ACTAUNIVERSITATIS

UPSALIENSISUPPSALA

2017

Digital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Science and Technology 1560

Non-Steroidal Anti-InflammatoryDrugs in Cyclooxygenases 1 and 2

Binding modes and mechanisms from computationalmethods and free energy calculations

YASMIN SHAMSUDIN KHAN

ISSN 1651-6214ISBN 978-91-513-0073-3urn:nbn:se:uu:diva-328478

Dissertation presented at Uppsala University to be publicly examined in B42, BMC,Husargatan 3, Uppsala, Thursday, 2 November 2017 at 10:15 for the degree of Doctorof Philosophy. The examination will be conducted in English. Faculty examiner: ICREAResearch Professor Xavier Barril (University of Barcelona).

AbstractShamsudin Khan, Y. 2017. Non-Steroidal Anti-Inflammatory Drugs in Cyclooxygenases1 and 2. Binding modes and mechanisms from computational methods and free energycalculations. Digital Comprehensive Summaries of Uppsala Dissertations from theFaculty of Science and Technology 1560. 55 pp. Uppsala: Acta Universitatis Upsaliensis.ISBN 978-91-513-0073-3.

Non-steroidal anti-inflammatory drugs (NSAIDs) are one of the most commonly used classesof drugs. They target the cyclooxygenases (COX) 1 and 2 to reduce the physiological responsesof pain, fever, and inflammation. Due to their role in inducing angiogenesis, COX proteins havealso been identified as targets in cancer therapies.

In this thesis, I describe computational protocols of molecular docking, molecular dynamicssimulations and free energy calculations. These methods were used in this thesis to determinestructure-activity relationships of a diverse set of NSAIDs in binding to their target proteinsCOX-1 and 2. Binding affinities were calculated and used to predict the binding modes. Basedon combinations of molecular dynamics simulations and free energy calculations, bindingmechanisms of sub-classes of NSAIDs were also proposed. Two stable conformations of COXwere probed to understand how they affect inhibitor affinities. Finally, a brief discussion onselectivity towards either COX isoform is discussed. These results will be useful in future denovo design and testing of third-generation NSAIDs.

Keywords: molecular dynamics simulations, binding free energy, molecular docking,cyclooxygenase, non-steroidal anti-inflammatory drugs, free energy perturbation, potentials ofmean force

Yasmin Shamsudin Khan, Department of Cell and Molecular Biology, Computational Biologyand Bioinformatics, Box 596, Uppsala University, SE-751 24 Uppsala, Sweden.

© Yasmin Shamsudin Khan 2017

ISSN 1651-6214ISBN 978-91-513-0073-3urn:nbn:se:uu:diva-328478 (http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-328478)

A day when you learn something new

is not a wasted day

List of Papers

This thesis is based on the following papers, which are referred to in the text by their Roman numerals.

I Shamsudin Khan, Y., Gutiérrez-de-Terán, H., Boukharta, L.,

Åqvist, J. (2014) Towards an optimal docking and free energy calculation scheme in ligand design with application to COX-1 inhibitors. J. Chem. Inf. Model., 54(5): 1488-1499

II Shamsudin Khan, Y., Kazemi, M., Gutiérrez-de-Terán, H., Åqvist, J. (2015) Origin of the enigmatic step-wise tight-binding inhibition of Cyclooxygenase-1. Biochemistry, 54(49): 7283-7291

III Shamsudin Khan, Y., Gutiérrez-de-Terán, H., Åqvist, J. (2017)

Probing the time-dependency of cyclooxygenase-1 inhibitors by computer simulations. Biochemistry, 56(13): 1911-1920

IV Shamsudin Khan, Y., Gutiérrez-de-Terán, H., Åqvist, J. (2017)

Molecular mechanisms in the selectivity of non-steroidal anti-in-flammatory drugs. Manuscript.

Reprints were made with permission from the respective publishers.

Appendix

i Diwakarla, S., Nylander, E., Grönbladh, A., Reddy Vanga, S., Shamsudin Khan, Y., Gutiérrez-de-Terán, H., Ng, L., Pham, V., Sävmarker, J., Lundbäck, T., Jenmalm-Jensen, A., Andersson, H., Engen, K., Rosenström, U., Larhed, M., Åqvist, J., Yeen Chai, S., Hallberg, M. (2016) Binding to and inhibition of insu-lin-regulated aminopeptidase (IRAP) by macrocyclic disulfides enhances spine density. Mol. Pharmacol., 89(45): 413-424

ii Diwakarla, S., Nylander, E., Grönbladh, A., Reddy Vanga, S.,

Shamsudin Khan, Y., Gutiérrez-de-Terán, H., Sävmarker, J., Ng, L., Pham, V., Lundbäck, T., Jenmalm-Jensen, A., Engen, K., Rosenström, U., Larhed, M., Åqvist, J., Yeen Chai, S., Hallberg, M. (2016) Aryl sulfonamide inhibitors of insulin-regulated ami-nopeptidase (IRAP) enhance spine density on hippocampal neu-rons. ACS Chem. Neurosc., 7(10): 1383-1392

Reprints were made with permission from the respective publishers.

Contents

Introduction ............................................................................................... 11

1. Methods................................................................................................. 131.1 Homology modelling ...................................................................... 131.2 Molecular mechanics and force fields ............................................ 131.3 Molecular docking .......................................................................... 15

1.3.1 GOLD ..................................................................................... 151.3.2 GLIDE .................................................................................... 15

1.4 Scoring functions ............................................................................ 161.4.1 Internal and external ranking .................................................. 161.4.2 Force field scoring functions .................................................. 161.4.3 Knowledge-based scoring functions ....................................... 171.4.4 Empirical scoring functions .................................................... 17

1.5 Molecular dynamics simulations .................................................... 181.5.1 Spherical boundary conditions ............................................... 19

1.6 Linear interaction energy ............................................................... 201.6.1 The non-polar term ................................................................. 201.6.2 The polar term ........................................................................ 201.6.3 The offset parameter γ ............................................................ 211.6.4 The electrostatic correction term ............................................ 21

1.7 Potential of mean force .................................................................. 221.8 Free energy perturbation ................................................................ 23

2. Computational protocols ....................................................................... 252.1 Optimizing a protocol for calculating binding free energies .......... 25

2.1.1 Performance of automated docking in COX-1 ....................... 292.1.2 Evaluating the LIE method ..................................................... 29

2.2 Combining free energy methods .................................................... 30

3. Cyclooxygenases ................................................................................... 323.1 Structure ......................................................................................... 32

3.1.1 The closed and open conformations of COX ......................... 333.2 Function .......................................................................................... 34

3.2.1 Arachidonic acid and prostaglandins ...................................... 35

4. Non-steroidal anti-inflammatory drugs ................................................. 374.1 Structural classes of NSAIDs ......................................................... 37

4.1.1 Salt-bridge and inverted binding modes of carboxylated inhibitors .......................................................................................... 394.1.2 S- and R enantiomers of propionic acid derivatives ............... 394.1.3 Effects of amine substitutions on the carboxylate .................. 404.1.4 Celecoxib analogues in the main and side-pocket .................. 41

4.2 Time-dependency and kinetic classes of NSAIDs ......................... 414.2.1 Time-dependency and tight-binding ....................................... 434.2.2 Time-dependency and selectivity ........................................... 44

5. Future perspectives ............................................................................... 46

6. Populärvetenskaplig sammanfattning ................................................... 48

Acknowledgements ................................................................................... 50

References ................................................................................................. 52

Abbreviations

AA Arachidonic acid ASA Acetylsalicylic acid (Aspirin) COX Cyclooxygenase (Prostaglandin H2 synthase) hCOX Human cyclooxygenase mCOX-2 Murine cyclooxygenase-2 oCOX-1 Ovine cyclooxygenase-1 FEP Free energy perturbation GLIDE Grid-based ligand docking with energetics GOLD Genetic optimization for ligand docking LIE Linear interaction energy MD Molecular dynamics MM Molecular mechanics NSAID Non-steroidal anti-inflammatory drug OPLS-AA Optimized potentials for liquid simulations all-atom PMF Potentials of mean force PGHS Prostaglandin H2 synthase SBC Spherical boundary conditions

11

Introduction

In recent years, computational methods have been developed for modelling chemical and biological systems. In 2013, the Nobel Prize in chemistry was awarded to three computational chemists for their work in developing algo-rithms and computational models and methods. Computational and quantum chemistry, based on quantum mechanics, are methods used to better under-stand chemical properties. Computational and systems biology are computa-tional fields through which biological units such as peptides, proteins, en-zymes, nucleic acids and molecular machines are researched and structure-activity relationships explored.

Non-steroidal anti-inflammatory drugs (NSAIDs) are the second most pre-scribed class of drugs in Sweden today, after antibiotics. While antibiotics are only sold in pharmacies as prescription drugs, NSAIDs can be bought either with a prescription or over-the-counter (OTC) in pharmacies and retail shops. In 2016 NSAIDs were prescribed to more than 1 million patients in Sweden1 and more than 300 million defined daily doses (DDD) were dispensed in a population of 10 million2.

NSAIDs inhibit the physiological responses of pain, inflammation and fe-ver by inhibiting the enzymes cyclooxygenase (COX) 1 and 2. These are isoforms of each other. The first-generation, or traditional, NSAIDs bind to the two isoforms non-selectively. They have side-effects including gastroin-testinal bleeding and renal problems.3-5 Nonetheless, they are indicated in medical conditions including arthritic conditions, acute musculoskeletal dis-orders, and other painful conditions resulting from trauma, including fractures and minor surgery.6

The second-generation NSAIDs, the coxibs, are selective only for COX-2. These have fewer of the side-effects associated with traditional NSAIDs, but increase the risk of myocardial infarctions due to thrombosis.7 However, apart from the anti-inflammatory and analgetic properties, these drugs also have an inhibitory effect on angiogenesis, a useful property in cancers therapies.

To counteract the side-effects of first and second generation NSAIDs, third generation NSAIDs could be developed. Minimization of side-effects and ef-ficient targeting of the desired symptoms could be achieved by improving the NSAID affinities and tailoring their ratio of binding to COX-1 and COX-2. However, to design drugs with these properties it is important to understand the binding modes and mechanisms of existing NSAIDs.

12

This thesis consists of three parts. The first chapter introduces theoretical models, computational, and free energy methods. The second chapter de-scribes computational protocols where these methods were used to obtain in-sights about the structure-activity relationships of NSAIDs and COX. These insights are described and discussed in the subsequent three chapters.

