Circadian patterns of gene expression in the human brain and disruption in major depressive disorder

Abstract

A cardinal symptom of major depressive disorder (MDD) is the disruption of circadian patterns. However, to date, there is no direct evidence of circadian clock dysregulation in the brains of patients who have MDD. Circadian rhythmicity of gene expression has been observed in animals and peripheral human tissues, but its presence and variability in the human brain were difficult to characterize. Here, we applied time-of-death analysis to gene expression data from high-quality postmortem brains, examining 24-h cyclic patterns in six cortical and limbic regions of 55 subjects with no history of psychiatric or neurological illnesses ("controls") and 34 patients with MDD. Our dataset covered ~12,000 transcripts in the dorsolateral prefrontal cortex, anterior cingulate cortex, hippocampus, amygdala, nucleus accumbens, and cerebellum. Several hundred transcripts in each region showed 24-h cyclic patterns in controls, and >100 transcripts exhibited consistent rhythmicity and phase synchrony across regions. Among the top-ranked rhythmic genes were the canonical clock genes BMAL1(ARNTL), PER1-2-3, NR1D1(REV-ERBa), DBP, BHLHE40 (DEC1), and BHLHE41(DEC2). The phasing of known circadian genes was consistent with data derived from other diurnal mammals. Cyclic patterns were much weaker in the brains of patients with MDD due to shifted peak timing and potentially disrupted phase relationships between individual circadian genes. This transcriptome-wide analysis of the human brain demonstrates a rhythmic rise and fall of gene expression in regions outside of the suprachiasmatic nucleus in control subjects. The description of its breakdown in MDD suggests potentially important molecular targets for treatment of mood disorders.

Keywords: circadian rhythms; depression; microarray.

Fig. 1.

Discovery of cyclic gene expression in the human brain: examples from the DLPFC. (A) TOD distribution in the controls (n = 52) and patients with MDD (n = 33 in the DLPFC). TODs (zeitgeber time, ZT) were individually adjusted by sunrise time. (B) Heat map of expression levels for top (P < 0.05) cyclic genes (n = 922) in DLPFC samples of 52 control subjects. Genes are shown in the vertical direction and ordered by inferred phase, and samples are shown along the horizontal direction and ordered by ZT across the 24-h day, where sunrise time is ZT = 0. Expression levels for each gene are rescaled by its observed SD. The color scale represents 0.25-fold to fourfold of SD. Red indicates higher expression, and blue indicates lower expression. (C) Expression (Exp) levels of six known circadian genes in samples ordered by TOD. P values and peak times are indicated above each panel. The red lines depict the best-fitting sinusoidal curves.

Fig. 2.

Characterization of the top cyclic genes in the human brain. (A) Comparison of statistical significance for the top cyclic genes across regions. Shown are P values of the top 50 genes across six regions, with the genes ordered by the average logged P value across the six regions. The 11 gene symbols that are highlighted in yellow were annotated as being part of the circadian rhythm pathway in the Kyoto Encyclopedia of Genes and Genomes (KEGG) or the Protein Information Resource (PIR). Among the 41 “core circadian genes” reviewed by Yan et al. (5), 38 were on the microarray platform used in our study and 8 (marked by *) overlapped with the 50 genes shown here. In addition, 5 genes among the 38 (TFRC, NAMPT, USP2, NR1D2, and CRY1) ranked among the top 5% in our study (ranked at 0.7%, 0.7%, 1.3%, 1.6%, and 4.2%, respectively). (B) Peak time of expression for PER genes in our study follows what might be predicted by the animal literature. PER1 expression peaks 0–2 h after sunrise, PER2 peaks in the afternoon, and PER3 peaks in the interval between PER1 and PER2 in all six brain regions.

Fig. 3.

Top cyclic genes show consistent rhythmicity, phasing, and amplitude across brain regions. (A) More than 100 genes exhibit consistently significant rhythmicity. The quantile–quantile plot compares the distribution of the combined P values across the six brain regions (using Fisher’s method) and a uniform distribution, showing that 100–200 genes had smaller combined P values than expected. The top 100 genes were colored in red, and the next 100 genes were colored in green. Gray lines indicate the sorted original P values in the six individual brain regions. The dotted red line indicates uniformly distributed P values. (B) Phasing of the top cyclic genes is consistent across brain regions, as indicated by a heat map of peak times. Genes are ordered from top to bottom by mean peak time. Genes of nonsignificant (P > 0.1) cyclic patterns in a given region were shown as missing (gray) because their peak times could not be accurately determined. (C) Amplitude of rhythms is similarly consistent across brain regions, as indicated by a heat map of the amplitude for 445 transcripts with P < 0.05 in at least two of six regions. Genes are ordered from top to bottom by mean amplitude. (D) Phasing of the top cyclic genes differs between species with different chronotypes (day-active human vs. night-active mouse). Shown is a comparison of peak times for genes that overlapped between a metaanalysis of circadian gene expression in the mouse (5) and our study (P < 0.01 in controls). The y axis shows the peak time in the mouse prefrontal cortex (PFR) or whole brain (WB). The line in the plot models a linear relationship using the 7 top genes (highlighted in red). When fit with robust linear modeling, they revealed a shift of 6.51 h and a slope of 1.18 (r = 0.88).

Fig. 4.

Disruption of cyclic pattern in patients with MDD. (A) Top 16 cyclic genes from controls are not rhythmic in the MDD group. The P values for the genes are formatted similar to Fig. 2A (ranked by the average logged P value across the six regions in controls). (B) Genes in patients with MDD do not exhibit consistently significant rhythmicity, as illustrated by a quantile–quantile plot comparing the combined P values across the six brain regions in MDD (using Fisher’s method) vs. the expected P values in a uniform distribution using the same style as in Fig. 3A. (C) Rhythms of patients with MDD are less synchronized with the solar day compared with controls. The predicted TOD in 55 controls (Left) and 34 patients with MDD (Right) are shown on the inner circle of a 24-h clock, and their documented TODs are shown on the outer circle. The deviations were smaller in controls than in patients with MDD (P = 0.012, Mann–Whitney nonparametric test). (D) Patterns of gene-gene correlations seen in controls (in-phase = positive correlation, out-of-phase = negative correlation) are only partially present in patients with MDD. Depicted are the correlation coefficients across the top 16 genes, calculated using DLPFC data for 52 controls (Left) and 33 MDD cases (Right). Genes are ordered by the peak time derived from the control dataset. Examples of gene pairs with significant differences between controls and patients with MDD are marked with an asterisk.