The glutamic acid residue at PB2 627, which is commonly seen in avian viruses, restricts viral growth in humans and monkeys, but a change to lysine restores virus replication in mammalian cells (33
). In this study we computed for every amino acid position (distributed in the 11 known influenza viral ORFs) an entropy value that represents how conserved an amino acid residue is at that given position. We found the entropy value –0.379 at 627 of PB2 and therefore used –0.4 as a threshold to discover other amino acid residues that might be potential determinants of host-cell tropism. Another 51 positions were found to be distinct or nearly distinct between human and avian viruses by this entropy threshold. Most of these (40 of 52) are located in viral ribonucleoproteins (RNPs) (PB2, PB1, PA, and NP), which are essential for viral replication. Taubenberger et al. reported 10 amino acid residues that distinguish human and avian influenza viral polymerases (3
). Six of them were also identified in this study. The entropy values of the 4 missing ones were also found close to the preset threshold (–0.4). For example, PB2 567 showed a human entropy of –0.039 and avian entropy of –0.490, PB1 375 with human entropy –0.165 and avian entropy –0.693, and PA 100 with human entropy –0.061 and avian entropy –0.406. All 3 positions were eliminated earlier from the stage of analyzing the 401-genome primary dataset. The fourth position, PB2 702, although in the first-round list, marginally failed in the subsequent validation with human entropy –0.057 and avian entropy –0.404.
We proposed a computational approach capable of indicating species-associated signatures in studying human versus avian influenza viral genomes. Although we intended to analyze a comprehensive set of avian versus human influenza A viral genomes, the available sequences are predominated by H5N1 in avian viruses and H3N2 in human viruses. The short supply of sequences other than those 2 subtypes may inevitably cause a certain amount of bias in our results. At the completion of this study, we noticed a recent article by Obenauer et al., who had made 169 newly sequenced avian influenza viral genomes available to GenBank on January 26, 2006 (34
); these were not included in our analysis. We checked on our 52 signature positions against these new genomes and found only 2 of them that showed an entropy value slightly over our threshold –0.4. These are PB1-F2 87 and HA 237, with entropy values of –0.522, and –0.692, respectively. The choice of entropy threshold would also affect the number of signatures found. Originally we chose –0.4 on the basis of the value –0.379, computed from PB2 627 by using 95 avian genomes. We noticed that this entropy value reduced to –0.299 at PB2 627 (see ) at the later validation stage, when we found 197 E and 19 K from a total of 215 avian PB2 sequences. If we chose to use a more stringent entropy threshold of –0.3, our analysis still showed 46 of those 52 reported signatures; missing were positions 73, 79, and 82 from PB1-F2, 409 from PA, and 237 and 389 from HA.
In addition to the data limitations, this approach of looking for species-associated signatures by entropy is less useful for HA and NA genes. The genetic diversity that exists in either human or avian viruses for these 2 gene segments can markedly boost their respective entropy to more negative values, thus making it difficult to find residues conserved enough for identifying such signatures. We additionally performed the analysis on human H1, H2, and H3 versus avian HA (). For NA we performed the analysis on human N1 and N2 versus avian NA. We compared 10 human H1, 3 human H2, and 293 human H3 with 95 avian HA sequences and found 13, 13, and 69 signatures (with entropy values for both human and avian within –0.4), respectively. This finding indicates that the human H1 and H2 strains are less distinct from avian strains (H5 dominant) than H3. For NA we found only 6 signatures, in comparison with 8 human N1 versus 95 avian (N1-dominant), and we found only 5 signatures when we compared 298 human N2 and 95 avian sequences. Entropy plots for these analyses can be seen in .
Two genetic alleles (allele A and B) have been described for the NS gene in avian influenza A virus. We decomposed those 95 avian NS genes into 43 in allele A and 52 in allele B and compared their amino acid sequences with 306 human NS genes. For NS1, 6 signatures were found between human viruses and avian allele A viruses, and 35 signatures were found between human viruses and avian allele B viruses. For NS2, 3 signatures were found between human viruses and allele A viruses, and 6 signatures were found between human viruses and allele B viruses. These results suggest that avian allele B viruses are more distinct from human viruses than are allele A viruses. Entropy plots and histograms for these analyses can be seen in and .
