This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
Author's personal copy
Recent progress in the study of G protein-coupled receptors with molecular
dynamics computer simulations
Department of Biochemistry and Biophysics, University of Rochester Medical Center, 601 Elmwood Ave., Box 712, Rochester, NY 14642, USA
a b s t r a c t a r t i c l e i n f o
Received 11 January 2011
Received in revised form 23 February 2011
Accepted 21 March 2011
Available online 3 April 2011
G protein-coupled receptor
G protein-coupled receptors (GPCRs) are a large, biomedically important family of proteins, and the recent
explosion of new high-resolution structural information about them has provided an enormous opportunity
for computational modeling to make major contributions. In particular, molecular dynamics simulations have
become a driving factor in many areas of GPCR biophysics, improving our understanding of lipid–protein
interaction, activation mechanisms, and internal hydration. Given that computers will continue to get faster
and more structures will be solved, the importance of computational methods will only continue to grow,
particularly as simulation research is more closely coupled to experiment.
© 2011 Elsevier B.V. All rights reserved.
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Rhodopsin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.Lipid–protein interactions in the dark state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.1. Polyunsaturated ω-3 lipids. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.1.2.Oligomerization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.Investigations of the activation mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1.Salt bridge to protonated Schiff base . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.2. Internal hydration changes upon activation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simulations of GPCR–G protein interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simulations of other GPCRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Assessing statistical errors in simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
G protein-coupled receptors (GPCRs) are arguably the most
important family of proteins currently studied. They are not only
numerous—GPCRs are the largest family of proteins in the human
genome—but exceptionally important biomedically. Indeed, it has
been estimated that more than a quarter of novel drugs target GPCRs
[1–3].As a result,GPCRshave drawnanenormousamountof scientific
attention, applying the entire arsenal of molecular biology, biochem-
istry, and biophysics. However, GPCRs, like many integral membrane
proteins, are difficult to handle experimentally; they require a
membrane-mimetic environment to remain folded. Thus, finding
conditions to overexpress and purify them remains challenging. As a
Biochimica et Biophysica Acta 1808 (2011) 1868–1878
E-mail address: firstname.lastname@example.org.
0005-2736/$ – see front matter © 2011 Elsevier B.V. All rights reserved.
Contents lists available at ScienceDirect
Biochimica et Biophysica Acta
journal homepage: www.elsevier.com/locate/bbamem
Author's personal copy
result, molecular level biophysical characterization of GPCR function
lags behind larger-scale functional approaches, yet this detailed
information will be needed to rationalize efforts in drug design and
other areas of GPCR research.
In 2000, the field took a major step forward with the publication of
the first high-resolution crystal structure for a GPCR, dark-state
bovine rhodopsin . This landmark achievement gave us the first
atomic level view of GPCR structure; however, the fact remains that
rhodopsin is a somewhat unusual GPCR, with a covalently bound (as
opposed to diffusable) ligand, and as such it was not clear how
applicable the insights from the structure would be to other GPCRs,
even those in the same subfamily.
In the next few years, several new structures were solved for
inactive rhodopsin [5–8], but it was not until 2007 that a second GPCR
crystal structure, this time for the β2-adrenergic receptor (B2AR), was
published [9–11]. This was followed in rapid succession by two more
GPCRs: the A2A-adenosine receptor (A2A)  and the β1-adrenergic
receptor (B1AR) . More recently, there have been a number of
structures that purport to capture a more “active” form of rhodopsin,
by crystallizing opsin (rhodopsin in the apo form, without its retinal
ligand) with  or without a G protein analog bound , or in the
presence of all-trans retinal, the agonist form of the ligand .
Finally, in 2010 new structures for the dopamine  and chemokine
 receptors were published.
