Research Article

Subsets of Slow Dynamic Modes Reveal

Global Information Sources as Allosteric
Sites

Bengi Altintel 1,2, Burcin Acar 2, Burak Erman 3 and Turkan Haliloglu 1,2⇑

1 - Department of Chemical Engineering, Bogazici University, 34342 Istanbul, Turkey

2 - Polymer Research Center, Bogazici University, 34342 Istanbul, Turkey
3 - Department of Chemical and Biological Engineering, Koc University, 34450 Istanbul, Turkey

Correspondence to Turkan Haliloglu: Department of Chemical Engineering, Bogazici University, 34342 Istanbul,
Turkey. bengi.altintel@boun.edu.tr (B. Altintel), burcinacar@boun.edu.tr (B. Acar), berman@ku.edu.tr (B. Erman),
halilogt@boun.edu.tr (T. Haliloglu), @bengialtintel (B. Altintel), @neseruzgari (B. Acar)
https://doi.org/10.1016/j.jmb.2022.167644
Edited by Ruth Nussinov

Abstract
Allostery is a key biological control mechanism, and dynamic information flow provides a perspective to
describe allosteric interactions in causal relationships. Here, as a novel implementation of the Gaussian
Network Model (GNM) based Transfer Entropy (TE) calculations, we show that the dissection of dynamic
information into subsets of slow dynamic modes discloses different layers of multi-directional allosteric
pathways inherent in a given protein structure. In these subsets of slow modes, the degree of collectivity
(Col) in the information transfer of residues with their TE values (TECol score) identifies distinct residues
as powerful effectors, global information sources; showing themselves with a high dynamic capacity to
collectively disseminate information to others. As exemplified on aspartate transcarbamoylase (ATCase),
Na+/K+-adenosine triphosphatase (Na+/K+-ATPase), and human transient receptor potential melastatin 2
(TRPM2) along with a dataset of 20 proteins, these specific residues are associated with known active
and allosteric sites. These information source residues, which collectively control others and lead allos-
teric communication pathways, hint at plausible binding sites for structure-based rational drug design.
Ó 2022 Elsevier Ltd. All rights reserved.

Introduction in years.5–14 Structural changes upon perturbations

Proteins execute various functions in cells while
adapting to different conditions and recruiting
several partners. This resilience is owed to their
dynamic nature through which protein activity can
be modulated without any direct interactions but
via dynamic allosteric communications with their
active sites. Allostery being an essential means in
regulating functions is of particular interest for
drug design efforts.1–3
Allostery was more of a structural term than a
dynamical one, when first introduced by Jacques
Monod and François Jacob in 1961.4 The role of
dynamics in allostery has become more apparent
are closely associated with dynamic fluctuations. In
the absence of conformational changes, however,
allosteric modulation mainly involves an entropic
effect, i.e., dynamic allostery, and is assumed
through a shift in dynamic modes and their frequen-
cies.10,14–18 Within the network of dynamic interac-
tions, allosteric pathways feature plausible routes
for the communication among functional sites, as
have also been considered in several graph-based
calculations.19–23 Communities of residues on
these interaction networks emerge as viable sites
for global transmission of allosteric signal and dis-
play functional importance.24–26

0022-2836/Ó 2022 Elsevier Ltd. All rights reserved. Journal of Molecular Biology 434 (2022) 167644
B. Altintel, B. Acar, B. Erman, et al.
Here, we use an information-theoretic approach
to illustrate allosteric communications between
residues in proteins. Correlated fluctuations
between residues could be described in terms of
mutual information shared by their fluctuations,
which means that the uncertainty in the
fluctuations of one is reduced with the information
of the fluctuations of another. Then, the allosteric
communication can simply be regarded as an
information transfer from one site to another,
which can be estimated by transfer entropy
(TE).27 TE definition uses Shannon’s entropy as a
measure of information and introduces a time delay
into the mutual correlation of residue movements to
capture causal allosteric interactions. Using transfer
entropy concept, information flow could be
extracted from Molecular Dynamics (MD) trajecto-
ries,28–33 yet the accessible MD simulation lengths
of functional importance are mostly questionable.
Recently, a coarse-grained analytic approach
based on Gaussian Network Model (GNM) inte-
grated with Schreiber’s information transfer formu-
lation has been proposed34 that we use in the
present TE calculations (Figure 1, Methods).
GNM predicts complex structural fluctuations in
an ensemble of substates, or in a spectrum of
dynamic modes, defined by the 3D protein
Figure 1. Schematic view of GNM-based informa-
tion transfer. Spheres represent Ca atoms of residues.
R0i and R0j with gray arrows represent equilibrium
position vectors of residues i and j. DRi and DRj with
dashed black arrows are respectively the fluctuation
vectors from the equilibrium positions of residues i and j
at time t and t + s. The information transfer between
residue pairs of i and j with s is described by green
arrows. The difference between Tij and Tji described by
red thunder reveals the net information transfer (TE)
from residue i to j at time s (DTij (s)).
2
Journal of Molecular Biology 434 (2022) 167644
structure (Figure 1). These substates, i.e.,
conformers, such as ligand bound, active/inactive,
and open/closed, are usually intrinsically
accessible particularly via a few slow
modes.10,17,35–38 Slow modes individually and/or
together have the potential to describe unique func-
tional motions by moving various sites simultane-
ously and collectively, i.e., creating
choreographies from basic dynamic elements. To
exemplify, the role of specific slow modes in the
conformational transition of vitamin B12 importer
protein BtuCD between the outward and inward
functional states was computationally illustrated,
which agreed with the observed behavior in the
experimental FRET distances and point muta-
tions.37,39 Thus, certain functional slow modes and
associated allosteric interactions may not be evi-
dent in average slow mode residue fluctuations
and remain hidden within the mask of more domi-
nant slowest/slower mode(s).
Allosteric interactions are embedded in slow
modes with their distinguished cooperativity
enabling concerted action required for the
formation of allosteric pathways. Allosteric effects
thus often arise by triggering or altering these pre-
existing global modes.10,17,37–38 In addition to pro-
viding the conformational transition between stable
functional states, global modes are also correlated
with specific directional allosteric pathways.10,37–
38,40–41
As an example, the slow modes enabling
the functional conformational changes between
the outward and inward facing structures of the
two ATP dependent ABC transporters BtuCD and
MalFGK render opposite directional allosteric com-
munication pathways in accordance with their type-
specific functional requirements.37 Consequently,
the dissection of dynamic behavior, here informa-
tion entropy, into individual or subsets of slow
modes is essential to view all likely causal interac-
tions and allosteric communication pathways inher-
ent in a given topology. We thus search to identify
the subsets of slow modes that maximize residues’
collective information transfer. That is each residue
has a certain degree of information content whose
transferrable information depends on the vibrational
modes of the protein34,37–38. Based on the concept
of transferrable information content and on the the-
ory of Bruschweiler,42 we develop a collectivity
measure that determines how globally a residue
transfers information to the rest of the protein, i.e.,
affects and controls the others. We use the term
global to express the high collectivity in transferring
this information.
Causal correlations accompanying allosteric
interactions between distinct sites rendered by
different subsets of slow modes might be pertinent
to the functional mechanism requiring a certain
type of motion. This may thus contribute to the
spot-on understanding of the allosteric
phenomenon with functional modularity and
enable the prediction of functional sites, either
B. Altintel, B. Acar, B. Erman, et al.
active and/or allosteric, within multi-/bi-directional
allosteric pathways.
Results & discussions
Slow modes maximizing collective information
transfer disclose functional sites as
information source residues
In basics, allosteric communication is an
information exchange between two distant protein
sites, as herein defined by the transfer entropy
measure from information theory.34 Our preliminary
analysis showed that known functional sites (either
active or allosteric residues) have the dynamic
capacity to transfer information collectively to the
others -as global information sources- in distinct
subsets of slow modes. This was a unique dynamic
behavior of functional residues revealed only by
specific slow modes, making them distinguishable
with their allosteric role in leading others and thus
controlling their motion.
To quantify the collectivity of residues in
information transfer and develop a general
protocol, a dynamic measure is defined by
adapting Bruschweiler’s collectivity definition42 to
search and identify the subsets of slow modes from
which it would be possible to reveal functional resi-
dues with high collectivities in the information trans-
fer. We have two -simple and combinatorial- search
schemes (Figure S1, Methods); the simple search
scheme considers the basic dissection of slow
modes into five subsets of slow modes by sequen-
tially removing the slowest modes in each and the
combinatorial search scheme takes the subsets of
the three-to-five slow mode combinations into
account.
Using either the simple or combinatorial search
scheme may yield more than one subset of slow
modes with which different functional sites or
common functional sites in a different dynamic
context could be disclosed. The proposed protocol
is applied on a dataset of 23 proteins (Table S1)
and we have shown that active and allosteric sites
are global information sources in multi-/bi-
directional allosteric pathways with the intrinsic
nature of allosteric communication.43–44
The predictions are elaborated for three
exemplary cases; aspartate transcarbamoylase
(ATCase), Na+/K+-adenosine triphosphatase (Na+/
K+-ATPase), and human transient receptor
potential melastatin 2 (TRPM2) and are
summarized below for the rest of the allosteric
proteins in the dataset.
Aspartate transcarbamoylase (ATCase). ATCase
catalyzes the synthesis of N-carbamoyl-L-aspartate
with inorganic phosphate from carbamoyl
phosphate and L-aspartate, an initial step in
pyrimidine biosynthesis in many prokaryotes, such
as Escherichia coli.45 It is a widely studied system;
a textbook example of cooperativity and allosteric
3
Journal of Molecular Biology 434 (2022) 167644
regulation.46–48 Further, due to the homology
shared with human ATCase, it is also of interest
as an anticancer target for allosteric drug design
studies.49 ATCase functions as a heterododecamer
of six catalytic subunits organized as two trimers
with three active sites at their interfaces50 and six
regulatory chains assembled as three dimers.51
While carbamoyl phosphate or L-aspartate occu-
pies the active site, nucleotides (i.e., ATP, CTP
and UTP) exert their allosteric effect through bind-
ing on the regulatory subunits. It is tightly regulated
to stabilize purine/pyrimidine ratio in cell. This is
maintained through the interplay of tense (T) and
relaxed (R) states, by which ATP induces activity
via stabilizing the R state and CTP and UTP inhibit
function by prioritizing the T state. While ATP and
CTP compete for the same binding site, UTP binds
to a different yet close position.
A popular inhibitor of ATCase, N-
Phosphonacetyl-L-Aspartate (PALA), acts through
binding to the active site on the catalytic subunits
and stabilizing the R-state.52 The PALA’s binding
affinity is highly affected by the presence of allos-
teric modulators ATP and CTP.53 Here, the PALA-
bound R-state structure (PDB: 1D0945) is used in
the TE calculations. The collectivity distributions of
the TE values of residues in five subsets of slow
modes display that the 1–10 and 4–10 slow modes
maximize the information transfer of residues (Fig-
ure S2). As shown in Figure 2, in these respective
subsets of slow modes, the residues with high
TECol score values coincide with the active PALA
(green) and allosteric ligand ATP/CTP (cyan) bind-
ing residues (Table S2). In other words, the PALA
binding residues appear as the most global informa-
tion source residues in terms of information transfer
when all ten slow modes are considered. However,
with the exclusion of the slowest to third slowest
modes, the allosteric ligand binding residues gradu-
ally gain the full dynamic capacity for the collective
information transfer in the 4–10 slow modes.
Here, the first three slow modes underlie the
prepotent role of active sites while the 4–10 slow
modes indicate potential stimulability upon a
perturbation (ligand binding, post-translational
modifications, etc.), causing certain slow modes to
efface, appear, and/or shuffle. Overall, the active
site residues might thus be expected to cause a
stronger adjustment in the dynamics of the
structure in this conformational state, which is
seen here with the stabilization of the R-state
through the PALA binding. Another global
information source with high collectivity yet smaller
TECol score values than active site residues in
the 1–10 slow modes appear with the C-terminal
end of Helix 12 (A298-N305) in the catalytic
subunit. Earlier experiments showed that
mutations at this part of the helix affect both the
stability of the trimer and their interactions with the
regulatory subunits through indirect global
adjustments.54 Thus, information source residues
B. Altintel, B. Acar, B. Erman, et al. Journal of Molecular Biology 434 (2022) 167644

