Vol.2, No.11, 1225-1246 (2010) Natural Science
http://dx.doi.org/10.4236/ns.2010.211150
Host markers and correlated mutations in the
overlapping genes of influenza viruses: M1, M2; NS1,
NS2; and PB1, PB1-F2
Wei Hu
Department of Computer Science, Houghton College, Houghton, New York, USA; [e-mail]
Received 10 August 2010; revised 13 September 2010; accepted 18 September 2010.
ABSTRACT
The influenza A viruses have three gene seg-
ments, M, NS, and PB1, which code for more
than one protein. The overlapping genes from
the same segment entail their interdependence,
which could be reflected in the evolutionary
constraints, host distinction, and co-mutations
of influenza. Most previous studies of overlap-
ping genes focused on their unique evolutionary
constraints, and very little was achieved to as-
sess the potential impact of the overlap on other
biological aspects of influenza. In this study, our
aim was to explore the mutual dependence in
host differentiation and co-mutations in M, NS,
and PB1 of avian, human, 2009 H1N1, and swine
viruses, with Random Forests, information en-
tropy, and mutual information. The host markers
and highly co-mutated individual sites and site
pairs (P values < 0.035) in the three gene seg-
ments were identified with their relative signi-
ficance between the overlapping genes calcu-
lated. Further, Random Forests predicted that
among the three stop codons in the current
PB1-F2 gene of 2009 H1N1, the significance of a
mutation at these sites for host differentiation
was, in order from most to least, that at 12, 58,
and 88, i.e., the closer to the start of the gene the
more important the mutation was. Finally, our
sequence analysis surprisingly revealed that the
full-length PB1-F2, if the three stop codons were
all mutated, would function more as a swine
protein than a human protein, although the PB1
of 2009 H1N1 was derived from human H3N2.
Keywords: 2009 H1N1; Co-Mutation; Correlation;
Host Marker; Influenza; Information Entropy; Mutual
Information; Mutation; Overlapping Genes; Random
Forests
1. INTRODUCTION
The influenza A virus genome comprises eight sepa-
rate gene segments. All segments code for one protein
except the M, NS and PB1 segments. The M gene seg-
ment encodes a structural matrix M1 protein and an ion
channel M2 protein, and the NS gene segment encodes a
nonstructural protein NS1 and a nuclear export protein
NS2. Together, NS2 and M2 export the viral ribonucleo-
protein complexes from the nucleus into the cytoplasm
for assembly into new virions. The PB1 gene segment
encodes the PB1 protein, a polymerase subunit, and a
nonstructural protein PB1-F2. The 2009 H1N1 virus has
truncated PB1-F2 and NS1, a feature worthy of further
exploration. A gene segment that can express multiple
proteins has evolved as a mechanism for producing more
proteins from the same genome length and for reducing
neutral mutations in a gene. It is apparent that the over-
lapping genes whose coding sequences overlap partially
or entirely in the same nucleotide sequence have to coo-
perate in their evolutionary constraints, host adaption,
and co-mutations. Naturally, this interplay between them
adds complexity as well as fascination to their studies.
M2 is involved in viral entry, virus assembly and bud-
ding. New evidence indicated that the M2 protein based
vaccines could confer cross-subtype protection [1]. This
protein is also the main target for antiviral drugs of the
adamantanes class, whose action is blocking the ion
channel activity of M2. In particular, two M2 inhibitors
(adamantine and rimantadine) affect two steps in the
replication cycle, viral uncoating and viral maturation.
There are five known adamantine-resistanant mutations
in M2 (L26F, V27A, A30V, A30T, S31N, and G34E).
The 2009 pandemic H1N1 virus contains a adaman-
tine-resistant mutation S31N in its M2 protein, thus
making the NA Inhibitors oseltamivir and zanamivir the
only options available to treat the infections caused by
the pandemic virus [2]. It also contains a mutation L43T
in M2, which is not present in seasonal, triple-reassortant
swine or H5N1 influenza viruses. The replacement of

W. Hu / Natural Science 2 (2010) 1225-1246
1226
the non-polar residue L43 by the polar residue T43 in
M2 may influence a nearby functional residue W44, the
channel lock and the binding site of rimantadine [3].
NS1 is involved in suppressing antiviral interferon
(IFN) induction during viral replication by preventing
activation of the latent transcription factors IRF-3 and
NF-κB [4], and thus is commonly regarded as a key fac-
tor by which all influenza viruses antagonize host im-
mune responses. Four C-terminal residues of NS1 could
influence pathogenicity [5], and typically avian viruses
have C-terminal ESEV (PDZ domain ligand motif, resi-
dues 227–230), while humans have RSKV [6]. The NS1
of 2009 H1N1 is truncated due to a stop codon at 220
and therefore lacks this PDZ ligand motif. Using reverse
genetics, the NS1 of A/California/04/09 (H1N1) was
restored to its full length and it had no impact on virus
replication and minimal effects on replication, pathoge-
nicity, and transmission [7]. The NS1 of 2009 H1N also
misses some other virulence factors, such as deletion of
residues 80-84 [8], a glutamic acid at position 92 [9].
PB1-F2 is a recently discovered protein of the in-
fluenza A virus. It is present in the nucleus, cytosol, and
the mitochondria membrane of infected cells. Structural
and functional analysis suggested that the biological
function of PB1-F2 is primarily related to its direct inte-
raction with membranes [10]. It can cause host immune
cell death, induce inflammation by recruitment of in-
flammatory cells, bind to PB1, increase the activity of
the virus polymerase, elicit an antibody response, and
form protein ion channels with no noticeable selectivity
in membranes [10-12]. In contrast to previous pandemic
viruses (1918 H1N1, 1957 H2N2, and 1968 H3N2), The
PB1 gene segment of 2009 H1N1 codes for a truncated
PB1-F2 protein. Full-length PB1-F2 proteins are com-
monly found in almost all avian influenza viruses, but
often become truncated during adaptation to mammalian
hosts [13]. A shorter PB1-F2, missing the mitochondrial
targeting sequence located at its C terminus, may result
in less efficient mitochondrial localization and PB1
binding, thus leading to less virus virulence. The low
pathogenicity of 2009 H1N1 could be attributed, in part,
to the absence of full-length PB1-F2 protein.
A specific amino acid (serine) at position 66 of
PB1-F2 (97 in PB1 numbering) was found correlated
with pathogenicity. S66 appeared in the three previous
pandemic influenza viruses, 1918 H1N1, 1957 H2N2,
and 1968 H3N2 [14,15], however, the significantly less
virulent strains in human, and swine typically had a N at
66. To change from low to high pathogenicity (from N to
S), a mutation a --> g was required at the second posi-
tion in codon 66. By generating two strains of 2009
H1N1 with productive PB1-F2 open reading frames
containing either an N at position 66 of PB1-F2 or a S at
position 66, one report found that PB1-F2 expression by
the these two strains could modulate the immune re-
sponse to infection while having a minimal effect on
virulence [16]. Some evidence also suggested that the
function of this protein might not universal, i.e., the im-
pact of PB1-F2 on polymerase activity is strain specific
and does not affect pathogenicity [11].
The virulence of influenza is a multigenic trait. It de-
pends on the interaction between the virus and host, the
specific virulence markers in the genes of the virus, and
the level of immune protection in the hosts. There are
five basic virulence factors of influenza viruses [17]. The
first is the PB1-F2 protein that targets the virus to the
inner and outer mitochondrial membranes, causing
damages to the cells and then inducing intrinsic apopto-
sis and cell death. The second is the nonstructural pro-
tein, NS1, which inhibits alpha-interferon to shut down
host defense against influenza. The third and the fourth
involve the protease susceptibility of HA and HA-NA
ratio that is determined by the binding of HA to neura-
minic acid. Lastly, the ability of the host immune system
to recognize the two major antigenic constituents, HA
and NA, of the virus is another factor. When the in-
fluenza virus leaps from animal species to humans, it
could generate a novel virus of pandemic potential with
high virulence, and the pandemic 2009 H1H1 was one
such jump.
Most previous studies of overlapping genes focused
on the evolution of influenza. By assuming that selection
operates independently on each gene in an overlapping
pair, the two genes PB1-F2 and NS1, overlapping with
PB1 and NS2 respectively, were found to have high
nonsynonymous/synonymous rate ratio (dN/dS) indica-
tive of positive selection [18-21]. Subsequent to these
reports, a new model that accounts for different selection
pressures on the genes in the overlap was developed. It
found no evidence for positive selection in these two
genes, illustrating that ignoring the unique evolutionary
constraints on overlapping regions could results in a
false appearance of positive selection [22].
In this study, we sought to understand the other bio-
logical characteristics of the overlapping genes in the M,
NS, and PB1 gene segments of avian, human, 2009
H1N1, and swine viruses: host distinction and co-muta-
tions, by employing Random Forests, information theory
including entropy and mutual information. Recently,
Random Forests were applied successfully to the study
of individual proteins and genes to identify novel host
markers in the influenza viruses [23,24]. These host
markers allow distinction of human virus from avian,
2009 H1N1, and swine viruses, and help to elucidate the
mechanism for animal viruses to cross species barriers to
infect humans. Considering all sites from their native

W. Hu / Natural Science 2 (2010) 1225-1246
1227
gene segment in the current study made it possible to
investigate the relative significance and contribution of
the two overlapping genes in host classification and
co-variation. There are three interesting consequences of
overlap to note. First, a single host marker or mutation in
the overlapping region might affect both genes. Second,
a synonymous mutation for one gene might be non-
synonymous for the other. Third, a single mutation might
confer a beneficial effect on one gene and the opposite
on the other. All of these could bring in a challenge to
the understanding of overlapping genes.
As mentioned above, an intriguing feature of the 2009
pandemic H1N1 virus was that its PB1 gene segment has
three stop codons in the second open reading frame that
cause the PB1-F2 protein truncated and nonfunctional.
However, the concern is that this novel virus might mu-
tate so its PB1-F2 protein will be functional. Our study
addressed this issue by quantifying the importance of
several key sites in the PB1-F2 gene that has the poten-
tial to change the current stop codons to produce a
full-length PB1-F2, thus affecting the virulence of this
virus.
2. MATERIALS AND METHODS
2.1. Sequence Data
The amino acid and nucleotide sequences of influenza
A virus employed in this study were downloaded from
the Influenza Virus Resource of the National Center for
Biotechnology Information (NCBI). All the sequences
utilized in the study were aligned with MAFFT [25].
2.2. Entropy and Mutual Information
In information theory [26], entropy is a measure of the
uncertainty associated with a random variable. Let
x
be a discrete random variable that has a set of possible
values
123
{,,,}
n
aaa a…
with probabilities
123
{,,,}
n
pppp…
where
( )
ii
Px ap= =
. The entropy H
of
x
is
( )
i
i
Hx p= −
∑
Log
i
p
The mutual information of two random variables is a
quantity that measures the mutual dependence of the two
variables or the average amount of information that
x
conveys about , which can defined as
()( )()
,(, )I xyH xHyHxy=+−
where
( )
Hx
is the entropy of , and
( )
,H xy
is the
joint entropy of
x
and
y
.
( )
,0I xy=
if and only if
x
and
y
are independent random variables.
In the current study, each of the N columns in a mul-
tiple sequence alignment of a set of influenza protein or
gene sequences of length N is considered as a discrete
random variable  (1 ≤i ≤N) that takes on one of the
20 (n=20) amino acid types or 4 (n=4) nucleotide types
with some probability.
( )
i
Hx
has its minimum value 0
if all the amino acids or nucleotides at position i are the
same, and achieves its maximum if all the amino acid or
nucleotide types appear with equal probability at posi-
tion i, which can be verified by the Lagrange multiplier
technique. A position of high entropy means that the
amino acids or nucleotides are often varied at this posi-
tion. While
( )
i
Hx
measures the genetic diversity at
position i in our current study,
(, )
ij
Ix y
measures the
correlation between amino acid or nucleotide substitu-
tions at positions i and j.
2.3. Mutual Information Evaluation
In order to assess the statistical significance of the
mutual information value of two positions in a multiple
sequence alignment, it is necessary to show that this
value is significantly higher than that based on random
sequences. For each pair of positions in a multiple pro-
tein/gene sequence alignment, we randomly permuted
the amino acids/nucleotides from different sequences at
the two positions and calculated the mutual information
of these random sequences. This procedure was repeated
1000 times. The P value was calculated as the percentage
of the mutual information values of the permuted se-
quences that were higher than those of the original se-
quences.
2.4. Random Forests
Random Forest, proposed by Leo Breiman in 1999
[27], is an ensemble classifier based on many decision
trees. Each tree is built on a bootstrap sample from the
original training set and is unpruned to obtain low-bias
trees. The variables used for splitting the tree nodes are a
random subset of the whole variable set. The classifica-
tion decision of a new instance is made by majority vot-
ing over all trees. About one-third of the instances are
left of the bootstrap sample and not used in the construc-
tion of the tree. These instances in the training set are
called “out-of-bag” instances and are used to evaluate
the performance of the classifier, which can achieve both
low bias and low variance with bagging and randomiza-
tion.
2.5. Feature Selection Using Random
Forests
Random Forest calculates several measures of varia-
ble importance. The mean decrease in accuracy measure
was employed in [28] to rank the importance of the fea-
tures in prediction. This measure is based on the de-
crease of classification accuracy when values of a varia-

