Rationale for More Diverse Inhibitors in Competition with Substrates
in HIV-1 Protease
Nevra Ozer,†‡Celia A. Schiffer,§* and Turkan Haliloglu†‡*
†Polymer Research Center and‡Chemical Engineering Department, Bogazici University, Istanbul, Turkey; and§Department of Biochemistry
and Molecular Pharmacology, University of Massachusetts Medical School, Worcester, Massachusetts
using the anisotropic network model. The directions of fluctuations in the most cooperative functional modes differ mainly around
the dynamically key regions, i.e., the hinge axes, which appear to be more flexible in substrate complexes. The flexibility of HIV-1
protease is likely optimized for the substrates’ turnover, resulting in substrate complexes being dynamic. In contrast, in an inhib-
itor complex, the inhibitor should bind and lock down to inactivate the active site. Protease and ligands are not independent.
Substrates are also more flexible than inhibitors and have the potential to meet the dynamic distributions that are inherent in
the protease. This may suggest a rationale and guidelines for designing inhibitors that can better fit the ensemble of binding sites
that are dynamically accessible to the protease.
The structural fluctuations of HIV-1 protease in interaction with its substrates versus inhibitors were analyzed
One of the most important factors in elucidating the patho-
genesis of HIV-1 is viral resistance; thus, it is important to
understand the development of this drug resistance to
improve the therapeutic management of AIDS (1). The
homodimeric HIV-1 protease is an effective therapeutic tar-
get of the most effective antiviral drugs for the treatment of
HIV-1 infection. The protease sequentially cleaves at least
10 asymmetric and nonhomologous sequences in the Gag
and Gag-Pol polyproteins, and allows for maturation of
the immature virion that facilitates the spread of the virus
(2). These peptidomimetic drugs are the result of struc-
ture-based drug design efforts on the part of both academia
and the pharmaceutical industry. Indeed, protease inhibitors
are considered the most potent drugs currently available for
the treatment of AIDS (1).
Protease inhibitors are all competitive inhibitors that bind
at the active site and compete directly with the enzyme’s
ability to recognize substrates (1,3). They all have large,
generally hydrophobic moieties that interact with the mainly
hydrophobic pockets in the active site (1). Unfortunately,
the medical efficacy of the current inhibitors is proving to
be short-lived, as viable mutant variants of HIV-1 protease
confer drug resistance. Drug resistance results from a subtle
change in the balance of recognition events between the
relative affinity of the enzyme to bind inhibitors and its
ability to bind and cleave substrates. Since HIV-1 protease
binds substrates and inhibitors at the same active site, the
change that alters inhibitor binding also alters substrate
binding. However, the substrate recognition does not seem
to be greatly altered when inhibitors contact the residues
that are not contacted extensively by the substrates (4).
This may not be the case for residues that are important
for both substrate and inhibitor binding. Although they
are chemically different, the three-dimensional shape and
electrostatic character of the protease inhibitors are fairly
similar. A small set of mutations can thus result in a protease
variant with multidrug resistance. This evolution of drug
resistance in HIV-1 protease presents a new challenge to
future structure-based drug design efforts (1).
The HIV-1 protease functions as a homodimer with
a single active site (residues 25–27 of each chain) that is
formed by the dimer interface and capped by two flexible
flaps (5). Despite the symmetry conferred on its active
site by being a homodimer, the enzyme recognizes a series
of nonhomologous asymmetric octomeric substrate sites
within the Gag and GagPol polyproteins. Yet, despite the
fact that the substrate sites are asymmetric, the currently
prescribed inhibitors are relatively symmetric around the
cleavage site. This allows a single mutation to impact the
inhibitor binding twice, while possibly impacting substrate
binding to a lesser extent. Two solvent-accessible loops of
the protease (residues 33–43 of each chain) followed by
the two flexible flaps (residues 44–62 of each chain) are
important for ligand-binding interactions (6). The terminal
residues 1–4 and 95–99 of each chain play a role in dimer-
ization and stabilization of the active protease (6). A large
conformational change occurs during ligand binding, which
consists of the opening and closing of the flaps over the
Molecular recognition in ligand binding is dependent on
the intrinsic dynamics of the protein (7). Although structural
changes have been observed experimentally with ligand
binding, the intrinsic dynamics of the protein, which is
likely evolutionarily optimized, is not well described. An
induced fit in ligand recognition is favored by long-range
interactions, whereas conformational selection in binding
Submitted November 13, 2009, and accepted for publication June 28, 2010.
*Correspondence: email@example.com or firstname.lastname@example.org
Nevra Ozer’s present address is Department of Bioengineering, Marmara
University, Istanbul, Turkey.
Editor: Ruth Nussinov.
? 2010 by the Biophysical Society
1650 Biophysical JournalVolume 99 September 20101650–1659
is favored by short-range interactions (7). The diversity of
conformations and the insufficient data on the energetics
of protein-ligand interactions make it very difficult to incor-
porate the intrinsic dynamics into drug discovery efforts.
