This journal is c the Owner Societies 2011Phys. Chem. Chem. Phys., 2011, 13,13709–13720 13709
Citethis: Phys. Chem. Chem. Phys.,2011,13,13709–13720
High temperatures enhance cooperative motions between CBM
and catalytic domains of a thermostable cellulase: mechanism
insights from essential dynamicsw
Paulo Ricardo Batista,z*abMauricio Garcia de Souza Costa,za
Pedro Geraldo Pascutti,acPaulo Mascarello Bischacand Wanderley de Souzaabc
Received 29th November 2010, Accepted 9th June 2011
Cellulases from thermophiles are capable of cleaving sugar chains from cellulose efficiently at high
temperatures. The thermo-resistant Cel9A-68 cellulase possesses two important domains: CBM
and a catalytic domain connected by a Pro/Ser/Thr rich linker. These domains act cooperatively
to allow efficient catalysis. Despite exhaustive efforts to characterize cellulase binding and
mechanism of action, a detailed description of the cellulose intrinsic flexibility is still lacking.
From computational simulations we studied the temperature influence on the enzyme plasticity,
prior to substrate binding. Interestingly, we observed an enhancement of collective motions at
high temperatures. These motions are the most representative and describe an intrinsic hinge
bending transition. A detailed analysis of these motions revealed an interdomain approximation
where D459 and G460, located at the linker region, are the hinge residues. Therefore, we propose
a new putative site for mutagenesis targeting the modulation of such conformational transition
that may be crucial for activity.
Plant biomass is a clean and renewable energy source that has
been extensively studied and used to produce biofuels.1,2
Cellulose, its major component, is a structural polysaccharide
component of the plant cell wall and is the most abundant
polymer in nature. It consists of a linear chain of multiple
b(1–4)linked D-glucose units. Due to the amphipathicity of this
molecule, cellulose is found to form three-dimensional (3D)
crystalline microfibrils. Its hydrophobic sites permit multi-
laminar arrangement while its multiple hydroxyl groups on
each glucose-unit form hydrogen bonds with the same or with
a neighbor chain (detailed in ref. 3).
The 3D supramolecular organization of cellulose is one of
the barriers for cleaving enzymes to access their substrates as
well as binding to lignin and other molecules.2Cellulases are
enzymes that catalyze the hydrolysis of cellulose’s glucosidic
bonds releasing shorter chains and also cellobiose/glucose units.
These proteins are widely spread in nature and present a
multidomain architecture consisting of strings of mobile
modules. They possess at least one catalytic domain (CD),
which contains the active site, linked to one or several non-
catalytic modules.4Among them, we emphasize the importance
of the carbohydrate-binding modules (CBM) that recognize,
interact and promote the association of the enzyme with the
substrate. One or multiple CBMs can be found in the same
Cellulases have been classified as endo- or exo-cellulases, on
the basis of their action mode on the substrate. One unusual
case is the cellulase Cel9A—formerly named E4—from the
filamentous soil thermophile Thermobifida fusca that presents
both exo- and endo-cellulase activities.6This bacterium produces
six structurally and functionally distinct cellulases—Cel9B,
Cel6A, Cel6B, Cel9A, Cel5A, Cel48A (formerly called from
E1 to E6, respectively7)—which act cooperatively to convert
insoluble cellulose into cellobiose and other soluble sugars.8
Cel9A-68, a 68 kDa fragment of Cel9A from T. fusca, had its
3D structure solved and revealed a family-9 CD and a family-III
CBM, connected by a short Pro/Ser/Thr rich linker.6
Recently, Ting et al. developed a mechanochemical model
for the dynamics of cellulase in which they considered CD and
CBM as two random walkers and showed that the linker
length and stiffness play a critical role in the cooperative action
of these domains.9However, their model is too simplistic and
did not take into account all-atom interactions nor it described
which motions are implicated in the cellulase function.
aInstituto de Biofı´sica Carlos Chagas Filho, Universidade Federal do
Rio de Janeiro, Rio de Janeiro, 21941-902, Brazil.
E-mail: email@example.com; Fax: +55 21 2280-8193;
Tel: +55 21 2562-6575
bInstituto Nacional de Metrologia, Normalizac ¸a ˜o e Qualidade
Industrial – DIPRO, Duque de Caxias, 25250-020, Brazil
cInstituto Nacional de Cieˆncia e Tecnologia em Biologia Estrutural e
Bioimagens – INBEB, Brazil
w Electronic supplementary information (ESI) available. See DOI:
z Contributed equally.
Dynamic Article Links
13710Phys. Chem. Chem. Phys., 2011, 13,13709–13720 This journal is c the Owner Societies 2011
Computational methods such as molecular dynamics (MD)
have been successfully applied to explore protein flexibility
considering detailed atomic interactions.10–12Nevertheless, a
limited number of modeling studies analyzed the interaction
between cellulase and cellulose.13–17In a recent study, the
energetics of carbohydrate product expulsion from a Cel7A
was probed using multiple free energy methods.18Particularly,
regarding the T. fusca Cel9A, only two MD studies focused on
the details of the enzyme interaction with the substrate19,20
without considering intrinsic conformational changes prior to
In the present work, we employed MD simulations to
investigate the influence of temperature on the stability of this
protein. Furthermore, we have conducted a straightforward
essential dynamics analysis to unravel the conformational
changes involving CD and CBM in order to establish their
relationship with cellulase function.
Results and discussion
Concerning protein–ligand associations, the existence of an
active conformer prior to ligand binding is postulated. This
conformation is in a pre-equilibrium with other low energy
structures and the binding of a ligand shifts the equilibrium in
favor of a most active conformer.21,22In general, the ligand
restricts the protein conformational space adopted. In contrast,
the apoprotein can visit easily a larger conformational space,
including those adopted when the protein is bound to ligands.
Therefore, protein conformational studies on the apo form can
better characterize the intrinsic plasticity of a protein.
Based on this concept, we present here a detailed analysis of
the temperature influence on the dynamics of a thermostable
Cel9A-68 cellulase from Thermobifida fusca in the apo form.
The three temperatures chosen were around the room tempera-
ture (300 K), the optimal temperature for activity (325 K) and
an extreme temperature in which the enzyme is still functional
Electrostatic surface of cellulase
Usually, proteins from thermophilic organisms present a
large proportion of charged residues.23,24This observation is
essentially related to the fact that at high temperatures intra-
protein electrostatic interactions become more favorable.25In
addition, the presence of charged groups may reduce the
protein configurational entropy due to repulsion between like
charges.24,25Analysis of the residue amino acid (a.a) composition
of Cel9A revealed a high incidence of charged residues at the
protein surface, mainly Arg and Glu. This observation is
consistent with the a.a residue composition of enzymes found
in thermophilic organisms.23Because of the importance of
electrostatic interactions in this class of enzymes, we investi-
gated the electrostatic potential at the solvent accessible surface
area (SASA) calculated with the Poisson–Boltzmann Surface
area (PBSA) software,26over the averaged structure from each
MD simulation (at 300, 325 and 350 K) (Fig. 1).
