Arg299 (see Table 4). The optimized snapX structureswith deprotonated Asp297 also lack this screening and thus havesome spin density on the A-prop carboxylate in RC/TSH/HYDwhen using small basis sets (0.08e/0.06e/0.06e for R2s/B1 and0.13e/0.08e/0.06e for R2s/Bg*), which, however, disappears uponbasis set improvement to R2s/B2 (consistent with similar findingsfor the original G/F geometries, see Table 2). The computedA-prop spin densities are thus sensitive both to the local environ-ment and the chosen basis set. They tend to vanish upon A-propscreening and for larger basis sets.

To check whether the protonation state of His355 (G/F, singlyprotonated; S/T, doubly protonated) affects energetics and propi-onate spin densities, we have performed a series of UB3LYP/CHARMM single-point calculations on the R1/B2 geometries ofsnapX by removing the H atom at the � nitrogen of His355. ForsnapX with deprotonated Asp297, barriers/endothermicities withdeprotonated (protonated) His355 are 17.4/5.0 kcal/mol (16.9/4.4kcal/mol) at R1p/B2 and 20.9/5.8 kcal/mol (19.7/5.0 kcal/mol) atR2s/B2. For snapX with protonated Asp297, they are 17.3/6.1kcal/mol (16.9/6.1 kcal/mol) at R1p/B2 and 19.5/6.3 kcal/mol(19.1/6.4 kcal/mol) at R2s/B2. Thus, barriers/endothermicities arealmost independent of the protonation states of His355 andAsp297 (both in MM region). However, A-prop carboxylate spindensities of RC/TSH/HYD for the system with deprotonatedAsp297 are affected by the protonation state of His355: zero withprotonated His355 (B2 basis); 0.11e/0.08e/0.08e (R1p/B2) and

0.19e/0.16e/0.16e (R2s/B2) with deprotonated His355 due to de-creased electron density on propionates. In the latter case (depro-tonated Asp297 and His355, both in MM region) there is Coulom-bic repulsion between His355 and the propionates (especially thenearby D-propionate), which results in some electron transfer tothe porphyrin and the appearance of some spin density at the lessscreened A-propionate. By contrast, there is no spin density on thepropionates for the snapX system with protonated Asp297, regard-less of the protonation state of His355. In summary, spin densitymay appear in the cases where carboxylates are only partiallyscreened, but there is no significant effect (typically less than 1kcal/mol) on the energetics.

Sensitivity of the Energetics to QM/MM Options

As already documented for the original G/F geometry12,13 (seeTable 1 and associate discussions), the choice of the QM regioninfluences the QM/MM results to some extent. The relative ener-gies for different QM regions and basis sets obtained with snapX(see Table 3) generally follow the same trend as those for the G/Fsystem. The inclusion of propionates into QM regions adds lessthan 1 kcal/mol to the barrier (see R1 vs. R1p, R1� vs. R1�p, andR2 vs. R2s). The inclusion of vinyl and methyl substituents on theheme into the QM region does not have any effect on the barrier/endothermicity (snapX with protonated Asp297: R2p/B2, 19.1/6.6kcal/mol; R2s/B2, 19.1/6.4 kcal/mol). When enlarging the proxi-mal ligand size from sulfhydryl to extended Cys representation(R1� 3 R2 or R1�p 3 R2s), the relative energies remain essen-tially the same. The methyl-mercaptide representation (OSCH3

�)is known to give artificially large state splittings in the gas phasedue to the strong electron-donating capability of the methylgroup2,36–38 but the polarizing effect of the enzyme environmentcompensates for the strong pushing effect of the methyl group.14

Rmp/B2//R1/B2 QM/MM calculations with protonated Asp297give a barrier/endothermicity of 19.6/6.9 kcal/mol, which is almostthe same as for R1�p/B2 (19.3/6.8 kcal/mol) and R2s/B2 (19.1/6.4kcal/mol). Thus, the methyl-mercaptide representation is appropri-

TABLE 4. Selected Distances (Å) Involving Environmental Residues in the Reactants.

SystemA-prop-Arg299O2A. . .N(H)a

Asp297-Arg299O. . .N(H)a

A-prop-Asp297O2A. . .Ob

A-prop-w62O2A. . .O(H1)

Thr101-w62O. . .O(H2)

X-ray (chain A)c 3.02(2.05) 4.11(3.42) 2.36(3.12) 3.12(2.14) 2.71(1.88)X-ray (chain B)c 3.00(2.06) 4.11(3.47) 2.73(3.28) 3.03(2.12) 2.75(1.78)G/F (original)d 3.45(2.95) 2.60(1.64) 3.59(3.64) 3.14(2.16) 2.76(1.77)G/F (optimized)e 2.72(1.73) 2.70(1.72) 4.38(4.09) 2.66(1.67) 2.78(1.93)SnapX (deprot. Asp297)e 3.86(3.50) 2.79(1.96) 4.75(4.77) 2.56(1.57) 3.03(2.34)SnapX (prot. Asp297)f 2.70(1.70) 3.61(3.15) 2.58(3.35) 3.21(2.28) 2.70(1.76)Snap40e 2.63(1.61) 2.72(1.83) 3.33(4.37) 2.73(1.75) 5.64(6.35)

Data with large deviations from X-ray values are shown in boldface.aSmallest distance between an NH2 unit of Arg299 and O2A or Asp297.bOne of the distances between terminal oxygens of Asp297 and O2A is given in parenthesis.cThe positions of H atoms were obtained with CHARMM minimization (pdb code: 1DZ9).dROB3LYP/OPLS Rg/Bg* geometry (deprotonated Asp297).eUB3LYP/CHARMM R1/B1 geometry (deprotonated Asp297).fUB3LYP/CHARMM R1/B1 geometry (protonated Asp297).



