Effects of aging on circadian patterns of gene expression in the human prefrontal cortex

Significance

Circadian rhythms are important in nearly all processes in the brain. Changes in rhythms that come with aging are associated with sleep problems, problems with cognition, and nighttime agitation in elderly people. In this manuscript, we identified transcripts genome-wide that have a circadian rhythm in expression in human prefrontal cortex. Moreover, we describe how these rhythms are changed during normal human aging. Interestingly, we also identified a set of previously unidentified transcripts that become rhythmic only in older individuals. This may represent a compensatory clock that becomes active with the loss of canonical clock function. These studies can help us to develop therapies in the future for older people who suffer from cognitive problems associated with a loss of normal rhythmicity.

Abstract

With aging, significant changes in circadian rhythms occur, including a shift in phase toward a “morning” chronotype and a loss of rhythmicity in circulating hormones. However, the effects of aging on molecular rhythms in the human brain have remained elusive. Here, we used a previously described time-of-death analysis to identify transcripts throughout the genome that have a significant circadian rhythm in expression in the human prefrontal cortex [Brodmann’s area 11 (BA11) and BA47]. Expression levels were determined by microarray analysis in 146 individuals. Rhythmicity in expression was found in ∼10% of detected transcripts (P < 0.05). Using a metaanalysis across the two brain areas, we identified a core set of 235 genes (q < 0.05) with significant circadian rhythms of expression. These 235 genes showed 92% concordance in the phase of expression between the two areas. In addition to the canonical core circadian genes, a number of other genes were found to exhibit rhythmic expression in the brain. Notably, we identified more than 1,000 genes (1,186 in BA11; 1,591 in BA47) that exhibited age-dependent rhythmicity or alterations in rhythmicity patterns with aging. Interestingly, a set of transcripts gained rhythmicity in older individuals, which may represent a compensatory mechanism due to a loss of canonical clock function. Thus, we confirm that rhythmic gene expression can be reliably measured in human brain and identified for the first time (to our knowledge) significant changes in molecular rhythms with aging that may contribute to altered cognition, sleep, and mood in later life.

Nearly all processes in the brain and body are controlled by a 24-h circadian rhythm. These rhythms are important in regulating the sleep/wake cycle, metabolism, alertness, cognition, and other processes (1). Environmental or genetic disruptions to circadian rhythms are strongly associated with chronic sleep problems, increased rates of cancer, lowered immune function, metabolic disorders, and psychiatric disorders (2, 3). The molecular clock is controlled by a transcriptional/translational feedback loop with the circadian locomotor output cycles kaput (CLOCK) and brain and muscle Arnt-like protein 1 (BMAL1; also known as ARNTL) proteins acting as the major transcriptional activators, and the Period (PER1, PER2, PER3) and Cryptochrome (CRY1, CRY2) proteins acting as the major repressors (4). This core circadian feedback loop regulates the diurnal expression patterns of many different genes as it is estimated that 10–20% of all transcripts have a circadian rhythm (1, 5). Although the master circadian pacemaker in the suprachiasmatic nucleus (SCN) of the hypothalamus synchronizes rhythms throughout the brain and body, the genes that control circadian rhythms are expressed in nearly every cell (6). In recent years, it has become apparent that these genes serve important functions in specific brain regions, including the control of daily rhythms in neuronal activity and the response to environmental stimuli (7–9).

Evidence from preclinical and clinical studies suggests rhythms in the prefrontal cortex (PFC) are particularly important for cognitive performance and executive function. Several studies in humans have reported diurnal differences in cognitive performance and a significant decrease in performance following circadian rhythm disruption (10–12). Interestingly, these measures vary by age with older adults performing better on cognitive tasks in the morning and getting worse throughout the day (12, 13). In older women, there is a direct correlation between weak circadian activity rhythms and poorer executive function (14). In preclinical studies, mice trained in cortical-driven cognitive tasks show pronounced diurnal differences in performance (15–17). Moreover, mice housed under a shortened day (20-h light–dark cycle of 10-h light, 10-h dark) displayed reduced cognitive flexibility and a loss of dendritic length and complexity in the PFC (18).

Physiological and activity rhythms are generally known to deteriorate with aging and show a phase advance toward early morning wakening (19, 20). Daily rhythms in hormones like melatonin and cortisol are decreased as are sleep and body temperature rhythms in older individuals (21). Interestingly, the recognized stimulation of alertness and cognition by blue light seen in young people is diminished in older people, suggesting decreased input to the clock (22). Moreover, the addition of serum from older people to cultured cells that express a circadian reporter construct (Bmal1-luciferase) leads to a shortening of molecular rhythms and a phase advance that is not seen with serum from younger people (23). This finding suggests the presence of circulating factors that alter molecular rhythms with age.

In the human brain, the investigation of brain region-specific transcriptional rhythms across the life span has been challenging due to the lack of control of environmental factors, such as time of death, sleep–wake cycles, and exposure to other factors known to influence rhythms. A recent study overcame some of these obstacles by ordering postmortem samples around a 24-h clock based on their time of death, essentially reconstructing a pseudo-time series by treating each individual sample as an independently sampled data point over one 24-h cycle (24). In 55 healthy controls and 34 patients with major depressive disorder (MDD), they found robust expression rhythms of hundreds of genes in several brain regions in control subjects, with many of these genes displaying significantly disrupted or altered expression rhythms in subjects with MDD. Here, we used a similar analytical approach to identify rhythmic transcripts in two areas of the PFC [Brodmann’s area 11 (BA11) and BA47] in a large sample of 146 subjects from across the life span with no history of psychiatric illness. This large sample allowed us to investigate the effects of normal aging on molecular rhythms in the human PFC and to test the prediction that older individuals display altered or even a loss of molecular rhythms in this brain region.

Materials and Methods

Human Postmortem Brain Samples.

