Effects of Target Length on the Hybridization Efficiency and Specificity of rRNA-Based Oligonucleotide Microarrays

Abstract

The effect of target size on microarray hybridization efficiencies and specificity was investigated using a set of 166 oligonucleotide probes targeting the 16S rRNA gene of Escherichia coli. The targets included unfragmented native rRNA, fragmented rRNA (∼20 to 100 bp), PCR amplicons (93 to 1,480 bp), and three synthetic single-stranded DNA oligonucleotides (45 to 56 bp). Fluorescence intensities of probes hybridized with targets were categorized into classes I (81 to 100% relative to the control probe), II (61 to 80%), III (41 to 60%), IV (21 to 40%), V (6 to 20%), and VI (0 to 5%). Good hybridization efficiency was defined for those probes conferring intensities in classes I to IV; those in classes V and VI were regarded as weak and false-negative signals, respectively. Using unfragmented native rRNA, 13.9% of the probes had fluorescence intensities in classes I to IV, whereas the majority (57.8%) exhibited false-negative signals. Similar trends were observed for the 1,480-bp PCR amplicon (6.6% of the probes were in classes I to IV). In contrast, after hybridization of fragmented rRNA, the percentage of probes in classes I to IV rose to 83.1%. Likewise, when DNA target sizes were reduced from 1,480 bp to 45 bp, this percentage increased approximately 14-fold. Overall, microarray hybridization efficiencies and specificity were improved with fragmented rRNA (20 to 100 bp), short PCR amplicons (<150 bp), and synthetic targets (45 to 56 bp). Such an understanding is important to the application of DNA microarray technology in microbial community studies.

DNA microarrays use hundreds or thousands of oligonucleotide probes (∼15 to 25 nucleotides [nt]) that are immobilized on a small area of a planar surface to measure functional genes and their expression, to detect genetic mutations, and to identify microorganisms (3, 36). In the last application (18, 28, 33), probes are targeted to phylogenetic markers (e.g., rRNAs). Simultaneous hybridization of these probes against labeled nucleic acids prepared from any given environmental sample allows the identification and monitoring of virtually all dominant microbial populations in a sample at a time (7). Nevertheless, the rRNA gene-based (phylogenetic) oligonucleotide microarray technique still faces technical challenges related to detection sensitivity and hybridization specificity. The former usually refers to the minimum amount of target that can be reproducibly detected by individual probes in a given environmental sample, and the latter refers to the ability of the DNA microarray technique to differentiate targets from nontargets or to discriminate closely related DNA or rRNA sequences that may possibly differ by only one base pair (17, 30).

Detection sensitivity is known to be influenced by various experimental factors (12, 28). Poor sensitivity (inefficient hybridization) is primarily due to the low abundance of nucleic acid targets from certain microbial populations (12) or to the effects of steric hindrance and surface electrostatic forces that impinge on the ability of the targets to access the probes (27, 31). Poor detection sensitivity resulting from low target concentrations is usually remedied by PCR amplification prior to hybridization (18, 32), even though these assays can be affected by the biases associated with enzymatic amplification (32). To reduce the effect of steric interference on hybridization of targets to planar surfaces (e.g., glass and silicon), spacer molecules with a length of more than 50 Å can be used to physically separate the probes from the microchip surface (24, 27).

The hybridization specificity of rRNA or amplified rRNA genes from unknown genetic backgrounds can also be compromised by the highly conservative nature of rRNA molecules in all microorganisms (36). This is mainly because the probes on a microarray are subjected to the same washing procedures (e.g., buffers, salt concentrations, and temperature). Because probes differ in sequence composition and thermodynamic characteristics, such a practice can reduce the resolution in differentiating targets from nontargets that are different by one nucleotide. Strategies to overcome such problems include the acquisition of melting curves for every individual probe (17); the addition of tetramethylammonium chloride, which equalizes the melting temperature of different probes by stabilizing the AT base pairs, to the hybridization solution (19); or the use of multiple probes to target a specific group of microorganisms (33).

Both detection sensitivity and hybridization specificity can be simultaneously affected by the secondary structures of the rRNA molecule and single-stranded DNA molecules in all rRNA gene-based microarrays. Secondary structure formation within the targets can reduce the binding constant of a specific probe by as much as 10⁵ to 10⁶ times (15), leading to an increase in false-negative signals and a decrease in hybridization specificity (1, 29). To prevent these from occurring, several approaches have been adopted. These include the use of helper oligonucleotides (24) and a two-probe proximal chaperon detection system (28) to mitigate the effects of target secondary structure hindrances, an appropriate labeling method (9), and a protocol to achieve optimized target lengths (23, 29, 34).

Since long targets can form secondary and tertiary structures that hinder efficient probe-target duplex formation, the sizes of the rRNA molecule and its amplicon are often reduced via chemical, enzymatic, or thermal fragmentation methods (13, 17, 23, 25, 28). A few studies have adopted this approach prior to hybridization (6, 11, 13, 17, 20, 28), but none of these studies have systematically investigated the effects of target length on the hybridization specificity and, possibly, detection sensitivity. To systematically address this question in this study, an array consisting of 166 probes encompassing almost the entire 16S rRNA gene of Escherichia coli K-12 (10) was prepared and hybridized with different targets, including unfragmented native rRNA, fragmented rRNA (∼20 to 100 bp), various PCR amplicons (ranging from 93 to 1,480 bp), and three synthetic single-stranded DNA targets (45 to 56 bp).

