Heidelberg Institute for Theoretical Studies, Schloss-Wolfsbrunnenweg 35, 69118 Heidelberg, Germany

CAS-MPG Partner Institute and Key Laboratory for Computational Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, 320 Yueyang Road, Shanghai 200031, China

Abstract

Background

Biomolecules or other complex macromolecules undergo conformational transitions upon exposure to an external perturbation such as ligand binding or mechanical force. To follow fluctuations in pairwise forces between atoms or residues during such conformational changes as observed in Molecular Dynamics (MD) simulations, we developed Time-Resolved Force Distribution Analysis (TRFDA).

Results

The implementation focuses on computational efficiency and low-memory usage and, along with the wide range of output options, makes possible time series analysis of pairwise forces variation in long MD simulations and for large molecular systems. It also provides an exact decomposition of pairwise forces resulting from 3- and 4-body potentials and a unified treatment of pairwise forces between atoms or residues. As a proof of concept, we present a stress analysis during unfolding of ubiquitin in a force-clamp MD simulation.

Conclusions

TRFDA can be used, among others, in tracking signal propagation at atomic level, for characterizing dynamical intermolecular interactions (

Background

Many biomolecular systems or other complex macromolecules can dynamically visit a broad range of conformational states. External perturbations such as a molecular interaction or a mechanical force can cause a molecule to dynamically transit between these conformational states. While the conformational space of biomolecules is typically analyzed by coordinate-based methods such as the detection of correlated motions, Force Distribution Analysis (FDA) has been recently developed as an alternative approach to analyze structure and structural transitions

FDA has been applied so far to averaged dynamical data from Molecular Dynamics (MD) simulations. However, the dynamics of the force distribution within proteins or other macromolecules,

Implementation

Overview

Time-Resolved Force Distribution Analysis (TRFDA) is based on the same concept of using pairwise forces as FDA, but focuses on their evolution during the MD simulation as well as on the analysis of large molecular systems. To achieve these goals, it stores in memory pairwise forces from only one integration time step at a time, such that the memory usage is independent from the length of the simulation. It also saves output data in new file formats, which are well suited for time series analysis. Similar to FDA, atomic pairwise forces are computed for all types of bonded interactions as well as from Coulomb and Lennard-Jones potentials; long-range electrostatic interactions computed on a grid (

Decomposition of forces from 3- and 4-body potentials

While two-body potentials such as bonds, Coulomb and Lennard-Jones potentials, can be directly used for the analysis of pairwise forces, many-body potentials which act on more than two atoms need to be decomposed into pairwise forces. TRFDA introduces a complete decomposition of the forces resulting from 3- and 4-body potentials (angle, dihedral angle, cross bond-bond, cross bond-angle), as described in the Additional file

**Multibody force decomposition.** Force decomposition for 3- and 4-body atomic interaction potentials.

Click here for file

The decomposition rectifies the shortcomings of the previous FDA code, which used approximations for computing pairwise forces resulting from 3- or 4-body potentials. For an angle formed by atoms

Internally computed pairwise forces between residues

Computing pairwise forces between residues allows a significant decrease in storage and computational cost for analysis, while providing a mapping of the interactions on the primary and possibly also secondary structure of a protein. The examination of pairwise forces between residues can also be used as a tool in the development of residue-level coarse grained models

where

A further advantage of TRFDA is the equal treatment of atoms and residues with respect to the output options. For example, the same vector to scalar transformations can be applied to both pairwise forces between atoms and pairwise forces between residues, punctual stress (see below) can be calculated both per atom and per residue, and the output file formats are the same for data referring to atoms or to residues.

Summed versus detailed pairwise forces

In TRFDA, the force decomposition described above allows several pairwise forces to be computed for the same atom pair. For example, two atoms forming a bond might also form angles and dihedral angles with neighboring atoms, resulting in multiple forces between them from different potentials. For two atoms, it is most often interesting to calculate the total interaction between them, independent of the underlying energy potentials. Only few applications, like force field development, might require a differentiation based on the potentials, with an associated increase in memory usage and output file size. TRFDA accommodates both scenarios: in the first case, it computes a vector sum from all pairwise forces between the two atoms, while in the second case it computes separate vector sums for pairwise forces resulting from different types of potentials.

In contrast, due to the approximations used for 3- and 4-body interactions in the previous FDA version, the bonded pairwise force between two atoms results from exactly one type of interaction.

Choice of output

TRFDA comes with a wide range of output options. Although internally forces are calculated and stored only in vector form, TRFDA can output pairwise forces as vector or scalar values, and several derived quantities based on them.

A scalar pairwise force is computed as the magnitude of the vector pairwise force, same as in FDA, or as the magnitude of the pairwise force projected onto the distance vector between the two atoms, same as in

From the absolute values of the scalar pairwise forces acting on an atom, TRFDA can compute the sum and average as well as select the minimum or maximum. The sum of the absolute values of scalar pairwise forces acting on an atom

measures the stress acting on that atom. It is reminiscent of the calculation of stress from continuum models used in mechanical engineering. We denote the stress defined in Eq. 2 as

For both pairwise forces and per atom quantities, the output consists of simple to parse text files. Their format allows an unlimited number of atoms, pairwise forces and simulation time steps, and is documented in the manual, which accompanies the code. We chose text as opposed to binary file formats to allow for loading the data in various analysis tools at the expense of the size of the files.