Figure 2. ATCase (PDB ID: 1D09). Net Transfer Entropy (TE) maps (a, c) and the corresponding collectivity and
TECol score plots (b, d) in the subsets of 1–10 and 4–10 slow modes. The active PALA (green) and the allosteric CTP
(cyan) binding residues are information sources with collective information transfer in the 1–10 and 4–10 slow modes,
respectively (b, d). All colored ribbon representations are from the highest (red) to lowest (blue) TECol score values.
CTP is taken from the CTP-bound T state structure (PDB ID: 1RAC73).

highlight not only ligand/substrate binding regions
but also other sites with potential functional roles
such as the stabilization of quaternary structure.
Na+/K+-adenosine triphosphatase (Na+/K+-
ATPase). Na /K -ATPase transports three Na+
+ +
ions in exchange with two K+ ions through the
cellular membrane by the ATP hydrolysis.55 This
requires a dynamic allosteric communication
between the intracellular ATP and transmembrane
Na+ and K+ binding sites along with the transloca-
tion tunnel. Also regulating the energy consuming
function against the ion gradient, it is an essential
protein for a variety of physiological processes to
maintain gradients of these ions across the cellular
membrane.
Na+/K+-ATPase is formed of three subunits, the
major catalytic a-subunit and two smaller subunits,
b and c. It shuttles between two major
conformations E1 and E2 to perform its function.
E1 binds ATP and three Na+ ions at the
cytoplasmic region and releases these ions upon
phosphorylation to the exterior and assumes to
E2. E2 binds two K+ ions at the extracellular
region followed by their dephosphorylation. The
ATP hydrolysis is required for the E2 to E1
conformational change to release trapped K+ ions
into the cell.56 During this transition, the b-subunit
undergoes significant conformational changes in
comparison to the a and c subunits.57
4
The crystal structure of Na+/K+-ATPase in the E1
state with bound ADP and Na+ (PDB: 4HQJ57) is
used in the TE calculations. The collectivity distribu-
tions of the TE values of residues in five subsets of
slow modes show that the 1–10 and 2–10 slow
modes maximize the information transfer (Fig-
ure S2). As shown in Figure 3, in these respective
subsets of slow modes, the residues with high
TECol score values appear at the ADP along with
tetrafluoroaluminate ion (ALF) and Mg2+ (green)
and allosteric Na+ (cyan) binding sites (Table S2).
With similar collectivities of the TE values, the 3–
10 and 5–10 slow modes also point to the allosteric
and active sites, respectively. The latter subsets of
slow modes behavior are respectively hidden in
the 2–10 and 1–10 slow modes.
Interestingly, highly collective residues in the TE
values also appear in the b-subunit, rather subtly
in the 1–10 slow modes and more significantly in
the 2–10 slow modes (Figure 3). This confirms the
expected regulatory behavior of the b-subunit as
stabilizer of the Na+-occluded E1-P state and
modulator of ion binding on the a-subunit.58 Specif-
ically, b-subunit harbors asparagine residues that
post-translationally bind oligosaccharides, related
with cell-specific functionality.59 In the 2–10 slow
modes, two of these oligosaccharide sites N158
and N193 appear as information sources (Figure 3
(d)). Besides, the cysteine residue pairs C159-
B. Altintel, B. Acar, B. Erman, et al.
Figure 3. Na+, K+-ATPase (PDB ID: 4HQJ). Net Transfer Entropy (TE) maps (a, c) and the corresponding
collectivity and TECol score plots (b, d) in the subsets of 1–10 and 2–10 slow modes. The active ADP-ALF-Mg2+
(green) and the allosteric Na+ (cyan) binding residues are information sources with collective information transfer in
the slow modes of 1–10 and 2–10, respectively (a, b). The active site reappears as an information source in the 5–10
slow modes. See also caption of Figure 2.
C175 and C213-C276 forming disulphide bridges
are information sources in the 2–10 slow modes,
while the latter is also apparent in the 1–10 slow
modes. These disulphide bonds along with
oligosaccharide binding sites are particularly impor-
tant in the assembly of the a/b subunits .60 They are
all located distant from the interface, yet possibly
providing an allosteric stabilizing effect. Nonethe-
less, in the 5–10 slow modes, the effect of the b-
subunit gives way to the unattended impact of the
ADP-ALF-Mg2+ binding sites as global information
sources. This shows the role of four slowest modes
on the allosteric signaling by the b-subunit.
Another interesting information source residue
with highest collectivity in the information transfer
is detected in the 2–10 slow modes is around
E818 (Figure 3(d)). A mutation on this site of the
human form was shown to disrupt the E1 > E2
conformational change and proton transport,
leading to a severe alternating hemiplegia of
childhood (AHC) disease.61 That the TE calcula-
tions provide a dynamic rationale may contribute
to decipher the unknown allosteric mechanism
involved in this mutation.
Human transient receptor potential melastatin 2
(TRPM2). TRPM2 is a nonselective Ca2+-
permeable cation channel on the cellular
membrane activated by heat, redox signals and/or
chemical binding of ADP-ribose (ADPR) and Ca2+.
It is a homotetramer formed of three tiers on top of
each other. ADPR and Ca2+ are respectively
bound to the bottom pier and top pier, while a
middle tier is located in between. To permeate
5
Journal of Molecular Biology 434 (2022) 167644
Ca2+, TRPM2 assumes a large conformational
change from closed to open conformation, whose
energy can be compensated only by the
synergistic effect of the allosteric communication
between ADPR and Ca2+.62 TRPM2 is essential
for various cells to deal with oxidative stress and
to regulate temperature.62 Yet, the complete chem-
ical mechanism of human TRPM2 activation is
unknown.
The apo structure of human TRPM2 in closed
state (PDB: 6MIX62) is used in the TE calculations.
The collectivity distributions of the residues’ TE val-
ues in five subsets of slow modes show that the
slow modes of 1–11 and 2–11 maximize the infor-
mation transfer (Figure S2). As shown in Figure 4,
in the respective subsets of slow modes, the infor-
mation sources with high TECol score values point
to important structural units (Table S2), the gating
S6 helix, the rotation of which is essential for the
channel opening and the ADPR binding sites (cyan)
whose priming influence on the Ca2+ binding con-
certedly rotate the gating S6 helix.62 Although both
identified sites are functional, the Ca2+ (green) bind-
ing sites are missing in these five subsets of slow
modes. Additionally, the slight distortion of the sym-
metry in the TE patterns in such a symmetric struc-
ture, if not due to a structural artifact, quest the
adequacy of the inclusion of the degenerate modes
with the same eigenvalues in the 1–11 slow modes.
Both require further investigation of the mode distri-
butions and combinations to possibly identify other
slow mode subsets with similar or higher collectivi-
B. Altintel, B. Acar, B. Erman, et al. Journal of Molecular Biology 434 (2022) 167644

