Each tiled prototype sequence was selected to have an intermediate level of sequence homology across a group of microbial or viral strains, which allowed for efficient hybridization and unique identification of most or all subtypes of targeted pathogenic species. For each relevant base of a given prototype sequence, the array contains eight 25mer probes (4 sense and 4 antisense). Two of 8 probes represent perfect matches, while the others correspond to possible mismatches at the central (13th) position of the 25mers. The prototype regions targeting influenza viruses were composed of partial sequences from HA genes of influenza A virus subtypes (H1, H3, and H5) and influenza B virus, NA genes of influenza A virus subtypes (N1 and N2) and influenza B virus, and the M genes of influenza A virus (full-length M1 and partial M2) and influenza B virus (Table 1). Both HA and NA regions encompassed a sufficient number of polymorphic sites to define subtypes. These regions were combined with prototype sequences for 22 other pathogens and tiled on 12.8-μm chips (Affymetrix Inc., Santa Clara, CA, USA), which contain ≈240 K 25mer probes and have the capacity to resolve 30,000 nucleotides. The design and content of RPM version 1 array have been previously described (14).

The influenza clinical specimens used in this study were collected through the Department of Defense Global Emerging Infections System during the 2004–2005 influenza season. Influenza throat swab specimens were collected in accordance with the case criteria previously described (15). Throat swabs were obtained within the first 72 h of the onset of symptoms, placed in viral transport medium (MicroTest M4, Remel Inc., Lenexa, KS, USA), and delivered by commercial carrier to the Air Force Institute for Operational Health in Brooks City Base, San Antonio, Texas, for culturing and molecular characterization. Specimens were passaged once through primary rhesus monkey kidney tissue culture (BioWhittaker, Walkersville, MD, USA). Cultures were tested for influenza A or B viruses by using the centrifugation-enhanced shell-vial technique with monoclonal antibody detection as previously described (16). Cultures testing positive for influenza A or B viruses were confirmed by using reverse transcription–polymerase chain reaction (RT-PCR) analysis with previously reported protocols (16,17). Total nucleic acids were extracted from 90-μL cultured samples or aliquots of live trivalent nasally administered influenza vaccine (FluMist 2004/05, MedImmune Inc., Gaithersburg, MD, USA) by using the MasterPure DNA purification kit (Epicentre Technologies, Madison, WI, USA) and dissolved in 30 μL of nuclease-free water.

Total RNA amplification from cultured samples using a random (RT-PCR) protocol was performed as previously described (18) with minor modifications. Briefly, 4 μL of total nucleic acids were reverse transcribed with 40 pmol of primer D (5´-GTTTCCCAGTAGGTCTCNNNNNNNNN-3´) and Superscript III reverse transcriptase (Invitrogen, Carlsbad, CA, USA) in a 20-μL volume at 42°C for 1 h, followed by heat denaturation at 70°C for 15 min. Aliquots (10 μL) of the RT reaction products were then amplified with the TaqPlus Long PCR System (Stratagene, La Jolla, CA, USA) for 35 cycles in a 50-μL volume consisting of 5 μL of 10× low salt buffer, 5 μL (25 mmol/L) MgCl₂, 2 μL (10 mmol/L) dNTP mix, 1 μL primer E (100 mmol/L) (5´-GTTTCCCAGTAGGTCTC-3´), and 0.5 μL TaqPlus Long polymerase (5 U/μL). Each cycle consisted of 94°C for 30 s, 40°C for 30 s, 50°C for 30 s, and 72°C for 2 min. This was followed by a final extension at 72°C for 7 min. After amplification, the PCR products were purified with the QIAquick PCR Purification Kit (Qiagen, Valencia, CA, USA) and eluted in 45 μL of EB buffer (10 mmol/L Tris, pH 8.5).

Purified DNA amplicons were adjusted to 2 μg in 35 μL of EB buffer, mixed with 15.1 μL of fragmentation cocktail buffer (5 μL NEB buffer 4, 5 μL 10 mmol/L Tris, pH 7.8, and 0.1 μL GeneChip fragmentation reagent [3 U/μL], Affymetrix Inc.), and incubated for 10 min at 37°C and 15 min at 95°C. The fragmented products were then biotin labeled with 1.5 μL of Biotin-N6-ddATP (PerkinElmer Life and Analytical Sciences, Boston, MA, USA) and 1 μL of terminal transferase (20 U/μL) (New England Biolabs, Beverly, MA, USA) for 45 min at 37°C and 15 min at 95°C. RPM version 1 arrays were prehybridized with 200 μL of prehybridization buffer (10 mmol/L Tris, pH 7.8, and 0.01% Tween 20) for 15 min at 45°C. After the prehybridization step, 167.5 μL of hybridization cocktail master mix (3 mol/L tetramethylammonium chloride, 10 mmol/L Tris, pH 7.8, 0.01% Tween 20, 0.5 mg/mL bovine serum albumin, 0.1 mg/mL herring sperm DNA [Promega, Madison, WI, USA], 50 pmol/L Oligo B2 [Affymetrix Inc.]) and biotin-labeled DNA fragments were heated for 5 min at 95°C, equilibrated for 5 min at 45°C, and added to RPM version 1. All hybridizations were incubated for 16 h at 45°C in the GeneChip hybridization oven 640 at 60 revolutions per minute. The microarrays were then washed and stained with the GeneChip Fluidics Station 450 and scanned with the GeneChip Scanner 300 according to the GeneChip CustomSeq array protocol.