Using the resources of the University of Pittsburgh’s Brain Tissue Donation Program, 210 subjects were identified. Samples were obtained after consent from next of kin during autopsies conducted at the Allegheny County Medical Examiner’s Office (Pittsburgh). The absence of lifetime psychiatric disorders was determined by an independent committee of experienced clinical research scientists using information from clinical records, toxicology results, and a standardized psychological autopsy. Subjects were removed from the study if death was not witnessed because their time of death (TOD) cannot be precisely determined. Subjects that did not meet the criteria of rapid death were also removed from the study. Sixty-four subjects were removed based on these criteria. The final sample consisted of 146 individuals with the following characteristics: mean (range) age of 50.7 (16–96) years, 78% male, 85% Caucasian, mean postmortem interval (PMI) for brain collection of 17.3 h (4.8–28 h), mean pH of 6.7 (5.8–7.6), and RNA integrity number (RIN) of 8.0 (5.9–9.6) (Table S1).

Table S1.

Cohort demographic and clinical information

HU	Age	PMI	pH	RIN	Sex	Race	Cause	Manner	TOD
510	63	12.4	6.5	7.5	M	W	Gastrointestinal bleeding	Natural	−5.83
516	20	14	7.1	8.4	M	B	Gunshot wound to chest	Homicide	16.66
545	65	13.1	6.9	6.2	M	B	Pancreatic carcinoma with pulmonary metastasis	Natural	14.90
546	37	23.5	6.7	8.6	F	W	ASCVD	Natural	6.07
551	61	16.4	6.6	8.3	M	W	Cardiac tamponade	Natural	11.35
567	46	15	6.8	8.9	F	W	Mitral valve prolapse	Natural	14.11
568	60	9.5	6.9	8.7	F	W	ASCVD	Natural	−5.47
575	55	11.3	6.8	9.6	F	B	ASCVD	Natural	15.78
585	26	16	6.7	8.3	M	W	Trauma	Accidental	15.60
604	39	19.3	7.1	8.6	M	W	Hypoplastic coronary artery	Natural	9.27
615	62	7.2	6.4	7.8	M	W	Ruptured abdominal aortic aneurysm	Natural	−3.38
627	43	14.1	7.1	7	F	B	Chronic obstructive pulmonary disease	Natural	14.62
630	65	21.2	7	9	M	W	ASCVD	Natural	6.39
634	52	16.2	7	8.5	M	W	ASCVD	Natural	13.54
643	50	24	6.2	8	M	W	ASCVD	Natural	4.34
645	69	23	6.7	8	M	W	ASCVD	Accidental	4.91
681	51	11.6	7.2	8.9	M	W	Hypertrophic cardiomyopathy	Natural	16.00
688	55	21.5	7	7.2	M	W	Hypertrophic cardiomyopathy	Natural	7.61
694	38	20.7	7	7.7	M	W	Subarachnoid hemorrhage	Natural	8.95
700	42	26.1	7	8.7	M	W	ASCVD	Natural	3.64
704	70	20.6	7	8.7	M	W	ASCVD	Natural	10.56
727	19	7	7.4	9.2	M	B	Multiple trauma	Accidental	−2.83
731	63	10.5	6.8	8.2	F	W	ASCVD	Natural	−4.11
737	66	18.2	7.3	7.5	M	W	ASCVD	Natural	10.81
739	40	15.8	6.9	8.4	M	W	ASCVD	Natural	11.25
746	64	20.8	6.7	7.6	F	B	Peritonitis	Natural	7.90
784	58	5.5	7.1	9.1	M	W	ASCVD	Natural	−2.70
794	52	20	6.8	8.4	M	B	End-stage dilated cardiomyopathy	Natural	6.68
795	68	11.8	6.8	8.2	M	W	Ruptured abdominal aortic aneurysm	Natural	15.13
806	57	24	6.9	7.8	M	W	Pulmonary embolism	Natural	4.31
812	55	23	6.7	8.5	M	W	ASCVD	Natural	4.37
818	67	24	7.1	8.4	F	W	Anaphylactic reaction to cef0tan	Accidental	5.35
825	59	16.9	7.1	7.8	M	W	Cardiac tamponade	Natural	12.60
838	58	16.5	7	8.5	M	W	ASCVD	Natural	12.43
839	74	5.3	7.1	8.9	F	W	Massive BFT	Accidental	−1.15
840	41	15.4	6.8	9.1	F	W	ASCVD	Natural	13.97
841	70	22.3	7.2	7.2	M	W	Hypertrophic cardiomyopathy	Natural	7.69
855	62	11.5	6.7	8.4	M	B	Hypertrophic cardiomyopathy	Natural	17.28
857	48	16.6	6.7	8.9	M	W	ASCVD	Natural	10.13
866	66	12.2	7	8.5	M	W	ASCVD	Natural	15.24
869	51	6.5	7.2	8.9	M	W	Severe ASCVD	Natural	−3.46
871	28	16.5	7.1	8.5	M	W	Trauma (chest)	Accidental	12.04
895	39	15.4	7.6	8.8	M	W	ASCVD	Natural	14.25
902	60	23.6	6.7	7.7	M	W	ASCVD	Natural	4.45
920	60	23	7.2	8.5	M	W	ASCVD	Natural	3.93
932	68	25.9	6.8	8.3	M	W	Cardiac tamponade	Natural	2.23
952	62	5.5	6.6	8.7	M	W	ASCVD	Natural	4.08
953	54	18.6	6.2	7.1	M	W	Extensive BFT (head, torso, and limbs)	Accidental	9.38
969	27	10.6	7	7.7	M	W	Trauma to trunk and extremities	Accidental	−4.91
970	42	25.9	6.4	7.2	M	W	ASCVD	Natural	4.38
972	72	16.4	6.3	7.7	M	W	Trauma (cervical/upper thoracic vertebra)	Accidental	11.71
980	52	11	6.4	8.3	M	W	ASCVD	Natural	−5.96
998	25	6.3	5.9	7.6	M	W	Pulmonary thromboembolism	Natural	2.24
1026	59	19.8	6.3	7.4	M	W	ASCVD	Natural	9.64
1030	96	13.2	6.3	6.8	F	W	Fractured cervical spinal column	Accidental	16.16
1031	53	23.2	6.8	8.9	M	W	Arteriosclerotic and hypertensive CVD	Natural	6.71
1038	56	14.6	6.5	7	M	W	Rupture of dissecting hematoma of aorta	Natural	14.64
1047	43	13.8	6.6	9	M	W	ASCVD	Natural	15.15
1054	44	17.1	6.6	8.8	F	W	ASCVD	Natural	10.20
1065	85	20.2	6.3	7.1	M	B	Pulmonary embolism	Accidental	7.32
1067	49	6	6.6	8.2	M	W	Hypertensive heart disease	Natural	−1.95
1083	20	19.9	6.6	8.8	M	W	Trauma (head and trunk)	Accidental	9.20
1091	58	16.5	6.6	8.1	M	W	ASCVD	Natural	6.67
1092	40	16.6	6.8	8	F	B	Mitral valve prolapse	Natural	11.81
1099	24	9.1	6.5	8.6	F	W	Cardiomyopathy not otherwise specified	Natural	1.37
1103	89	22.9	6.8	7.8	M	W	BFT of head and trunk	Accidental	8.12
1108	60	23.3	6.3	8.1	M	W	ASCVD	Natural	6.92
1114	29	11.4	6.8	9.4	F	W	Trauma (trunk)	Accidental	−5.57
