Mutual information networks unveil global properties of IFN▀ immediate transcriptional effects in humans
Guy Haskin Fernald
Last modified: March 31, 2006
Interferon beta (IFN▀) is an immunomodulatory drug commonly prescribed to control clinical exacerbations in multiple sclerosis (MS) patients. Although it effectively reduces the relapse rate and magnetic resonance imaging (MRI) activity in some patients, side effects are not uncommon and its efficacy is minimal in a significant proportion of treated individuals. Here we used high frequency sampling and DNA microarrays to analyze longitudinally the transcriptional profile of blood cells from two individuals only hours after IFN▀ administration and up to a week later. Blood samples from 4 healthy volunteers taken at similar times were also analyzed for comparison. We used traditional bioinformatics to identify genes regulated by IFN▀ in the patients, and normal fluctuation of gene expression in the healthy controls. We found that MX1, OAS1 and IFIT1 were upregulated within 3.5h of IFN▀ administration and their expression was sustained for at least the following 4 days. Interestingly, more genes were downregulated than upregulated in response to the drug. Among downregulated transcripts, we identify many involved in protein synthesis and folding, antigen presentation, and inhibition of apoptosis. This suggests a general reduction in cellular activity, resembling the endogenous antiviral response of interferons. In contrast, most differentially expressed genes in healthy individuals were involved in cellular physiological processes. We next used mutual information (MI) to build networks of co-regulated genes and analyzed their properties in both groups of samples. While both networks displayed scale-free properties, the connectivity distribution of nets generated with suboptimal MI values failed to fit a power law, and instead approximated a Poisson distribution. This strongly suggests that mutual information captures biologically relevant interactions. We also found important differences in network topologies and in the ontologies of their component genes. The networks generated in response to IFN▀ reveal a tight core of immune and apoptosis-related genes with higher values of MI. On the other hand, networks obtained from normal individuals mostly reflect cellular housekeeping functions. This is the first study that incorporates network analysis to investigate the gene regulation in response to a therapeutic drug in humans. This method could be used to create personalized models of response to therapy.