Gene expression is the process by which genetic information is translated into functional macromolecules. Transcription is the first step in this process, which involves the synthesis of messenger RNAs (mRNAs) from the corresponding gene DNA template. Over the past few decades, genome-wide transcriptional profiling approaches have made it possible to assess the expression levels of thousands of genes in parallel in various biological contexts. In statistical analyses, the expression of a gene is estimated by counting the number of occurrences of the corresponding mRNA sequences over a set of samples and is defined by two dimensions: the mean and the variance.
The overwhelming majority of transcriptomic analyses based on the study of expression profiles focus on the identification of genes whose mean expression changes significantly when comparing samples from different conditions, corresponding to biological processes of interest, such as developmental biology, disease etiology, therapeutic target discovery, etc. In this classical approach, variance is usually considered only as a noise parameter to be estimated before comparing expression means.. Yet, the estimation of expression variance can be relevant from a physiological point of view, since a change in this parameter can reflect a biological change in the regulation of gene expression. Fluctuations in gene expression may indeed be due to various sources intrinsic to cellular life, such as the stochastic nature of gene transcription, the phase of cell cycle, chromatin changes or mRNA degradation.
Researchers at IRIG propose to compare the performance of statistical methods that identify such differentially variant genes. They demonstrated the potential of this approach by analyzing expression datasets in various cancers, which identified key cellular functions in tumor progression that could not have been identified by comparing average expression.
The researchers evaluated four recent methods that detect differences in the mean and dispersion of RNA-seq data. By applying these methods on simulated data, they have characterized reliable parameters to detect genes with expression variance between two conditions. The Cancer Genome Atlas data were then subjected to these methods. Among the genes showing increased expression variance in tumors and no change in average expression, some key cellular functions were identified, the majority of which are related to catabolism and are overrepresented in most of the cancers analyzed.
It may be biologically and physiologically relevant to consider differential variance in gene expression. These results highlight autophagy in carcinogenesis, illustrating the potential of the differential variance approach to gain new insights into biological processes and to discover new biomarkers.