We included transcriptome data from studies in mouse models of physiological LVH induced by swim ming exercise, cardiac specific activation selleck KPT-330 of Akt, and cardiac specific activation of PI3K. This is the first study in cardiac hypertrophy at this scale and it may provide a basis for further understanding of both physiological and pathological LVH phenotypes. Results Generation of Microarray co expression Networks Gene expression profiles in heart tissue were investi gated under normal conditions, during physiological stress, and in two gene modified models of physiological LVH involving cardiac activation of the PI3K Akt pathway. To estimate the specificity of the hypertrophic gene signature, an additional dataset moni toring gene expression in healthy mouse organs was also used.

Four mouse microarray datasets totaling 141 arrays were obtained from ArrayExpress for further analysis. The Akt dataset was generated using a tetra cycline regulated transgenic system with the capacity to conditionally switch a constitutively active form of the Akt1 protein kinase on or off in the adult heart. This dataset consisted of normal heart tissue, short term, activated Akt1, and switched off Akt1. The PI3K dataset consisted of wild type hearts and hearts with expression of dominant negative PI3K or constitutively active PI3K. The Swimming dataset, containing 30 arrays, monitored expression in mouse hearts under normal conditions, swimming, and swimming fol lowed by 1 week of rest. Finally, the Normal dataset monitored transcript expression in healthy mouse tissues including bladder, bone, spleen, stomach, and the heart.

After pre processing, pair wise gene expression similarities were measured using the Pearson Correlation Coefficient. Co expression networks were undirected and, at PCC 0. 70, obeyed a power law, suggesting a scale free architecture dominated by a number of highly connected hub genes. The PCC threshold was set to 0. 70 on the basis of the following evidence, gene correlation profiles with PCC over 0. 60 were demonstrated to be more biologi cally relevant and similar studies of human gene co expression landscape have employed comparable threshold criteria. Additionally, below this cut off all networks were excessively large, suggesting a presence of false positive edges.

However, a more stringent PCC threshold was avoided, as further filtering has been implemented by selecting gene pairs that were correlated across all Dacomitinib three datasets. Finally, the data driven cut off approach was not deemed appropriate as it is intended primarily for comparison of multiple networks derived from differential phenotypes. At PCC 0. 70 it was noted that an increase of this cut off value removed weakly connected links from all networks while maintaining a constant number of genes.