Fluctuations of biomolecular complexes around their
native states are important for functional analysis in molec-
ular biophysics. Several features, such as entropy changes
upon binding, possible drug-binding sites, and the overall
stability, flexibility, and function, can be deduced from
detailed analyses of these fluctuations (8,9). There is a
significant correlation between cooperative motions of the
structure and its biological function (7). There are several
computational methods that can be used to identify these
dominant correlated motions. The common approach is
to decompose the dynamics into a collection of modes of
motion focusing on a few low-frequency/large-amplitude
modes that are expected to be relevant to function (10,11).
The process of extracting the dominant collective modes
from fluctuations in molecular-dynamics (MD) trajectories,
also called principal component analysis (PCA), is now an
established computational method for studying protein
dynamics. The major disadvantage of this method is the
sampling inefficiency of MD simulations, especially in large
molecular systems (12,13). Alternatively, the cooperative
motions can be studied by normal mode analysis, in which
the concerted motions of a protein are expressed in terms
of a set of collective variables (normal modes) (14–20).
There are also some geometry-based methods, such as
CONCOORD/tCONCOORD (21,22), and methods based
on constraint theory, such as FIRST (23), that can be used
to tackle protein flexibility and sample the conformational
space to predict functional behaviors. The similarities
between these approaches and elastic network models
were previously presented in detail (22,24).
Recently, elastic network models have gained consider-
able attention for studying the large-scale motion of protein
structures that are relevant to function (25–31). This sug-
gests that the structures evolved in such a way that the
intrinsic elastic low-frequency modes are the most efficient
way for the structures to function. Elastic network models
originated from the work of Tirion (32), who developed
a model in which a single uniform harmonic potential
reproduces the complex vibrational properties of macromo-
lecular systems. The fluctuations predicted by the isotropic
Gaussian network model (GNM) (33,34) applied to coarse-
grained proteins with one point mass per residue agree
significantly with experimental crystallographic B-factors
for many proteins. The model has been extended to the
anisotropic network model (ANM) (35), in which the direc-
tions as well as the magnitudes of motions are predicted.
GNM and ANM applied to the HIV-1 protease system
have also produced results that are highly in accord with
those of both experimental and MD simulations, despite
their simplicity (6,10,36–40). Computational studies of the
dynamics of HIV-1 protease suggest that the structural fluc-
tuations in low-frequency modes can be utilized for intrinsic
protein flexibility and motion to maintain function.
In protein-ligand interactions, the ligand prefers the con-
formations that best match its structural and dynamic prop-
erties among those intrinsically accessible to the unbound
protein (7). Here, to search for the determinants of structural
changes that accompany ligand binding in HIV-1 protease,
we performed a comparative analysis of the conformational
space spanned by the protease bound to different ligands
and its intrinsic dynamics. We analyzed the crystal struc-
tures of both substrate- (1,3,4) and inhibitor-bound (41–49)
proteases by means of a simple physics-based ANM (35)
constructed by incorporating all of the atoms (except
hydrogen atoms) of the structure. We elaborated the magni-
tude and orientation of motion of protease and peptide
residues in the low-frequency modes, emphasizing spe-
cific regions (i.e., the dimerization, active site, flap, and
substrate cleft regions), by performing a comparative anal-
ysis between different natural substrate and inhibitor com-
plex structures. The results of this analysis may help
elucidate both the binding and drug-resistance mechanisms
of HIV-1 protease.
MATERIALS AND METHODS
In this study, the crystal structures of both substrate (1f7a, ca-p2 (3); 1kj4,
ma-ca (1); 1tsu, nc-p1 (4); 1kjf, p1-p6 (1); 1kj7, p2-nc (1); 1kjh, rh-in
(1); 1kjg, rt-rh (1)) and inhibitor (1hpv, amprenavir-apv (41); 2fxe, atazana-
vir-atv (42); 1t3r, darunavir-drv (43); 1hsg, indinavir-idv (44); 1mui,
lopinavir-lpv (45); 1ohr, nelfinavir-nfv (46); 1hxw, ritonavir-rtv (47);
1hxb, saquinavir-sqv (48); 1d4y, tipranavir-tpv (49)) liganded protease
were analyzed by ANM (35). All atoms of the structure, except hydrogens,
were incorporated into the calculations.
Anisotropic network model
ANM (35) performs a harmonic vibrational analysis of protein structures
around their equilibrium states and predicts the directionalities of the
collective motions, as well as their magnitudes. The elastic network is
formed by connecting all neighboring atoms except hydrogen atoms, and
the conformations that describe the fluctuations of residues from the
average in the principal directions of motion are generated. The total poten-
tial energy for a system of N nodes is the summation over all harmonic
interactions of close-neighboring (i, j) pairs, calculated as:
V ¼ ðg=2Þ
where g is the harmonic force constant; Rijis the instantaneous distance;
Rij0is the equilibrium distance between sites i and j in the native structure;
h(rc? Rij) is the Heaviside step function (which is one if (rc? Rij) R 0, and
zero otherwise); and rcis the cutoff distance (here taken as 9 A˚, which leads
to a highercorrelationofthepredictedfluctuations withthe experimentalB-
factors compared to lower cutoff values tried). A 9 A˚cutoff distance has
successfully been used to account for interresidue interactions in all-atom
structure models (50). Higher cutoff distances are preferred in the ANM
compared to the GNM calculations, whereas lower cutoff distances are
Biophysical Journal 99(5) 1650–1659
Fluctuations in HIV-1 Protease Complexes1651
appropriate in all-atom models compared to a reduced description of resi-
dues with only a carbons (51,52).
