Investigating biological systems using first principles

Investigating biological systems using first principles Car–Parrinello molecular dynamics simulations Matteo Dal Peraro1, Paolo Ruggerone2, Simone Raugei3, Francesco Luigi Gervasio4 and Paolo Carloni3 Density functional theory (DFT)-based Car–Parrinello molecular dynamics (CPMD) simulations describe the time evolution of molecular systems without resorting to a predefined potential energy surface. CPMD and hybrid molecular mechanics/CPMD schemes have recently enabled the calculation of redox properties of electron transfer proteins in their complex biological environment. They provided structural and spectroscopic information on novel platinum-based anticancer drugs that target DNA, also setting the basis for the construction of force fields for the metal lesion. Molecular mechanics/CPMD also lead to mechanistic hypotheses for a variety of metalloenzymes. Recent advances that increase the accuracy of DFT and the efficiency of investigating rare events are further expanding the domain of CPMD applications to biomolecules. Addresses 1 Center for Molecular Modeling, Department of Chemistry, University of Pennsylvania, 231 South 34th Street, Philadelphia, PA 19104, USA 2 CNR-INFM-Slacs and Physics Department, University of Cagliari, SP Monserrato-Sestu km 0.700, 09042 Monserrato (CA), Italy 3 International School for Advanced Studies and INFM-Democritos, Via Beirut 2-4, 34100 Trieste, Italy 4 Computational Science, Department of Chemistry and Applied Biosciences, ETH Zurich, USI Campus, Via Giuseppe Buffi 13, CH-6900 Lugano, Switzerland Corresponding author: Carloni, Paolo ([email protected])

Current Opinion in Structural Biology 2007, 17:149–156 This review comes from a themed issue on Theory and simulation Edited by Richard Lavery and Kim A Sharp

0959-440X/$ – see front matter # 2007 Elsevier Ltd. All rights reserved. DOI 10.1016/j.sbi.2007.03.018

Introduction Density functional theory (DFT) methods are standard tools in quantum chemistry, in view of their favorable scaling with the number of electrons and their everimproving accuracy of functionals for the estimation of exchange and correlation effects. DFT was first applied within a molecular dynamics (MD) scheme some 20 years ago, when Car and Parrinello (CP) proposed to treat electronic degrees of freedom as dynamical variables and to couple the resulting fictitious electron dynamics with the classical dynamics of the nuclei [1]. Thus, CPMD describes the time evolution of molecular systems www.sciencedirect.com

(presently up to 102 heavy atoms for 102 ps) without resorting to a force field. Currently, Born–Oppenheimer approaches to first principles MD are also widely and efficiently used [2] (see also Update). To treat biologically relevant systems, which are invariably large for first principles calculations, hybrid molecular mechanics/Car–Parrinello molecular dynamics (MM/ CPMD) schemes have been introduced, subsequent to the quantum mechanics/molecular mechanics approach originally proposed by Warshel and Levitt [3]: a region of interest (e.g. an enzymatic active site) is described at the DFT level, dynamically and electrostatically coupled with the rest of the system, which is treated using biomolecular force fields [4–6]. Most applications presented here follow such an approach, as developed by Rothlisberger and co-workers [4], in which the Gromos96 program [7] is employed for the classical part. One of the main benefits of the CP approach is its ability to simulate complex reactions from first principles. The classical approach of first principles quantum chemistry is to determine local minima (which identify possible equilibrium configurations) and saddle points (which determine reaction pathways) on the potential energy surface. Unfortunately, this strategy might encounter difficulties when entropic effects are important and the free energy surface needs to be explored [8]. CPMD, which does include temperature effects, can benefit from the use of statistical mechanics methods (e.g. thermodynamic integration [9], metadynamics [10,11], steering dynamics [12], umbrella sampling) to investigate rare events, such as enzymatic reaction mechanisms. Particular emphasis is placed in the first section of this review on studies of metal-based enzymes. The interaction between a ligand and its target might depend on the electronic structure in such a subtle way that it is difficult to capture with force field based MD. In the second section, we report a few applications that address this issue. We focus in the third section on the electronic properties of DNA and electron transfer proteins. Several other excellent contributions (notably investigations of organic enzymes and ion channels [12,13,14–16,17]) are not reported here because of space limitations. We finally draw some general conclusions on current limitations and challenges of the method.

