INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

After the first reports describing DNA microchip arrays appeared (15, 16), microarray technology revolutionized the study of gene expression patterns in diverse organisms (reviewed in references 3 and 7). Arrays composed of oligonucleotides (15) or robotically spotted DNAs (16) permit genome scale analysis of gene expression patterns and have more recently been used in such applications as drug discovery (6), mutation detection (8, 9, 12), evolutionary studies (11, 21), and genome mapping (20). The application of DNA microarray technology as a diagnostic tool also shows great promise, since microarrays theoretically permit a simultaneous screen for any of tens of thousands of nucleic acid sequences. However, only recently have reports of these array-based applications started to appear. Oligonucleotide arrays have been used to search for mutations in cancer-linked genes (reviewed in reference 10) and human immunodeficiency virus mutations (13, 25) and in bacterial typing (1), but the high cost and limited availability of such tools continue to limit research in this area.

We have been interested in the development of new technologies capable of better identifying viral pathogens and have been using, as a model system, human influenza viruses (28, 29). Influenza viruses cause annually recurrent epidemics of moderate-to-severe respiratory disease, frequently associated with genetic variation termed drift and shift (5, 27). These viruses present an important diagnostic problem, and the rapid detection, typing, and subtyping of influenza A and B virus strains are of both clinical and epidemiological value. While antibody-based methods still form the foundations of routine diagnostic work, many reports over the last decade have demonstrated the utility and superiority of PCR-based diagnostic (2, 4, 14, 19, 22, 26) and retrospective (23) tests. Unfortunately PCR-based methods suffer from a problem in that simply producing DNA isn't sufficient evidence that one has amplified the right product. Other methods such as DNA sequencing (29), blotting (18), and fluorogenic PCR (19) are required if one desires proof that a PCR has amplified a bona fide nucleic acid target. While such confirmatory methods are certainly reliable, they still present something of a financial, technical, and logistical challenge to busy laboratories routinely screening for hundreds of different agents. They also become more difficult to apply when multiplex methods are being used to simultaneously amplify two or more PCR products from a mixture of templates.

DNA arrays offer a potential solution to these problems. The arrays can potentially encode tens of thousands of possible target sequences and thus provide a simple way of storing and indexing numerous hybridization probes. In principle one could draw this resource when needed to confirm the identity of one or more PCR products by hybridization, with only minimal modification to existing PCR-based protocols. We have tested this approach and show here that a model DNA array can be used to type and subtype influenza A and B virus strains. This shows that DNA arrays can provide multiply redundant confirmatory evidence that a PCR product encodes the sequence(s) it is expected to encode.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

Viruses. Human influenza virus strains (Table 1) were obtained from the American Type Culture Collection (Manassas, Va.) and propagated where necessary in 10-day-old embryonated chicken eggs (29). Crude viral RNA was extracted as described previously (29) and stored at 70°C.

Oligonucleotide primer design. Oligonucleotide primers were designed using RightPrimer, version 1.2, software and GenBank Blastn sequence alignments. Selected PCR primers meet the following criteria: (i) primers hybridize to highly conserved sequence elements, (ii) each amplicon spans approximately 500 bp, (iii) amplified segments collectively encompass an entire gene (thus 1.5 kb of virus DNA sequence required three primer pairs), and (iv) primer melting points generally fell between 50 and 54°C. Desalted primers were purchased from Gibco/BRL and used without further purification.

RT-PCR and cDNA cloning. A commercial kit was used to prepare viral cDNAs. Reaction mixtures contained 10 pg of RNA, 200 µM (each) deoxynucleoside triphosphate, 1.5 mM MgCl₂, 0.4 µM (each) primer, and other components as directed by the manufacturer (Titan one-tube reverse transcriptase PCR [RT-PCR] kit [Roche]). Reaction mixtures were incubated at 50°C for 30 min, denatured at 94°C for 3 min, and then subjected to 35 thermal cycles (94°C for 30 s, 45°C for 30 s, 68°C for 2 min). Following a final 10-min incubation at 68°C, the samples were chilled and the products were sized by electrophoresis. PCR-amplified viral cDNAs were subsequently cloned using a Topo TA cloning kit (Invitrogen). Recombinant plasmids were purified from lacZ mutant bacteria and sequenced as described previously (24).

