MicroRNAs (miRNAs) are extremely small RNA species that represent approximately 0.01% of the total mass of RNA in a sample (10 g total RNA contains ~1 ng miRNA). Due to their small size and low abundance, miRNAs can be extremely difficult to detect using standard array procedures. To address such limitations, Ambion scientists have developed the mirVana miRNA array system, consisting of:
Highly specific miRNA Probes (the mirVana miRNA Probe Set). The mirVana miRNA Probe Set includes amine-modified oligonucleotides with sequences complementary to the known human, mouse, and rat miRNAs. The probes have a polynucleotide spacer that ensures maximal hybridization efficiency to labeled miRNAs when the probes are arrayed on amine-reactive slides.
Robust miRNA Sample Labeling (the mirVana miRNA Labeling Kit). The mirVana miRNA Labeling Kit features a tailing procedure that adds 2050 nucleotides to the 3' ends of all miRNAs in a sample. This strategy maximizes signal from each miRNA without jeopardizing hybridization due to labeling. The 3' tail that is added is a mixture of standard and amine-modified nucleotides. Fluorescent dyes or other detectable moieties can be appended to the amine-modified miRNAs using NHS-ester derivatized dyes. The mixture of modified and unmodified nucleotides improves tailing efficiency and reduces fluorescence quenching that occurs when dye molecules are located on adjacent or nearby nucleotides.
Together the mirVana miRNA Labeling Kit and mirVana miRNA Probe Set provide the best combination of sensitivity and accuracy of any procedure for miRNA profiling. Here we validate key assay characteristics including sensitivity, specificity, and accuracy of this miRNA microarray procedure.
Highly Sensitive Limit of Detection
To determine sensitivity of the mirVana miRNA array system, we isolated the miRNA fraction from liver RNA samples then spiked in various amounts of miR-124, a brain-specific miRNA. Liver miRNA samples with and without added miR-124 were then labeled using the 3' tailing method, and each miR-124-containing sample was cohybridized with a differentially labeled liver miRNA sample missing miR-124. Analysis of raw signal intensities and the Log2(R/G Normalized Ratio) (Mean) of miR-124 expression indicated that signal significantly above slide background was detectable with femtomole amounts of miR-124 (Figure 1). This level of detection represents as little as 0.1% of the overall miRNA population in a 10 g total RNA sample.
Figure 1. miRNA Array Sensitivity. Raw intensities of miR-124 expression measured from liver miRNA samples with the indicated amounts of synthetic miR-124 spiked into the sample following fractionation. The signal is determined by background subtraction, and is depicted in signal/noise units. The heat map on the graph shows the Log2(R/G Normalized Ratio) (Mean) of miR-124 from the same samples.
High Target Specificity
A number of miRNAs have very similar sequences. For example, there are eight variants of the let-7 family that differ by one to four nucleotides. Part of the development of the mirVana miRNA array system involved creating probes and hybridization and wash conditions that minimized cross-hybridization between closely related miRNAs. While we were unable to distinguish single nucleotide mismatches without significantly reducing sensitivity, we were able to identify conditions that provide specificity and sensitivity. To test specificity, our model system included seven different arrays with probes targeting 40 different miRNAs. Each of the seven arrays included a different miR-16 probe--one was completely homologous to miR-16, two had distinct single mismatches, two had double mismatches, and two had three mismatches. miRNA from human tissue was isolated, labeled, and hybridized to each array. Signal from the miR-16 element was normalized using the signal from the other forty elements on each array. The hybridization and wash conditions used with the miRNA arrays provided over 10-fold more signal for perfect matches than for miRNAs with <90% target homology (Figure 2). Identical experiments with miR-23 probes yielded similar results.
Figure 2. miRNA Array Specificity. Probe specifi city was determined using probes designed with up to three mismatches. 10-fold more signal for perfect matches was obtained as compared to the same miRNA with <90% target homology. This experiment shows that individual miRNAs can be differentiated on arrays allowing detection of different miRNA isoforms that have three or more different nucleotides.
The simplest measure of precision for microarray analysis is self vs self analysis. For each tissue we split total RNA samples into two aliquots, and independently isolated miRNA from them. Each of the two aliquots was differentially labeled and co-hybridized to an miRNA array. Signal was measured for each element of the array. Since the differentially labeled samples were derived from the same total RNA sample, the two signals should have been approximately equal for each spot. We performed this experiment for a number of human samples; the average correlation of the data sets was routinely 9799% (Figure 3A).
