Research Details
Atomistic Simulations and Design in Biology
Research in our group utilises atomistic computer models to probe and provide mechanistic insights into fundamental processes in biology. We utilise the usual arsenal of tools: construction of models based on "imagination with a whiff of hand-waving", homology modelling, molecular dynamics, free energy, normal modes, reaction paths to examine shape shiftings in proteins, electrostatics, ligand-protein/protein-protein dockings including virtual screening (the docking program also includes the development of novel or modifications of existing scoring methods). On one hand, we examine how native and mutant forms of proteins may (mis)function while on the other, we have an extensive program that is directed towards ligand/drug discovery and protein/peptide design both from a therapeutic as well as a (bio)technological perspective. Our work is rooted in detailed computational biochemistry and we have extensive links with a variety of experimental labs, including our own attempts at "wetting our hands" so that our hypotheses are subject to rigorous testing and validations (EMBO Reports 9:144; 2008). Here we outline some of our findings.
(a) p53 pathway
(b) Cyclins
(c) Protein Kinase C
(d) Defensins
(e) Basic structural/computational biophysical chemistry
(a) The p53 Pathway (Collaboration with the p53 lab of Professor Sir David Lane, Institute of Molecular & Cell Biology, Singapore)
Computational findings and predictions: Cell Cycle (2007) 6:1-8
Experimental Validations: Cell Cycle (2008) March issue
The tumour suppressor protein p53 which guards cells against various forms of injury is known to be implicated in about 50% of all human cancers. Under normal conditions the protein is present at basal levels and is under the regulation of E3 ubiquitin ligase MDM2, which targets it for degradation. Upon cellular stress, DNA damage, hypoxia etc., the protein is stabilised and it induces pathways that cause cell cycle arrest. We are investigating aspects of this pathway, notably with questions such as (a) what is the nature of the interactions that help (de)stabilise p53 in its interactions with MDM2, (b) what are the physical forces that determine the stability of the p53 protein.

p53 interacts with MDM2 through its transactivation domain (TA), which assumes an amphipathic helix such that 3 conserved residues, Phe19, Trp23 and Leu26 (shown as sticks in Figure 1A) are presented to interact with a hydrophobic cleft in the N-terminal domain of MDM2 (shown as a surface in Figure 1A). At position 18 of the p53 TA region lies a Thr residue (shown as stick in Figure 1A) which is the site of phosphorylation by kinases CK1 and Chk2 upon cellular damage. Upon phosphorylation of Thr18, the interaction between MDM2 and p53 is abrogated, thus stabilising p53. The reason behind this has been proposed to be the reduction of the hydrogen bond between Thr18 and Asp21 (shown as sticks and h-bonded in yellow dashes in Figure 1A) as a result of the introduction of the charge-charge repulsion between the negative charge on Asp21 and that which develops on Thr18 as a result of its phosphorylation. This hydrogen bond has been thought to be key to the formation of the helical motif of p53 TA so essential to the optimal presentation of the 3 residues that interact with MDM2.
Figure 1A
Molecular dynamics simulations combined with Poisson-Boltzmann based electrostatic analyses and free energy computations carried out in our laboratory show that upon phosphorylation, this hydrogen bond is indeed broken, however only to be replaced by other hydrogen bonds and without drastic loss of the helical motif. Upon further analysis, we find that an anionic region on the MDM2 surface (shown circled region in Figure 1B) may generate a repulsive field that prevents phosphorylated p53 from binding tightly to MDM2. In order to test this hypothesis, we generated a triple mutant that removed this anionicity (shown in circled region in Figure 1C) and found that indeed, the computed binding affinity of phosphorylated p53 increased dramatically.
Figure 1B Figure 1C
This was subsequently tested in competitive binding assays with FAM labeled peptides and validated in the laboratory of Prof Sir David Lane at the Institute of Molecular & Cell Biology, Singapore. This study is important because it points to the possible evolution of MDM2 splice forms or mutants that could emerge in cancers with wild type p53 and over-ride the phosphorylation-dependant stabilisation of p53.