MATERIALS AND METHODS

Bacterial strain.

An Escherichia coli K-12 strain (NCIMB 10083) was obtained from the National Collection of Industrial, Marine and Food Bacteria (United Kingdom). E. coli cells were grown overnight on R2A agar at 37°C, harvested, and used for nucleic acid extraction.

Extraction, fragmentation, and labeling of native rRNA.

rRNA of E. coli cells was extracted and purified according to the protocol described by the manufacturer (RNAwiz, Ambion, TX). Initially, the cell pellet and 0.5 g of sterile glass beads (0.1 mm in diameter) were resuspended in 1 ml of RNA isolation reagent (RNAwiz) and mechanically disrupted using a Mini-BeadBeater (Biospec, Bartlesville, OK). After purification, the concentration and purity of extracted RNA were determined by measuring the absorbance with a spectrophotometer (Beckman Coulter, Fullerton, CA). Total RNA was visualized by agarose gel electrophoresis.

RNA fragmentation was carried out in duplicate by using alkaline- and metal ion-catalyzed methods (4, 5, 17). For alkaline-catalyzed fragmentation, 20 μl of RNA (∼10 μg) was digested with 50 mM of sodium hydroxide (NaOH) at 55°C for 20 min. The hydrolysis reaction was stopped by adding 1 μl of 1 N acetic acid. For metal ion-catalyzed fragmentations, the native RNA was incubated with 10 mM of zinc sulfate (ZnSO₄) or zinc chloride (ZnCl₂) in 25 mM of Tris-HCl (pH 7.4) at 60°C for 30 min. The reactions were stopped with 12 mM EDTA (pH 8). The fragmented RNA was then washed by ethanol precipitation. The length of fragmented RNA was determined by electrophoresis in a 10% polyacrylamide gel. Fragmented RNA (∼5 μg) was labeled using the Mirus Label IT Cy3 labeling kit (Mirus Bio Corporation, Madison, Wis.) according to the protocol provided by the manufacturer. The labeled RNA was purified through ethanol precipitation and dissolved in 20 μl of RNase-free water. The RNA concentration and labeling efficiency were evaluated by measurement of absorbance at 260 nm and 550 nm, respectively, with a spectrophotometer (Beckman Coulter).

DNA extraction, PCR amplification, and labeling of amplicons.

Genomic DNA of E. coli K-12 was prepared as described previously (16). In total, 13 different lengths of the 16S rRNA gene were prepared through PCR amplification using various sets of forward and reverse primers (Fig. 1A and B). These PCR products were terminally labeled using forward primers containing Cy3 at the 5′ end. PCR mixtures (50 μl) contained 1× PCR buffer (Invitrogen, San Diego, CA), 200 μM deoxynucleoside triphosphate mix, 2 mM MgCl₂, 0.2 μM of each primer, 1.25 U of Taq DNA polymerase (Promega, Madison, Wis.), and approximately 50 ng of DNA template. PCR amplification was carried out using a Bio-Rad iCycler (Hercules, CA) under the following thermal program: initial denaturation (95°C, 3 min); 30 cycles of 95°C (45 s), 54°C (45 s), and 72°C (45 s); and final extension (72°C, 3 min). PCR products were confirmed by electrophoresis using a 1% agarose gel and then concentrated using a Microcon YM-30 (Millipore, Billerica, MA) according to the protocol provided by the manufacturer. In addition, internal fluorescent labels were used for three different PCR products amplified from different 16S rRNA regions: bp 338 to 530 (193 bp), bp 338 to 431 (94 bp), and bp 431 to 530 (100 bp). The PCR conditions used were the same as described above except for the deoxynucleoside triphosphates, which included 200 μM of dATP, dTTP, and dGTP; 150 μM of dCTP; and 50 μM of Cy3-dCTP (Amersham Sciences, United Kingdom) in 50-μl PCR mixtures. All PCR amplicons are designated “T_x-y,” where x-y indicates the region of the 16S rRNA gene amplified (e.g., T_11-1490 for the PCR amplicon from bp 11 to 1490). Probes are designated “P_z,” where z stands for the probe number in Fig. 1C (e.g., P₁₀ represents probe 10).

FIG. 1.

(A) Forward and reverse primer sequences used during PCR amplification. (B) PCR amplicons with lengths varying from approximately 100 to 1,500 bp used as targets for hybridization.

In addition to PCR-amplified targets, three fluorescently labeled (Cy3 at the 5′ end) targets encompassing bp 431 to 486 (56 bp), bp 486 to 530 (45 bp), and bp 465 to 509 (45 bp) (i.e., T_431-486, T_486-530, and T_465-509) were used. These fluorescently labeled and unlabeled primers/targets were synthesized by Operon Inc. (Alameda, CA).

Oligonucleotide probes.

