Work leading up to Structure-Based Drug Design of Spermidine

Get Complete Project Material File(s) Now! »

Chapter 2 Work leading up to Structure-Based Drug Design of Spermidine Synthase: Structural and Mechanistic Insights into the Action of Plasmodium falciparum Spermidine Synthase

Introduction

Signi cant e orts have been invested over the last decade to increase the number of available 3D protein structures, including projects such as for example the Protein Struc-ture Initiative (PSI; http://www.structuralgenomics.org/) and the Structure Ge-nomics Consortium (SCG; http://www.sgc.utoronto.ca). Despite these e orts many therapeutically relevant structures remain unresolved (Nayeem et al., 2006). The time required for solving an average eukaryotic structure is estimated to take between one and three years and includes both protein expression, crystallization and solving the struc-ture. This time is thought to increase signi cantly with high risk targets such as viruses, molecular machinery and membrane proteins (Stevens, 2004). It is estimated that the use of 3D structures in the early drug discovery process for nding high quality leads could save up to 50% of the total costs (Stevens, 2004). Solving of malaria protein structures is notoriously di cult especially in P. falciparum with the main reason being the high A + T composition of 80% in coding regions (Gardner et al., 2002). The A + T composition can rise to as high as 90% in the intron and intergenic regions (Gardner et al., 2002). This A + T bias in the genome composition alters the codon usage and therefore makes it very di cult to express malarial proteins at su cient quantities for crystallization in most heterologous systems (Withers-Martinez et al., 1999). This challenge is currently being addressed by codon usage optimization and harmonization. Another factor complicating the expression of malarial proteins is the presence of low complexity segments and/or inserts. This can best be described as unstructured protein regions containing only a subset of the 20 amino acids, sometimes occurring as tandem repeat arrays (Xue and Forsdyke, 2003; Birkholtz et al., 2008b).
Protein structure prediction (i.e. homology modeling) can be seen as an alternative method to get structural information where experimental techniques have failed (Nayeem et al., 2006). Three widely accepted methods exist which can be used to predict 3D pro-tein structures from the amino acid sequence including comparative modeling, threading and ab initio prediction (Bourne and Wiessig, 2003). Comparative modeling is the most common method used and utilizes predetermined protein structures to predict the pro-tein conformation of unresolved protein structures with a similar amino acid sequence (Hillisch et al., 2004). The comparative methods are based on the assumption that the 3D structures of proteins are more conserved than their primary structures or amino acid sequences (Chothia and Lesk, 1986). In other words proteins with a certain degree of evolutionary-related sequence identity, have similar structures. Threading is also known as fold recognition and entails that a protein sequence be scored against a 3D database of known protein folds. The best- t structure is assumed to be the fold adopted by the protein sequence. This can then be used to nd the structure of the protein of interest. Bourne and Wiessig (2003) described ab initio structure prediction as a mixture between science and engineering with no reliable methods at present. The science part can be described as how the 3D structure of a protein is attained and the engineering part the deduction of a 3D structure from a given sequence. Homology modeling is currently the most reliable and mature of the three methods and hence was used in this study.
The uses of homology models are versatile and models can be used in nearly all stages of the drug design process (Figure 2.1). The quality of models derived from homology modeling is dependent on the relationship between the protein sequence identities of the known structure and the target protein to be modeled (Chothia and Lesk, 1986). Homologous proteins can be described as proteins having a common ancestor and thus an evolutionary relationship with a high probability of sharing a common 3D structure. Homologous proteins with a sequence identity of 30% and higher can often be used to construct reliable homology models. The relationship between the target and template sequence identity and the information content of the resulting homology models is pre-sented in Figure 2.2. Homology models based on a sequence identity of lower than 15% are thought to be speculative and could lead to misleading results. Models with a sequence identity of between 15 and 30% can be used in functional assignment of the target protein and to direct mutagenesis studies. However, sophisticated pro le-based methods should be used for alignment of the target and template sequences. Proteins having between 30 and 50% sequence identity can be used in structure-based methods to predict the target druggability and the design of mutagenesis experiments and in vitro test assays. Hillisch et al. (2004) proposed that in the presence of a 50% or higher sequence identity between the target and template, the resulting models are frequently of a su cient quality to be used in the prediction of protein-ligand interactions. This includes structure-based drug design and the prediction of preferred sites of metabolism for small molecules.
The homology modeling process usually comprises of four steps which include fold assignment, sequence structure alignment, model building and model re nement (Figure 2.3). The rst step, fold assignment, entails the template recognition and initial align-ment. The second step involves the crucial step of sequence alignment. This is followed by model building which includes backbone generation, loop modeling and side-chain modeling. The nal step involves the re nement or optimization of these models.
There are various software solutions which can be used to construct homology models including programs such as WHATIF (Vriend, 1990), MODELLER (Sali and Blundell, 1993), Swiss-Model (Gasteiger et al., 2003), Pro t (Martin, A.C.R. http://www.bioinf. org.uk/software/profit/), Prime (http://www.schrodinger.com), MOE (www.chemcomp. com) etc. MODELLER is a comparative protein modeling package which makes use of the satisfaction of spatial restraints to construct homology models and was used in this study. MODELLER proceeds by performing three steps to build a homology model (Sali and Blundell, 1993). These three steps include the alignment of target sequence with the template sequence, the extraction of spatial restraints and lastly the satisfaction of the spatial restraints. The protein sequence alignment is used to extract the distance and dihedral angles restraints for the model and is expressed as a probability density function which is used to describe the spatial restraints (Sali and Blundell, 1993). These restraints and energy terms which enforce proper stereochemistry are combined in an objective function. Optimization of the nal model is performed by using methods of conjugate gradients and molecular dynamics with simulated annealing (Sali et al., 1995).
The use of comparative protein modeling has been shown to be suitable for the prediction of protein structures in P. falciparum and subsequently used in this study to predict the structure of PfSpdSyn (McKie et al., 1998; Yuvaniyama et al., 2003; Birkholtz et al., 2004; de Beer et al., 2006).

