AND APPLIED ENVIRONMENTAL MICROBIOLOGY, May 2003, p. 28482856 0099-2240/03/$08.00 0 DOI: 10.1128/AEM.69.5.28482856.2003 Copyright 2003, American Society for Microbiology. All Rights Reserved.
Vol. 69, No. 5
Optimization of Single-Base-Pair Mismatch Discrimination in Oligonucleotide Microarrays
Hidetoshi Urakawa,1 Said El Fantroussi,1 Hauke Smidt,1 James C. Smoot,1 Erik H. Tribou,1 John J. Kelly,2 Peter A. Noble,1 and David A. Stahl1*
Department of Civil and Environmental Engineering, University of Washington, Seattle, Washington 98195,1 and Department of Biology, Loyola University Chicago, Chicago, Illinois 606262
Received 26 August 2002/Accepted 2 January 2003
The discrimination between perfect-match and single-base-pair-mismatched nucleic acid duplexes was investigated by using oligonucleotide DNA microarrays and nonequilibrium dissociation rates (melting proles). DNA and RNA versions of two synthetic targets corresponding to the 16S rRNA sequences of Staphylococcus epidermidis (38 nucleotides) and Nitrosomonas eutropha (39 nucleotides) were hybridized to perfectmatch probes (18-mer and 19-mer) and to a set of probes having all possible single-base-pair mismatches. The melting proles of all probe-target duplexes were determined in parallel by using an imposed temperature step gradient. We derived an optimum wash temperature for each probe and target by using a simple formula to calculate a discrimination index for each temperature of the step gradient. This optimum corresponded to the output of an independent analysis using a customized neural network program. These results together provide an experimental and analytical framework for optimizing mismatch discrimination among all probes on a DNA microarray.
DNA microarray technology provides parallel nucleic acid hybridizations for a large number of immobilized oligonucleotides or larger DNA fragments on a small surface area (21). In clinical and environmental microbiology, this technology has been used for assessing gene expression (19), characterizing whole genomes (5), identifying bacteria (8, 10, 28), and monitoring microbial populations (12, 22). We anticipate that, in the next several years, the application of DNA microarrays to environmental microbiology will greatly improve the understanding of complex microbial communities, which are typically composed of many microbial species. In general, oligonucleotide DNA microarrays containing 15to 25-mer oligonucleotide probes provide greater discrimination than microarrays composed of larger PCR-amplied DNA fragments. However, a central challenge to the application of DNA microarrays in environmental microbiology is achieving the specicity needed to resolve complex microbial populations, including discriminating between target and nontarget populations that differ by a single nucleotide (10). This level of specicity is needed to resolve variants of highly conserved genes (e.g., those encoding the rRNAs) and to distinguish between closely related target and nontarget microorganisms. In conventional hybridization assays, single-base-pair discrimination is achieved by adjusting the hybridization conditions (e.g., temperature, ionic strength, or formamide concentra* Corresponding author. Mailing address: Civil and Environmental Engineering, University of Washington, 302 More Hall, Box 352700, Seattle, WA 98195. Phone: [phone]. Fax: [phone]. Email: [e-mail]. Present address: National Institute for Environmental Studies, Tsukuba, Ibaraki 305-8606, Japan. Present address: Unit of Bioengineering, University of Louvain, B-1348 Louvain-la-Neuve, Belgium. Present address: Wageningen University, 6703 CT Wageningen, The Netherlands. 2848
tion) or washing conditions (dissociation) of the probe-target duplex (31). In DNA microarray assays, however, this approach is difcult to use since one set of hybridization and wash conditions does not provide optimal target discrimination for all probes on the microarray. We therefore have developed an alternative approach that uses differences in thermal dissociation rates of probe-target duplexes to resolve matched and mismatched probe-target duplexes (13, 25). The oligonucleotide DNA microarray used in this study is a variant of the more conventional format (15, 22, 29). Rather than being directly attached to glass, the probes are immobilized in three-dimensional polyacrylamide gel pads afxed to the glass (2, 6, 9, 10, 12, 13, 18, 2426, 30). The gel pads provide a format suitable for the determination of equilibrium and nonequilibrium dissociation kinetics (i.e., melting proles) of a large number of probe-target duplexes and for determining the dissociation temperature (Td), the temperature at which 50% of the duplexes remain during a specied wash period (13, 25). In this study, we used nonequilibrium dissociation kinetics to derive the optimum washing temperature for each probe, providing for maximum discrimination between target RNA or target DNA and all possible single-nucleotide-mismatch variants.