Preparation of target DNAs. M13 forward (5' GTAAAACGACGGCCAGTG 3') and reverse (5' CAGGAAACAGCTATGACC 3') primers were used to reamplify cloned viral cDNAs. The reverse primer incorporated an amine tag linked by a six-carbon spacer to the 5' end (Gibco/BRL). PCR mixtures (100 µl) contained 2 mM MgCl₂, 200 µM (each) deoxynucleoside triphosphate, 0.4 µM (each) primer, ~1 ng of purified plasmid DNA, 1/10-diluted enzyme buffer, and 1.5 U of Taq polymerase (Perkin-Elmer). Following 35 thermal cycles (typically 94°C for 30 s, 52°C for 30 s, 72°C for 1 min) the DNA was purified using MicroSpin S-400 columns (Amersham/Pharmacia Biotech), precipitated with ethanol, resuspended in 30 µl of 3× SSC (10× SSC is 87.6 g of NaCl/liter and 44.1 of sodium citrate/liter, pH 7.0), and the DNA concentration was adjusted to ~300 ng/µl for spotting purposes.

Microarray printing and processing. Amine-tagged target DNAs were distributed, in duplicate, into 384-well microtiter plates. A custom-built arrayer (Virtek) was used to spot DNA on aldehyde-activated silylated microscope slides (CEL Associates). Printed arrays were air dried for a few minutes at 50 to 60°C and then stored overnight at 20 to 37°C over desiccant. The arrays were rehydrated for 4 h in a humid atmosphere, dried briefly at 50°C on a heating block, washed once in 0.2% sodium dodecyl sulfate (SDS) and twice in water (1 min each), and treated with sodium borohydride (1.0 g of NaBH₄ [Sigma] dissolved in 300 ml of phosphate-buffered saline plus 100 ml of ethanol [17]) for 5 min. The DNA was denatured in water (2 min at 95°C) and then washed again (once in 0.2% SDS and once in water [1 min each]). Arrays were air dried and stored at room temperature.

Probe preparation and multiplex PCR. Two methods for incorporating fluoro-linked Cy3- and/or Cy5-dCTP (Amersham/Pharmacia Biotech) into fluorescent probes were devised. The simplest method involved adding 20 µM Cy3- or Cy5-dCTP to a standard 100-µl PCR mixture containing Taq polymerase; 200 µM (each) dATP, dGTP, and TTP plus 100 µM dCTP; and a single primer pair (see above). In subsequent experiments, probes were prepared using multiplex RT-PCR mixtures also supplemented with 20 µM Cy3- or Cy5-dCTP. In this case the primers were combined into three different groups, but in such a way that the combination of primers ensured that one influenza virus subtype could be amplified no matter which viral RNA or cloned cDNA was present. Labeled probes were purified using MicroSpin S-300 columns and heat denatured before hybridization. Fluorescent molecules were handled under dim lighting to minimize photobleaching.

Hybridization and data analysis. Microarrays were prehybridized in 20 µl of DIG Easy Hyb (Roche) containing 5 µg of denatured salmon sperm DNA at 62°C for 1 h under a 12- by 12-mm coverslip. The coverslip was washed off the slide in 0.1× SSC, and the slides were dried at room temperature. A fresh solution of DIG Easy Hyb, containing ~5 µl of denatured fluorescent probe plus 5 µg of salmon sperm DNA in a total volume of 20 µl, was then applied to the array and overlaid with a 12- by 12-mm coverslip. The arrays were incubated overnight at 58 to 62°C in a humid chamber and then washed for 5 min at 20°C in 1× SSC-0.1% SDS followed by 0.1× SSC-0.1% SDS. The arrays were rinsed in 0.1× SSC, dried, and stored in the dark. Arrays were analyzed using GenePix (Axon Instruments) or ChipReader (Virtek) confocal scanners, and the fluorescence was quantitated using ImaGene software (Biodiscovery). Gain settings produced a linear detector response. A "glass" background fluorescence reading was measured in the region surrounding each spot and subtracted, and the intensity was normalized to produce an average fluorescence reading at all nonhomologous spot positions equal to 50 U. The signal-to-background ratios reported here are the fluorescence intensities measured at homologous spots divided by an average measured at all other array locations.

