Effects of insufficient sleep on circadian rhythmicity and expression amplitude of the human blood transcriptome

Abstract

Insufficient sleep and circadian rhythm disruption are associated with negative health outcomes, including obesity, cardiovascular disease, and cognitive impairment, but the mechanisms involved remain largely unexplored. Twenty-six participants were exposed to 1 wk of insufficient sleep (sleep-restriction condition 5.70 h, SEM = 0.03 sleep per 24 h) and 1 wk of sufficient sleep (control condition 8.50 h sleep, SEM = 0.11). Immediately following each condition, 10 whole-blood RNA samples were collected from each participant, while controlling for the effects of light, activity, and food, during a period of total sleep deprivation. Transcriptome analysis revealed that 711 genes were up- or down-regulated by insufficient sleep. Insufficient sleep also reduced the number of genes with a circadian expression profile from 1,855 to 1,481, reduced the circadian amplitude of these genes, and led to an increase in the number of genes that responded to subsequent total sleep deprivation from 122 to 856. Genes affected by insufficient sleep were associated with circadian rhythms (PER1, PER2, PER3, CRY2, CLOCK, NR1D1, NR1D2, RORA, DEC1, CSNK1E), sleep homeostasis (IL6, STAT3, KCNV2, CAMK2D), oxidative stress (PRDX2, PRDX5), and metabolism (SLC2A3, SLC2A5, GHRL, ABCA1). Biological processes affected included chromatin modification, gene-expression regulation, macromolecular metabolism, and inflammatory, immune and stress responses. Thus, insufficient sleep affects the human blood transcriptome, disrupts its circadian regulation, and intensifies the effects of acute total sleep deprivation. The identified biological processes may be involved with the negative effects of sleep loss on health, and highlight the interrelatedness of sleep homeostasis, circadian rhythmicity, and metabolism.

Fig. 1.

Study protocol. The study protocol consisted of two 12-d laboratory sessions in a cross-over design. After two baseline/habituation nights, participants were scheduled to seven consecutive sleep opportunities of 6 h in the sleep-restriction condition and seven consecutive sleep opportunities of 10 h in the control condition. Following the final sleep restriction or control sleep opportunity, participants were subjected to a period of extended wakefulness (39–41 h of total sleep deprivation), which included hourly melatonin assessments, a well-established marker of circadian phase, and three hourly RNA samplings, under constant-routine conditions. Following a 12-h recovery sleep opportunity participants were discharged from the study.

Fig. 2.

Effects of chronic sleep restriction on the transcriptome. (A) Frequency distribution of expression fold-changes after sleep restriction relative to control. Histogram of changes in all transcripts (filled area; n = 31,685 probes that target 22,862 genes) and in transcripts identified as having a statistically significant (multiplicity corrected P value < 0.05) main effect of Sleep Condition (open area; n = 744 transcript that target 711 genes), plotted separately. (B) Example expression plots for genes with a significant main effect of Sleep Condition. MNFG (A_24_P224926) (P < 1 × 10⁻⁶), DCAF5 (A_24_P940396) (P < 1 × 10⁻⁶), RORA (CPID_186) (P < 1 × 10⁻⁶), and PRDX5 (A_24_P155378) (P < 1 × 10⁻⁶). Log₂ expression values are least-squares means ± SE (Procedure Mixed, SAS). Greyed area plots represent the melatonin profile averaged for the two conditions. Note that individual data were aligned relative to the individual melatonin rhythm and sorted into discrete circadian phase bins. Because of the shift in circadian phase after sleep restriction and to individual variation, the 10 melatonin samples covered 11 circadian phase bins after sleep restriction. (C) The top 10 enriched Gene Ontology Biological Processes and Molecular Functions within the statistically significant differentially expressed gene list as identified by WebGestalt when using the human genome as a background (66). Percentages are based on the number of unique gene symbols annotated as belonging to a specific biological process/molecular function compared with the number of unique gene symbols within the entire gene list. Color bars indicate the enrichment of a process/function, where red is the most enriched (top process/function) and yellow the least enriched (number 10 of the top 10). P values are the Benjamini and Hochberg (18) -corrected P values as calculated by WebGestalt (62).

Fig. 3.

Intersection of genes identified as circadian and time-awake–dependent in control and sleep-restriction (SR) conditions. (A) Venn diagram of prevalent circadian genes. (B) Venn diagram of genes identified as having a prevalent time-awake upward trend. (C) Venn diagram of genes identified as having a prevalent time-awake downward trend.

Fig. 4.