W. Hu / Natural Science 2 (2010) 1225-1246
1228
ble in a node of a tree are permuted randomly. In this
study, two packages of R, randomForest and varSelRF
[28], were utilized to compute the importance of the
nucleotides in a given gene sequence dataset. The effec-
tiveness and robustness of this technique as a feature
selection method has been demonstrated in various stu-
dies [29-34].
Random Forests produce non-deterministic outcomes.
To compensate this bias, the Random Forests algorithm
was run multiple times and then the average of the re-
sults was taken. The importance of each position in the
nucleotide sequences was based on the averaged calcula-
tions by using the function randomVarImpsRF in var-
SelRF repeated 5 times.
3. RESULTS
Before presenting the main findings of this work, it
was instructive to render some relevant sequence infor-
mation about the M, NS, and PB1 segments. The first
piece of such information was the Hamming distances of
M, NS, and PB1 of influenza viruses based on their
consensus nucleotide sequences, which measured the
similarities of the three gene segments. Notably, the dis-
tance information in Table 1 was in agreement with the
origins of the M, NS, and PB1 of 2009 H1N1 [35]. The
Hamming distances of individual proteins or genes of
influenza viruses were determined in [23,24].
The second piece of such information was mutation
rate. The host markers identified in [23,24] were either
synonymous or non-synonymous mutations between
human and avian, 2009 H1N1, and swine viruses. To
gain a global view of these mutations, synonymous and
non-synonymous rates (dS and dN) between human and
the other three viruses based on their consensus se-
quences were computed (Table 2). In general dS were
much higher than dN in all the genes. However, both dN
and dS of NS1 were very high in avian-human compari-
son, and dS of PB1-F2 was all zero in all avian-human,
2009 H1N1-human, and swine-human comparisons.
3.1. Host Markers in the M, NS, and PB1
Gene Segments
Random Forests were applied to all sites in each gene
segment to find host markers in the overlapping genes,
with their relative importance calculated. These host
markers allow distinction of human virus from avian,
2009 H1N1, and swine viruses. The distribution of top
host markers in the two overlapping genes demonstrated
the relative significance of each gene in host classifica-
tion, which could not be revealed if each gene was ana-
lyzed individually. Using the whole gene segment, sev-
eral new sites were detected when compared to those
found from each individual gene in [24]. It also con-
firmed that the most important sites found from individ-
ual genes remained the most important when measured
from the whole gene segment.
3.1.1. M Segment
M1 is an internal structural protein that forms a
shell-like structure beneath the viral envelope. Unlike
M1, M2 is a multifunctional protein, which has an ecto-
domain (residues 1-24) important for incorporation into
budding virus particles, the trans-membrane helix (resi-
dues 25-43), an amphiphilic C-terminal helix (residues
47-60), and a cytoplasmic tail (residues 61-97), facilitat-
ing virus assembly by binding to the M1 protein [2]. The
N-terminal region of the M gene segment affects viral
replication efficiency, and synonymous mutations intro-
duced there cause an increased reduction in virus repli-
cation [36].
The top 40 host markers in the M segment of different
viruses were discovered by Random Forests (Figure 1).
In comparison of the relative significance of M1 and M2
Table 1. Hamming distances of nucleotide sequences of M, NS, and PB1 of different origins. The minimum distance in each segment
is marked with an asterisk.
Gene
Dist(Avian, 2009_H1N1)
Dist(Human, 2009_H1N1) Dist(Swine, 2009_H1N1) Dist(Avian, Human) Dist(Avian, Swine)
Dist(Human, Swine)
M
71*
127
107
95
76
91
NS
131
153
61*
109
95
121
PB1
256
118*
184
215
158
152
Table 2. dN and dS of the overlapping genes in M, NS, and PB1 of different origins.
Gene M1 M2 NS1 NS2 PB1 PB1-F2
dN dS dN dS dN dS dN dS dN dS dN dS
Avian/Human 0.0176 0.5158 0.0645 0.2781 1.2508 2.5035 0.0230 0.4898 0.0077 0.5857 0.0887 0
2009 H1N1/Human 0.0364 0.6304
0.0740
0.3791
0.1124
0.6415
0.0504
0.7927
0.0085
0.2364
0.0872
0
Swine/Human 0.0140 0.3232 0.0456 0.3222 0.0944 0.4267 0.0302 0.5485 0.0062 0.3545 0.0738 0

W. Hu / Natural Science 2 (2010) 1225-1246
1229
Figure 1. Top 40 host markers in M of different origins. The positions in the figure are codons and the number in parenthesis is co-
don position, and positions marked with an asterisk are host markers found in the M1 or M2 protein individually in [24].
for host distinction in the M gene segment, the number
of sites selected in Figure 1 from avian, 2009 H1N1, and
swine viruses were considered. On average swine M2
contributed the most sites (18.55% of its sites), avian M2
the second (15.46%), and 2009 H1N1 M2 the last
(8.24%), and 2009 H1N1 M1 contributed the most sites
(12.70%), avian M1 the second (9.92%), and swine M1
the last (8.73%). The sites 103(3), 115(1), 115(3), 137(1),
137(3), 218(1), and 218(3) were shared by avian, 2009
H1N1, and swine. Further, sites 115, 137, and 218 had
two positions selected within the same codon, and sites
26 and 27 were selected as host markers in multiple in-
fluenza species.
Mutations at sites 26, 27, 30, 31, and 34, located in the
trans-membrane domain of M2, were found associated
with resistance to ion channel blockers such as adaman-
tane-base drugs. The M2 of 2009 H1N1 carries one such
mutation S31N [37] and another mutation C55F for en-
hanced transmission. Although these two sites were not
selected as host markers in 2009 H1N1, but site 55 was a
host marker in the avian virus. The outstanding host
markers selected in the M segment were: M1 115(1),
137(1), M2 54(2), 78(1) in avian virus, M1 28(3), 117(3),
166(3), 182(3), 214(3), and M2 43(2) in 2009 H1N1,
and M2 78(1) in swine virus (Figure 1).
The basic amino acids 101RKLKR105 of M1 were
involved in vRNP binding and nuclear localization [38],
and site 103(3) was within this motif with an amino acid
L at this site in avian, human, 2009 H1N1, and swine
viruses. The most significant site 43(2) in M2 of 2009
H1N1 in our analysis was also a host marker found in
[39]. Residues 22-46 formed the functional core of M2
proton-selective ion channel [2], and several of the top
host markers were in this region (Figure 1). The combi-
nation of two residues E14 and F55 in M2 was a deter-
minant of human-to-human transmission [40]. Random
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
*33(3)
*57(3)
84(3)
95(2)
97(3)
99(3)
*103(3)
106(3)
*115(1)
*115(3)
*120(3)
*121(1)
*137(1)
*149(3)
*150(3)
*152(3)
160(1)
162(3)
163(3)
*168(3)
*173(3)
*180(3)
209(3)
*218(1)
*218(3)
*20(2)
*25(3)
*26(3)
*47(3)
*51(3)
*53(3)
*54(2)
*55(3)
*57(1)
*58(3)
*69(3)
*77(3)
*78(1)
*86(2)
*93(2)
M1
M2
Importance
Top 40 important M positions separating avian from human viruses
0
0.0005
0.001
0.0015
0.002
0.0025
0.003
0.0035
*12(3)
*28(3)
*30(1)
*30(2)
*52(1)
*90(3)
*91(3)
94(3)
*103(3)
114(3)
*115(3)
116(1)
*117(3)
120(3)
137(3)
140(3)
*142(3)
*145(3)
147(3)
150(3)
157(3)
163(3)
*166(3)
167(3)
*182(3)
*186(3)
190(3)
*209(1)
*213(3)
*214(3)
*218(3)
232(3)
*13(2)
*27(3)
*28(1)
*29(3)
*43(2)
*50(3)
*71(3)
*77(2)
M1
M2
Importance
Top 40 important M positions separating 2009 pandemic H1N1 from human viruses
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
*24(3)
*33(3)
41(3)
51(3)
81(3)
*88(3)
97(3)
*100(3)
103(3)
106(3)
*115(1)
116(3)
*137(1)
*138(3)
*146(3)
156(3)
*173(3)
*180(3)
190(3)
209(3)
213(3)
*218(3)
*26(3)
*28(2)
*29(3)
45(3)
*54(2)
*57(1)
59(3)
62(3)
*67(3)
*68(3)
*78(1)
*79(1)
*86(2)
*89(2)
*93(2)
*95(2)
*96(1)
98(2)
M1
M2
Importance
Top 40 important M positions separating swine from human viruses

W. Hu / Natural Science 2 (2010) 1225-1246
1230
Forests revealed that these two sites 14 and 55 were host
markers in M2 of avian and swine viruses [24]. However,
when compared with all sites in the M segment in the
current study, these two sites were no longer significant
in host distinction (Figure 1).
Using nucleotide sequences, Random Forests were
able to identify synonymous mutations, which tend to
occur at the third codon position, since the first two po-
sitions are mainly determined by the amino acids they
encode. In nature, the proportions of nucleotide muta-
tions that are synonymous are ∼5%, 0%, and ∼70% for
the first, second, and third codon positions, respectively
[41]. Careful sequence examination suggested that syn-
onymous mutations composed of 45% of all the top 40
host markers in avian, 2009 H1N1, and swine viruses,
which was a sizable increase over those in the NS seg-
ment (see the subsection below). This finding was con-
sistent with the notation that the M segment is relatively
more conserved compared to other gene segments of
influenza viruses.
3.1.2. NS Segment
The multifunctional NS1 protein contains an
RNA-binding domain (residues 1-73), and its C-terminal
region (residues 74-237) contains the effector domain
that inhibits the maturation and exportation of the host
cellular antiviral mRNAs [42]. In avian, 2009 H1N1,
and swine viruses, multiple sites were selected in a nar-
row region of codons 78-90 (Figure 2), indicating their
functional relevancy in host distinction. The ami-
no-terminal region (residues 12-21) of the NS2 contains
a highly conserved nuclear export signal motif [3], and
sites 13(3) and 14(1) in Figure 2 were host markers in
this region. In contrast to the varied relative contribu-
tions of M1 and M2 to the host differentiation in the M
segment, the relative importance of NS1 and NS2 in the
NS segment appeared the same across all three species:
avian, 2009 H1N1, and swine viruses, with NS1 consis-
tently contributing more than NS2 (about 13% sites in
NS1 and about 8% in NS2) (Figure 2).
The outstanding host markers selected in the NS seg-
Figure 2. Top 40 host markers in NS of different origins. The positions in the figure are codons and the number in parenthesis is
codon position, and positions marked with an asterisk are host markers found in the M1 or M2 protein individually in [24].
0
0.005
0.01
0.015
0.02
0.025
0.03
21(2)
*22(1)
27(3)
34(3)
*50(3)
60(2)
60(3)
*70(1)
*79(2)
*79(3)
*80(3)
*81(3)
*82(3)
*84(2)
*84(3)
*87(3)
88(3)
95(1)
98(1)
*114(1)
125(3)
143(3)
146(1)
*147(3)
151(3)
*153(3)
*157(3)
158(3)
163(3)
166(1)
*14(1)
38(3)
*57(3)
*69(3)
*70(1)
*84(1)
*89(2)
*89(3)
*104(3)
*107(1)
NS1
NS2
Importance
Top 40 important NS positions separating avian from human viruses
0
0.001
0.002
0.003
0.004
0.005
2(3)
5(3)
*6(1)
*20(3)
25(3)
*43(1)
*52(3)
*53(3)
59(2)
*64(3)
67(1)
*68(3)
*69(1)
74(2)
78(2)
*80(3)
82(3)
86(1)
*90(3)
*91(1)
*100(1)
108(2)
112(2)
117(3)
118(1)
*119(1)
*128(3)
*129(1)
151(3)
*156(3)
*13(3)
*34(2)
*40(3)
*48(3)
*49(3)
*57(2)
*60(2)
*70(3)
*83(1)
89(1)
NS1
NS2
Importance
Top 40 important NS positions separating 2009 pandemic H1N1 from human viruses
0
0.002
0.004
0.006
0.008
0.01
0.012
18(1)
21(3)
*22(1)
*25(1)
*25(3)
49(3)
50(3)
56(1)
59(2)
*67(1)
73(3)
79(3)
*81(3)
*82(2)
*82(3)
*83(3)
*84(2)
85(3)
88(3)
91(1)
95(1)
112(1)
114(1)
117(3)
125(3)
136(3)
144(1)
*146(1)
163(3)
*14(1)
*32(1)
*48(3)
*57(2)
*57(3)
63(2)
*69(3)
*84(1)
*102(3)
*107(1)
117(3)
NS1
NS2
Importance
Top 40 important NS positions separating swine from human viruses