Nucleotide sequence accession numbers. Viral sequences obtained in this study have been assigned GenBank accession no. AF305216 to AF305220.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Viral genetic targets. Three types of human influenza viruses are commonly encountered (A, B, and C), of which the A and B types are of primary clinical interest. Type A strains are further subtyped as encoding one of three different hemagglutinins (HAs; H1, H2, or H3) and one of two different neuraminidases (N1 or N2). The HA and neuraminidase (NA) genes are principal pathogenic determinants that reside on separate subgenomic segments, and it is genetic reassortment of these segments which creates the six primary human influenza A virus subtypes which change over decades. Human type B influenza virus strains also encode HA and NA genes, but only a single major genetic variant of each is commonly encountered. Thus the typing and subtyping of influenza A and B virus strains require an ability to detect at least four HA and three NA genes plus their drifted allelic variants. Other viral genes are also useful in differentiating influenza A and B virus strains and offer the advantage of being more genetically stable. In this regard we have previously used the influenza matrix protein (MP) gene as a PCR target. Methods capable of differentiating a total of four HA, three NA, and two MP gene targets thus provide the capacity to type and subtype human influenza A and B virus strains with some degree of redundancy.

Preparation of cloned virus cDNAs. We initially cloned multiple separate fragments of genes for three influenza A virus HAs (A/HA1, A/HA2, and A/HA3), two influenza A virus NAs (A/NA1 and A/NA2), and an influenza A virus MP (A/MP) as well as an influenza B virus HA (B/HA), NA (B/NA), and MP (B/MP). The nine genes each spanned 1 to 1.5 kb and have little or no sequence homology. Primers were selected using a combination of primer design software and Blastn sequence alignments with the intent of locating each primer pair in maximally conserved sequence regions spaced about 500 bases apart. Table 2 lists the nine DNA sequences used to initiate homology searches and to design primers, and Table 3 shows the 52 primers generated by this approach. Based on available sequence data, the variation in nucleotide sequence within each influenza gene family ranges from 81 to 100% sequence identity (Table 2). These 26 primer pairs were used in RT-PCRs to amplify genes encoded by five different influenza virus strains (Table 1). We obtained 24 of 26 possible cDNA products (the A/HA1-2 and B/MP-2 primer sets did not work), cloned these cDNAs, and verified the DNA sequences.

Array fabrication. The 24 sequence-validated cDNA clones were reamplified using M13 universal primers, in the process adding a 5' amino tag to the reverse primer to permit covalent attachment to a modified glass support. A custom-built Stanford-type DNA arrayer was used to spot DNA, in duplicate, onto activated silylated glass slides at densities of 1,100 or 2,500 spots/cm² (300- or 200-µm spacing, respectively). The array pattern is shown in Table 4. In addition to virus cDNAs we added control spots composed of Escherichia coli DNA and 3× SSC buffer. Several replicate sets of the eight-by-eight grid were also printed elsewhere on the slide to facilitate statistical analysis of hybridization signals.

Hybridization of arrays with cloned cDNAs. To determine whether these arrayed cDNA targets retained their expected hybridization properties, we first tested whether fluorescent probes derived from the original cloned cDNA templates would hybridize to a cognate spot(s). Standard PCR mixtures were supplemented with Cy3-dCTP and used to reamplify each of the cloned cDNAs. These fluorescent probes were purified and individually hybridized to separate arrays, and then the hybridization signals were analyzed using Virtek or Axon chip readers. A selection of the resulting fluorescence images are shown in Fig. 1 (panels 1 to 6). These can be decoded by noting, for example, that a 520-bp A/HA3-1 probe (Table 3) produced strong hybridization signals at duplicate spots located in row 3, columns 1 and 2 (Fig. 1, panel 2). Overall, duplicate hybridization signals were clearly detected at all of the positions known to encode homologous DNA targets and little cross-hybridization to unrelated spots or background fluorescence was detected with these supports.

View larger version (98K): [in this window] [in a new window]   FIG. 1.   Identification of influenza cDNAs using glass-supported microarrays. DNA targets were spotted in duplicate in an eight-by-eight grid, and bound probes were detected using confocal fluorescence microscopy. The hybridization probe or probes used in each experiment are indicated below each image, and the array pattern is shown in Table 4. The images seen in panels 1 to 6 were obtained using short probes spanning mostly a single target (see text for a further discussion of this point), while panel 7 shows an array hybridized to a longer cDNA fragment spanning three contiguous targets (A/HA3-1, A/HA3-2, and A/HA3-3). The single array shown in panels 8 and 9 was hybridized to a mixture of Cy3-labeled A/HA2-1 probes and Cy5-labeled B/NA-1 probes and scanned simultaneously for Cy3 (panel 8) and Cy5 (panel 9) dyes at excitation wavelengths of 543 and 635 nm, respectively. For comparison, panel 10 shows another array hybridized to a mixture of Cy3-labeled A/HA2-1 and B/NA-1 probes. Panels 11 and 12 illustrate the use of multiplex RT-PCR and the ability to detect hybridization of heterologous A/Denver/1/57 and B/Hong Kong/5/72 probes to A/Port Chalmers/1/73 and B/Maryland/1/59 targets.