1119	57	20.2	6.8	8	M	W	ASCVD	Natural	8.62
1122	55	15.4	6.7	7.9	M	W	Cardiac tamponade	Natural	13.18
1129	54	21	6.8	9	M	W	ASCVD	Natural	7.91
1141	54	19.2	6.6	8.5	M	W	Arteriosclerotic and hypertensive CVD	Natural	9.48
1142	44	12.3	5.8	6.1	M	B	Hypertensive and arteriosclerotic CVD	Natural	15.98
1144	75	21.4	6.7	7.3	M	W	Arteriosclerotic and hypertensive CVD	Natural	7.75
1153	55	28	6.6	8	M	W	Atherosclerotic and hypertensive heart disease	Natural	3.69
1159	51	16.7	6.5	7.6	M	W	Hypertensive heart disease	Natural	11.81
1165	67	21	6.6	7.5	F	W	Cardiac tamponade	Natural	10.49
1171	56	19.2	6.5	8.5	M	W	Cardiac tamponade	Natural	8.55
1172	52	16.8	6.5	7.7	F	B	Hypertensive heart disease	Natural	11.79
1191	59	19.4	6.2	8.4	M	B	Atherosclerotic and hypertensive heart disease	Natural	9.24
1193	31	12.8	6.5	7.6	M	W	Myocardial infarction due to ASCVD	Natural	15.64
1194	27	16.4	6.4	8.8	M	W	ASCVD	Natural	12.63
1196	36	14.5	6.4	8.2	F	W	Positional asphyxia	Accidental	17.15
1201	52	16.4	6.2	8.1	F	W	ASCVD	Natural	12.51
1212	49	9.2	6.6	8.2	M	B	Massive subarachnoid hemorrhage	Natural	−4.42
1214	57	16.4	6.4	7.5	M	W	Arteriosclerotic and hypertensive CVD	Natural	12.56
1219	56	14.9	6.5	8.5	M	W	ASCVD	Natural	14.77
1237	64	8.7	6.4	7.7	M	W	ASCVD	Natural	0.70
1239	22	9.5	6.7	7.4	M	W	BFT to the head and neck	Accidental	−5.42
1245	59	21.6	6.6	7.4	M	W	ASCVD	Natural	7.37
1247	58	22.7	6.4	8.4	F	W	ASCVD	Natural	8.46
1264	38	21.1	6.7	8.6	M	W	Cardiomyopathy	Natural	5.60
1270	73	19.7	6.7	7.7	F	W	BFT of neck and trunk	Accidental	8.62
1276	57	14.9	6	6.1	F	W	Arteriosclerotic CVD	Natural	13.90
1280	50	23.5	6.7	7.7	F	W	Pulmonary thromboembolism	Natural	3.95
1288	59	19.7	6.9	7.4	M	W	Atherosclerotic and hypertensive heart disease	Natural	9.49
1293	65	18.5	6.6	7	F	W	BFT of trunk	Accidental	11.51
1298	48	24.5	6.8	7.9	M	W	ASCVD	Natural	4.78
1300	56	16.5	6	6.9	F	W	Pulmonary thromboembolism	Natural	13.22
1306	18	16	6.6	7.9	M	W	Congenital heart anomaly	Natural	13.54
1307	32	4.8	6.7	7.6	M	B	Hypertensive cardiomyopathy	Natural	4.03
1308	58	20.2	6	6.2	M	W	Hepatocellular carcinoma	Natural	8.43
1317	56	22.9	6.5	8.8	M	W	Arteriosclerotic and hypertensive CVD	Natural	4.80
1326	58	16.4	6.7	8	M	W	ASCVD	Natural	13.47
1333	46	18.2	6.8	8.7	M	W	Arteriosclerotic and hypertensive CVD	Natural	10.65
1335	18	14.6	7	8.7	M	W	BFT of head and trunk	Accidental	16.06
1350	21	24.2	6.4	7.3	M	W	BFT of trunk	Accidental	6.62
1371	33	13	6.2	8.5	M	B	Eosinophilic myocarditis	Natural	−5.61
1374	43	21.7	6.6	7.2	M	W	Coronary atherosclerotic disease	Natural	7.91
1378	53	20.1	6.5	7.8	F	B	Coronary artery disease	Natural	8.46
1382	73	25.3	6.6	7.9	F	W	BFT of head with ponto-midbrain transection	Accidental	2.87
1391	51	7.8	6.6	7.1	F	W	Atherosclerotic coronary artery disease	Natural	−0.59
1394	45	17.3	6.6	7.3	M	W	Atherosclerotic coronary artery disease	Natural	10.12
1400	50	16	5.8	6.8	M	W	Coronary atherosclerotic disease	Natural	11.40
1410	54	15.3	6.9	7.8	M	W	Atherosclerotic coronary artery disease	Natural	13.11
1433	65	24.1	6.8	7.5	M	W	Acute subdural hematoma with skull fracture	Accidental	7.18
1434	25	25.5	6.7	7.5	M	B	Cardiac arrhythmia while exercising	Natural	5.35
1441	57	22.3	6.5	8.5	M	W	Arteriosclerotic CVD	Natural	7.88
1444	46	22	6.5	8.4	M	W	Pulmonary thromboembolus	Natural	5.25
1447	51	16.2	6.5	8.5	M	W	Calcified atherosclerotic coronary artery disease	Natural	15.53
1460	53	21.3	6.2	7.3	M	W	Severe ASCVD	Natural	6.28
1462	47	17.2	6.6	8.5	M	W	ASCVD	Natural	10.58
1466	64	20	6.7	8.8	F	B	BFT to trunk	Accidental	9.33
1482	25	20.2	6.6	9.1	M	W	ASCVD	Natural	8.26
1488	39	21.5	6.4	8.7	M	B	Bilateral pulmonary thromboemboli	Natural	6.38
1497	38	8.5	6.3	9.1	M	W	ASCVD	Natural	−2.80
1511	48	23.5	6.8	7.2	M	W	Severe ASCVD	Natural	4.30
1514	60	25	6.4	6.6	F	W	Bilateral pulmonary emboli	Natural	2.91
1524	66	9.4	6.4	8.1	M	W	Acute small intestinal ischemic infarction	Natural	−4.21
1541	80	10.5	6.5	8	M	W	Pulmonary fibrosis	Natural	17.40
1554	50	23.2	6.5	7.6	M	W	Acute myocardial infarction	Natural	7.67
1555	17	15.1	6.9	7.9	M	W	BFT to head, thorax, abdomen, right	Accidental	16.70
1558	54	24.4	6.9	7.7	M	W	ASCVD	Natural	4.60
1559	45	8.9	6	6.9	M	W	Cardiomegaly	Natural	−1.66
1564	48	19.9	6.8	7.5	M	W	BFT to trunk	Accidental	8.99
1574	52	26.7	6.3	7	M	W	Bilateral pulmonary embolism	Natural	2.07
1583	58	19.1	6.8	8.2	M	W	BFT of the trunk	Accidental	9.10
1598	50	23.8	6.9	7.8	M	W	Acute myocardial infarction	Natural	10.12
1623	52	25.7	6.3	6.6	M	W	BFT of head and neck	Accidental	4.86
1637	46	16.6	6.9	8.2	M	W	Coronary artery atherosclerosis	Natural	15.86
10003	49	21.2	6.5	8.4	M	W	Multiple blunt force injuries	Accidental	9.82
10005	42	23.5	6.7	7.4	M	W	Multiple blunt force injuries	Accidental	7.43
10013	16	9.3	6.7	9	F	W	Trauma (multiple blunt force injuries)	Accidental	−3.88
10016	18	16.3	6.8	5.9	M	B	Gunshot wound of abdomen	Homicide	15.46
10021	38	12.7	6.3	7.2	M	W	Multiple blunt force injuries	Accidental	16.86
10027	53	13.1	6.2	6.6	F	B	Congestive heart failure	Natural	17.13

