Main

The genomes reported here comprise the initial results from the Influenza Genome Sequencing Project, a partnership between the US National Institute of Allergy and Infectious Diseases and collaborators from around the world4, whose goal is to sequence the genomes of thousands of influenza virus isolates. This study is the first, to our knowledge, to attempt to sequence strains that were not pre-selected for particular virulence or other unusual characteristics, and should therefore provide a relatively unbiased view of influenza virus strains in the population. Here we focus on a collection from New York State spanning several years, and subsequent studies will focus on samples from multiple, distant geographical sources across a longer time span. As our analysis shows, even within a geographically constrained set of isolates, we have found surprising genetic diversity, indicating that the reservoir of influenza A strains in the human population—and the concomitant potential for segment exchange between strains—may be greater than was previously suspected.

The genome of the influenza A virus (family Orthomyxoviridae) consists of eight single-stranded negative sense RNA molecules spanning approximately 13.5 kilobases (kb). The segments range in length from 890 to 2,341 nucleotides and encode a total of 11 proteins. Although a large number of partial influenza A virus sequences now exist in the public archives (for example, GenBank), relatively few complete genomes are available. In part this is due to the technical difficulty of constructing an efficient sequencing pipeline for an RNA-based organism. The bulk of the public data on influenza comprises short fragments from the haemagglutinin (HA) or neuraminidase (NA) segments of the genome, which encode the two main surface proteins and which, it is widely believed, are the source of most of the antigenic variation in the virus. As a result of this project, the number of complete human H3N2 influenza virus genomes in GenBank has already grown from just seven genomes to over 200.

We have completely sequenced all eight segments from 207 H3N2 isolates and two H1N2 isolates. In total, the finished sequence covers 2,821,103 bases, with an average of 13,498 bases per isolate. Table 1 shows the sequencing results for all isolates, broken down by segment. The polymerase chain reaction with reverse transcription (RT–PCR)-based sequencing strategy produces an average of 5.6 sequencing reads covering each nucleotide, as shown in Table 1. The average Phred quality value5 was 33, which at 5.6-fold coverage corresponds to an error rate of 3.2 × 10-19; however, regions of low coverage will have higher error rates. These regions can be inspected at the NCBI Assembly Archive6, which displays the raw data underlying every nucleotide in each genome. Note that the error rate in these genomes is likely to be considerably lower than previously sequenced influenza isolates, which in many cases reflect single-pass or two-pass sequencing. Assembly was performed using the Minimus assembler program7 followed by the AutoEditor program to correct erroneous bases8. Details of assembly and annotation are provided in the Supplementary Methods.

Table 1 Sequencing results for 209 complete genomes of H3N2 and H1N2 human influenza A viruses
Full size table

This is the first large-scale analysis of influenza isolates collected in a relatively unbiased manner, allowing a comprehensive look at an influenza virus population across several seasons within a constrained geographical area. Among RNA viruses, only human immunodeficiency virus (HIV) has been subjected to similar whole-genome analysis9,10. In this first set of 209 genomes, we have observed multiple, novel mutational events, including point mutations, deletions and segment exchange. By carefully cataloguing these events, we can begin to get the first real picture of the rate of mutational events underlying influenza A virus evolution.

Although extensive previous research has catalogued changes in the HA and NA segments, we found new mutations in these segments as well as multiple changes in the other six segments. Some of these changes are shown in Fig. 1 and discussed below, and a comprehensive list of amino acid mutations in all eight segments is provided in the Supplementary Data.

Figure 1: Sites with genetic changes across the ten main proteins in 207 influenza A viruses.
figure 1

Each row represents a single amino acid position in one protein. Amino acids (single-letter abbreviations are used) are colour-coded as shown in the key, so that mutations can be seen as changes in colour when scanning from left to right along a row. For simplicity, only amino acids that showed changes in at least three isolates are shown. Each column represents a single isolate, and columns are only a few pixels wide in order to display all 207 H3N2 isolates in this figure. Isolates are ordered along the columns chronologically according to the date of collection; boundaries between influenza seasons are indicated by gaps between columns. A more detailed version of this figure, showing positions that experienced any amino acid change and showing identifiers for the isolates in each column, is available as Supplementary Fig. 1.