ate for calculating QM/MM reaction energies in the protein envi-ronment. Partial inclusion of camphor in the QM region (R1 andR1p) leads to barriers that are ca. 3 kcal/mol lower than thoseobtained with full camphor in the QM region (R1� and R1�p). Thisdifference may be due to the H-bond between camphor and Tyr96that is stretched by ca. 0.05 Å in TSH compared to RC or HYD: theenergetic consequences of this stretching may be different depend-ing on whether the H-bond is completely in the MM region (R1and R1p) or not (R1� and R1�p). When Tyr96 is included in theR1� region as p-hydroxo toluene (snapX with protonated Asp297),both single-point calculations at the R1/B2 geometry (19.4/11.6kcal/mol) and geometry relaxations (18.6/10.7 kcal/mol, optimizedregion: 4 Å around the reactive center) with the B1 basis reproducethe R1�/B1//R1/B2 result of 18.4/10.3 kcal/mol, indicating thatthere is no significant charge transfer involving Tyr96 and nomajor structural reorganization. Concerning basis set improve-ment (B13 B2) for a given selection of the QM region (see Table3), the computed barriers are not affected much while the endo-thermicities decrease by ca. 4 kcal/mol (results on snapX withprotonated Asp297: R1, 16.9/10.1 kcal/mol 3 16.7/6.1 kcal/mol;R2s, 18.1/9.7 kcal/mol 3 19.1/6.4 kcal/mol).

The values in Table 3 for large QM regions refer to single-pointcalculations at R1/B2 geometries obtained by relaxing only theresidues that have atoms within 2.4 Å of the reactive center. To seeif the single-point results change with geometry optimizations, weoptimized R1/B2 geometries of snapX with protonated Asp297 atR1�/B2 and R2s/B2 (fixed OOH5exo distance for TSH) using alarge optimized region of 8 Å around the reactive center with bothprotonation states of His355. Barriers/endothermicities are as fol-lows: R1�/B2 with protonated (deprotonated) His355, 20.4/7.9(20.8/4.4) kcal/mol; R2s/B2 with protonated (deprotonated)His355, 21.0/7.7 (21.3/7.6) kcal/mol. The present optimizationsthus reproduce the single-point results at the geometries obtainedusing smaller QM and optimized regions (within 1 kcal/mol).Hence, the cost-effective optimized region of 2.4 Å around thereactive center seems reliable enough for the current system. Thisis confirmed by R2s/B2 geometry optimizations (protonatedHis355) with an optimized region of 2.4 Å around the reactivecenter (20.2/7.4 kcal/mol), whose results are again within 1 kcal/mol of those for an 8 Å optimized region (21.0/7.7 kcal/mol, seeabove).

Another concern of the current system setup is the presence ofcharged surface residues that yield a net charge of �9e for snapXwith protonated Asp297. During QM/MM calculations on thepentacoordinated P450cam complexes,34 we showed that redoxenergies could only be computed properly39 when the MM region(including propionates) has no net charge. For the current snapXsystem, there is no net charge on the QM regions (excludingpropionates which are screened by the environmental residues),and thus there should be no artificial long-range electrostaticinteractions between QM and MM regions. To check the effect ofnet charge, snapX with protonated Asp297 was neutralized asfollows: Asp 304 and 407 as well as Glu 306 and 329 wereprotonated; His 17, 347, and 391 were kept as in snapX and theother histidines were doubly protonated. CHARMM minimizationfor RC was performed keeping the reactive center and the outer 8Å solvent layer fixed. After a subsequent R1/B1 QM/MM mini-mization of RC, an approximate TSH was located (OOH5exo

distance: 1.189 Å at the highest point of PES) and then relaxed tofind HYD. The computed barrier/endothermicity (17.1/10.0 kcal/mol) is analogous to that obtained for the charged system (16.9/10.1 kcal/mol). Thus, the reaction energies with an uncharged QMregion do not suffer from any net charges due to surface residues.This conclusion is consistent with the single-point results at theoriginal G/F geometries (deprotonated Asp297, see above).

Extended QM Regions Including Environmental Residues

As the CHARMM force field evaluates the electrostatics fromfixed atomic charges, the calculations with the QM regions pre-sented so far cannot account for any charge transfer involving theenvironmental residues or for their polarization. Such effects maybe studied by incorporating the relevant residues into the QMregion. Therefore, we performed a number of single-point B1calculations at R1/B2-optimized geometries of snapX (protonatedHis355) extending the R2s QM region to include deprotonated(protonated) Asp297 [denoted as R2sAsp (R2sAspp)]. The B1barrier/endothermicity is computed to be 18.2/9.8 kcal/mol withprotonated Asp297 and 19.4/9.1 kcal/mol with deprotonatedAsp297 in the QM region. These values are quite similar to eachother and consistent with the results obtained without havingAsp297 in the QM region (R2s/B1: 18.1/9.7 kcal/mol for proton-ated Asp297 and 19.0/8.8 kcal/mol for deprotonated Asp297; seeTable 3). The system with protonated Asp297 (R2sAspp/B1) doesnot have any spin density on the propionates and Asp297. How-ever, for the system with deprotonated Asp297 (R2sAsp/B1), thereis some spin density on A-prop carboxylate (0.04e/0.03e/0.03e,analogous to R2s/B1 result), but much more on Asp297 (0.23e/0.21e/0.21e) for RC/TSH/HYD, arising from an electron transfer ofca. �0.25e from deprotonated Asp297 to the cationic porphyrinring in each case. Hence, if Asp297 is deprotonated, then it isnecessary to include it in any QM region that contains the sidechain propionates to account for such charge transfer.