Figure 4. Human TRPM2 Ion Channel (PDB ID: 6MIX). Net TE maps (a, c) and the corresponding collectivity and
TECol score plots (b, d) in the subsets of 1–11 and 2–11 slow modes. The channel gating S6 helix and the ADPR
(cyan) binding residues are information sources with collective information transfer in the 1–11 and 2–11 slow modes,
respectively (b, d). The Ca2+ (green) binding residues in the 1,4,8,9 slow modes, the pole helices of the channel in the
4,7,8 slow modes, and both the channel gating and the ADPR binding residues together in the 1,4,7,8 slow modes are
also information sources (e). The active and allosteric site ligands are taken from the Ca2+ and ADPR-bound structure
(PDB ID: 6PUU74). See also caption of Figure 2.

ties in the information transfer than the initial five
subsets of slow modes.
Indeed, we designated the 1,4,8,9 and 1,4,7,8
slow modes as the best slow mode combinations
maximizing the collective information transfer
(Figure S2), which disclose respectively the Ca2+
binding sites (green) and the ADPR binding sites
(cyan) together with the cytoplasmic gate of the
channel. Additionally, the channel pore region
extending from the extracellular to intracellular
side of the top tier appears strongly with the 4,7,8
slow modes. Residues in the outer pore are
critical for the inactivation of TRPM2 by Zn2+,
which is an endogenous allosteric modulator of
the protein.63 As seen, with the increasing number
of the slow mode subsets with high collectivity in
the information transfer, we may dispose an addi-
tional functional site or reobserve already identified
ones as fine-tuned and/or in different dynamic con-
text. For example, in the 1–11, 2–11, and 1,4,7,8
slow modes, while only the cytoplasmic gate and
only ADPR binding sites are mainly observed in
the first two subsets of slow modes, both functional
sites appear in the last slow mode combination
(Figure 4).
Directional allosteric pathways in ATCase, Na+/
+
K -ATPase, and TRPM2. As shown in Figure 5,
global information sources determine the patterns
of allosteric communications with a collective
influence on the others.
Based on the PALA bound active structure of
ATCase, residue A51 with highest collectivity in
6
the TE values at the active site of one catalytic
subunit collectively transfers information to the
other catalytic subunits in the 1–10 slow modes.
This implies cooperativity among catalytic
sites/subunits for the ligand binding, the PALA
binding, in this inhibited R-state. On the other
hand, residue V17 with high collectivity in the TE
values at the allosteric site collectively transfers
information to the active sites in the 4–10 slow
modes even in the absence of allosteric ligands.
This also provides a dynamic basis for the
inhibitory effect of ATP/CTP on the binding affinity
of PALA.53 Notably, this behavior is latent in the
1–10 slow modes in this functional state (the R-
state) of ATCase.
In Na+/K+-ATPase with bound ADP and Na+,
N476 is one of the residues with high collectivity in
its TE values of the 1–10 slow modes at the ADP
binding site and directs allosteric communications
pathways towards the allosteric region close to the
NA binding sites and the b-subunit. On the other
hand, D808 with high collectivity in the TE values
of the 2–10 slow modes at the NA binding site
collectively transfers information towards the ADP
binding sites and to the b-subunit. That the ADP
and NA binding sites exchange directed
information shows bidirectionality in their allosteric
communications. The altered directions are
revealed in two different subsets of slow modes,
indicating the importance of mode dissection for
layering of allosteric communication pathways.
B. Altintel, B. Acar, B. Erman, et al. Journal of Molecular Biology 434 (2022) 167644

Figure 5. Directional allosteric pathways. The information flow from a selected global information source residue
to the others in the structure, color-coded from the highest (red) to the lowest (blue) Net TE values with yellow dash
arrows displaying its direction (see Figures 2-4). ATCase. From the active ligand PALA binding residue A51 in the 1–
10 slow modes and from the allosteric ligand CTP binding residue V17 in the 4–10 slow modes to the other residues,
mainly to the catalytic sites of other subunits and to the active sites of all subunits, respectively (a). Na+/K+-
adenosine triphosphatase (Na+/K+-ATPase). From the active binding residue N476 in the 1–10 slow modes and
from the allosteric ADP-ALF-Mg binding residue D808 in the 2–10 slow modes to the other residues, mainly to the
allosteric and to the active sites, respectively (b). Human transient receptor potential melastatin 2 (TRPM2). From
the Ca2+ ligand binding residue E842 in the 1,4,8,9 slow modes and from the ADPR binding residue Y295 in the
1,4,7,8 slow modes to the other residues, mainly to the ADPR and to the Ca2+ binding residues, respectively (c).