Stability of the core DNA binding domain of p53
Computational findings and predictions: BMC Bioinf (2008) 9:S17
In a related study, we focus our attention to the enigmatic observation of the marginal stability of p53. The tumour suppressor protein p53 protein has a core domain that binds DNA and is the site for most oncogenic mutations. This domain is quite unstable compared to its homologs p63 and p73. Various efforts have been undertaken to understand the stability of this p53 core domain. In one such effort, Fersht & co-workers (PNAS, 2006, 103:2109) noticed that the core region of p53 is characterized by two polar residues, Tyr236 and Thr253 whose equivalents in p63/p73 are two apolar residues, Phe238 and Ile255. When they mutated these two polar residues of the p53 core domain, with the apolar equivalents from p63/p73 they found that the stability of the p53 core had increased by ~1.6 kcal/mol. To understand the molecular basis of this (in)stability, we carried out the same study computationally and additionally we mutated the core residues in p63/p73 by their equivalents in p53. While the crystal structure of the core domain of p53 (shown as a dimer bound to DNA in Figure 1D) is available, we relied on the homology between p53 and p63/p73 to construct homology models (shown as Figure 1D inset). These were then subject to computational mutagenesis and MD simulations followed by reaction path analyses.
Figure 1D
Molecular dynamics simulations suggest that mutations in p53 lead to increased conformational sampling of the phase space which stabilises the system entropically and thus the free energy. In contrast, reverse mutations of p63 and p73 core domains, Ile255 by Tyr and Thr respectively showed reduced conformational sampling although the change for p63 was much smaller than that for p73.

In order to deepen our understanding of the issues surrounding notions of stability, we decided to ask the question: what molecular events might be associated with the stability that is being measured experimentally. The experiments monitor the amount of urea that is used to unfold a protein: the lesser the urea needed, the higher the instability of the protein. Now urea is a molecule with a volume 75A3. This is about 6-fold larger than a water molecule suggesting that for a urea molecule to penetrate into the core of a protein to cause unfolding, sizeable fluctuations in the protein need to occur. To study these, we examined the rotational motions of sidechains containing aromatic rings (shown in Figure 1E) that would lead to some of the openings required for urea to penetrate the protein.
Figure 1E
We find that the barriers to the rotation of the aromatic rings were reduced several-fold when p53 was mutated; in contrast they increased when p73 was mutated and decreased by a small amount in p63. The rates of ring flips of a Tyrosine residue at the boundary of two domains display changes that seem to be correlated with the changes in stability and could hint at possible pathways of entry of agents that induce unfolding. We are currently awaiting experimental validation of these findings.
(b) Cyclins - A Story of Discrimination (Collaboration with the lab of Professor Sir David Lane, Institute of Molecular & Cell Biology, Singapore)
Computational findings and predictions: Cell Cycle (2007) 6:2219-2226
Cyclin dependant kinases combine with cyclins to regulate the progression of cell cycle and gene expression by phosphorylating specific proteins. This is rooted in the ability of different complexes to recognise different substrate proteins (p27, p21, p57, E2F1, p53, pRb, p107) which bind across a shallow hydrophobic groove called the "cyclin groove". All these proteins that bind to this groove have a characteristic ZRXL motif (Z and X are cationic) and have been shown in the literature to inhibit Cyclin A2 and E1 but not B1. Using a combination of homology modeling, molecular dynamics simulations, free energy analyses and continuum electrostatic analyses, we ask: what is the structural basis of discrimination displayed by cyclins A2, E1 and B1 that are otherwise structurally similar.

We first constructed a homology model of Cyclin B1 which turned to be remarkably close (0.6A over Ca atoms) to the crystal structure of the same that was published even as we submitted the first version of our story for publication. Figure 2A shows our homology model superposed on to the crystal structure and the similarity at the peptide (magenta) binding site is clear.
Figure 2A
Electrostatics of the surfaces of Cyclins A1 and B1 reveal that an anionic patch characterizes the cyclin groove in Cyclin A1, thus enabling rapid binding of the ZRXL containing peptides (Figure 2B) whereas this region is cationic in Cyclin B1, thus preventing the binding of the cationic peptides (Figure 2C). Appropriate mutations have been constructed that reverse this specificity and await experimental validation.
