My research involves the use of various computer-based methods to study proteins of biomedical importance. Proteins are central to almost all cellular functions and consequently play a significant role in most well-known diseases. For example, our own proteins can cause problems when they become defective (e.g. through mutations), while functional foreign proteins are critical to the survival of harmful infectious agents (e.g. HIV).
Fortunately, medicinal small molecules (i.e. drugs) have been discovered which can modulate protein function and are therefore used to help control these conditions. For example, the asthma drug salbutamol is a synthetic compound which causes dilation of the airways by binding to and stimulating receptor proteins in the lungs. In contrast, nevirapine is a HIV drug which blocks virus replication by binding to and inhibiting a crucial viral protein.
Unfortunately, existing drugs have several shortcomings, which include adverse side-effects, toxicity and resistance. Also, existing drugs only target an estimated 1% of potential "target proteins", the remainder of which are either "undrugged" or unidentified. Furthermore, the candidate drugs which have been synthesized for testing represent only a tiny fraction of theoretically possible molecules. Estimates of the size of this so-called "chemical space" vary widely, but the analogy of the current repertoire representing a "bucket in the ocean" gives a rough impression! It is clear that drug discovery is a promising endeavour which is still in it's infancy and computational approaches are innovations aimed at addressing some of these deficiencies.
Computer-Aided Drug Design (CADD)
I am using in silico techniques to both identify new drug molecules with which to control these target proteins and to understand how the proteins themselves carry out their functions. I am particularly interested in the use of CADD approaches to improve the drug discovery process. Traditionally, drug discovery has been confined to the experimental laboratory, whereby millions of possible compounds are screened for desirable effects on the target protein. However, recent advances in computer hardware and the development of "molecular modeling" software have enabled a more predictive/rational approach with which to drive experimental work.
A classic application is the use of "docking" algorithms to predict which of the candidate drugs will bind to the target protein with the highest affinity and thus warrant further examination. Using geometric and chemical features of the target protein and a potential drug, it is possible to estimate the strength of the interaction between them. Repeating this for a large set of drug molecules is known as a "virtual screen" and the strongest binders are typically selected for experimental validation. Many docking algorithms have been implemented, with leading commercial and academic packages (such as Schrödinger's "Glide" and TSRI's "Vina") used frequently in our group.
One of the major shortcomings of most docking studies is the "rigid" nature of the 3-dimensional protein structure used. Proteins need to move around in order to carry out their function and these changes in shape can also impact the topography of the drug binding site. Even modest alterations in the structure of the binding pocket can have dramatic effects on the protein-drug interaction and it is accepted that drugs may bind to conformations of the protein that occur relatively rarely and are therefore missed in the rigid approximation. To model the "dynamic" nature of protein motion, we typically use Molecular Dynamics (MD) simulation, which essentially generates an "animation" of protein movements, according to the laws of classical physics. I have been using MD simulation in a number of ways, with application to important drug targets. Traditionally, MD simulation has been used to make a connection between a protein's motions and aspects of it's function. For example, comparing the behaviour of a protein in the presence and absence of a drug molecule, or observing a domain movement which allows it to carry out a reaction or binding event. I have also been using MD simulation in combination with molecular docking, by performing virtual screens against a range of protein conformations (using a method known as the Relaxed Complex Scheme). Most of these protein "poses" are not captured experimentally, yet may be very important in drug binding and lead to the discovery of novel compounds. I have also been using such "ensemble" representations of proteins to predict completely new drug binding sites on the surface of target proteins. There is mounting evidence for multiple drug binding sites on target proteins - for example, "allosteric" sites. These additional sites open up new avenues for drug discovery, yet it is challenging to predict exactly where those sites are buried, especially when we are restricted to rigid structures.
Viral & Human Targets
My work is predominantly based on the viral HIV-1 Reverse Transcriptase (RT) and human G-protein Coupled Receptors (GPCRs):
RT is a water-soluble protein which is vital for the HIV life cycle, converting viral RNA into DNA (a form which can be integrated and concealed in human cells). The pharmacological block of this process is therefore an attractive means of controlling infection and has lead to the approval of 17 "anti-RT" drugs by the FDA. In collaboration with experimental laboratories in California & Maryland, I have been using docking with MD simulation, to discover new inhibitors at both an existing binding site and a potential novel site. I have also used MD simulation to clarify the mechanism by which inhibitors actually stop this enzyme working.
In contrast, GPCRs are eukaryotic proteins which are embedded in the cell membrane and play pivotal roles in signalling events. They are involved in diverse physiological processes and are implicated in numerous diseases, such as cancer, and those of the cardiovascular and central nervous systems. Testament to this, around 30% of existing drugs bind to GPCRs, yet many of these drugs are associated with side-effects which are thought to be caused by "off-target" activity. There has therefore been substantial interest in allosteric drugs which bind to novel pockets on the surface of GPCRs and offer more potential for receptor selectivity. I have been using MD simulation in combination with a mapping algorithm to predict where these allosteric sites may lie, using the Beta-1 and Beta-2 adrenergic receptors as case studies. The aim is that this method can be applied to any protein target for which a 3D structure is available and new druggability is sought, and it has also been applied to HIV-1 RT to find novel binding sites. I have also been using MD simulation to elucidate aspects of beta adrenergic receptor function - in particular the way drug accessibility to the principal binding site is "gated".
I am also involved with other collaborative projects, which include computational modeling of a preclinical polymer for the delivery of cancer drugs and proteins involved in Chagas' disease.