[10:05 25/8/2011 Bioinformatics-btr437.tex] Page: 2529 2529–2536
BIOINFORMATICS ORIGINAL PAPER
Vol. 27 no. 18 2011, pages 2529–2536
doi:10.1093/bioinformatics/btr437
Structural bioinformatics Advance Access publication July 22, 2011
Structural analysis of the hot spots in the binding between H1N1
HA and the 2D1 antibody: do mutations of H1N1 from 1918 to
2009 affect much on this binding?
Qian Liu1, Steven C. H. Hoi1, Chinh T. T. Su1, Zhenhua Li1, Chee-Keong Kwoh1,
Limsoon Wong2 and Jinyan Li2,3,∗
1School of Computer Engineering, Nanyang Technological University, Singapore, 2School of Computing, National
University of Singapore, Singapore and 3Advanced Analytics Institute, University of Technology Sydney, Australia
Associate Editor: Anna Tramontano
ABSTRACT
Motivation: Worldwide and substantial mortality caused by the 2009
H1N1 influenza A has stimulated a new surge of research on H1N1
viruses. An epitope conservation has been learned in the HA1 protein
that allows antibodies to cross-neutralize both 1918 and 2009 H1N1.
However, few works have thoroughly studied the binding hot spots
in those two antigen–antibody interfaces which are responsible for
the antibody cross-neutralization.
Results: We apply predictive methods to identify binding hot spots
at the epitope sites of the HA1 proteins and at the paratope sites
of the 2D1 antibody. We find that the six mutations at the HA1’s
epitope from 1918 to 2009 should not harm its binding to 2D1.
Instead, the change of binding free energy on the whole exhibits
an increased tendency after these mutations, making the binding
stronger. This is consistent with the observation that the 1918 H1N1
neutralizing antibody can cross-react with 2009 H1N1. We identified
three distinguished hot spot residues, including Lys166, common
between the two epitopes. These common hot spots again can
explain why 2D1 cross-reacted. We believe that these hot spot
residues are mutation candidates which may help H1N1 viruses to
evade the immune system. We also identified eight residues at the
paratope site of 2D1, five from its heavy chain and three from its
light chain, that are predicted to be energetically important in the
HA1 recognition. The identification of these hot spot residues and
their structural analysis are potentially useful to fight against H1N1
viruses.
Contact: jinyan.li@uts.edu.au
Availability: Z-score is available at http://155.69.2.25/liuqian/indexz.py
Supplementary information: Supplementary data are available at
Bioinformatics online.
Received on April 10, 2011; revised on June 19, 2011; accepted on
July 19, 2011
1 INTRODUCTION
The H1N1 influenza A caused two notable pandemics with
substantial mortality in 1918 and 2009. Fortunately, it has been
∗To whom correspondence should be addressed.
found that some antibodies can work against the Hemagglutinin
(HA) proteins in these two pandemics (Xu et al., 2010). HA is
a homotrimeric glycoprotein. HA monomers are synthesized as
precursors that are then cleaved into two proteins, HA1 and HA2,
which form the major surface proteins of influenza A viruses. The
infection is started by the binding of HA proteins to the sialic
acid-containing receptors of target cells and by fusing the viral
membrane with the endosomal membrane of the target cells. The
viral genome enters and infects the target cells after the binding. So,
inhibiting this binding by antibodies is an important way against
flu. Previous works have learned that there is an epitope (binding
site) conservation that exists between the 1918 and 2009 H1N1
HA proteins (Ekiert et al., 2009; Xu et al., 2010). Such epitope
conservation enables the older population to avoid infection from
2009 H1N1 because their pre-existing immunity against 1918 H1N1
can neutralize the 2009 H1N1 HA proteins. Thus, studies on these
antibody–HA binding interfaces are crucial to understand how the
antibodies recognize the antigens. However, there are few studies on
the energetic importance of the binding residues in the HA1 protein
in complex with the 2D1 antibody.
We apply predictive and comparative methods to examine the
interfaces between the 2D1 antibody and the HA1 proteins of 1918
and 2009 H1N1, and to investigate an assumed 2D1 binding to
the seasonal influenza virus A/Brisbane/59/2007 to understand why
2D1 did not bind to the 2007 strain (Krause et al., 2010; Xu
et al., 2010). This 2D1 antibody is a monoclonal antibody from
a survivor of the 1918 Spanish influenza (Yu et al., 2008), which
is believed to bind to HA1s in both of 1918 and 2009 H1N1.
Of particular interests, we identify binding hot spot residues from
the above mentioned two antibody–antigen interfaces. A binding
hot spot is a small fraction of interfacial residues that contribute
most to binding free energy (Bogan and Thorn, 1998; Clackson and
Wells, 1995). Their mutations—e.g. alanine mutations—can reduce
binding affinity remarkably (Clackson and Wells, 1995).