TRFDA can also write out scalar pairwise forces in the same format as FDA, preserving compatibility with the FDA R library

Internal organization

TRFDA defines two groups of atoms between which pairwise forces are computed. A pairwise force is only computed when the two atoms belong to different groups, while pairwise forces between atoms of the same group are not computed. This allows, for example, interactions between a protein and a ligand to be computed efficiently. If the interactions inside the protein and ligand are also of interest, the two groups should comprise the same atoms, namely those of both protein and ligand. This way of selection avoids the calculation of possibly unnecessary pairwise forces within parts of the molecules, in contrast to the previous implementation of FDA

For internal storage of pairwise forces, FDA uses a square matrix-like structure for which the memory usage depends on the atom numbering in the molecular system. TRFDA replaces this structure with lists for a much more efficient memory usage, opening the possibility of calculating pairwise forces for significantly larger molecular systems. Using lists also allows storing, for the same two atoms, of separate pairwise forces from the different potentials,

Results and discussion

Edwards

As a proof of concept, we here used TRFDA to follow the dynamics of the internal forces in ubiquitin during mechanical unfolding, as previously probed in single molecule force-clamp experiments

Punctual stress during unfolding in a force-clamp MD simulation of ubiquitin

**Punctual stress during unfolding in a force-clamp MD simulation of ubiquitin.** The end-to-end distance (overlay black line) remains relatively constant until around 28.8 ns after which it increases abruptly, indicating the start of the unfolding process. Stress values are given at 1 ps time intervals, and have not been smoothed or averaged.

Overall, the punctual stress fluctuates until around 28.6 ns suggesting that the network of pairwise forces is able to dynamically balance its load to resist the pulling force. We note that fluctuations in individual pairwise forces are large, but their time averages have been previously shown to converge on the nanosecond time scale of typical MD simulations (Figure 5 of _{
i
} in Eq. 2, varies much less in comparison. At a later time, significant variations in stress for many of the residues correlate with the stretching and unfolding of the structure. The stress gradually disappears from the

Conclusion

TRFDA opens the possibility of following the evolution of pairwise forces during the course of an MD simulation, bringing insights into dynamic processes like transitions between stable states or signal transmission. It contains a force decomposition scheme for 3- and 4-body potentials, and can accurately represent the distribution of a pulling force in the molecular structure. Its output ranges from very detailed atomic pairwise forces, separated by the interaction potential and useful in force field development, to summed pairwise forces between residues used to map interactions to the protein structure. TRFDA can calculate a per atom or per residue punctual stress, which highlights the points where forces accumulate in a molecular structure, thereby pinpointing important elements which contribute to the mechanical resistance of the structure. Even though pairwise forces or stresses within a molecular structure can currently not be assessed by experimental means, we believe these observables to prove useful in analyzing and understanding the mechanical response of a complex molecule like a protein, just as stress calculations in structural mechanics analyses of macroscopic objects proved useful in assisting the design process. While atomic-force microscopy experiments and MD simulations of protein unfolding allowed unprecedented insight into the mechanical stability and rupture mechanisms of proteins, TRFDA enables revealing the underlying molecular basis of the mechanical response observed. The implementation focuses on computational efficiency and low memory usage, making it suitable for the analysis of large molecular systems and long MD simulations.

Methods

The data used for the punctual stress analysis was ob- tained as follows:

All calculations were run with GROMACS 4.5.3 ^{+} and Cl ^{−} ions were added to obtain a 0.1 M salt concentration. A steepest descent minimization was carried out until the maximum atomic force was below 1000 kJ/mol nm. The molecular system was then equilibrated with a 0.1 ns NVT simulation followed by a 1 ns NpT simulation; during both these simulations, the heavy atoms of the protein were subjected to position restraints of 1000 kJ/mol nm. The short-range neighbor list, electrostatic and van der Waals cutoffs were set to 1 nm; long range electrostatic interactions were computed with the Particle Mesh Ewald method

To unfold the protein, a force-clamp MD simulation was then performed in the same conditions as above, except that only the bonds between heavy atoms and hydrogen atoms were constrained to their equilibrium values. A constant force of 400 kJ/mol nm was applied to the C _{
α
} atoms of the C- and N-terminal residues in the direction of the largest dimension of the water box; the simulation was stopped when the distance between the C _{
α
} atoms exceeded 10 nm.

TRFDA was carried out on the trajectory obtained from the force-clamp MD simulation. Residue stresses were obtained according to Eq. 2, with only Coulomb interactions taken into account.

Availability and requirements

Competing interests

The authors declare that they have no competing interests.

Authors’ contributions

BC developed and tested the software, analyzed the data, and wrote this manuscript. FG designed the experiment, and wrote this manuscript. Both authors read and approved the final manuscript.

Acknowledgements

The authors would like to thank Christian Seifert and Scott Edwards for useful discussions and suggestions. Financial support from the Klaus Tschira Foundation is acknowledged.