R2sAsp/B1 geometry optimizations (optimized region: 8 Åaround reactive center; OOH5exo distance refinement with 0.01 Åincrements for TSH) result in barriers/endothermicities of 20.0/11.5 kcal/mol with protonated His355 and of 23.6/13.3 kcal/molwith deprotonated His355. Thus, the inclusion of deprotonatedAsp297 in the R2s region changes the R2s/B1//R1/B2 barrier onlyby 1 kcal/mol with protonated His355 (barrier/endothermicity:19.0/8.8 kcal/mol). Furthermore, for protonated His355, the A-prop environment does not change upon inclusion of deprotonatedAsp297 in the QM region, that is, A-prop remains twisted, Asp297does not screen the O2A atom of A-prop (minimum distance of ca.3.65 Å between O2A atom and carboxylate O atoms of Asp297),propionate side chains do not have any unpaired spin density, andthe spin density on Asp297 is 0.43e/0.34e/0.36e for RC/TSH/HYD.

R2sAsp/B1 optimizations with deprotonated Asp297 and His355yield structures40 with surprisingly short O2AOO[Asp297] distancesof 2.20(2.88)/2.20(2.90)/2.20(2.88) Å for RC/TSH/HYD, which aresmaller than the X-ray (1DZ9) values of 2.36(3.12) Å for chain A,and much smaller than those of 2.73(3.28) Å for chain B (seeTable 4); the crystallographic B values indicate that Asp297 isbetter characterized for chain B. A minimum reactant structure ofthis kind was first found by V. Guallar, who communicated hisresults to us prior to publication.40 In the optimized R2sAsp/B1



structures with short O2AOO distances, two unpaired electronsare located on the FeO unit and the third one is shared between thecarboxylates of A-prop and Asp297, yielding spin densities of0.45e on A-prop carboxylate and of 0.55e on Asp297 for allstationary points. As a result, this species has an almost closed-shell porphyrin with spin densities of �0.08e/�0.06e/�0.07e.Single-point calculations with deprotonated His355 included in theR2sAsp QM region yield analogous relative energies (23.0/13.4kcal/mol) and spin densities on porphyrin (�0.08e/�0.06e/�0.06e), Asp297 (0.55e) and A-prop carboxylate (0.45e). Geom-etry optimizations with an optimized region of 8 Å around thereactive center do not change the single-point energies (23.6/13.3kcal/mol) and spin densities on porphyrin (�0.08e/�0.06e/�0.07e), Asp297 (0.54e), and A-prop carboxylate (0.45e). No spindensity appears on deprotonated His355. In terms of electronicstructure, the species with short O2AOO distances is actually CpdII with an additional unpaired electron in a �* orbital delocalizedover A-prop [O2A] and Asp297 [O] (weak ferromagnetic couplingto yield a quartet state). The H-abstraction barrier of these speciesis quite high (see above), that is, 3.6 kcal/mol higher than in theusual open-shell porphyrin case that is encountered with proton-ated His355.

To see whether these findings persist in gas-phase model cal-culations, we isolated R2s, R2sAsp, and R2sE (as in Fig. 1 butwith deprotonated Asp297) regions from the R2sAsp/B1-opti-mized snapX structures both for protonated and deprotonatedHis355. Single-point R2s, R2sAsp and R2sE calculations with theB1 basis set yield barriers/endothermicities of 23.2/18.0 (21.8/18.3), 23.5/17.3 (19.9/16.3), and 21.2/13.2 (17.7/11.4) kcal/molfor deprotonated (protonated) His355. Hence, R2s calculationsgive almost the same relative energies in both cases; inclusion ofAsp297 in the model has almost no effect for deprotonated His355,but lowers the relative energies by ca. 2 kcal/mol for protonatedHis355; enlarging the model to include also His355, Arg112,Arg299, and w903 lowers the barrier/endothermicity further by ca.2/5 kcal/mol in both cases; the R2sAsp and R2sE gas-phase modelcalculations yield barriers for deprotonated His355 (shortO2AOO distance) that are 3.6 kcal/mol higher than for proton-ated His355, as in the enzyme case (see above). Without anyscreening residues around propionates (R2s), one unpaired elec-tron is mainly localized on the A-prop and D-prop carboxylatesrather than in an a2u-like porphyrin orbital (porphyrin spin density:�0.07e). With the inclusion of Asp297 in the model, porphyrinspin density does not change much (less than 0.01e) but someamount of propionate spin density is transferred from the propi-onates to Asp297 carboxylate. For the species with shortO2AOO(Asp297) distances (deprotonated His355), D-prop has nospin density and one unpaired electron is mainly located in the�*(OOO) orbital of A-prop (0.61e) and Asp297 (0.39e), analo-gous to the enzyme case. For the protonated His355 case, R2sAspcalculations yield some spin densities on D-prop (ca. 0.23e) inaddition to A-prop (ca. 0.55e) and Asp297 (ca. 0.20e), but as soonas screening residues are included in the model (R2sE with pro-tonated His355), all spin density on propionate heme side chains ofRC/TSH/HYD is transferred to Asp297 (0.49e/0.42e/0.43e) andporphyrin (0.21e/0.28e/0.28e), analogous to the enzyme case (seeabove). For the deprotonated His355 case, gas-phase R2sE calcu-lations again give an electronic structure consistent with the en-

zyme (spin densities: 0.38e on A-prop carboxylate, 0.62e onAsp297, and small on porphyrin for each stationary point). Insummary, R2sE single-point calculations at the enzyme geometryreproduce all features of the full QM/MM calculations on theH-abstraction reaction, both with regard to energies and elec-tronic structure.

