Research Interests

Many important sub-cellular biological functions are governed by macromolecular complexes that form transient and fleeting structural states in crowded cellular environments. These processes are also often coupled and governed by networks of biochemical reactions. Understanding such phenomena challenges the use of conventional structural approaches and necessitates the use of integrative methods that can combine various types of experimental data with biophysical models that follow reaction-coupled precise biomolecular assembly events in both space and time. Our research is concerned with both the development and integration of such multi-scale computational biophysics approaches with a range of structural biology and deep learning methods. We work primarily with the HIV-1 maturation system, which is an excellent model to develop such approaches and contains several interesting areas of structural biology such as macromolecular polyprotein assemblies and RNA-regulated protein enzyme catalysis. We integrate high resolution X-ray, cryo-EM data and in vitro/in vivo imaging with all-atom, coarse-grained and ultra-coarse-grained (particle reaction diffusion) modeling to enable a spatiotemporal view of large macromolecular processes. This allows us to model timescales from the microsecond to minutes and spatial scales from the Angstrom up to 100s of nanometres. It also permits us to anticipate and predict novel and transient structures in the context of crowded environments, which have so far been too difficult to characterize experimentally, and thus steer experimental approaches for their determination.  We are also developing advanced multiscale statistical physics methods for determining binding free energy and molecular kinetics of biomolecules with application to a variety of protein-protein and protein-ligand systems. These efforts drive our aim to gain a deeper understanding of the physics of soft condensed matter systems, with an application towards both understanding the physical determinants of retrovirus maturation and control of synthetic retroviruses for gene delivery applications but also to inform therapeutic avenues for HIV-1 eradication therapy.

Current Research

Modelling the modulation of enzyme kinetics in phase-separated ribonucleoprotein condensates

Volkswagen Foundation ‘Experiment’ Grant

Proteins diffusing in and around an RNA granule inside the cell, against a backdrop of other RNA granules (artistic impression, picture: Ina Poehner, Kashif Sadiq).

Cells are thought of as the basic unit of life. Each is a dynamic micro-world of billions of molecules involved in complex biochemical reactions. Cells control many of these internal reactions by physically separating the required molecules into membrane-bound compartments. But, remarkably, when subjected to stress, such as extreme temperatures, mechanical damage and toxins, cells can also form membrane-less granules. These granules often contain self-assembled, condensed mixtures of proteins and long biopolymers like ribonucleic acids (RNAs) – together known as ribonucleoproteins (RNPs). Excitingly, granules can co-exist in different phases of matter: liquid, gel, solid or even somewhere in between. And their function is still largely unknown.

We ask whether the rate of enzymatic reactions in such granules might be accelerated. Our aim is to explore which factors affect and regulate their biomaterial properties and whether RNA is just a passenger or an active driver of this process. And if accelerated catalysis is possible, what is the basic physical mechanism that underpins it? Could specific macromolecular assemblies facilitate diffusion in these kinds of biomaterials?

We use ribonucleoprotein (RNP) condensation during HIV-1 maturation as a model system to explore these questions. We collaborate with experimental partners who characterize the dynamic PR-cleavage induced condensation of viral RNP through a) a set of AFM measurements of RNP condensation at different intermediate stages of nucleoprotein processing b) biochemical assays of PR in the absence and presence of different lengths of nucleic acid and c) transmission electron microscopy of freshly budded virion particles. We have found that cleavage of nucleocapsid intermediates by PR result in supercondensation of the RNP followed by partial relaxation. Long nucleic acids induce significant acceleration of the PR reaction kinetics – which are experimentally determined to be due to sequestration of the PR by the RNP. To understand this phenomenon, we have developed a kinetic polymer model to account for both acceleration and sequestration in terms of crowding effects that emerge due to the RNP.

Deep learning mesoscale simulation methods for reaction driven sub-cellular assembly processes

Coupled interplay between chemical reactions and assembly processes dominates soft condensed matter and yet understanding and modelling the underlying physical processes remains a great challenge. Retrovirus maturation is a remarkable example involving 100nm-scale self-assembly, auto-catalysed chemical degradation and architectural transformation of ~104 constituent biomolecules in minutes, in a sequence of steps that need to be carefully timed. Whilst reaction kinetics approaches (as above) are good for modeling the change of concentrations of various species over time, they provide no spatial description, nor do they handle diffusion. On the other hand, all-atom and coarse-grained molecular dynamics (MD) simulations struggle to access realistic timescale required to follow large scale macromolecular assembly processes, such as virus capsid assembly. Moreover they do not handle phenomenological chemical reactions, such as polyprotein cleavage.  Recent development of ultra-coarse-grained (UCG) interacting particle-based reaction diffusion (iPRD) simulation approaches has now made a range of reaction-coupled clustering problems accessible.