From the histograms, we found that some of the 11 genes vary greatly between human and avian viruses, while some others vary little. No boundaries were found between homopairs and heteropairs for HA, NA, and PB1 for human versus avian viruses. This finding seems reasonable because the 2 recent pandemic strains, the 1957 H2N2 and the 1968 H3N2, both originated from reassortment with avian influenza viruses (HA, NA, and PB1 gene segments were from avian influenza). On the other hand, because histograms of NP, followed by PA and PB2, may be used to distinguish human influenza viruses from avian influenza viruses, perhaps some biologic constraints against the occurrence of reassortment exist for these 3 genes. Both the M and NS genes are less differentiable between these 2 types of influenza A viruses.
NP not only displays a clear boundary between human and avian viruses from histogram analysis but also contains more species-associated amino acid signatures (15 of 52) than other ORFs. In addition to NP, polymerase proteins PB2, PB1, and PA also contain abundant species-associated signatures. Most signatures in these viral RNPs are located on the functional domains related to RNP-RNP interactions that are necessary to form replicase/transcriptase complex (3P and NP), which suggests that specific combinations of polymerase complex and NP would allow an influenza virus to replicate itself efficiently (). In addition to RNA-interacting domains, many species-associated amino acid signatures of 3P and NP are located in regions related to nuclear localization signals. Influenza viral replication is highly dependent on nuclear function (35
), making it worthwhile to further examine the roles of those amino acid signatures on nuclear localization of viral RNP in avian versus human cells. We also noticed that several amino acid signatures in NP are located in the regions that interact with cellular proteins, such as splicing factor (BAT1/UAP56) or MxA, which plays a certain role in cellular antiviral mechanisms. What species-specific host factors may affect influenza viral replication rates is not clear. Biologic experiments are required for further understanding the roles of those amino acid residues and related functional domains in the mechanism of interspecies infection.
PB1-F2 is a novel influenza viral protein translated from alternative initiation of PB1 gene. PB1-F2 of PR8 (H1N1) has been shown to target mitochondria and then trigger host cell apoptosis (36
). Our previous research has found that several strains contain truncated PB1-F2 (37
). In this study, 379 of 401 PB1 sequences (in the primary dataset) contained PB1-F2 >87 and <90 aa. For the other 22 sequences, 2 H3N2 strains missed a start codon, 3 H3N2 had the translation stopped at 11 aa, 1 H9N2 stopped at 8 aa, 5 H1N1 stopped at 57 aa, and 3 H9N2 and 7 H3N2 stopped at 79 aa. One H5N1 contained extra residues; its PB1-F2 was 101 aa. We also noted 5 species-associated signatures on PB1-F2; all of them are within the C-terminal domain, which is important for mitochondria targeting (15,16
). Further investigation of the mitochondria localization of those PB1-F2 variants and their abilities for triggering apoptosis in cells derived from different species is warranted.
How many mutations would make an avian virus capable of infecting humans efficiently, or how many mutations would render an influenza virus a pandemic strain, is difficult to predict. We have examined sequences from the 1918 strain, which is the only pandemic influenza virus that could be entirely derived from avian strains. Of the 52 species-associated positions, 16 have residues typical for human strains; the others remained as avian signatures. The result supports the hypothesis that the 1918 pandemic virus is more closely related to the avian influenza A virus than are other human influenza viruses (2
). From the 21 avian viruses isolated from humans in this study, we found 19 (90.5%) that contain >1 change at the species-associated sites. Upon examining signature changes from similarly sized sets of randomly selected human viruses, randomly selected avian viruses, and randomly selected viruses (avian plus human), we found 29.4%, 71.4%, and 47.1%, respectively, contain species-associated mutations. Although predicting the emergence of a pandemic strain is difficult, close monitoring of how those species-associated signature positions have changed from bird-specific to human-specific signatures may provide a measurement for the prediction of such events.