Benchmarking Molecular Dynamics Force Fields for All-Atom Simulations of Biological Condensates

Proteins containing intrinsically
disordered regions
are integral
parts of the cellular signaling pathways and common components of
biological condensates. Point mutations in the protein sequence, genetic
at birth or acquired through aging, can alter the properties of the
condensates, marking the onset of neurodegenerative diseases such
as ALS and dementia. While the all-atom molecular dynamics method
can, in principle, elucidate the conformational changes that arise
from point mutations, the applications of this method to protein condensate
systems is conditioned upon the availability of molecular force fields
that can accurately describe both structured and disordered regions
of such proteins. Using the special-purpose Anton 2 supercomputer,
we benchmarked the efficacy of nine presently available molecular
force fields in describing the structure and dynamics of a Fused in
sarcoma (FUS) protein. Five-microsecond simulations of the full-length
FUS protein characterized the effect of the force field on the global
conformation of the protein, self-interactions among its side chains,
solvent accessible surface area, and the diffusion constant. Using
the results of dynamic light scattering as a benchmark for the FUS
radius of gyration, we identified several force fields that produced
FUS conformations within the experimental range. Next, we used these
force fields to perform ten-microsecond simulations of two structured
RNA binding domains of FUS bound to their respective RNA targets,
finding the choice of the force field to affect stability of the RNA–FUS
complex. Taken together, our data suggest that a combination of protein
and RNA force fields sharing a common four-point water model provides
an optimal description of proteins containing both disordered and
structured regions and RNA–protein interactions. To make simulations
of such systems available beyond the Anton 2 machines, we describe
and validate implementation of the best performing force fields in
a publicly available molecular dynamics program NAMD. Our NAMD implementation
enables simulations of large (tens of millions of atoms) biological
condensate systems and makes such simulations accessible to a broader
scientific community.