The Hessian matrix H is a 3N ? 3N symmetric matrix composed of N ?
N super elements Hij, each of size 3 ? 3, given by the second derivatives of
the potential V. An orthogonal transformation of the real symmetric Hessian
matrix gives the normal modes of the elastic network with 3N ? 6 nonzero
eigenvalues liand corresponding eigenvectors ui:
The cross-correlation between the fluctuations of sites i and j is calcu-
where tr[H?1]ijis the trace of the ijthsubmatrix [H?1]ijof H?1. When i ¼ j,
the self-correlations between the components DRiare obtained. Here, the
knowledge of fluctuation vectors permits us to construct and explicitly
view pairs of alternative conformations sampled by the individual modes
simply by adding the fluctuation vectors 5DRito the equilibrium position
vectors in the respective modes.
RESULTS AND DISCUSSION
Motion in principal directions
The fluctuations in the low-frequency modes (principal
directions) refer to the main functional motion of the struc-
tures, and thus all HIV-1 protease complex structures that
are functional should display similar modes of motion.
These collective modes of motion should be highly robust
against sequence and structure variations.
Indeed, the motion in these most cooperative modes is
similar for all substrate- and inhibitor-bound structures.
In the first mode (Fig. 1 A), both monomers of the protease
rotate around two axes parallel to the z direction, and the
peptide fluctuates from the C-terminus to the N-terminus
in the negative y direction. In the second mode (Fig. 1 B),
there are two axes around which the monomers rotate
parallel to the x (longest axis along which the protease
lies) and z directions. The monomers rotate around the
x axis in opposite directions, and the motion in the substrate
is significant in the edges in the second mode. In the third-
slowest mode, the protease monomers rotate around two
axes parallel to the y direction (see Fig. S1 in the Supporting
Material). The contribution of each of these modes to the
overall protease-ligand dynamics on the average of all struc-
tures is 11%, 10%, and 6%, respectively, with the contribu-
tion of the next-slowest modes decaying very rapidly.
Diversity in orientational correlations
The fluctuations in the low-frequency modes are expected to
describe the functional motion of the ligand-bound HIV-1
protease structure. Therefore, it is of interest to elucidate
whether there is a difference between substrate- and inhib-
itor-bound HIV-1 protease structures in these functional
motions, and what such a difference might imply. The inner
products of fluctuation vectors for the two cooperative
modes are calculated to observe the orientational correla-
tions between the fluctuations of the same residues in dif-
ferent substrate- and inhibitor-bound complex structures.
The orientational correlations in the slowest two modes
are given in Figs. 2 and 3, where the normalized correlation
values range between 1 and ?1. The correlations for
all substrate- and inhibitor-bound complex structures are
included in the plots, which reflect the dynamic conforma-
tional ensemble of the protease complex structures. The
peaks with negative correlation values in the charts (the
troughs) indicate the residues that fluctuate more diversely,
i.e., the residues that display the maximum variations in
the direction of their fluctuations. The lower correlation
values in the substrate- compared to inhibitor-bound com-
plexes imply that the protease residues are able to depict
fluctuations in more diverse directions in interaction with
the substrates than with the inhibitors. The orientational
freedom of the protease residues is more restricted when
the protease is bound to the inhibitors. The standard devia-
tion from the average orientational correlation for the
substrate-bound structures is higher than that for the inhib-
itor-bound structures, which clearly indicates the richer
conformational space spanned by the former (Fig. S2).
est and (B) second-slowest modes. The fluctuation of the structure in each
mode is represented as the protease moving between the conformations
shown in green and red, and the peptide moving between the conformations
shown in green and blue.
Motion of HIV-1 protease complex structures in the (A) slow-
Biophysical Journal 99(5) 1650–1659
1652Ozer et al.
Also, the asymmetry in the fluctuations between the
two monomers of the substrate-bound protease struc-
tures is higher than that of the inhibitor-bound protease
The relatively restricted structural motion of inhibitor-
bound protease complex structures, and thus the limited
conformational space of the protease, may suggest tighter
binding as well as less diversity for dynamic freedom of
the protease in the bound state. Binding to different sub-
strates gives the protease sample a larger conformational
space, apparently due to the flexible nature of the substrate
structures, which in turn implies more flexibility of the
protease in its interactions with the substrates. In other
words, the substrates are better able to match the conforma-
tional states that are intrinsically accessible to the protease
structure. This is in agreement with recent work (7) in which
an analysis of a larger set of structures in inhibitor-bound
complex structures with the unbound structures revealed
that drugs select the conformation that best matches its ener-
getic and dynamic properties among those that are intrinsi-
cally accessible to the unbound protein structure.