The hybridization intensities were analyzed with the GeneChip operating software to generate raw image files (.DAT) and simplified image files (.CEL) with intensities assigned to each of the corresponding probe positions. GeneChip DNA analysis software version 3.0 (GDAS), which implements the ABACUS algorithm (7), was used to produce an estimate of corrected base calls file (.CHP). Base calls generated from each tiled region of the array were then exported from GDAS as Federal Acquisition Streamlining Act (FASTA)–formatted sequences.

Automated DNA sequencing was performed as previously described (17). HA nucleotide sequences for influenza strains used in this study are available at GenBank (accession nos. DQ265706–DG265730). The nucleotide sequences of primers used for amplification and sequencing are available upon request.

DNA sequences generated from RPM version 1 were searched against the Influenza Sequence Database (http://www.flu.lanl.gov/) (19) by using the BLAST algorithm (20). Advanced options for blastn search were set as follows: –W (word size) 7, –r (reward for a nucleotide match) 1, –q (penalty for a nucleotide mismatch) –1. These parameters were chosen to maximize sensitivity and allow sequences with as many as 50% ambiguous calls to still produce full-length searches. Sequence alignments were performed with the ClustalX program (ftp://ftp-igbmc.u-strasbg.fr/pub/ClustalX/).

To assess the performance of RPM version 1 with a real-world clinical isolate set, we tested 25 cultured strains collected from 4 continents during the 2004–2005 influenza season and previously diagnosed by culture and RT-PCR as influenza. One influenza subtype was identified in each tested sample based on the RPM version 1 hybridization profiles and sequence reads shown in Figure 1A–C and E. DNA fragments of HA1, NA1, and M genes randomly amplified from an H1N1 isolate specifically hybridized to their corresponding prototype regions on RPM version 1 (Figure 1A). Prototype regions of 1 influenza subtype exhibited no interference from other subtypes (Figure 1A–C), and prototype regions of other pathogens on RPM version 1 showed no cross-hybridization with any influenza virus segments (Figure 1D). Of the 25 isolates tested, we identified 12 A/H3N2, 12 influenza B, and 1 A/H1N1 (Table 2). The A/H1N1 and A/H3N2 subtypes effectively hybridized to the same prototype M sequence (derived from the A/NWS/33 H1N1 strain), confirming that M genes are conserved among different H/N subtypes of influenza A to allow the universal identification of influenza A subtypes with a single tiled prototype region. A computational hybridization simulation model we developed confirms this suggestion (A. Malanoski, unpub. data). Aside from the highly conserved matrix region, no cross-hybridization was observed between subtypes, which suggests that the more variable HA and NA tiles are subtype specific. The GDAS generated DNA sequences from 3 genes (HA, NA, and M) from each sample with 42%–92% of the prototype tiled sequences, resulting in unambiguous calls by the microarray (Table 2). To demonstrate the accuracy of microarray resequencing reads, the HA genes from all 25 samples were amplified by a specifically primed RT-PCR and subjected to conventional sequencing. The sequences produced by random amplification and RPM version 1 were identical to those identified by the conventional sequencing method with the exception of ambiguous base calls (Ns). That is, in cases where both methods assigned a base identity at a particular sequence position, those assignments were always identical (data not shown).

Microarray resequencing data and conventional sequencing data were searched by using the Influenza Sequence Database with the BLAST algorithm. Results for the highest bit scores were taken as strain identifications and are shown in Table 2.

Based on sequences of HA genes, which are routinely used for genetic and antigenic characterization, microarray strain identifications of all 13 influenza A isolates correlated with identifications from the conventional sequencing method. Although A/H3N2 isolates were sometimes matched with different specific strain sequences from the Influenza Sequence Database based on the top BLAST hits for each isolate, all were redundant representatives of the same A/Fujian/411/02 lineage identified by conventional sequencing. These results indicate that ambiguous calls (Ns) did not affect the accuracy of BLAST identification. At most, only 6 mismatches occurred between the actual sequence of each isolate and sequence of its top BLAST search hit (Table 2, column M1).