We adapted 166 probes that were used in the study by Fuchs et al. (10). They encompassed almost the entire 16S rRNA region of E. coli K-12 (positions 91 to 1454) and had an overlapping region of 5 to 13 nt between any two given adjacent probes. All probes were modified at the 5′ end with a C6 amino linker followed by a 15-mer dT spacer to minimize possible steric hindrance during hybridization. Excluding the spacer region, all probes are approximately 18-mers, with theoretical melting points of between 48 and 60°C. A control probe (5′-linker-(T)₁₅-GGGG-Cy3-3′) with a C6 amino linker at the 5′ end and a Cy3 label at the 3′end was synthesized. Both 16S rRNA probes and control probes were obtained from Operon Inc.

Microarray fabrication.

One nanoliter of each probe was spotted in triplicate in one location (see Fig. S1 in the supplemental material) onto a Euray Immobilizer microarray slide (Exiqon, Denmark) at a concentration of 25 pmol/μl, using the Biochip Arrayer (Perkin-Elmer, Wellesley, MA). The control probe was used as a position marker for the array and a reference for signal normalization. The printed microchips were incubated overnight in a humidity chamber containing filter paper prewetted with saturated NaCl solution. The slides were subsequently dried at 37°C for 2 h before storage at room temperature in a dark desiccator.

Hybridization.

All hybridization experiments with unfragmented native rRNA, fragmented native rRNA, PCR amplicons, and three different synthetic oligonucleotides were repeated at least twice. Prior to hybridization, these targets were treated with or without prior heat denaturation (i.e., heating at 95°C for 5 min [65°C for RNA targets] and chilling on ice for 2 min). The hybridization solution consisted of 3 μg of labeled targets, 20 mM Tris-HCl (pH 8.0), and 0.9 M NaCl. The hybridization solution was then applied to the microchip, and hybridization was carried out in a dark humidity chamber at 25°C for 16 h. The microchip was washed twice with a buffer solution containing 2× SSC (1× SSC is 0.15 M NaCl plus 0.015 M sodium citrate) and 0.1% sodium dodecyl sulfate for 2 min and then once with a solution containing 0.1× SSC and 0.1% sodium dodecyl sulfate for 2 min. A final rinse was carried out in a solution containing 0.1× SSC for 1 min. All washes were conducted at room temperature. Slides were then dried with nitrogen gas.

Quantitation of hybridization signals.

Hybridized microchips were scanned using ScanArray Express (Perkin-Elmer, Wellesley, MA) at a resolution of 5 μm at 550 nm. The photomultiplier tube gain was set at 60% for the scanning of all microchips to allow for unbiased comparison across experiments and to ensure that no signal saturation occurred with the Cy3-labeled control probes. Quantification and evaluation of spot intensities were carried out using the ScanArray Express software (Perkin Elmer). Spot intensities were obtained by subtracting background intensities from raw intensities for individual spots. The average intensity for the three replicates of each probe was calculated. The relative intensity was obtained by expressing the average intensity of each probe relative to that of the Cy3-labeled control probe. These relative intensities were subsequently classified into six classes of brightness: class I (81 to 100% relative to the control probe), class II (61 to 80%), class III (41 to 60%), class IV (21 to 40%), class V (6 to 20%), and class VI (0 to 5%) (10). For comparison purposes, good hybridization efficiency was defined as the accumulated percentage of probes in classes I to IV. Probes with relative intensities in classes V and VI were regarded as having weak and false-negative signals (inefficient hybridization), respectively. False-positive signals (cross-hybridization) were defined by expressing the number of probes (excluding those which target at T_x-y) with relative intensities in classes I to IV as a percentage of the total number of probes, not including those which target at T_x-y. Specificity was evaluated according to the extent and occurrence of both false-positive and false-negative results.

Probe affinity.

A probe affinity model (35) was used to evaluate the relationship between the normalized fluorescence intensity and the calculated Gibbs free energy changes of probe-RNA hybridization (ΔG°₁), probe folding (ΔG°₂), and 16S rRNA intramolecule folding (ΔG°₃) after hybridization with native rRNA. The normalized intensities of probes were obtained by the calculation of (raw intensity of individual probes − lowest raw intensity)/(highest raw intensity − lowest raw intensity) in individual hybridization. The values of individual Gibbs free energy changes were adopted from reference 35. Pearson product-moment correlation coefficients were used to quantify the degree of relationship.

Statistic analysis.

The paired two-sample Student t test and analysis of variance were performed using Microsoft Office Excel 2003 (α = 0.0001).

RESULTS

Extraction and fragmentation of native rRNA.

A clear separation among the 5S (ca. 120-nt), 16S (ca. 1,500-nt), and 23S (ca. 3,000-nt) rRNAs extracted from active E. coli cells was observed on an agarose gel (Fig. 2A). The length of the major fragmented rRNAs through NaOH- and ZnCl₂-catalyzed reactions was observed to be between 20 nt and 90 to 100 nt by using a 10% polyacrylamide gel (Fig. 2B and C). The ZnSO₄-catalyzed reaction yielded fragment sizes ranging from approximately 36 nt to a few hundred nucleotides, which were larger than those generated by the NaOH and ZnCl₂ methods.

