In this post, we showcase AMMo, a program that employs advanced biomolecular simulation methods to computationally profile ligands for assessing their allosteric effects.The rational design of ligands that modulate protein function via conformational changes presents new opportunities for drug discovery. AMMo has been developed by a team of researchers at the University of Edinburgh, Heriot-Watt University, and the biopharmaceutical company UCB.

Allosteric Ligands: Expanding Opportunities for Drug Discovery

Many ligands exert their effects by inducing conformational changes in protein structures upon binding, thereby modulating the biological function of the protein. In some cases, the ligand binds at a site remote from known functional regions—this site is called an allosteric site, and the ligand is referred to as an allosteric modulator.

Predicting in advance whether ligand binding to an allosteric site will activate or inhibit protein function is challenging, and most allosteric sites are discovered by chance. A computational methodology capable of robustly profiling ligands for their allosteric potential in silico would fill a crucial gap in the arsenal of molecular modeling tools available to drug designers.

AMMo: A Molecular Dynamics Workflow for Assessing Allosteric Modulators

AMMo (Allostery in Markov Models) employs steered molecular dynamics (sMD) simulations to efficiently compute plausible pathways that describe, in atomic detail, conformational changes in protein structures transitioning between functionally distinct states (typically, at least an ‘active’ and an ‘inactive’ state). While sMD is an efficient methodology, it does not provide information about the relative equilibrium probabilities of the sampled protein conformations.

To address this, AMMo then seeds ensembles of equilibrium molecular dynamics (MD) simulations from the sMD-generated pathway and analyzes the resulting swarm of MD trajectories using Markov State Models (MSMs). MSMs provide equilibrium probabilities but can be computationally intensive when used unguided to describe a conformational change of interest. The combined sMD/MSM approach in AMMo harnesses the advantages of both methods: computationally efficient simulation of conformational changes alongside rigorous estimation of equilibrium probabilities.

An AMMo project involves repeated simulations of putative conformational changes for a protein, its variants (typically point mutations), and the presence or absence of different ligands. By comparing the computed probabilities of the protein adopting a functionally ‘active’ state across these simulations, researchers can classify the effects of mutations or ligands as ‘activating’ or ‘inhibiting.’

AMMo is implemented as a highly automatable workflow, built on OpenBioSim’s BioSimSpace, which streamlines system preparation, steered MD, and MD simulations. MSM analyses are performed using the PyEmma library.

Case Study: Allosteric Ligands for PTP1B

AMMo was initially demonstrated by Hardie et al. on a dataset of allosteric ligands targeting the phosphatase enzyme PTP1B (protein tyrosine phosphatase 1B), a long-standing drug target for diabetes treatment. The active site of PTP1B contains numerous charged residues, making it challenging to identify ligands that can inhibit the enzyme by directly binding to the active site while maintaining favorable drug-like properties. As a result, there has been interest in identifying ligands that bind to alternative sites on the protein surface to allosterically inhibit PTP1B.

AMMo successfully ranked ligands bound to two distinct allosteric sites according to their inhibitory strength. A key advantage of AMMo is that the computed conformational ensembles can be analyzed to identify ligand interactions associated with allosteric effects, providing valuable insights for ligand optimization.

Case Study: Investigating the Activation Mechanism of EPAC1

More recently, AMMo was used by Hardie et al. to investigate the activation mechanism of the enzyme EPAC1 (Exchange Protein directly Activated by cAMP 1). As the name suggests, EPAC1 requires binding to the cofactor cyclic AMP (cAMP) to transition from an inactive to an active state. EPAC1 is a promising drug target for treating cardiac, metabolic, and inflammatory disorders, but developing a small molecule capable of activating EPAC1 more effectively than cAMP while maintaining good drug-like properties has proven challenging.

AMMo simulations successfully captured the activating effects of cAMP, as well as the effects of a point mutation that abolishes EPAC1’s sensitivity to cAMP. AMMo was then applied to understand why the drug-like small molecule I942 activates EPAC1 less effectively than cAMP. The simulations revealed that I942 does not engage in the same pattern of interactions as cAMP when EPAC1 reaches its active state. These insights led the team to propose additional protein-ligand interactions that could be engineered in I942 derivatives to fully activate EPAC1 using a drug-like small molecule. Experimental validation of this proposal will be reported in due course.

Future Applications

Thus far, AMMo has been successfully applied to model allosteric inhibitors and activators of enzymes. Ongoing work aims to extend its applicability to other critical drug target classes, such as kinases and ion channels. AMMo also highlights the benefits of combining multiple biomolecular simulation tools through BioSimSpace to construct scientific workflows of practical utility for drug discovery.

References

Written by Julien Michel