in the direction of their fluctuations between different struc-
tures in the slowest mode correspond to residues 56, 69, 78,
and 93 (Fig. 2). Residues 56 and 78 are close in space, 56
is the hinge point connecting the flap loop (residues 45–55)
to the solvent-exposed upper arm of the flap (residues
36–44), and 78 is the hinge point connecting the same flap
loop (residues 45–55) to the lower arm of the flap (residues
57–77). Residues 69 and 93 are also close in space; 69 is
the tip of the lower arm of the flap (residues 61–73) and 93
is the tip of another loop (residues 85–96) (Fig. 4 A). The
other minima observed, reflecting the diversity in the fluctu-
ations, yet less celebrated, are at residues 6, 11–13, 22, 33–
35, 44, 45, 57, 66–68, and 71. Of interest, these residues
which the protease monomers rotate in the slowest mode
(Fig. 1 A). In the second-slowest mode, on the other hand,
and 97 (Fig. 3). Residues 49–51 are at the flap tips. In addi-
tion, residues 25–27 are active-site residues at the tip of the
residues in the slowest mode in (A) substrate- and
(B) inhibitor-bound complex structures. The min-
imum value of orientational correlation among
the substrate-bound complexes is shown in red
bars in panel B.
Orientational correlation of protease
Biophysical Journal 99(5) 1650–1659
Fluctuations in HIV-1 Protease Complexes1653
to the 12–22 loop. Residues 84 and 97 are at the edges of the
85–96 loop (Fig. 4 B). Residues 2, 9, 16, 23–25, 28, 29, 39,
48, 52, 53, 61, 62, 74, and 75 also display diversity, albeit
to a lesser extent, in the fluctuation direction of the second
mode. These residues are also observed to lie along the rota-
tional axes (Fig. 4 B) around which the monomers rotate in
this mode (Fig. 1 B). The same applies for the third-slowest
mode, which has less significant motion compared to the
is observed at residues 13, 19, 20, 33, 64, 73, and 81, which
lie along the hinge axes that mediate the behavior in this
mode (Fig. S3 and Fig. S4). Apparently, the orientational
difference in fluctuations, and the asymmetry of fluctuations
between the monomers of the substrate- and inhibitor-bound
protease are caused mainly by these residues, which are part
motions of the complex structure.
The conformational freedom of the protease residues is
restricted when the protease is bound to the ligand. This is
expected because of the additional interaction between
protease residues and the ligand, which appears with
residues in the second-slowest mode in (A)
substrate- and (B) inhibitor-bound complex struc-
tures. The minimum value of orientational correla-
tion among the substrate-bound complexes is
shown in red bars in panel B.
Orientational correlation of protease
difference in the (A) slowest and (B) second-slow-
est modes. The protease monomers are displayed
in light blue and light pink, and the ligand is dis-
played in green. The residues that display maximal
variations in the direction of their fluctuations
between different complex structures are displayed
as blue spheres. The dashed lines indicate the rota-
tional axes around which the monomers rotate, i.e.,
the hinge axes along which the residues with
maximum orientational difference lie.
Regions that cause an orientational
Biophysical Journal 99(5) 1650–1659
1654Ozer et al.
additional springs in the elastic network description. Never-
theless, it should be noted that the same cutoff distance
value is used for both the substrate- and inhibitor-bound
structures studied, and the relative behavior of the protease
in interaction with the substrates versus inhibitors is of
interest here. Moreover, on average, the substrates comprise
a higher number of atoms than the inhibitors, indicating
relatively more additional springs for the substrate-bound
conformations; that is, although the number of springs
added in the ANM is larger for the substrates than for the
inhibitors, the proteases display more flexibility in interac-
tion with the substrates than the inhibitors.
Extent of mobility in the slowest modes of motion
The distribution of mobilities among residues driven by
different frequency modes can be represented by mode
shapes, such that the minima correspond to the hinge
regions (53). The residues that display the maximum differ-
ences in the direction of their fluctuations between different
structures in the respective modes have low mobility, i.e.,
low mean-square fluctuations, in the mode shape of the
bound HIV-1 protease structure. Hence, the distribution of
the mean-square fluctuations in the two low-frequency
modes shows that the residues with maximal orientational
differences between different structures in the first mode
(i.e., 56, 69, 78, and 93) and those in the second mode
(i.e., 25–27, 49–51, 84, and 97) are in the minima of the cor-
responding mode shapes presented in Fig. 5, where the
values are averaged over all of the substrate-bound protease
structures. A similar average mode shape is also revealed
by the inhibitor-bound protease structures. The residues
causing the orientational difference in the third-slowest
mode can also be seen in the fluctuation profile in the
over all the substrate-bound protease complex
structures in the (A) slowest and (B) second-slowest
modes. The fluctuations are calculated for all the
atoms in the structure, but the fluctuations of Ca
atoms are plotted for clearer representation, and
corresponding residue numbers are indicated on
the x axis.
Mean-square fluctuations averaged
Biophysical Journal 99(5) 1650–1659
Fluctuations in HIV-1 Protease Complexes1655
respective mode (Fig. S5). The crucial sites in the protease,
such as the dimerization region (residues 5–10), the active
site (residues 25–27), the flap (residues 45–55), and the
substrate cleft (residues 80–90), are observed to display
relatively low fluctuations in all three slowest mode shapes.
Yet, the slow fluctuations of the active site and flap residues
are emphasized more in the second mode (Fig. 5), as con-
firmed by the hinge axis along which they lie in this mode
(Fig. 4). The similar positions for the hinge regions are
also observed when conformation generation is performed
with geometric constraints using tCONCOORD (22).