View larger version (27K): [in this window] [in a new window]   FIG. 2.   Quantitation of hybridization signals. A glass background value was automatically subtracted from all measured fluorescence intensities, and the mean intensity and standard deviation (n = 4) were then calculated for each spot. To permit comparison between different arrays and probes, the average nonspecific cross-hybridization signal detemined at nonhomologous targets was assigned an arbitrary value of 50 fluorescence units. In all cases, a clear hybridization signal was readily differentiated from this nonspecific background.

Effect of target length on signal intensity. The primers used in this study were designed to create some overlap between the two or three target cDNAs derived from any given influenza virus gene. This made it possible to estimate what minimal target sequence length might be required to produce a hybridization signal. For example, targets A/HA2-1 and A/HA2-2 have 202 bp of common sequence and, by hybridizing a probe derived from A/HA2-1 to these arrays, one can measure the relative efficiency of probe binding to target A/HA2-1 (513 homologous base pairs) or to A/HA2-2 (202 homologous base pairs) (Fig. 1, panels 8 and 10). The results derived from this analysis are shown in Fig. 3. Generally it was noted that overlaps of less than ~100 bp produced poor or no hybridization signals, suggesting that this represents the minimal length of target required under these conditions. In contrast, overlapping sequences in excess of ~100 bp produced readily detectable fluorescence signals and the intensity appeared roughly dependent on the length of the overlapping region common to target and probe.

View larger version (27K): [in this window] [in a new window]   FIG. 3.   Hybridization signals depend on the length of homology. Hybridization signals were measured using a variety of probes and targets, and the fluorescence intensity was plotted as a function of the number of base pairs common to each probe and target. Positive hybridization signals were significantly differentiated from a nonspecific cross-hybridization background when the amount of base overlap exceeded 100 to 150 bp.

Hybridization using mixed probes and multiple dyes. A double-labeling experiment was also conducted to further test the discriminatory capacity of these arrays. Control experiments showed that a mixture of Cy3-labeled A/HA2-1 and B/NA-1 probes hybridized to the expected A/HA2-1 and B/NA-1 targets in a single array (Fig. 1, panel 10) producing fluorescent spots of comparable intensities. No cross-hybridization signals were detected beyond some additional binding to the A/HA2-2 locus (the A/HA2-1 probe has 202-bp homology with an A/HA2-2 target). We then used different fluorescent dyes to prepare A/HA2-1 probes labeled with Cy3-dCTP and B/NA-1 probes labeled with Cy5-dCTP. The probes were mixed together in equal amounts and hybridized to the array, and the bound fluorescent probes were then detected using 543 (Cy3) and 635 nm (Cy5) as the excitation wavelengths. The two resulting images are shown in Fig. 1, panels 8 and 9. Cy3- and Cy5-labeled probes seemed to hybridize with comparable efficiencies, producing fluorescence signals of comparable intensities. Moreover, the Cy3-labeled A/HA2-1 probes hybridized only to homologous A/HA2-1 and (overlapping) A/HA2-2 targets and the Cy-5-labeled B/NA-1 probe hybridized to the B/NA-1 target. These data show that one can differentiate a mixture of probes prepared using separate dyes and thus derive additional information regarding the probe composition.

Multiplex PCR. The preceding experiments showed that influenza virus cDNA arrays can be used to accurately identify different model probes either singly or in mixtures. However, the eventual goal of these experiments is to improve on multiplex methods capable of typing and subtyping unknown viruses. We thus screened our primers for primer combinations that are compatible in multiplex RT-PCRs. In recognition of the fact that some gene combinations are mutually exclusive (e.g., a virus is either H1 or H2 but can't be both), primer combinations that should produce a single diagnostically informative PCR product in a reaction with any given type or subtype of influenza virus were chosen. After testing large numbers of primer combinations against a battery of cDNAs templates, we eventually identified three multiplex primer combinations that collectively type and subtype influenza strains. These primer sets are summarized in Table 5. Figure 4 shows how these multiplex primer combinations can selectively amplify a particular gene target in reactions with one of the primer mixtures and a particular cloned cDNA target. For example, primer mixture A (Table 5) amplified an ~510-bp A/HA1-1 product in a PCR with a cloned A/HA1-1 template (Fig. 4, lane b) as well as ~490-bp A/HA2-3, ~700-bp A/HA3-3, and ~530-bp B/MP-1 products in PCRs with A/HA2-3, A/HA3-3, and B/MP-1 templates, respectively (Fig. 4, lanes e, h, and k).