This explosion of new structural information created an opportu-
nity for computational methods to make major contributions to our
understanding of GPCR function and dynamics. The rhodopsin
structures and, more recently, the structures of B2AR, B1AR, and
A2A have been used as starting points for a large variety of molecular
simulation projects. In this review, we will attempt to summarize the
work in this field, putting the results in the broader context of GPCR
function. We will not attempt a more general review of the GPCR field
as a whole or even to review the impact of the X-ray structures [19–
21], as this is too large a task for a single review. Moreover, we will not
attempt to cover all of the computational approaches applied to
GPCRs ; in order to control the scope of this manuscript, we will
largely exclude structure prediction, ligand docking, and other related
methods, in favor of a focus on more quantitative methods, primarily
molecular dynamics (MD) simulations. Finally, this manuscript is not
intended as an introduction to the field of GPCRs; although we will
cover some of the basics, we strongly suggest beginning readers also
consult the many review articles cited here. Basic GPCR behavior is
also covered in most standard textbooks on molecular cell biology
When referring to specific residues in GPCRs, we will use the
Ballesteros–Weinstein notation; in each helix, the most conserved
residue is numbered 50, and all other residues count from there .
Thus, the tryptophan residue involved in the rotamer toggle to
activation is Trp-265 in rhodopsin, found on helix 6, would be shown
as Trp-2656.48 in a discussion of rhodopsin, or as just Trp6.48 if we
were discussing the properties of that position independent of any
single protein. Although including the residue number for a specific
protein is not strictly part of Ballesteros–Weinstein notation, it will
simplify comparisons to other papers where the notation is not used.
For loops, where the residues are not well conserved, we will simply
include the loop identifier, e.g., “ECL2” for extracellular loop 2.
Rhodopsin, the dimlight receptor in the mammalian vision system,
in many ways functions as the hydrogen atom for GPCRs. It has been
studied extensively, with almost 8000 papers published on it
according to PubMed, starting with Wald and Clark in 1937 ; it
was actually discovered even earlier, in the late 1800s . Part of the
reason rhodopsin has been so heavily studied is its ready abundance;
in contrast to other GPCRs, which are typically found in very low
concentrations in the cell, rhodopsin is found in high concentrations
in the rod outer segment disks of the mammalian vision system. In
general, it is also more tolerant of high lipid–protein ratios than other
GPCRs, facilitating its study by biophysical methods, including
fluorescence, infrared spectroscopy, chemical labeling, electron
paramagnetic resonance,andsolid state NMR. This alsofacilitated
the protein's crystallization and led to the first high-resolution GPCR
structures in 2000 [4–8,28].
As a direct result, rhodopsin has also been heavily studied using
MD simulations, far more so than the other GPCRs. In particular, a
significant amount of work has focused on the role of lipid–protein
interactions in modulating rhodopsin function, including the possible
functional role of oligomers, and on the conformational changes
undergone during the activation process.
2.1. Lipid–protein interactions in the dark state
2.1.1. Polyunsaturated ω-3 lipids
The environment surrounding a biomolecule is critical to its
behavior, stabilizing the formation of a native state and modulating
the fluctuations that drive function. For membrane proteins, this
manifests in the ways that the lipid composition of a bilayer alters the
stability and efficiency proteins embedded in it [29–33].
Rhodopsin is a particularly interesting example of this phenom-
enon. Its native environment, the rod outer segment (ROS) mem-
branesof rod cellsof themammalianvisionsystem,havevery unusual
lipid compositions. First, they are highly enriched in polyunsaturated
ω-3 fatty acids ; given that natural lipids nearly always have a
saturated fatty acid in the sn-1 position, the results suggest that most
lipids have a polyunsaturated chain. This is remarkable considering
that mammals cannot synthesize ω-3 fatty acids themselves, and the
overall abundance of ω-3s at the organism level is more likely 5%.
Second, the cholesterol concentration is quite high in ROS membranes
but is not uniform . Rather, it is very high in immature ROS disks
and gradually drops as the disks migrate toward the top of the stack.