We address the problems whether the interfacial mutations from
1918 H1N1’s HA1 to 2009’s are hot spot residues and whether
these mutations make the binding stronger with 2D1. We explain
how the computational methods find those antigenic residues
that are energetically important in the antibody binding, such as
Asn129, Lys157 and Lys166. These three hot spot residues are
actually common between the 1918 and 2009 epitopes of HA1.
Their mutations may make the 2D1-antibody binding ineffective.
© The Author 2011. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com 2529
atMasarykUniversityonSeptember16,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[10:05 25/8/2011 Bioinformatics-btr437.tex] Page: 2530 2529–2536
Q.Liu et al.
Therefore, they are mutation candidates which may help H1N1
viruses evade the immune system. We also describe and characterize
hot spot residues at the paratope site of the 2D1 antibody—e.g.
Asp52 and Arg97 from the heavy chain and Asn31, Trp91 and Asp93
from the light chain. Knowledge gained from these binding hot spot
studies can be useful to fight against H1N1 viruses in future.
2 METHODS
2.1 The HA-2D1 binding structures
We retrieve from the PDB entry 3LZF the crystal structure of the 2D1
antibody binding to the HA proteins of 1918 H1N1 (1918HA1), and
from 3LZG the atomic coordinates of the HA proteins of the 2009
H1N1 A/California/04/2009 (Xu et al., 2010) (2009HA1). The structure
information of the HA1 proteins of A/Brisbane/59/2007 (2007HA1) is
taken from Igarashi et al. (2010). Our comparative analysis is on the three
interfaces: the interface between 2D1 and 1918HA1 (2D1-1918HA1), the
interface between 2D1 and 2009HA1 (2D1-2009HA1) and an assumed
artificial interface between 2D1 and 2007HA1 (2D1-2007HA1).
We use MAFFT to align the three HA1 sequences to examine the specific
mutations among 1918HA1, 2009HA1 and 2007HA1. We also take the
following steps to produce the structure of the 2D1-2009HA1 (and 2007HA1)
binding. First, we use PyMOL to align the HA protein structure of 2009HA1
(or 2007HA1) onto the HA protein structure of the 1918 H1N1 in the PDB
3LZF with the antibody coordinates. PyMOL aligns the two HA protein
structures to minimize root mean square deviation (RMSD) (Schrödinger,
LLC, 2010). After that, we obtain the computational binding interface
of 2009HA1 (or 2007HA1) with the 2D1 antibody by removing the HA
coordinates of 1918 H1N1. 2009HA1 (or 2007HA1) in this computational
binding interface remains in an unbound state with free side chains. So,
we use FoldX (Schymkowitz et al., 2005) to repair this interface when
fixing the antibody binding site. The repaired interfaces are then used for
our subsequent analysis.
2.2 Computational methods for predicting hot spots
Binding hot spot residues can be predicted by computational methods such
as by Robetta (Kortemme and Baker, 2002), FoldX (Schymkowitz et al.,
2005), KFC (Darnell et al., 2008), GCR (Li and Li, 2010) and a Z-score
method.
Robetta is a simple physical model for estimating the binding energy
of hot spots. This method uses all heavy atoms and polar hydrogens to
represent proteins and proposes a free energy function for linearly combining
such terms as Lennard–Jones potentials, an orientation-dependent hydrogen
bond potential, Coulomb electrostatics and an implicit solvation model.
Similarly, FoldX (Schymkowitz et al., 2005) uses a linear combination of
empirical terms to calculate free energy. The empirical terms are hydrophobic
and polar solvation, hydrogen bonds (water-intermediate hydrogen bonds
included), the Van der Waals terms, Coulomb electrostatics, and so on.
Meanwhile, KFC (Darnell et al., 2008) uses simple rules to identify binding
hot spots. The following features are employed by KFC to represent a residue:
physical and chemical features, shape specificity and biochemical contacts
such as atomic contacts, hydrogen bonds and salt bridges. Then it uses a
decision tree model to produce some rules for classifying hot spots. All
these computational methods achieved good prediction performance based
on experimental mutations. For example, the overall correlation between
the observed and Robetta-calculated changes in binding free energy has an
average unsigned error of 1.06 kcal/mol for interface mutations (Kortemme
and Baker, 2002).
Recently, a novel descriptor of atoms and residues, called burial level
by GCR (Li and Li, 2010), is also proposed to enhance hot spot prediction
performance. By this method, an atomic contact graph is built for a protein
complex, where vertices are atoms and edges are atom contacts. The burial
level of an atom in this graph is defined as the length of the shortest path from
this atom to its nearest exposed atom to the bulk solvent. The burial level of
an atom or a residue indicates the extent it is buried inside the complex. As
the hot spot residues are protected by O-rings (Bogan and Thorn, 1998; Li
and Liu, 2009), hot spot residues always have low solvent accessible surface
area (ASA) and high burial levels. But a high burial level is more sufficient
than ASA: there are very few highly buried interfacial residues that are not
hot spot residues. We have built a hot spot model (Li and Li, 2010) based on
this concept; and the model has achieved good performance.