In general, the electrostatic surface potential of the cellulase
for all averaged structures is dominated by negative charges.
This is consistent with the acidic pI (4.5) of the protein—note
the scale from ?4 to ?14KBT?1—that presents a total charge
of ?37. However, the regions of negative electrostatic potential
at the cellulase surface are spread in a non-homogeneous
fashion, since a large region of negative potential surrounding
the active site is observed.
Although the averaged structures obtained at 325 and 350 K
presented distinct conformations, the overall potential isocontours
and surface charge distribution are maintained. It was also
observed that the flat surface of cellulase, which was suggested
to be important for Cel9A-68 activity, is maintained even at
high temperatures (Fig. 1). Under these conditions, the increase
in thermal energy (BkT) allows the adoption of compact
conformations which appear to be inaccessible at 300 K due
of Cel9A-68 cellulase at different temperatures calculated with the
APBS17plugin in the Pymol64visualization software. For each tempera-
ture: the lateral (on the left) and top (on the right) views of the
electrostatic surface of cellulase; blue, red, and white indicate highly
charged regions, negative areas and neutral regions, respectively,
according to the scale indicating ?14 to ?4 KBT/ec, where KBis the
Boltzmann constant, T is the temperature and ecis the charge of one
electron. Bottom panel, the electrostatic isocontours: positive and
Electrostatic potential surface and positive/negative isocontours
This journal is c the Owner Societies 2011Phys. Chem. Chem. Phys., 2011, 13,13709–1372013711
to the repulsion between like charges. Remarkably, the potential
distribution and the shape of cellulase, which are crucial
features for the interaction with cellulose,6are maintained at
high temperatures even though the flexibility of the enzyme is
Temperature influences in global structural analysis
Structural thermal fluctuations of proteins are intimately
coupled to their functions27instead of merely representing
random events.28Herein are presented the results of explicit
solvent MD simulations during 50 ns of a thermostable
cellulase conducted at three distinct temperatures: 300, 325
and 350 K. All MD were carried out using the apo Cel9A-68
cellulase of T. fusca structure as initial coordinates (PDB code
1TF4). A priori, the system was heated until the desired
temperature and extensively equilibrated (see details in the
Methods section) so that the sampled trajectory used in the
analyses reflects structures belonging to a more stable stage of
the simulation, thus reducing artifacts due to differences between
the periodic water box (MD) and the crystal environments.29,30
We initially monitored the temperature dependence of global
structural parameters, starting from the analysis of the time
evolution of root mean square deviations (RMSD) of the
cellulase Ca-atoms during the simulation (Fig. 2A). In other
words, this analysis evaluates how much the protein deviates
over time from the initial structure. The protein at 300 K
presented the smallest deviation and adopted after 2 ns a
stable conformation not so far from the initial one (0.2 nm).
This can be seen more clearly in Fig. 2B where the pairwise
distribution of the RMS for the MD at 300 K presented a
narrow normal distribution centered in 0.2 nm. This indicates
that this system was very stable during the simulation, deviating
around a unique structure. The RMSD of the simulation at
325 K converged after 10 ns to a value around 0.4 nm,
remaining reasonably stable until the end of the simulation.
These results reveal substantial stability of cellulase at these
two temperatures (300 and 325 K). However, at 350 K the
RMSD values reached 0.6 nm at the end of the simulation.
RMSD distribution plots revealed that at 325 K a slightly
largerdistribution ofRMS values(from0.15to 0.45nm)isfound.
On the other hand, in the 350 K simulation, we clearly observed
a division between three populations around: (i) 0.25 nm—
very close to those observed at lower temperatures; (ii) 0.45 nm;
and (iii) high values of RMSD—not-well defined population
until 1.0 nm (Fig. 2B).
In Fig. 2C, we show the time-evolution of distinct clusters
of cellulase backbone conformations throughout the MD
simulations. Such analysis is more informative than a simple
RMSD plot against the initial structure; it can show not only
differences between each time step and the initial one but it can
also provide information about the structural transitions
occurred on the time-trajectory. Herein each cluster contains
conformations within an RMSD of 0.11 nm from its center
structure. If during a simulation the RMSD from the starting
structure increased gradually, the system probably visited
numerous clusters, which indicates that more conformational
states were accessed. In contrast, if a simulation visits only a
few (densely populated) clusters, it can be considered as being
more conformationally stable. Indeed, we found only two clusters
in the 300 K simulation (see the inset in Fig. 2C—red and blue
(initial) structures) in accordance with the narrow RMS distri-
bution plot analysis. Furthermore, in the 325 K simulation we
observed a progressive increase in clusters during the first 15 ns
of the simulation. Afterward this system stays in the same
cluster until the end of the 50 ns production period. This
indicates a shift towards a distinct conformation (see also the
inset for 325 K). In contrast, for the simulation at 350 K we
found a considerably higher number of different clusters (849)
with a continuous increase in RMSD backbone values.
evolution of the root mean square deviations (RMSD) of protein
Ca-atoms for the MD at 300 K (black line), 325 K (green) and 350 K
(red). In B, distribution of pairwise RMSD distances shown in A. In C,
the time-evolution of distinct clusters of cellulase backbone confor-
mations throughout the MD simulations, using the simple linkage
method (nearest neighbor) with a cut-off of 0.11 nm. The colored
spheres represent the initial position (blue) and the most representative
cluster structure (red). The inset figures (one for each temperature)
show the cartoon representation of the initial cellulase structure (blue)
fitted to the most representative structure within the cluster analysis,
represented by colored sphere position.
Global structural analysis of the MD trajectories. In A, time-
13712 Phys. Chem. Chem. Phys., 2011, 13,13709–13720This journal is c the Owner Societies 2011
From these results we have strong evidences that in the MD
simulation at 350 K the thermostable cellulase did not achieve
a conformational stabilization during the 50 ns. In order to
check if the protein at this temperature could enter in a
denaturing process we inspected the secondary structure content
during the simulations using DSSP. We found that the main
structural elements present on the enzyme (a-helix: CD;
b-sheet: CBM) were very stable throughout the entire simula-
tion, independently of the temperature simulated (Fig. S1,
right panel, ESIw). Additionally, we checked the time-evolution
of the solvent accessible surface (SAS) of the cellulase through-
out the simulations. The curves indicate that both hydrophobic
and hydrophilic exposed areas were stable at all temperatures
simulated (Fig. S1, left panel, ESIw). Together these analyses
revealed that despite the high fluctuations observed at 350 K,
the enzyme’s secondary structure and SAS for all the tempera-
tures simulated were stable during the entire simulations.