ASCVD, atherosclerotic cardiovascular disease; B, African American; BFT, blunt force trauma; CVD, cardiovascular disease; F, female; HU, subject no.; M, male; pH, brain tissue pH; PMI, postmortem interval; RIN, RNA integrity number; TOD, time of death (zeitgeber time); W, Caucasian.

For each subject, the right hemisphere was blocked coronally, frozen, and stored at −80 °C. Blocks containing BA11 or BA47 of the orbital PFC were cut on a cryostat, and cortical gray matter was collected for RNA extraction as previously described (25). All procedures were approved by the University of Pittsburgh’s Institutional Review Board for Biomedical Research and Committee for Research Involving the Dead.

Time of Death Analysis in the Zeitgeber Timescale.

To analyze rhythmic gene expression, the TOD for each subject was normalized to a zeitgeber time (ZT) scale. For each subject, both place and time of death was collected, where the time zone of the place of death was used to adjust the reported TOD from local time to coordinated universal time. Daylight savings time was also incorporated when appropriate. The sunrise time was calculated according to the date and the longitude and latitude of the place of death. Subject’s TOD was set as ZT = t hours after previous sunrise (if t < 18) or before next sunrise (if t ≥ −6).

Microarray Data Preprocessing.

All samples were analyzed using the Affymetrix Human Gene 1.1 ST microarray platform. Gene expression values were corrected, quantile-normalized, and log₂-transformed via Affymetrix Expression Console software (build 1.2.1.20) using RMA algorithm. A total of 33,297 probe sets were available on each array; only those probe sets with annotated gene symbols were selected (20,237 gene symbols). If a gene was represented by multiple probe sets, the one with the largest intensity interquartile region was selected to represent that gene. Microarray data for each brain region were analyzed separately. The raw and processed microarray data were deposited in the National Center for Biotechnology Information Gene Expression Omnibus database (GSE71620).

Detection of Circadian Gene Expression Patterns.

Nonlinear regression was used to detect circadian gene expression patterns. All individual samples were ordered by their TODs. Temporal expression was fitted to a sinusoidal curve using the nonlinear least-squares method. The coefficient of determination (R²) was used as a proxy of goodness-of-fit. The empirical P value was estimated from a null distribution of R² generated from 1,000 TOD-randomized expression datasets for BA11 and BA47 separately. Adaptive weighted Fisher method was adopted to combine multiple P values across brain regions (26). The q value was estimated using R package qvalue (27).

Analysis of Aging Effects on Expression Rhythmicity.

Subjects were classified as younger (<40 y, n = 31) or older (>60 y, n = 37) and analysis of rhythmic gene expression was performed on each group independently as described above. Age effect P values were estimated from 1,000 age-shuffled data matrixes for BA11 and BA47 separately (SI Materials and Methods).

SI Materials and Methods

Detection of Circadian Gene Expression Patterns.