W. Hu / Natural Science 2 (2010) 1225-1246
1231
ment were: NS1 22(1), 50(3), 84(2), NS2 57(3) in avian
virus, NS1 90(3), 91(1), 129(1), NS2 13(3) in 2009
H1N1, NS1 22(1), NS2 57(3) in swine virus. Sites
82(3)in NS1 and 57(2) and 57(3) in NS2 were shared
among avian, 2009 H1N1, and swine influenza. Among
the several dominant mutations found in 2009 H1N1 in
[4], a unique dominant mutation I123V occurred in the
NS1 protein. This mutation is located in the effector
domain of NS1, and its potential role in virulence and
host adaptation needs to be clarified. Mutation G92E in
avian H5N1 NS1 is known to promote the inhibition of
host immune responses [43], but was not selected as a
host marker in this analysis.
Sequence survey indicated that the synonymous muta-
tions occurred at the top 40 host markers in the NS seg-
ment comprised about 1/3 of all the top 40 sites in avian,
2009 H1N1, and swine viruses, thus rendering new and
subtle host markers that complemented the non-synony-
mous mutations identified using amino acid sequences in
[23]. Notably, there were a cluster of synonymous muta-
tions occurring within the codons 79 -- 88.
The identity of the nucleotide sequence of 2009 H1N1
NS1 is closest to swine NS1, and both of them have a
stop codon (tga) at 220 while avian and human NS1
genes have cgg(R) at 220, showing two mutations at
220(1) and 220(3) (62(3) and 63(2) in PB1-F2 number-
ing). Additionally, swine NS1 220(3) (63(2) in NS2
numbering) was selected as a host marker in the NS
segment in the current analysis. Interestingly, when con-
sidering the sites in NS1 and NS2 individually, NS1 sites
220(1) and 220(3) was not selected as a important host
marker in NS1, however, NS2 62(3) of 2009 H1N1 and
swine viruses was a host marker in NS2 [24]. This phe-
nomena suggested that a site in the overlapping region
might be significant for one gene, but not for the other.
Four C-terminal residues of the NS1 protein of highly
pathogenic avian virus form a potential PDZ ligand
binding motif of the ESEV type. By contrast, the NS1
proteins in most lowly pathogenic human viruses contain
a different motif (RSKV), which cannot bind
PDZ-containing proteins. The C-terminus of NS1 pro-
tein of 2009 H1N1 has a motif GTEI, similar to the
swine motif GSEI (Table 3). Thus, the 2009 H1N1 may
exhibit low pathogenicity [37]. The Hamming distances
of 2009 H1N1 motif to the motif of other species were:
dist (2009 H1N1, avian) = 3, dist(2009 H1N1, human) =
4, and dist(2009 H1N1, swine) = 1, reflecting the swine
origin of the NS of 2009 N1N1 in this respect.
3.1.3. PB1 Segment
The PB1 protein interacts with the PB2 and PA pro-
teins to form the viral polymerase, which is responsible
for the transcription and replication of the viral genome.
However, the PB1-F2 protein originating from an alter-
native reading frame in PB1 has completely different
functions. It has proapoptotic properties and plays a role
in the virulence of influenza. PB1 has five functional
domains including NLS and cRNA, vRNA, PA, and PB2
binding domains. Residues 1--25 and 600 -- 757 of PB1
bind to PA and PB2 respectively. In all the three in-
fluenza species, there was one site selected in the PA
binding region and 7-9 sites in the PB2 binding region.
Additionally, swine virus had a larger number of host
markers (9 of them) in the codons 300 -- 400 that were
in the cRNA binding domain of PB1, while avian and
2009 H1N1 viruses had only three host markers in the
same region (Figure 3).
The outstanding host markers selected in the PB1
segment were: 322(3) in avian virus, 62(3), 167(1),
435(2), 601(3) in 2009 H1N1, 60(3) in swine virus. The
PB1 protein of 2009 H1N1 appeared to have more out-
standing host markers than avian and swine viruses
(Figure 3), implying that the novel virus of 2009 H1N1
is still experiencing genetic drifting to make its human
host adaption stable. There were several common hast
markers in different viruses. Site 60(3) (29(2) in PB1-F2
numbering) was selected in avian, 2009 H1N1, and
swine viruses, and site 81(3) (50(2) in PB1-F2 number-
ing) was selected in 2009 H1N1and swine viruses.
In the PB1 gene segment, the 2009 H1N1 PB1-F2 had
more sites selected than avian and swine viruses (Figure
3), suggesting its critical role in host distinction for 2009
H1N1. It further demonstrated that the important host
markers found in the PB1-F2 sequences were also sig-
nificant in the whole PB1 sequences, though only few of
them were selected competitively.
Among the three stop codons in PB1-F2 of 2009
H1N1, two of them were selected as host markers in the
whole PB1 segment: 43(3) and 89(3) (12(2) and 58(2) in
PB1-F2 numbering), implying that the relative signific-
ance of host markers changed when measured from
PB1-F2 to PB1. In 2009 H1N1 PB1-F2, 12(2) and 58(2)
were the top two sites (Figure 4), but of the eight sites in
PB1-F2 selected from PB1, the top two sites were 62(3)
and 102(3) (31(2) and 71(2) in PB1-F2 numbering)
(Figure 3).
Table 3. Consensus nucleotides and amino acids at the PDZ ligand binding motif of different origins.
Avian
Human
2009 H1N1
Swine
Amino acids
ESEV
RSKV
GTEI
GSEI
Nucleotides
gag tca gaa gtt
agg tca aaa gtt
ggg aca gaa att
ggg tca gaa att

W. Hu / Natural Science 2 (2010) 1225-1246
1232
Figure 3. Top 40 host markers in PB1 of different origins. The positions marked with an asterisk are also found important in the
indivdual PB1-F2 protein in Figure 4.
3.1.4. PB1-F2 Gene
The newly discovered PB1-F2 is a proapoptotic pro-
tein. A specific amino acid (serine) at position 66 of
PB1-F2 (97 in PB1 numbering) was found correlated
with pathogenicity as described in the Introduction sec-
tion. In our analysis, Random Forests detected site 66(2)
as an important site for host distinction in avian, but not
in 2009 H1N1, and swine viruses (Figure 4).
The residues 46 to 75 of PB1-F2 were shown to be
both necessary and sufficient for mitochondrial targeting
[3], and sites 60(2) in avian, 58(2) in 2009 H1N1, and
71(2) in swine viruses were the most significant host
markers in this region. The number of host markers se-
lected in the PB1-F2 gene from the PB1 segment varied
in avian (one selected), 2009 H1N1 (eight selected), and
swine (five selected) viruses (Figure 3), illustrating the
relative importance of PB1-F2 in the PB1 segment of
2009 H1N1.
The avian and swine viruses had about four outstand-
ing host markers respectively, whereas 2009 H1N1 dis-
played about 3 such markers (Figure 4). There were four
sites in PB1-F2 shared by avian, 2009 H1N1, and swine:
29(2), 50(2), 60(2), and 82(2), reflecting their signific-
ance in host adaption (Figure 4). At the same time, the
sites 29(2) and 60(2) were the most important site in
swine and avian respectively, and 29(2) (60(3) in PB1
numbering) was also an important site when considering
all sites in the PB1 segment of avian, 2009 H1N1, and
swine viruses (Figure 3), and site 50(2) (81(3) in PB1
numbering) was selected both in 2009 H1N1 and swine
viruses.
Currently, there are three translation stop codons in
the PB1-F2 gene of 2009 H1N1 at positions 12, 58, and
88, preventing PB1-F2 protein expression. However,
0
0.005
0.01
0.015
0.02
0.025
10(3)
*60(3)
148(3)
211(1)
228(3)
232(3)
237(3)
240(3)
257(3)
259(3)
271(3)
275(3)
280(3)
298(3)
307(3)
322(3)
327(2)
333(3)
336(1)
337(1)
341(3)
361(3)
389(3)
401(3)
426(3)
439(3)
454(3)
502(3)
576(3)
581(3)
589(3)
607(3)
624(3)
661(3)
668(3)
695(1)
726(3)
727(1)
757(1)
757(2)
PB1
PB1-F2
PB1
Importance
Top 40 important PB1 positions separating avian from human viruses
0
0.0005
0.001
0.0015
0.002
0.0025
0.003
0.0035
0.004
12(1)
*43(3)
*60(3)
*62(3)
*65(3)
*81(3)
*89(3)
*102(3)
*105(3)
167(1)
167(3)
175(1)
183(3)
198(3)
212(1)
224(3)
225(3)
241(3)
244(3)
247(3)
250(3)
273(3)
339(3)
364(1)
364(3)
435(2)
443(3)
509(3)
541(3)
543(3)
563(3)
576(1)
587(2)
601(3)
618(3)
638(3)
664(3)
728(1)
728(3)
735(3)
PB1
PB1-F2
PB1
Importance
Top 40 important PB1 positions separating 2009 pandemic from human viruses
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
7(1)
58(3)
*60(3)
*65(3)
*81(3)
*102(3)
167(3)
191(3)
216(3)
217(3)
225(3)
234(3)
239(3)
271(3)
273(3)
275(3)
298(3)
299(3)
327(2)
330(3)
339(3)
341(3)
360(3)
361(2)
367(3)
379(3)
385(3)
473(3)
541(3)
566(3)
584(3)
592(3)
624(3)
627(3)
638(3)
654(3)
677(3)
695(1)
726(3)
757(2)
PB1
PB1-F2
PB1
Importance
Top 40 importance PB1 positions separting swine from human viruses

W. Hu / Natural Science 2 (2010) 1225-1246
1233
Figure 4. Top 20 host markers in PB1-F2 of different origins
The positions in the figure are codons and the number in pa-
renthesis is codon position.
PB1-F2 production could take place due to genetic mu-
tations. Thus, it is critical to investigate the potential
mutations that could influence the expression of the
PB1-F2 protein of 2009 H1N1. Random forests indi-
cated that the current three stop codons at 12, 58, and 88
in the PB1-F2 gene of 2009 H1N1 were crucial for its
adaption to humans, with condons 12 and 58 being the
two most significant sites when comparing them with
other host markers in PB1-F2 (Figure 4). Further, in
order from most to least, the importance of these three
sites was as follows: 12, 58, and 88, implying that the
closer to the start of the gene the more important the site
became. The avian and swine PB1-F2 shared the same
amino acid as human at these three sites (S at 12, W at
58 and 88), and therefore these sites were not considered
important for host differentiation by Random Forests.
The advantages of using nucleotide sequences over
amino acid sequences are their ability to detect syn-
onymous mutations and to pinpoint which position
within a codon is important for host distinction. The
consensus nucleotides and amino acids at the top 20 host
markers identified by Random Forests were found (Ta-
ble 4). Table 4 showed that 2009 H1N1 had taa(stop
codon) at 12, and it took a mutation a --> c at the second
position in the codon to became tca(S) as avian, human,
and swine viruses had at 12. Also 2009 H1N1 had
tag(stop codon) at 58, and it took a mutation a --> g at
the second position in the codon to make tgg(W) as
avian, human, and swine had at 58. Finally, 2009 H1N1
had tag (stop codon) at 88, and it took a mutation a --> g
to make tgg(W) as avian, human, and swine had at 88.
Different amino acids in the PB1-F2 of avian, human,
2009 H1N1, and swine viruses were identified (Table 5).
To further elucidate the origin of 2009 H1N1 PB1-F2, a
new strain from a recent human infection with
triple-reassortant swine (H1N1) virus (A/Saskatchewan/
5131/2009(H1N1)), termed as (triple/2009), was in-
cluded. The Hamming distances of PB1-F2 of different
origins (Table 6) suggested that the PB1-F2 of 2009
H1N1 was closer to swine than to avian and human vi-
ruses despite its human H3N2 origin [44], and the
triple/2009 was the closest to swine. If all the three mu-
tations occurred at the three stop codons in PB1-F2 of
2009 H1N1, then the distance of PB1-F2 of 2009 H1N1
to that of avian, human, and swine viruses will be re-
duced by three in each case. The novel sites of PB1-F2
of 2009 H1N1 that contained amino acids different from
the same amino acids at these sites of other species were
18, 22, 50, and 57. Lastly, the novel sites of triple/2009
were also of interest, and they were 5, 44, and 55.
The PB1-F2 of 2009 H1N1 is expected to be similar
to that of humans, because the PB1 of 2009 H1N1 was
of human H3N2 lineage [35], which could also be con-
firmed by its GC content (Figure 5). Surprisingly, the
GC content of PB1-F2 of 2009 H1N1 was more similar
to swine than to human viruses (Figure 6). We subse-
quently mutated the three stop codons in the 2009 H1N1
PB1-F2, and then found out that the GC content of the
mutated gene (full-length) was very similar to swine
(Figure 7). To further explore this peculiarity, PB1 of
A/swine/Indiana /9K035 /19 99 (Swine/1999) was used,
which is closest to that of A/California/04/2009 (H1N1)
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
3(2)
23(2)
25(2)
29(2)
31(2)
32(2)
41(3)
42(2)
50(2)
57(2)
57(3)
59(2)
60(2)
66(2)
70(2)
73(2)
76(2)
79(2)
82(2)
87(2)
Importance
Positions
Top 20 important PB1-F2 positions separating
avian from human viruses
0
0.01
0.02
0.03
0.04
0.05
0.06
12(2)
18(2)
22(2)
27(2)
29(2)
31(2)
34(2)
37(2)
47(2)
50(2)
57(2)
58(2)
60(2)
62(2)
70(2)
71(2)
74(2)
82(2)
88(2)
89(2)
Importance
Positions
Top 20 important PB1-F2 positions separating
2009 pandemic H1N1 from human viruses
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
21(2)
22(2)
23(2)
29(2)
34(2)
35(2)
37(2)
44(2)
45(2)
47(2)
49(2)
50(2)
60(2)
62(2)
71(2)
74(2)
81(3)
82(2)
87(2)
89(2)
Importance
Positions
Top 20 important PB1-F2 positions separating
swine from human viruses