Full size image

Perhaps the most dramatic finding in our data is the discovery of an epidemiologically significant reassortment that explains the appearance, during the 2003–2004 season, of the ‘Fujian/411/2002’-like strain, for which the existing vaccine had limited effectiveness. In a recent paper11, we described how phylogenetic analysis of 156 H3N2 genomes from our project revealed the clear presence of multiple, distinct clades circulating in the population. Through a reassortment event, a minor clade provided the haemagglutinin gene that later became part of the dominant strain after the 2002–2003 season. Two of our samples, A/New York/269/2003 (H3N2) and A/New York/32/2003 (H3N2), show that this minor clade continued to circulate in the 2003–2004 season, when most other isolates were reassortants. In these samples (Fig. 1, see columns for November 2003) the HA segment is clearly similar to the dominant clade, whereas the other segments all show numerous differences.

This finding illustrates not only that the influenza virus population contains multiple lineages at any given time, but also that alternate, minor lineages can contribute genetic variation to the dominant lineage, resulting in epidemiologically significant, antigenically novel strains. It is worth emphasizing that our sequence-based sampling approach—in contrast to traditional serologically based sampling—will reveal co-circulating strains even before they become antigenically novel.

Figure 1 illustrates five seasons' worth of mutations in all proteins from the 207 H3N2 influenza virus isolates included in this study. For clarity, amino acid positions are shown only if they underwent genetic changes in at least three isolates. Each mutation is indicated by a colour shift along a row in the figure. For example, the first row shows that the amino acid in position 5 of HA1 mutated from glycine (G, shown in light green) to valine (V, shown in burgundy) in November 1999, and then back to glycine in November 2001 and afterwards, except for three isolates in January to February 2002 that show a glutamic acid (E, shown in pink) at that position. In total, 186 positions experienced at least one amino acid change.

As the figure shows, mutations appear both during and between influenza seasons. For example, HA residues 5, 33 and 92 remained unchanged from May 1999 to October 1999, and then mutated in November 1999, leading to a permanent switch for the rest of that season. Notably, multiple changes in the internal segments, including those encoding the polymerase genes (PA, PB1 and PB2), the nucleocapsid protein (NP), and two non-structural proteins (NS1 and NS2), first appeared in the 2001–2002 season and became fixed thereafter.

Data from isolates collected in the spring of 2003 provide a glimpse of the transitional period before a major reassortment event. Many of the HA mutations that became dominant during the 2003–2004 influenza season first appeared in February 2003. Mutations to residues 155 and 156 of the HA1 domain (H155T and Q156H) show up in early 2003; these sites are accessible to antibodies and had an important role in the antigenic mismatch between the vaccine strain and the circulating viruses in the 2003–2004 season12. A different picture emerges in the other proteins, where mutations that appear in February 2003 remain only in a few isolates. This clearly indicates that a reassortment event brought in a new HA segment during or before the spring of 2003, and subsequent data show that this reassortant strain became dominant in the 2003–2004 influenza season11.

A number of important mutations found in our data may affect receptor-binding affinity and potentially increase viral replication efficiency. Studies have determined that changes in HA residues 183, 186 and 226 could affect HA receptor-binding affinity13, and residue positions 131, 222, 225 and 226 are important for efficient replication12. Mutation S186G appears in circulating viruses during the 2001–2002 influenza season, along with mutation V202I, and remains in the 2003–2004 season. Mutations A131T, W222R and G225D also emerge in February 2003. The HA1 T155H and H156Q mutations in our data are accompanied by a possibly correlated mutation at residue 25 (L25I).