MATERIALS AND METHODS Synthetic DNA and RNA targets. The 16S rDNA gene sequences of Staphylococcus epidermidis (accession no. L37605 and X75943) and Nitrosomonas eutropha (accession no. M96402) were obtained from GenBank in the National Center for Biotechnology Information. Single-strand complements to each probe, containing an additional 10 nucleotides of nontarget anking sequence at the 5 and 3 termini, were synthesized in DNA (Operon Technologies Inc., Alameda, Calif.) and RNA (Dharmacon Research Inc., Lafayette, Colo.) forms to avoid possible biases resulting from sample preparation using native rRNA (e.g., possible variability in the efciency of fragmentation and labeling) (2). The target molecules were uorescently labeled with Cy3 at the 5 terminus. The name of target, sequence, size, location of the sequence in the 16S rRNA gene
VOL. 69, 2003
(Escherichia coli numbering), and probe binding site (underlined) are as follows: Staphylococcus target, 5 -TCTGGTCTGTAACTGACGCTGATGTGCGAAA GCGTGGGG-3 , 39-mer, position 737 to 775; Nitrosomonas target, 5 -ACTAC AAAGCTAGAGTGCAGCAGAGGGGAGTGGAATTC-3 , 38-mer, position 643 to 680. Oligonucleotide probe design and synthesis. A 19-mer oligonucleotide probe (S-G-Staphy-0747-a-A-19) targeting Staphylococcus 16S rRNA was designed as described previously (25). An 18-mer oligonucleotide probe (S-*-Nsom-0653-aA-18) targeting halotolerant and obligate halophilic Nitrosomonas (27) was used for the Nitrosomonas target. These probes were complemented by a set of probes having all possible single-mismatch variants at each position (Table 1). Probes having two to ve mismatches were also incorporated on the microarray. All probes were synthesized with an amino linker at the 3 terminus as described previously (26). Microarray fabrication. The microarray matrix consisted of 100- by 100- by 20- m polyacrylamide gel pads at a 100- m spacing. The gel pads were xed to a glass slide by photopolymerization (9) and activated as described previously (18), and 1 pmol of probe was applied to each gel pad in one droplet (1 nl) of a 1 mM amino-oligonucleotide solution (24) with a robot arrayer (30). The oligonucleotide probes were immobilized through reductive coupling of a 3 amino group of the oligonucleotide with the aldehyde group of the activated gel pad on the microarrays (18). Microarray hybridization. Hybridizations were conducted at room temperature (20C) for 12 h in a hybridization chamber afxed to the surface of the glass slide (Grace BioLabs, Bend, Oreg.) containing 40 l of hybridization buffer (0.9 M NaCl, 20 mM Tris-HCl [pH 8.0], 40% formamide) and 1 l of Cy3-labeled target nucleic acids (each at 25 ng/ l). Following hybridization, the microarray was briey washed three times at room temperature with 100 l of wash buffer (20 mM Tris-HCl [pH 8.0], 5 mM EDTA, 4 mM NaCl). After the nal wash, 100 l of wash buffer was added to the wash chamber (Grace BioLabs) for uorescence monitoring. Image analysis was performed by using a custom-designed uorescence microscope (State Optical Institute, St. Petersburg, Russia) equipped with a cooled charge-coupled device camera (Princeton Instruments, Trenton, N.J.). Preliminary experiments revealed that there was no cross-hybridization of any probe-target duplexes when both target sequences were used (data not shown). Four microarray slides were used repeatedly in this study. Generation of melting proles. To generate melting proles, the microarray was xed to a thermal table mounted on the stage of the microscope. The thermal table was connected to a thermoelectric temperature controller (LFI3751; Wavelength Electronics, Inc., Bozeman, Mont.) and a water bath (Cole Parmer Instruments Co., Chicago, Ill.). Melting proles for all gel pads were generated by gradually increasing the temperature (1C/min) of the thermal table from 20 to 70C and recording the uorescence signal intensity of the gel pads at 2C intervals. Temperature, data acquisition, image processing, and analysis were controlled with custom software written in LabVIEW (version 5.1; National Instruments Co., Austin, Tex.). The signal intensity of each melting prole was normalized, and the Td was calculated by using Tdcalculator (http://stahl.ce .washington.edu) as described previously (25). Obtained Tds are listed in Table 1. Hybridization and melting prole analyses were repeated ve times for both DNA and RNA targets. DI. The optimum wash temperature, dened as that providing maximum discrimination between perfect-match duplexes and those containing mismatches, is generally determined empirically. To rene and systematize the determination of an optimum wash temperature, we introduced a discrimination index (DI). The DI for a specic wash temperature was determined by the following equation: DItemperature (pmtemperature/mmtemperature) (pmtemperature mmtemperature), where pmtemperature is the average signal intensity of perfectmatch duplexes at a specic wash temperature and mmtemperature is the average signal intensity of mismatched duplexes, excluding those duplexes which have terminal and next-to-terminal mismatches. Data for the NN. The input data set consisted of signal intensity (melting) proles, with each input