13

1. Methods

This section summarizes some of the theoretical and computational methods used in the work reported in this thesis.

1.1 Homology modelling Crystal and NMR structures of biomolecules are common starting points in computational methods such as molecular docking and simulations. However, when such structures are not available, an alternative is to use homology mod-elling. Structural fold is more conserved than sequences among homologous proteins.8 Thus, algorithms can be used to align a protein sequence to a tem-plate structure to create a new structure with similar 3D-structure.

In this thesis, the homology modelling software Modeller9 was used to cre-ate homology models of human COX-1 (hCOX-1) and COX-2 (hCOX-2). The human sequences were aligned to the highest resolution crystal structures available of either isoform.10,11 Thus, the template structures of ovine COX-1 (2.0 Å) and murine COX-2 (1.87 Å) were used. The sequence identity is 91% for ovine and human COX-1, and 87% for murine and human COX-2. The qualities of the generated models were assessed by visual comparison between the model and template structures, generating Ramachandran plots, and cal-culating heavy-atom RMSDs between the model and crystal structures.

1.2 Molecular mechanics and force fields Molecular mechanics (MM) is an important method for modelling and simu-lation of biomolecules. It is a simplified model for describing atoms and mol-ecules. Atoms follow the dynamics of classical mechanics. They are described as spheres with mass, van der Waals radii, and charges. Molecules are built by connecting atoms using harmonic or Morse potentials. Potential energy functions are then used to describe bonded and non-bonded interactions, such as bonds, angles, impropers, dihedrals, electrostatic, and Lennard-Jones inter-actions (Fig. 1).

14

Figure 1. Bonded and non-bonded interactions between atoms. Bond stretching (b), angle bending ( ), dihedral angle rotation ( ), improper dihedral bending ( ), Len-nard-Jones (A, B, r) and electrostatic (q, r) interactions.

The bonded terms describe the bond stretching, angle bending, improper di-hedral bending and dihedral rotation angles:

(1)

The non-bonded terms describe the electrostatic and van der Waals interac-tions:

(2)

Complete sets of the parameters in eq. 1 and 2 can be found in what are col-lectively called force fields, or MM force fields. Force fields are used to cal-culate the potential energy of the system. These are created by summarizing the bonded and non-bonded functions. The potential energies thus described by the force field correspond to the interactions between all atoms in any given system. This make them useful for scoring functions in docking programs, and for energetic evaluation of molecular simulations. In this thesis, the OPLS-AA force field12 was used in all simulations, and in some scoring functions used in molecular docking.

15

1.3 Molecular docking To estimate how well a drug would bind to its target molecule it is crucial to know where and how the drug binds. This can be done by manual docking using a molecular graphics program, or with varying degrees of automation using molecular docking programs. One of the advantages of molecular dock-ing programs is that they are quick. They can generate many binding modes for a single ligand in only a few seconds, and they can dock many ligands in minutes. Generally, the conformational search of each ligand is accompanied by rankings of the various poses. The program uses scoring functions to esti-mate the fit of each pose and suggest which one is the most probable candi-date. Thus, the docking programs have two important tasks to fulfill: confor-mational search of the ligand in the binding pocket and ranking the docking pose by estimating the corresponding interaction energies, using a scoring function.

In this thesis the two docking programs GOLD13 and GLIDE14 were used to dock inhibitors to the binding sites of hCOX-1 and hCOX-2.

1.3.1 GOLD GOLD13 is a docking program provided by the Cambridge Crystallographic Data Centre (CCDC). It uses a genetic search algorithm15 based on Mendelian genetics for finding possible binding modes. In the implementation of genetic algorithms to the docking problem, the process of evolution is mimicked by applying genetic operators to a pool of docking poses generated for the ligand, which are described by sets of 3D variables (“genes”). The first generation of conformations are generated and allowed to “evolve” by crossbreeding and mutation of the 3D variables (defining position, orientations, state of each ro-tatable bond, etc.). There is also the possibility of migrating an individual from one pool of conformations to another pool in order to introduce larger diver-sity. The fitness of the resulting docking poses is evaluated at each evolution-ary operation with a scoring function (see below). The new conformation will replace an earlier generation conformation in the pool of breeding confor-mations if the fitness improves. This cycle is then repeated until there is no more or only marginal improvement in fitness in the final generation of con-formations.

1.3.2 GLIDE

GLIDE14 is a docking program included in the Schrödinger Suite. It tries to mimic a full systematic search of the conformational, orientational, and posi-tional space of the docked ligand. This is done by using a series of hierarchical filters for determining the location of the ligand in the binding pocket. This is

16

followed by ligand torsion-angle minimization and a minimization of the lig-and inside the binding pocket using the OPLS-AA force field. The lowest en-ergy poses are then selected for a Monte Carlo procedure that examines nearby torsional minima. Finally, the poses are scored using a combination of a scor-ing function and the MM interaction energy, while taking into account the ligand strain energy.

1.4 Scoring functions Mathematical functions commonly known as scoring functions are used to quickly estimate the binding free energies of a given docking pose. The scor-ing functions can be categorized into three main groups: knowledge-based, empirical, and molecular mechanics force field scoring functions.16 Scoring functions only consider direct contact interactions between the ligand and the protein to save computational time (thus disregarding any long-range interac-tions). Furthermore, only a few scoring functions consider solvation effects and entropy effects, adding to their limitations for estimating accurate binding affinities.

1.4.1 Internal and external ranking Several different poses are generated for each docked molecule. Each pose is evaluated and scored using the selected scoring function. Thus, an internal ranking between all poses of the molecule in the site is created. The top ranked pose is the pose that the docking program suggests is the lowest energy (“the best”) pose of the molecule in that specific site. In theory, this would be the same pose the molecule would display in a crystal structure of the complex. In a similar way external rankings can be created from which the best binders can be determined. Thus, the highest scores obtained for each molecule can be compared instead of comparing different poses.

1.4.2 Force field scoring functions Force field scoring functions are based on or completely composed of molec-ular mechanics force fields such as AMBER17, CHARMM18, GROMOS19 and OPLS-AA12. Although many force field scoring functions are well parameter-ized, they are more computationally demanding than the knowledge-based, and empirical scoring functions. Thus, they are less likely to be used in dock-ing programs used in virtual screening. However, there are docking programs that have implemented “MM-like” scoring functions. One such example is GoldScore13, implemented in GOLD.

In GOLD up to version 5.0 the default scoring function is GoldScore, which uses only four terms to calculate the fitness score: ligand intramolecular (int),

17

and protein-ligand (ext) hydrogen-bond (hb), and van der Waals (vdw) score (S) according to the equation

GOLD Fitness=Shb_ext+ Svdw_ext+ Shb_int+Svdw_int (3)

1.4.3 Knowledge-based scoring functions Knowledge-based scoring functions are based on so called knowledge-based force fields and sometimes referred to as statistical potentials or potentials of mean force (PMF).20 These consist of several weighted molecular features re-lated to ligand-receptor binding modes. New protein-ligand complexes are as-sessed on the basis of similarities to the complexes found in the knowledge base, which is built on a large number of crystal structures of protein-ligand complexes. These kinds of functions therefor need to be constantly updated as the experimental database is expanded.

The Astex Statistical Potential (ASP) fitness function21 found in the GOLD docking program is a knowledge-based scoring function based on atom-atom distance potentials derived from a database of protein-ligand complexes found in protein data banks. ASP is the only one of its category used in this thesis, and also the scoring function with the weakest performance overall in ranking the docked poses and ligands to hCOX-1 among all the scoring functions used in this thesis.

1.4.4 Empirical scoring functions Empirical scoring functions use the sum of various empirical energy terms to approximate the binding energy. The majority of scoring functions used in this thesis are empirical scoring functions. These include ChemScore22 and ChemPLP23 in GOLD, and GlideScore SP14 and XP24 in GLIDE.

Chemscore is one of the oldest scoring functions still in use today. It is parameterized against known binding affinities. Chemscore includes terms for hydrogen bonding (Shbond), acceptor-metal (Smetal), and lipophilic (Slipo) inter-actions, while accounting for the loss of conformational entropy in the ligand upon binding to the protein (Hrot).

(4)

The ChemScore function implemented in GOLD adds extra terms in the form of a clash term, an intramolecular strain term, and a term for loss of covalent bonding:

(5)

18

ChemPLP uses the ChemScore hydrogen bonding term. It also uses several linear potentials (PLP, Piecewise Linear Potential) to model van der Waals and repulsive terms. It is a faster scoring function than GoldScore and is the current default scoring function in GOLD.

GLIDE has two scoring functions- GlideScore Standard Precision (SP) and GlideScore Extra Precision (XP). Both of these scoring functions are based on ChemScore. However, they add weighted components to the hydrogen-bond-ing term, a modification to the metal-ligand interaction term, and add solva-tion terms to account for solvation effects.14 The major difference between GlideScore SP and XP is that XP is a “harder” function. This means that it gives large penalties to poses that violate established physical chemistry prin-ciples, such as the lack of solvent exposure of charged or strongly polar groups. For GlideScore SP, the coefficients for these penalties are smaller, thus allowing ligands with a reasonable probability to bind in the active site to be docked, even if the structure is not optimally placed in the binding site.

1.5 Molecular dynamics simulations Although molecular docking is a quick and efficient method for estimating the binding mode of a ligand, there are more accurate methods for determining interaction energies. Molecular docking programs generally calculate their scores based on a single snapshot of the ligand-target complex. Thus, the re-sults are dependent on the target being in the right conformation, and that solv-ation is properly described in the system. However, this is not always the case. Flexible entities in the system, such as protein side-chains, could be displaced in such snapshots, and important water molecules could be entirely missing, causing less than optimal scores.

In molecular dynamics (MD) simulations, the system is simulated as a function of time. This approach results in an ensemble of thermally accessible configurations sampled according to the Boltzmann distribution, making it possible to obtain average potential energies. These are more comparable to experimentally measurable thermodynamic properties than potential energies of single snapshots.

In MD atoms are moved according to Newton’s laws of motion and the force of each atom i at the time t is calculated from the gradient ( ) of the force field potential energy function. The acceleration of each atom is calcu-lated using Newton’s second law (eq. 6) and the velocity and time can be ap-proximated from it using a truncated Taylor expansion.

ai t ∇iUpot (6)

19

The MD program Q25 is used for all MD simulations in this thesis. Q uses the Verlet leap-frog MD algorithm, with positions and velocities obtained from eq. 7 and 8 respectively.