With the apo structure of human TRPM2 in closed
state, E842 is one of the residues with high
collectivity in the TE values of the 1,4,8,9 slow
modes at the Ca2+ binding region and directs
allosteric communications pathways towards the
ADPR binding sites. On the other hand, Y295 with
high collectivity in the TE values of the 1,4,7,8
slow modes at the ADPR binding sites collectively
transfer information mainly towards the Ca2+
binding sites. That the ADPR binding sites at the
bottom tier and Ca2+ binding sites at the top tier
exchange directed information shows
bidirectionality in the allosteric crosstalk in the
close to open transition of the channel to
permeate the Ca2+ ions.
These three exemplary cases (Figures 2-5) show
that the simple search scheme is already powerful
to reveal the subsets of slow modes maximizing
the collective information transfer and active/
allosteric sites global information sources. Yet, as
in the case of human TRPM2, the combinatorial
search scheme may help identify new subsets of
slow modes with functional importance and fine-
tune the already notable subsets of slow modes.
7
Overview from allosteric proteins in the
dataset
In addition to the three exemplary cases, we
evaluate our predictions on 20 more allosteric
proteins. In the GNM based TE calculations, the
active state structures are mostly used without
explicitly including active and allosteric ligands or
their binding sites information. As presented in
Figures S4-24, the active and allosteric sites in
these allosteric proteins are revealed as global
information sources.
In the 18 cases, the simple search scheme of five
subsets of slow modes that sequentially exclude the
slowest mode disclose active and allosteric sites as
global information sources. For the remaining two
cases -phosphofructokinase (PDB: 4PFK) and
chemotaxis protein Che-Y (PDB: 1F4V)- these
subsets of slow modes partially detect functional
sites with relatively lower collectivities in their
information transfer (Figures S15 and S22).
However, the combinatorial search scheme of the
three-to-five slow mode combinations based on
the maximization of the collectivities of residues in
the transfer of information also give active and
B. Altintel, B. Acar, B. Erman, et al.
allosteric sites as information sources in these
structures.
For most of these cases, global information
sources overlap with active site residues in the
subsets of slow modes including the slowest
mode. However, when the effects of the slowest
mode and the next slower modes are sequentially
removed, latent allosteric communications in
which allosteric sites turn to the most pronounced
information sources appear. There are the
subsets of slow modes that we observe, both
active and allosteric sites are detected as globular
information sources. Appearance order of active
and allosteric sites in the slowest mode may alter
with the change in the conformational state,
prioritizing a certain functional behavior or an
information source over another.
The three-to-five slow mode combinations
besides the simple five subsets modes are also
evaluated for the 18 cases in the dataset. The
slowest mode in a subset of slow modes mainly
drives the general information transfer behavior.
However, it is also possible to see more than one
dynamic behavior in the subsets of different
combinations starting with the same slowest
mode. The averaging -the superimposition- of the
information transfer in multiple modes of a crude
dissection in such cases may preclude a clear
view of information transfer patterns with
eliminated “mode pollution” as well as lead to the
identification of some unique functional sites that
could not be observed otherwise.
The results of the statistical analysis for the
predictions of functional -active and allosteric-
sites are given in Table S2. 37 out of 46
predictions are statistically significant according to
their p-value analyses, corresponding to the
overall success of 80.4% of our methodology. The
predictions with p-values exceeding the
significance level of 0.05 belong to the cases of
small structures where the predictions may
include both active and allosteric sites as they are
in proximity on the structure. On the other hand,
for the threshold distance of 7
A, the values of
four other performance metrics reached as high
as 100%, 92%, 64%, and 85% respectively for
sensitivity, specificity, precision, and accuracy.
Overall, sensitivity and precision values are
relatively lower than specificity and accuracy
values. Relatively lower precision values may be
due to the fact that the predictions in a given
subset of slow modes may reveal both active and
allosteric sites while here we analyze on the
assumption of one functional group per each
statistical set. For example, in ADP-glucose
pyrophosphorylase (PDB: 1YP3), using both
groups of allosteric and active sites improved the
precision values. There could also be additional
unknown binding sites and/or other plausible
functional sites such as the ones important for
subunit folding and assembly. Furthermore,
8
Journal of Molecular Biology 434 (2022) 167644
neighboring information source residues of a
functional site are not counted when the threshold
distance is relatively small (<7
A). Thus, precision
increases up to 100% when a threshold distance
of 10
A is considered. Besides, relatively lower
precision values are partially balanced by the
higher specificity values for most of the cases,
meaning nonfunctional residues are defined
correctly. On the other hand, lower sensitivity
values reflect that some of the residues that make
up a functional site do not appear as global
information sources. This is expected that not all
ligand or substrate binding residues need to have
a dynamic capacity for collective information
transfer, some may be important for additional
roles such as the stabilization of the ligand–protein
interactions.
Nondegenerate and degenerate slow modes with
structural symmetry. Degenerate slow modes could
be functionally relevant as well as nondegenerate
slow modes that are essential for symmetric
functional motions of the structure, as recently
shown for toroidal proteins with different
oligomerization states.64 Although not revealing
high collectivities, the functional significance of
degenerate slow modes is evident in allosteric pro-
teins with structural symmetries (Table S3). Both
degenerate modes as well as nondegenerate slow
modes carry information on active or allosteric sites
(Table S3).
In TRPM2 (PDB: 6MIX), the subset of only
degenerate slow modes mainly reveals the ADPR
binding regions (Figure S25). Apparently,
degenerate slow modes dominate the dynamics in
the 2–11 slow modes that give the ADPR binding
regions in this basic subset (Figure 4). On the
other hand, however, in some cases, it is not
possible to identify either active or allosteric sites
with the subsets of only nondegenerate or only
degenerate slow modes. That is for which their
combinations can reveal functional sites with high
collectivities such as in the case of aspartate
transcarbamoylase (PDB: 1D09). For this case,
two nondegenerate slow modes (1 and 4) together
reveal allosteric sites (Figure S25(a)), while the
slowest mode discloses active sites with
contributions of some degenerate slow modes
(slow modes 2, 3, and 5–10) (Figures 2(b)).
Further, not all nondegenerate or degenerate
slow modes are of functional importance for a
particular functional motion. Thus, the subsets of
specific slow modes from nondegenerate or/and
degenerate slow modes may yield higher
collectivities in the information transfer and render
functional sites more distinguishable as global
information sources. The presence of the
nondegenerate slowest mode is essential in the
appearance of active sites in combinations with
different slow modes, while its elimination
discloses the allosteric sites in multiple subsets of
slow modes being a major determinant of
B. Altintel, B. Acar, B. Erman, et al.
allosteric behavior (Figures S7-10), such as in
fructose-1,6-bisphosphatase (PDB: 1EYI)
(Figure S7) and human caspase-1 (PDB: 2HBQ)
(Figure S10). For TRPM2 (PDB: 6MIX), this
observation becomes more complex, since
different subsets of slow modes yield distinct and
different functional sites such as channel entrance
along with ligand binding sites (Figures 4 and 5(c),
S25(c)-(d)).
As a general observation, only nondegenerate
slow modes mostly appoint high collectivities to
both active and allosteric sites, while only
degenerate slow modes are more likely to hint at
either active or allosteric sites, yielding
compartmentalization in the allosteric
communication. Thus, these degenerate slow
modes might be beneficial in dissecting the
dynamics into layers of allosteric communication
exclusive for entropy sources of either allosteric or
active sites. Upon coupling with other degenerate/
nondegenerate slow modes, this would help to
complete the allosteric circuit. Furthermore,
degenerate slow modes are likely those that
respond to internal/external perturbation through
splitting and their coupling to other slow modes
are plausible, which may lead to some functional
responses.
In combination with nondegenerate slow modes
in the present analysis, we pay attention not to set
apart the degenerate modes of a given
eigenvalue. On the other hand, it is still possible to
observe the subsets of slow modes in split
degeneracies with relatively maximum information
transfer of residues. However, we avoid using
them for functional site prediction.
Information flow in subunit cooperativity. In
multimeric proteins of identical subunits, the
dynamic characteristics of functional -active and
allosteric- sites as global information sources in
different subsets of slow modes could be manifold
as described below.
In tetrameric fructose-1,6 bisphosphatase (PDB:
1EYI) (Figure S7), the 1–10 slow modes result in
active sites as the only information sources to
integrate all subunits where the communication of
diagonal monomer pairs are more emphasized.
On the other hand, the 2–10 slow modes enable
the allosteric sites of each monomer to transfer
information to its neighboring monomer. In L-
lactate dehydrogenase (PDB: 1LTH) (Figure S9),
the 1–10 slow modes define information transfer
within the dimers as a general pattern. The active
site residues behave as exceptions to this general
behavior and communicate with the other dimer at
various positions. However, the 2–10 slow modes
integrate all subunits with the information transfer
from both active and allosteric sites. Similar
behavior is observed in glycogen phosphorylase
(PDB: 7GPB) (Figure S18) where the 1–10 slow
modes enable information transfer from both
active (relatively more collective) and allosteric
9
Journal of Molecular Biology 434 (2022) 167644
sites within the dimers. On the other hand, the 3–
10 slow modes having active sites with
information transfer capacity integrate all subunits.
The degeneracy shared by slow modes 1 and 2 is
apparently at play for the allosteric communication
on the level of dimers. ADP-glucose
phosphorylase (PDB: 1YP3) (Figure S14) in the
1–9 slow modes shows that active and allosteric
sites (relatively more collective) as global
information sources integrate all subunits yet
being more effective within dimers. Active and
allosteric sites are also information sources with
collective information transfer to all subunits in the
5–9 slow modes. A more repetitive communication
pattern is apparent in glutamate dehydrogenase
(PDB: 6DHD) (Figure S12), a hexameric protein.
Here, allosteric sites maintain the information flow
within each monomer in the 1–10 slow modes,
whereas active sites display information flow
within each dimer in the 4–10 slow modes. GlcN-
6-P deaminase (PDB: 1HOT) (Figure S13)
displays similar behavior to glutamate
dehydrogenase. In the 1–10 slow modes, the
information transfer is mainly within monomers in
general, yet allosteric sites act as information
sources and integrate all subunits. In the 4–10
slow modes, active site residues transfer
information within dimers.
The appearance of active and allosteric sites
alone or together and their control at different
subunit levels revealed in various subsets of slow
modes are based on the functional states
represented by the structures on which the
calculations are based. With conformational
changes or in another functional state, it would be
possible to see the slow modes/these subsets of
slow modes to efface, appear, and/or shuffle,
prioritizing certain functional motions according to
the specific functional state.
Additional new functional sites. It is also possible
to observe additional information source residues of
likely functional importance, such as in PDK1 (PDB:
3ORZ) and CHK1 (PDB: 2BRG) kinases. In these
two kinases, while the active and allosteric sites
appear in the 1–10 slow modes, the information
source residues also point to some other regions
of the structure in the 2–10 slow modes that we
see them overlapping with the drug bindings sites
from Kinase Atlas65 (Figures S20 and S21). Along,
in another kinase, human DNA-PK Holoenzyme
(PDB: 5Y3R), the allosteric DNA binding region
together with the catalytic sites in the kinase domain
appear in the 1–10 slow modes (Figure S23(b)).
However, other regions appear as additional allos-
teric sites in the 2–10 slow modes that we observe
one of these coinciding with the breast cancer 1
(BRCA1) binding sites66 (Figure S23(d)), which is
known for its role in the autophosphorylation of the
DNA-PK catalytic unit.67 Indeed, the information
flow from a selected global information source resi-
due S2056 (one of the BRCA1 binding sites) is seen
B. Altintel, B. Acar, B. Erman, et al.
towards the catalytic cavity and the DNA binding
sites associated with KU70/80 protein (Figure S23
(e)). These examples indicate that information
source residues are likely allosteric druggable sites
to be targeted.
Concluding remarks. Functional sites have
collective information transfer ability revealed in
various subsets of slow modes with which they
can lead and control plausible multi- and bi-
directional allosteric communication pathways
(Figure 6). The overlap between functional sites
and highly collective information sources might be
a way of evolution to optimize the protein topology
in the most economical way of allosteric
communication.
On a single protein level, to explore what the
information transfer is locally and globally in more
detail, considering all available structures of
functional states would reveal an extensive picture
of allosteric regulation, i.e., a more complete
dynamic allosteric landscape. Overall, this may
contribute to the understanding of innate dynamic
control mechanisms and likely modular control of
the protein function and suggest molecular
strategies particularly for structure based allosteric
drug design efforts. The layering of the
multifaceted allosteric communication, enabled by
the proposed approach, might give opportunity to
specifically target certain layers of allosteric
interactions while keeping some others.
Particularly, the allosteric interactions related with
binding specificity might be a potential target to
develop drugs with less side effects.
Figure 6. Active and allosteric sites as information sources for a test set of 23 proteins. Each ribbon structure
is color-coded from the highest (red) to lowest (blue) TECol score values of residues in subsets of slow modes of
maximal information transfer. Active (green) and allosteric (cyan) ligand binding sites are represented with spheres.
See Figures S4-23 for net TE maps and the corresponding collectivity and TECol score plots.
10
Journal of Molecular Biology 434 (2022) 167644
The efficiency and the robustness of the
presented approach enable to search not only for
the prediction of functional sites but also for
functional mechanisms in ever-increasing number
of available protein structures.
Materials and methods
Dataset
A dataset of the 20 known allosteric proteins from
Amor et al.21 and three additional allosteric cases of
interest are utilized for validation of the approach.
The corresponding PDB codes with known active
and allosteric sites are shown in Table S1. The
Amor dataset spans a wide area of cytosolic protein
space with consideration of various domain organi-
zations, multimerization states, ligand types, and
varying protein size from 147 to 3311.21 Incorpora-
tion of two membrane proteins, Na+/K+-ATPase
transporter (PDB: 4HQJ57) and human TRPM2 ion
channel (PDB: 6MIX62) of 1297 and 5348 amino
acids respectively, along with a large (4670 resi-
dues) repeat protein, human DNA-PK holoenzyme
(PDB: 5Y3R68), to this dataset adds more variety.
The Gaussian network Model (GNM) based
transfer entropy
GNM69–70 predicts residue fluctuations and their
correlations assuming a Gaussian probability distri-
bution for instantaneous residue fluctuations (DR).
It models the protein structure as a network of a-
carbon atoms of residues (Figure 1) interacting with
B. Altintel, B. Acar, B. Erman, et al.
a harmonic potential function within a threshold
radius (Rcut = 10