record consisting of a single prole of either a perfectmatch duplex, a duplex with a mismatch in the ultimate or penultimate position, or a duplex with an internal mismatch. The output data set consisted of one categorical variable that was coded 0 if the corresponding record was a perfectmatch duplex, 1 if the duplex had a mismatch in the ultimate or penultimate position, or 2 if the duplex had an internal mismatch. Prior to neural network (NN) analyses, the data were all normalized to have a mean of 0 and a standard deviation (SD) of 1. NN software and analyses. The NN software was custom designed by using Java software and was based on the leave one input out cross-validation model (3). Rather than leave one input out, we modied the model to use one input (e.g., single intensity values at a specic temperature) to predict a categorical
MELTING PROFILES OF NUCLEIC ACID DUPLEXES
2849
output (e.g., a perfect-match duplex, a duplex with a mismatch in the ultimate or penultimate position, or a duplex with an internal mismatch). We chose this approach because it was difcult to measure the importance of inputs that are statistically dependent (i.e., signal intensities within the same melting prole are highly correlated to one another). The software is available at a World Wide Web-based interface at http://stahl.ce.washington.edu under the heading Tools for data analyses. The network architecture consisted of one input layer, one hidden layer, and one output layer. Neurons in the hidden layer used a hyperbolic tangent activation function, while the neuron in the output layer used a standard purely linear activation function (11). All neurons included a bias term. The LevenbergMarquardt algorithm was used for training the NN rather than standard backpropagation and conjugate gradient methods because preliminary results showed that the Levenberg-Marquardt algorithm was superior in terms of the number of iterations needed to reach the error minima (11). Since preliminary analysis revealed that the minimum number of hidden neurons needed to produce the highest R2 results was two, only two hidden neurons were used for all NN analyses. A standard least-squares error function was used for training the NN since this function could be easily converted to R2 values. It should be noted that our method does not produce generalizable NNs since our specic objective was to identify with which inputs the NN learned best. Therefore, no data were used for testing or validation purposes. The NN was deemed to have reached minima (and consequently training was stopped) when the R2 did not increase by more than 0.001 U over a period of 10 s (i.e., approximately 200 megaops). For NN analyses, we generated an independent NN for each individual input. If one NN performed better with one input rather than another (i.e., it had a higher R2 value), the input having the better prediction was assumed to be more important. It is essential to recognize that this approach does not provide information on the optimal subsets of inputs but rather identies which inputs are most important for predicting outputs when presented independently. Since some NNs do not train properly because they reach local minima of their error space, a median of 11 NN runs was conducted for each input. We chose the median rather than the mean since the median minimizes local-minimum effects.
RESULTS AND DISCUSSION A typical image of the DNA microarray after DNA-DNA hybridization and the corresponding position of each probe on the microarray are shown in Fig. 1. To optimize hybridization and washing conditions, several different types of hybridization buffer containing 0 to 70% formamide and different compositions of wash buffers were tested (12, 25). The optimal conditions were achieved with a hybridization buffer containing 40% formamide and a wash buffer containing 4 mM NaCl. The signal intensities of ve probes having more than three mismatches were below detection under these hybridization and wash conditions. However, these conditions did not provide sufcient stringency for discriminating single- or double-base-pair mismatches. For this reason, we examined the melting proles of probe-target duplexes to determine if adequate discrimination can be attained for single- and double-base-pair mismatches. Typical normalized melting proles of perfect-match probetarget duplexes and those with one or two mismatched base pairs are shown in Fig. 2. For these duplexes, discrimination among perfect-match and mismatched probe-target duplexes was achieved by comparing the Tds. In general, the Td provides an important experimental parameter for distinguishing between probe-target duplexes with and without mismatches (31). For example, a previous study based on melting proles revealed that Tds provided excellent differentiation among ve closely related Bacillus species (13). However, complete resolution is achieved only when the Tds of the perfect-match probetarget and duplexes containing mismatches are sufciently different. The experimentally determined Tds for perfect-match and
2850
URAKAWA ET AL. TABLE 1. Oligonucleotide probes used in this study and their corresponding Tds
APPL. ENVIRON. MICROBIOL.