FIG. 2.

(A) Denaturing formaldehyde gel electrophoresis (1.2%) showing unfragmented native rRNA; (B) 10% polyacrylamide gel electrophoresis of rRNA, with 20 and 60 nt of synthetic oligonucleotides and 100 bp of DNA ladder as markers; (C) 10% polyacrylamide gel electrophoresis of rRNA, with 21, 36, and 90 nt of synthetic oligonucleotides as markers.

Hybridization with unfragmented and fragmented native RNAs.

Microchips were separately hybridized with nondenatured unfragmented native rRNA, denatured unfragmented native rRNA, and fragmented native rRNA. Clear improvement in hybridization signals was qualitatively observed with the scanning images of DNA microchips hybridized with nondenatured native rRNA and NaOH-fragmented rRNA (see Fig. S1 in the supplemental material). To systematically compare the hybridization efficiencies among those three treatments, the hybridization signal intensities of all probes were compared with that of the positive control probe and categorized into the six brightness classes (I to VI) that were used by Fuchs et al. (10). When nondenatured native rRNA was used as the target, none of the 166 probes had relative intensities in classes I and II, and only three probes (P₂, P₉₁, and P₉₂) exhibited relative intensities in class III. The widely used bacterial probe Eub338 (P₂₉) had a relative intensity of 23.9% and, together with 19 other probes, was categorized in class IV. In totality, only 23 (13.9%) of the 166 probes were within classes I to IV. For the remaining probes, 37.3% were in class V and 48.8% in class VI (false-negative signals).

Thermal denaturation was used to minimize possible inhibitory effects of secondary structures on hybridization efficiency. A significant change (P < 0.00001 by Student's t test) in the hybridization efficiency was observed (Fig. 3A). Out of the 166 probes, 10 probes (6%) exhibited more than a 10% increase in relative intensities compared to those without thermal denaturation, 14 probes showed decreases in relative intensities, and none had relative intensities in classes I to III. The percentage of probes in class IV increased slightly from 12.0% to 17.5%, that of probes in class V increased to 81.9%, and that of probes in class VI decreased to 0.6% (Table 1).

FIG. 3.

Comparison of relative intensities (against the Cy3 control probe) obtained from unfragmented (Unfrag) rRNA with and without heat denaturation (A) and from fragmented (Frag) rRNA (B). Probes with intensities in classes I to IV are above the dotted line.

TABLE 1.

Classification of relative intensities of microarray hybridizations with different sizes of E. coli 16S rRNA, PCR-amplified targets, and synthetic oligonucleotides

Target	Length (bp)	Probesa	% of total probes in brightness class:
			I	II	III	IV	V	VI	False positiveb
rRNA (denatured), unfragmented	∼1,500	P1-166	0	0	0	17.5	81.9	0.6	NAd
rRNA (undenatured)
Unfragmented	∼1,500	P1-166	0	0	1.8	12.0	37.3	48.8	NA
NaOH fragmented	20 to 90-100	P1-166	9.0	13.9	28.3	34.9	12.0	1.8	NA
ZnCl2 fragmented	20 to 90-100	P1-166	16.3	15.7	28.9	24.1	11.4	3.6	NA
ZnSO4 fragmented	36 to hundreds	P1-166	0	0	3.6	8.4	56.0	31.9	NA
DNA amplicon (denatured)
T11-1490	1,480	P1-166	0.6	1.2	0.6	4.2	13.9	79.5	NA
T11-530	520	P1-56	0	0	0	0	26.8	73.2	0
T519-907	389	P56-106	2.0	2.0	2.0	15.7	5.8	72.5	0
T785-1088	304	P92-126	0	14.3	8.6	5.7	11.4	60.0	2.3
T1100-1392	293	P128-160	3.0	0	6.1	6.1	18.2	66.6	1.5
T338-530	193	P29-56	3.6	0	28.6	25	28.5	14.3	3.6
T338-530c	193	P29-56	0	7.1	28.6	28.6	21.4	14.3	3.0
T155-338	184	P8-28	4.8	19	4.8	9.5	28.6	33.3	2.1
T1248-1392	145	P145-160	12.5	31.3	0	18.7	25	12.5	0.7
T645-785	141	P71-91	14.3	4.8	33.3	23.8	23.8	0	0
T431-530	100	P40-56	0	5.9	29.4	23.5	29.4	11.8	4.7
T431-530c	100	P40-56	0	0	5.9	58.8	29.4	5.9	3
T1392-1490	99	P160-166	14.3	0	42.8	14.3	28.6	0	0
T338-431	94	P29-39	9.1	9.1	18.2	27.3	27.3	9.0	1
T338-431c	94	P29-39	0	18.2	18.2	36.4	18.2	9.0	2.4
T996-1088	93	P116-126	0	27.3	18.2	36.3	18.2	0	1.3
Synthetic target (undenatured)
T431-486	56	P40-48	0	11.1	33.3	55.6	0	0	0
T465-509	45	P47-54	0	37.5	62.5	0	0	0	0
T486-530	45	P49-56	0	0	33.3	55.6	0	11.1	0

^aProbes were designed to target at the respective target fragment, T_x-y.

