Studies overestimate the extent of circadian rhythm reprogramming in response to dietary and genetic changes

Abstract

The circadian clock modulates key physiological processes in many organisms. This widespread role of circadian rhythms is typically characterized at the molecular level by profiling the transcriptome at multiple time points. Subsequent analysis identifies transcripts with altered rhythms between control and perturbed conditions, i.e., are differentially rhythmic (DiffR). Commonly, Venn Diagram analysis (VDA) compares lists of rhythmic transcripts to catalog transcripts with rhythms in both conditions or have gained or lost rhythms. However, unavoidable errors in the rhythmicity detection propagate to the final DiffR classification resulting in overestimated DiffR. We show using artificial experiments constructed from biological data that VDA indeed produces excessive false DiffR hits both in the presence and absence of true DiffR transcripts. We present a hypothesis testing and a model selection approaches in an R-package compareRhythms that instead compare circadian amplitude and phase of transcripts between the two conditions. These methods identify transcripts with gain, loss, change or have the same rhythms; the third category is missed by VDA. We reanalyzed three studies on the interplay between metabolism and the clock in the mouse liver that used VDA. We found not only fewer DiffR transcripts than originally reported, but VDA overlooked many relevant DiffR transcripts. Our analyses confirmed some and contradicted other conclusions in the original studies and also generated novel hypotheses. Our insights also generalize easily to studies using other -omics technologies. We trust that avoiding Venn Diagrams and using our R-package will contribute to improved reproducibility in chronobiology.