W. Hu / Natural Science 2 (2010) 1225-1246
1234
Table 4. Consensus nucleotides and amino acids at the top 20 host markers in PB1-F2 of avian, 2009 H1N1, and swine viruses. The
positions in the table are codons and the number in parenthesis is codon position.
Position 3(2) 12(2) 18(2) 21(2) 22(2) 23(2) 25(2) 27(2) 29(2) 31(2) 32(2) 34(2) 35(2)
Avian cag(Q) tca(S) att(I) agg(R) gga(G) aat(N) caa(Q) aca(T) aga(R) gag(E) cac(H) aac(N) tca(S)
Huma n cag(Q) tca(S) att(I) agg(R) gga(G) agt(S) cga(R) ata(I) aaa(K) ggg(G) cac(H) aac(N) tca(S)
2009 H1N1 cag(Q) taa(stop) act(T) agg(R) gaa(E) agt(S) cga(R) aca(T) aga(R) gtg(V) cac(H) agc(S) tca(S)
Swine cag(Q) tca(S) att(I) aag(K) gga(G) aat(N) cga(R) aca(T) aga(R) ggg(G) cac(H) agc(S) tca(S)
Position 37(2) 41(3) 42(2) 44(2) 45(2) 47(2) 49(2) 50(2) 57(2) 57(3) 58(2) 59(2) 60(2)
Avian caa(Q) cac(H) tac(Y) agg(R) ata(I) agc(S) gtg(V) gat(D) tcc(S) tcc(S) tgg(W) aag(K) caa(Q)
Huma n caa(Q) cac(H) tac(Y) agg(R) ata(I) agc(S) gtg(V) gat(D) tcc(S) tcc(S) tgg(W) agg(R) cta(L)
2009 H1N1 cga(R) cac(H) tac(Y) agg(R) ata(I) aac(N) gtg(V) ggt(G) ttc(F) ttc(F) tag(stop) agg(R) cta(L)
Swine cga(R) cac(H) tac(Y) agg(R) ata(I) aac(N) gtg(V) gat(D) tcc(S) tcc(S) tgg(W) agg(R) cca(P)
Position 62(2) 66(2) 70(2) 71(2) 73(2) 74(2) 76(2) 79(2) 81(3) 82(2) 87(2) 88(2) 89(2)
Avian ctt(L) agt(S) gaa(E) tct(S) aaa(K) act(T) gtc(V) cga(R) aag(K) ttg(L) gag(E) tgg(W) aca(T)
Huma n cct(P) aat(N) gga(G) tct(S) aga(R) act(T) gcc(A) caa(Q) aag(K) tcg(S) ggg(G) tgg(W) aca(T)
2009 H1N1 ctt(L) aat(N) gaa(E) tat(Y) aga(R) att(I) gcc(A) caa(Q) aag(K) ttg(L) ggg(G) tag(stop) ata(I)
Swine ctt(L) aat(N) gga(G) tat(Y) aga(R) att(I) gcc(A) caa(Q) aag(K) ttg(L) ggg(G) tgg(W) ata(I)
Table 5. Amino acid differences in the consensus sequences of PB1-F2 of different origins, where a 'X' represents the stop codon.
Table 6. Hamming distances of amino acid sequences of PB1-F2 of different origins.
Dist(Avian, 2009 H1N1) Dist(Human, 2009 H1N1) Dist(Swine, 2009 H1N1) Dist(Avian, Human) Dist(Avian, Swine)
25
20
13
18
20
Dist(Human, Swine) Dist(Triple 2009, Avian) Dist(Triple 2009, Human) Dist(Triple 2009, 2009 H1N1) Dist(Triple 2009, Swine)
15
24
17
17
4
Figure 5. AT/CG density curves of PB1 of different origins with sliding window size=sequence length/20.
among the triple-reassortants and has a complete PB1-F2
[45]. The PB1-F2 of Swine/1999 was very similar to the
full-length PB1-F2 of 2009 H1N1 (mutated version), and
their PB1s were similar as well (Figure 7). We also used
05001000 1500 20002500
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
Position
A-T C-G density (Avian PB1 Gene)
A-T
C-G
0500 1000 1500 2000 2500
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
Position
A-T C-G density (Human PB1 Gene)
A-T
C-G
0500 1000 1500 2000 2500
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
Position
A-T C-G density (2009 H1N1 PB1 Gene)
A-T
C-G
0500 1000 1500 2000 2500
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
Position
A-T C-G density (Swine PB1 Gene)
A-T
C-G
Position
5
6
12
16
18
21
22
23
25
27
29
31
34
37
44
47
50
55
Avian
Q
D
S
I
I
R
G
N
Q
T
R
E
N
Q
R
S
D
T
Human
Q
G
S
T
I
R
G
S
R
I
K
G
N
Q
R
S
D
T
2009 H1N1
Q
D
X
T
T
R
E
S
R
T
R
V
S
R
R
N
G
T
Swine
Q
D
S
T
I
K
G
N
R
T
R
G
S
R
R
N
D
T
2009 Triple
L
D
S
T
I
K
G
N
R
T
K
G
S
R
K
N
D
I
Position
57
58
59
60
62
66
70
71
73
74
75
76
79
82
83
87
88
89
Avian
S
W
K
Q
L
S
E
S
K
T
R
V
R
L
F
E
W
T
Human
S
W
R
L
P
N
G
S
R
T
H
A
Q
S
F
G
W
T
2009 H1N1
F
X
R
L
L
N
E
Y
R
I
H
A
Q
L
F
G
X
I
Swine
S
W
R
P
L
N
G
Y
R
I
H
A
Q
L
S
G
W
I
2009 Triple
S
W
R
P
L
N
G
Y
R
I
H
A
Q
L
S
G
W
I

W. Hu / Natural Science 2 (2010) 1225-1246
1235
Figure 6. AT/CG density curves of PB1-F2 of different origins with sliding window size=sequence length/20.
Figure 7. AT/CG density curves of PB1 and PB1-F2 of different origins with sliding window size=sequence length/20.
a consensus PB1 of all human H3N2 sequences in 1999
(Human H3N2/1999), which was similar to the PB1 of
Swine/1999 but its PB1-F2 was different (Figure 7). To
further confirm that the PB1-F2 of 2009 H1N1 is closest
to swine, we also used the PB1-F2 of A/Duck/Guangxi/
27/03 (Avian/2003), a highly pathogenic avian H5N1
virus [46] (Figure 7). After the comparison of PB1-F2 of
2009 H1N1 with that of other origins using consensus
sequences or specific strains, it seemed that the PB1-F2
of 2009 H1N1 was most similar to swine PB1-F2. Our
interpretation was that PB1-F2 is only a small portion of
PB1, and similar PB1s may not encode similar
PB1-F2s. The comparison of the GC content of 2009
H1N1 PB1-F2 with that of avian, human, and swine ori-
ginsimplied that if the PB1-F2 of 2009 H1N1 realized
the three mutations at the three stop codons at 12, 58,
and 88 to become full-length, then its function would be
more like a swine protein than a human protein.
After observing this unanticipated discrepancy between
the PB1 of 2009 H1N1 (of human H3N2 origin) and the
PB1-F2 of 2009 H1N1 (more like a swine PB1-F2), we
conducted a similar GC content analysis on NS. It showed
that the NS of 2009 H1N1 (of classical swine origin) had
its NS1 (NS2) closest to swine NS1 (NS2) when com-
pared to avian, human, and swine viruses.
3.2. Co-mutations in the M, NS, and PB1
Gene Segments
3.2.1. Highly Co-mutated Sites and Site Pairs in
M, NS, and PB1
In this section, mutual information (MI), a measure-
050 100 150 200 250 300
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Position
A-T C-G density (Avian PB1-F2 Gene)
A-T
C-G
050100 150 200 250 300
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Position
A-T C-G density (Human PB1-F2 Gene)
A-T
C-G
050100 150 200 250 300
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Position
A-T C-G density (2009 H1N1 PB1-F2 Gene)
A- T
C-G
050 100 150 200 250 300
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Position
A-T C-G density (Swine PB1-F2 Gene)
A- T
C-G
0500 1000 1500 2000 2500
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
Position
A-T C-G density (Swine 1999 PB1 Gene)
A-T
C-G
050100 150 200 250 300
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Position
A-T C-G density (Swine 1999 PB1-F2 Gene)
A-T
C-G
0500 1000 1500 2000 2500
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
Position
A-T C-G density (Human H3N2 1999 PB1 Gene)
A-T
C-G
050100 150200 250 300
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Position
A-T C-G density (Human H3N2 1999 PB1-F2 Gene)
A-T
C-G
050 100 150200250 300
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Position
A-T C-G density (2009 H1N1 PB1-F2 Gene, Three Stop Codons Mutated)
A-T
C-G
050 100 150 200 250 300
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Position
A-T C-G density (Avian 2003 PB1-F2 Gene)
A- T
C-G

W. Hu / Natural Science 2 (2010) 1225-1246
1236
ment of the correlation of two sites, was utilized to in-
vestigate the co-mutated sites and site pairs in the three
gene segments. To estimate how one site associates with
other sites, the MI values of this site with all other sites
in the same segment was summed, termed MI sum at
this site. As in the case of host marker identification, the
top 40 MI sites from each gene segment of avian, human,
2009 H1N1, and swine influenza were selected (Figures
8, 9, and 10). While the MI sum at a site represents how
well one site connects with others, it is also informative
to uncover the top highly co-mutated site pairs. For this
reason, the top 20 MI site pairs with smallest P values in
each gene segment were found (Tables 9, 11, and 13),
from which networks of co-mutated sites in each gene
segment could be inferred. The correlations between the
sites of 2009 H1N1 seemed lower than the other three
species, which resulted in multiple isolated networks of
highly co-mutated sites (Tables 10, 12, and 14).
Considering the percent of top MI sites selected from
each gene, it appeared that M1 contributed more to the
correlation of the M segment than M2 in avian and
swine viruses, M2 contributed more than M1 in human
viruses, and M1 and M2 made equal contribution in
2009 H1N1. However, as in the case of host distinction,
NS1 contributed more to the correlation of the NS seg-
ment than NS2 across all four influenza species. Finally,
the contribution of PB1-F2 to the correlation of PB1 was
the most prominent in 2009 H1N1 as in the case of host
classification (Figures 11 and 12).
To offer a comparative basis, the entropy of M, NS,
and PB1 was computed and its average over the whole
gene segment was compared to that of MI sum (Table 7).
In the M segment, M1 had a higher entropy than M2 of
avian, 2009 H1N1 and swine viruses, but the opposite
was found in humans. Further, the entropy of the over-
lapping region of M1 and M2 was high in avian, human,
and swine viruses, but was low in 2009 H1N1. In the NS
segment, the entropy of NS1 was always higher than
NS2 in all the four species, and the overlapping region
of NS1 and NS2 displayed an entropy comparable to that
of NS2. It appeared that the entropy of PB1 was higher
than that of PB1-F2 in all the four species.
Although entropy measures the variation of a single
site while MI measures the co-variation of two sites,
similar patterns of entropy and MI of M, NS, and PB1
were observed, which suggested that on average, when a
gene experienced high (low) sequence variation, its
co-variation tended to be high (low) as well. However,
there was no huge overlap between the host marker sites
and highly co-mutated sites within a gene, implying that
host distinction and co-mutation are two distinct features
of genes, and one complements the other in our view of
genes. Several sites were selected for both host distinc
tion and co-mutation, indicative of their significance. It
was remarkable that some selected codons contained one
position that was important for host differentiation, and
another for co-mutation.
In the beginning of Section 3, Hamming distances of
M, NS, and PB1 of different origins were calculated to
show the similarities of the nucleotide sequences of the
0
5
10
15
20
25
30
35
27(2)
41(3)
*84(3)
86(3)
*99(3)
101(2)
*103(3)
107(3)
112(3)
*115(3)
118(3)
119(3)
**121(3)
219(3)
136(3)
**137(3)
140(3)
143(3)
148(3)
*150(3)
155(3)
157(3)
166(2)
167(3)
*168(3)
183(3)
188(3)
193(3)
200(3)
201(3)
202(3)
205(3)
*218(3)
224(2)
225(3)
232(1)
*26(3)
28(1)
68(3)
96(1)
M1
M2
MI Sum
Top 40 MI sites in avian M gene segment
0
5
10
15
20
25
30
35
40
2(3)
10(3)
15(3)
16(3)
25(3)
35(3)
37(3)
(a*)84(3)
102(3)
127(3)
129(3)
(s*)138(3)
(p**)142(1)
(p**)142(2)
(s*)146(3)
(p**)147(1)
(a*,p*)150(3)
155(3)
159(3)
(a**)160(2)
167(1)
200(3)
229(3)
231(1)
239(1)
24(3)
(a*)25(3)
30(3)
35(3)
36(1)
(a*)51(3)
(a*)53(3)
(a**,s**)54(1)
(a*)55(3)
(a*)58(3)
(s*)59(3)
(a*)69(3)
(a*,s*)78(1)
88(3)
(s**)89(1)
M1
M2
MI Sum
Top 40 MI sites in human M gene segment

