Mining protein dynamics from sets of
crystal structures using ‘‘consensus
structures’’
Gerard J. P. van Westen,
1
Jo¨rg K. Wegner,
2
* Andreas Bender,
1
Adriaan P. IJzerman,
1
and Herman W. T. van Vlijmen
1,2
1
Division of Medicinal Chemistry, Leiden/Amsterdam Center for Drug Research, Einsteinweg 55, 2333 CC Leiden, Netherlands
2
Tibotec BVBA, Generaal de Wittelaan L11B3, 2800 Mechelen, Belgium
Received 14 July 2009; Revised 13 January 2010; Accepted 19 January 2010
DOI: 10.1002/pro.350
Published online 29 January 2010 proteinscience.org
Abstract: In this work, we describe two novel approaches to utilize the dynamic structure
information implicitly contained in large crystal structure data sets. The first approach visualizes
both consistent as well as variable ligand-induced changes in ligand-bound compared with apo
protein crystal structures. For this purpose, information was mined from B-factors and ligand-
induced residue displacements in multiple crystal structures, minimizing experimental error and
noise. With this approach, the mechanism of action of non-nucleoside reverse transcriptase
inhibitors (NNRTIs) as an inseparable combination of distortion of protein dynamics and
conformational changes of HIV-1 reverse transcriptase was corroborated (a combination of the
previously proposed ‘‘molecular arthritis" and ‘‘distorted site" mechanisms). The second approach
presented here uses ‘‘consensus structures" to map common binding features that are present in
a set of structures of NNRTI-bound HIV-1 reverse transcriptase. Consensus structures are based
on different levels of structural overlap of multiple crystal structures and are used to analyze
protein–ligand interactions. The structures are shown to yield information about conserved
hydrogen bonding interactions as well as binding-pocket flexibility, shape, and volume. From the
consensus structures, a common wild type NNRTI binding pocket emerges. Furthermore, we were
able to identify a conserved backbone hydrogen bond acceptor at P236 and a novel hydrophobic
subpocket, which are not yet utilized by current drugs. Our methods introduced here reinterpret
the atom information and make use of the data variability by using multiple structures,
complementing classical 3D structural information of single structures.
Keywords: ligand-induced conformational changes; pocket characterization; flexibility; B-factors;
working mechanism
Introduction
The availability of crystal structures in both public ar-
chives [such as the Protein Data Bank (PDB)
1
] as well
as proprietary repositories (such as within pharma-
ceutical companies) is growing at a phenomenal
speed. Crystal structures can provide a wealth of ex-
perimental data to the scientist, but the information
obtained is static and cannot accurately depict the
actual dynamic properties of the protein and its
ligand.
2
Additional information that can provide
insights into the dynamics is implicitly contained
within a larger group of crystal structures of the same
protein, as this set of structures captures (part of) the
dynamically accessible conformation space of the pro-
tein. The challenge resides in how to mine this wealth
of data. In the work presented here, we will introduce
two different methods for mining large sets of ligand–
protein crystal structures.
Additional Supporting Information may be found in the online
version of this article.
*Correspondence to: Jo¨rg K. Wegner, Tibotec-Virco BVBA,
Generaal de Wittelaan L11B3, 2800 Mechelen. E-mail:
jwegner@its.jnj.com
742 PROTEIN SCIENCE 2010 VOL 19:742—752 Published by Wiley-Blackwell. V
C
2010 The Protein Society

Our first approach mines data from B-factor val-
ues and ligand-induced residue displacements. In a
single structure, information concerning the dynam-
ics is provided by B-factors, which reflect the fluctu-
ations of atoms around their average position in the
crystal.
3
However, B-factors are also influenced by
experimental error, temperature and crystal quality.
Therefore, it is per se difficult to distinguish this
dynamic information from measurement errors and
artifacts in situations where only a single structure
is studied. The utilization of multiple structures can
alleviate this problem,
4
provided that the B-factor
values are normalized before comparing different
structures.
5
A closely related second approach is mapping
ligand-induced changes in residue orientation by
comparing apo structures with ligand bound struc-
tures.