(7)

vi t vi ai(t) (8)

Here, and are the position and velocity of the atom i, and is the time step. In Paper I the time step was consistently set to 1fs, but it is possible to also use a larger time step, as long as it is small enough to capture the fastest movements of the system.

1.5.1 Spherical boundary conditions It is important to consider what the most suitable system would be for studying the phenomena of interest when performing MD simulations. The most com-mon approach is to use so-called periodic boundary conditions (PBC). Here, the entire biological system of interest is placed in a box, which is periodically replicated in all directions of the space. Hence, as a molecule leaves the box through one of the walls of the box a new equivalent molecule will enter from the opposite wall to keep the number of molecules in the system constant. It is important that the box size is large enough to not only contain the biomole-cule of interest, but also to avoid artificial periodicity effects. PBC requires a large number of water molecules to avoid artificial distortions of the electric fields near the boundaries of the central image. However, this kind of setup is particularly useful e.g. to study the effect of large-scale conformational changes.

Another setup is possible: the spherical boundary conditions (SBC). Here, the boundary regions are constrained by using positional restraints while the water molecules used for solvating the system use polarization restraints sim-ilar to the SCAAS model in the boundary regions.26 The structure is solvated only in the regions of the protein where unrestrained MD simulations are ap-plied. Since the speed is quadratically dependent on the number of particles that need to be considered, the use of this model could reduce the simulation time compared to the equivalent simulations in larger boxes using PBC. The focus of the simulation sphere is on the binding site where the system is un-constrained, thus allowing free movement of solutes and solvent. At the edge of the sphere a half-harmonic potential is applied to keep the solvent mole-cules from leaving the sphere, thereby controlling the density of the solvent molecules inside the sphere in order to reproduce the density of bulk water. Since it is also possible to truncate systems using SBC simulation times can be altered by choosing a suitable sphere radius. Due to the lack of possible

20

conformational fluctuations distal to the binding site convergence could also be achieved faster.27 Thus, with a reduced system the electrostatic interactions could also in principle be considered explicitly. Though this is certainly the case for all ligand atoms, approximations such as the local reaction field28 to calculate the long-range interactions for other parts of the system contribute to further speed up the process.

In this thesis all MD simulations were carried out using the MD program Q, which is very well suited for ligand-binding simulations using SBC.

1.6 Linear interaction energy The linear interaction energy (LIE) method is a semi-empirical method used for calculating binding free energies.29 It uses the linear response approxima-tion (LRA), but adds a non-polar contribution term and an offset term. Here, one reference simulation is made of the ligand in a sphere of solvent (i.e. wa-ter). The second simulation has the ligand embedded in the solvated target structure. The sampled ligand-surrounding (l-s) energies of both states are then collected and the averages of the polar Ul-s

el and non-polar Ul-svdw inter-

actions are used as input in the LIE equation

(9)

1.6.1 The non-polar term The scaling factor for non-polar interaction energies, α, has an empirically deduced value of α = 0.18. It has been used in a wide variety of ligands and receptors with great success.

1.6.2 The polar term The scaling factor β scales the electrostatic polar interactions based on the LRA. 29,30

(10)

The LRA (eq. 10) is based on Zwanzigs formula31, which can be used for calculating free energies between two different states. In LIE, the two states are represented by the systems in which the electrostatic interactions are turned “off” and “on” respectively. In water it is possible to approximate the “off” term to 0 as the dipoles of the water molecules will orient themselves in a random manner around the ligand when the electrostatic interactions are turned off. In protein-ligand complexes the approximation has also been shown to be valid. However, it has also been demonstrated that in cases where

21

there is preorganization in the active site, this approximation is not completely accurate.32,33

In the first parameterization of the LIE method, β was set to 0.5. This is the theoretical exact value given from LRA for the binding of charged molecules. However, subsequent studies have shown that β will deviate from that value for uncharged molecules, and that β for such molecules will be dependent on the properties of the ligand.30,34,35 For uncharged ligands Hansson and co-workers30 proposed a classification based on the number of hydroxyl groups on the ligand as shown in Table 1.

Table 1. Classes of β proposed by Hansson and co-workers.

Class of compounds Number of OH-groups β

charged Not applicable 0.50

0 0.43

neutral 1 0.37 ≥ 2 0.33

There is also a model proposed by Almlöf and co-workers35 where more func-tional groups are included in the parameterization of β, as shown in Table 2.

Table 2. Classes of β proposed by Almlöf and co-workers.

1.6.3 The offset parameter γ The final constant, γ, can be used as an offset parameter to fix the absolute scale between the experimental and calculated binding free energies. γ is spe-cific to both the data set investigated and the conditions in the binding site, with a negative value indicative of a hydrophobic binding site.

1.6.4 The electrostatic correction term In our models, an electrostatic correction term36 was added to the LIE equation for charged ligands. This is only necessary when using SBC, since there are titratable residues manually set to the neutral (uncharged) state in this model.

Class of compounds Functional group β

Standard 0.43 Alcohols -0.06 1, 2-amines -0.04 1-amides -0.02 COOH -0.03 anions 0.02 cations 0.09

22

The correction term accounts for all long-range electrostatic interactions be-tween charged ligands and those residues that have been switched to their nat-ural states in the simulation setup. It corresponds to Coulomb’s law

(11)

where qp is the formal charge of the neutralized ionic group, ql is the partial charge of the ligand atom, and rpl is the distance between the ligand atom and the central atom in the ionic group. The effective dielectric constant of water (ε) was set to 80 in all papers included in this thesis.

1.7 Potential of mean force Sometimes it is useful to know how the free energies change as the system transitions from one state to another. The potential of mean force (PMF) is a graph of the free energies (ΔG) of a system as a function of a user-defined reaction coordinate. The system can be driven to explore all the configurations of interest in a relatively short amount of computational time using umbrella sampling. This is done by using biased potentials and restraining the configu-rations within the sampling windows to obtain sufficient sampling of these states. This way, statistically less probable states can be sampled for obtaining a continuous PMF. Unlike free energy perturbations—another method for transitioning between two states described in the next section—PMFs describe reaction paths between physical states.

In Papers II-IV, biased simulations were used to construct PMFs as cy-clooxygenases transitioned between two conformations, the closed and the open, driven by the rotation of an amino acid side-chain. Here, the Cγ−Cβ−Cα−N dihedral angle (χ1) of Leu531 (Fig. 2) was chosen as the reaction coordinate.

Figure 2. The Cγ−Cβ−Cα−N dihedral angle (χ1) of Leu531.

23

1.8 Free energy perturbation Long simulations would be needed to obtain reliable binding free energies of a ligand L going from free solution to being bound in a protein P. However, relative binding affinities can be calculated in a reasonable time by using the relationships obtained from the thermodynamic cycle in Fig. 3.

Figure 3. Thermodynamic cycle of two ligands (L and L’) binding to protein (P) to form ligand–protein complexes (PL and PL’).

Free energy perturbation (FEP) is a computational method used for introduc-ing point mutations into a system. It is then possible to calculate the relative free energy between the original and the mutated states. Since pressure is con-stant in biochemical processes such as ligand binding, and the PdV term is small, the distinction between Helmholtz and Gibbs free energy is disre-garded. Thus, the free energy difference (ΔG) between the two states A and B is calculated using the following perturbation formula

(12)

where k is Boltzmann’s constant, T is the temperature and UA and UB are the potentials used to describe states A and B.

The difference in potential energy of the two states can be evaluated by sampling the potentials of one of the states through MD or Monte Carlo (MC) simulations. Here, the configurations sampled on one of the potentials (UA) should also be probable on the other potential (UB), i.e. the potential energy difference between states A and B should not be too large, since this could cause large numerical errors. To estimate these errors and judge the conver-gence of the simulations the arbitrary labels of A and B are switched to give “forward” and “backward” calculations, which are then compared.

The potential energies of the two states of interest are usually too far apart in biological systems for the above equation to be useful and the generated numerical errors might dull the precision of the calculations. Thus, the trans-formation from one state to the other is done in a step-wise manner where the two states are connected through the introduction of intermediate alchemical

24

states. Potential energy functions of these intermediate states are linear com-binations of the potential energy functions representing the end points:

(13)

Here, m varies from 0 to 1 in n steps, creating n intermediate potential energy functions, which smoothly change the potential from state A to state B. The difference in the initial and final states can then be computed as the sum of the free energy differences of the intermediate states.

(14)

25

2. Computational protocols

Biological properties can be assessed and predicted by combining computa-tional and free energy methods. Computational methods can aid in the devel-opment of new pharmaceuticals, but how accurate are they?

Several computational methods and schemes have been applied in this the-sis. The quickest scheme consisted of fully automated molecular docking and scoring, while the most complex one combines docking with molecular dy-namics simulations and free energy calculations. Through these methods bind-ing affinities, modes, and mechanisms have been predicted. Where possible, these predictions have been compared with experimental data to assess the accuracy of the models and methods used.

2.1 Optimizing a protocol for calculating binding free energies In Paper I four computational schemes are presented for calculating binding affinities of a diverse set of ligands to COX-1. Each scheme starts with mo-lecular docking to the homology model hCOX-1 and the selection of initial ligand poses for the MD simulations. The interaction energies from these sim-ulations are then evaluated using the LIE method. The resulting binding free energies are then plotted against the experimental binding affinities. The var-ious schemes differ only in the selection of the best ligands pose, since this was identified as the limiting step for obtaining good estimates of binding af-finities. To design the optimal protocol several schemes were tested and eval-uated (Fig. 4).

The first scheme is a fully automated procedure in which docking programs are used for molecular docking and the built-in scoring functions decide which docked pose will be used as the initial ligand structure in MD simulations.

In the second scheme some manual intervention is used as the initial ligand structures for MD simulations are manually selected from the pool of docked ligands instead of relying on the ligand pose ranking of the scoring function.

26

Figure 4. A computational protocol combining molecular docking and scoring, mo-lecular dynamics simulations, and binding free energy calculations.

The third scheme is the scheme with most manual intervention as ligands used for MD simulations were manually docked, guided by pharmacophore knowledge gained from previous studies and the results of schemes 1 and 2. Here, the known salt-bridge interactions between carboxylated ligands and the Arg120 (Fig. 5) were prioritized among the possible ligand orientations. Fur-thermore, polar interactions with the Ser530 residue were also considered im-portant. For celecoxib and nimesulide analogues the side-pocket occupancy was identified as the most important feature. Thus, all such ligands were placed so as to occupy the side pocket.