Nonlinear regression was used to detect circadian gene expression patterns. All individual samples were ordered by their TODs, and for each gene, this temporal expression was fitted to a sinusoidal curve using the nonlinear least-squares method. The sinusoidal curve was defined as follows:

$$y=A\sin \left(Fx+\mathrm{phase}\right)+\mathrm{offset},$$

where A represents the amplitude and F represents the frequency, which was fixed at 24-h period. The nonlinear fitting was performed based on the Levenberg–Marquardt algorithm, which was implemented in SciPy’s Optimization package (release 0.13.2). The peak hour of the circadian pattern was calculated directly from the fitted curve.

The Choice of Cohort Age Cutoffs.

We chose age 40 and 60 as the cutoffs in the age stratification analysis for the following reasons:

• i)Age 60 or 65 are both widely used cutoffs to define elders, and they generally agree with the typical retirement age in the developed countries (e.g., United Nations has used 60+ to define old age). We also chose >60 because it generates a sample size (n = 37) that is comparable to the younger group (<40) (n = 31).
• ii)For this type of analysis, we require strong signals to distinguish different circadian gene expression patterns between the younger and the older groups. Therefore, we have to choose relatively extreme age groups while trying to keep a level of statistical power needed to detect significant differences (i.e., the group sizes cannot be too small and should be approximately equal sizes).
• iii)Moreover, current age cutoffs (40, 60) are close to the first quartile (Q1, 25%, 42.45 y) and the third quartile (Q3, 75%, 59.75 y), respectively. In addition, they are of approximately equal distance from the mean (50.67 y) and median (52.50 y).

Analysis of Aging Effects on Expression Rhythmicity.

First, subjects were classified as younger (<40 y, n = 31) or older (>60 y, n = 37) by age as mentioned above. Next, the procedures of detection of circadian gene expression patterns (mentioned in the main text) were applied to each group and each brain region separately. Then, for each gene, the absolute difference of the parameters (i.e., Δbase, Δamplitude, Δphase, and ΔR²) between the younger and the older group can be calculated by comparing their best-fitting circadian curves (P < 0.05) that we obtained at the previous step. After that, we generated 1,000 age-shuffled expression data matrixes (i.e., subjects were randomly assigned to either younger or older group) for BA11 and BA47 separately, by which we repeated all of the procedures to collect a null background of the delta values and estimated the empirical P values of each observation against the null background. The age effect on circadian expression patterns between the younger and the older groups were categorized as follows: change in the level of basal expression (Δbase); change in the amplitude of oscillation (Δamplitude); change in the acrophase of rhythmicity (Δphase); a loss of rhythmicity typically observed as a disruption of circadian rhythmicity of gene expression oscillation (ΔR²); and the gain of rhythmicity typically observed as a gain of circadian rhythmicity of gene expression oscillation (ΔR²). These categories are not mutually exclusive except for the loss and gain of rhythmicity. Note that the comparison of ΔR² was not limited to rhythmic or arrhythmic genes in either age group. Any gene that showed significant R² decrease or increase (empirical P < 0.05) in the older group would be annotated as “loss of rhythmicity” or “gain of rhythmicity,” respectively.

Results

Gene Expression Rhythms in Human BA11 and BA47.

In 146 subjects, we performed a comprehensive gene expression analysis in BA11 and BA47 of the PFC (Fig. S1A). In BA11 2,475 genes (∼12% of the genome) and in BA47 a total of 1,615 (∼8% of the genome) displayed detectable circadian rhythmicity (P < 0.05) (Dataset S1). To identify a set of genes with consistent circadian regulation of gene transcripts across BA11 and BA47, we applied a metaanalysis approach across the brain regions and limited the false-discovery rate to 5% (i.e., q < 0.05). A total of 235 genes with a significant circadian expression pattern were detected at q < 0.05 (Dataset S1 and Table S2). Notably, these genes displayed coordinated temporal expression patterns across BA11 and BA47, as illustrated by the nearly identical patterns observed across the two areas (Fig. 1 A and B). Indeed, the peak or acrophase of expression of those 235 genes was highly concordant (92%) between BA11 and BA47 (Pearson’s r = 0.95, P = 5.8e-120) (Fig. 1C). The core genes that make up the molecular clock showed robust expression rhythms in the PFC (Fig. 2). Their patterns of expression were strikingly similar across the two brain regions and also to those published previously (24), demonstrating the consistency in the use of these analyses to detect significant rhythms in gene expression in human postmortem tissue. The top 50 genes with significant circadian rhythms in expression are listed in Fig. 3. For example, C1ORF51 (more recently known as CIART) has the most robust expression rhythm (Fig. S2). The mouse ortholog of this gene, Gm129, has been recently identified as having a circadian pattern of expression in mice (28, 29). Many of these genes display circadian expression patterns in peripheral tissues of the mouse, such as the liver, heart, and/or skeletal muscle, and in the brain, including the SCN and cerebellum. For example, DUSP4, ANKRD12, TRIM24, and TMEM119 display expression rhythms in the mouse SCN (30), whereas DUSP11, USP2, SESN3, BACE2, and PEX1 are expressed rhythmically in other areas of the brain (30, 31). Notably, some of the genes we identified here as displaying rhythmicity in the human PFC are not known to be involved in circadian rhythms, or even to be expressed in a circadian pattern, suggesting these genes may have previously unidentified roles within or downstream of the molecular clock. For example, KCNH4 (potassium voltage-gated ion channel) is a top-ranked gene with strong circadian expression patterns across BA11 and BA47 (Fig. 3 and Fig. S2) and there are no previous reports of this transcript being regulated in a circadian manner. Other genes with a similarly strong pattern and no prior knowledge of rhythmicity include KIAA1370, RSP02, ZNF43, BACE2, ZBBX, and FGD5.

Fig. S1.

Overview of analysis flows. (A) Detection of clock-regulated genes in human prefrontal cortex (PFC) via metaanalysis. We first estimated the P values of expression rhythmicity for all of the genes in brain regions BA11 and BA47 independently. Then we combined these P values to meta-P values via AW-Fisher and corrected them into q values. A total of 235 cross-region PFC circadian genes were detected at q < 0.05. (B) Detection of age effect on circadian expression rhythmicity. The work flow was independent from A; i.e., the analysis in B did not rely on any a priori results yielded in A. It is possible that a gene shows expression rhythmicity only in younger individuals but not in older individuals (or could be vice versa). Therefore, if younger subjects were mixed with older subjects (e.g., when all 146 subjects were included as how we did in A), we might be unable to detect the expression rhythmicity of certain genes (see Fig. S4 for such examples).