Thus, these results confirm the experimentally demonstrated
thermostability of this protein as well as the absence of a
denaturing process during the 50 ns MD.
Next, we compared the root mean square fluctuations
(RMSF) of cellulase C-a atoms during the MD for each
condition simulated (Fig. 3: A—visually and B—numerically).
While at 300 and 325 K the RMSF values were very similar, at
350 K the fluctuations were considerably higher. Interestingly,
although the magnitude of the fluctuations was considerably
increased at high temperatures, as expected, the same fluctua-
tion pattern of cellulase residues was maintained in all
simulations performed. This result showed that CD is struc-
turally more stable than CBM. This indicates that temperature
increase influences enzyme flexibility in a global fashion,
instead of only inducing local perturbations. In addition, this
observation is consistent with the flexibility pattern inferred
from thermal B-factors obtained in the crystal structure6
(Fig. S2, ESIw).
Motions between CBM and CD govern protein shape changes
To assess if the observed displacements implicated in confor-
mational changes (shape changes) in the enzyme, we analyzed
the radius of gyration (Rg) of cellulase in each condition
(Fig. 4A). As expected from the RMSD analysis, at 300 K it
was not possible to observe significant changes in the Rgduring
the simulation. However, at 325 K we clearly observed a
structural transition starting at 10 ns, when the Rg fell to
2.85 nm and then jumped to 2.95 nm. Finally, at 350 K we
observed two prominent changes: one occurred at 17 ns, when
to the residue RMS fluctuation values (as in the legend) during the
50 ns MD of the apo Cel9A-68 cellulase from T. fusca. Besides
the differential colors, the thickness of the tube is also proportional
to the fluctuations. The principal domains are highlighted: in pink the
CBM, in black the ‘‘linker’’ and in blue the CD. In B, RMS fluctua-
tions plot for the MD at 300, 325 and 350 K colored as in Fig. 2.
Protein residues are numbered from 1–445 for CD, 446–461 for the
linker and 462–605 for the CBM domain.
In A, the visualization of backbone flexibility colored accordingly
the distance between the centers-of-mass of CD and CBM domains of
Cel9A-68 cellulase from T. fusca during the MD. In A, the calculation
of the protein Rgand in B the time evolution of the distance between
the centers-of-mass of CD and CBM domains of cellulase during the
50 ns of each MD, colored as in Fig. 2.
High correspondence between the radius of gyration (Rg) and
This journal is c the Owner Societies 2011 Phys. Chem. Chem. Phys., 2011, 13,13709–1372013713
the Rgstarted to increase from 2.95 nm to 3.1 nm and remained
stable until 40 ns; subsequently, the Rgfell to 2.98 nm and
remained close to this value until the end of the simulation.
Since the domain organization of cellulase is crucial for its
activity, we compared the distance between the centers of mass
of the two domains (CBM and CD) during the simulations. As
seen in Fig. 4B, at 300 K this distance was stable at approxi-
mately 5 nm throughout the entire simulation. On the other
hand, at 325 K we observed an interdomain approximation
(5 nm to 4.8 nm) at 13 ns followed by a rapid return to the
initial value which remained stable until the end of the
simulation. Moreover, at 350 K we observed a progressive
domain separation after the first 14 ns of simulation until a
distance around 5.4 nm is reached. From this point, the
interdomain distance remained stable during approximately
20 ns until it dropped down to 5.2 nm in the last 10 ns of the
Remarkably, the conformational transitions at high tempera-
tures (325 and 350 K) revealed by clustering and Rganalysis
(Fig. 2C and Fig. 4A, respectively) were coincident with the
fluctuations in the inter-domain distances (Fig. 4B). This
suggests that large scale domain motions may be the principal
event that underlies the conformational changes occurred
under these conditions.
Essential dynamics analysis of the MD trajectories
The essential subspace. In order to obtain a deep evaluation
of relevant large-scale motions occurred in the MD we applied
the quasi-harmonic analysis,31–33principal component analysis
(PCA) and subsequent essential dynamics (ED) analysis. With
this approach it is possible to obtain the directions and
amplitudes of the dominant protein motions implicated in
the conformational changes.34We thus diagonalized the atomic
positional covariance matrix to obtain the eigenvectors and
corresponding eigenvalues (see Methods section). By ordering
the eigenvalues in descending order, we can obtain the relative
contribution to positional fluctuations of the first principal
components with respect to the total fluctuations. Fig. S3A
(ESIw) shows the cumulative percentage contribution of the
total motions provided by the first 100 principal components
of each system. The first five (largest amplitude) principal
components recovered around: 65% of the total motions
at 300 K; 75% at 325 K and 85% at 350 K. We therefore
selected these five components as the essential subspace and
carried out the subsequent analyses focusing on understanding
the influence of temperature on the motions covered by this
The essential subspace obtained by our ED analysis represents
the dominant motions occurred during the MD trajectories.
To investigate the convergence of this essential subspace we
calculated the root mean square inner product (RMSIP)
between the two halves of each trajectory (see Methods). This
analysis provides an evaluation of how much the essential
motion space of the second half of the simulation is covered by
the subspace spanned by the first one.35Fig. S3C (ESIw) shows
that independently of the temperature simulated, the RMSIP
values were around 0.6–0.7, which indicates a satisfactory
convergence of the essential subspace.35
Fig. 5 shows the RMSF of each trajectory projected onto its
five most representative principal components. The first two
PCs were responsible for most of the atomic fluctuations.
Hence, the fluctuation patterns obtained by projecting the
MD trajectories onto these two components were very similar
(mainly for PC1) to those obtained from the protein backbone
(Fig. 3B), where the CBM presented the higher fluctuations.
Therefore we focused on the analysis of the movements
provided by these two components in each condition simulated.
We controlled the quality of our ED analysis by checking
the cosine content of the first principal components. This
analysis can exclude the interpretation of random diffusion
of atoms as a correlated motion. When the cosine content
of the first PCs is close to 1, the motions observed are
representative of a random diffusion behavior.36Additionally,
the cosine content analysis can provide information about
tures projected onto the five principal components. The principal domains
are highlighted as in Fig. 3. Plots are colored as in Fig. 2.