We have also proposed a Z-score biological significance for capturing
the probability of residues occurring in or contributing to protein binding
interfaces. This Z-score is actually intended to measure how far away certain
properties of a putative contact residue at a binding interface are from those
of crystal packing. So, we take crystal packing as the reference state to extract
residue pairwise potentials. Then, the potential score of a residue is defined
by using a knowledge-based potential function with ASA calculations. After
that, a null distribution of this potential score is generated from artifact crystal
packing contacts. Finally, the Z-score significance of a contact residue with
a specific potential score is determined according to this null distribution.
As binding hot spots contribute greatly to binding free energy, they should
have big Z-score values. Here, a contact residue is considered as a hot spot
residue if its Z-score is >1. Our evaluation on the ASEdb and BID datasets
(Cho et al., 2009) shows that Z-score is powerful for identifying protein
binding hot spots. The details of how to calculate Z-score are given in the
Supplementary Material.
2.3 A meta-learning approach to combine the
computational methods for predicting hot spots
We use the computational methods above to predict whether contact
residues are hot spot residues or not after alanine mutations in the three
interfaces—2D1-1918HA1, 2D1-2007HA1 and 2D1-2009HA1. We use
default parameters for the Robetta and KFC web servers and for the FoldX
software. Since Robetta and FoldX estimate G, we are interested in those
residues whose alanine mutation results in G ≥ 1 kcal/mol. After that,
we apply a meta-learning approach (Vilalta and Drissi, 2002) by combining
the Z-score method with the other methods. The reason is that the Z-score
method has a very high recall with low precision rate; however, the other
methods generally have a low recall but a high precision rate. Therefore, in
this work, we are interested in those hot spot residues that are predicted by
Z-score and are also confirmed by at least one of the other methods (Robetta,
FoldX or KFC). Meanwhile, we also trust with high confidence that non-hot
spot residues predicted by Z-score generally have insignificant contribution
to the binding. The hot spot residues which are predicted by a single method
only are considered having intermediate contribution to binding.
3 RESULTS
The sequence alignment among 1918HA1, 2007HA1 and 2009HA1
is shown in Figure 1a. There are a total of six interfacial mutations
between 1918HA1 and 2009HA1, namely E131D, T133N, S159N,
V169I, N171D and T242K. The structural alignment of 1918HA1
and 2009HA1 between their interfacial segments is shown in
Figure 1c where the structural match—based on the Cαs of these
interfacial residues—has an RMSD of 0.725 Å. Previous works have
reported that protein sequences with ∼50% identity or above in
crystallographic models can differ by ∼1 Å RMSD, while proteins
in NMR models can have even larger deviations (Chothia and Lesk,
1986; Schwede et al., 2000). In some cases, sequences with >95%
identity can also have an interface RMSD up to ∼1.2 Å (Kinjo and
Nakamura, 2010). Thus, the small 0.725 Å RMSD suggests that
the interfacial segments of the two HA1 proteins have a very good
match. This indicates that the mutations from 1918HA1 to 2009HA1
resulted in little structural change at the binding site, and that the
2530
atMasarykUniversityonSeptember16,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[10:05 25/8/2011 Bioinformatics-btr437.tex] Page: 2531 2529–2536
Binding hot spots between H1N1 HA1 protein and 2D1 antibody
Fig. 1. The epitope of HA1 in 1918, 2009 and 2007, and their sequence and structural alignment (better viewed in color). (a) The sequence alignment between
1918HA1(3LZF1918), 2009HA1(3LZG2009) and 2007HA1(Bris2007). Interface residues are in yellow, and the positions are in accordance to the 1918HA1
numbering. (b) The binding interface of 2D1-1918HA1. The HA1 epitope is in cyan and in sphere view, and the antibody paratope is in yellow and in stick
view. (c) The aligned structure of the epitopes between 1918HA1 (cyan) and 2009HA1 (green). In (b) and (c), residues in magenta are the mutations from
1918HA1 to 2009HA1.
computationally produced binding structure of 2009HA1 and the
2D1 antibody may not have any big deviation from the real one.