Series
Probea
Sequencee
DNA Td Mean SD T
b
RNA CV
c
Td Mean
SD
T
CV
S0 S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 S17 S18 S19 S20 S21 S22 S23 S24 S25 S26 S27 S28 S29 S30 S31 S32 S33 S34 S35 S36 S37 S38 S39 S40 S41 S42 S43 S44 S45 S46 S47 S48 S49 S50 S51 S52 S53 S54 S55 S56 S57 S58 S59 S60 S61 N0 N1 N2 N3
spm s1aa s1ga s1ca s2ag s2gg s2tg s3cc s3tc s3ac s4gg s4tg s4ag s5gt s5ct s5tt s6tg s6ag s6gg s7gt s7ct s7tt s8aa s8ga s8ca s9tg s9ag s9gg s10gt s10ct s10tt s11cc s11tc s11ac s12tg s12ag s12gg s13cc s13tc s13ac s14aa s14ga s14ca s15tg s15ag s15gg s16gt s16ct s16tt s17cc s17tc s17ac s18aa s18ga s18ca s19aa s19ga s19ca 1aa2gg 3cc4gg 3cc4gg11cc12gg 11cc12gg npm n1gg n1ag n1tg
TCGCACATCAGCGTCAGTT ACGCACATCAGCGTCAGTT GCGCACATCAGCGTCAGTT CCGCACATCAGCGTCAGTT TAGCACATCAGCGTCAGTT TGGCACATCAGCGTCAGTT TTGCACATCAGCGTCAGTT TCCCACATCAGCGTCAGTT TCTCACATCAGCGTCAGTT TCACACATCAGCGTCAGTT TCGGACATCAGCGTCAGTT TCGTACATCAGCGTCAGTT TCGAACATCAGCGTCAGTT TCGCGCATCAGCGTCAGTT TCGCCCATCAGCGTCAGTT TCGCTCATCAGCGTCAGTT TCGCATATCAGCGTCAGTT TCGCAAATCAGCGTCAGTT TCGCAGATCAGCGTCAGTT TCGCACGTCAGCGTCAGTT TCGCACCTCAGCGTCAGTT TCGCACTTCAGCGTCAGTT TCGCACAACAGCGTCAGTT TCGCACAGCAGCGTCAGTT TCGCACACCAGCGTCAGTT TCGCACATTAGCGTCAGTT TCGCACATAAGCGTCAGTT TCGCACATGAGCGTCAGTT TCGCACATCGGCGTCAGTT TCGCACATCCGCGTCAGTT TCGCACATCTGCGTCAGTT TCGCACATCACCGTCAGTT TCGCACATCATCGTCAGTT TCGCACATCAACGTCAGTT TCGCACATCAGTGTCAGTT TCGCACATCAGAGTCAGTT TCGCACATCAGGGTCAGTT TCGCACATCAGCCTCAGTT TCGCACATCAGCTTCAGTT TCGCACATCAGCATCAGTT TCGCACATCAGCGACAGTT TCGCACATCAGCGGCAGTT TCGCACATCAGCGCCAGTT TCGCACATCAGCGTTAGTT TCGCACATCAGCGTAAGTT TCGCACATCAGCGTGAGTT TCGCACATCAGCGTCGGTT TCGCACATCAGCGTCCGTT TCGCACATCAGCGTCTGTT TCGCACATCAGCGTCACTT TCGCACATCAGCGTCATTT TCGCACATCAGCGTCAATT TCGCACATCAGCGTCAGAT TCGCACATCAGCGTCAGGT TCGCACATCAGCGTCAGCT TCGCACATCAGCGTCAGTA TCGCACATCAGCGTCAGTG TCGCACATCAGCGTCAGTC AGGCACATCAGCGTCAGTT TCCGACATCAGCGTCAGTT TCCGACATCACGGTCAGTT TCGCACATCACGGTCAGTT CCCCTCTGCTGCACTCTA GCCCTCTGCTGCACTCTA ACCCTCTGCTGCACTCTA TCCCTCTGCTGCACTCTA
45.1 41.4 43.7 40.8 43.1 42.7 41.7 37.8 39.1 41.5 41.6 41.0 40.8 43.2 39.6 39.8 41.0 39.4 40.2 35.4 39.4 41.3 41.6 43.7 41.1 41.1 40.2 39.6 43.1 40.6 42.0 37.0 38.6 37.8 41.9 40.6 41.5 37.1 38.5 38.9 42.1 44.9 42.0 41.2 40.6 40.6 43.9 41.3 42.7 42.2 41.6 41.3 43.8 45.9 41.7 44.6 44.7 44.4 44.2 38.8 NDd 34.2 43.4 42.5 42.9 42.5
1.2 1.5 1.5 1.0 1.4 0.9 1.0 1.0 0.9 0.6 0.6 0.9 0.5 1.3 1.0 2.0 0.6 0.9 0.9 0.4 0.8 0.8 1.0 1.3 0.9 0.9 0.9 0.8 1.2 0.9 1.1 0.7 0.9 0.4 1.1 0.9 0.8 0.7 0.5 1.3 0.8 0.7 0.6 0.8 0.8 0.8 1.3 0.8 0.7 0.8 0.7 0.8 1.7 1.0 0.6 1.1 1.4 1.4 1.4 1.1 ND 0.8 1.3 0.8 2.2 2.2
0.0 3.7 1.4 4.3 2.0 2.4 3.4 7.3 6.0 3.6 3.5 4.1 4.3 1.9 5.5 5.3 4.1 5.7 4.9 9.7 5.7 3.8 3.5 1.4 4.0 4.0 4.9 5.5 2.0 4.5 3.1 8.1 6.5 7.3 3.2 4.5 3.6 8.0 6.6 6.2 3.0 0.2 3.1 3.9 4.5 4.5 1.2 3.8 2.4 2.9 3.5 3.8 1.3 0.8 3.4 0.5 0.4 0.7 0.9 6.3 ND 10.9 0.0 0.9 0.5 0.9
