Microarray construction. Microarrays were produced by Eurogentec S.A. (Belgium). A first series of PCR reactions was performed using gene-specific primers designed to amplify amplicons of 200 to 800 bp length from all ORFs of B. longum NCC2705 genome. Each forward primer carried at its 5’-terminus one of nine different 5’-sequence tags. Moreover, all reverse primers contained, in addition to gene-specific sequences, a unique common tag. These tags were used to perform a second round of PCR amplifications. These secondary PCR reactions were controlled by agarose gel electrophoresis and repeated if necessary. Finally, PCR products were purified using Millipore Multiscreen 96-wells filter plates, dried by evaporation and suspended in a suitable volume of 3 X SSC at a final concentration of 100-200 ng/µl prior to printing in duplicates onto CMT-GAPS II Corning glass slides.
Microarray hybridization. For each hybridization 2 µg of total RNA were labeled together with 10 ng of spiked luciferase control mRNA (Promega; used as a control for data normalization) using the 3DNA Array 350RP Genisphere kit (Genisphere Inc., Hatfield, U.S.A), following the protocol provided by the supplier. Following hybridization, the array slides were scanned using a Scanarray 4000 machine (Packard Biochip Technologies).
Microarray data analysis. Data were extracted from the scanned images using the software Imagene 5.6 (Biodiscovery). The local average background was subtracted from the average signal values of each spot. Spots displaying low signal intensities in both Cy3 and Cy5 channels (i.e. less than three fold the average local background) were excluded; however spots displaying low signal intensities in only one channel were retained, if the signal in the other channel was at least fivefold higher than the corresponding average local background signal. In order to remove from the expression measures any systematic trends that arise from the microarray technology, a scaling within-array normalisation was performed on the log2-transformed ratios (M-values). The normalization consisted on bringing the log-ratios median to zero within each array. Within array M-values from spots duplicate were averaged. The outcomes of this pre-processing step are M-values from four hybridization experiments for each comparison (lactose versus glucose and lactose+glucose versus glucose), obtained from two technical replicates (hybridizations) with RNAs from two biological replicates (fermentations). For more confidence in the results, only genes with M-value from at least two independent hybridizations were taken into account in the analysis. For the statistical estimation of the differential expression we used an empirical Bayes method (2, 3) to shrink the gene-wise sample variances towards a common values and, in this way, we avoided the problem of classical t-statistic driven by outliers in case of limited number of replicates (1). This method is implemented in the software package LIMMA for the R computing environment (www.r-project.org). LIMMA is part of the bioconductor project (http://www.bioconductor.org). In order to limit the expected number of false positive results to 1, we fixed the significance level a to 1/no, where no is the number of genes with an expression value in at least two out of four hybridization repeats. Finally, genes were considered as differentially expressed if their absolute log2-transformed signal ratios were greater than 1 and their P-values smaller than a (~0.002).
References
1. Hatfield, G. W., S. P. Hung, and P. Baldi. 2003. Differential analysis of DNA microarray gene expression data. Mol.Microbiol. 47:871-877.
2. Lönnstedt, I. and T. P. Speed. 2002. Linear models and empirical Bayes methods for assessing differential expression in microarray experiments. Statistica Sinica 12:31-46.
3. Smyth, G. K. 2004. Linear models and empirical Bayes methods for assessing differential expression in microarray experiments. Statistical Applications in Genetics and Molecular Biology 3.