We have recently developed a particle reaction diffusion approach to model retroviral infectivity that enables the spatiotemporal linkage between degradation of a truncated hexameric Gag lattice and diffusional clustering of envelope proteins to be followed. The model accounts for the basic physical determinants of infectivity and to our knowledge is the first-ever spatiotemporal model of partial retrovirus maturation processes that couples reaction to diffusion.

However, iPRD methods do not currently implement specific inter-particle orientations nor do they generally parameterise UCG inter-molecular interactions to lower coarse-grained and all-atom scales. This limits reliably modelling reactive assembly kinetics of orientation-specific 3D architectures. Therefore, we are developing orientation-specific UCG methods as well as deep learning multiscale intermolecular forcefield parameterisation methods for use in iPRD simulations. In collaboration with our experimental partners we are validating this approach on the model HIV-1 virion system using novel super-resolution microscopy technologies to track both the reaction kinetics and assembled structures within fully intact viruses. As well as future application in predictive synthetic retrovirology, these efforts aim to extend the applicability of computational methods to a wide range of reactive assembly problems at the mesoscale and thus a better understanding of the multiscale physics of biological condensed matter.

Multiscale methods for computing molecular binding kinetics

As part of the Kinetics for Drug Discovery project, we are developing multiscale computational biophysics methods for the prediction of protein-drug binding kinetics. In particular for forward binding rate constants (k_on), one of the major challenges is that the timescale for binding of clinically relevant inhibitors to pertinent molecular targets is considerably slower than what can currently be achieved by conventional unbiased molecular dynamics simulations. Therefore, we are developing methods that enable a rapid calculation of the diffusional and early encounter components of the complete binding process through flexible-body Brownian dynamics simulations, coupled with all-atom MD simulations and Markov state models that enable reconstructing the kinetics of the latter part of the binding process – particularly where both conformational selection and induced fit play non-negligible roles. A variety of systems are being studied including sets of inhibitors binding to HIV-1 protease, heat shock protein 90 (HSP-90), the model benzamidine-trypsin system and neuraminidase.

Novel computational approaches towards HIV-1 eradication therapy

Eradication of latently infected cellular reservoirs is a major barrier to the elimination of HIV-1. Notable strategies for targeting latent cells include the so-called ‘Shock-and-Kill’ methods that aim at drug-modulated reactivation of latent reservoirs (the Shock). Reactivated cells are then prone to death by virus production and cytotoxic immune response – the Kill. Recent advances in latency reversal suggest that the bottleneck of such an approach is not reactivation, but subsequent elimination, which is normally inefficient. We thus lack an effective ‘Kill’ component. Interestingly, non-nucleoside reverse transcriptase (RT) inhibitors (NNRTIs) of HIV-1 have the off-target effect of enhancing auto-activation of protease (PR) leading to highly cytotoxic premature PR release and subsequent infected cell death. The mechanistic and thermodynamic linkage between NNRTI binding and PR auto-activation within GagPol dimer precursors remains unclear. This limits rational design of NNRTI-site binders whose primary function is to act as PR auto-activation enhancers (PAEs).

We are developing a novel, rational and reproducible approach to design NNRTI-site-binding PAEs with optimal enhancement properties that induce infected cell death. Studies suggest RT is able to form a mature-like conformational heterodimer within the GagPol dimer precursor. We recently showed that N-terminal domains of mature-like RT are structurally compatible with catalytically viable C-terminal dimerized PR. Therefore even though PR activity is only ramped up when its N-termini are autocleaved, it cannot realistically get to that stage until its C-termini dimerize, whilst bound to RT. We have characterized the “nascent” event in HIV-1 PR autocatalysis. – this provides a mechanistic link for the therapeutic process and suggests a domino dimerization mechanism between PR and RT. Our recent development of a mathematical equilibrium domino dimerization (MEDD) model suggests that PAEs could potentially be reverse-engineered to fit a binding affinity window that yields optimum enhancement. Building on this work we are developing a structural model for a complete PR-RT precursor dimer based on all-atom molecular dynamics (MD) simulations.This forms the basis for a high-throughput virtual screening program combined with rigorous MD-based binding affinity calculations. Hits are experimentally tested by our collaborators for cytotoxicity and validated on an established cell-based assay that tests dose-dependent premature PR activation in eukaryotic cells.

