The aim of the study was to detect influenza A viruses isolated over a long period of time, from a range of hosts (humans, birds, horses, and pigs) and with a minimum of sequences tiled on the array. Four genes including PB2, M, HA and NA were chosen in order to type and subtype influenza viruses. Different strategies for selecting the region to resequence were adopted depending on the particular gene. Characteristics of the sequences tiled on the microarray are detailed in Table 1. Firstly, one partially conserved sequence of 218 bp covering the majority of influenza M segments, routinely employed in medical diagnostics, was used for genus identification, namely Influenzavirus A, B and C. Secondly, almost one hundred sequences of PB2 genes covering a large spectrum of influenza A viruses and hosts were selected from the GenBank database. Five typical PB2 sequences representative of the diversity of PB2 sequences over a long period of time and for different host reservoirs were finally retained. The five selected sequences are described in Table 1. The strategy adopted for the hemagglutinin (HA) and the neuraminidase (NA) genes involved the design of consensus sequences for the H1, H2, H3, H5, H7 and H9 genes and the N1, N2 and N7 genes. We have restricted the number of studied molecular subtypes of HA and NA in order to make the confirmation of our hypothesis more straightforward and demonstrative without aiming to cover the whole set of serotypes. The HA and NA subtypes chosen to be tiled on the microarray are shared by all the human viruses and highly pathogenic avian viruses, and the HAs are further distributed into phylogenetic groups that gather all the 16 serotypes 32. For each major HA and NA gene, sequences of influenza A virus were selected from different hosts including horses, pigs, humans and birds.

For each gene, a global alignment including all selected sequences was performed and clusters grouping a maximum number of sequences were established. In each sequence cluster, a common region with the lowest genetic divergence was identified. A consensus sequence was automatically obtained from this common region, characterised by a minimum size of 200 bases and a divergence between the consensus sequence and any given sequence belonging to the cluster of less than 15%. Twelve HA consensus sequences and 9 NA consensus sequences were identified and their positions are detailed in Table 1. Finally, the microarray based on the resequencing approach included 27 influenza viral sequences which corresponded to 7594 bases. For each relevant base of a given consensus or prototype sequence, the array contained eight 25 mer probes (4 sense and 4 antisense). Two of the eight probes represented perfect matches, while the others corresponded to possible mismatches at the central (13^th) position of the 25 mers. All together, 60752 probes were then tiled on the microarray, allowing subtyping of influenza viruses from different host species.

A total of 12 influenza virus samples were analysed by the resequencing microarray, including 2 swine, 3 human, 4 equine and 3 avian strains that originated from various locations worldwide and covered the period of time between 1956 and 2007. All samples were viral isolates, propagated either in embryonated chicken eggs or in MDCK cell cultures. They were chosen from sufficiently distant points in time to cover a large diversity of influenza viruses, especially for human and swine viruses, which were subject to major antigenic drift over the years. Avian strains were derived from cloacal samples collected from ducks in Baie de Somme, Marquenterre (France) in 2007. RNA extracts were amplified by WTA (Whole Transcriptome Amplification) and hybridised on the resequencing microarray (see Methods section). Results are summarised in Table 2. Nucleotide sequences were determined from the fluorescent signals as previously described 33 and compared with sequences in GenBank by BLASTN analysis, to identify and subtype influenza virus strains. Results for the highest hit scores were taken as strain identification. After hybridisation of WTA amplified material on the DNA microarray, all samples were positive for the presence of influenza A virus with a minimum call rate of 41% for the M gene and a call rate ranging from 35.5 to 97.4% for the PB2 gene, which is the percentage of bases determined by the resequencing algorithm. Subtypes of the different strains were identified with HA and NA sequences, giving a call rate ranging from 48.5% to 100%. Time ranges were obtained from BLASTN analysis. The sequences showing the highest alignment score were considered and duplicated sequences representing different isolates of the same strain were eliminated. The time range indicated the oldest isolate with high homology provided by the BLASTN algorithm and the most recent one. The median was calculated based upon the sequences showing the highest alignment scores (see Methods section).