Alignment of the HA peptide sequences translated from RPM version 1–obtained DNA sequences for 12 A/H3N2 isolates (Figure 2) showed that they all shared signature Fujian-like lineage amino acid substitutions (threonine and histidine) at positions 155 and 156 (17). Serine (position 227), which is located within antibody binding site D, was also conserved in these isolates, distinguishing them from the A/California/7/04 strain, which has proline at this position (17). In addition, isolate A/Ecuador/1968/04 shared similar amino acids with those observed in the A/Fujian/411/02 strain at antigenic sites A (lysine, position 145) and B (serine, position 189). Because of a more limited collection of NA and M gene sequences in the Influenza Sequence Database, strain identifications based on these 2 genes could only place them into clade A strains of H3N2 influenza A viruses sampled from New York State, which caused the A/Fujian/411/2002-like epidemic of the 2003–2004 influenza season (data not shown) (21). Although the only tiled M sequence was adopted from an A/H1N1 strain (A/NW/33), M results generated from the H3N2 isolates were still clearly identifiable as belonging to the A/H3N2 subtype and more specifically to the Fujian-like strain. The A/England/400/05 isolate was the only isolate appropriately identified as A/H1N1, and all 3 sequences (HA, NA, and M) generated from RPM version 1 and conventional sequencing for this isolate matched A/New York/227/2003 (H1N1). This is an A/New Caledonia/20/99-like strain that has been consistently circulating globally since 1999 (16).

The 12 influenza B isolates were classified as belonging to 2 distinct subgroups based on BLAST searches of 3 genes generated from RPM version 1 analysis. The top BLAST hits for the RPM-obtained sequences of the HA gene identified subgroup 1 isolates as either B/Milano/66/04 or B/Texas/3/2002, both of which are B/Shanghai/361/2002-like strains and belong to the B/Yamagata/16/88 lineage. BLAST queries of conventional sequencing data yielded similar identifications for these isolates. Subgroup 2 isolates were identified as B/Tehran/80/02 by microarray and as B/New York/1/2002 by conventional sequencing. The query results of both methods were similar (different identification can be attributed to ambiguous base calls), and all isolates were members of the B/Victoria/2/87 lineage. This lineage is not covered by the 2004–2005 influenza B vaccine (L. Daum, pers. comm.). These results correspond to a Centers for Disease Control and Prevention (Atlanta, GA, USA) influenza activity report documenting that both of the identified influenza B lineages were reported worldwide and that the Yamagata lineage viruses predominated in the 2004–2005 influenza season (22).

RPM version 1 can differentiate a broad number of variants based on a single-tiled "prototype" probe region without relying on predetermined hybridization patterns (9). A number of nucleotide mismatches that distinguished tested isolates from tiled prototype probe sequences were identified in each sample (Table 2, column M2). Some were unique with respect to existing influenza database-recorded sequences. All of these polymorphisms were verified by conventional sequencing (Table 3). Analysis of HA sequences generated from 12 A/H3N2 isolates by RPM version 1 showed that 4 of these nucleotide variations are common to 11 of the samples, excluding the outlying A/Ecuador/1968/04 isolate. Two of these common base substitutions, 313 G→A and 352 A→C, are at the third nucleotide of their respective codons and represent synonymous mutations. Such mutations do not code for amino acid changes and are usually selectively neutral and much more likely to be shared by common ancestry than by parallel evolution. These facts strongly support phylogenetic grouping of these 11 strains (Figure 3). In contrast, 393 A→T and 483 G→A are nonsynonymous mutations and code for critical amino acid changes. Analysis of conventional sequencing data confirmed that these 2 positions are in the antigenic site B and that the affected amino acids were changed from tyrosine to phenylalanine and from serine to asparagine, respectively. These 2 substitutions are both characteristic features of the A/California/7/04 strain that distinguish this group at both sequence and antigenic levels from other Fujian-like strains. The identified polymorphisms show that 11 of the 12 A/H3N2 isolates, although collected from 4 continents, are members of the same A/California/7/04 lineage, while the lone outlier, A/Ecuador/1968/04, is clearly identified as a member of the older A/Fujian/411/02 lineage. These observations demonstrate that RPM version 1 data can be effectively used for molecular epidemiologic tracking.

Nearly every isolate was shown to have unique base mutations, many of which resulted in amino acid substitutions. Identification of these mutations reaffirms common knowledge that genetic drift is a frequent event during circulation of influenza viruses and that the RPM version 1 gene chip is an effective tool for tracking unique genetic changes within influenza strains.

To test the capability of RPM version 1 to detect multiple pathogens with the random amplification protocol, we analyzed total nucleic acid isolated from trivalent FluMist intranasal vaccine. Figure 1D shows that 8 tiled influenza sequences on RPM version 1 were strongly hybridized by randomly amplified FluMist nucleic acids, and the resulting sequence data confirmed that FluMist includes immunogenic surface protein (HA and NA) genes from influenza H1N1, H3N2, and influenza B strains. Sequence analysis showed that these antigen-encoding genes matched those of 3 wild-type influenza strains recommended by WHO for making vaccine for the 2004–2005 season (Table 4). Two types of M genes from FluMist were identified by RPM version 1 as those in the cold-adapted Ann Arbor strains of influenza A and B, both of which are essential components in the cold-adapted master donor virus vaccine strain (23).