This clearly indicates that the cell is carefully controlling the ROS
membranes' lipid composition, and since almost all of the protein in
those membranes is rhodopsin it is unsurprising that experimental
work in vitro shows that both ω-3s and cholesterol have significant
effects on rhodopsin's activity. In particular, polyunsaturated lipids
enhance rhodopsin function, pushing the Meta-I/Meta-II equilibrium
toward the active Meta-II state , while cholesterol has the
opposite effect .
Given that both cholesterol and ω-3s alter the bulk properties of
liquid crystalline membranes—cholesterol increases order while ω-3s
are highly disordered , it would be interesting to know the
mechanism by which they each modulate rhodopsin. Specifically, do
they form specific interactions with the protein, or is their effect due
just to their effects on membrane elasticity or other bulk properties?
For the polyunsaturated lipids, the first evidence from simulations
came from a molecular dynamics simulation of rhodopsin in a 1-
stearoyl-2-docosahexaenoyl phosphatidylcholine (SDPC) membrane
by Feller et al. . Although the simulations were relatively short
(12.5 ns), there was a clear preference for the ω-3 docosahexanoyl
chains to interact with the protein, with a concomitant exclusion of
the saturated stearoyl chains.
Grossfield and coworkers followed this up by considering the
dynamics of rhodopsin in a realistic membrane composition contain-
ing SDPC, SDPE (phosphatidylethanolamine), and cholesterol [39,40].
The presence of multiple lipid species meant that lateral reorganiza-
tion of the membrane, which occurs on the microsecond scale or
slower, had to be taken into account. Accordingly, they chose to
perform 26 separate 100 ns simulations (as opposed to a single long
trajectory), rebuilding the bilayer from scratch each time to ensure
that a number of truly independent bilayer conformations were
explored. The results were consistent with those seen previously ;
A. Grossfield / Biochimica et Biophysica Acta 1808 (2011) 1868–1878
Author's personal copy
the density of the ω-3 chains at the protein surface was clearly
enhanced but the improvement in sampling allowed more detailed
analysis. In particular, the paper identified eight distinct sites on the
protein surface that repeatedly formed tight interactions with
docosahexaenoyl chains, consistent with solid state NMR results
suggesting specific binding [40,41]. Later analysis of the statistics of
chain states suggested that the preference for polyunsaturated chains
at the protein surface was entropically driven, in that the ω-3 chains
experience a far lower entropic penalty than saturated chains when
partitioning to the protein surface . A smaller number of less
populated clusters were identified for stearoyl chains and cholesterol.
Although previous simulation work suggested that cholesterol is
largely excluded from the protein surface [39,40], a more recent 1.6 μs
simulation of rhodopsin suggested that cholesterol–protein interac-
tions may modulate the kink angles of helix 7 .
There is also significant evidence that the headgroup composition
can also affect rhodopsin function, and that in particular PE head-
groups enhance rhodopsin function [44,45]. However, the time scale
of lateral reorganization appears too long for this to be directly
addressed by all-atom simulations at this time .
Over the last few years, the role of oligomerization in GPCR
function has been intensely debated . On one hand, there is
significant experimental evidence for homo- and hetero-dimerization
under some conditions [47–53]. Similarly, computational work from
Filizola and coworkers has argued that rhodopsin forms dimers 
and that other GPCRs form heterodimers . Along the same lines,
coarse-grained (CG) molecular dynamics simulations of rhodopsin in
a series of lipid bilayers with varying hydrophobic thicknesses clearly
demonstrated the role of lipid–protein interactions in modulating
On the other hand, the nature of the physiologically relevant state
is less clear, particularly since the nature of the lipid environment can
easily change the thermodynamic balance between monomer and
dimer . Early evidence suggested that rhodopsin functions as a
monomer [50,58,59], and more recent calorimetry experiments
indicate that in its native ROS membranes, rhodopsin is primarily
monomeric . Furthermore, rhodopsin is capable of binding G
protein in its monomeric form [61–66].