^bValues were obtained by expressing the number of probes (excluding those targeting at T_x-y) with normalized intensities in classes I to IV as a percentage of the total number of probes (excluding those targeting at T_x-y).

^cInternally labeled targets.

^dNA, not available.

The relative intensities obtained using unfragmented rRNAs and fragmented rRNAs generated by different methods were further compared (Fig. 3A and 3B). Hybridizations with NaOH- and ZnCl₂-fragmented RNA resulted in a significant improvement in hybridization efficiency (P < 0.00001 by Student's t test). A large proportion (86.1% and 84.9% for NaOH, and ZnCl₂, respectively) of the 166 probes exhibited relative intensities in classes I to IV (Table 1), a 6.1- to 6.2-fold increase compared to those with unfragmented rRNA. At the same time, ∼89- and ∼116-fold increases in the relative intensities were noted among the individual probes in the hybridization with NaOH- and ZnCl₂-fragmented RNAs, respectively. Comparing the relative intensities of individual probes using unfragmented rRNAs as targets, about 91.6 to 94.6% of the probes exhibited an increase in relative intensities of more than 10%, 4.8 to 6.0% had less than a 10% increase in relative intensities, and only 0.6 to 2.4% (i.e., 1 to 4 out of 166 probes) exhibited a slight drop (<4.7%) in relative intensities. These results strongly suggested that the target fragment length played a crucial role in the hybridization efficiency of oligonucleotide microarrays.

Comparison among hybridization patterns.

To further allow comparisons to be made among different hybridization patterns, the relative intensities of individual probes from individual hybridizations were normalized with respect to the highest relative intensity and lowest relative intensity (Fig. 4). The overall hybridization pattern obtained using unfragmented rRNA molecules was observed to be very similar, to some extent, to the results obtained using fluorescence in situ hybridization (FISH) (10). The hybridization patterns obtained using fragmented rRNA molecules were nearly identical. Five regions, i.e., helices 18 (E. coli nucleotide numbering, position 439 to 460), 23 and 24 (position 654 to 678), 21 and 26 (position 763 to 780), 45, 43, and 39 (position 1176 to 1193), and 49 (position 1410 to 1427), targeted by the probes were observed to have consistently high signal intensities (normalized intensities of >60%), and 25 segments had low intensities (normalized intensities of <20%). Figure 4B further indicates that within the same helix region, both high and low hybridization efficiencies could be observed. For example, the front region of helix 18 (E. coli nucleotide numbering, position 439 to 460) targeted by probes P_41-43 had normalized intensities of 62 to 82%, but the back region of helix 18 (E. coli nucleotide numbering, position 455 to 495) targeted by probes P_46-49 had normalized intensities of 0 to 15%. If the RNA was assumed to be randomly fragmented to 20 to 100 nt, the special intensity patterns of 166 probes hybridizing with fragmented native RNA of E. coli could be explained to be sequence dependent.

FIG. 4.

(A) Comparison between the normalized intensities of the 166 probes obtained from microarray hybridization using native RNA as the target and those obtained using whole-cell FISH (10). (B) Comparison among the normalized intensities of those 166 probes obtained from microarray hybridizations with fragmented RNAs prepared using three different methods. The numbers indicated the helix numbering of E. coli as reported by Fuchs et al. (10).

Effect of overall Gibbs free energy change.

To explain the consistent intensity patterns observed with fragmented 16S rRNA, the probe affinity model proposed by Yilmaz et al. (35) was used to evaluate the relationship between normalized intensity and overall Gibbs free energy (ΔG°_overall), a thermodynamic parameter that considers the occurrences of DNA-RNA (ΔG°₁), DNA-DNA (ΔG°₂), and RNA-RNA (ΔG°₃) interactions during a hybridization experiment. The overall measure of the thermodynamic affinity was obtained by calculating the ΔG°_overall as a function of ΔG°₁, ΔG°₂, and ΔG°₃. The lower (more negative) ΔG°_overall represented a greater potential for the formation of a probe-RNA duplex and thus higher brightness (35). Table 2 summaries the correlation coefficients between a calculated thermodynamic variable and the normalized intensity. The correlation coefficients for normalized intensities and ΔG°_{overall-mixed} were −0.39, and −0.41 to −0.45 for hybridization with unfragmented and fragmented rRNAs, respectively. These values were higher (more positive) than that observed in FISH experiments (−0.79) (35). As calculated by different methods, the values of ΔG°₃ were all positive using intact 16S rRNA as a target, but were shifted to negative with fragmented 16S rRNA, suggesting an effect of folding with the intact rRNA on hybridization efficiencies. Furthermore, positive values obtained for ΔG°₂ also suggested an effect of probe self-structure folding on hybridization efficiency. We further observed a moderate correlation between the predicted probe-RNA affinity (i.e., ΔG°₁) and the normalized intensity (−0.61, −0.57, and −0.54 with RNAs fragmented by NaOH, ZnCl₂, and ZnSO₄, respectively) (Fig. 5). The correlations were higher than those (−0.39 to −0.46) obtained using intact 16S rRNA in microarray and FISH analyses. These results supported that a reduction in the length of rRNA molecules could largely enhance the interaction between probes and fragmented rRNA, leading to higher hybridization intensities.

