DNA Methylation and Transcriptional Regulation in a Model Algae, Chlamydomonas reinhardtii

High-throughput gene expression data provides opportunities for studying gene functions on a global scale. It has been well established that genes within the same pathway or in the same functional complex often have similar expression patterns under diverse conditions. Gene co-expression networks, computed based on the expression similarity of gene pairs, are effective methods for the detection of biologically significant modules.  Many gene co-expression network methods have been developed and widely used to find modular patters in large datasets. Most gene co-expression networks are constructed based on either Pearson correlation coefficients, as in the network inference method WGCNA. However, typically gene co-expression networks are constructed based on a single dataset, which makes them sensitive to noise in expression estimates. Here we propose a method to integrate datasets that greatly improve the accuracy of the resulting network. We use 5 RNA-seq datasets (that correspond to experimental conditions involving acetate, anaerobiosis, cadmium, copper, a starch mutant, iron and bilin) from the alga Chlamydomonas reinhardtii to construct a co-expression network of genes. For each dataset, a Pearson correlation coefficient matrix was calculated and transformed into an adjacency matrix using the WGCNA R package. In order to identify highly conserved genes pairs, we combined the 5 datasets together and calculated the logical “AND” of the 5 adjacency matrixes to produce a single combined matrix. We use this matrix as the final combined adjacency matrix to construct the gene co-expression network and identify modules, thus effectively reducing the noise in any single dataset. We derived a network with 1424 nodes and 5305 edges which was visualized using Cytoscape. We attempted to validate the edges in the network based on whether the two connected genes were annotated within the same KEGG pathway and found a true positive rate of 86.9%. We found that 182 unconnected modules formed spontaneously. We first tested KEGG pathway and Gene Ontology enrichment in each module based on the Algal Functional Annotation Tool.  Many modules showed striking levels of functional enrichment.  For example, the genes in module 5 are all cytosolic ribosomal proteins, the genes in module 7 are all proteasome related, etc. To search for cis-regulatory elements within modules, we extracted about 200 bases upstream of genes in each module and used MEME  to search for motifs.  We found significant motifs in most of the modules. We also reasoned that unannotated genes within modules could potentially be functionally characterized based on the dominant module functions. We also extracted the transit peptide sequence of genes in each module, and uploaded them to the PredAlgo Server, to identify the localization in the cell of these genes.   We found that most modules were strongly enriched for one cellular localization. Lastly, to assess whether the expression of these modules was localized in time we compared them to time-series expression data from the day/night cycle in Chlamydomonas, and found that most of them were also synchronized in time and were coherently expressed during specific hours of the cycle.