RMS fluctuations of the cellulase trajectories at distinct tempera-
13714Phys. Chem. Chem. Phys., 2011, 13,13709–13720 This journal is c the Owner Societies 2011
conformational changes occurred during the MD.36,37Our
analysis revealed that the cosine content of PC1 was low at all
temperatures simulated (Fig. S3B, ESIw), thus confirming that
motions described by the first components represent actual
conformational transitions. While at 300 K the cosine content
was very low during all the simulations (less than 0.1), at
325 K we observed a peak at 20 ns in which this value reached
0.23 and then fell down to almost 0 until the end of the
simulation. At 350 K we observed two peaks with a maximum
value of the cosine content at 25 ns (0.4), and then the values
fell down as observed at 325 K. Remarkably, these peaks
were coincident with the conformational transition observed
previously by clustering (Fig. 2C), Rg(Fig. 4A) and inter-
domain distance (Fig. 4B) analyses. The cosine content of the
second principal component was also very low in all systems
simulated but it was higher than for PC1.
Biologically relevant motions. Large-amplitude collective
motions are important for describing long-timescale dynamics
of proteins, consisting in many cases of domain motions that
are related to biological function.30We inspected the dominant
motions occurred during the MD simulations by visualizing
the two extreme projections along the MD trajectory on the
average structure for PC1 and PC2 (Fig. 6). In this represen-
tation, the directions of C-a movements provided by these
extreme projections for PC1 and PC2 are represented by
arrows describing the mode directions (eigenvectors) for each
residue (Ca) in which their length is proportional to their
eigenvalues. The first PC represents the dominant motion
occurred during the MD. For all the temperatures, PC1 corres-
ponded toahingebendingmotion thatresultsina conformational
change implicating in an inter-domain approximation. Exami-
nation of the PC2 revealed a twisting motion in which the two
domains moved in planes parallel to each other (Fig. 6).
Remarkably, we observed a notable motion of residue Y420,
which is, along with W313, postulated to provide a platform
for the accommodation of the substrate6(data not shown).
Additionally, we carried out an analysis of cross correlation
coefficients between pairs of residues (Fig. S4, ESIw). This
analysis is largely applied in the identification of correlated
collective motions.38The correlation matrix represents the
linear correlation between pairs of C-a atoms as they move
about their average positions during dynamics. Positive
correlations are related to motions occurring in the same
direction whereas negative correlations indicate opposite
directions. Comparing the correlation matrices of the principal
components, which describe the hinge bending motion (PC1),
we found a similar pattern of both intra and inter-domain
correlated motions, at all of the simulated temperatures.
Furthermore, we measured the overlap between the first
principal components of the simulations at distinct tempera-
tures. Table 1 shows the overlap values for the first single PCs
and the RMSIP between the first five components (essential
subspace) of each simulation. According to previous visual
inspection of the projections of the trajectories onto the first PCs,
we observed that PC1 represents a very similar hinge bending
motion for all the temperatures simulated (Fig. S4, ESIw).
These similarities were confirmed by the observation of a high
overlap value between PC1 for all temperatures (Table 1):
300 and 325 K (0.69); 300 and 350 K (0.85) and 325 K and
350 K (0.72). We also observed high overlap values between
the principal components that describe the twisting motion
(PC2). Taken together, the similar global patterns of correlated
motions revealed by the inspection of the cross correlation
matrices and the analysis of the overlap between the first
PCs obtained at distinct temperatures strongly suggest that
the overall dynamics of cellulase is dominated by collective
motions that are intrinsic to the structure.
These results show a globally intrinsic dynamics between the
CD and CBM domains, in which the hinge bending motion
drives the most representative conformational transition
during the simulation. Ting et al. previously proposed that
the optimal hydrolysis rate would be obtained at a transition
between a compressed and an extended conformation through
a cooperative process.9Our results are consistent with this
hypothesis and provide a more detailed view revealing that the
dominant movement that occurred at all of the temperatures
simulated was a hinge bending motion that drives the transi-
tion from an initial ‘‘open’’ configuration to a ‘‘closed’’ state.
Irwin et al. have compared some functional properties of the
Cel9A-68 cellulase with other constructs with distinct domain
compositions.39In that occasion the relevance of CBM to the
activity and processivity of the enzyme was discussed. Further,
the ability of Cel9A to synergize with other cellulases from
T. fusca (Cel6B and Cel5A) was demonstrated. Although the
CBM domain is not directly involved in catalysis, constructs
lacking this domain are less functional. Thus the identification
of an intrinsic pattern of interdomain correlated motions
(Fig. S4, ESIw) provides a better understanding of why the
presence of a type IIIc CBM, as found in Cel9A-68, is crucial
to allow the access of the enzyme to internal cleavage sites
which was previously hypothesized but without a structural
provided by the first two PCs. The directions and amplitudes of the
motions are represented by red arrows.
Visualization of cellulase motions at distinct temperatures
This journal is c the Owner Societies 2011Phys. Chem. Chem. Phys., 2011, 13,13709–1372013715
Experimental results showed that Cel9A-68 cellulase activity
is increased at high temperatures.39,40Herein we demonstrate
an enhancement of the principal motions in such conditions
(Fig. 5), corroborating with the experimental data. Therefore
we suggest that collective motions intrinsic to the Cel9A-68
cellulase structure could be directly implicated in enzyme
Temporal conformational sampling along PC1/PC2 subspace.
To further understand the dynamics of cellulase Cel9A-68 at
distinct temperatures, we examined the conformational space
sampled by the two first principal components. Fig. 7 shows the
projections onto PC1 and PC2 of the MD trajectory for all
temperatures simulated. Here, each point represents a different
substate adopted by the enzyme during the simulations. We
highlighted the temporal sequence of frames to understand the
transitions between conformational states. The crystallographic
structure of the Cel9A-68 cellulase was also projected into these
conformational spaces (yellow triangle), illustrating the corres-
pondence between this structure and those adopted during the
simulations (Fig. 7).
Considering the conformational space sampled by the
Cel9A-68 backbone in the simulations, we observed that the
extent of sampling was significantly changed in each condition.
Firstly, we found that all conformations visited during MD at
300 K were located in a small region of the conformational
space. This is in agreement with the RMSD and clustering
results (Fig. 2B and C, respectively) showing that at this
temperature the system remained very stable during the MD
simulation. On the other hand, at 325 K, after transiting
between conformational clusters in the vicinity of the starting
structure (during the first 20 ns), we observed a progressive
transition towards a more distant region of the conformational
space (Fig. 7). Finally, at 350 K, we observed a similar behavior
but the extent of sampling was considerably higher.
Free energy landscape (FEL). The presented results of our
essential dynamics analysis revealed that the dynamics of cellulase
is governed by large scale collective motions. We have
also shown the convergence of the selected essential subspace
obtained by bi-dimensional projection of the trajectories onto the first
two PCs. In this representation each point represents a transient
conformer. The color-code represents the temporal sequence of frames.
