Article pubs.acs.org/biochemistry
Probing the Time Dependency of Cyclooxygenase‑1 Inhibitors by Computer Simulations Yasmin Shamsudin Khan, Hugo Gutiérrez-de-Terán,* and Johan Åqvist* Department of Cell and Molecular Biology, Uppsala University, BMC, Box 596, SE-751 24 Uppsala, Sweden S Supporting Information *
ABSTRACT: Time-dependent inhibition of the cyclooxygenases (COX) by a range of nonsteroidal anti-inflammatory drugs has been described since the first experimental assays of COX were performed. Slow tight-binding inhibitors of COX-1 bind in a two-step mechanism in which the EI → EI* transition is slow and practically irreversible. Since then, various properties of the inhibitors have been proposed to cause or affect the time dependency. Conformational changes in the enzyme have also been proposed to cause the time dependency, but no particular structural feature has been identified. Here, we investigated a series of inhibitors of COX1 that are either time-independent or time-dependent using a combination of molecular dynamics simulations, binding free energy calculations, and potential of mean force calculations. We find that the time-dependent inhibitors stabilize a conformational change in the enzyme mainly identified by the rotation of a leucine side chain adjacent to the binding pocket. The induced conformation has been previously shown to be essential for the high binding affinities of tight-binding inhibitors in COX-1. The results of this work show that the structural features of the enzyme involved in both time-dependent and tightbinding inhibition are identical and further identify a structural mechanism responsible for the transition between the two enzyme−inhibitor complexes characteristic of slow tight-binding COX-1 inhibitors.
T
drugs to the potential for new highly selective NSAIDs for adjuvant cancer therapies. The selectivity of NSAIDs has been attributed to a timedependent property. Selective and preferential inhibitors of one isoform have also been demonstrated to have the kinetic profiles of slow tight-binding inhibitors in that isoform.11,12 In COX-1, slow tight-binding inhibitors follow a two-step mechanism
he selectivity profile of the most common class of nonopiod anti-analgetics, the nonsteroidal anti-inflammatory drugs (NSAIDs), has implications for the associated physiological responses. The constitutive isoform of the target enzyme cyclooxygenase (COX-1) promotes the production of angiogenic growth factors and is overexpressed in ovarian cancer.1 The second, inducible isoform (COX-2) is overexpressed in several cancers, including lung, pancreatic, and colorectal cancer.2−5 Both isoforms catalyze the conversion of arachidonic acid into prostaglandins G2 and H2, which are important in the signaling pathway of inflammations, fevers, and pain. NSAIDs are thus traditionally used for reducing common fevers and inflammations and for treatment of a large range of intermittent and chronic pain, such as migraines and arthritic pain.6 Many of the traditional NSAIDs used for nonchronic pain, such as ibuprofen, naproxen, and diclofenac, are either nonselective or COX-1 preferential inhibitors, known to cause side effects such as gastric intestinal bleeding, peptic ulcer formation, and kidney problems.7−9 Although non-ulcerogenic COX-2 selective drugs, such as rofecoxib, have been developed and marketed, these drugs have been attributed to increased risks of myocardial infarctions caused by increased incidence of thrombosis, which led to the withdrawal of some of them from the market.10 Thus, the importance of understanding the binding mechanisms, and consequently, the selectivity of the different classes of NSAIDs extends from concerns over long-term side effects in these © XXXX American Chemical Society
E + I ⇄ EI → EI*
(1)
The first step is rapid and reversible, similar to the profiles of rapidly reversible inhibitors (E + I ⇄ EI), while the second step is slow and practically irreversible. Although the kinetic profiles of NSAIDs can be experimentally determined by assaying instant inhibition and inhibition after preincubation with the enzyme, the mechanisms regulating the observed time dependency have not been elucidated. However, it has been hypothesized that structural rearrangements in the enzyme cause the time dependency.11,13,14 In previous work, we attributed a rotation of the Leu531 side chain, located in the binding pocket between the key residues Arg120 and Ser530 in human COX-1 (Figure 1A), Received: October 3, 2016 Revised: March 7, 2017 Published: March 17, 2017 A
DOI: 10.1021/acs.biochem.6b01006 Biochemistry XXXX, XXX, XXX−XXX
Article
Biochemistry
Figure 1. (A) Detailed view of the binding site of COX-1. The closed conformation has the L531 side chain pointing toward the ligand (magenta), while the open conformation features L531 rotated away (beige) from the main binding pocket. (B) Overlay of the average structure of the final window of PMF calculations of 1a in the open conformation (beige) and the crystal structure of Protein Data Bank entry 1Q4G (multiple colors). (C) Chemical structures of the investigated data set containing flurbiprofen derivatives (1a−j) and ibuprofen derivatives (2a−c).