2.7 3.7 3.3 2.3 3.3 2.1 2.5 2.6 2.4 1.5 1.3 2.2 1.1 3.0 2.5 4.9 1.5 2.2 2.3 1.1 2.0 2.0 2.3 3.0 2.1 2.2 2.3 2.0 2.7 2.2 2.5 1.8 2.3 1.1 2.5 2.3 2.0 1.9 1.4 3.2 1.9 1.6 1.3 1.8 2.1 1.8 2.9 1.9 1.6 1.8 1.8 1.9 3.9 2.2 1.4 2.4 3.1 3.2 3.1 2.7 ND 2.3 6.5 2.0 5.0 5.3
44.8 40.0 42.0 40.4 41.3 40.7 40.7 35.6 37.6 38.5 38.4 39.4 36.7 40.2 38.0 39.1 40.1 37.2 36.1 36.0 36.8 38.5 37.9 38.8 38.5 39.4 35.6 36.3 39.9 38.0 38.8 35.5 36.0 34.5 40.1 36.3 38.1 35.8 36.3 38.0 38.4 40.5 39.8 39.3 36.3 36.8 41.2 38.7 39.2 39.8 39.0 38.7 40.0 43.7 37.2 42.8 41.8 41.5 43.5 36.9 ND 37.5 44.3 43.0 43.7 43.5
1.8 1.6 2.5 2.1 2.5 1.8 1.2 0.6 0.8 1.1 1.5 1.6 1.2 1.0 1.3 0.9 1.0 0.8 1.4 1.8 1.3 1.3 0.8 1.1 1.1 1.1 2.1 1.5 1.7 1.3 0.9 0.2 0.6 0.5 1.2 1.2 0.8 0.6 0.6 0.6 1.1 1.5 1.4 1.3 1.5 0.9 0.6 0.6 0.6 1.6 1.4 1.7 1.7 2.2 1.2 1.5 1.0 0.7 1.0 0.8 ND 2.1 2.9 2.5 2.7 3.2
0.0 4.8 2.8 4.4 3.5 4.1 4.1 9.2 7.2 6.3 6.4 5.4 8.1 4.6 6.8 5.7 4.7 7.6 8.7 8.8 8.0 6.3 6.9 6.0 6.3 5.4 9.2 8.5 4.9 6.8 6.0 9.3 8.8 10.3 4.7 8.5 6.7 9.0 8.5 6.8 6.4 4.3 5.0 5.5 8.5 8.0 3.6 6.1 5.6 5.0 5.8 6.1 4.8 1.1 7.6 2.0 3.0 3.3 1.3 7.9 ND 7.3 0.0 1.3 0.6 0.8
4.0 4.0 5.9 5.1 5.9 4.3 2.9 1.7 2.2 3.0 4.0 4.0 3.2 2.6 3.3 2.2 2.4 2.1 3.7 5.0 3.5 3.3 2.0 2.8 2.8 2.8 5.8 4.2 4.3 3.3 2.2 0.6 1.6 1.5 3.0 3.3 2.1 1.5 1.7 1.6 2.7 3.6 3.5 3.2 4.1 2.4 1.5 1.6 1.4 4.1 3.6 4.3 4.2 5.1 3.3 3.5 2.3 1.7 2.3 2.2 ND 5.5 6.5 5.7 6.1 7.4
Continued on following page
VOL. 69, 2003
MELTING PROFILES OF NUCLEIC ACID DUPLEXES TABLE 1Continued
2851
Series
Probea
Sequencee
DNA Td Mean SD T
b
RNA CV
c
Td Mean
SD
T
CV
N4 N5 N6 N7 N8 N9 N10 N11 N12 N13 N14 N15 N17 N16 N18 N19 N20 N21 N22 N23 N24 N25 N26 N27 N28 N29 N30 N31 N32 N33 N34 N35 N36 N37 N38 N39 N40 N41 N42 N43 N44 N45 N46 N47 N48 N49 N50 N51 N52 N53 N54 N55 N56 N57 N58 N59 N60 N61
n2gg n2ag n2tg n3gg n3ag n3tg n4gg n4ag n4tg n5ga n5aa n5ca ng6g n6ag n6gg n7aa n7ga n7ca n8cc n8tc n8ac n9tg n9ag n9gg n10aa n10ga n10ca n11cc n11tc n11ac n12tg n12ag n12gg n13gt n13ct n13tt n14tg n14ag n14gg n15aa n15ga n15ca n16tg n16ag n16gg n17aa n17ga n17ca n18gt n18ct n18tt 1gg2gg 3gg4gg 8cc9gg14gg 8cc9gg11cc12gg 6gg8cc9gg11cc12gg 8cc9gg14gg 6gg8cc9gg11cc12gg
CGCCTCTGCTGCACTCTA CACCTCTGCTGCACTCTA CTCCTCTGCTGCACTCTA CCGCTCTGCTGCACTCTA CCACTCTGCTGCACTCTA CCTCTCTGCTGCACTCTA CCCGTCTGCTGCACTCTA CCCATCTGCTGCACTCTA CCCTTCTGCTGCACTCTA CCCCGCTGCTGCACTCTA CCCCACTGCTGCACTCTA CCCCCCTGCTGCACTCTA CCCCTTTGCTGCACTCTA CCCCTATGCTGCACTCTA CCCCTGTGCTGCACTCTA CCCCTCAGCTGCACTCTA CCCCTCGGCTGCACTCTA CCCCTCCGCTGCACTCTA CCCCTCTCCTGCACTCTA CCCCTCTTCTGCACTCTA CCCCTCTACTGCACTCTA CCCCTCTGTTGCACTCTA CCCCTCTGATGCACTCTA CCCCTCTGGTGCACTCTA CCCCTCTGCAGCACTCTA