Network of cooperative fluctuations
The motion coordinated by the hinge axes results in a
network of cooperative fluctuations within a structure that
could have a plausible link to its function. When the inter-
and intrachain correlations in the protease dimer for
the average of the first 10 modes, which account for 48%
of the overall dynamics, are analyzed in both substrate-
and inhibitor-bound protease structures (Fig. 6), the dimer-
ization region (residues 5–10), the active sites (residues
25–27), the flaps (residues 45–55), and the substrate cleft
(residues 80–90) are observed to be highly positively corre-
lated with each other. The positive correlations of these
hinge regions across the interface account for the associa-
tion of the two chains. As for the interaction between the
protease and the peptide, the residues of the protease that
display positively correlated fluctuations with the fluctua-
tions of the peptide’s residues also correspond to these hinge
regions. On the other hand, the highly mobile regions corre-
sponding to the loops of residues 12–22 and 36–44 in the
two slowest modes display negative correlations with both
the hinge regions and the peptide in the average of the first
10 modes (Fig. 6). The mobile regions and residues that
are important for protein stability or that take part in the
key native contacts have been addressed for HIV-1 protease
in previous studies (13,37,38). Although the flap region
45–55 is part of the relatively mobile flap, it has reduced
mobility in the bound state due to its low tolerance to muta-
tions (37,38). Nevertheless, region 36–44 is located at the
solvent-exposed parts of the flap and has the highest
mobility. It is possible that these regions surround and
anchor the peptide in the cleft between the two monomers
(38). The coupled fluctuations of the flaps, the active site,
the dimerization interface, and the substrate cleft together
with the ligand, as emphasized in the cross-correlation map,
constitute a dynamic domain, suggesting that the enzymatic
activity is an entire property of the protease structure, rather
than localized in or near the active site (36).
The hinge regions composing the minima in the slowest
mode shapes (Fig. 5) that also cause diversity in orientation
(Fig. 4) serve as the mechanistically crucial sites that mainly
coordinate the intrinsic dynamics of the structure, as indi-
cated by the three-dimensional structural motion described
above (Fig. 1). The slowest two modes contribute to the
overall dynamics in similar amounts (11% and 10%, respec-
tively) and suggest the following when looked individually:
The hinges suggested by the slowest mode mostly coordi-
nate the intrachain cooperative motions, whereas the hinges
of the second-slowest mode are mostly responsible for the
correlations across the dimerization interface.
The correlated fluctuations (mainly referred to as posi-
tively correlated fluctuations here) between the two protease
protease in the 10 slowest ANM modes. (B) Protease regions significant for binding are color-coded by the correlations of protease residues to peptide resi-
dues. (C) Protease regions significant for dimerization are color-coded by the correlations between the residues of the two protease monomers. For clarity, the
peptide is not included.
Cross-correlations between the fluctuations of residues. (A) Correlations shown for the representative ca-p2 complex structure of HIV-1
Biophysical Journal 99(5) 1650–1659
1656 Ozer et al.
monomers, and those between each monomer and peptide
positions have some differences when studied separately.
The number of peptide atoms that are positively correlated
to each protease atom is higher for inhibitor-bound
than substrate-bound complex structures, even though the
substrates comprise a higher number of atoms. The sig-
nificant residues for binding in all complex structures are
mainly located in four regions (residues 8–10 (dimerization
region), 25–27 (active site), 45–55 (flap), and 80–90 (sub-
strate cleft)), yet the active sites and the flaps are more
emphasized compared to the other two regions. The well-
conserved residues (e.g., residues 25, 27–29, and 49 (54))
in these hinge regions have been demonstrated to be critical
for substrate binding (55,56), and thus resistance-evading
potent drugs should interact strongly with these residues.
In dimerization, on the other hand, the active sites of the
protease monomers interact strongly in the inhibitor-bound
complex structures, yet the interactions between the flaps
and between the N- and C-termini of the monomers are
weakened compared to the substrate-bound complex struc-
tures. The interactions between the two monomers in all
complex structures, i.e., the critical residues in dimerization,
are also mainly located in the same four specific regions as
in binding to the peptide. However, the dimerization region,
the active site, and the flaps are more emphasized in the
cooperative fluctuations compared to the substrate cleft.
In addition, the N- and C-termini of the monomers are
highlycorrelated with each other. The importanceof a dimer
interface for drug targeting was noted in previous works
(57,58), which demonstrated that inhibitors that act as allo-
steric inhibitors in binding at the dimer interface and alter
the conformation of the protease can indirectly reduce the
binding affinity of substrates.
