A database of tissue-specific rhythmically expressed human genes has potential applications in circadian medicine

Abstract

The discovery that half of the mammalian protein-coding genome is regulated by the circadian clock has clear implications for medicine. Recent studies demonstrated that the circadian clock influences therapeutic outcomes in human heart disease and cancer. However, biological time is rarely given clinical consideration. A key barrier is the absence of information on tissue-specific molecular rhythms in the human body. We have applied the cyclic ordering by periodic structure (CYCLOPS) algorithm, designed to reconstruct sample temporal order in the absence of time-of-day information, to the gene expression collection of 13 tissues from 632 human donors. We identified rhythms in gene expression across the body; nearly half of protein-coding genes were shown to be cycling in at least 1 of the 13 tissues analyzed. One thousand of these cycling genes encode proteins that either transport or metabolize drugs or are themselves drug targets. These results provide a useful resource for studying the role of circadian rhythms in medicine and support the idea that biological time might play a role in determining drug response.

Fig. 1.. Population-level rhythms in gene expression across human anatomy.

(A) Number of genes that met criteria for rhythmicity (FDR < 0.05; rAmp ≥ 0.1; R² ≥ 0.1) in at least 1 of the 13 tissues. Periodic analysis was performed by cosinor regression on expression values for 160 CYCLOPS-ordered samples per tissue, as discussed in fig. S1 and Materials and Methods. See fig. S2 for numbers of rhythmic genes discovered at varying FDR thresholds. (B) Set of 54 ubiquitous genes cycling in at least 8 of 13 human tissues, grouped by prior circadian context as reported in CircaDB (darkest blue, homolog cycles in 16 of 16 mouse tissues; lightest blue, not cycling in any of the 16 mouse tissues). (C) Average acrophases of core clock genes (external circle) across all human tissues compared to mouse [internal circle, average across 12 tissues reported by Zhang et al. (3)]. Average coefficient of determination (R², a measure of cycling robustness) is indicated by point size, and phase variability (Phase var) is indicated by color. Peak phase of ARNTL (human) or Arntl (mouse) was set as 0 for comparison. (D) Average acrophases of all other (noncore clock) human ubiquitous cycling genes. Genes located more distant from the center have larger average amplitudes of oscillation (rAmp) across tissues where they cycle.

Fig. 2.. Robust tissue-specific rhythms.

(A) Distribution of robustness (R², goodness of data fit to 24-hour sine wave) and (B) peak-to-trough ratio for all cyclers (gray) in each tissue; most clock genes (red) reside in the upper quartile in most tissues. (C) Distribution of acrophases for cyclers differs between tissue types. Light and dark shading represents inferred active and inactive phase based on Arntl expression, which is known to peak in anticipation of the inactive period in mammals. (D) UpSetR (18) to rank and visualize the intersection between multiple sets (tissues). Restricting input to the top 500 cyclers by FDR from each tissue, the largest intersections describe genes whose expression cycles only in a single tissue.

Fig. 3.. Pharmacological links to molecular rhythms in the human population.

(A) Of 7486 genes found to cycle in at least 1 of 13 human tissues sampled, 917 (12%) encode at least one drug target, transporter, or metabolizing enzyme (collectively referred to as “targets”). This represents a total of 2764 drug entities, both approved and experimental as logged in DrugBank (20), with targets that oscillate somewhere in the body. (B) Numbers of total (colored green) and pharmacologically active (colored blue) cycling drug targets by tissue type. (C) Cardiovascular-related tissues from ordered GTEx data. To enrich for high-amplitude cardio-cyclers, we limited to genes in the upper 50th percentile for rAmp within each tissue and FDR ≤0.1. Numbers of cardio-cyclers that meet these cycling criteria and are also drug targets are indicated in parentheses. (D) Cardio-cyclers targeted by select drug classes relevant to heart and vessel physiology. Plotted phase represents average peak phase across all cardiac tissues where that gene was found to cycle. (E) Expression values plotted as a function of sample phase for select cardio-cyclers. Coefficient of determination (R² ) and peak-to-trough ratio (ptr) are indicated. Right: Genes that are also rhythmically expressed in whole mouse heart, with peak phases indicated by orange bars. ARNTL (human) or Arntl (mouse) was set to 0 for comparison.