R2sAsp gas-phase optimizations for the species with shortO2AOO(Asp297) distances (OOH5exo distance refinement withan increment of 0.01 Å) yield analogous relative energies (ca. 1kcal/mol variation) and spin densities compared with single-pointQM calculations at the enzyme geometry: barrier/endothermicity,24.6/16.3 kcal/mol; spin densities, ca. �0.07e on porphyrin, 0.65e/0.61e/0.66e on carboxylate A-prop and 0.34e/0.38e/0.33e onAsp297. However, the proximal ligand is significantly reorganizedas noted in the previous studies,14,34 and the O2AOO(Asp297)distances of 2.30(3.34)/2.24(3.23)/2.31(3.34) Å are elongatedcompared with the QM/MM enzyme values of 2.20(2.88)/2.20(2.90)/2.20(2.88) Å (see above). Gas-phase R2sAsp calcula-tions with long O2AOO(Asp297) distances suffer from the freerotation of deprotonated Asp297.

To check for charge transfer effects involving His355, weperformed some QM/MM single-point B1 calculations includingHis355 into R2sAspp region, at the R1/B2 geometries of snapXwith protonated Asp297. The barriers/endothermicities are 18.3/9.8 kcal/mol for protonated His355 and 19.7/10.8 kcal/mol fordeprotonated His355, indicating that relative energies are again notaffected much by the protonation state of His355 (typically ca. 1kcal/mol) when His355 is included in the QM region. However,the electronic structures differ. For protonated His355, there is nospin density on propionates, His355 and Asp297 (spin density onporphyrin: 0.72e/0.72e/0.79e), while for deprotonated His355,there is some electron transfer, mainly from His355 to porphyrin,yielding spin densities of 0.40e/0.38e/0.38e on His355 and of0.39e/0.42e/0.47e on porphyrin (no spin density on propionatesand Asp297). Geometry optimizations with protonated Asp297and protonated (deprotonated) His355 do not change the single-point results much: barrier/endothermicity, 20.1/11.4 (21.5/11.9)kcal/mol; no spin density on propionates, Asp297 or His355, and0.71e/0.65e/0.78e on porphyrin (spin density of 0.40e/0.36e/0.36eon His355, 0.39e/0.41e/0.48e on porphyrin and none on propi-onates or Asp297). When His355 (deprotonated or protonated) isincluded in the QM region for snapX with deprotonated Asp297,neither the relative energies nor the electronic structures change(see Supplementary Material).

In another test calculation, we manually transferred the H-bonding hydrogen atoms from protonated Asp297 and doublyprotonated His355 to the O2A and O2D atoms of propionates. Thehydrogens returned to their original positions during QM/MMoptimization. Thus, H-transfer from environmental residues topropionates is not possible in the native enzyme environment.

Further Gas-Phase Model Calculations

To assess intrinsic QM effects, single-point UB3LYP calculationswere performed on various regions of the snapX system withprotonated Asp297 optimized at the R1/B2 level (see Table 5) andat the G/F geometries12,13 (see Supplementary Material). In thefollowing, we mainly focus on the results for snapX. The w and s



labels in Table 5 correspond to inclusion of w903 and heme sidechains in R1�, Rm and R2 regions, whose proximal ligand repre-sentations are OSH, OSCH3 and extended Cys, respectively.

All choices of proximal ligand yield very similar relative en-ergies in the gas phase. However, the spin density ratio of proximalligand/porphyrin (R1�, 4/6; Rm, 7/3; R2, 6/4) changes with theproximal ligand representation in the gas phase, contrary to theenzyme (always ca. 2/8). The gas-phase spin densities reproducethose in the enzyme best when the smallest sulfhydryl represen-tation is used.

R1/B1 (20.6/15.8 kcal/mol) and R1�/B1 (20.6/15.9 kcal/mol)results are the same in the gas-phase due to the absence of Tyr96.The presence of w903 lowers the barrier/endothermicity by 3–4/5–6 kcal/mol in the absence of propionates. Basis set improvementreduces the endothermicity by 2–4 kcal/mol. The calculations withpropionates in the absence of screening residues yield 1–2 kcal/mol higher relative energies compared with the models that do notcontain heme side chains. When the screening residues are in-cluded in the model (R2sE), the results of reduced models that donot contain heme side chains are reproduced as long as there isproper screening of the propionates by the neighboring residues,for example, snapX with protonated Asp297 (R2w/B1: 16.6/11.0kcal/mol vs. R2sE/B1: 17.4/10.7 kcal/mol). R2sE relative energiesare generally 1–2 kcal/mol lower than R2sw energies. Gas-phaseresults for models that contain propionate side chains withoutscreening residues cannot be compared with the protein QM/MMresults because the porphyrin spin density drifts to the propionateside chains in these models, yielding a Cpd II-type electronicstructure with an additional unpaired electron at the propionates.Hence, models for gas-phase calculations should either not containpropionate side chains or contain both propionate side chains andenvironmental screening residues.

In the presence of screening residues (R2sE) in the gas phase,there is no spin density on propionates and no charge transfer frompropionates during the reaction for snapX with protonated Asp297.In this case, the protonation state of His355 does not affect thebarrier/endothermicity (R2sE with protonated His355: 17.4/10.7kcal/mol; R2sE with deprotonated His355 [R2sEhsd]: 17.5/10.5kcal/mol) and propionates do not possess any spin density. Whenany of the screening residues are removed from the model, spindensity accumulates on the neighboring propionate side chain due

to unbalanced electrostatic interactions. For example, whenHis355 is removed from R2sE (R2sEnohis), D-prop carries sig-nificant spin densities of 0.14e/0.10e/0.11e with B1, but the bar-rier/endothermicity of R2sEnohis (17.8/10.8 kcal/mol) is almostthe same as in the largest models that do not have any spin densityon propionates (R2sE: 17.4/10.7 kcal/mol; R2sEhsd: 17.5/10.5kcal/mol). This emphasizes again that neither the amount nor thechange of propionate spin density during the reaction affect therelative energies much and the presence of His355 is not essentialfor an efficient catalysis.