CCCCTCTGCGGCACTCTA CCCCTCTGCCGCACTCTA CCCCTCTGCTCCACTCTA CCCCTCTGCTTCACTCTA CCCCTCTGCTACACTCTA CCCCTCTGCTGTACTCTA CCCCTCTGCTGAACTCTA CCCCTCTGCTGGACTCTA CCCCTCTGCTGCGCTCTA CCCCTCTGCTGCCCTCTA CCCCTCTGCTGCTCTCTA CCCCTCTGCTGCATTCTA CCCCTCTGCTGCAATCTA CCCCTCTGCTGCAGTCTA CCCCTCTGCTGCACACTA CCCCTCTGCTGCACGCTA CCCCTCTGCTGCACCCTA CCCCTCTGCTGCACTTTA CCCCTCTGCTGCACTATA CCCCTCTGCTGCACTGTA CCCCTCTGCTGCACTCAA CCCCTCTGCTGCACTCGA CCCCTCTGCTGCACTCCA CCCCTCTGCTGCACTCTG CCCCTCTGCTGCACTCTC CCCCTCTGCTGCACTCTT GGCCTCTGCTGCACTCTA CCGGTCTGCTGCACTCTA CCCCTCTCGTGCACTCTA CCCCTCTCGTCGACTCTA CCCCTGTCGTCGACTCTA CCCCTCTCGTCGAGTCTA CCCCTGTCGTCGAGTCTA
42.1 42.1 41.8 41.1 41.4 40.6 41.0 40.9 40.6 40.8 40.7 39.8 38.6 40.1 39.3 40.1 40.9 39.5 34.7 35.6 37.6 39.4 41.0 39.9 39.8 40.4 39.0 35.2 36.7 38.0 39.9 39.2 40.8 41.9 38.3 39.3 38.8 39.0 39.8 39.8 40.8 39.0 40.6 41.9 42.6 42.3 44.6 42.0 42.9 42.7 43.5 41.1 37.0 33.6 ND ND ND ND
2.1 2.3 2.3 2.0 2.4 1.9 1.4 1.2 1.0 1.0 1.5 1.8 1.5 2.1 1.3 2.0 2.1 1.6 0.6 1.5 0.6 0.9 1.3 1.3 1.4 1.6 1.3 0.7 0.9 1.1 1.3 1.4 0.8 1.0 0.5 0.8 1.2 0.9 0.9 0.4 0.8 1.2 0.9 0.6 0.3 1.3 1.5 1.1 1.7 1.7 1.4 1.0 0.9 1.5 ND ND ND ND
1.3 1.3 1.6 2.3 2.0 2.8 2.4 2.5 2.8 2.6 2.7 3.6 4.8 3.3 4.1 3.3 2.5 3.9 8.7 7.8 5.8 4.0 2.4 3.5 3.6 3.0 4.4 8.2 6.7 5.4 3.5 4.2 2.6 1.5 5.1 4.1 4.6 4.4 3.6 3.6 2.6 4.4 2.8 1.5 0.8 1.1 1.2 1.4 0.5 0.7 0.1 2.3 6.4 9.8 ND ND ND ND
5.0 5.5 5.4 4.8 5.8 4.8 3.4 2.9 2.3 2.4 3.8 4.4 4.0 5.2 3.4 5.0 5.1 4.0 1.6 4.3 1.7 2.2 3.2 3.3 3.5 4.0 3.4 1.9 2.5 2.9 3.2 3.6 2.0 2.3 1.2 2.1 3.1 2.3 2.2 1.0 1.8 3.1 2.3 1.4 0.7 3.1 3.4 2.6 3.9 3.9 3.2 2.5 2.4 4.5 ND ND ND ND
43.4 43.2 43.5 43.2 42.2 42.0 41.9 42.0 42.8 42.3 42.8 42.5 40.4 40.3 40.0 41.1 40.7 40.7 40.1 40.8 41.8 44.6 42.3 41.4 41.0 40.6 40.1 38.8 38.9 40.0 40.7 41.2 42.2 44.9 43.4 41.2 40.9 40.3 40.2 40.1 40.0 40.7 40.7 40.9 43.8 43.7 44.5 41.8 43.0 42.9 43.8 43.0 39.3 39.4 32.7 ND ND ND
3.0 3.0 2.8 2.3 1.8 1.3 1.4 1.0 2.7 1.7 2.8 3.0 2.8 2.6 2.3 2.9 2.1 1.9 1.3 1.6 1.5 2.3 1.7 2.0 1.7 1.2 1.6 1.3 1.6 2.0 2.2 1.7 1.6 3.5 1.9 1.6 2.2 1.8 1.6 1.4 1.4 1.5 1.3 1.8 1.8 1.8 1.4 1.7 1.3 1.4 1.4 1.8 0.8 1.1 2.1 ND ND ND
0.9 1.1 0.8 1.1 2.1 2.3 2.4 2.3 1.5 2.0 1.5 1.8 3.9 4.0 4.3 3.2 3.6 3.6 4.2 3.5 2.5 0.3 2.0 2.9 3.3 3.7 4.2 5.5 5.4 4.3 3.6 3.1 2.1 0.6 0.9 3.1 3.4 4.0 4.1 4.2 4.3 3.6 3.6 3.4 0.5 0.6 0.2 2.5 1.3 1.4 0.5 1.3 5.0 4.9 11.6 ND ND ND
6.8 6.9 6.3 5.3 4.2 3.1 3.2 2.4 6.2 3.9 6.6 7.0 7.0 6.5 5.8 7.0 5.1 4.7 3.3 3.9 3.5 5.1 4.1 4.7 4.2 2.9 3.9 3.2 4.1 5.0 5.4 4.1 3.8 7.8 4.4 3.8 5.4 4.5 4.0 3.4 3.5 3.6 3.2 4.3 4.1 4.1 3.2 4.0 3.0 3.3 3.2 4.1 1.9 2.8 6.3 ND ND ND