Table S2.

Numbers of significant circadian genes across BA11 and BA47 at different q value cutoffs after metaanalysis

Cutoff	No. of genes*	Concordant phase (%)†	Average R2
q < 0.01	84	80 (95%)	0.13
q < 0.05	235	217 (92%)	0.09
q < 0.1	465	411 (88%)	0.07

^*The total number of genes is 20,237.

^†Concordant phase was defined as the difference of peak hour ≤4 h between BA11 and BA47.

Fig. 1.

Heat map of expression levels and comparison of circadian acrophase for the top circadian genes (n = 235, q < 0.05) in BA11 and BA47. (A and B) Expression levels were Z-transformed for each gene. Red indicates higher expression level; green indicates lower expression levels. (C) The circadian phase (peak hours) of 235 circa genes derived from metaanalysis are plotted on TOD axes for BA11 (x axis) and BA47 (y axis). Red dashed line: 1:1 diagonal line. Green dashed lines: ±4-h phase concordance boundaries. Ninety-two percent of circa genes (217 of 235) are located within the concordance interval.

Fig. 2.

Circadian gene expression patterns of six canonical circadian genes in BA11 (A) and BA47 (B). The x axis denotes the time of death (TOD) on ZT scale (−6–18 h). Approximate day–night interval within a pseudoday is represented by yellow (day) and black (night) bands. Data points from 146 subjects are plotted in blue. The best-fitted sinusoidal curves are depicted in red. The empirical P value and the estimated peak hour are also reported above each panel.

Fig. 3.

Top 50 clock-regulated genes across two brain regions. Gene list is sorted by q value after adaptive weighted (AW)-Fisher procedure. Genes in red are known circadian-related genes according to their records in GeneCards database (59, 60).

Fig. S2.

Top eight rhythmic genes in BA11 (A) and BA47 (B). The x axis denotes the time of death (TOD) on ZT scale (−6–18 h). Approximate day–night interval within a pseudoday is represented by yellow (day) and black (night) bands. Data points from 146 subjects are plotted in blue. The best-fitted sinusoidal curves are depicted in red. The empirical P value and the estimated peak hour are also reported above each panel.

Effects of Aging on the PER Genes Across BA11 and BA47 of Human PFC.

To test the prediction of altered rhythmic patterns of gene expression with aging, we stratified our subjects into younger (<40 y, n = 31) and older (≥60 y, n = 37) groups (Fig. S1B and Fig. S3) and performed an analysis of circadian rhythms of canonical circadian genes of the molecular clock within the two subgroups. Sinusoidal curve analyses were used to investigate differences in total expression levels, amplitude, acrophase, and absolute changes in rhythmicity (loss or gain of rhythm). We found consistent effects of age (younger vs. older) on the circadian pattern of expression for PER1 and PER2, but very minimal changes in PER3, across BA11 and BA47 (Fig. 4). Specifically, the acrophases of PER1 rhythms were shifted from ZT5 (BA11) and ZT7 (BA47) to ZT1–ZT2 with a reduction in amplitude in both regions (Fig. 4 A and B), suggesting significantly disrupted temporal expression of PER1 in older individuals. Similar phase shifts in peak expression were observed for PER2 (Fig. 4 A and B)—acrophase was shifted from sunset (∼ZT12) to the middle of the day (ZT6). However, there were no obvious age effects on expression of PER3 in either BA11 or BA47 (Fig. 4 A and B).

Fig. S3.

Cohort age information in this study.

Fig. 4.

Aging has unique effects on the circadian gene expression patterns of period genes (PER family) in both BA11 (A) and BA47 (B). Older people have disrupted PER1 circadian expression patterns (P = 0.027 in BA11; P = 0.0005 in BA47), together with a phase advance of PER2 expression from sunset to noon (P = 0.005 in BA11; P = 0.004 in BA47), whereas their PER3 expression remained intact.

Effects of Aging on Rhythms of Gene Expression in BA11.

We next conducted separate, whole-genome analyses on BA11 and BA47 to determine whether the age effects on patterns of expression differed between these two regions beyond the canonical circadian genes of the molecular clock. We identified 1,186 genes in BA11 that exhibited age-dependent rhythmicity or alterations in rhythmicity patterns with aging (P < 0.05; Table S3), which can be categorized into five characteristics (Fig. 5): (i) a base shift (meaning a change in levels of expression but not necessarily a change in rhythm); (ii) a decrease (but not complete loss) in amplitude; (iii) a phase shift; (iv) a significant loss of rhythmicity; and (v) a gain of rhythmicity. A total of 201 genes in BA11 do not display any impairment in rhythmicity with aging, including many of the canonical clock genes (Dataset S2). A single gene can display multiple kinds of age-related effects on its rhythmicity pattern. For example, a gene may display both a base shift and a phase shift. Complete lists of genes in each category are in Datasets S2 and S3. When differences between groups were analyzed, we found that, in BA11, there were 30 genes that showed significant changes in level of expression (base shift) while retaining overall rhythmicity from younger to older individuals—for example, several of the most robustly rhythmic genes, PER1, KCNH4, and SPRY4 fell into this category (Fig. S4). There were 67 genes that displayed significant phase shifts in older individuals—for example, among the top circadian genes, PER2, BACE2, and SPRY4 are all significantly phase-advanced in older individuals (Fig. S4). An impressive 588 genes showed a complete loss of rhythmicity due to age, including ADRA1b of the top circadian genes, and a canonical circadian gene, CRY1. Only two genes had dampened amplitudes between age groups. These were C1ORF51 and LACRT. Finally, we identified 533 genes in BA11 that were not rhythmic in younger adults but appeared to gain rhythmicity with age, such as ATM, ME2, and MEPCE (Dataset S2).