6
Similarly, when only one pair of structures,
that is, one apo structure and one ligand-bound
structure are compared, it is difficult to distinguish
useful information from experimental artifacts. Our
hypothesis governing the current work was that the
simultaneous analysis of several apo and ligand-
bound structures will lead to a better understanding
of information common to all structures, highlight-
ing trends and distinguishing them from artifacts or
noise.
The third approach is to analyze the common
spatial and pharmacophoric interaction properties of
the available crystal structures, which we named
‘‘consensus structures.’’ Existing approaches to
derive a consensus structure have been aimed at
mapping common features of a group of known
ligands and creating a consensus pharmacophore.
7–9
However, this ligand-based approach does not take
any protein information into account, resulting in
the inability of such approaches to extract protein-
related information. Furthermore, it has already
been shown that consensus information derived from
several protein crystal structures can indeed extrap-
olate beyond the original data.
10,11
Consensus struc-
tures are based on the aligned ligand binding pock-
ets of multiple ligand-bound crystal structures and
allow analysis of the shape and pharmacophoric pat-
terns present in all of the structures, as well as the
differences between them. Consensus structures
combine information about different binding site
geometries to identify key features responsible for
ligand binding. Isocontour consensus surfaces that
visualize features common to a minimum percentage
of the total structures used are obtained from these
consensus structures and allow visualization of the
degree of conservation of the protein or protein
features.
In a case study, we applied the above methods
to a data set consisting of human immunodeficiency
virus Type 1 reverse transcriptase (HIV-1 RT) struc-
tures in complex with non-nucleoside reverse tran-
scriptase inhibitors (NNRTIs). HIV-1 RT is one of
the most studied drug targets known today and was
the first target identified in the treatment of infec-
tion with HIV-1.
12
As a result, a large number of
crystal structures are available in the PDB, render-
ing this target suitable for our first case study.
1
NNRTIs are noncompetitive inhibitors of HIV-1 RT
acting on an allosteric binding pocket with high
specificity. However, the nature of HIV-1 replication
leads to a quick onset of resistance of HIV-1 toward
NNRTIs.
13,14
This resistance forms an increasing
problem in the treatment of HIV-1 infection and is
mainly caused by point mutations in the protein.
13–
16
HIV-1 RT is a heterodimer, consisting of a large
560-residue subunit (p66) and a smaller 440-residue
subunit (p51). The catalytic site on the p66 unit con-
sists of a conserved YMDD motif and a third aspar-
tic acid (residues D110, Y183, M184, D185, and
D186). The rather flexible pocket is not present in
the apo form of the enzyme and is only created upon
binding of an NNRTI to HIV-1 RT, thus reducing
enzyme flexibility.
17
Furthermore, it has been shown
that the flexibility of HIV-1 RT depends on its liga-
tion state, and it is increased upon DNA binding.
18
The information mined from the B-factors and
ligand-induced changes of the HIV-RT crystal struc-
tures enabled us to explore the mechanism of
NNRTI inhibition in more detail. The consensus
structure analysis resulted in the identification of
conserved hydrophobic and hydrogen bonding fea-
tures that provided new insights and design options
for HIV-1 RT inhibition by NNRTIs.
Results and Discussion
B-factor analysis
Firstly, normalized B-factors of NNRTI-bound
enzymes were compared with the corresponding val-
ues in apo enzymes. Our results show a significant
decrease in B-factors of the entire pocket upon
NNRTI binding, indicating smaller fluctuations and
a stiffer protein backbone in this region. This reduc-
tion in flexibility is in agreement with earlier MD
simulations.
17,19
Although B-factors of most residues
respond variably upon ligand binding (see Fig. 1),
the loop containing two of the catalytic site residues
(D185 and D186) and the neighboring residues, that
is, residues 181 to 188, show a significant decrease
in flexibility (see also Supporting Information
Fig. S1). The restriction of conformational change of
this loop by NNRTIs, which was proposed to be the
mechanism of action by Das et al.,
20
is fully sup-
ported by our results on the significantly decreased
B-factors of this region. In contrast, the region
between residues L228 and L234 undergoes a con-
sistent average increase in flexibility. This region
contains the ‘‘primer grip" residue M230 and the
increase would be unfavorable for its function
van Westen et al. PROTEIN SCIENCE VOL 19:742—752 743