3.1 Energy change tendency of the six mutations
Using the Z-score method, three of the six mutations—T133N,
S159N and N171D—are predicted as non-hot spot residues in both
1918HA1 and 2009HA1. Two mutations—V169I and T242K—are
believed to contribute, though slightly, to the antibody binding in
2009HA1 only after the mutations. They may be newly formed hot
spots in the epitope of 2009HA1 after the mutations. The remaining
one of the six mutations—E131D—is predicted to contribute to the
binding free energy both before and after the mutation. Robetta,
FoldX and KFC predict all the mutations as non-hot spot residues in
both 1918HA1 and 2009HA1. However, G of V169I is predicted
to increase from 0.4 to 0.76 kcal/mol by Robetta and from 0.07 to
0.83 kcal/mol by FoldX. Hence, the six mutations from 1918HA1 to
2009HA1 do not appear to adversely affect the binding between the
2D1 antibody and the two HA1s. Instead, on the whole, the change
of binding free energy exhibits a possible increased tendency after
the mutation, making the binding stronger. This is consistent with the
result that the 1918HA1 neutralizing 2D1 antibody can cross-react
with 2009HA1 (Xu et al., 2010).
Geometrically, the six mutations are located at the rim of the
binding interface (Fig. 1b and c), forming a part of an O-ring
structure (Bogan and Thorn, 1998; Li and Liu, 2009). Most of
them have a large exposed portion to water and are not deeply
buried. Their absolute and relative ASA information and burial
levels are presented in Table 1. Only the residue at position 131
is buried with little exposure to water. The influential O-ring theory
(Bogan and Thorn, 1998; Li and Liu, 2009) stated that residues
on an O-ring, though of structural importance, are usually not
energetically important. Therefore, these six mutations can only
slightly destruct the antibody binding of 2009HA1 energetically if
they are adverse.
We closely examined the S159N mutation from 1918HA1 to
2009HA1. It was reported to have a high G ∼2 kcal/mol
(Xu et al., 2010). However, we believe that this residue does not
contribute greatly to the antibody binding either before or after the
mutation. First, Ser159 has no significant contact with the antibody;
see Figure 3a. Second, the S159N mutation makes the backbone
deviate far away—the Cα deviation is ∼1.1 Å. This results from
an increased flexibility for Asn159 side chain. At the side chain
of Ser159, OG has a hydrogen bond with its backbone O. The
distance between O and OG is 2.2 Å, and the angle of OG-H···O
is 151.0◦. Therefore, it is this hydrogen bond that confines Ser159
side chain. But the mutation makes Asn159 side chain free to contact
water molecules, which can drag and affect the backbone structure
a lot. Third, the mutation breaks the hydrogen bond of OG and the
backbone O at Ser159, releasing some binding free energy. In fact,
at least half of the 2 kcal/mol G of the mutation should come
from this hydrogen bond break. This can be seen from the mutation
2531
atMasarykUniversityonSeptember16,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[10:05 25/8/2011 Bioinformatics-btr437.tex] Page: 2532 2529–2536
Q.Liu et al.
Table 1. ASAs and burial levels of the six mutation residues from 1918HA1 to 2009HA1
Position 133 159 171 169 242 131
Residue Thra Asnb Sera Asnb Asna Aspb Vala Ileb Thra Lysb Glua Aspb
Absolute ASA (Å) 94.27 104.14 39.36 80.06 111.45 109.72 29.57 43.7 45.79 79.59 15.13 8.42
Relative ASA (%) 67.7 72.3 33.8 55.6 77.4 78.2 19.5 25.0 32.9 39.6 8.8 6.0
Burial level 0.43 0.5 0.83 0.63 0.5 0.63 0.86 0.75 0.57 0.78 1.22 1.25
aResidues in 1918HA1.
bResidues in 2009HA1.
Fig. 2. The binding hot spots in the HA-2D1 binding interfaces (better viewed in color). (a) The hot spots in the 1918HA1 epitope. (b) The hot spots in the
2009HA1 epitope. (c) The artificial hot spots in the 2007HA1 epitope and their glycosylation sites. (d) The hot spots in the antibody paratope of 2D1-1918HA1.
(e) The hot spots in the antibody paratope of 2D1-2009HA1. (f) The hot spots in the artificial paratope of 2D1-2007HA1. (a) and (d) form 2D1-1918HA1,
(b) and (e) form 2D1-2009HA1, and (c) and (f) form artificial 2D1-2007HA1. In (a)–(f), binding sites are in sphere view; residues in red (with yellow labels)
are predicted by Z-score to be hot spot residues and confirmed by at least one of other computational methods; residues in green (with light pink labels) are
predicted only by one method; residues in blue are predicted as non-hot spots by all methods.
of Ser159 to Gly159 (no side chain in Gly), whose G is bigger
than 1 kcal/mol (Xu et al., 2010). Therefore, the S159N mutation
from 1918HA1 to 2009HA1 did not greatly destroy the binding of
the 2D1 antibody to 2009HA1.