Figure 2B Figure 2C
(c) Protein Kinase-C (Collaboration with the lab of Prof Duan Wei, National University of Singapore & Deakin University, Australia):
Computational findings: J. Biol. Chem., 281, 30768-30781; Cellular Signalling, 18, 807-818
The Protein Kinase-C family is key in transducing extracellular signals generated by various growth factors, hormones and neurotransmitters. The various PKC isozymes have a C-terminal tail region - the V5 domain - that has relatively low sequence conservation across the family, and was thus thought to be unnecessary for function. Dr Duan's group has demonstrated experimentally that the V5 domain is crucial for catalytic activity and isozyme-specific function. For example, PKC-epsilon, when truncated by 7 residues, loses 90% of its catalytic activity, and a further one residue removed completely inactivates the kinase. Also, PKC-alpha loses 40% of activity when the C-terminal residue is removed and loses all kinase activity when 10 residues are removed. Given the classical view of the stability of kinases, this was quite an enigmatic finding. We apply computational methods to investigate the mechanistic basis of these findings. Our work involves (a) modelling the different isozymes, and (b) running molecular dynamics simulations on the native kinase and V5-truncated mutants to identify the cause of loss of action.

Our homology models have an RMSD to within 1-2Å of the template structures, and show good backbone stereochemistry. The V-5 domain for PKC-alpha is shown to wrap around the N-lobe and interact with it through several key hydrophobic, electrostatic, and hydrogen-bonding interactions (Figures 3A, 3B and 3C). Molecular dynamics studies of PKC-epsilon and PKC-alpha, along with their truncation mutants, have enabled us to explain the loss of catalytic activity of these enzymes. The PKC truncation mutants lose a number of key stabilising interactions between the V5 domain and the N-lobe of the kinase. The removal of these interactions causes two major effects - it structurally destabilises the V5 domain (Figure 3D, 3E, 3F and 3G) and causes the loss of anti-correlated motions which are required for enzyme activity (Figure 3H, 3I).
Figure 3
Figure 3H
Figure 3I
(d) Defensins (Collaboration with the lab of Prof Roger Beuerman, Singapore Eye Research Institute)
Defensins are peptides that are key components of the innate immune response of several organisms and are the first defensive response against microorganisms. They are 30-45 amino acids long with 3 intramolecular disulphide bridges. They consist of high cationicity and hydrophobicity and are thought to function by being electrostatically attracted to the negatively charged surfaces of certain bacteria where they then proceed to disrupt the cellular membranes by interactions with their hydrophobic parts. We have a multi-disciplinary team, SCAMP (Singapore Consortium for Antimicrobial Peptides) that aims to study, design and exploit defensins and their analogues for therapeutic purposes. We have a two-pronged approach:
  a) Designing potent analogues of defensins: we have, in an effort that included computational design, synthesis, characterization using analytical techniques such as HPLC, CD, Mass Spec and molecular biology and cell culture effects on growing bacteria and host cells, successfully designed a series of linearised defensin analogues by systematically and uniformly replacing the disulphide linked cysteines with a series of amino acids (in press Chem BioChem).
  b) Understanding the physical basis of defensin action: Using computational studies, we are asking questions such as: defensins are known to oligomerise; if so, given their high cationicity, what prevents the charge based repulsion. Using the crystal structure of human beta defensin-2 we find that in several mammals (including humans), there is a conserved pattern of amino acids that upon dimerisation leads to enhanced cationicity, thus increasing the "attraction" towards bacterial membranes (see Figure 4A) The gain in free energy that accrues as a result of the dimerisation-linked burial of a hydrophobic region far outweighs the charge-charge repulsion from cationicity of the monomers - see Figure 4B (BMC Bioinf (2006) 7:S17).
Figure 4A
Figure 4B
Resource development:
Nucleic Acid Res (2007) 35:D265
Biotech J (2007) 2:1353-1359
In an effort to collate the diverse pieces of information that are increasingly available on defensins, we have, together with SERI, set up the Defensins Knowledgebase ( in April 2006. It provides a comprehensive information resource for the general public in a standardised and searchable format, and feedback from the scientific community and other interested parties is incorporated. The knowledgebase is manually curated and regularly updated; a snapshot of the welcome page is shown below (Figure 4C). The niche of the Defensins Knowledgebase is in its wide range of aspects of defensins research, such as the grants and patents awarded, clinical studies in progress, other research groups working on defensins, and commercial companies manufacturing defensins for lab use. This clearly is a well sought after resource as it has been accessed by over half a million visits since its inception.