TABLE 2.

Correlation coefficients describing the relationship between calculated Gibbs free energy changes and normalized fluorescence intensities of probes hybridizing with E. coli 16S rRNA

Free energya	Correlation with intensity obtained with:
	Unfragmented rRNA	rRNA fragmented by:			FISHb
		NaOH	ZnCl2	ZnSO4
ΔG°1	−0.39	−0.61	−0.57	−0.54	−0.46
ΔG°2	0.43	0.31	0.26	0.29	0.11
ΔG°3-i	0.02	−0.09	−0.01	−0.04	0.18
ΔG°3-ii	0.04	−0.11	−0.03	−0.08	0.27
ΔG°3-iii	0.01	−0.14	−0.09	−0.15	0.30
ΔG°3-iv	0.04	−0.03	0.02	−0.03	0.30
ΔG°overall-i	−0.41	−0.50	−0.52	−0.47	−0.57
ΔG°overall-ii	−0.39	−0.44	−0.46	−0.40	−0.62
ΔG°overall-iii	−0.20	−0.14	−0.17	−0.09	−0.48
ΔG°overall-iv	−0.18	−0.18	−0.21	−0.16	−0.43
ΔG°overall-mixed	−0.39	−0.44	−0.45	−0.41	−0.79

^aSee calculated changes of Gibbs free energy in reference 35.

^bFrom reference 35.

FIG. 5.

Relationship between normalized intensities from microarray hybridizations with fragmented RNAs and Gibbs free energy changes (ΔG°₁) of probe-RNA duplex-forming reactions. The solid line is the linear regression line obtained.

Hybridization with PCR amplicons.

Thirteen sets of PCR amplicons with lengths varying from 93 to 1,480 bp were used as targets (Table 1). In general, the reduction in fragment lengths from a range of 293 to 1480 bp to 184 to 193 bp led to substantial improvements in hybridization efficiencies. The percentage of probes conferring intensities in class VI decreased from 60.0 to 79.5% to 14.3 to 33.3%. The fraction of probes classified in class VI was further reduced to 0 to 12.5% by reducing the target length to 93 to 145 bp. More than half of the probes conferred intensities in classes II to IV. However, the reduction in target length did not result in actual reductions of false-positive results. The highest percentage of false-positive signals was observed with target T_431-530 (4.7%), followed by T_785-1088 (2.3%), T_1100-1392 (1.5%), T_338-530 (3.6%), T_155-338 (2.1%), T_1248-1392 (0.7%), T_338-431 (1%), and T_996-1088 (1.3%).

The signal intensities were further compared among other end-labeled PCR amplicons sharing the same overlapping region. Figure 6 shows that T_11-1490 (1,480 bp), T_1100-1392 (293 bp), and T_1248-1392 (145 bp) share a region (position 1248 to 1392, [E. coli numbering]) specific for P_145-160. When T_11-1490 and T_1100-1392 served as the targets, the relative intensities of P_145-160 were all categorized in classes V and VI. In contrast, when T_1248-1392 was the target, the overall relative intensities were greatly improved: 10 (62.5%) of these 16 probes were in classes I to IV, 4 (25%) in class V, and only 1 (12.5%) in class VI.

FIG. 6.

Relative intensities obtained from the hybridization of various PCR amplicons with lengths varying from 94 to 1,480 bp.

Likewise, T_11-1490 (1,480 bp), T_11-530 (520 bp), and T_338-530 (193 bp) shared an overlapping region (position 338 to 530 [E. coli numbering]) covered by P_29-56 (Fig. 6 and 7). Using the longer targets T_11-1490 and T_11-530, the relative intensities conferred by P_29-56 were mostly categorized in classes V and VI. Of the 28 probes shared by these targets, 24 (85.7%) and 22 (78.6%) were scored as false-negative signals (class VI) with T_11-1490 and T_11-530, respectively. In contrast, when T_338-530 was the target, only four probes were in class VI. The majority of probes, 16 (57.2%), had relative intensities classified in classes I to IV, and 8 probes (28.5%) had relative intensities in class VI.

FIG. 7.

Relative intensities of P_29-56 with T_11-1490 (1,480 bp), T_11-530 (520 bp), T_338-530 (193 bp), T_338-431 (94 bp), T_431-530 (100 bp), T_431-486 (56 bp), T_465-509 (45 bp), and T_486-530 (45 bp).

The relative intensities of P_29-56 obtained using T_338-530 (193 bp) were further compared with those obtained using two shorter PCR amplicons, T_338-431 (94 bp) and T_431-530 (100 bp). The scanned image with T_338-530 (see Fig. S2A in the supplemental material) showed the occurrence of false-negative signals, which were subsequently improved using targets with shorter lengths. For T_338-431 (see Fig. S2B in the supplemental material), the number of those 11 probes in classes I to IV rose to 7 (63.7%), and only 1 probe (P₃₉) had a relative intensity (1.1%) in class VI. For T_431-530 (Fig. 7), 10 probes (58.8%) were categorized in classes I to IV, but 5 (29.4%) and 2 (11.8%) still remained in classes V and VI, respectively (e.g., P_47-49).