W. Hu / Natural Science 2 (2010) 1225-1246
1237
Figure 8. Top 40 MI sum sites in the M segment of different origins. The positions marked with one asterisk are host marker po-
sitions found in Section 3.1, and those marked with two asterisks are within the same codon of a host marker but different from
the host marker position. In the human M gene segment, a single letter “a” was attached to the position if it is a top avian MI sum
site, a “p” for pandemic 2009 H1N1, and a “s” for swine virus.
Table 7. Averaged entropy and MI sum of M, NS, and PB1.
Entropy Average
M
M1
M2
M1 and M2
NS
NS1
NS2
NS1 and NS2
PB1
PB1-F2
Avian
0.1221
0.1348
0.0750
0.0900
0.2575
0.2638
0.1825
0.1466
0.1564
0.1393
Human
0.0852
0.0789
0.0915
0.0380
0.1076
0.1030
0.0955
0.0715
0.1610
0.1399
2009 H1N1
0.0038
0.0042
0.0021
0
0.0048
0.0049
0.0034
0.0030
0.0036
0.0036
Swine
0.1239
0.1275
0.1087
0.1298
0.1912
0.1978
0.1450
0.1414
0.2114
0.1945
MI Average
M
M1
M2
M1 and M2
NS
NS1
NS2
NS1 and NS2
PB1
PB1-F2
Avian
2.35
2.61
1.35
1.54
27.08
28.05
18.33
15.56
7.59
6.23
Human
3.39
3.11
3.52
0.82
5.26
5.13
4.50
3.58
36.74
29.34
2009 H1N1
0.0114
0.0124
0.0070
0
0.0109
0.0118
0.0068
0.0042
0.0114
0.0146
Swine
4.71
4.87
3.85
4.07
9.53
10.04
6.81
6.59
31.98
27.33
Table 8. Pearson correlation coefficients of entropies or MI sums of avian, human, 2009 H1N1, and swine viruses. The
values marked with an asterisk are the minimum of the Pearson coefficients between human and avian, 2009 H1N1, and
swine viruses in each gene segment.
Entropy
Avian:2009 H1N1
Human:2009 H1N1
Swine:2009 H1N1
Avian:Human
Avian:Swine
Human:Swine
M 0.157843* 0.15122 0.126014 0.444767 0.712466 0.530262
NS 0.056013 0.023599 0.057873* 0.453913 0.676654 0.504818
PB1
0.066696
0.165166
0.167067*
0.334723
0.395312
0.805808
MI Avian:2009 H1N1 Human:2009 H1N1 Swine:2009 H1N1 Avian:Human Avian:Swine Human:Swine
M
0.143525*
0.123639
0.11097
0.405523
0.577274
0.47478
NS 0.074155 0.030983 0.075497* 0.320084 0.516167 0.465173
PB1 0.0619 0.145565* 0.138486 0.225252 0.248629 0.704232
three gene segments. Here the similarities of entropy and
MI patterns of the three segments were measured by
Pearson correlation coefficients (Table 8), which dem-
onstrated well the origin of 2009 H1N1 M, NS and PB1
except the entropy of PB1.
3.2.1.1. M Segment
A recent study on M gene identified sites of high se-
0
0.2
0.4
0.6
0.8
1
1.2
1.4
3(3)
13(2)
15(3)
16(3)
42(1)
45(1)
*52(1)
58(2)
58(3)
77(3)
**94(1)
98(1)
111(3)
113(2)
138(3)
**147(1)
**150(2)
164(3)
170(3)
185(3)
189(3)
200(3)
210(2)
215(3)
*218(3)
222(2)
227(1)
229(3)
*232(3)
24(3)
26(3)
34(3)
38(1)
54(2)
55(1)
57(3)
64(2)
68(3)
91(3)
95(1)
M1
M2
MI Sum
Top 40 MI sites in 2009 H1N1 M gene segment
0
10
20
30
40
50
15(1)
*24(3)
28(3)
*81(3)
84(3)
*88(3)
94(3)
95(2)
96(3)
101(2)
*103(3)
*115(3)
*116(3)
120(3)
121(1)
**137(3)
*138(3)
143(3)
149(3)
150(3)
152(3)
157(3)
160(3)
162(3)
167(3)
168(3)
181(1)
184(3)
191(3)
198(3)
222(3)
226(3)
20(2)
*28(2)
35(3)
52(3)
69(3)
*79(1)
*95(2)
**96(3)
M1
M2
MI Sum
Top 40 MI sites in swine M gene segment

W. Hu / Natural Science 2 (2010) 1225-1246
1238
Table 9. Top 20 MI site pairs in the M segment with smallest P-values.
Avian P value = 0 Hum an P value = 0 2009 H1N1 P value < 0.035 Swine P value = 0
Site Protein Site Protein Site Protein Site Protein Site Protein Site Protein Site Protein Site Protein
140(3) M1 193(3) M1 200(3) M1 54(1) M2 32(1) M1 29(3) M2 164(3) M1 200(3) M1
115(3) M1 140(3) M1 129(3) M1 146(3) M1 22(1) M1 51(2) M2 232(3) M1 68(3) M2
84(3) M1 140(3) M1 239(1) M1 55(3) M2 32(3) M2 51(2) M2 111(3) M1 185(3) M1
84(3) M1 168(3) M1 150(3) M1 239(1) M1 33(1) M2 51(2) M2 111(3) M1 138(3) M1
140(3) M1 168(3) M1 24(3) M2 53(3) M2 36(2) M2 51(2) M2 94(1) M1 232(3) M1
41(3) M1 84(3) M1 24(3) M2 30(3) M2 12(3) M1 17(2) M2 94(1) M1 68(3) M2
84(3) M1 143(3) M1 159(3) M1 88(3) M2 97(3) M1 14(2) M2 138(3) M1 185(3) M1
140(3) M1 143(3) M1 30(3) M2 53(3) M2 130(1) M1 136(3) M1 3(3) M1 98(1) M1
140(3) M1 150(3) M1 150(3) M1 55(3) M2 116(2) M1 136(3) M1 3(3) M1 170(3) M1
150(3) M1 218(3) M1 35(3) M1 150(3) M1 133(3) M1 136(3) M1 13(2) M1 45(1) M1
41(3) M1 168(3) M1 129(3) M1 155(3) M1 5(3) M1 29(3) M1 16(3) M1 218(3) M1
41(3) M1 140(3) M1 16(3) M1 150(3) M1 22(1) M1 118(3) M1 98(1) M1 170(3) M1
41(3) M1 143(3) M1 15(3) M1 88(3) M2 28(2) M1 45(3) M2 189(3) M1 210(2) M1
115(3) M1 143(3) M1 167(1) M1 88(3) M2 45(2) M1 224(2) M1 215(3) M1 55(1) M2
143(3) M1 168(3) M1 138(3) M1 239(1) M1 49(3) M1 237(1) M1 222(2) M1 57(3) M2
115(3) M1 168(3) M1 102(3) M1 53(3) M2 53(2) M1 60(2) M1 24(3) M2 38(1) M2
84(3) M1 115(3) M1 10(3) M1 88(3) M2 60(1) M1 101(3) M1 164(3) M1 185(3) M1
143(3) M1 150(3) M1 37(3) M1 53(3) M2 85(3) M1 133(1) M1 15(3) M1 54(2) M2
115(3) M1 193(3) M1 142(1) M1 160(2) M1 97(2) M1 43(1) M2 42(1) M1 94(1) M1
143(3) M1 218(3) M1 10(3) M1 167(1) M1 116(2) M1 133(3) M1 58(2) M1 58(3) M1
Table 10. Sites in the networks of highly co-mutated sites in the M segment of different origins.
Avian M1_168(3),M1_115(3),M1_41(3),M1_143(3),M1_218(3), M1_140(3), M1_150(3),M1_84(3), M1_193(3)
Human: first M1_150(3), M1_16(3), M1_35(3), M2_55(3),M1_239(1), M1_138(3)
Human: second M2_53(3), M2_24(3), M2_30(3), M1_102(3), M2_30(3),M1_37(3)
Human: third M2_88(3),M1_15(3), M1_159(3),M1_167(1), M1_10(3)
2009 H1N1:first M1_116(2), M1_136(3), M1_133(3), M1_130(1)
2009 H1N1:second M2_51(2), M2_36(2), M2_33(1), M2_32(3), M1_22(1), M1_118(3)
Swine:first M1_111(3), M1_138, M1_185(3), M1_164(3), M1_200(3)
Swine: second M1_3(3), M1_98(1),M1_170(3)
Swine: third M1_94(1), M1_232(3), M2_68(3)
Table 11. Top 20 MI site pairs in the NS segment with smallest P-values.
Avian
P value = 0
Human
P value = 0
2009 H1N1
P value < 0.015
Swine
P value = 0
Site
Protein
Site
Protein
Site
Protein
Site
Protein
Site
Protein
Site
Protein
Site
Protein
Site
Protein
42(3)
NS1
163(3)
NS1
113(3)
NS1
14(1)
NS2
32(1)
NS1
101(3)
NS2
52(3)
NS1
59(2)
NS1
116(3)
NS1
163(3)
NS1
52(3)
NS1
112(2)
NS1
12(3)
NS1
89(2)
NS2
52(3)
NS1
67(1)
NS1
137(3)
NS1
163(3)
NS1
112(2)
NS1
14(1)
NS2
97(3)
NS1
86(2)
NS2
59(2)
NS1
67(1)
NS1
147(3)
NS1
163(3)
NS1
19(3)
NS1
112(2)
NS1
116(2)
NS1
136(3)
NS1
67(1)
NS1
82(3)
NS1
71(3)
NS1
95(3)
NS1
19(3)
NS1
14(1)
NS2
133(3)
NS1
136(3)
NS1
25(1)
NS1
25(3)
NS1
95(3)
NS1
22(2)
NS2
59(2)
NS1
57(2)
NS2
5(3)
NS1
29(3)
NS1
59(2)
NS1
82(3)
NS1
157(3)
NS1
163(3)
NS1
19(3)
NS1
98(1)
NS1
22(1)
NS1
118(3)
NS1
52(3)
NS1
82(3)
NS1
116(3)
NS1
160(3)
NS1
67(2)
NS1
90(1)
NS1
28(2)
NS1
117(3)
NS2
67(1)
NS1
113(3)
NS1
116(3)
NS1
147(3)
NS1
8(3)
NS1
19(3)
NS1
45(1)
NS1
67(1)
NS2
25(3)
NS1
67(1)
NS1
67(1)
NS1
163(3)
NS1
98(1)
NS1
112(2)
NS1
49(3)
NS1
79(3)
NS2
82(3)
NS1
125(3)
NS1
95(3)
NS1
83(3)
NS2
8(3)
NS1
112(2)
NS1
53(2)
NS1
60(2)
NS1
82(3)
NS1
158(3)
NS1
42(3)
NS1
67(1)
NS1
8(3)
NS1
14(1)
NS2
60(1)
NS1
101(3)
NS1
67(1)
NS1
125(3)
NS1
71(3)
NS1
83(3)
NS2
98(1)
NS1
14(1)
NS2
68(3)
NS1
163(2)
NS1
25(3)
NS1
82(3)
NS1
42(3)
NS1
116(3)
NS1
67(2)
NS1
57(2)
NS2
85(3)
NS1
133(1)
NS1
76(1)
NS1
107(3)
NS1
67(1)
NS1
116(3)
NS1
90(1)
NS1
57(2)
NS2
97(2)
NS1
115(1)
NS2
91(1)
NS1
125(3)
NS1
116(3)
NS1
137(3)
NS1
52(3)
NS1
14(1)
NS2
116(2)
NS1
133(3)
NS1
52(3)
NS1
117(3)
NS1
115(3)
NS1
157(3)
NS1
28(1)
NS1
101(1)
NS1
163(3)
NS1
25(3)
NS2
82(3)
NS1
113(3)
NS1
122(3)
NS1
163(3)
NS1
19(3)
NS1
52(3)
NS1
104(3)
NS2
108(2)
NS2
67(1)
NS1
107(3)
NS1
127(2)
NS1
22(2)
NS2
8(3)
NS1
98(1)
NS1
105(1)
NS2
108(2)
NS2
67(1)
NS1
158(3)
NS1
42(3)
NS1
137(3)
NS1
125(3)
NS1
145(1)
NS1
84(3)
NS1
123(1)
NS1
67(1)
NS1
76(1)
NS1