Table S3.

Summary of detected age effects on circadian gene expression patterns in BA11 and BA47

		Age effects					No effect
Class	n	Base shift	Amp change	Phase shift	Loss of rhythmicity	Gain of rhythmicity
BA11
PFC circa genes	235	6	1	10	19	6	201
Others	20,002	24	1	57	569	527	18,849
Total	20,237	30	2	67	588	533	19,051
BA47
PFC circa genes	235	8	0	11	20	7	193
Others	20,002	27	3	83	1,043	427	18,453
Total	20,237	35	3	94	1,063	434	18,646

Note that the age effects are not necessary mutually exclusive. The total numbers of genes with age effects on expression rhythmicity are 20,237 − 19,051 = 1,186 in BA11 and 20,237 − 18,646 = 1,591.

Fig. 5.

Illustration of age effect on circadian gene expression patterns. The age effect on circadian gene expression patterns can be classified into five categories, as illustrated (rows 1–5). The number of instances and the top three most significant examples (sorted by P values) were also reported. See Datasets S2 and S3 for complete lists.

Fig. S4.

List of detected age effects on 235 PFC circadian genes in BA11. Upward arrow represents rise of circadian expression rhythmicity base in the older. Downward arrow represents drop of base or decrease of amplitude in the older. Phase advance means the peak hour of circadian expression rhythmicity comes earlier in the older. See Dataset S2 for full table.

Effect of Aging on Rhythms of Gene Expression in BA47.

Compared with BA11, there were many more alterations in patterns of gene expression in BA47 between older and younger individuals (1,591 genes), suggesting that this area might be more strongly influenced by age (Table S3 and Dataset S3). We identified 35 genes that had a change in overall level of expression including ARNTL (BMAL1), BHLHE40, DBP, KCNH4, NR1D1, and SPRY4 of the top 50 circadian genes (Fig. S5). Ninety-four genes displayed a significant shift in acrophase due to age, such as BHLHE40, DUSP4, PER2, and SPRY4 of the top circadian genes (Fig. S5). There were three genes that had a reduction in amplitude without a loss of rhythm, NOVA2, ADAMTS12, and HNRNPA3. A large number of genes (1,063) lost rhythmicity due to age in this rain region, with several of these being in the top circadian genes and also having primary roles in molecular clock function, such as PER1, and other genes, including ADRA1B, PDE7B, and RARA. Interestingly, there were 434 genes that gained rhythmicity in older individuals, including two microRNAs, miR128-1 and miR15a. Of note, 30 genes that gained rhythmicity in older individuals in BA47 overlapped with BA11, such as A2M, CDKN1A, and miR30a. Finally, 193 genes in BA47 do not display any impairment in rhythmicity with aging (Dataset S3).

Fig. S5.

List of detected age effects on PFC circadian genes in BA47. Upward arrows and downward arrows represent rise and drop of circadian expression rhythmicity base in the older, respectively. No instance was found with significant amplitude change. Phase advance means the peak hour of circadian expression rhythmicity comes earlier in the older. Phase delay means the peak hour comes later in the older. Phase reverse means the difference of peak hours between the younger and the older groups is nearly 12 h. See Dataset S3 for full table.

Several genes are overlapped in BA11 and BA47 in each age effect categories (Fig. S6). Most of the overlapped genes have the same direction of changes (Dataset S4). We also found significant co-occurrence between base shift and phase shift (Fig. S7). Notably, the top circadian genes with detectable age effects tend to have a downward base shift (base dropped) and/or an advanced phase (early peak hour) for the older subject group (Figs. S4 and S5).

Fig. S6.

Venn diagram of gene numbers in each age effect category between BA11 and BA47. Only those genes with P < 0.05 in each category were counted here (Datasets S2 and S3). See also Dataset S4 for complete lists. Shown are interactions between BA11 and BA47 in genes that show (A) base shift, (B) amplitude change, (C) phase shift, (D) loss of rhythmicity, or (E) gain of rhythmicity.

Fig. S7.

Co-occurrence of age effects on circadian expression rhythmicity. Shown in each cell is the number of co-occurrence and the associated P value (by Fisher’s exact test) between every two age effect categories. Note that the loss of rhythmicity and the gain of rhythmicity are two mutually exclusive categories by definition. Color code: blue cells, cases in BA11; green cells, cases in BA47.

Discussion

Our data are the first (to our knowledge) to indicate that age significantly alters the circadian rhythms of gene expression in the human PFC. Our data extend and replicate the findings of Li et al. (24) in larger cohort and further demonstrate that rhythms in gene expression in human postmortem tissue are robust and can be accurately measured. Moreover, the phase and amplitude of the clock-regulated genes analyzed here show remarkable consistency with the rhythms measured in these same genes in the Li et al. study. There are limitations to this type of analysis because we do not have information on these subjects regarding their lifestyle, actigraphy, and sleep time before death, which could contribute to changes in molecular rhythms. However, the independent confirmation and consistency between our cohort and the Li et al. study provides a high level of confidence in the overall results.

We decided to focus this study on two regions of the orbitofrontal cortex, BA11 and BA47. These are areas commonly associated with focused attention, executive functioning, and depression (32, 33). Although the Li et al. paper looked at genes that were rhythmic across six brain regions with more diverse cytoarchitectural structure (cortical, subcortical, cerebellum), we find many of the same genes to be rhythmic in BA11 and BA47 including most of the known circadian genes (i.e., ARNTL, PER1, PER2, PER3, NR1D1, NPAS2, etc.). Neither study identified CLOCK as a significantly rhythmic gene, which is consistent with its being constitutively expressed in other brain regions, the SCN in rodents and even in flies (34, 35). Interestingly, some of our most rhythmic genes identified, C1ORF51, KCNH4, and OPRL1, were not found in the Li et al. study, suggesting that perhaps their rhythmicity is specific to the areas of the brain that we investigated. Generally, we find very high concordance between rhythms in BA11 and BA47. This is interesting because studies in rodents have suggested that there can be differences in phase between adjacent brain regions (36, 37).