27

Figure 5. Detailed view of the binding pocket in COX-1. The blue sphere highlights the side-pocket, which is occupied by a coxib (cyan) and gated by Ile523. The main pocket is occupied by a carboxylated inhibitor (yellow). The natural substrate ara-chidonic acid (magenta) is superimposed. Both carboxylated ligands are oriented to-wards the anchoring side-chain Arg120.

The fourth (and most successful) scheme makes use of all LIE binding affini-ties calculated from schemes 1-3. For each ligand in each of the schemes the simulation that yields the lowest binding free energy is selected and plotted against the experimental affinity. Where possible, the average structure from these simulations were compared with crystal structures (Table 3).

Table 3. Comparison of average structures from MD simulations and crystal struc-tures.

Average unsigned errors were calculated using the LIE binding free energies from each scheme. From the results shown in Table 4, a correlation can be observed between initial ligand orientation and average unsigned error be-tween the calculated and experimental affinities. As the percentage of ligands in the salt-bridge orientation is increased between Scheme 1 and Scheme 2, the corresponding average unsigned error decreases. Furthermore, increasing the manual intervention with regards to initial ligand pose for MD simulations

Ligand XRAY RMSDa ΔΔ b

Celecoxib 3KK6 0.62 1.39 Flurbiprofen 2AYL 0.70 -1.73 Ibuprofen 1EQG 1.44 0.37 Naproxen 3NT1 1.88 -1.74 Nimesulide 3N8X 1.79 0.09 Salicylic Acid 1PTH 1.12 -1.77 Diclofenac 3N8Y 5.98 1.77

28

improves the overall results. For instance, Scheme 3, which was the approach that had the most user intervention, performed significantly better than Scheme 1, which was fully automated in this regard. However, although it was also noted that docking scores were fairly insensitive to the changes in orien-tation, binding affinities calculated with the LIE method was particularly sen-sitive to the salt-bridge interactions facilitated by a ligand orientation in which the carboxylate head is oriented towards the residue Arg120 (Fig. 5).

Table 4. A summary of the statistical results of the computational schemes 1-4.

Scheme Scoring Function Salt-bridgea ± 2kcal/mol,b <ǀerrǀ>c

Scheme 1

ASP 42% 11% 9.50

ChemScore 65% 26% 7.01

GoldScore 77% 32% 5.93

ChemPLP 73% 29% 5.83

GlideScore SP 88% 21% 5.49

GlideScore XP 65% 24% 8.55

Scheme 2

ASP 77% 18% 6.62

ChemScore 100% 29% 3.99

GoldScore 100% 37% 4.39

ChemPLP 96% 37% 3.70

Scheme 3 100% 63% (69%) 2.06 (1.75)

Scheme 4 100% 92% (100%) 1.36 (1.01) a The percentage of carboxylated ligands docked in the salt-bridge orientation. b The percentage of ligands with calculated LIE binding affinities within 2kcal/mol of the experimental binding affinity. c The average unsigned error between the LIE binding affinities and the experimental binding affinities. Numbers in parenthesis calculated after removal of the outliers discussed in Paper I.

29

2.1.1 Performance of automated docking in COX-1 Two problems were identified for the automated molecular docking and scor-ing in COX-1. First, docking modes were often inconsistent among a series of structurally similar ligands. The top ranked pose was not always the lowest energy pose. Although the success-rate of finding low-energy poses differed between different scoring functions, none of the tested scoring functions would consistently rank the lowest energy pose as the top ranked pose for all tested ligands. This was most obvious among the carboxylated inhibitors, which could be placed in two major orientations (see chapter 4.1.1). However, for most scoring functions, low energy conformations could be found among the top ten ranked poses for almost all ligands.

Second, the scoring functions used in the study had problems ranking in-hibitors; they usually gave larger ligands a higher score than smaller ones, which was not consistent with experimental affinities. They were also unable to distinguish between good and bad binders when the molecular weight was not significantly affected by substitutions on the ligand. It has been shown in previous papers that COX-1 is a difficult drug target for screening with auto-mated docking.37,38 In Paper I the Pearson correlation coefficient between ex-perimental and calculated affinities for the scoring functions considered ranged between 0.1 and 0.4, further demonstrating that none of the scoring functions used in the study could be relied upon to create an accurate ranking of the 45 non-steroidal inflammatory drugs (NSAIDs) tested. With regards to the relative performance of each single scoring function considered, it is ob-vious from Table 4 that the least successful was the knowledge-based scoring function ASP. However, none of the scoring functions could be relied upon to consistently dock all ligands in low-energy docking poses in COX-1.

2.1.2 Evaluating the LIE method The shortcomings of the automated docking programs can be remedied using MD simulations and LIE calculations. Subjecting docked ligands to MD sim-ulations gives them the possibility to make minor or major adjustments in the cavity. They are thus able to make rotations or translational movements, which would position them in more energetically favourable binding poses. LIE can be used as a scoring function to determine which pose gives the highest bind-ing affinities. If available, these poses can be compared with crystal or NMR structures to validate the model (Table 3). Once the highest affinity pose for a ligand has been identified it can be compared to other inhibitors to identify the best one. Hence, inhibitors can be ranked and structure-activity relation-ships can be identified. This approach was used in all papers included in this thesis with good results. The correlation between experimental and computa-tional affinities was estimated by calculating the average unsigned errors. These ranged between 0.5−1.0 kcal mol-1.

30

To check the robustness of the parameterization, α was freely parameter-ized on the data set used in Paper I. The resulting value was 0.2, which sup-ports the applicability of using the standard empirical value of 0.18 for the investigated system.

The optimized parameterization for β by Hansson and Almlöf was also tested and compared. There was a small improvement in the average unsigned error of 0.1 kcal mol-1 when using the Almlöf parameterization. However, the small improvement could not be tied to any overall improvements of any spe-cific sub-class of ligands. Thus, due to the complexity of calculating β using the Almlöf parameterization, it did not seem optimal for such a diverse data set. Consequently, β was set to 0.5 for all ligands bearing a charge, following linear response theory, while β for the uncharged ligands was determined ac-cording to the Hansson parameterization for all studies in this thesis (Table 1). The statistical results improved by using different values of β for the un-charged ligands according to the classification in Table 1, compared to using only one standard value for the whole dataset of uncharged ligands. Consid-ering the large structural variance in the dataset used in the study, these results further demonstrates the robustness of this classification for β.

Finally, the addition of an electrostatic correction term to the calculated binding free energies improved the statistical results of the dataset by decreas-ing the average unsigned error by 0.1-0.4 kcal/mol.

2.2 Combining free energy methods

Figure 6. An extended computational protocol combining homology modelling, mo-lecular docking, MD simulations, and free energy calculations. These include the LIE method, PMF, and FEP calculations.

31

Additional insights into the systems can be obtained by adding more compu-tational and free energy methods. To understand the conformational prefer-ence of cyclooxygenases in their apo states, and in complex with various in-hibitors, PMF calculations were used in papers II-IV. To further validate LIE calculations, and to understand how amino acid mutations affect the binding affinities of some probed ligands FEPs were calculated in papers III and IV. In themselves, each method provides insight into one aspect of the ligand-enzyme complexes. At times, one method can be more useful than another, although both can be used in the same system. However, synergy effects can be obtained by combining them. For instance, FEP calculations could have been used to calculate binding affinities for data sets and subsets of structur-ally homologous inhibitors. However, in structurally diverse data sets, LIE calculations are more practical. If both are combined, the results from one method can be used to validate the results of the other, and contribute to a deeper understanding of the system.

32

3. Cyclooxygenases

3.1 Structure Cyclooxygenases, also called prostaglandin H2 synthase (PGHS), are homodi-meric hemeproteins with half-of-site activity (Fig. 7). This means that only one monomer is necessary for catalytic activity. Each monomer contains a cyclooxygenase and a peroxidase site. COX exists in several isoforms; the most researched ones are COX-1 and COX-2.39,40

Figure 7. Overview of the dimeric COX-1. The dotted line marks the center of the dimer. Colored helices mark the entrance to the binding pocket. Heme are repre-sented as sticks (magenta).

COX-1 is found in all cells in the endoplasmic reticulum while COX-2 is at-tached to the cell membrane. The sequence identity of these isoforms is 58%, with multiple substitutions in the binding and side-pockets. It has been esti-mated that a V523I substitution between the main pocket and the side-pocket (Fig. 8) makes the COX-2 binding pocket 25% larger than in COX-1.41 Ini-tially, it was believed that this substitution made it possible for inhibitors to bind in the side-pocket of COX-2, but not in that of COX-1. However, subse-quent crystal structures have proven that inhibitors can bind in the side-pocket of COX-1.42 Experimental mutational studies have since been conducted to

33

understand how this mutation affects inhibitor binding, and to explain isoform preferences of the inhibitors.43

Figure 8. Detailed view of the side and main pockets of human COX-1 (gray) and COX-2 (magenta). Residues in common are green. The open and closed confor-mations of Leu531 are highlighted in yellow and blue. Labelled residues are dis-cussed in the thesis.

In this thesis, human COX-1 and COX-2 were modelled by homology model-ling using crystal structures of ovine and murine COX as templates (PDB codes 1Q4G10 and 3NT111). The sequence identity of the human and ovine COX-1 is 91 % while it is 87 % between human and murine COX-2. Available crystal structures of either isoform have very similar structures with near-iden-tical secondary and tertiary structures. Although the whole dimers were mod-elled, only one monomer was included in the simulation sphere.

3.1.1 The closed and open conformations of COX Through computer simulations we confirmed two stable protein confor-mations of human COX-1 and COX-2 as described in papers II-IV. The con-formations are easily distinguishable by their differences in the rotational states of a leucine side-chain found in the main binding pocket (Fig. 9). The stabilities of these conformations are affected by the bound inhibitors. The closed conformation is the most stable conformation in the apo enzyme, and in complex with rapidly reversible, time-independent inhibitors. The open conformation is induced as time-dependent inhibitors bind to the enzyme. In COX-1 all probed slowly reversible inhibitors stabilized the open confor-mation. In COX-2 only some slowly reversible inhibitors did the same. How-ever, slow tight-binding inhibitors are both stabilising and stabilised by the open conformation in both isoforms.