Methods

Homology Model Construction

Two crystal structures were used to construct the PfSpdSyn, which included the sper-midine synthase crystal structures of Thermotoga maritima (1.80 Resolution, PDBid: 1JQ3) and Arabidopsis thaliana (2.70 Resolution, PDBid: 1XJ5). The spermidine synthase crystal structure from Caenorhabditis elegans was omitted as a template since the gate-keeping loops were not resolved (Blundell et al., 2006). The spermidine synthase crystal structure from T. maritima (TmSpdSyn; PDBid 1JQ3) was co-crystallized with a combined substrate-product analogue, S-adenosyl-1,8-diamino-3-thiooctane (AdoDATO; Korolev et al. (2002)). In an attempt to nd an optimal alignment between the template and target sequences, the T-Co ee package was used to align SpdSyn sequences retrieved from the UniProt and PDB databases, resulting in a protein family alignment (Notredame et al., 2000; Apweiler et al., 2004; Breman, 2001). The experimentally aligned protein sequences of the spermidine synthases of T. maritima and A. thaliana were used as templates to construct the homology model of P. falciparum. PfSpdSyn (UniProt entry: Q9FS5) was indicated to have a sequence identity of 32% and 49% with T. maritima and A. thaliana, respectively. Homology models of PfSpdSyn containing the substrate analogue, AdoDATO, were constructed using MODELLER 6v3 (Sali and Blundell, 1993). The PfSpdSyn models were subjected to stereochemical analysis, using PROCHECK, to evaluate the quality of the model (Laskowski et al., 1993).
The PfSpdSyn homology models were further subjected to re nement by using the CHARMM (Chemistry at HARvard Molecular Mechanics) package (Brooks et al., 1983). The partial charges used in the construction of the residue topology le of AdoDATO were computed using the MOPAC module within the InsightII (Accelrys; www.accelrys.com) package. The PfSpdSyn models containing AdoDATO were subjected to 500 steps of steepest descent minimization followed by 50 steps of Adopted Basis Newton-Raphson (ABNR) minimization using the CHARMM27 all-atom empirical force eld for proteins and nucleic acids (Igarashi et al., 1982). The protein-ligand interactions between the PfSpdSyn and TmSpdSyn structures were then compared using LIGPLOT (Wallace et al., 1995).

Binding Site Analysis

AdoDATO was removed from the PfSpdSyn model and docked back into the model using Cerius2 (www.accelrys.com). The docked and build-in AdoDATO were com-pared with AdoDATO crystallized within TmSpdSyn. The comparison was made us-ing LIGPLOT and visual inspection (Wallace et al., 1995). The homology model of PfSpdSyn containing the build-in AdoDATO was used in further analysis. Evaluation of the binding cavity of PfSpdSyn was done using the LigandFit module of Cerius2 (www.accelrys.com). Two binding cavities could be distinguished, one for binding of dcAdoMet and the other for putrescine binding.