As the Z-score method has a high negative precision value for
predicting non-hot spot residues, the above residues predicted as
non-hot spot residues can be considered as energetically unimportant
to the antibody binding with high confidence. The additional
12 non-hot spot predictions in the HA1 epitopes are Pro(125
in 1918HA1, 122 in 2009HA1), Thr(125B in 1918HA1, 124 in
2009HA1), Ser(125C in 1918HA1, 125 in 2009HA1), Ser126, His130,
Gly158, Ser160, Leu164, Ser165, Ser167, Tyr168 and Thr248. Some of
these non-hot spot predictions can be verified by past non-alanine
mutation experiments (Xu et al., 2010). For example, G158E/D,
S160L or S165K cause only a small G (<1 kcal/mol). These also
happened between different residue-type groups, e.g. a mutation
from a polar uncharged residue Ser160 to a hydrophobic residue
Leu160. These suggest that these non-hot spot residues have little
contribution to the binding to either 1918HA1 or 2009HA1, just
as Z-score predicts. So, mutating these predicted non-hot spot
residues provides little chance for H1N1 to evade capture by the
2D1 antibody.
3.2 Hot spot residues at the epitopes of the two HA1s
The hot spot residues in 2D1-1918HA1 or 2D1-2009HA1 predicted
by Robetta, FoldX, KFC and Z-score are shown in Figure 2. All of
them are considered to have potential contribution to the antibody
2532
atMasarykUniversityonSeptember16,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[10:05 25/8/2011 Bioinformatics-btr437.tex] Page: 2533 2529–2536
Binding hot spots between H1N1 HA1 protein and 2D1 antibody
Table 2. Binding hot spots in the two epitopes predicted by Z-score and confirmed by other previous computational methods
Residues Robetta FoldX KFC P-value∗ Absolute ASA (Å) Relative ASA (%) Burial level
HA1a HA1b HA1a HA1b HA1a HA1b 2007HA HA1a HA1b HA1a HA1b HA1a HA1b HA1a HA1b
Pro128 0.83 0.9
√
No 0.02 0.0465 17.42 29.5 12.8 21.7 1.57 1.29
Asn129 1.4
√
No 0.0312 0.0234 27.37 23.01 19.0 16.0 1.25 1.25
Lys157 1.35 1.54
√ √
Yes153 <1E-324 <1E-324 9.69 13.38 4.8 6.7 1.44 1.55
Pro162 1.36 0.73 Yes158 0.0027 0.0018 0.07 0.00 0.1 0.0 2.00 2.00
Lys163 0.92 1.33 1.43 Yesc
159 <1E-324 <1E-324 35.81 32.43 17.8 16.2 1.33 1.11
Lys166 2.98 1.14 4.2 3.73 No <1E-324 <1E-324 4.19 3.05 2.1 1.5 1.67 1.67
The subscript number is the position in 2007HA1.
aResidues in 1918HA1.
bResidues in 2009HA1.
cAsn mutation creates a potential glycosylation site in 2007HA1 (Xu et al., 2010).
∗P-values of Z-score.
binding. In particular, Pro128, Asn129, Lys157, Pro162, Lys163 and
Lys166 are confirmed as hot spot residues at 2D1-1918HA1 or at
2D1-2009HA1 by Z-score and at least one of the other methods;
see Table 2 and Figure 2a and b. Three of them are common in
both 2D1-1918HA1 and 2D1-2009HA1. For the other three residues,
two of them (Pro128 and Pro162) are hot spot residues for 2D1-
1918HA1 and also with not low G in 2D1-2009HA1 (Table 2).
Similar observation can be found for the remaining one (Lys163)
in 2D1-2009HA1. These residues all have a very small ASA, and
are buried with a burial level up to 2.0 (Table 2). These doubleconfirmed
hot spot residues are believed to contribute greatly to
the antibody binding, as the combined prediction by Z-score and
the other computational methods has a much higher precision. So,
they are positions for mutations that can lead to H1N1’s escape from
2D1’s neutralization.
Some of these double-confirmed hot spot residues at 2D1-
1918HA1 or 2D1-2009HA1 are also supported by wet-lab
experiments. For example, the mutation ofAsn to Lys at position 129
and the mutation of Lys toAsn at position 163 in wet-lab experiments
resulted in >1 kcal/mol G (Xu et al., 2010). This fact indicates
that Asn129 and Lys163 are truly energetically important although
the mutations are non-alanine mutations.
Lys166 has been comprehensively studied in the past by wet-lab
experiments. It was found that this residue contributes greatly to this
antibody binding: its mutations to residue types such as its similar
hydrophilic residues Glu and Gln, or Pro resulted in >3 kcal/mol
G (Xu et al., 2010).As can be seen in Table 2, Lys166 is predicted
by three computational methods (Z-score, Robetta and FoldX)
as a hot spot residue in both 1918HA1 and 2009HA1 epitopes.