The neuraminidase protein has a box-shaped globular head with four catalytic sites that allow the cleavage of sialic acid linkages14. Amino acid positions important for antigenic drift have been identified for the N2 subtype14 as well as other regions likely to be involved in virus–host interactions and qualified as phylogenetically important regions15. Sequence data in our study indicate that once residue 197—an antigenic site14,16—mutated from 197H to 197D early in the 1999–2000 influenza season, it was accompanied by the mutation R249K. This residue is probably not in a functional site but may be functionally compensating by maintaining the accessibility of surface residues. Residue 199 interestingly switched (E199K) for the 2003–2004 influenza season for the majority of the isolates, except for the two isolates corresponding to the minor non-reassorted clade (A/New York/269/2003 and A/New York/32/2003)11.

Table 2 lists correlated mutations that may be co-mutations; that is, where there appears to be a balancing effect between two sites on the same protein, or between a site on an internal protein and one on a surface protein. The best example is seen for T392 in NA, which is present in the same eight isolates (appearing in 2001–2003) where there is an I463 mutation in PB2 (see Fig. 1).

Table 2 Correlated mutations within and between influenza virus proteins
Full size table

The fact that the minor Fujian-like clade has donated its HA to the previously dominant strain rather than itself becoming the dominant circulating virus indicates that there may be important amino acid co-substitutions in the other proteins essential for viral fitness11. When comparing the NA and internal proteins of the dominant circulating major clade present during the 2003–2004 influenza season with the previous dominant clade, there are a few substitutions that appear to be unique. NA Y40H and E199K, for example, occur in regions potentially affecting virus–host interactions15. On the basis of the NA alignments performed using data from the LANL influenza database (data not shown), the E199K mutation is also seen in southeast Asian isolates that were collected during the 2003–2004 influenza season when the Fujian-like variant was the dominant circulating virus in that region. Over the rest of the NA protein, however, the southeast Asian isolates resemble the non-reassortant North American clade B11. This indicates that this clade may have had deficiencies in certain residues that affected its ability to become the dominant virus. There is, unfortunately, no publicly available data for these proteins from the southeast Asian isolates with which to do a comparative analysis to the North American major and minor clades.

Recent reports17,18,19 have described a newly discovered protein, known as PB1-F2, encoded by a shifted reading frame in the PB1 gene. The data presented here more than double the total number of complete PB1 segments in the public archives, and we found that the PB1-F2 open reading frame is preserved in 206 out of 209 PB1 genes. In most cases (180 out of 206) the protein's length is 90 amino acids, but it is 87 amino acids in 23 isolates and 80 amino acids in three isolates. In three cases, an in-frame stop truncates the predicted protein after 11 amino acids. The translations of PB1-F2 for all 209 isolates are provided in the Supplementary Data.

Our project includes two clear examples of segment exchange between H1N1 and H3N2 viruses, both of which are H1N2 serotypes. In both isolates from our collection, only the haemagglutinin segment was exchanged, and these appear to be descendants of a human–swine recombinant, as has been reported previously20. Although segment exchange has been reported before21,22, no accurate data on the frequency of these events have been collected. Our observation of three events (two exchanges between different H3N2 clades11, and one exchange between H3N2 and H1N1) in 209 samples may provide an initial baseline for future estimates.

The Influenza Genome Sequencing Project is currently being expanded to include avian influenza, in an effort to establish how often these strains cross the species barrier and move into the human population. One possible cause for influenza pandemics is the mixing through reassortment of an avian influenza strain with a human strain via co-infection of a single host23. Recent reports of transmissions of avian influenza virus to humans24,25 have raised concerns that a new pandemic might emerge26. Despite the importance of the threat that influenza presents, no previous effort has been made to study its complete genome on a large scale. The protocols described here are being generalized to include large numbers of avian influenza isolates that, like the genomes reported here, will be deposited immediately in public archives.

Methods

All sequence data used in this study are available from GenBank, and also via a project page at http://www.tigr.org/flu. In addition, all 209 genomes and GenBank accession numbers are available as a single file in the Supplementary Data.

