In this Driving Project together with Gao Zhang and Meenhard Herlyn group from The Wistar Institute/UPenn we have explored gene expression data of human melanoma cells to investigate the cell transition from the autophagy to senescence. We have explored user’s gene expression experiment together with publically available gene expression data.
Our assumption is that the time course of melanoma growth can be divided into two stages: 1) autophagy that takes place up to 12 hrs of experiment and senescence, from 12 to 120 hrs. Our goal is to find critical players in the transition from autophagy to senescence; build an integrative gene regulatory network and change of this network with time; find out which known pathways are critical in autophagy-senescence transition. Stage differential expression is likely to provide clues to the autophagy-senescence transition that might point toward whole genome changes to (or consequences of) the trait with advancing age. To evaluate alterations in contributing biochemical pathways we statistically quantified these global pathway disturbances using the Kolmogorov-Smirnov goodness-of-fit test and found widespread, global, alterations in patterns of gene expression in diverse systems.
As a first step to determine critical players in the cell transition from the autophagy to senescence stage on gene expression levels, we performed a “between subject” T-test, based on the Welch t-test for small samples with unequal variance in the two groups (Welch, 1947). The T-test finds genes that are mostly differentially expressed between two stages: autophagy and senescence. Analysis of the scaled (globally normalized) microarray data was performed using T-test tool of BiologicalNetworks, where samples were assigned to two groups: Autophagy from 0h to12h and Senescence from 12hr to 120 hr. The T-test F-statistics P-values indicate statistical significance between stages (i.e, autophagy versus senescence) and produces two sets of genes that are: significantly and non-significantly different between the two stages. We performed sorting the probe sets into transition clusters based on T-test F-statistics. Probe sets not significant (at P-value<0.01) were excluded. Of 25160 genes in our experiment, 571 showed significant stage difference (P<0.01), that are thought to contribute mostly to the autophagy transition into a senescence stage and elevation of senescence cells with age. As expected in studying a complex multifactorial trait such as autophagy and senescence, most of the genes with stage significance show subtle fold changes between 0.5-fold and 2.0-fold; such modest fold changes reinforce the importance of a systems biology approach to analysing the data.
On the next step we extracted transcription factors and transcriptional regulators from the list of 571 genes products differentially expressed in autophagy and senescence stages. We found 11 transcription factors that can be critical players in pushing autophagy cells towards senescence. Next given set of genes and/or transcription factors we first searched for known promoter frameworks and known binding regions available in the IntegromeDB. Then, with a set of candidate genes, known and putative transcription factors and signal transduction molecules, the procedure searches the integrated database (e.g. GEO, ArrayExpress) for experiments showing strong (0.75 Pearson correlation threshold is used) co-expression patterns for input genes. Finally, the probabilistic method for discovering regulatory modules from gene expression data described in (Segal et. al., 2003) produces a partition of genes into modules and a regulation program for each module that explains the expression behaviour of genes in the module. The regulation program of a module specifies the set of regulators that control the module and the mRNA expression profile of the genes in the module as a function of the expression of the module’s regulators. For every experiment the procedure produces a list of modules and associated regulation programs: groups of co-regulated genes, their regulators and the conditions under which regulation takes place. The final network is integrated with known (from literature and public databases) protein-protein interactions and is functionally assessed for relevance of detected modules. The final pValue score for every module is produced that takes into account all three types of evidences (cis-regulatory motifs enrichment, Fisher’s pValue of co-expressed TF-gene pairs and network clustering pValue).
Among the top GO-clustered physiological processes among differentially expressed genes are: DNA replication, DNA damage, DNA damage response and DNA repair, Apoptosis and anti-apoptosis, autophagy, signal transduction resulting in apoptosis, MAPK and G protein-coupled receptor (GPCR) signalling. Stage transition differences included transcripts in several categories. Thus, the response (or adaptation) to genetic autophagy seems to involve activation (or deactivation) of a wide spectrum of gene expression programs. GO clusters give us a wider perspective on disruptions in biological processes within the authopagy cell that may lead to its senescence state (Figure 1). Among the important observations that arise from the stage differences in GO clusters: (1) the biological processes altered in the autophagy are large in number and diverse in process, perhaps not surprising in such a polygenic process. Nonetheless, the spectrum of clusters encompassing differentially expressed transcripts is quite broad.