34

Figure 9. The open and closed conformations of COX-1 (left) and COX-2 (right). The closed conformation has L531 pointing towards the binding pocket (grey, green) while the open conformation has the L531 side-chain parallel with the pocket (gold, beige).

Although the open conformation can be stabilized in both isoforms, there are structural differences between them. In COX-1, the open conformation in-cludes the rotation of the leucine side-chain, a shift in helix D, and water mol-ecules in the binding pocket. In COX-2 there is an additional displacement of Arg513 (Fig. 9), which is important for the binding of some COX-2 selective inhibitors. This selectivity is further discussed in chapter 4.2.2.

3.2 Function COX-1 is the constitutive isoform while COX-2 is induced during inflamma-tion. Inhibition of COX by traditional NSAIDs reduces pain, inflammations, and fevers, but causes side effects such as gastrointestinal bleeding and peptic ulcers. These side-effects are commonly attributed to COX-1 inhibition, while the pain-relieving function is believed to be the result of COX-2 inhibition. However, selective inhibition of COX-1 seems to inhibit thromboxane for-mation, which might have a protective effect on the heart. For instance, low-dose aspirin is a common prescription for patients at risk for thrombosis. It is a selective COX-1 inhibitor44 and one of the oldest traditional NSAIDs.

COX-1 also promotes the production of angiogenic growth factors and is overexpressed in ovarian cancers.45 Selective inhibition of COX-1 could therefore be used in therapies of certain cancer forms.

COX-2 can catalyse more substrates than COX-1. It is generally believed that COX-2 is the isoform responsible for pain inhibition by traditional NSAIDs. However, selective inhibition of this isoform does not result in the heart-protecting effects of aspirin. On the contrary, these inhibitors have been

35

linked to increases in myocardial infarctions. This has led to withdrawals of some COX-2 selective inhibitors from the market. However, these drugs have found an alternative market since it was discovered that COX-2 is overex-pressed in many cancer forms triggered by epidermic growth factor. These include colon, colorectal, pancreatic, and lung cancer.7,46-49

Thus, efficient inhibition, and particularly selective inhibition, of the two isoforms is useful. Therefore, in designing new anti-analgesic compounds considerations need to be made as to the binding of both COX-1 and COX-2. However, with a balanced binding profile to both enzymes better analgesics can be designed with fewer side-effects. New selective inhibitors for either isoform could also be potential cancer drugs.

3.2.1 Arachidonic acid and prostaglandins

Figure 10. Biochemical processes in cyclooxygenase and some products of the ara-chidonic acid cascade.

Pain, inflammation and fevers are common biological responses mediated by molecules called prostaglandins. In response to physical trauma and infec-tions, the cyclooxygenases (COX) catalyse the cyclooxygenation of fatty ac-ids called arachidonic acid (AA) into prostaglandin G2 (PGG2), and the sub-

36

sequent peroxidation of prostaglandin G2 into prostaglandin H2 (PGH2). Pros-taglandin H2 is further catalysed into a cascade of various prostaglandins, each responsible for different physiological responses (Fig. 10).

As an example, let us say that someone punches us in the arm. The imme-diate response is the feeling of pain, a response caused by nerve signals. A couple of minutes later, we are still in pain and the arm starts to bruise. These are the results of prostaglandins: in response to the punch prostaglandin E2 is produced. Among other things, it mediates pain and fever. Another prosta-glandin called thromboxane induces blood clotting (thrombosis) to prevent blood from leaking out of damaged blood vessels. A third type of prostaglan-din called prostacyclins, or prostaglandin I2, inhibits the platelet aggregation causing the blood clotting, thus preventing blood vessels from becomin com-pletely clogged up with clotted blood (infarctions).

Since the formation of these specialized prostaglandins (and their respec-tive physiological responses) start with the initial catalysis of prostaglandins from arachidonic acid, they can all be reduced or prevented by inhibition of the first catalytic steps, which happen in the cyclooxygenases.

37

4. Non-steroidal anti-inflammatory drugs

Non-steroidal anti-inflammatory drugs (NSAIDs) are COX inhibitors. They are common pharmaceuticals for treating pain, inflammation and fevers. Among the 10 most sold pharmaceuticals in the United States, there are two such inhibitors, the COX-2 selective Celebrex (celecoxib) and Vioxx (rofecoxib). Other well-known and commonly used NSAIDs are Aspirin (ac-etylsalicylic acid), ibuprofen, diclofenac, naproxen and indomethacin. Para-cetamol (or acetaminophen) is another common drug that inhibits pain and fevers. However, it is not considered an NSAID, due to the lack of anti-in-flammatory action.

4.1 Structural classes of NSAIDs NSAIDs can be grouped into structural classes. This classification of NSAIDs is common in the statistical tracking of NSAID drug use. Thus, we know from statistics reported by the Swedish pharmaceutical agency (Fig. 11) that the most commonly used NSAIDs in Sweden are derived from propionic acid. This structural class include the NSAIDs ibuprofen and naproxen (Fig. 12). Acetic acid derivatives, such as diclofenac, are also common. Second-gener-ation NSAIDs, such as celecoxib are becoming more popular, although their use is restricted due to the increased risks of myocardial infarctions.1

Figure 11. NSAID use in Sweden, divided by structural class.

38

Figure 12. Some NSAIDs probed in this thesis divided by structural class, except for the coxibs celecoxib and nimesulide, which are grouped together on the basis of their selectivity for COX-2.

39

4.1.1 Salt-bridge and inverted binding modes of carboxylated inhibitors Carboxylates are the most common scaffold seen in traditional NSAIDs. They mimic the AA carboxylate head for direct competitive inhibition of COX. The carboxylate head forms a strong salt-bridge with the Arg120 sidechain found at the entrance of the COX binding pockets (Fig. 13). Indeed, due to the hy-drophobic nature of the binding pocket all polar interactions are important. Specifically, the salt-bridge interaction was the most important ligand–en-zyme interaction for all carboxylated inhibitors in our data sets. Additional important polar interactions involve Tyr355, Ser530, and water molecules.

Figure 13. The main binding pocket of COX-1. Left: The salt-bridge (green) and the inverted (beige) orientations of diclofenac. Right: R-enantiomer (beige) and S-enan-tiomer (green) of flurbiprofen.

4.1.2 S- and R enantiomers of propionic acid derivatives The propionic acid derivatives are traditional NSAIDs. Although they are en-antiomeric it is usually only the S-enantiomer that is active.50 From MD sim-ulations and binding free energy calculations, we find that R-enantiomers have steric clashes between the methyl group of the inhibitor and Tyr355. For large and bulky ligands, this clash results in poor salt-bridge interactions between the ligand and the enzyme. In small and flexible ligands, it results in increased distances between the polar hydroxyl group of Tyr355 and the carboxylate. However, in the S-enantiomer the methyl group points in the other direction, which removes the steric clash, allowing the ligand to retain optimal salt-bridge interactions and to contribute to higher polar ligand–protein interac-tions.

40

4.1.3 Effects of amine substitutions on the carboxylate As mentioned in chapter 4.1.1, carboxylates are common functional groups in NSAIDs. Many high-affinity inhibitors are carboxylates, and in COX-1 most, if not all, known slow tight-binding inhibitors have the functional group. Sub-stitutions on the carboxylate result in the loss of time-dependency and a re-duction in affinities.

Figure 14. Four of the probed flurbiprofen analogues probed in paper III.

In paper III a series of flurbiprofen analogues were probed. These included substitutions of the fluorine atom and one of the carboxylate oxygens. The substitution on the carboxylate causes a destabilization of the inhibitor in the binding pocket (Fig. 14). This affects the ability of the inhibitor to stabilize the open conformation (see chapter 3.1.1), which abolishes the time-depend-ent feature of the original inhibitor. In the closed conformation, these inhibi-tors display lower affinities than in the open conformation. This is due to the loss of polar interactions, mainly with water molecules. For the amine and amide substituted inhibitors in the probed data set, the loss of polar interac-tions was the reason for decreased affinities. Some lost the important interac-tion with the Arg120 side-chain, while others lost interactions with the polar Ser530 hydroxy group.

41

4.1.4 Celecoxib analogues in the main and side-pocket

Figure 15. Two poses of celecoxib in the side pocket of closed (left) and open (right) COX-2.

Celecoxib and its analogues have been crystallized partially occupying the side-pocket in both COX-1 and COX-2.42,51 However, additional binding poses in which these inhibitors only occupy the main binding pocket were generated through automated docking. Energetic comparisons between these poses revealed that binding in the side-pocket was more favourable in both isoforms. In COX-2, these inhibitors induce a conformational change in the enzyme that affects the side-pocket. The conformational change induces an even higher binding affinity when these inhibitors are bound in the side-pocket.

4.2 Time-dependency and kinetic classes of NSAIDs Inhibition assays for COX have been conducted in various ways, which has resulted in large variances between different assays. Model systems vary, e.g. in/ex vivo (whole blood assays), microsomal assays, purified enzyme assays, and the tested species.52-54 Differences in pre-incubations times of the inhibi-tors are also common. These range from no pre-incubation (instant inhibition assays) up to an hour.54-59 It has been repeatedly demonstrated that for some inhibitors the pre-incubation time has a direct effect on the affinity of the in-hibitor, with an increase over time. Apart for aspirin, which is an irreversible inhibitor, this trend is seen among the slowly reversible and slow tight-binding inhibitors. The slow tight-binding inhibitors are rapidly reversible in the in-stant inhibition assays, but with increasing pre-incubation time they display slower reversibility until they are functionally irreversible. The phenomenon is well known since the earliest experimental assays conducted. 60,61 However,

42

the structural reasons for the time-dependent tight-binding in COX have been largely unknown. Since it is known that these inhibitors do not create covalent bonds with the protein, it has been speculated that the tight-binding is the re-sult of a conformational change. However, it is unknown what that conforma-tional change is.

Figure 16. A data set of NSAIDs colored based on their kinetic classes: rapidly re-versible (orange), two-step slow tight-binding (purple), slowly reversible (blue), ir-reversible (green), three-step slow tight-binding (red), and unknown class (gray).

Based on these observations, COX inhibitors have traditionally been catego-rized as either time-independent or time-dependent.60,61 A more rigorous clas-sification makes further distinctions in which five kinetic classes of NSAIDs have been identified (Fig. 16).62,63. These are: a) rapidly reversible, b) slowly reversible, c) two-step slow tight-binding, d) three-step slow tight-binding, and e) irreversible inhibitors. Except for the rapidly reversible inhibitors, all of these are time-dependent.

