1 clock gene. Future transcriptomic studies coupled to phylogenetic footprint and functional analysis should give insight into the transcriptional networks involved not only in diurnal regulation of gene expression but also in response to specific stresses of the marine environment such as phosphate, nitrogen or iron limitation, UV stress or in response to viral infection. Methods Slides construction and hybridization Genome-wide based Ostreococcus slides were manufactured in the Genopole Ouest Transcriptome Platform. Gene-specific 50-mer oligonucleotides were designed and synthesized by Eurogentec on the basis of January 2006 annotation. In the final annotation of the genome, 6369 genes were represented by at least one probe but 565 oligonucleotides did not match the genome anymore in BlastN. However 372 out these 565 probes gave a good and reproducible hybridization signal, were selected after ANOVA as differentially expressed genes. Therefore each probe was attributed a feature number with corresponding numbers in the two annotations. Cell AVE-8062 culture conditions, RNA extraction, labelling, hybridization and raw analysis have been previously described. Microarray data analysis Normalization was performed using the print-tip loess method and scaled with the Gquantile method. Time courses of gene expression were performed in triplicate, over 27 h, at 3 h intervals. Fifteen probes, where on more than 70% time points no data are available, were removed from the analysis. We first 
verified hybridization robustness by performing a hierarchical clustering on the 8041 selected probes using TiGRMeV4.0 suite. Technical triplicates were clustered. Therefore, for further analysis, we chose to work on the median value of each technical triplicate. Analysis of Variance and PubMed ID:http://www.ncbi.nlm.nih.gov/pubmed/19794329 Principal Component Analysis were performed using the GeneANOVA software and the limma from bioconductor. 6822 genes differentially expressed with a P value < 10-3 were selected using a 3 factors ANOVA. Correlations were found between gene expression and time points with PCA and we retained 2038 genes with best dispersion corresponding to maximized variance. Twelve gene expression clusters were highlighted with SOM 2D provided in TiGRMeV4.0 suite and analysed using FATIGO based on Arabidopsis functional annotation. Qualitative information was obtained about biological processes associated with specific times of the day. However, only a small number of homologues of Arabidopsis annotated genes were found. For this reason, these clusters were not further analyzed. Bayesian Fourier Clustering was used to cluster time series according to their expression profiles using the framework of a standard linear model. Curves were clustered together by BFC if they appeared to have been drawn from a joint distribution with parameters b and s2, where Y = Bb + and Y represents the logarithm of the expression levels. is a noise term, which is normally distributed with mean zero and variance s2. Thus the skewed time course of expressions of genes in each cluster is characterized by a different vector of Fourier coefficients b and associated variance s2. This technique is therefore a powerful way of uncovering a wide variety of shapes and respects the time ordering of expressions. This method was exceptionally fast because of the choice of distributions on the parameters, the settings of the hyperparameters and the hierarchical search among partition spaces. Each gene expression profile was initially a