Conformational sampling of cellulase at distinct temperatures
MD at 300, 325 and 350 K
Overlap and RMSIP between the PC eigenvectors from the
The values 40.45 are highlighted in grey.
13716Phys. Chem. Chem. Phys., 2011, 13,13709–13720 This journal is c the Owner Societies 2011
(first five principal components) and that the most relevant
motions presented high overlap values when comparing the
first two PCs obtained at distinct temperatures (Table 1). With
these results in mind, we were able to extract thermodynamic
properties from our simulations, such as free energy landscapes
(FEL), which may be derived as ensemble averages of a
sufficiently converged trajectory.41
Since the first PCs capture most of the total displace-
ment from the average protein structure throughout the
simulation,34,42we performed a free energy landscape (FEL)
analysis using as reaction coordinates the first two PCs obtained
by ED analysis. Fig. 8 shows the FEL for each temperature
simulated and also the structural alignment of the initial and
the lowest energy structure of each trajectory. At 300 K,
we found only one minimum centered in a small region
of the landscape accessed by cellulose in this condition. Visual
inspection of the lowest energy structure revealed a close
similarity to the starting structure (inset in Fig. 8). As a whole,
the projection of the trajectory onto its two first PCs and
subsequent FEL analysis suggests that the system at this
temperature was trapped at local minima.
Projection of the C-a trajectories at 325 and 350 K onto its
two principal components revealed a progressive conforma-
tional transition towards a distinct state (Fig. 7). Interestingly,
at 325 K we observed that the lowest energy region of the FEL
is far from the initial structure. Further, the access to this
region was only achieved at approximately 35–40 ns, suggesting
that the fluctuations observed during the first half of MD can
be described as transitions between metastable states (Fig. 8).
Visual inspection of the lowest energy structure revealed that
compared to the initial structure, it presents a more compact
conformation in which the CD and CBM domains are closer
to each other. The structural alignment of the lowest energy
structure and the initial one suggests that the access to this new
conformation would be mediated mainly by a hinge bending
motion, which was revealed as a principal motion (Fig. 8).
The FEL obtained at 350 K revealed that the lowest energy
region of the landscape was accessed at approximately 40 ns.
Again, the fluctuations occurred in the initial half of the
trajectory can be described as transitions between metastable
states. While at 325 K the system fluctuated around the basin
reached at 35–40 ns, at 350 K, the system jumped off the
minima and moved towards a distant region of the conforma-
tional space. Therefore, at this temperature, the system visits
low energy conformations that resemble those observed at
325 K. In addition, at 350 K the protein could jump off these
minima and explore additional regions in the conformational
space, even without reaching another region of minima.
Putative sites for engineering more effective cellulases.
Mutagenesis studies revealed the importance of selected Cel9A
residues in the catalytic mechanism of the enzyme.43The
authors bring out not only the roles of E424, D55 and D58
in catalysis, but also the importance of the aromatic residues
Y206 and W318 on binding to cellulose. A more recent study
confirmed these findings and also disclosed the importance of
both domains for processivity and binding.44Further, it was
also shown that a combination of mutations in both domains
(CBM and CD) may increase enzymatic activity on bacterial
cellulose and swollen cellulose.45Overall, these studies were
conducted focusing on describing mutations directly involved
in interactions with cellulose, rather than searching for muta-
tions that might alter the dynamics of Cel9A. Considering
that flexibility cannot be neglected, we discuss putative muta-
tion sites outside the active site aiming to modulate Cel9A
structures from the MD trajectories. The FELs were obtained using
as reaction coordinates the projections of cellulase C-a atoms onto the
first two principal components. Free energy values are given in kcal mol?1
as indicated by the color bar. For each plot, the representative
structures are represented in cartoon and are colored according to
the cellulase domain definitions: in red the CD; in blue the linker and
in green the CBM. The trajectories points (in ns) corresponding to
each structure are also indicated.
Free energy landscape (FEL) analysis and representative
This journal is c the Owner Societies 2011Phys. Chem. Chem. Phys., 2011, 13,13709–1372013717
collective motions, which according to our results could alter
binding and processivity of the enzyme.
To improve the understanding of the mechanism of inter-
domain approximation, we carried out a more detailed analysis
of this hinge motion with the Dyndom program.46Using
as input the extreme conformations from the PC1 at 350 K
(see Fig. 5) and no prior domain definitions, this analysis
identified the CD domain as ‘‘fixed’’ while CBM as the ‘‘moving’’
domain (Fig. 9). The moving domain (CBM) rotates 85.21
relative to the fixed domain. Interestingly, the residues of the
linker region (458–461) were assigned as bending. The approxi-
mation of CBM to CD is facilitated by a 23.71 rotation about
the j-dihedral axis of D459 and a 76.61 rotation about the
G460 c-dihedral axis. The adjacent regions to these two
bending residues were considerably more rigid.
Hinge bending motions were extensively described in a
variety of proteins, governing global motions including opening
and closure mechanisms of active site clefts.47–51Veltman et al.
have demonstrated the importance of conserved glycine residues
in hinge regions of thermolysin-like proteases.50Indeed, glycine
residues provide local flexibility required for a hinge bending
motion to occur, since replacement of a phenylalanine by a
glycine in Epac2 significantly increases hinge transitions and
Our results provide a structural/dynamical basis to guide
future mutational studies in which the modulation of the
overall flexibility/rigidity of the enzyme can be achieved by
changing key residues by others with distinct physicochemical
properties. Based on previous studies49–51and mostly in our
domain motion analysis, as G460 was assigned as the hinge
point we expect that mutations on this residue would result in
a drastic decrease of overall flexibility and consequently reduce
the enzymatic activity. For example, to investigate experi-
mentally the validity of our findings, we propose the insertion
of the G460P mutation on the Cel9A. This mutation should
constrain this hinge motion and have a direct negative influence
on the enzyme kinetics since proline residues give an exceptional
conformational rigidity in contrast to glycine.
Ting et al. hypothesized that a longer linker would result in
a higher hydrolysis rate.9Following this assumption, we propose
the insertion of an adjacent glycine residue into G460, once a
similar hinge point containing two adjacent glycines is found
also in thermolysin-like proteases,50thus facilitating functional
motions such as the interdomain approximation. This muta-
tion could also confer a higher activity to Cel9A at room
temperature, since our results revealed that the increase in
overall enzyme flexibility at 325 K and 350 K may be an
important event to explain at least in part the higher activity
observed at high temperatures.
