BioSimSpace diaries: steered MD simulations


Molecular dynamics (MD) simulations are frequently used to simulate conformational changes in protein structures at atomic resolution. The insights afforded by these simulations help drug designers identify opportunities for modulating protein function with drug-like molecules. However, the straightforward numerical integration of equations of motion at thermodynamic equilibrium often fails to generate large-scale protein motions relevant for drug design purposes with a reasonable amount of computing resources.

Steered MD simulations

Steered MD (sMD) simulations provide a simple yet powerful approach to ‘bias’ MD simulations and encourage biomolecular structures to undergo specific motions. Typically, a harmonic potential defined with the help of collective variables is added to the potential energy function to control motions of a subset of atoms in the protein of interest.

Here we illustrate how sMD simulations may be used to sample pathways for the closure or opening of the “WPD” loop of the protein tyrosine phosphatase 1B (PTP1B). This enzyme is an attractive drug target for type II diabetes treatments. Catalysis of the cleavage of a phosphate group from a phosphorylated Serine or Threonine residue by PTP1B requires motions of the WPD loop to close the active site after substrate binding. Allosteric inhibition of PTP1B function by preventing closure of the WPD loop has been proposed as an attractive drug design strategy for this target.

Experimental studies show that the WPD loop undergo spontaneous motions between open (yellow) and closed (red) loop conformations on timescales of hundreds of microseconds (Choy 2017), a duration that is very difficult to achieve with brute force MD.

steered MD simulation with BioSimSpace

sMD simulations with BioSimSpace: defining a collective variable

We start by loading a model of PTP1B. Here, the protein has been solvated and the system parameterized previously, so we have everything we need to start an MD simulation. Our goal is to implement a steered MD protocol to bias the conformation of the WPD loop by controlling the root mean squared deviation (RMSD) of the coordinates of atoms in the WPD loop from coordinates in a reference structure. To do so, we load a reference PDB structure and select protein residues to define a collective variable.

This information is then passed to BSS.Metadynamics.CollectiveVariable.RMSD to initialise the RMSD collective variable (CV).

Specifying a sMD protocol

We next define how quickly the biasing force is ramped up, how long it is applied to drive the system towards a target CV value, and how quickly the biasing force is ramped down once the steering stage is over. In this particular example, the protocol implements a sMD run with a total duration of 152 ns. The values of the force constants for the restraint have been tuned by trial and error to achieve gentle closure or opening of the WPD loop over a duration of 80-120 ns.

Faster loop motions can be enforced by increasing the value of the force constants, at the expense of increasingly unrealistic conformational changes.Once the biasing parameters have been defined we load a steering protocol with BioSimSpace.Protocol.Steering .

The sMD simulation can be run with a chosen backend engine. For instance for using Gromacs or Amber to run the simulation we would write:


Analysing a sMD trajectory

BioSimSpace uses the excellent PLUMED software behind the scenes to implement the steering protocol. We can write a few lines of code to parse the PLUMED output and assess whether the steering was successful. The COLVAR file contains all of the CV values (r1,t1,d1), as well as more information on the force applied and work done. PLUMED outputs time in picoseconds and RMSD in nanometers. For easier plotting, we change time to nanoseconds and distances to Angstrom.

We can see that the RMSD values for the WPD loop started around 6 Angstrom (when the loop is open), and gradually decrease to about  2 Angstrom over about 100 nanoseconds. This indicates that the loop conformations simulated by the end of the steered MD runs are very similar to closed loop conformation used as a reference for the steering protocol.

If you want to try this out for yourself, check out our detailed steered MD tutorials by logging in to our JupyterHub server at (GitHub account required) and following the instructions in the file TUTORIALS.txt.

This post was written  by Julien Michel by adapting materials produced by Adele Hardie.