A). Using the c force constant
and DR fluctuation vectors, potential energy V is
defined as
V ¼ ðc=2ÞDR T CDR
in which C is the Kirchhoff residue connectivity matrix.
With the pseudoinverse of C revealing N-1 intrinsic
modes of motion, the correlation between fluctuations
of residues i and j is given as


< DR i :DR j >¼ ð3k B T =cÞ C1 ij


¼ ð3k B T =cÞ UK1 U T ij
X transfer entropy
¼ ð3k B T =cÞ ½kk 1 U k U k T ij
k The residues with high net information transfer to
where K and U are respectively the eigenvalue and
eigenvector matrices of C with kk and Uk representing
the kth mode component, kB is the Boltzmann constant,
and T is the absolute temperature in degrees Kelvin.
Modes are ranked in ascending order of eigenvalues
with k = 1 is the slowest mode and k = N-1 is the
fastest mode. Following, the correlation of zero-time
fluctuations of i with future s time fluctuations of j can
be expressed as70


< DR i ð0Þ:DR j ðsÞ >¼ ð3k B T =cÞ C1 ij e kk s=s0
X residues with s, to determine the most collective
¼ ð3k B T =cÞ k
½kk 1 U k U k T ij e kk s=s0
Here s0 is a characteristic time of the vibrational
dynamics of all folded proteins.70
GNM based time-delayed correlations in
conditional Shannon entropies, defining the
amount of transfer entropy, i.e., information
transfer, Ti?j(s) from residue i to j in time delay s,
using the expression by Schreiber27 as
T i!j ðsÞ ¼ SðDR j ðt þ sÞjDR j ðsÞÞ  SðDR j ðt þ sÞjDR i ðsÞ; DR j ðsÞÞ
The conditional Shannon entropies are calculated
as described in34 under the equations of 2 and 3.
The net information transfer from residue i to j at
time s can be written as
DT i!j ðsÞ ¼ T i!j ðsÞ  T j!i ðsÞ
where DTi?j(s) estimates the direction of information flow
between residues i and j in a certain time delay s, yielding
the degree to which the present movement of residue i
decreases the amount of uncertainty for the future
movement of residue j. If, DTi?j(s) > 0 then the
dynamics of residue i affects the dynamics of residue j,
meaning a causal directional relationship between
fluctuations of residues i and j.
Additionally, the information transfer capability of
a residue can be defined as how much it transfers
information to the rest of the protein, as defined by
cumulative TE formulation as
X
N
DT i!rest ðsÞ ¼ T i!j ðsÞ
j¼1
Journal of Molecular Biology 434 (2022) 167644
Maximum and minimum cumulative TE
(Ti?rest(s)) values are regarded as entropy/
information sources and sinks, respectively.
Entropy sources send and entropy sinks receive
information to/from the rest of the protein. Two
ð1Þ
parameters are effective in TE values: time delay
s and slow mode component k. Here instead of all
modes, we develop a measure to identify distinct
sets of slow modes to disclose plausible layers of
allosteric communication pathways.
ð2Þ The degree of collectivity in the GNM-based
many other residues are likely powerful effectors
meaning that these residues affect most of the
others, as dynamically key sites that lead and
control rest of the protein. The motivation is thus
to dissect dynamic information by the
decomposition of the internal dynamics to disclose
the subsets of slow modes maximizing the
collectivities of residues in their information
transfer: the net TE (DTi?j(s)) values.
The degree of collectivity (Col) values can be
calculated using the positive net TE values of
ð3Þ information source residues (effectors) benefiting
from Bruschweiler’s study42 as
!
1 XN
ji;s ¼ exp  aðDT ij;k ðsÞÞ2 logðaðDT ij;k ðsÞÞ2 Þ ð7Þ
N j¼1
where i,s is the Col value of residue i in the
information transfer through the subset of slow
modes (s), N is the residue number, and DTij,k(s)
is the positive net transfer entropy from residue i
ð4Þ
to j in slow mode k for time delay s. In Eq. (8), a is
a normalization factor that is determined as
X
N
aðDT ij;k ðsÞÞ2 ¼ 1 ð8Þ
ð5Þ i¼1
One-dimensional plots comprised of the Col
values of residues in the information transfer yield
powerful effectors as global information source
residues, which affect collectively the others. The
multiplication of the Col values with cumulative
positive net TE values of residues reveals
powerful effectors collectively affecting the others
with higher entropy signals, which we call as
TECol score.
The time delay s is selected based on the
maximization of the degree of collectivities of net
TE values, which is 3  sopt. sopt is the time
window in which total TE of residues is
maximized. See Figure S3, illustrating the effect of
ð6Þ
different s values on the collectivities in the TE
values of residues for three main exemplary cases.
11
B. Altintel, B. Acar, B. Erman, et al.
Identification of subsets of slow dynamic
modes maximizing information transfer
To quantify the collectivity of residues in
information transfer and develop a general
protocol (Figure S1), we utilize a dynamic
measure defined by adapting Bruschweiler’s
collectivity definition42 given in Eq. (7). After the
mode decomposition, we have two -simple and
combinatorial- search schemes for the subsets of
slow modes of likely functional importance; the five
basic subsets of slow modes (i.e., n-Ns, with n being
the slow modes of 1 to 5) and the subsets of slow
mode combinations (i.e., three-to-five slow mode
combinations), where Ns represents the number of
slow modes in each case having the major contribu-
tion to the overall dynamics (Figure S1). Further-
more, Ns ranging from 8 to 11 slow modes over 23
structures, takes the degeneracy for the cases with
fold symmetries into account (Table S3). Around
ten slow modes on average are considered to rep-
resent the slow end of the dynamic spectrum.71–72
In the flow of this general protocol, we start with
the simple search scheme emphasizing the
contribution of slowest/slower modes in these
subsets as simply selected from the Ns slow
modes. If this search scheme does not yield
residues at least two subsets of slow modes with
collectivities of residues higher than a threshold
value (i.e., 0.45) or if we would like to have a more
refined form of these subsets and/or additional
subsets of slow modes with high information
transfer collectivities of residues, we follow the
combinatorial search scheme. Then, the residues
with maximum TECol score values are revealed
from these subsets of slow modes by both search
schemes as global information source residues.
Using either the simple or combinatorial search
scheme yields more than one subset of slow
modes with which unique functional sites or the
same functional sites in a different dynamic
context could be disclosed. The proposed protocol
is applied to a dataset of 23 proteins, and we have
shown that active and allosteric sites are
distinguishable based on their capacity to
disseminate information to others collectively as
global information sources.
Statistical significance analysis
The statistical significance of each functional site
prediction is examined using random sampling.
First, the average 3D distance between the TECol
score peaks above the peaks’ average and the
known active/allosteric binding residues are
calculated. Secondly, the average 3D distances
are calculated for 10,000 randomly selected
sample sets having the same number of residues
as the number of those TECol score peaks. Then,
Z score is calculated using the following definition:
ðX  lÞ
Z ¼ ð9Þ
r NATO Science for Peace Program with grant
12
Journal of Molecular Biology 434 (2022) 167644
where X is the mean distance between the above-
average TECol score peaks and the binding residues.
l and r are the mean and the standard deviation of
the population of random sampling, respectively. After
calculating Z scores of each prediction, p-values are
obtained by two-tailed hypothesis test using a
significance level of 0.05. The performance of the
functional site predictions is also measured in terms of
sensitivity, specificity, precision and accuracy with the
following formulations:
Sensitivity : SN ¼ TP=ðTP þ FNÞ ð10Þ
Specificity : SP ¼ TN=ðFP þ TNÞ ð11Þ
Precision : PRE ¼ TP=ðTP þ FPÞ ð12Þ
Accuracy : ACC ¼ ðTP þ TNÞ=ðTP þ FP þ TN þ FNÞ ð13Þ
where the number of true positives, the number of false
positives, the number of true negatives, and the
number of false negatives, are represented by TP, FP,
TN, and FN, respectively. For each functional site
prediction, TP corresponds to the number of above-
average TECol score peaks -in that subset of slow
mode- whose 3D distances to the known active/
allosteric ligand binding residues are below or equal to
the threshold distance, while FP corresponds to the
number of the ones with 3D distances above the
threshold. On the other hand, FN and TN correspond
to the numbers of below-average TECol score peaks
with 3D distances less and higher than the threshold,
respectively. These numbers are calculated using two
different threshold distances, 7
A and 10
A. The
corresponding performance metrics and p-values of all
predictions are given in the Table S2.
CRediT authorship contribution statement
Bengi Altintel: Formal analysis, Investigation,
Methodology, Software, Visualization, Writing –
original draft. Burcin Acar: Investigation,
Methodology, Software, Writing – original draft.
Burak Erman: Supervision, Methodology, Writing
– review & editing. Turkan Haliloglu:
Supervision, Methodology, Funding acquisition,
Resources, Writing – original draft, Writing –
review & editing.
DATA AVAILABILITY
Data will be made available on request.
Acknowledgement
This study was financed by funds from Scientific
Research Project Coordination of Bogazici
University, The Scientific and Technological
Research Council of Turkey (TUBITAK) and
B. Altintel, B. Acar, B. Erman, et al.
numbers 20A05P3, 119F392 and G5685,
respectively.
Declaration of Competing Interest
The authors declare that they have no known
competing financial interests or personal
relationships that could have appeared to
influence the work reported in this paper.
Appendix A. Supplementary material
Supplementary data to this article can be found
online at https://doi.org/10.1016/j.jmb.2022.
167644.
Received 31 December 2021;
Accepted 17 May 2022;
Available online 26 May 2022
Keywords:
allostery;
information transfer;
causality;
collectivity;
global information sources
Abbreviations:
GNM, Gaussian Network Model; TE, Transfer Entropy;
TECol, Score of collective information transfer; Col,
Collectivity in information transfer
References
1. J. Zhang, R. Nussinov, eds., Protein Allostery in Drug
Discovery, 1163 (2019). https://doi.org/10.1007/978-981-