For swine and human viruses, a minimum call rate of 35.5% allowed identification of influenza virus with a correct temporality. Indeed, for the majority of strains the BLASTN result was close to that of the sample strain, as outlined by median analysis. The median obtained for the NA gene from the A/Johannesburg/33/94 (H3N2) virus was relatively recent compared to the year of isolation of the strain, probably because the chosen consensus sequence for this gene was highly conserved and thus weakly discriminatory. The time range obtained for the A/Wyoming/3/2003 (H3N2) virus was narrow and limited to the year 2000, probably due to the marked antigenic drift from the year 2000 observed for the H3 gene in human influenza virus strains 34.

The epidemiology of swine influenza has become increasingly complex over the last decades. Three major influenza A virus subtypes are currently circulating in swine worldwide, but the origins and the antigenic and genetic characteristics of these swine influenza viruses differ with continents or regions of the world. A/swine/Gent/7625/99 (H1N2) virus is the result of a multiple reassortment involving a swine influenza virus with avian-influenza-like internal segments and human H1N1 and H3N2 viruses isolated in 1994 35. BLASTN analysis for H1N2 swine virus reflected this multiple reassortment with blasted sequences originating from human, avian and swine hosts (data not shown). Between 1983 and 1985, swine H3N2 viruses appeared in Italy. They contained avian-like internal genes and HA and NA genes from human H3N2 viruses 36. In BLASTN analysis, the vast majority of the sequences sorted by the algorithm for glycoprotein segments were from human or swine origins only. All equine influenza virus strains were identified and subtyped with the DNA microarray, with a minimum call rate of 48.5%. Medians were often heterogeneous and far from the original strain. This is probably due to (i) the paucity of equine influenza virus sequences available in the databases, (ii) the limited genetic evolution of those viruses leading to equivalent result scores after BLASTN analysis, (iii) the fact that HA and NA sequences tiled on the microarray were shorter than PB2 sequences, due to their extreme genetic diversity in influenza A viruses. For the H3N8 viruses, recent sequences corresponded to avian strains, outlining the fact that H3N8 equine viruses were reassortants between avian viruses and equine H7N7 viruses which have not been isolated in horses for more than 20 years.

All together, the results showed that the PathogenID v2.0 DNA microarray is able not only to provide correct full subtyping of various strains from different host origins but also to identify/detect, after BLASTN, an array of strains genetically related to them. In all cases, the strains output by BLASTN belonged to the same time period thus giving a strong insight into the variant to which it is genetically, and very probably antigenically, related.

Cloacal samples were collected from ducks in Baie de Somme, Marquenterre (France) in 2007 and analysed by qRT-PCR specific for the M gene. WTA amplification of viral RNA followed by hybridisation to the DNA microarray confirmed the identification of influenza viruses. Two samples were identified as H1N1 viruses and one sample as H3N2 virus. Influenza virus strain 221 was identified as H3N2 virus with a call rate of 84.7% for one of the H3 consensus sequences and 99.4% for one of the N2 consensus sequences. Influenza virus strains 223 and 224 were both identified as H1N1 virus with call rates of 90.7% and 94.9% for the N1 gene and call rates of 66.5% and 81.8% for the H1 gene, respectively (see Table 2). In order to confirm subtyping of the different viruses, primers were designed based upon resequencing sequences reconstructed by the microarray and used for RT-PCR (see additional file 1). Amplified products were cloned and analysed by classical sequencing. Sequencing results confirmed the subtyping by the DNA microarray of the 3 viral strains isolated from cloacal samples. Time ranges showed that the viral strains isolated in the environment are recent, except for the H1 gene of strain 223 (Table 2). The evolutionary rate of influenza viruses in the natural avian host is believed to be slower than in mammals, and this could explain the above result. Sequences were analysed with the BLASTN program and results with highest scores are represented in Table 3. The majority of the "classical" sequences were identical to those found after BLASTN analysis of sequences generated by the DNA microarray. The level of sequence information generated by the DNA resequencing microarray is sufficient and equivalent to that obtained by classical sequencing.