Thermophilic proteins are believed to be more rigid at room
temperature which is generally paralleled by a decreased
activity in such conditions.52From the crystal structure of
the Cel9A-68 cellulase, we performed MD simulations at three
different temperatures to understand the thermal influence on
enzyme flexibility and to gain insights into the mechanisms
intrinsic to the Cel9A-68 cellulase structure possibly involved
in its activity. As expected, the higher the temperature the
higher the protein atomic displacements (Fig. 2). However, the
overall pattern of deviations was not random even at high
temperatures (Fig. 3). The CBM was shown to be more flexible
than CD in a consistent fashion with the flexibility inferred
from crystallographic B-factors (Fig. S2, ESIw). This observa-
tion is also in agreement with previous studies reporting a
certain degree of rigidity in the proteins’ active site, while
regions that are involved in binding and recognition are taken
as more flexible domains.53
Although the large displacements were seen at 350 K MD,
the Cel9A-68 cellulase secondary structure and SASA remained
stable during all the simulations. This corroborates the experi-
mental results confirming the thermostable behavior of the
enzyme.54Asreportedin other studies, the structural transitions
observed in all simulations were governed by inter-domain
motions, as shown by the radius-of-gyration, inter-domain
distance analysis and PCA.
Essential dynamics analysis allowed us to identify that few
degrees of freedom are responsible for the large amplitude
collective motions enhanced at high temperatures (Fig. S4, ESIw).
Interestingly, we identified a conserved pattern of interdomanial
correlated displacements, at all temperatures simulated, which
provides a structural basis to understand the importance of
CBM to Cel9A activity although this domain is not directly
involved in catalysis.
Regarding the first two components, the motions provided
by PC1 and PC2 were very similar at all temperatures simulated,
as confirmed by the overlap calculations. The most represen-
tative motion describes a very collective bending movement,
which involves rotations about backbone torsion angles j
and c of residues D459 and G460 (Fig. 9). This finding
simulation at 350 K. We used as input the extreme conformations
from the PC1 and represented in cartoon. The initial projection is
represented with high transparency and the final conformation is
colored as in Fig. 2. The hinge bending residues are colored blue
and have represented their surfaces. The lines cross at the centre of
rotation, thus the bending axis is perpendicular to this crossing point.
Dyndom analysis of the hinge bending motion of the Cel9A-68
13718Phys. Chem. Chem. Phys., 2011, 13,13709–13720This journal is c the Owner Societies 2011
complements data obtained by Ting et al.9and provides a
structural understanding of the cooperative process between
CD and CBM. We suggest the involvement of such bending
motion in the activity and processivity of the enzyme, since
these properties (as well this motion) are enhanced at high
temperatures.40,54Future design studies may be carried out
targeting the modulation of this functional motion by the
development of cellulases with distinct linker composition
taking advantage of the flexibility provided by G460, which
was found to be involved directly in the hinge bending motion
The occurrence of a hinge bending transition is postulated
to be related to rugged energy landscapes in which low energy
barriers separate the minima wells.47Our FEL analysis
revealed that only at high temperatures Cel9A can reach relevant
conformations not accessible at 300 K, thus evidencing the
existence of energy barriers not transposable at room tempera-
ture but important for activity (and the hinge transition) under
extreme conditions (Fig. 8). This observation raises the
question if other thermophilic enzymes would present a similar
Previous data revealed that cellulases from thermophilic
organisms tend to outperform their mesophilic counterparts,
which highlights the benefits in using this class of enzymes.40
Our data may be crucial for the development of more efficient
cellulases through mutating the linker region, with the aim of
achieving overall changes in the flexibility of the enzyme that
modulates conformational changes important for activity.
Molecular dynamics (MD) simulations, energy minimization
and trajectory analyses were carried out with the GROMACS
4 package55,56using the GROMOS96 (G53a6) forcefield.57We
used the apo Cel9A-68 cellulase of the T. fusca crystallo-
graphic structure as initial coordinates (PDB code 1TF4), as
well as the crystallographic water coordinates. We removed
the Ca+2ions and we solvated the system with explicit SPC58
water molecules, in which a 0.1 nm layer of water molecules
was added around the solute molecules within a triclinic water
box, using periodic boundary conditions. Counter ions were
inserted for system neutralization in addition to NaCl in
0.15 M final concentration. LINCS59and SETTLE60were
applied to constrain solute and solvent bonds, respectively.
The temperature was kept at 300, 325 or 350 K with a
canonical rescaling approach;61and the pressure at 1 atm
using the Berendsen approach.62Electrostatic interactions
were calculated with the PME method,63using non-bonded
cutoffs of 1.0 nm for Coulomb and 1.2 nm for van der Waals.
The MD integration time-step was 2 fs.
A three-step energy minimization protocol was used to avoid
artifacts in atomic trajectories due to conversion of potential
into kinetic energies. Firstly, we applied the steepest-descent
algorithm: (i) 5000 steps with protein heavy atom positions
restrained to their initial positions using a harmonic constant
of 1 kJ mol?1nm?1in each Cartesian direction, allowing
unrestrained water and hydrogen movement; and (ii) 5000 steps
with all atoms free to move; subsequently, (iii) the conjugated-
gradient algorithm was applied for further energy minimization
until an energy gradient of 42 kJ mol?1nm?1is reached.
After that we performed a heating procedure from 20 K to
300, 325 or 350 K, for each temperature simulated. For this
proposal we performed a 500 ps MD using a ‘‘reverse’’
simulated annealing procedure keeping the protein heavy
atoms restrained to their initial positions (using the same
previous harmonic restraint potential). So the velocities were
sorted for an initial temperature of 20 K and we used the
‘‘annealing’’ option on the ‘‘.mdp’’ gromacs file to heat gradually
the system until the requested temperature (300, 325 or 350 K) is
reached, in contrast to the conventional cooling protocol.
Subsequently we performed an equilibration procedure consis-
ting in a preliminary MD (1.5 ns) reducing gradually the
positional restraint potential, in steps from 50 to 0 kJ mol?1nm?1
as follows: 100 ps MD steps for each potential—50, 25, 10, 5,
2.5, 1, 0.5, 0.25, 0.1 and 0.05 kJ mol?1nm?1—and 500 ps
with no positional restraint. This procedure permits the system
to achieve solvent equilibration and avoid artifacts as discussed.