13-8719-7.
2. Verkhivker, G.M., Agajanian, S., Hu, G., Tao, P., (2020).
Allosteric Regulation at the Crossroads of New
Technologies: Multiscale Modeling, Networks, and
Machine Learning. Front. Mol. Biosci. 7, 136. https://doi.
org/10.3389/FMOLB.2020.00136/BIBTEX.
3. Mannes, M., Martin, C., Menet, C., Ballet, S., (2021).
Wandering beyond small molecules: peptides as allosteric
protein modulators. Trends Pharmacol. Sci.. https://doi.org/
10.1016/J.TIPS.2021.10.011.
4. Monod, J., Jacob, F., (1961). Teleonomic mechanisms in
cellular metabolism, growth, and differentiation. Cold
Spring Harb. Symp. Quant. Biol. 26, 389–401. https://doi.
org/10.1101/sqb.1961.026.01.048.
5. Cooper, A., Dryden, D.T.F., (1984). Allostery without
conformational change - A plausible model. Eur. Biophys.
J. 11, 103–109. https://doi.org/10.1007/BF00276625.
6. Gunasekaran, K., Ma, B., Nussinov, R., (2004). Is allostery
an intrinsic property of all dynamic proteins? Proteins. 57,
433–443. https://doi.org/10.1002/PROT.20232.
7. Brunori, M., (2011). Allostery turns 50: Is the vintage yet
attractive? Protein Sci. 20, 1097. https://doi.org/10.1002/
PRO.660.
13
Journal of Molecular Biology 434 (2022) 167644
8. Tzeng, S.R., Kalodimos, C.G., (2011). Protein dynamics
and allostery: an NMR view. Curr. Opin. Struct. Biol. 21,
62–67. https://doi.org/10.1016/J.SBI.2010.10.007.
9. Tsai, C.J., Nussinov, R., (2014). A Unified View of “How
Allostery Works”. PLoS Comput. Biol. 10 https://doi.org/
10.1371/JOURNAL.PCBI.1003394.
10. Haliloglu, T., Bahar, I., (2015). Adaptability of protein
structures to enable functional interactions and
evolutionary implications. Curr. Opin. Struct. Biol. 35, 17–
23. https://doi.org/10.1016/j.sbi.2015.07.007.
11. Guarnera, E., Berezovsky, I.N., (2016). Allosteric sites:
Remote control in regulation of protein activity. Curr. Opin.
Struct. Biol. 37, 1–8. https://doi.org/10.1016/j.
sbi.2015.10.004.
12. Guo, J., Zhou, H.X., (2016). Protein Allostery and
Conformational Dynamics. Chem. Rev. 116, 6503–6515.
https://doi.org/10.1021/ACS.CHEMREV.5B00590.
13. Guarnera, E., Berezovsky, I.N., (2019). Toward
Comprehensive Allosteric Control over Protein Activity.
Structure. 27, 866–878.e1. https://doi.org/10.1016/J.
STR.2019.01.014.
14. Wodak, S.J., Paci, E., Dokholyan, N.V., Berezovsky, I.N.,
Horovitz, A., Li, J., Hilser, V.J., Bahar, I., Karanicolas, J.,
Stock, G., Hamm, P., Stote, R.H., Eberhardt, J., Chebaro,
Y., Dejaegere, A., Cecchini, M., Changeux, J.P., Bolhuis,
P.G., Vreede, J., Faccioli, P., Orioli, S., Ravasio, R., Yan,
L., Brito, C., Wyart, M., Gkeka, P., Rivalta, I., Palermo, G.,
McCammon, J.A., Panecka-Hofman, J., Wade, R.C., Di
Pizio, A., Niv, M.Y., Nussinov, R., Tsai, C.J., Jang, H.,
Padhorny, D., Kozakov, D., McLeish, T., (2019). Allostery
in its many disguises: from theory to applications.
Structure. 27, 566–578. https://doi.org/10.1016/J.
STR.2019.01.003.
15. Hawkins, R.J., McLeish, T.C.B., (2004). Coarse-grained
model of entropic allostery. Phys. Rev. Lett. 93, https://doi.
org/10.1103/PHYSREVLETT.93.098104/FIGURES/4/
MEDIUM 098104.
16. Popovych, N., Sun, S., Ebright, R.H., Kalodimos, C.G.,
(2006). Dynamically driven protein allostery. Nat. Struct.
Mol. Biol. 13, 831–838. https://doi.org/10.1038/NSMB1132.
17. Zhang, Y., Doruker, P., Kaynak, B., Zhang, S., Krieger, J.,
Li, H., Bahar, I., (2020). Intrinsic dynamics is evolutionarily
optimized to enable allosteric behavior. Curr. Opin. Struct.
Biol. 62, 14–21. https://doi.org/10.1016/J.SBI.2019.11.002.
18. Kutlu, Y., Ben-Tal, N., Haliloglu, T., (2021). Global
Dynamics Renders Protein Sites with High Functional
Response. J. Phys. Chem. B. 125, 4734–4745. https://
doi.org/10.1021/ACS.JPCB.1C02511.
19. Kaya, C., Armutlulu, A., Ekesan, S., Haliloglu, T., (2013).
MCPath: Monte Carlo path generation approach to predict
likely allosteric pathways and functional residues. Nucleic
Acids Res. 41, W249. https://doi.org/10.1093/NAR/
GKT284.
20. Van Den Bedem, H., Bhabha, G., Yang, K., Wright, P.E.,
Fraser, J.S., (2013). Automated identification of functional
dynamic networks from X-ray crystallography. Nat.
Methods. 10, 896. https://doi.org/10.1038/NMETH.2592.
21. Amor, B.R.C., Schaub, M.T., Yaliraki, S.N., Barahona, M.,
(2016). Prediction of allosteric sites and mediating
interactions through bond-to-bond propensities. Nat.
Commun. 7 https://doi.org/10.1038/ncomms12477.
22. J. Wang, A. Jain, L.R. McDonald, C. Gambogi, A.L. Lee, N.
V. Dokholyan, Mapping allosteric communications within
B. Altintel, B. Acar, B. Erman, et al.
individual proteins, Nat. Commun. 2020 111. 11 (2020) 1–
13. https://doi.org/10.1038/s41467-020-17618-2.
23. Sheik Amamuddy, O., Afriyie Boateng, R., Barozi, V.,
Wavinya Nyamai, D., Tastan Bishop, Ö., (2021). Novel
dynamic residue network analysis approaches to study
allosteric modulation: SARS-CoV-2 Mpro and its
evolutionary mutations as a case study. Comput Struct
Biotechnol J 19, 6431–6455. https://doi.org/10.1016/j.
csbj.2021.11.016.
24. Barozi, V., Musyoka, T.M., Sheik Amamuddy, O., Tastan
Bishop, Ö., (2022). Deciphering Isoniazid Drug Resistance
Mechanisms on Dimeric Mycobacterium tuberculosis KatG
via Post-molecular Dynamics Analyses Including
Combined Dynamic Residue Network Metrics. ACS
Omega 7, 13313–13332. https://doi.org/10.1021/
acsomega.2c01036.
25. McClendon, C.L., Kornev, A.P., Gilson, M.K., Taylora, S.
S., (2014). Dynamic architecture of a protein kinase. Proc.
Natl. Acad. Sci. U. S. A. 111, E4623–E4631. https://doi.org/
10.1073/PNAS.1418402111/SUPPL_FILE/
PNAS.1418402111.SM02.MOV.
26. Kornev, A.P., Taylor, S.S., (2015). Dynamics-Driven
Allostery in Protein Kinases. Trends Biochem. Sci. 40,
628–647. https://doi.org/10.1016/J.TIBS.2015.09.002.
27. Schreiber, T., (2000). Measuring information transfer.
Phys. Rev. Lett. 85, 461–464. https://doi.org/10.1103/
PhysRevLett.85.461.
28. Kamberaj, H., Van Der Vaart, A., (2009). Extracting the
causality of correlated motions from molecular dynamics
simulations. Biophys. J. 97, 1747–1755. https://doi.org/
10.1016/j.bpj.2009.07.019.
29. Barr, D., Oashi, T., Burkhard, K., Lucius, S., Samadani, R.,
Zhang, J., Shapiro, P., MacKerell, A.D., Van Der Vaart, A.,
(2011). Importance of domain closure for the autoactivation
of ERK2. Biochemistry 50, 8038–8048. https://doi.org/
10.1021/bi200503a.
30. Corrada, D., Morra, G., Colombo, G., (2013). Investigating
allostery in molecular recognition: insights from a
computational study of multiple antibody-antigen
complexes. J. Phys. Chem. B. 117, 535–552. https://doi.
org/10.1021/jp310753z.
31. Zhang, L., Centa, T., Buck, M., (2014). Structure and
dynamics analysis on plexin-B1 rho GTPase binding
domain as a monomer and dimer. J. Phys. Chem. B.
118, 7302–7311. https://doi.org/10.1021/jp503668k.
32. Vatansever, S., Gümüsß, Z.H., Erman, B., (2016). Intrinsic
K-Ras dynamics: A novel molecular dynamics data
analysis method shows causality between residue pair
motions. Sci. Rep. 6 https://doi.org/10.1038/srep37012.
33. Hacisuleyman, A., Erman, B., (2017). Entropy Transfer
between Residue Pairs and Allostery in Proteins:
Quantifying Allosteric Communication in Ubiquitin. PLOS
Comput. Biol. 13, https://doi.org/10.1371/JOURNAL.
PCBI.1005319 e1005319.
34. Hacisuleyman, A., Erman, B., (2017). Causality, transfer
entropy, and allosteric communication landscapes in
proteins with harmonic interactions. Proteins. 85, 1056–
1064. https://doi.org/10.1002/prot.25272.
35. Tama, F., Sanejouand, Y.H., (2001). Conformational change of
proteins arising from normal mode calculations. Protein Eng. 14,
1–6. https://doi.org/10.1093/PROTEIN/14.1.1.
36. Ikeguchi, M., Ueno, J., Sato, M., Kidera, A., (2005). Protein
structural change upon ligand binding: Linear response
theory. Phys. Rev. Lett. 94, https://doi.org/10.1103/
14
Journal of Molecular Biology 434 (2022) 167644
PHYSREVLETT.94.078102/FIGURES/3/MEDIUM
078102.
37. Acar, B., Rose, J., Aykac Fas, B., Ben-Tal, N., Lewinson,
O., Haliloglu, T., (2020). Distinct Allosteric Networks
Underlie Mechanistic Speciation of ABC Transporters.
Structure.. https://doi.org/10.1016/j.str.2020.03.014.
38. Ciftci, H., Tateishi, H., Koiwai, K., Koga, R., Anraku, K.,
Monde, K., Dağ, Ç., Destan, E., Yuksel, B., Ayan, E.,
Yildirim, G., Yigin, M., Ertem, F.B., Shafiei, A., Guven, O.,
Besler, S.O., Sierra, R.G., Yoon, C.H., Su, Z., Liang, M.,
Acar, B., Haliloglu, T., Otsuka, M., Yumoto, F., Fujita, M.,
Senda, T., DeMirci, H., (2021). Structural insight into host
plasma membrane association and assembly of HIV-1
matrix protein. Sci. Rep. 11 https://doi.org/10.1038/
S41598-021-95236-8.
39. Yang, M., Livnat Levanon, N., Acar, B., Aykac Fas, B.,
Masrati, G., Rose, J., Ben-Tal, N., Haliloglu, T., Zhao, Y.,
Lewinson, O., (2018). Single-molecule probing of the
conformational homogeneity of the ABC transporter
BtuCD. Nat. Chem. Biol. 14, 715–722. https://doi.org/
10.1038/s41589-018-0088-2.
40. Dror, R.O., Green, H.F., Valant, C., Borhani, D.W.,
Valcourt, J.R., Pan, A.C., Arlow, D.H., Canals, M., Lane,
J.R., Rahmani, R., Baell, J.B., Sexton, P.M., Christopoulos,
A., Shaw, D.E., (2013). Structural basis for modulation of a
G-protein-coupled receptor by allosteric drugs 2013
5037475. 503 Nat., 295–299. https://doi.org/
10.1038/nature12595.
41. Guarnera, E., Berezovsky, I.N., (2016). Structure-based
statistical mechanical model accounts for the causality and
energetics of allosteric communication. PLOS Comput.
Biol. 12, https://doi.org/10.1371/JOURNAL.PCBI.1004678
e1004678.
42. Brüschweiler, R., (1995). Collective protein dynamics and
nuclear spin relaxation. J. Chem. Phys. 102, 3396–3403.
https://doi.org/10.1063/1.469213.
43. Schulze, J.O., Saladino, G., Busschots, K., Neimanis, S.,
Süß, E., Odadzic, D., Zeuzem, S., Hindie, V., Herbrand, A.
K., Lisa, M.N., Alzari, P.M., Gervasio, F.L., Biondi, R.M.,
(2016). Bidirectional allosteric communication between the
ATP-binding site and the regulatory PIF Pocket in PDK1
protein kinase. Cell Chem. Biol. 23, 1193–1205. https://doi.
org/10.1016/J.CHEMBIOL.2016.06.017.
44. Leroux, A.E., Biondi, R.M., (2020). Renaissance of
Allostery to Disrupt Protein Kinase Interactions. Trends
Biochem. Sci. 45, 27–41. https://doi.org/10.1016/J.
TIBS.2019.09.007.
45. Jin, L., Stec, B., Lipscomb, W., Kantrowitz, E., (1999).
Insights into the mechanisms of catalysis and heterotropic
regulation of Escherichia coli aspartate transcarbamoylase
based upon a structure of the enzyme complexed with the
bisubstrate analogue N-phosphonacetyl-L-aspartate at 2.1