In R2sE and R2sEhsd gas-phase calculations on snapX withprotonated Asp297, there is no spin density on the environmentalresidues. When Asp297 is deprotonated (R2sEasp), one electron istransferred from Asp297 mainly to porphyrin (Asp297 charge:�0.3e), yielding a spin density of 0.66e/0.63e/0.63e for RC/TSH/HYD on Asp297 (A-prop spin densities: 0.11e/0.09e/0.10e; por-phyrin spin densities without propionates: 0.04e/0.08e/0.04e). Inthis Cpd II-like case, the barrier/endothermicity is 2/1 kcal/molhigher than for R2sE, showing that deprotonation of Asp297increases the barrier.

In conclusion, single-point gas-phase calculations for the rela-tive energies generally agree with the QM/MM results when usingthe smallest models without heme side chains or extended modelsthat include side chains and their screening residues. The changesin spin density distribution with proximal ligand representation(without heme side chains) do not affect the reaction energies, andthe sulfhydryl representation gives gas-phase spin density distri-butions that are closest to the QM/MM results.

Discussion and Conclusions

In recent years, there has been a rapid growth in the number ofQM/MM studies on biomolecules in general and on enzymaticreactions in particular. It is fair to state, however, that the QM/MMmethodology is still under development and that there is a need todevelop reliable and generally accepted QM/MM protocols forinvestigating such complex systems. Given this situation, it isobviously important to analyze and resolve any major discrepan-cies between QM/MM results that have appeared in the literature.In the present study, we have addressed the sources of contradic-

TABLE 5. UB3LYP Barrier/Endothermicity (kcal/mol) for Isolated QM Regions Extractedfrom R1/B2-Optimized B3LYP/CHARMM SnapX Geometries with Protonated Asp297.a

Without w903 With w903

R1� Rm R2 R1� Rm R2

B1 — 20.6/15.9 21.0/16.4 20.1/15.5 w 17.7/10.6 17.2/11.7 16.6/11.0s 21.6/17.6 21.2/17.2 20.8/16.9 sw 19.6/12.6 19.3/12.3 19.1/12.1

B2 — 20.9/11.8 20.4/12.9 20.4/10.9 w 18.8/7.9 18.7/9.2 19.0/7.4s 22.4/12.7 21.9/12.5 22.3/12.3 sw 21.3/9.2 20.9/9.1 20.7/8.8

aR2sE/B1: 17.4/10.7 kcal/mol; R2sEasp/B1: 19.0/11.8 kcal/mol (Asp297 is deprotonated; no relaxation); R2sEhsd/B1:17.5/10.5 kcal/mol (His355 is deprotonated; no relaxation); R2sEnohis/B1: 17.8/10.8 kcal/mol (His355 is removed fromthe model; no relaxation). The labels s and w indicate QM regions that include side chains and w903, respectively (seetext).



tory QM/MM findings for the P450cam-catalyzed H-abstractionreaction of camphor published by the S/T8 and G/F12,13 groups. Inthe course of this work, we have systematically investigated thefactors that influence the barrier and endothermocity of this reac-tion, and have thereby gained a number of new mechanistic in-sights.

On the technical side, we find that the ChemShell and QSiteprograms used by S/T and G/F, respectively, yield analogoussingle-point QM/MM results at given geometries, both for relativeenergies and spin densities, when comparable options are used(QM region, basis set, etc.). QM/MM geometry optimizations aremore problematic because of the difficulties of incomplete opti-mization and convergence to unconnected local minima, whichbecome more pressing for larger and more complex systems.Closer inspection of the G/F geometries, in particular, of thehydrogen bonds therein, suggests that they suffer from these prob-lems, and UB3LYP/CHARMM reoptimizations along pathwayswith a consistent protein environment indeed increase the com-puted barriers and endothermicities by about 5 kcal/mol, whichaccounts for about half of the reported discrepancies.8,12,13 An-other point of concern is the choice of the starting geometry forQM/MM optimizations: one strategy is to remain as close aspossible to available X-ray structures,12,13 while an alternativeapproach follows classical biomolecular simulation protocols anduses snapshots from equilibrated MD runs.8 In the present case, thelatter option leads to a migration of w903 away from its X-rayposition near the FeO moiety, which causes an increase of thecomputed relative energies by another 5 kcal/mol and is respon-sible roughly for the other half of the reported discrepancies.8,12,13

In the light of these findings, it seems reasonable to use both X-raylike geometries and MD snapshots as starting points in QM/MMgeometry optimizations, in order to cover a large diversity ofstructures.

The choice of the QM region is another important technicalissue. The common wisdom is that it is advisable to make the QMregion as large as possible to minimize any problems at theQM/MM boundary (link atom treatment, etc.). Although this istrue in general, one still needs to be careful. For example, the QMregion R2s includes the heme with all side chains, camphor, and anextended proximal Cys ligand (120 atoms, Fig. 2), but the twonegatively charged propionate side chains may cause problems ifthey are not properly screened. These problems can be overcomeby adopting the still larger QM region R2sE, which also incorpo-rates the screening residues (170 atoms, Fig. 1), but they may oftenalso be avoided by simply using a smaller QM region such as R1(51 atoms) without side chains. One of the lessons of the presentwork is that the hydrogen abstraction reaction is mostly influencedby the situation around the reacting atoms (FeO � � � HC), and lessby factors in more remote regions (such as the propionate sidechains and their environment; see below). These regions may playan essential role in other processes (e.g., electron transfer), andtheir inclusion into the QM part will then be mandatory, of course.The choice of the QM region should thus ultimately depend on theprocesses to be studied in QM/MM work.