As a result, the simulation community faces a significant degree of
ambiguity: what state should we simulate? The vast majority of
simulations of GPCRs have considered only the monomeric form, but
is this a good model for the in vivo behavior? This is a case where there
On one hand, we want to simulate the conditions that best resemble
those found experimentally in vitro and in vivo, which might argue in
favor of modeling dimers. However, modeling dimers brings with it a
number of major technical challenges. To start with, the simulation
system would need to be much larger, greatly increasing the
computational cost and significantly reducing the length of the
trajectory that can be run. At the same time, the protein system is
larger, and thus will have slower fluctuation modes that need to be
sampled, meaningthata longer trajectory would be required to acquire
equivalent statistical sampling. Generating good statistics is already a
longest all-atom trajectories ever run have been simulations of GPCRs
[40,67–71], careful analysis of the convergence has shown that the
microsecond-scale is almost certainly not long enough to draw firm
conclusions about many of the most interesting phenomena [72–74];
this will be discussed more extensively in Section 5.
The final and perhaps most important challenge is due to the
simple fact that we do not know exactly what dimer (or higher-order
oligomer) to use. Although a number of researchers have constructed
models of dimers (for one example, see the work of Filizola, Weinstein
and coworkers [54,55,75–77] and others ), the fact remains there
is no crystal structure of a biologically relevant dimer, and the overall
record of homology modeling and related efforts does not give us high
confidence in the atomic level accuracy of such models. Thus, even if
the overall orientations of the monomers in the dimer are correctly
predicted, it is almost certain that many smaller-level details are not.
This in turn means that we must rely on the simulations to fix the
problems for us, something not guaranteed to happen over the course
of a typical MD simulation. Indeed, the largest concern is that there
would be no real way to know if the starting structure was badly
flawed: one could easily imagine a mis-packed dimer persisting for
hundreds of nanoseconds or longer.
Perhaps the only exception to this concern is the judicious use of
coarse-graining, which would extend the simulated time scales
enough to allow dimers to spontaneously form and dissociate under
equilibrium conditions. Indeed, Periole et al. used precisely this
approach very successfully ; each of their CG simulations
contained 16 rhodopsin molecules, and they systematically examined
the effect on oligomerization of varying the bilayer thickness.
Although their primary conclusion—hydrophobic mismatch greatly
increases the propensity of rhodopsin to oligomerize—is not an
enormous surprise, their work is an excellent example of the potential
for simulations to vividly illustrate biophysical principles. While their
model contains some very significant approximations, including a
large number of restraints needed to keep the proteins near the native
state, it is wholly appropriate for the context in which it is used.
2.2. Investigations of the activation mechanism
One of the key questions in the rhodopsin field (and the GPCR field
in general) is the nature of the conformational changes that occur
during activation. The common model  of GPCR–G protein
function is that ligand binding from the extracellular face (or ligand
isomerization in the case of rhodopsin and retinal) drives a series of
conformational changes that propagate to the intracellular loops,
allowing the G protein trimer to bind and exchange GDP for GTP. As a
result, the G protein trimer dissociates into Gα and Gβγ subunits,
which in turn continue the signaling cascade by changing the
behavior of other proteins, for example adenylate cyclase.
Rhodopsin's structural changes during activation have been
studied extensively using a variety of techniques, including cysteine
scanning, EPR, NMR, and of course crystallography [27,79]. However,
there has been no crystal structure of a truly “active” GPCR, in part
because assessing activity in the context of a crystal is exceptionally
challenging, as it would arguably require crystallizing the full GPCR–G
protein complex. The closest examples to date are the crystal
structures of opsin on its own  and with a G protein fragment
As a result, the idea of using MD simulations to directly explore the
activation mechanism is highly attractive, and efforts began shortly
after publication of the original rhodopsin crystal structure. Schulten
and co-workers published a 10 ns MD simulation of rhodopsin in a 1-
palmitoyl-2-oleoyl phosphatidylcholine (POPC) membrane, focusing
primarily on the retinal conformation and the relaxation of the side
chains in the binding pocket . Unsurprisingly, they did not see
isomerization of Trp-2656.48, because the expected timescale for that
motion is significantly longer. However, they did observe significant
relocation of the retinal β-ionone ring, consistent with experimental
cross-linking data from Borhan et al. .