A. Undefined.. https://doi.org/10.1002/(SICI)1097-0134
(19991201)37:4.
46. Wang, J., Stieglitz, K.A., Cardia, J.P., Kantrowitz, E.R.,
(2005). Structural basis for ordered substrate binding and
cooperativity in aspartate transcarbamoylase. Proc. Natl.
Acad. Sci. 102, 8881–8886. https://doi.org/10.1073/
PNAS.0503742102.
47. Cockrell, G.M., Zheng, Y., Guo, W., Peterson, A.W.,
Truong, J.K., Kantrowitz, E.R., (2013). New paradigm for
allosteric regulation of Escherichia coli aspartate
transcarbamoylase. Biochemistry 52, 8036–8047. https://
doi.org/10.1021/BI401205N.
B. Altintel, B. Acar, B. Erman, et al.
48. Hodges, M., Barahona, M., Yaliraki, S.N., (2018). Allostery
and cooperativity in multimeric proteins: bond-to-bond
propensities in ATCase 2018 81. 8 Sci. Reports, 1–14.
https://doi.org/10.1038/s41598-018-27992-z.
49. Lei, Z., Wang, B., Lu, Z., Wang, N., Tan, H., Zheng, J., Jia,
Z., (2020). New regulatory mechanism-based inhibitors of
aspartate transcarbamoylase for potential anticancer drug
development. FEBS J. 287, 3579–3599. https://doi.org/
10.1111/FEBS.15220.
50. Lipscomb, W.N., Kantrowitz, E.R., (2012). Structure and
mechanisms of Escherichia coli. Acc. Chem. Res. 45, 444–
453. https://doi.org/10.1021/AR200166P.
51. Ke, H., Honzatko, R.B., Lipscomb, W.N., (1984). Structure
of unligated aspartate carbamoyltransferase of Escherichia
coli at 2.6-

