1.2 Classical
All-Atom MD, Elastic
Network Model,
and Enhanced
Sampling MD
Protein conformational energy landscapes are characterized by a
series of local minima (Fig. 1) [5], and transition between one
minimum to another corresponds to a high-energy transition
state. In a classical MD simulation, several minimization steps are
required to reach a local minimum; after that MD production runs
are started. However, given the characteristics of their energy landscape, proteins can be trapped in non-relevant local minima, even
during long molecular dynamics simulations [6, 7]. These findings
do not mean that classical MD simulations are meaningless, but
MD simulations need to be addressed to specific aims, such as:
minimization of protein models (e.g., membrane proteins) [8],
normal mode analysis of protein motions [9], Molecular Mechanics—Poisson Boltzmann Surface Area calculations [9, 10], and
protein contact network analysis [11]. The protein contact network
(PCN) formalism could be addressed to identification of hotspot
residues involved, for example, in ligand binding, protein–protein
interactions, or allosteric modulation of protein motion [12–
15]. However, PCN analysis of classical all-atom MD frames, for
example, belonging to an energy minimum (equilibrated MD), can
be characterized by non-relevant correlations of PCN metrics parameters vs time (e.g., degree, average shortest path, closeness centrality, betweenness centrality etc.,) [9]. Thus, PCN analysis can be
applied to two different protein conformations or to protein normal modes generated with Gaussian Network Model (GNM) [16]
or Anisotropic Network Model (ANM) [11]. GNM and ANM are
both based on Elastic Network Model (ENM) formalism, in which
the macromolecule is treated as a network (coarse-grained model),
where nodes are atoms, nucleotides or amino acids, that are linked
by edges treated as harmonic restraints on displacement from the
structure at equilibrium. ENM provides a fast calculation of
low-frequency normal modes [17]. Indeed, considering that the
ENM results are comparable to MD simulation analysis, ENM
approaches can provide valuable structural information at low
computational costs (minutes vs days) [17]. However, enhanced
sampling MD simulations are commonly used when large protein
motions with high-energy barriers (Fig. 2) are studied [18]. Additionally, enhanced sampling approaches are particularly suited for
search and identification of putative allosteric pockets that can be
hidden in an apo crystal structure [19]. In this perspective, largescale biased MD simulations can explore low-populated conformations, where hidden allosteric pockets might be unveiled
[20]. Accelerated MD simulations are characterized by application
of a positive potential in the protein potential energy surface,
leading to overcoming of high-energy barriers; with this approach,
novel allosteric pockets were identified in the IL-1R1 receptor
[21]. A technique closely related to accelerated MD is metadynamics, where a bias potential enforces the exploration of novel
unexplored conformations [22], this technique recently
Molecular Dynamics and Drug Discovery
247
All-Atom MD, Elastic
Network Model,
and Enhanced
Sampling MD
Protein conformational energy landscapes are characterized by a
series of local minima (Fig. 1) [5], and transition between one
minimum to another corresponds to a high-energy transition
state. In a classical MD simulation, several minimization steps are
required to reach a local minimum; after that MD production runs
are started. However, given the characteristics of their energy landscape, proteins can be trapped in non-relevant local minima, even
during long molecular dynamics simulations [6, 7]. These findings
do not mean that classical MD simulations are meaningless, but
MD simulations need to be addressed to specific aims, such as:
minimization of protein models (e.g., membrane proteins) [8],
normal mode analysis of protein motions [9], Molecular Mechanics—Poisson Boltzmann Surface Area calculations [9, 10], and
protein contact network analysis [11]. The protein contact network
(PCN) formalism could be addressed to identification of hotspot
residues involved, for example, in ligand binding, protein–protein
interactions, or allosteric modulation of protein motion [12–
15]. However, PCN analysis of classical all-atom MD frames, for
example, belonging to an energy minimum (equilibrated MD), can
be characterized by non-relevant correlations of PCN metrics parameters vs time (e.g., degree, average shortest path, closeness centrality, betweenness centrality etc.,) [9]. Thus, PCN analysis can be
applied to two different protein conformations or to protein normal modes generated with Gaussian Network Model (GNM) [16]
or Anisotropic Network Model (ANM) [11]. GNM and ANM are
both based on Elastic Network Model (ENM) formalism, in which
the macromolecule is treated as a network (coarse-grained model),
where nodes are atoms, nucleotides or amino acids, that are linked
by edges treated as harmonic restraints on displacement from the
structure at equilibrium. ENM provides a fast calculation of
low-frequency normal modes [17]. Indeed, considering that the
ENM results are comparable to MD simulation analysis, ENM
approaches can provide valuable structural information at low
computational costs (minutes vs days) [17]. However, enhanced
sampling MD simulations are commonly used when large protein
motions with high-energy barriers (Fig. 2) are studied [18]. Additionally, enhanced sampling approaches are particularly suited for
search and identification of putative allosteric pockets that can be
hidden in an apo crystal structure [19]. In this perspective, largescale biased MD simulations can explore low-populated conformations, where hidden allosteric pockets might be unveiled
[20]. Accelerated MD simulations are characterized by application
of a positive potential in the protein potential energy surface,
leading to overcoming of high-energy barriers; with this approach,
novel allosteric pockets were identified in the IL-1R1 receptor
[21]. A technique closely related to accelerated MD is metadynamics, where a bias potential enforces the exploration of novel
unexplored conformations [22], this technique recently
Molecular Dynamics and Drug Discovery
247