As a final technical remark, it is often inevitable in DFT/MMcalculations to use rather modest basis sets. The present workemphasizes the need to validate this choice by test calculationswith larger basis sets. Comparisons between the present B1 and B2

results clearly show that the computed spin densities are basis setdependent and tend to be significantly smaller with the larger basis(B2). Likewise, the computed endothermicities become smallerupon basis set extension from B1 to B2, while the calculatedbarriers are much less affected. This latter example also suggeststhat it may often not be advisable to extrapolate from computedendothermicities to barriers on the basis of the Hammond postu-late,41 because these two quantities may be affected in a differentmanner by various computational parameters.

We now turn to the chemical insights gained from the presentstudy. The most important chemical result concerns the role of thecrystal water molecule, w90320 that is liberated during the con-version of Cpd 0 to Cpd I and acts as a catalyst for hydrogenabstraction, by forming a hydrogen bond to the FeO oxo atom thatis stronger in the transition state than in the reactant. This autoca-talysis lowers the barrier by about 4 kcal/mol and thus contributesto the elusiveness of Cpd I in P450cam, which may help explainingwhy it has never been observed experimentally.

Previous QM/MM calculations by QSite have suggested thatthe A-propionate side chain of the heme carries some spin densityduring the hydrogen abstraction reaction by Cpd I. Single-pointChemShell calculations at the G/F geometries with small basis setsreproduce the reported A-prop spin densities12,13 which, however,diminish by ca. 0.1e, and thus become quite small, upon basis setextension. Moreover, it should be stressed that the A-prop car-boxylates are not fully screened by neighboring residues in the G/Fgeometries. The present systematic QM/MM calculations confirmprevious findings8,18 that there is no such spin density wheneverthe side chain carboxylates are properly screened. At ambienttemperatures, the protein will undergo thermal motion, and it isthus likely that it will encounter conformations where the screen-ing is incomplete and where A-prop spin density can develop.Hence, in our view, the occurrence of such spin density is adynamic phenomenon, but not a static feature of our system: in anequilibrium situation, the carboxylates will generally be screenedbecause this is energetically favorable, and they will thus not carrystatic spin density. In principle, this could be checked experimen-tally by EPR spectroscopy, which has unfortunately not yet beenpossible for Cpd I in P450cam due to its elusive nature. EPRmeasurements on related systems confirm, however, that there isno detectable spin density on side chain carboxylates: for example,the observed EPR spectra for Cpd I in chloroperoxidase (CPO)enzyme,42 which also has a proximal cysteinate ligand, do notindicate any side chain spin density. Thus, carboxylate spin densitycannot be a prerequisite for efficient catalysis by Cpd I.

The occurrence of A-prop spin density in the QM/MM calcu-lations depends on the choices made for the A-prop environment,in particular on the protonation states of the Asp297 and His355residues, which also influence the preferred conformational ar-rangement. Judging from the pKa values of Asp297 of 8.44(PROPKA) at the X-ray geometry (1DZ9), Asp297 should beprotonated. In this case, QM/MM optimizations yield an A-propenvironment close to the X-ray structure. Furthermore, previousMD simulations on the resting state24 and pentacoordinated com-plexes34 of P450cam with protonated Asp297 show that the A-propenvironment remains intact and close to the X-ray structure duringthe full MD run. Whenever Asp297 is chosen to be protonated, theA-propionate is well screened and carries no spin density.



In previous QM/MM studies,8,12,13 Asp297 was assigned to bedeprotonated. With this choice, the A-prop environment of theheme has significant conformational freedom, and the correspond-ing optimized conformations differ considerably from the X-raystructure. Even in this case, there is no spin density on the A-propcarboxylate if it is screened by Arg299 (as in the X-ray structure).8

Spin density is found on A-prop, however, when the H-bondinginteraction between the O2A atom of A-prop and Arg299 ismissing (as in the G/F geometries); such unscreened conforma-tional arrangements may yield propionate spin densities whenHis355 is chosen to be deprotonated.12,13

The preceding considerations show that the occurrence of A-prop spin density during protein dynamics depends in a delicatemanner on the A-prop environment that is capable of tuning theelectronic situation. It should be emphasized, however, that thecomputed hydrogen abstraction barriers are not affected much bythese factors. In the present systematic QM/MM calculations (withAsp297 and His355 treated at the MM level), the chosen protona-tion states and the amount of spin density at the propionates do nothave a significant influence on the barrier. There is no clearcorrelation between the computed A-prop spin densities and bar-riers, and hence no support for the proposed side-chain controlledtransition state stabilization mechanism.12,13

We finally address the unusual species with deprotonatedAsp297 and His355 that has been discovered in QM/MM calcu-lations with Asp297 included in the QM region. We confirm thatQM/MM optimizations for this setup give an A-prop environmentthat is consistent with the X-ray structure, with a short distance ofabout 2.2 Å between the oxygen atoms of Asp297 and A-propi-onate (O2A), both with formally negative charges. In this species,electron transfer occurs from deprotonated Asp297 to the porphy-rin ring, which acquires an almost closed-shell configuration. Thiselectronic structure corresponds to Cpd II coupled with an addi-tional unpaired electron shared between the A-prop and Asp297carboxylate oxygen atoms. It is well known experimentally thatCpd II species of heme complexes show sluggish C-H hydroxy-lation reactivity,43 and it is therefore not surprising that the hy-drogen abstraction barrier for this species is computed to be 3–4kcal/mol higher than for alternative Cpd I-like species with differ-ent protonation patterns, especially that with protonated Asp297and His355, which also has an A-prop environment consistent withthe X-ray structure.

