Refine
Has Fulltext
- yes (2)
Is part of the Bibliography
- yes (2)
Year of publication
- 2015 (2) (remove)
Document Type
- Doctoral Thesis (2)
Language
- English (2) (remove)
Keywords
- Molekulardynamik (2) (remove)
\textbf{Molecular Determinants of Drug-Target Residence Times of Bacterial Enoyl-ACP Reductases.} Whereas optimization processes of early drug discovery campaigns are often affinity-driven, the drug-target residence time $t_R$ should also be considered due to an often strong correlation with \textit{in vivo} efficacy of compounds. However, rational optimization of $t_R$ is not straightforward and generally hampered by the lack of structural information about the transition states of ligand association and dissociation. The enoyl-ACP reductase FabI of the fatty acid synthesis (FAS) type II is an important drug-target in antibiotic research. InhA is the FabI enzyme of \textit{Mycobacterium tuberculosis}, which is known to be inhibited by various compound classes. Slow-onset inhibition of InhA is assumed to be associated with the ordering of the most flexible protein region, the substrate binding loop (SBL). Diphenylethers are one class of InhA inhibitors that can promote such SBL ordering, resulting in long drug-target residence times. Although these inhibitors are energetically and kinetically well characterized, it is still unclear how the structural features of a ligand affect $t_R$.
Using classical molecular dynamics (MD) simulations, recurring conformational families of InhA protein-ligand complexes were detected and structural determinants of drug-target residence time of diphenyl\-ethers with different kinetic profiles were described. This information was used to deduce guidelines for efficacy improvement of InhA inhibitors, including 5'-substitution on the diphenylether B-ring. The validity of this suggestion was then analyzed by means of MD simulations.
Moreover, Steered MD (SMD) simulations were employed to analyze ligand dissociation of diphenylethers from the FabI enzyme of \textit{Staphylococcus aureus}. This approach resulted in a very accurate and quantitative linear regression model of the experimental $ln(t_R)$ of these inhibitors as a function of the calculated maximum free energy change of induced ligand extraction. This model can be used to predict the residence times of new potential inhibitors from crystal structures or valid docking poses.
Since correct structural characterization of the intermediate enzyme-inhibitor state (EI) and the final state (EI*) of two-step slow-onset inhibition is crucial for rational residence time optimization, the current view of the EI and EI* states of InhA was revisited by means of crystal structure analysis, MD and SMD simulations. Overall, the analyses affirmed that the EI* state is a conformation resembling the 2X23 crystal structure (with slow-onset inhibitor \textbf{PT70}), whereas a twist of residues Ile202 and Val203 with a further opened helix $\alpha 6$ corresponds to the EI state. Furthermore, MD simulations emphasized the influence of close contacts to symmetry mates in the SBL region on SBL stability, underlined by the observation that an MD simulation of \textbf{PT155} chain A with chain B' of a symmetry mate in close proximity of the SBL region showed significantly more stable loops, than a simulation of the tetrameric assembly. Closing Part I, SMD simulations were employed which allow the delimitation of slow-onset InhA inhibitors from rapid reversible ligands.
\textbf{Prediction of \textit{Mycobacterium tuberculosis} Cell Wall Permeability.} The cell wall of \textit{M. tuberculosis} hampers antimycobacterial drug design due to its unique composition, providing intrinsic antibiotic resistance against lipophilic and hydrophilic compounds. To assess the druggability space of this pathogen, a large-scale data mining endeavor was conducted, based on multivariate statistical analysis of differences in the physico-chemical composition of a normally distributed drug-like chemical space and a database of antimycobacterial--and thus very likely permeable--compounds. The approach resulted in the logistic regression model MycPermCheck, which is able to predict the permeability probability of small organic molecules based on their physico-chemical properties. Evaluation of MycPermCheck suggests a high predictive power. The model was implemented as a freely accessible online service and as a local stand-alone command-line version.
