Ageing is difficult to study because of the complexity and multi-factorial nature of traits that result from a combination of environmental, genetic, epigenetic and stochastic factors, each contributing to the overall phenotype. In light of this challenge, transcriptomic studies of aging organisms are of particular interest, since transcription is an intermediate step that links genotype and phenotype.
In recent years microarrays have been widely used for elucidation of changes that occur with age in the transcriptome in Caenorhabditis elegans. However, different microarray studies of C. elegans report sets of differentially expressed genes of varying consistence, with different functional annotations. Failures to find a consistent set of transcriptomic alterations may reflect the absence of a specific genetic program that would guide age-related changes but may also, to some extent, be a consequence of a small sample sizes and a lack of study power in transcriptomic researches. To tackle this issue we analyzed RNA sequences of samples from a time-series experiment of normal aging of C. elegans, performing the first, to our knowledge, NGS-based study of such kind. As a result, evidences were collected that promote a union of two competing theories: the theory of DNA damage accumulation and the theory of programmed aging.
Next, we applied two alternative methods, namely the Short Time-series Expression Mining and the Network Smoothing algorithm, in order to obtain and analyze sets of genes that represent distinct modules of age-related changes in the transcriptome. Besides characterization of age-related changes, we were also interested in assessment and validation of the Network Smoothing algorithm. Generally, results of clustering of smoothed scores are consistent with results of short time-series clustering, allowing robust elucidation of functions that are perturbed during aging.
At the last phase of the project we questioned if observed changes in the transcriptome can be controlled by specific transcription factors. Thus we used Chip-seq data to predict plausible transcription factor regulators of gene sets obtained using time series clustering and Network smoothing. On the one hand, all predicted transcription factors had documented relevance to aging. On the other hand, we did not achieve gene set specific prediction of transcription factors. In fact, genes with the opposite dynamics were predicted to respond to the same transcription factors.
To summarize, we characterized in details age-related changes in the transcriptome of C. elegans, validated the performance of the Network Smoothing algorithm and showed that integration of gene expression with Chip-seq data allows to predict transcription factors that are capable to modulate the lifespan of C. elegans.