The network of cooperative fluctuations can be analyzed
in terms of the positions of drug resistant mutations. In gen-
eral, drug resistance occurs when mutations in the target
protein enable it to retain function while no longer being
effectively inhibited by the drug (59). The drug-resistant
mutations in the HIV-1 protease render the variant protease
resistant to the inhibitor while allowing the enzyme to
cleave substrates (59). The invariant positions include the
active site (residues 25–27), some positions in close contact
with the active site or near the substrate cleft (residues 28,
29, 31, 80, 81), most of the N- and C-terminal sites on the
dimerization interface, and other positions associated with
maintaining the conformational flexibility, such as the
conserved glycines and the flap tips (residues 49, 51, and
52) (54). Drug resistance that alters the balance between
substrate recognition and inhibitor binding often occurs by
combinations of mutations both inside and outside the active
site. The most frequent mutations found in drug-resistant
isolates of protease involve positions 10, 36, 46, 54, 71,
77, 82, 84, and 90 (60). Of interest, positions 10, 36, 46,
71, and 77 correspond to either the residues or the first
neighbors of the residues lying along the hinge axes in the
slowest ANM mode, whereas positions 10, 54, 82, 84, and
90 are associated in a similar way with the second-slowest
mode (see Fig. S6, A and B). These hinge axes appear to
mostly coordinate the intrachain cooperative fluctuations
in the slowest mode, and the interchain cooperative fluctua-
tions together with ligand interactions in the second-slowest
mode. Color-coding of the cooperative fluctuations of pro-
tease relevant to intramolecular interactions, dimerization,
and substrate binding can be viewed with respect to these
frequent mutations in Fig. S6, C–E. Residues 10, 36, 46,
54, 71, 77, 82, 84, and 90 are highly positively correlated
in the intramolecular interactions. Residues 10, 46, 54, 82,
84, and 90 are associated with the highly positively corre-
lated regions specific to binding. Residues 10, 82, and 84,
which are in contact with the ligand, are positively corre-
lated at the dimerization interface. Residue positions 10,
36, 46, 54, 71, 77, and 90 are the common mutation sites
outside the active site and the substrate cleft. These may
not only impact inhibitor binding but also compensate for
the viability and fitness of the enzyme and thus increase
the growth rate of the mutant virus.
As presented in this study, the stronger binding of the inhib-
itors (55,56,58,61) results in restricted motion of the com-
plex structures compared to the substrates. By contrast,
because of their higher flexibility, the substrates are more
adaptable to the protease’s backbone rearrangements or con-
formational changes. The similarity in orientation of the
fluctuations of inhibitor complexes, together with their simi-
larity in three-dimensional shape and electrostatic character,
may also have implications for multidrug resistance. In
conclusion, our analysis of the fluctuations of the ligand-
bound HIV-1 protease structures using ANM identifies
a functionally plausible dynamic motion between substrate-
and inhibitor-bound complex structures. The results pre-
sented here may help elucidate the plasticity of the
ensemble of ligand-bound HIV-1 protease conformations
and aid in drug design.
Figures corresponding to the third-slowest mode of ANM of the HIV-1
protease complex structures, the standard deviations from the average
orientational correlations in the slowest and second-slowest modes, and
the location of most frequent drug-resistant mutations on the protease in
relation to the orientation and network of cooperative fluctuations are
available at http://www.biophysj.org/biophysj/supplemental/S0006-3495
This research was supported by the National Institutes of Health AIDS-
FIRCA RO3 TW006875-01. T.H. acknowledges Turkish State Planning
Organization grant 2009K120520 and Betil Fund. C.A.S. acknowledges
grant R01 GM65347.
Biophysical Journal 99(5) 1650–1659
Fluctuations in HIV-1 Protease Complexes1657
1. Prabu-Jeyabalan, M., E. Nalivaika, and C. A. Schiffer. 2002. Substrate
shape determines specificity of recognition for HIV-1 protease: anal-
ysis of crystal structures of six substrate complexes. Structure.
2. Chou, K. C. 1996. Prediction of human immunodeficiency virus
protease cleavage sites in proteins. Anal. Biochem. 233:1–14.
3. Prabu-Jeyabalan, M., E. Nalivaika, and C. A. Schiffer. 2000. How does
a symmetric dimer recognize an asymmetric substrate? A substrate
complex of HIV-1 protease. J. Mol. Biol. 301:1207–1220.
4. Prabu-Jeyabalan, M., E. A. Nalivaika, ., C. A. Schiffer. 2004. Struc-
tural basis for coevolution of a human immunodeficiency virus type 1
nucleocapsid-p1 cleavage site with a V82A drug-resistant mutation in
viral protease. J. Virol. 78:12446–12454.
5. Wlodawer, A., and J. W. Erickson. 1993. Structure-based inhibitors of
HIV-1 protease. Annu. Rev. Biochem. 62:543–585.
6. Yang, L., G. Song, ., R. L. Jernigan. 2008. Close correspondence
between the motions from principal component analysis of multiple
HIV-1 protease structures and elastic network modes. Structure.
7. Bakan,A., and I.Bahar. 2009.The intrinsic dynamicsofenzymes plays
a dominant role in determining the structural changes induced upon
inhibitor binding. Proc. Natl. Acad. Sci. USA. 106:14349–14354.
8. Micheletti, C., G. Lattanzi, and A. Maritan. 2002. Elastic properties of
proteins: insight on the folding process and evolutionary selection of
native structures. J. Mol. Biol. 321:909–921.
9. Keskin, O., S. R. Durell, ., D. G. Covell. 2002. Relating molecular
flexibility to function: a case study of tubulin. Biophys. J. 83:663–680.
10. Bahar, I., A. R. Atilgan, ., B. Erman. 1998. Vibrational dynamics of
folded proteins: significance of slow and fast motions in relation to
function and stability. Phys. Rev. Lett. 80:2733–2736.