not make the brominated inhibitor time-dependent.17 Similarly, the substitution of fluorine for hydrogen in flurbiprofen reduces the affinity of the inhibitor, but the substitution does not reverse the time dependency. In a recent work, we proposed a model in which tight binding is associated with the stability of a particular conformational state of the enzyme, denoted as the open conformation, which is most easily distinguished from the closed conformation by the rotameric state of Leu531 (Figure 1A).15 Flurbiprofen, a slow tight-binding inhibitor, was found to exclusively stabilize the otherwise unfavorable open conformation, while the lower-affinity and rapidly reversible inhibitor ibuprofen stabilized the intrinsic closed conformation. In this work, we expand that model and probe the underlying structural mechanisms of time dependency by examining the two-state binding hypothesis with a series of derivatives of flurbiprofen (Figure 1C, compounds 1a−j) and ibuprofen (Figure 1C, compounds 2a−c) bound to ovine COX-1 (oCOX-1). The ligands were subjected to molecular dynamics (MD) simulations in complex with COX-1 to estimate their binding affinities and to elucidate their interactions in each of the two enzyme conformations. The linear interaction energy (LIE) method was used to calculate free energies of binding
and the accompanying local structural rearrangements (Figure 1B), including the influx of water molecules into the binding pocket, to the high-affinity binding commonly associated with slow tight-binding COX-1 inhibitors.15 However, although the concepts of tight binding and time dependency have sometimes been used interdependently to describe some classes of COX inhibitors, the two terms are related but not exchangeable. For instance, a small group of NSAIDs have been classified as reversible time-dependent12 or slow reversible inhibitors,16 thus being time-dependent without the slow release characterizing tight-binding inhibitors. Several structural features have been proposed to affect the time dependency of COX inhibitors. One such feature is the substitution of carboxylates for methyl esters in time-dependent inhibitors. It has been experimentally determined that this substitution does indeed result in the formation of timeindependent inhibitors while simultaneously decreasing the affinities of the inhibitors.11 Furthermore, the presence of halogens in many time-dependent NSAIDs prompted the hypothesis that halogen substitution of time-independent inhibitors would convert them into time-dependent ones. However, the substitution of hydrogen for bromine in the timeindependent inhibitor ibuprofen increased the affinity but did B
DOI: 10.1021/acs.biochem.6b01006 Biochemistry XXXX, XXX, XXX−XXX
Article
Biochemistry
a pose similar to that of the cocrystallized inhibitor α-methyl-4biphenylacetic acid (BFL) in template 1Q4G.18 Binding free energies were calculated using the linear interaction energy (LIE) method22,26,27
and examine structure−activity relationships. In addition, potential of mean force (PMF) calculations were performed in which each enzyme−ligand complex was perturbed by gradually pushing the Leu531 side chain from the more commonly observed closed state to the open state previously associated with high-affinity tight-binding inhibitors.15 The results show a striking correlation between experimental and calculated binding free energies and provide a clear link between the experimentally determined time dependency of inhibitors and the stability of these in the open conformation. Altogether, our data suggest that for the class of slow tightbinding inhibitors of COX-1, inhibition in the closed conformation represents the reversible [EI] complex while binding to the open conformation would correspond to the subsequently formed irreversible [EI*] complex.
calc el ΔG bind = αΔ⟨Ulvdw −s ⟩ + β Δ⟨U l−s⟩ + γ
(2)
where α and β are predetermined parameters that scale the difference (Δ) in the average ligand-surrounding electrostatic ⟨Uell−s⟩ and nonpolar ⟨Uvdw l−s ⟩ interaction energies, as calculated from separate trajectories of 10 ns of MD simulations of the solvated protein−ligand complex and the free ligand in water. Empirical parameter α is a scaling coefficient for the nonpolar interactions, and in the standard parametrization of the LIE method, α = 0.18.26,27 Scaling factor β for the polar binding contribution for charged compounds was originally derived from the linear response approximation.26,27 Thus, for the carboxylated charged ligands, β = 0.5, while for noncarboxylated ligands, β has been determined to be 0.43 according to the standard LIE parametrization.22,27,28 γ is a constant offset parameter that fixes the scale for absolute binding free energies. It is obtained by regression fitting between experimental and calculated values for the whole ligand series, keeping α and β constant and considering only γ as a free parameter. The nature of this parameter has been related to hydrophobic descriptors of the binding site and is specific for the protein.29 Here, the value of γ was calculated to be −6.2 kcal mol−1, which is comparable to the value previously used for human COX-1 (γ = −6.4).15,22 An electrostatic correction term (ΔGelcorr = −0.34 kcal mol−1) was added to account for interactions of charged ligands with distant neglected ionized groups that were neutralized in the simulations.30 The calculated binding free energies are reported as average energies over replicates of independent MD simulations of 10 ns each, using the same conditions but with different initial random velocities, and the corresponding errors are reported as standard errors of the mean (sem). A minimal number of three replicates was used up to maximum of 10, where the requirement was to achieve an sem of