(Contacts and queries are welcome at ).
Figure 4C
(e) A Few Snapshots of Basic Physics/Chemistry that Underpins Biochemistry/Biology
1. Are hydrophobic cavities hydrated (Proteins: Str Fn Gen 2007 67:868)
For years there has been a debate that has characterised the structural community: can hydrophobic cavities be hydrated. With the publication of a paper by Clore & colleagues (Science 1995 267:1813) which evidenced the presence of water molecules at the core of interleukin-1b using NMR signals, this debate became ever so lively (Science 1995 270:1847); because prior to that all the structural (crystallographic) evidence pointed to an empty cavity. And no wonder too, as this core is very hydrophobic (see Figure 5A) and this observation challenged the whole notion of what then defines a cavity as hydrophobic if it doesn't repel water. A few years later a new twist to this story was added by Caspar & colleagues (PNAS 1999 96:9613) who decided to investigate the less resolved electron densities that characterised the core of this protein and found that indeed the cavity does contain some waters, albeit seemingly mobile.
Figure 5A
We entered this arena with our usual toolbox of simulation technology with the assumption that the cavity does contain water molecules. If so, what are the conditions under which it would escape detection within the limits of current technology? These limits being (a) if the water molecules are highly mobile and exhibit fluctuations that are larger than say 1A2, then no self-respecting crystallographer would wish to localise the resulting smear of electrons (b) what would the lifetime be of these waters within the cavity that would give rise to the sorts of signals that Clore & colleagues used to conclude that the cavity is hydrated.

And indeed, using MD simulations on configurations starting from cavity containing 0, 1, 2, 3 and 4 water molecules respectively, we find that the cavity remains hydrated for periods that can well exceed the lifetimes that give rise to NOEs. The water molecules exist in clusters, satisfying their hydrogen bond needs by holding hands and also by weakly stabilising themselves through weak alliances with the aromatic rings that line the cavity. This is a well choreographed dance that results in a variety of hydrogen bonded clusters (see Figure 5B).
Figure 5B
We next decided to extend our story into the context of the biological role of interleukin-1b. Covariance analysis of the motions suggests that regions of the protein that are far apart are brought into concert only during the presence of water molecules. There are regions that are implicated in "triggering" functional motions that excite the receptor into transmitting a signal. Our findings, that such motions seem to occur only when the cavity is hydrated, suggested that waters serve to "plug" the hole and extend the packing so that the region is coupled together into a thermodynamically stable whole. The fact that the signal is transmitted only when waters exist in our simulations suggests that perhaps evolutionarily this exerts a kinetic control over the signal transmission.
2. Portraits of active site catalytic dynamics (Biochemistry 2008 47:2025)
The catalytic triad/diad is very ubiquitous in the enzyme world and has been the focus of research using a variety of tools. One such focus is on the papain family of cysteine proteinases, where the active site is characterised by the activation of a Cysteine-Histidine pair prior to catalytic competence. This feature is under the control of various ionisations occurring at various pH levels in residues that lie at various distances from the active site including one much investigated Asp158 which is the closest ionisable residue (7A) from the catalytic pair (see Figure 5C).
Figure 5C
We have been using a multidisciplinary approach combining reactivity probe kinetics, catalysis, computer simulations to probe in incisive detail the mechanisms underpinning these enzymes. In the latest twist to this story, analysing the pH-dependant kinetics of hydrolysis of a series of cationic aminoalkyl 2-pyridyl disulfide probes (see Figure 5D) and complemented by normal mode analyses, we find that the local packing controls the characteristic motions of the cleft walls and that of Trp177 that "cover" the active site and modulates the "degree of solvation" of the catalytic pair thus providing a mechanistic basis for the observed variation in activity rather than the electrostatics of Asp158 (see
Figure 5D
Feedback Login Site Map