Electrostatic surface analysis
Electrostatic surface calculations were performed using APBS
software.26This analysis combines the solvent accessible
surface (SAS) calculation with the values of the electrostatic
potential for each atom at the surface. This latter calculation is
obtained by the resolution of the linearized Poisson–Boltzmann
equation to obtain the surface charge distribution. Images were
generated using PyMol.64
Principal component analysis (PCA) and ED analysis
We applied the g_covar module of GROMACS package to
obtain the covariance matrix of C-a atomic positions from the
50 ns trajectory of the cellulase. Rotation and translation
motions were removed prior to covariance matrix calculation
by least-squares superposition to the averaged-structure. Each
element of the covariance matrix C is represented by:31,33
Cij= hqi? hqiiihqj? hqjii,(1)
where qiand qjare the internal coordinates of atoms i and j.
All analyses were performed with the g_anaeig module of
Cosine content analysis of the first PCs
As often the first few principal components of simulations
of large proteins resemble cosines we evaluated the cosine
content ofthe first two principal components to exclude the inter-
pretation of the random diffuse motions.65
The cosine content ci of the principal component pi is
where T is the total simulation time. When ciis close to 1 the
large amplitude motions are not connected with the potential
energy landscape but with random diffusion. It has been
demonstrated that insufficient sampling can also lead to high
civalues, representative of random motions. Its analysis was
carried out with the g_analyse module of GROMACS.
This journal is c the Owner Societies 2011 Phys. Chem. Chem. Phys., 2011, 13,13709–1372013719
Overlap between PC motions
The overlap (O) between a given PC vector M and another PC
vector X is evaluated by their normalized projection,
O = M?X/JMJJXJ
where M is a PC mode vector from a MD at a particular
temperature and X a mode vector at a different temperature.
A perfect match yields an overlap value of 1.
We assume that the essential subspace of each system was
defined by the five eigenvectors with higher eigenvalues and
the overlap between the essential subspace of two different
groups was obtained from the RMSIP:
where niand vjare the eigenvectors of different simulations
(or subparts of the same simulation). The RMSIP measures
how well the subspace defined by a given set of modes (here we
consider the five lowest-frequency PC modes) from a system
(or a part of a MD trajectory) can include the motion
indicated by the essential subspace from the other given system
(or from other parts of the same MD).
Cross correlation analysis
We projected the MD trajectory onto the first principal
component, corresponding to the largest eigenvector, of the
covariance matrix in order to visualize the extreme structures
and the major fluctuations of the correlated motions. The
correlation matrix, an N ? N array whose i–j entry Corrij
summarizes the correlation between the motions of atoms
i and j, is obtained from the reduction and normalization of
the covariance matrix as follows:
i;aveÞ ? ðr
We applied the g_cluster module of GROMACS package to
calculate the RMS clusters of cellulase backbone conformations,
using the method of simple linkage (nearest neighbor) with a
RMSD cut-off of 0.11 nm.
Free energy landscape (FEL)
We applied the g_anaeig and g_sham modules of GROMACS
package to calculate the two-dimensional representation of
Ga ¼ ?KBT ln
where KBis the Boltzmann constant, T is the temperature of
simulation, P(qa) is an estimate of the probability density
function obtained from a histogram of the MD data and Pmax(q)
is the probability of the most probable state. Considering two
different reaction coordinates qiand qjthe two-dimensional
free energy landscapes were obtained from the joint probability
distributions P(qi, qj) of the system.
Domain motions analysis
Domain motions were analyzed with the DynDom software.46
This program compares two available structures of the same
protein and deduces the rigid-body movement of one domain
(the moving domain) relative to the other domain (the fixed
domain). We selected the extreme projections of the first principal
component (PC1) to be analyzed.
The authors declare no conflict of interest.
All authors wish to acknowledge the Brazilian agencies CNPq,
CAPES, FAPERJ, INBEB and CENPES-ANP for financial
1 C. M. Gowen and S. S. Fong, Chem. Biodiversity, 2010, 7,
2 M. E. Himmel, S. Y. Ding, D. K. Johnson, W. S. Adney,
M. R. Nimlos, J. W. Brady and T. D. Foust, Science, 2007, 315,
3 S. Pe ´ rez, D. Samain and H. Derek, Advances in Carbohydrate
Chemistry and Biochemistry, Academic Press, 2010, vol. 64,
4 M. Schulein, Biochim. Biophys. Acta, 2000, 1543, 239–252.
5 A. B. Boraston, D. N. Bolam, H. J. Gilbert and G. J. Davies,
Biochem. J., 2004, 382, 769–781.
6 J. Sakon, D. Irwin, D. B. Wilson and P. A. Karplus, Nat. Struct.
Biol., 1997, 4, 810–818.
7 B. Henrissat, T. T. Teeri and R. A. J. Warren, FEBS Lett., 1998,
8 N. A. Spiridonov and D. B. Wilson, J. Bacteriol., 1998, 180,
9 C. L. Ting, D. E. Makarov and Z. G. Wang, J. Phys. Chem. B,
2009, 113, 4970–4977.
10 W. R. Scott and C. A. Schiffer, Structure (London), 2000, 8,
11 E. Kim, S. Jang and Y. Pak, J. Chem. Phys., 2008, 128,
12 W. Treptow, S. J. Marrink and M. Tarek, J. Phys. Chem. B, 2008,
13 L. Tavagnacco, P. E. Mason, U. Schnupf, F. Pitici, L. Zhong,
M. E. Himmel, M. Crowley, A. Cesaro and J. W. Brady, Carbohydr.
Res., 2011, 346, 839–846.
14 J. Li, L. Du and L. Wang, J. Phys. Chem. B, 2011, 114,
15 M. Alahuhta, Q. Xu, Y. J. Bomble, R. Brunecky, W. S. Adney,
S. Y. Ding, M. E. Himmel and V. V. Lunin, J. Mol. Biol., 2011,
16 M. R. Nimlos, J. F. Matthews, M. F. Crowley, R. C. Walker,
G. Chukkapalli, J. W. Brady, W. S. Adney, J. M. Cleary, L. Zhong
and M. E. Himmel, Protein Eng., Des. Sel., 2007, 20, 179–187.
17 G. Andre, P. Kanchanawong, R. Palma, H. Cho, X. Deng,
D. Irwin, M. E. Himmel, D. B. Wilson and J. W. Brady,
Protein Eng., 2003, 16, 125–134.
18 L. Bu, G. T. Beckham, M. R. Shirts, M. R. Nimlos, W. S. Adney,
M. E. Himmel and M. F. Crowley, J. Biol. Chem., 2011.
19 O. V. Oliveira, L. C. Freitas, T. P. Straatsma and R. D. Lins,
J. Mol. Recognit., 2009, 22, 38–45.
20 J. M. Escovar-Kousen, D. Wilson and D. Irwin, Appl. Biochem.
Biotechnol., 2004, 113–116, 287–297.
21 A. P. Valente, C. A. Miyamoto and F. C. Almeida, Curr. Med.
Chem., 2006, 13, 3697–3703.