At its optimized minimum geometry, the Cpd II-type species ismore stable than the Cpd I-like alternative: the former is stabilizedby a two-center three-electron O � � � O bonding interaction, whilethe latter is strongly destabilized by Coulomb repulsion betweenthe negatively charged oxygen atoms. The attractive bonding in-teraction is short-ranged, while the Coulomb repulsion is longranged. During protein dynamics, it would thus seem likely thatany Cpd II-type species would spontaneously convert to the CpdI-like alternative as soon as the relevant O � � � O distance exceedsa certain threshold value, and it would then seem hard to return toCpd II-type geometries because of the strong electrostatic repul-sions experienced on the Cpd I surface along the relevant O � � �

O coordinate. Therefore, we believe that Cpd II-type conforma-tions with deprotonated Asp297 and His355 will appear only veryrarely during protein dynamics. More importantly, we do not

expect them to play a role in the hydrogen abstraction reactionbecause of their significantly higher barriers (see above).

We conclude on the basis of the present extensive QM/MMcalculations that the most probable scenario for the hydrogenabstraction reaction involves protonated Asp297 and His355: thecorresponding optimized QM/MM structures are consistent withthe available X-ray structures and have reasonable barriers (with-out showing A-prop spin density). However, these detailed con-siderations about the A-prop environment should not distract fromthe more important inclusion that the hydrogen abstraction barrieris not affected much by the A-prop environment. It is rather thepresence of the catalytic w903 molecule close to the FeO moietythat leads to a significant lowering of the barrier.

Acknowledgments

We thank R. A. Friesner and V. Guallar for supplying theirpublished structures. We are grateful to them as well as ShimritCohen and Jan C. Schoneboom for helpful discussions.

References

1. (a) Ortiz de Montellano, P. R., Ed. Cytochrome P450: Structure,Mechanism, and Biochemistry; Plenum Press: New York, 1995, 2nded.; (b) Ortiz de Montellano, P. R., Ed. Cytochrome P450: Structure,Mechanism, and Biochemistry; Kluwer Academic/Plenum Publishers:New York, 2004, 3rd ed.

2. Shaik, S.; Kumar, D.; de Visser, S. P.; Altun, A; Thiel, W. Chem Rev2005, 105, 2279.

3. Denisov, I. G.; Makris, T. M.; Sligar, S. G.; Schlichting I. Chem Rev2005, 105, 2253.

4. Groves, J. T.; Watanabe, Y. J Am Chem Soc 1988, 110, 8443.5. Gelb, M. H.; Heimbrook, D. C.; Malkonen, P.; Sligar, S. G. Biochem-

istry 1982, 21, 370.6. Hjelmeland, L. M.; Aronow, L.; Trudell, J Biochem Biophys Res

Commun 1977, 76, 541.7. Groves, J. T.; McClusky, G. A. J Am Chem Soc 1976, 98, 859.8. Schoneboom, J. C.; Cohen, S.; Lin, H.; Shaik, S.; Thiel, W. J Am

Chem Soc 2004, 126, 4017.9. Ogliaro, F.; Harris, N.; Cohen, S.; Filatov, M.; de Visser, S. P.; Shaik,

S. J Am Chem Soc 2000, 122, 8977.10. Shaik, S.; Cohen, S.; de Visser, S. P.; Sharma, P. K.; Kumar, D.;

Kozuch, S.; Ogliaro, F.; Danovich, D. Eur. J Inorg Chem 2004, 35,207.

11. Mueller, E. J.; Lioda, P. J.; Sligar, S. G. In Cytochrome P450: Struc-ture, Mechanism, and Biochemistry; Ortiz de Montellano, P. R., Ed.;Plenum Press: New York, 1995, vol. 2, Chap 3, 2nd ed.

12. Guallar, V.; Baik, M.; Lippard, S. J.; Friesner, R. A. Proc Natl AcadSci USA 2003, 100, 6998.

13. Guallar, V.; Friesner, R. A. J Am Chem Soc 2004, 126, 8501.14. Schoneboom, J. C.; Lin, H.; Reuter, N.; Thiel, W.; Cohen, S.; Ogliaro,

F.; Shaik, S. J Am Chem Soc 2002, 124, 8142.15. Schlichting, I.; Berendzen, J.; Chu, K.; Stock, A. M.; Maves, S. A.;

Benson, D. A.; Sweet, R. M.; Ringe, D.; Petsko, G. A.; Sligar, S. G.Science 2000, 287, 1615.

16. QSite; A QM/MM software, Schrodinger, Inc.: Portland, OR, 2001.17. Sherwood, P.; de Vries, A. H.; Guest, M. F.; Schreckenbach, G.;

Catlow, C. R. A.; French, S. A.; Sokol, A.A.; Bromley, S. T.; Thiel,W.; Turner, A. J.; Billeter, S.; Terstegen, F.; Thiel, S.; Kendrick, J.;



Rogers, S. C.; Casci, J.; Watson, M.; King, F.; Karlsen, E.; Sjovoll, M.;Fahmi, A.; Schafer, A.; Lennartz, C. J Mol Struct (Theochem) 2003,632, 1.