Lemaˆıtre et al. also explored the relaxation of the retinal post-
isomerization . They performed a 10 ns dark-state simulation as
equilibration, then used steered MD to flip the retinal torsion to the
transstate. They performedthis flipping threeindependenttimes, and
continued each simulation using conventional molecular dynamics
for 10 ns. The analysis focused on the details of the retinal
conformation, with careful comparison to solid state NMR experi-
ments from the same group. Although the calculations are very short
A. Grossfield / Biochimica et Biophysica Acta 1808 (2011) 1868–1878
Author's personal copy
by current standards, the paper is noteworthy for its use of multiple
trajectories: running three separate trajectories allowed the authors
to attempt to assess which of the conformational changesupon retinal
isomerization are critical parts of the activation mechanism.
Crozier et al. published a far more extensive simulation a few years
later ; by running for 150 ns (as opposed to 10 ns), they were able
tosee moreofthe protein's responsetoretinal isomerization. Asin the
earlier work by Saam et al. , they observed that the ionone ring
approached Ala-1694.58, consistent with chemical cross-linking data
; indeed, because the simulations were an order of magnitude
longer, they saw far more extensive motion in this regard, with the
final separation between moieties around 9 Å. They also observed a
number of changes in helical tilts and kink angles, although in the
absence of a comparable dark-state simulation it is difficult to know
how much of these changes is due to retinal isomerization.
Additionally, they saw the salt bridge break between Glu-1133.28
and the protonated Schiff base linking retinal to Lys-2967.43, as
expected during activation; see Section 2.2.1 for more discussion of
the salt bridge behavior.
Recently, Hornak et al. published a very interesting paper where
they used experimentally determined distance changes as energetic
restraints; applying these restraints in a simulation efficiently drove
rhodopsin toward the active Meta-II state . Essentially, they used
magic angle spinning NMR to measure a set of interatomic distances
in the Meta-II state, primarily between carbons in the retinal and the
surrounding residues, and selected those that differed significantly
from the crystal structure (PDB code 1U19). The selected distances
were used as additional restraints in a MD simulation of rhodopsin in
a bilayer-mimetic environment, in effect forcing it to become
consistent with the distances measured in the Meta-II state. This
allowed them to observe activation-like motions in very short
simulations, typically only a few ns in duration. In particular, they
observed significant motion by helix 6, including disruption of the
ionic lock,anddisplacement of extracellular loop 2(EL2), whichforms
the “lid” enclosing the retinal binding pocket.
Although the method is very clever and the results largely
persuasive, there is a significant reservation when interpreting this
kind of result, one common to most steered MD-type applications:
although the end points are reasonably well-determined, the path
taken during the trajectory is not guaranteed to be physiologically
relevant. In the cell, rhodopsin proceeds from dark state to Meta-I on
the microsecond scale, which then interconverts with Meta-II on the
millisecond scale (although much of this delay may be due to waiting
for the retinal Schiff base linkage to deprotonate, a necessary step for
Meta-II formation). Here, while we can be confident that the
restrained residues end up in conformations consistent with exper-
iment (and there are enough redundancies to strongly suggest that
the overallbinding pocketchangesare probablycorrect), it is notclear
that the remainder of the protein has “caught up.” That is, those
portions of the protein that are not altered by the experimental
restraints are rapidly perturbed by their application, and the
trajectories may not be long enough for them to relax, let alone
fully sample their equilibrium distributions. This is a major concern,
given that previous work has demonstrated that simulations two
orders of magnitude longer do not effectively converge the motions of
individual loops , let alone the protein as a whole (see Section 5
for further discussion).