W. Hu / Natural Science 2 (2010) 1225-1246
1239
Table 12. Sites in the networks of highly co-mutated sites in the NS segment of different origins.
Avian
NS1_163(3),NS1_116(3), NS1_160(3),NS1_147(3),NS1_67(1), NS1_122(3),NS1_42(3),NS1_137(3), NS1_157(3),NS1_115(3)
Huma n
NS2_14(1),NS1_112(2),NS1_52(3),NS1_98(1), NS1_113(3), NS1_19(3), NS1_8(3)
2009 H1N1:first
NS1_133(3), NS1_116(2), NS1_136(3)
2009 H1N1:second
NS2_105(1), NS2_108(2), NS2_104(3)
Swine
NS1_91(1), NS1_125(3), NS1_76(1), NS1_107(3), NS1_67(1), NS1_113(3), NS1_158(3), NS1_82(3), NS1_52(3), NS1_117(3), NS1_25(1),
NS1_25(3), NS1_59(2)
Figure 9. Top 40 MI sum sites in the NS segment of different origins.
100
105
110
115
120
125
8(3)
15(3)
19(3)
23(3)
26(3)
37(3)
42(3)
51(3)
54(3)
67(1)
71(3)
*79(3)
**80(1)
*83(3)
**84(1)
*84(2)
*84(3)
*87(3)
90(3)
95(3)
99(3)
103(2)
*114(1)
116(3)
122(3)
127(2)
130(3)
136(3)
137(3)
162(3)
163(3)
166(1)
166(3)
26(2)
63(2)
63(3)
83(1)
83(3)
86(2)
109(3)
Importance
Top 40 MI positions in avian NS gene segment
0
5
10
15
20
25
30
35
40
45
50
8(3)
9(3)
18(1)
19(3)
23(2)
26(3)
28(1)
41(2)
52(3)
55(1)
59(2)
66(3)
67(2)
73(3)
(a**, s**)84(1)
88(3)
(p**)90(1)
98(1)
(p**)100(3)
101(1)
105(3)
112(2)
113(3)
123(3)
125(3)
141(3)
143(3)
145(1)
(a*,s*)14(1)
(p*)40(1)
(a*,p*,s*)57(2)
(s*)68(3)
71(3)
75(3)
87(3)
(a*,s*)89(3)
90(3)
111(3)
(a*,p**)115(3)
117(3)
NS1
NS2
MI Sum
Top 40 MI positions in human NS gene segment
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
*5(3)
12(3)
22(1)
29(3)
32(1)
45(1)
48(3)
49(3)
**53(2)
55(2)
60(1)
60(2)
*68(3)
84(3)
85(3)
97(3)
101(3)
104(2)
116(2)
118(3)
122(3)
123(1)
130(1)
133(1)
133(3)
134(1)
136(3)
159(2)
163(3)
16(3)
25(3)
63(1)
79(3)
86(2)
89(2)
101(3)
104(3)
105(1)
108(2)
123(2)
NS1
NS2
MI Sum
Top 40 MI positions in 2009 H1N1 NS gene segment
0
10
20
30
40
50
60
70
*25(1)
*25(3)
27(1)
36(3)
49(3)
52(3)
56(1)
59(2)
*67(1)
74(1)
76(1)
78(3)
*82(3)
86(1)
87(3)
91(1)
107(3)
108(2)
110(3)
113(3)
116(3)
117(3)
118(1)
122(3)
125(3)
144(1)
147(3)
151(3)
*153(3)
*158(3)
160(3)
*48(3)
*57(2)
59(3)
*62(3)
63(2)
64(3)
67(3)
*70(3)
*89(3)
NS1
NS2
MI Sum
Top 40 MI positions in swine NS gene segment

W. Hu / Natural Science 2 (2010) 1225-1246
1240
Figure 10. Top 40 MI sum sites in the PB1 segment of different origins.
Figure 11. Averaged MI within and between overlapping genes of different origins.
0
20
40
60
80
100
12(3)
17(3)
19(3)
23(3)
79(3)
*88(3)
89(1)
130(3)
142(3)
186(3)
231(3)
239(1)
262(3)
275(3)
300(3)
302(3)
325(3)
340(3)
364(3)
373(3)
383(3)
389(3)
395(3)
426(3)
434(3)
473(3)
482(3)
486(1)
528(1)
582(1)
608(1)
623(1)
625(1)
638(1)
662(1)
666(1)
678(1)
683(1)
727(1)
734(2)
PB1
PB1-F2
PB1
MI Sum
Top 40 MI positions in avian PB1 gene segment
210
215
220
225
230
235
240
245
250
12(3)
(a*, p*, s*)81(3)
123(3)
126(3)
149(3)
154(3)
160(3)
206(3)
238(3)
248(3)
259(3)
262(3)
275(3)
280(3)
307(3)
320(3)
322(3)
352(3)
361(3)
364(3)
399(3)
417(3)
422(3)
434(3)
440(3)
454(3)
498(3)
502(3)
517(3)
520(3)
521(3)
527(3)
540(3)
564(3)
591(3)
624(3)
627(3)
648(3)
662(3)
672(3)
PB1
PB1-F2
PB1
Mi Sum
Top 40 MI positions in human PB1 gene segment
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
14(3)
45(3)
71(3)
80(3)
84(3)
100(3)
108(1)
119(2)
*119(3)
154(3)
155(3)
192(3)
201(3)
211(3)
212(3)
288(3)
291(2)
301(3)
353(2)
363(3)
370(2)
387(3)
422(3)
466(2)
480(2)
486(3)
491(3)
504(3)
517(1)
533(3)
564(3)
566(1)
568(3)
587(2)
633(3)
646(1)
665(2)
667(2)
692(3)
743(3)
PB1
PB1-F2
PB1
MI Sum
Top 40 MI positions in 2009 H1N1 PB1 gene segment
0
50
100
150
200
250
20(3)
*66(3)
*81(3)
101(3)
126(3)
149(3)
160(3)
170(3)
206(3)
248(3)
269(3)
352(3)
364(3)
385(3)
394(3)
399(3)
400(3)
417(3)
418(3)
423(3)
434(3)
454(3)
463(3)
473(3)
488(3)
502(3)
509(3)
520(3)
527(3)
566(3)
568(3)
592(3)
627(3)
639(3)
643(3)
648(3)
672(3)
719(3)
735(3)
752(3)
PB1
PB1-F2
PB1
MI Sum
Top 40 MI positions in swine PB1 gene segment
0
0.01
0.02
0.03
0.04
0.05
0.06
Averaged MI
Avian
Human
2009 H1N1
Swine
0
0.001
0.002
0.003
0.004
0.005
M1_M1
M1_M2
M2_M2
M1_and_M2
NS1_NS1
NS1_NS2
NS2_NS2
NS1_and_NS2
PB1_PB1
PB1_F2_PB1_F2
PB1_PB1_F2
Averaged MI
2009 H1N1

W. Hu / Natural Science 2 (2010) 1225-1246
1241
Table 13. Top 20 MI site pairs in the PB1 segment with smallest P-values.
Avian P value = 0 Huma n P value = 0 2009 H1N1 P value <0.009 Swine P value = 0
Site Protein Site Protein Site Protein Site Protein Site Protein Site Protein Site Protein Site Protein
12(3) PB1 734(2) PB1 123(3) PB1 154(3) PB1 100(3) PB1-F2 566(1) PB1 149(3) PB1 627(3) PB1
12(3) PB1 434(3) PB1 123(3) PB1 498(3) PB1 100(3) PB1-F2 353(2) PB1 170(3) PB1 592(3) PB1
12(3) PB1 373(3) PB1 149(3) PB1 502(3) PB1 115(3) PB1-F2 480(2) PB1 20(3) PB1 639(3) PB1
12(3) PB1 275(3) PB1 149(3) PB1 648(3) PB1 154(3) PB1 201(3) PB1 364(3) PB1 639(3) PB1
142(3) PB1 239(1) PB1 149(3) PB1 423(3) PB1 154(3) PB1 692(3) PB1 364(3) PB1 672(3) PB1
275(3) PB1 364(3) PB1 149(3) PB1 440(3) PB1 154(3) PB1 735(1) PB1 364(3) PB1 454(3) PB1
275(3) PB1 373(3) PB1 154(3) PB1 498(3) PB1 155(3) PB1 566(1) PB1 364(3) PB1 520(3) PB1
275(3) PB1 434(3) PB1 154(3) PB1 206(3) PB1 155(3) PB1 353(2) PB1 364(3) PB1 417(3) PB1
275(3) PB1 734(2) PB1 160(3) PB1 488(3) PB1 201(3) PB1 692(3) PB1 364(3) PB1 434(3) PB1
364(3)
PB1 373(3) PB1 160(3) PB1 592(3) PB1 212(3) PB1 667(2) PB1 364(3) PB1 423(3) PB1
364(3) PB1 434(3) PB1 206(3) PB1 498(3) PB1 235(3) PB1 619(3) PB1 364(3) PB1 627(3) PB1
364(3) PB1 734(2) PB1 234(3) PB1 568(3) PB1 353(2) PB1 566(1) PB1 394(3) PB1 454(3) PB1
373(3)
PB1 434(3) PB1 248(3) PB1 517(3) PB1 384(1) PB1 533(3) PB1 394(3) PB1 400(3) PB1
373(3) PB1 734(2) PB1 248(3) PB1 648(3) PB1 384(1) PB1 633(3) PB1 400(3) PB1 639(3) PB1
373(3) PB1 623(1) PB1 361(3) PB1 502(3) PB1 436(3) PB1 564(3) PB1 400(3) PB1 454(3) PB1
373(3)
PB1 426(3) PB1 440(3) PB1 502(3) PB1 533(3) PB1 633(3) PB1 417(3) PB1 643(3) PB1
426(3) PB1 638(1) PB1 502(3) PB1 648(3) PB1 57(3) PB1-F2 393(2) PB1 417(3) PB1 672(3) PB1
434(3) PB1 734(2) PB1 517(3) PB1 591(3) PB1 64(1) PB1-F2 473(3) PB1 417(3) PB1 566(3) PB1
88(3) PB1-F2 395(3) PB1 517(3) PB1 648(3) PB1 71(3) PB1-F2 633(3) PB1 454(3) PB1 639(3) PB1
88(3) PB1-F2 666(1) PB1 591(3) PB1 648(3) PB1 71(3) PB1-F2 533(3) PB1 488(3) PB1 639(3) PB1
Table 14. Sites in the networks of highly co-mutated sites in the PB1 segment of different origins.
Avian
PB1_275(3),PB1_364(3), PB1_434(3),PB1_12(3),PB1_373(3),PB1_623(1),PB1_734(2),PB1_426(3),PB1_638(1)
Human
PB1_361(3),PB1_502(3),PB1_440(3),PB1_149(3),PB1_423(3), PB1_648(3),PB1_591(3), PB1_517(3),PB1_248(3)
2009 H1N1
: first
PB1_F2_71(3),
PB1_633(3), PB1_533(3), PB1_384(1)
2009 H1N1
: second
PB1_F2_100(3), PB1_566(1), PB1_155(3), PB1_353(2)
2009 H1N1
: third
PB1_692(3), PB1_154(3), PB1_201(3), PB1_735(1)
Swine
PB1_520(3),PB1_434(3),PB1_20(3),PB1_454(3),PB1_364(3),PB1_639(3),PB1_488(3),PB1_423(3),PB1_672(3),PB1_41
7(3),PB1_643(3),PB1_400(3),PB1_394(3),PB1_566(3)
lective pressure between human and avian influenza,
which were 115, 121, 137 in M1, and 11, 16, 20, 54, 57,
78, 86, and 93 in M2 [47]. Furthermore, M1 sites 115,
121, and 137 and M2 sites 11, 14, 20, 28, 54, 55, 57, 78,
and 86 were host markers in [48], illustrating the signi-
ficance of M1 sites 115, 121, and 137, and M2 sites 11,
20, 54, and 78. In our analysis, avian and swine M1 sites
115, 121, 137, swine M2 site 20, human and 2009 H1N1
M2 site 54, and, 2009 H1N1 M2 site 57 were selected as
highly co-mutated sites (Figure 8). The M2 sites be-
tween 10 and 22 overlapped with M1. Although M2 is a
short gene, it had 15 sites selected as highly co-mutated
sites in human with many of these sites selected by other
viruses as well, and the second contributor was 2009
H1N1.
The outstanding highly co-mutated sites in M were:
115(3) and 140(3) in avian virus (140(3) a new site),
614(3) and 200(3) in 2009 H1N1. Human and swine
viruses appeared to have homogenous MI sums of their
top 40 sites, so no clear outstanding sites were detected.
The 2009 pandemic H1N1 virus contains the adaman-
tine-resistant mutation S31N in its M2 protein. However,
this site was not selected as a host marker in Section 3.1
and as a highly co-mutated site here. M2 55 is a site
known for enhanced transmission, and it was selected as
a highly co-mutated site in human and 2009 H1N1 vi-
ruse s.
Some selected codons in the M segment contained one
position that was important for host differentiation, and
another for co-mutation, a feature not shared by the PB1
segment. Avian M1 121(1) and 137(1) were host mark-
ers and 121(3) and 137(3) were highly co-mutated sites.

