approaches which drive transitions between crystal structures, but a
particularly interesting and automated example, called normal-
mode analysis and umbrella sampling molecular dynamics or
NUMD, uses an ENM model to build a transition coordinate
between structures and then calculates the PMF using umbrella
sampling [96].
SAX data is relatively low resolution compared to crystallogra-
phy but can still be used to drive exploratory biosimulations both
with biasing or correction potentials and adaptive sampling of a
common classical force-field [97–99].
DEER is a pulse electron paramagnetic resonance (EPR) tech-
nique that provides distance distributions which relate to the dis-
tances between nitroxide spin labels (probes) introduced into the
target protein via site-directed mutagenesis. There are various lim-
itations to the size of the distances and the number of probes which
can be used at once, but DEER offers average interspin distribu-
tions and so can detect hidden but important conformations
[100]. DEER has also been combined with MD to correct easily
sampled ensembles which may be inaccurate owing to an imperfect
force-field [101, 102].
NMR offers dynamic detail of proteins from experiment at
biologically relevant timescales and hence can offer information to
drive molecular simulations. Granata et al. have used NMR chemi-
cal shifts as a bias-exchange metadynamics collective variable
through their camshift metric, which measures the difference
between calculated and experimental chemical shifts [103]. This
combination provides a considerable speed up for their case of
folding a small protein. However, lightly populated conformations
cannot be observed with normal NMR spectra. Specialized NMR
approaches such as chemical-exchange saturation transfer (CEST),
Carr–Purcell–Meiboom–Gill (CPMG) relaxation dispersion, or
paramagnetic relaxation enhancement (PRE) are able to find struc-
tural information on these transient states, although they are each
constrained to particular ranges of timescales [104]. A recent study
has been able to measure an interesting transient state in T4 lyso-
zyme and build a NMR restrained MD simulation of the related
enzyme catalytic process [105]. Despite some success in finding
hidden conformations with NMR, it remains a challenge for rou-
tine applications due to difficulties in interpreting NMR data (see
Note 17). In addition, the methods for evaluating consistency
between structural conformations and chemical shift data are of
limited accuracy [106]. Furthermore, NMR is often not able to
generate data for larger proteins or homomultimers, which, along
with the time it often takes to assign spectra, is a major limitation
for applications in drug discovery [107]. This suggests that
NMR-driven MD alone is not the answer to routinely finding
hidden conformations.
Computational Study of Protein Conformational Transitions 349