View larger version (81K): [in this window] [in a new window]   FIG. 4.   Multiplex PCR amplification of influenza A and B virus strains. The three multiplex primer sets described in Table 5 were tested for the capacity to amplify a single, appropriately sized, DNA product in reactions with the indicated cloned templates. For example, primer mixture A (Table 5) produced a 510-bp product in a reaction with an A/HA1-1 template (lane b), a 490-bp product in a reaction with an A/HA2-3 template (lane e), primarily a 710-bp product in a reaction with an A/HA3-3 template (lane h), and a 530-bp product in a reaction with B/MP-1 template (lane k). PCR products were separated using a 1.2% agarose gel and stained with ethidium bromide.

Typing and subtyping influenza viruses by RT-PCR and array hybridization. In a now more realistic test of this method we examined whether the multiplex primers described in Table 5 can actually type and subtype different influenza viruses. Three viral RNA strains were tested (A/Denver/1/57 [H1N1], A/Victoria/3/75 [H3N2], and B/Hong Kong/5/72), with A/Denver/1/57 and B/Hong Kong/5/72 being different from those strains used to prepare arrayed targets (Table 1). None of the sequences targeted in strains A/Denver/1/57 and B/Hong Kong/5/72 were known to us when the work started, nor was the sequence of the MP gene of A/Victoria/3/75. Multiplex RT-PCR succeeded in producing all six of the six possible probes from mixtures containing influenza A virus RNA (HA1-1, HA3-3, NA1-1, NA2-2, and two MP-1 gene fragments) and two of the three possible probes from mixtures containing influenza B virus RNA (HA-1 and NA-1). In all cases only a single product was detected by agarose gel electrophoresis (data not shown). These cDNA probes correctly hybridized to targets specific for each particular type and subtype of virus (e.g., Fig. 1, panels 11 and 12), and the hybridization signals were again well above the background detected at nonhomologous spots (Fig. 5).

View larger version (34K): [in this window] [in a new window]   FIG. 5.   RT-PCR detection and differentiation of influenza virus RNA templates. The multiplex primers indicated in Table 5 were tested for the capacity to type and subtype the indicated influenza virus strains. Probes were prepared in reaction mixtures containing crude viral RNA, Cy3-dCTP, and an appropriate multiplex primer mixture and hybridized overnight to microarrays. The signal intensity was calculated as described in Materials and Methods (n = 4). Signals marked with an asterisk derive from perfectly matched probe-and-target pairs (i.e., the probe is identical to the arrayed target), and the remaining signals represent heterologous probe-and-target combinations (i.e., the target DNA came from a strain different from that tested in this experiment).

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

Although DNA arrays have been most widely utilized by the genomics research community to study gene expression patterns and identify new genes, many authorities have speculated that they might also prove useful in DNA-based diagnostics. In this paper we have shown that robotically spotted DNA chips, when used in conjunction with multiplex PCR methods, can be used to facilitate typing and subtyping human influenza viruses. Since the method exploits the specificity of both PCR and DNA hybridization reactions, it offers a degree of accuracy superior to those of many other methods commonly used to detect PCR-amplified DNAs. Thus it should be attractive in situations, such as medical diagnostics and forensics, where further identification of a PCR-amplified DNA may be required. The method is obviously applicable to any pathogen that can be cloned and arrayed and offers the advantage that one can detect and differentiate DNAs contained in mixtures of fluorescent PCR products (Fig. 1, panel 10). This suggests that DNA arrays could streamline the detection of multiple agents through parallel analysis of pools of PCR-amplified DNAs.

Several features of gene chips, PCR, and fluorescent probes facilitated the work outlined in this paper. First, PCR is well known for its extraordinary sensitivity and can be used to generate large quantities of probe DNA. Second, nucleotide-linked Cy3 and Cy5 dyes are readily incorporated into PCR products without greatly interfering with the efficiency of amplification or the yield of the probe. Third, these dyes are intensely fluorescent and hybridization is easily detected at levels well above background using modern confocally based array readers. Finally, the hybridization and wash protocols differ little from the more-traditional methods used with other nucleic acid binding membranes, and thus experience with traditional methods is directly applicable to the glass arrays used in these studies.