18. Bathelt, C. M.; Zurek, J. L.; Mulholland, A. J.; Harvey, J. N. J AmChem Soc 2005, 127, 12900.

19. Derat, E.; Cohen, S.; Shaik, S.; Altun, A.; Thiel, W. J Am Chem Soc2005, 127, 13611.

20. Altun, A; Guallar, V.; Friesner, R. A.; Shaik, S.; Thiel, W. J Am ChemSoc, 2006, 128, 3924.

21. MacKerell, A. D., Jr.; Bashford, D.; Bellott, M.; Dunbrack R. L., Jr.;Evanseck, J. D.; Field, M. J.; Fischer, S.; Gao, J.; Guo, H.; Ha, S.;Joseph–McCarthy, D.; Kuchnir, L.; Kuczera, K.; Lau, F. T. K.; Mattos,C.; Michnick, S.; Ngo, T.; Nguyen, D. T.; Prodhom, B.; Reiher, W. E.,III; Roux, B.; Schlenkrich, M.; Smith, J. C.; Stote, R.; Straub, J.;Watanabe, M.; Wiorkiewicz–Kuczera, J.; Yin, D.; Karplus, M. J PhysChem B 1998, 102, 3586.

22. Lounnas, V.; Wade, R. C. Biochemistry 1997, 36, 5402.23. Li, H.; Robertson, A. D.; Jensen, J. H. Proteins Struct Funct Bioinfor

2005, 61, 704.24. Schoneboom, J. C.; Thiel, W. J Phys Chem B 2004, 108, 7468.25. (a) Ahlrichs, R.; Bar, M.; Haser, M.; Horn, H.; Kolmel, C. Chem Phys

Lett 1989, 162, 165; (b) Ahlrichs, R.; Bar, M.; Baron, H.-P.; Bauern-schmitt, R.; Bocker, S.; Ehrig, M.; Eichkorn, K.; Elliot, S.; Furche, F.;Haser, M.; Horn, H.; Hattig, C.; Huber, C.; Huniar, U.; Kattanneck,M.; Kohn, A.; Kolmel, C.; Kollwitz, M.; May, K.; Ochsenfeld, C.;Ohm, H.; Schafer, A.; Schneider, U.; Treutler, O.; v. Arnim, M.;Weigend, F.; Weis, P.; Weiss, H. TURBOMOLE 5.5, University ofKarlsruhe, 2002.

26. (a) Becke, A. D. Phys Rev A 1988, 38, 3098; (b) Lee, C.; Yang, W.;Parr, R. G. Phys Rev B 1988, 37, 785; (c) Becke, A. D. J Chem Phys1993, 98, 5648.

27. Smith, W.; Forester, T. J Mol Graph 1996, 14, 136.

28. Billeter, S. R.; Turner, A. J.; Thiel, W. Phys Chem Chem Phys 2000,2, 2177.

29. Bakowies, D.; Thiel, W. J Phys Chem 1996, 100, 10580.30. Antes, I.; Thiel, W. Hybrid Quantum Mechanical and Molecular

Mechanical Methods; Gao, J., Ed.; ACS Symp. Ser. 712; AmericanChemical Society: Washington, DC, 1998, p. 50.

31. de Vries, A. H.; Sherwood, P.; Collins, S. J.; Rigby, A. M.; Rigutto,M.; Kramer, G. J. J Phys Chem B 1999, 103, 6133.

32. Hay, J. P.; Wadt, W. R. J Chem Phys 1985, 82, 299.33. (a) Wachters, A. J. H. J Chem Phys 1970, 52, 1033; (b) Hay, P. J.

J Chem Phys 1977, 66, 4377; (c) Bauschlicher, C. W., Jr.; Langhoff,S. R.; Barnes, L. A. J Chem Phys 1989, 91, 2399.

34. Altun, A.; Thiel, W. J Phys Chem B 2005, 109, 1268.35. (a) Ditchfield, R.; Hehre, W. J.; Pople, J. A. J Chem Phys 1971, 54,

724; (b) Hehre, W. J.; Ditchfield, R.; Pople, J. A. J Chem Phys 1972,56, 2257; (c) Hariharan, P. C.; Pople, J. A. Theor Chim Acta 1973, 28,213; (d) Clark, T.; Chandrasekhar, J.; Spitznagel, G. W.; Schleyer, P.v. R. J Comp Chem 1983, 4, 294.

36. Harris, D. L. Curr Opin Chem Biol 2001, 5, 724.37. Ogliaro, F.; Harris, N.; Cohen, S.; Filatov, M.; de Visser, S. P.; Shaik,

S. J Am Chem Soc 2000, 122, 8977.38. Ogliaro, F.; Cohen, S.; Filatov, M.; Harris, N.; Shaik, S. Angew Chem

Int Ed 2000, 39, 3851.39. Li, G.; Zhang, X.; Cui, Q. J Phys Chem 2003, 107, 8643.40. Guallar, V.; Olsen, B. J Inorg Biochem, 2006, 100, 755.41. Hammond, G. S. J Am Chem Soc 1955, 77, 334.42. Rutter, R.; Hager, L. P.; Dhonau, H.; Hendrich, M.; Valentine, M.;

Debrunner, P. Biochemistry 1984, 23, 6809.43. (a) Groves, J. T.; Gross, Z.; Stern, M. K. Inorg Chem. 1994, 33, 5065;

(b) Nam, W.; Park, S.-E.; Lim, I. K.; Lim, M. H.; Hong, J.; Kim, J.J Am Chem Soc 2003, 125, 14674.



Documents

Systematic QM/MM investigation of factors that affect the cytochrome P450-catalyzed hydrogen abstraction of camphor