A resolution. Proc. Natl. Acad. Sci. U. S. A. 81,
4037–4040. https://doi.org/10.1073/PNAS.81.13.4037.
52. Moody, M.F., Vachette, P., Foote, A.M., (1979). Changes
in the X-ray solution scattering of aspartate
transcarbamylase following the allosteric transition. J.
Mol. Biol. 133, 517–532. https://doi.org/10.1016/0022-
2836(79)90405-4.
53. Newell, J.O., Markby, D.W., Schachmann, H.K., (1989).
Cooperative binding of the bisubstrate analog N-
(phosphonacetyl)-L-aspartate to aspartate
transcarbamoylase and the heterotropic effects of ATP
and CTP. J. Biol. Chem. 264, 2476–2481. https://doi.org/
10.1016/S0021-9258(19)81638-1.
54. Peterson, C.B., Schachman, H.K., (1992). Long range
effects of amino acid substitutions in the catalytic chain of
aspartate transcarbamoylase. Localized replacements in
the carboxyl-terminal alpha-helix cause marked alterations
in allosteric properties and intersubunit interactions. J. Biol.
Chem. 267, 2443–2450. https://doi.org/10.1016/S0021-
9258(18)45899-1.
55. Clausen, M.V., Hilbers, F., Poulsen, H., (2017). The
structure and function of the Na, K-ATPase isoforms in
health and disease. Front. Physiol. 8, 371. https://doi.org/
10.3389/FPHYS.2017.00371/BIBTEX.
56. Glynn, I.M., (1993). Annual review prize lecture. ‘All hands
to the sodium pump’. J. Physiol. 462, 1–30. https://doi.org/
10.1113/JPHYSIOL.1993.SP019540.
57. Nyblom, M., Poulsen, H., Gourdon, P., Reinhard, L.,
Andersson, M., Lindahl, E., Fedosova, N., Nissen, P.,
(2013). Crystal structure of Na+, K(+)-ATPase in the Na(+)-
bound state. Science 342, 123–127. https://doi.org/
10.1126/SCIENCE.1243352.
58. Hilbers, F., Kopec, W., Isaksen, T.J., Holm, T.H., Lykke-
Hartmann, K., Nissen, P., Khandelia, H., Poulsen, H.,
(2016). Tuning of the Na, K-ATPase by the beta subunit
2016 61. 6 Sci. Rep., 1–11. https://doi.org/10.1038/
srep20442.
59. Schmalzing, G., Gloor, S., (1994). Na+/K+-Pump Beta
Subunits: Structure and Functions. Cell. Physiol. Biochem.
4, 96–114. https://doi.org/10.1159/000154715.
60. Beggah, A.T., Jaunin, P., Geering, K., (1997). Role of
glycosylation and disulfide bond formation in the beta
subunit in the folding and functional expression of Na, K-
ATPase. J. Biol. Chem. 272, 10318–10326. https://doi.org/
10.1074/JBC.272.15.10318.
61. Li, M., Jazayeri, D., Corry, B., McSweeney, K.M., Heinzen,
E.L., Goldstein, D.B., Petrou, S., (2015). A functional
15
Journal of Molecular Biology 434 (2022) 167644
correlate of severity in alternating hemiplegia of childhood.
Neurobiol. Dis. 77, 88–93. https://doi.org/10.1016/J.
NBD.2015.02.002.
62. Wang, L., Fu, T.M., Zhou, Y., Xia, S., Greka, A., Wu, H.,
(2018). Structures and gating mechanism of human
TRPM2. Science (80-. ), 362. https://doi.org/10.1126/
SCIENCE.AAV4809.
63. Yang, W., Manna, P.T., Zou, J., Luo, J., Beech, D.J.,
Sivaprasadarao, A., Jiang, L.H., (2011). Zinc Inactivates
Melastatin Transient Receptor Potential 2 Channels via the
Outer Pore *. J. Biol. Chem. 286, 23789–23798. https://doi.
org/10.1074/JBC.M111.247478.
64. Li, H., Doruker, P., Hu, G., Bahar, I., (2020). Modulation of
Toroidal Proteins Dynamics in Favor of Functional
Mechanisms upon Ligand Binding. Biophys. J. 118,
1782–1794. https://doi.org/10.1016/J.BPJ.2020.01.046.
65. Yueh, C., Rettenmaier, J., Xia, B., Hall, D.R., Alekseenko,
A., Porter, K.A., Barkovich, K., Keseru, G., Whitty, A.,
Wells, J.A., Vajda, S., Kozakov, D., (2019). Kinase Atlas:
Druggability Analysis of Potential Allosteric Sites in
Kinases. J. Med. Chem. 62, 6512–6524. https://doi.org/
10.1021/ACS.JMEDCHEM.9B00089.
66. Sibanda, B.L., Chirgadze, D.Y., Ascher, D.B., Blundell, T.
L., (2017). DNA-PKcs structure suggests an allosteric
mechanism modulating DNA double-strand break repair.
Science 355, 520–524. https://doi.org/10.1126/SCIENCE.
AAK9654.
67. Davis, A.J., Chen, B.P.C., Chen, D.J., (2014). DNA-PK: a
dynamic enzyme in a versatile DSB repair pathway. DNA
Repair (Amst). 17, 21–29. https://doi.org/10.1016/J.
DNAREP.2014.02.020.
68. Yin, X., Liu, M., Tian, Y., Wang, J., Xu, Y., (2017). Cryo-EM
structure of human DNA-PK holoenzyme. Cell Res. 27,
1341–1350. https://doi.org/10.1038/CR.2017.110.
69. Bahar, I., Atilgan, A.R., Erman, B., (1997). Direct
evaluation of thermal fluctuations in proteins using a
single-parameter harmonic potential. Fold. Des. 2, 173–
181. https://doi.org/10.1016/S1359-0278(97)00024-2.
70. Haliloglu, T., Bahar, I., Erman, B., (1997). Gaussian
Dynamics of Folded Proteins. Phys. Rev. Lett. 79, 3090–
3093. https://doi.org/10.1103/PhysRevLett.79.3090.
71. Kaynak, B.T., Doruker, P., (2019). Protein-Ligand
Complexes as Constrained Dynamical Systems. J. Chem.
Inf. Model. 59, 2352–2358. https://doi.org/10.1021/ACS.
JCIM.8B00946/ASSET/IMAGES/ACS.JCIM.8B00946.
SOCIAL.JPEG_V03.
72. Kaynak, B.T., Bahar, I., Doruker, P., (2020). Essential Site
Scanning Analysis: a new approach for detecting sites that
modulate the dispersion of protein global motions. Comput.
Struct. Biotechnol. J.. https://doi.org/10.1016/j.
csbj.2020.06.020.
73. Kosman, R.P., Gouaux, J.E., Lipscomb, W.N., (1993).
Crystal structure of CTP-ligated T state aspartate
transcarbamoylase at 2.5
A resolution: Implications for
ATCase mutants and the mechanism of negative
cooperativity. Proteins Struct. Funct. Bioinforma. 15, 147–
176. https://doi.org/10.1002/PROT.340150206.
74. Huang, Y., Roth, B., Lü, W., Du, J., (2019). Ligand
recognition and gating mechanism through three Ligand-
binding sites of human TRPM2 channel. Elife. 8 https://doi.
org/10.7554/ELIFE.50175.