11. Bahar, I. 1999. Dynamics of proteins and biomolecular complexes:
inferring functional motions from structure. Rev. Chem. Eng. 15:
12. Doruker, P., A. R. Atilgan, and I. Bahar. 2000. Dynamics of proteins
approaches: application to a-amylase inhibitor. Proteins. 40:512–524.
13. Micheletti, C., F. Cecconi, ., A. Maritan. 2002. Crucial stages of
protein folding through a solvable model: predicting target sites for
enzyme-inhibiting drugs. Protein Sci. 11:1878–1887.
14. Go, N., T. Noguti, and T. Nishikawa. 1983. Dynamics of a small glob-
ular protein in terms of low-frequency vibrational modes. Proc. Natl.
Acad. Sci. USA. 80:3696–3700.
15. Brooks, B., and M. Karplus. 1983. Harmonic dynamics of proteins:
normal modes and fluctuations in bovine pancreatic trypsin inhibitor.
Proc. Natl. Acad. Sci. USA. 80:6571–6575.
16. Case, D. A. 1994. Normal mode analysis of protein dynamics. Curr.
Opin. Struct. Biol. 4:285–290.
17. Hinsen, K. 1998. Analysis of domain motions by approximate normal
mode calculations. Proteins. 33:417–429.
18. Tama, F., and Y. H. Sanejouand. 2001. Conformational change of
proteins arising from normal mode calculations. Protein Eng. 14:1–6.
19. Ma, J. 2005. Usefulness and limitations of normal mode analysis
in modeling dynamics of biomolecular complexes. Structure.
20. Cui, Q., and I. Bahar. 2005. Normal Mode Analysis: Theory and
Applications to Biological and Chemical Systems. CRC Press, Boca
21. de Groot, B. L., D. M. F. van Aalten, ., H. J. Berendsen. 1997. Predic-
tion of protein conformational freedom from distance constraints.
22. Seeliger, D., J. Haas, and B. L. de Groot. 2007. Geometry-based
sampling of conformational transitions in proteins. Structure.
23. Thorpe, M. F., M. Lei, ., L. A. Kuhn. 2001. Protein flexibility and
dynamics using constraint theory. J. Mol. Graph. Model. 19:60–69.
24. Rader, A. J., and I. Bahar. 2004. Folding core predictions from
networks models of proteins. Polymer (Guildf.). 45:659–668.
25. Bahar, I., and A. J. Rader. 2005. Coarse-grained normal mode analysis
in structural biology. Curr. Opin. Struct. Biol. 15:586–592.
26. Schuyler, A. D., and G. S. Chirikjian. 2005. Efficient determination of
low-frequency normal modes of large protein structures by cluster-
NMA. J. Mol. Graph. Model. 24:46–58.
27. Nicolay, S., and Y. H. Sanejouand. 2006. Functional modes of proteins
are among the most robust. Phys. Rev. Lett. 96:078104.
28. Kim,M. K., Y. Jang, andJ. I. Jeong.2006.Usingharmonicanalysis and
optimization to study macromolecular dynamics. Int. J. Control Autom.
29. Leherte, L., and D. P. Vercauteren. 2008. Collective motions in protein
structures: applications of elastic network models built from electron
density distributions. Comput. Phys. Commun. 179:171–180.
30. Csermely, P. 2008. Creative elements: network-based predictions of
active centres in proteins and cellular and social networks. Trends Bio-
chem. Sci. 33:569–576.
31. Zheng, W. 2008. A unification of the elastic network model and the
Gaussian network model for optimal description of protein conforma-
tional motions and fluctuations. Biophys. J. 94:3853–3857.
32. Tirion, M. M. 1996. Large amplitude elastic motions in protein from
a single-parameter atomic analysis. Phys. Rev. Lett. 77:1905–1908.
33. Bahar, I., A. R. Atilgan, and B. Erman. 1997. Direct evaluation of
thermal fluctuations in proteins using a single-parameter harmonic
potential. Fold. Des. 2:173–181.
34. Haliloglu, T., I. Bahar, and B. Erman. 1997. Gaussian dynamics of
folded proteins. Phys. Rev. Lett. 79:3090–3093.
35. Atilgan, A. R., S. R. Durell, ., I. Bahar. 2001. Anisotropy of fluctua-
tion dynamics of proteins with an elastic network model. Biophys. J.
36. Zoete, V., O. Michielin, and M. Karplus. 2002. Relation between
sequence and structure of HIV-1 protease inhibitor complexes: a model
system for the analysis of protein flexibility. J. Mol. Biol. 315:21–52.
37. Kurt, N., W. R. Scott, ., T. Haliloglu. 2003. Cooperative fluctuations
of unliganded and substrate-bound HIV-1 protease: a structure-based
analysison avariety of conformations fromcrystallographyand molec-
ular dynamics simulations. Proteins. 51:409–422.
38. Micheletti, C., P. Carloni, and A. Maritan. 2004. Accurate and efficient
description of protein vibrational dynamics: comparing molecular
dynamics and Gaussian models. Proteins. 55:635–645.