Effect of sleep restriction on the phase of circadian genes. (A) Genes with a prevalent circadian variation during the constant routine/total sleep deprivation after the control condition (2,103 probes that target 1,855 genes, FDR <5%). Heatmap rows correspond to the median of the melatonin-aligned probe values across all participants in the control condition. Rows are clustered based on a circular self-organizing map. Cluster means are plotted above as time-series and the number of genes per cluster (C1–C5) is indicated in parenthesis (genes belonging to multiple clusters are counted in each cluster independently). Color codes to the left of the heatmap correspond to the colors of the clusters. Sampling times and melatonin profile shown correspond to the average values across all participants in the control condition. Genes related to circadian rhythmicity and sleep (according to Gene Ontology) are indicated in the heatmap (colors indicate cluster location). (B) Phase histogram of melatonin-aligned peak times of prevalent circadian genes following control (black contour; 2,103 probes circadian in an average of 11.56 participants, n = 24,311) and sleep restriction (red contour; 1,644 probes circadian in an average of 10.50 participants, n = 17,276). The distribution of the phases is significantly different between conditions (on 1-h binned data, χ² = 1305.785, df = 23, P < 2.2 × 10⁻¹⁶). Histogram bins are 1-h-wide and bin heights are normalized to the maximum bin height per variable. The relative clock times and melatonin profile shown correspond to the average values across all participants and sleep conditions. (C) The top 10 enriched Gene Ontology Biological Processes and Molecular Functions within the circadian gene list of the control condition as identified by WebGestalt when using the human genome as a background (62). Percentages are based on the number of unique gene symbols annotated as belonging to a specific biological process/molecular function compared with the number of unique gene symbols within the entire gene list. Color bars indicate the enrichment of a process/function, where red is the most enriched (top process/function) and yellow the least enriched (number 10 of the top 10). P values are the Benjamini and Hochberg (18) -corrected P values as calculated by WebGestalt (62).

Fig. 5.

Circadian variations in the transcriptome following control and sleep restriction. Genes with a prevalent circadian variation during the constant routine/total sleep deprivation after control and/or sleep restriction (n = 2,859 probes that target 2,510 genes). (A) Heatmap rows correspond to the median of the melatonin-aligned probe values across all participants per sleep condition. Rows are clustered based on a circular SOM. Color codes on the left side of the heat map identify the clusters. Relative clock times and melatonin profiles are average values across all participants per sleep condition. Genes related to circadian rhythmicity and sleep (according to Gene Ontology) are indicated in the heatmap (gene colors indicate cluster location). (B) Examples of genes with a significant difference in circadian amplitude: GHRL (A_23_P40956) (pair-wise comparison across participants: P = 0.0040), IDS (A_24_P285032) (P = 0.0042), AVIL (A_23_P390157) (P = 0.0109), and CEACAM3 (A_23_P358244) (P = 0.0004). Log₂ expression values are least-squares means ± SE (Procedure Mixed, SAS). (C) Comparison of width at mid-amplitude for the night hours (trough) in the melatonin-aligned median profiles of day-active probes [n = 1,356 paired values, estimated mean of the difference −0.199 h (95% CI -0.288, −0.109), P < 1.416 × 10⁻⁵; density of the paired differences and 95% CI are shown in orange] and comparison of width at mid-amplitude for the night hours (crest) in the night-active probes [n = 1,469 paired values, estimated mean of the difference −0.572 h (95% CI −0.656, −0.489), P < 2.2 × 10⁻¹⁶; density of the paired differences and 95% CI are shown in blue]. (D) Density plot of circadian amplitudes per participant for the prevalent circadian genes in control and sleep restriction [control n = 29,568 (black solid line) corresponding to 2,859 probes circadian in an average of 10.34 participants; sleep restriction n = 24,354 (red solid line) corresponding to 2,859 probes circadian in an average of 8.51 participants]. The estimated mean for the circadian amplitude is 0.313 in control (black broken line) and 0.273 in sleep restriction (red broken line), 95% CI of the difference (−0.042, −0.037), P < 2.2 × 10⁻¹⁶.

Fig. 6.

Time-awake–dependent variations in the transcriptome. (A) Genes with a prevalent time-awake–dependent variation during the constant routine/total sleep deprivation following the control condition (124 probes that target 122 genes, FDR <5%). Medians of the melatonin-aligned probe values across all participants in the control condition are clustered based on a circular SOM. Cluster means are plotted as time-series and the number of genes per cluster is indicated in parenthesis (genes belonging to multiple clusters are counted in each cluster independently). Sampling times and melatonin profile shown correspond to the average values across all participants in the control condition. (B–D) Data are based on genes with a prevalent time-awake variation during the constant routine/total sleep deprivation after sleep restriction and not significantly prevalent after control (363 probes that target 361 genes with cumulative upward trend, and 470 probes that target 444 genes with cumulative downward trend; see Fig. 3 B and C). (B) Heatmap rows correspond to the median of the melatonin-aligned probe values across all participants per sleep condition. Rows are clustered based on a circular SOM of the sleep-restriction profiles. Color codes to the left of the heat map identify the clusters. Relative clock time and melatonin profile are the average values across all participants per condition. Genes related to circadian rhythmicity and sleep (according to Gene Ontology) are indicated in the heatmap (gene colors indicate cluster location). (C) Smoothing spline (64) of the average of melatonin-aligned median profiles (shown in B) of probes with an increasing trend and of probes with a decreasing trend. (D) Density plot of cumulative trend angle differences between sleep restriction and control. A total of 1,838 paired trend angles (363 probes significant in an average of 5.06 participants) were used for the comparison of upward trends, and a total of 2,454 paired trend angles (470 probes significant in an average of 5.22 participants) were used for the comparison of downward trends. For upward trend (orange), the estimated mean of the differences is 20.1085° [95% CI (19.3552, 20.8618), indicated by orange broken lines; t test P < 2.2 × 10⁻¹⁶]. For downward trend (blue), the estimated mean of the differences is −21.1062° [95% CI (−21.7344, −20.4781), indicated by blue broken lines; t test P < 2.2 × 10⁻¹⁶].