Figure 3. miRNA Array Reproducibility. (A) Raw intensities from each channel (Cy3 and Cy5) for two self vs. self arrays. The arrays were performed on two separate days. The average correlation of such samples routinely ranges between 9799% indicating precision with the array system. (B) The average Log2(R/G Normalized Ratio) of each spot on the array from six independent prostate vs colon samples is plotted as a circle on the graph. The average standard deviation of each spot is indicated by the error bars. The average correlation for the 6 replicates is 98% indicating reproducibility with the array system.
We also tested the reproducibility of our miRNA array system by comparing miRNA profiles obtained from human colon vs. human prostate in six independent array experiments. Total RNA was isolated from a single human prostate and colon sample. Each total RNA sample was split into six replicates. miRNAs in each sample were independently isolated and labeled. Each of the six Cy3-labeled prostate miRNA samples was combined with a Cy5-labeled colon sample and the six prostate/colon pairs were hybridized to miRNA arrays. The Log2(R/G Normalized Ratio) of each spot is shown in Figure 3B. The average correlation between the six independent experiments was 98%, indicating that the miRNA array process was highly reproducible.
The most important measurement for a microarray procedure is whether the quantitative array data truly reflect the abundance of miRNAs in total RNA samples. We set up an extensive experimental analysis using several tissue samples to address accuracy. miRNA was purified from FirstChoice Total RNA from human bladder, lung, and uterus.
The miRNAs were labeled and hybridized with arrays to compare the expression of more than 160 miRNAs in bladder, lung, and uterus.
The relative abundance of seven different miRNAs was measured by Northern analysis or a hybridization ass ay in the same bladder, lung, and uterus total RNA samples that were used for array analysis. The mature miRNAs on the Northern blots were quantified by phosphorimaging. The hybridization assays used chemiluminescent reactions and detection. The mirVana miRNA array system accurately reflected the expression level of each miRNA as compared to the hybridization assay and Northern analysis data obtained from total RNA (Figure 4).
Figure 4. miRNA Array Quantitation. The fold change in expression of seven different miRNAs within three different pairs of tissues was determined using three independent methods of quantitation--Northern analysis, solution hybridization/digestion assay (Hyb/Dig'n), and microarray analysis. The high correlation between the three methods validates the miRNA expression data obtained from the array system.
A final validation of the procedure as well as a confirmation of the biological significance of miRNAs in adult human tissues involved profiling miRNA expression in biological samples. We isolated miRNA from 26 normal human tissues. Half of each sample was used to create an miRNA pool containing miRNA from all 26 samples. The miRNAs from the pooled sample as well as each of the 26 single-tissue samples were fluorescently labeled with Cy3 or Cy5 using the mirVana miRNA Labeling Kit. Each single tissue miRNA sample was hybridized to the miRNA arrays along with a pooled sample to act as a common reference. The relative signal intensities of the single and pooled samples were compared for each miRNA. Each of the 26 normal human tissues that were analyzed had a unique pattern of miRNA expression (Figure 5). Hierarchical clustering showed that miRNA profiles were similar between related tissues and distinct between unrelated tissues, e.g. heart and skeletal muscle miRNA profiles were very similar; and expression in digestive tract tissues clustered, as did that of reproductive organ tissues. Interestingly, the brain miRNA profile was clearly distinct from the other tissues analyzed. The observation that adult tissues possessed unique miRNA profiles suggests that miRNAs function in tissue differentiation. Furthermore, similar miRNA profiles among related tissues provides additional evidence that miRNAs play a major role in defining adult tissues.
Figure 5. Biological Validation of the miRNA Array System. Heat map of miRNA expression from 26 normal human tissues. Each sample was hybridized on the array against a pool of all samples. Red indicates that the miRNA is over-expressed relative to sample pool; green indicates that it is under-expressed relative to the sample pool. Hierarchical clustering (right) shows that miRNA expression profiles are similar between related tissues and distinct between unrelated tissues.
The mirVana miRNA array procedure was sensitive, specific, and accurate enough to detect these similarities and differences between adult tissues, providing the opportunity to study the subtle and not so subtle differences between other, more interesting samples.
Jaclyn Shingara, Kerri Keiger, Jeffrey Shelton, Ila Wolf, Emmanu el Labourier, David Brown Ambion, Inc.