a Probe names incorporate the type of target (s, Staphylococcus; n, Nitrosomonas), position of the mismatch from the 5 terminus, and the type of mismatch (probe-target). spm and npm, perfect-match probes for Staphylococcus and Nitrosomonas targets, respectively. b T, difference of the Td between the perfect-match probe and each mismatch probe. c CV, coefcient of variation. d ND, not determined due to faint uorescence signal. e Mismatches are underlined.
all single-base-pair-mismatch duplexes are listed in Table 1. The mean Tds for some single-base-pair-mismatch duplexes were slightly higher than the Td for perfect-match duplexes. For example, the Td for probe N50 was 44.6C (SD, 1.5C) in
DNA duplexes while the Td for perfect-match probe N0 was 43.4C (SD, 1.3C). In this case, the difference between Tds was not sufcient to adequately resolve perfect-match duplexes and duplexes having a single-base-pair mismatch.
2852
URAKAWA ET AL.
APPL. ENVIRON. MICROBIOL.
FIG. 1. Typical image of a DNA microarray after hybridization with DNA target sequences (A) and the locations of the oligonucleotide probes (B). Probe labels are as in Table 1. Hybridization and wash conditions are described in the text. Exposure time was 1.0 s. White boxes (A) indicate probes that did not yield detectable uorescence signals after the wash at 20C.
In this study we evaluated the inclusion of signal intensity data to optimize discrimination among perfect-match and mismatched probe-target duplexes. Considering intensity data alone, an optimum corresponds to hybridization and washing conditions at which the signal intensity of mismatches reaches (or approaches) background and the perfect-match duplex maintains a detectable signal. Often these conditions are determined empirically, as represented by Fig. 3. This gure shows the signal intensity for each probe duplex (color) measured at 2C increments during the thermal dissociation. An empirical estimate of the optimum wash temperature for each probe-target duplex is shown (left section of each panel), and the corresponding intensity data are shown in the adjacent section. For perfect-match duplexes, signal intensities at each empirically dened optimum were approximately 20% of the initial signal intensities. For example, the signal intensity of the perfect-match DNA-DNA duplex of Staphylococcus was 1.11 U at 20C, while the signal intensity was 0.16 U at the empirically determined optimal wash temperature (52C) (Fig. 3A, right section). These intensity measurements corresponded to those achieved in a separate experiment in which the microarray was washed at the identied temperature optimum (Fig. 4). However, it was not possible to fully resolve perfect-match probe-target duplexes and those with mismatches at the ultimate or penultimate position. These results were in accordance with the conclusion derived from Td analysis reported previously (25).