Furthermore, strong false-positive signals were observed outside the targeted hybridization regions of T_338-431, T_431-530, and T_338-530. These include, for example, probe P₂₈, which had only 13 out of 17 nucleotides complementary to the 5′ end of T_338-431 (see Fig. S2B in the supplemental material); probe P₁₆₁ (E. coli numbering, position 1392 to 1409) after being hybridized with T_338-431 and T_338-530; and P₁₃₁ and P₁₃₂ when the target was changed from T_338-530 to T_431-530 (see Fig. S2B and S2C in the supplemental material). It is likely that a decrease in target fragment size could lead to an increase in cross-hybridization.

We further investigated whether the signal intensities of P_29-56 could be improved by using internally labeled T_338-530 (193 bp), T_338-431 (94 bp), and T_431-530 (100 bp). However, only slight increases in relative intensities were observed for some of the probes when internally labeled targets were hybridized (Table 1). Overall, the use of internal labeling did not significantly enhance probe intensities for this region (P = 0.57 by Student's t test). Due to the high cost involved with internal labeling, the use of end-labeled products is preferred.

Hybridization with synthetic DNA oligonucleotides.

To further understand the effect of length on the low hybridization efficiency for the region from position 468 to 495 (P_47-49), three synthesized oligonucleotides, i.e., T_431-486 (56 bp), T_465-509 (45 bp), and T_486-530 (45 bp), encompassing positions 431 to 530 were used in microarray hybridization. As shown in Fig. 7, the relative intensities of P₄₇ and P₄₈ were greatly improved, from 6.2% and 2.4%, respectively, using T_431-530 as the target to 54% and 61.4%, respectively, using T_431-486 as the target. Similarly, using T_465-509, relative intensities for P_47-54 were consistently high and were in the range of 51.9 to 69.1%. These intensities were approximately 1.5-fold higher than those obtained with T_431-530 (100 bp). However, using single-stranded target T_486-530, the relative intensities of P_49-52 still could not be substantially improved compared to those using T_431-530. These observations suggested that a reduction in fragment length to approximately 50 nt could greatly enhance but not completely resolve the issue related to low hybridization efficiency.

DISCUSSION

In DNA microarray studies, the secondary structures and the lengths of nucleic acid molecules are known to affect the rate and efficiency of target-probe duplex formation (8, 13, 20, 24), leading to reduced hybridization efficiencies and false-negative signals. To address this, solutions focusing on target fragmentation, target denaturation, and helper oligonucleotide probes have been proposed and demonstrated (20, 24, 25, 28). In this study, the effect of target fragmentation and length on microarray hybridization efficiency was carefully evaluated by minimizing any possible effect on signal intensities due to other factors. To do so, all probes were modified with a same spacer (15 dTs) to physically separate the oligonucleotide probe from the slide surface, alleviating possible steric interference (27) with hybridization efficiencies and specificities. All probe and target concentrations and the hybridization temperature were also kept constant. The probes were designed with theoretical melting temperatures ranging from 48 to 60°C (10). As a result, approximately two-thirds of the ΔG°₂ values (−2.9 to 4.6 kcal/mol) are positive, suggesting that probe self-folding should have a minimal effect on hybridization efficiency (35). Thus, the cause of any variance in hybridization efficiencies and specificities should be primarily related to target length and type (i.e., RNA or DNA amplicons/fragments) and target secondary structure.

Our results clearly indicate that hybridization efficiency and detection sensitivity can be greatly improved by shortening amplified 16S rRNA gene and native rRNA targets to smaller sizes. Fragmenting native rRNA targets to 20 to 100 nt by using NaOH-and ZnCl₂-catalyzed methods not only enhanced hybridization signal intensities by a factor of 6.1 to 6.2 but also reduced false-negative signals. It is interesting to observe that different fragmentation methods could result a significant difference in hybridization signal intensities (Fig. 3) (P < 0.00001 by analysis of variance), but the normalized hybridization patterns were nearly identical with those 166 probes (Fig. 4). The correlation analysis in Table 2 revealed that the hybridization intensity with fragmented RNA correlated better with ΔG°₁ than with ΔG°_overall. Since ΔG°_overall is a function of ΔG°₁, ΔG°₂, and ΔG°₃ (35), this result suggested that microarray hybridizations with short rRNA fragments were more dependent on target sequence than on the competition between probe-target interaction and RNA self-folding. In addition to RNA folding interactions, hybridization kinetics could also play a critical part in determining the hybridization efficiencies (35) and could be the primary factor contributing to the large scattering of the data in Fig. 5. Other possible factors included nonspecific hybridization and intermolecular RNA-RNA interactions.

