Total RNA was isolated from six 10 µm tissue sections using the RecoverAll Total Nucleic Acid Isolation Kit (Applied Biosystems/Ambion, USA) according to manufacturer guidelines. Total RNA quantity and quality were checked by spectrophotometer (Nanodrop ND-1000).
Label
Hy3
Label protocol
From each sample 100 ng of total RNA was labeled with Hy3 fluorescent dye using the miRCURY LNA Array power labeling kit (Exiqon, Denmark). All samples were labeled the same day with the same master mix, in order to minimize technical variation.
Hybridization protocol
The Hy3-labelled samples were hybridized to miRCURY LNA arrays (v11.0) (Exiqon, Denmark), containing capture probes targeting all human miRNAs registered in the miRBASE version 15.0 at the Sanger Institute. The hybridization was performed overnight at 56°C according to manufacturer specifications using a Tecan HS4800 hybridization station (Tecan, Austria). Since it was not possible to hybridize all arrays in one go, samples were randomly split into 5 batches as to minimize day to day variation in the hybridization process.
Scan protocol
After hybridization the microarray slides were scanned using an Agilent G2565BA Microarray Scanner System (Agilent Technologies, Inc., USA) at 5µm resolution, and the resulting TIFF images were analyzed using the ImaGene 8.0 software on standard settings (BioDiscovery, Inc., USA).
Data processing
Probe signals were background corrected by fitting a convolution of normal and exponential distributions to the foreground intensities using the background intensities as a covariate 4. Four technical replicate spots for each probe were combined to produce one signal by taking the logarithmic base-2 mean of reliable spots. If all four replicates for a given probe were judged unreliable that probe was removed from further analysis. A reference data vector R was calculated as the median signal of each probe across all samples. For all probe signals in a given sample, represented by the sample data vector S, a curve F was determined by locally weighted polynomial regression so as to provide the best fit between S and R. A normalized sample vector M was calculated from this by transforming it with the function F, so that M = F(S). In this manner, all samples were normalized to the reference R. This normalization procedure largely follows that outlined in Rosenfeld et al., 2008. Remote data points (probes in sparsely sampled intensity regions with less than 15 probes per signal unit) were considered unreliably adjusted by this method and removed before further analysis.