Protein-substrate Interactions

Information obtained from the binding site analysis was subsequently used to eluci-date protein-substrate interactions. The moieties of the substrate analogue AdoDATO were converted into dcAdoMet and putrescine using InsightII (www.accelrys.com). The attacking nitrogen of putrescine was in the deprotonated state since it needs to per-form a nucleophilic attack on the electrophilic carbon of dcAdoMet. The PfSpdSyn model containing the newly formed substrates were then subjected to 100 steps of steep-est descent minimization using CHARMM (Brooks et al., 1983). Putrescine adopted a strongly angular conformation as an artifact of the minimization conditions and was restored to a linear conformation using InsightII (Accelrys). The PfSpdSyn, AdoDATO complex was then further minimized for 400 steps using steepest descent minimization. To determine the protein-ligand interactions between PfSpdSyn and the protonated pu-trescine a further 100 minimization steps were performed with putrescine in the protonated state. Protein-substrate interactions were evaluated using LIGPLOT and visual inspection (Wallace et al., 1995).

Molecular Dynamics

Molecular dynamics (MD) was performed on the homology models containing the substrate analogue AdoDATO as well as the substrates, putrescine and dcAdoMet. The protein was solvated with TIP3 water molecules. Molecular dynamics was started by 5 000 steps of steepest descent minimization followed by 200 steps of ABNR minimization. The system was then heated to 310K in steps of 5K every 100 steps and left to equilibrate for 10 picoseconds (ps). The molecular dynamics simulation was subsequently performed for 1 nanosecond (ns). VMD was used to visually inspect the molecular dynamics simulations of the homology models (Humphrey et al., 1996). The site-directed mutagenesis models were also subjected to molecular dynamics and the same procedure as above was followed. Molecular dynamics simulations were also performed on models containing the products, spermidine and MTA as well as the SpdSyn inhibitor 4MCHA under the same conditions.

Declaration 
Acknowledgments 
List of Figures
List of Tables 
List of Abbreviations 
Chapter 1. Malaria, Drug Discovery and Polyamines
1.1. Malaria
1.1.1. Antimalarials
1.1.1.1. Quinolines
1.1.1.2. Antifolates
1.1.1.3. Artemisinin compounds
1.1.1.4. Multidrug Resistance
1.1.2. The Fight Against Malaria
1.1.2.1. Vaccines and Vector Contol
1.1.2.2. Malaria and Drug Discovery
1.2. Polyamines
1.2.1. Polyamine metabolism in mammals and P. falciparum
1.2.2. Polyamine Metabolism Inhibitors and Parasitic Protozoa
1.2.3. Spermidine Synthase
1.3. Aims
Chapter 2. Work leading up to Structure-Based Drug Design of Spermidine
Synthase: Structural and Mechanistic Insights into the Action of Plasmodium falciparum Spermidine Synthase
2.1. Introduction
2.2. Methods
2.2.1. Homology Model Construction
2.2.2. Binding Site Analysis
2.2.3. Protein-substrate Interactions
2.2.4. Molecular Dynamics
2.2.5. Validation of homology model by site-directed mutagenesis
2.2.6. Site-directed mutagenesis and functional analysis of recombinant PfSpdSyn
2.3. Results and Discussion
2.3.1. Comparative Modeling of PfSpdSyn
2.3.2. Binding Cavity Analysis
2.3.3. Dynamic protein-substrate interaction analyses .
2.3.4. Proposed mechanism of action of PfSpdSyn mediated by a gate-keeping
loop
2.3.5. In vitro validation by site-directed mutagenesis
2.3.6. Inhibitor studies
2.4. Conclusion
Chapter 3. The Development of a Dynamic Receptor-Based Pharmacophore Model of Plasmodium falciparum Spermidine Synthase 
3.1. Introduction to Pharmacophores
3.2. Methods and Methodology
3.3. Results and Discussion
3.3.1. Protein Structure Quality Assessment and pKa Predictions
3.3.1.1. Protein Selection and Structure Quality Assessment
3.3.1.2. pKa Prediction
3.3.2. Phase Space Sampling
3.3.2.1. Molecular Dynamics
3.3.2.2. Clustering of Molecular Dynamics Trajectory
3.3.3. Negative Image Construction, Hit Analysis and In Vitro Testing
3.3.3.1. MIF Analysis and Pharmacophore Feature Identication
3.3.3.2. Exclusion volumes
3.3.3.3. Pharmacophore Model Selection
3.3.3.4. DPM1 Binding Cavity
3.3.3.5. DPM2 Binding Cavity
3.3.3.6. DPM3 Binding Cavity
3.3.3.7. DPM4 Binding Cavity
3.4. Conclusion
Chapter 4. Concluding Discussion 
Summary
Bibliography
GET THE COMPLETE PROJECT