W. Hu / Natural Science 2 (2010) 1225-1246
1242
Figure 12. Averaged MI within the functional domains in the overlapping genes of different origins.
The M1 of 2009 H1N1 had sites 94(1), 147(1), 150(2) as
highly co-mutated sites and 94(3), 147(3), 150(3) host
markers. Swine M1 137(3) and M2 96(3) were highly
co-mutated sites and M1 137(1) and M2 96(1) were host
markers. Also some sites were found in multiple viruses.
M1 150(3), and M2 54(1) and 78(1) were not only high-
ly co-mutated in human virus, but also in avian, 2009
H1N1, and swine viruses.
The positions marked with one asterisk are host
marker positions found in Section 3.1, and those marked
with two asterisks are within the same codon of a host
marker but different from the host marker position. In
the human M gene segment, a single letter “a” was at-
tached to the position if it is a top avian MI sum site, a
“p” for pandemic 2009 H1N1, and a “s” for swine virus.
The sites in M1 and M2 of different origins were both
present in the top 20 highly co-mutated site pairs with
the exception of avian virus (Table 9). From these top 20
site pairs, networks of co-mutated sites could be induced,
and the sites in these networks were displayed in Table
10. Table 9 showed that the most connected sites were
M1 143(3), M1 168(3), M1 115(3), M1 84(3), M1
140(3), and M1 41(3) in avian virus, M1 150(3) in the
first network, M2 53(3) in the second network, and M2
88(3) and M1 10(3) in the third network of human virus,
M1 116(2), M1 133(3), M1 136(3) in the first network
and M2 51(2) in the second network of 2009 H1N1, and
M1 113(3), M1 138(3), and 185(3) in the first network,
M1 3(3), M1 98(1), and M 170(3) in the second network,
and M1 94(1), M1 68(3), and M1 232(3) in the third
network in swine virus.
3.2.1.2. NS Segment
Following the same pattern of their relative contribu-
tions to host distinction (Figure 2), the NS1 and NS2 of
different origins made similar contributions to the top
MI sites, with NS1 contributing more than NS2 (about
13% from NS1 and about 7% from NS2). It seemed
that the sites in NS1 were more correlated than NS2, and
were more important than NS2 in host adaptation. Some
selected codons in the NS segment contained one posi-
tion that was important for host differentiation, and
another for co-mutation. Avian NS1 80(1) 84(1), 84(2),
84(3) were highly co-mutated sites, and NS1 80(3),
84(2), 84(3) as host markers. NS1 of 2009 H1N1 had
site 53(2) as a highly co-mutated site, and 53(3) as a host
marker. Along a similar line, some selected sites were
shared by different viruses. NS1 84(1) and NS2 14(1),
57(2), 89(3), and 115(3) were highly co-mutated in hu-
man virus as well as in avian, 2009 H1N1, and swine
viruses (Figure 9).
The outstanding highly co-mutated sites selected in
NS were: NS1 163(3) in avian, NS1 123(1) and NS2
123(2) in 2009 H1N1. Human and swine viruses had
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
Averaged MI
Avian
Human
2009 H1N1
Swine
0
0.001
0.002
0.003
0.004
0.005
Averaged MI
2009 H1N1

W. Hu / Natural Science 2 (2010) 1225-1246
1243
homogenous MI sums of their top 40 sites. In particular,
the NS segment of 2009 H1N1 had NS1123 as its num-
ber one site and carries a novel I123V mutation [49]. It
was observed that avian NS1 had two clusters of highly
co-mutated sites: the first located around codons 79 -- 90
that was in the eIF4G binding domain and the second
around 162--166 that was in the SH2/SH3 binding do-
main [48]. Mutation G92E in avian H5N1 NS1 is known
to promote the inhibition of host immune responses, but
was not selected as a highly co-mutated site. The amino
acids at avian NS1 42 were a key determinant of viru-
lence of H5N1 [46], and site 42 was selected as a highly
co-mutated position in avian NS1 (Figure 9).
Some selected sites in the NS segment overlapped
with the host markers found in Section 3.1, which was in
clear contrast to the PB1 segment. NS1 sites 22(1), 81(3),
84(2), 84(3), 215(1) (57(3) in NS2 numbering), and NS2
sites 60(1), 60(2), 70(1), 70(3), and 107(1) were host
markers in [24,48]. NS1 sites 22(1) in 2009 H1N1,
84(1) in avian and human viruses, 84(2) in avian virus,
84(3) in avian and 2009 H1N1 viruses, NS2 57(2) in
human and swine viruses, 70(3) in swine virus were se-
lected as highly co-mutated sites. The NS2 sites between
12 and 74 overlapped with NS1. NS1 220 (63 in NS2
numbering) was an interesting site, since the NS1 genes
of 2009 H1N1 and swine both had a stop codon (tga) at
220 but avian and human NS1 genes have cgg(R) at 220.
Additionally, avian and swine NS2 site 63(2), avian and
2009 NS2 63(1) were selected as highly co-mutated sites
in this study.
The positions marked with one asterisk are host
marker positions found in Section 3.1, and those marked
with two asterisks are within the same codon of a host
marker but different from the host marker position. In
the human M gene segment, a single letter “a” was at-
tached to the position if it is a top avian MI sum site, a
“p” for pandemic 2009 H1N1, and a “s” for swine virus.
The sites in NS1 and NS2 of different origins were
both present in the top 20 highly co-mutated site pairs
with the exception of swine virus (Table 11). From these
top 20 site pairs, networks of co-mutated sites could be
induced, and the sites in these networks were displayed
in Table 12. Table 11 showed that the most connected
sites were NS1 116(3) and NS1 163(3) in avian virus,
NS1 112(2), NS1 19(3), NS1 98(1), and NS2 14(1) in
human virus, NS1 116(2), NS1 133(3), NS1 136(3) in
the first network and NS2 104(3), NS2 105(1), NS2
108(2) in the second network of 2009 H1N1, and NS1
25(3), NS1 52(3), NS1 67(1), and NS1 82(3) in swine
virus.
3.2.1.3. PB1 Segment
As the core subunit of viral polymerase of influenza,
the regions at the start and end of PB1 (residues 1--25
and 600 -- 757) bind to PA and PB2 respectively. As in
the case of host markers in Section 3.1, there was one
highly co-mutated site selected in the PA binding region
in human, 2009 H1N1, and swine viruses, and four in
avian virus. However, avian and swine PB1 had a great
number of sites selected (8 --10) in the PB2 binding re-
gion. It was of note that swine PB1 had nine highly
co-mutated sites in the range of codon 400 to codon 500,
which were in the cRNA binding domain (Figure 10).
The outstanding highly co-mutated sites in PB1 were:
149(3) in human virus, 201(3), 353(2), and 692(3) in
2009 H1N1. Avian and swine viruses had homogenous
MI sums of their top 40 sites, so no clear outstanding
sites were seen. Further, there were several sites in PB1
that were shared between viruses of different origins,
which were significant sites as well.
The PB1 segment seemed to have a much smaller
overlap between host marker sites and highly co-mutated
sites than the M and NS segments. PB1 81(3) (50(2) in
PB1-F2 numbering) was the only site in the PB1-F2 re-
gion selected as a highly co-mutated site in the human
virus, and it also was selected in the swine virus. Further,
PB1-F2 50(2) was a host marker in avian, 2009 H1N1
and swine viruses (Figure 4). Although there were sev-
eral other key host markers in PB1-F2 such as 27(2)
(58(3) in PB1 numbering) and 60(2) (91(3) in PB1
numbering) (Figure 4), they were not selected as highly
co-mutated sites here. As in the case of host markers,
PB1-F2 of 2009 H1N1 made a prominent contribution to
the correlation of the whole PB1 segment with eight
selected sites when compared to other species. PB1-F2
of 2009 H1N1 was more important in host distinction
and was more correlated than that of avian, human, and
swine viruses.
The positions marked with one asterisk are host
marker positions found in Section 3.1, and those marked
with two asterisks are within the same codon of a host
marker but different from the host marker position. In
the human M gene segment, a single letter “a” was at-
tached to the position if it is a top avian MI sum site, a
“p” for pandemic 2009 H1N1, and a “s” for swine virus.
The sites in PB1-F2 and PB1 of avian and 2009 H1N1
origins were both present in the top 20 highly
co-mutated site pairs, but none of the selected sites of
human and swine viruses contained the PB1-F2 sites
(Table 13). From these top 20 site pairs, networks of
co-mutated sites could be induced, and the sites in these
networks were displayed in Table 14. Table 13 showed
that the most connected sites were PB1 12(3), PB1
275(3), PB1 373(3), PB1 364(3), PB1 434(3), and PB1
734(2) in avian virus, PB1 149(3) and PB1 648(3) in
human virus, PB1 533(3) and PB1 633(3) in the first
network, PB1 353(2) and PB1 566(1) in the second net-

W. Hu / Natural Science 2 (2010) 1225-1246
1244
work , and PB1 154(3) in the third network of 2009
H1N1, and PB1 364(3), PB1 417(3), and PB1 454(3) in
swine virus.
3.2.2. Correlations within and between
Overlapping Genes in M, NS, and PB1
In addition to the highly co-mutated sites and site
pairs in the overlapping genes found in Section 3.2.1,
mutual information was employed to understand the
co-mutations within and between these genes. The for-
mer reflects the local correlation whereas the latter
represents the global correlation of the genes. Our analy-
sis indicated that the correlation within M1 was the
highest, between M1 and M2 the second, and within M2
the least across avian, human, 2009 H1N1, and swine
viruses. However, this correlation pattern was not ob-
served in NS1 and NS2. PB1-F2 was found to be more
correlated within itself than with PB1 of avian, human,
and swine viruses, and exhibited similar correlations
within itself and with PB1 of 2009 H1N1 (Figure 11).
Furthermore, a similar analysis was applied to the
co-mutations in the functional domains of a gene. It re-
vealed that in general each gene had its own highest
correlating domain (Figure 12). Among the various do-
mains in M1, the highest correlating domain was NS2
binding in avian virus, NLS in human and swine viruses,
RNP binding in 2009 H1N1. In M2, the co-mutation rate
was higher in trans-membrane than extracellular domain
in avian, human, 2009 H1N1, and swine viruses. In NS1,
the highest correlating domain was RNA binding in
avian virus, PABII binding in human virus, NLS in 2009
H1N1, and PDZ ligand motif in swine virus. In PB1, the
highest correlating domain was PB2 binding in avian
virus, NLS in human virus, PB1-F2 Mitochondrial
targeting binding in 2009 H1N1, and cRNA binding in
swine virus. There is only one functional domain exist-
ing in PB1-F2, and the co-mutation rate was high in
2009 H1N1 and swine viruses relative to avian and hu-
man viruses.
4. DISCUSSION
The emergence of the pandemic 2009 H1N1 virus
highlighted the importance of understanding the mole-
cular basis for zoonotic influenza viruses to cross species
barrier to infect humans. The segmented genome of in-
fluenza makes it easy to generate a new virus through
multiple reassortment with several influenza species,
with 2009 H1N1 as the most recent example. Influenza
viruses have several gene segments that encode multiple
proteins, which provide ideal data for study of the inte-
ractions between the proteins expressed from the same
gene segment. The NS1 and PB1-F2 proteins are two
key virulence factors of influenza. Through the integra-
tive approach developed in this study, these two genes
were investigated in the context of their native gene
segments, NS and PB1 respectively, and of the interac-
tion between their overlapping genes. Furthermore, the
three stop codons in the PB1-F2 of 2009 H1N1 were
identified as important host markers in 2009 H1H1. It is
well known that the functions of a protein depend on the
coordinated interactions among its residues. Mutations
occurred at one site in a protein may need to be com-
pensated by mutations at other sites, in order to maintain
proper protein structure and function. While past re-
search on overlapping genes targeted primarily the evo-
lution of influenza, our current study focused on host
markers and co-mutations in the overlapping genes of
the M, NS, and PB1 segments, which revealed new in-
sights that could not be gained through the study of these
genes individually.
5. CONCLUSIONS
We explored two aspects of the overlapping genes:
host markers and co-mutations. In co-mutations, our
analysis was further divided into local correlations of
sites and site pairs in a gene and global correlations
within and between genes. Besides the host markers
discovered in the overlapping genes along with their
relative significance calculated, Random Forests were
able to predict the significance of the mutations at the
three notable stop codons 12, 58, and 88 in the PB1-F2
gene of 2009 H1N1, revealing that the closer to the start
of the gene the more important the mutation was. Also
GC content analysis unexpectedly demonstrated that the
full length PB1-F2 protein of 2009 H1N1 if all the three
mutations occurred would function more like a swine
protein than a human protein despite the human H3N2
origin of its PB1 segment. In the co-mutations of each
gene segment, the top 40 highly co-mutated sites, top 20
highly co-mutated site pairs with P values less than
0.035, and networks of highly co-mutated sites were
found. The correlation of the sites in M1 was higher than
that of M2 of all the four viruses of avian, human, 2009
H1N1, and swine origins. However, the correlations of
NS1 and NS2 of different origins varied. Finally,
PB1-F2 was more correlated within itself than with PB1
of avian, human, and swine viruses, and exhibited simi-
lar correlations within itself and with PB1 of 2009
H1N1.
6. ACKNOWLEDGEMENTS
We thank Houghton College for its financial support.
REFERENCES
[1] Betakova, T. (2007) M2 Protein–a proton channel of

