posted on 2021-09-10, 19:38authored byLily Wang, Megan L. O’Mara
Molecular dynamics (MD) simulations
have been used extensively
to study P-glycoprotein (P-gp), a flexible multidrug transporter that
is a key player in the development of multidrug resistance to chemotherapeutics.
A substantial body of literature has grown from simulation studies
that have employed various simulation conditions and parameters, including
AMBER, CHARMM, OPLS, GROMOS, and coarse-grained force fields, drawing
conclusions from simulations spanning hundreds of nanoseconds. Each
force field is typically parametrized and validated on different data
and observables, usually of small molecules and peptides; there have
been few comparisons of force field performance on large protein–membrane
systems. Here we compare the conformational ensembles of P-gp embedded
in a POPC/cholesterol bilayer generated over 500 ns of replicate simulation
with five force fields from popular biomolecular families: AMBER 99SB-ILDN,
CHARMM 36, OPLS-AA/L, GROMOS 54A7, and MARTINI. We find considerable
differences among the ensembles with little conformational overlap,
although they correspond to similar extents to structural data obtained
from electron paramagnetic resonance and cross-linking studies. Moreover,
each trajectory was still sampling new conformations at a high rate
after 500 ns of simulation, suggesting the need for more sampling.
This work highlights the need to consider known limitations of the
force field used (e.g., biases toward certain secondary structures)
and the simulation itself (e.g., whether sufficient sampling has been
achieved) when interpreting accumulated results of simulation studies
of P-gp and other transport proteins.