Methodologies and findings from both parts of this thesis were combined to conduct a virtual screening for antimycobacterial substances. MycPermCheck was employed to screen the chemical permeability space of \textit{M. tuberculosis} from the entire ZINC12 drug-like database. After subsequent filtering steps regarding ADMET properties, InhA was chosen as an exemplary target. Docking to InhA led to a principal hit compound, which was further optimized. The quality of the interaction of selected derivatives with InhA was subsequently evaluated using MD and SMD simulations in terms of protein and ligand stability, as well as maximum free energy change of induced ligand egress. The results of the presented computational experiments suggest that compounds with an indole-3-acethydrazide scaffold might constitute a novel class of InhA inhibitors, worthwhile of further investigation.
The aim of the present work is the development and implementation of new simulation
possibilities for the CAST program package. Development included, among other
things, the partial parallelization of the already existing force fields, extension of the
treatment of electrostatic interactions and implementation of molecular dynamics and
free energy algorithms.
The most time consuming part of force field calculations is the evaluation of the nonbonded
interactions. The calculation of these interactions has been parallelized and
it could be shown to yield a significant speed up for multi-core calculations compared
to the serial execution on only one CPU. For both, simple energy/gradient as well as
molecular dynamics simulations the computational time could be significantly reduced.
To further increase the performance of calculations employing a cutoff radius, a linkedcell
algorithm was implemented which is able to build up the non-bonded interaction
list up to 7 times faster than the original algorithm.
To provide access to dynamic properties based on the natural time evolution of a system,
a molecular dynamics code has been implemented. The MD implementation features
two integration schemes for the equations of motion which are able to generate stable
trajectories. The basic MD algorithm as described in Section 1.2 leads to the sampling
in the microcanonical (NVE) ensemble. The practical use of NVE simulations is limited
though because it does not correspond to any experimentally realistic situation.
More realistic simulation conditions are found in the isothermal (NVT) and isothermalisobaric
(NPT) ensembles. To generate those ensembles, temperature and pressure
control has been implemented. The temperature can be controlled in two ways: by direct
velocity scaling and by a Nose-Hoover thermostat which produces a real canonical
ensemble. The pressure coupling is realized by implementation of a Berendsen barostat.
The pressure coupling can be used for isotropic or anisotropic box dimensions with the
restriction that the angles of the box need to be 90. A crucial simulation parameter in
MD simulations is the length of the timestep. The timestep is usually in the rang of 1fs.
Increasing the timestep beyond 1fs can lead to unstable trajectories since the fastest
motion in the system, usually the H-X stretch vibration can not be sampled anymore.
A way to allow for bigger timesteps is the use of a constraint algorithm which constrains the H-X bonds to the equilibrium distance. For this the RATTLE algorithm has been
implemented in the CAST program. The velocity Verlet algorithm in combination with
the RATTLE algorithm has been shown to yield stable trajectories for an arbitrary
length of simulation time. In a first application the MD implementation is used in conjunction
with the MOPAC interface for the investigation of PBI sidechains and their
rigidity. The theoretical investigations show a nice agreement with experimentally obtained
results. Based on the MD techniques two algorithms for the determination of free
energy differences have been implemented. The umbrella sampling algorithm can be
used to determine the free energy change along a reaction coordinate based on distances
or dihedral angles. The implementation was tested on the stretching of a deca-L-alanine
and the rotation barrier of butane in vacuum. The results are in nearly perfect agreement
with literature values. For the FEP implementation calculations were performed
for a zero-sum transformation of ethane in explicit solvent, the charging of a sodium
ion in explicit solvent and the transformations of a tripeptide in explicit solvent. All
results are in agreement with benchmark calculations of the NAMD program as well
as literature values. The FEP formalism was then applied to determine the relative
binding free energies between two inhibitors in an inhibitor-protein complex.
Next to force fields, ab-initio methods can be used for simulations and global optimizations.
Since the performance of such methods is usually significantly poorer than force
field applications, the use for global optimizations is limited. Nevertheless significant
progress has been made by porting these codes to GPUs. In order to make use of these
developments a MPI interface has been implemented into CAST for communication
with the DFT code TeraChem. The CAST/TeraChem combination has been tested
on the $H_2 O_{10}$ cluster as well as the polypeptide met-Enkephalin. The pure ab-initio
calculations showed a superior behavior compared to the standard procedure where the
force field results are usually refined using quantum chemical methods.