The preparation of DNA arrays used now fairly standard technology. We fixed DNA to the slide surface using a 5'-amine-tagged oligonucleotide and carbodiimide cross-linker. This method seems to provide a more stable array and works well if, as was done here, cloned inserts can be reamplified using a single pair of primers directed against flanking vector sequences. However, the high cost of amine-linked primers renders the method impractical if many different modified primers are desired. In such situations we have employed noncovalent binding methods with comparable success (J. Li and D. H. Evans, unpublished data). To prepare DNA arrays capable of typing and subtyping human influenza A and B virus strains, we designed 52 primers theoretically capable of amplifying 26 different portions of the influenza A and B virus HA, NA, and MP genes (Table 3). These primers amplified all but two of the intended targets in standard RT-PCRs. One of the two failures (B/MP-2) required primers that bind to sequences poorly represented in the databases and thus may have been badly designed; the failure of the other primer set (A/HA1-2) is inexplicable. The 24 successfully amplified viral cDNAs nevertheless provided a sufficiently redundant probe set for our purposes and were cloned, sequenced, and arrayed.

These arrayed DNAs, each about 500 bp long, were readily hybridized to probes prepared using the same cDNA templates (Fig. 1 and 2), with signal-to-background ratios ranging from 6 to 30 (Fig. 2). No obvious hybridization was detected at nonhomologous targets, and no cross-hybridization was detected among the different subtypes of the viruses. There was some expected cross-hybridization seen when probes encoded sequences shared by two or more target sequences. Further analysis of this phenomenon showed that >100 bp of homology was required to produce any detectable binding (Fig. 3). This effect was almost certainly due to the stringent annealing conditions we used to ensure maximal specificity, with the hybridization temperature set ~15°C higher than is recommended by the manufacturer of the hybridization buffer (Roche) for use in Southern blotting applications. This feature is useful, because it ensures that primer-derived sequences common to both PCR products and chip-bound targets will not cross-hybridize if illegitimate sequences have accidentally misamplified. We also noted that sensitivity was improved by increasing the size of the target spot and by increasing the length of the fluorescent probe (Fig. 1, panel 7). This last experiment also illustrates an important advantage of arraying partially redundant gene fragments rather than very large single cDNAs. The fact that a 1.6-kbp probe hybridized to three separate HA3 gene fragments showed that the cDNA encoded sequences spanning the entire gene. This provided an additional check on the identity of the probe.

It was not obvious how sequence mismatches affected hybridization efficiency. Using perfectly matched probes (i.e., probes derived from the same cDNA clones as target DNAs) we noted signal-to-noise ratios varying from 6 to 30 over a variety of gene fragments (Fig. 2). A similar range of signal-to-noise ratios (5 to 24) was noted when heterologous probes were prepared from viral isolates belonging only to the same viral type and subtype as the target virus cDNAs (Fig. 5). Analyzing the cause of this variation is difficult because it is hard to measure the specific activities of fluorescent probes and because there also appear to be target-specific variations in hybridization efficiency even with identical probe-target pairs (Fig. 2). However, it is encouraging to note that both targets A/HA1-1 and A/NA-1 (derived from A/New Jersey/8/76) hybridized to probes derived from virus strain A/Denver/1/57 with signal-to-background ratios of 5.5 and 5.8, respectively. Upon the conclusion of these experiments we directly sequenced portions of these A/Denver/1/57-derived probes and found that the sequenced portions had ~86% sequence identity with the target DNAs (GenBank accession no. AF305216 and AF305218). Similarly, B/HA-1 and B/NA-1 probes (derived from strain B/Hong Kong/5/72) hybridized to targets derived from strain B/Maryland/1/59 with signal-to-background ratios of 19 and 24, respectively. In this case sequencing detected ~96% sequence identity between probe and target (GenBank accession no. AF305219 and AF305220), suggesting that mismatches may reduce the amount of probe bound under these stringent hybridization conditions. Nevertheless, it seems clear that, if PCR primers of sufficiently broad utility can be designed, the great variation in influenza virus gene sequences (Table 2) will not seriously interfere with the application of this technology. We are currently expanding the content of these arrays to permit detection of a much greater selection of viral and other pathogens.