To investigate why this residue is energetically so important, we
examine its contacts using Figure 3b. First, this residue is deeply
buried with a small ASA and high burial level. Second, Lys166
has several hydrogen bonds with its NZ atom as the donor: one
hydrogen bond forms with the backbone O of the Ser126 from the
same chain as Lys166, and the other two form with the side chain
O of Asp93 and of Asn31 from the light chain of the antibody.
Third, CE of Lys166 has a π-involving contact with Trp127 from
the same chain as Lys166. These contacts suggest that this residue
contributes greatly to the antibody binding and to the antigen folding
by the π-involving contact with Trp127 and the hydrogen bond
with the Ser126. In fact, Ser126 is also at the epitope site, and it is
predicted as a hot spot residue by FoldX and Robetta in the 2009HA1
epitope site but as a non-hot spot residue in the 1918HA1 by all the
methods; so this hydrogen bond contributes to the antibody binding
indirectly.
The residue Lys166 was also reported as a selected escape
mutation at 2D1-antibody by several viruses including 2009 H1N1,
1918 H1N1 and the 1930 swine viruses (Krause et al., 2010; Xu
et al., 2010; Yu et al., 2008). What is more interesting is that
Lys166’s two hydrogen bond contact residues, Asp93 and Asn31, are
all predicted to be hot spot residues by Z-score and more than one
other methods, instantiating the hot spot coupling property (Halperin
et al., 2004).
Although we have not found wet-lab evidence and report for
Pro128, Lys157 or Pro162, we suggest that all of the six doubleconfirmed
hot spot residues—Pro128, Asn129, Lys157, Pro162,
Lys163 and Lys166—are potential escape mutations for H1N1 to
elude the 2D1 antibody.
3.3 Hot spot residues at 2D1’s paratope
We also studied those residues in the paratope (the antigen binding
site) of the 2D1 antibody that can contribute greatly to the binding.
These antibody hot spot residues can uncover how the 2D1 antibody
captures the H1N1 viruses.
Using the Z-score method, all and only six hot spot residues
are predicted in the antibody light chain which are also common
between 2D1-1918HA1 and 2D1-2009HA1. Meanwhile, eight hot
spot residues in the antibody heavy chain are identified in 2D1-
1918HA1, and seven in 2D1-2009HA1. These predicted paratope
hot spots are depicted in Figure 2d and e. Among them, five from
the heavy chain and three from the light chain are confirmed by more
than one existing computational methods (Table 3). In the antibody
light chain, the hot spot residues Asp93 and Asn31 have significant
contacts with the antigen hot spot residue Lys166 as we discussed
above. We believe that they are mainly responsible for the binding
to the antigen.
In the antibody heavy chain, we are interested in the predicted hot
spot residue Arg97, as it is predicted to be energetically important by
three methods in both 2D1-1918HA1 and 2D1-2009HA1 (Table 3).
2533
atMasarykUniversityonSeptember16,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[10:05 25/8/2011 Bioinformatics-btr437.tex] Page: 2534 2529–2536
Q.Liu et al.
Fig. 3. Three examples of (non-)hot spot predictions in 2D1-1918HA1 and 2D1-2009HA1 (better viewed in color). (a) The mutation of Ser159 to Asn159
from 1918HA1 (cyan) to 2009HA1 (brown) when binding to 2D1 (the heavy chain in black); position 159 is in red and in stick view. (b) The hot spot residue
Lys166 when both 1918HA1 (black) and 2009HA1 (brown) bind to 2D1 (the light chain in purple); the residues Lys166 are in red and in stick view. (c) The
hot spot residues Asp52 and Arg97 in magenta in the paratope of the antibody heavy chain. (d) The cavity in the binding interface surrounded by the binding
residues. In (c) and (d), the whole complex and the core interface are shown in surface view, and so the epitope and the paratope have no surface; 1918HA1,
the antibody heavy chain and the antibody light chain are in cyan, green and brown, respectively.
Its close contact with Asp52 is shown in Figure 3c. Asp52 is also
from the antibody heavy chain and confirmed as a hot spot residue
by Robetta and FoldX.As seen in Figure 3c,Arg97 has two hydrogen
bonds with Asp52 with NH1 and NH2 as donors and OD1 and OD2
as acceptors. Furthermore, Arg97’s NH2 has a hydrogen bond with
OD1 from Asp53, and Arg97’s NH1 has another hydrogen bond
with OD1 from the antigen residue Asn129. Asn129 is considered
to contribute greatly to the antibody binding by the two existing
methods (Table 2). These contacts form a hydrogen-bond network
which is believed to generate a favorable electrostatic contribution to
the protein binding that can strongly stabilize the protein complexes
(Sheinerman and Honig, 2002).Another finding is that the side chain
of Arg97 confines at least three side chains: the side chain of Arg97
which is positively electrically charged, and the side chains ofAsp52
and Asp53 which are negatively charged. These side chains should
prefer solvent water molecules if they are free. So, confining by
the side chain of Arg97 can make remarkable contribution to the
antibody binding by removing the freedom of these three charged
side chains.
