The aim of the project
Table 1: Structure of neurokinins.
A pharmacophore model is a geometrical description of the chemical functionalities required of a ligand to interact with the receptor. The ligand needs to be both sterical and electronically complementary to the receptor binding cavity in order to attain activity. The first step in construction of the pharmacophore model is selection of a set of high affinity ligands. The set should be selected from the criteria of structural diversity and low flexibility. Next, functional groups considered essential for biological activity (pharmacophore elements) are identified. The most important step is identification of the receptor bound conformation. Once this is accomplished the ligands are superimposed so the pharmacophore elements overlap and a common template, i.e. the pharmacophore model is identified.5 The pharmacophore model is used to rationalize SAR and as a tool for designing new high affinity ligands.
Since the biological environment of a drug molecule is mostly aqueous calculations should be made using a solvation model. Otherwise polar intramolecular interactions will be overestimated and electrostatic collapse may occur. The most widely used solvation models are the dielectric continuum models. Also the protonation state of the ligands in the biological environment have to be considered. This is necessary since the protonation of a molecule will alter both its sterical and electronic properties.
RMS values are used as a measure of how well the proposed bioactive conformations fit the pharmacophore. For certain pharmacophore elements centroids and site-points should be used instead of superimposing the ligands atoms.5 For hydrogen donors/acceptors a dummy atom 2.8Å from the heteroatom along the hydrogen/electron lone pair will emphasize the directional aspect of the ligand-receptor interaction. Aromatic rings should be represented as a centroid. If all the five or six atoms were selected for superimpositioning the ring would be weighted much higher than other pharmacophore elements and the RMS value misleading.
Information about the receptor cavity can be obtained by analysing the ligands in their bioactive conformation. Low affinity ligands might intrude into regions that are already occupied by the receptor resulting in unfavourable ligand-receptor interactions. The space occupied by the receptor is referred to as the receptor-essential volume. If such areas can be identified they can be included in the pharmacophore model. Receptor-essential volumes are often necessary pharmacophore elements in models discriminating between receptor subtypes.
3D Quantitative Structure-Activity Relationship (3D-QSAR) is a statistical method for predicting the biological activity of ligands. Generation of the model requires a test set of ligands with known IC50 values, ranging from highly active to almost inactive.
In order to create a predictive model, it is crucial that the molecular alignment is correctly established. This can be done either manually or by an automatic procedure. The aligned set of molecules is positioned inside a grid box. At each grid point the interaction energy of a probe and each molecule is calculated. For each probe a map is generated from the interaction energies, displaying favourable and unfavourable interactions with the ligands. The probes are small molecules like water, ammonia and methane. Depending on the probe, the map will display properties, such as steric, hydrophobic or hydrophilic interactions.
Most of the variables do not contribute to the correlation between molecular structure and biological activity. These are discarded, but the system will still be massively over-determined, i.e. many more variables than molecules. Therefore classical regression methods can not be used. Partial Least Squares (PLS) and Principal Component Analysis (PCR) are the methods of choice for this problem. Both these methods reduce the variables into a few components, each linear combinations of the original variables. These components are constructed so they are orthogonal and contain as much of the remaining variance as possible.6
Finally the model is cross-validated, often by the random groups procedure. This involves omitting one structure from the data set, re-deriving the model and predicting the activity of the omitted compound. This procedure is repeated for every compound in the data set. The derived metric for the comparison is called q2. A q2 value of 1.0 corresponds to a perfect model, and a value of 0.0 to a model with no predictive power. A model with a q2 below 0.0 is worse than no model at all.6
If an experimental receptor structure is unavailable, a model can be build based on the structure of a homologous receptor. (Two proteins are homologous if they are related by natural evolutionary processes) This requires that the amino acid sequence of the receptor be known. The method is based on the observation that homologous proteins have similar 3D structures even for low sequence similarity. i.e. shape is better conserved than sequence.7
Forming an accurate alignment of the sequences is absolutely essential for deriving a useful model. Alignment is an automated procedure that requires a sequence similarity greater than about 20%. Structurally Conserved Regions (SCR) and Structurally Variable Regions (SVR) are identified. The SVRs are generally loops that connect regions of secondary structure. The main chain conformation and spatial relationship for the SCRs are taken from the known structure to which they were aligned. For the SVRs conformations that have the correct endpoint geometry and length are obtained from a database search or energy minimisation. The SCRs and SVRs are fused so they do not clash sterically. To this backbone the side chains are added and their conformations are refined.7
Contents of the project
H. Lundbeck A/S and Royal
Danish School of Pharmacy have a collaboration concerning the use of
advanced techniques within the field of computational chemistry applied to
specific subtypes of GPCRs.
In the elaboration of the 3D-pharmacophore models, state-of-the-art methods will be applied. Molecular mechanics programs such as FLO96 and the software package MacroModel will be used for exhaustive investigations of the conformational space of the ligands. Recently developed computational methods for the inclusion of solvent, e.g. water, in the calculations will be extensively used. This is a decisive importance for the calculations of conformational energies of bioactive conformations.
To investigate the electronic properties of the compounds they will be analysed using the quantum mechanical programs GAUSSIAN, JAGUAR and SPARTAN. Affinity data from the literature as well as Lundbeck's in house database will be used for 3D-QSAR studies with the programs GRID/GOLPE. For database search the program package CATALYST will be used.
On the basis of the developed 3D-pharmacophore models, the amino acid sequence, hydrophobicity plots, and site-directed mutagenesis data a NK-1 or NK-2 receptor model will be developed using Molecular Modelling techniques.
Read my Ph.D. Thesis: Pharmacophore and receptor models for Neurokinin receptors
1)Nutt, D., Substance-P antagonists: a new treatment for depression?, The
Lancet, 1998, 3521644-1646.
2)Raffa, R. B., Possible Role(s) of Neurokinins in CNS Development and Neurodegenerative or other Disorders, Neuroscience and Behavioral Reviews, 1998, 22, 6, 789-813.
3)File, S. E., Anxiolytic Action of a Neurokinin-1 Receptor Antagonist in the Social Interaction Test, Pharmacology Biochemistry and Behaviour, 1997, 58, 3, 747-752.
4)Regioli, D.; Boudon, A.; Fauchére, J.-L., Receptor and Antagonists for Substance P and Related Peptides, Pharmacological Reviews, 1994, 46, 4, 551-599.
5)Liljefors, T.; Pettersson, I. Computer-Aided Development of Three-Dimensional Pharmacophore Models; Krogsgaard-Larsen, P., Liljefors, T. and Madsen, U., Ed.; Harwood Academic Publishers: Amsterdam, 1996, pp 60-92.
6)Högberg, T.; Norinder, U. Quantitative Structure-Activity Relationships and Exprimental Design; Krogsgaard-Larsen, P., Liljefors, T. and Madsen, U., Ed.; Harwood Academic Publishers: Amsterdam, 1996, pp 94-130.
7)Hubbard, T. J. P.; Lesk, A. M. Modelling Protein Structures; Goodfellow, J. M., Ed.; VCH: Weinheim, 1995, pp 10-34.