Defining and detecting global transcriptional amplitude in circadian gene expression

Many genes exhibit circadian rhythms in expression. The amplitude of oscillation, both in core clock and circadian output genes, may differ from person to person. Mutations in core clock genes are known to alter global rhythmic properties, and researchers often informally discuss "circadian amplitude." Yet, it remains unclear whether, in the general population, differences in transcriptional amplitude are largely gene-specific, or if they reflect a global, transcriptome-wide pattern -- whether some individuals have globally higher or lower amplitude across the set of all rhythmic genes. We used Cosinor regression to reanalyze four human skin time-series transcriptomic datasets (paired epidermis/dermis samples, N=11, N=19) and found that, using either absolute or relative amplitude measures, distributions of gene amplitudes tended to cluster by subject. Using a non-parametric, permutation-based statistical test, we found that in many subjects this global amplitude trend was statistically significant (p ≤ 0.01). Furthermore, we found that when rhythmic genes were divided into two sets based on peak time (genes peaking before-noon and after-noon), the subjects' global amplitude in one set predicted global amplitude in the other set (p ≤ 0.05). We also found that in the paired epidermis/dermis datasets, subjects' global amplitude in epidermis predicted their global amplitude in the dermis (p ≤ 0.05). After identifying these trends in the skin datasets, we then found that evidence for subject-specific transcriptional rhythm strength replicated across 6 additional human time-course datasets from adipose, muscle, and blood. Perhaps surprisingly, across datasets, we found that neither established metrics of core clock transcriptional organization nor the amplitude of core clock transcription were strongly correlated with subject-specific global amplitude.