W. Hu / Natural Science 2 (2010) 1225-1246
1245
influenza a virus. Current Pharmaceutical Design, 13,
3231-3235.
[2] Ma, C., Polishchuk, A.L., Ohigashi, Y., Stouffer, A.L.,
Schön, A., Magavern, E., Jing, X., Lear, J.D., Freire, E.
and Lamb, R.A., (2009) Identification of the functional
core of the influenza A virus A/M2 proton-selective ion
channel. Proceedings of the National Academy of
Sciences of the United States of America, 106, 12283-
12288.
[3] Dua, Q.S., Wang, S.Q., Huang, R.B. and Chou, K.C.
(2010) Computational 3D structures of drug-targeting
proteins in the 2009-H1N1 influenza A virus. Chemical
Physics Letters, 485, 191-195.
[4] Pan, C., Cheung, B., Tan, S., Li, C., Li, L., et al. (2010)
Genomic signature and mutation trend analysis of pan-
demic (H1N1) 2009 influenza A virus. PLoS ONE, 5,
e9549.
[5] Jackson, D., Hossain, M.J., Hickman, D., Perez, D.R. and
Lamb, R.A. (2008) A new influenza virus virulence de-
terminant: The NS1 protein four C-terminal residues
modulate pathogenicity. Proceedings of the National
Academy of Sciences of the United States of America,
105, 4381-4386.
[6] Soubies, S.M., Volmer, C., Croville, G., Loupias, J.,
Peralta, B., Costes, P., Lacroux, C., Guérin, J.L. and
Volmer, R. (2010) Species-specific contribution of the
four C-terminal amino acids of influenza A virus NS1
protein to virulence. The Journal of Virology, 84,
6733-6747.
[7] Hale, B.G., Steel, J., Manicassamy, B., Medina, R.A., Ye,
J., Hickman, D., Lowen, A.C., Perez, D.R. and García-
Sastre, A. (2010) Mutations in the NS1 C-terminal tail do
not enhance replication or virulence of the 2009 pan-
demic H1N1 influenza A virus. The Journal of General
Virology, 91, 1737-1742.
[8] Long, J.X., Peng, D.X., Liu, Y.L., Wu, Y.T. and Liu, X.F.
(2008). Virulence of H5N1 avian influenza virus en-
hanced by a 15-nucleotide deletion in the viral nonstruc-
tural gene. Virus Genes. 36, 471-478.
[9] Seo, S.H., Hoffmann, E. and Webster, R.G. (2002) Le-
thal H5N1 influenza viruses escape host anti-viral cyto-
kine responses. Nature Medicine, 8, 950-954.
[10] Henkel, M., Mitzner, D., Henklein, P., Meyer-Almes, F-J.,
Moroni, A. et al., (2010) The proapoptotic influenza A
virus protein PB1-F2 forms a nonselective ion channel.
PLoS ONE, 5, e 11112.
[11] McAuley, J.L., Zhang, K. and McCullers, J.A. (2010)
The effects of influenza a virus PB1-F2 protein on poly-
merase activity are strain specific and do not impact pa-
thogenesis. Journal of Virology, 84, 558-56 4.
[12] Krejnusová, I. , Gocníková, H., By strick á, M., Bennink,
H.J. and Russ, G. (2009) Antibodies to PB1-F2 protein
are induced in response to influenza a virus infection .
Archives of Virology, 154, 1599-1604
[13] Zell, R., Krumbholz, A., Eitner, A., Krieg, R., Halbhuber,
K.J. and Wutzler, P. (2007) Prevalence of PB1-F2 of in-
fluenza A viruses. Journal General Vi ro logy , 88,
536-546.
[14] Conenello, G., Zamarin, D., Perrone, L., Tumpey, T., and
Palese, P. (2007). A single mutation in the PB1-F2 of
H5N1 (HK/97) and 1918 influenza A viruses contributes
to increased virulence. PLoS Pathogens, 3, 1414-1421.
[15] Mcauley, J., Hornung, F., Boy d, K., Smith, A., Mckeon,
R., Bennink, J., Yewdell, J., and Mccullers, J. (2007) Ex-
pression of the 1918 influenza A virus PB1-F2 enhances
the pathogenesis of viral and secondary bacterial pneu-
monia. Cell Host & Microbe, 2, 240-249.
[16] Hai, R., Schmolke, M., Varga, Z.T., Manicassamy, B.,
Wang, T.T., Belser, J.A., Pearce, M.B., García-Sastre , A.,
Tumpey, T.M. and Palese, P. (2010) PB1-F2 expression
by the 2009 pandemic H1N1 influenza virus has minimal
impact on virulence in animal models. Journal Virology,
84, 4442-4450.
[17] Opal, S. (2010) Understanding viral zoonoses: H1N1
influenza. Veterinary Medicine, 3, 131-135.
[18] Pavesi, A. (2007) Pattern of nucleotide substitution in the
overlapping nonstructural genes of influenza A virus and
implication for the genetic diversity of the H5N1 subtype.
Gene, 402, 28-34.
[19] Campitelli, L., Ciccozzi, M., Salemi, M., Taglia, F., Bo-
ros, S., et al. (2006) H5N1 influenza virus evolution: A
comparison of different epidemics in birds and humans
(1997-2004). Journal of General Virolo gy, 87, 955-960.
[20] Obenauer, J.C., Denson, J., Mehta, P.K., Su, X., Mukati-
ra, S., et al. (2006) Large-scale sequence analysis of
avian influenza isolates. Science, 311, 1576-1580.
[21] Li, K.S., Guan, Y., Wang, J., Smith, G.J., Xu, K.M., et al.
(2004) Genesis of a highly pathogenic and potentially
pandemic H5N1 influenza virus in eastern Asia. Nature,
430, 209-213.
[22] Sabath, N., Landan, G. and Graur, D. (2008) A Method
for the simultaneous estimation of selection intensities in
overlapping genes. PLoS ONE, 3, e3996.
[23] Hu, W. (2010) Novel host markers in the 2009 pandemic
H1N1 influenza A virus. Journal of Biomedical Science
and Engineering, 3, 584-601.
[24] Hu, W. (2010) Nucleotide host markers in the influenza
A viruses. Journal of Biomedical Science and Engineer-
ing, 3, 684-699.
[25] Katoh, K., Kuma, K., Toh, H. and Miyata, T. (2005)
MAFFT version 5: Improvement in accuracy of multiple
sequence alignment. Nucleic Acids Research, 33, 511-
518.
[26] MacKay, D. (2003) Information theory, inference, and
learning algorithms. Cambridge University Press, UK.
[27] Breiman, L., (2001) Random forests. Machine Learning,
45, 5-32.
[28] Díaz-Uriarte, R. and Alvarez de Andrés, S. (2006) Gene
selection and classification of microarray data using
random forest. BMC Bioinformatics, 7, 3.
[29] Kellie, J.A. and Ryan, V.K. (2008) Empirical characteri-
zation of random forest variable importance measures.
Computational Statistics and Data Analysis, 52, 2249-
2260.
[30] Reif, D.M., Motsinger, A.A., McKinney, B.A., Crowe,
J.E., Moore, J.H. (2006) Feature Selection using a ran-
dom forests classifier for the integrated analysis of mul-
tiple data type. Proceedings of 2006 IEEE Symposium on
Computational Intelligence and Bioinformatics and
Computational Biology, CIBCB '06, 1-8.
[31] Pablo, M.G., Furlanellob, C., Biasiolia, F. and Gasperia,
F. (2006) Recursive feature elimination with random
forest for PTR-MS analysis of agroindustrial products.
Chemometrics and Intelligent Laboratory Systems, 83,

W. Hu / Natural Science 2 (2010) 1225-1246
1246
83-90.
[32] Bjoern, H.M., Kelm, B.M., Masuch, R., Himmelreich, U.,
Bachert, P., Petrich, W. and Hamprecht, F.A. (2009) A
comparison of random forest and its Gini importance
with standard chemometric methods for the feature se-
lection and classification of spectral data. BMC Bioin-
formatics, 10, 213.
[33] Gao, D., Zhang, Y.X., Zhao, Y.H. (2009) Random forest
algorithm for classification of multi-wavelength data.
Research in Astronomy and Astrophysics, 9, 220-226.
[34] Hu, W. (2009) Identifying predictive markers of chemo-
sensitivity of breast cancer with random forests. Journal
of Biomedical Science and Engineering, 3, 59-64.
[35] Garten, R.J., Davis, C.T., Russell, C.A., Shu, B., et al.,
(2009) Antigenic and genetic characteristics of
swine-origin 2009 A(H1N1) influenza viruses circulating
in humans. Science, 325, 197-201.
[36] Ozawa, M., Maeda, J., Iwatsuki-Horimoto, K., Watanabe,
S., Goto, H., Horimoto, T. and Kawaoka , Y. (2009) Nuc-
leotide sequence requirements at the 5' end of the in-
fluenza A virus M RNA segment for efficient virus repli-
cation. J Virol. 83(7):3384-8.
[37] Wu B, Wang C M, Dong G Y, et al. (2009) Molecular
characterization of H1N1 influenza A viruses from hu-
man cases in North America. Chinese Sciences Bulletin,
54, 2179-2192
[38] Liu, T. and Ye, Z.P. (2005) Attenuating mutations of the
matrix gene of influenza A/WSN/33 Virus. Journal of
Virology, 79, 1918-1923.
[39] Dey de, V.M., Sheu, T.G., Trujillo, A.A., Oko-
mo-Adhiambo, M., Garten, R., Klimov, A.I., Gubareva,
L.V. (2010) Detection of molecular markers of drug re-
sistance in 2009 pandemic influenza A (H1N1) viruses
by pyrosequencing. Antimicrob Agents Chemother, 54,
1102-1110.
[40] Pan, C.G. and Jian g, S.B. (2009) E14-F55 combination
in M2 protein: A putative molecular determinant respon-
sible for swine-origin influenza A virus transmission in
humans. PLoS Currents Influenza, 29, RRN1044.
[41] Sabath, N., Landan, G. and Graur, D. (2008) A method
for the simultaneous estimation of selection intensities in
overlapping genes. PLoS ONE, 3, e3996.
[42] Lin, D., Lan, J. and Zhang, Z. (2007) Structure and func-
tion of the NS1 protein of influenza A virus. Acta Bio-
chim Biophys Sin (Shanghai), 39, 155-162.
[43] Suwannakhon, N., Pookorn, S., Sanguansermsri, D.,
Chamnanpood, C., Chamnanpood, P., Wongvilairat, R.,
Pongcharoen, S., Niumsup, P.R., Kunthalert, D. and
Sanguansermsri, P. (2008) Genetic characterization of
nonstructural genes of H5N1 avian influenza viruses iso-
lated in Thailand in 2004-2005. Southeast Asian Journal
of Tropical Medicine and Public Health, 39, 837-847.
[44] Scalera, N.M. and Mossad, S.B. (2009) The first pan-
demic of the 21st century: A review of the 2009 pan-
demic variant influenza A (H1N1) virus. Postgraduate
Medicine, 121, 43-47.
[45] Gibbs, A.J., Armstrong, J.S., Downie, J.C. (2009) From
where did the 2009 'swine-origin' influenza A virus
(H1N1) emerge? Journal of Virology, 6, 207.
[46] Jiao, P.R., Tian, G.B., Li, Y.B., Deng, G.H., Jiang, Y.P.,
Liu, C., Liu, W.L., Bu, Z.G., Kawaoka, Y. and Chen,
H.L. (2008) A single-amino-acid substitution in the NS1
protein changes the pathogenicity of H5N1 avian in-
fluenza viruses in mice. Journal of Virology , 82,
1146-1154.
[47] Betakova1, T. and Hay, A.J. (2009) Stability and func-
tion of the influenza A virus M2 ion channel protein is
determined by both extracellular and cytoplasmic do-
mains. Journal Archives of Virology, 154, 147-151.
[48] Miotto, O., Heiny, A.T., Albrecht, R., García-Sastre, A.,
Tan, T.W., August, J.T. and Brusic, V. (2010) Com-
plete-proteome mapping of human influenza a adaptive
mutations: Implications for human transmissibility of
zoonotic strains. PLoS ONE, 5, e9025.
[49] Maurer-Stroh, S., Raphael, T.C.L., Eisenhaber, F., Lin C.,
Shiau, P. P. and Raymond, T.P.L. (2010) A new common
mutation in the hemagglutinin of the 2009 (H1N1) in-
fluenza A virus. PLoS Currents Influenza.