2.2.1. Salt bridge to protonated Schiff base
In all of the rhodopsin crystal structures, the salt bridge between
Glu-1133.28 and the protonated Schiff base linkage between Lys-
2967.43 and the retinal is easily observed. Moreover, it is well known
that this interaction is at least partially disrupted during protein
activation. However, the exact mechanism by which this occurs has
been the subject of significant controversy. Two competing models
exist for the role of two internal glutamates (Glu-1133.28and Glu-
181ECL2). One model, commonly referred to as the counterion switch
model , argues that the latter glutamate is protonated in dark-
state rhodopsin. After retinal isomerization, this proton is transferred
to Glu-1133.28, causing the salt bridge to break. After some
reorganization, the Schiff base forms a salt bridge to Glu-181ECL2,
signaling formation of the Meta-I state. By contrast, the complex
counterion model of Lu¨deke et al.  argues that both glutamates
are deprotonated throughout the process and that upon retinal
isomerization the Schiff base is effectively “shared” by the two
Röhrig, Rothlisberger and coworkers looked extensively at this
issue, systematically using Poisson–Boltzmann methods, classical
molecular dynamics, and mixed quantum/classical calculations
[87,88]. Their results were most consistent with the complex
counterion model, although the trajectories (particularly the QM-
MM ones) are fairly short. More recently, Ro¨hrig and Sebastiani
compared computed chemical shifts from a QM-MM trajectory to
experiments, but were unable to definitively determine which model
is more likely .
Recent advances in supercomputer technology have allowed
enormous increases in the scale of classical molecular dynamics
simulations [90,91]. Grossfield and coworkers took advantage of these
gains to test both mechanisms directly, using classical MD methods
[67,92]. They performed two separate all-atom simulations of the
activation process. In the first, designed to test the counterion switch
model, they began with an equilibrated model with Glu-181ECL2
protonated ; they induced the retinal torsion to flip from cis to
trans and ran for 500 ns. They then stopped the simulation, manually
moved the proton from Glu-181ECL2 to Glu-1133.28, and continued
the simulation for an additional 1500 ns. To test the complex
counterion model, they similarly generated an equilibrated structure
with both glutamates in the charged state , flipped the retinal
torsion, and ran for an additional 1500 ns. At the time of publication,
each of these simulations was longer than any previously published
all-atom trajectory of a membrane protein (or any other comparably
These results indicated that both trajectories behaved essentially
the way their underlying models suggested they should. In particular,
Grossfield et al. tracked the distances between the glutamate carbonyl
oxygens and the Schiff base nitrogen in both trajectories, as seen in
Fig. 1 . In Part A, showing the counterion shift trajectory, the Glu-
1133.28–Schiff base salt bridge remains stable until the 500 ns point,
where the proton was moved to Glu-181ECL2. The interaction then
quickly breaks and, a few hundred nanoseconds later, is replaced by a
salt bridge to the now-charged Glu-181ECL2. In Part B, the complex
counterion trajectory, the salt bridge breaks after about 100 ns, and
the Schiff base gradually approaches Glu-181ECL2, although both
glutamates make strong interactions with it.
On one hand, these calculations strongly suggest that both models
approached something like the Meta-I state; in each case, the
glutamates did precisely what the underlying models said they
should. On the other, there is not enough information to determine
which of the two models is more likely correct. To do so, Mart´ınez-
Mayorga et al. compared the trajectories to solid state NMR results
. Brown and coworkers previously reconstituted rhodopsin with
retinal specifically deuterated at particular methyl groups. From the
deuterium spectra, they predicted the orientation of the ionone ringin
the dark state and Meta-I [93,94] by finding the single conformation
that produced a theoretical line shape  most consistent with the
experimental spectrum. Mart´ınez-Mayorga et al. in a sense reversed
this approach, computing theoretical NMR spectra from the MD
trajectories by generating histograms of the methyl group orienta-
tions and using them to compute a weighted average spectrum. As
shown in Fig. 2, one trajectory—the complex counterion—produced
extremely accurate results, while the counterion switch trajectory
produced spectra that differ substantially from experiment,
A. Grossfield / Biochimica et Biophysica Acta 1808 (2011) 1868–1878