Previous Research Projects

Developing and applying molecular dynamics (MD) simulations and biomolecular binding free energy calculation approaches

MD simulations of drug resistance in HIV-1 protease

We discovered a mutation-assisted lateral drug escape mechanism from the HIV-1 protease active site. Characteristic drug resistant mutant strains were shown to take advantage of extra coupling between the secondary-structure elements known as the “flaps” of the protease to induce the first stages of lateral dissociation from the active site. This led to the notion that a fully open active site is not necessary for inhibitor dissociation and revealed a new mechanism by which mutation-accelerated dissociation may occur.

Accurate and reproducible binding affinity calculation methodologies

We also developed a methodology for determining absolute inhibitor binding free energies to HIV-1 protease from MD simulations. The sensitivity of the approach allowed drug resistant mutations to be thermodynamically distinguished in excellent agreement with experiment. This guided the development and optimization of a rigorous absolute binding affinity methodology using explicit solvent and umbrella sampling techniques that made use of distributed computing resources and which came within 1 kcal/mol of absolute accuracy to experiment.

Development of automated binding affinity calculator for drug resistance in HIV for use in patient-specific clinical decision making software

The above methodological work formed the basis for the molecular component of the EU 6th Framework project “ViroLab” where computed thermodynamic data on inhibitor-mutant binding is integrated with additional resistance determination software to provide enhanced clinical decision support. We developed a computational framework that automated the calculation of protein-ligand binding affinities and which formed the basis of an initiative to provide enhanced phenotyopic clinical decision support on a patient specific basis, specifically by using viral genotypic information unique to a given patient and high performance computing resources. This required achieving successful management of the challenges faced in using state-of-the-art supercomputing infrastructures. The research has intersected with the EU 7th framework project “VPH”, forming an example of direct utilization of molecular level scientific models to address higher scale physiological behavior. The framework has subsequently been used to successfully distinguish between outputs from two discordant decision support software analyzing real patient genotypes.

MD-based discovery of structural and kinetic mechanisms for improving molecular therapy

Critical self-activation of immature HIV-1 protease

We investigated the structural basis for the self-activation of HIV-1 protease. The precursor protease emerges from an embedded dimerized form of the polyproteins that it cleaves. Using large-scale molecular dynamics simulations on the GPUGRID supercomputing infrastructure coupled to the development of novel analysis algorithms, we first demonstrated that the enzyme can exist in multiple conformations including the substrate bound conformation, even in the absence of the substrate. Then, by incorporating theoretical descriptions in transition path theory, we determined the structural and kinetic pathway of self-activation. The N-terminal of the protease, which is usually tightly folded in the dimer interface far from the active site, can in the immature protease have enough flexibility to bind to the active site thus triggering self-activation. This was the first all-atom structural description of the process. The structures that emerge from this work form the basis for a new target for HIV-1 therapy.

Movie of MD simulation of self-association of N-terminal tail of HIV-1 PR to its own active site

Domain rearrangements in NNRTI-bound HIV-1 Reverse Transcriptase (RT)

Function of RT requires the ability for one of its domains, the thumb domain to open and close during reverse transcription. A widely held view about the mechanism of NNRTIs was that they locked the “thumb” domain in an open conformation. We demonstrated using all-atom MD simulations that the thumb domain can close even in NNRTI-bound HIV-1 RT. This means that rather than an inducing an absolute locking mechanism NNRTIs must change the conformational equilibrium between open and closed forms of the thumb.            

Investigating the intrinsic disorder in HIV-1 fusion peptide

Many proteins do not have a stable tertiary structure yet are still functional, either because they transiently form folded structures or because they exhibit a pattern of dynamical interactions. The fusion peptide of HIV is an example. It is unstructured in solvent yet must form alpha-helical and/or beta-sheets to embed in host cell membranes thus forming the viral anchor that leads to membrane fusion. By designing and implementing large-scale MD simulations and developing novel analysis algorithms for clustering secondary structures, in collaboration with KU Leuven and UPF, we have shown that the fusion peptide forms a small complex ensemble of structures even in solvent as well as unstructured conformations. This implies that viral fusion is coordinated by the kinetic interplay between these conformers.

Systems biology approaches towards HIV maturation

Development of mathematical framework for enzyme-catalyzed (Gag and GagPol) cleavage

We derived a general theoretical reaction kinetics formulation for the complete degradation of heteropolymers. This can be applied to HIV Gag chains and allows prediction of the liberation rate of key components involved in HIV maturation such as capsid proteins. This has been extended to the degradation of Gag and GagPol polyproteins, including self-activation of the protease and constitutes the first detailed reaction kinetics model of HIV maturation.


Switch to the German homepage or stay on this page