43

4.2.1 Time-dependency and tight-binding The high affinities associated with time-dependent inhibitors were replicated in COX-1 and COX-2 after modelling the Leu531 side chain (Fig. 8) in a novel conformation, termed open. PMF calculations showed that time-de-pendent inhibitors stabilize the open conformation while time-independent in-hibitors stabilize the intrinsic closed conformation observed in most crystal structures (Fig. 17).

Figure 17. Left: PMF of apo COX-1 (black), in complex with the rapidly reversible ibuprofen (beige), and the slow tight-binding flurbiprofen (red). Right: Correspond-ing PMF of COX-2.

Calculated binding affinities are consistently higher in the open than closed conformation for slow tight-binding inhibitors in both COX-1 and COX-2 (Fig. 18). For these inhibitors, there is a clear correlation between binding free energies in the open conformation and pre-incubated experimental assays. Similarly, binding in the closed conformation correlates with data from instant inhibition assays. These results indicate that the open conformation is the con-formation slow tight-binding inhibitors bind to after the time-dependent event.

Since tight-binding has been linked to time-dependency in COX ligands the two terms are often used interchangeably. However, although the two terms are obviously related, they are not synonymous. For instance, the group of slowly reversible inhibitors are time-dependent, but they are not tight-bind-ing. Flurbiprofen is a slow tight-binding inhibitor while piroxicam is slowly reversible. Both inhibitors stabilise the open conformation in both COX-1 and COX-2. However, to probe the time-dependency of COX inhibitors a series of flurbiprofen analogues of flurbiprofen were analysed. Consequently, we found that only the time-dependent inhibitors in the series were able to stabi-lise the open conformation of COX-1. The induced conformational changes create a very stable ligand-enzyme complex. Thus, the time-dependent event precedes or coincides with the high affinities indicative of tight-binding for these ligands.

44

Figure 18. Instant and pre-incubated inhibition compared with calculated binding free energies in the open and closed conformation of COX-1 for a series of inhibi-tors belonging to the classes of two-step slow tight-binding (Class I), rapidly reversi-ble (Class II), and slowly reversible inhibitors (Class III).

4.2.2 Time-dependency and selectivity The selectivity of NSAIDs towards the two COX isoforms has implications for their side-effects. It is believed that preference for COX-1 leads to the side-effects commonly associated with NSAIDs, such as gastrointestinal and kid-ney problems. On the other hand, COX-2 selective NSAIDs increase the risk of myocardiac infarctions, which has led to the withdrawal of such drugs from the market.

In paper IV, the binding of NSAIDs in COX-2 were probed to better un-derstand the underlying mechanisms of selectivity. It has been speculated that it is the mysterious properties associated with time-dependency that results in selectivity. Time-dependent inhibitors in COX-1 are also time-dependent in COX-2. However, the reverse is not always true. When the kinetic class of an inhibitor differs between the two isoforms it results in selectivity.

In COX-1 time-dependent slow tight-binding inhibitors bind through a two-step mechanism where the second step is very slowly reversible:

E + I EI EI* (15)

Although the class of inhibitors that bind according to eq. 15 in COX-1 bind in the same way in COX-2, they differ in affinities. For this class, selectivity towards either isoform depends on how the inhibitor makes use of the in-creased space of the binding pocket as Leu531 rotates into the open confor-mation (see chapter 3.1.1). For instance, small and flexible molecules will be preferential towards COX-1. With a smaller gate-keeping residue (Val523) compared to COX-1, these inhibitors have a higher mobility towards the side-pocket. This in turn can destabilize the open conformation, resulting in a rota-tion of the Leu531 side-chain.

45

Larger, flexible inhibitors of this class, such as diclofenac (Fig. 16), are preferential towards COX-2. Diclofenac can take advantage of the larger bind-ing pocket in COX-2, which gives it increased flexibility and the ability to create optimal interactions with the protein.

There are other inhibitors which are slow, tight-binding, but COX-2 selec-tive. These bind to COX-2 in a three-step mechanism:

E + I EI EI’ EI’’ (16)

The first and second steps are rapidly reversible while the third and final step is time-dependent. Since this final step is only slowly reversible, it leads to tight-binding inhibition of COX-2. Time-dependent selective COX-2 inhibi-tors bind in the side-pocket of both isoforms. However, the slowly, reversible step is not present in COX-1.

It has been hypothesized that the difference in binding between COX-1 and COX-2 for this class of ligands is due to the smaller binding pocket in COX-1 caused by the bulkier gate-keeping residue between the main and side-pock-ets (Fig. 8). However, although this substitution affects binding, we find that it is a combination of several residues that contribute to the time-dependent binding of coxibs to COX-2. More importantly, the previously described open and closed conformations of the enzyme affect the binding affinities. In the open conformation of COX-2 there are similar shifts as described in COX-1. However, with an additional displaced Arg513 side chain in the open confor-mation, COX-2 is able to interact optimally with the sulphonamide (or corre-sponding) group of celecoxib (Fig. 16) and its structural analogues. Since Arg513 is substituted for His513 in COX-1 (Fig. 8) the analogous interaction cannot occur there. Thus, we can propose the mechanisms involved in the three-step selective COX-2 inhibition (eq. 16) and explain why these inhibi-tors bind through a different mechanism in COX-1.

46

5. Future perspectives

In the field of NSAIDs, new drugs could be designed based on the open-closed model and the structure-activity relationships described in this thesis. Since the computational models of Papers I-IV have been extensively tested for the COX enzymes, they could be used when designing new molecules. By using variations of the computational methods or protocols from this thesis it should be straight-forward to design and probe new molecules against both COX isoforms. The computational screening could then identify potential leads that could be experimentally validated. Since simulations can be done relatively fast, money and time could be saved in experimental labs. This way, drugs with a range of different COX-1/COX-2 ratios could be designed and screened for ulcerogenic and thrombotic properties to pinpoint the window of ratios where side-effects are kept at a minimum. High-affinity selective inhibitors could also be designed to prevent angiogenesis, which is vital to tumor growth.

In the future, selectivity towards either isoform can be built into NSAIDs by using the known differences in binding pocket size and behaviour. Previous attempts have induced selectivity for COX-2 by simply adding bulky substi-tutions to inhibitors so they no longer fit in COX-1. However, these substitu-tions have mostly resulted in low-affinity binders of COX-2. With the knowledge of the open and closed conformations, new selective NSAIDs could be designed that take advantage of the mechanisms described. Aspirin, which is an irreversible inhibitor, acetylates Ser530. This is enough to prevent cyclooxygenation in COX-1 but not in COX-2. To create non-irreversible high-affinity COX-1-selective inhibitors, ligands need to be time-dependent in COX-1 while rapidly reversible in COX-2. Thus, small and flexible ligands could be designed that follow the reaction mechanism of eq. 15. These would have to be able to stabilise the open conformation of COX-1, but not COX-2.

To create COX-2-selective inhibitors the side-pocket is the obvious target. Much effort has been made to create inhibitors that bind there. However, since there is a risk of selective COX-2 inhibitors causing myocardiac infarctions it would be of interest to create less selective COX-2 inhibitors. Such drugs could be designed either by tuning down the interactions with the side-pocket, or by building upon the diclofenac, or related, scaffolds.

Although the work in this thesis contributes to the field, more work is still needed to fully understand the behavior of COX. Among other things, the class of slowly reversible inhibitors should be probed further to understand how different substructures in the group behave. The mechanisms behind the

47

half-of-site activity could be investigated. The differences in catalytic ability between the two isoforms needs to be understood. When these question marks have been solved, new ones are sure to emerge. In the meantime, drugs can be designed based on the ones we already have.

48

6. Populärvetenskaplig sammanfattning

Vad är egentligen skillnaden mellan en Treo, Ipren, Voltaren, Naproxen och en Alvedon? Vi använder dessa läkemedel mot olika smärttillstånd eller vid feber, men hur fungerar de egentligen och är det viktigt att veta det?

Av de nämnda läkemedlen ovan tillhör alla utom Alvedon gruppen ickeste-roida antiinflammatoriska läkemedel, eller NSAID. Dessa är febernedsät-tande, smärtstillande och, som namnet antyder, antiinflammatoriska läkeme-del som numera kan köpas både i apotek och i dagligvaruhandeln. Alvedon med det verksamma ämnet paracetamol (eller acetaminofen i USA) är både febernedsättande och smärtstillande, men inte antiinflammatorisk, vilket gör att den inte klassificeras som ett NSAID.

Alla NSAID har ungefär samma verkningsmekanism. I kroppens byggste-nar, cellerna, finns det en mängd olika molekyler och större molekylära ma-skiner som är livsviktiga för att vi ska må bra. Bland dessa finns de två prote-inerna, eller enzymerna, cyclooxygenas 1 och 2. Cyclooxygenas 1 är aktiva i alla celler medan cyclooxygenas 2 aktiveras vid inflammationer. NSAID ver-kar genom att ta sig in i dessa proteiner och stannar där olika länge beroende på läkemedelssubstansen. På så vis hämmas funktionen av dessa två proteiner. När dessa proteiner hämmas kan de inte längre tillverka de molekyler som senare kommer att ombildas till andra molekyler, som i sin tur signalerar krop-pen att ge symptomen för bland annat feber, inflammation och smärta. Ef-tersom NSAID tävlar med de molekyler som vanligtvis går in i enzymen, de naturliga substraten, kommer effektiviteten av dessa läkemedel att bero på hur länge de klarar av att stanna i enzymen. Om de endast stannar en kort stund kommer man behöva mer läkemedel för att effektivt hålla de naturliga sub-straten borta från enzymen. Stannar de däremot väldigt länge behövs det mindre mängder, vilket är bättre för kroppen då mindre mängder läkemedel oftare leder till mindre biverkningar. Det är främst dessa skillnader som finns mellan de NSAID som nämndes i den första paragrafen. För att tillverka nya och bättre läkemedel behöver man därför veta vad det är som gör att dessa läkemedel stannar länge i enzymet för att därigenom kunna bygga de mest funktionella och effektiva läkemedlen.

I den här avhandlingen har jag tittat på hur existerande NSAID binder till COX-1 och COX-2 genom datorsimuleringar och fria-energiberäkningar. Därigenom har jag kunnat identifiera olika strukturella egenskaper i dessa mo-lekyler samt några mekanismer som tillsammans påverkar hur starkt olika NSAID binder till de olika proteinerna.