Similar to the RNA results described above and to previous findings (27, 29), this study also showed enhanced hybridization efficiencies and reduced false-negative signals for DNA fragments when the target length was shortened (Table 1). A reduction in the length of the DNA target from 1,480 bp to approximately 184 to 193 bp could minimize false-negative signals by a factor of 2.4 to 5.6. This factor further increased to at least 6.7 or 8.8 when the target length was reduced to 93 to 145 bp (DNA amplicons) or 45 to 56 nt (synthetic targets), respectively. Interestingly, it was observed that P_47-49 targeting the region from nt 468 to 495 of E. coli rRNA consistently displayed low intensities, at classes V and VI, when fragmented rRNA targets (Fig. 3B) and DNA fragments of >100 bp in length (Fig. 7) were used. Similar observations were reported when FISH with rRNA-targeted oligonucleotides was used (2, 10). These low hybridization efficiencies were likely associated with the high (more positive) ΔG°₁ values (−14.6 to −17.2 kcal/mol) predicted for the probe-RNA target duplexes. The use of a short single-stranded oligonucleotide as the target improved the hybridization intensities.

Furthermore, a reduction in target length did not necessarily ensure probe specificity, as a slight increase in false-positive signals, up to 4.7%, of those 166 probes was observed with defined regions of DNA fragments (Table 1; Fig. 7). To minimize the occurrence of false-positive signals, studies have increased the stringency of washing conditions, conducted melting curve analysis for all probe-target duplexes (17), or used multiple probes to target a specific group of microorganisms (33).

Our results, as well as others (27, 29), disagree with the findings of Lane and coworkers (14), which suggested that hybridization efficiency could not be improved by reducing the length of PCR amplicons. In the study by Lane and coworkers, fewer probes were used (n = 8), and thus the result was insufficient to reflect the range of hybridization efficiencies with different PCR amplicons. Second, the PCR amplicons were too long (162 to 1,517 bp) to achieve good hybridization efficiency based on our findings (Table 1). Third, their microarray used oligonucleotide probes without a spacer to mitigate steric hindrance, which usually prevents the targets from approaching the probes. In fact, Lane and coworkers presented findings, which also support our conclusion, that random fragmentation (i.e., a reduction in target length) and labeling through nick translation chemistry could disrupt secondary structure interference and thus greatly improve hybridization efficiency.

Currently, different chemical methods are used to fragment target RNA and DNA molecules. For RNA fragmentation, fragmentation catalyzed by alkali (e.g., NaOH) and divalent ions such as Zn²⁺ is a simple and inexpensive strategy and is commonly used (6, 20, 25, 28). Alternatively, RNA could be fragmented using radical-generating complexes such as 1,10-phenanthroline-Cu(II) and Fe(II)-EDTA (13). Although the constant hybridization pattern observed for three fragmentation methods could be attributed to the dependence of sequences, there is a likelihood that the fragmentation procedures used here might not result in random fragmentation and could be affected by the loop-and-stem structure associated with the target.

Fragmenting of DNA targets or PCR amplicons has been achieved by incubation under acidic conditions (e.g., in the presence of formic acid) (11, 25, 26) or by cleavage with enzymatic nucleases (e.g., DNase I, endonuclease V, and uracil-DNA-glycosylase) (21, 22, 32, 33) or with “chemical nucleases” (e.g., hydroxyl radicals) (13). Fragmentation by DNase I generally gave a fragment size of about 50 to 450 bp and good hybridization efficiencies (32, 33). The use of hydroxyl radicals for simultaneous labeling and fragmentation was simple and inexpensive but could potentially damage the nucleobases and yield small amounts of fragmented DNA targets (32), subsequently affecting the hybridization between the fragmented targets and oligonucleotide probes. Furthermore, short DNA targets of 50 to 450 bp in length could be produced through Klenow amplification to improve the microarray hybridization efficiency (32).

A few studies have also used thermal denaturation of RNA targets prior to DNA microarray hybridization to reduce any possible interference posed by secondary structures (12, 28). Our results indicate that thermal denaturation of RNA targets could indeed lead to changes in hybridization efficiencies and probe specificity (Fig. 3A), even though 82.5% of the probes still exhibited hybridization intensities in classes V and VI. This observation also suggested that possible reassociation of RNA stem-loop structures could have occurred before or during the hybridization event (29). Unlabeled “helper” oligonucleotides are another alternative to minimize the impact of secondary structures on hybridization specificity and efficiencies (24). By binding to the site adjacent to the probe-targeting region, these “helper” molecules resolve and reorganize the secondary structures of the target, leading to an increase in the accessibility of the target sites to the respective probes. However, this approach is limited by the high cost associated with the additional helper probes. Furthermore, designing helper probes with the same specificity as the corresponding probe is not always possible. Therefore, the target fragmentation procedure is still the preferred approach to improve the hybridization efficiency and sensitivity of DNA microarray analysis.

In summary, substantial increases in hybridization efficiencies could be achieved with short RNAs (20 to 100 nt), DNA amplicons (93 to 145 bp), and synthetic targets (45 to 56 bp). A reduction in target length also resulted in a reduction in false-negative signals, but extremely short fragments could adversely lead to a slight increase in false-positive signals. Thus, there is a need to compromise between good hybridization efficiencies and hybridization specificities.