Metal-based enzymes Metal-containing proteins represent almost half of the proteome of living organisms. In the past few years, MM/ Current Opinion in Structural Biology 2007, 17:149–156

150 Theory and simulation

CPMD simulations combined with classical MD (used to sample fluctuations occurring on timescales not accessible by first principles calculations alone) have contributed important insights into the catalytic mechanisms and structural features of a variety of metalloenzymes [18,1920,21,2223,24,25,26]. The bacterial expression of zinc metallo b-lactamases (MbLs) represents a key resistance mechanism against the action of b-lactam antibiotics [27]. MbLs hydrolyze the b-lactam N–C bond of the antibiotic aided by one or two Zn2+ ions, which bind to the nucleophilic agent of the reaction, a hydroxide (Figure 1). MM/CPMD studies of mononuclear and binuclear MbLs (from Bacillus cereus [18] and Bacteroides fragilis [19], respectively) suggest that the insertion of a second equivalent Zn2+ preserves the substrate-binding ability while improving the efficiency of b-lactam hydrolysis. Moreover, Zn2+ coordination sphere

flexibility plays a key role, regardless of the metal content, in activating the nucleophile and a water molecule (buried at the metal site) for b-lactam ring scission (Figure 1). Further studies of other metallohydrolases sharing the same fold (such as human glyoxalase II; M Dal Peraro, ML Klein, unpublished) have led to the suggestion that binuclear MbLs and related hydrolases could use metal ions as dynamic, flexible anchors for the rearrangement and pKa activation of reactive moieties. The role of water dissociation in Mg2+-dependent phosphatase catalysis has been recently highlighted by several calculations [20,21,22]. For the N-terminal domain of soluble epoxide hydrolase, MM/CPMD calculations point out the crucial role of an extended hydrogen-bond network of water molecules at the active site in enzymatic phosphoryl transfer. The calculations also reveal the importance of the electrostatic effects of the metal ion,

Figure 1

Binuclear B1 MbL from B. fragilis in complex with a common substrate, cefotaxime. The inset sketches the reaction mechanism, which is proposed by MM/CPMD calculations to involve two Zn2+ ions [19]. In a single concerted step, the nucleophile (OH–) attacks the b-lactam ring, assisted by Zn1 (black arrow), and the catalytic water (activated by Zn2), protonates the b-lactam nitrogen, promoting the opening of the ring (red arrow) and eventually replaces OH_ bridging the two Zn2+ ions (blue arrow). In mononuclear B1 MbL from B. cereus, instead the metal ion distorts the tetrahedral polyhedron at the first transition state and accepts the catalytic water, which is then activated in the second step for blactam ring scission [18]. The second Zn2+ ion in the binuclear enzyme permits an elegant and concerted single-step pathway by activating the catalytic water at the same time as nucleophilic attack [19]. Thus, by efficiently stabilizing the negative charge developed at the b-lactam nitrogen during the reaction, this metal gives binuclear MbLs a functional advantage over mononuclear species (calculated DF# = 18 versus 21 kcal mol 1, compared with experimental 17 kcal mol 1). Current Opinion in Structural Biology 2007, 17:149–156

www.sciencedirect.com

Car–Parrinello molecular dynamics simulations Dal Peraro et al. 151

which promote water dissociation, leading to high enzymatic efficiency [20,21]. Moreover, extensive MM/ CPMD metadynamics calculations on bovine Hsc70 ATPase have pinpointed, for the first time, the unique role of a specific Mg2+-bound water molecule, which acts as a trigger in the initial phase of the ATP hydrolysis reaction, aided by the cooperative action of Mg2+ and K+ cations at the active site [22].

Figure 2

A new mechanism for the process of formic acid binding has been proposed for the Fe-catalase enzyme: the reaction intermediate is, in fact, described as a hydroxoferryl species instead of the classic oxoferryl intermediate [23,24]. Finally, in vanadium haloperoxidases, extremely efficient oxidants of halides to hypohalous acids, MM/CPMD metadynamic simulations [25] suggest that the addition of hydrogen peroxide to the enzyme does not involve initial protonation of the cofactor, in contrast to previous proposals. Instead, the hydrogen peroxide directly attacks the axial hydroxo group and the subsequent intermediate is promptly protonated to form a peroxo species, which in turn reacts with the halogen moiety to yield a hypohalogen vanadate.

Ligand–target interactions Among transition-metal-based anticancer drugs that target DNA, cisplatin is the most widely used [28,29]. MM/ CPMD-based modeling of a cisplatin–DNA adduct [30], in which the metal binds to two adjacent guanines, recovered some structural features of the NMR structure [31]. This approach also provided new insights into the structure of a new generation of platinum complexes (1 and 2 in Figure 2), which distort DNA much less than cisplatin [32]. DNA-binding organo-ruthenium compounds (3 and 4 in Figure 2) constitute a promising alternative approach to platinum-based therapies, as the low toxicity of ruthenium is associated with pH-dependent cancer cell specificity [33]. MM/CPMD simulations show that compound 3 might bind to the DNA major groove via a cisplatin-like complexation motif involving two adjacent guanine N7 atoms, inducing a bending angle similar to cisplatin [34]. Additionally, compound 4 causes breaks in Watson–Crick base pairing adjacent to the ruthenium-binding site, in agreement with experiment [35]. Force field MD simulations might encounter difficulties when a ligand forms unusual types of interactions with its target. This is the case for protein inhibitors that form ultrashort hydrogen bonds (acceptor–donor distance