22 L. C. James and D. S. Tawfik, Trends Biochem. Sci., 2003, 28,
23 P. J. Haney, J. H. Badger, G. L. Buldak, C. I. Reich, C. R.
Woese and G. J. Olsen, Proc. Natl. Acad. Sci. U. S. A., 1999, 96,
24 T. Simonson, Rep. Prog. Phys., 2003, 66, 737–787.
13720Phys. Chem. Chem. Phys., 2011, 13,13709–13720 This journal is c the Owner Societies 2011 Download full-text
25 R. Sterner and W. Liebl, Crit. Rev. Biochem. Mol. Biol., 2001, 36,
26 N. A. Baker, D. Sept, S. Joseph, M. J. Holst and J. A. McCammon,
Proc. Natl. Acad. Sci. U. S. A., 2001, 98, 10037–10041.
27 A. Liwo, C. Czaplewski, S. Oldziej and H. A. Scheraga, Curr. Opin.
Struct. Biol., 2008, 18, 134–139.
28 I. Bahar, T. R. Lezon, A. Bakan and I. H. Shrivastava, Chem. Rev.,
2010, 110, 1463–1497.
29 J. Janin and F. Rodier, Proteins, 1995, 23, 580–587.
30 P. R. Batista, C. H. Robert, J. D. Marechal, M. Ben Hamida-Rebai,
P. G. Pascutti, P. M. Bisch and D. Perahia, Phys. Chem. Chem.
Phys., 2010, 12, 2850–2859.
31 M. Karplus and J. N. Kushick, Macromolecules, 1981, 14,
32 R. M. Levy, D. Perahia and M. Karplus, Proc. Natl. Acad. Sci.
U. S. A., 1982, 79, 1346–1350.
33 R. M. Levy, M. Karplus, J. Kushick and D. Perahia, Macromolecules,
1984, 17, 1370–1374.
34 A. Amadei, A. B. Linssen and H. J. Berendsen, Proteins, 1993, 17,
35 A. Amadei, M. A. Ceruso and A. Di Nola, Proteins, 1999, 36,
36 B. Hess, Phys. Rev. E: Stat. Nonlinear Soft Matter Phys., 2002,
37 J. F. Taly, A. Marin and J. F. Gibrat, BMC Bioinf., 2008, 9, 6.
38 P. H. Hunenberger, A. E. Mark and W. F. Vangunsteren,
J. Mol. Biol., 1995, 252, 492–503.
39 D. Irwin, D. H. Shin, S. Zhang, B. K. Barr, J. Sakon, P. A. Karplus
and D. B. Wilson, J. Bacteriol., 1998, 180, 1709–1714.
40 F. Mingardon, J. D. Bagert, C. Maisonnier, D. L. Trudeau and
F. H. Arnold, Appl. Environ. Microbiol., 2011, 77, 1436–1442.
41 S. Hayward and B. L. de Groot, Molecular Modeling of Proteins,
2008, vol. 443.
42 G. Maisuradze, A. Liwo and H. Scheraga, J. Chem. Theory
Comput., 2010, 6, 583–595.
43 W. Zhou, D. C. Irwin, J. Escovar-Kousen and D. B. Wilson,
Biochemistry, 2004, 43, 9655–9663.
44 Y. Li, D. C. Irwin and D. B. Wilson, Appl. Environ. Microbiol.,
2007, 73, 3165–3172.
45 Y. Li, D. C. Irwin and D. B. Wilson, Appl. Environ. Microbiol.,
2010, 76, 2582–2588.
46 S. Hayward and H. J. Berendsen, Proteins, 1998, 30, 144–154.
47 N. Sinha, S. Kumar and R. Nussinov, Structure (London), 2001, 9,
48 B. L. de Groot, S. Hayward, D. M. van Aalten, A. Amadei and
H. J. Berendsen, Proteins, 1998, 31, 116–127.
49 A. J. Stokka, R. N. Carvalho, J. F. Barroso and T. Flatmark,
J. Biol. Chem., 2004, 279, 26571–26580.
50 O. R. Veltman, V. G. Eijsink, G. Vriend, A. de Kreij, G. Venema
and B. Van den Burg, Biochemistry, 1998, 37, 5305–5311.
51 T. Tsalkova, D. K. Blumenthal, F. C. Mei, M. A. White and
X. Cheng, J. Biol. Chem., 2009, 284, 23644–23651.
52 R. Jaenicke, Proc. Natl. Acad. Sci. U. S. A., 2000, 97, 2962–2964.
53 I. Luque and E. Freire, Proteins: Struct., Funct., Genet., 2000, 41,
54 D. Irwin, D. H. Shin, S. Zhang, B. K. Barr, J. Sakon, P. A. Karplus
and D. B. Wilson, J. Bacteriol., 1998, 180, 1709–1714.
55 B. Hess, C. Kutzner, D. Van Der Spoel and E. Lindahl, J. Chem.
Theory Comput., 2008, 4, 435.
56 D. Van der Spoel, E. Lindahl, B. Hess, G. Groenhof, A. E. Mark
and H. J. C. Berendsen, J. Comput. Chem., 2005, 26, 1701–1718.
57 C. Oostenbrink, A. Villa, A. E. Mark and W. F. van Gunsteren,
J. Comput. Chem., 2004, 25, 1656–1676.
58 H. J. C. Berendsen, J. P. M. Postma, W. F. Van Gusteren,
J. Hermans, in Intermolecular Forces, ed. Pullman, Dordrecht,
Reidel, 1981, 331–342.
59 B. Hess, H. Bekker, H. J. C. Berendsen and J. G. E. M. Fraaije,
J. Comput. Chem., 1997, 18, 1463–1472.
60 S. Miyamoto and P. A. Kollman, J. Comput. Chem., 1992, 13,
61 G. Bussi, D. Donadio and M. Parrinello, J. Chem. Phys., 2007,
62 H. J. C. Berendsen, J. P. M. Postma, W. F. Vangunsteren,
A. Dinola and J. R. Haak, J. Chem. Phys., 1984, 81, 3684–3690.
63 E. Ulrich, P. Lalith, L. B. Max, D. Tom, L. Hsing and G. P. Lee,
J. Chem. Phys., 1995, 103, 8577–8593.
64 W. L. Delano, The PyMOL Molecular Graphics System,
Version 1.2r3pre, Schro ¨ dinger, LLC.
65 B. Hess, Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat. Interdiscip.
Top., 2000, 62, 8438–8448.
66 E. Papaleo, L. Russo, N. Shaikh, L. Cipolla, P. Fantucci and
L. De Gioia, J. Mol. Graphics Modell., 2010, 28, 834–841.