NVT or NPT ensemble, respectively for some nanoseconds. The
final step before the production run is an equilibration in the NVT
ensemble of at least 5 ns. We suggest to perform the productive
MD simulations using LINCS algorithm [117] to constrain heavyatom bonds, allowing for a 2 fs time-step. Long-range electrostatic
interactions can be described by the Particle-mesh Ewald summation Scheme [118]. Van der Waals and Coulomb interactions can
be truncated at 0.9 nm.
3.3 Classical MD
Simulations
Classical MD simulations should be carried out for some hundreds
of ns up to microseconds if possible, for the first exploration of
distal coupling between different protein regions. They can be
carried out at 300 K using, for example, a velocity-rescale
thermostat [119].
We suggest evaluating the radius of gyration and the secondary
structure content to ensure that they do not dramatically deviate
from the corresponding values in the known experimental structures used as starting structures for the simulations. These are best
practices and common sense in the field. It is also important to
monitor the main-chain Root Mean Square Deviation (RMSD) of
the folded regions of the protein with respect to the initial structure, and if some ns are required to reach convergence of this
property, the first part of the simulation can be discarded from
further analyses. In a case such as p53, where a metal ion is bound
to the structure, it is also important to evaluate the distances
between the metal atom and its coordinating groups over the
simulation time, as well as the coordination geometry. In a case
such as the one published by us, we carried out multiple MD
replicates of each p53 variant to better explore the conformational
space. This can be achieved, for example, using different random
initial velocities associated with the atoms at the beginning of each
simulation. If multiple replicates are used, the equilibrated portions
of each trajectory can then be concatenated in a unique macrotrajectory for the analyses.
3.4 Dimensionality
Reduction Methods
In this example, we will illustrate the usage of Principal Component
Analysis (PCA) [120], but also other methods can be used for the
same purpose such as higher-order statistics methods, or other
metrics for ensemble comparison that have been cited in the paragraphs above. The Gromacstools for PCA such as g_covar and
g_anaeig can be applied for this step. The PCA of MD trajectories
allows to identify the eigenvectors (also called principal components) of the mass-weighted covariance matrix of the atomic positional fluctuations, and it is generally carried out using the Cα
atoms. If different variants are under comparison, it is important
to concatenate the trajectories of the different system so that the
comparison can be made in the same PCA subspace. With a proper
MD sampling, the first three PCs generally accounts alone for more
Dynamics of p53
233
final step before the production run is an equilibration in the NVT
ensemble of at least 5 ns. We suggest to perform the productive
MD simulations using LINCS algorithm [117] to constrain heavyatom bonds, allowing for a 2 fs time-step. Long-range electrostatic
interactions can be described by the Particle-mesh Ewald summation Scheme [118]. Van der Waals and Coulomb interactions can
be truncated at 0.9 nm.
3.3 Classical MD
Simulations
Classical MD simulations should be carried out for some hundreds
of ns up to microseconds if possible, for the first exploration of
distal coupling between different protein regions. They can be
carried out at 300 K using, for example, a velocity-rescale
thermostat [119].
We suggest evaluating the radius of gyration and the secondary
structure content to ensure that they do not dramatically deviate
from the corresponding values in the known experimental structures used as starting structures for the simulations. These are best
practices and common sense in the field. It is also important to
monitor the main-chain Root Mean Square Deviation (RMSD) of
the folded regions of the protein with respect to the initial structure, and if some ns are required to reach convergence of this
property, the first part of the simulation can be discarded from
further analyses. In a case such as p53, where a metal ion is bound
to the structure, it is also important to evaluate the distances
between the metal atom and its coordinating groups over the
simulation time, as well as the coordination geometry. In a case
such as the one published by us, we carried out multiple MD
replicates of each p53 variant to better explore the conformational
space. This can be achieved, for example, using different random
initial velocities associated with the atoms at the beginning of each
simulation. If multiple replicates are used, the equilibrated portions
of each trajectory can then be concatenated in a unique macrotrajectory for the analyses.
3.4 Dimensionality
Reduction Methods
In this example, we will illustrate the usage of Principal Component
Analysis (PCA) [120], but also other methods can be used for the
same purpose such as higher-order statistics methods, or other
metrics for ensemble comparison that have been cited in the paragraphs above. The Gromacstools for PCA such as g_covar and
g_anaeig can be applied for this step. The PCA of MD trajectories
allows to identify the eigenvectors (also called principal components) of the mass-weighted covariance matrix of the atomic positional fluctuations, and it is generally carried out using the Cα
atoms. If different variants are under comparison, it is important
to concatenate the trajectories of the different system so that the
comparison can be made in the same PCA subspace. With a proper
MD sampling, the first three PCs generally accounts alone for more
Dynamics of p53
233