Our investigation also finds a large cavity at the core of the binding
interface, as seen in Figure 3c and d. This cavity has a surface
surrounded by the binding residues. Its narrowest part is >9 Å wide,
which is equivalent to more than three water molecule diameters
(2.75 Å). What is more important is that the side chains of Arg97
and Asp52 are in the rim of the cavity, and some side chain atoms of
Asp53 contact the solvent as seen in Figure 3c and d. Thus, removing
the side chains ofArg97 and/orAsp52 would increase the chance that
this cavity is open to contact the solvent. In other words, the binding
2534
atMasarykUniversityonSeptember16,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[10:05 25/8/2011 Bioinformatics-btr437.tex] Page: 2535 2529–2536
Binding hot spots between H1N1 HA1 protein and 2D1 antibody
Table 3. Binding hot spots in the antibody paratopes predicted by Z-score and confirmed by other previous computational methods
Residues Robetta FoldX KFC P-value∗ Absolute ASA (Å) Relative ASA (%) Burial level
2D1a 2D1b 2D1a 2D1b 2D1a 2D1b 2007HA 2D1a 2D1b 2D1a 2D1b 2D1a 2D1b 2D1a 2D1b
Asp52 3.93 2.85 1.06 No 0.2522 0.1735 2.44 2.06 1.7 1.5 1.88 1.88
Tyr58 0.96 1.35 No 0.0585 0.1691 23.57 26.04 11.1 12.2 1.33 1.33
Arg97 4.5 4.57 1.32 3.53 Yesc 0.0075 0.0106 0.71 0.69 0.3 0.3 2.00 2.00
Tyr100B 2.28 1.14 Yesc 0.0178 0.015 26.35 33.65 12.4 15.8 1.17 1.08
dVal100C 1.02 1.16 Noc 0.2539 0.3271 16.97 16.5 11.2 10.9 1.57 1.57
eAsn31 2.44 1.56 1.68 1.86 No 0.3236 0.3251 11.30 10.62 7.9 7.4 1.5 1.88
Trp91 3.05 3.02 2.96 2.7
√
Yes <1E-324 <1E-324 12.39 10.99 5.0 4.4 1.29 2.14
Asp93 3.08 3.32 2.78
√
No 0.016 0.0065 13.34 11.13 9.5 7.9 1.25 1.25
Italic are for the antibody heavy chain, and bold for the light chain.
aResidues in 2D1-1918HA1.
bResidues in 2D1-2009HA1.
cParatope residues which are close to potential glycosylation sites in 2D1-2007HA1.
dZ-score failed to identify this hot spot in 2D1-2009HA1.
e Z-score failed to identify this hot spot.
∗P-values of Z-score.
interface would have a larger open empty or solvent core. This makes
it impossible to have stable binding. Hence, we believe that both
Arg97 and Asp52 are energetically significant in the binding.
3.4 Analysis on the assumed 2D1-2007HA1 binding
The hot spot prediction results on the assumed artificial 2D1-
2007HA1 interface are also presented in Tables 2, 3 and Figure 2c
and f. We find that more than half of the predicted hot spots in the
binding of 2D1-1918HA1 or 2D1-2009HA1 are not predicted to
make contribution to the artificial binding 2D1-2007HA1, although
Leu156 is newly predicted as a hot spot residue at the epitope of
2D1-2007HA1.
We have two interesting remarks about this assumed binding.
One remark is about Lys162 in 2007HA1 (Lys166 in 1918HA1 and
2009HA1) which is predicted as a non-hot spot residue. First, this
residue is conserved in the three HA1s. Second, it is predicted as a
hot spot residue in 1918HA1 and 2009HA1 by all the computational
methods here. Third, its binding importance in 2D1-1918HA1 and
2D1-2009HA1 has been demonstrated by Xu et al. (2010). So,
the reduced contribution of this Lys in 2D1-2007HA1 indicates an
escape of 2007HA1 from 2D1. This together with the less number
of predicted hot spots in 2D1-2007HA1 suggests a very small
occurrence probability of 2D1-2007HA1. The other remark is that
although Asn159 in 2007HA1 (Lys163 in 1918HA1 and 2009HA1)
is predicted to contribute to the artificial 2D1-2007HA1 binding, this
mutation creates a potential N-glycosylation site (Xu et al., 2010) as
shown in Figure 2c. 2007HA1 contains another glycosylation site
Asn125 (Xu et al., 2010); see Figure 2c. To better understand the
assumed binding, we use Figure 2f to depict the binding region of
2D1’s paratope to the glycosylation sites, and this region covers
two of the three predicted hot spot residues of the paratope. It
can be observed that the glycosylation sites mask the surface of
2007HA1 to block the cross-neutralization by 2D1 (Xu et al., 2010).