All samples for this study were collected by the Virus Reference and Surveillance Laboratory of the Wadsworth Center in Albany, New York, which maintains a repository of human influenza samples dating back to 1992. Virus samples were received as part of outbreak investigations, through the reference function of the laboratory, and, since 2001, as part of a sentinel physician influenza programme. Use of the diagnostic samples in this study was approved by the New York State Department of Health Institutional Review Board.

Viral RNA isolation

Isolates were amplified in tube cultures of primary rhesus monkey kidney (pRhMK) cells before extracting 140 µl of culture supernatant. Viral RNA was extracted from clarified supernatant fluid using the Qiagen BioRobot M48 workstation with the MagAttract Viral RNA M48 kit (Qiagen).

RNA ligation

RNA was circularized overnight at 4 °C with T4 RNA ligase (Epicentre). Before the ligation step, the RNA was first treated with tobacco acid pyrophosphatase (20 U TAP in a 15-µl reaction, incubated at 37 °C for 1 h). TAP treatment is usually used to remove molecules from the 5′ end of RNA, mostly plus-strand RNA. Although no such molecules are expected to be present on the influenza genomic RNA segments, ligation was more efficient with this treatment than without. The circularized RNA was cleaned again with the RNeasy Mini kit (Qiagen).

RT–PCR and sequencing

The first step in the high-throughput sequencing pipeline uses reverse transcription followed by polymerase chain reaction amplification to generate overlapping DNA amplicons covering each segment of the influenza virus genome. Overlapping primers were designed approximately every 200–250 nucleotides along the genome; degenerate primers allow the pipeline to tolerate sequence variation. In order to capture the extreme ends of each segment, we used an RNA circularization step before the RT–PCR27. We then used RT–PCR to amplify a chimaeric product that contained the sequence from both ends of the segment.

Complementary DNA synthesis

RT–PCRs were performed with a OneStep RT–PCR kit (Qiagen). Ninety-five reactions were performed per RNA sample. Degenerate primers were designed based upon the alignment of selected human H3N2 sequences. For most of the segments, all full-length and nearly full-length sequences from 1980 to the present were aligned and used for primer design. For others, more stringent criteria were used in order to reduce the number of sequences in the set to a more manageable number. An M13 sequence tag was added to the 5′ end of each primer to be used for sequencing (F primers: TGTAAAACGACGGCCAGT; R primers: CAGGAAACAGCTATGACC). Eight pairs of primers were designed to span the ligated ends of each segment to capture the end sequences. Four of the reactions were analysed on an agarose gel for quality control purposes. Primer sequences are included as a separate table in the Supplementary Data.

Amplicons were prepared for sequencing by incubating them at 37 °C for 60 min with 0.5 U of shrimp alkaline phosphatase (Amersham) and 1 U of exonuclease I (Amersham) to inactivate remaining dNTPs and to digest the single-stranded primers. The enzymes were inactivated by incubation at 72 °C for 15 min.

Sequencing

Sequencing reactions were performed on a standard high-throughput sequencing system using Big Dye Terminator chemistry (Applied Biosystems) with 2 µl of template cDNA. Each amplicon was sequenced from each end using M13 primers (F primer: TGTAAAACGACGGCCAGT; R primer: CAGGAAACAGCTATGACC). Sequencing reactions were analysed on an Applied Biosystems 3730 ABI sequencer. Each influenza isolate was processed on its own 96-well plate to minimize the possibility of sample mix-ups.

Data release

Raw traces were submitted to the NCBI Trace Archive. The finished assembly of each isolate, showing how the traces are aligned to one another and to the finished sequence, was deposited in the NCBI Assembly Archive6, which allows scientists to investigate the data supporting every nucleotide of each genome. An annotation pipeline developed at NCBI (see Supplementary Methods) was run to make gene assignments, and finished genomes with annotation were deposited without delay in GenBank.