Probe design and specificity. (A) Schematic of the miRNA microarray probe design. Unmodified microarray probe (black) hybridized to miRNA target (red) is shown in (1), where the hybridizing sequence synthesized on the microarray (the probe) is connected to the glass surface by a T₁₀ stilt (squiggly line). By adding a G on the 5′ end of the probe, one more G-C pair is added to the probe–target interaction (2). When necessary, destabilization of the probe–target hybrid is achieved by eliminating base-pairing from the 5′ end of the miRNA, by shortening the probe from the 3′ end (3). All probes were synthesized with and without hairpins (4), which can increase the specificity toward the target miRNA and can potentially increase the stability of probe–target interactions. (B) Probe–target sequence specificity of the human let-7 family. Each miRNA in the let-7 family was individually labeled and hybridized. The total signals reported by all the probes were normalized to the perfect match probe–target hybrid for each microarray. (C) Hybridization of FlashPAGE-enriched small RNA versus total RNA from placenta. Background-subtracted microarray signals from 60 ng of FlashPAGE-separated small RNAs (from 570 ng of total RNA) (x-axis) are plotted versus signals from 120 ng of total RNA (y-axis). Pearson correlation of the two samples is 0.992. The FlashPAGE separated and the total RNA were from the same placenta tissue and were labeled using the methods described here. Data are from both hairpin and nonhairpin probes.

miRNA expression profiles of human tissue total RNAs. All data were background-subtracted signals from hairpin and nonhairpin probes, obtained from 120 ng of total RNA. (A) miRNA expression profiles of human brain tissue from two separate labeling reactions. Pearson correlation for the data shown is 0.994. Highly reproducible miRNA profiles were obtained for all tissues examined. (B) Differential expression profile between placenta and brain. Pearson correlation between the samples is 0.238. (C) Heat map of miRNA expression profiles from human brain, breast, heart, liver, thymus, placenta, and skeletal muscle tissues. The color is proportional to the logarithm of measured signals, from blue (no detectable signal) to yellow (highest signal). Signals from Drosophila miRNA probes are on the bottom four rows of the map; the spiked-in Drosophila sequences appear as uniform yellow lines across the heat map while the sequences that were not spiked-in appear as uniform blue lines.

Validation of labeling efficiency. (A) Effect of DMSO. Five synthetic miRNAs containing different sequences were either directly labeled (blue) at 0%, 15%, or 25% DMSO or heated prior to labeling (red) at 0%, 15%, 25%, or 30% DMSO. At 25% DMSO, high labeling yields were observed independent of heat denaturation. Fifty-five separate reactions are shown. (B) Evaluation of dye bias. Box plots of labeling efficiencies of 53 different synthetic miRNAs labeled individually with Cy5-pCp and Cy3-pCp are shown. Results from 220 miRNA ligations are shown (see Supplement 1 for all detailed ligation data). The mean yield and standard error was 80±2 for Cy3 and 81±1 for Cy5. ANOVA p value of the Cy3 and Cy5 labeling efficiency distributions was 0.70, showing no difference between the labeling efficiencies of the two dyes. (C) Evaluation of sequence bias. Box plots of ligation efficiencies of the 220 reactions from B categorized according to 3′ nucleotide are shown. The mean percent yields of the labeled product and the standard errors were 82 ± 3 for A, 86 ± 2 for C, 78 ± 2 for G, and 78 ± 2 for U. ANOVA p value distinguishing the four distributions was 0.015, indicating that the labeling bias, while slight, is statistically significant. Ligations in A and B with low yields (<60%) were not reproducible, as subsequent ligations generally resulted in significantly higher yields (>80%). The reasons for these occasional instances of low yields are unclear. Infrequency and lack of reproducibility suggest such low yields are atypical experimental variations. High yields (>90%) were very reproducible.

Linear dynamic range of microarray signals. An equimolar mixture of 57 synthetic miRNAs was labeled and hybridized in 0.2 amol to 2 fmol aliquots on individual microarrays. Prelabeled 0.2 fmol of dme-miR-6 was added as a control. All miRNAs behaved similarly, except for the three indicated. Both miR-126* and miR-384 had exceptionally low calculated T_m. The unusual sequence content of miR-296 inhibits hybridization at low concentrations, resulting in a nonlinear response. The slopes of the linear portions of the curves for all miRNAs are 1.04 ± 0.01 (1 SD). Data are background subtracted signals, not normalized, summed over all probes to each miRNA. The list of miRNAs and the data used to generate the figure are included in Supplement 6.

Comparison of quantitative RT-PCR (qRT-PCR) and microarray expression profiles of 10 miRNAs. Relative expression of 10 human miRNAs was determined for seven different human tissues using both qRT-PCR and our microarray assay. Results are individually plotted for each miRNA, with qRT-PCR results in orange and microarray results in blue. The miRNA levels in each tissue are reported as the fraction of the expression level in the tissue in which that miRNA is most abundant. The qRT-PCR and the microarray results agreed on which tissue expressed the highest levels of each miRNA. The qRT-PCR reactions and microarray hybridizations were each repeated four times for each miRNA. Error bars indicate one standard deviation.