39. Hamacher, K., and J. A. McCammon. 2006. Computing the amino acid
specificity of fluctuations in biomolecular systems. J. Chem. Theory
40. Hamacher, K. 2009. Temperature dependence of fluctuations in HIV-1
protease. Eur. Biophys. J. 39:1051–1056.
41. Kim, E. E., C. T. Baker, ., M. A. Navia. 1995. Crystal structure of
HIV-1 protease in complex with Vx-478, a potent and orally bioavail-
able inhibitor of the enzyme. J. Am. Chem. Soc. 117:1181–1182.
42. Klei, H. E., K. Kish, ., S. Sheriff. 2007. X-ray crystal structures of
human immunodeficiency virus type 1 protease mutants complexed
with atazanavir. J. Virol. 81:9525–9535.
43. Surleraux, D. L., A. Tahri, ., P. B. Wigerinck. 2005. Discovery and
selection of TMC114, a next generation HIV-1 protease inhibitor.
J. Med. Chem. 48:1813–1822.
44. Chen, Z., Y. Li, ., L. C. Kuo. 1994. Crystal structure at 1.9-A resolu-
tion of human immunodeficiency virus (HIV) II protease complexed
with L-735,524, an orally bioavailable inhibitor of the HIV proteases.
J. Biol. Chem. 269:26344–26348.
45. Stoll, V., W. Qin, ., D. W. Norbeck. 2002. X-ray crystallographic
structure of ABT-378 (lopinavir) bound to HIV-1 protease. Bioorg.
Med. Chem. 10:2803–2806.
Biophysical Journal 99(5) 1650–1659
1658Ozer et al.
46. Kaldor, S. W., V. J. Kalish, ., J. H. Tatlock. 1997. Viracept (nelfinavir Download full-text
mesylate, AG1343): a potent, orally bioavailable inhibitor of HIV-1
protease. J. Med. Chem. 40:3979–3985.
47. Kempf, D. J., K. C. Marsh, J. F. Denissen, E. McDonald, S. Vasava-
nonda, C. A. Flentge, B. E. Green, L. Fino, C. H. Park, X. P.
Kong., 1995. ABT-538 is a potent inhibitor of human immunodefi-
ciency virus protease and has high oral bioavailability in humans.
Proc. Natl. Acad. Sci. USA. 92:2484–2488.
48. Krohn,A.,S. Redshaw, .,M. H.Hatada.1991.Novel bindingmodeof
highly potent HIV-proteinase inhibitors incorporating the (R)-hydrox-
yethylamine isostere. J. Med. Chem. 34:3340–3342.
49. Thaisrivongs, S., ., H. I. Skulnick, K. D. Watenpaugh. 1996. Struc-
ture-based design of HIV protease inhibitors: sulfonamide-containing
5,6-dihydro-4-hydroxy-2-pyrones as non-peptidic inhibitors. J. Med.
50. Kurkcuoglu, O., R. L. Jernigan, and P. Doruker. 2006. Loop motions of
triosephosphate isomerase observed with elastic networks. Biochem-
51. Pandey, B. P., C. Zhang, ., Y. Zhou. 2005. Protein flexibility predic-
tion by an all-atom mean-field statistical theory. Protein Sci. 14:1772–
52. Sen, T. Z., Y. Feng, ., R. L. Jernigan. 2006. The extent of cooperativ-
ity of protein motions observed with elastic network models is similar
for atomic and coarser-grained models. J. Chem. Theory Comput.
53. Liu, X., H. A. Karimi, ., I. Bahar. 2004. Protein functional motion
query and visualization. Proc. Annu. Int. COMPSAC, 28th. 86–89.
54. Wu, T. D., C. A. Schiffer, ., R. W. Shafer. 2003. Mutationpatterns and
structural correlates in human immunodeficiency virus type 1 protease
following different protease inhibitor treatments. J. Virol. 77:4836–
55. Wang, W., and P. A. Kollman. 2001. Computational study of protein
specificity: the molecular basis of HIV-1 protease drug resistance.
Proc. Natl. Acad. Sci. USA. 98:14937–14942.
56. Hou, T., W. A. McLaughlin, and W. Wang. 2008. Evaluating the
potency of HIV-1 protease drugs to combat resistance. Proteins.
57. Bowman, M. J., S. Byrne, and J. Chmielewski. 2005. Switching
between allosteric and dimerization inhibition of HIV-1 protease.
Chem. Biol. 12:439–444.
58. Hornak, V., and C. Simmerling. 2007. Targeting structural flexibility in
HIV-1 protease inhibitor binding. Drug Discov. Today. 12:132–138.
59. King, N. M., M. Prabu-Jeyabalan, ., C. A. Schiffer. 2004. Combating
susceptibility to drug resistance: lessons from HIV-1 protease. Chem.
60. Hertogs, K.,S. Bloor, .,B. A.Larder.2000.Phenotypicandgenotypic
analysis of clinical HIV-1 isolates reveals extensive protease inhibitor
cross-resistance: a survey of over 6000 samples. AIDS. 14:1203–1210.
61. Luque, I., M. J. Todd, ., E. Freire. 1998. Molecular basis of resistance
to HIV-1 protease inhibition: a plausible hypothesis. Biochemistry.
Biophysical Journal 99(5) 1650–1659
Fluctuations in HIV-1 Protease Complexes1659