49

Genom att kombinera tillgängliga datorsimuleringstekniker och beräk-ningsmetoder med kunskaper i biokemi och strukturbiologi har vi fått större möjligheter att förstå hur biomolekyler fungerar, ner på atomnivå. Med denna kunskap kan vi i nästa steg utveckla framtidens läkemedel, inte bara mot smärta, feber och inflammation, utan även för att fördröja tumörbildning, vil-ket är en viktig del i utvecklingen av cancer. De beräkningsmetoder som har använts i denna avhandling kan även med fördel användas i utvecklandet av läkemedel som verkar mot andra sjukdomar. Jag hoppas därmed att denna av-handling kommer vara till nytta för många i framtiden och tacka för att du tog dig tid att läsa den.

50

Acknowledgements

To past and present members of the Åqvist group: Thank you to Professor Johan Åqvist for accepting me into the group and

giving me the opportunity to follow through. Assoc. Prof. Hugo Gutiérrez de Terán for signing up as my assistant supervisor and thoroughly checking my work. Göran Wallin for introducing me to the group and the field. Assoc. Prof. Jens Carlsson for teaching me the tricks of the trade and setting me on this path. Dr. Martin Andér for teaching me how to parameterize things in ways Macromodel couldn’t. Dr. Johan Sund for teaching me how to do molecular docking. Dr. Lars Boukharta for teaching me how to do homology modeling. Dr. Stefan Trobro for answering my questions about everything and nothing. Dr. Henrik Keränen for teaching me how to do step-wise free energy pertur-bations, and for all the conversations, advice, and that time we shared an of-fice. Dr. Masoud Kazemi for teaching me how to do potentials of mean force calculations. Sudarsan Reddy Vanga for introducing me to IRAP and navi-gating me through that field. Dr. Ana L. Oliveira for sharing office, gym time, gossip and cakes with me. Dr. Mauricio Esguerra for teaching me how to use R and for the computer-magic you did for me. Dr. Priyadarshi Satpati, Chris-toffer Lind, Dr. Jessica Sallander, Silvana Vasile, Jaka Socan, Willem Jespers and Dr. Alexei Siretskiy for all the lunches, fika and conversations. To the people at XRAY, CSB, and ICM:

To TEX, for proofreading, listening, great conversation and advice. Prof. Alwyn T Jones, Prof. Stefan Knight, Dr. Christoffer Björklid, Dr. Xiao Huo and Annika Söderholm for all the important discussions on crystal structures and crystallography. To the people at CSB, especially Dr. Daniel Larsson for introducing me to GROMACS. Dr. Fernanda Duarte for all the support, con-versations, conferences, and the computational science group for women. To professor David van der Spoel, Dr. Nina Fischer, and Muhammad Ghahremanpour: thanks for the lunches, conversations, and support. To Erling for Kingkong, licenses and storage solutions. Prof. Lynn Kamerlin for valua-ble conversations and advice.

To the people I have had fika and conversations with through the years. To the Malay community:

To Kak Leni & Nazli, Kakja & Ipen, Kak Linda & Abe Rahim, Kak Saz & Zikri, Kak Ami, Wan Salman, Mona & Duen, Wan and Poje, Kak Nurul &

51

Abg Long, Shimah & Kamarul, Siti, Fadhli & Alia, Abe Shahar & Sikin, Nasir & An, and Shaifful. To Hanim and all the rest of the Malay-speaking commu-nity- thanks for introducing a bit of Asia to this otherwise cold country. Finally, to my Family. Thanks for Everything, and more; past, present, future, always.

52

References

(1) Socialstyrelsens statistikdatabas. Socialstyrelsen. http://www.socialstyrel-sen.se/statistik/statistikdatabas/lakemedel

(2) E-hälsomyndighetens läkemedelsstatistik för egenvård. https://www.ehalsomyn-digheten.se/tjanster/lakemedelsstatistik/sok-statistik-for-egenvard/

(3) Steinmeyer, J. (2000) Pharmacological basis for the therapy of pain and inflam-mation with nonsteroidal anti-inflammatory drugs. Arthritis Res. 2, 379–385.

(4) Winkelmayer, W. C., Waikar, S. S., and Mogun, H. (2008) Nonselective and cy-clooxygenase-2-selective NSAIDs and acute kidney injury. Am. J. Med. 121,1092–1098.

(5) Schneider, V., Lévesque, L. E., Zhang, B., Hutchinson, T., and Brophy, J. M.(2006) Association of selective and conventional nonsteroidal antiinflammatorydrugs with acute renal failure: A population-based, nested case-control analysis.Am. J. Epidemiol. 164, 881–889.

(6) European Medical Agency. http://www.ema.europa.eu/ema/index.jsp?curl=pages/news_and_events/news/2012/10/news_detail_001637.jsp&mid=WC0b01ac058004d5c1

(7) Bresalier, R. S., Sandler, R. S., Quan, H., Bolognese, J. A., Stat, M., Oxenius, B.,Horgan, K., Lines, C., Riddell, R., Morton, D., Lanas, A., Konstam, M. A., andBaron, J. A. (2006) Cardiovascular events associated with rofecoxib in a colorec-tal adenoma chemoprevention trial. N. Engl. J. Med. 355, 221–221.

(8) Chothia, C., and Lesk, A. M. (1986) The relation between the divergence of se-quence and structure in proteins. The EMBO Journal 5, 823–826.

(9) Eswar, N., Webb, B., Marti-Renom, M. A., Madhusudhan, M. S., Eramian, D.,Shen, M.-Y., Pieper, U., and Sali, A. (2006) Comparative protein structure mod-eling using Modeller. Curr Protoc Bioinformatics Chapter 5, Unit–5.6.

(10) Gupta, K., Selinsky, B. S., Kaub, C. J., and Katz, A. K. (2004) The 2.0 Å Reso-lution Crystal Structure of Prostaglandin H 2 Synthase-1: Structural Insights intoan Unusual Peroxidase. J. Mol. Biol. 335, 503–518.

(11) Duggan, K. C., Walters, M. J., Musee, J., Harp, J. M., Kiefer, J. R., Oates, J. A.,and Marnett, L. J. (2010) Molecular Basis for Cyclooxygenase Inhibition by theNon-steroidal Anti-inflammatory Drug Naproxen. J. Biol. Chem. 285, 34950–34959.

(12) Jorgensen, W. L., Maxwell, D. S., and Tirado-Rives, J. (1996) Development andTesting of the OPLS All-Atom Force Field on Conformational Energetics andProperties of Organic Liquids. J. Am. Chem. Soc. 118, 11225–11236.

(13) Verdonk, M. L., Cole, J. C., and Hartshorn, M. J. (2003) Improved protein–liganddocking using GOLD. Proteins: Struct., Funct., Genet. 52, 609–623.

(14) Friesner, R. A., Banks, J. L., Murphy, R. B., Halgren, T. A., Klicic, J. J., Mainz,D. T., Repasky, M. P., Knoll, E. H., Shelley, M., Perry, J. K., Shaw, D. E., Fran-cis, P., and Shenkin, P. S. (2004) Glide: A new approach for rapid, accuratedocking and scoring. 1. Method and assessment of docking accuracy. J. Med.Chem. 47, 1739–1749.

53

(15) Jones, G., Willett, P., Glen, R. C., Leach, A. R., and Taylor, R. (1997) Develop-ment and validation of a genetic algorithm for flexible docking. J. Mol. Biol. 267, 727–748.

(16) McInnes, C. (2007) Virtual screening strategies in drug discovery. Curr. Opin. Chem. Biol. 11, 494–502.

(17) Cornell, W. D., Cieplak, P., Bayly, C. I., Gould, I. R., Merz, K. M., Ferguson, D. M., Spellmeyer, D. C., Fox, T., Caldwell, J. W., and Kollman, P. A. (1995) A second generation force field for the simulation of proteins, nucleic acids, and organic molecules. J. Am. Chem. Soc. 117, 5179–5197.

(18) MacKerell, A. D., Bashford, D., Bellott, Dunbrack, R. L., 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., Roux, B., Schlenkrich, M., Smith, J. C., Stote, R., Straub, J., Watanabe, M., Wiórkiewicz-Kuczera, J., Yin, D., and Karplus, M. (1998) All-atom empirical potential for molecular modeling and dynamics stud-ies of proteins. J. Phys. Chem. B 102, 3586–3616.

(19) Kräutler, V., and Oostenbrink, C. (2005) The GROMOS software for biomolec-ular simulation: GROMOS05. J. Comput. Chem. 26, 1719–1751.

(20) Sippl, M. J. (1995) Knowledge-based potentials for proteins. Curr. Opin. Struct. Biol. 5, 229–235.

(21) Mooij, W. T. M., and Verdonk, M. L. (2005) General and targeted statistical po-tentials for protein-ligand interactions. Proteins 61, 272–287.

(22) Eldridge, M. D., Murray, C. W., Auton, T. R., Paolini, G. V., and Mee, R. P. (1997) Empirical scoring functions: I. The development of a fast empirical scor-ing function to estimate the binding affinity of ligands in receptor complexes. J. Comput. Aided. Mol. Des. 11, 425–445.

(23) Korb, O., Stützle, T., and Exner, T. E. (2009) Empirical Scoring Functions for Advanced Protein−Ligand Docking with PLANTS. J. Chem. Inf. Model. 49, 84–96.

(24) Friesner, R. A., Murphy, R. B., Repasky, M. P., Frye, L. L., Greenwood, J. R., Halgren, T. A., Sanschagrin, P. C., and Mainz, D. T. (2006) Extra Precision Glide: Docking and Scoring Incorporating a Model of Hydrophobic Enclosure for Protein−Ligand Complexes. J. Med. Chem. 49, 6177–6196.

(25) Marelius, J., Kolmodin, K., Feierberg, I., and Åqvist, J. (1998) Q: a molecular dynamics program for free energy calculations and empirical valence bond sim-ulations in biomolecular systems. J. Mol. Graphics Mod. 16, 213–225.

(26) King, G., and Warshel, A. (1989) A surface constrained all-atom solvent model for effective simulations of polar solutions 91, 3647–3661.

(27) Bjelic, S., Brandsdal, B. O., and Åqvist, J. (2008) Cold adaptation of enzyme reaction rates. Biochemistry 47, 10049–10057.