We also observed contrasting relative stabilities of DNADNA versus RNA-DNA duplexes for these two probes. For Staphylococcus, the empirically determined optimum wash temperature for DNA-DNA duplexes was higher than that for RNA-DNA duplexes (52 versus 48C) (Fig. 3A and B). How-
FIG. 2. Typical normalized melting proles of DNA-DNA duplexes of Staphylococcus. S0, perfect-match duplex; S30, single-basepair-mismatched duplex containing a tt mismatch (probe-target) at position 10 from the 5 terminus; S59, double-base-pair-mismatched duplex containing cc and gg mismatches at positions 3 and 4, respectively. Error bars, SDs of the data (S0, n 10; S30, n 5; S59, n 4).
VOL. 69, 2003
MELTING PROFILES OF NUCLEIC ACID DUPLEXES
2853
FIG. 3. Signal intensity prole of probe-target duplexes with temperature gradient (color sections; A.U., arbitrary units of uorescence intensities) and signal intensities at empirically determined optimum wash temperatures (bars). Red triangles, optimum wash temperatures. (A and B) Staphylococcus target DNA (A) and target RNA (B); (C and D) Nitrosomonas target DNA (C) and target RNA (D). Data represent the mean signal intensities of ve melting prole analyses and SDs (error bars).
2854
URAKAWA ET AL.
APPL. ENVIRON. MICROBIOL.
FIG. 4. Typical images of DNA microarrays washed at optimum temperatures. Melting proling was terminated at the empirically determined optimum wash temperature. The optimum wash temperatures (shown in Fig. 3) were 52C for the Staphylococcus DNA probe-target duplexes (A) and 50C for the Nitrosomonas DNA probe-target duplexes (B). pm, perfect-match probes for Staphylococcus target (S0 in Fig. 1B) or the Nitrosomonas target (N0 in Fig. 1B); 1st, probes dene ultimate position; 2nd, probes dene penultimate position.
ever, for Nitrosomonas, the optimum for the DNA-DNA duplexes was lower than that for RNA-DNA duplexes (50 versus 58C) (Fig. 3C and D). These contrasting results, also supported by comparing their Tds (P 0.0001), indicate that the stability of DNA-DNA duplexes and RNA-DNA duplexes is sequence dependent (20, 23) and underscore the difculty in a priori prediction of duplex stability using currently available models (H. Urakawa et al., unpublished data). To rene and systematize the above-described empirical approach, we introduced a DI, which is calculated by the formula given in Materials and Methods. This index is dened as the product of difference and ratio of the signal intensities for perfect-match and mismatched duplexes at a given wash temperature. The temperature with the maximum DI is dened as the optimum wash temperature. As shown in Fig. 5, optimum wash temperatures were calculated by using the DI from the signal intensity proles in the range of 20 to 64C (Fig. 3). For hybridization with the Staphylococcus target, DI-based and empirically determined optimum wash temperatures for DNA-
DNA and RNA-DNA hybridizations were identical (Fig. 3 and 5). For Nitrosomonas, DI-based and empirically determined optimum wash temperatures were within 2C of each other (i.e., triangle and peak DI values occur at around the same temperature in Fig. 5), suggesting a reasonable match between DI-based prediction and the empirical determination. NN analyses were used to further investigate the relationship between terminal and internal mismatches (Fig. 5). The NNs were able to adequately discriminate between perfectmatch probe-target duplexes and duplexes with internal mismatches (R2 0.90) and between perfect-match probe-target duplexes and duplexes with mismatches at any position (R2 0.70) within the temperature intervals indicated. These results are in good agreement with the empirically determined optimum wash temperatures and maxima of DI proles (Fig. 3 and 5) and support the use of the DI to identify optimum washing temperatures. The application of NNs to the analysis of complex data in microbiology is relatively new (1). NNs have been used to
VOL. 69, 2003
MELTING PROFILES OF NUCLEIC ACID DUPLEXES
2855
FIG. 5. Inferred optimum wash temperatures for the discrimination of perfect and mismatched duplexes. (A and B) Staphylococcus target DNA (sDNA; A) and target RNA (sRNA; B); (C and D) Nitrosomonas target DNA (nDNA; C) and target RNA (nRNA; D). DI was calculated by using the formula given in Materials and Methods. Triangles, temperatures empirically inferred from melting proles. Light gray zones, temperature intervals allowing for mismatch discrimination as deduced from NN analysis using all data sets (R2 0.7); dark gray zones, temperature intervals deduced from NN analysis using data sets excluding data from ultimate and penultimate positions (R2 0.9).
identify the restriction enzyme proles for E. coli O156:H7 (4), the pyrolysis mass spectra for Mycobacterium tuberculosis complex species (7), bacterial species from randomly amplied polymorphic DNA patterns (14), fatty acid proles of microbial communities (17), stable low-molecular-weight rRNA from gel electrophoresis patterns (16), and Td from microarray data (25). However, to our knowledge, no study has used the method outlined in this paper to determine the relative importance of inputs to outputs. In conclusion, our studies have established an analytical approach to achieving optimum discrimination between target and nontarget duplex structures. Although this objective is important in any application of DNA microarrays to sequence analysis (e.g., identication of point mutations), we n...