From many available influenza virus sequences, gene-specific databases containing only the sequences of a specific gene or subtype (e.g., PB2 or H1 subtype) were created and converted to the FASTA format. Then, a multiple-sequence alignment was performed with the data set with BioNumerics software for windows (version 5.1, Applied Maths). Additionally, a neighbor-joining phylogenetic tree was created. Consensus sequences were then designed based upon global alignments of this large number of influenza virus sequences available in databanks.

Influenza virus samples were all viral isolates propagated either in embryonated chicken eggs or in MDCK cell cultures. The virus type was determined by hemagglutination inhibition assay 39.

Cloacal samples collected from ducks in Baie de Somme, Marquenterre (France) were inoculated via the allantoic route on 11-day-old embryonated chicken eggs. Eggs were incubated at 35°C and allantoic fluids were harvested on day 3 postinfection. Virus titration of the fluids was performed by hemagglutination by using guinea pig erythrocytes. Influenza virus isolation was confirmed by qRT-PCR specific for the M gene.

Viral RNA from influenza samples and allantoic fluids were extracted using QIAamp Viral RNA Mini Kit (Qiagen) according to the manufacturer's instructions. Nucleic acid amplification was performed by WTA as described previously 14.

DNA amounts obtained after amplification were quantified by Quantit (Invitrogen(r)). Five micrograms of DNA was fragmented (GeneChip(r) Resequencing Assay Kit, Affymetrix Inc.) and labelled according to the GeneChip(r) Mapping 100K Assay Manual (Affymetrix Inc.). Microarray hybridisation was conducted at 45°C and array processing was carried out according to the protocol recommended by the manufacturer (Affymetrix Inc.) as previously described 40.

After hybridisation to the DNA microarray, the nucleotide sequences of the different genes were determined from the signals as described previously 33 and compared with sequences in GenBank by BLASTN analysis.

All sequences generated by the microarray were characterised by the call rate, which is the percentage of bases determined by the resequencing algorithm. For determining the time of emergence of the sample strain, median analysis was performed as described: the sequence generated by the microarray showing the highest call rate value was blasted against sequences available in Genbank using the BLASTN 2.2.24+ program 41. All homologous sequences within the group of sequences showing the highest score (as categorised by the BLASTN program) were selected and their year of circulation used for median calculation and determination of the time range. If the number of these sequences was higher than 100, median analysis was performed on the 100 first sequences only. For avian influenza strains, median analysis was performed as described above and the first sequence(s) showing the highest scores were also detailed in Table 3.

After identification and subtyping of avian influenza strains by the microarray, oligonucleotides specific for each tiled influenza virus gene were designed and used in RT-PCR assays. Five microliters of extracted RNA was amplified by SuperScript(tm) One-Step RT-PCR system with Platinum(r) Taq DNA polymerase (Invitrogen, Life Technologies, Carlsbad, CA). One-Step RT-PCR was performed with 25 μl reaction volume. The extracted RNA sample (5 μl) was added to RT-PCR mixture (20 μl) consisting of 1X Reaction Mix (a buffer containing 0.4 mM of each dNTP, 2.4 mM MgSO₄), 0.5 μl of SuperScript(tm) II RT/Platinum(r) Taq Mix, 0.4 μM primer Forward, 0.4 μM primer Reverse, 0.2 U/μl of RNaseOUT Recombinant Ribonuclease Inhibitor and nuclease-free water. The RT and PCR were carried out with following steps: RT at 45°C for 30 min; 55°C for 15 min, 94°C for 5 min, then PCR cycle 1-30, 94°C for 30 sec, 45°C for 30 sec, 72°C for 30 sec; final extension, 72°C for 5 min.

Amplified products were extracted using QIAquick Gel Extraction kit (Qiagen) according to the manufacturer's instructions and cloned in pCR^(r)II.1-TOPO^(r) TA cloning vector (Invitrogen). Sequences of influenza virus genes were obtained from 3 individual clones for each gene.