(28) Lee, F. S., and Warshel, A. (1992) A local reaction field method for fast evalua-tion of long-range electrostatic interactions in molecular simulations. J. Chem. Phys. 97, 3100–3107.

(29) Åqvist, J., Medina, C., and Samuelsson, J.-E. (1994) A new method for predicting binding affinity in computer-aided drug design. Protein. Eng. Des. Sel. 7, 385–391.

(30) Hansson, T., Marelius, J., and Åqvist, J. (1998) Ligand binding affinity prediction by linear interaction energy methods. J. Comput. Aided. Mol. Des. 12, 27–35.

(31) Zwanzig, R. W. (1954) High-Temperature Equation of State by a Perturbation Method. I. Nonpolar Gases. J. Chem. Phys. 22, 1420–1426.

54

(32) Almlöf, M., Åqvist, J., Smalås, A. O., and Brandsdal, B. O. (2006) Probing the Effect of Point Mutations at Protein-Protein Interfaces with Free Energy Calcu-lations. Biophys. J. 90, 433–442.

(33) Mishra, S. K., Sund, J., Åqvist, J., and Koča, J. (2012) Computational prediction of monosaccharide binding free energies to lectins with linear interaction energy models. J. Comput. Chem. 33, 2340–2350.

(34) Åqvist, J., and Hansson, T. (1996) On the validity of electrostatic linear response in polar solvents. J. Phys. Chem. 100, 9512–9521.

(35) Almlöf, M., Carlsson, J., and Åqvist, J. (2007) Improving the accuracy of the linear interaction energy method for solvation free energies. J. Chem. Theory Comput. 3, 2162–2175.

(36) Åqvist, J. (1996) Calculation of absolute binding free energies for charged lig-ands and effects of long-range electrostatic interactions. J. Comput. Chem. 17, 1587–1597.

(37) Ravitz, O., Zsoldos, Z., and Simon, A. (2011) Improving molecular docking through eHiTS’ tunable scoring function. J. Comput. Aided. Mol. Des. 25, 1033–1051.

(38) Huang, N., Shoichet, B. K., and Irwin, J. J. (2006) Benchmarking Sets for Mo-lecular Docking. J. Med. Chem. 49, 6789–6801.

(39) Picot, D., Loll, P. J., and Garavito, R. M. (1994) The X-ray crystal structure of the membrane protein prostaglandin H2 synthase-1. Nature 367, 243–249.

(40) Kurumbail, R. G., Stevens, A. M., Gierse, J. K., and McDonald, J. J. (1996) Struc-tural basis for selective inhibition of cyclooxygenase-2 by anti-inflammatory agents. Nature 384, 644–648.

(41) Luong, C., Miller, A., Barnett, J., Chow, J., Ramesha, C., and Browner, M. F. (1996) Flexibility of the NSAID binding site in the structure of human cycloox-ygenase-2. Nat. Struct. Mol. Biol. 3, 927–933.

(42) Rimon, G., Sidhu, R. S., Lauver, D. A., Lee, J. Y., Sharma, N. P., Yuan, C., Frieler, R. A., Trievel, R. C., Lucchesi, B. R., and Smith, W. L. (2010) Coxibs interfere with the action of aspirin by binding tightly to one monomer of cycloox-ygenase-1. Proc. Natl. Acad. Sci. U.S.A 107, 28–33.

(43) Gierse, J. K., McDonald, J. J., Hauser, S. D., Rangwala, S. H., Koboldt, C. M., and Seibert, K. (1996) A single amino acid difference between cyclooxygenase-1 (COX-1) and -2 (COX-2) reverses the selectivity of COX-2 specific inhibitors. J. Biol. Chem. 271, 15810–15814.

(44) Smith, J. B., and Willis, A. L. (1971) Aspirin selectively inhibits prostaglandin production in human platelets. Nature New Biol. 231, 235–237.

(45) Gupta, R. A., Tejada, L. V., Tong, B. J., Das, S. K., Morrow, J. D., Dey, S. K., and DuBois, R. N. (2003) Cyclooxygenase-1 is overexpressed and promotes an-giogenic growth factor production in ovarian cancer. Cancer Res. 63, 906–911.

(46) Solomon, S. D., McMurray, J., and Pfeffer, M. A. (2005) Cardiovascular risk associated with celecoxib in a clinical trial for colorectal adenoma prevention. N. Engl. J. Med. 352, 1071–1080.

(47) Gupta, R. A., and DuBois, R. N. (2001) Colorectal cancer prevention and treat-ment by inhibition of cyclooxygenase-2. Nat. Rev. Cancer 1, 11–21.

(48) Tucker, O. N., Dannenberg, A. J., Yang, E. K., Zhang, F., Teng, L., Daly, J. M., Soslow, R. A., Masferrer, J. L., Woerner, B. M., Koki, A. T., and Fahey, T. J. (1999) Cyclooxygenase-2 expression is up-regulated in human pancreatic cancer. Cancer Res. 59, 987–990.

(49) Sano, H., Kawahito, Y., Wilder, R. L., Hashiramoto, A., Mukai, S., Asai, K., Ki-mura, S., Kato, H., Kondo, M., and Hla, T. (1995) Expression of cyclooxygenase-1 and -2 in human colorectal cancer. Cancer Res. 55, 3785–3789.

55

(50) Knihinicki, R. D., Williams, K. M., and Day, R. O. (1989) Chiral inversion of 2-arylpropionic acid non-steroidal anti-inflammatory drugs—1: In vitro studies ofibuprofen and flurbiprofen. Biochem. Pharmacol. 38, 4389–4395.

(51) Orlando, B. J., and Malkowski, M. G. (2016) Crystal structure of rofecoxib boundto human cyclooxygenase-2. Acta Crystallogr F Struct Biol Commun 72, 772–776.

(52) Riendeau, D., Charleson, S., Cromlish, W., Mancini, J. A., Wong, E., and Guay,J. (1997) Comparison of the cyclooxygenase-1 inhibitory properties of nonsteroi-dal anti-inflammatory drugs (NSAIDs) and selective COX-2 inhibitors, usingsensitive microsomal and platelet assays. Can. J. Physiol. Pharmacol. 75, 1088–1095.

(53) Kato, M., Nishida, S., Kitasato, H., Sakata, N., and Kawai, S. (2001) Cyclooxy-genase-1 and cyclooxygenase-2 selectivity of non-steroidal anti-inflammatorydrugs: investigation using human peripheral monocytes. J. Pharm. Pharmacol.53, 1679–1685.

(54) Warner, T. D., Giuliano, F., Vojnovic, I., Bukasa, A., Mitchell, J. A., and Vane,J. R. (1999) Nonsteroid drug selectivities for cyclo-oxygenase-1 rather than cy-clo-oxygenase-2 are associated with human gastrointestinal toxicity: A full invitro analysis. Proc. Natl. Acad. Sci. U.S.A 96, 7563–7568.

(55) Laneuville, O., Breuer, D. K., Dewitt, D. L., Hla, T., Funk, C. D., and Smith, W.L. (1994) Differential inhibition of human prostaglandin endoperoxide H syn-thases-1 and-2 by nonsteroidal anti-inflammatory drugs. J. Pharmacol. Exp.Ther. 271, 927–934.

(56) O'Neill, G. P., Mancini, J. A., Kargman, S., Yergey, J., Kwan, M. Y., Falgueyret,J. P., Abramovitz, M., Kennedy, B. P., Ouellet, M., and Cromlish, W. (1993)Overexpression of human prostaglandin G/H synthase-1 and -2 by recombinantvaccinia virus: inhibition by nonsteroidal anti-inflammatory drugs and biosynthe-sis of 15-hydroxyeicosatetraenoic acid. Mol. Pharmacol. 45, 245–254.

(57) Brideau, C., Kargman, S., Liu, S., Dallob, A. L., Ehrich, E. W., Rodger, I. W.,and Chan, C. C. (1996) A human whole blood assay for clinical evaluation ofbiochemical efficacy of cyclooxygenase inhibitors. Inflamm Res 45, 68–74.

(58) Glaser, K., Sung, M.-L., O'Neill, K., Belfast, M., Hartman, D., Carlson, R., Kreft,A., Kubrak, D., Hsiao, C.-L., and Weichman, B. (1995) Etodolac selectively in-hibits human prostaglandin G/H synthase 2 (PGHS-2) versus human PGHS-1.Eur. J. Pharmacol. 281, 107–111.

(59) Cryer, B., and Feldman, M. (1998) Cyclooxygenase-1 and Cyclooxygenase-2 Se-lectivity of Widely Used Nonsteroidal Anti-Inflammatory Drugs. Am. J. Med.104, 413–421.

(60) Smith, W. L., and Lands, W. E. (1971) Stimulation and blockade of prostaglandinbiosynthesis. J. Biol. Chem. 246, 6700–6702.

(61) Rome, L. H., and Lands, W. E. (1975) Structural requirements for time-dependentinhibition of prostaglandin biosynthesis by anti-inflammatory drugs. Proc. Natl.Acad. Sci. U.S.A 72, 4863–4865.

(62) Gierse, J. K., Koboldt, C. M., Walker, M. C., Seibert, K., and Isakson, P. C.(1999) Kinetic basis for selective inhibition of cyclo-oxygenases. Biochem. J. 339( Pt 3), 607–614.

(63) Walker, M. C., Kurumbail, R. G., Kiefer, J. R., Moreland, K. T., Koboldt, C. M.,Isakson, P. C., Seibert, K., and Gierse, J. K. (2001) A three-step kinetic mecha-nism for selective inhibition of cyclo-oxygenase-2 by diarylheterocyclic inhibi-tors. Biochem. J. 357, 709–718.

Acta Universitatis UpsaliensisDigital Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Science and Technology 1560

Editor: The Dean of the Faculty of Science and Technology

A doctoral dissertation from the Faculty of Science andTechnology, Uppsala University, is usually a summary of anumber of papers. A few copies of the complete dissertationare kept at major Swedish research libraries, while thesummary alone is distributed internationally throughthe series Digital Comprehensive Summaries of UppsalaDissertations from the Faculty of Science and Technology.(Prior to January, 2005, the series was published under thetitle “Comprehensive Summaries of Uppsala Dissertationsfrom the Faculty of Science and Technology”.)

Distribution: publications.uu.seurn:nbn:se:uu:diva-328478

ACTAUNIVERSITATIS

UPSALIENSISUPPSALA

2017