The computational methods did make some predictions of hot spots
in 2D1-2007HA1, because none of them considers the potential
glycosylation sites but only the residue information. In fact, these hot
spot predictions are not true if the glycosylation sites are considered.
In summary, 2D1 cannot recognize 2007HA1 for neutralization.
4 CONCLUSION
We have done a structural analysis on the interfaces between the
2D1 antibody and the HA1 proteins of 2009 H1N1 and 1918 H1N1.
The cross-neutralization of this antibody is clearly demonstrated by
the hot spot residues common in the two binding interfaces. Our
comprehensive investigation suggests that there are six outstanding
epitope residues whose mutations will help H1N1 evade capture
by this antibody. We further pinpointed the hot spot residues at the
paratope site of the 2D1 antibody which are responsible for the
antigen recognition. The understanding of these hot spot residues
can potentially facilitate drug design to neutralize influenza viruses.
Funding: Singapore MOE Tier-2 funding grants (T208B2203 and
MOE2009-T2-2-004 in part).
Conflict of Interest: none declared.
REFERENCES
Bogan,A.A. and Thorn,K.S. (1998) Anatomy of hot spots in protein interfaces. J. Mol.
Biol., 280, 1–9.
Cho,K.-I. et al. (2009)Afeature-based approach to modeling protein-protein interaction
hot spots. Nucleic Acids Res., 37, 2672–2687.
Chothia,C. and Lesk,A.M. (1986) The relation between the divergence of sequence and
structure in proteins. EMBO J., 5, 823–826.
Clackson,T. and Wells,J. (1995) A hot spot of binding energy in a hormone-receptor
interface. Science, 267, 383–386.
Darnell,S.J.J. et al. (2008) KFC server: interactive forecasting of protein interaction hot
spots. Nucleic Acids Res., 36, W265–W269.
Ekiert,D.C. et al. (2009) Antibody recognition of a highly conserved influenza virus
epitope. Science, 324, 246–251.
Halperin,I. et al. (2004) Protein-protein interactions: coupling of structurally conserved
residues and of hot spots across interfaces. Implications for docking. Structure, 12,
1027–1038.
Igarashi,M. et al. (2010) Predicting the antigenic structure of the pandemic (H1N1)
2009 influenza virus hemagglutinin. PLoS One, 5, e8553.
Kinjo,A.R. and Nakamura,H. (2010) Geometric similarities of protein-protein interfaces
at atomic resolution are only observed within homologous families: an exhaustive
structural classification study. J. Mol. Biol., 399, 526–540.
2535
atMasarykUniversityonSeptember16,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[10:05 25/8/2011 Bioinformatics-btr437.tex] Page: 2536 2529–2536
Q.Liu et al.
Kortemme,T. and Baker,D. (2002)Asimple physical model for binding energy hot spots
in protein-protein complexes. Proc. Natl Acad. Sci. USA, 99, 14116–14121.
Krause,J.C. et al. (2010) Naturally occurring human monoclonal antibodies neutralize
both 1918 and 2009 pandemic influenza a (H1N1) viruses. J. Virol., 84,
3127–3130.
Li,J. and Liu,Q. (2009) ‘Double water exclusion’: a hypothesis refining the O-ring
theory for the hot spots at protein interfaces. Bioinformatics, 25, 743–750.
Li,Z. and Li,J. (2010) Geometrically centered region: a “wet” model of protein binding
hot spots not excluding water molecules. Proteins, 78, 3304–3316.
Schrödinger,LLC (2010) The PyMOL molecular graphics system, version 1.3r1.
Available at http://pymol.sourceforge.net/faq.html#CITE (last accessed date April
2011).
Schwede,T. et al. (2000) Protein structure computing in the genomic era. Res.
Microbiol., 151, 107–112.
Schymkowitz,J. et al. (2005) The FoldX web server: an online force field. Nucleic Acids
Res., 33, W382–W388.
Sheinerman,F.B. and Honig,B. (2002) On the role of electrostatic interactions in the
design of protein-protein interfaces. J. Mol. Biol., 318, 161–177.
Vilalta,R. and Drissi,Y. (2002) A perspective view and survey of Meta-Learning. Art.
Intell. Rev., 18, 77–95.
Xu,R. et al. (2010) Structural basis of preexisting immunity to the 2009 H1N1 Pandemic
Influenza Virus. Science, 328, 357–360.
Yu,X. et al. (2008) Neutralizing antibodies derived from the B cells of 1918 influenza
pandemic survivors. Nature, 455, 532–536.
2536
atMasarykUniversityonSeptember16,2011bioinformatics.oxfordjournals.orgDownloadedfrom