Several of the genes that have a significant rhythm in expression have not been described previously as core circadian genes or clock-regulated genes. One example is KCNH4, which is a voltage-gated potassium channel in the ether-a-go-go family (38). Its expression is largely restricted to the brain, although little is known about its function. Another is PDZRN3, a ubiquitin ligase known to be critically involved in cell differentiation via Wnt signaling in a variety of cell types throughout development (39, 40). We also identified certain genes that are known to play a role in rhythm regulation in the SCN or other tissues in rodents. For example, OPRL1 (also known as nociception receptor) is an opiate receptor that is strongly expressed in the mouse SCN where it binds nociception/orphanin FQ, which suppresses 88% of SCN neurons (41). Moreover, OPRL1 activation down-regulates PER2 in the SCN and accelerates reentrainment of rhythms following a shift in the light–dark cycle (42). The importance of the rhythms in this protein in BA11 and BA47 function in human brain is not known. Another example is C1ORF51 (also known as Gm129, CHRONO, and CIART). Annayev et al. (29) recently found that this protein acts as a novel transcriptional repressor with high-amplitude oscillations in the mouse liver that directly interacts with BMAL1 to repress CLOCK/BMAL1 function in this tissue. Here, we find that this transcript has the highest level of rhythmicity of any gene that we identified in human cortex, suggesting it may have a similar important circadian function in human brain. Interestingly, RNF115 [also known as breast cancer-associated gene 2 (BCA2)] was strongly rhythmic in our study. RNF115 is another ubiquitin E3 ligase that is strongly associated with increased tumor growth, particularly in response to estrogen (43). Many studies have found significant correlations between disrupted circadian rhythms and increased risk for certain cancers, including breast cancer (44). Ubiquitination regulates the stability of core clock proteins; thus, it is possible that this protein interacts with core circadian proteins to regulate rhythms in cell growth and differentiation (45). Again, the role of this protein in the human PFC is not clear.

Age-stratified analysis enables us to identify age group-specific circadian genes that are undetectable from a cohort of heterogeneous age composition. For example, FKBP5 shows a reversal of circadian rhythmicity between the younger and the older subjects (Fig. S8A). Its rhythmicity thus cannot be detected from a mixed age cohort composed of both younger and older subjects. Another example is IGF1, which also shows distinct rhythmicity between the younger and the older (Fig. S8B).

Fig. S8.

Rhythmicity expression profiles before and after age stratification. (A) FKBP5. (B) IGF1. (Upper) Before age stratification (n = 146). (Lower) After age stratification (n = 31 in younger and n = 37 in older).

With aging, we found a disruption in several rhythmic transcripts, with the effects somewhat more pronounced in BA47 than in BA11. The reason for this difference is not clear. Samples were processed in parallel from dissection to array hybridization, making technical differences unlikely. Thus, differences may reflect distinct functions and underlying biology of the two regions. The most common deficit that we identified was a complete or partial loss of rhythmicity in transcripts that were rhythmic in younger people. We also identified certain transcripts that had a shift in phase. This is interesting because it is well documented that healthy older individuals tend to experience a shift toward “morningness” where they prefer to wake early in the morning and go to sleep relatively early in the evening (20, 46). There are natural variations in chronotype at all ages with some showing morning preference and some showing evening preference. Extreme chronotypes are associated with multiple psychiatric disorders (47–50), and it would be interesting in future studies to determine how molecular rhythms correlate with chronotype. Interestingly, a condition called “sundowning” or “sundown syndrome” affects some 20–40% of older people with dementia or Alzheimer’s disease where they become confused, delirious, anxious, and agitated in the evening when the sun goes down, causing them to wander, become combative, and have difficulties sleeping (51). Sundowning is thought to be directly related to the breakdown in circadian rhythms in these individuals (52). Although our study examined only healthy individuals, it is possible that some of the genes that experience a loss of rhythmicity with aging might be linked to temporal changes in cognition.

Somewhat surprisingly, we identified a number of transcripts that gain rhythmicity with aging. Very interestingly, two of these transcripts were micro-RNAs miR128 and miR15a. miR128 is involved in the regulation of amyloid-β degradation in Alzheimer’s disease and is involved in tumor suppression (53, 54), and miR15a is also involved in tumor growth and its expression is correlated with plaque score in Alzheimer’s disease (55, 56). A survey of age-related microRNA expression changes in the neuronal stem cell niches of the short-lived annual fish Nothobranchius furzeri found that miR15a was abundant only in older animals (57). It will be interesting in future studies to examine postmortem samples from subjects with conditions such as Alzheimer’s disease to determine whether there are more extreme changes in circadian gene expression. The gain of rhythmicity in these micro-RNAs could impact the cyclic expression of a whole host of proteins, perhaps leading to a novel, compensatory clock that is activated when the canonical clock breaks down. A recent study by Eckel-Mahan et al. (58) had a similar finding in the liver of mice fed a high-fat diet. Mice on regular chow showed normal cycling of canonical circadian genes in the liver and the rhythms in these genes were severely disrupted with a high-fat diet. However, another set of genes became rhythmic with the high-fat diet, suggesting a reprogramming of the liver to allow it to handle this change in diet. It is possible that a similar mechanism occurs in the brain of older individuals, concurrently with a partial breakdown of normal rhythmicity. More studies will need to be done, however, to determine the true significance and potential mechanisms regulating the gain of rhythmicity in these transcripts.

Conclusions

In conclusion, our study demonstrates the power of TOD transcriptomic analysis to identify transcripts with a circadian rhythm in human postmortem brain. We find many genes to have a significant rhythm in BA11 and BA47 with a high concordance in their phase. Furthermore, we have identified age-related changes in rhythmic transcript expression and somewhat surprisingly have uncovered a previously unidentified set of genes that gain rhythmicity in older individuals. These studies will help us better understand how rhythms change during aging and how targeted therapies that enhance molecular rhythmicity might be developed to prevent conditions like sundowning or enhance cognition. Moreover, in the future, we can use this approach to identify changes in molecular rhythms in a variety of brain regions or specific cell types that associate with psychiatric and neurological diseases.